xref: /openbmc/linux/mm/filemap.c (revision 5aaa4024)
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/syscalls.h>
26 #include <linux/mman.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/uio.h>
30 #include <linux/error-injection.h>
31 #include <linux/hash.h>
32 #include <linux/writeback.h>
33 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h>
35 #include <linux/security.h>
36 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include <linux/migrate.h>
46 #include <linux/pipe_fs_i.h>
47 #include <linux/splice.h>
48 #include <asm/pgalloc.h>
49 #include <asm/tlbflush.h>
50 #include "internal.h"
51 
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/filemap.h>
54 
55 /*
56  * FIXME: remove all knowledge of the buffer layer from the core VM
57  */
58 #include <linux/buffer_head.h> /* for try_to_free_buffers */
59 
60 #include <asm/mman.h>
61 
62 #include "swap.h"
63 
64 /*
65  * Shared mappings implemented 30.11.1994. It's not fully working yet,
66  * though.
67  *
68  * Shared mappings now work. 15.8.1995  Bruno.
69  *
70  * finished 'unifying' the page and buffer cache and SMP-threaded the
71  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
72  *
73  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74  */
75 
76 /*
77  * Lock ordering:
78  *
79  *  ->i_mmap_rwsem		(truncate_pagecache)
80  *    ->private_lock		(__free_pte->block_dirty_folio)
81  *      ->swap_lock		(exclusive_swap_page, others)
82  *        ->i_pages lock
83  *
84  *  ->i_rwsem
85  *    ->invalidate_lock		(acquired by fs in truncate path)
86  *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
87  *
88  *  ->mmap_lock
89  *    ->i_mmap_rwsem
90  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
91  *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
92  *
93  *  ->mmap_lock
94  *    ->invalidate_lock		(filemap_fault)
95  *      ->lock_page		(filemap_fault, access_process_vm)
96  *
97  *  ->i_rwsem			(generic_perform_write)
98  *    ->mmap_lock		(fault_in_readable->do_page_fault)
99  *
100  *  bdi->wb.list_lock
101  *    sb_lock			(fs/fs-writeback.c)
102  *    ->i_pages lock		(__sync_single_inode)
103  *
104  *  ->i_mmap_rwsem
105  *    ->anon_vma.lock		(vma_merge)
106  *
107  *  ->anon_vma.lock
108  *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
109  *
110  *  ->page_table_lock or pte_lock
111  *    ->swap_lock		(try_to_unmap_one)
112  *    ->private_lock		(try_to_unmap_one)
113  *    ->i_pages lock		(try_to_unmap_one)
114  *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
115  *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
116  *    ->private_lock		(page_remove_rmap->set_page_dirty)
117  *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
118  *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
119  *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
120  *    ->memcg->move_lock	(page_remove_rmap->folio_memcg_lock)
121  *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
122  *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
123  *    ->private_lock		(zap_pte_range->block_dirty_folio)
124  *
125  * ->i_mmap_rwsem
126  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
127  */
128 
129 static void page_cache_delete(struct address_space *mapping,
130 				   struct folio *folio, void *shadow)
131 {
132 	XA_STATE(xas, &mapping->i_pages, folio->index);
133 	long nr = 1;
134 
135 	mapping_set_update(&xas, mapping);
136 
137 	/* hugetlb pages are represented by a single entry in the xarray */
138 	if (!folio_test_hugetlb(folio)) {
139 		xas_set_order(&xas, folio->index, folio_order(folio));
140 		nr = folio_nr_pages(folio);
141 	}
142 
143 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
144 
145 	xas_store(&xas, shadow);
146 	xas_init_marks(&xas);
147 
148 	folio->mapping = NULL;
149 	/* Leave page->index set: truncation lookup relies upon it */
150 	mapping->nrpages -= nr;
151 }
152 
153 static void filemap_unaccount_folio(struct address_space *mapping,
154 		struct folio *folio)
155 {
156 	long nr;
157 
158 	VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
159 	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
160 		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
161 			 current->comm, folio_pfn(folio));
162 		dump_page(&folio->page, "still mapped when deleted");
163 		dump_stack();
164 		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
165 
166 		if (mapping_exiting(mapping) && !folio_test_large(folio)) {
167 			int mapcount = page_mapcount(&folio->page);
168 
169 			if (folio_ref_count(folio) >= mapcount + 2) {
170 				/*
171 				 * All vmas have already been torn down, so it's
172 				 * a good bet that actually the page is unmapped
173 				 * and we'd rather not leak it: if we're wrong,
174 				 * another bad page check should catch it later.
175 				 */
176 				page_mapcount_reset(&folio->page);
177 				folio_ref_sub(folio, mapcount);
178 			}
179 		}
180 	}
181 
182 	/* hugetlb folios do not participate in page cache accounting. */
183 	if (folio_test_hugetlb(folio))
184 		return;
185 
186 	nr = folio_nr_pages(folio);
187 
188 	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
189 	if (folio_test_swapbacked(folio)) {
190 		__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
191 		if (folio_test_pmd_mappable(folio))
192 			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
193 	} else if (folio_test_pmd_mappable(folio)) {
194 		__lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
195 		filemap_nr_thps_dec(mapping);
196 	}
197 
198 	/*
199 	 * At this point folio must be either written or cleaned by
200 	 * truncate.  Dirty folio here signals a bug and loss of
201 	 * unwritten data - on ordinary filesystems.
202 	 *
203 	 * But it's harmless on in-memory filesystems like tmpfs; and can
204 	 * occur when a driver which did get_user_pages() sets page dirty
205 	 * before putting it, while the inode is being finally evicted.
206 	 *
207 	 * Below fixes dirty accounting after removing the folio entirely
208 	 * but leaves the dirty flag set: it has no effect for truncated
209 	 * folio and anyway will be cleared before returning folio to
210 	 * buddy allocator.
211 	 */
212 	if (WARN_ON_ONCE(folio_test_dirty(folio) &&
213 			 mapping_can_writeback(mapping)))
214 		folio_account_cleaned(folio, inode_to_wb(mapping->host));
215 }
216 
217 /*
218  * Delete a page from the page cache and free it. Caller has to make
219  * sure the page is locked and that nobody else uses it - or that usage
220  * is safe.  The caller must hold the i_pages lock.
221  */
222 void __filemap_remove_folio(struct folio *folio, void *shadow)
223 {
224 	struct address_space *mapping = folio->mapping;
225 
226 	trace_mm_filemap_delete_from_page_cache(folio);
227 	filemap_unaccount_folio(mapping, folio);
228 	page_cache_delete(mapping, folio, shadow);
229 }
230 
231 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
232 {
233 	void (*free_folio)(struct folio *);
234 	int refs = 1;
235 
236 	free_folio = mapping->a_ops->free_folio;
237 	if (free_folio)
238 		free_folio(folio);
239 
240 	if (folio_test_large(folio) && !folio_test_hugetlb(folio))
241 		refs = folio_nr_pages(folio);
242 	folio_put_refs(folio, refs);
243 }
244 
245 /**
246  * filemap_remove_folio - Remove folio from page cache.
247  * @folio: The folio.
248  *
249  * This must be called only on folios that are locked and have been
250  * verified to be in the page cache.  It will never put the folio into
251  * the free list because the caller has a reference on the page.
252  */
253 void filemap_remove_folio(struct folio *folio)
254 {
255 	struct address_space *mapping = folio->mapping;
256 
257 	BUG_ON(!folio_test_locked(folio));
258 	spin_lock(&mapping->host->i_lock);
259 	xa_lock_irq(&mapping->i_pages);
260 	__filemap_remove_folio(folio, NULL);
261 	xa_unlock_irq(&mapping->i_pages);
262 	if (mapping_shrinkable(mapping))
263 		inode_add_lru(mapping->host);
264 	spin_unlock(&mapping->host->i_lock);
265 
266 	filemap_free_folio(mapping, folio);
267 }
268 
269 /*
270  * page_cache_delete_batch - delete several folios from page cache
271  * @mapping: the mapping to which folios belong
272  * @fbatch: batch of folios to delete
273  *
274  * The function walks over mapping->i_pages and removes folios passed in
275  * @fbatch from the mapping. The function expects @fbatch to be sorted
276  * by page index and is optimised for it to be dense.
277  * It tolerates holes in @fbatch (mapping entries at those indices are not
278  * modified).
279  *
280  * The function expects the i_pages lock to be held.
281  */
282 static void page_cache_delete_batch(struct address_space *mapping,
283 			     struct folio_batch *fbatch)
284 {
285 	XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
286 	long total_pages = 0;
287 	int i = 0;
288 	struct folio *folio;
289 
290 	mapping_set_update(&xas, mapping);
291 	xas_for_each(&xas, folio, ULONG_MAX) {
292 		if (i >= folio_batch_count(fbatch))
293 			break;
294 
295 		/* A swap/dax/shadow entry got inserted? Skip it. */
296 		if (xa_is_value(folio))
297 			continue;
298 		/*
299 		 * A page got inserted in our range? Skip it. We have our
300 		 * pages locked so they are protected from being removed.
301 		 * If we see a page whose index is higher than ours, it
302 		 * means our page has been removed, which shouldn't be
303 		 * possible because we're holding the PageLock.
304 		 */
305 		if (folio != fbatch->folios[i]) {
306 			VM_BUG_ON_FOLIO(folio->index >
307 					fbatch->folios[i]->index, folio);
308 			continue;
309 		}
310 
311 		WARN_ON_ONCE(!folio_test_locked(folio));
312 
313 		folio->mapping = NULL;
314 		/* Leave folio->index set: truncation lookup relies on it */
315 
316 		i++;
317 		xas_store(&xas, NULL);
318 		total_pages += folio_nr_pages(folio);
319 	}
320 	mapping->nrpages -= total_pages;
321 }
322 
323 void delete_from_page_cache_batch(struct address_space *mapping,
324 				  struct folio_batch *fbatch)
325 {
326 	int i;
327 
328 	if (!folio_batch_count(fbatch))
329 		return;
330 
331 	spin_lock(&mapping->host->i_lock);
332 	xa_lock_irq(&mapping->i_pages);
333 	for (i = 0; i < folio_batch_count(fbatch); i++) {
334 		struct folio *folio = fbatch->folios[i];
335 
336 		trace_mm_filemap_delete_from_page_cache(folio);
337 		filemap_unaccount_folio(mapping, folio);
338 	}
339 	page_cache_delete_batch(mapping, fbatch);
340 	xa_unlock_irq(&mapping->i_pages);
341 	if (mapping_shrinkable(mapping))
342 		inode_add_lru(mapping->host);
343 	spin_unlock(&mapping->host->i_lock);
344 
345 	for (i = 0; i < folio_batch_count(fbatch); i++)
346 		filemap_free_folio(mapping, fbatch->folios[i]);
347 }
348 
349 int filemap_check_errors(struct address_space *mapping)
350 {
351 	int ret = 0;
352 	/* Check for outstanding write errors */
353 	if (test_bit(AS_ENOSPC, &mapping->flags) &&
354 	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
355 		ret = -ENOSPC;
356 	if (test_bit(AS_EIO, &mapping->flags) &&
357 	    test_and_clear_bit(AS_EIO, &mapping->flags))
358 		ret = -EIO;
359 	return ret;
360 }
361 EXPORT_SYMBOL(filemap_check_errors);
362 
363 static int filemap_check_and_keep_errors(struct address_space *mapping)
364 {
365 	/* Check for outstanding write errors */
366 	if (test_bit(AS_EIO, &mapping->flags))
367 		return -EIO;
368 	if (test_bit(AS_ENOSPC, &mapping->flags))
369 		return -ENOSPC;
370 	return 0;
371 }
372 
373 /**
374  * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
375  * @mapping:	address space structure to write
376  * @wbc:	the writeback_control controlling the writeout
377  *
378  * Call writepages on the mapping using the provided wbc to control the
379  * writeout.
380  *
381  * Return: %0 on success, negative error code otherwise.
382  */
383 int filemap_fdatawrite_wbc(struct address_space *mapping,
384 			   struct writeback_control *wbc)
385 {
386 	int ret;
387 
388 	if (!mapping_can_writeback(mapping) ||
389 	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
390 		return 0;
391 
392 	wbc_attach_fdatawrite_inode(wbc, mapping->host);
393 	ret = do_writepages(mapping, wbc);
394 	wbc_detach_inode(wbc);
395 	return ret;
396 }
397 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
398 
399 /**
400  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
401  * @mapping:	address space structure to write
402  * @start:	offset in bytes where the range starts
403  * @end:	offset in bytes where the range ends (inclusive)
404  * @sync_mode:	enable synchronous operation
405  *
406  * Start writeback against all of a mapping's dirty pages that lie
407  * within the byte offsets <start, end> inclusive.
408  *
409  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
410  * opposed to a regular memory cleansing writeback.  The difference between
411  * these two operations is that if a dirty page/buffer is encountered, it must
412  * be waited upon, and not just skipped over.
413  *
414  * Return: %0 on success, negative error code otherwise.
415  */
416 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
417 				loff_t end, int sync_mode)
418 {
419 	struct writeback_control wbc = {
420 		.sync_mode = sync_mode,
421 		.nr_to_write = LONG_MAX,
422 		.range_start = start,
423 		.range_end = end,
424 	};
425 
426 	return filemap_fdatawrite_wbc(mapping, &wbc);
427 }
428 
429 static inline int __filemap_fdatawrite(struct address_space *mapping,
430 	int sync_mode)
431 {
432 	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
433 }
434 
435 int filemap_fdatawrite(struct address_space *mapping)
436 {
437 	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
438 }
439 EXPORT_SYMBOL(filemap_fdatawrite);
440 
441 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
442 				loff_t end)
443 {
444 	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
445 }
446 EXPORT_SYMBOL(filemap_fdatawrite_range);
447 
448 /**
449  * filemap_flush - mostly a non-blocking flush
450  * @mapping:	target address_space
451  *
452  * This is a mostly non-blocking flush.  Not suitable for data-integrity
453  * purposes - I/O may not be started against all dirty pages.
454  *
455  * Return: %0 on success, negative error code otherwise.
456  */
457 int filemap_flush(struct address_space *mapping)
458 {
459 	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
460 }
461 EXPORT_SYMBOL(filemap_flush);
462 
463 /**
464  * filemap_range_has_page - check if a page exists in range.
465  * @mapping:           address space within which to check
466  * @start_byte:        offset in bytes where the range starts
467  * @end_byte:          offset in bytes where the range ends (inclusive)
468  *
469  * Find at least one page in the range supplied, usually used to check if
470  * direct writing in this range will trigger a writeback.
471  *
472  * Return: %true if at least one page exists in the specified range,
473  * %false otherwise.
474  */
475 bool filemap_range_has_page(struct address_space *mapping,
476 			   loff_t start_byte, loff_t end_byte)
477 {
478 	struct folio *folio;
479 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
480 	pgoff_t max = end_byte >> PAGE_SHIFT;
481 
482 	if (end_byte < start_byte)
483 		return false;
484 
485 	rcu_read_lock();
486 	for (;;) {
487 		folio = xas_find(&xas, max);
488 		if (xas_retry(&xas, folio))
489 			continue;
490 		/* Shadow entries don't count */
491 		if (xa_is_value(folio))
492 			continue;
493 		/*
494 		 * We don't need to try to pin this page; we're about to
495 		 * release the RCU lock anyway.  It is enough to know that
496 		 * there was a page here recently.
497 		 */
498 		break;
499 	}
500 	rcu_read_unlock();
501 
502 	return folio != NULL;
503 }
504 EXPORT_SYMBOL(filemap_range_has_page);
505 
506 static void __filemap_fdatawait_range(struct address_space *mapping,
507 				     loff_t start_byte, loff_t end_byte)
508 {
509 	pgoff_t index = start_byte >> PAGE_SHIFT;
510 	pgoff_t end = end_byte >> PAGE_SHIFT;
511 	struct folio_batch fbatch;
512 	unsigned nr_folios;
513 
514 	folio_batch_init(&fbatch);
515 
516 	while (index <= end) {
517 		unsigned i;
518 
519 		nr_folios = filemap_get_folios_tag(mapping, &index, end,
520 				PAGECACHE_TAG_WRITEBACK, &fbatch);
521 
522 		if (!nr_folios)
523 			break;
524 
525 		for (i = 0; i < nr_folios; i++) {
526 			struct folio *folio = fbatch.folios[i];
527 
528 			folio_wait_writeback(folio);
529 			folio_clear_error(folio);
530 		}
531 		folio_batch_release(&fbatch);
532 		cond_resched();
533 	}
534 }
535 
536 /**
537  * filemap_fdatawait_range - wait for writeback to complete
538  * @mapping:		address space structure to wait for
539  * @start_byte:		offset in bytes where the range starts
540  * @end_byte:		offset in bytes where the range ends (inclusive)
541  *
542  * Walk the list of under-writeback pages of the given address space
543  * in the given range and wait for all of them.  Check error status of
544  * the address space and return it.
545  *
546  * Since the error status of the address space is cleared by this function,
547  * callers are responsible for checking the return value and handling and/or
548  * reporting the error.
549  *
550  * Return: error status of the address space.
551  */
552 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
553 			    loff_t end_byte)
554 {
555 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
556 	return filemap_check_errors(mapping);
557 }
558 EXPORT_SYMBOL(filemap_fdatawait_range);
559 
560 /**
561  * filemap_fdatawait_range_keep_errors - wait for writeback to complete
562  * @mapping:		address space structure to wait for
563  * @start_byte:		offset in bytes where the range starts
564  * @end_byte:		offset in bytes where the range ends (inclusive)
565  *
566  * Walk the list of under-writeback pages of the given address space in the
567  * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
568  * this function does not clear error status of the address space.
569  *
570  * Use this function if callers don't handle errors themselves.  Expected
571  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
572  * fsfreeze(8)
573  */
574 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
575 		loff_t start_byte, loff_t end_byte)
576 {
577 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
578 	return filemap_check_and_keep_errors(mapping);
579 }
580 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
581 
582 /**
583  * file_fdatawait_range - wait for writeback to complete
584  * @file:		file pointing to address space structure to wait for
585  * @start_byte:		offset in bytes where the range starts
586  * @end_byte:		offset in bytes where the range ends (inclusive)
587  *
588  * Walk the list of under-writeback pages of the address space that file
589  * refers to, in the given range and wait for all of them.  Check error
590  * status of the address space vs. the file->f_wb_err cursor and return it.
591  *
592  * Since the error status of the file is advanced by this function,
593  * callers are responsible for checking the return value and handling and/or
594  * reporting the error.
595  *
596  * Return: error status of the address space vs. the file->f_wb_err cursor.
597  */
598 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
599 {
600 	struct address_space *mapping = file->f_mapping;
601 
602 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
603 	return file_check_and_advance_wb_err(file);
604 }
605 EXPORT_SYMBOL(file_fdatawait_range);
606 
607 /**
608  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
609  * @mapping: address space structure to wait for
610  *
611  * Walk the list of under-writeback pages of the given address space
612  * and wait for all of them.  Unlike filemap_fdatawait(), this function
613  * does not clear error status of the address space.
614  *
615  * Use this function if callers don't handle errors themselves.  Expected
616  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
617  * fsfreeze(8)
618  *
619  * Return: error status of the address space.
620  */
621 int filemap_fdatawait_keep_errors(struct address_space *mapping)
622 {
623 	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
624 	return filemap_check_and_keep_errors(mapping);
625 }
626 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
627 
628 /* Returns true if writeback might be needed or already in progress. */
629 static bool mapping_needs_writeback(struct address_space *mapping)
630 {
631 	return mapping->nrpages;
632 }
633 
634 bool filemap_range_has_writeback(struct address_space *mapping,
635 				 loff_t start_byte, loff_t end_byte)
636 {
637 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
638 	pgoff_t max = end_byte >> PAGE_SHIFT;
639 	struct folio *folio;
640 
641 	if (end_byte < start_byte)
642 		return false;
643 
644 	rcu_read_lock();
645 	xas_for_each(&xas, folio, max) {
646 		if (xas_retry(&xas, folio))
647 			continue;
648 		if (xa_is_value(folio))
649 			continue;
650 		if (folio_test_dirty(folio) || folio_test_locked(folio) ||
651 				folio_test_writeback(folio))
652 			break;
653 	}
654 	rcu_read_unlock();
655 	return folio != NULL;
656 }
657 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
658 
659 /**
660  * filemap_write_and_wait_range - write out & wait on a file range
661  * @mapping:	the address_space for the pages
662  * @lstart:	offset in bytes where the range starts
663  * @lend:	offset in bytes where the range ends (inclusive)
664  *
665  * Write out and wait upon file offsets lstart->lend, inclusive.
666  *
667  * Note that @lend is inclusive (describes the last byte to be written) so
668  * that this function can be used to write to the very end-of-file (end = -1).
669  *
670  * Return: error status of the address space.
671  */
672 int filemap_write_and_wait_range(struct address_space *mapping,
673 				 loff_t lstart, loff_t lend)
674 {
675 	int err = 0, err2;
676 
677 	if (lend < lstart)
678 		return 0;
679 
680 	if (mapping_needs_writeback(mapping)) {
681 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
682 						 WB_SYNC_ALL);
683 		/*
684 		 * Even if the above returned error, the pages may be
685 		 * written partially (e.g. -ENOSPC), so we wait for it.
686 		 * But the -EIO is special case, it may indicate the worst
687 		 * thing (e.g. bug) happened, so we avoid waiting for it.
688 		 */
689 		if (err != -EIO)
690 			__filemap_fdatawait_range(mapping, lstart, lend);
691 	}
692 	err2 = filemap_check_errors(mapping);
693 	if (!err)
694 		err = err2;
695 	return err;
696 }
697 EXPORT_SYMBOL(filemap_write_and_wait_range);
698 
699 void __filemap_set_wb_err(struct address_space *mapping, int err)
700 {
701 	errseq_t eseq = errseq_set(&mapping->wb_err, err);
702 
703 	trace_filemap_set_wb_err(mapping, eseq);
704 }
705 EXPORT_SYMBOL(__filemap_set_wb_err);
706 
707 /**
708  * file_check_and_advance_wb_err - report wb error (if any) that was previously
709  * 				   and advance wb_err to current one
710  * @file: struct file on which the error is being reported
711  *
712  * When userland calls fsync (or something like nfsd does the equivalent), we
713  * want to report any writeback errors that occurred since the last fsync (or
714  * since the file was opened if there haven't been any).
715  *
716  * Grab the wb_err from the mapping. If it matches what we have in the file,
717  * then just quickly return 0. The file is all caught up.
718  *
719  * If it doesn't match, then take the mapping value, set the "seen" flag in
720  * it and try to swap it into place. If it works, or another task beat us
721  * to it with the new value, then update the f_wb_err and return the error
722  * portion. The error at this point must be reported via proper channels
723  * (a'la fsync, or NFS COMMIT operation, etc.).
724  *
725  * While we handle mapping->wb_err with atomic operations, the f_wb_err
726  * value is protected by the f_lock since we must ensure that it reflects
727  * the latest value swapped in for this file descriptor.
728  *
729  * Return: %0 on success, negative error code otherwise.
730  */
731 int file_check_and_advance_wb_err(struct file *file)
732 {
733 	int err = 0;
734 	errseq_t old = READ_ONCE(file->f_wb_err);
735 	struct address_space *mapping = file->f_mapping;
736 
737 	/* Locklessly handle the common case where nothing has changed */
738 	if (errseq_check(&mapping->wb_err, old)) {
739 		/* Something changed, must use slow path */
740 		spin_lock(&file->f_lock);
741 		old = file->f_wb_err;
742 		err = errseq_check_and_advance(&mapping->wb_err,
743 						&file->f_wb_err);
744 		trace_file_check_and_advance_wb_err(file, old);
745 		spin_unlock(&file->f_lock);
746 	}
747 
748 	/*
749 	 * We're mostly using this function as a drop in replacement for
750 	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
751 	 * that the legacy code would have had on these flags.
752 	 */
753 	clear_bit(AS_EIO, &mapping->flags);
754 	clear_bit(AS_ENOSPC, &mapping->flags);
755 	return err;
756 }
757 EXPORT_SYMBOL(file_check_and_advance_wb_err);
758 
759 /**
760  * file_write_and_wait_range - write out & wait on a file range
761  * @file:	file pointing to address_space with pages
762  * @lstart:	offset in bytes where the range starts
763  * @lend:	offset in bytes where the range ends (inclusive)
764  *
765  * Write out and wait upon file offsets lstart->lend, inclusive.
766  *
767  * Note that @lend is inclusive (describes the last byte to be written) so
768  * that this function can be used to write to the very end-of-file (end = -1).
769  *
770  * After writing out and waiting on the data, we check and advance the
771  * f_wb_err cursor to the latest value, and return any errors detected there.
772  *
773  * Return: %0 on success, negative error code otherwise.
774  */
775 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
776 {
777 	int err = 0, err2;
778 	struct address_space *mapping = file->f_mapping;
779 
780 	if (lend < lstart)
781 		return 0;
782 
783 	if (mapping_needs_writeback(mapping)) {
784 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
785 						 WB_SYNC_ALL);
786 		/* See comment of filemap_write_and_wait() */
787 		if (err != -EIO)
788 			__filemap_fdatawait_range(mapping, lstart, lend);
789 	}
790 	err2 = file_check_and_advance_wb_err(file);
791 	if (!err)
792 		err = err2;
793 	return err;
794 }
795 EXPORT_SYMBOL(file_write_and_wait_range);
796 
797 /**
798  * replace_page_cache_folio - replace a pagecache folio with a new one
799  * @old:	folio to be replaced
800  * @new:	folio to replace with
801  *
802  * This function replaces a folio in the pagecache with a new one.  On
803  * success it acquires the pagecache reference for the new folio and
804  * drops it for the old folio.  Both the old and new folios must be
805  * locked.  This function does not add the new folio to the LRU, the
806  * caller must do that.
807  *
808  * The remove + add is atomic.  This function cannot fail.
809  */
810 void replace_page_cache_folio(struct folio *old, struct folio *new)
811 {
812 	struct address_space *mapping = old->mapping;
813 	void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
814 	pgoff_t offset = old->index;
815 	XA_STATE(xas, &mapping->i_pages, offset);
816 
817 	VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
818 	VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
819 	VM_BUG_ON_FOLIO(new->mapping, new);
820 
821 	folio_get(new);
822 	new->mapping = mapping;
823 	new->index = offset;
824 
825 	mem_cgroup_migrate(old, new);
826 
827 	xas_lock_irq(&xas);
828 	xas_store(&xas, new);
829 
830 	old->mapping = NULL;
831 	/* hugetlb pages do not participate in page cache accounting. */
832 	if (!folio_test_hugetlb(old))
833 		__lruvec_stat_sub_folio(old, NR_FILE_PAGES);
834 	if (!folio_test_hugetlb(new))
835 		__lruvec_stat_add_folio(new, NR_FILE_PAGES);
836 	if (folio_test_swapbacked(old))
837 		__lruvec_stat_sub_folio(old, NR_SHMEM);
838 	if (folio_test_swapbacked(new))
839 		__lruvec_stat_add_folio(new, NR_SHMEM);
840 	xas_unlock_irq(&xas);
841 	if (free_folio)
842 		free_folio(old);
843 	folio_put(old);
844 }
845 EXPORT_SYMBOL_GPL(replace_page_cache_folio);
846 
847 noinline int __filemap_add_folio(struct address_space *mapping,
848 		struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
849 {
850 	XA_STATE(xas, &mapping->i_pages, index);
851 	int huge = folio_test_hugetlb(folio);
852 	bool charged = false;
853 	long nr = 1;
854 
855 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
856 	VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
857 	mapping_set_update(&xas, mapping);
858 
859 	if (!huge) {
860 		int error = mem_cgroup_charge(folio, NULL, gfp);
861 		VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
862 		if (error)
863 			return error;
864 		charged = true;
865 		xas_set_order(&xas, index, folio_order(folio));
866 		nr = folio_nr_pages(folio);
867 	}
868 
869 	gfp &= GFP_RECLAIM_MASK;
870 	folio_ref_add(folio, nr);
871 	folio->mapping = mapping;
872 	folio->index = xas.xa_index;
873 
874 	do {
875 		unsigned int order = xa_get_order(xas.xa, xas.xa_index);
876 		void *entry, *old = NULL;
877 
878 		if (order > folio_order(folio))
879 			xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
880 					order, gfp);
881 		xas_lock_irq(&xas);
882 		xas_for_each_conflict(&xas, entry) {
883 			old = entry;
884 			if (!xa_is_value(entry)) {
885 				xas_set_err(&xas, -EEXIST);
886 				goto unlock;
887 			}
888 		}
889 
890 		if (old) {
891 			if (shadowp)
892 				*shadowp = old;
893 			/* entry may have been split before we acquired lock */
894 			order = xa_get_order(xas.xa, xas.xa_index);
895 			if (order > folio_order(folio)) {
896 				/* How to handle large swap entries? */
897 				BUG_ON(shmem_mapping(mapping));
898 				xas_split(&xas, old, order);
899 				xas_reset(&xas);
900 			}
901 		}
902 
903 		xas_store(&xas, folio);
904 		if (xas_error(&xas))
905 			goto unlock;
906 
907 		mapping->nrpages += nr;
908 
909 		/* hugetlb pages do not participate in page cache accounting */
910 		if (!huge) {
911 			__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
912 			if (folio_test_pmd_mappable(folio))
913 				__lruvec_stat_mod_folio(folio,
914 						NR_FILE_THPS, nr);
915 		}
916 unlock:
917 		xas_unlock_irq(&xas);
918 	} while (xas_nomem(&xas, gfp));
919 
920 	if (xas_error(&xas))
921 		goto error;
922 
923 	trace_mm_filemap_add_to_page_cache(folio);
924 	return 0;
925 error:
926 	if (charged)
927 		mem_cgroup_uncharge(folio);
928 	folio->mapping = NULL;
929 	/* Leave page->index set: truncation relies upon it */
930 	folio_put_refs(folio, nr);
931 	return xas_error(&xas);
932 }
933 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
934 
935 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
936 				pgoff_t index, gfp_t gfp)
937 {
938 	void *shadow = NULL;
939 	int ret;
940 
941 	__folio_set_locked(folio);
942 	ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
943 	if (unlikely(ret))
944 		__folio_clear_locked(folio);
945 	else {
946 		/*
947 		 * The folio might have been evicted from cache only
948 		 * recently, in which case it should be activated like
949 		 * any other repeatedly accessed folio.
950 		 * The exception is folios getting rewritten; evicting other
951 		 * data from the working set, only to cache data that will
952 		 * get overwritten with something else, is a waste of memory.
953 		 */
954 		WARN_ON_ONCE(folio_test_active(folio));
955 		if (!(gfp & __GFP_WRITE) && shadow)
956 			workingset_refault(folio, shadow);
957 		folio_add_lru(folio);
958 	}
959 	return ret;
960 }
961 EXPORT_SYMBOL_GPL(filemap_add_folio);
962 
963 #ifdef CONFIG_NUMA
964 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
965 {
966 	int n;
967 	struct folio *folio;
968 
969 	if (cpuset_do_page_mem_spread()) {
970 		unsigned int cpuset_mems_cookie;
971 		do {
972 			cpuset_mems_cookie = read_mems_allowed_begin();
973 			n = cpuset_mem_spread_node();
974 			folio = __folio_alloc_node(gfp, order, n);
975 		} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
976 
977 		return folio;
978 	}
979 	return folio_alloc(gfp, order);
980 }
981 EXPORT_SYMBOL(filemap_alloc_folio);
982 #endif
983 
984 /*
985  * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
986  *
987  * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
988  *
989  * @mapping1: the first mapping to lock
990  * @mapping2: the second mapping to lock
991  */
992 void filemap_invalidate_lock_two(struct address_space *mapping1,
993 				 struct address_space *mapping2)
994 {
995 	if (mapping1 > mapping2)
996 		swap(mapping1, mapping2);
997 	if (mapping1)
998 		down_write(&mapping1->invalidate_lock);
999 	if (mapping2 && mapping1 != mapping2)
1000 		down_write_nested(&mapping2->invalidate_lock, 1);
1001 }
1002 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1003 
1004 /*
1005  * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1006  *
1007  * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1008  *
1009  * @mapping1: the first mapping to unlock
1010  * @mapping2: the second mapping to unlock
1011  */
1012 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1013 				   struct address_space *mapping2)
1014 {
1015 	if (mapping1)
1016 		up_write(&mapping1->invalidate_lock);
1017 	if (mapping2 && mapping1 != mapping2)
1018 		up_write(&mapping2->invalidate_lock);
1019 }
1020 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1021 
1022 /*
1023  * In order to wait for pages to become available there must be
1024  * waitqueues associated with pages. By using a hash table of
1025  * waitqueues where the bucket discipline is to maintain all
1026  * waiters on the same queue and wake all when any of the pages
1027  * become available, and for the woken contexts to check to be
1028  * sure the appropriate page became available, this saves space
1029  * at a cost of "thundering herd" phenomena during rare hash
1030  * collisions.
1031  */
1032 #define PAGE_WAIT_TABLE_BITS 8
1033 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1034 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1035 
1036 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1037 {
1038 	return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1039 }
1040 
1041 void __init pagecache_init(void)
1042 {
1043 	int i;
1044 
1045 	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1046 		init_waitqueue_head(&folio_wait_table[i]);
1047 
1048 	page_writeback_init();
1049 }
1050 
1051 /*
1052  * The page wait code treats the "wait->flags" somewhat unusually, because
1053  * we have multiple different kinds of waits, not just the usual "exclusive"
1054  * one.
1055  *
1056  * We have:
1057  *
1058  *  (a) no special bits set:
1059  *
1060  *	We're just waiting for the bit to be released, and when a waker
1061  *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1062  *	and remove it from the wait queue.
1063  *
1064  *	Simple and straightforward.
1065  *
1066  *  (b) WQ_FLAG_EXCLUSIVE:
1067  *
1068  *	The waiter is waiting to get the lock, and only one waiter should
1069  *	be woken up to avoid any thundering herd behavior. We'll set the
1070  *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1071  *
1072  *	This is the traditional exclusive wait.
1073  *
1074  *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1075  *
1076  *	The waiter is waiting to get the bit, and additionally wants the
1077  *	lock to be transferred to it for fair lock behavior. If the lock
1078  *	cannot be taken, we stop walking the wait queue without waking
1079  *	the waiter.
1080  *
1081  *	This is the "fair lock handoff" case, and in addition to setting
1082  *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1083  *	that it now has the lock.
1084  */
1085 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1086 {
1087 	unsigned int flags;
1088 	struct wait_page_key *key = arg;
1089 	struct wait_page_queue *wait_page
1090 		= container_of(wait, struct wait_page_queue, wait);
1091 
1092 	if (!wake_page_match(wait_page, key))
1093 		return 0;
1094 
1095 	/*
1096 	 * If it's a lock handoff wait, we get the bit for it, and
1097 	 * stop walking (and do not wake it up) if we can't.
1098 	 */
1099 	flags = wait->flags;
1100 	if (flags & WQ_FLAG_EXCLUSIVE) {
1101 		if (test_bit(key->bit_nr, &key->folio->flags))
1102 			return -1;
1103 		if (flags & WQ_FLAG_CUSTOM) {
1104 			if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1105 				return -1;
1106 			flags |= WQ_FLAG_DONE;
1107 		}
1108 	}
1109 
1110 	/*
1111 	 * We are holding the wait-queue lock, but the waiter that
1112 	 * is waiting for this will be checking the flags without
1113 	 * any locking.
1114 	 *
1115 	 * So update the flags atomically, and wake up the waiter
1116 	 * afterwards to avoid any races. This store-release pairs
1117 	 * with the load-acquire in folio_wait_bit_common().
1118 	 */
1119 	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1120 	wake_up_state(wait->private, mode);
1121 
1122 	/*
1123 	 * Ok, we have successfully done what we're waiting for,
1124 	 * and we can unconditionally remove the wait entry.
1125 	 *
1126 	 * Note that this pairs with the "finish_wait()" in the
1127 	 * waiter, and has to be the absolute last thing we do.
1128 	 * After this list_del_init(&wait->entry) the wait entry
1129 	 * might be de-allocated and the process might even have
1130 	 * exited.
1131 	 */
1132 	list_del_init_careful(&wait->entry);
1133 	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1134 }
1135 
1136 static void folio_wake_bit(struct folio *folio, int bit_nr)
1137 {
1138 	wait_queue_head_t *q = folio_waitqueue(folio);
1139 	struct wait_page_key key;
1140 	unsigned long flags;
1141 	wait_queue_entry_t bookmark;
1142 
1143 	key.folio = folio;
1144 	key.bit_nr = bit_nr;
1145 	key.page_match = 0;
1146 
1147 	bookmark.flags = 0;
1148 	bookmark.private = NULL;
1149 	bookmark.func = NULL;
1150 	INIT_LIST_HEAD(&bookmark.entry);
1151 
1152 	spin_lock_irqsave(&q->lock, flags);
1153 	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1154 
1155 	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1156 		/*
1157 		 * Take a breather from holding the lock,
1158 		 * allow pages that finish wake up asynchronously
1159 		 * to acquire the lock and remove themselves
1160 		 * from wait queue
1161 		 */
1162 		spin_unlock_irqrestore(&q->lock, flags);
1163 		cpu_relax();
1164 		spin_lock_irqsave(&q->lock, flags);
1165 		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1166 	}
1167 
1168 	/*
1169 	 * It's possible to miss clearing waiters here, when we woke our page
1170 	 * waiters, but the hashed waitqueue has waiters for other pages on it.
1171 	 * That's okay, it's a rare case. The next waker will clear it.
1172 	 *
1173 	 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1174 	 * other), the flag may be cleared in the course of freeing the page;
1175 	 * but that is not required for correctness.
1176 	 */
1177 	if (!waitqueue_active(q) || !key.page_match)
1178 		folio_clear_waiters(folio);
1179 
1180 	spin_unlock_irqrestore(&q->lock, flags);
1181 }
1182 
1183 static void folio_wake(struct folio *folio, int bit)
1184 {
1185 	if (!folio_test_waiters(folio))
1186 		return;
1187 	folio_wake_bit(folio, bit);
1188 }
1189 
1190 /*
1191  * A choice of three behaviors for folio_wait_bit_common():
1192  */
1193 enum behavior {
1194 	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1195 			 * __folio_lock() waiting on then setting PG_locked.
1196 			 */
1197 	SHARED,		/* Hold ref to page and check the bit when woken, like
1198 			 * folio_wait_writeback() waiting on PG_writeback.
1199 			 */
1200 	DROP,		/* Drop ref to page before wait, no check when woken,
1201 			 * like folio_put_wait_locked() on PG_locked.
1202 			 */
1203 };
1204 
1205 /*
1206  * Attempt to check (or get) the folio flag, and mark us done
1207  * if successful.
1208  */
1209 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1210 					struct wait_queue_entry *wait)
1211 {
1212 	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1213 		if (test_and_set_bit(bit_nr, &folio->flags))
1214 			return false;
1215 	} else if (test_bit(bit_nr, &folio->flags))
1216 		return false;
1217 
1218 	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1219 	return true;
1220 }
1221 
1222 /* How many times do we accept lock stealing from under a waiter? */
1223 int sysctl_page_lock_unfairness = 5;
1224 
1225 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1226 		int state, enum behavior behavior)
1227 {
1228 	wait_queue_head_t *q = folio_waitqueue(folio);
1229 	int unfairness = sysctl_page_lock_unfairness;
1230 	struct wait_page_queue wait_page;
1231 	wait_queue_entry_t *wait = &wait_page.wait;
1232 	bool thrashing = false;
1233 	unsigned long pflags;
1234 	bool in_thrashing;
1235 
1236 	if (bit_nr == PG_locked &&
1237 	    !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1238 		delayacct_thrashing_start(&in_thrashing);
1239 		psi_memstall_enter(&pflags);
1240 		thrashing = true;
1241 	}
1242 
1243 	init_wait(wait);
1244 	wait->func = wake_page_function;
1245 	wait_page.folio = folio;
1246 	wait_page.bit_nr = bit_nr;
1247 
1248 repeat:
1249 	wait->flags = 0;
1250 	if (behavior == EXCLUSIVE) {
1251 		wait->flags = WQ_FLAG_EXCLUSIVE;
1252 		if (--unfairness < 0)
1253 			wait->flags |= WQ_FLAG_CUSTOM;
1254 	}
1255 
1256 	/*
1257 	 * Do one last check whether we can get the
1258 	 * page bit synchronously.
1259 	 *
1260 	 * Do the folio_set_waiters() marking before that
1261 	 * to let any waker we _just_ missed know they
1262 	 * need to wake us up (otherwise they'll never
1263 	 * even go to the slow case that looks at the
1264 	 * page queue), and add ourselves to the wait
1265 	 * queue if we need to sleep.
1266 	 *
1267 	 * This part needs to be done under the queue
1268 	 * lock to avoid races.
1269 	 */
1270 	spin_lock_irq(&q->lock);
1271 	folio_set_waiters(folio);
1272 	if (!folio_trylock_flag(folio, bit_nr, wait))
1273 		__add_wait_queue_entry_tail(q, wait);
1274 	spin_unlock_irq(&q->lock);
1275 
1276 	/*
1277 	 * From now on, all the logic will be based on
1278 	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1279 	 * see whether the page bit testing has already
1280 	 * been done by the wake function.
1281 	 *
1282 	 * We can drop our reference to the folio.
1283 	 */
1284 	if (behavior == DROP)
1285 		folio_put(folio);
1286 
1287 	/*
1288 	 * Note that until the "finish_wait()", or until
1289 	 * we see the WQ_FLAG_WOKEN flag, we need to
1290 	 * be very careful with the 'wait->flags', because
1291 	 * we may race with a waker that sets them.
1292 	 */
1293 	for (;;) {
1294 		unsigned int flags;
1295 
1296 		set_current_state(state);
1297 
1298 		/* Loop until we've been woken or interrupted */
1299 		flags = smp_load_acquire(&wait->flags);
1300 		if (!(flags & WQ_FLAG_WOKEN)) {
1301 			if (signal_pending_state(state, current))
1302 				break;
1303 
1304 			io_schedule();
1305 			continue;
1306 		}
1307 
1308 		/* If we were non-exclusive, we're done */
1309 		if (behavior != EXCLUSIVE)
1310 			break;
1311 
1312 		/* If the waker got the lock for us, we're done */
1313 		if (flags & WQ_FLAG_DONE)
1314 			break;
1315 
1316 		/*
1317 		 * Otherwise, if we're getting the lock, we need to
1318 		 * try to get it ourselves.
1319 		 *
1320 		 * And if that fails, we'll have to retry this all.
1321 		 */
1322 		if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1323 			goto repeat;
1324 
1325 		wait->flags |= WQ_FLAG_DONE;
1326 		break;
1327 	}
1328 
1329 	/*
1330 	 * If a signal happened, this 'finish_wait()' may remove the last
1331 	 * waiter from the wait-queues, but the folio waiters bit will remain
1332 	 * set. That's ok. The next wakeup will take care of it, and trying
1333 	 * to do it here would be difficult and prone to races.
1334 	 */
1335 	finish_wait(q, wait);
1336 
1337 	if (thrashing) {
1338 		delayacct_thrashing_end(&in_thrashing);
1339 		psi_memstall_leave(&pflags);
1340 	}
1341 
1342 	/*
1343 	 * NOTE! The wait->flags weren't stable until we've done the
1344 	 * 'finish_wait()', and we could have exited the loop above due
1345 	 * to a signal, and had a wakeup event happen after the signal
1346 	 * test but before the 'finish_wait()'.
1347 	 *
1348 	 * So only after the finish_wait() can we reliably determine
1349 	 * if we got woken up or not, so we can now figure out the final
1350 	 * return value based on that state without races.
1351 	 *
1352 	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1353 	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1354 	 */
1355 	if (behavior == EXCLUSIVE)
1356 		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1357 
1358 	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1359 }
1360 
1361 #ifdef CONFIG_MIGRATION
1362 /**
1363  * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1364  * @entry: migration swap entry.
1365  * @ptl: already locked ptl. This function will drop the lock.
1366  *
1367  * Wait for a migration entry referencing the given page to be removed. This is
1368  * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1369  * this can be called without taking a reference on the page. Instead this
1370  * should be called while holding the ptl for the migration entry referencing
1371  * the page.
1372  *
1373  * Returns after unlocking the ptl.
1374  *
1375  * This follows the same logic as folio_wait_bit_common() so see the comments
1376  * there.
1377  */
1378 void migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
1379 	__releases(ptl)
1380 {
1381 	struct wait_page_queue wait_page;
1382 	wait_queue_entry_t *wait = &wait_page.wait;
1383 	bool thrashing = false;
1384 	unsigned long pflags;
1385 	bool in_thrashing;
1386 	wait_queue_head_t *q;
1387 	struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1388 
1389 	q = folio_waitqueue(folio);
1390 	if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1391 		delayacct_thrashing_start(&in_thrashing);
1392 		psi_memstall_enter(&pflags);
1393 		thrashing = true;
1394 	}
1395 
1396 	init_wait(wait);
1397 	wait->func = wake_page_function;
1398 	wait_page.folio = folio;
1399 	wait_page.bit_nr = PG_locked;
1400 	wait->flags = 0;
1401 
1402 	spin_lock_irq(&q->lock);
1403 	folio_set_waiters(folio);
1404 	if (!folio_trylock_flag(folio, PG_locked, wait))
1405 		__add_wait_queue_entry_tail(q, wait);
1406 	spin_unlock_irq(&q->lock);
1407 
1408 	/*
1409 	 * If a migration entry exists for the page the migration path must hold
1410 	 * a valid reference to the page, and it must take the ptl to remove the
1411 	 * migration entry. So the page is valid until the ptl is dropped.
1412 	 */
1413 	spin_unlock(ptl);
1414 
1415 	for (;;) {
1416 		unsigned int flags;
1417 
1418 		set_current_state(TASK_UNINTERRUPTIBLE);
1419 
1420 		/* Loop until we've been woken or interrupted */
1421 		flags = smp_load_acquire(&wait->flags);
1422 		if (!(flags & WQ_FLAG_WOKEN)) {
1423 			if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1424 				break;
1425 
1426 			io_schedule();
1427 			continue;
1428 		}
1429 		break;
1430 	}
1431 
1432 	finish_wait(q, wait);
1433 
1434 	if (thrashing) {
1435 		delayacct_thrashing_end(&in_thrashing);
1436 		psi_memstall_leave(&pflags);
1437 	}
1438 }
1439 #endif
1440 
1441 void folio_wait_bit(struct folio *folio, int bit_nr)
1442 {
1443 	folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1444 }
1445 EXPORT_SYMBOL(folio_wait_bit);
1446 
1447 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1448 {
1449 	return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1450 }
1451 EXPORT_SYMBOL(folio_wait_bit_killable);
1452 
1453 /**
1454  * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1455  * @folio: The folio to wait for.
1456  * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1457  *
1458  * The caller should hold a reference on @folio.  They expect the page to
1459  * become unlocked relatively soon, but do not wish to hold up migration
1460  * (for example) by holding the reference while waiting for the folio to
1461  * come unlocked.  After this function returns, the caller should not
1462  * dereference @folio.
1463  *
1464  * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1465  */
1466 static int folio_put_wait_locked(struct folio *folio, int state)
1467 {
1468 	return folio_wait_bit_common(folio, PG_locked, state, DROP);
1469 }
1470 
1471 /**
1472  * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1473  * @folio: Folio defining the wait queue of interest
1474  * @waiter: Waiter to add to the queue
1475  *
1476  * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1477  */
1478 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1479 {
1480 	wait_queue_head_t *q = folio_waitqueue(folio);
1481 	unsigned long flags;
1482 
1483 	spin_lock_irqsave(&q->lock, flags);
1484 	__add_wait_queue_entry_tail(q, waiter);
1485 	folio_set_waiters(folio);
1486 	spin_unlock_irqrestore(&q->lock, flags);
1487 }
1488 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1489 
1490 #ifndef clear_bit_unlock_is_negative_byte
1491 
1492 /*
1493  * PG_waiters is the high bit in the same byte as PG_lock.
1494  *
1495  * On x86 (and on many other architectures), we can clear PG_lock and
1496  * test the sign bit at the same time. But if the architecture does
1497  * not support that special operation, we just do this all by hand
1498  * instead.
1499  *
1500  * The read of PG_waiters has to be after (or concurrently with) PG_locked
1501  * being cleared, but a memory barrier should be unnecessary since it is
1502  * in the same byte as PG_locked.
1503  */
1504 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1505 {
1506 	clear_bit_unlock(nr, mem);
1507 	/* smp_mb__after_atomic(); */
1508 	return test_bit(PG_waiters, mem);
1509 }
1510 
1511 #endif
1512 
1513 /**
1514  * folio_unlock - Unlock a locked folio.
1515  * @folio: The folio.
1516  *
1517  * Unlocks the folio and wakes up any thread sleeping on the page lock.
1518  *
1519  * Context: May be called from interrupt or process context.  May not be
1520  * called from NMI context.
1521  */
1522 void folio_unlock(struct folio *folio)
1523 {
1524 	/* Bit 7 allows x86 to check the byte's sign bit */
1525 	BUILD_BUG_ON(PG_waiters != 7);
1526 	BUILD_BUG_ON(PG_locked > 7);
1527 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1528 	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1529 		folio_wake_bit(folio, PG_locked);
1530 }
1531 EXPORT_SYMBOL(folio_unlock);
1532 
1533 /**
1534  * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1535  * @folio: The folio.
1536  *
1537  * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1538  * it.  The folio reference held for PG_private_2 being set is released.
1539  *
1540  * This is, for example, used when a netfs folio is being written to a local
1541  * disk cache, thereby allowing writes to the cache for the same folio to be
1542  * serialised.
1543  */
1544 void folio_end_private_2(struct folio *folio)
1545 {
1546 	VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1547 	clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1548 	folio_wake_bit(folio, PG_private_2);
1549 	folio_put(folio);
1550 }
1551 EXPORT_SYMBOL(folio_end_private_2);
1552 
1553 /**
1554  * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1555  * @folio: The folio to wait on.
1556  *
1557  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1558  */
1559 void folio_wait_private_2(struct folio *folio)
1560 {
1561 	while (folio_test_private_2(folio))
1562 		folio_wait_bit(folio, PG_private_2);
1563 }
1564 EXPORT_SYMBOL(folio_wait_private_2);
1565 
1566 /**
1567  * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1568  * @folio: The folio to wait on.
1569  *
1570  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1571  * fatal signal is received by the calling task.
1572  *
1573  * Return:
1574  * - 0 if successful.
1575  * - -EINTR if a fatal signal was encountered.
1576  */
1577 int folio_wait_private_2_killable(struct folio *folio)
1578 {
1579 	int ret = 0;
1580 
1581 	while (folio_test_private_2(folio)) {
1582 		ret = folio_wait_bit_killable(folio, PG_private_2);
1583 		if (ret < 0)
1584 			break;
1585 	}
1586 
1587 	return ret;
1588 }
1589 EXPORT_SYMBOL(folio_wait_private_2_killable);
1590 
1591 /**
1592  * folio_end_writeback - End writeback against a folio.
1593  * @folio: The folio.
1594  */
1595 void folio_end_writeback(struct folio *folio)
1596 {
1597 	/*
1598 	 * folio_test_clear_reclaim() could be used here but it is an
1599 	 * atomic operation and overkill in this particular case. Failing
1600 	 * to shuffle a folio marked for immediate reclaim is too mild
1601 	 * a gain to justify taking an atomic operation penalty at the
1602 	 * end of every folio writeback.
1603 	 */
1604 	if (folio_test_reclaim(folio)) {
1605 		folio_clear_reclaim(folio);
1606 		folio_rotate_reclaimable(folio);
1607 	}
1608 
1609 	/*
1610 	 * Writeback does not hold a folio reference of its own, relying
1611 	 * on truncation to wait for the clearing of PG_writeback.
1612 	 * But here we must make sure that the folio is not freed and
1613 	 * reused before the folio_wake().
1614 	 */
1615 	folio_get(folio);
1616 	if (!__folio_end_writeback(folio))
1617 		BUG();
1618 
1619 	smp_mb__after_atomic();
1620 	folio_wake(folio, PG_writeback);
1621 	acct_reclaim_writeback(folio);
1622 	folio_put(folio);
1623 }
1624 EXPORT_SYMBOL(folio_end_writeback);
1625 
1626 /**
1627  * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1628  * @folio: The folio to lock
1629  */
1630 void __folio_lock(struct folio *folio)
1631 {
1632 	folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1633 				EXCLUSIVE);
1634 }
1635 EXPORT_SYMBOL(__folio_lock);
1636 
1637 int __folio_lock_killable(struct folio *folio)
1638 {
1639 	return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1640 					EXCLUSIVE);
1641 }
1642 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1643 
1644 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1645 {
1646 	struct wait_queue_head *q = folio_waitqueue(folio);
1647 	int ret = 0;
1648 
1649 	wait->folio = folio;
1650 	wait->bit_nr = PG_locked;
1651 
1652 	spin_lock_irq(&q->lock);
1653 	__add_wait_queue_entry_tail(q, &wait->wait);
1654 	folio_set_waiters(folio);
1655 	ret = !folio_trylock(folio);
1656 	/*
1657 	 * If we were successful now, we know we're still on the
1658 	 * waitqueue as we're still under the lock. This means it's
1659 	 * safe to remove and return success, we know the callback
1660 	 * isn't going to trigger.
1661 	 */
1662 	if (!ret)
1663 		__remove_wait_queue(q, &wait->wait);
1664 	else
1665 		ret = -EIOCBQUEUED;
1666 	spin_unlock_irq(&q->lock);
1667 	return ret;
1668 }
1669 
1670 /*
1671  * Return values:
1672  * true - folio is locked; mmap_lock is still held.
1673  * false - folio is not locked.
1674  *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1675  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1676  *     which case mmap_lock is still held.
1677  *
1678  * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1679  * with the folio locked and the mmap_lock unperturbed.
1680  */
1681 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1682 			 unsigned int flags)
1683 {
1684 	if (fault_flag_allow_retry_first(flags)) {
1685 		/*
1686 		 * CAUTION! In this case, mmap_lock is not released
1687 		 * even though return 0.
1688 		 */
1689 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1690 			return false;
1691 
1692 		mmap_read_unlock(mm);
1693 		if (flags & FAULT_FLAG_KILLABLE)
1694 			folio_wait_locked_killable(folio);
1695 		else
1696 			folio_wait_locked(folio);
1697 		return false;
1698 	}
1699 	if (flags & FAULT_FLAG_KILLABLE) {
1700 		bool ret;
1701 
1702 		ret = __folio_lock_killable(folio);
1703 		if (ret) {
1704 			mmap_read_unlock(mm);
1705 			return false;
1706 		}
1707 	} else {
1708 		__folio_lock(folio);
1709 	}
1710 
1711 	return true;
1712 }
1713 
1714 /**
1715  * page_cache_next_miss() - Find the next gap in the page cache.
1716  * @mapping: Mapping.
1717  * @index: Index.
1718  * @max_scan: Maximum range to search.
1719  *
1720  * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1721  * gap with the lowest index.
1722  *
1723  * This function may be called under the rcu_read_lock.  However, this will
1724  * not atomically search a snapshot of the cache at a single point in time.
1725  * For example, if a gap is created at index 5, then subsequently a gap is
1726  * created at index 10, page_cache_next_miss covering both indices may
1727  * return 10 if called under the rcu_read_lock.
1728  *
1729  * Return: The index of the gap if found, otherwise an index outside the
1730  * range specified (in which case 'return - index >= max_scan' will be true).
1731  * In the rare case of index wrap-around, 0 will be returned.
1732  */
1733 pgoff_t page_cache_next_miss(struct address_space *mapping,
1734 			     pgoff_t index, unsigned long max_scan)
1735 {
1736 	XA_STATE(xas, &mapping->i_pages, index);
1737 
1738 	while (max_scan--) {
1739 		void *entry = xas_next(&xas);
1740 		if (!entry || xa_is_value(entry))
1741 			break;
1742 		if (xas.xa_index == 0)
1743 			break;
1744 	}
1745 
1746 	return xas.xa_index;
1747 }
1748 EXPORT_SYMBOL(page_cache_next_miss);
1749 
1750 /**
1751  * page_cache_prev_miss() - Find the previous gap in the page cache.
1752  * @mapping: Mapping.
1753  * @index: Index.
1754  * @max_scan: Maximum range to search.
1755  *
1756  * Search the range [max(index - max_scan + 1, 0), index] for the
1757  * gap with the highest index.
1758  *
1759  * This function may be called under the rcu_read_lock.  However, this will
1760  * not atomically search a snapshot of the cache at a single point in time.
1761  * For example, if a gap is created at index 10, then subsequently a gap is
1762  * created at index 5, page_cache_prev_miss() covering both indices may
1763  * return 5 if called under the rcu_read_lock.
1764  *
1765  * Return: The index of the gap if found, otherwise an index outside the
1766  * range specified (in which case 'index - return >= max_scan' will be true).
1767  * In the rare case of wrap-around, ULONG_MAX will be returned.
1768  */
1769 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1770 			     pgoff_t index, unsigned long max_scan)
1771 {
1772 	XA_STATE(xas, &mapping->i_pages, index);
1773 
1774 	while (max_scan--) {
1775 		void *entry = xas_prev(&xas);
1776 		if (!entry || xa_is_value(entry))
1777 			break;
1778 		if (xas.xa_index == ULONG_MAX)
1779 			break;
1780 	}
1781 
1782 	return xas.xa_index;
1783 }
1784 EXPORT_SYMBOL(page_cache_prev_miss);
1785 
1786 /*
1787  * Lockless page cache protocol:
1788  * On the lookup side:
1789  * 1. Load the folio from i_pages
1790  * 2. Increment the refcount if it's not zero
1791  * 3. If the folio is not found by xas_reload(), put the refcount and retry
1792  *
1793  * On the removal side:
1794  * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1795  * B. Remove the page from i_pages
1796  * C. Return the page to the page allocator
1797  *
1798  * This means that any page may have its reference count temporarily
1799  * increased by a speculative page cache (or fast GUP) lookup as it can
1800  * be allocated by another user before the RCU grace period expires.
1801  * Because the refcount temporarily acquired here may end up being the
1802  * last refcount on the page, any page allocation must be freeable by
1803  * folio_put().
1804  */
1805 
1806 /*
1807  * filemap_get_entry - Get a page cache entry.
1808  * @mapping: the address_space to search
1809  * @index: The page cache index.
1810  *
1811  * Looks up the page cache entry at @mapping & @index.  If it is a folio,
1812  * it is returned with an increased refcount.  If it is a shadow entry
1813  * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1814  * it is returned without further action.
1815  *
1816  * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1817  */
1818 void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1819 {
1820 	XA_STATE(xas, &mapping->i_pages, index);
1821 	struct folio *folio;
1822 
1823 	rcu_read_lock();
1824 repeat:
1825 	xas_reset(&xas);
1826 	folio = xas_load(&xas);
1827 	if (xas_retry(&xas, folio))
1828 		goto repeat;
1829 	/*
1830 	 * A shadow entry of a recently evicted page, or a swap entry from
1831 	 * shmem/tmpfs.  Return it without attempting to raise page count.
1832 	 */
1833 	if (!folio || xa_is_value(folio))
1834 		goto out;
1835 
1836 	if (!folio_try_get_rcu(folio))
1837 		goto repeat;
1838 
1839 	if (unlikely(folio != xas_reload(&xas))) {
1840 		folio_put(folio);
1841 		goto repeat;
1842 	}
1843 out:
1844 	rcu_read_unlock();
1845 
1846 	return folio;
1847 }
1848 
1849 /**
1850  * __filemap_get_folio - Find and get a reference to a folio.
1851  * @mapping: The address_space to search.
1852  * @index: The page index.
1853  * @fgp_flags: %FGP flags modify how the folio is returned.
1854  * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1855  *
1856  * Looks up the page cache entry at @mapping & @index.
1857  *
1858  * @fgp_flags can be zero or more of these flags:
1859  *
1860  * * %FGP_ACCESSED - The folio will be marked accessed.
1861  * * %FGP_LOCK - The folio is returned locked.
1862  * * %FGP_CREAT - If no page is present then a new page is allocated using
1863  *   @gfp and added to the page cache and the VM's LRU list.
1864  *   The page is returned locked and with an increased refcount.
1865  * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1866  *   page is already in cache.  If the page was allocated, unlock it before
1867  *   returning so the caller can do the same dance.
1868  * * %FGP_WRITE - The page will be written to by the caller.
1869  * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1870  * * %FGP_NOWAIT - Don't get blocked by page lock.
1871  * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1872  *
1873  * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1874  * if the %GFP flags specified for %FGP_CREAT are atomic.
1875  *
1876  * If there is a page cache page, it is returned with an increased refcount.
1877  *
1878  * Return: The found folio or an ERR_PTR() otherwise.
1879  */
1880 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1881 		int fgp_flags, gfp_t gfp)
1882 {
1883 	struct folio *folio;
1884 
1885 repeat:
1886 	folio = filemap_get_entry(mapping, index);
1887 	if (xa_is_value(folio))
1888 		folio = NULL;
1889 	if (!folio)
1890 		goto no_page;
1891 
1892 	if (fgp_flags & FGP_LOCK) {
1893 		if (fgp_flags & FGP_NOWAIT) {
1894 			if (!folio_trylock(folio)) {
1895 				folio_put(folio);
1896 				return ERR_PTR(-EAGAIN);
1897 			}
1898 		} else {
1899 			folio_lock(folio);
1900 		}
1901 
1902 		/* Has the page been truncated? */
1903 		if (unlikely(folio->mapping != mapping)) {
1904 			folio_unlock(folio);
1905 			folio_put(folio);
1906 			goto repeat;
1907 		}
1908 		VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1909 	}
1910 
1911 	if (fgp_flags & FGP_ACCESSED)
1912 		folio_mark_accessed(folio);
1913 	else if (fgp_flags & FGP_WRITE) {
1914 		/* Clear idle flag for buffer write */
1915 		if (folio_test_idle(folio))
1916 			folio_clear_idle(folio);
1917 	}
1918 
1919 	if (fgp_flags & FGP_STABLE)
1920 		folio_wait_stable(folio);
1921 no_page:
1922 	if (!folio && (fgp_flags & FGP_CREAT)) {
1923 		int err;
1924 		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1925 			gfp |= __GFP_WRITE;
1926 		if (fgp_flags & FGP_NOFS)
1927 			gfp &= ~__GFP_FS;
1928 		if (fgp_flags & FGP_NOWAIT) {
1929 			gfp &= ~GFP_KERNEL;
1930 			gfp |= GFP_NOWAIT | __GFP_NOWARN;
1931 		}
1932 
1933 		folio = filemap_alloc_folio(gfp, 0);
1934 		if (!folio)
1935 			return ERR_PTR(-ENOMEM);
1936 
1937 		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1938 			fgp_flags |= FGP_LOCK;
1939 
1940 		/* Init accessed so avoid atomic mark_page_accessed later */
1941 		if (fgp_flags & FGP_ACCESSED)
1942 			__folio_set_referenced(folio);
1943 
1944 		err = filemap_add_folio(mapping, folio, index, gfp);
1945 		if (unlikely(err)) {
1946 			folio_put(folio);
1947 			folio = NULL;
1948 			if (err == -EEXIST)
1949 				goto repeat;
1950 		}
1951 
1952 		/*
1953 		 * filemap_add_folio locks the page, and for mmap
1954 		 * we expect an unlocked page.
1955 		 */
1956 		if (folio && (fgp_flags & FGP_FOR_MMAP))
1957 			folio_unlock(folio);
1958 	}
1959 
1960 	if (!folio)
1961 		return ERR_PTR(-ENOENT);
1962 	return folio;
1963 }
1964 EXPORT_SYMBOL(__filemap_get_folio);
1965 
1966 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1967 		xa_mark_t mark)
1968 {
1969 	struct folio *folio;
1970 
1971 retry:
1972 	if (mark == XA_PRESENT)
1973 		folio = xas_find(xas, max);
1974 	else
1975 		folio = xas_find_marked(xas, max, mark);
1976 
1977 	if (xas_retry(xas, folio))
1978 		goto retry;
1979 	/*
1980 	 * A shadow entry of a recently evicted page, a swap
1981 	 * entry from shmem/tmpfs or a DAX entry.  Return it
1982 	 * without attempting to raise page count.
1983 	 */
1984 	if (!folio || xa_is_value(folio))
1985 		return folio;
1986 
1987 	if (!folio_try_get_rcu(folio))
1988 		goto reset;
1989 
1990 	if (unlikely(folio != xas_reload(xas))) {
1991 		folio_put(folio);
1992 		goto reset;
1993 	}
1994 
1995 	return folio;
1996 reset:
1997 	xas_reset(xas);
1998 	goto retry;
1999 }
2000 
2001 /**
2002  * find_get_entries - gang pagecache lookup
2003  * @mapping:	The address_space to search
2004  * @start:	The starting page cache index
2005  * @end:	The final page index (inclusive).
2006  * @fbatch:	Where the resulting entries are placed.
2007  * @indices:	The cache indices corresponding to the entries in @entries
2008  *
2009  * find_get_entries() will search for and return a batch of entries in
2010  * the mapping.  The entries are placed in @fbatch.  find_get_entries()
2011  * takes a reference on any actual folios it returns.
2012  *
2013  * The entries have ascending indexes.  The indices may not be consecutive
2014  * due to not-present entries or large folios.
2015  *
2016  * Any shadow entries of evicted folios, or swap entries from
2017  * shmem/tmpfs, are included in the returned array.
2018  *
2019  * Return: The number of entries which were found.
2020  */
2021 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2022 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2023 {
2024 	XA_STATE(xas, &mapping->i_pages, *start);
2025 	struct folio *folio;
2026 
2027 	rcu_read_lock();
2028 	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2029 		indices[fbatch->nr] = xas.xa_index;
2030 		if (!folio_batch_add(fbatch, folio))
2031 			break;
2032 	}
2033 	rcu_read_unlock();
2034 
2035 	if (folio_batch_count(fbatch)) {
2036 		unsigned long nr = 1;
2037 		int idx = folio_batch_count(fbatch) - 1;
2038 
2039 		folio = fbatch->folios[idx];
2040 		if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2041 			nr = folio_nr_pages(folio);
2042 		*start = indices[idx] + nr;
2043 	}
2044 	return folio_batch_count(fbatch);
2045 }
2046 
2047 /**
2048  * find_lock_entries - Find a batch of pagecache entries.
2049  * @mapping:	The address_space to search.
2050  * @start:	The starting page cache index.
2051  * @end:	The final page index (inclusive).
2052  * @fbatch:	Where the resulting entries are placed.
2053  * @indices:	The cache indices of the entries in @fbatch.
2054  *
2055  * find_lock_entries() will return a batch of entries from @mapping.
2056  * Swap, shadow and DAX entries are included.  Folios are returned
2057  * locked and with an incremented refcount.  Folios which are locked
2058  * by somebody else or under writeback are skipped.  Folios which are
2059  * partially outside the range are not returned.
2060  *
2061  * The entries have ascending indexes.  The indices may not be consecutive
2062  * due to not-present entries, large folios, folios which could not be
2063  * locked or folios under writeback.
2064  *
2065  * Return: The number of entries which were found.
2066  */
2067 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2068 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2069 {
2070 	XA_STATE(xas, &mapping->i_pages, *start);
2071 	struct folio *folio;
2072 
2073 	rcu_read_lock();
2074 	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2075 		if (!xa_is_value(folio)) {
2076 			if (folio->index < *start)
2077 				goto put;
2078 			if (folio->index + folio_nr_pages(folio) - 1 > end)
2079 				goto put;
2080 			if (!folio_trylock(folio))
2081 				goto put;
2082 			if (folio->mapping != mapping ||
2083 			    folio_test_writeback(folio))
2084 				goto unlock;
2085 			VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2086 					folio);
2087 		}
2088 		indices[fbatch->nr] = xas.xa_index;
2089 		if (!folio_batch_add(fbatch, folio))
2090 			break;
2091 		continue;
2092 unlock:
2093 		folio_unlock(folio);
2094 put:
2095 		folio_put(folio);
2096 	}
2097 	rcu_read_unlock();
2098 
2099 	if (folio_batch_count(fbatch)) {
2100 		unsigned long nr = 1;
2101 		int idx = folio_batch_count(fbatch) - 1;
2102 
2103 		folio = fbatch->folios[idx];
2104 		if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2105 			nr = folio_nr_pages(folio);
2106 		*start = indices[idx] + nr;
2107 	}
2108 	return folio_batch_count(fbatch);
2109 }
2110 
2111 /**
2112  * filemap_get_folios - Get a batch of folios
2113  * @mapping:	The address_space to search
2114  * @start:	The starting page index
2115  * @end:	The final page index (inclusive)
2116  * @fbatch:	The batch to fill.
2117  *
2118  * Search for and return a batch of folios in the mapping starting at
2119  * index @start and up to index @end (inclusive).  The folios are returned
2120  * in @fbatch with an elevated reference count.
2121  *
2122  * The first folio may start before @start; if it does, it will contain
2123  * @start.  The final folio may extend beyond @end; if it does, it will
2124  * contain @end.  The folios have ascending indices.  There may be gaps
2125  * between the folios if there are indices which have no folio in the
2126  * page cache.  If folios are added to or removed from the page cache
2127  * while this is running, they may or may not be found by this call.
2128  *
2129  * Return: The number of folios which were found.
2130  * We also update @start to index the next folio for the traversal.
2131  */
2132 unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2133 		pgoff_t end, struct folio_batch *fbatch)
2134 {
2135 	XA_STATE(xas, &mapping->i_pages, *start);
2136 	struct folio *folio;
2137 
2138 	rcu_read_lock();
2139 	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2140 		/* Skip over shadow, swap and DAX entries */
2141 		if (xa_is_value(folio))
2142 			continue;
2143 		if (!folio_batch_add(fbatch, folio)) {
2144 			unsigned long nr = folio_nr_pages(folio);
2145 
2146 			if (folio_test_hugetlb(folio))
2147 				nr = 1;
2148 			*start = folio->index + nr;
2149 			goto out;
2150 		}
2151 	}
2152 
2153 	/*
2154 	 * We come here when there is no page beyond @end. We take care to not
2155 	 * overflow the index @start as it confuses some of the callers. This
2156 	 * breaks the iteration when there is a page at index -1 but that is
2157 	 * already broken anyway.
2158 	 */
2159 	if (end == (pgoff_t)-1)
2160 		*start = (pgoff_t)-1;
2161 	else
2162 		*start = end + 1;
2163 out:
2164 	rcu_read_unlock();
2165 
2166 	return folio_batch_count(fbatch);
2167 }
2168 EXPORT_SYMBOL(filemap_get_folios);
2169 
2170 static inline
2171 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2172 {
2173 	if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2174 		return false;
2175 	if (index >= max)
2176 		return false;
2177 	return index < folio->index + folio_nr_pages(folio) - 1;
2178 }
2179 
2180 /**
2181  * filemap_get_folios_contig - Get a batch of contiguous folios
2182  * @mapping:	The address_space to search
2183  * @start:	The starting page index
2184  * @end:	The final page index (inclusive)
2185  * @fbatch:	The batch to fill
2186  *
2187  * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2188  * except the returned folios are guaranteed to be contiguous. This may
2189  * not return all contiguous folios if the batch gets filled up.
2190  *
2191  * Return: The number of folios found.
2192  * Also update @start to be positioned for traversal of the next folio.
2193  */
2194 
2195 unsigned filemap_get_folios_contig(struct address_space *mapping,
2196 		pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2197 {
2198 	XA_STATE(xas, &mapping->i_pages, *start);
2199 	unsigned long nr;
2200 	struct folio *folio;
2201 
2202 	rcu_read_lock();
2203 
2204 	for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2205 			folio = xas_next(&xas)) {
2206 		if (xas_retry(&xas, folio))
2207 			continue;
2208 		/*
2209 		 * If the entry has been swapped out, we can stop looking.
2210 		 * No current caller is looking for DAX entries.
2211 		 */
2212 		if (xa_is_value(folio))
2213 			goto update_start;
2214 
2215 		if (!folio_try_get_rcu(folio))
2216 			goto retry;
2217 
2218 		if (unlikely(folio != xas_reload(&xas)))
2219 			goto put_folio;
2220 
2221 		if (!folio_batch_add(fbatch, folio)) {
2222 			nr = folio_nr_pages(folio);
2223 
2224 			if (folio_test_hugetlb(folio))
2225 				nr = 1;
2226 			*start = folio->index + nr;
2227 			goto out;
2228 		}
2229 		continue;
2230 put_folio:
2231 		folio_put(folio);
2232 
2233 retry:
2234 		xas_reset(&xas);
2235 	}
2236 
2237 update_start:
2238 	nr = folio_batch_count(fbatch);
2239 
2240 	if (nr) {
2241 		folio = fbatch->folios[nr - 1];
2242 		if (folio_test_hugetlb(folio))
2243 			*start = folio->index + 1;
2244 		else
2245 			*start = folio->index + folio_nr_pages(folio);
2246 	}
2247 out:
2248 	rcu_read_unlock();
2249 	return folio_batch_count(fbatch);
2250 }
2251 EXPORT_SYMBOL(filemap_get_folios_contig);
2252 
2253 /**
2254  * filemap_get_folios_tag - Get a batch of folios matching @tag
2255  * @mapping:    The address_space to search
2256  * @start:      The starting page index
2257  * @end:        The final page index (inclusive)
2258  * @tag:        The tag index
2259  * @fbatch:     The batch to fill
2260  *
2261  * Same as filemap_get_folios(), but only returning folios tagged with @tag.
2262  *
2263  * Return: The number of folios found.
2264  * Also update @start to index the next folio for traversal.
2265  */
2266 unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2267 			pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2268 {
2269 	XA_STATE(xas, &mapping->i_pages, *start);
2270 	struct folio *folio;
2271 
2272 	rcu_read_lock();
2273 	while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2274 		/*
2275 		 * Shadow entries should never be tagged, but this iteration
2276 		 * is lockless so there is a window for page reclaim to evict
2277 		 * a page we saw tagged. Skip over it.
2278 		 */
2279 		if (xa_is_value(folio))
2280 			continue;
2281 		if (!folio_batch_add(fbatch, folio)) {
2282 			unsigned long nr = folio_nr_pages(folio);
2283 
2284 			if (folio_test_hugetlb(folio))
2285 				nr = 1;
2286 			*start = folio->index + nr;
2287 			goto out;
2288 		}
2289 	}
2290 	/*
2291 	 * We come here when there is no page beyond @end. We take care to not
2292 	 * overflow the index @start as it confuses some of the callers. This
2293 	 * breaks the iteration when there is a page at index -1 but that is
2294 	 * already broke anyway.
2295 	 */
2296 	if (end == (pgoff_t)-1)
2297 		*start = (pgoff_t)-1;
2298 	else
2299 		*start = end + 1;
2300 out:
2301 	rcu_read_unlock();
2302 
2303 	return folio_batch_count(fbatch);
2304 }
2305 EXPORT_SYMBOL(filemap_get_folios_tag);
2306 
2307 /*
2308  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2309  * a _large_ part of the i/o request. Imagine the worst scenario:
2310  *
2311  *      ---R__________________________________________B__________
2312  *         ^ reading here                             ^ bad block(assume 4k)
2313  *
2314  * read(R) => miss => readahead(R...B) => media error => frustrating retries
2315  * => failing the whole request => read(R) => read(R+1) =>
2316  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2317  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2318  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2319  *
2320  * It is going insane. Fix it by quickly scaling down the readahead size.
2321  */
2322 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2323 {
2324 	ra->ra_pages /= 4;
2325 }
2326 
2327 /*
2328  * filemap_get_read_batch - Get a batch of folios for read
2329  *
2330  * Get a batch of folios which represent a contiguous range of bytes in
2331  * the file.  No exceptional entries will be returned.  If @index is in
2332  * the middle of a folio, the entire folio will be returned.  The last
2333  * folio in the batch may have the readahead flag set or the uptodate flag
2334  * clear so that the caller can take the appropriate action.
2335  */
2336 static void filemap_get_read_batch(struct address_space *mapping,
2337 		pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2338 {
2339 	XA_STATE(xas, &mapping->i_pages, index);
2340 	struct folio *folio;
2341 
2342 	rcu_read_lock();
2343 	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2344 		if (xas_retry(&xas, folio))
2345 			continue;
2346 		if (xas.xa_index > max || xa_is_value(folio))
2347 			break;
2348 		if (xa_is_sibling(folio))
2349 			break;
2350 		if (!folio_try_get_rcu(folio))
2351 			goto retry;
2352 
2353 		if (unlikely(folio != xas_reload(&xas)))
2354 			goto put_folio;
2355 
2356 		if (!folio_batch_add(fbatch, folio))
2357 			break;
2358 		if (!folio_test_uptodate(folio))
2359 			break;
2360 		if (folio_test_readahead(folio))
2361 			break;
2362 		xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2363 		continue;
2364 put_folio:
2365 		folio_put(folio);
2366 retry:
2367 		xas_reset(&xas);
2368 	}
2369 	rcu_read_unlock();
2370 }
2371 
2372 static int filemap_read_folio(struct file *file, filler_t filler,
2373 		struct folio *folio)
2374 {
2375 	bool workingset = folio_test_workingset(folio);
2376 	unsigned long pflags;
2377 	int error;
2378 
2379 	/*
2380 	 * A previous I/O error may have been due to temporary failures,
2381 	 * eg. multipath errors.  PG_error will be set again if read_folio
2382 	 * fails.
2383 	 */
2384 	folio_clear_error(folio);
2385 
2386 	/* Start the actual read. The read will unlock the page. */
2387 	if (unlikely(workingset))
2388 		psi_memstall_enter(&pflags);
2389 	error = filler(file, folio);
2390 	if (unlikely(workingset))
2391 		psi_memstall_leave(&pflags);
2392 	if (error)
2393 		return error;
2394 
2395 	error = folio_wait_locked_killable(folio);
2396 	if (error)
2397 		return error;
2398 	if (folio_test_uptodate(folio))
2399 		return 0;
2400 	if (file)
2401 		shrink_readahead_size_eio(&file->f_ra);
2402 	return -EIO;
2403 }
2404 
2405 static bool filemap_range_uptodate(struct address_space *mapping,
2406 		loff_t pos, size_t count, struct folio *folio,
2407 		bool need_uptodate)
2408 {
2409 	if (folio_test_uptodate(folio))
2410 		return true;
2411 	/* pipes can't handle partially uptodate pages */
2412 	if (need_uptodate)
2413 		return false;
2414 	if (!mapping->a_ops->is_partially_uptodate)
2415 		return false;
2416 	if (mapping->host->i_blkbits >= folio_shift(folio))
2417 		return false;
2418 
2419 	if (folio_pos(folio) > pos) {
2420 		count -= folio_pos(folio) - pos;
2421 		pos = 0;
2422 	} else {
2423 		pos -= folio_pos(folio);
2424 	}
2425 
2426 	return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2427 }
2428 
2429 static int filemap_update_page(struct kiocb *iocb,
2430 		struct address_space *mapping, size_t count,
2431 		struct folio *folio, bool need_uptodate)
2432 {
2433 	int error;
2434 
2435 	if (iocb->ki_flags & IOCB_NOWAIT) {
2436 		if (!filemap_invalidate_trylock_shared(mapping))
2437 			return -EAGAIN;
2438 	} else {
2439 		filemap_invalidate_lock_shared(mapping);
2440 	}
2441 
2442 	if (!folio_trylock(folio)) {
2443 		error = -EAGAIN;
2444 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2445 			goto unlock_mapping;
2446 		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2447 			filemap_invalidate_unlock_shared(mapping);
2448 			/*
2449 			 * This is where we usually end up waiting for a
2450 			 * previously submitted readahead to finish.
2451 			 */
2452 			folio_put_wait_locked(folio, TASK_KILLABLE);
2453 			return AOP_TRUNCATED_PAGE;
2454 		}
2455 		error = __folio_lock_async(folio, iocb->ki_waitq);
2456 		if (error)
2457 			goto unlock_mapping;
2458 	}
2459 
2460 	error = AOP_TRUNCATED_PAGE;
2461 	if (!folio->mapping)
2462 		goto unlock;
2463 
2464 	error = 0;
2465 	if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2466 				   need_uptodate))
2467 		goto unlock;
2468 
2469 	error = -EAGAIN;
2470 	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2471 		goto unlock;
2472 
2473 	error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2474 			folio);
2475 	goto unlock_mapping;
2476 unlock:
2477 	folio_unlock(folio);
2478 unlock_mapping:
2479 	filemap_invalidate_unlock_shared(mapping);
2480 	if (error == AOP_TRUNCATED_PAGE)
2481 		folio_put(folio);
2482 	return error;
2483 }
2484 
2485 static int filemap_create_folio(struct file *file,
2486 		struct address_space *mapping, pgoff_t index,
2487 		struct folio_batch *fbatch)
2488 {
2489 	struct folio *folio;
2490 	int error;
2491 
2492 	folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2493 	if (!folio)
2494 		return -ENOMEM;
2495 
2496 	/*
2497 	 * Protect against truncate / hole punch. Grabbing invalidate_lock
2498 	 * here assures we cannot instantiate and bring uptodate new
2499 	 * pagecache folios after evicting page cache during truncate
2500 	 * and before actually freeing blocks.	Note that we could
2501 	 * release invalidate_lock after inserting the folio into
2502 	 * the page cache as the locked folio would then be enough to
2503 	 * synchronize with hole punching. But there are code paths
2504 	 * such as filemap_update_page() filling in partially uptodate
2505 	 * pages or ->readahead() that need to hold invalidate_lock
2506 	 * while mapping blocks for IO so let's hold the lock here as
2507 	 * well to keep locking rules simple.
2508 	 */
2509 	filemap_invalidate_lock_shared(mapping);
2510 	error = filemap_add_folio(mapping, folio, index,
2511 			mapping_gfp_constraint(mapping, GFP_KERNEL));
2512 	if (error == -EEXIST)
2513 		error = AOP_TRUNCATED_PAGE;
2514 	if (error)
2515 		goto error;
2516 
2517 	error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2518 	if (error)
2519 		goto error;
2520 
2521 	filemap_invalidate_unlock_shared(mapping);
2522 	folio_batch_add(fbatch, folio);
2523 	return 0;
2524 error:
2525 	filemap_invalidate_unlock_shared(mapping);
2526 	folio_put(folio);
2527 	return error;
2528 }
2529 
2530 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2531 		struct address_space *mapping, struct folio *folio,
2532 		pgoff_t last_index)
2533 {
2534 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2535 
2536 	if (iocb->ki_flags & IOCB_NOIO)
2537 		return -EAGAIN;
2538 	page_cache_async_ra(&ractl, folio, last_index - folio->index);
2539 	return 0;
2540 }
2541 
2542 static int filemap_get_pages(struct kiocb *iocb, size_t count,
2543 		struct folio_batch *fbatch, bool need_uptodate)
2544 {
2545 	struct file *filp = iocb->ki_filp;
2546 	struct address_space *mapping = filp->f_mapping;
2547 	struct file_ra_state *ra = &filp->f_ra;
2548 	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2549 	pgoff_t last_index;
2550 	struct folio *folio;
2551 	int err = 0;
2552 
2553 	/* "last_index" is the index of the page beyond the end of the read */
2554 	last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2555 retry:
2556 	if (fatal_signal_pending(current))
2557 		return -EINTR;
2558 
2559 	filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2560 	if (!folio_batch_count(fbatch)) {
2561 		if (iocb->ki_flags & IOCB_NOIO)
2562 			return -EAGAIN;
2563 		page_cache_sync_readahead(mapping, ra, filp, index,
2564 				last_index - index);
2565 		filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2566 	}
2567 	if (!folio_batch_count(fbatch)) {
2568 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2569 			return -EAGAIN;
2570 		err = filemap_create_folio(filp, mapping,
2571 				iocb->ki_pos >> PAGE_SHIFT, fbatch);
2572 		if (err == AOP_TRUNCATED_PAGE)
2573 			goto retry;
2574 		return err;
2575 	}
2576 
2577 	folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2578 	if (folio_test_readahead(folio)) {
2579 		err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2580 		if (err)
2581 			goto err;
2582 	}
2583 	if (!folio_test_uptodate(folio)) {
2584 		if ((iocb->ki_flags & IOCB_WAITQ) &&
2585 		    folio_batch_count(fbatch) > 1)
2586 			iocb->ki_flags |= IOCB_NOWAIT;
2587 		err = filemap_update_page(iocb, mapping, count, folio,
2588 					  need_uptodate);
2589 		if (err)
2590 			goto err;
2591 	}
2592 
2593 	return 0;
2594 err:
2595 	if (err < 0)
2596 		folio_put(folio);
2597 	if (likely(--fbatch->nr))
2598 		return 0;
2599 	if (err == AOP_TRUNCATED_PAGE)
2600 		goto retry;
2601 	return err;
2602 }
2603 
2604 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2605 {
2606 	unsigned int shift = folio_shift(folio);
2607 
2608 	return (pos1 >> shift == pos2 >> shift);
2609 }
2610 
2611 /**
2612  * filemap_read - Read data from the page cache.
2613  * @iocb: The iocb to read.
2614  * @iter: Destination for the data.
2615  * @already_read: Number of bytes already read by the caller.
2616  *
2617  * Copies data from the page cache.  If the data is not currently present,
2618  * uses the readahead and read_folio address_space operations to fetch it.
2619  *
2620  * Return: Total number of bytes copied, including those already read by
2621  * the caller.  If an error happens before any bytes are copied, returns
2622  * a negative error number.
2623  */
2624 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2625 		ssize_t already_read)
2626 {
2627 	struct file *filp = iocb->ki_filp;
2628 	struct file_ra_state *ra = &filp->f_ra;
2629 	struct address_space *mapping = filp->f_mapping;
2630 	struct inode *inode = mapping->host;
2631 	struct folio_batch fbatch;
2632 	int i, error = 0;
2633 	bool writably_mapped;
2634 	loff_t isize, end_offset;
2635 
2636 	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2637 		return 0;
2638 	if (unlikely(!iov_iter_count(iter)))
2639 		return 0;
2640 
2641 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2642 	folio_batch_init(&fbatch);
2643 
2644 	do {
2645 		cond_resched();
2646 
2647 		/*
2648 		 * If we've already successfully copied some data, then we
2649 		 * can no longer safely return -EIOCBQUEUED. Hence mark
2650 		 * an async read NOWAIT at that point.
2651 		 */
2652 		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2653 			iocb->ki_flags |= IOCB_NOWAIT;
2654 
2655 		if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2656 			break;
2657 
2658 		error = filemap_get_pages(iocb, iter->count, &fbatch, false);
2659 		if (error < 0)
2660 			break;
2661 
2662 		/*
2663 		 * i_size must be checked after we know the pages are Uptodate.
2664 		 *
2665 		 * Checking i_size after the check allows us to calculate
2666 		 * the correct value for "nr", which means the zero-filled
2667 		 * part of the page is not copied back to userspace (unless
2668 		 * another truncate extends the file - this is desired though).
2669 		 */
2670 		isize = i_size_read(inode);
2671 		if (unlikely(iocb->ki_pos >= isize))
2672 			goto put_folios;
2673 		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2674 
2675 		/*
2676 		 * Once we start copying data, we don't want to be touching any
2677 		 * cachelines that might be contended:
2678 		 */
2679 		writably_mapped = mapping_writably_mapped(mapping);
2680 
2681 		/*
2682 		 * When a read accesses the same folio several times, only
2683 		 * mark it as accessed the first time.
2684 		 */
2685 		if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1,
2686 							fbatch.folios[0]))
2687 			folio_mark_accessed(fbatch.folios[0]);
2688 
2689 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2690 			struct folio *folio = fbatch.folios[i];
2691 			size_t fsize = folio_size(folio);
2692 			size_t offset = iocb->ki_pos & (fsize - 1);
2693 			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2694 					     fsize - offset);
2695 			size_t copied;
2696 
2697 			if (end_offset < folio_pos(folio))
2698 				break;
2699 			if (i > 0)
2700 				folio_mark_accessed(folio);
2701 			/*
2702 			 * If users can be writing to this folio using arbitrary
2703 			 * virtual addresses, take care of potential aliasing
2704 			 * before reading the folio on the kernel side.
2705 			 */
2706 			if (writably_mapped)
2707 				flush_dcache_folio(folio);
2708 
2709 			copied = copy_folio_to_iter(folio, offset, bytes, iter);
2710 
2711 			already_read += copied;
2712 			iocb->ki_pos += copied;
2713 			ra->prev_pos = iocb->ki_pos;
2714 
2715 			if (copied < bytes) {
2716 				error = -EFAULT;
2717 				break;
2718 			}
2719 		}
2720 put_folios:
2721 		for (i = 0; i < folio_batch_count(&fbatch); i++)
2722 			folio_put(fbatch.folios[i]);
2723 		folio_batch_init(&fbatch);
2724 	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2725 
2726 	file_accessed(filp);
2727 
2728 	return already_read ? already_read : error;
2729 }
2730 EXPORT_SYMBOL_GPL(filemap_read);
2731 
2732 int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2733 {
2734 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2735 	loff_t pos = iocb->ki_pos;
2736 	loff_t end = pos + count - 1;
2737 
2738 	if (iocb->ki_flags & IOCB_NOWAIT) {
2739 		if (filemap_range_needs_writeback(mapping, pos, end))
2740 			return -EAGAIN;
2741 		return 0;
2742 	}
2743 
2744 	return filemap_write_and_wait_range(mapping, pos, end);
2745 }
2746 
2747 int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2748 {
2749 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2750 	loff_t pos = iocb->ki_pos;
2751 	loff_t end = pos + count - 1;
2752 	int ret;
2753 
2754 	if (iocb->ki_flags & IOCB_NOWAIT) {
2755 		/* we could block if there are any pages in the range */
2756 		if (filemap_range_has_page(mapping, pos, end))
2757 			return -EAGAIN;
2758 	} else {
2759 		ret = filemap_write_and_wait_range(mapping, pos, end);
2760 		if (ret)
2761 			return ret;
2762 	}
2763 
2764 	/*
2765 	 * After a write we want buffered reads to be sure to go to disk to get
2766 	 * the new data.  We invalidate clean cached page from the region we're
2767 	 * about to write.  We do this *before* the write so that we can return
2768 	 * without clobbering -EIOCBQUEUED from ->direct_IO().
2769 	 */
2770 	return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2771 					     end >> PAGE_SHIFT);
2772 }
2773 
2774 /**
2775  * generic_file_read_iter - generic filesystem read routine
2776  * @iocb:	kernel I/O control block
2777  * @iter:	destination for the data read
2778  *
2779  * This is the "read_iter()" routine for all filesystems
2780  * that can use the page cache directly.
2781  *
2782  * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2783  * be returned when no data can be read without waiting for I/O requests
2784  * to complete; it doesn't prevent readahead.
2785  *
2786  * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2787  * requests shall be made for the read or for readahead.  When no data
2788  * can be read, -EAGAIN shall be returned.  When readahead would be
2789  * triggered, a partial, possibly empty read shall be returned.
2790  *
2791  * Return:
2792  * * number of bytes copied, even for partial reads
2793  * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2794  */
2795 ssize_t
2796 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2797 {
2798 	size_t count = iov_iter_count(iter);
2799 	ssize_t retval = 0;
2800 
2801 	if (!count)
2802 		return 0; /* skip atime */
2803 
2804 	if (iocb->ki_flags & IOCB_DIRECT) {
2805 		struct file *file = iocb->ki_filp;
2806 		struct address_space *mapping = file->f_mapping;
2807 		struct inode *inode = mapping->host;
2808 
2809 		retval = kiocb_write_and_wait(iocb, count);
2810 		if (retval < 0)
2811 			return retval;
2812 		file_accessed(file);
2813 
2814 		retval = mapping->a_ops->direct_IO(iocb, iter);
2815 		if (retval >= 0) {
2816 			iocb->ki_pos += retval;
2817 			count -= retval;
2818 		}
2819 		if (retval != -EIOCBQUEUED)
2820 			iov_iter_revert(iter, count - iov_iter_count(iter));
2821 
2822 		/*
2823 		 * Btrfs can have a short DIO read if we encounter
2824 		 * compressed extents, so if there was an error, or if
2825 		 * we've already read everything we wanted to, or if
2826 		 * there was a short read because we hit EOF, go ahead
2827 		 * and return.  Otherwise fallthrough to buffered io for
2828 		 * the rest of the read.  Buffered reads will not work for
2829 		 * DAX files, so don't bother trying.
2830 		 */
2831 		if (retval < 0 || !count || IS_DAX(inode))
2832 			return retval;
2833 		if (iocb->ki_pos >= i_size_read(inode))
2834 			return retval;
2835 	}
2836 
2837 	return filemap_read(iocb, iter, retval);
2838 }
2839 EXPORT_SYMBOL(generic_file_read_iter);
2840 
2841 /*
2842  * Splice subpages from a folio into a pipe.
2843  */
2844 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2845 			      struct folio *folio, loff_t fpos, size_t size)
2846 {
2847 	struct page *page;
2848 	size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2849 
2850 	page = folio_page(folio, offset / PAGE_SIZE);
2851 	size = min(size, folio_size(folio) - offset);
2852 	offset %= PAGE_SIZE;
2853 
2854 	while (spliced < size &&
2855 	       !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2856 		struct pipe_buffer *buf = pipe_head_buf(pipe);
2857 		size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2858 
2859 		*buf = (struct pipe_buffer) {
2860 			.ops	= &page_cache_pipe_buf_ops,
2861 			.page	= page,
2862 			.offset	= offset,
2863 			.len	= part,
2864 		};
2865 		folio_get(folio);
2866 		pipe->head++;
2867 		page++;
2868 		spliced += part;
2869 		offset = 0;
2870 	}
2871 
2872 	return spliced;
2873 }
2874 
2875 /**
2876  * filemap_splice_read -  Splice data from a file's pagecache into a pipe
2877  * @in: The file to read from
2878  * @ppos: Pointer to the file position to read from
2879  * @pipe: The pipe to splice into
2880  * @len: The amount to splice
2881  * @flags: The SPLICE_F_* flags
2882  *
2883  * This function gets folios from a file's pagecache and splices them into the
2884  * pipe.  Readahead will be called as necessary to fill more folios.  This may
2885  * be used for blockdevs also.
2886  *
2887  * Return: On success, the number of bytes read will be returned and *@ppos
2888  * will be updated if appropriate; 0 will be returned if there is no more data
2889  * to be read; -EAGAIN will be returned if the pipe had no space, and some
2890  * other negative error code will be returned on error.  A short read may occur
2891  * if the pipe has insufficient space, we reach the end of the data or we hit a
2892  * hole.
2893  */
2894 ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2895 			    struct pipe_inode_info *pipe,
2896 			    size_t len, unsigned int flags)
2897 {
2898 	struct folio_batch fbatch;
2899 	struct kiocb iocb;
2900 	size_t total_spliced = 0, used, npages;
2901 	loff_t isize, end_offset;
2902 	bool writably_mapped;
2903 	int i, error = 0;
2904 
2905 	if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
2906 		return 0;
2907 
2908 	init_sync_kiocb(&iocb, in);
2909 	iocb.ki_pos = *ppos;
2910 
2911 	/* Work out how much data we can actually add into the pipe */
2912 	used = pipe_occupancy(pipe->head, pipe->tail);
2913 	npages = max_t(ssize_t, pipe->max_usage - used, 0);
2914 	len = min_t(size_t, len, npages * PAGE_SIZE);
2915 
2916 	folio_batch_init(&fbatch);
2917 
2918 	do {
2919 		cond_resched();
2920 
2921 		if (*ppos >= i_size_read(in->f_mapping->host))
2922 			break;
2923 
2924 		iocb.ki_pos = *ppos;
2925 		error = filemap_get_pages(&iocb, len, &fbatch, true);
2926 		if (error < 0)
2927 			break;
2928 
2929 		/*
2930 		 * i_size must be checked after we know the pages are Uptodate.
2931 		 *
2932 		 * Checking i_size after the check allows us to calculate
2933 		 * the correct value for "nr", which means the zero-filled
2934 		 * part of the page is not copied back to userspace (unless
2935 		 * another truncate extends the file - this is desired though).
2936 		 */
2937 		isize = i_size_read(in->f_mapping->host);
2938 		if (unlikely(*ppos >= isize))
2939 			break;
2940 		end_offset = min_t(loff_t, isize, *ppos + len);
2941 
2942 		/*
2943 		 * Once we start copying data, we don't want to be touching any
2944 		 * cachelines that might be contended:
2945 		 */
2946 		writably_mapped = mapping_writably_mapped(in->f_mapping);
2947 
2948 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2949 			struct folio *folio = fbatch.folios[i];
2950 			size_t n;
2951 
2952 			if (folio_pos(folio) >= end_offset)
2953 				goto out;
2954 			folio_mark_accessed(folio);
2955 
2956 			/*
2957 			 * If users can be writing to this folio using arbitrary
2958 			 * virtual addresses, take care of potential aliasing
2959 			 * before reading the folio on the kernel side.
2960 			 */
2961 			if (writably_mapped)
2962 				flush_dcache_folio(folio);
2963 
2964 			n = min_t(loff_t, len, isize - *ppos);
2965 			n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2966 			if (!n)
2967 				goto out;
2968 			len -= n;
2969 			total_spliced += n;
2970 			*ppos += n;
2971 			in->f_ra.prev_pos = *ppos;
2972 			if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2973 				goto out;
2974 		}
2975 
2976 		folio_batch_release(&fbatch);
2977 	} while (len);
2978 
2979 out:
2980 	folio_batch_release(&fbatch);
2981 	file_accessed(in);
2982 
2983 	return total_spliced ? total_spliced : error;
2984 }
2985 EXPORT_SYMBOL(filemap_splice_read);
2986 
2987 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2988 		struct address_space *mapping, struct folio *folio,
2989 		loff_t start, loff_t end, bool seek_data)
2990 {
2991 	const struct address_space_operations *ops = mapping->a_ops;
2992 	size_t offset, bsz = i_blocksize(mapping->host);
2993 
2994 	if (xa_is_value(folio) || folio_test_uptodate(folio))
2995 		return seek_data ? start : end;
2996 	if (!ops->is_partially_uptodate)
2997 		return seek_data ? end : start;
2998 
2999 	xas_pause(xas);
3000 	rcu_read_unlock();
3001 	folio_lock(folio);
3002 	if (unlikely(folio->mapping != mapping))
3003 		goto unlock;
3004 
3005 	offset = offset_in_folio(folio, start) & ~(bsz - 1);
3006 
3007 	do {
3008 		if (ops->is_partially_uptodate(folio, offset, bsz) ==
3009 							seek_data)
3010 			break;
3011 		start = (start + bsz) & ~(bsz - 1);
3012 		offset += bsz;
3013 	} while (offset < folio_size(folio));
3014 unlock:
3015 	folio_unlock(folio);
3016 	rcu_read_lock();
3017 	return start;
3018 }
3019 
3020 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3021 {
3022 	if (xa_is_value(folio))
3023 		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3024 	return folio_size(folio);
3025 }
3026 
3027 /**
3028  * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3029  * @mapping: Address space to search.
3030  * @start: First byte to consider.
3031  * @end: Limit of search (exclusive).
3032  * @whence: Either SEEK_HOLE or SEEK_DATA.
3033  *
3034  * If the page cache knows which blocks contain holes and which blocks
3035  * contain data, your filesystem can use this function to implement
3036  * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
3037  * entirely memory-based such as tmpfs, and filesystems which support
3038  * unwritten extents.
3039  *
3040  * Return: The requested offset on success, or -ENXIO if @whence specifies
3041  * SEEK_DATA and there is no data after @start.  There is an implicit hole
3042  * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3043  * and @end contain data.
3044  */
3045 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3046 		loff_t end, int whence)
3047 {
3048 	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3049 	pgoff_t max = (end - 1) >> PAGE_SHIFT;
3050 	bool seek_data = (whence == SEEK_DATA);
3051 	struct folio *folio;
3052 
3053 	if (end <= start)
3054 		return -ENXIO;
3055 
3056 	rcu_read_lock();
3057 	while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3058 		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3059 		size_t seek_size;
3060 
3061 		if (start < pos) {
3062 			if (!seek_data)
3063 				goto unlock;
3064 			start = pos;
3065 		}
3066 
3067 		seek_size = seek_folio_size(&xas, folio);
3068 		pos = round_up((u64)pos + 1, seek_size);
3069 		start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3070 				seek_data);
3071 		if (start < pos)
3072 			goto unlock;
3073 		if (start >= end)
3074 			break;
3075 		if (seek_size > PAGE_SIZE)
3076 			xas_set(&xas, pos >> PAGE_SHIFT);
3077 		if (!xa_is_value(folio))
3078 			folio_put(folio);
3079 	}
3080 	if (seek_data)
3081 		start = -ENXIO;
3082 unlock:
3083 	rcu_read_unlock();
3084 	if (folio && !xa_is_value(folio))
3085 		folio_put(folio);
3086 	if (start > end)
3087 		return end;
3088 	return start;
3089 }
3090 
3091 #ifdef CONFIG_MMU
3092 #define MMAP_LOTSAMISS  (100)
3093 /*
3094  * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3095  * @vmf - the vm_fault for this fault.
3096  * @folio - the folio to lock.
3097  * @fpin - the pointer to the file we may pin (or is already pinned).
3098  *
3099  * This works similar to lock_folio_or_retry in that it can drop the
3100  * mmap_lock.  It differs in that it actually returns the folio locked
3101  * if it returns 1 and 0 if it couldn't lock the folio.  If we did have
3102  * to drop the mmap_lock then fpin will point to the pinned file and
3103  * needs to be fput()'ed at a later point.
3104  */
3105 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3106 				     struct file **fpin)
3107 {
3108 	if (folio_trylock(folio))
3109 		return 1;
3110 
3111 	/*
3112 	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3113 	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3114 	 * is supposed to work. We have way too many special cases..
3115 	 */
3116 	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3117 		return 0;
3118 
3119 	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3120 	if (vmf->flags & FAULT_FLAG_KILLABLE) {
3121 		if (__folio_lock_killable(folio)) {
3122 			/*
3123 			 * We didn't have the right flags to drop the mmap_lock,
3124 			 * but all fault_handlers only check for fatal signals
3125 			 * if we return VM_FAULT_RETRY, so we need to drop the
3126 			 * mmap_lock here and return 0 if we don't have a fpin.
3127 			 */
3128 			if (*fpin == NULL)
3129 				mmap_read_unlock(vmf->vma->vm_mm);
3130 			return 0;
3131 		}
3132 	} else
3133 		__folio_lock(folio);
3134 
3135 	return 1;
3136 }
3137 
3138 /*
3139  * Synchronous readahead happens when we don't even find a page in the page
3140  * cache at all.  We don't want to perform IO under the mmap sem, so if we have
3141  * to drop the mmap sem we return the file that was pinned in order for us to do
3142  * that.  If we didn't pin a file then we return NULL.  The file that is
3143  * returned needs to be fput()'ed when we're done with it.
3144  */
3145 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3146 {
3147 	struct file *file = vmf->vma->vm_file;
3148 	struct file_ra_state *ra = &file->f_ra;
3149 	struct address_space *mapping = file->f_mapping;
3150 	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3151 	struct file *fpin = NULL;
3152 	unsigned long vm_flags = vmf->vma->vm_flags;
3153 	unsigned int mmap_miss;
3154 
3155 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3156 	/* Use the readahead code, even if readahead is disabled */
3157 	if (vm_flags & VM_HUGEPAGE) {
3158 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3159 		ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3160 		ra->size = HPAGE_PMD_NR;
3161 		/*
3162 		 * Fetch two PMD folios, so we get the chance to actually
3163 		 * readahead, unless we've been told not to.
3164 		 */
3165 		if (!(vm_flags & VM_RAND_READ))
3166 			ra->size *= 2;
3167 		ra->async_size = HPAGE_PMD_NR;
3168 		page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3169 		return fpin;
3170 	}
3171 #endif
3172 
3173 	/* If we don't want any read-ahead, don't bother */
3174 	if (vm_flags & VM_RAND_READ)
3175 		return fpin;
3176 	if (!ra->ra_pages)
3177 		return fpin;
3178 
3179 	if (vm_flags & VM_SEQ_READ) {
3180 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3181 		page_cache_sync_ra(&ractl, ra->ra_pages);
3182 		return fpin;
3183 	}
3184 
3185 	/* Avoid banging the cache line if not needed */
3186 	mmap_miss = READ_ONCE(ra->mmap_miss);
3187 	if (mmap_miss < MMAP_LOTSAMISS * 10)
3188 		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3189 
3190 	/*
3191 	 * Do we miss much more than hit in this file? If so,
3192 	 * stop bothering with read-ahead. It will only hurt.
3193 	 */
3194 	if (mmap_miss > MMAP_LOTSAMISS)
3195 		return fpin;
3196 
3197 	/*
3198 	 * mmap read-around
3199 	 */
3200 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3201 	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3202 	ra->size = ra->ra_pages;
3203 	ra->async_size = ra->ra_pages / 4;
3204 	ractl._index = ra->start;
3205 	page_cache_ra_order(&ractl, ra, 0);
3206 	return fpin;
3207 }
3208 
3209 /*
3210  * Asynchronous readahead happens when we find the page and PG_readahead,
3211  * so we want to possibly extend the readahead further.  We return the file that
3212  * was pinned if we have to drop the mmap_lock in order to do IO.
3213  */
3214 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3215 					    struct folio *folio)
3216 {
3217 	struct file *file = vmf->vma->vm_file;
3218 	struct file_ra_state *ra = &file->f_ra;
3219 	DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3220 	struct file *fpin = NULL;
3221 	unsigned int mmap_miss;
3222 
3223 	/* If we don't want any read-ahead, don't bother */
3224 	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3225 		return fpin;
3226 
3227 	mmap_miss = READ_ONCE(ra->mmap_miss);
3228 	if (mmap_miss)
3229 		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3230 
3231 	if (folio_test_readahead(folio)) {
3232 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3233 		page_cache_async_ra(&ractl, folio, ra->ra_pages);
3234 	}
3235 	return fpin;
3236 }
3237 
3238 /**
3239  * filemap_fault - read in file data for page fault handling
3240  * @vmf:	struct vm_fault containing details of the fault
3241  *
3242  * filemap_fault() is invoked via the vma operations vector for a
3243  * mapped memory region to read in file data during a page fault.
3244  *
3245  * The goto's are kind of ugly, but this streamlines the normal case of having
3246  * it in the page cache, and handles the special cases reasonably without
3247  * having a lot of duplicated code.
3248  *
3249  * vma->vm_mm->mmap_lock must be held on entry.
3250  *
3251  * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3252  * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3253  *
3254  * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3255  * has not been released.
3256  *
3257  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3258  *
3259  * Return: bitwise-OR of %VM_FAULT_ codes.
3260  */
3261 vm_fault_t filemap_fault(struct vm_fault *vmf)
3262 {
3263 	int error;
3264 	struct file *file = vmf->vma->vm_file;
3265 	struct file *fpin = NULL;
3266 	struct address_space *mapping = file->f_mapping;
3267 	struct inode *inode = mapping->host;
3268 	pgoff_t max_idx, index = vmf->pgoff;
3269 	struct folio *folio;
3270 	vm_fault_t ret = 0;
3271 	bool mapping_locked = false;
3272 
3273 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3274 	if (unlikely(index >= max_idx))
3275 		return VM_FAULT_SIGBUS;
3276 
3277 	/*
3278 	 * Do we have something in the page cache already?
3279 	 */
3280 	folio = filemap_get_folio(mapping, index);
3281 	if (likely(!IS_ERR(folio))) {
3282 		/*
3283 		 * We found the page, so try async readahead before waiting for
3284 		 * the lock.
3285 		 */
3286 		if (!(vmf->flags & FAULT_FLAG_TRIED))
3287 			fpin = do_async_mmap_readahead(vmf, folio);
3288 		if (unlikely(!folio_test_uptodate(folio))) {
3289 			filemap_invalidate_lock_shared(mapping);
3290 			mapping_locked = true;
3291 		}
3292 	} else {
3293 		/* No page in the page cache at all */
3294 		count_vm_event(PGMAJFAULT);
3295 		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3296 		ret = VM_FAULT_MAJOR;
3297 		fpin = do_sync_mmap_readahead(vmf);
3298 retry_find:
3299 		/*
3300 		 * See comment in filemap_create_folio() why we need
3301 		 * invalidate_lock
3302 		 */
3303 		if (!mapping_locked) {
3304 			filemap_invalidate_lock_shared(mapping);
3305 			mapping_locked = true;
3306 		}
3307 		folio = __filemap_get_folio(mapping, index,
3308 					  FGP_CREAT|FGP_FOR_MMAP,
3309 					  vmf->gfp_mask);
3310 		if (IS_ERR(folio)) {
3311 			if (fpin)
3312 				goto out_retry;
3313 			filemap_invalidate_unlock_shared(mapping);
3314 			return VM_FAULT_OOM;
3315 		}
3316 	}
3317 
3318 	if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3319 		goto out_retry;
3320 
3321 	/* Did it get truncated? */
3322 	if (unlikely(folio->mapping != mapping)) {
3323 		folio_unlock(folio);
3324 		folio_put(folio);
3325 		goto retry_find;
3326 	}
3327 	VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3328 
3329 	/*
3330 	 * We have a locked page in the page cache, now we need to check
3331 	 * that it's up-to-date. If not, it is going to be due to an error.
3332 	 */
3333 	if (unlikely(!folio_test_uptodate(folio))) {
3334 		/*
3335 		 * The page was in cache and uptodate and now it is not.
3336 		 * Strange but possible since we didn't hold the page lock all
3337 		 * the time. Let's drop everything get the invalidate lock and
3338 		 * try again.
3339 		 */
3340 		if (!mapping_locked) {
3341 			folio_unlock(folio);
3342 			folio_put(folio);
3343 			goto retry_find;
3344 		}
3345 		goto page_not_uptodate;
3346 	}
3347 
3348 	/*
3349 	 * We've made it this far and we had to drop our mmap_lock, now is the
3350 	 * time to return to the upper layer and have it re-find the vma and
3351 	 * redo the fault.
3352 	 */
3353 	if (fpin) {
3354 		folio_unlock(folio);
3355 		goto out_retry;
3356 	}
3357 	if (mapping_locked)
3358 		filemap_invalidate_unlock_shared(mapping);
3359 
3360 	/*
3361 	 * Found the page and have a reference on it.
3362 	 * We must recheck i_size under page lock.
3363 	 */
3364 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3365 	if (unlikely(index >= max_idx)) {
3366 		folio_unlock(folio);
3367 		folio_put(folio);
3368 		return VM_FAULT_SIGBUS;
3369 	}
3370 
3371 	vmf->page = folio_file_page(folio, index);
3372 	return ret | VM_FAULT_LOCKED;
3373 
3374 page_not_uptodate:
3375 	/*
3376 	 * Umm, take care of errors if the page isn't up-to-date.
3377 	 * Try to re-read it _once_. We do this synchronously,
3378 	 * because there really aren't any performance issues here
3379 	 * and we need to check for errors.
3380 	 */
3381 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3382 	error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3383 	if (fpin)
3384 		goto out_retry;
3385 	folio_put(folio);
3386 
3387 	if (!error || error == AOP_TRUNCATED_PAGE)
3388 		goto retry_find;
3389 	filemap_invalidate_unlock_shared(mapping);
3390 
3391 	return VM_FAULT_SIGBUS;
3392 
3393 out_retry:
3394 	/*
3395 	 * We dropped the mmap_lock, we need to return to the fault handler to
3396 	 * re-find the vma and come back and find our hopefully still populated
3397 	 * page.
3398 	 */
3399 	if (!IS_ERR(folio))
3400 		folio_put(folio);
3401 	if (mapping_locked)
3402 		filemap_invalidate_unlock_shared(mapping);
3403 	if (fpin)
3404 		fput(fpin);
3405 	return ret | VM_FAULT_RETRY;
3406 }
3407 EXPORT_SYMBOL(filemap_fault);
3408 
3409 static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3410 		pgoff_t start)
3411 {
3412 	struct mm_struct *mm = vmf->vma->vm_mm;
3413 
3414 	/* Huge page is mapped? No need to proceed. */
3415 	if (pmd_trans_huge(*vmf->pmd)) {
3416 		folio_unlock(folio);
3417 		folio_put(folio);
3418 		return true;
3419 	}
3420 
3421 	if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3422 		struct page *page = folio_file_page(folio, start);
3423 		vm_fault_t ret = do_set_pmd(vmf, page);
3424 		if (!ret) {
3425 			/* The page is mapped successfully, reference consumed. */
3426 			folio_unlock(folio);
3427 			return true;
3428 		}
3429 	}
3430 
3431 	if (pmd_none(*vmf->pmd))
3432 		pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3433 
3434 	return false;
3435 }
3436 
3437 static struct folio *next_uptodate_page(struct folio *folio,
3438 				       struct address_space *mapping,
3439 				       struct xa_state *xas, pgoff_t end_pgoff)
3440 {
3441 	unsigned long max_idx;
3442 
3443 	do {
3444 		if (!folio)
3445 			return NULL;
3446 		if (xas_retry(xas, folio))
3447 			continue;
3448 		if (xa_is_value(folio))
3449 			continue;
3450 		if (folio_test_locked(folio))
3451 			continue;
3452 		if (!folio_try_get_rcu(folio))
3453 			continue;
3454 		/* Has the page moved or been split? */
3455 		if (unlikely(folio != xas_reload(xas)))
3456 			goto skip;
3457 		if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3458 			goto skip;
3459 		if (!folio_trylock(folio))
3460 			goto skip;
3461 		if (folio->mapping != mapping)
3462 			goto unlock;
3463 		if (!folio_test_uptodate(folio))
3464 			goto unlock;
3465 		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3466 		if (xas->xa_index >= max_idx)
3467 			goto unlock;
3468 		return folio;
3469 unlock:
3470 		folio_unlock(folio);
3471 skip:
3472 		folio_put(folio);
3473 	} while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3474 
3475 	return NULL;
3476 }
3477 
3478 static inline struct folio *first_map_page(struct address_space *mapping,
3479 					  struct xa_state *xas,
3480 					  pgoff_t end_pgoff)
3481 {
3482 	return next_uptodate_page(xas_find(xas, end_pgoff),
3483 				  mapping, xas, end_pgoff);
3484 }
3485 
3486 static inline struct folio *next_map_page(struct address_space *mapping,
3487 					 struct xa_state *xas,
3488 					 pgoff_t end_pgoff)
3489 {
3490 	return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3491 				  mapping, xas, end_pgoff);
3492 }
3493 
3494 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3495 			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3496 {
3497 	struct vm_area_struct *vma = vmf->vma;
3498 	struct file *file = vma->vm_file;
3499 	struct address_space *mapping = file->f_mapping;
3500 	pgoff_t last_pgoff = start_pgoff;
3501 	unsigned long addr;
3502 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3503 	struct folio *folio;
3504 	struct page *page;
3505 	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3506 	vm_fault_t ret = 0;
3507 
3508 	rcu_read_lock();
3509 	folio = first_map_page(mapping, &xas, end_pgoff);
3510 	if (!folio)
3511 		goto out;
3512 
3513 	if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3514 		ret = VM_FAULT_NOPAGE;
3515 		goto out;
3516 	}
3517 
3518 	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3519 	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3520 	if (!vmf->pte) {
3521 		folio_unlock(folio);
3522 		folio_put(folio);
3523 		goto out;
3524 	}
3525 	do {
3526 again:
3527 		page = folio_file_page(folio, xas.xa_index);
3528 		if (PageHWPoison(page))
3529 			goto unlock;
3530 
3531 		if (mmap_miss > 0)
3532 			mmap_miss--;
3533 
3534 		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3535 		vmf->pte += xas.xa_index - last_pgoff;
3536 		last_pgoff = xas.xa_index;
3537 
3538 		/*
3539 		 * NOTE: If there're PTE markers, we'll leave them to be
3540 		 * handled in the specific fault path, and it'll prohibit the
3541 		 * fault-around logic.
3542 		 */
3543 		if (!pte_none(ptep_get(vmf->pte)))
3544 			goto unlock;
3545 
3546 		/* We're about to handle the fault */
3547 		if (vmf->address == addr)
3548 			ret = VM_FAULT_NOPAGE;
3549 
3550 		do_set_pte(vmf, page, addr);
3551 		/* no need to invalidate: a not-present page won't be cached */
3552 		update_mmu_cache(vma, addr, vmf->pte);
3553 		if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3554 			xas.xa_index++;
3555 			folio_ref_inc(folio);
3556 			goto again;
3557 		}
3558 		folio_unlock(folio);
3559 		continue;
3560 unlock:
3561 		if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3562 			xas.xa_index++;
3563 			goto again;
3564 		}
3565 		folio_unlock(folio);
3566 		folio_put(folio);
3567 	} while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3568 	pte_unmap_unlock(vmf->pte, vmf->ptl);
3569 out:
3570 	rcu_read_unlock();
3571 	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3572 	return ret;
3573 }
3574 EXPORT_SYMBOL(filemap_map_pages);
3575 
3576 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3577 {
3578 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3579 	struct folio *folio = page_folio(vmf->page);
3580 	vm_fault_t ret = VM_FAULT_LOCKED;
3581 
3582 	sb_start_pagefault(mapping->host->i_sb);
3583 	file_update_time(vmf->vma->vm_file);
3584 	folio_lock(folio);
3585 	if (folio->mapping != mapping) {
3586 		folio_unlock(folio);
3587 		ret = VM_FAULT_NOPAGE;
3588 		goto out;
3589 	}
3590 	/*
3591 	 * We mark the folio dirty already here so that when freeze is in
3592 	 * progress, we are guaranteed that writeback during freezing will
3593 	 * see the dirty folio and writeprotect it again.
3594 	 */
3595 	folio_mark_dirty(folio);
3596 	folio_wait_stable(folio);
3597 out:
3598 	sb_end_pagefault(mapping->host->i_sb);
3599 	return ret;
3600 }
3601 
3602 const struct vm_operations_struct generic_file_vm_ops = {
3603 	.fault		= filemap_fault,
3604 	.map_pages	= filemap_map_pages,
3605 	.page_mkwrite	= filemap_page_mkwrite,
3606 };
3607 
3608 /* This is used for a general mmap of a disk file */
3609 
3610 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3611 {
3612 	struct address_space *mapping = file->f_mapping;
3613 
3614 	if (!mapping->a_ops->read_folio)
3615 		return -ENOEXEC;
3616 	file_accessed(file);
3617 	vma->vm_ops = &generic_file_vm_ops;
3618 	return 0;
3619 }
3620 
3621 /*
3622  * This is for filesystems which do not implement ->writepage.
3623  */
3624 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3625 {
3626 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3627 		return -EINVAL;
3628 	return generic_file_mmap(file, vma);
3629 }
3630 #else
3631 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3632 {
3633 	return VM_FAULT_SIGBUS;
3634 }
3635 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3636 {
3637 	return -ENOSYS;
3638 }
3639 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3640 {
3641 	return -ENOSYS;
3642 }
3643 #endif /* CONFIG_MMU */
3644 
3645 EXPORT_SYMBOL(filemap_page_mkwrite);
3646 EXPORT_SYMBOL(generic_file_mmap);
3647 EXPORT_SYMBOL(generic_file_readonly_mmap);
3648 
3649 static struct folio *do_read_cache_folio(struct address_space *mapping,
3650 		pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3651 {
3652 	struct folio *folio;
3653 	int err;
3654 
3655 	if (!filler)
3656 		filler = mapping->a_ops->read_folio;
3657 repeat:
3658 	folio = filemap_get_folio(mapping, index);
3659 	if (IS_ERR(folio)) {
3660 		folio = filemap_alloc_folio(gfp, 0);
3661 		if (!folio)
3662 			return ERR_PTR(-ENOMEM);
3663 		err = filemap_add_folio(mapping, folio, index, gfp);
3664 		if (unlikely(err)) {
3665 			folio_put(folio);
3666 			if (err == -EEXIST)
3667 				goto repeat;
3668 			/* Presumably ENOMEM for xarray node */
3669 			return ERR_PTR(err);
3670 		}
3671 
3672 		goto filler;
3673 	}
3674 	if (folio_test_uptodate(folio))
3675 		goto out;
3676 
3677 	if (!folio_trylock(folio)) {
3678 		folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3679 		goto repeat;
3680 	}
3681 
3682 	/* Folio was truncated from mapping */
3683 	if (!folio->mapping) {
3684 		folio_unlock(folio);
3685 		folio_put(folio);
3686 		goto repeat;
3687 	}
3688 
3689 	/* Someone else locked and filled the page in a very small window */
3690 	if (folio_test_uptodate(folio)) {
3691 		folio_unlock(folio);
3692 		goto out;
3693 	}
3694 
3695 filler:
3696 	err = filemap_read_folio(file, filler, folio);
3697 	if (err) {
3698 		folio_put(folio);
3699 		if (err == AOP_TRUNCATED_PAGE)
3700 			goto repeat;
3701 		return ERR_PTR(err);
3702 	}
3703 
3704 out:
3705 	folio_mark_accessed(folio);
3706 	return folio;
3707 }
3708 
3709 /**
3710  * read_cache_folio - Read into page cache, fill it if needed.
3711  * @mapping: The address_space to read from.
3712  * @index: The index to read.
3713  * @filler: Function to perform the read, or NULL to use aops->read_folio().
3714  * @file: Passed to filler function, may be NULL if not required.
3715  *
3716  * Read one page into the page cache.  If it succeeds, the folio returned
3717  * will contain @index, but it may not be the first page of the folio.
3718  *
3719  * If the filler function returns an error, it will be returned to the
3720  * caller.
3721  *
3722  * Context: May sleep.  Expects mapping->invalidate_lock to be held.
3723  * Return: An uptodate folio on success, ERR_PTR() on failure.
3724  */
3725 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3726 		filler_t filler, struct file *file)
3727 {
3728 	return do_read_cache_folio(mapping, index, filler, file,
3729 			mapping_gfp_mask(mapping));
3730 }
3731 EXPORT_SYMBOL(read_cache_folio);
3732 
3733 /**
3734  * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3735  * @mapping:	The address_space for the folio.
3736  * @index:	The index that the allocated folio will contain.
3737  * @gfp:	The page allocator flags to use if allocating.
3738  *
3739  * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3740  * any new memory allocations done using the specified allocation flags.
3741  *
3742  * The most likely error from this function is EIO, but ENOMEM is
3743  * possible and so is EINTR.  If ->read_folio returns another error,
3744  * that will be returned to the caller.
3745  *
3746  * The function expects mapping->invalidate_lock to be already held.
3747  *
3748  * Return: Uptodate folio on success, ERR_PTR() on failure.
3749  */
3750 struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3751 		pgoff_t index, gfp_t gfp)
3752 {
3753 	return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3754 }
3755 EXPORT_SYMBOL(mapping_read_folio_gfp);
3756 
3757 static struct page *do_read_cache_page(struct address_space *mapping,
3758 		pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3759 {
3760 	struct folio *folio;
3761 
3762 	folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3763 	if (IS_ERR(folio))
3764 		return &folio->page;
3765 	return folio_file_page(folio, index);
3766 }
3767 
3768 struct page *read_cache_page(struct address_space *mapping,
3769 			pgoff_t index, filler_t *filler, struct file *file)
3770 {
3771 	return do_read_cache_page(mapping, index, filler, file,
3772 			mapping_gfp_mask(mapping));
3773 }
3774 EXPORT_SYMBOL(read_cache_page);
3775 
3776 /**
3777  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3778  * @mapping:	the page's address_space
3779  * @index:	the page index
3780  * @gfp:	the page allocator flags to use if allocating
3781  *
3782  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3783  * any new page allocations done using the specified allocation flags.
3784  *
3785  * If the page does not get brought uptodate, return -EIO.
3786  *
3787  * The function expects mapping->invalidate_lock to be already held.
3788  *
3789  * Return: up to date page on success, ERR_PTR() on failure.
3790  */
3791 struct page *read_cache_page_gfp(struct address_space *mapping,
3792 				pgoff_t index,
3793 				gfp_t gfp)
3794 {
3795 	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3796 }
3797 EXPORT_SYMBOL(read_cache_page_gfp);
3798 
3799 /*
3800  * Warn about a page cache invalidation failure during a direct I/O write.
3801  */
3802 static void dio_warn_stale_pagecache(struct file *filp)
3803 {
3804 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3805 	char pathname[128];
3806 	char *path;
3807 
3808 	errseq_set(&filp->f_mapping->wb_err, -EIO);
3809 	if (__ratelimit(&_rs)) {
3810 		path = file_path(filp, pathname, sizeof(pathname));
3811 		if (IS_ERR(path))
3812 			path = "(unknown)";
3813 		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3814 		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3815 			current->comm);
3816 	}
3817 }
3818 
3819 void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
3820 {
3821 	struct address_space *mapping = iocb->ki_filp->f_mapping;
3822 
3823 	if (mapping->nrpages &&
3824 	    invalidate_inode_pages2_range(mapping,
3825 			iocb->ki_pos >> PAGE_SHIFT,
3826 			(iocb->ki_pos + count - 1) >> PAGE_SHIFT))
3827 		dio_warn_stale_pagecache(iocb->ki_filp);
3828 }
3829 
3830 ssize_t
3831 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3832 {
3833 	struct address_space *mapping = iocb->ki_filp->f_mapping;
3834 	size_t write_len = iov_iter_count(from);
3835 	ssize_t written;
3836 
3837 	/*
3838 	 * If a page can not be invalidated, return 0 to fall back
3839 	 * to buffered write.
3840 	 */
3841 	written = kiocb_invalidate_pages(iocb, write_len);
3842 	if (written) {
3843 		if (written == -EBUSY)
3844 			return 0;
3845 		return written;
3846 	}
3847 
3848 	written = mapping->a_ops->direct_IO(iocb, from);
3849 
3850 	/*
3851 	 * Finally, try again to invalidate clean pages which might have been
3852 	 * cached by non-direct readahead, or faulted in by get_user_pages()
3853 	 * if the source of the write was an mmap'ed region of the file
3854 	 * we're writing.  Either one is a pretty crazy thing to do,
3855 	 * so we don't support it 100%.  If this invalidation
3856 	 * fails, tough, the write still worked...
3857 	 *
3858 	 * Most of the time we do not need this since dio_complete() will do
3859 	 * the invalidation for us. However there are some file systems that
3860 	 * do not end up with dio_complete() being called, so let's not break
3861 	 * them by removing it completely.
3862 	 *
3863 	 * Noticeable example is a blkdev_direct_IO().
3864 	 *
3865 	 * Skip invalidation for async writes or if mapping has no pages.
3866 	 */
3867 	if (written > 0) {
3868 		struct inode *inode = mapping->host;
3869 		loff_t pos = iocb->ki_pos;
3870 
3871 		kiocb_invalidate_post_direct_write(iocb, written);
3872 		pos += written;
3873 		write_len -= written;
3874 		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3875 			i_size_write(inode, pos);
3876 			mark_inode_dirty(inode);
3877 		}
3878 		iocb->ki_pos = pos;
3879 	}
3880 	if (written != -EIOCBQUEUED)
3881 		iov_iter_revert(from, write_len - iov_iter_count(from));
3882 	return written;
3883 }
3884 EXPORT_SYMBOL(generic_file_direct_write);
3885 
3886 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3887 {
3888 	struct file *file = iocb->ki_filp;
3889 	loff_t pos = iocb->ki_pos;
3890 	struct address_space *mapping = file->f_mapping;
3891 	const struct address_space_operations *a_ops = mapping->a_ops;
3892 	long status = 0;
3893 	ssize_t written = 0;
3894 
3895 	do {
3896 		struct page *page;
3897 		unsigned long offset;	/* Offset into pagecache page */
3898 		unsigned long bytes;	/* Bytes to write to page */
3899 		size_t copied;		/* Bytes copied from user */
3900 		void *fsdata = NULL;
3901 
3902 		offset = (pos & (PAGE_SIZE - 1));
3903 		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3904 						iov_iter_count(i));
3905 
3906 again:
3907 		/*
3908 		 * Bring in the user page that we will copy from _first_.
3909 		 * Otherwise there's a nasty deadlock on copying from the
3910 		 * same page as we're writing to, without it being marked
3911 		 * up-to-date.
3912 		 */
3913 		if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3914 			status = -EFAULT;
3915 			break;
3916 		}
3917 
3918 		if (fatal_signal_pending(current)) {
3919 			status = -EINTR;
3920 			break;
3921 		}
3922 
3923 		status = a_ops->write_begin(file, mapping, pos, bytes,
3924 						&page, &fsdata);
3925 		if (unlikely(status < 0))
3926 			break;
3927 
3928 		if (mapping_writably_mapped(mapping))
3929 			flush_dcache_page(page);
3930 
3931 		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3932 		flush_dcache_page(page);
3933 
3934 		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3935 						page, fsdata);
3936 		if (unlikely(status != copied)) {
3937 			iov_iter_revert(i, copied - max(status, 0L));
3938 			if (unlikely(status < 0))
3939 				break;
3940 		}
3941 		cond_resched();
3942 
3943 		if (unlikely(status == 0)) {
3944 			/*
3945 			 * A short copy made ->write_end() reject the
3946 			 * thing entirely.  Might be memory poisoning
3947 			 * halfway through, might be a race with munmap,
3948 			 * might be severe memory pressure.
3949 			 */
3950 			if (copied)
3951 				bytes = copied;
3952 			goto again;
3953 		}
3954 		pos += status;
3955 		written += status;
3956 
3957 		balance_dirty_pages_ratelimited(mapping);
3958 	} while (iov_iter_count(i));
3959 
3960 	if (!written)
3961 		return status;
3962 	iocb->ki_pos += written;
3963 	return written;
3964 }
3965 EXPORT_SYMBOL(generic_perform_write);
3966 
3967 /**
3968  * __generic_file_write_iter - write data to a file
3969  * @iocb:	IO state structure (file, offset, etc.)
3970  * @from:	iov_iter with data to write
3971  *
3972  * This function does all the work needed for actually writing data to a
3973  * file. It does all basic checks, removes SUID from the file, updates
3974  * modification times and calls proper subroutines depending on whether we
3975  * do direct IO or a standard buffered write.
3976  *
3977  * It expects i_rwsem to be grabbed unless we work on a block device or similar
3978  * object which does not need locking at all.
3979  *
3980  * This function does *not* take care of syncing data in case of O_SYNC write.
3981  * A caller has to handle it. This is mainly due to the fact that we want to
3982  * avoid syncing under i_rwsem.
3983  *
3984  * Return:
3985  * * number of bytes written, even for truncated writes
3986  * * negative error code if no data has been written at all
3987  */
3988 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3989 {
3990 	struct file *file = iocb->ki_filp;
3991 	struct address_space *mapping = file->f_mapping;
3992 	struct inode *inode = mapping->host;
3993 	ssize_t ret;
3994 
3995 	ret = file_remove_privs(file);
3996 	if (ret)
3997 		return ret;
3998 
3999 	ret = file_update_time(file);
4000 	if (ret)
4001 		return ret;
4002 
4003 	if (iocb->ki_flags & IOCB_DIRECT) {
4004 		ret = generic_file_direct_write(iocb, from);
4005 		/*
4006 		 * If the write stopped short of completing, fall back to
4007 		 * buffered writes.  Some filesystems do this for writes to
4008 		 * holes, for example.  For DAX files, a buffered write will
4009 		 * not succeed (even if it did, DAX does not handle dirty
4010 		 * page-cache pages correctly).
4011 		 */
4012 		if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4013 			return ret;
4014 		return direct_write_fallback(iocb, from, ret,
4015 				generic_perform_write(iocb, from));
4016 	}
4017 
4018 	return generic_perform_write(iocb, from);
4019 }
4020 EXPORT_SYMBOL(__generic_file_write_iter);
4021 
4022 /**
4023  * generic_file_write_iter - write data to a file
4024  * @iocb:	IO state structure
4025  * @from:	iov_iter with data to write
4026  *
4027  * This is a wrapper around __generic_file_write_iter() to be used by most
4028  * filesystems. It takes care of syncing the file in case of O_SYNC file
4029  * and acquires i_rwsem as needed.
4030  * Return:
4031  * * negative error code if no data has been written at all of
4032  *   vfs_fsync_range() failed for a synchronous write
4033  * * number of bytes written, even for truncated writes
4034  */
4035 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4036 {
4037 	struct file *file = iocb->ki_filp;
4038 	struct inode *inode = file->f_mapping->host;
4039 	ssize_t ret;
4040 
4041 	inode_lock(inode);
4042 	ret = generic_write_checks(iocb, from);
4043 	if (ret > 0)
4044 		ret = __generic_file_write_iter(iocb, from);
4045 	inode_unlock(inode);
4046 
4047 	if (ret > 0)
4048 		ret = generic_write_sync(iocb, ret);
4049 	return ret;
4050 }
4051 EXPORT_SYMBOL(generic_file_write_iter);
4052 
4053 /**
4054  * filemap_release_folio() - Release fs-specific metadata on a folio.
4055  * @folio: The folio which the kernel is trying to free.
4056  * @gfp: Memory allocation flags (and I/O mode).
4057  *
4058  * The address_space is trying to release any data attached to a folio
4059  * (presumably at folio->private).
4060  *
4061  * This will also be called if the private_2 flag is set on a page,
4062  * indicating that the folio has other metadata associated with it.
4063  *
4064  * The @gfp argument specifies whether I/O may be performed to release
4065  * this page (__GFP_IO), and whether the call may block
4066  * (__GFP_RECLAIM & __GFP_FS).
4067  *
4068  * Return: %true if the release was successful, otherwise %false.
4069  */
4070 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4071 {
4072 	struct address_space * const mapping = folio->mapping;
4073 
4074 	BUG_ON(!folio_test_locked(folio));
4075 	if (folio_test_writeback(folio))
4076 		return false;
4077 
4078 	if (mapping && mapping->a_ops->release_folio)
4079 		return mapping->a_ops->release_folio(folio, gfp);
4080 	return try_to_free_buffers(folio);
4081 }
4082 EXPORT_SYMBOL(filemap_release_folio);
4083 
4084 #ifdef CONFIG_CACHESTAT_SYSCALL
4085 /**
4086  * filemap_cachestat() - compute the page cache statistics of a mapping
4087  * @mapping:	The mapping to compute the statistics for.
4088  * @first_index:	The starting page cache index.
4089  * @last_index:	The final page index (inclusive).
4090  * @cs:	the cachestat struct to write the result to.
4091  *
4092  * This will query the page cache statistics of a mapping in the
4093  * page range of [first_index, last_index] (inclusive). The statistics
4094  * queried include: number of dirty pages, number of pages marked for
4095  * writeback, and the number of (recently) evicted pages.
4096  */
4097 static void filemap_cachestat(struct address_space *mapping,
4098 		pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4099 {
4100 	XA_STATE(xas, &mapping->i_pages, first_index);
4101 	struct folio *folio;
4102 
4103 	rcu_read_lock();
4104 	xas_for_each(&xas, folio, last_index) {
4105 		unsigned long nr_pages;
4106 		pgoff_t folio_first_index, folio_last_index;
4107 
4108 		if (xas_retry(&xas, folio))
4109 			continue;
4110 
4111 		if (xa_is_value(folio)) {
4112 			/* page is evicted */
4113 			void *shadow = (void *)folio;
4114 			bool workingset; /* not used */
4115 			int order = xa_get_order(xas.xa, xas.xa_index);
4116 
4117 			nr_pages = 1 << order;
4118 			folio_first_index = round_down(xas.xa_index, 1 << order);
4119 			folio_last_index = folio_first_index + nr_pages - 1;
4120 
4121 			/* Folios might straddle the range boundaries, only count covered pages */
4122 			if (folio_first_index < first_index)
4123 				nr_pages -= first_index - folio_first_index;
4124 
4125 			if (folio_last_index > last_index)
4126 				nr_pages -= folio_last_index - last_index;
4127 
4128 			cs->nr_evicted += nr_pages;
4129 
4130 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4131 			if (shmem_mapping(mapping)) {
4132 				/* shmem file - in swap cache */
4133 				swp_entry_t swp = radix_to_swp_entry(folio);
4134 
4135 				shadow = get_shadow_from_swap_cache(swp);
4136 			}
4137 #endif
4138 			if (workingset_test_recent(shadow, true, &workingset))
4139 				cs->nr_recently_evicted += nr_pages;
4140 
4141 			goto resched;
4142 		}
4143 
4144 		nr_pages = folio_nr_pages(folio);
4145 		folio_first_index = folio_pgoff(folio);
4146 		folio_last_index = folio_first_index + nr_pages - 1;
4147 
4148 		/* Folios might straddle the range boundaries, only count covered pages */
4149 		if (folio_first_index < first_index)
4150 			nr_pages -= first_index - folio_first_index;
4151 
4152 		if (folio_last_index > last_index)
4153 			nr_pages -= folio_last_index - last_index;
4154 
4155 		/* page is in cache */
4156 		cs->nr_cache += nr_pages;
4157 
4158 		if (folio_test_dirty(folio))
4159 			cs->nr_dirty += nr_pages;
4160 
4161 		if (folio_test_writeback(folio))
4162 			cs->nr_writeback += nr_pages;
4163 
4164 resched:
4165 		if (need_resched()) {
4166 			xas_pause(&xas);
4167 			cond_resched_rcu();
4168 		}
4169 	}
4170 	rcu_read_unlock();
4171 }
4172 
4173 /*
4174  * The cachestat(2) system call.
4175  *
4176  * cachestat() returns the page cache statistics of a file in the
4177  * bytes range specified by `off` and `len`: number of cached pages,
4178  * number of dirty pages, number of pages marked for writeback,
4179  * number of evicted pages, and number of recently evicted pages.
4180  *
4181  * An evicted page is a page that is previously in the page cache
4182  * but has been evicted since. A page is recently evicted if its last
4183  * eviction was recent enough that its reentry to the cache would
4184  * indicate that it is actively being used by the system, and that
4185  * there is memory pressure on the system.
4186  *
4187  * `off` and `len` must be non-negative integers. If `len` > 0,
4188  * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4189  * we will query in the range from `off` to the end of the file.
4190  *
4191  * The `flags` argument is unused for now, but is included for future
4192  * extensibility. User should pass 0 (i.e no flag specified).
4193  *
4194  * Currently, hugetlbfs is not supported.
4195  *
4196  * Because the status of a page can change after cachestat() checks it
4197  * but before it returns to the application, the returned values may
4198  * contain stale information.
4199  *
4200  * return values:
4201  *  zero        - success
4202  *  -EFAULT     - cstat or cstat_range points to an illegal address
4203  *  -EINVAL     - invalid flags
4204  *  -EBADF      - invalid file descriptor
4205  *  -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4206  */
4207 SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4208 		struct cachestat_range __user *, cstat_range,
4209 		struct cachestat __user *, cstat, unsigned int, flags)
4210 {
4211 	struct fd f = fdget(fd);
4212 	struct address_space *mapping;
4213 	struct cachestat_range csr;
4214 	struct cachestat cs;
4215 	pgoff_t first_index, last_index;
4216 
4217 	if (!f.file)
4218 		return -EBADF;
4219 
4220 	if (copy_from_user(&csr, cstat_range,
4221 			sizeof(struct cachestat_range))) {
4222 		fdput(f);
4223 		return -EFAULT;
4224 	}
4225 
4226 	/* hugetlbfs is not supported */
4227 	if (is_file_hugepages(f.file)) {
4228 		fdput(f);
4229 		return -EOPNOTSUPP;
4230 	}
4231 
4232 	if (flags != 0) {
4233 		fdput(f);
4234 		return -EINVAL;
4235 	}
4236 
4237 	first_index = csr.off >> PAGE_SHIFT;
4238 	last_index =
4239 		csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4240 	memset(&cs, 0, sizeof(struct cachestat));
4241 	mapping = f.file->f_mapping;
4242 	filemap_cachestat(mapping, first_index, last_index, &cs);
4243 	fdput(f);
4244 
4245 	if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4246 		return -EFAULT;
4247 
4248 	return 0;
4249 }
4250 #endif /* CONFIG_CACHESTAT_SYSCALL */
4251