xref: /openbmc/linux/mm/filemap.c (revision e7f127b2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	linux/mm/filemap.c
4  *
5  * Copyright (C) 1994-1999  Linus Torvalds
6  */
7 
8 /*
9  * This file handles the generic file mmap semantics used by
10  * most "normal" filesystems (but you don't /have/ to use this:
11  * the NFS filesystem used to do this differently, for example)
12  */
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
16 #include <linux/fs.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
22 #include <linux/mm.h>
23 #include <linux/swap.h>
24 #include <linux/swapops.h>
25 #include <linux/mman.h>
26 #include <linux/pagemap.h>
27 #include <linux/file.h>
28 #include <linux/uio.h>
29 #include <linux/error-injection.h>
30 #include <linux/hash.h>
31 #include <linux/writeback.h>
32 #include <linux/backing-dev.h>
33 #include <linux/pagevec.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <linux/migrate.h>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
47 #include "internal.h"
48 
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
51 
52 /*
53  * FIXME: remove all knowledge of the buffer layer from the core VM
54  */
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
56 
57 #include <asm/mman.h>
58 
59 /*
60  * Shared mappings implemented 30.11.1994. It's not fully working yet,
61  * though.
62  *
63  * Shared mappings now work. 15.8.1995  Bruno.
64  *
65  * finished 'unifying' the page and buffer cache and SMP-threaded the
66  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
67  *
68  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
69  */
70 
71 /*
72  * Lock ordering:
73  *
74  *  ->i_mmap_rwsem		(truncate_pagecache)
75  *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
76  *      ->swap_lock		(exclusive_swap_page, others)
77  *        ->i_pages lock
78  *
79  *  ->i_rwsem
80  *    ->invalidate_lock		(acquired by fs in truncate path)
81  *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
82  *
83  *  ->mmap_lock
84  *    ->i_mmap_rwsem
85  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
86  *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
87  *
88  *  ->mmap_lock
89  *    ->invalidate_lock		(filemap_fault)
90  *      ->lock_page		(filemap_fault, access_process_vm)
91  *
92  *  ->i_rwsem			(generic_perform_write)
93  *    ->mmap_lock		(fault_in_readable->do_page_fault)
94  *
95  *  bdi->wb.list_lock
96  *    sb_lock			(fs/fs-writeback.c)
97  *    ->i_pages lock		(__sync_single_inode)
98  *
99  *  ->i_mmap_rwsem
100  *    ->anon_vma.lock		(vma_adjust)
101  *
102  *  ->anon_vma.lock
103  *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
104  *
105  *  ->page_table_lock or pte_lock
106  *    ->swap_lock		(try_to_unmap_one)
107  *    ->private_lock		(try_to_unmap_one)
108  *    ->i_pages lock		(try_to_unmap_one)
109  *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
110  *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
111  *    ->private_lock		(page_remove_rmap->set_page_dirty)
112  *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
113  *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
114  *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
115  *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
116  *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
117  *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
118  *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
119  *
120  * ->i_mmap_rwsem
121  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
122  */
123 
124 static void page_cache_delete(struct address_space *mapping,
125 				   struct folio *folio, void *shadow)
126 {
127 	XA_STATE(xas, &mapping->i_pages, folio->index);
128 	long nr = 1;
129 
130 	mapping_set_update(&xas, mapping);
131 
132 	/* hugetlb pages are represented by a single entry in the xarray */
133 	if (!folio_test_hugetlb(folio)) {
134 		xas_set_order(&xas, folio->index, folio_order(folio));
135 		nr = folio_nr_pages(folio);
136 	}
137 
138 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
139 
140 	xas_store(&xas, shadow);
141 	xas_init_marks(&xas);
142 
143 	folio->mapping = NULL;
144 	/* Leave page->index set: truncation lookup relies upon it */
145 	mapping->nrpages -= nr;
146 }
147 
148 static void filemap_unaccount_folio(struct address_space *mapping,
149 		struct folio *folio)
150 {
151 	long nr;
152 
153 	VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
154 	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
155 		int mapcount;
156 
157 		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
158 			 current->comm, folio_pfn(folio));
159 		dump_page(&folio->page, "still mapped when deleted");
160 		dump_stack();
161 		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
162 
163 		mapcount = page_mapcount(&folio->page);
164 		if (mapping_exiting(mapping) &&
165 		    folio_ref_count(folio) >= mapcount + 2) {
166 			/*
167 			 * All vmas have already been torn down, so it's
168 			 * a good bet that actually the folio is unmapped,
169 			 * and we'd prefer not to leak it: if we're wrong,
170 			 * some other bad page check should catch it later.
171 			 */
172 			page_mapcount_reset(&folio->page);
173 			folio_ref_sub(folio, mapcount);
174 		}
175 	}
176 
177 	/* hugetlb folios do not participate in page cache accounting. */
178 	if (folio_test_hugetlb(folio))
179 		return;
180 
181 	nr = folio_nr_pages(folio);
182 
183 	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
184 	if (folio_test_swapbacked(folio)) {
185 		__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
186 		if (folio_test_pmd_mappable(folio))
187 			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
188 	} else if (folio_test_pmd_mappable(folio)) {
189 		__lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
190 		filemap_nr_thps_dec(mapping);
191 	}
192 
193 	/*
194 	 * At this point folio must be either written or cleaned by
195 	 * truncate.  Dirty folio here signals a bug and loss of
196 	 * unwritten data.
197 	 *
198 	 * This fixes dirty accounting after removing the folio entirely
199 	 * but leaves the dirty flag set: it has no effect for truncated
200 	 * folio and anyway will be cleared before returning folio to
201 	 * buddy allocator.
202 	 */
203 	if (WARN_ON_ONCE(folio_test_dirty(folio)))
204 		folio_account_cleaned(folio, mapping,
205 					inode_to_wb(mapping->host));
206 }
207 
208 /*
209  * Delete a page from the page cache and free it. Caller has to make
210  * sure the page is locked and that nobody else uses it - or that usage
211  * is safe.  The caller must hold the i_pages lock.
212  */
213 void __filemap_remove_folio(struct folio *folio, void *shadow)
214 {
215 	struct address_space *mapping = folio->mapping;
216 
217 	trace_mm_filemap_delete_from_page_cache(folio);
218 	filemap_unaccount_folio(mapping, folio);
219 	page_cache_delete(mapping, folio, shadow);
220 }
221 
222 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
223 {
224 	void (*freepage)(struct page *);
225 	int refs = 1;
226 
227 	freepage = mapping->a_ops->freepage;
228 	if (freepage)
229 		freepage(&folio->page);
230 
231 	if (folio_test_large(folio) && !folio_test_hugetlb(folio))
232 		refs = folio_nr_pages(folio);
233 	folio_put_refs(folio, refs);
234 }
235 
236 /**
237  * filemap_remove_folio - Remove folio from page cache.
238  * @folio: The folio.
239  *
240  * This must be called only on folios that are locked and have been
241  * verified to be in the page cache.  It will never put the folio into
242  * the free list because the caller has a reference on the page.
243  */
244 void filemap_remove_folio(struct folio *folio)
245 {
246 	struct address_space *mapping = folio->mapping;
247 
248 	BUG_ON(!folio_test_locked(folio));
249 	spin_lock(&mapping->host->i_lock);
250 	xa_lock_irq(&mapping->i_pages);
251 	__filemap_remove_folio(folio, NULL);
252 	xa_unlock_irq(&mapping->i_pages);
253 	if (mapping_shrinkable(mapping))
254 		inode_add_lru(mapping->host);
255 	spin_unlock(&mapping->host->i_lock);
256 
257 	filemap_free_folio(mapping, folio);
258 }
259 
260 /*
261  * page_cache_delete_batch - delete several folios from page cache
262  * @mapping: the mapping to which folios belong
263  * @fbatch: batch of folios to delete
264  *
265  * The function walks over mapping->i_pages and removes folios passed in
266  * @fbatch from the mapping. The function expects @fbatch to be sorted
267  * by page index and is optimised for it to be dense.
268  * It tolerates holes in @fbatch (mapping entries at those indices are not
269  * modified).
270  *
271  * The function expects the i_pages lock to be held.
272  */
273 static void page_cache_delete_batch(struct address_space *mapping,
274 			     struct folio_batch *fbatch)
275 {
276 	XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
277 	long total_pages = 0;
278 	int i = 0;
279 	struct folio *folio;
280 
281 	mapping_set_update(&xas, mapping);
282 	xas_for_each(&xas, folio, ULONG_MAX) {
283 		if (i >= folio_batch_count(fbatch))
284 			break;
285 
286 		/* A swap/dax/shadow entry got inserted? Skip it. */
287 		if (xa_is_value(folio))
288 			continue;
289 		/*
290 		 * A page got inserted in our range? Skip it. We have our
291 		 * pages locked so they are protected from being removed.
292 		 * If we see a page whose index is higher than ours, it
293 		 * means our page has been removed, which shouldn't be
294 		 * possible because we're holding the PageLock.
295 		 */
296 		if (folio != fbatch->folios[i]) {
297 			VM_BUG_ON_FOLIO(folio->index >
298 					fbatch->folios[i]->index, folio);
299 			continue;
300 		}
301 
302 		WARN_ON_ONCE(!folio_test_locked(folio));
303 
304 		folio->mapping = NULL;
305 		/* Leave folio->index set: truncation lookup relies on it */
306 
307 		i++;
308 		xas_store(&xas, NULL);
309 		total_pages += folio_nr_pages(folio);
310 	}
311 	mapping->nrpages -= total_pages;
312 }
313 
314 void delete_from_page_cache_batch(struct address_space *mapping,
315 				  struct folio_batch *fbatch)
316 {
317 	int i;
318 
319 	if (!folio_batch_count(fbatch))
320 		return;
321 
322 	spin_lock(&mapping->host->i_lock);
323 	xa_lock_irq(&mapping->i_pages);
324 	for (i = 0; i < folio_batch_count(fbatch); i++) {
325 		struct folio *folio = fbatch->folios[i];
326 
327 		trace_mm_filemap_delete_from_page_cache(folio);
328 		filemap_unaccount_folio(mapping, folio);
329 	}
330 	page_cache_delete_batch(mapping, fbatch);
331 	xa_unlock_irq(&mapping->i_pages);
332 	if (mapping_shrinkable(mapping))
333 		inode_add_lru(mapping->host);
334 	spin_unlock(&mapping->host->i_lock);
335 
336 	for (i = 0; i < folio_batch_count(fbatch); i++)
337 		filemap_free_folio(mapping, fbatch->folios[i]);
338 }
339 
340 int filemap_check_errors(struct address_space *mapping)
341 {
342 	int ret = 0;
343 	/* Check for outstanding write errors */
344 	if (test_bit(AS_ENOSPC, &mapping->flags) &&
345 	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
346 		ret = -ENOSPC;
347 	if (test_bit(AS_EIO, &mapping->flags) &&
348 	    test_and_clear_bit(AS_EIO, &mapping->flags))
349 		ret = -EIO;
350 	return ret;
351 }
352 EXPORT_SYMBOL(filemap_check_errors);
353 
354 static int filemap_check_and_keep_errors(struct address_space *mapping)
355 {
356 	/* Check for outstanding write errors */
357 	if (test_bit(AS_EIO, &mapping->flags))
358 		return -EIO;
359 	if (test_bit(AS_ENOSPC, &mapping->flags))
360 		return -ENOSPC;
361 	return 0;
362 }
363 
364 /**
365  * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
366  * @mapping:	address space structure to write
367  * @wbc:	the writeback_control controlling the writeout
368  *
369  * Call writepages on the mapping using the provided wbc to control the
370  * writeout.
371  *
372  * Return: %0 on success, negative error code otherwise.
373  */
374 int filemap_fdatawrite_wbc(struct address_space *mapping,
375 			   struct writeback_control *wbc)
376 {
377 	int ret;
378 
379 	if (!mapping_can_writeback(mapping) ||
380 	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
381 		return 0;
382 
383 	wbc_attach_fdatawrite_inode(wbc, mapping->host);
384 	ret = do_writepages(mapping, wbc);
385 	wbc_detach_inode(wbc);
386 	return ret;
387 }
388 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
389 
390 /**
391  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
392  * @mapping:	address space structure to write
393  * @start:	offset in bytes where the range starts
394  * @end:	offset in bytes where the range ends (inclusive)
395  * @sync_mode:	enable synchronous operation
396  *
397  * Start writeback against all of a mapping's dirty pages that lie
398  * within the byte offsets <start, end> inclusive.
399  *
400  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
401  * opposed to a regular memory cleansing writeback.  The difference between
402  * these two operations is that if a dirty page/buffer is encountered, it must
403  * be waited upon, and not just skipped over.
404  *
405  * Return: %0 on success, negative error code otherwise.
406  */
407 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
408 				loff_t end, int sync_mode)
409 {
410 	struct writeback_control wbc = {
411 		.sync_mode = sync_mode,
412 		.nr_to_write = LONG_MAX,
413 		.range_start = start,
414 		.range_end = end,
415 	};
416 
417 	return filemap_fdatawrite_wbc(mapping, &wbc);
418 }
419 
420 static inline int __filemap_fdatawrite(struct address_space *mapping,
421 	int sync_mode)
422 {
423 	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
424 }
425 
426 int filemap_fdatawrite(struct address_space *mapping)
427 {
428 	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
429 }
430 EXPORT_SYMBOL(filemap_fdatawrite);
431 
432 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
433 				loff_t end)
434 {
435 	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
436 }
437 EXPORT_SYMBOL(filemap_fdatawrite_range);
438 
439 /**
440  * filemap_flush - mostly a non-blocking flush
441  * @mapping:	target address_space
442  *
443  * This is a mostly non-blocking flush.  Not suitable for data-integrity
444  * purposes - I/O may not be started against all dirty pages.
445  *
446  * Return: %0 on success, negative error code otherwise.
447  */
448 int filemap_flush(struct address_space *mapping)
449 {
450 	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
451 }
452 EXPORT_SYMBOL(filemap_flush);
453 
454 /**
455  * filemap_range_has_page - check if a page exists in range.
456  * @mapping:           address space within which to check
457  * @start_byte:        offset in bytes where the range starts
458  * @end_byte:          offset in bytes where the range ends (inclusive)
459  *
460  * Find at least one page in the range supplied, usually used to check if
461  * direct writing in this range will trigger a writeback.
462  *
463  * Return: %true if at least one page exists in the specified range,
464  * %false otherwise.
465  */
466 bool filemap_range_has_page(struct address_space *mapping,
467 			   loff_t start_byte, loff_t end_byte)
468 {
469 	struct page *page;
470 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
471 	pgoff_t max = end_byte >> PAGE_SHIFT;
472 
473 	if (end_byte < start_byte)
474 		return false;
475 
476 	rcu_read_lock();
477 	for (;;) {
478 		page = xas_find(&xas, max);
479 		if (xas_retry(&xas, page))
480 			continue;
481 		/* Shadow entries don't count */
482 		if (xa_is_value(page))
483 			continue;
484 		/*
485 		 * We don't need to try to pin this page; we're about to
486 		 * release the RCU lock anyway.  It is enough to know that
487 		 * there was a page here recently.
488 		 */
489 		break;
490 	}
491 	rcu_read_unlock();
492 
493 	return page != NULL;
494 }
495 EXPORT_SYMBOL(filemap_range_has_page);
496 
497 static void __filemap_fdatawait_range(struct address_space *mapping,
498 				     loff_t start_byte, loff_t end_byte)
499 {
500 	pgoff_t index = start_byte >> PAGE_SHIFT;
501 	pgoff_t end = end_byte >> PAGE_SHIFT;
502 	struct pagevec pvec;
503 	int nr_pages;
504 
505 	if (end_byte < start_byte)
506 		return;
507 
508 	pagevec_init(&pvec);
509 	while (index <= end) {
510 		unsigned i;
511 
512 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
513 				end, PAGECACHE_TAG_WRITEBACK);
514 		if (!nr_pages)
515 			break;
516 
517 		for (i = 0; i < nr_pages; i++) {
518 			struct page *page = pvec.pages[i];
519 
520 			wait_on_page_writeback(page);
521 			ClearPageError(page);
522 		}
523 		pagevec_release(&pvec);
524 		cond_resched();
525 	}
526 }
527 
528 /**
529  * filemap_fdatawait_range - wait for writeback to complete
530  * @mapping:		address space structure to wait for
531  * @start_byte:		offset in bytes where the range starts
532  * @end_byte:		offset in bytes where the range ends (inclusive)
533  *
534  * Walk the list of under-writeback pages of the given address space
535  * in the given range and wait for all of them.  Check error status of
536  * the address space and return it.
537  *
538  * Since the error status of the address space is cleared by this function,
539  * callers are responsible for checking the return value and handling and/or
540  * reporting the error.
541  *
542  * Return: error status of the address space.
543  */
544 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
545 			    loff_t end_byte)
546 {
547 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
548 	return filemap_check_errors(mapping);
549 }
550 EXPORT_SYMBOL(filemap_fdatawait_range);
551 
552 /**
553  * filemap_fdatawait_range_keep_errors - wait for writeback to complete
554  * @mapping:		address space structure to wait for
555  * @start_byte:		offset in bytes where the range starts
556  * @end_byte:		offset in bytes where the range ends (inclusive)
557  *
558  * Walk the list of under-writeback pages of the given address space in the
559  * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
560  * this function does not clear error status of the address space.
561  *
562  * Use this function if callers don't handle errors themselves.  Expected
563  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
564  * fsfreeze(8)
565  */
566 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
567 		loff_t start_byte, loff_t end_byte)
568 {
569 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
570 	return filemap_check_and_keep_errors(mapping);
571 }
572 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
573 
574 /**
575  * file_fdatawait_range - wait for writeback to complete
576  * @file:		file pointing to address space structure to wait for
577  * @start_byte:		offset in bytes where the range starts
578  * @end_byte:		offset in bytes where the range ends (inclusive)
579  *
580  * Walk the list of under-writeback pages of the address space that file
581  * refers to, in the given range and wait for all of them.  Check error
582  * status of the address space vs. the file->f_wb_err cursor and return it.
583  *
584  * Since the error status of the file is advanced by this function,
585  * callers are responsible for checking the return value and handling and/or
586  * reporting the error.
587  *
588  * Return: error status of the address space vs. the file->f_wb_err cursor.
589  */
590 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
591 {
592 	struct address_space *mapping = file->f_mapping;
593 
594 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
595 	return file_check_and_advance_wb_err(file);
596 }
597 EXPORT_SYMBOL(file_fdatawait_range);
598 
599 /**
600  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
601  * @mapping: address space structure to wait for
602  *
603  * Walk the list of under-writeback pages of the given address space
604  * and wait for all of them.  Unlike filemap_fdatawait(), this function
605  * does not clear error status of the address space.
606  *
607  * Use this function if callers don't handle errors themselves.  Expected
608  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
609  * fsfreeze(8)
610  *
611  * Return: error status of the address space.
612  */
613 int filemap_fdatawait_keep_errors(struct address_space *mapping)
614 {
615 	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
616 	return filemap_check_and_keep_errors(mapping);
617 }
618 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
619 
620 /* Returns true if writeback might be needed or already in progress. */
621 static bool mapping_needs_writeback(struct address_space *mapping)
622 {
623 	return mapping->nrpages;
624 }
625 
626 bool filemap_range_has_writeback(struct address_space *mapping,
627 				 loff_t start_byte, loff_t end_byte)
628 {
629 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
630 	pgoff_t max = end_byte >> PAGE_SHIFT;
631 	struct page *page;
632 
633 	if (end_byte < start_byte)
634 		return false;
635 
636 	rcu_read_lock();
637 	xas_for_each(&xas, page, max) {
638 		if (xas_retry(&xas, page))
639 			continue;
640 		if (xa_is_value(page))
641 			continue;
642 		if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
643 			break;
644 	}
645 	rcu_read_unlock();
646 	return page != NULL;
647 }
648 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
649 
650 /**
651  * filemap_write_and_wait_range - write out & wait on a file range
652  * @mapping:	the address_space for the pages
653  * @lstart:	offset in bytes where the range starts
654  * @lend:	offset in bytes where the range ends (inclusive)
655  *
656  * Write out and wait upon file offsets lstart->lend, inclusive.
657  *
658  * Note that @lend is inclusive (describes the last byte to be written) so
659  * that this function can be used to write to the very end-of-file (end = -1).
660  *
661  * Return: error status of the address space.
662  */
663 int filemap_write_and_wait_range(struct address_space *mapping,
664 				 loff_t lstart, loff_t lend)
665 {
666 	int err = 0;
667 
668 	if (mapping_needs_writeback(mapping)) {
669 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
670 						 WB_SYNC_ALL);
671 		/*
672 		 * Even if the above returned error, the pages may be
673 		 * written partially (e.g. -ENOSPC), so we wait for it.
674 		 * But the -EIO is special case, it may indicate the worst
675 		 * thing (e.g. bug) happened, so we avoid waiting for it.
676 		 */
677 		if (err != -EIO) {
678 			int err2 = filemap_fdatawait_range(mapping,
679 						lstart, lend);
680 			if (!err)
681 				err = err2;
682 		} else {
683 			/* Clear any previously stored errors */
684 			filemap_check_errors(mapping);
685 		}
686 	} else {
687 		err = filemap_check_errors(mapping);
688 	}
689 	return err;
690 }
691 EXPORT_SYMBOL(filemap_write_and_wait_range);
692 
693 void __filemap_set_wb_err(struct address_space *mapping, int err)
694 {
695 	errseq_t eseq = errseq_set(&mapping->wb_err, err);
696 
697 	trace_filemap_set_wb_err(mapping, eseq);
698 }
699 EXPORT_SYMBOL(__filemap_set_wb_err);
700 
701 /**
702  * file_check_and_advance_wb_err - report wb error (if any) that was previously
703  * 				   and advance wb_err to current one
704  * @file: struct file on which the error is being reported
705  *
706  * When userland calls fsync (or something like nfsd does the equivalent), we
707  * want to report any writeback errors that occurred since the last fsync (or
708  * since the file was opened if there haven't been any).
709  *
710  * Grab the wb_err from the mapping. If it matches what we have in the file,
711  * then just quickly return 0. The file is all caught up.
712  *
713  * If it doesn't match, then take the mapping value, set the "seen" flag in
714  * it and try to swap it into place. If it works, or another task beat us
715  * to it with the new value, then update the f_wb_err and return the error
716  * portion. The error at this point must be reported via proper channels
717  * (a'la fsync, or NFS COMMIT operation, etc.).
718  *
719  * While we handle mapping->wb_err with atomic operations, the f_wb_err
720  * value is protected by the f_lock since we must ensure that it reflects
721  * the latest value swapped in for this file descriptor.
722  *
723  * Return: %0 on success, negative error code otherwise.
724  */
725 int file_check_and_advance_wb_err(struct file *file)
726 {
727 	int err = 0;
728 	errseq_t old = READ_ONCE(file->f_wb_err);
729 	struct address_space *mapping = file->f_mapping;
730 
731 	/* Locklessly handle the common case where nothing has changed */
732 	if (errseq_check(&mapping->wb_err, old)) {
733 		/* Something changed, must use slow path */
734 		spin_lock(&file->f_lock);
735 		old = file->f_wb_err;
736 		err = errseq_check_and_advance(&mapping->wb_err,
737 						&file->f_wb_err);
738 		trace_file_check_and_advance_wb_err(file, old);
739 		spin_unlock(&file->f_lock);
740 	}
741 
742 	/*
743 	 * We're mostly using this function as a drop in replacement for
744 	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
745 	 * that the legacy code would have had on these flags.
746 	 */
747 	clear_bit(AS_EIO, &mapping->flags);
748 	clear_bit(AS_ENOSPC, &mapping->flags);
749 	return err;
750 }
751 EXPORT_SYMBOL(file_check_and_advance_wb_err);
752 
753 /**
754  * file_write_and_wait_range - write out & wait on a file range
755  * @file:	file pointing to address_space with pages
756  * @lstart:	offset in bytes where the range starts
757  * @lend:	offset in bytes where the range ends (inclusive)
758  *
759  * Write out and wait upon file offsets lstart->lend, inclusive.
760  *
761  * Note that @lend is inclusive (describes the last byte to be written) so
762  * that this function can be used to write to the very end-of-file (end = -1).
763  *
764  * After writing out and waiting on the data, we check and advance the
765  * f_wb_err cursor to the latest value, and return any errors detected there.
766  *
767  * Return: %0 on success, negative error code otherwise.
768  */
769 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
770 {
771 	int err = 0, err2;
772 	struct address_space *mapping = file->f_mapping;
773 
774 	if (mapping_needs_writeback(mapping)) {
775 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
776 						 WB_SYNC_ALL);
777 		/* See comment of filemap_write_and_wait() */
778 		if (err != -EIO)
779 			__filemap_fdatawait_range(mapping, lstart, lend);
780 	}
781 	err2 = file_check_and_advance_wb_err(file);
782 	if (!err)
783 		err = err2;
784 	return err;
785 }
786 EXPORT_SYMBOL(file_write_and_wait_range);
787 
788 /**
789  * replace_page_cache_page - replace a pagecache page with a new one
790  * @old:	page to be replaced
791  * @new:	page to replace with
792  *
793  * This function replaces a page in the pagecache with a new one.  On
794  * success it acquires the pagecache reference for the new page and
795  * drops it for the old page.  Both the old and new pages must be
796  * locked.  This function does not add the new page to the LRU, the
797  * caller must do that.
798  *
799  * The remove + add is atomic.  This function cannot fail.
800  */
801 void replace_page_cache_page(struct page *old, struct page *new)
802 {
803 	struct folio *fold = page_folio(old);
804 	struct folio *fnew = page_folio(new);
805 	struct address_space *mapping = old->mapping;
806 	void (*freepage)(struct page *) = mapping->a_ops->freepage;
807 	pgoff_t offset = old->index;
808 	XA_STATE(xas, &mapping->i_pages, offset);
809 
810 	VM_BUG_ON_PAGE(!PageLocked(old), old);
811 	VM_BUG_ON_PAGE(!PageLocked(new), new);
812 	VM_BUG_ON_PAGE(new->mapping, new);
813 
814 	get_page(new);
815 	new->mapping = mapping;
816 	new->index = offset;
817 
818 	mem_cgroup_migrate(fold, fnew);
819 
820 	xas_lock_irq(&xas);
821 	xas_store(&xas, new);
822 
823 	old->mapping = NULL;
824 	/* hugetlb pages do not participate in page cache accounting. */
825 	if (!PageHuge(old))
826 		__dec_lruvec_page_state(old, NR_FILE_PAGES);
827 	if (!PageHuge(new))
828 		__inc_lruvec_page_state(new, NR_FILE_PAGES);
829 	if (PageSwapBacked(old))
830 		__dec_lruvec_page_state(old, NR_SHMEM);
831 	if (PageSwapBacked(new))
832 		__inc_lruvec_page_state(new, NR_SHMEM);
833 	xas_unlock_irq(&xas);
834 	if (freepage)
835 		freepage(old);
836 	put_page(old);
837 }
838 EXPORT_SYMBOL_GPL(replace_page_cache_page);
839 
840 noinline int __filemap_add_folio(struct address_space *mapping,
841 		struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
842 {
843 	XA_STATE(xas, &mapping->i_pages, index);
844 	int huge = folio_test_hugetlb(folio);
845 	int error;
846 	bool charged = false;
847 
848 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
849 	VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
850 	mapping_set_update(&xas, mapping);
851 
852 	folio_get(folio);
853 	folio->mapping = mapping;
854 	folio->index = index;
855 
856 	if (!huge) {
857 		error = mem_cgroup_charge(folio, NULL, gfp);
858 		VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
859 		if (error)
860 			goto error;
861 		charged = true;
862 	}
863 
864 	gfp &= GFP_RECLAIM_MASK;
865 
866 	do {
867 		unsigned int order = xa_get_order(xas.xa, xas.xa_index);
868 		void *entry, *old = NULL;
869 
870 		if (order > folio_order(folio))
871 			xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
872 					order, gfp);
873 		xas_lock_irq(&xas);
874 		xas_for_each_conflict(&xas, entry) {
875 			old = entry;
876 			if (!xa_is_value(entry)) {
877 				xas_set_err(&xas, -EEXIST);
878 				goto unlock;
879 			}
880 		}
881 
882 		if (old) {
883 			if (shadowp)
884 				*shadowp = old;
885 			/* entry may have been split before we acquired lock */
886 			order = xa_get_order(xas.xa, xas.xa_index);
887 			if (order > folio_order(folio)) {
888 				xas_split(&xas, old, order);
889 				xas_reset(&xas);
890 			}
891 		}
892 
893 		xas_store(&xas, folio);
894 		if (xas_error(&xas))
895 			goto unlock;
896 
897 		mapping->nrpages++;
898 
899 		/* hugetlb pages do not participate in page cache accounting */
900 		if (!huge)
901 			__lruvec_stat_add_folio(folio, NR_FILE_PAGES);
902 unlock:
903 		xas_unlock_irq(&xas);
904 	} while (xas_nomem(&xas, gfp));
905 
906 	if (xas_error(&xas)) {
907 		error = xas_error(&xas);
908 		if (charged)
909 			mem_cgroup_uncharge(folio);
910 		goto error;
911 	}
912 
913 	trace_mm_filemap_add_to_page_cache(folio);
914 	return 0;
915 error:
916 	folio->mapping = NULL;
917 	/* Leave page->index set: truncation relies upon it */
918 	folio_put(folio);
919 	return error;
920 }
921 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
922 
923 /**
924  * add_to_page_cache_locked - add a locked page to the pagecache
925  * @page:	page to add
926  * @mapping:	the page's address_space
927  * @offset:	page index
928  * @gfp_mask:	page allocation mode
929  *
930  * This function is used to add a page to the pagecache. It must be locked.
931  * This function does not add the page to the LRU.  The caller must do that.
932  *
933  * Return: %0 on success, negative error code otherwise.
934  */
935 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
936 		pgoff_t offset, gfp_t gfp_mask)
937 {
938 	return __filemap_add_folio(mapping, page_folio(page), offset,
939 					  gfp_mask, NULL);
940 }
941 EXPORT_SYMBOL(add_to_page_cache_locked);
942 
943 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
944 				pgoff_t index, gfp_t gfp)
945 {
946 	void *shadow = NULL;
947 	int ret;
948 
949 	__folio_set_locked(folio);
950 	ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
951 	if (unlikely(ret))
952 		__folio_clear_locked(folio);
953 	else {
954 		/*
955 		 * The folio might have been evicted from cache only
956 		 * recently, in which case it should be activated like
957 		 * any other repeatedly accessed folio.
958 		 * The exception is folios getting rewritten; evicting other
959 		 * data from the working set, only to cache data that will
960 		 * get overwritten with something else, is a waste of memory.
961 		 */
962 		WARN_ON_ONCE(folio_test_active(folio));
963 		if (!(gfp & __GFP_WRITE) && shadow)
964 			workingset_refault(folio, shadow);
965 		folio_add_lru(folio);
966 	}
967 	return ret;
968 }
969 EXPORT_SYMBOL_GPL(filemap_add_folio);
970 
971 #ifdef CONFIG_NUMA
972 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
973 {
974 	int n;
975 	struct folio *folio;
976 
977 	if (cpuset_do_page_mem_spread()) {
978 		unsigned int cpuset_mems_cookie;
979 		do {
980 			cpuset_mems_cookie = read_mems_allowed_begin();
981 			n = cpuset_mem_spread_node();
982 			folio = __folio_alloc_node(gfp, order, n);
983 		} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
984 
985 		return folio;
986 	}
987 	return folio_alloc(gfp, order);
988 }
989 EXPORT_SYMBOL(filemap_alloc_folio);
990 #endif
991 
992 /*
993  * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
994  *
995  * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
996  *
997  * @mapping1: the first mapping to lock
998  * @mapping2: the second mapping to lock
999  */
1000 void filemap_invalidate_lock_two(struct address_space *mapping1,
1001 				 struct address_space *mapping2)
1002 {
1003 	if (mapping1 > mapping2)
1004 		swap(mapping1, mapping2);
1005 	if (mapping1)
1006 		down_write(&mapping1->invalidate_lock);
1007 	if (mapping2 && mapping1 != mapping2)
1008 		down_write_nested(&mapping2->invalidate_lock, 1);
1009 }
1010 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1011 
1012 /*
1013  * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1014  *
1015  * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1016  *
1017  * @mapping1: the first mapping to unlock
1018  * @mapping2: the second mapping to unlock
1019  */
1020 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1021 				   struct address_space *mapping2)
1022 {
1023 	if (mapping1)
1024 		up_write(&mapping1->invalidate_lock);
1025 	if (mapping2 && mapping1 != mapping2)
1026 		up_write(&mapping2->invalidate_lock);
1027 }
1028 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1029 
1030 /*
1031  * In order to wait for pages to become available there must be
1032  * waitqueues associated with pages. By using a hash table of
1033  * waitqueues where the bucket discipline is to maintain all
1034  * waiters on the same queue and wake all when any of the pages
1035  * become available, and for the woken contexts to check to be
1036  * sure the appropriate page became available, this saves space
1037  * at a cost of "thundering herd" phenomena during rare hash
1038  * collisions.
1039  */
1040 #define PAGE_WAIT_TABLE_BITS 8
1041 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1042 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1043 
1044 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1045 {
1046 	return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1047 }
1048 
1049 void __init pagecache_init(void)
1050 {
1051 	int i;
1052 
1053 	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1054 		init_waitqueue_head(&folio_wait_table[i]);
1055 
1056 	page_writeback_init();
1057 }
1058 
1059 /*
1060  * The page wait code treats the "wait->flags" somewhat unusually, because
1061  * we have multiple different kinds of waits, not just the usual "exclusive"
1062  * one.
1063  *
1064  * We have:
1065  *
1066  *  (a) no special bits set:
1067  *
1068  *	We're just waiting for the bit to be released, and when a waker
1069  *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1070  *	and remove it from the wait queue.
1071  *
1072  *	Simple and straightforward.
1073  *
1074  *  (b) WQ_FLAG_EXCLUSIVE:
1075  *
1076  *	The waiter is waiting to get the lock, and only one waiter should
1077  *	be woken up to avoid any thundering herd behavior. We'll set the
1078  *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1079  *
1080  *	This is the traditional exclusive wait.
1081  *
1082  *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1083  *
1084  *	The waiter is waiting to get the bit, and additionally wants the
1085  *	lock to be transferred to it for fair lock behavior. If the lock
1086  *	cannot be taken, we stop walking the wait queue without waking
1087  *	the waiter.
1088  *
1089  *	This is the "fair lock handoff" case, and in addition to setting
1090  *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1091  *	that it now has the lock.
1092  */
1093 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1094 {
1095 	unsigned int flags;
1096 	struct wait_page_key *key = arg;
1097 	struct wait_page_queue *wait_page
1098 		= container_of(wait, struct wait_page_queue, wait);
1099 
1100 	if (!wake_page_match(wait_page, key))
1101 		return 0;
1102 
1103 	/*
1104 	 * If it's a lock handoff wait, we get the bit for it, and
1105 	 * stop walking (and do not wake it up) if we can't.
1106 	 */
1107 	flags = wait->flags;
1108 	if (flags & WQ_FLAG_EXCLUSIVE) {
1109 		if (test_bit(key->bit_nr, &key->folio->flags))
1110 			return -1;
1111 		if (flags & WQ_FLAG_CUSTOM) {
1112 			if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1113 				return -1;
1114 			flags |= WQ_FLAG_DONE;
1115 		}
1116 	}
1117 
1118 	/*
1119 	 * We are holding the wait-queue lock, but the waiter that
1120 	 * is waiting for this will be checking the flags without
1121 	 * any locking.
1122 	 *
1123 	 * So update the flags atomically, and wake up the waiter
1124 	 * afterwards to avoid any races. This store-release pairs
1125 	 * with the load-acquire in folio_wait_bit_common().
1126 	 */
1127 	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1128 	wake_up_state(wait->private, mode);
1129 
1130 	/*
1131 	 * Ok, we have successfully done what we're waiting for,
1132 	 * and we can unconditionally remove the wait entry.
1133 	 *
1134 	 * Note that this pairs with the "finish_wait()" in the
1135 	 * waiter, and has to be the absolute last thing we do.
1136 	 * After this list_del_init(&wait->entry) the wait entry
1137 	 * might be de-allocated and the process might even have
1138 	 * exited.
1139 	 */
1140 	list_del_init_careful(&wait->entry);
1141 	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1142 }
1143 
1144 static void folio_wake_bit(struct folio *folio, int bit_nr)
1145 {
1146 	wait_queue_head_t *q = folio_waitqueue(folio);
1147 	struct wait_page_key key;
1148 	unsigned long flags;
1149 	wait_queue_entry_t bookmark;
1150 
1151 	key.folio = folio;
1152 	key.bit_nr = bit_nr;
1153 	key.page_match = 0;
1154 
1155 	bookmark.flags = 0;
1156 	bookmark.private = NULL;
1157 	bookmark.func = NULL;
1158 	INIT_LIST_HEAD(&bookmark.entry);
1159 
1160 	spin_lock_irqsave(&q->lock, flags);
1161 	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1162 
1163 	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1164 		/*
1165 		 * Take a breather from holding the lock,
1166 		 * allow pages that finish wake up asynchronously
1167 		 * to acquire the lock and remove themselves
1168 		 * from wait queue
1169 		 */
1170 		spin_unlock_irqrestore(&q->lock, flags);
1171 		cpu_relax();
1172 		spin_lock_irqsave(&q->lock, flags);
1173 		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1174 	}
1175 
1176 	/*
1177 	 * It is possible for other pages to have collided on the waitqueue
1178 	 * hash, so in that case check for a page match. That prevents a long-
1179 	 * term waiter
1180 	 *
1181 	 * It is still possible to miss a case here, when we woke page waiters
1182 	 * and removed them from the waitqueue, but there are still other
1183 	 * page waiters.
1184 	 */
1185 	if (!waitqueue_active(q) || !key.page_match) {
1186 		folio_clear_waiters(folio);
1187 		/*
1188 		 * It's possible to miss clearing Waiters here, when we woke
1189 		 * our page waiters, but the hashed waitqueue has waiters for
1190 		 * other pages on it.
1191 		 *
1192 		 * That's okay, it's a rare case. The next waker will clear it.
1193 		 */
1194 	}
1195 	spin_unlock_irqrestore(&q->lock, flags);
1196 }
1197 
1198 static void folio_wake(struct folio *folio, int bit)
1199 {
1200 	if (!folio_test_waiters(folio))
1201 		return;
1202 	folio_wake_bit(folio, bit);
1203 }
1204 
1205 /*
1206  * A choice of three behaviors for folio_wait_bit_common():
1207  */
1208 enum behavior {
1209 	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1210 			 * __folio_lock() waiting on then setting PG_locked.
1211 			 */
1212 	SHARED,		/* Hold ref to page and check the bit when woken, like
1213 			 * folio_wait_writeback() waiting on PG_writeback.
1214 			 */
1215 	DROP,		/* Drop ref to page before wait, no check when woken,
1216 			 * like folio_put_wait_locked() on PG_locked.
1217 			 */
1218 };
1219 
1220 /*
1221  * Attempt to check (or get) the folio flag, and mark us done
1222  * if successful.
1223  */
1224 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1225 					struct wait_queue_entry *wait)
1226 {
1227 	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1228 		if (test_and_set_bit(bit_nr, &folio->flags))
1229 			return false;
1230 	} else if (test_bit(bit_nr, &folio->flags))
1231 		return false;
1232 
1233 	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1234 	return true;
1235 }
1236 
1237 /* How many times do we accept lock stealing from under a waiter? */
1238 int sysctl_page_lock_unfairness = 5;
1239 
1240 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1241 		int state, enum behavior behavior)
1242 {
1243 	wait_queue_head_t *q = folio_waitqueue(folio);
1244 	int unfairness = sysctl_page_lock_unfairness;
1245 	struct wait_page_queue wait_page;
1246 	wait_queue_entry_t *wait = &wait_page.wait;
1247 	bool thrashing = false;
1248 	bool delayacct = false;
1249 	unsigned long pflags;
1250 
1251 	if (bit_nr == PG_locked &&
1252 	    !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1253 		if (!folio_test_swapbacked(folio)) {
1254 			delayacct_thrashing_start();
1255 			delayacct = true;
1256 		}
1257 		psi_memstall_enter(&pflags);
1258 		thrashing = true;
1259 	}
1260 
1261 	init_wait(wait);
1262 	wait->func = wake_page_function;
1263 	wait_page.folio = folio;
1264 	wait_page.bit_nr = bit_nr;
1265 
1266 repeat:
1267 	wait->flags = 0;
1268 	if (behavior == EXCLUSIVE) {
1269 		wait->flags = WQ_FLAG_EXCLUSIVE;
1270 		if (--unfairness < 0)
1271 			wait->flags |= WQ_FLAG_CUSTOM;
1272 	}
1273 
1274 	/*
1275 	 * Do one last check whether we can get the
1276 	 * page bit synchronously.
1277 	 *
1278 	 * Do the folio_set_waiters() marking before that
1279 	 * to let any waker we _just_ missed know they
1280 	 * need to wake us up (otherwise they'll never
1281 	 * even go to the slow case that looks at the
1282 	 * page queue), and add ourselves to the wait
1283 	 * queue if we need to sleep.
1284 	 *
1285 	 * This part needs to be done under the queue
1286 	 * lock to avoid races.
1287 	 */
1288 	spin_lock_irq(&q->lock);
1289 	folio_set_waiters(folio);
1290 	if (!folio_trylock_flag(folio, bit_nr, wait))
1291 		__add_wait_queue_entry_tail(q, wait);
1292 	spin_unlock_irq(&q->lock);
1293 
1294 	/*
1295 	 * From now on, all the logic will be based on
1296 	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1297 	 * see whether the page bit testing has already
1298 	 * been done by the wake function.
1299 	 *
1300 	 * We can drop our reference to the folio.
1301 	 */
1302 	if (behavior == DROP)
1303 		folio_put(folio);
1304 
1305 	/*
1306 	 * Note that until the "finish_wait()", or until
1307 	 * we see the WQ_FLAG_WOKEN flag, we need to
1308 	 * be very careful with the 'wait->flags', because
1309 	 * we may race with a waker that sets them.
1310 	 */
1311 	for (;;) {
1312 		unsigned int flags;
1313 
1314 		set_current_state(state);
1315 
1316 		/* Loop until we've been woken or interrupted */
1317 		flags = smp_load_acquire(&wait->flags);
1318 		if (!(flags & WQ_FLAG_WOKEN)) {
1319 			if (signal_pending_state(state, current))
1320 				break;
1321 
1322 			io_schedule();
1323 			continue;
1324 		}
1325 
1326 		/* If we were non-exclusive, we're done */
1327 		if (behavior != EXCLUSIVE)
1328 			break;
1329 
1330 		/* If the waker got the lock for us, we're done */
1331 		if (flags & WQ_FLAG_DONE)
1332 			break;
1333 
1334 		/*
1335 		 * Otherwise, if we're getting the lock, we need to
1336 		 * try to get it ourselves.
1337 		 *
1338 		 * And if that fails, we'll have to retry this all.
1339 		 */
1340 		if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1341 			goto repeat;
1342 
1343 		wait->flags |= WQ_FLAG_DONE;
1344 		break;
1345 	}
1346 
1347 	/*
1348 	 * If a signal happened, this 'finish_wait()' may remove the last
1349 	 * waiter from the wait-queues, but the folio waiters bit will remain
1350 	 * set. That's ok. The next wakeup will take care of it, and trying
1351 	 * to do it here would be difficult and prone to races.
1352 	 */
1353 	finish_wait(q, wait);
1354 
1355 	if (thrashing) {
1356 		if (delayacct)
1357 			delayacct_thrashing_end();
1358 		psi_memstall_leave(&pflags);
1359 	}
1360 
1361 	/*
1362 	 * NOTE! The wait->flags weren't stable until we've done the
1363 	 * 'finish_wait()', and we could have exited the loop above due
1364 	 * to a signal, and had a wakeup event happen after the signal
1365 	 * test but before the 'finish_wait()'.
1366 	 *
1367 	 * So only after the finish_wait() can we reliably determine
1368 	 * if we got woken up or not, so we can now figure out the final
1369 	 * return value based on that state without races.
1370 	 *
1371 	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1372 	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1373 	 */
1374 	if (behavior == EXCLUSIVE)
1375 		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1376 
1377 	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1378 }
1379 
1380 #ifdef CONFIG_MIGRATION
1381 /**
1382  * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1383  * @entry: migration swap entry.
1384  * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1385  *        for pte entries, pass NULL for pmd entries.
1386  * @ptl: already locked ptl. This function will drop the lock.
1387  *
1388  * Wait for a migration entry referencing the given page to be removed. This is
1389  * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1390  * this can be called without taking a reference on the page. Instead this
1391  * should be called while holding the ptl for the migration entry referencing
1392  * the page.
1393  *
1394  * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1395  *
1396  * This follows the same logic as folio_wait_bit_common() so see the comments
1397  * there.
1398  */
1399 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1400 				spinlock_t *ptl)
1401 {
1402 	struct wait_page_queue wait_page;
1403 	wait_queue_entry_t *wait = &wait_page.wait;
1404 	bool thrashing = false;
1405 	bool delayacct = false;
1406 	unsigned long pflags;
1407 	wait_queue_head_t *q;
1408 	struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1409 
1410 	q = folio_waitqueue(folio);
1411 	if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1412 		if (!folio_test_swapbacked(folio)) {
1413 			delayacct_thrashing_start();
1414 			delayacct = true;
1415 		}
1416 		psi_memstall_enter(&pflags);
1417 		thrashing = true;
1418 	}
1419 
1420 	init_wait(wait);
1421 	wait->func = wake_page_function;
1422 	wait_page.folio = folio;
1423 	wait_page.bit_nr = PG_locked;
1424 	wait->flags = 0;
1425 
1426 	spin_lock_irq(&q->lock);
1427 	folio_set_waiters(folio);
1428 	if (!folio_trylock_flag(folio, PG_locked, wait))
1429 		__add_wait_queue_entry_tail(q, wait);
1430 	spin_unlock_irq(&q->lock);
1431 
1432 	/*
1433 	 * If a migration entry exists for the page the migration path must hold
1434 	 * a valid reference to the page, and it must take the ptl to remove the
1435 	 * migration entry. So the page is valid until the ptl is dropped.
1436 	 */
1437 	if (ptep)
1438 		pte_unmap_unlock(ptep, ptl);
1439 	else
1440 		spin_unlock(ptl);
1441 
1442 	for (;;) {
1443 		unsigned int flags;
1444 
1445 		set_current_state(TASK_UNINTERRUPTIBLE);
1446 
1447 		/* Loop until we've been woken or interrupted */
1448 		flags = smp_load_acquire(&wait->flags);
1449 		if (!(flags & WQ_FLAG_WOKEN)) {
1450 			if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1451 				break;
1452 
1453 			io_schedule();
1454 			continue;
1455 		}
1456 		break;
1457 	}
1458 
1459 	finish_wait(q, wait);
1460 
1461 	if (thrashing) {
1462 		if (delayacct)
1463 			delayacct_thrashing_end();
1464 		psi_memstall_leave(&pflags);
1465 	}
1466 }
1467 #endif
1468 
1469 void folio_wait_bit(struct folio *folio, int bit_nr)
1470 {
1471 	folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1472 }
1473 EXPORT_SYMBOL(folio_wait_bit);
1474 
1475 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1476 {
1477 	return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1478 }
1479 EXPORT_SYMBOL(folio_wait_bit_killable);
1480 
1481 /**
1482  * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1483  * @folio: The folio to wait for.
1484  * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1485  *
1486  * The caller should hold a reference on @folio.  They expect the page to
1487  * become unlocked relatively soon, but do not wish to hold up migration
1488  * (for example) by holding the reference while waiting for the folio to
1489  * come unlocked.  After this function returns, the caller should not
1490  * dereference @folio.
1491  *
1492  * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1493  */
1494 int folio_put_wait_locked(struct folio *folio, int state)
1495 {
1496 	return folio_wait_bit_common(folio, PG_locked, state, DROP);
1497 }
1498 
1499 /**
1500  * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1501  * @folio: Folio defining the wait queue of interest
1502  * @waiter: Waiter to add to the queue
1503  *
1504  * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1505  */
1506 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1507 {
1508 	wait_queue_head_t *q = folio_waitqueue(folio);
1509 	unsigned long flags;
1510 
1511 	spin_lock_irqsave(&q->lock, flags);
1512 	__add_wait_queue_entry_tail(q, waiter);
1513 	folio_set_waiters(folio);
1514 	spin_unlock_irqrestore(&q->lock, flags);
1515 }
1516 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1517 
1518 #ifndef clear_bit_unlock_is_negative_byte
1519 
1520 /*
1521  * PG_waiters is the high bit in the same byte as PG_lock.
1522  *
1523  * On x86 (and on many other architectures), we can clear PG_lock and
1524  * test the sign bit at the same time. But if the architecture does
1525  * not support that special operation, we just do this all by hand
1526  * instead.
1527  *
1528  * The read of PG_waiters has to be after (or concurrently with) PG_locked
1529  * being cleared, but a memory barrier should be unnecessary since it is
1530  * in the same byte as PG_locked.
1531  */
1532 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1533 {
1534 	clear_bit_unlock(nr, mem);
1535 	/* smp_mb__after_atomic(); */
1536 	return test_bit(PG_waiters, mem);
1537 }
1538 
1539 #endif
1540 
1541 /**
1542  * folio_unlock - Unlock a locked folio.
1543  * @folio: The folio.
1544  *
1545  * Unlocks the folio and wakes up any thread sleeping on the page lock.
1546  *
1547  * Context: May be called from interrupt or process context.  May not be
1548  * called from NMI context.
1549  */
1550 void folio_unlock(struct folio *folio)
1551 {
1552 	/* Bit 7 allows x86 to check the byte's sign bit */
1553 	BUILD_BUG_ON(PG_waiters != 7);
1554 	BUILD_BUG_ON(PG_locked > 7);
1555 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1556 	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1557 		folio_wake_bit(folio, PG_locked);
1558 }
1559 EXPORT_SYMBOL(folio_unlock);
1560 
1561 /**
1562  * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1563  * @folio: The folio.
1564  *
1565  * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1566  * it.  The folio reference held for PG_private_2 being set is released.
1567  *
1568  * This is, for example, used when a netfs folio is being written to a local
1569  * disk cache, thereby allowing writes to the cache for the same folio to be
1570  * serialised.
1571  */
1572 void folio_end_private_2(struct folio *folio)
1573 {
1574 	VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1575 	clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1576 	folio_wake_bit(folio, PG_private_2);
1577 	folio_put(folio);
1578 }
1579 EXPORT_SYMBOL(folio_end_private_2);
1580 
1581 /**
1582  * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1583  * @folio: The folio to wait on.
1584  *
1585  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1586  */
1587 void folio_wait_private_2(struct folio *folio)
1588 {
1589 	while (folio_test_private_2(folio))
1590 		folio_wait_bit(folio, PG_private_2);
1591 }
1592 EXPORT_SYMBOL(folio_wait_private_2);
1593 
1594 /**
1595  * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1596  * @folio: The folio to wait on.
1597  *
1598  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1599  * fatal signal is received by the calling task.
1600  *
1601  * Return:
1602  * - 0 if successful.
1603  * - -EINTR if a fatal signal was encountered.
1604  */
1605 int folio_wait_private_2_killable(struct folio *folio)
1606 {
1607 	int ret = 0;
1608 
1609 	while (folio_test_private_2(folio)) {
1610 		ret = folio_wait_bit_killable(folio, PG_private_2);
1611 		if (ret < 0)
1612 			break;
1613 	}
1614 
1615 	return ret;
1616 }
1617 EXPORT_SYMBOL(folio_wait_private_2_killable);
1618 
1619 /**
1620  * folio_end_writeback - End writeback against a folio.
1621  * @folio: The folio.
1622  */
1623 void folio_end_writeback(struct folio *folio)
1624 {
1625 	/*
1626 	 * folio_test_clear_reclaim() could be used here but it is an
1627 	 * atomic operation and overkill in this particular case. Failing
1628 	 * to shuffle a folio marked for immediate reclaim is too mild
1629 	 * a gain to justify taking an atomic operation penalty at the
1630 	 * end of every folio writeback.
1631 	 */
1632 	if (folio_test_reclaim(folio)) {
1633 		folio_clear_reclaim(folio);
1634 		folio_rotate_reclaimable(folio);
1635 	}
1636 
1637 	/*
1638 	 * Writeback does not hold a folio reference of its own, relying
1639 	 * on truncation to wait for the clearing of PG_writeback.
1640 	 * But here we must make sure that the folio is not freed and
1641 	 * reused before the folio_wake().
1642 	 */
1643 	folio_get(folio);
1644 	if (!__folio_end_writeback(folio))
1645 		BUG();
1646 
1647 	smp_mb__after_atomic();
1648 	folio_wake(folio, PG_writeback);
1649 	acct_reclaim_writeback(folio);
1650 	folio_put(folio);
1651 }
1652 EXPORT_SYMBOL(folio_end_writeback);
1653 
1654 /*
1655  * After completing I/O on a page, call this routine to update the page
1656  * flags appropriately
1657  */
1658 void page_endio(struct page *page, bool is_write, int err)
1659 {
1660 	if (!is_write) {
1661 		if (!err) {
1662 			SetPageUptodate(page);
1663 		} else {
1664 			ClearPageUptodate(page);
1665 			SetPageError(page);
1666 		}
1667 		unlock_page(page);
1668 	} else {
1669 		if (err) {
1670 			struct address_space *mapping;
1671 
1672 			SetPageError(page);
1673 			mapping = page_mapping(page);
1674 			if (mapping)
1675 				mapping_set_error(mapping, err);
1676 		}
1677 		end_page_writeback(page);
1678 	}
1679 }
1680 EXPORT_SYMBOL_GPL(page_endio);
1681 
1682 /**
1683  * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1684  * @folio: The folio to lock
1685  */
1686 void __folio_lock(struct folio *folio)
1687 {
1688 	folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1689 				EXCLUSIVE);
1690 }
1691 EXPORT_SYMBOL(__folio_lock);
1692 
1693 int __folio_lock_killable(struct folio *folio)
1694 {
1695 	return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1696 					EXCLUSIVE);
1697 }
1698 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1699 
1700 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1701 {
1702 	struct wait_queue_head *q = folio_waitqueue(folio);
1703 	int ret = 0;
1704 
1705 	wait->folio = folio;
1706 	wait->bit_nr = PG_locked;
1707 
1708 	spin_lock_irq(&q->lock);
1709 	__add_wait_queue_entry_tail(q, &wait->wait);
1710 	folio_set_waiters(folio);
1711 	ret = !folio_trylock(folio);
1712 	/*
1713 	 * If we were successful now, we know we're still on the
1714 	 * waitqueue as we're still under the lock. This means it's
1715 	 * safe to remove and return success, we know the callback
1716 	 * isn't going to trigger.
1717 	 */
1718 	if (!ret)
1719 		__remove_wait_queue(q, &wait->wait);
1720 	else
1721 		ret = -EIOCBQUEUED;
1722 	spin_unlock_irq(&q->lock);
1723 	return ret;
1724 }
1725 
1726 /*
1727  * Return values:
1728  * true - folio is locked; mmap_lock is still held.
1729  * false - folio is not locked.
1730  *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1731  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1732  *     which case mmap_lock is still held.
1733  *
1734  * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1735  * with the folio locked and the mmap_lock unperturbed.
1736  */
1737 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1738 			 unsigned int flags)
1739 {
1740 	if (fault_flag_allow_retry_first(flags)) {
1741 		/*
1742 		 * CAUTION! In this case, mmap_lock is not released
1743 		 * even though return 0.
1744 		 */
1745 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1746 			return false;
1747 
1748 		mmap_read_unlock(mm);
1749 		if (flags & FAULT_FLAG_KILLABLE)
1750 			folio_wait_locked_killable(folio);
1751 		else
1752 			folio_wait_locked(folio);
1753 		return false;
1754 	}
1755 	if (flags & FAULT_FLAG_KILLABLE) {
1756 		bool ret;
1757 
1758 		ret = __folio_lock_killable(folio);
1759 		if (ret) {
1760 			mmap_read_unlock(mm);
1761 			return false;
1762 		}
1763 	} else {
1764 		__folio_lock(folio);
1765 	}
1766 
1767 	return true;
1768 }
1769 
1770 /**
1771  * page_cache_next_miss() - Find the next gap in the page cache.
1772  * @mapping: Mapping.
1773  * @index: Index.
1774  * @max_scan: Maximum range to search.
1775  *
1776  * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1777  * gap with the lowest index.
1778  *
1779  * This function may be called under the rcu_read_lock.  However, this will
1780  * not atomically search a snapshot of the cache at a single point in time.
1781  * For example, if a gap is created at index 5, then subsequently a gap is
1782  * created at index 10, page_cache_next_miss covering both indices may
1783  * return 10 if called under the rcu_read_lock.
1784  *
1785  * Return: The index of the gap if found, otherwise an index outside the
1786  * range specified (in which case 'return - index >= max_scan' will be true).
1787  * In the rare case of index wrap-around, 0 will be returned.
1788  */
1789 pgoff_t page_cache_next_miss(struct address_space *mapping,
1790 			     pgoff_t index, unsigned long max_scan)
1791 {
1792 	XA_STATE(xas, &mapping->i_pages, index);
1793 
1794 	while (max_scan--) {
1795 		void *entry = xas_next(&xas);
1796 		if (!entry || xa_is_value(entry))
1797 			break;
1798 		if (xas.xa_index == 0)
1799 			break;
1800 	}
1801 
1802 	return xas.xa_index;
1803 }
1804 EXPORT_SYMBOL(page_cache_next_miss);
1805 
1806 /**
1807  * page_cache_prev_miss() - Find the previous gap in the page cache.
1808  * @mapping: Mapping.
1809  * @index: Index.
1810  * @max_scan: Maximum range to search.
1811  *
1812  * Search the range [max(index - max_scan + 1, 0), index] for the
1813  * gap with the highest index.
1814  *
1815  * This function may be called under the rcu_read_lock.  However, this will
1816  * not atomically search a snapshot of the cache at a single point in time.
1817  * For example, if a gap is created at index 10, then subsequently a gap is
1818  * created at index 5, page_cache_prev_miss() covering both indices may
1819  * return 5 if called under the rcu_read_lock.
1820  *
1821  * Return: The index of the gap if found, otherwise an index outside the
1822  * range specified (in which case 'index - return >= max_scan' will be true).
1823  * In the rare case of wrap-around, ULONG_MAX will be returned.
1824  */
1825 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1826 			     pgoff_t index, unsigned long max_scan)
1827 {
1828 	XA_STATE(xas, &mapping->i_pages, index);
1829 
1830 	while (max_scan--) {
1831 		void *entry = xas_prev(&xas);
1832 		if (!entry || xa_is_value(entry))
1833 			break;
1834 		if (xas.xa_index == ULONG_MAX)
1835 			break;
1836 	}
1837 
1838 	return xas.xa_index;
1839 }
1840 EXPORT_SYMBOL(page_cache_prev_miss);
1841 
1842 /*
1843  * Lockless page cache protocol:
1844  * On the lookup side:
1845  * 1. Load the folio from i_pages
1846  * 2. Increment the refcount if it's not zero
1847  * 3. If the folio is not found by xas_reload(), put the refcount and retry
1848  *
1849  * On the removal side:
1850  * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1851  * B. Remove the page from i_pages
1852  * C. Return the page to the page allocator
1853  *
1854  * This means that any page may have its reference count temporarily
1855  * increased by a speculative page cache (or fast GUP) lookup as it can
1856  * be allocated by another user before the RCU grace period expires.
1857  * Because the refcount temporarily acquired here may end up being the
1858  * last refcount on the page, any page allocation must be freeable by
1859  * folio_put().
1860  */
1861 
1862 /*
1863  * mapping_get_entry - Get a page cache entry.
1864  * @mapping: the address_space to search
1865  * @index: The page cache index.
1866  *
1867  * Looks up the page cache entry at @mapping & @index.  If it is a folio,
1868  * it is returned with an increased refcount.  If it is a shadow entry
1869  * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1870  * it is returned without further action.
1871  *
1872  * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1873  */
1874 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1875 {
1876 	XA_STATE(xas, &mapping->i_pages, index);
1877 	struct folio *folio;
1878 
1879 	rcu_read_lock();
1880 repeat:
1881 	xas_reset(&xas);
1882 	folio = xas_load(&xas);
1883 	if (xas_retry(&xas, folio))
1884 		goto repeat;
1885 	/*
1886 	 * A shadow entry of a recently evicted page, or a swap entry from
1887 	 * shmem/tmpfs.  Return it without attempting to raise page count.
1888 	 */
1889 	if (!folio || xa_is_value(folio))
1890 		goto out;
1891 
1892 	if (!folio_try_get_rcu(folio))
1893 		goto repeat;
1894 
1895 	if (unlikely(folio != xas_reload(&xas))) {
1896 		folio_put(folio);
1897 		goto repeat;
1898 	}
1899 out:
1900 	rcu_read_unlock();
1901 
1902 	return folio;
1903 }
1904 
1905 /**
1906  * __filemap_get_folio - Find and get a reference to a folio.
1907  * @mapping: The address_space to search.
1908  * @index: The page index.
1909  * @fgp_flags: %FGP flags modify how the folio is returned.
1910  * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1911  *
1912  * Looks up the page cache entry at @mapping & @index.
1913  *
1914  * @fgp_flags can be zero or more of these flags:
1915  *
1916  * * %FGP_ACCESSED - The folio will be marked accessed.
1917  * * %FGP_LOCK - The folio is returned locked.
1918  * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1919  *   instead of allocating a new folio to replace it.
1920  * * %FGP_CREAT - If no page is present then a new page is allocated using
1921  *   @gfp and added to the page cache and the VM's LRU list.
1922  *   The page is returned locked and with an increased refcount.
1923  * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1924  *   page is already in cache.  If the page was allocated, unlock it before
1925  *   returning so the caller can do the same dance.
1926  * * %FGP_WRITE - The page will be written to by the caller.
1927  * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1928  * * %FGP_NOWAIT - Don't get blocked by page lock.
1929  * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1930  *
1931  * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1932  * if the %GFP flags specified for %FGP_CREAT are atomic.
1933  *
1934  * If there is a page cache page, it is returned with an increased refcount.
1935  *
1936  * Return: The found folio or %NULL otherwise.
1937  */
1938 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1939 		int fgp_flags, gfp_t gfp)
1940 {
1941 	struct folio *folio;
1942 
1943 repeat:
1944 	folio = mapping_get_entry(mapping, index);
1945 	if (xa_is_value(folio)) {
1946 		if (fgp_flags & FGP_ENTRY)
1947 			return folio;
1948 		folio = NULL;
1949 	}
1950 	if (!folio)
1951 		goto no_page;
1952 
1953 	if (fgp_flags & FGP_LOCK) {
1954 		if (fgp_flags & FGP_NOWAIT) {
1955 			if (!folio_trylock(folio)) {
1956 				folio_put(folio);
1957 				return NULL;
1958 			}
1959 		} else {
1960 			folio_lock(folio);
1961 		}
1962 
1963 		/* Has the page been truncated? */
1964 		if (unlikely(folio->mapping != mapping)) {
1965 			folio_unlock(folio);
1966 			folio_put(folio);
1967 			goto repeat;
1968 		}
1969 		VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1970 	}
1971 
1972 	if (fgp_flags & FGP_ACCESSED)
1973 		folio_mark_accessed(folio);
1974 	else if (fgp_flags & FGP_WRITE) {
1975 		/* Clear idle flag for buffer write */
1976 		if (folio_test_idle(folio))
1977 			folio_clear_idle(folio);
1978 	}
1979 
1980 	if (fgp_flags & FGP_STABLE)
1981 		folio_wait_stable(folio);
1982 no_page:
1983 	if (!folio && (fgp_flags & FGP_CREAT)) {
1984 		int err;
1985 		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1986 			gfp |= __GFP_WRITE;
1987 		if (fgp_flags & FGP_NOFS)
1988 			gfp &= ~__GFP_FS;
1989 
1990 		folio = filemap_alloc_folio(gfp, 0);
1991 		if (!folio)
1992 			return NULL;
1993 
1994 		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1995 			fgp_flags |= FGP_LOCK;
1996 
1997 		/* Init accessed so avoid atomic mark_page_accessed later */
1998 		if (fgp_flags & FGP_ACCESSED)
1999 			__folio_set_referenced(folio);
2000 
2001 		err = filemap_add_folio(mapping, folio, index, gfp);
2002 		if (unlikely(err)) {
2003 			folio_put(folio);
2004 			folio = NULL;
2005 			if (err == -EEXIST)
2006 				goto repeat;
2007 		}
2008 
2009 		/*
2010 		 * filemap_add_folio locks the page, and for mmap
2011 		 * we expect an unlocked page.
2012 		 */
2013 		if (folio && (fgp_flags & FGP_FOR_MMAP))
2014 			folio_unlock(folio);
2015 	}
2016 
2017 	return folio;
2018 }
2019 EXPORT_SYMBOL(__filemap_get_folio);
2020 
2021 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2022 		xa_mark_t mark)
2023 {
2024 	struct folio *folio;
2025 
2026 retry:
2027 	if (mark == XA_PRESENT)
2028 		folio = xas_find(xas, max);
2029 	else
2030 		folio = xas_find_marked(xas, max, mark);
2031 
2032 	if (xas_retry(xas, folio))
2033 		goto retry;
2034 	/*
2035 	 * A shadow entry of a recently evicted page, a swap
2036 	 * entry from shmem/tmpfs or a DAX entry.  Return it
2037 	 * without attempting to raise page count.
2038 	 */
2039 	if (!folio || xa_is_value(folio))
2040 		return folio;
2041 
2042 	if (!folio_try_get_rcu(folio))
2043 		goto reset;
2044 
2045 	if (unlikely(folio != xas_reload(xas))) {
2046 		folio_put(folio);
2047 		goto reset;
2048 	}
2049 
2050 	return folio;
2051 reset:
2052 	xas_reset(xas);
2053 	goto retry;
2054 }
2055 
2056 /**
2057  * find_get_entries - gang pagecache lookup
2058  * @mapping:	The address_space to search
2059  * @start:	The starting page cache index
2060  * @end:	The final page index (inclusive).
2061  * @fbatch:	Where the resulting entries are placed.
2062  * @indices:	The cache indices corresponding to the entries in @entries
2063  *
2064  * find_get_entries() will search for and return a batch of entries in
2065  * the mapping.  The entries are placed in @fbatch.  find_get_entries()
2066  * takes a reference on any actual folios it returns.
2067  *
2068  * The entries have ascending indexes.  The indices may not be consecutive
2069  * due to not-present entries or large folios.
2070  *
2071  * Any shadow entries of evicted folios, or swap entries from
2072  * shmem/tmpfs, are included in the returned array.
2073  *
2074  * Return: The number of entries which were found.
2075  */
2076 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2077 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2078 {
2079 	XA_STATE(xas, &mapping->i_pages, start);
2080 	struct folio *folio;
2081 
2082 	rcu_read_lock();
2083 	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2084 		indices[fbatch->nr] = xas.xa_index;
2085 		if (!folio_batch_add(fbatch, folio))
2086 			break;
2087 	}
2088 	rcu_read_unlock();
2089 
2090 	return folio_batch_count(fbatch);
2091 }
2092 
2093 /**
2094  * find_lock_entries - Find a batch of pagecache entries.
2095  * @mapping:	The address_space to search.
2096  * @start:	The starting page cache index.
2097  * @end:	The final page index (inclusive).
2098  * @fbatch:	Where the resulting entries are placed.
2099  * @indices:	The cache indices of the entries in @fbatch.
2100  *
2101  * find_lock_entries() will return a batch of entries from @mapping.
2102  * Swap, shadow and DAX entries are included.  Folios are returned
2103  * locked and with an incremented refcount.  Folios which are locked
2104  * by somebody else or under writeback are skipped.  Folios which are
2105  * partially outside the range are not returned.
2106  *
2107  * The entries have ascending indexes.  The indices may not be consecutive
2108  * due to not-present entries, large folios, folios which could not be
2109  * locked or folios under writeback.
2110  *
2111  * Return: The number of entries which were found.
2112  */
2113 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2114 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2115 {
2116 	XA_STATE(xas, &mapping->i_pages, start);
2117 	struct folio *folio;
2118 
2119 	rcu_read_lock();
2120 	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2121 		if (!xa_is_value(folio)) {
2122 			if (folio->index < start)
2123 				goto put;
2124 			if (folio->index + folio_nr_pages(folio) - 1 > end)
2125 				goto put;
2126 			if (!folio_trylock(folio))
2127 				goto put;
2128 			if (folio->mapping != mapping ||
2129 			    folio_test_writeback(folio))
2130 				goto unlock;
2131 			VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2132 					folio);
2133 		}
2134 		indices[fbatch->nr] = xas.xa_index;
2135 		if (!folio_batch_add(fbatch, folio))
2136 			break;
2137 		continue;
2138 unlock:
2139 		folio_unlock(folio);
2140 put:
2141 		folio_put(folio);
2142 	}
2143 	rcu_read_unlock();
2144 
2145 	return folio_batch_count(fbatch);
2146 }
2147 
2148 static inline
2149 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2150 {
2151 	if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2152 		return false;
2153 	if (index >= max)
2154 		return false;
2155 	return index < folio->index + folio_nr_pages(folio) - 1;
2156 }
2157 
2158 /**
2159  * find_get_pages_range - gang pagecache lookup
2160  * @mapping:	The address_space to search
2161  * @start:	The starting page index
2162  * @end:	The final page index (inclusive)
2163  * @nr_pages:	The maximum number of pages
2164  * @pages:	Where the resulting pages are placed
2165  *
2166  * find_get_pages_range() will search for and return a group of up to @nr_pages
2167  * pages in the mapping starting at index @start and up to index @end
2168  * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
2169  * a reference against the returned pages.
2170  *
2171  * The search returns a group of mapping-contiguous pages with ascending
2172  * indexes.  There may be holes in the indices due to not-present pages.
2173  * We also update @start to index the next page for the traversal.
2174  *
2175  * Return: the number of pages which were found. If this number is
2176  * smaller than @nr_pages, the end of specified range has been
2177  * reached.
2178  */
2179 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2180 			      pgoff_t end, unsigned int nr_pages,
2181 			      struct page **pages)
2182 {
2183 	XA_STATE(xas, &mapping->i_pages, *start);
2184 	struct folio *folio;
2185 	unsigned ret = 0;
2186 
2187 	if (unlikely(!nr_pages))
2188 		return 0;
2189 
2190 	rcu_read_lock();
2191 	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2192 		/* Skip over shadow, swap and DAX entries */
2193 		if (xa_is_value(folio))
2194 			continue;
2195 
2196 again:
2197 		pages[ret] = folio_file_page(folio, xas.xa_index);
2198 		if (++ret == nr_pages) {
2199 			*start = xas.xa_index + 1;
2200 			goto out;
2201 		}
2202 		if (folio_more_pages(folio, xas.xa_index, end)) {
2203 			xas.xa_index++;
2204 			folio_ref_inc(folio);
2205 			goto again;
2206 		}
2207 	}
2208 
2209 	/*
2210 	 * We come here when there is no page beyond @end. We take care to not
2211 	 * overflow the index @start as it confuses some of the callers. This
2212 	 * breaks the iteration when there is a page at index -1 but that is
2213 	 * already broken anyway.
2214 	 */
2215 	if (end == (pgoff_t)-1)
2216 		*start = (pgoff_t)-1;
2217 	else
2218 		*start = end + 1;
2219 out:
2220 	rcu_read_unlock();
2221 
2222 	return ret;
2223 }
2224 
2225 /**
2226  * find_get_pages_contig - gang contiguous pagecache lookup
2227  * @mapping:	The address_space to search
2228  * @index:	The starting page index
2229  * @nr_pages:	The maximum number of pages
2230  * @pages:	Where the resulting pages are placed
2231  *
2232  * find_get_pages_contig() works exactly like find_get_pages(), except
2233  * that the returned number of pages are guaranteed to be contiguous.
2234  *
2235  * Return: the number of pages which were found.
2236  */
2237 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2238 			       unsigned int nr_pages, struct page **pages)
2239 {
2240 	XA_STATE(xas, &mapping->i_pages, index);
2241 	struct folio *folio;
2242 	unsigned int ret = 0;
2243 
2244 	if (unlikely(!nr_pages))
2245 		return 0;
2246 
2247 	rcu_read_lock();
2248 	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2249 		if (xas_retry(&xas, folio))
2250 			continue;
2251 		/*
2252 		 * If the entry has been swapped out, we can stop looking.
2253 		 * No current caller is looking for DAX entries.
2254 		 */
2255 		if (xa_is_value(folio))
2256 			break;
2257 
2258 		if (!folio_try_get_rcu(folio))
2259 			goto retry;
2260 
2261 		if (unlikely(folio != xas_reload(&xas)))
2262 			goto put_page;
2263 
2264 again:
2265 		pages[ret] = folio_file_page(folio, xas.xa_index);
2266 		if (++ret == nr_pages)
2267 			break;
2268 		if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
2269 			xas.xa_index++;
2270 			folio_ref_inc(folio);
2271 			goto again;
2272 		}
2273 		continue;
2274 put_page:
2275 		folio_put(folio);
2276 retry:
2277 		xas_reset(&xas);
2278 	}
2279 	rcu_read_unlock();
2280 	return ret;
2281 }
2282 EXPORT_SYMBOL(find_get_pages_contig);
2283 
2284 /**
2285  * find_get_pages_range_tag - Find and return head pages matching @tag.
2286  * @mapping:	the address_space to search
2287  * @index:	the starting page index
2288  * @end:	The final page index (inclusive)
2289  * @tag:	the tag index
2290  * @nr_pages:	the maximum number of pages
2291  * @pages:	where the resulting pages are placed
2292  *
2293  * Like find_get_pages(), except we only return head pages which are tagged
2294  * with @tag.  @index is updated to the index immediately after the last
2295  * page we return, ready for the next iteration.
2296  *
2297  * Return: the number of pages which were found.
2298  */
2299 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2300 			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2301 			struct page **pages)
2302 {
2303 	XA_STATE(xas, &mapping->i_pages, *index);
2304 	struct folio *folio;
2305 	unsigned ret = 0;
2306 
2307 	if (unlikely(!nr_pages))
2308 		return 0;
2309 
2310 	rcu_read_lock();
2311 	while ((folio = find_get_entry(&xas, end, tag))) {
2312 		/*
2313 		 * Shadow entries should never be tagged, but this iteration
2314 		 * is lockless so there is a window for page reclaim to evict
2315 		 * a page we saw tagged.  Skip over it.
2316 		 */
2317 		if (xa_is_value(folio))
2318 			continue;
2319 
2320 		pages[ret] = &folio->page;
2321 		if (++ret == nr_pages) {
2322 			*index = folio->index + folio_nr_pages(folio);
2323 			goto out;
2324 		}
2325 	}
2326 
2327 	/*
2328 	 * We come here when we got to @end. We take care to not overflow the
2329 	 * index @index as it confuses some of the callers. This breaks the
2330 	 * iteration when there is a page at index -1 but that is already
2331 	 * broken anyway.
2332 	 */
2333 	if (end == (pgoff_t)-1)
2334 		*index = (pgoff_t)-1;
2335 	else
2336 		*index = end + 1;
2337 out:
2338 	rcu_read_unlock();
2339 
2340 	return ret;
2341 }
2342 EXPORT_SYMBOL(find_get_pages_range_tag);
2343 
2344 /*
2345  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2346  * a _large_ part of the i/o request. Imagine the worst scenario:
2347  *
2348  *      ---R__________________________________________B__________
2349  *         ^ reading here                             ^ bad block(assume 4k)
2350  *
2351  * read(R) => miss => readahead(R...B) => media error => frustrating retries
2352  * => failing the whole request => read(R) => read(R+1) =>
2353  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2354  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2355  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2356  *
2357  * It is going insane. Fix it by quickly scaling down the readahead size.
2358  */
2359 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2360 {
2361 	ra->ra_pages /= 4;
2362 }
2363 
2364 /*
2365  * filemap_get_read_batch - Get a batch of folios for read
2366  *
2367  * Get a batch of folios which represent a contiguous range of bytes in
2368  * the file.  No exceptional entries will be returned.  If @index is in
2369  * the middle of a folio, the entire folio will be returned.  The last
2370  * folio in the batch may have the readahead flag set or the uptodate flag
2371  * clear so that the caller can take the appropriate action.
2372  */
2373 static void filemap_get_read_batch(struct address_space *mapping,
2374 		pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2375 {
2376 	XA_STATE(xas, &mapping->i_pages, index);
2377 	struct folio *folio;
2378 
2379 	rcu_read_lock();
2380 	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2381 		if (xas_retry(&xas, folio))
2382 			continue;
2383 		if (xas.xa_index > max || xa_is_value(folio))
2384 			break;
2385 		if (!folio_try_get_rcu(folio))
2386 			goto retry;
2387 
2388 		if (unlikely(folio != xas_reload(&xas)))
2389 			goto put_folio;
2390 
2391 		if (!folio_batch_add(fbatch, folio))
2392 			break;
2393 		if (!folio_test_uptodate(folio))
2394 			break;
2395 		if (folio_test_readahead(folio))
2396 			break;
2397 		xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2398 		continue;
2399 put_folio:
2400 		folio_put(folio);
2401 retry:
2402 		xas_reset(&xas);
2403 	}
2404 	rcu_read_unlock();
2405 }
2406 
2407 static int filemap_read_folio(struct file *file, struct address_space *mapping,
2408 		struct folio *folio)
2409 {
2410 	int error;
2411 
2412 	/*
2413 	 * A previous I/O error may have been due to temporary failures,
2414 	 * eg. multipath errors.  PG_error will be set again if readpage
2415 	 * fails.
2416 	 */
2417 	folio_clear_error(folio);
2418 	/* Start the actual read. The read will unlock the page. */
2419 	error = mapping->a_ops->readpage(file, &folio->page);
2420 	if (error)
2421 		return error;
2422 
2423 	error = folio_wait_locked_killable(folio);
2424 	if (error)
2425 		return error;
2426 	if (folio_test_uptodate(folio))
2427 		return 0;
2428 	shrink_readahead_size_eio(&file->f_ra);
2429 	return -EIO;
2430 }
2431 
2432 static bool filemap_range_uptodate(struct address_space *mapping,
2433 		loff_t pos, struct iov_iter *iter, struct folio *folio)
2434 {
2435 	int count;
2436 
2437 	if (folio_test_uptodate(folio))
2438 		return true;
2439 	/* pipes can't handle partially uptodate pages */
2440 	if (iov_iter_is_pipe(iter))
2441 		return false;
2442 	if (!mapping->a_ops->is_partially_uptodate)
2443 		return false;
2444 	if (mapping->host->i_blkbits >= folio_shift(folio))
2445 		return false;
2446 
2447 	count = iter->count;
2448 	if (folio_pos(folio) > pos) {
2449 		count -= folio_pos(folio) - pos;
2450 		pos = 0;
2451 	} else {
2452 		pos -= folio_pos(folio);
2453 	}
2454 
2455 	return mapping->a_ops->is_partially_uptodate(&folio->page, pos, count);
2456 }
2457 
2458 static int filemap_update_page(struct kiocb *iocb,
2459 		struct address_space *mapping, struct iov_iter *iter,
2460 		struct folio *folio)
2461 {
2462 	int error;
2463 
2464 	if (iocb->ki_flags & IOCB_NOWAIT) {
2465 		if (!filemap_invalidate_trylock_shared(mapping))
2466 			return -EAGAIN;
2467 	} else {
2468 		filemap_invalidate_lock_shared(mapping);
2469 	}
2470 
2471 	if (!folio_trylock(folio)) {
2472 		error = -EAGAIN;
2473 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2474 			goto unlock_mapping;
2475 		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2476 			filemap_invalidate_unlock_shared(mapping);
2477 			/*
2478 			 * This is where we usually end up waiting for a
2479 			 * previously submitted readahead to finish.
2480 			 */
2481 			folio_put_wait_locked(folio, TASK_KILLABLE);
2482 			return AOP_TRUNCATED_PAGE;
2483 		}
2484 		error = __folio_lock_async(folio, iocb->ki_waitq);
2485 		if (error)
2486 			goto unlock_mapping;
2487 	}
2488 
2489 	error = AOP_TRUNCATED_PAGE;
2490 	if (!folio->mapping)
2491 		goto unlock;
2492 
2493 	error = 0;
2494 	if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2495 		goto unlock;
2496 
2497 	error = -EAGAIN;
2498 	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2499 		goto unlock;
2500 
2501 	error = filemap_read_folio(iocb->ki_filp, mapping, folio);
2502 	goto unlock_mapping;
2503 unlock:
2504 	folio_unlock(folio);
2505 unlock_mapping:
2506 	filemap_invalidate_unlock_shared(mapping);
2507 	if (error == AOP_TRUNCATED_PAGE)
2508 		folio_put(folio);
2509 	return error;
2510 }
2511 
2512 static int filemap_create_folio(struct file *file,
2513 		struct address_space *mapping, pgoff_t index,
2514 		struct folio_batch *fbatch)
2515 {
2516 	struct folio *folio;
2517 	int error;
2518 
2519 	folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2520 	if (!folio)
2521 		return -ENOMEM;
2522 
2523 	/*
2524 	 * Protect against truncate / hole punch. Grabbing invalidate_lock
2525 	 * here assures we cannot instantiate and bring uptodate new
2526 	 * pagecache folios after evicting page cache during truncate
2527 	 * and before actually freeing blocks.	Note that we could
2528 	 * release invalidate_lock after inserting the folio into
2529 	 * the page cache as the locked folio would then be enough to
2530 	 * synchronize with hole punching. But there are code paths
2531 	 * such as filemap_update_page() filling in partially uptodate
2532 	 * pages or ->readpages() that need to hold invalidate_lock
2533 	 * while mapping blocks for IO so let's hold the lock here as
2534 	 * well to keep locking rules simple.
2535 	 */
2536 	filemap_invalidate_lock_shared(mapping);
2537 	error = filemap_add_folio(mapping, folio, index,
2538 			mapping_gfp_constraint(mapping, GFP_KERNEL));
2539 	if (error == -EEXIST)
2540 		error = AOP_TRUNCATED_PAGE;
2541 	if (error)
2542 		goto error;
2543 
2544 	error = filemap_read_folio(file, mapping, folio);
2545 	if (error)
2546 		goto error;
2547 
2548 	filemap_invalidate_unlock_shared(mapping);
2549 	folio_batch_add(fbatch, folio);
2550 	return 0;
2551 error:
2552 	filemap_invalidate_unlock_shared(mapping);
2553 	folio_put(folio);
2554 	return error;
2555 }
2556 
2557 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2558 		struct address_space *mapping, struct folio *folio,
2559 		pgoff_t last_index)
2560 {
2561 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2562 
2563 	if (iocb->ki_flags & IOCB_NOIO)
2564 		return -EAGAIN;
2565 	page_cache_async_ra(&ractl, folio, last_index - folio->index);
2566 	return 0;
2567 }
2568 
2569 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2570 		struct folio_batch *fbatch)
2571 {
2572 	struct file *filp = iocb->ki_filp;
2573 	struct address_space *mapping = filp->f_mapping;
2574 	struct file_ra_state *ra = &filp->f_ra;
2575 	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2576 	pgoff_t last_index;
2577 	struct folio *folio;
2578 	int err = 0;
2579 
2580 	last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2581 retry:
2582 	if (fatal_signal_pending(current))
2583 		return -EINTR;
2584 
2585 	filemap_get_read_batch(mapping, index, last_index, fbatch);
2586 	if (!folio_batch_count(fbatch)) {
2587 		if (iocb->ki_flags & IOCB_NOIO)
2588 			return -EAGAIN;
2589 		page_cache_sync_readahead(mapping, ra, filp, index,
2590 				last_index - index);
2591 		filemap_get_read_batch(mapping, index, last_index, fbatch);
2592 	}
2593 	if (!folio_batch_count(fbatch)) {
2594 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2595 			return -EAGAIN;
2596 		err = filemap_create_folio(filp, mapping,
2597 				iocb->ki_pos >> PAGE_SHIFT, fbatch);
2598 		if (err == AOP_TRUNCATED_PAGE)
2599 			goto retry;
2600 		return err;
2601 	}
2602 
2603 	folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2604 	if (folio_test_readahead(folio)) {
2605 		err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2606 		if (err)
2607 			goto err;
2608 	}
2609 	if (!folio_test_uptodate(folio)) {
2610 		if ((iocb->ki_flags & IOCB_WAITQ) &&
2611 		    folio_batch_count(fbatch) > 1)
2612 			iocb->ki_flags |= IOCB_NOWAIT;
2613 		err = filemap_update_page(iocb, mapping, iter, folio);
2614 		if (err)
2615 			goto err;
2616 	}
2617 
2618 	return 0;
2619 err:
2620 	if (err < 0)
2621 		folio_put(folio);
2622 	if (likely(--fbatch->nr))
2623 		return 0;
2624 	if (err == AOP_TRUNCATED_PAGE)
2625 		goto retry;
2626 	return err;
2627 }
2628 
2629 /**
2630  * filemap_read - Read data from the page cache.
2631  * @iocb: The iocb to read.
2632  * @iter: Destination for the data.
2633  * @already_read: Number of bytes already read by the caller.
2634  *
2635  * Copies data from the page cache.  If the data is not currently present,
2636  * uses the readahead and readpage address_space operations to fetch it.
2637  *
2638  * Return: Total number of bytes copied, including those already read by
2639  * the caller.  If an error happens before any bytes are copied, returns
2640  * a negative error number.
2641  */
2642 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2643 		ssize_t already_read)
2644 {
2645 	struct file *filp = iocb->ki_filp;
2646 	struct file_ra_state *ra = &filp->f_ra;
2647 	struct address_space *mapping = filp->f_mapping;
2648 	struct inode *inode = mapping->host;
2649 	struct folio_batch fbatch;
2650 	int i, error = 0;
2651 	bool writably_mapped;
2652 	loff_t isize, end_offset;
2653 
2654 	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2655 		return 0;
2656 	if (unlikely(!iov_iter_count(iter)))
2657 		return 0;
2658 
2659 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2660 	folio_batch_init(&fbatch);
2661 
2662 	do {
2663 		cond_resched();
2664 
2665 		/*
2666 		 * If we've already successfully copied some data, then we
2667 		 * can no longer safely return -EIOCBQUEUED. Hence mark
2668 		 * an async read NOWAIT at that point.
2669 		 */
2670 		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2671 			iocb->ki_flags |= IOCB_NOWAIT;
2672 
2673 		if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2674 			break;
2675 
2676 		error = filemap_get_pages(iocb, iter, &fbatch);
2677 		if (error < 0)
2678 			break;
2679 
2680 		/*
2681 		 * i_size must be checked after we know the pages are Uptodate.
2682 		 *
2683 		 * Checking i_size after the check allows us to calculate
2684 		 * the correct value for "nr", which means the zero-filled
2685 		 * part of the page is not copied back to userspace (unless
2686 		 * another truncate extends the file - this is desired though).
2687 		 */
2688 		isize = i_size_read(inode);
2689 		if (unlikely(iocb->ki_pos >= isize))
2690 			goto put_folios;
2691 		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2692 
2693 		/*
2694 		 * Once we start copying data, we don't want to be touching any
2695 		 * cachelines that might be contended:
2696 		 */
2697 		writably_mapped = mapping_writably_mapped(mapping);
2698 
2699 		/*
2700 		 * When a sequential read accesses a page several times, only
2701 		 * mark it as accessed the first time.
2702 		 */
2703 		if (iocb->ki_pos >> PAGE_SHIFT !=
2704 		    ra->prev_pos >> PAGE_SHIFT)
2705 			folio_mark_accessed(fbatch.folios[0]);
2706 
2707 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2708 			struct folio *folio = fbatch.folios[i];
2709 			size_t fsize = folio_size(folio);
2710 			size_t offset = iocb->ki_pos & (fsize - 1);
2711 			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2712 					     fsize - offset);
2713 			size_t copied;
2714 
2715 			if (end_offset < folio_pos(folio))
2716 				break;
2717 			if (i > 0)
2718 				folio_mark_accessed(folio);
2719 			/*
2720 			 * If users can be writing to this folio using arbitrary
2721 			 * virtual addresses, take care of potential aliasing
2722 			 * before reading the folio on the kernel side.
2723 			 */
2724 			if (writably_mapped)
2725 				flush_dcache_folio(folio);
2726 
2727 			copied = copy_folio_to_iter(folio, offset, bytes, iter);
2728 
2729 			already_read += copied;
2730 			iocb->ki_pos += copied;
2731 			ra->prev_pos = iocb->ki_pos;
2732 
2733 			if (copied < bytes) {
2734 				error = -EFAULT;
2735 				break;
2736 			}
2737 		}
2738 put_folios:
2739 		for (i = 0; i < folio_batch_count(&fbatch); i++)
2740 			folio_put(fbatch.folios[i]);
2741 		folio_batch_init(&fbatch);
2742 	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2743 
2744 	file_accessed(filp);
2745 
2746 	return already_read ? already_read : error;
2747 }
2748 EXPORT_SYMBOL_GPL(filemap_read);
2749 
2750 /**
2751  * generic_file_read_iter - generic filesystem read routine
2752  * @iocb:	kernel I/O control block
2753  * @iter:	destination for the data read
2754  *
2755  * This is the "read_iter()" routine for all filesystems
2756  * that can use the page cache directly.
2757  *
2758  * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2759  * be returned when no data can be read without waiting for I/O requests
2760  * to complete; it doesn't prevent readahead.
2761  *
2762  * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2763  * requests shall be made for the read or for readahead.  When no data
2764  * can be read, -EAGAIN shall be returned.  When readahead would be
2765  * triggered, a partial, possibly empty read shall be returned.
2766  *
2767  * Return:
2768  * * number of bytes copied, even for partial reads
2769  * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2770  */
2771 ssize_t
2772 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2773 {
2774 	size_t count = iov_iter_count(iter);
2775 	ssize_t retval = 0;
2776 
2777 	if (!count)
2778 		return 0; /* skip atime */
2779 
2780 	if (iocb->ki_flags & IOCB_DIRECT) {
2781 		struct file *file = iocb->ki_filp;
2782 		struct address_space *mapping = file->f_mapping;
2783 		struct inode *inode = mapping->host;
2784 
2785 		if (iocb->ki_flags & IOCB_NOWAIT) {
2786 			if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2787 						iocb->ki_pos + count - 1))
2788 				return -EAGAIN;
2789 		} else {
2790 			retval = filemap_write_and_wait_range(mapping,
2791 						iocb->ki_pos,
2792 					        iocb->ki_pos + count - 1);
2793 			if (retval < 0)
2794 				return retval;
2795 		}
2796 
2797 		file_accessed(file);
2798 
2799 		retval = mapping->a_ops->direct_IO(iocb, iter);
2800 		if (retval >= 0) {
2801 			iocb->ki_pos += retval;
2802 			count -= retval;
2803 		}
2804 		if (retval != -EIOCBQUEUED)
2805 			iov_iter_revert(iter, count - iov_iter_count(iter));
2806 
2807 		/*
2808 		 * Btrfs can have a short DIO read if we encounter
2809 		 * compressed extents, so if there was an error, or if
2810 		 * we've already read everything we wanted to, or if
2811 		 * there was a short read because we hit EOF, go ahead
2812 		 * and return.  Otherwise fallthrough to buffered io for
2813 		 * the rest of the read.  Buffered reads will not work for
2814 		 * DAX files, so don't bother trying.
2815 		 */
2816 		if (retval < 0 || !count || IS_DAX(inode))
2817 			return retval;
2818 		if (iocb->ki_pos >= i_size_read(inode))
2819 			return retval;
2820 	}
2821 
2822 	return filemap_read(iocb, iter, retval);
2823 }
2824 EXPORT_SYMBOL(generic_file_read_iter);
2825 
2826 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2827 		struct address_space *mapping, struct folio *folio,
2828 		loff_t start, loff_t end, bool seek_data)
2829 {
2830 	const struct address_space_operations *ops = mapping->a_ops;
2831 	size_t offset, bsz = i_blocksize(mapping->host);
2832 
2833 	if (xa_is_value(folio) || folio_test_uptodate(folio))
2834 		return seek_data ? start : end;
2835 	if (!ops->is_partially_uptodate)
2836 		return seek_data ? end : start;
2837 
2838 	xas_pause(xas);
2839 	rcu_read_unlock();
2840 	folio_lock(folio);
2841 	if (unlikely(folio->mapping != mapping))
2842 		goto unlock;
2843 
2844 	offset = offset_in_folio(folio, start) & ~(bsz - 1);
2845 
2846 	do {
2847 		if (ops->is_partially_uptodate(&folio->page, offset, bsz) ==
2848 							seek_data)
2849 			break;
2850 		start = (start + bsz) & ~(bsz - 1);
2851 		offset += bsz;
2852 	} while (offset < folio_size(folio));
2853 unlock:
2854 	folio_unlock(folio);
2855 	rcu_read_lock();
2856 	return start;
2857 }
2858 
2859 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2860 {
2861 	if (xa_is_value(folio))
2862 		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2863 	return folio_size(folio);
2864 }
2865 
2866 /**
2867  * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2868  * @mapping: Address space to search.
2869  * @start: First byte to consider.
2870  * @end: Limit of search (exclusive).
2871  * @whence: Either SEEK_HOLE or SEEK_DATA.
2872  *
2873  * If the page cache knows which blocks contain holes and which blocks
2874  * contain data, your filesystem can use this function to implement
2875  * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
2876  * entirely memory-based such as tmpfs, and filesystems which support
2877  * unwritten extents.
2878  *
2879  * Return: The requested offset on success, or -ENXIO if @whence specifies
2880  * SEEK_DATA and there is no data after @start.  There is an implicit hole
2881  * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2882  * and @end contain data.
2883  */
2884 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2885 		loff_t end, int whence)
2886 {
2887 	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2888 	pgoff_t max = (end - 1) >> PAGE_SHIFT;
2889 	bool seek_data = (whence == SEEK_DATA);
2890 	struct folio *folio;
2891 
2892 	if (end <= start)
2893 		return -ENXIO;
2894 
2895 	rcu_read_lock();
2896 	while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
2897 		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2898 		size_t seek_size;
2899 
2900 		if (start < pos) {
2901 			if (!seek_data)
2902 				goto unlock;
2903 			start = pos;
2904 		}
2905 
2906 		seek_size = seek_folio_size(&xas, folio);
2907 		pos = round_up((u64)pos + 1, seek_size);
2908 		start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
2909 				seek_data);
2910 		if (start < pos)
2911 			goto unlock;
2912 		if (start >= end)
2913 			break;
2914 		if (seek_size > PAGE_SIZE)
2915 			xas_set(&xas, pos >> PAGE_SHIFT);
2916 		if (!xa_is_value(folio))
2917 			folio_put(folio);
2918 	}
2919 	if (seek_data)
2920 		start = -ENXIO;
2921 unlock:
2922 	rcu_read_unlock();
2923 	if (folio && !xa_is_value(folio))
2924 		folio_put(folio);
2925 	if (start > end)
2926 		return end;
2927 	return start;
2928 }
2929 
2930 #ifdef CONFIG_MMU
2931 #define MMAP_LOTSAMISS  (100)
2932 /*
2933  * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2934  * @vmf - the vm_fault for this fault.
2935  * @folio - the folio to lock.
2936  * @fpin - the pointer to the file we may pin (or is already pinned).
2937  *
2938  * This works similar to lock_folio_or_retry in that it can drop the
2939  * mmap_lock.  It differs in that it actually returns the folio locked
2940  * if it returns 1 and 0 if it couldn't lock the folio.  If we did have
2941  * to drop the mmap_lock then fpin will point to the pinned file and
2942  * needs to be fput()'ed at a later point.
2943  */
2944 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
2945 				     struct file **fpin)
2946 {
2947 	if (folio_trylock(folio))
2948 		return 1;
2949 
2950 	/*
2951 	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2952 	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2953 	 * is supposed to work. We have way too many special cases..
2954 	 */
2955 	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2956 		return 0;
2957 
2958 	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2959 	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2960 		if (__folio_lock_killable(folio)) {
2961 			/*
2962 			 * We didn't have the right flags to drop the mmap_lock,
2963 			 * but all fault_handlers only check for fatal signals
2964 			 * if we return VM_FAULT_RETRY, so we need to drop the
2965 			 * mmap_lock here and return 0 if we don't have a fpin.
2966 			 */
2967 			if (*fpin == NULL)
2968 				mmap_read_unlock(vmf->vma->vm_mm);
2969 			return 0;
2970 		}
2971 	} else
2972 		__folio_lock(folio);
2973 
2974 	return 1;
2975 }
2976 
2977 /*
2978  * Synchronous readahead happens when we don't even find a page in the page
2979  * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2980  * to drop the mmap sem we return the file that was pinned in order for us to do
2981  * that.  If we didn't pin a file then we return NULL.  The file that is
2982  * returned needs to be fput()'ed when we're done with it.
2983  */
2984 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2985 {
2986 	struct file *file = vmf->vma->vm_file;
2987 	struct file_ra_state *ra = &file->f_ra;
2988 	struct address_space *mapping = file->f_mapping;
2989 	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2990 	struct file *fpin = NULL;
2991 	unsigned int mmap_miss;
2992 
2993 	/* If we don't want any read-ahead, don't bother */
2994 	if (vmf->vma->vm_flags & VM_RAND_READ)
2995 		return fpin;
2996 	if (!ra->ra_pages)
2997 		return fpin;
2998 
2999 	if (vmf->vma->vm_flags & VM_SEQ_READ) {
3000 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3001 		page_cache_sync_ra(&ractl, ra->ra_pages);
3002 		return fpin;
3003 	}
3004 
3005 	/* Avoid banging the cache line if not needed */
3006 	mmap_miss = READ_ONCE(ra->mmap_miss);
3007 	if (mmap_miss < MMAP_LOTSAMISS * 10)
3008 		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3009 
3010 	/*
3011 	 * Do we miss much more than hit in this file? If so,
3012 	 * stop bothering with read-ahead. It will only hurt.
3013 	 */
3014 	if (mmap_miss > MMAP_LOTSAMISS)
3015 		return fpin;
3016 
3017 	/*
3018 	 * mmap read-around
3019 	 */
3020 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3021 	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3022 	ra->size = ra->ra_pages;
3023 	ra->async_size = ra->ra_pages / 4;
3024 	ractl._index = ra->start;
3025 	do_page_cache_ra(&ractl, ra->size, ra->async_size);
3026 	return fpin;
3027 }
3028 
3029 /*
3030  * Asynchronous readahead happens when we find the page and PG_readahead,
3031  * so we want to possibly extend the readahead further.  We return the file that
3032  * was pinned if we have to drop the mmap_lock in order to do IO.
3033  */
3034 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3035 					    struct folio *folio)
3036 {
3037 	struct file *file = vmf->vma->vm_file;
3038 	struct file_ra_state *ra = &file->f_ra;
3039 	DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3040 	struct file *fpin = NULL;
3041 	unsigned int mmap_miss;
3042 
3043 	/* If we don't want any read-ahead, don't bother */
3044 	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3045 		return fpin;
3046 
3047 	mmap_miss = READ_ONCE(ra->mmap_miss);
3048 	if (mmap_miss)
3049 		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3050 
3051 	if (folio_test_readahead(folio)) {
3052 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3053 		page_cache_async_ra(&ractl, folio, ra->ra_pages);
3054 	}
3055 	return fpin;
3056 }
3057 
3058 /**
3059  * filemap_fault - read in file data for page fault handling
3060  * @vmf:	struct vm_fault containing details of the fault
3061  *
3062  * filemap_fault() is invoked via the vma operations vector for a
3063  * mapped memory region to read in file data during a page fault.
3064  *
3065  * The goto's are kind of ugly, but this streamlines the normal case of having
3066  * it in the page cache, and handles the special cases reasonably without
3067  * having a lot of duplicated code.
3068  *
3069  * vma->vm_mm->mmap_lock must be held on entry.
3070  *
3071  * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3072  * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3073  *
3074  * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3075  * has not been released.
3076  *
3077  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3078  *
3079  * Return: bitwise-OR of %VM_FAULT_ codes.
3080  */
3081 vm_fault_t filemap_fault(struct vm_fault *vmf)
3082 {
3083 	int error;
3084 	struct file *file = vmf->vma->vm_file;
3085 	struct file *fpin = NULL;
3086 	struct address_space *mapping = file->f_mapping;
3087 	struct inode *inode = mapping->host;
3088 	pgoff_t max_idx, index = vmf->pgoff;
3089 	struct folio *folio;
3090 	vm_fault_t ret = 0;
3091 	bool mapping_locked = false;
3092 
3093 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3094 	if (unlikely(index >= max_idx))
3095 		return VM_FAULT_SIGBUS;
3096 
3097 	/*
3098 	 * Do we have something in the page cache already?
3099 	 */
3100 	folio = filemap_get_folio(mapping, index);
3101 	if (likely(folio)) {
3102 		/*
3103 		 * We found the page, so try async readahead before waiting for
3104 		 * the lock.
3105 		 */
3106 		if (!(vmf->flags & FAULT_FLAG_TRIED))
3107 			fpin = do_async_mmap_readahead(vmf, folio);
3108 		if (unlikely(!folio_test_uptodate(folio))) {
3109 			filemap_invalidate_lock_shared(mapping);
3110 			mapping_locked = true;
3111 		}
3112 	} else {
3113 		/* No page in the page cache at all */
3114 		count_vm_event(PGMAJFAULT);
3115 		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3116 		ret = VM_FAULT_MAJOR;
3117 		fpin = do_sync_mmap_readahead(vmf);
3118 retry_find:
3119 		/*
3120 		 * See comment in filemap_create_folio() why we need
3121 		 * invalidate_lock
3122 		 */
3123 		if (!mapping_locked) {
3124 			filemap_invalidate_lock_shared(mapping);
3125 			mapping_locked = true;
3126 		}
3127 		folio = __filemap_get_folio(mapping, index,
3128 					  FGP_CREAT|FGP_FOR_MMAP,
3129 					  vmf->gfp_mask);
3130 		if (!folio) {
3131 			if (fpin)
3132 				goto out_retry;
3133 			filemap_invalidate_unlock_shared(mapping);
3134 			return VM_FAULT_OOM;
3135 		}
3136 	}
3137 
3138 	if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3139 		goto out_retry;
3140 
3141 	/* Did it get truncated? */
3142 	if (unlikely(folio->mapping != mapping)) {
3143 		folio_unlock(folio);
3144 		folio_put(folio);
3145 		goto retry_find;
3146 	}
3147 	VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3148 
3149 	/*
3150 	 * We have a locked page in the page cache, now we need to check
3151 	 * that it's up-to-date. If not, it is going to be due to an error.
3152 	 */
3153 	if (unlikely(!folio_test_uptodate(folio))) {
3154 		/*
3155 		 * The page was in cache and uptodate and now it is not.
3156 		 * Strange but possible since we didn't hold the page lock all
3157 		 * the time. Let's drop everything get the invalidate lock and
3158 		 * try again.
3159 		 */
3160 		if (!mapping_locked) {
3161 			folio_unlock(folio);
3162 			folio_put(folio);
3163 			goto retry_find;
3164 		}
3165 		goto page_not_uptodate;
3166 	}
3167 
3168 	/*
3169 	 * We've made it this far and we had to drop our mmap_lock, now is the
3170 	 * time to return to the upper layer and have it re-find the vma and
3171 	 * redo the fault.
3172 	 */
3173 	if (fpin) {
3174 		folio_unlock(folio);
3175 		goto out_retry;
3176 	}
3177 	if (mapping_locked)
3178 		filemap_invalidate_unlock_shared(mapping);
3179 
3180 	/*
3181 	 * Found the page and have a reference on it.
3182 	 * We must recheck i_size under page lock.
3183 	 */
3184 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3185 	if (unlikely(index >= max_idx)) {
3186 		folio_unlock(folio);
3187 		folio_put(folio);
3188 		return VM_FAULT_SIGBUS;
3189 	}
3190 
3191 	vmf->page = folio_file_page(folio, index);
3192 	return ret | VM_FAULT_LOCKED;
3193 
3194 page_not_uptodate:
3195 	/*
3196 	 * Umm, take care of errors if the page isn't up-to-date.
3197 	 * Try to re-read it _once_. We do this synchronously,
3198 	 * because there really aren't any performance issues here
3199 	 * and we need to check for errors.
3200 	 */
3201 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3202 	error = filemap_read_folio(file, mapping, folio);
3203 	if (fpin)
3204 		goto out_retry;
3205 	folio_put(folio);
3206 
3207 	if (!error || error == AOP_TRUNCATED_PAGE)
3208 		goto retry_find;
3209 	filemap_invalidate_unlock_shared(mapping);
3210 
3211 	return VM_FAULT_SIGBUS;
3212 
3213 out_retry:
3214 	/*
3215 	 * We dropped the mmap_lock, we need to return to the fault handler to
3216 	 * re-find the vma and come back and find our hopefully still populated
3217 	 * page.
3218 	 */
3219 	if (folio)
3220 		folio_put(folio);
3221 	if (mapping_locked)
3222 		filemap_invalidate_unlock_shared(mapping);
3223 	if (fpin)
3224 		fput(fpin);
3225 	return ret | VM_FAULT_RETRY;
3226 }
3227 EXPORT_SYMBOL(filemap_fault);
3228 
3229 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3230 {
3231 	struct mm_struct *mm = vmf->vma->vm_mm;
3232 
3233 	/* Huge page is mapped? No need to proceed. */
3234 	if (pmd_trans_huge(*vmf->pmd)) {
3235 		unlock_page(page);
3236 		put_page(page);
3237 		return true;
3238 	}
3239 
3240 	if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3241 		vm_fault_t ret = do_set_pmd(vmf, page);
3242 		if (!ret) {
3243 			/* The page is mapped successfully, reference consumed. */
3244 			unlock_page(page);
3245 			return true;
3246 		}
3247 	}
3248 
3249 	if (pmd_none(*vmf->pmd))
3250 		pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3251 
3252 	/* See comment in handle_pte_fault() */
3253 	if (pmd_devmap_trans_unstable(vmf->pmd)) {
3254 		unlock_page(page);
3255 		put_page(page);
3256 		return true;
3257 	}
3258 
3259 	return false;
3260 }
3261 
3262 static struct folio *next_uptodate_page(struct folio *folio,
3263 				       struct address_space *mapping,
3264 				       struct xa_state *xas, pgoff_t end_pgoff)
3265 {
3266 	unsigned long max_idx;
3267 
3268 	do {
3269 		if (!folio)
3270 			return NULL;
3271 		if (xas_retry(xas, folio))
3272 			continue;
3273 		if (xa_is_value(folio))
3274 			continue;
3275 		if (folio_test_locked(folio))
3276 			continue;
3277 		if (!folio_try_get_rcu(folio))
3278 			continue;
3279 		/* Has the page moved or been split? */
3280 		if (unlikely(folio != xas_reload(xas)))
3281 			goto skip;
3282 		if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3283 			goto skip;
3284 		if (!folio_trylock(folio))
3285 			goto skip;
3286 		if (folio->mapping != mapping)
3287 			goto unlock;
3288 		if (!folio_test_uptodate(folio))
3289 			goto unlock;
3290 		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3291 		if (xas->xa_index >= max_idx)
3292 			goto unlock;
3293 		return folio;
3294 unlock:
3295 		folio_unlock(folio);
3296 skip:
3297 		folio_put(folio);
3298 	} while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3299 
3300 	return NULL;
3301 }
3302 
3303 static inline struct folio *first_map_page(struct address_space *mapping,
3304 					  struct xa_state *xas,
3305 					  pgoff_t end_pgoff)
3306 {
3307 	return next_uptodate_page(xas_find(xas, end_pgoff),
3308 				  mapping, xas, end_pgoff);
3309 }
3310 
3311 static inline struct folio *next_map_page(struct address_space *mapping,
3312 					 struct xa_state *xas,
3313 					 pgoff_t end_pgoff)
3314 {
3315 	return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3316 				  mapping, xas, end_pgoff);
3317 }
3318 
3319 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3320 			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3321 {
3322 	struct vm_area_struct *vma = vmf->vma;
3323 	struct file *file = vma->vm_file;
3324 	struct address_space *mapping = file->f_mapping;
3325 	pgoff_t last_pgoff = start_pgoff;
3326 	unsigned long addr;
3327 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3328 	struct folio *folio;
3329 	struct page *page;
3330 	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3331 	vm_fault_t ret = 0;
3332 
3333 	rcu_read_lock();
3334 	folio = first_map_page(mapping, &xas, end_pgoff);
3335 	if (!folio)
3336 		goto out;
3337 
3338 	if (filemap_map_pmd(vmf, &folio->page)) {
3339 		ret = VM_FAULT_NOPAGE;
3340 		goto out;
3341 	}
3342 
3343 	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3344 	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3345 	do {
3346 again:
3347 		page = folio_file_page(folio, xas.xa_index);
3348 		if (PageHWPoison(page))
3349 			goto unlock;
3350 
3351 		if (mmap_miss > 0)
3352 			mmap_miss--;
3353 
3354 		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3355 		vmf->pte += xas.xa_index - last_pgoff;
3356 		last_pgoff = xas.xa_index;
3357 
3358 		if (!pte_none(*vmf->pte))
3359 			goto unlock;
3360 
3361 		/* We're about to handle the fault */
3362 		if (vmf->address == addr)
3363 			ret = VM_FAULT_NOPAGE;
3364 
3365 		do_set_pte(vmf, page, addr);
3366 		/* no need to invalidate: a not-present page won't be cached */
3367 		update_mmu_cache(vma, addr, vmf->pte);
3368 		if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3369 			xas.xa_index++;
3370 			folio_ref_inc(folio);
3371 			goto again;
3372 		}
3373 		folio_unlock(folio);
3374 		continue;
3375 unlock:
3376 		if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3377 			xas.xa_index++;
3378 			goto again;
3379 		}
3380 		folio_unlock(folio);
3381 		folio_put(folio);
3382 	} while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3383 	pte_unmap_unlock(vmf->pte, vmf->ptl);
3384 out:
3385 	rcu_read_unlock();
3386 	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3387 	return ret;
3388 }
3389 EXPORT_SYMBOL(filemap_map_pages);
3390 
3391 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3392 {
3393 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3394 	struct folio *folio = page_folio(vmf->page);
3395 	vm_fault_t ret = VM_FAULT_LOCKED;
3396 
3397 	sb_start_pagefault(mapping->host->i_sb);
3398 	file_update_time(vmf->vma->vm_file);
3399 	folio_lock(folio);
3400 	if (folio->mapping != mapping) {
3401 		folio_unlock(folio);
3402 		ret = VM_FAULT_NOPAGE;
3403 		goto out;
3404 	}
3405 	/*
3406 	 * We mark the folio dirty already here so that when freeze is in
3407 	 * progress, we are guaranteed that writeback during freezing will
3408 	 * see the dirty folio and writeprotect it again.
3409 	 */
3410 	folio_mark_dirty(folio);
3411 	folio_wait_stable(folio);
3412 out:
3413 	sb_end_pagefault(mapping->host->i_sb);
3414 	return ret;
3415 }
3416 
3417 const struct vm_operations_struct generic_file_vm_ops = {
3418 	.fault		= filemap_fault,
3419 	.map_pages	= filemap_map_pages,
3420 	.page_mkwrite	= filemap_page_mkwrite,
3421 };
3422 
3423 /* This is used for a general mmap of a disk file */
3424 
3425 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3426 {
3427 	struct address_space *mapping = file->f_mapping;
3428 
3429 	if (!mapping->a_ops->readpage)
3430 		return -ENOEXEC;
3431 	file_accessed(file);
3432 	vma->vm_ops = &generic_file_vm_ops;
3433 	return 0;
3434 }
3435 
3436 /*
3437  * This is for filesystems which do not implement ->writepage.
3438  */
3439 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3440 {
3441 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3442 		return -EINVAL;
3443 	return generic_file_mmap(file, vma);
3444 }
3445 #else
3446 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3447 {
3448 	return VM_FAULT_SIGBUS;
3449 }
3450 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3451 {
3452 	return -ENOSYS;
3453 }
3454 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3455 {
3456 	return -ENOSYS;
3457 }
3458 #endif /* CONFIG_MMU */
3459 
3460 EXPORT_SYMBOL(filemap_page_mkwrite);
3461 EXPORT_SYMBOL(generic_file_mmap);
3462 EXPORT_SYMBOL(generic_file_readonly_mmap);
3463 
3464 static struct folio *do_read_cache_folio(struct address_space *mapping,
3465 		pgoff_t index, filler_t filler, void *data, gfp_t gfp)
3466 {
3467 	struct folio *folio;
3468 	int err;
3469 repeat:
3470 	folio = filemap_get_folio(mapping, index);
3471 	if (!folio) {
3472 		folio = filemap_alloc_folio(gfp, 0);
3473 		if (!folio)
3474 			return ERR_PTR(-ENOMEM);
3475 		err = filemap_add_folio(mapping, folio, index, gfp);
3476 		if (unlikely(err)) {
3477 			folio_put(folio);
3478 			if (err == -EEXIST)
3479 				goto repeat;
3480 			/* Presumably ENOMEM for xarray node */
3481 			return ERR_PTR(err);
3482 		}
3483 
3484 filler:
3485 		if (filler)
3486 			err = filler(data, &folio->page);
3487 		else
3488 			err = mapping->a_ops->readpage(data, &folio->page);
3489 
3490 		if (err < 0) {
3491 			folio_put(folio);
3492 			return ERR_PTR(err);
3493 		}
3494 
3495 		folio_wait_locked(folio);
3496 		if (!folio_test_uptodate(folio)) {
3497 			folio_put(folio);
3498 			return ERR_PTR(-EIO);
3499 		}
3500 
3501 		goto out;
3502 	}
3503 	if (folio_test_uptodate(folio))
3504 		goto out;
3505 
3506 	if (!folio_trylock(folio)) {
3507 		folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3508 		goto repeat;
3509 	}
3510 
3511 	/* Folio was truncated from mapping */
3512 	if (!folio->mapping) {
3513 		folio_unlock(folio);
3514 		folio_put(folio);
3515 		goto repeat;
3516 	}
3517 
3518 	/* Someone else locked and filled the page in a very small window */
3519 	if (folio_test_uptodate(folio)) {
3520 		folio_unlock(folio);
3521 		goto out;
3522 	}
3523 
3524 	/*
3525 	 * A previous I/O error may have been due to temporary
3526 	 * failures.
3527 	 * Clear page error before actual read, PG_error will be
3528 	 * set again if read page fails.
3529 	 */
3530 	folio_clear_error(folio);
3531 	goto filler;
3532 
3533 out:
3534 	folio_mark_accessed(folio);
3535 	return folio;
3536 }
3537 
3538 /**
3539  * read_cache_folio - read into page cache, fill it if needed
3540  * @mapping:	the page's address_space
3541  * @index:	the page index
3542  * @filler:	function to perform the read
3543  * @data:	first arg to filler(data, page) function, often left as NULL
3544  *
3545  * Read into the page cache. If a page already exists, and PageUptodate() is
3546  * not set, try to fill the page and wait for it to become unlocked.
3547  *
3548  * If the page does not get brought uptodate, return -EIO.
3549  *
3550  * The function expects mapping->invalidate_lock to be already held.
3551  *
3552  * Return: up to date page on success, ERR_PTR() on failure.
3553  */
3554 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3555 		filler_t filler, void *data)
3556 {
3557 	return do_read_cache_folio(mapping, index, filler, data,
3558 			mapping_gfp_mask(mapping));
3559 }
3560 EXPORT_SYMBOL(read_cache_folio);
3561 
3562 static struct page *do_read_cache_page(struct address_space *mapping,
3563 		pgoff_t index, filler_t *filler, void *data, gfp_t gfp)
3564 {
3565 	struct folio *folio;
3566 
3567 	folio = do_read_cache_folio(mapping, index, filler, data, gfp);
3568 	if (IS_ERR(folio))
3569 		return &folio->page;
3570 	return folio_file_page(folio, index);
3571 }
3572 
3573 struct page *read_cache_page(struct address_space *mapping,
3574 				pgoff_t index, filler_t *filler, void *data)
3575 {
3576 	return do_read_cache_page(mapping, index, filler, data,
3577 			mapping_gfp_mask(mapping));
3578 }
3579 EXPORT_SYMBOL(read_cache_page);
3580 
3581 /**
3582  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3583  * @mapping:	the page's address_space
3584  * @index:	the page index
3585  * @gfp:	the page allocator flags to use if allocating
3586  *
3587  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3588  * any new page allocations done using the specified allocation flags.
3589  *
3590  * If the page does not get brought uptodate, return -EIO.
3591  *
3592  * The function expects mapping->invalidate_lock to be already held.
3593  *
3594  * Return: up to date page on success, ERR_PTR() on failure.
3595  */
3596 struct page *read_cache_page_gfp(struct address_space *mapping,
3597 				pgoff_t index,
3598 				gfp_t gfp)
3599 {
3600 	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3601 }
3602 EXPORT_SYMBOL(read_cache_page_gfp);
3603 
3604 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3605 				loff_t pos, unsigned len, unsigned flags,
3606 				struct page **pagep, void **fsdata)
3607 {
3608 	const struct address_space_operations *aops = mapping->a_ops;
3609 
3610 	return aops->write_begin(file, mapping, pos, len, flags,
3611 							pagep, fsdata);
3612 }
3613 EXPORT_SYMBOL(pagecache_write_begin);
3614 
3615 int pagecache_write_end(struct file *file, struct address_space *mapping,
3616 				loff_t pos, unsigned len, unsigned copied,
3617 				struct page *page, void *fsdata)
3618 {
3619 	const struct address_space_operations *aops = mapping->a_ops;
3620 
3621 	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3622 }
3623 EXPORT_SYMBOL(pagecache_write_end);
3624 
3625 /*
3626  * Warn about a page cache invalidation failure during a direct I/O write.
3627  */
3628 void dio_warn_stale_pagecache(struct file *filp)
3629 {
3630 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3631 	char pathname[128];
3632 	char *path;
3633 
3634 	errseq_set(&filp->f_mapping->wb_err, -EIO);
3635 	if (__ratelimit(&_rs)) {
3636 		path = file_path(filp, pathname, sizeof(pathname));
3637 		if (IS_ERR(path))
3638 			path = "(unknown)";
3639 		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3640 		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3641 			current->comm);
3642 	}
3643 }
3644 
3645 ssize_t
3646 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3647 {
3648 	struct file	*file = iocb->ki_filp;
3649 	struct address_space *mapping = file->f_mapping;
3650 	struct inode	*inode = mapping->host;
3651 	loff_t		pos = iocb->ki_pos;
3652 	ssize_t		written;
3653 	size_t		write_len;
3654 	pgoff_t		end;
3655 
3656 	write_len = iov_iter_count(from);
3657 	end = (pos + write_len - 1) >> PAGE_SHIFT;
3658 
3659 	if (iocb->ki_flags & IOCB_NOWAIT) {
3660 		/* If there are pages to writeback, return */
3661 		if (filemap_range_has_page(file->f_mapping, pos,
3662 					   pos + write_len - 1))
3663 			return -EAGAIN;
3664 	} else {
3665 		written = filemap_write_and_wait_range(mapping, pos,
3666 							pos + write_len - 1);
3667 		if (written)
3668 			goto out;
3669 	}
3670 
3671 	/*
3672 	 * After a write we want buffered reads to be sure to go to disk to get
3673 	 * the new data.  We invalidate clean cached page from the region we're
3674 	 * about to write.  We do this *before* the write so that we can return
3675 	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3676 	 */
3677 	written = invalidate_inode_pages2_range(mapping,
3678 					pos >> PAGE_SHIFT, end);
3679 	/*
3680 	 * If a page can not be invalidated, return 0 to fall back
3681 	 * to buffered write.
3682 	 */
3683 	if (written) {
3684 		if (written == -EBUSY)
3685 			return 0;
3686 		goto out;
3687 	}
3688 
3689 	written = mapping->a_ops->direct_IO(iocb, from);
3690 
3691 	/*
3692 	 * Finally, try again to invalidate clean pages which might have been
3693 	 * cached by non-direct readahead, or faulted in by get_user_pages()
3694 	 * if the source of the write was an mmap'ed region of the file
3695 	 * we're writing.  Either one is a pretty crazy thing to do,
3696 	 * so we don't support it 100%.  If this invalidation
3697 	 * fails, tough, the write still worked...
3698 	 *
3699 	 * Most of the time we do not need this since dio_complete() will do
3700 	 * the invalidation for us. However there are some file systems that
3701 	 * do not end up with dio_complete() being called, so let's not break
3702 	 * them by removing it completely.
3703 	 *
3704 	 * Noticeable example is a blkdev_direct_IO().
3705 	 *
3706 	 * Skip invalidation for async writes or if mapping has no pages.
3707 	 */
3708 	if (written > 0 && mapping->nrpages &&
3709 	    invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3710 		dio_warn_stale_pagecache(file);
3711 
3712 	if (written > 0) {
3713 		pos += written;
3714 		write_len -= written;
3715 		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3716 			i_size_write(inode, pos);
3717 			mark_inode_dirty(inode);
3718 		}
3719 		iocb->ki_pos = pos;
3720 	}
3721 	if (written != -EIOCBQUEUED)
3722 		iov_iter_revert(from, write_len - iov_iter_count(from));
3723 out:
3724 	return written;
3725 }
3726 EXPORT_SYMBOL(generic_file_direct_write);
3727 
3728 ssize_t generic_perform_write(struct file *file,
3729 				struct iov_iter *i, loff_t pos)
3730 {
3731 	struct address_space *mapping = file->f_mapping;
3732 	const struct address_space_operations *a_ops = mapping->a_ops;
3733 	long status = 0;
3734 	ssize_t written = 0;
3735 	unsigned int flags = 0;
3736 
3737 	do {
3738 		struct page *page;
3739 		unsigned long offset;	/* Offset into pagecache page */
3740 		unsigned long bytes;	/* Bytes to write to page */
3741 		size_t copied;		/* Bytes copied from user */
3742 		void *fsdata;
3743 
3744 		offset = (pos & (PAGE_SIZE - 1));
3745 		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3746 						iov_iter_count(i));
3747 
3748 again:
3749 		/*
3750 		 * Bring in the user page that we will copy from _first_.
3751 		 * Otherwise there's a nasty deadlock on copying from the
3752 		 * same page as we're writing to, without it being marked
3753 		 * up-to-date.
3754 		 */
3755 		if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3756 			status = -EFAULT;
3757 			break;
3758 		}
3759 
3760 		if (fatal_signal_pending(current)) {
3761 			status = -EINTR;
3762 			break;
3763 		}
3764 
3765 		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3766 						&page, &fsdata);
3767 		if (unlikely(status < 0))
3768 			break;
3769 
3770 		if (mapping_writably_mapped(mapping))
3771 			flush_dcache_page(page);
3772 
3773 		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3774 		flush_dcache_page(page);
3775 
3776 		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3777 						page, fsdata);
3778 		if (unlikely(status != copied)) {
3779 			iov_iter_revert(i, copied - max(status, 0L));
3780 			if (unlikely(status < 0))
3781 				break;
3782 		}
3783 		cond_resched();
3784 
3785 		if (unlikely(status == 0)) {
3786 			/*
3787 			 * A short copy made ->write_end() reject the
3788 			 * thing entirely.  Might be memory poisoning
3789 			 * halfway through, might be a race with munmap,
3790 			 * might be severe memory pressure.
3791 			 */
3792 			if (copied)
3793 				bytes = copied;
3794 			goto again;
3795 		}
3796 		pos += status;
3797 		written += status;
3798 
3799 		balance_dirty_pages_ratelimited(mapping);
3800 	} while (iov_iter_count(i));
3801 
3802 	return written ? written : status;
3803 }
3804 EXPORT_SYMBOL(generic_perform_write);
3805 
3806 /**
3807  * __generic_file_write_iter - write data to a file
3808  * @iocb:	IO state structure (file, offset, etc.)
3809  * @from:	iov_iter with data to write
3810  *
3811  * This function does all the work needed for actually writing data to a
3812  * file. It does all basic checks, removes SUID from the file, updates
3813  * modification times and calls proper subroutines depending on whether we
3814  * do direct IO or a standard buffered write.
3815  *
3816  * It expects i_rwsem to be grabbed unless we work on a block device or similar
3817  * object which does not need locking at all.
3818  *
3819  * This function does *not* take care of syncing data in case of O_SYNC write.
3820  * A caller has to handle it. This is mainly due to the fact that we want to
3821  * avoid syncing under i_rwsem.
3822  *
3823  * Return:
3824  * * number of bytes written, even for truncated writes
3825  * * negative error code if no data has been written at all
3826  */
3827 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3828 {
3829 	struct file *file = iocb->ki_filp;
3830 	struct address_space *mapping = file->f_mapping;
3831 	struct inode 	*inode = mapping->host;
3832 	ssize_t		written = 0;
3833 	ssize_t		err;
3834 	ssize_t		status;
3835 
3836 	/* We can write back this queue in page reclaim */
3837 	current->backing_dev_info = inode_to_bdi(inode);
3838 	err = file_remove_privs(file);
3839 	if (err)
3840 		goto out;
3841 
3842 	err = file_update_time(file);
3843 	if (err)
3844 		goto out;
3845 
3846 	if (iocb->ki_flags & IOCB_DIRECT) {
3847 		loff_t pos, endbyte;
3848 
3849 		written = generic_file_direct_write(iocb, from);
3850 		/*
3851 		 * If the write stopped short of completing, fall back to
3852 		 * buffered writes.  Some filesystems do this for writes to
3853 		 * holes, for example.  For DAX files, a buffered write will
3854 		 * not succeed (even if it did, DAX does not handle dirty
3855 		 * page-cache pages correctly).
3856 		 */
3857 		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3858 			goto out;
3859 
3860 		status = generic_perform_write(file, from, pos = iocb->ki_pos);
3861 		/*
3862 		 * If generic_perform_write() returned a synchronous error
3863 		 * then we want to return the number of bytes which were
3864 		 * direct-written, or the error code if that was zero.  Note
3865 		 * that this differs from normal direct-io semantics, which
3866 		 * will return -EFOO even if some bytes were written.
3867 		 */
3868 		if (unlikely(status < 0)) {
3869 			err = status;
3870 			goto out;
3871 		}
3872 		/*
3873 		 * We need to ensure that the page cache pages are written to
3874 		 * disk and invalidated to preserve the expected O_DIRECT
3875 		 * semantics.
3876 		 */
3877 		endbyte = pos + status - 1;
3878 		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3879 		if (err == 0) {
3880 			iocb->ki_pos = endbyte + 1;
3881 			written += status;
3882 			invalidate_mapping_pages(mapping,
3883 						 pos >> PAGE_SHIFT,
3884 						 endbyte >> PAGE_SHIFT);
3885 		} else {
3886 			/*
3887 			 * We don't know how much we wrote, so just return
3888 			 * the number of bytes which were direct-written
3889 			 */
3890 		}
3891 	} else {
3892 		written = generic_perform_write(file, from, iocb->ki_pos);
3893 		if (likely(written > 0))
3894 			iocb->ki_pos += written;
3895 	}
3896 out:
3897 	current->backing_dev_info = NULL;
3898 	return written ? written : err;
3899 }
3900 EXPORT_SYMBOL(__generic_file_write_iter);
3901 
3902 /**
3903  * generic_file_write_iter - write data to a file
3904  * @iocb:	IO state structure
3905  * @from:	iov_iter with data to write
3906  *
3907  * This is a wrapper around __generic_file_write_iter() to be used by most
3908  * filesystems. It takes care of syncing the file in case of O_SYNC file
3909  * and acquires i_rwsem as needed.
3910  * Return:
3911  * * negative error code if no data has been written at all of
3912  *   vfs_fsync_range() failed for a synchronous write
3913  * * number of bytes written, even for truncated writes
3914  */
3915 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3916 {
3917 	struct file *file = iocb->ki_filp;
3918 	struct inode *inode = file->f_mapping->host;
3919 	ssize_t ret;
3920 
3921 	inode_lock(inode);
3922 	ret = generic_write_checks(iocb, from);
3923 	if (ret > 0)
3924 		ret = __generic_file_write_iter(iocb, from);
3925 	inode_unlock(inode);
3926 
3927 	if (ret > 0)
3928 		ret = generic_write_sync(iocb, ret);
3929 	return ret;
3930 }
3931 EXPORT_SYMBOL(generic_file_write_iter);
3932 
3933 /**
3934  * filemap_release_folio() - Release fs-specific metadata on a folio.
3935  * @folio: The folio which the kernel is trying to free.
3936  * @gfp: Memory allocation flags (and I/O mode).
3937  *
3938  * The address_space is trying to release any data attached to a folio
3939  * (presumably at folio->private).
3940  *
3941  * This will also be called if the private_2 flag is set on a page,
3942  * indicating that the folio has other metadata associated with it.
3943  *
3944  * The @gfp argument specifies whether I/O may be performed to release
3945  * this page (__GFP_IO), and whether the call may block
3946  * (__GFP_RECLAIM & __GFP_FS).
3947  *
3948  * Return: %true if the release was successful, otherwise %false.
3949  */
3950 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
3951 {
3952 	struct address_space * const mapping = folio->mapping;
3953 
3954 	BUG_ON(!folio_test_locked(folio));
3955 	if (folio_test_writeback(folio))
3956 		return false;
3957 
3958 	if (mapping && mapping->a_ops->releasepage)
3959 		return mapping->a_ops->releasepage(&folio->page, gfp);
3960 	return try_to_free_buffers(&folio->page);
3961 }
3962 EXPORT_SYMBOL(filemap_release_folio);
3963