xref: /openbmc/linux/mm/filemap.c (revision 1f012283)
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/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/security.h>
34 #include <linux/cpuset.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.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 <asm/pgalloc.h>
45 #include <asm/tlbflush.h>
46 #include "internal.h"
47 
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/filemap.h>
50 
51 /*
52  * FIXME: remove all knowledge of the buffer layer from the core VM
53  */
54 #include <linux/buffer_head.h> /* for try_to_free_buffers */
55 
56 #include <asm/mman.h>
57 
58 /*
59  * Shared mappings implemented 30.11.1994. It's not fully working yet,
60  * though.
61  *
62  * Shared mappings now work. 15.8.1995  Bruno.
63  *
64  * finished 'unifying' the page and buffer cache and SMP-threaded the
65  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
66  *
67  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
68  */
69 
70 /*
71  * Lock ordering:
72  *
73  *  ->i_mmap_rwsem		(truncate_pagecache)
74  *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
75  *      ->swap_lock		(exclusive_swap_page, others)
76  *        ->i_pages lock
77  *
78  *  ->i_rwsem
79  *    ->invalidate_lock		(acquired by fs in truncate path)
80  *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
81  *
82  *  ->mmap_lock
83  *    ->i_mmap_rwsem
84  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
85  *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
86  *
87  *  ->mmap_lock
88  *    ->invalidate_lock		(filemap_fault)
89  *      ->lock_page		(filemap_fault, access_process_vm)
90  *
91  *  ->i_rwsem			(generic_perform_write)
92  *    ->mmap_lock		(fault_in_readable->do_page_fault)
93  *
94  *  bdi->wb.list_lock
95  *    sb_lock			(fs/fs-writeback.c)
96  *    ->i_pages lock		(__sync_single_inode)
97  *
98  *  ->i_mmap_rwsem
99  *    ->anon_vma.lock		(vma_adjust)
100  *
101  *  ->anon_vma.lock
102  *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
103  *
104  *  ->page_table_lock or pte_lock
105  *    ->swap_lock		(try_to_unmap_one)
106  *    ->private_lock		(try_to_unmap_one)
107  *    ->i_pages lock		(try_to_unmap_one)
108  *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
109  *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
110  *    ->private_lock		(page_remove_rmap->set_page_dirty)
111  *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
112  *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
113  *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
114  *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
115  *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
116  *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
117  *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
118  *
119  * ->i_mmap_rwsem
120  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
121  */
122 
123 static void page_cache_delete(struct address_space *mapping,
124 				   struct page *page, void *shadow)
125 {
126 	XA_STATE(xas, &mapping->i_pages, page->index);
127 	unsigned int nr = 1;
128 
129 	mapping_set_update(&xas, mapping);
130 
131 	/* hugetlb pages are represented by a single entry in the xarray */
132 	if (!PageHuge(page)) {
133 		xas_set_order(&xas, page->index, compound_order(page));
134 		nr = compound_nr(page);
135 	}
136 
137 	VM_BUG_ON_PAGE(!PageLocked(page), page);
138 	VM_BUG_ON_PAGE(PageTail(page), page);
139 	VM_BUG_ON_PAGE(nr != 1 && shadow, page);
140 
141 	xas_store(&xas, shadow);
142 	xas_init_marks(&xas);
143 
144 	page->mapping = NULL;
145 	/* Leave page->index set: truncation lookup relies upon it */
146 	mapping->nrpages -= nr;
147 }
148 
149 static void unaccount_page_cache_page(struct address_space *mapping,
150 				      struct page *page)
151 {
152 	int nr;
153 
154 	/*
155 	 * if we're uptodate, flush out into the cleancache, otherwise
156 	 * invalidate any existing cleancache entries.  We can't leave
157 	 * stale data around in the cleancache once our page is gone
158 	 */
159 	if (PageUptodate(page) && PageMappedToDisk(page))
160 		cleancache_put_page(page);
161 	else
162 		cleancache_invalidate_page(mapping, page);
163 
164 	VM_BUG_ON_PAGE(PageTail(page), page);
165 	VM_BUG_ON_PAGE(page_mapped(page), page);
166 	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
167 		int mapcount;
168 
169 		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
170 			 current->comm, page_to_pfn(page));
171 		dump_page(page, "still mapped when deleted");
172 		dump_stack();
173 		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
174 
175 		mapcount = page_mapcount(page);
176 		if (mapping_exiting(mapping) &&
177 		    page_count(page) >= mapcount + 2) {
178 			/*
179 			 * All vmas have already been torn down, so it's
180 			 * a good bet that actually the page is unmapped,
181 			 * and we'd prefer not to leak it: if we're wrong,
182 			 * some other bad page check should catch it later.
183 			 */
184 			page_mapcount_reset(page);
185 			page_ref_sub(page, mapcount);
186 		}
187 	}
188 
189 	/* hugetlb pages do not participate in page cache accounting. */
190 	if (PageHuge(page))
191 		return;
192 
193 	nr = thp_nr_pages(page);
194 
195 	__mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
196 	if (PageSwapBacked(page)) {
197 		__mod_lruvec_page_state(page, NR_SHMEM, -nr);
198 		if (PageTransHuge(page))
199 			__mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
200 	} else if (PageTransHuge(page)) {
201 		__mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
202 		filemap_nr_thps_dec(mapping);
203 	}
204 
205 	/*
206 	 * At this point page must be either written or cleaned by
207 	 * truncate.  Dirty page here signals a bug and loss of
208 	 * unwritten data.
209 	 *
210 	 * This fixes dirty accounting after removing the page entirely
211 	 * but leaves PageDirty set: it has no effect for truncated
212 	 * page and anyway will be cleared before returning page into
213 	 * buddy allocator.
214 	 */
215 	if (WARN_ON_ONCE(PageDirty(page)))
216 		account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
217 }
218 
219 /*
220  * Delete a page from the page cache and free it. Caller has to make
221  * sure the page is locked and that nobody else uses it - or that usage
222  * is safe.  The caller must hold the i_pages lock.
223  */
224 void __delete_from_page_cache(struct page *page, void *shadow)
225 {
226 	struct address_space *mapping = page->mapping;
227 
228 	trace_mm_filemap_delete_from_page_cache(page);
229 
230 	unaccount_page_cache_page(mapping, page);
231 	page_cache_delete(mapping, page, shadow);
232 }
233 
234 static void page_cache_free_page(struct address_space *mapping,
235 				struct page *page)
236 {
237 	void (*freepage)(struct page *);
238 
239 	freepage = mapping->a_ops->freepage;
240 	if (freepage)
241 		freepage(page);
242 
243 	if (PageTransHuge(page) && !PageHuge(page)) {
244 		page_ref_sub(page, thp_nr_pages(page));
245 		VM_BUG_ON_PAGE(page_count(page) <= 0, page);
246 	} else {
247 		put_page(page);
248 	}
249 }
250 
251 /**
252  * delete_from_page_cache - delete page from page cache
253  * @page: the page which the kernel is trying to remove from page cache
254  *
255  * This must be called only on pages that have been verified to be in the page
256  * cache and locked.  It will never put the page into the free list, the caller
257  * has a reference on the page.
258  */
259 void delete_from_page_cache(struct page *page)
260 {
261 	struct address_space *mapping = page_mapping(page);
262 
263 	BUG_ON(!PageLocked(page));
264 	spin_lock(&mapping->host->i_lock);
265 	xa_lock_irq(&mapping->i_pages);
266 	__delete_from_page_cache(page, NULL);
267 	xa_unlock_irq(&mapping->i_pages);
268 	if (mapping_shrinkable(mapping))
269 		inode_add_lru(mapping->host);
270 	spin_unlock(&mapping->host->i_lock);
271 
272 	page_cache_free_page(mapping, page);
273 }
274 EXPORT_SYMBOL(delete_from_page_cache);
275 
276 /*
277  * page_cache_delete_batch - delete several pages from page cache
278  * @mapping: the mapping to which pages belong
279  * @pvec: pagevec with pages to delete
280  *
281  * The function walks over mapping->i_pages and removes pages passed in @pvec
282  * from the mapping. The function expects @pvec to be sorted by page index
283  * and is optimised for it to be dense.
284  * It tolerates holes in @pvec (mapping entries at those indices are not
285  * modified). The function expects only THP head pages to be present in the
286  * @pvec.
287  *
288  * The function expects the i_pages lock to be held.
289  */
290 static void page_cache_delete_batch(struct address_space *mapping,
291 			     struct pagevec *pvec)
292 {
293 	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
294 	int total_pages = 0;
295 	int i = 0;
296 	struct page *page;
297 
298 	mapping_set_update(&xas, mapping);
299 	xas_for_each(&xas, page, ULONG_MAX) {
300 		if (i >= pagevec_count(pvec))
301 			break;
302 
303 		/* A swap/dax/shadow entry got inserted? Skip it. */
304 		if (xa_is_value(page))
305 			continue;
306 		/*
307 		 * A page got inserted in our range? Skip it. We have our
308 		 * pages locked so they are protected from being removed.
309 		 * If we see a page whose index is higher than ours, it
310 		 * means our page has been removed, which shouldn't be
311 		 * possible because we're holding the PageLock.
312 		 */
313 		if (page != pvec->pages[i]) {
314 			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
315 					page);
316 			continue;
317 		}
318 
319 		WARN_ON_ONCE(!PageLocked(page));
320 
321 		if (page->index == xas.xa_index)
322 			page->mapping = NULL;
323 		/* Leave page->index set: truncation lookup relies on it */
324 
325 		/*
326 		 * Move to the next page in the vector if this is a regular
327 		 * page or the index is of the last sub-page of this compound
328 		 * page.
329 		 */
330 		if (page->index + compound_nr(page) - 1 == xas.xa_index)
331 			i++;
332 		xas_store(&xas, NULL);
333 		total_pages++;
334 	}
335 	mapping->nrpages -= total_pages;
336 }
337 
338 void delete_from_page_cache_batch(struct address_space *mapping,
339 				  struct pagevec *pvec)
340 {
341 	int i;
342 
343 	if (!pagevec_count(pvec))
344 		return;
345 
346 	spin_lock(&mapping->host->i_lock);
347 	xa_lock_irq(&mapping->i_pages);
348 	for (i = 0; i < pagevec_count(pvec); i++) {
349 		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
350 
351 		unaccount_page_cache_page(mapping, pvec->pages[i]);
352 	}
353 	page_cache_delete_batch(mapping, pvec);
354 	xa_unlock_irq(&mapping->i_pages);
355 	if (mapping_shrinkable(mapping))
356 		inode_add_lru(mapping->host);
357 	spin_unlock(&mapping->host->i_lock);
358 
359 	for (i = 0; i < pagevec_count(pvec); i++)
360 		page_cache_free_page(mapping, pvec->pages[i]);
361 }
362 
363 int filemap_check_errors(struct address_space *mapping)
364 {
365 	int ret = 0;
366 	/* Check for outstanding write errors */
367 	if (test_bit(AS_ENOSPC, &mapping->flags) &&
368 	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
369 		ret = -ENOSPC;
370 	if (test_bit(AS_EIO, &mapping->flags) &&
371 	    test_and_clear_bit(AS_EIO, &mapping->flags))
372 		ret = -EIO;
373 	return ret;
374 }
375 EXPORT_SYMBOL(filemap_check_errors);
376 
377 static int filemap_check_and_keep_errors(struct address_space *mapping)
378 {
379 	/* Check for outstanding write errors */
380 	if (test_bit(AS_EIO, &mapping->flags))
381 		return -EIO;
382 	if (test_bit(AS_ENOSPC, &mapping->flags))
383 		return -ENOSPC;
384 	return 0;
385 }
386 
387 /**
388  * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
389  * @mapping:	address space structure to write
390  * @wbc:	the writeback_control controlling the writeout
391  *
392  * Call writepages on the mapping using the provided wbc to control the
393  * writeout.
394  *
395  * Return: %0 on success, negative error code otherwise.
396  */
397 int filemap_fdatawrite_wbc(struct address_space *mapping,
398 			   struct writeback_control *wbc)
399 {
400 	int ret;
401 
402 	if (!mapping_can_writeback(mapping) ||
403 	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
404 		return 0;
405 
406 	wbc_attach_fdatawrite_inode(wbc, mapping->host);
407 	ret = do_writepages(mapping, wbc);
408 	wbc_detach_inode(wbc);
409 	return ret;
410 }
411 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
412 
413 /**
414  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
415  * @mapping:	address space structure to write
416  * @start:	offset in bytes where the range starts
417  * @end:	offset in bytes where the range ends (inclusive)
418  * @sync_mode:	enable synchronous operation
419  *
420  * Start writeback against all of a mapping's dirty pages that lie
421  * within the byte offsets <start, end> inclusive.
422  *
423  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
424  * opposed to a regular memory cleansing writeback.  The difference between
425  * these two operations is that if a dirty page/buffer is encountered, it must
426  * be waited upon, and not just skipped over.
427  *
428  * Return: %0 on success, negative error code otherwise.
429  */
430 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
431 				loff_t end, int sync_mode)
432 {
433 	struct writeback_control wbc = {
434 		.sync_mode = sync_mode,
435 		.nr_to_write = LONG_MAX,
436 		.range_start = start,
437 		.range_end = end,
438 	};
439 
440 	return filemap_fdatawrite_wbc(mapping, &wbc);
441 }
442 
443 static inline int __filemap_fdatawrite(struct address_space *mapping,
444 	int sync_mode)
445 {
446 	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
447 }
448 
449 int filemap_fdatawrite(struct address_space *mapping)
450 {
451 	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
452 }
453 EXPORT_SYMBOL(filemap_fdatawrite);
454 
455 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
456 				loff_t end)
457 {
458 	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
459 }
460 EXPORT_SYMBOL(filemap_fdatawrite_range);
461 
462 /**
463  * filemap_flush - mostly a non-blocking flush
464  * @mapping:	target address_space
465  *
466  * This is a mostly non-blocking flush.  Not suitable for data-integrity
467  * purposes - I/O may not be started against all dirty pages.
468  *
469  * Return: %0 on success, negative error code otherwise.
470  */
471 int filemap_flush(struct address_space *mapping)
472 {
473 	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
474 }
475 EXPORT_SYMBOL(filemap_flush);
476 
477 /**
478  * filemap_range_has_page - check if a page exists in range.
479  * @mapping:           address space within which to check
480  * @start_byte:        offset in bytes where the range starts
481  * @end_byte:          offset in bytes where the range ends (inclusive)
482  *
483  * Find at least one page in the range supplied, usually used to check if
484  * direct writing in this range will trigger a writeback.
485  *
486  * Return: %true if at least one page exists in the specified range,
487  * %false otherwise.
488  */
489 bool filemap_range_has_page(struct address_space *mapping,
490 			   loff_t start_byte, loff_t end_byte)
491 {
492 	struct page *page;
493 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
494 	pgoff_t max = end_byte >> PAGE_SHIFT;
495 
496 	if (end_byte < start_byte)
497 		return false;
498 
499 	rcu_read_lock();
500 	for (;;) {
501 		page = xas_find(&xas, max);
502 		if (xas_retry(&xas, page))
503 			continue;
504 		/* Shadow entries don't count */
505 		if (xa_is_value(page))
506 			continue;
507 		/*
508 		 * We don't need to try to pin this page; we're about to
509 		 * release the RCU lock anyway.  It is enough to know that
510 		 * there was a page here recently.
511 		 */
512 		break;
513 	}
514 	rcu_read_unlock();
515 
516 	return page != NULL;
517 }
518 EXPORT_SYMBOL(filemap_range_has_page);
519 
520 static void __filemap_fdatawait_range(struct address_space *mapping,
521 				     loff_t start_byte, loff_t end_byte)
522 {
523 	pgoff_t index = start_byte >> PAGE_SHIFT;
524 	pgoff_t end = end_byte >> PAGE_SHIFT;
525 	struct pagevec pvec;
526 	int nr_pages;
527 
528 	if (end_byte < start_byte)
529 		return;
530 
531 	pagevec_init(&pvec);
532 	while (index <= end) {
533 		unsigned i;
534 
535 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
536 				end, PAGECACHE_TAG_WRITEBACK);
537 		if (!nr_pages)
538 			break;
539 
540 		for (i = 0; i < nr_pages; i++) {
541 			struct page *page = pvec.pages[i];
542 
543 			wait_on_page_writeback(page);
544 			ClearPageError(page);
545 		}
546 		pagevec_release(&pvec);
547 		cond_resched();
548 	}
549 }
550 
551 /**
552  * filemap_fdatawait_range - wait for writeback to complete
553  * @mapping:		address space structure to wait for
554  * @start_byte:		offset in bytes where the range starts
555  * @end_byte:		offset in bytes where the range ends (inclusive)
556  *
557  * Walk the list of under-writeback pages of the given address space
558  * in the given range and wait for all of them.  Check error status of
559  * the address space and return it.
560  *
561  * Since the error status of the address space is cleared by this function,
562  * callers are responsible for checking the return value and handling and/or
563  * reporting the error.
564  *
565  * Return: error status of the address space.
566  */
567 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
568 			    loff_t end_byte)
569 {
570 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
571 	return filemap_check_errors(mapping);
572 }
573 EXPORT_SYMBOL(filemap_fdatawait_range);
574 
575 /**
576  * filemap_fdatawait_range_keep_errors - wait for writeback to complete
577  * @mapping:		address space structure to wait for
578  * @start_byte:		offset in bytes where the range starts
579  * @end_byte:		offset in bytes where the range ends (inclusive)
580  *
581  * Walk the list of under-writeback pages of the given address space in the
582  * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
583  * this function does not clear error status of the address space.
584  *
585  * Use this function if callers don't handle errors themselves.  Expected
586  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
587  * fsfreeze(8)
588  */
589 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
590 		loff_t start_byte, loff_t end_byte)
591 {
592 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
593 	return filemap_check_and_keep_errors(mapping);
594 }
595 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
596 
597 /**
598  * file_fdatawait_range - wait for writeback to complete
599  * @file:		file pointing to address space structure to wait for
600  * @start_byte:		offset in bytes where the range starts
601  * @end_byte:		offset in bytes where the range ends (inclusive)
602  *
603  * Walk the list of under-writeback pages of the address space that file
604  * refers to, in the given range and wait for all of them.  Check error
605  * status of the address space vs. the file->f_wb_err cursor and return it.
606  *
607  * Since the error status of the file is advanced by this function,
608  * callers are responsible for checking the return value and handling and/or
609  * reporting the error.
610  *
611  * Return: error status of the address space vs. the file->f_wb_err cursor.
612  */
613 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
614 {
615 	struct address_space *mapping = file->f_mapping;
616 
617 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
618 	return file_check_and_advance_wb_err(file);
619 }
620 EXPORT_SYMBOL(file_fdatawait_range);
621 
622 /**
623  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
624  * @mapping: address space structure to wait for
625  *
626  * Walk the list of under-writeback pages of the given address space
627  * and wait for all of them.  Unlike filemap_fdatawait(), this function
628  * does not clear error status of the address space.
629  *
630  * Use this function if callers don't handle errors themselves.  Expected
631  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
632  * fsfreeze(8)
633  *
634  * Return: error status of the address space.
635  */
636 int filemap_fdatawait_keep_errors(struct address_space *mapping)
637 {
638 	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
639 	return filemap_check_and_keep_errors(mapping);
640 }
641 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
642 
643 /* Returns true if writeback might be needed or already in progress. */
644 static bool mapping_needs_writeback(struct address_space *mapping)
645 {
646 	return mapping->nrpages;
647 }
648 
649 static bool filemap_range_has_writeback(struct address_space *mapping,
650 					loff_t start_byte, loff_t end_byte)
651 {
652 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
653 	pgoff_t max = end_byte >> PAGE_SHIFT;
654 	struct page *page;
655 
656 	if (end_byte < start_byte)
657 		return false;
658 
659 	rcu_read_lock();
660 	xas_for_each(&xas, page, max) {
661 		if (xas_retry(&xas, page))
662 			continue;
663 		if (xa_is_value(page))
664 			continue;
665 		if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
666 			break;
667 	}
668 	rcu_read_unlock();
669 	return page != NULL;
670 
671 }
672 
673 /**
674  * filemap_range_needs_writeback - check if range potentially needs writeback
675  * @mapping:           address space within which to check
676  * @start_byte:        offset in bytes where the range starts
677  * @end_byte:          offset in bytes where the range ends (inclusive)
678  *
679  * Find at least one page in the range supplied, usually used to check if
680  * direct writing in this range will trigger a writeback. Used by O_DIRECT
681  * read/write with IOCB_NOWAIT, to see if the caller needs to do
682  * filemap_write_and_wait_range() before proceeding.
683  *
684  * Return: %true if the caller should do filemap_write_and_wait_range() before
685  * doing O_DIRECT to a page in this range, %false otherwise.
686  */
687 bool filemap_range_needs_writeback(struct address_space *mapping,
688 				   loff_t start_byte, loff_t end_byte)
689 {
690 	if (!mapping_needs_writeback(mapping))
691 		return false;
692 	if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
693 	    !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
694 		return false;
695 	return filemap_range_has_writeback(mapping, start_byte, end_byte);
696 }
697 EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);
698 
699 /**
700  * filemap_write_and_wait_range - write out & wait on a file range
701  * @mapping:	the address_space for the pages
702  * @lstart:	offset in bytes where the range starts
703  * @lend:	offset in bytes where the range ends (inclusive)
704  *
705  * Write out and wait upon file offsets lstart->lend, inclusive.
706  *
707  * Note that @lend is inclusive (describes the last byte to be written) so
708  * that this function can be used to write to the very end-of-file (end = -1).
709  *
710  * Return: error status of the address space.
711  */
712 int filemap_write_and_wait_range(struct address_space *mapping,
713 				 loff_t lstart, loff_t lend)
714 {
715 	int err = 0;
716 
717 	if (mapping_needs_writeback(mapping)) {
718 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
719 						 WB_SYNC_ALL);
720 		/*
721 		 * Even if the above returned error, the pages may be
722 		 * written partially (e.g. -ENOSPC), so we wait for it.
723 		 * But the -EIO is special case, it may indicate the worst
724 		 * thing (e.g. bug) happened, so we avoid waiting for it.
725 		 */
726 		if (err != -EIO) {
727 			int err2 = filemap_fdatawait_range(mapping,
728 						lstart, lend);
729 			if (!err)
730 				err = err2;
731 		} else {
732 			/* Clear any previously stored errors */
733 			filemap_check_errors(mapping);
734 		}
735 	} else {
736 		err = filemap_check_errors(mapping);
737 	}
738 	return err;
739 }
740 EXPORT_SYMBOL(filemap_write_and_wait_range);
741 
742 void __filemap_set_wb_err(struct address_space *mapping, int err)
743 {
744 	errseq_t eseq = errseq_set(&mapping->wb_err, err);
745 
746 	trace_filemap_set_wb_err(mapping, eseq);
747 }
748 EXPORT_SYMBOL(__filemap_set_wb_err);
749 
750 /**
751  * file_check_and_advance_wb_err - report wb error (if any) that was previously
752  * 				   and advance wb_err to current one
753  * @file: struct file on which the error is being reported
754  *
755  * When userland calls fsync (or something like nfsd does the equivalent), we
756  * want to report any writeback errors that occurred since the last fsync (or
757  * since the file was opened if there haven't been any).
758  *
759  * Grab the wb_err from the mapping. If it matches what we have in the file,
760  * then just quickly return 0. The file is all caught up.
761  *
762  * If it doesn't match, then take the mapping value, set the "seen" flag in
763  * it and try to swap it into place. If it works, or another task beat us
764  * to it with the new value, then update the f_wb_err and return the error
765  * portion. The error at this point must be reported via proper channels
766  * (a'la fsync, or NFS COMMIT operation, etc.).
767  *
768  * While we handle mapping->wb_err with atomic operations, the f_wb_err
769  * value is protected by the f_lock since we must ensure that it reflects
770  * the latest value swapped in for this file descriptor.
771  *
772  * Return: %0 on success, negative error code otherwise.
773  */
774 int file_check_and_advance_wb_err(struct file *file)
775 {
776 	int err = 0;
777 	errseq_t old = READ_ONCE(file->f_wb_err);
778 	struct address_space *mapping = file->f_mapping;
779 
780 	/* Locklessly handle the common case where nothing has changed */
781 	if (errseq_check(&mapping->wb_err, old)) {
782 		/* Something changed, must use slow path */
783 		spin_lock(&file->f_lock);
784 		old = file->f_wb_err;
785 		err = errseq_check_and_advance(&mapping->wb_err,
786 						&file->f_wb_err);
787 		trace_file_check_and_advance_wb_err(file, old);
788 		spin_unlock(&file->f_lock);
789 	}
790 
791 	/*
792 	 * We're mostly using this function as a drop in replacement for
793 	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
794 	 * that the legacy code would have had on these flags.
795 	 */
796 	clear_bit(AS_EIO, &mapping->flags);
797 	clear_bit(AS_ENOSPC, &mapping->flags);
798 	return err;
799 }
800 EXPORT_SYMBOL(file_check_and_advance_wb_err);
801 
802 /**
803  * file_write_and_wait_range - write out & wait on a file range
804  * @file:	file pointing to address_space with pages
805  * @lstart:	offset in bytes where the range starts
806  * @lend:	offset in bytes where the range ends (inclusive)
807  *
808  * Write out and wait upon file offsets lstart->lend, inclusive.
809  *
810  * Note that @lend is inclusive (describes the last byte to be written) so
811  * that this function can be used to write to the very end-of-file (end = -1).
812  *
813  * After writing out and waiting on the data, we check and advance the
814  * f_wb_err cursor to the latest value, and return any errors detected there.
815  *
816  * Return: %0 on success, negative error code otherwise.
817  */
818 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
819 {
820 	int err = 0, err2;
821 	struct address_space *mapping = file->f_mapping;
822 
823 	if (mapping_needs_writeback(mapping)) {
824 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
825 						 WB_SYNC_ALL);
826 		/* See comment of filemap_write_and_wait() */
827 		if (err != -EIO)
828 			__filemap_fdatawait_range(mapping, lstart, lend);
829 	}
830 	err2 = file_check_and_advance_wb_err(file);
831 	if (!err)
832 		err = err2;
833 	return err;
834 }
835 EXPORT_SYMBOL(file_write_and_wait_range);
836 
837 /**
838  * replace_page_cache_page - replace a pagecache page with a new one
839  * @old:	page to be replaced
840  * @new:	page to replace with
841  *
842  * This function replaces a page in the pagecache with a new one.  On
843  * success it acquires the pagecache reference for the new page and
844  * drops it for the old page.  Both the old and new pages must be
845  * locked.  This function does not add the new page to the LRU, the
846  * caller must do that.
847  *
848  * The remove + add is atomic.  This function cannot fail.
849  */
850 void replace_page_cache_page(struct page *old, struct page *new)
851 {
852 	struct folio *fold = page_folio(old);
853 	struct folio *fnew = page_folio(new);
854 	struct address_space *mapping = old->mapping;
855 	void (*freepage)(struct page *) = mapping->a_ops->freepage;
856 	pgoff_t offset = old->index;
857 	XA_STATE(xas, &mapping->i_pages, offset);
858 
859 	VM_BUG_ON_PAGE(!PageLocked(old), old);
860 	VM_BUG_ON_PAGE(!PageLocked(new), new);
861 	VM_BUG_ON_PAGE(new->mapping, new);
862 
863 	get_page(new);
864 	new->mapping = mapping;
865 	new->index = offset;
866 
867 	mem_cgroup_migrate(fold, fnew);
868 
869 	xas_lock_irq(&xas);
870 	xas_store(&xas, new);
871 
872 	old->mapping = NULL;
873 	/* hugetlb pages do not participate in page cache accounting. */
874 	if (!PageHuge(old))
875 		__dec_lruvec_page_state(old, NR_FILE_PAGES);
876 	if (!PageHuge(new))
877 		__inc_lruvec_page_state(new, NR_FILE_PAGES);
878 	if (PageSwapBacked(old))
879 		__dec_lruvec_page_state(old, NR_SHMEM);
880 	if (PageSwapBacked(new))
881 		__inc_lruvec_page_state(new, NR_SHMEM);
882 	xas_unlock_irq(&xas);
883 	if (freepage)
884 		freepage(old);
885 	put_page(old);
886 }
887 EXPORT_SYMBOL_GPL(replace_page_cache_page);
888 
889 noinline int __filemap_add_folio(struct address_space *mapping,
890 		struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
891 {
892 	XA_STATE(xas, &mapping->i_pages, index);
893 	int huge = folio_test_hugetlb(folio);
894 	int error;
895 	bool charged = false;
896 
897 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
898 	VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
899 	mapping_set_update(&xas, mapping);
900 
901 	folio_get(folio);
902 	folio->mapping = mapping;
903 	folio->index = index;
904 
905 	if (!huge) {
906 		error = mem_cgroup_charge(folio, NULL, gfp);
907 		VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
908 		if (error)
909 			goto error;
910 		charged = true;
911 	}
912 
913 	gfp &= GFP_RECLAIM_MASK;
914 
915 	do {
916 		unsigned int order = xa_get_order(xas.xa, xas.xa_index);
917 		void *entry, *old = NULL;
918 
919 		if (order > folio_order(folio))
920 			xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
921 					order, gfp);
922 		xas_lock_irq(&xas);
923 		xas_for_each_conflict(&xas, entry) {
924 			old = entry;
925 			if (!xa_is_value(entry)) {
926 				xas_set_err(&xas, -EEXIST);
927 				goto unlock;
928 			}
929 		}
930 
931 		if (old) {
932 			if (shadowp)
933 				*shadowp = old;
934 			/* entry may have been split before we acquired lock */
935 			order = xa_get_order(xas.xa, xas.xa_index);
936 			if (order > folio_order(folio)) {
937 				xas_split(&xas, old, order);
938 				xas_reset(&xas);
939 			}
940 		}
941 
942 		xas_store(&xas, folio);
943 		if (xas_error(&xas))
944 			goto unlock;
945 
946 		mapping->nrpages++;
947 
948 		/* hugetlb pages do not participate in page cache accounting */
949 		if (!huge)
950 			__lruvec_stat_add_folio(folio, NR_FILE_PAGES);
951 unlock:
952 		xas_unlock_irq(&xas);
953 	} while (xas_nomem(&xas, gfp));
954 
955 	if (xas_error(&xas)) {
956 		error = xas_error(&xas);
957 		if (charged)
958 			mem_cgroup_uncharge(folio);
959 		goto error;
960 	}
961 
962 	trace_mm_filemap_add_to_page_cache(&folio->page);
963 	return 0;
964 error:
965 	folio->mapping = NULL;
966 	/* Leave page->index set: truncation relies upon it */
967 	folio_put(folio);
968 	return error;
969 }
970 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
971 
972 /**
973  * add_to_page_cache_locked - add a locked page to the pagecache
974  * @page:	page to add
975  * @mapping:	the page's address_space
976  * @offset:	page index
977  * @gfp_mask:	page allocation mode
978  *
979  * This function is used to add a page to the pagecache. It must be locked.
980  * This function does not add the page to the LRU.  The caller must do that.
981  *
982  * Return: %0 on success, negative error code otherwise.
983  */
984 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
985 		pgoff_t offset, gfp_t gfp_mask)
986 {
987 	return __filemap_add_folio(mapping, page_folio(page), offset,
988 					  gfp_mask, NULL);
989 }
990 EXPORT_SYMBOL(add_to_page_cache_locked);
991 
992 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
993 				pgoff_t index, gfp_t gfp)
994 {
995 	void *shadow = NULL;
996 	int ret;
997 
998 	__folio_set_locked(folio);
999 	ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
1000 	if (unlikely(ret))
1001 		__folio_clear_locked(folio);
1002 	else {
1003 		/*
1004 		 * The folio might have been evicted from cache only
1005 		 * recently, in which case it should be activated like
1006 		 * any other repeatedly accessed folio.
1007 		 * The exception is folios getting rewritten; evicting other
1008 		 * data from the working set, only to cache data that will
1009 		 * get overwritten with something else, is a waste of memory.
1010 		 */
1011 		WARN_ON_ONCE(folio_test_active(folio));
1012 		if (!(gfp & __GFP_WRITE) && shadow)
1013 			workingset_refault(folio, shadow);
1014 		folio_add_lru(folio);
1015 	}
1016 	return ret;
1017 }
1018 EXPORT_SYMBOL_GPL(filemap_add_folio);
1019 
1020 #ifdef CONFIG_NUMA
1021 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
1022 {
1023 	int n;
1024 	struct folio *folio;
1025 
1026 	if (cpuset_do_page_mem_spread()) {
1027 		unsigned int cpuset_mems_cookie;
1028 		do {
1029 			cpuset_mems_cookie = read_mems_allowed_begin();
1030 			n = cpuset_mem_spread_node();
1031 			folio = __folio_alloc_node(gfp, order, n);
1032 		} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
1033 
1034 		return folio;
1035 	}
1036 	return folio_alloc(gfp, order);
1037 }
1038 EXPORT_SYMBOL(filemap_alloc_folio);
1039 #endif
1040 
1041 /*
1042  * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1043  *
1044  * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1045  *
1046  * @mapping1: the first mapping to lock
1047  * @mapping2: the second mapping to lock
1048  */
1049 void filemap_invalidate_lock_two(struct address_space *mapping1,
1050 				 struct address_space *mapping2)
1051 {
1052 	if (mapping1 > mapping2)
1053 		swap(mapping1, mapping2);
1054 	if (mapping1)
1055 		down_write(&mapping1->invalidate_lock);
1056 	if (mapping2 && mapping1 != mapping2)
1057 		down_write_nested(&mapping2->invalidate_lock, 1);
1058 }
1059 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1060 
1061 /*
1062  * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1063  *
1064  * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1065  *
1066  * @mapping1: the first mapping to unlock
1067  * @mapping2: the second mapping to unlock
1068  */
1069 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1070 				   struct address_space *mapping2)
1071 {
1072 	if (mapping1)
1073 		up_write(&mapping1->invalidate_lock);
1074 	if (mapping2 && mapping1 != mapping2)
1075 		up_write(&mapping2->invalidate_lock);
1076 }
1077 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1078 
1079 /*
1080  * In order to wait for pages to become available there must be
1081  * waitqueues associated with pages. By using a hash table of
1082  * waitqueues where the bucket discipline is to maintain all
1083  * waiters on the same queue and wake all when any of the pages
1084  * become available, and for the woken contexts to check to be
1085  * sure the appropriate page became available, this saves space
1086  * at a cost of "thundering herd" phenomena during rare hash
1087  * collisions.
1088  */
1089 #define PAGE_WAIT_TABLE_BITS 8
1090 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1091 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1092 
1093 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1094 {
1095 	return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1096 }
1097 
1098 void __init pagecache_init(void)
1099 {
1100 	int i;
1101 
1102 	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1103 		init_waitqueue_head(&folio_wait_table[i]);
1104 
1105 	page_writeback_init();
1106 }
1107 
1108 /*
1109  * The page wait code treats the "wait->flags" somewhat unusually, because
1110  * we have multiple different kinds of waits, not just the usual "exclusive"
1111  * one.
1112  *
1113  * We have:
1114  *
1115  *  (a) no special bits set:
1116  *
1117  *	We're just waiting for the bit to be released, and when a waker
1118  *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1119  *	and remove it from the wait queue.
1120  *
1121  *	Simple and straightforward.
1122  *
1123  *  (b) WQ_FLAG_EXCLUSIVE:
1124  *
1125  *	The waiter is waiting to get the lock, and only one waiter should
1126  *	be woken up to avoid any thundering herd behavior. We'll set the
1127  *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1128  *
1129  *	This is the traditional exclusive wait.
1130  *
1131  *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1132  *
1133  *	The waiter is waiting to get the bit, and additionally wants the
1134  *	lock to be transferred to it for fair lock behavior. If the lock
1135  *	cannot be taken, we stop walking the wait queue without waking
1136  *	the waiter.
1137  *
1138  *	This is the "fair lock handoff" case, and in addition to setting
1139  *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1140  *	that it now has the lock.
1141  */
1142 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1143 {
1144 	unsigned int flags;
1145 	struct wait_page_key *key = arg;
1146 	struct wait_page_queue *wait_page
1147 		= container_of(wait, struct wait_page_queue, wait);
1148 
1149 	if (!wake_page_match(wait_page, key))
1150 		return 0;
1151 
1152 	/*
1153 	 * If it's a lock handoff wait, we get the bit for it, and
1154 	 * stop walking (and do not wake it up) if we can't.
1155 	 */
1156 	flags = wait->flags;
1157 	if (flags & WQ_FLAG_EXCLUSIVE) {
1158 		if (test_bit(key->bit_nr, &key->folio->flags))
1159 			return -1;
1160 		if (flags & WQ_FLAG_CUSTOM) {
1161 			if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1162 				return -1;
1163 			flags |= WQ_FLAG_DONE;
1164 		}
1165 	}
1166 
1167 	/*
1168 	 * We are holding the wait-queue lock, but the waiter that
1169 	 * is waiting for this will be checking the flags without
1170 	 * any locking.
1171 	 *
1172 	 * So update the flags atomically, and wake up the waiter
1173 	 * afterwards to avoid any races. This store-release pairs
1174 	 * with the load-acquire in folio_wait_bit_common().
1175 	 */
1176 	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1177 	wake_up_state(wait->private, mode);
1178 
1179 	/*
1180 	 * Ok, we have successfully done what we're waiting for,
1181 	 * and we can unconditionally remove the wait entry.
1182 	 *
1183 	 * Note that this pairs with the "finish_wait()" in the
1184 	 * waiter, and has to be the absolute last thing we do.
1185 	 * After this list_del_init(&wait->entry) the wait entry
1186 	 * might be de-allocated and the process might even have
1187 	 * exited.
1188 	 */
1189 	list_del_init_careful(&wait->entry);
1190 	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1191 }
1192 
1193 static void folio_wake_bit(struct folio *folio, int bit_nr)
1194 {
1195 	wait_queue_head_t *q = folio_waitqueue(folio);
1196 	struct wait_page_key key;
1197 	unsigned long flags;
1198 	wait_queue_entry_t bookmark;
1199 
1200 	key.folio = folio;
1201 	key.bit_nr = bit_nr;
1202 	key.page_match = 0;
1203 
1204 	bookmark.flags = 0;
1205 	bookmark.private = NULL;
1206 	bookmark.func = NULL;
1207 	INIT_LIST_HEAD(&bookmark.entry);
1208 
1209 	spin_lock_irqsave(&q->lock, flags);
1210 	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1211 
1212 	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1213 		/*
1214 		 * Take a breather from holding the lock,
1215 		 * allow pages that finish wake up asynchronously
1216 		 * to acquire the lock and remove themselves
1217 		 * from wait queue
1218 		 */
1219 		spin_unlock_irqrestore(&q->lock, flags);
1220 		cpu_relax();
1221 		spin_lock_irqsave(&q->lock, flags);
1222 		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1223 	}
1224 
1225 	/*
1226 	 * It is possible for other pages to have collided on the waitqueue
1227 	 * hash, so in that case check for a page match. That prevents a long-
1228 	 * term waiter
1229 	 *
1230 	 * It is still possible to miss a case here, when we woke page waiters
1231 	 * and removed them from the waitqueue, but there are still other
1232 	 * page waiters.
1233 	 */
1234 	if (!waitqueue_active(q) || !key.page_match) {
1235 		folio_clear_waiters(folio);
1236 		/*
1237 		 * It's possible to miss clearing Waiters here, when we woke
1238 		 * our page waiters, but the hashed waitqueue has waiters for
1239 		 * other pages on it.
1240 		 *
1241 		 * That's okay, it's a rare case. The next waker will clear it.
1242 		 */
1243 	}
1244 	spin_unlock_irqrestore(&q->lock, flags);
1245 }
1246 
1247 static void folio_wake(struct folio *folio, int bit)
1248 {
1249 	if (!folio_test_waiters(folio))
1250 		return;
1251 	folio_wake_bit(folio, bit);
1252 }
1253 
1254 /*
1255  * A choice of three behaviors for folio_wait_bit_common():
1256  */
1257 enum behavior {
1258 	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1259 			 * __folio_lock() waiting on then setting PG_locked.
1260 			 */
1261 	SHARED,		/* Hold ref to page and check the bit when woken, like
1262 			 * wait_on_page_writeback() waiting on PG_writeback.
1263 			 */
1264 	DROP,		/* Drop ref to page before wait, no check when woken,
1265 			 * like put_and_wait_on_page_locked() on PG_locked.
1266 			 */
1267 };
1268 
1269 /*
1270  * Attempt to check (or get) the folio flag, and mark us done
1271  * if successful.
1272  */
1273 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1274 					struct wait_queue_entry *wait)
1275 {
1276 	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1277 		if (test_and_set_bit(bit_nr, &folio->flags))
1278 			return false;
1279 	} else if (test_bit(bit_nr, &folio->flags))
1280 		return false;
1281 
1282 	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1283 	return true;
1284 }
1285 
1286 /* How many times do we accept lock stealing from under a waiter? */
1287 int sysctl_page_lock_unfairness = 5;
1288 
1289 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1290 		int state, enum behavior behavior)
1291 {
1292 	wait_queue_head_t *q = folio_waitqueue(folio);
1293 	int unfairness = sysctl_page_lock_unfairness;
1294 	struct wait_page_queue wait_page;
1295 	wait_queue_entry_t *wait = &wait_page.wait;
1296 	bool thrashing = false;
1297 	bool delayacct = false;
1298 	unsigned long pflags;
1299 
1300 	if (bit_nr == PG_locked &&
1301 	    !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1302 		if (!folio_test_swapbacked(folio)) {
1303 			delayacct_thrashing_start();
1304 			delayacct = true;
1305 		}
1306 		psi_memstall_enter(&pflags);
1307 		thrashing = true;
1308 	}
1309 
1310 	init_wait(wait);
1311 	wait->func = wake_page_function;
1312 	wait_page.folio = folio;
1313 	wait_page.bit_nr = bit_nr;
1314 
1315 repeat:
1316 	wait->flags = 0;
1317 	if (behavior == EXCLUSIVE) {
1318 		wait->flags = WQ_FLAG_EXCLUSIVE;
1319 		if (--unfairness < 0)
1320 			wait->flags |= WQ_FLAG_CUSTOM;
1321 	}
1322 
1323 	/*
1324 	 * Do one last check whether we can get the
1325 	 * page bit synchronously.
1326 	 *
1327 	 * Do the folio_set_waiters() marking before that
1328 	 * to let any waker we _just_ missed know they
1329 	 * need to wake us up (otherwise they'll never
1330 	 * even go to the slow case that looks at the
1331 	 * page queue), and add ourselves to the wait
1332 	 * queue if we need to sleep.
1333 	 *
1334 	 * This part needs to be done under the queue
1335 	 * lock to avoid races.
1336 	 */
1337 	spin_lock_irq(&q->lock);
1338 	folio_set_waiters(folio);
1339 	if (!folio_trylock_flag(folio, bit_nr, wait))
1340 		__add_wait_queue_entry_tail(q, wait);
1341 	spin_unlock_irq(&q->lock);
1342 
1343 	/*
1344 	 * From now on, all the logic will be based on
1345 	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1346 	 * see whether the page bit testing has already
1347 	 * been done by the wake function.
1348 	 *
1349 	 * We can drop our reference to the folio.
1350 	 */
1351 	if (behavior == DROP)
1352 		folio_put(folio);
1353 
1354 	/*
1355 	 * Note that until the "finish_wait()", or until
1356 	 * we see the WQ_FLAG_WOKEN flag, we need to
1357 	 * be very careful with the 'wait->flags', because
1358 	 * we may race with a waker that sets them.
1359 	 */
1360 	for (;;) {
1361 		unsigned int flags;
1362 
1363 		set_current_state(state);
1364 
1365 		/* Loop until we've been woken or interrupted */
1366 		flags = smp_load_acquire(&wait->flags);
1367 		if (!(flags & WQ_FLAG_WOKEN)) {
1368 			if (signal_pending_state(state, current))
1369 				break;
1370 
1371 			io_schedule();
1372 			continue;
1373 		}
1374 
1375 		/* If we were non-exclusive, we're done */
1376 		if (behavior != EXCLUSIVE)
1377 			break;
1378 
1379 		/* If the waker got the lock for us, we're done */
1380 		if (flags & WQ_FLAG_DONE)
1381 			break;
1382 
1383 		/*
1384 		 * Otherwise, if we're getting the lock, we need to
1385 		 * try to get it ourselves.
1386 		 *
1387 		 * And if that fails, we'll have to retry this all.
1388 		 */
1389 		if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1390 			goto repeat;
1391 
1392 		wait->flags |= WQ_FLAG_DONE;
1393 		break;
1394 	}
1395 
1396 	/*
1397 	 * If a signal happened, this 'finish_wait()' may remove the last
1398 	 * waiter from the wait-queues, but the folio waiters bit will remain
1399 	 * set. That's ok. The next wakeup will take care of it, and trying
1400 	 * to do it here would be difficult and prone to races.
1401 	 */
1402 	finish_wait(q, wait);
1403 
1404 	if (thrashing) {
1405 		if (delayacct)
1406 			delayacct_thrashing_end();
1407 		psi_memstall_leave(&pflags);
1408 	}
1409 
1410 	/*
1411 	 * NOTE! The wait->flags weren't stable until we've done the
1412 	 * 'finish_wait()', and we could have exited the loop above due
1413 	 * to a signal, and had a wakeup event happen after the signal
1414 	 * test but before the 'finish_wait()'.
1415 	 *
1416 	 * So only after the finish_wait() can we reliably determine
1417 	 * if we got woken up or not, so we can now figure out the final
1418 	 * return value based on that state without races.
1419 	 *
1420 	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1421 	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1422 	 */
1423 	if (behavior == EXCLUSIVE)
1424 		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1425 
1426 	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1427 }
1428 
1429 void folio_wait_bit(struct folio *folio, int bit_nr)
1430 {
1431 	folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1432 }
1433 EXPORT_SYMBOL(folio_wait_bit);
1434 
1435 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1436 {
1437 	return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1438 }
1439 EXPORT_SYMBOL(folio_wait_bit_killable);
1440 
1441 /**
1442  * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1443  * @page: The page to wait for.
1444  * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1445  *
1446  * The caller should hold a reference on @page.  They expect the page to
1447  * become unlocked relatively soon, but do not wish to hold up migration
1448  * (for example) by holding the reference while waiting for the page to
1449  * come unlocked.  After this function returns, the caller should not
1450  * dereference @page.
1451  *
1452  * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1453  */
1454 int put_and_wait_on_page_locked(struct page *page, int state)
1455 {
1456 	return folio_wait_bit_common(page_folio(page), PG_locked, state,
1457 			DROP);
1458 }
1459 
1460 /**
1461  * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1462  * @folio: Folio defining the wait queue of interest
1463  * @waiter: Waiter to add to the queue
1464  *
1465  * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1466  */
1467 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1468 {
1469 	wait_queue_head_t *q = folio_waitqueue(folio);
1470 	unsigned long flags;
1471 
1472 	spin_lock_irqsave(&q->lock, flags);
1473 	__add_wait_queue_entry_tail(q, waiter);
1474 	folio_set_waiters(folio);
1475 	spin_unlock_irqrestore(&q->lock, flags);
1476 }
1477 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1478 
1479 #ifndef clear_bit_unlock_is_negative_byte
1480 
1481 /*
1482  * PG_waiters is the high bit in the same byte as PG_lock.
1483  *
1484  * On x86 (and on many other architectures), we can clear PG_lock and
1485  * test the sign bit at the same time. But if the architecture does
1486  * not support that special operation, we just do this all by hand
1487  * instead.
1488  *
1489  * The read of PG_waiters has to be after (or concurrently with) PG_locked
1490  * being cleared, but a memory barrier should be unnecessary since it is
1491  * in the same byte as PG_locked.
1492  */
1493 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1494 {
1495 	clear_bit_unlock(nr, mem);
1496 	/* smp_mb__after_atomic(); */
1497 	return test_bit(PG_waiters, mem);
1498 }
1499 
1500 #endif
1501 
1502 /**
1503  * folio_unlock - Unlock a locked folio.
1504  * @folio: The folio.
1505  *
1506  * Unlocks the folio and wakes up any thread sleeping on the page lock.
1507  *
1508  * Context: May be called from interrupt or process context.  May not be
1509  * called from NMI context.
1510  */
1511 void folio_unlock(struct folio *folio)
1512 {
1513 	/* Bit 7 allows x86 to check the byte's sign bit */
1514 	BUILD_BUG_ON(PG_waiters != 7);
1515 	BUILD_BUG_ON(PG_locked > 7);
1516 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1517 	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1518 		folio_wake_bit(folio, PG_locked);
1519 }
1520 EXPORT_SYMBOL(folio_unlock);
1521 
1522 /**
1523  * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1524  * @folio: The folio.
1525  *
1526  * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1527  * it.  The folio reference held for PG_private_2 being set is released.
1528  *
1529  * This is, for example, used when a netfs folio is being written to a local
1530  * disk cache, thereby allowing writes to the cache for the same folio to be
1531  * serialised.
1532  */
1533 void folio_end_private_2(struct folio *folio)
1534 {
1535 	VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1536 	clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1537 	folio_wake_bit(folio, PG_private_2);
1538 	folio_put(folio);
1539 }
1540 EXPORT_SYMBOL(folio_end_private_2);
1541 
1542 /**
1543  * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1544  * @folio: The folio to wait on.
1545  *
1546  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1547  */
1548 void folio_wait_private_2(struct folio *folio)
1549 {
1550 	while (folio_test_private_2(folio))
1551 		folio_wait_bit(folio, PG_private_2);
1552 }
1553 EXPORT_SYMBOL(folio_wait_private_2);
1554 
1555 /**
1556  * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1557  * @folio: The folio to wait on.
1558  *
1559  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1560  * fatal signal is received by the calling task.
1561  *
1562  * Return:
1563  * - 0 if successful.
1564  * - -EINTR if a fatal signal was encountered.
1565  */
1566 int folio_wait_private_2_killable(struct folio *folio)
1567 {
1568 	int ret = 0;
1569 
1570 	while (folio_test_private_2(folio)) {
1571 		ret = folio_wait_bit_killable(folio, PG_private_2);
1572 		if (ret < 0)
1573 			break;
1574 	}
1575 
1576 	return ret;
1577 }
1578 EXPORT_SYMBOL(folio_wait_private_2_killable);
1579 
1580 /**
1581  * folio_end_writeback - End writeback against a folio.
1582  * @folio: The folio.
1583  */
1584 void folio_end_writeback(struct folio *folio)
1585 {
1586 	/*
1587 	 * folio_test_clear_reclaim() could be used here but it is an
1588 	 * atomic operation and overkill in this particular case. Failing
1589 	 * to shuffle a folio marked for immediate reclaim is too mild
1590 	 * a gain to justify taking an atomic operation penalty at the
1591 	 * end of every folio writeback.
1592 	 */
1593 	if (folio_test_reclaim(folio)) {
1594 		folio_clear_reclaim(folio);
1595 		folio_rotate_reclaimable(folio);
1596 	}
1597 
1598 	/*
1599 	 * Writeback does not hold a folio reference of its own, relying
1600 	 * on truncation to wait for the clearing of PG_writeback.
1601 	 * But here we must make sure that the folio is not freed and
1602 	 * reused before the folio_wake().
1603 	 */
1604 	folio_get(folio);
1605 	if (!__folio_end_writeback(folio))
1606 		BUG();
1607 
1608 	smp_mb__after_atomic();
1609 	folio_wake(folio, PG_writeback);
1610 	acct_reclaim_writeback(folio);
1611 	folio_put(folio);
1612 }
1613 EXPORT_SYMBOL(folio_end_writeback);
1614 
1615 /*
1616  * After completing I/O on a page, call this routine to update the page
1617  * flags appropriately
1618  */
1619 void page_endio(struct page *page, bool is_write, int err)
1620 {
1621 	if (!is_write) {
1622 		if (!err) {
1623 			SetPageUptodate(page);
1624 		} else {
1625 			ClearPageUptodate(page);
1626 			SetPageError(page);
1627 		}
1628 		unlock_page(page);
1629 	} else {
1630 		if (err) {
1631 			struct address_space *mapping;
1632 
1633 			SetPageError(page);
1634 			mapping = page_mapping(page);
1635 			if (mapping)
1636 				mapping_set_error(mapping, err);
1637 		}
1638 		end_page_writeback(page);
1639 	}
1640 }
1641 EXPORT_SYMBOL_GPL(page_endio);
1642 
1643 /**
1644  * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1645  * @folio: The folio to lock
1646  */
1647 void __folio_lock(struct folio *folio)
1648 {
1649 	folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1650 				EXCLUSIVE);
1651 }
1652 EXPORT_SYMBOL(__folio_lock);
1653 
1654 int __folio_lock_killable(struct folio *folio)
1655 {
1656 	return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1657 					EXCLUSIVE);
1658 }
1659 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1660 
1661 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1662 {
1663 	struct wait_queue_head *q = folio_waitqueue(folio);
1664 	int ret = 0;
1665 
1666 	wait->folio = folio;
1667 	wait->bit_nr = PG_locked;
1668 
1669 	spin_lock_irq(&q->lock);
1670 	__add_wait_queue_entry_tail(q, &wait->wait);
1671 	folio_set_waiters(folio);
1672 	ret = !folio_trylock(folio);
1673 	/*
1674 	 * If we were successful now, we know we're still on the
1675 	 * waitqueue as we're still under the lock. This means it's
1676 	 * safe to remove and return success, we know the callback
1677 	 * isn't going to trigger.
1678 	 */
1679 	if (!ret)
1680 		__remove_wait_queue(q, &wait->wait);
1681 	else
1682 		ret = -EIOCBQUEUED;
1683 	spin_unlock_irq(&q->lock);
1684 	return ret;
1685 }
1686 
1687 /*
1688  * Return values:
1689  * true - folio is locked; mmap_lock is still held.
1690  * false - folio is not locked.
1691  *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1692  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1693  *     which case mmap_lock is still held.
1694  *
1695  * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1696  * with the folio locked and the mmap_lock unperturbed.
1697  */
1698 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1699 			 unsigned int flags)
1700 {
1701 	if (fault_flag_allow_retry_first(flags)) {
1702 		/*
1703 		 * CAUTION! In this case, mmap_lock is not released
1704 		 * even though return 0.
1705 		 */
1706 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1707 			return false;
1708 
1709 		mmap_read_unlock(mm);
1710 		if (flags & FAULT_FLAG_KILLABLE)
1711 			folio_wait_locked_killable(folio);
1712 		else
1713 			folio_wait_locked(folio);
1714 		return false;
1715 	}
1716 	if (flags & FAULT_FLAG_KILLABLE) {
1717 		bool ret;
1718 
1719 		ret = __folio_lock_killable(folio);
1720 		if (ret) {
1721 			mmap_read_unlock(mm);
1722 			return false;
1723 		}
1724 	} else {
1725 		__folio_lock(folio);
1726 	}
1727 
1728 	return true;
1729 }
1730 
1731 /**
1732  * page_cache_next_miss() - Find the next gap in the page cache.
1733  * @mapping: Mapping.
1734  * @index: Index.
1735  * @max_scan: Maximum range to search.
1736  *
1737  * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1738  * gap with the lowest index.
1739  *
1740  * This function may be called under the rcu_read_lock.  However, this will
1741  * not atomically search a snapshot of the cache at a single point in time.
1742  * For example, if a gap is created at index 5, then subsequently a gap is
1743  * created at index 10, page_cache_next_miss covering both indices may
1744  * return 10 if called under the rcu_read_lock.
1745  *
1746  * Return: The index of the gap if found, otherwise an index outside the
1747  * range specified (in which case 'return - index >= max_scan' will be true).
1748  * In the rare case of index wrap-around, 0 will be returned.
1749  */
1750 pgoff_t page_cache_next_miss(struct address_space *mapping,
1751 			     pgoff_t index, unsigned long max_scan)
1752 {
1753 	XA_STATE(xas, &mapping->i_pages, index);
1754 
1755 	while (max_scan--) {
1756 		void *entry = xas_next(&xas);
1757 		if (!entry || xa_is_value(entry))
1758 			break;
1759 		if (xas.xa_index == 0)
1760 			break;
1761 	}
1762 
1763 	return xas.xa_index;
1764 }
1765 EXPORT_SYMBOL(page_cache_next_miss);
1766 
1767 /**
1768  * page_cache_prev_miss() - Find the previous gap in the page cache.
1769  * @mapping: Mapping.
1770  * @index: Index.
1771  * @max_scan: Maximum range to search.
1772  *
1773  * Search the range [max(index - max_scan + 1, 0), index] for the
1774  * gap with the highest index.
1775  *
1776  * This function may be called under the rcu_read_lock.  However, this will
1777  * not atomically search a snapshot of the cache at a single point in time.
1778  * For example, if a gap is created at index 10, then subsequently a gap is
1779  * created at index 5, page_cache_prev_miss() covering both indices may
1780  * return 5 if called under the rcu_read_lock.
1781  *
1782  * Return: The index of the gap if found, otherwise an index outside the
1783  * range specified (in which case 'index - return >= max_scan' will be true).
1784  * In the rare case of wrap-around, ULONG_MAX will be returned.
1785  */
1786 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1787 			     pgoff_t index, unsigned long max_scan)
1788 {
1789 	XA_STATE(xas, &mapping->i_pages, index);
1790 
1791 	while (max_scan--) {
1792 		void *entry = xas_prev(&xas);
1793 		if (!entry || xa_is_value(entry))
1794 			break;
1795 		if (xas.xa_index == ULONG_MAX)
1796 			break;
1797 	}
1798 
1799 	return xas.xa_index;
1800 }
1801 EXPORT_SYMBOL(page_cache_prev_miss);
1802 
1803 /*
1804  * Lockless page cache protocol:
1805  * On the lookup side:
1806  * 1. Load the folio from i_pages
1807  * 2. Increment the refcount if it's not zero
1808  * 3. If the folio is not found by xas_reload(), put the refcount and retry
1809  *
1810  * On the removal side:
1811  * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1812  * B. Remove the page from i_pages
1813  * C. Return the page to the page allocator
1814  *
1815  * This means that any page may have its reference count temporarily
1816  * increased by a speculative page cache (or fast GUP) lookup as it can
1817  * be allocated by another user before the RCU grace period expires.
1818  * Because the refcount temporarily acquired here may end up being the
1819  * last refcount on the page, any page allocation must be freeable by
1820  * folio_put().
1821  */
1822 
1823 /*
1824  * mapping_get_entry - Get a page cache entry.
1825  * @mapping: the address_space to search
1826  * @index: The page cache index.
1827  *
1828  * Looks up the page cache entry at @mapping & @index.  If it is a folio,
1829  * it is returned with an increased refcount.  If it is a shadow entry
1830  * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1831  * it is returned without further action.
1832  *
1833  * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1834  */
1835 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1836 {
1837 	XA_STATE(xas, &mapping->i_pages, index);
1838 	struct folio *folio;
1839 
1840 	rcu_read_lock();
1841 repeat:
1842 	xas_reset(&xas);
1843 	folio = xas_load(&xas);
1844 	if (xas_retry(&xas, folio))
1845 		goto repeat;
1846 	/*
1847 	 * A shadow entry of a recently evicted page, or a swap entry from
1848 	 * shmem/tmpfs.  Return it without attempting to raise page count.
1849 	 */
1850 	if (!folio || xa_is_value(folio))
1851 		goto out;
1852 
1853 	if (!folio_try_get_rcu(folio))
1854 		goto repeat;
1855 
1856 	if (unlikely(folio != xas_reload(&xas))) {
1857 		folio_put(folio);
1858 		goto repeat;
1859 	}
1860 out:
1861 	rcu_read_unlock();
1862 
1863 	return folio;
1864 }
1865 
1866 /**
1867  * __filemap_get_folio - Find and get a reference to a folio.
1868  * @mapping: The address_space to search.
1869  * @index: The page index.
1870  * @fgp_flags: %FGP flags modify how the folio is returned.
1871  * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1872  *
1873  * Looks up the page cache entry at @mapping & @index.
1874  *
1875  * @fgp_flags can be zero or more of these flags:
1876  *
1877  * * %FGP_ACCESSED - The folio will be marked accessed.
1878  * * %FGP_LOCK - The folio is returned locked.
1879  * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1880  *   instead of allocating a new folio to replace it.
1881  * * %FGP_CREAT - If no page is present then a new page is allocated using
1882  *   @gfp and added to the page cache and the VM's LRU list.
1883  *   The page is returned locked and with an increased refcount.
1884  * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1885  *   page is already in cache.  If the page was allocated, unlock it before
1886  *   returning so the caller can do the same dance.
1887  * * %FGP_WRITE - The page will be written to by the caller.
1888  * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1889  * * %FGP_NOWAIT - Don't get blocked by page lock.
1890  * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1891  *
1892  * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1893  * if the %GFP flags specified for %FGP_CREAT are atomic.
1894  *
1895  * If there is a page cache page, it is returned with an increased refcount.
1896  *
1897  * Return: The found folio or %NULL otherwise.
1898  */
1899 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1900 		int fgp_flags, gfp_t gfp)
1901 {
1902 	struct folio *folio;
1903 
1904 repeat:
1905 	folio = mapping_get_entry(mapping, index);
1906 	if (xa_is_value(folio)) {
1907 		if (fgp_flags & FGP_ENTRY)
1908 			return folio;
1909 		folio = NULL;
1910 	}
1911 	if (!folio)
1912 		goto no_page;
1913 
1914 	if (fgp_flags & FGP_LOCK) {
1915 		if (fgp_flags & FGP_NOWAIT) {
1916 			if (!folio_trylock(folio)) {
1917 				folio_put(folio);
1918 				return NULL;
1919 			}
1920 		} else {
1921 			folio_lock(folio);
1922 		}
1923 
1924 		/* Has the page been truncated? */
1925 		if (unlikely(folio->mapping != mapping)) {
1926 			folio_unlock(folio);
1927 			folio_put(folio);
1928 			goto repeat;
1929 		}
1930 		VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1931 	}
1932 
1933 	if (fgp_flags & FGP_ACCESSED)
1934 		folio_mark_accessed(folio);
1935 	else if (fgp_flags & FGP_WRITE) {
1936 		/* Clear idle flag for buffer write */
1937 		if (folio_test_idle(folio))
1938 			folio_clear_idle(folio);
1939 	}
1940 
1941 	if (fgp_flags & FGP_STABLE)
1942 		folio_wait_stable(folio);
1943 no_page:
1944 	if (!folio && (fgp_flags & FGP_CREAT)) {
1945 		int err;
1946 		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1947 			gfp |= __GFP_WRITE;
1948 		if (fgp_flags & FGP_NOFS)
1949 			gfp &= ~__GFP_FS;
1950 
1951 		folio = filemap_alloc_folio(gfp, 0);
1952 		if (!folio)
1953 			return NULL;
1954 
1955 		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1956 			fgp_flags |= FGP_LOCK;
1957 
1958 		/* Init accessed so avoid atomic mark_page_accessed later */
1959 		if (fgp_flags & FGP_ACCESSED)
1960 			__folio_set_referenced(folio);
1961 
1962 		err = filemap_add_folio(mapping, folio, index, gfp);
1963 		if (unlikely(err)) {
1964 			folio_put(folio);
1965 			folio = NULL;
1966 			if (err == -EEXIST)
1967 				goto repeat;
1968 		}
1969 
1970 		/*
1971 		 * filemap_add_folio locks the page, and for mmap
1972 		 * we expect an unlocked page.
1973 		 */
1974 		if (folio && (fgp_flags & FGP_FOR_MMAP))
1975 			folio_unlock(folio);
1976 	}
1977 
1978 	return folio;
1979 }
1980 EXPORT_SYMBOL(__filemap_get_folio);
1981 
1982 static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1983 		xa_mark_t mark)
1984 {
1985 	struct page *page;
1986 
1987 retry:
1988 	if (mark == XA_PRESENT)
1989 		page = xas_find(xas, max);
1990 	else
1991 		page = xas_find_marked(xas, max, mark);
1992 
1993 	if (xas_retry(xas, page))
1994 		goto retry;
1995 	/*
1996 	 * A shadow entry of a recently evicted page, a swap
1997 	 * entry from shmem/tmpfs or a DAX entry.  Return it
1998 	 * without attempting to raise page count.
1999 	 */
2000 	if (!page || xa_is_value(page))
2001 		return page;
2002 
2003 	if (!page_cache_get_speculative(page))
2004 		goto reset;
2005 
2006 	/* Has the page moved or been split? */
2007 	if (unlikely(page != xas_reload(xas))) {
2008 		put_page(page);
2009 		goto reset;
2010 	}
2011 
2012 	return page;
2013 reset:
2014 	xas_reset(xas);
2015 	goto retry;
2016 }
2017 
2018 /**
2019  * find_get_entries - gang pagecache lookup
2020  * @mapping:	The address_space to search
2021  * @start:	The starting page cache index
2022  * @end:	The final page index (inclusive).
2023  * @pvec:	Where the resulting entries are placed.
2024  * @indices:	The cache indices corresponding to the entries in @entries
2025  *
2026  * find_get_entries() will search for and return a batch of entries in
2027  * the mapping.  The entries are placed in @pvec.  find_get_entries()
2028  * takes a reference on any actual pages it returns.
2029  *
2030  * The search returns a group of mapping-contiguous page cache entries
2031  * with ascending indexes.  There may be holes in the indices due to
2032  * not-present pages.
2033  *
2034  * Any shadow entries of evicted pages, or swap entries from
2035  * shmem/tmpfs, are included in the returned array.
2036  *
2037  * If it finds a Transparent Huge Page, head or tail, find_get_entries()
2038  * stops at that page: the caller is likely to have a better way to handle
2039  * the compound page as a whole, and then skip its extent, than repeatedly
2040  * calling find_get_entries() to return all its tails.
2041  *
2042  * Return: the number of pages and shadow entries which were found.
2043  */
2044 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2045 		pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2046 {
2047 	XA_STATE(xas, &mapping->i_pages, start);
2048 	struct page *page;
2049 	unsigned int ret = 0;
2050 	unsigned nr_entries = PAGEVEC_SIZE;
2051 
2052 	rcu_read_lock();
2053 	while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2054 		/*
2055 		 * Terminate early on finding a THP, to allow the caller to
2056 		 * handle it all at once; but continue if this is hugetlbfs.
2057 		 */
2058 		if (!xa_is_value(page) && PageTransHuge(page) &&
2059 				!PageHuge(page)) {
2060 			page = find_subpage(page, xas.xa_index);
2061 			nr_entries = ret + 1;
2062 		}
2063 
2064 		indices[ret] = xas.xa_index;
2065 		pvec->pages[ret] = page;
2066 		if (++ret == nr_entries)
2067 			break;
2068 	}
2069 	rcu_read_unlock();
2070 
2071 	pvec->nr = ret;
2072 	return ret;
2073 }
2074 
2075 /**
2076  * find_lock_entries - Find a batch of pagecache entries.
2077  * @mapping:	The address_space to search.
2078  * @start:	The starting page cache index.
2079  * @end:	The final page index (inclusive).
2080  * @pvec:	Where the resulting entries are placed.
2081  * @indices:	The cache indices of the entries in @pvec.
2082  *
2083  * find_lock_entries() will return a batch of entries from @mapping.
2084  * Swap, shadow and DAX entries are included.  Pages are returned
2085  * locked and with an incremented refcount.  Pages which are locked by
2086  * somebody else or under writeback are skipped.  Only the head page of
2087  * a THP is returned.  Pages which are partially outside the range are
2088  * not returned.
2089  *
2090  * The entries have ascending indexes.  The indices may not be consecutive
2091  * due to not-present entries, THP pages, pages which could not be locked
2092  * or pages under writeback.
2093  *
2094  * Return: The number of entries which were found.
2095  */
2096 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2097 		pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2098 {
2099 	XA_STATE(xas, &mapping->i_pages, start);
2100 	struct page *page;
2101 
2102 	rcu_read_lock();
2103 	while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2104 		if (!xa_is_value(page)) {
2105 			if (page->index < start)
2106 				goto put;
2107 			if (page->index + thp_nr_pages(page) - 1 > end)
2108 				goto put;
2109 			if (!trylock_page(page))
2110 				goto put;
2111 			if (page->mapping != mapping || PageWriteback(page))
2112 				goto unlock;
2113 			VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
2114 					page);
2115 		}
2116 		indices[pvec->nr] = xas.xa_index;
2117 		if (!pagevec_add(pvec, page))
2118 			break;
2119 		goto next;
2120 unlock:
2121 		unlock_page(page);
2122 put:
2123 		put_page(page);
2124 next:
2125 		if (!xa_is_value(page) && PageTransHuge(page)) {
2126 			unsigned int nr_pages = thp_nr_pages(page);
2127 
2128 			/* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
2129 			xas_set(&xas, page->index + nr_pages);
2130 			if (xas.xa_index < nr_pages)
2131 				break;
2132 		}
2133 	}
2134 	rcu_read_unlock();
2135 
2136 	return pagevec_count(pvec);
2137 }
2138 
2139 /**
2140  * find_get_pages_range - gang pagecache lookup
2141  * @mapping:	The address_space to search
2142  * @start:	The starting page index
2143  * @end:	The final page index (inclusive)
2144  * @nr_pages:	The maximum number of pages
2145  * @pages:	Where the resulting pages are placed
2146  *
2147  * find_get_pages_range() will search for and return a group of up to @nr_pages
2148  * pages in the mapping starting at index @start and up to index @end
2149  * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
2150  * a reference against the returned pages.
2151  *
2152  * The search returns a group of mapping-contiguous pages with ascending
2153  * indexes.  There may be holes in the indices due to not-present pages.
2154  * We also update @start to index the next page for the traversal.
2155  *
2156  * Return: the number of pages which were found. If this number is
2157  * smaller than @nr_pages, the end of specified range has been
2158  * reached.
2159  */
2160 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2161 			      pgoff_t end, unsigned int nr_pages,
2162 			      struct page **pages)
2163 {
2164 	XA_STATE(xas, &mapping->i_pages, *start);
2165 	struct page *page;
2166 	unsigned ret = 0;
2167 
2168 	if (unlikely(!nr_pages))
2169 		return 0;
2170 
2171 	rcu_read_lock();
2172 	while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2173 		/* Skip over shadow, swap and DAX entries */
2174 		if (xa_is_value(page))
2175 			continue;
2176 
2177 		pages[ret] = find_subpage(page, xas.xa_index);
2178 		if (++ret == nr_pages) {
2179 			*start = xas.xa_index + 1;
2180 			goto out;
2181 		}
2182 	}
2183 
2184 	/*
2185 	 * We come here when there is no page beyond @end. We take care to not
2186 	 * overflow the index @start as it confuses some of the callers. This
2187 	 * breaks the iteration when there is a page at index -1 but that is
2188 	 * already broken anyway.
2189 	 */
2190 	if (end == (pgoff_t)-1)
2191 		*start = (pgoff_t)-1;
2192 	else
2193 		*start = end + 1;
2194 out:
2195 	rcu_read_unlock();
2196 
2197 	return ret;
2198 }
2199 
2200 /**
2201  * find_get_pages_contig - gang contiguous pagecache lookup
2202  * @mapping:	The address_space to search
2203  * @index:	The starting page index
2204  * @nr_pages:	The maximum number of pages
2205  * @pages:	Where the resulting pages are placed
2206  *
2207  * find_get_pages_contig() works exactly like find_get_pages(), except
2208  * that the returned number of pages are guaranteed to be contiguous.
2209  *
2210  * Return: the number of pages which were found.
2211  */
2212 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2213 			       unsigned int nr_pages, struct page **pages)
2214 {
2215 	XA_STATE(xas, &mapping->i_pages, index);
2216 	struct page *page;
2217 	unsigned int ret = 0;
2218 
2219 	if (unlikely(!nr_pages))
2220 		return 0;
2221 
2222 	rcu_read_lock();
2223 	for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2224 		if (xas_retry(&xas, page))
2225 			continue;
2226 		/*
2227 		 * If the entry has been swapped out, we can stop looking.
2228 		 * No current caller is looking for DAX entries.
2229 		 */
2230 		if (xa_is_value(page))
2231 			break;
2232 
2233 		if (!page_cache_get_speculative(page))
2234 			goto retry;
2235 
2236 		/* Has the page moved or been split? */
2237 		if (unlikely(page != xas_reload(&xas)))
2238 			goto put_page;
2239 
2240 		pages[ret] = find_subpage(page, xas.xa_index);
2241 		if (++ret == nr_pages)
2242 			break;
2243 		continue;
2244 put_page:
2245 		put_page(page);
2246 retry:
2247 		xas_reset(&xas);
2248 	}
2249 	rcu_read_unlock();
2250 	return ret;
2251 }
2252 EXPORT_SYMBOL(find_get_pages_contig);
2253 
2254 /**
2255  * find_get_pages_range_tag - Find and return head pages matching @tag.
2256  * @mapping:	the address_space to search
2257  * @index:	the starting page index
2258  * @end:	The final page index (inclusive)
2259  * @tag:	the tag index
2260  * @nr_pages:	the maximum number of pages
2261  * @pages:	where the resulting pages are placed
2262  *
2263  * Like find_get_pages(), except we only return head pages which are tagged
2264  * with @tag.  @index is updated to the index immediately after the last
2265  * page we return, ready for the next iteration.
2266  *
2267  * Return: the number of pages which were found.
2268  */
2269 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2270 			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2271 			struct page **pages)
2272 {
2273 	XA_STATE(xas, &mapping->i_pages, *index);
2274 	struct page *page;
2275 	unsigned ret = 0;
2276 
2277 	if (unlikely(!nr_pages))
2278 		return 0;
2279 
2280 	rcu_read_lock();
2281 	while ((page = find_get_entry(&xas, end, tag))) {
2282 		/*
2283 		 * Shadow entries should never be tagged, but this iteration
2284 		 * is lockless so there is a window for page reclaim to evict
2285 		 * a page we saw tagged.  Skip over it.
2286 		 */
2287 		if (xa_is_value(page))
2288 			continue;
2289 
2290 		pages[ret] = page;
2291 		if (++ret == nr_pages) {
2292 			*index = page->index + thp_nr_pages(page);
2293 			goto out;
2294 		}
2295 	}
2296 
2297 	/*
2298 	 * We come here when we got to @end. We take care to not overflow the
2299 	 * index @index as it confuses some of the callers. This breaks the
2300 	 * iteration when there is a page at index -1 but that is already
2301 	 * broken anyway.
2302 	 */
2303 	if (end == (pgoff_t)-1)
2304 		*index = (pgoff_t)-1;
2305 	else
2306 		*index = end + 1;
2307 out:
2308 	rcu_read_unlock();
2309 
2310 	return ret;
2311 }
2312 EXPORT_SYMBOL(find_get_pages_range_tag);
2313 
2314 /*
2315  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2316  * a _large_ part of the i/o request. Imagine the worst scenario:
2317  *
2318  *      ---R__________________________________________B__________
2319  *         ^ reading here                             ^ bad block(assume 4k)
2320  *
2321  * read(R) => miss => readahead(R...B) => media error => frustrating retries
2322  * => failing the whole request => read(R) => read(R+1) =>
2323  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2324  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2325  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2326  *
2327  * It is going insane. Fix it by quickly scaling down the readahead size.
2328  */
2329 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2330 {
2331 	ra->ra_pages /= 4;
2332 }
2333 
2334 /*
2335  * filemap_get_read_batch - Get a batch of pages for read
2336  *
2337  * Get a batch of pages which represent a contiguous range of bytes
2338  * in the file.  No tail pages will be returned.  If @index is in the
2339  * middle of a THP, the entire THP will be returned.  The last page in
2340  * the batch may have Readahead set or be not Uptodate so that the
2341  * caller can take the appropriate action.
2342  */
2343 static void filemap_get_read_batch(struct address_space *mapping,
2344 		pgoff_t index, pgoff_t max, struct pagevec *pvec)
2345 {
2346 	XA_STATE(xas, &mapping->i_pages, index);
2347 	struct page *head;
2348 
2349 	rcu_read_lock();
2350 	for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2351 		if (xas_retry(&xas, head))
2352 			continue;
2353 		if (xas.xa_index > max || xa_is_value(head))
2354 			break;
2355 		if (!page_cache_get_speculative(head))
2356 			goto retry;
2357 
2358 		/* Has the page moved or been split? */
2359 		if (unlikely(head != xas_reload(&xas)))
2360 			goto put_page;
2361 
2362 		if (!pagevec_add(pvec, head))
2363 			break;
2364 		if (!PageUptodate(head))
2365 			break;
2366 		if (PageReadahead(head))
2367 			break;
2368 		xas.xa_index = head->index + thp_nr_pages(head) - 1;
2369 		xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2370 		continue;
2371 put_page:
2372 		put_page(head);
2373 retry:
2374 		xas_reset(&xas);
2375 	}
2376 	rcu_read_unlock();
2377 }
2378 
2379 static int filemap_read_page(struct file *file, struct address_space *mapping,
2380 		struct page *page)
2381 {
2382 	int error;
2383 
2384 	/*
2385 	 * A previous I/O error may have been due to temporary failures,
2386 	 * eg. multipath errors.  PG_error will be set again if readpage
2387 	 * fails.
2388 	 */
2389 	ClearPageError(page);
2390 	/* Start the actual read. The read will unlock the page. */
2391 	error = mapping->a_ops->readpage(file, page);
2392 	if (error)
2393 		return error;
2394 
2395 	error = wait_on_page_locked_killable(page);
2396 	if (error)
2397 		return error;
2398 	if (PageUptodate(page))
2399 		return 0;
2400 	shrink_readahead_size_eio(&file->f_ra);
2401 	return -EIO;
2402 }
2403 
2404 static bool filemap_range_uptodate(struct address_space *mapping,
2405 		loff_t pos, struct iov_iter *iter, struct page *page)
2406 {
2407 	int count;
2408 
2409 	if (PageUptodate(page))
2410 		return true;
2411 	/* pipes can't handle partially uptodate pages */
2412 	if (iov_iter_is_pipe(iter))
2413 		return false;
2414 	if (!mapping->a_ops->is_partially_uptodate)
2415 		return false;
2416 	if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2417 		return false;
2418 
2419 	count = iter->count;
2420 	if (page_offset(page) > pos) {
2421 		count -= page_offset(page) - pos;
2422 		pos = 0;
2423 	} else {
2424 		pos -= page_offset(page);
2425 	}
2426 
2427 	return mapping->a_ops->is_partially_uptodate(page, pos, count);
2428 }
2429 
2430 static int filemap_update_page(struct kiocb *iocb,
2431 		struct address_space *mapping, struct iov_iter *iter,
2432 		struct page *page)
2433 {
2434 	struct folio *folio = page_folio(page);
2435 	int error;
2436 
2437 	if (iocb->ki_flags & IOCB_NOWAIT) {
2438 		if (!filemap_invalidate_trylock_shared(mapping))
2439 			return -EAGAIN;
2440 	} else {
2441 		filemap_invalidate_lock_shared(mapping);
2442 	}
2443 
2444 	if (!folio_trylock(folio)) {
2445 		error = -EAGAIN;
2446 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2447 			goto unlock_mapping;
2448 		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2449 			filemap_invalidate_unlock_shared(mapping);
2450 			put_and_wait_on_page_locked(&folio->page, TASK_KILLABLE);
2451 			return AOP_TRUNCATED_PAGE;
2452 		}
2453 		error = __folio_lock_async(folio, iocb->ki_waitq);
2454 		if (error)
2455 			goto unlock_mapping;
2456 	}
2457 
2458 	error = AOP_TRUNCATED_PAGE;
2459 	if (!folio->mapping)
2460 		goto unlock;
2461 
2462 	error = 0;
2463 	if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, &folio->page))
2464 		goto unlock;
2465 
2466 	error = -EAGAIN;
2467 	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2468 		goto unlock;
2469 
2470 	error = filemap_read_page(iocb->ki_filp, mapping, &folio->page);
2471 	goto unlock_mapping;
2472 unlock:
2473 	folio_unlock(folio);
2474 unlock_mapping:
2475 	filemap_invalidate_unlock_shared(mapping);
2476 	if (error == AOP_TRUNCATED_PAGE)
2477 		folio_put(folio);
2478 	return error;
2479 }
2480 
2481 static int filemap_create_page(struct file *file,
2482 		struct address_space *mapping, pgoff_t index,
2483 		struct pagevec *pvec)
2484 {
2485 	struct page *page;
2486 	int error;
2487 
2488 	page = page_cache_alloc(mapping);
2489 	if (!page)
2490 		return -ENOMEM;
2491 
2492 	/*
2493 	 * Protect against truncate / hole punch. Grabbing invalidate_lock here
2494 	 * assures we cannot instantiate and bring uptodate new pagecache pages
2495 	 * after evicting page cache during truncate and before actually
2496 	 * freeing blocks.  Note that we could release invalidate_lock after
2497 	 * inserting the page into page cache as the locked page would then be
2498 	 * enough to synchronize with hole punching. But there are code paths
2499 	 * such as filemap_update_page() filling in partially uptodate pages or
2500 	 * ->readpages() that need to hold invalidate_lock while mapping blocks
2501 	 * for IO so let's hold the lock here as well to keep locking rules
2502 	 * simple.
2503 	 */
2504 	filemap_invalidate_lock_shared(mapping);
2505 	error = add_to_page_cache_lru(page, mapping, index,
2506 			mapping_gfp_constraint(mapping, GFP_KERNEL));
2507 	if (error == -EEXIST)
2508 		error = AOP_TRUNCATED_PAGE;
2509 	if (error)
2510 		goto error;
2511 
2512 	error = filemap_read_page(file, mapping, page);
2513 	if (error)
2514 		goto error;
2515 
2516 	filemap_invalidate_unlock_shared(mapping);
2517 	pagevec_add(pvec, page);
2518 	return 0;
2519 error:
2520 	filemap_invalidate_unlock_shared(mapping);
2521 	put_page(page);
2522 	return error;
2523 }
2524 
2525 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2526 		struct address_space *mapping, struct page *page,
2527 		pgoff_t last_index)
2528 {
2529 	if (iocb->ki_flags & IOCB_NOIO)
2530 		return -EAGAIN;
2531 	page_cache_async_readahead(mapping, &file->f_ra, file, page,
2532 			page->index, last_index - page->index);
2533 	return 0;
2534 }
2535 
2536 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2537 		struct pagevec *pvec)
2538 {
2539 	struct file *filp = iocb->ki_filp;
2540 	struct address_space *mapping = filp->f_mapping;
2541 	struct file_ra_state *ra = &filp->f_ra;
2542 	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2543 	pgoff_t last_index;
2544 	struct page *page;
2545 	int err = 0;
2546 
2547 	last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2548 retry:
2549 	if (fatal_signal_pending(current))
2550 		return -EINTR;
2551 
2552 	filemap_get_read_batch(mapping, index, last_index, pvec);
2553 	if (!pagevec_count(pvec)) {
2554 		if (iocb->ki_flags & IOCB_NOIO)
2555 			return -EAGAIN;
2556 		page_cache_sync_readahead(mapping, ra, filp, index,
2557 				last_index - index);
2558 		filemap_get_read_batch(mapping, index, last_index, pvec);
2559 	}
2560 	if (!pagevec_count(pvec)) {
2561 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2562 			return -EAGAIN;
2563 		err = filemap_create_page(filp, mapping,
2564 				iocb->ki_pos >> PAGE_SHIFT, pvec);
2565 		if (err == AOP_TRUNCATED_PAGE)
2566 			goto retry;
2567 		return err;
2568 	}
2569 
2570 	page = pvec->pages[pagevec_count(pvec) - 1];
2571 	if (PageReadahead(page)) {
2572 		err = filemap_readahead(iocb, filp, mapping, page, last_index);
2573 		if (err)
2574 			goto err;
2575 	}
2576 	if (!PageUptodate(page)) {
2577 		if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2578 			iocb->ki_flags |= IOCB_NOWAIT;
2579 		err = filemap_update_page(iocb, mapping, iter, page);
2580 		if (err)
2581 			goto err;
2582 	}
2583 
2584 	return 0;
2585 err:
2586 	if (err < 0)
2587 		put_page(page);
2588 	if (likely(--pvec->nr))
2589 		return 0;
2590 	if (err == AOP_TRUNCATED_PAGE)
2591 		goto retry;
2592 	return err;
2593 }
2594 
2595 /**
2596  * filemap_read - Read data from the page cache.
2597  * @iocb: The iocb to read.
2598  * @iter: Destination for the data.
2599  * @already_read: Number of bytes already read by the caller.
2600  *
2601  * Copies data from the page cache.  If the data is not currently present,
2602  * uses the readahead and readpage address_space operations to fetch it.
2603  *
2604  * Return: Total number of bytes copied, including those already read by
2605  * the caller.  If an error happens before any bytes are copied, returns
2606  * a negative error number.
2607  */
2608 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2609 		ssize_t already_read)
2610 {
2611 	struct file *filp = iocb->ki_filp;
2612 	struct file_ra_state *ra = &filp->f_ra;
2613 	struct address_space *mapping = filp->f_mapping;
2614 	struct inode *inode = mapping->host;
2615 	struct pagevec pvec;
2616 	int i, error = 0;
2617 	bool writably_mapped;
2618 	loff_t isize, end_offset;
2619 
2620 	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2621 		return 0;
2622 	if (unlikely(!iov_iter_count(iter)))
2623 		return 0;
2624 
2625 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2626 	pagevec_init(&pvec);
2627 
2628 	do {
2629 		cond_resched();
2630 
2631 		/*
2632 		 * If we've already successfully copied some data, then we
2633 		 * can no longer safely return -EIOCBQUEUED. Hence mark
2634 		 * an async read NOWAIT at that point.
2635 		 */
2636 		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2637 			iocb->ki_flags |= IOCB_NOWAIT;
2638 
2639 		if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2640 			break;
2641 
2642 		error = filemap_get_pages(iocb, iter, &pvec);
2643 		if (error < 0)
2644 			break;
2645 
2646 		/*
2647 		 * i_size must be checked after we know the pages are Uptodate.
2648 		 *
2649 		 * Checking i_size after the check allows us to calculate
2650 		 * the correct value for "nr", which means the zero-filled
2651 		 * part of the page is not copied back to userspace (unless
2652 		 * another truncate extends the file - this is desired though).
2653 		 */
2654 		isize = i_size_read(inode);
2655 		if (unlikely(iocb->ki_pos >= isize))
2656 			goto put_pages;
2657 		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2658 
2659 		/*
2660 		 * Once we start copying data, we don't want to be touching any
2661 		 * cachelines that might be contended:
2662 		 */
2663 		writably_mapped = mapping_writably_mapped(mapping);
2664 
2665 		/*
2666 		 * When a sequential read accesses a page several times, only
2667 		 * mark it as accessed the first time.
2668 		 */
2669 		if (iocb->ki_pos >> PAGE_SHIFT !=
2670 		    ra->prev_pos >> PAGE_SHIFT)
2671 			mark_page_accessed(pvec.pages[0]);
2672 
2673 		for (i = 0; i < pagevec_count(&pvec); i++) {
2674 			struct page *page = pvec.pages[i];
2675 			size_t page_size = thp_size(page);
2676 			size_t offset = iocb->ki_pos & (page_size - 1);
2677 			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2678 					     page_size - offset);
2679 			size_t copied;
2680 
2681 			if (end_offset < page_offset(page))
2682 				break;
2683 			if (i > 0)
2684 				mark_page_accessed(page);
2685 			/*
2686 			 * If users can be writing to this page using arbitrary
2687 			 * virtual addresses, take care about potential aliasing
2688 			 * before reading the page on the kernel side.
2689 			 */
2690 			if (writably_mapped) {
2691 				int j;
2692 
2693 				for (j = 0; j < thp_nr_pages(page); j++)
2694 					flush_dcache_page(page + j);
2695 			}
2696 
2697 			copied = copy_page_to_iter(page, offset, bytes, iter);
2698 
2699 			already_read += copied;
2700 			iocb->ki_pos += copied;
2701 			ra->prev_pos = iocb->ki_pos;
2702 
2703 			if (copied < bytes) {
2704 				error = -EFAULT;
2705 				break;
2706 			}
2707 		}
2708 put_pages:
2709 		for (i = 0; i < pagevec_count(&pvec); i++)
2710 			put_page(pvec.pages[i]);
2711 		pagevec_reinit(&pvec);
2712 	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2713 
2714 	file_accessed(filp);
2715 
2716 	return already_read ? already_read : error;
2717 }
2718 EXPORT_SYMBOL_GPL(filemap_read);
2719 
2720 /**
2721  * generic_file_read_iter - generic filesystem read routine
2722  * @iocb:	kernel I/O control block
2723  * @iter:	destination for the data read
2724  *
2725  * This is the "read_iter()" routine for all filesystems
2726  * that can use the page cache directly.
2727  *
2728  * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2729  * be returned when no data can be read without waiting for I/O requests
2730  * to complete; it doesn't prevent readahead.
2731  *
2732  * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2733  * requests shall be made for the read or for readahead.  When no data
2734  * can be read, -EAGAIN shall be returned.  When readahead would be
2735  * triggered, a partial, possibly empty read shall be returned.
2736  *
2737  * Return:
2738  * * number of bytes copied, even for partial reads
2739  * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2740  */
2741 ssize_t
2742 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2743 {
2744 	size_t count = iov_iter_count(iter);
2745 	ssize_t retval = 0;
2746 
2747 	if (!count)
2748 		return 0; /* skip atime */
2749 
2750 	if (iocb->ki_flags & IOCB_DIRECT) {
2751 		struct file *file = iocb->ki_filp;
2752 		struct address_space *mapping = file->f_mapping;
2753 		struct inode *inode = mapping->host;
2754 
2755 		if (iocb->ki_flags & IOCB_NOWAIT) {
2756 			if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2757 						iocb->ki_pos + count - 1))
2758 				return -EAGAIN;
2759 		} else {
2760 			retval = filemap_write_and_wait_range(mapping,
2761 						iocb->ki_pos,
2762 					        iocb->ki_pos + count - 1);
2763 			if (retval < 0)
2764 				return retval;
2765 		}
2766 
2767 		file_accessed(file);
2768 
2769 		retval = mapping->a_ops->direct_IO(iocb, iter);
2770 		if (retval >= 0) {
2771 			iocb->ki_pos += retval;
2772 			count -= retval;
2773 		}
2774 		if (retval != -EIOCBQUEUED)
2775 			iov_iter_revert(iter, count - iov_iter_count(iter));
2776 
2777 		/*
2778 		 * Btrfs can have a short DIO read if we encounter
2779 		 * compressed extents, so if there was an error, or if
2780 		 * we've already read everything we wanted to, or if
2781 		 * there was a short read because we hit EOF, go ahead
2782 		 * and return.  Otherwise fallthrough to buffered io for
2783 		 * the rest of the read.  Buffered reads will not work for
2784 		 * DAX files, so don't bother trying.
2785 		 */
2786 		if (retval < 0 || !count || IS_DAX(inode))
2787 			return retval;
2788 		if (iocb->ki_pos >= i_size_read(inode))
2789 			return retval;
2790 	}
2791 
2792 	return filemap_read(iocb, iter, retval);
2793 }
2794 EXPORT_SYMBOL(generic_file_read_iter);
2795 
2796 static inline loff_t page_seek_hole_data(struct xa_state *xas,
2797 		struct address_space *mapping, struct page *page,
2798 		loff_t start, loff_t end, bool seek_data)
2799 {
2800 	const struct address_space_operations *ops = mapping->a_ops;
2801 	size_t offset, bsz = i_blocksize(mapping->host);
2802 
2803 	if (xa_is_value(page) || PageUptodate(page))
2804 		return seek_data ? start : end;
2805 	if (!ops->is_partially_uptodate)
2806 		return seek_data ? end : start;
2807 
2808 	xas_pause(xas);
2809 	rcu_read_unlock();
2810 	lock_page(page);
2811 	if (unlikely(page->mapping != mapping))
2812 		goto unlock;
2813 
2814 	offset = offset_in_thp(page, start) & ~(bsz - 1);
2815 
2816 	do {
2817 		if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
2818 			break;
2819 		start = (start + bsz) & ~(bsz - 1);
2820 		offset += bsz;
2821 	} while (offset < thp_size(page));
2822 unlock:
2823 	unlock_page(page);
2824 	rcu_read_lock();
2825 	return start;
2826 }
2827 
2828 static inline
2829 unsigned int seek_page_size(struct xa_state *xas, struct page *page)
2830 {
2831 	if (xa_is_value(page))
2832 		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2833 	return thp_size(page);
2834 }
2835 
2836 /**
2837  * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2838  * @mapping: Address space to search.
2839  * @start: First byte to consider.
2840  * @end: Limit of search (exclusive).
2841  * @whence: Either SEEK_HOLE or SEEK_DATA.
2842  *
2843  * If the page cache knows which blocks contain holes and which blocks
2844  * contain data, your filesystem can use this function to implement
2845  * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
2846  * entirely memory-based such as tmpfs, and filesystems which support
2847  * unwritten extents.
2848  *
2849  * Return: The requested offset on success, or -ENXIO if @whence specifies
2850  * SEEK_DATA and there is no data after @start.  There is an implicit hole
2851  * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2852  * and @end contain data.
2853  */
2854 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2855 		loff_t end, int whence)
2856 {
2857 	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2858 	pgoff_t max = (end - 1) >> PAGE_SHIFT;
2859 	bool seek_data = (whence == SEEK_DATA);
2860 	struct page *page;
2861 
2862 	if (end <= start)
2863 		return -ENXIO;
2864 
2865 	rcu_read_lock();
2866 	while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
2867 		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2868 		unsigned int seek_size;
2869 
2870 		if (start < pos) {
2871 			if (!seek_data)
2872 				goto unlock;
2873 			start = pos;
2874 		}
2875 
2876 		seek_size = seek_page_size(&xas, page);
2877 		pos = round_up(pos + 1, seek_size);
2878 		start = page_seek_hole_data(&xas, mapping, page, start, pos,
2879 				seek_data);
2880 		if (start < pos)
2881 			goto unlock;
2882 		if (start >= end)
2883 			break;
2884 		if (seek_size > PAGE_SIZE)
2885 			xas_set(&xas, pos >> PAGE_SHIFT);
2886 		if (!xa_is_value(page))
2887 			put_page(page);
2888 	}
2889 	if (seek_data)
2890 		start = -ENXIO;
2891 unlock:
2892 	rcu_read_unlock();
2893 	if (page && !xa_is_value(page))
2894 		put_page(page);
2895 	if (start > end)
2896 		return end;
2897 	return start;
2898 }
2899 
2900 #ifdef CONFIG_MMU
2901 #define MMAP_LOTSAMISS  (100)
2902 /*
2903  * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2904  * @vmf - the vm_fault for this fault.
2905  * @page - the page to lock.
2906  * @fpin - the pointer to the file we may pin (or is already pinned).
2907  *
2908  * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2909  * It differs in that it actually returns the page locked if it returns 1 and 0
2910  * if it couldn't lock the page.  If we did have to drop the mmap_lock then fpin
2911  * will point to the pinned file and needs to be fput()'ed at a later point.
2912  */
2913 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2914 				     struct file **fpin)
2915 {
2916 	struct folio *folio = page_folio(page);
2917 
2918 	if (folio_trylock(folio))
2919 		return 1;
2920 
2921 	/*
2922 	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2923 	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2924 	 * is supposed to work. We have way too many special cases..
2925 	 */
2926 	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2927 		return 0;
2928 
2929 	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2930 	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2931 		if (__folio_lock_killable(folio)) {
2932 			/*
2933 			 * We didn't have the right flags to drop the mmap_lock,
2934 			 * but all fault_handlers only check for fatal signals
2935 			 * if we return VM_FAULT_RETRY, so we need to drop the
2936 			 * mmap_lock here and return 0 if we don't have a fpin.
2937 			 */
2938 			if (*fpin == NULL)
2939 				mmap_read_unlock(vmf->vma->vm_mm);
2940 			return 0;
2941 		}
2942 	} else
2943 		__folio_lock(folio);
2944 
2945 	return 1;
2946 }
2947 
2948 /*
2949  * Synchronous readahead happens when we don't even find a page in the page
2950  * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2951  * to drop the mmap sem we return the file that was pinned in order for us to do
2952  * that.  If we didn't pin a file then we return NULL.  The file that is
2953  * returned needs to be fput()'ed when we're done with it.
2954  */
2955 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2956 {
2957 	struct file *file = vmf->vma->vm_file;
2958 	struct file_ra_state *ra = &file->f_ra;
2959 	struct address_space *mapping = file->f_mapping;
2960 	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2961 	struct file *fpin = NULL;
2962 	unsigned int mmap_miss;
2963 
2964 	/* If we don't want any read-ahead, don't bother */
2965 	if (vmf->vma->vm_flags & VM_RAND_READ)
2966 		return fpin;
2967 	if (!ra->ra_pages)
2968 		return fpin;
2969 
2970 	if (vmf->vma->vm_flags & VM_SEQ_READ) {
2971 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2972 		page_cache_sync_ra(&ractl, ra->ra_pages);
2973 		return fpin;
2974 	}
2975 
2976 	/* Avoid banging the cache line if not needed */
2977 	mmap_miss = READ_ONCE(ra->mmap_miss);
2978 	if (mmap_miss < MMAP_LOTSAMISS * 10)
2979 		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2980 
2981 	/*
2982 	 * Do we miss much more than hit in this file? If so,
2983 	 * stop bothering with read-ahead. It will only hurt.
2984 	 */
2985 	if (mmap_miss > MMAP_LOTSAMISS)
2986 		return fpin;
2987 
2988 	/*
2989 	 * mmap read-around
2990 	 */
2991 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2992 	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2993 	ra->size = ra->ra_pages;
2994 	ra->async_size = ra->ra_pages / 4;
2995 	ractl._index = ra->start;
2996 	do_page_cache_ra(&ractl, ra->size, ra->async_size);
2997 	return fpin;
2998 }
2999 
3000 /*
3001  * Asynchronous readahead happens when we find the page and PG_readahead,
3002  * so we want to possibly extend the readahead further.  We return the file that
3003  * was pinned if we have to drop the mmap_lock in order to do IO.
3004  */
3005 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3006 					    struct page *page)
3007 {
3008 	struct file *file = vmf->vma->vm_file;
3009 	struct file_ra_state *ra = &file->f_ra;
3010 	struct address_space *mapping = file->f_mapping;
3011 	struct file *fpin = NULL;
3012 	unsigned int mmap_miss;
3013 	pgoff_t offset = vmf->pgoff;
3014 
3015 	/* If we don't want any read-ahead, don't bother */
3016 	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3017 		return fpin;
3018 	mmap_miss = READ_ONCE(ra->mmap_miss);
3019 	if (mmap_miss)
3020 		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3021 	if (PageReadahead(page)) {
3022 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3023 		page_cache_async_readahead(mapping, ra, file,
3024 					   page, offset, ra->ra_pages);
3025 	}
3026 	return fpin;
3027 }
3028 
3029 /**
3030  * filemap_fault - read in file data for page fault handling
3031  * @vmf:	struct vm_fault containing details of the fault
3032  *
3033  * filemap_fault() is invoked via the vma operations vector for a
3034  * mapped memory region to read in file data during a page fault.
3035  *
3036  * The goto's are kind of ugly, but this streamlines the normal case of having
3037  * it in the page cache, and handles the special cases reasonably without
3038  * having a lot of duplicated code.
3039  *
3040  * vma->vm_mm->mmap_lock must be held on entry.
3041  *
3042  * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3043  * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
3044  *
3045  * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3046  * has not been released.
3047  *
3048  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3049  *
3050  * Return: bitwise-OR of %VM_FAULT_ codes.
3051  */
3052 vm_fault_t filemap_fault(struct vm_fault *vmf)
3053 {
3054 	int error;
3055 	struct file *file = vmf->vma->vm_file;
3056 	struct file *fpin = NULL;
3057 	struct address_space *mapping = file->f_mapping;
3058 	struct inode *inode = mapping->host;
3059 	pgoff_t offset = vmf->pgoff;
3060 	pgoff_t max_off;
3061 	struct page *page;
3062 	vm_fault_t ret = 0;
3063 	bool mapping_locked = false;
3064 
3065 	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3066 	if (unlikely(offset >= max_off))
3067 		return VM_FAULT_SIGBUS;
3068 
3069 	/*
3070 	 * Do we have something in the page cache already?
3071 	 */
3072 	page = find_get_page(mapping, offset);
3073 	if (likely(page)) {
3074 		/*
3075 		 * We found the page, so try async readahead before waiting for
3076 		 * the lock.
3077 		 */
3078 		if (!(vmf->flags & FAULT_FLAG_TRIED))
3079 			fpin = do_async_mmap_readahead(vmf, page);
3080 		if (unlikely(!PageUptodate(page))) {
3081 			filemap_invalidate_lock_shared(mapping);
3082 			mapping_locked = true;
3083 		}
3084 	} else {
3085 		/* No page in the page cache at all */
3086 		count_vm_event(PGMAJFAULT);
3087 		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3088 		ret = VM_FAULT_MAJOR;
3089 		fpin = do_sync_mmap_readahead(vmf);
3090 retry_find:
3091 		/*
3092 		 * See comment in filemap_create_page() why we need
3093 		 * invalidate_lock
3094 		 */
3095 		if (!mapping_locked) {
3096 			filemap_invalidate_lock_shared(mapping);
3097 			mapping_locked = true;
3098 		}
3099 		page = pagecache_get_page(mapping, offset,
3100 					  FGP_CREAT|FGP_FOR_MMAP,
3101 					  vmf->gfp_mask);
3102 		if (!page) {
3103 			if (fpin)
3104 				goto out_retry;
3105 			filemap_invalidate_unlock_shared(mapping);
3106 			return VM_FAULT_OOM;
3107 		}
3108 	}
3109 
3110 	if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3111 		goto out_retry;
3112 
3113 	/* Did it get truncated? */
3114 	if (unlikely(compound_head(page)->mapping != mapping)) {
3115 		unlock_page(page);
3116 		put_page(page);
3117 		goto retry_find;
3118 	}
3119 	VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3120 
3121 	/*
3122 	 * We have a locked page in the page cache, now we need to check
3123 	 * that it's up-to-date. If not, it is going to be due to an error.
3124 	 */
3125 	if (unlikely(!PageUptodate(page))) {
3126 		/*
3127 		 * The page was in cache and uptodate and now it is not.
3128 		 * Strange but possible since we didn't hold the page lock all
3129 		 * the time. Let's drop everything get the invalidate lock and
3130 		 * try again.
3131 		 */
3132 		if (!mapping_locked) {
3133 			unlock_page(page);
3134 			put_page(page);
3135 			goto retry_find;
3136 		}
3137 		goto page_not_uptodate;
3138 	}
3139 
3140 	/*
3141 	 * We've made it this far and we had to drop our mmap_lock, now is the
3142 	 * time to return to the upper layer and have it re-find the vma and
3143 	 * redo the fault.
3144 	 */
3145 	if (fpin) {
3146 		unlock_page(page);
3147 		goto out_retry;
3148 	}
3149 	if (mapping_locked)
3150 		filemap_invalidate_unlock_shared(mapping);
3151 
3152 	/*
3153 	 * Found the page and have a reference on it.
3154 	 * We must recheck i_size under page lock.
3155 	 */
3156 	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3157 	if (unlikely(offset >= max_off)) {
3158 		unlock_page(page);
3159 		put_page(page);
3160 		return VM_FAULT_SIGBUS;
3161 	}
3162 
3163 	vmf->page = page;
3164 	return ret | VM_FAULT_LOCKED;
3165 
3166 page_not_uptodate:
3167 	/*
3168 	 * Umm, take care of errors if the page isn't up-to-date.
3169 	 * Try to re-read it _once_. We do this synchronously,
3170 	 * because there really aren't any performance issues here
3171 	 * and we need to check for errors.
3172 	 */
3173 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3174 	error = filemap_read_page(file, mapping, page);
3175 	if (fpin)
3176 		goto out_retry;
3177 	put_page(page);
3178 
3179 	if (!error || error == AOP_TRUNCATED_PAGE)
3180 		goto retry_find;
3181 	filemap_invalidate_unlock_shared(mapping);
3182 
3183 	return VM_FAULT_SIGBUS;
3184 
3185 out_retry:
3186 	/*
3187 	 * We dropped the mmap_lock, we need to return to the fault handler to
3188 	 * re-find the vma and come back and find our hopefully still populated
3189 	 * page.
3190 	 */
3191 	if (page)
3192 		put_page(page);
3193 	if (mapping_locked)
3194 		filemap_invalidate_unlock_shared(mapping);
3195 	if (fpin)
3196 		fput(fpin);
3197 	return ret | VM_FAULT_RETRY;
3198 }
3199 EXPORT_SYMBOL(filemap_fault);
3200 
3201 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3202 {
3203 	struct mm_struct *mm = vmf->vma->vm_mm;
3204 
3205 	/* Huge page is mapped? No need to proceed. */
3206 	if (pmd_trans_huge(*vmf->pmd)) {
3207 		unlock_page(page);
3208 		put_page(page);
3209 		return true;
3210 	}
3211 
3212 	if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3213 		vm_fault_t ret = do_set_pmd(vmf, page);
3214 		if (!ret) {
3215 			/* The page is mapped successfully, reference consumed. */
3216 			unlock_page(page);
3217 			return true;
3218 		}
3219 	}
3220 
3221 	if (pmd_none(*vmf->pmd))
3222 		pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3223 
3224 	/* See comment in handle_pte_fault() */
3225 	if (pmd_devmap_trans_unstable(vmf->pmd)) {
3226 		unlock_page(page);
3227 		put_page(page);
3228 		return true;
3229 	}
3230 
3231 	return false;
3232 }
3233 
3234 static struct page *next_uptodate_page(struct page *page,
3235 				       struct address_space *mapping,
3236 				       struct xa_state *xas, pgoff_t end_pgoff)
3237 {
3238 	unsigned long max_idx;
3239 
3240 	do {
3241 		if (!page)
3242 			return NULL;
3243 		if (xas_retry(xas, page))
3244 			continue;
3245 		if (xa_is_value(page))
3246 			continue;
3247 		if (PageLocked(page))
3248 			continue;
3249 		if (!page_cache_get_speculative(page))
3250 			continue;
3251 		/* Has the page moved or been split? */
3252 		if (unlikely(page != xas_reload(xas)))
3253 			goto skip;
3254 		if (!PageUptodate(page) || PageReadahead(page))
3255 			goto skip;
3256 		if (PageHWPoison(page))
3257 			goto skip;
3258 		if (!trylock_page(page))
3259 			goto skip;
3260 		if (page->mapping != mapping)
3261 			goto unlock;
3262 		if (!PageUptodate(page))
3263 			goto unlock;
3264 		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3265 		if (xas->xa_index >= max_idx)
3266 			goto unlock;
3267 		return page;
3268 unlock:
3269 		unlock_page(page);
3270 skip:
3271 		put_page(page);
3272 	} while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3273 
3274 	return NULL;
3275 }
3276 
3277 static inline struct page *first_map_page(struct address_space *mapping,
3278 					  struct xa_state *xas,
3279 					  pgoff_t end_pgoff)
3280 {
3281 	return next_uptodate_page(xas_find(xas, end_pgoff),
3282 				  mapping, xas, end_pgoff);
3283 }
3284 
3285 static inline struct page *next_map_page(struct address_space *mapping,
3286 					 struct xa_state *xas,
3287 					 pgoff_t end_pgoff)
3288 {
3289 	return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3290 				  mapping, xas, end_pgoff);
3291 }
3292 
3293 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3294 			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3295 {
3296 	struct vm_area_struct *vma = vmf->vma;
3297 	struct file *file = vma->vm_file;
3298 	struct address_space *mapping = file->f_mapping;
3299 	pgoff_t last_pgoff = start_pgoff;
3300 	unsigned long addr;
3301 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3302 	struct page *head, *page;
3303 	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3304 	vm_fault_t ret = 0;
3305 
3306 	rcu_read_lock();
3307 	head = first_map_page(mapping, &xas, end_pgoff);
3308 	if (!head)
3309 		goto out;
3310 
3311 	if (filemap_map_pmd(vmf, head)) {
3312 		ret = VM_FAULT_NOPAGE;
3313 		goto out;
3314 	}
3315 
3316 	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3317 	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3318 	do {
3319 		page = find_subpage(head, xas.xa_index);
3320 		if (PageHWPoison(page))
3321 			goto unlock;
3322 
3323 		if (mmap_miss > 0)
3324 			mmap_miss--;
3325 
3326 		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3327 		vmf->pte += xas.xa_index - last_pgoff;
3328 		last_pgoff = xas.xa_index;
3329 
3330 		if (!pte_none(*vmf->pte))
3331 			goto unlock;
3332 
3333 		/* We're about to handle the fault */
3334 		if (vmf->address == addr)
3335 			ret = VM_FAULT_NOPAGE;
3336 
3337 		do_set_pte(vmf, page, addr);
3338 		/* no need to invalidate: a not-present page won't be cached */
3339 		update_mmu_cache(vma, addr, vmf->pte);
3340 		unlock_page(head);
3341 		continue;
3342 unlock:
3343 		unlock_page(head);
3344 		put_page(head);
3345 	} while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3346 	pte_unmap_unlock(vmf->pte, vmf->ptl);
3347 out:
3348 	rcu_read_unlock();
3349 	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3350 	return ret;
3351 }
3352 EXPORT_SYMBOL(filemap_map_pages);
3353 
3354 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3355 {
3356 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3357 	struct page *page = vmf->page;
3358 	vm_fault_t ret = VM_FAULT_LOCKED;
3359 
3360 	sb_start_pagefault(mapping->host->i_sb);
3361 	file_update_time(vmf->vma->vm_file);
3362 	lock_page(page);
3363 	if (page->mapping != mapping) {
3364 		unlock_page(page);
3365 		ret = VM_FAULT_NOPAGE;
3366 		goto out;
3367 	}
3368 	/*
3369 	 * We mark the page dirty already here so that when freeze is in
3370 	 * progress, we are guaranteed that writeback during freezing will
3371 	 * see the dirty page and writeprotect it again.
3372 	 */
3373 	set_page_dirty(page);
3374 	wait_for_stable_page(page);
3375 out:
3376 	sb_end_pagefault(mapping->host->i_sb);
3377 	return ret;
3378 }
3379 
3380 const struct vm_operations_struct generic_file_vm_ops = {
3381 	.fault		= filemap_fault,
3382 	.map_pages	= filemap_map_pages,
3383 	.page_mkwrite	= filemap_page_mkwrite,
3384 };
3385 
3386 /* This is used for a general mmap of a disk file */
3387 
3388 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3389 {
3390 	struct address_space *mapping = file->f_mapping;
3391 
3392 	if (!mapping->a_ops->readpage)
3393 		return -ENOEXEC;
3394 	file_accessed(file);
3395 	vma->vm_ops = &generic_file_vm_ops;
3396 	return 0;
3397 }
3398 
3399 /*
3400  * This is for filesystems which do not implement ->writepage.
3401  */
3402 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3403 {
3404 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3405 		return -EINVAL;
3406 	return generic_file_mmap(file, vma);
3407 }
3408 #else
3409 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3410 {
3411 	return VM_FAULT_SIGBUS;
3412 }
3413 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3414 {
3415 	return -ENOSYS;
3416 }
3417 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3418 {
3419 	return -ENOSYS;
3420 }
3421 #endif /* CONFIG_MMU */
3422 
3423 EXPORT_SYMBOL(filemap_page_mkwrite);
3424 EXPORT_SYMBOL(generic_file_mmap);
3425 EXPORT_SYMBOL(generic_file_readonly_mmap);
3426 
3427 static struct page *wait_on_page_read(struct page *page)
3428 {
3429 	if (!IS_ERR(page)) {
3430 		wait_on_page_locked(page);
3431 		if (!PageUptodate(page)) {
3432 			put_page(page);
3433 			page = ERR_PTR(-EIO);
3434 		}
3435 	}
3436 	return page;
3437 }
3438 
3439 static struct page *do_read_cache_page(struct address_space *mapping,
3440 				pgoff_t index,
3441 				int (*filler)(void *, struct page *),
3442 				void *data,
3443 				gfp_t gfp)
3444 {
3445 	struct page *page;
3446 	int err;
3447 repeat:
3448 	page = find_get_page(mapping, index);
3449 	if (!page) {
3450 		page = __page_cache_alloc(gfp);
3451 		if (!page)
3452 			return ERR_PTR(-ENOMEM);
3453 		err = add_to_page_cache_lru(page, mapping, index, gfp);
3454 		if (unlikely(err)) {
3455 			put_page(page);
3456 			if (err == -EEXIST)
3457 				goto repeat;
3458 			/* Presumably ENOMEM for xarray node */
3459 			return ERR_PTR(err);
3460 		}
3461 
3462 filler:
3463 		if (filler)
3464 			err = filler(data, page);
3465 		else
3466 			err = mapping->a_ops->readpage(data, page);
3467 
3468 		if (err < 0) {
3469 			put_page(page);
3470 			return ERR_PTR(err);
3471 		}
3472 
3473 		page = wait_on_page_read(page);
3474 		if (IS_ERR(page))
3475 			return page;
3476 		goto out;
3477 	}
3478 	if (PageUptodate(page))
3479 		goto out;
3480 
3481 	/*
3482 	 * Page is not up to date and may be locked due to one of the following
3483 	 * case a: Page is being filled and the page lock is held
3484 	 * case b: Read/write error clearing the page uptodate status
3485 	 * case c: Truncation in progress (page locked)
3486 	 * case d: Reclaim in progress
3487 	 *
3488 	 * Case a, the page will be up to date when the page is unlocked.
3489 	 *    There is no need to serialise on the page lock here as the page
3490 	 *    is pinned so the lock gives no additional protection. Even if the
3491 	 *    page is truncated, the data is still valid if PageUptodate as
3492 	 *    it's a race vs truncate race.
3493 	 * Case b, the page will not be up to date
3494 	 * Case c, the page may be truncated but in itself, the data may still
3495 	 *    be valid after IO completes as it's a read vs truncate race. The
3496 	 *    operation must restart if the page is not uptodate on unlock but
3497 	 *    otherwise serialising on page lock to stabilise the mapping gives
3498 	 *    no additional guarantees to the caller as the page lock is
3499 	 *    released before return.
3500 	 * Case d, similar to truncation. If reclaim holds the page lock, it
3501 	 *    will be a race with remove_mapping that determines if the mapping
3502 	 *    is valid on unlock but otherwise the data is valid and there is
3503 	 *    no need to serialise with page lock.
3504 	 *
3505 	 * As the page lock gives no additional guarantee, we optimistically
3506 	 * wait on the page to be unlocked and check if it's up to date and
3507 	 * use the page if it is. Otherwise, the page lock is required to
3508 	 * distinguish between the different cases. The motivation is that we
3509 	 * avoid spurious serialisations and wakeups when multiple processes
3510 	 * wait on the same page for IO to complete.
3511 	 */
3512 	wait_on_page_locked(page);
3513 	if (PageUptodate(page))
3514 		goto out;
3515 
3516 	/* Distinguish between all the cases under the safety of the lock */
3517 	lock_page(page);
3518 
3519 	/* Case c or d, restart the operation */
3520 	if (!page->mapping) {
3521 		unlock_page(page);
3522 		put_page(page);
3523 		goto repeat;
3524 	}
3525 
3526 	/* Someone else locked and filled the page in a very small window */
3527 	if (PageUptodate(page)) {
3528 		unlock_page(page);
3529 		goto out;
3530 	}
3531 
3532 	/*
3533 	 * A previous I/O error may have been due to temporary
3534 	 * failures.
3535 	 * Clear page error before actual read, PG_error will be
3536 	 * set again if read page fails.
3537 	 */
3538 	ClearPageError(page);
3539 	goto filler;
3540 
3541 out:
3542 	mark_page_accessed(page);
3543 	return page;
3544 }
3545 
3546 /**
3547  * read_cache_page - read into page cache, fill it if needed
3548  * @mapping:	the page's address_space
3549  * @index:	the page index
3550  * @filler:	function to perform the read
3551  * @data:	first arg to filler(data, page) function, often left as NULL
3552  *
3553  * Read into the page cache. If a page already exists, and PageUptodate() is
3554  * not set, try to fill the page and wait for it to become unlocked.
3555  *
3556  * If the page does not get brought uptodate, return -EIO.
3557  *
3558  * The function expects mapping->invalidate_lock to be already held.
3559  *
3560  * Return: up to date page on success, ERR_PTR() on failure.
3561  */
3562 struct page *read_cache_page(struct address_space *mapping,
3563 				pgoff_t index,
3564 				int (*filler)(void *, struct page *),
3565 				void *data)
3566 {
3567 	return do_read_cache_page(mapping, index, filler, data,
3568 			mapping_gfp_mask(mapping));
3569 }
3570 EXPORT_SYMBOL(read_cache_page);
3571 
3572 /**
3573  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3574  * @mapping:	the page's address_space
3575  * @index:	the page index
3576  * @gfp:	the page allocator flags to use if allocating
3577  *
3578  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3579  * any new page allocations done using the specified allocation flags.
3580  *
3581  * If the page does not get brought uptodate, return -EIO.
3582  *
3583  * The function expects mapping->invalidate_lock to be already held.
3584  *
3585  * Return: up to date page on success, ERR_PTR() on failure.
3586  */
3587 struct page *read_cache_page_gfp(struct address_space *mapping,
3588 				pgoff_t index,
3589 				gfp_t gfp)
3590 {
3591 	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3592 }
3593 EXPORT_SYMBOL(read_cache_page_gfp);
3594 
3595 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3596 				loff_t pos, unsigned len, unsigned flags,
3597 				struct page **pagep, void **fsdata)
3598 {
3599 	const struct address_space_operations *aops = mapping->a_ops;
3600 
3601 	return aops->write_begin(file, mapping, pos, len, flags,
3602 							pagep, fsdata);
3603 }
3604 EXPORT_SYMBOL(pagecache_write_begin);
3605 
3606 int pagecache_write_end(struct file *file, struct address_space *mapping,
3607 				loff_t pos, unsigned len, unsigned copied,
3608 				struct page *page, void *fsdata)
3609 {
3610 	const struct address_space_operations *aops = mapping->a_ops;
3611 
3612 	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3613 }
3614 EXPORT_SYMBOL(pagecache_write_end);
3615 
3616 /*
3617  * Warn about a page cache invalidation failure during a direct I/O write.
3618  */
3619 void dio_warn_stale_pagecache(struct file *filp)
3620 {
3621 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3622 	char pathname[128];
3623 	char *path;
3624 
3625 	errseq_set(&filp->f_mapping->wb_err, -EIO);
3626 	if (__ratelimit(&_rs)) {
3627 		path = file_path(filp, pathname, sizeof(pathname));
3628 		if (IS_ERR(path))
3629 			path = "(unknown)";
3630 		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3631 		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3632 			current->comm);
3633 	}
3634 }
3635 
3636 ssize_t
3637 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3638 {
3639 	struct file	*file = iocb->ki_filp;
3640 	struct address_space *mapping = file->f_mapping;
3641 	struct inode	*inode = mapping->host;
3642 	loff_t		pos = iocb->ki_pos;
3643 	ssize_t		written;
3644 	size_t		write_len;
3645 	pgoff_t		end;
3646 
3647 	write_len = iov_iter_count(from);
3648 	end = (pos + write_len - 1) >> PAGE_SHIFT;
3649 
3650 	if (iocb->ki_flags & IOCB_NOWAIT) {
3651 		/* If there are pages to writeback, return */
3652 		if (filemap_range_has_page(file->f_mapping, pos,
3653 					   pos + write_len - 1))
3654 			return -EAGAIN;
3655 	} else {
3656 		written = filemap_write_and_wait_range(mapping, pos,
3657 							pos + write_len - 1);
3658 		if (written)
3659 			goto out;
3660 	}
3661 
3662 	/*
3663 	 * After a write we want buffered reads to be sure to go to disk to get
3664 	 * the new data.  We invalidate clean cached page from the region we're
3665 	 * about to write.  We do this *before* the write so that we can return
3666 	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3667 	 */
3668 	written = invalidate_inode_pages2_range(mapping,
3669 					pos >> PAGE_SHIFT, end);
3670 	/*
3671 	 * If a page can not be invalidated, return 0 to fall back
3672 	 * to buffered write.
3673 	 */
3674 	if (written) {
3675 		if (written == -EBUSY)
3676 			return 0;
3677 		goto out;
3678 	}
3679 
3680 	written = mapping->a_ops->direct_IO(iocb, from);
3681 
3682 	/*
3683 	 * Finally, try again to invalidate clean pages which might have been
3684 	 * cached by non-direct readahead, or faulted in by get_user_pages()
3685 	 * if the source of the write was an mmap'ed region of the file
3686 	 * we're writing.  Either one is a pretty crazy thing to do,
3687 	 * so we don't support it 100%.  If this invalidation
3688 	 * fails, tough, the write still worked...
3689 	 *
3690 	 * Most of the time we do not need this since dio_complete() will do
3691 	 * the invalidation for us. However there are some file systems that
3692 	 * do not end up with dio_complete() being called, so let's not break
3693 	 * them by removing it completely.
3694 	 *
3695 	 * Noticeable example is a blkdev_direct_IO().
3696 	 *
3697 	 * Skip invalidation for async writes or if mapping has no pages.
3698 	 */
3699 	if (written > 0 && mapping->nrpages &&
3700 	    invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3701 		dio_warn_stale_pagecache(file);
3702 
3703 	if (written > 0) {
3704 		pos += written;
3705 		write_len -= written;
3706 		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3707 			i_size_write(inode, pos);
3708 			mark_inode_dirty(inode);
3709 		}
3710 		iocb->ki_pos = pos;
3711 	}
3712 	if (written != -EIOCBQUEUED)
3713 		iov_iter_revert(from, write_len - iov_iter_count(from));
3714 out:
3715 	return written;
3716 }
3717 EXPORT_SYMBOL(generic_file_direct_write);
3718 
3719 ssize_t generic_perform_write(struct file *file,
3720 				struct iov_iter *i, loff_t pos)
3721 {
3722 	struct address_space *mapping = file->f_mapping;
3723 	const struct address_space_operations *a_ops = mapping->a_ops;
3724 	long status = 0;
3725 	ssize_t written = 0;
3726 	unsigned int flags = 0;
3727 
3728 	do {
3729 		struct page *page;
3730 		unsigned long offset;	/* Offset into pagecache page */
3731 		unsigned long bytes;	/* Bytes to write to page */
3732 		size_t copied;		/* Bytes copied from user */
3733 		void *fsdata;
3734 
3735 		offset = (pos & (PAGE_SIZE - 1));
3736 		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3737 						iov_iter_count(i));
3738 
3739 again:
3740 		/*
3741 		 * Bring in the user page that we will copy from _first_.
3742 		 * Otherwise there's a nasty deadlock on copying from the
3743 		 * same page as we're writing to, without it being marked
3744 		 * up-to-date.
3745 		 */
3746 		if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3747 			status = -EFAULT;
3748 			break;
3749 		}
3750 
3751 		if (fatal_signal_pending(current)) {
3752 			status = -EINTR;
3753 			break;
3754 		}
3755 
3756 		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3757 						&page, &fsdata);
3758 		if (unlikely(status < 0))
3759 			break;
3760 
3761 		if (mapping_writably_mapped(mapping))
3762 			flush_dcache_page(page);
3763 
3764 		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3765 		flush_dcache_page(page);
3766 
3767 		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3768 						page, fsdata);
3769 		if (unlikely(status != copied)) {
3770 			iov_iter_revert(i, copied - max(status, 0L));
3771 			if (unlikely(status < 0))
3772 				break;
3773 		}
3774 		cond_resched();
3775 
3776 		if (unlikely(status == 0)) {
3777 			/*
3778 			 * A short copy made ->write_end() reject the
3779 			 * thing entirely.  Might be memory poisoning
3780 			 * halfway through, might be a race with munmap,
3781 			 * might be severe memory pressure.
3782 			 */
3783 			if (copied)
3784 				bytes = copied;
3785 			goto again;
3786 		}
3787 		pos += status;
3788 		written += status;
3789 
3790 		balance_dirty_pages_ratelimited(mapping);
3791 	} while (iov_iter_count(i));
3792 
3793 	return written ? written : status;
3794 }
3795 EXPORT_SYMBOL(generic_perform_write);
3796 
3797 /**
3798  * __generic_file_write_iter - write data to a file
3799  * @iocb:	IO state structure (file, offset, etc.)
3800  * @from:	iov_iter with data to write
3801  *
3802  * This function does all the work needed for actually writing data to a
3803  * file. It does all basic checks, removes SUID from the file, updates
3804  * modification times and calls proper subroutines depending on whether we
3805  * do direct IO or a standard buffered write.
3806  *
3807  * It expects i_rwsem to be grabbed unless we work on a block device or similar
3808  * object which does not need locking at all.
3809  *
3810  * This function does *not* take care of syncing data in case of O_SYNC write.
3811  * A caller has to handle it. This is mainly due to the fact that we want to
3812  * avoid syncing under i_rwsem.
3813  *
3814  * Return:
3815  * * number of bytes written, even for truncated writes
3816  * * negative error code if no data has been written at all
3817  */
3818 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3819 {
3820 	struct file *file = iocb->ki_filp;
3821 	struct address_space *mapping = file->f_mapping;
3822 	struct inode 	*inode = mapping->host;
3823 	ssize_t		written = 0;
3824 	ssize_t		err;
3825 	ssize_t		status;
3826 
3827 	/* We can write back this queue in page reclaim */
3828 	current->backing_dev_info = inode_to_bdi(inode);
3829 	err = file_remove_privs(file);
3830 	if (err)
3831 		goto out;
3832 
3833 	err = file_update_time(file);
3834 	if (err)
3835 		goto out;
3836 
3837 	if (iocb->ki_flags & IOCB_DIRECT) {
3838 		loff_t pos, endbyte;
3839 
3840 		written = generic_file_direct_write(iocb, from);
3841 		/*
3842 		 * If the write stopped short of completing, fall back to
3843 		 * buffered writes.  Some filesystems do this for writes to
3844 		 * holes, for example.  For DAX files, a buffered write will
3845 		 * not succeed (even if it did, DAX does not handle dirty
3846 		 * page-cache pages correctly).
3847 		 */
3848 		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3849 			goto out;
3850 
3851 		status = generic_perform_write(file, from, pos = iocb->ki_pos);
3852 		/*
3853 		 * If generic_perform_write() returned a synchronous error
3854 		 * then we want to return the number of bytes which were
3855 		 * direct-written, or the error code if that was zero.  Note
3856 		 * that this differs from normal direct-io semantics, which
3857 		 * will return -EFOO even if some bytes were written.
3858 		 */
3859 		if (unlikely(status < 0)) {
3860 			err = status;
3861 			goto out;
3862 		}
3863 		/*
3864 		 * We need to ensure that the page cache pages are written to
3865 		 * disk and invalidated to preserve the expected O_DIRECT
3866 		 * semantics.
3867 		 */
3868 		endbyte = pos + status - 1;
3869 		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3870 		if (err == 0) {
3871 			iocb->ki_pos = endbyte + 1;
3872 			written += status;
3873 			invalidate_mapping_pages(mapping,
3874 						 pos >> PAGE_SHIFT,
3875 						 endbyte >> PAGE_SHIFT);
3876 		} else {
3877 			/*
3878 			 * We don't know how much we wrote, so just return
3879 			 * the number of bytes which were direct-written
3880 			 */
3881 		}
3882 	} else {
3883 		written = generic_perform_write(file, from, iocb->ki_pos);
3884 		if (likely(written > 0))
3885 			iocb->ki_pos += written;
3886 	}
3887 out:
3888 	current->backing_dev_info = NULL;
3889 	return written ? written : err;
3890 }
3891 EXPORT_SYMBOL(__generic_file_write_iter);
3892 
3893 /**
3894  * generic_file_write_iter - write data to a file
3895  * @iocb:	IO state structure
3896  * @from:	iov_iter with data to write
3897  *
3898  * This is a wrapper around __generic_file_write_iter() to be used by most
3899  * filesystems. It takes care of syncing the file in case of O_SYNC file
3900  * and acquires i_rwsem as needed.
3901  * Return:
3902  * * negative error code if no data has been written at all of
3903  *   vfs_fsync_range() failed for a synchronous write
3904  * * number of bytes written, even for truncated writes
3905  */
3906 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3907 {
3908 	struct file *file = iocb->ki_filp;
3909 	struct inode *inode = file->f_mapping->host;
3910 	ssize_t ret;
3911 
3912 	inode_lock(inode);
3913 	ret = generic_write_checks(iocb, from);
3914 	if (ret > 0)
3915 		ret = __generic_file_write_iter(iocb, from);
3916 	inode_unlock(inode);
3917 
3918 	if (ret > 0)
3919 		ret = generic_write_sync(iocb, ret);
3920 	return ret;
3921 }
3922 EXPORT_SYMBOL(generic_file_write_iter);
3923 
3924 /**
3925  * try_to_release_page() - release old fs-specific metadata on a page
3926  *
3927  * @page: the page which the kernel is trying to free
3928  * @gfp_mask: memory allocation flags (and I/O mode)
3929  *
3930  * The address_space is to try to release any data against the page
3931  * (presumably at page->private).
3932  *
3933  * This may also be called if PG_fscache is set on a page, indicating that the
3934  * page is known to the local caching routines.
3935  *
3936  * The @gfp_mask argument specifies whether I/O may be performed to release
3937  * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3938  *
3939  * Return: %1 if the release was successful, otherwise return zero.
3940  */
3941 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3942 {
3943 	struct address_space * const mapping = page->mapping;
3944 
3945 	BUG_ON(!PageLocked(page));
3946 	if (PageWriteback(page))
3947 		return 0;
3948 
3949 	if (mapping && mapping->a_ops->releasepage)
3950 		return mapping->a_ops->releasepage(page, gfp_mask);
3951 	return try_to_free_buffers(page);
3952 }
3953 
3954 EXPORT_SYMBOL(try_to_release_page);
3955