1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	linux/mm/filemap.c
4  *
5  * Copyright (C) 1994-1999  Linus Torvalds
6  */
7 
8 /*
9  * This file handles the generic file mmap semantics used by
10  * most "normal" filesystems (but you don't /have/ to use this:
11  * the NFS filesystem used to do this differently, for example)
12  */
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
16 #include <linux/fs.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
22 #include <linux/mm.h>
23 #include <linux/swap.h>
24 #include <linux/swapops.h>
25 #include <linux/mman.h>
26 #include <linux/pagemap.h>
27 #include <linux/file.h>
28 #include <linux/uio.h>
29 #include <linux/error-injection.h>
30 #include <linux/hash.h>
31 #include <linux/writeback.h>
32 #include <linux/backing-dev.h>
33 #include <linux/pagevec.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <linux/migrate.h>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
47 #include "internal.h"
48 
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
51 
52 /*
53  * FIXME: remove all knowledge of the buffer layer from the core VM
54  */
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
56 
57 #include <asm/mman.h>
58 
59 /*
60  * Shared mappings implemented 30.11.1994. It's not fully working yet,
61  * though.
62  *
63  * Shared mappings now work. 15.8.1995  Bruno.
64  *
65  * finished 'unifying' the page and buffer cache and SMP-threaded the
66  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
67  *
68  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
69  */
70 
71 /*
72  * Lock ordering:
73  *
74  *  ->i_mmap_rwsem		(truncate_pagecache)
75  *    ->private_lock		(__free_pte->block_dirty_folio)
76  *      ->swap_lock		(exclusive_swap_page, others)
77  *        ->i_pages lock
78  *
79  *  ->i_rwsem
80  *    ->invalidate_lock		(acquired by fs in truncate path)
81  *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
82  *
83  *  ->mmap_lock
84  *    ->i_mmap_rwsem
85  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
86  *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
87  *
88  *  ->mmap_lock
89  *    ->invalidate_lock		(filemap_fault)
90  *      ->lock_page		(filemap_fault, access_process_vm)
91  *
92  *  ->i_rwsem			(generic_perform_write)
93  *    ->mmap_lock		(fault_in_readable->do_page_fault)
94  *
95  *  bdi->wb.list_lock
96  *    sb_lock			(fs/fs-writeback.c)
97  *    ->i_pages lock		(__sync_single_inode)
98  *
99  *  ->i_mmap_rwsem
100  *    ->anon_vma.lock		(vma_adjust)
101  *
102  *  ->anon_vma.lock
103  *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
104  *
105  *  ->page_table_lock or pte_lock
106  *    ->swap_lock		(try_to_unmap_one)
107  *    ->private_lock		(try_to_unmap_one)
108  *    ->i_pages lock		(try_to_unmap_one)
109  *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
110  *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
111  *    ->private_lock		(page_remove_rmap->set_page_dirty)
112  *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
113  *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
114  *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
115  *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
116  *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
117  *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
118  *    ->private_lock		(zap_pte_range->block_dirty_folio)
119  *
120  * ->i_mmap_rwsem
121  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
122  */
123 
page_cache_delete(struct address_space * mapping,struct folio * folio,void * shadow)124 static void page_cache_delete(struct address_space *mapping,
125 				   struct folio *folio, void *shadow)
126 {
127 	XA_STATE(xas, &mapping->i_pages, folio->index);
128 	long nr = 1;
129 
130 	mapping_set_update(&xas, mapping);
131 
132 	/* hugetlb pages are represented by a single entry in the xarray */
133 	if (!folio_test_hugetlb(folio)) {
134 		xas_set_order(&xas, folio->index, folio_order(folio));
135 		nr = folio_nr_pages(folio);
136 	}
137 
138 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
139 
140 	xas_store(&xas, shadow);
141 	xas_init_marks(&xas);
142 
143 	folio->mapping = NULL;
144 	/* Leave page->index set: truncation lookup relies upon it */
145 	mapping->nrpages -= nr;
146 }
147 
filemap_unaccount_folio(struct address_space * mapping,struct folio * folio)148 static void filemap_unaccount_folio(struct address_space *mapping,
149 		struct folio *folio)
150 {
151 	long nr;
152 
153 	VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
154 	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
155 		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
156 			 current->comm, folio_pfn(folio));
157 		dump_page(&folio->page, "still mapped when deleted");
158 		dump_stack();
159 		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
160 
161 		if (mapping_exiting(mapping) && !folio_test_large(folio)) {
162 			int mapcount = page_mapcount(&folio->page);
163 
164 			if (folio_ref_count(folio) >= mapcount + 2) {
165 				/*
166 				 * All vmas have already been torn down, so it's
167 				 * a good bet that actually the page is unmapped
168 				 * and we'd rather not leak it: if we're wrong,
169 				 * another bad page check should catch it later.
170 				 */
171 				page_mapcount_reset(&folio->page);
172 				folio_ref_sub(folio, mapcount);
173 			}
174 		}
175 	}
176 
177 	/* hugetlb folios do not participate in page cache accounting. */
178 	if (folio_test_hugetlb(folio))
179 		return;
180 
181 	nr = folio_nr_pages(folio);
182 
183 	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
184 	if (folio_test_swapbacked(folio)) {
185 		__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
186 		if (folio_test_pmd_mappable(folio))
187 			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
188 	} else if (folio_test_pmd_mappable(folio)) {
189 		__lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
190 		filemap_nr_thps_dec(mapping);
191 	}
192 
193 	/*
194 	 * At this point folio must be either written or cleaned by
195 	 * truncate.  Dirty folio here signals a bug and loss of
196 	 * unwritten data - on ordinary filesystems.
197 	 *
198 	 * But it's harmless on in-memory filesystems like tmpfs; and can
199 	 * occur when a driver which did get_user_pages() sets page dirty
200 	 * before putting it, while the inode is being finally evicted.
201 	 *
202 	 * Below fixes dirty accounting after removing the folio entirely
203 	 * but leaves the dirty flag set: it has no effect for truncated
204 	 * folio and anyway will be cleared before returning folio to
205 	 * buddy allocator.
206 	 */
207 	if (WARN_ON_ONCE(folio_test_dirty(folio) &&
208 			 mapping_can_writeback(mapping)))
209 		folio_account_cleaned(folio, inode_to_wb(mapping->host));
210 }
211 
212 /*
213  * Delete a page from the page cache and free it. Caller has to make
214  * sure the page is locked and that nobody else uses it - or that usage
215  * is safe.  The caller must hold the i_pages lock.
216  */
__filemap_remove_folio(struct folio * folio,void * shadow)217 void __filemap_remove_folio(struct folio *folio, void *shadow)
218 {
219 	struct address_space *mapping = folio->mapping;
220 
221 	trace_mm_filemap_delete_from_page_cache(folio);
222 	filemap_unaccount_folio(mapping, folio);
223 	page_cache_delete(mapping, folio, shadow);
224 }
225 
filemap_free_folio(struct address_space * mapping,struct folio * folio)226 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
227 {
228 	void (*free_folio)(struct folio *);
229 	int refs = 1;
230 
231 	free_folio = mapping->a_ops->free_folio;
232 	if (free_folio)
233 		free_folio(folio);
234 
235 	if (folio_test_large(folio) && !folio_test_hugetlb(folio))
236 		refs = folio_nr_pages(folio);
237 	folio_put_refs(folio, refs);
238 }
239 
240 /**
241  * filemap_remove_folio - Remove folio from page cache.
242  * @folio: The folio.
243  *
244  * This must be called only on folios that are locked and have been
245  * verified to be in the page cache.  It will never put the folio into
246  * the free list because the caller has a reference on the page.
247  */
filemap_remove_folio(struct folio * folio)248 void filemap_remove_folio(struct folio *folio)
249 {
250 	struct address_space *mapping = folio->mapping;
251 
252 	BUG_ON(!folio_test_locked(folio));
253 	spin_lock(&mapping->host->i_lock);
254 	xa_lock_irq(&mapping->i_pages);
255 	__filemap_remove_folio(folio, NULL);
256 	xa_unlock_irq(&mapping->i_pages);
257 	if (mapping_shrinkable(mapping))
258 		inode_add_lru(mapping->host);
259 	spin_unlock(&mapping->host->i_lock);
260 
261 	filemap_free_folio(mapping, folio);
262 }
263 
264 /*
265  * page_cache_delete_batch - delete several folios from page cache
266  * @mapping: the mapping to which folios belong
267  * @fbatch: batch of folios to delete
268  *
269  * The function walks over mapping->i_pages and removes folios passed in
270  * @fbatch from the mapping. The function expects @fbatch to be sorted
271  * by page index and is optimised for it to be dense.
272  * It tolerates holes in @fbatch (mapping entries at those indices are not
273  * modified).
274  *
275  * The function expects the i_pages lock to be held.
276  */
page_cache_delete_batch(struct address_space * mapping,struct folio_batch * fbatch)277 static void page_cache_delete_batch(struct address_space *mapping,
278 			     struct folio_batch *fbatch)
279 {
280 	XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
281 	long total_pages = 0;
282 	int i = 0;
283 	struct folio *folio;
284 
285 	mapping_set_update(&xas, mapping);
286 	xas_for_each(&xas, folio, ULONG_MAX) {
287 		if (i >= folio_batch_count(fbatch))
288 			break;
289 
290 		/* A swap/dax/shadow entry got inserted? Skip it. */
291 		if (xa_is_value(folio))
292 			continue;
293 		/*
294 		 * A page got inserted in our range? Skip it. We have our
295 		 * pages locked so they are protected from being removed.
296 		 * If we see a page whose index is higher than ours, it
297 		 * means our page has been removed, which shouldn't be
298 		 * possible because we're holding the PageLock.
299 		 */
300 		if (folio != fbatch->folios[i]) {
301 			VM_BUG_ON_FOLIO(folio->index >
302 					fbatch->folios[i]->index, folio);
303 			continue;
304 		}
305 
306 		WARN_ON_ONCE(!folio_test_locked(folio));
307 
308 		folio->mapping = NULL;
309 		/* Leave folio->index set: truncation lookup relies on it */
310 
311 		i++;
312 		xas_store(&xas, NULL);
313 		total_pages += folio_nr_pages(folio);
314 	}
315 	mapping->nrpages -= total_pages;
316 }
317 
delete_from_page_cache_batch(struct address_space * mapping,struct folio_batch * fbatch)318 void delete_from_page_cache_batch(struct address_space *mapping,
319 				  struct folio_batch *fbatch)
320 {
321 	int i;
322 
323 	if (!folio_batch_count(fbatch))
324 		return;
325 
326 	spin_lock(&mapping->host->i_lock);
327 	xa_lock_irq(&mapping->i_pages);
328 	for (i = 0; i < folio_batch_count(fbatch); i++) {
329 		struct folio *folio = fbatch->folios[i];
330 
331 		trace_mm_filemap_delete_from_page_cache(folio);
332 		filemap_unaccount_folio(mapping, folio);
333 	}
334 	page_cache_delete_batch(mapping, fbatch);
335 	xa_unlock_irq(&mapping->i_pages);
336 	if (mapping_shrinkable(mapping))
337 		inode_add_lru(mapping->host);
338 	spin_unlock(&mapping->host->i_lock);
339 
340 	for (i = 0; i < folio_batch_count(fbatch); i++)
341 		filemap_free_folio(mapping, fbatch->folios[i]);
342 }
343 
filemap_check_errors(struct address_space * mapping)344 int filemap_check_errors(struct address_space *mapping)
345 {
346 	int ret = 0;
347 	/* Check for outstanding write errors */
348 	if (test_bit(AS_ENOSPC, &mapping->flags) &&
349 	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
350 		ret = -ENOSPC;
351 	if (test_bit(AS_EIO, &mapping->flags) &&
352 	    test_and_clear_bit(AS_EIO, &mapping->flags))
353 		ret = -EIO;
354 	return ret;
355 }
356 EXPORT_SYMBOL(filemap_check_errors);
357 
filemap_check_and_keep_errors(struct address_space * mapping)358 static int filemap_check_and_keep_errors(struct address_space *mapping)
359 {
360 	/* Check for outstanding write errors */
361 	if (test_bit(AS_EIO, &mapping->flags))
362 		return -EIO;
363 	if (test_bit(AS_ENOSPC, &mapping->flags))
364 		return -ENOSPC;
365 	return 0;
366 }
367 
368 /**
369  * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
370  * @mapping:	address space structure to write
371  * @wbc:	the writeback_control controlling the writeout
372  *
373  * Call writepages on the mapping using the provided wbc to control the
374  * writeout.
375  *
376  * Return: %0 on success, negative error code otherwise.
377  */
filemap_fdatawrite_wbc(struct address_space * mapping,struct writeback_control * wbc)378 int filemap_fdatawrite_wbc(struct address_space *mapping,
379 			   struct writeback_control *wbc)
380 {
381 	int ret;
382 
383 	if (!mapping_can_writeback(mapping) ||
384 	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
385 		return 0;
386 
387 	wbc_attach_fdatawrite_inode(wbc, mapping->host);
388 	ret = do_writepages(mapping, wbc);
389 	wbc_detach_inode(wbc);
390 	return ret;
391 }
392 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
393 
394 /**
395  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
396  * @mapping:	address space structure to write
397  * @start:	offset in bytes where the range starts
398  * @end:	offset in bytes where the range ends (inclusive)
399  * @sync_mode:	enable synchronous operation
400  *
401  * Start writeback against all of a mapping's dirty pages that lie
402  * within the byte offsets <start, end> inclusive.
403  *
404  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
405  * opposed to a regular memory cleansing writeback.  The difference between
406  * these two operations is that if a dirty page/buffer is encountered, it must
407  * be waited upon, and not just skipped over.
408  *
409  * Return: %0 on success, negative error code otherwise.
410  */
__filemap_fdatawrite_range(struct address_space * mapping,loff_t start,loff_t end,int sync_mode)411 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
412 				loff_t end, int sync_mode)
413 {
414 	struct writeback_control wbc = {
415 		.sync_mode = sync_mode,
416 		.nr_to_write = LONG_MAX,
417 		.range_start = start,
418 		.range_end = end,
419 	};
420 
421 	return filemap_fdatawrite_wbc(mapping, &wbc);
422 }
423 
__filemap_fdatawrite(struct address_space * mapping,int sync_mode)424 static inline int __filemap_fdatawrite(struct address_space *mapping,
425 	int sync_mode)
426 {
427 	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
428 }
429 
filemap_fdatawrite(struct address_space * mapping)430 int filemap_fdatawrite(struct address_space *mapping)
431 {
432 	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
433 }
434 EXPORT_SYMBOL(filemap_fdatawrite);
435 
filemap_fdatawrite_range(struct address_space * mapping,loff_t start,loff_t end)436 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
437 				loff_t end)
438 {
439 	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
440 }
441 EXPORT_SYMBOL(filemap_fdatawrite_range);
442 
443 /**
444  * filemap_flush - mostly a non-blocking flush
445  * @mapping:	target address_space
446  *
447  * This is a mostly non-blocking flush.  Not suitable for data-integrity
448  * purposes - I/O may not be started against all dirty pages.
449  *
450  * Return: %0 on success, negative error code otherwise.
451  */
filemap_flush(struct address_space * mapping)452 int filemap_flush(struct address_space *mapping)
453 {
454 	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
455 }
456 EXPORT_SYMBOL(filemap_flush);
457 
458 /**
459  * filemap_range_has_page - check if a page exists in range.
460  * @mapping:           address space within which to check
461  * @start_byte:        offset in bytes where the range starts
462  * @end_byte:          offset in bytes where the range ends (inclusive)
463  *
464  * Find at least one page in the range supplied, usually used to check if
465  * direct writing in this range will trigger a writeback.
466  *
467  * Return: %true if at least one page exists in the specified range,
468  * %false otherwise.
469  */
filemap_range_has_page(struct address_space * mapping,loff_t start_byte,loff_t end_byte)470 bool filemap_range_has_page(struct address_space *mapping,
471 			   loff_t start_byte, loff_t end_byte)
472 {
473 	struct page *page;
474 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
475 	pgoff_t max = end_byte >> PAGE_SHIFT;
476 
477 	if (end_byte < start_byte)
478 		return false;
479 
480 	rcu_read_lock();
481 	for (;;) {
482 		page = xas_find(&xas, max);
483 		if (xas_retry(&xas, page))
484 			continue;
485 		/* Shadow entries don't count */
486 		if (xa_is_value(page))
487 			continue;
488 		/*
489 		 * We don't need to try to pin this page; we're about to
490 		 * release the RCU lock anyway.  It is enough to know that
491 		 * there was a page here recently.
492 		 */
493 		break;
494 	}
495 	rcu_read_unlock();
496 
497 	return page != NULL;
498 }
499 EXPORT_SYMBOL(filemap_range_has_page);
500 
__filemap_fdatawait_range(struct address_space * mapping,loff_t start_byte,loff_t end_byte)501 static void __filemap_fdatawait_range(struct address_space *mapping,
502 				     loff_t start_byte, loff_t end_byte)
503 {
504 	pgoff_t index = start_byte >> PAGE_SHIFT;
505 	pgoff_t end = end_byte >> PAGE_SHIFT;
506 	struct pagevec pvec;
507 	int nr_pages;
508 
509 	if (end_byte < start_byte)
510 		return;
511 
512 	pagevec_init(&pvec);
513 	while (index <= end) {
514 		unsigned i;
515 
516 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
517 				end, PAGECACHE_TAG_WRITEBACK);
518 		if (!nr_pages)
519 			break;
520 
521 		for (i = 0; i < nr_pages; i++) {
522 			struct page *page = pvec.pages[i];
523 
524 			wait_on_page_writeback(page);
525 			ClearPageError(page);
526 		}
527 		pagevec_release(&pvec);
528 		cond_resched();
529 	}
530 }
531 
532 /**
533  * filemap_fdatawait_range - wait for writeback to complete
534  * @mapping:		address space structure to wait for
535  * @start_byte:		offset in bytes where the range starts
536  * @end_byte:		offset in bytes where the range ends (inclusive)
537  *
538  * Walk the list of under-writeback pages of the given address space
539  * in the given range and wait for all of them.  Check error status of
540  * the address space and return it.
541  *
542  * Since the error status of the address space is cleared by this function,
543  * callers are responsible for checking the return value and handling and/or
544  * reporting the error.
545  *
546  * Return: error status of the address space.
547  */
filemap_fdatawait_range(struct address_space * mapping,loff_t start_byte,loff_t end_byte)548 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
549 			    loff_t end_byte)
550 {
551 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
552 	return filemap_check_errors(mapping);
553 }
554 EXPORT_SYMBOL(filemap_fdatawait_range);
555 
556 /**
557  * filemap_fdatawait_range_keep_errors - wait for writeback to complete
558  * @mapping:		address space structure to wait for
559  * @start_byte:		offset in bytes where the range starts
560  * @end_byte:		offset in bytes where the range ends (inclusive)
561  *
562  * Walk the list of under-writeback pages of the given address space in the
563  * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
564  * this function does not clear error status of the address space.
565  *
566  * Use this function if callers don't handle errors themselves.  Expected
567  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
568  * fsfreeze(8)
569  */
filemap_fdatawait_range_keep_errors(struct address_space * mapping,loff_t start_byte,loff_t end_byte)570 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
571 		loff_t start_byte, loff_t end_byte)
572 {
573 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
574 	return filemap_check_and_keep_errors(mapping);
575 }
576 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
577 
578 /**
579  * file_fdatawait_range - wait for writeback to complete
580  * @file:		file pointing to address space structure to wait for
581  * @start_byte:		offset in bytes where the range starts
582  * @end_byte:		offset in bytes where the range ends (inclusive)
583  *
584  * Walk the list of under-writeback pages of the address space that file
585  * refers to, in the given range and wait for all of them.  Check error
586  * status of the address space vs. the file->f_wb_err cursor and return it.
587  *
588  * Since the error status of the file is advanced by this function,
589  * callers are responsible for checking the return value and handling and/or
590  * reporting the error.
591  *
592  * Return: error status of the address space vs. the file->f_wb_err cursor.
593  */
file_fdatawait_range(struct file * file,loff_t start_byte,loff_t end_byte)594 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
595 {
596 	struct address_space *mapping = file->f_mapping;
597 
598 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
599 	return file_check_and_advance_wb_err(file);
600 }
601 EXPORT_SYMBOL(file_fdatawait_range);
602 
603 /**
604  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
605  * @mapping: address space structure to wait for
606  *
607  * Walk the list of under-writeback pages of the given address space
608  * and wait for all of them.  Unlike filemap_fdatawait(), this function
609  * does not clear error status of the address space.
610  *
611  * Use this function if callers don't handle errors themselves.  Expected
612  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
613  * fsfreeze(8)
614  *
615  * Return: error status of the address space.
616  */
filemap_fdatawait_keep_errors(struct address_space * mapping)617 int filemap_fdatawait_keep_errors(struct address_space *mapping)
618 {
619 	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
620 	return filemap_check_and_keep_errors(mapping);
621 }
622 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
623 
624 /* Returns true if writeback might be needed or already in progress. */
mapping_needs_writeback(struct address_space * mapping)625 static bool mapping_needs_writeback(struct address_space *mapping)
626 {
627 	return mapping->nrpages;
628 }
629 
filemap_range_has_writeback(struct address_space * mapping,loff_t start_byte,loff_t end_byte)630 bool filemap_range_has_writeback(struct address_space *mapping,
631 				 loff_t start_byte, loff_t end_byte)
632 {
633 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
634 	pgoff_t max = end_byte >> PAGE_SHIFT;
635 	struct page *page;
636 
637 	if (end_byte < start_byte)
638 		return false;
639 
640 	rcu_read_lock();
641 	xas_for_each(&xas, page, max) {
642 		if (xas_retry(&xas, page))
643 			continue;
644 		if (xa_is_value(page))
645 			continue;
646 		if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
647 			break;
648 	}
649 	rcu_read_unlock();
650 	return page != NULL;
651 }
652 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
653 
654 /**
655  * filemap_write_and_wait_range - write out & wait on a file range
656  * @mapping:	the address_space for the pages
657  * @lstart:	offset in bytes where the range starts
658  * @lend:	offset in bytes where the range ends (inclusive)
659  *
660  * Write out and wait upon file offsets lstart->lend, inclusive.
661  *
662  * Note that @lend is inclusive (describes the last byte to be written) so
663  * that this function can be used to write to the very end-of-file (end = -1).
664  *
665  * Return: error status of the address space.
666  */
filemap_write_and_wait_range(struct address_space * mapping,loff_t lstart,loff_t lend)667 int filemap_write_and_wait_range(struct address_space *mapping,
668 				 loff_t lstart, loff_t lend)
669 {
670 	int err = 0;
671 
672 	if (mapping_needs_writeback(mapping)) {
673 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
674 						 WB_SYNC_ALL);
675 		/*
676 		 * Even if the above returned error, the pages may be
677 		 * written partially (e.g. -ENOSPC), so we wait for it.
678 		 * But the -EIO is special case, it may indicate the worst
679 		 * thing (e.g. bug) happened, so we avoid waiting for it.
680 		 */
681 		if (err != -EIO) {
682 			int err2 = filemap_fdatawait_range(mapping,
683 						lstart, lend);
684 			if (!err)
685 				err = err2;
686 		} else {
687 			/* Clear any previously stored errors */
688 			filemap_check_errors(mapping);
689 		}
690 	} else {
691 		err = filemap_check_errors(mapping);
692 	}
693 	return err;
694 }
695 EXPORT_SYMBOL(filemap_write_and_wait_range);
696 
__filemap_set_wb_err(struct address_space * mapping,int err)697 void __filemap_set_wb_err(struct address_space *mapping, int err)
698 {
699 	errseq_t eseq = errseq_set(&mapping->wb_err, err);
700 
701 	trace_filemap_set_wb_err(mapping, eseq);
702 }
703 EXPORT_SYMBOL(__filemap_set_wb_err);
704 
705 /**
706  * file_check_and_advance_wb_err - report wb error (if any) that was previously
707  * 				   and advance wb_err to current one
708  * @file: struct file on which the error is being reported
709  *
710  * When userland calls fsync (or something like nfsd does the equivalent), we
711  * want to report any writeback errors that occurred since the last fsync (or
712  * since the file was opened if there haven't been any).
713  *
714  * Grab the wb_err from the mapping. If it matches what we have in the file,
715  * then just quickly return 0. The file is all caught up.
716  *
717  * If it doesn't match, then take the mapping value, set the "seen" flag in
718  * it and try to swap it into place. If it works, or another task beat us
719  * to it with the new value, then update the f_wb_err and return the error
720  * portion. The error at this point must be reported via proper channels
721  * (a'la fsync, or NFS COMMIT operation, etc.).
722  *
723  * While we handle mapping->wb_err with atomic operations, the f_wb_err
724  * value is protected by the f_lock since we must ensure that it reflects
725  * the latest value swapped in for this file descriptor.
726  *
727  * Return: %0 on success, negative error code otherwise.
728  */
file_check_and_advance_wb_err(struct file * file)729 int file_check_and_advance_wb_err(struct file *file)
730 {
731 	int err = 0;
732 	errseq_t old = READ_ONCE(file->f_wb_err);
733 	struct address_space *mapping = file->f_mapping;
734 
735 	/* Locklessly handle the common case where nothing has changed */
736 	if (errseq_check(&mapping->wb_err, old)) {
737 		/* Something changed, must use slow path */
738 		spin_lock(&file->f_lock);
739 		old = file->f_wb_err;
740 		err = errseq_check_and_advance(&mapping->wb_err,
741 						&file->f_wb_err);
742 		trace_file_check_and_advance_wb_err(file, old);
743 		spin_unlock(&file->f_lock);
744 	}
745 
746 	/*
747 	 * We're mostly using this function as a drop in replacement for
748 	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
749 	 * that the legacy code would have had on these flags.
750 	 */
751 	clear_bit(AS_EIO, &mapping->flags);
752 	clear_bit(AS_ENOSPC, &mapping->flags);
753 	return err;
754 }
755 EXPORT_SYMBOL(file_check_and_advance_wb_err);
756 
757 /**
758  * file_write_and_wait_range - write out & wait on a file range
759  * @file:	file pointing to address_space with pages
760  * @lstart:	offset in bytes where the range starts
761  * @lend:	offset in bytes where the range ends (inclusive)
762  *
763  * Write out and wait upon file offsets lstart->lend, inclusive.
764  *
765  * Note that @lend is inclusive (describes the last byte to be written) so
766  * that this function can be used to write to the very end-of-file (end = -1).
767  *
768  * After writing out and waiting on the data, we check and advance the
769  * f_wb_err cursor to the latest value, and return any errors detected there.
770  *
771  * Return: %0 on success, negative error code otherwise.
772  */
file_write_and_wait_range(struct file * file,loff_t lstart,loff_t lend)773 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
774 {
775 	int err = 0, err2;
776 	struct address_space *mapping = file->f_mapping;
777 
778 	if (mapping_needs_writeback(mapping)) {
779 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
780 						 WB_SYNC_ALL);
781 		/* See comment of filemap_write_and_wait() */
782 		if (err != -EIO)
783 			__filemap_fdatawait_range(mapping, lstart, lend);
784 	}
785 	err2 = file_check_and_advance_wb_err(file);
786 	if (!err)
787 		err = err2;
788 	return err;
789 }
790 EXPORT_SYMBOL(file_write_and_wait_range);
791 
792 /**
793  * replace_page_cache_page - replace a pagecache page with a new one
794  * @old:	page to be replaced
795  * @new:	page to replace with
796  *
797  * This function replaces a page in the pagecache with a new one.  On
798  * success it acquires the pagecache reference for the new page and
799  * drops it for the old page.  Both the old and new pages must be
800  * locked.  This function does not add the new page to the LRU, the
801  * caller must do that.
802  *
803  * The remove + add is atomic.  This function cannot fail.
804  */
replace_page_cache_page(struct page * old,struct page * new)805 void replace_page_cache_page(struct page *old, struct page *new)
806 {
807 	struct folio *fold = page_folio(old);
808 	struct folio *fnew = page_folio(new);
809 	struct address_space *mapping = old->mapping;
810 	void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
811 	pgoff_t offset = old->index;
812 	XA_STATE(xas, &mapping->i_pages, offset);
813 
814 	VM_BUG_ON_PAGE(!PageLocked(old), old);
815 	VM_BUG_ON_PAGE(!PageLocked(new), new);
816 	VM_BUG_ON_PAGE(new->mapping, new);
817 
818 	get_page(new);
819 	new->mapping = mapping;
820 	new->index = offset;
821 
822 	mem_cgroup_migrate(fold, fnew);
823 
824 	xas_lock_irq(&xas);
825 	xas_store(&xas, new);
826 
827 	old->mapping = NULL;
828 	/* hugetlb pages do not participate in page cache accounting. */
829 	if (!PageHuge(old))
830 		__dec_lruvec_page_state(old, NR_FILE_PAGES);
831 	if (!PageHuge(new))
832 		__inc_lruvec_page_state(new, NR_FILE_PAGES);
833 	if (PageSwapBacked(old))
834 		__dec_lruvec_page_state(old, NR_SHMEM);
835 	if (PageSwapBacked(new))
836 		__inc_lruvec_page_state(new, NR_SHMEM);
837 	xas_unlock_irq(&xas);
838 	if (free_folio)
839 		free_folio(fold);
840 	folio_put(fold);
841 }
842 EXPORT_SYMBOL_GPL(replace_page_cache_page);
843 
__filemap_add_folio(struct address_space * mapping,struct folio * folio,pgoff_t index,gfp_t gfp,void ** shadowp)844 noinline int __filemap_add_folio(struct address_space *mapping,
845 		struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
846 {
847 	XA_STATE(xas, &mapping->i_pages, index);
848 	int huge = folio_test_hugetlb(folio);
849 	bool charged = false;
850 	long nr = 1;
851 
852 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
853 	VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
854 	mapping_set_update(&xas, mapping);
855 
856 	if (!huge) {
857 		int error = mem_cgroup_charge(folio, NULL, gfp);
858 		VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
859 		if (error)
860 			return error;
861 		charged = true;
862 		xas_set_order(&xas, index, folio_order(folio));
863 		nr = folio_nr_pages(folio);
864 	}
865 
866 	gfp &= GFP_RECLAIM_MASK;
867 	folio_ref_add(folio, nr);
868 	folio->mapping = mapping;
869 	folio->index = xas.xa_index;
870 
871 	do {
872 		unsigned int order = xa_get_order(xas.xa, xas.xa_index);
873 		void *entry, *old = NULL;
874 
875 		if (order > folio_order(folio))
876 			xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
877 					order, gfp);
878 		xas_lock_irq(&xas);
879 		xas_for_each_conflict(&xas, entry) {
880 			old = entry;
881 			if (!xa_is_value(entry)) {
882 				xas_set_err(&xas, -EEXIST);
883 				goto unlock;
884 			}
885 		}
886 
887 		if (old) {
888 			if (shadowp)
889 				*shadowp = old;
890 			/* entry may have been split before we acquired lock */
891 			order = xa_get_order(xas.xa, xas.xa_index);
892 			if (order > folio_order(folio)) {
893 				/* How to handle large swap entries? */
894 				BUG_ON(shmem_mapping(mapping));
895 				xas_split(&xas, old, order);
896 				xas_reset(&xas);
897 			}
898 		}
899 
900 		xas_store(&xas, folio);
901 		if (xas_error(&xas))
902 			goto unlock;
903 
904 		mapping->nrpages += nr;
905 
906 		/* hugetlb pages do not participate in page cache accounting */
907 		if (!huge) {
908 			__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
909 			if (folio_test_pmd_mappable(folio))
910 				__lruvec_stat_mod_folio(folio,
911 						NR_FILE_THPS, nr);
912 		}
913 unlock:
914 		xas_unlock_irq(&xas);
915 	} while (xas_nomem(&xas, gfp));
916 
917 	if (xas_error(&xas))
918 		goto error;
919 
920 	trace_mm_filemap_add_to_page_cache(folio);
921 	return 0;
922 error:
923 	if (charged)
924 		mem_cgroup_uncharge(folio);
925 	folio->mapping = NULL;
926 	/* Leave page->index set: truncation relies upon it */
927 	folio_put_refs(folio, nr);
928 	return xas_error(&xas);
929 }
930 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
931 
932 /**
933  * add_to_page_cache_locked - add a locked page to the pagecache
934  * @page:	page to add
935  * @mapping:	the page's address_space
936  * @offset:	page index
937  * @gfp_mask:	page allocation mode
938  *
939  * This function is used to add a page to the pagecache. It must be locked.
940  * This function does not add the page to the LRU.  The caller must do that.
941  *
942  * Return: %0 on success, negative error code otherwise.
943  */
add_to_page_cache_locked(struct page * page,struct address_space * mapping,pgoff_t offset,gfp_t gfp_mask)944 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
945 		pgoff_t offset, gfp_t gfp_mask)
946 {
947 	return __filemap_add_folio(mapping, page_folio(page), offset,
948 					  gfp_mask, NULL);
949 }
950 EXPORT_SYMBOL(add_to_page_cache_locked);
951 
filemap_add_folio(struct address_space * mapping,struct folio * folio,pgoff_t index,gfp_t gfp)952 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
953 				pgoff_t index, gfp_t gfp)
954 {
955 	void *shadow = NULL;
956 	int ret;
957 
958 	__folio_set_locked(folio);
959 	ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
960 	if (unlikely(ret))
961 		__folio_clear_locked(folio);
962 	else {
963 		/*
964 		 * The folio might have been evicted from cache only
965 		 * recently, in which case it should be activated like
966 		 * any other repeatedly accessed folio.
967 		 * The exception is folios getting rewritten; evicting other
968 		 * data from the working set, only to cache data that will
969 		 * get overwritten with something else, is a waste of memory.
970 		 */
971 		WARN_ON_ONCE(folio_test_active(folio));
972 		if (!(gfp & __GFP_WRITE) && shadow)
973 			workingset_refault(folio, shadow);
974 		folio_add_lru(folio);
975 	}
976 	return ret;
977 }
978 EXPORT_SYMBOL_GPL(filemap_add_folio);
979 
980 #ifdef CONFIG_NUMA
filemap_alloc_folio(gfp_t gfp,unsigned int order)981 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
982 {
983 	int n;
984 	struct folio *folio;
985 
986 	if (cpuset_do_page_mem_spread()) {
987 		unsigned int cpuset_mems_cookie;
988 		do {
989 			cpuset_mems_cookie = read_mems_allowed_begin();
990 			n = cpuset_mem_spread_node();
991 			folio = __folio_alloc_node(gfp, order, n);
992 		} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
993 
994 		return folio;
995 	}
996 	return folio_alloc(gfp, order);
997 }
998 EXPORT_SYMBOL(filemap_alloc_folio);
999 #endif
1000 
1001 /*
1002  * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1003  *
1004  * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1005  *
1006  * @mapping1: the first mapping to lock
1007  * @mapping2: the second mapping to lock
1008  */
filemap_invalidate_lock_two(struct address_space * mapping1,struct address_space * mapping2)1009 void filemap_invalidate_lock_two(struct address_space *mapping1,
1010 				 struct address_space *mapping2)
1011 {
1012 	if (mapping1 > mapping2)
1013 		swap(mapping1, mapping2);
1014 	if (mapping1)
1015 		down_write(&mapping1->invalidate_lock);
1016 	if (mapping2 && mapping1 != mapping2)
1017 		down_write_nested(&mapping2->invalidate_lock, 1);
1018 }
1019 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1020 
1021 /*
1022  * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1023  *
1024  * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1025  *
1026  * @mapping1: the first mapping to unlock
1027  * @mapping2: the second mapping to unlock
1028  */
filemap_invalidate_unlock_two(struct address_space * mapping1,struct address_space * mapping2)1029 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1030 				   struct address_space *mapping2)
1031 {
1032 	if (mapping1)
1033 		up_write(&mapping1->invalidate_lock);
1034 	if (mapping2 && mapping1 != mapping2)
1035 		up_write(&mapping2->invalidate_lock);
1036 }
1037 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1038 
1039 /*
1040  * In order to wait for pages to become available there must be
1041  * waitqueues associated with pages. By using a hash table of
1042  * waitqueues where the bucket discipline is to maintain all
1043  * waiters on the same queue and wake all when any of the pages
1044  * become available, and for the woken contexts to check to be
1045  * sure the appropriate page became available, this saves space
1046  * at a cost of "thundering herd" phenomena during rare hash
1047  * collisions.
1048  */
1049 #define PAGE_WAIT_TABLE_BITS 8
1050 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1051 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1052 
folio_waitqueue(struct folio * folio)1053 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1054 {
1055 	return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1056 }
1057 
pagecache_init(void)1058 void __init pagecache_init(void)
1059 {
1060 	int i;
1061 
1062 	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1063 		init_waitqueue_head(&folio_wait_table[i]);
1064 
1065 	page_writeback_init();
1066 }
1067 
1068 /*
1069  * The page wait code treats the "wait->flags" somewhat unusually, because
1070  * we have multiple different kinds of waits, not just the usual "exclusive"
1071  * one.
1072  *
1073  * We have:
1074  *
1075  *  (a) no special bits set:
1076  *
1077  *	We're just waiting for the bit to be released, and when a waker
1078  *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1079  *	and remove it from the wait queue.
1080  *
1081  *	Simple and straightforward.
1082  *
1083  *  (b) WQ_FLAG_EXCLUSIVE:
1084  *
1085  *	The waiter is waiting to get the lock, and only one waiter should
1086  *	be woken up to avoid any thundering herd behavior. We'll set the
1087  *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1088  *
1089  *	This is the traditional exclusive wait.
1090  *
1091  *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1092  *
1093  *	The waiter is waiting to get the bit, and additionally wants the
1094  *	lock to be transferred to it for fair lock behavior. If the lock
1095  *	cannot be taken, we stop walking the wait queue without waking
1096  *	the waiter.
1097  *
1098  *	This is the "fair lock handoff" case, and in addition to setting
1099  *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1100  *	that it now has the lock.
1101  */
wake_page_function(wait_queue_entry_t * wait,unsigned mode,int sync,void * arg)1102 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1103 {
1104 	unsigned int flags;
1105 	struct wait_page_key *key = arg;
1106 	struct wait_page_queue *wait_page
1107 		= container_of(wait, struct wait_page_queue, wait);
1108 
1109 	if (!wake_page_match(wait_page, key))
1110 		return 0;
1111 
1112 	/*
1113 	 * If it's a lock handoff wait, we get the bit for it, and
1114 	 * stop walking (and do not wake it up) if we can't.
1115 	 */
1116 	flags = wait->flags;
1117 	if (flags & WQ_FLAG_EXCLUSIVE) {
1118 		if (test_bit(key->bit_nr, &key->folio->flags))
1119 			return -1;
1120 		if (flags & WQ_FLAG_CUSTOM) {
1121 			if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1122 				return -1;
1123 			flags |= WQ_FLAG_DONE;
1124 		}
1125 	}
1126 
1127 	/*
1128 	 * We are holding the wait-queue lock, but the waiter that
1129 	 * is waiting for this will be checking the flags without
1130 	 * any locking.
1131 	 *
1132 	 * So update the flags atomically, and wake up the waiter
1133 	 * afterwards to avoid any races. This store-release pairs
1134 	 * with the load-acquire in folio_wait_bit_common().
1135 	 */
1136 	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1137 	wake_up_state(wait->private, mode);
1138 
1139 	/*
1140 	 * Ok, we have successfully done what we're waiting for,
1141 	 * and we can unconditionally remove the wait entry.
1142 	 *
1143 	 * Note that this pairs with the "finish_wait()" in the
1144 	 * waiter, and has to be the absolute last thing we do.
1145 	 * After this list_del_init(&wait->entry) the wait entry
1146 	 * might be de-allocated and the process might even have
1147 	 * exited.
1148 	 */
1149 	list_del_init_careful(&wait->entry);
1150 	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1151 }
1152 
folio_wake_bit(struct folio * folio,int bit_nr)1153 static void folio_wake_bit(struct folio *folio, int bit_nr)
1154 {
1155 	wait_queue_head_t *q = folio_waitqueue(folio);
1156 	struct wait_page_key key;
1157 	unsigned long flags;
1158 	wait_queue_entry_t bookmark;
1159 
1160 	key.folio = folio;
1161 	key.bit_nr = bit_nr;
1162 	key.page_match = 0;
1163 
1164 	bookmark.flags = 0;
1165 	bookmark.private = NULL;
1166 	bookmark.func = NULL;
1167 	INIT_LIST_HEAD(&bookmark.entry);
1168 
1169 	spin_lock_irqsave(&q->lock, flags);
1170 	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1171 
1172 	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1173 		/*
1174 		 * Take a breather from holding the lock,
1175 		 * allow pages that finish wake up asynchronously
1176 		 * to acquire the lock and remove themselves
1177 		 * from wait queue
1178 		 */
1179 		spin_unlock_irqrestore(&q->lock, flags);
1180 		cpu_relax();
1181 		spin_lock_irqsave(&q->lock, flags);
1182 		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1183 	}
1184 
1185 	/*
1186 	 * It's possible to miss clearing waiters here, when we woke our page
1187 	 * waiters, but the hashed waitqueue has waiters for other pages on it.
1188 	 * That's okay, it's a rare case. The next waker will clear it.
1189 	 *
1190 	 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1191 	 * other), the flag may be cleared in the course of freeing the page;
1192 	 * but that is not required for correctness.
1193 	 */
1194 	if (!waitqueue_active(q) || !key.page_match)
1195 		folio_clear_waiters(folio);
1196 
1197 	spin_unlock_irqrestore(&q->lock, flags);
1198 }
1199 
folio_wake(struct folio * folio,int bit)1200 static void folio_wake(struct folio *folio, int bit)
1201 {
1202 	if (!folio_test_waiters(folio))
1203 		return;
1204 	folio_wake_bit(folio, bit);
1205 }
1206 
1207 /*
1208  * A choice of three behaviors for folio_wait_bit_common():
1209  */
1210 enum behavior {
1211 	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1212 			 * __folio_lock() waiting on then setting PG_locked.
1213 			 */
1214 	SHARED,		/* Hold ref to page and check the bit when woken, like
1215 			 * folio_wait_writeback() waiting on PG_writeback.
1216 			 */
1217 	DROP,		/* Drop ref to page before wait, no check when woken,
1218 			 * like folio_put_wait_locked() on PG_locked.
1219 			 */
1220 };
1221 
1222 /*
1223  * Attempt to check (or get) the folio flag, and mark us done
1224  * if successful.
1225  */
folio_trylock_flag(struct folio * folio,int bit_nr,struct wait_queue_entry * wait)1226 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1227 					struct wait_queue_entry *wait)
1228 {
1229 	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1230 		if (test_and_set_bit(bit_nr, &folio->flags))
1231 			return false;
1232 	} else if (test_bit(bit_nr, &folio->flags))
1233 		return false;
1234 
1235 	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1236 	return true;
1237 }
1238 
1239 /* How many times do we accept lock stealing from under a waiter? */
1240 int sysctl_page_lock_unfairness = 5;
1241 
folio_wait_bit_common(struct folio * folio,int bit_nr,int state,enum behavior behavior)1242 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1243 		int state, enum behavior behavior)
1244 {
1245 	wait_queue_head_t *q = folio_waitqueue(folio);
1246 	int unfairness = sysctl_page_lock_unfairness;
1247 	struct wait_page_queue wait_page;
1248 	wait_queue_entry_t *wait = &wait_page.wait;
1249 	bool thrashing = false;
1250 	bool delayacct = false;
1251 	unsigned long pflags;
1252 
1253 	if (bit_nr == PG_locked &&
1254 	    !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1255 		if (!folio_test_swapbacked(folio)) {
1256 			delayacct_thrashing_start();
1257 			delayacct = true;
1258 		}
1259 		psi_memstall_enter(&pflags);
1260 		thrashing = true;
1261 	}
1262 
1263 	init_wait(wait);
1264 	wait->func = wake_page_function;
1265 	wait_page.folio = folio;
1266 	wait_page.bit_nr = bit_nr;
1267 
1268 repeat:
1269 	wait->flags = 0;
1270 	if (behavior == EXCLUSIVE) {
1271 		wait->flags = WQ_FLAG_EXCLUSIVE;
1272 		if (--unfairness < 0)
1273 			wait->flags |= WQ_FLAG_CUSTOM;
1274 	}
1275 
1276 	/*
1277 	 * Do one last check whether we can get the
1278 	 * page bit synchronously.
1279 	 *
1280 	 * Do the folio_set_waiters() marking before that
1281 	 * to let any waker we _just_ missed know they
1282 	 * need to wake us up (otherwise they'll never
1283 	 * even go to the slow case that looks at the
1284 	 * page queue), and add ourselves to the wait
1285 	 * queue if we need to sleep.
1286 	 *
1287 	 * This part needs to be done under the queue
1288 	 * lock to avoid races.
1289 	 */
1290 	spin_lock_irq(&q->lock);
1291 	folio_set_waiters(folio);
1292 	if (!folio_trylock_flag(folio, bit_nr, wait))
1293 		__add_wait_queue_entry_tail(q, wait);
1294 	spin_unlock_irq(&q->lock);
1295 
1296 	/*
1297 	 * From now on, all the logic will be based on
1298 	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1299 	 * see whether the page bit testing has already
1300 	 * been done by the wake function.
1301 	 *
1302 	 * We can drop our reference to the folio.
1303 	 */
1304 	if (behavior == DROP)
1305 		folio_put(folio);
1306 
1307 	/*
1308 	 * Note that until the "finish_wait()", or until
1309 	 * we see the WQ_FLAG_WOKEN flag, we need to
1310 	 * be very careful with the 'wait->flags', because
1311 	 * we may race with a waker that sets them.
1312 	 */
1313 	for (;;) {
1314 		unsigned int flags;
1315 
1316 		set_current_state(state);
1317 
1318 		/* Loop until we've been woken or interrupted */
1319 		flags = smp_load_acquire(&wait->flags);
1320 		if (!(flags & WQ_FLAG_WOKEN)) {
1321 			if (signal_pending_state(state, current))
1322 				break;
1323 
1324 			io_schedule();
1325 			continue;
1326 		}
1327 
1328 		/* If we were non-exclusive, we're done */
1329 		if (behavior != EXCLUSIVE)
1330 			break;
1331 
1332 		/* If the waker got the lock for us, we're done */
1333 		if (flags & WQ_FLAG_DONE)
1334 			break;
1335 
1336 		/*
1337 		 * Otherwise, if we're getting the lock, we need to
1338 		 * try to get it ourselves.
1339 		 *
1340 		 * And if that fails, we'll have to retry this all.
1341 		 */
1342 		if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1343 			goto repeat;
1344 
1345 		wait->flags |= WQ_FLAG_DONE;
1346 		break;
1347 	}
1348 
1349 	/*
1350 	 * If a signal happened, this 'finish_wait()' may remove the last
1351 	 * waiter from the wait-queues, but the folio waiters bit will remain
1352 	 * set. That's ok. The next wakeup will take care of it, and trying
1353 	 * to do it here would be difficult and prone to races.
1354 	 */
1355 	finish_wait(q, wait);
1356 
1357 	if (thrashing) {
1358 		if (delayacct)
1359 			delayacct_thrashing_end();
1360 		psi_memstall_leave(&pflags);
1361 	}
1362 
1363 	/*
1364 	 * NOTE! The wait->flags weren't stable until we've done the
1365 	 * 'finish_wait()', and we could have exited the loop above due
1366 	 * to a signal, and had a wakeup event happen after the signal
1367 	 * test but before the 'finish_wait()'.
1368 	 *
1369 	 * So only after the finish_wait() can we reliably determine
1370 	 * if we got woken up or not, so we can now figure out the final
1371 	 * return value based on that state without races.
1372 	 *
1373 	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1374 	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1375 	 */
1376 	if (behavior == EXCLUSIVE)
1377 		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1378 
1379 	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1380 }
1381 
1382 #ifdef CONFIG_MIGRATION
1383 /**
1384  * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1385  * @entry: migration swap entry.
1386  * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1387  *        for pte entries, pass NULL for pmd entries.
1388  * @ptl: already locked ptl. This function will drop the lock.
1389  *
1390  * Wait for a migration entry referencing the given page to be removed. This is
1391  * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1392  * this can be called without taking a reference on the page. Instead this
1393  * should be called while holding the ptl for the migration entry referencing
1394  * the page.
1395  *
1396  * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1397  *
1398  * This follows the same logic as folio_wait_bit_common() so see the comments
1399  * there.
1400  */
migration_entry_wait_on_locked(swp_entry_t entry,pte_t * ptep,spinlock_t * ptl)1401 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1402 				spinlock_t *ptl)
1403 {
1404 	struct wait_page_queue wait_page;
1405 	wait_queue_entry_t *wait = &wait_page.wait;
1406 	bool thrashing = false;
1407 	bool delayacct = false;
1408 	unsigned long pflags;
1409 	wait_queue_head_t *q;
1410 	struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1411 
1412 	q = folio_waitqueue(folio);
1413 	if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1414 		if (!folio_test_swapbacked(folio)) {
1415 			delayacct_thrashing_start();
1416 			delayacct = true;
1417 		}
1418 		psi_memstall_enter(&pflags);
1419 		thrashing = true;
1420 	}
1421 
1422 	init_wait(wait);
1423 	wait->func = wake_page_function;
1424 	wait_page.folio = folio;
1425 	wait_page.bit_nr = PG_locked;
1426 	wait->flags = 0;
1427 
1428 	spin_lock_irq(&q->lock);
1429 	folio_set_waiters(folio);
1430 	if (!folio_trylock_flag(folio, PG_locked, wait))
1431 		__add_wait_queue_entry_tail(q, wait);
1432 	spin_unlock_irq(&q->lock);
1433 
1434 	/*
1435 	 * If a migration entry exists for the page the migration path must hold
1436 	 * a valid reference to the page, and it must take the ptl to remove the
1437 	 * migration entry. So the page is valid until the ptl is dropped.
1438 	 */
1439 	if (ptep)
1440 		pte_unmap_unlock(ptep, ptl);
1441 	else
1442 		spin_unlock(ptl);
1443 
1444 	for (;;) {
1445 		unsigned int flags;
1446 
1447 		set_current_state(TASK_UNINTERRUPTIBLE);
1448 
1449 		/* Loop until we've been woken or interrupted */
1450 		flags = smp_load_acquire(&wait->flags);
1451 		if (!(flags & WQ_FLAG_WOKEN)) {
1452 			if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1453 				break;
1454 
1455 			io_schedule();
1456 			continue;
1457 		}
1458 		break;
1459 	}
1460 
1461 	finish_wait(q, wait);
1462 
1463 	if (thrashing) {
1464 		if (delayacct)
1465 			delayacct_thrashing_end();
1466 		psi_memstall_leave(&pflags);
1467 	}
1468 }
1469 #endif
1470 
folio_wait_bit(struct folio * folio,int bit_nr)1471 void folio_wait_bit(struct folio *folio, int bit_nr)
1472 {
1473 	folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1474 }
1475 EXPORT_SYMBOL(folio_wait_bit);
1476 
folio_wait_bit_killable(struct folio * folio,int bit_nr)1477 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1478 {
1479 	return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1480 }
1481 EXPORT_SYMBOL(folio_wait_bit_killable);
1482 
1483 /**
1484  * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1485  * @folio: The folio to wait for.
1486  * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1487  *
1488  * The caller should hold a reference on @folio.  They expect the page to
1489  * become unlocked relatively soon, but do not wish to hold up migration
1490  * (for example) by holding the reference while waiting for the folio to
1491  * come unlocked.  After this function returns, the caller should not
1492  * dereference @folio.
1493  *
1494  * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1495  */
folio_put_wait_locked(struct folio * folio,int state)1496 int folio_put_wait_locked(struct folio *folio, int state)
1497 {
1498 	return folio_wait_bit_common(folio, PG_locked, state, DROP);
1499 }
1500 
1501 /**
1502  * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1503  * @folio: Folio defining the wait queue of interest
1504  * @waiter: Waiter to add to the queue
1505  *
1506  * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1507  */
folio_add_wait_queue(struct folio * folio,wait_queue_entry_t * waiter)1508 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1509 {
1510 	wait_queue_head_t *q = folio_waitqueue(folio);
1511 	unsigned long flags;
1512 
1513 	spin_lock_irqsave(&q->lock, flags);
1514 	__add_wait_queue_entry_tail(q, waiter);
1515 	folio_set_waiters(folio);
1516 	spin_unlock_irqrestore(&q->lock, flags);
1517 }
1518 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1519 
1520 #ifndef clear_bit_unlock_is_negative_byte
1521 
1522 /*
1523  * PG_waiters is the high bit in the same byte as PG_lock.
1524  *
1525  * On x86 (and on many other architectures), we can clear PG_lock and
1526  * test the sign bit at the same time. But if the architecture does
1527  * not support that special operation, we just do this all by hand
1528  * instead.
1529  *
1530  * The read of PG_waiters has to be after (or concurrently with) PG_locked
1531  * being cleared, but a memory barrier should be unnecessary since it is
1532  * in the same byte as PG_locked.
1533  */
clear_bit_unlock_is_negative_byte(long nr,volatile void * mem)1534 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1535 {
1536 	clear_bit_unlock(nr, mem);
1537 	/* smp_mb__after_atomic(); */
1538 	return test_bit(PG_waiters, mem);
1539 }
1540 
1541 #endif
1542 
1543 /**
1544  * folio_unlock - Unlock a locked folio.
1545  * @folio: The folio.
1546  *
1547  * Unlocks the folio and wakes up any thread sleeping on the page lock.
1548  *
1549  * Context: May be called from interrupt or process context.  May not be
1550  * called from NMI context.
1551  */
folio_unlock(struct folio * folio)1552 void folio_unlock(struct folio *folio)
1553 {
1554 	/* Bit 7 allows x86 to check the byte's sign bit */
1555 	BUILD_BUG_ON(PG_waiters != 7);
1556 	BUILD_BUG_ON(PG_locked > 7);
1557 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1558 	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1559 		folio_wake_bit(folio, PG_locked);
1560 }
1561 EXPORT_SYMBOL(folio_unlock);
1562 
1563 /**
1564  * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1565  * @folio: The folio.
1566  *
1567  * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1568  * it.  The folio reference held for PG_private_2 being set is released.
1569  *
1570  * This is, for example, used when a netfs folio is being written to a local
1571  * disk cache, thereby allowing writes to the cache for the same folio to be
1572  * serialised.
1573  */
folio_end_private_2(struct folio * folio)1574 void folio_end_private_2(struct folio *folio)
1575 {
1576 	VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1577 	clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1578 	folio_wake_bit(folio, PG_private_2);
1579 	folio_put(folio);
1580 }
1581 EXPORT_SYMBOL(folio_end_private_2);
1582 
1583 /**
1584  * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1585  * @folio: The folio to wait on.
1586  *
1587  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1588  */
folio_wait_private_2(struct folio * folio)1589 void folio_wait_private_2(struct folio *folio)
1590 {
1591 	while (folio_test_private_2(folio))
1592 		folio_wait_bit(folio, PG_private_2);
1593 }
1594 EXPORT_SYMBOL(folio_wait_private_2);
1595 
1596 /**
1597  * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1598  * @folio: The folio to wait on.
1599  *
1600  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1601  * fatal signal is received by the calling task.
1602  *
1603  * Return:
1604  * - 0 if successful.
1605  * - -EINTR if a fatal signal was encountered.
1606  */
folio_wait_private_2_killable(struct folio * folio)1607 int folio_wait_private_2_killable(struct folio *folio)
1608 {
1609 	int ret = 0;
1610 
1611 	while (folio_test_private_2(folio)) {
1612 		ret = folio_wait_bit_killable(folio, PG_private_2);
1613 		if (ret < 0)
1614 			break;
1615 	}
1616 
1617 	return ret;
1618 }
1619 EXPORT_SYMBOL(folio_wait_private_2_killable);
1620 
1621 /**
1622  * folio_end_writeback - End writeback against a folio.
1623  * @folio: The folio.
1624  */
folio_end_writeback(struct folio * folio)1625 void folio_end_writeback(struct folio *folio)
1626 {
1627 	/*
1628 	 * folio_test_clear_reclaim() could be used here but it is an
1629 	 * atomic operation and overkill in this particular case. Failing
1630 	 * to shuffle a folio marked for immediate reclaim is too mild
1631 	 * a gain to justify taking an atomic operation penalty at the
1632 	 * end of every folio writeback.
1633 	 */
1634 	if (folio_test_reclaim(folio)) {
1635 		folio_clear_reclaim(folio);
1636 		folio_rotate_reclaimable(folio);
1637 	}
1638 
1639 	/*
1640 	 * Writeback does not hold a folio reference of its own, relying
1641 	 * on truncation to wait for the clearing of PG_writeback.
1642 	 * But here we must make sure that the folio is not freed and
1643 	 * reused before the folio_wake().
1644 	 */
1645 	folio_get(folio);
1646 	if (!__folio_end_writeback(folio))
1647 		BUG();
1648 
1649 	smp_mb__after_atomic();
1650 	folio_wake(folio, PG_writeback);
1651 	acct_reclaim_writeback(folio);
1652 	folio_put(folio);
1653 }
1654 EXPORT_SYMBOL(folio_end_writeback);
1655 
1656 /*
1657  * After completing I/O on a page, call this routine to update the page
1658  * flags appropriately
1659  */
page_endio(struct page * page,bool is_write,int err)1660 void page_endio(struct page *page, bool is_write, int err)
1661 {
1662 	if (!is_write) {
1663 		if (!err) {
1664 			SetPageUptodate(page);
1665 		} else {
1666 			ClearPageUptodate(page);
1667 			SetPageError(page);
1668 		}
1669 		unlock_page(page);
1670 	} else {
1671 		if (err) {
1672 			struct address_space *mapping;
1673 
1674 			SetPageError(page);
1675 			mapping = page_mapping(page);
1676 			if (mapping)
1677 				mapping_set_error(mapping, err);
1678 		}
1679 		end_page_writeback(page);
1680 	}
1681 }
1682 EXPORT_SYMBOL_GPL(page_endio);
1683 
1684 /**
1685  * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1686  * @folio: The folio to lock
1687  */
__folio_lock(struct folio * folio)1688 void __folio_lock(struct folio *folio)
1689 {
1690 	folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1691 				EXCLUSIVE);
1692 }
1693 EXPORT_SYMBOL(__folio_lock);
1694 
__folio_lock_killable(struct folio * folio)1695 int __folio_lock_killable(struct folio *folio)
1696 {
1697 	return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1698 					EXCLUSIVE);
1699 }
1700 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1701 
__folio_lock_async(struct folio * folio,struct wait_page_queue * wait)1702 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1703 {
1704 	struct wait_queue_head *q = folio_waitqueue(folio);
1705 	int ret = 0;
1706 
1707 	wait->folio = folio;
1708 	wait->bit_nr = PG_locked;
1709 
1710 	spin_lock_irq(&q->lock);
1711 	__add_wait_queue_entry_tail(q, &wait->wait);
1712 	folio_set_waiters(folio);
1713 	ret = !folio_trylock(folio);
1714 	/*
1715 	 * If we were successful now, we know we're still on the
1716 	 * waitqueue as we're still under the lock. This means it's
1717 	 * safe to remove and return success, we know the callback
1718 	 * isn't going to trigger.
1719 	 */
1720 	if (!ret)
1721 		__remove_wait_queue(q, &wait->wait);
1722 	else
1723 		ret = -EIOCBQUEUED;
1724 	spin_unlock_irq(&q->lock);
1725 	return ret;
1726 }
1727 
1728 /*
1729  * Return values:
1730  * true - folio is locked; mmap_lock is still held.
1731  * false - folio is not locked.
1732  *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1733  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1734  *     which case mmap_lock is still held.
1735  *
1736  * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1737  * with the folio locked and the mmap_lock unperturbed.
1738  */
__folio_lock_or_retry(struct folio * folio,struct mm_struct * mm,unsigned int flags)1739 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1740 			 unsigned int flags)
1741 {
1742 	if (fault_flag_allow_retry_first(flags)) {
1743 		/*
1744 		 * CAUTION! In this case, mmap_lock is not released
1745 		 * even though return 0.
1746 		 */
1747 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1748 			return false;
1749 
1750 		mmap_read_unlock(mm);
1751 		if (flags & FAULT_FLAG_KILLABLE)
1752 			folio_wait_locked_killable(folio);
1753 		else
1754 			folio_wait_locked(folio);
1755 		return false;
1756 	}
1757 	if (flags & FAULT_FLAG_KILLABLE) {
1758 		bool ret;
1759 
1760 		ret = __folio_lock_killable(folio);
1761 		if (ret) {
1762 			mmap_read_unlock(mm);
1763 			return false;
1764 		}
1765 	} else {
1766 		__folio_lock(folio);
1767 	}
1768 
1769 	return true;
1770 }
1771 
1772 /**
1773  * page_cache_next_miss() - Find the next gap in the page cache.
1774  * @mapping: Mapping.
1775  * @index: Index.
1776  * @max_scan: Maximum range to search.
1777  *
1778  * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1779  * gap with the lowest index.
1780  *
1781  * This function may be called under the rcu_read_lock.  However, this will
1782  * not atomically search a snapshot of the cache at a single point in time.
1783  * For example, if a gap is created at index 5, then subsequently a gap is
1784  * created at index 10, page_cache_next_miss covering both indices may
1785  * return 10 if called under the rcu_read_lock.
1786  *
1787  * Return: The index of the gap if found, otherwise an index outside the
1788  * range specified (in which case 'return - index >= max_scan' will be true).
1789  * In the rare case of index wrap-around, 0 will be returned.
1790  */
page_cache_next_miss(struct address_space * mapping,pgoff_t index,unsigned long max_scan)1791 pgoff_t page_cache_next_miss(struct address_space *mapping,
1792 			     pgoff_t index, unsigned long max_scan)
1793 {
1794 	XA_STATE(xas, &mapping->i_pages, index);
1795 
1796 	while (max_scan--) {
1797 		void *entry = xas_next(&xas);
1798 		if (!entry || xa_is_value(entry))
1799 			break;
1800 		if (xas.xa_index == 0)
1801 			break;
1802 	}
1803 
1804 	return xas.xa_index;
1805 }
1806 EXPORT_SYMBOL(page_cache_next_miss);
1807 
1808 /**
1809  * page_cache_prev_miss() - Find the previous gap in the page cache.
1810  * @mapping: Mapping.
1811  * @index: Index.
1812  * @max_scan: Maximum range to search.
1813  *
1814  * Search the range [max(index - max_scan + 1, 0), index] for the
1815  * gap with the highest index.
1816  *
1817  * This function may be called under the rcu_read_lock.  However, this will
1818  * not atomically search a snapshot of the cache at a single point in time.
1819  * For example, if a gap is created at index 10, then subsequently a gap is
1820  * created at index 5, page_cache_prev_miss() covering both indices may
1821  * return 5 if called under the rcu_read_lock.
1822  *
1823  * Return: The index of the gap if found, otherwise an index outside the
1824  * range specified (in which case 'index - return >= max_scan' will be true).
1825  * In the rare case of wrap-around, ULONG_MAX will be returned.
1826  */
page_cache_prev_miss(struct address_space * mapping,pgoff_t index,unsigned long max_scan)1827 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1828 			     pgoff_t index, unsigned long max_scan)
1829 {
1830 	XA_STATE(xas, &mapping->i_pages, index);
1831 
1832 	while (max_scan--) {
1833 		void *entry = xas_prev(&xas);
1834 		if (!entry || xa_is_value(entry))
1835 			break;
1836 		if (xas.xa_index == ULONG_MAX)
1837 			break;
1838 	}
1839 
1840 	return xas.xa_index;
1841 }
1842 EXPORT_SYMBOL(page_cache_prev_miss);
1843 
1844 /*
1845  * Lockless page cache protocol:
1846  * On the lookup side:
1847  * 1. Load the folio from i_pages
1848  * 2. Increment the refcount if it's not zero
1849  * 3. If the folio is not found by xas_reload(), put the refcount and retry
1850  *
1851  * On the removal side:
1852  * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1853  * B. Remove the page from i_pages
1854  * C. Return the page to the page allocator
1855  *
1856  * This means that any page may have its reference count temporarily
1857  * increased by a speculative page cache (or fast GUP) lookup as it can
1858  * be allocated by another user before the RCU grace period expires.
1859  * Because the refcount temporarily acquired here may end up being the
1860  * last refcount on the page, any page allocation must be freeable by
1861  * folio_put().
1862  */
1863 
1864 /*
1865  * mapping_get_entry - Get a page cache entry.
1866  * @mapping: the address_space to search
1867  * @index: The page cache index.
1868  *
1869  * Looks up the page cache entry at @mapping & @index.  If it is a folio,
1870  * it is returned with an increased refcount.  If it is a shadow entry
1871  * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1872  * it is returned without further action.
1873  *
1874  * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1875  */
mapping_get_entry(struct address_space * mapping,pgoff_t index)1876 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1877 {
1878 	XA_STATE(xas, &mapping->i_pages, index);
1879 	struct folio *folio;
1880 
1881 	rcu_read_lock();
1882 repeat:
1883 	xas_reset(&xas);
1884 	folio = xas_load(&xas);
1885 	if (xas_retry(&xas, folio))
1886 		goto repeat;
1887 	/*
1888 	 * A shadow entry of a recently evicted page, or a swap entry from
1889 	 * shmem/tmpfs.  Return it without attempting to raise page count.
1890 	 */
1891 	if (!folio || xa_is_value(folio))
1892 		goto out;
1893 
1894 	if (!folio_try_get_rcu(folio))
1895 		goto repeat;
1896 
1897 	if (unlikely(folio != xas_reload(&xas))) {
1898 		folio_put(folio);
1899 		goto repeat;
1900 	}
1901 out:
1902 	rcu_read_unlock();
1903 
1904 	return folio;
1905 }
1906 
1907 /**
1908  * __filemap_get_folio - Find and get a reference to a folio.
1909  * @mapping: The address_space to search.
1910  * @index: The page index.
1911  * @fgp_flags: %FGP flags modify how the folio is returned.
1912  * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1913  *
1914  * Looks up the page cache entry at @mapping & @index.
1915  *
1916  * @fgp_flags can be zero or more of these flags:
1917  *
1918  * * %FGP_ACCESSED - The folio will be marked accessed.
1919  * * %FGP_LOCK - The folio is returned locked.
1920  * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1921  *   instead of allocating a new folio to replace it.
1922  * * %FGP_CREAT - If no page is present then a new page is allocated using
1923  *   @gfp and added to the page cache and the VM's LRU list.
1924  *   The page is returned locked and with an increased refcount.
1925  * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1926  *   page is already in cache.  If the page was allocated, unlock it before
1927  *   returning so the caller can do the same dance.
1928  * * %FGP_WRITE - The page will be written to by the caller.
1929  * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1930  * * %FGP_NOWAIT - Don't get blocked by page lock.
1931  * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1932  *
1933  * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1934  * if the %GFP flags specified for %FGP_CREAT are atomic.
1935  *
1936  * If there is a page cache page, it is returned with an increased refcount.
1937  *
1938  * Return: The found folio or %NULL otherwise.
1939  */
__filemap_get_folio(struct address_space * mapping,pgoff_t index,int fgp_flags,gfp_t gfp)1940 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1941 		int fgp_flags, gfp_t gfp)
1942 {
1943 	struct folio *folio;
1944 
1945 repeat:
1946 	folio = mapping_get_entry(mapping, index);
1947 	if (xa_is_value(folio)) {
1948 		if (fgp_flags & FGP_ENTRY)
1949 			return folio;
1950 		folio = NULL;
1951 	}
1952 	if (!folio)
1953 		goto no_page;
1954 
1955 	if (fgp_flags & FGP_LOCK) {
1956 		if (fgp_flags & FGP_NOWAIT) {
1957 			if (!folio_trylock(folio)) {
1958 				folio_put(folio);
1959 				return NULL;
1960 			}
1961 		} else {
1962 			folio_lock(folio);
1963 		}
1964 
1965 		/* Has the page been truncated? */
1966 		if (unlikely(folio->mapping != mapping)) {
1967 			folio_unlock(folio);
1968 			folio_put(folio);
1969 			goto repeat;
1970 		}
1971 		VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1972 	}
1973 
1974 	if (fgp_flags & FGP_ACCESSED)
1975 		folio_mark_accessed(folio);
1976 	else if (fgp_flags & FGP_WRITE) {
1977 		/* Clear idle flag for buffer write */
1978 		if (folio_test_idle(folio))
1979 			folio_clear_idle(folio);
1980 	}
1981 
1982 	if (fgp_flags & FGP_STABLE)
1983 		folio_wait_stable(folio);
1984 no_page:
1985 	if (!folio && (fgp_flags & FGP_CREAT)) {
1986 		int err;
1987 		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1988 			gfp |= __GFP_WRITE;
1989 		if (fgp_flags & FGP_NOFS)
1990 			gfp &= ~__GFP_FS;
1991 
1992 		folio = filemap_alloc_folio(gfp, 0);
1993 		if (!folio)
1994 			return NULL;
1995 
1996 		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1997 			fgp_flags |= FGP_LOCK;
1998 
1999 		/* Init accessed so avoid atomic mark_page_accessed later */
2000 		if (fgp_flags & FGP_ACCESSED)
2001 			__folio_set_referenced(folio);
2002 
2003 		err = filemap_add_folio(mapping, folio, index, gfp);
2004 		if (unlikely(err)) {
2005 			folio_put(folio);
2006 			folio = NULL;
2007 			if (err == -EEXIST)
2008 				goto repeat;
2009 		}
2010 
2011 		/*
2012 		 * filemap_add_folio locks the page, and for mmap
2013 		 * we expect an unlocked page.
2014 		 */
2015 		if (folio && (fgp_flags & FGP_FOR_MMAP))
2016 			folio_unlock(folio);
2017 	}
2018 
2019 	return folio;
2020 }
2021 EXPORT_SYMBOL(__filemap_get_folio);
2022 
find_get_entry(struct xa_state * xas,pgoff_t max,xa_mark_t mark)2023 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2024 		xa_mark_t mark)
2025 {
2026 	struct folio *folio;
2027 
2028 retry:
2029 	if (mark == XA_PRESENT)
2030 		folio = xas_find(xas, max);
2031 	else
2032 		folio = xas_find_marked(xas, max, mark);
2033 
2034 	if (xas_retry(xas, folio))
2035 		goto retry;
2036 	/*
2037 	 * A shadow entry of a recently evicted page, a swap
2038 	 * entry from shmem/tmpfs or a DAX entry.  Return it
2039 	 * without attempting to raise page count.
2040 	 */
2041 	if (!folio || xa_is_value(folio))
2042 		return folio;
2043 
2044 	if (!folio_try_get_rcu(folio))
2045 		goto reset;
2046 
2047 	if (unlikely(folio != xas_reload(xas))) {
2048 		folio_put(folio);
2049 		goto reset;
2050 	}
2051 
2052 	return folio;
2053 reset:
2054 	xas_reset(xas);
2055 	goto retry;
2056 }
2057 
2058 /**
2059  * find_get_entries - gang pagecache lookup
2060  * @mapping:	The address_space to search
2061  * @start:	The starting page cache index
2062  * @end:	The final page index (inclusive).
2063  * @fbatch:	Where the resulting entries are placed.
2064  * @indices:	The cache indices corresponding to the entries in @entries
2065  *
2066  * find_get_entries() will search for and return a batch of entries in
2067  * the mapping.  The entries are placed in @fbatch.  find_get_entries()
2068  * takes a reference on any actual folios it returns.
2069  *
2070  * The entries have ascending indexes.  The indices may not be consecutive
2071  * due to not-present entries or large folios.
2072  *
2073  * Any shadow entries of evicted folios, or swap entries from
2074  * shmem/tmpfs, are included in the returned array.
2075  *
2076  * Return: The number of entries which were found.
2077  */
find_get_entries(struct address_space * mapping,pgoff_t start,pgoff_t end,struct folio_batch * fbatch,pgoff_t * indices)2078 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2079 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2080 {
2081 	XA_STATE(xas, &mapping->i_pages, start);
2082 	struct folio *folio;
2083 
2084 	rcu_read_lock();
2085 	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2086 		indices[fbatch->nr] = xas.xa_index;
2087 		if (!folio_batch_add(fbatch, folio))
2088 			break;
2089 	}
2090 	rcu_read_unlock();
2091 
2092 	return folio_batch_count(fbatch);
2093 }
2094 
2095 /**
2096  * find_lock_entries - Find a batch of pagecache entries.
2097  * @mapping:	The address_space to search.
2098  * @start:	The starting page cache index.
2099  * @end:	The final page index (inclusive).
2100  * @fbatch:	Where the resulting entries are placed.
2101  * @indices:	The cache indices of the entries in @fbatch.
2102  *
2103  * find_lock_entries() will return a batch of entries from @mapping.
2104  * Swap, shadow and DAX entries are included.  Folios are returned
2105  * locked and with an incremented refcount.  Folios which are locked
2106  * by somebody else or under writeback are skipped.  Folios which are
2107  * partially outside the range are not returned.
2108  *
2109  * The entries have ascending indexes.  The indices may not be consecutive
2110  * due to not-present entries, large folios, folios which could not be
2111  * locked or folios under writeback.
2112  *
2113  * Return: The number of entries which were found.
2114  */
find_lock_entries(struct address_space * mapping,pgoff_t start,pgoff_t end,struct folio_batch * fbatch,pgoff_t * indices)2115 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2116 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2117 {
2118 	XA_STATE(xas, &mapping->i_pages, start);
2119 	struct folio *folio;
2120 
2121 	rcu_read_lock();
2122 	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2123 		if (!xa_is_value(folio)) {
2124 			if (folio->index < start)
2125 				goto put;
2126 			if (folio->index + folio_nr_pages(folio) - 1 > end)
2127 				goto put;
2128 			if (!folio_trylock(folio))
2129 				goto put;
2130 			if (folio->mapping != mapping ||
2131 			    folio_test_writeback(folio))
2132 				goto unlock;
2133 			VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2134 					folio);
2135 		}
2136 		indices[fbatch->nr] = xas.xa_index;
2137 		if (!folio_batch_add(fbatch, folio))
2138 			break;
2139 		continue;
2140 unlock:
2141 		folio_unlock(folio);
2142 put:
2143 		folio_put(folio);
2144 	}
2145 	rcu_read_unlock();
2146 
2147 	return folio_batch_count(fbatch);
2148 }
2149 
2150 static inline
folio_more_pages(struct folio * folio,pgoff_t index,pgoff_t max)2151 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2152 {
2153 	if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2154 		return false;
2155 	if (index >= max)
2156 		return false;
2157 	return index < folio->index + folio_nr_pages(folio) - 1;
2158 }
2159 
2160 /**
2161  * find_get_pages_range - gang pagecache lookup
2162  * @mapping:	The address_space to search
2163  * @start:	The starting page index
2164  * @end:	The final page index (inclusive)
2165  * @nr_pages:	The maximum number of pages
2166  * @pages:	Where the resulting pages are placed
2167  *
2168  * find_get_pages_range() will search for and return a group of up to @nr_pages
2169  * pages in the mapping starting at index @start and up to index @end
2170  * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
2171  * a reference against the returned pages.
2172  *
2173  * The search returns a group of mapping-contiguous pages with ascending
2174  * indexes.  There may be holes in the indices due to not-present pages.
2175  * We also update @start to index the next page for the traversal.
2176  *
2177  * Return: the number of pages which were found. If this number is
2178  * smaller than @nr_pages, the end of specified range has been
2179  * reached.
2180  */
find_get_pages_range(struct address_space * mapping,pgoff_t * start,pgoff_t end,unsigned int nr_pages,struct page ** pages)2181 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2182 			      pgoff_t end, unsigned int nr_pages,
2183 			      struct page **pages)
2184 {
2185 	XA_STATE(xas, &mapping->i_pages, *start);
2186 	struct folio *folio;
2187 	unsigned ret = 0;
2188 
2189 	if (unlikely(!nr_pages))
2190 		return 0;
2191 
2192 	rcu_read_lock();
2193 	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2194 		/* Skip over shadow, swap and DAX entries */
2195 		if (xa_is_value(folio))
2196 			continue;
2197 
2198 again:
2199 		pages[ret] = folio_file_page(folio, xas.xa_index);
2200 		if (++ret == nr_pages) {
2201 			*start = xas.xa_index + 1;
2202 			goto out;
2203 		}
2204 		if (folio_more_pages(folio, xas.xa_index, end)) {
2205 			xas.xa_index++;
2206 			folio_ref_inc(folio);
2207 			goto again;
2208 		}
2209 	}
2210 
2211 	/*
2212 	 * We come here when there is no page beyond @end. We take care to not
2213 	 * overflow the index @start as it confuses some of the callers. This
2214 	 * breaks the iteration when there is a page at index -1 but that is
2215 	 * already broken anyway.
2216 	 */
2217 	if (end == (pgoff_t)-1)
2218 		*start = (pgoff_t)-1;
2219 	else
2220 		*start = end + 1;
2221 out:
2222 	rcu_read_unlock();
2223 
2224 	return ret;
2225 }
2226 
2227 /**
2228  * find_get_pages_contig - gang contiguous pagecache lookup
2229  * @mapping:	The address_space to search
2230  * @index:	The starting page index
2231  * @nr_pages:	The maximum number of pages
2232  * @pages:	Where the resulting pages are placed
2233  *
2234  * find_get_pages_contig() works exactly like find_get_pages_range(),
2235  * except that the returned number of pages are guaranteed to be
2236  * contiguous.
2237  *
2238  * Return: the number of pages which were found.
2239  */
find_get_pages_contig(struct address_space * mapping,pgoff_t index,unsigned int nr_pages,struct page ** pages)2240 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2241 			       unsigned int nr_pages, struct page **pages)
2242 {
2243 	XA_STATE(xas, &mapping->i_pages, index);
2244 	struct folio *folio;
2245 	unsigned int ret = 0;
2246 
2247 	if (unlikely(!nr_pages))
2248 		return 0;
2249 
2250 	rcu_read_lock();
2251 	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2252 		if (xas_retry(&xas, folio))
2253 			continue;
2254 		/*
2255 		 * If the entry has been swapped out, we can stop looking.
2256 		 * No current caller is looking for DAX entries.
2257 		 */
2258 		if (xa_is_value(folio))
2259 			break;
2260 
2261 		if (!folio_try_get_rcu(folio))
2262 			goto retry;
2263 
2264 		if (unlikely(folio != xas_reload(&xas)))
2265 			goto put_page;
2266 
2267 again:
2268 		pages[ret] = folio_file_page(folio, xas.xa_index);
2269 		if (++ret == nr_pages)
2270 			break;
2271 		if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
2272 			xas.xa_index++;
2273 			folio_ref_inc(folio);
2274 			goto again;
2275 		}
2276 		continue;
2277 put_page:
2278 		folio_put(folio);
2279 retry:
2280 		xas_reset(&xas);
2281 	}
2282 	rcu_read_unlock();
2283 	return ret;
2284 }
2285 EXPORT_SYMBOL(find_get_pages_contig);
2286 
2287 /**
2288  * find_get_pages_range_tag - Find and return head pages matching @tag.
2289  * @mapping:	the address_space to search
2290  * @index:	the starting page index
2291  * @end:	The final page index (inclusive)
2292  * @tag:	the tag index
2293  * @nr_pages:	the maximum number of pages
2294  * @pages:	where the resulting pages are placed
2295  *
2296  * Like find_get_pages_range(), except we only return head pages which are
2297  * tagged with @tag.  @index is updated to the index immediately after the
2298  * last page we return, ready for the next iteration.
2299  *
2300  * Return: the number of pages which were found.
2301  */
find_get_pages_range_tag(struct address_space * mapping,pgoff_t * index,pgoff_t end,xa_mark_t tag,unsigned int nr_pages,struct page ** pages)2302 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2303 			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2304 			struct page **pages)
2305 {
2306 	XA_STATE(xas, &mapping->i_pages, *index);
2307 	struct folio *folio;
2308 	unsigned ret = 0;
2309 
2310 	if (unlikely(!nr_pages))
2311 		return 0;
2312 
2313 	rcu_read_lock();
2314 	while ((folio = find_get_entry(&xas, end, tag))) {
2315 		/*
2316 		 * Shadow entries should never be tagged, but this iteration
2317 		 * is lockless so there is a window for page reclaim to evict
2318 		 * a page we saw tagged.  Skip over it.
2319 		 */
2320 		if (xa_is_value(folio))
2321 			continue;
2322 
2323 		pages[ret] = &folio->page;
2324 		if (++ret == nr_pages) {
2325 			*index = folio->index + folio_nr_pages(folio);
2326 			goto out;
2327 		}
2328 	}
2329 
2330 	/*
2331 	 * We come here when we got to @end. We take care to not overflow the
2332 	 * index @index as it confuses some of the callers. This breaks the
2333 	 * iteration when there is a page at index -1 but that is already
2334 	 * broken anyway.
2335 	 */
2336 	if (end == (pgoff_t)-1)
2337 		*index = (pgoff_t)-1;
2338 	else
2339 		*index = end + 1;
2340 out:
2341 	rcu_read_unlock();
2342 
2343 	return ret;
2344 }
2345 EXPORT_SYMBOL(find_get_pages_range_tag);
2346 
2347 /*
2348  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2349  * a _large_ part of the i/o request. Imagine the worst scenario:
2350  *
2351  *      ---R__________________________________________B__________
2352  *         ^ reading here                             ^ bad block(assume 4k)
2353  *
2354  * read(R) => miss => readahead(R...B) => media error => frustrating retries
2355  * => failing the whole request => read(R) => read(R+1) =>
2356  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2357  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2358  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2359  *
2360  * It is going insane. Fix it by quickly scaling down the readahead size.
2361  */
shrink_readahead_size_eio(struct file_ra_state * ra)2362 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2363 {
2364 	ra->ra_pages /= 4;
2365 }
2366 
2367 /*
2368  * filemap_get_read_batch - Get a batch of folios for read
2369  *
2370  * Get a batch of folios which represent a contiguous range of bytes in
2371  * the file.  No exceptional entries will be returned.  If @index is in
2372  * the middle of a folio, the entire folio will be returned.  The last
2373  * folio in the batch may have the readahead flag set or the uptodate flag
2374  * clear so that the caller can take the appropriate action.
2375  */
filemap_get_read_batch(struct address_space * mapping,pgoff_t index,pgoff_t max,struct folio_batch * fbatch)2376 static void filemap_get_read_batch(struct address_space *mapping,
2377 		pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2378 {
2379 	XA_STATE(xas, &mapping->i_pages, index);
2380 	struct folio *folio;
2381 
2382 	rcu_read_lock();
2383 	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2384 		if (xas_retry(&xas, folio))
2385 			continue;
2386 		if (xas.xa_index > max || xa_is_value(folio))
2387 			break;
2388 		if (xa_is_sibling(folio))
2389 			break;
2390 		if (!folio_try_get_rcu(folio))
2391 			goto retry;
2392 
2393 		if (unlikely(folio != xas_reload(&xas)))
2394 			goto put_folio;
2395 
2396 		if (!folio_batch_add(fbatch, folio))
2397 			break;
2398 		if (!folio_test_uptodate(folio))
2399 			break;
2400 		if (folio_test_readahead(folio))
2401 			break;
2402 		xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2403 		continue;
2404 put_folio:
2405 		folio_put(folio);
2406 retry:
2407 		xas_reset(&xas);
2408 	}
2409 	rcu_read_unlock();
2410 }
2411 
filemap_read_folio(struct file * file,struct address_space * mapping,struct folio * folio)2412 static int filemap_read_folio(struct file *file, struct address_space *mapping,
2413 		struct folio *folio)
2414 {
2415 	int error;
2416 
2417 	/*
2418 	 * A previous I/O error may have been due to temporary failures,
2419 	 * eg. multipath errors.  PG_error will be set again if read_folio
2420 	 * fails.
2421 	 */
2422 	folio_clear_error(folio);
2423 	/* Start the actual read. The read will unlock the page. */
2424 	error = mapping->a_ops->read_folio(file, folio);
2425 	if (error)
2426 		return error;
2427 
2428 	error = folio_wait_locked_killable(folio);
2429 	if (error)
2430 		return error;
2431 	if (folio_test_uptodate(folio))
2432 		return 0;
2433 	shrink_readahead_size_eio(&file->f_ra);
2434 	return -EIO;
2435 }
2436 
filemap_range_uptodate(struct address_space * mapping,loff_t pos,struct iov_iter * iter,struct folio * folio)2437 static bool filemap_range_uptodate(struct address_space *mapping,
2438 		loff_t pos, struct iov_iter *iter, struct folio *folio)
2439 {
2440 	int count;
2441 
2442 	if (folio_test_uptodate(folio))
2443 		return true;
2444 	/* pipes can't handle partially uptodate pages */
2445 	if (iov_iter_is_pipe(iter))
2446 		return false;
2447 	if (!mapping->a_ops->is_partially_uptodate)
2448 		return false;
2449 	if (mapping->host->i_blkbits >= folio_shift(folio))
2450 		return false;
2451 
2452 	count = iter->count;
2453 	if (folio_pos(folio) > pos) {
2454 		count -= folio_pos(folio) - pos;
2455 		pos = 0;
2456 	} else {
2457 		pos -= folio_pos(folio);
2458 	}
2459 
2460 	return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2461 }
2462 
filemap_update_page(struct kiocb * iocb,struct address_space * mapping,struct iov_iter * iter,struct folio * folio)2463 static int filemap_update_page(struct kiocb *iocb,
2464 		struct address_space *mapping, struct iov_iter *iter,
2465 		struct folio *folio)
2466 {
2467 	int error;
2468 
2469 	if (iocb->ki_flags & IOCB_NOWAIT) {
2470 		if (!filemap_invalidate_trylock_shared(mapping))
2471 			return -EAGAIN;
2472 	} else {
2473 		filemap_invalidate_lock_shared(mapping);
2474 	}
2475 
2476 	if (!folio_trylock(folio)) {
2477 		error = -EAGAIN;
2478 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2479 			goto unlock_mapping;
2480 		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2481 			filemap_invalidate_unlock_shared(mapping);
2482 			/*
2483 			 * This is where we usually end up waiting for a
2484 			 * previously submitted readahead to finish.
2485 			 */
2486 			folio_put_wait_locked(folio, TASK_KILLABLE);
2487 			return AOP_TRUNCATED_PAGE;
2488 		}
2489 		error = __folio_lock_async(folio, iocb->ki_waitq);
2490 		if (error)
2491 			goto unlock_mapping;
2492 	}
2493 
2494 	error = AOP_TRUNCATED_PAGE;
2495 	if (!folio->mapping)
2496 		goto unlock;
2497 
2498 	error = 0;
2499 	if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2500 		goto unlock;
2501 
2502 	error = -EAGAIN;
2503 	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2504 		goto unlock;
2505 
2506 	error = filemap_read_folio(iocb->ki_filp, mapping, folio);
2507 	goto unlock_mapping;
2508 unlock:
2509 	folio_unlock(folio);
2510 unlock_mapping:
2511 	filemap_invalidate_unlock_shared(mapping);
2512 	if (error == AOP_TRUNCATED_PAGE)
2513 		folio_put(folio);
2514 	return error;
2515 }
2516 
filemap_create_folio(struct file * file,struct address_space * mapping,pgoff_t index,struct folio_batch * fbatch)2517 static int filemap_create_folio(struct file *file,
2518 		struct address_space *mapping, pgoff_t index,
2519 		struct folio_batch *fbatch)
2520 {
2521 	struct folio *folio;
2522 	int error;
2523 
2524 	folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2525 	if (!folio)
2526 		return -ENOMEM;
2527 
2528 	/*
2529 	 * Protect against truncate / hole punch. Grabbing invalidate_lock
2530 	 * here assures we cannot instantiate and bring uptodate new
2531 	 * pagecache folios after evicting page cache during truncate
2532 	 * and before actually freeing blocks.	Note that we could
2533 	 * release invalidate_lock after inserting the folio into
2534 	 * the page cache as the locked folio would then be enough to
2535 	 * synchronize with hole punching. But there are code paths
2536 	 * such as filemap_update_page() filling in partially uptodate
2537 	 * pages or ->readahead() that need to hold invalidate_lock
2538 	 * while mapping blocks for IO so let's hold the lock here as
2539 	 * well to keep locking rules simple.
2540 	 */
2541 	filemap_invalidate_lock_shared(mapping);
2542 	error = filemap_add_folio(mapping, folio, index,
2543 			mapping_gfp_constraint(mapping, GFP_KERNEL));
2544 	if (error == -EEXIST)
2545 		error = AOP_TRUNCATED_PAGE;
2546 	if (error)
2547 		goto error;
2548 
2549 	error = filemap_read_folio(file, mapping, folio);
2550 	if (error)
2551 		goto error;
2552 
2553 	filemap_invalidate_unlock_shared(mapping);
2554 	folio_batch_add(fbatch, folio);
2555 	return 0;
2556 error:
2557 	filemap_invalidate_unlock_shared(mapping);
2558 	folio_put(folio);
2559 	return error;
2560 }
2561 
filemap_readahead(struct kiocb * iocb,struct file * file,struct address_space * mapping,struct folio * folio,pgoff_t last_index)2562 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2563 		struct address_space *mapping, struct folio *folio,
2564 		pgoff_t last_index)
2565 {
2566 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2567 
2568 	if (iocb->ki_flags & IOCB_NOIO)
2569 		return -EAGAIN;
2570 	page_cache_async_ra(&ractl, folio, last_index - folio->index);
2571 	return 0;
2572 }
2573 
filemap_get_pages(struct kiocb * iocb,struct iov_iter * iter,struct folio_batch * fbatch)2574 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2575 		struct folio_batch *fbatch)
2576 {
2577 	struct file *filp = iocb->ki_filp;
2578 	struct address_space *mapping = filp->f_mapping;
2579 	struct file_ra_state *ra = &filp->f_ra;
2580 	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2581 	pgoff_t last_index;
2582 	struct folio *folio;
2583 	int err = 0;
2584 
2585 	last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2586 retry:
2587 	if (fatal_signal_pending(current))
2588 		return -EINTR;
2589 
2590 	filemap_get_read_batch(mapping, index, last_index, fbatch);
2591 	if (!folio_batch_count(fbatch)) {
2592 		if (iocb->ki_flags & IOCB_NOIO)
2593 			return -EAGAIN;
2594 		page_cache_sync_readahead(mapping, ra, filp, index,
2595 				last_index - index);
2596 		filemap_get_read_batch(mapping, index, last_index, fbatch);
2597 	}
2598 	if (!folio_batch_count(fbatch)) {
2599 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2600 			return -EAGAIN;
2601 		err = filemap_create_folio(filp, mapping,
2602 				iocb->ki_pos >> PAGE_SHIFT, fbatch);
2603 		if (err == AOP_TRUNCATED_PAGE)
2604 			goto retry;
2605 		return err;
2606 	}
2607 
2608 	folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2609 	if (folio_test_readahead(folio)) {
2610 		err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2611 		if (err)
2612 			goto err;
2613 	}
2614 	if (!folio_test_uptodate(folio)) {
2615 		if ((iocb->ki_flags & IOCB_WAITQ) &&
2616 		    folio_batch_count(fbatch) > 1)
2617 			iocb->ki_flags |= IOCB_NOWAIT;
2618 		err = filemap_update_page(iocb, mapping, iter, folio);
2619 		if (err)
2620 			goto err;
2621 	}
2622 
2623 	return 0;
2624 err:
2625 	if (err < 0)
2626 		folio_put(folio);
2627 	if (likely(--fbatch->nr))
2628 		return 0;
2629 	if (err == AOP_TRUNCATED_PAGE)
2630 		goto retry;
2631 	return err;
2632 }
2633 
pos_same_folio(loff_t pos1,loff_t pos2,struct folio * folio)2634 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2635 {
2636 	unsigned int shift = folio_shift(folio);
2637 
2638 	return (pos1 >> shift == pos2 >> shift);
2639 }
2640 
2641 /**
2642  * filemap_read - Read data from the page cache.
2643  * @iocb: The iocb to read.
2644  * @iter: Destination for the data.
2645  * @already_read: Number of bytes already read by the caller.
2646  *
2647  * Copies data from the page cache.  If the data is not currently present,
2648  * uses the readahead and read_folio address_space operations to fetch it.
2649  *
2650  * Return: Total number of bytes copied, including those already read by
2651  * the caller.  If an error happens before any bytes are copied, returns
2652  * a negative error number.
2653  */
filemap_read(struct kiocb * iocb,struct iov_iter * iter,ssize_t already_read)2654 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2655 		ssize_t already_read)
2656 {
2657 	struct file *filp = iocb->ki_filp;
2658 	struct file_ra_state *ra = &filp->f_ra;
2659 	struct address_space *mapping = filp->f_mapping;
2660 	struct inode *inode = mapping->host;
2661 	struct folio_batch fbatch;
2662 	int i, error = 0;
2663 	bool writably_mapped;
2664 	loff_t isize, end_offset;
2665 
2666 	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2667 		return 0;
2668 	if (unlikely(!iov_iter_count(iter)))
2669 		return 0;
2670 
2671 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2672 	folio_batch_init(&fbatch);
2673 
2674 	do {
2675 		cond_resched();
2676 
2677 		/*
2678 		 * If we've already successfully copied some data, then we
2679 		 * can no longer safely return -EIOCBQUEUED. Hence mark
2680 		 * an async read NOWAIT at that point.
2681 		 */
2682 		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2683 			iocb->ki_flags |= IOCB_NOWAIT;
2684 
2685 		if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2686 			break;
2687 
2688 		error = filemap_get_pages(iocb, iter, &fbatch);
2689 		if (error < 0)
2690 			break;
2691 
2692 		/*
2693 		 * i_size must be checked after we know the pages are Uptodate.
2694 		 *
2695 		 * Checking i_size after the check allows us to calculate
2696 		 * the correct value for "nr", which means the zero-filled
2697 		 * part of the page is not copied back to userspace (unless
2698 		 * another truncate extends the file - this is desired though).
2699 		 */
2700 		isize = i_size_read(inode);
2701 		if (unlikely(iocb->ki_pos >= isize))
2702 			goto put_folios;
2703 		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2704 
2705 		/*
2706 		 * Once we start copying data, we don't want to be touching any
2707 		 * cachelines that might be contended:
2708 		 */
2709 		writably_mapped = mapping_writably_mapped(mapping);
2710 
2711 		/*
2712 		 * When a read accesses the same folio several times, only
2713 		 * mark it as accessed the first time.
2714 		 */
2715 		if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1,
2716 							fbatch.folios[0]))
2717 			folio_mark_accessed(fbatch.folios[0]);
2718 
2719 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2720 			struct folio *folio = fbatch.folios[i];
2721 			size_t fsize = folio_size(folio);
2722 			size_t offset = iocb->ki_pos & (fsize - 1);
2723 			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2724 					     fsize - offset);
2725 			size_t copied;
2726 
2727 			if (end_offset < folio_pos(folio))
2728 				break;
2729 			if (i > 0)
2730 				folio_mark_accessed(folio);
2731 			/*
2732 			 * If users can be writing to this folio using arbitrary
2733 			 * virtual addresses, take care of potential aliasing
2734 			 * before reading the folio on the kernel side.
2735 			 */
2736 			if (writably_mapped)
2737 				flush_dcache_folio(folio);
2738 
2739 			copied = copy_folio_to_iter(folio, offset, bytes, iter);
2740 
2741 			already_read += copied;
2742 			iocb->ki_pos += copied;
2743 			ra->prev_pos = iocb->ki_pos;
2744 
2745 			if (copied < bytes) {
2746 				error = -EFAULT;
2747 				break;
2748 			}
2749 		}
2750 put_folios:
2751 		for (i = 0; i < folio_batch_count(&fbatch); i++)
2752 			folio_put(fbatch.folios[i]);
2753 		folio_batch_init(&fbatch);
2754 	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2755 
2756 	file_accessed(filp);
2757 
2758 	return already_read ? already_read : error;
2759 }
2760 EXPORT_SYMBOL_GPL(filemap_read);
2761 
2762 /**
2763  * generic_file_read_iter - generic filesystem read routine
2764  * @iocb:	kernel I/O control block
2765  * @iter:	destination for the data read
2766  *
2767  * This is the "read_iter()" routine for all filesystems
2768  * that can use the page cache directly.
2769  *
2770  * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2771  * be returned when no data can be read without waiting for I/O requests
2772  * to complete; it doesn't prevent readahead.
2773  *
2774  * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2775  * requests shall be made for the read or for readahead.  When no data
2776  * can be read, -EAGAIN shall be returned.  When readahead would be
2777  * triggered, a partial, possibly empty read shall be returned.
2778  *
2779  * Return:
2780  * * number of bytes copied, even for partial reads
2781  * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2782  */
2783 ssize_t
generic_file_read_iter(struct kiocb * iocb,struct iov_iter * iter)2784 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2785 {
2786 	size_t count = iov_iter_count(iter);
2787 	ssize_t retval = 0;
2788 
2789 	if (!count)
2790 		return 0; /* skip atime */
2791 
2792 	if (iocb->ki_flags & IOCB_DIRECT) {
2793 		struct file *file = iocb->ki_filp;
2794 		struct address_space *mapping = file->f_mapping;
2795 		struct inode *inode = mapping->host;
2796 
2797 		if (iocb->ki_flags & IOCB_NOWAIT) {
2798 			if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2799 						iocb->ki_pos + count - 1))
2800 				return -EAGAIN;
2801 		} else {
2802 			retval = filemap_write_and_wait_range(mapping,
2803 						iocb->ki_pos,
2804 					        iocb->ki_pos + count - 1);
2805 			if (retval < 0)
2806 				return retval;
2807 		}
2808 
2809 		file_accessed(file);
2810 
2811 		retval = mapping->a_ops->direct_IO(iocb, iter);
2812 		if (retval >= 0) {
2813 			iocb->ki_pos += retval;
2814 			count -= retval;
2815 		}
2816 		if (retval != -EIOCBQUEUED)
2817 			iov_iter_revert(iter, count - iov_iter_count(iter));
2818 
2819 		/*
2820 		 * Btrfs can have a short DIO read if we encounter
2821 		 * compressed extents, so if there was an error, or if
2822 		 * we've already read everything we wanted to, or if
2823 		 * there was a short read because we hit EOF, go ahead
2824 		 * and return.  Otherwise fallthrough to buffered io for
2825 		 * the rest of the read.  Buffered reads will not work for
2826 		 * DAX files, so don't bother trying.
2827 		 */
2828 		if (retval < 0 || !count || IS_DAX(inode))
2829 			return retval;
2830 		if (iocb->ki_pos >= i_size_read(inode))
2831 			return retval;
2832 	}
2833 
2834 	return filemap_read(iocb, iter, retval);
2835 }
2836 EXPORT_SYMBOL(generic_file_read_iter);
2837 
folio_seek_hole_data(struct xa_state * xas,struct address_space * mapping,struct folio * folio,loff_t start,loff_t end,bool seek_data)2838 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2839 		struct address_space *mapping, struct folio *folio,
2840 		loff_t start, loff_t end, bool seek_data)
2841 {
2842 	const struct address_space_operations *ops = mapping->a_ops;
2843 	size_t offset, bsz = i_blocksize(mapping->host);
2844 
2845 	if (xa_is_value(folio) || folio_test_uptodate(folio))
2846 		return seek_data ? start : end;
2847 	if (!ops->is_partially_uptodate)
2848 		return seek_data ? end : start;
2849 
2850 	xas_pause(xas);
2851 	rcu_read_unlock();
2852 	folio_lock(folio);
2853 	if (unlikely(folio->mapping != mapping))
2854 		goto unlock;
2855 
2856 	offset = offset_in_folio(folio, start) & ~(bsz - 1);
2857 
2858 	do {
2859 		if (ops->is_partially_uptodate(folio, offset, bsz) ==
2860 							seek_data)
2861 			break;
2862 		start = (start + bsz) & ~(bsz - 1);
2863 		offset += bsz;
2864 	} while (offset < folio_size(folio));
2865 unlock:
2866 	folio_unlock(folio);
2867 	rcu_read_lock();
2868 	return start;
2869 }
2870 
seek_folio_size(struct xa_state * xas,struct folio * folio)2871 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2872 {
2873 	if (xa_is_value(folio))
2874 		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2875 	return folio_size(folio);
2876 }
2877 
2878 /**
2879  * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2880  * @mapping: Address space to search.
2881  * @start: First byte to consider.
2882  * @end: Limit of search (exclusive).
2883  * @whence: Either SEEK_HOLE or SEEK_DATA.
2884  *
2885  * If the page cache knows which blocks contain holes and which blocks
2886  * contain data, your filesystem can use this function to implement
2887  * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
2888  * entirely memory-based such as tmpfs, and filesystems which support
2889  * unwritten extents.
2890  *
2891  * Return: The requested offset on success, or -ENXIO if @whence specifies
2892  * SEEK_DATA and there is no data after @start.  There is an implicit hole
2893  * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2894  * and @end contain data.
2895  */
mapping_seek_hole_data(struct address_space * mapping,loff_t start,loff_t end,int whence)2896 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2897 		loff_t end, int whence)
2898 {
2899 	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2900 	pgoff_t max = (end - 1) >> PAGE_SHIFT;
2901 	bool seek_data = (whence == SEEK_DATA);
2902 	struct folio *folio;
2903 
2904 	if (end <= start)
2905 		return -ENXIO;
2906 
2907 	rcu_read_lock();
2908 	while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
2909 		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2910 		size_t seek_size;
2911 
2912 		if (start < pos) {
2913 			if (!seek_data)
2914 				goto unlock;
2915 			start = pos;
2916 		}
2917 
2918 		seek_size = seek_folio_size(&xas, folio);
2919 		pos = round_up((u64)pos + 1, seek_size);
2920 		start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
2921 				seek_data);
2922 		if (start < pos)
2923 			goto unlock;
2924 		if (start >= end)
2925 			break;
2926 		if (seek_size > PAGE_SIZE)
2927 			xas_set(&xas, pos >> PAGE_SHIFT);
2928 		if (!xa_is_value(folio))
2929 			folio_put(folio);
2930 	}
2931 	if (seek_data)
2932 		start = -ENXIO;
2933 unlock:
2934 	rcu_read_unlock();
2935 	if (folio && !xa_is_value(folio))
2936 		folio_put(folio);
2937 	if (start > end)
2938 		return end;
2939 	return start;
2940 }
2941 
2942 #ifdef CONFIG_MMU
2943 #define MMAP_LOTSAMISS  (100)
2944 /*
2945  * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2946  * @vmf - the vm_fault for this fault.
2947  * @folio - the folio to lock.
2948  * @fpin - the pointer to the file we may pin (or is already pinned).
2949  *
2950  * This works similar to lock_folio_or_retry in that it can drop the
2951  * mmap_lock.  It differs in that it actually returns the folio locked
2952  * if it returns 1 and 0 if it couldn't lock the folio.  If we did have
2953  * to drop the mmap_lock then fpin will point to the pinned file and
2954  * needs to be fput()'ed at a later point.
2955  */
lock_folio_maybe_drop_mmap(struct vm_fault * vmf,struct folio * folio,struct file ** fpin)2956 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
2957 				     struct file **fpin)
2958 {
2959 	if (folio_trylock(folio))
2960 		return 1;
2961 
2962 	/*
2963 	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2964 	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2965 	 * is supposed to work. We have way too many special cases..
2966 	 */
2967 	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2968 		return 0;
2969 
2970 	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2971 	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2972 		if (__folio_lock_killable(folio)) {
2973 			/*
2974 			 * We didn't have the right flags to drop the mmap_lock,
2975 			 * but all fault_handlers only check for fatal signals
2976 			 * if we return VM_FAULT_RETRY, so we need to drop the
2977 			 * mmap_lock here and return 0 if we don't have a fpin.
2978 			 */
2979 			if (*fpin == NULL)
2980 				mmap_read_unlock(vmf->vma->vm_mm);
2981 			return 0;
2982 		}
2983 	} else
2984 		__folio_lock(folio);
2985 
2986 	return 1;
2987 }
2988 
2989 /*
2990  * Synchronous readahead happens when we don't even find a page in the page
2991  * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2992  * to drop the mmap sem we return the file that was pinned in order for us to do
2993  * that.  If we didn't pin a file then we return NULL.  The file that is
2994  * returned needs to be fput()'ed when we're done with it.
2995  */
do_sync_mmap_readahead(struct vm_fault * vmf)2996 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2997 {
2998 	struct file *file = vmf->vma->vm_file;
2999 	struct file_ra_state *ra = &file->f_ra;
3000 	struct address_space *mapping = file->f_mapping;
3001 	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3002 	struct file *fpin = NULL;
3003 	unsigned long vm_flags = vmf->vma->vm_flags;
3004 	unsigned int mmap_miss;
3005 
3006 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3007 	/* Use the readahead code, even if readahead is disabled */
3008 	if (vm_flags & VM_HUGEPAGE) {
3009 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3010 		ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3011 		ra->size = HPAGE_PMD_NR;
3012 		/*
3013 		 * Fetch two PMD folios, so we get the chance to actually
3014 		 * readahead, unless we've been told not to.
3015 		 */
3016 		if (!(vm_flags & VM_RAND_READ))
3017 			ra->size *= 2;
3018 		ra->async_size = HPAGE_PMD_NR;
3019 		page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3020 		return fpin;
3021 	}
3022 #endif
3023 
3024 	/* If we don't want any read-ahead, don't bother */
3025 	if (vm_flags & VM_RAND_READ)
3026 		return fpin;
3027 	if (!ra->ra_pages)
3028 		return fpin;
3029 
3030 	if (vm_flags & VM_SEQ_READ) {
3031 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3032 		page_cache_sync_ra(&ractl, ra->ra_pages);
3033 		return fpin;
3034 	}
3035 
3036 	/* Avoid banging the cache line if not needed */
3037 	mmap_miss = READ_ONCE(ra->mmap_miss);
3038 	if (mmap_miss < MMAP_LOTSAMISS * 10)
3039 		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3040 
3041 	/*
3042 	 * Do we miss much more than hit in this file? If so,
3043 	 * stop bothering with read-ahead. It will only hurt.
3044 	 */
3045 	if (mmap_miss > MMAP_LOTSAMISS)
3046 		return fpin;
3047 
3048 	/*
3049 	 * mmap read-around
3050 	 */
3051 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3052 	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3053 	ra->size = ra->ra_pages;
3054 	ra->async_size = ra->ra_pages / 4;
3055 	ractl._index = ra->start;
3056 	page_cache_ra_order(&ractl, ra, 0);
3057 	return fpin;
3058 }
3059 
3060 /*
3061  * Asynchronous readahead happens when we find the page and PG_readahead,
3062  * so we want to possibly extend the readahead further.  We return the file that
3063  * was pinned if we have to drop the mmap_lock in order to do IO.
3064  */
do_async_mmap_readahead(struct vm_fault * vmf,struct folio * folio)3065 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3066 					    struct folio *folio)
3067 {
3068 	struct file *file = vmf->vma->vm_file;
3069 	struct file_ra_state *ra = &file->f_ra;
3070 	DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3071 	struct file *fpin = NULL;
3072 	unsigned int mmap_miss;
3073 
3074 	/* If we don't want any read-ahead, don't bother */
3075 	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3076 		return fpin;
3077 
3078 	mmap_miss = READ_ONCE(ra->mmap_miss);
3079 	if (mmap_miss)
3080 		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3081 
3082 	if (folio_test_readahead(folio)) {
3083 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3084 		page_cache_async_ra(&ractl, folio, ra->ra_pages);
3085 	}
3086 	return fpin;
3087 }
3088 
3089 /**
3090  * filemap_fault - read in file data for page fault handling
3091  * @vmf:	struct vm_fault containing details of the fault
3092  *
3093  * filemap_fault() is invoked via the vma operations vector for a
3094  * mapped memory region to read in file data during a page fault.
3095  *
3096  * The goto's are kind of ugly, but this streamlines the normal case of having
3097  * it in the page cache, and handles the special cases reasonably without
3098  * having a lot of duplicated code.
3099  *
3100  * vma->vm_mm->mmap_lock must be held on entry.
3101  *
3102  * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3103  * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3104  *
3105  * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3106  * has not been released.
3107  *
3108  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3109  *
3110  * Return: bitwise-OR of %VM_FAULT_ codes.
3111  */
filemap_fault(struct vm_fault * vmf)3112 vm_fault_t filemap_fault(struct vm_fault *vmf)
3113 {
3114 	int error;
3115 	struct file *file = vmf->vma->vm_file;
3116 	struct file *fpin = NULL;
3117 	struct address_space *mapping = file->f_mapping;
3118 	struct inode *inode = mapping->host;
3119 	pgoff_t max_idx, index = vmf->pgoff;
3120 	struct folio *folio;
3121 	vm_fault_t ret = 0;
3122 	bool mapping_locked = false;
3123 
3124 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3125 	if (unlikely(index >= max_idx))
3126 		return VM_FAULT_SIGBUS;
3127 
3128 	/*
3129 	 * Do we have something in the page cache already?
3130 	 */
3131 	folio = filemap_get_folio(mapping, index);
3132 	if (likely(folio)) {
3133 		/*
3134 		 * We found the page, so try async readahead before waiting for
3135 		 * the lock.
3136 		 */
3137 		if (!(vmf->flags & FAULT_FLAG_TRIED))
3138 			fpin = do_async_mmap_readahead(vmf, folio);
3139 		if (unlikely(!folio_test_uptodate(folio))) {
3140 			filemap_invalidate_lock_shared(mapping);
3141 			mapping_locked = true;
3142 		}
3143 	} else {
3144 		/* No page in the page cache at all */
3145 		count_vm_event(PGMAJFAULT);
3146 		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3147 		ret = VM_FAULT_MAJOR;
3148 		fpin = do_sync_mmap_readahead(vmf);
3149 retry_find:
3150 		/*
3151 		 * See comment in filemap_create_folio() why we need
3152 		 * invalidate_lock
3153 		 */
3154 		if (!mapping_locked) {
3155 			filemap_invalidate_lock_shared(mapping);
3156 			mapping_locked = true;
3157 		}
3158 		folio = __filemap_get_folio(mapping, index,
3159 					  FGP_CREAT|FGP_FOR_MMAP,
3160 					  vmf->gfp_mask);
3161 		if (!folio) {
3162 			if (fpin)
3163 				goto out_retry;
3164 			filemap_invalidate_unlock_shared(mapping);
3165 			return VM_FAULT_OOM;
3166 		}
3167 	}
3168 
3169 	if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3170 		goto out_retry;
3171 
3172 	/* Did it get truncated? */
3173 	if (unlikely(folio->mapping != mapping)) {
3174 		folio_unlock(folio);
3175 		folio_put(folio);
3176 		goto retry_find;
3177 	}
3178 	VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3179 
3180 	/*
3181 	 * We have a locked page in the page cache, now we need to check
3182 	 * that it's up-to-date. If not, it is going to be due to an error.
3183 	 */
3184 	if (unlikely(!folio_test_uptodate(folio))) {
3185 		/*
3186 		 * The page was in cache and uptodate and now it is not.
3187 		 * Strange but possible since we didn't hold the page lock all
3188 		 * the time. Let's drop everything get the invalidate lock and
3189 		 * try again.
3190 		 */
3191 		if (!mapping_locked) {
3192 			folio_unlock(folio);
3193 			folio_put(folio);
3194 			goto retry_find;
3195 		}
3196 		goto page_not_uptodate;
3197 	}
3198 
3199 	/*
3200 	 * We've made it this far and we had to drop our mmap_lock, now is the
3201 	 * time to return to the upper layer and have it re-find the vma and
3202 	 * redo the fault.
3203 	 */
3204 	if (fpin) {
3205 		folio_unlock(folio);
3206 		goto out_retry;
3207 	}
3208 	if (mapping_locked)
3209 		filemap_invalidate_unlock_shared(mapping);
3210 
3211 	/*
3212 	 * Found the page and have a reference on it.
3213 	 * We must recheck i_size under page lock.
3214 	 */
3215 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3216 	if (unlikely(index >= max_idx)) {
3217 		folio_unlock(folio);
3218 		folio_put(folio);
3219 		return VM_FAULT_SIGBUS;
3220 	}
3221 
3222 	vmf->page = folio_file_page(folio, index);
3223 	return ret | VM_FAULT_LOCKED;
3224 
3225 page_not_uptodate:
3226 	/*
3227 	 * Umm, take care of errors if the page isn't up-to-date.
3228 	 * Try to re-read it _once_. We do this synchronously,
3229 	 * because there really aren't any performance issues here
3230 	 * and we need to check for errors.
3231 	 */
3232 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3233 	error = filemap_read_folio(file, mapping, folio);
3234 	if (fpin)
3235 		goto out_retry;
3236 	folio_put(folio);
3237 
3238 	if (!error || error == AOP_TRUNCATED_PAGE)
3239 		goto retry_find;
3240 	filemap_invalidate_unlock_shared(mapping);
3241 
3242 	return VM_FAULT_SIGBUS;
3243 
3244 out_retry:
3245 	/*
3246 	 * We dropped the mmap_lock, we need to return to the fault handler to
3247 	 * re-find the vma and come back and find our hopefully still populated
3248 	 * page.
3249 	 */
3250 	if (folio)
3251 		folio_put(folio);
3252 	if (mapping_locked)
3253 		filemap_invalidate_unlock_shared(mapping);
3254 	if (fpin)
3255 		fput(fpin);
3256 	return ret | VM_FAULT_RETRY;
3257 }
3258 EXPORT_SYMBOL(filemap_fault);
3259 
filemap_map_pmd(struct vm_fault * vmf,struct page * page)3260 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3261 {
3262 	struct mm_struct *mm = vmf->vma->vm_mm;
3263 
3264 	/* Huge page is mapped? No need to proceed. */
3265 	if (pmd_trans_huge(*vmf->pmd)) {
3266 		unlock_page(page);
3267 		put_page(page);
3268 		return true;
3269 	}
3270 
3271 	if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3272 		vm_fault_t ret = do_set_pmd(vmf, page);
3273 		if (!ret) {
3274 			/* The page is mapped successfully, reference consumed. */
3275 			unlock_page(page);
3276 			return true;
3277 		}
3278 	}
3279 
3280 	if (pmd_none(*vmf->pmd))
3281 		pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3282 
3283 	/* See comment in handle_pte_fault() */
3284 	if (pmd_devmap_trans_unstable(vmf->pmd)) {
3285 		unlock_page(page);
3286 		put_page(page);
3287 		return true;
3288 	}
3289 
3290 	return false;
3291 }
3292 
next_uptodate_page(struct folio * folio,struct address_space * mapping,struct xa_state * xas,pgoff_t end_pgoff)3293 static struct folio *next_uptodate_page(struct folio *folio,
3294 				       struct address_space *mapping,
3295 				       struct xa_state *xas, pgoff_t end_pgoff)
3296 {
3297 	unsigned long max_idx;
3298 
3299 	do {
3300 		if (!folio)
3301 			return NULL;
3302 		if (xas_retry(xas, folio))
3303 			continue;
3304 		if (xa_is_value(folio))
3305 			continue;
3306 		if (folio_test_locked(folio))
3307 			continue;
3308 		if (!folio_try_get_rcu(folio))
3309 			continue;
3310 		/* Has the page moved or been split? */
3311 		if (unlikely(folio != xas_reload(xas)))
3312 			goto skip;
3313 		if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3314 			goto skip;
3315 		if (!folio_trylock(folio))
3316 			goto skip;
3317 		if (folio->mapping != mapping)
3318 			goto unlock;
3319 		if (!folio_test_uptodate(folio))
3320 			goto unlock;
3321 		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3322 		if (xas->xa_index >= max_idx)
3323 			goto unlock;
3324 		return folio;
3325 unlock:
3326 		folio_unlock(folio);
3327 skip:
3328 		folio_put(folio);
3329 	} while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3330 
3331 	return NULL;
3332 }
3333 
first_map_page(struct address_space * mapping,struct xa_state * xas,pgoff_t end_pgoff)3334 static inline struct folio *first_map_page(struct address_space *mapping,
3335 					  struct xa_state *xas,
3336 					  pgoff_t end_pgoff)
3337 {
3338 	return next_uptodate_page(xas_find(xas, end_pgoff),
3339 				  mapping, xas, end_pgoff);
3340 }
3341 
next_map_page(struct address_space * mapping,struct xa_state * xas,pgoff_t end_pgoff)3342 static inline struct folio *next_map_page(struct address_space *mapping,
3343 					 struct xa_state *xas,
3344 					 pgoff_t end_pgoff)
3345 {
3346 	return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3347 				  mapping, xas, end_pgoff);
3348 }
3349 
filemap_map_pages(struct vm_fault * vmf,pgoff_t start_pgoff,pgoff_t end_pgoff)3350 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3351 			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3352 {
3353 	struct vm_area_struct *vma = vmf->vma;
3354 	struct file *file = vma->vm_file;
3355 	struct address_space *mapping = file->f_mapping;
3356 	pgoff_t last_pgoff = start_pgoff;
3357 	unsigned long addr;
3358 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3359 	struct folio *folio;
3360 	struct page *page;
3361 	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3362 	vm_fault_t ret = 0;
3363 
3364 	rcu_read_lock();
3365 	folio = first_map_page(mapping, &xas, end_pgoff);
3366 	if (!folio)
3367 		goto out;
3368 
3369 	if (filemap_map_pmd(vmf, &folio->page)) {
3370 		ret = VM_FAULT_NOPAGE;
3371 		goto out;
3372 	}
3373 
3374 	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3375 	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3376 	do {
3377 again:
3378 		page = folio_file_page(folio, xas.xa_index);
3379 		if (PageHWPoison(page))
3380 			goto unlock;
3381 
3382 		if (mmap_miss > 0)
3383 			mmap_miss--;
3384 
3385 		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3386 		vmf->pte += xas.xa_index - last_pgoff;
3387 		last_pgoff = xas.xa_index;
3388 
3389 		/*
3390 		 * NOTE: If there're PTE markers, we'll leave them to be
3391 		 * handled in the specific fault path, and it'll prohibit the
3392 		 * fault-around logic.
3393 		 */
3394 		if (!pte_none(*vmf->pte))
3395 			goto unlock;
3396 
3397 		/* We're about to handle the fault */
3398 		if (vmf->address == addr)
3399 			ret = VM_FAULT_NOPAGE;
3400 
3401 		do_set_pte(vmf, page, addr);
3402 		/* no need to invalidate: a not-present page won't be cached */
3403 		update_mmu_cache(vma, addr, vmf->pte);
3404 		if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3405 			xas.xa_index++;
3406 			folio_ref_inc(folio);
3407 			goto again;
3408 		}
3409 		folio_unlock(folio);
3410 		continue;
3411 unlock:
3412 		if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3413 			xas.xa_index++;
3414 			goto again;
3415 		}
3416 		folio_unlock(folio);
3417 		folio_put(folio);
3418 	} while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3419 	pte_unmap_unlock(vmf->pte, vmf->ptl);
3420 out:
3421 	rcu_read_unlock();
3422 	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3423 	return ret;
3424 }
3425 EXPORT_SYMBOL(filemap_map_pages);
3426 
filemap_page_mkwrite(struct vm_fault * vmf)3427 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3428 {
3429 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3430 	struct folio *folio = page_folio(vmf->page);
3431 	vm_fault_t ret = VM_FAULT_LOCKED;
3432 
3433 	sb_start_pagefault(mapping->host->i_sb);
3434 	file_update_time(vmf->vma->vm_file);
3435 	folio_lock(folio);
3436 	if (folio->mapping != mapping) {
3437 		folio_unlock(folio);
3438 		ret = VM_FAULT_NOPAGE;
3439 		goto out;
3440 	}
3441 	/*
3442 	 * We mark the folio dirty already here so that when freeze is in
3443 	 * progress, we are guaranteed that writeback during freezing will
3444 	 * see the dirty folio and writeprotect it again.
3445 	 */
3446 	folio_mark_dirty(folio);
3447 	folio_wait_stable(folio);
3448 out:
3449 	sb_end_pagefault(mapping->host->i_sb);
3450 	return ret;
3451 }
3452 
3453 const struct vm_operations_struct generic_file_vm_ops = {
3454 	.fault		= filemap_fault,
3455 	.map_pages	= filemap_map_pages,
3456 	.page_mkwrite	= filemap_page_mkwrite,
3457 };
3458 
3459 /* This is used for a general mmap of a disk file */
3460 
generic_file_mmap(struct file * file,struct vm_area_struct * vma)3461 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3462 {
3463 	struct address_space *mapping = file->f_mapping;
3464 
3465 	if (!mapping->a_ops->read_folio)
3466 		return -ENOEXEC;
3467 	file_accessed(file);
3468 	vma->vm_ops = &generic_file_vm_ops;
3469 	return 0;
3470 }
3471 
3472 /*
3473  * This is for filesystems which do not implement ->writepage.
3474  */
generic_file_readonly_mmap(struct file * file,struct vm_area_struct * vma)3475 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3476 {
3477 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3478 		return -EINVAL;
3479 	return generic_file_mmap(file, vma);
3480 }
3481 #else
filemap_page_mkwrite(struct vm_fault * vmf)3482 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3483 {
3484 	return VM_FAULT_SIGBUS;
3485 }
generic_file_mmap(struct file * file,struct vm_area_struct * vma)3486 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3487 {
3488 	return -ENOSYS;
3489 }
generic_file_readonly_mmap(struct file * file,struct vm_area_struct * vma)3490 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3491 {
3492 	return -ENOSYS;
3493 }
3494 #endif /* CONFIG_MMU */
3495 
3496 EXPORT_SYMBOL(filemap_page_mkwrite);
3497 EXPORT_SYMBOL(generic_file_mmap);
3498 EXPORT_SYMBOL(generic_file_readonly_mmap);
3499 
do_read_cache_folio(struct address_space * mapping,pgoff_t index,filler_t filler,struct file * file,gfp_t gfp)3500 static struct folio *do_read_cache_folio(struct address_space *mapping,
3501 		pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3502 {
3503 	struct folio *folio;
3504 	int err;
3505 
3506 	if (!filler)
3507 		filler = mapping->a_ops->read_folio;
3508 repeat:
3509 	folio = filemap_get_folio(mapping, index);
3510 	if (!folio) {
3511 		folio = filemap_alloc_folio(gfp, 0);
3512 		if (!folio)
3513 			return ERR_PTR(-ENOMEM);
3514 		err = filemap_add_folio(mapping, folio, index, gfp);
3515 		if (unlikely(err)) {
3516 			folio_put(folio);
3517 			if (err == -EEXIST)
3518 				goto repeat;
3519 			/* Presumably ENOMEM for xarray node */
3520 			return ERR_PTR(err);
3521 		}
3522 
3523 filler:
3524 		err = filler(file, folio);
3525 		if (err < 0) {
3526 			folio_put(folio);
3527 			return ERR_PTR(err);
3528 		}
3529 
3530 		folio_wait_locked(folio);
3531 		if (!folio_test_uptodate(folio)) {
3532 			folio_put(folio);
3533 			return ERR_PTR(-EIO);
3534 		}
3535 
3536 		goto out;
3537 	}
3538 	if (folio_test_uptodate(folio))
3539 		goto out;
3540 
3541 	if (!folio_trylock(folio)) {
3542 		folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3543 		goto repeat;
3544 	}
3545 
3546 	/* Folio was truncated from mapping */
3547 	if (!folio->mapping) {
3548 		folio_unlock(folio);
3549 		folio_put(folio);
3550 		goto repeat;
3551 	}
3552 
3553 	/* Someone else locked and filled the page in a very small window */
3554 	if (folio_test_uptodate(folio)) {
3555 		folio_unlock(folio);
3556 		goto out;
3557 	}
3558 
3559 	/*
3560 	 * A previous I/O error may have been due to temporary
3561 	 * failures.
3562 	 * Clear page error before actual read, PG_error will be
3563 	 * set again if read page fails.
3564 	 */
3565 	folio_clear_error(folio);
3566 	goto filler;
3567 
3568 out:
3569 	folio_mark_accessed(folio);
3570 	return folio;
3571 }
3572 
3573 /**
3574  * read_cache_folio - Read into page cache, fill it if needed.
3575  * @mapping: The address_space to read from.
3576  * @index: The index to read.
3577  * @filler: Function to perform the read, or NULL to use aops->read_folio().
3578  * @file: Passed to filler function, may be NULL if not required.
3579  *
3580  * Read one page into the page cache.  If it succeeds, the folio returned
3581  * will contain @index, but it may not be the first page of the folio.
3582  *
3583  * If the filler function returns an error, it will be returned to the
3584  * caller.
3585  *
3586  * Context: May sleep.  Expects mapping->invalidate_lock to be held.
3587  * Return: An uptodate folio on success, ERR_PTR() on failure.
3588  */
read_cache_folio(struct address_space * mapping,pgoff_t index,filler_t filler,struct file * file)3589 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3590 		filler_t filler, struct file *file)
3591 {
3592 	return do_read_cache_folio(mapping, index, filler, file,
3593 			mapping_gfp_mask(mapping));
3594 }
3595 EXPORT_SYMBOL(read_cache_folio);
3596 
do_read_cache_page(struct address_space * mapping,pgoff_t index,filler_t * filler,struct file * file,gfp_t gfp)3597 static struct page *do_read_cache_page(struct address_space *mapping,
3598 		pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3599 {
3600 	struct folio *folio;
3601 
3602 	folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3603 	if (IS_ERR(folio))
3604 		return &folio->page;
3605 	return folio_file_page(folio, index);
3606 }
3607 
read_cache_page(struct address_space * mapping,pgoff_t index,filler_t * filler,struct file * file)3608 struct page *read_cache_page(struct address_space *mapping,
3609 			pgoff_t index, filler_t *filler, struct file *file)
3610 {
3611 	return do_read_cache_page(mapping, index, filler, file,
3612 			mapping_gfp_mask(mapping));
3613 }
3614 EXPORT_SYMBOL(read_cache_page);
3615 
3616 /**
3617  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3618  * @mapping:	the page's address_space
3619  * @index:	the page index
3620  * @gfp:	the page allocator flags to use if allocating
3621  *
3622  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3623  * any new page allocations done using the specified allocation flags.
3624  *
3625  * If the page does not get brought uptodate, return -EIO.
3626  *
3627  * The function expects mapping->invalidate_lock to be already held.
3628  *
3629  * Return: up to date page on success, ERR_PTR() on failure.
3630  */
read_cache_page_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)3631 struct page *read_cache_page_gfp(struct address_space *mapping,
3632 				pgoff_t index,
3633 				gfp_t gfp)
3634 {
3635 	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3636 }
3637 EXPORT_SYMBOL(read_cache_page_gfp);
3638 
3639 /*
3640  * Warn about a page cache invalidation failure during a direct I/O write.
3641  */
dio_warn_stale_pagecache(struct file * filp)3642 void dio_warn_stale_pagecache(struct file *filp)
3643 {
3644 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3645 	char pathname[128];
3646 	char *path;
3647 
3648 	errseq_set(&filp->f_mapping->wb_err, -EIO);
3649 	if (__ratelimit(&_rs)) {
3650 		path = file_path(filp, pathname, sizeof(pathname));
3651 		if (IS_ERR(path))
3652 			path = "(unknown)";
3653 		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3654 		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3655 			current->comm);
3656 	}
3657 }
3658 
3659 ssize_t
generic_file_direct_write(struct kiocb * iocb,struct iov_iter * from)3660 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3661 {
3662 	struct file	*file = iocb->ki_filp;
3663 	struct address_space *mapping = file->f_mapping;
3664 	struct inode	*inode = mapping->host;
3665 	loff_t		pos = iocb->ki_pos;
3666 	ssize_t		written;
3667 	size_t		write_len;
3668 	pgoff_t		end;
3669 
3670 	write_len = iov_iter_count(from);
3671 	end = (pos + write_len - 1) >> PAGE_SHIFT;
3672 
3673 	if (iocb->ki_flags & IOCB_NOWAIT) {
3674 		/* If there are pages to writeback, return */
3675 		if (filemap_range_has_page(file->f_mapping, pos,
3676 					   pos + write_len - 1))
3677 			return -EAGAIN;
3678 	} else {
3679 		written = filemap_write_and_wait_range(mapping, pos,
3680 							pos + write_len - 1);
3681 		if (written)
3682 			goto out;
3683 	}
3684 
3685 	/*
3686 	 * After a write we want buffered reads to be sure to go to disk to get
3687 	 * the new data.  We invalidate clean cached page from the region we're
3688 	 * about to write.  We do this *before* the write so that we can return
3689 	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3690 	 */
3691 	written = invalidate_inode_pages2_range(mapping,
3692 					pos >> PAGE_SHIFT, end);
3693 	/*
3694 	 * If a page can not be invalidated, return 0 to fall back
3695 	 * to buffered write.
3696 	 */
3697 	if (written) {
3698 		if (written == -EBUSY)
3699 			return 0;
3700 		goto out;
3701 	}
3702 
3703 	written = mapping->a_ops->direct_IO(iocb, from);
3704 
3705 	/*
3706 	 * Finally, try again to invalidate clean pages which might have been
3707 	 * cached by non-direct readahead, or faulted in by get_user_pages()
3708 	 * if the source of the write was an mmap'ed region of the file
3709 	 * we're writing.  Either one is a pretty crazy thing to do,
3710 	 * so we don't support it 100%.  If this invalidation
3711 	 * fails, tough, the write still worked...
3712 	 *
3713 	 * Most of the time we do not need this since dio_complete() will do
3714 	 * the invalidation for us. However there are some file systems that
3715 	 * do not end up with dio_complete() being called, so let's not break
3716 	 * them by removing it completely.
3717 	 *
3718 	 * Noticeable example is a blkdev_direct_IO().
3719 	 *
3720 	 * Skip invalidation for async writes or if mapping has no pages.
3721 	 */
3722 	if (written > 0 && mapping->nrpages &&
3723 	    invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3724 		dio_warn_stale_pagecache(file);
3725 
3726 	if (written > 0) {
3727 		pos += written;
3728 		write_len -= written;
3729 		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3730 			i_size_write(inode, pos);
3731 			mark_inode_dirty(inode);
3732 		}
3733 		iocb->ki_pos = pos;
3734 	}
3735 	if (written != -EIOCBQUEUED)
3736 		iov_iter_revert(from, write_len - iov_iter_count(from));
3737 out:
3738 	return written;
3739 }
3740 EXPORT_SYMBOL(generic_file_direct_write);
3741 
generic_perform_write(struct kiocb * iocb,struct iov_iter * i)3742 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3743 {
3744 	struct file *file = iocb->ki_filp;
3745 	loff_t pos = iocb->ki_pos;
3746 	struct address_space *mapping = file->f_mapping;
3747 	const struct address_space_operations *a_ops = mapping->a_ops;
3748 	long status = 0;
3749 	ssize_t written = 0;
3750 
3751 	do {
3752 		struct page *page;
3753 		unsigned long offset;	/* Offset into pagecache page */
3754 		unsigned long bytes;	/* Bytes to write to page */
3755 		size_t copied;		/* Bytes copied from user */
3756 		void *fsdata;
3757 
3758 		offset = (pos & (PAGE_SIZE - 1));
3759 		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3760 						iov_iter_count(i));
3761 
3762 again:
3763 		/*
3764 		 * Bring in the user page that we will copy from _first_.
3765 		 * Otherwise there's a nasty deadlock on copying from the
3766 		 * same page as we're writing to, without it being marked
3767 		 * up-to-date.
3768 		 */
3769 		if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3770 			status = -EFAULT;
3771 			break;
3772 		}
3773 
3774 		if (fatal_signal_pending(current)) {
3775 			status = -EINTR;
3776 			break;
3777 		}
3778 
3779 		status = a_ops->write_begin(file, mapping, pos, bytes,
3780 						&page, &fsdata);
3781 		if (unlikely(status < 0))
3782 			break;
3783 
3784 		if (mapping_writably_mapped(mapping))
3785 			flush_dcache_page(page);
3786 
3787 		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3788 		flush_dcache_page(page);
3789 
3790 		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3791 						page, fsdata);
3792 		if (unlikely(status != copied)) {
3793 			iov_iter_revert(i, copied - max(status, 0L));
3794 			if (unlikely(status < 0))
3795 				break;
3796 		}
3797 		cond_resched();
3798 
3799 		if (unlikely(status == 0)) {
3800 			/*
3801 			 * A short copy made ->write_end() reject the
3802 			 * thing entirely.  Might be memory poisoning
3803 			 * halfway through, might be a race with munmap,
3804 			 * might be severe memory pressure.
3805 			 */
3806 			if (copied)
3807 				bytes = copied;
3808 			goto again;
3809 		}
3810 		pos += status;
3811 		written += status;
3812 
3813 		balance_dirty_pages_ratelimited(mapping);
3814 	} while (iov_iter_count(i));
3815 
3816 	return written ? written : status;
3817 }
3818 EXPORT_SYMBOL(generic_perform_write);
3819 
3820 /**
3821  * __generic_file_write_iter - write data to a file
3822  * @iocb:	IO state structure (file, offset, etc.)
3823  * @from:	iov_iter with data to write
3824  *
3825  * This function does all the work needed for actually writing data to a
3826  * file. It does all basic checks, removes SUID from the file, updates
3827  * modification times and calls proper subroutines depending on whether we
3828  * do direct IO or a standard buffered write.
3829  *
3830  * It expects i_rwsem to be grabbed unless we work on a block device or similar
3831  * object which does not need locking at all.
3832  *
3833  * This function does *not* take care of syncing data in case of O_SYNC write.
3834  * A caller has to handle it. This is mainly due to the fact that we want to
3835  * avoid syncing under i_rwsem.
3836  *
3837  * Return:
3838  * * number of bytes written, even for truncated writes
3839  * * negative error code if no data has been written at all
3840  */
__generic_file_write_iter(struct kiocb * iocb,struct iov_iter * from)3841 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3842 {
3843 	struct file *file = iocb->ki_filp;
3844 	struct address_space *mapping = file->f_mapping;
3845 	struct inode 	*inode = mapping->host;
3846 	ssize_t		written = 0;
3847 	ssize_t		err;
3848 	ssize_t		status;
3849 
3850 	/* We can write back this queue in page reclaim */
3851 	current->backing_dev_info = inode_to_bdi(inode);
3852 	err = file_remove_privs(file);
3853 	if (err)
3854 		goto out;
3855 
3856 	err = file_update_time(file);
3857 	if (err)
3858 		goto out;
3859 
3860 	if (iocb->ki_flags & IOCB_DIRECT) {
3861 		loff_t pos, endbyte;
3862 
3863 		written = generic_file_direct_write(iocb, from);
3864 		/*
3865 		 * If the write stopped short of completing, fall back to
3866 		 * buffered writes.  Some filesystems do this for writes to
3867 		 * holes, for example.  For DAX files, a buffered write will
3868 		 * not succeed (even if it did, DAX does not handle dirty
3869 		 * page-cache pages correctly).
3870 		 */
3871 		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3872 			goto out;
3873 
3874 		pos = iocb->ki_pos;
3875 		status = generic_perform_write(iocb, from);
3876 		/*
3877 		 * If generic_perform_write() returned a synchronous error
3878 		 * then we want to return the number of bytes which were
3879 		 * direct-written, or the error code if that was zero.  Note
3880 		 * that this differs from normal direct-io semantics, which
3881 		 * will return -EFOO even if some bytes were written.
3882 		 */
3883 		if (unlikely(status < 0)) {
3884 			err = status;
3885 			goto out;
3886 		}
3887 		/*
3888 		 * We need to ensure that the page cache pages are written to
3889 		 * disk and invalidated to preserve the expected O_DIRECT
3890 		 * semantics.
3891 		 */
3892 		endbyte = pos + status - 1;
3893 		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3894 		if (err == 0) {
3895 			iocb->ki_pos = endbyte + 1;
3896 			written += status;
3897 			invalidate_mapping_pages(mapping,
3898 						 pos >> PAGE_SHIFT,
3899 						 endbyte >> PAGE_SHIFT);
3900 		} else {
3901 			/*
3902 			 * We don't know how much we wrote, so just return
3903 			 * the number of bytes which were direct-written
3904 			 */
3905 		}
3906 	} else {
3907 		written = generic_perform_write(iocb, from);
3908 		if (likely(written > 0))
3909 			iocb->ki_pos += written;
3910 	}
3911 out:
3912 	current->backing_dev_info = NULL;
3913 	return written ? written : err;
3914 }
3915 EXPORT_SYMBOL(__generic_file_write_iter);
3916 
3917 /**
3918  * generic_file_write_iter - write data to a file
3919  * @iocb:	IO state structure
3920  * @from:	iov_iter with data to write
3921  *
3922  * This is a wrapper around __generic_file_write_iter() to be used by most
3923  * filesystems. It takes care of syncing the file in case of O_SYNC file
3924  * and acquires i_rwsem as needed.
3925  * Return:
3926  * * negative error code if no data has been written at all of
3927  *   vfs_fsync_range() failed for a synchronous write
3928  * * number of bytes written, even for truncated writes
3929  */
generic_file_write_iter(struct kiocb * iocb,struct iov_iter * from)3930 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3931 {
3932 	struct file *file = iocb->ki_filp;
3933 	struct inode *inode = file->f_mapping->host;
3934 	ssize_t ret;
3935 
3936 	inode_lock(inode);
3937 	ret = generic_write_checks(iocb, from);
3938 	if (ret > 0)
3939 		ret = __generic_file_write_iter(iocb, from);
3940 	inode_unlock(inode);
3941 
3942 	if (ret > 0)
3943 		ret = generic_write_sync(iocb, ret);
3944 	return ret;
3945 }
3946 EXPORT_SYMBOL(generic_file_write_iter);
3947 
3948 /**
3949  * filemap_release_folio() - Release fs-specific metadata on a folio.
3950  * @folio: The folio which the kernel is trying to free.
3951  * @gfp: Memory allocation flags (and I/O mode).
3952  *
3953  * The address_space is trying to release any data attached to a folio
3954  * (presumably at folio->private).
3955  *
3956  * This will also be called if the private_2 flag is set on a page,
3957  * indicating that the folio has other metadata associated with it.
3958  *
3959  * The @gfp argument specifies whether I/O may be performed to release
3960  * this page (__GFP_IO), and whether the call may block
3961  * (__GFP_RECLAIM & __GFP_FS).
3962  *
3963  * Return: %true if the release was successful, otherwise %false.
3964  */
filemap_release_folio(struct folio * folio,gfp_t gfp)3965 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
3966 {
3967 	struct address_space * const mapping = folio->mapping;
3968 
3969 	BUG_ON(!folio_test_locked(folio));
3970 	if (folio_test_writeback(folio))
3971 		return false;
3972 
3973 	if (mapping && mapping->a_ops->release_folio)
3974 		return mapping->a_ops->release_folio(folio, gfp);
3975 	return try_to_free_buffers(folio);
3976 }
3977 EXPORT_SYMBOL(filemap_release_folio);
3978