1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	linux/mm/filemap.c
4  *
5  * Copyright (C) 1994-1999  Linus Torvalds
6  */
7 
8 /*
9  * This file handles the generic file mmap semantics used by
10  * most "normal" filesystems (but you don't /have/ to use this:
11  * the NFS filesystem used to do this differently, for example)
12  */
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
16 #include <linux/fs.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
22 #include <linux/mm.h>
23 #include <linux/swap.h>
24 #include <linux/swapops.h>
25 #include <linux/syscalls.h>
26 #include <linux/mman.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/uio.h>
30 #include <linux/error-injection.h>
31 #include <linux/hash.h>
32 #include <linux/writeback.h>
33 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h>
35 #include <linux/security.h>
36 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include <linux/migrate.h>
46 #include <linux/pipe_fs_i.h>
47 #include <linux/splice.h>
48 #include <asm/pgalloc.h>
49 #include <asm/tlbflush.h>
50 #include "internal.h"
51 
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/filemap.h>
54 
55 /*
56  * FIXME: remove all knowledge of the buffer layer from the core VM
57  */
58 #include <linux/buffer_head.h> /* for try_to_free_buffers */
59 
60 #include <asm/mman.h>
61 
62 #include "swap.h"
63 
64 /*
65  * Shared mappings implemented 30.11.1994. It's not fully working yet,
66  * though.
67  *
68  * Shared mappings now work. 15.8.1995  Bruno.
69  *
70  * finished 'unifying' the page and buffer cache and SMP-threaded the
71  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
72  *
73  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74  */
75 
76 /*
77  * Lock ordering:
78  *
79  *  ->i_mmap_rwsem		(truncate_pagecache)
80  *    ->private_lock		(__free_pte->block_dirty_folio)
81  *      ->swap_lock		(exclusive_swap_page, others)
82  *        ->i_pages lock
83  *
84  *  ->i_rwsem
85  *    ->invalidate_lock		(acquired by fs in truncate path)
86  *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
87  *
88  *  ->mmap_lock
89  *    ->i_mmap_rwsem
90  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
91  *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
92  *
93  *  ->mmap_lock
94  *    ->invalidate_lock		(filemap_fault)
95  *      ->lock_page		(filemap_fault, access_process_vm)
96  *
97  *  ->i_rwsem			(generic_perform_write)
98  *    ->mmap_lock		(fault_in_readable->do_page_fault)
99  *
100  *  bdi->wb.list_lock
101  *    sb_lock			(fs/fs-writeback.c)
102  *    ->i_pages lock		(__sync_single_inode)
103  *
104  *  ->i_mmap_rwsem
105  *    ->anon_vma.lock		(vma_merge)
106  *
107  *  ->anon_vma.lock
108  *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
109  *
110  *  ->page_table_lock or pte_lock
111  *    ->swap_lock		(try_to_unmap_one)
112  *    ->private_lock		(try_to_unmap_one)
113  *    ->i_pages lock		(try_to_unmap_one)
114  *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
115  *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
116  *    ->private_lock		(page_remove_rmap->set_page_dirty)
117  *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
118  *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
119  *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
120  *    ->memcg->move_lock	(page_remove_rmap->folio_memcg_lock)
121  *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
122  *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
123  *    ->private_lock		(zap_pte_range->block_dirty_folio)
124  */
125 
page_cache_delete(struct address_space * mapping,struct folio * folio,void * shadow)126 static void page_cache_delete(struct address_space *mapping,
127 				   struct folio *folio, void *shadow)
128 {
129 	XA_STATE(xas, &mapping->i_pages, folio->index);
130 	long nr = 1;
131 
132 	mapping_set_update(&xas, mapping);
133 
134 	/* hugetlb pages are represented by a single entry in the xarray */
135 	if (!folio_test_hugetlb(folio)) {
136 		xas_set_order(&xas, folio->index, folio_order(folio));
137 		nr = folio_nr_pages(folio);
138 	}
139 
140 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
141 
142 	xas_store(&xas, shadow);
143 	xas_init_marks(&xas);
144 
145 	folio->mapping = NULL;
146 	/* Leave page->index set: truncation lookup relies upon it */
147 	mapping->nrpages -= nr;
148 }
149 
filemap_unaccount_folio(struct address_space * mapping,struct folio * folio)150 static void filemap_unaccount_folio(struct address_space *mapping,
151 		struct folio *folio)
152 {
153 	long nr;
154 
155 	VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
156 	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
157 		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
158 			 current->comm, folio_pfn(folio));
159 		dump_page(&folio->page, "still mapped when deleted");
160 		dump_stack();
161 		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
162 
163 		if (mapping_exiting(mapping) && !folio_test_large(folio)) {
164 			int mapcount = page_mapcount(&folio->page);
165 
166 			if (folio_ref_count(folio) >= mapcount + 2) {
167 				/*
168 				 * All vmas have already been torn down, so it's
169 				 * a good bet that actually the page is unmapped
170 				 * and we'd rather not leak it: if we're wrong,
171 				 * another bad page check should catch it later.
172 				 */
173 				page_mapcount_reset(&folio->page);
174 				folio_ref_sub(folio, mapcount);
175 			}
176 		}
177 	}
178 
179 	/* hugetlb folios do not participate in page cache accounting. */
180 	if (folio_test_hugetlb(folio))
181 		return;
182 
183 	nr = folio_nr_pages(folio);
184 
185 	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
186 	if (folio_test_swapbacked(folio)) {
187 		__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
188 		if (folio_test_pmd_mappable(folio))
189 			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
190 	} else if (folio_test_pmd_mappable(folio)) {
191 		__lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
192 		filemap_nr_thps_dec(mapping);
193 	}
194 
195 	/*
196 	 * At this point folio must be either written or cleaned by
197 	 * truncate.  Dirty folio here signals a bug and loss of
198 	 * unwritten data - on ordinary filesystems.
199 	 *
200 	 * But it's harmless on in-memory filesystems like tmpfs; and can
201 	 * occur when a driver which did get_user_pages() sets page dirty
202 	 * before putting it, while the inode is being finally evicted.
203 	 *
204 	 * Below fixes dirty accounting after removing the folio entirely
205 	 * but leaves the dirty flag set: it has no effect for truncated
206 	 * folio and anyway will be cleared before returning folio to
207 	 * buddy allocator.
208 	 */
209 	if (WARN_ON_ONCE(folio_test_dirty(folio) &&
210 			 mapping_can_writeback(mapping)))
211 		folio_account_cleaned(folio, inode_to_wb(mapping->host));
212 }
213 
214 /*
215  * Delete a page from the page cache and free it. Caller has to make
216  * sure the page is locked and that nobody else uses it - or that usage
217  * is safe.  The caller must hold the i_pages lock.
218  */
__filemap_remove_folio(struct folio * folio,void * shadow)219 void __filemap_remove_folio(struct folio *folio, void *shadow)
220 {
221 	struct address_space *mapping = folio->mapping;
222 
223 	trace_mm_filemap_delete_from_page_cache(folio);
224 	filemap_unaccount_folio(mapping, folio);
225 	page_cache_delete(mapping, folio, shadow);
226 }
227 
filemap_free_folio(struct address_space * mapping,struct folio * folio)228 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
229 {
230 	void (*free_folio)(struct folio *);
231 	int refs = 1;
232 
233 	free_folio = mapping->a_ops->free_folio;
234 	if (free_folio)
235 		free_folio(folio);
236 
237 	if (folio_test_large(folio) && !folio_test_hugetlb(folio))
238 		refs = folio_nr_pages(folio);
239 	folio_put_refs(folio, refs);
240 }
241 
242 /**
243  * filemap_remove_folio - Remove folio from page cache.
244  * @folio: The folio.
245  *
246  * This must be called only on folios that are locked and have been
247  * verified to be in the page cache.  It will never put the folio into
248  * the free list because the caller has a reference on the page.
249  */
filemap_remove_folio(struct folio * folio)250 void filemap_remove_folio(struct folio *folio)
251 {
252 	struct address_space *mapping = folio->mapping;
253 
254 	BUG_ON(!folio_test_locked(folio));
255 	spin_lock(&mapping->host->i_lock);
256 	xa_lock_irq(&mapping->i_pages);
257 	__filemap_remove_folio(folio, NULL);
258 	xa_unlock_irq(&mapping->i_pages);
259 	if (mapping_shrinkable(mapping))
260 		inode_add_lru(mapping->host);
261 	spin_unlock(&mapping->host->i_lock);
262 
263 	filemap_free_folio(mapping, folio);
264 }
265 
266 /*
267  * page_cache_delete_batch - delete several folios from page cache
268  * @mapping: the mapping to which folios belong
269  * @fbatch: batch of folios to delete
270  *
271  * The function walks over mapping->i_pages and removes folios passed in
272  * @fbatch from the mapping. The function expects @fbatch to be sorted
273  * by page index and is optimised for it to be dense.
274  * It tolerates holes in @fbatch (mapping entries at those indices are not
275  * modified).
276  *
277  * The function expects the i_pages lock to be held.
278  */
page_cache_delete_batch(struct address_space * mapping,struct folio_batch * fbatch)279 static void page_cache_delete_batch(struct address_space *mapping,
280 			     struct folio_batch *fbatch)
281 {
282 	XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
283 	long total_pages = 0;
284 	int i = 0;
285 	struct folio *folio;
286 
287 	mapping_set_update(&xas, mapping);
288 	xas_for_each(&xas, folio, ULONG_MAX) {
289 		if (i >= folio_batch_count(fbatch))
290 			break;
291 
292 		/* A swap/dax/shadow entry got inserted? Skip it. */
293 		if (xa_is_value(folio))
294 			continue;
295 		/*
296 		 * A page got inserted in our range? Skip it. We have our
297 		 * pages locked so they are protected from being removed.
298 		 * If we see a page whose index is higher than ours, it
299 		 * means our page has been removed, which shouldn't be
300 		 * possible because we're holding the PageLock.
301 		 */
302 		if (folio != fbatch->folios[i]) {
303 			VM_BUG_ON_FOLIO(folio->index >
304 					fbatch->folios[i]->index, folio);
305 			continue;
306 		}
307 
308 		WARN_ON_ONCE(!folio_test_locked(folio));
309 
310 		folio->mapping = NULL;
311 		/* Leave folio->index set: truncation lookup relies on it */
312 
313 		i++;
314 		xas_store(&xas, NULL);
315 		total_pages += folio_nr_pages(folio);
316 	}
317 	mapping->nrpages -= total_pages;
318 }
319 
delete_from_page_cache_batch(struct address_space * mapping,struct folio_batch * fbatch)320 void delete_from_page_cache_batch(struct address_space *mapping,
321 				  struct folio_batch *fbatch)
322 {
323 	int i;
324 
325 	if (!folio_batch_count(fbatch))
326 		return;
327 
328 	spin_lock(&mapping->host->i_lock);
329 	xa_lock_irq(&mapping->i_pages);
330 	for (i = 0; i < folio_batch_count(fbatch); i++) {
331 		struct folio *folio = fbatch->folios[i];
332 
333 		trace_mm_filemap_delete_from_page_cache(folio);
334 		filemap_unaccount_folio(mapping, folio);
335 	}
336 	page_cache_delete_batch(mapping, fbatch);
337 	xa_unlock_irq(&mapping->i_pages);
338 	if (mapping_shrinkable(mapping))
339 		inode_add_lru(mapping->host);
340 	spin_unlock(&mapping->host->i_lock);
341 
342 	for (i = 0; i < folio_batch_count(fbatch); i++)
343 		filemap_free_folio(mapping, fbatch->folios[i]);
344 }
345 
filemap_check_errors(struct address_space * mapping)346 int filemap_check_errors(struct address_space *mapping)
347 {
348 	int ret = 0;
349 	/* Check for outstanding write errors */
350 	if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
352 		ret = -ENOSPC;
353 	if (test_bit(AS_EIO, &mapping->flags) &&
354 	    test_and_clear_bit(AS_EIO, &mapping->flags))
355 		ret = -EIO;
356 	return ret;
357 }
358 EXPORT_SYMBOL(filemap_check_errors);
359 
filemap_check_and_keep_errors(struct address_space * mapping)360 static int filemap_check_and_keep_errors(struct address_space *mapping)
361 {
362 	/* Check for outstanding write errors */
363 	if (test_bit(AS_EIO, &mapping->flags))
364 		return -EIO;
365 	if (test_bit(AS_ENOSPC, &mapping->flags))
366 		return -ENOSPC;
367 	return 0;
368 }
369 
370 /**
371  * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
372  * @mapping:	address space structure to write
373  * @wbc:	the writeback_control controlling the writeout
374  *
375  * Call writepages on the mapping using the provided wbc to control the
376  * writeout.
377  *
378  * Return: %0 on success, negative error code otherwise.
379  */
filemap_fdatawrite_wbc(struct address_space * mapping,struct writeback_control * wbc)380 int filemap_fdatawrite_wbc(struct address_space *mapping,
381 			   struct writeback_control *wbc)
382 {
383 	int ret;
384 
385 	if (!mapping_can_writeback(mapping) ||
386 	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
387 		return 0;
388 
389 	wbc_attach_fdatawrite_inode(wbc, mapping->host);
390 	ret = do_writepages(mapping, wbc);
391 	wbc_detach_inode(wbc);
392 	return ret;
393 }
394 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
395 
396 /**
397  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
398  * @mapping:	address space structure to write
399  * @start:	offset in bytes where the range starts
400  * @end:	offset in bytes where the range ends (inclusive)
401  * @sync_mode:	enable synchronous operation
402  *
403  * Start writeback against all of a mapping's dirty pages that lie
404  * within the byte offsets <start, end> inclusive.
405  *
406  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
407  * opposed to a regular memory cleansing writeback.  The difference between
408  * these two operations is that if a dirty page/buffer is encountered, it must
409  * be waited upon, and not just skipped over.
410  *
411  * Return: %0 on success, negative error code otherwise.
412  */
__filemap_fdatawrite_range(struct address_space * mapping,loff_t start,loff_t end,int sync_mode)413 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
414 				loff_t end, int sync_mode)
415 {
416 	struct writeback_control wbc = {
417 		.sync_mode = sync_mode,
418 		.nr_to_write = LONG_MAX,
419 		.range_start = start,
420 		.range_end = end,
421 	};
422 
423 	return filemap_fdatawrite_wbc(mapping, &wbc);
424 }
425 
__filemap_fdatawrite(struct address_space * mapping,int sync_mode)426 static inline int __filemap_fdatawrite(struct address_space *mapping,
427 	int sync_mode)
428 {
429 	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
430 }
431 
filemap_fdatawrite(struct address_space * mapping)432 int filemap_fdatawrite(struct address_space *mapping)
433 {
434 	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435 }
436 EXPORT_SYMBOL(filemap_fdatawrite);
437 
filemap_fdatawrite_range(struct address_space * mapping,loff_t start,loff_t end)438 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
439 				loff_t end)
440 {
441 	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442 }
443 EXPORT_SYMBOL(filemap_fdatawrite_range);
444 
445 /**
446  * filemap_flush - mostly a non-blocking flush
447  * @mapping:	target address_space
448  *
449  * This is a mostly non-blocking flush.  Not suitable for data-integrity
450  * purposes - I/O may not be started against all dirty pages.
451  *
452  * Return: %0 on success, negative error code otherwise.
453  */
filemap_flush(struct address_space * mapping)454 int filemap_flush(struct address_space *mapping)
455 {
456 	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457 }
458 EXPORT_SYMBOL(filemap_flush);
459 
460 /**
461  * filemap_range_has_page - check if a page exists in range.
462  * @mapping:           address space within which to check
463  * @start_byte:        offset in bytes where the range starts
464  * @end_byte:          offset in bytes where the range ends (inclusive)
465  *
466  * Find at least one page in the range supplied, usually used to check if
467  * direct writing in this range will trigger a writeback.
468  *
469  * Return: %true if at least one page exists in the specified range,
470  * %false otherwise.
471  */
filemap_range_has_page(struct address_space * mapping,loff_t start_byte,loff_t end_byte)472 bool filemap_range_has_page(struct address_space *mapping,
473 			   loff_t start_byte, loff_t end_byte)
474 {
475 	struct folio *folio;
476 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477 	pgoff_t max = end_byte >> PAGE_SHIFT;
478 
479 	if (end_byte < start_byte)
480 		return false;
481 
482 	rcu_read_lock();
483 	for (;;) {
484 		folio = xas_find(&xas, max);
485 		if (xas_retry(&xas, folio))
486 			continue;
487 		/* Shadow entries don't count */
488 		if (xa_is_value(folio))
489 			continue;
490 		/*
491 		 * We don't need to try to pin this page; we're about to
492 		 * release the RCU lock anyway.  It is enough to know that
493 		 * there was a page here recently.
494 		 */
495 		break;
496 	}
497 	rcu_read_unlock();
498 
499 	return folio != NULL;
500 }
501 EXPORT_SYMBOL(filemap_range_has_page);
502 
__filemap_fdatawait_range(struct address_space * mapping,loff_t start_byte,loff_t end_byte)503 static void __filemap_fdatawait_range(struct address_space *mapping,
504 				     loff_t start_byte, loff_t end_byte)
505 {
506 	pgoff_t index = start_byte >> PAGE_SHIFT;
507 	pgoff_t end = end_byte >> PAGE_SHIFT;
508 	struct folio_batch fbatch;
509 	unsigned nr_folios;
510 
511 	folio_batch_init(&fbatch);
512 
513 	while (index <= end) {
514 		unsigned i;
515 
516 		nr_folios = filemap_get_folios_tag(mapping, &index, end,
517 				PAGECACHE_TAG_WRITEBACK, &fbatch);
518 
519 		if (!nr_folios)
520 			break;
521 
522 		for (i = 0; i < nr_folios; i++) {
523 			struct folio *folio = fbatch.folios[i];
524 
525 			folio_wait_writeback(folio);
526 			folio_clear_error(folio);
527 		}
528 		folio_batch_release(&fbatch);
529 		cond_resched();
530 	}
531 }
532 
533 /**
534  * filemap_fdatawait_range - wait for writeback to complete
535  * @mapping:		address space structure to wait for
536  * @start_byte:		offset in bytes where the range starts
537  * @end_byte:		offset in bytes where the range ends (inclusive)
538  *
539  * Walk the list of under-writeback pages of the given address space
540  * in the given range and wait for all of them.  Check error status of
541  * the address space and return it.
542  *
543  * Since the error status of the address space is cleared by this function,
544  * callers are responsible for checking the return value and handling and/or
545  * reporting the error.
546  *
547  * Return: error status of the address space.
548  */
filemap_fdatawait_range(struct address_space * mapping,loff_t start_byte,loff_t end_byte)549 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
550 			    loff_t end_byte)
551 {
552 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
553 	return filemap_check_errors(mapping);
554 }
555 EXPORT_SYMBOL(filemap_fdatawait_range);
556 
557 /**
558  * filemap_fdatawait_range_keep_errors - wait for writeback to complete
559  * @mapping:		address space structure to wait for
560  * @start_byte:		offset in bytes where the range starts
561  * @end_byte:		offset in bytes where the range ends (inclusive)
562  *
563  * Walk the list of under-writeback pages of the given address space in the
564  * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
565  * this function does not clear error status of the address space.
566  *
567  * Use this function if callers don't handle errors themselves.  Expected
568  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
569  * fsfreeze(8)
570  */
filemap_fdatawait_range_keep_errors(struct address_space * mapping,loff_t start_byte,loff_t end_byte)571 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
572 		loff_t start_byte, loff_t end_byte)
573 {
574 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
575 	return filemap_check_and_keep_errors(mapping);
576 }
577 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
578 
579 /**
580  * file_fdatawait_range - wait for writeback to complete
581  * @file:		file pointing to address space structure to wait for
582  * @start_byte:		offset in bytes where the range starts
583  * @end_byte:		offset in bytes where the range ends (inclusive)
584  *
585  * Walk the list of under-writeback pages of the address space that file
586  * refers to, in the given range and wait for all of them.  Check error
587  * status of the address space vs. the file->f_wb_err cursor and return it.
588  *
589  * Since the error status of the file is advanced by this function,
590  * callers are responsible for checking the return value and handling and/or
591  * reporting the error.
592  *
593  * Return: error status of the address space vs. the file->f_wb_err cursor.
594  */
file_fdatawait_range(struct file * file,loff_t start_byte,loff_t end_byte)595 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
596 {
597 	struct address_space *mapping = file->f_mapping;
598 
599 	__filemap_fdatawait_range(mapping, start_byte, end_byte);
600 	return file_check_and_advance_wb_err(file);
601 }
602 EXPORT_SYMBOL(file_fdatawait_range);
603 
604 /**
605  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
606  * @mapping: address space structure to wait for
607  *
608  * Walk the list of under-writeback pages of the given address space
609  * and wait for all of them.  Unlike filemap_fdatawait(), this function
610  * does not clear error status of the address space.
611  *
612  * Use this function if callers don't handle errors themselves.  Expected
613  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
614  * fsfreeze(8)
615  *
616  * Return: error status of the address space.
617  */
filemap_fdatawait_keep_errors(struct address_space * mapping)618 int filemap_fdatawait_keep_errors(struct address_space *mapping)
619 {
620 	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
621 	return filemap_check_and_keep_errors(mapping);
622 }
623 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
624 
625 /* Returns true if writeback might be needed or already in progress. */
mapping_needs_writeback(struct address_space * mapping)626 static bool mapping_needs_writeback(struct address_space *mapping)
627 {
628 	return mapping->nrpages;
629 }
630 
filemap_range_has_writeback(struct address_space * mapping,loff_t start_byte,loff_t end_byte)631 bool filemap_range_has_writeback(struct address_space *mapping,
632 				 loff_t start_byte, loff_t end_byte)
633 {
634 	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
635 	pgoff_t max = end_byte >> PAGE_SHIFT;
636 	struct folio *folio;
637 
638 	if (end_byte < start_byte)
639 		return false;
640 
641 	rcu_read_lock();
642 	xas_for_each(&xas, folio, max) {
643 		if (xas_retry(&xas, folio))
644 			continue;
645 		if (xa_is_value(folio))
646 			continue;
647 		if (folio_test_dirty(folio) || folio_test_locked(folio) ||
648 				folio_test_writeback(folio))
649 			break;
650 	}
651 	rcu_read_unlock();
652 	return folio != NULL;
653 }
654 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
655 
656 /**
657  * filemap_write_and_wait_range - write out & wait on a file range
658  * @mapping:	the address_space for the pages
659  * @lstart:	offset in bytes where the range starts
660  * @lend:	offset in bytes where the range ends (inclusive)
661  *
662  * Write out and wait upon file offsets lstart->lend, inclusive.
663  *
664  * Note that @lend is inclusive (describes the last byte to be written) so
665  * that this function can be used to write to the very end-of-file (end = -1).
666  *
667  * Return: error status of the address space.
668  */
filemap_write_and_wait_range(struct address_space * mapping,loff_t lstart,loff_t lend)669 int filemap_write_and_wait_range(struct address_space *mapping,
670 				 loff_t lstart, loff_t lend)
671 {
672 	int err = 0, err2;
673 
674 	if (lend < lstart)
675 		return 0;
676 
677 	if (mapping_needs_writeback(mapping)) {
678 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
679 						 WB_SYNC_ALL);
680 		/*
681 		 * Even if the above returned error, the pages may be
682 		 * written partially (e.g. -ENOSPC), so we wait for it.
683 		 * But the -EIO is special case, it may indicate the worst
684 		 * thing (e.g. bug) happened, so we avoid waiting for it.
685 		 */
686 		if (err != -EIO)
687 			__filemap_fdatawait_range(mapping, lstart, lend);
688 	}
689 	err2 = filemap_check_errors(mapping);
690 	if (!err)
691 		err = err2;
692 	return err;
693 }
694 EXPORT_SYMBOL(filemap_write_and_wait_range);
695 
__filemap_set_wb_err(struct address_space * mapping,int err)696 void __filemap_set_wb_err(struct address_space *mapping, int err)
697 {
698 	errseq_t eseq = errseq_set(&mapping->wb_err, err);
699 
700 	trace_filemap_set_wb_err(mapping, eseq);
701 }
702 EXPORT_SYMBOL(__filemap_set_wb_err);
703 
704 /**
705  * file_check_and_advance_wb_err - report wb error (if any) that was previously
706  * 				   and advance wb_err to current one
707  * @file: struct file on which the error is being reported
708  *
709  * When userland calls fsync (or something like nfsd does the equivalent), we
710  * want to report any writeback errors that occurred since the last fsync (or
711  * since the file was opened if there haven't been any).
712  *
713  * Grab the wb_err from the mapping. If it matches what we have in the file,
714  * then just quickly return 0. The file is all caught up.
715  *
716  * If it doesn't match, then take the mapping value, set the "seen" flag in
717  * it and try to swap it into place. If it works, or another task beat us
718  * to it with the new value, then update the f_wb_err and return the error
719  * portion. The error at this point must be reported via proper channels
720  * (a'la fsync, or NFS COMMIT operation, etc.).
721  *
722  * While we handle mapping->wb_err with atomic operations, the f_wb_err
723  * value is protected by the f_lock since we must ensure that it reflects
724  * the latest value swapped in for this file descriptor.
725  *
726  * Return: %0 on success, negative error code otherwise.
727  */
file_check_and_advance_wb_err(struct file * file)728 int file_check_and_advance_wb_err(struct file *file)
729 {
730 	int err = 0;
731 	errseq_t old = READ_ONCE(file->f_wb_err);
732 	struct address_space *mapping = file->f_mapping;
733 
734 	/* Locklessly handle the common case where nothing has changed */
735 	if (errseq_check(&mapping->wb_err, old)) {
736 		/* Something changed, must use slow path */
737 		spin_lock(&file->f_lock);
738 		old = file->f_wb_err;
739 		err = errseq_check_and_advance(&mapping->wb_err,
740 						&file->f_wb_err);
741 		trace_file_check_and_advance_wb_err(file, old);
742 		spin_unlock(&file->f_lock);
743 	}
744 
745 	/*
746 	 * We're mostly using this function as a drop in replacement for
747 	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
748 	 * that the legacy code would have had on these flags.
749 	 */
750 	clear_bit(AS_EIO, &mapping->flags);
751 	clear_bit(AS_ENOSPC, &mapping->flags);
752 	return err;
753 }
754 EXPORT_SYMBOL(file_check_and_advance_wb_err);
755 
756 /**
757  * file_write_and_wait_range - write out & wait on a file range
758  * @file:	file pointing to address_space with pages
759  * @lstart:	offset in bytes where the range starts
760  * @lend:	offset in bytes where the range ends (inclusive)
761  *
762  * Write out and wait upon file offsets lstart->lend, inclusive.
763  *
764  * Note that @lend is inclusive (describes the last byte to be written) so
765  * that this function can be used to write to the very end-of-file (end = -1).
766  *
767  * After writing out and waiting on the data, we check and advance the
768  * f_wb_err cursor to the latest value, and return any errors detected there.
769  *
770  * Return: %0 on success, negative error code otherwise.
771  */
file_write_and_wait_range(struct file * file,loff_t lstart,loff_t lend)772 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
773 {
774 	int err = 0, err2;
775 	struct address_space *mapping = file->f_mapping;
776 
777 	if (lend < lstart)
778 		return 0;
779 
780 	if (mapping_needs_writeback(mapping)) {
781 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
782 						 WB_SYNC_ALL);
783 		/* See comment of filemap_write_and_wait() */
784 		if (err != -EIO)
785 			__filemap_fdatawait_range(mapping, lstart, lend);
786 	}
787 	err2 = file_check_and_advance_wb_err(file);
788 	if (!err)
789 		err = err2;
790 	return err;
791 }
792 EXPORT_SYMBOL(file_write_and_wait_range);
793 
794 /**
795  * replace_page_cache_folio - replace a pagecache folio with a new one
796  * @old:	folio to be replaced
797  * @new:	folio to replace with
798  *
799  * This function replaces a folio in the pagecache with a new one.  On
800  * success it acquires the pagecache reference for the new folio and
801  * drops it for the old folio.  Both the old and new folios must be
802  * locked.  This function does not add the new folio to the LRU, the
803  * caller must do that.
804  *
805  * The remove + add is atomic.  This function cannot fail.
806  */
replace_page_cache_folio(struct folio * old,struct folio * new)807 void replace_page_cache_folio(struct folio *old, struct folio *new)
808 {
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_FOLIO(!folio_test_locked(old), old);
815 	VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
816 	VM_BUG_ON_FOLIO(new->mapping, new);
817 
818 	folio_get(new);
819 	new->mapping = mapping;
820 	new->index = offset;
821 
822 	mem_cgroup_migrate(old, new);
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 (!folio_test_hugetlb(old))
830 		__lruvec_stat_sub_folio(old, NR_FILE_PAGES);
831 	if (!folio_test_hugetlb(new))
832 		__lruvec_stat_add_folio(new, NR_FILE_PAGES);
833 	if (folio_test_swapbacked(old))
834 		__lruvec_stat_sub_folio(old, NR_SHMEM);
835 	if (folio_test_swapbacked(new))
836 		__lruvec_stat_add_folio(new, NR_SHMEM);
837 	xas_unlock_irq(&xas);
838 	if (free_folio)
839 		free_folio(old);
840 	folio_put(old);
841 }
842 EXPORT_SYMBOL_GPL(replace_page_cache_folio);
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 
filemap_add_folio(struct address_space * mapping,struct folio * folio,pgoff_t index,gfp_t gfp)932 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
933 				pgoff_t index, gfp_t gfp)
934 {
935 	void *shadow = NULL;
936 	int ret;
937 
938 	__folio_set_locked(folio);
939 	ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
940 	if (unlikely(ret))
941 		__folio_clear_locked(folio);
942 	else {
943 		/*
944 		 * The folio might have been evicted from cache only
945 		 * recently, in which case it should be activated like
946 		 * any other repeatedly accessed folio.
947 		 * The exception is folios getting rewritten; evicting other
948 		 * data from the working set, only to cache data that will
949 		 * get overwritten with something else, is a waste of memory.
950 		 */
951 		WARN_ON_ONCE(folio_test_active(folio));
952 		if (!(gfp & __GFP_WRITE) && shadow)
953 			workingset_refault(folio, shadow);
954 		folio_add_lru(folio);
955 	}
956 	return ret;
957 }
958 EXPORT_SYMBOL_GPL(filemap_add_folio);
959 
960 #ifdef CONFIG_NUMA
filemap_alloc_folio(gfp_t gfp,unsigned int order)961 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
962 {
963 	int n;
964 	struct folio *folio;
965 
966 	if (cpuset_do_page_mem_spread()) {
967 		unsigned int cpuset_mems_cookie;
968 		do {
969 			cpuset_mems_cookie = read_mems_allowed_begin();
970 			n = cpuset_mem_spread_node();
971 			folio = __folio_alloc_node(gfp, order, n);
972 		} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
973 
974 		return folio;
975 	}
976 	return folio_alloc(gfp, order);
977 }
978 EXPORT_SYMBOL(filemap_alloc_folio);
979 #endif
980 
981 /*
982  * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
983  *
984  * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
985  *
986  * @mapping1: the first mapping to lock
987  * @mapping2: the second mapping to lock
988  */
filemap_invalidate_lock_two(struct address_space * mapping1,struct address_space * mapping2)989 void filemap_invalidate_lock_two(struct address_space *mapping1,
990 				 struct address_space *mapping2)
991 {
992 	if (mapping1 > mapping2)
993 		swap(mapping1, mapping2);
994 	if (mapping1)
995 		down_write(&mapping1->invalidate_lock);
996 	if (mapping2 && mapping1 != mapping2)
997 		down_write_nested(&mapping2->invalidate_lock, 1);
998 }
999 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1000 
1001 /*
1002  * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1003  *
1004  * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1005  *
1006  * @mapping1: the first mapping to unlock
1007  * @mapping2: the second mapping to unlock
1008  */
filemap_invalidate_unlock_two(struct address_space * mapping1,struct address_space * mapping2)1009 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1010 				   struct address_space *mapping2)
1011 {
1012 	if (mapping1)
1013 		up_write(&mapping1->invalidate_lock);
1014 	if (mapping2 && mapping1 != mapping2)
1015 		up_write(&mapping2->invalidate_lock);
1016 }
1017 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1018 
1019 /*
1020  * In order to wait for pages to become available there must be
1021  * waitqueues associated with pages. By using a hash table of
1022  * waitqueues where the bucket discipline is to maintain all
1023  * waiters on the same queue and wake all when any of the pages
1024  * become available, and for the woken contexts to check to be
1025  * sure the appropriate page became available, this saves space
1026  * at a cost of "thundering herd" phenomena during rare hash
1027  * collisions.
1028  */
1029 #define PAGE_WAIT_TABLE_BITS 8
1030 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1031 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1032 
folio_waitqueue(struct folio * folio)1033 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1034 {
1035 	return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1036 }
1037 
pagecache_init(void)1038 void __init pagecache_init(void)
1039 {
1040 	int i;
1041 
1042 	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1043 		init_waitqueue_head(&folio_wait_table[i]);
1044 
1045 	page_writeback_init();
1046 }
1047 
1048 /*
1049  * The page wait code treats the "wait->flags" somewhat unusually, because
1050  * we have multiple different kinds of waits, not just the usual "exclusive"
1051  * one.
1052  *
1053  * We have:
1054  *
1055  *  (a) no special bits set:
1056  *
1057  *	We're just waiting for the bit to be released, and when a waker
1058  *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1059  *	and remove it from the wait queue.
1060  *
1061  *	Simple and straightforward.
1062  *
1063  *  (b) WQ_FLAG_EXCLUSIVE:
1064  *
1065  *	The waiter is waiting to get the lock, and only one waiter should
1066  *	be woken up to avoid any thundering herd behavior. We'll set the
1067  *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1068  *
1069  *	This is the traditional exclusive wait.
1070  *
1071  *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1072  *
1073  *	The waiter is waiting to get the bit, and additionally wants the
1074  *	lock to be transferred to it for fair lock behavior. If the lock
1075  *	cannot be taken, we stop walking the wait queue without waking
1076  *	the waiter.
1077  *
1078  *	This is the "fair lock handoff" case, and in addition to setting
1079  *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1080  *	that it now has the lock.
1081  */
wake_page_function(wait_queue_entry_t * wait,unsigned mode,int sync,void * arg)1082 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1083 {
1084 	unsigned int flags;
1085 	struct wait_page_key *key = arg;
1086 	struct wait_page_queue *wait_page
1087 		= container_of(wait, struct wait_page_queue, wait);
1088 
1089 	if (!wake_page_match(wait_page, key))
1090 		return 0;
1091 
1092 	/*
1093 	 * If it's a lock handoff wait, we get the bit for it, and
1094 	 * stop walking (and do not wake it up) if we can't.
1095 	 */
1096 	flags = wait->flags;
1097 	if (flags & WQ_FLAG_EXCLUSIVE) {
1098 		if (test_bit(key->bit_nr, &key->folio->flags))
1099 			return -1;
1100 		if (flags & WQ_FLAG_CUSTOM) {
1101 			if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1102 				return -1;
1103 			flags |= WQ_FLAG_DONE;
1104 		}
1105 	}
1106 
1107 	/*
1108 	 * We are holding the wait-queue lock, but the waiter that
1109 	 * is waiting for this will be checking the flags without
1110 	 * any locking.
1111 	 *
1112 	 * So update the flags atomically, and wake up the waiter
1113 	 * afterwards to avoid any races. This store-release pairs
1114 	 * with the load-acquire in folio_wait_bit_common().
1115 	 */
1116 	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1117 	wake_up_state(wait->private, mode);
1118 
1119 	/*
1120 	 * Ok, we have successfully done what we're waiting for,
1121 	 * and we can unconditionally remove the wait entry.
1122 	 *
1123 	 * Note that this pairs with the "finish_wait()" in the
1124 	 * waiter, and has to be the absolute last thing we do.
1125 	 * After this list_del_init(&wait->entry) the wait entry
1126 	 * might be de-allocated and the process might even have
1127 	 * exited.
1128 	 */
1129 	list_del_init_careful(&wait->entry);
1130 	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1131 }
1132 
folio_wake_bit(struct folio * folio,int bit_nr)1133 static void folio_wake_bit(struct folio *folio, int bit_nr)
1134 {
1135 	wait_queue_head_t *q = folio_waitqueue(folio);
1136 	struct wait_page_key key;
1137 	unsigned long flags;
1138 	wait_queue_entry_t bookmark;
1139 
1140 	key.folio = folio;
1141 	key.bit_nr = bit_nr;
1142 	key.page_match = 0;
1143 
1144 	bookmark.flags = 0;
1145 	bookmark.private = NULL;
1146 	bookmark.func = NULL;
1147 	INIT_LIST_HEAD(&bookmark.entry);
1148 
1149 	spin_lock_irqsave(&q->lock, flags);
1150 	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1151 
1152 	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1153 		/*
1154 		 * Take a breather from holding the lock,
1155 		 * allow pages that finish wake up asynchronously
1156 		 * to acquire the lock and remove themselves
1157 		 * from wait queue
1158 		 */
1159 		spin_unlock_irqrestore(&q->lock, flags);
1160 		cpu_relax();
1161 		spin_lock_irqsave(&q->lock, flags);
1162 		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1163 	}
1164 
1165 	/*
1166 	 * It's possible to miss clearing waiters here, when we woke our page
1167 	 * waiters, but the hashed waitqueue has waiters for other pages on it.
1168 	 * That's okay, it's a rare case. The next waker will clear it.
1169 	 *
1170 	 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1171 	 * other), the flag may be cleared in the course of freeing the page;
1172 	 * but that is not required for correctness.
1173 	 */
1174 	if (!waitqueue_active(q) || !key.page_match)
1175 		folio_clear_waiters(folio);
1176 
1177 	spin_unlock_irqrestore(&q->lock, flags);
1178 }
1179 
folio_wake(struct folio * folio,int bit)1180 static void folio_wake(struct folio *folio, int bit)
1181 {
1182 	if (!folio_test_waiters(folio))
1183 		return;
1184 	folio_wake_bit(folio, bit);
1185 }
1186 
1187 /*
1188  * A choice of three behaviors for folio_wait_bit_common():
1189  */
1190 enum behavior {
1191 	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1192 			 * __folio_lock() waiting on then setting PG_locked.
1193 			 */
1194 	SHARED,		/* Hold ref to page and check the bit when woken, like
1195 			 * folio_wait_writeback() waiting on PG_writeback.
1196 			 */
1197 	DROP,		/* Drop ref to page before wait, no check when woken,
1198 			 * like folio_put_wait_locked() on PG_locked.
1199 			 */
1200 };
1201 
1202 /*
1203  * Attempt to check (or get) the folio flag, and mark us done
1204  * if successful.
1205  */
folio_trylock_flag(struct folio * folio,int bit_nr,struct wait_queue_entry * wait)1206 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1207 					struct wait_queue_entry *wait)
1208 {
1209 	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1210 		if (test_and_set_bit(bit_nr, &folio->flags))
1211 			return false;
1212 	} else if (test_bit(bit_nr, &folio->flags))
1213 		return false;
1214 
1215 	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1216 	return true;
1217 }
1218 
1219 /* How many times do we accept lock stealing from under a waiter? */
1220 int sysctl_page_lock_unfairness = 5;
1221 
folio_wait_bit_common(struct folio * folio,int bit_nr,int state,enum behavior behavior)1222 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1223 		int state, enum behavior behavior)
1224 {
1225 	wait_queue_head_t *q = folio_waitqueue(folio);
1226 	int unfairness = sysctl_page_lock_unfairness;
1227 	struct wait_page_queue wait_page;
1228 	wait_queue_entry_t *wait = &wait_page.wait;
1229 	bool thrashing = false;
1230 	unsigned long pflags;
1231 	bool in_thrashing;
1232 
1233 	if (bit_nr == PG_locked &&
1234 	    !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1235 		delayacct_thrashing_start(&in_thrashing);
1236 		psi_memstall_enter(&pflags);
1237 		thrashing = true;
1238 	}
1239 
1240 	init_wait(wait);
1241 	wait->func = wake_page_function;
1242 	wait_page.folio = folio;
1243 	wait_page.bit_nr = bit_nr;
1244 
1245 repeat:
1246 	wait->flags = 0;
1247 	if (behavior == EXCLUSIVE) {
1248 		wait->flags = WQ_FLAG_EXCLUSIVE;
1249 		if (--unfairness < 0)
1250 			wait->flags |= WQ_FLAG_CUSTOM;
1251 	}
1252 
1253 	/*
1254 	 * Do one last check whether we can get the
1255 	 * page bit synchronously.
1256 	 *
1257 	 * Do the folio_set_waiters() marking before that
1258 	 * to let any waker we _just_ missed know they
1259 	 * need to wake us up (otherwise they'll never
1260 	 * even go to the slow case that looks at the
1261 	 * page queue), and add ourselves to the wait
1262 	 * queue if we need to sleep.
1263 	 *
1264 	 * This part needs to be done under the queue
1265 	 * lock to avoid races.
1266 	 */
1267 	spin_lock_irq(&q->lock);
1268 	folio_set_waiters(folio);
1269 	if (!folio_trylock_flag(folio, bit_nr, wait))
1270 		__add_wait_queue_entry_tail(q, wait);
1271 	spin_unlock_irq(&q->lock);
1272 
1273 	/*
1274 	 * From now on, all the logic will be based on
1275 	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1276 	 * see whether the page bit testing has already
1277 	 * been done by the wake function.
1278 	 *
1279 	 * We can drop our reference to the folio.
1280 	 */
1281 	if (behavior == DROP)
1282 		folio_put(folio);
1283 
1284 	/*
1285 	 * Note that until the "finish_wait()", or until
1286 	 * we see the WQ_FLAG_WOKEN flag, we need to
1287 	 * be very careful with the 'wait->flags', because
1288 	 * we may race with a waker that sets them.
1289 	 */
1290 	for (;;) {
1291 		unsigned int flags;
1292 
1293 		set_current_state(state);
1294 
1295 		/* Loop until we've been woken or interrupted */
1296 		flags = smp_load_acquire(&wait->flags);
1297 		if (!(flags & WQ_FLAG_WOKEN)) {
1298 			if (signal_pending_state(state, current))
1299 				break;
1300 
1301 			io_schedule();
1302 			continue;
1303 		}
1304 
1305 		/* If we were non-exclusive, we're done */
1306 		if (behavior != EXCLUSIVE)
1307 			break;
1308 
1309 		/* If the waker got the lock for us, we're done */
1310 		if (flags & WQ_FLAG_DONE)
1311 			break;
1312 
1313 		/*
1314 		 * Otherwise, if we're getting the lock, we need to
1315 		 * try to get it ourselves.
1316 		 *
1317 		 * And if that fails, we'll have to retry this all.
1318 		 */
1319 		if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1320 			goto repeat;
1321 
1322 		wait->flags |= WQ_FLAG_DONE;
1323 		break;
1324 	}
1325 
1326 	/*
1327 	 * If a signal happened, this 'finish_wait()' may remove the last
1328 	 * waiter from the wait-queues, but the folio waiters bit will remain
1329 	 * set. That's ok. The next wakeup will take care of it, and trying
1330 	 * to do it here would be difficult and prone to races.
1331 	 */
1332 	finish_wait(q, wait);
1333 
1334 	if (thrashing) {
1335 		delayacct_thrashing_end(&in_thrashing);
1336 		psi_memstall_leave(&pflags);
1337 	}
1338 
1339 	/*
1340 	 * NOTE! The wait->flags weren't stable until we've done the
1341 	 * 'finish_wait()', and we could have exited the loop above due
1342 	 * to a signal, and had a wakeup event happen after the signal
1343 	 * test but before the 'finish_wait()'.
1344 	 *
1345 	 * So only after the finish_wait() can we reliably determine
1346 	 * if we got woken up or not, so we can now figure out the final
1347 	 * return value based on that state without races.
1348 	 *
1349 	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1350 	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1351 	 */
1352 	if (behavior == EXCLUSIVE)
1353 		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1354 
1355 	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1356 }
1357 
1358 #ifdef CONFIG_MIGRATION
1359 /**
1360  * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1361  * @entry: migration swap entry.
1362  * @ptl: already locked ptl. This function will drop the lock.
1363  *
1364  * Wait for a migration entry referencing the given page to be removed. This is
1365  * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1366  * this can be called without taking a reference on the page. Instead this
1367  * should be called while holding the ptl for the migration entry referencing
1368  * the page.
1369  *
1370  * Returns after unlocking the ptl.
1371  *
1372  * This follows the same logic as folio_wait_bit_common() so see the comments
1373  * there.
1374  */
migration_entry_wait_on_locked(swp_entry_t entry,spinlock_t * ptl)1375 void migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
1376 	__releases(ptl)
1377 {
1378 	struct wait_page_queue wait_page;
1379 	wait_queue_entry_t *wait = &wait_page.wait;
1380 	bool thrashing = false;
1381 	unsigned long pflags;
1382 	bool in_thrashing;
1383 	wait_queue_head_t *q;
1384 	struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1385 
1386 	q = folio_waitqueue(folio);
1387 	if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1388 		delayacct_thrashing_start(&in_thrashing);
1389 		psi_memstall_enter(&pflags);
1390 		thrashing = true;
1391 	}
1392 
1393 	init_wait(wait);
1394 	wait->func = wake_page_function;
1395 	wait_page.folio = folio;
1396 	wait_page.bit_nr = PG_locked;
1397 	wait->flags = 0;
1398 
1399 	spin_lock_irq(&q->lock);
1400 	folio_set_waiters(folio);
1401 	if (!folio_trylock_flag(folio, PG_locked, wait))
1402 		__add_wait_queue_entry_tail(q, wait);
1403 	spin_unlock_irq(&q->lock);
1404 
1405 	/*
1406 	 * If a migration entry exists for the page the migration path must hold
1407 	 * a valid reference to the page, and it must take the ptl to remove the
1408 	 * migration entry. So the page is valid until the ptl is dropped.
1409 	 */
1410 	spin_unlock(ptl);
1411 
1412 	for (;;) {
1413 		unsigned int flags;
1414 
1415 		set_current_state(TASK_UNINTERRUPTIBLE);
1416 
1417 		/* Loop until we've been woken or interrupted */
1418 		flags = smp_load_acquire(&wait->flags);
1419 		if (!(flags & WQ_FLAG_WOKEN)) {
1420 			if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1421 				break;
1422 
1423 			io_schedule();
1424 			continue;
1425 		}
1426 		break;
1427 	}
1428 
1429 	finish_wait(q, wait);
1430 
1431 	if (thrashing) {
1432 		delayacct_thrashing_end(&in_thrashing);
1433 		psi_memstall_leave(&pflags);
1434 	}
1435 }
1436 #endif
1437 
folio_wait_bit(struct folio * folio,int bit_nr)1438 void folio_wait_bit(struct folio *folio, int bit_nr)
1439 {
1440 	folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1441 }
1442 EXPORT_SYMBOL(folio_wait_bit);
1443 
folio_wait_bit_killable(struct folio * folio,int bit_nr)1444 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1445 {
1446 	return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1447 }
1448 EXPORT_SYMBOL(folio_wait_bit_killable);
1449 
1450 /**
1451  * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1452  * @folio: The folio to wait for.
1453  * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1454  *
1455  * The caller should hold a reference on @folio.  They expect the page to
1456  * become unlocked relatively soon, but do not wish to hold up migration
1457  * (for example) by holding the reference while waiting for the folio to
1458  * come unlocked.  After this function returns, the caller should not
1459  * dereference @folio.
1460  *
1461  * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1462  */
folio_put_wait_locked(struct folio * folio,int state)1463 static int folio_put_wait_locked(struct folio *folio, int state)
1464 {
1465 	return folio_wait_bit_common(folio, PG_locked, state, DROP);
1466 }
1467 
1468 /**
1469  * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1470  * @folio: Folio defining the wait queue of interest
1471  * @waiter: Waiter to add to the queue
1472  *
1473  * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1474  */
folio_add_wait_queue(struct folio * folio,wait_queue_entry_t * waiter)1475 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1476 {
1477 	wait_queue_head_t *q = folio_waitqueue(folio);
1478 	unsigned long flags;
1479 
1480 	spin_lock_irqsave(&q->lock, flags);
1481 	__add_wait_queue_entry_tail(q, waiter);
1482 	folio_set_waiters(folio);
1483 	spin_unlock_irqrestore(&q->lock, flags);
1484 }
1485 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1486 
1487 #ifndef clear_bit_unlock_is_negative_byte
1488 
1489 /*
1490  * PG_waiters is the high bit in the same byte as PG_lock.
1491  *
1492  * On x86 (and on many other architectures), we can clear PG_lock and
1493  * test the sign bit at the same time. But if the architecture does
1494  * not support that special operation, we just do this all by hand
1495  * instead.
1496  *
1497  * The read of PG_waiters has to be after (or concurrently with) PG_locked
1498  * being cleared, but a memory barrier should be unnecessary since it is
1499  * in the same byte as PG_locked.
1500  */
clear_bit_unlock_is_negative_byte(long nr,volatile void * mem)1501 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1502 {
1503 	clear_bit_unlock(nr, mem);
1504 	/* smp_mb__after_atomic(); */
1505 	return test_bit(PG_waiters, mem);
1506 }
1507 
1508 #endif
1509 
1510 /**
1511  * folio_unlock - Unlock a locked folio.
1512  * @folio: The folio.
1513  *
1514  * Unlocks the folio and wakes up any thread sleeping on the page lock.
1515  *
1516  * Context: May be called from interrupt or process context.  May not be
1517  * called from NMI context.
1518  */
folio_unlock(struct folio * folio)1519 void folio_unlock(struct folio *folio)
1520 {
1521 	/* Bit 7 allows x86 to check the byte's sign bit */
1522 	BUILD_BUG_ON(PG_waiters != 7);
1523 	BUILD_BUG_ON(PG_locked > 7);
1524 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1525 	if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1526 		folio_wake_bit(folio, PG_locked);
1527 }
1528 EXPORT_SYMBOL(folio_unlock);
1529 
1530 /**
1531  * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1532  * @folio: The folio.
1533  *
1534  * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1535  * it.  The folio reference held for PG_private_2 being set is released.
1536  *
1537  * This is, for example, used when a netfs folio is being written to a local
1538  * disk cache, thereby allowing writes to the cache for the same folio to be
1539  * serialised.
1540  */
folio_end_private_2(struct folio * folio)1541 void folio_end_private_2(struct folio *folio)
1542 {
1543 	VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1544 	clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1545 	folio_wake_bit(folio, PG_private_2);
1546 	folio_put(folio);
1547 }
1548 EXPORT_SYMBOL(folio_end_private_2);
1549 
1550 /**
1551  * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1552  * @folio: The folio to wait on.
1553  *
1554  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1555  */
folio_wait_private_2(struct folio * folio)1556 void folio_wait_private_2(struct folio *folio)
1557 {
1558 	while (folio_test_private_2(folio))
1559 		folio_wait_bit(folio, PG_private_2);
1560 }
1561 EXPORT_SYMBOL(folio_wait_private_2);
1562 
1563 /**
1564  * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1565  * @folio: The folio to wait on.
1566  *
1567  * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1568  * fatal signal is received by the calling task.
1569  *
1570  * Return:
1571  * - 0 if successful.
1572  * - -EINTR if a fatal signal was encountered.
1573  */
folio_wait_private_2_killable(struct folio * folio)1574 int folio_wait_private_2_killable(struct folio *folio)
1575 {
1576 	int ret = 0;
1577 
1578 	while (folio_test_private_2(folio)) {
1579 		ret = folio_wait_bit_killable(folio, PG_private_2);
1580 		if (ret < 0)
1581 			break;
1582 	}
1583 
1584 	return ret;
1585 }
1586 EXPORT_SYMBOL(folio_wait_private_2_killable);
1587 
1588 /**
1589  * folio_end_writeback - End writeback against a folio.
1590  * @folio: The folio.
1591  */
folio_end_writeback(struct folio * folio)1592 void folio_end_writeback(struct folio *folio)
1593 {
1594 	/*
1595 	 * folio_test_clear_reclaim() could be used here but it is an
1596 	 * atomic operation and overkill in this particular case. Failing
1597 	 * to shuffle a folio marked for immediate reclaim is too mild
1598 	 * a gain to justify taking an atomic operation penalty at the
1599 	 * end of every folio writeback.
1600 	 */
1601 	if (folio_test_reclaim(folio)) {
1602 		folio_clear_reclaim(folio);
1603 		folio_rotate_reclaimable(folio);
1604 	}
1605 
1606 	/*
1607 	 * Writeback does not hold a folio reference of its own, relying
1608 	 * on truncation to wait for the clearing of PG_writeback.
1609 	 * But here we must make sure that the folio is not freed and
1610 	 * reused before the folio_wake().
1611 	 */
1612 	folio_get(folio);
1613 	if (!__folio_end_writeback(folio))
1614 		BUG();
1615 
1616 	smp_mb__after_atomic();
1617 	folio_wake(folio, PG_writeback);
1618 	acct_reclaim_writeback(folio);
1619 	folio_put(folio);
1620 }
1621 EXPORT_SYMBOL(folio_end_writeback);
1622 
1623 /**
1624  * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1625  * @folio: The folio to lock
1626  */
__folio_lock(struct folio * folio)1627 void __folio_lock(struct folio *folio)
1628 {
1629 	folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1630 				EXCLUSIVE);
1631 }
1632 EXPORT_SYMBOL(__folio_lock);
1633 
__folio_lock_killable(struct folio * folio)1634 int __folio_lock_killable(struct folio *folio)
1635 {
1636 	return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1637 					EXCLUSIVE);
1638 }
1639 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1640 
__folio_lock_async(struct folio * folio,struct wait_page_queue * wait)1641 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1642 {
1643 	struct wait_queue_head *q = folio_waitqueue(folio);
1644 	int ret = 0;
1645 
1646 	wait->folio = folio;
1647 	wait->bit_nr = PG_locked;
1648 
1649 	spin_lock_irq(&q->lock);
1650 	__add_wait_queue_entry_tail(q, &wait->wait);
1651 	folio_set_waiters(folio);
1652 	ret = !folio_trylock(folio);
1653 	/*
1654 	 * If we were successful now, we know we're still on the
1655 	 * waitqueue as we're still under the lock. This means it's
1656 	 * safe to remove and return success, we know the callback
1657 	 * isn't going to trigger.
1658 	 */
1659 	if (!ret)
1660 		__remove_wait_queue(q, &wait->wait);
1661 	else
1662 		ret = -EIOCBQUEUED;
1663 	spin_unlock_irq(&q->lock);
1664 	return ret;
1665 }
1666 
1667 /*
1668  * Return values:
1669  * 0 - folio is locked.
1670  * non-zero - folio is not locked.
1671  *     mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
1672  *     vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
1673  *     FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
1674  *
1675  * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
1676  * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
1677  */
__folio_lock_or_retry(struct folio * folio,struct vm_fault * vmf)1678 vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
1679 {
1680 	unsigned int flags = vmf->flags;
1681 
1682 	if (fault_flag_allow_retry_first(flags)) {
1683 		/*
1684 		 * CAUTION! In this case, mmap_lock/per-VMA lock is not
1685 		 * released even though returning VM_FAULT_RETRY.
1686 		 */
1687 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1688 			return VM_FAULT_RETRY;
1689 
1690 		release_fault_lock(vmf);
1691 		if (flags & FAULT_FLAG_KILLABLE)
1692 			folio_wait_locked_killable(folio);
1693 		else
1694 			folio_wait_locked(folio);
1695 		return VM_FAULT_RETRY;
1696 	}
1697 	if (flags & FAULT_FLAG_KILLABLE) {
1698 		bool ret;
1699 
1700 		ret = __folio_lock_killable(folio);
1701 		if (ret) {
1702 			release_fault_lock(vmf);
1703 			return VM_FAULT_RETRY;
1704 		}
1705 	} else {
1706 		__folio_lock(folio);
1707 	}
1708 
1709 	return 0;
1710 }
1711 
1712 /**
1713  * page_cache_next_miss() - Find the next gap in the page cache.
1714  * @mapping: Mapping.
1715  * @index: Index.
1716  * @max_scan: Maximum range to search.
1717  *
1718  * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1719  * gap with the lowest index.
1720  *
1721  * This function may be called under the rcu_read_lock.  However, this will
1722  * not atomically search a snapshot of the cache at a single point in time.
1723  * For example, if a gap is created at index 5, then subsequently a gap is
1724  * created at index 10, page_cache_next_miss covering both indices may
1725  * return 10 if called under the rcu_read_lock.
1726  *
1727  * Return: The index of the gap if found, otherwise an index outside the
1728  * range specified (in which case 'return - index >= max_scan' will be true).
1729  * In the rare case of index wrap-around, 0 will be returned.
1730  */
page_cache_next_miss(struct address_space * mapping,pgoff_t index,unsigned long max_scan)1731 pgoff_t page_cache_next_miss(struct address_space *mapping,
1732 			     pgoff_t index, unsigned long max_scan)
1733 {
1734 	XA_STATE(xas, &mapping->i_pages, index);
1735 
1736 	while (max_scan--) {
1737 		void *entry = xas_next(&xas);
1738 		if (!entry || xa_is_value(entry))
1739 			break;
1740 		if (xas.xa_index == 0)
1741 			break;
1742 	}
1743 
1744 	return xas.xa_index;
1745 }
1746 EXPORT_SYMBOL(page_cache_next_miss);
1747 
1748 /**
1749  * page_cache_prev_miss() - Find the previous gap in the page cache.
1750  * @mapping: Mapping.
1751  * @index: Index.
1752  * @max_scan: Maximum range to search.
1753  *
1754  * Search the range [max(index - max_scan + 1, 0), index] for the
1755  * gap with the highest index.
1756  *
1757  * This function may be called under the rcu_read_lock.  However, this will
1758  * not atomically search a snapshot of the cache at a single point in time.
1759  * For example, if a gap is created at index 10, then subsequently a gap is
1760  * created at index 5, page_cache_prev_miss() covering both indices may
1761  * return 5 if called under the rcu_read_lock.
1762  *
1763  * Return: The index of the gap if found, otherwise an index outside the
1764  * range specified (in which case 'index - return >= max_scan' will be true).
1765  * In the rare case of wrap-around, ULONG_MAX will be returned.
1766  */
page_cache_prev_miss(struct address_space * mapping,pgoff_t index,unsigned long max_scan)1767 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1768 			     pgoff_t index, unsigned long max_scan)
1769 {
1770 	XA_STATE(xas, &mapping->i_pages, index);
1771 
1772 	while (max_scan--) {
1773 		void *entry = xas_prev(&xas);
1774 		if (!entry || xa_is_value(entry))
1775 			break;
1776 		if (xas.xa_index == ULONG_MAX)
1777 			break;
1778 	}
1779 
1780 	return xas.xa_index;
1781 }
1782 EXPORT_SYMBOL(page_cache_prev_miss);
1783 
1784 /*
1785  * Lockless page cache protocol:
1786  * On the lookup side:
1787  * 1. Load the folio from i_pages
1788  * 2. Increment the refcount if it's not zero
1789  * 3. If the folio is not found by xas_reload(), put the refcount and retry
1790  *
1791  * On the removal side:
1792  * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1793  * B. Remove the page from i_pages
1794  * C. Return the page to the page allocator
1795  *
1796  * This means that any page may have its reference count temporarily
1797  * increased by a speculative page cache (or fast GUP) lookup as it can
1798  * be allocated by another user before the RCU grace period expires.
1799  * Because the refcount temporarily acquired here may end up being the
1800  * last refcount on the page, any page allocation must be freeable by
1801  * folio_put().
1802  */
1803 
1804 /*
1805  * filemap_get_entry - Get a page cache entry.
1806  * @mapping: the address_space to search
1807  * @index: The page cache index.
1808  *
1809  * Looks up the page cache entry at @mapping & @index.  If it is a folio,
1810  * it is returned with an increased refcount.  If it is a shadow entry
1811  * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1812  * it is returned without further action.
1813  *
1814  * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1815  */
filemap_get_entry(struct address_space * mapping,pgoff_t index)1816 void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1817 {
1818 	XA_STATE(xas, &mapping->i_pages, index);
1819 	struct folio *folio;
1820 
1821 	rcu_read_lock();
1822 repeat:
1823 	xas_reset(&xas);
1824 	folio = xas_load(&xas);
1825 	if (xas_retry(&xas, folio))
1826 		goto repeat;
1827 	/*
1828 	 * A shadow entry of a recently evicted page, or a swap entry from
1829 	 * shmem/tmpfs.  Return it without attempting to raise page count.
1830 	 */
1831 	if (!folio || xa_is_value(folio))
1832 		goto out;
1833 
1834 	if (!folio_try_get_rcu(folio))
1835 		goto repeat;
1836 
1837 	if (unlikely(folio != xas_reload(&xas))) {
1838 		folio_put(folio);
1839 		goto repeat;
1840 	}
1841 out:
1842 	rcu_read_unlock();
1843 
1844 	return folio;
1845 }
1846 
1847 /**
1848  * __filemap_get_folio - Find and get a reference to a folio.
1849  * @mapping: The address_space to search.
1850  * @index: The page index.
1851  * @fgp_flags: %FGP flags modify how the folio is returned.
1852  * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1853  *
1854  * Looks up the page cache entry at @mapping & @index.
1855  *
1856  * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1857  * if the %GFP flags specified for %FGP_CREAT are atomic.
1858  *
1859  * If this function returns a folio, it is returned with an increased refcount.
1860  *
1861  * Return: The found folio or an ERR_PTR() otherwise.
1862  */
__filemap_get_folio(struct address_space * mapping,pgoff_t index,fgf_t fgp_flags,gfp_t gfp)1863 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1864 		fgf_t fgp_flags, gfp_t gfp)
1865 {
1866 	struct folio *folio;
1867 
1868 repeat:
1869 	folio = filemap_get_entry(mapping, index);
1870 	if (xa_is_value(folio))
1871 		folio = NULL;
1872 	if (!folio)
1873 		goto no_page;
1874 
1875 	if (fgp_flags & FGP_LOCK) {
1876 		if (fgp_flags & FGP_NOWAIT) {
1877 			if (!folio_trylock(folio)) {
1878 				folio_put(folio);
1879 				return ERR_PTR(-EAGAIN);
1880 			}
1881 		} else {
1882 			folio_lock(folio);
1883 		}
1884 
1885 		/* Has the page been truncated? */
1886 		if (unlikely(folio->mapping != mapping)) {
1887 			folio_unlock(folio);
1888 			folio_put(folio);
1889 			goto repeat;
1890 		}
1891 		VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1892 	}
1893 
1894 	if (fgp_flags & FGP_ACCESSED)
1895 		folio_mark_accessed(folio);
1896 	else if (fgp_flags & FGP_WRITE) {
1897 		/* Clear idle flag for buffer write */
1898 		if (folio_test_idle(folio))
1899 			folio_clear_idle(folio);
1900 	}
1901 
1902 	if (fgp_flags & FGP_STABLE)
1903 		folio_wait_stable(folio);
1904 no_page:
1905 	if (!folio && (fgp_flags & FGP_CREAT)) {
1906 		unsigned order = FGF_GET_ORDER(fgp_flags);
1907 		int err;
1908 
1909 		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1910 			gfp |= __GFP_WRITE;
1911 		if (fgp_flags & FGP_NOFS)
1912 			gfp &= ~__GFP_FS;
1913 		if (fgp_flags & FGP_NOWAIT) {
1914 			gfp &= ~GFP_KERNEL;
1915 			gfp |= GFP_NOWAIT | __GFP_NOWARN;
1916 		}
1917 		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1918 			fgp_flags |= FGP_LOCK;
1919 
1920 		if (!mapping_large_folio_support(mapping))
1921 			order = 0;
1922 		if (order > MAX_PAGECACHE_ORDER)
1923 			order = MAX_PAGECACHE_ORDER;
1924 		/* If we're not aligned, allocate a smaller folio */
1925 		if (index & ((1UL << order) - 1))
1926 			order = __ffs(index);
1927 
1928 		do {
1929 			gfp_t alloc_gfp = gfp;
1930 
1931 			err = -ENOMEM;
1932 			if (order == 1)
1933 				order = 0;
1934 			if (order > 0)
1935 				alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
1936 			folio = filemap_alloc_folio(alloc_gfp, order);
1937 			if (!folio)
1938 				continue;
1939 
1940 			/* Init accessed so avoid atomic mark_page_accessed later */
1941 			if (fgp_flags & FGP_ACCESSED)
1942 				__folio_set_referenced(folio);
1943 
1944 			err = filemap_add_folio(mapping, folio, index, gfp);
1945 			if (!err)
1946 				break;
1947 			folio_put(folio);
1948 			folio = NULL;
1949 		} while (order-- > 0);
1950 
1951 		if (err == -EEXIST)
1952 			goto repeat;
1953 		if (err)
1954 			return ERR_PTR(err);
1955 		/*
1956 		 * filemap_add_folio locks the page, and for mmap
1957 		 * we expect an unlocked page.
1958 		 */
1959 		if (folio && (fgp_flags & FGP_FOR_MMAP))
1960 			folio_unlock(folio);
1961 	}
1962 
1963 	if (!folio)
1964 		return ERR_PTR(-ENOENT);
1965 	return folio;
1966 }
1967 EXPORT_SYMBOL(__filemap_get_folio);
1968 
find_get_entry(struct xa_state * xas,pgoff_t max,xa_mark_t mark)1969 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1970 		xa_mark_t mark)
1971 {
1972 	struct folio *folio;
1973 
1974 retry:
1975 	if (mark == XA_PRESENT)
1976 		folio = xas_find(xas, max);
1977 	else
1978 		folio = xas_find_marked(xas, max, mark);
1979 
1980 	if (xas_retry(xas, folio))
1981 		goto retry;
1982 	/*
1983 	 * A shadow entry of a recently evicted page, a swap
1984 	 * entry from shmem/tmpfs or a DAX entry.  Return it
1985 	 * without attempting to raise page count.
1986 	 */
1987 	if (!folio || xa_is_value(folio))
1988 		return folio;
1989 
1990 	if (!folio_try_get_rcu(folio))
1991 		goto reset;
1992 
1993 	if (unlikely(folio != xas_reload(xas))) {
1994 		folio_put(folio);
1995 		goto reset;
1996 	}
1997 
1998 	return folio;
1999 reset:
2000 	xas_reset(xas);
2001 	goto retry;
2002 }
2003 
2004 /**
2005  * find_get_entries - gang pagecache lookup
2006  * @mapping:	The address_space to search
2007  * @start:	The starting page cache index
2008  * @end:	The final page index (inclusive).
2009  * @fbatch:	Where the resulting entries are placed.
2010  * @indices:	The cache indices corresponding to the entries in @entries
2011  *
2012  * find_get_entries() will search for and return a batch of entries in
2013  * the mapping.  The entries are placed in @fbatch.  find_get_entries()
2014  * takes a reference on any actual folios it returns.
2015  *
2016  * The entries have ascending indexes.  The indices may not be consecutive
2017  * due to not-present entries or large folios.
2018  *
2019  * Any shadow entries of evicted folios, or swap entries from
2020  * shmem/tmpfs, are included in the returned array.
2021  *
2022  * Return: The number of entries which were found.
2023  */
find_get_entries(struct address_space * mapping,pgoff_t * start,pgoff_t end,struct folio_batch * fbatch,pgoff_t * indices)2024 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2025 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2026 {
2027 	XA_STATE(xas, &mapping->i_pages, *start);
2028 	struct folio *folio;
2029 
2030 	rcu_read_lock();
2031 	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2032 		indices[fbatch->nr] = xas.xa_index;
2033 		if (!folio_batch_add(fbatch, folio))
2034 			break;
2035 	}
2036 	rcu_read_unlock();
2037 
2038 	if (folio_batch_count(fbatch)) {
2039 		unsigned long nr = 1;
2040 		int idx = folio_batch_count(fbatch) - 1;
2041 
2042 		folio = fbatch->folios[idx];
2043 		if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2044 			nr = folio_nr_pages(folio);
2045 		*start = indices[idx] + nr;
2046 	}
2047 	return folio_batch_count(fbatch);
2048 }
2049 
2050 /**
2051  * find_lock_entries - Find a batch of pagecache entries.
2052  * @mapping:	The address_space to search.
2053  * @start:	The starting page cache index.
2054  * @end:	The final page index (inclusive).
2055  * @fbatch:	Where the resulting entries are placed.
2056  * @indices:	The cache indices of the entries in @fbatch.
2057  *
2058  * find_lock_entries() will return a batch of entries from @mapping.
2059  * Swap, shadow and DAX entries are included.  Folios are returned
2060  * locked and with an incremented refcount.  Folios which are locked
2061  * by somebody else or under writeback are skipped.  Folios which are
2062  * partially outside the range are not returned.
2063  *
2064  * The entries have ascending indexes.  The indices may not be consecutive
2065  * due to not-present entries, large folios, folios which could not be
2066  * locked or folios under writeback.
2067  *
2068  * Return: The number of entries which were found.
2069  */
find_lock_entries(struct address_space * mapping,pgoff_t * start,pgoff_t end,struct folio_batch * fbatch,pgoff_t * indices)2070 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2071 		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2072 {
2073 	XA_STATE(xas, &mapping->i_pages, *start);
2074 	struct folio *folio;
2075 
2076 	rcu_read_lock();
2077 	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2078 		if (!xa_is_value(folio)) {
2079 			if (folio->index < *start)
2080 				goto put;
2081 			if (folio_next_index(folio) - 1 > end)
2082 				goto put;
2083 			if (!folio_trylock(folio))
2084 				goto put;
2085 			if (folio->mapping != mapping ||
2086 			    folio_test_writeback(folio))
2087 				goto unlock;
2088 			VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2089 					folio);
2090 		}
2091 		indices[fbatch->nr] = xas.xa_index;
2092 		if (!folio_batch_add(fbatch, folio))
2093 			break;
2094 		continue;
2095 unlock:
2096 		folio_unlock(folio);
2097 put:
2098 		folio_put(folio);
2099 	}
2100 	rcu_read_unlock();
2101 
2102 	if (folio_batch_count(fbatch)) {
2103 		unsigned long nr = 1;
2104 		int idx = folio_batch_count(fbatch) - 1;
2105 
2106 		folio = fbatch->folios[idx];
2107 		if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2108 			nr = folio_nr_pages(folio);
2109 		*start = indices[idx] + nr;
2110 	}
2111 	return folio_batch_count(fbatch);
2112 }
2113 
2114 /**
2115  * filemap_get_folios - Get a batch of folios
2116  * @mapping:	The address_space to search
2117  * @start:	The starting page index
2118  * @end:	The final page index (inclusive)
2119  * @fbatch:	The batch to fill.
2120  *
2121  * Search for and return a batch of folios in the mapping starting at
2122  * index @start and up to index @end (inclusive).  The folios are returned
2123  * in @fbatch with an elevated reference count.
2124  *
2125  * The first folio may start before @start; if it does, it will contain
2126  * @start.  The final folio may extend beyond @end; if it does, it will
2127  * contain @end.  The folios have ascending indices.  There may be gaps
2128  * between the folios if there are indices which have no folio in the
2129  * page cache.  If folios are added to or removed from the page cache
2130  * while this is running, they may or may not be found by this call.
2131  *
2132  * Return: The number of folios which were found.
2133  * We also update @start to index the next folio for the traversal.
2134  */
filemap_get_folios(struct address_space * mapping,pgoff_t * start,pgoff_t end,struct folio_batch * fbatch)2135 unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2136 		pgoff_t end, struct folio_batch *fbatch)
2137 {
2138 	XA_STATE(xas, &mapping->i_pages, *start);
2139 	struct folio *folio;
2140 
2141 	rcu_read_lock();
2142 	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2143 		/* Skip over shadow, swap and DAX entries */
2144 		if (xa_is_value(folio))
2145 			continue;
2146 		if (!folio_batch_add(fbatch, folio)) {
2147 			unsigned long nr = folio_nr_pages(folio);
2148 
2149 			if (folio_test_hugetlb(folio))
2150 				nr = 1;
2151 			*start = folio->index + nr;
2152 			goto out;
2153 		}
2154 	}
2155 
2156 	/*
2157 	 * We come here when there is no page beyond @end. We take care to not
2158 	 * overflow the index @start as it confuses some of the callers. This
2159 	 * breaks the iteration when there is a page at index -1 but that is
2160 	 * already broken anyway.
2161 	 */
2162 	if (end == (pgoff_t)-1)
2163 		*start = (pgoff_t)-1;
2164 	else
2165 		*start = end + 1;
2166 out:
2167 	rcu_read_unlock();
2168 
2169 	return folio_batch_count(fbatch);
2170 }
2171 EXPORT_SYMBOL(filemap_get_folios);
2172 
2173 /**
2174  * filemap_get_folios_contig - Get a batch of contiguous folios
2175  * @mapping:	The address_space to search
2176  * @start:	The starting page index
2177  * @end:	The final page index (inclusive)
2178  * @fbatch:	The batch to fill
2179  *
2180  * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2181  * except the returned folios are guaranteed to be contiguous. This may
2182  * not return all contiguous folios if the batch gets filled up.
2183  *
2184  * Return: The number of folios found.
2185  * Also update @start to be positioned for traversal of the next folio.
2186  */
2187 
filemap_get_folios_contig(struct address_space * mapping,pgoff_t * start,pgoff_t end,struct folio_batch * fbatch)2188 unsigned filemap_get_folios_contig(struct address_space *mapping,
2189 		pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2190 {
2191 	XA_STATE(xas, &mapping->i_pages, *start);
2192 	unsigned long nr;
2193 	struct folio *folio;
2194 
2195 	rcu_read_lock();
2196 
2197 	for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2198 			folio = xas_next(&xas)) {
2199 		if (xas_retry(&xas, folio))
2200 			continue;
2201 		/*
2202 		 * If the entry has been swapped out, we can stop looking.
2203 		 * No current caller is looking for DAX entries.
2204 		 */
2205 		if (xa_is_value(folio))
2206 			goto update_start;
2207 
2208 		if (!folio_try_get_rcu(folio))
2209 			goto retry;
2210 
2211 		if (unlikely(folio != xas_reload(&xas)))
2212 			goto put_folio;
2213 
2214 		if (!folio_batch_add(fbatch, folio)) {
2215 			nr = folio_nr_pages(folio);
2216 
2217 			if (folio_test_hugetlb(folio))
2218 				nr = 1;
2219 			*start = folio->index + nr;
2220 			goto out;
2221 		}
2222 		continue;
2223 put_folio:
2224 		folio_put(folio);
2225 
2226 retry:
2227 		xas_reset(&xas);
2228 	}
2229 
2230 update_start:
2231 	nr = folio_batch_count(fbatch);
2232 
2233 	if (nr) {
2234 		folio = fbatch->folios[nr - 1];
2235 		if (folio_test_hugetlb(folio))
2236 			*start = folio->index + 1;
2237 		else
2238 			*start = folio_next_index(folio);
2239 	}
2240 out:
2241 	rcu_read_unlock();
2242 	return folio_batch_count(fbatch);
2243 }
2244 EXPORT_SYMBOL(filemap_get_folios_contig);
2245 
2246 /**
2247  * filemap_get_folios_tag - Get a batch of folios matching @tag
2248  * @mapping:    The address_space to search
2249  * @start:      The starting page index
2250  * @end:        The final page index (inclusive)
2251  * @tag:        The tag index
2252  * @fbatch:     The batch to fill
2253  *
2254  * Same as filemap_get_folios(), but only returning folios tagged with @tag.
2255  *
2256  * Return: The number of folios found.
2257  * Also update @start to index the next folio for traversal.
2258  */
filemap_get_folios_tag(struct address_space * mapping,pgoff_t * start,pgoff_t end,xa_mark_t tag,struct folio_batch * fbatch)2259 unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2260 			pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2261 {
2262 	XA_STATE(xas, &mapping->i_pages, *start);
2263 	struct folio *folio;
2264 
2265 	rcu_read_lock();
2266 	while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2267 		/*
2268 		 * Shadow entries should never be tagged, but this iteration
2269 		 * is lockless so there is a window for page reclaim to evict
2270 		 * a page we saw tagged. Skip over it.
2271 		 */
2272 		if (xa_is_value(folio))
2273 			continue;
2274 		if (!folio_batch_add(fbatch, folio)) {
2275 			unsigned long nr = folio_nr_pages(folio);
2276 
2277 			if (folio_test_hugetlb(folio))
2278 				nr = 1;
2279 			*start = folio->index + nr;
2280 			goto out;
2281 		}
2282 	}
2283 	/*
2284 	 * We come here when there is no page beyond @end. We take care to not
2285 	 * overflow the index @start as it confuses some of the callers. This
2286 	 * breaks the iteration when there is a page at index -1 but that is
2287 	 * already broke anyway.
2288 	 */
2289 	if (end == (pgoff_t)-1)
2290 		*start = (pgoff_t)-1;
2291 	else
2292 		*start = end + 1;
2293 out:
2294 	rcu_read_unlock();
2295 
2296 	return folio_batch_count(fbatch);
2297 }
2298 EXPORT_SYMBOL(filemap_get_folios_tag);
2299 
2300 /*
2301  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2302  * a _large_ part of the i/o request. Imagine the worst scenario:
2303  *
2304  *      ---R__________________________________________B__________
2305  *         ^ reading here                             ^ bad block(assume 4k)
2306  *
2307  * read(R) => miss => readahead(R...B) => media error => frustrating retries
2308  * => failing the whole request => read(R) => read(R+1) =>
2309  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2310  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2311  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2312  *
2313  * It is going insane. Fix it by quickly scaling down the readahead size.
2314  */
shrink_readahead_size_eio(struct file_ra_state * ra)2315 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2316 {
2317 	ra->ra_pages /= 4;
2318 }
2319 
2320 /*
2321  * filemap_get_read_batch - Get a batch of folios for read
2322  *
2323  * Get a batch of folios which represent a contiguous range of bytes in
2324  * the file.  No exceptional entries will be returned.  If @index is in
2325  * the middle of a folio, the entire folio will be returned.  The last
2326  * folio in the batch may have the readahead flag set or the uptodate flag
2327  * clear so that the caller can take the appropriate action.
2328  */
filemap_get_read_batch(struct address_space * mapping,pgoff_t index,pgoff_t max,struct folio_batch * fbatch)2329 static void filemap_get_read_batch(struct address_space *mapping,
2330 		pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2331 {
2332 	XA_STATE(xas, &mapping->i_pages, index);
2333 	struct folio *folio;
2334 
2335 	rcu_read_lock();
2336 	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2337 		if (xas_retry(&xas, folio))
2338 			continue;
2339 		if (xas.xa_index > max || xa_is_value(folio))
2340 			break;
2341 		if (xa_is_sibling(folio))
2342 			break;
2343 		if (!folio_try_get_rcu(folio))
2344 			goto retry;
2345 
2346 		if (unlikely(folio != xas_reload(&xas)))
2347 			goto put_folio;
2348 
2349 		if (!folio_batch_add(fbatch, folio))
2350 			break;
2351 		if (!folio_test_uptodate(folio))
2352 			break;
2353 		if (folio_test_readahead(folio))
2354 			break;
2355 		xas_advance(&xas, folio_next_index(folio) - 1);
2356 		continue;
2357 put_folio:
2358 		folio_put(folio);
2359 retry:
2360 		xas_reset(&xas);
2361 	}
2362 	rcu_read_unlock();
2363 }
2364 
filemap_read_folio(struct file * file,filler_t filler,struct folio * folio)2365 static int filemap_read_folio(struct file *file, filler_t filler,
2366 		struct folio *folio)
2367 {
2368 	bool workingset = folio_test_workingset(folio);
2369 	unsigned long pflags;
2370 	int error;
2371 
2372 	/*
2373 	 * A previous I/O error may have been due to temporary failures,
2374 	 * eg. multipath errors.  PG_error will be set again if read_folio
2375 	 * fails.
2376 	 */
2377 	folio_clear_error(folio);
2378 
2379 	/* Start the actual read. The read will unlock the page. */
2380 	if (unlikely(workingset))
2381 		psi_memstall_enter(&pflags);
2382 	error = filler(file, folio);
2383 	if (unlikely(workingset))
2384 		psi_memstall_leave(&pflags);
2385 	if (error)
2386 		return error;
2387 
2388 	error = folio_wait_locked_killable(folio);
2389 	if (error)
2390 		return error;
2391 	if (folio_test_uptodate(folio))
2392 		return 0;
2393 	if (file)
2394 		shrink_readahead_size_eio(&file->f_ra);
2395 	return -EIO;
2396 }
2397 
filemap_range_uptodate(struct address_space * mapping,loff_t pos,size_t count,struct folio * folio,bool need_uptodate)2398 static bool filemap_range_uptodate(struct address_space *mapping,
2399 		loff_t pos, size_t count, struct folio *folio,
2400 		bool need_uptodate)
2401 {
2402 	if (folio_test_uptodate(folio))
2403 		return true;
2404 	/* pipes can't handle partially uptodate pages */
2405 	if (need_uptodate)
2406 		return false;
2407 	if (!mapping->a_ops->is_partially_uptodate)
2408 		return false;
2409 	if (mapping->host->i_blkbits >= folio_shift(folio))
2410 		return false;
2411 
2412 	if (folio_pos(folio) > pos) {
2413 		count -= folio_pos(folio) - pos;
2414 		pos = 0;
2415 	} else {
2416 		pos -= folio_pos(folio);
2417 	}
2418 
2419 	return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2420 }
2421 
filemap_update_page(struct kiocb * iocb,struct address_space * mapping,size_t count,struct folio * folio,bool need_uptodate)2422 static int filemap_update_page(struct kiocb *iocb,
2423 		struct address_space *mapping, size_t count,
2424 		struct folio *folio, bool need_uptodate)
2425 {
2426 	int error;
2427 
2428 	if (iocb->ki_flags & IOCB_NOWAIT) {
2429 		if (!filemap_invalidate_trylock_shared(mapping))
2430 			return -EAGAIN;
2431 	} else {
2432 		filemap_invalidate_lock_shared(mapping);
2433 	}
2434 
2435 	if (!folio_trylock(folio)) {
2436 		error = -EAGAIN;
2437 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2438 			goto unlock_mapping;
2439 		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2440 			filemap_invalidate_unlock_shared(mapping);
2441 			/*
2442 			 * This is where we usually end up waiting for a
2443 			 * previously submitted readahead to finish.
2444 			 */
2445 			folio_put_wait_locked(folio, TASK_KILLABLE);
2446 			return AOP_TRUNCATED_PAGE;
2447 		}
2448 		error = __folio_lock_async(folio, iocb->ki_waitq);
2449 		if (error)
2450 			goto unlock_mapping;
2451 	}
2452 
2453 	error = AOP_TRUNCATED_PAGE;
2454 	if (!folio->mapping)
2455 		goto unlock;
2456 
2457 	error = 0;
2458 	if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2459 				   need_uptodate))
2460 		goto unlock;
2461 
2462 	error = -EAGAIN;
2463 	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2464 		goto unlock;
2465 
2466 	error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2467 			folio);
2468 	goto unlock_mapping;
2469 unlock:
2470 	folio_unlock(folio);
2471 unlock_mapping:
2472 	filemap_invalidate_unlock_shared(mapping);
2473 	if (error == AOP_TRUNCATED_PAGE)
2474 		folio_put(folio);
2475 	return error;
2476 }
2477 
filemap_create_folio(struct file * file,struct address_space * mapping,pgoff_t index,struct folio_batch * fbatch)2478 static int filemap_create_folio(struct file *file,
2479 		struct address_space *mapping, pgoff_t index,
2480 		struct folio_batch *fbatch)
2481 {
2482 	struct folio *folio;
2483 	int error;
2484 
2485 	folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2486 	if (!folio)
2487 		return -ENOMEM;
2488 
2489 	/*
2490 	 * Protect against truncate / hole punch. Grabbing invalidate_lock
2491 	 * here assures we cannot instantiate and bring uptodate new
2492 	 * pagecache folios after evicting page cache during truncate
2493 	 * and before actually freeing blocks.	Note that we could
2494 	 * release invalidate_lock after inserting the folio into
2495 	 * the page cache as the locked folio would then be enough to
2496 	 * synchronize with hole punching. But there are code paths
2497 	 * such as filemap_update_page() filling in partially uptodate
2498 	 * pages or ->readahead() that need to hold invalidate_lock
2499 	 * while mapping blocks for IO so let's hold the lock here as
2500 	 * well to keep locking rules simple.
2501 	 */
2502 	filemap_invalidate_lock_shared(mapping);
2503 	error = filemap_add_folio(mapping, folio, index,
2504 			mapping_gfp_constraint(mapping, GFP_KERNEL));
2505 	if (error == -EEXIST)
2506 		error = AOP_TRUNCATED_PAGE;
2507 	if (error)
2508 		goto error;
2509 
2510 	error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2511 	if (error)
2512 		goto error;
2513 
2514 	filemap_invalidate_unlock_shared(mapping);
2515 	folio_batch_add(fbatch, folio);
2516 	return 0;
2517 error:
2518 	filemap_invalidate_unlock_shared(mapping);
2519 	folio_put(folio);
2520 	return error;
2521 }
2522 
filemap_readahead(struct kiocb * iocb,struct file * file,struct address_space * mapping,struct folio * folio,pgoff_t last_index)2523 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2524 		struct address_space *mapping, struct folio *folio,
2525 		pgoff_t last_index)
2526 {
2527 	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2528 
2529 	if (iocb->ki_flags & IOCB_NOIO)
2530 		return -EAGAIN;
2531 	page_cache_async_ra(&ractl, folio, last_index - folio->index);
2532 	return 0;
2533 }
2534 
filemap_get_pages(struct kiocb * iocb,size_t count,struct folio_batch * fbatch,bool need_uptodate)2535 static int filemap_get_pages(struct kiocb *iocb, size_t count,
2536 		struct folio_batch *fbatch, bool need_uptodate)
2537 {
2538 	struct file *filp = iocb->ki_filp;
2539 	struct address_space *mapping = filp->f_mapping;
2540 	struct file_ra_state *ra = &filp->f_ra;
2541 	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2542 	pgoff_t last_index;
2543 	struct folio *folio;
2544 	int err = 0;
2545 
2546 	/* "last_index" is the index of the page beyond the end of the read */
2547 	last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2548 retry:
2549 	if (fatal_signal_pending(current))
2550 		return -EINTR;
2551 
2552 	filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2553 	if (!folio_batch_count(fbatch)) {
2554 		if (iocb->ki_flags & IOCB_NOIO)
2555 			return -EAGAIN;
2556 		page_cache_sync_readahead(mapping, ra, filp, index,
2557 				last_index - index);
2558 		filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2559 	}
2560 	if (!folio_batch_count(fbatch)) {
2561 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2562 			return -EAGAIN;
2563 		err = filemap_create_folio(filp, mapping,
2564 				iocb->ki_pos >> PAGE_SHIFT, fbatch);
2565 		if (err == AOP_TRUNCATED_PAGE)
2566 			goto retry;
2567 		return err;
2568 	}
2569 
2570 	folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2571 	if (folio_test_readahead(folio)) {
2572 		err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2573 		if (err)
2574 			goto err;
2575 	}
2576 	if (!folio_test_uptodate(folio)) {
2577 		if ((iocb->ki_flags & IOCB_WAITQ) &&
2578 		    folio_batch_count(fbatch) > 1)
2579 			iocb->ki_flags |= IOCB_NOWAIT;
2580 		err = filemap_update_page(iocb, mapping, count, folio,
2581 					  need_uptodate);
2582 		if (err)
2583 			goto err;
2584 	}
2585 
2586 	return 0;
2587 err:
2588 	if (err < 0)
2589 		folio_put(folio);
2590 	if (likely(--fbatch->nr))
2591 		return 0;
2592 	if (err == AOP_TRUNCATED_PAGE)
2593 		goto retry;
2594 	return err;
2595 }
2596 
pos_same_folio(loff_t pos1,loff_t pos2,struct folio * folio)2597 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2598 {
2599 	unsigned int shift = folio_shift(folio);
2600 
2601 	return (pos1 >> shift == pos2 >> shift);
2602 }
2603 
2604 /**
2605  * filemap_read - Read data from the page cache.
2606  * @iocb: The iocb to read.
2607  * @iter: Destination for the data.
2608  * @already_read: Number of bytes already read by the caller.
2609  *
2610  * Copies data from the page cache.  If the data is not currently present,
2611  * uses the readahead and read_folio address_space operations to fetch it.
2612  *
2613  * Return: Total number of bytes copied, including those already read by
2614  * the caller.  If an error happens before any bytes are copied, returns
2615  * a negative error number.
2616  */
filemap_read(struct kiocb * iocb,struct iov_iter * iter,ssize_t already_read)2617 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2618 		ssize_t already_read)
2619 {
2620 	struct file *filp = iocb->ki_filp;
2621 	struct file_ra_state *ra = &filp->f_ra;
2622 	struct address_space *mapping = filp->f_mapping;
2623 	struct inode *inode = mapping->host;
2624 	struct folio_batch fbatch;
2625 	int i, error = 0;
2626 	bool writably_mapped;
2627 	loff_t isize, end_offset;
2628 	loff_t last_pos = ra->prev_pos;
2629 
2630 	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2631 		return 0;
2632 	if (unlikely(!iov_iter_count(iter)))
2633 		return 0;
2634 
2635 	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2636 	folio_batch_init(&fbatch);
2637 
2638 	do {
2639 		cond_resched();
2640 
2641 		/*
2642 		 * If we've already successfully copied some data, then we
2643 		 * can no longer safely return -EIOCBQUEUED. Hence mark
2644 		 * an async read NOWAIT at that point.
2645 		 */
2646 		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2647 			iocb->ki_flags |= IOCB_NOWAIT;
2648 
2649 		if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2650 			break;
2651 
2652 		error = filemap_get_pages(iocb, iter->count, &fbatch, false);
2653 		if (error < 0)
2654 			break;
2655 
2656 		/*
2657 		 * i_size must be checked after we know the pages are Uptodate.
2658 		 *
2659 		 * Checking i_size after the check allows us to calculate
2660 		 * the correct value for "nr", which means the zero-filled
2661 		 * part of the page is not copied back to userspace (unless
2662 		 * another truncate extends the file - this is desired though).
2663 		 */
2664 		isize = i_size_read(inode);
2665 		if (unlikely(iocb->ki_pos >= isize))
2666 			goto put_folios;
2667 		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2668 
2669 		/*
2670 		 * Pairs with a barrier in
2671 		 * block_write_end()->mark_buffer_dirty() or other page
2672 		 * dirtying routines like iomap_write_end() to ensure
2673 		 * changes to page contents are visible before we see
2674 		 * increased inode size.
2675 		 */
2676 		smp_rmb();
2677 
2678 		/*
2679 		 * Once we start copying data, we don't want to be touching any
2680 		 * cachelines that might be contended:
2681 		 */
2682 		writably_mapped = mapping_writably_mapped(mapping);
2683 
2684 		/*
2685 		 * When a read accesses the same folio several times, only
2686 		 * mark it as accessed the first time.
2687 		 */
2688 		if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
2689 				    fbatch.folios[0]))
2690 			folio_mark_accessed(fbatch.folios[0]);
2691 
2692 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2693 			struct folio *folio = fbatch.folios[i];
2694 			size_t fsize = folio_size(folio);
2695 			size_t offset = iocb->ki_pos & (fsize - 1);
2696 			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2697 					     fsize - offset);
2698 			size_t copied;
2699 
2700 			if (end_offset < folio_pos(folio))
2701 				break;
2702 			if (i > 0)
2703 				folio_mark_accessed(folio);
2704 			/*
2705 			 * If users can be writing to this folio using arbitrary
2706 			 * virtual addresses, take care of potential aliasing
2707 			 * before reading the folio on the kernel side.
2708 			 */
2709 			if (writably_mapped)
2710 				flush_dcache_folio(folio);
2711 
2712 			copied = copy_folio_to_iter(folio, offset, bytes, iter);
2713 
2714 			already_read += copied;
2715 			iocb->ki_pos += copied;
2716 			last_pos = iocb->ki_pos;
2717 
2718 			if (copied < bytes) {
2719 				error = -EFAULT;
2720 				break;
2721 			}
2722 		}
2723 put_folios:
2724 		for (i = 0; i < folio_batch_count(&fbatch); i++)
2725 			folio_put(fbatch.folios[i]);
2726 		folio_batch_init(&fbatch);
2727 	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2728 
2729 	file_accessed(filp);
2730 	ra->prev_pos = last_pos;
2731 	return already_read ? already_read : error;
2732 }
2733 EXPORT_SYMBOL_GPL(filemap_read);
2734 
kiocb_write_and_wait(struct kiocb * iocb,size_t count)2735 int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2736 {
2737 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2738 	loff_t pos = iocb->ki_pos;
2739 	loff_t end = pos + count - 1;
2740 
2741 	if (iocb->ki_flags & IOCB_NOWAIT) {
2742 		if (filemap_range_needs_writeback(mapping, pos, end))
2743 			return -EAGAIN;
2744 		return 0;
2745 	}
2746 
2747 	return filemap_write_and_wait_range(mapping, pos, end);
2748 }
2749 
kiocb_invalidate_pages(struct kiocb * iocb,size_t count)2750 int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2751 {
2752 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2753 	loff_t pos = iocb->ki_pos;
2754 	loff_t end = pos + count - 1;
2755 	int ret;
2756 
2757 	if (iocb->ki_flags & IOCB_NOWAIT) {
2758 		/* we could block if there are any pages in the range */
2759 		if (filemap_range_has_page(mapping, pos, end))
2760 			return -EAGAIN;
2761 	} else {
2762 		ret = filemap_write_and_wait_range(mapping, pos, end);
2763 		if (ret)
2764 			return ret;
2765 	}
2766 
2767 	/*
2768 	 * After a write we want buffered reads to be sure to go to disk to get
2769 	 * the new data.  We invalidate clean cached page from the region we're
2770 	 * about to write.  We do this *before* the write so that we can return
2771 	 * without clobbering -EIOCBQUEUED from ->direct_IO().
2772 	 */
2773 	return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2774 					     end >> PAGE_SHIFT);
2775 }
2776 
2777 /**
2778  * generic_file_read_iter - generic filesystem read routine
2779  * @iocb:	kernel I/O control block
2780  * @iter:	destination for the data read
2781  *
2782  * This is the "read_iter()" routine for all filesystems
2783  * that can use the page cache directly.
2784  *
2785  * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2786  * be returned when no data can be read without waiting for I/O requests
2787  * to complete; it doesn't prevent readahead.
2788  *
2789  * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2790  * requests shall be made for the read or for readahead.  When no data
2791  * can be read, -EAGAIN shall be returned.  When readahead would be
2792  * triggered, a partial, possibly empty read shall be returned.
2793  *
2794  * Return:
2795  * * number of bytes copied, even for partial reads
2796  * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2797  */
2798 ssize_t
generic_file_read_iter(struct kiocb * iocb,struct iov_iter * iter)2799 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2800 {
2801 	size_t count = iov_iter_count(iter);
2802 	ssize_t retval = 0;
2803 
2804 	if (!count)
2805 		return 0; /* skip atime */
2806 
2807 	if (iocb->ki_flags & IOCB_DIRECT) {
2808 		struct file *file = iocb->ki_filp;
2809 		struct address_space *mapping = file->f_mapping;
2810 		struct inode *inode = mapping->host;
2811 
2812 		retval = kiocb_write_and_wait(iocb, count);
2813 		if (retval < 0)
2814 			return retval;
2815 		file_accessed(file);
2816 
2817 		retval = mapping->a_ops->direct_IO(iocb, iter);
2818 		if (retval >= 0) {
2819 			iocb->ki_pos += retval;
2820 			count -= retval;
2821 		}
2822 		if (retval != -EIOCBQUEUED)
2823 			iov_iter_revert(iter, count - iov_iter_count(iter));
2824 
2825 		/*
2826 		 * Btrfs can have a short DIO read if we encounter
2827 		 * compressed extents, so if there was an error, or if
2828 		 * we've already read everything we wanted to, or if
2829 		 * there was a short read because we hit EOF, go ahead
2830 		 * and return.  Otherwise fallthrough to buffered io for
2831 		 * the rest of the read.  Buffered reads will not work for
2832 		 * DAX files, so don't bother trying.
2833 		 */
2834 		if (retval < 0 || !count || IS_DAX(inode))
2835 			return retval;
2836 		if (iocb->ki_pos >= i_size_read(inode))
2837 			return retval;
2838 	}
2839 
2840 	return filemap_read(iocb, iter, retval);
2841 }
2842 EXPORT_SYMBOL(generic_file_read_iter);
2843 
2844 /*
2845  * Splice subpages from a folio into a pipe.
2846  */
splice_folio_into_pipe(struct pipe_inode_info * pipe,struct folio * folio,loff_t fpos,size_t size)2847 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2848 			      struct folio *folio, loff_t fpos, size_t size)
2849 {
2850 	struct page *page;
2851 	size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2852 
2853 	page = folio_page(folio, offset / PAGE_SIZE);
2854 	size = min(size, folio_size(folio) - offset);
2855 	offset %= PAGE_SIZE;
2856 
2857 	while (spliced < size &&
2858 	       !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2859 		struct pipe_buffer *buf = pipe_head_buf(pipe);
2860 		size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2861 
2862 		*buf = (struct pipe_buffer) {
2863 			.ops	= &page_cache_pipe_buf_ops,
2864 			.page	= page,
2865 			.offset	= offset,
2866 			.len	= part,
2867 		};
2868 		folio_get(folio);
2869 		pipe->head++;
2870 		page++;
2871 		spliced += part;
2872 		offset = 0;
2873 	}
2874 
2875 	return spliced;
2876 }
2877 
2878 /**
2879  * filemap_splice_read -  Splice data from a file's pagecache into a pipe
2880  * @in: The file to read from
2881  * @ppos: Pointer to the file position to read from
2882  * @pipe: The pipe to splice into
2883  * @len: The amount to splice
2884  * @flags: The SPLICE_F_* flags
2885  *
2886  * This function gets folios from a file's pagecache and splices them into the
2887  * pipe.  Readahead will be called as necessary to fill more folios.  This may
2888  * be used for blockdevs also.
2889  *
2890  * Return: On success, the number of bytes read will be returned and *@ppos
2891  * will be updated if appropriate; 0 will be returned if there is no more data
2892  * to be read; -EAGAIN will be returned if the pipe had no space, and some
2893  * other negative error code will be returned on error.  A short read may occur
2894  * if the pipe has insufficient space, we reach the end of the data or we hit a
2895  * hole.
2896  */
filemap_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)2897 ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2898 			    struct pipe_inode_info *pipe,
2899 			    size_t len, unsigned int flags)
2900 {
2901 	struct folio_batch fbatch;
2902 	struct kiocb iocb;
2903 	size_t total_spliced = 0, used, npages;
2904 	loff_t isize, end_offset;
2905 	bool writably_mapped;
2906 	int i, error = 0;
2907 
2908 	if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
2909 		return 0;
2910 
2911 	init_sync_kiocb(&iocb, in);
2912 	iocb.ki_pos = *ppos;
2913 
2914 	/* Work out how much data we can actually add into the pipe */
2915 	used = pipe_occupancy(pipe->head, pipe->tail);
2916 	npages = max_t(ssize_t, pipe->max_usage - used, 0);
2917 	len = min_t(size_t, len, npages * PAGE_SIZE);
2918 
2919 	folio_batch_init(&fbatch);
2920 
2921 	do {
2922 		cond_resched();
2923 
2924 		if (*ppos >= i_size_read(in->f_mapping->host))
2925 			break;
2926 
2927 		iocb.ki_pos = *ppos;
2928 		error = filemap_get_pages(&iocb, len, &fbatch, true);
2929 		if (error < 0)
2930 			break;
2931 
2932 		/*
2933 		 * i_size must be checked after we know the pages are Uptodate.
2934 		 *
2935 		 * Checking i_size after the check allows us to calculate
2936 		 * the correct value for "nr", which means the zero-filled
2937 		 * part of the page is not copied back to userspace (unless
2938 		 * another truncate extends the file - this is desired though).
2939 		 */
2940 		isize = i_size_read(in->f_mapping->host);
2941 		if (unlikely(*ppos >= isize))
2942 			break;
2943 		end_offset = min_t(loff_t, isize, *ppos + len);
2944 
2945 		/*
2946 		 * Once we start copying data, we don't want to be touching any
2947 		 * cachelines that might be contended:
2948 		 */
2949 		writably_mapped = mapping_writably_mapped(in->f_mapping);
2950 
2951 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2952 			struct folio *folio = fbatch.folios[i];
2953 			size_t n;
2954 
2955 			if (folio_pos(folio) >= end_offset)
2956 				goto out;
2957 			folio_mark_accessed(folio);
2958 
2959 			/*
2960 			 * If users can be writing to this folio using arbitrary
2961 			 * virtual addresses, take care of potential aliasing
2962 			 * before reading the folio on the kernel side.
2963 			 */
2964 			if (writably_mapped)
2965 				flush_dcache_folio(folio);
2966 
2967 			n = min_t(loff_t, len, isize - *ppos);
2968 			n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2969 			if (!n)
2970 				goto out;
2971 			len -= n;
2972 			total_spliced += n;
2973 			*ppos += n;
2974 			in->f_ra.prev_pos = *ppos;
2975 			if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2976 				goto out;
2977 		}
2978 
2979 		folio_batch_release(&fbatch);
2980 	} while (len);
2981 
2982 out:
2983 	folio_batch_release(&fbatch);
2984 	file_accessed(in);
2985 
2986 	return total_spliced ? total_spliced : error;
2987 }
2988 EXPORT_SYMBOL(filemap_splice_read);
2989 
folio_seek_hole_data(struct xa_state * xas,struct address_space * mapping,struct folio * folio,loff_t start,loff_t end,bool seek_data)2990 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2991 		struct address_space *mapping, struct folio *folio,
2992 		loff_t start, loff_t end, bool seek_data)
2993 {
2994 	const struct address_space_operations *ops = mapping->a_ops;
2995 	size_t offset, bsz = i_blocksize(mapping->host);
2996 
2997 	if (xa_is_value(folio) || folio_test_uptodate(folio))
2998 		return seek_data ? start : end;
2999 	if (!ops->is_partially_uptodate)
3000 		return seek_data ? end : start;
3001 
3002 	xas_pause(xas);
3003 	rcu_read_unlock();
3004 	folio_lock(folio);
3005 	if (unlikely(folio->mapping != mapping))
3006 		goto unlock;
3007 
3008 	offset = offset_in_folio(folio, start) & ~(bsz - 1);
3009 
3010 	do {
3011 		if (ops->is_partially_uptodate(folio, offset, bsz) ==
3012 							seek_data)
3013 			break;
3014 		start = (start + bsz) & ~(bsz - 1);
3015 		offset += bsz;
3016 	} while (offset < folio_size(folio));
3017 unlock:
3018 	folio_unlock(folio);
3019 	rcu_read_lock();
3020 	return start;
3021 }
3022 
seek_folio_size(struct xa_state * xas,struct folio * folio)3023 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3024 {
3025 	if (xa_is_value(folio))
3026 		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3027 	return folio_size(folio);
3028 }
3029 
3030 /**
3031  * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3032  * @mapping: Address space to search.
3033  * @start: First byte to consider.
3034  * @end: Limit of search (exclusive).
3035  * @whence: Either SEEK_HOLE or SEEK_DATA.
3036  *
3037  * If the page cache knows which blocks contain holes and which blocks
3038  * contain data, your filesystem can use this function to implement
3039  * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
3040  * entirely memory-based such as tmpfs, and filesystems which support
3041  * unwritten extents.
3042  *
3043  * Return: The requested offset on success, or -ENXIO if @whence specifies
3044  * SEEK_DATA and there is no data after @start.  There is an implicit hole
3045  * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3046  * and @end contain data.
3047  */
mapping_seek_hole_data(struct address_space * mapping,loff_t start,loff_t end,int whence)3048 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3049 		loff_t end, int whence)
3050 {
3051 	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3052 	pgoff_t max = (end - 1) >> PAGE_SHIFT;
3053 	bool seek_data = (whence == SEEK_DATA);
3054 	struct folio *folio;
3055 
3056 	if (end <= start)
3057 		return -ENXIO;
3058 
3059 	rcu_read_lock();
3060 	while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3061 		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3062 		size_t seek_size;
3063 
3064 		if (start < pos) {
3065 			if (!seek_data)
3066 				goto unlock;
3067 			start = pos;
3068 		}
3069 
3070 		seek_size = seek_folio_size(&xas, folio);
3071 		pos = round_up((u64)pos + 1, seek_size);
3072 		start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3073 				seek_data);
3074 		if (start < pos)
3075 			goto unlock;
3076 		if (start >= end)
3077 			break;
3078 		if (seek_size > PAGE_SIZE)
3079 			xas_set(&xas, pos >> PAGE_SHIFT);
3080 		if (!xa_is_value(folio))
3081 			folio_put(folio);
3082 	}
3083 	if (seek_data)
3084 		start = -ENXIO;
3085 unlock:
3086 	rcu_read_unlock();
3087 	if (folio && !xa_is_value(folio))
3088 		folio_put(folio);
3089 	if (start > end)
3090 		return end;
3091 	return start;
3092 }
3093 
3094 #ifdef CONFIG_MMU
3095 #define MMAP_LOTSAMISS  (100)
3096 /*
3097  * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3098  * @vmf - the vm_fault for this fault.
3099  * @folio - the folio to lock.
3100  * @fpin - the pointer to the file we may pin (or is already pinned).
3101  *
3102  * This works similar to lock_folio_or_retry in that it can drop the
3103  * mmap_lock.  It differs in that it actually returns the folio locked
3104  * if it returns 1 and 0 if it couldn't lock the folio.  If we did have
3105  * to drop the mmap_lock then fpin will point to the pinned file and
3106  * needs to be fput()'ed at a later point.
3107  */
lock_folio_maybe_drop_mmap(struct vm_fault * vmf,struct folio * folio,struct file ** fpin)3108 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3109 				     struct file **fpin)
3110 {
3111 	if (folio_trylock(folio))
3112 		return 1;
3113 
3114 	/*
3115 	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3116 	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3117 	 * is supposed to work. We have way too many special cases..
3118 	 */
3119 	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3120 		return 0;
3121 
3122 	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3123 	if (vmf->flags & FAULT_FLAG_KILLABLE) {
3124 		if (__folio_lock_killable(folio)) {
3125 			/*
3126 			 * We didn't have the right flags to drop the mmap_lock,
3127 			 * but all fault_handlers only check for fatal signals
3128 			 * if we return VM_FAULT_RETRY, so we need to drop the
3129 			 * mmap_lock here and return 0 if we don't have a fpin.
3130 			 */
3131 			if (*fpin == NULL)
3132 				mmap_read_unlock(vmf->vma->vm_mm);
3133 			return 0;
3134 		}
3135 	} else
3136 		__folio_lock(folio);
3137 
3138 	return 1;
3139 }
3140 
3141 /*
3142  * Synchronous readahead happens when we don't even find a page in the page
3143  * cache at all.  We don't want to perform IO under the mmap sem, so if we have
3144  * to drop the mmap sem we return the file that was pinned in order for us to do
3145  * that.  If we didn't pin a file then we return NULL.  The file that is
3146  * returned needs to be fput()'ed when we're done with it.
3147  */
do_sync_mmap_readahead(struct vm_fault * vmf)3148 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3149 {
3150 	struct file *file = vmf->vma->vm_file;
3151 	struct file_ra_state *ra = &file->f_ra;
3152 	struct address_space *mapping = file->f_mapping;
3153 	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3154 	struct file *fpin = NULL;
3155 	unsigned long vm_flags = vmf->vma->vm_flags;
3156 	unsigned int mmap_miss;
3157 
3158 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3159 	/* Use the readahead code, even if readahead is disabled */
3160 	if (vm_flags & VM_HUGEPAGE) {
3161 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3162 		ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3163 		ra->size = HPAGE_PMD_NR;
3164 		/*
3165 		 * Fetch two PMD folios, so we get the chance to actually
3166 		 * readahead, unless we've been told not to.
3167 		 */
3168 		if (!(vm_flags & VM_RAND_READ))
3169 			ra->size *= 2;
3170 		ra->async_size = HPAGE_PMD_NR;
3171 		page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3172 		return fpin;
3173 	}
3174 #endif
3175 
3176 	/* If we don't want any read-ahead, don't bother */
3177 	if (vm_flags & VM_RAND_READ)
3178 		return fpin;
3179 	if (!ra->ra_pages)
3180 		return fpin;
3181 
3182 	if (vm_flags & VM_SEQ_READ) {
3183 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3184 		page_cache_sync_ra(&ractl, ra->ra_pages);
3185 		return fpin;
3186 	}
3187 
3188 	/* Avoid banging the cache line if not needed */
3189 	mmap_miss = READ_ONCE(ra->mmap_miss);
3190 	if (mmap_miss < MMAP_LOTSAMISS * 10)
3191 		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3192 
3193 	/*
3194 	 * Do we miss much more than hit in this file? If so,
3195 	 * stop bothering with read-ahead. It will only hurt.
3196 	 */
3197 	if (mmap_miss > MMAP_LOTSAMISS)
3198 		return fpin;
3199 
3200 	/*
3201 	 * mmap read-around
3202 	 */
3203 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3204 	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3205 	ra->size = ra->ra_pages;
3206 	ra->async_size = ra->ra_pages / 4;
3207 	ractl._index = ra->start;
3208 	page_cache_ra_order(&ractl, ra, 0);
3209 	return fpin;
3210 }
3211 
3212 /*
3213  * Asynchronous readahead happens when we find the page and PG_readahead,
3214  * so we want to possibly extend the readahead further.  We return the file that
3215  * was pinned if we have to drop the mmap_lock in order to do IO.
3216  */
do_async_mmap_readahead(struct vm_fault * vmf,struct folio * folio)3217 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3218 					    struct folio *folio)
3219 {
3220 	struct file *file = vmf->vma->vm_file;
3221 	struct file_ra_state *ra = &file->f_ra;
3222 	DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3223 	struct file *fpin = NULL;
3224 	unsigned int mmap_miss;
3225 
3226 	/* If we don't want any read-ahead, don't bother */
3227 	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3228 		return fpin;
3229 
3230 	mmap_miss = READ_ONCE(ra->mmap_miss);
3231 	if (mmap_miss)
3232 		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3233 
3234 	if (folio_test_readahead(folio)) {
3235 		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3236 		page_cache_async_ra(&ractl, folio, ra->ra_pages);
3237 	}
3238 	return fpin;
3239 }
3240 
3241 /**
3242  * filemap_fault - read in file data for page fault handling
3243  * @vmf:	struct vm_fault containing details of the fault
3244  *
3245  * filemap_fault() is invoked via the vma operations vector for a
3246  * mapped memory region to read in file data during a page fault.
3247  *
3248  * The goto's are kind of ugly, but this streamlines the normal case of having
3249  * it in the page cache, and handles the special cases reasonably without
3250  * having a lot of duplicated code.
3251  *
3252  * vma->vm_mm->mmap_lock must be held on entry.
3253  *
3254  * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3255  * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3256  *
3257  * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3258  * has not been released.
3259  *
3260  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3261  *
3262  * Return: bitwise-OR of %VM_FAULT_ codes.
3263  */
filemap_fault(struct vm_fault * vmf)3264 vm_fault_t filemap_fault(struct vm_fault *vmf)
3265 {
3266 	int error;
3267 	struct file *file = vmf->vma->vm_file;
3268 	struct file *fpin = NULL;
3269 	struct address_space *mapping = file->f_mapping;
3270 	struct inode *inode = mapping->host;
3271 	pgoff_t max_idx, index = vmf->pgoff;
3272 	struct folio *folio;
3273 	vm_fault_t ret = 0;
3274 	bool mapping_locked = false;
3275 
3276 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3277 	if (unlikely(index >= max_idx))
3278 		return VM_FAULT_SIGBUS;
3279 
3280 	/*
3281 	 * Do we have something in the page cache already?
3282 	 */
3283 	folio = filemap_get_folio(mapping, index);
3284 	if (likely(!IS_ERR(folio))) {
3285 		/*
3286 		 * We found the page, so try async readahead before waiting for
3287 		 * the lock.
3288 		 */
3289 		if (!(vmf->flags & FAULT_FLAG_TRIED))
3290 			fpin = do_async_mmap_readahead(vmf, folio);
3291 		if (unlikely(!folio_test_uptodate(folio))) {
3292 			filemap_invalidate_lock_shared(mapping);
3293 			mapping_locked = true;
3294 		}
3295 	} else {
3296 		/* No page in the page cache at all */
3297 		count_vm_event(PGMAJFAULT);
3298 		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3299 		ret = VM_FAULT_MAJOR;
3300 		fpin = do_sync_mmap_readahead(vmf);
3301 retry_find:
3302 		/*
3303 		 * See comment in filemap_create_folio() why we need
3304 		 * invalidate_lock
3305 		 */
3306 		if (!mapping_locked) {
3307 			filemap_invalidate_lock_shared(mapping);
3308 			mapping_locked = true;
3309 		}
3310 		folio = __filemap_get_folio(mapping, index,
3311 					  FGP_CREAT|FGP_FOR_MMAP,
3312 					  vmf->gfp_mask);
3313 		if (IS_ERR(folio)) {
3314 			if (fpin)
3315 				goto out_retry;
3316 			filemap_invalidate_unlock_shared(mapping);
3317 			return VM_FAULT_OOM;
3318 		}
3319 	}
3320 
3321 	if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3322 		goto out_retry;
3323 
3324 	/* Did it get truncated? */
3325 	if (unlikely(folio->mapping != mapping)) {
3326 		folio_unlock(folio);
3327 		folio_put(folio);
3328 		goto retry_find;
3329 	}
3330 	VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3331 
3332 	/*
3333 	 * We have a locked page in the page cache, now we need to check
3334 	 * that it's up-to-date. If not, it is going to be due to an error.
3335 	 */
3336 	if (unlikely(!folio_test_uptodate(folio))) {
3337 		/*
3338 		 * The page was in cache and uptodate and now it is not.
3339 		 * Strange but possible since we didn't hold the page lock all
3340 		 * the time. Let's drop everything get the invalidate lock and
3341 		 * try again.
3342 		 */
3343 		if (!mapping_locked) {
3344 			folio_unlock(folio);
3345 			folio_put(folio);
3346 			goto retry_find;
3347 		}
3348 		goto page_not_uptodate;
3349 	}
3350 
3351 	/*
3352 	 * We've made it this far and we had to drop our mmap_lock, now is the
3353 	 * time to return to the upper layer and have it re-find the vma and
3354 	 * redo the fault.
3355 	 */
3356 	if (fpin) {
3357 		folio_unlock(folio);
3358 		goto out_retry;
3359 	}
3360 	if (mapping_locked)
3361 		filemap_invalidate_unlock_shared(mapping);
3362 
3363 	/*
3364 	 * Found the page and have a reference on it.
3365 	 * We must recheck i_size under page lock.
3366 	 */
3367 	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3368 	if (unlikely(index >= max_idx)) {
3369 		folio_unlock(folio);
3370 		folio_put(folio);
3371 		return VM_FAULT_SIGBUS;
3372 	}
3373 
3374 	vmf->page = folio_file_page(folio, index);
3375 	return ret | VM_FAULT_LOCKED;
3376 
3377 page_not_uptodate:
3378 	/*
3379 	 * Umm, take care of errors if the page isn't up-to-date.
3380 	 * Try to re-read it _once_. We do this synchronously,
3381 	 * because there really aren't any performance issues here
3382 	 * and we need to check for errors.
3383 	 */
3384 	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3385 	error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3386 	if (fpin)
3387 		goto out_retry;
3388 	folio_put(folio);
3389 
3390 	if (!error || error == AOP_TRUNCATED_PAGE)
3391 		goto retry_find;
3392 	filemap_invalidate_unlock_shared(mapping);
3393 
3394 	return VM_FAULT_SIGBUS;
3395 
3396 out_retry:
3397 	/*
3398 	 * We dropped the mmap_lock, we need to return to the fault handler to
3399 	 * re-find the vma and come back and find our hopefully still populated
3400 	 * page.
3401 	 */
3402 	if (!IS_ERR(folio))
3403 		folio_put(folio);
3404 	if (mapping_locked)
3405 		filemap_invalidate_unlock_shared(mapping);
3406 	if (fpin)
3407 		fput(fpin);
3408 	return ret | VM_FAULT_RETRY;
3409 }
3410 EXPORT_SYMBOL(filemap_fault);
3411 
filemap_map_pmd(struct vm_fault * vmf,struct folio * folio,pgoff_t start)3412 static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3413 		pgoff_t start)
3414 {
3415 	struct mm_struct *mm = vmf->vma->vm_mm;
3416 
3417 	/* Huge page is mapped? No need to proceed. */
3418 	if (pmd_trans_huge(*vmf->pmd)) {
3419 		folio_unlock(folio);
3420 		folio_put(folio);
3421 		return true;
3422 	}
3423 
3424 	if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3425 		struct page *page = folio_file_page(folio, start);
3426 		vm_fault_t ret = do_set_pmd(vmf, page);
3427 		if (!ret) {
3428 			/* The page is mapped successfully, reference consumed. */
3429 			folio_unlock(folio);
3430 			return true;
3431 		}
3432 	}
3433 
3434 	if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
3435 		pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3436 
3437 	return false;
3438 }
3439 
next_uptodate_folio(struct xa_state * xas,struct address_space * mapping,pgoff_t end_pgoff)3440 static struct folio *next_uptodate_folio(struct xa_state *xas,
3441 		struct address_space *mapping, pgoff_t end_pgoff)
3442 {
3443 	struct folio *folio = xas_next_entry(xas, end_pgoff);
3444 	unsigned long max_idx;
3445 
3446 	do {
3447 		if (!folio)
3448 			return NULL;
3449 		if (xas_retry(xas, folio))
3450 			continue;
3451 		if (xa_is_value(folio))
3452 			continue;
3453 		if (folio_test_locked(folio))
3454 			continue;
3455 		if (!folio_try_get_rcu(folio))
3456 			continue;
3457 		/* Has the page moved or been split? */
3458 		if (unlikely(folio != xas_reload(xas)))
3459 			goto skip;
3460 		if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3461 			goto skip;
3462 		if (!folio_trylock(folio))
3463 			goto skip;
3464 		if (folio->mapping != mapping)
3465 			goto unlock;
3466 		if (!folio_test_uptodate(folio))
3467 			goto unlock;
3468 		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3469 		if (xas->xa_index >= max_idx)
3470 			goto unlock;
3471 		return folio;
3472 unlock:
3473 		folio_unlock(folio);
3474 skip:
3475 		folio_put(folio);
3476 	} while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3477 
3478 	return NULL;
3479 }
3480 
3481 /*
3482  * Map page range [start_page, start_page + nr_pages) of folio.
3483  * start_page is gotten from start by folio_page(folio, start)
3484  */
filemap_map_folio_range(struct vm_fault * vmf,struct folio * folio,unsigned long start,unsigned long addr,unsigned int nr_pages,unsigned int * mmap_miss)3485 static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
3486 			struct folio *folio, unsigned long start,
3487 			unsigned long addr, unsigned int nr_pages,
3488 			unsigned int *mmap_miss)
3489 {
3490 	vm_fault_t ret = 0;
3491 	struct page *page = folio_page(folio, start);
3492 	unsigned int count = 0;
3493 	pte_t *old_ptep = vmf->pte;
3494 
3495 	do {
3496 		if (PageHWPoison(page + count))
3497 			goto skip;
3498 
3499 		(*mmap_miss)++;
3500 
3501 		/*
3502 		 * NOTE: If there're PTE markers, we'll leave them to be
3503 		 * handled in the specific fault path, and it'll prohibit the
3504 		 * fault-around logic.
3505 		 */
3506 		if (!pte_none(vmf->pte[count]))
3507 			goto skip;
3508 
3509 		count++;
3510 		continue;
3511 skip:
3512 		if (count) {
3513 			set_pte_range(vmf, folio, page, count, addr);
3514 			folio_ref_add(folio, count);
3515 			if (in_range(vmf->address, addr, count * PAGE_SIZE))
3516 				ret = VM_FAULT_NOPAGE;
3517 		}
3518 
3519 		count++;
3520 		page += count;
3521 		vmf->pte += count;
3522 		addr += count * PAGE_SIZE;
3523 		count = 0;
3524 	} while (--nr_pages > 0);
3525 
3526 	if (count) {
3527 		set_pte_range(vmf, folio, page, count, addr);
3528 		folio_ref_add(folio, count);
3529 		if (in_range(vmf->address, addr, count * PAGE_SIZE))
3530 			ret = VM_FAULT_NOPAGE;
3531 	}
3532 
3533 	vmf->pte = old_ptep;
3534 
3535 	return ret;
3536 }
3537 
filemap_map_order0_folio(struct vm_fault * vmf,struct folio * folio,unsigned long addr,unsigned int * mmap_miss)3538 static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
3539 		struct folio *folio, unsigned long addr,
3540 		unsigned int *mmap_miss)
3541 {
3542 	vm_fault_t ret = 0;
3543 	struct page *page = &folio->page;
3544 
3545 	if (PageHWPoison(page))
3546 		return ret;
3547 
3548 	(*mmap_miss)++;
3549 
3550 	/*
3551 	 * NOTE: If there're PTE markers, we'll leave them to be
3552 	 * handled in the specific fault path, and it'll prohibit
3553 	 * the fault-around logic.
3554 	 */
3555 	if (!pte_none(ptep_get(vmf->pte)))
3556 		return ret;
3557 
3558 	if (vmf->address == addr)
3559 		ret = VM_FAULT_NOPAGE;
3560 
3561 	set_pte_range(vmf, folio, page, 1, addr);
3562 	folio_ref_inc(folio);
3563 
3564 	return ret;
3565 }
3566 
filemap_map_pages(struct vm_fault * vmf,pgoff_t start_pgoff,pgoff_t end_pgoff)3567 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3568 			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3569 {
3570 	struct vm_area_struct *vma = vmf->vma;
3571 	struct file *file = vma->vm_file;
3572 	struct address_space *mapping = file->f_mapping;
3573 	pgoff_t last_pgoff = start_pgoff;
3574 	unsigned long addr;
3575 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3576 	struct folio *folio;
3577 	vm_fault_t ret = 0;
3578 	unsigned int nr_pages = 0, mmap_miss = 0, mmap_miss_saved;
3579 
3580 	rcu_read_lock();
3581 	folio = next_uptodate_folio(&xas, mapping, end_pgoff);
3582 	if (!folio)
3583 		goto out;
3584 
3585 	if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3586 		ret = VM_FAULT_NOPAGE;
3587 		goto out;
3588 	}
3589 
3590 	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3591 	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3592 	if (!vmf->pte) {
3593 		folio_unlock(folio);
3594 		folio_put(folio);
3595 		goto out;
3596 	}
3597 	do {
3598 		unsigned long end;
3599 
3600 		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3601 		vmf->pte += xas.xa_index - last_pgoff;
3602 		last_pgoff = xas.xa_index;
3603 		end = folio->index + folio_nr_pages(folio) - 1;
3604 		nr_pages = min(end, end_pgoff) - xas.xa_index + 1;
3605 
3606 		if (!folio_test_large(folio))
3607 			ret |= filemap_map_order0_folio(vmf,
3608 					folio, addr, &mmap_miss);
3609 		else
3610 			ret |= filemap_map_folio_range(vmf, folio,
3611 					xas.xa_index - folio->index, addr,
3612 					nr_pages, &mmap_miss);
3613 
3614 		folio_unlock(folio);
3615 		folio_put(folio);
3616 	} while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
3617 	pte_unmap_unlock(vmf->pte, vmf->ptl);
3618 out:
3619 	rcu_read_unlock();
3620 
3621 	mmap_miss_saved = READ_ONCE(file->f_ra.mmap_miss);
3622 	if (mmap_miss >= mmap_miss_saved)
3623 		WRITE_ONCE(file->f_ra.mmap_miss, 0);
3624 	else
3625 		WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss_saved - mmap_miss);
3626 
3627 	return ret;
3628 }
3629 EXPORT_SYMBOL(filemap_map_pages);
3630 
filemap_page_mkwrite(struct vm_fault * vmf)3631 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3632 {
3633 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3634 	struct folio *folio = page_folio(vmf->page);
3635 	vm_fault_t ret = VM_FAULT_LOCKED;
3636 
3637 	sb_start_pagefault(mapping->host->i_sb);
3638 	file_update_time(vmf->vma->vm_file);
3639 	folio_lock(folio);
3640 	if (folio->mapping != mapping) {
3641 		folio_unlock(folio);
3642 		ret = VM_FAULT_NOPAGE;
3643 		goto out;
3644 	}
3645 	/*
3646 	 * We mark the folio dirty already here so that when freeze is in
3647 	 * progress, we are guaranteed that writeback during freezing will
3648 	 * see the dirty folio and writeprotect it again.
3649 	 */
3650 	folio_mark_dirty(folio);
3651 	folio_wait_stable(folio);
3652 out:
3653 	sb_end_pagefault(mapping->host->i_sb);
3654 	return ret;
3655 }
3656 
3657 const struct vm_operations_struct generic_file_vm_ops = {
3658 	.fault		= filemap_fault,
3659 	.map_pages	= filemap_map_pages,
3660 	.page_mkwrite	= filemap_page_mkwrite,
3661 };
3662 
3663 /* This is used for a general mmap of a disk file */
3664 
generic_file_mmap(struct file * file,struct vm_area_struct * vma)3665 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3666 {
3667 	struct address_space *mapping = file->f_mapping;
3668 
3669 	if (!mapping->a_ops->read_folio)
3670 		return -ENOEXEC;
3671 	file_accessed(file);
3672 	vma->vm_ops = &generic_file_vm_ops;
3673 	return 0;
3674 }
3675 
3676 /*
3677  * This is for filesystems which do not implement ->writepage.
3678  */
generic_file_readonly_mmap(struct file * file,struct vm_area_struct * vma)3679 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3680 {
3681 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3682 		return -EINVAL;
3683 	return generic_file_mmap(file, vma);
3684 }
3685 #else
filemap_page_mkwrite(struct vm_fault * vmf)3686 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3687 {
3688 	return VM_FAULT_SIGBUS;
3689 }
generic_file_mmap(struct file * file,struct vm_area_struct * vma)3690 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3691 {
3692 	return -ENOSYS;
3693 }
generic_file_readonly_mmap(struct file * file,struct vm_area_struct * vma)3694 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3695 {
3696 	return -ENOSYS;
3697 }
3698 #endif /* CONFIG_MMU */
3699 
3700 EXPORT_SYMBOL(filemap_page_mkwrite);
3701 EXPORT_SYMBOL(generic_file_mmap);
3702 EXPORT_SYMBOL(generic_file_readonly_mmap);
3703 
do_read_cache_folio(struct address_space * mapping,pgoff_t index,filler_t filler,struct file * file,gfp_t gfp)3704 static struct folio *do_read_cache_folio(struct address_space *mapping,
3705 		pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3706 {
3707 	struct folio *folio;
3708 	int err;
3709 
3710 	if (!filler)
3711 		filler = mapping->a_ops->read_folio;
3712 repeat:
3713 	folio = filemap_get_folio(mapping, index);
3714 	if (IS_ERR(folio)) {
3715 		folio = filemap_alloc_folio(gfp, 0);
3716 		if (!folio)
3717 			return ERR_PTR(-ENOMEM);
3718 		err = filemap_add_folio(mapping, folio, index, gfp);
3719 		if (unlikely(err)) {
3720 			folio_put(folio);
3721 			if (err == -EEXIST)
3722 				goto repeat;
3723 			/* Presumably ENOMEM for xarray node */
3724 			return ERR_PTR(err);
3725 		}
3726 
3727 		goto filler;
3728 	}
3729 	if (folio_test_uptodate(folio))
3730 		goto out;
3731 
3732 	if (!folio_trylock(folio)) {
3733 		folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3734 		goto repeat;
3735 	}
3736 
3737 	/* Folio was truncated from mapping */
3738 	if (!folio->mapping) {
3739 		folio_unlock(folio);
3740 		folio_put(folio);
3741 		goto repeat;
3742 	}
3743 
3744 	/* Someone else locked and filled the page in a very small window */
3745 	if (folio_test_uptodate(folio)) {
3746 		folio_unlock(folio);
3747 		goto out;
3748 	}
3749 
3750 filler:
3751 	err = filemap_read_folio(file, filler, folio);
3752 	if (err) {
3753 		folio_put(folio);
3754 		if (err == AOP_TRUNCATED_PAGE)
3755 			goto repeat;
3756 		return ERR_PTR(err);
3757 	}
3758 
3759 out:
3760 	folio_mark_accessed(folio);
3761 	return folio;
3762 }
3763 
3764 /**
3765  * read_cache_folio - Read into page cache, fill it if needed.
3766  * @mapping: The address_space to read from.
3767  * @index: The index to read.
3768  * @filler: Function to perform the read, or NULL to use aops->read_folio().
3769  * @file: Passed to filler function, may be NULL if not required.
3770  *
3771  * Read one page into the page cache.  If it succeeds, the folio returned
3772  * will contain @index, but it may not be the first page of the folio.
3773  *
3774  * If the filler function returns an error, it will be returned to the
3775  * caller.
3776  *
3777  * Context: May sleep.  Expects mapping->invalidate_lock to be held.
3778  * Return: An uptodate folio on success, ERR_PTR() on failure.
3779  */
read_cache_folio(struct address_space * mapping,pgoff_t index,filler_t filler,struct file * file)3780 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3781 		filler_t filler, struct file *file)
3782 {
3783 	return do_read_cache_folio(mapping, index, filler, file,
3784 			mapping_gfp_mask(mapping));
3785 }
3786 EXPORT_SYMBOL(read_cache_folio);
3787 
3788 /**
3789  * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3790  * @mapping:	The address_space for the folio.
3791  * @index:	The index that the allocated folio will contain.
3792  * @gfp:	The page allocator flags to use if allocating.
3793  *
3794  * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3795  * any new memory allocations done using the specified allocation flags.
3796  *
3797  * The most likely error from this function is EIO, but ENOMEM is
3798  * possible and so is EINTR.  If ->read_folio returns another error,
3799  * that will be returned to the caller.
3800  *
3801  * The function expects mapping->invalidate_lock to be already held.
3802  *
3803  * Return: Uptodate folio on success, ERR_PTR() on failure.
3804  */
mapping_read_folio_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)3805 struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3806 		pgoff_t index, gfp_t gfp)
3807 {
3808 	return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3809 }
3810 EXPORT_SYMBOL(mapping_read_folio_gfp);
3811 
do_read_cache_page(struct address_space * mapping,pgoff_t index,filler_t * filler,struct file * file,gfp_t gfp)3812 static struct page *do_read_cache_page(struct address_space *mapping,
3813 		pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3814 {
3815 	struct folio *folio;
3816 
3817 	folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3818 	if (IS_ERR(folio))
3819 		return &folio->page;
3820 	return folio_file_page(folio, index);
3821 }
3822 
read_cache_page(struct address_space * mapping,pgoff_t index,filler_t * filler,struct file * file)3823 struct page *read_cache_page(struct address_space *mapping,
3824 			pgoff_t index, filler_t *filler, struct file *file)
3825 {
3826 	return do_read_cache_page(mapping, index, filler, file,
3827 			mapping_gfp_mask(mapping));
3828 }
3829 EXPORT_SYMBOL(read_cache_page);
3830 
3831 /**
3832  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3833  * @mapping:	the page's address_space
3834  * @index:	the page index
3835  * @gfp:	the page allocator flags to use if allocating
3836  *
3837  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3838  * any new page allocations done using the specified allocation flags.
3839  *
3840  * If the page does not get brought uptodate, return -EIO.
3841  *
3842  * The function expects mapping->invalidate_lock to be already held.
3843  *
3844  * Return: up to date page on success, ERR_PTR() on failure.
3845  */
read_cache_page_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)3846 struct page *read_cache_page_gfp(struct address_space *mapping,
3847 				pgoff_t index,
3848 				gfp_t gfp)
3849 {
3850 	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3851 }
3852 EXPORT_SYMBOL(read_cache_page_gfp);
3853 
3854 /*
3855  * Warn about a page cache invalidation failure during a direct I/O write.
3856  */
dio_warn_stale_pagecache(struct file * filp)3857 static void dio_warn_stale_pagecache(struct file *filp)
3858 {
3859 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3860 	char pathname[128];
3861 	char *path;
3862 
3863 	errseq_set(&filp->f_mapping->wb_err, -EIO);
3864 	if (__ratelimit(&_rs)) {
3865 		path = file_path(filp, pathname, sizeof(pathname));
3866 		if (IS_ERR(path))
3867 			path = "(unknown)";
3868 		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3869 		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3870 			current->comm);
3871 	}
3872 }
3873 
kiocb_invalidate_post_direct_write(struct kiocb * iocb,size_t count)3874 void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
3875 {
3876 	struct address_space *mapping = iocb->ki_filp->f_mapping;
3877 
3878 	if (mapping->nrpages &&
3879 	    invalidate_inode_pages2_range(mapping,
3880 			iocb->ki_pos >> PAGE_SHIFT,
3881 			(iocb->ki_pos + count - 1) >> PAGE_SHIFT))
3882 		dio_warn_stale_pagecache(iocb->ki_filp);
3883 }
3884 
3885 ssize_t
generic_file_direct_write(struct kiocb * iocb,struct iov_iter * from)3886 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3887 {
3888 	struct address_space *mapping = iocb->ki_filp->f_mapping;
3889 	size_t write_len = iov_iter_count(from);
3890 	ssize_t written;
3891 
3892 	/*
3893 	 * If a page can not be invalidated, return 0 to fall back
3894 	 * to buffered write.
3895 	 */
3896 	written = kiocb_invalidate_pages(iocb, write_len);
3897 	if (written) {
3898 		if (written == -EBUSY)
3899 			return 0;
3900 		return written;
3901 	}
3902 
3903 	written = mapping->a_ops->direct_IO(iocb, from);
3904 
3905 	/*
3906 	 * Finally, try again to invalidate clean pages which might have been
3907 	 * cached by non-direct readahead, or faulted in by get_user_pages()
3908 	 * if the source of the write was an mmap'ed region of the file
3909 	 * we're writing.  Either one is a pretty crazy thing to do,
3910 	 * so we don't support it 100%.  If this invalidation
3911 	 * fails, tough, the write still worked...
3912 	 *
3913 	 * Most of the time we do not need this since dio_complete() will do
3914 	 * the invalidation for us. However there are some file systems that
3915 	 * do not end up with dio_complete() being called, so let's not break
3916 	 * them by removing it completely.
3917 	 *
3918 	 * Noticeable example is a blkdev_direct_IO().
3919 	 *
3920 	 * Skip invalidation for async writes or if mapping has no pages.
3921 	 */
3922 	if (written > 0) {
3923 		struct inode *inode = mapping->host;
3924 		loff_t pos = iocb->ki_pos;
3925 
3926 		kiocb_invalidate_post_direct_write(iocb, written);
3927 		pos += written;
3928 		write_len -= written;
3929 		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3930 			i_size_write(inode, pos);
3931 			mark_inode_dirty(inode);
3932 		}
3933 		iocb->ki_pos = pos;
3934 	}
3935 	if (written != -EIOCBQUEUED)
3936 		iov_iter_revert(from, write_len - iov_iter_count(from));
3937 	return written;
3938 }
3939 EXPORT_SYMBOL(generic_file_direct_write);
3940 
generic_perform_write(struct kiocb * iocb,struct iov_iter * i)3941 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3942 {
3943 	struct file *file = iocb->ki_filp;
3944 	loff_t pos = iocb->ki_pos;
3945 	struct address_space *mapping = file->f_mapping;
3946 	const struct address_space_operations *a_ops = mapping->a_ops;
3947 	long status = 0;
3948 	ssize_t written = 0;
3949 
3950 	do {
3951 		struct page *page;
3952 		unsigned long offset;	/* Offset into pagecache page */
3953 		unsigned long bytes;	/* Bytes to write to page */
3954 		size_t copied;		/* Bytes copied from user */
3955 		void *fsdata = NULL;
3956 
3957 		offset = (pos & (PAGE_SIZE - 1));
3958 		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3959 						iov_iter_count(i));
3960 
3961 again:
3962 		/*
3963 		 * Bring in the user page that we will copy from _first_.
3964 		 * Otherwise there's a nasty deadlock on copying from the
3965 		 * same page as we're writing to, without it being marked
3966 		 * up-to-date.
3967 		 */
3968 		if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3969 			status = -EFAULT;
3970 			break;
3971 		}
3972 
3973 		if (fatal_signal_pending(current)) {
3974 			status = -EINTR;
3975 			break;
3976 		}
3977 
3978 		status = a_ops->write_begin(file, mapping, pos, bytes,
3979 						&page, &fsdata);
3980 		if (unlikely(status < 0))
3981 			break;
3982 
3983 		if (mapping_writably_mapped(mapping))
3984 			flush_dcache_page(page);
3985 
3986 		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3987 		flush_dcache_page(page);
3988 
3989 		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3990 						page, fsdata);
3991 		if (unlikely(status != copied)) {
3992 			iov_iter_revert(i, copied - max(status, 0L));
3993 			if (unlikely(status < 0))
3994 				break;
3995 		}
3996 		cond_resched();
3997 
3998 		if (unlikely(status == 0)) {
3999 			/*
4000 			 * A short copy made ->write_end() reject the
4001 			 * thing entirely.  Might be memory poisoning
4002 			 * halfway through, might be a race with munmap,
4003 			 * might be severe memory pressure.
4004 			 */
4005 			if (copied)
4006 				bytes = copied;
4007 			goto again;
4008 		}
4009 		pos += status;
4010 		written += status;
4011 
4012 		balance_dirty_pages_ratelimited(mapping);
4013 	} while (iov_iter_count(i));
4014 
4015 	if (!written)
4016 		return status;
4017 	iocb->ki_pos += written;
4018 	return written;
4019 }
4020 EXPORT_SYMBOL(generic_perform_write);
4021 
4022 /**
4023  * __generic_file_write_iter - write data to a file
4024  * @iocb:	IO state structure (file, offset, etc.)
4025  * @from:	iov_iter with data to write
4026  *
4027  * This function does all the work needed for actually writing data to a
4028  * file. It does all basic checks, removes SUID from the file, updates
4029  * modification times and calls proper subroutines depending on whether we
4030  * do direct IO or a standard buffered write.
4031  *
4032  * It expects i_rwsem to be grabbed unless we work on a block device or similar
4033  * object which does not need locking at all.
4034  *
4035  * This function does *not* take care of syncing data in case of O_SYNC write.
4036  * A caller has to handle it. This is mainly due to the fact that we want to
4037  * avoid syncing under i_rwsem.
4038  *
4039  * Return:
4040  * * number of bytes written, even for truncated writes
4041  * * negative error code if no data has been written at all
4042  */
__generic_file_write_iter(struct kiocb * iocb,struct iov_iter * from)4043 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4044 {
4045 	struct file *file = iocb->ki_filp;
4046 	struct address_space *mapping = file->f_mapping;
4047 	struct inode *inode = mapping->host;
4048 	ssize_t ret;
4049 
4050 	ret = file_remove_privs(file);
4051 	if (ret)
4052 		return ret;
4053 
4054 	ret = file_update_time(file);
4055 	if (ret)
4056 		return ret;
4057 
4058 	if (iocb->ki_flags & IOCB_DIRECT) {
4059 		ret = generic_file_direct_write(iocb, from);
4060 		/*
4061 		 * If the write stopped short of completing, fall back to
4062 		 * buffered writes.  Some filesystems do this for writes to
4063 		 * holes, for example.  For DAX files, a buffered write will
4064 		 * not succeed (even if it did, DAX does not handle dirty
4065 		 * page-cache pages correctly).
4066 		 */
4067 		if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4068 			return ret;
4069 		return direct_write_fallback(iocb, from, ret,
4070 				generic_perform_write(iocb, from));
4071 	}
4072 
4073 	return generic_perform_write(iocb, from);
4074 }
4075 EXPORT_SYMBOL(__generic_file_write_iter);
4076 
4077 /**
4078  * generic_file_write_iter - write data to a file
4079  * @iocb:	IO state structure
4080  * @from:	iov_iter with data to write
4081  *
4082  * This is a wrapper around __generic_file_write_iter() to be used by most
4083  * filesystems. It takes care of syncing the file in case of O_SYNC file
4084  * and acquires i_rwsem as needed.
4085  * Return:
4086  * * negative error code if no data has been written at all of
4087  *   vfs_fsync_range() failed for a synchronous write
4088  * * number of bytes written, even for truncated writes
4089  */
generic_file_write_iter(struct kiocb * iocb,struct iov_iter * from)4090 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4091 {
4092 	struct file *file = iocb->ki_filp;
4093 	struct inode *inode = file->f_mapping->host;
4094 	ssize_t ret;
4095 
4096 	inode_lock(inode);
4097 	ret = generic_write_checks(iocb, from);
4098 	if (ret > 0)
4099 		ret = __generic_file_write_iter(iocb, from);
4100 	inode_unlock(inode);
4101 
4102 	if (ret > 0)
4103 		ret = generic_write_sync(iocb, ret);
4104 	return ret;
4105 }
4106 EXPORT_SYMBOL(generic_file_write_iter);
4107 
4108 /**
4109  * filemap_release_folio() - Release fs-specific metadata on a folio.
4110  * @folio: The folio which the kernel is trying to free.
4111  * @gfp: Memory allocation flags (and I/O mode).
4112  *
4113  * The address_space is trying to release any data attached to a folio
4114  * (presumably at folio->private).
4115  *
4116  * This will also be called if the private_2 flag is set on a page,
4117  * indicating that the folio has other metadata associated with it.
4118  *
4119  * The @gfp argument specifies whether I/O may be performed to release
4120  * this page (__GFP_IO), and whether the call may block
4121  * (__GFP_RECLAIM & __GFP_FS).
4122  *
4123  * Return: %true if the release was successful, otherwise %false.
4124  */
filemap_release_folio(struct folio * folio,gfp_t gfp)4125 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4126 {
4127 	struct address_space * const mapping = folio->mapping;
4128 
4129 	BUG_ON(!folio_test_locked(folio));
4130 	if (!folio_needs_release(folio))
4131 		return true;
4132 	if (folio_test_writeback(folio))
4133 		return false;
4134 
4135 	if (mapping && mapping->a_ops->release_folio)
4136 		return mapping->a_ops->release_folio(folio, gfp);
4137 	return try_to_free_buffers(folio);
4138 }
4139 EXPORT_SYMBOL(filemap_release_folio);
4140 
4141 #ifdef CONFIG_CACHESTAT_SYSCALL
4142 /**
4143  * filemap_cachestat() - compute the page cache statistics of a mapping
4144  * @mapping:	The mapping to compute the statistics for.
4145  * @first_index:	The starting page cache index.
4146  * @last_index:	The final page index (inclusive).
4147  * @cs:	the cachestat struct to write the result to.
4148  *
4149  * This will query the page cache statistics of a mapping in the
4150  * page range of [first_index, last_index] (inclusive). The statistics
4151  * queried include: number of dirty pages, number of pages marked for
4152  * writeback, and the number of (recently) evicted pages.
4153  */
filemap_cachestat(struct address_space * mapping,pgoff_t first_index,pgoff_t last_index,struct cachestat * cs)4154 static void filemap_cachestat(struct address_space *mapping,
4155 		pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4156 {
4157 	XA_STATE(xas, &mapping->i_pages, first_index);
4158 	struct folio *folio;
4159 
4160 	rcu_read_lock();
4161 	xas_for_each(&xas, folio, last_index) {
4162 		int order;
4163 		unsigned long nr_pages;
4164 		pgoff_t folio_first_index, folio_last_index;
4165 
4166 		/*
4167 		 * Don't deref the folio. It is not pinned, and might
4168 		 * get freed (and reused) underneath us.
4169 		 *
4170 		 * We *could* pin it, but that would be expensive for
4171 		 * what should be a fast and lightweight syscall.
4172 		 *
4173 		 * Instead, derive all information of interest from
4174 		 * the rcu-protected xarray.
4175 		 */
4176 
4177 		if (xas_retry(&xas, folio))
4178 			continue;
4179 
4180 		order = xa_get_order(xas.xa, xas.xa_index);
4181 		nr_pages = 1 << order;
4182 		folio_first_index = round_down(xas.xa_index, 1 << order);
4183 		folio_last_index = folio_first_index + nr_pages - 1;
4184 
4185 		/* Folios might straddle the range boundaries, only count covered pages */
4186 		if (folio_first_index < first_index)
4187 			nr_pages -= first_index - folio_first_index;
4188 
4189 		if (folio_last_index > last_index)
4190 			nr_pages -= folio_last_index - last_index;
4191 
4192 		if (xa_is_value(folio)) {
4193 			/* page is evicted */
4194 			void *shadow = (void *)folio;
4195 			bool workingset; /* not used */
4196 
4197 			cs->nr_evicted += nr_pages;
4198 
4199 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4200 			if (shmem_mapping(mapping)) {
4201 				/* shmem file - in swap cache */
4202 				swp_entry_t swp = radix_to_swp_entry(folio);
4203 
4204 				shadow = get_shadow_from_swap_cache(swp);
4205 			}
4206 #endif
4207 			if (workingset_test_recent(shadow, true, &workingset))
4208 				cs->nr_recently_evicted += nr_pages;
4209 
4210 			goto resched;
4211 		}
4212 
4213 		/* page is in cache */
4214 		cs->nr_cache += nr_pages;
4215 
4216 		if (xas_get_mark(&xas, PAGECACHE_TAG_DIRTY))
4217 			cs->nr_dirty += nr_pages;
4218 
4219 		if (xas_get_mark(&xas, PAGECACHE_TAG_WRITEBACK))
4220 			cs->nr_writeback += nr_pages;
4221 
4222 resched:
4223 		if (need_resched()) {
4224 			xas_pause(&xas);
4225 			cond_resched_rcu();
4226 		}
4227 	}
4228 	rcu_read_unlock();
4229 }
4230 
4231 /*
4232  * The cachestat(2) system call.
4233  *
4234  * cachestat() returns the page cache statistics of a file in the
4235  * bytes range specified by `off` and `len`: number of cached pages,
4236  * number of dirty pages, number of pages marked for writeback,
4237  * number of evicted pages, and number of recently evicted pages.
4238  *
4239  * An evicted page is a page that is previously in the page cache
4240  * but has been evicted since. A page is recently evicted if its last
4241  * eviction was recent enough that its reentry to the cache would
4242  * indicate that it is actively being used by the system, and that
4243  * there is memory pressure on the system.
4244  *
4245  * `off` and `len` must be non-negative integers. If `len` > 0,
4246  * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4247  * we will query in the range from `off` to the end of the file.
4248  *
4249  * The `flags` argument is unused for now, but is included for future
4250  * extensibility. User should pass 0 (i.e no flag specified).
4251  *
4252  * Currently, hugetlbfs is not supported.
4253  *
4254  * Because the status of a page can change after cachestat() checks it
4255  * but before it returns to the application, the returned values may
4256  * contain stale information.
4257  *
4258  * return values:
4259  *  zero        - success
4260  *  -EFAULT     - cstat or cstat_range points to an illegal address
4261  *  -EINVAL     - invalid flags
4262  *  -EBADF      - invalid file descriptor
4263  *  -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4264  */
SYSCALL_DEFINE4(cachestat,unsigned int,fd,struct cachestat_range __user *,cstat_range,struct cachestat __user *,cstat,unsigned int,flags)4265 SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4266 		struct cachestat_range __user *, cstat_range,
4267 		struct cachestat __user *, cstat, unsigned int, flags)
4268 {
4269 	struct fd f = fdget(fd);
4270 	struct address_space *mapping;
4271 	struct cachestat_range csr;
4272 	struct cachestat cs;
4273 	pgoff_t first_index, last_index;
4274 
4275 	if (!f.file)
4276 		return -EBADF;
4277 
4278 	if (copy_from_user(&csr, cstat_range,
4279 			sizeof(struct cachestat_range))) {
4280 		fdput(f);
4281 		return -EFAULT;
4282 	}
4283 
4284 	/* hugetlbfs is not supported */
4285 	if (is_file_hugepages(f.file)) {
4286 		fdput(f);
4287 		return -EOPNOTSUPP;
4288 	}
4289 
4290 	if (flags != 0) {
4291 		fdput(f);
4292 		return -EINVAL;
4293 	}
4294 
4295 	first_index = csr.off >> PAGE_SHIFT;
4296 	last_index =
4297 		csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4298 	memset(&cs, 0, sizeof(struct cachestat));
4299 	mapping = f.file->f_mapping;
4300 	filemap_cachestat(mapping, first_index, last_index, &cs);
4301 	fdput(f);
4302 
4303 	if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4304 		return -EFAULT;
4305 
4306 	return 0;
4307 }
4308 #endif /* CONFIG_CACHESTAT_SYSCALL */
4309