readahead.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * mm/readahead.c - address_space-level file readahead.
  4 *
  5 * Copyright (C) 2002, Linus Torvalds
  6 *
  7 * 09Apr2002	Andrew Morton
  8 *		Initial version.
  9 */
 10
 11/**
 12 * DOC: Readahead Overview
 13 *
 14 * Readahead is used to read content into the page cache before it is
 15 * explicitly requested by the application.  Readahead only ever
 16 * attempts to read folios that are not yet in the page cache.  If a
 17 * folio is present but not up-to-date, readahead will not try to read
 18 * it. In that case a simple ->read_folio() will be requested.
 19 *
 20 * Readahead is triggered when an application read request (whether a
 21 * system call or a page fault) finds that the requested folio is not in
 22 * the page cache, or that it is in the page cache and has the
 23 * readahead flag set.  This flag indicates that the folio was read
 24 * as part of a previous readahead request and now that it has been
 25 * accessed, it is time for the next readahead.
 26 *
 27 * Each readahead request is partly synchronous read, and partly async
 28 * readahead.  This is reflected in the struct file_ra_state which
 29 * contains ->size being the total number of pages, and ->async_size
 30 * which is the number of pages in the async section.  The readahead
 31 * flag will be set on the first folio in this async section to trigger
 32 * a subsequent readahead.  Once a series of sequential reads has been
 33 * established, there should be no need for a synchronous component and
 34 * all readahead request will be fully asynchronous.
 35 *
 36 * When either of the triggers causes a readahead, three numbers need
 37 * to be determined: the start of the region to read, the size of the
 38 * region, and the size of the async tail.
 39 *
 40 * The start of the region is simply the first page address at or after
 41 * the accessed address, which is not currently populated in the page
 42 * cache.  This is found with a simple search in the page cache.
 43 *
 44 * The size of the async tail is determined by subtracting the size that
 45 * was explicitly requested from the determined request size, unless
 46 * this would be less than zero - then zero is used.  NOTE THIS
 47 * CALCULATION IS WRONG WHEN THE START OF THE REGION IS NOT THE ACCESSED
 48 * PAGE.  ALSO THIS CALCULATION IS NOT USED CONSISTENTLY.
 49 *
 50 * The size of the region is normally determined from the size of the
 51 * previous readahead which loaded the preceding pages.  This may be
 52 * discovered from the struct file_ra_state for simple sequential reads,
 53 * or from examining the state of the page cache when multiple
 54 * sequential reads are interleaved.  Specifically: where the readahead
 55 * was triggered by the readahead flag, the size of the previous
 56 * readahead is assumed to be the number of pages from the triggering
 57 * page to the start of the new readahead.  In these cases, the size of
 58 * the previous readahead is scaled, often doubled, for the new
 59 * readahead, though see get_next_ra_size() for details.
 60 *
 61 * If the size of the previous read cannot be determined, the number of
 62 * preceding pages in the page cache is used to estimate the size of
 63 * a previous read.  This estimate could easily be misled by random
 64 * reads being coincidentally adjacent, so it is ignored unless it is
 65 * larger than the current request, and it is not scaled up, unless it
 66 * is at the start of file.
 67 *
 68 * In general readahead is accelerated at the start of the file, as
 69 * reads from there are often sequential.  There are other minor
 70 * adjustments to the readahead size in various special cases and these
 71 * are best discovered by reading the code.
 72 *
 73 * The above calculation, based on the previous readahead size,
 74 * determines the size of the readahead, to which any requested read
 75 * size may be added.
 76 *
 77 * Readahead requests are sent to the filesystem using the ->readahead()
 78 * address space operation, for which mpage_readahead() is a canonical
 79 * implementation.  ->readahead() should normally initiate reads on all
 80 * folios, but may fail to read any or all folios without causing an I/O
 81 * error.  The page cache reading code will issue a ->read_folio() request
 82 * for any folio which ->readahead() did not read, and only an error
 83 * from this will be final.
 84 *
 85 * ->readahead() will generally call readahead_folio() repeatedly to get
 86 * each folio from those prepared for readahead.  It may fail to read a
 87 * folio by:
 88 *
 89 * * not calling readahead_folio() sufficiently many times, effectively
 90 *   ignoring some folios, as might be appropriate if the path to
 91 *   storage is congested.
 92 *
 93 * * failing to actually submit a read request for a given folio,
 94 *   possibly due to insufficient resources, or
 95 *
 96 * * getting an error during subsequent processing of a request.
 97 *
 98 * In the last two cases, the folio should be unlocked by the filesystem
 99 * to indicate that the read attempt has failed.  In the first case the
100 * folio will be unlocked by the VFS.
101 *
102 * Those folios not in the final ``async_size`` of the request should be
103 * considered to be important and ->readahead() should not fail them due
104 * to congestion or temporary resource unavailability, but should wait
105 * for necessary resources (e.g.  memory or indexing information) to
106 * become available.  Folios in the final ``async_size`` may be
107 * considered less urgent and failure to read them is more acceptable.
108 * In this case it is best to use filemap_remove_folio() to remove the
109 * folios from the page cache as is automatically done for folios that
110 * were not fetched with readahead_folio().  This will allow a
111 * subsequent synchronous readahead request to try them again.  If they
112 * are left in the page cache, then they will be read individually using
113 * ->read_folio() which may be less efficient.
114 */
115
116#include <linux/blkdev.h>
117#include <linux/kernel.h>
118#include <linux/dax.h>
119#include <linux/gfp.h>
120#include <linux/export.h>
121#include <linux/backing-dev.h>
122#include <linux/task_io_accounting_ops.h>
123#include <linux/pagemap.h>
124#include <linux/psi.h>
125#include <linux/syscalls.h>
126#include <linux/file.h>
127#include <linux/mm_inline.h>
128#include <linux/blk-cgroup.h>
129#include <linux/fadvise.h>
130#include <linux/sched/mm.h>
131
132#define CREATE_TRACE_POINTS
133#include <trace/events/readahead.h>
134
135#include "internal.h"
136
137/*
138 * Initialise a struct file's readahead state.  Assumes that the caller has
139 * memset *ra to zero.
140 */
141void
142file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
143{
144	ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages;
145	ra->prev_pos = -1;
146}
147EXPORT_SYMBOL_GPL(file_ra_state_init);
148
149static void read_pages(struct readahead_control *rac)
150{
151	const struct address_space_operations *aops = rac->mapping->a_ops;
152	struct folio *folio;
153	struct blk_plug plug;
154
155	if (!readahead_count(rac))
156		return;
157
158	if (unlikely(rac->_workingset))
159		psi_memstall_enter(&rac->_pflags);
160	blk_start_plug(&plug);
161
162	if (aops->readahead) {
163		aops->readahead(rac);
164		/* Clean up the remaining folios. */
165		while ((folio = readahead_folio(rac)) != NULL) {
166			folio_get(folio);
167			filemap_remove_folio(folio);
168			folio_unlock(folio);
169			folio_put(folio);
170		}
171	} else {
172		while ((folio = readahead_folio(rac)) != NULL)
173			aops->read_folio(rac->file, folio);
174	}
175
176	blk_finish_plug(&plug);
177	if (unlikely(rac->_workingset))
178		psi_memstall_leave(&rac->_pflags);
179	rac->_workingset = false;
180
181	BUG_ON(readahead_count(rac));
182}
183
184static struct folio *ractl_alloc_folio(struct readahead_control *ractl,
185				       gfp_t gfp_mask, unsigned int order)
186{
187	struct folio *folio;
188
189	folio = filemap_alloc_folio(gfp_mask, order, NULL);
190	if (folio && ractl->dropbehind)
191		__folio_set_dropbehind(folio);
192
193	return folio;
194}
195
196/**
197 * page_cache_ra_unbounded - Start unchecked readahead.
198 * @ractl: Readahead control.
199 * @nr_to_read: The number of pages to read.
200 * @lookahead_size: Where to start the next readahead.
201 *
202 * This function is for filesystems to call when they want to start
203 * readahead beyond a file's stated i_size.  This is almost certainly
204 * not the function you want to call.  Use page_cache_async_readahead()
205 * or page_cache_sync_readahead() instead.
206 *
207 * Context: File is referenced by caller.  Mutexes may be held by caller.
208 * May sleep, but will not reenter filesystem to reclaim memory.
209 */
210void page_cache_ra_unbounded(struct readahead_control *ractl,
211		unsigned long nr_to_read, unsigned long lookahead_size)
212{
213	struct address_space *mapping = ractl->mapping;
214	unsigned long index = readahead_index(ractl);
215	gfp_t gfp_mask = readahead_gfp_mask(mapping);
216	unsigned long mark = ULONG_MAX, i = 0;
217	unsigned int min_nrpages = mapping_min_folio_nrpages(mapping);
218
219	/*
220	 * Partway through the readahead operation, we will have added
221	 * locked pages to the page cache, but will not yet have submitted
222	 * them for I/O.  Adding another page may need to allocate memory,
223	 * which can trigger memory reclaim.  Telling the VM we're in
224	 * the middle of a filesystem operation will cause it to not
225	 * touch file-backed pages, preventing a deadlock.  Most (all?)
226	 * filesystems already specify __GFP_NOFS in their mapping's
227	 * gfp_mask, but let's be explicit here.
228	 */
229	unsigned int nofs = memalloc_nofs_save();
230
231	trace_page_cache_ra_unbounded(mapping->host, index, nr_to_read,
232				      lookahead_size);
233	filemap_invalidate_lock_shared(mapping);
234	index = mapping_align_index(mapping, index);
235
236	/*
237	 * As iterator `i` is aligned to min_nrpages, round_up the
238	 * difference between nr_to_read and lookahead_size to mark the
239	 * index that only has lookahead or "async_region" to set the
240	 * readahead flag.
241	 */
242	if (lookahead_size <= nr_to_read) {
243		unsigned long ra_folio_index;
244
245		ra_folio_index = round_up(readahead_index(ractl) +
246					  nr_to_read - lookahead_size,
247					  min_nrpages);
248		mark = ra_folio_index - index;
249	}
250	nr_to_read += readahead_index(ractl) - index;
251	ractl->_index = index;
252
253	/*
254	 * Preallocate as many pages as we will need.
255	 */
256	while (i < nr_to_read) {
257		struct folio *folio = xa_load(&mapping->i_pages, index + i);
258		int ret;
259
260		if (folio && !xa_is_value(folio)) {
261			/*
262			 * Page already present?  Kick off the current batch
263			 * of contiguous pages before continuing with the
264			 * next batch.  This page may be the one we would
265			 * have intended to mark as Readahead, but we don't
266			 * have a stable reference to this page, and it's
267			 * not worth getting one just for that.
268			 */
269			read_pages(ractl);
270			ractl->_index += min_nrpages;
271			i = ractl->_index + ractl->_nr_pages - index;
272			continue;
273		}
274
275		folio = ractl_alloc_folio(ractl, gfp_mask,
276					mapping_min_folio_order(mapping));
277		if (!folio)
278			break;
279
280		ret = filemap_add_folio(mapping, folio, index + i, gfp_mask);
281		if (ret < 0) {
282			folio_put(folio);
283			if (ret == -ENOMEM)
284				break;
285			read_pages(ractl);
286			ractl->_index += min_nrpages;
287			i = ractl->_index + ractl->_nr_pages - index;
288			continue;
289		}
290		if (i == mark)
291			folio_set_readahead(folio);
292		ractl->_workingset |= folio_test_workingset(folio);
293		ractl->_nr_pages += min_nrpages;
294		i += min_nrpages;
295	}
296
297	/*
298	 * Now start the IO.  We ignore I/O errors - if the folio is not
299	 * uptodate then the caller will launch read_folio again, and
300	 * will then handle the error.
301	 */
302	read_pages(ractl);
303	filemap_invalidate_unlock_shared(mapping);
304	memalloc_nofs_restore(nofs);
305}
306EXPORT_SYMBOL_GPL(page_cache_ra_unbounded);
307
308/*
309 * do_page_cache_ra() actually reads a chunk of disk.  It allocates
310 * the pages first, then submits them for I/O. This avoids the very bad
311 * behaviour which would occur if page allocations are causing VM writeback.
312 * We really don't want to intermingle reads and writes like that.
313 */
314static void do_page_cache_ra(struct readahead_control *ractl,
315		unsigned long nr_to_read, unsigned long lookahead_size)
316{
317	struct inode *inode = ractl->mapping->host;
318	unsigned long index = readahead_index(ractl);
319	loff_t isize = i_size_read(inode);
320	pgoff_t end_index;	/* The last page we want to read */
321
322	if (isize == 0)
323		return;
324
325	end_index = (isize - 1) >> PAGE_SHIFT;
326	if (index > end_index)
327		return;
328	/* Don't read past the page containing the last byte of the file */
329	if (nr_to_read > end_index - index)
330		nr_to_read = end_index - index + 1;
331
332	page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size);
333}
334
335/*
336 * Chunk the readahead into 2 megabyte units, so that we don't pin too much
337 * memory at once.
338 */
339void force_page_cache_ra(struct readahead_control *ractl,
340		unsigned long nr_to_read)
341{
342	struct address_space *mapping = ractl->mapping;
343	struct file_ra_state *ra = ractl->ra;
344	struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
345	unsigned long max_pages;
346
347	if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead))
348		return;
349
350	/*
351	 * If the request exceeds the readahead window, allow the read to
352	 * be up to the optimal hardware IO size
353	 */
354	max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
355	nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
356	while (nr_to_read) {
357		unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
358
359		if (this_chunk > nr_to_read)
360			this_chunk = nr_to_read;
361		do_page_cache_ra(ractl, this_chunk, 0);
362
363		nr_to_read -= this_chunk;
364	}
365}
366
367/*
368 * Set the initial window size, round to next power of 2 and square
369 * for small size, x 4 for medium, and x 2 for large
370 * for 128k (32 page) max ra
371 * 1-2 page = 16k, 3-4 page 32k, 5-8 page = 64k, > 8 page = 128k initial
372 */
373static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
374{
375	unsigned long newsize = roundup_pow_of_two(size);
376
377	if (newsize <= max / 32)
378		newsize = newsize * 4;
379	else if (newsize <= max / 4)
380		newsize = newsize * 2;
381	else
382		newsize = max;
383
384	return newsize;
385}
386
387/*
388 *  Get the previous window size, ramp it up, and
389 *  return it as the new window size.
390 */
391static unsigned long get_next_ra_size(struct file_ra_state *ra,
392				      unsigned long max)
393{
394	unsigned long cur = ra->size;
395
396	if (cur < max / 16)
397		return 4 * cur;
398	if (cur <= max / 2)
399		return 2 * cur;
400	return max;
401}
402
403/*
404 * On-demand readahead design.
405 *
406 * The fields in struct file_ra_state represent the most-recently-executed
407 * readahead attempt:
408 *
409 *                        |<----- async_size ---------|
410 *     |------------------- size -------------------->|
411 *     |==================#===========================|
412 *     ^start             ^page marked with PG_readahead
413 *
414 * To overlap application thinking time and disk I/O time, we do
415 * `readahead pipelining': Do not wait until the application consumed all
416 * readahead pages and stalled on the missing page at readahead_index;
417 * Instead, submit an asynchronous readahead I/O as soon as there are
418 * only async_size pages left in the readahead window. Normally async_size
419 * will be equal to size, for maximum pipelining.
420 *
421 * In interleaved sequential reads, concurrent streams on the same fd can
422 * be invalidating each other's readahead state. So we flag the new readahead
423 * page at (start+size-async_size) with PG_readahead, and use it as readahead
424 * indicator. The flag won't be set on already cached pages, to avoid the
425 * readahead-for-nothing fuss, saving pointless page cache lookups.
426 *
427 * prev_pos tracks the last visited byte in the _previous_ read request.
428 * It should be maintained by the caller, and will be used for detecting
429 * small random reads. Note that the readahead algorithm checks loosely
430 * for sequential patterns. Hence interleaved reads might be served as
431 * sequential ones.
432 *
433 * There is a special-case: if the first page which the application tries to
434 * read happens to be the first page of the file, it is assumed that a linear
435 * read is about to happen and the window is immediately set to the initial size
436 * based on I/O request size and the max_readahead.
437 *
438 * The code ramps up the readahead size aggressively at first, but slow down as
439 * it approaches max_readhead.
440 */
441
442static inline int ra_alloc_folio(struct readahead_control *ractl, pgoff_t index,
443		pgoff_t mark, unsigned int order, gfp_t gfp)
444{
445	int err;
446	struct folio *folio = ractl_alloc_folio(ractl, gfp, order);
447
448	if (!folio)
449		return -ENOMEM;
450	mark = round_down(mark, 1UL << order);
451	if (index == mark)
452		folio_set_readahead(folio);
453	err = filemap_add_folio(ractl->mapping, folio, index, gfp);
454	if (err) {
455		folio_put(folio);
456		return err;
457	}
458
459	ractl->_nr_pages += 1UL << order;
460	ractl->_workingset |= folio_test_workingset(folio);
461	return 0;
462}
463
464void page_cache_ra_order(struct readahead_control *ractl,
465		struct file_ra_state *ra)
466{
467	struct address_space *mapping = ractl->mapping;
468	pgoff_t start = readahead_index(ractl);
469	pgoff_t index = start;
470	unsigned int min_order = mapping_min_folio_order(mapping);
471	pgoff_t limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT;
472	pgoff_t mark = index + ra->size - ra->async_size;
473	unsigned int nofs;
474	int err = 0;
475	gfp_t gfp = readahead_gfp_mask(mapping);
476	unsigned int new_order = ra->order;
477
478	trace_page_cache_ra_order(mapping->host, start, ra);
479	if (!mapping_large_folio_support(mapping)) {
480		ra->order = 0;
481		goto fallback;
482	}
483
484	limit = min(limit, index + ra->size - 1);
485
486	new_order = min(mapping_max_folio_order(mapping), new_order);
487	new_order = min_t(unsigned int, new_order, ilog2(ra->size));
488	new_order = max(new_order, min_order);
489
490	ra->order = new_order;
491
492	/* See comment in page_cache_ra_unbounded() */
493	nofs = memalloc_nofs_save();
494	filemap_invalidate_lock_shared(mapping);
495	/*
496	 * If the new_order is greater than min_order and index is
497	 * already aligned to new_order, then this will be noop as index
498	 * aligned to new_order should also be aligned to min_order.
499	 */
500	ractl->_index = mapping_align_index(mapping, index);
501	index = readahead_index(ractl);
502
503	while (index <= limit) {
504		unsigned int order = new_order;
505
506		/* Align with smaller pages if needed */
507		if (index & ((1UL << order) - 1))
508			order = __ffs(index);
509		/* Don't allocate pages past EOF */
510		while (order > min_order && index + (1UL << order) - 1 > limit)
511			order--;
512		err = ra_alloc_folio(ractl, index, mark, order, gfp);
513		if (err)
514			break;
515		index += 1UL << order;
516	}
517
518	read_pages(ractl);
519	filemap_invalidate_unlock_shared(mapping);
520	memalloc_nofs_restore(nofs);
521
522	/*
523	 * If there were already pages in the page cache, then we may have
524	 * left some gaps.  Let the regular readahead code take care of this
525	 * situation below.
526	 */
527	if (!err)
528		return;
529fallback:
530	/*
531	 * ->readahead() may have updated readahead window size so we have to
532	 * check there's still something to read.
533	 */
534	if (ra->size > index - start)
535		do_page_cache_ra(ractl, ra->size - (index - start),
536				 ra->async_size);
537}
538
539static unsigned long ractl_max_pages(struct readahead_control *ractl,
540		unsigned long req_size)
541{
542	struct backing_dev_info *bdi = inode_to_bdi(ractl->mapping->host);
543	unsigned long max_pages = ractl->ra->ra_pages;
544
545	/*
546	 * If the request exceeds the readahead window, allow the read to
547	 * be up to the optimal hardware IO size
548	 */
549	if (req_size > max_pages && bdi->io_pages > max_pages)
550		max_pages = min(req_size, bdi->io_pages);
551	return max_pages;
552}
553
554void page_cache_sync_ra(struct readahead_control *ractl,
555		unsigned long req_count)
556{
557	pgoff_t index = readahead_index(ractl);
558	bool do_forced_ra = ractl->file && (ractl->file->f_mode & FMODE_RANDOM);
559	struct file_ra_state *ra = ractl->ra;
560	unsigned long max_pages, contig_count;
561	pgoff_t prev_index, miss;
562
563	trace_page_cache_sync_ra(ractl->mapping->host, index, ra, req_count);
564	/*
565	 * Even if readahead is disabled, issue this request as readahead
566	 * as we'll need it to satisfy the requested range. The forced
567	 * readahead will do the right thing and limit the read to just the
568	 * requested range, which we'll set to 1 page for this case.
569	 */
570	if (!ra->ra_pages || blk_cgroup_congested()) {
571		if (!ractl->file)
572			return;
573		req_count = 1;
574		do_forced_ra = true;
575	}
576
577	/* be dumb */
578	if (do_forced_ra) {
579		force_page_cache_ra(ractl, req_count);
580		return;
581	}
582
583	max_pages = ractl_max_pages(ractl, req_count);
584	prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT;
585	/*
586	 * A start of file, oversized read, or sequential cache miss:
587	 * trivial case: (index - prev_index) == 1
588	 * unaligned reads: (index - prev_index) == 0
589	 */
590	if (!index || req_count > max_pages || index - prev_index <= 1UL) {
591		ra->start = index;
592		ra->size = get_init_ra_size(req_count, max_pages);
593		ra->async_size = ra->size > req_count ? ra->size - req_count :
594							ra->size >> 1;
595		goto readit;
596	}
597
598	/*
599	 * Query the page cache and look for the traces(cached history pages)
600	 * that a sequential stream would leave behind.
601	 */
602	rcu_read_lock();
603	miss = page_cache_prev_miss(ractl->mapping, index - 1, max_pages);
604	rcu_read_unlock();
605	contig_count = index - miss - 1;
606	/*
607	 * Standalone, small random read. Read as is, and do not pollute the
608	 * readahead state.
609	 */
610	if (contig_count <= req_count) {
611		do_page_cache_ra(ractl, req_count, 0);
612		return;
613	}
614	/*
615	 * File cached from the beginning:
616	 * it is a strong indication of long-run stream (or whole-file-read)
617	 */
618	if (miss == ULONG_MAX)
619		contig_count *= 2;
620	ra->start = index;
621	ra->size = min(contig_count + req_count, max_pages);
622	ra->async_size = 1;
623readit:
624	ra->order = 0;
625	ractl->_index = ra->start;
626	page_cache_ra_order(ractl, ra);
627}
628EXPORT_SYMBOL_GPL(page_cache_sync_ra);
629
630void page_cache_async_ra(struct readahead_control *ractl,
631		struct folio *folio, unsigned long req_count)
632{
633	unsigned long max_pages;
634	struct file_ra_state *ra = ractl->ra;
635	pgoff_t index = readahead_index(ractl);
636	pgoff_t expected, start, end, aligned_end, align;
637
638	/* no readahead */
639	if (!ra->ra_pages)
640		return;
641
642	/*
643	 * Same bit is used for PG_readahead and PG_reclaim.
644	 */
645	if (folio_test_writeback(folio))
646		return;
647
648	trace_page_cache_async_ra(ractl->mapping->host, index, ra, req_count);
649	folio_clear_readahead(folio);
650
651	if (blk_cgroup_congested())
652		return;
653
654	max_pages = ractl_max_pages(ractl, req_count);
655	/*
656	 * It's the expected callback index, assume sequential access.
657	 * Ramp up sizes, and push forward the readahead window.
658	 */
659	expected = round_down(ra->start + ra->size - ra->async_size,
660			folio_nr_pages(folio));
661	if (index == expected) {
662		ra->start += ra->size;
663		/*
664		 * In the case of MADV_HUGEPAGE, the actual size might exceed
665		 * the readahead window.
666		 */
667		ra->size = max(ra->size, get_next_ra_size(ra, max_pages));
668		goto readit;
669	}
670
671	/*
672	 * Hit a marked folio without valid readahead state.
673	 * E.g. interleaved reads.
674	 * Query the pagecache for async_size, which normally equals to
675	 * readahead size. Ramp it up and use it as the new readahead size.
676	 */
677	rcu_read_lock();
678	start = page_cache_next_miss(ractl->mapping, index + 1, max_pages);
679	rcu_read_unlock();
680
681	if (!start || start - index > max_pages)
682		return;
683
684	ra->start = start;
685	ra->size = start - index;	/* old async_size */
686	ra->size += req_count;
687	ra->size = get_next_ra_size(ra, max_pages);
688readit:
689	ra->order += 2;
690	align = 1UL << min(ra->order, ffs(max_pages) - 1);
691	end = ra->start + ra->size;
692	aligned_end = round_down(end, align);
693	if (aligned_end > ra->start)
694		ra->size -= end - aligned_end;
695	ra->async_size = ra->size;
696	ractl->_index = ra->start;
697	page_cache_ra_order(ractl, ra);
698}
699EXPORT_SYMBOL_GPL(page_cache_async_ra);
700
701ssize_t ksys_readahead(int fd, loff_t offset, size_t count)
702{
703	struct file *file;
704	const struct inode *inode;
705
706	CLASS(fd, f)(fd);
707	if (fd_empty(f))
708		return -EBADF;
709
710	file = fd_file(f);
711	if (!(file->f_mode & FMODE_READ))
712		return -EBADF;
713
714	/*
715	 * The readahead() syscall is intended to run only on files
716	 * that can execute readahead. If readahead is not possible
717	 * on this file, then we must return -EINVAL.
718	 */
719	if (!file->f_mapping)
720		return -EINVAL;
721	if (!file->f_mapping->a_ops)
722		return -EINVAL;
723
724	inode = file_inode(file);
725	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
726		return -EINVAL;
727	if (IS_ANON_FILE(inode))
728		return -EINVAL;
729
730	return vfs_fadvise(fd_file(f), offset, count, POSIX_FADV_WILLNEED);
731}
732
733SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
734{
735	return ksys_readahead(fd, offset, count);
736}
737
738#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_READAHEAD)
739COMPAT_SYSCALL_DEFINE4(readahead, int, fd, compat_arg_u64_dual(offset), size_t, count)
740{
741	return ksys_readahead(fd, compat_arg_u64_glue(offset), count);
742}
743#endif
744
745/**
746 * readahead_expand - Expand a readahead request
747 * @ractl: The request to be expanded
748 * @new_start: The revised start
749 * @new_len: The revised size of the request
750 *
751 * Attempt to expand a readahead request outwards from the current size to the
752 * specified size by inserting locked pages before and after the current window
753 * to increase the size to the new window.  This may involve the insertion of
754 * THPs, in which case the window may get expanded even beyond what was
755 * requested.
756 *
757 * The algorithm will stop if it encounters a conflicting page already in the
758 * pagecache and leave a smaller expansion than requested.
759 *
760 * The caller must check for this by examining the revised @ractl object for a
761 * different expansion than was requested.
762 */
763void readahead_expand(struct readahead_control *ractl,
764		      loff_t new_start, size_t new_len)
765{
766	struct address_space *mapping = ractl->mapping;
767	struct file_ra_state *ra = ractl->ra;
768	pgoff_t new_index, new_nr_pages;
769	gfp_t gfp_mask = readahead_gfp_mask(mapping);
770	unsigned long min_nrpages = mapping_min_folio_nrpages(mapping);
771	unsigned int min_order = mapping_min_folio_order(mapping);
772
773	new_index = new_start / PAGE_SIZE;
774	/*
775	 * Readahead code should have aligned the ractl->_index to
776	 * min_nrpages before calling readahead aops.
777	 */
778	VM_BUG_ON(!IS_ALIGNED(ractl->_index, min_nrpages));
779
780	/* Expand the leading edge downwards */
781	while (ractl->_index > new_index) {
782		unsigned long index = ractl->_index - 1;
783		struct folio *folio = xa_load(&mapping->i_pages, index);
784
785		if (folio && !xa_is_value(folio))
786			return; /* Folio apparently present */
787
788		folio = ractl_alloc_folio(ractl, gfp_mask, min_order);
789		if (!folio)
790			return;
791
792		index = mapping_align_index(mapping, index);
793		if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) {
794			folio_put(folio);
795			return;
796		}
797		if (unlikely(folio_test_workingset(folio)) &&
798				!ractl->_workingset) {
799			ractl->_workingset = true;
800			psi_memstall_enter(&ractl->_pflags);
801		}
802		ractl->_nr_pages += min_nrpages;
803		ractl->_index = folio->index;
804	}
805
806	new_len += new_start - readahead_pos(ractl);
807	new_nr_pages = DIV_ROUND_UP(new_len, PAGE_SIZE);
808
809	/* Expand the trailing edge upwards */
810	while (ractl->_nr_pages < new_nr_pages) {
811		unsigned long index = ractl->_index + ractl->_nr_pages;
812		struct folio *folio = xa_load(&mapping->i_pages, index);
813
814		if (folio && !xa_is_value(folio))
815			return; /* Folio apparently present */
816
817		folio = ractl_alloc_folio(ractl, gfp_mask, min_order);
818		if (!folio)
819			return;
820
821		index = mapping_align_index(mapping, index);
822		if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) {
823			folio_put(folio);
824			return;
825		}
826		if (unlikely(folio_test_workingset(folio)) &&
827				!ractl->_workingset) {
828			ractl->_workingset = true;
829			psi_memstall_enter(&ractl->_pflags);
830		}
831		ractl->_nr_pages += min_nrpages;
832		if (ra) {
833			ra->size += min_nrpages;
834			ra->async_size += min_nrpages;
835		}
836	}
837}
838EXPORT_SYMBOL(readahead_expand);