Line data Source code
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 : #include "internal.h"
133 :
134 : /*
135 : * Initialise a struct file's readahead state. Assumes that the caller has
136 : * memset *ra to zero.
137 : */
138 : void
139 677286174 : file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
140 : {
141 677286174 : ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages;
142 677252077 : ra->prev_pos = -1;
143 677252077 : }
144 : EXPORT_SYMBOL_GPL(file_ra_state_init);
145 :
146 23530614 : static void read_pages(struct readahead_control *rac)
147 : {
148 23530614 : const struct address_space_operations *aops = rac->mapping->a_ops;
149 23530614 : struct folio *folio;
150 23530614 : struct blk_plug plug;
151 :
152 23530614 : if (!readahead_count(rac))
153 9406770 : return;
154 :
155 14123844 : if (unlikely(rac->_workingset))
156 : psi_memstall_enter(&rac->_pflags);
157 14123844 : blk_start_plug(&plug);
158 :
159 14123813 : if (aops->readahead) {
160 14123813 : aops->readahead(rac);
161 : /*
162 : * Clean up the remaining folios. The sizes in ->ra
163 : * may be used to size the next readahead, so make sure
164 : * they accurately reflect what happened.
165 : */
166 14123921 : while ((folio = readahead_folio(rac)) != NULL) {
167 108 : unsigned long nr = folio_nr_pages(folio);
168 :
169 108 : folio_get(folio);
170 108 : rac->ra->size -= nr;
171 108 : if (rac->ra->async_size >= nr) {
172 29 : rac->ra->async_size -= nr;
173 29 : filemap_remove_folio(folio);
174 : }
175 108 : folio_unlock(folio);
176 108 : folio_put(folio);
177 : }
178 : } else {
179 0 : while ((folio = readahead_folio(rac)) != NULL)
180 0 : aops->read_folio(rac->file, folio);
181 : }
182 :
183 14123764 : blk_finish_plug(&plug);
184 14123876 : if (unlikely(rac->_workingset))
185 : psi_memstall_leave(&rac->_pflags);
186 14123876 : rac->_workingset = false;
187 :
188 14123876 : BUG_ON(readahead_count(rac));
189 : }
190 :
191 : /**
192 : * page_cache_ra_unbounded - Start unchecked readahead.
193 : * @ractl: Readahead control.
194 : * @nr_to_read: The number of pages to read.
195 : * @lookahead_size: Where to start the next readahead.
196 : *
197 : * This function is for filesystems to call when they want to start
198 : * readahead beyond a file's stated i_size. This is almost certainly
199 : * not the function you want to call. Use page_cache_async_readahead()
200 : * or page_cache_sync_readahead() instead.
201 : *
202 : * Context: File is referenced by caller. Mutexes may be held by caller.
203 : * May sleep, but will not reenter filesystem to reclaim memory.
204 : */
205 10173482 : void page_cache_ra_unbounded(struct readahead_control *ractl,
206 : unsigned long nr_to_read, unsigned long lookahead_size)
207 : {
208 10173482 : struct address_space *mapping = ractl->mapping;
209 10173482 : unsigned long index = readahead_index(ractl);
210 10173482 : gfp_t gfp_mask = readahead_gfp_mask(mapping);
211 10173482 : unsigned long i;
212 :
213 : /*
214 : * Partway through the readahead operation, we will have added
215 : * locked pages to the page cache, but will not yet have submitted
216 : * them for I/O. Adding another page may need to allocate memory,
217 : * which can trigger memory reclaim. Telling the VM we're in
218 : * the middle of a filesystem operation will cause it to not
219 : * touch file-backed pages, preventing a deadlock. Most (all?)
220 : * filesystems already specify __GFP_NOFS in their mapping's
221 : * gfp_mask, but let's be explicit here.
222 : */
223 10173482 : unsigned int nofs = memalloc_nofs_save();
224 :
225 10173482 : filemap_invalidate_lock_shared(mapping);
226 : /*
227 : * Preallocate as many pages as we will need.
228 : */
229 59070883 : for (i = 0; i < nr_to_read; i++) {
230 38723936 : struct folio *folio = xa_load(&mapping->i_pages, index + i);
231 :
232 38723757 : if (folio && !xa_is_value(folio)) {
233 : /*
234 : * Page already present? Kick off the current batch
235 : * of contiguous pages before continuing with the
236 : * next batch. This page may be the one we would
237 : * have intended to mark as Readahead, but we don't
238 : * have a stable reference to this page, and it's
239 : * not worth getting one just for that.
240 : */
241 8924178 : read_pages(ractl);
242 8924175 : ractl->_index++;
243 8924175 : i = ractl->_index + ractl->_nr_pages - index - 1;
244 8924175 : continue;
245 : }
246 :
247 29799579 : folio = filemap_alloc_folio(gfp_mask, 0);
248 29799431 : if (!folio)
249 : break;
250 29799259 : if (filemap_add_folio(mapping, folio, index + i,
251 : gfp_mask) < 0) {
252 908 : folio_put(folio);
253 908 : read_pages(ractl);
254 908 : ractl->_index++;
255 908 : i = ractl->_index + ractl->_nr_pages - index - 1;
256 908 : continue;
257 : }
258 29798775 : if (i == nr_to_read - lookahead_size)
259 740298 : folio_set_readahead(folio);
260 29798783 : ractl->_workingset |= folio_test_workingset(folio);
261 29798783 : ractl->_nr_pages++;
262 : }
263 :
264 : /*
265 : * Now start the IO. We ignore I/O errors - if the folio is not
266 : * uptodate then the caller will launch read_folio again, and
267 : * will then handle the error.
268 : */
269 10173637 : read_pages(ractl);
270 10173618 : filemap_invalidate_unlock_shared(mapping);
271 10173637 : memalloc_nofs_restore(nofs);
272 10173637 : }
273 : EXPORT_SYMBOL_GPL(page_cache_ra_unbounded);
274 :
275 : /*
276 : * do_page_cache_ra() actually reads a chunk of disk. It allocates
277 : * the pages first, then submits them for I/O. This avoids the very bad
278 : * behaviour which would occur if page allocations are causing VM writeback.
279 : * We really don't want to intermingle reads and writes like that.
280 : */
281 10183704 : static void do_page_cache_ra(struct readahead_control *ractl,
282 : unsigned long nr_to_read, unsigned long lookahead_size)
283 : {
284 10183704 : struct inode *inode = ractl->mapping->host;
285 10183704 : unsigned long index = readahead_index(ractl);
286 10183704 : loff_t isize = i_size_read(inode);
287 10183704 : pgoff_t end_index; /* The last page we want to read */
288 :
289 10183704 : if (isize == 0)
290 : return;
291 :
292 10183704 : end_index = (isize - 1) >> PAGE_SHIFT;
293 10183704 : if (index > end_index)
294 : return;
295 : /* Don't read past the page containing the last byte of the file */
296 10173451 : if (nr_to_read > end_index - index)
297 1815017 : nr_to_read = end_index - index + 1;
298 :
299 10173451 : page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size);
300 : }
301 :
302 : /*
303 : * Chunk the readahead into 2 megabyte units, so that we don't pin too much
304 : * memory at once.
305 : */
306 790647 : void force_page_cache_ra(struct readahead_control *ractl,
307 : unsigned long nr_to_read)
308 : {
309 790647 : struct address_space *mapping = ractl->mapping;
310 790647 : struct file_ra_state *ra = ractl->ra;
311 790647 : struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
312 790639 : unsigned long max_pages, index;
313 :
314 790639 : if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead))
315 : return;
316 :
317 : /*
318 : * If the request exceeds the readahead window, allow the read to
319 : * be up to the optimal hardware IO size
320 : */
321 790639 : index = readahead_index(ractl);
322 790639 : max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
323 790639 : nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
324 1581337 : while (nr_to_read) {
325 790639 : unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
326 :
327 790639 : if (this_chunk > nr_to_read)
328 790657 : this_chunk = nr_to_read;
329 790639 : ractl->_index = index;
330 790639 : do_page_cache_ra(ractl, this_chunk, 0);
331 :
332 790698 : index += this_chunk;
333 790698 : nr_to_read -= this_chunk;
334 : }
335 : }
336 :
337 : /*
338 : * Set the initial window size, round to next power of 2 and square
339 : * for small size, x 4 for medium, and x 2 for large
340 : * for 128k (32 page) max ra
341 : * 1-2 page = 16k, 3-4 page 32k, 5-8 page = 64k, > 8 page = 128k initial
342 : */
343 1757827 : static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
344 : {
345 1757827 : unsigned long newsize = roundup_pow_of_two(size);
346 :
347 1757827 : if (newsize <= max / 32)
348 377948 : newsize = newsize * 4;
349 1379879 : else if (newsize <= max / 4)
350 1140035 : newsize = newsize * 2;
351 : else
352 : newsize = max;
353 :
354 1757827 : return newsize;
355 : }
356 :
357 : /*
358 : * Get the previous window size, ramp it up, and
359 : * return it as the new window size.
360 : */
361 : static unsigned long get_next_ra_size(struct file_ra_state *ra,
362 : unsigned long max)
363 : {
364 453408 : unsigned long cur = ra->size;
365 :
366 453408 : if (cur < max / 16)
367 280 : return 4 * cur;
368 453128 : if (cur <= max / 2)
369 234128 : return 2 * cur;
370 : return max;
371 : }
372 :
373 : /*
374 : * On-demand readahead design.
375 : *
376 : * The fields in struct file_ra_state represent the most-recently-executed
377 : * readahead attempt:
378 : *
379 : * |<----- async_size ---------|
380 : * |------------------- size -------------------->|
381 : * |==================#===========================|
382 : * ^start ^page marked with PG_readahead
383 : *
384 : * To overlap application thinking time and disk I/O time, we do
385 : * `readahead pipelining': Do not wait until the application consumed all
386 : * readahead pages and stalled on the missing page at readahead_index;
387 : * Instead, submit an asynchronous readahead I/O as soon as there are
388 : * only async_size pages left in the readahead window. Normally async_size
389 : * will be equal to size, for maximum pipelining.
390 : *
391 : * In interleaved sequential reads, concurrent streams on the same fd can
392 : * be invalidating each other's readahead state. So we flag the new readahead
393 : * page at (start+size-async_size) with PG_readahead, and use it as readahead
394 : * indicator. The flag won't be set on already cached pages, to avoid the
395 : * readahead-for-nothing fuss, saving pointless page cache lookups.
396 : *
397 : * prev_pos tracks the last visited byte in the _previous_ read request.
398 : * It should be maintained by the caller, and will be used for detecting
399 : * small random reads. Note that the readahead algorithm checks loosely
400 : * for sequential patterns. Hence interleaved reads might be served as
401 : * sequential ones.
402 : *
403 : * There is a special-case: if the first page which the application tries to
404 : * read happens to be the first page of the file, it is assumed that a linear
405 : * read is about to happen and the window is immediately set to the initial size
406 : * based on I/O request size and the max_readahead.
407 : *
408 : * The code ramps up the readahead size aggressively at first, but slow down as
409 : * it approaches max_readhead.
410 : */
411 :
412 : /*
413 : * Count contiguously cached pages from @index-1 to @index-@max,
414 : * this count is a conservative estimation of
415 : * - length of the sequential read sequence, or
416 : * - thrashing threshold in memory tight systems
417 : */
418 7261842 : static pgoff_t count_history_pages(struct address_space *mapping,
419 : pgoff_t index, unsigned long max)
420 : {
421 7261842 : pgoff_t head;
422 :
423 7261842 : rcu_read_lock();
424 7261857 : head = page_cache_prev_miss(mapping, index - 1, max);
425 7261800 : rcu_read_unlock();
426 :
427 7261801 : return index - 1 - head;
428 : }
429 :
430 : /*
431 : * page cache context based readahead
432 : */
433 7261856 : static int try_context_readahead(struct address_space *mapping,
434 : struct file_ra_state *ra,
435 : pgoff_t index,
436 : unsigned long req_size,
437 : unsigned long max)
438 : {
439 7261856 : pgoff_t size;
440 :
441 7261856 : size = count_history_pages(mapping, index, max);
442 :
443 : /*
444 : * not enough history pages:
445 : * it could be a random read
446 : */
447 7261800 : if (size <= req_size)
448 : return 0;
449 :
450 : /*
451 : * starts from beginning of file:
452 : * it is a strong indication of long-run stream (or whole-file-read)
453 : */
454 73889 : if (size >= index)
455 15719 : size *= 2;
456 :
457 73889 : ra->start = index;
458 73889 : ra->size = min(size + req_size, max);
459 73889 : ra->async_size = 1;
460 :
461 73889 : return 1;
462 : }
463 :
464 : /*
465 : * There are some parts of the kernel which assume that PMD entries
466 : * are exactly HPAGE_PMD_ORDER. Those should be fixed, but until then,
467 : * limit the maximum allocation order to PMD size. I'm not aware of any
468 : * assumptions about maximum order if THP are disabled, but 8 seems like
469 : * a good order (that's 1MB if you're using 4kB pages)
470 : */
471 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 : #define MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER
473 : #else
474 : #define MAX_PAGECACHE_ORDER 8
475 : #endif
476 :
477 19422759 : static inline int ra_alloc_folio(struct readahead_control *ractl, pgoff_t index,
478 : pgoff_t mark, unsigned int order, gfp_t gfp)
479 : {
480 19422759 : int err;
481 19422759 : struct folio *folio = filemap_alloc_folio(gfp, order);
482 :
483 19422659 : if (!folio)
484 : return -ENOMEM;
485 19422533 : mark = round_up(mark, 1UL << order);
486 19422533 : if (index == mark)
487 1614362 : folio_set_readahead(folio);
488 19422530 : err = filemap_add_folio(ractl->mapping, folio, index, gfp);
489 19422695 : if (err) {
490 1890046 : folio_put(folio);
491 1890046 : return err;
492 : }
493 :
494 17532649 : ractl->_nr_pages += 1UL << order;
495 17532649 : ractl->_workingset |= folio_test_workingset(folio);
496 17532649 : return 0;
497 : }
498 :
499 4746853 : void page_cache_ra_order(struct readahead_control *ractl,
500 : struct file_ra_state *ra, unsigned int new_order)
501 : {
502 4746853 : struct address_space *mapping = ractl->mapping;
503 4746853 : pgoff_t index = readahead_index(ractl);
504 4746853 : pgoff_t limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT;
505 4746853 : pgoff_t mark = index + ra->size - ra->async_size;
506 4746853 : int err = 0;
507 4746853 : gfp_t gfp = readahead_gfp_mask(mapping);
508 :
509 9493706 : if (!mapping_large_folio_support(mapping) || ra->size < 4)
510 314951 : goto fallback;
511 :
512 4431902 : limit = min(limit, index + ra->size - 1);
513 :
514 4431902 : if (new_order < MAX_PAGECACHE_ORDER) {
515 4431900 : new_order += 2;
516 4431900 : if (new_order > MAX_PAGECACHE_ORDER)
517 0 : new_order = MAX_PAGECACHE_ORDER;
518 4511477 : while ((1 << new_order) > ra->size)
519 79577 : new_order--;
520 : }
521 :
522 4431902 : filemap_invalidate_lock_shared(mapping);
523 21964535 : while (index <= limit) {
524 19422752 : unsigned int order = new_order;
525 :
526 : /* Align with smaller pages if needed */
527 19422752 : if (index & ((1UL << order) - 1)) {
528 5626969 : order = __ffs(index);
529 5626969 : if (order == 1)
530 1857024 : order = 0;
531 : }
532 : /* Don't allocate pages past EOF */
533 21346906 : while (index + (1UL << order) - 1 > limit) {
534 1924154 : if (--order == 1)
535 : order = 0;
536 : }
537 19422752 : err = ra_alloc_folio(ractl, index, mark, order, gfp);
538 19422793 : if (err)
539 : break;
540 : index += 1UL << order;
541 : }
542 :
543 4431957 : if (index > limit) {
544 2541783 : ra->size += index - limit - 1;
545 2541783 : ra->async_size += index - limit - 1;
546 : }
547 :
548 4431957 : read_pages(ractl);
549 4431950 : filemap_invalidate_unlock_shared(mapping);
550 :
551 : /*
552 : * If there were already pages in the page cache, then we may have
553 : * left some gaps. Let the regular readahead code take care of this
554 : * situation.
555 : */
556 4431958 : if (!err)
557 : return;
558 1890174 : fallback:
559 2205125 : do_page_cache_ra(ractl, ra->size, ra->async_size);
560 : }
561 :
562 : /*
563 : * A minimal readahead algorithm for trivial sequential/random reads.
564 : */
565 9440055 : static void ondemand_readahead(struct readahead_control *ractl,
566 : struct folio *folio, unsigned long req_size)
567 : {
568 9440055 : struct backing_dev_info *bdi = inode_to_bdi(ractl->mapping->host);
569 9440173 : struct file_ra_state *ra = ractl->ra;
570 9440173 : unsigned long max_pages = ra->ra_pages;
571 9440173 : unsigned long add_pages;
572 9440173 : pgoff_t index = readahead_index(ractl);
573 9440173 : pgoff_t expected, prev_index;
574 9440173 : unsigned int order = folio ? folio_order(folio) : 0;
575 :
576 : /*
577 : * If the request exceeds the readahead window, allow the read to
578 : * be up to the optimal hardware IO size
579 : */
580 9440173 : if (req_size > max_pages && bdi->io_pages > max_pages)
581 19635 : max_pages = min(req_size, bdi->io_pages);
582 :
583 : /*
584 : * start of file
585 : */
586 9440173 : if (!index)
587 642065 : goto initial_readahead;
588 :
589 : /*
590 : * It's the expected callback index, assume sequential access.
591 : * Ramp up sizes, and push forward the readahead window.
592 : */
593 8798108 : expected = round_up(ra->start + ra->size - ra->async_size,
594 : 1UL << order);
595 8798108 : if (index == expected || index == (ra->start + ra->size)) {
596 282006 : ra->start += ra->size;
597 282006 : ra->size = get_next_ra_size(ra, max_pages);
598 282006 : ra->async_size = ra->size;
599 282006 : goto readit;
600 : }
601 :
602 : /*
603 : * Hit a marked folio without valid readahead state.
604 : * E.g. interleaved reads.
605 : * Query the pagecache for async_size, which normally equals to
606 : * readahead size. Ramp it up and use it as the new readahead size.
607 : */
608 8516102 : if (folio) {
609 138490 : pgoff_t start;
610 :
611 138490 : rcu_read_lock();
612 138491 : start = page_cache_next_miss(ractl->mapping, index + 1,
613 : max_pages);
614 138491 : rcu_read_unlock();
615 :
616 138487 : if (!start || start - index > max_pages)
617 : return;
618 :
619 138487 : ra->start = start;
620 138487 : ra->size = start - index; /* old async_size */
621 138487 : ra->size += req_size;
622 138487 : ra->size = get_next_ra_size(ra, max_pages);
623 138487 : ra->async_size = ra->size;
624 138487 : goto readit;
625 : }
626 :
627 : /*
628 : * oversize read
629 : */
630 8377612 : if (req_size > max_pages)
631 1124 : goto initial_readahead;
632 :
633 : /*
634 : * sequential cache miss
635 : * trivial case: (index - prev_index) == 1
636 : * unaligned reads: (index - prev_index) == 0
637 : */
638 8376488 : prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT;
639 8376488 : if (index - prev_index <= 1UL)
640 1114619 : goto initial_readahead;
641 :
642 : /*
643 : * Query the page cache and look for the traces(cached history pages)
644 : * that a sequential stream would leave behind.
645 : */
646 7261869 : if (try_context_readahead(ractl->mapping, ra, index, req_size,
647 : max_pages))
648 73889 : goto readit;
649 :
650 : /*
651 : * standalone, small random read
652 : * Read as is, and do not pollute the readahead state.
653 : */
654 7187897 : do_page_cache_ra(ractl, req_size, 0);
655 7187897 : return;
656 :
657 1757808 : initial_readahead:
658 1757808 : ra->start = index;
659 1757808 : ra->size = get_init_ra_size(req_size, max_pages);
660 1757808 : ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
661 :
662 2252190 : readit:
663 : /*
664 : * Will this read hit the readahead marker made by itself?
665 : * If so, trigger the readahead marker hit now, and merge
666 : * the resulted next readahead window into the current one.
667 : * Take care of maximum IO pages as above.
668 : */
669 2252190 : if (index == ra->start && ra->size == ra->async_size) {
670 32915 : add_pages = get_next_ra_size(ra, max_pages);
671 32915 : if (ra->size + add_pages <= max_pages) {
672 14439 : ra->async_size = add_pages;
673 14439 : ra->size += add_pages;
674 : } else {
675 18476 : ra->size = max_pages;
676 18476 : ra->async_size = max_pages >> 1;
677 : }
678 : }
679 :
680 2252190 : ractl->_index = ra->start;
681 2252190 : page_cache_ra_order(ractl, ra, order);
682 : }
683 :
684 9846275 : void page_cache_sync_ra(struct readahead_control *ractl,
685 : unsigned long req_count)
686 : {
687 9846275 : bool do_forced_ra = ractl->file && (ractl->file->f_mode & FMODE_RANDOM);
688 :
689 : /*
690 : * Even if readahead is disabled, issue this request as readahead
691 : * as we'll need it to satisfy the requested range. The forced
692 : * readahead will do the right thing and limit the read to just the
693 : * requested range, which we'll set to 1 page for this case.
694 : */
695 9846275 : if (!ractl->ra->ra_pages || blk_cgroup_congested()) {
696 2 : if (!ractl->file)
697 : return;
698 : req_count = 1;
699 : do_forced_ra = true;
700 : }
701 :
702 : /* be dumb */
703 9846194 : if (do_forced_ra) {
704 790612 : force_page_cache_ra(ractl, req_count);
705 790612 : return;
706 : }
707 :
708 9055505 : ondemand_readahead(ractl, NULL, req_count);
709 : }
710 : EXPORT_SYMBOL_GPL(page_cache_sync_ra);
711 :
712 384618 : void page_cache_async_ra(struct readahead_control *ractl,
713 : struct folio *folio, unsigned long req_count)
714 : {
715 : /* no readahead */
716 384618 : if (!ractl->ra->ra_pages)
717 : return;
718 :
719 : /*
720 : * Same bit is used for PG_readahead and PG_reclaim.
721 : */
722 384618 : if (folio_test_writeback(folio))
723 : return;
724 :
725 384549 : folio_clear_readahead(folio);
726 :
727 384549 : if (blk_cgroup_congested())
728 : return;
729 :
730 384535 : ondemand_readahead(ractl, folio, req_count);
731 : }
732 : EXPORT_SYMBOL_GPL(page_cache_async_ra);
733 :
734 0 : ssize_t ksys_readahead(int fd, loff_t offset, size_t count)
735 : {
736 0 : ssize_t ret;
737 0 : struct fd f;
738 :
739 0 : ret = -EBADF;
740 0 : f = fdget(fd);
741 0 : if (!f.file || !(f.file->f_mode & FMODE_READ))
742 0 : goto out;
743 :
744 : /*
745 : * The readahead() syscall is intended to run only on files
746 : * that can execute readahead. If readahead is not possible
747 : * on this file, then we must return -EINVAL.
748 : */
749 0 : ret = -EINVAL;
750 0 : if (!f.file->f_mapping || !f.file->f_mapping->a_ops ||
751 0 : !S_ISREG(file_inode(f.file)->i_mode))
752 0 : goto out;
753 :
754 0 : ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED);
755 0 : out:
756 0 : fdput(f);
757 0 : return ret;
758 : }
759 :
760 0 : SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
761 : {
762 0 : return ksys_readahead(fd, offset, count);
763 : }
764 :
765 : #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_READAHEAD)
766 : COMPAT_SYSCALL_DEFINE4(readahead, int, fd, compat_arg_u64_dual(offset), size_t, count)
767 : {
768 : return ksys_readahead(fd, compat_arg_u64_glue(offset), count);
769 : }
770 : #endif
771 :
772 : /**
773 : * readahead_expand - Expand a readahead request
774 : * @ractl: The request to be expanded
775 : * @new_start: The revised start
776 : * @new_len: The revised size of the request
777 : *
778 : * Attempt to expand a readahead request outwards from the current size to the
779 : * specified size by inserting locked pages before and after the current window
780 : * to increase the size to the new window. This may involve the insertion of
781 : * THPs, in which case the window may get expanded even beyond what was
782 : * requested.
783 : *
784 : * The algorithm will stop if it encounters a conflicting page already in the
785 : * pagecache and leave a smaller expansion than requested.
786 : *
787 : * The caller must check for this by examining the revised @ractl object for a
788 : * different expansion than was requested.
789 : */
790 0 : void readahead_expand(struct readahead_control *ractl,
791 : loff_t new_start, size_t new_len)
792 : {
793 0 : struct address_space *mapping = ractl->mapping;
794 0 : struct file_ra_state *ra = ractl->ra;
795 0 : pgoff_t new_index, new_nr_pages;
796 0 : gfp_t gfp_mask = readahead_gfp_mask(mapping);
797 :
798 0 : new_index = new_start / PAGE_SIZE;
799 :
800 : /* Expand the leading edge downwards */
801 0 : while (ractl->_index > new_index) {
802 0 : unsigned long index = ractl->_index - 1;
803 0 : struct folio *folio = xa_load(&mapping->i_pages, index);
804 :
805 0 : if (folio && !xa_is_value(folio))
806 : return; /* Folio apparently present */
807 :
808 0 : folio = filemap_alloc_folio(gfp_mask, 0);
809 0 : if (!folio)
810 : return;
811 0 : if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) {
812 0 : folio_put(folio);
813 0 : return;
814 : }
815 0 : if (unlikely(folio_test_workingset(folio)) &&
816 0 : !ractl->_workingset) {
817 0 : ractl->_workingset = true;
818 0 : psi_memstall_enter(&ractl->_pflags);
819 : }
820 0 : ractl->_nr_pages++;
821 0 : ractl->_index = folio->index;
822 : }
823 :
824 0 : new_len += new_start - readahead_pos(ractl);
825 0 : new_nr_pages = DIV_ROUND_UP(new_len, PAGE_SIZE);
826 :
827 : /* Expand the trailing edge upwards */
828 0 : while (ractl->_nr_pages < new_nr_pages) {
829 0 : unsigned long index = ractl->_index + ractl->_nr_pages;
830 0 : struct folio *folio = xa_load(&mapping->i_pages, index);
831 :
832 0 : if (folio && !xa_is_value(folio))
833 : return; /* Folio apparently present */
834 :
835 0 : folio = filemap_alloc_folio(gfp_mask, 0);
836 0 : if (!folio)
837 : return;
838 0 : if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) {
839 0 : folio_put(folio);
840 0 : return;
841 : }
842 0 : if (unlikely(folio_test_workingset(folio)) &&
843 0 : !ractl->_workingset) {
844 0 : ractl->_workingset = true;
845 0 : psi_memstall_enter(&ractl->_pflags);
846 : }
847 0 : ractl->_nr_pages++;
848 0 : if (ra) {
849 0 : ra->size++;
850 0 : ra->async_size++;
851 : }
852 : }
853 : }
854 : EXPORT_SYMBOL(readahead_expand);
|
