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 ->readpage() 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 ->readpage() 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 : * ->readpage() which may be less efficient.
114 : */
115 :
116 : #include <linux/kernel.h>
117 : #include <linux/dax.h>
118 : #include <linux/gfp.h>
119 : #include <linux/export.h>
120 : #include <linux/backing-dev.h>
121 : #include <linux/task_io_accounting_ops.h>
122 : #include <linux/pagevec.h>
123 : #include <linux/pagemap.h>
124 : #include <linux/syscalls.h>
125 : #include <linux/file.h>
126 : #include <linux/mm_inline.h>
127 : #include <linux/blk-cgroup.h>
128 : #include <linux/fadvise.h>
129 : #include <linux/sched/mm.h>
130 :
131 : #include "internal.h"
132 :
133 : /*
134 : * Initialise a struct file's readahead state. Assumes that the caller has
135 : * memset *ra to zero.
136 : */
137 : void
138 0 : file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
139 : {
140 0 : ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages;
141 0 : ra->prev_pos = -1;
142 0 : }
143 : EXPORT_SYMBOL_GPL(file_ra_state_init);
144 :
145 0 : static void read_pages(struct readahead_control *rac)
146 : {
147 0 : const struct address_space_operations *aops = rac->mapping->a_ops;
148 : struct page *page;
149 : struct blk_plug plug;
150 :
151 0 : if (!readahead_count(rac))
152 0 : return;
153 :
154 0 : blk_start_plug(&plug);
155 :
156 0 : if (aops->readahead) {
157 0 : aops->readahead(rac);
158 : /*
159 : * Clean up the remaining pages. The sizes in ->ra
160 : * may be used to size the next readahead, so make sure
161 : * they accurately reflect what happened.
162 : */
163 0 : while ((page = readahead_page(rac))) {
164 0 : rac->ra->size -= 1;
165 0 : if (rac->ra->async_size > 0) {
166 0 : rac->ra->async_size -= 1;
167 0 : delete_from_page_cache(page);
168 : }
169 0 : unlock_page(page);
170 0 : put_page(page);
171 : }
172 : } else {
173 0 : while ((page = readahead_page(rac))) {
174 0 : aops->readpage(rac->file, page);
175 0 : put_page(page);
176 : }
177 : }
178 :
179 0 : blk_finish_plug(&plug);
180 :
181 0 : BUG_ON(readahead_count(rac));
182 : }
183 :
184 : /**
185 : * page_cache_ra_unbounded - Start unchecked readahead.
186 : * @ractl: Readahead control.
187 : * @nr_to_read: The number of pages to read.
188 : * @lookahead_size: Where to start the next readahead.
189 : *
190 : * This function is for filesystems to call when they want to start
191 : * readahead beyond a file's stated i_size. This is almost certainly
192 : * not the function you want to call. Use page_cache_async_readahead()
193 : * or page_cache_sync_readahead() instead.
194 : *
195 : * Context: File is referenced by caller. Mutexes may be held by caller.
196 : * May sleep, but will not reenter filesystem to reclaim memory.
197 : */
198 0 : void page_cache_ra_unbounded(struct readahead_control *ractl,
199 : unsigned long nr_to_read, unsigned long lookahead_size)
200 : {
201 0 : struct address_space *mapping = ractl->mapping;
202 0 : unsigned long index = readahead_index(ractl);
203 0 : gfp_t gfp_mask = readahead_gfp_mask(mapping);
204 : unsigned long i;
205 :
206 : /*
207 : * Partway through the readahead operation, we will have added
208 : * locked pages to the page cache, but will not yet have submitted
209 : * them for I/O. Adding another page may need to allocate memory,
210 : * which can trigger memory reclaim. Telling the VM we're in
211 : * the middle of a filesystem operation will cause it to not
212 : * touch file-backed pages, preventing a deadlock. Most (all?)
213 : * filesystems already specify __GFP_NOFS in their mapping's
214 : * gfp_mask, but let's be explicit here.
215 : */
216 0 : unsigned int nofs = memalloc_nofs_save();
217 :
218 0 : filemap_invalidate_lock_shared(mapping);
219 : /*
220 : * Preallocate as many pages as we will need.
221 : */
222 0 : for (i = 0; i < nr_to_read; i++) {
223 0 : struct folio *folio = xa_load(&mapping->i_pages, index + i);
224 :
225 0 : if (folio && !xa_is_value(folio)) {
226 : /*
227 : * Page already present? Kick off the current batch
228 : * of contiguous pages before continuing with the
229 : * next batch. This page may be the one we would
230 : * have intended to mark as Readahead, but we don't
231 : * have a stable reference to this page, and it's
232 : * not worth getting one just for that.
233 : */
234 0 : read_pages(ractl);
235 0 : ractl->_index++;
236 0 : i = ractl->_index + ractl->_nr_pages - index - 1;
237 0 : continue;
238 : }
239 :
240 0 : folio = filemap_alloc_folio(gfp_mask, 0);
241 0 : if (!folio)
242 : break;
243 0 : if (filemap_add_folio(mapping, folio, index + i,
244 : gfp_mask) < 0) {
245 0 : folio_put(folio);
246 0 : read_pages(ractl);
247 0 : ractl->_index++;
248 0 : i = ractl->_index + ractl->_nr_pages - index - 1;
249 0 : continue;
250 : }
251 0 : if (i == nr_to_read - lookahead_size)
252 : folio_set_readahead(folio);
253 0 : ractl->_nr_pages++;
254 : }
255 :
256 : /*
257 : * Now start the IO. We ignore I/O errors - if the page is not
258 : * uptodate then the caller will launch readpage again, and
259 : * will then handle the error.
260 : */
261 0 : read_pages(ractl);
262 0 : filemap_invalidate_unlock_shared(mapping);
263 0 : memalloc_nofs_restore(nofs);
264 0 : }
265 : EXPORT_SYMBOL_GPL(page_cache_ra_unbounded);
266 :
267 : /*
268 : * do_page_cache_ra() actually reads a chunk of disk. It allocates
269 : * the pages first, then submits them for I/O. This avoids the very bad
270 : * behaviour which would occur if page allocations are causing VM writeback.
271 : * We really don't want to intermingle reads and writes like that.
272 : */
273 0 : static void do_page_cache_ra(struct readahead_control *ractl,
274 : unsigned long nr_to_read, unsigned long lookahead_size)
275 : {
276 0 : struct inode *inode = ractl->mapping->host;
277 0 : unsigned long index = readahead_index(ractl);
278 0 : loff_t isize = i_size_read(inode);
279 : pgoff_t end_index; /* The last page we want to read */
280 :
281 0 : if (isize == 0)
282 : return;
283 :
284 0 : end_index = (isize - 1) >> PAGE_SHIFT;
285 0 : if (index > end_index)
286 : return;
287 : /* Don't read past the page containing the last byte of the file */
288 0 : if (nr_to_read > end_index - index)
289 0 : nr_to_read = end_index - index + 1;
290 :
291 0 : page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size);
292 : }
293 :
294 : /*
295 : * Chunk the readahead into 2 megabyte units, so that we don't pin too much
296 : * memory at once.
297 : */
298 0 : void force_page_cache_ra(struct readahead_control *ractl,
299 : unsigned long nr_to_read)
300 : {
301 0 : struct address_space *mapping = ractl->mapping;
302 0 : struct file_ra_state *ra = ractl->ra;
303 0 : struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
304 : unsigned long max_pages, index;
305 :
306 0 : if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readahead))
307 : return;
308 :
309 : /*
310 : * If the request exceeds the readahead window, allow the read to
311 : * be up to the optimal hardware IO size
312 : */
313 0 : index = readahead_index(ractl);
314 0 : max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
315 0 : nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
316 0 : while (nr_to_read) {
317 0 : unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
318 :
319 0 : if (this_chunk > nr_to_read)
320 0 : this_chunk = nr_to_read;
321 0 : ractl->_index = index;
322 0 : do_page_cache_ra(ractl, this_chunk, 0);
323 :
324 0 : index += this_chunk;
325 0 : nr_to_read -= this_chunk;
326 : }
327 : }
328 :
329 : /*
330 : * Set the initial window size, round to next power of 2 and square
331 : * for small size, x 4 for medium, and x 2 for large
332 : * for 128k (32 page) max ra
333 : * 1-2 page = 16k, 3-4 page 32k, 5-8 page = 64k, > 8 page = 128k initial
334 : */
335 0 : static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
336 : {
337 0 : unsigned long newsize = roundup_pow_of_two(size);
338 :
339 0 : if (newsize <= max / 32)
340 0 : newsize = newsize * 4;
341 0 : else if (newsize <= max / 4)
342 0 : newsize = newsize * 2;
343 : else
344 : newsize = max;
345 :
346 0 : return newsize;
347 : }
348 :
349 : /*
350 : * Get the previous window size, ramp it up, and
351 : * return it as the new window size.
352 : */
353 : static unsigned long get_next_ra_size(struct file_ra_state *ra,
354 : unsigned long max)
355 : {
356 0 : unsigned long cur = ra->size;
357 :
358 0 : if (cur < max / 16)
359 0 : return 4 * cur;
360 0 : if (cur <= max / 2)
361 0 : return 2 * cur;
362 : return max;
363 : }
364 :
365 : /*
366 : * On-demand readahead design.
367 : *
368 : * The fields in struct file_ra_state represent the most-recently-executed
369 : * readahead attempt:
370 : *
371 : * |<----- async_size ---------|
372 : * |------------------- size -------------------->|
373 : * |==================#===========================|
374 : * ^start ^page marked with PG_readahead
375 : *
376 : * To overlap application thinking time and disk I/O time, we do
377 : * `readahead pipelining': Do not wait until the application consumed all
378 : * readahead pages and stalled on the missing page at readahead_index;
379 : * Instead, submit an asynchronous readahead I/O as soon as there are
380 : * only async_size pages left in the readahead window. Normally async_size
381 : * will be equal to size, for maximum pipelining.
382 : *
383 : * In interleaved sequential reads, concurrent streams on the same fd can
384 : * be invalidating each other's readahead state. So we flag the new readahead
385 : * page at (start+size-async_size) with PG_readahead, and use it as readahead
386 : * indicator. The flag won't be set on already cached pages, to avoid the
387 : * readahead-for-nothing fuss, saving pointless page cache lookups.
388 : *
389 : * prev_pos tracks the last visited byte in the _previous_ read request.
390 : * It should be maintained by the caller, and will be used for detecting
391 : * small random reads. Note that the readahead algorithm checks loosely
392 : * for sequential patterns. Hence interleaved reads might be served as
393 : * sequential ones.
394 : *
395 : * There is a special-case: if the first page which the application tries to
396 : * read happens to be the first page of the file, it is assumed that a linear
397 : * read is about to happen and the window is immediately set to the initial size
398 : * based on I/O request size and the max_readahead.
399 : *
400 : * The code ramps up the readahead size aggressively at first, but slow down as
401 : * it approaches max_readhead.
402 : */
403 :
404 : /*
405 : * Count contiguously cached pages from @index-1 to @index-@max,
406 : * this count is a conservative estimation of
407 : * - length of the sequential read sequence, or
408 : * - thrashing threshold in memory tight systems
409 : */
410 : static pgoff_t count_history_pages(struct address_space *mapping,
411 : pgoff_t index, unsigned long max)
412 : {
413 : pgoff_t head;
414 :
415 : rcu_read_lock();
416 0 : head = page_cache_prev_miss(mapping, index - 1, max);
417 : rcu_read_unlock();
418 :
419 0 : return index - 1 - head;
420 : }
421 :
422 : /*
423 : * page cache context based readahead
424 : */
425 0 : static int try_context_readahead(struct address_space *mapping,
426 : struct file_ra_state *ra,
427 : pgoff_t index,
428 : unsigned long req_size,
429 : unsigned long max)
430 : {
431 : pgoff_t size;
432 :
433 0 : size = count_history_pages(mapping, index, max);
434 :
435 : /*
436 : * not enough history pages:
437 : * it could be a random read
438 : */
439 0 : if (size <= req_size)
440 : return 0;
441 :
442 : /*
443 : * starts from beginning of file:
444 : * it is a strong indication of long-run stream (or whole-file-read)
445 : */
446 0 : if (size >= index)
447 0 : size *= 2;
448 :
449 0 : ra->start = index;
450 0 : ra->size = min(size + req_size, max);
451 0 : ra->async_size = 1;
452 :
453 0 : return 1;
454 : }
455 :
456 : /*
457 : * There are some parts of the kernel which assume that PMD entries
458 : * are exactly HPAGE_PMD_ORDER. Those should be fixed, but until then,
459 : * limit the maximum allocation order to PMD size. I'm not aware of any
460 : * assumptions about maximum order if THP are disabled, but 8 seems like
461 : * a good order (that's 1MB if you're using 4kB pages)
462 : */
463 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
464 : #define MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER
465 : #else
466 : #define MAX_PAGECACHE_ORDER 8
467 : #endif
468 :
469 : static inline int ra_alloc_folio(struct readahead_control *ractl, pgoff_t index,
470 : pgoff_t mark, unsigned int order, gfp_t gfp)
471 : {
472 : int err;
473 : struct folio *folio = filemap_alloc_folio(gfp, order);
474 :
475 : if (!folio)
476 : return -ENOMEM;
477 : mark = round_up(mark, 1UL << order);
478 : if (index == mark)
479 : folio_set_readahead(folio);
480 : err = filemap_add_folio(ractl->mapping, folio, index, gfp);
481 : if (err)
482 : folio_put(folio);
483 : else
484 : ractl->_nr_pages += 1UL << order;
485 : return err;
486 : }
487 :
488 0 : void page_cache_ra_order(struct readahead_control *ractl,
489 : struct file_ra_state *ra, unsigned int new_order)
490 : {
491 0 : struct address_space *mapping = ractl->mapping;
492 0 : pgoff_t index = readahead_index(ractl);
493 0 : pgoff_t limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT;
494 0 : pgoff_t mark = index + ra->size - ra->async_size;
495 0 : int err = 0;
496 0 : gfp_t gfp = readahead_gfp_mask(mapping);
497 :
498 0 : if (!mapping_large_folio_support(mapping) || ra->size < 4)
499 : goto fallback;
500 :
501 : limit = min(limit, index + ra->size - 1);
502 :
503 : if (new_order < MAX_PAGECACHE_ORDER) {
504 : new_order += 2;
505 : if (new_order > MAX_PAGECACHE_ORDER)
506 : new_order = MAX_PAGECACHE_ORDER;
507 : while ((1 << new_order) > ra->size)
508 : new_order--;
509 : }
510 :
511 : while (index <= limit) {
512 : unsigned int order = new_order;
513 :
514 : /* Align with smaller pages if needed */
515 : if (index & ((1UL << order) - 1)) {
516 : order = __ffs(index);
517 : if (order == 1)
518 : order = 0;
519 : }
520 : /* Don't allocate pages past EOF */
521 : while (index + (1UL << order) - 1 > limit) {
522 : if (--order == 1)
523 : order = 0;
524 : }
525 : err = ra_alloc_folio(ractl, index, mark, order, gfp);
526 : if (err)
527 : break;
528 : index += 1UL << order;
529 : }
530 :
531 : if (index > limit) {
532 : ra->size += index - limit - 1;
533 : ra->async_size += index - limit - 1;
534 : }
535 :
536 : read_pages(ractl);
537 :
538 : /*
539 : * If there were already pages in the page cache, then we may have
540 : * left some gaps. Let the regular readahead code take care of this
541 : * situation.
542 : */
543 : if (!err)
544 : return;
545 : fallback:
546 0 : do_page_cache_ra(ractl, ra->size, ra->async_size);
547 : }
548 :
549 : /*
550 : * A minimal readahead algorithm for trivial sequential/random reads.
551 : */
552 0 : static void ondemand_readahead(struct readahead_control *ractl,
553 : struct folio *folio, unsigned long req_size)
554 : {
555 0 : struct backing_dev_info *bdi = inode_to_bdi(ractl->mapping->host);
556 0 : struct file_ra_state *ra = ractl->ra;
557 0 : unsigned long max_pages = ra->ra_pages;
558 : unsigned long add_pages;
559 0 : pgoff_t index = readahead_index(ractl);
560 : pgoff_t expected, prev_index;
561 0 : unsigned int order = folio ? folio_order(folio) : 0;
562 :
563 : /*
564 : * If the request exceeds the readahead window, allow the read to
565 : * be up to the optimal hardware IO size
566 : */
567 0 : if (req_size > max_pages && bdi->io_pages > max_pages)
568 0 : max_pages = min(req_size, bdi->io_pages);
569 :
570 : /*
571 : * start of file
572 : */
573 0 : if (!index)
574 : goto initial_readahead;
575 :
576 : /*
577 : * It's the expected callback index, assume sequential access.
578 : * Ramp up sizes, and push forward the readahead window.
579 : */
580 0 : expected = round_up(ra->start + ra->size - ra->async_size,
581 : 1UL << order);
582 0 : if (index == expected || index == (ra->start + ra->size)) {
583 0 : ra->start += ra->size;
584 0 : ra->size = get_next_ra_size(ra, max_pages);
585 0 : ra->async_size = ra->size;
586 0 : goto readit;
587 : }
588 :
589 : /*
590 : * Hit a marked folio without valid readahead state.
591 : * E.g. interleaved reads.
592 : * Query the pagecache for async_size, which normally equals to
593 : * readahead size. Ramp it up and use it as the new readahead size.
594 : */
595 0 : if (folio) {
596 : pgoff_t start;
597 :
598 : rcu_read_lock();
599 0 : start = page_cache_next_miss(ractl->mapping, index + 1,
600 : max_pages);
601 : rcu_read_unlock();
602 :
603 0 : if (!start || start - index > max_pages)
604 : return;
605 :
606 0 : ra->start = start;
607 0 : ra->size = start - index; /* old async_size */
608 0 : ra->size += req_size;
609 0 : ra->size = get_next_ra_size(ra, max_pages);
610 0 : ra->async_size = ra->size;
611 0 : goto readit;
612 : }
613 :
614 : /*
615 : * oversize read
616 : */
617 0 : if (req_size > max_pages)
618 : goto initial_readahead;
619 :
620 : /*
621 : * sequential cache miss
622 : * trivial case: (index - prev_index) == 1
623 : * unaligned reads: (index - prev_index) == 0
624 : */
625 0 : prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT;
626 0 : if (index - prev_index <= 1UL)
627 : goto initial_readahead;
628 :
629 : /*
630 : * Query the page cache and look for the traces(cached history pages)
631 : * that a sequential stream would leave behind.
632 : */
633 0 : if (try_context_readahead(ractl->mapping, ra, index, req_size,
634 : max_pages))
635 : goto readit;
636 :
637 : /*
638 : * standalone, small random read
639 : * Read as is, and do not pollute the readahead state.
640 : */
641 0 : do_page_cache_ra(ractl, req_size, 0);
642 0 : return;
643 :
644 : initial_readahead:
645 0 : ra->start = index;
646 0 : ra->size = get_init_ra_size(req_size, max_pages);
647 0 : ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
648 :
649 : readit:
650 : /*
651 : * Will this read hit the readahead marker made by itself?
652 : * If so, trigger the readahead marker hit now, and merge
653 : * the resulted next readahead window into the current one.
654 : * Take care of maximum IO pages as above.
655 : */
656 0 : if (index == ra->start && ra->size == ra->async_size) {
657 0 : add_pages = get_next_ra_size(ra, max_pages);
658 0 : if (ra->size + add_pages <= max_pages) {
659 0 : ra->async_size = add_pages;
660 0 : ra->size += add_pages;
661 : } else {
662 0 : ra->size = max_pages;
663 0 : ra->async_size = max_pages >> 1;
664 : }
665 : }
666 :
667 0 : ractl->_index = ra->start;
668 : page_cache_ra_order(ractl, ra, order);
669 : }
670 :
671 0 : void page_cache_sync_ra(struct readahead_control *ractl,
672 : unsigned long req_count)
673 : {
674 0 : bool do_forced_ra = ractl->file && (ractl->file->f_mode & FMODE_RANDOM);
675 :
676 : /*
677 : * Even if readahead is disabled, issue this request as readahead
678 : * as we'll need it to satisfy the requested range. The forced
679 : * readahead will do the right thing and limit the read to just the
680 : * requested range, which we'll set to 1 page for this case.
681 : */
682 0 : if (!ractl->ra->ra_pages || blk_cgroup_congested()) {
683 0 : if (!ractl->file)
684 : return;
685 : req_count = 1;
686 : do_forced_ra = true;
687 : }
688 :
689 : /* be dumb */
690 0 : if (do_forced_ra) {
691 0 : force_page_cache_ra(ractl, req_count);
692 0 : return;
693 : }
694 :
695 0 : ondemand_readahead(ractl, NULL, req_count);
696 : }
697 : EXPORT_SYMBOL_GPL(page_cache_sync_ra);
698 :
699 0 : void page_cache_async_ra(struct readahead_control *ractl,
700 : struct folio *folio, unsigned long req_count)
701 : {
702 : /* no readahead */
703 0 : if (!ractl->ra->ra_pages)
704 : return;
705 :
706 : /*
707 : * Same bit is used for PG_readahead and PG_reclaim.
708 : */
709 0 : if (folio_test_writeback(folio))
710 : return;
711 :
712 0 : folio_clear_readahead(folio);
713 :
714 : if (blk_cgroup_congested())
715 : return;
716 :
717 0 : ondemand_readahead(ractl, folio, req_count);
718 : }
719 : EXPORT_SYMBOL_GPL(page_cache_async_ra);
720 :
721 0 : ssize_t ksys_readahead(int fd, loff_t offset, size_t count)
722 : {
723 : ssize_t ret;
724 : struct fd f;
725 :
726 0 : ret = -EBADF;
727 0 : f = fdget(fd);
728 0 : if (!f.file || !(f.file->f_mode & FMODE_READ))
729 : goto out;
730 :
731 : /*
732 : * The readahead() syscall is intended to run only on files
733 : * that can execute readahead. If readahead is not possible
734 : * on this file, then we must return -EINVAL.
735 : */
736 0 : ret = -EINVAL;
737 0 : if (!f.file->f_mapping || !f.file->f_mapping->a_ops ||
738 0 : !S_ISREG(file_inode(f.file)->i_mode))
739 : goto out;
740 :
741 0 : ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED);
742 : out:
743 0 : fdput(f);
744 0 : return ret;
745 : }
746 :
747 0 : SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
748 : {
749 0 : return ksys_readahead(fd, offset, count);
750 : }
751 :
752 : /**
753 : * readahead_expand - Expand a readahead request
754 : * @ractl: The request to be expanded
755 : * @new_start: The revised start
756 : * @new_len: The revised size of the request
757 : *
758 : * Attempt to expand a readahead request outwards from the current size to the
759 : * specified size by inserting locked pages before and after the current window
760 : * to increase the size to the new window. This may involve the insertion of
761 : * THPs, in which case the window may get expanded even beyond what was
762 : * requested.
763 : *
764 : * The algorithm will stop if it encounters a conflicting page already in the
765 : * pagecache and leave a smaller expansion than requested.
766 : *
767 : * The caller must check for this by examining the revised @ractl object for a
768 : * different expansion than was requested.
769 : */
770 0 : void readahead_expand(struct readahead_control *ractl,
771 : loff_t new_start, size_t new_len)
772 : {
773 0 : struct address_space *mapping = ractl->mapping;
774 0 : struct file_ra_state *ra = ractl->ra;
775 : pgoff_t new_index, new_nr_pages;
776 0 : gfp_t gfp_mask = readahead_gfp_mask(mapping);
777 :
778 0 : new_index = new_start / PAGE_SIZE;
779 :
780 : /* Expand the leading edge downwards */
781 0 : while (ractl->_index > new_index) {
782 0 : unsigned long index = ractl->_index - 1;
783 0 : struct page *page = xa_load(&mapping->i_pages, index);
784 :
785 0 : if (page && !xa_is_value(page))
786 : return; /* Page apparently present */
787 :
788 0 : page = __page_cache_alloc(gfp_mask);
789 0 : if (!page)
790 : return;
791 0 : if (add_to_page_cache_lru(page, mapping, index, gfp_mask) < 0) {
792 0 : put_page(page);
793 0 : return;
794 : }
795 :
796 0 : ractl->_nr_pages++;
797 0 : ractl->_index = page->index;
798 : }
799 :
800 0 : new_len += new_start - readahead_pos(ractl);
801 0 : new_nr_pages = DIV_ROUND_UP(new_len, PAGE_SIZE);
802 :
803 : /* Expand the trailing edge upwards */
804 0 : while (ractl->_nr_pages < new_nr_pages) {
805 0 : unsigned long index = ractl->_index + ractl->_nr_pages;
806 0 : struct page *page = xa_load(&mapping->i_pages, index);
807 :
808 0 : if (page && !xa_is_value(page))
809 : return; /* Page apparently present */
810 :
811 0 : page = __page_cache_alloc(gfp_mask);
812 0 : if (!page)
813 : return;
814 0 : if (add_to_page_cache_lru(page, mapping, index, gfp_mask) < 0) {
815 0 : put_page(page);
816 0 : return;
817 : }
818 0 : ractl->_nr_pages++;
819 0 : if (ra) {
820 0 : ra->size++;
821 0 : ra->async_size++;
822 : }
823 : }
824 : }
825 : EXPORT_SYMBOL(readahead_expand);
|
