blob: 65ec288dc057e874a661fd372501454c32cc2bd4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/readahead.c - address_space-level file readahead.
3 *
4 * Copyright (C) 2002, Linus Torvalds
5 *
Francois Camie1f8e872008-10-15 22:01:59 -07006 * 09Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * Initial version.
8 */
9
10#include <linux/kernel.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090011#include <linux/gfp.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/blkdev.h>
14#include <linux/backing-dev.h>
Andrew Morton8bde37f2006-12-10 02:19:40 -080015#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/pagevec.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020017#include <linux/pagemap.h>
Cong Wang782182e2012-05-29 15:06:43 -070018#include <linux/syscalls.h>
19#include <linux/file.h>
Geliang Tangd72ee912016-01-14 15:22:01 -080020#include <linux/mm_inline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Fabian Frederick29f175d2014-04-07 15:37:55 -070022#include "internal.h"
23
Linus Torvalds1da177e2005-04-16 15:20:36 -070024/*
25 * Initialise a struct file's readahead state. Assumes that the caller has
26 * memset *ra to zero.
27 */
28void
29file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
30{
Christoph Hellwigde1414a2015-01-14 10:42:36 +010031 ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages;
Fengguang Wuf4e6b492007-10-16 01:24:33 -070032 ra->prev_pos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070033}
Steven Whitehoused41cc702006-01-30 08:53:33 +000034EXPORT_SYMBOL_GPL(file_ra_state_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
David Howells03fb3d22009-04-03 16:42:35 +010036/*
37 * see if a page needs releasing upon read_cache_pages() failure
David Howells266cf652009-04-03 16:42:36 +010038 * - the caller of read_cache_pages() may have set PG_private or PG_fscache
39 * before calling, such as the NFS fs marking pages that are cached locally
40 * on disk, thus we need to give the fs a chance to clean up in the event of
41 * an error
David Howells03fb3d22009-04-03 16:42:35 +010042 */
43static void read_cache_pages_invalidate_page(struct address_space *mapping,
44 struct page *page)
45{
David Howells266cf652009-04-03 16:42:36 +010046 if (page_has_private(page)) {
David Howells03fb3d22009-04-03 16:42:35 +010047 if (!trylock_page(page))
48 BUG();
49 page->mapping = mapping;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030050 do_invalidatepage(page, 0, PAGE_SIZE);
David Howells03fb3d22009-04-03 16:42:35 +010051 page->mapping = NULL;
52 unlock_page(page);
53 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030054 put_page(page);
David Howells03fb3d22009-04-03 16:42:35 +010055}
56
57/*
58 * release a list of pages, invalidating them first if need be
59 */
60static void read_cache_pages_invalidate_pages(struct address_space *mapping,
61 struct list_head *pages)
62{
63 struct page *victim;
64
65 while (!list_empty(pages)) {
Geliang Tangc8ad6302016-01-14 15:20:51 -080066 victim = lru_to_page(pages);
David Howells03fb3d22009-04-03 16:42:35 +010067 list_del(&victim->lru);
68 read_cache_pages_invalidate_page(mapping, victim);
69 }
70}
71
Linus Torvalds1da177e2005-04-16 15:20:36 -070072/**
Randy Dunlapbd40cdd2006-06-25 05:48:08 -070073 * read_cache_pages - populate an address space with some pages & start reads against them
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * @mapping: the address_space
75 * @pages: The address of a list_head which contains the target pages. These
76 * pages have their ->index populated and are otherwise uninitialised.
77 * @filler: callback routine for filling a single page.
78 * @data: private data for the callback routine.
79 *
80 * Hides the details of the LRU cache etc from the filesystems.
81 */
82int read_cache_pages(struct address_space *mapping, struct list_head *pages,
83 int (*filler)(void *, struct page *), void *data)
84{
85 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 int ret = 0;
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 while (!list_empty(pages)) {
Geliang Tangc8ad6302016-01-14 15:20:51 -080089 page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 list_del(&page->lru);
Michal Hocko063d99b2015-10-15 15:28:24 -070091 if (add_to_page_cache_lru(page, mapping, page->index,
Michal Hocko8a5c7432016-07-26 15:24:53 -070092 readahead_gfp_mask(mapping))) {
David Howells03fb3d22009-04-03 16:42:35 +010093 read_cache_pages_invalidate_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 continue;
95 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030096 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -070097
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 ret = filler(data, page);
Nick Piggineb2be182007-10-16 01:24:57 -070099 if (unlikely(ret)) {
David Howells03fb3d22009-04-03 16:42:35 +0100100 read_cache_pages_invalidate_pages(mapping, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 break;
102 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300103 task_io_account_read(PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 return ret;
106}
107
108EXPORT_SYMBOL(read_cache_pages);
109
110static int read_pages(struct address_space *mapping, struct file *filp,
Michal Hocko8a5c7432016-07-26 15:24:53 -0700111 struct list_head *pages, unsigned int nr_pages, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
Jens Axboe5b417b12010-04-19 10:04:38 +0200113 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 unsigned page_idx;
Zach Brown994fc28c2005-12-15 14:28:17 -0800115 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Jens Axboe5b417b12010-04-19 10:04:38 +0200117 blk_start_plug(&plug);
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 if (mapping->a_ops->readpages) {
120 ret = mapping->a_ops->readpages(filp, mapping, pages, nr_pages);
OGAWA Hirofumi029e3322006-11-02 22:07:06 -0800121 /* Clean up the remaining pages */
122 put_pages_list(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 goto out;
124 }
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Geliang Tangc8ad6302016-01-14 15:20:51 -0800127 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 list_del(&page->lru);
Michal Hocko8a5c7432016-07-26 15:24:53 -0700129 if (!add_to_page_cache_lru(page, mapping, page->index, gfp))
Zach Brown9f1a3cf2006-06-25 05:46:46 -0700130 mapping->a_ops->readpage(filp, page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300131 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800133 ret = 0;
Jens Axboe5b417b12010-04-19 10:04:38 +0200134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135out:
Jens Axboe5b417b12010-04-19 10:04:38 +0200136 blk_finish_plug(&plug);
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 return ret;
139}
140
141/*
Wu Fengguangd30a1102009-06-16 15:31:30 -0700142 * __do_page_cache_readahead() actually reads a chunk of disk. It allocates all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 * the pages first, then submits them all for I/O. This avoids the very bad
144 * behaviour which would occur if page allocations are causing VM writeback.
145 * We really don't want to intermingle reads and writes like that.
146 *
147 * Returns the number of pages requested, or the maximum amount of I/O allowed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 */
Fabian Frederick29f175d2014-04-07 15:37:55 -0700149int __do_page_cache_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700150 pgoff_t offset, unsigned long nr_to_read,
151 unsigned long lookahead_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
153 struct inode *inode = mapping->host;
154 struct page *page;
155 unsigned long end_index; /* The last page we want to read */
156 LIST_HEAD(page_pool);
157 int page_idx;
158 int ret = 0;
159 loff_t isize = i_size_read(inode);
Michal Hocko8a5c7432016-07-26 15:24:53 -0700160 gfp_t gfp_mask = readahead_gfp_mask(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 if (isize == 0)
163 goto out;
164
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300165 end_index = ((isize - 1) >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 /*
168 * Preallocate as many pages as we will need.
169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 for (page_idx = 0; page_idx < nr_to_read; page_idx++) {
Andrew Morton7361f4d2005-11-07 00:59:28 -0800171 pgoff_t page_offset = offset + page_idx;
Fengguang Wuc743d962007-07-19 01:48:04 -0700172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 if (page_offset > end_index)
174 break;
175
Nick Piggin00128182007-10-16 01:24:40 -0700176 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 page = radix_tree_lookup(&mapping->page_tree, page_offset);
Nick Piggin00128182007-10-16 01:24:40 -0700178 rcu_read_unlock();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700179 if (page && !radix_tree_exceptional_entry(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 continue;
181
Michal Hocko8a5c7432016-07-26 15:24:53 -0700182 page = __page_cache_alloc(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (!page)
184 break;
185 page->index = page_offset;
186 list_add(&page->lru, &page_pool);
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700187 if (page_idx == nr_to_read - lookahead_size)
188 SetPageReadahead(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 ret++;
190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
192 /*
193 * Now start the IO. We ignore I/O errors - if the page is not
194 * uptodate then the caller will launch readpage again, and
195 * will then handle the error.
196 */
197 if (ret)
Michal Hocko8a5c7432016-07-26 15:24:53 -0700198 read_pages(mapping, filp, &page_pool, ret, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 BUG_ON(!list_empty(&page_pool));
200out:
201 return ret;
202}
203
204/*
205 * Chunk the readahead into 2 megabyte units, so that we don't pin too much
206 * memory at once.
207 */
208int force_page_cache_readahead(struct address_space *mapping, struct file *filp,
Andrew Morton7361f4d2005-11-07 00:59:28 -0800209 pgoff_t offset, unsigned long nr_to_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages))
212 return -EINVAL;
213
Roman Gushchin600e19a2015-11-05 18:47:08 -0800214 nr_to_read = min(nr_to_read, inode_to_bdi(mapping->host)->ra_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 while (nr_to_read) {
216 int err;
217
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300218 unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220 if (this_chunk > nr_to_read)
221 this_chunk = nr_to_read;
222 err = __do_page_cache_readahead(mapping, filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700223 offset, this_chunk, 0);
Mark Rutland58d56402014-01-29 14:05:51 -0800224 if (err < 0)
225 return err;
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 offset += this_chunk;
228 nr_to_read -= this_chunk;
229 }
Mark Rutland58d56402014-01-29 14:05:51 -0800230 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
Fengguang Wu5ce11102007-07-19 01:47:59 -0700233/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700234 * Set the initial window size, round to next power of 2 and square
235 * for small size, x 4 for medium, and x 2 for large
236 * for 128k (32 page) max ra
237 * 1-8 page = 32k initial, > 8 page = 128k initial
238 */
239static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
240{
241 unsigned long newsize = roundup_pow_of_two(size);
242
243 if (newsize <= max / 32)
244 newsize = newsize * 4;
245 else if (newsize <= max / 4)
246 newsize = newsize * 2;
247 else
248 newsize = max;
249
250 return newsize;
251}
252
253/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700254 * Get the previous window size, ramp it up, and
255 * return it as the new window size.
256 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700257static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700258 unsigned long max)
259{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700260 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700261 unsigned long newsize;
262
263 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700264 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700265 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700266 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700267
268 return min(newsize, max);
269}
270
271/*
272 * On-demand readahead design.
273 *
274 * The fields in struct file_ra_state represent the most-recently-executed
275 * readahead attempt:
276 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700277 * |<----- async_size ---------|
278 * |------------------- size -------------------->|
279 * |==================#===========================|
280 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700281 *
282 * To overlap application thinking time and disk I/O time, we do
283 * `readahead pipelining': Do not wait until the application consumed all
284 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700285 * Instead, submit an asynchronous readahead I/O as soon as there are
286 * only async_size pages left in the readahead window. Normally async_size
287 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700288 *
289 * In interleaved sequential reads, concurrent streams on the same fd can
290 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700291 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700292 * indicator. The flag won't be set on already cached pages, to avoid the
293 * readahead-for-nothing fuss, saving pointless page cache lookups.
294 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700295 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700296 * It should be maintained by the caller, and will be used for detecting
297 * small random reads. Note that the readahead algorithm checks loosely
298 * for sequential patterns. Hence interleaved reads might be served as
299 * sequential ones.
300 *
301 * There is a special-case: if the first page which the application tries to
302 * read happens to be the first page of the file, it is assumed that a linear
303 * read is about to happen and the window is immediately set to the initial size
304 * based on I/O request size and the max_readahead.
305 *
306 * The code ramps up the readahead size aggressively at first, but slow down as
307 * it approaches max_readhead.
308 */
309
310/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700311 * Count contiguously cached pages from @offset-1 to @offset-@max,
312 * this count is a conservative estimation of
313 * - length of the sequential read sequence, or
314 * - thrashing threshold in memory tight systems
315 */
316static pgoff_t count_history_pages(struct address_space *mapping,
Wu Fengguang10be0b32009-06-16 15:31:36 -0700317 pgoff_t offset, unsigned long max)
318{
319 pgoff_t head;
320
321 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700322 head = page_cache_prev_hole(mapping, offset - 1, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700323 rcu_read_unlock();
324
325 return offset - 1 - head;
326}
327
328/*
329 * page cache context based read-ahead
330 */
331static int try_context_readahead(struct address_space *mapping,
332 struct file_ra_state *ra,
333 pgoff_t offset,
334 unsigned long req_size,
335 unsigned long max)
336{
337 pgoff_t size;
338
Fabian Frederick3e2faa082014-08-06 16:04:55 -0700339 size = count_history_pages(mapping, offset, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700340
341 /*
Fengguang Wu2cad4012013-09-11 14:21:47 -0700342 * not enough history pages:
Wu Fengguang10be0b32009-06-16 15:31:36 -0700343 * it could be a random read
344 */
Fengguang Wu2cad4012013-09-11 14:21:47 -0700345 if (size <= req_size)
Wu Fengguang10be0b32009-06-16 15:31:36 -0700346 return 0;
347
348 /*
349 * starts from beginning of file:
350 * it is a strong indication of long-run stream (or whole-file-read)
351 */
352 if (size >= offset)
353 size *= 2;
354
355 ra->start = offset;
Fengguang Wu2cad4012013-09-11 14:21:47 -0700356 ra->size = min(size + req_size, max);
357 ra->async_size = 1;
Wu Fengguang10be0b32009-06-16 15:31:36 -0700358
359 return 1;
360}
361
362/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700363 * A minimal readahead algorithm for trivial sequential/random reads.
364 */
365static unsigned long
366ondemand_readahead(struct address_space *mapping,
367 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700368 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700369 unsigned long req_size)
370{
Roman Gushchin600e19a2015-11-05 18:47:08 -0800371 unsigned long max = ra->ra_pages;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800372 pgoff_t prev_offset;
Wu Fengguang045a2522009-06-16 15:31:33 -0700373
374 /*
375 * start of file
376 */
377 if (!offset)
378 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700379
380 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700381 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700382 * Ramp up sizes, and push forward the readahead window.
383 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700384 if ((offset == (ra->start + ra->size - ra->async_size) ||
385 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700386 ra->start += ra->size;
387 ra->size = get_next_ra_size(ra, max);
388 ra->async_size = ra->size;
389 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700390 }
391
Fengguang Wu122a21d2007-07-19 01:48:01 -0700392 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700393 * Hit a marked page without valid readahead state.
394 * E.g. interleaved reads.
395 * Query the pagecache for async_size, which normally equals to
396 * readahead size. Ramp it up and use it as the new readahead size.
397 */
398 if (hit_readahead_marker) {
399 pgoff_t start;
400
Nick Piggin30002ed2008-07-25 19:45:28 -0700401 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700402 start = page_cache_next_hole(mapping, offset + 1, max);
Nick Piggin30002ed2008-07-25 19:45:28 -0700403 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700404
405 if (!start || start - offset > max)
406 return 0;
407
408 ra->start = start;
409 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700410 ra->size += req_size;
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700411 ra->size = get_next_ra_size(ra, max);
412 ra->async_size = ra->size;
413 goto readit;
414 }
415
416 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700417 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700418 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700419 if (req_size > max)
420 goto initial_readahead;
421
422 /*
423 * sequential cache miss
Damien Ramondaaf248a02013-11-12 15:08:16 -0800424 * trivial case: (offset - prev_offset) == 1
425 * unaligned reads: (offset - prev_offset) == 0
Wu Fengguang045a2522009-06-16 15:31:33 -0700426 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300427 prev_offset = (unsigned long long)ra->prev_pos >> PAGE_SHIFT;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800428 if (offset - prev_offset <= 1UL)
Wu Fengguang045a2522009-06-16 15:31:33 -0700429 goto initial_readahead;
430
431 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700432 * Query the page cache and look for the traces(cached history pages)
433 * that a sequential stream would leave behind.
434 */
435 if (try_context_readahead(mapping, ra, offset, req_size, max))
436 goto readit;
437
438 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700439 * standalone, small random read
440 * Read as is, and do not pollute the readahead state.
441 */
442 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
443
444initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700445 ra->start = offset;
446 ra->size = get_init_ra_size(req_size, max);
447 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700448
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700449readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700450 /*
451 * Will this read hit the readahead marker made by itself?
452 * If so, trigger the readahead marker hit now, and merge
453 * the resulted next readahead window into the current one.
454 */
455 if (offset == ra->start && ra->size == ra->async_size) {
456 ra->async_size = get_next_ra_size(ra, max);
457 ra->size += ra->async_size;
458 }
459
Fengguang Wu122a21d2007-07-19 01:48:01 -0700460 return ra_submit(ra, mapping, filp);
461}
462
463/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700464 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700465 * @mapping: address_space which holds the pagecache and I/O vectors
466 * @ra: file_ra_state which holds the readahead state
467 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700468 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700469 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700470 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700471 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700472 * page_cache_sync_readahead() should be called when a cache miss happened:
473 * it will submit the read. The readahead logic may decide to piggyback more
474 * pages onto the read request if access patterns suggest it will improve
475 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700476 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700477void page_cache_sync_readahead(struct address_space *mapping,
478 struct file_ra_state *ra, struct file *filp,
479 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700480{
481 /* no read-ahead */
482 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700483 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700484
Wu Fengguang01414502010-03-05 13:42:03 -0800485 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700486 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800487 force_page_cache_readahead(mapping, filp, offset, req_size);
488 return;
489 }
490
Fengguang Wu122a21d2007-07-19 01:48:01 -0700491 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700492 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700493}
Rusty Russellcf914a72007-07-19 01:48:08 -0700494EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
495
496/**
497 * page_cache_async_readahead - file readahead for marked pages
498 * @mapping: address_space which holds the pagecache and I/O vectors
499 * @ra: file_ra_state which holds the readahead state
500 * @filp: passed on to ->readpage() and ->readpages()
501 * @page: the page at @offset which has the PG_readahead flag set
502 * @offset: start offset into @mapping, in pagecache page-sized units
503 * @req_size: hint: total size of the read which the caller is performing in
504 * pagecache pages
505 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700506 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700507 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700508 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700509 * more pages.
510 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700511void
512page_cache_async_readahead(struct address_space *mapping,
513 struct file_ra_state *ra, struct file *filp,
514 struct page *page, pgoff_t offset,
515 unsigned long req_size)
516{
517 /* no read-ahead */
518 if (!ra->ra_pages)
519 return;
520
521 /*
522 * Same bit is used for PG_readahead and PG_reclaim.
523 */
524 if (PageWriteback(page))
525 return;
526
527 ClearPageReadahead(page);
528
529 /*
530 * Defer asynchronous read-ahead on IO congestion.
531 */
Tejun Heo703c2702015-05-22 17:13:44 -0400532 if (inode_read_congested(mapping->host))
Rusty Russellcf914a72007-07-19 01:48:08 -0700533 return;
534
535 /* do read-ahead */
536 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
537}
538EXPORT_SYMBOL_GPL(page_cache_async_readahead);
Cong Wang782182e2012-05-29 15:06:43 -0700539
540static ssize_t
541do_readahead(struct address_space *mapping, struct file *filp,
542 pgoff_t index, unsigned long nr)
543{
Andrew Morton63d0f0a32013-11-12 15:07:09 -0800544 if (!mapping || !mapping->a_ops)
Cong Wang782182e2012-05-29 15:06:43 -0700545 return -EINVAL;
546
Mark Rutland58d56402014-01-29 14:05:51 -0800547 return force_page_cache_readahead(mapping, filp, index, nr);
Cong Wang782182e2012-05-29 15:06:43 -0700548}
549
Al Viro4a0fd5b2013-01-21 15:16:58 -0500550SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
Cong Wang782182e2012-05-29 15:06:43 -0700551{
552 ssize_t ret;
Al Viro2903ff02012-08-28 12:52:22 -0400553 struct fd f;
Cong Wang782182e2012-05-29 15:06:43 -0700554
555 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400556 f = fdget(fd);
557 if (f.file) {
558 if (f.file->f_mode & FMODE_READ) {
559 struct address_space *mapping = f.file->f_mapping;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300560 pgoff_t start = offset >> PAGE_SHIFT;
561 pgoff_t end = (offset + count - 1) >> PAGE_SHIFT;
Cong Wang782182e2012-05-29 15:06:43 -0700562 unsigned long len = end - start + 1;
Al Viro2903ff02012-08-28 12:52:22 -0400563 ret = do_readahead(mapping, f.file, start, len);
Cong Wang782182e2012-05-29 15:06:43 -0700564 }
Al Viro2903ff02012-08-28 12:52:22 -0400565 fdput(f);
Cong Wang782182e2012-05-29 15:06:43 -0700566 }
567 return ret;
568}