blob: c4ca702392333288448cc7ae4ba6b410ddb93fef [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>
Ross Zwisler11bd9692016-08-25 15:17:17 -070011#include <linux/dax.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/gfp.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040013#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/blkdev.h>
15#include <linux/backing-dev.h>
Andrew Morton8bde37f2006-12-10 02:19:40 -080016#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/pagevec.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020018#include <linux/pagemap.h>
Cong Wang782182e2012-05-29 15:06:43 -070019#include <linux/syscalls.h>
20#include <linux/file.h>
Geliang Tangd72ee912016-01-14 15:22:01 -080021#include <linux/mm_inline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
Fabian Frederick29f175d2014-04-07 15:37:55 -070023#include "internal.h"
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025/*
26 * Initialise a struct file's readahead state. Assumes that the caller has
27 * memset *ra to zero.
28 */
29void
30file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
31{
Christoph Hellwigde1414a2015-01-14 10:42:36 +010032 ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages;
Fengguang Wuf4e6b492007-10-16 01:24:33 -070033 ra->prev_pos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034}
Steven Whitehoused41cc702006-01-30 08:53:33 +000035EXPORT_SYMBOL_GPL(file_ra_state_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
David Howells03fb3d22009-04-03 16:42:35 +010037/*
38 * see if a page needs releasing upon read_cache_pages() failure
David Howells266cf652009-04-03 16:42:36 +010039 * - the caller of read_cache_pages() may have set PG_private or PG_fscache
40 * before calling, such as the NFS fs marking pages that are cached locally
41 * on disk, thus we need to give the fs a chance to clean up in the event of
42 * an error
David Howells03fb3d22009-04-03 16:42:35 +010043 */
44static void read_cache_pages_invalidate_page(struct address_space *mapping,
45 struct page *page)
46{
David Howells266cf652009-04-03 16:42:36 +010047 if (page_has_private(page)) {
David Howells03fb3d22009-04-03 16:42:35 +010048 if (!trylock_page(page))
49 BUG();
50 page->mapping = mapping;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030051 do_invalidatepage(page, 0, PAGE_SIZE);
David Howells03fb3d22009-04-03 16:42:35 +010052 page->mapping = NULL;
53 unlock_page(page);
54 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030055 put_page(page);
David Howells03fb3d22009-04-03 16:42:35 +010056}
57
58/*
59 * release a list of pages, invalidating them first if need be
60 */
61static void read_cache_pages_invalidate_pages(struct address_space *mapping,
62 struct list_head *pages)
63{
64 struct page *victim;
65
66 while (!list_empty(pages)) {
Geliang Tangc8ad6302016-01-14 15:20:51 -080067 victim = lru_to_page(pages);
David Howells03fb3d22009-04-03 16:42:35 +010068 list_del(&victim->lru);
69 read_cache_pages_invalidate_page(mapping, victim);
70 }
71}
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073/**
Randy Dunlapbd40cdd2006-06-25 05:48:08 -070074 * read_cache_pages - populate an address space with some pages & start reads against them
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * @mapping: the address_space
76 * @pages: The address of a list_head which contains the target pages. These
77 * pages have their ->index populated and are otherwise uninitialised.
78 * @filler: callback routine for filling a single page.
79 * @data: private data for the callback routine.
80 *
81 * Hides the details of the LRU cache etc from the filesystems.
82 */
83int read_cache_pages(struct address_space *mapping, struct list_head *pages,
84 int (*filler)(void *, struct page *), void *data)
85{
86 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 int ret = 0;
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 while (!list_empty(pages)) {
Geliang Tangc8ad6302016-01-14 15:20:51 -080090 page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 list_del(&page->lru);
Michal Hocko063d99b2015-10-15 15:28:24 -070092 if (add_to_page_cache_lru(page, mapping, page->index,
Michal Hocko8a5c7432016-07-26 15:24:53 -070093 readahead_gfp_mask(mapping))) {
David Howells03fb3d22009-04-03 16:42:35 +010094 read_cache_pages_invalidate_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 continue;
96 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030097 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -070098
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 ret = filler(data, page);
Nick Piggineb2be182007-10-16 01:24:57 -0700100 if (unlikely(ret)) {
David Howells03fb3d22009-04-03 16:42:35 +0100101 read_cache_pages_invalidate_pages(mapping, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 break;
103 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300104 task_io_account_read(PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 return ret;
107}
108
109EXPORT_SYMBOL(read_cache_pages);
110
111static int read_pages(struct address_space *mapping, struct file *filp,
Michal Hocko8a5c7432016-07-26 15:24:53 -0700112 struct list_head *pages, unsigned int nr_pages, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
Jens Axboe5b417b12010-04-19 10:04:38 +0200114 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 unsigned page_idx;
Zach Brown994fc28c2005-12-15 14:28:17 -0800116 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Jens Axboe5b417b12010-04-19 10:04:38 +0200118 blk_start_plug(&plug);
119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 if (mapping->a_ops->readpages) {
121 ret = mapping->a_ops->readpages(filp, mapping, pages, nr_pages);
OGAWA Hirofumi029e3322006-11-02 22:07:06 -0800122 /* Clean up the remaining pages */
123 put_pages_list(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 goto out;
125 }
126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Geliang Tangc8ad6302016-01-14 15:20:51 -0800128 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 list_del(&page->lru);
Michal Hocko8a5c7432016-07-26 15:24:53 -0700130 if (!add_to_page_cache_lru(page, mapping, page->index, gfp))
Zach Brown9f1a3cf2006-06-25 05:46:46 -0700131 mapping->a_ops->readpage(filp, page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300132 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800134 ret = 0;
Jens Axboe5b417b12010-04-19 10:04:38 +0200135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136out:
Jens Axboe5b417b12010-04-19 10:04:38 +0200137 blk_finish_plug(&plug);
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return ret;
140}
141
142/*
Wu Fengguangd30a1102009-06-16 15:31:30 -0700143 * __do_page_cache_readahead() actually reads a chunk of disk. It allocates all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 * the pages first, then submits them all for I/O. This avoids the very bad
145 * behaviour which would occur if page allocations are causing VM writeback.
146 * We really don't want to intermingle reads and writes like that.
147 *
148 * Returns the number of pages requested, or the maximum amount of I/O allowed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
Fabian Frederick29f175d2014-04-07 15:37:55 -0700150int __do_page_cache_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700151 pgoff_t offset, unsigned long nr_to_read,
152 unsigned long lookahead_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153{
154 struct inode *inode = mapping->host;
155 struct page *page;
156 unsigned long end_index; /* The last page we want to read */
157 LIST_HEAD(page_pool);
158 int page_idx;
159 int ret = 0;
160 loff_t isize = i_size_read(inode);
Michal Hocko8a5c7432016-07-26 15:24:53 -0700161 gfp_t gfp_mask = readahead_gfp_mask(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 if (isize == 0)
164 goto out;
165
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300166 end_index = ((isize - 1) >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168 /*
169 * Preallocate as many pages as we will need.
170 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 for (page_idx = 0; page_idx < nr_to_read; page_idx++) {
Andrew Morton7361f4d2005-11-07 00:59:28 -0800172 pgoff_t page_offset = offset + page_idx;
Fengguang Wuc743d962007-07-19 01:48:04 -0700173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 if (page_offset > end_index)
175 break;
176
Nick Piggin00128182007-10-16 01:24:40 -0700177 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 page = radix_tree_lookup(&mapping->page_tree, page_offset);
Nick Piggin00128182007-10-16 01:24:40 -0700179 rcu_read_unlock();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700180 if (page && !radix_tree_exceptional_entry(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 continue;
182
Michal Hocko8a5c7432016-07-26 15:24:53 -0700183 page = __page_cache_alloc(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 if (!page)
185 break;
186 page->index = page_offset;
187 list_add(&page->lru, &page_pool);
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700188 if (page_idx == nr_to_read - lookahead_size)
189 SetPageReadahead(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 ret++;
191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193 /*
194 * Now start the IO. We ignore I/O errors - if the page is not
195 * uptodate then the caller will launch readpage again, and
196 * will then handle the error.
197 */
198 if (ret)
Michal Hocko8a5c7432016-07-26 15:24:53 -0700199 read_pages(mapping, filp, &page_pool, ret, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 BUG_ON(!list_empty(&page_pool));
201out:
202 return ret;
203}
204
205/*
206 * Chunk the readahead into 2 megabyte units, so that we don't pin too much
207 * memory at once.
208 */
209int force_page_cache_readahead(struct address_space *mapping, struct file *filp,
Jens Axboe9491ae42016-12-12 16:43:26 -0800210 pgoff_t offset, unsigned long nr_to_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Jens Axboe9491ae42016-12-12 16:43:26 -0800212 struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
213 struct file_ra_state *ra = &filp->f_ra;
214 unsigned long max_pages;
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages))
217 return -EINVAL;
218
Jens Axboe9491ae42016-12-12 16:43:26 -0800219 /*
220 * If the request exceeds the readahead window, allow the read to
221 * be up to the optimal hardware IO size
222 */
223 max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
224 nr_to_read = min(nr_to_read, max_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 while (nr_to_read) {
226 int err;
227
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300228 unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 if (this_chunk > nr_to_read)
231 this_chunk = nr_to_read;
232 err = __do_page_cache_readahead(mapping, filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700233 offset, this_chunk, 0);
Mark Rutland58d56402014-01-29 14:05:51 -0800234 if (err < 0)
235 return err;
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 offset += this_chunk;
238 nr_to_read -= this_chunk;
239 }
Mark Rutland58d56402014-01-29 14:05:51 -0800240 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Fengguang Wu5ce11102007-07-19 01:47:59 -0700243/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700244 * Set the initial window size, round to next power of 2 and square
245 * for small size, x 4 for medium, and x 2 for large
246 * for 128k (32 page) max ra
247 * 1-8 page = 32k initial, > 8 page = 128k initial
248 */
249static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
250{
251 unsigned long newsize = roundup_pow_of_two(size);
252
253 if (newsize <= max / 32)
254 newsize = newsize * 4;
255 else if (newsize <= max / 4)
256 newsize = newsize * 2;
257 else
258 newsize = max;
259
260 return newsize;
261}
262
263/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700264 * Get the previous window size, ramp it up, and
265 * return it as the new window size.
266 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700267static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700268 unsigned long max)
269{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700270 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700271 unsigned long newsize;
272
273 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700274 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700275 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700276 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700277
278 return min(newsize, max);
279}
280
281/*
282 * On-demand readahead design.
283 *
284 * The fields in struct file_ra_state represent the most-recently-executed
285 * readahead attempt:
286 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700287 * |<----- async_size ---------|
288 * |------------------- size -------------------->|
289 * |==================#===========================|
290 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700291 *
292 * To overlap application thinking time and disk I/O time, we do
293 * `readahead pipelining': Do not wait until the application consumed all
294 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700295 * Instead, submit an asynchronous readahead I/O as soon as there are
296 * only async_size pages left in the readahead window. Normally async_size
297 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700298 *
299 * In interleaved sequential reads, concurrent streams on the same fd can
300 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700301 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700302 * indicator. The flag won't be set on already cached pages, to avoid the
303 * readahead-for-nothing fuss, saving pointless page cache lookups.
304 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700305 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700306 * It should be maintained by the caller, and will be used for detecting
307 * small random reads. Note that the readahead algorithm checks loosely
308 * for sequential patterns. Hence interleaved reads might be served as
309 * sequential ones.
310 *
311 * There is a special-case: if the first page which the application tries to
312 * read happens to be the first page of the file, it is assumed that a linear
313 * read is about to happen and the window is immediately set to the initial size
314 * based on I/O request size and the max_readahead.
315 *
316 * The code ramps up the readahead size aggressively at first, but slow down as
317 * it approaches max_readhead.
318 */
319
320/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700321 * Count contiguously cached pages from @offset-1 to @offset-@max,
322 * this count is a conservative estimation of
323 * - length of the sequential read sequence, or
324 * - thrashing threshold in memory tight systems
325 */
326static pgoff_t count_history_pages(struct address_space *mapping,
Wu Fengguang10be0b32009-06-16 15:31:36 -0700327 pgoff_t offset, unsigned long max)
328{
329 pgoff_t head;
330
331 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700332 head = page_cache_prev_hole(mapping, offset - 1, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700333 rcu_read_unlock();
334
335 return offset - 1 - head;
336}
337
338/*
339 * page cache context based read-ahead
340 */
341static int try_context_readahead(struct address_space *mapping,
342 struct file_ra_state *ra,
343 pgoff_t offset,
344 unsigned long req_size,
345 unsigned long max)
346{
347 pgoff_t size;
348
Fabian Frederick3e2faa082014-08-06 16:04:55 -0700349 size = count_history_pages(mapping, offset, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700350
351 /*
Fengguang Wu2cad4012013-09-11 14:21:47 -0700352 * not enough history pages:
Wu Fengguang10be0b32009-06-16 15:31:36 -0700353 * it could be a random read
354 */
Fengguang Wu2cad4012013-09-11 14:21:47 -0700355 if (size <= req_size)
Wu Fengguang10be0b32009-06-16 15:31:36 -0700356 return 0;
357
358 /*
359 * starts from beginning of file:
360 * it is a strong indication of long-run stream (or whole-file-read)
361 */
362 if (size >= offset)
363 size *= 2;
364
365 ra->start = offset;
Fengguang Wu2cad4012013-09-11 14:21:47 -0700366 ra->size = min(size + req_size, max);
367 ra->async_size = 1;
Wu Fengguang10be0b32009-06-16 15:31:36 -0700368
369 return 1;
370}
371
372/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700373 * A minimal readahead algorithm for trivial sequential/random reads.
374 */
375static unsigned long
376ondemand_readahead(struct address_space *mapping,
377 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700378 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700379 unsigned long req_size)
380{
Jens Axboe9491ae42016-12-12 16:43:26 -0800381 struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
382 unsigned long max_pages = ra->ra_pages;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800383 pgoff_t prev_offset;
Wu Fengguang045a2522009-06-16 15:31:33 -0700384
385 /*
Jens Axboe9491ae42016-12-12 16:43:26 -0800386 * If the request exceeds the readahead window, allow the read to
387 * be up to the optimal hardware IO size
388 */
389 if (req_size > max_pages && bdi->io_pages > max_pages)
390 max_pages = min(req_size, bdi->io_pages);
391
392 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700393 * start of file
394 */
395 if (!offset)
396 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700397
398 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700399 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700400 * Ramp up sizes, and push forward the readahead window.
401 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700402 if ((offset == (ra->start + ra->size - ra->async_size) ||
403 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700404 ra->start += ra->size;
Jens Axboe9491ae42016-12-12 16:43:26 -0800405 ra->size = get_next_ra_size(ra, max_pages);
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700406 ra->async_size = ra->size;
407 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700408 }
409
Fengguang Wu122a21d2007-07-19 01:48:01 -0700410 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700411 * Hit a marked page without valid readahead state.
412 * E.g. interleaved reads.
413 * Query the pagecache for async_size, which normally equals to
414 * readahead size. Ramp it up and use it as the new readahead size.
415 */
416 if (hit_readahead_marker) {
417 pgoff_t start;
418
Nick Piggin30002ed2008-07-25 19:45:28 -0700419 rcu_read_lock();
Jens Axboe9491ae42016-12-12 16:43:26 -0800420 start = page_cache_next_hole(mapping, offset + 1, max_pages);
Nick Piggin30002ed2008-07-25 19:45:28 -0700421 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700422
Jens Axboe9491ae42016-12-12 16:43:26 -0800423 if (!start || start - offset > max_pages)
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700424 return 0;
425
426 ra->start = start;
427 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700428 ra->size += req_size;
Jens Axboe9491ae42016-12-12 16:43:26 -0800429 ra->size = get_next_ra_size(ra, max_pages);
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700430 ra->async_size = ra->size;
431 goto readit;
432 }
433
434 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700435 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700436 */
Jens Axboe9491ae42016-12-12 16:43:26 -0800437 if (req_size > max_pages)
Wu Fengguang045a2522009-06-16 15:31:33 -0700438 goto initial_readahead;
439
440 /*
441 * sequential cache miss
Damien Ramondaaf248a02013-11-12 15:08:16 -0800442 * trivial case: (offset - prev_offset) == 1
443 * unaligned reads: (offset - prev_offset) == 0
Wu Fengguang045a2522009-06-16 15:31:33 -0700444 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300445 prev_offset = (unsigned long long)ra->prev_pos >> PAGE_SHIFT;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800446 if (offset - prev_offset <= 1UL)
Wu Fengguang045a2522009-06-16 15:31:33 -0700447 goto initial_readahead;
448
449 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700450 * Query the page cache and look for the traces(cached history pages)
451 * that a sequential stream would leave behind.
452 */
Jens Axboe9491ae42016-12-12 16:43:26 -0800453 if (try_context_readahead(mapping, ra, offset, req_size, max_pages))
Wu Fengguang10be0b32009-06-16 15:31:36 -0700454 goto readit;
455
456 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700457 * standalone, small random read
458 * Read as is, and do not pollute the readahead state.
459 */
460 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
461
462initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700463 ra->start = offset;
Jens Axboe9491ae42016-12-12 16:43:26 -0800464 ra->size = get_init_ra_size(req_size, max_pages);
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700465 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700466
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700467readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700468 /*
469 * Will this read hit the readahead marker made by itself?
470 * If so, trigger the readahead marker hit now, and merge
471 * the resulted next readahead window into the current one.
472 */
473 if (offset == ra->start && ra->size == ra->async_size) {
Jens Axboe9491ae42016-12-12 16:43:26 -0800474 ra->async_size = get_next_ra_size(ra, max_pages);
Wu Fengguang51daa882009-06-16 15:31:24 -0700475 ra->size += ra->async_size;
476 }
477
Fengguang Wu122a21d2007-07-19 01:48:01 -0700478 return ra_submit(ra, mapping, filp);
479}
480
481/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700482 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700483 * @mapping: address_space which holds the pagecache and I/O vectors
484 * @ra: file_ra_state which holds the readahead state
485 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700486 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700487 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700488 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700489 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700490 * page_cache_sync_readahead() should be called when a cache miss happened:
491 * it will submit the read. The readahead logic may decide to piggyback more
492 * pages onto the read request if access patterns suggest it will improve
493 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700494 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700495void page_cache_sync_readahead(struct address_space *mapping,
496 struct file_ra_state *ra, struct file *filp,
497 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700498{
499 /* no read-ahead */
500 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700501 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700502
Wu Fengguang01414502010-03-05 13:42:03 -0800503 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700504 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800505 force_page_cache_readahead(mapping, filp, offset, req_size);
506 return;
507 }
508
Fengguang Wu122a21d2007-07-19 01:48:01 -0700509 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700510 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700511}
Rusty Russellcf914a72007-07-19 01:48:08 -0700512EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
513
514/**
515 * page_cache_async_readahead - file readahead for marked pages
516 * @mapping: address_space which holds the pagecache and I/O vectors
517 * @ra: file_ra_state which holds the readahead state
518 * @filp: passed on to ->readpage() and ->readpages()
519 * @page: the page at @offset which has the PG_readahead flag set
520 * @offset: start offset into @mapping, in pagecache page-sized units
521 * @req_size: hint: total size of the read which the caller is performing in
522 * pagecache pages
523 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700524 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700525 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700526 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700527 * more pages.
528 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700529void
530page_cache_async_readahead(struct address_space *mapping,
531 struct file_ra_state *ra, struct file *filp,
532 struct page *page, pgoff_t offset,
533 unsigned long req_size)
534{
535 /* no read-ahead */
536 if (!ra->ra_pages)
537 return;
538
539 /*
540 * Same bit is used for PG_readahead and PG_reclaim.
541 */
542 if (PageWriteback(page))
543 return;
544
545 ClearPageReadahead(page);
546
547 /*
548 * Defer asynchronous read-ahead on IO congestion.
549 */
Tejun Heo703c2702015-05-22 17:13:44 -0400550 if (inode_read_congested(mapping->host))
Rusty Russellcf914a72007-07-19 01:48:08 -0700551 return;
552
553 /* do read-ahead */
554 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
555}
556EXPORT_SYMBOL_GPL(page_cache_async_readahead);
Cong Wang782182e2012-05-29 15:06:43 -0700557
558static ssize_t
559do_readahead(struct address_space *mapping, struct file *filp,
560 pgoff_t index, unsigned long nr)
561{
Andrew Morton63d0f0a32013-11-12 15:07:09 -0800562 if (!mapping || !mapping->a_ops)
Cong Wang782182e2012-05-29 15:06:43 -0700563 return -EINVAL;
564
Ross Zwisler11bd9692016-08-25 15:17:17 -0700565 /*
566 * Readahead doesn't make sense for DAX inodes, but we don't want it
567 * to report a failure either. Instead, we just return success and
568 * don't do any work.
569 */
570 if (dax_mapping(mapping))
571 return 0;
572
Mark Rutland58d56402014-01-29 14:05:51 -0800573 return force_page_cache_readahead(mapping, filp, index, nr);
Cong Wang782182e2012-05-29 15:06:43 -0700574}
575
Al Viro4a0fd5b2013-01-21 15:16:58 -0500576SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
Cong Wang782182e2012-05-29 15:06:43 -0700577{
578 ssize_t ret;
Al Viro2903ff02012-08-28 12:52:22 -0400579 struct fd f;
Cong Wang782182e2012-05-29 15:06:43 -0700580
581 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400582 f = fdget(fd);
583 if (f.file) {
584 if (f.file->f_mode & FMODE_READ) {
585 struct address_space *mapping = f.file->f_mapping;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300586 pgoff_t start = offset >> PAGE_SHIFT;
587 pgoff_t end = (offset + count - 1) >> PAGE_SHIFT;
Cong Wang782182e2012-05-29 15:06:43 -0700588 unsigned long len = end - start + 1;
Al Viro2903ff02012-08-28 12:52:22 -0400589 ret = do_readahead(mapping, f.file, start, len);
Cong Wang782182e2012-05-29 15:06:43 -0700590 }
Al Viro2903ff02012-08-28 12:52:22 -0400591 fdput(f);
Cong Wang782182e2012-05-29 15:06:43 -0700592 }
593 return ret;
594}