blob: 60cd846a9a4401f73a6a51539507ecc22fb308d4 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Fabian Frederick29f175d2014-04-07 15:37:55 -070021#include "internal.h"
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023/*
24 * Initialise a struct file's readahead state. Assumes that the caller has
25 * memset *ra to zero.
26 */
27void
28file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
29{
Christoph Hellwigde1414a2015-01-14 10:42:36 +010030 ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages;
Fengguang Wuf4e6b492007-10-16 01:24:33 -070031 ra->prev_pos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070032}
Steven Whitehoused41cc702006-01-30 08:53:33 +000033EXPORT_SYMBOL_GPL(file_ra_state_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#define list_to_page(head) (list_entry((head)->prev, struct page, lru))
36
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;
Lukas Czernerd47992f2013-05-21 23:17:23 -040051 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
David Howells03fb3d22009-04-03 16:42:35 +010052 page->mapping = NULL;
53 unlock_page(page);
54 }
55 page_cache_release(page);
56}
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)) {
67 victim = list_to_page(pages);
68 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)) {
90 page = list_to_page(pages);
91 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -070092 if (add_to_page_cache_lru(page, mapping,
93 page->index, GFP_KERNEL)) {
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 }
Nick Piggineb2be182007-10-16 01:24:57 -070097 page_cache_release(page);
98
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 }
Andrew Morton8bde37f2006-12-10 02:19:40 -0800104 task_io_account_read(PAGE_CACHE_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,
112 struct list_head *pages, unsigned nr_pages)
113{
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++) {
128 struct page *page = list_to_page(pages);
129 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700130 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 page->index, GFP_KERNEL)) {
Zach Brown9f1a3cf2006-06-25 05:46:46 -0700132 mapping->a_ops->readpage(filp, page);
Nick Piggineb2be182007-10-16 01:24:57 -0700133 }
134 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800136 ret = 0;
Jens Axboe5b417b12010-04-19 10:04:38 +0200137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138out:
Jens Axboe5b417b12010-04-19 10:04:38 +0200139 blk_finish_plug(&plug);
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 return ret;
142}
143
144/*
Wu Fengguangd30a1102009-06-16 15:31:30 -0700145 * __do_page_cache_readahead() actually reads a chunk of disk. It allocates all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 * the pages first, then submits them all for I/O. This avoids the very bad
147 * behaviour which would occur if page allocations are causing VM writeback.
148 * We really don't want to intermingle reads and writes like that.
149 *
150 * Returns the number of pages requested, or the maximum amount of I/O allowed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 */
Fabian Frederick29f175d2014-04-07 15:37:55 -0700152int __do_page_cache_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700153 pgoff_t offset, unsigned long nr_to_read,
154 unsigned long lookahead_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
156 struct inode *inode = mapping->host;
157 struct page *page;
158 unsigned long end_index; /* The last page we want to read */
159 LIST_HEAD(page_pool);
160 int page_idx;
161 int ret = 0;
162 loff_t isize = i_size_read(inode);
163
164 if (isize == 0)
165 goto out;
166
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700167 end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
169 /*
170 * Preallocate as many pages as we will need.
171 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 for (page_idx = 0; page_idx < nr_to_read; page_idx++) {
Andrew Morton7361f4d2005-11-07 00:59:28 -0800173 pgoff_t page_offset = offset + page_idx;
Fengguang Wuc743d962007-07-19 01:48:04 -0700174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 if (page_offset > end_index)
176 break;
177
Nick Piggin00128182007-10-16 01:24:40 -0700178 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 page = radix_tree_lookup(&mapping->page_tree, page_offset);
Nick Piggin00128182007-10-16 01:24:40 -0700180 rcu_read_unlock();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700181 if (page && !radix_tree_exceptional_entry(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 continue;
183
Wu Fengguang7b1de582011-05-24 17:12:25 -0700184 page = page_cache_alloc_readahead(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 if (!page)
186 break;
187 page->index = page_offset;
188 list_add(&page->lru, &page_pool);
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700189 if (page_idx == nr_to_read - lookahead_size)
190 SetPageReadahead(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 ret++;
192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 /*
195 * Now start the IO. We ignore I/O errors - if the page is not
196 * uptodate then the caller will launch readpage again, and
197 * will then handle the error.
198 */
199 if (ret)
200 read_pages(mapping, filp, &page_pool, ret);
201 BUG_ON(!list_empty(&page_pool));
202out:
203 return ret;
204}
205
206/*
207 * Chunk the readahead into 2 megabyte units, so that we don't pin too much
208 * memory at once.
209 */
210int force_page_cache_readahead(struct address_space *mapping, struct file *filp,
Andrew Morton7361f4d2005-11-07 00:59:28 -0800211 pgoff_t offset, unsigned long nr_to_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages))
214 return -EINVAL;
215
Wu Fengguangf7e839d2009-06-16 15:31:20 -0700216 nr_to_read = max_sane_readahead(nr_to_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 while (nr_to_read) {
218 int err;
219
220 unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_CACHE_SIZE;
221
222 if (this_chunk > nr_to_read)
223 this_chunk = nr_to_read;
224 err = __do_page_cache_readahead(mapping, filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700225 offset, this_chunk, 0);
Mark Rutland58d56402014-01-29 14:05:51 -0800226 if (err < 0)
227 return err;
228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 offset += this_chunk;
230 nr_to_read -= this_chunk;
231 }
Mark Rutland58d56402014-01-29 14:05:51 -0800232 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
Raghavendra K T6d2be912014-04-03 14:48:23 -0700235#define MAX_READAHEAD ((512*4096)/PAGE_CACHE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 * Given a desired number of PAGE_CACHE_SIZE readahead pages, return a
238 * sensible upper limit.
239 */
240unsigned long max_sane_readahead(unsigned long nr)
241{
Raghavendra K T6d2be912014-04-03 14:48:23 -0700242 return min(nr, MAX_READAHEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
Fengguang Wu5ce11102007-07-19 01:47:59 -0700244
245/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700246 * Set the initial window size, round to next power of 2 and square
247 * for small size, x 4 for medium, and x 2 for large
248 * for 128k (32 page) max ra
249 * 1-8 page = 32k initial, > 8 page = 128k initial
250 */
251static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
252{
253 unsigned long newsize = roundup_pow_of_two(size);
254
255 if (newsize <= max / 32)
256 newsize = newsize * 4;
257 else if (newsize <= max / 4)
258 newsize = newsize * 2;
259 else
260 newsize = max;
261
262 return newsize;
263}
264
265/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700266 * Get the previous window size, ramp it up, and
267 * return it as the new window size.
268 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700269static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700270 unsigned long max)
271{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700272 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700273 unsigned long newsize;
274
275 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700276 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700277 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700278 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700279
280 return min(newsize, max);
281}
282
283/*
284 * On-demand readahead design.
285 *
286 * The fields in struct file_ra_state represent the most-recently-executed
287 * readahead attempt:
288 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700289 * |<----- async_size ---------|
290 * |------------------- size -------------------->|
291 * |==================#===========================|
292 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700293 *
294 * To overlap application thinking time and disk I/O time, we do
295 * `readahead pipelining': Do not wait until the application consumed all
296 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700297 * Instead, submit an asynchronous readahead I/O as soon as there are
298 * only async_size pages left in the readahead window. Normally async_size
299 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700300 *
301 * In interleaved sequential reads, concurrent streams on the same fd can
302 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700303 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700304 * indicator. The flag won't be set on already cached pages, to avoid the
305 * readahead-for-nothing fuss, saving pointless page cache lookups.
306 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700307 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700308 * It should be maintained by the caller, and will be used for detecting
309 * small random reads. Note that the readahead algorithm checks loosely
310 * for sequential patterns. Hence interleaved reads might be served as
311 * sequential ones.
312 *
313 * There is a special-case: if the first page which the application tries to
314 * read happens to be the first page of the file, it is assumed that a linear
315 * read is about to happen and the window is immediately set to the initial size
316 * based on I/O request size and the max_readahead.
317 *
318 * The code ramps up the readahead size aggressively at first, but slow down as
319 * it approaches max_readhead.
320 */
321
322/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700323 * Count contiguously cached pages from @offset-1 to @offset-@max,
324 * this count is a conservative estimation of
325 * - length of the sequential read sequence, or
326 * - thrashing threshold in memory tight systems
327 */
328static pgoff_t count_history_pages(struct address_space *mapping,
Wu Fengguang10be0b32009-06-16 15:31:36 -0700329 pgoff_t offset, unsigned long max)
330{
331 pgoff_t head;
332
333 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700334 head = page_cache_prev_hole(mapping, offset - 1, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700335 rcu_read_unlock();
336
337 return offset - 1 - head;
338}
339
340/*
341 * page cache context based read-ahead
342 */
343static int try_context_readahead(struct address_space *mapping,
344 struct file_ra_state *ra,
345 pgoff_t offset,
346 unsigned long req_size,
347 unsigned long max)
348{
349 pgoff_t size;
350
Fabian Frederick3e2faa082014-08-06 16:04:55 -0700351 size = count_history_pages(mapping, offset, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700352
353 /*
Fengguang Wu2cad4012013-09-11 14:21:47 -0700354 * not enough history pages:
Wu Fengguang10be0b32009-06-16 15:31:36 -0700355 * it could be a random read
356 */
Fengguang Wu2cad4012013-09-11 14:21:47 -0700357 if (size <= req_size)
Wu Fengguang10be0b32009-06-16 15:31:36 -0700358 return 0;
359
360 /*
361 * starts from beginning of file:
362 * it is a strong indication of long-run stream (or whole-file-read)
363 */
364 if (size >= offset)
365 size *= 2;
366
367 ra->start = offset;
Fengguang Wu2cad4012013-09-11 14:21:47 -0700368 ra->size = min(size + req_size, max);
369 ra->async_size = 1;
Wu Fengguang10be0b32009-06-16 15:31:36 -0700370
371 return 1;
372}
373
374/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700375 * A minimal readahead algorithm for trivial sequential/random reads.
376 */
377static unsigned long
378ondemand_readahead(struct address_space *mapping,
379 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700380 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700381 unsigned long req_size)
382{
Wu Fengguangfc31d162009-06-16 15:31:21 -0700383 unsigned long max = max_sane_readahead(ra->ra_pages);
Damien Ramondaaf248a02013-11-12 15:08:16 -0800384 pgoff_t prev_offset;
Wu Fengguang045a2522009-06-16 15:31:33 -0700385
386 /*
387 * start of file
388 */
389 if (!offset)
390 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700391
392 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700393 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700394 * Ramp up sizes, and push forward the readahead window.
395 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700396 if ((offset == (ra->start + ra->size - ra->async_size) ||
397 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700398 ra->start += ra->size;
399 ra->size = get_next_ra_size(ra, max);
400 ra->async_size = ra->size;
401 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700402 }
403
Fengguang Wu122a21d2007-07-19 01:48:01 -0700404 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700405 * Hit a marked page without valid readahead state.
406 * E.g. interleaved reads.
407 * Query the pagecache for async_size, which normally equals to
408 * readahead size. Ramp it up and use it as the new readahead size.
409 */
410 if (hit_readahead_marker) {
411 pgoff_t start;
412
Nick Piggin30002ed2008-07-25 19:45:28 -0700413 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700414 start = page_cache_next_hole(mapping, offset + 1, max);
Nick Piggin30002ed2008-07-25 19:45:28 -0700415 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700416
417 if (!start || start - offset > max)
418 return 0;
419
420 ra->start = start;
421 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700422 ra->size += req_size;
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700423 ra->size = get_next_ra_size(ra, max);
424 ra->async_size = ra->size;
425 goto readit;
426 }
427
428 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700429 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700430 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700431 if (req_size > max)
432 goto initial_readahead;
433
434 /*
435 * sequential cache miss
Damien Ramondaaf248a02013-11-12 15:08:16 -0800436 * trivial case: (offset - prev_offset) == 1
437 * unaligned reads: (offset - prev_offset) == 0
Wu Fengguang045a2522009-06-16 15:31:33 -0700438 */
Damien Ramondaaf248a02013-11-12 15:08:16 -0800439 prev_offset = (unsigned long long)ra->prev_pos >> PAGE_CACHE_SHIFT;
440 if (offset - prev_offset <= 1UL)
Wu Fengguang045a2522009-06-16 15:31:33 -0700441 goto initial_readahead;
442
443 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700444 * Query the page cache and look for the traces(cached history pages)
445 * that a sequential stream would leave behind.
446 */
447 if (try_context_readahead(mapping, ra, offset, req_size, max))
448 goto readit;
449
450 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700451 * standalone, small random read
452 * Read as is, and do not pollute the readahead state.
453 */
454 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
455
456initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700457 ra->start = offset;
458 ra->size = get_init_ra_size(req_size, max);
459 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700460
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700461readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700462 /*
463 * Will this read hit the readahead marker made by itself?
464 * If so, trigger the readahead marker hit now, and merge
465 * the resulted next readahead window into the current one.
466 */
467 if (offset == ra->start && ra->size == ra->async_size) {
468 ra->async_size = get_next_ra_size(ra, max);
469 ra->size += ra->async_size;
470 }
471
Fengguang Wu122a21d2007-07-19 01:48:01 -0700472 return ra_submit(ra, mapping, filp);
473}
474
475/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700476 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700477 * @mapping: address_space which holds the pagecache and I/O vectors
478 * @ra: file_ra_state which holds the readahead state
479 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700480 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700481 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700482 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700483 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700484 * page_cache_sync_readahead() should be called when a cache miss happened:
485 * it will submit the read. The readahead logic may decide to piggyback more
486 * pages onto the read request if access patterns suggest it will improve
487 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700488 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700489void page_cache_sync_readahead(struct address_space *mapping,
490 struct file_ra_state *ra, struct file *filp,
491 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700492{
493 /* no read-ahead */
494 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700495 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700496
Wu Fengguang01414502010-03-05 13:42:03 -0800497 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700498 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800499 force_page_cache_readahead(mapping, filp, offset, req_size);
500 return;
501 }
502
Fengguang Wu122a21d2007-07-19 01:48:01 -0700503 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700504 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700505}
Rusty Russellcf914a72007-07-19 01:48:08 -0700506EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
507
508/**
509 * page_cache_async_readahead - file readahead for marked pages
510 * @mapping: address_space which holds the pagecache and I/O vectors
511 * @ra: file_ra_state which holds the readahead state
512 * @filp: passed on to ->readpage() and ->readpages()
513 * @page: the page at @offset which has the PG_readahead flag set
514 * @offset: start offset into @mapping, in pagecache page-sized units
515 * @req_size: hint: total size of the read which the caller is performing in
516 * pagecache pages
517 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700518 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700519 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700520 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700521 * more pages.
522 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700523void
524page_cache_async_readahead(struct address_space *mapping,
525 struct file_ra_state *ra, struct file *filp,
526 struct page *page, pgoff_t offset,
527 unsigned long req_size)
528{
529 /* no read-ahead */
530 if (!ra->ra_pages)
531 return;
532
533 /*
534 * Same bit is used for PG_readahead and PG_reclaim.
535 */
536 if (PageWriteback(page))
537 return;
538
539 ClearPageReadahead(page);
540
541 /*
542 * Defer asynchronous read-ahead on IO congestion.
543 */
Tejun Heo703c2702015-05-22 17:13:44 -0400544 if (inode_read_congested(mapping->host))
Rusty Russellcf914a72007-07-19 01:48:08 -0700545 return;
546
547 /* do read-ahead */
548 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
549}
550EXPORT_SYMBOL_GPL(page_cache_async_readahead);
Cong Wang782182e2012-05-29 15:06:43 -0700551
552static ssize_t
553do_readahead(struct address_space *mapping, struct file *filp,
554 pgoff_t index, unsigned long nr)
555{
Andrew Morton63d0f0a32013-11-12 15:07:09 -0800556 if (!mapping || !mapping->a_ops)
Cong Wang782182e2012-05-29 15:06:43 -0700557 return -EINVAL;
558
Mark Rutland58d56402014-01-29 14:05:51 -0800559 return force_page_cache_readahead(mapping, filp, index, nr);
Cong Wang782182e2012-05-29 15:06:43 -0700560}
561
Al Viro4a0fd5b2013-01-21 15:16:58 -0500562SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
Cong Wang782182e2012-05-29 15:06:43 -0700563{
564 ssize_t ret;
Al Viro2903ff02012-08-28 12:52:22 -0400565 struct fd f;
Cong Wang782182e2012-05-29 15:06:43 -0700566
567 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400568 f = fdget(fd);
569 if (f.file) {
570 if (f.file->f_mode & FMODE_READ) {
571 struct address_space *mapping = f.file->f_mapping;
Cong Wang782182e2012-05-29 15:06:43 -0700572 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
573 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
574 unsigned long len = end - start + 1;
Al Viro2903ff02012-08-28 12:52:22 -0400575 ret = do_readahead(mapping, f.file, start, len);
Cong Wang782182e2012-05-29 15:06:43 -0700576 }
Al Viro2903ff02012-08-28 12:52:22 -0400577 fdput(f);
Cong Wang782182e2012-05-29 15:06:43 -0700578 }
579 return ret;
580}