blob: 998ad592f408332135d107c442e0beb879eb5d58 [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);
Michal Hocko063d99b2015-10-15 15:28:24 -070092 if (add_to_page_cache_lru(page, mapping, page->index,
93 GFP_KERNEL & mapping_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 }
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);
Michal Hocko063d99b2015-10-15 15:28:24 -0700130 if (!add_to_page_cache_lru(page, mapping, page->index,
131 GFP_KERNEL & mapping_gfp_mask(mapping))) {
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
Roman Gushchin600e19a2015-11-05 18:47:08 -0800216 nr_to_read = min(nr_to_read, inode_to_bdi(mapping->host)->ra_pages);
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
Fengguang Wu5ce11102007-07-19 01:47:59 -0700235/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700236 * Set the initial window size, round to next power of 2 and square
237 * for small size, x 4 for medium, and x 2 for large
238 * for 128k (32 page) max ra
239 * 1-8 page = 32k initial, > 8 page = 128k initial
240 */
241static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
242{
243 unsigned long newsize = roundup_pow_of_two(size);
244
245 if (newsize <= max / 32)
246 newsize = newsize * 4;
247 else if (newsize <= max / 4)
248 newsize = newsize * 2;
249 else
250 newsize = max;
251
252 return newsize;
253}
254
255/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700256 * Get the previous window size, ramp it up, and
257 * return it as the new window size.
258 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700259static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700260 unsigned long max)
261{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700262 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700263 unsigned long newsize;
264
265 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700266 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700267 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700268 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700269
270 return min(newsize, max);
271}
272
273/*
274 * On-demand readahead design.
275 *
276 * The fields in struct file_ra_state represent the most-recently-executed
277 * readahead attempt:
278 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700279 * |<----- async_size ---------|
280 * |------------------- size -------------------->|
281 * |==================#===========================|
282 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700283 *
284 * To overlap application thinking time and disk I/O time, we do
285 * `readahead pipelining': Do not wait until the application consumed all
286 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700287 * Instead, submit an asynchronous readahead I/O as soon as there are
288 * only async_size pages left in the readahead window. Normally async_size
289 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700290 *
291 * In interleaved sequential reads, concurrent streams on the same fd can
292 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700293 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700294 * indicator. The flag won't be set on already cached pages, to avoid the
295 * readahead-for-nothing fuss, saving pointless page cache lookups.
296 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700297 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700298 * It should be maintained by the caller, and will be used for detecting
299 * small random reads. Note that the readahead algorithm checks loosely
300 * for sequential patterns. Hence interleaved reads might be served as
301 * sequential ones.
302 *
303 * There is a special-case: if the first page which the application tries to
304 * read happens to be the first page of the file, it is assumed that a linear
305 * read is about to happen and the window is immediately set to the initial size
306 * based on I/O request size and the max_readahead.
307 *
308 * The code ramps up the readahead size aggressively at first, but slow down as
309 * it approaches max_readhead.
310 */
311
312/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700313 * Count contiguously cached pages from @offset-1 to @offset-@max,
314 * this count is a conservative estimation of
315 * - length of the sequential read sequence, or
316 * - thrashing threshold in memory tight systems
317 */
318static pgoff_t count_history_pages(struct address_space *mapping,
Wu Fengguang10be0b32009-06-16 15:31:36 -0700319 pgoff_t offset, unsigned long max)
320{
321 pgoff_t head;
322
323 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700324 head = page_cache_prev_hole(mapping, offset - 1, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700325 rcu_read_unlock();
326
327 return offset - 1 - head;
328}
329
330/*
331 * page cache context based read-ahead
332 */
333static int try_context_readahead(struct address_space *mapping,
334 struct file_ra_state *ra,
335 pgoff_t offset,
336 unsigned long req_size,
337 unsigned long max)
338{
339 pgoff_t size;
340
Fabian Frederick3e2faa082014-08-06 16:04:55 -0700341 size = count_history_pages(mapping, offset, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700342
343 /*
Fengguang Wu2cad4012013-09-11 14:21:47 -0700344 * not enough history pages:
Wu Fengguang10be0b32009-06-16 15:31:36 -0700345 * it could be a random read
346 */
Fengguang Wu2cad4012013-09-11 14:21:47 -0700347 if (size <= req_size)
Wu Fengguang10be0b32009-06-16 15:31:36 -0700348 return 0;
349
350 /*
351 * starts from beginning of file:
352 * it is a strong indication of long-run stream (or whole-file-read)
353 */
354 if (size >= offset)
355 size *= 2;
356
357 ra->start = offset;
Fengguang Wu2cad4012013-09-11 14:21:47 -0700358 ra->size = min(size + req_size, max);
359 ra->async_size = 1;
Wu Fengguang10be0b32009-06-16 15:31:36 -0700360
361 return 1;
362}
363
364/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700365 * A minimal readahead algorithm for trivial sequential/random reads.
366 */
367static unsigned long
368ondemand_readahead(struct address_space *mapping,
369 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700370 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700371 unsigned long req_size)
372{
Roman Gushchin600e19a2015-11-05 18:47:08 -0800373 unsigned long max = ra->ra_pages;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800374 pgoff_t prev_offset;
Wu Fengguang045a2522009-06-16 15:31:33 -0700375
376 /*
377 * start of file
378 */
379 if (!offset)
380 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700381
382 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700383 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700384 * Ramp up sizes, and push forward the readahead window.
385 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700386 if ((offset == (ra->start + ra->size - ra->async_size) ||
387 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700388 ra->start += ra->size;
389 ra->size = get_next_ra_size(ra, max);
390 ra->async_size = ra->size;
391 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700392 }
393
Fengguang Wu122a21d2007-07-19 01:48:01 -0700394 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700395 * Hit a marked page without valid readahead state.
396 * E.g. interleaved reads.
397 * Query the pagecache for async_size, which normally equals to
398 * readahead size. Ramp it up and use it as the new readahead size.
399 */
400 if (hit_readahead_marker) {
401 pgoff_t start;
402
Nick Piggin30002ed2008-07-25 19:45:28 -0700403 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700404 start = page_cache_next_hole(mapping, offset + 1, max);
Nick Piggin30002ed2008-07-25 19:45:28 -0700405 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700406
407 if (!start || start - offset > max)
408 return 0;
409
410 ra->start = start;
411 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700412 ra->size += req_size;
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700413 ra->size = get_next_ra_size(ra, max);
414 ra->async_size = ra->size;
415 goto readit;
416 }
417
418 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700419 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700420 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700421 if (req_size > max)
422 goto initial_readahead;
423
424 /*
425 * sequential cache miss
Damien Ramondaaf248a02013-11-12 15:08:16 -0800426 * trivial case: (offset - prev_offset) == 1
427 * unaligned reads: (offset - prev_offset) == 0
Wu Fengguang045a2522009-06-16 15:31:33 -0700428 */
Damien Ramondaaf248a02013-11-12 15:08:16 -0800429 prev_offset = (unsigned long long)ra->prev_pos >> PAGE_CACHE_SHIFT;
430 if (offset - prev_offset <= 1UL)
Wu Fengguang045a2522009-06-16 15:31:33 -0700431 goto initial_readahead;
432
433 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700434 * Query the page cache and look for the traces(cached history pages)
435 * that a sequential stream would leave behind.
436 */
437 if (try_context_readahead(mapping, ra, offset, req_size, max))
438 goto readit;
439
440 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700441 * standalone, small random read
442 * Read as is, and do not pollute the readahead state.
443 */
444 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
445
446initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700447 ra->start = offset;
448 ra->size = get_init_ra_size(req_size, max);
449 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700450
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700451readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700452 /*
453 * Will this read hit the readahead marker made by itself?
454 * If so, trigger the readahead marker hit now, and merge
455 * the resulted next readahead window into the current one.
456 */
457 if (offset == ra->start && ra->size == ra->async_size) {
458 ra->async_size = get_next_ra_size(ra, max);
459 ra->size += ra->async_size;
460 }
461
Fengguang Wu122a21d2007-07-19 01:48:01 -0700462 return ra_submit(ra, mapping, filp);
463}
464
465/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700466 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700467 * @mapping: address_space which holds the pagecache and I/O vectors
468 * @ra: file_ra_state which holds the readahead state
469 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700470 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700471 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700472 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700473 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700474 * page_cache_sync_readahead() should be called when a cache miss happened:
475 * it will submit the read. The readahead logic may decide to piggyback more
476 * pages onto the read request if access patterns suggest it will improve
477 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700478 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700479void page_cache_sync_readahead(struct address_space *mapping,
480 struct file_ra_state *ra, struct file *filp,
481 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700482{
483 /* no read-ahead */
484 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700485 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700486
Wu Fengguang01414502010-03-05 13:42:03 -0800487 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700488 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800489 force_page_cache_readahead(mapping, filp, offset, req_size);
490 return;
491 }
492
Fengguang Wu122a21d2007-07-19 01:48:01 -0700493 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700494 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700495}
Rusty Russellcf914a72007-07-19 01:48:08 -0700496EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
497
498/**
499 * page_cache_async_readahead - file readahead for marked pages
500 * @mapping: address_space which holds the pagecache and I/O vectors
501 * @ra: file_ra_state which holds the readahead state
502 * @filp: passed on to ->readpage() and ->readpages()
503 * @page: the page at @offset which has the PG_readahead flag set
504 * @offset: start offset into @mapping, in pagecache page-sized units
505 * @req_size: hint: total size of the read which the caller is performing in
506 * pagecache pages
507 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700508 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700509 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700510 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700511 * more pages.
512 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700513void
514page_cache_async_readahead(struct address_space *mapping,
515 struct file_ra_state *ra, struct file *filp,
516 struct page *page, pgoff_t offset,
517 unsigned long req_size)
518{
519 /* no read-ahead */
520 if (!ra->ra_pages)
521 return;
522
523 /*
524 * Same bit is used for PG_readahead and PG_reclaim.
525 */
526 if (PageWriteback(page))
527 return;
528
529 ClearPageReadahead(page);
530
531 /*
532 * Defer asynchronous read-ahead on IO congestion.
533 */
Tejun Heo703c2702015-05-22 17:13:44 -0400534 if (inode_read_congested(mapping->host))
Rusty Russellcf914a72007-07-19 01:48:08 -0700535 return;
536
537 /* do read-ahead */
538 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
539}
540EXPORT_SYMBOL_GPL(page_cache_async_readahead);
Cong Wang782182e2012-05-29 15:06:43 -0700541
542static ssize_t
543do_readahead(struct address_space *mapping, struct file *filp,
544 pgoff_t index, unsigned long nr)
545{
Andrew Morton63d0f0a32013-11-12 15:07:09 -0800546 if (!mapping || !mapping->a_ops)
Cong Wang782182e2012-05-29 15:06:43 -0700547 return -EINVAL;
548
Mark Rutland58d56402014-01-29 14:05:51 -0800549 return force_page_cache_readahead(mapping, filp, index, nr);
Cong Wang782182e2012-05-29 15:06:43 -0700550}
551
Al Viro4a0fd5b2013-01-21 15:16:58 -0500552SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
Cong Wang782182e2012-05-29 15:06:43 -0700553{
554 ssize_t ret;
Al Viro2903ff02012-08-28 12:52:22 -0400555 struct fd f;
Cong Wang782182e2012-05-29 15:06:43 -0700556
557 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400558 f = fdget(fd);
559 if (f.file) {
560 if (f.file->f_mode & FMODE_READ) {
561 struct address_space *mapping = f.file->f_mapping;
Cong Wang782182e2012-05-29 15:06:43 -0700562 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
563 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
564 unsigned long len = end - start + 1;
Al Viro2903ff02012-08-28 12:52:22 -0400565 ret = do_readahead(mapping, f.file, start, len);
Cong Wang782182e2012-05-29 15:06:43 -0700566 }
Al Viro2903ff02012-08-28 12:52:22 -0400567 fdput(f);
Cong Wang782182e2012-05-29 15:06:43 -0700568 }
569 return ret;
570}