blob: 40be3ae0afe3061e2bf762cc033f3142972c8ffd [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 Hockoc62d2552015-11-06 16:28:49 -080092 mapping_gfp_constraint(mapping, GFP_KERNEL))) {
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,
111 struct list_head *pages, unsigned nr_pages)
112{
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 Hocko063d99b2015-10-15 15:28:24 -0700129 if (!add_to_page_cache_lru(page, mapping, page->index,
Michal Hockoc62d2552015-11-06 16:28:49 -0800130 mapping_gfp_constraint(mapping, GFP_KERNEL))) {
Zach Brown9f1a3cf2006-06-25 05:46:46 -0700131 mapping->a_ops->readpage(filp, page);
Nick Piggineb2be182007-10-16 01:24:57 -0700132 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300133 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800135 ret = 0;
Jens Axboe5b417b12010-04-19 10:04:38 +0200136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137out:
Jens Axboe5b417b12010-04-19 10:04:38 +0200138 blk_finish_plug(&plug);
139
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 return ret;
141}
142
143/*
Wu Fengguangd30a1102009-06-16 15:31:30 -0700144 * __do_page_cache_readahead() actually reads a chunk of disk. It allocates all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 * the pages first, then submits them all for I/O. This avoids the very bad
146 * behaviour which would occur if page allocations are causing VM writeback.
147 * We really don't want to intermingle reads and writes like that.
148 *
149 * Returns the number of pages requested, or the maximum amount of I/O allowed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 */
Fabian Frederick29f175d2014-04-07 15:37:55 -0700151int __do_page_cache_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700152 pgoff_t offset, unsigned long nr_to_read,
153 unsigned long lookahead_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
155 struct inode *inode = mapping->host;
156 struct page *page;
157 unsigned long end_index; /* The last page we want to read */
158 LIST_HEAD(page_pool);
159 int page_idx;
160 int ret = 0;
161 loff_t isize = i_size_read(inode);
162
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
Wu Fengguang7b1de582011-05-24 17:12:25 -0700183 page = page_cache_alloc_readahead(mapping);
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)
199 read_pages(mapping, filp, &page_pool, ret);
200 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,
Andrew Morton7361f4d2005-11-07 00:59:28 -0800210 pgoff_t offset, unsigned long nr_to_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages))
213 return -EINVAL;
214
Roman Gushchin600e19a2015-11-05 18:47:08 -0800215 nr_to_read = min(nr_to_read, inode_to_bdi(mapping->host)->ra_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 while (nr_to_read) {
217 int err;
218
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300219 unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 if (this_chunk > nr_to_read)
222 this_chunk = nr_to_read;
223 err = __do_page_cache_readahead(mapping, filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700224 offset, this_chunk, 0);
Mark Rutland58d56402014-01-29 14:05:51 -0800225 if (err < 0)
226 return err;
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 offset += this_chunk;
229 nr_to_read -= this_chunk;
230 }
Mark Rutland58d56402014-01-29 14:05:51 -0800231 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
Fengguang Wu5ce11102007-07-19 01:47:59 -0700234/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700235 * Set the initial window size, round to next power of 2 and square
236 * for small size, x 4 for medium, and x 2 for large
237 * for 128k (32 page) max ra
238 * 1-8 page = 32k initial, > 8 page = 128k initial
239 */
240static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
241{
242 unsigned long newsize = roundup_pow_of_two(size);
243
244 if (newsize <= max / 32)
245 newsize = newsize * 4;
246 else if (newsize <= max / 4)
247 newsize = newsize * 2;
248 else
249 newsize = max;
250
251 return newsize;
252}
253
254/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700255 * Get the previous window size, ramp it up, and
256 * return it as the new window size.
257 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700258static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700259 unsigned long max)
260{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700261 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700262 unsigned long newsize;
263
264 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700265 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700266 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700267 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700268
269 return min(newsize, max);
270}
271
272/*
273 * On-demand readahead design.
274 *
275 * The fields in struct file_ra_state represent the most-recently-executed
276 * readahead attempt:
277 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700278 * |<----- async_size ---------|
279 * |------------------- size -------------------->|
280 * |==================#===========================|
281 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700282 *
283 * To overlap application thinking time and disk I/O time, we do
284 * `readahead pipelining': Do not wait until the application consumed all
285 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700286 * Instead, submit an asynchronous readahead I/O as soon as there are
287 * only async_size pages left in the readahead window. Normally async_size
288 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700289 *
290 * In interleaved sequential reads, concurrent streams on the same fd can
291 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700292 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700293 * indicator. The flag won't be set on already cached pages, to avoid the
294 * readahead-for-nothing fuss, saving pointless page cache lookups.
295 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700296 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700297 * It should be maintained by the caller, and will be used for detecting
298 * small random reads. Note that the readahead algorithm checks loosely
299 * for sequential patterns. Hence interleaved reads might be served as
300 * sequential ones.
301 *
302 * There is a special-case: if the first page which the application tries to
303 * read happens to be the first page of the file, it is assumed that a linear
304 * read is about to happen and the window is immediately set to the initial size
305 * based on I/O request size and the max_readahead.
306 *
307 * The code ramps up the readahead size aggressively at first, but slow down as
308 * it approaches max_readhead.
309 */
310
311/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700312 * Count contiguously cached pages from @offset-1 to @offset-@max,
313 * this count is a conservative estimation of
314 * - length of the sequential read sequence, or
315 * - thrashing threshold in memory tight systems
316 */
317static pgoff_t count_history_pages(struct address_space *mapping,
Wu Fengguang10be0b32009-06-16 15:31:36 -0700318 pgoff_t offset, unsigned long max)
319{
320 pgoff_t head;
321
322 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700323 head = page_cache_prev_hole(mapping, offset - 1, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700324 rcu_read_unlock();
325
326 return offset - 1 - head;
327}
328
329/*
330 * page cache context based read-ahead
331 */
332static int try_context_readahead(struct address_space *mapping,
333 struct file_ra_state *ra,
334 pgoff_t offset,
335 unsigned long req_size,
336 unsigned long max)
337{
338 pgoff_t size;
339
Fabian Frederick3e2faa082014-08-06 16:04:55 -0700340 size = count_history_pages(mapping, offset, max);
Wu Fengguang10be0b32009-06-16 15:31:36 -0700341
342 /*
Fengguang Wu2cad4012013-09-11 14:21:47 -0700343 * not enough history pages:
Wu Fengguang10be0b32009-06-16 15:31:36 -0700344 * it could be a random read
345 */
Fengguang Wu2cad4012013-09-11 14:21:47 -0700346 if (size <= req_size)
Wu Fengguang10be0b32009-06-16 15:31:36 -0700347 return 0;
348
349 /*
350 * starts from beginning of file:
351 * it is a strong indication of long-run stream (or whole-file-read)
352 */
353 if (size >= offset)
354 size *= 2;
355
356 ra->start = offset;
Fengguang Wu2cad4012013-09-11 14:21:47 -0700357 ra->size = min(size + req_size, max);
358 ra->async_size = 1;
Wu Fengguang10be0b32009-06-16 15:31:36 -0700359
360 return 1;
361}
362
363/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700364 * A minimal readahead algorithm for trivial sequential/random reads.
365 */
366static unsigned long
367ondemand_readahead(struct address_space *mapping,
368 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700369 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700370 unsigned long req_size)
371{
Roman Gushchin600e19a2015-11-05 18:47:08 -0800372 unsigned long max = ra->ra_pages;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800373 pgoff_t prev_offset;
Wu Fengguang045a2522009-06-16 15:31:33 -0700374
375 /*
376 * start of file
377 */
378 if (!offset)
379 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700380
381 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700382 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700383 * Ramp up sizes, and push forward the readahead window.
384 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700385 if ((offset == (ra->start + ra->size - ra->async_size) ||
386 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700387 ra->start += ra->size;
388 ra->size = get_next_ra_size(ra, max);
389 ra->async_size = ra->size;
390 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700391 }
392
Fengguang Wu122a21d2007-07-19 01:48:01 -0700393 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700394 * Hit a marked page without valid readahead state.
395 * E.g. interleaved reads.
396 * Query the pagecache for async_size, which normally equals to
397 * readahead size. Ramp it up and use it as the new readahead size.
398 */
399 if (hit_readahead_marker) {
400 pgoff_t start;
401
Nick Piggin30002ed2008-07-25 19:45:28 -0700402 rcu_read_lock();
Johannes Weinere7b563b2014-04-03 14:47:44 -0700403 start = page_cache_next_hole(mapping, offset + 1, max);
Nick Piggin30002ed2008-07-25 19:45:28 -0700404 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700405
406 if (!start || start - offset > max)
407 return 0;
408
409 ra->start = start;
410 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700411 ra->size += req_size;
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700412 ra->size = get_next_ra_size(ra, max);
413 ra->async_size = ra->size;
414 goto readit;
415 }
416
417 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700418 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700419 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700420 if (req_size > max)
421 goto initial_readahead;
422
423 /*
424 * sequential cache miss
Damien Ramondaaf248a02013-11-12 15:08:16 -0800425 * trivial case: (offset - prev_offset) == 1
426 * unaligned reads: (offset - prev_offset) == 0
Wu Fengguang045a2522009-06-16 15:31:33 -0700427 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300428 prev_offset = (unsigned long long)ra->prev_pos >> PAGE_SHIFT;
Damien Ramondaaf248a02013-11-12 15:08:16 -0800429 if (offset - prev_offset <= 1UL)
Wu Fengguang045a2522009-06-16 15:31:33 -0700430 goto initial_readahead;
431
432 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700433 * Query the page cache and look for the traces(cached history pages)
434 * that a sequential stream would leave behind.
435 */
436 if (try_context_readahead(mapping, ra, offset, req_size, max))
437 goto readit;
438
439 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700440 * standalone, small random read
441 * Read as is, and do not pollute the readahead state.
442 */
443 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
444
445initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700446 ra->start = offset;
447 ra->size = get_init_ra_size(req_size, max);
448 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700449
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700450readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700451 /*
452 * Will this read hit the readahead marker made by itself?
453 * If so, trigger the readahead marker hit now, and merge
454 * the resulted next readahead window into the current one.
455 */
456 if (offset == ra->start && ra->size == ra->async_size) {
457 ra->async_size = get_next_ra_size(ra, max);
458 ra->size += ra->async_size;
459 }
460
Fengguang Wu122a21d2007-07-19 01:48:01 -0700461 return ra_submit(ra, mapping, filp);
462}
463
464/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700465 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700466 * @mapping: address_space which holds the pagecache and I/O vectors
467 * @ra: file_ra_state which holds the readahead state
468 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700469 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700470 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700471 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700472 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700473 * page_cache_sync_readahead() should be called when a cache miss happened:
474 * it will submit the read. The readahead logic may decide to piggyback more
475 * pages onto the read request if access patterns suggest it will improve
476 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700477 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700478void page_cache_sync_readahead(struct address_space *mapping,
479 struct file_ra_state *ra, struct file *filp,
480 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700481{
482 /* no read-ahead */
483 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700484 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700485
Wu Fengguang01414502010-03-05 13:42:03 -0800486 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700487 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800488 force_page_cache_readahead(mapping, filp, offset, req_size);
489 return;
490 }
491
Fengguang Wu122a21d2007-07-19 01:48:01 -0700492 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700493 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700494}
Rusty Russellcf914a72007-07-19 01:48:08 -0700495EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
496
497/**
498 * page_cache_async_readahead - file readahead for marked pages
499 * @mapping: address_space which holds the pagecache and I/O vectors
500 * @ra: file_ra_state which holds the readahead state
501 * @filp: passed on to ->readpage() and ->readpages()
502 * @page: the page at @offset which has the PG_readahead flag set
503 * @offset: start offset into @mapping, in pagecache page-sized units
504 * @req_size: hint: total size of the read which the caller is performing in
505 * pagecache pages
506 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700507 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700508 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700509 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700510 * more pages.
511 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700512void
513page_cache_async_readahead(struct address_space *mapping,
514 struct file_ra_state *ra, struct file *filp,
515 struct page *page, pgoff_t offset,
516 unsigned long req_size)
517{
518 /* no read-ahead */
519 if (!ra->ra_pages)
520 return;
521
522 /*
523 * Same bit is used for PG_readahead and PG_reclaim.
524 */
525 if (PageWriteback(page))
526 return;
527
528 ClearPageReadahead(page);
529
530 /*
531 * Defer asynchronous read-ahead on IO congestion.
532 */
Tejun Heo703c2702015-05-22 17:13:44 -0400533 if (inode_read_congested(mapping->host))
Rusty Russellcf914a72007-07-19 01:48:08 -0700534 return;
535
536 /* do read-ahead */
537 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
538}
539EXPORT_SYMBOL_GPL(page_cache_async_readahead);
Cong Wang782182e2012-05-29 15:06:43 -0700540
541static ssize_t
542do_readahead(struct address_space *mapping, struct file *filp,
543 pgoff_t index, unsigned long nr)
544{
Andrew Morton63d0f0a32013-11-12 15:07:09 -0800545 if (!mapping || !mapping->a_ops)
Cong Wang782182e2012-05-29 15:06:43 -0700546 return -EINVAL;
547
Mark Rutland58d56402014-01-29 14:05:51 -0800548 return force_page_cache_readahead(mapping, filp, index, nr);
Cong Wang782182e2012-05-29 15:06:43 -0700549}
550
Al Viro4a0fd5b2013-01-21 15:16:58 -0500551SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
Cong Wang782182e2012-05-29 15:06:43 -0700552{
553 ssize_t ret;
Al Viro2903ff02012-08-28 12:52:22 -0400554 struct fd f;
Cong Wang782182e2012-05-29 15:06:43 -0700555
556 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400557 f = fdget(fd);
558 if (f.file) {
559 if (f.file->f_mode & FMODE_READ) {
560 struct address_space *mapping = f.file->f_mapping;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300561 pgoff_t start = offset >> PAGE_SHIFT;
562 pgoff_t end = (offset + count - 1) >> PAGE_SHIFT;
Cong Wang782182e2012-05-29 15:06:43 -0700563 unsigned long len = end - start + 1;
Al Viro2903ff02012-08-28 12:52:22 -0400564 ret = do_readahead(mapping, f.file, start, len);
Cong Wang782182e2012-05-29 15:06:43 -0700565 }
Al Viro2903ff02012-08-28 12:52:22 -0400566 fdput(f);
Cong Wang782182e2012-05-29 15:06:43 -0700567 }
568 return ret;
569}