blob: 7cdbb44aa90bd99bef05f43ee27cc13b2121ed7a [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>
11#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040014#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/blkdev.h>
16#include <linux/backing-dev.h>
Andrew Morton8bde37f2006-12-10 02:19:40 -080017#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/pagevec.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020019#include <linux/pagemap.h>
Cong Wang782182e2012-05-29 15:06:43 -070020#include <linux/syscalls.h>
21#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
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{
30 ra->ra_pages = mapping->backing_dev_info->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 */
152static int
153__do_page_cache_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700154 pgoff_t offset, unsigned long nr_to_read,
155 unsigned long lookahead_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156{
157 struct inode *inode = mapping->host;
158 struct page *page;
159 unsigned long end_index; /* The last page we want to read */
160 LIST_HEAD(page_pool);
161 int page_idx;
162 int ret = 0;
163 loff_t isize = i_size_read(inode);
164
165 if (isize == 0)
166 goto out;
167
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700168 end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170 /*
171 * Preallocate as many pages as we will need.
172 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 for (page_idx = 0; page_idx < nr_to_read; page_idx++) {
Andrew Morton7361f4d2005-11-07 00:59:28 -0800174 pgoff_t page_offset = offset + page_idx;
Fengguang Wuc743d962007-07-19 01:48:04 -0700175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 if (page_offset > end_index)
177 break;
178
Nick Piggin00128182007-10-16 01:24:40 -0700179 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 page = radix_tree_lookup(&mapping->page_tree, page_offset);
Nick Piggin00128182007-10-16 01:24:40 -0700181 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 if (page)
183 continue;
184
Wu Fengguang7b1de582011-05-24 17:12:25 -0700185 page = page_cache_alloc_readahead(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 if (!page)
187 break;
188 page->index = page_offset;
189 list_add(&page->lru, &page_pool);
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700190 if (page_idx == nr_to_read - lookahead_size)
191 SetPageReadahead(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 ret++;
193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 /*
196 * Now start the IO. We ignore I/O errors - if the page is not
197 * uptodate then the caller will launch readpage again, and
198 * will then handle the error.
199 */
200 if (ret)
201 read_pages(mapping, filp, &page_pool, ret);
202 BUG_ON(!list_empty(&page_pool));
203out:
204 return ret;
205}
206
207/*
208 * Chunk the readahead into 2 megabyte units, so that we don't pin too much
209 * memory at once.
210 */
211int force_page_cache_readahead(struct address_space *mapping, struct file *filp,
Andrew Morton7361f4d2005-11-07 00:59:28 -0800212 pgoff_t offset, unsigned long nr_to_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 int ret = 0;
215
216 if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages))
217 return -EINVAL;
218
Wu Fengguangf7e839d2009-06-16 15:31:20 -0700219 nr_to_read = max_sane_readahead(nr_to_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 while (nr_to_read) {
221 int err;
222
223 unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_CACHE_SIZE;
224
225 if (this_chunk > nr_to_read)
226 this_chunk = nr_to_read;
227 err = __do_page_cache_readahead(mapping, filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700228 offset, this_chunk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 if (err < 0) {
230 ret = err;
231 break;
232 }
233 ret += err;
234 offset += this_chunk;
235 nr_to_read -= this_chunk;
236 }
237 return ret;
238}
239
240/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 * Given a desired number of PAGE_CACHE_SIZE readahead pages, return a
242 * sensible upper limit.
243 */
244unsigned long max_sane_readahead(unsigned long nr)
245{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700246 return min(nr, (node_page_state(numa_node_id(), NR_INACTIVE_FILE)
Christoph Lameter05a04162007-02-10 01:43:05 -0800247 + node_page_state(numa_node_id(), NR_FREE_PAGES)) / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
Fengguang Wu5ce11102007-07-19 01:47:59 -0700249
250/*
251 * Submit IO for the read-ahead request in file_ra_state.
252 */
Wu Fengguangd30a1102009-06-16 15:31:30 -0700253unsigned long ra_submit(struct file_ra_state *ra,
Fengguang Wu5ce11102007-07-19 01:47:59 -0700254 struct address_space *mapping, struct file *filp)
255{
Fengguang Wu5ce11102007-07-19 01:47:59 -0700256 int actual;
257
Fengguang Wu5ce11102007-07-19 01:47:59 -0700258 actual = __do_page_cache_readahead(mapping, filp,
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700259 ra->start, ra->size, ra->async_size);
Fengguang Wu5ce11102007-07-19 01:47:59 -0700260
261 return actual;
262}
Fengguang Wu122a21d2007-07-19 01:48:01 -0700263
264/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700265 * Set the initial window size, round to next power of 2 and square
266 * for small size, x 4 for medium, and x 2 for large
267 * for 128k (32 page) max ra
268 * 1-8 page = 32k initial, > 8 page = 128k initial
269 */
270static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
271{
272 unsigned long newsize = roundup_pow_of_two(size);
273
274 if (newsize <= max / 32)
275 newsize = newsize * 4;
276 else if (newsize <= max / 4)
277 newsize = newsize * 2;
278 else
279 newsize = max;
280
281 return newsize;
282}
283
284/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700285 * Get the previous window size, ramp it up, and
286 * return it as the new window size.
287 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700288static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700289 unsigned long max)
290{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700291 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700292 unsigned long newsize;
293
294 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700295 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700296 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700297 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700298
299 return min(newsize, max);
300}
301
302/*
303 * On-demand readahead design.
304 *
305 * The fields in struct file_ra_state represent the most-recently-executed
306 * readahead attempt:
307 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700308 * |<----- async_size ---------|
309 * |------------------- size -------------------->|
310 * |==================#===========================|
311 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700312 *
313 * To overlap application thinking time and disk I/O time, we do
314 * `readahead pipelining': Do not wait until the application consumed all
315 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700316 * Instead, submit an asynchronous readahead I/O as soon as there are
317 * only async_size pages left in the readahead window. Normally async_size
318 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700319 *
320 * In interleaved sequential reads, concurrent streams on the same fd can
321 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700322 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700323 * indicator. The flag won't be set on already cached pages, to avoid the
324 * readahead-for-nothing fuss, saving pointless page cache lookups.
325 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700326 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700327 * It should be maintained by the caller, and will be used for detecting
328 * small random reads. Note that the readahead algorithm checks loosely
329 * for sequential patterns. Hence interleaved reads might be served as
330 * sequential ones.
331 *
332 * There is a special-case: if the first page which the application tries to
333 * read happens to be the first page of the file, it is assumed that a linear
334 * read is about to happen and the window is immediately set to the initial size
335 * based on I/O request size and the max_readahead.
336 *
337 * The code ramps up the readahead size aggressively at first, but slow down as
338 * it approaches max_readhead.
339 */
340
341/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700342 * Count contiguously cached pages from @offset-1 to @offset-@max,
343 * this count is a conservative estimation of
344 * - length of the sequential read sequence, or
345 * - thrashing threshold in memory tight systems
346 */
347static pgoff_t count_history_pages(struct address_space *mapping,
348 struct file_ra_state *ra,
349 pgoff_t offset, unsigned long max)
350{
351 pgoff_t head;
352
353 rcu_read_lock();
354 head = radix_tree_prev_hole(&mapping->page_tree, offset - 1, max);
355 rcu_read_unlock();
356
357 return offset - 1 - head;
358}
359
360/*
361 * page cache context based read-ahead
362 */
363static int try_context_readahead(struct address_space *mapping,
364 struct file_ra_state *ra,
365 pgoff_t offset,
366 unsigned long req_size,
367 unsigned long max)
368{
369 pgoff_t size;
370
371 size = count_history_pages(mapping, ra, offset, max);
372
373 /*
Fengguang Wu2cad4012013-09-11 14:21:47 -0700374 * not enough history pages:
Wu Fengguang10be0b32009-06-16 15:31:36 -0700375 * it could be a random read
376 */
Fengguang Wu2cad4012013-09-11 14:21:47 -0700377 if (size <= req_size)
Wu Fengguang10be0b32009-06-16 15:31:36 -0700378 return 0;
379
380 /*
381 * starts from beginning of file:
382 * it is a strong indication of long-run stream (or whole-file-read)
383 */
384 if (size >= offset)
385 size *= 2;
386
387 ra->start = offset;
Fengguang Wu2cad4012013-09-11 14:21:47 -0700388 ra->size = min(size + req_size, max);
389 ra->async_size = 1;
Wu Fengguang10be0b32009-06-16 15:31:36 -0700390
391 return 1;
392}
393
394/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700395 * A minimal readahead algorithm for trivial sequential/random reads.
396 */
397static unsigned long
398ondemand_readahead(struct address_space *mapping,
399 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700400 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700401 unsigned long req_size)
402{
Wu Fengguangfc31d162009-06-16 15:31:21 -0700403 unsigned long max = max_sane_readahead(ra->ra_pages);
Damien Ramondaaf248a02013-11-12 15:08:16 -0800404 pgoff_t prev_offset;
Wu Fengguang045a2522009-06-16 15:31:33 -0700405
406 /*
407 * start of file
408 */
409 if (!offset)
410 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700411
412 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700413 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700414 * Ramp up sizes, and push forward the readahead window.
415 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700416 if ((offset == (ra->start + ra->size - ra->async_size) ||
417 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700418 ra->start += ra->size;
419 ra->size = get_next_ra_size(ra, max);
420 ra->async_size = ra->size;
421 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700422 }
423
Fengguang Wu122a21d2007-07-19 01:48:01 -0700424 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700425 * Hit a marked page without valid readahead state.
426 * E.g. interleaved reads.
427 * Query the pagecache for async_size, which normally equals to
428 * readahead size. Ramp it up and use it as the new readahead size.
429 */
430 if (hit_readahead_marker) {
431 pgoff_t start;
432
Nick Piggin30002ed2008-07-25 19:45:28 -0700433 rcu_read_lock();
Wu Fengguangcaca7cb2009-06-16 15:31:21 -0700434 start = radix_tree_next_hole(&mapping->page_tree, offset+1,max);
Nick Piggin30002ed2008-07-25 19:45:28 -0700435 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700436
437 if (!start || start - offset > max)
438 return 0;
439
440 ra->start = start;
441 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700442 ra->size += req_size;
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700443 ra->size = get_next_ra_size(ra, max);
444 ra->async_size = ra->size;
445 goto readit;
446 }
447
448 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700449 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700450 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700451 if (req_size > max)
452 goto initial_readahead;
453
454 /*
455 * sequential cache miss
Damien Ramondaaf248a02013-11-12 15:08:16 -0800456 * trivial case: (offset - prev_offset) == 1
457 * unaligned reads: (offset - prev_offset) == 0
Wu Fengguang045a2522009-06-16 15:31:33 -0700458 */
Damien Ramondaaf248a02013-11-12 15:08:16 -0800459 prev_offset = (unsigned long long)ra->prev_pos >> PAGE_CACHE_SHIFT;
460 if (offset - prev_offset <= 1UL)
Wu Fengguang045a2522009-06-16 15:31:33 -0700461 goto initial_readahead;
462
463 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700464 * Query the page cache and look for the traces(cached history pages)
465 * that a sequential stream would leave behind.
466 */
467 if (try_context_readahead(mapping, ra, offset, req_size, max))
468 goto readit;
469
470 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700471 * standalone, small random read
472 * Read as is, and do not pollute the readahead state.
473 */
474 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
475
476initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700477 ra->start = offset;
478 ra->size = get_init_ra_size(req_size, max);
479 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700480
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700481readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700482 /*
483 * Will this read hit the readahead marker made by itself?
484 * If so, trigger the readahead marker hit now, and merge
485 * the resulted next readahead window into the current one.
486 */
487 if (offset == ra->start && ra->size == ra->async_size) {
488 ra->async_size = get_next_ra_size(ra, max);
489 ra->size += ra->async_size;
490 }
491
Fengguang Wu122a21d2007-07-19 01:48:01 -0700492 return ra_submit(ra, mapping, filp);
493}
494
495/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700496 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700497 * @mapping: address_space which holds the pagecache and I/O vectors
498 * @ra: file_ra_state which holds the readahead state
499 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700500 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700501 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700502 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700503 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700504 * page_cache_sync_readahead() should be called when a cache miss happened:
505 * it will submit the read. The readahead logic may decide to piggyback more
506 * pages onto the read request if access patterns suggest it will improve
507 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700508 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700509void page_cache_sync_readahead(struct address_space *mapping,
510 struct file_ra_state *ra, struct file *filp,
511 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700512{
513 /* no read-ahead */
514 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700515 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700516
Wu Fengguang01414502010-03-05 13:42:03 -0800517 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700518 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800519 force_page_cache_readahead(mapping, filp, offset, req_size);
520 return;
521 }
522
Fengguang Wu122a21d2007-07-19 01:48:01 -0700523 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700524 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700525}
Rusty Russellcf914a72007-07-19 01:48:08 -0700526EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
527
528/**
529 * page_cache_async_readahead - file readahead for marked pages
530 * @mapping: address_space which holds the pagecache and I/O vectors
531 * @ra: file_ra_state which holds the readahead state
532 * @filp: passed on to ->readpage() and ->readpages()
533 * @page: the page at @offset which has the PG_readahead flag set
534 * @offset: start offset into @mapping, in pagecache page-sized units
535 * @req_size: hint: total size of the read which the caller is performing in
536 * pagecache pages
537 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700538 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700539 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700540 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700541 * more pages.
542 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700543void
544page_cache_async_readahead(struct address_space *mapping,
545 struct file_ra_state *ra, struct file *filp,
546 struct page *page, pgoff_t offset,
547 unsigned long req_size)
548{
549 /* no read-ahead */
550 if (!ra->ra_pages)
551 return;
552
553 /*
554 * Same bit is used for PG_readahead and PG_reclaim.
555 */
556 if (PageWriteback(page))
557 return;
558
559 ClearPageReadahead(page);
560
561 /*
562 * Defer asynchronous read-ahead on IO congestion.
563 */
564 if (bdi_read_congested(mapping->backing_dev_info))
565 return;
566
567 /* do read-ahead */
568 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
569}
570EXPORT_SYMBOL_GPL(page_cache_async_readahead);
Cong Wang782182e2012-05-29 15:06:43 -0700571
572static ssize_t
573do_readahead(struct address_space *mapping, struct file *filp,
574 pgoff_t index, unsigned long nr)
575{
Andrew Morton63d0f0a32013-11-12 15:07:09 -0800576 if (!mapping || !mapping->a_ops)
Cong Wang782182e2012-05-29 15:06:43 -0700577 return -EINVAL;
578
579 force_page_cache_readahead(mapping, filp, index, nr);
580 return 0;
581}
582
Al Viro4a0fd5b2013-01-21 15:16:58 -0500583SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count)
Cong Wang782182e2012-05-29 15:06:43 -0700584{
585 ssize_t ret;
Al Viro2903ff02012-08-28 12:52:22 -0400586 struct fd f;
Cong Wang782182e2012-05-29 15:06:43 -0700587
588 ret = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400589 f = fdget(fd);
590 if (f.file) {
591 if (f.file->f_mode & FMODE_READ) {
592 struct address_space *mapping = f.file->f_mapping;
Cong Wang782182e2012-05-29 15:06:43 -0700593 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
594 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
595 unsigned long len = end - start + 1;
Al Viro2903ff02012-08-28 12:52:22 -0400596 ret = do_readahead(mapping, f.file, start, len);
Cong Wang782182e2012-05-29 15:06:43 -0700597 }
Al Viro2903ff02012-08-28 12:52:22 -0400598 fdput(f);
Cong Wang782182e2012-05-29 15:06:43 -0700599 }
600 return ret;
601}