blob: 2c0cc489e2880cb319a92b905a5ee7b292039b7f [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>
14#include <linux/module.h>
15#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Linus Torvalds1da177e2005-04-16 15:20:36 -070021/*
22 * Initialise a struct file's readahead state. Assumes that the caller has
23 * memset *ra to zero.
24 */
25void
26file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping)
27{
28 ra->ra_pages = mapping->backing_dev_info->ra_pages;
Fengguang Wuf4e6b492007-10-16 01:24:33 -070029 ra->prev_pos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070030}
Steven Whitehoused41cc702006-01-30 08:53:33 +000031EXPORT_SYMBOL_GPL(file_ra_state_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#define list_to_page(head) (list_entry((head)->prev, struct page, lru))
34
David Howells03fb3d22009-04-03 16:42:35 +010035/*
36 * see if a page needs releasing upon read_cache_pages() failure
David Howells266cf652009-04-03 16:42:36 +010037 * - the caller of read_cache_pages() may have set PG_private or PG_fscache
38 * before calling, such as the NFS fs marking pages that are cached locally
39 * on disk, thus we need to give the fs a chance to clean up in the event of
40 * an error
David Howells03fb3d22009-04-03 16:42:35 +010041 */
42static void read_cache_pages_invalidate_page(struct address_space *mapping,
43 struct page *page)
44{
David Howells266cf652009-04-03 16:42:36 +010045 if (page_has_private(page)) {
David Howells03fb3d22009-04-03 16:42:35 +010046 if (!trylock_page(page))
47 BUG();
48 page->mapping = mapping;
49 do_invalidatepage(page, 0);
50 page->mapping = NULL;
51 unlock_page(page);
52 }
53 page_cache_release(page);
54}
55
56/*
57 * release a list of pages, invalidating them first if need be
58 */
59static void read_cache_pages_invalidate_pages(struct address_space *mapping,
60 struct list_head *pages)
61{
62 struct page *victim;
63
64 while (!list_empty(pages)) {
65 victim = list_to_page(pages);
66 list_del(&victim->lru);
67 read_cache_pages_invalidate_page(mapping, victim);
68 }
69}
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071/**
Randy Dunlapbd40cdd2006-06-25 05:48:08 -070072 * read_cache_pages - populate an address space with some pages & start reads against them
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * @mapping: the address_space
74 * @pages: The address of a list_head which contains the target pages. These
75 * pages have their ->index populated and are otherwise uninitialised.
76 * @filler: callback routine for filling a single page.
77 * @data: private data for the callback routine.
78 *
79 * Hides the details of the LRU cache etc from the filesystems.
80 */
81int read_cache_pages(struct address_space *mapping, struct list_head *pages,
82 int (*filler)(void *, struct page *), void *data)
83{
84 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 int ret = 0;
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 while (!list_empty(pages)) {
88 page = list_to_page(pages);
89 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -070090 if (add_to_page_cache_lru(page, mapping,
91 page->index, GFP_KERNEL)) {
David Howells03fb3d22009-04-03 16:42:35 +010092 read_cache_pages_invalidate_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 continue;
94 }
Nick Piggineb2be182007-10-16 01:24:57 -070095 page_cache_release(page);
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 ret = filler(data, page);
Nick Piggineb2be182007-10-16 01:24:57 -070098 if (unlikely(ret)) {
David Howells03fb3d22009-04-03 16:42:35 +010099 read_cache_pages_invalidate_pages(mapping, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 break;
101 }
Andrew Morton8bde37f2006-12-10 02:19:40 -0800102 task_io_account_read(PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 return ret;
105}
106
107EXPORT_SYMBOL(read_cache_pages);
108
109static int read_pages(struct address_space *mapping, struct file *filp,
110 struct list_head *pages, unsigned nr_pages)
111{
Jens Axboe5b417b12010-04-19 10:04:38 +0200112 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 unsigned page_idx;
Zach Brown994fc28c2005-12-15 14:28:17 -0800114 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Jens Axboe5b417b12010-04-19 10:04:38 +0200116 blk_start_plug(&plug);
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 if (mapping->a_ops->readpages) {
119 ret = mapping->a_ops->readpages(filp, mapping, pages, nr_pages);
OGAWA Hirofumi029e3322006-11-02 22:07:06 -0800120 /* Clean up the remaining pages */
121 put_pages_list(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 goto out;
123 }
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
126 struct page *page = list_to_page(pages);
127 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700128 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 page->index, GFP_KERNEL)) {
Zach Brown9f1a3cf2006-06-25 05:46:46 -0700130 mapping->a_ops->readpage(filp, page);
Nick Piggineb2be182007-10-16 01:24:57 -0700131 }
132 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 }
Zach Brown994fc28c2005-12-15 14:28:17 -0800134 ret = 0;
Jens Axboe5b417b12010-04-19 10:04:38 +0200135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136out:
Jens Axboe5b417b12010-04-19 10:04:38 +0200137 blk_finish_plug(&plug);
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return ret;
140}
141
142/*
Wu Fengguangd30a1102009-06-16 15:31:30 -0700143 * __do_page_cache_readahead() actually reads a chunk of disk. It allocates all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 * the pages first, then submits them all for I/O. This avoids the very bad
145 * behaviour which would occur if page allocations are causing VM writeback.
146 * We really don't want to intermingle reads and writes like that.
147 *
148 * Returns the number of pages requested, or the maximum amount of I/O allowed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
150static int
151__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
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700166 end_index = ((isize - 1) >> PAGE_CACHE_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();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 if (page)
181 continue;
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 page = page_cache_alloc_cold(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{
212 int ret = 0;
213
214 if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages))
215 return -EINVAL;
216
Wu Fengguangf7e839d2009-06-16 15:31:20 -0700217 nr_to_read = max_sane_readahead(nr_to_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 while (nr_to_read) {
219 int err;
220
221 unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_CACHE_SIZE;
222
223 if (this_chunk > nr_to_read)
224 this_chunk = nr_to_read;
225 err = __do_page_cache_readahead(mapping, filp,
Fengguang Wu46fc3e72007-07-19 01:47:57 -0700226 offset, this_chunk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 if (err < 0) {
228 ret = err;
229 break;
230 }
231 ret += err;
232 offset += this_chunk;
233 nr_to_read -= this_chunk;
234 }
235 return ret;
236}
237
238/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 * Given a desired number of PAGE_CACHE_SIZE readahead pages, return a
240 * sensible upper limit.
241 */
242unsigned long max_sane_readahead(unsigned long nr)
243{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700244 return min(nr, (node_page_state(numa_node_id(), NR_INACTIVE_FILE)
Christoph Lameter05a04162007-02-10 01:43:05 -0800245 + node_page_state(numa_node_id(), NR_FREE_PAGES)) / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
Fengguang Wu5ce11102007-07-19 01:47:59 -0700247
248/*
249 * Submit IO for the read-ahead request in file_ra_state.
250 */
Wu Fengguangd30a1102009-06-16 15:31:30 -0700251unsigned long ra_submit(struct file_ra_state *ra,
Fengguang Wu5ce11102007-07-19 01:47:59 -0700252 struct address_space *mapping, struct file *filp)
253{
Fengguang Wu5ce11102007-07-19 01:47:59 -0700254 int actual;
255
Fengguang Wu5ce11102007-07-19 01:47:59 -0700256 actual = __do_page_cache_readahead(mapping, filp,
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700257 ra->start, ra->size, ra->async_size);
Fengguang Wu5ce11102007-07-19 01:47:59 -0700258
259 return actual;
260}
Fengguang Wu122a21d2007-07-19 01:48:01 -0700261
262/*
Fengguang Wuc743d962007-07-19 01:48:04 -0700263 * Set the initial window size, round to next power of 2 and square
264 * for small size, x 4 for medium, and x 2 for large
265 * for 128k (32 page) max ra
266 * 1-8 page = 32k initial, > 8 page = 128k initial
267 */
268static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
269{
270 unsigned long newsize = roundup_pow_of_two(size);
271
272 if (newsize <= max / 32)
273 newsize = newsize * 4;
274 else if (newsize <= max / 4)
275 newsize = newsize * 2;
276 else
277 newsize = max;
278
279 return newsize;
280}
281
282/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700283 * Get the previous window size, ramp it up, and
284 * return it as the new window size.
285 */
Fengguang Wuc743d962007-07-19 01:48:04 -0700286static unsigned long get_next_ra_size(struct file_ra_state *ra,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700287 unsigned long max)
288{
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700289 unsigned long cur = ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700290 unsigned long newsize;
291
292 if (cur < max / 16)
Fengguang Wuc743d962007-07-19 01:48:04 -0700293 newsize = 4 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700294 else
Fengguang Wuc743d962007-07-19 01:48:04 -0700295 newsize = 2 * cur;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700296
297 return min(newsize, max);
298}
299
300/*
301 * On-demand readahead design.
302 *
303 * The fields in struct file_ra_state represent the most-recently-executed
304 * readahead attempt:
305 *
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700306 * |<----- async_size ---------|
307 * |------------------- size -------------------->|
308 * |==================#===========================|
309 * ^start ^page marked with PG_readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700310 *
311 * To overlap application thinking time and disk I/O time, we do
312 * `readahead pipelining': Do not wait until the application consumed all
313 * readahead pages and stalled on the missing page at readahead_index;
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700314 * Instead, submit an asynchronous readahead I/O as soon as there are
315 * only async_size pages left in the readahead window. Normally async_size
316 * will be equal to size, for maximum pipelining.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700317 *
318 * In interleaved sequential reads, concurrent streams on the same fd can
319 * be invalidating each other's readahead state. So we flag the new readahead
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700320 * page at (start+size-async_size) with PG_readahead, and use it as readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700321 * indicator. The flag won't be set on already cached pages, to avoid the
322 * readahead-for-nothing fuss, saving pointless page cache lookups.
323 *
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700324 * prev_pos tracks the last visited byte in the _previous_ read request.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700325 * It should be maintained by the caller, and will be used for detecting
326 * small random reads. Note that the readahead algorithm checks loosely
327 * for sequential patterns. Hence interleaved reads might be served as
328 * sequential ones.
329 *
330 * There is a special-case: if the first page which the application tries to
331 * read happens to be the first page of the file, it is assumed that a linear
332 * read is about to happen and the window is immediately set to the initial size
333 * based on I/O request size and the max_readahead.
334 *
335 * The code ramps up the readahead size aggressively at first, but slow down as
336 * it approaches max_readhead.
337 */
338
339/*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700340 * Count contiguously cached pages from @offset-1 to @offset-@max,
341 * this count is a conservative estimation of
342 * - length of the sequential read sequence, or
343 * - thrashing threshold in memory tight systems
344 */
345static pgoff_t count_history_pages(struct address_space *mapping,
346 struct file_ra_state *ra,
347 pgoff_t offset, unsigned long max)
348{
349 pgoff_t head;
350
351 rcu_read_lock();
352 head = radix_tree_prev_hole(&mapping->page_tree, offset - 1, max);
353 rcu_read_unlock();
354
355 return offset - 1 - head;
356}
357
358/*
359 * page cache context based read-ahead
360 */
361static int try_context_readahead(struct address_space *mapping,
362 struct file_ra_state *ra,
363 pgoff_t offset,
364 unsigned long req_size,
365 unsigned long max)
366{
367 pgoff_t size;
368
369 size = count_history_pages(mapping, ra, offset, max);
370
371 /*
372 * no history pages:
373 * it could be a random read
374 */
375 if (!size)
376 return 0;
377
378 /*
379 * starts from beginning of file:
380 * it is a strong indication of long-run stream (or whole-file-read)
381 */
382 if (size >= offset)
383 size *= 2;
384
385 ra->start = offset;
386 ra->size = get_init_ra_size(size + req_size, max);
387 ra->async_size = ra->size;
388
389 return 1;
390}
391
392/*
Fengguang Wu122a21d2007-07-19 01:48:01 -0700393 * A minimal readahead algorithm for trivial sequential/random reads.
394 */
395static unsigned long
396ondemand_readahead(struct address_space *mapping,
397 struct file_ra_state *ra, struct file *filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700398 bool hit_readahead_marker, pgoff_t offset,
Fengguang Wu122a21d2007-07-19 01:48:01 -0700399 unsigned long req_size)
400{
Wu Fengguangfc31d162009-06-16 15:31:21 -0700401 unsigned long max = max_sane_readahead(ra->ra_pages);
Wu Fengguang045a2522009-06-16 15:31:33 -0700402
403 /*
404 * start of file
405 */
406 if (!offset)
407 goto initial_readahead;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700408
409 /*
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700410 * It's the expected callback offset, assume sequential access.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700411 * Ramp up sizes, and push forward the readahead window.
412 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700413 if ((offset == (ra->start + ra->size - ra->async_size) ||
414 offset == (ra->start + ra->size))) {
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700415 ra->start += ra->size;
416 ra->size = get_next_ra_size(ra, max);
417 ra->async_size = ra->size;
418 goto readit;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700419 }
420
Fengguang Wu122a21d2007-07-19 01:48:01 -0700421 /*
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700422 * Hit a marked page without valid readahead state.
423 * E.g. interleaved reads.
424 * Query the pagecache for async_size, which normally equals to
425 * readahead size. Ramp it up and use it as the new readahead size.
426 */
427 if (hit_readahead_marker) {
428 pgoff_t start;
429
Nick Piggin30002ed2008-07-25 19:45:28 -0700430 rcu_read_lock();
Wu Fengguangcaca7cb2009-06-16 15:31:21 -0700431 start = radix_tree_next_hole(&mapping->page_tree, offset+1,max);
Nick Piggin30002ed2008-07-25 19:45:28 -0700432 rcu_read_unlock();
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700433
434 if (!start || start - offset > max)
435 return 0;
436
437 ra->start = start;
438 ra->size = start - offset; /* old async_size */
Wu Fengguang160334a2009-06-16 15:31:23 -0700439 ra->size += req_size;
Fengguang Wu6b10c6c2007-10-16 01:24:34 -0700440 ra->size = get_next_ra_size(ra, max);
441 ra->async_size = ra->size;
442 goto readit;
443 }
444
445 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700446 * oversize read
Fengguang Wu122a21d2007-07-19 01:48:01 -0700447 */
Wu Fengguang045a2522009-06-16 15:31:33 -0700448 if (req_size > max)
449 goto initial_readahead;
450
451 /*
452 * sequential cache miss
453 */
454 if (offset - (ra->prev_pos >> PAGE_CACHE_SHIFT) <= 1UL)
455 goto initial_readahead;
456
457 /*
Wu Fengguang10be0b32009-06-16 15:31:36 -0700458 * Query the page cache and look for the traces(cached history pages)
459 * that a sequential stream would leave behind.
460 */
461 if (try_context_readahead(mapping, ra, offset, req_size, max))
462 goto readit;
463
464 /*
Wu Fengguang045a2522009-06-16 15:31:33 -0700465 * standalone, small random read
466 * Read as is, and do not pollute the readahead state.
467 */
468 return __do_page_cache_readahead(mapping, filp, offset, req_size, 0);
469
470initial_readahead:
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700471 ra->start = offset;
472 ra->size = get_init_ra_size(req_size, max);
473 ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700474
Fengguang Wuf9acc8c2007-07-19 01:48:08 -0700475readit:
Wu Fengguang51daa882009-06-16 15:31:24 -0700476 /*
477 * Will this read hit the readahead marker made by itself?
478 * If so, trigger the readahead marker hit now, and merge
479 * the resulted next readahead window into the current one.
480 */
481 if (offset == ra->start && ra->size == ra->async_size) {
482 ra->async_size = get_next_ra_size(ra, max);
483 ra->size += ra->async_size;
484 }
485
Fengguang Wu122a21d2007-07-19 01:48:01 -0700486 return ra_submit(ra, mapping, filp);
487}
488
489/**
Rusty Russellcf914a72007-07-19 01:48:08 -0700490 * page_cache_sync_readahead - generic file readahead
Fengguang Wu122a21d2007-07-19 01:48:01 -0700491 * @mapping: address_space which holds the pagecache and I/O vectors
492 * @ra: file_ra_state which holds the readahead state
493 * @filp: passed on to ->readpage() and ->readpages()
Rusty Russellcf914a72007-07-19 01:48:08 -0700494 * @offset: start offset into @mapping, in pagecache page-sized units
Fengguang Wu122a21d2007-07-19 01:48:01 -0700495 * @req_size: hint: total size of the read which the caller is performing in
Rusty Russellcf914a72007-07-19 01:48:08 -0700496 * pagecache pages
Fengguang Wu122a21d2007-07-19 01:48:01 -0700497 *
Rusty Russellcf914a72007-07-19 01:48:08 -0700498 * page_cache_sync_readahead() should be called when a cache miss happened:
499 * it will submit the read. The readahead logic may decide to piggyback more
500 * pages onto the read request if access patterns suggest it will improve
501 * performance.
Fengguang Wu122a21d2007-07-19 01:48:01 -0700502 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700503void page_cache_sync_readahead(struct address_space *mapping,
504 struct file_ra_state *ra, struct file *filp,
505 pgoff_t offset, unsigned long req_size)
Fengguang Wu122a21d2007-07-19 01:48:01 -0700506{
507 /* no read-ahead */
508 if (!ra->ra_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700509 return;
Fengguang Wu122a21d2007-07-19 01:48:01 -0700510
Wu Fengguang01414502010-03-05 13:42:03 -0800511 /* be dumb */
Wu Fengguang70655c02010-04-06 14:34:53 -0700512 if (filp && (filp->f_mode & FMODE_RANDOM)) {
Wu Fengguang01414502010-03-05 13:42:03 -0800513 force_page_cache_readahead(mapping, filp, offset, req_size);
514 return;
515 }
516
Fengguang Wu122a21d2007-07-19 01:48:01 -0700517 /* do read-ahead */
Rusty Russellcf914a72007-07-19 01:48:08 -0700518 ondemand_readahead(mapping, ra, filp, false, offset, req_size);
Fengguang Wu122a21d2007-07-19 01:48:01 -0700519}
Rusty Russellcf914a72007-07-19 01:48:08 -0700520EXPORT_SYMBOL_GPL(page_cache_sync_readahead);
521
522/**
523 * page_cache_async_readahead - file readahead for marked pages
524 * @mapping: address_space which holds the pagecache and I/O vectors
525 * @ra: file_ra_state which holds the readahead state
526 * @filp: passed on to ->readpage() and ->readpages()
527 * @page: the page at @offset which has the PG_readahead flag set
528 * @offset: start offset into @mapping, in pagecache page-sized units
529 * @req_size: hint: total size of the read which the caller is performing in
530 * pagecache pages
531 *
Huang Shijiebf8abe82010-05-24 14:32:36 -0700532 * page_cache_async_readahead() should be called when a page is used which
Randy Dunlapf7850d92008-03-19 17:01:02 -0700533 * has the PG_readahead flag; this is a marker to suggest that the application
Rusty Russellcf914a72007-07-19 01:48:08 -0700534 * has used up enough of the readahead window that we should start pulling in
Randy Dunlapf7850d92008-03-19 17:01:02 -0700535 * more pages.
536 */
Rusty Russellcf914a72007-07-19 01:48:08 -0700537void
538page_cache_async_readahead(struct address_space *mapping,
539 struct file_ra_state *ra, struct file *filp,
540 struct page *page, pgoff_t offset,
541 unsigned long req_size)
542{
543 /* no read-ahead */
544 if (!ra->ra_pages)
545 return;
546
547 /*
548 * Same bit is used for PG_readahead and PG_reclaim.
549 */
550 if (PageWriteback(page))
551 return;
552
553 ClearPageReadahead(page);
554
555 /*
556 * Defer asynchronous read-ahead on IO congestion.
557 */
558 if (bdi_read_congested(mapping->backing_dev_info))
559 return;
560
561 /* do read-ahead */
562 ondemand_readahead(mapping, ra, filp, true, offset, req_size);
563}
564EXPORT_SYMBOL_GPL(page_cache_async_readahead);