2 * mm/readahead.c - address_space-level file readahead.
4 * Copyright (C) 2002, Linus Torvalds
6 * 09Apr2002 akpm@zip.com.au
10 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/blkdev.h>
15 #include <linux/backing-dev.h>
16 #include <linux/task_io_accounting_ops.h>
17 #include <linux/pagevec.h>
19 void default_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
22 EXPORT_SYMBOL(default_unplug_io_fn
);
25 * Convienent macros for min/max read-ahead pages.
26 * Note that MAX_RA_PAGES is rounded down, while MIN_RA_PAGES is rounded up.
27 * The latter is necessary for systems with large page size(i.e. 64k).
29 #define MAX_RA_PAGES (VM_MAX_READAHEAD*1024 / PAGE_CACHE_SIZE)
30 #define MIN_RA_PAGES DIV_ROUND_UP(VM_MIN_READAHEAD*1024, PAGE_CACHE_SIZE)
32 struct backing_dev_info default_backing_dev_info
= {
33 .ra_pages
= MAX_RA_PAGES
,
35 .capabilities
= BDI_CAP_MAP_COPY
,
36 .unplug_io_fn
= default_unplug_io_fn
,
38 EXPORT_SYMBOL_GPL(default_backing_dev_info
);
41 * Initialise a struct file's readahead state. Assumes that the caller has
45 file_ra_state_init(struct file_ra_state
*ra
, struct address_space
*mapping
)
47 ra
->ra_pages
= mapping
->backing_dev_info
->ra_pages
;
50 EXPORT_SYMBOL_GPL(file_ra_state_init
);
52 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
55 * read_cache_pages - populate an address space with some pages & start reads against them
56 * @mapping: the address_space
57 * @pages: The address of a list_head which contains the target pages. These
58 * pages have their ->index populated and are otherwise uninitialised.
59 * @filler: callback routine for filling a single page.
60 * @data: private data for the callback routine.
62 * Hides the details of the LRU cache etc from the filesystems.
64 int read_cache_pages(struct address_space
*mapping
, struct list_head
*pages
,
65 int (*filler
)(void *, struct page
*), void *data
)
68 struct pagevec lru_pvec
;
71 pagevec_init(&lru_pvec
, 0);
73 while (!list_empty(pages
)) {
74 page
= list_to_page(pages
);
76 if (add_to_page_cache(page
, mapping
, page
->index
, GFP_KERNEL
)) {
77 page_cache_release(page
);
80 ret
= filler(data
, page
);
81 if (!pagevec_add(&lru_pvec
, page
))
82 __pagevec_lru_add(&lru_pvec
);
84 put_pages_list(pages
);
87 task_io_account_read(PAGE_CACHE_SIZE
);
89 pagevec_lru_add(&lru_pvec
);
93 EXPORT_SYMBOL(read_cache_pages
);
95 static int read_pages(struct address_space
*mapping
, struct file
*filp
,
96 struct list_head
*pages
, unsigned nr_pages
)
99 struct pagevec lru_pvec
;
102 if (mapping
->a_ops
->readpages
) {
103 ret
= mapping
->a_ops
->readpages(filp
, mapping
, pages
, nr_pages
);
104 /* Clean up the remaining pages */
105 put_pages_list(pages
);
109 pagevec_init(&lru_pvec
, 0);
110 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
111 struct page
*page
= list_to_page(pages
);
112 list_del(&page
->lru
);
113 if (!add_to_page_cache(page
, mapping
,
114 page
->index
, GFP_KERNEL
)) {
115 mapping
->a_ops
->readpage(filp
, page
);
116 if (!pagevec_add(&lru_pvec
, page
))
117 __pagevec_lru_add(&lru_pvec
);
119 page_cache_release(page
);
121 pagevec_lru_add(&lru_pvec
);
128 * do_page_cache_readahead actually reads a chunk of disk. It allocates all
129 * the pages first, then submits them all for I/O. This avoids the very bad
130 * behaviour which would occur if page allocations are causing VM writeback.
131 * We really don't want to intermingle reads and writes like that.
133 * Returns the number of pages requested, or the maximum amount of I/O allowed.
135 * do_page_cache_readahead() returns -1 if it encountered request queue
139 __do_page_cache_readahead(struct address_space
*mapping
, struct file
*filp
,
140 pgoff_t offset
, unsigned long nr_to_read
,
141 unsigned long lookahead_size
)
143 struct inode
*inode
= mapping
->host
;
145 unsigned long end_index
; /* The last page we want to read */
146 LIST_HEAD(page_pool
);
149 loff_t isize
= i_size_read(inode
);
154 end_index
= ((isize
- 1) >> PAGE_CACHE_SHIFT
);
157 * Preallocate as many pages as we will need.
159 read_lock_irq(&mapping
->tree_lock
);
160 for (page_idx
= 0; page_idx
< nr_to_read
; page_idx
++) {
161 pgoff_t page_offset
= offset
+ page_idx
;
163 if (page_offset
> end_index
)
166 page
= radix_tree_lookup(&mapping
->page_tree
, page_offset
);
170 read_unlock_irq(&mapping
->tree_lock
);
171 page
= page_cache_alloc_cold(mapping
);
172 read_lock_irq(&mapping
->tree_lock
);
175 page
->index
= page_offset
;
176 list_add(&page
->lru
, &page_pool
);
177 if (page_idx
== nr_to_read
- lookahead_size
)
178 SetPageReadahead(page
);
181 read_unlock_irq(&mapping
->tree_lock
);
184 * Now start the IO. We ignore I/O errors - if the page is not
185 * uptodate then the caller will launch readpage again, and
186 * will then handle the error.
189 read_pages(mapping
, filp
, &page_pool
, ret
);
190 BUG_ON(!list_empty(&page_pool
));
196 * Chunk the readahead into 2 megabyte units, so that we don't pin too much
199 int force_page_cache_readahead(struct address_space
*mapping
, struct file
*filp
,
200 pgoff_t offset
, unsigned long nr_to_read
)
204 if (unlikely(!mapping
->a_ops
->readpage
&& !mapping
->a_ops
->readpages
))
210 unsigned long this_chunk
= (2 * 1024 * 1024) / PAGE_CACHE_SIZE
;
212 if (this_chunk
> nr_to_read
)
213 this_chunk
= nr_to_read
;
214 err
= __do_page_cache_readahead(mapping
, filp
,
215 offset
, this_chunk
, 0);
221 offset
+= this_chunk
;
222 nr_to_read
-= this_chunk
;
228 * This version skips the IO if the queue is read-congested, and will tell the
229 * block layer to abandon the readahead if request allocation would block.
231 * force_page_cache_readahead() will ignore queue congestion and will block on
234 int do_page_cache_readahead(struct address_space
*mapping
, struct file
*filp
,
235 pgoff_t offset
, unsigned long nr_to_read
)
237 if (bdi_read_congested(mapping
->backing_dev_info
))
240 return __do_page_cache_readahead(mapping
, filp
, offset
, nr_to_read
, 0);
244 * Given a desired number of PAGE_CACHE_SIZE readahead pages, return a
245 * sensible upper limit.
247 unsigned long max_sane_readahead(unsigned long nr
)
249 return min(nr
, (node_page_state(numa_node_id(), NR_INACTIVE
)
250 + node_page_state(numa_node_id(), NR_FREE_PAGES
)) / 2);
254 * Submit IO for the read-ahead request in file_ra_state.
256 unsigned long ra_submit(struct file_ra_state
*ra
,
257 struct address_space
*mapping
, struct file
*filp
)
259 unsigned long ra_size
;
260 unsigned long la_size
;
263 ra_size
= ra_readahead_size(ra
);
264 la_size
= ra_lookahead_size(ra
);
265 actual
= __do_page_cache_readahead(mapping
, filp
,
266 ra
->ra_index
, ra_size
, la_size
);
270 EXPORT_SYMBOL_GPL(ra_submit
);
273 * Set the initial window size, round to next power of 2 and square
274 * for small size, x 4 for medium, and x 2 for large
275 * for 128k (32 page) max ra
276 * 1-8 page = 32k initial, > 8 page = 128k initial
278 static unsigned long get_init_ra_size(unsigned long size
, unsigned long max
)
280 unsigned long newsize
= roundup_pow_of_two(size
);
282 if (newsize
<= max
/ 32)
283 newsize
= newsize
* 4;
284 else if (newsize
<= max
/ 4)
285 newsize
= newsize
* 2;
293 * Get the previous window size, ramp it up, and
294 * return it as the new window size.
296 static unsigned long get_next_ra_size(struct file_ra_state
*ra
,
299 unsigned long cur
= ra
->readahead_index
- ra
->ra_index
;
300 unsigned long newsize
;
307 return min(newsize
, max
);
311 * On-demand readahead design.
313 * The fields in struct file_ra_state represent the most-recently-executed
316 * |-------- last readahead window -------->|
317 * |-- application walking here -->|
318 * ======#============|==================#=====================|
319 * ^la_index ^ra_index ^lookahead_index ^readahead_index
321 * [ra_index, readahead_index) represents the last readahead window.
323 * [la_index, lookahead_index] is where the application would be walking(in
324 * the common case of cache-cold sequential reads): the last window was
325 * established when the application was at la_index, and the next window will
326 * be bring in when the application reaches lookahead_index.
328 * To overlap application thinking time and disk I/O time, we do
329 * `readahead pipelining': Do not wait until the application consumed all
330 * readahead pages and stalled on the missing page at readahead_index;
331 * Instead, submit an asynchronous readahead I/O as early as the application
332 * reads on the page at lookahead_index. Normally lookahead_index will be
333 * equal to ra_index, for maximum pipelining.
335 * In interleaved sequential reads, concurrent streams on the same fd can
336 * be invalidating each other's readahead state. So we flag the new readahead
337 * page at lookahead_index with PG_readahead, and use it as readahead
338 * indicator. The flag won't be set on already cached pages, to avoid the
339 * readahead-for-nothing fuss, saving pointless page cache lookups.
341 * prev_index tracks the last visited page in the _previous_ read request.
342 * It should be maintained by the caller, and will be used for detecting
343 * small random reads. Note that the readahead algorithm checks loosely
344 * for sequential patterns. Hence interleaved reads might be served as
347 * There is a special-case: if the first page which the application tries to
348 * read happens to be the first page of the file, it is assumed that a linear
349 * read is about to happen and the window is immediately set to the initial size
350 * based on I/O request size and the max_readahead.
352 * The code ramps up the readahead size aggressively at first, but slow down as
353 * it approaches max_readhead.
357 * A minimal readahead algorithm for trivial sequential/random reads.
360 ondemand_readahead(struct address_space
*mapping
,
361 struct file_ra_state
*ra
, struct file
*filp
,
362 struct page
*page
, pgoff_t offset
,
363 unsigned long req_size
)
365 unsigned long max
; /* max readahead pages */
366 pgoff_t ra_index
; /* readahead index */
367 unsigned long ra_size
; /* readahead size */
368 unsigned long la_size
; /* lookahead size */
372 sequential
= (offset
- ra
->prev_index
<= 1UL) || (req_size
> max
);
375 * Lookahead/readahead hit, assume sequential access.
376 * Ramp up sizes, and push forward the readahead window.
378 if (offset
&& (offset
== ra
->lookahead_index
||
379 offset
== ra
->readahead_index
)) {
380 ra_index
= ra
->readahead_index
;
381 ra_size
= get_next_ra_size(ra
, max
);
387 * Standalone, small read.
388 * Read as is, and do not pollute the readahead state.
390 if (!page
&& !sequential
) {
391 return __do_page_cache_readahead(mapping
, filp
,
392 offset
, req_size
, 0);
397 * - first read on start of file
398 * - sequential cache miss
399 * - oversize random read
400 * Start readahead for it.
403 ra_size
= get_init_ra_size(req_size
, max
);
404 la_size
= ra_size
> req_size
? ra_size
- req_size
: ra_size
;
407 * Hit on a lookahead page without valid readahead state.
408 * E.g. interleaved reads.
409 * Not knowing its readahead pos/size, bet on the minimal possible one.
413 ra_size
= min(4 * ra_size
, max
);
417 ra_set_index(ra
, offset
, ra_index
);
418 ra_set_size(ra
, ra_size
, la_size
);
420 return ra_submit(ra
, mapping
, filp
);
424 * page_cache_readahead_ondemand - generic file readahead
425 * @mapping: address_space which holds the pagecache and I/O vectors
426 * @ra: file_ra_state which holds the readahead state
427 * @filp: passed on to ->readpage() and ->readpages()
428 * @page: the page at @offset, or NULL if non-present
429 * @offset: start offset into @mapping, in PAGE_CACHE_SIZE units
430 * @req_size: hint: total size of the read which the caller is performing in
431 * PAGE_CACHE_SIZE units
433 * page_cache_readahead_ondemand() is the entry point of readahead logic.
434 * This function should be called when it is time to perform readahead:
436 * A cache miss happened, time for synchronous readahead.
437 * 2) @page != NULL && PageReadahead(@page)
438 * A look-ahead hit occured, time for asynchronous readahead.
441 page_cache_readahead_ondemand(struct address_space
*mapping
,
442 struct file_ra_state
*ra
, struct file
*filp
,
443 struct page
*page
, pgoff_t offset
,
444 unsigned long req_size
)
452 * It can be PG_reclaim.
454 if (PageWriteback(page
))
457 ClearPageReadahead(page
);
460 * Defer asynchronous read-ahead on IO congestion.
462 if (bdi_read_congested(mapping
->backing_dev_info
))
467 return ondemand_readahead(mapping
, ra
, filp
, page
,
470 EXPORT_SYMBOL_GPL(page_cache_readahead_ondemand
);