]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/readahead.c - address_space-level file readahead. | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
6 | * 09Apr2002 akpm@zip.com.au | |
7 | * Initial version. | |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
11 | #include <linux/fs.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/blkdev.h> | |
15 | #include <linux/backing-dev.h> | |
8bde37f0 | 16 | #include <linux/task_io_accounting_ops.h> |
1da177e4 LT |
17 | #include <linux/pagevec.h> |
18 | ||
19 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) | |
20 | { | |
21 | } | |
22 | EXPORT_SYMBOL(default_unplug_io_fn); | |
23 | ||
f615bfca FW |
24 | /* |
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). | |
28 | */ | |
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) | |
31 | ||
1da177e4 | 32 | struct backing_dev_info default_backing_dev_info = { |
f615bfca | 33 | .ra_pages = MAX_RA_PAGES, |
1da177e4 LT |
34 | .state = 0, |
35 | .capabilities = BDI_CAP_MAP_COPY, | |
36 | .unplug_io_fn = default_unplug_io_fn, | |
37 | }; | |
38 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
39 | ||
40 | /* | |
41 | * Initialise a struct file's readahead state. Assumes that the caller has | |
42 | * memset *ra to zero. | |
43 | */ | |
44 | void | |
45 | file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping) | |
46 | { | |
47 | ra->ra_pages = mapping->backing_dev_info->ra_pages; | |
6ce745ed | 48 | ra->prev_index = -1; |
1da177e4 | 49 | } |
d41cc702 | 50 | EXPORT_SYMBOL_GPL(file_ra_state_init); |
1da177e4 LT |
51 | |
52 | /* | |
53 | * Return max readahead size for this inode in number-of-pages. | |
54 | */ | |
55 | static inline unsigned long get_max_readahead(struct file_ra_state *ra) | |
56 | { | |
57 | return ra->ra_pages; | |
58 | } | |
59 | ||
60 | static inline unsigned long get_min_readahead(struct file_ra_state *ra) | |
61 | { | |
f615bfca | 62 | return MIN_RA_PAGES; |
1da177e4 LT |
63 | } |
64 | ||
a564da39 ON |
65 | static inline void reset_ahead_window(struct file_ra_state *ra) |
66 | { | |
67 | /* | |
68 | * ... but preserve ahead_start + ahead_size value, | |
69 | * see 'recheck:' label in page_cache_readahead(). | |
70 | * Note: We never use ->ahead_size as rvalue without | |
71 | * checking ->ahead_start != 0 first. | |
72 | */ | |
73 | ra->ahead_size += ra->ahead_start; | |
74 | ra->ahead_start = 0; | |
75 | } | |
76 | ||
1da177e4 LT |
77 | static inline void ra_off(struct file_ra_state *ra) |
78 | { | |
79 | ra->start = 0; | |
80 | ra->flags = 0; | |
81 | ra->size = 0; | |
a564da39 | 82 | reset_ahead_window(ra); |
1da177e4 LT |
83 | return; |
84 | } | |
85 | ||
86 | /* | |
87 | * Set the initial window size, round to next power of 2 and square | |
88 | * for small size, x 4 for medium, and x 2 for large | |
89 | * for 128k (32 page) max ra | |
90 | * 1-8 page = 32k initial, > 8 page = 128k initial | |
91 | */ | |
92 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) | |
93 | { | |
94 | unsigned long newsize = roundup_pow_of_two(size); | |
95 | ||
aed75ff3 SP |
96 | if (newsize <= max / 32) |
97 | newsize = newsize * 4; | |
1da177e4 | 98 | else if (newsize <= max / 4) |
aed75ff3 | 99 | newsize = newsize * 2; |
1da177e4 LT |
100 | else |
101 | newsize = max; | |
102 | return newsize; | |
103 | } | |
104 | ||
105 | /* | |
106 | * Set the new window size, this is called only when I/O is to be submitted, | |
107 | * not for each call to readahead. If a cache miss occured, reduce next I/O | |
108 | * size, else increase depending on how close to max we are. | |
109 | */ | |
110 | static inline unsigned long get_next_ra_size(struct file_ra_state *ra) | |
111 | { | |
112 | unsigned long max = get_max_readahead(ra); | |
113 | unsigned long min = get_min_readahead(ra); | |
114 | unsigned long cur = ra->size; | |
115 | unsigned long newsize; | |
116 | ||
117 | if (ra->flags & RA_FLAG_MISS) { | |
118 | ra->flags &= ~RA_FLAG_MISS; | |
119 | newsize = max((cur - 2), min); | |
120 | } else if (cur < max / 16) { | |
121 | newsize = 4 * cur; | |
122 | } else { | |
123 | newsize = 2 * cur; | |
124 | } | |
125 | return min(newsize, max); | |
126 | } | |
127 | ||
128 | #define list_to_page(head) (list_entry((head)->prev, struct page, lru)) | |
129 | ||
130 | /** | |
bd40cdda | 131 | * read_cache_pages - populate an address space with some pages & start reads against them |
1da177e4 LT |
132 | * @mapping: the address_space |
133 | * @pages: The address of a list_head which contains the target pages. These | |
134 | * pages have their ->index populated and are otherwise uninitialised. | |
135 | * @filler: callback routine for filling a single page. | |
136 | * @data: private data for the callback routine. | |
137 | * | |
138 | * Hides the details of the LRU cache etc from the filesystems. | |
139 | */ | |
140 | int read_cache_pages(struct address_space *mapping, struct list_head *pages, | |
141 | int (*filler)(void *, struct page *), void *data) | |
142 | { | |
143 | struct page *page; | |
144 | struct pagevec lru_pvec; | |
145 | int ret = 0; | |
146 | ||
147 | pagevec_init(&lru_pvec, 0); | |
148 | ||
149 | while (!list_empty(pages)) { | |
150 | page = list_to_page(pages); | |
151 | list_del(&page->lru); | |
152 | if (add_to_page_cache(page, mapping, page->index, GFP_KERNEL)) { | |
153 | page_cache_release(page); | |
154 | continue; | |
155 | } | |
156 | ret = filler(data, page); | |
157 | if (!pagevec_add(&lru_pvec, page)) | |
158 | __pagevec_lru_add(&lru_pvec); | |
159 | if (ret) { | |
38da288b | 160 | put_pages_list(pages); |
1da177e4 LT |
161 | break; |
162 | } | |
8bde37f0 | 163 | task_io_account_read(PAGE_CACHE_SIZE); |
1da177e4 LT |
164 | } |
165 | pagevec_lru_add(&lru_pvec); | |
166 | return ret; | |
167 | } | |
168 | ||
169 | EXPORT_SYMBOL(read_cache_pages); | |
170 | ||
171 | static int read_pages(struct address_space *mapping, struct file *filp, | |
172 | struct list_head *pages, unsigned nr_pages) | |
173 | { | |
174 | unsigned page_idx; | |
175 | struct pagevec lru_pvec; | |
994fc28c | 176 | int ret; |
1da177e4 LT |
177 | |
178 | if (mapping->a_ops->readpages) { | |
179 | ret = mapping->a_ops->readpages(filp, mapping, pages, nr_pages); | |
029e332e OH |
180 | /* Clean up the remaining pages */ |
181 | put_pages_list(pages); | |
1da177e4 LT |
182 | goto out; |
183 | } | |
184 | ||
185 | pagevec_init(&lru_pvec, 0); | |
186 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
187 | struct page *page = list_to_page(pages); | |
188 | list_del(&page->lru); | |
189 | if (!add_to_page_cache(page, mapping, | |
190 | page->index, GFP_KERNEL)) { | |
9f1a3cfc ZB |
191 | mapping->a_ops->readpage(filp, page); |
192 | if (!pagevec_add(&lru_pvec, page)) | |
193 | __pagevec_lru_add(&lru_pvec); | |
194 | } else | |
195 | page_cache_release(page); | |
1da177e4 LT |
196 | } |
197 | pagevec_lru_add(&lru_pvec); | |
994fc28c | 198 | ret = 0; |
1da177e4 LT |
199 | out: |
200 | return ret; | |
201 | } | |
202 | ||
203 | /* | |
204 | * Readahead design. | |
205 | * | |
206 | * The fields in struct file_ra_state represent the most-recently-executed | |
207 | * readahead attempt: | |
208 | * | |
209 | * start: Page index at which we started the readahead | |
210 | * size: Number of pages in that read | |
211 | * Together, these form the "current window". | |
212 | * Together, start and size represent the `readahead window'. | |
6ce745ed | 213 | * prev_index: The page which the readahead algorithm most-recently inspected. |
1da177e4 LT |
214 | * It is mainly used to detect sequential file reading. |
215 | * If page_cache_readahead sees that it is again being called for | |
216 | * a page which it just looked at, it can return immediately without | |
217 | * making any state changes. | |
6ce745ed | 218 | * offset: Offset in the prev_index where the last read ended - used for |
ec0f1637 | 219 | * detection of sequential file reading. |
1da177e4 LT |
220 | * ahead_start, |
221 | * ahead_size: Together, these form the "ahead window". | |
222 | * ra_pages: The externally controlled max readahead for this fd. | |
223 | * | |
224 | * When readahead is in the off state (size == 0), readahead is disabled. | |
6ce745ed | 225 | * In this state, prev_index is used to detect the resumption of sequential I/O. |
1da177e4 LT |
226 | * |
227 | * The readahead code manages two windows - the "current" and the "ahead" | |
228 | * windows. The intent is that while the application is walking the pages | |
229 | * in the current window, I/O is underway on the ahead window. When the | |
230 | * current window is fully traversed, it is replaced by the ahead window | |
231 | * and the ahead window is invalidated. When this copying happens, the | |
232 | * new current window's pages are probably still locked. So | |
233 | * we submit a new batch of I/O immediately, creating a new ahead window. | |
234 | * | |
235 | * So: | |
236 | * | |
237 | * ----|----------------|----------------|----- | |
238 | * ^start ^start+size | |
239 | * ^ahead_start ^ahead_start+ahead_size | |
240 | * | |
241 | * ^ When this page is read, we submit I/O for the | |
242 | * ahead window. | |
243 | * | |
244 | * A `readahead hit' occurs when a read request is made against a page which is | |
245 | * the next sequential page. Ahead window calculations are done only when it | |
246 | * is time to submit a new IO. The code ramps up the size agressively at first, | |
247 | * but slow down as it approaches max_readhead. | |
248 | * | |
249 | * Any seek/ramdom IO will result in readahead being turned off. It will resume | |
250 | * at the first sequential access. | |
251 | * | |
252 | * There is a special-case: if the first page which the application tries to | |
253 | * read happens to be the first page of the file, it is assumed that a linear | |
254 | * read is about to happen and the window is immediately set to the initial size | |
255 | * based on I/O request size and the max_readahead. | |
256 | * | |
257 | * This function is to be called for every read request, rather than when | |
258 | * it is time to perform readahead. It is called only once for the entire I/O | |
259 | * regardless of size unless readahead is unable to start enough I/O to satisfy | |
260 | * the request (I/O request > max_readahead). | |
261 | */ | |
262 | ||
263 | /* | |
264 | * do_page_cache_readahead actually reads a chunk of disk. It allocates all | |
265 | * the pages first, then submits them all for I/O. This avoids the very bad | |
266 | * behaviour which would occur if page allocations are causing VM writeback. | |
267 | * We really don't want to intermingle reads and writes like that. | |
268 | * | |
269 | * Returns the number of pages requested, or the maximum amount of I/O allowed. | |
270 | * | |
271 | * do_page_cache_readahead() returns -1 if it encountered request queue | |
272 | * congestion. | |
273 | */ | |
274 | static int | |
275 | __do_page_cache_readahead(struct address_space *mapping, struct file *filp, | |
46fc3e7b FW |
276 | pgoff_t offset, unsigned long nr_to_read, |
277 | unsigned long lookahead_size) | |
1da177e4 LT |
278 | { |
279 | struct inode *inode = mapping->host; | |
280 | struct page *page; | |
281 | unsigned long end_index; /* The last page we want to read */ | |
282 | LIST_HEAD(page_pool); | |
283 | int page_idx; | |
284 | int ret = 0; | |
285 | loff_t isize = i_size_read(inode); | |
286 | ||
287 | if (isize == 0) | |
288 | goto out; | |
289 | ||
46fc3e7b | 290 | end_index = ((isize - 1) >> PAGE_CACHE_SHIFT); |
1da177e4 LT |
291 | |
292 | /* | |
293 | * Preallocate as many pages as we will need. | |
294 | */ | |
295 | read_lock_irq(&mapping->tree_lock); | |
296 | for (page_idx = 0; page_idx < nr_to_read; page_idx++) { | |
7361f4d8 | 297 | pgoff_t page_offset = offset + page_idx; |
1da177e4 LT |
298 | |
299 | if (page_offset > end_index) | |
300 | break; | |
301 | ||
302 | page = radix_tree_lookup(&mapping->page_tree, page_offset); | |
303 | if (page) | |
304 | continue; | |
305 | ||
306 | read_unlock_irq(&mapping->tree_lock); | |
307 | page = page_cache_alloc_cold(mapping); | |
308 | read_lock_irq(&mapping->tree_lock); | |
309 | if (!page) | |
310 | break; | |
311 | page->index = page_offset; | |
312 | list_add(&page->lru, &page_pool); | |
46fc3e7b FW |
313 | if (page_idx == nr_to_read - lookahead_size) |
314 | SetPageReadahead(page); | |
1da177e4 LT |
315 | ret++; |
316 | } | |
317 | read_unlock_irq(&mapping->tree_lock); | |
318 | ||
319 | /* | |
320 | * Now start the IO. We ignore I/O errors - if the page is not | |
321 | * uptodate then the caller will launch readpage again, and | |
322 | * will then handle the error. | |
323 | */ | |
324 | if (ret) | |
325 | read_pages(mapping, filp, &page_pool, ret); | |
326 | BUG_ON(!list_empty(&page_pool)); | |
327 | out: | |
328 | return ret; | |
329 | } | |
330 | ||
331 | /* | |
332 | * Chunk the readahead into 2 megabyte units, so that we don't pin too much | |
333 | * memory at once. | |
334 | */ | |
335 | int force_page_cache_readahead(struct address_space *mapping, struct file *filp, | |
7361f4d8 | 336 | pgoff_t offset, unsigned long nr_to_read) |
1da177e4 LT |
337 | { |
338 | int ret = 0; | |
339 | ||
340 | if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages)) | |
341 | return -EINVAL; | |
342 | ||
343 | while (nr_to_read) { | |
344 | int err; | |
345 | ||
346 | unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_CACHE_SIZE; | |
347 | ||
348 | if (this_chunk > nr_to_read) | |
349 | this_chunk = nr_to_read; | |
350 | err = __do_page_cache_readahead(mapping, filp, | |
46fc3e7b | 351 | offset, this_chunk, 0); |
1da177e4 LT |
352 | if (err < 0) { |
353 | ret = err; | |
354 | break; | |
355 | } | |
356 | ret += err; | |
357 | offset += this_chunk; | |
358 | nr_to_read -= this_chunk; | |
359 | } | |
360 | return ret; | |
361 | } | |
362 | ||
363 | /* | |
364 | * Check how effective readahead is being. If the amount of started IO is | |
365 | * less than expected then the file is partly or fully in pagecache and | |
366 | * readahead isn't helping. | |
367 | * | |
368 | */ | |
369 | static inline int check_ra_success(struct file_ra_state *ra, | |
370 | unsigned long nr_to_read, unsigned long actual) | |
371 | { | |
372 | if (actual == 0) { | |
373 | ra->cache_hit += nr_to_read; | |
374 | if (ra->cache_hit >= VM_MAX_CACHE_HIT) { | |
375 | ra_off(ra); | |
376 | ra->flags |= RA_FLAG_INCACHE; | |
377 | return 0; | |
378 | } | |
379 | } else { | |
380 | ra->cache_hit=0; | |
381 | } | |
382 | return 1; | |
383 | } | |
384 | ||
385 | /* | |
386 | * This version skips the IO if the queue is read-congested, and will tell the | |
387 | * block layer to abandon the readahead if request allocation would block. | |
388 | * | |
389 | * force_page_cache_readahead() will ignore queue congestion and will block on | |
390 | * request queues. | |
391 | */ | |
392 | int do_page_cache_readahead(struct address_space *mapping, struct file *filp, | |
7361f4d8 | 393 | pgoff_t offset, unsigned long nr_to_read) |
1da177e4 LT |
394 | { |
395 | if (bdi_read_congested(mapping->backing_dev_info)) | |
396 | return -1; | |
397 | ||
46fc3e7b | 398 | return __do_page_cache_readahead(mapping, filp, offset, nr_to_read, 0); |
1da177e4 LT |
399 | } |
400 | ||
401 | /* | |
402 | * Read 'nr_to_read' pages starting at page 'offset'. If the flag 'block' | |
403 | * is set wait till the read completes. Otherwise attempt to read without | |
404 | * blocking. | |
d6e05edc AM |
405 | * Returns 1 meaning 'success' if read is successful without switching off |
406 | * readahead mode. Otherwise return failure. | |
1da177e4 LT |
407 | */ |
408 | static int | |
409 | blockable_page_cache_readahead(struct address_space *mapping, struct file *filp, | |
7361f4d8 | 410 | pgoff_t offset, unsigned long nr_to_read, |
1da177e4 LT |
411 | struct file_ra_state *ra, int block) |
412 | { | |
413 | int actual; | |
414 | ||
415 | if (!block && bdi_read_congested(mapping->backing_dev_info)) | |
416 | return 0; | |
417 | ||
46fc3e7b | 418 | actual = __do_page_cache_readahead(mapping, filp, offset, nr_to_read, 0); |
1da177e4 LT |
419 | |
420 | return check_ra_success(ra, nr_to_read, actual); | |
421 | } | |
422 | ||
423 | static int make_ahead_window(struct address_space *mapping, struct file *filp, | |
424 | struct file_ra_state *ra, int force) | |
425 | { | |
426 | int block, ret; | |
427 | ||
428 | ra->ahead_size = get_next_ra_size(ra); | |
429 | ra->ahead_start = ra->start + ra->size; | |
430 | ||
6ce745ed | 431 | block = force || (ra->prev_index >= ra->ahead_start); |
1da177e4 LT |
432 | ret = blockable_page_cache_readahead(mapping, filp, |
433 | ra->ahead_start, ra->ahead_size, ra, block); | |
434 | ||
435 | if (!ret && !force) { | |
436 | /* A read failure in blocking mode, implies pages are | |
437 | * all cached. So we can safely assume we have taken | |
438 | * care of all the pages requested in this call. | |
439 | * A read failure in non-blocking mode, implies we are | |
440 | * reading more pages than requested in this call. So | |
441 | * we safely assume we have taken care of all the pages | |
442 | * requested in this call. | |
443 | * | |
444 | * Just reset the ahead window in case we failed due to | |
445 | * congestion. The ahead window will any way be closed | |
446 | * in case we failed due to excessive page cache hits. | |
447 | */ | |
a564da39 | 448 | reset_ahead_window(ra); |
1da177e4 LT |
449 | } |
450 | ||
451 | return ret; | |
452 | } | |
453 | ||
7361f4d8 AM |
454 | /** |
455 | * page_cache_readahead - generic adaptive readahead | |
456 | * @mapping: address_space which holds the pagecache and I/O vectors | |
457 | * @ra: file_ra_state which holds the readahead state | |
458 | * @filp: passed on to ->readpage() and ->readpages() | |
459 | * @offset: start offset into @mapping, in PAGE_CACHE_SIZE units | |
460 | * @req_size: hint: total size of the read which the caller is performing in | |
461 | * PAGE_CACHE_SIZE units | |
462 | * | |
46fc3e7b | 463 | * page_cache_readahead() is the main function. It performs the adaptive |
1da177e4 | 464 | * readahead window size management and submits the readahead I/O. |
7361f4d8 AM |
465 | * |
466 | * Note that @filp is purely used for passing on to the ->readpage[s]() | |
467 | * handler: it may refer to a different file from @mapping (so we may not use | |
e9536ae7 | 468 | * @filp->f_mapping or @filp->f_path.dentry->d_inode here). |
7361f4d8 AM |
469 | * Also, @ra may not be equal to &@filp->f_ra. |
470 | * | |
1da177e4 LT |
471 | */ |
472 | unsigned long | |
473 | page_cache_readahead(struct address_space *mapping, struct file_ra_state *ra, | |
7361f4d8 | 474 | struct file *filp, pgoff_t offset, unsigned long req_size) |
1da177e4 LT |
475 | { |
476 | unsigned long max, newsize; | |
477 | int sequential; | |
478 | ||
479 | /* | |
480 | * We avoid doing extra work and bogusly perturbing the readahead | |
481 | * window expansion logic. | |
482 | */ | |
6ce745ed | 483 | if (offset == ra->prev_index && --req_size) |
1da177e4 LT |
484 | ++offset; |
485 | ||
6ce745ed JK |
486 | /* Note that prev_index == -1 if it is a first read */ |
487 | sequential = (offset == ra->prev_index + 1); | |
488 | ra->prev_index = offset; | |
489 | ra->prev_offset = 0; | |
1da177e4 LT |
490 | |
491 | max = get_max_readahead(ra); | |
492 | newsize = min(req_size, max); | |
493 | ||
494 | /* No readahead or sub-page sized read or file already in cache */ | |
495 | if (newsize == 0 || (ra->flags & RA_FLAG_INCACHE)) | |
496 | goto out; | |
497 | ||
6ce745ed | 498 | ra->prev_index += newsize - 1; |
1da177e4 LT |
499 | |
500 | /* | |
501 | * Special case - first read at start of file. We'll assume it's | |
502 | * a whole-file read and grow the window fast. Or detect first | |
503 | * sequential access | |
504 | */ | |
505 | if (sequential && ra->size == 0) { | |
506 | ra->size = get_init_ra_size(newsize, max); | |
507 | ra->start = offset; | |
508 | if (!blockable_page_cache_readahead(mapping, filp, offset, | |
509 | ra->size, ra, 1)) | |
510 | goto out; | |
511 | ||
512 | /* | |
513 | * If the request size is larger than our max readahead, we | |
514 | * at least want to be sure that we get 2 IOs in flight and | |
515 | * we know that we will definitly need the new I/O. | |
516 | * once we do this, subsequent calls should be able to overlap | |
517 | * IOs,* thus preventing stalls. so issue the ahead window | |
518 | * immediately. | |
519 | */ | |
520 | if (req_size >= max) | |
521 | make_ahead_window(mapping, filp, ra, 1); | |
522 | ||
523 | goto out; | |
524 | } | |
525 | ||
526 | /* | |
527 | * Now handle the random case: | |
528 | * partial page reads and first access were handled above, | |
529 | * so this must be the next page otherwise it is random | |
530 | */ | |
531 | if (!sequential) { | |
532 | ra_off(ra); | |
533 | blockable_page_cache_readahead(mapping, filp, offset, | |
534 | newsize, ra, 1); | |
535 | goto out; | |
536 | } | |
537 | ||
538 | /* | |
539 | * If we get here we are doing sequential IO and this was not the first | |
540 | * occurence (ie we have an existing window) | |
541 | */ | |
1da177e4 LT |
542 | if (ra->ahead_start == 0) { /* no ahead window yet */ |
543 | if (!make_ahead_window(mapping, filp, ra, 0)) | |
a564da39 | 544 | goto recheck; |
1da177e4 | 545 | } |
a564da39 | 546 | |
1da177e4 LT |
547 | /* |
548 | * Already have an ahead window, check if we crossed into it. | |
549 | * If so, shift windows and issue a new ahead window. | |
550 | * Only return the #pages that are in the current window, so that | |
551 | * we get called back on the first page of the ahead window which | |
552 | * will allow us to submit more IO. | |
553 | */ | |
6ce745ed | 554 | if (ra->prev_index >= ra->ahead_start) { |
1da177e4 LT |
555 | ra->start = ra->ahead_start; |
556 | ra->size = ra->ahead_size; | |
557 | make_ahead_window(mapping, filp, ra, 0); | |
a564da39 | 558 | recheck: |
6ce745ed JK |
559 | /* prev_index shouldn't overrun the ahead window */ |
560 | ra->prev_index = min(ra->prev_index, | |
a564da39 | 561 | ra->ahead_start + ra->ahead_size - 1); |
1da177e4 LT |
562 | } |
563 | ||
564 | out: | |
6ce745ed | 565 | return ra->prev_index + 1; |
1da177e4 | 566 | } |
d8733c29 | 567 | EXPORT_SYMBOL_GPL(page_cache_readahead); |
1da177e4 LT |
568 | |
569 | /* | |
570 | * handle_ra_miss() is called when it is known that a page which should have | |
571 | * been present in the pagecache (we just did some readahead there) was in fact | |
572 | * not found. This will happen if it was evicted by the VM (readahead | |
573 | * thrashing) | |
574 | * | |
575 | * Turn on the cache miss flag in the RA struct, this will cause the RA code | |
576 | * to reduce the RA size on the next read. | |
577 | */ | |
578 | void handle_ra_miss(struct address_space *mapping, | |
579 | struct file_ra_state *ra, pgoff_t offset) | |
580 | { | |
581 | ra->flags |= RA_FLAG_MISS; | |
582 | ra->flags &= ~RA_FLAG_INCACHE; | |
3b30bbd9 | 583 | ra->cache_hit = 0; |
1da177e4 LT |
584 | } |
585 | ||
586 | /* | |
587 | * Given a desired number of PAGE_CACHE_SIZE readahead pages, return a | |
588 | * sensible upper limit. | |
589 | */ | |
590 | unsigned long max_sane_readahead(unsigned long nr) | |
591 | { | |
05a0416b CL |
592 | return min(nr, (node_page_state(numa_node_id(), NR_INACTIVE) |
593 | + node_page_state(numa_node_id(), NR_FREE_PAGES)) / 2); | |
1da177e4 | 594 | } |
5ce1110b FW |
595 | |
596 | /* | |
597 | * Submit IO for the read-ahead request in file_ra_state. | |
598 | */ | |
599 | unsigned long ra_submit(struct file_ra_state *ra, | |
600 | struct address_space *mapping, struct file *filp) | |
601 | { | |
602 | unsigned long ra_size; | |
603 | unsigned long la_size; | |
604 | int actual; | |
605 | ||
606 | ra_size = ra_readahead_size(ra); | |
607 | la_size = ra_lookahead_size(ra); | |
608 | actual = __do_page_cache_readahead(mapping, filp, | |
609 | ra->ra_index, ra_size, la_size); | |
610 | ||
611 | return actual; | |
612 | } | |
613 | EXPORT_SYMBOL_GPL(ra_submit); | |
122a21d1 FW |
614 | |
615 | /* | |
616 | * Get the previous window size, ramp it up, and | |
617 | * return it as the new window size. | |
618 | */ | |
619 | static unsigned long get_next_ra_size2(struct file_ra_state *ra, | |
620 | unsigned long max) | |
621 | { | |
622 | unsigned long cur = ra->readahead_index - ra->ra_index; | |
623 | unsigned long newsize; | |
624 | ||
625 | if (cur < max / 16) | |
626 | newsize = cur * 4; | |
627 | else | |
628 | newsize = cur * 2; | |
629 | ||
630 | return min(newsize, max); | |
631 | } | |
632 | ||
633 | /* | |
634 | * On-demand readahead design. | |
635 | * | |
636 | * The fields in struct file_ra_state represent the most-recently-executed | |
637 | * readahead attempt: | |
638 | * | |
639 | * |-------- last readahead window -------->| | |
640 | * |-- application walking here -->| | |
641 | * ======#============|==================#=====================| | |
642 | * ^la_index ^ra_index ^lookahead_index ^readahead_index | |
643 | * | |
644 | * [ra_index, readahead_index) represents the last readahead window. | |
645 | * | |
646 | * [la_index, lookahead_index] is where the application would be walking(in | |
647 | * the common case of cache-cold sequential reads): the last window was | |
648 | * established when the application was at la_index, and the next window will | |
649 | * be bring in when the application reaches lookahead_index. | |
650 | * | |
651 | * To overlap application thinking time and disk I/O time, we do | |
652 | * `readahead pipelining': Do not wait until the application consumed all | |
653 | * readahead pages and stalled on the missing page at readahead_index; | |
654 | * Instead, submit an asynchronous readahead I/O as early as the application | |
655 | * reads on the page at lookahead_index. Normally lookahead_index will be | |
656 | * equal to ra_index, for maximum pipelining. | |
657 | * | |
658 | * In interleaved sequential reads, concurrent streams on the same fd can | |
659 | * be invalidating each other's readahead state. So we flag the new readahead | |
660 | * page at lookahead_index with PG_readahead, and use it as readahead | |
661 | * indicator. The flag won't be set on already cached pages, to avoid the | |
662 | * readahead-for-nothing fuss, saving pointless page cache lookups. | |
663 | * | |
664 | * prev_index tracks the last visited page in the _previous_ read request. | |
665 | * It should be maintained by the caller, and will be used for detecting | |
666 | * small random reads. Note that the readahead algorithm checks loosely | |
667 | * for sequential patterns. Hence interleaved reads might be served as | |
668 | * sequential ones. | |
669 | * | |
670 | * There is a special-case: if the first page which the application tries to | |
671 | * read happens to be the first page of the file, it is assumed that a linear | |
672 | * read is about to happen and the window is immediately set to the initial size | |
673 | * based on I/O request size and the max_readahead. | |
674 | * | |
675 | * The code ramps up the readahead size aggressively at first, but slow down as | |
676 | * it approaches max_readhead. | |
677 | */ | |
678 | ||
679 | /* | |
680 | * A minimal readahead algorithm for trivial sequential/random reads. | |
681 | */ | |
682 | static unsigned long | |
683 | ondemand_readahead(struct address_space *mapping, | |
684 | struct file_ra_state *ra, struct file *filp, | |
685 | struct page *page, pgoff_t offset, | |
686 | unsigned long req_size) | |
687 | { | |
688 | unsigned long max; /* max readahead pages */ | |
689 | pgoff_t ra_index; /* readahead index */ | |
690 | unsigned long ra_size; /* readahead size */ | |
691 | unsigned long la_size; /* lookahead size */ | |
692 | int sequential; | |
693 | ||
694 | max = ra->ra_pages; | |
695 | sequential = (offset - ra->prev_index <= 1UL) || (req_size > max); | |
696 | ||
697 | /* | |
698 | * Lookahead/readahead hit, assume sequential access. | |
699 | * Ramp up sizes, and push forward the readahead window. | |
700 | */ | |
701 | if (offset && (offset == ra->lookahead_index || | |
702 | offset == ra->readahead_index)) { | |
703 | ra_index = ra->readahead_index; | |
704 | ra_size = get_next_ra_size2(ra, max); | |
705 | la_size = ra_size; | |
706 | goto fill_ra; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Standalone, small read. | |
711 | * Read as is, and do not pollute the readahead state. | |
712 | */ | |
713 | if (!page && !sequential) { | |
714 | return __do_page_cache_readahead(mapping, filp, | |
715 | offset, req_size, 0); | |
716 | } | |
717 | ||
718 | /* | |
719 | * It may be one of | |
720 | * - first read on start of file | |
721 | * - sequential cache miss | |
722 | * - oversize random read | |
723 | * Start readahead for it. | |
724 | */ | |
725 | ra_index = offset; | |
726 | ra_size = get_init_ra_size(req_size, max); | |
727 | la_size = ra_size > req_size ? ra_size - req_size : ra_size; | |
728 | ||
729 | /* | |
730 | * Hit on a lookahead page without valid readahead state. | |
731 | * E.g. interleaved reads. | |
732 | * Not knowing its readahead pos/size, bet on the minimal possible one. | |
733 | */ | |
734 | if (page) { | |
735 | ra_index++; | |
736 | ra_size = min(4 * ra_size, max); | |
737 | } | |
738 | ||
739 | fill_ra: | |
740 | ra_set_index(ra, offset, ra_index); | |
741 | ra_set_size(ra, ra_size, la_size); | |
742 | ||
743 | return ra_submit(ra, mapping, filp); | |
744 | } | |
745 | ||
746 | /** | |
747 | * page_cache_readahead_ondemand - generic file readahead | |
748 | * @mapping: address_space which holds the pagecache and I/O vectors | |
749 | * @ra: file_ra_state which holds the readahead state | |
750 | * @filp: passed on to ->readpage() and ->readpages() | |
751 | * @page: the page at @offset, or NULL if non-present | |
752 | * @offset: start offset into @mapping, in PAGE_CACHE_SIZE units | |
753 | * @req_size: hint: total size of the read which the caller is performing in | |
754 | * PAGE_CACHE_SIZE units | |
755 | * | |
756 | * page_cache_readahead_ondemand() is the entry point of readahead logic. | |
757 | * This function should be called when it is time to perform readahead: | |
758 | * 1) @page == NULL | |
759 | * A cache miss happened, time for synchronous readahead. | |
760 | * 2) @page != NULL && PageReadahead(@page) | |
761 | * A look-ahead hit occured, time for asynchronous readahead. | |
762 | */ | |
763 | unsigned long | |
764 | page_cache_readahead_ondemand(struct address_space *mapping, | |
765 | struct file_ra_state *ra, struct file *filp, | |
766 | struct page *page, pgoff_t offset, | |
767 | unsigned long req_size) | |
768 | { | |
769 | /* no read-ahead */ | |
770 | if (!ra->ra_pages) | |
771 | return 0; | |
772 | ||
773 | if (page) { | |
774 | ClearPageReadahead(page); | |
775 | ||
776 | /* | |
777 | * Defer asynchronous read-ahead on IO congestion. | |
778 | */ | |
779 | if (bdi_read_congested(mapping->backing_dev_info)) | |
780 | return 0; | |
781 | } | |
782 | ||
783 | /* do read-ahead */ | |
784 | return ondemand_readahead(mapping, ra, filp, page, | |
785 | offset, req_size); | |
786 | } | |
787 | EXPORT_SYMBOL_GPL(page_cache_readahead_ondemand); |