]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/readahead.c - address_space-level file readahead. | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
e1f8e874 | 6 | * 09Apr2002 Andrew Morton |
1da177e4 LT |
7 | * Initial version. |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
11bd969f | 11 | #include <linux/dax.h> |
5a0e3ad6 | 12 | #include <linux/gfp.h> |
b95f1b31 | 13 | #include <linux/export.h> |
1da177e4 LT |
14 | #include <linux/blkdev.h> |
15 | #include <linux/backing-dev.h> | |
8bde37f0 | 16 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 17 | #include <linux/pagevec.h> |
f5ff8422 | 18 | #include <linux/pagemap.h> |
782182e5 CW |
19 | #include <linux/syscalls.h> |
20 | #include <linux/file.h> | |
d72ee911 | 21 | #include <linux/mm_inline.h> |
ca47e8c7 | 22 | #include <linux/blk-cgroup.h> |
3d8f7615 | 23 | #include <linux/fadvise.h> |
1da177e4 | 24 | |
29f175d1 FF |
25 | #include "internal.h" |
26 | ||
1da177e4 LT |
27 | /* |
28 | * Initialise a struct file's readahead state. Assumes that the caller has | |
29 | * memset *ra to zero. | |
30 | */ | |
31 | void | |
32 | file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping) | |
33 | { | |
de1414a6 | 34 | ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages; |
f4e6b498 | 35 | ra->prev_pos = -1; |
1da177e4 | 36 | } |
d41cc702 | 37 | EXPORT_SYMBOL_GPL(file_ra_state_init); |
1da177e4 | 38 | |
03fb3d2a DH |
39 | /* |
40 | * see if a page needs releasing upon read_cache_pages() failure | |
266cf658 DH |
41 | * - the caller of read_cache_pages() may have set PG_private or PG_fscache |
42 | * before calling, such as the NFS fs marking pages that are cached locally | |
43 | * on disk, thus we need to give the fs a chance to clean up in the event of | |
44 | * an error | |
03fb3d2a DH |
45 | */ |
46 | static void read_cache_pages_invalidate_page(struct address_space *mapping, | |
47 | struct page *page) | |
48 | { | |
266cf658 | 49 | if (page_has_private(page)) { |
03fb3d2a DH |
50 | if (!trylock_page(page)) |
51 | BUG(); | |
52 | page->mapping = mapping; | |
09cbfeaf | 53 | do_invalidatepage(page, 0, PAGE_SIZE); |
03fb3d2a DH |
54 | page->mapping = NULL; |
55 | unlock_page(page); | |
56 | } | |
09cbfeaf | 57 | put_page(page); |
03fb3d2a DH |
58 | } |
59 | ||
60 | /* | |
61 | * release a list of pages, invalidating them first if need be | |
62 | */ | |
63 | static void read_cache_pages_invalidate_pages(struct address_space *mapping, | |
64 | struct list_head *pages) | |
65 | { | |
66 | struct page *victim; | |
67 | ||
68 | while (!list_empty(pages)) { | |
c8ad6302 | 69 | victim = lru_to_page(pages); |
03fb3d2a DH |
70 | list_del(&victim->lru); |
71 | read_cache_pages_invalidate_page(mapping, victim); | |
72 | } | |
73 | } | |
74 | ||
1da177e4 | 75 | /** |
bd40cdda | 76 | * read_cache_pages - populate an address space with some pages & start reads against them |
1da177e4 LT |
77 | * @mapping: the address_space |
78 | * @pages: The address of a list_head which contains the target pages. These | |
79 | * pages have their ->index populated and are otherwise uninitialised. | |
80 | * @filler: callback routine for filling a single page. | |
81 | * @data: private data for the callback routine. | |
82 | * | |
83 | * Hides the details of the LRU cache etc from the filesystems. | |
84 | */ | |
85 | int read_cache_pages(struct address_space *mapping, struct list_head *pages, | |
86 | int (*filler)(void *, struct page *), void *data) | |
87 | { | |
88 | struct page *page; | |
1da177e4 LT |
89 | int ret = 0; |
90 | ||
1da177e4 | 91 | while (!list_empty(pages)) { |
c8ad6302 | 92 | page = lru_to_page(pages); |
1da177e4 | 93 | list_del(&page->lru); |
063d99b4 | 94 | if (add_to_page_cache_lru(page, mapping, page->index, |
8a5c743e | 95 | readahead_gfp_mask(mapping))) { |
03fb3d2a | 96 | read_cache_pages_invalidate_page(mapping, page); |
1da177e4 LT |
97 | continue; |
98 | } | |
09cbfeaf | 99 | put_page(page); |
eb2be189 | 100 | |
1da177e4 | 101 | ret = filler(data, page); |
eb2be189 | 102 | if (unlikely(ret)) { |
03fb3d2a | 103 | read_cache_pages_invalidate_pages(mapping, pages); |
1da177e4 LT |
104 | break; |
105 | } | |
09cbfeaf | 106 | task_io_account_read(PAGE_SIZE); |
1da177e4 | 107 | } |
1da177e4 LT |
108 | return ret; |
109 | } | |
110 | ||
111 | EXPORT_SYMBOL(read_cache_pages); | |
112 | ||
113 | static int read_pages(struct address_space *mapping, struct file *filp, | |
8a5c743e | 114 | struct list_head *pages, unsigned int nr_pages, gfp_t gfp) |
1da177e4 | 115 | { |
5b417b18 | 116 | struct blk_plug plug; |
1da177e4 | 117 | unsigned page_idx; |
994fc28c | 118 | int ret; |
1da177e4 | 119 | |
5b417b18 JA |
120 | blk_start_plug(&plug); |
121 | ||
1da177e4 LT |
122 | if (mapping->a_ops->readpages) { |
123 | ret = mapping->a_ops->readpages(filp, mapping, pages, nr_pages); | |
029e332e OH |
124 | /* Clean up the remaining pages */ |
125 | put_pages_list(pages); | |
1da177e4 LT |
126 | goto out; |
127 | } | |
128 | ||
1da177e4 | 129 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
c8ad6302 | 130 | struct page *page = lru_to_page(pages); |
1da177e4 | 131 | list_del(&page->lru); |
8a5c743e | 132 | if (!add_to_page_cache_lru(page, mapping, page->index, gfp)) |
9f1a3cfc | 133 | mapping->a_ops->readpage(filp, page); |
09cbfeaf | 134 | put_page(page); |
1da177e4 | 135 | } |
994fc28c | 136 | ret = 0; |
5b417b18 | 137 | |
1da177e4 | 138 | out: |
5b417b18 JA |
139 | blk_finish_plug(&plug); |
140 | ||
1da177e4 LT |
141 | return ret; |
142 | } | |
143 | ||
1da177e4 | 144 | /* |
b3751e6a CH |
145 | * __do_page_cache_readahead() actually reads a chunk of disk. It allocates |
146 | * the pages first, then submits them for I/O. This avoids the very bad | |
1da177e4 LT |
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. | |
1da177e4 | 151 | */ |
c534aa3f CH |
152 | unsigned int __do_page_cache_readahead(struct address_space *mapping, |
153 | struct file *filp, pgoff_t offset, unsigned long nr_to_read, | |
154 | unsigned long lookahead_size) | |
1da177e4 LT |
155 | { |
156 | struct inode *inode = mapping->host; | |
157 | struct page *page; | |
158 | unsigned long end_index; /* The last page we want to read */ | |
159 | LIST_HEAD(page_pool); | |
160 | int page_idx; | |
c534aa3f | 161 | unsigned int nr_pages = 0; |
1da177e4 | 162 | loff_t isize = i_size_read(inode); |
8a5c743e | 163 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
1da177e4 LT |
164 | |
165 | if (isize == 0) | |
166 | goto out; | |
167 | ||
09cbfeaf | 168 | end_index = ((isize - 1) >> PAGE_SHIFT); |
1da177e4 LT |
169 | |
170 | /* | |
171 | * Preallocate as many pages as we will need. | |
172 | */ | |
1da177e4 | 173 | for (page_idx = 0; page_idx < nr_to_read; page_idx++) { |
7361f4d8 | 174 | pgoff_t page_offset = offset + page_idx; |
c743d96b | 175 | |
1da177e4 LT |
176 | if (page_offset > end_index) |
177 | break; | |
178 | ||
00128188 | 179 | rcu_read_lock(); |
b93b0163 | 180 | page = radix_tree_lookup(&mapping->i_pages, page_offset); |
00128188 | 181 | rcu_read_unlock(); |
b3751e6a CH |
182 | if (page && !radix_tree_exceptional_entry(page)) { |
183 | /* | |
184 | * Page already present? Kick off the current batch of | |
185 | * contiguous pages before continuing with the next | |
186 | * batch. | |
187 | */ | |
188 | if (nr_pages) | |
189 | read_pages(mapping, filp, &page_pool, nr_pages, | |
190 | gfp_mask); | |
191 | nr_pages = 0; | |
1da177e4 | 192 | continue; |
b3751e6a | 193 | } |
1da177e4 | 194 | |
8a5c743e | 195 | page = __page_cache_alloc(gfp_mask); |
1da177e4 LT |
196 | if (!page) |
197 | break; | |
198 | page->index = page_offset; | |
199 | list_add(&page->lru, &page_pool); | |
46fc3e7b FW |
200 | if (page_idx == nr_to_read - lookahead_size) |
201 | SetPageReadahead(page); | |
836978b3 | 202 | nr_pages++; |
1da177e4 | 203 | } |
1da177e4 LT |
204 | |
205 | /* | |
206 | * Now start the IO. We ignore I/O errors - if the page is not | |
207 | * uptodate then the caller will launch readpage again, and | |
208 | * will then handle the error. | |
209 | */ | |
836978b3 CH |
210 | if (nr_pages) |
211 | read_pages(mapping, filp, &page_pool, nr_pages, gfp_mask); | |
1da177e4 LT |
212 | BUG_ON(!list_empty(&page_pool)); |
213 | out: | |
836978b3 | 214 | return nr_pages; |
1da177e4 LT |
215 | } |
216 | ||
217 | /* | |
218 | * Chunk the readahead into 2 megabyte units, so that we don't pin too much | |
219 | * memory at once. | |
220 | */ | |
221 | int force_page_cache_readahead(struct address_space *mapping, struct file *filp, | |
9491ae4a | 222 | pgoff_t offset, unsigned long nr_to_read) |
1da177e4 | 223 | { |
9491ae4a JA |
224 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
225 | struct file_ra_state *ra = &filp->f_ra; | |
226 | unsigned long max_pages; | |
227 | ||
1da177e4 LT |
228 | if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages)) |
229 | return -EINVAL; | |
230 | ||
9491ae4a JA |
231 | /* |
232 | * If the request exceeds the readahead window, allow the read to | |
233 | * be up to the optimal hardware IO size | |
234 | */ | |
235 | max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages); | |
236 | nr_to_read = min(nr_to_read, max_pages); | |
1da177e4 | 237 | while (nr_to_read) { |
09cbfeaf | 238 | unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE; |
1da177e4 LT |
239 | |
240 | if (this_chunk > nr_to_read) | |
241 | this_chunk = nr_to_read; | |
c534aa3f | 242 | __do_page_cache_readahead(mapping, filp, offset, this_chunk, 0); |
58d5640e | 243 | |
1da177e4 LT |
244 | offset += this_chunk; |
245 | nr_to_read -= this_chunk; | |
246 | } | |
58d5640e | 247 | return 0; |
1da177e4 LT |
248 | } |
249 | ||
c743d96b FW |
250 | /* |
251 | * Set the initial window size, round to next power of 2 and square | |
252 | * for small size, x 4 for medium, and x 2 for large | |
253 | * for 128k (32 page) max ra | |
254 | * 1-8 page = 32k initial, > 8 page = 128k initial | |
255 | */ | |
256 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) | |
257 | { | |
258 | unsigned long newsize = roundup_pow_of_two(size); | |
259 | ||
260 | if (newsize <= max / 32) | |
261 | newsize = newsize * 4; | |
262 | else if (newsize <= max / 4) | |
263 | newsize = newsize * 2; | |
264 | else | |
265 | newsize = max; | |
266 | ||
267 | return newsize; | |
268 | } | |
269 | ||
122a21d1 FW |
270 | /* |
271 | * Get the previous window size, ramp it up, and | |
272 | * return it as the new window size. | |
273 | */ | |
c743d96b | 274 | static unsigned long get_next_ra_size(struct file_ra_state *ra, |
122a21d1 FW |
275 | unsigned long max) |
276 | { | |
f9acc8c7 | 277 | unsigned long cur = ra->size; |
122a21d1 FW |
278 | unsigned long newsize; |
279 | ||
280 | if (cur < max / 16) | |
c743d96b | 281 | newsize = 4 * cur; |
122a21d1 | 282 | else |
c743d96b | 283 | newsize = 2 * cur; |
122a21d1 FW |
284 | |
285 | return min(newsize, max); | |
286 | } | |
287 | ||
288 | /* | |
289 | * On-demand readahead design. | |
290 | * | |
291 | * The fields in struct file_ra_state represent the most-recently-executed | |
292 | * readahead attempt: | |
293 | * | |
f9acc8c7 FW |
294 | * |<----- async_size ---------| |
295 | * |------------------- size -------------------->| | |
296 | * |==================#===========================| | |
297 | * ^start ^page marked with PG_readahead | |
122a21d1 FW |
298 | * |
299 | * To overlap application thinking time and disk I/O time, we do | |
300 | * `readahead pipelining': Do not wait until the application consumed all | |
301 | * readahead pages and stalled on the missing page at readahead_index; | |
f9acc8c7 FW |
302 | * Instead, submit an asynchronous readahead I/O as soon as there are |
303 | * only async_size pages left in the readahead window. Normally async_size | |
304 | * will be equal to size, for maximum pipelining. | |
122a21d1 FW |
305 | * |
306 | * In interleaved sequential reads, concurrent streams on the same fd can | |
307 | * be invalidating each other's readahead state. So we flag the new readahead | |
f9acc8c7 | 308 | * page at (start+size-async_size) with PG_readahead, and use it as readahead |
122a21d1 FW |
309 | * indicator. The flag won't be set on already cached pages, to avoid the |
310 | * readahead-for-nothing fuss, saving pointless page cache lookups. | |
311 | * | |
f4e6b498 | 312 | * prev_pos tracks the last visited byte in the _previous_ read request. |
122a21d1 FW |
313 | * It should be maintained by the caller, and will be used for detecting |
314 | * small random reads. Note that the readahead algorithm checks loosely | |
315 | * for sequential patterns. Hence interleaved reads might be served as | |
316 | * sequential ones. | |
317 | * | |
318 | * There is a special-case: if the first page which the application tries to | |
319 | * read happens to be the first page of the file, it is assumed that a linear | |
320 | * read is about to happen and the window is immediately set to the initial size | |
321 | * based on I/O request size and the max_readahead. | |
322 | * | |
323 | * The code ramps up the readahead size aggressively at first, but slow down as | |
324 | * it approaches max_readhead. | |
325 | */ | |
326 | ||
10be0b37 WF |
327 | /* |
328 | * Count contiguously cached pages from @offset-1 to @offset-@max, | |
329 | * this count is a conservative estimation of | |
330 | * - length of the sequential read sequence, or | |
331 | * - thrashing threshold in memory tight systems | |
332 | */ | |
333 | static pgoff_t count_history_pages(struct address_space *mapping, | |
10be0b37 WF |
334 | pgoff_t offset, unsigned long max) |
335 | { | |
336 | pgoff_t head; | |
337 | ||
338 | rcu_read_lock(); | |
e7b563bb | 339 | head = page_cache_prev_hole(mapping, offset - 1, max); |
10be0b37 WF |
340 | rcu_read_unlock(); |
341 | ||
342 | return offset - 1 - head; | |
343 | } | |
344 | ||
345 | /* | |
346 | * page cache context based read-ahead | |
347 | */ | |
348 | static int try_context_readahead(struct address_space *mapping, | |
349 | struct file_ra_state *ra, | |
350 | pgoff_t offset, | |
351 | unsigned long req_size, | |
352 | unsigned long max) | |
353 | { | |
354 | pgoff_t size; | |
355 | ||
3e2faa08 | 356 | size = count_history_pages(mapping, offset, max); |
10be0b37 WF |
357 | |
358 | /* | |
2cad4018 | 359 | * not enough history pages: |
10be0b37 WF |
360 | * it could be a random read |
361 | */ | |
2cad4018 | 362 | if (size <= req_size) |
10be0b37 WF |
363 | return 0; |
364 | ||
365 | /* | |
366 | * starts from beginning of file: | |
367 | * it is a strong indication of long-run stream (or whole-file-read) | |
368 | */ | |
369 | if (size >= offset) | |
370 | size *= 2; | |
371 | ||
372 | ra->start = offset; | |
2cad4018 FW |
373 | ra->size = min(size + req_size, max); |
374 | ra->async_size = 1; | |
10be0b37 WF |
375 | |
376 | return 1; | |
377 | } | |
378 | ||
122a21d1 FW |
379 | /* |
380 | * A minimal readahead algorithm for trivial sequential/random reads. | |
381 | */ | |
382 | static unsigned long | |
383 | ondemand_readahead(struct address_space *mapping, | |
384 | struct file_ra_state *ra, struct file *filp, | |
cf914a7d | 385 | bool hit_readahead_marker, pgoff_t offset, |
122a21d1 FW |
386 | unsigned long req_size) |
387 | { | |
9491ae4a JA |
388 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
389 | unsigned long max_pages = ra->ra_pages; | |
dc30b96a | 390 | unsigned long add_pages; |
af248a0c | 391 | pgoff_t prev_offset; |
045a2529 | 392 | |
9491ae4a JA |
393 | /* |
394 | * If the request exceeds the readahead window, allow the read to | |
395 | * be up to the optimal hardware IO size | |
396 | */ | |
397 | if (req_size > max_pages && bdi->io_pages > max_pages) | |
398 | max_pages = min(req_size, bdi->io_pages); | |
399 | ||
045a2529 WF |
400 | /* |
401 | * start of file | |
402 | */ | |
403 | if (!offset) | |
404 | goto initial_readahead; | |
122a21d1 FW |
405 | |
406 | /* | |
f9acc8c7 | 407 | * It's the expected callback offset, assume sequential access. |
122a21d1 FW |
408 | * Ramp up sizes, and push forward the readahead window. |
409 | */ | |
045a2529 WF |
410 | if ((offset == (ra->start + ra->size - ra->async_size) || |
411 | offset == (ra->start + ra->size))) { | |
f9acc8c7 | 412 | ra->start += ra->size; |
9491ae4a | 413 | ra->size = get_next_ra_size(ra, max_pages); |
f9acc8c7 FW |
414 | ra->async_size = ra->size; |
415 | goto readit; | |
122a21d1 FW |
416 | } |
417 | ||
6b10c6c9 FW |
418 | /* |
419 | * Hit a marked page without valid readahead state. | |
420 | * E.g. interleaved reads. | |
421 | * Query the pagecache for async_size, which normally equals to | |
422 | * readahead size. Ramp it up and use it as the new readahead size. | |
423 | */ | |
424 | if (hit_readahead_marker) { | |
425 | pgoff_t start; | |
426 | ||
30002ed2 | 427 | rcu_read_lock(); |
9491ae4a | 428 | start = page_cache_next_hole(mapping, offset + 1, max_pages); |
30002ed2 | 429 | rcu_read_unlock(); |
6b10c6c9 | 430 | |
9491ae4a | 431 | if (!start || start - offset > max_pages) |
6b10c6c9 FW |
432 | return 0; |
433 | ||
434 | ra->start = start; | |
435 | ra->size = start - offset; /* old async_size */ | |
160334a0 | 436 | ra->size += req_size; |
9491ae4a | 437 | ra->size = get_next_ra_size(ra, max_pages); |
6b10c6c9 FW |
438 | ra->async_size = ra->size; |
439 | goto readit; | |
440 | } | |
441 | ||
122a21d1 | 442 | /* |
045a2529 | 443 | * oversize read |
122a21d1 | 444 | */ |
9491ae4a | 445 | if (req_size > max_pages) |
045a2529 WF |
446 | goto initial_readahead; |
447 | ||
448 | /* | |
449 | * sequential cache miss | |
af248a0c DR |
450 | * trivial case: (offset - prev_offset) == 1 |
451 | * unaligned reads: (offset - prev_offset) == 0 | |
045a2529 | 452 | */ |
09cbfeaf | 453 | prev_offset = (unsigned long long)ra->prev_pos >> PAGE_SHIFT; |
af248a0c | 454 | if (offset - prev_offset <= 1UL) |
045a2529 WF |
455 | goto initial_readahead; |
456 | ||
10be0b37 WF |
457 | /* |
458 | * Query the page cache and look for the traces(cached history pages) | |
459 | * that a sequential stream would leave behind. | |
460 | */ | |
9491ae4a | 461 | if (try_context_readahead(mapping, ra, offset, req_size, max_pages)) |
10be0b37 WF |
462 | goto readit; |
463 | ||
045a2529 WF |
464 | /* |
465 | * 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 | ||
470 | initial_readahead: | |
f9acc8c7 | 471 | ra->start = offset; |
9491ae4a | 472 | ra->size = get_init_ra_size(req_size, max_pages); |
f9acc8c7 | 473 | ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size; |
122a21d1 | 474 | |
f9acc8c7 | 475 | readit: |
51daa88e WF |
476 | /* |
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. | |
dc30b96a | 480 | * Take care of maximum IO pages as above. |
51daa88e WF |
481 | */ |
482 | if (offset == ra->start && ra->size == ra->async_size) { | |
dc30b96a MS |
483 | add_pages = get_next_ra_size(ra, max_pages); |
484 | if (ra->size + add_pages <= max_pages) { | |
485 | ra->async_size = add_pages; | |
486 | ra->size += add_pages; | |
487 | } else { | |
488 | ra->size = max_pages; | |
489 | ra->async_size = max_pages >> 1; | |
490 | } | |
51daa88e WF |
491 | } |
492 | ||
122a21d1 FW |
493 | return ra_submit(ra, mapping, filp); |
494 | } | |
495 | ||
496 | /** | |
cf914a7d | 497 | * page_cache_sync_readahead - generic file readahead |
122a21d1 FW |
498 | * @mapping: address_space which holds the pagecache and I/O vectors |
499 | * @ra: file_ra_state which holds the readahead state | |
500 | * @filp: passed on to ->readpage() and ->readpages() | |
cf914a7d | 501 | * @offset: start offset into @mapping, in pagecache page-sized units |
122a21d1 | 502 | * @req_size: hint: total size of the read which the caller is performing in |
cf914a7d | 503 | * pagecache pages |
122a21d1 | 504 | * |
cf914a7d RR |
505 | * page_cache_sync_readahead() should be called when a cache miss happened: |
506 | * it will submit the read. The readahead logic may decide to piggyback more | |
507 | * pages onto the read request if access patterns suggest it will improve | |
508 | * performance. | |
122a21d1 | 509 | */ |
cf914a7d RR |
510 | void page_cache_sync_readahead(struct address_space *mapping, |
511 | struct file_ra_state *ra, struct file *filp, | |
512 | pgoff_t offset, unsigned long req_size) | |
122a21d1 FW |
513 | { |
514 | /* no read-ahead */ | |
515 | if (!ra->ra_pages) | |
cf914a7d RR |
516 | return; |
517 | ||
ca47e8c7 JB |
518 | if (blk_cgroup_congested()) |
519 | return; | |
520 | ||
0141450f | 521 | /* be dumb */ |
70655c06 | 522 | if (filp && (filp->f_mode & FMODE_RANDOM)) { |
0141450f WF |
523 | force_page_cache_readahead(mapping, filp, offset, req_size); |
524 | return; | |
525 | } | |
526 | ||
cf914a7d RR |
527 | /* do read-ahead */ |
528 | ondemand_readahead(mapping, ra, filp, false, offset, req_size); | |
529 | } | |
530 | EXPORT_SYMBOL_GPL(page_cache_sync_readahead); | |
531 | ||
532 | /** | |
533 | * page_cache_async_readahead - file readahead for marked pages | |
534 | * @mapping: address_space which holds the pagecache and I/O vectors | |
535 | * @ra: file_ra_state which holds the readahead state | |
536 | * @filp: passed on to ->readpage() and ->readpages() | |
537 | * @page: the page at @offset which has the PG_readahead flag set | |
538 | * @offset: start offset into @mapping, in pagecache page-sized units | |
539 | * @req_size: hint: total size of the read which the caller is performing in | |
540 | * pagecache pages | |
541 | * | |
bf8abe8b | 542 | * page_cache_async_readahead() should be called when a page is used which |
f7850d93 | 543 | * has the PG_readahead flag; this is a marker to suggest that the application |
cf914a7d | 544 | * has used up enough of the readahead window that we should start pulling in |
f7850d93 RD |
545 | * more pages. |
546 | */ | |
cf914a7d RR |
547 | void |
548 | page_cache_async_readahead(struct address_space *mapping, | |
549 | struct file_ra_state *ra, struct file *filp, | |
550 | struct page *page, pgoff_t offset, | |
551 | unsigned long req_size) | |
552 | { | |
553 | /* no read-ahead */ | |
554 | if (!ra->ra_pages) | |
555 | return; | |
556 | ||
557 | /* | |
558 | * Same bit is used for PG_readahead and PG_reclaim. | |
559 | */ | |
560 | if (PageWriteback(page)) | |
561 | return; | |
562 | ||
563 | ClearPageReadahead(page); | |
564 | ||
565 | /* | |
566 | * Defer asynchronous read-ahead on IO congestion. | |
567 | */ | |
703c2708 | 568 | if (inode_read_congested(mapping->host)) |
cf914a7d | 569 | return; |
122a21d1 | 570 | |
ca47e8c7 JB |
571 | if (blk_cgroup_congested()) |
572 | return; | |
573 | ||
122a21d1 | 574 | /* do read-ahead */ |
cf914a7d | 575 | ondemand_readahead(mapping, ra, filp, true, offset, req_size); |
122a21d1 | 576 | } |
cf914a7d | 577 | EXPORT_SYMBOL_GPL(page_cache_async_readahead); |
782182e5 | 578 | |
c7b95d51 | 579 | ssize_t ksys_readahead(int fd, loff_t offset, size_t count) |
782182e5 CW |
580 | { |
581 | ssize_t ret; | |
2903ff01 | 582 | struct fd f; |
782182e5 CW |
583 | |
584 | ret = -EBADF; | |
2903ff01 | 585 | f = fdget(fd); |
3d8f7615 AG |
586 | if (!f.file || !(f.file->f_mode & FMODE_READ)) |
587 | goto out; | |
588 | ||
589 | /* | |
590 | * The readahead() syscall is intended to run only on files | |
591 | * that can execute readahead. If readahead is not possible | |
592 | * on this file, then we must return -EINVAL. | |
593 | */ | |
594 | ret = -EINVAL; | |
595 | if (!f.file->f_mapping || !f.file->f_mapping->a_ops || | |
596 | !S_ISREG(file_inode(f.file)->i_mode)) | |
597 | goto out; | |
598 | ||
599 | ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED); | |
600 | out: | |
601 | fdput(f); | |
782182e5 CW |
602 | return ret; |
603 | } | |
c7b95d51 DB |
604 | |
605 | SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count) | |
606 | { | |
607 | return ksys_readahead(fd, offset, count); | |
608 | } |