Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * mm/readahead.c - address_space-level file readahead. | |
4 | * | |
5 | * Copyright (C) 2002, Linus Torvalds | |
6 | * | |
e1f8e874 | 7 | * 09Apr2002 Andrew Morton |
1da177e4 LT |
8 | * Initial version. |
9 | */ | |
10 | ||
11 | #include <linux/kernel.h> | |
11bd969f | 12 | #include <linux/dax.h> |
5a0e3ad6 | 13 | #include <linux/gfp.h> |
b95f1b31 | 14 | #include <linux/export.h> |
1da177e4 LT |
15 | #include <linux/blkdev.h> |
16 | #include <linux/backing-dev.h> | |
8bde37f0 | 17 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 18 | #include <linux/pagevec.h> |
f5ff8422 | 19 | #include <linux/pagemap.h> |
782182e5 CW |
20 | #include <linux/syscalls.h> |
21 | #include <linux/file.h> | |
d72ee911 | 22 | #include <linux/mm_inline.h> |
ca47e8c7 | 23 | #include <linux/blk-cgroup.h> |
3d8f7615 | 24 | #include <linux/fadvise.h> |
1da177e4 | 25 | |
29f175d1 FF |
26 | #include "internal.h" |
27 | ||
1da177e4 LT |
28 | /* |
29 | * Initialise a struct file's readahead state. Assumes that the caller has | |
30 | * memset *ra to zero. | |
31 | */ | |
32 | void | |
33 | file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping) | |
34 | { | |
de1414a6 | 35 | ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages; |
f4e6b498 | 36 | ra->prev_pos = -1; |
1da177e4 | 37 | } |
d41cc702 | 38 | EXPORT_SYMBOL_GPL(file_ra_state_init); |
1da177e4 | 39 | |
03fb3d2a DH |
40 | /* |
41 | * see if a page needs releasing upon read_cache_pages() failure | |
266cf658 DH |
42 | * - the caller of read_cache_pages() may have set PG_private or PG_fscache |
43 | * before calling, such as the NFS fs marking pages that are cached locally | |
44 | * on disk, thus we need to give the fs a chance to clean up in the event of | |
45 | * an error | |
03fb3d2a DH |
46 | */ |
47 | static void read_cache_pages_invalidate_page(struct address_space *mapping, | |
48 | struct page *page) | |
49 | { | |
266cf658 | 50 | if (page_has_private(page)) { |
03fb3d2a DH |
51 | if (!trylock_page(page)) |
52 | BUG(); | |
53 | page->mapping = mapping; | |
09cbfeaf | 54 | do_invalidatepage(page, 0, PAGE_SIZE); |
03fb3d2a DH |
55 | page->mapping = NULL; |
56 | unlock_page(page); | |
57 | } | |
09cbfeaf | 58 | put_page(page); |
03fb3d2a DH |
59 | } |
60 | ||
61 | /* | |
62 | * release a list of pages, invalidating them first if need be | |
63 | */ | |
64 | static void read_cache_pages_invalidate_pages(struct address_space *mapping, | |
65 | struct list_head *pages) | |
66 | { | |
67 | struct page *victim; | |
68 | ||
69 | while (!list_empty(pages)) { | |
c8ad6302 | 70 | victim = lru_to_page(pages); |
03fb3d2a DH |
71 | list_del(&victim->lru); |
72 | read_cache_pages_invalidate_page(mapping, victim); | |
73 | } | |
74 | } | |
75 | ||
1da177e4 | 76 | /** |
bd40cdda | 77 | * read_cache_pages - populate an address space with some pages & start reads against them |
1da177e4 LT |
78 | * @mapping: the address_space |
79 | * @pages: The address of a list_head which contains the target pages. These | |
80 | * pages have their ->index populated and are otherwise uninitialised. | |
81 | * @filler: callback routine for filling a single page. | |
82 | * @data: private data for the callback routine. | |
83 | * | |
84 | * Hides the details of the LRU cache etc from the filesystems. | |
a862f68a MR |
85 | * |
86 | * Returns: %0 on success, error return by @filler otherwise | |
1da177e4 LT |
87 | */ |
88 | int read_cache_pages(struct address_space *mapping, struct list_head *pages, | |
89 | int (*filler)(void *, struct page *), void *data) | |
90 | { | |
91 | struct page *page; | |
1da177e4 LT |
92 | int ret = 0; |
93 | ||
1da177e4 | 94 | while (!list_empty(pages)) { |
c8ad6302 | 95 | page = lru_to_page(pages); |
1da177e4 | 96 | list_del(&page->lru); |
063d99b4 | 97 | if (add_to_page_cache_lru(page, mapping, page->index, |
8a5c743e | 98 | readahead_gfp_mask(mapping))) { |
03fb3d2a | 99 | read_cache_pages_invalidate_page(mapping, page); |
1da177e4 LT |
100 | continue; |
101 | } | |
09cbfeaf | 102 | put_page(page); |
eb2be189 | 103 | |
1da177e4 | 104 | ret = filler(data, page); |
eb2be189 | 105 | if (unlikely(ret)) { |
03fb3d2a | 106 | read_cache_pages_invalidate_pages(mapping, pages); |
1da177e4 LT |
107 | break; |
108 | } | |
09cbfeaf | 109 | task_io_account_read(PAGE_SIZE); |
1da177e4 | 110 | } |
1da177e4 LT |
111 | return ret; |
112 | } | |
113 | ||
114 | EXPORT_SYMBOL(read_cache_pages); | |
115 | ||
a4d96536 MWO |
116 | static void read_pages(struct readahead_control *rac, struct list_head *pages, |
117 | gfp_t gfp) | |
1da177e4 | 118 | { |
a4d96536 | 119 | const struct address_space_operations *aops = rac->mapping->a_ops; |
5b417b18 | 120 | struct blk_plug plug; |
1da177e4 | 121 | unsigned page_idx; |
1da177e4 | 122 | |
a4d96536 | 123 | if (!readahead_count(rac)) |
ad4ae1c7 MWO |
124 | return; |
125 | ||
5b417b18 JA |
126 | blk_start_plug(&plug); |
127 | ||
a4d96536 MWO |
128 | if (aops->readpages) { |
129 | aops->readpages(rac->file, rac->mapping, pages, | |
130 | readahead_count(rac)); | |
029e332e OH |
131 | /* Clean up the remaining pages */ |
132 | put_pages_list(pages); | |
1da177e4 LT |
133 | goto out; |
134 | } | |
135 | ||
a4d96536 | 136 | for (page_idx = 0; page_idx < readahead_count(rac); page_idx++) { |
c8ad6302 | 137 | struct page *page = lru_to_page(pages); |
1da177e4 | 138 | list_del(&page->lru); |
a4d96536 MWO |
139 | if (!add_to_page_cache_lru(page, rac->mapping, page->index, |
140 | gfp)) | |
141 | aops->readpage(rac->file, page); | |
09cbfeaf | 142 | put_page(page); |
1da177e4 | 143 | } |
5b417b18 | 144 | |
1da177e4 | 145 | out: |
5b417b18 | 146 | blk_finish_plug(&plug); |
ad4ae1c7 MWO |
147 | |
148 | BUG_ON(!list_empty(pages)); | |
a4d96536 | 149 | rac->_nr_pages = 0; |
1da177e4 LT |
150 | } |
151 | ||
1da177e4 | 152 | /* |
b3751e6a CH |
153 | * __do_page_cache_readahead() actually reads a chunk of disk. It allocates |
154 | * the pages first, then submits them for I/O. This avoids the very bad | |
1da177e4 LT |
155 | * behaviour which would occur if page allocations are causing VM writeback. |
156 | * We really don't want to intermingle reads and writes like that. | |
1da177e4 | 157 | */ |
9a42823a | 158 | void __do_page_cache_readahead(struct address_space *mapping, |
c534aa3f CH |
159 | struct file *filp, pgoff_t offset, unsigned long nr_to_read, |
160 | unsigned long lookahead_size) | |
1da177e4 LT |
161 | { |
162 | struct inode *inode = mapping->host; | |
163 | struct page *page; | |
164 | unsigned long end_index; /* The last page we want to read */ | |
165 | LIST_HEAD(page_pool); | |
166 | int page_idx; | |
1da177e4 | 167 | loff_t isize = i_size_read(inode); |
8a5c743e | 168 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
a4d96536 MWO |
169 | struct readahead_control rac = { |
170 | .mapping = mapping, | |
171 | .file = filp, | |
172 | }; | |
1da177e4 LT |
173 | |
174 | if (isize == 0) | |
9a42823a | 175 | return; |
1da177e4 | 176 | |
09cbfeaf | 177 | end_index = ((isize - 1) >> PAGE_SHIFT); |
1da177e4 LT |
178 | |
179 | /* | |
180 | * Preallocate as many pages as we will need. | |
181 | */ | |
1da177e4 | 182 | for (page_idx = 0; page_idx < nr_to_read; page_idx++) { |
7361f4d8 | 183 | pgoff_t page_offset = offset + page_idx; |
c743d96b | 184 | |
1da177e4 LT |
185 | if (page_offset > end_index) |
186 | break; | |
187 | ||
560d454b | 188 | page = xa_load(&mapping->i_pages, page_offset); |
3159f943 | 189 | if (page && !xa_is_value(page)) { |
b3751e6a CH |
190 | /* |
191 | * Page already present? Kick off the current batch of | |
192 | * contiguous pages before continuing with the next | |
193 | * batch. | |
194 | */ | |
a4d96536 | 195 | read_pages(&rac, &page_pool, gfp_mask); |
1da177e4 | 196 | continue; |
b3751e6a | 197 | } |
1da177e4 | 198 | |
8a5c743e | 199 | page = __page_cache_alloc(gfp_mask); |
1da177e4 LT |
200 | if (!page) |
201 | break; | |
202 | page->index = page_offset; | |
203 | list_add(&page->lru, &page_pool); | |
46fc3e7b FW |
204 | if (page_idx == nr_to_read - lookahead_size) |
205 | SetPageReadahead(page); | |
a4d96536 | 206 | rac._nr_pages++; |
1da177e4 | 207 | } |
1da177e4 LT |
208 | |
209 | /* | |
210 | * Now start the IO. We ignore I/O errors - if the page is not | |
211 | * uptodate then the caller will launch readpage again, and | |
212 | * will then handle the error. | |
213 | */ | |
a4d96536 | 214 | read_pages(&rac, &page_pool, gfp_mask); |
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 | */ | |
9a42823a MWO |
221 | void force_page_cache_readahead(struct address_space *mapping, |
222 | struct file *filp, 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 | 228 | if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages)) |
9a42823a | 229 | return; |
1da177e4 | 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 | } | |
1da177e4 LT |
247 | } |
248 | ||
c743d96b FW |
249 | /* |
250 | * Set the initial window size, round to next power of 2 and square | |
251 | * for small size, x 4 for medium, and x 2 for large | |
252 | * for 128k (32 page) max ra | |
253 | * 1-8 page = 32k initial, > 8 page = 128k initial | |
254 | */ | |
255 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) | |
256 | { | |
257 | unsigned long newsize = roundup_pow_of_two(size); | |
258 | ||
259 | if (newsize <= max / 32) | |
260 | newsize = newsize * 4; | |
261 | else if (newsize <= max / 4) | |
262 | newsize = newsize * 2; | |
263 | else | |
264 | newsize = max; | |
265 | ||
266 | return newsize; | |
267 | } | |
268 | ||
122a21d1 FW |
269 | /* |
270 | * Get the previous window size, ramp it up, and | |
271 | * return it as the new window size. | |
272 | */ | |
c743d96b | 273 | static unsigned long get_next_ra_size(struct file_ra_state *ra, |
20ff1c95 | 274 | unsigned long max) |
122a21d1 | 275 | { |
f9acc8c7 | 276 | unsigned long cur = ra->size; |
122a21d1 FW |
277 | |
278 | if (cur < max / 16) | |
20ff1c95 GX |
279 | return 4 * cur; |
280 | if (cur <= max / 2) | |
281 | return 2 * cur; | |
282 | return max; | |
122a21d1 FW |
283 | } |
284 | ||
285 | /* | |
286 | * On-demand readahead design. | |
287 | * | |
288 | * The fields in struct file_ra_state represent the most-recently-executed | |
289 | * readahead attempt: | |
290 | * | |
f9acc8c7 FW |
291 | * |<----- async_size ---------| |
292 | * |------------------- size -------------------->| | |
293 | * |==================#===========================| | |
294 | * ^start ^page marked with PG_readahead | |
122a21d1 FW |
295 | * |
296 | * To overlap application thinking time and disk I/O time, we do | |
297 | * `readahead pipelining': Do not wait until the application consumed all | |
298 | * readahead pages and stalled on the missing page at readahead_index; | |
f9acc8c7 FW |
299 | * Instead, submit an asynchronous readahead I/O as soon as there are |
300 | * only async_size pages left in the readahead window. Normally async_size | |
301 | * will be equal to size, for maximum pipelining. | |
122a21d1 FW |
302 | * |
303 | * In interleaved sequential reads, concurrent streams on the same fd can | |
304 | * be invalidating each other's readahead state. So we flag the new readahead | |
f9acc8c7 | 305 | * page at (start+size-async_size) with PG_readahead, and use it as readahead |
122a21d1 FW |
306 | * indicator. The flag won't be set on already cached pages, to avoid the |
307 | * readahead-for-nothing fuss, saving pointless page cache lookups. | |
308 | * | |
f4e6b498 | 309 | * prev_pos tracks the last visited byte in the _previous_ read request. |
122a21d1 FW |
310 | * It should be maintained by the caller, and will be used for detecting |
311 | * small random reads. Note that the readahead algorithm checks loosely | |
312 | * for sequential patterns. Hence interleaved reads might be served as | |
313 | * sequential ones. | |
314 | * | |
315 | * There is a special-case: if the first page which the application tries to | |
316 | * read happens to be the first page of the file, it is assumed that a linear | |
317 | * read is about to happen and the window is immediately set to the initial size | |
318 | * based on I/O request size and the max_readahead. | |
319 | * | |
320 | * The code ramps up the readahead size aggressively at first, but slow down as | |
321 | * it approaches max_readhead. | |
322 | */ | |
323 | ||
10be0b37 WF |
324 | /* |
325 | * Count contiguously cached pages from @offset-1 to @offset-@max, | |
326 | * this count is a conservative estimation of | |
327 | * - length of the sequential read sequence, or | |
328 | * - thrashing threshold in memory tight systems | |
329 | */ | |
330 | static pgoff_t count_history_pages(struct address_space *mapping, | |
10be0b37 WF |
331 | pgoff_t offset, unsigned long max) |
332 | { | |
333 | pgoff_t head; | |
334 | ||
335 | rcu_read_lock(); | |
0d3f9296 | 336 | head = page_cache_prev_miss(mapping, offset - 1, max); |
10be0b37 WF |
337 | rcu_read_unlock(); |
338 | ||
339 | return offset - 1 - head; | |
340 | } | |
341 | ||
342 | /* | |
343 | * page cache context based read-ahead | |
344 | */ | |
345 | static int try_context_readahead(struct address_space *mapping, | |
346 | struct file_ra_state *ra, | |
347 | pgoff_t offset, | |
348 | unsigned long req_size, | |
349 | unsigned long max) | |
350 | { | |
351 | pgoff_t size; | |
352 | ||
3e2faa08 | 353 | size = count_history_pages(mapping, offset, max); |
10be0b37 WF |
354 | |
355 | /* | |
2cad4018 | 356 | * not enough history pages: |
10be0b37 WF |
357 | * it could be a random read |
358 | */ | |
2cad4018 | 359 | if (size <= req_size) |
10be0b37 WF |
360 | return 0; |
361 | ||
362 | /* | |
363 | * starts from beginning of file: | |
364 | * it is a strong indication of long-run stream (or whole-file-read) | |
365 | */ | |
366 | if (size >= offset) | |
367 | size *= 2; | |
368 | ||
369 | ra->start = offset; | |
2cad4018 FW |
370 | ra->size = min(size + req_size, max); |
371 | ra->async_size = 1; | |
10be0b37 WF |
372 | |
373 | return 1; | |
374 | } | |
375 | ||
122a21d1 FW |
376 | /* |
377 | * A minimal readahead algorithm for trivial sequential/random reads. | |
378 | */ | |
9a42823a MWO |
379 | static void ondemand_readahead(struct address_space *mapping, |
380 | struct file_ra_state *ra, struct file *filp, | |
381 | bool hit_readahead_marker, pgoff_t offset, | |
382 | unsigned long req_size) | |
122a21d1 | 383 | { |
9491ae4a JA |
384 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
385 | unsigned long max_pages = ra->ra_pages; | |
dc30b96a | 386 | unsigned long add_pages; |
af248a0c | 387 | pgoff_t prev_offset; |
045a2529 | 388 | |
9491ae4a JA |
389 | /* |
390 | * If the request exceeds the readahead window, allow the read to | |
391 | * be up to the optimal hardware IO size | |
392 | */ | |
393 | if (req_size > max_pages && bdi->io_pages > max_pages) | |
394 | max_pages = min(req_size, bdi->io_pages); | |
395 | ||
045a2529 WF |
396 | /* |
397 | * start of file | |
398 | */ | |
399 | if (!offset) | |
400 | goto initial_readahead; | |
122a21d1 FW |
401 | |
402 | /* | |
f9acc8c7 | 403 | * It's the expected callback offset, assume sequential access. |
122a21d1 FW |
404 | * Ramp up sizes, and push forward the readahead window. |
405 | */ | |
045a2529 WF |
406 | if ((offset == (ra->start + ra->size - ra->async_size) || |
407 | offset == (ra->start + ra->size))) { | |
f9acc8c7 | 408 | ra->start += ra->size; |
9491ae4a | 409 | ra->size = get_next_ra_size(ra, max_pages); |
f9acc8c7 FW |
410 | ra->async_size = ra->size; |
411 | goto readit; | |
122a21d1 FW |
412 | } |
413 | ||
6b10c6c9 FW |
414 | /* |
415 | * Hit a marked page without valid readahead state. | |
416 | * E.g. interleaved reads. | |
417 | * Query the pagecache for async_size, which normally equals to | |
418 | * readahead size. Ramp it up and use it as the new readahead size. | |
419 | */ | |
420 | if (hit_readahead_marker) { | |
421 | pgoff_t start; | |
422 | ||
30002ed2 | 423 | rcu_read_lock(); |
0d3f9296 | 424 | start = page_cache_next_miss(mapping, offset + 1, max_pages); |
30002ed2 | 425 | rcu_read_unlock(); |
6b10c6c9 | 426 | |
9491ae4a | 427 | if (!start || start - offset > max_pages) |
9a42823a | 428 | return; |
6b10c6c9 FW |
429 | |
430 | ra->start = start; | |
431 | ra->size = start - offset; /* old async_size */ | |
160334a0 | 432 | ra->size += req_size; |
9491ae4a | 433 | ra->size = get_next_ra_size(ra, max_pages); |
6b10c6c9 FW |
434 | ra->async_size = ra->size; |
435 | goto readit; | |
436 | } | |
437 | ||
122a21d1 | 438 | /* |
045a2529 | 439 | * oversize read |
122a21d1 | 440 | */ |
9491ae4a | 441 | if (req_size > max_pages) |
045a2529 WF |
442 | goto initial_readahead; |
443 | ||
444 | /* | |
445 | * sequential cache miss | |
af248a0c DR |
446 | * trivial case: (offset - prev_offset) == 1 |
447 | * unaligned reads: (offset - prev_offset) == 0 | |
045a2529 | 448 | */ |
09cbfeaf | 449 | prev_offset = (unsigned long long)ra->prev_pos >> PAGE_SHIFT; |
af248a0c | 450 | if (offset - prev_offset <= 1UL) |
045a2529 WF |
451 | goto initial_readahead; |
452 | ||
10be0b37 WF |
453 | /* |
454 | * Query the page cache and look for the traces(cached history pages) | |
455 | * that a sequential stream would leave behind. | |
456 | */ | |
9491ae4a | 457 | if (try_context_readahead(mapping, ra, offset, req_size, max_pages)) |
10be0b37 WF |
458 | goto readit; |
459 | ||
045a2529 WF |
460 | /* |
461 | * standalone, small random read | |
462 | * Read as is, and do not pollute the readahead state. | |
463 | */ | |
9a42823a MWO |
464 | __do_page_cache_readahead(mapping, filp, offset, req_size, 0); |
465 | return; | |
045a2529 WF |
466 | |
467 | initial_readahead: | |
f9acc8c7 | 468 | ra->start = offset; |
9491ae4a | 469 | ra->size = get_init_ra_size(req_size, max_pages); |
f9acc8c7 | 470 | ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size; |
122a21d1 | 471 | |
f9acc8c7 | 472 | readit: |
51daa88e WF |
473 | /* |
474 | * Will this read hit the readahead marker made by itself? | |
475 | * If so, trigger the readahead marker hit now, and merge | |
476 | * the resulted next readahead window into the current one. | |
dc30b96a | 477 | * Take care of maximum IO pages as above. |
51daa88e WF |
478 | */ |
479 | if (offset == ra->start && ra->size == ra->async_size) { | |
dc30b96a MS |
480 | add_pages = get_next_ra_size(ra, max_pages); |
481 | if (ra->size + add_pages <= max_pages) { | |
482 | ra->async_size = add_pages; | |
483 | ra->size += add_pages; | |
484 | } else { | |
485 | ra->size = max_pages; | |
486 | ra->async_size = max_pages >> 1; | |
487 | } | |
51daa88e WF |
488 | } |
489 | ||
9a42823a | 490 | ra_submit(ra, mapping, filp); |
122a21d1 FW |
491 | } |
492 | ||
493 | /** | |
cf914a7d | 494 | * page_cache_sync_readahead - generic file readahead |
122a21d1 FW |
495 | * @mapping: address_space which holds the pagecache and I/O vectors |
496 | * @ra: file_ra_state which holds the readahead state | |
497 | * @filp: passed on to ->readpage() and ->readpages() | |
cf914a7d | 498 | * @offset: start offset into @mapping, in pagecache page-sized units |
122a21d1 | 499 | * @req_size: hint: total size of the read which the caller is performing in |
cf914a7d | 500 | * pagecache pages |
122a21d1 | 501 | * |
cf914a7d RR |
502 | * page_cache_sync_readahead() should be called when a cache miss happened: |
503 | * it will submit the read. The readahead logic may decide to piggyback more | |
504 | * pages onto the read request if access patterns suggest it will improve | |
505 | * performance. | |
122a21d1 | 506 | */ |
cf914a7d RR |
507 | void page_cache_sync_readahead(struct address_space *mapping, |
508 | struct file_ra_state *ra, struct file *filp, | |
509 | pgoff_t offset, unsigned long req_size) | |
122a21d1 FW |
510 | { |
511 | /* no read-ahead */ | |
512 | if (!ra->ra_pages) | |
cf914a7d RR |
513 | return; |
514 | ||
ca47e8c7 JB |
515 | if (blk_cgroup_congested()) |
516 | return; | |
517 | ||
0141450f | 518 | /* be dumb */ |
70655c06 | 519 | if (filp && (filp->f_mode & FMODE_RANDOM)) { |
0141450f WF |
520 | force_page_cache_readahead(mapping, filp, offset, req_size); |
521 | return; | |
522 | } | |
523 | ||
cf914a7d RR |
524 | /* do read-ahead */ |
525 | ondemand_readahead(mapping, ra, filp, false, offset, req_size); | |
526 | } | |
527 | EXPORT_SYMBOL_GPL(page_cache_sync_readahead); | |
528 | ||
529 | /** | |
530 | * page_cache_async_readahead - file readahead for marked pages | |
531 | * @mapping: address_space which holds the pagecache and I/O vectors | |
532 | * @ra: file_ra_state which holds the readahead state | |
533 | * @filp: passed on to ->readpage() and ->readpages() | |
534 | * @page: the page at @offset which has the PG_readahead flag set | |
535 | * @offset: start offset into @mapping, in pagecache page-sized units | |
536 | * @req_size: hint: total size of the read which the caller is performing in | |
537 | * pagecache pages | |
538 | * | |
bf8abe8b | 539 | * page_cache_async_readahead() should be called when a page is used which |
f7850d93 | 540 | * has the PG_readahead flag; this is a marker to suggest that the application |
cf914a7d | 541 | * has used up enough of the readahead window that we should start pulling in |
f7850d93 RD |
542 | * more pages. |
543 | */ | |
cf914a7d RR |
544 | void |
545 | page_cache_async_readahead(struct address_space *mapping, | |
546 | struct file_ra_state *ra, struct file *filp, | |
547 | struct page *page, pgoff_t offset, | |
548 | unsigned long req_size) | |
549 | { | |
550 | /* no read-ahead */ | |
551 | if (!ra->ra_pages) | |
552 | return; | |
553 | ||
554 | /* | |
555 | * Same bit is used for PG_readahead and PG_reclaim. | |
556 | */ | |
557 | if (PageWriteback(page)) | |
558 | return; | |
559 | ||
560 | ClearPageReadahead(page); | |
561 | ||
562 | /* | |
563 | * Defer asynchronous read-ahead on IO congestion. | |
564 | */ | |
703c2708 | 565 | if (inode_read_congested(mapping->host)) |
cf914a7d | 566 | return; |
122a21d1 | 567 | |
ca47e8c7 JB |
568 | if (blk_cgroup_congested()) |
569 | return; | |
570 | ||
122a21d1 | 571 | /* do read-ahead */ |
cf914a7d | 572 | ondemand_readahead(mapping, ra, filp, true, offset, req_size); |
122a21d1 | 573 | } |
cf914a7d | 574 | EXPORT_SYMBOL_GPL(page_cache_async_readahead); |
782182e5 | 575 | |
c7b95d51 | 576 | ssize_t ksys_readahead(int fd, loff_t offset, size_t count) |
782182e5 CW |
577 | { |
578 | ssize_t ret; | |
2903ff01 | 579 | struct fd f; |
782182e5 CW |
580 | |
581 | ret = -EBADF; | |
2903ff01 | 582 | f = fdget(fd); |
3d8f7615 AG |
583 | if (!f.file || !(f.file->f_mode & FMODE_READ)) |
584 | goto out; | |
585 | ||
586 | /* | |
587 | * The readahead() syscall is intended to run only on files | |
588 | * that can execute readahead. If readahead is not possible | |
589 | * on this file, then we must return -EINVAL. | |
590 | */ | |
591 | ret = -EINVAL; | |
592 | if (!f.file->f_mapping || !f.file->f_mapping->a_ops || | |
593 | !S_ISREG(file_inode(f.file)->i_mode)) | |
594 | goto out; | |
595 | ||
596 | ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED); | |
597 | out: | |
598 | fdput(f); | |
782182e5 CW |
599 | return ret; |
600 | } | |
c7b95d51 DB |
601 | |
602 | SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count) | |
603 | { | |
604 | return ksys_readahead(fd, offset, count); | |
605 | } |