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 | 116 | static void read_pages(struct readahead_control *rac, struct list_head *pages, |
c1f6925e | 117 | bool skip_page) |
1da177e4 | 118 | { |
a4d96536 | 119 | const struct address_space_operations *aops = rac->mapping->a_ops; |
c1f6925e | 120 | struct page *page; |
5b417b18 | 121 | struct blk_plug plug; |
1da177e4 | 122 | |
a4d96536 | 123 | if (!readahead_count(rac)) |
c1f6925e | 124 | goto out; |
ad4ae1c7 | 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); | |
c1f6925e MWO |
133 | rac->_index += rac->_nr_pages; |
134 | rac->_nr_pages = 0; | |
135 | } else { | |
136 | while ((page = readahead_page(rac))) { | |
a4d96536 | 137 | aops->readpage(rac->file, page); |
c1f6925e MWO |
138 | put_page(page); |
139 | } | |
1da177e4 | 140 | } |
5b417b18 | 141 | |
5b417b18 | 142 | blk_finish_plug(&plug); |
ad4ae1c7 MWO |
143 | |
144 | BUG_ON(!list_empty(pages)); | |
c1f6925e MWO |
145 | BUG_ON(readahead_count(rac)); |
146 | ||
147 | out: | |
148 | if (skip_page) | |
149 | rac->_index++; | |
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, |
08eb9658 | 159 | struct file *filp, pgoff_t index, unsigned long nr_to_read, |
c534aa3f | 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); | |
1da177e4 | 166 | loff_t isize = i_size_read(inode); |
8a5c743e | 167 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
a4d96536 MWO |
168 | struct readahead_control rac = { |
169 | .mapping = mapping, | |
170 | .file = filp, | |
c1f6925e | 171 | ._index = index, |
a4d96536 | 172 | }; |
c2c7ad74 | 173 | unsigned long i; |
1da177e4 LT |
174 | |
175 | if (isize == 0) | |
9a42823a | 176 | return; |
1da177e4 | 177 | |
09cbfeaf | 178 | end_index = ((isize - 1) >> PAGE_SHIFT); |
1da177e4 LT |
179 | |
180 | /* | |
181 | * Preallocate as many pages as we will need. | |
182 | */ | |
c2c7ad74 | 183 | for (i = 0; i < nr_to_read; i++) { |
ef8153b6 | 184 | if (index + i > end_index) |
1da177e4 LT |
185 | break; |
186 | ||
c1f6925e MWO |
187 | BUG_ON(index + i != rac._index + rac._nr_pages); |
188 | ||
ef8153b6 | 189 | page = xa_load(&mapping->i_pages, index + i); |
3159f943 | 190 | if (page && !xa_is_value(page)) { |
b3751e6a CH |
191 | /* |
192 | * Page already present? Kick off the current batch of | |
193 | * contiguous pages before continuing with the next | |
194 | * batch. | |
195 | */ | |
c1f6925e | 196 | read_pages(&rac, &page_pool, true); |
1da177e4 | 197 | continue; |
b3751e6a | 198 | } |
1da177e4 | 199 | |
8a5c743e | 200 | page = __page_cache_alloc(gfp_mask); |
1da177e4 LT |
201 | if (!page) |
202 | break; | |
c1f6925e MWO |
203 | if (mapping->a_ops->readpages) { |
204 | page->index = index + i; | |
205 | list_add(&page->lru, &page_pool); | |
206 | } else if (add_to_page_cache_lru(page, mapping, index + i, | |
207 | gfp_mask) < 0) { | |
208 | put_page(page); | |
209 | read_pages(&rac, &page_pool, true); | |
210 | continue; | |
211 | } | |
c2c7ad74 | 212 | if (i == nr_to_read - lookahead_size) |
46fc3e7b | 213 | SetPageReadahead(page); |
a4d96536 | 214 | rac._nr_pages++; |
1da177e4 | 215 | } |
1da177e4 LT |
216 | |
217 | /* | |
218 | * Now start the IO. We ignore I/O errors - if the page is not | |
219 | * uptodate then the caller will launch readpage again, and | |
220 | * will then handle the error. | |
221 | */ | |
c1f6925e | 222 | read_pages(&rac, &page_pool, false); |
1da177e4 LT |
223 | } |
224 | ||
225 | /* | |
226 | * Chunk the readahead into 2 megabyte units, so that we don't pin too much | |
227 | * memory at once. | |
228 | */ | |
9a42823a | 229 | void force_page_cache_readahead(struct address_space *mapping, |
08eb9658 | 230 | struct file *filp, pgoff_t index, unsigned long nr_to_read) |
1da177e4 | 231 | { |
9491ae4a JA |
232 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
233 | struct file_ra_state *ra = &filp->f_ra; | |
234 | unsigned long max_pages; | |
235 | ||
1da177e4 | 236 | if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages)) |
9a42823a | 237 | return; |
1da177e4 | 238 | |
9491ae4a JA |
239 | /* |
240 | * If the request exceeds the readahead window, allow the read to | |
241 | * be up to the optimal hardware IO size | |
242 | */ | |
243 | max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages); | |
244 | nr_to_read = min(nr_to_read, max_pages); | |
1da177e4 | 245 | while (nr_to_read) { |
09cbfeaf | 246 | unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE; |
1da177e4 LT |
247 | |
248 | if (this_chunk > nr_to_read) | |
249 | this_chunk = nr_to_read; | |
08eb9658 | 250 | __do_page_cache_readahead(mapping, filp, index, this_chunk, 0); |
58d5640e | 251 | |
08eb9658 | 252 | index += this_chunk; |
1da177e4 LT |
253 | nr_to_read -= this_chunk; |
254 | } | |
1da177e4 LT |
255 | } |
256 | ||
c743d96b FW |
257 | /* |
258 | * Set the initial window size, round to next power of 2 and square | |
259 | * for small size, x 4 for medium, and x 2 for large | |
260 | * for 128k (32 page) max ra | |
261 | * 1-8 page = 32k initial, > 8 page = 128k initial | |
262 | */ | |
263 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) | |
264 | { | |
265 | unsigned long newsize = roundup_pow_of_two(size); | |
266 | ||
267 | if (newsize <= max / 32) | |
268 | newsize = newsize * 4; | |
269 | else if (newsize <= max / 4) | |
270 | newsize = newsize * 2; | |
271 | else | |
272 | newsize = max; | |
273 | ||
274 | return newsize; | |
275 | } | |
276 | ||
122a21d1 FW |
277 | /* |
278 | * Get the previous window size, ramp it up, and | |
279 | * return it as the new window size. | |
280 | */ | |
c743d96b | 281 | static unsigned long get_next_ra_size(struct file_ra_state *ra, |
20ff1c95 | 282 | unsigned long max) |
122a21d1 | 283 | { |
f9acc8c7 | 284 | unsigned long cur = ra->size; |
122a21d1 FW |
285 | |
286 | if (cur < max / 16) | |
20ff1c95 GX |
287 | return 4 * cur; |
288 | if (cur <= max / 2) | |
289 | return 2 * cur; | |
290 | return max; | |
122a21d1 FW |
291 | } |
292 | ||
293 | /* | |
294 | * On-demand readahead design. | |
295 | * | |
296 | * The fields in struct file_ra_state represent the most-recently-executed | |
297 | * readahead attempt: | |
298 | * | |
f9acc8c7 FW |
299 | * |<----- async_size ---------| |
300 | * |------------------- size -------------------->| | |
301 | * |==================#===========================| | |
302 | * ^start ^page marked with PG_readahead | |
122a21d1 FW |
303 | * |
304 | * To overlap application thinking time and disk I/O time, we do | |
305 | * `readahead pipelining': Do not wait until the application consumed all | |
306 | * readahead pages and stalled on the missing page at readahead_index; | |
f9acc8c7 FW |
307 | * Instead, submit an asynchronous readahead I/O as soon as there are |
308 | * only async_size pages left in the readahead window. Normally async_size | |
309 | * will be equal to size, for maximum pipelining. | |
122a21d1 FW |
310 | * |
311 | * In interleaved sequential reads, concurrent streams on the same fd can | |
312 | * be invalidating each other's readahead state. So we flag the new readahead | |
f9acc8c7 | 313 | * page at (start+size-async_size) with PG_readahead, and use it as readahead |
122a21d1 FW |
314 | * indicator. The flag won't be set on already cached pages, to avoid the |
315 | * readahead-for-nothing fuss, saving pointless page cache lookups. | |
316 | * | |
f4e6b498 | 317 | * prev_pos tracks the last visited byte in the _previous_ read request. |
122a21d1 FW |
318 | * It should be maintained by the caller, and will be used for detecting |
319 | * small random reads. Note that the readahead algorithm checks loosely | |
320 | * for sequential patterns. Hence interleaved reads might be served as | |
321 | * sequential ones. | |
322 | * | |
323 | * There is a special-case: if the first page which the application tries to | |
324 | * read happens to be the first page of the file, it is assumed that a linear | |
325 | * read is about to happen and the window is immediately set to the initial size | |
326 | * based on I/O request size and the max_readahead. | |
327 | * | |
328 | * The code ramps up the readahead size aggressively at first, but slow down as | |
329 | * it approaches max_readhead. | |
330 | */ | |
331 | ||
10be0b37 | 332 | /* |
08eb9658 | 333 | * Count contiguously cached pages from @index-1 to @index-@max, |
10be0b37 WF |
334 | * this count is a conservative estimation of |
335 | * - length of the sequential read sequence, or | |
336 | * - thrashing threshold in memory tight systems | |
337 | */ | |
338 | static pgoff_t count_history_pages(struct address_space *mapping, | |
08eb9658 | 339 | pgoff_t index, unsigned long max) |
10be0b37 WF |
340 | { |
341 | pgoff_t head; | |
342 | ||
343 | rcu_read_lock(); | |
08eb9658 | 344 | head = page_cache_prev_miss(mapping, index - 1, max); |
10be0b37 WF |
345 | rcu_read_unlock(); |
346 | ||
08eb9658 | 347 | return index - 1 - head; |
10be0b37 WF |
348 | } |
349 | ||
350 | /* | |
351 | * page cache context based read-ahead | |
352 | */ | |
353 | static int try_context_readahead(struct address_space *mapping, | |
354 | struct file_ra_state *ra, | |
08eb9658 | 355 | pgoff_t index, |
10be0b37 WF |
356 | unsigned long req_size, |
357 | unsigned long max) | |
358 | { | |
359 | pgoff_t size; | |
360 | ||
08eb9658 | 361 | size = count_history_pages(mapping, index, max); |
10be0b37 WF |
362 | |
363 | /* | |
2cad4018 | 364 | * not enough history pages: |
10be0b37 WF |
365 | * it could be a random read |
366 | */ | |
2cad4018 | 367 | if (size <= req_size) |
10be0b37 WF |
368 | return 0; |
369 | ||
370 | /* | |
371 | * starts from beginning of file: | |
372 | * it is a strong indication of long-run stream (or whole-file-read) | |
373 | */ | |
08eb9658 | 374 | if (size >= index) |
10be0b37 WF |
375 | size *= 2; |
376 | ||
08eb9658 | 377 | ra->start = index; |
2cad4018 FW |
378 | ra->size = min(size + req_size, max); |
379 | ra->async_size = 1; | |
10be0b37 WF |
380 | |
381 | return 1; | |
382 | } | |
383 | ||
122a21d1 FW |
384 | /* |
385 | * A minimal readahead algorithm for trivial sequential/random reads. | |
386 | */ | |
9a42823a MWO |
387 | static void ondemand_readahead(struct address_space *mapping, |
388 | struct file_ra_state *ra, struct file *filp, | |
08eb9658 | 389 | bool hit_readahead_marker, pgoff_t index, |
9a42823a | 390 | unsigned long req_size) |
122a21d1 | 391 | { |
9491ae4a JA |
392 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
393 | unsigned long max_pages = ra->ra_pages; | |
dc30b96a | 394 | unsigned long add_pages; |
08eb9658 | 395 | pgoff_t prev_index; |
045a2529 | 396 | |
9491ae4a JA |
397 | /* |
398 | * If the request exceeds the readahead window, allow the read to | |
399 | * be up to the optimal hardware IO size | |
400 | */ | |
401 | if (req_size > max_pages && bdi->io_pages > max_pages) | |
402 | max_pages = min(req_size, bdi->io_pages); | |
403 | ||
045a2529 WF |
404 | /* |
405 | * start of file | |
406 | */ | |
08eb9658 | 407 | if (!index) |
045a2529 | 408 | goto initial_readahead; |
122a21d1 FW |
409 | |
410 | /* | |
08eb9658 | 411 | * It's the expected callback index, assume sequential access. |
122a21d1 FW |
412 | * Ramp up sizes, and push forward the readahead window. |
413 | */ | |
08eb9658 MWO |
414 | if ((index == (ra->start + ra->size - ra->async_size) || |
415 | index == (ra->start + ra->size))) { | |
f9acc8c7 | 416 | ra->start += ra->size; |
9491ae4a | 417 | ra->size = get_next_ra_size(ra, max_pages); |
f9acc8c7 FW |
418 | ra->async_size = ra->size; |
419 | goto readit; | |
122a21d1 FW |
420 | } |
421 | ||
6b10c6c9 FW |
422 | /* |
423 | * Hit a marked page without valid readahead state. | |
424 | * E.g. interleaved reads. | |
425 | * Query the pagecache for async_size, which normally equals to | |
426 | * readahead size. Ramp it up and use it as the new readahead size. | |
427 | */ | |
428 | if (hit_readahead_marker) { | |
429 | pgoff_t start; | |
430 | ||
30002ed2 | 431 | rcu_read_lock(); |
08eb9658 | 432 | start = page_cache_next_miss(mapping, index + 1, max_pages); |
30002ed2 | 433 | rcu_read_unlock(); |
6b10c6c9 | 434 | |
08eb9658 | 435 | if (!start || start - index > max_pages) |
9a42823a | 436 | return; |
6b10c6c9 FW |
437 | |
438 | ra->start = start; | |
08eb9658 | 439 | ra->size = start - index; /* old async_size */ |
160334a0 | 440 | ra->size += req_size; |
9491ae4a | 441 | ra->size = get_next_ra_size(ra, max_pages); |
6b10c6c9 FW |
442 | ra->async_size = ra->size; |
443 | goto readit; | |
444 | } | |
445 | ||
122a21d1 | 446 | /* |
045a2529 | 447 | * oversize read |
122a21d1 | 448 | */ |
9491ae4a | 449 | if (req_size > max_pages) |
045a2529 WF |
450 | goto initial_readahead; |
451 | ||
452 | /* | |
453 | * sequential cache miss | |
08eb9658 MWO |
454 | * trivial case: (index - prev_index) == 1 |
455 | * unaligned reads: (index - prev_index) == 0 | |
045a2529 | 456 | */ |
08eb9658 MWO |
457 | prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT; |
458 | if (index - prev_index <= 1UL) | |
045a2529 WF |
459 | goto initial_readahead; |
460 | ||
10be0b37 WF |
461 | /* |
462 | * Query the page cache and look for the traces(cached history pages) | |
463 | * that a sequential stream would leave behind. | |
464 | */ | |
08eb9658 | 465 | if (try_context_readahead(mapping, ra, index, req_size, max_pages)) |
10be0b37 WF |
466 | goto readit; |
467 | ||
045a2529 WF |
468 | /* |
469 | * standalone, small random read | |
470 | * Read as is, and do not pollute the readahead state. | |
471 | */ | |
08eb9658 | 472 | __do_page_cache_readahead(mapping, filp, index, req_size, 0); |
9a42823a | 473 | return; |
045a2529 WF |
474 | |
475 | initial_readahead: | |
08eb9658 | 476 | ra->start = index; |
9491ae4a | 477 | ra->size = get_init_ra_size(req_size, max_pages); |
f9acc8c7 | 478 | ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size; |
122a21d1 | 479 | |
f9acc8c7 | 480 | readit: |
51daa88e WF |
481 | /* |
482 | * Will this read hit the readahead marker made by itself? | |
483 | * If so, trigger the readahead marker hit now, and merge | |
484 | * the resulted next readahead window into the current one. | |
dc30b96a | 485 | * Take care of maximum IO pages as above. |
51daa88e | 486 | */ |
08eb9658 | 487 | if (index == ra->start && ra->size == ra->async_size) { |
dc30b96a MS |
488 | add_pages = get_next_ra_size(ra, max_pages); |
489 | if (ra->size + add_pages <= max_pages) { | |
490 | ra->async_size = add_pages; | |
491 | ra->size += add_pages; | |
492 | } else { | |
493 | ra->size = max_pages; | |
494 | ra->async_size = max_pages >> 1; | |
495 | } | |
51daa88e WF |
496 | } |
497 | ||
9a42823a | 498 | ra_submit(ra, mapping, filp); |
122a21d1 FW |
499 | } |
500 | ||
501 | /** | |
cf914a7d | 502 | * page_cache_sync_readahead - generic file readahead |
122a21d1 FW |
503 | * @mapping: address_space which holds the pagecache and I/O vectors |
504 | * @ra: file_ra_state which holds the readahead state | |
505 | * @filp: passed on to ->readpage() and ->readpages() | |
08eb9658 MWO |
506 | * @index: Index of first page to be read. |
507 | * @req_count: Total number of pages being read by the caller. | |
122a21d1 | 508 | * |
cf914a7d RR |
509 | * page_cache_sync_readahead() should be called when a cache miss happened: |
510 | * it will submit the read. The readahead logic may decide to piggyback more | |
511 | * pages onto the read request if access patterns suggest it will improve | |
512 | * performance. | |
122a21d1 | 513 | */ |
cf914a7d RR |
514 | void page_cache_sync_readahead(struct address_space *mapping, |
515 | struct file_ra_state *ra, struct file *filp, | |
08eb9658 | 516 | pgoff_t index, unsigned long req_count) |
122a21d1 FW |
517 | { |
518 | /* no read-ahead */ | |
519 | if (!ra->ra_pages) | |
cf914a7d RR |
520 | return; |
521 | ||
ca47e8c7 JB |
522 | if (blk_cgroup_congested()) |
523 | return; | |
524 | ||
0141450f | 525 | /* be dumb */ |
70655c06 | 526 | if (filp && (filp->f_mode & FMODE_RANDOM)) { |
08eb9658 | 527 | force_page_cache_readahead(mapping, filp, index, req_count); |
0141450f WF |
528 | return; |
529 | } | |
530 | ||
cf914a7d | 531 | /* do read-ahead */ |
08eb9658 | 532 | ondemand_readahead(mapping, ra, filp, false, index, req_count); |
cf914a7d RR |
533 | } |
534 | EXPORT_SYMBOL_GPL(page_cache_sync_readahead); | |
535 | ||
536 | /** | |
537 | * page_cache_async_readahead - file readahead for marked pages | |
538 | * @mapping: address_space which holds the pagecache and I/O vectors | |
539 | * @ra: file_ra_state which holds the readahead state | |
540 | * @filp: passed on to ->readpage() and ->readpages() | |
08eb9658 MWO |
541 | * @page: The page at @index which triggered the readahead call. |
542 | * @index: Index of first page to be read. | |
543 | * @req_count: Total number of pages being read by the caller. | |
cf914a7d | 544 | * |
bf8abe8b | 545 | * page_cache_async_readahead() should be called when a page is used which |
08eb9658 | 546 | * is marked as PageReadahead; this is a marker to suggest that the application |
cf914a7d | 547 | * has used up enough of the readahead window that we should start pulling in |
f7850d93 RD |
548 | * more pages. |
549 | */ | |
cf914a7d RR |
550 | void |
551 | page_cache_async_readahead(struct address_space *mapping, | |
552 | struct file_ra_state *ra, struct file *filp, | |
08eb9658 MWO |
553 | struct page *page, pgoff_t index, |
554 | unsigned long req_count) | |
cf914a7d RR |
555 | { |
556 | /* no read-ahead */ | |
557 | if (!ra->ra_pages) | |
558 | return; | |
559 | ||
560 | /* | |
561 | * Same bit is used for PG_readahead and PG_reclaim. | |
562 | */ | |
563 | if (PageWriteback(page)) | |
564 | return; | |
565 | ||
566 | ClearPageReadahead(page); | |
567 | ||
568 | /* | |
569 | * Defer asynchronous read-ahead on IO congestion. | |
570 | */ | |
703c2708 | 571 | if (inode_read_congested(mapping->host)) |
cf914a7d | 572 | return; |
122a21d1 | 573 | |
ca47e8c7 JB |
574 | if (blk_cgroup_congested()) |
575 | return; | |
576 | ||
122a21d1 | 577 | /* do read-ahead */ |
08eb9658 | 578 | ondemand_readahead(mapping, ra, filp, true, index, req_count); |
122a21d1 | 579 | } |
cf914a7d | 580 | EXPORT_SYMBOL_GPL(page_cache_async_readahead); |
782182e5 | 581 | |
c7b95d51 | 582 | ssize_t ksys_readahead(int fd, loff_t offset, size_t count) |
782182e5 CW |
583 | { |
584 | ssize_t ret; | |
2903ff01 | 585 | struct fd f; |
782182e5 CW |
586 | |
587 | ret = -EBADF; | |
2903ff01 | 588 | f = fdget(fd); |
3d8f7615 AG |
589 | if (!f.file || !(f.file->f_mode & FMODE_READ)) |
590 | goto out; | |
591 | ||
592 | /* | |
593 | * The readahead() syscall is intended to run only on files | |
594 | * that can execute readahead. If readahead is not possible | |
595 | * on this file, then we must return -EINVAL. | |
596 | */ | |
597 | ret = -EINVAL; | |
598 | if (!f.file->f_mapping || !f.file->f_mapping->a_ops || | |
599 | !S_ISREG(file_inode(f.file)->i_mode)) | |
600 | goto out; | |
601 | ||
602 | ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED); | |
603 | out: | |
604 | fdput(f); | |
782182e5 CW |
605 | return ret; |
606 | } | |
c7b95d51 DB |
607 | |
608 | SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count) | |
609 | { | |
610 | return ksys_readahead(fd, offset, count); | |
611 | } |