]>
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> |
f2c817be | 25 | #include <linux/sched/mm.h> |
1da177e4 | 26 | |
29f175d1 FF |
27 | #include "internal.h" |
28 | ||
1da177e4 LT |
29 | /* |
30 | * Initialise a struct file's readahead state. Assumes that the caller has | |
31 | * memset *ra to zero. | |
32 | */ | |
33 | void | |
34 | file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping) | |
35 | { | |
de1414a6 | 36 | ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages; |
f4e6b498 | 37 | ra->prev_pos = -1; |
1da177e4 | 38 | } |
d41cc702 | 39 | EXPORT_SYMBOL_GPL(file_ra_state_init); |
1da177e4 | 40 | |
03fb3d2a DH |
41 | /* |
42 | * see if a page needs releasing upon read_cache_pages() failure | |
266cf658 DH |
43 | * - the caller of read_cache_pages() may have set PG_private or PG_fscache |
44 | * before calling, such as the NFS fs marking pages that are cached locally | |
45 | * on disk, thus we need to give the fs a chance to clean up in the event of | |
46 | * an error | |
03fb3d2a DH |
47 | */ |
48 | static void read_cache_pages_invalidate_page(struct address_space *mapping, | |
49 | struct page *page) | |
50 | { | |
266cf658 | 51 | if (page_has_private(page)) { |
03fb3d2a DH |
52 | if (!trylock_page(page)) |
53 | BUG(); | |
54 | page->mapping = mapping; | |
09cbfeaf | 55 | do_invalidatepage(page, 0, PAGE_SIZE); |
03fb3d2a DH |
56 | page->mapping = NULL; |
57 | unlock_page(page); | |
58 | } | |
09cbfeaf | 59 | put_page(page); |
03fb3d2a DH |
60 | } |
61 | ||
62 | /* | |
63 | * release a list of pages, invalidating them first if need be | |
64 | */ | |
65 | static void read_cache_pages_invalidate_pages(struct address_space *mapping, | |
66 | struct list_head *pages) | |
67 | { | |
68 | struct page *victim; | |
69 | ||
70 | while (!list_empty(pages)) { | |
c8ad6302 | 71 | victim = lru_to_page(pages); |
03fb3d2a DH |
72 | list_del(&victim->lru); |
73 | read_cache_pages_invalidate_page(mapping, victim); | |
74 | } | |
75 | } | |
76 | ||
1da177e4 | 77 | /** |
bd40cdda | 78 | * read_cache_pages - populate an address space with some pages & start reads against them |
1da177e4 LT |
79 | * @mapping: the address_space |
80 | * @pages: The address of a list_head which contains the target pages. These | |
81 | * pages have their ->index populated and are otherwise uninitialised. | |
82 | * @filler: callback routine for filling a single page. | |
83 | * @data: private data for the callback routine. | |
84 | * | |
85 | * Hides the details of the LRU cache etc from the filesystems. | |
a862f68a MR |
86 | * |
87 | * Returns: %0 on success, error return by @filler otherwise | |
1da177e4 LT |
88 | */ |
89 | int read_cache_pages(struct address_space *mapping, struct list_head *pages, | |
90 | int (*filler)(void *, struct page *), void *data) | |
91 | { | |
92 | struct page *page; | |
1da177e4 LT |
93 | int ret = 0; |
94 | ||
1da177e4 | 95 | while (!list_empty(pages)) { |
c8ad6302 | 96 | page = lru_to_page(pages); |
1da177e4 | 97 | list_del(&page->lru); |
063d99b4 | 98 | if (add_to_page_cache_lru(page, mapping, page->index, |
8a5c743e | 99 | readahead_gfp_mask(mapping))) { |
03fb3d2a | 100 | read_cache_pages_invalidate_page(mapping, page); |
1da177e4 LT |
101 | continue; |
102 | } | |
09cbfeaf | 103 | put_page(page); |
eb2be189 | 104 | |
1da177e4 | 105 | ret = filler(data, page); |
eb2be189 | 106 | if (unlikely(ret)) { |
03fb3d2a | 107 | read_cache_pages_invalidate_pages(mapping, pages); |
1da177e4 LT |
108 | break; |
109 | } | |
09cbfeaf | 110 | task_io_account_read(PAGE_SIZE); |
1da177e4 | 111 | } |
1da177e4 LT |
112 | return ret; |
113 | } | |
114 | ||
115 | EXPORT_SYMBOL(read_cache_pages); | |
116 | ||
a4d96536 | 117 | static void read_pages(struct readahead_control *rac, struct list_head *pages, |
c1f6925e | 118 | bool skip_page) |
1da177e4 | 119 | { |
a4d96536 | 120 | const struct address_space_operations *aops = rac->mapping->a_ops; |
c1f6925e | 121 | struct page *page; |
5b417b18 | 122 | struct blk_plug plug; |
1da177e4 | 123 | |
a4d96536 | 124 | if (!readahead_count(rac)) |
c1f6925e | 125 | goto out; |
ad4ae1c7 | 126 | |
5b417b18 JA |
127 | blk_start_plug(&plug); |
128 | ||
8151b4c8 MWO |
129 | if (aops->readahead) { |
130 | aops->readahead(rac); | |
131 | /* Clean up the remaining pages */ | |
132 | while ((page = readahead_page(rac))) { | |
133 | unlock_page(page); | |
134 | put_page(page); | |
135 | } | |
136 | } else if (aops->readpages) { | |
a4d96536 MWO |
137 | aops->readpages(rac->file, rac->mapping, pages, |
138 | readahead_count(rac)); | |
029e332e OH |
139 | /* Clean up the remaining pages */ |
140 | put_pages_list(pages); | |
c1f6925e MWO |
141 | rac->_index += rac->_nr_pages; |
142 | rac->_nr_pages = 0; | |
143 | } else { | |
144 | while ((page = readahead_page(rac))) { | |
a4d96536 | 145 | aops->readpage(rac->file, page); |
c1f6925e MWO |
146 | put_page(page); |
147 | } | |
1da177e4 | 148 | } |
5b417b18 | 149 | |
5b417b18 | 150 | blk_finish_plug(&plug); |
ad4ae1c7 MWO |
151 | |
152 | BUG_ON(!list_empty(pages)); | |
c1f6925e MWO |
153 | BUG_ON(readahead_count(rac)); |
154 | ||
155 | out: | |
156 | if (skip_page) | |
157 | rac->_index++; | |
1da177e4 LT |
158 | } |
159 | ||
2c684234 | 160 | /** |
73bb49da MWO |
161 | * page_cache_ra_unbounded - Start unchecked readahead. |
162 | * @ractl: Readahead control. | |
2c684234 MWO |
163 | * @nr_to_read: The number of pages to read. |
164 | * @lookahead_size: Where to start the next readahead. | |
165 | * | |
166 | * This function is for filesystems to call when they want to start | |
167 | * readahead beyond a file's stated i_size. This is almost certainly | |
168 | * not the function you want to call. Use page_cache_async_readahead() | |
169 | * or page_cache_sync_readahead() instead. | |
170 | * | |
171 | * Context: File is referenced by caller. Mutexes may be held by caller. | |
172 | * May sleep, but will not reenter filesystem to reclaim memory. | |
1da177e4 | 173 | */ |
73bb49da MWO |
174 | void page_cache_ra_unbounded(struct readahead_control *ractl, |
175 | unsigned long nr_to_read, unsigned long lookahead_size) | |
1da177e4 | 176 | { |
73bb49da MWO |
177 | struct address_space *mapping = ractl->mapping; |
178 | unsigned long index = readahead_index(ractl); | |
1da177e4 | 179 | LIST_HEAD(page_pool); |
8a5c743e | 180 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
c2c7ad74 | 181 | unsigned long i; |
1da177e4 | 182 | |
f2c817be MWO |
183 | /* |
184 | * Partway through the readahead operation, we will have added | |
185 | * locked pages to the page cache, but will not yet have submitted | |
186 | * them for I/O. Adding another page may need to allocate memory, | |
187 | * which can trigger memory reclaim. Telling the VM we're in | |
188 | * the middle of a filesystem operation will cause it to not | |
189 | * touch file-backed pages, preventing a deadlock. Most (all?) | |
190 | * filesystems already specify __GFP_NOFS in their mapping's | |
191 | * gfp_mask, but let's be explicit here. | |
192 | */ | |
193 | unsigned int nofs = memalloc_nofs_save(); | |
194 | ||
730633f0 | 195 | filemap_invalidate_lock_shared(mapping); |
1da177e4 LT |
196 | /* |
197 | * Preallocate as many pages as we will need. | |
198 | */ | |
c2c7ad74 | 199 | for (i = 0; i < nr_to_read; i++) { |
b0f31d78 | 200 | struct page *page = xa_load(&mapping->i_pages, index + i); |
1da177e4 | 201 | |
3159f943 | 202 | if (page && !xa_is_value(page)) { |
b3751e6a | 203 | /* |
2d8163e4 MWO |
204 | * Page already present? Kick off the current batch |
205 | * of contiguous pages before continuing with the | |
206 | * next batch. This page may be the one we would | |
207 | * have intended to mark as Readahead, but we don't | |
208 | * have a stable reference to this page, and it's | |
209 | * not worth getting one just for that. | |
b3751e6a | 210 | */ |
73bb49da | 211 | read_pages(ractl, &page_pool, true); |
f615bd5c | 212 | i = ractl->_index + ractl->_nr_pages - index - 1; |
1da177e4 | 213 | continue; |
b3751e6a | 214 | } |
1da177e4 | 215 | |
8a5c743e | 216 | page = __page_cache_alloc(gfp_mask); |
1da177e4 LT |
217 | if (!page) |
218 | break; | |
c1f6925e MWO |
219 | if (mapping->a_ops->readpages) { |
220 | page->index = index + i; | |
221 | list_add(&page->lru, &page_pool); | |
222 | } else if (add_to_page_cache_lru(page, mapping, index + i, | |
223 | gfp_mask) < 0) { | |
224 | put_page(page); | |
73bb49da | 225 | read_pages(ractl, &page_pool, true); |
f615bd5c | 226 | i = ractl->_index + ractl->_nr_pages - index - 1; |
c1f6925e MWO |
227 | continue; |
228 | } | |
c2c7ad74 | 229 | if (i == nr_to_read - lookahead_size) |
46fc3e7b | 230 | SetPageReadahead(page); |
73bb49da | 231 | ractl->_nr_pages++; |
1da177e4 | 232 | } |
1da177e4 LT |
233 | |
234 | /* | |
235 | * Now start the IO. We ignore I/O errors - if the page is not | |
236 | * uptodate then the caller will launch readpage again, and | |
237 | * will then handle the error. | |
238 | */ | |
73bb49da | 239 | read_pages(ractl, &page_pool, false); |
730633f0 | 240 | filemap_invalidate_unlock_shared(mapping); |
f2c817be | 241 | memalloc_nofs_restore(nofs); |
1da177e4 | 242 | } |
73bb49da | 243 | EXPORT_SYMBOL_GPL(page_cache_ra_unbounded); |
2c684234 MWO |
244 | |
245 | /* | |
8238287e | 246 | * do_page_cache_ra() actually reads a chunk of disk. It allocates |
2c684234 MWO |
247 | * the pages first, then submits them for I/O. This avoids the very bad |
248 | * behaviour which would occur if page allocations are causing VM writeback. | |
249 | * We really don't want to intermingle reads and writes like that. | |
250 | */ | |
8238287e MWO |
251 | void do_page_cache_ra(struct readahead_control *ractl, |
252 | unsigned long nr_to_read, unsigned long lookahead_size) | |
2c684234 | 253 | { |
8238287e MWO |
254 | struct inode *inode = ractl->mapping->host; |
255 | unsigned long index = readahead_index(ractl); | |
2c684234 MWO |
256 | loff_t isize = i_size_read(inode); |
257 | pgoff_t end_index; /* The last page we want to read */ | |
258 | ||
259 | if (isize == 0) | |
260 | return; | |
261 | ||
262 | end_index = (isize - 1) >> PAGE_SHIFT; | |
263 | if (index > end_index) | |
264 | return; | |
265 | /* Don't read past the page containing the last byte of the file */ | |
266 | if (nr_to_read > end_index - index) | |
267 | nr_to_read = end_index - index + 1; | |
268 | ||
8238287e | 269 | page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size); |
2c684234 | 270 | } |
1da177e4 LT |
271 | |
272 | /* | |
273 | * Chunk the readahead into 2 megabyte units, so that we don't pin too much | |
274 | * memory at once. | |
275 | */ | |
7b3df3b9 | 276 | void force_page_cache_ra(struct readahead_control *ractl, |
fcd9ae4f | 277 | unsigned long nr_to_read) |
1da177e4 | 278 | { |
7b3df3b9 | 279 | struct address_space *mapping = ractl->mapping; |
fcd9ae4f | 280 | struct file_ra_state *ra = ractl->ra; |
9491ae4a | 281 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
7b3df3b9 | 282 | unsigned long max_pages, index; |
9491ae4a | 283 | |
8151b4c8 MWO |
284 | if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages && |
285 | !mapping->a_ops->readahead)) | |
9a42823a | 286 | return; |
1da177e4 | 287 | |
9491ae4a JA |
288 | /* |
289 | * If the request exceeds the readahead window, allow the read to | |
290 | * be up to the optimal hardware IO size | |
291 | */ | |
7b3df3b9 | 292 | index = readahead_index(ractl); |
9491ae4a | 293 | max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages); |
7b3df3b9 | 294 | nr_to_read = min_t(unsigned long, nr_to_read, max_pages); |
1da177e4 | 295 | while (nr_to_read) { |
09cbfeaf | 296 | unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE; |
1da177e4 LT |
297 | |
298 | if (this_chunk > nr_to_read) | |
299 | this_chunk = nr_to_read; | |
7b3df3b9 DH |
300 | ractl->_index = index; |
301 | do_page_cache_ra(ractl, this_chunk, 0); | |
58d5640e | 302 | |
08eb9658 | 303 | index += this_chunk; |
1da177e4 LT |
304 | nr_to_read -= this_chunk; |
305 | } | |
1da177e4 LT |
306 | } |
307 | ||
c743d96b FW |
308 | /* |
309 | * Set the initial window size, round to next power of 2 and square | |
310 | * for small size, x 4 for medium, and x 2 for large | |
311 | * for 128k (32 page) max ra | |
312 | * 1-8 page = 32k initial, > 8 page = 128k initial | |
313 | */ | |
314 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) | |
315 | { | |
316 | unsigned long newsize = roundup_pow_of_two(size); | |
317 | ||
318 | if (newsize <= max / 32) | |
319 | newsize = newsize * 4; | |
320 | else if (newsize <= max / 4) | |
321 | newsize = newsize * 2; | |
322 | else | |
323 | newsize = max; | |
324 | ||
325 | return newsize; | |
326 | } | |
327 | ||
122a21d1 FW |
328 | /* |
329 | * Get the previous window size, ramp it up, and | |
330 | * return it as the new window size. | |
331 | */ | |
c743d96b | 332 | static unsigned long get_next_ra_size(struct file_ra_state *ra, |
20ff1c95 | 333 | unsigned long max) |
122a21d1 | 334 | { |
f9acc8c7 | 335 | unsigned long cur = ra->size; |
122a21d1 FW |
336 | |
337 | if (cur < max / 16) | |
20ff1c95 GX |
338 | return 4 * cur; |
339 | if (cur <= max / 2) | |
340 | return 2 * cur; | |
341 | return max; | |
122a21d1 FW |
342 | } |
343 | ||
344 | /* | |
345 | * On-demand readahead design. | |
346 | * | |
347 | * The fields in struct file_ra_state represent the most-recently-executed | |
348 | * readahead attempt: | |
349 | * | |
f9acc8c7 FW |
350 | * |<----- async_size ---------| |
351 | * |------------------- size -------------------->| | |
352 | * |==================#===========================| | |
353 | * ^start ^page marked with PG_readahead | |
122a21d1 FW |
354 | * |
355 | * To overlap application thinking time and disk I/O time, we do | |
356 | * `readahead pipelining': Do not wait until the application consumed all | |
357 | * readahead pages and stalled on the missing page at readahead_index; | |
f9acc8c7 FW |
358 | * Instead, submit an asynchronous readahead I/O as soon as there are |
359 | * only async_size pages left in the readahead window. Normally async_size | |
360 | * will be equal to size, for maximum pipelining. | |
122a21d1 FW |
361 | * |
362 | * In interleaved sequential reads, concurrent streams on the same fd can | |
363 | * be invalidating each other's readahead state. So we flag the new readahead | |
f9acc8c7 | 364 | * page at (start+size-async_size) with PG_readahead, and use it as readahead |
122a21d1 FW |
365 | * indicator. The flag won't be set on already cached pages, to avoid the |
366 | * readahead-for-nothing fuss, saving pointless page cache lookups. | |
367 | * | |
f4e6b498 | 368 | * prev_pos tracks the last visited byte in the _previous_ read request. |
122a21d1 FW |
369 | * It should be maintained by the caller, and will be used for detecting |
370 | * small random reads. Note that the readahead algorithm checks loosely | |
371 | * for sequential patterns. Hence interleaved reads might be served as | |
372 | * sequential ones. | |
373 | * | |
374 | * There is a special-case: if the first page which the application tries to | |
375 | * read happens to be the first page of the file, it is assumed that a linear | |
376 | * read is about to happen and the window is immediately set to the initial size | |
377 | * based on I/O request size and the max_readahead. | |
378 | * | |
379 | * The code ramps up the readahead size aggressively at first, but slow down as | |
380 | * it approaches max_readhead. | |
381 | */ | |
382 | ||
10be0b37 | 383 | /* |
08eb9658 | 384 | * Count contiguously cached pages from @index-1 to @index-@max, |
10be0b37 WF |
385 | * this count is a conservative estimation of |
386 | * - length of the sequential read sequence, or | |
387 | * - thrashing threshold in memory tight systems | |
388 | */ | |
389 | static pgoff_t count_history_pages(struct address_space *mapping, | |
08eb9658 | 390 | pgoff_t index, unsigned long max) |
10be0b37 WF |
391 | { |
392 | pgoff_t head; | |
393 | ||
394 | rcu_read_lock(); | |
08eb9658 | 395 | head = page_cache_prev_miss(mapping, index - 1, max); |
10be0b37 WF |
396 | rcu_read_unlock(); |
397 | ||
08eb9658 | 398 | return index - 1 - head; |
10be0b37 WF |
399 | } |
400 | ||
401 | /* | |
402 | * page cache context based read-ahead | |
403 | */ | |
404 | static int try_context_readahead(struct address_space *mapping, | |
405 | struct file_ra_state *ra, | |
08eb9658 | 406 | pgoff_t index, |
10be0b37 WF |
407 | unsigned long req_size, |
408 | unsigned long max) | |
409 | { | |
410 | pgoff_t size; | |
411 | ||
08eb9658 | 412 | size = count_history_pages(mapping, index, max); |
10be0b37 WF |
413 | |
414 | /* | |
2cad4018 | 415 | * not enough history pages: |
10be0b37 WF |
416 | * it could be a random read |
417 | */ | |
2cad4018 | 418 | if (size <= req_size) |
10be0b37 WF |
419 | return 0; |
420 | ||
421 | /* | |
422 | * starts from beginning of file: | |
423 | * it is a strong indication of long-run stream (or whole-file-read) | |
424 | */ | |
08eb9658 | 425 | if (size >= index) |
10be0b37 WF |
426 | size *= 2; |
427 | ||
08eb9658 | 428 | ra->start = index; |
2cad4018 FW |
429 | ra->size = min(size + req_size, max); |
430 | ra->async_size = 1; | |
10be0b37 WF |
431 | |
432 | return 1; | |
433 | } | |
434 | ||
122a21d1 FW |
435 | /* |
436 | * A minimal readahead algorithm for trivial sequential/random reads. | |
437 | */ | |
6e4af69a | 438 | static void ondemand_readahead(struct readahead_control *ractl, |
fcd9ae4f | 439 | bool hit_readahead_marker, unsigned long req_size) |
122a21d1 | 440 | { |
6e4af69a | 441 | struct backing_dev_info *bdi = inode_to_bdi(ractl->mapping->host); |
fcd9ae4f | 442 | struct file_ra_state *ra = ractl->ra; |
9491ae4a | 443 | unsigned long max_pages = ra->ra_pages; |
dc30b96a | 444 | unsigned long add_pages; |
6e4af69a | 445 | unsigned long index = readahead_index(ractl); |
08eb9658 | 446 | pgoff_t prev_index; |
045a2529 | 447 | |
9491ae4a JA |
448 | /* |
449 | * If the request exceeds the readahead window, allow the read to | |
450 | * be up to the optimal hardware IO size | |
451 | */ | |
452 | if (req_size > max_pages && bdi->io_pages > max_pages) | |
453 | max_pages = min(req_size, bdi->io_pages); | |
454 | ||
045a2529 WF |
455 | /* |
456 | * start of file | |
457 | */ | |
08eb9658 | 458 | if (!index) |
045a2529 | 459 | goto initial_readahead; |
122a21d1 FW |
460 | |
461 | /* | |
08eb9658 | 462 | * It's the expected callback index, assume sequential access. |
122a21d1 FW |
463 | * Ramp up sizes, and push forward the readahead window. |
464 | */ | |
08eb9658 MWO |
465 | if ((index == (ra->start + ra->size - ra->async_size) || |
466 | index == (ra->start + ra->size))) { | |
f9acc8c7 | 467 | ra->start += ra->size; |
9491ae4a | 468 | ra->size = get_next_ra_size(ra, max_pages); |
f9acc8c7 FW |
469 | ra->async_size = ra->size; |
470 | goto readit; | |
122a21d1 FW |
471 | } |
472 | ||
6b10c6c9 FW |
473 | /* |
474 | * Hit a marked page without valid readahead state. | |
475 | * E.g. interleaved reads. | |
476 | * Query the pagecache for async_size, which normally equals to | |
477 | * readahead size. Ramp it up and use it as the new readahead size. | |
478 | */ | |
479 | if (hit_readahead_marker) { | |
480 | pgoff_t start; | |
481 | ||
30002ed2 | 482 | rcu_read_lock(); |
6e4af69a DH |
483 | start = page_cache_next_miss(ractl->mapping, index + 1, |
484 | max_pages); | |
30002ed2 | 485 | rcu_read_unlock(); |
6b10c6c9 | 486 | |
08eb9658 | 487 | if (!start || start - index > max_pages) |
9a42823a | 488 | return; |
6b10c6c9 FW |
489 | |
490 | ra->start = start; | |
08eb9658 | 491 | ra->size = start - index; /* old async_size */ |
160334a0 | 492 | ra->size += req_size; |
9491ae4a | 493 | ra->size = get_next_ra_size(ra, max_pages); |
6b10c6c9 FW |
494 | ra->async_size = ra->size; |
495 | goto readit; | |
496 | } | |
497 | ||
122a21d1 | 498 | /* |
045a2529 | 499 | * oversize read |
122a21d1 | 500 | */ |
9491ae4a | 501 | if (req_size > max_pages) |
045a2529 WF |
502 | goto initial_readahead; |
503 | ||
504 | /* | |
505 | * sequential cache miss | |
08eb9658 MWO |
506 | * trivial case: (index - prev_index) == 1 |
507 | * unaligned reads: (index - prev_index) == 0 | |
045a2529 | 508 | */ |
08eb9658 MWO |
509 | prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT; |
510 | if (index - prev_index <= 1UL) | |
045a2529 WF |
511 | goto initial_readahead; |
512 | ||
10be0b37 WF |
513 | /* |
514 | * Query the page cache and look for the traces(cached history pages) | |
515 | * that a sequential stream would leave behind. | |
516 | */ | |
6e4af69a DH |
517 | if (try_context_readahead(ractl->mapping, ra, index, req_size, |
518 | max_pages)) | |
10be0b37 WF |
519 | goto readit; |
520 | ||
045a2529 WF |
521 | /* |
522 | * standalone, small random read | |
523 | * Read as is, and do not pollute the readahead state. | |
524 | */ | |
6e4af69a | 525 | do_page_cache_ra(ractl, req_size, 0); |
9a42823a | 526 | return; |
045a2529 WF |
527 | |
528 | initial_readahead: | |
08eb9658 | 529 | ra->start = index; |
9491ae4a | 530 | ra->size = get_init_ra_size(req_size, max_pages); |
f9acc8c7 | 531 | ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size; |
122a21d1 | 532 | |
f9acc8c7 | 533 | readit: |
51daa88e WF |
534 | /* |
535 | * Will this read hit the readahead marker made by itself? | |
536 | * If so, trigger the readahead marker hit now, and merge | |
537 | * the resulted next readahead window into the current one. | |
dc30b96a | 538 | * Take care of maximum IO pages as above. |
51daa88e | 539 | */ |
08eb9658 | 540 | if (index == ra->start && ra->size == ra->async_size) { |
dc30b96a MS |
541 | add_pages = get_next_ra_size(ra, max_pages); |
542 | if (ra->size + add_pages <= max_pages) { | |
543 | ra->async_size = add_pages; | |
544 | ra->size += add_pages; | |
545 | } else { | |
546 | ra->size = max_pages; | |
547 | ra->async_size = max_pages >> 1; | |
548 | } | |
51daa88e WF |
549 | } |
550 | ||
6e4af69a DH |
551 | ractl->_index = ra->start; |
552 | do_page_cache_ra(ractl, ra->size, ra->async_size); | |
122a21d1 FW |
553 | } |
554 | ||
fefa7c47 | 555 | void page_cache_sync_ra(struct readahead_control *ractl, |
fcd9ae4f | 556 | unsigned long req_count) |
122a21d1 | 557 | { |
324bcf54 | 558 | bool do_forced_ra = ractl->file && (ractl->file->f_mode & FMODE_RANDOM); |
cf914a7d | 559 | |
324bcf54 JA |
560 | /* |
561 | * Even if read-ahead is disabled, issue this request as read-ahead | |
562 | * as we'll need it to satisfy the requested range. The forced | |
563 | * read-ahead will do the right thing and limit the read to just the | |
564 | * requested range, which we'll set to 1 page for this case. | |
565 | */ | |
fcd9ae4f | 566 | if (!ractl->ra->ra_pages || blk_cgroup_congested()) { |
324bcf54 JA |
567 | if (!ractl->file) |
568 | return; | |
569 | req_count = 1; | |
570 | do_forced_ra = true; | |
571 | } | |
ca47e8c7 | 572 | |
0141450f | 573 | /* be dumb */ |
324bcf54 | 574 | if (do_forced_ra) { |
fcd9ae4f | 575 | force_page_cache_ra(ractl, req_count); |
0141450f WF |
576 | return; |
577 | } | |
578 | ||
cf914a7d | 579 | /* do read-ahead */ |
fcd9ae4f | 580 | ondemand_readahead(ractl, false, req_count); |
cf914a7d | 581 | } |
fefa7c47 | 582 | EXPORT_SYMBOL_GPL(page_cache_sync_ra); |
cf914a7d | 583 | |
fefa7c47 | 584 | void page_cache_async_ra(struct readahead_control *ractl, |
fcd9ae4f | 585 | struct page *page, unsigned long req_count) |
cf914a7d RR |
586 | { |
587 | /* no read-ahead */ | |
fcd9ae4f | 588 | if (!ractl->ra->ra_pages) |
cf914a7d RR |
589 | return; |
590 | ||
591 | /* | |
592 | * Same bit is used for PG_readahead and PG_reclaim. | |
593 | */ | |
594 | if (PageWriteback(page)) | |
595 | return; | |
596 | ||
597 | ClearPageReadahead(page); | |
598 | ||
599 | /* | |
600 | * Defer asynchronous read-ahead on IO congestion. | |
601 | */ | |
fefa7c47 | 602 | if (inode_read_congested(ractl->mapping->host)) |
cf914a7d | 603 | return; |
122a21d1 | 604 | |
ca47e8c7 JB |
605 | if (blk_cgroup_congested()) |
606 | return; | |
607 | ||
122a21d1 | 608 | /* do read-ahead */ |
fcd9ae4f | 609 | ondemand_readahead(ractl, true, req_count); |
122a21d1 | 610 | } |
fefa7c47 | 611 | EXPORT_SYMBOL_GPL(page_cache_async_ra); |
782182e5 | 612 | |
c7b95d51 | 613 | ssize_t ksys_readahead(int fd, loff_t offset, size_t count) |
782182e5 CW |
614 | { |
615 | ssize_t ret; | |
2903ff01 | 616 | struct fd f; |
782182e5 CW |
617 | |
618 | ret = -EBADF; | |
2903ff01 | 619 | f = fdget(fd); |
3d8f7615 AG |
620 | if (!f.file || !(f.file->f_mode & FMODE_READ)) |
621 | goto out; | |
622 | ||
623 | /* | |
624 | * The readahead() syscall is intended to run only on files | |
625 | * that can execute readahead. If readahead is not possible | |
626 | * on this file, then we must return -EINVAL. | |
627 | */ | |
628 | ret = -EINVAL; | |
629 | if (!f.file->f_mapping || !f.file->f_mapping->a_ops || | |
630 | !S_ISREG(file_inode(f.file)->i_mode)) | |
631 | goto out; | |
632 | ||
633 | ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED); | |
634 | out: | |
635 | fdput(f); | |
782182e5 CW |
636 | return ret; |
637 | } | |
c7b95d51 DB |
638 | |
639 | SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count) | |
640 | { | |
641 | return ksys_readahead(fd, offset, count); | |
642 | } | |
3ca23644 DH |
643 | |
644 | /** | |
645 | * readahead_expand - Expand a readahead request | |
646 | * @ractl: The request to be expanded | |
647 | * @new_start: The revised start | |
648 | * @new_len: The revised size of the request | |
649 | * | |
650 | * Attempt to expand a readahead request outwards from the current size to the | |
651 | * specified size by inserting locked pages before and after the current window | |
652 | * to increase the size to the new window. This may involve the insertion of | |
653 | * THPs, in which case the window may get expanded even beyond what was | |
654 | * requested. | |
655 | * | |
656 | * The algorithm will stop if it encounters a conflicting page already in the | |
657 | * pagecache and leave a smaller expansion than requested. | |
658 | * | |
659 | * The caller must check for this by examining the revised @ractl object for a | |
660 | * different expansion than was requested. | |
661 | */ | |
662 | void readahead_expand(struct readahead_control *ractl, | |
663 | loff_t new_start, size_t new_len) | |
664 | { | |
665 | struct address_space *mapping = ractl->mapping; | |
666 | struct file_ra_state *ra = ractl->ra; | |
667 | pgoff_t new_index, new_nr_pages; | |
668 | gfp_t gfp_mask = readahead_gfp_mask(mapping); | |
669 | ||
670 | new_index = new_start / PAGE_SIZE; | |
671 | ||
672 | /* Expand the leading edge downwards */ | |
673 | while (ractl->_index > new_index) { | |
674 | unsigned long index = ractl->_index - 1; | |
675 | struct page *page = xa_load(&mapping->i_pages, index); | |
676 | ||
677 | if (page && !xa_is_value(page)) | |
678 | return; /* Page apparently present */ | |
679 | ||
680 | page = __page_cache_alloc(gfp_mask); | |
681 | if (!page) | |
682 | return; | |
683 | if (add_to_page_cache_lru(page, mapping, index, gfp_mask) < 0) { | |
684 | put_page(page); | |
685 | return; | |
686 | } | |
687 | ||
688 | ractl->_nr_pages++; | |
689 | ractl->_index = page->index; | |
690 | } | |
691 | ||
692 | new_len += new_start - readahead_pos(ractl); | |
693 | new_nr_pages = DIV_ROUND_UP(new_len, PAGE_SIZE); | |
694 | ||
695 | /* Expand the trailing edge upwards */ | |
696 | while (ractl->_nr_pages < new_nr_pages) { | |
697 | unsigned long index = ractl->_index + ractl->_nr_pages; | |
698 | struct page *page = xa_load(&mapping->i_pages, index); | |
699 | ||
700 | if (page && !xa_is_value(page)) | |
701 | return; /* Page apparently present */ | |
702 | ||
703 | page = __page_cache_alloc(gfp_mask); | |
704 | if (!page) | |
705 | return; | |
706 | if (add_to_page_cache_lru(page, mapping, index, gfp_mask) < 0) { | |
707 | put_page(page); | |
708 | return; | |
709 | } | |
710 | ractl->_nr_pages++; | |
711 | if (ra) { | |
712 | ra->size++; | |
713 | ra->async_size++; | |
714 | } | |
715 | } | |
716 | } | |
717 | EXPORT_SYMBOL(readahead_expand); |