| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * mm/readahead.c - address_space-level file readahead. |
| 4 | * |
| 5 | * Copyright (C) 2002, Linus Torvalds |
| 6 | * |
| 7 | * 09Apr2002 Andrew Morton |
| 8 | * Initial version. |
| 9 | */ |
| 10 | |
| 11 | /** |
| 12 | * DOC: Readahead Overview |
| 13 | * |
| 14 | * Readahead is used to read content into the page cache before it is |
| 15 | * explicitly requested by the application. Readahead only ever |
| 16 | * attempts to read folios that are not yet in the page cache. If a |
| 17 | * folio is present but not up-to-date, readahead will not try to read |
| 18 | * it. In that case a simple ->read_folio() will be requested. |
| 19 | * |
| 20 | * Readahead is triggered when an application read request (whether a |
| 21 | * system call or a page fault) finds that the requested folio is not in |
| 22 | * the page cache, or that it is in the page cache and has the |
| 23 | * readahead flag set. This flag indicates that the folio was read |
| 24 | * as part of a previous readahead request and now that it has been |
| 25 | * accessed, it is time for the next readahead. |
| 26 | * |
| 27 | * Each readahead request is partly synchronous read, and partly async |
| 28 | * readahead. This is reflected in the struct file_ra_state which |
| 29 | * contains ->size being the total number of pages, and ->async_size |
| 30 | * which is the number of pages in the async section. The readahead |
| 31 | * flag will be set on the first folio in this async section to trigger |
| 32 | * a subsequent readahead. Once a series of sequential reads has been |
| 33 | * established, there should be no need for a synchronous component and |
| 34 | * all readahead request will be fully asynchronous. |
| 35 | * |
| 36 | * When either of the triggers causes a readahead, three numbers need |
| 37 | * to be determined: the start of the region to read, the size of the |
| 38 | * region, and the size of the async tail. |
| 39 | * |
| 40 | * The start of the region is simply the first page address at or after |
| 41 | * the accessed address, which is not currently populated in the page |
| 42 | * cache. This is found with a simple search in the page cache. |
| 43 | * |
| 44 | * The size of the async tail is determined by subtracting the size that |
| 45 | * was explicitly requested from the determined request size, unless |
| 46 | * this would be less than zero - then zero is used. NOTE THIS |
| 47 | * CALCULATION IS WRONG WHEN THE START OF THE REGION IS NOT THE ACCESSED |
| 48 | * PAGE. ALSO THIS CALCULATION IS NOT USED CONSISTENTLY. |
| 49 | * |
| 50 | * The size of the region is normally determined from the size of the |
| 51 | * previous readahead which loaded the preceding pages. This may be |
| 52 | * discovered from the struct file_ra_state for simple sequential reads, |
| 53 | * or from examining the state of the page cache when multiple |
| 54 | * sequential reads are interleaved. Specifically: where the readahead |
| 55 | * was triggered by the readahead flag, the size of the previous |
| 56 | * readahead is assumed to be the number of pages from the triggering |
| 57 | * page to the start of the new readahead. In these cases, the size of |
| 58 | * the previous readahead is scaled, often doubled, for the new |
| 59 | * readahead, though see get_next_ra_size() for details. |
| 60 | * |
| 61 | * If the size of the previous read cannot be determined, the number of |
| 62 | * preceding pages in the page cache is used to estimate the size of |
| 63 | * a previous read. This estimate could easily be misled by random |
| 64 | * reads being coincidentally adjacent, so it is ignored unless it is |
| 65 | * larger than the current request, and it is not scaled up, unless it |
| 66 | * is at the start of file. |
| 67 | * |
| 68 | * In general readahead is accelerated at the start of the file, as |
| 69 | * reads from there are often sequential. There are other minor |
| 70 | * adjustments to the readahead size in various special cases and these |
| 71 | * are best discovered by reading the code. |
| 72 | * |
| 73 | * The above calculation, based on the previous readahead size, |
| 74 | * determines the size of the readahead, to which any requested read |
| 75 | * size may be added. |
| 76 | * |
| 77 | * Readahead requests are sent to the filesystem using the ->readahead() |
| 78 | * address space operation, for which mpage_readahead() is a canonical |
| 79 | * implementation. ->readahead() should normally initiate reads on all |
| 80 | * folios, but may fail to read any or all folios without causing an I/O |
| 81 | * error. The page cache reading code will issue a ->read_folio() request |
| 82 | * for any folio which ->readahead() did not read, and only an error |
| 83 | * from this will be final. |
| 84 | * |
| 85 | * ->readahead() will generally call readahead_folio() repeatedly to get |
| 86 | * each folio from those prepared for readahead. It may fail to read a |
| 87 | * folio by: |
| 88 | * |
| 89 | * * not calling readahead_folio() sufficiently many times, effectively |
| 90 | * ignoring some folios, as might be appropriate if the path to |
| 91 | * storage is congested. |
| 92 | * |
| 93 | * * failing to actually submit a read request for a given folio, |
| 94 | * possibly due to insufficient resources, or |
| 95 | * |
| 96 | * * getting an error during subsequent processing of a request. |
| 97 | * |
| 98 | * In the last two cases, the folio should be unlocked by the filesystem |
| 99 | * to indicate that the read attempt has failed. In the first case the |
| 100 | * folio will be unlocked by the VFS. |
| 101 | * |
| 102 | * Those folios not in the final ``async_size`` of the request should be |
| 103 | * considered to be important and ->readahead() should not fail them due |
| 104 | * to congestion or temporary resource unavailability, but should wait |
| 105 | * for necessary resources (e.g. memory or indexing information) to |
| 106 | * become available. Folios in the final ``async_size`` may be |
| 107 | * considered less urgent and failure to read them is more acceptable. |
| 108 | * In this case it is best to use filemap_remove_folio() to remove the |
| 109 | * folios from the page cache as is automatically done for folios that |
| 110 | * were not fetched with readahead_folio(). This will allow a |
| 111 | * subsequent synchronous readahead request to try them again. If they |
| 112 | * are left in the page cache, then they will be read individually using |
| 113 | * ->read_folio() which may be less efficient. |
| 114 | */ |
| 115 | |
| 116 | #include <linux/blkdev.h> |
| 117 | #include <linux/kernel.h> |
| 118 | #include <linux/dax.h> |
| 119 | #include <linux/gfp.h> |
| 120 | #include <linux/export.h> |
| 121 | #include <linux/backing-dev.h> |
| 122 | #include <linux/task_io_accounting_ops.h> |
| 123 | #include <linux/pagemap.h> |
| 124 | #include <linux/psi.h> |
| 125 | #include <linux/syscalls.h> |
| 126 | #include <linux/file.h> |
| 127 | #include <linux/mm_inline.h> |
| 128 | #include <linux/blk-cgroup.h> |
| 129 | #include <linux/fadvise.h> |
| 130 | #include <linux/sched/mm.h> |
| 131 | |
| 132 | #include "internal.h" |
| 133 | |
| 134 | /* |
| 135 | * Initialise a struct file's readahead state. Assumes that the caller has |
| 136 | * memset *ra to zero. |
| 137 | */ |
| 138 | void |
| 139 | file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping) |
| 140 | { |
| 141 | ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages; |
| 142 | ra->prev_pos = -1; |
| 143 | } |
| 144 | EXPORT_SYMBOL_GPL(file_ra_state_init); |
| 145 | |
| 146 | static void read_pages(struct readahead_control *rac) |
| 147 | { |
| 148 | const struct address_space_operations *aops = rac->mapping->a_ops; |
| 149 | struct folio *folio; |
| 150 | struct blk_plug plug; |
| 151 | |
| 152 | if (!readahead_count(rac)) |
| 153 | return; |
| 154 | |
| 155 | if (unlikely(rac->_workingset)) |
| 156 | psi_memstall_enter(&rac->_pflags); |
| 157 | blk_start_plug(&plug); |
| 158 | |
| 159 | if (aops->readahead) { |
| 160 | aops->readahead(rac); |
| 161 | /* |
| 162 | * Clean up the remaining folios. The sizes in ->ra |
| 163 | * may be used to size the next readahead, so make sure |
| 164 | * they accurately reflect what happened. |
| 165 | */ |
| 166 | while ((folio = readahead_folio(rac)) != NULL) { |
| 167 | unsigned long nr = folio_nr_pages(folio); |
| 168 | |
| 169 | folio_get(folio); |
| 170 | rac->ra->size -= nr; |
| 171 | if (rac->ra->async_size >= nr) { |
| 172 | rac->ra->async_size -= nr; |
| 173 | filemap_remove_folio(folio); |
| 174 | } |
| 175 | folio_unlock(folio); |
| 176 | folio_put(folio); |
| 177 | } |
| 178 | } else { |
| 179 | while ((folio = readahead_folio(rac)) != NULL) |
| 180 | aops->read_folio(rac->file, folio); |
| 181 | } |
| 182 | |
| 183 | blk_finish_plug(&plug); |
| 184 | if (unlikely(rac->_workingset)) |
| 185 | psi_memstall_leave(&rac->_pflags); |
| 186 | rac->_workingset = false; |
| 187 | |
| 188 | BUG_ON(readahead_count(rac)); |
| 189 | } |
| 190 | |
| 191 | /** |
| 192 | * page_cache_ra_unbounded - Start unchecked readahead. |
| 193 | * @ractl: Readahead control. |
| 194 | * @nr_to_read: The number of pages to read. |
| 195 | * @lookahead_size: Where to start the next readahead. |
| 196 | * |
| 197 | * This function is for filesystems to call when they want to start |
| 198 | * readahead beyond a file's stated i_size. This is almost certainly |
| 199 | * not the function you want to call. Use page_cache_async_readahead() |
| 200 | * or page_cache_sync_readahead() instead. |
| 201 | * |
| 202 | * Context: File is referenced by caller. Mutexes may be held by caller. |
| 203 | * May sleep, but will not reenter filesystem to reclaim memory. |
| 204 | */ |
| 205 | void page_cache_ra_unbounded(struct readahead_control *ractl, |
| 206 | unsigned long nr_to_read, unsigned long lookahead_size) |
| 207 | { |
| 208 | struct address_space *mapping = ractl->mapping; |
| 209 | unsigned long ra_folio_index, index = readahead_index(ractl); |
| 210 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
| 211 | unsigned long mark, i = 0; |
| 212 | unsigned int min_nrpages = mapping_min_folio_nrpages(mapping); |
| 213 | |
| 214 | /* |
| 215 | * Partway through the readahead operation, we will have added |
| 216 | * locked pages to the page cache, but will not yet have submitted |
| 217 | * them for I/O. Adding another page may need to allocate memory, |
| 218 | * which can trigger memory reclaim. Telling the VM we're in |
| 219 | * the middle of a filesystem operation will cause it to not |
| 220 | * touch file-backed pages, preventing a deadlock. Most (all?) |
| 221 | * filesystems already specify __GFP_NOFS in their mapping's |
| 222 | * gfp_mask, but let's be explicit here. |
| 223 | */ |
| 224 | unsigned int nofs = memalloc_nofs_save(); |
| 225 | |
| 226 | filemap_invalidate_lock_shared(mapping); |
| 227 | index = mapping_align_index(mapping, index); |
| 228 | |
| 229 | /* |
| 230 | * As iterator `i` is aligned to min_nrpages, round_up the |
| 231 | * difference between nr_to_read and lookahead_size to mark the |
| 232 | * index that only has lookahead or "async_region" to set the |
| 233 | * readahead flag. |
| 234 | */ |
| 235 | ra_folio_index = round_up(readahead_index(ractl) + nr_to_read - lookahead_size, |
| 236 | min_nrpages); |
| 237 | mark = ra_folio_index - index; |
| 238 | nr_to_read += readahead_index(ractl) - index; |
| 239 | ractl->_index = index; |
| 240 | |
| 241 | /* |
| 242 | * Preallocate as many pages as we will need. |
| 243 | */ |
| 244 | while (i < nr_to_read) { |
| 245 | struct folio *folio = xa_load(&mapping->i_pages, index + i); |
| 246 | int ret; |
| 247 | |
| 248 | if (folio && !xa_is_value(folio)) { |
| 249 | /* |
| 250 | * Page already present? Kick off the current batch |
| 251 | * of contiguous pages before continuing with the |
| 252 | * next batch. This page may be the one we would |
| 253 | * have intended to mark as Readahead, but we don't |
| 254 | * have a stable reference to this page, and it's |
| 255 | * not worth getting one just for that. |
| 256 | */ |
| 257 | read_pages(ractl); |
| 258 | ractl->_index += min_nrpages; |
| 259 | i = ractl->_index + ractl->_nr_pages - index; |
| 260 | continue; |
| 261 | } |
| 262 | |
| 263 | folio = filemap_alloc_folio(gfp_mask, |
| 264 | mapping_min_folio_order(mapping)); |
| 265 | if (!folio) |
| 266 | break; |
| 267 | |
| 268 | ret = filemap_add_folio(mapping, folio, index + i, gfp_mask); |
| 269 | if (ret < 0) { |
| 270 | folio_put(folio); |
| 271 | if (ret == -ENOMEM) |
| 272 | break; |
| 273 | read_pages(ractl); |
| 274 | ractl->_index += min_nrpages; |
| 275 | i = ractl->_index + ractl->_nr_pages - index; |
| 276 | continue; |
| 277 | } |
| 278 | if (i == mark) |
| 279 | folio_set_readahead(folio); |
| 280 | ractl->_workingset |= folio_test_workingset(folio); |
| 281 | ractl->_nr_pages += min_nrpages; |
| 282 | i += min_nrpages; |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * Now start the IO. We ignore I/O errors - if the folio is not |
| 287 | * uptodate then the caller will launch read_folio again, and |
| 288 | * will then handle the error. |
| 289 | */ |
| 290 | read_pages(ractl); |
| 291 | filemap_invalidate_unlock_shared(mapping); |
| 292 | memalloc_nofs_restore(nofs); |
| 293 | } |
| 294 | EXPORT_SYMBOL_GPL(page_cache_ra_unbounded); |
| 295 | |
| 296 | /* |
| 297 | * do_page_cache_ra() actually reads a chunk of disk. It allocates |
| 298 | * the pages first, then submits them for I/O. This avoids the very bad |
| 299 | * behaviour which would occur if page allocations are causing VM writeback. |
| 300 | * We really don't want to intermingle reads and writes like that. |
| 301 | */ |
| 302 | static void do_page_cache_ra(struct readahead_control *ractl, |
| 303 | unsigned long nr_to_read, unsigned long lookahead_size) |
| 304 | { |
| 305 | struct inode *inode = ractl->mapping->host; |
| 306 | unsigned long index = readahead_index(ractl); |
| 307 | loff_t isize = i_size_read(inode); |
| 308 | pgoff_t end_index; /* The last page we want to read */ |
| 309 | |
| 310 | if (isize == 0) |
| 311 | return; |
| 312 | |
| 313 | end_index = (isize - 1) >> PAGE_SHIFT; |
| 314 | if (index > end_index) |
| 315 | return; |
| 316 | /* Don't read past the page containing the last byte of the file */ |
| 317 | if (nr_to_read > end_index - index) |
| 318 | nr_to_read = end_index - index + 1; |
| 319 | |
| 320 | page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size); |
| 321 | } |
| 322 | |
| 323 | /* |
| 324 | * Chunk the readahead into 2 megabyte units, so that we don't pin too much |
| 325 | * memory at once. |
| 326 | */ |
| 327 | void force_page_cache_ra(struct readahead_control *ractl, |
| 328 | unsigned long nr_to_read) |
| 329 | { |
| 330 | struct address_space *mapping = ractl->mapping; |
| 331 | struct file_ra_state *ra = ractl->ra; |
| 332 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
| 333 | unsigned long max_pages; |
| 334 | |
| 335 | if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead)) |
| 336 | return; |
| 337 | |
| 338 | /* |
| 339 | * If the request exceeds the readahead window, allow the read to |
| 340 | * be up to the optimal hardware IO size |
| 341 | */ |
| 342 | max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages); |
| 343 | nr_to_read = min_t(unsigned long, nr_to_read, max_pages); |
| 344 | while (nr_to_read) { |
| 345 | unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE; |
| 346 | |
| 347 | if (this_chunk > nr_to_read) |
| 348 | this_chunk = nr_to_read; |
| 349 | do_page_cache_ra(ractl, this_chunk, 0); |
| 350 | |
| 351 | nr_to_read -= this_chunk; |
| 352 | } |
| 353 | } |
| 354 | |
| 355 | /* |
| 356 | * Set the initial window size, round to next power of 2 and square |
| 357 | * for small size, x 4 for medium, and x 2 for large |
| 358 | * for 128k (32 page) max ra |
| 359 | * 1-2 page = 16k, 3-4 page 32k, 5-8 page = 64k, > 8 page = 128k initial |
| 360 | */ |
| 361 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) |
| 362 | { |
| 363 | unsigned long newsize = roundup_pow_of_two(size); |
| 364 | |
| 365 | if (newsize <= max / 32) |
| 366 | newsize = newsize * 4; |
| 367 | else if (newsize <= max / 4) |
| 368 | newsize = newsize * 2; |
| 369 | else |
| 370 | newsize = max; |
| 371 | |
| 372 | return newsize; |
| 373 | } |
| 374 | |
| 375 | /* |
| 376 | * Get the previous window size, ramp it up, and |
| 377 | * return it as the new window size. |
| 378 | */ |
| 379 | static unsigned long get_next_ra_size(struct file_ra_state *ra, |
| 380 | unsigned long max) |
| 381 | { |
| 382 | unsigned long cur = ra->size; |
| 383 | |
| 384 | if (cur < max / 16) |
| 385 | return 4 * cur; |
| 386 | if (cur <= max / 2) |
| 387 | return 2 * cur; |
| 388 | return max; |
| 389 | } |
| 390 | |
| 391 | /* |
| 392 | * On-demand readahead design. |
| 393 | * |
| 394 | * The fields in struct file_ra_state represent the most-recently-executed |
| 395 | * readahead attempt: |
| 396 | * |
| 397 | * |<----- async_size ---------| |
| 398 | * |------------------- size -------------------->| |
| 399 | * |==================#===========================| |
| 400 | * ^start ^page marked with PG_readahead |
| 401 | * |
| 402 | * To overlap application thinking time and disk I/O time, we do |
| 403 | * `readahead pipelining': Do not wait until the application consumed all |
| 404 | * readahead pages and stalled on the missing page at readahead_index; |
| 405 | * Instead, submit an asynchronous readahead I/O as soon as there are |
| 406 | * only async_size pages left in the readahead window. Normally async_size |
| 407 | * will be equal to size, for maximum pipelining. |
| 408 | * |
| 409 | * In interleaved sequential reads, concurrent streams on the same fd can |
| 410 | * be invalidating each other's readahead state. So we flag the new readahead |
| 411 | * page at (start+size-async_size) with PG_readahead, and use it as readahead |
| 412 | * indicator. The flag won't be set on already cached pages, to avoid the |
| 413 | * readahead-for-nothing fuss, saving pointless page cache lookups. |
| 414 | * |
| 415 | * prev_pos tracks the last visited byte in the _previous_ read request. |
| 416 | * It should be maintained by the caller, and will be used for detecting |
| 417 | * small random reads. Note that the readahead algorithm checks loosely |
| 418 | * for sequential patterns. Hence interleaved reads might be served as |
| 419 | * sequential ones. |
| 420 | * |
| 421 | * There is a special-case: if the first page which the application tries to |
| 422 | * read happens to be the first page of the file, it is assumed that a linear |
| 423 | * read is about to happen and the window is immediately set to the initial size |
| 424 | * based on I/O request size and the max_readahead. |
| 425 | * |
| 426 | * The code ramps up the readahead size aggressively at first, but slow down as |
| 427 | * it approaches max_readhead. |
| 428 | */ |
| 429 | |
| 430 | static inline int ra_alloc_folio(struct readahead_control *ractl, pgoff_t index, |
| 431 | pgoff_t mark, unsigned int order, gfp_t gfp) |
| 432 | { |
| 433 | int err; |
| 434 | struct folio *folio = filemap_alloc_folio(gfp, order); |
| 435 | |
| 436 | if (!folio) |
| 437 | return -ENOMEM; |
| 438 | mark = round_down(mark, 1UL << order); |
| 439 | if (index == mark) |
| 440 | folio_set_readahead(folio); |
| 441 | err = filemap_add_folio(ractl->mapping, folio, index, gfp); |
| 442 | if (err) { |
| 443 | folio_put(folio); |
| 444 | return err; |
| 445 | } |
| 446 | |
| 447 | ractl->_nr_pages += 1UL << order; |
| 448 | ractl->_workingset |= folio_test_workingset(folio); |
| 449 | return 0; |
| 450 | } |
| 451 | |
| 452 | void page_cache_ra_order(struct readahead_control *ractl, |
| 453 | struct file_ra_state *ra, unsigned int new_order) |
| 454 | { |
| 455 | struct address_space *mapping = ractl->mapping; |
| 456 | pgoff_t start = readahead_index(ractl); |
| 457 | pgoff_t index = start; |
| 458 | unsigned int min_order = mapping_min_folio_order(mapping); |
| 459 | pgoff_t limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT; |
| 460 | pgoff_t mark = index + ra->size - ra->async_size; |
| 461 | unsigned int nofs; |
| 462 | int err = 0; |
| 463 | gfp_t gfp = readahead_gfp_mask(mapping); |
| 464 | unsigned int min_ra_size = max(4, mapping_min_folio_nrpages(mapping)); |
| 465 | |
| 466 | /* |
| 467 | * Fallback when size < min_nrpages as each folio should be |
| 468 | * at least min_nrpages anyway. |
| 469 | */ |
| 470 | if (!mapping_large_folio_support(mapping) || ra->size < min_ra_size) |
| 471 | goto fallback; |
| 472 | |
| 473 | limit = min(limit, index + ra->size - 1); |
| 474 | |
| 475 | if (new_order < mapping_max_folio_order(mapping)) |
| 476 | new_order += 2; |
| 477 | |
| 478 | new_order = min(mapping_max_folio_order(mapping), new_order); |
| 479 | new_order = min_t(unsigned int, new_order, ilog2(ra->size)); |
| 480 | new_order = max(new_order, min_order); |
| 481 | |
| 482 | /* See comment in page_cache_ra_unbounded() */ |
| 483 | nofs = memalloc_nofs_save(); |
| 484 | filemap_invalidate_lock_shared(mapping); |
| 485 | /* |
| 486 | * If the new_order is greater than min_order and index is |
| 487 | * already aligned to new_order, then this will be noop as index |
| 488 | * aligned to new_order should also be aligned to min_order. |
| 489 | */ |
| 490 | ractl->_index = mapping_align_index(mapping, index); |
| 491 | index = readahead_index(ractl); |
| 492 | |
| 493 | while (index <= limit) { |
| 494 | unsigned int order = new_order; |
| 495 | |
| 496 | /* Align with smaller pages if needed */ |
| 497 | if (index & ((1UL << order) - 1)) |
| 498 | order = __ffs(index); |
| 499 | /* Don't allocate pages past EOF */ |
| 500 | while (order > min_order && index + (1UL << order) - 1 > limit) |
| 501 | order--; |
| 502 | err = ra_alloc_folio(ractl, index, mark, order, gfp); |
| 503 | if (err) |
| 504 | break; |
| 505 | index += 1UL << order; |
| 506 | } |
| 507 | |
| 508 | read_pages(ractl); |
| 509 | filemap_invalidate_unlock_shared(mapping); |
| 510 | memalloc_nofs_restore(nofs); |
| 511 | |
| 512 | /* |
| 513 | * If there were already pages in the page cache, then we may have |
| 514 | * left some gaps. Let the regular readahead code take care of this |
| 515 | * situation. |
| 516 | */ |
| 517 | if (!err) |
| 518 | return; |
| 519 | fallback: |
| 520 | do_page_cache_ra(ractl, ra->size - (index - start), ra->async_size); |
| 521 | } |
| 522 | |
| 523 | static unsigned long ractl_max_pages(struct readahead_control *ractl, |
| 524 | unsigned long req_size) |
| 525 | { |
| 526 | struct backing_dev_info *bdi = inode_to_bdi(ractl->mapping->host); |
| 527 | unsigned long max_pages = ractl->ra->ra_pages; |
| 528 | |
| 529 | /* |
| 530 | * If the request exceeds the readahead window, allow the read to |
| 531 | * be up to the optimal hardware IO size |
| 532 | */ |
| 533 | if (req_size > max_pages && bdi->io_pages > max_pages) |
| 534 | max_pages = min(req_size, bdi->io_pages); |
| 535 | return max_pages; |
| 536 | } |
| 537 | |
| 538 | void page_cache_sync_ra(struct readahead_control *ractl, |
| 539 | unsigned long req_count) |
| 540 | { |
| 541 | pgoff_t index = readahead_index(ractl); |
| 542 | bool do_forced_ra = ractl->file && (ractl->file->f_mode & FMODE_RANDOM); |
| 543 | struct file_ra_state *ra = ractl->ra; |
| 544 | unsigned long max_pages, contig_count; |
| 545 | pgoff_t prev_index, miss; |
| 546 | |
| 547 | /* |
| 548 | * Even if readahead is disabled, issue this request as readahead |
| 549 | * as we'll need it to satisfy the requested range. The forced |
| 550 | * readahead will do the right thing and limit the read to just the |
| 551 | * requested range, which we'll set to 1 page for this case. |
| 552 | */ |
| 553 | if (!ra->ra_pages || blk_cgroup_congested()) { |
| 554 | if (!ractl->file) |
| 555 | return; |
| 556 | req_count = 1; |
| 557 | do_forced_ra = true; |
| 558 | } |
| 559 | |
| 560 | /* be dumb */ |
| 561 | if (do_forced_ra) { |
| 562 | force_page_cache_ra(ractl, req_count); |
| 563 | return; |
| 564 | } |
| 565 | |
| 566 | max_pages = ractl_max_pages(ractl, req_count); |
| 567 | prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT; |
| 568 | /* |
| 569 | * A start of file, oversized read, or sequential cache miss: |
| 570 | * trivial case: (index - prev_index) == 1 |
| 571 | * unaligned reads: (index - prev_index) == 0 |
| 572 | */ |
| 573 | if (!index || req_count > max_pages || index - prev_index <= 1UL) { |
| 574 | ra->start = index; |
| 575 | ra->size = get_init_ra_size(req_count, max_pages); |
| 576 | ra->async_size = ra->size > req_count ? ra->size - req_count : |
| 577 | ra->size >> 1; |
| 578 | goto readit; |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * Query the page cache and look for the traces(cached history pages) |
| 583 | * that a sequential stream would leave behind. |
| 584 | */ |
| 585 | rcu_read_lock(); |
| 586 | miss = page_cache_prev_miss(ractl->mapping, index - 1, max_pages); |
| 587 | rcu_read_unlock(); |
| 588 | contig_count = index - miss - 1; |
| 589 | /* |
| 590 | * Standalone, small random read. Read as is, and do not pollute the |
| 591 | * readahead state. |
| 592 | */ |
| 593 | if (contig_count <= req_count) { |
| 594 | do_page_cache_ra(ractl, req_count, 0); |
| 595 | return; |
| 596 | } |
| 597 | /* |
| 598 | * File cached from the beginning: |
| 599 | * it is a strong indication of long-run stream (or whole-file-read) |
| 600 | */ |
| 601 | if (miss == ULONG_MAX) |
| 602 | contig_count *= 2; |
| 603 | ra->start = index; |
| 604 | ra->size = min(contig_count + req_count, max_pages); |
| 605 | ra->async_size = 1; |
| 606 | readit: |
| 607 | ractl->_index = ra->start; |
| 608 | page_cache_ra_order(ractl, ra, 0); |
| 609 | } |
| 610 | EXPORT_SYMBOL_GPL(page_cache_sync_ra); |
| 611 | |
| 612 | void page_cache_async_ra(struct readahead_control *ractl, |
| 613 | struct folio *folio, unsigned long req_count) |
| 614 | { |
| 615 | unsigned long max_pages; |
| 616 | struct file_ra_state *ra = ractl->ra; |
| 617 | pgoff_t index = readahead_index(ractl); |
| 618 | pgoff_t expected, start; |
| 619 | unsigned int order = folio_order(folio); |
| 620 | |
| 621 | /* no readahead */ |
| 622 | if (!ra->ra_pages) |
| 623 | return; |
| 624 | |
| 625 | /* |
| 626 | * Same bit is used for PG_readahead and PG_reclaim. |
| 627 | */ |
| 628 | if (folio_test_writeback(folio)) |
| 629 | return; |
| 630 | |
| 631 | folio_clear_readahead(folio); |
| 632 | |
| 633 | if (blk_cgroup_congested()) |
| 634 | return; |
| 635 | |
| 636 | max_pages = ractl_max_pages(ractl, req_count); |
| 637 | /* |
| 638 | * It's the expected callback index, assume sequential access. |
| 639 | * Ramp up sizes, and push forward the readahead window. |
| 640 | */ |
| 641 | expected = round_down(ra->start + ra->size - ra->async_size, |
| 642 | 1UL << order); |
| 643 | if (index == expected) { |
| 644 | ra->start += ra->size; |
| 645 | ra->size = get_next_ra_size(ra, max_pages); |
| 646 | ra->async_size = ra->size; |
| 647 | goto readit; |
| 648 | } |
| 649 | |
| 650 | /* |
| 651 | * Hit a marked folio without valid readahead state. |
| 652 | * E.g. interleaved reads. |
| 653 | * Query the pagecache for async_size, which normally equals to |
| 654 | * readahead size. Ramp it up and use it as the new readahead size. |
| 655 | */ |
| 656 | rcu_read_lock(); |
| 657 | start = page_cache_next_miss(ractl->mapping, index + 1, max_pages); |
| 658 | rcu_read_unlock(); |
| 659 | |
| 660 | if (!start || start - index > max_pages) |
| 661 | return; |
| 662 | |
| 663 | ra->start = start; |
| 664 | ra->size = start - index; /* old async_size */ |
| 665 | ra->size += req_count; |
| 666 | ra->size = get_next_ra_size(ra, max_pages); |
| 667 | ra->async_size = ra->size; |
| 668 | readit: |
| 669 | ractl->_index = ra->start; |
| 670 | page_cache_ra_order(ractl, ra, order); |
| 671 | } |
| 672 | EXPORT_SYMBOL_GPL(page_cache_async_ra); |
| 673 | |
| 674 | ssize_t ksys_readahead(int fd, loff_t offset, size_t count) |
| 675 | { |
| 676 | ssize_t ret; |
| 677 | struct fd f; |
| 678 | |
| 679 | ret = -EBADF; |
| 680 | f = fdget(fd); |
| 681 | if (!fd_file(f) || !(fd_file(f)->f_mode & FMODE_READ)) |
| 682 | goto out; |
| 683 | |
| 684 | /* |
| 685 | * The readahead() syscall is intended to run only on files |
| 686 | * that can execute readahead. If readahead is not possible |
| 687 | * on this file, then we must return -EINVAL. |
| 688 | */ |
| 689 | ret = -EINVAL; |
| 690 | if (!fd_file(f)->f_mapping || !fd_file(f)->f_mapping->a_ops || |
| 691 | (!S_ISREG(file_inode(fd_file(f))->i_mode) && |
| 692 | !S_ISBLK(file_inode(fd_file(f))->i_mode))) |
| 693 | goto out; |
| 694 | |
| 695 | ret = vfs_fadvise(fd_file(f), offset, count, POSIX_FADV_WILLNEED); |
| 696 | out: |
| 697 | fdput(f); |
| 698 | return ret; |
| 699 | } |
| 700 | |
| 701 | SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count) |
| 702 | { |
| 703 | return ksys_readahead(fd, offset, count); |
| 704 | } |
| 705 | |
| 706 | #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_READAHEAD) |
| 707 | COMPAT_SYSCALL_DEFINE4(readahead, int, fd, compat_arg_u64_dual(offset), size_t, count) |
| 708 | { |
| 709 | return ksys_readahead(fd, compat_arg_u64_glue(offset), count); |
| 710 | } |
| 711 | #endif |
| 712 | |
| 713 | /** |
| 714 | * readahead_expand - Expand a readahead request |
| 715 | * @ractl: The request to be expanded |
| 716 | * @new_start: The revised start |
| 717 | * @new_len: The revised size of the request |
| 718 | * |
| 719 | * Attempt to expand a readahead request outwards from the current size to the |
| 720 | * specified size by inserting locked pages before and after the current window |
| 721 | * to increase the size to the new window. This may involve the insertion of |
| 722 | * THPs, in which case the window may get expanded even beyond what was |
| 723 | * requested. |
| 724 | * |
| 725 | * The algorithm will stop if it encounters a conflicting page already in the |
| 726 | * pagecache and leave a smaller expansion than requested. |
| 727 | * |
| 728 | * The caller must check for this by examining the revised @ractl object for a |
| 729 | * different expansion than was requested. |
| 730 | */ |
| 731 | void readahead_expand(struct readahead_control *ractl, |
| 732 | loff_t new_start, size_t new_len) |
| 733 | { |
| 734 | struct address_space *mapping = ractl->mapping; |
| 735 | struct file_ra_state *ra = ractl->ra; |
| 736 | pgoff_t new_index, new_nr_pages; |
| 737 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
| 738 | unsigned long min_nrpages = mapping_min_folio_nrpages(mapping); |
| 739 | unsigned int min_order = mapping_min_folio_order(mapping); |
| 740 | |
| 741 | new_index = new_start / PAGE_SIZE; |
| 742 | /* |
| 743 | * Readahead code should have aligned the ractl->_index to |
| 744 | * min_nrpages before calling readahead aops. |
| 745 | */ |
| 746 | VM_BUG_ON(!IS_ALIGNED(ractl->_index, min_nrpages)); |
| 747 | |
| 748 | /* Expand the leading edge downwards */ |
| 749 | while (ractl->_index > new_index) { |
| 750 | unsigned long index = ractl->_index - 1; |
| 751 | struct folio *folio = xa_load(&mapping->i_pages, index); |
| 752 | |
| 753 | if (folio && !xa_is_value(folio)) |
| 754 | return; /* Folio apparently present */ |
| 755 | |
| 756 | folio = filemap_alloc_folio(gfp_mask, min_order); |
| 757 | if (!folio) |
| 758 | return; |
| 759 | |
| 760 | index = mapping_align_index(mapping, index); |
| 761 | if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) { |
| 762 | folio_put(folio); |
| 763 | return; |
| 764 | } |
| 765 | if (unlikely(folio_test_workingset(folio)) && |
| 766 | !ractl->_workingset) { |
| 767 | ractl->_workingset = true; |
| 768 | psi_memstall_enter(&ractl->_pflags); |
| 769 | } |
| 770 | ractl->_nr_pages += min_nrpages; |
| 771 | ractl->_index = folio->index; |
| 772 | } |
| 773 | |
| 774 | new_len += new_start - readahead_pos(ractl); |
| 775 | new_nr_pages = DIV_ROUND_UP(new_len, PAGE_SIZE); |
| 776 | |
| 777 | /* Expand the trailing edge upwards */ |
| 778 | while (ractl->_nr_pages < new_nr_pages) { |
| 779 | unsigned long index = ractl->_index + ractl->_nr_pages; |
| 780 | struct folio *folio = xa_load(&mapping->i_pages, index); |
| 781 | |
| 782 | if (folio && !xa_is_value(folio)) |
| 783 | return; /* Folio apparently present */ |
| 784 | |
| 785 | folio = filemap_alloc_folio(gfp_mask, min_order); |
| 786 | if (!folio) |
| 787 | return; |
| 788 | |
| 789 | index = mapping_align_index(mapping, index); |
| 790 | if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) { |
| 791 | folio_put(folio); |
| 792 | return; |
| 793 | } |
| 794 | if (unlikely(folio_test_workingset(folio)) && |
| 795 | !ractl->_workingset) { |
| 796 | ractl->_workingset = true; |
| 797 | psi_memstall_enter(&ractl->_pflags); |
| 798 | } |
| 799 | ractl->_nr_pages += min_nrpages; |
| 800 | if (ra) { |
| 801 | ra->size += min_nrpages; |
| 802 | ra->async_size += min_nrpages; |
| 803 | } |
| 804 | } |
| 805 | } |
| 806 | EXPORT_SYMBOL(readahead_expand); |