page-writeback.c revision 245b2e70eabd797932adb263a65da0bab3711753
1/* 2 * mm/page-writeback.c 3 * 4 * Copyright (C) 2002, Linus Torvalds. 5 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> 6 * 7 * Contains functions related to writing back dirty pages at the 8 * address_space level. 9 * 10 * 10Apr2002 Andrew Morton 11 * Initial version 12 */ 13 14#include <linux/kernel.h> 15#include <linux/module.h> 16#include <linux/spinlock.h> 17#include <linux/fs.h> 18#include <linux/mm.h> 19#include <linux/swap.h> 20#include <linux/slab.h> 21#include <linux/pagemap.h> 22#include <linux/writeback.h> 23#include <linux/init.h> 24#include <linux/backing-dev.h> 25#include <linux/task_io_accounting_ops.h> 26#include <linux/blkdev.h> 27#include <linux/mpage.h> 28#include <linux/rmap.h> 29#include <linux/percpu.h> 30#include <linux/notifier.h> 31#include <linux/smp.h> 32#include <linux/sysctl.h> 33#include <linux/cpu.h> 34#include <linux/syscalls.h> 35#include <linux/buffer_head.h> 36#include <linux/pagevec.h> 37 38/* 39 * The maximum number of pages to writeout in a single bdflush/kupdate 40 * operation. We do this so we don't hold I_SYNC against an inode for 41 * enormous amounts of time, which would block a userspace task which has 42 * been forced to throttle against that inode. Also, the code reevaluates 43 * the dirty each time it has written this many pages. 44 */ 45#define MAX_WRITEBACK_PAGES 1024 46 47/* 48 * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited 49 * will look to see if it needs to force writeback or throttling. 50 */ 51static long ratelimit_pages = 32; 52 53/* 54 * When balance_dirty_pages decides that the caller needs to perform some 55 * non-background writeback, this is how many pages it will attempt to write. 56 * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably 57 * large amounts of I/O are submitted. 58 */ 59static inline long sync_writeback_pages(void) 60{ 61 return ratelimit_pages + ratelimit_pages / 2; 62} 63 64/* The following parameters are exported via /proc/sys/vm */ 65 66/* 67 * Start background writeback (via pdflush) at this percentage 68 */ 69int dirty_background_ratio = 10; 70 71/* 72 * dirty_background_bytes starts at 0 (disabled) so that it is a function of 73 * dirty_background_ratio * the amount of dirtyable memory 74 */ 75unsigned long dirty_background_bytes; 76 77/* 78 * free highmem will not be subtracted from the total free memory 79 * for calculating free ratios if vm_highmem_is_dirtyable is true 80 */ 81int vm_highmem_is_dirtyable; 82 83/* 84 * The generator of dirty data starts writeback at this percentage 85 */ 86int vm_dirty_ratio = 20; 87 88/* 89 * vm_dirty_bytes starts at 0 (disabled) so that it is a function of 90 * vm_dirty_ratio * the amount of dirtyable memory 91 */ 92unsigned long vm_dirty_bytes; 93 94/* 95 * The interval between `kupdate'-style writebacks 96 */ 97unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */ 98 99/* 100 * The longest time for which data is allowed to remain dirty 101 */ 102unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */ 103 104/* 105 * Flag that makes the machine dump writes/reads and block dirtyings. 106 */ 107int block_dump; 108 109/* 110 * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies: 111 * a full sync is triggered after this time elapses without any disk activity. 112 */ 113int laptop_mode; 114 115EXPORT_SYMBOL(laptop_mode); 116 117/* End of sysctl-exported parameters */ 118 119 120static void background_writeout(unsigned long _min_pages); 121 122/* 123 * Scale the writeback cache size proportional to the relative writeout speeds. 124 * 125 * We do this by keeping a floating proportion between BDIs, based on page 126 * writeback completions [end_page_writeback()]. Those devices that write out 127 * pages fastest will get the larger share, while the slower will get a smaller 128 * share. 129 * 130 * We use page writeout completions because we are interested in getting rid of 131 * dirty pages. Having them written out is the primary goal. 132 * 133 * We introduce a concept of time, a period over which we measure these events, 134 * because demand can/will vary over time. The length of this period itself is 135 * measured in page writeback completions. 136 * 137 */ 138static struct prop_descriptor vm_completions; 139static struct prop_descriptor vm_dirties; 140 141/* 142 * couple the period to the dirty_ratio: 143 * 144 * period/2 ~ roundup_pow_of_two(dirty limit) 145 */ 146static int calc_period_shift(void) 147{ 148 unsigned long dirty_total; 149 150 if (vm_dirty_bytes) 151 dirty_total = vm_dirty_bytes / PAGE_SIZE; 152 else 153 dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / 154 100; 155 return 2 + ilog2(dirty_total - 1); 156} 157 158/* 159 * update the period when the dirty threshold changes. 160 */ 161static void update_completion_period(void) 162{ 163 int shift = calc_period_shift(); 164 prop_change_shift(&vm_completions, shift); 165 prop_change_shift(&vm_dirties, shift); 166} 167 168int dirty_background_ratio_handler(struct ctl_table *table, int write, 169 struct file *filp, void __user *buffer, size_t *lenp, 170 loff_t *ppos) 171{ 172 int ret; 173 174 ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); 175 if (ret == 0 && write) 176 dirty_background_bytes = 0; 177 return ret; 178} 179 180int dirty_background_bytes_handler(struct ctl_table *table, int write, 181 struct file *filp, void __user *buffer, size_t *lenp, 182 loff_t *ppos) 183{ 184 int ret; 185 186 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); 187 if (ret == 0 && write) 188 dirty_background_ratio = 0; 189 return ret; 190} 191 192int dirty_ratio_handler(struct ctl_table *table, int write, 193 struct file *filp, void __user *buffer, size_t *lenp, 194 loff_t *ppos) 195{ 196 int old_ratio = vm_dirty_ratio; 197 int ret; 198 199 ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); 200 if (ret == 0 && write && vm_dirty_ratio != old_ratio) { 201 update_completion_period(); 202 vm_dirty_bytes = 0; 203 } 204 return ret; 205} 206 207 208int dirty_bytes_handler(struct ctl_table *table, int write, 209 struct file *filp, void __user *buffer, size_t *lenp, 210 loff_t *ppos) 211{ 212 unsigned long old_bytes = vm_dirty_bytes; 213 int ret; 214 215 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); 216 if (ret == 0 && write && vm_dirty_bytes != old_bytes) { 217 update_completion_period(); 218 vm_dirty_ratio = 0; 219 } 220 return ret; 221} 222 223/* 224 * Increment the BDI's writeout completion count and the global writeout 225 * completion count. Called from test_clear_page_writeback(). 226 */ 227static inline void __bdi_writeout_inc(struct backing_dev_info *bdi) 228{ 229 __prop_inc_percpu_max(&vm_completions, &bdi->completions, 230 bdi->max_prop_frac); 231} 232 233void bdi_writeout_inc(struct backing_dev_info *bdi) 234{ 235 unsigned long flags; 236 237 local_irq_save(flags); 238 __bdi_writeout_inc(bdi); 239 local_irq_restore(flags); 240} 241EXPORT_SYMBOL_GPL(bdi_writeout_inc); 242 243void task_dirty_inc(struct task_struct *tsk) 244{ 245 prop_inc_single(&vm_dirties, &tsk->dirties); 246} 247 248/* 249 * Obtain an accurate fraction of the BDI's portion. 250 */ 251static void bdi_writeout_fraction(struct backing_dev_info *bdi, 252 long *numerator, long *denominator) 253{ 254 if (bdi_cap_writeback_dirty(bdi)) { 255 prop_fraction_percpu(&vm_completions, &bdi->completions, 256 numerator, denominator); 257 } else { 258 *numerator = 0; 259 *denominator = 1; 260 } 261} 262 263/* 264 * Clip the earned share of dirty pages to that which is actually available. 265 * This avoids exceeding the total dirty_limit when the floating averages 266 * fluctuate too quickly. 267 */ 268static void clip_bdi_dirty_limit(struct backing_dev_info *bdi, 269 unsigned long dirty, unsigned long *pbdi_dirty) 270{ 271 unsigned long avail_dirty; 272 273 avail_dirty = global_page_state(NR_FILE_DIRTY) + 274 global_page_state(NR_WRITEBACK) + 275 global_page_state(NR_UNSTABLE_NFS) + 276 global_page_state(NR_WRITEBACK_TEMP); 277 278 if (avail_dirty < dirty) 279 avail_dirty = dirty - avail_dirty; 280 else 281 avail_dirty = 0; 282 283 avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) + 284 bdi_stat(bdi, BDI_WRITEBACK); 285 286 *pbdi_dirty = min(*pbdi_dirty, avail_dirty); 287} 288 289static inline void task_dirties_fraction(struct task_struct *tsk, 290 long *numerator, long *denominator) 291{ 292 prop_fraction_single(&vm_dirties, &tsk->dirties, 293 numerator, denominator); 294} 295 296/* 297 * scale the dirty limit 298 * 299 * task specific dirty limit: 300 * 301 * dirty -= (dirty/8) * p_{t} 302 */ 303static void task_dirty_limit(struct task_struct *tsk, unsigned long *pdirty) 304{ 305 long numerator, denominator; 306 unsigned long dirty = *pdirty; 307 u64 inv = dirty >> 3; 308 309 task_dirties_fraction(tsk, &numerator, &denominator); 310 inv *= numerator; 311 do_div(inv, denominator); 312 313 dirty -= inv; 314 if (dirty < *pdirty/2) 315 dirty = *pdirty/2; 316 317 *pdirty = dirty; 318} 319 320/* 321 * 322 */ 323static DEFINE_SPINLOCK(bdi_lock); 324static unsigned int bdi_min_ratio; 325 326int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) 327{ 328 int ret = 0; 329 unsigned long flags; 330 331 spin_lock_irqsave(&bdi_lock, flags); 332 if (min_ratio > bdi->max_ratio) { 333 ret = -EINVAL; 334 } else { 335 min_ratio -= bdi->min_ratio; 336 if (bdi_min_ratio + min_ratio < 100) { 337 bdi_min_ratio += min_ratio; 338 bdi->min_ratio += min_ratio; 339 } else { 340 ret = -EINVAL; 341 } 342 } 343 spin_unlock_irqrestore(&bdi_lock, flags); 344 345 return ret; 346} 347 348int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) 349{ 350 unsigned long flags; 351 int ret = 0; 352 353 if (max_ratio > 100) 354 return -EINVAL; 355 356 spin_lock_irqsave(&bdi_lock, flags); 357 if (bdi->min_ratio > max_ratio) { 358 ret = -EINVAL; 359 } else { 360 bdi->max_ratio = max_ratio; 361 bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; 362 } 363 spin_unlock_irqrestore(&bdi_lock, flags); 364 365 return ret; 366} 367EXPORT_SYMBOL(bdi_set_max_ratio); 368 369/* 370 * Work out the current dirty-memory clamping and background writeout 371 * thresholds. 372 * 373 * The main aim here is to lower them aggressively if there is a lot of mapped 374 * memory around. To avoid stressing page reclaim with lots of unreclaimable 375 * pages. It is better to clamp down on writers than to start swapping, and 376 * performing lots of scanning. 377 * 378 * We only allow 1/2 of the currently-unmapped memory to be dirtied. 379 * 380 * We don't permit the clamping level to fall below 5% - that is getting rather 381 * excessive. 382 * 383 * We make sure that the background writeout level is below the adjusted 384 * clamping level. 385 */ 386 387static unsigned long highmem_dirtyable_memory(unsigned long total) 388{ 389#ifdef CONFIG_HIGHMEM 390 int node; 391 unsigned long x = 0; 392 393 for_each_node_state(node, N_HIGH_MEMORY) { 394 struct zone *z = 395 &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; 396 397 x += zone_page_state(z, NR_FREE_PAGES) + zone_lru_pages(z); 398 } 399 /* 400 * Make sure that the number of highmem pages is never larger 401 * than the number of the total dirtyable memory. This can only 402 * occur in very strange VM situations but we want to make sure 403 * that this does not occur. 404 */ 405 return min(x, total); 406#else 407 return 0; 408#endif 409} 410 411/** 412 * determine_dirtyable_memory - amount of memory that may be used 413 * 414 * Returns the numebr of pages that can currently be freed and used 415 * by the kernel for direct mappings. 416 */ 417unsigned long determine_dirtyable_memory(void) 418{ 419 unsigned long x; 420 421 x = global_page_state(NR_FREE_PAGES) + global_lru_pages(); 422 423 if (!vm_highmem_is_dirtyable) 424 x -= highmem_dirtyable_memory(x); 425 426 return x + 1; /* Ensure that we never return 0 */ 427} 428 429void 430get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty, 431 unsigned long *pbdi_dirty, struct backing_dev_info *bdi) 432{ 433 unsigned long background; 434 unsigned long dirty; 435 unsigned long available_memory = determine_dirtyable_memory(); 436 struct task_struct *tsk; 437 438 if (vm_dirty_bytes) 439 dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE); 440 else { 441 int dirty_ratio; 442 443 dirty_ratio = vm_dirty_ratio; 444 if (dirty_ratio < 5) 445 dirty_ratio = 5; 446 dirty = (dirty_ratio * available_memory) / 100; 447 } 448 449 if (dirty_background_bytes) 450 background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE); 451 else 452 background = (dirty_background_ratio * available_memory) / 100; 453 454 if (background >= dirty) 455 background = dirty / 2; 456 tsk = current; 457 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { 458 background += background / 4; 459 dirty += dirty / 4; 460 } 461 *pbackground = background; 462 *pdirty = dirty; 463 464 if (bdi) { 465 u64 bdi_dirty; 466 long numerator, denominator; 467 468 /* 469 * Calculate this BDI's share of the dirty ratio. 470 */ 471 bdi_writeout_fraction(bdi, &numerator, &denominator); 472 473 bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; 474 bdi_dirty *= numerator; 475 do_div(bdi_dirty, denominator); 476 bdi_dirty += (dirty * bdi->min_ratio) / 100; 477 if (bdi_dirty > (dirty * bdi->max_ratio) / 100) 478 bdi_dirty = dirty * bdi->max_ratio / 100; 479 480 *pbdi_dirty = bdi_dirty; 481 clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); 482 task_dirty_limit(current, pbdi_dirty); 483 } 484} 485 486/* 487 * balance_dirty_pages() must be called by processes which are generating dirty 488 * data. It looks at the number of dirty pages in the machine and will force 489 * the caller to perform writeback if the system is over `vm_dirty_ratio'. 490 * If we're over `background_thresh' then pdflush is woken to perform some 491 * writeout. 492 */ 493static void balance_dirty_pages(struct address_space *mapping) 494{ 495 long nr_reclaimable, bdi_nr_reclaimable; 496 long nr_writeback, bdi_nr_writeback; 497 unsigned long background_thresh; 498 unsigned long dirty_thresh; 499 unsigned long bdi_thresh; 500 unsigned long pages_written = 0; 501 unsigned long write_chunk = sync_writeback_pages(); 502 503 struct backing_dev_info *bdi = mapping->backing_dev_info; 504 505 for (;;) { 506 struct writeback_control wbc = { 507 .bdi = bdi, 508 .sync_mode = WB_SYNC_NONE, 509 .older_than_this = NULL, 510 .nr_to_write = write_chunk, 511 .range_cyclic = 1, 512 }; 513 514 get_dirty_limits(&background_thresh, &dirty_thresh, 515 &bdi_thresh, bdi); 516 517 nr_reclaimable = global_page_state(NR_FILE_DIRTY) + 518 global_page_state(NR_UNSTABLE_NFS); 519 nr_writeback = global_page_state(NR_WRITEBACK); 520 521 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); 522 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); 523 524 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) 525 break; 526 527 /* 528 * Throttle it only when the background writeback cannot 529 * catch-up. This avoids (excessively) small writeouts 530 * when the bdi limits are ramping up. 531 */ 532 if (nr_reclaimable + nr_writeback < 533 (background_thresh + dirty_thresh) / 2) 534 break; 535 536 if (!bdi->dirty_exceeded) 537 bdi->dirty_exceeded = 1; 538 539 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. 540 * Unstable writes are a feature of certain networked 541 * filesystems (i.e. NFS) in which data may have been 542 * written to the server's write cache, but has not yet 543 * been flushed to permanent storage. 544 */ 545 if (bdi_nr_reclaimable) { 546 writeback_inodes(&wbc); 547 pages_written += write_chunk - wbc.nr_to_write; 548 get_dirty_limits(&background_thresh, &dirty_thresh, 549 &bdi_thresh, bdi); 550 } 551 552 /* 553 * In order to avoid the stacked BDI deadlock we need 554 * to ensure we accurately count the 'dirty' pages when 555 * the threshold is low. 556 * 557 * Otherwise it would be possible to get thresh+n pages 558 * reported dirty, even though there are thresh-m pages 559 * actually dirty; with m+n sitting in the percpu 560 * deltas. 561 */ 562 if (bdi_thresh < 2*bdi_stat_error(bdi)) { 563 bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); 564 bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); 565 } else if (bdi_nr_reclaimable) { 566 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); 567 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); 568 } 569 570 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) 571 break; 572 if (pages_written >= write_chunk) 573 break; /* We've done our duty */ 574 575 congestion_wait(WRITE, HZ/10); 576 } 577 578 if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh && 579 bdi->dirty_exceeded) 580 bdi->dirty_exceeded = 0; 581 582 if (writeback_in_progress(bdi)) 583 return; /* pdflush is already working this queue */ 584 585 /* 586 * In laptop mode, we wait until hitting the higher threshold before 587 * starting background writeout, and then write out all the way down 588 * to the lower threshold. So slow writers cause minimal disk activity. 589 * 590 * In normal mode, we start background writeout at the lower 591 * background_thresh, to keep the amount of dirty memory low. 592 */ 593 if ((laptop_mode && pages_written) || 594 (!laptop_mode && (global_page_state(NR_FILE_DIRTY) 595 + global_page_state(NR_UNSTABLE_NFS) 596 > background_thresh))) 597 pdflush_operation(background_writeout, 0); 598} 599 600void set_page_dirty_balance(struct page *page, int page_mkwrite) 601{ 602 if (set_page_dirty(page) || page_mkwrite) { 603 struct address_space *mapping = page_mapping(page); 604 605 if (mapping) 606 balance_dirty_pages_ratelimited(mapping); 607 } 608} 609 610static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0; 611 612/** 613 * balance_dirty_pages_ratelimited_nr - balance dirty memory state 614 * @mapping: address_space which was dirtied 615 * @nr_pages_dirtied: number of pages which the caller has just dirtied 616 * 617 * Processes which are dirtying memory should call in here once for each page 618 * which was newly dirtied. The function will periodically check the system's 619 * dirty state and will initiate writeback if needed. 620 * 621 * On really big machines, get_writeback_state is expensive, so try to avoid 622 * calling it too often (ratelimiting). But once we're over the dirty memory 623 * limit we decrease the ratelimiting by a lot, to prevent individual processes 624 * from overshooting the limit by (ratelimit_pages) each. 625 */ 626void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, 627 unsigned long nr_pages_dirtied) 628{ 629 unsigned long ratelimit; 630 unsigned long *p; 631 632 ratelimit = ratelimit_pages; 633 if (mapping->backing_dev_info->dirty_exceeded) 634 ratelimit = 8; 635 636 /* 637 * Check the rate limiting. Also, we do not want to throttle real-time 638 * tasks in balance_dirty_pages(). Period. 639 */ 640 preempt_disable(); 641 p = &__get_cpu_var(bdp_ratelimits); 642 *p += nr_pages_dirtied; 643 if (unlikely(*p >= ratelimit)) { 644 *p = 0; 645 preempt_enable(); 646 balance_dirty_pages(mapping); 647 return; 648 } 649 preempt_enable(); 650} 651EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); 652 653void throttle_vm_writeout(gfp_t gfp_mask) 654{ 655 unsigned long background_thresh; 656 unsigned long dirty_thresh; 657 658 for ( ; ; ) { 659 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 660 661 /* 662 * Boost the allowable dirty threshold a bit for page 663 * allocators so they don't get DoS'ed by heavy writers 664 */ 665 dirty_thresh += dirty_thresh / 10; /* wheeee... */ 666 667 if (global_page_state(NR_UNSTABLE_NFS) + 668 global_page_state(NR_WRITEBACK) <= dirty_thresh) 669 break; 670 congestion_wait(WRITE, HZ/10); 671 672 /* 673 * The caller might hold locks which can prevent IO completion 674 * or progress in the filesystem. So we cannot just sit here 675 * waiting for IO to complete. 676 */ 677 if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) 678 break; 679 } 680} 681 682/* 683 * writeback at least _min_pages, and keep writing until the amount of dirty 684 * memory is less than the background threshold, or until we're all clean. 685 */ 686static void background_writeout(unsigned long _min_pages) 687{ 688 long min_pages = _min_pages; 689 struct writeback_control wbc = { 690 .bdi = NULL, 691 .sync_mode = WB_SYNC_NONE, 692 .older_than_this = NULL, 693 .nr_to_write = 0, 694 .nonblocking = 1, 695 .range_cyclic = 1, 696 }; 697 698 for ( ; ; ) { 699 unsigned long background_thresh; 700 unsigned long dirty_thresh; 701 702 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 703 if (global_page_state(NR_FILE_DIRTY) + 704 global_page_state(NR_UNSTABLE_NFS) < background_thresh 705 && min_pages <= 0) 706 break; 707 wbc.more_io = 0; 708 wbc.encountered_congestion = 0; 709 wbc.nr_to_write = MAX_WRITEBACK_PAGES; 710 wbc.pages_skipped = 0; 711 writeback_inodes(&wbc); 712 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 713 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { 714 /* Wrote less than expected */ 715 if (wbc.encountered_congestion || wbc.more_io) 716 congestion_wait(WRITE, HZ/10); 717 else 718 break; 719 } 720 } 721} 722 723/* 724 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 725 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns 726 * -1 if all pdflush threads were busy. 727 */ 728int wakeup_pdflush(long nr_pages) 729{ 730 if (nr_pages == 0) 731 nr_pages = global_page_state(NR_FILE_DIRTY) + 732 global_page_state(NR_UNSTABLE_NFS); 733 return pdflush_operation(background_writeout, nr_pages); 734} 735 736static void wb_timer_fn(unsigned long unused); 737static void laptop_timer_fn(unsigned long unused); 738 739static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0); 740static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0); 741 742/* 743 * Periodic writeback of "old" data. 744 * 745 * Define "old": the first time one of an inode's pages is dirtied, we mark the 746 * dirtying-time in the inode's address_space. So this periodic writeback code 747 * just walks the superblock inode list, writing back any inodes which are 748 * older than a specific point in time. 749 * 750 * Try to run once per dirty_writeback_interval. But if a writeback event 751 * takes longer than a dirty_writeback_interval interval, then leave a 752 * one-second gap. 753 * 754 * older_than_this takes precedence over nr_to_write. So we'll only write back 755 * all dirty pages if they are all attached to "old" mappings. 756 */ 757static void wb_kupdate(unsigned long arg) 758{ 759 unsigned long oldest_jif; 760 unsigned long start_jif; 761 unsigned long next_jif; 762 long nr_to_write; 763 struct writeback_control wbc = { 764 .bdi = NULL, 765 .sync_mode = WB_SYNC_NONE, 766 .older_than_this = &oldest_jif, 767 .nr_to_write = 0, 768 .nonblocking = 1, 769 .for_kupdate = 1, 770 .range_cyclic = 1, 771 }; 772 773 sync_supers(); 774 775 oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10); 776 start_jif = jiffies; 777 next_jif = start_jif + msecs_to_jiffies(dirty_writeback_interval * 10); 778 nr_to_write = global_page_state(NR_FILE_DIRTY) + 779 global_page_state(NR_UNSTABLE_NFS) + 780 (inodes_stat.nr_inodes - inodes_stat.nr_unused); 781 while (nr_to_write > 0) { 782 wbc.more_io = 0; 783 wbc.encountered_congestion = 0; 784 wbc.nr_to_write = MAX_WRITEBACK_PAGES; 785 writeback_inodes(&wbc); 786 if (wbc.nr_to_write > 0) { 787 if (wbc.encountered_congestion || wbc.more_io) 788 congestion_wait(WRITE, HZ/10); 789 else 790 break; /* All the old data is written */ 791 } 792 nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 793 } 794 if (time_before(next_jif, jiffies + HZ)) 795 next_jif = jiffies + HZ; 796 if (dirty_writeback_interval) 797 mod_timer(&wb_timer, next_jif); 798} 799 800/* 801 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs 802 */ 803int dirty_writeback_centisecs_handler(ctl_table *table, int write, 804 struct file *file, void __user *buffer, size_t *length, loff_t *ppos) 805{ 806 proc_dointvec(table, write, file, buffer, length, ppos); 807 if (dirty_writeback_interval) 808 mod_timer(&wb_timer, jiffies + 809 msecs_to_jiffies(dirty_writeback_interval * 10)); 810 else 811 del_timer(&wb_timer); 812 return 0; 813} 814 815static void wb_timer_fn(unsigned long unused) 816{ 817 if (pdflush_operation(wb_kupdate, 0) < 0) 818 mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */ 819} 820 821static void laptop_flush(unsigned long unused) 822{ 823 sys_sync(); 824} 825 826static void laptop_timer_fn(unsigned long unused) 827{ 828 pdflush_operation(laptop_flush, 0); 829} 830 831/* 832 * We've spun up the disk and we're in laptop mode: schedule writeback 833 * of all dirty data a few seconds from now. If the flush is already scheduled 834 * then push it back - the user is still using the disk. 835 */ 836void laptop_io_completion(void) 837{ 838 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode); 839} 840 841/* 842 * We're in laptop mode and we've just synced. The sync's writes will have 843 * caused another writeback to be scheduled by laptop_io_completion. 844 * Nothing needs to be written back anymore, so we unschedule the writeback. 845 */ 846void laptop_sync_completion(void) 847{ 848 del_timer(&laptop_mode_wb_timer); 849} 850 851/* 852 * If ratelimit_pages is too high then we can get into dirty-data overload 853 * if a large number of processes all perform writes at the same time. 854 * If it is too low then SMP machines will call the (expensive) 855 * get_writeback_state too often. 856 * 857 * Here we set ratelimit_pages to a level which ensures that when all CPUs are 858 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory 859 * thresholds before writeback cuts in. 860 * 861 * But the limit should not be set too high. Because it also controls the 862 * amount of memory which the balance_dirty_pages() caller has to write back. 863 * If this is too large then the caller will block on the IO queue all the 864 * time. So limit it to four megabytes - the balance_dirty_pages() caller 865 * will write six megabyte chunks, max. 866 */ 867 868void writeback_set_ratelimit(void) 869{ 870 ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); 871 if (ratelimit_pages < 16) 872 ratelimit_pages = 16; 873 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) 874 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; 875} 876 877static int __cpuinit 878ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) 879{ 880 writeback_set_ratelimit(); 881 return NOTIFY_DONE; 882} 883 884static struct notifier_block __cpuinitdata ratelimit_nb = { 885 .notifier_call = ratelimit_handler, 886 .next = NULL, 887}; 888 889/* 890 * Called early on to tune the page writeback dirty limits. 891 * 892 * We used to scale dirty pages according to how total memory 893 * related to pages that could be allocated for buffers (by 894 * comparing nr_free_buffer_pages() to vm_total_pages. 895 * 896 * However, that was when we used "dirty_ratio" to scale with 897 * all memory, and we don't do that any more. "dirty_ratio" 898 * is now applied to total non-HIGHPAGE memory (by subtracting 899 * totalhigh_pages from vm_total_pages), and as such we can't 900 * get into the old insane situation any more where we had 901 * large amounts of dirty pages compared to a small amount of 902 * non-HIGHMEM memory. 903 * 904 * But we might still want to scale the dirty_ratio by how 905 * much memory the box has.. 906 */ 907void __init page_writeback_init(void) 908{ 909 int shift; 910 911 mod_timer(&wb_timer, 912 jiffies + msecs_to_jiffies(dirty_writeback_interval * 10)); 913 writeback_set_ratelimit(); 914 register_cpu_notifier(&ratelimit_nb); 915 916 shift = calc_period_shift(); 917 prop_descriptor_init(&vm_completions, shift); 918 prop_descriptor_init(&vm_dirties, shift); 919} 920 921/** 922 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. 923 * @mapping: address space structure to write 924 * @wbc: subtract the number of written pages from *@wbc->nr_to_write 925 * @writepage: function called for each page 926 * @data: data passed to writepage function 927 * 928 * If a page is already under I/O, write_cache_pages() skips it, even 929 * if it's dirty. This is desirable behaviour for memory-cleaning writeback, 930 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() 931 * and msync() need to guarantee that all the data which was dirty at the time 932 * the call was made get new I/O started against them. If wbc->sync_mode is 933 * WB_SYNC_ALL then we were called for data integrity and we must wait for 934 * existing IO to complete. 935 */ 936int write_cache_pages(struct address_space *mapping, 937 struct writeback_control *wbc, writepage_t writepage, 938 void *data) 939{ 940 struct backing_dev_info *bdi = mapping->backing_dev_info; 941 int ret = 0; 942 int done = 0; 943 struct pagevec pvec; 944 int nr_pages; 945 pgoff_t uninitialized_var(writeback_index); 946 pgoff_t index; 947 pgoff_t end; /* Inclusive */ 948 pgoff_t done_index; 949 int cycled; 950 int range_whole = 0; 951 long nr_to_write = wbc->nr_to_write; 952 953 if (wbc->nonblocking && bdi_write_congested(bdi)) { 954 wbc->encountered_congestion = 1; 955 return 0; 956 } 957 958 pagevec_init(&pvec, 0); 959 if (wbc->range_cyclic) { 960 writeback_index = mapping->writeback_index; /* prev offset */ 961 index = writeback_index; 962 if (index == 0) 963 cycled = 1; 964 else 965 cycled = 0; 966 end = -1; 967 } else { 968 index = wbc->range_start >> PAGE_CACHE_SHIFT; 969 end = wbc->range_end >> PAGE_CACHE_SHIFT; 970 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 971 range_whole = 1; 972 cycled = 1; /* ignore range_cyclic tests */ 973 } 974retry: 975 done_index = index; 976 while (!done && (index <= end)) { 977 int i; 978 979 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 980 PAGECACHE_TAG_DIRTY, 981 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 982 if (nr_pages == 0) 983 break; 984 985 for (i = 0; i < nr_pages; i++) { 986 struct page *page = pvec.pages[i]; 987 988 /* 989 * At this point, the page may be truncated or 990 * invalidated (changing page->mapping to NULL), or 991 * even swizzled back from swapper_space to tmpfs file 992 * mapping. However, page->index will not change 993 * because we have a reference on the page. 994 */ 995 if (page->index > end) { 996 /* 997 * can't be range_cyclic (1st pass) because 998 * end == -1 in that case. 999 */ 1000 done = 1; 1001 break; 1002 } 1003 1004 done_index = page->index + 1; 1005 1006 lock_page(page); 1007 1008 /* 1009 * Page truncated or invalidated. We can freely skip it 1010 * then, even for data integrity operations: the page 1011 * has disappeared concurrently, so there could be no 1012 * real expectation of this data interity operation 1013 * even if there is now a new, dirty page at the same 1014 * pagecache address. 1015 */ 1016 if (unlikely(page->mapping != mapping)) { 1017continue_unlock: 1018 unlock_page(page); 1019 continue; 1020 } 1021 1022 if (!PageDirty(page)) { 1023 /* someone wrote it for us */ 1024 goto continue_unlock; 1025 } 1026 1027 if (PageWriteback(page)) { 1028 if (wbc->sync_mode != WB_SYNC_NONE) 1029 wait_on_page_writeback(page); 1030 else 1031 goto continue_unlock; 1032 } 1033 1034 BUG_ON(PageWriteback(page)); 1035 if (!clear_page_dirty_for_io(page)) 1036 goto continue_unlock; 1037 1038 ret = (*writepage)(page, wbc, data); 1039 if (unlikely(ret)) { 1040 if (ret == AOP_WRITEPAGE_ACTIVATE) { 1041 unlock_page(page); 1042 ret = 0; 1043 } else { 1044 /* 1045 * done_index is set past this page, 1046 * so media errors will not choke 1047 * background writeout for the entire 1048 * file. This has consequences for 1049 * range_cyclic semantics (ie. it may 1050 * not be suitable for data integrity 1051 * writeout). 1052 */ 1053 done = 1; 1054 break; 1055 } 1056 } 1057 1058 if (nr_to_write > 0) { 1059 nr_to_write--; 1060 if (nr_to_write == 0 && 1061 wbc->sync_mode == WB_SYNC_NONE) { 1062 /* 1063 * We stop writing back only if we are 1064 * not doing integrity sync. In case of 1065 * integrity sync we have to keep going 1066 * because someone may be concurrently 1067 * dirtying pages, and we might have 1068 * synced a lot of newly appeared dirty 1069 * pages, but have not synced all of the 1070 * old dirty pages. 1071 */ 1072 done = 1; 1073 break; 1074 } 1075 } 1076 1077 if (wbc->nonblocking && bdi_write_congested(bdi)) { 1078 wbc->encountered_congestion = 1; 1079 done = 1; 1080 break; 1081 } 1082 } 1083 pagevec_release(&pvec); 1084 cond_resched(); 1085 } 1086 if (!cycled && !done) { 1087 /* 1088 * range_cyclic: 1089 * We hit the last page and there is more work to be done: wrap 1090 * back to the start of the file 1091 */ 1092 cycled = 1; 1093 index = 0; 1094 end = writeback_index - 1; 1095 goto retry; 1096 } 1097 if (!wbc->no_nrwrite_index_update) { 1098 if (wbc->range_cyclic || (range_whole && nr_to_write > 0)) 1099 mapping->writeback_index = done_index; 1100 wbc->nr_to_write = nr_to_write; 1101 } 1102 1103 return ret; 1104} 1105EXPORT_SYMBOL(write_cache_pages); 1106 1107/* 1108 * Function used by generic_writepages to call the real writepage 1109 * function and set the mapping flags on error 1110 */ 1111static int __writepage(struct page *page, struct writeback_control *wbc, 1112 void *data) 1113{ 1114 struct address_space *mapping = data; 1115 int ret = mapping->a_ops->writepage(page, wbc); 1116 mapping_set_error(mapping, ret); 1117 return ret; 1118} 1119 1120/** 1121 * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. 1122 * @mapping: address space structure to write 1123 * @wbc: subtract the number of written pages from *@wbc->nr_to_write 1124 * 1125 * This is a library function, which implements the writepages() 1126 * address_space_operation. 1127 */ 1128int generic_writepages(struct address_space *mapping, 1129 struct writeback_control *wbc) 1130{ 1131 /* deal with chardevs and other special file */ 1132 if (!mapping->a_ops->writepage) 1133 return 0; 1134 1135 return write_cache_pages(mapping, wbc, __writepage, mapping); 1136} 1137 1138EXPORT_SYMBOL(generic_writepages); 1139 1140int do_writepages(struct address_space *mapping, struct writeback_control *wbc) 1141{ 1142 int ret; 1143 1144 if (wbc->nr_to_write <= 0) 1145 return 0; 1146 wbc->for_writepages = 1; 1147 if (mapping->a_ops->writepages) 1148 ret = mapping->a_ops->writepages(mapping, wbc); 1149 else 1150 ret = generic_writepages(mapping, wbc); 1151 wbc->for_writepages = 0; 1152 return ret; 1153} 1154 1155/** 1156 * write_one_page - write out a single page and optionally wait on I/O 1157 * @page: the page to write 1158 * @wait: if true, wait on writeout 1159 * 1160 * The page must be locked by the caller and will be unlocked upon return. 1161 * 1162 * write_one_page() returns a negative error code if I/O failed. 1163 */ 1164int write_one_page(struct page *page, int wait) 1165{ 1166 struct address_space *mapping = page->mapping; 1167 int ret = 0; 1168 struct writeback_control wbc = { 1169 .sync_mode = WB_SYNC_ALL, 1170 .nr_to_write = 1, 1171 }; 1172 1173 BUG_ON(!PageLocked(page)); 1174 1175 if (wait) 1176 wait_on_page_writeback(page); 1177 1178 if (clear_page_dirty_for_io(page)) { 1179 page_cache_get(page); 1180 ret = mapping->a_ops->writepage(page, &wbc); 1181 if (ret == 0 && wait) { 1182 wait_on_page_writeback(page); 1183 if (PageError(page)) 1184 ret = -EIO; 1185 } 1186 page_cache_release(page); 1187 } else { 1188 unlock_page(page); 1189 } 1190 return ret; 1191} 1192EXPORT_SYMBOL(write_one_page); 1193 1194/* 1195 * For address_spaces which do not use buffers nor write back. 1196 */ 1197int __set_page_dirty_no_writeback(struct page *page) 1198{ 1199 if (!PageDirty(page)) 1200 SetPageDirty(page); 1201 return 0; 1202} 1203 1204/* 1205 * Helper function for set_page_dirty family. 1206 * NOTE: This relies on being atomic wrt interrupts. 1207 */ 1208void account_page_dirtied(struct page *page, struct address_space *mapping) 1209{ 1210 if (mapping_cap_account_dirty(mapping)) { 1211 __inc_zone_page_state(page, NR_FILE_DIRTY); 1212 __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); 1213 task_dirty_inc(current); 1214 task_io_account_write(PAGE_CACHE_SIZE); 1215 } 1216} 1217 1218/* 1219 * For address_spaces which do not use buffers. Just tag the page as dirty in 1220 * its radix tree. 1221 * 1222 * This is also used when a single buffer is being dirtied: we want to set the 1223 * page dirty in that case, but not all the buffers. This is a "bottom-up" 1224 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. 1225 * 1226 * Most callers have locked the page, which pins the address_space in memory. 1227 * But zap_pte_range() does not lock the page, however in that case the 1228 * mapping is pinned by the vma's ->vm_file reference. 1229 * 1230 * We take care to handle the case where the page was truncated from the 1231 * mapping by re-checking page_mapping() inside tree_lock. 1232 */ 1233int __set_page_dirty_nobuffers(struct page *page) 1234{ 1235 if (!TestSetPageDirty(page)) { 1236 struct address_space *mapping = page_mapping(page); 1237 struct address_space *mapping2; 1238 1239 if (!mapping) 1240 return 1; 1241 1242 spin_lock_irq(&mapping->tree_lock); 1243 mapping2 = page_mapping(page); 1244 if (mapping2) { /* Race with truncate? */ 1245 BUG_ON(mapping2 != mapping); 1246 WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); 1247 account_page_dirtied(page, mapping); 1248 radix_tree_tag_set(&mapping->page_tree, 1249 page_index(page), PAGECACHE_TAG_DIRTY); 1250 } 1251 spin_unlock_irq(&mapping->tree_lock); 1252 if (mapping->host) { 1253 /* !PageAnon && !swapper_space */ 1254 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1255 } 1256 return 1; 1257 } 1258 return 0; 1259} 1260EXPORT_SYMBOL(__set_page_dirty_nobuffers); 1261 1262/* 1263 * When a writepage implementation decides that it doesn't want to write this 1264 * page for some reason, it should redirty the locked page via 1265 * redirty_page_for_writepage() and it should then unlock the page and return 0 1266 */ 1267int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) 1268{ 1269 wbc->pages_skipped++; 1270 return __set_page_dirty_nobuffers(page); 1271} 1272EXPORT_SYMBOL(redirty_page_for_writepage); 1273 1274/* 1275 * If the mapping doesn't provide a set_page_dirty a_op, then 1276 * just fall through and assume that it wants buffer_heads. 1277 */ 1278int set_page_dirty(struct page *page) 1279{ 1280 struct address_space *mapping = page_mapping(page); 1281 1282 if (likely(mapping)) { 1283 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; 1284#ifdef CONFIG_BLOCK 1285 if (!spd) 1286 spd = __set_page_dirty_buffers; 1287#endif 1288 return (*spd)(page); 1289 } 1290 if (!PageDirty(page)) { 1291 if (!TestSetPageDirty(page)) 1292 return 1; 1293 } 1294 return 0; 1295} 1296EXPORT_SYMBOL(set_page_dirty); 1297 1298/* 1299 * set_page_dirty() is racy if the caller has no reference against 1300 * page->mapping->host, and if the page is unlocked. This is because another 1301 * CPU could truncate the page off the mapping and then free the mapping. 1302 * 1303 * Usually, the page _is_ locked, or the caller is a user-space process which 1304 * holds a reference on the inode by having an open file. 1305 * 1306 * In other cases, the page should be locked before running set_page_dirty(). 1307 */ 1308int set_page_dirty_lock(struct page *page) 1309{ 1310 int ret; 1311 1312 lock_page_nosync(page); 1313 ret = set_page_dirty(page); 1314 unlock_page(page); 1315 return ret; 1316} 1317EXPORT_SYMBOL(set_page_dirty_lock); 1318 1319/* 1320 * Clear a page's dirty flag, while caring for dirty memory accounting. 1321 * Returns true if the page was previously dirty. 1322 * 1323 * This is for preparing to put the page under writeout. We leave the page 1324 * tagged as dirty in the radix tree so that a concurrent write-for-sync 1325 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage 1326 * implementation will run either set_page_writeback() or set_page_dirty(), 1327 * at which stage we bring the page's dirty flag and radix-tree dirty tag 1328 * back into sync. 1329 * 1330 * This incoherency between the page's dirty flag and radix-tree tag is 1331 * unfortunate, but it only exists while the page is locked. 1332 */ 1333int clear_page_dirty_for_io(struct page *page) 1334{ 1335 struct address_space *mapping = page_mapping(page); 1336 1337 BUG_ON(!PageLocked(page)); 1338 1339 ClearPageReclaim(page); 1340 if (mapping && mapping_cap_account_dirty(mapping)) { 1341 /* 1342 * Yes, Virginia, this is indeed insane. 1343 * 1344 * We use this sequence to make sure that 1345 * (a) we account for dirty stats properly 1346 * (b) we tell the low-level filesystem to 1347 * mark the whole page dirty if it was 1348 * dirty in a pagetable. Only to then 1349 * (c) clean the page again and return 1 to 1350 * cause the writeback. 1351 * 1352 * This way we avoid all nasty races with the 1353 * dirty bit in multiple places and clearing 1354 * them concurrently from different threads. 1355 * 1356 * Note! Normally the "set_page_dirty(page)" 1357 * has no effect on the actual dirty bit - since 1358 * that will already usually be set. But we 1359 * need the side effects, and it can help us 1360 * avoid races. 1361 * 1362 * We basically use the page "master dirty bit" 1363 * as a serialization point for all the different 1364 * threads doing their things. 1365 */ 1366 if (page_mkclean(page)) 1367 set_page_dirty(page); 1368 /* 1369 * We carefully synchronise fault handlers against 1370 * installing a dirty pte and marking the page dirty 1371 * at this point. We do this by having them hold the 1372 * page lock at some point after installing their 1373 * pte, but before marking the page dirty. 1374 * Pages are always locked coming in here, so we get 1375 * the desired exclusion. See mm/memory.c:do_wp_page() 1376 * for more comments. 1377 */ 1378 if (TestClearPageDirty(page)) { 1379 dec_zone_page_state(page, NR_FILE_DIRTY); 1380 dec_bdi_stat(mapping->backing_dev_info, 1381 BDI_RECLAIMABLE); 1382 return 1; 1383 } 1384 return 0; 1385 } 1386 return TestClearPageDirty(page); 1387} 1388EXPORT_SYMBOL(clear_page_dirty_for_io); 1389 1390int test_clear_page_writeback(struct page *page) 1391{ 1392 struct address_space *mapping = page_mapping(page); 1393 int ret; 1394 1395 if (mapping) { 1396 struct backing_dev_info *bdi = mapping->backing_dev_info; 1397 unsigned long flags; 1398 1399 spin_lock_irqsave(&mapping->tree_lock, flags); 1400 ret = TestClearPageWriteback(page); 1401 if (ret) { 1402 radix_tree_tag_clear(&mapping->page_tree, 1403 page_index(page), 1404 PAGECACHE_TAG_WRITEBACK); 1405 if (bdi_cap_account_writeback(bdi)) { 1406 __dec_bdi_stat(bdi, BDI_WRITEBACK); 1407 __bdi_writeout_inc(bdi); 1408 } 1409 } 1410 spin_unlock_irqrestore(&mapping->tree_lock, flags); 1411 } else { 1412 ret = TestClearPageWriteback(page); 1413 } 1414 if (ret) 1415 dec_zone_page_state(page, NR_WRITEBACK); 1416 return ret; 1417} 1418 1419int test_set_page_writeback(struct page *page) 1420{ 1421 struct address_space *mapping = page_mapping(page); 1422 int ret; 1423 1424 if (mapping) { 1425 struct backing_dev_info *bdi = mapping->backing_dev_info; 1426 unsigned long flags; 1427 1428 spin_lock_irqsave(&mapping->tree_lock, flags); 1429 ret = TestSetPageWriteback(page); 1430 if (!ret) { 1431 radix_tree_tag_set(&mapping->page_tree, 1432 page_index(page), 1433 PAGECACHE_TAG_WRITEBACK); 1434 if (bdi_cap_account_writeback(bdi)) 1435 __inc_bdi_stat(bdi, BDI_WRITEBACK); 1436 } 1437 if (!PageDirty(page)) 1438 radix_tree_tag_clear(&mapping->page_tree, 1439 page_index(page), 1440 PAGECACHE_TAG_DIRTY); 1441 spin_unlock_irqrestore(&mapping->tree_lock, flags); 1442 } else { 1443 ret = TestSetPageWriteback(page); 1444 } 1445 if (!ret) 1446 inc_zone_page_state(page, NR_WRITEBACK); 1447 return ret; 1448 1449} 1450EXPORT_SYMBOL(test_set_page_writeback); 1451 1452/* 1453 * Return true if any of the pages in the mapping are marked with the 1454 * passed tag. 1455 */ 1456int mapping_tagged(struct address_space *mapping, int tag) 1457{ 1458 int ret; 1459 rcu_read_lock(); 1460 ret = radix_tree_tagged(&mapping->page_tree, tag); 1461 rcu_read_unlock(); 1462 return ret; 1463} 1464EXPORT_SYMBOL(mapping_tagged); 1465