fs-writeback.c (36b2e922b5acd291051fab25bc7535274ce49532) fs-writeback.c (d46db3d58233be4be980eb1e42eebe7808bcabab)
1/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the

--- 16 unchanged lines hidden (view full) ---

25#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/buffer_head.h>
29#include <linux/tracepoint.h>
30#include "internal.h"
31
32/*
1/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the

--- 16 unchanged lines hidden (view full) ---

25#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/buffer_head.h>
29#include <linux/tracepoint.h>
30#include "internal.h"
31
32/*
33 * The maximum number of pages to writeout in a single bdi flush/kupdate
34 * operation. We do this so we don't hold I_SYNC against an inode for
35 * enormous amounts of time, which would block a userspace task which has
36 * been forced to throttle against that inode. Also, the code reevaluates
37 * the dirty each time it has written this many pages.
38 */
39#define MAX_WRITEBACK_PAGES 1024L
40
41/*
33 * Passed into wb_writeback(), essentially a subset of writeback_control
34 */
35struct wb_writeback_work {
36 long nr_pages;
37 struct super_block *sb;
42 * Passed into wb_writeback(), essentially a subset of writeback_control
43 */
44struct wb_writeback_work {
45 long nr_pages;
46 struct super_block *sb;
47 unsigned long *older_than_this;
38 enum writeback_sync_modes sync_mode;
48 enum writeback_sync_modes sync_mode;
49 unsigned int tagged_writepages:1;
39 unsigned int for_kupdate:1;
40 unsigned int range_cyclic:1;
41 unsigned int for_background:1;
42
43 struct list_head list; /* pending work list */
44 struct completion *done; /* set if the caller waits */
45};
46

--- 128 unchanged lines hidden (view full) ---

175 spin_unlock_bh(&bdi->wb_lock);
176}
177
178/*
179 * Remove the inode from the writeback list it is on.
180 */
181void inode_wb_list_del(struct inode *inode)
182{
50 unsigned int for_kupdate:1;
51 unsigned int range_cyclic:1;
52 unsigned int for_background:1;
53
54 struct list_head list; /* pending work list */
55 struct completion *done; /* set if the caller waits */
56};
57

--- 128 unchanged lines hidden (view full) ---

186 spin_unlock_bh(&bdi->wb_lock);
187}
188
189/*
190 * Remove the inode from the writeback list it is on.
191 */
192void inode_wb_list_del(struct inode *inode)
193{
183 spin_lock(&inode_wb_list_lock);
194 struct backing_dev_info *bdi = inode_to_bdi(inode);
195
196 spin_lock(&bdi->wb.list_lock);
184 list_del_init(&inode->i_wb_list);
197 list_del_init(&inode->i_wb_list);
185 spin_unlock(&inode_wb_list_lock);
198 spin_unlock(&bdi->wb.list_lock);
186}
187
199}
200
188
189/*
190 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
191 * furthest end of its superblock's dirty-inode list.
192 *
193 * Before stamping the inode's ->dirtied_when, we check to see whether it is
194 * already the most-recently-dirtied inode on the b_dirty list. If that is
195 * the case then the inode must have been redirtied while it was being written
196 * out and we don't reset its dirtied_when.
197 */
201/*
202 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
203 * furthest end of its superblock's dirty-inode list.
204 *
205 * Before stamping the inode's ->dirtied_when, we check to see whether it is
206 * already the most-recently-dirtied inode on the b_dirty list. If that is
207 * the case then the inode must have been redirtied while it was being written
208 * out and we don't reset its dirtied_when.
209 */
198static void redirty_tail(struct inode *inode)
210static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
199{
211{
200 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
201
202 assert_spin_locked(&inode_wb_list_lock);
212 assert_spin_locked(&wb->list_lock);
203 if (!list_empty(&wb->b_dirty)) {
204 struct inode *tail;
205
206 tail = wb_inode(wb->b_dirty.next);
207 if (time_before(inode->dirtied_when, tail->dirtied_when))
208 inode->dirtied_when = jiffies;
209 }
210 list_move(&inode->i_wb_list, &wb->b_dirty);
211}
212
213/*
214 * requeue inode for re-scanning after bdi->b_io list is exhausted.
215 */
213 if (!list_empty(&wb->b_dirty)) {
214 struct inode *tail;
215
216 tail = wb_inode(wb->b_dirty.next);
217 if (time_before(inode->dirtied_when, tail->dirtied_when))
218 inode->dirtied_when = jiffies;
219 }
220 list_move(&inode->i_wb_list, &wb->b_dirty);
221}
222
223/*
224 * requeue inode for re-scanning after bdi->b_io list is exhausted.
225 */
216static void requeue_io(struct inode *inode)
226static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
217{
227{
218 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
219
220 assert_spin_locked(&inode_wb_list_lock);
228 assert_spin_locked(&wb->list_lock);
221 list_move(&inode->i_wb_list, &wb->b_more_io);
222}
223
224static void inode_sync_complete(struct inode *inode)
225{
226 /*
227 * Prevent speculative execution through
229 list_move(&inode->i_wb_list, &wb->b_more_io);
230}
231
232static void inode_sync_complete(struct inode *inode)
233{
234 /*
235 * Prevent speculative execution through
228 * spin_unlock(&inode_wb_list_lock);
236 * spin_unlock(&wb->list_lock);
229 */
230
231 smp_mb();
232 wake_up_bit(&inode->i_state, __I_SYNC);
233}
234
235static bool inode_dirtied_after(struct inode *inode, unsigned long t)
236{

--- 8 unchanged lines hidden (view full) ---

245 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
246#endif
247 return ret;
248}
249
250/*
251 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
252 */
237 */
238
239 smp_mb();
240 wake_up_bit(&inode->i_state, __I_SYNC);
241}
242
243static bool inode_dirtied_after(struct inode *inode, unsigned long t)
244{

--- 8 unchanged lines hidden (view full) ---

253 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
254#endif
255 return ret;
256}
257
258/*
259 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
260 */
253static void move_expired_inodes(struct list_head *delaying_queue,
261static int move_expired_inodes(struct list_head *delaying_queue,
254 struct list_head *dispatch_queue,
262 struct list_head *dispatch_queue,
255 unsigned long *older_than_this)
263 unsigned long *older_than_this)
256{
257 LIST_HEAD(tmp);
258 struct list_head *pos, *node;
259 struct super_block *sb = NULL;
260 struct inode *inode;
261 int do_sb_sort = 0;
264{
265 LIST_HEAD(tmp);
266 struct list_head *pos, *node;
267 struct super_block *sb = NULL;
268 struct inode *inode;
269 int do_sb_sort = 0;
270 int moved = 0;
262
263 while (!list_empty(delaying_queue)) {
264 inode = wb_inode(delaying_queue->prev);
265 if (older_than_this &&
266 inode_dirtied_after(inode, *older_than_this))
267 break;
268 if (sb && sb != inode->i_sb)
269 do_sb_sort = 1;
270 sb = inode->i_sb;
271 list_move(&inode->i_wb_list, &tmp);
271
272 while (!list_empty(delaying_queue)) {
273 inode = wb_inode(delaying_queue->prev);
274 if (older_than_this &&
275 inode_dirtied_after(inode, *older_than_this))
276 break;
277 if (sb && sb != inode->i_sb)
278 do_sb_sort = 1;
279 sb = inode->i_sb;
280 list_move(&inode->i_wb_list, &tmp);
281 moved++;
272 }
273
274 /* just one sb in list, splice to dispatch_queue and we're done */
275 if (!do_sb_sort) {
276 list_splice(&tmp, dispatch_queue);
282 }
283
284 /* just one sb in list, splice to dispatch_queue and we're done */
285 if (!do_sb_sort) {
286 list_splice(&tmp, dispatch_queue);
277 return;
287 goto out;
278 }
279
280 /* Move inodes from one superblock together */
281 while (!list_empty(&tmp)) {
282 sb = wb_inode(tmp.prev)->i_sb;
283 list_for_each_prev_safe(pos, node, &tmp) {
284 inode = wb_inode(pos);
285 if (inode->i_sb == sb)
286 list_move(&inode->i_wb_list, dispatch_queue);
287 }
288 }
288 }
289
290 /* Move inodes from one superblock together */
291 while (!list_empty(&tmp)) {
292 sb = wb_inode(tmp.prev)->i_sb;
293 list_for_each_prev_safe(pos, node, &tmp) {
294 inode = wb_inode(pos);
295 if (inode->i_sb == sb)
296 list_move(&inode->i_wb_list, dispatch_queue);
297 }
298 }
299out:
300 return moved;
289}
290
291/*
292 * Queue all expired dirty inodes for io, eldest first.
293 * Before
294 * newly dirtied b_dirty b_io b_more_io
295 * =============> gf edc BA
296 * After
297 * newly dirtied b_dirty b_io b_more_io
298 * =============> g fBAedc
299 * |
300 * +--> dequeue for IO
301 */
302static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
303{
301}
302
303/*
304 * Queue all expired dirty inodes for io, eldest first.
305 * Before
306 * newly dirtied b_dirty b_io b_more_io
307 * =============> gf edc BA
308 * After
309 * newly dirtied b_dirty b_io b_more_io
310 * =============> g fBAedc
311 * |
312 * +--> dequeue for IO
313 */
314static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
315{
304 assert_spin_locked(&inode_wb_list_lock);
316 int moved;
317 assert_spin_locked(&wb->list_lock);
305 list_splice_init(&wb->b_more_io, &wb->b_io);
318 list_splice_init(&wb->b_more_io, &wb->b_io);
306 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
319 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
320 trace_writeback_queue_io(wb, older_than_this, moved);
307}
308
309static int write_inode(struct inode *inode, struct writeback_control *wbc)
310{
311 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
312 return inode->i_sb->s_op->write_inode(inode, wbc);
313 return 0;
314}
315
316/*
317 * Wait for writeback on an inode to complete.
318 */
321}
322
323static int write_inode(struct inode *inode, struct writeback_control *wbc)
324{
325 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
326 return inode->i_sb->s_op->write_inode(inode, wbc);
327 return 0;
328}
329
330/*
331 * Wait for writeback on an inode to complete.
332 */
319static void inode_wait_for_writeback(struct inode *inode)
333static void inode_wait_for_writeback(struct inode *inode,
334 struct bdi_writeback *wb)
320{
321 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
322 wait_queue_head_t *wqh;
323
324 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
325 while (inode->i_state & I_SYNC) {
326 spin_unlock(&inode->i_lock);
335{
336 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
337 wait_queue_head_t *wqh;
338
339 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
340 while (inode->i_state & I_SYNC) {
341 spin_unlock(&inode->i_lock);
327 spin_unlock(&inode_wb_list_lock);
342 spin_unlock(&wb->list_lock);
328 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
343 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
329 spin_lock(&inode_wb_list_lock);
344 spin_lock(&wb->list_lock);
330 spin_lock(&inode->i_lock);
331 }
332}
333
334/*
345 spin_lock(&inode->i_lock);
346 }
347}
348
349/*
335 * Write out an inode's dirty pages. Called under inode_wb_list_lock and
350 * Write out an inode's dirty pages. Called under wb->list_lock and
336 * inode->i_lock. Either the caller has an active reference on the inode or
337 * the inode has I_WILL_FREE set.
338 *
339 * If `wait' is set, wait on the writeout.
340 *
341 * The whole writeout design is quite complex and fragile. We want to avoid
342 * starvation of particular inodes when others are being redirtied, prevent
343 * livelocks, etc.
344 */
345static int
351 * inode->i_lock. Either the caller has an active reference on the inode or
352 * the inode has I_WILL_FREE set.
353 *
354 * If `wait' is set, wait on the writeout.
355 *
356 * The whole writeout design is quite complex and fragile. We want to avoid
357 * starvation of particular inodes when others are being redirtied, prevent
358 * livelocks, etc.
359 */
360static int
346writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
361writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
362 struct writeback_control *wbc)
347{
348 struct address_space *mapping = inode->i_mapping;
363{
364 struct address_space *mapping = inode->i_mapping;
365 long nr_to_write = wbc->nr_to_write;
349 unsigned dirty;
350 int ret;
351
366 unsigned dirty;
367 int ret;
368
352 assert_spin_locked(&inode_wb_list_lock);
369 assert_spin_locked(&wb->list_lock);
353 assert_spin_locked(&inode->i_lock);
354
355 if (!atomic_read(&inode->i_count))
356 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
357 else
358 WARN_ON(inode->i_state & I_WILL_FREE);
359
360 if (inode->i_state & I_SYNC) {
361 /*
362 * If this inode is locked for writeback and we are not doing
363 * writeback-for-data-integrity, move it to b_more_io so that
364 * writeback can proceed with the other inodes on s_io.
365 *
366 * We'll have another go at writing back this inode when we
367 * completed a full scan of b_io.
368 */
369 if (wbc->sync_mode != WB_SYNC_ALL) {
370 assert_spin_locked(&inode->i_lock);
371
372 if (!atomic_read(&inode->i_count))
373 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
374 else
375 WARN_ON(inode->i_state & I_WILL_FREE);
376
377 if (inode->i_state & I_SYNC) {
378 /*
379 * If this inode is locked for writeback and we are not doing
380 * writeback-for-data-integrity, move it to b_more_io so that
381 * writeback can proceed with the other inodes on s_io.
382 *
383 * We'll have another go at writing back this inode when we
384 * completed a full scan of b_io.
385 */
386 if (wbc->sync_mode != WB_SYNC_ALL) {
370 requeue_io(inode);
387 requeue_io(inode, wb);
388 trace_writeback_single_inode_requeue(inode, wbc,
389 nr_to_write);
371 return 0;
372 }
373
374 /*
375 * It's a data-integrity sync. We must wait.
376 */
390 return 0;
391 }
392
393 /*
394 * It's a data-integrity sync. We must wait.
395 */
377 inode_wait_for_writeback(inode);
396 inode_wait_for_writeback(inode, wb);
378 }
379
380 BUG_ON(inode->i_state & I_SYNC);
381
382 /* Set I_SYNC, reset I_DIRTY_PAGES */
383 inode->i_state |= I_SYNC;
384 inode->i_state &= ~I_DIRTY_PAGES;
385 spin_unlock(&inode->i_lock);
397 }
398
399 BUG_ON(inode->i_state & I_SYNC);
400
401 /* Set I_SYNC, reset I_DIRTY_PAGES */
402 inode->i_state |= I_SYNC;
403 inode->i_state &= ~I_DIRTY_PAGES;
404 spin_unlock(&inode->i_lock);
386 spin_unlock(&inode_wb_list_lock);
405 spin_unlock(&wb->list_lock);
387
388 ret = do_writepages(mapping, wbc);
389
390 /*
391 * Make sure to wait on the data before writing out the metadata.
392 * This is important for filesystems that modify metadata on data
393 * I/O completion.
394 */

--- 14 unchanged lines hidden (view full) ---

409 spin_unlock(&inode->i_lock);
410 /* Don't write the inode if only I_DIRTY_PAGES was set */
411 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
412 int err = write_inode(inode, wbc);
413 if (ret == 0)
414 ret = err;
415 }
416
406
407 ret = do_writepages(mapping, wbc);
408
409 /*
410 * Make sure to wait on the data before writing out the metadata.
411 * This is important for filesystems that modify metadata on data
412 * I/O completion.
413 */

--- 14 unchanged lines hidden (view full) ---

428 spin_unlock(&inode->i_lock);
429 /* Don't write the inode if only I_DIRTY_PAGES was set */
430 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
431 int err = write_inode(inode, wbc);
432 if (ret == 0)
433 ret = err;
434 }
435
417 spin_lock(&inode_wb_list_lock);
436 spin_lock(&wb->list_lock);
418 spin_lock(&inode->i_lock);
419 inode->i_state &= ~I_SYNC;
420 if (!(inode->i_state & I_FREEING)) {
437 spin_lock(&inode->i_lock);
438 inode->i_state &= ~I_SYNC;
439 if (!(inode->i_state & I_FREEING)) {
440 /*
441 * Sync livelock prevention. Each inode is tagged and synced in
442 * one shot. If still dirty, it will be redirty_tail()'ed below.
443 * Update the dirty time to prevent enqueue and sync it again.
444 */
445 if ((inode->i_state & I_DIRTY) &&
446 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
447 inode->dirtied_when = jiffies;
448
421 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
422 /*
423 * We didn't write back all the pages. nfs_writepages()
424 * sometimes bales out without doing anything.
425 */
426 inode->i_state |= I_DIRTY_PAGES;
427 if (wbc->nr_to_write <= 0) {
428 /*
429 * slice used up: queue for next turn
430 */
449 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
450 /*
451 * We didn't write back all the pages. nfs_writepages()
452 * sometimes bales out without doing anything.
453 */
454 inode->i_state |= I_DIRTY_PAGES;
455 if (wbc->nr_to_write <= 0) {
456 /*
457 * slice used up: queue for next turn
458 */
431 requeue_io(inode);
459 requeue_io(inode, wb);
432 } else {
433 /*
434 * Writeback blocked by something other than
435 * congestion. Delay the inode for some time to
436 * avoid spinning on the CPU (100% iowait)
437 * retrying writeback of the dirty page/inode
438 * that cannot be performed immediately.
439 */
460 } else {
461 /*
462 * Writeback blocked by something other than
463 * congestion. Delay the inode for some time to
464 * avoid spinning on the CPU (100% iowait)
465 * retrying writeback of the dirty page/inode
466 * that cannot be performed immediately.
467 */
440 redirty_tail(inode);
468 redirty_tail(inode, wb);
441 }
442 } else if (inode->i_state & I_DIRTY) {
443 /*
444 * Filesystems can dirty the inode during writeback
445 * operations, such as delayed allocation during
446 * submission or metadata updates after data IO
447 * completion.
448 */
469 }
470 } else if (inode->i_state & I_DIRTY) {
471 /*
472 * Filesystems can dirty the inode during writeback
473 * operations, such as delayed allocation during
474 * submission or metadata updates after data IO
475 * completion.
476 */
449 redirty_tail(inode);
477 redirty_tail(inode, wb);
450 } else {
451 /*
452 * The inode is clean. At this point we either have
453 * a reference to the inode or it's on it's way out.
454 * No need to add it back to the LRU.
455 */
456 list_del_init(&inode->i_wb_list);
457 }
458 }
459 inode_sync_complete(inode);
478 } else {
479 /*
480 * The inode is clean. At this point we either have
481 * a reference to the inode or it's on it's way out.
482 * No need to add it back to the LRU.
483 */
484 list_del_init(&inode->i_wb_list);
485 }
486 }
487 inode_sync_complete(inode);
488 trace_writeback_single_inode(inode, wbc, nr_to_write);
460 return ret;
461}
462
463/*
464 * For background writeback the caller does not have the sb pinned
465 * before calling writeback. So make sure that we do pin it, so it doesn't
466 * go away while we are writing inodes from it.
467 */

--- 13 unchanged lines hidden (view full) ---

481 return true;
482 up_read(&sb->s_umount);
483 }
484
485 put_super(sb);
486 return false;
487}
488
489 return ret;
490}
491
492/*
493 * For background writeback the caller does not have the sb pinned
494 * before calling writeback. So make sure that we do pin it, so it doesn't
495 * go away while we are writing inodes from it.
496 */

--- 13 unchanged lines hidden (view full) ---

510 return true;
511 up_read(&sb->s_umount);
512 }
513
514 put_super(sb);
515 return false;
516}
517
518static long writeback_chunk_size(struct wb_writeback_work *work)
519{
520 long pages;
521
522 /*
523 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
524 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
525 * here avoids calling into writeback_inodes_wb() more than once.
526 *
527 * The intended call sequence for WB_SYNC_ALL writeback is:
528 *
529 * wb_writeback()
530 * writeback_sb_inodes() <== called only once
531 * write_cache_pages() <== called once for each inode
532 * (quickly) tag currently dirty pages
533 * (maybe slowly) sync all tagged pages
534 */
535 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
536 pages = LONG_MAX;
537 else
538 pages = min(MAX_WRITEBACK_PAGES, work->nr_pages);
539
540 return pages;
541}
542
489/*
490 * Write a portion of b_io inodes which belong to @sb.
491 *
492 * If @only_this_sb is true, then find and write all such
493 * inodes. Otherwise write only ones which go sequentially
494 * in reverse order.
495 *
543/*
544 * Write a portion of b_io inodes which belong to @sb.
545 *
546 * If @only_this_sb is true, then find and write all such
547 * inodes. Otherwise write only ones which go sequentially
548 * in reverse order.
549 *
496 * Return 1, if the caller writeback routine should be
497 * interrupted. Otherwise return 0.
550 * Return the number of pages and/or inodes written.
498 */
551 */
499static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
500 struct writeback_control *wbc, bool only_this_sb)
552static long writeback_sb_inodes(struct super_block *sb,
553 struct bdi_writeback *wb,
554 struct wb_writeback_work *work)
501{
555{
556 struct writeback_control wbc = {
557 .sync_mode = work->sync_mode,
558 .tagged_writepages = work->tagged_writepages,
559 .for_kupdate = work->for_kupdate,
560 .for_background = work->for_background,
561 .range_cyclic = work->range_cyclic,
562 .range_start = 0,
563 .range_end = LLONG_MAX,
564 };
565 unsigned long start_time = jiffies;
566 long write_chunk;
567 long wrote = 0; /* count both pages and inodes */
568
502 while (!list_empty(&wb->b_io)) {
569 while (!list_empty(&wb->b_io)) {
503 long pages_skipped;
504 struct inode *inode = wb_inode(wb->b_io.prev);
505
506 if (inode->i_sb != sb) {
570 struct inode *inode = wb_inode(wb->b_io.prev);
571
572 if (inode->i_sb != sb) {
507 if (only_this_sb) {
573 if (work->sb) {
508 /*
509 * We only want to write back data for this
510 * superblock, move all inodes not belonging
511 * to it back onto the dirty list.
512 */
574 /*
575 * We only want to write back data for this
576 * superblock, move all inodes not belonging
577 * to it back onto the dirty list.
578 */
513 redirty_tail(inode);
579 redirty_tail(inode, wb);
514 continue;
515 }
516
517 /*
518 * The inode belongs to a different superblock.
519 * Bounce back to the caller to unpin this and
520 * pin the next superblock.
521 */
580 continue;
581 }
582
583 /*
584 * The inode belongs to a different superblock.
585 * Bounce back to the caller to unpin this and
586 * pin the next superblock.
587 */
522 return 0;
588 break;
523 }
524
525 /*
526 * Don't bother with new inodes or inodes beeing freed, first
527 * kind does not need peridic writeout yet, and for the latter
528 * kind writeout is handled by the freer.
529 */
530 spin_lock(&inode->i_lock);
531 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
532 spin_unlock(&inode->i_lock);
589 }
590
591 /*
592 * Don't bother with new inodes or inodes beeing freed, first
593 * kind does not need peridic writeout yet, and for the latter
594 * kind writeout is handled by the freer.
595 */
596 spin_lock(&inode->i_lock);
597 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
598 spin_unlock(&inode->i_lock);
533 requeue_io(inode);
599 requeue_io(inode, wb);
534 continue;
535 }
600 continue;
601 }
536
537 /*
538 * Was this inode dirtied after sync_sb_inodes was called?
539 * This keeps sync from extra jobs and livelock.
540 */
541 if (inode_dirtied_after(inode, wbc->wb_start)) {
542 spin_unlock(&inode->i_lock);
543 return 1;
544 }
545
546 __iget(inode);
602 __iget(inode);
603 write_chunk = writeback_chunk_size(work);
604 wbc.nr_to_write = write_chunk;
605 wbc.pages_skipped = 0;
547
606
548 pages_skipped = wbc->pages_skipped;
549 writeback_single_inode(inode, wbc);
550 if (wbc->pages_skipped != pages_skipped) {
607 writeback_single_inode(inode, wb, &wbc);
608
609 work->nr_pages -= write_chunk - wbc.nr_to_write;
610 wrote += write_chunk - wbc.nr_to_write;
611 if (!(inode->i_state & I_DIRTY))
612 wrote++;
613 if (wbc.pages_skipped) {
551 /*
552 * writeback is not making progress due to locked
553 * buffers. Skip this inode for now.
554 */
614 /*
615 * writeback is not making progress due to locked
616 * buffers. Skip this inode for now.
617 */
555 redirty_tail(inode);
618 redirty_tail(inode, wb);
556 }
557 spin_unlock(&inode->i_lock);
619 }
620 spin_unlock(&inode->i_lock);
558 spin_unlock(&inode_wb_list_lock);
621 spin_unlock(&wb->list_lock);
559 iput(inode);
560 cond_resched();
622 iput(inode);
623 cond_resched();
561 spin_lock(&inode_wb_list_lock);
562 if (wbc->nr_to_write <= 0) {
563 wbc->more_io = 1;
564 return 1;
624 spin_lock(&wb->list_lock);
625 /*
626 * bail out to wb_writeback() often enough to check
627 * background threshold and other termination conditions.
628 */
629 if (wrote) {
630 if (time_is_before_jiffies(start_time + HZ / 10UL))
631 break;
632 if (work->nr_pages <= 0)
633 break;
565 }
634 }
566 if (!list_empty(&wb->b_more_io))
567 wbc->more_io = 1;
568 }
635 }
569 /* b_io is empty */
570 return 1;
636 return wrote;
571}
572
637}
638
573void writeback_inodes_wb(struct bdi_writeback *wb,
574 struct writeback_control *wbc)
639static long __writeback_inodes_wb(struct bdi_writeback *wb,
640 struct wb_writeback_work *work)
575{
641{
576 int ret = 0;
642 unsigned long start_time = jiffies;
643 long wrote = 0;
577
644
578 if (!wbc->wb_start)
579 wbc->wb_start = jiffies; /* livelock avoidance */
580 spin_lock(&inode_wb_list_lock);
581 if (!wbc->for_kupdate || list_empty(&wb->b_io))
582 queue_io(wb, wbc->older_than_this);
583
584 while (!list_empty(&wb->b_io)) {
585 struct inode *inode = wb_inode(wb->b_io.prev);
586 struct super_block *sb = inode->i_sb;
587
588 if (!pin_sb_for_writeback(sb)) {
645 while (!list_empty(&wb->b_io)) {
646 struct inode *inode = wb_inode(wb->b_io.prev);
647 struct super_block *sb = inode->i_sb;
648
649 if (!pin_sb_for_writeback(sb)) {
589 requeue_io(inode);
650 requeue_io(inode, wb);
590 continue;
591 }
651 continue;
652 }
592 ret = writeback_sb_inodes(sb, wb, wbc, false);
653 wrote += writeback_sb_inodes(sb, wb, work);
593 drop_super(sb);
594
654 drop_super(sb);
655
595 if (ret)
596 break;
656 /* refer to the same tests at the end of writeback_sb_inodes */
657 if (wrote) {
658 if (time_is_before_jiffies(start_time + HZ / 10UL))
659 break;
660 if (work->nr_pages <= 0)
661 break;
662 }
597 }
663 }
598 spin_unlock(&inode_wb_list_lock);
599 /* Leave any unwritten inodes on b_io */
664 /* Leave any unwritten inodes on b_io */
665 return wrote;
600}
601
666}
667
602static void __writeback_inodes_sb(struct super_block *sb,
603 struct bdi_writeback *wb, struct writeback_control *wbc)
668long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages)
604{
669{
605 WARN_ON(!rwsem_is_locked(&sb->s_umount));
670 struct wb_writeback_work work = {
671 .nr_pages = nr_pages,
672 .sync_mode = WB_SYNC_NONE,
673 .range_cyclic = 1,
674 };
606
675
607 spin_lock(&inode_wb_list_lock);
608 if (!wbc->for_kupdate || list_empty(&wb->b_io))
609 queue_io(wb, wbc->older_than_this);
610 writeback_sb_inodes(sb, wb, wbc, true);
611 spin_unlock(&inode_wb_list_lock);
676 spin_lock(&wb->list_lock);
677 if (list_empty(&wb->b_io))
678 queue_io(wb, NULL);
679 __writeback_inodes_wb(wb, &work);
680 spin_unlock(&wb->list_lock);
681
682 return nr_pages - work.nr_pages;
612}
613
683}
684
614/*
615 * The maximum number of pages to writeout in a single bdi flush/kupdate
616 * operation. We do this so we don't hold I_SYNC against an inode for
617 * enormous amounts of time, which would block a userspace task which has
618 * been forced to throttle against that inode. Also, the code reevaluates
619 * the dirty each time it has written this many pages.
620 */
621#define MAX_WRITEBACK_PAGES 1024
622
623static inline bool over_bground_thresh(void)
624{
625 unsigned long background_thresh, dirty_thresh;
626
627 global_dirty_limits(&background_thresh, &dirty_thresh);
628
629 return (global_page_state(NR_FILE_DIRTY) +
630 global_page_state(NR_UNSTABLE_NFS) > background_thresh);

--- 12 unchanged lines hidden (view full) ---

643 * one-second gap.
644 *
645 * older_than_this takes precedence over nr_to_write. So we'll only write back
646 * all dirty pages if they are all attached to "old" mappings.
647 */
648static long wb_writeback(struct bdi_writeback *wb,
649 struct wb_writeback_work *work)
650{
685static inline bool over_bground_thresh(void)
686{
687 unsigned long background_thresh, dirty_thresh;
688
689 global_dirty_limits(&background_thresh, &dirty_thresh);
690
691 return (global_page_state(NR_FILE_DIRTY) +
692 global_page_state(NR_UNSTABLE_NFS) > background_thresh);

--- 12 unchanged lines hidden (view full) ---

705 * one-second gap.
706 *
707 * older_than_this takes precedence over nr_to_write. So we'll only write back
708 * all dirty pages if they are all attached to "old" mappings.
709 */
710static long wb_writeback(struct bdi_writeback *wb,
711 struct wb_writeback_work *work)
712{
651 struct writeback_control wbc = {
652 .sync_mode = work->sync_mode,
653 .older_than_this = NULL,
654 .for_kupdate = work->for_kupdate,
655 .for_background = work->for_background,
656 .range_cyclic = work->range_cyclic,
657 };
713 long nr_pages = work->nr_pages;
658 unsigned long oldest_jif;
714 unsigned long oldest_jif;
659 long wrote = 0;
660 long write_chunk;
661 struct inode *inode;
715 struct inode *inode;
716 long progress;
662
717
663 if (wbc.for_kupdate) {
664 wbc.older_than_this = &oldest_jif;
665 oldest_jif = jiffies -
666 msecs_to_jiffies(dirty_expire_interval * 10);
667 }
668 if (!wbc.range_cyclic) {
669 wbc.range_start = 0;
670 wbc.range_end = LLONG_MAX;
671 }
718 oldest_jif = jiffies;
719 work->older_than_this = &oldest_jif;
672
720
673 /*
674 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
675 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
676 * here avoids calling into writeback_inodes_wb() more than once.
677 *
678 * The intended call sequence for WB_SYNC_ALL writeback is:
679 *
680 * wb_writeback()
681 * __writeback_inodes_sb() <== called only once
682 * write_cache_pages() <== called once for each inode
683 * (quickly) tag currently dirty pages
684 * (maybe slowly) sync all tagged pages
685 */
686 if (wbc.sync_mode == WB_SYNC_NONE)
687 write_chunk = MAX_WRITEBACK_PAGES;
688 else
689 write_chunk = LONG_MAX;
690
691 wbc.wb_start = jiffies; /* livelock avoidance */
721 spin_lock(&wb->list_lock);
692 for (;;) {
693 /*
694 * Stop writeback when nr_pages has been consumed
695 */
696 if (work->nr_pages <= 0)
697 break;
698
699 /*

--- 8 unchanged lines hidden (view full) ---

708
709 /*
710 * For background writeout, stop when we are below the
711 * background dirty threshold
712 */
713 if (work->for_background && !over_bground_thresh())
714 break;
715
722 for (;;) {
723 /*
724 * Stop writeback when nr_pages has been consumed
725 */
726 if (work->nr_pages <= 0)
727 break;
728
729 /*

--- 8 unchanged lines hidden (view full) ---

738
739 /*
740 * For background writeout, stop when we are below the
741 * background dirty threshold
742 */
743 if (work->for_background && !over_bground_thresh())
744 break;
745
716 wbc.more_io = 0;
717 wbc.nr_to_write = write_chunk;
718 wbc.pages_skipped = 0;
746 if (work->for_kupdate) {
747 oldest_jif = jiffies -
748 msecs_to_jiffies(dirty_expire_interval * 10);
749 work->older_than_this = &oldest_jif;
750 }
719
751
720 trace_wbc_writeback_start(&wbc, wb->bdi);
752 trace_writeback_start(wb->bdi, work);
753 if (list_empty(&wb->b_io))
754 queue_io(wb, work->older_than_this);
721 if (work->sb)
755 if (work->sb)
722 __writeback_inodes_sb(work->sb, wb, &wbc);
756 progress = writeback_sb_inodes(work->sb, wb, work);
723 else
757 else
724 writeback_inodes_wb(wb, &wbc);
725 trace_wbc_writeback_written(&wbc, wb->bdi);
758 progress = __writeback_inodes_wb(wb, work);
759 trace_writeback_written(wb->bdi, work);
726
760
727 work->nr_pages -= write_chunk - wbc.nr_to_write;
728 wrote += write_chunk - wbc.nr_to_write;
729
730 /*
761 /*
731 * If we consumed everything, see if we have more
762 * Did we write something? Try for more
763 *
764 * Dirty inodes are moved to b_io for writeback in batches.
765 * The completion of the current batch does not necessarily
766 * mean the overall work is done. So we keep looping as long
767 * as made some progress on cleaning pages or inodes.
732 */
768 */
733 if (wbc.nr_to_write <= 0)
769 if (progress)
734 continue;
735 /*
770 continue;
771 /*
736 * Didn't write everything and we don't have more IO, bail
772 * No more inodes for IO, bail
737 */
773 */
738 if (!wbc.more_io)
774 if (list_empty(&wb->b_more_io))
739 break;
740 /*
775 break;
776 /*
741 * Did we write something? Try for more
742 */
743 if (wbc.nr_to_write < write_chunk)
744 continue;
745 /*
746 * Nothing written. Wait for some inode to
747 * become available for writeback. Otherwise
748 * we'll just busyloop.
749 */
777 * Nothing written. Wait for some inode to
778 * become available for writeback. Otherwise
779 * we'll just busyloop.
780 */
750 spin_lock(&inode_wb_list_lock);
751 if (!list_empty(&wb->b_more_io)) {
781 if (!list_empty(&wb->b_more_io)) {
782 trace_writeback_wait(wb->bdi, work);
752 inode = wb_inode(wb->b_more_io.prev);
783 inode = wb_inode(wb->b_more_io.prev);
753 trace_wbc_writeback_wait(&wbc, wb->bdi);
754 spin_lock(&inode->i_lock);
784 spin_lock(&inode->i_lock);
755 inode_wait_for_writeback(inode);
785 inode_wait_for_writeback(inode, wb);
756 spin_unlock(&inode->i_lock);
757 }
786 spin_unlock(&inode->i_lock);
787 }
758 spin_unlock(&inode_wb_list_lock);
759 }
788 }
789 spin_unlock(&wb->list_lock);
760
790
761 return wrote;
791 return nr_pages - work->nr_pages;
762}
763
764/*
765 * Return the next wb_writeback_work struct that hasn't been processed yet.
766 */
767static struct wb_writeback_work *
768get_next_work_item(struct backing_dev_info *bdi)
769{

--- 314 unchanged lines hidden (view full) ---

1084 * bdi thread to make sure background
1085 * write-back happens later.
1086 */
1087 if (!wb_has_dirty_io(&bdi->wb))
1088 wakeup_bdi = true;
1089 }
1090
1091 spin_unlock(&inode->i_lock);
792}
793
794/*
795 * Return the next wb_writeback_work struct that hasn't been processed yet.
796 */
797static struct wb_writeback_work *
798get_next_work_item(struct backing_dev_info *bdi)
799{

--- 314 unchanged lines hidden (view full) ---

1114 * bdi thread to make sure background
1115 * write-back happens later.
1116 */
1117 if (!wb_has_dirty_io(&bdi->wb))
1118 wakeup_bdi = true;
1119 }
1120
1121 spin_unlock(&inode->i_lock);
1092 spin_lock(&inode_wb_list_lock);
1122 spin_lock(&bdi->wb.list_lock);
1093 inode->dirtied_when = jiffies;
1094 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1123 inode->dirtied_when = jiffies;
1124 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1095 spin_unlock(&inode_wb_list_lock);
1125 spin_unlock(&bdi->wb.list_lock);
1096
1097 if (wakeup_bdi)
1098 bdi_wakeup_thread_delayed(bdi);
1099 return;
1100 }
1101 }
1102out_unlock_inode:
1103 spin_unlock(&inode->i_lock);

--- 79 unchanged lines hidden (view full) ---

1183 * Start writeback on some inodes on this super_block. No guarantees are made
1184 * on how many (if any) will be written, and this function does not wait
1185 * for IO completion of submitted IO.
1186 */
1187void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1188{
1189 DECLARE_COMPLETION_ONSTACK(done);
1190 struct wb_writeback_work work = {
1126
1127 if (wakeup_bdi)
1128 bdi_wakeup_thread_delayed(bdi);
1129 return;
1130 }
1131 }
1132out_unlock_inode:
1133 spin_unlock(&inode->i_lock);

--- 79 unchanged lines hidden (view full) ---

1213 * Start writeback on some inodes on this super_block. No guarantees are made
1214 * on how many (if any) will be written, and this function does not wait
1215 * for IO completion of submitted IO.
1216 */
1217void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1218{
1219 DECLARE_COMPLETION_ONSTACK(done);
1220 struct wb_writeback_work work = {
1191 .sb = sb,
1192 .sync_mode = WB_SYNC_NONE,
1193 .done = &done,
1194 .nr_pages = nr,
1221 .sb = sb,
1222 .sync_mode = WB_SYNC_NONE,
1223 .tagged_writepages = 1,
1224 .done = &done,
1225 .nr_pages = nr,
1195 };
1196
1197 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1198 bdi_queue_work(sb->s_bdi, &work);
1199 wait_for_completion(&done);
1200}
1201EXPORT_SYMBOL(writeback_inodes_sb_nr);
1202

--- 85 unchanged lines hidden (view full) ---

1288 *
1289 * This function commits an inode to disk immediately if it is dirty. This is
1290 * primarily needed by knfsd.
1291 *
1292 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1293 */
1294int write_inode_now(struct inode *inode, int sync)
1295{
1226 };
1227
1228 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1229 bdi_queue_work(sb->s_bdi, &work);
1230 wait_for_completion(&done);
1231}
1232EXPORT_SYMBOL(writeback_inodes_sb_nr);
1233

--- 85 unchanged lines hidden (view full) ---

1319 *
1320 * This function commits an inode to disk immediately if it is dirty. This is
1321 * primarily needed by knfsd.
1322 *
1323 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1324 */
1325int write_inode_now(struct inode *inode, int sync)
1326{
1327 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1296 int ret;
1297 struct writeback_control wbc = {
1298 .nr_to_write = LONG_MAX,
1299 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1300 .range_start = 0,
1301 .range_end = LLONG_MAX,
1302 };
1303
1304 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1305 wbc.nr_to_write = 0;
1306
1307 might_sleep();
1328 int ret;
1329 struct writeback_control wbc = {
1330 .nr_to_write = LONG_MAX,
1331 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1332 .range_start = 0,
1333 .range_end = LLONG_MAX,
1334 };
1335
1336 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1337 wbc.nr_to_write = 0;
1338
1339 might_sleep();
1308 spin_lock(&inode_wb_list_lock);
1340 spin_lock(&wb->list_lock);
1309 spin_lock(&inode->i_lock);
1341 spin_lock(&inode->i_lock);
1310 ret = writeback_single_inode(inode, &wbc);
1342 ret = writeback_single_inode(inode, wb, &wbc);
1311 spin_unlock(&inode->i_lock);
1343 spin_unlock(&inode->i_lock);
1312 spin_unlock(&inode_wb_list_lock);
1344 spin_unlock(&wb->list_lock);
1313 if (sync)
1314 inode_sync_wait(inode);
1315 return ret;
1316}
1317EXPORT_SYMBOL(write_inode_now);
1318
1319/**
1320 * sync_inode - write an inode and its pages to disk.
1321 * @inode: the inode to sync
1322 * @wbc: controls the writeback mode
1323 *
1324 * sync_inode() will write an inode and its pages to disk. It will also
1325 * correctly update the inode on its superblock's dirty inode lists and will
1326 * update inode->i_state.
1327 *
1328 * The caller must have a ref on the inode.
1329 */
1330int sync_inode(struct inode *inode, struct writeback_control *wbc)
1331{
1345 if (sync)
1346 inode_sync_wait(inode);
1347 return ret;
1348}
1349EXPORT_SYMBOL(write_inode_now);
1350
1351/**
1352 * sync_inode - write an inode and its pages to disk.
1353 * @inode: the inode to sync
1354 * @wbc: controls the writeback mode
1355 *
1356 * sync_inode() will write an inode and its pages to disk. It will also
1357 * correctly update the inode on its superblock's dirty inode lists and will
1358 * update inode->i_state.
1359 *
1360 * The caller must have a ref on the inode.
1361 */
1362int sync_inode(struct inode *inode, struct writeback_control *wbc)
1363{
1364 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1332 int ret;
1333
1365 int ret;
1366
1334 spin_lock(&inode_wb_list_lock);
1367 spin_lock(&wb->list_lock);
1335 spin_lock(&inode->i_lock);
1368 spin_lock(&inode->i_lock);
1336 ret = writeback_single_inode(inode, wbc);
1369 ret = writeback_single_inode(inode, wb, wbc);
1337 spin_unlock(&inode->i_lock);
1370 spin_unlock(&inode->i_lock);
1338 spin_unlock(&inode_wb_list_lock);
1371 spin_unlock(&wb->list_lock);
1339 return ret;
1340}
1341EXPORT_SYMBOL(sync_inode);
1342
1343/**
1344 * sync_inode_metadata - write an inode to disk
1345 * @inode: the inode to sync
1346 * @wait: wait for I/O to complete.

--- 15 unchanged lines hidden --- 		1372 return ret;
1373}
1374EXPORT_SYMBOL(sync_inode);
1375
1376/**
1377 * sync_inode_metadata - write an inode to disk
1378 * @inode: the inode to sync
1379 * @wait: wait for I/O to complete.

--- 15 unchanged lines hidden ---