4 * Copyright (C) 2002, Linus Torvalds.
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
9 * inode itself is not handled here.
11 * 10Apr2002 akpm@zip.com.au
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/spinlock.h>
18 #include <linux/sched.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/backing-dev.h>
24 #include <linux/buffer_head.h>
26 extern struct super_block
*blockdev_superblock
;
29 * __mark_inode_dirty - internal function
30 * @inode: inode to mark
31 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
32 * Mark an inode as dirty. Callers should use mark_inode_dirty or
33 * mark_inode_dirty_sync.
35 * Put the inode on the super block's dirty list.
37 * CAREFUL! We mark it dirty unconditionally, but move it onto the
38 * dirty list only if it is hashed or if it refers to a blockdev.
39 * If it was not hashed, it will never be added to the dirty list
40 * even if it is later hashed, as it will have been marked dirty already.
42 * In short, make sure you hash any inodes _before_ you start marking
45 * This function *must* be atomic for the I_DIRTY_PAGES case -
46 * set_page_dirty() is called under spinlock in several places.
48 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
49 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
50 * the kernel-internal blockdev inode represents the dirtying time of the
51 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
52 * page->mapping->host, so the page-dirtying time is recorded in the internal
55 void __mark_inode_dirty(struct inode
*inode
, int flags
)
57 struct super_block
*sb
= inode
->i_sb
;
60 * Don't do this for I_DIRTY_PAGES - that doesn't actually
61 * dirty the inode itself
63 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
64 if (sb
->s_op
->dirty_inode
)
65 sb
->s_op
->dirty_inode(inode
);
69 * make sure that changes are seen by all cpus before we test i_state
74 /* avoid the locking if we can */
75 if ((inode
->i_state
& flags
) == flags
)
78 if (unlikely(block_dump
)) {
79 struct dentry
*dentry
= NULL
;
80 const char *name
= "?";
82 if (!list_empty(&inode
->i_dentry
)) {
83 dentry
= list_entry(inode
->i_dentry
.next
,
84 struct dentry
, d_alias
);
85 if (dentry
&& dentry
->d_name
.name
)
86 name
= (const char *) dentry
->d_name
.name
;
89 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev"))
91 "%s(%d): dirtied inode %lu (%s) on %s\n",
92 current
->comm
, current
->pid
, inode
->i_ino
,
93 name
, inode
->i_sb
->s_id
);
96 spin_lock(&inode_lock
);
97 if ((inode
->i_state
& flags
) != flags
) {
98 const int was_dirty
= inode
->i_state
& I_DIRTY
;
100 inode
->i_state
|= flags
;
103 * If the inode is locked, just update its dirty state.
104 * The unlocker will place the inode on the appropriate
105 * superblock list, based upon its state.
107 if (inode
->i_state
& I_LOCK
)
111 * Only add valid (hashed) inodes to the superblock's
112 * dirty list. Add blockdev inodes as well.
114 if (!S_ISBLK(inode
->i_mode
)) {
115 if (hlist_unhashed(&inode
->i_hash
))
118 if (inode
->i_state
& (I_FREEING
|I_CLEAR
))
122 * If the inode was already on s_dirty or s_io, don't
123 * reposition it (that would break s_dirty time-ordering).
126 inode
->dirtied_when
= jiffies
;
127 list_move(&inode
->i_list
, &sb
->s_dirty
);
131 spin_unlock(&inode_lock
);
134 EXPORT_SYMBOL(__mark_inode_dirty
);
136 static int write_inode(struct inode
*inode
, int sync
)
138 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
139 return inode
->i_sb
->s_op
->write_inode(inode
, sync
);
144 * Write a single inode's dirty pages and inode data out to disk.
145 * If `wait' is set, wait on the writeout.
147 * The whole writeout design is quite complex and fragile. We want to avoid
148 * starvation of particular inodes when others are being redirtied, prevent
151 * Called under inode_lock.
154 __sync_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
157 struct address_space
*mapping
= inode
->i_mapping
;
158 struct super_block
*sb
= inode
->i_sb
;
159 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
162 BUG_ON(inode
->i_state
& I_LOCK
);
164 /* Set I_LOCK, reset I_DIRTY */
165 dirty
= inode
->i_state
& I_DIRTY
;
166 inode
->i_state
|= I_LOCK
;
167 inode
->i_state
&= ~I_DIRTY
;
169 spin_unlock(&inode_lock
);
171 ret
= do_writepages(mapping
, wbc
);
173 /* Don't write the inode if only I_DIRTY_PAGES was set */
174 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
175 int err
= write_inode(inode
, wait
);
181 int err
= filemap_fdatawait(mapping
);
186 spin_lock(&inode_lock
);
187 inode
->i_state
&= ~I_LOCK
;
188 if (!(inode
->i_state
& I_FREEING
)) {
189 if (!(inode
->i_state
& I_DIRTY
) &&
190 mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
192 * We didn't write back all the pages. nfs_writepages()
193 * sometimes bales out without doing anything. Redirty
194 * the inode. It is still on sb->s_io.
196 if (wbc
->for_kupdate
) {
198 * For the kupdate function we leave the inode
199 * at the head of sb_dirty so it will get more
200 * writeout as soon as the queue becomes
203 inode
->i_state
|= I_DIRTY_PAGES
;
204 list_move_tail(&inode
->i_list
, &sb
->s_dirty
);
207 * Otherwise fully redirty the inode so that
208 * other inodes on this superblock will get some
209 * writeout. Otherwise heavy writing to one
210 * file would indefinitely suspend writeout of
211 * all the other files.
213 inode
->i_state
|= I_DIRTY_PAGES
;
214 inode
->dirtied_when
= jiffies
;
215 list_move(&inode
->i_list
, &sb
->s_dirty
);
217 } else if (inode
->i_state
& I_DIRTY
) {
219 * Someone redirtied the inode while were writing back
222 list_move(&inode
->i_list
, &sb
->s_dirty
);
223 } else if (atomic_read(&inode
->i_count
)) {
225 * The inode is clean, inuse
227 list_move(&inode
->i_list
, &inode_in_use
);
230 * The inode is clean, unused
232 list_move(&inode
->i_list
, &inode_unused
);
235 wake_up_inode(inode
);
240 * Write out an inode's dirty pages. Called under inode_lock. Either the
241 * caller has ref on the inode (either via __iget or via syscall against an fd)
242 * or the inode has I_WILL_FREE set (via generic_forget_inode)
245 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
247 wait_queue_head_t
*wqh
;
249 if (!atomic_read(&inode
->i_count
))
250 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
252 WARN_ON(inode
->i_state
& I_WILL_FREE
);
254 if ((wbc
->sync_mode
!= WB_SYNC_ALL
) && (inode
->i_state
& I_LOCK
)) {
255 list_move(&inode
->i_list
, &inode
->i_sb
->s_dirty
);
260 * It's a data-integrity sync. We must wait.
262 if (inode
->i_state
& I_LOCK
) {
263 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_LOCK
);
265 wqh
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
267 spin_unlock(&inode_lock
);
268 __wait_on_bit(wqh
, &wq
, inode_wait
,
269 TASK_UNINTERRUPTIBLE
);
270 spin_lock(&inode_lock
);
271 } while (inode
->i_state
& I_LOCK
);
273 return __sync_single_inode(inode
, wbc
);
277 * Write out a superblock's list of dirty inodes. A wait will be performed
278 * upon no inodes, all inodes or the final one, depending upon sync_mode.
280 * If older_than_this is non-NULL, then only write out inodes which
281 * had their first dirtying at a time earlier than *older_than_this.
283 * If we're a pdlfush thread, then implement pdflush collision avoidance
284 * against the entire list.
286 * WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so
287 * that it can be located for waiting on in __writeback_single_inode().
289 * Called under inode_lock.
291 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
292 * This function assumes that the blockdev superblock's inodes are backed by
293 * a variety of queues, so all inodes are searched. For other superblocks,
294 * assume that all inodes are backed by the same queue.
296 * FIXME: this linear search could get expensive with many fileystems. But
297 * how to fix? We need to go from an address_space to all inodes which share
298 * a queue with that address_space. (Easy: have a global "dirty superblocks"
301 * The inodes to be written are parked on sb->s_io. They are moved back onto
302 * sb->s_dirty as they are selected for writing. This way, none can be missed
303 * on the writer throttling path, and we get decent balancing between many
304 * throttled threads: we don't want them all piling up on __wait_on_inode.
307 sync_sb_inodes(struct super_block
*sb
, struct writeback_control
*wbc
)
309 const unsigned long start
= jiffies
; /* livelock avoidance */
311 if (!wbc
->for_kupdate
|| list_empty(&sb
->s_io
))
312 list_splice_init(&sb
->s_dirty
, &sb
->s_io
);
314 while (!list_empty(&sb
->s_io
)) {
315 struct inode
*inode
= list_entry(sb
->s_io
.prev
,
316 struct inode
, i_list
);
317 struct address_space
*mapping
= inode
->i_mapping
;
318 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
321 if (!bdi_cap_writeback_dirty(bdi
)) {
322 list_move(&inode
->i_list
, &sb
->s_dirty
);
323 if (sb
== blockdev_superblock
) {
325 * Dirty memory-backed blockdev: the ramdisk
326 * driver does this. Skip just this inode
331 * Dirty memory-backed inode against a filesystem other
332 * than the kernel-internal bdev filesystem. Skip the
338 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
339 wbc
->encountered_congestion
= 1;
340 if (sb
!= blockdev_superblock
)
341 break; /* Skip a congested fs */
342 list_move(&inode
->i_list
, &sb
->s_dirty
);
343 continue; /* Skip a congested blockdev */
346 if (wbc
->bdi
&& bdi
!= wbc
->bdi
) {
347 if (sb
!= blockdev_superblock
)
348 break; /* fs has the wrong queue */
349 list_move(&inode
->i_list
, &sb
->s_dirty
);
350 continue; /* blockdev has wrong queue */
353 /* Was this inode dirtied after sync_sb_inodes was called? */
354 if (time_after(inode
->dirtied_when
, start
))
357 /* Was this inode dirtied too recently? */
358 if (wbc
->older_than_this
&& time_after(inode
->dirtied_when
,
359 *wbc
->older_than_this
))
362 /* Is another pdflush already flushing this queue? */
363 if (current_is_pdflush() && !writeback_acquire(bdi
))
366 BUG_ON(inode
->i_state
& I_FREEING
);
368 pages_skipped
= wbc
->pages_skipped
;
369 __writeback_single_inode(inode
, wbc
);
370 if (wbc
->sync_mode
== WB_SYNC_HOLD
) {
371 inode
->dirtied_when
= jiffies
;
372 list_move(&inode
->i_list
, &sb
->s_dirty
);
374 if (current_is_pdflush())
375 writeback_release(bdi
);
376 if (wbc
->pages_skipped
!= pages_skipped
) {
378 * writeback is not making progress due to locked
379 * buffers. Skip this inode for now.
381 list_move(&inode
->i_list
, &sb
->s_dirty
);
383 spin_unlock(&inode_lock
);
386 spin_lock(&inode_lock
);
387 if (wbc
->nr_to_write
<= 0)
390 return; /* Leave any unwritten inodes on s_io */
394 * Start writeback of dirty pagecache data against all unlocked inodes.
397 * We don't need to grab a reference to superblock here. If it has non-empty
398 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
399 * past sync_inodes_sb() until both the ->s_dirty and ->s_io lists are
400 * empty. Since __sync_single_inode() regains inode_lock before it finally moves
401 * inode from superblock lists we are OK.
403 * If `older_than_this' is non-zero then only flush inodes which have a
404 * flushtime older than *older_than_this.
406 * If `bdi' is non-zero then we will scan the first inode against each
407 * superblock until we find the matching ones. One group will be the dirty
408 * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
409 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
410 * super-efficient but we're about to do a ton of I/O...
413 writeback_inodes(struct writeback_control
*wbc
)
415 struct super_block
*sb
;
420 sb
= sb_entry(super_blocks
.prev
);
421 for (; sb
!= sb_entry(&super_blocks
); sb
= sb_entry(sb
->s_list
.prev
)) {
422 if (!list_empty(&sb
->s_dirty
) || !list_empty(&sb
->s_io
)) {
423 /* we're making our own get_super here */
425 spin_unlock(&sb_lock
);
427 * If we can't get the readlock, there's no sense in
428 * waiting around, most of the time the FS is going to
429 * be unmounted by the time it is released.
431 if (down_read_trylock(&sb
->s_umount
)) {
433 spin_lock(&inode_lock
);
434 sync_sb_inodes(sb
, wbc
);
435 spin_unlock(&inode_lock
);
437 up_read(&sb
->s_umount
);
440 if (__put_super_and_need_restart(sb
))
443 if (wbc
->nr_to_write
<= 0)
446 spin_unlock(&sb_lock
);
450 * writeback and wait upon the filesystem's dirty inodes. The caller will
451 * do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is
452 * used to park the written inodes on sb->s_dirty for the wait pass.
454 * A finite limit is set on the number of pages which will be written.
455 * To prevent infinite livelock of sys_sync().
457 * We add in the number of potentially dirty inodes, because each inode write
458 * can dirty pagecache in the underlying blockdev.
460 void sync_inodes_sb(struct super_block
*sb
, int wait
)
462 struct writeback_control wbc
= {
463 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_HOLD
,
465 unsigned long nr_dirty
= read_page_state(nr_dirty
);
466 unsigned long nr_unstable
= read_page_state(nr_unstable
);
468 wbc
.nr_to_write
= nr_dirty
+ nr_unstable
+
469 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
) +
470 nr_dirty
+ nr_unstable
;
471 wbc
.nr_to_write
+= wbc
.nr_to_write
/ 2; /* Bit more for luck */
472 spin_lock(&inode_lock
);
473 sync_sb_inodes(sb
, &wbc
);
474 spin_unlock(&inode_lock
);
478 * Rather lame livelock avoidance.
480 static void set_sb_syncing(int val
)
482 struct super_block
*sb
;
484 sb
= sb_entry(super_blocks
.prev
);
485 for (; sb
!= sb_entry(&super_blocks
); sb
= sb_entry(sb
->s_list
.prev
)) {
488 spin_unlock(&sb_lock
);
492 * sync_inodes - writes all inodes to disk
493 * @wait: wait for completion
495 * sync_inodes() goes through each super block's dirty inode list, writes the
496 * inodes out, waits on the writeout and puts the inodes back on the normal
499 * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle
500 * part of the sync functions is that the blockdev "superblock" is processed
501 * last. This is because the write_inode() function of a typical fs will
502 * perform no I/O, but will mark buffers in the blockdev mapping as dirty.
503 * What we want to do is to perform all that dirtying first, and then write
504 * back all those inode blocks via the blockdev mapping in one sweep. So the
505 * additional (somewhat redundant) sync_blockdev() calls here are to make
506 * sure that really happens. Because if we call sync_inodes_sb(wait=1) with
507 * outstanding dirty inodes, the writeback goes block-at-a-time within the
508 * filesystem's write_inode(). This is extremely slow.
510 static void __sync_inodes(int wait
)
512 struct super_block
*sb
;
516 list_for_each_entry(sb
, &super_blocks
, s_list
) {
521 spin_unlock(&sb_lock
);
522 down_read(&sb
->s_umount
);
524 sync_inodes_sb(sb
, wait
);
525 sync_blockdev(sb
->s_bdev
);
527 up_read(&sb
->s_umount
);
529 if (__put_super_and_need_restart(sb
))
532 spin_unlock(&sb_lock
);
535 void sync_inodes(int wait
)
547 * write_inode_now - write an inode to disk
548 * @inode: inode to write to disk
549 * @sync: whether the write should be synchronous or not
551 * This function commits an inode to disk immediately if it is dirty. This is
552 * primarily needed by knfsd.
554 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
556 int write_inode_now(struct inode
*inode
, int sync
)
559 struct writeback_control wbc
= {
560 .nr_to_write
= LONG_MAX
,
561 .sync_mode
= WB_SYNC_ALL
,
564 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
568 spin_lock(&inode_lock
);
569 ret
= __writeback_single_inode(inode
, &wbc
);
570 spin_unlock(&inode_lock
);
572 wait_on_inode(inode
);
575 EXPORT_SYMBOL(write_inode_now
);
578 * sync_inode - write an inode and its pages to disk.
579 * @inode: the inode to sync
580 * @wbc: controls the writeback mode
582 * sync_inode() will write an inode and its pages to disk. It will also
583 * correctly update the inode on its superblock's dirty inode lists and will
584 * update inode->i_state.
586 * The caller must have a ref on the inode.
588 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
592 spin_lock(&inode_lock
);
593 ret
= __writeback_single_inode(inode
, wbc
);
594 spin_unlock(&inode_lock
);
597 EXPORT_SYMBOL(sync_inode
);
600 * generic_osync_inode - flush all dirty data for a given inode to disk
601 * @inode: inode to write
602 * @mapping: the address_space that should be flushed
603 * @what: what to write and wait upon
605 * This can be called by file_write functions for files which have the
606 * O_SYNC flag set, to flush dirty writes to disk.
608 * @what is a bitmask, specifying which part of the inode's data should be
609 * written and waited upon.
611 * OSYNC_DATA: i_mapping's dirty data
612 * OSYNC_METADATA: the buffers at i_mapping->private_list
613 * OSYNC_INODE: the inode itself
616 int generic_osync_inode(struct inode
*inode
, struct address_space
*mapping
, int what
)
619 int need_write_inode_now
= 0;
622 current
->flags
|= PF_SYNCWRITE
;
623 if (what
& OSYNC_DATA
)
624 err
= filemap_fdatawrite(mapping
);
625 if (what
& (OSYNC_METADATA
|OSYNC_DATA
)) {
626 err2
= sync_mapping_buffers(mapping
);
630 if (what
& OSYNC_DATA
) {
631 err2
= filemap_fdatawait(mapping
);
635 current
->flags
&= ~PF_SYNCWRITE
;
637 spin_lock(&inode_lock
);
638 if ((inode
->i_state
& I_DIRTY
) &&
639 ((what
& OSYNC_INODE
) || (inode
->i_state
& I_DIRTY_DATASYNC
)))
640 need_write_inode_now
= 1;
641 spin_unlock(&inode_lock
);
643 if (need_write_inode_now
) {
644 err2
= write_inode_now(inode
, 1);
649 wait_on_inode(inode
);
654 EXPORT_SYMBOL(generic_osync_inode
);
657 * writeback_acquire: attempt to get exclusive writeback access to a device
658 * @bdi: the device's backing_dev_info structure
660 * It is a waste of resources to have more than one pdflush thread blocked on
661 * a single request queue. Exclusion at the request_queue level is obtained
662 * via a flag in the request_queue's backing_dev_info.state.
664 * Non-request_queue-backed address_spaces will share default_backing_dev_info,
665 * unless they implement their own. Which is somewhat inefficient, as this
666 * may prevent concurrent writeback against multiple devices.
668 int writeback_acquire(struct backing_dev_info
*bdi
)
670 return !test_and_set_bit(BDI_pdflush
, &bdi
->state
);
674 * writeback_in_progress: determine whether there is writeback in progress
675 * @bdi: the device's backing_dev_info structure.
677 * Determine whether there is writeback in progress against a backing device.
679 int writeback_in_progress(struct backing_dev_info
*bdi
)
681 return test_bit(BDI_pdflush
, &bdi
->state
);
685 * writeback_release: relinquish exclusive writeback access against a device.
686 * @bdi: the device's backing_dev_info structure
688 void writeback_release(struct backing_dev_info
*bdi
)
690 BUG_ON(!writeback_in_progress(bdi
));
691 clear_bit(BDI_pdflush
, &bdi
->state
);
This page took 0.0560929999999999 seconds and 5 git commands to generate.