]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/fs-writeback.c
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>
28 * __mark_inode_dirty - internal function
29 * @inode: inode to mark
30 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
31 * Mark an inode as dirty. Callers should use mark_inode_dirty or
32 * mark_inode_dirty_sync.
34 * Put the inode on the super block's dirty list.
36 * CAREFUL! We mark it dirty unconditionally, but move it onto the
37 * dirty list only if it is hashed or if it refers to a blockdev.
38 * If it was not hashed, it will never be added to the dirty list
39 * even if it is later hashed, as it will have been marked dirty already.
41 * In short, make sure you hash any inodes _before_ you start marking
44 * This function *must* be atomic for the I_DIRTY_PAGES case -
45 * set_page_dirty() is called under spinlock in several places.
47 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
48 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
49 * the kernel-internal blockdev inode represents the dirtying time of the
50 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
51 * page->mapping->host, so the page-dirtying time is recorded in the internal
54 void __mark_inode_dirty(struct inode
*inode
, int flags
)
56 struct super_block
*sb
= inode
->i_sb
;
59 * Don't do this for I_DIRTY_PAGES - that doesn't actually
60 * dirty the inode itself
62 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
63 if (sb
->s_op
->dirty_inode
)
64 sb
->s_op
->dirty_inode(inode
);
68 * make sure that changes are seen by all cpus before we test i_state
73 /* avoid the locking if we can */
74 if ((inode
->i_state
& flags
) == flags
)
77 if (unlikely(block_dump
)) {
78 struct dentry
*dentry
= NULL
;
79 const char *name
= "?";
81 if (!list_empty(&inode
->i_dentry
)) {
82 dentry
= list_entry(inode
->i_dentry
.next
,
83 struct dentry
, d_alias
);
84 if (dentry
&& dentry
->d_name
.name
)
85 name
= (const char *) dentry
->d_name
.name
;
88 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev"))
90 "%s(%d): dirtied inode %lu (%s) on %s\n",
91 current
->comm
, current
->pid
, inode
->i_ino
,
92 name
, inode
->i_sb
->s_id
);
95 spin_lock(&inode_lock
);
96 if ((inode
->i_state
& flags
) != flags
) {
97 const int was_dirty
= inode
->i_state
& I_DIRTY
;
99 inode
->i_state
|= flags
;
102 * If the inode is locked, just update its dirty state.
103 * The unlocker will place the inode on the appropriate
104 * superblock list, based upon its state.
106 if (inode
->i_state
& I_LOCK
)
110 * Only add valid (hashed) inodes to the superblock's
111 * dirty list. Add blockdev inodes as well.
113 if (!S_ISBLK(inode
->i_mode
)) {
114 if (hlist_unhashed(&inode
->i_hash
))
117 if (inode
->i_state
& (I_FREEING
|I_CLEAR
))
121 * If the inode was already on s_dirty or s_io, don't
122 * reposition it (that would break s_dirty time-ordering).
125 inode
->dirtied_when
= jiffies
;
126 list_move(&inode
->i_list
, &sb
->s_dirty
);
130 spin_unlock(&inode_lock
);
133 EXPORT_SYMBOL(__mark_inode_dirty
);
135 static int write_inode(struct inode
*inode
, int sync
)
137 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
138 return inode
->i_sb
->s_op
->write_inode(inode
, sync
);
143 * Write a single inode's dirty pages and inode data out to disk.
144 * If `wait' is set, wait on the writeout.
146 * The whole writeout design is quite complex and fragile. We want to avoid
147 * starvation of particular inodes when others are being redirtied, prevent
150 * Called under inode_lock.
153 __sync_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
156 struct address_space
*mapping
= inode
->i_mapping
;
157 struct super_block
*sb
= inode
->i_sb
;
158 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
161 BUG_ON(inode
->i_state
& I_LOCK
);
163 /* Set I_LOCK, reset I_DIRTY */
164 dirty
= inode
->i_state
& I_DIRTY
;
165 inode
->i_state
|= I_LOCK
;
166 inode
->i_state
&= ~I_DIRTY
;
168 spin_unlock(&inode_lock
);
170 ret
= do_writepages(mapping
, wbc
);
172 /* Don't write the inode if only I_DIRTY_PAGES was set */
173 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
174 int err
= write_inode(inode
, wait
);
180 int err
= filemap_fdatawait(mapping
);
185 spin_lock(&inode_lock
);
186 inode
->i_state
&= ~I_LOCK
;
187 if (!(inode
->i_state
& I_FREEING
)) {
188 if (!(inode
->i_state
& I_DIRTY
) &&
189 mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
191 * We didn't write back all the pages. nfs_writepages()
192 * sometimes bales out without doing anything. Redirty
193 * the inode. It is still on sb->s_io.
195 if (wbc
->for_kupdate
) {
197 * For the kupdate function we leave the inode
198 * at the head of sb_dirty so it will get more
199 * writeout as soon as the queue becomes
202 inode
->i_state
|= I_DIRTY_PAGES
;
203 list_move_tail(&inode
->i_list
, &sb
->s_dirty
);
206 * Otherwise fully redirty the inode so that
207 * other inodes on this superblock will get some
208 * writeout. Otherwise heavy writing to one
209 * file would indefinitely suspend writeout of
210 * all the other files.
212 inode
->i_state
|= I_DIRTY_PAGES
;
213 inode
->dirtied_when
= jiffies
;
214 list_move(&inode
->i_list
, &sb
->s_dirty
);
216 } else if (inode
->i_state
& I_DIRTY
) {
218 * Someone redirtied the inode while were writing back
221 list_move(&inode
->i_list
, &sb
->s_dirty
);
222 } else if (atomic_read(&inode
->i_count
)) {
224 * The inode is clean, inuse
226 list_move(&inode
->i_list
, &inode_in_use
);
229 * The inode is clean, unused
231 list_move(&inode
->i_list
, &inode_unused
);
234 wake_up_inode(inode
);
239 * Write out an inode's dirty pages. Called under inode_lock. Either the
240 * caller has ref on the inode (either via __iget or via syscall against an fd)
241 * or the inode has I_WILL_FREE set (via generic_forget_inode)
244 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
246 wait_queue_head_t
*wqh
;
248 if (!atomic_read(&inode
->i_count
))
249 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
251 WARN_ON(inode
->i_state
& I_WILL_FREE
);
253 if ((wbc
->sync_mode
!= WB_SYNC_ALL
) && (inode
->i_state
& I_LOCK
)) {
254 struct address_space
*mapping
= inode
->i_mapping
;
257 list_move(&inode
->i_list
, &inode
->i_sb
->s_dirty
);
260 * Even if we don't actually write the inode itself here,
261 * we can at least start some of the data writeout..
263 spin_unlock(&inode_lock
);
264 ret
= do_writepages(mapping
, wbc
);
265 spin_lock(&inode_lock
);
270 * It's a data-integrity sync. We must wait.
272 if (inode
->i_state
& I_LOCK
) {
273 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_LOCK
);
275 wqh
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
277 spin_unlock(&inode_lock
);
278 __wait_on_bit(wqh
, &wq
, inode_wait
,
279 TASK_UNINTERRUPTIBLE
);
280 spin_lock(&inode_lock
);
281 } while (inode
->i_state
& I_LOCK
);
283 return __sync_single_inode(inode
, wbc
);
287 * Write out a superblock's list of dirty inodes. A wait will be performed
288 * upon no inodes, all inodes or the final one, depending upon sync_mode.
290 * If older_than_this is non-NULL, then only write out inodes which
291 * had their first dirtying at a time earlier than *older_than_this.
293 * If we're a pdlfush thread, then implement pdflush collision avoidance
294 * against the entire list.
296 * WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so
297 * that it can be located for waiting on in __writeback_single_inode().
299 * Called under inode_lock.
301 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
302 * This function assumes that the blockdev superblock's inodes are backed by
303 * a variety of queues, so all inodes are searched. For other superblocks,
304 * assume that all inodes are backed by the same queue.
306 * FIXME: this linear search could get expensive with many fileystems. But
307 * how to fix? We need to go from an address_space to all inodes which share
308 * a queue with that address_space. (Easy: have a global "dirty superblocks"
311 * The inodes to be written are parked on sb->s_io. They are moved back onto
312 * sb->s_dirty as they are selected for writing. This way, none can be missed
313 * on the writer throttling path, and we get decent balancing between many
314 * throttled threads: we don't want them all piling up on __wait_on_inode.
317 sync_sb_inodes(struct super_block
*sb
, struct writeback_control
*wbc
)
319 const unsigned long start
= jiffies
; /* livelock avoidance */
321 if (!wbc
->for_kupdate
|| list_empty(&sb
->s_io
))
322 list_splice_init(&sb
->s_dirty
, &sb
->s_io
);
324 while (!list_empty(&sb
->s_io
)) {
325 struct inode
*inode
= list_entry(sb
->s_io
.prev
,
326 struct inode
, i_list
);
327 struct address_space
*mapping
= inode
->i_mapping
;
328 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
331 if (!bdi_cap_writeback_dirty(bdi
)) {
332 list_move(&inode
->i_list
, &sb
->s_dirty
);
333 if (sb_is_blkdev_sb(sb
)) {
335 * Dirty memory-backed blockdev: the ramdisk
336 * driver does this. Skip just this inode
341 * Dirty memory-backed inode against a filesystem other
342 * than the kernel-internal bdev filesystem. Skip the
348 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
349 wbc
->encountered_congestion
= 1;
350 if (!sb_is_blkdev_sb(sb
))
351 break; /* Skip a congested fs */
352 list_move(&inode
->i_list
, &sb
->s_dirty
);
353 continue; /* Skip a congested blockdev */
356 if (wbc
->bdi
&& bdi
!= wbc
->bdi
) {
357 if (!sb_is_blkdev_sb(sb
))
358 break; /* fs has the wrong queue */
359 list_move(&inode
->i_list
, &sb
->s_dirty
);
360 continue; /* blockdev has wrong queue */
363 /* Was this inode dirtied after sync_sb_inodes was called? */
364 if (time_after(inode
->dirtied_when
, start
))
367 /* Was this inode dirtied too recently? */
368 if (wbc
->older_than_this
&& time_after(inode
->dirtied_when
,
369 *wbc
->older_than_this
))
372 /* Is another pdflush already flushing this queue? */
373 if (current_is_pdflush() && !writeback_acquire(bdi
))
376 BUG_ON(inode
->i_state
& I_FREEING
);
378 pages_skipped
= wbc
->pages_skipped
;
379 __writeback_single_inode(inode
, wbc
);
380 if (wbc
->sync_mode
== WB_SYNC_HOLD
) {
381 inode
->dirtied_when
= jiffies
;
382 list_move(&inode
->i_list
, &sb
->s_dirty
);
384 if (current_is_pdflush())
385 writeback_release(bdi
);
386 if (wbc
->pages_skipped
!= pages_skipped
) {
388 * writeback is not making progress due to locked
389 * buffers. Skip this inode for now.
391 list_move(&inode
->i_list
, &sb
->s_dirty
);
393 spin_unlock(&inode_lock
);
396 spin_lock(&inode_lock
);
397 if (wbc
->nr_to_write
<= 0)
400 return; /* Leave any unwritten inodes on s_io */
404 * Start writeback of dirty pagecache data against all unlocked inodes.
407 * We don't need to grab a reference to superblock here. If it has non-empty
408 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
409 * past sync_inodes_sb() until both the ->s_dirty and ->s_io lists are
410 * empty. Since __sync_single_inode() regains inode_lock before it finally moves
411 * inode from superblock lists we are OK.
413 * If `older_than_this' is non-zero then only flush inodes which have a
414 * flushtime older than *older_than_this.
416 * If `bdi' is non-zero then we will scan the first inode against each
417 * superblock until we find the matching ones. One group will be the dirty
418 * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
419 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
420 * super-efficient but we're about to do a ton of I/O...
423 writeback_inodes(struct writeback_control
*wbc
)
425 struct super_block
*sb
;
430 sb
= sb_entry(super_blocks
.prev
);
431 for (; sb
!= sb_entry(&super_blocks
); sb
= sb_entry(sb
->s_list
.prev
)) {
432 if (!list_empty(&sb
->s_dirty
) || !list_empty(&sb
->s_io
)) {
433 /* we're making our own get_super here */
435 spin_unlock(&sb_lock
);
437 * If we can't get the readlock, there's no sense in
438 * waiting around, most of the time the FS is going to
439 * be unmounted by the time it is released.
441 if (down_read_trylock(&sb
->s_umount
)) {
443 spin_lock(&inode_lock
);
444 sync_sb_inodes(sb
, wbc
);
445 spin_unlock(&inode_lock
);
447 up_read(&sb
->s_umount
);
450 if (__put_super_and_need_restart(sb
))
453 if (wbc
->nr_to_write
<= 0)
456 spin_unlock(&sb_lock
);
460 * writeback and wait upon the filesystem's dirty inodes. The caller will
461 * do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is
462 * used to park the written inodes on sb->s_dirty for the wait pass.
464 * A finite limit is set on the number of pages which will be written.
465 * To prevent infinite livelock of sys_sync().
467 * We add in the number of potentially dirty inodes, because each inode write
468 * can dirty pagecache in the underlying blockdev.
470 void sync_inodes_sb(struct super_block
*sb
, int wait
)
472 struct writeback_control wbc
= {
473 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_HOLD
,
475 .range_end
= LLONG_MAX
,
477 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
478 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
480 wbc
.nr_to_write
= nr_dirty
+ nr_unstable
+
481 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
) +
482 nr_dirty
+ nr_unstable
;
483 wbc
.nr_to_write
+= wbc
.nr_to_write
/ 2; /* Bit more for luck */
484 spin_lock(&inode_lock
);
485 sync_sb_inodes(sb
, &wbc
);
486 spin_unlock(&inode_lock
);
490 * Rather lame livelock avoidance.
492 static void set_sb_syncing(int val
)
494 struct super_block
*sb
;
496 sb
= sb_entry(super_blocks
.prev
);
497 for (; sb
!= sb_entry(&super_blocks
); sb
= sb_entry(sb
->s_list
.prev
)) {
500 spin_unlock(&sb_lock
);
504 * sync_inodes - writes all inodes to disk
505 * @wait: wait for completion
507 * sync_inodes() goes through each super block's dirty inode list, writes the
508 * inodes out, waits on the writeout and puts the inodes back on the normal
511 * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle
512 * part of the sync functions is that the blockdev "superblock" is processed
513 * last. This is because the write_inode() function of a typical fs will
514 * perform no I/O, but will mark buffers in the blockdev mapping as dirty.
515 * What we want to do is to perform all that dirtying first, and then write
516 * back all those inode blocks via the blockdev mapping in one sweep. So the
517 * additional (somewhat redundant) sync_blockdev() calls here are to make
518 * sure that really happens. Because if we call sync_inodes_sb(wait=1) with
519 * outstanding dirty inodes, the writeback goes block-at-a-time within the
520 * filesystem's write_inode(). This is extremely slow.
522 static void __sync_inodes(int wait
)
524 struct super_block
*sb
;
528 list_for_each_entry(sb
, &super_blocks
, s_list
) {
533 spin_unlock(&sb_lock
);
534 down_read(&sb
->s_umount
);
536 sync_inodes_sb(sb
, wait
);
537 sync_blockdev(sb
->s_bdev
);
539 up_read(&sb
->s_umount
);
541 if (__put_super_and_need_restart(sb
))
544 spin_unlock(&sb_lock
);
547 void sync_inodes(int wait
)
559 * write_inode_now - write an inode to disk
560 * @inode: inode to write to disk
561 * @sync: whether the write should be synchronous or not
563 * This function commits an inode to disk immediately if it is dirty. This is
564 * primarily needed by knfsd.
566 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
568 int write_inode_now(struct inode
*inode
, int sync
)
571 struct writeback_control wbc
= {
572 .nr_to_write
= LONG_MAX
,
573 .sync_mode
= WB_SYNC_ALL
,
575 .range_end
= LLONG_MAX
,
578 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
582 spin_lock(&inode_lock
);
583 ret
= __writeback_single_inode(inode
, &wbc
);
584 spin_unlock(&inode_lock
);
586 wait_on_inode(inode
);
589 EXPORT_SYMBOL(write_inode_now
);
592 * sync_inode - write an inode and its pages to disk.
593 * @inode: the inode to sync
594 * @wbc: controls the writeback mode
596 * sync_inode() will write an inode and its pages to disk. It will also
597 * correctly update the inode on its superblock's dirty inode lists and will
598 * update inode->i_state.
600 * The caller must have a ref on the inode.
602 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
606 spin_lock(&inode_lock
);
607 ret
= __writeback_single_inode(inode
, wbc
);
608 spin_unlock(&inode_lock
);
611 EXPORT_SYMBOL(sync_inode
);
614 * generic_osync_inode - flush all dirty data for a given inode to disk
615 * @inode: inode to write
616 * @mapping: the address_space that should be flushed
617 * @what: what to write and wait upon
619 * This can be called by file_write functions for files which have the
620 * O_SYNC flag set, to flush dirty writes to disk.
622 * @what is a bitmask, specifying which part of the inode's data should be
623 * written and waited upon.
625 * OSYNC_DATA: i_mapping's dirty data
626 * OSYNC_METADATA: the buffers at i_mapping->private_list
627 * OSYNC_INODE: the inode itself
630 int generic_osync_inode(struct inode
*inode
, struct address_space
*mapping
, int what
)
633 int need_write_inode_now
= 0;
636 if (what
& OSYNC_DATA
)
637 err
= filemap_fdatawrite(mapping
);
638 if (what
& (OSYNC_METADATA
|OSYNC_DATA
)) {
639 err2
= sync_mapping_buffers(mapping
);
643 if (what
& OSYNC_DATA
) {
644 err2
= filemap_fdatawait(mapping
);
649 spin_lock(&inode_lock
);
650 if ((inode
->i_state
& I_DIRTY
) &&
651 ((what
& OSYNC_INODE
) || (inode
->i_state
& I_DIRTY_DATASYNC
)))
652 need_write_inode_now
= 1;
653 spin_unlock(&inode_lock
);
655 if (need_write_inode_now
) {
656 err2
= write_inode_now(inode
, 1);
661 wait_on_inode(inode
);
666 EXPORT_SYMBOL(generic_osync_inode
);
669 * writeback_acquire: attempt to get exclusive writeback access to a device
670 * @bdi: the device's backing_dev_info structure
672 * It is a waste of resources to have more than one pdflush thread blocked on
673 * a single request queue. Exclusion at the request_queue level is obtained
674 * via a flag in the request_queue's backing_dev_info.state.
676 * Non-request_queue-backed address_spaces will share default_backing_dev_info,
677 * unless they implement their own. Which is somewhat inefficient, as this
678 * may prevent concurrent writeback against multiple devices.
680 int writeback_acquire(struct backing_dev_info
*bdi
)
682 return !test_and_set_bit(BDI_pdflush
, &bdi
->state
);
686 * writeback_in_progress: determine whether there is writeback in progress
687 * @bdi: the device's backing_dev_info structure.
689 * Determine whether there is writeback in progress against a backing device.
691 int writeback_in_progress(struct backing_dev_info
*bdi
)
693 return test_bit(BDI_pdflush
, &bdi
->state
);
697 * writeback_release: relinquish exclusive writeback access against a device.
698 * @bdi: the device's backing_dev_info structure
700 void writeback_release(struct backing_dev_info
*bdi
)
702 BUG_ON(!writeback_in_progress(bdi
));
703 clear_bit(BDI_pdflush
, &bdi
->state
);