2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_dinode.h"
32 #include "xfs_error.h"
33 #include "xfs_filestream.h"
34 #include "xfs_vnodeops.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_quota.h"
37 #include "xfs_trace.h"
38 #include "xfs_fsops.h"
40 #include <linux/kthread.h>
41 #include <linux/freezer.h>
43 struct workqueue_struct
*xfs_syncd_wq
; /* sync workqueue */
46 * The inode lookup is done in batches to keep the amount of lock traffic and
47 * radix tree lookups to a minimum. The batch size is a trade off between
48 * lookup reduction and stack usage. This is in the reclaim path, so we can't
51 #define XFS_LOOKUP_BATCH 32
54 xfs_inode_ag_walk_grab(
57 struct inode
*inode
= VFS_I(ip
);
59 ASSERT(rcu_read_lock_held());
62 * check for stale RCU freed inode
64 * If the inode has been reallocated, it doesn't matter if it's not in
65 * the AG we are walking - we are walking for writeback, so if it
66 * passes all the "valid inode" checks and is dirty, then we'll write
67 * it back anyway. If it has been reallocated and still being
68 * initialised, the XFS_INEW check below will catch it.
70 spin_lock(&ip
->i_flags_lock
);
72 goto out_unlock_noent
;
74 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
75 if (__xfs_iflags_test(ip
, XFS_INEW
| XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
76 goto out_unlock_noent
;
77 spin_unlock(&ip
->i_flags_lock
);
79 /* nothing to sync during shutdown */
80 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
83 /* If we can't grab the inode, it must on it's way to reclaim. */
87 if (is_bad_inode(inode
)) {
96 spin_unlock(&ip
->i_flags_lock
);
102 struct xfs_mount
*mp
,
103 struct xfs_perag
*pag
,
104 int (*execute
)(struct xfs_inode
*ip
,
105 struct xfs_perag
*pag
, int flags
),
108 uint32_t first_index
;
120 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
125 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
126 (void **)batch
, first_index
,
134 * Grab the inodes before we drop the lock. if we found
135 * nothing, nr == 0 and the loop will be skipped.
137 for (i
= 0; i
< nr_found
; i
++) {
138 struct xfs_inode
*ip
= batch
[i
];
140 if (done
|| xfs_inode_ag_walk_grab(ip
))
144 * Update the index for the next lookup. Catch
145 * overflows into the next AG range which can occur if
146 * we have inodes in the last block of the AG and we
147 * are currently pointing to the last inode.
149 * Because we may see inodes that are from the wrong AG
150 * due to RCU freeing and reallocation, only update the
151 * index if it lies in this AG. It was a race that lead
152 * us to see this inode, so another lookup from the
153 * same index will not find it again.
155 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
157 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
158 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
162 /* unlock now we've grabbed the inodes. */
165 for (i
= 0; i
< nr_found
; i
++) {
168 error
= execute(batch
[i
], pag
, flags
);
170 if (error
== EAGAIN
) {
174 if (error
&& last_error
!= EFSCORRUPTED
)
178 /* bail out if the filesystem is corrupted. */
179 if (error
== EFSCORRUPTED
)
184 } while (nr_found
&& !done
);
194 xfs_inode_ag_iterator(
195 struct xfs_mount
*mp
,
196 int (*execute
)(struct xfs_inode
*ip
,
197 struct xfs_perag
*pag
, int flags
),
200 struct xfs_perag
*pag
;
206 while ((pag
= xfs_perag_get(mp
, ag
))) {
207 ag
= pag
->pag_agno
+ 1;
208 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
);
212 if (error
== EFSCORRUPTED
)
216 return XFS_ERROR(last_error
);
221 struct xfs_inode
*ip
,
222 struct xfs_perag
*pag
,
225 struct inode
*inode
= VFS_I(ip
);
226 struct address_space
*mapping
= inode
->i_mapping
;
229 if (!mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
232 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
)) {
233 if (flags
& SYNC_TRYLOCK
)
235 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
238 error
= xfs_flush_pages(ip
, 0, -1, (flags
& SYNC_WAIT
) ?
239 0 : XBF_ASYNC
, FI_NONE
);
240 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
246 struct xfs_inode
*ip
,
247 struct xfs_perag
*pag
,
252 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
253 if (xfs_inode_clean(ip
))
255 if (!xfs_iflock_nowait(ip
)) {
256 if (!(flags
& SYNC_WAIT
))
261 if (xfs_inode_clean(ip
)) {
266 error
= xfs_iflush(ip
, flags
);
269 * We don't want to try again on non-blocking flushes that can't run
270 * again immediately. If an inode really must be written, then that's
271 * what the SYNC_WAIT flag is for.
273 if (error
== EAGAIN
) {
274 ASSERT(!(flags
& SYNC_WAIT
));
279 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
284 * Write out pagecache data for the whole filesystem.
288 struct xfs_mount
*mp
,
293 ASSERT((flags
& ~(SYNC_TRYLOCK
|SYNC_WAIT
)) == 0);
295 error
= xfs_inode_ag_iterator(mp
, xfs_sync_inode_data
, flags
);
297 return XFS_ERROR(error
);
299 xfs_log_force(mp
, (flags
& SYNC_WAIT
) ? XFS_LOG_SYNC
: 0);
304 * Write out inode metadata (attributes) for the whole filesystem.
308 struct xfs_mount
*mp
,
311 ASSERT((flags
& ~SYNC_WAIT
) == 0);
313 return xfs_inode_ag_iterator(mp
, xfs_sync_inode_attr
, flags
);
318 struct xfs_mount
*mp
)
324 * If the buffer is pinned then push on the log so we won't get stuck
325 * waiting in the write for someone, maybe ourselves, to flush the log.
327 * Even though we just pushed the log above, we did not have the
328 * superblock buffer locked at that point so it can become pinned in
329 * between there and here.
331 bp
= xfs_getsb(mp
, 0);
332 if (xfs_buf_ispinned(bp
))
333 xfs_log_force(mp
, 0);
334 error
= xfs_bwrite(bp
);
340 * When remounting a filesystem read-only or freezing the filesystem, we have
341 * two phases to execute. This first phase is syncing the data before we
342 * quiesce the filesystem, and the second is flushing all the inodes out after
343 * we've waited for all the transactions created by the first phase to
344 * complete. The second phase ensures that the inodes are written to their
345 * location on disk rather than just existing in transactions in the log. This
346 * means after a quiesce there is no log replay required to write the inodes to
347 * disk (this is the main difference between a sync and a quiesce).
350 * First stage of freeze - no writers will make progress now we are here,
351 * so we flush delwri and delalloc buffers here, then wait for all I/O to
352 * complete. Data is frozen at that point. Metadata is not frozen,
353 * transactions can still occur here so don't bother flushing the buftarg
354 * because it'll just get dirty again.
358 struct xfs_mount
*mp
)
360 int error
, error2
= 0;
362 /* force out the log */
363 xfs_log_force(mp
, XFS_LOG_SYNC
);
365 /* write superblock and hoover up shutdown errors */
366 error
= xfs_sync_fsdata(mp
);
368 /* make sure all delwri buffers are written out */
369 xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
371 /* mark the log as covered if needed */
372 if (xfs_log_need_covered(mp
))
373 error2
= xfs_fs_log_dummy(mp
);
375 /* flush data-only devices */
376 if (mp
->m_rtdev_targp
)
377 xfs_flush_buftarg(mp
->m_rtdev_targp
, 1);
379 return error
? error
: error2
;
384 struct xfs_mount
*mp
)
386 int count
= 0, pincount
;
388 xfs_reclaim_inodes(mp
, 0);
389 xfs_flush_buftarg(mp
->m_ddev_targp
, 0);
392 * This loop must run at least twice. The first instance of the loop
393 * will flush most meta data but that will generate more meta data
394 * (typically directory updates). Which then must be flushed and
395 * logged before we can write the unmount record. We also so sync
396 * reclaim of inodes to catch any that the above delwri flush skipped.
399 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
400 xfs_sync_attr(mp
, SYNC_WAIT
);
401 pincount
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
410 * Second stage of a quiesce. The data is already synced, now we have to take
411 * care of the metadata. New transactions are already blocked, so we need to
412 * wait for any remaining transactions to drain out before proceeding.
416 struct xfs_mount
*mp
)
420 /* wait for all modifications to complete */
421 while (atomic_read(&mp
->m_active_trans
) > 0)
424 /* flush inodes and push all remaining buffers out to disk */
428 * Just warn here till VFS can correctly support
429 * read-only remount without racing.
431 WARN_ON(atomic_read(&mp
->m_active_trans
) != 0);
433 /* Push the superblock and write an unmount record */
434 error
= xfs_log_sbcount(mp
);
436 xfs_warn(mp
, "xfs_attr_quiesce: failed to log sb changes. "
437 "Frozen image may not be consistent.");
438 xfs_log_unmount_write(mp
);
439 xfs_unmountfs_writesb(mp
);
443 xfs_syncd_queue_sync(
444 struct xfs_mount
*mp
)
446 queue_delayed_work(xfs_syncd_wq
, &mp
->m_sync_work
,
447 msecs_to_jiffies(xfs_syncd_centisecs
* 10));
451 * Every sync period we need to unpin all items, reclaim inodes and sync
452 * disk quotas. We might need to cover the log to indicate that the
453 * filesystem is idle and not frozen.
457 struct work_struct
*work
)
459 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
460 struct xfs_mount
, m_sync_work
);
463 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
464 /* dgc: errors ignored here */
465 if (mp
->m_super
->s_frozen
== SB_UNFROZEN
&&
466 xfs_log_need_covered(mp
))
467 error
= xfs_fs_log_dummy(mp
);
469 xfs_log_force(mp
, 0);
471 /* start pushing all the metadata that is currently dirty */
472 xfs_ail_push_all(mp
->m_ail
);
475 /* queue us up again */
476 xfs_syncd_queue_sync(mp
);
480 * Queue a new inode reclaim pass if there are reclaimable inodes and there
481 * isn't a reclaim pass already in progress. By default it runs every 5s based
482 * on the xfs syncd work default of 30s. Perhaps this should have it's own
483 * tunable, but that can be done if this method proves to be ineffective or too
487 xfs_syncd_queue_reclaim(
488 struct xfs_mount
*mp
)
492 * We can have inodes enter reclaim after we've shut down the syncd
493 * workqueue during unmount, so don't allow reclaim work to be queued
496 if (!(mp
->m_super
->s_flags
& MS_ACTIVE
))
500 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
501 queue_delayed_work(xfs_syncd_wq
, &mp
->m_reclaim_work
,
502 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
508 * This is a fast pass over the inode cache to try to get reclaim moving on as
509 * many inodes as possible in a short period of time. It kicks itself every few
510 * seconds, as well as being kicked by the inode cache shrinker when memory
511 * goes low. It scans as quickly as possible avoiding locked inodes or those
512 * already being flushed, and once done schedules a future pass.
516 struct work_struct
*work
)
518 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
519 struct xfs_mount
, m_reclaim_work
);
521 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
522 xfs_syncd_queue_reclaim(mp
);
526 * Flush delayed allocate data, attempting to free up reserved space
527 * from existing allocations. At this point a new allocation attempt
528 * has failed with ENOSPC and we are in the process of scratching our
529 * heads, looking about for more room.
531 * Queue a new data flush if there isn't one already in progress and
532 * wait for completion of the flush. This means that we only ever have one
533 * inode flush in progress no matter how many ENOSPC events are occurring and
534 * so will prevent the system from bogging down due to every concurrent
535 * ENOSPC event scanning all the active inodes in the system for writeback.
539 struct xfs_inode
*ip
)
541 struct xfs_mount
*mp
= ip
->i_mount
;
543 queue_work(xfs_syncd_wq
, &mp
->m_flush_work
);
544 flush_work_sync(&mp
->m_flush_work
);
549 struct work_struct
*work
)
551 struct xfs_mount
*mp
= container_of(work
,
552 struct xfs_mount
, m_flush_work
);
554 xfs_sync_data(mp
, SYNC_TRYLOCK
);
555 xfs_sync_data(mp
, SYNC_TRYLOCK
| SYNC_WAIT
);
560 struct xfs_mount
*mp
)
562 INIT_WORK(&mp
->m_flush_work
, xfs_flush_worker
);
563 INIT_DELAYED_WORK(&mp
->m_sync_work
, xfs_sync_worker
);
564 INIT_DELAYED_WORK(&mp
->m_reclaim_work
, xfs_reclaim_worker
);
566 xfs_syncd_queue_sync(mp
);
567 xfs_syncd_queue_reclaim(mp
);
574 struct xfs_mount
*mp
)
576 cancel_delayed_work_sync(&mp
->m_sync_work
);
577 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
578 cancel_work_sync(&mp
->m_flush_work
);
582 __xfs_inode_set_reclaim_tag(
583 struct xfs_perag
*pag
,
584 struct xfs_inode
*ip
)
586 radix_tree_tag_set(&pag
->pag_ici_root
,
587 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
588 XFS_ICI_RECLAIM_TAG
);
590 if (!pag
->pag_ici_reclaimable
) {
591 /* propagate the reclaim tag up into the perag radix tree */
592 spin_lock(&ip
->i_mount
->m_perag_lock
);
593 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
594 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
595 XFS_ICI_RECLAIM_TAG
);
596 spin_unlock(&ip
->i_mount
->m_perag_lock
);
598 /* schedule periodic background inode reclaim */
599 xfs_syncd_queue_reclaim(ip
->i_mount
);
601 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
604 pag
->pag_ici_reclaimable
++;
608 * We set the inode flag atomically with the radix tree tag.
609 * Once we get tag lookups on the radix tree, this inode flag
613 xfs_inode_set_reclaim_tag(
616 struct xfs_mount
*mp
= ip
->i_mount
;
617 struct xfs_perag
*pag
;
619 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
620 spin_lock(&pag
->pag_ici_lock
);
621 spin_lock(&ip
->i_flags_lock
);
622 __xfs_inode_set_reclaim_tag(pag
, ip
);
623 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
624 spin_unlock(&ip
->i_flags_lock
);
625 spin_unlock(&pag
->pag_ici_lock
);
630 __xfs_inode_clear_reclaim(
634 pag
->pag_ici_reclaimable
--;
635 if (!pag
->pag_ici_reclaimable
) {
636 /* clear the reclaim tag from the perag radix tree */
637 spin_lock(&ip
->i_mount
->m_perag_lock
);
638 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
639 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
640 XFS_ICI_RECLAIM_TAG
);
641 spin_unlock(&ip
->i_mount
->m_perag_lock
);
642 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
648 __xfs_inode_clear_reclaim_tag(
653 radix_tree_tag_clear(&pag
->pag_ici_root
,
654 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
655 __xfs_inode_clear_reclaim(pag
, ip
);
659 * Grab the inode for reclaim exclusively.
660 * Return 0 if we grabbed it, non-zero otherwise.
663 xfs_reclaim_inode_grab(
664 struct xfs_inode
*ip
,
667 ASSERT(rcu_read_lock_held());
669 /* quick check for stale RCU freed inode */
674 * If we are asked for non-blocking operation, do unlocked checks to
675 * see if the inode already is being flushed or in reclaim to avoid
678 if ((flags
& SYNC_TRYLOCK
) &&
679 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
683 * The radix tree lock here protects a thread in xfs_iget from racing
684 * with us starting reclaim on the inode. Once we have the
685 * XFS_IRECLAIM flag set it will not touch us.
687 * Due to RCU lookup, we may find inodes that have been freed and only
688 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
689 * aren't candidates for reclaim at all, so we must check the
690 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
692 spin_lock(&ip
->i_flags_lock
);
693 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
694 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
695 /* not a reclaim candidate. */
696 spin_unlock(&ip
->i_flags_lock
);
699 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
700 spin_unlock(&ip
->i_flags_lock
);
705 * Inodes in different states need to be treated differently, and the return
706 * value of xfs_iflush is not sufficient to get this right. The following table
707 * lists the inode states and the reclaim actions necessary for non-blocking
711 * inode state iflush ret required action
712 * --------------- ---------- ---------------
714 * shutdown EIO unpin and reclaim
715 * clean, unpinned 0 reclaim
716 * stale, unpinned 0 reclaim
717 * clean, pinned(*) 0 requeue
718 * stale, pinned EAGAIN requeue
719 * dirty, delwri ok 0 requeue
720 * dirty, delwri blocked EAGAIN requeue
721 * dirty, sync flush 0 reclaim
723 * (*) dgc: I don't think the clean, pinned state is possible but it gets
724 * handled anyway given the order of checks implemented.
726 * As can be seen from the table, the return value of xfs_iflush() is not
727 * sufficient to correctly decide the reclaim action here. The checks in
728 * xfs_iflush() might look like duplicates, but they are not.
730 * Also, because we get the flush lock first, we know that any inode that has
731 * been flushed delwri has had the flush completed by the time we check that
732 * the inode is clean. The clean inode check needs to be done before flushing
733 * the inode delwri otherwise we would loop forever requeuing clean inodes as
734 * we cannot tell apart a successful delwri flush and a clean inode from the
735 * return value of xfs_iflush().
737 * Note that because the inode is flushed delayed write by background
738 * writeback, the flush lock may already be held here and waiting on it can
739 * result in very long latencies. Hence for sync reclaims, where we wait on the
740 * flush lock, the caller should push out delayed write inodes first before
741 * trying to reclaim them to minimise the amount of time spent waiting. For
742 * background relaim, we just requeue the inode for the next pass.
744 * Hence the order of actions after gaining the locks should be:
746 * shutdown => unpin and reclaim
747 * pinned, delwri => requeue
748 * pinned, sync => unpin
751 * dirty, delwri => flush and requeue
752 * dirty, sync => flush, wait and reclaim
756 struct xfs_inode
*ip
,
757 struct xfs_perag
*pag
,
764 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
765 if (!xfs_iflock_nowait(ip
)) {
766 if (!(sync_mode
& SYNC_WAIT
))
770 * If we only have a single dirty inode in a cluster there is
771 * a fair chance that the AIL push may have pushed it into
772 * the buffer, but xfsbufd won't touch it until 30 seconds
773 * from now, and thus we will lock up here.
775 * Promote the inode buffer to the front of the delwri list
776 * and wake up xfsbufd now.
778 xfs_promote_inode(ip
);
782 if (is_bad_inode(VFS_I(ip
)))
784 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
788 if (xfs_ipincount(ip
)) {
789 if (!(sync_mode
& SYNC_WAIT
)) {
795 if (xfs_iflags_test(ip
, XFS_ISTALE
))
797 if (xfs_inode_clean(ip
))
801 * Now we have an inode that needs flushing.
803 * We do a nonblocking flush here even if we are doing a SYNC_WAIT
804 * reclaim as we can deadlock with inode cluster removal.
805 * xfs_ifree_cluster() can lock the inode buffer before it locks the
806 * ip->i_lock, and we are doing the exact opposite here. As a result,
807 * doing a blocking xfs_itobp() to get the cluster buffer will result
808 * in an ABBA deadlock with xfs_ifree_cluster().
810 * As xfs_ifree_cluser() must gather all inodes that are active in the
811 * cache to mark them stale, if we hit this case we don't actually want
812 * to do IO here - we want the inode marked stale so we can simply
813 * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush,
814 * just unlock the inode, back off and try again. Hopefully the next
815 * pass through will see the stale flag set on the inode.
817 error
= xfs_iflush(ip
, SYNC_TRYLOCK
| sync_mode
);
818 if (sync_mode
& SYNC_WAIT
) {
819 if (error
== EAGAIN
) {
820 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
821 /* backoff longer than in xfs_ifree_cluster */
830 * When we have to flush an inode but don't have SYNC_WAIT set, we
831 * flush the inode out using a delwri buffer and wait for the next
832 * call into reclaim to find it in a clean state instead of waiting for
833 * it now. We also don't return errors here - if the error is transient
834 * then the next reclaim pass will flush the inode, and if the error
835 * is permanent then the next sync reclaim will reclaim the inode and
838 if (error
&& error
!= EAGAIN
&& !XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
839 xfs_warn(ip
->i_mount
,
840 "inode 0x%llx background reclaim flush failed with %d",
841 (long long)ip
->i_ino
, error
);
844 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
845 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
847 * We could return EAGAIN here to make reclaim rescan the inode tree in
848 * a short while. However, this just burns CPU time scanning the tree
849 * waiting for IO to complete and xfssyncd never goes back to the idle
850 * state. Instead, return 0 to let the next scheduled background reclaim
851 * attempt to reclaim the inode again.
857 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
859 XFS_STATS_INC(xs_ig_reclaims
);
861 * Remove the inode from the per-AG radix tree.
863 * Because radix_tree_delete won't complain even if the item was never
864 * added to the tree assert that it's been there before to catch
865 * problems with the inode life time early on.
867 spin_lock(&pag
->pag_ici_lock
);
868 if (!radix_tree_delete(&pag
->pag_ici_root
,
869 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
871 __xfs_inode_clear_reclaim(pag
, ip
);
872 spin_unlock(&pag
->pag_ici_lock
);
875 * Here we do an (almost) spurious inode lock in order to coordinate
876 * with inode cache radix tree lookups. This is because the lookup
877 * can reference the inodes in the cache without taking references.
879 * We make that OK here by ensuring that we wait until the inode is
880 * unlocked after the lookup before we go ahead and free it.
882 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
884 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
892 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
893 * corrupted, we still want to try to reclaim all the inodes. If we don't,
894 * then a shut down during filesystem unmount reclaim walk leak all the
895 * unreclaimed inodes.
898 xfs_reclaim_inodes_ag(
899 struct xfs_mount
*mp
,
903 struct xfs_perag
*pag
;
907 int trylock
= flags
& SYNC_TRYLOCK
;
913 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
914 unsigned long first_index
= 0;
918 ag
= pag
->pag_agno
+ 1;
921 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
926 first_index
= pag
->pag_ici_reclaim_cursor
;
928 mutex_lock(&pag
->pag_ici_reclaim_lock
);
931 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
935 nr_found
= radix_tree_gang_lookup_tag(
937 (void **)batch
, first_index
,
939 XFS_ICI_RECLAIM_TAG
);
947 * Grab the inodes before we drop the lock. if we found
948 * nothing, nr == 0 and the loop will be skipped.
950 for (i
= 0; i
< nr_found
; i
++) {
951 struct xfs_inode
*ip
= batch
[i
];
953 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
957 * Update the index for the next lookup. Catch
958 * overflows into the next AG range which can
959 * occur if we have inodes in the last block of
960 * the AG and we are currently pointing to the
963 * Because we may see inodes that are from the
964 * wrong AG due to RCU freeing and
965 * reallocation, only update the index if it
966 * lies in this AG. It was a race that lead us
967 * to see this inode, so another lookup from
968 * the same index will not find it again.
970 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
973 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
974 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
978 /* unlock now we've grabbed the inodes. */
981 for (i
= 0; i
< nr_found
; i
++) {
984 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
985 if (error
&& last_error
!= EFSCORRUPTED
)
989 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
993 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
995 if (trylock
&& !done
)
996 pag
->pag_ici_reclaim_cursor
= first_index
;
998 pag
->pag_ici_reclaim_cursor
= 0;
999 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1004 * if we skipped any AG, and we still have scan count remaining, do
1005 * another pass this time using blocking reclaim semantics (i.e
1006 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1007 * ensure that when we get more reclaimers than AGs we block rather
1008 * than spin trying to execute reclaim.
1010 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1014 return XFS_ERROR(last_error
);
1022 int nr_to_scan
= INT_MAX
;
1024 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1028 * Scan a certain number of inodes for reclaim.
1030 * When called we make sure that there is a background (fast) inode reclaim in
1031 * progress, while we will throttle the speed of reclaim via doing synchronous
1032 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1033 * them to be cleaned, which we hope will not be very long due to the
1034 * background walker having already kicked the IO off on those dirty inodes.
1037 xfs_reclaim_inodes_nr(
1038 struct xfs_mount
*mp
,
1041 /* kick background reclaimer and push the AIL */
1042 xfs_syncd_queue_reclaim(mp
);
1043 xfs_ail_push_all(mp
->m_ail
);
1045 xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1049 * Return the number of reclaimable inodes in the filesystem for
1050 * the shrinker to determine how much to reclaim.
1053 xfs_reclaim_inodes_count(
1054 struct xfs_mount
*mp
)
1056 struct xfs_perag
*pag
;
1057 xfs_agnumber_t ag
= 0;
1058 int reclaimable
= 0;
1060 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1061 ag
= pag
->pag_agno
+ 1;
1062 reclaimable
+= pag
->pag_ici_reclaimable
;