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
);
243 if (flags
& SYNC_WAIT
)
250 struct xfs_inode
*ip
,
251 struct xfs_perag
*pag
,
256 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
257 if (xfs_inode_clean(ip
))
259 if (!xfs_iflock_nowait(ip
)) {
260 if (!(flags
& SYNC_WAIT
))
265 if (xfs_inode_clean(ip
)) {
270 error
= xfs_iflush(ip
, flags
);
273 * We don't want to try again on non-blocking flushes that can't run
274 * again immediately. If an inode really must be written, then that's
275 * what the SYNC_WAIT flag is for.
277 if (error
== EAGAIN
) {
278 ASSERT(!(flags
& SYNC_WAIT
));
283 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
288 * Write out pagecache data for the whole filesystem.
292 struct xfs_mount
*mp
,
297 ASSERT((flags
& ~(SYNC_TRYLOCK
|SYNC_WAIT
)) == 0);
299 error
= xfs_inode_ag_iterator(mp
, xfs_sync_inode_data
, flags
);
301 return XFS_ERROR(error
);
303 xfs_log_force(mp
, (flags
& SYNC_WAIT
) ? XFS_LOG_SYNC
: 0);
308 * Write out inode metadata (attributes) for the whole filesystem.
312 struct xfs_mount
*mp
,
315 ASSERT((flags
& ~SYNC_WAIT
) == 0);
317 return xfs_inode_ag_iterator(mp
, xfs_sync_inode_attr
, flags
);
322 struct xfs_mount
*mp
)
327 * If the buffer is pinned then push on the log so we won't get stuck
328 * waiting in the write for someone, maybe ourselves, to flush the log.
330 * Even though we just pushed the log above, we did not have the
331 * superblock buffer locked at that point so it can become pinned in
332 * between there and here.
334 bp
= xfs_getsb(mp
, 0);
335 if (XFS_BUF_ISPINNED(bp
))
336 xfs_log_force(mp
, 0);
338 return xfs_bwrite(mp
, bp
);
342 * When remounting a filesystem read-only or freezing the filesystem, we have
343 * two phases to execute. This first phase is syncing the data before we
344 * quiesce the filesystem, and the second is flushing all the inodes out after
345 * we've waited for all the transactions created by the first phase to
346 * complete. The second phase ensures that the inodes are written to their
347 * location on disk rather than just existing in transactions in the log. This
348 * means after a quiesce there is no log replay required to write the inodes to
349 * disk (this is the main difference between a sync and a quiesce).
352 * First stage of freeze - no writers will make progress now we are here,
353 * so we flush delwri and delalloc buffers here, then wait for all I/O to
354 * complete. Data is frozen at that point. Metadata is not frozen,
355 * transactions can still occur here so don't bother flushing the buftarg
356 * because it'll just get dirty again.
360 struct xfs_mount
*mp
)
362 int error
, error2
= 0;
364 /* push non-blocking */
365 xfs_sync_data(mp
, 0);
366 xfs_qm_sync(mp
, SYNC_TRYLOCK
);
368 /* push and block till complete */
369 xfs_sync_data(mp
, SYNC_WAIT
);
370 xfs_qm_sync(mp
, SYNC_WAIT
);
372 /* write superblock and hoover up shutdown errors */
373 error
= xfs_sync_fsdata(mp
);
375 /* make sure all delwri buffers are written out */
376 xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
378 /* mark the log as covered if needed */
379 if (xfs_log_need_covered(mp
))
380 error2
= xfs_fs_log_dummy(mp
);
382 /* flush data-only devices */
383 if (mp
->m_rtdev_targp
)
384 XFS_bflush(mp
->m_rtdev_targp
);
386 return error
? error
: error2
;
391 struct xfs_mount
*mp
)
393 int count
= 0, pincount
;
395 xfs_reclaim_inodes(mp
, 0);
396 xfs_flush_buftarg(mp
->m_ddev_targp
, 0);
399 * This loop must run at least twice. The first instance of the loop
400 * will flush most meta data but that will generate more meta data
401 * (typically directory updates). Which then must be flushed and
402 * logged before we can write the unmount record. We also so sync
403 * reclaim of inodes to catch any that the above delwri flush skipped.
406 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
407 xfs_sync_attr(mp
, SYNC_WAIT
);
408 pincount
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
417 * Second stage of a quiesce. The data is already synced, now we have to take
418 * care of the metadata. New transactions are already blocked, so we need to
419 * wait for any remaining transactions to drain out before proceeding.
423 struct xfs_mount
*mp
)
427 /* wait for all modifications to complete */
428 while (atomic_read(&mp
->m_active_trans
) > 0)
431 /* flush inodes and push all remaining buffers out to disk */
435 * Just warn here till VFS can correctly support
436 * read-only remount without racing.
438 WARN_ON(atomic_read(&mp
->m_active_trans
) != 0);
440 /* Push the superblock and write an unmount record */
441 error
= xfs_log_sbcount(mp
, 1);
443 xfs_warn(mp
, "xfs_attr_quiesce: failed to log sb changes. "
444 "Frozen image may not be consistent.");
445 xfs_log_unmount_write(mp
);
446 xfs_unmountfs_writesb(mp
);
450 xfs_syncd_queue_sync(
451 struct xfs_mount
*mp
)
453 queue_delayed_work(xfs_syncd_wq
, &mp
->m_sync_work
,
454 msecs_to_jiffies(xfs_syncd_centisecs
* 10));
458 * Every sync period we need to unpin all items, reclaim inodes and sync
459 * disk quotas. We might need to cover the log to indicate that the
460 * filesystem is idle and not frozen.
464 struct work_struct
*work
)
466 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
467 struct xfs_mount
, m_sync_work
);
470 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
471 /* dgc: errors ignored here */
472 if (mp
->m_super
->s_frozen
== SB_UNFROZEN
&&
473 xfs_log_need_covered(mp
))
474 error
= xfs_fs_log_dummy(mp
);
476 xfs_log_force(mp
, 0);
477 error
= xfs_qm_sync(mp
, SYNC_TRYLOCK
);
479 /* start pushing all the metadata that is currently dirty */
480 xfs_ail_push_all(mp
->m_ail
);
483 /* queue us up again */
484 xfs_syncd_queue_sync(mp
);
488 * Queue a new inode reclaim pass if there are reclaimable inodes and there
489 * isn't a reclaim pass already in progress. By default it runs every 5s based
490 * on the xfs syncd work default of 30s. Perhaps this should have it's own
491 * tunable, but that can be done if this method proves to be ineffective or too
495 xfs_syncd_queue_reclaim(
496 struct xfs_mount
*mp
)
500 * We can have inodes enter reclaim after we've shut down the syncd
501 * workqueue during unmount, so don't allow reclaim work to be queued
504 if (!(mp
->m_super
->s_flags
& MS_ACTIVE
))
508 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
509 queue_delayed_work(xfs_syncd_wq
, &mp
->m_reclaim_work
,
510 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
516 * This is a fast pass over the inode cache to try to get reclaim moving on as
517 * many inodes as possible in a short period of time. It kicks itself every few
518 * seconds, as well as being kicked by the inode cache shrinker when memory
519 * goes low. It scans as quickly as possible avoiding locked inodes or those
520 * already being flushed, and once done schedules a future pass.
524 struct work_struct
*work
)
526 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
527 struct xfs_mount
, m_reclaim_work
);
529 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
530 xfs_syncd_queue_reclaim(mp
);
534 * Flush delayed allocate data, attempting to free up reserved space
535 * from existing allocations. At this point a new allocation attempt
536 * has failed with ENOSPC and we are in the process of scratching our
537 * heads, looking about for more room.
539 * Queue a new data flush if there isn't one already in progress and
540 * wait for completion of the flush. This means that we only ever have one
541 * inode flush in progress no matter how many ENOSPC events are occurring and
542 * so will prevent the system from bogging down due to every concurrent
543 * ENOSPC event scanning all the active inodes in the system for writeback.
547 struct xfs_inode
*ip
)
549 struct xfs_mount
*mp
= ip
->i_mount
;
551 queue_work(xfs_syncd_wq
, &mp
->m_flush_work
);
552 flush_work_sync(&mp
->m_flush_work
);
557 struct work_struct
*work
)
559 struct xfs_mount
*mp
= container_of(work
,
560 struct xfs_mount
, m_flush_work
);
562 xfs_sync_data(mp
, SYNC_TRYLOCK
);
563 xfs_sync_data(mp
, SYNC_TRYLOCK
| SYNC_WAIT
);
568 struct xfs_mount
*mp
)
570 INIT_WORK(&mp
->m_flush_work
, xfs_flush_worker
);
571 INIT_DELAYED_WORK(&mp
->m_sync_work
, xfs_sync_worker
);
572 INIT_DELAYED_WORK(&mp
->m_reclaim_work
, xfs_reclaim_worker
);
574 xfs_syncd_queue_sync(mp
);
575 xfs_syncd_queue_reclaim(mp
);
582 struct xfs_mount
*mp
)
584 cancel_delayed_work_sync(&mp
->m_sync_work
);
585 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
586 cancel_work_sync(&mp
->m_flush_work
);
590 __xfs_inode_set_reclaim_tag(
591 struct xfs_perag
*pag
,
592 struct xfs_inode
*ip
)
594 radix_tree_tag_set(&pag
->pag_ici_root
,
595 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
596 XFS_ICI_RECLAIM_TAG
);
598 if (!pag
->pag_ici_reclaimable
) {
599 /* propagate the reclaim tag up into the perag radix tree */
600 spin_lock(&ip
->i_mount
->m_perag_lock
);
601 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
602 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
603 XFS_ICI_RECLAIM_TAG
);
604 spin_unlock(&ip
->i_mount
->m_perag_lock
);
606 /* schedule periodic background inode reclaim */
607 xfs_syncd_queue_reclaim(ip
->i_mount
);
609 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
612 pag
->pag_ici_reclaimable
++;
616 * We set the inode flag atomically with the radix tree tag.
617 * Once we get tag lookups on the radix tree, this inode flag
621 xfs_inode_set_reclaim_tag(
624 struct xfs_mount
*mp
= ip
->i_mount
;
625 struct xfs_perag
*pag
;
627 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
628 spin_lock(&pag
->pag_ici_lock
);
629 spin_lock(&ip
->i_flags_lock
);
630 __xfs_inode_set_reclaim_tag(pag
, ip
);
631 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
632 spin_unlock(&ip
->i_flags_lock
);
633 spin_unlock(&pag
->pag_ici_lock
);
638 __xfs_inode_clear_reclaim(
642 pag
->pag_ici_reclaimable
--;
643 if (!pag
->pag_ici_reclaimable
) {
644 /* clear the reclaim tag from the perag radix tree */
645 spin_lock(&ip
->i_mount
->m_perag_lock
);
646 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
647 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
648 XFS_ICI_RECLAIM_TAG
);
649 spin_unlock(&ip
->i_mount
->m_perag_lock
);
650 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
656 __xfs_inode_clear_reclaim_tag(
661 radix_tree_tag_clear(&pag
->pag_ici_root
,
662 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
663 __xfs_inode_clear_reclaim(pag
, ip
);
667 * Grab the inode for reclaim exclusively.
668 * Return 0 if we grabbed it, non-zero otherwise.
671 xfs_reclaim_inode_grab(
672 struct xfs_inode
*ip
,
675 ASSERT(rcu_read_lock_held());
677 /* quick check for stale RCU freed inode */
682 * do some unlocked checks first to avoid unnecessary lock traffic.
683 * The first is a flush lock check, the second is a already in reclaim
684 * check. Only do these checks if we are not going to block on locks.
686 if ((flags
& SYNC_TRYLOCK
) &&
687 (!ip
->i_flush
.done
|| __xfs_iflags_test(ip
, XFS_IRECLAIM
))) {
692 * The radix tree lock here protects a thread in xfs_iget from racing
693 * with us starting reclaim on the inode. Once we have the
694 * XFS_IRECLAIM flag set it will not touch us.
696 * Due to RCU lookup, we may find inodes that have been freed and only
697 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
698 * aren't candidates for reclaim at all, so we must check the
699 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
701 spin_lock(&ip
->i_flags_lock
);
702 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
703 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
704 /* not a reclaim candidate. */
705 spin_unlock(&ip
->i_flags_lock
);
708 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
709 spin_unlock(&ip
->i_flags_lock
);
714 * Inodes in different states need to be treated differently, and the return
715 * value of xfs_iflush is not sufficient to get this right. The following table
716 * lists the inode states and the reclaim actions necessary for non-blocking
720 * inode state iflush ret required action
721 * --------------- ---------- ---------------
723 * shutdown EIO unpin and reclaim
724 * clean, unpinned 0 reclaim
725 * stale, unpinned 0 reclaim
726 * clean, pinned(*) 0 requeue
727 * stale, pinned EAGAIN requeue
728 * dirty, delwri ok 0 requeue
729 * dirty, delwri blocked EAGAIN requeue
730 * dirty, sync flush 0 reclaim
732 * (*) dgc: I don't think the clean, pinned state is possible but it gets
733 * handled anyway given the order of checks implemented.
735 * As can be seen from the table, the return value of xfs_iflush() is not
736 * sufficient to correctly decide the reclaim action here. The checks in
737 * xfs_iflush() might look like duplicates, but they are not.
739 * Also, because we get the flush lock first, we know that any inode that has
740 * been flushed delwri has had the flush completed by the time we check that
741 * the inode is clean. The clean inode check needs to be done before flushing
742 * the inode delwri otherwise we would loop forever requeuing clean inodes as
743 * we cannot tell apart a successful delwri flush and a clean inode from the
744 * return value of xfs_iflush().
746 * Note that because the inode is flushed delayed write by background
747 * writeback, the flush lock may already be held here and waiting on it can
748 * result in very long latencies. Hence for sync reclaims, where we wait on the
749 * flush lock, the caller should push out delayed write inodes first before
750 * trying to reclaim them to minimise the amount of time spent waiting. For
751 * background relaim, we just requeue the inode for the next pass.
753 * Hence the order of actions after gaining the locks should be:
755 * shutdown => unpin and reclaim
756 * pinned, delwri => requeue
757 * pinned, sync => unpin
760 * dirty, delwri => flush and requeue
761 * dirty, sync => flush, wait and reclaim
765 struct xfs_inode
*ip
,
766 struct xfs_perag
*pag
,
773 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
774 if (!xfs_iflock_nowait(ip
)) {
775 if (!(sync_mode
& SYNC_WAIT
))
780 if (is_bad_inode(VFS_I(ip
)))
782 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
786 if (xfs_ipincount(ip
)) {
787 if (!(sync_mode
& SYNC_WAIT
)) {
793 if (xfs_iflags_test(ip
, XFS_ISTALE
))
795 if (xfs_inode_clean(ip
))
799 * Now we have an inode that needs flushing.
801 * We do a nonblocking flush here even if we are doing a SYNC_WAIT
802 * reclaim as we can deadlock with inode cluster removal.
803 * xfs_ifree_cluster() can lock the inode buffer before it locks the
804 * ip->i_lock, and we are doing the exact opposite here. As a result,
805 * doing a blocking xfs_itobp() to get the cluster buffer will result
806 * in an ABBA deadlock with xfs_ifree_cluster().
808 * As xfs_ifree_cluser() must gather all inodes that are active in the
809 * cache to mark them stale, if we hit this case we don't actually want
810 * to do IO here - we want the inode marked stale so we can simply
811 * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush,
812 * just unlock the inode, back off and try again. Hopefully the next
813 * pass through will see the stale flag set on the inode.
815 error
= xfs_iflush(ip
, SYNC_TRYLOCK
| sync_mode
);
816 if (sync_mode
& SYNC_WAIT
) {
817 if (error
== EAGAIN
) {
818 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
819 /* backoff longer than in xfs_ifree_cluster */
828 * When we have to flush an inode but don't have SYNC_WAIT set, we
829 * flush the inode out using a delwri buffer and wait for the next
830 * call into reclaim to find it in a clean state instead of waiting for
831 * it now. We also don't return errors here - if the error is transient
832 * then the next reclaim pass will flush the inode, and if the error
833 * is permanent then the next sync reclaim will reclaim the inode and
836 if (error
&& error
!= EAGAIN
&& !XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
837 xfs_warn(ip
->i_mount
,
838 "inode 0x%llx background reclaim flush failed with %d",
839 (long long)ip
->i_ino
, error
);
842 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
843 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
845 * We could return EAGAIN here to make reclaim rescan the inode tree in
846 * a short while. However, this just burns CPU time scanning the tree
847 * waiting for IO to complete and xfssyncd never goes back to the idle
848 * state. Instead, return 0 to let the next scheduled background reclaim
849 * attempt to reclaim the inode again.
855 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
857 XFS_STATS_INC(xs_ig_reclaims
);
859 * Remove the inode from the per-AG radix tree.
861 * Because radix_tree_delete won't complain even if the item was never
862 * added to the tree assert that it's been there before to catch
863 * problems with the inode life time early on.
865 spin_lock(&pag
->pag_ici_lock
);
866 if (!radix_tree_delete(&pag
->pag_ici_root
,
867 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
869 __xfs_inode_clear_reclaim(pag
, ip
);
870 spin_unlock(&pag
->pag_ici_lock
);
873 * Here we do an (almost) spurious inode lock in order to coordinate
874 * with inode cache radix tree lookups. This is because the lookup
875 * can reference the inodes in the cache without taking references.
877 * We make that OK here by ensuring that we wait until the inode is
878 * unlocked after the lookup before we go ahead and free it. We get
879 * both the ilock and the iolock because the code may need to drop the
880 * ilock one but will still hold the iolock.
882 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
884 xfs_iunlock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_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
;