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
);
245 * Write out pagecache data for the whole filesystem.
249 struct xfs_mount
*mp
,
254 ASSERT((flags
& ~(SYNC_TRYLOCK
|SYNC_WAIT
)) == 0);
256 error
= xfs_inode_ag_iterator(mp
, xfs_sync_inode_data
, flags
);
258 return XFS_ERROR(error
);
260 xfs_log_force(mp
, (flags
& SYNC_WAIT
) ? XFS_LOG_SYNC
: 0);
266 struct xfs_mount
*mp
)
272 * If the buffer is pinned then push on the log so we won't get stuck
273 * waiting in the write for someone, maybe ourselves, to flush the log.
275 * Even though we just pushed the log above, we did not have the
276 * superblock buffer locked at that point so it can become pinned in
277 * between there and here.
279 bp
= xfs_getsb(mp
, 0);
280 if (xfs_buf_ispinned(bp
))
281 xfs_log_force(mp
, 0);
282 error
= xfs_bwrite(bp
);
288 * When remounting a filesystem read-only or freezing the filesystem, we have
289 * two phases to execute. This first phase is syncing the data before we
290 * quiesce the filesystem, and the second is flushing all the inodes out after
291 * we've waited for all the transactions created by the first phase to
292 * complete. The second phase ensures that the inodes are written to their
293 * location on disk rather than just existing in transactions in the log. This
294 * means after a quiesce there is no log replay required to write the inodes to
295 * disk (this is the main difference between a sync and a quiesce).
298 * First stage of freeze - no writers will make progress now we are here,
299 * so we flush delwri and delalloc buffers here, then wait for all I/O to
300 * complete. Data is frozen at that point. Metadata is not frozen,
301 * transactions can still occur here so don't bother emptying the AIL
302 * because it'll just get dirty again.
306 struct xfs_mount
*mp
)
308 int error
, error2
= 0;
310 /* force out the log */
311 xfs_log_force(mp
, XFS_LOG_SYNC
);
313 /* write superblock and hoover up shutdown errors */
314 error
= xfs_sync_fsdata(mp
);
316 /* mark the log as covered if needed */
317 if (xfs_log_need_covered(mp
))
318 error2
= xfs_fs_log_dummy(mp
);
320 return error
? error
: error2
;
324 * Second stage of a quiesce. The data is already synced, now we have to take
325 * care of the metadata. New transactions are already blocked, so we need to
326 * wait for any remaining transactions to drain out before proceeding.
330 struct xfs_mount
*mp
)
334 /* wait for all modifications to complete */
335 while (atomic_read(&mp
->m_active_trans
) > 0)
338 /* reclaim inodes to do any IO before the freeze completes */
339 xfs_reclaim_inodes(mp
, 0);
340 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
342 /* flush all pending changes from the AIL */
343 xfs_ail_push_all_sync(mp
->m_ail
);
346 * Just warn here till VFS can correctly support
347 * read-only remount without racing.
349 WARN_ON(atomic_read(&mp
->m_active_trans
) != 0);
351 /* Push the superblock and write an unmount record */
352 error
= xfs_log_sbcount(mp
);
354 xfs_warn(mp
, "xfs_attr_quiesce: failed to log sb changes. "
355 "Frozen image may not be consistent.");
356 xfs_log_unmount_write(mp
);
359 * At this point we might have modified the superblock again and thus
360 * added an item to the AIL, thus flush it again.
362 xfs_ail_push_all_sync(mp
->m_ail
);
366 xfs_syncd_queue_sync(
367 struct xfs_mount
*mp
)
369 queue_delayed_work(xfs_syncd_wq
, &mp
->m_sync_work
,
370 msecs_to_jiffies(xfs_syncd_centisecs
* 10));
374 * Every sync period we need to unpin all items, reclaim inodes and sync
375 * disk quotas. We might need to cover the log to indicate that the
376 * filesystem is idle and not frozen.
380 struct work_struct
*work
)
382 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
383 struct xfs_mount
, m_sync_work
);
387 * We shouldn't write/force the log if we are in the mount/unmount
388 * process or on a read only filesystem. The workqueue still needs to be
389 * active in both cases, however, because it is used for inode reclaim
390 * during these times. hence use the MS_ACTIVE flag to avoid doing
391 * anything in these periods.
393 if (!(mp
->m_super
->s_flags
& MS_ACTIVE
) &&
394 !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
395 /* dgc: errors ignored here */
396 if (mp
->m_super
->s_frozen
== SB_UNFROZEN
&&
397 xfs_log_need_covered(mp
))
398 error
= xfs_fs_log_dummy(mp
);
400 xfs_log_force(mp
, 0);
402 /* start pushing all the metadata that is currently dirty */
403 xfs_ail_push_all(mp
->m_ail
);
406 /* queue us up again */
407 xfs_syncd_queue_sync(mp
);
411 * Queue a new inode reclaim pass if there are reclaimable inodes and there
412 * isn't a reclaim pass already in progress. By default it runs every 5s based
413 * on the xfs syncd work default of 30s. Perhaps this should have it's own
414 * tunable, but that can be done if this method proves to be ineffective or too
418 xfs_syncd_queue_reclaim(
419 struct xfs_mount
*mp
)
423 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
424 queue_delayed_work(xfs_syncd_wq
, &mp
->m_reclaim_work
,
425 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
431 * This is a fast pass over the inode cache to try to get reclaim moving on as
432 * many inodes as possible in a short period of time. It kicks itself every few
433 * seconds, as well as being kicked by the inode cache shrinker when memory
434 * goes low. It scans as quickly as possible avoiding locked inodes or those
435 * already being flushed, and once done schedules a future pass.
439 struct work_struct
*work
)
441 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
442 struct xfs_mount
, m_reclaim_work
);
444 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
445 xfs_syncd_queue_reclaim(mp
);
449 * Flush delayed allocate data, attempting to free up reserved space
450 * from existing allocations. At this point a new allocation attempt
451 * has failed with ENOSPC and we are in the process of scratching our
452 * heads, looking about for more room.
454 * Queue a new data flush if there isn't one already in progress and
455 * wait for completion of the flush. This means that we only ever have one
456 * inode flush in progress no matter how many ENOSPC events are occurring and
457 * so will prevent the system from bogging down due to every concurrent
458 * ENOSPC event scanning all the active inodes in the system for writeback.
462 struct xfs_inode
*ip
)
464 struct xfs_mount
*mp
= ip
->i_mount
;
466 queue_work(xfs_syncd_wq
, &mp
->m_flush_work
);
467 flush_work_sync(&mp
->m_flush_work
);
472 struct work_struct
*work
)
474 struct xfs_mount
*mp
= container_of(work
,
475 struct xfs_mount
, m_flush_work
);
477 xfs_sync_data(mp
, SYNC_TRYLOCK
);
478 xfs_sync_data(mp
, SYNC_TRYLOCK
| SYNC_WAIT
);
483 struct xfs_mount
*mp
)
485 INIT_WORK(&mp
->m_flush_work
, xfs_flush_worker
);
486 INIT_DELAYED_WORK(&mp
->m_sync_work
, xfs_sync_worker
);
487 INIT_DELAYED_WORK(&mp
->m_reclaim_work
, xfs_reclaim_worker
);
489 xfs_syncd_queue_sync(mp
);
496 struct xfs_mount
*mp
)
498 cancel_delayed_work_sync(&mp
->m_sync_work
);
499 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
500 cancel_work_sync(&mp
->m_flush_work
);
504 __xfs_inode_set_reclaim_tag(
505 struct xfs_perag
*pag
,
506 struct xfs_inode
*ip
)
508 radix_tree_tag_set(&pag
->pag_ici_root
,
509 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
510 XFS_ICI_RECLAIM_TAG
);
512 if (!pag
->pag_ici_reclaimable
) {
513 /* propagate the reclaim tag up into the perag radix tree */
514 spin_lock(&ip
->i_mount
->m_perag_lock
);
515 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
516 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
517 XFS_ICI_RECLAIM_TAG
);
518 spin_unlock(&ip
->i_mount
->m_perag_lock
);
520 /* schedule periodic background inode reclaim */
521 xfs_syncd_queue_reclaim(ip
->i_mount
);
523 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
526 pag
->pag_ici_reclaimable
++;
530 * We set the inode flag atomically with the radix tree tag.
531 * Once we get tag lookups on the radix tree, this inode flag
535 xfs_inode_set_reclaim_tag(
538 struct xfs_mount
*mp
= ip
->i_mount
;
539 struct xfs_perag
*pag
;
541 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
542 spin_lock(&pag
->pag_ici_lock
);
543 spin_lock(&ip
->i_flags_lock
);
544 __xfs_inode_set_reclaim_tag(pag
, ip
);
545 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
546 spin_unlock(&ip
->i_flags_lock
);
547 spin_unlock(&pag
->pag_ici_lock
);
552 __xfs_inode_clear_reclaim(
556 pag
->pag_ici_reclaimable
--;
557 if (!pag
->pag_ici_reclaimable
) {
558 /* clear the reclaim tag from the perag radix tree */
559 spin_lock(&ip
->i_mount
->m_perag_lock
);
560 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
561 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
562 XFS_ICI_RECLAIM_TAG
);
563 spin_unlock(&ip
->i_mount
->m_perag_lock
);
564 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
570 __xfs_inode_clear_reclaim_tag(
575 radix_tree_tag_clear(&pag
->pag_ici_root
,
576 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
577 __xfs_inode_clear_reclaim(pag
, ip
);
581 * Grab the inode for reclaim exclusively.
582 * Return 0 if we grabbed it, non-zero otherwise.
585 xfs_reclaim_inode_grab(
586 struct xfs_inode
*ip
,
589 ASSERT(rcu_read_lock_held());
591 /* quick check for stale RCU freed inode */
596 * If we are asked for non-blocking operation, do unlocked checks to
597 * see if the inode already is being flushed or in reclaim to avoid
600 if ((flags
& SYNC_TRYLOCK
) &&
601 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
605 * The radix tree lock here protects a thread in xfs_iget from racing
606 * with us starting reclaim on the inode. Once we have the
607 * XFS_IRECLAIM flag set it will not touch us.
609 * Due to RCU lookup, we may find inodes that have been freed and only
610 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
611 * aren't candidates for reclaim at all, so we must check the
612 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
614 spin_lock(&ip
->i_flags_lock
);
615 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
616 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
617 /* not a reclaim candidate. */
618 spin_unlock(&ip
->i_flags_lock
);
621 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
622 spin_unlock(&ip
->i_flags_lock
);
627 * Inodes in different states need to be treated differently. The following
628 * table lists the inode states and the reclaim actions necessary:
630 * inode state iflush ret required action
631 * --------------- ---------- ---------------
633 * shutdown EIO unpin and reclaim
634 * clean, unpinned 0 reclaim
635 * stale, unpinned 0 reclaim
636 * clean, pinned(*) 0 requeue
637 * stale, pinned EAGAIN requeue
638 * dirty, async - requeue
639 * dirty, sync 0 reclaim
641 * (*) dgc: I don't think the clean, pinned state is possible but it gets
642 * handled anyway given the order of checks implemented.
644 * Also, because we get the flush lock first, we know that any inode that has
645 * been flushed delwri has had the flush completed by the time we check that
646 * the inode is clean.
648 * Note that because the inode is flushed delayed write by AIL pushing, the
649 * flush lock may already be held here and waiting on it can result in very
650 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
651 * the caller should push the AIL first before trying to reclaim inodes to
652 * minimise the amount of time spent waiting. For background relaim, we only
653 * bother to reclaim clean inodes anyway.
655 * Hence the order of actions after gaining the locks should be:
657 * shutdown => unpin and reclaim
658 * pinned, async => requeue
659 * pinned, sync => unpin
662 * dirty, async => requeue
663 * dirty, sync => flush, wait and reclaim
667 struct xfs_inode
*ip
,
668 struct xfs_perag
*pag
,
671 struct xfs_buf
*bp
= NULL
;
676 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
677 if (!xfs_iflock_nowait(ip
)) {
678 if (!(sync_mode
& SYNC_WAIT
))
683 if (is_bad_inode(VFS_I(ip
)))
685 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
687 xfs_iflush_abort(ip
, false);
690 if (xfs_ipincount(ip
)) {
691 if (!(sync_mode
& SYNC_WAIT
))
695 if (xfs_iflags_test(ip
, XFS_ISTALE
))
697 if (xfs_inode_clean(ip
))
701 * Never flush out dirty data during non-blocking reclaim, as it would
702 * just contend with AIL pushing trying to do the same job.
704 if (!(sync_mode
& SYNC_WAIT
))
708 * Now we have an inode that needs flushing.
710 * Note that xfs_iflush will never block on the inode buffer lock, as
711 * xfs_ifree_cluster() can lock the inode buffer before it locks the
712 * ip->i_lock, and we are doing the exact opposite here. As a result,
713 * doing a blocking xfs_itobp() to get the cluster buffer would result
714 * in an ABBA deadlock with xfs_ifree_cluster().
716 * As xfs_ifree_cluser() must gather all inodes that are active in the
717 * cache to mark them stale, if we hit this case we don't actually want
718 * to do IO here - we want the inode marked stale so we can simply
719 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
720 * inode, back off and try again. Hopefully the next pass through will
721 * see the stale flag set on the inode.
723 error
= xfs_iflush(ip
, &bp
);
724 if (error
== EAGAIN
) {
725 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
726 /* backoff longer than in xfs_ifree_cluster */
732 error
= xfs_bwrite(bp
);
739 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
741 XFS_STATS_INC(xs_ig_reclaims
);
743 * Remove the inode from the per-AG radix tree.
745 * Because radix_tree_delete won't complain even if the item was never
746 * added to the tree assert that it's been there before to catch
747 * problems with the inode life time early on.
749 spin_lock(&pag
->pag_ici_lock
);
750 if (!radix_tree_delete(&pag
->pag_ici_root
,
751 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
753 __xfs_inode_clear_reclaim(pag
, ip
);
754 spin_unlock(&pag
->pag_ici_lock
);
757 * Here we do an (almost) spurious inode lock in order to coordinate
758 * with inode cache radix tree lookups. This is because the lookup
759 * can reference the inodes in the cache without taking references.
761 * We make that OK here by ensuring that we wait until the inode is
762 * unlocked after the lookup before we go ahead and free it.
764 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
766 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
774 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
775 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
777 * We could return EAGAIN here to make reclaim rescan the inode tree in
778 * a short while. However, this just burns CPU time scanning the tree
779 * waiting for IO to complete and xfssyncd never goes back to the idle
780 * state. Instead, return 0 to let the next scheduled background reclaim
781 * attempt to reclaim the inode again.
787 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
788 * corrupted, we still want to try to reclaim all the inodes. If we don't,
789 * then a shut down during filesystem unmount reclaim walk leak all the
790 * unreclaimed inodes.
793 xfs_reclaim_inodes_ag(
794 struct xfs_mount
*mp
,
798 struct xfs_perag
*pag
;
802 int trylock
= flags
& SYNC_TRYLOCK
;
808 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
809 unsigned long first_index
= 0;
813 ag
= pag
->pag_agno
+ 1;
816 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
821 first_index
= pag
->pag_ici_reclaim_cursor
;
823 mutex_lock(&pag
->pag_ici_reclaim_lock
);
826 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
830 nr_found
= radix_tree_gang_lookup_tag(
832 (void **)batch
, first_index
,
834 XFS_ICI_RECLAIM_TAG
);
842 * Grab the inodes before we drop the lock. if we found
843 * nothing, nr == 0 and the loop will be skipped.
845 for (i
= 0; i
< nr_found
; i
++) {
846 struct xfs_inode
*ip
= batch
[i
];
848 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
852 * Update the index for the next lookup. Catch
853 * overflows into the next AG range which can
854 * occur if we have inodes in the last block of
855 * the AG and we are currently pointing to the
858 * Because we may see inodes that are from the
859 * wrong AG due to RCU freeing and
860 * reallocation, only update the index if it
861 * lies in this AG. It was a race that lead us
862 * to see this inode, so another lookup from
863 * the same index will not find it again.
865 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
868 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
869 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
873 /* unlock now we've grabbed the inodes. */
876 for (i
= 0; i
< nr_found
; i
++) {
879 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
880 if (error
&& last_error
!= EFSCORRUPTED
)
884 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
888 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
890 if (trylock
&& !done
)
891 pag
->pag_ici_reclaim_cursor
= first_index
;
893 pag
->pag_ici_reclaim_cursor
= 0;
894 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
899 * if we skipped any AG, and we still have scan count remaining, do
900 * another pass this time using blocking reclaim semantics (i.e
901 * waiting on the reclaim locks and ignoring the reclaim cursors). This
902 * ensure that when we get more reclaimers than AGs we block rather
903 * than spin trying to execute reclaim.
905 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
909 return XFS_ERROR(last_error
);
917 int nr_to_scan
= INT_MAX
;
919 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
923 * Scan a certain number of inodes for reclaim.
925 * When called we make sure that there is a background (fast) inode reclaim in
926 * progress, while we will throttle the speed of reclaim via doing synchronous
927 * reclaim of inodes. That means if we come across dirty inodes, we wait for
928 * them to be cleaned, which we hope will not be very long due to the
929 * background walker having already kicked the IO off on those dirty inodes.
932 xfs_reclaim_inodes_nr(
933 struct xfs_mount
*mp
,
936 /* kick background reclaimer and push the AIL */
937 xfs_syncd_queue_reclaim(mp
);
938 xfs_ail_push_all(mp
->m_ail
);
940 xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
944 * Return the number of reclaimable inodes in the filesystem for
945 * the shrinker to determine how much to reclaim.
948 xfs_reclaim_inodes_count(
949 struct xfs_mount
*mp
)
951 struct xfs_perag
*pag
;
952 xfs_agnumber_t ag
= 0;
955 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
956 ag
= pag
->pag_agno
+ 1;
957 reclaimable
+= pag
->pag_ici_reclaimable
;
This page took 0.049511 seconds and 5 git commands to generate.