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
)
182 } while (nr_found
&& !done
);
192 xfs_inode_ag_iterator(
193 struct xfs_mount
*mp
,
194 int (*execute
)(struct xfs_inode
*ip
,
195 struct xfs_perag
*pag
, int flags
),
198 struct xfs_perag
*pag
;
204 while ((pag
= xfs_perag_get(mp
, ag
))) {
205 ag
= pag
->pag_agno
+ 1;
206 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
);
210 if (error
== EFSCORRUPTED
)
214 return XFS_ERROR(last_error
);
219 struct xfs_inode
*ip
,
220 struct xfs_perag
*pag
,
223 struct inode
*inode
= VFS_I(ip
);
224 struct address_space
*mapping
= inode
->i_mapping
;
227 if (!mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
230 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
)) {
231 if (flags
& SYNC_TRYLOCK
)
233 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
236 error
= xfs_flush_pages(ip
, 0, -1, (flags
& SYNC_WAIT
) ?
237 0 : XBF_ASYNC
, FI_NONE
);
238 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
241 if (flags
& SYNC_WAIT
)
248 struct xfs_inode
*ip
,
249 struct xfs_perag
*pag
,
254 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
255 if (xfs_inode_clean(ip
))
257 if (!xfs_iflock_nowait(ip
)) {
258 if (!(flags
& SYNC_WAIT
))
263 if (xfs_inode_clean(ip
)) {
268 error
= xfs_iflush(ip
, flags
);
271 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
276 * Write out pagecache data for the whole filesystem.
280 struct xfs_mount
*mp
,
285 ASSERT((flags
& ~(SYNC_TRYLOCK
|SYNC_WAIT
)) == 0);
287 error
= xfs_inode_ag_iterator(mp
, xfs_sync_inode_data
, flags
);
289 return XFS_ERROR(error
);
291 xfs_log_force(mp
, (flags
& SYNC_WAIT
) ? XFS_LOG_SYNC
: 0);
296 * Write out inode metadata (attributes) for the whole filesystem.
300 struct xfs_mount
*mp
,
303 ASSERT((flags
& ~SYNC_WAIT
) == 0);
305 return xfs_inode_ag_iterator(mp
, xfs_sync_inode_attr
, flags
);
310 struct xfs_mount
*mp
)
315 * If the buffer is pinned then push on the log so we won't get stuck
316 * waiting in the write for someone, maybe ourselves, to flush the log.
318 * Even though we just pushed the log above, we did not have the
319 * superblock buffer locked at that point so it can become pinned in
320 * between there and here.
322 bp
= xfs_getsb(mp
, 0);
323 if (XFS_BUF_ISPINNED(bp
))
324 xfs_log_force(mp
, 0);
326 return xfs_bwrite(mp
, bp
);
330 * When remounting a filesystem read-only or freezing the filesystem, we have
331 * two phases to execute. This first phase is syncing the data before we
332 * quiesce the filesystem, and the second is flushing all the inodes out after
333 * we've waited for all the transactions created by the first phase to
334 * complete. The second phase ensures that the inodes are written to their
335 * location on disk rather than just existing in transactions in the log. This
336 * means after a quiesce there is no log replay required to write the inodes to
337 * disk (this is the main difference between a sync and a quiesce).
340 * First stage of freeze - no writers will make progress now we are here,
341 * so we flush delwri and delalloc buffers here, then wait for all I/O to
342 * complete. Data is frozen at that point. Metadata is not frozen,
343 * transactions can still occur here so don't bother flushing the buftarg
344 * because it'll just get dirty again.
348 struct xfs_mount
*mp
)
350 int error
, error2
= 0;
352 /* push non-blocking */
353 xfs_sync_data(mp
, 0);
354 xfs_qm_sync(mp
, SYNC_TRYLOCK
);
356 /* push and block till complete */
357 xfs_sync_data(mp
, SYNC_WAIT
);
358 xfs_qm_sync(mp
, SYNC_WAIT
);
360 /* write superblock and hoover up shutdown errors */
361 error
= xfs_sync_fsdata(mp
);
363 /* make sure all delwri buffers are written out */
364 xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
366 /* mark the log as covered if needed */
367 if (xfs_log_need_covered(mp
))
368 error2
= xfs_fs_log_dummy(mp
);
370 /* flush data-only devices */
371 if (mp
->m_rtdev_targp
)
372 XFS_bflush(mp
->m_rtdev_targp
);
374 return error
? error
: error2
;
379 struct xfs_mount
*mp
)
381 int count
= 0, pincount
;
383 xfs_reclaim_inodes(mp
, 0);
384 xfs_flush_buftarg(mp
->m_ddev_targp
, 0);
387 * This loop must run at least twice. The first instance of the loop
388 * will flush most meta data but that will generate more meta data
389 * (typically directory updates). Which then must be flushed and
390 * logged before we can write the unmount record. We also so sync
391 * reclaim of inodes to catch any that the above delwri flush skipped.
394 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
395 xfs_sync_attr(mp
, SYNC_WAIT
);
396 pincount
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
405 * Second stage of a quiesce. The data is already synced, now we have to take
406 * care of the metadata. New transactions are already blocked, so we need to
407 * wait for any remaining transactions to drain out before proceeding.
411 struct xfs_mount
*mp
)
415 /* wait for all modifications to complete */
416 while (atomic_read(&mp
->m_active_trans
) > 0)
419 /* flush inodes and push all remaining buffers out to disk */
423 * Just warn here till VFS can correctly support
424 * read-only remount without racing.
426 WARN_ON(atomic_read(&mp
->m_active_trans
) != 0);
428 /* Push the superblock and write an unmount record */
429 error
= xfs_log_sbcount(mp
, 1);
431 xfs_warn(mp
, "xfs_attr_quiesce: failed to log sb changes. "
432 "Frozen image may not be consistent.");
433 xfs_log_unmount_write(mp
);
434 xfs_unmountfs_writesb(mp
);
438 xfs_syncd_queue_sync(
439 struct xfs_mount
*mp
)
441 queue_delayed_work(xfs_syncd_wq
, &mp
->m_sync_work
,
442 msecs_to_jiffies(xfs_syncd_centisecs
* 10));
446 * Every sync period we need to unpin all items, reclaim inodes and sync
447 * disk quotas. We might need to cover the log to indicate that the
448 * filesystem is idle and not frozen.
452 struct work_struct
*work
)
454 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
455 struct xfs_mount
, m_sync_work
);
458 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
459 /* dgc: errors ignored here */
460 if (mp
->m_super
->s_frozen
== SB_UNFROZEN
&&
461 xfs_log_need_covered(mp
))
462 error
= xfs_fs_log_dummy(mp
);
464 xfs_log_force(mp
, 0);
465 error
= xfs_qm_sync(mp
, SYNC_TRYLOCK
);
467 /* start pushing all the metadata that is currently dirty */
468 xfs_ail_push_all(mp
->m_ail
);
471 /* queue us up again */
472 xfs_syncd_queue_sync(mp
);
476 * Queue a new inode reclaim pass if there are reclaimable inodes and there
477 * isn't a reclaim pass already in progress. By default it runs every 5s based
478 * on the xfs syncd work default of 30s. Perhaps this should have it's own
479 * tunable, but that can be done if this method proves to be ineffective or too
483 xfs_syncd_queue_reclaim(
484 struct xfs_mount
*mp
)
488 * We can have inodes enter reclaim after we've shut down the syncd
489 * workqueue during unmount, so don't allow reclaim work to be queued
492 if (!(mp
->m_super
->s_flags
& MS_ACTIVE
))
496 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
497 queue_delayed_work(xfs_syncd_wq
, &mp
->m_reclaim_work
,
498 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
504 * This is a fast pass over the inode cache to try to get reclaim moving on as
505 * many inodes as possible in a short period of time. It kicks itself every few
506 * seconds, as well as being kicked by the inode cache shrinker when memory
507 * goes low. It scans as quickly as possible avoiding locked inodes or those
508 * already being flushed, and once done schedules a future pass.
512 struct work_struct
*work
)
514 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
515 struct xfs_mount
, m_reclaim_work
);
517 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
518 xfs_syncd_queue_reclaim(mp
);
522 * Flush delayed allocate data, attempting to free up reserved space
523 * from existing allocations. At this point a new allocation attempt
524 * has failed with ENOSPC and we are in the process of scratching our
525 * heads, looking about for more room.
527 * Queue a new data flush if there isn't one already in progress and
528 * wait for completion of the flush. This means that we only ever have one
529 * inode flush in progress no matter how many ENOSPC events are occurring and
530 * so will prevent the system from bogging down due to every concurrent
531 * ENOSPC event scanning all the active inodes in the system for writeback.
535 struct xfs_inode
*ip
)
537 struct xfs_mount
*mp
= ip
->i_mount
;
539 queue_work(xfs_syncd_wq
, &mp
->m_flush_work
);
540 flush_work_sync(&mp
->m_flush_work
);
545 struct work_struct
*work
)
547 struct xfs_mount
*mp
= container_of(work
,
548 struct xfs_mount
, m_flush_work
);
550 xfs_sync_data(mp
, SYNC_TRYLOCK
);
551 xfs_sync_data(mp
, SYNC_TRYLOCK
| SYNC_WAIT
);
556 struct xfs_mount
*mp
)
558 INIT_WORK(&mp
->m_flush_work
, xfs_flush_worker
);
559 INIT_DELAYED_WORK(&mp
->m_sync_work
, xfs_sync_worker
);
560 INIT_DELAYED_WORK(&mp
->m_reclaim_work
, xfs_reclaim_worker
);
562 xfs_syncd_queue_sync(mp
);
563 xfs_syncd_queue_reclaim(mp
);
570 struct xfs_mount
*mp
)
572 cancel_delayed_work_sync(&mp
->m_sync_work
);
573 cancel_delayed_work_sync(&mp
->m_reclaim_work
);
574 cancel_work_sync(&mp
->m_flush_work
);
578 __xfs_inode_set_reclaim_tag(
579 struct xfs_perag
*pag
,
580 struct xfs_inode
*ip
)
582 radix_tree_tag_set(&pag
->pag_ici_root
,
583 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
584 XFS_ICI_RECLAIM_TAG
);
586 if (!pag
->pag_ici_reclaimable
) {
587 /* propagate the reclaim tag up into the perag radix tree */
588 spin_lock(&ip
->i_mount
->m_perag_lock
);
589 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
590 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
591 XFS_ICI_RECLAIM_TAG
);
592 spin_unlock(&ip
->i_mount
->m_perag_lock
);
594 /* schedule periodic background inode reclaim */
595 xfs_syncd_queue_reclaim(ip
->i_mount
);
597 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
600 pag
->pag_ici_reclaimable
++;
604 * We set the inode flag atomically with the radix tree tag.
605 * Once we get tag lookups on the radix tree, this inode flag
609 xfs_inode_set_reclaim_tag(
612 struct xfs_mount
*mp
= ip
->i_mount
;
613 struct xfs_perag
*pag
;
615 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
616 spin_lock(&pag
->pag_ici_lock
);
617 spin_lock(&ip
->i_flags_lock
);
618 __xfs_inode_set_reclaim_tag(pag
, ip
);
619 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
620 spin_unlock(&ip
->i_flags_lock
);
621 spin_unlock(&pag
->pag_ici_lock
);
626 __xfs_inode_clear_reclaim(
630 pag
->pag_ici_reclaimable
--;
631 if (!pag
->pag_ici_reclaimable
) {
632 /* clear the reclaim tag from the perag radix tree */
633 spin_lock(&ip
->i_mount
->m_perag_lock
);
634 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
635 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
636 XFS_ICI_RECLAIM_TAG
);
637 spin_unlock(&ip
->i_mount
->m_perag_lock
);
638 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
644 __xfs_inode_clear_reclaim_tag(
649 radix_tree_tag_clear(&pag
->pag_ici_root
,
650 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
651 __xfs_inode_clear_reclaim(pag
, ip
);
655 * Grab the inode for reclaim exclusively.
656 * Return 0 if we grabbed it, non-zero otherwise.
659 xfs_reclaim_inode_grab(
660 struct xfs_inode
*ip
,
663 ASSERT(rcu_read_lock_held());
665 /* quick check for stale RCU freed inode */
670 * do some unlocked checks first to avoid unnecessary lock traffic.
671 * The first is a flush lock check, the second is a already in reclaim
672 * check. Only do these checks if we are not going to block on locks.
674 if ((flags
& SYNC_TRYLOCK
) &&
675 (!ip
->i_flush
.done
|| __xfs_iflags_test(ip
, XFS_IRECLAIM
))) {
680 * The radix tree lock here protects a thread in xfs_iget from racing
681 * with us starting reclaim on the inode. Once we have the
682 * XFS_IRECLAIM flag set it will not touch us.
684 * Due to RCU lookup, we may find inodes that have been freed and only
685 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
686 * aren't candidates for reclaim at all, so we must check the
687 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
689 spin_lock(&ip
->i_flags_lock
);
690 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
691 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
692 /* not a reclaim candidate. */
693 spin_unlock(&ip
->i_flags_lock
);
696 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
697 spin_unlock(&ip
->i_flags_lock
);
702 * Inodes in different states need to be treated differently, and the return
703 * value of xfs_iflush is not sufficient to get this right. The following table
704 * lists the inode states and the reclaim actions necessary for non-blocking
708 * inode state iflush ret required action
709 * --------------- ---------- ---------------
711 * shutdown EIO unpin and reclaim
712 * clean, unpinned 0 reclaim
713 * stale, unpinned 0 reclaim
714 * clean, pinned(*) 0 requeue
715 * stale, pinned EAGAIN requeue
716 * dirty, delwri ok 0 requeue
717 * dirty, delwri blocked EAGAIN requeue
718 * dirty, sync flush 0 reclaim
720 * (*) dgc: I don't think the clean, pinned state is possible but it gets
721 * handled anyway given the order of checks implemented.
723 * As can be seen from the table, the return value of xfs_iflush() is not
724 * sufficient to correctly decide the reclaim action here. The checks in
725 * xfs_iflush() might look like duplicates, but they are not.
727 * Also, because we get the flush lock first, we know that any inode that has
728 * been flushed delwri has had the flush completed by the time we check that
729 * the inode is clean. The clean inode check needs to be done before flushing
730 * the inode delwri otherwise we would loop forever requeuing clean inodes as
731 * we cannot tell apart a successful delwri flush and a clean inode from the
732 * return value of xfs_iflush().
734 * Note that because the inode is flushed delayed write by background
735 * writeback, the flush lock may already be held here and waiting on it can
736 * result in very long latencies. Hence for sync reclaims, where we wait on the
737 * flush lock, the caller should push out delayed write inodes first before
738 * trying to reclaim them to minimise the amount of time spent waiting. For
739 * background relaim, we just requeue the inode for the next pass.
741 * Hence the order of actions after gaining the locks should be:
743 * shutdown => unpin and reclaim
744 * pinned, delwri => requeue
745 * pinned, sync => unpin
748 * dirty, delwri => flush and requeue
749 * dirty, sync => flush, wait and reclaim
753 struct xfs_inode
*ip
,
754 struct xfs_perag
*pag
,
761 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
762 if (!xfs_iflock_nowait(ip
)) {
763 if (!(sync_mode
& SYNC_WAIT
))
768 if (is_bad_inode(VFS_I(ip
)))
770 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
774 if (xfs_ipincount(ip
)) {
775 if (!(sync_mode
& SYNC_WAIT
)) {
781 if (xfs_iflags_test(ip
, XFS_ISTALE
))
783 if (xfs_inode_clean(ip
))
787 * Now we have an inode that needs flushing.
789 * We do a nonblocking flush here even if we are doing a SYNC_WAIT
790 * reclaim as we can deadlock with inode cluster removal.
791 * xfs_ifree_cluster() can lock the inode buffer before it locks the
792 * ip->i_lock, and we are doing the exact opposite here. As a result,
793 * doing a blocking xfs_itobp() to get the cluster buffer will result
794 * in an ABBA deadlock with xfs_ifree_cluster().
796 * As xfs_ifree_cluser() must gather all inodes that are active in the
797 * cache to mark them stale, if we hit this case we don't actually want
798 * to do IO here - we want the inode marked stale so we can simply
799 * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush,
800 * just unlock the inode, back off and try again. Hopefully the next
801 * pass through will see the stale flag set on the inode.
803 error
= xfs_iflush(ip
, SYNC_TRYLOCK
| sync_mode
);
804 if (sync_mode
& SYNC_WAIT
) {
805 if (error
== EAGAIN
) {
806 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
807 /* backoff longer than in xfs_ifree_cluster */
816 * When we have to flush an inode but don't have SYNC_WAIT set, we
817 * flush the inode out using a delwri buffer and wait for the next
818 * call into reclaim to find it in a clean state instead of waiting for
819 * it now. We also don't return errors here - if the error is transient
820 * then the next reclaim pass will flush the inode, and if the error
821 * is permanent then the next sync reclaim will reclaim the inode and
824 if (error
&& error
!= EAGAIN
&& !XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
825 xfs_warn(ip
->i_mount
,
826 "inode 0x%llx background reclaim flush failed with %d",
827 (long long)ip
->i_ino
, error
);
830 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
831 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
833 * We could return EAGAIN here to make reclaim rescan the inode tree in
834 * a short while. However, this just burns CPU time scanning the tree
835 * waiting for IO to complete and xfssyncd never goes back to the idle
836 * state. Instead, return 0 to let the next scheduled background reclaim
837 * attempt to reclaim the inode again.
843 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
845 XFS_STATS_INC(xs_ig_reclaims
);
847 * Remove the inode from the per-AG radix tree.
849 * Because radix_tree_delete won't complain even if the item was never
850 * added to the tree assert that it's been there before to catch
851 * problems with the inode life time early on.
853 spin_lock(&pag
->pag_ici_lock
);
854 if (!radix_tree_delete(&pag
->pag_ici_root
,
855 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
857 __xfs_inode_clear_reclaim(pag
, ip
);
858 spin_unlock(&pag
->pag_ici_lock
);
861 * Here we do an (almost) spurious inode lock in order to coordinate
862 * with inode cache radix tree lookups. This is because the lookup
863 * can reference the inodes in the cache without taking references.
865 * We make that OK here by ensuring that we wait until the inode is
866 * unlocked after the lookup before we go ahead and free it. We get
867 * both the ilock and the iolock because the code may need to drop the
868 * ilock one but will still hold the iolock.
870 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
872 xfs_iunlock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
880 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
881 * corrupted, we still want to try to reclaim all the inodes. If we don't,
882 * then a shut down during filesystem unmount reclaim walk leak all the
883 * unreclaimed inodes.
886 xfs_reclaim_inodes_ag(
887 struct xfs_mount
*mp
,
891 struct xfs_perag
*pag
;
895 int trylock
= flags
& SYNC_TRYLOCK
;
901 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
902 unsigned long first_index
= 0;
906 ag
= pag
->pag_agno
+ 1;
909 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
914 first_index
= pag
->pag_ici_reclaim_cursor
;
916 mutex_lock(&pag
->pag_ici_reclaim_lock
);
919 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
923 nr_found
= radix_tree_gang_lookup_tag(
925 (void **)batch
, first_index
,
927 XFS_ICI_RECLAIM_TAG
);
935 * Grab the inodes before we drop the lock. if we found
936 * nothing, nr == 0 and the loop will be skipped.
938 for (i
= 0; i
< nr_found
; i
++) {
939 struct xfs_inode
*ip
= batch
[i
];
941 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
945 * Update the index for the next lookup. Catch
946 * overflows into the next AG range which can
947 * occur if we have inodes in the last block of
948 * the AG and we are currently pointing to the
951 * Because we may see inodes that are from the
952 * wrong AG due to RCU freeing and
953 * reallocation, only update the index if it
954 * lies in this AG. It was a race that lead us
955 * to see this inode, so another lookup from
956 * the same index will not find it again.
958 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
961 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
962 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
966 /* unlock now we've grabbed the inodes. */
969 for (i
= 0; i
< nr_found
; i
++) {
972 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
973 if (error
&& last_error
!= EFSCORRUPTED
)
977 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
979 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
981 if (trylock
&& !done
)
982 pag
->pag_ici_reclaim_cursor
= first_index
;
984 pag
->pag_ici_reclaim_cursor
= 0;
985 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
990 * if we skipped any AG, and we still have scan count remaining, do
991 * another pass this time using blocking reclaim semantics (i.e
992 * waiting on the reclaim locks and ignoring the reclaim cursors). This
993 * ensure that when we get more reclaimers than AGs we block rather
994 * than spin trying to execute reclaim.
996 if (trylock
&& skipped
&& *nr_to_scan
> 0) {
1000 return XFS_ERROR(last_error
);
1008 int nr_to_scan
= INT_MAX
;
1010 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1014 * Inode cache shrinker.
1016 * When called we make sure that there is a background (fast) inode reclaim in
1017 * progress, while we will throttle the speed of reclaim via doiing synchronous
1018 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1019 * them to be cleaned, which we hope will not be very long due to the
1020 * background walker having already kicked the IO off on those dirty inodes.
1023 xfs_reclaim_inode_shrink(
1024 struct shrinker
*shrink
,
1028 struct xfs_mount
*mp
;
1029 struct xfs_perag
*pag
;
1033 mp
= container_of(shrink
, struct xfs_mount
, m_inode_shrink
);
1035 /* kick background reclaimer and push the AIL */
1036 xfs_syncd_queue_reclaim(mp
);
1037 xfs_ail_push_all(mp
->m_ail
);
1039 if (!(gfp_mask
& __GFP_FS
))
1042 xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
,
1044 /* terminate if we don't exhaust the scan */
1051 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1052 ag
= pag
->pag_agno
+ 1;
1053 reclaimable
+= pag
->pag_ici_reclaimable
;
1060 xfs_inode_shrinker_register(
1061 struct xfs_mount
*mp
)
1063 mp
->m_inode_shrink
.shrink
= xfs_reclaim_inode_shrink
;
1064 mp
->m_inode_shrink
.seeks
= DEFAULT_SEEKS
;
1065 register_shrinker(&mp
->m_inode_shrink
);
1069 xfs_inode_shrinker_unregister(
1070 struct xfs_mount
*mp
)
1072 unregister_shrinker(&mp
->m_inode_shrink
);