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"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_dinode.h"
31 #include "xfs_error.h"
32 #include "xfs_filestream.h"
33 #include "xfs_vnodeops.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_quota.h"
36 #include "xfs_trace.h"
37 #include "xfs_fsops.h"
39 #include <linux/kthread.h>
40 #include <linux/freezer.h>
43 * The inode lookup is done in batches to keep the amount of lock traffic and
44 * radix tree lookups to a minimum. The batch size is a trade off between
45 * lookup reduction and stack usage. This is in the reclaim path, so we can't
48 #define XFS_LOOKUP_BATCH 32
51 xfs_inode_ag_walk_grab(
54 struct inode
*inode
= VFS_I(ip
);
56 /* nothing to sync during shutdown */
57 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
60 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
61 if (xfs_iflags_test(ip
, XFS_INEW
| XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
64 /* If we can't grab the inode, it must on it's way to reclaim. */
68 if (is_bad_inode(inode
)) {
80 struct xfs_perag
*pag
,
81 int (*execute
)(struct xfs_inode
*ip
,
82 struct xfs_perag
*pag
, int flags
),
97 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
101 read_lock(&pag
->pag_ici_lock
);
102 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
103 (void **)batch
, first_index
,
106 read_unlock(&pag
->pag_ici_lock
);
111 * Grab the inodes before we drop the lock. if we found
112 * nothing, nr == 0 and the loop will be skipped.
114 for (i
= 0; i
< nr_found
; i
++) {
115 struct xfs_inode
*ip
= batch
[i
];
117 if (done
|| xfs_inode_ag_walk_grab(ip
))
121 * Update the index for the next lookup. Catch overflows
122 * into the next AG range which can occur if we have inodes
123 * in the last block of the AG and we are currently
124 * pointing to the last inode.
126 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
127 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
131 /* unlock now we've grabbed the inodes. */
132 read_unlock(&pag
->pag_ici_lock
);
134 for (i
= 0; i
< nr_found
; i
++) {
137 error
= execute(batch
[i
], pag
, flags
);
139 if (error
== EAGAIN
) {
143 if (error
&& last_error
!= EFSCORRUPTED
)
147 /* bail out if the filesystem is corrupted. */
148 if (error
== EFSCORRUPTED
)
151 } while (nr_found
&& !done
);
161 xfs_inode_ag_iterator(
162 struct xfs_mount
*mp
,
163 int (*execute
)(struct xfs_inode
*ip
,
164 struct xfs_perag
*pag
, int flags
),
167 struct xfs_perag
*pag
;
173 while ((pag
= xfs_perag_get(mp
, ag
))) {
174 ag
= pag
->pag_agno
+ 1;
175 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
);
179 if (error
== EFSCORRUPTED
)
183 return XFS_ERROR(last_error
);
188 struct xfs_inode
*ip
,
189 struct xfs_perag
*pag
,
192 struct inode
*inode
= VFS_I(ip
);
193 struct address_space
*mapping
= inode
->i_mapping
;
196 if (!mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
199 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
)) {
200 if (flags
& SYNC_TRYLOCK
)
202 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
205 error
= xfs_flush_pages(ip
, 0, -1, (flags
& SYNC_WAIT
) ?
206 0 : XBF_ASYNC
, FI_NONE
);
207 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
210 if (flags
& SYNC_WAIT
)
217 struct xfs_inode
*ip
,
218 struct xfs_perag
*pag
,
223 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
224 if (xfs_inode_clean(ip
))
226 if (!xfs_iflock_nowait(ip
)) {
227 if (!(flags
& SYNC_WAIT
))
232 if (xfs_inode_clean(ip
)) {
237 error
= xfs_iflush(ip
, flags
);
240 xfs_iunlock(ip
, XFS_ILOCK_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);
265 * Write out inode metadata (attributes) for the whole filesystem.
269 struct xfs_mount
*mp
,
272 ASSERT((flags
& ~SYNC_WAIT
) == 0);
274 return xfs_inode_ag_iterator(mp
, xfs_sync_inode_attr
, flags
);
279 struct xfs_mount
*mp
)
284 * If the buffer is pinned then push on the log so we won't get stuck
285 * waiting in the write for someone, maybe ourselves, to flush the log.
287 * Even though we just pushed the log above, we did not have the
288 * superblock buffer locked at that point so it can become pinned in
289 * between there and here.
291 bp
= xfs_getsb(mp
, 0);
292 if (XFS_BUF_ISPINNED(bp
))
293 xfs_log_force(mp
, 0);
295 return xfs_bwrite(mp
, bp
);
299 * When remounting a filesystem read-only or freezing the filesystem, we have
300 * two phases to execute. This first phase is syncing the data before we
301 * quiesce the filesystem, and the second is flushing all the inodes out after
302 * we've waited for all the transactions created by the first phase to
303 * complete. The second phase ensures that the inodes are written to their
304 * location on disk rather than just existing in transactions in the log. This
305 * means after a quiesce there is no log replay required to write the inodes to
306 * disk (this is the main difference between a sync and a quiesce).
309 * First stage of freeze - no writers will make progress now we are here,
310 * so we flush delwri and delalloc buffers here, then wait for all I/O to
311 * complete. Data is frozen at that point. Metadata is not frozen,
312 * transactions can still occur here so don't bother flushing the buftarg
313 * because it'll just get dirty again.
317 struct xfs_mount
*mp
)
319 int error
, error2
= 0;
321 /* push non-blocking */
322 xfs_sync_data(mp
, 0);
323 xfs_qm_sync(mp
, SYNC_TRYLOCK
);
325 /* push and block till complete */
326 xfs_sync_data(mp
, SYNC_WAIT
);
327 xfs_qm_sync(mp
, SYNC_WAIT
);
329 /* write superblock and hoover up shutdown errors */
330 error
= xfs_sync_fsdata(mp
);
332 /* make sure all delwri buffers are written out */
333 xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
335 /* mark the log as covered if needed */
336 if (xfs_log_need_covered(mp
))
337 error2
= xfs_fs_log_dummy(mp
, SYNC_WAIT
);
339 /* flush data-only devices */
340 if (mp
->m_rtdev_targp
)
341 XFS_bflush(mp
->m_rtdev_targp
);
343 return error
? error
: error2
;
348 struct xfs_mount
*mp
)
350 int count
= 0, pincount
;
352 xfs_reclaim_inodes(mp
, 0);
353 xfs_flush_buftarg(mp
->m_ddev_targp
, 0);
356 * This loop must run at least twice. The first instance of the loop
357 * will flush most meta data but that will generate more meta data
358 * (typically directory updates). Which then must be flushed and
359 * logged before we can write the unmount record. We also so sync
360 * reclaim of inodes to catch any that the above delwri flush skipped.
363 xfs_reclaim_inodes(mp
, SYNC_WAIT
);
364 xfs_sync_attr(mp
, SYNC_WAIT
);
365 pincount
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
374 * Second stage of a quiesce. The data is already synced, now we have to take
375 * care of the metadata. New transactions are already blocked, so we need to
376 * wait for any remaining transactions to drain out before proceding.
380 struct xfs_mount
*mp
)
384 /* wait for all modifications to complete */
385 while (atomic_read(&mp
->m_active_trans
) > 0)
388 /* flush inodes and push all remaining buffers out to disk */
392 * Just warn here till VFS can correctly support
393 * read-only remount without racing.
395 WARN_ON(atomic_read(&mp
->m_active_trans
) != 0);
397 /* Push the superblock and write an unmount record */
398 error
= xfs_log_sbcount(mp
, 1);
400 xfs_fs_cmn_err(CE_WARN
, mp
,
401 "xfs_attr_quiesce: failed to log sb changes. "
402 "Frozen image may not be consistent.");
403 xfs_log_unmount_write(mp
);
404 xfs_unmountfs_writesb(mp
);
408 * Enqueue a work item to be picked up by the vfs xfssyncd thread.
409 * Doing this has two advantages:
410 * - It saves on stack space, which is tight in certain situations
411 * - It can be used (with care) as a mechanism to avoid deadlocks.
412 * Flushing while allocating in a full filesystem requires both.
415 xfs_syncd_queue_work(
416 struct xfs_mount
*mp
,
418 void (*syncer
)(struct xfs_mount
*, void *),
419 struct completion
*completion
)
421 struct xfs_sync_work
*work
;
423 work
= kmem_alloc(sizeof(struct xfs_sync_work
), KM_SLEEP
);
424 INIT_LIST_HEAD(&work
->w_list
);
425 work
->w_syncer
= syncer
;
428 work
->w_completion
= completion
;
429 spin_lock(&mp
->m_sync_lock
);
430 list_add_tail(&work
->w_list
, &mp
->m_sync_list
);
431 spin_unlock(&mp
->m_sync_lock
);
432 wake_up_process(mp
->m_sync_task
);
436 * Flush delayed allocate data, attempting to free up reserved space
437 * from existing allocations. At this point a new allocation attempt
438 * has failed with ENOSPC and we are in the process of scratching our
439 * heads, looking about for more room...
442 xfs_flush_inodes_work(
443 struct xfs_mount
*mp
,
446 struct inode
*inode
= arg
;
447 xfs_sync_data(mp
, SYNC_TRYLOCK
);
448 xfs_sync_data(mp
, SYNC_TRYLOCK
| SYNC_WAIT
);
456 struct inode
*inode
= VFS_I(ip
);
457 DECLARE_COMPLETION_ONSTACK(completion
);
460 xfs_syncd_queue_work(ip
->i_mount
, inode
, xfs_flush_inodes_work
, &completion
);
461 wait_for_completion(&completion
);
462 xfs_log_force(ip
->i_mount
, XFS_LOG_SYNC
);
466 * Every sync period we need to unpin all items, reclaim inodes and sync
467 * disk quotas. We might need to cover the log to indicate that the
468 * filesystem is idle and not frozen.
472 struct xfs_mount
*mp
,
477 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
478 xfs_log_force(mp
, 0);
479 xfs_reclaim_inodes(mp
, 0);
480 /* dgc: errors ignored here */
481 error
= xfs_qm_sync(mp
, SYNC_TRYLOCK
);
482 if (mp
->m_super
->s_frozen
== SB_UNFROZEN
&&
483 xfs_log_need_covered(mp
))
484 error
= xfs_fs_log_dummy(mp
, 0);
487 wake_up(&mp
->m_wait_single_sync_task
);
494 struct xfs_mount
*mp
= arg
;
496 xfs_sync_work_t
*work
, *n
;
500 timeleft
= xfs_syncd_centisecs
* msecs_to_jiffies(10);
502 if (list_empty(&mp
->m_sync_list
))
503 timeleft
= schedule_timeout_interruptible(timeleft
);
506 if (kthread_should_stop() && list_empty(&mp
->m_sync_list
))
509 spin_lock(&mp
->m_sync_lock
);
511 * We can get woken by laptop mode, to do a sync -
512 * that's the (only!) case where the list would be
513 * empty with time remaining.
515 if (!timeleft
|| list_empty(&mp
->m_sync_list
)) {
517 timeleft
= xfs_syncd_centisecs
*
518 msecs_to_jiffies(10);
519 INIT_LIST_HEAD(&mp
->m_sync_work
.w_list
);
520 list_add_tail(&mp
->m_sync_work
.w_list
,
523 list_splice_init(&mp
->m_sync_list
, &tmp
);
524 spin_unlock(&mp
->m_sync_lock
);
526 list_for_each_entry_safe(work
, n
, &tmp
, w_list
) {
527 (*work
->w_syncer
)(mp
, work
->w_data
);
528 list_del(&work
->w_list
);
529 if (work
== &mp
->m_sync_work
)
531 if (work
->w_completion
)
532 complete(work
->w_completion
);
542 struct xfs_mount
*mp
)
544 mp
->m_sync_work
.w_syncer
= xfs_sync_worker
;
545 mp
->m_sync_work
.w_mount
= mp
;
546 mp
->m_sync_work
.w_completion
= NULL
;
547 mp
->m_sync_task
= kthread_run(xfssyncd
, mp
, "xfssyncd/%s", mp
->m_fsname
);
548 if (IS_ERR(mp
->m_sync_task
))
549 return -PTR_ERR(mp
->m_sync_task
);
555 struct xfs_mount
*mp
)
557 kthread_stop(mp
->m_sync_task
);
561 __xfs_inode_set_reclaim_tag(
562 struct xfs_perag
*pag
,
563 struct xfs_inode
*ip
)
565 radix_tree_tag_set(&pag
->pag_ici_root
,
566 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
567 XFS_ICI_RECLAIM_TAG
);
569 if (!pag
->pag_ici_reclaimable
) {
570 /* propagate the reclaim tag up into the perag radix tree */
571 spin_lock(&ip
->i_mount
->m_perag_lock
);
572 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
573 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
574 XFS_ICI_RECLAIM_TAG
);
575 spin_unlock(&ip
->i_mount
->m_perag_lock
);
576 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
579 pag
->pag_ici_reclaimable
++;
583 * We set the inode flag atomically with the radix tree tag.
584 * Once we get tag lookups on the radix tree, this inode flag
588 xfs_inode_set_reclaim_tag(
591 struct xfs_mount
*mp
= ip
->i_mount
;
592 struct xfs_perag
*pag
;
594 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
595 write_lock(&pag
->pag_ici_lock
);
596 spin_lock(&ip
->i_flags_lock
);
597 __xfs_inode_set_reclaim_tag(pag
, ip
);
598 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
599 spin_unlock(&ip
->i_flags_lock
);
600 write_unlock(&pag
->pag_ici_lock
);
605 __xfs_inode_clear_reclaim(
609 pag
->pag_ici_reclaimable
--;
610 if (!pag
->pag_ici_reclaimable
) {
611 /* clear the reclaim tag from the perag radix tree */
612 spin_lock(&ip
->i_mount
->m_perag_lock
);
613 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
614 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
615 XFS_ICI_RECLAIM_TAG
);
616 spin_unlock(&ip
->i_mount
->m_perag_lock
);
617 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
623 __xfs_inode_clear_reclaim_tag(
628 radix_tree_tag_clear(&pag
->pag_ici_root
,
629 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
630 __xfs_inode_clear_reclaim(pag
, ip
);
634 * Grab the inode for reclaim exclusively.
635 * Return 0 if we grabbed it, non-zero otherwise.
638 xfs_reclaim_inode_grab(
639 struct xfs_inode
*ip
,
644 * do some unlocked checks first to avoid unnecceary lock traffic.
645 * The first is a flush lock check, the second is a already in reclaim
646 * check. Only do these checks if we are not going to block on locks.
648 if ((flags
& SYNC_TRYLOCK
) &&
649 (!ip
->i_flush
.done
|| __xfs_iflags_test(ip
, XFS_IRECLAIM
))) {
654 * The radix tree lock here protects a thread in xfs_iget from racing
655 * with us starting reclaim on the inode. Once we have the
656 * XFS_IRECLAIM flag set it will not touch us.
658 spin_lock(&ip
->i_flags_lock
);
659 ASSERT_ALWAYS(__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
));
660 if (__xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
661 /* ignore as it is already under reclaim */
662 spin_unlock(&ip
->i_flags_lock
);
665 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
666 spin_unlock(&ip
->i_flags_lock
);
671 * Inodes in different states need to be treated differently, and the return
672 * value of xfs_iflush is not sufficient to get this right. The following table
673 * lists the inode states and the reclaim actions necessary for non-blocking
677 * inode state iflush ret required action
678 * --------------- ---------- ---------------
680 * shutdown EIO unpin and reclaim
681 * clean, unpinned 0 reclaim
682 * stale, unpinned 0 reclaim
683 * clean, pinned(*) 0 requeue
684 * stale, pinned EAGAIN requeue
685 * dirty, delwri ok 0 requeue
686 * dirty, delwri blocked EAGAIN requeue
687 * dirty, sync flush 0 reclaim
689 * (*) dgc: I don't think the clean, pinned state is possible but it gets
690 * handled anyway given the order of checks implemented.
692 * As can be seen from the table, the return value of xfs_iflush() is not
693 * sufficient to correctly decide the reclaim action here. The checks in
694 * xfs_iflush() might look like duplicates, but they are not.
696 * Also, because we get the flush lock first, we know that any inode that has
697 * been flushed delwri has had the flush completed by the time we check that
698 * the inode is clean. The clean inode check needs to be done before flushing
699 * the inode delwri otherwise we would loop forever requeuing clean inodes as
700 * we cannot tell apart a successful delwri flush and a clean inode from the
701 * return value of xfs_iflush().
703 * Note that because the inode is flushed delayed write by background
704 * writeback, the flush lock may already be held here and waiting on it can
705 * result in very long latencies. Hence for sync reclaims, where we wait on the
706 * flush lock, the caller should push out delayed write inodes first before
707 * trying to reclaim them to minimise the amount of time spent waiting. For
708 * background relaim, we just requeue the inode for the next pass.
710 * Hence the order of actions after gaining the locks should be:
712 * shutdown => unpin and reclaim
713 * pinned, delwri => requeue
714 * pinned, sync => unpin
717 * dirty, delwri => flush and requeue
718 * dirty, sync => flush, wait and reclaim
722 struct xfs_inode
*ip
,
723 struct xfs_perag
*pag
,
728 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
729 if (!xfs_iflock_nowait(ip
)) {
730 if (!(sync_mode
& SYNC_WAIT
))
735 if (is_bad_inode(VFS_I(ip
)))
737 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
741 if (xfs_ipincount(ip
)) {
742 if (!(sync_mode
& SYNC_WAIT
)) {
748 if (xfs_iflags_test(ip
, XFS_ISTALE
))
750 if (xfs_inode_clean(ip
))
753 /* Now we have an inode that needs flushing */
754 error
= xfs_iflush(ip
, sync_mode
);
755 if (sync_mode
& SYNC_WAIT
) {
761 * When we have to flush an inode but don't have SYNC_WAIT set, we
762 * flush the inode out using a delwri buffer and wait for the next
763 * call into reclaim to find it in a clean state instead of waiting for
764 * it now. We also don't return errors here - if the error is transient
765 * then the next reclaim pass will flush the inode, and if the error
766 * is permanent then the next sync reclaim will reclaim the inode and
769 if (error
&& error
!= EAGAIN
&& !XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
770 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
771 "inode 0x%llx background reclaim flush failed with %d",
772 (long long)ip
->i_ino
, error
);
775 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
776 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
778 * We could return EAGAIN here to make reclaim rescan the inode tree in
779 * a short while. However, this just burns CPU time scanning the tree
780 * waiting for IO to complete and xfssyncd never goes back to the idle
781 * state. Instead, return 0 to let the next scheduled background reclaim
782 * attempt to reclaim the inode again.
788 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
790 XFS_STATS_INC(xs_ig_reclaims
);
792 * Remove the inode from the per-AG radix tree.
794 * Because radix_tree_delete won't complain even if the item was never
795 * added to the tree assert that it's been there before to catch
796 * problems with the inode life time early on.
798 write_lock(&pag
->pag_ici_lock
);
799 if (!radix_tree_delete(&pag
->pag_ici_root
,
800 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
802 __xfs_inode_clear_reclaim(pag
, ip
);
803 write_unlock(&pag
->pag_ici_lock
);
806 * Here we do an (almost) spurious inode lock in order to coordinate
807 * with inode cache radix tree lookups. This is because the lookup
808 * can reference the inodes in the cache without taking references.
810 * We make that OK here by ensuring that we wait until the inode is
811 * unlocked after the lookup before we go ahead and free it. We get
812 * both the ilock and the iolock because the code may need to drop the
813 * ilock one but will still hold the iolock.
815 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
817 xfs_iunlock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
825 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
826 * corrupted, we still want to try to reclaim all the inodes. If we don't,
827 * then a shut down during filesystem unmount reclaim walk leak all the
828 * unreclaimed inodes.
831 xfs_reclaim_inodes_ag(
832 struct xfs_mount
*mp
,
836 struct xfs_perag
*pag
;
840 int trylock
= flags
& SYNC_TRYLOCK
;
846 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
847 unsigned long first_index
= 0;
851 ag
= pag
->pag_agno
+ 1;
854 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
859 first_index
= pag
->pag_ici_reclaim_cursor
;
861 mutex_lock(&pag
->pag_ici_reclaim_lock
);
864 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
867 write_lock(&pag
->pag_ici_lock
);
868 nr_found
= radix_tree_gang_lookup_tag(
870 (void **)batch
, first_index
,
872 XFS_ICI_RECLAIM_TAG
);
874 write_unlock(&pag
->pag_ici_lock
);
879 * Grab the inodes before we drop the lock. if we found
880 * nothing, nr == 0 and the loop will be skipped.
882 for (i
= 0; i
< nr_found
; i
++) {
883 struct xfs_inode
*ip
= batch
[i
];
885 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
889 * Update the index for the next lookup. Catch
890 * overflows into the next AG range which can
891 * occur if we have inodes in the last block of
892 * the AG and we are currently pointing to the
895 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
896 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
900 /* unlock now we've grabbed the inodes. */
901 write_unlock(&pag
->pag_ici_lock
);
903 for (i
= 0; i
< nr_found
; i
++) {
906 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
907 if (error
&& last_error
!= EFSCORRUPTED
)
911 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
913 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
915 if (trylock
&& !done
)
916 pag
->pag_ici_reclaim_cursor
= first_index
;
918 pag
->pag_ici_reclaim_cursor
= 0;
919 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
924 * if we skipped any AG, and we still have scan count remaining, do
925 * another pass this time using blocking reclaim semantics (i.e
926 * waiting on the reclaim locks and ignoring the reclaim cursors). This
927 * ensure that when we get more reclaimers than AGs we block rather
928 * than spin trying to execute reclaim.
930 if (trylock
&& skipped
&& *nr_to_scan
> 0) {
934 return XFS_ERROR(last_error
);
942 int nr_to_scan
= INT_MAX
;
944 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
948 * Shrinker infrastructure.
951 xfs_reclaim_inode_shrink(
952 struct shrinker
*shrink
,
956 struct xfs_mount
*mp
;
957 struct xfs_perag
*pag
;
961 mp
= container_of(shrink
, struct xfs_mount
, m_inode_shrink
);
963 if (!(gfp_mask
& __GFP_FS
))
966 xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
, &nr_to_scan
);
967 /* terminate if we don't exhaust the scan */
974 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
975 ag
= pag
->pag_agno
+ 1;
976 reclaimable
+= pag
->pag_ici_reclaimable
;
983 xfs_inode_shrinker_register(
984 struct xfs_mount
*mp
)
986 mp
->m_inode_shrink
.shrink
= xfs_reclaim_inode_shrink
;
987 mp
->m_inode_shrink
.seeks
= DEFAULT_SEEKS
;
988 register_shrinker(&mp
->m_inode_shrink
);
992 xfs_inode_shrinker_unregister(
993 struct xfs_mount
*mp
)
995 unregister_shrinker(&mp
->m_inode_shrink
);