2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_dinode.h"
32 #include "xfs_error.h"
33 #include "xfs_filestream.h"
34 #include "xfs_vnodeops.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_quota.h"
37 #include "xfs_trace.h"
38 #include "xfs_fsops.h"
40 #include <linux/kthread.h>
41 #include <linux/freezer.h>
43 struct workqueue_struct *xfs_syncd_wq; /* sync workqueue */
46 * The inode lookup is done in batches to keep the amount of lock traffic and
47 * radix tree lookups to a minimum. The batch size is a trade off between
48 * lookup reduction and stack usage. This is in the reclaim path, so we can't
51 #define XFS_LOOKUP_BATCH 32
54 xfs_inode_ag_walk_grab(
57 struct inode *inode = VFS_I(ip);
59 ASSERT(rcu_read_lock_held());
62 * check for stale RCU freed inode
64 * If the inode has been reallocated, it doesn't matter if it's not in
65 * the AG we are walking - we are walking for writeback, so if it
66 * passes all the "valid inode" checks and is dirty, then we'll write
67 * it back anyway. If it has been reallocated and still being
68 * initialised, the XFS_INEW check below will catch it.
70 spin_lock(&ip->i_flags_lock);
72 goto out_unlock_noent;
74 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
75 if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM))
76 goto out_unlock_noent;
77 spin_unlock(&ip->i_flags_lock);
79 /* nothing to sync during shutdown */
80 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
83 /* If we can't grab the inode, it must on it's way to reclaim. */
87 if (is_bad_inode(inode)) {
96 spin_unlock(&ip->i_flags_lock);
102 struct xfs_mount *mp,
103 struct xfs_perag *pag,
104 int (*execute)(struct xfs_inode *ip,
105 struct xfs_perag *pag, int flags),
108 uint32_t first_index;
120 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
125 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
126 (void **)batch, first_index,
134 * Grab the inodes before we drop the lock. if we found
135 * nothing, nr == 0 and the loop will be skipped.
137 for (i = 0; i < nr_found; i++) {
138 struct xfs_inode *ip = batch[i];
140 if (done || xfs_inode_ag_walk_grab(ip))
144 * Update the index for the next lookup. Catch
145 * overflows into the next AG range which can occur if
146 * we have inodes in the last block of the AG and we
147 * are currently pointing to the last inode.
149 * Because we may see inodes that are from the wrong AG
150 * due to RCU freeing and reallocation, only update the
151 * index if it lies in this AG. It was a race that lead
152 * us to see this inode, so another lookup from the
153 * same index will not find it again.
155 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
157 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
158 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
162 /* unlock now we've grabbed the inodes. */
165 for (i = 0; i < nr_found; i++) {
168 error = execute(batch[i], pag, flags);
170 if (error == EAGAIN) {
174 if (error && last_error != EFSCORRUPTED)
178 /* bail out if the filesystem is corrupted. */
179 if (error == EFSCORRUPTED)
184 } while (nr_found && !done);
194 xfs_inode_ag_iterator(
195 struct xfs_mount *mp,
196 int (*execute)(struct xfs_inode *ip,
197 struct xfs_perag *pag, int flags),
200 struct xfs_perag *pag;
206 while ((pag = xfs_perag_get(mp, ag))) {
207 ag = pag->pag_agno + 1;
208 error = xfs_inode_ag_walk(mp, pag, execute, flags);
212 if (error == EFSCORRUPTED)
216 return XFS_ERROR(last_error);
221 struct xfs_inode *ip,
222 struct xfs_perag *pag,
225 struct inode *inode = VFS_I(ip);
226 struct address_space *mapping = inode->i_mapping;
229 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
232 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) {
233 if (flags & SYNC_TRYLOCK)
235 xfs_ilock(ip, XFS_IOLOCK_SHARED);
238 error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ?
239 0 : XBF_ASYNC, FI_NONE);
240 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
246 struct xfs_inode *ip,
247 struct xfs_perag *pag,
252 xfs_ilock(ip, XFS_ILOCK_SHARED);
253 if (xfs_inode_clean(ip))
255 if (!xfs_iflock_nowait(ip)) {
256 if (!(flags & SYNC_WAIT))
261 if (xfs_inode_clean(ip)) {
266 error = xfs_iflush(ip, flags);
269 * We don't want to try again on non-blocking flushes that can't run
270 * again immediately. If an inode really must be written, then that's
271 * what the SYNC_WAIT flag is for.
273 if (error == EAGAIN) {
274 ASSERT(!(flags & SYNC_WAIT));
279 xfs_iunlock(ip, XFS_ILOCK_SHARED);
284 * Write out pagecache data for the whole filesystem.
288 struct xfs_mount *mp,
293 ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0);
295 error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags);
297 return XFS_ERROR(error);
299 xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0);
304 * Write out inode metadata (attributes) for the whole filesystem.
308 struct xfs_mount *mp,
311 ASSERT((flags & ~SYNC_WAIT) == 0);
313 return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags);
318 struct xfs_mount *mp)
324 * If the buffer is pinned then push on the log so we won't get stuck
325 * waiting in the write for someone, maybe ourselves, to flush the log.
327 * Even though we just pushed the log above, we did not have the
328 * superblock buffer locked at that point so it can become pinned in
329 * between there and here.
331 bp = xfs_getsb(mp, 0);
332 if (xfs_buf_ispinned(bp))
333 xfs_log_force(mp, 0);
334 error = xfs_bwrite(bp);
340 * When remounting a filesystem read-only or freezing the filesystem, we have
341 * two phases to execute. This first phase is syncing the data before we
342 * quiesce the filesystem, and the second is flushing all the inodes out after
343 * we've waited for all the transactions created by the first phase to
344 * complete. The second phase ensures that the inodes are written to their
345 * location on disk rather than just existing in transactions in the log. This
346 * means after a quiesce there is no log replay required to write the inodes to
347 * disk (this is the main difference between a sync and a quiesce).
350 * First stage of freeze - no writers will make progress now we are here,
351 * so we flush delwri and delalloc buffers here, then wait for all I/O to
352 * complete. Data is frozen at that point. Metadata is not frozen,
353 * transactions can still occur here so don't bother flushing the buftarg
354 * because it'll just get dirty again.
358 struct xfs_mount *mp)
360 int error, error2 = 0;
362 xfs_qm_sync(mp, SYNC_TRYLOCK);
363 xfs_qm_sync(mp, SYNC_WAIT);
365 /* force out the newly dirtied log buffers */
366 xfs_log_force(mp, XFS_LOG_SYNC);
368 /* write superblock and hoover up shutdown errors */
369 error = xfs_sync_fsdata(mp);
371 /* make sure all delwri buffers are written out */
372 xfs_flush_buftarg(mp->m_ddev_targp, 1);
374 /* mark the log as covered if needed */
375 if (xfs_log_need_covered(mp))
376 error2 = xfs_fs_log_dummy(mp);
378 /* flush data-only devices */
379 if (mp->m_rtdev_targp)
380 xfs_flush_buftarg(mp->m_rtdev_targp, 1);
382 return error ? error : error2;
387 struct xfs_mount *mp)
389 int count = 0, pincount;
391 xfs_reclaim_inodes(mp, 0);
392 xfs_flush_buftarg(mp->m_ddev_targp, 0);
395 * This loop must run at least twice. The first instance of the loop
396 * will flush most meta data but that will generate more meta data
397 * (typically directory updates). Which then must be flushed and
398 * logged before we can write the unmount record. We also so sync
399 * reclaim of inodes to catch any that the above delwri flush skipped.
402 xfs_reclaim_inodes(mp, SYNC_WAIT);
403 xfs_sync_attr(mp, SYNC_WAIT);
404 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
413 * Second stage of a quiesce. The data is already synced, now we have to take
414 * care of the metadata. New transactions are already blocked, so we need to
415 * wait for any remaining transactions to drain out before proceeding.
419 struct xfs_mount *mp)
423 /* wait for all modifications to complete */
424 while (atomic_read(&mp->m_active_trans) > 0)
427 /* flush inodes and push all remaining buffers out to disk */
431 * Just warn here till VFS can correctly support
432 * read-only remount without racing.
434 WARN_ON(atomic_read(&mp->m_active_trans) != 0);
436 /* Push the superblock and write an unmount record */
437 error = xfs_log_sbcount(mp);
439 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
440 "Frozen image may not be consistent.");
441 xfs_log_unmount_write(mp);
442 xfs_unmountfs_writesb(mp);
446 xfs_syncd_queue_sync(
447 struct xfs_mount *mp)
449 queue_delayed_work(xfs_syncd_wq, &mp->m_sync_work,
450 msecs_to_jiffies(xfs_syncd_centisecs * 10));
454 * Every sync period we need to unpin all items, reclaim inodes and sync
455 * disk quotas. We might need to cover the log to indicate that the
456 * filesystem is idle and not frozen.
460 struct work_struct *work)
462 struct xfs_mount *mp = container_of(to_delayed_work(work),
463 struct xfs_mount, m_sync_work);
466 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
467 /* dgc: errors ignored here */
468 if (mp->m_super->s_frozen == SB_UNFROZEN &&
469 xfs_log_need_covered(mp))
470 error = xfs_fs_log_dummy(mp);
472 xfs_log_force(mp, 0);
473 error = xfs_qm_sync(mp, SYNC_TRYLOCK);
475 /* start pushing all the metadata that is currently dirty */
476 xfs_ail_push_all(mp->m_ail);
479 /* queue us up again */
480 xfs_syncd_queue_sync(mp);
484 * Queue a new inode reclaim pass if there are reclaimable inodes and there
485 * isn't a reclaim pass already in progress. By default it runs every 5s based
486 * on the xfs syncd work default of 30s. Perhaps this should have it's own
487 * tunable, but that can be done if this method proves to be ineffective or too
491 xfs_syncd_queue_reclaim(
492 struct xfs_mount *mp)
496 * We can have inodes enter reclaim after we've shut down the syncd
497 * workqueue during unmount, so don't allow reclaim work to be queued
500 if (!(mp->m_super->s_flags & MS_ACTIVE))
504 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
505 queue_delayed_work(xfs_syncd_wq, &mp->m_reclaim_work,
506 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
512 * This is a fast pass over the inode cache to try to get reclaim moving on as
513 * many inodes as possible in a short period of time. It kicks itself every few
514 * seconds, as well as being kicked by the inode cache shrinker when memory
515 * goes low. It scans as quickly as possible avoiding locked inodes or those
516 * already being flushed, and once done schedules a future pass.
520 struct work_struct *work)
522 struct xfs_mount *mp = container_of(to_delayed_work(work),
523 struct xfs_mount, m_reclaim_work);
525 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
526 xfs_syncd_queue_reclaim(mp);
530 * Flush delayed allocate data, attempting to free up reserved space
531 * from existing allocations. At this point a new allocation attempt
532 * has failed with ENOSPC and we are in the process of scratching our
533 * heads, looking about for more room.
535 * Queue a new data flush if there isn't one already in progress and
536 * wait for completion of the flush. This means that we only ever have one
537 * inode flush in progress no matter how many ENOSPC events are occurring and
538 * so will prevent the system from bogging down due to every concurrent
539 * ENOSPC event scanning all the active inodes in the system for writeback.
543 struct xfs_inode *ip)
545 struct xfs_mount *mp = ip->i_mount;
547 queue_work(xfs_syncd_wq, &mp->m_flush_work);
548 flush_work_sync(&mp->m_flush_work);
553 struct work_struct *work)
555 struct xfs_mount *mp = container_of(work,
556 struct xfs_mount, m_flush_work);
558 xfs_sync_data(mp, SYNC_TRYLOCK);
559 xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT);
564 struct xfs_mount *mp)
566 INIT_WORK(&mp->m_flush_work, xfs_flush_worker);
567 INIT_DELAYED_WORK(&mp->m_sync_work, xfs_sync_worker);
568 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
570 xfs_syncd_queue_sync(mp);
571 xfs_syncd_queue_reclaim(mp);
578 struct xfs_mount *mp)
580 cancel_delayed_work_sync(&mp->m_sync_work);
581 cancel_delayed_work_sync(&mp->m_reclaim_work);
582 cancel_work_sync(&mp->m_flush_work);
586 __xfs_inode_set_reclaim_tag(
587 struct xfs_perag *pag,
588 struct xfs_inode *ip)
590 radix_tree_tag_set(&pag->pag_ici_root,
591 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
592 XFS_ICI_RECLAIM_TAG);
594 if (!pag->pag_ici_reclaimable) {
595 /* propagate the reclaim tag up into the perag radix tree */
596 spin_lock(&ip->i_mount->m_perag_lock);
597 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
598 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
599 XFS_ICI_RECLAIM_TAG);
600 spin_unlock(&ip->i_mount->m_perag_lock);
602 /* schedule periodic background inode reclaim */
603 xfs_syncd_queue_reclaim(ip->i_mount);
605 trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno,
608 pag->pag_ici_reclaimable++;
612 * We set the inode flag atomically with the radix tree tag.
613 * Once we get tag lookups on the radix tree, this inode flag
617 xfs_inode_set_reclaim_tag(
620 struct xfs_mount *mp = ip->i_mount;
621 struct xfs_perag *pag;
623 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
624 spin_lock(&pag->pag_ici_lock);
625 spin_lock(&ip->i_flags_lock);
626 __xfs_inode_set_reclaim_tag(pag, ip);
627 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
628 spin_unlock(&ip->i_flags_lock);
629 spin_unlock(&pag->pag_ici_lock);
634 __xfs_inode_clear_reclaim(
638 pag->pag_ici_reclaimable--;
639 if (!pag->pag_ici_reclaimable) {
640 /* clear the reclaim tag from the perag radix tree */
641 spin_lock(&ip->i_mount->m_perag_lock);
642 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
643 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
644 XFS_ICI_RECLAIM_TAG);
645 spin_unlock(&ip->i_mount->m_perag_lock);
646 trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno,
652 __xfs_inode_clear_reclaim_tag(
657 radix_tree_tag_clear(&pag->pag_ici_root,
658 XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
659 __xfs_inode_clear_reclaim(pag, ip);
663 * Grab the inode for reclaim exclusively.
664 * Return 0 if we grabbed it, non-zero otherwise.
667 xfs_reclaim_inode_grab(
668 struct xfs_inode *ip,
671 ASSERT(rcu_read_lock_held());
673 /* quick check for stale RCU freed inode */
678 * do some unlocked checks first to avoid unnecessary lock traffic.
679 * The first is a flush lock check, the second is a already in reclaim
680 * check. Only do these checks if we are not going to block on locks.
682 if ((flags & SYNC_TRYLOCK) &&
683 (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) {
688 * The radix tree lock here protects a thread in xfs_iget from racing
689 * with us starting reclaim on the inode. Once we have the
690 * XFS_IRECLAIM flag set it will not touch us.
692 * Due to RCU lookup, we may find inodes that have been freed and only
693 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
694 * aren't candidates for reclaim at all, so we must check the
695 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
697 spin_lock(&ip->i_flags_lock);
698 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
699 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
700 /* not a reclaim candidate. */
701 spin_unlock(&ip->i_flags_lock);
704 __xfs_iflags_set(ip, XFS_IRECLAIM);
705 spin_unlock(&ip->i_flags_lock);
710 * Inodes in different states need to be treated differently, and the return
711 * value of xfs_iflush is not sufficient to get this right. The following table
712 * lists the inode states and the reclaim actions necessary for non-blocking
716 * inode state iflush ret required action
717 * --------------- ---------- ---------------
719 * shutdown EIO unpin and reclaim
720 * clean, unpinned 0 reclaim
721 * stale, unpinned 0 reclaim
722 * clean, pinned(*) 0 requeue
723 * stale, pinned EAGAIN requeue
724 * dirty, delwri ok 0 requeue
725 * dirty, delwri blocked EAGAIN requeue
726 * dirty, sync flush 0 reclaim
728 * (*) dgc: I don't think the clean, pinned state is possible but it gets
729 * handled anyway given the order of checks implemented.
731 * As can be seen from the table, the return value of xfs_iflush() is not
732 * sufficient to correctly decide the reclaim action here. The checks in
733 * xfs_iflush() might look like duplicates, but they are not.
735 * Also, because we get the flush lock first, we know that any inode that has
736 * been flushed delwri has had the flush completed by the time we check that
737 * the inode is clean. The clean inode check needs to be done before flushing
738 * the inode delwri otherwise we would loop forever requeuing clean inodes as
739 * we cannot tell apart a successful delwri flush and a clean inode from the
740 * return value of xfs_iflush().
742 * Note that because the inode is flushed delayed write by background
743 * writeback, the flush lock may already be held here and waiting on it can
744 * result in very long latencies. Hence for sync reclaims, where we wait on the
745 * flush lock, the caller should push out delayed write inodes first before
746 * trying to reclaim them to minimise the amount of time spent waiting. For
747 * background relaim, we just requeue the inode for the next pass.
749 * Hence the order of actions after gaining the locks should be:
751 * shutdown => unpin and reclaim
752 * pinned, delwri => requeue
753 * pinned, sync => unpin
756 * dirty, delwri => flush and requeue
757 * dirty, sync => flush, wait and reclaim
761 struct xfs_inode *ip,
762 struct xfs_perag *pag,
769 xfs_ilock(ip, XFS_ILOCK_EXCL);
770 if (!xfs_iflock_nowait(ip)) {
771 if (!(sync_mode & SYNC_WAIT))
775 * If we only have a single dirty inode in a cluster there is
776 * a fair chance that the AIL push may have pushed it into
777 * the buffer, but xfsbufd won't touch it until 30 seconds
778 * from now, and thus we will lock up here.
780 * Promote the inode buffer to the front of the delwri list
781 * and wake up xfsbufd now.
783 xfs_promote_inode(ip);
787 if (is_bad_inode(VFS_I(ip)))
789 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
793 if (xfs_ipincount(ip)) {
794 if (!(sync_mode & SYNC_WAIT)) {
800 if (xfs_iflags_test(ip, XFS_ISTALE))
802 if (xfs_inode_clean(ip))
806 * Now we have an inode that needs flushing.
808 * We do a nonblocking flush here even if we are doing a SYNC_WAIT
809 * reclaim as we can deadlock with inode cluster removal.
810 * xfs_ifree_cluster() can lock the inode buffer before it locks the
811 * ip->i_lock, and we are doing the exact opposite here. As a result,
812 * doing a blocking xfs_itobp() to get the cluster buffer will result
813 * in an ABBA deadlock with xfs_ifree_cluster().
815 * As xfs_ifree_cluser() must gather all inodes that are active in the
816 * cache to mark them stale, if we hit this case we don't actually want
817 * to do IO here - we want the inode marked stale so we can simply
818 * reclaim it. Hence if we get an EAGAIN error on a SYNC_WAIT flush,
819 * just unlock the inode, back off and try again. Hopefully the next
820 * pass through will see the stale flag set on the inode.
822 error = xfs_iflush(ip, SYNC_TRYLOCK | sync_mode);
823 if (sync_mode & SYNC_WAIT) {
824 if (error == EAGAIN) {
825 xfs_iunlock(ip, XFS_ILOCK_EXCL);
826 /* backoff longer than in xfs_ifree_cluster */
835 * When we have to flush an inode but don't have SYNC_WAIT set, we
836 * flush the inode out using a delwri buffer and wait for the next
837 * call into reclaim to find it in a clean state instead of waiting for
838 * it now. We also don't return errors here - if the error is transient
839 * then the next reclaim pass will flush the inode, and if the error
840 * is permanent then the next sync reclaim will reclaim the inode and
843 if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) {
844 xfs_warn(ip->i_mount,
845 "inode 0x%llx background reclaim flush failed with %d",
846 (long long)ip->i_ino, error);
849 xfs_iflags_clear(ip, XFS_IRECLAIM);
850 xfs_iunlock(ip, XFS_ILOCK_EXCL);
852 * We could return EAGAIN here to make reclaim rescan the inode tree in
853 * a short while. However, this just burns CPU time scanning the tree
854 * waiting for IO to complete and xfssyncd never goes back to the idle
855 * state. Instead, return 0 to let the next scheduled background reclaim
856 * attempt to reclaim the inode again.
862 xfs_iunlock(ip, XFS_ILOCK_EXCL);
864 XFS_STATS_INC(xs_ig_reclaims);
866 * Remove the inode from the per-AG radix tree.
868 * Because radix_tree_delete won't complain even if the item was never
869 * added to the tree assert that it's been there before to catch
870 * problems with the inode life time early on.
872 spin_lock(&pag->pag_ici_lock);
873 if (!radix_tree_delete(&pag->pag_ici_root,
874 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino)))
876 __xfs_inode_clear_reclaim(pag, ip);
877 spin_unlock(&pag->pag_ici_lock);
880 * Here we do an (almost) spurious inode lock in order to coordinate
881 * with inode cache radix tree lookups. This is because the lookup
882 * can reference the inodes in the cache without taking references.
884 * We make that OK here by ensuring that we wait until the inode is
885 * unlocked after the lookup before we go ahead and free it. We get
886 * both the ilock and the iolock because the code may need to drop the
887 * ilock one but will still hold the iolock.
889 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
891 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
899 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
900 * corrupted, we still want to try to reclaim all the inodes. If we don't,
901 * then a shut down during filesystem unmount reclaim walk leak all the
902 * unreclaimed inodes.
905 xfs_reclaim_inodes_ag(
906 struct xfs_mount *mp,
910 struct xfs_perag *pag;
914 int trylock = flags & SYNC_TRYLOCK;
920 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
921 unsigned long first_index = 0;
925 ag = pag->pag_agno + 1;
928 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
933 first_index = pag->pag_ici_reclaim_cursor;
935 mutex_lock(&pag->pag_ici_reclaim_lock);
938 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
942 nr_found = radix_tree_gang_lookup_tag(
944 (void **)batch, first_index,
946 XFS_ICI_RECLAIM_TAG);
954 * Grab the inodes before we drop the lock. if we found
955 * nothing, nr == 0 and the loop will be skipped.
957 for (i = 0; i < nr_found; i++) {
958 struct xfs_inode *ip = batch[i];
960 if (done || xfs_reclaim_inode_grab(ip, flags))
964 * Update the index for the next lookup. Catch
965 * overflows into the next AG range which can
966 * occur if we have inodes in the last block of
967 * the AG and we are currently pointing to the
970 * Because we may see inodes that are from the
971 * wrong AG due to RCU freeing and
972 * reallocation, only update the index if it
973 * lies in this AG. It was a race that lead us
974 * to see this inode, so another lookup from
975 * the same index will not find it again.
977 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
980 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
981 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
985 /* unlock now we've grabbed the inodes. */
988 for (i = 0; i < nr_found; i++) {
991 error = xfs_reclaim_inode(batch[i], pag, flags);
992 if (error && last_error != EFSCORRUPTED)
996 *nr_to_scan -= XFS_LOOKUP_BATCH;
1000 } while (nr_found && !done && *nr_to_scan > 0);
1002 if (trylock && !done)
1003 pag->pag_ici_reclaim_cursor = first_index;
1005 pag->pag_ici_reclaim_cursor = 0;
1006 mutex_unlock(&pag->pag_ici_reclaim_lock);
1011 * if we skipped any AG, and we still have scan count remaining, do
1012 * another pass this time using blocking reclaim semantics (i.e
1013 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1014 * ensure that when we get more reclaimers than AGs we block rather
1015 * than spin trying to execute reclaim.
1017 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1021 return XFS_ERROR(last_error);
1029 int nr_to_scan = INT_MAX;
1031 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1035 * Scan a certain number of inodes for reclaim.
1037 * When called we make sure that there is a background (fast) inode reclaim in
1038 * progress, while we will throttle the speed of reclaim via doing synchronous
1039 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1040 * them to be cleaned, which we hope will not be very long due to the
1041 * background walker having already kicked the IO off on those dirty inodes.
1044 xfs_reclaim_inodes_nr(
1045 struct xfs_mount *mp,
1048 /* kick background reclaimer and push the AIL */
1049 xfs_syncd_queue_reclaim(mp);
1050 xfs_ail_push_all(mp->m_ail);
1052 xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1056 * Return the number of reclaimable inodes in the filesystem for
1057 * the shrinker to determine how much to reclaim.
1060 xfs_reclaim_inodes_count(
1061 struct xfs_mount *mp)
1063 struct xfs_perag *pag;
1064 xfs_agnumber_t ag = 0;
1065 int reclaimable = 0;
1067 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1068 ag = pag->pag_agno + 1;
1069 reclaimable += pag->pag_ici_reclaimable;