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_format.h"
21 #include "xfs_types.h"
23 #include "xfs_log_priv.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_dinode.h"
33 #include "xfs_error.h"
34 #include "xfs_filestream.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_quota.h"
37 #include "xfs_trace.h"
38 #include "xfs_fsops.h"
39 #include "xfs_icache.h"
40 #include "xfs_bmap_util.h"
42 #include <linux/kthread.h>
43 #include <linux/freezer.h>
45 STATIC
void __xfs_inode_clear_reclaim_tag(struct xfs_mount
*mp
,
46 struct xfs_perag
*pag
, struct xfs_inode
*ip
);
49 * Allocate and initialise an xfs_inode.
51 STATIC
struct xfs_inode
*
59 * if this didn't occur in transactions, we could use
60 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
61 * code up to do this anyway.
63 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
66 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
67 kmem_zone_free(xfs_inode_zone
, ip
);
71 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
72 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
73 ASSERT(!xfs_isiflocked(ip
));
74 ASSERT(ip
->i_ino
== 0);
76 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
78 /* initialise the xfs inode */
81 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
83 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
85 ip
->i_delayed_blks
= 0;
86 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
92 xfs_inode_free_callback(
93 struct rcu_head
*head
)
95 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
96 struct xfs_inode
*ip
= XFS_I(inode
);
98 kmem_zone_free(xfs_inode_zone
, ip
);
103 struct xfs_inode
*ip
)
105 switch (ip
->i_d
.di_mode
& S_IFMT
) {
109 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
114 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
117 ASSERT(!(ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
));
118 xfs_inode_item_destroy(ip
);
122 /* asserts to verify all state is correct here */
123 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
124 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
125 ASSERT(!xfs_isiflocked(ip
));
128 * Because we use RCU freeing we need to ensure the inode always
129 * appears to be reclaimed with an invalid inode number when in the
130 * free state. The ip->i_flags_lock provides the barrier against lookup
133 spin_lock(&ip
->i_flags_lock
);
134 ip
->i_flags
= XFS_IRECLAIM
;
136 spin_unlock(&ip
->i_flags_lock
);
138 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
142 * Check the validity of the inode we just found it the cache
146 struct xfs_perag
*pag
,
147 struct xfs_inode
*ip
,
150 int lock_flags
) __releases(RCU
)
152 struct inode
*inode
= VFS_I(ip
);
153 struct xfs_mount
*mp
= ip
->i_mount
;
157 * check for re-use of an inode within an RCU grace period due to the
158 * radix tree nodes not being updated yet. We monitor for this by
159 * setting the inode number to zero before freeing the inode structure.
160 * If the inode has been reallocated and set up, then the inode number
161 * will not match, so check for that, too.
163 spin_lock(&ip
->i_flags_lock
);
164 if (ip
->i_ino
!= ino
) {
165 trace_xfs_iget_skip(ip
);
166 XFS_STATS_INC(xs_ig_frecycle
);
173 * If we are racing with another cache hit that is currently
174 * instantiating this inode or currently recycling it out of
175 * reclaimabe state, wait for the initialisation to complete
178 * XXX(hch): eventually we should do something equivalent to
179 * wait_on_inode to wait for these flags to be cleared
180 * instead of polling for it.
182 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
183 trace_xfs_iget_skip(ip
);
184 XFS_STATS_INC(xs_ig_frecycle
);
190 * If lookup is racing with unlink return an error immediately.
192 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
198 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
199 * Need to carefully get it back into useable state.
201 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
202 trace_xfs_iget_reclaim(ip
);
205 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
206 * from stomping over us while we recycle the inode. We can't
207 * clear the radix tree reclaimable tag yet as it requires
208 * pag_ici_lock to be held exclusive.
210 ip
->i_flags
|= XFS_IRECLAIM
;
212 spin_unlock(&ip
->i_flags_lock
);
215 error
= -inode_init_always(mp
->m_super
, inode
);
218 * Re-initializing the inode failed, and we are in deep
219 * trouble. Try to re-add it to the reclaim list.
222 spin_lock(&ip
->i_flags_lock
);
224 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
225 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
226 trace_xfs_iget_reclaim_fail(ip
);
230 spin_lock(&pag
->pag_ici_lock
);
231 spin_lock(&ip
->i_flags_lock
);
234 * Clear the per-lifetime state in the inode as we are now
235 * effectively a new inode and need to return to the initial
236 * state before reuse occurs.
238 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
239 ip
->i_flags
|= XFS_INEW
;
240 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
241 inode
->i_state
= I_NEW
;
243 ASSERT(!rwsem_is_locked(&ip
->i_iolock
.mr_lock
));
244 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
246 spin_unlock(&ip
->i_flags_lock
);
247 spin_unlock(&pag
->pag_ici_lock
);
249 /* If the VFS inode is being torn down, pause and try again. */
251 trace_xfs_iget_skip(ip
);
256 /* We've got a live one. */
257 spin_unlock(&ip
->i_flags_lock
);
259 trace_xfs_iget_hit(ip
);
263 xfs_ilock(ip
, lock_flags
);
265 xfs_iflags_clear(ip
, XFS_ISTALE
| XFS_IDONTCACHE
);
266 XFS_STATS_INC(xs_ig_found
);
271 spin_unlock(&ip
->i_flags_lock
);
279 struct xfs_mount
*mp
,
280 struct xfs_perag
*pag
,
283 struct xfs_inode
**ipp
,
287 struct xfs_inode
*ip
;
289 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
292 ip
= xfs_inode_alloc(mp
, ino
);
296 error
= xfs_iread(mp
, tp
, ip
, flags
);
300 trace_xfs_iget_miss(ip
);
302 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
308 * Preload the radix tree so we can insert safely under the
309 * write spinlock. Note that we cannot sleep inside the preload
310 * region. Since we can be called from transaction context, don't
311 * recurse into the file system.
313 if (radix_tree_preload(GFP_NOFS
)) {
319 * Because the inode hasn't been added to the radix-tree yet it can't
320 * be found by another thread, so we can do the non-sleeping lock here.
323 if (!xfs_ilock_nowait(ip
, lock_flags
))
328 * These values must be set before inserting the inode into the radix
329 * tree as the moment it is inserted a concurrent lookup (allowed by the
330 * RCU locking mechanism) can find it and that lookup must see that this
331 * is an inode currently under construction (i.e. that XFS_INEW is set).
332 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
333 * memory barrier that ensures this detection works correctly at lookup
337 if (flags
& XFS_IGET_DONTCACHE
)
338 iflags
|= XFS_IDONTCACHE
;
342 xfs_iflags_set(ip
, iflags
);
344 /* insert the new inode */
345 spin_lock(&pag
->pag_ici_lock
);
346 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
347 if (unlikely(error
)) {
348 WARN_ON(error
!= -EEXIST
);
349 XFS_STATS_INC(xs_ig_dup
);
351 goto out_preload_end
;
353 spin_unlock(&pag
->pag_ici_lock
);
354 radix_tree_preload_end();
360 spin_unlock(&pag
->pag_ici_lock
);
361 radix_tree_preload_end();
363 xfs_iunlock(ip
, lock_flags
);
365 __destroy_inode(VFS_I(ip
));
371 * Look up an inode by number in the given file system.
372 * The inode is looked up in the cache held in each AG.
373 * If the inode is found in the cache, initialise the vfs inode
376 * If it is not in core, read it in from the file system's device,
377 * add it to the cache and initialise the vfs inode.
379 * The inode is locked according to the value of the lock_flags parameter.
380 * This flag parameter indicates how and if the inode's IO lock and inode lock
383 * mp -- the mount point structure for the current file system. It points
384 * to the inode hash table.
385 * tp -- a pointer to the current transaction if there is one. This is
386 * simply passed through to the xfs_iread() call.
387 * ino -- the number of the inode desired. This is the unique identifier
388 * within the file system for the inode being requested.
389 * lock_flags -- flags indicating how to lock the inode. See the comment
390 * for xfs_ilock() for a list of valid values.
407 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
408 * doesn't get freed while it's being referenced during a
409 * radix tree traversal here. It assumes this function
410 * aqcuires only the ILOCK (and therefore it has no need to
411 * involve the IOLOCK in this synchronization).
413 ASSERT((lock_flags
& (XFS_IOLOCK_EXCL
| XFS_IOLOCK_SHARED
)) == 0);
415 /* reject inode numbers outside existing AGs */
416 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
419 /* get the perag structure and ensure that it's inode capable */
420 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
421 agino
= XFS_INO_TO_AGINO(mp
, ino
);
426 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
429 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
431 goto out_error_or_again
;
434 XFS_STATS_INC(xs_ig_missed
);
436 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
439 goto out_error_or_again
;
446 * If we have a real type for an on-disk inode, we can set ops(&unlock)
447 * now. If it's a new inode being created, xfs_ialloc will handle it.
449 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
454 if (error
== EAGAIN
) {
463 * The inode lookup is done in batches to keep the amount of lock traffic and
464 * radix tree lookups to a minimum. The batch size is a trade off between
465 * lookup reduction and stack usage. This is in the reclaim path, so we can't
468 #define XFS_LOOKUP_BATCH 32
471 xfs_inode_ag_walk_grab(
472 struct xfs_inode
*ip
)
474 struct inode
*inode
= VFS_I(ip
);
476 ASSERT(rcu_read_lock_held());
479 * check for stale RCU freed inode
481 * If the inode has been reallocated, it doesn't matter if it's not in
482 * the AG we are walking - we are walking for writeback, so if it
483 * passes all the "valid inode" checks and is dirty, then we'll write
484 * it back anyway. If it has been reallocated and still being
485 * initialised, the XFS_INEW check below will catch it.
487 spin_lock(&ip
->i_flags_lock
);
489 goto out_unlock_noent
;
491 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
492 if (__xfs_iflags_test(ip
, XFS_INEW
| XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
493 goto out_unlock_noent
;
494 spin_unlock(&ip
->i_flags_lock
);
496 /* nothing to sync during shutdown */
497 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
500 /* If we can't grab the inode, it must on it's way to reclaim. */
504 if (is_bad_inode(inode
)) {
513 spin_unlock(&ip
->i_flags_lock
);
519 struct xfs_mount
*mp
,
520 struct xfs_perag
*pag
,
521 int (*execute
)(struct xfs_inode
*ip
,
522 struct xfs_perag
*pag
, int flags
,
528 uint32_t first_index
;
540 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
547 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
548 (void **)batch
, first_index
,
551 nr_found
= radix_tree_gang_lookup_tag(
553 (void **) batch
, first_index
,
554 XFS_LOOKUP_BATCH
, tag
);
562 * Grab the inodes before we drop the lock. if we found
563 * nothing, nr == 0 and the loop will be skipped.
565 for (i
= 0; i
< nr_found
; i
++) {
566 struct xfs_inode
*ip
= batch
[i
];
568 if (done
|| xfs_inode_ag_walk_grab(ip
))
572 * Update the index for the next lookup. Catch
573 * overflows into the next AG range which can occur if
574 * we have inodes in the last block of the AG and we
575 * are currently pointing to the last inode.
577 * Because we may see inodes that are from the wrong AG
578 * due to RCU freeing and reallocation, only update the
579 * index if it lies in this AG. It was a race that lead
580 * us to see this inode, so another lookup from the
581 * same index will not find it again.
583 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
585 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
586 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
590 /* unlock now we've grabbed the inodes. */
593 for (i
= 0; i
< nr_found
; i
++) {
596 error
= execute(batch
[i
], pag
, flags
, args
);
598 if (error
== EAGAIN
) {
602 if (error
&& last_error
!= EFSCORRUPTED
)
606 /* bail out if the filesystem is corrupted. */
607 if (error
== EFSCORRUPTED
)
612 } while (nr_found
&& !done
);
622 * Background scanning to trim post-EOF preallocated space. This is queued
623 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
627 struct xfs_mount
*mp
)
630 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_EOFBLOCKS_TAG
))
631 queue_delayed_work(mp
->m_eofblocks_workqueue
,
632 &mp
->m_eofblocks_work
,
633 msecs_to_jiffies(xfs_eofb_secs
* 1000));
638 xfs_eofblocks_worker(
639 struct work_struct
*work
)
641 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
642 struct xfs_mount
, m_eofblocks_work
);
643 xfs_icache_free_eofblocks(mp
, NULL
);
644 xfs_queue_eofblocks(mp
);
648 xfs_inode_ag_iterator(
649 struct xfs_mount
*mp
,
650 int (*execute
)(struct xfs_inode
*ip
,
651 struct xfs_perag
*pag
, int flags
,
656 struct xfs_perag
*pag
;
662 while ((pag
= xfs_perag_get(mp
, ag
))) {
663 ag
= pag
->pag_agno
+ 1;
664 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, -1);
668 if (error
== EFSCORRUPTED
)
672 return XFS_ERROR(last_error
);
676 xfs_inode_ag_iterator_tag(
677 struct xfs_mount
*mp
,
678 int (*execute
)(struct xfs_inode
*ip
,
679 struct xfs_perag
*pag
, int flags
,
685 struct xfs_perag
*pag
;
691 while ((pag
= xfs_perag_get_tag(mp
, ag
, tag
))) {
692 ag
= pag
->pag_agno
+ 1;
693 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, tag
);
697 if (error
== EFSCORRUPTED
)
701 return XFS_ERROR(last_error
);
705 * Queue a new inode reclaim pass if there are reclaimable inodes and there
706 * isn't a reclaim pass already in progress. By default it runs every 5s based
707 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
708 * tunable, but that can be done if this method proves to be ineffective or too
712 xfs_reclaim_work_queue(
713 struct xfs_mount
*mp
)
717 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
718 queue_delayed_work(mp
->m_reclaim_workqueue
, &mp
->m_reclaim_work
,
719 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
725 * This is a fast pass over the inode cache to try to get reclaim moving on as
726 * many inodes as possible in a short period of time. It kicks itself every few
727 * seconds, as well as being kicked by the inode cache shrinker when memory
728 * goes low. It scans as quickly as possible avoiding locked inodes or those
729 * already being flushed, and once done schedules a future pass.
733 struct work_struct
*work
)
735 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
736 struct xfs_mount
, m_reclaim_work
);
738 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
739 xfs_reclaim_work_queue(mp
);
743 __xfs_inode_set_reclaim_tag(
744 struct xfs_perag
*pag
,
745 struct xfs_inode
*ip
)
747 radix_tree_tag_set(&pag
->pag_ici_root
,
748 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
749 XFS_ICI_RECLAIM_TAG
);
751 if (!pag
->pag_ici_reclaimable
) {
752 /* propagate the reclaim tag up into the perag radix tree */
753 spin_lock(&ip
->i_mount
->m_perag_lock
);
754 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
755 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
756 XFS_ICI_RECLAIM_TAG
);
757 spin_unlock(&ip
->i_mount
->m_perag_lock
);
759 /* schedule periodic background inode reclaim */
760 xfs_reclaim_work_queue(ip
->i_mount
);
762 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
765 pag
->pag_ici_reclaimable
++;
769 * We set the inode flag atomically with the radix tree tag.
770 * Once we get tag lookups on the radix tree, this inode flag
774 xfs_inode_set_reclaim_tag(
777 struct xfs_mount
*mp
= ip
->i_mount
;
778 struct xfs_perag
*pag
;
780 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
781 spin_lock(&pag
->pag_ici_lock
);
782 spin_lock(&ip
->i_flags_lock
);
783 __xfs_inode_set_reclaim_tag(pag
, ip
);
784 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
785 spin_unlock(&ip
->i_flags_lock
);
786 spin_unlock(&pag
->pag_ici_lock
);
791 __xfs_inode_clear_reclaim(
795 pag
->pag_ici_reclaimable
--;
796 if (!pag
->pag_ici_reclaimable
) {
797 /* clear the reclaim tag from the perag radix tree */
798 spin_lock(&ip
->i_mount
->m_perag_lock
);
799 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
800 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
801 XFS_ICI_RECLAIM_TAG
);
802 spin_unlock(&ip
->i_mount
->m_perag_lock
);
803 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
809 __xfs_inode_clear_reclaim_tag(
814 radix_tree_tag_clear(&pag
->pag_ici_root
,
815 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
816 __xfs_inode_clear_reclaim(pag
, ip
);
820 * Grab the inode for reclaim exclusively.
821 * Return 0 if we grabbed it, non-zero otherwise.
824 xfs_reclaim_inode_grab(
825 struct xfs_inode
*ip
,
828 ASSERT(rcu_read_lock_held());
830 /* quick check for stale RCU freed inode */
835 * If we are asked for non-blocking operation, do unlocked checks to
836 * see if the inode already is being flushed or in reclaim to avoid
839 if ((flags
& SYNC_TRYLOCK
) &&
840 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
844 * The radix tree lock here protects a thread in xfs_iget from racing
845 * with us starting reclaim on the inode. Once we have the
846 * XFS_IRECLAIM flag set it will not touch us.
848 * Due to RCU lookup, we may find inodes that have been freed and only
849 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
850 * aren't candidates for reclaim at all, so we must check the
851 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
853 spin_lock(&ip
->i_flags_lock
);
854 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
855 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
856 /* not a reclaim candidate. */
857 spin_unlock(&ip
->i_flags_lock
);
860 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
861 spin_unlock(&ip
->i_flags_lock
);
866 * Inodes in different states need to be treated differently. The following
867 * table lists the inode states and the reclaim actions necessary:
869 * inode state iflush ret required action
870 * --------------- ---------- ---------------
872 * shutdown EIO unpin and reclaim
873 * clean, unpinned 0 reclaim
874 * stale, unpinned 0 reclaim
875 * clean, pinned(*) 0 requeue
876 * stale, pinned EAGAIN requeue
877 * dirty, async - requeue
878 * dirty, sync 0 reclaim
880 * (*) dgc: I don't think the clean, pinned state is possible but it gets
881 * handled anyway given the order of checks implemented.
883 * Also, because we get the flush lock first, we know that any inode that has
884 * been flushed delwri has had the flush completed by the time we check that
885 * the inode is clean.
887 * Note that because the inode is flushed delayed write by AIL pushing, the
888 * flush lock may already be held here and waiting on it can result in very
889 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
890 * the caller should push the AIL first before trying to reclaim inodes to
891 * minimise the amount of time spent waiting. For background relaim, we only
892 * bother to reclaim clean inodes anyway.
894 * Hence the order of actions after gaining the locks should be:
896 * shutdown => unpin and reclaim
897 * pinned, async => requeue
898 * pinned, sync => unpin
901 * dirty, async => requeue
902 * dirty, sync => flush, wait and reclaim
906 struct xfs_inode
*ip
,
907 struct xfs_perag
*pag
,
910 struct xfs_buf
*bp
= NULL
;
915 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
916 if (!xfs_iflock_nowait(ip
)) {
917 if (!(sync_mode
& SYNC_WAIT
))
922 if (is_bad_inode(VFS_I(ip
)))
924 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
926 xfs_iflush_abort(ip
, false);
929 if (xfs_ipincount(ip
)) {
930 if (!(sync_mode
& SYNC_WAIT
))
934 if (xfs_iflags_test(ip
, XFS_ISTALE
))
936 if (xfs_inode_clean(ip
))
940 * Never flush out dirty data during non-blocking reclaim, as it would
941 * just contend with AIL pushing trying to do the same job.
943 if (!(sync_mode
& SYNC_WAIT
))
947 * Now we have an inode that needs flushing.
949 * Note that xfs_iflush will never block on the inode buffer lock, as
950 * xfs_ifree_cluster() can lock the inode buffer before it locks the
951 * ip->i_lock, and we are doing the exact opposite here. As a result,
952 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
953 * result in an ABBA deadlock with xfs_ifree_cluster().
955 * As xfs_ifree_cluser() must gather all inodes that are active in the
956 * cache to mark them stale, if we hit this case we don't actually want
957 * to do IO here - we want the inode marked stale so we can simply
958 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
959 * inode, back off and try again. Hopefully the next pass through will
960 * see the stale flag set on the inode.
962 error
= xfs_iflush(ip
, &bp
);
963 if (error
== EAGAIN
) {
964 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
965 /* backoff longer than in xfs_ifree_cluster */
971 error
= xfs_bwrite(bp
);
978 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
980 XFS_STATS_INC(xs_ig_reclaims
);
982 * Remove the inode from the per-AG radix tree.
984 * Because radix_tree_delete won't complain even if the item was never
985 * added to the tree assert that it's been there before to catch
986 * problems with the inode life time early on.
988 spin_lock(&pag
->pag_ici_lock
);
989 if (!radix_tree_delete(&pag
->pag_ici_root
,
990 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
992 __xfs_inode_clear_reclaim(pag
, ip
);
993 spin_unlock(&pag
->pag_ici_lock
);
996 * Here we do an (almost) spurious inode lock in order to coordinate
997 * with inode cache radix tree lookups. This is because the lookup
998 * can reference the inodes in the cache without taking references.
1000 * We make that OK here by ensuring that we wait until the inode is
1001 * unlocked after the lookup before we go ahead and free it.
1003 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1004 xfs_qm_dqdetach(ip
);
1005 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1013 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
1014 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1016 * We could return EAGAIN here to make reclaim rescan the inode tree in
1017 * a short while. However, this just burns CPU time scanning the tree
1018 * waiting for IO to complete and the reclaim work never goes back to
1019 * the idle state. Instead, return 0 to let the next scheduled
1020 * background reclaim attempt to reclaim the inode again.
1026 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1027 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1028 * then a shut down during filesystem unmount reclaim walk leak all the
1029 * unreclaimed inodes.
1032 xfs_reclaim_inodes_ag(
1033 struct xfs_mount
*mp
,
1037 struct xfs_perag
*pag
;
1041 int trylock
= flags
& SYNC_TRYLOCK
;
1047 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1048 unsigned long first_index
= 0;
1052 ag
= pag
->pag_agno
+ 1;
1055 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
1060 first_index
= pag
->pag_ici_reclaim_cursor
;
1062 mutex_lock(&pag
->pag_ici_reclaim_lock
);
1065 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
1069 nr_found
= radix_tree_gang_lookup_tag(
1071 (void **)batch
, first_index
,
1073 XFS_ICI_RECLAIM_TAG
);
1081 * Grab the inodes before we drop the lock. if we found
1082 * nothing, nr == 0 and the loop will be skipped.
1084 for (i
= 0; i
< nr_found
; i
++) {
1085 struct xfs_inode
*ip
= batch
[i
];
1087 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
1091 * Update the index for the next lookup. Catch
1092 * overflows into the next AG range which can
1093 * occur if we have inodes in the last block of
1094 * the AG and we are currently pointing to the
1097 * Because we may see inodes that are from the
1098 * wrong AG due to RCU freeing and
1099 * reallocation, only update the index if it
1100 * lies in this AG. It was a race that lead us
1101 * to see this inode, so another lookup from
1102 * the same index will not find it again.
1104 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
1107 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
1108 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
1112 /* unlock now we've grabbed the inodes. */
1115 for (i
= 0; i
< nr_found
; i
++) {
1118 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
1119 if (error
&& last_error
!= EFSCORRUPTED
)
1123 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
1127 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
1129 if (trylock
&& !done
)
1130 pag
->pag_ici_reclaim_cursor
= first_index
;
1132 pag
->pag_ici_reclaim_cursor
= 0;
1133 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1138 * if we skipped any AG, and we still have scan count remaining, do
1139 * another pass this time using blocking reclaim semantics (i.e
1140 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1141 * ensure that when we get more reclaimers than AGs we block rather
1142 * than spin trying to execute reclaim.
1144 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1148 return XFS_ERROR(last_error
);
1156 int nr_to_scan
= INT_MAX
;
1158 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1162 * Scan a certain number of inodes for reclaim.
1164 * When called we make sure that there is a background (fast) inode reclaim in
1165 * progress, while we will throttle the speed of reclaim via doing synchronous
1166 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1167 * them to be cleaned, which we hope will not be very long due to the
1168 * background walker having already kicked the IO off on those dirty inodes.
1171 xfs_reclaim_inodes_nr(
1172 struct xfs_mount
*mp
,
1175 /* kick background reclaimer and push the AIL */
1176 xfs_reclaim_work_queue(mp
);
1177 xfs_ail_push_all(mp
->m_ail
);
1179 xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1183 * Return the number of reclaimable inodes in the filesystem for
1184 * the shrinker to determine how much to reclaim.
1187 xfs_reclaim_inodes_count(
1188 struct xfs_mount
*mp
)
1190 struct xfs_perag
*pag
;
1191 xfs_agnumber_t ag
= 0;
1192 int reclaimable
= 0;
1194 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1195 ag
= pag
->pag_agno
+ 1;
1196 reclaimable
+= pag
->pag_ici_reclaimable
;
1204 struct xfs_inode
*ip
,
1205 struct xfs_eofblocks
*eofb
)
1207 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1208 !uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1211 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1212 !gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1215 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1216 xfs_get_projid(ip
) != eofb
->eof_prid
)
1223 xfs_inode_free_eofblocks(
1224 struct xfs_inode
*ip
,
1225 struct xfs_perag
*pag
,
1230 struct xfs_eofblocks
*eofb
= args
;
1232 if (!xfs_can_free_eofblocks(ip
, false)) {
1233 /* inode could be preallocated or append-only */
1234 trace_xfs_inode_free_eofblocks_invalid(ip
);
1235 xfs_inode_clear_eofblocks_tag(ip
);
1240 * If the mapping is dirty the operation can block and wait for some
1241 * time. Unless we are waiting, skip it.
1243 if (!(flags
& SYNC_WAIT
) &&
1244 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
))
1248 if (!xfs_inode_match_id(ip
, eofb
))
1251 /* skip the inode if the file size is too small */
1252 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1253 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1257 ret
= xfs_free_eofblocks(ip
->i_mount
, ip
, true);
1259 /* don't revisit the inode if we're not waiting */
1260 if (ret
== EAGAIN
&& !(flags
& SYNC_WAIT
))
1267 xfs_icache_free_eofblocks(
1268 struct xfs_mount
*mp
,
1269 struct xfs_eofblocks
*eofb
)
1271 int flags
= SYNC_TRYLOCK
;
1273 if (eofb
&& (eofb
->eof_flags
& XFS_EOF_FLAGS_SYNC
))
1276 return xfs_inode_ag_iterator_tag(mp
, xfs_inode_free_eofblocks
, flags
,
1277 eofb
, XFS_ICI_EOFBLOCKS_TAG
);
1281 xfs_inode_set_eofblocks_tag(
1284 struct xfs_mount
*mp
= ip
->i_mount
;
1285 struct xfs_perag
*pag
;
1288 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1289 spin_lock(&pag
->pag_ici_lock
);
1290 trace_xfs_inode_set_eofblocks_tag(ip
);
1292 tagged
= radix_tree_tagged(&pag
->pag_ici_root
,
1293 XFS_ICI_EOFBLOCKS_TAG
);
1294 radix_tree_tag_set(&pag
->pag_ici_root
,
1295 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
1296 XFS_ICI_EOFBLOCKS_TAG
);
1298 /* propagate the eofblocks tag up into the perag radix tree */
1299 spin_lock(&ip
->i_mount
->m_perag_lock
);
1300 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
1301 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1302 XFS_ICI_EOFBLOCKS_TAG
);
1303 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1305 /* kick off background trimming */
1306 xfs_queue_eofblocks(ip
->i_mount
);
1308 trace_xfs_perag_set_eofblocks(ip
->i_mount
, pag
->pag_agno
,
1312 spin_unlock(&pag
->pag_ici_lock
);
1317 xfs_inode_clear_eofblocks_tag(
1320 struct xfs_mount
*mp
= ip
->i_mount
;
1321 struct xfs_perag
*pag
;
1323 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1324 spin_lock(&pag
->pag_ici_lock
);
1325 trace_xfs_inode_clear_eofblocks_tag(ip
);
1327 radix_tree_tag_clear(&pag
->pag_ici_root
,
1328 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
1329 XFS_ICI_EOFBLOCKS_TAG
);
1330 if (!radix_tree_tagged(&pag
->pag_ici_root
, XFS_ICI_EOFBLOCKS_TAG
)) {
1331 /* clear the eofblocks tag from the perag radix tree */
1332 spin_lock(&ip
->i_mount
->m_perag_lock
);
1333 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
1334 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1335 XFS_ICI_EOFBLOCKS_TAG
);
1336 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1337 trace_xfs_perag_clear_eofblocks(ip
->i_mount
, pag
->pag_agno
,
1341 spin_unlock(&pag
->pag_ici_lock
);