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"
22 #include "xfs_log_priv.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"
39 #include "xfs_icache.h"
41 #include <linux/kthread.h>
42 #include <linux/freezer.h>
44 STATIC
void __xfs_inode_clear_reclaim_tag(struct xfs_mount
*mp
,
45 struct xfs_perag
*pag
, struct xfs_inode
*ip
);
48 * Allocate and initialise an xfs_inode.
50 STATIC
struct xfs_inode
*
58 * if this didn't occur in transactions, we could use
59 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
60 * code up to do this anyway.
62 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
65 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
66 kmem_zone_free(xfs_inode_zone
, ip
);
70 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
71 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
72 ASSERT(!xfs_isiflocked(ip
));
73 ASSERT(ip
->i_ino
== 0);
75 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
77 /* initialise the xfs inode */
80 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
82 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
84 ip
->i_delayed_blks
= 0;
85 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
91 xfs_inode_free_callback(
92 struct rcu_head
*head
)
94 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
95 struct xfs_inode
*ip
= XFS_I(inode
);
97 kmem_zone_free(xfs_inode_zone
, ip
);
102 struct xfs_inode
*ip
)
104 switch (ip
->i_d
.di_mode
& S_IFMT
) {
108 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
113 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
116 ASSERT(!(ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
));
117 xfs_inode_item_destroy(ip
);
121 /* asserts to verify all state is correct here */
122 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
123 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
124 ASSERT(!xfs_isiflocked(ip
));
127 * Because we use RCU freeing we need to ensure the inode always
128 * appears to be reclaimed with an invalid inode number when in the
129 * free state. The ip->i_flags_lock provides the barrier against lookup
132 spin_lock(&ip
->i_flags_lock
);
133 ip
->i_flags
= XFS_IRECLAIM
;
135 spin_unlock(&ip
->i_flags_lock
);
137 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
141 * Check the validity of the inode we just found it the cache
145 struct xfs_perag
*pag
,
146 struct xfs_inode
*ip
,
149 int lock_flags
) __releases(RCU
)
151 struct inode
*inode
= VFS_I(ip
);
152 struct xfs_mount
*mp
= ip
->i_mount
;
156 * check for re-use of an inode within an RCU grace period due to the
157 * radix tree nodes not being updated yet. We monitor for this by
158 * setting the inode number to zero before freeing the inode structure.
159 * If the inode has been reallocated and set up, then the inode number
160 * will not match, so check for that, too.
162 spin_lock(&ip
->i_flags_lock
);
163 if (ip
->i_ino
!= ino
) {
164 trace_xfs_iget_skip(ip
);
165 XFS_STATS_INC(xs_ig_frecycle
);
172 * If we are racing with another cache hit that is currently
173 * instantiating this inode or currently recycling it out of
174 * reclaimabe state, wait for the initialisation to complete
177 * XXX(hch): eventually we should do something equivalent to
178 * wait_on_inode to wait for these flags to be cleared
179 * instead of polling for it.
181 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
182 trace_xfs_iget_skip(ip
);
183 XFS_STATS_INC(xs_ig_frecycle
);
189 * If lookup is racing with unlink return an error immediately.
191 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
197 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
198 * Need to carefully get it back into useable state.
200 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
201 trace_xfs_iget_reclaim(ip
);
204 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
205 * from stomping over us while we recycle the inode. We can't
206 * clear the radix tree reclaimable tag yet as it requires
207 * pag_ici_lock to be held exclusive.
209 ip
->i_flags
|= XFS_IRECLAIM
;
211 spin_unlock(&ip
->i_flags_lock
);
214 error
= -inode_init_always(mp
->m_super
, inode
);
217 * Re-initializing the inode failed, and we are in deep
218 * trouble. Try to re-add it to the reclaim list.
221 spin_lock(&ip
->i_flags_lock
);
223 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
224 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
225 trace_xfs_iget_reclaim_fail(ip
);
229 spin_lock(&pag
->pag_ici_lock
);
230 spin_lock(&ip
->i_flags_lock
);
233 * Clear the per-lifetime state in the inode as we are now
234 * effectively a new inode and need to return to the initial
235 * state before reuse occurs.
237 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
238 ip
->i_flags
|= XFS_INEW
;
239 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
240 inode
->i_state
= I_NEW
;
242 ASSERT(!rwsem_is_locked(&ip
->i_iolock
.mr_lock
));
243 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
245 spin_unlock(&ip
->i_flags_lock
);
246 spin_unlock(&pag
->pag_ici_lock
);
248 /* If the VFS inode is being torn down, pause and try again. */
250 trace_xfs_iget_skip(ip
);
255 /* We've got a live one. */
256 spin_unlock(&ip
->i_flags_lock
);
258 trace_xfs_iget_hit(ip
);
262 xfs_ilock(ip
, lock_flags
);
264 xfs_iflags_clear(ip
, XFS_ISTALE
| XFS_IDONTCACHE
);
265 XFS_STATS_INC(xs_ig_found
);
270 spin_unlock(&ip
->i_flags_lock
);
278 struct xfs_mount
*mp
,
279 struct xfs_perag
*pag
,
282 struct xfs_inode
**ipp
,
286 struct xfs_inode
*ip
;
288 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
291 ip
= xfs_inode_alloc(mp
, ino
);
295 error
= xfs_iread(mp
, tp
, ip
, flags
);
299 trace_xfs_iget_miss(ip
);
301 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
307 * Preload the radix tree so we can insert safely under the
308 * write spinlock. Note that we cannot sleep inside the preload
309 * region. Since we can be called from transaction context, don't
310 * recurse into the file system.
312 if (radix_tree_preload(GFP_NOFS
)) {
318 * Because the inode hasn't been added to the radix-tree yet it can't
319 * be found by another thread, so we can do the non-sleeping lock here.
322 if (!xfs_ilock_nowait(ip
, lock_flags
))
327 * These values must be set before inserting the inode into the radix
328 * tree as the moment it is inserted a concurrent lookup (allowed by the
329 * RCU locking mechanism) can find it and that lookup must see that this
330 * is an inode currently under construction (i.e. that XFS_INEW is set).
331 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
332 * memory barrier that ensures this detection works correctly at lookup
336 if (flags
& XFS_IGET_DONTCACHE
)
337 iflags
|= XFS_IDONTCACHE
;
338 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
339 xfs_iflags_set(ip
, iflags
);
341 /* insert the new inode */
342 spin_lock(&pag
->pag_ici_lock
);
343 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
344 if (unlikely(error
)) {
345 WARN_ON(error
!= -EEXIST
);
346 XFS_STATS_INC(xs_ig_dup
);
348 goto out_preload_end
;
350 spin_unlock(&pag
->pag_ici_lock
);
351 radix_tree_preload_end();
357 spin_unlock(&pag
->pag_ici_lock
);
358 radix_tree_preload_end();
360 xfs_iunlock(ip
, lock_flags
);
362 __destroy_inode(VFS_I(ip
));
368 * Look up an inode by number in the given file system.
369 * The inode is looked up in the cache held in each AG.
370 * If the inode is found in the cache, initialise the vfs inode
373 * If it is not in core, read it in from the file system's device,
374 * add it to the cache and initialise the vfs inode.
376 * The inode is locked according to the value of the lock_flags parameter.
377 * This flag parameter indicates how and if the inode's IO lock and inode lock
380 * mp -- the mount point structure for the current file system. It points
381 * to the inode hash table.
382 * tp -- a pointer to the current transaction if there is one. This is
383 * simply passed through to the xfs_iread() call.
384 * ino -- the number of the inode desired. This is the unique identifier
385 * within the file system for the inode being requested.
386 * lock_flags -- flags indicating how to lock the inode. See the comment
387 * for xfs_ilock() for a list of valid values.
404 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
405 * doesn't get freed while it's being referenced during a
406 * radix tree traversal here. It assumes this function
407 * aqcuires only the ILOCK (and therefore it has no need to
408 * involve the IOLOCK in this synchronization).
410 ASSERT((lock_flags
& (XFS_IOLOCK_EXCL
| XFS_IOLOCK_SHARED
)) == 0);
412 /* reject inode numbers outside existing AGs */
413 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
416 /* get the perag structure and ensure that it's inode capable */
417 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
418 agino
= XFS_INO_TO_AGINO(mp
, ino
);
423 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
426 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
428 goto out_error_or_again
;
431 XFS_STATS_INC(xs_ig_missed
);
433 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
436 goto out_error_or_again
;
443 * If we have a real type for an on-disk inode, we can set ops(&unlock)
444 * now. If it's a new inode being created, xfs_ialloc will handle it.
446 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
451 if (error
== EAGAIN
) {
460 * The inode lookup is done in batches to keep the amount of lock traffic and
461 * radix tree lookups to a minimum. The batch size is a trade off between
462 * lookup reduction and stack usage. This is in the reclaim path, so we can't
465 #define XFS_LOOKUP_BATCH 32
468 xfs_inode_ag_walk_grab(
469 struct xfs_inode
*ip
)
471 struct inode
*inode
= VFS_I(ip
);
473 ASSERT(rcu_read_lock_held());
476 * check for stale RCU freed inode
478 * If the inode has been reallocated, it doesn't matter if it's not in
479 * the AG we are walking - we are walking for writeback, so if it
480 * passes all the "valid inode" checks and is dirty, then we'll write
481 * it back anyway. If it has been reallocated and still being
482 * initialised, the XFS_INEW check below will catch it.
484 spin_lock(&ip
->i_flags_lock
);
486 goto out_unlock_noent
;
488 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
489 if (__xfs_iflags_test(ip
, XFS_INEW
| XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
490 goto out_unlock_noent
;
491 spin_unlock(&ip
->i_flags_lock
);
493 /* nothing to sync during shutdown */
494 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
497 /* If we can't grab the inode, it must on it's way to reclaim. */
501 if (is_bad_inode(inode
)) {
510 spin_unlock(&ip
->i_flags_lock
);
516 struct xfs_mount
*mp
,
517 struct xfs_perag
*pag
,
518 int (*execute
)(struct xfs_inode
*ip
,
519 struct xfs_perag
*pag
, int flags
,
525 uint32_t first_index
;
537 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
544 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
545 (void **)batch
, first_index
,
548 nr_found
= radix_tree_gang_lookup_tag(
550 (void **) batch
, first_index
,
551 XFS_LOOKUP_BATCH
, tag
);
559 * Grab the inodes before we drop the lock. if we found
560 * nothing, nr == 0 and the loop will be skipped.
562 for (i
= 0; i
< nr_found
; i
++) {
563 struct xfs_inode
*ip
= batch
[i
];
565 if (done
|| xfs_inode_ag_walk_grab(ip
))
569 * Update the index for the next lookup. Catch
570 * overflows into the next AG range which can occur if
571 * we have inodes in the last block of the AG and we
572 * are currently pointing to the last inode.
574 * Because we may see inodes that are from the wrong AG
575 * due to RCU freeing and reallocation, only update the
576 * index if it lies in this AG. It was a race that lead
577 * us to see this inode, so another lookup from the
578 * same index will not find it again.
580 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
582 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
583 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
587 /* unlock now we've grabbed the inodes. */
590 for (i
= 0; i
< nr_found
; i
++) {
593 error
= execute(batch
[i
], pag
, flags
, args
);
595 if (error
== EAGAIN
) {
599 if (error
&& last_error
!= EFSCORRUPTED
)
603 /* bail out if the filesystem is corrupted. */
604 if (error
== EFSCORRUPTED
)
609 } while (nr_found
&& !done
);
619 * Background scanning to trim post-EOF preallocated space. This is queued
620 * based on the 'background_prealloc_discard_period' tunable (5m by default).
624 struct xfs_mount
*mp
)
627 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_EOFBLOCKS_TAG
))
628 queue_delayed_work(mp
->m_eofblocks_workqueue
,
629 &mp
->m_eofblocks_work
,
630 msecs_to_jiffies(xfs_eofb_secs
* 1000));
635 xfs_eofblocks_worker(
636 struct work_struct
*work
)
638 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
639 struct xfs_mount
, m_eofblocks_work
);
640 xfs_icache_free_eofblocks(mp
, NULL
);
641 xfs_queue_eofblocks(mp
);
645 xfs_inode_ag_iterator(
646 struct xfs_mount
*mp
,
647 int (*execute
)(struct xfs_inode
*ip
,
648 struct xfs_perag
*pag
, int flags
,
653 struct xfs_perag
*pag
;
659 while ((pag
= xfs_perag_get(mp
, ag
))) {
660 ag
= pag
->pag_agno
+ 1;
661 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, -1);
665 if (error
== EFSCORRUPTED
)
669 return XFS_ERROR(last_error
);
673 xfs_inode_ag_iterator_tag(
674 struct xfs_mount
*mp
,
675 int (*execute
)(struct xfs_inode
*ip
,
676 struct xfs_perag
*pag
, int flags
,
682 struct xfs_perag
*pag
;
688 while ((pag
= xfs_perag_get_tag(mp
, ag
, tag
))) {
689 ag
= pag
->pag_agno
+ 1;
690 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, tag
);
694 if (error
== EFSCORRUPTED
)
698 return XFS_ERROR(last_error
);
702 * Queue a new inode reclaim pass if there are reclaimable inodes and there
703 * isn't a reclaim pass already in progress. By default it runs every 5s based
704 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
705 * tunable, but that can be done if this method proves to be ineffective or too
709 xfs_reclaim_work_queue(
710 struct xfs_mount
*mp
)
714 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
715 queue_delayed_work(mp
->m_reclaim_workqueue
, &mp
->m_reclaim_work
,
716 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
722 * This is a fast pass over the inode cache to try to get reclaim moving on as
723 * many inodes as possible in a short period of time. It kicks itself every few
724 * seconds, as well as being kicked by the inode cache shrinker when memory
725 * goes low. It scans as quickly as possible avoiding locked inodes or those
726 * already being flushed, and once done schedules a future pass.
730 struct work_struct
*work
)
732 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
733 struct xfs_mount
, m_reclaim_work
);
735 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
736 xfs_reclaim_work_queue(mp
);
740 __xfs_inode_set_reclaim_tag(
741 struct xfs_perag
*pag
,
742 struct xfs_inode
*ip
)
744 radix_tree_tag_set(&pag
->pag_ici_root
,
745 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
746 XFS_ICI_RECLAIM_TAG
);
748 if (!pag
->pag_ici_reclaimable
) {
749 /* propagate the reclaim tag up into the perag radix tree */
750 spin_lock(&ip
->i_mount
->m_perag_lock
);
751 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
752 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
753 XFS_ICI_RECLAIM_TAG
);
754 spin_unlock(&ip
->i_mount
->m_perag_lock
);
756 /* schedule periodic background inode reclaim */
757 xfs_reclaim_work_queue(ip
->i_mount
);
759 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
762 pag
->pag_ici_reclaimable
++;
766 * We set the inode flag atomically with the radix tree tag.
767 * Once we get tag lookups on the radix tree, this inode flag
771 xfs_inode_set_reclaim_tag(
774 struct xfs_mount
*mp
= ip
->i_mount
;
775 struct xfs_perag
*pag
;
777 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
778 spin_lock(&pag
->pag_ici_lock
);
779 spin_lock(&ip
->i_flags_lock
);
780 __xfs_inode_set_reclaim_tag(pag
, ip
);
781 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
782 spin_unlock(&ip
->i_flags_lock
);
783 spin_unlock(&pag
->pag_ici_lock
);
788 __xfs_inode_clear_reclaim(
792 pag
->pag_ici_reclaimable
--;
793 if (!pag
->pag_ici_reclaimable
) {
794 /* clear the reclaim tag from the perag radix tree */
795 spin_lock(&ip
->i_mount
->m_perag_lock
);
796 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
797 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
798 XFS_ICI_RECLAIM_TAG
);
799 spin_unlock(&ip
->i_mount
->m_perag_lock
);
800 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
806 __xfs_inode_clear_reclaim_tag(
811 radix_tree_tag_clear(&pag
->pag_ici_root
,
812 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
813 __xfs_inode_clear_reclaim(pag
, ip
);
817 * Grab the inode for reclaim exclusively.
818 * Return 0 if we grabbed it, non-zero otherwise.
821 xfs_reclaim_inode_grab(
822 struct xfs_inode
*ip
,
825 ASSERT(rcu_read_lock_held());
827 /* quick check for stale RCU freed inode */
832 * If we are asked for non-blocking operation, do unlocked checks to
833 * see if the inode already is being flushed or in reclaim to avoid
836 if ((flags
& SYNC_TRYLOCK
) &&
837 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
841 * The radix tree lock here protects a thread in xfs_iget from racing
842 * with us starting reclaim on the inode. Once we have the
843 * XFS_IRECLAIM flag set it will not touch us.
845 * Due to RCU lookup, we may find inodes that have been freed and only
846 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
847 * aren't candidates for reclaim at all, so we must check the
848 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
850 spin_lock(&ip
->i_flags_lock
);
851 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
852 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
853 /* not a reclaim candidate. */
854 spin_unlock(&ip
->i_flags_lock
);
857 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
858 spin_unlock(&ip
->i_flags_lock
);
863 * Inodes in different states need to be treated differently. The following
864 * table lists the inode states and the reclaim actions necessary:
866 * inode state iflush ret required action
867 * --------------- ---------- ---------------
869 * shutdown EIO unpin and reclaim
870 * clean, unpinned 0 reclaim
871 * stale, unpinned 0 reclaim
872 * clean, pinned(*) 0 requeue
873 * stale, pinned EAGAIN requeue
874 * dirty, async - requeue
875 * dirty, sync 0 reclaim
877 * (*) dgc: I don't think the clean, pinned state is possible but it gets
878 * handled anyway given the order of checks implemented.
880 * Also, because we get the flush lock first, we know that any inode that has
881 * been flushed delwri has had the flush completed by the time we check that
882 * the inode is clean.
884 * Note that because the inode is flushed delayed write by AIL pushing, the
885 * flush lock may already be held here and waiting on it can result in very
886 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
887 * the caller should push the AIL first before trying to reclaim inodes to
888 * minimise the amount of time spent waiting. For background relaim, we only
889 * bother to reclaim clean inodes anyway.
891 * Hence the order of actions after gaining the locks should be:
893 * shutdown => unpin and reclaim
894 * pinned, async => requeue
895 * pinned, sync => unpin
898 * dirty, async => requeue
899 * dirty, sync => flush, wait and reclaim
903 struct xfs_inode
*ip
,
904 struct xfs_perag
*pag
,
907 struct xfs_buf
*bp
= NULL
;
912 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
913 if (!xfs_iflock_nowait(ip
)) {
914 if (!(sync_mode
& SYNC_WAIT
))
919 if (is_bad_inode(VFS_I(ip
)))
921 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
923 xfs_iflush_abort(ip
, false);
926 if (xfs_ipincount(ip
)) {
927 if (!(sync_mode
& SYNC_WAIT
))
931 if (xfs_iflags_test(ip
, XFS_ISTALE
))
933 if (xfs_inode_clean(ip
))
937 * Never flush out dirty data during non-blocking reclaim, as it would
938 * just contend with AIL pushing trying to do the same job.
940 if (!(sync_mode
& SYNC_WAIT
))
944 * Now we have an inode that needs flushing.
946 * Note that xfs_iflush will never block on the inode buffer lock, as
947 * xfs_ifree_cluster() can lock the inode buffer before it locks the
948 * ip->i_lock, and we are doing the exact opposite here. As a result,
949 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
950 * result in an ABBA deadlock with xfs_ifree_cluster().
952 * As xfs_ifree_cluser() must gather all inodes that are active in the
953 * cache to mark them stale, if we hit this case we don't actually want
954 * to do IO here - we want the inode marked stale so we can simply
955 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
956 * inode, back off and try again. Hopefully the next pass through will
957 * see the stale flag set on the inode.
959 error
= xfs_iflush(ip
, &bp
);
960 if (error
== EAGAIN
) {
961 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
962 /* backoff longer than in xfs_ifree_cluster */
968 error
= xfs_bwrite(bp
);
975 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
977 XFS_STATS_INC(xs_ig_reclaims
);
979 * Remove the inode from the per-AG radix tree.
981 * Because radix_tree_delete won't complain even if the item was never
982 * added to the tree assert that it's been there before to catch
983 * problems with the inode life time early on.
985 spin_lock(&pag
->pag_ici_lock
);
986 if (!radix_tree_delete(&pag
->pag_ici_root
,
987 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
989 __xfs_inode_clear_reclaim(pag
, ip
);
990 spin_unlock(&pag
->pag_ici_lock
);
993 * Here we do an (almost) spurious inode lock in order to coordinate
994 * with inode cache radix tree lookups. This is because the lookup
995 * can reference the inodes in the cache without taking references.
997 * We make that OK here by ensuring that we wait until the inode is
998 * unlocked after the lookup before we go ahead and free it.
1000 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1001 xfs_qm_dqdetach(ip
);
1002 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1010 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
1011 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1013 * We could return EAGAIN here to make reclaim rescan the inode tree in
1014 * a short while. However, this just burns CPU time scanning the tree
1015 * waiting for IO to complete and the reclaim work never goes back to
1016 * the idle state. Instead, return 0 to let the next scheduled
1017 * background reclaim attempt to reclaim the inode again.
1023 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1024 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1025 * then a shut down during filesystem unmount reclaim walk leak all the
1026 * unreclaimed inodes.
1029 xfs_reclaim_inodes_ag(
1030 struct xfs_mount
*mp
,
1034 struct xfs_perag
*pag
;
1038 int trylock
= flags
& SYNC_TRYLOCK
;
1044 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1045 unsigned long first_index
= 0;
1049 ag
= pag
->pag_agno
+ 1;
1052 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
1057 first_index
= pag
->pag_ici_reclaim_cursor
;
1059 mutex_lock(&pag
->pag_ici_reclaim_lock
);
1062 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
1066 nr_found
= radix_tree_gang_lookup_tag(
1068 (void **)batch
, first_index
,
1070 XFS_ICI_RECLAIM_TAG
);
1078 * Grab the inodes before we drop the lock. if we found
1079 * nothing, nr == 0 and the loop will be skipped.
1081 for (i
= 0; i
< nr_found
; i
++) {
1082 struct xfs_inode
*ip
= batch
[i
];
1084 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
1088 * Update the index for the next lookup. Catch
1089 * overflows into the next AG range which can
1090 * occur if we have inodes in the last block of
1091 * the AG and we are currently pointing to the
1094 * Because we may see inodes that are from the
1095 * wrong AG due to RCU freeing and
1096 * reallocation, only update the index if it
1097 * lies in this AG. It was a race that lead us
1098 * to see this inode, so another lookup from
1099 * the same index will not find it again.
1101 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
1104 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
1105 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
1109 /* unlock now we've grabbed the inodes. */
1112 for (i
= 0; i
< nr_found
; i
++) {
1115 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
1116 if (error
&& last_error
!= EFSCORRUPTED
)
1120 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
1124 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
1126 if (trylock
&& !done
)
1127 pag
->pag_ici_reclaim_cursor
= first_index
;
1129 pag
->pag_ici_reclaim_cursor
= 0;
1130 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1135 * if we skipped any AG, and we still have scan count remaining, do
1136 * another pass this time using blocking reclaim semantics (i.e
1137 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1138 * ensure that when we get more reclaimers than AGs we block rather
1139 * than spin trying to execute reclaim.
1141 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1145 return XFS_ERROR(last_error
);
1153 int nr_to_scan
= INT_MAX
;
1155 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1159 * Scan a certain number of inodes for reclaim.
1161 * When called we make sure that there is a background (fast) inode reclaim in
1162 * progress, while we will throttle the speed of reclaim via doing synchronous
1163 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1164 * them to be cleaned, which we hope will not be very long due to the
1165 * background walker having already kicked the IO off on those dirty inodes.
1168 xfs_reclaim_inodes_nr(
1169 struct xfs_mount
*mp
,
1172 /* kick background reclaimer and push the AIL */
1173 xfs_reclaim_work_queue(mp
);
1174 xfs_ail_push_all(mp
->m_ail
);
1176 xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1180 * Return the number of reclaimable inodes in the filesystem for
1181 * the shrinker to determine how much to reclaim.
1184 xfs_reclaim_inodes_count(
1185 struct xfs_mount
*mp
)
1187 struct xfs_perag
*pag
;
1188 xfs_agnumber_t ag
= 0;
1189 int reclaimable
= 0;
1191 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1192 ag
= pag
->pag_agno
+ 1;
1193 reclaimable
+= pag
->pag_ici_reclaimable
;
1201 struct xfs_inode
*ip
,
1202 struct xfs_eofblocks
*eofb
)
1204 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UID
&&
1205 ip
->i_d
.di_uid
!= eofb
->eof_uid
)
1208 if (eofb
->eof_flags
& XFS_EOF_FLAGS_GID
&&
1209 ip
->i_d
.di_gid
!= eofb
->eof_gid
)
1212 if (eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
&&
1213 xfs_get_projid(ip
) != eofb
->eof_prid
)
1220 xfs_inode_free_eofblocks(
1221 struct xfs_inode
*ip
,
1222 struct xfs_perag
*pag
,
1227 struct xfs_eofblocks
*eofb
= args
;
1229 if (!xfs_can_free_eofblocks(ip
, false)) {
1230 /* inode could be preallocated or append-only */
1231 trace_xfs_inode_free_eofblocks_invalid(ip
);
1232 xfs_inode_clear_eofblocks_tag(ip
);
1237 * If the mapping is dirty the operation can block and wait for some
1238 * time. Unless we are waiting, skip it.
1240 if (!(flags
& SYNC_WAIT
) &&
1241 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
))
1245 if (!xfs_inode_match_id(ip
, eofb
))
1248 /* skip the inode if the file size is too small */
1249 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1250 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1254 ret
= xfs_free_eofblocks(ip
->i_mount
, ip
, true);
1256 /* don't revisit the inode if we're not waiting */
1257 if (ret
== EAGAIN
&& !(flags
& SYNC_WAIT
))
1264 xfs_icache_free_eofblocks(
1265 struct xfs_mount
*mp
,
1266 struct xfs_eofblocks
*eofb
)
1268 int flags
= SYNC_TRYLOCK
;
1270 if (eofb
&& (eofb
->eof_flags
& XFS_EOF_FLAGS_SYNC
))
1273 return xfs_inode_ag_iterator_tag(mp
, xfs_inode_free_eofblocks
, flags
,
1274 eofb
, XFS_ICI_EOFBLOCKS_TAG
);
1278 xfs_inode_set_eofblocks_tag(
1281 struct xfs_mount
*mp
= ip
->i_mount
;
1282 struct xfs_perag
*pag
;
1285 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1286 spin_lock(&pag
->pag_ici_lock
);
1287 trace_xfs_inode_set_eofblocks_tag(ip
);
1289 tagged
= radix_tree_tagged(&pag
->pag_ici_root
,
1290 XFS_ICI_EOFBLOCKS_TAG
);
1291 radix_tree_tag_set(&pag
->pag_ici_root
,
1292 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
1293 XFS_ICI_EOFBLOCKS_TAG
);
1295 /* propagate the eofblocks tag up into the perag radix tree */
1296 spin_lock(&ip
->i_mount
->m_perag_lock
);
1297 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
1298 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1299 XFS_ICI_EOFBLOCKS_TAG
);
1300 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1302 /* kick off background trimming */
1303 xfs_queue_eofblocks(ip
->i_mount
);
1305 trace_xfs_perag_set_eofblocks(ip
->i_mount
, pag
->pag_agno
,
1309 spin_unlock(&pag
->pag_ici_lock
);
1314 xfs_inode_clear_eofblocks_tag(
1317 struct xfs_mount
*mp
= ip
->i_mount
;
1318 struct xfs_perag
*pag
;
1320 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1321 spin_lock(&pag
->pag_ici_lock
);
1322 trace_xfs_inode_clear_eofblocks_tag(ip
);
1324 radix_tree_tag_clear(&pag
->pag_ici_root
,
1325 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
1326 XFS_ICI_EOFBLOCKS_TAG
);
1327 if (!radix_tree_tagged(&pag
->pag_ici_root
, XFS_ICI_EOFBLOCKS_TAG
)) {
1328 /* clear the eofblocks tag from the perag radix tree */
1329 spin_lock(&ip
->i_mount
->m_perag_lock
);
1330 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
1331 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1332 XFS_ICI_EOFBLOCKS_TAG
);
1333 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1334 trace_xfs_perag_clear_eofblocks(ip
->i_mount
, pag
->pag_agno
,
1338 spin_unlock(&pag
->pag_ici_lock
);