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_log_format.h"
22 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_inode.h"
26 #include "xfs_error.h"
27 #include "xfs_trans.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_inode_item.h"
30 #include "xfs_quota.h"
31 #include "xfs_trace.h"
32 #include "xfs_icache.h"
33 #include "xfs_bmap_util.h"
34 #include "xfs_dquot_item.h"
35 #include "xfs_dquot.h"
37 #include <linux/kthread.h>
38 #include <linux/freezer.h>
40 STATIC
void __xfs_inode_clear_reclaim_tag(struct xfs_mount
*mp
,
41 struct xfs_perag
*pag
, struct xfs_inode
*ip
);
44 * Allocate and initialise an xfs_inode.
54 * if this didn't occur in transactions, we could use
55 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
56 * code up to do this anyway.
58 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
61 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
62 kmem_zone_free(xfs_inode_zone
, ip
);
66 XFS_STATS_INC(mp
, vn_active
);
67 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
68 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
69 ASSERT(!xfs_isiflocked(ip
));
70 ASSERT(ip
->i_ino
== 0);
72 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
74 /* initialise the xfs inode */
77 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
79 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
81 ip
->i_delayed_blks
= 0;
82 memset(&ip
->i_d
, 0, sizeof(ip
->i_d
));
88 xfs_inode_free_callback(
89 struct rcu_head
*head
)
91 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
92 struct xfs_inode
*ip
= XFS_I(inode
);
94 kmem_zone_free(xfs_inode_zone
, ip
);
101 switch (ip
->i_d
.di_mode
& S_IFMT
) {
105 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
110 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
113 ASSERT(!(ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
));
114 xfs_inode_item_destroy(ip
);
119 * Because we use RCU freeing we need to ensure the inode always
120 * appears to be reclaimed with an invalid inode number when in the
121 * free state. The ip->i_flags_lock provides the barrier against lookup
124 spin_lock(&ip
->i_flags_lock
);
125 ip
->i_flags
= XFS_IRECLAIM
;
127 spin_unlock(&ip
->i_flags_lock
);
129 /* asserts to verify all state is correct here */
130 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
131 ASSERT(!xfs_isiflocked(ip
));
132 XFS_STATS_DEC(ip
->i_mount
, vn_active
);
134 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
138 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
139 * part of the structure. This is made more complex by the fact we store
140 * information about the on-disk values in the VFS inode and so we can't just
141 * overwrite it's values unconditionally. Hence we save the parameters we
142 * need to retain across reinitialisation, and rewrite them into the VFS inode
143 * after resetting it's state even if resetting fails.
147 struct xfs_mount
*mp
,
151 uint32_t nlink
= inode
->i_nlink
;
152 uint32_t generation
= inode
->i_generation
;
154 error
= inode_init_always(mp
->m_super
, inode
);
156 set_nlink(inode
, nlink
);
157 inode
->i_generation
= generation
;
162 * Check the validity of the inode we just found it the cache
166 struct xfs_perag
*pag
,
167 struct xfs_inode
*ip
,
170 int lock_flags
) __releases(RCU
)
172 struct inode
*inode
= VFS_I(ip
);
173 struct xfs_mount
*mp
= ip
->i_mount
;
177 * check for re-use of an inode within an RCU grace period due to the
178 * radix tree nodes not being updated yet. We monitor for this by
179 * setting the inode number to zero before freeing the inode structure.
180 * If the inode has been reallocated and set up, then the inode number
181 * will not match, so check for that, too.
183 spin_lock(&ip
->i_flags_lock
);
184 if (ip
->i_ino
!= ino
) {
185 trace_xfs_iget_skip(ip
);
186 XFS_STATS_INC(mp
, xs_ig_frecycle
);
193 * If we are racing with another cache hit that is currently
194 * instantiating this inode or currently recycling it out of
195 * reclaimabe state, wait for the initialisation to complete
198 * XXX(hch): eventually we should do something equivalent to
199 * wait_on_inode to wait for these flags to be cleared
200 * instead of polling for it.
202 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
203 trace_xfs_iget_skip(ip
);
204 XFS_STATS_INC(mp
, xs_ig_frecycle
);
210 * If lookup is racing with unlink return an error immediately.
212 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
218 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
219 * Need to carefully get it back into useable state.
221 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
222 trace_xfs_iget_reclaim(ip
);
225 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
226 * from stomping over us while we recycle the inode. We can't
227 * clear the radix tree reclaimable tag yet as it requires
228 * pag_ici_lock to be held exclusive.
230 ip
->i_flags
|= XFS_IRECLAIM
;
232 spin_unlock(&ip
->i_flags_lock
);
235 error
= xfs_reinit_inode(mp
, inode
);
238 * Re-initializing the inode failed, and we are in deep
239 * trouble. Try to re-add it to the reclaim list.
242 spin_lock(&ip
->i_flags_lock
);
244 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
245 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
246 trace_xfs_iget_reclaim_fail(ip
);
250 spin_lock(&pag
->pag_ici_lock
);
251 spin_lock(&ip
->i_flags_lock
);
254 * Clear the per-lifetime state in the inode as we are now
255 * effectively a new inode and need to return to the initial
256 * state before reuse occurs.
258 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
259 ip
->i_flags
|= XFS_INEW
;
260 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
261 inode
->i_state
= I_NEW
;
263 ASSERT(!rwsem_is_locked(&ip
->i_iolock
.mr_lock
));
264 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
266 spin_unlock(&ip
->i_flags_lock
);
267 spin_unlock(&pag
->pag_ici_lock
);
269 /* If the VFS inode is being torn down, pause and try again. */
271 trace_xfs_iget_skip(ip
);
276 /* We've got a live one. */
277 spin_unlock(&ip
->i_flags_lock
);
279 trace_xfs_iget_hit(ip
);
283 xfs_ilock(ip
, lock_flags
);
285 xfs_iflags_clear(ip
, XFS_ISTALE
| XFS_IDONTCACHE
);
286 XFS_STATS_INC(mp
, xs_ig_found
);
291 spin_unlock(&ip
->i_flags_lock
);
299 struct xfs_mount
*mp
,
300 struct xfs_perag
*pag
,
303 struct xfs_inode
**ipp
,
307 struct xfs_inode
*ip
;
309 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
312 ip
= xfs_inode_alloc(mp
, ino
);
316 error
= xfs_iread(mp
, tp
, ip
, flags
);
320 trace_xfs_iget_miss(ip
);
322 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
328 * Preload the radix tree so we can insert safely under the
329 * write spinlock. Note that we cannot sleep inside the preload
330 * region. Since we can be called from transaction context, don't
331 * recurse into the file system.
333 if (radix_tree_preload(GFP_NOFS
)) {
339 * Because the inode hasn't been added to the radix-tree yet it can't
340 * be found by another thread, so we can do the non-sleeping lock here.
343 if (!xfs_ilock_nowait(ip
, lock_flags
))
348 * These values must be set before inserting the inode into the radix
349 * tree as the moment it is inserted a concurrent lookup (allowed by the
350 * RCU locking mechanism) can find it and that lookup must see that this
351 * is an inode currently under construction (i.e. that XFS_INEW is set).
352 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
353 * memory barrier that ensures this detection works correctly at lookup
357 if (flags
& XFS_IGET_DONTCACHE
)
358 iflags
|= XFS_IDONTCACHE
;
362 xfs_iflags_set(ip
, iflags
);
364 /* insert the new inode */
365 spin_lock(&pag
->pag_ici_lock
);
366 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
367 if (unlikely(error
)) {
368 WARN_ON(error
!= -EEXIST
);
369 XFS_STATS_INC(mp
, xs_ig_dup
);
371 goto out_preload_end
;
373 spin_unlock(&pag
->pag_ici_lock
);
374 radix_tree_preload_end();
380 spin_unlock(&pag
->pag_ici_lock
);
381 radix_tree_preload_end();
383 xfs_iunlock(ip
, lock_flags
);
385 __destroy_inode(VFS_I(ip
));
391 * Look up an inode by number in the given file system.
392 * The inode is looked up in the cache held in each AG.
393 * If the inode is found in the cache, initialise the vfs inode
396 * If it is not in core, read it in from the file system's device,
397 * add it to the cache and initialise the vfs inode.
399 * The inode is locked according to the value of the lock_flags parameter.
400 * This flag parameter indicates how and if the inode's IO lock and inode lock
403 * mp -- the mount point structure for the current file system. It points
404 * to the inode hash table.
405 * tp -- a pointer to the current transaction if there is one. This is
406 * simply passed through to the xfs_iread() call.
407 * ino -- the number of the inode desired. This is the unique identifier
408 * within the file system for the inode being requested.
409 * lock_flags -- flags indicating how to lock the inode. See the comment
410 * for xfs_ilock() for a list of valid values.
427 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
428 * doesn't get freed while it's being referenced during a
429 * radix tree traversal here. It assumes this function
430 * aqcuires only the ILOCK (and therefore it has no need to
431 * involve the IOLOCK in this synchronization).
433 ASSERT((lock_flags
& (XFS_IOLOCK_EXCL
| XFS_IOLOCK_SHARED
)) == 0);
435 /* reject inode numbers outside existing AGs */
436 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
439 XFS_STATS_INC(mp
, xs_ig_attempts
);
441 /* get the perag structure and ensure that it's inode capable */
442 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
443 agino
= XFS_INO_TO_AGINO(mp
, ino
);
448 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
451 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
453 goto out_error_or_again
;
456 XFS_STATS_INC(mp
, xs_ig_missed
);
458 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
461 goto out_error_or_again
;
468 * If we have a real type for an on-disk inode, we can setup the inode
469 * now. If it's a new inode being created, xfs_ialloc will handle it.
471 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
472 xfs_setup_existing_inode(ip
);
476 if (error
== -EAGAIN
) {
485 * The inode lookup is done in batches to keep the amount of lock traffic and
486 * radix tree lookups to a minimum. The batch size is a trade off between
487 * lookup reduction and stack usage. This is in the reclaim path, so we can't
490 #define XFS_LOOKUP_BATCH 32
493 xfs_inode_ag_walk_grab(
494 struct xfs_inode
*ip
)
496 struct inode
*inode
= VFS_I(ip
);
498 ASSERT(rcu_read_lock_held());
501 * check for stale RCU freed inode
503 * If the inode has been reallocated, it doesn't matter if it's not in
504 * the AG we are walking - we are walking for writeback, so if it
505 * passes all the "valid inode" checks and is dirty, then we'll write
506 * it back anyway. If it has been reallocated and still being
507 * initialised, the XFS_INEW check below will catch it.
509 spin_lock(&ip
->i_flags_lock
);
511 goto out_unlock_noent
;
513 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
514 if (__xfs_iflags_test(ip
, XFS_INEW
| XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
515 goto out_unlock_noent
;
516 spin_unlock(&ip
->i_flags_lock
);
518 /* nothing to sync during shutdown */
519 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
520 return -EFSCORRUPTED
;
522 /* If we can't grab the inode, it must on it's way to reclaim. */
530 spin_unlock(&ip
->i_flags_lock
);
536 struct xfs_mount
*mp
,
537 struct xfs_perag
*pag
,
538 int (*execute
)(struct xfs_inode
*ip
, int flags
,
544 uint32_t first_index
;
556 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
563 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
564 (void **)batch
, first_index
,
567 nr_found
= radix_tree_gang_lookup_tag(
569 (void **) batch
, first_index
,
570 XFS_LOOKUP_BATCH
, tag
);
578 * Grab the inodes before we drop the lock. if we found
579 * nothing, nr == 0 and the loop will be skipped.
581 for (i
= 0; i
< nr_found
; i
++) {
582 struct xfs_inode
*ip
= batch
[i
];
584 if (done
|| xfs_inode_ag_walk_grab(ip
))
588 * Update the index for the next lookup. Catch
589 * overflows into the next AG range which can occur if
590 * we have inodes in the last block of the AG and we
591 * are currently pointing to the last inode.
593 * Because we may see inodes that are from the wrong AG
594 * due to RCU freeing and reallocation, only update the
595 * index if it lies in this AG. It was a race that lead
596 * us to see this inode, so another lookup from the
597 * same index will not find it again.
599 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
601 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
602 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
606 /* unlock now we've grabbed the inodes. */
609 for (i
= 0; i
< nr_found
; i
++) {
612 error
= execute(batch
[i
], flags
, args
);
614 if (error
== -EAGAIN
) {
618 if (error
&& last_error
!= -EFSCORRUPTED
)
622 /* bail out if the filesystem is corrupted. */
623 if (error
== -EFSCORRUPTED
)
628 } while (nr_found
&& !done
);
638 * Background scanning to trim post-EOF preallocated space. This is queued
639 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
643 struct xfs_mount
*mp
)
646 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_EOFBLOCKS_TAG
))
647 queue_delayed_work(mp
->m_eofblocks_workqueue
,
648 &mp
->m_eofblocks_work
,
649 msecs_to_jiffies(xfs_eofb_secs
* 1000));
654 xfs_eofblocks_worker(
655 struct work_struct
*work
)
657 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
658 struct xfs_mount
, m_eofblocks_work
);
659 xfs_icache_free_eofblocks(mp
, NULL
);
660 xfs_queue_eofblocks(mp
);
664 xfs_inode_ag_iterator(
665 struct xfs_mount
*mp
,
666 int (*execute
)(struct xfs_inode
*ip
, int flags
,
671 struct xfs_perag
*pag
;
677 while ((pag
= xfs_perag_get(mp
, ag
))) {
678 ag
= pag
->pag_agno
+ 1;
679 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, -1);
683 if (error
== -EFSCORRUPTED
)
691 xfs_inode_ag_iterator_tag(
692 struct xfs_mount
*mp
,
693 int (*execute
)(struct xfs_inode
*ip
, int flags
,
699 struct xfs_perag
*pag
;
705 while ((pag
= xfs_perag_get_tag(mp
, ag
, tag
))) {
706 ag
= pag
->pag_agno
+ 1;
707 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, tag
);
711 if (error
== -EFSCORRUPTED
)
719 * Queue a new inode reclaim pass if there are reclaimable inodes and there
720 * isn't a reclaim pass already in progress. By default it runs every 5s based
721 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
722 * tunable, but that can be done if this method proves to be ineffective or too
726 xfs_reclaim_work_queue(
727 struct xfs_mount
*mp
)
731 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
732 queue_delayed_work(mp
->m_reclaim_workqueue
, &mp
->m_reclaim_work
,
733 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
739 * This is a fast pass over the inode cache to try to get reclaim moving on as
740 * many inodes as possible in a short period of time. It kicks itself every few
741 * seconds, as well as being kicked by the inode cache shrinker when memory
742 * goes low. It scans as quickly as possible avoiding locked inodes or those
743 * already being flushed, and once done schedules a future pass.
747 struct work_struct
*work
)
749 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
750 struct xfs_mount
, m_reclaim_work
);
752 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
753 xfs_reclaim_work_queue(mp
);
757 __xfs_inode_set_reclaim_tag(
758 struct xfs_perag
*pag
,
759 struct xfs_inode
*ip
)
761 radix_tree_tag_set(&pag
->pag_ici_root
,
762 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
763 XFS_ICI_RECLAIM_TAG
);
765 if (!pag
->pag_ici_reclaimable
) {
766 /* propagate the reclaim tag up into the perag radix tree */
767 spin_lock(&ip
->i_mount
->m_perag_lock
);
768 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
769 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
770 XFS_ICI_RECLAIM_TAG
);
771 spin_unlock(&ip
->i_mount
->m_perag_lock
);
773 /* schedule periodic background inode reclaim */
774 xfs_reclaim_work_queue(ip
->i_mount
);
776 trace_xfs_perag_set_reclaim(ip
->i_mount
, pag
->pag_agno
,
779 pag
->pag_ici_reclaimable
++;
783 * We set the inode flag atomically with the radix tree tag.
784 * Once we get tag lookups on the radix tree, this inode flag
788 xfs_inode_set_reclaim_tag(
791 struct xfs_mount
*mp
= ip
->i_mount
;
792 struct xfs_perag
*pag
;
794 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
795 spin_lock(&pag
->pag_ici_lock
);
796 spin_lock(&ip
->i_flags_lock
);
797 __xfs_inode_set_reclaim_tag(pag
, ip
);
798 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
799 spin_unlock(&ip
->i_flags_lock
);
800 spin_unlock(&pag
->pag_ici_lock
);
805 __xfs_inode_clear_reclaim(
809 pag
->pag_ici_reclaimable
--;
810 if (!pag
->pag_ici_reclaimable
) {
811 /* clear the reclaim tag from the perag radix tree */
812 spin_lock(&ip
->i_mount
->m_perag_lock
);
813 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
814 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
815 XFS_ICI_RECLAIM_TAG
);
816 spin_unlock(&ip
->i_mount
->m_perag_lock
);
817 trace_xfs_perag_clear_reclaim(ip
->i_mount
, pag
->pag_agno
,
823 __xfs_inode_clear_reclaim_tag(
828 radix_tree_tag_clear(&pag
->pag_ici_root
,
829 XFS_INO_TO_AGINO(mp
, ip
->i_ino
), XFS_ICI_RECLAIM_TAG
);
830 __xfs_inode_clear_reclaim(pag
, ip
);
834 * Grab the inode for reclaim exclusively.
835 * Return 0 if we grabbed it, non-zero otherwise.
838 xfs_reclaim_inode_grab(
839 struct xfs_inode
*ip
,
842 ASSERT(rcu_read_lock_held());
844 /* quick check for stale RCU freed inode */
849 * If we are asked for non-blocking operation, do unlocked checks to
850 * see if the inode already is being flushed or in reclaim to avoid
853 if ((flags
& SYNC_TRYLOCK
) &&
854 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
858 * The radix tree lock here protects a thread in xfs_iget from racing
859 * with us starting reclaim on the inode. Once we have the
860 * XFS_IRECLAIM flag set it will not touch us.
862 * Due to RCU lookup, we may find inodes that have been freed and only
863 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
864 * aren't candidates for reclaim at all, so we must check the
865 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
867 spin_lock(&ip
->i_flags_lock
);
868 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
869 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
870 /* not a reclaim candidate. */
871 spin_unlock(&ip
->i_flags_lock
);
874 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
875 spin_unlock(&ip
->i_flags_lock
);
880 * Inodes in different states need to be treated differently. The following
881 * table lists the inode states and the reclaim actions necessary:
883 * inode state iflush ret required action
884 * --------------- ---------- ---------------
886 * shutdown EIO unpin and reclaim
887 * clean, unpinned 0 reclaim
888 * stale, unpinned 0 reclaim
889 * clean, pinned(*) 0 requeue
890 * stale, pinned EAGAIN requeue
891 * dirty, async - requeue
892 * dirty, sync 0 reclaim
894 * (*) dgc: I don't think the clean, pinned state is possible but it gets
895 * handled anyway given the order of checks implemented.
897 * Also, because we get the flush lock first, we know that any inode that has
898 * been flushed delwri has had the flush completed by the time we check that
899 * the inode is clean.
901 * Note that because the inode is flushed delayed write by AIL pushing, the
902 * flush lock may already be held here and waiting on it can result in very
903 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
904 * the caller should push the AIL first before trying to reclaim inodes to
905 * minimise the amount of time spent waiting. For background relaim, we only
906 * bother to reclaim clean inodes anyway.
908 * Hence the order of actions after gaining the locks should be:
910 * shutdown => unpin and reclaim
911 * pinned, async => requeue
912 * pinned, sync => unpin
915 * dirty, async => requeue
916 * dirty, sync => flush, wait and reclaim
920 struct xfs_inode
*ip
,
921 struct xfs_perag
*pag
,
924 struct xfs_buf
*bp
= NULL
;
929 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
930 if (!xfs_iflock_nowait(ip
)) {
931 if (!(sync_mode
& SYNC_WAIT
))
936 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
938 xfs_iflush_abort(ip
, false);
941 if (xfs_ipincount(ip
)) {
942 if (!(sync_mode
& SYNC_WAIT
))
946 if (xfs_iflags_test(ip
, XFS_ISTALE
))
948 if (xfs_inode_clean(ip
))
952 * Never flush out dirty data during non-blocking reclaim, as it would
953 * just contend with AIL pushing trying to do the same job.
955 if (!(sync_mode
& SYNC_WAIT
))
959 * Now we have an inode that needs flushing.
961 * Note that xfs_iflush will never block on the inode buffer lock, as
962 * xfs_ifree_cluster() can lock the inode buffer before it locks the
963 * ip->i_lock, and we are doing the exact opposite here. As a result,
964 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
965 * result in an ABBA deadlock with xfs_ifree_cluster().
967 * As xfs_ifree_cluser() must gather all inodes that are active in the
968 * cache to mark them stale, if we hit this case we don't actually want
969 * to do IO here - we want the inode marked stale so we can simply
970 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
971 * inode, back off and try again. Hopefully the next pass through will
972 * see the stale flag set on the inode.
974 error
= xfs_iflush(ip
, &bp
);
975 if (error
== -EAGAIN
) {
976 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
977 /* backoff longer than in xfs_ifree_cluster */
983 error
= xfs_bwrite(bp
);
990 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
992 XFS_STATS_INC(ip
->i_mount
, xs_ig_reclaims
);
994 * Remove the inode from the per-AG radix tree.
996 * Because radix_tree_delete won't complain even if the item was never
997 * added to the tree assert that it's been there before to catch
998 * problems with the inode life time early on.
1000 spin_lock(&pag
->pag_ici_lock
);
1001 if (!radix_tree_delete(&pag
->pag_ici_root
,
1002 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
)))
1004 __xfs_inode_clear_reclaim(pag
, ip
);
1005 spin_unlock(&pag
->pag_ici_lock
);
1008 * Here we do an (almost) spurious inode lock in order to coordinate
1009 * with inode cache radix tree lookups. This is because the lookup
1010 * can reference the inodes in the cache without taking references.
1012 * We make that OK here by ensuring that we wait until the inode is
1013 * unlocked after the lookup before we go ahead and free it.
1015 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1016 xfs_qm_dqdetach(ip
);
1017 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1025 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
1026 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1028 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1029 * a short while. However, this just burns CPU time scanning the tree
1030 * waiting for IO to complete and the reclaim work never goes back to
1031 * the idle state. Instead, return 0 to let the next scheduled
1032 * background reclaim attempt to reclaim the inode again.
1038 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1039 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1040 * then a shut down during filesystem unmount reclaim walk leak all the
1041 * unreclaimed inodes.
1044 xfs_reclaim_inodes_ag(
1045 struct xfs_mount
*mp
,
1049 struct xfs_perag
*pag
;
1053 int trylock
= flags
& SYNC_TRYLOCK
;
1059 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1060 unsigned long first_index
= 0;
1064 ag
= pag
->pag_agno
+ 1;
1067 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
1072 first_index
= pag
->pag_ici_reclaim_cursor
;
1074 mutex_lock(&pag
->pag_ici_reclaim_lock
);
1077 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
1081 nr_found
= radix_tree_gang_lookup_tag(
1083 (void **)batch
, first_index
,
1085 XFS_ICI_RECLAIM_TAG
);
1093 * Grab the inodes before we drop the lock. if we found
1094 * nothing, nr == 0 and the loop will be skipped.
1096 for (i
= 0; i
< nr_found
; i
++) {
1097 struct xfs_inode
*ip
= batch
[i
];
1099 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
1103 * Update the index for the next lookup. Catch
1104 * overflows into the next AG range which can
1105 * occur if we have inodes in the last block of
1106 * the AG and we are currently pointing to the
1109 * Because we may see inodes that are from the
1110 * wrong AG due to RCU freeing and
1111 * reallocation, only update the index if it
1112 * lies in this AG. It was a race that lead us
1113 * to see this inode, so another lookup from
1114 * the same index will not find it again.
1116 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
1119 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
1120 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
1124 /* unlock now we've grabbed the inodes. */
1127 for (i
= 0; i
< nr_found
; i
++) {
1130 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
1131 if (error
&& last_error
!= -EFSCORRUPTED
)
1135 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
1139 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
1141 if (trylock
&& !done
)
1142 pag
->pag_ici_reclaim_cursor
= first_index
;
1144 pag
->pag_ici_reclaim_cursor
= 0;
1145 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1150 * if we skipped any AG, and we still have scan count remaining, do
1151 * another pass this time using blocking reclaim semantics (i.e
1152 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1153 * ensure that when we get more reclaimers than AGs we block rather
1154 * than spin trying to execute reclaim.
1156 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1168 int nr_to_scan
= INT_MAX
;
1170 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1174 * Scan a certain number of inodes for reclaim.
1176 * When called we make sure that there is a background (fast) inode reclaim in
1177 * progress, while we will throttle the speed of reclaim via doing synchronous
1178 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1179 * them to be cleaned, which we hope will not be very long due to the
1180 * background walker having already kicked the IO off on those dirty inodes.
1183 xfs_reclaim_inodes_nr(
1184 struct xfs_mount
*mp
,
1187 /* kick background reclaimer and push the AIL */
1188 xfs_reclaim_work_queue(mp
);
1189 xfs_ail_push_all(mp
->m_ail
);
1191 return xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1195 * Return the number of reclaimable inodes in the filesystem for
1196 * the shrinker to determine how much to reclaim.
1199 xfs_reclaim_inodes_count(
1200 struct xfs_mount
*mp
)
1202 struct xfs_perag
*pag
;
1203 xfs_agnumber_t ag
= 0;
1204 int reclaimable
= 0;
1206 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1207 ag
= pag
->pag_agno
+ 1;
1208 reclaimable
+= pag
->pag_ici_reclaimable
;
1216 struct xfs_inode
*ip
,
1217 struct xfs_eofblocks
*eofb
)
1219 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1220 !uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1223 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1224 !gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1227 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1228 xfs_get_projid(ip
) != eofb
->eof_prid
)
1235 * A union-based inode filtering algorithm. Process the inode if any of the
1236 * criteria match. This is for global/internal scans only.
1239 xfs_inode_match_id_union(
1240 struct xfs_inode
*ip
,
1241 struct xfs_eofblocks
*eofb
)
1243 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1244 uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1247 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1248 gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1251 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1252 xfs_get_projid(ip
) == eofb
->eof_prid
)
1259 xfs_inode_free_eofblocks(
1260 struct xfs_inode
*ip
,
1265 struct xfs_eofblocks
*eofb
= args
;
1266 bool need_iolock
= true;
1269 ASSERT(!eofb
|| (eofb
&& eofb
->eof_scan_owner
!= 0));
1271 if (!xfs_can_free_eofblocks(ip
, false)) {
1272 /* inode could be preallocated or append-only */
1273 trace_xfs_inode_free_eofblocks_invalid(ip
);
1274 xfs_inode_clear_eofblocks_tag(ip
);
1279 * If the mapping is dirty the operation can block and wait for some
1280 * time. Unless we are waiting, skip it.
1282 if (!(flags
& SYNC_WAIT
) &&
1283 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
))
1287 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1288 match
= xfs_inode_match_id_union(ip
, eofb
);
1290 match
= xfs_inode_match_id(ip
, eofb
);
1294 /* skip the inode if the file size is too small */
1295 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1296 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1300 * A scan owner implies we already hold the iolock. Skip it in
1301 * xfs_free_eofblocks() to avoid deadlock. This also eliminates
1302 * the possibility of EAGAIN being returned.
1304 if (eofb
->eof_scan_owner
== ip
->i_ino
)
1305 need_iolock
= false;
1308 ret
= xfs_free_eofblocks(ip
->i_mount
, ip
, need_iolock
);
1310 /* don't revisit the inode if we're not waiting */
1311 if (ret
== -EAGAIN
&& !(flags
& SYNC_WAIT
))
1318 xfs_icache_free_eofblocks(
1319 struct xfs_mount
*mp
,
1320 struct xfs_eofblocks
*eofb
)
1322 int flags
= SYNC_TRYLOCK
;
1324 if (eofb
&& (eofb
->eof_flags
& XFS_EOF_FLAGS_SYNC
))
1327 return xfs_inode_ag_iterator_tag(mp
, xfs_inode_free_eofblocks
, flags
,
1328 eofb
, XFS_ICI_EOFBLOCKS_TAG
);
1332 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1333 * multiple quotas, we don't know exactly which quota caused an allocation
1334 * failure. We make a best effort by including each quota under low free space
1335 * conditions (less than 1% free space) in the scan.
1338 xfs_inode_free_quota_eofblocks(
1339 struct xfs_inode
*ip
)
1342 struct xfs_eofblocks eofb
= {0};
1343 struct xfs_dquot
*dq
;
1345 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1348 * Set the scan owner to avoid a potential livelock. Otherwise, the scan
1349 * can repeatedly trylock on the inode we're currently processing. We
1350 * run a sync scan to increase effectiveness and use the union filter to
1351 * cover all applicable quotas in a single scan.
1353 eofb
.eof_scan_owner
= ip
->i_ino
;
1354 eofb
.eof_flags
= XFS_EOF_FLAGS_UNION
|XFS_EOF_FLAGS_SYNC
;
1356 if (XFS_IS_UQUOTA_ENFORCED(ip
->i_mount
)) {
1357 dq
= xfs_inode_dquot(ip
, XFS_DQ_USER
);
1358 if (dq
&& xfs_dquot_lowsp(dq
)) {
1359 eofb
.eof_uid
= VFS_I(ip
)->i_uid
;
1360 eofb
.eof_flags
|= XFS_EOF_FLAGS_UID
;
1365 if (XFS_IS_GQUOTA_ENFORCED(ip
->i_mount
)) {
1366 dq
= xfs_inode_dquot(ip
, XFS_DQ_GROUP
);
1367 if (dq
&& xfs_dquot_lowsp(dq
)) {
1368 eofb
.eof_gid
= VFS_I(ip
)->i_gid
;
1369 eofb
.eof_flags
|= XFS_EOF_FLAGS_GID
;
1375 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
1381 xfs_inode_set_eofblocks_tag(
1384 struct xfs_mount
*mp
= ip
->i_mount
;
1385 struct xfs_perag
*pag
;
1388 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1389 spin_lock(&pag
->pag_ici_lock
);
1390 trace_xfs_inode_set_eofblocks_tag(ip
);
1392 tagged
= radix_tree_tagged(&pag
->pag_ici_root
,
1393 XFS_ICI_EOFBLOCKS_TAG
);
1394 radix_tree_tag_set(&pag
->pag_ici_root
,
1395 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
1396 XFS_ICI_EOFBLOCKS_TAG
);
1398 /* propagate the eofblocks tag up into the perag radix tree */
1399 spin_lock(&ip
->i_mount
->m_perag_lock
);
1400 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
1401 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1402 XFS_ICI_EOFBLOCKS_TAG
);
1403 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1405 /* kick off background trimming */
1406 xfs_queue_eofblocks(ip
->i_mount
);
1408 trace_xfs_perag_set_eofblocks(ip
->i_mount
, pag
->pag_agno
,
1412 spin_unlock(&pag
->pag_ici_lock
);
1417 xfs_inode_clear_eofblocks_tag(
1420 struct xfs_mount
*mp
= ip
->i_mount
;
1421 struct xfs_perag
*pag
;
1423 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1424 spin_lock(&pag
->pag_ici_lock
);
1425 trace_xfs_inode_clear_eofblocks_tag(ip
);
1427 radix_tree_tag_clear(&pag
->pag_ici_root
,
1428 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
),
1429 XFS_ICI_EOFBLOCKS_TAG
);
1430 if (!radix_tree_tagged(&pag
->pag_ici_root
, XFS_ICI_EOFBLOCKS_TAG
)) {
1431 /* clear the eofblocks tag from the perag radix tree */
1432 spin_lock(&ip
->i_mount
->m_perag_lock
);
1433 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
1434 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1435 XFS_ICI_EOFBLOCKS_TAG
);
1436 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1437 trace_xfs_perag_clear_eofblocks(ip
->i_mount
, pag
->pag_agno
,
1441 spin_unlock(&pag
->pag_ici_lock
);