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"
36 #include "xfs_reflink.h"
38 #include <linux/kthread.h>
39 #include <linux/freezer.h>
42 * Allocate and initialise an xfs_inode.
52 * if this didn't occur in transactions, we could use
53 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
54 * code up to do this anyway.
56 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
59 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
60 kmem_zone_free(xfs_inode_zone
, ip
);
64 /* VFS doesn't initialise i_mode! */
65 VFS_I(ip
)->i_mode
= 0;
67 XFS_STATS_INC(mp
, vn_active
);
68 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
69 ASSERT(!xfs_isiflocked(ip
));
70 ASSERT(ip
->i_ino
== 0);
72 /* initialise the xfs inode */
75 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
79 ip
->i_cformat
= XFS_DINODE_FMT_EXTENTS
;
80 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
82 ip
->i_delayed_blks
= 0;
83 memset(&ip
->i_d
, 0, sizeof(ip
->i_d
));
89 xfs_inode_free_callback(
90 struct rcu_head
*head
)
92 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
93 struct xfs_inode
*ip
= XFS_I(inode
);
95 switch (VFS_I(ip
)->i_mode
& S_IFMT
) {
99 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
104 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
106 xfs_idestroy_fork(ip
, XFS_COW_FORK
);
109 ASSERT(!(ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
));
110 xfs_inode_item_destroy(ip
);
114 kmem_zone_free(xfs_inode_zone
, ip
);
119 struct xfs_inode
*ip
)
121 /* asserts to verify all state is correct here */
122 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
123 XFS_STATS_DEC(ip
->i_mount
, vn_active
);
125 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
130 struct xfs_inode
*ip
)
132 ASSERT(!xfs_isiflocked(ip
));
135 * Because we use RCU freeing we need to ensure the inode always
136 * appears to be reclaimed with an invalid inode number when in the
137 * free state. The ip->i_flags_lock provides the barrier against lookup
140 spin_lock(&ip
->i_flags_lock
);
141 ip
->i_flags
= XFS_IRECLAIM
;
143 spin_unlock(&ip
->i_flags_lock
);
145 __xfs_inode_free(ip
);
149 * Queue a new inode reclaim pass if there are reclaimable inodes and there
150 * isn't a reclaim pass already in progress. By default it runs every 5s based
151 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
152 * tunable, but that can be done if this method proves to be ineffective or too
156 xfs_reclaim_work_queue(
157 struct xfs_mount
*mp
)
161 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
162 queue_delayed_work(mp
->m_reclaim_workqueue
, &mp
->m_reclaim_work
,
163 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
169 * This is a fast pass over the inode cache to try to get reclaim moving on as
170 * many inodes as possible in a short period of time. It kicks itself every few
171 * seconds, as well as being kicked by the inode cache shrinker when memory
172 * goes low. It scans as quickly as possible avoiding locked inodes or those
173 * already being flushed, and once done schedules a future pass.
177 struct work_struct
*work
)
179 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
180 struct xfs_mount
, m_reclaim_work
);
182 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
183 xfs_reclaim_work_queue(mp
);
187 xfs_perag_set_reclaim_tag(
188 struct xfs_perag
*pag
)
190 struct xfs_mount
*mp
= pag
->pag_mount
;
192 lockdep_assert_held(&pag
->pag_ici_lock
);
193 if (pag
->pag_ici_reclaimable
++)
196 /* propagate the reclaim tag up into the perag radix tree */
197 spin_lock(&mp
->m_perag_lock
);
198 radix_tree_tag_set(&mp
->m_perag_tree
, pag
->pag_agno
,
199 XFS_ICI_RECLAIM_TAG
);
200 spin_unlock(&mp
->m_perag_lock
);
202 /* schedule periodic background inode reclaim */
203 xfs_reclaim_work_queue(mp
);
205 trace_xfs_perag_set_reclaim(mp
, pag
->pag_agno
, -1, _RET_IP_
);
209 xfs_perag_clear_reclaim_tag(
210 struct xfs_perag
*pag
)
212 struct xfs_mount
*mp
= pag
->pag_mount
;
214 lockdep_assert_held(&pag
->pag_ici_lock
);
215 if (--pag
->pag_ici_reclaimable
)
218 /* clear the reclaim tag from the perag radix tree */
219 spin_lock(&mp
->m_perag_lock
);
220 radix_tree_tag_clear(&mp
->m_perag_tree
, pag
->pag_agno
,
221 XFS_ICI_RECLAIM_TAG
);
222 spin_unlock(&mp
->m_perag_lock
);
223 trace_xfs_perag_clear_reclaim(mp
, pag
->pag_agno
, -1, _RET_IP_
);
228 * We set the inode flag atomically with the radix tree tag.
229 * Once we get tag lookups on the radix tree, this inode flag
233 xfs_inode_set_reclaim_tag(
234 struct xfs_inode
*ip
)
236 struct xfs_mount
*mp
= ip
->i_mount
;
237 struct xfs_perag
*pag
;
239 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
240 spin_lock(&pag
->pag_ici_lock
);
241 spin_lock(&ip
->i_flags_lock
);
243 radix_tree_tag_set(&pag
->pag_ici_root
, XFS_INO_TO_AGINO(mp
, ip
->i_ino
),
244 XFS_ICI_RECLAIM_TAG
);
245 xfs_perag_set_reclaim_tag(pag
);
246 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
248 spin_unlock(&ip
->i_flags_lock
);
249 spin_unlock(&pag
->pag_ici_lock
);
254 xfs_inode_clear_reclaim_tag(
255 struct xfs_perag
*pag
,
258 radix_tree_tag_clear(&pag
->pag_ici_root
,
259 XFS_INO_TO_AGINO(pag
->pag_mount
, ino
),
260 XFS_ICI_RECLAIM_TAG
);
261 xfs_perag_clear_reclaim_tag(pag
);
266 struct xfs_inode
*ip
)
268 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_INEW_BIT
);
269 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_INEW_BIT
);
272 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
273 if (!xfs_iflags_test(ip
, XFS_INEW
))
277 finish_wait(wq
, &wait
.wq_entry
);
281 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
282 * part of the structure. This is made more complex by the fact we store
283 * information about the on-disk values in the VFS inode and so we can't just
284 * overwrite the values unconditionally. Hence we save the parameters we
285 * need to retain across reinitialisation, and rewrite them into the VFS inode
286 * after reinitialisation even if it fails.
290 struct xfs_mount
*mp
,
294 uint32_t nlink
= inode
->i_nlink
;
295 uint32_t generation
= inode
->i_generation
;
296 uint64_t version
= inode
->i_version
;
297 umode_t mode
= inode
->i_mode
;
299 error
= inode_init_always(mp
->m_super
, inode
);
301 set_nlink(inode
, nlink
);
302 inode
->i_generation
= generation
;
303 inode
->i_version
= version
;
304 inode
->i_mode
= mode
;
309 * Check the validity of the inode we just found it the cache
313 struct xfs_perag
*pag
,
314 struct xfs_inode
*ip
,
317 int lock_flags
) __releases(RCU
)
319 struct inode
*inode
= VFS_I(ip
);
320 struct xfs_mount
*mp
= ip
->i_mount
;
324 * check for re-use of an inode within an RCU grace period due to the
325 * radix tree nodes not being updated yet. We monitor for this by
326 * setting the inode number to zero before freeing the inode structure.
327 * If the inode has been reallocated and set up, then the inode number
328 * will not match, so check for that, too.
330 spin_lock(&ip
->i_flags_lock
);
331 if (ip
->i_ino
!= ino
) {
332 trace_xfs_iget_skip(ip
);
333 XFS_STATS_INC(mp
, xs_ig_frecycle
);
340 * If we are racing with another cache hit that is currently
341 * instantiating this inode or currently recycling it out of
342 * reclaimabe state, wait for the initialisation to complete
345 * XXX(hch): eventually we should do something equivalent to
346 * wait_on_inode to wait for these flags to be cleared
347 * instead of polling for it.
349 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
350 trace_xfs_iget_skip(ip
);
351 XFS_STATS_INC(mp
, xs_ig_frecycle
);
357 * If lookup is racing with unlink return an error immediately.
359 if (VFS_I(ip
)->i_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
365 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
366 * Need to carefully get it back into useable state.
368 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
369 trace_xfs_iget_reclaim(ip
);
371 if (flags
& XFS_IGET_INCORE
) {
377 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
378 * from stomping over us while we recycle the inode. We can't
379 * clear the radix tree reclaimable tag yet as it requires
380 * pag_ici_lock to be held exclusive.
382 ip
->i_flags
|= XFS_IRECLAIM
;
384 spin_unlock(&ip
->i_flags_lock
);
387 error
= xfs_reinit_inode(mp
, inode
);
391 * Re-initializing the inode failed, and we are in deep
392 * trouble. Try to re-add it to the reclaim list.
395 spin_lock(&ip
->i_flags_lock
);
396 wake
= !!__xfs_iflags_test(ip
, XFS_INEW
);
397 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
399 wake_up_bit(&ip
->i_flags
, __XFS_INEW_BIT
);
400 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
401 trace_xfs_iget_reclaim_fail(ip
);
405 spin_lock(&pag
->pag_ici_lock
);
406 spin_lock(&ip
->i_flags_lock
);
409 * Clear the per-lifetime state in the inode as we are now
410 * effectively a new inode and need to return to the initial
411 * state before reuse occurs.
413 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
414 ip
->i_flags
|= XFS_INEW
;
415 xfs_inode_clear_reclaim_tag(pag
, ip
->i_ino
);
416 inode
->i_state
= I_NEW
;
418 ASSERT(!rwsem_is_locked(&inode
->i_rwsem
));
419 init_rwsem(&inode
->i_rwsem
);
421 spin_unlock(&ip
->i_flags_lock
);
422 spin_unlock(&pag
->pag_ici_lock
);
424 /* If the VFS inode is being torn down, pause and try again. */
426 trace_xfs_iget_skip(ip
);
431 /* We've got a live one. */
432 spin_unlock(&ip
->i_flags_lock
);
434 trace_xfs_iget_hit(ip
);
438 xfs_ilock(ip
, lock_flags
);
440 if (!(flags
& XFS_IGET_INCORE
))
441 xfs_iflags_clear(ip
, XFS_ISTALE
| XFS_IDONTCACHE
);
442 XFS_STATS_INC(mp
, xs_ig_found
);
447 spin_unlock(&ip
->i_flags_lock
);
455 struct xfs_mount
*mp
,
456 struct xfs_perag
*pag
,
459 struct xfs_inode
**ipp
,
463 struct xfs_inode
*ip
;
465 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
468 ip
= xfs_inode_alloc(mp
, ino
);
472 error
= xfs_iread(mp
, tp
, ip
, flags
);
476 trace_xfs_iget_miss(ip
);
478 if ((VFS_I(ip
)->i_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
484 * Preload the radix tree so we can insert safely under the
485 * write spinlock. Note that we cannot sleep inside the preload
486 * region. Since we can be called from transaction context, don't
487 * recurse into the file system.
489 if (radix_tree_preload(GFP_NOFS
)) {
495 * Because the inode hasn't been added to the radix-tree yet it can't
496 * be found by another thread, so we can do the non-sleeping lock here.
499 if (!xfs_ilock_nowait(ip
, lock_flags
))
504 * These values must be set before inserting the inode into the radix
505 * tree as the moment it is inserted a concurrent lookup (allowed by the
506 * RCU locking mechanism) can find it and that lookup must see that this
507 * is an inode currently under construction (i.e. that XFS_INEW is set).
508 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
509 * memory barrier that ensures this detection works correctly at lookup
513 if (flags
& XFS_IGET_DONTCACHE
)
514 iflags
|= XFS_IDONTCACHE
;
518 xfs_iflags_set(ip
, iflags
);
520 /* insert the new inode */
521 spin_lock(&pag
->pag_ici_lock
);
522 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
523 if (unlikely(error
)) {
524 WARN_ON(error
!= -EEXIST
);
525 XFS_STATS_INC(mp
, xs_ig_dup
);
527 goto out_preload_end
;
529 spin_unlock(&pag
->pag_ici_lock
);
530 radix_tree_preload_end();
536 spin_unlock(&pag
->pag_ici_lock
);
537 radix_tree_preload_end();
539 xfs_iunlock(ip
, lock_flags
);
541 __destroy_inode(VFS_I(ip
));
547 * Look up an inode by number in the given file system.
548 * The inode is looked up in the cache held in each AG.
549 * If the inode is found in the cache, initialise the vfs inode
552 * If it is not in core, read it in from the file system's device,
553 * add it to the cache and initialise the vfs inode.
555 * The inode is locked according to the value of the lock_flags parameter.
556 * This flag parameter indicates how and if the inode's IO lock and inode lock
559 * mp -- the mount point structure for the current file system. It points
560 * to the inode hash table.
561 * tp -- a pointer to the current transaction if there is one. This is
562 * simply passed through to the xfs_iread() call.
563 * ino -- the number of the inode desired. This is the unique identifier
564 * within the file system for the inode being requested.
565 * lock_flags -- flags indicating how to lock the inode. See the comment
566 * for xfs_ilock() for a list of valid values.
583 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
584 * doesn't get freed while it's being referenced during a
585 * radix tree traversal here. It assumes this function
586 * aqcuires only the ILOCK (and therefore it has no need to
587 * involve the IOLOCK in this synchronization).
589 ASSERT((lock_flags
& (XFS_IOLOCK_EXCL
| XFS_IOLOCK_SHARED
)) == 0);
591 /* reject inode numbers outside existing AGs */
592 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
595 XFS_STATS_INC(mp
, xs_ig_attempts
);
597 /* get the perag structure and ensure that it's inode capable */
598 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
599 agino
= XFS_INO_TO_AGINO(mp
, ino
);
604 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
607 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
609 goto out_error_or_again
;
612 if (flags
& XFS_IGET_INCORE
) {
614 goto out_error_or_again
;
616 XFS_STATS_INC(mp
, xs_ig_missed
);
618 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
621 goto out_error_or_again
;
628 * If we have a real type for an on-disk inode, we can setup the inode
629 * now. If it's a new inode being created, xfs_ialloc will handle it.
631 if (xfs_iflags_test(ip
, XFS_INEW
) && VFS_I(ip
)->i_mode
!= 0)
632 xfs_setup_existing_inode(ip
);
636 if (!(flags
& XFS_IGET_INCORE
) && error
== -EAGAIN
) {
645 * "Is this a cached inode that's also allocated?"
647 * Look up an inode by number in the given file system. If the inode is
648 * in cache and isn't in purgatory, return 1 if the inode is allocated
649 * and 0 if it is not. For all other cases (not in cache, being torn
650 * down, etc.), return a negative error code.
652 * The caller has to prevent inode allocation and freeing activity,
653 * presumably by locking the AGI buffer. This is to ensure that an
654 * inode cannot transition from allocated to freed until the caller is
655 * ready to allow that. If the inode is in an intermediate state (new,
656 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
657 * inode is not in the cache, -ENOENT will be returned. The caller must
658 * deal with these scenarios appropriately.
660 * This is a specialized use case for the online scrubber; if you're
661 * reading this, you probably want xfs_iget.
664 xfs_icache_inode_is_allocated(
665 struct xfs_mount
*mp
,
666 struct xfs_trans
*tp
,
670 struct xfs_inode
*ip
;
673 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_INCORE
, 0, &ip
);
677 *inuse
= !!(VFS_I(ip
)->i_mode
);
683 * The inode lookup is done in batches to keep the amount of lock traffic and
684 * radix tree lookups to a minimum. The batch size is a trade off between
685 * lookup reduction and stack usage. This is in the reclaim path, so we can't
688 #define XFS_LOOKUP_BATCH 32
691 xfs_inode_ag_walk_grab(
692 struct xfs_inode
*ip
,
695 struct inode
*inode
= VFS_I(ip
);
696 bool newinos
= !!(flags
& XFS_AGITER_INEW_WAIT
);
698 ASSERT(rcu_read_lock_held());
701 * check for stale RCU freed inode
703 * If the inode has been reallocated, it doesn't matter if it's not in
704 * the AG we are walking - we are walking for writeback, so if it
705 * passes all the "valid inode" checks and is dirty, then we'll write
706 * it back anyway. If it has been reallocated and still being
707 * initialised, the XFS_INEW check below will catch it.
709 spin_lock(&ip
->i_flags_lock
);
711 goto out_unlock_noent
;
713 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
714 if ((!newinos
&& __xfs_iflags_test(ip
, XFS_INEW
)) ||
715 __xfs_iflags_test(ip
, XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
716 goto out_unlock_noent
;
717 spin_unlock(&ip
->i_flags_lock
);
719 /* nothing to sync during shutdown */
720 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
721 return -EFSCORRUPTED
;
723 /* If we can't grab the inode, it must on it's way to reclaim. */
731 spin_unlock(&ip
->i_flags_lock
);
737 struct xfs_mount
*mp
,
738 struct xfs_perag
*pag
,
739 int (*execute
)(struct xfs_inode
*ip
, int flags
,
746 uint32_t first_index
;
758 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
765 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
766 (void **)batch
, first_index
,
769 nr_found
= radix_tree_gang_lookup_tag(
771 (void **) batch
, first_index
,
772 XFS_LOOKUP_BATCH
, tag
);
780 * Grab the inodes before we drop the lock. if we found
781 * nothing, nr == 0 and the loop will be skipped.
783 for (i
= 0; i
< nr_found
; i
++) {
784 struct xfs_inode
*ip
= batch
[i
];
786 if (done
|| xfs_inode_ag_walk_grab(ip
, iter_flags
))
790 * Update the index for the next lookup. Catch
791 * overflows into the next AG range which can occur if
792 * we have inodes in the last block of the AG and we
793 * are currently pointing to the last inode.
795 * Because we may see inodes that are from the wrong AG
796 * due to RCU freeing and reallocation, only update the
797 * index if it lies in this AG. It was a race that lead
798 * us to see this inode, so another lookup from the
799 * same index will not find it again.
801 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
803 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
804 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
808 /* unlock now we've grabbed the inodes. */
811 for (i
= 0; i
< nr_found
; i
++) {
814 if ((iter_flags
& XFS_AGITER_INEW_WAIT
) &&
815 xfs_iflags_test(batch
[i
], XFS_INEW
))
816 xfs_inew_wait(batch
[i
]);
817 error
= execute(batch
[i
], flags
, args
);
819 if (error
== -EAGAIN
) {
823 if (error
&& last_error
!= -EFSCORRUPTED
)
827 /* bail out if the filesystem is corrupted. */
828 if (error
== -EFSCORRUPTED
)
833 } while (nr_found
&& !done
);
843 * Background scanning to trim post-EOF preallocated space. This is queued
844 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
848 struct xfs_mount
*mp
)
851 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_EOFBLOCKS_TAG
))
852 queue_delayed_work(mp
->m_eofblocks_workqueue
,
853 &mp
->m_eofblocks_work
,
854 msecs_to_jiffies(xfs_eofb_secs
* 1000));
859 xfs_eofblocks_worker(
860 struct work_struct
*work
)
862 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
863 struct xfs_mount
, m_eofblocks_work
);
864 xfs_icache_free_eofblocks(mp
, NULL
);
865 xfs_queue_eofblocks(mp
);
869 * Background scanning to trim preallocated CoW space. This is queued
870 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
871 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
875 struct xfs_mount
*mp
)
878 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_COWBLOCKS_TAG
))
879 queue_delayed_work(mp
->m_eofblocks_workqueue
,
880 &mp
->m_cowblocks_work
,
881 msecs_to_jiffies(xfs_cowb_secs
* 1000));
886 xfs_cowblocks_worker(
887 struct work_struct
*work
)
889 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
890 struct xfs_mount
, m_cowblocks_work
);
891 xfs_icache_free_cowblocks(mp
, NULL
);
892 xfs_queue_cowblocks(mp
);
896 xfs_inode_ag_iterator_flags(
897 struct xfs_mount
*mp
,
898 int (*execute
)(struct xfs_inode
*ip
, int flags
,
904 struct xfs_perag
*pag
;
910 while ((pag
= xfs_perag_get(mp
, ag
))) {
911 ag
= pag
->pag_agno
+ 1;
912 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, -1,
917 if (error
== -EFSCORRUPTED
)
925 xfs_inode_ag_iterator(
926 struct xfs_mount
*mp
,
927 int (*execute
)(struct xfs_inode
*ip
, int flags
,
932 return xfs_inode_ag_iterator_flags(mp
, execute
, flags
, args
, 0);
936 xfs_inode_ag_iterator_tag(
937 struct xfs_mount
*mp
,
938 int (*execute
)(struct xfs_inode
*ip
, int flags
,
944 struct xfs_perag
*pag
;
950 while ((pag
= xfs_perag_get_tag(mp
, ag
, tag
))) {
951 ag
= pag
->pag_agno
+ 1;
952 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, tag
,
957 if (error
== -EFSCORRUPTED
)
965 * Grab the inode for reclaim exclusively.
966 * Return 0 if we grabbed it, non-zero otherwise.
969 xfs_reclaim_inode_grab(
970 struct xfs_inode
*ip
,
973 ASSERT(rcu_read_lock_held());
975 /* quick check for stale RCU freed inode */
980 * If we are asked for non-blocking operation, do unlocked checks to
981 * see if the inode already is being flushed or in reclaim to avoid
984 if ((flags
& SYNC_TRYLOCK
) &&
985 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
989 * The radix tree lock here protects a thread in xfs_iget from racing
990 * with us starting reclaim on the inode. Once we have the
991 * XFS_IRECLAIM flag set it will not touch us.
993 * Due to RCU lookup, we may find inodes that have been freed and only
994 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
995 * aren't candidates for reclaim at all, so we must check the
996 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
998 spin_lock(&ip
->i_flags_lock
);
999 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
1000 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
1001 /* not a reclaim candidate. */
1002 spin_unlock(&ip
->i_flags_lock
);
1005 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
1006 spin_unlock(&ip
->i_flags_lock
);
1011 * Inodes in different states need to be treated differently. The following
1012 * table lists the inode states and the reclaim actions necessary:
1014 * inode state iflush ret required action
1015 * --------------- ---------- ---------------
1017 * shutdown EIO unpin and reclaim
1018 * clean, unpinned 0 reclaim
1019 * stale, unpinned 0 reclaim
1020 * clean, pinned(*) 0 requeue
1021 * stale, pinned EAGAIN requeue
1022 * dirty, async - requeue
1023 * dirty, sync 0 reclaim
1025 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1026 * handled anyway given the order of checks implemented.
1028 * Also, because we get the flush lock first, we know that any inode that has
1029 * been flushed delwri has had the flush completed by the time we check that
1030 * the inode is clean.
1032 * Note that because the inode is flushed delayed write by AIL pushing, the
1033 * flush lock may already be held here and waiting on it can result in very
1034 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1035 * the caller should push the AIL first before trying to reclaim inodes to
1036 * minimise the amount of time spent waiting. For background relaim, we only
1037 * bother to reclaim clean inodes anyway.
1039 * Hence the order of actions after gaining the locks should be:
1041 * shutdown => unpin and reclaim
1042 * pinned, async => requeue
1043 * pinned, sync => unpin
1046 * dirty, async => requeue
1047 * dirty, sync => flush, wait and reclaim
1051 struct xfs_inode
*ip
,
1052 struct xfs_perag
*pag
,
1055 struct xfs_buf
*bp
= NULL
;
1056 xfs_ino_t ino
= ip
->i_ino
; /* for radix_tree_delete */
1061 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1062 if (!xfs_iflock_nowait(ip
)) {
1063 if (!(sync_mode
& SYNC_WAIT
))
1068 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1069 xfs_iunpin_wait(ip
);
1070 /* xfs_iflush_abort() drops the flush lock */
1071 xfs_iflush_abort(ip
, false);
1074 if (xfs_ipincount(ip
)) {
1075 if (!(sync_mode
& SYNC_WAIT
))
1077 xfs_iunpin_wait(ip
);
1079 if (xfs_iflags_test(ip
, XFS_ISTALE
) || xfs_inode_clean(ip
)) {
1085 * Never flush out dirty data during non-blocking reclaim, as it would
1086 * just contend with AIL pushing trying to do the same job.
1088 if (!(sync_mode
& SYNC_WAIT
))
1092 * Now we have an inode that needs flushing.
1094 * Note that xfs_iflush will never block on the inode buffer lock, as
1095 * xfs_ifree_cluster() can lock the inode buffer before it locks the
1096 * ip->i_lock, and we are doing the exact opposite here. As a result,
1097 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1098 * result in an ABBA deadlock with xfs_ifree_cluster().
1100 * As xfs_ifree_cluser() must gather all inodes that are active in the
1101 * cache to mark them stale, if we hit this case we don't actually want
1102 * to do IO here - we want the inode marked stale so we can simply
1103 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1104 * inode, back off and try again. Hopefully the next pass through will
1105 * see the stale flag set on the inode.
1107 error
= xfs_iflush(ip
, &bp
);
1108 if (error
== -EAGAIN
) {
1109 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1110 /* backoff longer than in xfs_ifree_cluster */
1116 error
= xfs_bwrite(bp
);
1121 ASSERT(!xfs_isiflocked(ip
));
1124 * Because we use RCU freeing we need to ensure the inode always appears
1125 * to be reclaimed with an invalid inode number when in the free state.
1126 * We do this as early as possible under the ILOCK so that
1127 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1128 * detect races with us here. By doing this, we guarantee that once
1129 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1130 * it will see either a valid inode that will serialise correctly, or it
1131 * will see an invalid inode that it can skip.
1133 spin_lock(&ip
->i_flags_lock
);
1134 ip
->i_flags
= XFS_IRECLAIM
;
1136 spin_unlock(&ip
->i_flags_lock
);
1138 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1140 XFS_STATS_INC(ip
->i_mount
, xs_ig_reclaims
);
1142 * Remove the inode from the per-AG radix tree.
1144 * Because radix_tree_delete won't complain even if the item was never
1145 * added to the tree assert that it's been there before to catch
1146 * problems with the inode life time early on.
1148 spin_lock(&pag
->pag_ici_lock
);
1149 if (!radix_tree_delete(&pag
->pag_ici_root
,
1150 XFS_INO_TO_AGINO(ip
->i_mount
, ino
)))
1152 xfs_perag_clear_reclaim_tag(pag
);
1153 spin_unlock(&pag
->pag_ici_lock
);
1156 * Here we do an (almost) spurious inode lock in order to coordinate
1157 * with inode cache radix tree lookups. This is because the lookup
1158 * can reference the inodes in the cache without taking references.
1160 * We make that OK here by ensuring that we wait until the inode is
1161 * unlocked after the lookup before we go ahead and free it.
1163 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1164 xfs_qm_dqdetach(ip
);
1165 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1167 __xfs_inode_free(ip
);
1173 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
1174 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1176 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1177 * a short while. However, this just burns CPU time scanning the tree
1178 * waiting for IO to complete and the reclaim work never goes back to
1179 * the idle state. Instead, return 0 to let the next scheduled
1180 * background reclaim attempt to reclaim the inode again.
1186 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1187 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1188 * then a shut down during filesystem unmount reclaim walk leak all the
1189 * unreclaimed inodes.
1192 xfs_reclaim_inodes_ag(
1193 struct xfs_mount
*mp
,
1197 struct xfs_perag
*pag
;
1201 int trylock
= flags
& SYNC_TRYLOCK
;
1207 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1208 unsigned long first_index
= 0;
1212 ag
= pag
->pag_agno
+ 1;
1215 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
1220 first_index
= pag
->pag_ici_reclaim_cursor
;
1222 mutex_lock(&pag
->pag_ici_reclaim_lock
);
1225 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
1229 nr_found
= radix_tree_gang_lookup_tag(
1231 (void **)batch
, first_index
,
1233 XFS_ICI_RECLAIM_TAG
);
1241 * Grab the inodes before we drop the lock. if we found
1242 * nothing, nr == 0 and the loop will be skipped.
1244 for (i
= 0; i
< nr_found
; i
++) {
1245 struct xfs_inode
*ip
= batch
[i
];
1247 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
1251 * Update the index for the next lookup. Catch
1252 * overflows into the next AG range which can
1253 * occur if we have inodes in the last block of
1254 * the AG and we are currently pointing to the
1257 * Because we may see inodes that are from the
1258 * wrong AG due to RCU freeing and
1259 * reallocation, only update the index if it
1260 * lies in this AG. It was a race that lead us
1261 * to see this inode, so another lookup from
1262 * the same index will not find it again.
1264 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
1267 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
1268 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
1272 /* unlock now we've grabbed the inodes. */
1275 for (i
= 0; i
< nr_found
; i
++) {
1278 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
1279 if (error
&& last_error
!= -EFSCORRUPTED
)
1283 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
1287 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
1289 if (trylock
&& !done
)
1290 pag
->pag_ici_reclaim_cursor
= first_index
;
1292 pag
->pag_ici_reclaim_cursor
= 0;
1293 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1298 * if we skipped any AG, and we still have scan count remaining, do
1299 * another pass this time using blocking reclaim semantics (i.e
1300 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1301 * ensure that when we get more reclaimers than AGs we block rather
1302 * than spin trying to execute reclaim.
1304 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1316 int nr_to_scan
= INT_MAX
;
1318 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1322 * Scan a certain number of inodes for reclaim.
1324 * When called we make sure that there is a background (fast) inode reclaim in
1325 * progress, while we will throttle the speed of reclaim via doing synchronous
1326 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1327 * them to be cleaned, which we hope will not be very long due to the
1328 * background walker having already kicked the IO off on those dirty inodes.
1331 xfs_reclaim_inodes_nr(
1332 struct xfs_mount
*mp
,
1335 /* kick background reclaimer and push the AIL */
1336 xfs_reclaim_work_queue(mp
);
1337 xfs_ail_push_all(mp
->m_ail
);
1339 return xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1343 * Return the number of reclaimable inodes in the filesystem for
1344 * the shrinker to determine how much to reclaim.
1347 xfs_reclaim_inodes_count(
1348 struct xfs_mount
*mp
)
1350 struct xfs_perag
*pag
;
1351 xfs_agnumber_t ag
= 0;
1352 int reclaimable
= 0;
1354 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1355 ag
= pag
->pag_agno
+ 1;
1356 reclaimable
+= pag
->pag_ici_reclaimable
;
1364 struct xfs_inode
*ip
,
1365 struct xfs_eofblocks
*eofb
)
1367 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1368 !uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1371 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1372 !gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1375 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1376 xfs_get_projid(ip
) != eofb
->eof_prid
)
1383 * A union-based inode filtering algorithm. Process the inode if any of the
1384 * criteria match. This is for global/internal scans only.
1387 xfs_inode_match_id_union(
1388 struct xfs_inode
*ip
,
1389 struct xfs_eofblocks
*eofb
)
1391 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1392 uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1395 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1396 gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1399 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1400 xfs_get_projid(ip
) == eofb
->eof_prid
)
1407 xfs_inode_free_eofblocks(
1408 struct xfs_inode
*ip
,
1413 struct xfs_eofblocks
*eofb
= args
;
1416 if (!xfs_can_free_eofblocks(ip
, false)) {
1417 /* inode could be preallocated or append-only */
1418 trace_xfs_inode_free_eofblocks_invalid(ip
);
1419 xfs_inode_clear_eofblocks_tag(ip
);
1424 * If the mapping is dirty the operation can block and wait for some
1425 * time. Unless we are waiting, skip it.
1427 if (!(flags
& SYNC_WAIT
) &&
1428 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
))
1432 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1433 match
= xfs_inode_match_id_union(ip
, eofb
);
1435 match
= xfs_inode_match_id(ip
, eofb
);
1439 /* skip the inode if the file size is too small */
1440 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1441 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1446 * If the caller is waiting, return -EAGAIN to keep the background
1447 * scanner moving and revisit the inode in a subsequent pass.
1449 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
1450 if (flags
& SYNC_WAIT
)
1454 ret
= xfs_free_eofblocks(ip
);
1455 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1461 __xfs_icache_free_eofblocks(
1462 struct xfs_mount
*mp
,
1463 struct xfs_eofblocks
*eofb
,
1464 int (*execute
)(struct xfs_inode
*ip
, int flags
,
1468 int flags
= SYNC_TRYLOCK
;
1470 if (eofb
&& (eofb
->eof_flags
& XFS_EOF_FLAGS_SYNC
))
1473 return xfs_inode_ag_iterator_tag(mp
, execute
, flags
,
1478 xfs_icache_free_eofblocks(
1479 struct xfs_mount
*mp
,
1480 struct xfs_eofblocks
*eofb
)
1482 return __xfs_icache_free_eofblocks(mp
, eofb
, xfs_inode_free_eofblocks
,
1483 XFS_ICI_EOFBLOCKS_TAG
);
1487 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1488 * multiple quotas, we don't know exactly which quota caused an allocation
1489 * failure. We make a best effort by including each quota under low free space
1490 * conditions (less than 1% free space) in the scan.
1493 __xfs_inode_free_quota_eofblocks(
1494 struct xfs_inode
*ip
,
1495 int (*execute
)(struct xfs_mount
*mp
,
1496 struct xfs_eofblocks
*eofb
))
1499 struct xfs_eofblocks eofb
= {0};
1500 struct xfs_dquot
*dq
;
1503 * Run a sync scan to increase effectiveness and use the union filter to
1504 * cover all applicable quotas in a single scan.
1506 eofb
.eof_flags
= XFS_EOF_FLAGS_UNION
|XFS_EOF_FLAGS_SYNC
;
1508 if (XFS_IS_UQUOTA_ENFORCED(ip
->i_mount
)) {
1509 dq
= xfs_inode_dquot(ip
, XFS_DQ_USER
);
1510 if (dq
&& xfs_dquot_lowsp(dq
)) {
1511 eofb
.eof_uid
= VFS_I(ip
)->i_uid
;
1512 eofb
.eof_flags
|= XFS_EOF_FLAGS_UID
;
1517 if (XFS_IS_GQUOTA_ENFORCED(ip
->i_mount
)) {
1518 dq
= xfs_inode_dquot(ip
, XFS_DQ_GROUP
);
1519 if (dq
&& xfs_dquot_lowsp(dq
)) {
1520 eofb
.eof_gid
= VFS_I(ip
)->i_gid
;
1521 eofb
.eof_flags
|= XFS_EOF_FLAGS_GID
;
1527 execute(ip
->i_mount
, &eofb
);
1533 xfs_inode_free_quota_eofblocks(
1534 struct xfs_inode
*ip
)
1536 return __xfs_inode_free_quota_eofblocks(ip
, xfs_icache_free_eofblocks
);
1540 __xfs_inode_set_eofblocks_tag(
1542 void (*execute
)(struct xfs_mount
*mp
),
1543 void (*set_tp
)(struct xfs_mount
*mp
, xfs_agnumber_t agno
,
1544 int error
, unsigned long caller_ip
),
1547 struct xfs_mount
*mp
= ip
->i_mount
;
1548 struct xfs_perag
*pag
;
1552 * Don't bother locking the AG and looking up in the radix trees
1553 * if we already know that we have the tag set.
1555 if (ip
->i_flags
& XFS_IEOFBLOCKS
)
1557 spin_lock(&ip
->i_flags_lock
);
1558 ip
->i_flags
|= XFS_IEOFBLOCKS
;
1559 spin_unlock(&ip
->i_flags_lock
);
1561 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1562 spin_lock(&pag
->pag_ici_lock
);
1564 tagged
= radix_tree_tagged(&pag
->pag_ici_root
, tag
);
1565 radix_tree_tag_set(&pag
->pag_ici_root
,
1566 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
), tag
);
1568 /* propagate the eofblocks tag up into the perag radix tree */
1569 spin_lock(&ip
->i_mount
->m_perag_lock
);
1570 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
1571 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1573 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1575 /* kick off background trimming */
1576 execute(ip
->i_mount
);
1578 set_tp(ip
->i_mount
, pag
->pag_agno
, -1, _RET_IP_
);
1581 spin_unlock(&pag
->pag_ici_lock
);
1586 xfs_inode_set_eofblocks_tag(
1589 trace_xfs_inode_set_eofblocks_tag(ip
);
1590 return __xfs_inode_set_eofblocks_tag(ip
, xfs_queue_eofblocks
,
1591 trace_xfs_perag_set_eofblocks
,
1592 XFS_ICI_EOFBLOCKS_TAG
);
1596 __xfs_inode_clear_eofblocks_tag(
1598 void (*clear_tp
)(struct xfs_mount
*mp
, xfs_agnumber_t agno
,
1599 int error
, unsigned long caller_ip
),
1602 struct xfs_mount
*mp
= ip
->i_mount
;
1603 struct xfs_perag
*pag
;
1605 spin_lock(&ip
->i_flags_lock
);
1606 ip
->i_flags
&= ~XFS_IEOFBLOCKS
;
1607 spin_unlock(&ip
->i_flags_lock
);
1609 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1610 spin_lock(&pag
->pag_ici_lock
);
1612 radix_tree_tag_clear(&pag
->pag_ici_root
,
1613 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
), tag
);
1614 if (!radix_tree_tagged(&pag
->pag_ici_root
, tag
)) {
1615 /* clear the eofblocks tag from the perag radix tree */
1616 spin_lock(&ip
->i_mount
->m_perag_lock
);
1617 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
1618 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1620 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1621 clear_tp(ip
->i_mount
, pag
->pag_agno
, -1, _RET_IP_
);
1624 spin_unlock(&pag
->pag_ici_lock
);
1629 xfs_inode_clear_eofblocks_tag(
1632 trace_xfs_inode_clear_eofblocks_tag(ip
);
1633 return __xfs_inode_clear_eofblocks_tag(ip
,
1634 trace_xfs_perag_clear_eofblocks
, XFS_ICI_EOFBLOCKS_TAG
);
1638 * Automatic CoW Reservation Freeing
1640 * These functions automatically garbage collect leftover CoW reservations
1641 * that were made on behalf of a cowextsize hint when we start to run out
1642 * of quota or when the reservations sit around for too long. If the file
1643 * has dirty pages or is undergoing writeback, its CoW reservations will
1646 * The actual garbage collection piggybacks off the same code that runs
1647 * the speculative EOF preallocation garbage collector.
1650 xfs_inode_free_cowblocks(
1651 struct xfs_inode
*ip
,
1656 struct xfs_eofblocks
*eofb
= args
;
1658 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
1661 * Just clear the tag if we have an empty cow fork or none at all. It's
1662 * possible the inode was fully unshared since it was originally tagged.
1664 if (!xfs_is_reflink_inode(ip
) || !ifp
->if_bytes
) {
1665 trace_xfs_inode_free_cowblocks_invalid(ip
);
1666 xfs_inode_clear_cowblocks_tag(ip
);
1671 * If the mapping is dirty or under writeback we cannot touch the
1672 * CoW fork. Leave it alone if we're in the midst of a directio.
1674 if ((VFS_I(ip
)->i_state
& I_DIRTY_PAGES
) ||
1675 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
) ||
1676 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_WRITEBACK
) ||
1677 atomic_read(&VFS_I(ip
)->i_dio_count
))
1681 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1682 match
= xfs_inode_match_id_union(ip
, eofb
);
1684 match
= xfs_inode_match_id(ip
, eofb
);
1688 /* skip the inode if the file size is too small */
1689 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1690 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1694 /* Free the CoW blocks */
1695 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1696 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
1698 ret
= xfs_reflink_cancel_cow_range(ip
, 0, NULLFILEOFF
, false);
1700 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1701 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1707 xfs_icache_free_cowblocks(
1708 struct xfs_mount
*mp
,
1709 struct xfs_eofblocks
*eofb
)
1711 return __xfs_icache_free_eofblocks(mp
, eofb
, xfs_inode_free_cowblocks
,
1712 XFS_ICI_COWBLOCKS_TAG
);
1716 xfs_inode_free_quota_cowblocks(
1717 struct xfs_inode
*ip
)
1719 return __xfs_inode_free_quota_eofblocks(ip
, xfs_icache_free_cowblocks
);
1723 xfs_inode_set_cowblocks_tag(
1726 trace_xfs_inode_set_cowblocks_tag(ip
);
1727 return __xfs_inode_set_eofblocks_tag(ip
, xfs_queue_cowblocks
,
1728 trace_xfs_perag_set_cowblocks
,
1729 XFS_ICI_COWBLOCKS_TAG
);
1733 xfs_inode_clear_cowblocks_tag(
1736 trace_xfs_inode_clear_cowblocks_tag(ip
);
1737 return __xfs_inode_clear_eofblocks_tag(ip
,
1738 trace_xfs_perag_clear_cowblocks
, XFS_ICI_COWBLOCKS_TAG
);