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
;
298 dev_t dev
= inode
->i_rdev
;
300 error
= inode_init_always(mp
->m_super
, inode
);
302 set_nlink(inode
, nlink
);
303 inode
->i_generation
= generation
;
304 inode
->i_version
= version
;
305 inode
->i_mode
= mode
;
311 * Check the validity of the inode we just found it the cache
315 struct xfs_perag
*pag
,
316 struct xfs_inode
*ip
,
319 int lock_flags
) __releases(RCU
)
321 struct inode
*inode
= VFS_I(ip
);
322 struct xfs_mount
*mp
= ip
->i_mount
;
326 * check for re-use of an inode within an RCU grace period due to the
327 * radix tree nodes not being updated yet. We monitor for this by
328 * setting the inode number to zero before freeing the inode structure.
329 * If the inode has been reallocated and set up, then the inode number
330 * will not match, so check for that, too.
332 spin_lock(&ip
->i_flags_lock
);
333 if (ip
->i_ino
!= ino
) {
334 trace_xfs_iget_skip(ip
);
335 XFS_STATS_INC(mp
, xs_ig_frecycle
);
342 * If we are racing with another cache hit that is currently
343 * instantiating this inode or currently recycling it out of
344 * reclaimabe state, wait for the initialisation to complete
347 * XXX(hch): eventually we should do something equivalent to
348 * wait_on_inode to wait for these flags to be cleared
349 * instead of polling for it.
351 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
352 trace_xfs_iget_skip(ip
);
353 XFS_STATS_INC(mp
, xs_ig_frecycle
);
359 * If lookup is racing with unlink return an error immediately.
361 if (VFS_I(ip
)->i_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
367 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
368 * Need to carefully get it back into useable state.
370 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
371 trace_xfs_iget_reclaim(ip
);
373 if (flags
& XFS_IGET_INCORE
) {
379 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
380 * from stomping over us while we recycle the inode. We can't
381 * clear the radix tree reclaimable tag yet as it requires
382 * pag_ici_lock to be held exclusive.
384 ip
->i_flags
|= XFS_IRECLAIM
;
386 spin_unlock(&ip
->i_flags_lock
);
389 error
= xfs_reinit_inode(mp
, inode
);
393 * Re-initializing the inode failed, and we are in deep
394 * trouble. Try to re-add it to the reclaim list.
397 spin_lock(&ip
->i_flags_lock
);
398 wake
= !!__xfs_iflags_test(ip
, XFS_INEW
);
399 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
401 wake_up_bit(&ip
->i_flags
, __XFS_INEW_BIT
);
402 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
403 trace_xfs_iget_reclaim_fail(ip
);
407 spin_lock(&pag
->pag_ici_lock
);
408 spin_lock(&ip
->i_flags_lock
);
411 * Clear the per-lifetime state in the inode as we are now
412 * effectively a new inode and need to return to the initial
413 * state before reuse occurs.
415 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
416 ip
->i_flags
|= XFS_INEW
;
417 xfs_inode_clear_reclaim_tag(pag
, ip
->i_ino
);
418 inode
->i_state
= I_NEW
;
420 ASSERT(!rwsem_is_locked(&inode
->i_rwsem
));
421 init_rwsem(&inode
->i_rwsem
);
423 spin_unlock(&ip
->i_flags_lock
);
424 spin_unlock(&pag
->pag_ici_lock
);
426 /* If the VFS inode is being torn down, pause and try again. */
428 trace_xfs_iget_skip(ip
);
433 /* We've got a live one. */
434 spin_unlock(&ip
->i_flags_lock
);
436 trace_xfs_iget_hit(ip
);
440 xfs_ilock(ip
, lock_flags
);
442 if (!(flags
& XFS_IGET_INCORE
))
443 xfs_iflags_clear(ip
, XFS_ISTALE
| XFS_IDONTCACHE
);
444 XFS_STATS_INC(mp
, xs_ig_found
);
449 spin_unlock(&ip
->i_flags_lock
);
457 struct xfs_mount
*mp
,
458 struct xfs_perag
*pag
,
461 struct xfs_inode
**ipp
,
465 struct xfs_inode
*ip
;
467 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
470 ip
= xfs_inode_alloc(mp
, ino
);
474 error
= xfs_iread(mp
, tp
, ip
, flags
);
478 trace_xfs_iget_miss(ip
);
482 * If we are allocating a new inode, then check what was returned is
483 * actually a free, empty inode. If we are not allocating an inode,
484 * the check we didn't find a free inode.
486 if (flags
& XFS_IGET_CREATE
) {
487 if (VFS_I(ip
)->i_mode
!= 0) {
489 "Corruption detected! Free inode 0x%llx not marked free on disk",
491 error
= -EFSCORRUPTED
;
494 if (ip
->i_d
.di_nblocks
!= 0) {
496 "Corruption detected! Free inode 0x%llx has blocks allocated!",
498 error
= -EFSCORRUPTED
;
501 } else if (VFS_I(ip
)->i_mode
== 0) {
507 * Preload the radix tree so we can insert safely under the
508 * write spinlock. Note that we cannot sleep inside the preload
509 * region. Since we can be called from transaction context, don't
510 * recurse into the file system.
512 if (radix_tree_preload(GFP_NOFS
)) {
518 * Because the inode hasn't been added to the radix-tree yet it can't
519 * be found by another thread, so we can do the non-sleeping lock here.
522 if (!xfs_ilock_nowait(ip
, lock_flags
))
527 * These values must be set before inserting the inode into the radix
528 * tree as the moment it is inserted a concurrent lookup (allowed by the
529 * RCU locking mechanism) can find it and that lookup must see that this
530 * is an inode currently under construction (i.e. that XFS_INEW is set).
531 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
532 * memory barrier that ensures this detection works correctly at lookup
536 if (flags
& XFS_IGET_DONTCACHE
)
537 iflags
|= XFS_IDONTCACHE
;
541 xfs_iflags_set(ip
, iflags
);
543 /* insert the new inode */
544 spin_lock(&pag
->pag_ici_lock
);
545 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
546 if (unlikely(error
)) {
547 WARN_ON(error
!= -EEXIST
);
548 XFS_STATS_INC(mp
, xs_ig_dup
);
550 goto out_preload_end
;
552 spin_unlock(&pag
->pag_ici_lock
);
553 radix_tree_preload_end();
559 spin_unlock(&pag
->pag_ici_lock
);
560 radix_tree_preload_end();
562 xfs_iunlock(ip
, lock_flags
);
564 __destroy_inode(VFS_I(ip
));
570 * Look up an inode by number in the given file system.
571 * The inode is looked up in the cache held in each AG.
572 * If the inode is found in the cache, initialise the vfs inode
575 * If it is not in core, read it in from the file system's device,
576 * add it to the cache and initialise the vfs inode.
578 * The inode is locked according to the value of the lock_flags parameter.
579 * This flag parameter indicates how and if the inode's IO lock and inode lock
582 * mp -- the mount point structure for the current file system. It points
583 * to the inode hash table.
584 * tp -- a pointer to the current transaction if there is one. This is
585 * simply passed through to the xfs_iread() call.
586 * ino -- the number of the inode desired. This is the unique identifier
587 * within the file system for the inode being requested.
588 * lock_flags -- flags indicating how to lock the inode. See the comment
589 * for xfs_ilock() for a list of valid values.
606 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
607 * doesn't get freed while it's being referenced during a
608 * radix tree traversal here. It assumes this function
609 * aqcuires only the ILOCK (and therefore it has no need to
610 * involve the IOLOCK in this synchronization).
612 ASSERT((lock_flags
& (XFS_IOLOCK_EXCL
| XFS_IOLOCK_SHARED
)) == 0);
614 /* reject inode numbers outside existing AGs */
615 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
618 XFS_STATS_INC(mp
, xs_ig_attempts
);
620 /* get the perag structure and ensure that it's inode capable */
621 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
622 agino
= XFS_INO_TO_AGINO(mp
, ino
);
627 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
630 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
632 goto out_error_or_again
;
635 if (flags
& XFS_IGET_INCORE
) {
637 goto out_error_or_again
;
639 XFS_STATS_INC(mp
, xs_ig_missed
);
641 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
644 goto out_error_or_again
;
651 * If we have a real type for an on-disk inode, we can setup the inode
652 * now. If it's a new inode being created, xfs_ialloc will handle it.
654 if (xfs_iflags_test(ip
, XFS_INEW
) && VFS_I(ip
)->i_mode
!= 0)
655 xfs_setup_existing_inode(ip
);
659 if (!(flags
& XFS_IGET_INCORE
) && error
== -EAGAIN
) {
668 * "Is this a cached inode that's also allocated?"
670 * Look up an inode by number in the given file system. If the inode is
671 * in cache and isn't in purgatory, return 1 if the inode is allocated
672 * and 0 if it is not. For all other cases (not in cache, being torn
673 * down, etc.), return a negative error code.
675 * The caller has to prevent inode allocation and freeing activity,
676 * presumably by locking the AGI buffer. This is to ensure that an
677 * inode cannot transition from allocated to freed until the caller is
678 * ready to allow that. If the inode is in an intermediate state (new,
679 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
680 * inode is not in the cache, -ENOENT will be returned. The caller must
681 * deal with these scenarios appropriately.
683 * This is a specialized use case for the online scrubber; if you're
684 * reading this, you probably want xfs_iget.
687 xfs_icache_inode_is_allocated(
688 struct xfs_mount
*mp
,
689 struct xfs_trans
*tp
,
693 struct xfs_inode
*ip
;
696 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_INCORE
, 0, &ip
);
700 *inuse
= !!(VFS_I(ip
)->i_mode
);
706 * The inode lookup is done in batches to keep the amount of lock traffic and
707 * radix tree lookups to a minimum. The batch size is a trade off between
708 * lookup reduction and stack usage. This is in the reclaim path, so we can't
711 #define XFS_LOOKUP_BATCH 32
714 xfs_inode_ag_walk_grab(
715 struct xfs_inode
*ip
,
718 struct inode
*inode
= VFS_I(ip
);
719 bool newinos
= !!(flags
& XFS_AGITER_INEW_WAIT
);
721 ASSERT(rcu_read_lock_held());
724 * check for stale RCU freed inode
726 * If the inode has been reallocated, it doesn't matter if it's not in
727 * the AG we are walking - we are walking for writeback, so if it
728 * passes all the "valid inode" checks and is dirty, then we'll write
729 * it back anyway. If it has been reallocated and still being
730 * initialised, the XFS_INEW check below will catch it.
732 spin_lock(&ip
->i_flags_lock
);
734 goto out_unlock_noent
;
736 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
737 if ((!newinos
&& __xfs_iflags_test(ip
, XFS_INEW
)) ||
738 __xfs_iflags_test(ip
, XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
739 goto out_unlock_noent
;
740 spin_unlock(&ip
->i_flags_lock
);
742 /* nothing to sync during shutdown */
743 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
744 return -EFSCORRUPTED
;
746 /* If we can't grab the inode, it must on it's way to reclaim. */
754 spin_unlock(&ip
->i_flags_lock
);
760 struct xfs_mount
*mp
,
761 struct xfs_perag
*pag
,
762 int (*execute
)(struct xfs_inode
*ip
, int flags
,
769 uint32_t first_index
;
781 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
788 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
789 (void **)batch
, first_index
,
792 nr_found
= radix_tree_gang_lookup_tag(
794 (void **) batch
, first_index
,
795 XFS_LOOKUP_BATCH
, tag
);
803 * Grab the inodes before we drop the lock. if we found
804 * nothing, nr == 0 and the loop will be skipped.
806 for (i
= 0; i
< nr_found
; i
++) {
807 struct xfs_inode
*ip
= batch
[i
];
809 if (done
|| xfs_inode_ag_walk_grab(ip
, iter_flags
))
813 * Update the index for the next lookup. Catch
814 * overflows into the next AG range which can occur if
815 * we have inodes in the last block of the AG and we
816 * are currently pointing to the last inode.
818 * Because we may see inodes that are from the wrong AG
819 * due to RCU freeing and reallocation, only update the
820 * index if it lies in this AG. It was a race that lead
821 * us to see this inode, so another lookup from the
822 * same index will not find it again.
824 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
826 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
827 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
831 /* unlock now we've grabbed the inodes. */
834 for (i
= 0; i
< nr_found
; i
++) {
837 if ((iter_flags
& XFS_AGITER_INEW_WAIT
) &&
838 xfs_iflags_test(batch
[i
], XFS_INEW
))
839 xfs_inew_wait(batch
[i
]);
840 error
= execute(batch
[i
], flags
, args
);
842 if (error
== -EAGAIN
) {
846 if (error
&& last_error
!= -EFSCORRUPTED
)
850 /* bail out if the filesystem is corrupted. */
851 if (error
== -EFSCORRUPTED
)
856 } while (nr_found
&& !done
);
866 * Background scanning to trim post-EOF preallocated space. This is queued
867 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
871 struct xfs_mount
*mp
)
874 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_EOFBLOCKS_TAG
))
875 queue_delayed_work(mp
->m_eofblocks_workqueue
,
876 &mp
->m_eofblocks_work
,
877 msecs_to_jiffies(xfs_eofb_secs
* 1000));
882 xfs_eofblocks_worker(
883 struct work_struct
*work
)
885 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
886 struct xfs_mount
, m_eofblocks_work
);
887 xfs_icache_free_eofblocks(mp
, NULL
);
888 xfs_queue_eofblocks(mp
);
892 * Background scanning to trim preallocated CoW space. This is queued
893 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
894 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
898 struct xfs_mount
*mp
)
901 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_COWBLOCKS_TAG
))
902 queue_delayed_work(mp
->m_eofblocks_workqueue
,
903 &mp
->m_cowblocks_work
,
904 msecs_to_jiffies(xfs_cowb_secs
* 1000));
909 xfs_cowblocks_worker(
910 struct work_struct
*work
)
912 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
913 struct xfs_mount
, m_cowblocks_work
);
914 xfs_icache_free_cowblocks(mp
, NULL
);
915 xfs_queue_cowblocks(mp
);
919 xfs_inode_ag_iterator_flags(
920 struct xfs_mount
*mp
,
921 int (*execute
)(struct xfs_inode
*ip
, int flags
,
927 struct xfs_perag
*pag
;
933 while ((pag
= xfs_perag_get(mp
, ag
))) {
934 ag
= pag
->pag_agno
+ 1;
935 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, -1,
940 if (error
== -EFSCORRUPTED
)
948 xfs_inode_ag_iterator(
949 struct xfs_mount
*mp
,
950 int (*execute
)(struct xfs_inode
*ip
, int flags
,
955 return xfs_inode_ag_iterator_flags(mp
, execute
, flags
, args
, 0);
959 xfs_inode_ag_iterator_tag(
960 struct xfs_mount
*mp
,
961 int (*execute
)(struct xfs_inode
*ip
, int flags
,
967 struct xfs_perag
*pag
;
973 while ((pag
= xfs_perag_get_tag(mp
, ag
, tag
))) {
974 ag
= pag
->pag_agno
+ 1;
975 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, tag
,
980 if (error
== -EFSCORRUPTED
)
988 * Grab the inode for reclaim exclusively.
989 * Return 0 if we grabbed it, non-zero otherwise.
992 xfs_reclaim_inode_grab(
993 struct xfs_inode
*ip
,
996 ASSERT(rcu_read_lock_held());
998 /* quick check for stale RCU freed inode */
1003 * If we are asked for non-blocking operation, do unlocked checks to
1004 * see if the inode already is being flushed or in reclaim to avoid
1007 if ((flags
& SYNC_TRYLOCK
) &&
1008 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
1012 * The radix tree lock here protects a thread in xfs_iget from racing
1013 * with us starting reclaim on the inode. Once we have the
1014 * XFS_IRECLAIM flag set it will not touch us.
1016 * Due to RCU lookup, we may find inodes that have been freed and only
1017 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
1018 * aren't candidates for reclaim at all, so we must check the
1019 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
1021 spin_lock(&ip
->i_flags_lock
);
1022 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
1023 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
1024 /* not a reclaim candidate. */
1025 spin_unlock(&ip
->i_flags_lock
);
1028 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
1029 spin_unlock(&ip
->i_flags_lock
);
1034 * Inodes in different states need to be treated differently. The following
1035 * table lists the inode states and the reclaim actions necessary:
1037 * inode state iflush ret required action
1038 * --------------- ---------- ---------------
1040 * shutdown EIO unpin and reclaim
1041 * clean, unpinned 0 reclaim
1042 * stale, unpinned 0 reclaim
1043 * clean, pinned(*) 0 requeue
1044 * stale, pinned EAGAIN requeue
1045 * dirty, async - requeue
1046 * dirty, sync 0 reclaim
1048 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1049 * handled anyway given the order of checks implemented.
1051 * Also, because we get the flush lock first, we know that any inode that has
1052 * been flushed delwri has had the flush completed by the time we check that
1053 * the inode is clean.
1055 * Note that because the inode is flushed delayed write by AIL pushing, the
1056 * flush lock may already be held here and waiting on it can result in very
1057 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1058 * the caller should push the AIL first before trying to reclaim inodes to
1059 * minimise the amount of time spent waiting. For background relaim, we only
1060 * bother to reclaim clean inodes anyway.
1062 * Hence the order of actions after gaining the locks should be:
1064 * shutdown => unpin and reclaim
1065 * pinned, async => requeue
1066 * pinned, sync => unpin
1069 * dirty, async => requeue
1070 * dirty, sync => flush, wait and reclaim
1074 struct xfs_inode
*ip
,
1075 struct xfs_perag
*pag
,
1078 struct xfs_buf
*bp
= NULL
;
1079 xfs_ino_t ino
= ip
->i_ino
; /* for radix_tree_delete */
1084 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1085 if (!xfs_iflock_nowait(ip
)) {
1086 if (!(sync_mode
& SYNC_WAIT
))
1091 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1092 xfs_iunpin_wait(ip
);
1093 /* xfs_iflush_abort() drops the flush lock */
1094 xfs_iflush_abort(ip
, false);
1097 if (xfs_ipincount(ip
)) {
1098 if (!(sync_mode
& SYNC_WAIT
))
1100 xfs_iunpin_wait(ip
);
1102 if (xfs_iflags_test(ip
, XFS_ISTALE
) || xfs_inode_clean(ip
)) {
1108 * Never flush out dirty data during non-blocking reclaim, as it would
1109 * just contend with AIL pushing trying to do the same job.
1111 if (!(sync_mode
& SYNC_WAIT
))
1115 * Now we have an inode that needs flushing.
1117 * Note that xfs_iflush will never block on the inode buffer lock, as
1118 * xfs_ifree_cluster() can lock the inode buffer before it locks the
1119 * ip->i_lock, and we are doing the exact opposite here. As a result,
1120 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1121 * result in an ABBA deadlock with xfs_ifree_cluster().
1123 * As xfs_ifree_cluser() must gather all inodes that are active in the
1124 * cache to mark them stale, if we hit this case we don't actually want
1125 * to do IO here - we want the inode marked stale so we can simply
1126 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1127 * inode, back off and try again. Hopefully the next pass through will
1128 * see the stale flag set on the inode.
1130 error
= xfs_iflush(ip
, &bp
);
1131 if (error
== -EAGAIN
) {
1132 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1133 /* backoff longer than in xfs_ifree_cluster */
1139 error
= xfs_bwrite(bp
);
1144 ASSERT(!xfs_isiflocked(ip
));
1147 * Because we use RCU freeing we need to ensure the inode always appears
1148 * to be reclaimed with an invalid inode number when in the free state.
1149 * We do this as early as possible under the ILOCK so that
1150 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1151 * detect races with us here. By doing this, we guarantee that once
1152 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1153 * it will see either a valid inode that will serialise correctly, or it
1154 * will see an invalid inode that it can skip.
1156 spin_lock(&ip
->i_flags_lock
);
1157 ip
->i_flags
= XFS_IRECLAIM
;
1159 spin_unlock(&ip
->i_flags_lock
);
1161 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1163 XFS_STATS_INC(ip
->i_mount
, xs_ig_reclaims
);
1165 * Remove the inode from the per-AG radix tree.
1167 * Because radix_tree_delete won't complain even if the item was never
1168 * added to the tree assert that it's been there before to catch
1169 * problems with the inode life time early on.
1171 spin_lock(&pag
->pag_ici_lock
);
1172 if (!radix_tree_delete(&pag
->pag_ici_root
,
1173 XFS_INO_TO_AGINO(ip
->i_mount
, ino
)))
1175 xfs_perag_clear_reclaim_tag(pag
);
1176 spin_unlock(&pag
->pag_ici_lock
);
1179 * Here we do an (almost) spurious inode lock in order to coordinate
1180 * with inode cache radix tree lookups. This is because the lookup
1181 * can reference the inodes in the cache without taking references.
1183 * We make that OK here by ensuring that we wait until the inode is
1184 * unlocked after the lookup before we go ahead and free it.
1186 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1187 xfs_qm_dqdetach(ip
);
1188 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1190 __xfs_inode_free(ip
);
1196 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
1197 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1199 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1200 * a short while. However, this just burns CPU time scanning the tree
1201 * waiting for IO to complete and the reclaim work never goes back to
1202 * the idle state. Instead, return 0 to let the next scheduled
1203 * background reclaim attempt to reclaim the inode again.
1209 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1210 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1211 * then a shut down during filesystem unmount reclaim walk leak all the
1212 * unreclaimed inodes.
1215 xfs_reclaim_inodes_ag(
1216 struct xfs_mount
*mp
,
1220 struct xfs_perag
*pag
;
1224 int trylock
= flags
& SYNC_TRYLOCK
;
1230 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1231 unsigned long first_index
= 0;
1235 ag
= pag
->pag_agno
+ 1;
1238 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
1243 first_index
= pag
->pag_ici_reclaim_cursor
;
1245 mutex_lock(&pag
->pag_ici_reclaim_lock
);
1248 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
1252 nr_found
= radix_tree_gang_lookup_tag(
1254 (void **)batch
, first_index
,
1256 XFS_ICI_RECLAIM_TAG
);
1264 * Grab the inodes before we drop the lock. if we found
1265 * nothing, nr == 0 and the loop will be skipped.
1267 for (i
= 0; i
< nr_found
; i
++) {
1268 struct xfs_inode
*ip
= batch
[i
];
1270 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
1274 * Update the index for the next lookup. Catch
1275 * overflows into the next AG range which can
1276 * occur if we have inodes in the last block of
1277 * the AG and we are currently pointing to the
1280 * Because we may see inodes that are from the
1281 * wrong AG due to RCU freeing and
1282 * reallocation, only update the index if it
1283 * lies in this AG. It was a race that lead us
1284 * to see this inode, so another lookup from
1285 * the same index will not find it again.
1287 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
1290 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
1291 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
1295 /* unlock now we've grabbed the inodes. */
1298 for (i
= 0; i
< nr_found
; i
++) {
1301 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
1302 if (error
&& last_error
!= -EFSCORRUPTED
)
1306 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
1310 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
1312 if (trylock
&& !done
)
1313 pag
->pag_ici_reclaim_cursor
= first_index
;
1315 pag
->pag_ici_reclaim_cursor
= 0;
1316 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1321 * if we skipped any AG, and we still have scan count remaining, do
1322 * another pass this time using blocking reclaim semantics (i.e
1323 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1324 * ensure that when we get more reclaimers than AGs we block rather
1325 * than spin trying to execute reclaim.
1327 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1339 int nr_to_scan
= INT_MAX
;
1341 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1345 * Scan a certain number of inodes for reclaim.
1347 * When called we make sure that there is a background (fast) inode reclaim in
1348 * progress, while we will throttle the speed of reclaim via doing synchronous
1349 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1350 * them to be cleaned, which we hope will not be very long due to the
1351 * background walker having already kicked the IO off on those dirty inodes.
1354 xfs_reclaim_inodes_nr(
1355 struct xfs_mount
*mp
,
1358 /* kick background reclaimer and push the AIL */
1359 xfs_reclaim_work_queue(mp
);
1360 xfs_ail_push_all(mp
->m_ail
);
1362 return xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1366 * Return the number of reclaimable inodes in the filesystem for
1367 * the shrinker to determine how much to reclaim.
1370 xfs_reclaim_inodes_count(
1371 struct xfs_mount
*mp
)
1373 struct xfs_perag
*pag
;
1374 xfs_agnumber_t ag
= 0;
1375 int reclaimable
= 0;
1377 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1378 ag
= pag
->pag_agno
+ 1;
1379 reclaimable
+= pag
->pag_ici_reclaimable
;
1387 struct xfs_inode
*ip
,
1388 struct xfs_eofblocks
*eofb
)
1390 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1391 !uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1394 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1395 !gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1398 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1399 xfs_get_projid(ip
) != eofb
->eof_prid
)
1406 * A union-based inode filtering algorithm. Process the inode if any of the
1407 * criteria match. This is for global/internal scans only.
1410 xfs_inode_match_id_union(
1411 struct xfs_inode
*ip
,
1412 struct xfs_eofblocks
*eofb
)
1414 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1415 uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1418 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1419 gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1422 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1423 xfs_get_projid(ip
) == eofb
->eof_prid
)
1430 xfs_inode_free_eofblocks(
1431 struct xfs_inode
*ip
,
1436 struct xfs_eofblocks
*eofb
= args
;
1439 if (!xfs_can_free_eofblocks(ip
, false)) {
1440 /* inode could be preallocated or append-only */
1441 trace_xfs_inode_free_eofblocks_invalid(ip
);
1442 xfs_inode_clear_eofblocks_tag(ip
);
1447 * If the mapping is dirty the operation can block and wait for some
1448 * time. Unless we are waiting, skip it.
1450 if (!(flags
& SYNC_WAIT
) &&
1451 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
))
1455 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1456 match
= xfs_inode_match_id_union(ip
, eofb
);
1458 match
= xfs_inode_match_id(ip
, eofb
);
1462 /* skip the inode if the file size is too small */
1463 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1464 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1469 * If the caller is waiting, return -EAGAIN to keep the background
1470 * scanner moving and revisit the inode in a subsequent pass.
1472 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
1473 if (flags
& SYNC_WAIT
)
1477 ret
= xfs_free_eofblocks(ip
);
1478 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1484 __xfs_icache_free_eofblocks(
1485 struct xfs_mount
*mp
,
1486 struct xfs_eofblocks
*eofb
,
1487 int (*execute
)(struct xfs_inode
*ip
, int flags
,
1491 int flags
= SYNC_TRYLOCK
;
1493 if (eofb
&& (eofb
->eof_flags
& XFS_EOF_FLAGS_SYNC
))
1496 return xfs_inode_ag_iterator_tag(mp
, execute
, flags
,
1501 xfs_icache_free_eofblocks(
1502 struct xfs_mount
*mp
,
1503 struct xfs_eofblocks
*eofb
)
1505 return __xfs_icache_free_eofblocks(mp
, eofb
, xfs_inode_free_eofblocks
,
1506 XFS_ICI_EOFBLOCKS_TAG
);
1510 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1511 * multiple quotas, we don't know exactly which quota caused an allocation
1512 * failure. We make a best effort by including each quota under low free space
1513 * conditions (less than 1% free space) in the scan.
1516 __xfs_inode_free_quota_eofblocks(
1517 struct xfs_inode
*ip
,
1518 int (*execute
)(struct xfs_mount
*mp
,
1519 struct xfs_eofblocks
*eofb
))
1522 struct xfs_eofblocks eofb
= {0};
1523 struct xfs_dquot
*dq
;
1526 * Run a sync scan to increase effectiveness and use the union filter to
1527 * cover all applicable quotas in a single scan.
1529 eofb
.eof_flags
= XFS_EOF_FLAGS_UNION
|XFS_EOF_FLAGS_SYNC
;
1531 if (XFS_IS_UQUOTA_ENFORCED(ip
->i_mount
)) {
1532 dq
= xfs_inode_dquot(ip
, XFS_DQ_USER
);
1533 if (dq
&& xfs_dquot_lowsp(dq
)) {
1534 eofb
.eof_uid
= VFS_I(ip
)->i_uid
;
1535 eofb
.eof_flags
|= XFS_EOF_FLAGS_UID
;
1540 if (XFS_IS_GQUOTA_ENFORCED(ip
->i_mount
)) {
1541 dq
= xfs_inode_dquot(ip
, XFS_DQ_GROUP
);
1542 if (dq
&& xfs_dquot_lowsp(dq
)) {
1543 eofb
.eof_gid
= VFS_I(ip
)->i_gid
;
1544 eofb
.eof_flags
|= XFS_EOF_FLAGS_GID
;
1550 execute(ip
->i_mount
, &eofb
);
1556 xfs_inode_free_quota_eofblocks(
1557 struct xfs_inode
*ip
)
1559 return __xfs_inode_free_quota_eofblocks(ip
, xfs_icache_free_eofblocks
);
1562 static inline unsigned long
1567 case XFS_ICI_EOFBLOCKS_TAG
:
1568 return XFS_IEOFBLOCKS
;
1569 case XFS_ICI_COWBLOCKS_TAG
:
1570 return XFS_ICOWBLOCKS
;
1578 __xfs_inode_set_blocks_tag(
1580 void (*execute
)(struct xfs_mount
*mp
),
1581 void (*set_tp
)(struct xfs_mount
*mp
, xfs_agnumber_t agno
,
1582 int error
, unsigned long caller_ip
),
1585 struct xfs_mount
*mp
= ip
->i_mount
;
1586 struct xfs_perag
*pag
;
1590 * Don't bother locking the AG and looking up in the radix trees
1591 * if we already know that we have the tag set.
1593 if (ip
->i_flags
& xfs_iflag_for_tag(tag
))
1595 spin_lock(&ip
->i_flags_lock
);
1596 ip
->i_flags
|= xfs_iflag_for_tag(tag
);
1597 spin_unlock(&ip
->i_flags_lock
);
1599 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1600 spin_lock(&pag
->pag_ici_lock
);
1602 tagged
= radix_tree_tagged(&pag
->pag_ici_root
, tag
);
1603 radix_tree_tag_set(&pag
->pag_ici_root
,
1604 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
), tag
);
1606 /* propagate the eofblocks tag up into the perag radix tree */
1607 spin_lock(&ip
->i_mount
->m_perag_lock
);
1608 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
1609 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1611 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1613 /* kick off background trimming */
1614 execute(ip
->i_mount
);
1616 set_tp(ip
->i_mount
, pag
->pag_agno
, -1, _RET_IP_
);
1619 spin_unlock(&pag
->pag_ici_lock
);
1624 xfs_inode_set_eofblocks_tag(
1627 trace_xfs_inode_set_eofblocks_tag(ip
);
1628 return __xfs_inode_set_blocks_tag(ip
, xfs_queue_eofblocks
,
1629 trace_xfs_perag_set_eofblocks
,
1630 XFS_ICI_EOFBLOCKS_TAG
);
1634 __xfs_inode_clear_blocks_tag(
1636 void (*clear_tp
)(struct xfs_mount
*mp
, xfs_agnumber_t agno
,
1637 int error
, unsigned long caller_ip
),
1640 struct xfs_mount
*mp
= ip
->i_mount
;
1641 struct xfs_perag
*pag
;
1643 spin_lock(&ip
->i_flags_lock
);
1644 ip
->i_flags
&= ~xfs_iflag_for_tag(tag
);
1645 spin_unlock(&ip
->i_flags_lock
);
1647 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1648 spin_lock(&pag
->pag_ici_lock
);
1650 radix_tree_tag_clear(&pag
->pag_ici_root
,
1651 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
), tag
);
1652 if (!radix_tree_tagged(&pag
->pag_ici_root
, tag
)) {
1653 /* clear the eofblocks tag from the perag radix tree */
1654 spin_lock(&ip
->i_mount
->m_perag_lock
);
1655 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
1656 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1658 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1659 clear_tp(ip
->i_mount
, pag
->pag_agno
, -1, _RET_IP_
);
1662 spin_unlock(&pag
->pag_ici_lock
);
1667 xfs_inode_clear_eofblocks_tag(
1670 trace_xfs_inode_clear_eofblocks_tag(ip
);
1671 return __xfs_inode_clear_blocks_tag(ip
,
1672 trace_xfs_perag_clear_eofblocks
, XFS_ICI_EOFBLOCKS_TAG
);
1676 * Automatic CoW Reservation Freeing
1678 * These functions automatically garbage collect leftover CoW reservations
1679 * that were made on behalf of a cowextsize hint when we start to run out
1680 * of quota or when the reservations sit around for too long. If the file
1681 * has dirty pages or is undergoing writeback, its CoW reservations will
1684 * The actual garbage collection piggybacks off the same code that runs
1685 * the speculative EOF preallocation garbage collector.
1688 xfs_inode_free_cowblocks(
1689 struct xfs_inode
*ip
,
1694 struct xfs_eofblocks
*eofb
= args
;
1696 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
1699 * Just clear the tag if we have an empty cow fork or none at all. It's
1700 * possible the inode was fully unshared since it was originally tagged.
1702 if (!xfs_is_reflink_inode(ip
) || !ifp
->if_bytes
) {
1703 trace_xfs_inode_free_cowblocks_invalid(ip
);
1704 xfs_inode_clear_cowblocks_tag(ip
);
1709 * If the mapping is dirty or under writeback we cannot touch the
1710 * CoW fork. Leave it alone if we're in the midst of a directio.
1712 if ((VFS_I(ip
)->i_state
& I_DIRTY_PAGES
) ||
1713 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
) ||
1714 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_WRITEBACK
) ||
1715 atomic_read(&VFS_I(ip
)->i_dio_count
))
1719 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1720 match
= xfs_inode_match_id_union(ip
, eofb
);
1722 match
= xfs_inode_match_id(ip
, eofb
);
1726 /* skip the inode if the file size is too small */
1727 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1728 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1732 /* Free the CoW blocks */
1733 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1734 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
1736 ret
= xfs_reflink_cancel_cow_range(ip
, 0, NULLFILEOFF
, false);
1738 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1739 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1745 xfs_icache_free_cowblocks(
1746 struct xfs_mount
*mp
,
1747 struct xfs_eofblocks
*eofb
)
1749 return __xfs_icache_free_eofblocks(mp
, eofb
, xfs_inode_free_cowblocks
,
1750 XFS_ICI_COWBLOCKS_TAG
);
1754 xfs_inode_free_quota_cowblocks(
1755 struct xfs_inode
*ip
)
1757 return __xfs_inode_free_quota_eofblocks(ip
, xfs_icache_free_cowblocks
);
1761 xfs_inode_set_cowblocks_tag(
1764 trace_xfs_inode_set_cowblocks_tag(ip
);
1765 return __xfs_inode_set_blocks_tag(ip
, xfs_queue_cowblocks
,
1766 trace_xfs_perag_set_cowblocks
,
1767 XFS_ICI_COWBLOCKS_TAG
);
1771 xfs_inode_clear_cowblocks_tag(
1774 trace_xfs_inode_clear_cowblocks_tag(ip
);
1775 return __xfs_inode_clear_blocks_tag(ip
,
1776 trace_xfs_perag_clear_cowblocks
, XFS_ICI_COWBLOCKS_TAG
);