1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_trans_priv.h"
17 #include "xfs_inode_item.h"
18 #include "xfs_quota.h"
19 #include "xfs_trace.h"
20 #include "xfs_icache.h"
21 #include "xfs_bmap_util.h"
22 #include "xfs_dquot_item.h"
23 #include "xfs_dquot.h"
24 #include "xfs_reflink.h"
26 #include <linux/iversion.h>
29 * Allocate and initialise an xfs_inode.
39 * if this didn't occur in transactions, we could use
40 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
41 * code up to do this anyway.
43 ip
= kmem_zone_alloc(xfs_inode_zone
, 0);
46 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
47 kmem_cache_free(xfs_inode_zone
, ip
);
51 /* VFS doesn't initialise i_mode! */
52 VFS_I(ip
)->i_mode
= 0;
54 XFS_STATS_INC(mp
, vn_active
);
55 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
56 ASSERT(!xfs_isiflocked(ip
));
57 ASSERT(ip
->i_ino
== 0);
59 /* initialise the xfs inode */
62 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
66 ip
->i_cformat
= XFS_DINODE_FMT_EXTENTS
;
67 memset(&ip
->i_df
, 0, sizeof(ip
->i_df
));
69 ip
->i_delayed_blks
= 0;
70 memset(&ip
->i_d
, 0, sizeof(ip
->i_d
));
73 INIT_WORK(&ip
->i_ioend_work
, xfs_end_io
);
74 INIT_LIST_HEAD(&ip
->i_ioend_list
);
75 spin_lock_init(&ip
->i_ioend_lock
);
81 xfs_inode_free_callback(
82 struct rcu_head
*head
)
84 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
85 struct xfs_inode
*ip
= XFS_I(inode
);
87 switch (VFS_I(ip
)->i_mode
& S_IFMT
) {
91 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
96 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
98 xfs_idestroy_fork(ip
, XFS_COW_FORK
);
101 ASSERT(!test_bit(XFS_LI_IN_AIL
,
102 &ip
->i_itemp
->ili_item
.li_flags
));
103 xfs_inode_item_destroy(ip
);
107 kmem_cache_free(xfs_inode_zone
, ip
);
112 struct xfs_inode
*ip
)
114 /* asserts to verify all state is correct here */
115 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
116 XFS_STATS_DEC(ip
->i_mount
, vn_active
);
118 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
123 struct xfs_inode
*ip
)
125 ASSERT(!xfs_isiflocked(ip
));
128 * Because we use RCU freeing we need to ensure the inode always
129 * appears to be reclaimed with an invalid inode number when in the
130 * free state. The ip->i_flags_lock provides the barrier against lookup
133 spin_lock(&ip
->i_flags_lock
);
134 ip
->i_flags
= XFS_IRECLAIM
;
136 spin_unlock(&ip
->i_flags_lock
);
138 __xfs_inode_free(ip
);
142 * Queue a new inode reclaim pass if there are reclaimable inodes and there
143 * isn't a reclaim pass already in progress. By default it runs every 5s based
144 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
145 * tunable, but that can be done if this method proves to be ineffective or too
149 xfs_reclaim_work_queue(
150 struct xfs_mount
*mp
)
154 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_RECLAIM_TAG
)) {
155 queue_delayed_work(mp
->m_reclaim_workqueue
, &mp
->m_reclaim_work
,
156 msecs_to_jiffies(xfs_syncd_centisecs
/ 6 * 10));
162 * This is a fast pass over the inode cache to try to get reclaim moving on as
163 * many inodes as possible in a short period of time. It kicks itself every few
164 * seconds, as well as being kicked by the inode cache shrinker when memory
165 * goes low. It scans as quickly as possible avoiding locked inodes or those
166 * already being flushed, and once done schedules a future pass.
170 struct work_struct
*work
)
172 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
173 struct xfs_mount
, m_reclaim_work
);
175 xfs_reclaim_inodes(mp
, SYNC_TRYLOCK
);
176 xfs_reclaim_work_queue(mp
);
180 xfs_perag_set_reclaim_tag(
181 struct xfs_perag
*pag
)
183 struct xfs_mount
*mp
= pag
->pag_mount
;
185 lockdep_assert_held(&pag
->pag_ici_lock
);
186 if (pag
->pag_ici_reclaimable
++)
189 /* propagate the reclaim tag up into the perag radix tree */
190 spin_lock(&mp
->m_perag_lock
);
191 radix_tree_tag_set(&mp
->m_perag_tree
, pag
->pag_agno
,
192 XFS_ICI_RECLAIM_TAG
);
193 spin_unlock(&mp
->m_perag_lock
);
195 /* schedule periodic background inode reclaim */
196 xfs_reclaim_work_queue(mp
);
198 trace_xfs_perag_set_reclaim(mp
, pag
->pag_agno
, -1, _RET_IP_
);
202 xfs_perag_clear_reclaim_tag(
203 struct xfs_perag
*pag
)
205 struct xfs_mount
*mp
= pag
->pag_mount
;
207 lockdep_assert_held(&pag
->pag_ici_lock
);
208 if (--pag
->pag_ici_reclaimable
)
211 /* clear the reclaim tag from the perag radix tree */
212 spin_lock(&mp
->m_perag_lock
);
213 radix_tree_tag_clear(&mp
->m_perag_tree
, pag
->pag_agno
,
214 XFS_ICI_RECLAIM_TAG
);
215 spin_unlock(&mp
->m_perag_lock
);
216 trace_xfs_perag_clear_reclaim(mp
, pag
->pag_agno
, -1, _RET_IP_
);
221 * We set the inode flag atomically with the radix tree tag.
222 * Once we get tag lookups on the radix tree, this inode flag
226 xfs_inode_set_reclaim_tag(
227 struct xfs_inode
*ip
)
229 struct xfs_mount
*mp
= ip
->i_mount
;
230 struct xfs_perag
*pag
;
232 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
233 spin_lock(&pag
->pag_ici_lock
);
234 spin_lock(&ip
->i_flags_lock
);
236 radix_tree_tag_set(&pag
->pag_ici_root
, XFS_INO_TO_AGINO(mp
, ip
->i_ino
),
237 XFS_ICI_RECLAIM_TAG
);
238 xfs_perag_set_reclaim_tag(pag
);
239 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
241 spin_unlock(&ip
->i_flags_lock
);
242 spin_unlock(&pag
->pag_ici_lock
);
247 xfs_inode_clear_reclaim_tag(
248 struct xfs_perag
*pag
,
251 radix_tree_tag_clear(&pag
->pag_ici_root
,
252 XFS_INO_TO_AGINO(pag
->pag_mount
, ino
),
253 XFS_ICI_RECLAIM_TAG
);
254 xfs_perag_clear_reclaim_tag(pag
);
259 struct xfs_inode
*ip
)
261 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_INEW_BIT
);
262 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_INEW_BIT
);
265 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
266 if (!xfs_iflags_test(ip
, XFS_INEW
))
270 finish_wait(wq
, &wait
.wq_entry
);
274 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
275 * part of the structure. This is made more complex by the fact we store
276 * information about the on-disk values in the VFS inode and so we can't just
277 * overwrite the values unconditionally. Hence we save the parameters we
278 * need to retain across reinitialisation, and rewrite them into the VFS inode
279 * after reinitialisation even if it fails.
283 struct xfs_mount
*mp
,
287 uint32_t nlink
= inode
->i_nlink
;
288 uint32_t generation
= inode
->i_generation
;
289 uint64_t version
= inode_peek_iversion(inode
);
290 umode_t mode
= inode
->i_mode
;
291 dev_t dev
= inode
->i_rdev
;
293 error
= inode_init_always(mp
->m_super
, inode
);
295 set_nlink(inode
, nlink
);
296 inode
->i_generation
= generation
;
297 inode_set_iversion_queried(inode
, version
);
298 inode
->i_mode
= mode
;
304 * If we are allocating a new inode, then check what was returned is
305 * actually a free, empty inode. If we are not allocating an inode,
306 * then check we didn't find a free inode.
309 * 0 if the inode free state matches the lookup context
310 * -ENOENT if the inode is free and we are not allocating
311 * -EFSCORRUPTED if there is any state mismatch at all
314 xfs_iget_check_free_state(
315 struct xfs_inode
*ip
,
318 if (flags
& XFS_IGET_CREATE
) {
319 /* should be a free inode */
320 if (VFS_I(ip
)->i_mode
!= 0) {
321 xfs_warn(ip
->i_mount
,
322 "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
323 ip
->i_ino
, VFS_I(ip
)->i_mode
);
324 return -EFSCORRUPTED
;
327 if (ip
->i_d
.di_nblocks
!= 0) {
328 xfs_warn(ip
->i_mount
,
329 "Corruption detected! Free inode 0x%llx has blocks allocated!",
331 return -EFSCORRUPTED
;
336 /* should be an allocated inode */
337 if (VFS_I(ip
)->i_mode
== 0)
344 * Check the validity of the inode we just found it the cache
348 struct xfs_perag
*pag
,
349 struct xfs_inode
*ip
,
352 int lock_flags
) __releases(RCU
)
354 struct inode
*inode
= VFS_I(ip
);
355 struct xfs_mount
*mp
= ip
->i_mount
;
359 * check for re-use of an inode within an RCU grace period due to the
360 * radix tree nodes not being updated yet. We monitor for this by
361 * setting the inode number to zero before freeing the inode structure.
362 * If the inode has been reallocated and set up, then the inode number
363 * will not match, so check for that, too.
365 spin_lock(&ip
->i_flags_lock
);
366 if (ip
->i_ino
!= ino
) {
367 trace_xfs_iget_skip(ip
);
368 XFS_STATS_INC(mp
, xs_ig_frecycle
);
375 * If we are racing with another cache hit that is currently
376 * instantiating this inode or currently recycling it out of
377 * reclaimabe state, wait for the initialisation to complete
380 * XXX(hch): eventually we should do something equivalent to
381 * wait_on_inode to wait for these flags to be cleared
382 * instead of polling for it.
384 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
385 trace_xfs_iget_skip(ip
);
386 XFS_STATS_INC(mp
, xs_ig_frecycle
);
392 * Check the inode free state is valid. This also detects lookup
393 * racing with unlinks.
395 error
= xfs_iget_check_free_state(ip
, flags
);
400 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
401 * Need to carefully get it back into useable state.
403 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
404 trace_xfs_iget_reclaim(ip
);
406 if (flags
& XFS_IGET_INCORE
) {
412 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
413 * from stomping over us while we recycle the inode. We can't
414 * clear the radix tree reclaimable tag yet as it requires
415 * pag_ici_lock to be held exclusive.
417 ip
->i_flags
|= XFS_IRECLAIM
;
419 spin_unlock(&ip
->i_flags_lock
);
422 error
= xfs_reinit_inode(mp
, inode
);
426 * Re-initializing the inode failed, and we are in deep
427 * trouble. Try to re-add it to the reclaim list.
430 spin_lock(&ip
->i_flags_lock
);
431 wake
= !!__xfs_iflags_test(ip
, XFS_INEW
);
432 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
434 wake_up_bit(&ip
->i_flags
, __XFS_INEW_BIT
);
435 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
436 trace_xfs_iget_reclaim_fail(ip
);
440 spin_lock(&pag
->pag_ici_lock
);
441 spin_lock(&ip
->i_flags_lock
);
444 * Clear the per-lifetime state in the inode as we are now
445 * effectively a new inode and need to return to the initial
446 * state before reuse occurs.
448 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
449 ip
->i_flags
|= XFS_INEW
;
450 xfs_inode_clear_reclaim_tag(pag
, ip
->i_ino
);
451 inode
->i_state
= I_NEW
;
455 ASSERT(!rwsem_is_locked(&inode
->i_rwsem
));
456 init_rwsem(&inode
->i_rwsem
);
458 spin_unlock(&ip
->i_flags_lock
);
459 spin_unlock(&pag
->pag_ici_lock
);
461 /* If the VFS inode is being torn down, pause and try again. */
463 trace_xfs_iget_skip(ip
);
468 /* We've got a live one. */
469 spin_unlock(&ip
->i_flags_lock
);
471 trace_xfs_iget_hit(ip
);
475 xfs_ilock(ip
, lock_flags
);
477 if (!(flags
& XFS_IGET_INCORE
))
478 xfs_iflags_clear(ip
, XFS_ISTALE
| XFS_IDONTCACHE
);
479 XFS_STATS_INC(mp
, xs_ig_found
);
484 spin_unlock(&ip
->i_flags_lock
);
492 struct xfs_mount
*mp
,
493 struct xfs_perag
*pag
,
496 struct xfs_inode
**ipp
,
500 struct xfs_inode
*ip
;
502 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
505 ip
= xfs_inode_alloc(mp
, ino
);
509 error
= xfs_iread(mp
, tp
, ip
, flags
);
513 if (!xfs_inode_verify_forks(ip
)) {
514 error
= -EFSCORRUPTED
;
518 trace_xfs_iget_miss(ip
);
522 * Check the inode free state is valid. This also detects lookup
523 * racing with unlinks.
525 error
= xfs_iget_check_free_state(ip
, flags
);
530 * Preload the radix tree so we can insert safely under the
531 * write spinlock. Note that we cannot sleep inside the preload
532 * region. Since we can be called from transaction context, don't
533 * recurse into the file system.
535 if (radix_tree_preload(GFP_NOFS
)) {
541 * Because the inode hasn't been added to the radix-tree yet it can't
542 * be found by another thread, so we can do the non-sleeping lock here.
545 if (!xfs_ilock_nowait(ip
, lock_flags
))
550 * These values must be set before inserting the inode into the radix
551 * tree as the moment it is inserted a concurrent lookup (allowed by the
552 * RCU locking mechanism) can find it and that lookup must see that this
553 * is an inode currently under construction (i.e. that XFS_INEW is set).
554 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
555 * memory barrier that ensures this detection works correctly at lookup
559 if (flags
& XFS_IGET_DONTCACHE
)
560 iflags
|= XFS_IDONTCACHE
;
564 xfs_iflags_set(ip
, iflags
);
566 /* insert the new inode */
567 spin_lock(&pag
->pag_ici_lock
);
568 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
569 if (unlikely(error
)) {
570 WARN_ON(error
!= -EEXIST
);
571 XFS_STATS_INC(mp
, xs_ig_dup
);
573 goto out_preload_end
;
575 spin_unlock(&pag
->pag_ici_lock
);
576 radix_tree_preload_end();
582 spin_unlock(&pag
->pag_ici_lock
);
583 radix_tree_preload_end();
585 xfs_iunlock(ip
, lock_flags
);
587 __destroy_inode(VFS_I(ip
));
593 * Look up an inode by number in the given file system.
594 * The inode is looked up in the cache held in each AG.
595 * If the inode is found in the cache, initialise the vfs inode
598 * If it is not in core, read it in from the file system's device,
599 * add it to the cache and initialise the vfs inode.
601 * The inode is locked according to the value of the lock_flags parameter.
602 * This flag parameter indicates how and if the inode's IO lock and inode lock
605 * mp -- the mount point structure for the current file system. It points
606 * to the inode hash table.
607 * tp -- a pointer to the current transaction if there is one. This is
608 * simply passed through to the xfs_iread() call.
609 * ino -- the number of the inode desired. This is the unique identifier
610 * within the file system for the inode being requested.
611 * lock_flags -- flags indicating how to lock the inode. See the comment
612 * for xfs_ilock() for a list of valid values.
629 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
630 * doesn't get freed while it's being referenced during a
631 * radix tree traversal here. It assumes this function
632 * aqcuires only the ILOCK (and therefore it has no need to
633 * involve the IOLOCK in this synchronization).
635 ASSERT((lock_flags
& (XFS_IOLOCK_EXCL
| XFS_IOLOCK_SHARED
)) == 0);
637 /* reject inode numbers outside existing AGs */
638 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
641 XFS_STATS_INC(mp
, xs_ig_attempts
);
643 /* get the perag structure and ensure that it's inode capable */
644 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
645 agino
= XFS_INO_TO_AGINO(mp
, ino
);
650 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
653 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
655 goto out_error_or_again
;
658 if (flags
& XFS_IGET_INCORE
) {
660 goto out_error_or_again
;
662 XFS_STATS_INC(mp
, xs_ig_missed
);
664 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
667 goto out_error_or_again
;
674 * If we have a real type for an on-disk inode, we can setup the inode
675 * now. If it's a new inode being created, xfs_ialloc will handle it.
677 if (xfs_iflags_test(ip
, XFS_INEW
) && VFS_I(ip
)->i_mode
!= 0)
678 xfs_setup_existing_inode(ip
);
682 if (!(flags
& XFS_IGET_INCORE
) && error
== -EAGAIN
) {
691 * "Is this a cached inode that's also allocated?"
693 * Look up an inode by number in the given file system. If the inode is
694 * in cache and isn't in purgatory, return 1 if the inode is allocated
695 * and 0 if it is not. For all other cases (not in cache, being torn
696 * down, etc.), return a negative error code.
698 * The caller has to prevent inode allocation and freeing activity,
699 * presumably by locking the AGI buffer. This is to ensure that an
700 * inode cannot transition from allocated to freed until the caller is
701 * ready to allow that. If the inode is in an intermediate state (new,
702 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
703 * inode is not in the cache, -ENOENT will be returned. The caller must
704 * deal with these scenarios appropriately.
706 * This is a specialized use case for the online scrubber; if you're
707 * reading this, you probably want xfs_iget.
710 xfs_icache_inode_is_allocated(
711 struct xfs_mount
*mp
,
712 struct xfs_trans
*tp
,
716 struct xfs_inode
*ip
;
719 error
= xfs_iget(mp
, tp
, ino
, XFS_IGET_INCORE
, 0, &ip
);
723 *inuse
= !!(VFS_I(ip
)->i_mode
);
729 * The inode lookup is done in batches to keep the amount of lock traffic and
730 * radix tree lookups to a minimum. The batch size is a trade off between
731 * lookup reduction and stack usage. This is in the reclaim path, so we can't
734 #define XFS_LOOKUP_BATCH 32
737 xfs_inode_ag_walk_grab(
738 struct xfs_inode
*ip
,
741 struct inode
*inode
= VFS_I(ip
);
742 bool newinos
= !!(flags
& XFS_AGITER_INEW_WAIT
);
744 ASSERT(rcu_read_lock_held());
747 * check for stale RCU freed inode
749 * If the inode has been reallocated, it doesn't matter if it's not in
750 * the AG we are walking - we are walking for writeback, so if it
751 * passes all the "valid inode" checks and is dirty, then we'll write
752 * it back anyway. If it has been reallocated and still being
753 * initialised, the XFS_INEW check below will catch it.
755 spin_lock(&ip
->i_flags_lock
);
757 goto out_unlock_noent
;
759 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
760 if ((!newinos
&& __xfs_iflags_test(ip
, XFS_INEW
)) ||
761 __xfs_iflags_test(ip
, XFS_IRECLAIMABLE
| XFS_IRECLAIM
))
762 goto out_unlock_noent
;
763 spin_unlock(&ip
->i_flags_lock
);
765 /* nothing to sync during shutdown */
766 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
767 return -EFSCORRUPTED
;
769 /* If we can't grab the inode, it must on it's way to reclaim. */
777 spin_unlock(&ip
->i_flags_lock
);
783 struct xfs_mount
*mp
,
784 struct xfs_perag
*pag
,
785 int (*execute
)(struct xfs_inode
*ip
, int flags
,
792 uint32_t first_index
;
804 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
811 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
,
812 (void **)batch
, first_index
,
815 nr_found
= radix_tree_gang_lookup_tag(
817 (void **) batch
, first_index
,
818 XFS_LOOKUP_BATCH
, tag
);
826 * Grab the inodes before we drop the lock. if we found
827 * nothing, nr == 0 and the loop will be skipped.
829 for (i
= 0; i
< nr_found
; i
++) {
830 struct xfs_inode
*ip
= batch
[i
];
832 if (done
|| xfs_inode_ag_walk_grab(ip
, iter_flags
))
836 * Update the index for the next lookup. Catch
837 * overflows into the next AG range which can occur if
838 * we have inodes in the last block of the AG and we
839 * are currently pointing to the last inode.
841 * Because we may see inodes that are from the wrong AG
842 * due to RCU freeing and reallocation, only update the
843 * index if it lies in this AG. It was a race that lead
844 * us to see this inode, so another lookup from the
845 * same index will not find it again.
847 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) != pag
->pag_agno
)
849 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
850 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
854 /* unlock now we've grabbed the inodes. */
857 for (i
= 0; i
< nr_found
; i
++) {
860 if ((iter_flags
& XFS_AGITER_INEW_WAIT
) &&
861 xfs_iflags_test(batch
[i
], XFS_INEW
))
862 xfs_inew_wait(batch
[i
]);
863 error
= execute(batch
[i
], flags
, args
);
865 if (error
== -EAGAIN
) {
869 if (error
&& last_error
!= -EFSCORRUPTED
)
873 /* bail out if the filesystem is corrupted. */
874 if (error
== -EFSCORRUPTED
)
879 } while (nr_found
&& !done
);
889 * Background scanning to trim post-EOF preallocated space. This is queued
890 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
894 struct xfs_mount
*mp
)
897 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_EOFBLOCKS_TAG
))
898 queue_delayed_work(mp
->m_eofblocks_workqueue
,
899 &mp
->m_eofblocks_work
,
900 msecs_to_jiffies(xfs_eofb_secs
* 1000));
905 xfs_eofblocks_worker(
906 struct work_struct
*work
)
908 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
909 struct xfs_mount
, m_eofblocks_work
);
910 xfs_icache_free_eofblocks(mp
, NULL
);
911 xfs_queue_eofblocks(mp
);
915 * Background scanning to trim preallocated CoW space. This is queued
916 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
917 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
921 struct xfs_mount
*mp
)
924 if (radix_tree_tagged(&mp
->m_perag_tree
, XFS_ICI_COWBLOCKS_TAG
))
925 queue_delayed_work(mp
->m_eofblocks_workqueue
,
926 &mp
->m_cowblocks_work
,
927 msecs_to_jiffies(xfs_cowb_secs
* 1000));
932 xfs_cowblocks_worker(
933 struct work_struct
*work
)
935 struct xfs_mount
*mp
= container_of(to_delayed_work(work
),
936 struct xfs_mount
, m_cowblocks_work
);
937 xfs_icache_free_cowblocks(mp
, NULL
);
938 xfs_queue_cowblocks(mp
);
942 xfs_inode_ag_iterator_flags(
943 struct xfs_mount
*mp
,
944 int (*execute
)(struct xfs_inode
*ip
, int flags
,
950 struct xfs_perag
*pag
;
956 while ((pag
= xfs_perag_get(mp
, ag
))) {
957 ag
= pag
->pag_agno
+ 1;
958 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, -1,
963 if (error
== -EFSCORRUPTED
)
971 xfs_inode_ag_iterator(
972 struct xfs_mount
*mp
,
973 int (*execute
)(struct xfs_inode
*ip
, int flags
,
978 return xfs_inode_ag_iterator_flags(mp
, execute
, flags
, args
, 0);
982 xfs_inode_ag_iterator_tag(
983 struct xfs_mount
*mp
,
984 int (*execute
)(struct xfs_inode
*ip
, int flags
,
990 struct xfs_perag
*pag
;
996 while ((pag
= xfs_perag_get_tag(mp
, ag
, tag
))) {
997 ag
= pag
->pag_agno
+ 1;
998 error
= xfs_inode_ag_walk(mp
, pag
, execute
, flags
, args
, tag
,
1003 if (error
== -EFSCORRUPTED
)
1011 * Grab the inode for reclaim exclusively.
1012 * Return 0 if we grabbed it, non-zero otherwise.
1015 xfs_reclaim_inode_grab(
1016 struct xfs_inode
*ip
,
1019 ASSERT(rcu_read_lock_held());
1021 /* quick check for stale RCU freed inode */
1026 * If we are asked for non-blocking operation, do unlocked checks to
1027 * see if the inode already is being flushed or in reclaim to avoid
1030 if ((flags
& SYNC_TRYLOCK
) &&
1031 __xfs_iflags_test(ip
, XFS_IFLOCK
| XFS_IRECLAIM
))
1035 * The radix tree lock here protects a thread in xfs_iget from racing
1036 * with us starting reclaim on the inode. Once we have the
1037 * XFS_IRECLAIM flag set it will not touch us.
1039 * Due to RCU lookup, we may find inodes that have been freed and only
1040 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
1041 * aren't candidates for reclaim at all, so we must check the
1042 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
1044 spin_lock(&ip
->i_flags_lock
);
1045 if (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) ||
1046 __xfs_iflags_test(ip
, XFS_IRECLAIM
)) {
1047 /* not a reclaim candidate. */
1048 spin_unlock(&ip
->i_flags_lock
);
1051 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
1052 spin_unlock(&ip
->i_flags_lock
);
1057 * Inodes in different states need to be treated differently. The following
1058 * table lists the inode states and the reclaim actions necessary:
1060 * inode state iflush ret required action
1061 * --------------- ---------- ---------------
1063 * shutdown EIO unpin and reclaim
1064 * clean, unpinned 0 reclaim
1065 * stale, unpinned 0 reclaim
1066 * clean, pinned(*) 0 requeue
1067 * stale, pinned EAGAIN requeue
1068 * dirty, async - requeue
1069 * dirty, sync 0 reclaim
1071 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1072 * handled anyway given the order of checks implemented.
1074 * Also, because we get the flush lock first, we know that any inode that has
1075 * been flushed delwri has had the flush completed by the time we check that
1076 * the inode is clean.
1078 * Note that because the inode is flushed delayed write by AIL pushing, the
1079 * flush lock may already be held here and waiting on it can result in very
1080 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1081 * the caller should push the AIL first before trying to reclaim inodes to
1082 * minimise the amount of time spent waiting. For background relaim, we only
1083 * bother to reclaim clean inodes anyway.
1085 * Hence the order of actions after gaining the locks should be:
1087 * shutdown => unpin and reclaim
1088 * pinned, async => requeue
1089 * pinned, sync => unpin
1092 * dirty, async => requeue
1093 * dirty, sync => flush, wait and reclaim
1097 struct xfs_inode
*ip
,
1098 struct xfs_perag
*pag
,
1101 struct xfs_buf
*bp
= NULL
;
1102 xfs_ino_t ino
= ip
->i_ino
; /* for radix_tree_delete */
1107 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1108 if (!xfs_iflock_nowait(ip
)) {
1109 if (!(sync_mode
& SYNC_WAIT
))
1114 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1115 xfs_iunpin_wait(ip
);
1116 /* xfs_iflush_abort() drops the flush lock */
1117 xfs_iflush_abort(ip
, false);
1120 if (xfs_ipincount(ip
)) {
1121 if (!(sync_mode
& SYNC_WAIT
))
1123 xfs_iunpin_wait(ip
);
1125 if (xfs_iflags_test(ip
, XFS_ISTALE
) || xfs_inode_clean(ip
)) {
1131 * Never flush out dirty data during non-blocking reclaim, as it would
1132 * just contend with AIL pushing trying to do the same job.
1134 if (!(sync_mode
& SYNC_WAIT
))
1138 * Now we have an inode that needs flushing.
1140 * Note that xfs_iflush will never block on the inode buffer lock, as
1141 * xfs_ifree_cluster() can lock the inode buffer before it locks the
1142 * ip->i_lock, and we are doing the exact opposite here. As a result,
1143 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1144 * result in an ABBA deadlock with xfs_ifree_cluster().
1146 * As xfs_ifree_cluser() must gather all inodes that are active in the
1147 * cache to mark them stale, if we hit this case we don't actually want
1148 * to do IO here - we want the inode marked stale so we can simply
1149 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1150 * inode, back off and try again. Hopefully the next pass through will
1151 * see the stale flag set on the inode.
1153 error
= xfs_iflush(ip
, &bp
);
1154 if (error
== -EAGAIN
) {
1155 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1156 /* backoff longer than in xfs_ifree_cluster */
1162 error
= xfs_bwrite(bp
);
1167 ASSERT(!xfs_isiflocked(ip
));
1170 * Because we use RCU freeing we need to ensure the inode always appears
1171 * to be reclaimed with an invalid inode number when in the free state.
1172 * We do this as early as possible under the ILOCK so that
1173 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1174 * detect races with us here. By doing this, we guarantee that once
1175 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1176 * it will see either a valid inode that will serialise correctly, or it
1177 * will see an invalid inode that it can skip.
1179 spin_lock(&ip
->i_flags_lock
);
1180 ip
->i_flags
= XFS_IRECLAIM
;
1182 spin_unlock(&ip
->i_flags_lock
);
1184 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1186 XFS_STATS_INC(ip
->i_mount
, xs_ig_reclaims
);
1188 * Remove the inode from the per-AG radix tree.
1190 * Because radix_tree_delete won't complain even if the item was never
1191 * added to the tree assert that it's been there before to catch
1192 * problems with the inode life time early on.
1194 spin_lock(&pag
->pag_ici_lock
);
1195 if (!radix_tree_delete(&pag
->pag_ici_root
,
1196 XFS_INO_TO_AGINO(ip
->i_mount
, ino
)))
1198 xfs_perag_clear_reclaim_tag(pag
);
1199 spin_unlock(&pag
->pag_ici_lock
);
1202 * Here we do an (almost) spurious inode lock in order to coordinate
1203 * with inode cache radix tree lookups. This is because the lookup
1204 * can reference the inodes in the cache without taking references.
1206 * We make that OK here by ensuring that we wait until the inode is
1207 * unlocked after the lookup before we go ahead and free it.
1209 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1210 xfs_qm_dqdetach(ip
);
1211 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1213 __xfs_inode_free(ip
);
1219 xfs_iflags_clear(ip
, XFS_IRECLAIM
);
1220 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1222 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1223 * a short while. However, this just burns CPU time scanning the tree
1224 * waiting for IO to complete and the reclaim work never goes back to
1225 * the idle state. Instead, return 0 to let the next scheduled
1226 * background reclaim attempt to reclaim the inode again.
1232 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1233 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1234 * then a shut down during filesystem unmount reclaim walk leak all the
1235 * unreclaimed inodes.
1238 xfs_reclaim_inodes_ag(
1239 struct xfs_mount
*mp
,
1243 struct xfs_perag
*pag
;
1247 int trylock
= flags
& SYNC_TRYLOCK
;
1253 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1254 unsigned long first_index
= 0;
1258 ag
= pag
->pag_agno
+ 1;
1261 if (!mutex_trylock(&pag
->pag_ici_reclaim_lock
)) {
1266 first_index
= pag
->pag_ici_reclaim_cursor
;
1268 mutex_lock(&pag
->pag_ici_reclaim_lock
);
1271 struct xfs_inode
*batch
[XFS_LOOKUP_BATCH
];
1275 nr_found
= radix_tree_gang_lookup_tag(
1277 (void **)batch
, first_index
,
1279 XFS_ICI_RECLAIM_TAG
);
1287 * Grab the inodes before we drop the lock. if we found
1288 * nothing, nr == 0 and the loop will be skipped.
1290 for (i
= 0; i
< nr_found
; i
++) {
1291 struct xfs_inode
*ip
= batch
[i
];
1293 if (done
|| xfs_reclaim_inode_grab(ip
, flags
))
1297 * Update the index for the next lookup. Catch
1298 * overflows into the next AG range which can
1299 * occur if we have inodes in the last block of
1300 * the AG and we are currently pointing to the
1303 * Because we may see inodes that are from the
1304 * wrong AG due to RCU freeing and
1305 * reallocation, only update the index if it
1306 * lies in this AG. It was a race that lead us
1307 * to see this inode, so another lookup from
1308 * the same index will not find it again.
1310 if (XFS_INO_TO_AGNO(mp
, ip
->i_ino
) !=
1313 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
+ 1);
1314 if (first_index
< XFS_INO_TO_AGINO(mp
, ip
->i_ino
))
1318 /* unlock now we've grabbed the inodes. */
1321 for (i
= 0; i
< nr_found
; i
++) {
1324 error
= xfs_reclaim_inode(batch
[i
], pag
, flags
);
1325 if (error
&& last_error
!= -EFSCORRUPTED
)
1329 *nr_to_scan
-= XFS_LOOKUP_BATCH
;
1333 } while (nr_found
&& !done
&& *nr_to_scan
> 0);
1335 if (trylock
&& !done
)
1336 pag
->pag_ici_reclaim_cursor
= first_index
;
1338 pag
->pag_ici_reclaim_cursor
= 0;
1339 mutex_unlock(&pag
->pag_ici_reclaim_lock
);
1344 * if we skipped any AG, and we still have scan count remaining, do
1345 * another pass this time using blocking reclaim semantics (i.e
1346 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1347 * ensure that when we get more reclaimers than AGs we block rather
1348 * than spin trying to execute reclaim.
1350 if (skipped
&& (flags
& SYNC_WAIT
) && *nr_to_scan
> 0) {
1362 int nr_to_scan
= INT_MAX
;
1364 return xfs_reclaim_inodes_ag(mp
, mode
, &nr_to_scan
);
1368 * Scan a certain number of inodes for reclaim.
1370 * When called we make sure that there is a background (fast) inode reclaim in
1371 * progress, while we will throttle the speed of reclaim via doing synchronous
1372 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1373 * them to be cleaned, which we hope will not be very long due to the
1374 * background walker having already kicked the IO off on those dirty inodes.
1377 xfs_reclaim_inodes_nr(
1378 struct xfs_mount
*mp
,
1381 /* kick background reclaimer and push the AIL */
1382 xfs_reclaim_work_queue(mp
);
1383 xfs_ail_push_all(mp
->m_ail
);
1385 return xfs_reclaim_inodes_ag(mp
, SYNC_TRYLOCK
| SYNC_WAIT
, &nr_to_scan
);
1389 * Return the number of reclaimable inodes in the filesystem for
1390 * the shrinker to determine how much to reclaim.
1393 xfs_reclaim_inodes_count(
1394 struct xfs_mount
*mp
)
1396 struct xfs_perag
*pag
;
1397 xfs_agnumber_t ag
= 0;
1398 int reclaimable
= 0;
1400 while ((pag
= xfs_perag_get_tag(mp
, ag
, XFS_ICI_RECLAIM_TAG
))) {
1401 ag
= pag
->pag_agno
+ 1;
1402 reclaimable
+= pag
->pag_ici_reclaimable
;
1410 struct xfs_inode
*ip
,
1411 struct xfs_eofblocks
*eofb
)
1413 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1414 !uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1417 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1418 !gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1421 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1422 ip
->i_d
.di_projid
!= eofb
->eof_prid
)
1429 * A union-based inode filtering algorithm. Process the inode if any of the
1430 * criteria match. This is for global/internal scans only.
1433 xfs_inode_match_id_union(
1434 struct xfs_inode
*ip
,
1435 struct xfs_eofblocks
*eofb
)
1437 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_UID
) &&
1438 uid_eq(VFS_I(ip
)->i_uid
, eofb
->eof_uid
))
1441 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_GID
) &&
1442 gid_eq(VFS_I(ip
)->i_gid
, eofb
->eof_gid
))
1445 if ((eofb
->eof_flags
& XFS_EOF_FLAGS_PRID
) &&
1446 ip
->i_d
.di_projid
== eofb
->eof_prid
)
1453 xfs_inode_free_eofblocks(
1454 struct xfs_inode
*ip
,
1459 struct xfs_eofblocks
*eofb
= args
;
1462 if (!xfs_can_free_eofblocks(ip
, false)) {
1463 /* inode could be preallocated or append-only */
1464 trace_xfs_inode_free_eofblocks_invalid(ip
);
1465 xfs_inode_clear_eofblocks_tag(ip
);
1470 * If the mapping is dirty the operation can block and wait for some
1471 * time. Unless we are waiting, skip it.
1473 if (!(flags
& SYNC_WAIT
) &&
1474 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
))
1478 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1479 match
= xfs_inode_match_id_union(ip
, eofb
);
1481 match
= xfs_inode_match_id(ip
, eofb
);
1485 /* skip the inode if the file size is too small */
1486 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1487 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1492 * If the caller is waiting, return -EAGAIN to keep the background
1493 * scanner moving and revisit the inode in a subsequent pass.
1495 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
1496 if (flags
& SYNC_WAIT
)
1500 ret
= xfs_free_eofblocks(ip
);
1501 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1507 __xfs_icache_free_eofblocks(
1508 struct xfs_mount
*mp
,
1509 struct xfs_eofblocks
*eofb
,
1510 int (*execute
)(struct xfs_inode
*ip
, int flags
,
1514 int flags
= SYNC_TRYLOCK
;
1516 if (eofb
&& (eofb
->eof_flags
& XFS_EOF_FLAGS_SYNC
))
1519 return xfs_inode_ag_iterator_tag(mp
, execute
, flags
,
1524 xfs_icache_free_eofblocks(
1525 struct xfs_mount
*mp
,
1526 struct xfs_eofblocks
*eofb
)
1528 return __xfs_icache_free_eofblocks(mp
, eofb
, xfs_inode_free_eofblocks
,
1529 XFS_ICI_EOFBLOCKS_TAG
);
1533 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1534 * multiple quotas, we don't know exactly which quota caused an allocation
1535 * failure. We make a best effort by including each quota under low free space
1536 * conditions (less than 1% free space) in the scan.
1539 __xfs_inode_free_quota_eofblocks(
1540 struct xfs_inode
*ip
,
1541 int (*execute
)(struct xfs_mount
*mp
,
1542 struct xfs_eofblocks
*eofb
))
1545 struct xfs_eofblocks eofb
= {0};
1546 struct xfs_dquot
*dq
;
1549 * Run a sync scan to increase effectiveness and use the union filter to
1550 * cover all applicable quotas in a single scan.
1552 eofb
.eof_flags
= XFS_EOF_FLAGS_UNION
|XFS_EOF_FLAGS_SYNC
;
1554 if (XFS_IS_UQUOTA_ENFORCED(ip
->i_mount
)) {
1555 dq
= xfs_inode_dquot(ip
, XFS_DQ_USER
);
1556 if (dq
&& xfs_dquot_lowsp(dq
)) {
1557 eofb
.eof_uid
= VFS_I(ip
)->i_uid
;
1558 eofb
.eof_flags
|= XFS_EOF_FLAGS_UID
;
1563 if (XFS_IS_GQUOTA_ENFORCED(ip
->i_mount
)) {
1564 dq
= xfs_inode_dquot(ip
, XFS_DQ_GROUP
);
1565 if (dq
&& xfs_dquot_lowsp(dq
)) {
1566 eofb
.eof_gid
= VFS_I(ip
)->i_gid
;
1567 eofb
.eof_flags
|= XFS_EOF_FLAGS_GID
;
1573 execute(ip
->i_mount
, &eofb
);
1579 xfs_inode_free_quota_eofblocks(
1580 struct xfs_inode
*ip
)
1582 return __xfs_inode_free_quota_eofblocks(ip
, xfs_icache_free_eofblocks
);
1585 static inline unsigned long
1590 case XFS_ICI_EOFBLOCKS_TAG
:
1591 return XFS_IEOFBLOCKS
;
1592 case XFS_ICI_COWBLOCKS_TAG
:
1593 return XFS_ICOWBLOCKS
;
1601 __xfs_inode_set_blocks_tag(
1603 void (*execute
)(struct xfs_mount
*mp
),
1604 void (*set_tp
)(struct xfs_mount
*mp
, xfs_agnumber_t agno
,
1605 int error
, unsigned long caller_ip
),
1608 struct xfs_mount
*mp
= ip
->i_mount
;
1609 struct xfs_perag
*pag
;
1613 * Don't bother locking the AG and looking up in the radix trees
1614 * if we already know that we have the tag set.
1616 if (ip
->i_flags
& xfs_iflag_for_tag(tag
))
1618 spin_lock(&ip
->i_flags_lock
);
1619 ip
->i_flags
|= xfs_iflag_for_tag(tag
);
1620 spin_unlock(&ip
->i_flags_lock
);
1622 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1623 spin_lock(&pag
->pag_ici_lock
);
1625 tagged
= radix_tree_tagged(&pag
->pag_ici_root
, tag
);
1626 radix_tree_tag_set(&pag
->pag_ici_root
,
1627 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
), tag
);
1629 /* propagate the eofblocks tag up into the perag radix tree */
1630 spin_lock(&ip
->i_mount
->m_perag_lock
);
1631 radix_tree_tag_set(&ip
->i_mount
->m_perag_tree
,
1632 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1634 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1636 /* kick off background trimming */
1637 execute(ip
->i_mount
);
1639 set_tp(ip
->i_mount
, pag
->pag_agno
, -1, _RET_IP_
);
1642 spin_unlock(&pag
->pag_ici_lock
);
1647 xfs_inode_set_eofblocks_tag(
1650 trace_xfs_inode_set_eofblocks_tag(ip
);
1651 return __xfs_inode_set_blocks_tag(ip
, xfs_queue_eofblocks
,
1652 trace_xfs_perag_set_eofblocks
,
1653 XFS_ICI_EOFBLOCKS_TAG
);
1657 __xfs_inode_clear_blocks_tag(
1659 void (*clear_tp
)(struct xfs_mount
*mp
, xfs_agnumber_t agno
,
1660 int error
, unsigned long caller_ip
),
1663 struct xfs_mount
*mp
= ip
->i_mount
;
1664 struct xfs_perag
*pag
;
1666 spin_lock(&ip
->i_flags_lock
);
1667 ip
->i_flags
&= ~xfs_iflag_for_tag(tag
);
1668 spin_unlock(&ip
->i_flags_lock
);
1670 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
1671 spin_lock(&pag
->pag_ici_lock
);
1673 radix_tree_tag_clear(&pag
->pag_ici_root
,
1674 XFS_INO_TO_AGINO(ip
->i_mount
, ip
->i_ino
), tag
);
1675 if (!radix_tree_tagged(&pag
->pag_ici_root
, tag
)) {
1676 /* clear the eofblocks tag from the perag radix tree */
1677 spin_lock(&ip
->i_mount
->m_perag_lock
);
1678 radix_tree_tag_clear(&ip
->i_mount
->m_perag_tree
,
1679 XFS_INO_TO_AGNO(ip
->i_mount
, ip
->i_ino
),
1681 spin_unlock(&ip
->i_mount
->m_perag_lock
);
1682 clear_tp(ip
->i_mount
, pag
->pag_agno
, -1, _RET_IP_
);
1685 spin_unlock(&pag
->pag_ici_lock
);
1690 xfs_inode_clear_eofblocks_tag(
1693 trace_xfs_inode_clear_eofblocks_tag(ip
);
1694 return __xfs_inode_clear_blocks_tag(ip
,
1695 trace_xfs_perag_clear_eofblocks
, XFS_ICI_EOFBLOCKS_TAG
);
1699 * Set ourselves up to free CoW blocks from this file. If it's already clean
1700 * then we can bail out quickly, but otherwise we must back off if the file
1701 * is undergoing some kind of write.
1704 xfs_prep_free_cowblocks(
1705 struct xfs_inode
*ip
)
1708 * Just clear the tag if we have an empty cow fork or none at all. It's
1709 * possible the inode was fully unshared since it was originally tagged.
1711 if (!xfs_inode_has_cow_data(ip
)) {
1712 trace_xfs_inode_free_cowblocks_invalid(ip
);
1713 xfs_inode_clear_cowblocks_tag(ip
);
1718 * If the mapping is dirty or under writeback we cannot touch the
1719 * CoW fork. Leave it alone if we're in the midst of a directio.
1721 if ((VFS_I(ip
)->i_state
& I_DIRTY_PAGES
) ||
1722 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_DIRTY
) ||
1723 mapping_tagged(VFS_I(ip
)->i_mapping
, PAGECACHE_TAG_WRITEBACK
) ||
1724 atomic_read(&VFS_I(ip
)->i_dio_count
))
1731 * Automatic CoW Reservation Freeing
1733 * These functions automatically garbage collect leftover CoW reservations
1734 * that were made on behalf of a cowextsize hint when we start to run out
1735 * of quota or when the reservations sit around for too long. If the file
1736 * has dirty pages or is undergoing writeback, its CoW reservations will
1739 * The actual garbage collection piggybacks off the same code that runs
1740 * the speculative EOF preallocation garbage collector.
1743 xfs_inode_free_cowblocks(
1744 struct xfs_inode
*ip
,
1748 struct xfs_eofblocks
*eofb
= args
;
1752 if (!xfs_prep_free_cowblocks(ip
))
1756 if (eofb
->eof_flags
& XFS_EOF_FLAGS_UNION
)
1757 match
= xfs_inode_match_id_union(ip
, eofb
);
1759 match
= xfs_inode_match_id(ip
, eofb
);
1763 /* skip the inode if the file size is too small */
1764 if (eofb
->eof_flags
& XFS_EOF_FLAGS_MINFILESIZE
&&
1765 XFS_ISIZE(ip
) < eofb
->eof_min_file_size
)
1769 /* Free the CoW blocks */
1770 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1771 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
1774 * Check again, nobody else should be able to dirty blocks or change
1775 * the reflink iflag now that we have the first two locks held.
1777 if (xfs_prep_free_cowblocks(ip
))
1778 ret
= xfs_reflink_cancel_cow_range(ip
, 0, NULLFILEOFF
, false);
1780 xfs_iunlock(ip
, XFS_MMAPLOCK_EXCL
);
1781 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1787 xfs_icache_free_cowblocks(
1788 struct xfs_mount
*mp
,
1789 struct xfs_eofblocks
*eofb
)
1791 return __xfs_icache_free_eofblocks(mp
, eofb
, xfs_inode_free_cowblocks
,
1792 XFS_ICI_COWBLOCKS_TAG
);
1796 xfs_inode_free_quota_cowblocks(
1797 struct xfs_inode
*ip
)
1799 return __xfs_inode_free_quota_eofblocks(ip
, xfs_icache_free_cowblocks
);
1803 xfs_inode_set_cowblocks_tag(
1806 trace_xfs_inode_set_cowblocks_tag(ip
);
1807 return __xfs_inode_set_blocks_tag(ip
, xfs_queue_cowblocks
,
1808 trace_xfs_perag_set_cowblocks
,
1809 XFS_ICI_COWBLOCKS_TAG
);
1813 xfs_inode_clear_cowblocks_tag(
1816 trace_xfs_inode_clear_cowblocks_tag(ip
);
1817 return __xfs_inode_clear_blocks_tag(ip
,
1818 trace_xfs_perag_clear_cowblocks
, XFS_ICI_COWBLOCKS_TAG
);
1821 /* Disable post-EOF and CoW block auto-reclamation. */
1823 xfs_stop_block_reaping(
1824 struct xfs_mount
*mp
)
1826 cancel_delayed_work_sync(&mp
->m_eofblocks_work
);
1827 cancel_delayed_work_sync(&mp
->m_cowblocks_work
);
1830 /* Enable post-EOF and CoW block auto-reclamation. */
1832 xfs_start_block_reaping(
1833 struct xfs_mount
*mp
)
1835 xfs_queue_eofblocks(mp
);
1836 xfs_queue_cowblocks(mp
);