1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_error.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/kthread.h>
27 #include <linux/freezer.h>
28 #include <linux/iversion.h>
31 * Allocate and initialise an xfs_inode.
41 * if this didn't occur in transactions, we could use
42 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
43 * code up to do this anyway.
45 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
48 if (inode_init_always(mp->m_super, VFS_I(ip))) {
49 kmem_zone_free(xfs_inode_zone, ip);
53 /* VFS doesn't initialise i_mode! */
54 VFS_I(ip)->i_mode = 0;
56 XFS_STATS_INC(mp, vn_active);
57 ASSERT(atomic_read(&ip->i_pincount) == 0);
58 ASSERT(!xfs_isiflocked(ip));
59 ASSERT(ip->i_ino == 0);
61 /* initialise the xfs inode */
64 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
68 ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
69 memset(&ip->i_df, 0, sizeof(ip->i_df));
71 ip->i_delayed_blks = 0;
72 memset(&ip->i_d, 0, sizeof(ip->i_d));
78 xfs_inode_free_callback(
79 struct rcu_head *head)
81 struct inode *inode = container_of(head, struct inode, i_rcu);
82 struct xfs_inode *ip = XFS_I(inode);
84 switch (VFS_I(ip)->i_mode & S_IFMT) {
88 xfs_idestroy_fork(ip, XFS_DATA_FORK);
93 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
95 xfs_idestroy_fork(ip, XFS_COW_FORK);
98 ASSERT(!test_bit(XFS_LI_IN_AIL,
99 &ip->i_itemp->ili_item.li_flags));
100 xfs_inode_item_destroy(ip);
104 kmem_zone_free(xfs_inode_zone, ip);
109 struct xfs_inode *ip)
111 /* asserts to verify all state is correct here */
112 ASSERT(atomic_read(&ip->i_pincount) == 0);
113 XFS_STATS_DEC(ip->i_mount, vn_active);
115 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
120 struct xfs_inode *ip)
122 ASSERT(!xfs_isiflocked(ip));
125 * Because we use RCU freeing we need to ensure the inode always
126 * appears to be reclaimed with an invalid inode number when in the
127 * free state. The ip->i_flags_lock provides the barrier against lookup
130 spin_lock(&ip->i_flags_lock);
131 ip->i_flags = XFS_IRECLAIM;
133 spin_unlock(&ip->i_flags_lock);
135 __xfs_inode_free(ip);
139 * Queue a new inode reclaim pass if there are reclaimable inodes and there
140 * isn't a reclaim pass already in progress. By default it runs every 5s based
141 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
142 * tunable, but that can be done if this method proves to be ineffective or too
146 xfs_reclaim_work_queue(
147 struct xfs_mount *mp)
151 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
152 queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
153 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
159 * This is a fast pass over the inode cache to try to get reclaim moving on as
160 * many inodes as possible in a short period of time. It kicks itself every few
161 * seconds, as well as being kicked by the inode cache shrinker when memory
162 * goes low. It scans as quickly as possible avoiding locked inodes or those
163 * already being flushed, and once done schedules a future pass.
167 struct work_struct *work)
169 struct xfs_mount *mp = container_of(to_delayed_work(work),
170 struct xfs_mount, m_reclaim_work);
172 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
173 xfs_reclaim_work_queue(mp);
177 xfs_perag_set_reclaim_tag(
178 struct xfs_perag *pag)
180 struct xfs_mount *mp = pag->pag_mount;
182 lockdep_assert_held(&pag->pag_ici_lock);
183 if (pag->pag_ici_reclaimable++)
186 /* propagate the reclaim tag up into the perag radix tree */
187 spin_lock(&mp->m_perag_lock);
188 radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno,
189 XFS_ICI_RECLAIM_TAG);
190 spin_unlock(&mp->m_perag_lock);
192 /* schedule periodic background inode reclaim */
193 xfs_reclaim_work_queue(mp);
195 trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
199 xfs_perag_clear_reclaim_tag(
200 struct xfs_perag *pag)
202 struct xfs_mount *mp = pag->pag_mount;
204 lockdep_assert_held(&pag->pag_ici_lock);
205 if (--pag->pag_ici_reclaimable)
208 /* clear the reclaim tag from the perag radix tree */
209 spin_lock(&mp->m_perag_lock);
210 radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno,
211 XFS_ICI_RECLAIM_TAG);
212 spin_unlock(&mp->m_perag_lock);
213 trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
218 * We set the inode flag atomically with the radix tree tag.
219 * Once we get tag lookups on the radix tree, this inode flag
223 xfs_inode_set_reclaim_tag(
224 struct xfs_inode *ip)
226 struct xfs_mount *mp = ip->i_mount;
227 struct xfs_perag *pag;
229 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
230 spin_lock(&pag->pag_ici_lock);
231 spin_lock(&ip->i_flags_lock);
233 radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino),
234 XFS_ICI_RECLAIM_TAG);
235 xfs_perag_set_reclaim_tag(pag);
236 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
238 spin_unlock(&ip->i_flags_lock);
239 spin_unlock(&pag->pag_ici_lock);
244 xfs_inode_clear_reclaim_tag(
245 struct xfs_perag *pag,
248 radix_tree_tag_clear(&pag->pag_ici_root,
249 XFS_INO_TO_AGINO(pag->pag_mount, ino),
250 XFS_ICI_RECLAIM_TAG);
251 xfs_perag_clear_reclaim_tag(pag);
256 struct xfs_inode *ip)
258 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT);
259 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT);
262 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
263 if (!xfs_iflags_test(ip, XFS_INEW))
267 finish_wait(wq, &wait.wq_entry);
271 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
272 * part of the structure. This is made more complex by the fact we store
273 * information about the on-disk values in the VFS inode and so we can't just
274 * overwrite the values unconditionally. Hence we save the parameters we
275 * need to retain across reinitialisation, and rewrite them into the VFS inode
276 * after reinitialisation even if it fails.
280 struct xfs_mount *mp,
284 uint32_t nlink = inode->i_nlink;
285 uint32_t generation = inode->i_generation;
286 uint64_t version = inode_peek_iversion(inode);
287 umode_t mode = inode->i_mode;
288 dev_t dev = inode->i_rdev;
290 error = inode_init_always(mp->m_super, inode);
292 set_nlink(inode, nlink);
293 inode->i_generation = generation;
294 inode_set_iversion_queried(inode, version);
295 inode->i_mode = mode;
301 * If we are allocating a new inode, then check what was returned is
302 * actually a free, empty inode. If we are not allocating an inode,
303 * then check we didn't find a free inode.
306 * 0 if the inode free state matches the lookup context
307 * -ENOENT if the inode is free and we are not allocating
308 * -EFSCORRUPTED if there is any state mismatch at all
311 xfs_iget_check_free_state(
312 struct xfs_inode *ip,
315 if (flags & XFS_IGET_CREATE) {
316 /* should be a free inode */
317 if (VFS_I(ip)->i_mode != 0) {
318 xfs_warn(ip->i_mount,
319 "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
320 ip->i_ino, VFS_I(ip)->i_mode);
321 return -EFSCORRUPTED;
324 if (ip->i_d.di_nblocks != 0) {
325 xfs_warn(ip->i_mount,
326 "Corruption detected! Free inode 0x%llx has blocks allocated!",
328 return -EFSCORRUPTED;
333 /* should be an allocated inode */
334 if (VFS_I(ip)->i_mode == 0)
341 * Check the validity of the inode we just found it the cache
345 struct xfs_perag *pag,
346 struct xfs_inode *ip,
349 int lock_flags) __releases(RCU)
351 struct inode *inode = VFS_I(ip);
352 struct xfs_mount *mp = ip->i_mount;
356 * check for re-use of an inode within an RCU grace period due to the
357 * radix tree nodes not being updated yet. We monitor for this by
358 * setting the inode number to zero before freeing the inode structure.
359 * If the inode has been reallocated and set up, then the inode number
360 * will not match, so check for that, too.
362 spin_lock(&ip->i_flags_lock);
363 if (ip->i_ino != ino) {
364 trace_xfs_iget_skip(ip);
365 XFS_STATS_INC(mp, xs_ig_frecycle);
372 * If we are racing with another cache hit that is currently
373 * instantiating this inode or currently recycling it out of
374 * reclaimabe state, wait for the initialisation to complete
377 * XXX(hch): eventually we should do something equivalent to
378 * wait_on_inode to wait for these flags to be cleared
379 * instead of polling for it.
381 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
382 trace_xfs_iget_skip(ip);
383 XFS_STATS_INC(mp, xs_ig_frecycle);
389 * Check the inode free state is valid. This also detects lookup
390 * racing with unlinks.
392 error = xfs_iget_check_free_state(ip, flags);
397 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
398 * Need to carefully get it back into useable state.
400 if (ip->i_flags & XFS_IRECLAIMABLE) {
401 trace_xfs_iget_reclaim(ip);
403 if (flags & XFS_IGET_INCORE) {
409 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
410 * from stomping over us while we recycle the inode. We can't
411 * clear the radix tree reclaimable tag yet as it requires
412 * pag_ici_lock to be held exclusive.
414 ip->i_flags |= XFS_IRECLAIM;
416 spin_unlock(&ip->i_flags_lock);
419 error = xfs_reinit_inode(mp, inode);
423 * Re-initializing the inode failed, and we are in deep
424 * trouble. Try to re-add it to the reclaim list.
427 spin_lock(&ip->i_flags_lock);
428 wake = !!__xfs_iflags_test(ip, XFS_INEW);
429 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
431 wake_up_bit(&ip->i_flags, __XFS_INEW_BIT);
432 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
433 trace_xfs_iget_reclaim_fail(ip);
437 spin_lock(&pag->pag_ici_lock);
438 spin_lock(&ip->i_flags_lock);
441 * Clear the per-lifetime state in the inode as we are now
442 * effectively a new inode and need to return to the initial
443 * state before reuse occurs.
445 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
446 ip->i_flags |= XFS_INEW;
447 xfs_inode_clear_reclaim_tag(pag, ip->i_ino);
448 inode->i_state = I_NEW;
450 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
451 init_rwsem(&inode->i_rwsem);
453 spin_unlock(&ip->i_flags_lock);
454 spin_unlock(&pag->pag_ici_lock);
456 /* If the VFS inode is being torn down, pause and try again. */
458 trace_xfs_iget_skip(ip);
463 /* We've got a live one. */
464 spin_unlock(&ip->i_flags_lock);
466 trace_xfs_iget_hit(ip);
470 xfs_ilock(ip, lock_flags);
472 if (!(flags & XFS_IGET_INCORE))
473 xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
474 XFS_STATS_INC(mp, xs_ig_found);
479 spin_unlock(&ip->i_flags_lock);
487 struct xfs_mount *mp,
488 struct xfs_perag *pag,
491 struct xfs_inode **ipp,
495 struct xfs_inode *ip;
497 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
500 ip = xfs_inode_alloc(mp, ino);
504 error = xfs_iread(mp, tp, ip, flags);
508 if (!xfs_inode_verify_forks(ip)) {
509 error = -EFSCORRUPTED;
513 trace_xfs_iget_miss(ip);
517 * Check the inode free state is valid. This also detects lookup
518 * racing with unlinks.
520 error = xfs_iget_check_free_state(ip, flags);
525 * Preload the radix tree so we can insert safely under the
526 * write spinlock. Note that we cannot sleep inside the preload
527 * region. Since we can be called from transaction context, don't
528 * recurse into the file system.
530 if (radix_tree_preload(GFP_NOFS)) {
536 * Because the inode hasn't been added to the radix-tree yet it can't
537 * be found by another thread, so we can do the non-sleeping lock here.
540 if (!xfs_ilock_nowait(ip, lock_flags))
545 * These values must be set before inserting the inode into the radix
546 * tree as the moment it is inserted a concurrent lookup (allowed by the
547 * RCU locking mechanism) can find it and that lookup must see that this
548 * is an inode currently under construction (i.e. that XFS_INEW is set).
549 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
550 * memory barrier that ensures this detection works correctly at lookup
554 if (flags & XFS_IGET_DONTCACHE)
555 iflags |= XFS_IDONTCACHE;
559 xfs_iflags_set(ip, iflags);
561 /* insert the new inode */
562 spin_lock(&pag->pag_ici_lock);
563 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
564 if (unlikely(error)) {
565 WARN_ON(error != -EEXIST);
566 XFS_STATS_INC(mp, xs_ig_dup);
568 goto out_preload_end;
570 spin_unlock(&pag->pag_ici_lock);
571 radix_tree_preload_end();
577 spin_unlock(&pag->pag_ici_lock);
578 radix_tree_preload_end();
580 xfs_iunlock(ip, lock_flags);
582 __destroy_inode(VFS_I(ip));
588 * Look up an inode by number in the given file system.
589 * The inode is looked up in the cache held in each AG.
590 * If the inode is found in the cache, initialise the vfs inode
593 * If it is not in core, read it in from the file system's device,
594 * add it to the cache and initialise the vfs inode.
596 * The inode is locked according to the value of the lock_flags parameter.
597 * This flag parameter indicates how and if the inode's IO lock and inode lock
600 * mp -- the mount point structure for the current file system. It points
601 * to the inode hash table.
602 * tp -- a pointer to the current transaction if there is one. This is
603 * simply passed through to the xfs_iread() call.
604 * ino -- the number of the inode desired. This is the unique identifier
605 * within the file system for the inode being requested.
606 * lock_flags -- flags indicating how to lock the inode. See the comment
607 * for xfs_ilock() for a list of valid values.
624 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
625 * doesn't get freed while it's being referenced during a
626 * radix tree traversal here. It assumes this function
627 * aqcuires only the ILOCK (and therefore it has no need to
628 * involve the IOLOCK in this synchronization).
630 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
632 /* reject inode numbers outside existing AGs */
633 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
636 XFS_STATS_INC(mp, xs_ig_attempts);
638 /* get the perag structure and ensure that it's inode capable */
639 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
640 agino = XFS_INO_TO_AGINO(mp, ino);
645 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
648 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
650 goto out_error_or_again;
653 if (flags & XFS_IGET_INCORE) {
655 goto out_error_or_again;
657 XFS_STATS_INC(mp, xs_ig_missed);
659 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
662 goto out_error_or_again;
669 * If we have a real type for an on-disk inode, we can setup the inode
670 * now. If it's a new inode being created, xfs_ialloc will handle it.
672 if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
673 xfs_setup_existing_inode(ip);
677 if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) {
686 * "Is this a cached inode that's also allocated?"
688 * Look up an inode by number in the given file system. If the inode is
689 * in cache and isn't in purgatory, return 1 if the inode is allocated
690 * and 0 if it is not. For all other cases (not in cache, being torn
691 * down, etc.), return a negative error code.
693 * The caller has to prevent inode allocation and freeing activity,
694 * presumably by locking the AGI buffer. This is to ensure that an
695 * inode cannot transition from allocated to freed until the caller is
696 * ready to allow that. If the inode is in an intermediate state (new,
697 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
698 * inode is not in the cache, -ENOENT will be returned. The caller must
699 * deal with these scenarios appropriately.
701 * This is a specialized use case for the online scrubber; if you're
702 * reading this, you probably want xfs_iget.
705 xfs_icache_inode_is_allocated(
706 struct xfs_mount *mp,
707 struct xfs_trans *tp,
711 struct xfs_inode *ip;
714 error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip);
718 *inuse = !!(VFS_I(ip)->i_mode);
724 * The inode lookup is done in batches to keep the amount of lock traffic and
725 * radix tree lookups to a minimum. The batch size is a trade off between
726 * lookup reduction and stack usage. This is in the reclaim path, so we can't
729 #define XFS_LOOKUP_BATCH 32
732 xfs_inode_ag_walk_grab(
733 struct xfs_inode *ip,
736 struct inode *inode = VFS_I(ip);
737 bool newinos = !!(flags & XFS_AGITER_INEW_WAIT);
739 ASSERT(rcu_read_lock_held());
742 * check for stale RCU freed inode
744 * If the inode has been reallocated, it doesn't matter if it's not in
745 * the AG we are walking - we are walking for writeback, so if it
746 * passes all the "valid inode" checks and is dirty, then we'll write
747 * it back anyway. If it has been reallocated and still being
748 * initialised, the XFS_INEW check below will catch it.
750 spin_lock(&ip->i_flags_lock);
752 goto out_unlock_noent;
754 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
755 if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) ||
756 __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM))
757 goto out_unlock_noent;
758 spin_unlock(&ip->i_flags_lock);
760 /* nothing to sync during shutdown */
761 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
762 return -EFSCORRUPTED;
764 /* If we can't grab the inode, it must on it's way to reclaim. */
772 spin_unlock(&ip->i_flags_lock);
778 struct xfs_mount *mp,
779 struct xfs_perag *pag,
780 int (*execute)(struct xfs_inode *ip, int flags,
787 uint32_t first_index;
799 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
806 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
807 (void **)batch, first_index,
810 nr_found = radix_tree_gang_lookup_tag(
812 (void **) batch, first_index,
813 XFS_LOOKUP_BATCH, tag);
821 * Grab the inodes before we drop the lock. if we found
822 * nothing, nr == 0 and the loop will be skipped.
824 for (i = 0; i < nr_found; i++) {
825 struct xfs_inode *ip = batch[i];
827 if (done || xfs_inode_ag_walk_grab(ip, iter_flags))
831 * Update the index for the next lookup. Catch
832 * overflows into the next AG range which can occur if
833 * we have inodes in the last block of the AG and we
834 * are currently pointing to the last inode.
836 * Because we may see inodes that are from the wrong AG
837 * due to RCU freeing and reallocation, only update the
838 * index if it lies in this AG. It was a race that lead
839 * us to see this inode, so another lookup from the
840 * same index will not find it again.
842 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
844 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
845 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
849 /* unlock now we've grabbed the inodes. */
852 for (i = 0; i < nr_found; i++) {
855 if ((iter_flags & XFS_AGITER_INEW_WAIT) &&
856 xfs_iflags_test(batch[i], XFS_INEW))
857 xfs_inew_wait(batch[i]);
858 error = execute(batch[i], flags, args);
860 if (error == -EAGAIN) {
864 if (error && last_error != -EFSCORRUPTED)
868 /* bail out if the filesystem is corrupted. */
869 if (error == -EFSCORRUPTED)
874 } while (nr_found && !done);
884 * Background scanning to trim post-EOF preallocated space. This is queued
885 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
889 struct xfs_mount *mp)
892 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
893 queue_delayed_work(mp->m_eofblocks_workqueue,
894 &mp->m_eofblocks_work,
895 msecs_to_jiffies(xfs_eofb_secs * 1000));
900 xfs_eofblocks_worker(
901 struct work_struct *work)
903 struct xfs_mount *mp = container_of(to_delayed_work(work),
904 struct xfs_mount, m_eofblocks_work);
905 xfs_icache_free_eofblocks(mp, NULL);
906 xfs_queue_eofblocks(mp);
910 * Background scanning to trim preallocated CoW space. This is queued
911 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
912 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
916 struct xfs_mount *mp)
919 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
920 queue_delayed_work(mp->m_eofblocks_workqueue,
921 &mp->m_cowblocks_work,
922 msecs_to_jiffies(xfs_cowb_secs * 1000));
927 xfs_cowblocks_worker(
928 struct work_struct *work)
930 struct xfs_mount *mp = container_of(to_delayed_work(work),
931 struct xfs_mount, m_cowblocks_work);
932 xfs_icache_free_cowblocks(mp, NULL);
933 xfs_queue_cowblocks(mp);
937 xfs_inode_ag_iterator_flags(
938 struct xfs_mount *mp,
939 int (*execute)(struct xfs_inode *ip, int flags,
945 struct xfs_perag *pag;
951 while ((pag = xfs_perag_get(mp, ag))) {
952 ag = pag->pag_agno + 1;
953 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1,
958 if (error == -EFSCORRUPTED)
966 xfs_inode_ag_iterator(
967 struct xfs_mount *mp,
968 int (*execute)(struct xfs_inode *ip, int flags,
973 return xfs_inode_ag_iterator_flags(mp, execute, flags, args, 0);
977 xfs_inode_ag_iterator_tag(
978 struct xfs_mount *mp,
979 int (*execute)(struct xfs_inode *ip, int flags,
985 struct xfs_perag *pag;
991 while ((pag = xfs_perag_get_tag(mp, ag, tag))) {
992 ag = pag->pag_agno + 1;
993 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag,
998 if (error == -EFSCORRUPTED)
1006 * Grab the inode for reclaim exclusively.
1007 * Return 0 if we grabbed it, non-zero otherwise.
1010 xfs_reclaim_inode_grab(
1011 struct xfs_inode *ip,
1014 ASSERT(rcu_read_lock_held());
1016 /* quick check for stale RCU freed inode */
1021 * If we are asked for non-blocking operation, do unlocked checks to
1022 * see if the inode already is being flushed or in reclaim to avoid
1025 if ((flags & SYNC_TRYLOCK) &&
1026 __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
1030 * The radix tree lock here protects a thread in xfs_iget from racing
1031 * with us starting reclaim on the inode. Once we have the
1032 * XFS_IRECLAIM flag set it will not touch us.
1034 * Due to RCU lookup, we may find inodes that have been freed and only
1035 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
1036 * aren't candidates for reclaim at all, so we must check the
1037 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
1039 spin_lock(&ip->i_flags_lock);
1040 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
1041 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
1042 /* not a reclaim candidate. */
1043 spin_unlock(&ip->i_flags_lock);
1046 __xfs_iflags_set(ip, XFS_IRECLAIM);
1047 spin_unlock(&ip->i_flags_lock);
1052 * Inodes in different states need to be treated differently. The following
1053 * table lists the inode states and the reclaim actions necessary:
1055 * inode state iflush ret required action
1056 * --------------- ---------- ---------------
1058 * shutdown EIO unpin and reclaim
1059 * clean, unpinned 0 reclaim
1060 * stale, unpinned 0 reclaim
1061 * clean, pinned(*) 0 requeue
1062 * stale, pinned EAGAIN requeue
1063 * dirty, async - requeue
1064 * dirty, sync 0 reclaim
1066 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1067 * handled anyway given the order of checks implemented.
1069 * Also, because we get the flush lock first, we know that any inode that has
1070 * been flushed delwri has had the flush completed by the time we check that
1071 * the inode is clean.
1073 * Note that because the inode is flushed delayed write by AIL pushing, the
1074 * flush lock may already be held here and waiting on it can result in very
1075 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1076 * the caller should push the AIL first before trying to reclaim inodes to
1077 * minimise the amount of time spent waiting. For background relaim, we only
1078 * bother to reclaim clean inodes anyway.
1080 * Hence the order of actions after gaining the locks should be:
1082 * shutdown => unpin and reclaim
1083 * pinned, async => requeue
1084 * pinned, sync => unpin
1087 * dirty, async => requeue
1088 * dirty, sync => flush, wait and reclaim
1092 struct xfs_inode *ip,
1093 struct xfs_perag *pag,
1096 struct xfs_buf *bp = NULL;
1097 xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
1102 xfs_ilock(ip, XFS_ILOCK_EXCL);
1103 if (!xfs_iflock_nowait(ip)) {
1104 if (!(sync_mode & SYNC_WAIT))
1109 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1110 xfs_iunpin_wait(ip);
1111 /* xfs_iflush_abort() drops the flush lock */
1112 xfs_iflush_abort(ip, false);
1115 if (xfs_ipincount(ip)) {
1116 if (!(sync_mode & SYNC_WAIT))
1118 xfs_iunpin_wait(ip);
1120 if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) {
1126 * Never flush out dirty data during non-blocking reclaim, as it would
1127 * just contend with AIL pushing trying to do the same job.
1129 if (!(sync_mode & SYNC_WAIT))
1133 * Now we have an inode that needs flushing.
1135 * Note that xfs_iflush will never block on the inode buffer lock, as
1136 * xfs_ifree_cluster() can lock the inode buffer before it locks the
1137 * ip->i_lock, and we are doing the exact opposite here. As a result,
1138 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1139 * result in an ABBA deadlock with xfs_ifree_cluster().
1141 * As xfs_ifree_cluser() must gather all inodes that are active in the
1142 * cache to mark them stale, if we hit this case we don't actually want
1143 * to do IO here - we want the inode marked stale so we can simply
1144 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1145 * inode, back off and try again. Hopefully the next pass through will
1146 * see the stale flag set on the inode.
1148 error = xfs_iflush(ip, &bp);
1149 if (error == -EAGAIN) {
1150 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1151 /* backoff longer than in xfs_ifree_cluster */
1157 error = xfs_bwrite(bp);
1162 ASSERT(!xfs_isiflocked(ip));
1165 * Because we use RCU freeing we need to ensure the inode always appears
1166 * to be reclaimed with an invalid inode number when in the free state.
1167 * We do this as early as possible under the ILOCK so that
1168 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1169 * detect races with us here. By doing this, we guarantee that once
1170 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1171 * it will see either a valid inode that will serialise correctly, or it
1172 * will see an invalid inode that it can skip.
1174 spin_lock(&ip->i_flags_lock);
1175 ip->i_flags = XFS_IRECLAIM;
1177 spin_unlock(&ip->i_flags_lock);
1179 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1181 XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
1183 * Remove the inode from the per-AG radix tree.
1185 * Because radix_tree_delete won't complain even if the item was never
1186 * added to the tree assert that it's been there before to catch
1187 * problems with the inode life time early on.
1189 spin_lock(&pag->pag_ici_lock);
1190 if (!radix_tree_delete(&pag->pag_ici_root,
1191 XFS_INO_TO_AGINO(ip->i_mount, ino)))
1193 xfs_perag_clear_reclaim_tag(pag);
1194 spin_unlock(&pag->pag_ici_lock);
1197 * Here we do an (almost) spurious inode lock in order to coordinate
1198 * with inode cache radix tree lookups. This is because the lookup
1199 * can reference the inodes in the cache without taking references.
1201 * We make that OK here by ensuring that we wait until the inode is
1202 * unlocked after the lookup before we go ahead and free it.
1204 xfs_ilock(ip, XFS_ILOCK_EXCL);
1205 xfs_qm_dqdetach(ip);
1206 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1208 __xfs_inode_free(ip);
1214 xfs_iflags_clear(ip, XFS_IRECLAIM);
1215 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1217 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1218 * a short while. However, this just burns CPU time scanning the tree
1219 * waiting for IO to complete and the reclaim work never goes back to
1220 * the idle state. Instead, return 0 to let the next scheduled
1221 * background reclaim attempt to reclaim the inode again.
1227 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1228 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1229 * then a shut down during filesystem unmount reclaim walk leak all the
1230 * unreclaimed inodes.
1233 xfs_reclaim_inodes_ag(
1234 struct xfs_mount *mp,
1238 struct xfs_perag *pag;
1242 int trylock = flags & SYNC_TRYLOCK;
1248 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1249 unsigned long first_index = 0;
1253 ag = pag->pag_agno + 1;
1256 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1261 first_index = pag->pag_ici_reclaim_cursor;
1263 mutex_lock(&pag->pag_ici_reclaim_lock);
1266 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1270 nr_found = radix_tree_gang_lookup_tag(
1272 (void **)batch, first_index,
1274 XFS_ICI_RECLAIM_TAG);
1282 * Grab the inodes before we drop the lock. if we found
1283 * nothing, nr == 0 and the loop will be skipped.
1285 for (i = 0; i < nr_found; i++) {
1286 struct xfs_inode *ip = batch[i];
1288 if (done || xfs_reclaim_inode_grab(ip, flags))
1292 * Update the index for the next lookup. Catch
1293 * overflows into the next AG range which can
1294 * occur if we have inodes in the last block of
1295 * the AG and we are currently pointing to the
1298 * Because we may see inodes that are from the
1299 * wrong AG due to RCU freeing and
1300 * reallocation, only update the index if it
1301 * lies in this AG. It was a race that lead us
1302 * to see this inode, so another lookup from
1303 * the same index will not find it again.
1305 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1308 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1309 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1313 /* unlock now we've grabbed the inodes. */
1316 for (i = 0; i < nr_found; i++) {
1319 error = xfs_reclaim_inode(batch[i], pag, flags);
1320 if (error && last_error != -EFSCORRUPTED)
1324 *nr_to_scan -= XFS_LOOKUP_BATCH;
1328 } while (nr_found && !done && *nr_to_scan > 0);
1330 if (trylock && !done)
1331 pag->pag_ici_reclaim_cursor = first_index;
1333 pag->pag_ici_reclaim_cursor = 0;
1334 mutex_unlock(&pag->pag_ici_reclaim_lock);
1339 * if we skipped any AG, and we still have scan count remaining, do
1340 * another pass this time using blocking reclaim semantics (i.e
1341 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1342 * ensure that when we get more reclaimers than AGs we block rather
1343 * than spin trying to execute reclaim.
1345 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1357 int nr_to_scan = INT_MAX;
1359 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1363 * Scan a certain number of inodes for reclaim.
1365 * When called we make sure that there is a background (fast) inode reclaim in
1366 * progress, while we will throttle the speed of reclaim via doing synchronous
1367 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1368 * them to be cleaned, which we hope will not be very long due to the
1369 * background walker having already kicked the IO off on those dirty inodes.
1372 xfs_reclaim_inodes_nr(
1373 struct xfs_mount *mp,
1376 /* kick background reclaimer and push the AIL */
1377 xfs_reclaim_work_queue(mp);
1378 xfs_ail_push_all(mp->m_ail);
1380 return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1384 * Return the number of reclaimable inodes in the filesystem for
1385 * the shrinker to determine how much to reclaim.
1388 xfs_reclaim_inodes_count(
1389 struct xfs_mount *mp)
1391 struct xfs_perag *pag;
1392 xfs_agnumber_t ag = 0;
1393 int reclaimable = 0;
1395 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1396 ag = pag->pag_agno + 1;
1397 reclaimable += pag->pag_ici_reclaimable;
1405 struct xfs_inode *ip,
1406 struct xfs_eofblocks *eofb)
1408 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1409 !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1412 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1413 !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1416 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1417 xfs_get_projid(ip) != eofb->eof_prid)
1424 * A union-based inode filtering algorithm. Process the inode if any of the
1425 * criteria match. This is for global/internal scans only.
1428 xfs_inode_match_id_union(
1429 struct xfs_inode *ip,
1430 struct xfs_eofblocks *eofb)
1432 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1433 uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1436 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1437 gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1440 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1441 xfs_get_projid(ip) == eofb->eof_prid)
1448 xfs_inode_free_eofblocks(
1449 struct xfs_inode *ip,
1454 struct xfs_eofblocks *eofb = args;
1457 if (!xfs_can_free_eofblocks(ip, false)) {
1458 /* inode could be preallocated or append-only */
1459 trace_xfs_inode_free_eofblocks_invalid(ip);
1460 xfs_inode_clear_eofblocks_tag(ip);
1465 * If the mapping is dirty the operation can block and wait for some
1466 * time. Unless we are waiting, skip it.
1468 if (!(flags & SYNC_WAIT) &&
1469 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1473 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1474 match = xfs_inode_match_id_union(ip, eofb);
1476 match = xfs_inode_match_id(ip, eofb);
1480 /* skip the inode if the file size is too small */
1481 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1482 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1487 * If the caller is waiting, return -EAGAIN to keep the background
1488 * scanner moving and revisit the inode in a subsequent pass.
1490 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1491 if (flags & SYNC_WAIT)
1495 ret = xfs_free_eofblocks(ip);
1496 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1502 __xfs_icache_free_eofblocks(
1503 struct xfs_mount *mp,
1504 struct xfs_eofblocks *eofb,
1505 int (*execute)(struct xfs_inode *ip, int flags,
1509 int flags = SYNC_TRYLOCK;
1511 if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
1514 return xfs_inode_ag_iterator_tag(mp, execute, flags,
1519 xfs_icache_free_eofblocks(
1520 struct xfs_mount *mp,
1521 struct xfs_eofblocks *eofb)
1523 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks,
1524 XFS_ICI_EOFBLOCKS_TAG);
1528 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1529 * multiple quotas, we don't know exactly which quota caused an allocation
1530 * failure. We make a best effort by including each quota under low free space
1531 * conditions (less than 1% free space) in the scan.
1534 __xfs_inode_free_quota_eofblocks(
1535 struct xfs_inode *ip,
1536 int (*execute)(struct xfs_mount *mp,
1537 struct xfs_eofblocks *eofb))
1540 struct xfs_eofblocks eofb = {0};
1541 struct xfs_dquot *dq;
1544 * Run a sync scan to increase effectiveness and use the union filter to
1545 * cover all applicable quotas in a single scan.
1547 eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC;
1549 if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) {
1550 dq = xfs_inode_dquot(ip, XFS_DQ_USER);
1551 if (dq && xfs_dquot_lowsp(dq)) {
1552 eofb.eof_uid = VFS_I(ip)->i_uid;
1553 eofb.eof_flags |= XFS_EOF_FLAGS_UID;
1558 if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) {
1559 dq = xfs_inode_dquot(ip, XFS_DQ_GROUP);
1560 if (dq && xfs_dquot_lowsp(dq)) {
1561 eofb.eof_gid = VFS_I(ip)->i_gid;
1562 eofb.eof_flags |= XFS_EOF_FLAGS_GID;
1568 execute(ip->i_mount, &eofb);
1574 xfs_inode_free_quota_eofblocks(
1575 struct xfs_inode *ip)
1577 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks);
1580 static inline unsigned long
1585 case XFS_ICI_EOFBLOCKS_TAG:
1586 return XFS_IEOFBLOCKS;
1587 case XFS_ICI_COWBLOCKS_TAG:
1588 return XFS_ICOWBLOCKS;
1596 __xfs_inode_set_blocks_tag(
1598 void (*execute)(struct xfs_mount *mp),
1599 void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1600 int error, unsigned long caller_ip),
1603 struct xfs_mount *mp = ip->i_mount;
1604 struct xfs_perag *pag;
1608 * Don't bother locking the AG and looking up in the radix trees
1609 * if we already know that we have the tag set.
1611 if (ip->i_flags & xfs_iflag_for_tag(tag))
1613 spin_lock(&ip->i_flags_lock);
1614 ip->i_flags |= xfs_iflag_for_tag(tag);
1615 spin_unlock(&ip->i_flags_lock);
1617 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1618 spin_lock(&pag->pag_ici_lock);
1620 tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
1621 radix_tree_tag_set(&pag->pag_ici_root,
1622 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1624 /* propagate the eofblocks tag up into the perag radix tree */
1625 spin_lock(&ip->i_mount->m_perag_lock);
1626 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1627 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1629 spin_unlock(&ip->i_mount->m_perag_lock);
1631 /* kick off background trimming */
1632 execute(ip->i_mount);
1634 set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
1637 spin_unlock(&pag->pag_ici_lock);
1642 xfs_inode_set_eofblocks_tag(
1645 trace_xfs_inode_set_eofblocks_tag(ip);
1646 return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks,
1647 trace_xfs_perag_set_eofblocks,
1648 XFS_ICI_EOFBLOCKS_TAG);
1652 __xfs_inode_clear_blocks_tag(
1654 void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1655 int error, unsigned long caller_ip),
1658 struct xfs_mount *mp = ip->i_mount;
1659 struct xfs_perag *pag;
1661 spin_lock(&ip->i_flags_lock);
1662 ip->i_flags &= ~xfs_iflag_for_tag(tag);
1663 spin_unlock(&ip->i_flags_lock);
1665 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1666 spin_lock(&pag->pag_ici_lock);
1668 radix_tree_tag_clear(&pag->pag_ici_root,
1669 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1670 if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
1671 /* clear the eofblocks tag from the perag radix tree */
1672 spin_lock(&ip->i_mount->m_perag_lock);
1673 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1674 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1676 spin_unlock(&ip->i_mount->m_perag_lock);
1677 clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
1680 spin_unlock(&pag->pag_ici_lock);
1685 xfs_inode_clear_eofblocks_tag(
1688 trace_xfs_inode_clear_eofblocks_tag(ip);
1689 return __xfs_inode_clear_blocks_tag(ip,
1690 trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
1694 * Set ourselves up to free CoW blocks from this file. If it's already clean
1695 * then we can bail out quickly, but otherwise we must back off if the file
1696 * is undergoing some kind of write.
1699 xfs_prep_free_cowblocks(
1700 struct xfs_inode *ip)
1703 * Just clear the tag if we have an empty cow fork or none at all. It's
1704 * possible the inode was fully unshared since it was originally tagged.
1706 if (!xfs_inode_has_cow_data(ip)) {
1707 trace_xfs_inode_free_cowblocks_invalid(ip);
1708 xfs_inode_clear_cowblocks_tag(ip);
1713 * If the mapping is dirty or under writeback we cannot touch the
1714 * CoW fork. Leave it alone if we're in the midst of a directio.
1716 if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1717 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1718 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1719 atomic_read(&VFS_I(ip)->i_dio_count))
1726 * Automatic CoW Reservation Freeing
1728 * These functions automatically garbage collect leftover CoW reservations
1729 * that were made on behalf of a cowextsize hint when we start to run out
1730 * of quota or when the reservations sit around for too long. If the file
1731 * has dirty pages or is undergoing writeback, its CoW reservations will
1734 * The actual garbage collection piggybacks off the same code that runs
1735 * the speculative EOF preallocation garbage collector.
1738 xfs_inode_free_cowblocks(
1739 struct xfs_inode *ip,
1743 struct xfs_eofblocks *eofb = args;
1747 if (!xfs_prep_free_cowblocks(ip))
1751 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1752 match = xfs_inode_match_id_union(ip, eofb);
1754 match = xfs_inode_match_id(ip, eofb);
1758 /* skip the inode if the file size is too small */
1759 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1760 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1764 /* Free the CoW blocks */
1765 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1766 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1769 * Check again, nobody else should be able to dirty blocks or change
1770 * the reflink iflag now that we have the first two locks held.
1772 if (xfs_prep_free_cowblocks(ip))
1773 ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1775 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1776 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1782 xfs_icache_free_cowblocks(
1783 struct xfs_mount *mp,
1784 struct xfs_eofblocks *eofb)
1786 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks,
1787 XFS_ICI_COWBLOCKS_TAG);
1791 xfs_inode_free_quota_cowblocks(
1792 struct xfs_inode *ip)
1794 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks);
1798 xfs_inode_set_cowblocks_tag(
1801 trace_xfs_inode_set_cowblocks_tag(ip);
1802 return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks,
1803 trace_xfs_perag_set_cowblocks,
1804 XFS_ICI_COWBLOCKS_TAG);
1808 xfs_inode_clear_cowblocks_tag(
1811 trace_xfs_inode_clear_cowblocks_tag(ip);
1812 return __xfs_inode_clear_blocks_tag(ip,
1813 trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG);
1816 /* Disable post-EOF and CoW block auto-reclamation. */
1818 xfs_icache_disable_reclaim(
1819 struct xfs_mount *mp)
1821 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1822 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1825 /* Enable post-EOF and CoW block auto-reclamation. */
1827 xfs_icache_enable_reclaim(
1828 struct xfs_mount *mp)
1830 xfs_queue_eofblocks(mp);
1831 xfs_queue_cowblocks(mp);