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));
75 INIT_WORK(&ip->i_ioend_work, xfs_end_io);
76 INIT_LIST_HEAD(&ip->i_ioend_list);
77 spin_lock_init(&ip->i_ioend_lock);
83 xfs_inode_free_callback(
84 struct rcu_head *head)
86 struct inode *inode = container_of(head, struct inode, i_rcu);
87 struct xfs_inode *ip = XFS_I(inode);
89 switch (VFS_I(ip)->i_mode & S_IFMT) {
93 xfs_idestroy_fork(ip, XFS_DATA_FORK);
98 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
100 xfs_idestroy_fork(ip, XFS_COW_FORK);
103 ASSERT(!test_bit(XFS_LI_IN_AIL,
104 &ip->i_itemp->ili_item.li_flags));
105 xfs_inode_item_destroy(ip);
109 kmem_zone_free(xfs_inode_zone, ip);
114 struct xfs_inode *ip)
116 /* asserts to verify all state is correct here */
117 ASSERT(atomic_read(&ip->i_pincount) == 0);
118 XFS_STATS_DEC(ip->i_mount, vn_active);
120 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
125 struct xfs_inode *ip)
127 ASSERT(!xfs_isiflocked(ip));
130 * Because we use RCU freeing we need to ensure the inode always
131 * appears to be reclaimed with an invalid inode number when in the
132 * free state. The ip->i_flags_lock provides the barrier against lookup
135 spin_lock(&ip->i_flags_lock);
136 ip->i_flags = XFS_IRECLAIM;
138 spin_unlock(&ip->i_flags_lock);
140 __xfs_inode_free(ip);
144 * Queue a new inode reclaim pass if there are reclaimable inodes and there
145 * isn't a reclaim pass already in progress. By default it runs every 5s based
146 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
147 * tunable, but that can be done if this method proves to be ineffective or too
151 xfs_reclaim_work_queue(
152 struct xfs_mount *mp)
156 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
157 queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
158 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
164 * This is a fast pass over the inode cache to try to get reclaim moving on as
165 * many inodes as possible in a short period of time. It kicks itself every few
166 * seconds, as well as being kicked by the inode cache shrinker when memory
167 * goes low. It scans as quickly as possible avoiding locked inodes or those
168 * already being flushed, and once done schedules a future pass.
172 struct work_struct *work)
174 struct xfs_mount *mp = container_of(to_delayed_work(work),
175 struct xfs_mount, m_reclaim_work);
177 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
178 xfs_reclaim_work_queue(mp);
182 xfs_perag_set_reclaim_tag(
183 struct xfs_perag *pag)
185 struct xfs_mount *mp = pag->pag_mount;
187 lockdep_assert_held(&pag->pag_ici_lock);
188 if (pag->pag_ici_reclaimable++)
191 /* propagate the reclaim tag up into the perag radix tree */
192 spin_lock(&mp->m_perag_lock);
193 radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno,
194 XFS_ICI_RECLAIM_TAG);
195 spin_unlock(&mp->m_perag_lock);
197 /* schedule periodic background inode reclaim */
198 xfs_reclaim_work_queue(mp);
200 trace_xfs_perag_set_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
204 xfs_perag_clear_reclaim_tag(
205 struct xfs_perag *pag)
207 struct xfs_mount *mp = pag->pag_mount;
209 lockdep_assert_held(&pag->pag_ici_lock);
210 if (--pag->pag_ici_reclaimable)
213 /* clear the reclaim tag from the perag radix tree */
214 spin_lock(&mp->m_perag_lock);
215 radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno,
216 XFS_ICI_RECLAIM_TAG);
217 spin_unlock(&mp->m_perag_lock);
218 trace_xfs_perag_clear_reclaim(mp, pag->pag_agno, -1, _RET_IP_);
223 * We set the inode flag atomically with the radix tree tag.
224 * Once we get tag lookups on the radix tree, this inode flag
228 xfs_inode_set_reclaim_tag(
229 struct xfs_inode *ip)
231 struct xfs_mount *mp = ip->i_mount;
232 struct xfs_perag *pag;
234 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
235 spin_lock(&pag->pag_ici_lock);
236 spin_lock(&ip->i_flags_lock);
238 radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino),
239 XFS_ICI_RECLAIM_TAG);
240 xfs_perag_set_reclaim_tag(pag);
241 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
243 spin_unlock(&ip->i_flags_lock);
244 spin_unlock(&pag->pag_ici_lock);
249 xfs_inode_clear_reclaim_tag(
250 struct xfs_perag *pag,
253 radix_tree_tag_clear(&pag->pag_ici_root,
254 XFS_INO_TO_AGINO(pag->pag_mount, ino),
255 XFS_ICI_RECLAIM_TAG);
256 xfs_perag_clear_reclaim_tag(pag);
261 struct xfs_inode *ip)
263 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT);
264 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT);
267 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
268 if (!xfs_iflags_test(ip, XFS_INEW))
272 finish_wait(wq, &wait.wq_entry);
276 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
277 * part of the structure. This is made more complex by the fact we store
278 * information about the on-disk values in the VFS inode and so we can't just
279 * overwrite the values unconditionally. Hence we save the parameters we
280 * need to retain across reinitialisation, and rewrite them into the VFS inode
281 * after reinitialisation even if it fails.
285 struct xfs_mount *mp,
289 uint32_t nlink = inode->i_nlink;
290 uint32_t generation = inode->i_generation;
291 uint64_t version = inode_peek_iversion(inode);
292 umode_t mode = inode->i_mode;
293 dev_t dev = inode->i_rdev;
295 error = inode_init_always(mp->m_super, inode);
297 set_nlink(inode, nlink);
298 inode->i_generation = generation;
299 inode_set_iversion_queried(inode, version);
300 inode->i_mode = mode;
306 * If we are allocating a new inode, then check what was returned is
307 * actually a free, empty inode. If we are not allocating an inode,
308 * then check we didn't find a free inode.
311 * 0 if the inode free state matches the lookup context
312 * -ENOENT if the inode is free and we are not allocating
313 * -EFSCORRUPTED if there is any state mismatch at all
316 xfs_iget_check_free_state(
317 struct xfs_inode *ip,
320 if (flags & XFS_IGET_CREATE) {
321 /* should be a free inode */
322 if (VFS_I(ip)->i_mode != 0) {
323 xfs_warn(ip->i_mount,
324 "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
325 ip->i_ino, VFS_I(ip)->i_mode);
326 return -EFSCORRUPTED;
329 if (ip->i_d.di_nblocks != 0) {
330 xfs_warn(ip->i_mount,
331 "Corruption detected! Free inode 0x%llx has blocks allocated!",
333 return -EFSCORRUPTED;
338 /* should be an allocated inode */
339 if (VFS_I(ip)->i_mode == 0)
346 * Check the validity of the inode we just found it the cache
350 struct xfs_perag *pag,
351 struct xfs_inode *ip,
354 int lock_flags) __releases(RCU)
356 struct inode *inode = VFS_I(ip);
357 struct xfs_mount *mp = ip->i_mount;
361 * check for re-use of an inode within an RCU grace period due to the
362 * radix tree nodes not being updated yet. We monitor for this by
363 * setting the inode number to zero before freeing the inode structure.
364 * If the inode has been reallocated and set up, then the inode number
365 * will not match, so check for that, too.
367 spin_lock(&ip->i_flags_lock);
368 if (ip->i_ino != ino) {
369 trace_xfs_iget_skip(ip);
370 XFS_STATS_INC(mp, xs_ig_frecycle);
377 * If we are racing with another cache hit that is currently
378 * instantiating this inode or currently recycling it out of
379 * reclaimabe state, wait for the initialisation to complete
382 * XXX(hch): eventually we should do something equivalent to
383 * wait_on_inode to wait for these flags to be cleared
384 * instead of polling for it.
386 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
387 trace_xfs_iget_skip(ip);
388 XFS_STATS_INC(mp, xs_ig_frecycle);
394 * Check the inode free state is valid. This also detects lookup
395 * racing with unlinks.
397 error = xfs_iget_check_free_state(ip, flags);
402 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
403 * Need to carefully get it back into useable state.
405 if (ip->i_flags & XFS_IRECLAIMABLE) {
406 trace_xfs_iget_reclaim(ip);
408 if (flags & XFS_IGET_INCORE) {
414 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
415 * from stomping over us while we recycle the inode. We can't
416 * clear the radix tree reclaimable tag yet as it requires
417 * pag_ici_lock to be held exclusive.
419 ip->i_flags |= XFS_IRECLAIM;
421 spin_unlock(&ip->i_flags_lock);
424 error = xfs_reinit_inode(mp, inode);
428 * Re-initializing the inode failed, and we are in deep
429 * trouble. Try to re-add it to the reclaim list.
432 spin_lock(&ip->i_flags_lock);
433 wake = !!__xfs_iflags_test(ip, XFS_INEW);
434 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
436 wake_up_bit(&ip->i_flags, __XFS_INEW_BIT);
437 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
438 trace_xfs_iget_reclaim_fail(ip);
442 spin_lock(&pag->pag_ici_lock);
443 spin_lock(&ip->i_flags_lock);
446 * Clear the per-lifetime state in the inode as we are now
447 * effectively a new inode and need to return to the initial
448 * state before reuse occurs.
450 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
451 ip->i_flags |= XFS_INEW;
452 xfs_inode_clear_reclaim_tag(pag, ip->i_ino);
453 inode->i_state = I_NEW;
457 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
458 init_rwsem(&inode->i_rwsem);
460 spin_unlock(&ip->i_flags_lock);
461 spin_unlock(&pag->pag_ici_lock);
463 /* If the VFS inode is being torn down, pause and try again. */
465 trace_xfs_iget_skip(ip);
470 /* We've got a live one. */
471 spin_unlock(&ip->i_flags_lock);
473 trace_xfs_iget_hit(ip);
477 xfs_ilock(ip, lock_flags);
479 if (!(flags & XFS_IGET_INCORE))
480 xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
481 XFS_STATS_INC(mp, xs_ig_found);
486 spin_unlock(&ip->i_flags_lock);
494 struct xfs_mount *mp,
495 struct xfs_perag *pag,
498 struct xfs_inode **ipp,
502 struct xfs_inode *ip;
504 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
507 ip = xfs_inode_alloc(mp, ino);
511 error = xfs_iread(mp, tp, ip, flags);
515 if (!xfs_inode_verify_forks(ip)) {
516 error = -EFSCORRUPTED;
520 trace_xfs_iget_miss(ip);
524 * Check the inode free state is valid. This also detects lookup
525 * racing with unlinks.
527 error = xfs_iget_check_free_state(ip, flags);
532 * Preload the radix tree so we can insert safely under the
533 * write spinlock. Note that we cannot sleep inside the preload
534 * region. Since we can be called from transaction context, don't
535 * recurse into the file system.
537 if (radix_tree_preload(GFP_NOFS)) {
543 * Because the inode hasn't been added to the radix-tree yet it can't
544 * be found by another thread, so we can do the non-sleeping lock here.
547 if (!xfs_ilock_nowait(ip, lock_flags))
552 * These values must be set before inserting the inode into the radix
553 * tree as the moment it is inserted a concurrent lookup (allowed by the
554 * RCU locking mechanism) can find it and that lookup must see that this
555 * is an inode currently under construction (i.e. that XFS_INEW is set).
556 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
557 * memory barrier that ensures this detection works correctly at lookup
561 if (flags & XFS_IGET_DONTCACHE)
562 iflags |= XFS_IDONTCACHE;
566 xfs_iflags_set(ip, iflags);
568 /* insert the new inode */
569 spin_lock(&pag->pag_ici_lock);
570 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
571 if (unlikely(error)) {
572 WARN_ON(error != -EEXIST);
573 XFS_STATS_INC(mp, xs_ig_dup);
575 goto out_preload_end;
577 spin_unlock(&pag->pag_ici_lock);
578 radix_tree_preload_end();
584 spin_unlock(&pag->pag_ici_lock);
585 radix_tree_preload_end();
587 xfs_iunlock(ip, lock_flags);
589 __destroy_inode(VFS_I(ip));
595 * Look up an inode by number in the given file system.
596 * The inode is looked up in the cache held in each AG.
597 * If the inode is found in the cache, initialise the vfs inode
600 * If it is not in core, read it in from the file system's device,
601 * add it to the cache and initialise the vfs inode.
603 * The inode is locked according to the value of the lock_flags parameter.
604 * This flag parameter indicates how and if the inode's IO lock and inode lock
607 * mp -- the mount point structure for the current file system. It points
608 * to the inode hash table.
609 * tp -- a pointer to the current transaction if there is one. This is
610 * simply passed through to the xfs_iread() call.
611 * ino -- the number of the inode desired. This is the unique identifier
612 * within the file system for the inode being requested.
613 * lock_flags -- flags indicating how to lock the inode. See the comment
614 * for xfs_ilock() for a list of valid values.
631 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
632 * doesn't get freed while it's being referenced during a
633 * radix tree traversal here. It assumes this function
634 * aqcuires only the ILOCK (and therefore it has no need to
635 * involve the IOLOCK in this synchronization).
637 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
639 /* reject inode numbers outside existing AGs */
640 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
643 XFS_STATS_INC(mp, xs_ig_attempts);
645 /* get the perag structure and ensure that it's inode capable */
646 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
647 agino = XFS_INO_TO_AGINO(mp, ino);
652 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
655 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
657 goto out_error_or_again;
660 if (flags & XFS_IGET_INCORE) {
662 goto out_error_or_again;
664 XFS_STATS_INC(mp, xs_ig_missed);
666 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
669 goto out_error_or_again;
676 * If we have a real type for an on-disk inode, we can setup the inode
677 * now. If it's a new inode being created, xfs_ialloc will handle it.
679 if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
680 xfs_setup_existing_inode(ip);
684 if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) {
693 * "Is this a cached inode that's also allocated?"
695 * Look up an inode by number in the given file system. If the inode is
696 * in cache and isn't in purgatory, return 1 if the inode is allocated
697 * and 0 if it is not. For all other cases (not in cache, being torn
698 * down, etc.), return a negative error code.
700 * The caller has to prevent inode allocation and freeing activity,
701 * presumably by locking the AGI buffer. This is to ensure that an
702 * inode cannot transition from allocated to freed until the caller is
703 * ready to allow that. If the inode is in an intermediate state (new,
704 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
705 * inode is not in the cache, -ENOENT will be returned. The caller must
706 * deal with these scenarios appropriately.
708 * This is a specialized use case for the online scrubber; if you're
709 * reading this, you probably want xfs_iget.
712 xfs_icache_inode_is_allocated(
713 struct xfs_mount *mp,
714 struct xfs_trans *tp,
718 struct xfs_inode *ip;
721 error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip);
725 *inuse = !!(VFS_I(ip)->i_mode);
731 * The inode lookup is done in batches to keep the amount of lock traffic and
732 * radix tree lookups to a minimum. The batch size is a trade off between
733 * lookup reduction and stack usage. This is in the reclaim path, so we can't
736 #define XFS_LOOKUP_BATCH 32
739 xfs_inode_ag_walk_grab(
740 struct xfs_inode *ip,
743 struct inode *inode = VFS_I(ip);
744 bool newinos = !!(flags & XFS_AGITER_INEW_WAIT);
746 ASSERT(rcu_read_lock_held());
749 * check for stale RCU freed inode
751 * If the inode has been reallocated, it doesn't matter if it's not in
752 * the AG we are walking - we are walking for writeback, so if it
753 * passes all the "valid inode" checks and is dirty, then we'll write
754 * it back anyway. If it has been reallocated and still being
755 * initialised, the XFS_INEW check below will catch it.
757 spin_lock(&ip->i_flags_lock);
759 goto out_unlock_noent;
761 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
762 if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) ||
763 __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM))
764 goto out_unlock_noent;
765 spin_unlock(&ip->i_flags_lock);
767 /* nothing to sync during shutdown */
768 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
769 return -EFSCORRUPTED;
771 /* If we can't grab the inode, it must on it's way to reclaim. */
779 spin_unlock(&ip->i_flags_lock);
785 struct xfs_mount *mp,
786 struct xfs_perag *pag,
787 int (*execute)(struct xfs_inode *ip, int flags,
794 uint32_t first_index;
806 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
813 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
814 (void **)batch, first_index,
817 nr_found = radix_tree_gang_lookup_tag(
819 (void **) batch, first_index,
820 XFS_LOOKUP_BATCH, tag);
828 * Grab the inodes before we drop the lock. if we found
829 * nothing, nr == 0 and the loop will be skipped.
831 for (i = 0; i < nr_found; i++) {
832 struct xfs_inode *ip = batch[i];
834 if (done || xfs_inode_ag_walk_grab(ip, iter_flags))
838 * Update the index for the next lookup. Catch
839 * overflows into the next AG range which can occur if
840 * we have inodes in the last block of the AG and we
841 * are currently pointing to the last inode.
843 * Because we may see inodes that are from the wrong AG
844 * due to RCU freeing and reallocation, only update the
845 * index if it lies in this AG. It was a race that lead
846 * us to see this inode, so another lookup from the
847 * same index will not find it again.
849 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
851 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
852 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
856 /* unlock now we've grabbed the inodes. */
859 for (i = 0; i < nr_found; i++) {
862 if ((iter_flags & XFS_AGITER_INEW_WAIT) &&
863 xfs_iflags_test(batch[i], XFS_INEW))
864 xfs_inew_wait(batch[i]);
865 error = execute(batch[i], flags, args);
867 if (error == -EAGAIN) {
871 if (error && last_error != -EFSCORRUPTED)
875 /* bail out if the filesystem is corrupted. */
876 if (error == -EFSCORRUPTED)
881 } while (nr_found && !done);
891 * Background scanning to trim post-EOF preallocated space. This is queued
892 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
896 struct xfs_mount *mp)
899 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
900 queue_delayed_work(mp->m_eofblocks_workqueue,
901 &mp->m_eofblocks_work,
902 msecs_to_jiffies(xfs_eofb_secs * 1000));
907 xfs_eofblocks_worker(
908 struct work_struct *work)
910 struct xfs_mount *mp = container_of(to_delayed_work(work),
911 struct xfs_mount, m_eofblocks_work);
912 xfs_icache_free_eofblocks(mp, NULL);
913 xfs_queue_eofblocks(mp);
917 * Background scanning to trim preallocated CoW space. This is queued
918 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
919 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
923 struct xfs_mount *mp)
926 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
927 queue_delayed_work(mp->m_eofblocks_workqueue,
928 &mp->m_cowblocks_work,
929 msecs_to_jiffies(xfs_cowb_secs * 1000));
934 xfs_cowblocks_worker(
935 struct work_struct *work)
937 struct xfs_mount *mp = container_of(to_delayed_work(work),
938 struct xfs_mount, m_cowblocks_work);
939 xfs_icache_free_cowblocks(mp, NULL);
940 xfs_queue_cowblocks(mp);
944 xfs_inode_ag_iterator_flags(
945 struct xfs_mount *mp,
946 int (*execute)(struct xfs_inode *ip, int flags,
952 struct xfs_perag *pag;
958 while ((pag = xfs_perag_get(mp, ag))) {
959 ag = pag->pag_agno + 1;
960 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1,
965 if (error == -EFSCORRUPTED)
973 xfs_inode_ag_iterator(
974 struct xfs_mount *mp,
975 int (*execute)(struct xfs_inode *ip, int flags,
980 return xfs_inode_ag_iterator_flags(mp, execute, flags, args, 0);
984 xfs_inode_ag_iterator_tag(
985 struct xfs_mount *mp,
986 int (*execute)(struct xfs_inode *ip, int flags,
992 struct xfs_perag *pag;
998 while ((pag = xfs_perag_get_tag(mp, ag, tag))) {
999 ag = pag->pag_agno + 1;
1000 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag,
1005 if (error == -EFSCORRUPTED)
1013 * Grab the inode for reclaim exclusively.
1014 * Return 0 if we grabbed it, non-zero otherwise.
1017 xfs_reclaim_inode_grab(
1018 struct xfs_inode *ip,
1021 ASSERT(rcu_read_lock_held());
1023 /* quick check for stale RCU freed inode */
1028 * If we are asked for non-blocking operation, do unlocked checks to
1029 * see if the inode already is being flushed or in reclaim to avoid
1032 if ((flags & SYNC_TRYLOCK) &&
1033 __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
1037 * The radix tree lock here protects a thread in xfs_iget from racing
1038 * with us starting reclaim on the inode. Once we have the
1039 * XFS_IRECLAIM flag set it will not touch us.
1041 * Due to RCU lookup, we may find inodes that have been freed and only
1042 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
1043 * aren't candidates for reclaim at all, so we must check the
1044 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
1046 spin_lock(&ip->i_flags_lock);
1047 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
1048 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
1049 /* not a reclaim candidate. */
1050 spin_unlock(&ip->i_flags_lock);
1053 __xfs_iflags_set(ip, XFS_IRECLAIM);
1054 spin_unlock(&ip->i_flags_lock);
1059 * Inodes in different states need to be treated differently. The following
1060 * table lists the inode states and the reclaim actions necessary:
1062 * inode state iflush ret required action
1063 * --------------- ---------- ---------------
1065 * shutdown EIO unpin and reclaim
1066 * clean, unpinned 0 reclaim
1067 * stale, unpinned 0 reclaim
1068 * clean, pinned(*) 0 requeue
1069 * stale, pinned EAGAIN requeue
1070 * dirty, async - requeue
1071 * dirty, sync 0 reclaim
1073 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1074 * handled anyway given the order of checks implemented.
1076 * Also, because we get the flush lock first, we know that any inode that has
1077 * been flushed delwri has had the flush completed by the time we check that
1078 * the inode is clean.
1080 * Note that because the inode is flushed delayed write by AIL pushing, the
1081 * flush lock may already be held here and waiting on it can result in very
1082 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1083 * the caller should push the AIL first before trying to reclaim inodes to
1084 * minimise the amount of time spent waiting. For background relaim, we only
1085 * bother to reclaim clean inodes anyway.
1087 * Hence the order of actions after gaining the locks should be:
1089 * shutdown => unpin and reclaim
1090 * pinned, async => requeue
1091 * pinned, sync => unpin
1094 * dirty, async => requeue
1095 * dirty, sync => flush, wait and reclaim
1099 struct xfs_inode *ip,
1100 struct xfs_perag *pag,
1103 struct xfs_buf *bp = NULL;
1104 xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
1109 xfs_ilock(ip, XFS_ILOCK_EXCL);
1110 if (!xfs_iflock_nowait(ip)) {
1111 if (!(sync_mode & SYNC_WAIT))
1116 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1117 xfs_iunpin_wait(ip);
1118 /* xfs_iflush_abort() drops the flush lock */
1119 xfs_iflush_abort(ip, false);
1122 if (xfs_ipincount(ip)) {
1123 if (!(sync_mode & SYNC_WAIT))
1125 xfs_iunpin_wait(ip);
1127 if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) {
1133 * Never flush out dirty data during non-blocking reclaim, as it would
1134 * just contend with AIL pushing trying to do the same job.
1136 if (!(sync_mode & SYNC_WAIT))
1140 * Now we have an inode that needs flushing.
1142 * Note that xfs_iflush will never block on the inode buffer lock, as
1143 * xfs_ifree_cluster() can lock the inode buffer before it locks the
1144 * ip->i_lock, and we are doing the exact opposite here. As a result,
1145 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1146 * result in an ABBA deadlock with xfs_ifree_cluster().
1148 * As xfs_ifree_cluser() must gather all inodes that are active in the
1149 * cache to mark them stale, if we hit this case we don't actually want
1150 * to do IO here - we want the inode marked stale so we can simply
1151 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1152 * inode, back off and try again. Hopefully the next pass through will
1153 * see the stale flag set on the inode.
1155 error = xfs_iflush(ip, &bp);
1156 if (error == -EAGAIN) {
1157 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1158 /* backoff longer than in xfs_ifree_cluster */
1164 error = xfs_bwrite(bp);
1169 ASSERT(!xfs_isiflocked(ip));
1172 * Because we use RCU freeing we need to ensure the inode always appears
1173 * to be reclaimed with an invalid inode number when in the free state.
1174 * We do this as early as possible under the ILOCK so that
1175 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1176 * detect races with us here. By doing this, we guarantee that once
1177 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1178 * it will see either a valid inode that will serialise correctly, or it
1179 * will see an invalid inode that it can skip.
1181 spin_lock(&ip->i_flags_lock);
1182 ip->i_flags = XFS_IRECLAIM;
1184 spin_unlock(&ip->i_flags_lock);
1186 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1188 XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
1190 * Remove the inode from the per-AG radix tree.
1192 * Because radix_tree_delete won't complain even if the item was never
1193 * added to the tree assert that it's been there before to catch
1194 * problems with the inode life time early on.
1196 spin_lock(&pag->pag_ici_lock);
1197 if (!radix_tree_delete(&pag->pag_ici_root,
1198 XFS_INO_TO_AGINO(ip->i_mount, ino)))
1200 xfs_perag_clear_reclaim_tag(pag);
1201 spin_unlock(&pag->pag_ici_lock);
1204 * Here we do an (almost) spurious inode lock in order to coordinate
1205 * with inode cache radix tree lookups. This is because the lookup
1206 * can reference the inodes in the cache without taking references.
1208 * We make that OK here by ensuring that we wait until the inode is
1209 * unlocked after the lookup before we go ahead and free it.
1211 xfs_ilock(ip, XFS_ILOCK_EXCL);
1212 xfs_qm_dqdetach(ip);
1213 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1215 __xfs_inode_free(ip);
1221 xfs_iflags_clear(ip, XFS_IRECLAIM);
1222 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1224 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1225 * a short while. However, this just burns CPU time scanning the tree
1226 * waiting for IO to complete and the reclaim work never goes back to
1227 * the idle state. Instead, return 0 to let the next scheduled
1228 * background reclaim attempt to reclaim the inode again.
1234 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1235 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1236 * then a shut down during filesystem unmount reclaim walk leak all the
1237 * unreclaimed inodes.
1240 xfs_reclaim_inodes_ag(
1241 struct xfs_mount *mp,
1245 struct xfs_perag *pag;
1249 int trylock = flags & SYNC_TRYLOCK;
1255 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1256 unsigned long first_index = 0;
1260 ag = pag->pag_agno + 1;
1263 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1268 first_index = pag->pag_ici_reclaim_cursor;
1270 mutex_lock(&pag->pag_ici_reclaim_lock);
1273 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1277 nr_found = radix_tree_gang_lookup_tag(
1279 (void **)batch, first_index,
1281 XFS_ICI_RECLAIM_TAG);
1289 * Grab the inodes before we drop the lock. if we found
1290 * nothing, nr == 0 and the loop will be skipped.
1292 for (i = 0; i < nr_found; i++) {
1293 struct xfs_inode *ip = batch[i];
1295 if (done || xfs_reclaim_inode_grab(ip, flags))
1299 * Update the index for the next lookup. Catch
1300 * overflows into the next AG range which can
1301 * occur if we have inodes in the last block of
1302 * the AG and we are currently pointing to the
1305 * Because we may see inodes that are from the
1306 * wrong AG due to RCU freeing and
1307 * reallocation, only update the index if it
1308 * lies in this AG. It was a race that lead us
1309 * to see this inode, so another lookup from
1310 * the same index will not find it again.
1312 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1315 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1316 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1320 /* unlock now we've grabbed the inodes. */
1323 for (i = 0; i < nr_found; i++) {
1326 error = xfs_reclaim_inode(batch[i], pag, flags);
1327 if (error && last_error != -EFSCORRUPTED)
1331 *nr_to_scan -= XFS_LOOKUP_BATCH;
1335 } while (nr_found && !done && *nr_to_scan > 0);
1337 if (trylock && !done)
1338 pag->pag_ici_reclaim_cursor = first_index;
1340 pag->pag_ici_reclaim_cursor = 0;
1341 mutex_unlock(&pag->pag_ici_reclaim_lock);
1346 * if we skipped any AG, and we still have scan count remaining, do
1347 * another pass this time using blocking reclaim semantics (i.e
1348 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1349 * ensure that when we get more reclaimers than AGs we block rather
1350 * than spin trying to execute reclaim.
1352 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1364 int nr_to_scan = INT_MAX;
1366 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1370 * Scan a certain number of inodes for reclaim.
1372 * When called we make sure that there is a background (fast) inode reclaim in
1373 * progress, while we will throttle the speed of reclaim via doing synchronous
1374 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1375 * them to be cleaned, which we hope will not be very long due to the
1376 * background walker having already kicked the IO off on those dirty inodes.
1379 xfs_reclaim_inodes_nr(
1380 struct xfs_mount *mp,
1383 /* kick background reclaimer and push the AIL */
1384 xfs_reclaim_work_queue(mp);
1385 xfs_ail_push_all(mp->m_ail);
1387 return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1391 * Return the number of reclaimable inodes in the filesystem for
1392 * the shrinker to determine how much to reclaim.
1395 xfs_reclaim_inodes_count(
1396 struct xfs_mount *mp)
1398 struct xfs_perag *pag;
1399 xfs_agnumber_t ag = 0;
1400 int reclaimable = 0;
1402 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1403 ag = pag->pag_agno + 1;
1404 reclaimable += pag->pag_ici_reclaimable;
1412 struct xfs_inode *ip,
1413 struct xfs_eofblocks *eofb)
1415 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1416 !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1419 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1420 !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1423 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1424 xfs_get_projid(ip) != eofb->eof_prid)
1431 * A union-based inode filtering algorithm. Process the inode if any of the
1432 * criteria match. This is for global/internal scans only.
1435 xfs_inode_match_id_union(
1436 struct xfs_inode *ip,
1437 struct xfs_eofblocks *eofb)
1439 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1440 uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1443 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1444 gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1447 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1448 xfs_get_projid(ip) == eofb->eof_prid)
1455 xfs_inode_free_eofblocks(
1456 struct xfs_inode *ip,
1461 struct xfs_eofblocks *eofb = args;
1464 if (!xfs_can_free_eofblocks(ip, false)) {
1465 /* inode could be preallocated or append-only */
1466 trace_xfs_inode_free_eofblocks_invalid(ip);
1467 xfs_inode_clear_eofblocks_tag(ip);
1472 * If the mapping is dirty the operation can block and wait for some
1473 * time. Unless we are waiting, skip it.
1475 if (!(flags & SYNC_WAIT) &&
1476 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1480 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1481 match = xfs_inode_match_id_union(ip, eofb);
1483 match = xfs_inode_match_id(ip, eofb);
1487 /* skip the inode if the file size is too small */
1488 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1489 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1494 * If the caller is waiting, return -EAGAIN to keep the background
1495 * scanner moving and revisit the inode in a subsequent pass.
1497 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1498 if (flags & SYNC_WAIT)
1502 ret = xfs_free_eofblocks(ip);
1503 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1509 __xfs_icache_free_eofblocks(
1510 struct xfs_mount *mp,
1511 struct xfs_eofblocks *eofb,
1512 int (*execute)(struct xfs_inode *ip, int flags,
1516 int flags = SYNC_TRYLOCK;
1518 if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
1521 return xfs_inode_ag_iterator_tag(mp, execute, flags,
1526 xfs_icache_free_eofblocks(
1527 struct xfs_mount *mp,
1528 struct xfs_eofblocks *eofb)
1530 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_eofblocks,
1531 XFS_ICI_EOFBLOCKS_TAG);
1535 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1536 * multiple quotas, we don't know exactly which quota caused an allocation
1537 * failure. We make a best effort by including each quota under low free space
1538 * conditions (less than 1% free space) in the scan.
1541 __xfs_inode_free_quota_eofblocks(
1542 struct xfs_inode *ip,
1543 int (*execute)(struct xfs_mount *mp,
1544 struct xfs_eofblocks *eofb))
1547 struct xfs_eofblocks eofb = {0};
1548 struct xfs_dquot *dq;
1551 * Run a sync scan to increase effectiveness and use the union filter to
1552 * cover all applicable quotas in a single scan.
1554 eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC;
1556 if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) {
1557 dq = xfs_inode_dquot(ip, XFS_DQ_USER);
1558 if (dq && xfs_dquot_lowsp(dq)) {
1559 eofb.eof_uid = VFS_I(ip)->i_uid;
1560 eofb.eof_flags |= XFS_EOF_FLAGS_UID;
1565 if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) {
1566 dq = xfs_inode_dquot(ip, XFS_DQ_GROUP);
1567 if (dq && xfs_dquot_lowsp(dq)) {
1568 eofb.eof_gid = VFS_I(ip)->i_gid;
1569 eofb.eof_flags |= XFS_EOF_FLAGS_GID;
1575 execute(ip->i_mount, &eofb);
1581 xfs_inode_free_quota_eofblocks(
1582 struct xfs_inode *ip)
1584 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks);
1587 static inline unsigned long
1592 case XFS_ICI_EOFBLOCKS_TAG:
1593 return XFS_IEOFBLOCKS;
1594 case XFS_ICI_COWBLOCKS_TAG:
1595 return XFS_ICOWBLOCKS;
1603 __xfs_inode_set_blocks_tag(
1605 void (*execute)(struct xfs_mount *mp),
1606 void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1607 int error, unsigned long caller_ip),
1610 struct xfs_mount *mp = ip->i_mount;
1611 struct xfs_perag *pag;
1615 * Don't bother locking the AG and looking up in the radix trees
1616 * if we already know that we have the tag set.
1618 if (ip->i_flags & xfs_iflag_for_tag(tag))
1620 spin_lock(&ip->i_flags_lock);
1621 ip->i_flags |= xfs_iflag_for_tag(tag);
1622 spin_unlock(&ip->i_flags_lock);
1624 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1625 spin_lock(&pag->pag_ici_lock);
1627 tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
1628 radix_tree_tag_set(&pag->pag_ici_root,
1629 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1631 /* propagate the eofblocks tag up into the perag radix tree */
1632 spin_lock(&ip->i_mount->m_perag_lock);
1633 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1634 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1636 spin_unlock(&ip->i_mount->m_perag_lock);
1638 /* kick off background trimming */
1639 execute(ip->i_mount);
1641 set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
1644 spin_unlock(&pag->pag_ici_lock);
1649 xfs_inode_set_eofblocks_tag(
1652 trace_xfs_inode_set_eofblocks_tag(ip);
1653 return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks,
1654 trace_xfs_perag_set_eofblocks,
1655 XFS_ICI_EOFBLOCKS_TAG);
1659 __xfs_inode_clear_blocks_tag(
1661 void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1662 int error, unsigned long caller_ip),
1665 struct xfs_mount *mp = ip->i_mount;
1666 struct xfs_perag *pag;
1668 spin_lock(&ip->i_flags_lock);
1669 ip->i_flags &= ~xfs_iflag_for_tag(tag);
1670 spin_unlock(&ip->i_flags_lock);
1672 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1673 spin_lock(&pag->pag_ici_lock);
1675 radix_tree_tag_clear(&pag->pag_ici_root,
1676 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1677 if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
1678 /* clear the eofblocks tag from the perag radix tree */
1679 spin_lock(&ip->i_mount->m_perag_lock);
1680 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1681 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1683 spin_unlock(&ip->i_mount->m_perag_lock);
1684 clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
1687 spin_unlock(&pag->pag_ici_lock);
1692 xfs_inode_clear_eofblocks_tag(
1695 trace_xfs_inode_clear_eofblocks_tag(ip);
1696 return __xfs_inode_clear_blocks_tag(ip,
1697 trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
1701 * Set ourselves up to free CoW blocks from this file. If it's already clean
1702 * then we can bail out quickly, but otherwise we must back off if the file
1703 * is undergoing some kind of write.
1706 xfs_prep_free_cowblocks(
1707 struct xfs_inode *ip)
1710 * Just clear the tag if we have an empty cow fork or none at all. It's
1711 * possible the inode was fully unshared since it was originally tagged.
1713 if (!xfs_inode_has_cow_data(ip)) {
1714 trace_xfs_inode_free_cowblocks_invalid(ip);
1715 xfs_inode_clear_cowblocks_tag(ip);
1720 * If the mapping is dirty or under writeback we cannot touch the
1721 * CoW fork. Leave it alone if we're in the midst of a directio.
1723 if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1724 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1725 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1726 atomic_read(&VFS_I(ip)->i_dio_count))
1733 * Automatic CoW Reservation Freeing
1735 * These functions automatically garbage collect leftover CoW reservations
1736 * that were made on behalf of a cowextsize hint when we start to run out
1737 * of quota or when the reservations sit around for too long. If the file
1738 * has dirty pages or is undergoing writeback, its CoW reservations will
1741 * The actual garbage collection piggybacks off the same code that runs
1742 * the speculative EOF preallocation garbage collector.
1745 xfs_inode_free_cowblocks(
1746 struct xfs_inode *ip,
1750 struct xfs_eofblocks *eofb = args;
1754 if (!xfs_prep_free_cowblocks(ip))
1758 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1759 match = xfs_inode_match_id_union(ip, eofb);
1761 match = xfs_inode_match_id(ip, eofb);
1765 /* skip the inode if the file size is too small */
1766 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1767 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1771 /* Free the CoW blocks */
1772 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1773 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1776 * Check again, nobody else should be able to dirty blocks or change
1777 * the reflink iflag now that we have the first two locks held.
1779 if (xfs_prep_free_cowblocks(ip))
1780 ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1782 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1783 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1789 xfs_icache_free_cowblocks(
1790 struct xfs_mount *mp,
1791 struct xfs_eofblocks *eofb)
1793 return __xfs_icache_free_eofblocks(mp, eofb, xfs_inode_free_cowblocks,
1794 XFS_ICI_COWBLOCKS_TAG);
1798 xfs_inode_free_quota_cowblocks(
1799 struct xfs_inode *ip)
1801 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks);
1805 xfs_inode_set_cowblocks_tag(
1808 trace_xfs_inode_set_cowblocks_tag(ip);
1809 return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks,
1810 trace_xfs_perag_set_cowblocks,
1811 XFS_ICI_COWBLOCKS_TAG);
1815 xfs_inode_clear_cowblocks_tag(
1818 trace_xfs_inode_clear_cowblocks_tag(ip);
1819 return __xfs_inode_clear_blocks_tag(ip,
1820 trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG);
1823 /* Disable post-EOF and CoW block auto-reclamation. */
1825 xfs_stop_block_reaping(
1826 struct xfs_mount *mp)
1828 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1829 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1832 /* Enable post-EOF and CoW block auto-reclamation. */
1834 xfs_start_block_reaping(
1835 struct xfs_mount *mp)
1837 xfs_queue_eofblocks(mp);
1838 xfs_queue_cowblocks(mp);