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"
25 #include "xfs_ialloc.h"
27 #include <linux/iversion.h>
30 * Allocate and initialise an xfs_inode.
40 * if this didn't occur in transactions, we could use
41 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
42 * code up to do this anyway.
44 ip = kmem_zone_alloc(xfs_inode_zone, 0);
47 if (inode_init_always(mp->m_super, VFS_I(ip))) {
48 kmem_cache_free(xfs_inode_zone, ip);
52 /* VFS doesn't initialise i_mode! */
53 VFS_I(ip)->i_mode = 0;
55 XFS_STATS_INC(mp, vn_active);
56 ASSERT(atomic_read(&ip->i_pincount) == 0);
57 ASSERT(!xfs_isiflocked(ip));
58 ASSERT(ip->i_ino == 0);
60 /* initialise the xfs inode */
63 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
66 memset(&ip->i_df, 0, sizeof(ip->i_df));
68 ip->i_delayed_blks = 0;
69 memset(&ip->i_d, 0, sizeof(ip->i_d));
72 INIT_WORK(&ip->i_ioend_work, xfs_end_io);
73 INIT_LIST_HEAD(&ip->i_ioend_list);
74 spin_lock_init(&ip->i_ioend_lock);
80 xfs_inode_free_callback(
81 struct rcu_head *head)
83 struct inode *inode = container_of(head, struct inode, i_rcu);
84 struct xfs_inode *ip = XFS_I(inode);
86 switch (VFS_I(ip)->i_mode & S_IFMT) {
90 xfs_idestroy_fork(&ip->i_df);
95 xfs_idestroy_fork(ip->i_afp);
96 kmem_cache_free(xfs_ifork_zone, ip->i_afp);
99 xfs_idestroy_fork(ip->i_cowfp);
100 kmem_cache_free(xfs_ifork_zone, ip->i_cowfp);
103 ASSERT(!test_bit(XFS_LI_IN_AIL,
104 &ip->i_itemp->ili_item.li_flags));
105 xfs_inode_item_destroy(ip);
109 kmem_cache_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;
294 kuid_t uid = inode->i_uid;
295 kgid_t gid = inode->i_gid;
297 error = inode_init_always(mp->m_super, inode);
299 set_nlink(inode, nlink);
300 inode->i_generation = generation;
301 inode_set_iversion_queried(inode, version);
302 inode->i_mode = mode;
310 * If we are allocating a new inode, then check what was returned is
311 * actually a free, empty inode. If we are not allocating an inode,
312 * then check we didn't find a free inode.
315 * 0 if the inode free state matches the lookup context
316 * -ENOENT if the inode is free and we are not allocating
317 * -EFSCORRUPTED if there is any state mismatch at all
320 xfs_iget_check_free_state(
321 struct xfs_inode *ip,
324 if (flags & XFS_IGET_CREATE) {
325 /* should be a free inode */
326 if (VFS_I(ip)->i_mode != 0) {
327 xfs_warn(ip->i_mount,
328 "Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
329 ip->i_ino, VFS_I(ip)->i_mode);
330 return -EFSCORRUPTED;
333 if (ip->i_d.di_nblocks != 0) {
334 xfs_warn(ip->i_mount,
335 "Corruption detected! Free inode 0x%llx has blocks allocated!",
337 return -EFSCORRUPTED;
342 /* should be an allocated inode */
343 if (VFS_I(ip)->i_mode == 0)
350 * Check the validity of the inode we just found it the cache
354 struct xfs_perag *pag,
355 struct xfs_inode *ip,
358 int lock_flags) __releases(RCU)
360 struct inode *inode = VFS_I(ip);
361 struct xfs_mount *mp = ip->i_mount;
365 * check for re-use of an inode within an RCU grace period due to the
366 * radix tree nodes not being updated yet. We monitor for this by
367 * setting the inode number to zero before freeing the inode structure.
368 * If the inode has been reallocated and set up, then the inode number
369 * will not match, so check for that, too.
371 spin_lock(&ip->i_flags_lock);
372 if (ip->i_ino != ino) {
373 trace_xfs_iget_skip(ip);
374 XFS_STATS_INC(mp, xs_ig_frecycle);
381 * If we are racing with another cache hit that is currently
382 * instantiating this inode or currently recycling it out of
383 * reclaimabe state, wait for the initialisation to complete
386 * XXX(hch): eventually we should do something equivalent to
387 * wait_on_inode to wait for these flags to be cleared
388 * instead of polling for it.
390 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
391 trace_xfs_iget_skip(ip);
392 XFS_STATS_INC(mp, xs_ig_frecycle);
398 * Check the inode free state is valid. This also detects lookup
399 * racing with unlinks.
401 error = xfs_iget_check_free_state(ip, flags);
406 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
407 * Need to carefully get it back into useable state.
409 if (ip->i_flags & XFS_IRECLAIMABLE) {
410 trace_xfs_iget_reclaim(ip);
412 if (flags & XFS_IGET_INCORE) {
418 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
419 * from stomping over us while we recycle the inode. We can't
420 * clear the radix tree reclaimable tag yet as it requires
421 * pag_ici_lock to be held exclusive.
423 ip->i_flags |= XFS_IRECLAIM;
425 spin_unlock(&ip->i_flags_lock);
428 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
429 error = xfs_reinit_inode(mp, inode);
433 * Re-initializing the inode failed, and we are in deep
434 * trouble. Try to re-add it to the reclaim list.
437 spin_lock(&ip->i_flags_lock);
438 wake = !!__xfs_iflags_test(ip, XFS_INEW);
439 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
441 wake_up_bit(&ip->i_flags, __XFS_INEW_BIT);
442 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
443 trace_xfs_iget_reclaim_fail(ip);
447 spin_lock(&pag->pag_ici_lock);
448 spin_lock(&ip->i_flags_lock);
451 * Clear the per-lifetime state in the inode as we are now
452 * effectively a new inode and need to return to the initial
453 * state before reuse occurs.
455 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
456 ip->i_flags |= XFS_INEW;
457 xfs_inode_clear_reclaim_tag(pag, ip->i_ino);
458 inode->i_state = I_NEW;
462 spin_unlock(&ip->i_flags_lock);
463 spin_unlock(&pag->pag_ici_lock);
465 /* If the VFS inode is being torn down, pause and try again. */
467 trace_xfs_iget_skip(ip);
472 /* We've got a live one. */
473 spin_unlock(&ip->i_flags_lock);
475 trace_xfs_iget_hit(ip);
479 xfs_ilock(ip, lock_flags);
481 if (!(flags & XFS_IGET_INCORE))
482 xfs_iflags_clear(ip, XFS_ISTALE);
483 XFS_STATS_INC(mp, xs_ig_found);
488 spin_unlock(&ip->i_flags_lock);
496 struct xfs_mount *mp,
497 struct xfs_perag *pag,
500 struct xfs_inode **ipp,
504 struct xfs_inode *ip;
506 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
509 ip = xfs_inode_alloc(mp, ino);
513 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, flags);
518 * For version 5 superblocks, if we are initialising a new inode and we
519 * are not utilising the XFS_MOUNT_IKEEP inode cluster mode, we can
520 * simply build the new inode core with a random generation number.
522 * For version 4 (and older) superblocks, log recovery is dependent on
523 * the di_flushiter field being initialised from the current on-disk
524 * value and hence we must also read the inode off disk even when
525 * initializing new inodes.
527 if (xfs_sb_version_has_v3inode(&mp->m_sb) &&
528 (flags & XFS_IGET_CREATE) && !(mp->m_flags & XFS_MOUNT_IKEEP)) {
529 VFS_I(ip)->i_generation = prandom_u32();
531 struct xfs_dinode *dip;
534 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0);
538 error = xfs_inode_from_disk(ip, dip);
540 xfs_buf_set_ref(bp, XFS_INO_REF);
541 xfs_trans_brelse(tp, bp);
547 trace_xfs_iget_miss(ip);
550 * Check the inode free state is valid. This also detects lookup
551 * racing with unlinks.
553 error = xfs_iget_check_free_state(ip, flags);
558 * Preload the radix tree so we can insert safely under the
559 * write spinlock. Note that we cannot sleep inside the preload
560 * region. Since we can be called from transaction context, don't
561 * recurse into the file system.
563 if (radix_tree_preload(GFP_NOFS)) {
569 * Because the inode hasn't been added to the radix-tree yet it can't
570 * be found by another thread, so we can do the non-sleeping lock here.
573 if (!xfs_ilock_nowait(ip, lock_flags))
578 * These values must be set before inserting the inode into the radix
579 * tree as the moment it is inserted a concurrent lookup (allowed by the
580 * RCU locking mechanism) can find it and that lookup must see that this
581 * is an inode currently under construction (i.e. that XFS_INEW is set).
582 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
583 * memory barrier that ensures this detection works correctly at lookup
587 if (flags & XFS_IGET_DONTCACHE)
588 d_mark_dontcache(VFS_I(ip));
592 xfs_iflags_set(ip, iflags);
594 /* insert the new inode */
595 spin_lock(&pag->pag_ici_lock);
596 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
597 if (unlikely(error)) {
598 WARN_ON(error != -EEXIST);
599 XFS_STATS_INC(mp, xs_ig_dup);
601 goto out_preload_end;
603 spin_unlock(&pag->pag_ici_lock);
604 radix_tree_preload_end();
610 spin_unlock(&pag->pag_ici_lock);
611 radix_tree_preload_end();
613 xfs_iunlock(ip, lock_flags);
615 __destroy_inode(VFS_I(ip));
621 * Look up an inode by number in the given file system.
622 * The inode is looked up in the cache held in each AG.
623 * If the inode is found in the cache, initialise the vfs inode
626 * If it is not in core, read it in from the file system's device,
627 * add it to the cache and initialise the vfs inode.
629 * The inode is locked according to the value of the lock_flags parameter.
630 * This flag parameter indicates how and if the inode's IO lock and inode lock
633 * mp -- the mount point structure for the current file system. It points
634 * to the inode hash table.
635 * tp -- a pointer to the current transaction if there is one. This is
636 * simply passed through to the xfs_iread() call.
637 * ino -- the number of the inode desired. This is the unique identifier
638 * within the file system for the inode being requested.
639 * lock_flags -- flags indicating how to lock the inode. See the comment
640 * for xfs_ilock() for a list of valid values.
657 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
658 * doesn't get freed while it's being referenced during a
659 * radix tree traversal here. It assumes this function
660 * aqcuires only the ILOCK (and therefore it has no need to
661 * involve the IOLOCK in this synchronization).
663 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
665 /* reject inode numbers outside existing AGs */
666 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
669 XFS_STATS_INC(mp, xs_ig_attempts);
671 /* get the perag structure and ensure that it's inode capable */
672 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
673 agino = XFS_INO_TO_AGINO(mp, ino);
678 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
681 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
683 goto out_error_or_again;
686 if (flags & XFS_IGET_INCORE) {
688 goto out_error_or_again;
690 XFS_STATS_INC(mp, xs_ig_missed);
692 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
695 goto out_error_or_again;
702 * If we have a real type for an on-disk inode, we can setup the inode
703 * now. If it's a new inode being created, xfs_ialloc will handle it.
705 if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
706 xfs_setup_existing_inode(ip);
710 if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) {
719 * "Is this a cached inode that's also allocated?"
721 * Look up an inode by number in the given file system. If the inode is
722 * in cache and isn't in purgatory, return 1 if the inode is allocated
723 * and 0 if it is not. For all other cases (not in cache, being torn
724 * down, etc.), return a negative error code.
726 * The caller has to prevent inode allocation and freeing activity,
727 * presumably by locking the AGI buffer. This is to ensure that an
728 * inode cannot transition from allocated to freed until the caller is
729 * ready to allow that. If the inode is in an intermediate state (new,
730 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
731 * inode is not in the cache, -ENOENT will be returned. The caller must
732 * deal with these scenarios appropriately.
734 * This is a specialized use case for the online scrubber; if you're
735 * reading this, you probably want xfs_iget.
738 xfs_icache_inode_is_allocated(
739 struct xfs_mount *mp,
740 struct xfs_trans *tp,
744 struct xfs_inode *ip;
747 error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip);
751 *inuse = !!(VFS_I(ip)->i_mode);
757 * The inode lookup is done in batches to keep the amount of lock traffic and
758 * radix tree lookups to a minimum. The batch size is a trade off between
759 * lookup reduction and stack usage. This is in the reclaim path, so we can't
762 #define XFS_LOOKUP_BATCH 32
765 * Decide if the given @ip is eligible to be a part of the inode walk, and
766 * grab it if so. Returns true if it's ready to go or false if we should just
770 xfs_inode_walk_ag_grab(
771 struct xfs_inode *ip,
774 struct inode *inode = VFS_I(ip);
775 bool newinos = !!(flags & XFS_INODE_WALK_INEW_WAIT);
777 ASSERT(rcu_read_lock_held());
780 * check for stale RCU freed inode
782 * If the inode has been reallocated, it doesn't matter if it's not in
783 * the AG we are walking - we are walking for writeback, so if it
784 * passes all the "valid inode" checks and is dirty, then we'll write
785 * it back anyway. If it has been reallocated and still being
786 * initialised, the XFS_INEW check below will catch it.
788 spin_lock(&ip->i_flags_lock);
790 goto out_unlock_noent;
792 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
793 if ((!newinos && __xfs_iflags_test(ip, XFS_INEW)) ||
794 __xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM))
795 goto out_unlock_noent;
796 spin_unlock(&ip->i_flags_lock);
798 /* nothing to sync during shutdown */
799 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
802 /* If we can't grab the inode, it must on it's way to reclaim. */
810 spin_unlock(&ip->i_flags_lock);
815 * For a given per-AG structure @pag, grab, @execute, and rele all incore
816 * inodes with the given radix tree @tag.
820 struct xfs_perag *pag,
822 int (*execute)(struct xfs_inode *ip, void *args),
826 struct xfs_mount *mp = pag->pag_mount;
827 uint32_t first_index;
839 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
845 if (tag == XFS_ICI_NO_TAG)
846 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
847 (void **)batch, first_index,
850 nr_found = radix_tree_gang_lookup_tag(
852 (void **) batch, first_index,
853 XFS_LOOKUP_BATCH, tag);
861 * Grab the inodes before we drop the lock. if we found
862 * nothing, nr == 0 and the loop will be skipped.
864 for (i = 0; i < nr_found; i++) {
865 struct xfs_inode *ip = batch[i];
867 if (done || !xfs_inode_walk_ag_grab(ip, iter_flags))
871 * Update the index for the next lookup. Catch
872 * overflows into the next AG range which can occur if
873 * we have inodes in the last block of the AG and we
874 * are currently pointing to the last inode.
876 * Because we may see inodes that are from the wrong AG
877 * due to RCU freeing and reallocation, only update the
878 * index if it lies in this AG. It was a race that lead
879 * us to see this inode, so another lookup from the
880 * same index will not find it again.
882 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
884 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
885 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
889 /* unlock now we've grabbed the inodes. */
892 for (i = 0; i < nr_found; i++) {
895 if ((iter_flags & XFS_INODE_WALK_INEW_WAIT) &&
896 xfs_iflags_test(batch[i], XFS_INEW))
897 xfs_inew_wait(batch[i]);
898 error = execute(batch[i], args);
900 if (error == -EAGAIN) {
904 if (error && last_error != -EFSCORRUPTED)
908 /* bail out if the filesystem is corrupted. */
909 if (error == -EFSCORRUPTED)
914 } while (nr_found && !done);
923 /* Fetch the next (possibly tagged) per-AG structure. */
924 static inline struct xfs_perag *
925 xfs_inode_walk_get_perag(
926 struct xfs_mount *mp,
930 if (tag == XFS_ICI_NO_TAG)
931 return xfs_perag_get(mp, agno);
932 return xfs_perag_get_tag(mp, agno, tag);
936 * Call the @execute function on all incore inodes matching the radix tree
941 struct xfs_mount *mp,
943 int (*execute)(struct xfs_inode *ip, void *args),
947 struct xfs_perag *pag;
953 while ((pag = xfs_inode_walk_get_perag(mp, ag, tag))) {
954 ag = pag->pag_agno + 1;
955 error = xfs_inode_walk_ag(pag, iter_flags, execute, args, tag);
959 if (error == -EFSCORRUPTED)
967 * Background scanning to trim post-EOF preallocated space. This is queued
968 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
972 struct xfs_mount *mp)
975 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
976 queue_delayed_work(mp->m_eofblocks_workqueue,
977 &mp->m_eofblocks_work,
978 msecs_to_jiffies(xfs_eofb_secs * 1000));
983 xfs_eofblocks_worker(
984 struct work_struct *work)
986 struct xfs_mount *mp = container_of(to_delayed_work(work),
987 struct xfs_mount, m_eofblocks_work);
989 if (!sb_start_write_trylock(mp->m_super))
991 xfs_icache_free_eofblocks(mp, NULL);
992 sb_end_write(mp->m_super);
994 xfs_queue_eofblocks(mp);
998 * Background scanning to trim preallocated CoW space. This is queued
999 * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
1000 * (We'll just piggyback on the post-EOF prealloc space workqueue.)
1003 xfs_queue_cowblocks(
1004 struct xfs_mount *mp)
1007 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
1008 queue_delayed_work(mp->m_eofblocks_workqueue,
1009 &mp->m_cowblocks_work,
1010 msecs_to_jiffies(xfs_cowb_secs * 1000));
1015 xfs_cowblocks_worker(
1016 struct work_struct *work)
1018 struct xfs_mount *mp = container_of(to_delayed_work(work),
1019 struct xfs_mount, m_cowblocks_work);
1021 if (!sb_start_write_trylock(mp->m_super))
1023 xfs_icache_free_cowblocks(mp, NULL);
1024 sb_end_write(mp->m_super);
1026 xfs_queue_cowblocks(mp);
1030 * Grab the inode for reclaim exclusively.
1031 * Return 0 if we grabbed it, non-zero otherwise.
1034 xfs_reclaim_inode_grab(
1035 struct xfs_inode *ip,
1038 ASSERT(rcu_read_lock_held());
1040 /* quick check for stale RCU freed inode */
1045 * If we are asked for non-blocking operation, do unlocked checks to
1046 * see if the inode already is being flushed or in reclaim to avoid
1049 if ((flags & SYNC_TRYLOCK) &&
1050 __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
1054 * The radix tree lock here protects a thread in xfs_iget from racing
1055 * with us starting reclaim on the inode. Once we have the
1056 * XFS_IRECLAIM flag set it will not touch us.
1058 * Due to RCU lookup, we may find inodes that have been freed and only
1059 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
1060 * aren't candidates for reclaim at all, so we must check the
1061 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
1063 spin_lock(&ip->i_flags_lock);
1064 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
1065 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
1066 /* not a reclaim candidate. */
1067 spin_unlock(&ip->i_flags_lock);
1070 __xfs_iflags_set(ip, XFS_IRECLAIM);
1071 spin_unlock(&ip->i_flags_lock);
1076 * Inodes in different states need to be treated differently. The following
1077 * table lists the inode states and the reclaim actions necessary:
1079 * inode state iflush ret required action
1080 * --------------- ---------- ---------------
1082 * shutdown EIO unpin and reclaim
1083 * clean, unpinned 0 reclaim
1084 * stale, unpinned 0 reclaim
1085 * clean, pinned(*) 0 requeue
1086 * stale, pinned EAGAIN requeue
1087 * dirty, async - requeue
1088 * dirty, sync 0 reclaim
1090 * (*) dgc: I don't think the clean, pinned state is possible but it gets
1091 * handled anyway given the order of checks implemented.
1093 * Also, because we get the flush lock first, we know that any inode that has
1094 * been flushed delwri has had the flush completed by the time we check that
1095 * the inode is clean.
1097 * Note that because the inode is flushed delayed write by AIL pushing, the
1098 * flush lock may already be held here and waiting on it can result in very
1099 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
1100 * the caller should push the AIL first before trying to reclaim inodes to
1101 * minimise the amount of time spent waiting. For background relaim, we only
1102 * bother to reclaim clean inodes anyway.
1104 * Hence the order of actions after gaining the locks should be:
1106 * shutdown => unpin and reclaim
1107 * pinned, async => requeue
1108 * pinned, sync => unpin
1111 * dirty, async => requeue
1112 * dirty, sync => flush, wait and reclaim
1116 struct xfs_inode *ip,
1117 struct xfs_perag *pag,
1120 struct xfs_buf *bp = NULL;
1121 xfs_ino_t ino = ip->i_ino; /* for radix_tree_delete */
1126 xfs_ilock(ip, XFS_ILOCK_EXCL);
1127 if (!xfs_iflock_nowait(ip)) {
1128 if (!(sync_mode & SYNC_WAIT))
1133 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1134 xfs_iunpin_wait(ip);
1135 /* xfs_iflush_abort() drops the flush lock */
1136 xfs_iflush_abort(ip);
1139 if (xfs_ipincount(ip)) {
1140 if (!(sync_mode & SYNC_WAIT))
1142 xfs_iunpin_wait(ip);
1144 if (xfs_iflags_test(ip, XFS_ISTALE) || xfs_inode_clean(ip)) {
1150 * Never flush out dirty data during non-blocking reclaim, as it would
1151 * just contend with AIL pushing trying to do the same job.
1153 if (!(sync_mode & SYNC_WAIT))
1157 * Now we have an inode that needs flushing.
1159 * Note that xfs_iflush will never block on the inode buffer lock, as
1160 * xfs_ifree_cluster() can lock the inode buffer before it locks the
1161 * ip->i_lock, and we are doing the exact opposite here. As a result,
1162 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
1163 * result in an ABBA deadlock with xfs_ifree_cluster().
1165 * As xfs_ifree_cluser() must gather all inodes that are active in the
1166 * cache to mark them stale, if we hit this case we don't actually want
1167 * to do IO here - we want the inode marked stale so we can simply
1168 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
1169 * inode, back off and try again. Hopefully the next pass through will
1170 * see the stale flag set on the inode.
1172 error = xfs_iflush(ip, &bp);
1173 if (error == -EAGAIN) {
1174 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1175 /* backoff longer than in xfs_ifree_cluster */
1181 error = xfs_bwrite(bp);
1186 ASSERT(!xfs_isiflocked(ip));
1189 * Because we use RCU freeing we need to ensure the inode always appears
1190 * to be reclaimed with an invalid inode number when in the free state.
1191 * We do this as early as possible under the ILOCK so that
1192 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
1193 * detect races with us here. By doing this, we guarantee that once
1194 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
1195 * it will see either a valid inode that will serialise correctly, or it
1196 * will see an invalid inode that it can skip.
1198 spin_lock(&ip->i_flags_lock);
1199 ip->i_flags = XFS_IRECLAIM;
1201 spin_unlock(&ip->i_flags_lock);
1203 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1205 XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
1207 * Remove the inode from the per-AG radix tree.
1209 * Because radix_tree_delete won't complain even if the item was never
1210 * added to the tree assert that it's been there before to catch
1211 * problems with the inode life time early on.
1213 spin_lock(&pag->pag_ici_lock);
1214 if (!radix_tree_delete(&pag->pag_ici_root,
1215 XFS_INO_TO_AGINO(ip->i_mount, ino)))
1217 xfs_perag_clear_reclaim_tag(pag);
1218 spin_unlock(&pag->pag_ici_lock);
1221 * Here we do an (almost) spurious inode lock in order to coordinate
1222 * with inode cache radix tree lookups. This is because the lookup
1223 * can reference the inodes in the cache without taking references.
1225 * We make that OK here by ensuring that we wait until the inode is
1226 * unlocked after the lookup before we go ahead and free it.
1228 xfs_ilock(ip, XFS_ILOCK_EXCL);
1229 xfs_qm_dqdetach(ip);
1230 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1232 __xfs_inode_free(ip);
1238 xfs_iflags_clear(ip, XFS_IRECLAIM);
1239 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1241 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1242 * a short while. However, this just burns CPU time scanning the tree
1243 * waiting for IO to complete and the reclaim work never goes back to
1244 * the idle state. Instead, return 0 to let the next scheduled
1245 * background reclaim attempt to reclaim the inode again.
1251 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1252 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1253 * then a shut down during filesystem unmount reclaim walk leak all the
1254 * unreclaimed inodes.
1257 xfs_reclaim_inodes_ag(
1258 struct xfs_mount *mp,
1262 struct xfs_perag *pag;
1266 int trylock = flags & SYNC_TRYLOCK;
1272 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1273 unsigned long first_index = 0;
1277 ag = pag->pag_agno + 1;
1280 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1285 first_index = pag->pag_ici_reclaim_cursor;
1287 mutex_lock(&pag->pag_ici_reclaim_lock);
1290 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1294 nr_found = radix_tree_gang_lookup_tag(
1296 (void **)batch, first_index,
1298 XFS_ICI_RECLAIM_TAG);
1306 * Grab the inodes before we drop the lock. if we found
1307 * nothing, nr == 0 and the loop will be skipped.
1309 for (i = 0; i < nr_found; i++) {
1310 struct xfs_inode *ip = batch[i];
1312 if (done || xfs_reclaim_inode_grab(ip, flags))
1316 * Update the index for the next lookup. Catch
1317 * overflows into the next AG range which can
1318 * occur if we have inodes in the last block of
1319 * the AG and we are currently pointing to the
1322 * Because we may see inodes that are from the
1323 * wrong AG due to RCU freeing and
1324 * reallocation, only update the index if it
1325 * lies in this AG. It was a race that lead us
1326 * to see this inode, so another lookup from
1327 * the same index will not find it again.
1329 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1332 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1333 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1337 /* unlock now we've grabbed the inodes. */
1340 for (i = 0; i < nr_found; i++) {
1343 error = xfs_reclaim_inode(batch[i], pag, flags);
1344 if (error && last_error != -EFSCORRUPTED)
1348 *nr_to_scan -= XFS_LOOKUP_BATCH;
1352 } while (nr_found && !done && *nr_to_scan > 0);
1354 if (trylock && !done)
1355 pag->pag_ici_reclaim_cursor = first_index;
1357 pag->pag_ici_reclaim_cursor = 0;
1358 mutex_unlock(&pag->pag_ici_reclaim_lock);
1363 * if we skipped any AG, and we still have scan count remaining, do
1364 * another pass this time using blocking reclaim semantics (i.e
1365 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1366 * ensure that when we get more reclaimers than AGs we block rather
1367 * than spin trying to execute reclaim.
1369 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1381 int nr_to_scan = INT_MAX;
1383 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1387 * Scan a certain number of inodes for reclaim.
1389 * When called we make sure that there is a background (fast) inode reclaim in
1390 * progress, while we will throttle the speed of reclaim via doing synchronous
1391 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1392 * them to be cleaned, which we hope will not be very long due to the
1393 * background walker having already kicked the IO off on those dirty inodes.
1396 xfs_reclaim_inodes_nr(
1397 struct xfs_mount *mp,
1400 /* kick background reclaimer and push the AIL */
1401 xfs_reclaim_work_queue(mp);
1402 xfs_ail_push_all(mp->m_ail);
1404 return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1408 * Return the number of reclaimable inodes in the filesystem for
1409 * the shrinker to determine how much to reclaim.
1412 xfs_reclaim_inodes_count(
1413 struct xfs_mount *mp)
1415 struct xfs_perag *pag;
1416 xfs_agnumber_t ag = 0;
1417 int reclaimable = 0;
1419 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1420 ag = pag->pag_agno + 1;
1421 reclaimable += pag->pag_ici_reclaimable;
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 ip->i_d.di_projid != eofb->eof_prid)
1448 * A union-based inode filtering algorithm. Process the inode if any of the
1449 * criteria match. This is for global/internal scans only.
1452 xfs_inode_match_id_union(
1453 struct xfs_inode *ip,
1454 struct xfs_eofblocks *eofb)
1456 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1457 uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1460 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1461 gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1464 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1465 ip->i_d.di_projid == eofb->eof_prid)
1472 * Is this inode @ip eligible for eof/cow block reclamation, given some
1473 * filtering parameters @eofb? The inode is eligible if @eofb is null or
1474 * if the predicate functions match.
1477 xfs_inode_matches_eofb(
1478 struct xfs_inode *ip,
1479 struct xfs_eofblocks *eofb)
1486 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1487 match = xfs_inode_match_id_union(ip, eofb);
1489 match = xfs_inode_match_id(ip, eofb);
1493 /* skip the inode if the file size is too small */
1494 if ((eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE) &&
1495 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1502 xfs_inode_free_eofblocks(
1503 struct xfs_inode *ip,
1506 struct xfs_eofblocks *eofb = args;
1510 wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC);
1512 if (!xfs_can_free_eofblocks(ip, false)) {
1513 /* inode could be preallocated or append-only */
1514 trace_xfs_inode_free_eofblocks_invalid(ip);
1515 xfs_inode_clear_eofblocks_tag(ip);
1520 * If the mapping is dirty the operation can block and wait for some
1521 * time. Unless we are waiting, skip it.
1523 if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1526 if (!xfs_inode_matches_eofb(ip, eofb))
1530 * If the caller is waiting, return -EAGAIN to keep the background
1531 * scanner moving and revisit the inode in a subsequent pass.
1533 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1539 ret = xfs_free_eofblocks(ip);
1540 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1546 xfs_icache_free_eofblocks(
1547 struct xfs_mount *mp,
1548 struct xfs_eofblocks *eofb)
1550 return xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, eofb,
1551 XFS_ICI_EOFBLOCKS_TAG);
1555 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1556 * multiple quotas, we don't know exactly which quota caused an allocation
1557 * failure. We make a best effort by including each quota under low free space
1558 * conditions (less than 1% free space) in the scan.
1561 __xfs_inode_free_quota_eofblocks(
1562 struct xfs_inode *ip,
1563 int (*execute)(struct xfs_mount *mp,
1564 struct xfs_eofblocks *eofb))
1567 struct xfs_eofblocks eofb = {0};
1568 struct xfs_dquot *dq;
1571 * Run a sync scan to increase effectiveness and use the union filter to
1572 * cover all applicable quotas in a single scan.
1574 eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC;
1576 if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) {
1577 dq = xfs_inode_dquot(ip, XFS_DQ_USER);
1578 if (dq && xfs_dquot_lowsp(dq)) {
1579 eofb.eof_uid = VFS_I(ip)->i_uid;
1580 eofb.eof_flags |= XFS_EOF_FLAGS_UID;
1585 if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) {
1586 dq = xfs_inode_dquot(ip, XFS_DQ_GROUP);
1587 if (dq && xfs_dquot_lowsp(dq)) {
1588 eofb.eof_gid = VFS_I(ip)->i_gid;
1589 eofb.eof_flags |= XFS_EOF_FLAGS_GID;
1595 execute(ip->i_mount, &eofb);
1601 xfs_inode_free_quota_eofblocks(
1602 struct xfs_inode *ip)
1604 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_eofblocks);
1607 static inline unsigned long
1612 case XFS_ICI_EOFBLOCKS_TAG:
1613 return XFS_IEOFBLOCKS;
1614 case XFS_ICI_COWBLOCKS_TAG:
1615 return XFS_ICOWBLOCKS;
1623 __xfs_inode_set_blocks_tag(
1625 void (*execute)(struct xfs_mount *mp),
1626 void (*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1627 int error, unsigned long caller_ip),
1630 struct xfs_mount *mp = ip->i_mount;
1631 struct xfs_perag *pag;
1635 * Don't bother locking the AG and looking up in the radix trees
1636 * if we already know that we have the tag set.
1638 if (ip->i_flags & xfs_iflag_for_tag(tag))
1640 spin_lock(&ip->i_flags_lock);
1641 ip->i_flags |= xfs_iflag_for_tag(tag);
1642 spin_unlock(&ip->i_flags_lock);
1644 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1645 spin_lock(&pag->pag_ici_lock);
1647 tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
1648 radix_tree_tag_set(&pag->pag_ici_root,
1649 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1651 /* propagate the eofblocks tag up into the perag radix tree */
1652 spin_lock(&ip->i_mount->m_perag_lock);
1653 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1654 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1656 spin_unlock(&ip->i_mount->m_perag_lock);
1658 /* kick off background trimming */
1659 execute(ip->i_mount);
1661 set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
1664 spin_unlock(&pag->pag_ici_lock);
1669 xfs_inode_set_eofblocks_tag(
1672 trace_xfs_inode_set_eofblocks_tag(ip);
1673 return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks,
1674 trace_xfs_perag_set_eofblocks,
1675 XFS_ICI_EOFBLOCKS_TAG);
1679 __xfs_inode_clear_blocks_tag(
1681 void (*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
1682 int error, unsigned long caller_ip),
1685 struct xfs_mount *mp = ip->i_mount;
1686 struct xfs_perag *pag;
1688 spin_lock(&ip->i_flags_lock);
1689 ip->i_flags &= ~xfs_iflag_for_tag(tag);
1690 spin_unlock(&ip->i_flags_lock);
1692 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1693 spin_lock(&pag->pag_ici_lock);
1695 radix_tree_tag_clear(&pag->pag_ici_root,
1696 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
1697 if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
1698 /* clear the eofblocks tag from the perag radix tree */
1699 spin_lock(&ip->i_mount->m_perag_lock);
1700 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1701 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1703 spin_unlock(&ip->i_mount->m_perag_lock);
1704 clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
1707 spin_unlock(&pag->pag_ici_lock);
1712 xfs_inode_clear_eofblocks_tag(
1715 trace_xfs_inode_clear_eofblocks_tag(ip);
1716 return __xfs_inode_clear_blocks_tag(ip,
1717 trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
1721 * Set ourselves up to free CoW blocks from this file. If it's already clean
1722 * then we can bail out quickly, but otherwise we must back off if the file
1723 * is undergoing some kind of write.
1726 xfs_prep_free_cowblocks(
1727 struct xfs_inode *ip)
1730 * Just clear the tag if we have an empty cow fork or none at all. It's
1731 * possible the inode was fully unshared since it was originally tagged.
1733 if (!xfs_inode_has_cow_data(ip)) {
1734 trace_xfs_inode_free_cowblocks_invalid(ip);
1735 xfs_inode_clear_cowblocks_tag(ip);
1740 * If the mapping is dirty or under writeback we cannot touch the
1741 * CoW fork. Leave it alone if we're in the midst of a directio.
1743 if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1744 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1745 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1746 atomic_read(&VFS_I(ip)->i_dio_count))
1753 * Automatic CoW Reservation Freeing
1755 * These functions automatically garbage collect leftover CoW reservations
1756 * that were made on behalf of a cowextsize hint when we start to run out
1757 * of quota or when the reservations sit around for too long. If the file
1758 * has dirty pages or is undergoing writeback, its CoW reservations will
1761 * The actual garbage collection piggybacks off the same code that runs
1762 * the speculative EOF preallocation garbage collector.
1765 xfs_inode_free_cowblocks(
1766 struct xfs_inode *ip,
1769 struct xfs_eofblocks *eofb = args;
1772 if (!xfs_prep_free_cowblocks(ip))
1775 if (!xfs_inode_matches_eofb(ip, eofb))
1778 /* Free the CoW blocks */
1779 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1780 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1783 * Check again, nobody else should be able to dirty blocks or change
1784 * the reflink iflag now that we have the first two locks held.
1786 if (xfs_prep_free_cowblocks(ip))
1787 ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1789 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1790 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1796 xfs_icache_free_cowblocks(
1797 struct xfs_mount *mp,
1798 struct xfs_eofblocks *eofb)
1800 return xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, eofb,
1801 XFS_ICI_COWBLOCKS_TAG);
1805 xfs_inode_free_quota_cowblocks(
1806 struct xfs_inode *ip)
1808 return __xfs_inode_free_quota_eofblocks(ip, xfs_icache_free_cowblocks);
1812 xfs_inode_set_cowblocks_tag(
1815 trace_xfs_inode_set_cowblocks_tag(ip);
1816 return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks,
1817 trace_xfs_perag_set_cowblocks,
1818 XFS_ICI_COWBLOCKS_TAG);
1822 xfs_inode_clear_cowblocks_tag(
1825 trace_xfs_inode_clear_cowblocks_tag(ip);
1826 return __xfs_inode_clear_blocks_tag(ip,
1827 trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG);
1830 /* Disable post-EOF and CoW block auto-reclamation. */
1832 xfs_stop_block_reaping(
1833 struct xfs_mount *mp)
1835 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1836 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1839 /* Enable post-EOF and CoW block auto-reclamation. */
1841 xfs_start_block_reaping(
1842 struct xfs_mount *mp)
1844 xfs_queue_eofblocks(mp);
1845 xfs_queue_cowblocks(mp);