2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_error.h"
29 #include "xfs_trans.h"
30 #include "xfs_trans_priv.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_quota.h"
33 #include "xfs_trace.h"
34 #include "xfs_icache.h"
35 #include "xfs_bmap_util.h"
36 #include "xfs_quota.h"
37 #include "xfs_dquot_item.h"
38 #include "xfs_dquot.h"
40 #include <linux/kthread.h>
41 #include <linux/freezer.h>
43 STATIC void __xfs_inode_clear_reclaim_tag(struct xfs_mount *mp,
44 struct xfs_perag *pag, struct xfs_inode *ip);
47 * Allocate and initialise an xfs_inode.
57 * if this didn't occur in transactions, we could use
58 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
59 * code up to do this anyway.
61 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
64 if (inode_init_always(mp->m_super, VFS_I(ip))) {
65 kmem_zone_free(xfs_inode_zone, ip);
69 ASSERT(atomic_read(&ip->i_pincount) == 0);
70 ASSERT(!spin_is_locked(&ip->i_flags_lock));
71 ASSERT(!xfs_isiflocked(ip));
72 ASSERT(ip->i_ino == 0);
74 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
76 /* initialise the xfs inode */
79 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
81 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
83 ip->i_delayed_blks = 0;
84 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
90 xfs_inode_free_callback(
91 struct rcu_head *head)
93 struct inode *inode = container_of(head, struct inode, i_rcu);
94 struct xfs_inode *ip = XFS_I(inode);
96 kmem_zone_free(xfs_inode_zone, ip);
101 struct xfs_inode *ip)
103 switch (ip->i_d.di_mode & S_IFMT) {
107 xfs_idestroy_fork(ip, XFS_DATA_FORK);
112 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
115 ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL));
116 xfs_inode_item_destroy(ip);
121 * Because we use RCU freeing we need to ensure the inode always
122 * appears to be reclaimed with an invalid inode number when in the
123 * free state. The ip->i_flags_lock provides the barrier against lookup
126 spin_lock(&ip->i_flags_lock);
127 ip->i_flags = XFS_IRECLAIM;
129 spin_unlock(&ip->i_flags_lock);
131 /* asserts to verify all state is correct here */
132 ASSERT(atomic_read(&ip->i_pincount) == 0);
133 ASSERT(!xfs_isiflocked(ip));
135 call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
139 * Check the validity of the inode we just found it the cache
143 struct xfs_perag *pag,
144 struct xfs_inode *ip,
147 int lock_flags) __releases(RCU)
149 struct inode *inode = VFS_I(ip);
150 struct xfs_mount *mp = ip->i_mount;
154 * check for re-use of an inode within an RCU grace period due to the
155 * radix tree nodes not being updated yet. We monitor for this by
156 * setting the inode number to zero before freeing the inode structure.
157 * If the inode has been reallocated and set up, then the inode number
158 * will not match, so check for that, too.
160 spin_lock(&ip->i_flags_lock);
161 if (ip->i_ino != ino) {
162 trace_xfs_iget_skip(ip);
163 XFS_STATS_INC(xs_ig_frecycle);
170 * If we are racing with another cache hit that is currently
171 * instantiating this inode or currently recycling it out of
172 * reclaimabe state, wait for the initialisation to complete
175 * XXX(hch): eventually we should do something equivalent to
176 * wait_on_inode to wait for these flags to be cleared
177 * instead of polling for it.
179 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
180 trace_xfs_iget_skip(ip);
181 XFS_STATS_INC(xs_ig_frecycle);
187 * If lookup is racing with unlink return an error immediately.
189 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
195 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
196 * Need to carefully get it back into useable state.
198 if (ip->i_flags & XFS_IRECLAIMABLE) {
199 trace_xfs_iget_reclaim(ip);
202 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
203 * from stomping over us while we recycle the inode. We can't
204 * clear the radix tree reclaimable tag yet as it requires
205 * pag_ici_lock to be held exclusive.
207 ip->i_flags |= XFS_IRECLAIM;
209 spin_unlock(&ip->i_flags_lock);
212 error = inode_init_always(mp->m_super, inode);
215 * Re-initializing the inode failed, and we are in deep
216 * trouble. Try to re-add it to the reclaim list.
219 spin_lock(&ip->i_flags_lock);
221 ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
222 ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
223 trace_xfs_iget_reclaim_fail(ip);
227 spin_lock(&pag->pag_ici_lock);
228 spin_lock(&ip->i_flags_lock);
231 * Clear the per-lifetime state in the inode as we are now
232 * effectively a new inode and need to return to the initial
233 * state before reuse occurs.
235 ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
236 ip->i_flags |= XFS_INEW;
237 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
238 inode->i_state = I_NEW;
240 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
241 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
243 spin_unlock(&ip->i_flags_lock);
244 spin_unlock(&pag->pag_ici_lock);
246 /* If the VFS inode is being torn down, pause and try again. */
248 trace_xfs_iget_skip(ip);
253 /* We've got a live one. */
254 spin_unlock(&ip->i_flags_lock);
256 trace_xfs_iget_hit(ip);
260 xfs_ilock(ip, lock_flags);
262 xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
263 XFS_STATS_INC(xs_ig_found);
268 spin_unlock(&ip->i_flags_lock);
276 struct xfs_mount *mp,
277 struct xfs_perag *pag,
280 struct xfs_inode **ipp,
284 struct xfs_inode *ip;
286 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
289 ip = xfs_inode_alloc(mp, ino);
293 error = xfs_iread(mp, tp, ip, flags);
297 trace_xfs_iget_miss(ip);
299 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
305 * Preload the radix tree so we can insert safely under the
306 * write spinlock. Note that we cannot sleep inside the preload
307 * region. Since we can be called from transaction context, don't
308 * recurse into the file system.
310 if (radix_tree_preload(GFP_NOFS)) {
316 * Because the inode hasn't been added to the radix-tree yet it can't
317 * be found by another thread, so we can do the non-sleeping lock here.
320 if (!xfs_ilock_nowait(ip, lock_flags))
325 * These values must be set before inserting the inode into the radix
326 * tree as the moment it is inserted a concurrent lookup (allowed by the
327 * RCU locking mechanism) can find it and that lookup must see that this
328 * is an inode currently under construction (i.e. that XFS_INEW is set).
329 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
330 * memory barrier that ensures this detection works correctly at lookup
334 if (flags & XFS_IGET_DONTCACHE)
335 iflags |= XFS_IDONTCACHE;
339 xfs_iflags_set(ip, iflags);
341 /* insert the new inode */
342 spin_lock(&pag->pag_ici_lock);
343 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
344 if (unlikely(error)) {
345 WARN_ON(error != -EEXIST);
346 XFS_STATS_INC(xs_ig_dup);
348 goto out_preload_end;
350 spin_unlock(&pag->pag_ici_lock);
351 radix_tree_preload_end();
357 spin_unlock(&pag->pag_ici_lock);
358 radix_tree_preload_end();
360 xfs_iunlock(ip, lock_flags);
362 __destroy_inode(VFS_I(ip));
368 * Look up an inode by number in the given file system.
369 * The inode is looked up in the cache held in each AG.
370 * If the inode is found in the cache, initialise the vfs inode
373 * If it is not in core, read it in from the file system's device,
374 * add it to the cache and initialise the vfs inode.
376 * The inode is locked according to the value of the lock_flags parameter.
377 * This flag parameter indicates how and if the inode's IO lock and inode lock
380 * mp -- the mount point structure for the current file system. It points
381 * to the inode hash table.
382 * tp -- a pointer to the current transaction if there is one. This is
383 * simply passed through to the xfs_iread() call.
384 * ino -- the number of the inode desired. This is the unique identifier
385 * within the file system for the inode being requested.
386 * lock_flags -- flags indicating how to lock the inode. See the comment
387 * for xfs_ilock() for a list of valid values.
404 * xfs_reclaim_inode() uses the ILOCK to ensure an inode
405 * doesn't get freed while it's being referenced during a
406 * radix tree traversal here. It assumes this function
407 * aqcuires only the ILOCK (and therefore it has no need to
408 * involve the IOLOCK in this synchronization).
410 ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
412 /* reject inode numbers outside existing AGs */
413 if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
416 /* get the perag structure and ensure that it's inode capable */
417 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
418 agino = XFS_INO_TO_AGINO(mp, ino);
423 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
426 error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
428 goto out_error_or_again;
431 XFS_STATS_INC(xs_ig_missed);
433 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
436 goto out_error_or_again;
443 * If we have a real type for an on-disk inode, we can set ops(&unlock)
444 * now. If it's a new inode being created, xfs_ialloc will handle it.
446 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
451 if (error == -EAGAIN) {
460 * The inode lookup is done in batches to keep the amount of lock traffic and
461 * radix tree lookups to a minimum. The batch size is a trade off between
462 * lookup reduction and stack usage. This is in the reclaim path, so we can't
465 #define XFS_LOOKUP_BATCH 32
468 xfs_inode_ag_walk_grab(
469 struct xfs_inode *ip)
471 struct inode *inode = VFS_I(ip);
473 ASSERT(rcu_read_lock_held());
476 * check for stale RCU freed inode
478 * If the inode has been reallocated, it doesn't matter if it's not in
479 * the AG we are walking - we are walking for writeback, so if it
480 * passes all the "valid inode" checks and is dirty, then we'll write
481 * it back anyway. If it has been reallocated and still being
482 * initialised, the XFS_INEW check below will catch it.
484 spin_lock(&ip->i_flags_lock);
486 goto out_unlock_noent;
488 /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
489 if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM))
490 goto out_unlock_noent;
491 spin_unlock(&ip->i_flags_lock);
493 /* nothing to sync during shutdown */
494 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
495 return -EFSCORRUPTED;
497 /* If we can't grab the inode, it must on it's way to reclaim. */
505 spin_unlock(&ip->i_flags_lock);
511 struct xfs_mount *mp,
512 struct xfs_perag *pag,
513 int (*execute)(struct xfs_inode *ip, int flags,
519 uint32_t first_index;
531 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
538 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
539 (void **)batch, first_index,
542 nr_found = radix_tree_gang_lookup_tag(
544 (void **) batch, first_index,
545 XFS_LOOKUP_BATCH, tag);
553 * Grab the inodes before we drop the lock. if we found
554 * nothing, nr == 0 and the loop will be skipped.
556 for (i = 0; i < nr_found; i++) {
557 struct xfs_inode *ip = batch[i];
559 if (done || xfs_inode_ag_walk_grab(ip))
563 * Update the index for the next lookup. Catch
564 * overflows into the next AG range which can occur if
565 * we have inodes in the last block of the AG and we
566 * are currently pointing to the last inode.
568 * Because we may see inodes that are from the wrong AG
569 * due to RCU freeing and reallocation, only update the
570 * index if it lies in this AG. It was a race that lead
571 * us to see this inode, so another lookup from the
572 * same index will not find it again.
574 if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
576 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
577 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
581 /* unlock now we've grabbed the inodes. */
584 for (i = 0; i < nr_found; i++) {
587 error = execute(batch[i], flags, args);
589 if (error == -EAGAIN) {
593 if (error && last_error != -EFSCORRUPTED)
597 /* bail out if the filesystem is corrupted. */
598 if (error == -EFSCORRUPTED)
603 } while (nr_found && !done);
613 * Background scanning to trim post-EOF preallocated space. This is queued
614 * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
618 struct xfs_mount *mp)
621 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
622 queue_delayed_work(mp->m_eofblocks_workqueue,
623 &mp->m_eofblocks_work,
624 msecs_to_jiffies(xfs_eofb_secs * 1000));
629 xfs_eofblocks_worker(
630 struct work_struct *work)
632 struct xfs_mount *mp = container_of(to_delayed_work(work),
633 struct xfs_mount, m_eofblocks_work);
634 xfs_icache_free_eofblocks(mp, NULL);
635 xfs_queue_eofblocks(mp);
639 xfs_inode_ag_iterator(
640 struct xfs_mount *mp,
641 int (*execute)(struct xfs_inode *ip, int flags,
646 struct xfs_perag *pag;
652 while ((pag = xfs_perag_get(mp, ag))) {
653 ag = pag->pag_agno + 1;
654 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1);
658 if (error == -EFSCORRUPTED)
666 xfs_inode_ag_iterator_tag(
667 struct xfs_mount *mp,
668 int (*execute)(struct xfs_inode *ip, int flags,
674 struct xfs_perag *pag;
680 while ((pag = xfs_perag_get_tag(mp, ag, tag))) {
681 ag = pag->pag_agno + 1;
682 error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag);
686 if (error == -EFSCORRUPTED)
694 * Queue a new inode reclaim pass if there are reclaimable inodes and there
695 * isn't a reclaim pass already in progress. By default it runs every 5s based
696 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
697 * tunable, but that can be done if this method proves to be ineffective or too
701 xfs_reclaim_work_queue(
702 struct xfs_mount *mp)
706 if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
707 queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
708 msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
714 * This is a fast pass over the inode cache to try to get reclaim moving on as
715 * many inodes as possible in a short period of time. It kicks itself every few
716 * seconds, as well as being kicked by the inode cache shrinker when memory
717 * goes low. It scans as quickly as possible avoiding locked inodes or those
718 * already being flushed, and once done schedules a future pass.
722 struct work_struct *work)
724 struct xfs_mount *mp = container_of(to_delayed_work(work),
725 struct xfs_mount, m_reclaim_work);
727 xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
728 xfs_reclaim_work_queue(mp);
732 __xfs_inode_set_reclaim_tag(
733 struct xfs_perag *pag,
734 struct xfs_inode *ip)
736 radix_tree_tag_set(&pag->pag_ici_root,
737 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
738 XFS_ICI_RECLAIM_TAG);
740 if (!pag->pag_ici_reclaimable) {
741 /* propagate the reclaim tag up into the perag radix tree */
742 spin_lock(&ip->i_mount->m_perag_lock);
743 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
744 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
745 XFS_ICI_RECLAIM_TAG);
746 spin_unlock(&ip->i_mount->m_perag_lock);
748 /* schedule periodic background inode reclaim */
749 xfs_reclaim_work_queue(ip->i_mount);
751 trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno,
754 pag->pag_ici_reclaimable++;
758 * We set the inode flag atomically with the radix tree tag.
759 * Once we get tag lookups on the radix tree, this inode flag
763 xfs_inode_set_reclaim_tag(
766 struct xfs_mount *mp = ip->i_mount;
767 struct xfs_perag *pag;
769 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
770 spin_lock(&pag->pag_ici_lock);
771 spin_lock(&ip->i_flags_lock);
772 __xfs_inode_set_reclaim_tag(pag, ip);
773 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
774 spin_unlock(&ip->i_flags_lock);
775 spin_unlock(&pag->pag_ici_lock);
780 __xfs_inode_clear_reclaim(
784 pag->pag_ici_reclaimable--;
785 if (!pag->pag_ici_reclaimable) {
786 /* clear the reclaim tag from the perag radix tree */
787 spin_lock(&ip->i_mount->m_perag_lock);
788 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
789 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
790 XFS_ICI_RECLAIM_TAG);
791 spin_unlock(&ip->i_mount->m_perag_lock);
792 trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno,
798 __xfs_inode_clear_reclaim_tag(
803 radix_tree_tag_clear(&pag->pag_ici_root,
804 XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
805 __xfs_inode_clear_reclaim(pag, ip);
809 * Grab the inode for reclaim exclusively.
810 * Return 0 if we grabbed it, non-zero otherwise.
813 xfs_reclaim_inode_grab(
814 struct xfs_inode *ip,
817 ASSERT(rcu_read_lock_held());
819 /* quick check for stale RCU freed inode */
824 * If we are asked for non-blocking operation, do unlocked checks to
825 * see if the inode already is being flushed or in reclaim to avoid
828 if ((flags & SYNC_TRYLOCK) &&
829 __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
833 * The radix tree lock here protects a thread in xfs_iget from racing
834 * with us starting reclaim on the inode. Once we have the
835 * XFS_IRECLAIM flag set it will not touch us.
837 * Due to RCU lookup, we may find inodes that have been freed and only
838 * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that
839 * aren't candidates for reclaim at all, so we must check the
840 * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
842 spin_lock(&ip->i_flags_lock);
843 if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
844 __xfs_iflags_test(ip, XFS_IRECLAIM)) {
845 /* not a reclaim candidate. */
846 spin_unlock(&ip->i_flags_lock);
849 __xfs_iflags_set(ip, XFS_IRECLAIM);
850 spin_unlock(&ip->i_flags_lock);
855 * Inodes in different states need to be treated differently. The following
856 * table lists the inode states and the reclaim actions necessary:
858 * inode state iflush ret required action
859 * --------------- ---------- ---------------
861 * shutdown EIO unpin and reclaim
862 * clean, unpinned 0 reclaim
863 * stale, unpinned 0 reclaim
864 * clean, pinned(*) 0 requeue
865 * stale, pinned EAGAIN requeue
866 * dirty, async - requeue
867 * dirty, sync 0 reclaim
869 * (*) dgc: I don't think the clean, pinned state is possible but it gets
870 * handled anyway given the order of checks implemented.
872 * Also, because we get the flush lock first, we know that any inode that has
873 * been flushed delwri has had the flush completed by the time we check that
874 * the inode is clean.
876 * Note that because the inode is flushed delayed write by AIL pushing, the
877 * flush lock may already be held here and waiting on it can result in very
878 * long latencies. Hence for sync reclaims, where we wait on the flush lock,
879 * the caller should push the AIL first before trying to reclaim inodes to
880 * minimise the amount of time spent waiting. For background relaim, we only
881 * bother to reclaim clean inodes anyway.
883 * Hence the order of actions after gaining the locks should be:
885 * shutdown => unpin and reclaim
886 * pinned, async => requeue
887 * pinned, sync => unpin
890 * dirty, async => requeue
891 * dirty, sync => flush, wait and reclaim
895 struct xfs_inode *ip,
896 struct xfs_perag *pag,
899 struct xfs_buf *bp = NULL;
904 xfs_ilock(ip, XFS_ILOCK_EXCL);
905 if (!xfs_iflock_nowait(ip)) {
906 if (!(sync_mode & SYNC_WAIT))
911 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
913 xfs_iflush_abort(ip, false);
916 if (xfs_ipincount(ip)) {
917 if (!(sync_mode & SYNC_WAIT))
921 if (xfs_iflags_test(ip, XFS_ISTALE))
923 if (xfs_inode_clean(ip))
927 * Never flush out dirty data during non-blocking reclaim, as it would
928 * just contend with AIL pushing trying to do the same job.
930 if (!(sync_mode & SYNC_WAIT))
934 * Now we have an inode that needs flushing.
936 * Note that xfs_iflush will never block on the inode buffer lock, as
937 * xfs_ifree_cluster() can lock the inode buffer before it locks the
938 * ip->i_lock, and we are doing the exact opposite here. As a result,
939 * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
940 * result in an ABBA deadlock with xfs_ifree_cluster().
942 * As xfs_ifree_cluser() must gather all inodes that are active in the
943 * cache to mark them stale, if we hit this case we don't actually want
944 * to do IO here - we want the inode marked stale so we can simply
945 * reclaim it. Hence if we get an EAGAIN error here, just unlock the
946 * inode, back off and try again. Hopefully the next pass through will
947 * see the stale flag set on the inode.
949 error = xfs_iflush(ip, &bp);
950 if (error == -EAGAIN) {
951 xfs_iunlock(ip, XFS_ILOCK_EXCL);
952 /* backoff longer than in xfs_ifree_cluster */
958 error = xfs_bwrite(bp);
965 xfs_iunlock(ip, XFS_ILOCK_EXCL);
967 XFS_STATS_INC(xs_ig_reclaims);
969 * Remove the inode from the per-AG radix tree.
971 * Because radix_tree_delete won't complain even if the item was never
972 * added to the tree assert that it's been there before to catch
973 * problems with the inode life time early on.
975 spin_lock(&pag->pag_ici_lock);
976 if (!radix_tree_delete(&pag->pag_ici_root,
977 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino)))
979 __xfs_inode_clear_reclaim(pag, ip);
980 spin_unlock(&pag->pag_ici_lock);
983 * Here we do an (almost) spurious inode lock in order to coordinate
984 * with inode cache radix tree lookups. This is because the lookup
985 * can reference the inodes in the cache without taking references.
987 * We make that OK here by ensuring that we wait until the inode is
988 * unlocked after the lookup before we go ahead and free it.
990 xfs_ilock(ip, XFS_ILOCK_EXCL);
992 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1000 xfs_iflags_clear(ip, XFS_IRECLAIM);
1001 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1003 * We could return -EAGAIN here to make reclaim rescan the inode tree in
1004 * a short while. However, this just burns CPU time scanning the tree
1005 * waiting for IO to complete and the reclaim work never goes back to
1006 * the idle state. Instead, return 0 to let the next scheduled
1007 * background reclaim attempt to reclaim the inode again.
1013 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1014 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1015 * then a shut down during filesystem unmount reclaim walk leak all the
1016 * unreclaimed inodes.
1019 xfs_reclaim_inodes_ag(
1020 struct xfs_mount *mp,
1024 struct xfs_perag *pag;
1028 int trylock = flags & SYNC_TRYLOCK;
1034 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1035 unsigned long first_index = 0;
1039 ag = pag->pag_agno + 1;
1042 if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1047 first_index = pag->pag_ici_reclaim_cursor;
1049 mutex_lock(&pag->pag_ici_reclaim_lock);
1052 struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1056 nr_found = radix_tree_gang_lookup_tag(
1058 (void **)batch, first_index,
1060 XFS_ICI_RECLAIM_TAG);
1068 * Grab the inodes before we drop the lock. if we found
1069 * nothing, nr == 0 and the loop will be skipped.
1071 for (i = 0; i < nr_found; i++) {
1072 struct xfs_inode *ip = batch[i];
1074 if (done || xfs_reclaim_inode_grab(ip, flags))
1078 * Update the index for the next lookup. Catch
1079 * overflows into the next AG range which can
1080 * occur if we have inodes in the last block of
1081 * the AG and we are currently pointing to the
1084 * Because we may see inodes that are from the
1085 * wrong AG due to RCU freeing and
1086 * reallocation, only update the index if it
1087 * lies in this AG. It was a race that lead us
1088 * to see this inode, so another lookup from
1089 * the same index will not find it again.
1091 if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1094 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1095 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1099 /* unlock now we've grabbed the inodes. */
1102 for (i = 0; i < nr_found; i++) {
1105 error = xfs_reclaim_inode(batch[i], pag, flags);
1106 if (error && last_error != -EFSCORRUPTED)
1110 *nr_to_scan -= XFS_LOOKUP_BATCH;
1114 } while (nr_found && !done && *nr_to_scan > 0);
1116 if (trylock && !done)
1117 pag->pag_ici_reclaim_cursor = first_index;
1119 pag->pag_ici_reclaim_cursor = 0;
1120 mutex_unlock(&pag->pag_ici_reclaim_lock);
1125 * if we skipped any AG, and we still have scan count remaining, do
1126 * another pass this time using blocking reclaim semantics (i.e
1127 * waiting on the reclaim locks and ignoring the reclaim cursors). This
1128 * ensure that when we get more reclaimers than AGs we block rather
1129 * than spin trying to execute reclaim.
1131 if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1143 int nr_to_scan = INT_MAX;
1145 return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1149 * Scan a certain number of inodes for reclaim.
1151 * When called we make sure that there is a background (fast) inode reclaim in
1152 * progress, while we will throttle the speed of reclaim via doing synchronous
1153 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1154 * them to be cleaned, which we hope will not be very long due to the
1155 * background walker having already kicked the IO off on those dirty inodes.
1158 xfs_reclaim_inodes_nr(
1159 struct xfs_mount *mp,
1162 /* kick background reclaimer and push the AIL */
1163 xfs_reclaim_work_queue(mp);
1164 xfs_ail_push_all(mp->m_ail);
1166 return xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1170 * Return the number of reclaimable inodes in the filesystem for
1171 * the shrinker to determine how much to reclaim.
1174 xfs_reclaim_inodes_count(
1175 struct xfs_mount *mp)
1177 struct xfs_perag *pag;
1178 xfs_agnumber_t ag = 0;
1179 int reclaimable = 0;
1181 while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1182 ag = pag->pag_agno + 1;
1183 reclaimable += pag->pag_ici_reclaimable;
1191 struct xfs_inode *ip,
1192 struct xfs_eofblocks *eofb)
1194 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1195 !uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1198 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1199 !gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1202 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1203 xfs_get_projid(ip) != eofb->eof_prid)
1210 * A union-based inode filtering algorithm. Process the inode if any of the
1211 * criteria match. This is for global/internal scans only.
1214 xfs_inode_match_id_union(
1215 struct xfs_inode *ip,
1216 struct xfs_eofblocks *eofb)
1218 if ((eofb->eof_flags & XFS_EOF_FLAGS_UID) &&
1219 uid_eq(VFS_I(ip)->i_uid, eofb->eof_uid))
1222 if ((eofb->eof_flags & XFS_EOF_FLAGS_GID) &&
1223 gid_eq(VFS_I(ip)->i_gid, eofb->eof_gid))
1226 if ((eofb->eof_flags & XFS_EOF_FLAGS_PRID) &&
1227 xfs_get_projid(ip) == eofb->eof_prid)
1234 xfs_inode_free_eofblocks(
1235 struct xfs_inode *ip,
1240 struct xfs_eofblocks *eofb = args;
1241 bool need_iolock = true;
1244 ASSERT(!eofb || (eofb && eofb->eof_scan_owner != 0));
1246 if (!xfs_can_free_eofblocks(ip, false)) {
1247 /* inode could be preallocated or append-only */
1248 trace_xfs_inode_free_eofblocks_invalid(ip);
1249 xfs_inode_clear_eofblocks_tag(ip);
1254 * If the mapping is dirty the operation can block and wait for some
1255 * time. Unless we are waiting, skip it.
1257 if (!(flags & SYNC_WAIT) &&
1258 mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1262 if (eofb->eof_flags & XFS_EOF_FLAGS_UNION)
1263 match = xfs_inode_match_id_union(ip, eofb);
1265 match = xfs_inode_match_id(ip, eofb);
1269 /* skip the inode if the file size is too small */
1270 if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1271 XFS_ISIZE(ip) < eofb->eof_min_file_size)
1275 * A scan owner implies we already hold the iolock. Skip it in
1276 * xfs_free_eofblocks() to avoid deadlock. This also eliminates
1277 * the possibility of EAGAIN being returned.
1279 if (eofb->eof_scan_owner == ip->i_ino)
1280 need_iolock = false;
1283 ret = xfs_free_eofblocks(ip->i_mount, ip, need_iolock);
1285 /* don't revisit the inode if we're not waiting */
1286 if (ret == -EAGAIN && !(flags & SYNC_WAIT))
1293 xfs_icache_free_eofblocks(
1294 struct xfs_mount *mp,
1295 struct xfs_eofblocks *eofb)
1297 int flags = SYNC_TRYLOCK;
1299 if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
1302 return xfs_inode_ag_iterator_tag(mp, xfs_inode_free_eofblocks, flags,
1303 eofb, XFS_ICI_EOFBLOCKS_TAG);
1307 * Run eofblocks scans on the quotas applicable to the inode. For inodes with
1308 * multiple quotas, we don't know exactly which quota caused an allocation
1309 * failure. We make a best effort by including each quota under low free space
1310 * conditions (less than 1% free space) in the scan.
1313 xfs_inode_free_quota_eofblocks(
1314 struct xfs_inode *ip)
1317 struct xfs_eofblocks eofb = {0};
1318 struct xfs_dquot *dq;
1320 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1323 * Set the scan owner to avoid a potential livelock. Otherwise, the scan
1324 * can repeatedly trylock on the inode we're currently processing. We
1325 * run a sync scan to increase effectiveness and use the union filter to
1326 * cover all applicable quotas in a single scan.
1328 eofb.eof_scan_owner = ip->i_ino;
1329 eofb.eof_flags = XFS_EOF_FLAGS_UNION|XFS_EOF_FLAGS_SYNC;
1331 if (XFS_IS_UQUOTA_ENFORCED(ip->i_mount)) {
1332 dq = xfs_inode_dquot(ip, XFS_DQ_USER);
1333 if (dq && xfs_dquot_lowsp(dq)) {
1334 eofb.eof_uid = VFS_I(ip)->i_uid;
1335 eofb.eof_flags |= XFS_EOF_FLAGS_UID;
1340 if (XFS_IS_GQUOTA_ENFORCED(ip->i_mount)) {
1341 dq = xfs_inode_dquot(ip, XFS_DQ_GROUP);
1342 if (dq && xfs_dquot_lowsp(dq)) {
1343 eofb.eof_gid = VFS_I(ip)->i_gid;
1344 eofb.eof_flags |= XFS_EOF_FLAGS_GID;
1350 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
1356 xfs_inode_set_eofblocks_tag(
1359 struct xfs_mount *mp = ip->i_mount;
1360 struct xfs_perag *pag;
1363 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1364 spin_lock(&pag->pag_ici_lock);
1365 trace_xfs_inode_set_eofblocks_tag(ip);
1367 tagged = radix_tree_tagged(&pag->pag_ici_root,
1368 XFS_ICI_EOFBLOCKS_TAG);
1369 radix_tree_tag_set(&pag->pag_ici_root,
1370 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
1371 XFS_ICI_EOFBLOCKS_TAG);
1373 /* propagate the eofblocks tag up into the perag radix tree */
1374 spin_lock(&ip->i_mount->m_perag_lock);
1375 radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1376 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1377 XFS_ICI_EOFBLOCKS_TAG);
1378 spin_unlock(&ip->i_mount->m_perag_lock);
1380 /* kick off background trimming */
1381 xfs_queue_eofblocks(ip->i_mount);
1383 trace_xfs_perag_set_eofblocks(ip->i_mount, pag->pag_agno,
1387 spin_unlock(&pag->pag_ici_lock);
1392 xfs_inode_clear_eofblocks_tag(
1395 struct xfs_mount *mp = ip->i_mount;
1396 struct xfs_perag *pag;
1398 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1399 spin_lock(&pag->pag_ici_lock);
1400 trace_xfs_inode_clear_eofblocks_tag(ip);
1402 radix_tree_tag_clear(&pag->pag_ici_root,
1403 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
1404 XFS_ICI_EOFBLOCKS_TAG);
1405 if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_EOFBLOCKS_TAG)) {
1406 /* clear the eofblocks tag from the perag radix tree */
1407 spin_lock(&ip->i_mount->m_perag_lock);
1408 radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1409 XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1410 XFS_ICI_EOFBLOCKS_TAG);
1411 spin_unlock(&ip->i_mount->m_perag_lock);
1412 trace_xfs_perag_clear_eofblocks(ip->i_mount, pag->pag_agno,
1416 spin_unlock(&pag->pag_ici_lock);