[XFS] Radix tree based inode caching
One of the perpetual scaling problems XFS has is indexing it's incore
inodes. We currently uses hashes and the default hash sizes chosen can
only ever be a tradeoff between memory consumption and the maximum
realistic size of the cache.
As a result, anyone who has millions of inodes cached on a filesystem
needs to tunes the size of the cache via the ihashsize mount option to
allow decent scalability with inode cache operations.
A further problem is the separate inode cluster hash, whose size is based
on the ihashsize but is smaller, and so under certain conditions (sparse
cluster cache population) this can become a limitation long before the
inode hash is causing issues.
The following patchset removes the inode hash and cluster hash and
replaces them with radix trees to avoid the scalability limitations of the
hashes. It also reduces the size of the inodes by 3 pointers....
SGI-PV: 969561
SGI-Modid: xfs-linux-melb:xfs-kern:29481a
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Tim Shimmin <tes@sgi.com>
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index 41a0c73..c1b917b 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -52,7 +52,7 @@
kmem_zone_t *xfs_ifork_zone;
kmem_zone_t *xfs_inode_zone;
-kmem_zone_t *xfs_chashlist_zone;
+kmem_zone_t *xfs_icluster_zone;
/*
* Used in xfs_itruncate(). This is the maximum number of extents
@@ -2182,10 +2182,10 @@
int i, j, found, pre_flushed;
xfs_daddr_t blkno;
xfs_buf_t *bp;
- xfs_ihash_t *ih;
xfs_inode_t *ip, **ip_found;
xfs_inode_log_item_t *iip;
xfs_log_item_t *lip;
+ xfs_perag_t *pag = xfs_get_perag(mp, inum);
SPLDECL(s);
if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
@@ -2220,23 +2220,20 @@
*/
found = 0;
for (i = 0; i < ninodes; i++) {
- ih = XFS_IHASH(mp, inum + i);
- read_lock(&ih->ih_lock);
- for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
- if (ip->i_ino == inum + i)
- break;
- }
+ read_lock(&pag->pag_ici_lock);
+ ip = radix_tree_lookup(&pag->pag_ici_root,
+ XFS_INO_TO_AGINO(mp, (inum + i)));
/* Inode not in memory or we found it already,
* nothing to do
*/
if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
- read_unlock(&ih->ih_lock);
+ read_unlock(&pag->pag_ici_lock);
continue;
}
if (xfs_inode_clean(ip)) {
- read_unlock(&ih->ih_lock);
+ read_unlock(&pag->pag_ici_lock);
continue;
}
@@ -2259,7 +2256,7 @@
ip_found[found++] = ip;
}
}
- read_unlock(&ih->ih_lock);
+ read_unlock(&pag->pag_ici_lock);
continue;
}
@@ -2277,8 +2274,7 @@
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
}
-
- read_unlock(&ih->ih_lock);
+ read_unlock(&pag->pag_ici_lock);
}
bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
@@ -2333,6 +2329,7 @@
}
kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
+ xfs_put_perag(mp, pag);
}
/*
@@ -3050,12 +3047,11 @@
xfs_mount_t *mp;
int error;
/* REFERENCED */
- xfs_chash_t *ch;
xfs_inode_t *iq;
int clcount; /* count of inodes clustered */
int bufwasdelwri;
+ struct hlist_node *entry;
enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
- SPLDECL(s);
XFS_STATS_INC(xs_iflush_count);
@@ -3169,14 +3165,14 @@
* inode clustering:
* see if other inodes can be gathered into this write
*/
-
- ip->i_chash->chl_buf = bp;
-
- ch = XFS_CHASH(mp, ip->i_blkno);
- s = mutex_spinlock(&ch->ch_lock);
+ spin_lock(&ip->i_cluster->icl_lock);
+ ip->i_cluster->icl_buf = bp;
clcount = 0;
- for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
+ hlist_for_each_entry(iq, entry, &ip->i_cluster->icl_inodes, i_cnode) {
+ if (iq == ip)
+ continue;
+
/*
* Do an un-protected check to see if the inode is dirty and
* is a candidate for flushing. These checks will be repeated
@@ -3227,7 +3223,7 @@
xfs_iunlock(iq, XFS_ILOCK_SHARED);
}
}
- mutex_spinunlock(&ch->ch_lock, s);
+ spin_unlock(&ip->i_cluster->icl_lock);
if (clcount) {
XFS_STATS_INC(xs_icluster_flushcnt);
@@ -3264,7 +3260,7 @@
/* Corruption detected in the clustering loop. Invalidate the
* inode buffer and shut down the filesystem.
*/
- mutex_spinunlock(&ch->ch_lock, s);
+ spin_unlock(&ip->i_cluster->icl_lock);
/*
* Clean up the buffer. If it was B_DELWRI, just release it --