[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>

20 files changed, 284 insertions, 503 deletions

diff --git a/fs/xfs/linux-2.6/xfs_export.c b/fs/xfs/linux-2.6/xfs_export.c
index e3a5fedac1ba..f6e99fa7a683 100644
--- a/fs/xfs/linux-2.6/xfs_export.c
+++ b/fs/xfs/linux-2.6/xfs_export.c
@@ -17,10 +17,12 @@
  */
 #include "xfs.h"
 #include "xfs_types.h"
-#include "xfs_dmapi.h"
+#include "xfs_inum.h"
 #include "xfs_log.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
+#include "xfs_dmapi.h"
 #include "xfs_mount.h"
 #include "xfs_export.h"
 
diff --git a/fs/xfs/xfs_ag.h b/fs/xfs/xfs_ag.h
index 51c09c114a20..9381b0360c4b 100644
--- a/fs/xfs/xfs_ag.h
+++ b/fs/xfs/xfs_ag.h
@@ -197,6 +197,10 @@ typedef struct xfs_perag
 #endif
 	xfs_perag_busy_t *pagb_list;	/* unstable blocks */
 	atomic_t        pagf_fstrms;    /* # of filestreams active in this AG */
+
+	int		pag_ici_init;	/* incore inode cache initialised */
+	rwlock_t	pag_ici_lock;	/* incore inode lock */
+	struct radix_tree_root pag_ici_root;	/* incore inode cache root */
 } xfs_perag_t;
 
 #define	XFS_AG_MAXLEVELS(mp)		((mp)->m_ag_maxlevels)
diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index b0667cb27d66..c8f2c2886fe4 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -23,6 +23,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
 #include "xfs_buf_item.h"
diff --git a/fs/xfs/xfs_clnt.h b/fs/xfs/xfs_clnt.h
index f89196cb08d2..d16c1b971074 100644
--- a/fs/xfs/xfs_clnt.h
+++ b/fs/xfs/xfs_clnt.h
@@ -89,7 +89,6 @@ struct xfs_mount_args {
 #define XFSMNT_IDELETE		0x08000000	/* inode cluster delete */
 #define XFSMNT_SWALLOC		0x10000000	/* turn on stripe width
 						 * allocation */
-#define XFSMNT_IHASHSIZE	0x20000000	/* inode hash table size */
 #define XFSMNT_DIRSYNC		0x40000000	/* sync creat,link,unlink,rename
 						 * symlink,mkdir,rmdir,mknod */
 #define XFSMNT_FLAGS2		0x80000000	/* more flags set in flags2 */
diff --git a/fs/xfs/xfs_dir2_block.c b/fs/xfs/xfs_dir2_block.c
index f6b919af7b82..c171767e242a 100644
--- a/fs/xfs/xfs_dir2_block.c
+++ b/fs/xfs/xfs_dir2_block.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
diff --git a/fs/xfs/xfs_dir2_data.c b/fs/xfs/xfs_dir2_data.c
index 7ebe295bd6d3..d2452699e9b1 100644
--- a/fs/xfs/xfs_dir2_data.c
+++ b/fs/xfs/xfs_dir2_data.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
diff --git a/fs/xfs/xfs_dir2_node.c b/fs/xfs/xfs_dir2_node.c
index 91c61d9632c8..eb18e399e836 100644
--- a/fs/xfs/xfs_dir2_node.c
+++ b/fs/xfs/xfs_dir2_node.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
diff --git a/fs/xfs/xfs_dir2_sf.c b/fs/xfs/xfs_dir2_sf.c
index c67d73572905..182c70315ad1 100644
--- a/fs/xfs/xfs_dir2_sf.c
+++ b/fs/xfs/xfs_dir2_sf.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
diff --git a/fs/xfs/xfs_error.c b/fs/xfs/xfs_error.c
index 8c4331631337..339f9d4a49cd 100644
--- a/fs/xfs/xfs_error.c
+++ b/fs/xfs/xfs_error.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
diff --git a/fs/xfs/xfs_extfree_item.c b/fs/xfs/xfs_extfree_item.c
index 3b14427ee123..f938a51be81b 100644
--- a/fs/xfs/xfs_extfree_item.c
+++ b/fs/xfs/xfs_extfree_item.c
@@ -23,6 +23,7 @@
 #include "xfs_trans.h"
 #include "xfs_buf_item.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
 #include "xfs_trans_priv.h"
diff --git a/fs/xfs/xfs_iget.c b/fs/xfs/xfs_iget.c
index 114433a22baa..e07dcc1b70a6 100644
--- a/fs/xfs/xfs_iget.c
+++ b/fs/xfs/xfs_iget.c
@@ -40,131 +40,13 @@
 #include "xfs_utils.h"
 
 /*
- * Initialize the inode hash table for the newly mounted file system.
- * Choose an initial table size based on user specified value, else
- * use a simple algorithm using the maximum number of inodes as an
- * indicator for table size, and clamp it between one and some large
- * number of pages.
- */
-void
-xfs_ihash_init(xfs_mount_t *mp)
-{
-	__uint64_t	icount;
-	uint		i;
-
-	if (!mp->m_ihsize) {
-		icount = mp->m_maxicount ? mp->m_maxicount :
-			 (mp->m_sb.sb_dblocks << mp->m_sb.sb_inopblog);
-		mp->m_ihsize = 1 << max_t(uint, 8,
-					(xfs_highbit64(icount) + 1) / 2);
-		mp->m_ihsize = min_t(uint, mp->m_ihsize,
-					(64 * NBPP) / sizeof(xfs_ihash_t));
-	}
-
-	mp->m_ihash = kmem_zalloc_greedy(&mp->m_ihsize,
-					 NBPC * sizeof(xfs_ihash_t),
-					 mp->m_ihsize * sizeof(xfs_ihash_t),
-					 KM_SLEEP | KM_MAYFAIL | KM_LARGE);
-	mp->m_ihsize /= sizeof(xfs_ihash_t);
-	for (i = 0; i < mp->m_ihsize; i++)
-		rwlock_init(&(mp->m_ihash[i].ih_lock));
-}
-
-/*
- * Free up structures allocated by xfs_ihash_init, at unmount time.
- */
-void
-xfs_ihash_free(xfs_mount_t *mp)
-{
-	kmem_free(mp->m_ihash, mp->m_ihsize * sizeof(xfs_ihash_t));
-	mp->m_ihash = NULL;
-}
-
-/*
- * Initialize the inode cluster hash table for the newly mounted file system.
- * Its size is derived from the ihash table size.
- */
-void
-xfs_chash_init(xfs_mount_t *mp)
-{
-	uint	i;
-
-	mp->m_chsize = max_t(uint, 1, mp->m_ihsize /
-			 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog));
-	mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize);
-	mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize
-						 * sizeof(xfs_chash_t),
-						 KM_SLEEP | KM_LARGE);
-	for (i = 0; i < mp->m_chsize; i++) {
-		spinlock_init(&mp->m_chash[i].ch_lock,"xfshash");
-	}
-}
-
-/*
- * Free up structures allocated by xfs_chash_init, at unmount time.
- */
-void
-xfs_chash_free(xfs_mount_t *mp)
-{
-	int	i;
-
-	for (i = 0; i < mp->m_chsize; i++) {
-		spinlock_destroy(&mp->m_chash[i].ch_lock);
-	}
-
-	kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t));
-	mp->m_chash = NULL;
-}
-
-/*
- * Try to move an inode to the front of its hash list if possible
- * (and if its not there already).  Called right after obtaining
- * the list version number and then dropping the read_lock on the
- * hash list in question (which is done right after looking up the
- * inode in question...).
- */
-STATIC void
-xfs_ihash_promote(
-	xfs_ihash_t	*ih,
-	xfs_inode_t	*ip,
-	ulong		version)
-{
-	xfs_inode_t	*iq;
-
-	if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) {
-		if (likely(version == ih->ih_version)) {
-			/* remove from list */
-			if ((iq = ip->i_next)) {
-				iq->i_prevp = ip->i_prevp;
-			}
-			*ip->i_prevp = iq;
-
-			/* insert at list head */
-			iq = ih->ih_next;
-			iq->i_prevp = &ip->i_next;
-			ip->i_next = iq;
-			ip->i_prevp = &ih->ih_next;
-			ih->ih_next = ip;
-		}
-		write_unlock(&ih->ih_lock);
-	}
-}
-
-/*
  * Look up an inode by number in the given file system.
- * The inode is looked up in the hash table for the file system
- * represented by the mount point parameter mp.  Each bucket of
- * the hash table is guarded by an individual semaphore.
- *
- * If the inode is found in the hash table, its corresponding vnode
- * is obtained with a call to vn_get().  This call takes care of
- * coordination with the reclamation of the inode and vnode.  Note
- * that the vmap structure is filled in while holding the hash lock.
- * This gives us the state of the inode/vnode when we found it and
- * is used for coordination in vn_get().
+ * The inode is looked up in the cache held in each AG.
+ * If the inode is found in the cache, attach it to the provided
+ * vnode.
  *
- * If it is not in core, read it in from the file system's device and
- * add the inode into the hash table.
+ * If it is not in core, read it in from the file system's device,
+ * add it to the cache and attach the provided vnode.
  *
  * The inode is locked according to the value of the lock_flags parameter.
  * This flag parameter indicates how and if the inode's IO lock and inode lock
@@ -192,274 +74,241 @@ xfs_iget_core(
 	xfs_inode_t	**ipp,
 	xfs_daddr_t	bno)
 {
-	xfs_ihash_t	*ih;
 	xfs_inode_t	*ip;
 	xfs_inode_t	*iq;
 	bhv_vnode_t	*inode_vp;
-	ulong		version;
 	int		error;
-	/* REFERENCED */
-	xfs_chash_t	*ch;
-	xfs_chashlist_t	*chl, *chlnew;
-	SPLDECL(s);
+	xfs_icluster_t	*icl, *new_icl = NULL;
+	unsigned long	first_index, mask;
+	xfs_perag_t	*pag;
+	xfs_agino_t	agino;
+
+	/* the radix tree exists only in inode capable AGs */
+	if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
+		return EINVAL;
+
+	/* get the perag structure and ensure that it's inode capable */
+	pag = xfs_get_perag(mp, ino);
+	if (!pag->pagi_inodeok)
+		return EINVAL;
+	ASSERT(pag->pag_ici_init);
+	agino = XFS_INO_TO_AGINO(mp, ino);
 
+again:
+	read_lock(&pag->pag_ici_lock);
+	ip = radix_tree_lookup(&pag->pag_ici_root, agino);
 
-	ih = XFS_IHASH(mp, ino);
+	if (ip != NULL) {
+		/*
+		 * If INEW is set this inode is being set up
+		 * we need to pause and try again.
+		 */
+		if (xfs_iflags_test(ip, XFS_INEW)) {
+			read_unlock(&pag->pag_ici_lock);
+			delay(1);
+			XFS_STATS_INC(xs_ig_frecycle);
 
-again:
-	read_lock(&ih->ih_lock);
+			goto again;
+		}
 
-	for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
-		if (ip->i_ino == ino) {
+		inode_vp = XFS_ITOV_NULL(ip);
+		if (inode_vp == NULL) {
 			/*
-			 * If INEW is set this inode is being set up
+			 * If IRECLAIM is set this inode is
+			 * on its way out of the system,
 			 * we need to pause and try again.
 			 */
-			if (xfs_iflags_test(ip, XFS_INEW)) {
-				read_unlock(&ih->ih_lock);
+			if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
+				read_unlock(&pag->pag_ici_lock);
 				delay(1);
 				XFS_STATS_INC(xs_ig_frecycle);
 
 				goto again;
 			}
+			ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 
-			inode_vp = XFS_ITOV_NULL(ip);
-			if (inode_vp == NULL) {
-				/*
-				 * If IRECLAIM is set this inode is
-				 * on its way out of the system,
-				 * we need to pause and try again.
-				 */
-				if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
-					read_unlock(&ih->ih_lock);
-					delay(1);
-					XFS_STATS_INC(xs_ig_frecycle);
-
-					goto again;
-				}
-				ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
-
-				/*
-				 * If lookup is racing with unlink, then we
-				 * should return an error immediately so we
-				 * don't remove it from the reclaim list and
-				 * potentially leak the inode.
-				 */
-				if ((ip->i_d.di_mode == 0) &&
-				    !(flags & XFS_IGET_CREATE)) {
-					read_unlock(&ih->ih_lock);
-					return ENOENT;
-				}
-
-				/*
-				 * There may be transactions sitting in the
-				 * incore log buffers or being flushed to disk
-				 * at this time.  We can't clear the
-				 * XFS_IRECLAIMABLE flag until these
-				 * transactions have hit the disk, otherwise we
-				 * will void the guarantee the flag provides
-				 * xfs_iunpin()
-				 */
-				if (xfs_ipincount(ip)) {
-					read_unlock(&ih->ih_lock);
-					xfs_log_force(mp, 0,
-						XFS_LOG_FORCE|XFS_LOG_SYNC);
-					XFS_STATS_INC(xs_ig_frecycle);
-					goto again;
-				}
-
-				vn_trace_exit(vp, "xfs_iget.alloc",
-					(inst_t *)__return_address);
+			/*
+			 * If lookup is racing with unlink, then we
+			 * should return an error immediately so we
+			 * don't remove it from the reclaim list and
+			 * potentially leak the inode.
+			 */
+			if ((ip->i_d.di_mode == 0) &&
+			    !(flags & XFS_IGET_CREATE)) {
+				read_unlock(&pag->pag_ici_lock);
+				xfs_put_perag(mp, pag);
+				return ENOENT;
+			}
 
-				XFS_STATS_INC(xs_ig_found);
+			/*
+			 * There may be transactions sitting in the
+			 * incore log buffers or being flushed to disk
+			 * at this time.  We can't clear the
+			 * XFS_IRECLAIMABLE flag until these
+			 * transactions have hit the disk, otherwise we
+			 * will void the guarantee the flag provides
+			 * xfs_iunpin()
+			 */
+			if (xfs_ipincount(ip)) {
+				read_unlock(&pag->pag_ici_lock);
+				xfs_log_force(mp, 0,
+					XFS_LOG_FORCE|XFS_LOG_SYNC);
+				XFS_STATS_INC(xs_ig_frecycle);
+				goto again;
+			}
 
-				xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
-				version = ih->ih_version;
-				read_unlock(&ih->ih_lock);
-				xfs_ihash_promote(ih, ip, version);
+			vn_trace_exit(vp, "xfs_iget.alloc",
+				(inst_t *)__return_address);
 
-				XFS_MOUNT_ILOCK(mp);
-				list_del_init(&ip->i_reclaim);
-				XFS_MOUNT_IUNLOCK(mp);
+			XFS_STATS_INC(xs_ig_found);
 
-				goto finish_inode;
+			xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
+			read_unlock(&pag->pag_ici_lock);
 
-			} else if (vp != inode_vp) {
-				struct inode *inode = vn_to_inode(inode_vp);
+			XFS_MOUNT_ILOCK(mp);
+			list_del_init(&ip->i_reclaim);
+			XFS_MOUNT_IUNLOCK(mp);
 
-				/* The inode is being torn down, pause and
-				 * try again.
-				 */
-				if (inode->i_state & (I_FREEING | I_CLEAR)) {
-					read_unlock(&ih->ih_lock);
-					delay(1);
-					XFS_STATS_INC(xs_ig_frecycle);
+			goto finish_inode;
 
-					goto again;
-				}
-/* Chances are the other vnode (the one in the inode) is being torn
- * down right now, and we landed on top of it. Question is, what do
- * we do? Unhook the old inode and hook up the new one?
- */
-				cmn_err(CE_PANIC,
-			"xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
-						inode_vp, vp);
-			}
+		} else if (vp != inode_vp) {
+			struct inode *inode = vn_to_inode(inode_vp);
 
-			/*
-			 * Inode cache hit: if ip is not at the front of
-			 * its hash chain, move it there now.
-			 * Do this with the lock held for update, but
-			 * do statistics after releasing the lock.
+			/* The inode is being torn down, pause and
+			 * try again.
 			 */
-			version = ih->ih_version;
-			read_unlock(&ih->ih_lock);
-			xfs_ihash_promote(ih, ip, version);
-			XFS_STATS_INC(xs_ig_found);
+			if (inode->i_state & (I_FREEING | I_CLEAR)) {
+				read_unlock(&pag->pag_ici_lock);
+				delay(1);
+				XFS_STATS_INC(xs_ig_frecycle);
 
-finish_inode:
-			if (ip->i_d.di_mode == 0) {
-				if (!(flags & XFS_IGET_CREATE))
-					return ENOENT;
-				xfs_iocore_inode_reinit(ip);
+				goto again;
 			}
+/* Chances are the other vnode (the one in the inode) is being torn
+* down right now, and we landed on top of it. Question is, what do
+* we do? Unhook the old inode and hook up the new one?
+*/
+			cmn_err(CE_PANIC,
+		"xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
+					inode_vp, vp);
+		}
 
-			if (lock_flags != 0)
-				xfs_ilock(ip, lock_flags);
+		/*
+		 * Inode cache hit
+		 */
+		read_unlock(&pag->pag_ici_lock);
+		XFS_STATS_INC(xs_ig_found);
 
-			xfs_iflags_clear(ip, XFS_ISTALE);
-			vn_trace_exit(vp, "xfs_iget.found",
-						(inst_t *)__return_address);
-			goto return_ip;
+finish_inode:
+		if (ip->i_d.di_mode == 0) {
+			if (!(flags & XFS_IGET_CREATE)) {
+				xfs_put_perag(mp, pag);
+				return ENOENT;
+			}
+			xfs_iocore_inode_reinit(ip);
 		}
+
+		if (lock_flags != 0)
+			xfs_ilock(ip, lock_flags);
+
+		xfs_iflags_clear(ip, XFS_ISTALE);
+		vn_trace_exit(vp, "xfs_iget.found",
+					(inst_t *)__return_address);
+		goto return_ip;
 	}
 
 	/*
-	 * Inode cache miss: save the hash chain version stamp and unlock
-	 * the chain, so we don't deadlock in vn_alloc.
+	 * Inode cache miss
 	 */
+	read_unlock(&pag->pag_ici_lock);
 	XFS_STATS_INC(xs_ig_missed);
 
-	version = ih->ih_version;
-
-	read_unlock(&ih->ih_lock);
-
 	/*
 	 * Read the disk inode attributes into a new inode structure and get
 	 * a new vnode for it. This should also initialize i_ino and i_mount.
 	 */
 	error = xfs_iread(mp, tp, ino, &ip, bno,
 			  (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
-	if (error)
+	if (error) {
+		xfs_put_perag(mp, pag);
 		return error;
+	}
 
 	vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address);
 
 	xfs_inode_lock_init(ip, vp);
 	xfs_iocore_inode_init(ip);
-
 	if (lock_flags)
 		xfs_ilock(ip, lock_flags);
 
 	if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
 		xfs_idestroy(ip);
+		xfs_put_perag(mp, pag);
 		return ENOENT;
 	}
 
 	/*
-	 * Put ip on its hash chain, unless someone else hashed a duplicate
-	 * after we released the hash lock.
+	 * This is a bit messy - we preallocate everything we _might_
+	 * need before we pick up the ici lock. That way we don't have to
+	 * juggle locks and go all the way back to the start.
 	 */
-	write_lock(&ih->ih_lock);
+	new_icl = kmem_zone_alloc(xfs_icluster_zone, KM_SLEEP);
+	if (radix_tree_preload(GFP_KERNEL)) {
+		delay(1);
+		goto again;
+	}
+	mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
+	first_index = agino & mask;
+	write_lock(&pag->pag_ici_lock);
 
-	if (ih->ih_version != version) {
-		for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) {
-			if (iq->i_ino == ino) {
-				write_unlock(&ih->ih_lock);
-				xfs_idestroy(ip);
+	/*
+	 * Find the cluster if it exists
+	 */
+	icl = NULL;
+	if (radix_tree_gang_lookup(&pag->pag_ici_root, (void**)&iq,
+							first_index, 1)) {
+		if ((iq->i_ino & mask) == first_index)
+			icl = iq->i_cluster;
+	}
 
-				XFS_STATS_INC(xs_ig_dup);
-				goto again;
-			}
-		}
+	/*
+	 * insert the new inode
+	 */
+	error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
+	if (unlikely(error)) {
+		BUG_ON(error != -EEXIST);
+		write_unlock(&pag->pag_ici_lock);
+		radix_tree_preload_end();
+		xfs_idestroy(ip);
+		XFS_STATS_INC(xs_ig_dup);
+		goto again;
 	}
 
 	/*
 	 * These values _must_ be set before releasing ihlock!
 	 */
-	ip->i_hash = ih;
-	if ((iq = ih->ih_next)) {
-		iq->i_prevp = &ip->i_next;
-	}
-	ip->i_next = iq;
-	ip->i_prevp = &ih->ih_next;
-	ih->ih_next = ip;
 	ip->i_udquot = ip->i_gdquot = NULL;
-	ih->ih_version++;
 	xfs_iflags_set(ip, XFS_INEW);
-	write_unlock(&ih->ih_lock);
 
-	/*
-	 * put ip on its cluster's hash chain
-	 */
-	ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL &&
-	       ip->i_cnext == NULL);
-
-	chlnew = NULL;
-	ch = XFS_CHASH(mp, ip->i_blkno);
- chlredo:
-	s = mutex_spinlock(&ch->ch_lock);
-	for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) {
-		if (chl->chl_blkno == ip->i_blkno) {
-
-			/* insert this inode into the doubly-linked list
-			 * where chl points */
-			if ((iq = chl->chl_ip)) {
-				ip->i_cprev = iq->i_cprev;
-				iq->i_cprev->i_cnext = ip;
-				iq->i_cprev = ip;
-				ip->i_cnext = iq;
-			} else {
-				ip->i_cnext = ip;
-				ip->i_cprev = ip;
-			}
-			chl->chl_ip = ip;
-			ip->i_chash = chl;
-			break;
-		}
-	}
+	ASSERT(ip->i_cluster == NULL);
 
-	/* no hash list found for this block; add a new hash list */
-	if (chl == NULL)  {
-		if (chlnew == NULL) {
-			mutex_spinunlock(&ch->ch_lock, s);
-			ASSERT(xfs_chashlist_zone != NULL);
-			chlnew = (xfs_chashlist_t *)
-					kmem_zone_alloc(xfs_chashlist_zone,
-						KM_SLEEP);
-			ASSERT(chlnew != NULL);
-			goto chlredo;
-		} else {
-			ip->i_cnext = ip;
-			ip->i_cprev = ip;
-			ip->i_chash = chlnew;
-			chlnew->chl_ip = ip;
-			chlnew->chl_blkno = ip->i_blkno;
-			if (ch->ch_list)
-				ch->ch_list->chl_prev = chlnew;
-			chlnew->chl_next = ch->ch_list;
-			chlnew->chl_prev = NULL;
-			ch->ch_list = chlnew;
-			chlnew = NULL;
-		}
+	if (!icl) {
+		spin_lock_init(&new_icl->icl_lock);
+		INIT_HLIST_HEAD(&new_icl->icl_inodes);
+		icl = new_icl;
+		new_icl = NULL;
 	} else {
-		if (chlnew != NULL) {
-			kmem_zone_free(xfs_chashlist_zone, chlnew);
-		}
+		ASSERT(!hlist_empty(&icl->icl_inodes));
 	}
+	spin_lock(&icl->icl_lock);
+	hlist_add_head(&ip->i_cnode, &icl->icl_inodes);
+	ip->i_cluster = icl;
+	spin_unlock(&icl->icl_lock);
 
-	mutex_spinunlock(&ch->ch_lock, s);
-
+	write_unlock(&pag->pag_ici_lock);
+	radix_tree_preload_end();
+	if (new_icl)
+		kmem_zone_free(xfs_icluster_zone, new_icl);
 
 	/*
 	 * Link ip to its mount and thread it on the mount's inode list.
@@ -478,6 +327,7 @@ finish_inode:
 	mp->m_inodes = ip;
 
 	XFS_MOUNT_IUNLOCK(mp);
+	xfs_put_perag(mp, pag);
 
  return_ip:
 	ASSERT(ip->i_df.if_ext_max ==
@@ -587,32 +437,19 @@ xfs_inode_incore(xfs_mount_t	*mp,
 		 xfs_ino_t	ino,
 		 xfs_trans_t	*tp)
 {
-	xfs_ihash_t	*ih;
 	xfs_inode_t	*ip;
-	ulong		version;
-
-	ih = XFS_IHASH(mp, ino);
-	read_lock(&ih->ih_lock);
-	for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
-		if (ip->i_ino == ino) {
-			/*
-			 * If we find it and tp matches, return it.
-			 * Also move it to the front of the hash list
-			 * if we find it and it is not already there.
-			 * Otherwise break from the loop and return
-			 * NULL.
-			 */
-			if (ip->i_transp == tp) {
-				version = ih->ih_version;
-				read_unlock(&ih->ih_lock);
-				xfs_ihash_promote(ih, ip, version);
-				return (ip);
-			}
-			break;
-		}
-	}
-	read_unlock(&ih->ih_lock);
-	return (NULL);
+	xfs_perag_t	*pag;
+
+	pag = xfs_get_perag(mp, ino);
+	read_lock(&pag->pag_ici_lock);
+	ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
+	read_unlock(&pag->pag_ici_lock);
+	xfs_put_perag(mp, pag);
+
+	/* the returned inode must match the transaction */
+	if (ip && (ip->i_transp != tp))
+		return NULL;
+	return ip;
 }
 
 /*
@@ -718,58 +555,26 @@ void
 xfs_iextract(
 	xfs_inode_t	*ip)
 {
-	xfs_ihash_t	*ih;
+	xfs_mount_t	*mp = ip->i_mount;
+	xfs_perag_t	*pag = xfs_get_perag(mp, ip->i_ino);
 	xfs_inode_t	*iq;
-	xfs_mount_t	*mp;
-	xfs_chash_t	*ch;
-	xfs_chashlist_t *chl, *chm;
-	SPLDECL(s);
-
-	ih = ip->i_hash;
-	write_lock(&ih->ih_lock);
-	if ((iq = ip->i_next)) {
-		iq->i_prevp = ip->i_prevp;
-	}
-	*ip->i_prevp = iq;
-	ih->ih_version++;
-	write_unlock(&ih->ih_lock);
+
+	write_lock(&pag->pag_ici_lock);
+	radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
+	write_unlock(&pag->pag_ici_lock);
+	xfs_put_perag(mp, pag);
 
 	/*
-	 * Remove from cluster hash list
-	 *   1) delete the chashlist if this is the last inode on the chashlist
-	 *   2) unchain from list of inodes
-	 *   3) point chashlist->chl_ip to 'chl_next' if to this inode.
+	 * Remove from cluster list
 	 */
 	mp = ip->i_mount;
-	ch = XFS_CHASH(mp, ip->i_blkno);
-	s = mutex_spinlock(&ch->ch_lock);
-
-	if (ip->i_cnext == ip) {
-		/* Last inode on chashlist */
-		ASSERT(ip->i_cnext == ip && ip->i_cprev == ip);
-		ASSERT(ip->i_chash != NULL);
-		chm=NULL;
-		chl = ip->i_chash;
-		if (chl->chl_prev)
-			chl->chl_prev->chl_next = chl->chl_next;
-		else
-			ch->ch_list = chl->chl_next;
-		if (chl->chl_next)
-			chl->chl_next->chl_prev = chl->chl_prev;
-		kmem_zone_free(xfs_chashlist_zone, chl);
-	} else {
-		/* delete one inode from a non-empty list */
-		iq = ip->i_cnext;
-		iq->i_cprev = ip->i_cprev;
-		ip->i_cprev->i_cnext = iq;
-		if (ip->i_chash->chl_ip == ip) {
-			ip->i_chash->chl_ip = iq;
-		}
-		ip->i_chash = __return_address;
-		ip->i_cprev = __return_address;
-		ip->i_cnext = __return_address;
-	}
-	mutex_spinunlock(&ch->ch_lock, s);
+	spin_lock(&ip->i_cluster->icl_lock);
+	hlist_del(&ip->i_cnode);
+	spin_unlock(&ip->i_cluster->icl_lock);
+
+	/* was last inode in cluster? */
+	if (hlist_empty(&ip->i_cluster->icl_inodes))
+		kmem_zone_free(xfs_icluster_zone, ip->i_cluster);
 
 	/*
 	 * Remove from mount's inode list.
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index 41a0c73b601a..c1b917bd5951 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 @@ xfs_ifree_cluster(
 	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 @@ xfs_ifree_cluster(
 		 */
 		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 @@ xfs_ifree_cluster(
 						ip_found[found++] = ip;
 					}
 				}
-				read_unlock(&ih->ih_lock);
+				read_unlock(&pag->pag_ici_lock);
 				continue;
 			}
 
@@ -2277,8 +2274,7 @@ xfs_ifree_cluster(
 					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 @@ xfs_ifree_cluster(
 	}
 
 	kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
+	xfs_put_perag(mp, pag);
 }
 
 /*
@@ -3050,12 +3047,11 @@ xfs_iflush(
 	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 @@ xfs_iflush(
 	 * 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_iflush(
 			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 @@ cluster_corrupt_out:
 	/* 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 --
diff --git a/fs/xfs/xfs_inode.h b/fs/xfs/xfs_inode.h
index 873b9f783d29..b6dd23d9b3d6 100644
--- a/fs/xfs/xfs_inode.h
+++ b/fs/xfs/xfs_inode.h
@@ -172,41 +172,18 @@ typedef struct xfs_iocore {
 extern void xfs_iocore_inode_init(struct xfs_inode *);
 extern void xfs_iocore_inode_reinit(struct xfs_inode *);
 
-
-/*
- * This is the type used in the xfs inode hash table.
- * An array of these is allocated for each mounted
- * file system to hash the inodes for that file system.
- */
-typedef struct xfs_ihash {
-	struct xfs_inode	*ih_next;
-	rwlock_t		ih_lock;
-	uint			ih_version;
-} xfs_ihash_t;
-
-#define XFS_IHASH(mp,ino) ((mp)->m_ihash + (((uint)(ino)) % (mp)->m_ihsize))
-
 /*
- * This is the xfs inode cluster hash.  This hash is used by xfs_iflush to
- * find inodes that share a cluster and can be flushed to disk at the same
- * time.
+ * This is the xfs inode cluster structure.  This structure is used by
+ * xfs_iflush to find inodes that share a cluster and can be flushed to disk at
+ * the same time.
  */
-typedef struct xfs_chashlist {
-	struct xfs_chashlist	*chl_next;
-	struct xfs_chashlist	*chl_prev;
-	struct xfs_inode	*chl_ip;
-	xfs_daddr_t		chl_blkno;	/* starting block number of
+typedef struct xfs_icluster {
+	struct hlist_head	icl_inodes;	/* list of inodes on cluster */
+	xfs_daddr_t		icl_blkno;	/* starting block number of
 						 * the cluster */
-	struct xfs_buf		*chl_buf;	/* the inode buffer */
-} xfs_chashlist_t;
-
-typedef struct xfs_chash {
-	xfs_chashlist_t		*ch_list;
-	lock_t			ch_lock;
-} xfs_chash_t;
-
-#define XFS_CHASH(mp,blk) ((mp)->m_chash + (((uint)blk) % (mp)->m_chsize))
-
+	struct xfs_buf		*icl_buf;	/* the inode buffer */
+	lock_t			icl_lock;	/* inode list lock */
+} xfs_icluster_t;
 
 /*
  * This is the xfs in-core inode structure.
@@ -269,21 +246,15 @@ typedef struct xfs_icdinode {
 } xfs_icdinode_t;
 
 typedef struct {
-	struct xfs_ihash	*ip_hash;	/* pointer to hash header */
-	struct xfs_inode	*ip_next;	/* inode hash link forw */
 	struct xfs_inode	*ip_mnext;	/* next inode in mount list */
 	struct xfs_inode	*ip_mprev;	/* ptr to prev inode */
-	struct xfs_inode	**ip_prevp;	/* ptr to prev i_next */
 	struct xfs_mount	*ip_mount;	/* fs mount struct ptr */
 } xfs_iptr_t;
 
 typedef struct xfs_inode {
 	/* Inode linking and identification information. */
-	struct xfs_ihash	*i_hash;	/* pointer to hash header */
-	struct xfs_inode	*i_next;	/* inode hash link forw */
 	struct xfs_inode	*i_mnext;	/* next inode in mount list */
 	struct xfs_inode	*i_mprev;	/* ptr to prev inode */
-	struct xfs_inode	**i_prevp;	/* ptr to prev i_next */
 	struct xfs_mount	*i_mount;	/* fs mount struct ptr */
 	struct list_head	i_reclaim;	/* reclaim list */
 	struct bhv_desc		i_bhv_desc;	/* inode behavior descriptor*/
@@ -324,9 +295,8 @@ typedef struct xfs_inode {
 	unsigned int		i_delayed_blks;	/* count of delay alloc blks */
 
 	xfs_icdinode_t		i_d;		/* most of ondisk inode */
-	xfs_chashlist_t		*i_chash;	/* cluster hash list header */
-	struct xfs_inode	*i_cnext;	/* cluster hash link forward */
-	struct xfs_inode	*i_cprev;	/* cluster hash link backward */
+	xfs_icluster_t		*i_cluster;	/* cluster list header */
+	struct hlist_node	i_cnode;	/* cluster link node */
 
 	xfs_fsize_t		i_size;		/* in-memory size */
 	/* Trace buffers per inode. */
@@ -521,8 +491,6 @@ xfs_iflags_test(xfs_inode_t *ip, unsigned short flags)
  */
 void		xfs_ihash_init(struct xfs_mount *);
 void		xfs_ihash_free(struct xfs_mount *);
-void		xfs_chash_init(struct xfs_mount *);
-void		xfs_chash_free(struct xfs_mount *);
 xfs_inode_t	*xfs_inode_incore(struct xfs_mount *, xfs_ino_t,
 				  struct xfs_trans *);
 void            xfs_inode_lock_init(xfs_inode_t *, struct bhv_vnode *);
@@ -633,7 +601,7 @@ void		xfs_inobp_check(struct xfs_mount *, struct xfs_buf *);
 #define	xfs_inobp_check(mp, bp)
 #endif /* DEBUG */
 
-extern struct kmem_zone	*xfs_chashlist_zone;
+extern struct kmem_zone	*xfs_icluster_zone;
 extern struct kmem_zone	*xfs_ifork_zone;
 extern struct kmem_zone	*xfs_inode_zone;
 extern struct kmem_zone	*xfs_ili_zone;
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index f4daf1ec9931..71f25947251d 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -160,11 +160,6 @@ xfs_mount_free(
 	xfs_mount_t	*mp,
 	int		remove_bhv)
 {
-	if (mp->m_ihash)
-		xfs_ihash_free(mp);
-	if (mp->m_chash)
-		xfs_chash_free(mp);
-
 	if (mp->m_perag) {
 		int	agno;
 
@@ -342,6 +337,17 @@ xfs_mount_validate_sb(
 	return 0;
 }
 
+STATIC void
+xfs_initialize_perag_icache(
+	xfs_perag_t	*pag)
+{
+	if (!pag->pag_ici_init) {
+		rwlock_init(&pag->pag_ici_lock);
+		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
+		pag->pag_ici_init = 1;
+	}
+}
+
 xfs_agnumber_t
 xfs_initialize_perag(
 	bhv_vfs_t	*vfs,
@@ -396,12 +402,14 @@ xfs_initialize_perag(
 			pag->pagi_inodeok = 1;
 			if (index < max_metadata)
 				pag->pagf_metadata = 1;
+			xfs_initialize_perag_icache(pag);
 		}
 	} else {
 		/* Setup default behavior for smaller filesystems */
 		for (index = 0; index < agcount; index++) {
 			pag = &mp->m_perag[index];
 			pag->pagi_inodeok = 1;
+			xfs_initialize_perag_icache(pag);
 		}
 	}
 	return index;
@@ -1033,13 +1041,6 @@ xfs_mountfs(
 	xfs_trans_init(mp);
 
 	/*
-	 * Allocate and initialize the inode hash table for this
-	 * file system.
-	 */
-	xfs_ihash_init(mp);
-	xfs_chash_init(mp);
-
-	/*
 	 * Allocate and initialize the per-ag data.
 	 */
 	init_rwsem(&mp->m_peraglock);
@@ -1190,8 +1191,6 @@ xfs_mountfs(
  error3:
 	xfs_log_unmount_dealloc(mp);
  error2:
-	xfs_ihash_free(mp);
-	xfs_chash_free(mp);
 	for (agno = 0; agno < sbp->sb_agcount; agno++)
 		if (mp->m_perag[agno].pagb_list)
 			kmem_free(mp->m_perag[agno].pagb_list,
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index 9ceff40326d0..bc23cb407701 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -57,10 +57,7 @@ struct log;
 struct bhv_vfs;
 struct bhv_vnode;
 struct xfs_mount_args;
-struct xfs_ihash;
-struct xfs_chash;
 struct xfs_inode;
-struct xfs_perag;
 struct xfs_iocore;
 struct xfs_bmbt_irec;
 struct xfs_bmap_free;
@@ -335,8 +332,6 @@ typedef struct xfs_mount {
 	xfs_agnumber_t		m_agirotor;	/* last ag dir inode alloced */
 	lock_t			m_agirotor_lock;/* .. and lock protecting it */
 	xfs_agnumber_t		m_maxagi;	/* highest inode alloc group */
-	size_t			m_ihsize;	/* size of next field */
-	struct xfs_ihash	*m_ihash;	/* fs private inode hash table*/
 	struct xfs_inode	*m_inodes;	/* active inode list */
 	struct list_head	m_del_inodes;	/* inodes to reclaim */
 	mutex_t			m_ilock;	/* inode list mutex */
@@ -458,7 +453,7 @@ typedef struct xfs_mount {
 #define XFS_MOUNT_IDELETE	(1ULL << 18)	/* delete empty inode clusters*/
 #define XFS_MOUNT_SWALLOC	(1ULL << 19)	/* turn on stripe width
 						 * allocation */
-#define XFS_MOUNT_IHASHSIZE	(1ULL << 20)	/* inode hash table size */
+			     /*	(1ULL << 20)	-- currently unused */
 #define XFS_MOUNT_DIRSYNC	(1ULL << 21)	/* synchronous directory ops */
 #define XFS_MOUNT_COMPAT_IOSIZE	(1ULL << 22)	/* don't report large preferred
 						 * I/O size in stat() */
@@ -572,6 +567,21 @@ xfs_daddr_to_agbno(struct xfs_mount *mp, xfs_daddr_t d)
 }
 
 /*
+ * perag get/put wrappers for eventual ref counting
+ */
+static inline xfs_perag_t *
+xfs_get_perag(struct xfs_mount *mp, xfs_ino_t ino)
+{
+	return &mp->m_perag[XFS_INO_TO_AGNO(mp, ino)];
+}
+
+static inline void
+xfs_put_perag(struct xfs_mount *mp, xfs_perag_t *pag)
+{
+	/* nothing to see here, move along */
+}
+
+/*
  * Per-cpu superblock locking functions
  */
 #ifdef HAVE_PERCPU_SB
diff --git a/fs/xfs/xfs_rename.c b/fs/xfs/xfs_rename.c
index 53d9600af4a4..187318e8d259 100644
--- a/fs/xfs/xfs_rename.c
+++ b/fs/xfs/xfs_rename.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
diff --git a/fs/xfs/xfs_trans_ail.c b/fs/xfs/xfs_trans_ail.c
index ceb4f6e99960..5b2ff59f19cf 100644
--- a/fs/xfs/xfs_trans_ail.c
+++ b/fs/xfs/xfs_trans_ail.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
 #include "xfs_trans_priv.h"
diff --git a/fs/xfs/xfs_trans_extfree.c b/fs/xfs/xfs_trans_extfree.c
index b290270dd4a6..27cce2a9c7e9 100644
--- a/fs/xfs/xfs_trans_extfree.c
+++ b/fs/xfs/xfs_trans_extfree.c
@@ -22,6 +22,7 @@
 #include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
+#include "xfs_ag.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
 #include "xfs_trans_priv.h"
diff --git a/fs/xfs/xfs_vfsops.c b/fs/xfs/xfs_vfsops.c
index 4a27648b5446..1644be14a144 100644
--- a/fs/xfs/xfs_vfsops.c
+++ b/fs/xfs/xfs_vfsops.c
@@ -117,8 +117,8 @@ xfs_init(void)
 	xfs_ili_zone =
 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 					KM_ZONE_SPREAD, NULL);
-	xfs_chashlist_zone =
-		kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
+	xfs_icluster_zone =
+		kmem_zone_init_flags(sizeof(xfs_icluster_t), "xfs_icluster",
 					KM_ZONE_SPREAD, NULL);
 
 	/*
@@ -163,7 +163,7 @@ xfs_cleanup(void)
 	extern kmem_zone_t	*xfs_efd_zone;
 	extern kmem_zone_t	*xfs_efi_zone;
 	extern kmem_zone_t	*xfs_buf_item_zone;
-	extern kmem_zone_t	*xfs_chashlist_zone;
+	extern kmem_zone_t	*xfs_icluster_zone;
 
 	xfs_cleanup_procfs();
 	xfs_sysctl_unregister();
@@ -199,7 +199,7 @@ xfs_cleanup(void)
 	kmem_zone_destroy(xfs_efi_zone);
 	kmem_zone_destroy(xfs_ifork_zone);
 	kmem_zone_destroy(xfs_ili_zone);
-	kmem_zone_destroy(xfs_chashlist_zone);
+	kmem_zone_destroy(xfs_icluster_zone);
 }
 
 /*
@@ -246,7 +246,6 @@ xfs_start_flags(
 			ap->logbufsize);
 		return XFS_ERROR(EINVAL);
 	}
-	mp->m_ihsize = ap->ihashsize;
 	mp->m_logbsize = ap->logbufsize;
 	mp->m_fsname_len = strlen(ap->fsname) + 1;
 	mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
@@ -293,8 +292,6 @@ xfs_start_flags(
 		mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 	}
 
-	if (ap->flags & XFSMNT_IHASHSIZE)
-		mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 	if (ap->flags & XFSMNT_IDELETE)
 		mp->m_flags |= XFS_MOUNT_IDELETE;
 	if (ap->flags & XFSMNT_DIRSYNC)
@@ -1673,7 +1670,6 @@ xfs_vget(
 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
-#define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
 #define MNTOPT_BARRIER	"barrier"	/* use writer barriers for log write and
 					 * unwritten extent conversion */
@@ -1799,15 +1795,6 @@ xfs_parseargs(
 			iosize = suffix_strtoul(value, &eov, 10);
 			args->flags |= XFSMNT_IOSIZE;
 			args->iosizelog = ffs(iosize) - 1;
-		} else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
-			if (!value || !*value) {
-				cmn_err(CE_WARN,
-					"XFS: %s option requires an argument",
-					this_char);
-				return EINVAL;
-			}
-			args->flags |= XFSMNT_IHASHSIZE;
-			args->ihashsize = simple_strtoul(value, &eov, 10);
 		} else if (!strcmp(this_char, MNTOPT_GRPID) ||
 			   !strcmp(this_char, MNTOPT_BSDGROUPS)) {
 			vfsp->vfs_flag |= VFS_GRPID;
@@ -1876,6 +1863,9 @@ xfs_parseargs(
 			args->flags &= ~XFSMNT_ATTR2;
 		} else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
 			args->flags2 |= XFSMNT2_FILESTREAMS;
+		} else if (!strcmp(this_char, "ihashsize")) {
+			cmn_err(CE_WARN,
+	"XFS: ihashsize no longer used, option is deprecated.");
 		} else if (!strcmp(this_char, "osyncisdsync")) {
 			/* no-op, this is now the default */
 			cmn_err(CE_WARN,
@@ -1966,9 +1956,6 @@ xfs_showargs(
 			seq_puts(m, xfs_infop->str);
 	}
 
-	if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
-		seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", (int)mp->m_ihsize);
-
 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
 		seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
 				(int)(1 << mp->m_writeio_log) >> 10);
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index bde4a1ad90f2..15bc01b2d6a0 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -3876,7 +3876,7 @@ xfs_finish_reclaim(
 	int		locked,
 	int		sync_mode)
 {
-	xfs_ihash_t	*ih = ip->i_hash;
+	xfs_perag_t	*pag = xfs_get_perag(ip->i_mount, ip->i_ino);
 	bhv_vnode_t	*vp = XFS_ITOV_NULL(ip);
 	int		error;
 
@@ -3888,12 +3888,12 @@ xfs_finish_reclaim(
 	 * Once we have the XFS_IRECLAIM flag set it will not touch
 	 * us.
 	 */
-	write_lock(&ih->ih_lock);
+	write_lock(&pag->pag_ici_lock);
 	spin_lock(&ip->i_flags_lock);
 	if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
 	    (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
 		spin_unlock(&ip->i_flags_lock);
-		write_unlock(&ih->ih_lock);
+		write_unlock(&pag->pag_ici_lock);
 		if (locked) {
 			xfs_ifunlock(ip);
 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
@@ -3902,7 +3902,8 @@ xfs_finish_reclaim(
 	}
 	__xfs_iflags_set(ip, XFS_IRECLAIM);
 	spin_unlock(&ip->i_flags_lock);
-	write_unlock(&ih->ih_lock);
+	write_unlock(&pag->pag_ici_lock);
+	xfs_put_perag(ip->i_mount, pag);
 
 	/*
 	 * If the inode is still dirty, then flush it out.  If the inode