XFS: replace the mount inode list with radix tree traversals V4

index 7d4bdfd..aba1cf0 100644
@@ -58,6 +58,7 @@
 #include "xfs_extfree_item.h"
 #include "xfs_mru_cache.h"
 #include "xfs_inode_item.h"
+#include "xfs_sync.h"
 
 #include <linux/namei.h>
 #include <linux/init.h>

605 lines not on this page · raw message

new file mode 100644
index 0000000..f4c3b1e
@@ -0,0 +1,7 @@
+#ifndef XFS_SYNC_H
+#define XFS_SYNC_H 1
+
+int xfs_sync(struct xfs_mount *mp, int flags);
+int xfs_syncsub(struct xfs_mount *mp, int flags, int *bypassed);
+
+#endif

569 lines not on this page · raw message

index a74b050..6701d0e 100644
@@ -8,7 +8,6 @@ struct kstatfs;
 struct xfs_mount;
 struct xfs_mount_args;
 
-int xfs_sync(struct xfs_mount *mp, int flags);
 void xfs_do_force_shutdown(struct xfs_mount *mp, int flags, char *fname,
 		int lnnum);
 void xfs_attr_quiesce(struct xfs_mount *mp);

index aba1cf0..59f6209 100644
@@ -944,146 +944,6 @@ xfs_fs_clear_inode(
 	ASSERT(XFS_I(inode) == NULL);
 }
 
-/*
- * Enqueue a work item to be picked up by the vfs xfssyncd thread.
- * Doing this has two advantages:
- * - It saves on stack space, which is tight in certain situations
- * - It can be used (with care) as a mechanism to avoid deadlocks.
- * Flushing while allocating in a full filesystem requires both.
- */
-STATIC void
-xfs_syncd_queue_work(
-	struct xfs_mount *mp,
-	void		*data,
-	void		(*syncer)(struct xfs_mount *, void *))
-{
-	struct bhv_vfs_sync_work *work;
-
-	work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
-	INIT_LIST_HEAD(&work->w_list);
-	work->w_syncer = syncer;
-	work->w_data = data;
-	work->w_mount = mp;
-	spin_lock(&mp->m_sync_lock);
-	list_add_tail(&work->w_list, &mp->m_sync_list);
-	spin_unlock(&mp->m_sync_lock);
-	wake_up_process(mp->m_sync_task);
-}
-
-/*
- * Flush delayed allocate data, attempting to free up reserved space
- * from existing allocations.  At this point a new allocation attempt
- * has failed with ENOSPC and we are in the process of scratching our
- * heads, looking about for more room...
- */
-STATIC void
-xfs_flush_inode_work(
-	struct xfs_mount *mp,
-	void		*arg)
-{
-	struct inode	*inode = arg;
-	filemap_flush(inode->i_mapping);
-	iput(inode);
-}
-
-void
-xfs_flush_inode(
-	xfs_inode_t	*ip)
-{
-	struct inode	*inode = VFS_I(ip);
-
-	igrab(inode);
-	xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inode_work);
-	delay(msecs_to_jiffies(500));
-}
-
-/*
- * This is the "bigger hammer" version of xfs_flush_inode_work...
- * (IOW, "If at first you don't succeed, use a Bigger Hammer").
- */
-STATIC void
-xfs_flush_device_work(
-	struct xfs_mount *mp,
-	void		*arg)
-{
-	struct inode	*inode = arg;
-	sync_blockdev(mp->m_super->s_bdev);
-	iput(inode);
-}
-
-void
-xfs_flush_device(
-	xfs_inode_t	*ip)
-{
-	struct inode	*inode = VFS_I(ip);
-
-	igrab(inode);
-	xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_device_work);
-	delay(msecs_to_jiffies(500));
-	xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
-}
-
-STATIC void
-xfs_sync_worker(
-	struct xfs_mount *mp,
-	void		*unused)
-{
-	int		error;
-
-	if (!(mp->m_flags & XFS_MOUNT_RDONLY))
-		error = xfs_sync(mp, SYNC_FSDATA | SYNC_BDFLUSH | SYNC_ATTR);
-	mp->m_sync_seq++;
-	wake_up(&mp->m_wait_single_sync_task);
-}
-
-STATIC int
-xfssyncd(
-	void			*arg)
-{
-	struct xfs_mount	*mp = arg;
-	long			timeleft;
-	bhv_vfs_sync_work_t	*work, *n;
-	LIST_HEAD		(tmp);
-
-	set_freezable();
-	timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
-	for (;;) {
-		timeleft = schedule_timeout_interruptible(timeleft);
-		/* swsusp */
-		try_to_freeze();
-		if (kthread_should_stop() && list_empty(&mp->m_sync_list))
-			break;
-
-		spin_lock(&mp->m_sync_lock);
-		/*
-		 * We can get woken by laptop mode, to do a sync -
-		 * that's the (only!) case where the list would be
-		 * empty with time remaining.
-		 */
-		if (!timeleft || list_empty(&mp->m_sync_list)) {
-			if (!timeleft)
-				timeleft = xfs_syncd_centisecs *
-							msecs_to_jiffies(10);
-			INIT_LIST_HEAD(&mp->m_sync_work.w_list);
-			list_add_tail(&mp->m_sync_work.w_list,
-					&mp->m_sync_list);
-		}
-		list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list)
-			list_move(&work->w_list, &tmp);
-		spin_unlock(&mp->m_sync_lock);
-
-		list_for_each_entry_safe(work, n, &tmp, w_list) {
-			(*work->w_syncer)(mp, work->w_data);
-			list_del(&work->w_list);
-			if (work == &mp->m_sync_work)
-				continue;
-			kmem_free(work);
-		}
-	}
-
-	return 0;
-}
-
 STATIC void
 xfs_free_fsname(
 	struct xfs_mount	*mp)

@@ -1102,8 +962,7 @@ xfs_fs_put_super(
 	int			unmount_event_flags = 0;
 	int			error;
 
-	kthread_stop(mp->m_sync_task);
-
+	xfs_syncd_stop(mp);
 	xfs_sync(mp, SYNC_ATTR | SYNC_DELWRI);
 
 #ifdef HAVE_DMAPI

@@ -1753,13 +1612,9 @@ xfs_fs_fill_super(
 		goto fail_vnrele;
 	}
 
-	mp->m_sync_work.w_syncer = xfs_sync_worker;
-	mp->m_sync_work.w_mount = mp;
-	mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd");
-	if (IS_ERR(mp->m_sync_task)) {
-		error = -PTR_ERR(mp->m_sync_task);
+	error = xfs_syncd_init(mp);
+	if (error)
 		goto fail_vnrele;
-	}
 
 	xfs_itrace_exit(XFS_I(sb->s_root->d_inode));
 

index fe2ef4e..56dc48a 100644
@@ -101,9 +101,6 @@ struct block_device;
 
 extern __uint64_t xfs_max_file_offset(unsigned int);
 
-extern void xfs_flush_inode(struct xfs_inode *);
-extern void xfs_flush_device(struct xfs_inode *);
-
 extern void xfs_blkdev_issue_flush(struct xfs_buftarg *);
 
 extern const struct export_operations xfs_export_operations;

172 lines not on this page · raw message

63 lines not on this page · raw message

index 7e60c77..0ab60bc 100644
@@ -33,37 +33,6 @@ struct xfs_mount_args;
 
 typedef struct kstatfs	bhv_statvfs_t;
 
-typedef struct bhv_vfs_sync_work {
-	struct list_head	w_list;
-	struct xfs_mount	*w_mount;
-	void			*w_data;	/* syncer routine argument */
-	void			(*w_syncer)(struct xfs_mount *, void *);
-} bhv_vfs_sync_work_t;
-
-#define SYNC_ATTR		0x0001	/* sync attributes */
-#define SYNC_CLOSE		0x0002	/* close file system down */
-#define SYNC_DELWRI		0x0004	/* look at delayed writes */
-#define SYNC_WAIT		0x0008	/* wait for i/o to complete */
-#define SYNC_BDFLUSH		0x0010	/* BDFLUSH is calling -- don't block */
-#define SYNC_FSDATA		0x0020	/* flush fs data (e.g. superblocks) */
-#define SYNC_REFCACHE		0x0040  /* prune some of the nfs ref cache */
-#define SYNC_REMOUNT		0x0080  /* remount readonly, no dummy LRs */
-#define SYNC_IOWAIT		0x0100  /* wait for all I/O to complete */
-
-/*
- * When remounting a filesystem read-only or freezing the filesystem,
- * we have two phases to execute. This first phase is syncing the data
- * before we quiesce the fielsystem, and the second is flushing all the
- * inodes out after we've waited for all the transactions created by
- * the first phase to complete. The second phase uses SYNC_INODE_QUIESCE
- * to ensure that the inodes are written to their location on disk
- * rather than just existing in transactions in the log. This means
- * after a quiesce there is no log replay required to write the inodes
- * to disk (this is the main difference between a sync and a quiesce).
- */
-#define SYNC_DATA_QUIESCE	(SYNC_DELWRI|SYNC_FSDATA|SYNC_WAIT|SYNC_IOWAIT)
-#define SYNC_INODE_QUIESCE	(SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
-
 #define SHUTDOWN_META_IO_ERROR	0x0001	/* write attempt to metadata failed */
 #define SHUTDOWN_LOG_IO_ERROR	0x0002	/* write attempt to the log failed */
 #define SHUTDOWN_FORCE_UMOUNT	0x0004	/* shutdown from a forced unmount */

index 64116b5..16b5211 100644
@@ -18,6 +18,7 @@
 #ifndef __XFS_MOUNT_H__
 #define	__XFS_MOUNT_H__
 
+#include "xfs_sync.h"
 
 typedef struct xfs_trans_reservations {
 	uint	tr_write;	/* extent alloc trans */

index a51534c..cd82ba5 100644
@@ -504,7 +504,7 @@ xfs_syncsub(
 
 	if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
 		if (flags & SYNC_BDFLUSH)
-			xfs_finish_reclaim_all(mp, 1);
+			xfs_finish_reclaim_all(mp, 1, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
 		else
 			error = xfs_sync_inodes(mp, flags, bypassed);
 	}

index 00e80df..98361bb 100644
@@ -3459,41 +3459,6 @@ corrupt_out:
 }
 
 
-/*
- * Flush all inactive inodes in mp.
- */
-void
-xfs_iflush_all(
-	xfs_mount_t	*mp)
-{
-	xfs_inode_t	*ip;
-
- again:
-	XFS_MOUNT_ILOCK(mp);
-	ip = mp->m_inodes;
-	if (ip == NULL)
-		goto out;
-
-	do {
-		/* Make sure we skip markers inserted by sync */
-		if (ip->i_mount == NULL) {
-			ip = ip->i_mnext;
-			continue;
-		}
-
-		if (!VFS_I(ip)) {
-			XFS_MOUNT_IUNLOCK(mp);
-			xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
-			goto again;
-		}
-
-		ASSERT(vn_count(VFS_I(ip)) == 0);
-
-		ip = ip->i_mnext;
-	} while (ip != mp->m_inodes);
- out:
-	XFS_MOUNT_IUNLOCK(mp);
-}
 
 #ifdef XFS_ILOCK_TRACE
 ktrace_t	*xfs_ilock_trace_buf;

index 9559e0b..ff68540 100644
@@ -484,7 +484,7 @@ uint		xfs_ilock_map_shared(xfs_inode_t *);
 void		xfs_iunlock_map_shared(xfs_inode_t *, uint);
 void		xfs_ireclaim(xfs_inode_t *);
 int		xfs_finish_reclaim(xfs_inode_t *, int, int);
-int		xfs_finish_reclaim_all(struct xfs_mount *, int);
+int		xfs_finish_reclaim_all(struct xfs_mount *, int, int);
 
 /*
  * xfs_inode.c prototypes.

@@ -522,7 +522,6 @@ void		xfs_ipin(xfs_inode_t *);
 void		xfs_iunpin(xfs_inode_t *);
 int		xfs_iextents_copy(xfs_inode_t *, xfs_bmbt_rec_t *, int);
 int		xfs_iflush(xfs_inode_t *, uint);
-void		xfs_iflush_all(struct xfs_mount *);
 void		xfs_ichgtime(xfs_inode_t *, int);
 xfs_fsize_t	xfs_file_last_byte(xfs_inode_t *);
 void		xfs_lock_inodes(xfs_inode_t **, int, uint);

index a4503f5..e222ab9 100644
@@ -1241,7 +1241,7 @@ xfs_unmountfs(
 	 * need to force the log first.
 	 */
 	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
-	xfs_iflush_all(mp);
+	xfs_finish_reclaim_all(mp, 0, XFS_IFLUSH_ASYNC);
 
 	XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
 

index 01e274b..0c5ee5e 100644
@@ -66,7 +66,7 @@ xfs_quiesce_fs(
 	int			count = 0, pincount;
 
 	xfs_flush_buftarg(mp->m_ddev_targp, 0);
-	xfs_finish_reclaim_all(mp, 0);
+	xfs_finish_reclaim_all(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
 
 	/* This loop must run at least twice.
 	 * The first instance of the loop will flush

54 lines not on this page · raw message

index cd82ba5..53d85ec 100644
@@ -121,356 +121,197 @@ xfs_sync(
 }
 
 /*
- * xfs sync routine for internal use
- *
- * This routine supports all of the flags defined for the generic vfs_sync
- * interface as explained above under xfs_sync.
- *
+ * Sync all the inodes in the given AG according to the
+ * direction given by the flags.
  */
-int
-xfs_sync_inodes(
+STATIC int
+xfs_sync_inodes_ag(
 	xfs_mount_t	*mp,
+	int		ag,
 	int		flags,
-	int             *bypassed)
+	int		*bypassed)
 {
 	xfs_inode_t	*ip = NULL;
 	struct inode	*vp = NULL;
-	int		error;
-	int		last_error;
-	uint64_t	fflag;
-	uint		lock_flags;
-	uint		base_lock_flags;
-	boolean_t	mount_locked;
-	boolean_t	vnode_refed;
-	int		preempt;
-	xfs_iptr_t	*ipointer;
-#ifdef DEBUG
-	boolean_t	ipointer_in = B_FALSE;
-
-#define IPOINTER_SET	ipointer_in = B_TRUE
-#define IPOINTER_CLR	ipointer_in = B_FALSE
-#else
-#define IPOINTER_SET
-#define IPOINTER_CLR
-#endif
-
-
-/* Insert a marker record into the inode list after inode ip. The list
- * must be locked when this is called. After the call the list will no
- * longer be locked.
- */
-#define IPOINTER_INSERT(ip, mp)	{ \
-		ASSERT(ipointer_in == B_FALSE); \
-		ipointer->ip_mnext = ip->i_mnext; \
-		ipointer->ip_mprev = ip; \
-		ip->i_mnext = (xfs_inode_t *)ipointer; \
-		ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
-		preempt = 0; \
-		XFS_MOUNT_IUNLOCK(mp); \
-		mount_locked = B_FALSE; \
-		IPOINTER_SET; \
-	}
-
-/* Remove the marker from the inode list. If the marker was the only item
- * in the list then there are no remaining inodes and we should zero out
- * the whole list. If we are the current head of the list then move the head
- * past us.
- */
-#define IPOINTER_REMOVE(ip, mp)	{ \
-		ASSERT(ipointer_in == B_TRUE); \
-		if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
-			ip = ipointer->ip_mnext; \
-			ip->i_mprev = ipointer->ip_mprev; \
-			ipointer->ip_mprev->i_mnext = ip; \
-			if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
-				mp->m_inodes = ip; \
-			} \
-		} else { \
-			ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
-			mp->m_inodes = NULL; \
-			ip = NULL; \
-		} \
-		IPOINTER_CLR; \
-	}
-
-#define XFS_PREEMPT_MASK	0x7f
-
-	ASSERT(!(flags & SYNC_BDFLUSH));
-
-	if (bypassed)
-		*bypassed = 0;
-	if (mp->m_flags & XFS_MOUNT_RDONLY)
-		return 0;
-	error = 0;
-	last_error = 0;
-	preempt = 0;
-
-	/* Allocate a reference marker */
-	ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
+	xfs_perag_t	*pag = &mp->m_perag[ag];
+	boolean_t	vnode_refed = B_FALSE;
+	int		nr_found;
+	int		first_index = 0;
+	int		error = 0;
+	int		last_error = 0;
+	int		fflag = XFS_B_ASYNC;
+	int		lock_flags = XFS_ILOCK_SHARED;
 
-	fflag = XFS_B_ASYNC;		/* default is don't wait */
 	if (flags & SYNC_DELWRI)
 		fflag = XFS_B_DELWRI;
 	if (flags & SYNC_WAIT)
 		fflag = 0;		/* synchronous overrides all */
 
-	base_lock_flags = XFS_ILOCK_SHARED;
 	if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 		/*
 		 * We need the I/O lock if we're going to call any of
 		 * the flush/inval routines.
 		 */
-		base_lock_flags |= XFS_IOLOCK_SHARED;
+		lock_flags |= XFS_IOLOCK_SHARED;
 	}
 
-	XFS_MOUNT_ILOCK(mp);
-
-	ip = mp->m_inodes;
-
-	mount_locked = B_TRUE;
-	vnode_refed  = B_FALSE;
-
-	IPOINTER_CLR;
-
 	do {
-		ASSERT(ipointer_in == B_FALSE);
-		ASSERT(vnode_refed == B_FALSE);
-
-		lock_flags = base_lock_flags;
-
 		/*
-		 * There were no inodes in the list, just break out
-		 * of the loop.
+		 * use a gang lookup to find the next inode in the tree
+		 * as the tree is sparse and a gang lookup walks to find
+		 * the number of objects requested.
 		 */
-		if (ip == NULL) {
-			break;
-		}
+		read_lock(&pag->pag_ici_lock);
+		nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
+				(void**)&ip, first_index, 1);
 
-		/*
-		 * We found another sync thread marker - skip it
-		 */
-		if (ip->i_mount == NULL) {
-			ip = ip->i_mnext;
-			continue;
+		if (!nr_found) {
+			read_unlock(&pag->pag_ici_lock);
+			break;
 		}
 
-		vp = VFS_I(ip);
+		/* update the index for the next lookup */
+		first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
 
 		/*
-		 * If the vnode is gone then this is being torn down,
-		 * call reclaim if it is flushed, else let regular flush
-		 * code deal with it later in the loop.
+		 * skip inodes in reclaim. Let xfs_syncsub do that for
+		 * us so we don't need to worry.
 		 */
-
-		if (vp == NULL) {
-			/* Skip ones already in reclaim */
-			if (ip->i_flags & XFS_IRECLAIM) {
-				ip = ip->i_mnext;
-				continue;
-			}
-			if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
-				ip = ip->i_mnext;
-			} else if ((xfs_ipincount(ip) == 0) &&
-				    xfs_iflock_nowait(ip)) {
-				IPOINTER_INSERT(ip, mp);
-
-				xfs_finish_reclaim(ip, 1,
-						XFS_IFLUSH_DELWRI_ELSE_ASYNC);
-
-				XFS_MOUNT_ILOCK(mp);
-				mount_locked = B_TRUE;
-				IPOINTER_REMOVE(ip, mp);
-			} else {
-				xfs_iunlock(ip, XFS_ILOCK_EXCL);
-				ip = ip->i_mnext;
-			}
+		vp = VFS_I(ip);
+		if (!vp) {
+			read_unlock(&pag->pag_ici_lock);
 			continue;
 		}
 
+		/* bad inodes are dealt with elsewhere */
 		if (VN_BAD(vp)) {
-			ip = ip->i_mnext;
+			read_unlock(&pag->pag_ici_lock);
 			continue;
 		}
 
+		/* nothing to sync during shutdown */
 		if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
-			XFS_MOUNT_IUNLOCK(mp);
-			kmem_free(ipointer);
+			read_unlock(&pag->pag_ici_lock);
 			return 0;
 		}
 
 		/*
-		 * Try to lock without sleeping.  We're out of order with
-		 * the inode list lock here, so if we fail we need to drop
-		 * the mount lock and try again.  If we're called from
-		 * bdflush() here, then don't bother.
-		 *
-		 * The inode lock here actually coordinates with the
-		 * almost spurious inode lock in xfs_ireclaim() to prevent
-		 * the vnode we handle here without a reference from
-		 * being freed while we reference it.  If we lock the inode
-		 * while it's on the mount list here, then the spurious inode
-		 * lock in xfs_ireclaim() after the inode is pulled from
-		 * the mount list will sleep until we release it here.
-		 * This keeps the vnode from being freed while we reference
-		 * it.
+		 * The inode lock here actually coordinates with the almost
+		 * spurious inode lock in xfs_ireclaim() to prevent the vnode
+		 * we handle here without a reference from being freed while we
+		 * reference it.  If we lock the inode while it's on the mount
+		 * list here, then the spurious inode lock in xfs_ireclaim()
+		 * after the inode is pulled from the mount list will sleep
+		 * until we release it here.  This keeps the vnode from being
+		 * freed while we reference it.
 		 */
 		if (xfs_ilock_nowait(ip, lock_flags) == 0) {
-			if (vp == NULL) {
-				ip = ip->i_mnext;
-				continue;
-			}
-
 			vp = vn_grab(vp);
-			if (vp == NULL) {
-				ip = ip->i_mnext;
+			read_unlock(&pag->pag_ici_lock);
+			if (!vp)
 				continue;
-			}
-
-			IPOINTER_INSERT(ip, mp);
 			xfs_ilock(ip, lock_flags);
 
 			ASSERT(vp == VFS_I(ip));
 			ASSERT(ip->i_mount == mp);
 
 			vnode_refed = B_TRUE;
+		} else {
+			/* safe to unlock here as we have a reference */
+			read_unlock(&pag->pag_ici_lock);
 		}
-
-		/* From here on in the loop we may have a marker record
-		 * in the inode list.
-		 */
-
 		/*
 		 * If we have to flush data or wait for I/O completion
 		 * we need to drop the ilock that we currently hold.
 		 * If we need to drop the lock, insert a marker if we
 		 * have not already done so.
 		 */
-		if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
-		    ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
-			if (mount_locked) {
-				IPOINTER_INSERT(ip, mp);
-			}
+		if (flags & SYNC_CLOSE) {
 			xfs_iunlock(ip, XFS_ILOCK_SHARED);
-
-			if (flags & SYNC_CLOSE) {
-				/* Shutdown case. Flush and invalidate. */
-				if (XFS_FORCED_SHUTDOWN(mp))
-					xfs_tosspages(ip, 0, -1,
-							     FI_REMAPF);
-				else
-					error = xfs_flushinval_pages(ip,
-							0, -1, FI_REMAPF);
-			} else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
-				error = xfs_flush_pages(ip, 0,
-							-1, fflag, FI_NONE);
-			}
-
-			/*
-			 * When freezing, we need to wait ensure all I/O (including direct
-			 * I/O) is complete to ensure no further data modification can take
-			 * place after this point
-			 */
+			if (XFS_FORCED_SHUTDOWN(mp))
+				xfs_tosspages(ip, 0, -1, FI_REMAPF);
+			else
+				error = xfs_flushinval_pages(ip, 0, -1,
+							FI_REMAPF);
+			/* wait for I/O on freeze */
 			if (flags & SYNC_IOWAIT)
 				vn_iowait(ip);
 
 			xfs_ilock(ip, XFS_ILOCK_SHARED);
 		}
 
-		if ((flags & SYNC_ATTR) &&
-		    (ip->i_update_core ||
-		     (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
-			if (mount_locked)
-				IPOINTER_INSERT(ip, mp);
+		if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
+			xfs_iunlock(ip, XFS_ILOCK_SHARED);
+			error = xfs_flush_pages(ip, 0, -1, fflag, FI_NONE);
+			if (flags & SYNC_IOWAIT)
+				vn_iowait(ip);
+			xfs_ilock(ip, XFS_ILOCK_SHARED);
+		}
 
+		if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) {
 			if (flags & SYNC_WAIT) {
 				xfs_iflock(ip);
-				error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
-
-			/*
-			 * If we can't acquire the flush lock, then the inode
-			 * is already being flushed so don't bother waiting.
-			 *
-			 * If we can lock it then do a delwri flush so we can
-			 * combine multiple inode flushes in each disk write.
-			 */
+				if (!xfs_inode_clean(ip))
+					error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
+				else
+					xfs_ifunlock(ip);
 			} else if (xfs_iflock_nowait(ip)) {
-				error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
+				if (!xfs_inode_clean(ip))
+					error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
+				else
+					xfs_ifunlock(ip);
 			} else if (bypassed) {
 				(*bypassed)++;
 			}
 		}
 
-		if (lock_flags != 0) {
+		if (lock_flags)
 			xfs_iunlock(ip, lock_flags);
-		}
 
 		if (vnode_refed) {
-			/*
-			 * If we had to take a reference on the vnode
-			 * above, then wait until after we've unlocked
-			 * the inode to release the reference.  This is
-			 * because we can be already holding the inode
-			 * lock when IRELE() calls xfs_inactive().
-			 *
-			 * Make sure to drop the mount lock before calling
-			 * IRELE() so that we don't trip over ourselves if
-			 * we have to go for the mount lock again in the
-			 * inactive code.
-			 */
-			if (mount_locked) {
-				IPOINTER_INSERT(ip, mp);
-			}
-
 			IRELE(ip);
-
 			vnode_refed = B_FALSE;
 		}
 
-		if (error) {
+		if (error)
 			last_error = error;
-		}
-
 		/*
 		 * bail out if the filesystem is corrupted.
 		 */
-		if (error == EFSCORRUPTED)  {
-			if (!mount_locked) {
-				XFS_MOUNT_ILOCK(mp);
-				IPOINTER_REMOVE(ip, mp);
-			}
-			XFS_MOUNT_IUNLOCK(mp);
-			ASSERT(ipointer_in == B_FALSE);
-			kmem_free(ipointer);
+		if (error == EFSCORRUPTED)
 			return XFS_ERROR(error);
-		}
-
-		/* Let other threads have a chance at the mount lock
-		 * if we have looped many times without dropping the
-		 * lock.
-		 */
-		if ((++preempt & XFS_PREEMPT_MASK) == 0) {
-			if (mount_locked) {
-				IPOINTER_INSERT(ip, mp);
-			}
-		}
-
-		if (mount_locked == B_FALSE) {
-			XFS_MOUNT_ILOCK(mp);
-			mount_locked = B_TRUE;
-			IPOINTER_REMOVE(ip, mp);
-			continue;
-		}
 
-		ASSERT(ipointer_in == B_FALSE);
-		ip = ip->i_mnext;
+	} while (nr_found);
 
-	} while (ip != mp->m_inodes);
+	return last_error;
+}
 
-	XFS_MOUNT_IUNLOCK(mp);
+int
+xfs_sync_inodes(
+	xfs_mount_t	*mp,
+	int		flags,
+	int             *bypassed)
+{
+	int		error;
+	int		last_error;
+	int		i;
 
-	ASSERT(ipointer_in == B_FALSE);
+	if (bypassed)
+		*bypassed = 0;
+	if (mp->m_flags & XFS_MOUNT_RDONLY)
+		return 0;
+	error = 0;
+	last_error = 0;
 
-	kmem_free(ipointer);
+	for (i = 0; i < mp->m_sb.sb_agcount; i++) {
+		if (!mp->m_perag[i].pag_ici_init)
+			continue;
+		error = xfs_sync_inodes_ag(mp, i, flags, bypassed);
+		if (error)
+			last_error = error;
+		if (error == EFSCORRUPTED)
+			break;
+	}
 	return XFS_ERROR(last_error);
 }
 

index 1a3b803..26152b9 100644
@@ -1022,101 +1022,92 @@ xfs_qm_export_flags(
 
 
 /*
- * Go thru all the inodes in the file system, releasing their dquots.
- * Note that the mount structure gets modified to indicate that quotas are off
- * AFTER this, in the case of quotaoff. This also gets called from
- * xfs_rootumount.
+ * Release all the dquots on the inodes in an AG.
  */
-void
-xfs_qm_dqrele_all_inodes(
-	struct xfs_mount *mp,
-	uint		 flags)
+STATIC void
+xfs_qm_dqrele_inodes_ag(
+	xfs_mount_t	*mp,
+	int		ag,
+	uint		flags)
 {
-	xfs_inode_t	*ip, *topino;
-	uint		ireclaims;
-	struct inode	*vp;
-	boolean_t	vnode_refd;
+	xfs_inode_t	*ip = NULL;
+	struct inode	*vp = NULL;
+	xfs_perag_t	*pag = &mp->m_perag[ag];
+	int		first_index = 0;
+	int		nr_found;
 
-	ASSERT(mp->m_quotainfo);
-
-	XFS_MOUNT_ILOCK(mp);
-again:
-	ip = mp->m_inodes;
-	if (ip == NULL) {
-		XFS_MOUNT_IUNLOCK(mp);
-		return;
-	}
 	do {
-		/* Skip markers inserted by xfs_sync */
-		if (ip->i_mount == NULL) {
-			ip = ip->i_mnext;
-			continue;
-		}
-		/* Root inode, rbmip and rsumip have associated blocks */
-		if (ip == XFS_QI_UQIP(mp) || ip == XFS_QI_GQIP(mp)) {
-			ASSERT(ip->i_udquot == NULL);
-			ASSERT(ip->i_gdquot == NULL);
-			ip = ip->i_mnext;
-			continue;
+		boolean_t	vnode_refd = B_FALSE;
+
+		/*
+		 * use a gang lookup to find the next inode in the tree
+		 * as the tree is sparse and a gang lookup walks to find
+		 * the number of objects requested.
+		 */
+		read_lock(&pag->pag_ici_lock);
+		nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
+				(void**)&ip, first_index, 1);
+
+		if (!nr_found) {
+			read_unlock(&pag->pag_ici_lock);
+			break;
 		}
+
+		/* update the index for the next lookup */
+		first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
+
+		/* skip quota inodes and those in reclaim */
 		vp = VFS_I(ip);
-		if (!vp) {
+		if (!vp || ip == XFS_QI_UQIP(mp) || ip == XFS_QI_GQIP(mp)) {
 			ASSERT(ip->i_udquot == NULL);
 			ASSERT(ip->i_gdquot == NULL);
-			ip = ip->i_mnext;
+			read_unlock(&pag->pag_ici_lock);
 			continue;
 		}
-		vnode_refd = B_FALSE;
 		if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
-			ireclaims = mp->m_ireclaims;
-			topino = mp->m_inodes;
 			vp = vn_grab(vp);
+			read_unlock(&pag->pag_ici_lock);
 			if (!vp)
-				goto again;
-
-			XFS_MOUNT_IUNLOCK(mp);
-			/* XXX restart limit ? */
-			xfs_ilock(ip, XFS_ILOCK_EXCL);
+				continue;
 			vnode_refd = B_TRUE;
+			xfs_ilock(ip, XFS_ILOCK_EXCL);
 		} else {
-			ireclaims = mp->m_ireclaims;
-			topino = mp->m_inodes;
-			XFS_MOUNT_IUNLOCK(mp);
+			read_unlock(&pag->pag_ici_lock);
 		}
-
-		/*
-		 * We don't keep the mountlock across the dqrele() call,
-		 * since it can take a while..
-		 */
 		if ((flags & XFS_UQUOTA_ACCT) && ip->i_udquot) {
 			xfs_qm_dqrele(ip->i_udquot);
 			ip->i_udquot = NULL;
 		}
-		if (flags & (XFS_PQUOTA_ACCT|XFS_GQUOTA_ACCT) && ip->i_gdquot) {
+		if (flags & (XFS_PQUOTA_ACCT|XFS_GQUOTA_ACCT) &&
+		    ip->i_gdquot) {
 			xfs_qm_dqrele(ip->i_gdquot);
 			ip->i_gdquot = NULL;
 		}
 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
-		/*
-		 * Wait until we've dropped the ilock and mountlock to
-		 * do the vn_rele. Or be condemned to an eternity in the
-		 * inactive code in hell.
-		 */
 		if (vnode_refd)
 			IRELE(ip);
-		XFS_MOUNT_ILOCK(mp);
-		/*
-		 * If an inode was inserted or removed, we gotta
-		 * start over again.
-		 */
-		if (topino != mp->m_inodes || mp->m_ireclaims != ireclaims) {
-			/* XXX use a sentinel */
-			goto again;
-		}
-		ip = ip->i_mnext;
-	} while (ip != mp->m_inodes);
+	} while (nr_found);
+}
 
-	XFS_MOUNT_IUNLOCK(mp);
+/*
+ * Go thru all the inodes in the file system, releasing their dquots.
+ * Note that the mount structure gets modified to indicate that quotas are off
+ * AFTER this, in the case of quotaoff. This also gets called from
+ * xfs_rootumount.
+ */
+void
+xfs_qm_dqrele_all_inodes(
+	struct xfs_mount *mp,
+	uint		 flags)
+{
+	int		i;
+
+	ASSERT(mp->m_quotainfo);
+	for (i = 0; i < mp->m_sb.sb_agcount; i++) {
+		if (!mp->m_perag[i].pag_ici_init)
+			continue;
+		xfs_qm_dqrele_inodes_ag(mp, i, flags);
+	}
 }
 
 /*------------------------------------------------------------------------*/

index e229e9e..60554ce 100644
@@ -76,7 +76,6 @@ xfs_iget_core(
 {
 	struct inode	*old_inode;
 	xfs_inode_t	*ip;
-	xfs_inode_t	*iq;
 	int		error;
 	unsigned long	first_index, mask;
 	xfs_perag_t	*pag;

@@ -267,24 +266,6 @@ finish_inode:
 
 	write_unlock(&pag->pag_ici_lock);
 	radix_tree_preload_end();
-
-	/*
-	 * Link ip to its mount and thread it on the mount's inode list.
-	 */
-	XFS_MOUNT_ILOCK(mp);
-	if ((iq = mp->m_inodes)) {
-		ASSERT(iq->i_mprev->i_mnext == iq);
-		ip->i_mprev = iq->i_mprev;
-		iq->i_mprev->i_mnext = ip;
-		iq->i_mprev = ip;
-		ip->i_mnext = iq;
-	} else {
-		ip->i_mnext = ip;
-		ip->i_mprev = ip;
-	}
-	mp->m_inodes = ip;
-
-	XFS_MOUNT_IUNLOCK(mp);
 	xfs_put_perag(mp, pag);
 
  return_ip:

@@ -505,36 +486,15 @@ xfs_iextract(
 {
 	xfs_mount_t	*mp = ip->i_mount;
 	xfs_perag_t	*pag = xfs_get_perag(mp, ip->i_ino);
-	xfs_inode_t	*iq;
 
 	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 mount's inode list.
-	 */
-	XFS_MOUNT_ILOCK(mp);
-	ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
-	iq = ip->i_mnext;
-	iq->i_mprev = ip->i_mprev;
-	ip->i_mprev->i_mnext = iq;
-
-	/*
-	 * Fix up the head pointer if it points to the inode being deleted.
-	 */
-	if (mp->m_inodes == ip) {
-		if (ip == iq) {
-			mp->m_inodes = NULL;
-		} else {
-			mp->m_inodes = iq;
-		}
-	}
-
 	/* Deal with the deleted inodes list */
+	XFS_MOUNT_ILOCK(mp);
 	list_del_init(&ip->i_reclaim);
-
 	mp->m_ireclaims++;
 	XFS_MOUNT_IUNLOCK(mp);
 }

index ff68540..7b50eb4 100644
@@ -191,16 +191,8 @@ typedef struct xfs_icdinode {
 	__uint32_t	di_gen;		/* generation number */
 } xfs_icdinode_t;
 
-typedef struct {
-	struct xfs_inode	*ip_mnext;	/* next inode in mount list */
-	struct xfs_inode	*ip_mprev;	/* ptr to prev inode */
-	struct xfs_mount	*ip_mount;	/* fs mount struct ptr */
-} xfs_iptr_t;
-
 typedef struct xfs_inode {
 	/* Inode linking and identification information. */
-	struct xfs_inode	*i_mnext;	/* next inode in mount list */
-	struct xfs_inode	*i_mprev;	/* ptr to prev inode */
 	struct xfs_mount	*i_mount;	/* fs mount struct ptr */
 	struct list_head	i_reclaim;	/* reclaim list */
 	struct inode		*i_vnode;	/* vnode backpointer */

index e222ab9..8372b3d 100644
@@ -1285,11 +1285,6 @@ xfs_unmountfs(
 	xfs_unmountfs_wait(mp); 		/* wait for async bufs */
 	xfs_log_unmount(mp);			/* Done! No more fs ops. */
 
-	/*
-	 * All inodes from this mount point should be freed.
-	 */
-	ASSERT(mp->m_inodes == NULL);
-
 	if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
 		uuid_table_remove(&mp->m_sb.sb_uuid);
 

index 16b5211..e9f2d1a 100644
@@ -248,7 +248,6 @@ typedef struct xfs_mount {
 	xfs_agnumber_t		m_agirotor;	/* last ag dir inode alloced */
 	spinlock_t		m_agirotor_lock;/* .. and lock protecting it */
 	xfs_agnumber_t		m_maxagi;	/* highest inode alloc group */
-	struct xfs_inode	*m_inodes;	/* active inode list */
 	struct list_head	m_del_inodes;	/* inodes to reclaim */
 	mutex_t			m_ilock;	/* inode list mutex */
 	uint			m_ireclaims;	/* count of calls to reclaim*/