[XFS] merge Steve's sync changes over to 2.5

SGI Modid: 2.5.x-xfs:slinx:147932a

fs/xfs/Makefile

@@ -125,6 +125,7 @@ xfs-y				+= $(addprefix linux/, \
 				   xfs_iomap.o \
 				   xfs_iops.o \
 				   xfs_lrw.o \
+				   xfs_syncd.o \
 				   xfs_super.o \
 				   xfs_vfs.o \
 				   xfs_vnode.o)

fs/xfs/linux/xfs_globals.c

@@ -48,7 +48,7 @@ unsigned long xfs_physmem;
  * Tunable XFS parameters.  xfs_params is required even when CONFIG_SYSCTL=n,
  * other XFS code uses these values.
  */
-xfs_param_t xfs_params = { 0, 1, 0, 0, 0, 3 };
+xfs_param_t xfs_params = { 0, 1, 0, 0, 0, 3, 30 * HZ };
 
 /*
  * Global system credential structure.

fs/xfs/linux/xfs_iops.c

@@ -152,8 +152,6 @@ linvfs_mknod(
 			ip->i_rdev = to_kdev_t(rdev);
 		validate_fields(dir);
 		d_instantiate(dentry, ip);
-		mark_inode_dirty_sync(ip);
-		mark_inode_dirty_sync(dir);
 	}
 
 	if (!error && have_default_acl) {
@@ -240,7 +238,6 @@ linvfs_link(
 		VN_HOLD(vp);
 		validate_fields(ip);
 		d_instantiate(dentry, ip);
-		mark_inode_dirty_sync(ip);
 	}
 	return -error;
 }
@@ -261,8 +258,6 @@ linvfs_unlink(
 	if (!error) {
 		validate_fields(dir);	/* For size only */
 		validate_fields(inode);
-		mark_inode_dirty_sync(inode);
-		mark_inode_dirty_sync(dir);
 	}
 
 	return -error;
@@ -296,8 +291,6 @@ linvfs_symlink(
 		d_instantiate(dentry, ip);
 		validate_fields(dir);
 		validate_fields(ip); /* size needs update */
-		mark_inode_dirty_sync(ip);
-		mark_inode_dirty_sync(dir);
 	}
 	return -error;
 }
@@ -315,8 +308,6 @@ linvfs_rmdir(
 	if (!error) {
 		validate_fields(inode);
 		validate_fields(dir);
-		mark_inode_dirty_sync(inode);
-		mark_inode_dirty_sync(dir);
 	}
 	return -error;
 }
@@ -346,7 +337,6 @@ linvfs_rename(
 	validate_fields(odir);
 	if (ndir != odir)
 		validate_fields(ndir);
-	mark_inode_dirty(ndir);
 	return 0;
 }
 
@@ -520,7 +510,6 @@ linvfs_setattr(
 
 	if (!error) {
 		vn_revalidate(vp);
-		mark_inode_dirty_sync(inode);
 	}
 	return error;
 }

fs/xfs/linux/xfs_super.c

@@ -331,9 +331,10 @@ destroy_inodecache( void )
 }
 
 /*
- * We do not actually write the inode here, just mark the
- * super block dirty so that sync_supers calls us and
- * forces the flush.
+ * Attempt to flush the inode, this will actually fail
+ * if the inode is pinned, but we dirty the inode again
+ * at the point when it is unpinned after a log write,
+ * since this is when the inode itself becomes flushable. 
  */
 STATIC void
 linvfs_write_inode(
@@ -348,8 +349,6 @@ linvfs_write_inode(
 		if (sync)
 			flags |= FLUSH_SYNC;
 		VOP_IFLUSH(vp, flags, error);
-		if (error == EAGAIN)
-			inode->i_sb->s_dirt = 1;
 	}
 }
 
@@ -376,9 +375,11 @@ linvfs_put_super(
 	vfs_t			*vfsp = LINVFS_GET_VFS(sb);
 	int			error;
 
+	linvfs_stop_syncd(vfsp);
 	VFS_SYNC(vfsp, SYNC_ATTR|SYNC_DELWRI, NULL, error);
-	if (error == 0)
+	if (error == 0) {
 		VFS_UNMOUNT(vfsp, 0, NULL, error);
+	}
 	if (error) {
 		printk("XFS unmount got error %d\n", error);
 		printk("%s: vfsp/0x%p left dangling!\n", __FUNCTION__, vfsp);
@@ -395,10 +396,13 @@ linvfs_write_super(
 	vfs_t			*vfsp = LINVFS_GET_VFS(sb);
 	int			error;
 
-	sb->s_dirt = 0;
-	if (sb->s_flags & MS_RDONLY)
+	if (sb->s_flags & MS_RDONLY) {
+		sb->s_dirt = 0; /* paranoia */
 		return;
-	VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
+	}
+	/* Push the log and superblock a little */
+	VFS_SYNC(vfsp, SYNC_FSDATA, NULL, error);
+	sb->s_dirt = 0;
 }
 
 STATIC int
@@ -652,7 +656,8 @@ linvfs_fill_super(
 		goto fail_vnrele;
 	if (is_bad_inode(sb->s_root->d_inode))
 		goto fail_vnrele;
-
+	if (linvfs_start_syncd(vfsp))
+		goto fail_vnrele;
 	vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 
 	kmem_free(args, sizeof(*args));

fs/xfs/linux/xfs_super.h

@@ -104,5 +104,7 @@ extern void xfs_free_buftarg(struct pb_target *);
 
 extern void xfs_setsize_buftarg(struct pb_target *, unsigned int, unsigned int);
 extern unsigned int xfs_getsize_buftarg(struct pb_target *);
+extern int linvfs_start_syncd(vfs_t *);
+extern void linvfs_stop_syncd(vfs_t *);
 
 #endif	/* __XFS_SUPER_H__ */

fs/xfs/linux/xfs_syncd.c

+/*
+ * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * Further, this software is distributed without any warranty that it is
+ * free of the rightful claim of any third person regarding infringement
+ * or the like.  Any license provided herein, whether implied or
+ * otherwise, applies only to this software file.  Patent licenses, if
+ * any, provided herein do not apply to combinations of this program with
+ * other software, or any other product whatsoever.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write the Free Software Foundation, Inc., 59
+ * Temple Place - Suite 330, Boston MA 02111-1307, USA.
+ *
+ * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
+ * Mountain View, CA  94043, or:
+ *
+ * http://www.sgi.com
+ *
+ * For further information regarding this notice, see:
+ *
+ * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
+ */
+
+#include <xfs.h>
+
+static void sync_timeout(unsigned long __data)
+{
+	struct task_struct * p = (struct task_struct *) __data;
+
+	wake_up_process(p);
+}
+
+#define SYNCD_FLAGS	(SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR)
+
+int syncd(void *arg)
+{
+	vfs_t			*vfsp = (vfs_t *) arg;
+	int			error;
+	struct timer_list	timer;
+
+	daemonize("xfs_syncd");
+
+	vfsp->vfs_sync_task = current;
+	wmb();
+	wake_up(&vfsp->vfs_wait_sync_task);
+
+	init_timer(&timer);
+	timer.data = (unsigned long)current;
+	timer.function = sync_timeout;
+
+	do {
+		mod_timer(&timer, jiffies + xfs_params.sync_interval);
+		interruptible_sleep_on(&vfsp->vfs_sync);
+
+		if (!(vfsp->vfs_flag & VFS_RDONLY))
+			VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
+	} while (!(vfsp->vfs_flag & VFS_UMOUNT));
+
+	del_timer_sync(&timer);
+	vfsp->vfs_sync_task = NULL;
+	wmb();
+	wake_up(&vfsp->vfs_wait_sync_task);
+	return 0;
+}
+
+int
+linvfs_start_syncd(vfs_t *vfsp)
+{
+	int pid;
+
+	pid = kernel_thread(syncd, (void *) vfsp,
+			CLONE_VM | CLONE_FS | CLONE_FILES);
+	if (pid < 0)
+		return pid;
+	wait_event(vfsp->vfs_wait_sync_task, vfsp->vfs_sync_task);
+	return 0;
+}
+
+void
+linvfs_stop_syncd(vfs_t *vfsp)
+{
+	vfsp->vfs_flag |= VFS_UMOUNT;
+	wmb();
+
+	wake_up(&vfsp->vfs_sync);
+	wait_event(vfsp->vfs_wait_sync_task, !vfsp->vfs_sync_task);
+}

fs/xfs/linux/xfs_sysctl.c

@@ -36,8 +36,8 @@
 #include <linux/proc_fs.h>
 
 
-STATIC ulong xfs_min[XFS_PARAM] = { 0, 0, 0, 0, 0,   0 };
-STATIC ulong xfs_max[XFS_PARAM] = { 1, 1, 1, 1, 127, 3 };
+STATIC ulong xfs_min[XFS_PARAM] = { 0, 0, 0, 0, 0,   0, HZ };
+STATIC ulong xfs_max[XFS_PARAM] = { 1, 1, 1, 1, 127, 3, HZ * 60 };
 
 static struct ctl_table_header *xfs_table_header;
 
@@ -92,6 +92,10 @@ STATIC ctl_table xfs_table[] = {
 	sizeof(ulong), 0644, NULL, &proc_doulongvec_minmax,
 	&sysctl_intvec, NULL, &xfs_min[5], &xfs_max[5]},
 
+	{XFS_SYNC_INTERVAL, "sync_interval", &xfs_params.sync_interval,
+	sizeof(ulong), 0644, NULL, &proc_doulongvec_minmax,
+	&sysctl_intvec, NULL, &xfs_min[6], &xfs_max[6]},
+
 	{0}
 };
 

fs/xfs/linux/xfs_sysctl.h

@@ -49,6 +49,7 @@ typedef struct xfs_param {
 	ulong	symlink_mode;	/* Symlink creat mode affected by umask. */
 	ulong	panic_mask;	/* bitmask to specify panics on errors.  */
 	ulong	error_level;	/* Degree of reporting for internal probs*/
+	ulong	sync_interval;	/* time between sync calls		 */
 } xfs_param_t;
 
 /*
@@ -73,6 +74,7 @@ enum {
 	XFS_SYMLINK_MODE = 4,
 	XFS_PANIC_MASK = 5,
 	XFS_ERRLEVEL = 6,
+	XFS_SYNC_INTERVAL = 7,
 };
 
 extern xfs_param_t	xfs_params;

fs/xfs/linux/xfs_vfs.c

@@ -238,6 +238,8 @@ vfs_allocate( void )
 
 	vfsp = kmem_zalloc(sizeof(vfs_t), KM_SLEEP);
 	bhv_head_init(VFS_BHVHEAD(vfsp), "vfs");
+	init_waitqueue_head(&vfsp->vfs_wait_sync_task);
+	init_waitqueue_head(&vfsp->vfs_sync);
 	return vfsp;
 }
 

fs/xfs/linux/xfs_vfs.h

@@ -48,6 +48,9 @@ typedef struct vfs {
 	fsid_t			*vfs_altfsid;	/* An ID fixed for life of FS */
 	bhv_head_t		vfs_bh;		/* head of vfs behavior chain */
 	struct super_block	*vfs_super;	/* Linux superblock structure */
+	struct task_struct	*vfs_sync_task;
+	wait_queue_head_t	vfs_sync;
+	wait_queue_head_t	vfs_wait_sync_task;
 } vfs_t;
 
 #define vfs_fbhv		vfs_bh.bh_first	/* 1st on vfs behavior chain */
@@ -78,7 +81,8 @@ typedef enum {
 #define VFS_RDONLY		0x0001	/* read-only vfs */
 #define VFS_GRPID		0x0002	/* group-ID assigned from directory */
 #define VFS_DMI			0x0004	/* filesystem has the DMI enabled */
-#define VFS_END			0x0004	/* max flag */
+#define VFS_UMOUNT		0x0008	/* unmount in progress */
+#define VFS_END			0x0008	/* max flag */
 
 #define SYNC_ATTR		0x0001	/* sync attributes */
 #define SYNC_CLOSE		0x0002	/* close file system down */
@@ -87,6 +91,7 @@ typedef enum {
 #define SYNC_FSDATA		0x0020	/* flush fs data (e.g. superblocks) */
 #define SYNC_BDFLUSH		0x0010	/* BDFLUSH is calling -- don't block */
 
+
 typedef int	(*vfs_mount_t)(bhv_desc_t *,
 				struct xfs_mount_args *, struct cred *);
 typedef int	(*vfs_parseargs_t)(bhv_desc_t *, char *,

fs/xfs/linux/xfs_vnode.h

@@ -562,8 +562,7 @@ static __inline__ void vn_flagclr(struct vnode *vp, uint flag)
 	(!list_empty(&(LINVFS_GET_IP(vp)->i_mapping->i_mmap_shared))))
 #define VN_CACHED(vp)	(LINVFS_GET_IP(vp)->i_mapping->nrpages)
 #define VN_DIRTY(vp)	(!list_empty(&(LINVFS_GET_IP(vp)->i_mapping->dirty_pages)))
-#define VMODIFY(vp)	{ VN_FLAGSET(vp, VMODIFIED); \
-			mark_inode_dirty(LINVFS_GET_IP(vp)); }
+#define VMODIFY(vp)	VN_FLAGSET(vp, VMODIFIED)
 #define VUNMODIFY(vp)	VN_FLAGCLR(vp, VMODIFIED)
 
 /*

fs/xfs/xfs_iget.c

@@ -252,6 +252,11 @@ xfs_iget_core(
 			if (newnode) {
 				xfs_iocore_inode_reinit(ip);
 			}
+
+			XFS_MOUNT_ILOCK(mp);
+			list_del_init(&ip->i_reclaim);
+			XFS_MOUNT_IUNLOCK(mp);
+
 			vn_trace_exit(vp, "xfs_iget.found",
 						(inst_t *)__return_address);
 			goto return_ip;
@@ -467,8 +472,10 @@ xfs_iget(
 			}
 
 			bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops);
-			if (bdp == NULL)
+			if (bdp == NULL) {
+				XFS_STATS_INC(xfsstats.xs_ig_dup);
 				goto inode_allocate;
+			}
 			ip = XFS_BHVTOI(bdp);
 			if (lock_flags != 0)
 				xfs_ilock(ip, lock_flags);
@@ -720,6 +727,9 @@ xfs_iextract(
 		}
 	}
 
+	/* Deal with the deleted inodes list */
+	list_del_init(&ip->i_reclaim);
+
 	mp->m_ireclaims++;
 	XFS_MOUNT_IUNLOCK(mp);
 }

fs/xfs/xfs_inode.c

@@ -976,6 +976,8 @@ xfs_iread(
 			XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
 	}
 
+	INIT_LIST_HEAD(&ip->i_reclaim);
+
 	/*
 	 * The inode format changed when we moved the link count and
 	 * made it 32 bits long.  If this is an old format inode,
@@ -2625,6 +2627,15 @@ xfs_iunpin(
 	ASSERT(atomic_read(&ip->i_pincount) > 0);
 
 	if (atomic_dec_and_test(&ip->i_pincount)) {
+		vnode_t	*vp = XFS_ITOV_NULL(ip);
+
+		/* make sync come back and flush this inode */
+		if (vp) {
+			struct inode	*inode = LINVFS_GET_IP(vp);
+
+			mark_inode_dirty_sync(inode);
+		}
+
 		wake_up(&ip->i_ipin_wait);
 	}
 }
@@ -3640,6 +3651,8 @@ xfs_ichgtime(xfs_inode_t *ip,
 	 */
 	SYNCHRONIZE();
 	ip->i_update_core = 1;
+	if (!(inode->i_state & I_LOCK))
+		mark_inode_dirty(inode);
 }
 
 #ifdef XFS_ILOCK_TRACE

fs/xfs/xfs_inode.h

@@ -243,6 +243,7 @@ typedef struct xfs_inode {
 	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*/
 	struct xfs_dquot	*i_udquot;	/* user dquot */
 	struct xfs_dquot	*i_gdquot;	/* group dquot */
@@ -477,7 +478,7 @@ void		xfs_iunlock_map_shared(xfs_inode_t *, uint);
 void		xfs_ifunlock(xfs_inode_t *);
 void		xfs_ireclaim(xfs_inode_t *);
 int		xfs_finish_reclaim(xfs_inode_t *, int, int);
-int		xfs_finish_reclaim_all(struct xfs_mount *);
+int		xfs_finish_reclaim_all(struct xfs_mount *, int);
 
 /*
  * xfs_inode.c prototypes.

fs/xfs/xfs_mount.c

@@ -546,6 +546,7 @@ xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
 	mp->m_blockmask = sbp->sb_blocksize - 1;
 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
 	mp->m_blockwmask = mp->m_blockwsize - 1;
+	INIT_LIST_HEAD(&mp->m_del_inodes);
 
 
 	if (XFS_SB_VERSION_HASLOGV2(sbp)) {
@@ -601,7 +602,6 @@ xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
 					sbp->sb_inopblock);
 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 }
-
 /*
  * xfs_mountfs
  *

fs/xfs/xfs_mount.h

@@ -296,6 +296,7 @@ typedef struct xfs_mount {
 	int			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 */
 	uint			m_ireclaims;	/* count of calls to reclaim*/
 	uint			m_readio_log;	/* min read size log bytes */

fs/xfs/xfs_vfsops.c

@@ -620,7 +620,7 @@ xfs_mntupdate(
 
 	if (*flags & MS_RDONLY) {
 		pagebuf_delwri_flush(mp->m_ddev_targp, 0, NULL);
-		xfs_finish_reclaim_all(mp);
+		xfs_finish_reclaim_all(mp, 0);
 
 		do {
 			VFS_SYNC(vfsp, SYNC_ATTR|SYNC_WAIT, NULL, error);
@@ -849,19 +849,14 @@ 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 xys_sync.  In the interests of not
+ * interface as explained above under xfs_sync.  In the interests of not
  * changing interfaces within the 6.5 family, additional internallly-
  * required functions are specified within a separate xflags parameter,
  * only available by calling this routine.
  *
- * xflags:
- *	XFS_XSYNC_RELOC - Sync for relocation.  Don't try to get behavior
- *                        locks as this will cause you to hang.  Not all
- *                        combinations of flags are necessarily supported
- *                        when this is specified.
  */
-int
-xfs_syncsub(
+STATIC int
+xfs_sync_inodes(
 	xfs_mount_t	*mp,
 	int		flags,
 	int             xflags,
@@ -877,12 +872,10 @@ xfs_syncsub(
 	uint64_t	fflag;
 	uint		lock_flags;
 	uint		base_lock_flags;
-	uint		log_flags;
 	boolean_t	mount_locked;
 	boolean_t	vnode_refed;
 	int		preempt;
 	xfs_dinode_t	*dip;
-	xfs_buf_log_item_t	*bip;
 	xfs_iptr_t	*ipointer;
 #ifdef DEBUG
 	boolean_t	ipointer_in = B_FALSE;
@@ -961,16 +954,6 @@ xfs_syncsub(
 		base_lock_flags |= XFS_IOLOCK_SHARED;
 	}
 
-	/*
-	 * Sync out the log.  This ensures that the log is periodically
-	 * flushed even if there is not enough activity to fill it up.
-	 */
-	if (flags & SYNC_WAIT) {
-		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
-	} else {
-		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
-	}
-
 	XFS_MOUNT_ILOCK(mp);
 
 	ip = mp->m_inodes;
@@ -1016,27 +999,23 @@ xfs_syncsub(
 				ip = ip->i_mnext;
 				continue;
 			}
-			if ((ip->i_update_core == 0) &&
-			    ((ip->i_itemp == NULL) ||
-			    !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL))) {
-				if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
-					ip = ip->i_mnext;
-				} else if ((xfs_ipincount(ip) == 0) &&
+			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);
+				IPOINTER_INSERT(ip, mp);
 
-					xfs_finish_reclaim(ip, 1,
-						XFS_IFLUSH_DELWRI_ELSE_SYNC);
+				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;
-				}
-				continue;
+				XFS_MOUNT_ILOCK(mp);
+				mount_locked = B_TRUE;
+				IPOINTER_REMOVE(ip, mp);
+			} else {
+				xfs_iunlock(ip, XFS_ILOCK_EXCL);
+				ip = ip->i_mnext;
 			}
+			continue;
 		}
 
 		if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
@@ -1148,21 +1127,9 @@ xfs_syncsub(
 			xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 			if (XFS_FORCED_SHUTDOWN(mp)) {
-				if (xflags & XFS_XSYNC_RELOC) {
-					fs_tosspages(XFS_ITOBHV(ip), 0, -1,
-						     FI_REMAPF);
-				}
-				else {
-					VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
-				}
+				VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
 			} else {
-				if (xflags & XFS_XSYNC_RELOC) {
-					fs_flushinval_pages(XFS_ITOBHV(ip),
-							    0, -1, FI_REMAPF);
-				}
-				else {
-					VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
-				}
+				VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
 			}
 
 			xfs_ilock(ip, XFS_ILOCK_SHARED);
@@ -1418,16 +1385,55 @@ xfs_syncsub(
 
 	ASSERT(ipointer_in == B_FALSE);
 
+	kmem_free(ipointer, sizeof(xfs_iptr_t));
+	return XFS_ERROR(last_error);
+}
+
+/*
+ * 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.  In the interests of not
+ * changing interfaces within the 6.5 family, additional internallly-
+ * required functions are specified within a separate xflags parameter,
+ * only available by calling this routine.
+ *
+ */
+int
+xfs_syncsub(
+	xfs_mount_t	*mp,
+	int		flags,
+	int             xflags,
+	int             *bypassed)
+{
+	int		error = 0;
+	int		last_error = 0;
+	uint		log_flags = XFS_LOG_FORCE;
+	xfs_buf_t	*bp;
+	xfs_buf_log_item_t	*bip;
+
+	/*
+	 * Sync out the log.  This ensures that the log is periodically
+	 * flushed even if there is not enough activity to fill it up.
+	 */
+	if (flags & SYNC_WAIT)
+		log_flags |= XFS_LOG_SYNC;
+
+	xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
+
+	if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
+		if (flags & SYNC_BDFLUSH)
+			xfs_finish_reclaim_all(mp, 1);
+		else
+			error = xfs_sync_inodes(mp, flags, xflags, bypassed);
+	}
+
 	/*
 	 * Flushing out dirty data above probably generated more
 	 * log activity, so if this isn't vfs_sync() then flush
-	 * the log again.  If SYNC_WAIT is set then do it synchronously.
+	 * the log again.
 	 */
-	if (!(flags & SYNC_BDFLUSH)) {
-		log_flags = XFS_LOG_FORCE;
-		if (flags & SYNC_WAIT) {
-			log_flags |= XFS_LOG_SYNC;
-		}
+	if (flags & SYNC_DELWRI) {
 		xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
 	}
 
@@ -1463,11 +1469,10 @@ xfs_syncsub(
 			 * that point so it can become pinned in between
 			 * there and here.
 			 */
-			if (XFS_BUF_ISPINNED(bp)) {
-				xfs_log_force(mp, (xfs_lsn_t)0,
-					      XFS_LOG_FORCE);
-			}
-			XFS_BUF_BFLAGS(bp) |= fflag;
+			if (XFS_BUF_ISPINNED(bp))
+				xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
+			if (!(flags & SYNC_WAIT))
+				XFS_BUF_BFLAGS(bp) |= XFS_B_ASYNC;
 			error = xfs_bwrite(mp, bp);
 		}
 		if (error) {
@@ -1478,9 +1483,9 @@ xfs_syncsub(
 	/*
 	 * Now check to see if the log needs a "dummy" transaction.
 	 */
-
 	if (xfs_log_need_covered(mp)) {
 		xfs_trans_t *tp;
+		xfs_inode_t *ip;
 
 		/*
 		 * Put a dummy transaction in the log to tell
@@ -1491,7 +1496,6 @@ xfs_syncsub(
 				XFS_ICHANGE_LOG_RES(mp),
 				0, 0, 0)))  {
 			xfs_trans_cancel(tp, 0);
-			kmem_free(ipointer, sizeof(xfs_iptr_t));
 			return error;
 		}
 
@@ -1503,6 +1507,7 @@ xfs_syncsub(
 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 		error = xfs_trans_commit(tp, 0, NULL);
 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
+		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
 	}
 
 	/*
@@ -1516,7 +1521,6 @@ xfs_syncsub(
 		}
 	}
 
-	kmem_free(ipointer, sizeof(xfs_iptr_t));
 	return XFS_ERROR(last_error);
 }
 

fs/xfs/xfs_vnodeops.c

@@ -3786,27 +3786,30 @@ xfs_inode_flush(
 					flush_flags = XFS_IFLUSH_SYNC;
 				else
 #endif
-					flush_flags = XFS_IFLUSH_DELWRI;
+				flush_flags = XFS_IFLUSH_DELWRI_ELSE_ASYNC;
 
 				xfs_ifunlock(ip);
 				xfs_iunlock(ip, XFS_ILOCK_SHARED);
 				error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0);
 				if (error)
-					goto eagain;
+					return error;
 				xfs_buf_relse(bp);
 
 				if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED) == 0)
-					goto eagain;
+					return EAGAIN;
+
+				if (xfs_ipincount(ip) ||
+				    !xfs_iflock_nowait(ip)) {
+					xfs_iunlock(ip, XFS_ILOCK_SHARED);
+					return EAGAIN;
+				}
 
-				if ((xfs_ipincount(ip) == 0) &&
-				     xfs_iflock_nowait(ip))
-					error = xfs_iflush(ip, flush_flags);
+				error = xfs_iflush(ip, flush_flags);
 			} else {
 				error = EAGAIN;
 			}
 			xfs_iunlock(ip, XFS_ILOCK_SHARED);
 		} else {
-eagain:
 			error = EAGAIN;
 		}
 	}
@@ -3934,6 +3937,8 @@ xfs_reclaim(
 		/* Protect sync from us */
 		XFS_MOUNT_ILOCK(mp);
 		vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
+		list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
+
 		XFS_MOUNT_IUNLOCK(mp);
 	}
 	return 0;
@@ -4010,40 +4015,33 @@ xfs_finish_reclaim(
 }
 
 int
-xfs_finish_reclaim_all(xfs_mount_t *mp)
+xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
 {
 	int		purged;
+	struct list_head	*curr, *next;
 	xfs_inode_t	*ip;
-	vnode_t		*vp;
 	int		done = 0;
 
 	while (!done) {
 		purged = 0;
 		XFS_MOUNT_ILOCK(mp);
-		ip = mp->m_inodes;
-		if (ip == NULL) {
+		list_for_each_safe(curr, next, &mp->m_del_inodes) {
+			ip = list_entry(curr, xfs_inode_t, i_reclaim);
+			if (noblock) {
+				if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
+					continue;
+				if (xfs_ipincount(ip) ||
+				    !xfs_iflock_nowait(ip)) {
+					xfs_iunlock(ip, XFS_ILOCK_EXCL);
+					continue;
+				}
+			}
+			XFS_MOUNT_IUNLOCK(mp);
+			xfs_finish_reclaim(ip, noblock,
+				XFS_IFLUSH_DELWRI_ELSE_ASYNC);
+			purged = 1;
 			break;
 		}
-		do {
-			/* Make sure we skip markers inserted by sync */
-			if (ip->i_mount == NULL) {
-				ip = ip->i_mnext;
-				continue;
-			}
-
-			/*
-			 * It's up to our caller to purge the root
-			 * and quota vnodes later.
-			 */
-			vp = XFS_ITOV_NULL(ip);
-
-			if (!vp) {
-				XFS_MOUNT_IUNLOCK(mp);
-				xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
-				purged = 1;
-				break;
-			}
-		} while (ip != mp->m_inodes);
 
 		done = !purged;
 	}