Merge

fs/xfs/Makefile

@@ -132,7 +132,6 @@ xfs-y				+= $(addprefix linux/, \
 # Objects in support/
 xfs-y				+= $(addprefix support/, \
 				   debug.o \
-				   kmem.o \
 				   ktrace.o \
 				   move.o \
 				   mrlock.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_lrw.c

@@ -887,29 +887,23 @@ xfsbdstrat(
 	return (xfs_bioerror_relse(bp));
 }
 
-
-void
-XFS_bflush(xfs_buftarg_t *target)
-{
-	pagebuf_delwri_flush(target, PBDF_WAIT, NULL);
-}
-
 /*
- * If the underlying (log or data) device is readonly, there are some
+ * If the underlying (data/log/rt) device is readonly, there are some
  * operations that cannot proceed.
  */
 int
-xfs_dev_is_read_only(xfs_mount_t *mp, char *message)
+xfs_dev_is_read_only(
+	xfs_mount_t		*mp,
+	char			*message)
 {
-	if (bdev_read_only(mp->m_ddev_targp->pbr_bdev) ||
-	    bdev_read_only(mp->m_logdev_targp->pbr_bdev) ||
-	   (mp->m_rtdev_targp && bdev_read_only(mp->m_rtdev_targp->pbr_bdev))) {
+	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
+	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
+	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
 		cmn_err(CE_NOTE,
 			"XFS: %s required on read-only device.", message);
 		cmn_err(CE_NOTE,
 			"XFS: write access unavailable, cannot proceed.");
 		return EROFS;
 	}
-
 	return 0;
 }

fs/xfs/linux/xfs_super.c

@@ -217,13 +217,27 @@ xfs_blkdev_put(
 }
 
 void
-xfs_free_buftarg(
+xfs_flush_buftarg(
 	xfs_buftarg_t		*btp)
 {
 	pagebuf_delwri_flush(btp, PBDF_WAIT, NULL);
+}
+
+void
+xfs_free_buftarg(
+	xfs_buftarg_t		*btp)
+{
+	xfs_flush_buftarg(btp);
 	kmem_free(btp, sizeof(*btp));
 }
 
+int
+xfs_readonly_buftarg(
+	xfs_buftarg_t		*btp)
+{
+	return bdev_read_only(btp->pbr_bdev);
+}
+
 void
 xfs_relse_buftarg(
 	xfs_buftarg_t		*btp)
@@ -331,9 +345,10 @@ 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 +363,6 @@ linvfs_write_inode(
 		if (sync)
 			flags |= FLUSH_SYNC;
 		VOP_IFLUSH(vp, flags, error);
-		if (error == EAGAIN)
-			inode->i_sb->s_dirt = 1;
 	}
 }
 
@@ -369,6 +382,61 @@ linvfs_clear_inode(
 	}
 }
 
+
+#define SYNCD_FLAGS	(SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR)
+
+STATIC int
+syncd(void *arg)
+{
+	vfs_t			*vfsp = (vfs_t *) arg;
+	int			error;
+
+	daemonize("xfs_syncd");
+
+	vfsp->vfs_sync_task = current;
+	wmb();
+	wake_up(&vfsp->vfs_wait_sync_task);
+
+	for (;;) {
+		set_current_state(TASK_INTERRUPTIBLE);
+		schedule_timeout(xfs_params.sync_interval);
+		if (vfsp->vfs_flag & VFS_UMOUNT)
+			break;
+		if (vfsp->vfs_flag & VFS_RDONLY)
+			continue;
+		VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
+	}
+
+	vfsp->vfs_sync_task = NULL;
+	wmb();
+	wake_up(&vfsp->vfs_wait_sync_task);
+
+	return 0;
+}
+
+STATIC 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;
+}
+
+STATIC void
+linvfs_stop_syncd(vfs_t *vfsp)
+{
+	vfsp->vfs_flag |= VFS_UMOUNT;
+	wmb();
+
+	wake_up_process(vfsp->vfs_sync_task);
+	wait_event(vfsp->vfs_wait_sync_task, !vfsp->vfs_sync_task);
+}
+
 STATIC void
 linvfs_put_super(
 	struct super_block	*sb)
@@ -376,8 +444,9 @@ 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)
 		VFS_UNMOUNT(vfsp, 0, NULL, error);
 	if (error) {
 		printk("XFS unmount got error %d\n", error);
@@ -395,10 +464,13 @@ 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
@@ -424,12 +496,8 @@ linvfs_remount(
 	int			error;
 
 	VFS_PARSEARGS(vfsp, options, args, 1, error);
-	if (error)
-		goto out;
-
-	VFS_MNTUPDATE(vfsp, flags, args, error);
-
-out:
+	if (!error)
+		VFS_MNTUPDATE(vfsp, flags, args, error);
 	kmem_free(args, sizeof(*args));
 	return error;
 }
@@ -438,11 +506,10 @@ STATIC void
 linvfs_freeze_fs(
 	struct super_block	*sb)
 {
-	vfs_t			*vfsp;
+	vfs_t			*vfsp = LINVFS_GET_VFS(sb);
 	vnode_t			*vp;
 	int			error;
 
-	vfsp = LINVFS_GET_VFS(sb);
 	if (sb->s_flags & MS_RDONLY)
 		return;
 	VFS_ROOT(vfsp, &vp, error);
@@ -454,11 +521,10 @@ STATIC void
 linvfs_unfreeze_fs(
 	struct super_block	*sb)
 {
-	vfs_t			*vfsp;
+	vfs_t			*vfsp = LINVFS_GET_VFS(sb);
 	vnode_t			*vp;
 	int			error;
 
-	vfsp = LINVFS_GET_VFS(sb);
 	VFS_ROOT(vfsp, &vp, error);
 	VOP_IOCTL(vp, LINVFS_GET_IP(vp), NULL, XFS_IOC_THAW, 0, error);
 	VN_RELE(vp);
@@ -652,7 +718,8 @@ 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

@@ -101,7 +101,8 @@ extern void xfs_blkdev_put(struct block_device *);
 extern struct pb_target *xfs_alloc_buftarg(struct block_device *);
 extern void xfs_relse_buftarg(struct pb_target *);
 extern void xfs_free_buftarg(struct pb_target *);
-
+extern void xfs_flush_buftarg(struct pb_target *);
+extern int xfs_readonly_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 *);
 

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>
+
+#define SYNCD_FLAGS	(SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR)
+
+int syncd(void *arg)
+{
+	vfs_t			*vfsp = (vfs_t *) arg;
+	int			error;
+
+	daemonize("xfs_syncd");
+
+	vfsp->vfs_sync_task = current;
+	wmb();
+	wake_up(&vfsp->vfs_wait_sync_task);
+
+	for (;;) {
+		set_current_state(TASK_INTERRUPTIBLE);
+		schedule_timeout(xfs_params.sync_interval);
+		if (vfsp->vfs_flag & VFS_UMOUNT)
+			break;
+		if (vfsp->vfs_flag & VFS_RDONLY);
+			continue;
+		VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
+	}
+
+	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_process(vfsp->vfs_sync_task);
+	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 +238,7 @@ 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);
 	return vfsp;
 }
 

fs/xfs/linux/xfs_vfs.h

@@ -48,6 +48,8 @@ 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_wait_sync_task;
 } vfs_t;
 
 #define vfs_fbhv		vfs_bh.bh_first	/* 1st on vfs behavior chain */
@@ -78,7 +80,8 @@ 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 +90,7 @@ 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/pagebuf/page_buf.c

@@ -68,7 +68,7 @@
 #define BN_ALIGN_MASK	((1 << (PAGE_CACHE_SHIFT - BBSHIFT)) - 1)
 
 #ifndef GFP_READAHEAD
-#define GFP_READAHEAD	__GFP_NOWARN
+#define GFP_READAHEAD	(__GFP_NOWARN|__GFP_NORETRY)
 #endif
 
 /*
@@ -76,11 +76,11 @@
  */
 
 #ifdef PAGEBUF_TRACE
-static	spinlock_t		pb_trace_lock = SPIN_LOCK_UNLOCKED;
+static spinlock_t		pb_trace_lock = SPIN_LOCK_UNLOCKED;
 struct pagebuf_trace_buf	pb_trace;
 EXPORT_SYMBOL(pb_trace);
 EXPORT_SYMBOL(pb_trace_func);
-#define CIRC_INC(i)     (((i) + 1) & (PB_TRACE_BUFSIZE - 1))
+#define CIRC_INC(i)	(((i) + 1) & (PB_TRACE_BUFSIZE - 1))
 
 void
 pb_trace_func(
@@ -181,7 +181,7 @@ _bhash(
 	 * dev_t is 16 bits, loff_t is always 64 bits
 	 */
 	base ^= dev;
-	for (bit = hval = 0; base != 0 && bit < sizeof(base) * 8; bit += NBITS) {
+	for (bit = hval = 0; base && bit < sizeof(base) * 8; bit += NBITS) {
 		hval ^= (int)base & (NHASH-1);
 		base >>= NBITS;
 	}
@@ -189,18 +189,18 @@ _bhash(
 }
 
 /*
- * Mapping of multi-page buffers into contingous virtual space
+ * Mapping of multi-page buffers into contiguous virtual space
  */
 
 STATIC void *pagebuf_mapout_locked(page_buf_t *);
 
-STATIC  spinlock_t              as_lock = SPIN_LOCK_UNLOCKED;
 typedef struct a_list {
-	void	*vm_addr;
+	void		*vm_addr;
 	struct a_list	*next;
 } a_list_t;
-STATIC  a_list_t	*as_free_head;
-STATIC  int		as_list_len;
+STATIC a_list_t		*as_free_head;
+STATIC int		as_list_len;
+STATIC spinlock_t	as_lock = SPIN_LOCK_UNLOCKED;
 
 
 /*
@@ -1897,13 +1897,6 @@ pagebuf_readstats(
 }
 #endif  /* CONFIG_PROC_FS */
 
-STATIC void
-pagebuf_shaker(void)
-{
-	pagebuf_daemon_wakeup(1);
-}
-
-
 /*
  *	Initialization and Termination
  */
@@ -1943,7 +1936,6 @@ pagebuf_init(void)
 #endif
 
 	pagebuf_daemon_start();
-	kmem_shake_register(pagebuf_shaker);
 	return 0;
 }
 
@@ -1959,7 +1951,6 @@ pagebuf_terminate(void)
 	pagebuf_daemon_stop();
 
 	kmem_cache_destroy(pagebuf_cache);
-	kmem_shake_deregister(pagebuf_shaker);
 
 	unregister_sysctl_table(pagebuf_table_header);
 #ifdef  CONFIG_PROC_FS

fs/xfs/quota/xfs_qm.c

@@ -88,7 +88,7 @@ STATIC void	xfs_qm_list_destroy(xfs_dqlist_t *);
 STATIC int	xfs_qm_quotacheck(xfs_mount_t *);
 
 STATIC int	xfs_qm_init_quotainos(xfs_mount_t *);
-STATIC void	xfs_qm_shake(void);
+STATIC int	xfs_qm_shake(int, unsigned int);
 
 #ifdef DEBUG
 extern mutex_t	qcheck_lock;
@@ -112,6 +112,8 @@ extern mutex_t	qcheck_lock;
 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
 #endif
 
+struct shrinker *xfs_qm_shrinker;
+
 /*
  * Initialize the XQM structure.
  * Note that there is not one quota manager per file system.
@@ -161,7 +163,7 @@ xfs_Gqm_init(void)
 	} else
 		xqm->qm_dqzone = qm_dqzone;
 
-	kmem_shake_register(xfs_qm_shake);
+	xfs_qm_shrinker = set_shrinker(DEFAULT_SEEKS, xfs_qm_shake);
 
 	/*
 	 * The t_dqinfo portion of transactions.
@@ -193,7 +195,8 @@ xfs_qm_destroy(
 
 	ASSERT(xqm != NULL);
 	ASSERT(xqm->qm_nrefs == 0);
-	kmem_shake_deregister(xfs_qm_shake);
+
+	remove_shrinker(xfs_qm_shrinker);
 	hsize = xqm->qm_dqhashmask + 1;
 	for (i = 0; i < hsize; i++) {
 		xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
@@ -2088,7 +2091,7 @@ xfs_qm_shake_freelist(
 			xfs_dqunlock(dqp);
 			xfs_qm_freelist_unlock(xfs_Gqm);
 			if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
-				return (nreclaimed != howmany);
+				goto out;
 			XQM_STATS_INC(xqmstats.xs_qm_dqwants);
 			goto tryagain;
 		}
@@ -2163,7 +2166,7 @@ xfs_qm_shake_freelist(
 			XFS_DQ_HASH_UNLOCK(hash);
 			xfs_qm_freelist_unlock(xfs_Gqm);
 			if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
-				return (nreclaimed != howmany);
+				goto out;
 			goto tryagain;
 		}
 		xfs_dqtrace_entry(dqp, "DQSHAKE: UNLINKING");
@@ -2188,7 +2191,8 @@ xfs_qm_shake_freelist(
 		dqp = nextdqp;
 	}
 	xfs_qm_freelist_unlock(xfs_Gqm);
-	return (nreclaimed != howmany);
+ out:
+	return nreclaimed;
 }
 
 
@@ -2197,13 +2201,15 @@ xfs_qm_shake_freelist(
  * running low.
  */
 /* ARGSUSED */
-STATIC void
-xfs_qm_shake(void)
+STATIC int
+xfs_qm_shake(int nr_to_scan, unsigned int gfp_mask)
 {
 	int	ndqused, nfree, n;
 
+	if (!(gfp_mask & __GFP_WAIT))
+		return 0;
 	if (!xfs_Gqm)
-		return;
+		return 0;
 
 	nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
 	/* incore dquots in all f/s's */
@@ -2212,12 +2218,12 @@ xfs_qm_shake(void)
 	ASSERT(ndqused >= 0);
 
 	if (nfree <= ndqused && nfree < ndquot)
-		return;
+		return 0;
 
 	ndqused *= xfs_Gqm->qm_dqfree_ratio;	/* target # of free dquots */
 	n = nfree - ndqused - ndquot;		/* # over target */
 
-	(void) xfs_qm_shake_freelist(MAX(nfree, n));
+	return xfs_qm_shake_freelist(MAX(nfree, n));
 }
 
 

fs/xfs/support/kmem.h

@@ -32,31 +32,118 @@
 #ifndef __XFS_SUPPORT_KMEM_H__
 #define __XFS_SUPPORT_KMEM_H__
 
+#include <linux/mm.h>
+#include <linux/highmem.h>
 #include <linux/slab.h>
+#include <linux/vmalloc.h>
 
 /*
- * memory management routines
+ * Cutoff point to use vmalloc instead of kmalloc.
  */
+#define MAX_SLAB_SIZE	0x10000
+
+/*
+ * XFS uses slightly different names for these due to the
+ * IRIX heritage.
+ */
+#define	kmem_zone	kmem_cache_s
+#define kmem_zone_t	kmem_cache_t
+
 #define KM_SLEEP	0x0001
 #define KM_NOSLEEP	0x0002
 #define KM_NOFS		0x0004
 
-#define	kmem_zone	kmem_cache_s
-#define kmem_zone_t	kmem_cache_t
 
-extern kmem_zone_t  *kmem_zone_init(int, char *);
-extern void	    *kmem_zone_zalloc(kmem_zone_t *, int);
-extern void	    *kmem_zone_alloc(kmem_zone_t *, int);
-extern void         kmem_zone_free(kmem_zone_t *, void *);
+/*
+ * XXX get rid of the unconditional  __GFP_NOFAIL by adding
+ * a KM_FAIL flag and using it where we're allowed to fail.
+ */
+static __inline unsigned int
+flag_convert(int flags)
+{
+#if DEBUG
+	if (unlikely(flags & ~(KM_SLEEP|KM_NOSLEEP|KM_NOFS))) {
+		printk(KERN_WARNING
+		    "XFS: memory allocation with wrong flags (%x)\n", flags);
+		BUG();
+	}
+#endif
+
+	if (flags & KM_NOSLEEP)
+		return GFP_ATOMIC;
+	/* If we're in a transaction, FS activity is not ok */
+	else if ((current->flags & PF_FSTRANS) || (flags & KM_NOFS))
+		return GFP_NOFS | __GFP_NOFAIL;
+	return GFP_KERNEL | __GFP_NOFAIL;
+}
+
+static __inline void *
+kmem_alloc(size_t size, int flags)
+{
+	if (unlikely(MAX_SLAB_SIZE < size))
+		/* Avoid doing filesystem sensitive stuff to get this */
+		return __vmalloc(size, flag_convert(flags), PAGE_KERNEL);
+	return kmalloc(size, flag_convert(flags));
+}
+
+static __inline void *
+kmem_zalloc(size_t size, int flags)
+{
+	void *ptr = kmem_alloc(size, flags);
+	if (likely(ptr != NULL))
+		memset(ptr, 0, size);
+	return ptr;
+}
+
+static __inline void
+kmem_free(void *ptr, size_t size)
+{
+	if (unlikely((unsigned long)ptr < VMALLOC_START ||
+		     (unsigned long)ptr >= VMALLOC_END))
+		kfree(ptr);
+	else
+		vfree(ptr);
+}
+
+static __inline void *
+kmem_realloc(void *ptr, size_t newsize, size_t oldsize, int flags)
+{
+	void *new = kmem_alloc(newsize, flags);
+
+	if (likely(ptr != NULL)) {
+		if (likely(new != NULL))
+			memcpy(new, ptr, min(oldsize, newsize));
+		kmem_free(ptr, oldsize);
+	}
+
+	return new;
+}
+
+static __inline kmem_zone_t *
+kmem_zone_init(int size, char *zone_name)
+{
+	return kmem_cache_create(zone_name, size, 0, 0, NULL, NULL);
+}
 
-extern void	    *kmem_alloc(size_t, int);
-extern void	    *kmem_realloc(void *, size_t, size_t, int);
-extern void	    *kmem_zalloc(size_t, int);
-extern void         kmem_free(void *, size_t);
+static __inline void *
+kmem_zone_alloc(kmem_zone_t *zone, int flags)
+{
+	return kmem_cache_alloc(zone, flag_convert(flags));
+}
 
-typedef void	    (*kmem_shake_func_t)(void);
+static __inline void *
+kmem_zone_zalloc(kmem_zone_t *zone, int flags)
+{
+	void *ptr = kmem_zone_alloc(zone, flags);
+	if (likely(ptr != NULL))
+		memset(ptr, 0, kmem_cache_size(zone));
+	return ptr;
+}
 
-extern void	    kmem_shake_register(kmem_shake_func_t);
-extern void	    kmem_shake_deregister(kmem_shake_func_t);
+static __inline void
+kmem_zone_free(kmem_zone_t *zone, void *ptr)
+{
+	kmem_cache_free(zone, ptr);
+}
 
 #endif /* __XFS_SUPPORT_KMEM_H__ */

fs/xfs/xfs_ag.h

 /*
- * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
+ * 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
@@ -185,9 +185,8 @@ xfs_agblock_t xfs_agfl_block(struct xfs_mount *mp);
 #endif
 #define XFS_AGFL_SIZE(mp)	((mp)->m_sb.sb_sectsize / sizeof(xfs_agblock_t))
 
-/* -- nathans TODO ... use of BBSIZE here - should be sector size -- */
 typedef struct xfs_agfl {
-	xfs_agblock_t	agfl_bno[BBSIZE/sizeof(xfs_agblock_t)];
+	xfs_agblock_t	agfl_bno[1];	/* actually XFS_AGFL_SIZE(mp) */
 } xfs_agfl_t;
 
 /*

fs/xfs/xfs_buf.h

 /*
- * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
+ * 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
@@ -283,7 +283,6 @@ static inline int	XFS_bwrite(page_buf_t *pb)
 	return error;
 }
 
-
 #define XFS_bdwrite(pb)		     \
 	    pagebuf_iostart(pb, PBF_DELWRI | PBF_ASYNC)
 
@@ -307,15 +306,15 @@ static inline int xfs_bdwrite(void *mp, page_buf_t *bp)
  * of its metadata.
  */
 
-extern void XFS_bflush(xfs_buftarg_t *);
-#define xfs_binval(buftarg) XFS_bflush(buftarg)
+#define xfs_binval(buftarg)	xfs_flush_buftarg(buftarg)
+
+#define XFS_bflush(buftarg)	xfs_flush_buftarg(buftarg)
 
 #define xfs_incore_relse(buftarg,delwri_only,wait)	\
 	xfs_relse_buftarg(buftarg)
 
 #define xfs_baread(target, rablkno, ralen)  \
-	pagebuf_readahead((target), (rablkno), \
-			  (ralen), PBF_DONT_BLOCK)
+	pagebuf_readahead((target), (rablkno), (ralen), PBF_DONT_BLOCK)
 
 #define XFS_getrbuf(sleep,mp)	\
 	pagebuf_get_empty((mp)->m_ddev_targp)

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

@@ -656,7 +656,9 @@ xfs_iformat_extents(
 	int		nex;
 	int		real_size;
 	int		size;
+#if ARCH_CONVERT != ARCH_NOCONVERT
 	int		i;
+#endif
 
 	ifp = XFS_IFORK_PTR(ip, whichfork);
 	nex = XFS_DFORK_NEXTENTS_ARCH(dip, whichfork, ARCH_CONVERT);
@@ -976,6 +978,8 @@ 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 +2629,15 @@ 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 +3653,8 @@ 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_log.c

 /*
- * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
+ * 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
@@ -413,19 +413,6 @@ xfs_log_release_iclog(xfs_mount_t *mp,
 	return 0;
 }
 
-/*
- * Initialize log manager data.  This routine is intended to be called when
- * a system boots up.  It is not a per filesystem initialization.
- *
- * As you can see, we currently do nothing.
- */
-int
-xfs_log_init(void)
-{
-	return( 0 );
-}
-
-
 /*
  *  1. Reserve an amount of on-disk log space and return a ticket corresponding
  *	to the reservation.
@@ -497,8 +484,6 @@ xfs_log_mount(xfs_mount_t	*mp,
 	      xfs_daddr_t	blk_offset,
 	      int		num_bblks)
 {
-	xlog_t *log;
-
 	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
 		cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname);
 	else {
@@ -508,7 +493,7 @@ xfs_log_mount(xfs_mount_t	*mp,
 		ASSERT(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY);
 	}
 
-	mp->m_log = log = xlog_alloc_log(mp, log_dev, blk_offset, num_bblks);
+	mp->m_log = xlog_alloc_log(mp, log_dev, blk_offset, num_bblks);
 
 #if defined(DEBUG) || defined(XLOG_NOLOG)
 	if (! xlog_debug) {
@@ -528,19 +513,19 @@ xfs_log_mount(xfs_mount_t	*mp,
 		if (readonly)
 			vfsp->vfs_flag &= ~VFS_RDONLY;
 
-		error = xlog_recover(log, readonly);
+		error = xlog_recover(mp->m_log, readonly);
 
 		if (readonly)
 			vfsp->vfs_flag |= VFS_RDONLY;
 		if (error) {
 			cmn_err(CE_WARN, "XFS: log mount/recovery failed");
-			xlog_unalloc_log(log);
+			xlog_unalloc_log(mp->m_log);
 			return error;
 		}
 	}
 
 	/* Normal transactions can now occur */
-	log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
+	mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
 
 	/* End mounting message in xfs_log_mount_finish */
 	return 0;
@@ -1191,28 +1176,39 @@ xlog_alloc_log(xfs_mount_t	*mp,
 	int			i;
 	int			iclogsize;
 
-	log = (void *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP);
+	log = (xlog_t *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP);
 
 	log->l_mp	   = mp;
 	log->l_dev	   = log_dev;
 	log->l_logsize     = BBTOB(num_bblks);
 	log->l_logBBstart  = blk_offset;
 	log->l_logBBsize   = num_bblks;
-	log->l_roundoff	   = 0;
 	log->l_covered_state = XLOG_STATE_COVER_IDLE;
 	log->l_flags	   |= XLOG_ACTIVE_RECOVERY;
 
 	log->l_prev_block  = -1;
 	ASSIGN_ANY_LSN(log->l_tail_lsn, 1, 0, ARCH_NOCONVERT);
-	/* log->l_tail_lsn    = 0x100000000LL; cycle = 1; current block = 0 */
+	/* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
 	log->l_last_sync_lsn = log->l_tail_lsn;
 	log->l_curr_cycle  = 1;	    /* 0 is bad since this is initial value */
-	log->l_curr_block  = 0;		/* filled in by xlog_recover */
-	log->l_grant_reserve_bytes = 0;
 	log->l_grant_reserve_cycle = 1;
-	log->l_grant_write_bytes = 0;
 	log->l_grant_write_cycle = 1;
-	log->l_quotaoffs_flag = 0;      /* XFS_LI_QUOTAOFF logitems */
+
+	if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
+		if (mp->m_sb.sb_logsunit <= 1) {
+			log->l_stripemask = 1;
+		} else {
+			log->l_stripemask = 1 <<
+				xfs_highbit32(mp->m_sb.sb_logsunit >> BBSHIFT);
+		}
+	}
+	if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb)) {
+		log->l_sectbb_log = mp->m_sb.sb_logsectlog - BBSHIFT;
+		ASSERT(log->l_sectbb_log <= mp->m_sectbb_log);
+		ASSERT(XFS_SB_VERSION_HASLOGV2(&mp->m_sb));
+		ASSERT(mp->m_sb.sb_logsectlog >= BBSHIFT);
+	}
+	log->l_sectbb_mask = (1 << log->l_sectbb_log) - 1;
 
 	xlog_get_iclog_buffer_size(mp, log);
 
@@ -2811,10 +2807,9 @@ xlog_state_switch_iclogs(xlog_t		*log,
 
 	/* Round up to next log-sunit */
 	if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
-		if (log->l_curr_block & (log->l_mp->m_lstripemask - 1)) {
-			roundup = log->l_mp->m_lstripemask -
-				(log->l_curr_block &
-				 (log->l_mp->m_lstripemask - 1));
+		if (log->l_curr_block & (log->l_stripemask - 1)) {
+			roundup = log->l_stripemask -
+				(log->l_curr_block & (log->l_stripemask - 1));
 		} else {
 			roundup = 0;
 		}
@@ -3293,15 +3288,17 @@ xlog_verify_disk_cycle_no(xlog_t	 *log,
 {
     xfs_buf_t	*bp;
     uint	cycle_no;
+    xfs_caddr_t ptr;
     xfs_daddr_t	i;
 
     if (BLOCK_LSN(iclog->ic_header.h_lsn, ARCH_CONVERT) < 10) {
 	cycle_no = CYCLE_LSN(iclog->ic_header.h_lsn, ARCH_CONVERT);
-	bp = xlog_get_bp(1, log->l_mp);
+	bp = xlog_get_bp(log, 1);
 	ASSERT(bp);
 	for (i = 0; i < BLOCK_LSN(iclog->ic_header.h_lsn, ARCH_CONVERT); i++) {
 	    xlog_bread(log, i, 1, bp);
-	    if (GET_CYCLE(XFS_BUF_PTR(bp), ARCH_CONVERT) != cycle_no)
+	    ptr = xlog_align(log, i, 1, bp);
+	    if (GET_CYCLE(ptr, ARCH_CONVERT) != cycle_no)
 		xlog_warn("XFS: xlog_verify_disk_cycle_no: bad cycle no");
 	}
 	xlog_put_bp(bp);

fs/xfs/xfs_log.h

 /*
- * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
+ * 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
@@ -153,7 +153,6 @@ xfs_lsn_t xfs_log_done(struct xfs_mount *mp,
 int	  xfs_log_force(struct xfs_mount *mp,
 			xfs_lsn_t	 lsn,
 			uint		 flags);
-int	  xfs_log_init(void);
 int	  xfs_log_mount(struct xfs_mount *mp,
 			dev_t		 log_dev,
 			xfs_daddr_t		 start_block,

fs/xfs/xfs_log_priv.h

 /*
- * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
+ * 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
@@ -73,6 +73,9 @@ int xlog_btolrbb(int b);
 
 #define XLOG_HEADER_SIZE	512
 
+#define XLOG_REC_SHIFT(log) \
+	BTOBB(1 << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
+	 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
 #define XLOG_TOTAL_REC_SHIFT(log) \
 	BTOBB(XLOG_MAX_ICLOGS << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
 	 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
@@ -202,9 +205,9 @@ void xlog_grant_add_space(struct log *log, int bytes, int type);
 #define LOG_LOCK(log)		mutex_spinlock(&(log)->l_icloglock)
 #define LOG_UNLOCK(log, s)	mutex_spinunlock(&(log)->l_icloglock, s)
 
-#define xlog_panic(s)		{cmn_err(CE_PANIC, s); }
-#define xlog_exit(s)		{cmn_err(CE_PANIC, s); }
-#define xlog_warn(s)		{cmn_err(CE_WARN, s); }
+#define xlog_panic(args...)	cmn_err(CE_PANIC, ## args)
+#define xlog_exit(args...)	cmn_err(CE_PANIC, ## args)
+#define xlog_warn(args...)	cmn_err(CE_WARN, ## args)
 
 /*
  * In core log state
@@ -403,6 +406,7 @@ typedef struct xlog_rec_ext_header {
 	uint	  xh_cycle;	/* write cycle of log			: 4 */
 	uint	  xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /*	: 256 */
 } xlog_rec_ext_header_t;
+
 #ifdef __KERNEL__
 /*
  * - A log record header is 512 bytes.  There is plenty of room to grow the
@@ -441,12 +445,10 @@ typedef struct xlog_iclog_fields {
 	char			*ic_datap;	/* pointer to iclog data */
 } xlog_iclog_fields_t;
 
-typedef struct xlog_in_core2 {
-	union {
-		xlog_rec_header_t hic_header;
-		xlog_rec_ext_header_t hic_xheader;
-		char		  hic_sector[XLOG_HEADER_SIZE];
-	} ic_h;
+typedef union xlog_in_core2 {
+	xlog_rec_header_t	hic_header;
+	xlog_rec_ext_header_t	hic_xheader;
+	char			hic_sector[XLOG_HEADER_SIZE];
 } xlog_in_core_2_t;
 
 typedef struct xlog_in_core {
@@ -473,7 +475,7 @@ typedef struct xlog_in_core {
 #define	ic_bwritecnt	hic_fields.ic_bwritecnt
 #define	ic_state	hic_fields.ic_state
 #define ic_datap	hic_fields.ic_datap
-#define ic_header	hic_data->ic_h.hic_header
+#define ic_header	hic_data->hic_header
 
 /*
  * The reservation head lsn is not made up of a cycle number and block number.
@@ -530,8 +532,11 @@ typedef struct log {
     uint		l_flags;
     uint		l_quotaoffs_flag;/* XFS_DQ_*, if QUOTAOFFs found */
     struct xfs_buf_cancel **l_buf_cancel_table;
+    int			l_stripemask;	/* log stripe mask */
     int			l_iclog_hsize;  /* size of iclog header */
     int			l_iclog_heads;  /* number of iclog header sectors */
+    uint		l_sectbb_log;   /* log2 of sector size in bbs */
+    uint		l_sectbb_mask;  /* sector size in bbs alignment mask */
 } xlog_t;
 
 
@@ -546,11 +551,13 @@ extern int	 xlog_print_find_oldest(xlog_t *log, xfs_daddr_t *last_blk);
 extern int	 xlog_recover(xlog_t *log, int readonly);
 extern int	 xlog_recover_finish(xlog_t *log, int mfsi_flags);
 extern void	 xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog);
-extern struct xfs_buf *xlog_get_bp(int,xfs_mount_t *);
-extern void	 xlog_put_bp(struct xfs_buf *);
-extern int	 xlog_bread(xlog_t *, xfs_daddr_t blkno, int bblks, struct xfs_buf *bp);
 extern void	 xlog_recover_process_iunlinks(xlog_t *log);
 
+extern struct xfs_buf *xlog_get_bp(xlog_t *, int);
+extern void	 xlog_put_bp(struct xfs_buf *);
+extern int	 xlog_bread(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
+extern xfs_caddr_t xlog_align(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
+
 #define XLOG_TRACE_GRAB_FLUSH  1
 #define XLOG_TRACE_REL_FLUSH   2
 #define XLOG_TRACE_SLEEP_FLUSH 3

fs/xfs/xfs_mount.c

@@ -467,7 +467,11 @@ xfs_readsb(xfs_mount_t *mp)
 
 	bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
 				BTOBB(sector_size), extra_flags);
-	ASSERT(bp);
+	if (!bp || XFS_BUF_ISERROR(bp)) {
+		cmn_err(CE_WARN, "XFS: SB read failed");
+		error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
+		goto fail;
+	}
 	ASSERT(XFS_BUF_ISBUSY(bp));
 	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
 
@@ -482,9 +486,7 @@ xfs_readsb(xfs_mount_t *mp)
 	error = xfs_mount_validate_sb(mp, &(mp->m_sb));
 	if (error) {
 		cmn_err(CE_WARN, "XFS: SB validate failed");
-		XFS_BUF_UNMANAGE(bp);
-		xfs_buf_relse(bp);
-		return error;
+		goto fail;
 	}
 
 	/*
@@ -494,9 +496,8 @@ xfs_readsb(xfs_mount_t *mp)
 		cmn_err(CE_WARN,
 			"XFS: device supports only %u byte sectors (not %u)",
 			sector_size, mp->m_sb.sb_sectsize);
-		XFS_BUF_UNMANAGE(bp);
-		xfs_buf_relse(bp);
-		return XFS_ERROR(ENOSYS);
+		error = ENOSYS;
+		goto fail;
 	}
 
 	/*
@@ -509,7 +510,11 @@ xfs_readsb(xfs_mount_t *mp)
 		sector_size = mp->m_sb.sb_sectsize;
 		bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
 					BTOBB(sector_size), extra_flags);
-		ASSERT(bp);
+		if (!bp || XFS_BUF_ISERROR(bp)) {
+			cmn_err(CE_WARN, "XFS: SB re-read failed");
+			error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
+			goto fail;
+		}
 		ASSERT(XFS_BUF_ISBUSY(bp));
 		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
 	}
@@ -518,6 +523,13 @@ xfs_readsb(xfs_mount_t *mp)
 	xfs_buf_relse(bp);
 	ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
 	return 0;
+
+ fail:
+	if (bp) {
+		XFS_BUF_UNMANAGE(bp);
+		xfs_buf_relse(bp);
+	}
+	return error;
 }
 
 
@@ -546,16 +558,7 @@ 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;
-
-
-	if (XFS_SB_VERSION_HASLOGV2(sbp)) {
-		if (sbp->sb_logsunit <= 1) {
-			mp->m_lstripemask = 1;
-		} else {
-			mp->m_lstripemask =
-				1 << xfs_highbit32(sbp->sb_logsunit >> BBSHIFT);
-		}
-	}
+	INIT_LIST_HEAD(&mp->m_del_inodes);
 
 	/*
 	 * Setup for attributes, in case they get created.
@@ -601,7 +604,6 @@ 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

@@ -68,6 +68,7 @@ typedef struct xfs_trans_reservations {
 	((xfs_agblock_t)(XFS_BB_TO_FSBT(mp, d) % (mp)->m_sb.sb_agblocks))
 #else
 struct cred;
+struct log;
 struct vfs;
 struct vnode;
 struct xfs_mount_args;
@@ -296,13 +297,14 @@ 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 */
 	uint			m_readio_blocks; /* min read size blocks */
 	uint			m_writeio_log;	/* min write size log bytes */
 	uint			m_writeio_blocks; /* min write size blocks */
-	void			*m_log;		/* log specific stuff */
+	struct log		*m_log;		/* log specific stuff */
 	int			m_logbufs;	/* number of log buffers */
 	int			m_logbsize;	/* size of each log buffer */
 	uint			m_rsumlevels;	/* rt summary levels */
@@ -357,7 +359,6 @@ typedef struct xfs_mount {
 #endif
 	int			m_dalign;	/* stripe unit */
 	int			m_swidth;	/* stripe width */
-	int			m_lstripemask;	/* log stripe mask */
 	int			m_sinoalign;	/* stripe unit inode alignmnt */
 	int			m_attr_magicpct;/* 37% of the blocksize */
 	int			m_dir_magicpct;	/* 37% of the dir blocksize */

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

@@ -658,7 +658,7 @@ xfs_setattr(
 		if (vap->va_size > ip->i_d.di_size) {
 			code = xfs_igrow_start(ip, vap->va_size, credp);
 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
-		} else if (vap->va_size < ip->i_d.di_size) {
+		} else if (vap->va_size <= ip->i_d.di_size) {
 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
 			xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
 					    (xfs_fsize_t)vap->va_size);
@@ -701,7 +701,7 @@ xfs_setattr(
 		if (vap->va_size > ip->i_d.di_size) {
 			xfs_igrow_finish(tp, ip, vap->va_size,
 			    !(flags & ATTR_DMI));
-		} else if ((vap->va_size < ip->i_d.di_size) ||
+		} else if ((vap->va_size <= ip->i_d.di_size) ||
 			   ((vap->va_size == 0) && ip->i_d.di_nextents)) {
 			/*
 			 * signal a sync transaction unless
@@ -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;
 	}

fs/xfs/xfsidbg.c

@@ -4269,8 +4269,10 @@ xfsidbg_xlog(xlog_t *log)
 	kdb_printf("iclog_bak: 0x%p  iclog_size: 0x%x (%d)  num iclogs: %d\n",
 		log->l_iclog_bak, log->l_iclog_size, log->l_iclog_size,
 		log->l_iclog_bufs);
-	kdb_printf("l_iclog_hsize %d l_iclog_heads %d\n",
-		log->l_iclog_hsize, log->l_iclog_heads);
+	kdb_printf("l_stripemask %d l_iclog_hsize %d l_iclog_heads %d\n",
+		log->l_stripemask, log->l_iclog_hsize, log->l_iclog_heads);
+	kdb_printf("l_sectbb_log %u l_sectbb_mask %u\n",
+		log->l_sectbb_log, log->l_sectbb_mask);
 	kdb_printf("&grant_lock: 0x%p  resHeadQ: 0x%p  wrHeadQ: 0x%p\n",
 		&log->l_grant_lock, log->l_reserve_headq, log->l_write_headq);
 	kdb_printf("GResCycle: %d  GResBytes: %d  GWrCycle: %d  GWrBytes: %d\n",
@@ -4712,7 +4714,6 @@ xfsidbg_xmount(xfs_mount_t *mp)
 		(xfs_dfiloff_t)mp->m_dirfreeblk);
 	kdb_printf("chsize %d chash 0x%p\n",
 		mp->m_chsize, mp->m_chash);
-	kdb_printf("m_lstripemask %d\n", mp->m_lstripemask);
 	kdb_printf("m_frozen %d m_active_trans %d\n",
 		mp->m_frozen, mp->m_active_trans.counter);
 	if (mp->m_fsname != NULL)