Merge bk://jfs.bkbits.net/linux-2.5

into home.transmeta.com:/home/torvalds/v2.5/linux

Documentation/filesystems/jfs.txt

@@ -17,11 +17,16 @@ iocharset=name	Character set to use for converting from Unicode to
 		translations.  This requires CONFIG_NLS_UTF8 to be set
 		in the kernel .config file.
 
+resize=value	Resize the volume to <value> blocks.  JFS only supports
+		growing a volume, not shrinking it.  This option is only
+		valid during a remount, when the volume is mounted
+		read-write.  The resize keyword with no value will grow
+		the volume to the full size of the partition.
+
 JFS TODO list:
 
 Plans for our near term development items
 
-   - implement online resize for extending JFS volumes
    - enhance support for logfile on dedicated partition
    - get access control list functionality operational
    - get extended attributes functionality operational

fs/jfs/Makefile

@@ -8,7 +8,7 @@ jfs-objs := super.o file.o inode.o namei.o jfs_mount.o jfs_umount.o \
 	    jfs_xtree.o jfs_imap.o jfs_debug.o jfs_dmap.o \
 	    jfs_unicode.o jfs_dtree.o jfs_inode.o \
 	    jfs_extent.o symlink.o jfs_metapage.o \
-	    jfs_logmgr.o jfs_txnmgr.o jfs_uniupr.o
+	    jfs_logmgr.o jfs_txnmgr.o jfs_uniupr.o resize.o
 
 EXTRA_CFLAGS += -D_JFS_4K
 

fs/jfs/file.c

@@ -38,9 +38,7 @@ int jfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
 		return rc;
 
-	IWRITE_LOCK(inode);
 	rc |= jfs_commit_inode(inode, 1);
-	IWRITE_UNLOCK(inode);
 
 	return rc ? -EIO : 0;
 }
@@ -64,10 +62,19 @@ void jfs_truncate_nolock(struct inode *ip, loff_t length)
 	do {
 		tid = txBegin(ip->i_sb, 0);
 
+		/*
+		 * The commit_sem cannot be taken before txBegin.
+		 * txBegin may block and there is a chance the inode
+		 * could be marked dirty and need to be committed
+		 * before txBegin unblocks
+		 */
+		down(&JFS_IP(ip)->commit_sem);
+
 		newsize = xtTruncate(tid, ip, length,
 				     COMMIT_TRUNCATE | COMMIT_PWMAP);
 		if (newsize < 0) {
 			txEnd(tid);
+			up(&JFS_IP(ip)->commit_sem);
 			break;
 		}
 
@@ -76,6 +83,7 @@ void jfs_truncate_nolock(struct inode *ip, loff_t length)
 
 		txCommit(tid, 1, &ip, 0);
 		txEnd(tid);
+		up(&JFS_IP(ip)->commit_sem);
 	} while (newsize > length);	/* Truncate isn't always atomic */
 }
 

fs/jfs/inode.c

@@ -107,8 +107,10 @@ int jfs_commit_inode(struct inode *inode, int wait)
 	}
 
 	tid = txBegin(inode->i_sb, COMMIT_INODE);
+	down(&JFS_IP(inode)->commit_sem);
 	rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
 	txEnd(tid);
+	up(&JFS_IP(inode)->commit_sem);
 	return -rc;
 }
 
@@ -123,25 +125,19 @@ void jfs_write_inode(struct inode *inode, int wait)
 	    !test_cflag(COMMIT_Dirty, inode))
 		return;
 
-	IWRITE_LOCK(inode);
-
 	if (jfs_commit_inode(inode, wait)) {
 		jERROR(1, ("jfs_write_inode: jfs_commit_inode failed!\n"));
 	}
-
-	IWRITE_UNLOCK(inode);
 }
 
 void jfs_delete_inode(struct inode *inode)
 {
 	jFYI(1, ("In jfs_delete_inode, inode = 0x%p\n", inode));
 
-	IWRITE_LOCK(inode);
 	if (test_cflag(COMMIT_Freewmap, inode))
 		freeZeroLink(inode);
 
 	diFree(inode);
-	IWRITE_UNLOCK(inode);
 
 	clear_inode(inode);
 }
@@ -203,8 +199,7 @@ static int jfs_get_block(struct inode *ip, sector_t lblock,
 
 	if ((no_size_check ||
 	     ((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size)) &&
-	    (xtLookup
-	     (ip, lblock64, 1, &xflag, &xaddr, &xlen, no_size_check)
+	    (xtLookup(ip, lblock64, 1, &xflag, &xaddr, &xlen, no_size_check)
 	     == 0) && xlen) {
 		if (xflag & XAD_NOTRECORDED) {
 			if (!create)
@@ -241,8 +236,7 @@ static int jfs_get_block(struct inode *ip, sector_t lblock,
 	 * Allocate a new block
 	 */
 #ifdef _JFS_4K
-	if ((rc =
-	     extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
+	if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
 		goto unlock;
 	rc = extAlloc(ip, 1, lblock64, &xad, FALSE);
 	if (rc)

fs/jfs/jfs_dmap.c

@@ -330,7 +330,7 @@ int dbSync(struct inode *ipbmap)
 	filemap_fdatawait(ipbmap->i_mapping);
 
 	ipbmap->i_state |= I_DIRTY;
-	diWriteSpecial(ipbmap);
+	diWriteSpecial(ipbmap, 0);
 
 	return (0);
 }
@@ -3175,7 +3175,7 @@ static int dbAllocDmapBU(bmap_t * bmp, dmap_t * dp, s64 blkno, int nblocks)
 			dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb)
 						      >> wbitno);
 
-			word += 1;
+			word++;
 		} else {
 			/* one or more dmap words are fully contained
 			 * within the block range.  determine how many
@@ -3187,6 +3187,7 @@ static int dbAllocDmapBU(bmap_t * bmp, dmap_t * dp, s64 blkno, int nblocks)
 
 			/* determine how many bits */
 			nb = nwords << L2DBWORD;
+			word += nwords;
 		}
 	}
 

fs/jfs/jfs_extent.c

@@ -96,6 +96,9 @@ extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, boolean_t abnr)
 	/* This blocks if we are low on resources */
 	txBeginAnon(ip->i_sb);
 
+	/* Avoid race with jfs_commit_inode() */
+	down(&JFS_IP(ip)->commit_sem);
+
 	/* validate extent length */
 	if (xlen > MAXXLEN)
 		xlen = MAXXLEN;
@@ -138,8 +141,8 @@ extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, boolean_t abnr)
 	 * is smaller than the number of blocks per page.
 	 */
 	nxlen = xlen;
-	if ((rc =
-	     extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
+	if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
+		up(&JFS_IP(ip)->commit_sem);
 		return (rc);
 	}
 
@@ -160,6 +163,7 @@ extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, boolean_t abnr)
 	 */
 	if (rc) {
 		dbFree(ip, nxaddr, nxlen);
+		up(&JFS_IP(ip)->commit_sem);
 		return (rc);
 	}
 
@@ -174,6 +178,7 @@ extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, boolean_t abnr)
 
 	mark_inode_dirty(ip);
 
+	up(&JFS_IP(ip)->commit_sem);
 	/*
 	 * COMMIT_SyncList flags an anonymous tlock on page that is on
 	 * sync list.
@@ -217,6 +222,7 @@ int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
 	/* This blocks if we are low on resources */
 	txBeginAnon(ip->i_sb);
 
+	down(&JFS_IP(ip)->commit_sem);
 	/* validate extent length */
 	if (nxlen > MAXXLEN)
 		nxlen = MAXXLEN;
@@ -235,7 +241,7 @@ int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
 	if ((xp->flag & XAD_NOTRECORDED) && !abnr) {
 		xp->flag = 0;
 		if ((rc = xtUpdate(0, ip, xp)))
-			return (rc);
+			goto exit;
 	}
 
 	/* try to allocated the request number of blocks for the
@@ -247,7 +253,7 @@ int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
 	 * space as to satisfy the extend page.
 	 */
 	if ((rc = extBrealloc(ip, xaddr, xlen, &nxlen, &nxaddr)))
-		return (rc);
+		goto exit;
 
 	delta = nxlen - xlen;
 
@@ -284,7 +290,7 @@ int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
 		/* extend the extent */
 		if ((rc = xtExtend(0, ip, xoff + xlen, (int) nextend, 0))) {
 			dbFree(ip, xaddr + xlen, delta);
-			return (rc);
+			goto exit;
 		}
 	} else {
 		/*
@@ -294,7 +300,7 @@ int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
 		 */
 		if ((rc = xtTailgate(0, ip, xoff, (int) ntail, nxaddr, 0))) {
 			dbFree(ip, nxaddr, nxlen);
-			return (rc);
+			goto exit;
 		}
 	}
 
@@ -325,8 +331,9 @@ int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
 	xp->flag = xflag;
 
 	mark_inode_dirty(ip);
-
-	return (0);
+exit:
+	up(&JFS_IP(ip)->commit_sem);
+	return (rc);
 }
 
 
@@ -423,19 +430,13 @@ int extRecord(struct inode *ip, xad_t * xp)
 
 	txBeginAnon(ip->i_sb);
 
-	/* update the extent */
-	if ((rc = xtUpdate(0, ip, xp)))
-		return (rc);
-
-#ifdef _STILL_TO_PORT
-	/* no longer abnr */
-	cp->cm_abnr = FALSE;
+	down(&JFS_IP(ip)->commit_sem);
 
-	/* mark the cbuf as modified */
-	cp->cm_modified = TRUE;
-#endif				/*  _STILL_TO_PORT */
+	/* update the extent */
+	rc = xtUpdate(0, ip, xp);
 
-	return (0);
+	up(&JFS_IP(ip)->commit_sem);
+	return (rc);
 }
 
 

fs/jfs/jfs_imap.c

@@ -285,7 +285,7 @@ int diSync(struct inode *ipimap)
 	filemap_fdatawrite(ipimap->i_mapping);
 	filemap_fdatawait(ipimap->i_mapping);
 
-	diWriteSpecial(ipimap);
+	diWriteSpecial(ipimap, 0);
 
 	return (0);
 }
@@ -450,12 +450,13 @@ int diRead(struct inode *ip)
  * PARAMETERS:
  *      sb - filesystem superblock
  *	inum - aggregate inode number
+ *	secondary - 1 if secondary aggregate inode table
  *
  * RETURN VALUES:
  *      new inode	- success
  *      NULL		- i/o error.
  */
-struct inode *diReadSpecial(struct super_block *sb, ino_t inum)
+struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
 {
 	struct jfs_sb_info *sbi = JFS_SBI(sb);
 	uint address;
@@ -470,21 +471,16 @@ struct inode *diReadSpecial(struct super_block *sb, ino_t inum)
 		return ip;
 	}
 
-	/*
-	 * If ip->i_number >= 32 (INOSPEREXT), then read from secondary
-	 * aggregate inode table.
-	 */
-
-	if (inum >= INOSPEREXT) {
-		address =
-		    addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
-		inum -= INOSPEREXT;
-		ASSERT(inum < INOSPEREXT);
+	if (secondary) {
+		address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
 		JFS_IP(ip)->ipimap = sbi->ipaimap2;
 	} else {
 		address = AITBL_OFF >> L2PSIZE;
 		JFS_IP(ip)->ipimap = sbi->ipaimap;
 	}
+
+	ASSERT(inum < INOSPEREXT);
+
 	ip->i_ino = inum;
 
 	address += inum >> 3;	/* 8 inodes per 4K page */
@@ -538,11 +534,12 @@ struct inode *diReadSpecial(struct super_block *sb, ino_t inum)
  *
  * PARAMETERS:
  *      ip - special inode
+ *	secondary - 1 if secondary aggregate inode table
  *
  * RETURN VALUES: none
  */
 
-void diWriteSpecial(struct inode *ip)
+void diWriteSpecial(struct inode *ip, int secondary)
 {
 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 	uint address;
@@ -550,24 +547,14 @@ void diWriteSpecial(struct inode *ip)
 	ino_t inum = ip->i_ino;
 	metapage_t *mp;
 
-	/*
-	 * If ip->i_number >= 32 (INOSPEREXT), then write to secondary
-	 * aggregate inode table.
-	 */
-
-	if (!(ip->i_state & I_DIRTY))
-		return;
-
 	ip->i_state &= ~I_DIRTY;
 
-	if (inum >= INOSPEREXT) {
-		address =
-		    addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
-		inum -= INOSPEREXT;
-		ASSERT(inum < INOSPEREXT);
-	} else {
+	if (secondary)
+		address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
+	else
 		address = AITBL_OFF >> L2PSIZE;
-	}
+
+	ASSERT(inum < INOSPEREXT);
 
 	address += inum >> 3;	/* 8 inodes per 4K page */
 
@@ -2996,7 +2983,7 @@ duplicateIXtree(struct super_block *sb, s64 blkno, int xlen, s64 * xaddr)
 	/* if AIT2 ipmap2 is bad, do not try to update it */
 	if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT)	/* s_flag */
 		return;
-	ip = diReadSpecial(sb, FILESYSTEM_I + INOSPEREXT);
+	ip = diReadSpecial(sb, FILESYSTEM_I, 1);
 	if (ip == 0) {
 		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
 		if ((rc = readSuper(sb, &mpsuper)))

fs/jfs/jfs_imap.h

@@ -146,15 +146,12 @@ extern int diSync(struct inode *);
 /* external references */
 extern int diUpdatePMap(struct inode *ipimap, unsigned long inum,
 			boolean_t is_free, tblock_t * tblk);
-#ifdef _STILL_TO_PORT
-extern int diExtendFS(inode_t * ipimap, inode_t * ipbmap);
-#endif				/* _STILL_TO_PORT */
-
+extern int diExtendFS(struct inode *ipimap, struct inode *ipbmap);
 extern int diMount(struct inode *);
 extern int diUnmount(struct inode *, int);
 extern int diRead(struct inode *);
-extern struct inode *diReadSpecial(struct super_block *, ino_t);
-extern void diWriteSpecial(struct inode *);
+extern struct inode *diReadSpecial(struct super_block *, ino_t, int);
+extern void diWriteSpecial(struct inode *, int);
 extern void diFreeSpecial(struct inode *);
 extern int diWrite(tid_t tid, struct inode *);
 #endif				/* _H_JFS_IMAP */

fs/jfs/jfs_incore.h

@@ -19,6 +19,7 @@
 #ifndef _H_JFS_INCORE
 #define _H_JFS_INCORE
 
+#include <linux/rwsem.h>
 #include <linux/slab.h>
 #include <asm/bitops.h>
 #include "jfs_types.h"
@@ -30,14 +31,6 @@
  */
 #define JFS_SUPER_MAGIC 0x3153464a /* "JFS1" */
 
-/*
- * Due to header ordering problems this can't be in jfs_lock.h
- */
-typedef struct	jfs_rwlock {
-	struct rw_semaphore rw_sem;
-	atomic_t in_use;	/* for hacked implementation of trylock */
-} jfs_rwlock_t;
-
 /*
  * JFS-private inode information
  */
@@ -62,7 +55,19 @@ struct jfs_inode_info {
 	lid_t	atltail;	/* anonymous tlock list tail	*/
 	struct list_head anon_inode_list; /* inodes having anonymous txns */
 	struct list_head mp_list; /* metapages in inode's address space */
-	jfs_rwlock_t rdwrlock;	/* read/write lock	*/
+	/*
+	 * rdwrlock serializes xtree between reads & writes and synchronizes
+	 * changes to special inodes.  It's use would be redundant on
+	 * directories since the i_sem taken in the VFS is sufficient.
+	 */
+	struct rw_semaphore rdwrlock;
+	/*
+	 * commit_sem serializes transaction processing on an inode.
+	 * It must be taken after beginning a transaction (txBegin), since
+	 * dirty inodes may be committed while a new transaction on the
+	 * inode is blocked in txBegin or TxBeginAnon
+	 */
+	struct semaphore commit_sem;
 	lid_t	xtlid;		/* lid of xtree lock on directory */
 	union {
 		struct {
@@ -87,6 +92,12 @@ struct jfs_inode_info {
 #define i_dtroot u.dir._dtroot
 #define i_inline u.link._inline
 
+
+#define IREAD_LOCK(ip)		down_read(&JFS_IP(ip)->rdwrlock)
+#define IREAD_UNLOCK(ip)	up_read(&JFS_IP(ip)->rdwrlock)
+#define IWRITE_LOCK(ip)		down_write(&JFS_IP(ip)->rdwrlock)
+#define IWRITE_UNLOCK(ip)	up_write(&JFS_IP(ip)->rdwrlock)
+
 /*
  * cflag
  */
@@ -125,6 +136,7 @@ struct jfs_sb_info {
 	u32		logdev;		/* 2: external log device	*/
 	uint		aggregate;	/* volume identifier in log record */
 	pxd_t		logpxd;		/* 8: pxd describing log	*/
+	pxd_t		fsckpxd;	/* 8: pxd describing fsck wkspc */
 	pxd_t		ait2;		/* 8: pxd describing AIT copy	*/
 	char		uuid[16];	/* 16: 128-bit uuid for volume	*/
 	char		loguuid[16];	/* 16: 128-bit uuid for log	*/

fs/jfs/jfs_inode.c

@@ -91,40 +91,3 @@ struct inode *ialloc(struct inode *parent, umode_t mode)
 
 	return inode;
 }
-
-/*
- * NAME:	iwritelocklist()
- *
- * FUNCTION:	Lock multiple inodes in sorted order to avoid deadlock
- *
- */
-void iwritelocklist(int n, ...)
-{
-	va_list ilist;
-	struct inode *sort[4];
-	struct inode *ip;
-	int k, m;
-
-	va_start(ilist, n);
-	for (k = 0; k < n; k++)
-		sort[k] = va_arg(ilist, struct inode *);
-	va_end(ilist);
-
-	/* Bubble sort in descending order */
-	do {
-		m = 0;
-		for (k = 0; k < n; k++)
-			if ((k + 1) < n
-			    && sort[k + 1]->i_ino > sort[k]->i_ino) {
-				ip = sort[k];
-				sort[k] = sort[k + 1];
-				sort[k + 1] = ip;
-				m++;
-			}
-	} while (m);
-
-	/* Lock them */
-	for (k = 0; k < n; k++) {
-		IWRITE_LOCK(sort[k]);
-	}
-}

fs/jfs/jfs_lock.h

@@ -24,63 +24,7 @@
 
 /*
  *	jfs_lock.h
- *
- * JFS lock definition for globally referenced locks
- */
-
-/* readers/writer lock: thread-thread */
-
-/*
- * RW semaphores do not currently have a trylock function.  Since the
- * implementation varies by platform, I have implemented a platform-independent
- * wrapper around the rw_semaphore routines.  If this turns out to be the best
- * way of avoiding our locking problems, I will push to get a trylock
- * implemented in the kernel, but I'd rather find a way to avoid having to
- * use it.
  */
-#define RDWRLOCK_T jfs_rwlock_t
-static inline void RDWRLOCK_INIT(jfs_rwlock_t * Lock)
-{
-	init_rwsem(&Lock->rw_sem);
-	atomic_set(&Lock->in_use, 0);
-}
-static inline void READ_LOCK(jfs_rwlock_t * Lock)
-{
-	atomic_inc(&Lock->in_use);
-	down_read(&Lock->rw_sem);
-}
-static inline void READ_UNLOCK(jfs_rwlock_t * Lock)
-{
-	up_read(&Lock->rw_sem);
-	atomic_dec(&Lock->in_use);
-}
-static inline void WRITE_LOCK(jfs_rwlock_t * Lock)
-{
-	atomic_inc(&Lock->in_use);
-	down_write(&Lock->rw_sem);
-}
-
-static inline int WRITE_TRYLOCK(jfs_rwlock_t * Lock)
-{
-	if (atomic_read(&Lock->in_use))
-		return 0;
-	WRITE_LOCK(Lock);
-	return 1;
-}
-static inline void WRITE_UNLOCK(jfs_rwlock_t * Lock)
-{
-	up_write(&Lock->rw_sem);
-	atomic_dec(&Lock->in_use);
-}
-
-#define IREAD_LOCK(ip)		READ_LOCK(&JFS_IP(ip)->rdwrlock)
-#define IREAD_UNLOCK(ip)	READ_UNLOCK(&JFS_IP(ip)->rdwrlock)
-#define IWRITE_LOCK(ip)		WRITE_LOCK(&JFS_IP(ip)->rdwrlock)
-#define IWRITE_TRYLOCK(ip)	WRITE_TRYLOCK(&JFS_IP(ip)->rdwrlock)
-#define IWRITE_UNLOCK(ip)	WRITE_UNLOCK(&JFS_IP(ip)->rdwrlock)
-#define IWRITE_LOCK_LIST	iwritelocklist
-
-extern void iwritelocklist(int, ...);
 
 /*
  * Conditional sleep where condition is protected by spinlock

fs/jfs/jfs_logmgr.c

@@ -172,8 +172,6 @@ static int lmWriteRecord(log_t * log, tblock_t * tblk, lrd_t * lrd,
 
 static int lmNextPage(log_t * log);
 static int lmLogFileSystem(log_t * log, char *uuid, int activate);
-static int lmLogInit(log_t * log);
-static int lmLogShutdown(log_t * log);
 
 static int lbmLogInit(log_t * log);
 static void lbmLogShutdown(log_t * log);
@@ -1037,7 +1035,7 @@ int lmLogSync(log_t * log, int nosyncwait)
 	 * by setting syncbarrier flag.
 	 */
 	if (written > LOGSYNC_BARRIER(logsize) && logsize > 32 * LOGPSIZE) {
-		log->syncbarrier = 1;
+		set_bit(log_SYNCBARRIER, &log->flag);
 		jFYI(1, ("log barrier on: lsn=0x%x syncpt=0x%x\n", lsn,
 			 log->syncpt));
 	}
@@ -1068,6 +1066,7 @@ int lmLogOpen(struct super_block *sb, log_t ** logptr)
 	if (!(log = kmalloc(sizeof(log_t), GFP_KERNEL)))
 		return ENOMEM;
 	memset(log, 0, sizeof(log_t));
+	init_waitqueue_head(&log->syncwait);
 
 	log->sb = sb;		/* This should be a list */
 
@@ -1080,7 +1079,7 @@ int lmLogOpen(struct super_block *sb, log_t ** logptr)
 	 * file system to log have 1-to-1 relationship;
 	 */
 
-	log->flag = JFS_INLINELOG;
+	set_bit(log_INLINELOG, &log->flag);
 	log->bdev = sb->s_bdev;
 	log->base = addressPXD(&JFS_SBI(sb)->logpxd);
 	log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
@@ -1175,7 +1174,7 @@ int lmLogOpen(struct super_block *sb, log_t ** logptr)
  *			
  * serialization: single first open thread
  */
-static int lmLogInit(log_t * log)
+int lmLogInit(log_t * log)
 {
 	int rc = 0;
 	lrd_t lrd;
@@ -1203,7 +1202,7 @@ static int lmLogInit(log_t * log)
 	 */
 
 
-	if (!(log->flag & JFS_INLINELOG))
+	if (!test_bit(log_INLINELOG, &log->flag))
 		log->l2bsize = 12;	/* XXX kludge alert XXX */
 	if ((rc = lbmRead(log, 1, &bpsuper)))
 		goto errout10;
@@ -1224,7 +1223,7 @@ static int lmLogInit(log_t * log)
 	}
 
 	/* initialize log inode from log superblock */
-	if (log->flag & JFS_INLINELOG) {
+	if (test_bit(log_INLINELOG,&log->flag)) {
 		if (log->size != le32_to_cpu(logsuper->size)) {
 			rc = EINVAL;
 			goto errout20;
@@ -1244,10 +1243,6 @@ static int lmLogInit(log_t * log)
 		      log, (unsigned long long) log->base, log->size));
 	}
 
-	log->flag |= JFS_GROUPCOMMIT;
-/*
-	log->flag |= JFS_LAZYCOMMIT;
-*/
 	log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
 	log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
 
@@ -1309,7 +1304,6 @@ static int lmLogInit(log_t * log)
 	log->syncpt = lsn;
 	log->sync = log->syncpt;
 	log->nextsync = LOGSYNC_DELTA(log->logsize);
-	init_waitqueue_head(&log->syncwait);
 
 	jFYI(1, ("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x\n",
 		 log->lsn, log->syncpt, log->sync));
@@ -1377,7 +1371,7 @@ int lmLogClose(struct super_block *sb, log_t * log)
 
 	jFYI(1, ("lmLogClose: log:0x%p\n", log));
 
-	if (!(log->flag & JFS_INLINELOG))
+	if (!test_bit(log_INLINELOG, &log->flag))
 		goto externalLog;
 	
 	/*
@@ -1445,7 +1439,7 @@ void lmLogWait(log_t *log)
  *			
  * serialization: single last close thread
  */
-static int lmLogShutdown(log_t * log)
+int lmLogShutdown(log_t * log)
 {
 	int rc;
 	lrd_t lrd;
@@ -1524,8 +1518,6 @@ static int lmLogShutdown(log_t * log)
  *
  * RETURN:	0	- success
  *		errors returned by vms_iowait().
- *			
- * serialization: IWRITE_LOCK(log inode) held on entry/exit
  */
 static int lmLogFileSystem(log_t * log, char *uuid, int activate)
 {
@@ -1578,37 +1570,6 @@ static int lmLogFileSystem(log_t * log, char *uuid, int activate)
 	return rc;
 }
 
-
-/*
- *	lmLogQuiesce()
- */
-int lmLogQuiesce(log_t * log)
-{
-	int rc;
-
-	rc = lmLogShutdown(log);
-
-	return rc;
-}
-
-
-/*
- *	lmLogResume()
- */
-int lmLogResume(log_t * log, struct super_block *sb)
-{
-	struct jfs_sb_info *sbi = JFS_SBI(sb);
-	int rc;
-
-	log->base = addressPXD(&sbi->logpxd);
-	log->size =
-	    (lengthPXD(&sbi->logpxd) << sb->s_blocksize_bits) >> L2LOGPSIZE;
-	rc = lmLogInit(log);
-
-	return rc;
-}
-
-
 /*
  *		log buffer manager (lbm)
  *		------------------------
@@ -2192,42 +2153,40 @@ int jfsIOWait(void *arg)
 	return 0;
 }
 
-
-#ifdef _STILL_TO_PORT
 /*
  * NAME:	lmLogFormat()/jfs_logform()
  *
- * FUNCTION:	format file system log (ref. jfs_logform()).
+ * FUNCTION:	format file system log
  *
  * PARAMETERS:
- *	log	- log inode (with common mount inode base);
- *	logAddress - start address of log space in FS block;
+ *      log	- volume log
+ *	logAddress - start address of log space in FS block
  *	logSize	- length of log space in FS block;
  *
  * RETURN:	0	- success
- *		-1 -	i/o error
+ *		-EIO	- i/o error
+ *
+ * XXX: We're synchronously writing one page at a time.  This needs to
+ *	be improved by writing multiple pages at once.
  */
-int lmLogFormat(inode_t * ipmnt, s64 logAddress, int logSize)
+int lmLogFormat(log_t *log, s64 logAddress, int logSize)
 {
-	int rc = 0;
-	cbuf_t *bp;
+	int rc = -EIO;
+	struct jfs_sb_info *sbi = JFS_SBI(log->sb);
 	logsuper_t *logsuper;
 	logpage_t *lp;
 	int lspn;		/* log sequence page number */
 	struct lrd *lrd_ptr;
-	int npbperpage, npages;
+	int npages = 0;
+	lbuf_t *bp;
 
 	jFYI(0, ("lmLogFormat: logAddress:%Ld logSize:%d\n",
-		 logAddress, logSize));
-
-	/* allocate a JFS buffer */
-	bp = rawAllocate();
+		 (long long)logAddress, logSize));
 
-	/* map the logical block address to physical block address */
-	bp->cm_blkno = logAddress << ipmnt->i_l2bfactor;
+	/* allocate a log buffer */
+	bp = lbmAllocate(log, 1);
 
-	npbperpage = LOGPSIZE >> ipmnt->i_l2pbsize;
-	npages = logSize / (LOGPSIZE >> ipmnt->i_l2bsize);
+	npages = logSize >> sbi->l2nbperpage;
 
 	/*
 	 *      log space:
@@ -2241,20 +2200,22 @@ int lmLogFormat(inode_t * ipmnt, s64 logAddress, int logSize)
 	/*
 	 *      init log superblock: log page 1
 	 */
-	logsuper = (logsuper_t *) bp->cm_cdata;
+	logsuper = (logsuper_t *) bp->l_ldata;
 
 	logsuper->magic = cpu_to_le32(LOGMAGIC);
 	logsuper->version = cpu_to_le32(LOGVERSION);
 	logsuper->state = cpu_to_le32(LOGREDONE);
-	logsuper->flag = cpu_to_le32(ipmnt->i_mntflag);	/* ? */
+	logsuper->flag = cpu_to_le32(sbi->mntflag);	/* ? */
 	logsuper->size = cpu_to_le32(npages);
-	logsuper->bsize = cpu_to_le32(ipmnt->i_bsize);
-	logsuper->l2bsize = cpu_to_le32(ipmnt->i_l2bsize);
-	logsuper->end =
-	    cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
+	logsuper->bsize = cpu_to_le32(sbi->bsize);
+	logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
+	logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
 
-	bp->cm_blkno += npbperpage;
-	rawWrite(ipmnt, bp, 0);
+	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
+	bp->l_blkno = logAddress + sbi->nbperpage;
+	lbmStartIO(bp);
+	if ((rc = lbmIOWait(bp, 0)))
+		goto exit;
 
 	/*
 	 *      init pages 2 to npages-1 as log data pages:
@@ -2270,7 +2231,6 @@ int lmLogFormat(inode_t * ipmnt, s64 logAddress, int logSize)
 	 * a circular file for the log records;
 	 * lpsn grows by 1 monotonically as each log page is written
 	 * to the circular file of the log;
-	 * Since the AIX DUMMY log record is dropped for this XJFS,
 	 * and setLogpage() will not reset the page number even if
 	 * the eor is equal to LOGPHDRSIZE. In order for binary search
 	 * still work in find log end process, we have to simulate the
@@ -2279,8 +2239,7 @@ int lmLogFormat(inode_t * ipmnt, s64 logAddress, int logSize)
 	 * the succeeding log pages will have ascending order of
 	 * the lspn starting from 0, ... (N-2)
 	 */
-	lp = (logpage_t *) bp->cm_cdata;
-
+	lp = (logpage_t *) bp->l_ldata;
 	/*
 	 * initialize 1st log page to be written: lpsn = N - 1,
 	 * write a SYNCPT log record is written to this page
@@ -2295,8 +2254,11 @@ int lmLogFormat(inode_t * ipmnt, s64 logAddress, int logSize)
 	lrd_ptr->length = 0;
 	lrd_ptr->log.syncpt.sync = 0;
 
-	bp->cm_blkno += npbperpage;
-	rawWrite(ipmnt, bp, 0);
+	bp->l_blkno += sbi->nbperpage;
+	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
+	lbmStartIO(bp);
+	if ((rc = lbmIOWait(bp, 0)))
+		goto exit;
 
 	/*
 	 *      initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
@@ -2305,20 +2267,23 @@ int lmLogFormat(inode_t * ipmnt, s64 logAddress, int logSize)
 		lp->h.page = lp->t.page = cpu_to_le32(lspn);
 		lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
 
-		bp->cm_blkno += npbperpage;
-		rawWrite(ipmnt, bp, 0);
+		bp->l_blkno += sbi->nbperpage;
+		bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
+		lbmStartIO(bp);
+		if ((rc = lbmIOWait(bp, 0)))
+			goto exit;
 	}
 
+	rc = 0;
+exit:
 	/*
 	 *      finalize log
 	 */
 	/* release the buffer */
-	rawRelease(bp);
+	lbmFree(bp);
 
 	return rc;
 }
-#endif				/* _STILL_TO_PORT */
-
 
 #ifdef CONFIG_JFS_STATISTICS
 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,

fs/jfs/jfs_logmgr.h

@@ -379,8 +379,7 @@ typedef struct jfs_log {
 	int size;		/* 4: log size in log page (in page) */
 	int l2bsize;		/* 4: log2 of bsize */
 
-	uint flag;		/* 4: flag */
-	uint state;		/* 4: state */
+	long flag;		/* 4: flag */
 
 	struct lbuf *lbuf_free;	/* 4: free lbufs */
 	wait_queue_head_t free_wait;	/* 4: */
@@ -396,7 +395,6 @@ typedef struct jfs_log {
 	/* syncpt */
 	int nextsync;		/* 4: bytes to write before next syncpt */
 	int active;		/* 4: */
-	int syncbarrier;	/* 4: */
 	wait_queue_head_t syncwait;	/* 4: */
 
 	/* commit */
@@ -420,6 +418,13 @@ typedef struct jfs_log {
 	char uuid[16];		/* 16: 128-bit uuid of log device */
 } log_t;
 
+/*
+ * Log flag
+ */
+#define log_INLINELOG	1
+#define log_SYNCBARRIER	2
+#define log_QUIESCE	3
+
 /*
  * group commit flag
  */
@@ -499,8 +504,8 @@ extern int lmLogOpen(struct super_block *sb, log_t ** log);
 extern void lmLogWait(log_t * log);
 extern int lmLogClose(struct super_block *sb, log_t * log);
 extern int lmLogSync(log_t * log, int nosyncwait);
-extern int lmLogQuiesce(log_t * log);
-extern int lmLogResume(log_t * log, struct super_block *sb);
-extern int lmLogFormat(struct super_block *sb, s64 logAddress, int logSize);
+extern int lmLogShutdown(log_t * log);
+extern int lmLogInit(log_t * log);
+extern int lmLogFormat(log_t *log, s64 logAddress, int logSize);
 
 #endif				/* _H_JFS_LOGMGR */

fs/jfs/jfs_metapage.c

@@ -20,6 +20,7 @@
 #include <linux/fs.h>
 #include <linux/init.h>
 #include <linux/buffer_head.h>
+#include <linux/mempool.h>
 #include "jfs_incore.h"
 #include "jfs_filsys.h"
 #include "jfs_metapage.h"
@@ -27,11 +28,6 @@
 #include "jfs_debug.h"
 
 extern struct task_struct *jfsCommitTask;
-static unsigned int metapages = 1024;	/* ??? Need a better number */
-static unsigned int free_metapages;
-static metapage_t *metapage_buf;
-static unsigned long meta_order;
-static metapage_t *meta_free_list = NULL;
 static spinlock_t meta_lock = SPIN_LOCK_UNLOCKED;
 static wait_queue_head_t meta_wait;
 
@@ -93,12 +89,51 @@ static inline void lock_metapage(struct metapage *mp)
 		__lock_metapage(mp);
 }
 
-int __init metapage_init(void)
+#define METAPOOL_MIN_PAGES 32
+static kmem_cache_t *metapage_cache;
+static mempool_t *metapage_mempool;
+
+static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 {
-	int i;
-	metapage_t *last = NULL;
-	metapage_t *mp;
+	metapage_t *mp = (metapage_t *)foo;
 
+	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+	    SLAB_CTOR_CONSTRUCTOR) {
+		mp->lid = 0;
+		mp->lsn = 0;
+		mp->flag = 0;
+		mp->data = NULL;
+		mp->clsn = 0;
+		mp->log = NULL;
+		set_bit(META_free, &mp->flag);
+		init_waitqueue_head(&mp->wait);
+	}
+}
+
+static inline metapage_t *alloc_metapage(int no_wait)
+{
+	return mempool_alloc(metapage_mempool, no_wait ? GFP_ATOMIC : GFP_NOFS);
+}
+
+static inline void free_metapage(metapage_t *mp)
+{
+	mp->flag = 0;
+	set_bit(META_free, &mp->flag);
+
+	mempool_free(mp, metapage_mempool);
+}
+
+static void *mp_mempool_alloc(int gfp_mask, void *pool_data)
+{
+	return kmem_cache_alloc(metapage_cache, gfp_mask);
+}
+static void mp_mempool_free(void *element, void *pool_data)
+{
+	return kmem_cache_free(metapage_cache, element);
+}
+
+int __init metapage_init(void)
+{
 	/*
 	 * Initialize wait queue
 	 */
@@ -107,30 +142,18 @@ int __init metapage_init(void)
 	/*
 	 * Allocate the metapage structures
 	 */
-	for (meta_order = 0;
-	     ((PAGE_SIZE << meta_order) / sizeof(metapage_t)) < metapages;
-	     meta_order++);
-	metapages = (PAGE_SIZE << meta_order) / sizeof(metapage_t);
-
-	jFYI(1, ("metapage_init: metapage size = %Zd, metapages = %d\n",
-		 sizeof(metapage_t), metapages));
+	metapage_cache = kmem_cache_create("jfs_mp", sizeof(metapage_t), 0, 0,
+					   init_once, NULL);
+	if (metapage_cache == NULL)
+		return -ENOMEM;
 
-	metapage_buf =
-	    (metapage_t *) __get_free_pages(GFP_KERNEL, meta_order);
-	assert(metapage_buf);
-	memset(metapage_buf, 0, PAGE_SIZE << meta_order);
+	metapage_mempool = mempool_create(METAPOOL_MIN_PAGES, mp_mempool_alloc,
+					  mp_mempool_free, NULL);
 
-	mp = metapage_buf;
-	for (i = 0; i < metapages; i++, mp++) {
-		mp->flag = 0;
-		set_bit(META_free, &mp->flag);
-		init_waitqueue_head(&mp->wait);
-		mp->hash_next = last;
-		last = mp;
+	if (metapage_mempool == NULL) {
+		kmem_cache_destroy(metapage_cache);
+		return -ENOMEM;
 	}
-	meta_free_list = last;
-	free_metapages = metapages;
-
 	/*
 	 * Now the hash list
 	 */
@@ -147,64 +170,8 @@ int __init metapage_init(void)
 
 void metapage_exit(void)
 {
-	free_pages((unsigned long) metapage_buf, meta_order);
-	free_pages((unsigned long) hash_table, hash_order);
-	metapage_buf = 0;	/* This is a signal to the jfsIOwait thread */
-}
-
-/*
- * Get metapage structure from freelist
- * 
- * Caller holds meta_lock
- */
-static metapage_t *alloc_metapage(int *dropped_lock)
-{
-	metapage_t *new;
-
-	*dropped_lock = FALSE;
-
-	/*
-	 * Reserve two metapages for the lazy commit thread.  Otherwise
-	 * we may deadlock with holders of metapages waiting for tlocks
-	 * that lazy thread should be freeing.
-	 */
-	if ((free_metapages < 3) && (current != jfsCommitTask)) {
-		INCREMENT(mpStat.allocwait);
-		*dropped_lock = TRUE;
-		__SLEEP_COND(meta_wait, (free_metapages > 2),
-			     spin_lock(&meta_lock), spin_unlock(&meta_lock));
-	}
-
-	assert(meta_free_list);
-
-	new = meta_free_list;
-	meta_free_list = new->hash_next;
-	free_metapages--;
-
-	return new;
-}
-
-/*
- * Put metapage on freelist (holding meta_lock)
- */
-static inline void __free_metapage(metapage_t * mp)
-{
-	mp->flag = 0;
-	set_bit(META_free, &mp->flag);
-	mp->hash_next = meta_free_list;
-	meta_free_list = mp;
-	free_metapages++;
-	wake_up(&meta_wait);
-}
-
-/*
- * Put metapage on freelist (not holding meta_lock)
- */
-static inline void free_metapage(metapage_t * mp)
-{
-	spin_lock(&meta_lock);
-	__free_metapage(mp);
-	spin_unlock(&meta_lock);
+	mempool_destroy(metapage_mempool);
+	kmem_cache_destroy(metapage_cache);
 }
 
 /*
@@ -295,19 +262,18 @@ static int direct_bmap(struct address_space *mapping, long block)
 }
 
 struct address_space_operations direct_aops = {
-	readpage:	direct_readpage,
-	writepage:	direct_writepage,
-	sync_page:	block_sync_page,
-	prepare_write:	direct_prepare_write,
-	commit_write:	generic_commit_write,
-	bmap:		direct_bmap,
+	.readpage	= direct_readpage,
+	.writepage	= direct_writepage,
+	.sync_page	= block_sync_page,
+	.prepare_write	= direct_prepare_write,
+	.commit_write	= generic_commit_write,
+	.bmap		= direct_bmap,
 };
 
 metapage_t *__get_metapage(struct inode *inode,
 			   unsigned long lblock, unsigned int size,
 			   int absolute, unsigned long new)
 {
-	int dropped_lock;
 	metapage_t **hash_ptr;
 	int l2BlocksPerPage;
 	int l2bsize;
@@ -354,16 +320,42 @@ metapage_t *__get_metapage(struct inode *inode,
 			return NULL;
 		}
 		
-		mp = alloc_metapage(&dropped_lock);
-		if (dropped_lock) {
-			/* alloc_metapage blocked, we need to search the hash
-			 * again.  (The goto is ugly, maybe we'll clean this
-			 * up in the future.)
+		/*
+		 * Locks held on aggregate inode pages are usually
+		 * not held long, and they are taken in critical code
+		 * paths (committing dirty inodes, txCommit thread) 
+		 * 
+		 * Attempt to get metapage without blocking, tapping into
+		 * reserves if necessary.
+		 */
+		mp = NULL;
+		if (JFS_IP(inode)->fileset == AGGREGATE_I) {
+			mp =  mempool_alloc(metapage_mempool, GFP_ATOMIC);
+			if (!mp) {
+				/*
+				 * mempool is supposed to protect us from
+				 * failing here.  We will try a blocking
+				 * call, but a deadlock is possible here
 				 */
+				printk(KERN_WARNING
+				       "__get_metapage: atomic call to mempool_alloc failed.\n");
+				printk(KERN_WARNING
+				       "Will attempt blocking call\n");
+			}
+		}
+		if (!mp) {
 			metapage_t *mp2;
+
+			spin_unlock(&meta_lock);
+			mp =  mempool_alloc(metapage_mempool, GFP_NOFS);
+			spin_lock(&meta_lock);
+
+			/* we dropped the meta_lock, we need to search the
+			 * hash again.
+			 */
 			mp2 = search_hash(hash_ptr, mapping, lblock);
 			if (mp2) {
-				__free_metapage(mp);
+				free_metapage(mp);
 				mp = mp2;
 				goto page_found;
 			}
@@ -416,7 +408,7 @@ metapage_t *__get_metapage(struct inode *inode,
 	remove_from_hash(mp, hash_ptr);
 	if (!absolute)
 		list_del(&mp->inode_list);
-	__free_metapage(mp);
+	free_metapage(mp);
 	spin_unlock(&meta_lock);
 	return NULL;
 }
@@ -631,12 +623,10 @@ int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length,
 	len += sprintf(buffer,
 		       "JFS Metapage statistics\n"
 		       "=======================\n"
-		       "metapages in use = %d\n"
 		       "page allocations = %d\n"
 		       "page frees = %d\n"
 		       "lock waits = %d\n"
 		       "allocation waits = %d\n",
-		       metapages - free_metapages,
 		       mpStat.pagealloc,
 		       mpStat.pagefree,
 		       mpStat.lockwait,

fs/jfs/jfs_mount.c

@@ -95,7 +95,7 @@ int jfs_mount(struct super_block *sb)
 		goto errout20;
 	}
 
-	ipaimap = diReadSpecial(sb, AGGREGATE_I);
+	ipaimap = diReadSpecial(sb, AGGREGATE_I, 0);
 	if (ipaimap == NULL) {
 		jERROR(1, ("jfs_mount: Faild to read AGGREGATE_I\n"));
 		rc = EIO;
@@ -118,7 +118,7 @@ int jfs_mount(struct super_block *sb)
 	/*
 	 * open aggregate block allocation map
 	 */
-	ipbmap = diReadSpecial(sb, BMAP_I);
+	ipbmap = diReadSpecial(sb, BMAP_I, 0);
 	if (ipbmap == NULL) {
 		rc = EIO;
 		goto errout22;
@@ -148,7 +148,7 @@ int jfs_mount(struct super_block *sb)
 	 * table.
 	 */
 	if ((sbi->mntflag & JFS_BAD_SAIT) == 0) {
-		ipaimap2 = diReadSpecial(sb, AGGREGATE_I + INOSPEREXT);
+		ipaimap2 = diReadSpecial(sb, AGGREGATE_I, 1);
 		if (ipaimap2 == 0) {
 			jERROR(1,
 			       ("jfs_mount: Faild to read AGGREGATE_I\n"));
@@ -178,7 +178,7 @@ int jfs_mount(struct super_block *sb)
 	/*
 	 * open fileset inode allocation map (aka fileset inode)
 	 */
-	ipimap = diReadSpecial(sb, FILESYSTEM_I);
+	ipimap = diReadSpecial(sb, FILESYSTEM_I, 0);
 	if (ipimap == NULL) {
 		jERROR(1, ("jfs_mount: Failed to read FILESYSTEM_I\n"));
 		/* open fileset secondary inode allocation map */
@@ -410,6 +410,7 @@ static int chkSuper(struct super_block *sb)
 		memcpy(sbi->uuid, j_sb->s_uuid, sizeof(sbi->uuid));
 		memcpy(sbi->loguuid, j_sb->s_loguuid, sizeof(sbi->uuid));
 	}
+	sbi->fsckpxd = j_sb->s_fsckpxd;
 	sbi->ait2 = j_sb->s_ait2;
 
       out:

fs/jfs/jfs_txnmgr.c

@@ -318,11 +318,12 @@ tid_t txBegin(struct super_block *sb, int flag)
 	TXN_LOCK();
 
       retry:
-	if (flag != COMMIT_FORCE) {
+	if (!(flag & COMMIT_FORCE)) {
 		/*
 		 * synchronize with logsync barrier
 		 */
-		if (log->syncbarrier) {
+		if (test_bit(log_SYNCBARRIER, &log->flag) ||
+		    test_bit(log_QUIESCE, &log->flag)) {
 			TXN_SLEEP(&log->syncwait);
 			goto retry;
 		}
@@ -330,8 +331,8 @@ tid_t txBegin(struct super_block *sb, int flag)
 	if (flag == 0) {
 		/*
 		 * Don't begin transaction if we're getting starved for tlocks
-		 * unless COMMIT_FORCE (imap changes) or COMMIT_INODE (which
-		 * may ultimately free tlocks)
+		 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
+		 * free tlocks)
 		 */
 		if (TlocksLow) {
 			TXN_SLEEP(&TxAnchor.lowlockwait);
@@ -411,7 +412,8 @@ void txBeginAnon(struct super_block *sb)
 	/*
 	 * synchronize with logsync barrier
 	 */
-	if (log->syncbarrier) {
+	if (test_bit(log_SYNCBARRIER, &log->flag) ||
+	    test_bit(log_QUIESCE, &log->flag)) {
 		TXN_SLEEP(&log->syncwait);
 		goto retry;
 	}
@@ -490,14 +492,14 @@ void txEnd(tid_t tid)
 	/*
 	 * synchronize with logsync barrier
 	 */
-	if (log->syncbarrier && log->active == 0) {
+	if (test_bit(log_SYNCBARRIER, &log->flag) && log->active == 0) {
 		/* forward log syncpt */
 		/* lmSync(log); */
 
 		jFYI(1, ("     log barrier off: 0x%x\n", log->lsn));
 
 		/* enable new transactions start */
-		log->syncbarrier = 0;
+		clear_bit(log_SYNCBARRIER, &log->flag);
 
 		/* wakeup all waitors for logsync barrier */
 		TXN_WAKEUP(&log->syncwait);
@@ -823,36 +825,21 @@ static void txRelease(tblock_t * tblk)
  *
  * FUNCTION:    Initiates pageout of pages modified by tid in journalled
  *              objects and frees their lockwords.
- *
- * PARAMETER:
- *              flag    -
- *
- * RETURN:      Errors from subroutines.
  */
-static void txUnlock(tblock_t * tblk, int flag)
+static void txUnlock(tblock_t * tblk)
 {
 	tlock_t *tlck;
 	linelock_t *linelock;
 	lid_t lid, next, llid, k;
 	metapage_t *mp;
 	log_t *log;
-	int force;
 	int difft, diffp;
 
 	jFYI(1, ("txUnlock: tblk = 0x%p\n", tblk));
 	log = (log_t *) JFS_SBI(tblk->sb)->log;
-	force = flag & COMMIT_FLUSH;
-	if (log->syncbarrier)
-		force |= COMMIT_FORCE;
 
 	/*
 	 * mark page under tlock homeok (its log has been written):
-	 * if caller has specified FORCE (e.g., iRecycle()), or
-	 * if syncwait for the log is set (i.e., the log sync point
-	 * has fallen behind), or
-	 * if syncpt is set for the page, or
-	 * if the page is new, initiate pageout;
-	 * otherwise, leave the page in memory.
 	 */
 	for (lid = tblk->next; lid; lid = next) {
 		tlck = lid_to_tlock(lid);
@@ -1268,7 +1255,7 @@ int txCommit(tid_t tid,		/* transaction identifier */
 	txRelease(tblk);
 
 	if ((tblk->flag & tblkGC_LAZY) == 0)
-		txUnlock(tblk, flag);
+		txUnlock(tblk);
 
 
 	/*
@@ -2753,7 +2740,7 @@ void txLazyCommit(tblock_t * tblk)
 	spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
 
 	if (tblk->flag & tblkGC_LAZY) {
-		txUnlock(tblk, 0);
+		txUnlock(tblk);
 		tblk->flag &= ~tblkGC_LAZY;
 		txEnd(tblk - TxBlock);	/* Convert back to tid */
 	}
@@ -2887,6 +2874,77 @@ static void LogSyncRelease(metapage_t * mp)
 	release_metapage(mp);
 }
 
+/*
+ *	txQuiesce
+ *
+ *	Block all new transactions and push anonymous transactions to
+ *	completion
+ *
+ *	This does almost the same thing as jfs_sync below.  We don't
+ *	worry about deadlocking when TlocksLow is set, since we would
+ *	expect jfs_sync to get us out of that jam.
+ */
+void txQuiesce(struct super_block *sb)
+{
+	struct inode *ip;
+	struct jfs_inode_info *jfs_ip;
+	log_t *log = JFS_SBI(sb)->log;
+	int rc;
+	tid_t tid;
+
+	set_bit(log_QUIESCE, &log->flag);
+
+	TXN_LOCK();
+restart:
+	while (!list_empty(&TxAnchor.anon_list)) {
+		jfs_ip = list_entry(TxAnchor.anon_list.next,
+				    struct jfs_inode_info,
+				    anon_inode_list);
+		ip = &jfs_ip->vfs_inode;
+
+		/*
+		 * inode will be removed from anonymous list
+		 * when it is committed
+		 */
+		TXN_UNLOCK();
+		tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
+		down(&jfs_ip->commit_sem);
+		rc = txCommit(tid, 1, &ip, 0);
+		txEnd(tid);
+		up(&jfs_ip->commit_sem);
+		/*
+		 * Just to be safe.  I don't know how
+		 * long we can run without blocking
+		 */
+		cond_resched();
+		TXN_LOCK();
+	}
+
+	/*
+	 * If jfs_sync is running in parallel, there could be some inodes
+	 * on anon_list2.  Let's check.
+	 */
+	if (!list_empty(&TxAnchor.anon_list2)) {
+		list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list);
+		INIT_LIST_HEAD(&TxAnchor.anon_list2);
+		goto restart;
+	}
+	TXN_UNLOCK();
+}
+
+/*
+ * txResume()
+ *
+ * Allows transactions to start again following txQuiesce
+ */
+void txResume(struct super_block *sb)
+{
+	log_t *log = JFS_SBI(sb)->log;
+
+	clear_bit(log_QUIESCE, &log->flag);
+	TXN_WAKEUP(&log->syncwait);
+}
+
 /*
  *      jfs_sync(void)
  *
@@ -2898,6 +2956,8 @@ int jfs_sync(void)
 {
 	struct inode *ip;
 	struct jfs_inode_info *jfs_ip;
+	int rc;
+	tid_t tid;
 
 	lock_kernel();
 
@@ -2927,17 +2987,20 @@ int jfs_sync(void)
 			ip = &jfs_ip->vfs_inode;
 
 			/*
-			 * We must release the TXN_LOCK since our
-			 * IWRITE_TRYLOCK implementation may still block
+			 * down_trylock returns 0 on success.  This is
+			 * inconsistent with spin_trylock.
 			 */
-			TXN_UNLOCK();
-			if (IWRITE_TRYLOCK(ip)) {
+			if (! down_trylock(&jfs_ip->commit_sem)) {
 				/*
 				 * inode will be removed from anonymous list
 				 * when it is committed
 				 */
-				jfs_commit_inode(ip, 0);
-				IWRITE_UNLOCK(ip);
+				TXN_UNLOCK();
+				tid = txBegin(ip->i_sb,
+					      COMMIT_INODE | COMMIT_FORCE);
+				rc = txCommit(tid, 1, &ip, 0);
+				txEnd(tid);
+				up(&jfs_ip->commit_sem);
 				/*
 				 * Just to be safe.  I don't know how
 				 * long we can run without blocking
@@ -2945,21 +3008,12 @@ int jfs_sync(void)
 				cond_resched();
 				TXN_LOCK();
 			} else {
-				/* We can't get the write lock.  It may
+				/* We can't get the commit semaphore.  It may
 				 * be held by a thread waiting for tlock's
 				 * so let's not block here.  Save it to
 				 * put back on the anon_list.
 				 */
 
-				/*
-				 * We released TXN_LOCK, let's make sure
-				 * this inode is still there
-				 */
-				TXN_LOCK();
-				if (TxAnchor.anon_list.next !=
-				    &jfs_ip->anon_inode_list)
-					continue;
-
 				/* Take off anon_list */
 				list_del(&jfs_ip->anon_inode_list);
 

fs/jfs/jfs_txnmgr.h

@@ -310,4 +310,7 @@ extern void txFreelock(struct inode *ip);
 
 extern int lmLog(log_t * log, tblock_t * tblk, lrd_t * lrd, tlock_t * tlck);
 
+extern void txQuiesce(struct super_block *sb);
+
+extern void txResume(struct super_block *sb);
 #endif				/* _H_JFS_TXNMGR */

fs/jfs/jfs_xtree.c

@@ -2373,7 +2373,6 @@ printf("xtUpdate.updateLeft.split p:0x%p\n", p);
 }
 
 
-#ifdef _STILL_TO_PORT
 /*
  *      xtAppend()
  *
@@ -2392,7 +2391,7 @@ printf("xtUpdate.updateLeft.split p:0x%p\n", p);
  * return:
  */
 int xtAppend(tid_t tid,		/* transaction id */
-	     struct inode *ip, int xflag, s64 xoff, s32 maxblocks,	/* @GD1 */
+	     struct inode *ip, int xflag, s64 xoff, s32 maxblocks,	
 	     s32 * xlenp,	/* (in/out) */
 	     s64 * xaddrp,	/* (in/out) */
 	     int flag)
@@ -2460,7 +2459,7 @@ int xtAppend(tid_t tid,		/* transaction id */
 	pxdlist.maxnpxd = pxdlist.npxd = 0;
 	pxd = &pxdlist.pxd[0];
 	nblocks = JFS_SBI(ip->i_sb)->nbperpage;
-	for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) {	/* @GD1 */
+	for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) {	
 		if ((rc = dbAllocBottomUp(ip, xaddr, (s64) nblocks)) == 0) {
 			PXDaddress(pxd, xaddr);
 			PXDlength(pxd, nblocks);
@@ -2475,7 +2474,7 @@ int xtAppend(tid_t tid,		/* transaction id */
 		goto out;
 	}
 
-	xlen = min(xlen, maxblocks);	/* @GD1 */
+	xlen = min(xlen, maxblocks);	
 
 	/*
 	 * allocate data extent requested
@@ -2528,7 +2527,7 @@ int xtAppend(tid_t tid,		/* transaction id */
 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
 
 	xtlck->lwm.offset =
-	    (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index;
+	    (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index;
 	xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
 	    xtlck->lwm.offset;
 
@@ -2541,7 +2540,7 @@ int xtAppend(tid_t tid,		/* transaction id */
 
 	return rc;
 }
-
+#ifdef _STILL_TO_PORT
 
 /* - TBD for defragmentaion/reorganization -
  *

fs/jfs/super.c

@@ -71,6 +71,7 @@ extern void jfs_delete_inode(struct inode *inode);
 extern void jfs_write_inode(struct inode *inode, int wait);
 
 extern struct dentry *jfs_get_parent(struct dentry *dentry);
+extern int jfs_extendfs(struct super_block *, s64, int);
 
 #ifdef PROC_FS_JFS		/* see jfs_debug.h */
 extern void jfs_proc_init(void);
@@ -119,7 +120,7 @@ static int jfs_statfs(struct super_block *sb, struct statfs *buf)
 	 */
 	maxinodes = min((s64) atomic_read(&imap->im_numinos) +
 			((sbi->bmap->db_nfree >> imap->im_l2nbperiext)
-			 << L2INOSPEREXT), (s64)0xffffffffLL);
+			 << L2INOSPEREXT), (s64) 0xffffffffLL);
 	buf->f_files = maxinodes;
 	buf->f_ffree = maxinodes - (atomic_read(&imap->im_numinos) -
 				    atomic_read(&imap->im_numfree));
@@ -156,20 +157,23 @@ static void jfs_put_super(struct super_block *sb)
 	kfree(sbi);
 }
 
-static int parse_options (char * options, struct jfs_sb_info *sbi)
+static int parse_options(char *options, struct super_block *sb, s64 *newLVSize)
 {
 	void *nls_map = NULL;
-	char * this_char;
-	char * value;
+	char *this_char;
+	char *value;
+	struct jfs_sb_info *sbi = JFS_SBI(sb);
+
+	*newLVSize = 0;
 
 	if (!options)
 		return 1;
-	while ((this_char = strsep (&options, ",")) != NULL) {
+	while ((this_char = strsep(&options, ",")) != NULL) {
 		if (!*this_char)
 			continue;
-		if ((value = strchr (this_char, '=')) != NULL)
+		if ((value = strchr(this_char, '=')) != NULL)
 			*value++ = 0;
-		if (!strcmp (this_char, "iocharset")) {
+		if (!strcmp(this_char, "iocharset")) {
 			if (!value || !*value)
 				goto needs_arg;
 			if (nls_map)	/* specified iocharset twice! */
@@ -179,14 +183,25 @@ static int parse_options (char * options, struct jfs_sb_info *sbi)
 				printk(KERN_ERR "JFS: charset not found\n");
 				goto cleanup;
 			}
+		} else if (!strcmp(this_char, "resize")) {
+			if (!value || !*value) {
+				*newLVSize = sb->s_bdev->bd_inode->i_size >>
+					sb->s_blocksize_bits;
+				if (*newLVSize == 0)
+					printk(KERN_ERR
+					 "JFS: Cannot determine volume size\n");
+			} else
+				*newLVSize = simple_strtoull(value, &value, 0);
+
 			/* Silently ignore the quota options */
-		} else if (!strcmp (this_char, "grpquota")
-		         || !strcmp (this_char, "noquota")
-		         || !strcmp (this_char, "quota")
-		         || !strcmp (this_char, "usrquota"))
+		} else if (!strcmp(this_char, "grpquota")
+			   || !strcmp(this_char, "noquota")
+			   || !strcmp(this_char, "quota")
+			   || !strcmp(this_char, "usrquota"))
 			/* Don't do anything ;-) */ ;
 		else {
-			printk ("jfs: Unrecognized mount option %s\n", this_char);
+			printk("jfs: Unrecognized mount option %s\n",
+			       this_char);
 			goto cleanup;
 		}
 	}
@@ -208,10 +223,22 @@ static int parse_options (char * options, struct jfs_sb_info *sbi)
 int jfs_remount(struct super_block *sb, int *flags, char *data)
 {
 	struct jfs_sb_info *sbi = JFS_SBI(sb);
+	s64 newLVSize = 0;
+	int rc = 0;
 
-	if (!parse_options(data, sbi)) {
+	if (!parse_options(data, sb, &newLVSize)) {
 		return -EINVAL;
 	}
+	if (newLVSize) {
+		if (sb->s_flags & MS_RDONLY) {
+			printk(KERN_ERR
+		  "JFS: resize requires volume to be mounted read-write\n");
+			return -EROFS;
+		}
+		rc = jfs_extendfs(sb, newLVSize, 0);
+		if (rc)
+			return rc;
+	}
 
 	if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
 		/*
@@ -232,20 +259,26 @@ static int jfs_fill_super(struct super_block *sb, void *data, int silent)
 	struct jfs_sb_info *sbi;
 	struct inode *inode;
 	int rc;
+	s64 newLVSize = 0;
 
 	jFYI(1, ("In jfs_read_super: s_flags=0x%lx\n", sb->s_flags));
 
-	sbi = kmalloc(sizeof(struct jfs_sb_info), GFP_KERNEL);
+	sbi = kmalloc(sizeof (struct jfs_sb_info), GFP_KERNEL);
 	if (!sbi)
 		return -ENOSPC;
-	memset(sbi, 0, sizeof(struct jfs_sb_info));
+	memset(sbi, 0, sizeof (struct jfs_sb_info));
 	sb->u.generic_sbp = sbi;
 
-	if (!parse_options((char *)data, sbi)) {
+	if (!parse_options((char *) data, sb, &newLVSize)) {
 		kfree(sbi);
 		return -EINVAL;
 	}
 
+	if (newLVSize) {
+		printk(KERN_ERR "resize option for remount only\n");
+		return -EINVAL;
+	}
+
 	/*
 	 * Initialize blocksize to 4K.
 	 */
@@ -276,8 +309,7 @@ static int jfs_fill_super(struct super_block *sb, void *data, int silent)
 	if (rc) {
 		if (!silent) {
 			jERROR(1,
-			       ("jfs_mount failed w/return code = %d\n",
-				rc));
+			       ("jfs_mount failed w/return code = %d\n", rc));
 		}
 		goto out_mount_failed;
 	}
@@ -314,7 +346,7 @@ static int jfs_fill_super(struct super_block *sb, void *data, int silent)
 	 * Page cache is indexed by long.
 	 * I would use MAX_LFS_FILESIZE, but it's only half as big
 	 */
-	sb->s_maxbytes = min(((u64)PAGE_CACHE_SIZE << 32) - 1, sb->s_maxbytes);
+	sb->s_maxbytes = min(((u64) PAGE_CACHE_SIZE << 32) - 1, sb->s_maxbytes);
 #endif
 
 	return 0;
@@ -379,15 +411,17 @@ extern int txInit(void);
 extern void txExit(void);
 extern void metapage_exit(void);
 
-static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
+static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 {
 	struct jfs_inode_info *jfs_ip = (struct jfs_inode_info *) foo;
 
-	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+	if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 	    SLAB_CTOR_CONSTRUCTOR) {
 		INIT_LIST_HEAD(&jfs_ip->anon_inode_list);
 		INIT_LIST_HEAD(&jfs_ip->mp_list);
-		RDWRLOCK_INIT(&jfs_ip->rdwrlock);
+		init_rwsem(&jfs_ip->rdwrlock);
+		init_MUTEX(&jfs_ip->commit_sem);
+		jfs_ip->atlhead = 0;
 		inode_init_once(&jfs_ip->vfs_inode);
 	}
 }
@@ -397,9 +431,8 @@ static int __init init_jfs_fs(void)
 	int rc;
 
 	jfs_inode_cachep =
-	    kmem_cache_create("jfs_ip",
-			    sizeof(struct jfs_inode_info),
-			    0, 0, init_once, NULL);
+	    kmem_cache_create("jfs_ip", sizeof(struct jfs_inode_info), 0, 0,
+			      init_once, NULL);
 	if (jfs_inode_cachep == NULL)
 		return -ENOMEM;
 
@@ -425,37 +458,32 @@ static int __init init_jfs_fs(void)
 	 * I/O completion thread (endio)
 	 */
 	jfsIOthread = kernel_thread(jfsIOWait, 0,
-				    CLONE_FS | CLONE_FILES |
-				    CLONE_SIGHAND);
+				    CLONE_FS | CLONE_FILES | CLONE_SIGHAND);
 	if (jfsIOthread < 0) {
 		jERROR(1,
-		       ("init_jfs_fs: fork failed w/rc = %d\n",
-			jfsIOthread));
+		       ("init_jfs_fs: fork failed w/rc = %d\n", jfsIOthread));
 		goto end_txmngr;
 	}
-	wait_for_completion(&jfsIOwait);	/* Wait until IO thread starts */
+	wait_for_completion(&jfsIOwait);	/* Wait until thread starts */
 
 	jfsCommitThread = kernel_thread(jfs_lazycommit, 0,
-					CLONE_FS | CLONE_FILES |
-					CLONE_SIGHAND);
+					CLONE_FS | CLONE_FILES | CLONE_SIGHAND);
 	if (jfsCommitThread < 0) {
 		jERROR(1,
 		       ("init_jfs_fs: fork failed w/rc = %d\n",
 			jfsCommitThread));
 		goto kill_iotask;
 	}
-	wait_for_completion(&jfsIOwait);	/* Wait until IO thread starts */
+	wait_for_completion(&jfsIOwait);	/* Wait until thread starts */
 
 	jfsSyncThread = kernel_thread(jfs_sync, 0,
-				      CLONE_FS | CLONE_FILES |
-				      CLONE_SIGHAND);
+				      CLONE_FS | CLONE_FILES | CLONE_SIGHAND);
 	if (jfsSyncThread < 0) {
 		jERROR(1,
-		       ("init_jfs_fs: fork failed w/rc = %d\n",
-			jfsSyncThread));
+		       ("init_jfs_fs: fork failed w/rc = %d\n", jfsSyncThread));
 		goto kill_committask;
 	}
-	wait_for_completion(&jfsIOwait);	/* Wait until IO thread starts */
+	wait_for_completion(&jfsIOwait);	/* Wait until thread starts */
 
 #ifdef PROC_FS_JFS
 	jfs_proc_init();
@@ -463,15 +491,14 @@ static int __init init_jfs_fs(void)
 
 	return register_filesystem(&jfs_fs_type);
 
-
 kill_committask:
 	jfs_stop_threads = 1;
 	wake_up(&jfs_commit_thread_wait);
-	wait_for_completion(&jfsIOwait);	/* Wait until Commit thread exits */
+	wait_for_completion(&jfsIOwait);	/* Wait for thread exit */
 kill_iotask:
 	jfs_stop_threads = 1;
 	wake_up(&jfs_IO_thread_wait);
-	wait_for_completion(&jfsIOwait);	/* Wait until IO thread exits */
+	wait_for_completion(&jfsIOwait);	/* Wait for thread exit */
 end_txmngr:
 	txExit();
 free_metapage: