Add support for external JFS journal

Submitted by Christoph Hellwig & Dave Kleikamp

fs/jfs/jfs_incore.h

@@ -123,7 +123,7 @@ struct jfs_sb_info {
 	short		nbperpage;	/* 2: blocks per page		*/
 	short		l2nbperpage;	/* 2: log2 blocks per page	*/
 	short		l2niperblk;	/* 2: log2 inodes per page	*/
-	short		reserved;	/* 2: log2 inodes per page	*/
+	kdev_t		logdev;		/* 2: external log device	*/
 	pxd_t		logpxd;		/* 8: pxd describing log	*/
 	pxd_t		ait2;		/* 8: pxd describing AIT copy	*/
 	/* Formerly in ipimap */

fs/jfs/jfs_logmgr.c

@@ -159,7 +159,6 @@ do {						\
 /*
  * external references
  */
-extern void vPut(struct inode *ip);
 extern void txLazyUnlock(tblock_t * tblk);
 extern int jfs_thread_stopped(void);
 extern struct task_struct *jfsIOtask;
@@ -172,6 +171,7 @@ static int lmWriteRecord(log_t * log, tblock_t * tblk, lrd_t * lrd,
 			 tlock_t * tlck);
 
 static int lmNextPage(log_t * log);
+static int lmLogFileSystem(log_t * log, kdev_t fsdev, int activate);
 static int lmLogInit(log_t * log);
 static int lmLogShutdown(log_t * log);
 
@@ -1057,46 +1057,36 @@ int lmLogSync(log_t * log, int nosyncwait)
 int lmLogOpen(struct super_block *sb, log_t ** logptr)
 {
 	int rc;
-	kdev_t logdev;		/* dev_t of log device */
+	struct block_device *bdev;
 	log_t *log;
 
-	logdev = sb->s_dev;
+	if (!(log = kmalloc(sizeof(log_t), GFP_KERNEL)))
+		return ENOMEM;
+	memset(log, 0, sizeof(log_t));
 
-#ifdef _STILL_TO_PORT
-	/*
-	 * open the inode representing the log device (aka log inode)
-	 */
-	if (logdev != fsdev)
+	if (!(JFS_SBI(sb)->mntflag & JFS_INLINELOG))
 		goto externalLog;
-#endif				/* _STILL_TO_PORT */
 
 	/*
 	 *      in-line log in host file system
 	 *
 	 * file system to log have 1-to-1 relationship;
 	 */
-//    inlineLog:
-
-	*logptr = log = kmalloc(sizeof(log_t), GFP_KERNEL);
-	if (log == 0)
-		return ENOMEM;
 
-	memset(log, 0, sizeof(log_t));
 	log->sb = sb;		/* This should be a list */
 	log->flag = JFS_INLINELOG;
-	log->dev = logdev;
+	log->dev = sb->s_dev;
 	log->base = addressPXD(&JFS_SBI(sb)->logpxd);
 	log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
 	    (L2LOGPSIZE - sb->s_blocksize_bits);
 	log->l2bsize = sb->s_blocksize_bits;
 	ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
+
 	/*
 	 * initialize log.
 	 */
 	if ((rc = lmLogInit(log)))
 		goto errout10;
-
-#ifdef _STILL_TO_PORT
 	goto out;
 
 	/*
@@ -1105,103 +1095,47 @@ int lmLogOpen(struct super_block *sb, log_t ** logptr)
 	 * file systems to log may have n-to-1 relationship;
 	 */
       externalLog:
-	/*
-	 * open log inode
-	 *
-	 * log inode is reserved inode of (dev_t = log device,
-	 * fileset number = 0, i_number = 0), which acquire
-	 * one i_count for each open by file system.
-	 *
-	 * hand craft dummy vfs to force iget() the special case of
-	 * an in-memory inode allocation without on-disk inode
-	 */
-	memset(&dummyvfs, 0, sizeof(struct vfs));
-	dummyvfs.filesetvfs.vfs_data = NULL;
-	dummyvfs.dummyvfs.dev = logdev;
-	dummyvfs.dummyvfs.ipmnt = NULL;
-	ICACHE_LOCK();
-	rc = iget((struct vfs *) &dummyvfs, 0, (inode_t **) & log, 0);
-	ICACHE_UNLOCK();
-	if (rc)
-		return rc;
-
-	log->flag = 0;
-	log->dev = logdev;
-	log->base = 0;
-	log->size = 0;
-
-	/*
-	 * serialize open/close between multiple file systems
-	 * bound with the log;
-	 */
-	ip = (inode_t *) log;
-	IWRITE_LOCK(ip);
-
-	/*
-	 * subsequent open: add file system to log active file system list
-	 */
-#ifdef	_JFS_OS2
-	if (log->strat2p)
-#endif				/* _JFS_OS2 */
-	{
-		if (rc = lmLogFileSystem(log, fsdev, 1))
-			goto errout10;
-
-		IWRITE_UNLOCK(ip);
-
-		*iplog = ip;
-		jFYI(1, ("lmLogOpen: exit(0)\n"));
-		return 0;
+	if (!(bdev = bdget(kdev_t_to_nr(JFS_SBI(sb)->logdev)))) {
+		rc = ENODEV;
+		goto errout10;
 	}
 
-	/* decouple log inode from dummy vfs */
-	vPut(ip);
-
-	/*
-	 * first open:
-	 */
-#ifdef	_JFS_OS2
-	/*
-	 * establish access to the single/shared (already open) log device
-	 */
-	logdevfp = (void *) logStrat2;
-	log->strat2p = logStrat2;
-	log->strat3p = logStrat3;
-
-	log->l2pbsize = 9;	/* todo: when OS/2 have multiple external log */
-#endif				/* _JFS_OS2 */
+	if ((rc = blkdev_get(bdev, FMODE_READ|FMODE_WRITE, 0, BDEV_FS))) {
+		rc = -rc;
+		goto errout10;
+	}
 
+	log->sb = sb;		/* This should be a list */
+	log->dev = JFS_SBI(sb)->logdev;
+	log->bdev = bdev;
+	
 	/*
 	 * initialize log:
 	 */
-	if (rc = lmLogInit(log))
+	if ((rc = lmLogInit(log)))
 		goto errout20;
 
 	/*
 	 * add file system to log active file system list
 	 */
-	if (rc = lmLogFileSystem(log, fsdev, 1))
+	if ((rc = lmLogFileSystem(log, sb->s_dev, 1)))
 		goto errout30;
 
-	/*
-	 *      insert log device into log device list
-	 */
       out:
-#endif				/*  _STILL_TO_PORT */
 	jFYI(1, ("lmLogOpen: exit(0)\n"));
+	*logptr = log;
 	return 0;
 
 	/*
 	 *      unwind on error
 	 */
-#ifdef _STILL_TO_PORT
       errout30:		/* unwind lbmLogInit() */
 	lbmLogShutdown(log);
 
       errout20:		/* close external log device */
+	blkdev_put(bdev, BDEV_FS);
 
-#endif				/* _STILL_TO_PORT */
-      errout10:		/* free log inode */
+      errout10:		/* free log descriptor */
 	kfree(log);
 
 	jFYI(1, ("lmLogOpen: exit(%d)\n", rc));
@@ -1253,6 +1187,10 @@ static int lmLogInit(log_t * log)
 	/*
 	 * validate log superblock
 	 */
+
+
+	if (!(log->flag & JFS_INLINELOG))
+		log->l2bsize = 12;	/* XXX kludge alert XXX */
 	if ((rc = lbmRead(log, 1, &bpsuper)))
 		goto errout10;
 
@@ -1282,6 +1220,7 @@ static int lmLogInit(log_t * log)
 		      log, (unsigned long long) log->base, log->size));
 	} else {
 		log->size = le32_to_cpu(logsuper->size);
+		log->l2bsize = le32_to_cpu(logsuper->l2bsize);
 		jFYI(0,
 		     ("lmLogInit: external log:0x%p base:0x%Lx size:0x%x\n",
 		      log, (unsigned long long) log->base, log->size));
@@ -1376,6 +1315,7 @@ static int lmLogInit(log_t * log)
 	logsuper->state = cpu_to_le32(LOGMOUNT);
 	log->serial = le32_to_cpu(logsuper->serial) + 1;
 	logsuper->serial = cpu_to_le32(log->serial);
+	logsuper->device = cpu_to_le32(kdev_t_to_nr(log->dev));
 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
 	if ((rc = lbmIOWait(bpsuper, lbmFREE)))
 		goto errout30;
@@ -1419,47 +1359,24 @@ int lmLogClose(struct super_block *sb, log_t * log)
 
 	jFYI(1, ("lmLogClose: log:0x%p\n", log));
 
+	if (!(log->flag & JFS_INLINELOG))
+		goto externalLog;
+	
 	/*
 	 *      in-line log in host file system
 	 */
-//      inlineLog:
-#ifdef _STILL_TO_PORT
-	if (log->flag & JFS_INLINELOG) {
-		rc = lmLogShutdown(log);
-
-		goto out1;
-	}
+	rc = lmLogShutdown(log);
+	goto out;
 
 	/*
 	 *      external log as separate logical volume
 	 */
       externalLog:
-
-	/* serialize open/close between multiple file systems
-	 * associated with the log
-	 */
-	IWRITE_LOCK(iplog);
-
-	/*
-	 * remove file system from log active file system list
-	 */
-	rc = lmLogFileSystem(log, fsdev, 0);
-
-	if (iplog->i_count > 1)
-		goto out2;
-
-	/*
-	 *      last close: shut down log
-	 */
-	rc = ((rc1 = lmLogShutdown(log)) && rc == 0) ? rc1 : rc;
-
-      out1:
-#else				/* _STILL_TO_PORT */
+	lmLogFileSystem(log, sb->s_dev, 0);
 	rc = lmLogShutdown(log);
-#endif				/* _STILL_TO_PORT */
-
-//      out2:
+	blkdev_put(log->bdev, BDEV_FS);
 
+      out:
 	jFYI(0, ("lmLogClose: exit(%d)\n", rc));
 	return rc;
 }
@@ -1561,7 +1478,6 @@ static int lmLogShutdown(log_t * log)
 }
 
 
-#ifdef _STILL_TO_PORT
 /*
  * NAME:	lmLogFileSystem()
  *
@@ -1569,7 +1485,7 @@ static int lmLogShutdown(log_t * log)
  *	file system into/from log active file system list.
  *
  * PARAMETE:	log	- pointer to logs inode.
- *		fsdev	- dev_t of filesystem.
+ *		fsdev	- kdev_t of filesystem.
  *		serial  - pointer to returned log serial number
  *		activate - insert/remove device from active list.
  *
@@ -1578,10 +1494,11 @@ static int lmLogShutdown(log_t * log)
  *			
  * serialization: IWRITE_LOCK(log inode) held on entry/exit
  */
-static int lmLogFileSystem(log_t * log, dev_t fsdev, int activate)
+static int lmLogFileSystem(log_t * log, kdev_t fsdev, int activate)
 {
 	int rc = 0;
-	int bit, word;
+	int i;
+	u32 dev_le = cpu_to_le32(kdev_t_to_nr(fsdev));
 	logsuper_t *logsuper;
 	lbuf_t *bpsuper;
 
@@ -1592,15 +1509,25 @@ static int lmLogFileSystem(log_t * log, dev_t fsdev, int activate)
 		return rc;
 
 	logsuper = (logsuper_t *) bpsuper->l_ldata;
-	bit = minor(fsdev);
-	word = bit / 32;
-	bit -= 32 * word;
-	if (activate)
-		logsuper->active[word] |=
-		    cpu_to_le32((LEFTMOSTONE >> bit));
-	else
-		logsuper->active[word] &=
-		    cpu_to_le32((~(LEFTMOSTONE >> bit)));
+	if (activate) {
+		for (i = 0; i < MAX_ACTIVE; i++)
+			if (logsuper->active[i] == 0) {
+				logsuper->active[i] = dev_le;
+				break;
+			}
+		if (i == MAX_ACTIVE) {
+			jERROR(1,("Too many file systems sharing journal!\n"));
+			lbmFree(bpsuper);
+			return EMFILE;	/* Is there a better rc? */
+		}
+	} else {
+		for (i = 0; i < MAX_ACTIVE; i++)
+			if (logsuper->active[i] == dev_le) {
+				logsuper->active[i] = 0;
+				break;
+			}
+		assert(i < MAX_ACTIVE);
+	}
 
 	/*
 	 * synchronous write log superblock:
@@ -1618,7 +1545,6 @@ static int lmLogFileSystem(log_t * log, dev_t fsdev, int activate)
 
 	return rc;
 }
-#endif				/* _STILL_TO_PORT */
 
 
 /*

fs/jfs/jfs_logmgr.h

@@ -60,6 +60,8 @@
 #define	LOGMAGIC	0x87654321
 #define	LOGVERSION	1
 
+#define MAX_ACTIVE	512	/* Max active file systems sharing log */
+
 typedef struct {
 	u32 magic;		/* 4: log lv identifier */
 	s32 version;		/* 4: version number */
@@ -72,8 +74,8 @@ typedef struct {
 	u32 state;		/* 4: state - see below */
 
 	s32 end;		/* 4: addr of last log record set by logredo */
-	u32 active[8];		/* 32: active file systems bit vector */
-	s32 rsrvd[LOGPSIZE / 4 - 17];
+	u32 device;		/* 4: save device in case location changes */
+	u32 active[MAX_ACTIVE];	/* 2048: active file systems list */
 } logsuper_t;
 
 /* log flag: commit option (see jfs_filsys.h) */
@@ -200,7 +202,7 @@ typedef struct lrd {
 	s32 backchain;		/* 4: ptr to prev record of same transaction */
 	u16 type;		/* 2: record type */
 	s16 length;		/* 2: length of data in record (in byte) */
-	s32 aggregate;		/* 4: file system lv/aggregate */
+	u32 aggregate;		/* 4: file system lv/aggregate */
 	/* (16) */
 
 	/*
@@ -367,7 +369,7 @@ typedef struct jfs_log {
 				 *    the log between fs's
 				 */
 	kdev_t dev;		/* 4: log lv number */
-	struct file *devfp;	/* 4: log device file */
+	struct block_device *bdev; /* 4: log lv pointer */
 	s32 serial;		/* 4: log mount serial number */
 
 	s64 base;		/* @8: log extent address (inline log ) */

fs/jfs/jfs_mount.c

@@ -283,24 +283,6 @@ int jfs_mount_rw(struct super_block *sb, int remount)
 			return rc;
 		}
 	}
-#ifdef _STILL_TO_PORT
-	/*
-	 * get log device associated with the fs being mounted;
-	 */
-	if (ipmnt->i_mntflag & JFS_INLINELOG) {
-		vfsp->vfs_logVPB = vfsp->vfs_hVPB;
-		vfsp->vfs_logvpfs = vfsp->vfs_vpfsi;
-	} else if (vfsp->vfs_logvpfs == NULL) {
-		/*
-		 * XXX: there's only one external log per system;
-		 */
-		jERROR(1, ("jfs_mount: Mount Failure! No Log Device.\n"));
-		goto errout30;
-	}
-
-	logdev = vfsp->vfs_logvpfs->vpi_unit;
-	ipmnt->i_logdev = logdev;
-#endif				/* _STILL_TO_PORT */
 
 	/*
 	 * open/initialize log
@@ -360,7 +342,7 @@ static int chkSuper(struct super_block *sb)
 	 */
 	/* validate fs signature */
 	if (strncmp(j_sb->s_magic, JFS_MAGIC, 4) ||
-	    j_sb->s_version != cpu_to_le32(JFS_VERSION)) {
+	    j_sb->s_version > cpu_to_le32(JFS_VERSION)) {
 		//rc = EFORMAT;
 		rc = EINVAL;
 		goto out;
@@ -398,10 +380,6 @@ static int chkSuper(struct super_block *sb)
 			j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
 	}
 
-	/* in release 1, the flag MUST reflect inline log, and group commit */
-	if ((j_sb->s_flag & cpu_to_le32(JFS_INLINELOG)) !=
-	    cpu_to_le32(JFS_INLINELOG))
-		j_sb->s_flag |= cpu_to_le32(JFS_INLINELOG);
 	if ((j_sb->s_flag & cpu_to_le32(JFS_GROUPCOMMIT)) !=
 	    cpu_to_le32(JFS_GROUPCOMMIT))
 		j_sb->s_flag |= cpu_to_le32(JFS_GROUPCOMMIT);
@@ -437,6 +415,8 @@ static int chkSuper(struct super_block *sb)
 	sbi->l2niperblk = sbi->l2bsize - L2DISIZE;
 	if (sbi->mntflag & JFS_INLINELOG)
 		sbi->logpxd = j_sb->s_logpxd;
+	else
+		sbi->logdev = to_kdev_t(le32_to_cpu(j_sb->s_logdev));
 	sbi->ait2 = j_sb->s_ait2;
 
       out:
@@ -476,6 +456,10 @@ int updateSuper(struct super_block *sb, uint state)
 		j_sb->s_logdev =
 			cpu_to_le32(kdev_t_to_nr(JFS_SBI(sb)->log->dev));
 		j_sb->s_logserial = cpu_to_le32(JFS_SBI(sb)->log->serial);
+		/* record our own device number in case the location
+		 * changes after a reboot
+		 */
+		j_sb->s_device = cpu_to_le32(kdev_t_to_nr(sb->s_dev));
 	} else if (state == FM_CLEAN) {
 		/*
 		 * If this volume is shared with OS/2, OS/2 will need to

fs/jfs/jfs_superblock.h

@@ -25,9 +25,9 @@
 /*
  * make the magic number something a human could read
  */
-#define JFS_MAGIC 	"JFS1"	/* Magic word: Version 1 */
+#define JFS_MAGIC 	"JFS1"	/* Magic word */
 
-#define JFS_VERSION	1	/* Version number: Version 1 */
+#define JFS_VERSION	2	/* Version number: Version 2 */
 
 #define LV_NAME_SIZE	11	/* MUST BE 11 for OS/2 boot sector */
 
@@ -103,38 +103,10 @@ struct jfs_superblock {
 	pxd_t s_xlogpxd;	/* 8: extendfs logpxd */
 	/* - 128 byte boundary - */
 
-	/*
-	 *      DFS VFS support (preliminary) 
-	 */
-	char s_attach;		/* 1: VFS: flag: set when aggregate is attached
+	u32 s_device;		/* Store device in case location changes
+				 * between reboots
 				 */
-	u8 rsrvd4[7];		/* 7: reserved - set to 0 */
 
-	u64 totalUsable;	/* 8: VFS: total of 1K blocks which are
-				 * available to "normal" (non-root) users.
-				 */
-	u64 minFree;		/* 8: VFS: # of 1K blocks held in reserve for 
-				 * exclusive use of root.  This value can be 0,
-				 * and if it is then totalUsable will be equal 
-				 * to # of blocks in aggregate.  I believe this
-				 * means that minFree + totalUsable = # blocks.
-				 * In that case, we don't need to store both 
-				 * totalUsable and minFree since we can compute
-				 * one from the other.  I would guess minFree 
-				 * would be the one we should store, and 
-				 * totalUsable would be the one we should 
-				 * compute.  (Just a guess...)
-				 */
-
-	u64 realFree;		/* 8: VFS: # of free 1K blocks can be used by 
-				 * "normal" users.  It may be this is something
-				 * we should compute when asked for instead of 
-				 * storing in the superblock.  I don't know how
-				 * often this information is needed.
-				 */
-	/*
-	 *      graffiti area
-	 */
 };
 
 extern int readSuper(struct super_block *, struct metapage **);