Linux-2.6.12-rc2v2.6.12-rc2

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!

1 files changed, 2524 insertions, 0 deletions

diff --git a/fs/jfs/jfs_logmgr.c b/fs/jfs/jfs_logmgr.c
new file mode 100644
index 00000000000..b6a6869ebb4
--- /dev/null
+++ b/fs/jfs/jfs_logmgr.c
@@ -0,0 +1,2524 @@
+/*
+ *   Copyright (C) International Business Machines Corp., 2000-2004
+ *   Portions Copyright (C) Christoph Hellwig, 2001-2002
+ *
+ *   This program is free software;  you can redistribute it and/or modify
+ *   it under the terms of the GNU General Public License as published by
+ *   the Free Software Foundation; either version 2 of the License, or 
+ *   (at your option) any later version.
+ * 
+ *   This program is distributed in the hope that it will be useful,
+ *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
+ *   the GNU General Public License for more details.
+ *
+ *   You should have received a copy of the GNU General Public License
+ *   along with this program;  if not, write to the Free Software 
+ *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+/*
+ *	jfs_logmgr.c: log manager
+ *
+ * for related information, see transaction manager (jfs_txnmgr.c), and
+ * recovery manager (jfs_logredo.c).
+ *
+ * note: for detail, RTFS.
+ *
+ *	log buffer manager:
+ * special purpose buffer manager supporting log i/o requirements.
+ * per log serial pageout of logpage
+ * queuing i/o requests and redrive i/o at iodone
+ * maintain current logpage buffer
+ * no caching since append only
+ * appropriate jfs buffer cache buffers as needed
+ *
+ *	group commit:
+ * transactions which wrote COMMIT records in the same in-memory
+ * log page during the pageout of previous/current log page(s) are
+ * committed together by the pageout of the page.
+ *
+ *	TBD lazy commit:
+ * transactions are committed asynchronously when the log page
+ * containing it COMMIT is paged out when it becomes full;
+ *
+ *	serialization:
+ * . a per log lock serialize log write.
+ * . a per log lock serialize group commit.
+ * . a per log lock serialize log open/close;
+ *
+ *	TBD log integrity:
+ * careful-write (ping-pong) of last logpage to recover from crash
+ * in overwrite.
+ * detection of split (out-of-order) write of physical sectors
+ * of last logpage via timestamp at end of each sector
+ * with its mirror data array at trailer).
+ *
+ *	alternatives:
+ * lsn - 64-bit monotonically increasing integer vs
+ * 32-bit lspn and page eor.
+ */
+
+#include <linux/fs.h>
+#include <linux/blkdev.h>
+#include <linux/interrupt.h>
+#include <linux/smp_lock.h>
+#include <linux/completion.h>
+#include <linux/buffer_head.h>		/* for sync_blockdev() */
+#include <linux/bio.h>
+#include <linux/suspend.h>
+#include <linux/delay.h>
+#include "jfs_incore.h"
+#include "jfs_filsys.h"
+#include "jfs_metapage.h"
+#include "jfs_txnmgr.h"
+#include "jfs_debug.h"
+
+
+/*
+ * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
+ */
+static struct lbuf *log_redrive_list;
+static DEFINE_SPINLOCK(log_redrive_lock);
+DECLARE_WAIT_QUEUE_HEAD(jfs_IO_thread_wait);
+
+
+/*
+ *	log read/write serialization (per log)
+ */
+#define LOG_LOCK_INIT(log)	init_MUTEX(&(log)->loglock)
+#define LOG_LOCK(log)		down(&((log)->loglock))
+#define LOG_UNLOCK(log)		up(&((log)->loglock))
+
+
+/*
+ *	log group commit serialization (per log)
+ */
+
+#define LOGGC_LOCK_INIT(log)	spin_lock_init(&(log)->gclock)
+#define LOGGC_LOCK(log)		spin_lock_irq(&(log)->gclock)
+#define LOGGC_UNLOCK(log)	spin_unlock_irq(&(log)->gclock)
+#define LOGGC_WAKEUP(tblk)	wake_up_all(&(tblk)->gcwait)
+
+/*
+ *	log sync serialization (per log)
+ */
+#define	LOGSYNC_DELTA(logsize)		min((logsize)/8, 128*LOGPSIZE)
+#define	LOGSYNC_BARRIER(logsize)	((logsize)/4)
+/*
+#define	LOGSYNC_DELTA(logsize)		min((logsize)/4, 256*LOGPSIZE)
+#define	LOGSYNC_BARRIER(logsize)	((logsize)/2)
+*/
+
+
+/*
+ *	log buffer cache synchronization
+ */
+static DEFINE_SPINLOCK(jfsLCacheLock);
+
+#define	LCACHE_LOCK(flags)	spin_lock_irqsave(&jfsLCacheLock, flags)
+#define	LCACHE_UNLOCK(flags)	spin_unlock_irqrestore(&jfsLCacheLock, flags)
+
+/*
+ * See __SLEEP_COND in jfs_locks.h
+ */
+#define LCACHE_SLEEP_COND(wq, cond, flags)	\
+do {						\
+	if (cond)				\
+		break;				\
+	__SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
+} while (0)
+
+#define	LCACHE_WAKEUP(event)	wake_up(event)
+
+
+/*
+ *	lbuf buffer cache (lCache) control
+ */
+/* log buffer manager pageout control (cumulative, inclusive) */
+#define	lbmREAD		0x0001
+#define	lbmWRITE	0x0002	/* enqueue at tail of write queue;
+				 * init pageout if at head of queue;
+				 */
+#define	lbmRELEASE	0x0004	/* remove from write queue
+				 * at completion of pageout;
+				 * do not free/recycle it yet:
+				 * caller will free it;
+				 */
+#define	lbmSYNC		0x0008	/* do not return to freelist
+				 * when removed from write queue;
+				 */
+#define lbmFREE		0x0010	/* return to freelist
+				 * at completion of pageout;
+				 * the buffer may be recycled;
+				 */
+#define	lbmDONE		0x0020
+#define	lbmERROR	0x0040
+#define lbmGC		0x0080	/* lbmIODone to perform post-GC processing
+				 * of log page
+				 */
+#define lbmDIRECT	0x0100
+
+/*
+ * Global list of active external journals
+ */
+static LIST_HEAD(jfs_external_logs);
+static struct jfs_log *dummy_log = NULL;
+static DECLARE_MUTEX(jfs_log_sem);
+
+/*
+ * external references
+ */
+extern void txLazyUnlock(struct tblock * tblk);
+extern int jfs_stop_threads;
+extern struct completion jfsIOwait;
+extern int jfs_tlocks_low;
+
+/*
+ * forward references
+ */
+static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
+			 struct lrd * lrd, struct tlock * tlck);
+
+static int lmNextPage(struct jfs_log * log);
+static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
+			   int activate);
+
+static int open_inline_log(struct super_block *sb);
+static int open_dummy_log(struct super_block *sb);
+static int lbmLogInit(struct jfs_log * log);
+static void lbmLogShutdown(struct jfs_log * log);
+static struct lbuf *lbmAllocate(struct jfs_log * log, int);
+static void lbmFree(struct lbuf * bp);
+static void lbmfree(struct lbuf * bp);
+static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
+static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
+static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
+static int lbmIOWait(struct lbuf * bp, int flag);
+static bio_end_io_t lbmIODone;
+static void lbmStartIO(struct lbuf * bp);
+static void lmGCwrite(struct jfs_log * log, int cant_block);
+static int lmLogSync(struct jfs_log * log, int nosyncwait);
+
+
+
+/*
+ *	statistics
+ */
+#ifdef CONFIG_JFS_STATISTICS
+static struct lmStat {
+	uint commit;		/* # of commit */
+	uint pagedone;		/* # of page written */
+	uint submitted;		/* # of pages submitted */
+	uint full_page;		/* # of full pages submitted */
+	uint partial_page;	/* # of partial pages submitted */
+} lmStat;
+#endif
+
+
+/*
+ * NAME:	lmLog()
+ *
+ * FUNCTION:	write a log record;
+ *
+ * PARAMETER:
+ *
+ * RETURN:	lsn - offset to the next log record to write (end-of-log);
+ *		-1  - error;
+ *
+ * note: todo: log error handler
+ */
+int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
+	  struct tlock * tlck)
+{
+	int lsn;
+	int diffp, difft;
+	struct metapage *mp = NULL;
+
+	jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
+		 log, tblk, lrd, tlck);
+
+	LOG_LOCK(log);
+
+	/* log by (out-of-transaction) JFS ? */
+	if (tblk == NULL)
+		goto writeRecord;
+
+	/* log from page ? */
+	if (tlck == NULL ||
+	    tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
+		goto writeRecord;
+
+	/*
+	 *      initialize/update page/transaction recovery lsn
+	 */
+	lsn = log->lsn;
+
+	LOGSYNC_LOCK(log);
+
+	/*
+	 * initialize page lsn if first log write of the page
+	 */
+	if (mp->lsn == 0) {
+		mp->log = log;
+		mp->lsn = lsn;
+		log->count++;
+
+		/* insert page at tail of logsynclist */
+		list_add_tail(&mp->synclist, &log->synclist);
+	}
+
+	/*
+	 *      initialize/update lsn of tblock of the page
+	 *
+	 * transaction inherits oldest lsn of pages associated
+	 * with allocation/deallocation of resources (their
+	 * log records are used to reconstruct allocation map
+	 * at recovery time: inode for inode allocation map,
+	 * B+-tree index of extent descriptors for block
+	 * allocation map);
+	 * allocation map pages inherit transaction lsn at
+	 * commit time to allow forwarding log syncpt past log
+	 * records associated with allocation/deallocation of
+	 * resources only after persistent map of these map pages
+	 * have been updated and propagated to home.
+	 */
+	/*
+	 * initialize transaction lsn:
+	 */
+	if (tblk->lsn == 0) {
+		/* inherit lsn of its first page logged */
+		tblk->lsn = mp->lsn;
+		log->count++;
+
+		/* insert tblock after the page on logsynclist */
+		list_add(&tblk->synclist, &mp->synclist);
+	}
+	/*
+	 * update transaction lsn:
+	 */
+	else {
+		/* inherit oldest/smallest lsn of page */
+		logdiff(diffp, mp->lsn, log);
+		logdiff(difft, tblk->lsn, log);
+		if (diffp < difft) {
+			/* update tblock lsn with page lsn */
+			tblk->lsn = mp->lsn;
+
+			/* move tblock after page on logsynclist */
+			list_move(&tblk->synclist, &mp->synclist);
+		}
+	}
+
+	LOGSYNC_UNLOCK(log);
+
+	/*
+	 *      write the log record
+	 */
+      writeRecord:
+	lsn = lmWriteRecord(log, tblk, lrd, tlck);
+
+	/*
+	 * forward log syncpt if log reached next syncpt trigger
+	 */
+	logdiff(diffp, lsn, log);
+	if (diffp >= log->nextsync)
+		lsn = lmLogSync(log, 0);
+
+	/* update end-of-log lsn */
+	log->lsn = lsn;
+
+	LOG_UNLOCK(log);
+
+	/* return end-of-log address */
+	return lsn;
+}
+
+
+/*
+ * NAME:	lmWriteRecord()
+ *
+ * FUNCTION:	move the log record to current log page
+ *
+ * PARAMETER:	cd	- commit descriptor
+ *
+ * RETURN:	end-of-log address
+ *			
+ * serialization: LOG_LOCK() held on entry/exit
+ */
+static int
+lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
+	      struct tlock * tlck)
+{
+	int lsn = 0;		/* end-of-log address */
+	struct lbuf *bp;	/* dst log page buffer */
+	struct logpage *lp;	/* dst log page */
+	caddr_t dst;		/* destination address in log page */
+	int dstoffset;		/* end-of-log offset in log page */
+	int freespace;		/* free space in log page */
+	caddr_t p;		/* src meta-data page */
+	caddr_t src;
+	int srclen;
+	int nbytes;		/* number of bytes to move */
+	int i;
+	int len;
+	struct linelock *linelock;
+	struct lv *lv;
+	struct lvd *lvd;
+	int l2linesize;
+
+	len = 0;
+
+	/* retrieve destination log page to write */
+	bp = (struct lbuf *) log->bp;
+	lp = (struct logpage *) bp->l_ldata;
+	dstoffset = log->eor;
+
+	/* any log data to write ? */
+	if (tlck == NULL)
+		goto moveLrd;
+
+	/*
+	 *      move log record data
+	 */
+	/* retrieve source meta-data page to log */
+	if (tlck->flag & tlckPAGELOCK) {
+		p = (caddr_t) (tlck->mp->data);
+		linelock = (struct linelock *) & tlck->lock;
+	}
+	/* retrieve source in-memory inode to log */
+	else if (tlck->flag & tlckINODELOCK) {
+		if (tlck->type & tlckDTREE)
+			p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
+		else
+			p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
+		linelock = (struct linelock *) & tlck->lock;
+	}
+#ifdef	_JFS_WIP
+	else if (tlck->flag & tlckINLINELOCK) {
+
+		inlinelock = (struct inlinelock *) & tlck;
+		p = (caddr_t) & inlinelock->pxd;
+		linelock = (struct linelock *) & tlck;
+	}
+#endif				/* _JFS_WIP */
+	else {
+		jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
+		return 0;	/* Probably should trap */
+	}
+	l2linesize = linelock->l2linesize;
+
+      moveData:
+	ASSERT(linelock->index <= linelock->maxcnt);
+
+	lv = linelock->lv;
+	for (i = 0; i < linelock->index; i++, lv++) {
+		if (lv->length == 0)
+			continue;
+
+		/* is page full ? */
+		if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
+			/* page become full: move on to next page */
+			lmNextPage(log);
+
+			bp = log->bp;
+			lp = (struct logpage *) bp->l_ldata;
+			dstoffset = LOGPHDRSIZE;
+		}
+
+		/*
+		 * move log vector data
+		 */
+		src = (u8 *) p + (lv->offset << l2linesize);
+		srclen = lv->length << l2linesize;
+		len += srclen;
+		while (srclen > 0) {
+			freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
+			nbytes = min(freespace, srclen);
+			dst = (caddr_t) lp + dstoffset;
+			memcpy(dst, src, nbytes);
+			dstoffset += nbytes;
+
+			/* is page not full ? */
+			if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
+				break;
+
+			/* page become full: move on to next page */
+			lmNextPage(log);
+
+			bp = (struct lbuf *) log->bp;
+			lp = (struct logpage *) bp->l_ldata;
+			dstoffset = LOGPHDRSIZE;
+
+			srclen -= nbytes;
+			src += nbytes;
+		}
+
+		/*
+		 * move log vector descriptor
+		 */
+		len += 4;
+		lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
+		lvd->offset = cpu_to_le16(lv->offset);
+		lvd->length = cpu_to_le16(lv->length);
+		dstoffset += 4;
+		jfs_info("lmWriteRecord: lv offset:%d length:%d",
+			 lv->offset, lv->length);
+	}
+
+	if ((i = linelock->next)) {
+		linelock = (struct linelock *) lid_to_tlock(i);
+		goto moveData;
+	}
+
+	/*
+	 *      move log record descriptor
+	 */
+      moveLrd:
+	lrd->length = cpu_to_le16(len);
+
+	src = (caddr_t) lrd;
+	srclen = LOGRDSIZE;
+
+	while (srclen > 0) {
+		freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
+		nbytes = min(freespace, srclen);
+		dst = (caddr_t) lp + dstoffset;
+		memcpy(dst, src, nbytes);
+
+		dstoffset += nbytes;
+		srclen -= nbytes;
+
+		/* are there more to move than freespace of page ? */
+		if (srclen)
+			goto pageFull;
+
+		/*
+		 * end of log record descriptor
+		 */
+
+		/* update last log record eor */
+		log->eor = dstoffset;
+		bp->l_eor = dstoffset;
+		lsn = (log->page << L2LOGPSIZE) + dstoffset;
+
+		if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
+			tblk->clsn = lsn;
+			jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
+				 bp->l_eor);
+
+			INCREMENT(lmStat.commit);	/* # of commit */
+
+			/*
+			 * enqueue tblock for group commit:
+			 *
+			 * enqueue tblock of non-trivial/synchronous COMMIT
+			 * at tail of group commit queue
+			 * (trivial/asynchronous COMMITs are ignored by
+			 * group commit.)
+			 */
+			LOGGC_LOCK(log);
+
+			/* init tblock gc state */
+			tblk->flag = tblkGC_QUEUE;
+			tblk->bp = log->bp;
+			tblk->pn = log->page;
+			tblk->eor = log->eor;
+
+			/* enqueue transaction to commit queue */
+			list_add_tail(&tblk->cqueue, &log->cqueue);
+
+			LOGGC_UNLOCK(log);
+		}
+
+		jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
+			le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
+
+		/* page not full ? */
+		if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
+			return lsn;
+
+	      pageFull:
+		/* page become full: move on to next page */
+		lmNextPage(log);
+
+		bp = (struct lbuf *) log->bp;
+		lp = (struct logpage *) bp->l_ldata;
+		dstoffset = LOGPHDRSIZE;
+		src += nbytes;
+	}
+
+	return lsn;
+}
+
+
+/*
+ * NAME:	lmNextPage()
+ *
+ * FUNCTION:	write current page and allocate next page.
+ *
+ * PARAMETER:	log
+ *
+ * RETURN:	0
+ *			
+ * serialization: LOG_LOCK() held on entry/exit
+ */
+static int lmNextPage(struct jfs_log * log)
+{
+	struct logpage *lp;
+	int lspn;		/* log sequence page number */
+	int pn;			/* current page number */
+	struct lbuf *bp;
+	struct lbuf *nextbp;
+	struct tblock *tblk;
+
+	/* get current log page number and log sequence page number */
+	pn = log->page;
+	bp = log->bp;
+	lp = (struct logpage *) bp->l_ldata;
+	lspn = le32_to_cpu(lp->h.page);
+
+	LOGGC_LOCK(log);
+
+	/*
+	 *      write or queue the full page at the tail of write queue
+	 */
+	/* get the tail tblk on commit queue */
+	if (list_empty(&log->cqueue))
+		tblk = NULL;
+	else
+		tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
+
+	/* every tblk who has COMMIT record on the current page,
+	 * and has not been committed, must be on commit queue
+	 * since tblk is queued at commit queueu at the time
+	 * of writing its COMMIT record on the page before
+	 * page becomes full (even though the tblk thread
+	 * who wrote COMMIT record may have been suspended
+	 * currently);
+	 */
+
+	/* is page bound with outstanding tail tblk ? */
+	if (tblk && tblk->pn == pn) {
+		/* mark tblk for end-of-page */
+		tblk->flag |= tblkGC_EOP;
+
+		if (log->cflag & logGC_PAGEOUT) {
+			/* if page is not already on write queue,
+			 * just enqueue (no lbmWRITE to prevent redrive)
+			 * buffer to wqueue to ensure correct serial order
+			 * of the pages since log pages will be added
+			 * continuously
+			 */
+			if (bp->l_wqnext == NULL)
+				lbmWrite(log, bp, 0, 0);
+		} else {
+			/*
+			 * No current GC leader, initiate group commit
+			 */
+			log->cflag |= logGC_PAGEOUT;
+			lmGCwrite(log, 0);
+		}
+	}
+	/* page is not bound with outstanding tblk:
+	 * init write or mark it to be redriven (lbmWRITE)
+	 */
+	else {
+		/* finalize the page */
+		bp->l_ceor = bp->l_eor;
+		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
+		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
+	}
+	LOGGC_UNLOCK(log);
+
+	/*
+	 *      allocate/initialize next page
+	 */
+	/* if log wraps, the first data page of log is 2
+	 * (0 never used, 1 is superblock).
+	 */
+	log->page = (pn == log->size - 1) ? 2 : pn + 1;
+	log->eor = LOGPHDRSIZE;	/* ? valid page empty/full at logRedo() */
+
+	/* allocate/initialize next log page buffer */
+	nextbp = lbmAllocate(log, log->page);
+	nextbp->l_eor = log->eor;
+	log->bp = nextbp;
+
+	/* initialize next log page */
+	lp = (struct logpage *) nextbp->l_ldata;
+	lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
+	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
+
+	return 0;
+}
+
+
+/*
+ * NAME:	lmGroupCommit()
+ *
+ * FUNCTION:	group commit
+ *	initiate pageout of the pages with COMMIT in the order of
+ *	page number - redrive pageout of the page at the head of
+ *	pageout queue until full page has been written.
+ *
+ * RETURN:	
+ *
+ * NOTE:
+ *	LOGGC_LOCK serializes log group commit queue, and
+ *	transaction blocks on the commit queue.
+ *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
+ */
+int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
+{
+	int rc = 0;
+
+	LOGGC_LOCK(log);
+
+	/* group committed already ? */
+	if (tblk->flag & tblkGC_COMMITTED) {
+		if (tblk->flag & tblkGC_ERROR)
+			rc = -EIO;
+
+		LOGGC_UNLOCK(log);
+		return rc;
+	}
+	jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
+
+	if (tblk->xflag & COMMIT_LAZY)
+		tblk->flag |= tblkGC_LAZY;
+
+	if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
+	    (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
+	     || jfs_tlocks_low)) {
+		/*
+		 * No pageout in progress
+		 *
+		 * start group commit as its group leader.
+		 */
+		log->cflag |= logGC_PAGEOUT;
+
+		lmGCwrite(log, 0);
+	}
+
+	if (tblk->xflag & COMMIT_LAZY) {
+		/*
+		 * Lazy transactions can leave now
+		 */
+		LOGGC_UNLOCK(log);
+		return 0;
+	}
+
+	/* lmGCwrite gives up LOGGC_LOCK, check again */
+
+	if (tblk->flag & tblkGC_COMMITTED) {
+		if (tblk->flag & tblkGC_ERROR)
+			rc = -EIO;
+
+		LOGGC_UNLOCK(log);
+		return rc;
+	}
+
+	/* upcount transaction waiting for completion
+	 */
+	log->gcrtc++;
+	tblk->flag |= tblkGC_READY;
+
+	__SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
+		     LOGGC_LOCK(log), LOGGC_UNLOCK(log));
+
+	/* removed from commit queue */
+	if (tblk->flag & tblkGC_ERROR)
+		rc = -EIO;
+
+	LOGGC_UNLOCK(log);
+	return rc;
+}
+
+/*
+ * NAME:	lmGCwrite()
+ *
+ * FUNCTION:	group commit write
+ *	initiate write of log page, building a group of all transactions
+ *	with commit records on that page.
+ *
+ * RETURN:	None
+ *
+ * NOTE:
+ *	LOGGC_LOCK must be held by caller.
+ *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
+ */
+static void lmGCwrite(struct jfs_log * log, int cant_write)
+{
+	struct lbuf *bp;
+	struct logpage *lp;
+	int gcpn;		/* group commit page number */
+	struct tblock *tblk;
+	struct tblock *xtblk = NULL;
+
+	/*
+	 * build the commit group of a log page
+	 *
+	 * scan commit queue and make a commit group of all
+	 * transactions with COMMIT records on the same log page.
+	 */
+	/* get the head tblk on the commit queue */
+	gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
+
+	list_for_each_entry(tblk, &log->cqueue, cqueue) {
+		if (tblk->pn != gcpn)
+			break;
+
+		xtblk = tblk;
+
+		/* state transition: (QUEUE, READY) -> COMMIT */
+		tblk->flag |= tblkGC_COMMIT;
+	}
+	tblk = xtblk;		/* last tblk of the page */
+
+	/*
+	 * pageout to commit transactions on the log page.
+	 */
+	bp = (struct lbuf *) tblk->bp;
+	lp = (struct logpage *) bp->l_ldata;
+	/* is page already full ? */
+	if (tblk->flag & tblkGC_EOP) {
+		/* mark page to free at end of group commit of the page */
+		tblk->flag &= ~tblkGC_EOP;
+		tblk->flag |= tblkGC_FREE;
+		bp->l_ceor = bp->l_eor;
+		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
+		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
+			 cant_write);
+		INCREMENT(lmStat.full_page);
+	}
+	/* page is not yet full */
+	else {
+		bp->l_ceor = tblk->eor;	/* ? bp->l_ceor = bp->l_eor; */
+		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
+		lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
+		INCREMENT(lmStat.partial_page);
+	}
+}
+
+/*
+ * NAME:	lmPostGC()
+ *
+ * FUNCTION:	group commit post-processing
+ *	Processes transactions after their commit records have been written
+ *	to disk, redriving log I/O if necessary.
+ *
+ * RETURN:	None
+ *
+ * NOTE:
+ *	This routine is called a interrupt time by lbmIODone
+ */
+static void lmPostGC(struct lbuf * bp)
+{
+	unsigned long flags;
+	struct jfs_log *log = bp->l_log;
+	struct logpage *lp;
+	struct tblock *tblk, *temp;
+
+	//LOGGC_LOCK(log);
+	spin_lock_irqsave(&log->gclock, flags);
+	/*
+	 * current pageout of group commit completed.
+	 *
+	 * remove/wakeup transactions from commit queue who were
+	 * group committed with the current log page
+	 */
+	list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
+		if (!(tblk->flag & tblkGC_COMMIT))
+			break;
+		/* if transaction was marked GC_COMMIT then
+		 * it has been shipped in the current pageout
+		 * and made it to disk - it is committed.
+		 */
+
+		if (bp->l_flag & lbmERROR)
+			tblk->flag |= tblkGC_ERROR;
+
+		/* remove it from the commit queue */
+		list_del(&tblk->cqueue);
+		tblk->flag &= ~tblkGC_QUEUE;
+
+		if (tblk == log->flush_tblk) {
+			/* we can stop flushing the log now */
+			clear_bit(log_FLUSH, &log->flag);
+			log->flush_tblk = NULL;
+		}
+
+		jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
+			 tblk->flag);
+
+		if (!(tblk->xflag & COMMIT_FORCE))
+			/*
+			 * Hand tblk over to lazy commit thread
+			 */
+			txLazyUnlock(tblk);
+		else {
+			/* state transition: COMMIT -> COMMITTED */
+			tblk->flag |= tblkGC_COMMITTED;
+
+			if (tblk->flag & tblkGC_READY)
+				log->gcrtc--;
+
+			LOGGC_WAKEUP(tblk);
+		}
+
+		/* was page full before pageout ?
+		 * (and this is the last tblk bound with the page)
+		 */
+		if (tblk->flag & tblkGC_FREE)
+			lbmFree(bp);
+		/* did page become full after pageout ?
+		 * (and this is the last tblk bound with the page)
+		 */
+		else if (tblk->flag & tblkGC_EOP) {
+			/* finalize the page */
+			lp = (struct logpage *) bp->l_ldata;
+			bp->l_ceor = bp->l_eor;
+			lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
+			jfs_info("lmPostGC: calling lbmWrite");
+			lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
+				 1);
+		}
+
+	}
+
+	/* are there any transactions who have entered lnGroupCommit()
+	 * (whose COMMITs are after that of the last log page written.
+	 * They are waiting for new group commit (above at (SLEEP 1))
+	 * or lazy transactions are on a full (queued) log page,
+	 * select the latest ready transaction as new group leader and
+	 * wake her up to lead her group.
+	 */
+	if ((!list_empty(&log->cqueue)) &&
+	    ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
+	     test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
+		/*
+		 * Call lmGCwrite with new group leader
+		 */
+		lmGCwrite(log, 1);
+
+	/* no transaction are ready yet (transactions are only just
+	 * queued (GC_QUEUE) and not entered for group commit yet).
+	 * the first transaction entering group commit
+	 * will elect herself as new group leader.
+	 */
+	else
+		log->cflag &= ~logGC_PAGEOUT;
+
+	//LOGGC_UNLOCK(log);
+	spin_unlock_irqrestore(&log->gclock, flags);
+	return;
+}
+
+/*
+ * NAME:	lmLogSync()
+ *
+ * FUNCTION:	write log SYNCPT record for specified log
+ *	if new sync address is available
+ *	(normally the case if sync() is executed by back-ground
+ *	process).
+ *	if not, explicitly run jfs_blogsync() to initiate
+ *	getting of new sync address.
+ *	calculate new value of i_nextsync which determines when
+ *	this code is called again.
+ *
+ *	this is called only from lmLog().
+ *
+ * PARAMETER:	ip	- pointer to logs inode.
+ *
+ * RETURN:	0
+ *			
+ * serialization: LOG_LOCK() held on entry/exit
+ */
+static int lmLogSync(struct jfs_log * log, int nosyncwait)
+{
+	int logsize;
+	int written;		/* written since last syncpt */
+	int free;		/* free space left available */
+	int delta;		/* additional delta to write normally */
+	int more;		/* additional write granted */
+	struct lrd lrd;
+	int lsn;
+	struct logsyncblk *lp;
+
+	/*
+	 *      forward syncpt
+	 */
+	/* if last sync is same as last syncpt,
+	 * invoke sync point forward processing to update sync.
+	 */
+
+	if (log->sync == log->syncpt) {
+		LOGSYNC_LOCK(log);
+		/* ToDo: push dirty metapages out to disk */
+//              bmLogSync(log);
+
+		if (list_empty(&log->synclist))
+			log->sync = log->lsn;
+		else {
+			lp = list_entry(log->synclist.next,
+					struct logsyncblk, synclist);
+			log->sync = lp->lsn;
+		}
+		LOGSYNC_UNLOCK(log);
+
+	}
+
+	/* if sync is different from last syncpt,
+	 * write a SYNCPT record with syncpt = sync.
+	 * reset syncpt = sync
+	 */
+	if (log->sync != log->syncpt) {
+		struct jfs_sb_info *sbi;
+
+		/*
+		 * We need to make sure all of the "written" metapages
+		 * actually make it to disk
+		 */
+		list_for_each_entry(sbi, &log->sb_list, log_list) {
+			if (sbi->flag & JFS_NOINTEGRITY)
+				continue;
+			filemap_fdatawrite(sbi->ipbmap->i_mapping);
+			filemap_fdatawrite(sbi->ipimap->i_mapping);
+			filemap_fdatawrite(sbi->sb->s_bdev->bd_inode->i_mapping);
+		}
+		list_for_each_entry(sbi, &log->sb_list, log_list) {
+			if (sbi->flag & JFS_NOINTEGRITY)
+				continue;
+			filemap_fdatawait(sbi->ipbmap->i_mapping);
+			filemap_fdatawait(sbi->ipimap->i_mapping);
+			filemap_fdatawait(sbi->sb->s_bdev->bd_inode->i_mapping);
+		}
+
+		lrd.logtid = 0;
+		lrd.backchain = 0;
+		lrd.type = cpu_to_le16(LOG_SYNCPT);
+		lrd.length = 0;
+		lrd.log.syncpt.sync = cpu_to_le32(log->sync);
+		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
+
+		log->syncpt = log->sync;
+	} else
+		lsn = log->lsn;
+
+	/*
+	 *      setup next syncpt trigger (SWAG)
+	 */
+	logsize = log->logsize;
+
+	logdiff(written, lsn, log);
+	free = logsize - written;
+	delta = LOGSYNC_DELTA(logsize);
+	more = min(free / 2, delta);
+	if (more < 2 * LOGPSIZE) {
+		jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
+		/*
+		 *      log wrapping
+		 *
+		 * option 1 - panic ? No.!
+		 * option 2 - shutdown file systems
+		 *            associated with log ?
+		 * option 3 - extend log ?
+		 */
+		/*
+		 * option 4 - second chance
+		 *
+		 * mark log wrapped, and continue.
+		 * when all active transactions are completed,
+		 * mark log vaild for recovery.
+		 * if crashed during invalid state, log state
+		 * implies invald log, forcing fsck().
+		 */
+		/* mark log state log wrap in log superblock */
+		/* log->state = LOGWRAP; */
+
+		/* reset sync point computation */
+		log->syncpt = log->sync = lsn;
+		log->nextsync = delta;
+	} else
+		/* next syncpt trigger = written + more */
+		log->nextsync = written + more;
+
+	/* return if lmLogSync() from outside of transaction, e.g., sync() */
+	if (nosyncwait)
+		return lsn;
+
+	/* if number of bytes written from last sync point is more
+	 * than 1/4 of the log size, stop new transactions from
+	 * starting until all current transactions are completed
+	 * by setting syncbarrier flag.
+	 */
+	if (written > LOGSYNC_BARRIER(logsize) && logsize > 32 * LOGPSIZE) {
+		set_bit(log_SYNCBARRIER, &log->flag);
+		jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
+			 log->syncpt);
+		/*
+		 * We may have to initiate group commit
+		 */
+		jfs_flush_journal(log, 0);
+	}
+
+	return lsn;
+}
+
+
+/*
+ * NAME:	lmLogOpen()
+ *
+ * FUNCTION:    open the log on first open;
+ *	insert filesystem in the active list of the log.
+ *
+ * PARAMETER:	ipmnt	- file system mount inode
+ *		iplog 	- log inode (out)
+ *
+ * RETURN:
+ *
+ * serialization:
+ */
+int lmLogOpen(struct super_block *sb)
+{
+	int rc;
+	struct block_device *bdev;
+	struct jfs_log *log;
+	struct jfs_sb_info *sbi = JFS_SBI(sb);
+
+	if (sbi->flag & JFS_NOINTEGRITY)
+		return open_dummy_log(sb);
+	
+	if (sbi->mntflag & JFS_INLINELOG)
+		return open_inline_log(sb);
+
+	down(&jfs_log_sem);
+	list_for_each_entry(log, &jfs_external_logs, journal_list) {
+		if (log->bdev->bd_dev == sbi->logdev) {
+			if (memcmp(log->uuid, sbi->loguuid,
+				   sizeof(log->uuid))) {
+				jfs_warn("wrong uuid on JFS journal\n");
+				up(&jfs_log_sem);
+				return -EINVAL;
+			}
+			/*
+			 * add file system to log active file system list
+			 */
+			if ((rc = lmLogFileSystem(log, sbi, 1))) {
+				up(&jfs_log_sem);
+				return rc;
+			}
+			goto journal_found;
+		}
+	}
+
+	if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
+		up(&jfs_log_sem);
+		return -ENOMEM;
+	}
+	memset(log, 0, sizeof(struct jfs_log));
+	INIT_LIST_HEAD(&log->sb_list);
+	init_waitqueue_head(&log->syncwait);
+
+	/*
+	 *      external log as separate logical volume
+	 *
+	 * file systems to log may have n-to-1 relationship;
+	 */
+
+	bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
+	if (IS_ERR(bdev)) {
+		rc = -PTR_ERR(bdev);
+		goto free;
+	}
+
+	if ((rc = bd_claim(bdev, log))) {
+		goto close;
+	}
+
+	log->bdev = bdev;
+	memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
+	
+	/*
+	 * initialize log:
+	 */
+	if ((rc = lmLogInit(log)))
+		goto unclaim;
+
+	list_add(&log->journal_list, &jfs_external_logs);
+
+	/*
+	 * add file system to log active file system list
+	 */
+	if ((rc = lmLogFileSystem(log, sbi, 1)))
+		goto shutdown;
+
+journal_found:
+	LOG_LOCK(log);
+	list_add(&sbi->log_list, &log->sb_list);
+	sbi->log = log;
+	LOG_UNLOCK(log);
+
+	up(&jfs_log_sem);
+	return 0;
+
+	/*
+	 *      unwind on error
+	 */
+      shutdown:		/* unwind lbmLogInit() */
+	list_del(&log->journal_list);
+	lbmLogShutdown(log);
+
+      unclaim:
+	bd_release(bdev);
+
+      close:		/* close external log device */
+	blkdev_put(bdev);
+
+      free:		/* free log descriptor */
+	up(&jfs_log_sem);
+	kfree(log);
+
+	jfs_warn("lmLogOpen: exit(%d)", rc);
+	return rc;
+}
+
+static int open_inline_log(struct super_block *sb)
+{
+	struct jfs_log *log;
+	int rc;
+
+	if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL)))
+		return -ENOMEM;
+	memset(log, 0, sizeof(struct jfs_log));
+	INIT_LIST_HEAD(&log->sb_list);
+	init_waitqueue_head(&am