Merge jfs@jfs.bkbits.net:linux-2.5

into austin.ibm.com:/shaggy/bk/jfs-2.5

fs/jfs/jfs_incore.h

@@ -152,6 +152,11 @@ struct jfs_sb_info {
 	pxd_t		ait2;		/* pxd describing AIT copy	*/
 	char		uuid[16];	/* 128-bit uuid for volume	*/
 	char		loguuid[16];	/* 128-bit uuid for log	*/
+	/*
+	 * commit_state is used for synchronization of the jfs_commit
+	 * threads.  It is protected by LAZY_LOCK().
+	 */
+	int		commit_state;	/* commit state */
 	/* Formerly in ipimap */
 	uint		gengen;		/* inode generation generator*/
 	uint		inostamp;	/* shows inode belongs to fileset*/
@@ -164,6 +169,9 @@ struct jfs_sb_info {
 	uint		p_state;	/* state prior to going no integrity */
 };
 
+/* jfs_sb_info commit_state */
+#define IN_LAZYCOMMIT 1
+
 static inline struct jfs_inode_info *JFS_IP(struct inode *inode)
 {
 	return list_entry(inode, struct jfs_inode_info, vfs_inode);

fs/jfs/jfs_logmgr.c

@@ -171,6 +171,7 @@ DECLARE_MUTEX(jfs_log_sem);
 extern void txLazyUnlock(struct tblock * tblk);
 extern int jfs_stop_threads;
 extern struct completion jfsIOwait;
+extern int jfs_tlocks_low;
 
 /*
  * forward references
@@ -524,12 +525,7 @@ lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
 			tblk->eor = log->eor;
 
 			/* enqueue transaction to commit queue */
-			tblk->cqnext = NULL;
-			if (log->cqueue.head) {
-				log->cqueue.tail->cqnext = tblk;
-				log->cqueue.tail = tblk;
-			} else
-				log->cqueue.head = log->cqueue.tail = tblk;
+			list_add_tail(&tblk->cqueue, &log->cqueue);
 
 			LOGGC_UNLOCK(log);
 		}
@@ -587,7 +583,10 @@ static int lmNextPage(struct jfs_log * log)
 	 *      write or queue the full page at the tail of write queue
 	 */
 	/* get the tail tblk on commit queue */
-	tblk = log->cqueue.tail;
+	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
@@ -688,8 +687,9 @@ int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
 	if (tblk->xflag & COMMIT_LAZY)
 		tblk->flag |= tblkGC_LAZY;
 
-	if ((!(log->cflag & logGC_PAGEOUT)) && log->cqueue.head &&
-	    (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag))) {
+	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
 		 *
@@ -753,7 +753,7 @@ static void lmGCwrite(struct jfs_log * log, int cant_write)
 	struct logpage *lp;
 	int gcpn;		/* group commit page number */
 	struct tblock *tblk;
-	struct tblock *xtblk;
+	struct tblock *xtblk = NULL;
 
 	/*
 	 * build the commit group of a log page
@@ -762,15 +762,16 @@ static void lmGCwrite(struct jfs_log * log, int cant_write)
 	 * transactions with COMMIT records on the same log page.
 	 */
 	/* get the head tblk on the commit queue */
-	tblk = xtblk = log->cqueue.head;
-	gcpn = tblk->pn;
+	gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
+
+	list_for_each_entry(tblk, &log->cqueue, cqueue) {
+		if (tblk->pn != gcpn)
+			break;
 
-	while (tblk && tblk->pn == gcpn) {
 		xtblk = tblk;
 
 		/* state transition: (QUEUE, READY) -> COMMIT */
 		tblk->flag |= tblkGC_COMMIT;
-		tblk = tblk->cqnext;
 	}
 	tblk = xtblk;		/* last tblk of the page */
 
@@ -816,7 +817,7 @@ static void lmPostGC(struct lbuf * bp)
 	unsigned long flags;
 	struct jfs_log *log = bp->l_log;
 	struct logpage *lp;
-	struct tblock *tblk;
+	struct tblock *tblk, *temp;
 
 	//LOGGC_LOCK(log);
 	spin_lock_irqsave(&log->gclock, flags);
@@ -826,7 +827,9 @@ static void lmPostGC(struct lbuf * bp)
 	 * remove/wakeup transactions from commit queue who were
 	 * group committed with the current log page
 	 */
-	while ((tblk = log->cqueue.head) && (tblk->flag & tblkGC_COMMIT)) {
+	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.
@@ -836,11 +839,8 @@ static void lmPostGC(struct lbuf * bp)
 			tblk->flag |= tblkGC_ERROR;
 
 		/* remove it from the commit queue */
-		log->cqueue.head = tblk->cqnext;
-		if (log->cqueue.head == NULL)
-			log->cqueue.tail = NULL;
+		list_del(&tblk->cqueue);
 		tblk->flag &= ~tblkGC_QUEUE;
-		tblk->cqnext = 0;
 
 		if (tblk == log->flush_tblk) {
 			/* we can stop flushing the log now */
@@ -893,9 +893,9 @@ static void lmPostGC(struct lbuf * bp)
 	 * select the latest ready transaction as new group leader and
 	 * wake her up to lead her group.
 	 */
-	if ((tblk = log->cqueue.head) &&
+	if ((!list_empty(&log->cqueue)) &&
 	    ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
-	     test_bit(log_FLUSH, &log->flag)))
+	     test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
 		/*
 		 * Call lmGCwrite with new group leader
 		 */
@@ -1288,7 +1288,7 @@ int lmLogInit(struct jfs_log * log)
 
 	init_waitqueue_head(&log->syncwait);
 
-	log->cqueue.head = log->cqueue.tail = NULL;
+	INIT_LIST_HEAD(&log->cqueue);
 	log->flush_tblk = NULL;
 
 	log->count = 0;
@@ -1486,6 +1486,7 @@ int lmLogClose(struct super_block *sb)
 		 *      in-line log in host file system
 		 */
 		rc = lmLogShutdown(log);
+		kfree(log);
 		goto out;
 	}
 
@@ -1515,6 +1516,8 @@ int lmLogClose(struct super_block *sb)
 	bd_release(bdev);
 	blkdev_put(bdev);
 
+	kfree(log);
+
       out:
 	up(&jfs_log_sem);
 	jfs_info("lmLogClose: exit(%d)", rc);
@@ -1535,7 +1538,7 @@ int lmLogClose(struct super_block *sb)
 void jfs_flush_journal(struct jfs_log *log, int wait)
 {
 	int i;
-	struct tblock *target;
+	struct tblock *target = NULL;
 
 	/* jfs_write_inode may call us during read-only mount */
 	if (!log)
@@ -1545,13 +1548,12 @@ void jfs_flush_journal(struct jfs_log *log, int wait)
 
 	LOGGC_LOCK(log);
 
-	target = log->cqueue.head;
-
-	if (target) {
+	if (!list_empty(&log->cqueue)) {
 		/*
 		 * This ensures that we will keep writing to the journal as long
 		 * as there are unwritten commit records
 		 */
+		target = list_entry(log->cqueue.prev, struct tblock, cqueue);
 
 		if (test_bit(log_FLUSH, &log->flag)) {
 			/*
@@ -1602,16 +1604,16 @@ void jfs_flush_journal(struct jfs_log *log, int wait)
 	 * If there was recent activity, we may need to wait
 	 * for the lazycommit thread to catch up
 	 */
-	if (log->cqueue.head || !list_empty(&log->synclist)) {
+	if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
 		for (i = 0; i < 800; i++) {	/* Too much? */
 			current->state = TASK_INTERRUPTIBLE;
 			schedule_timeout(HZ / 4);
-			if ((log->cqueue.head == NULL) &&
+			if (list_empty(&log->cqueue) &&
 			    list_empty(&log->synclist))
 				break;
 		}
 	}
-	assert(log->cqueue.head == NULL);
+	assert(list_empty(&log->cqueue));
 	assert(list_empty(&log->synclist));
 	clear_bit(log_FLUSH, &log->flag);
 }

fs/jfs/jfs_logmgr.h

@@ -398,10 +398,7 @@ struct jfs_log {
 
 	/* commit */
 	uint cflag;		/* 4: */
-	struct {		/* 8: FIFO commit queue header */
-		struct tblock *head;
-		struct tblock *tail;
-	} cqueue;
+	struct list_head	cqueue; /* FIFO commit queue */
 	struct tblock *flush_tblk; /* tblk we're waiting on for flush */
 	int gcrtc;		/* 4: GC_READY transaction count */
 	struct tblock *gclrt;	/* 4: latest GC_READY transaction */

fs/jfs/jfs_mount.c

@@ -288,6 +288,11 @@ int jfs_mount_rw(struct super_block *sb, int remount)
 	 */
 	logMOUNT(sb);
 
+	/*
+	 * Set page cache allocation policy
+	 */
+	mapping_set_gfp_mask(sb->s_bdev->bd_inode->i_mapping, GFP_NOFS);
+
 	return rc;
 }
 

fs/jfs/jfs_txnmgr.c

@@ -48,6 +48,8 @@
 #include <linux/smp_lock.h>
 #include <linux/completion.h>
 #include <linux/suspend.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
 #include "jfs_incore.h"
 #include "jfs_filsys.h"
 #include "jfs_metapage.h"
@@ -61,25 +63,22 @@
  *      transaction management structures
  */
 static struct {
-	/* tblock */
 	int freetid;		/* index of a free tid structure */
-	wait_queue_head_t freewait;	/* eventlist of free tblock */
-
-	/* tlock */
 	int freelock;		/* index first free lock word */
+	wait_queue_head_t freewait;	/* eventlist of free tblock */
 	wait_queue_head_t freelockwait;	/* eventlist of free tlock */
 	wait_queue_head_t lowlockwait;	/* eventlist of ample tlocks */
 	int tlocksInUse;	/* Number of tlocks in use */
-	int TlocksLow;		/* Indicates low number of available tlocks */
 	spinlock_t LazyLock;	/* synchronize sync_queue & unlock_queue */
 /*	struct tblock *sync_queue; * Transactions waiting for data sync */
-	struct tblock *unlock_queue;	/* Txns waiting to be released */
-	struct tblock *unlock_tail;	/* Tail of unlock_queue */
+	struct list_head unlock_queue;	/* Txns waiting to be released */
 	struct list_head anon_list;	/* inodes having anonymous txns */
 	struct list_head anon_list2;	/* inodes having anonymous txns
 					   that couldn't be sync'ed */
 } TxAnchor;
 
+int jfs_tlocks_low;		/* Indicates low number of available tlocks */
+
 #ifdef CONFIG_JFS_STATISTICS
 struct {
 	uint txBegin;
@@ -95,11 +94,19 @@ struct {
 #endif
 
 static int nTxBlock = 512;	/* number of transaction blocks */
-struct tblock *TxBlock;	        /* transaction block table */
+module_param(nTxBlock, int, 0);
+MODULE_PARM_DESC(nTxBlock,
+		 "Number of transaction blocks (default:512, max:65536)");
 
 static int nTxLock = 4096;	/* number of transaction locks */
-static int TxLockLWM = 4096*.4;	/* Low water mark for number of txLocks used */
-static int TxLockHWM = 4096*.8;	/* High water mark for number of txLocks used */
+module_param(nTxLock, int, 0);
+MODULE_PARM_DESC(nTxLock,
+		 "Number of transaction locks (default:4096, max:65536)");
+
+struct tblock *TxBlock;	        /* transaction block table */
+static int TxLockLWM;		/* Low water mark for number of txLocks used */
+static int TxLockHWM;		/* High water mark for number of txLocks used */
+static int TxLockVHWM;		/* Very High water mark */
 struct tlock *TxLock;           /* transaction lock table */
 
 
@@ -162,7 +169,6 @@ extern void lmSync(struct jfs_log *);
 extern int jfs_commit_inode(struct inode *, int);
 extern int jfs_stop_threads;
 
-struct task_struct *jfsCommitTask;
 extern struct completion jfsIOwait;
 
 /*
@@ -210,9 +216,9 @@ static lid_t txLockAlloc(void)
 		TXN_SLEEP(&TxAnchor.freelockwait);
 	TxAnchor.freelock = TxLock[lid].next;
 	HIGHWATERMARK(stattx.maxlid, lid);
-	if ((++TxAnchor.tlocksInUse > TxLockHWM) && (TxAnchor.TlocksLow == 0)) {
-		jfs_info("txLockAlloc TlocksLow");
-		TxAnchor.TlocksLow = 1;
+	if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
+		jfs_info("txLockAlloc tlocks low");
+		jfs_tlocks_low = 1;
 		wake_up(&jfs_sync_thread_wait);
 	}
 
@@ -224,9 +230,9 @@ static void txLockFree(lid_t lid)
 	TxLock[lid].next = TxAnchor.freelock;
 	TxAnchor.freelock = lid;
 	TxAnchor.tlocksInUse--;
-	if (TxAnchor.TlocksLow && (TxAnchor.tlocksInUse < TxLockLWM)) {
-		jfs_info("txLockFree TlocksLow no more");
-		TxAnchor.TlocksLow = 0;
+	if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
+		jfs_info("txLockFree jfs_tlocks_low no more");
+		jfs_tlocks_low = 0;
 		TXN_WAKEUP(&TxAnchor.lowlockwait);
 	}
 	TXN_WAKEUP(&TxAnchor.freelockwait);
@@ -245,12 +251,25 @@ int txInit(void)
 {
 	int k, size;
 
+	/* Verify tunable parameters */
+	if (nTxBlock < 16)
+		nTxBlock = 16;	/* No one should set it this low */
+	if (nTxBlock > 65536)
+		nTxBlock = 65536;
+	if (nTxLock < 256)
+		nTxLock = 256;	/* No one should set it this low */
+	if (nTxLock > 65536)
+		nTxLock = 65536;
 	/*
 	 * initialize transaction block (tblock) table
 	 *
 	 * transaction id (tid) = tblock index
 	 * tid = 0 is reserved.
 	 */
+	TxLockLWM = (nTxLock * 4) / 10;
+	TxLockHWM = (nTxLock * 8) / 10;
+	TxLockVHWM = (nTxLock * 9) / 10;
+
 	size = sizeof(struct tblock) * nTxBlock;
 	TxBlock = (struct tblock *) vmalloc(size);
 	if (TxBlock == NULL)
@@ -295,6 +314,9 @@ int txInit(void)
 	INIT_LIST_HEAD(&TxAnchor.anon_list);
 	INIT_LIST_HEAD(&TxAnchor.anon_list2);
 
+	LAZY_LOCK_INIT();
+	INIT_LIST_HEAD(&TxAnchor.unlock_queue);
+
 	stattx.maxlid = 1;	/* statistics */
 
 	return 0;
@@ -358,7 +380,7 @@ tid_t txBegin(struct super_block *sb, int flag)
 		 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
 		 * free tlocks)
 		 */
-		if (TxAnchor.TlocksLow) {
+		if (TxAnchor.tlocksInUse > TxLockVHWM) {
 			INCREMENT(TxStat.txBegin_lockslow);
 			TXN_SLEEP(&TxAnchor.lowlockwait);
 			goto retry;
@@ -450,7 +472,7 @@ void txBeginAnon(struct super_block *sb)
 	/*
 	 * Don't begin transaction if we're getting starved for tlocks
 	 */
-	if (TxAnchor.TlocksLow) {
+	if (TxAnchor.tlocksInUse > TxLockVHWM) {
 		INCREMENT(TxStat.txBeginAnon_lockslow);
 		TXN_SLEEP(&TxAnchor.lowlockwait);
 		goto retry;
@@ -2559,6 +2581,7 @@ void txFreelock(struct inode *ip)
 	if (!jfs_ip->atlhead)
 		return;
 
+	TXN_LOCK();
 	xtlck = (struct tlock *) &jfs_ip->atlhead;
 
 	while ((lid = xtlck->next)) {
@@ -2579,10 +2602,9 @@ void txFreelock(struct inode *ip)
 		/*
 		 * If inode was on anon_list, remove it
 		 */
-		TXN_LOCK();
 		list_del_init(&jfs_ip->anon_inode_list);
-		TXN_UNLOCK();
 	}
+	TXN_UNLOCK();
 }
 
 
@@ -2707,50 +2729,54 @@ int jfs_lazycommit(void *arg)
 	int WorkDone;
 	struct tblock *tblk;
 	unsigned long flags;
+	struct jfs_sb_info *sbi;
 
 	daemonize("jfsCommit");
 
-	jfsCommitTask = current;
-
-	LAZY_LOCK_INIT();
-	TxAnchor.unlock_queue = TxAnchor.unlock_tail = 0;
-
 	complete(&jfsIOwait);
 
 	do {
 		LAZY_LOCK(flags);
-restart:
+		while (!list_empty(&TxAnchor.unlock_queue)) {
 			WorkDone = 0;
-		while ((tblk = TxAnchor.unlock_queue)) {
+			list_for_each_entry(tblk, &TxAnchor.unlock_queue,
+					    cqueue) {
+
+				sbi = JFS_SBI(tblk->sb);
 				/*
-			 * We can't get ahead of user thread.  Spinning is
-			 * simpler than blocking/waking.  We shouldn't spin
-			 * very long, since user thread shouldn't be blocking
-			 * between lmGroupCommit & txEnd.
+				 * For each volume, the transactions must be
+				 * handled in order.  If another commit thread
+				 * is handling a tblk for this superblock,
+				 * skip it
 				 */
+				if (sbi->commit_state & IN_LAZYCOMMIT)
+					continue;
+
+				sbi->commit_state |= IN_LAZYCOMMIT;
 				WorkDone = 1;
 
 				/*
-			 * Remove first transaction from queue
+				 * Remove transaction from queue
 				 */
-			TxAnchor.unlock_queue = tblk->cqnext;
-			tblk->cqnext = 0;
-			if (TxAnchor.unlock_tail == tblk)
-				TxAnchor.unlock_tail = 0;
+				list_del(&tblk->cqueue);
 
 				LAZY_UNLOCK(flags);
 				txLazyCommit(tblk);
+				LAZY_LOCK(flags);
 
+				sbi->commit_state &= ~IN_LAZYCOMMIT;
 				/*
-			 * We can be running indefinitely if other processors
-			 * are adding transactions to this list
+				 * Don't continue in the for loop.  (We can't
+				 * anyway, it's unsafe!)  We want to go back to
+				 * the beginning of the list.
 				 */
-			cond_resched();
-			LAZY_LOCK(flags);
+				break;
 			}
 
-		if (WorkDone)
-			goto restart;
+			/* If there was nothing to do, don't continue */
+			if (!WorkDone)
+				break;
+		}
 
 		if (current->flags & PF_FREEZE) {
 			LAZY_UNLOCK(flags);
@@ -2767,7 +2793,7 @@ int jfs_lazycommit(void *arg)
 		}
 	} while (!jfs_stop_threads);
 
-	if (TxAnchor.unlock_queue)
+	if (!list_empty(&TxAnchor.unlock_queue))
 		jfs_err("jfs_lazycommit being killed w/pending transactions!");
 	else
 		jfs_info("jfs_lazycommit being killed\n");
@@ -2780,14 +2806,14 @@ void txLazyUnlock(struct tblock * tblk)
 
 	LAZY_LOCK(flags);
 
-	if (TxAnchor.unlock_tail)
-		TxAnchor.unlock_tail->cqnext = tblk;
-	else
-		TxAnchor.unlock_queue = tblk;
-	TxAnchor.unlock_tail = tblk;
-	tblk->cqnext = 0;
-	LAZY_UNLOCK(flags);
+	list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
+	/*
+	 * Don't wake up a commit thread if there is already one servicing
+	 * this superblock.
+	 */
+	if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT))
 		wake_up(&jfs_commit_thread_wait);
+	LAZY_UNLOCK(flags);
 }
 
 static void LogSyncRelease(struct metapage * mp)
@@ -2821,7 +2847,7 @@ static void LogSyncRelease(struct metapage * mp)
  *	completion
  *
  *	This does almost the same thing as jfs_sync below.  We don't
- *	worry about deadlocking when TlocksLow is set, since we would
+ *	worry about deadlocking when jfs_tlocks_low is set, since we would
  *	expect jfs_sync to get us out of that jam.
  */
 void txQuiesce(struct super_block *sb)
@@ -2912,7 +2938,7 @@ int jfs_sync(void *arg)
 		 * write each inode on the anonymous inode list
 		 */
 		TXN_LOCK();
-		while (TxAnchor.TlocksLow && !list_empty(&TxAnchor.anon_list)) {
+		while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
 			jfs_ip = list_entry(TxAnchor.anon_list.next,
 					    struct jfs_inode_info,
 					    anon_inode_list);
@@ -3008,18 +3034,16 @@ int jfs_txanchor_read(char *buffer, char **start, off_t offset, int length,
 		       "freelockwait = %s\n"
 		       "lowlockwait = %s\n"
 		       "tlocksInUse = %d\n"
-		       "TlocksLow = %d\n"
-		       "unlock_queue = 0x%p\n"
-		       "unlock_tail = 0x%p\n",
+		       "jfs_tlocks_low = %d\n"
+		       "unlock_queue is %sempty\n",
 		       TxAnchor.freetid,
 		       freewait,
 		       TxAnchor.freelock,
 		       freelockwait,
 		       lowlockwait,
 		       TxAnchor.tlocksInUse,
-		       TxAnchor.TlocksLow,
-		       TxAnchor.unlock_queue,
-		       TxAnchor.unlock_tail);
+		       jfs_tlocks_low,
+		       list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
 
 	begin = offset;
 	*start = buffer + begin;

fs/jfs/jfs_txnmgr.h

 /*
- *   Copyright (c) International Business Machines Corp., 2000-2002
+ *   Copyright (C) International Business Machines Corp., 2000-2004
  *
  *   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
@@ -53,7 +53,7 @@ struct tblock {
 	u32 logtid;		/* log transaction id */
 
 	/* commit management */
-	struct tblock *cqnext;	/* commit queue link */
+	struct list_head cqueue;	/* commit queue list */
 	s32 clsn;		/* commit lsn */
 	struct lbuf *bp;
 	s32 pn;			/* commit record log page number */
@@ -93,16 +93,16 @@ extern struct tblock *TxBlock;	/* transaction block table */
  *	transaction lock
  */
 struct tlock {
-	lid_t next;		/* index next lockword on tid locklist
+	lid_t next;		/* 2: index next lockword on tid locklist
 				 *          next lockword on freelist
 				 */
-	tid_t tid;		/* transaction id holding lock */
+	tid_t tid;		/* 2: transaction id holding lock */
 
 	u16 flag;		/* 2: lock control */
 	u16 type;		/* 2: log type */
 
-	struct metapage *mp;	/* 4: object page buffer locked */
-	struct inode *ip;	/* 4: object */
+	struct metapage *mp;	/* 4/8: object page buffer locked */
+	struct inode *ip;	/* 4/8: object */
 	/* (16) */
 
 	s16 lock[24];		/* 48: overlay area */
@@ -167,7 +167,7 @@ struct lv {
 #define	TLOCKLONG	28
 
 struct linelock {
-	u16 next;		/* 2: next linelock */
+	lid_t next;		/* 2: next linelock */
 
 	s8 maxcnt;		/* 1: */
 	s8 index;		/* 1: */
@@ -183,7 +183,7 @@ struct linelock {
 #define dt_lock	linelock
 
 struct xtlock {
-	u16 next;		/* 2: */
+	lid_t next;		/* 2: */
 
 	s8 maxcnt;		/* 1: */
 	s8 index;		/* 1: */
@@ -214,7 +214,7 @@ struct xtlock {
  * free maplock (i.e., number of maplock) in the tlock; 
  */
 struct maplock {
-	u16 next;		/* 2: */
+	lid_t next;		/* 2: */
 
 	u8 maxcnt;		/* 2: */
 	u8 index;		/* 2: next free maplock index */
@@ -242,7 +242,7 @@ struct maplock {
 #define	pxd_lock	maplock
 
 struct xdlistlock {
-	u16 next;		/* 2: */
+	lid_t next;		/* 2: */
 
 	u8 maxcnt;		/* 2: */
 	u8 index;		/* 2: */

fs/jfs/jfs_types.h

 /*
- *   Copyright (c) International Business Machines Corp., 2000-2002
+ *   Copyright (C) International Business Machines Corp., 2000-2004
  *
  *   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
@@ -34,9 +34,12 @@
 
 /*
  * transaction and lock id's
+ *
+ * Don't change these without carefully considering the impact on the
+ * size and alignment of all of the linelock variants
  */
-typedef uint tid_t;
-typedef uint lid_t;
+typedef u16 tid_t;
+typedef u16 lid_t;
 
 /*
  * Almost identical to Linux's timespec, but not quite

fs/jfs/jfs_umount.c

@@ -121,7 +121,10 @@ int jfs_umount(struct super_block *sb)
 	 * list (to signify skip logredo()).
 	 */
 	if (log) {		/* log = NULL if read-only mount */
-		rc = updateSuper(sb, FM_CLEAN);
+		updateSuper(sb, FM_CLEAN);
+
+		/* Restore default gfp_mask for bdev */
+		mapping_set_gfp_mask(bdev_mapping, GFP_USER);
 
 		/*
 		 * close log: 
@@ -168,5 +171,8 @@ int jfs_umount_rw(struct super_block *sb)
 
 	updateSuper(sb, FM_CLEAN);
 
+	/* Restore default gfp_mask for bdev */
+	mapping_set_gfp_mask(bdev_mapping, GFP_USER);
+
 	return lmLogClose(sb);
 }

fs/jfs/namei.c

@@ -135,7 +135,6 @@ static int jfs_create(struct inode *dip, struct dentry *dentry, int mode,
 
 	insert_inode_hash(ip);
 	mark_inode_dirty(ip);
-	d_instantiate(dentry, ip);
 
 	dip->i_ctime = dip->i_mtime = CURRENT_TIME;
 
@@ -150,7 +149,8 @@ static int jfs_create(struct inode *dip, struct dentry *dentry, int mode,
 	if (rc) {
 		ip->i_nlink = 0;
 		iput(ip);
-	}
+	} else
+		d_instantiate(dentry, ip);
 
       out2:
 	free_UCSname(&dname);
@@ -265,7 +265,6 @@ static int jfs_mkdir(struct inode *dip, struct dentry *dentry, int mode)
 
 	insert_inode_hash(ip);
 	mark_inode_dirty(ip);
-	d_instantiate(dentry, ip);
 
 	/* update parent directory inode */
 	dip->i_nlink++;		/* for '..' from child directory */
@@ -281,7 +280,8 @@ static int jfs_mkdir(struct inode *dip, struct dentry *dentry, int mode)
 	if (rc) {
 		ip->i_nlink = 0;
 		iput(ip);
-	}
+	} else
+		d_instantiate(dentry, ip);
 
       out2:
 	free_UCSname(&dname);
@@ -806,12 +806,14 @@ static int jfs_link(struct dentry *old_dentry,
 	ip->i_ctime = CURRENT_TIME;
 	mark_inode_dirty(dir);
 	atomic_inc(&ip->i_count);
-	d_instantiate(dentry, ip);
 
 	iplist[0] = ip;
 	iplist[1] = dir;
 	rc = txCommit(tid, 2, &iplist[0], 0);
 
+	if (!rc)
+		d_instantiate(dentry, ip);
+
       free_dname:
 	free_UCSname(&dname);
 
@@ -999,7 +1001,6 @@ static int jfs_symlink(struct inode *dip, struct dentry *dentry,
 
 	insert_inode_hash(ip);
 	mark_inode_dirty(ip);
-	d_instantiate(dentry, ip);
 
 	/*
 	 * commit update of parent directory and link object
@@ -1028,7 +1029,8 @@ static int jfs_symlink(struct inode *dip, struct dentry *dentry,
 	if (rc) {
 		ip->i_nlink = 0;
 		iput(ip);
-	}
+	} else
+		d_instantiate(dentry, ip);
 
       out2:
 	free_UCSname(&dname);
@@ -1368,7 +1370,6 @@ static int jfs_mknod(struct inode *dir, struct dentry *dentry,
 
 	insert_inode_hash(ip);
 	mark_inode_dirty(ip);
-	d_instantiate(dentry, ip);
 
 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 
@@ -1385,7 +1386,8 @@ static int jfs_mknod(struct inode *dir, struct dentry *dentry,
 	if (rc) {
 		ip->i_nlink = 0;
 		iput(ip);
-	}
+	} else
+		d_instantiate(dentry, ip);
 
       out1:
 	free_UCSname(&dname);

fs/jfs/super.c

@@ -23,6 +23,7 @@
 #include <linux/parser.h>
 #include <linux/completion.h>
 #include <linux/vfs.h>
+#include <linux/moduleparam.h>
 #include <asm/uaccess.h>
 
 #include "jfs_incore.h"
@@ -44,15 +45,20 @@ static struct super_operations jfs_super_operations;
 static struct export_operations jfs_export_operations;
 static struct file_system_type jfs_fs_type;
 
+#define MAX_COMMIT_THREADS 64
+static int commit_threads = 0;
+module_param(commit_threads, int, 0);
+MODULE_PARM_DESC(commit_threads, "Number of commit threads");
+
 int jfs_stop_threads;
 static pid_t jfsIOthread;
-static pid_t jfsCommitThread;
+static pid_t jfsCommitThread[MAX_COMMIT_THREADS];
 static pid_t jfsSyncThread;
 DECLARE_COMPLETION(jfsIOwait);
 
 #ifdef CONFIG_JFS_DEBUG
 int jfsloglevel = JFS_LOGLEVEL_WARN;
-MODULE_PARM(jfsloglevel, "i");
+module_param(jfsloglevel, int, 644);
 MODULE_PARM_DESC(jfsloglevel, "Specify JFS loglevel (0, 1 or 2)");
 #endif
 
@@ -564,6 +570,7 @@ static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 
 static int __init init_jfs_fs(void)
 {
+	int i;
 	int rc;
 
 	jfs_inode_cachep =
@@ -600,12 +607,23 @@ static int __init init_jfs_fs(void)
 	}
 	wait_for_completion(&jfsIOwait);	/* Wait until thread starts */
 
-	jfsCommitThread = kernel_thread(jfs_lazycommit, 0, CLONE_KERNEL);
-	if (jfsCommitThread < 0) {
-		jfs_err("init_jfs_fs: fork failed w/rc = %d", jfsCommitThread);
-		goto kill_iotask;
+	if (commit_threads < 1)
+		commit_threads = num_online_cpus();
+	else if (commit_threads > MAX_COMMIT_THREADS)
+		commit_threads = MAX_COMMIT_THREADS;
+
+	for (i = 0; i < commit_threads; i++) {
+		jfsCommitThread[i] = kernel_thread(jfs_lazycommit, 0,
+						   CLONE_KERNEL);
+		if (jfsCommitThread[i] < 0) {
+			jfs_err("init_jfs_fs: fork failed w/rc = %d",
+				jfsCommitThread[i]);
+			commit_threads = i;
+			goto kill_committask;
+		}
+		/* Wait until thread starts */
+		wait_for_completion(&jfsIOwait);
 	}
-	wait_for_completion(&jfsIOwait);	/* Wait until thread starts */
 
 	jfsSyncThread = kernel_thread(jfs_sync, 0, CLONE_KERNEL);
 	if (jfsSyncThread < 0) {
@@ -622,10 +640,10 @@ static int __init init_jfs_fs(void)
 
 kill_committask:
 	jfs_stop_threads = 1;
-	wake_up(&jfs_commit_thread_wait);
-	wait_for_completion(&jfsIOwait);	/* Wait for thread exit */
-kill_iotask:
-	jfs_stop_threads = 1;
+	wake_up_all(&jfs_commit_thread_wait);
+	for (i = 0; i < commit_threads; i++)
+		wait_for_completion(&jfsIOwait);
+
 	wake_up(&jfs_IO_thread_wait);
 	wait_for_completion(&jfsIOwait);	/* Wait for thread exit */
 end_txmngr:
@@ -639,6 +657,8 @@ static int __init init_jfs_fs(void)
 
 static void __exit exit_jfs_fs(void)
 {
+	int i;
+
 	jfs_info("exit_jfs_fs called");
 
 	jfs_stop_threads = 1;
@@ -646,8 +666,9 @@ static void __exit exit_jfs_fs(void)
 	metapage_exit();
 	wake_up(&jfs_IO_thread_wait);
 	wait_for_completion(&jfsIOwait);	/* Wait until IO thread exits */
-	wake_up(&jfs_commit_thread_wait);
-	wait_for_completion(&jfsIOwait);	/* Wait until Commit thread exits */
+	wake_up_all(&jfs_commit_thread_wait);
+	for (i = 0; i < commit_threads; i++)
+		wait_for_completion(&jfsIOwait);
 	wake_up(&jfs_sync_thread_wait);
 	wait_for_completion(&jfsIOwait);	/* Wait until Sync thread exits */
 #ifdef PROC_FS_JFS

fs/jfs/xattr.c

@@ -550,7 +550,8 @@ static int ea_get(struct inode *inode, struct ea_buffer *ea_buf, int min_size)
 	}
 	ea_buf->flag = EA_EXTENT;
 	ea_buf->mp = read_metapage(inode, addressDXD(&ji->ea),
-				   lengthDXD(&ji->ea), 1);
+				   lengthDXD(&ji->ea) << sb->s_blocksize_bits,
+				   1);
 	if (ea_buf->mp == NULL)
 		return -EIO;
 	ea_buf->xattr = ea_buf->mp->data;
@@ -737,11 +738,7 @@ static int can_set_xattr(struct inode *inode, const char *name,
 	    (!S_ISDIR(inode->i_mode) || inode->i_mode &S_ISVTX))
 		return -EPERM;
 
-#ifdef CONFIG_JFS_POSIX_ACL
-	return jfs_permission(inode, MAY_WRITE, NULL);
-#else
 	return permission(inode, MAY_WRITE, NULL);
-#endif
 }
 
 int __jfs_setxattr(struct inode *inode, const char *name, const void *value,
@@ -900,13 +897,9 @@ int jfs_setxattr(struct dentry *dentry, const char *name, const void *value,
 
 static int can_get_xattr(struct inode *inode, const char *name)
 {
-#ifdef CONFIG_JFS_POSIX_ACL
 	if(strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) == 0)
 		return 0;
-	return jfs_permission(inode, MAY_READ, NULL);
-#else
 	return permission(inode, MAY_READ, NULL);
-#endif
 }
 
 ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,