fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / fs / jfs / jfs_txnmgr.c
index f48fdea..6988a10 100644 (file)
@@ -1,19 +1,19 @@
 /*
- *   Copyright (C) International Business Machines Corp., 2000-2004
+ *   Copyright (C) International Business Machines Corp., 2000-2005
  *   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 
+ *   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 
+ *   along with this program;  if not, write to the Free Software
  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  */
 
  * hold on to mp+lock thru update of maps
  */
 
-
 #include <linux/fs.h>
 #include <linux/vmalloc.h>
 #include <linux/smp_lock.h>
 #include <linux/completion.h>
-#include <linux/suspend.h>
+#include <linux/freezer.h>
 #include <linux/module.h>
 #include <linux/moduleparam.h>
+#include <linux/kthread.h>
 #include "jfs_incore.h"
+#include "jfs_inode.h"
 #include "jfs_filsys.h"
 #include "jfs_metapage.h"
 #include "jfs_dinode.h"
@@ -80,7 +81,7 @@ static struct {
 int jfs_tlocks_low;            /* Indicates low number of available tlocks */
 
 #ifdef CONFIG_JFS_STATISTICS
-struct {
+static struct {
        uint txBegin;
        uint txBegin_barrier;
        uint txBegin_lockslow;
@@ -93,15 +94,15 @@ struct {
 } TxStat;
 #endif
 
-static int nTxBlock = 512;     /* number of transaction blocks */
+static int nTxBlock = -1;      /* number of transaction blocks */
 module_param(nTxBlock, int, 0);
 MODULE_PARM_DESC(nTxBlock,
-                "Number of transaction blocks (default:512, max:65536)");
+                "Number of transaction blocks (max:65536)");
 
-static int nTxLock = 4096;     /* number of transaction locks */
+static int nTxLock = -1;       /* number of transaction locks */
 module_param(nTxLock, int, 0);
 MODULE_PARM_DESC(nTxLock,
-                "Number of transaction locks (default:4096, max:65536)");
+                "Number of transaction locks (max:65536)");
 
 struct tblock *TxBlock;                /* transaction block table */
 static int TxLockLWM;          /* Low water mark for number of txLocks used */
@@ -109,11 +110,10 @@ static int TxLockHWM;             /* High water mark for number of txLocks used */
 static int TxLockVHWM;         /* Very High water mark */
 struct tlock *TxLock;           /* transaction lock table */
 
-
 /*
  *      transaction management lock
  */
-static spinlock_t jfsTxnLock = SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(jfsTxnLock);
 
 #define TXN_LOCK()              spin_lock(&jfsTxnLock)
 #define TXN_UNLOCK()            spin_unlock(&jfsTxnLock)
@@ -122,8 +122,8 @@ static spinlock_t jfsTxnLock = SPIN_LOCK_UNLOCKED;
 #define LAZY_LOCK(flags)       spin_lock_irqsave(&TxAnchor.LazyLock, flags)
 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
 
-DECLARE_WAIT_QUEUE_HEAD(jfs_sync_thread_wait);
-DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
+static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
+static int jfs_commit_thread_waking;
 
 /*
  * Retry logic exist outside these macros to protect from spurrious wakeups.
@@ -135,7 +135,7 @@ static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
        add_wait_queue(event, &wait);
        set_current_state(TASK_UNINTERRUPTIBLE);
        TXN_UNLOCK();
-       schedule();
+       io_schedule();
        current->state = TASK_RUNNING;
        remove_wait_queue(event, &wait);
 }
@@ -148,11 +148,10 @@ static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
 
 #define TXN_WAKEUP(event) wake_up_all(event)
 
-
 /*
  *      statistics
  */
-struct {
+static struct {
        tid_t maxtid;           /* 4: biggest tid ever used */
        lid_t maxlid;           /* 4: biggest lid ever used */
        int ntid;               /* 4: # of transactions performed */
@@ -160,17 +159,6 @@ struct {
        int waitlock;           /* 4: # of tlock wait */
 } stattx;
 
-
-/*
- * external references
- */
-extern int lmGroupCommit(struct jfs_log *, struct tblock *);
-extern void lmSync(struct jfs_log *);
-extern int jfs_commit_inode(struct inode *, int);
-extern int jfs_stop_threads;
-
-extern struct completion jfsIOwait;
-
 /*
  * forward references
  */
@@ -219,7 +207,7 @@ static lid_t txLockAlloc(void)
        if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
                jfs_info("txLockAlloc tlocks low");
                jfs_tlocks_low = 1;
-               wake_up(&jfs_sync_thread_wait);
+               wake_up_process(jfsSyncThread);
        }
 
        return lid;
@@ -227,6 +215,7 @@ static lid_t txLockAlloc(void)
 
 static void txLockFree(lid_t lid)
 {
+       TxLock[lid].tid = 0;
        TxLock[lid].next = TxAnchor.freelock;
        TxAnchor.freelock = lid;
        TxAnchor.tlocksInUse--;
@@ -250,6 +239,25 @@ static void txLockFree(lid_t lid)
 int txInit(void)
 {
        int k, size;
+       struct sysinfo si;
+
+       /* Set defaults for nTxLock and nTxBlock if unset */
+
+       if (nTxLock == -1) {
+               if (nTxBlock == -1) {
+                       /* Base default on memory size */
+                       si_meminfo(&si);
+                       if (si.totalram > (256 * 1024)) /* 1 GB */
+                               nTxLock = 64 * 1024;
+                       else
+                               nTxLock = si.totalram >> 2;
+               } else if (nTxBlock > (8 * 1024))
+                       nTxLock = 64 * 1024;
+               else
+                       nTxLock = nTxBlock << 3;
+       }
+       if (nTxBlock == -1)
+               nTxBlock = nTxLock >> 3;
 
        /* Verify tunable parameters */
        if (nTxBlock < 16)
@@ -260,6 +268,9 @@ int txInit(void)
                nTxLock = 256;  /* No one should set it this low */
        if (nTxLock > 65536)
                nTxLock = 65536;
+
+       printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
+              nTxBlock, nTxLock);
        /*
         * initialize transaction block (tblock) table
         *
@@ -267,11 +278,11 @@ int txInit(void)
         * tid = 0 is reserved.
         */
        TxLockLWM = (nTxLock * 4) / 10;
-       TxLockHWM = (nTxLock * 8) / 10;
-       TxLockVHWM = (nTxLock * 9) / 10;
+       TxLockHWM = (nTxLock * 7) / 10;
+       TxLockVHWM = (nTxLock * 8) / 10;
 
        size = sizeof(struct tblock) * nTxBlock;
-       TxBlock = (struct tblock *) vmalloc(size);
+       TxBlock = vmalloc(size);
        if (TxBlock == NULL)
                return -ENOMEM;
 
@@ -296,7 +307,7 @@ int txInit(void)
         * tlock id = 0 is reserved.
         */
        size = sizeof(struct tlock) * nTxLock;
-       TxLock = (struct tlock *) vmalloc(size);
+       TxLock = vmalloc(size);
        if (TxLock == NULL) {
                vfree(TxBlock);
                return -ENOMEM;
@@ -330,12 +341,11 @@ int txInit(void)
 void txExit(void)
 {
        vfree(TxLock);
-       TxLock = 0;
+       TxLock = NULL;
        vfree(TxBlock);
-       TxBlock = 0;
+       TxBlock = NULL;
 }
 
-
 /*
  * NAME:        txBegin()
  *
@@ -437,7 +447,6 @@ tid_t txBegin(struct super_block *sb, int flag)
        return t;
 }
 
-
 /*
  * NAME:        txBeginAnon()
  *
@@ -480,7 +489,6 @@ void txBeginAnon(struct super_block *sb)
        TXN_UNLOCK();
 }
 
-
 /*
  *      txEnd()
  *
@@ -544,8 +552,10 @@ void txEnd(tid_t tid)
                 * synchronize with logsync barrier
                 */
                if (test_bit(log_SYNCBARRIER, &log->flag)) {
-                       /* forward log syncpt */
-                       /* lmSync(log); */
+                       TXN_UNLOCK();
+
+                       /* write dirty metadata & forward log syncpt */
+                       jfs_syncpt(log, 1);
 
                        jfs_info("log barrier off: 0x%x", log->lsn);
 
@@ -554,18 +564,19 @@ void txEnd(tid_t tid)
 
                        /* wakeup all waitors for logsync barrier */
                        TXN_WAKEUP(&log->syncwait);
+
+                       goto wakeup;
                }
        }
 
+       TXN_UNLOCK();
+wakeup:
        /*
         * wakeup all waitors for a free tblock
         */
        TXN_WAKEUP(&TxAnchor.freewait);
-
-       TXN_UNLOCK();
 }
 
-
 /*
  *      txLock()
  *
@@ -611,8 +622,10 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
 
        /* is page locked by the requester transaction ? */
        tlck = lid_to_tlock(lid);
-       if ((xtid = tlck->tid) == tid)
+       if ((xtid = tlck->tid) == tid) {
+               TXN_UNLOCK();
                goto grantLock;
+       }
 
        /*
         * is page locked by anonymous transaction/lock ?
@@ -627,6 +640,7 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
         */
        if (xtid == 0) {
                tlck->tid = tid;
+               TXN_UNLOCK();
                tblk = tid_to_tblock(tid);
                /*
                 * The order of the tlocks in the transaction is important
@@ -643,7 +657,9 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
                                /* only anonymous txn.
                                 * Remove from anon_list
                                 */
+                               TXN_LOCK();
                                list_del_init(&jfs_ip->anon_inode_list);
+                               TXN_UNLOCK();
                        }
                        jfs_ip->atlhead = tlck->next;
                } else {
@@ -684,17 +700,18 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
         */
        tlck->tid = tid;
 
+       TXN_UNLOCK();
+
        /* mark tlock for meta-data page */
        if (mp->xflag & COMMIT_PAGE) {
 
                tlck->flag = tlckPAGELOCK;
 
                /* mark the page dirty and nohomeok */
-               mark_metapage_dirty(mp);
-               atomic_inc(&mp->nohomeok);
+               metapage_nohomeok(mp);
 
                jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
-                        mp, atomic_read(&mp->nohomeok), tid, tlck);
+                        mp, mp->nohomeok, tid, tlck);
 
                /* if anonymous transaction, and buffer is on the group
                 * commit synclist, mark inode to show this.  This will
@@ -708,6 +725,9 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
        else
                tlck->flag = tlckINODELOCK;
 
+       if (S_ISDIR(ip->i_mode))
+               tlck->flag |= tlckDIRECTORY;
+
        tlck->type = 0;
 
        /* bind the tlock and the page */
@@ -740,8 +760,10 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
                if (tlck->next == 0) {
                        /* This inode's first anonymous transaction */
                        jfs_ip->atltail = lid;
+                       TXN_LOCK();
                        list_add_tail(&jfs_ip->anon_inode_list,
                                      &TxAnchor.anon_list);
+                       TXN_UNLOCK();
                }
        }
 
@@ -799,8 +821,6 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
       grantLock:
        tlck->type |= type;
 
-       TXN_UNLOCK();
-
        return tlck;
 
        /*
@@ -819,17 +839,24 @@ struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
                BUG();
        }
        INCREMENT(stattx.waitlock);     /* statistics */
+       TXN_UNLOCK();
        release_metapage(mp);
+       TXN_LOCK();
+       xtid = tlck->tid;       /* reacquire after dropping TXN_LOCK */
 
        jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
                 tid, xtid, lid);
-       TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
+
+       /* Recheck everything since dropping TXN_LOCK */
+       if (xtid && (tlck->mp == mp) && (mp->lid == lid))
+               TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
+       else
+               TXN_UNLOCK();
        jfs_info("txLock: awakened     tid = %d, lid = %d", tid, lid);
 
        return NULL;
 }
 
-
 /*
  * NAME:        txRelease()
  *
@@ -869,7 +896,6 @@ static void txRelease(struct tblock * tblk)
        TXN_UNLOCK();
 }
 
-
 /*
  * NAME:        txUnlock()
  *
@@ -884,6 +910,7 @@ static void txUnlock(struct tblock * tblk)
        struct metapage *mp;
        struct jfs_log *log;
        int difft, diffp;
+       unsigned long flags;
 
        jfs_info("txUnlock: tblk = 0x%p", tblk);
        log = JFS_SBI(tblk->sb)->log;
@@ -903,19 +930,14 @@ static void txUnlock(struct tblock * tblk)
                        assert(mp->xflag & COMMIT_PAGE);
 
                        /* hold buffer
-                        *
-                        * It's possible that someone else has the metapage.
-                        * The only things were changing are nohomeok, which
-                        * is handled atomically, and clsn which is protected
-                        * by the LOGSYNC_LOCK.
                         */
-                       hold_metapage(mp, 1);
+                       hold_metapage(mp);
 
-                       assert(atomic_read(&mp->nohomeok) > 0);
-                       atomic_dec(&mp->nohomeok);
+                       assert(mp->nohomeok > 0);
+                       _metapage_homeok(mp);
 
                        /* inherit younger/larger clsn */
-                       LOGSYNC_LOCK(log);
+                       LOGSYNC_LOCK(log, flags);
                        if (mp->clsn) {
                                logdiff(difft, tblk->clsn, log);
                                logdiff(diffp, mp->clsn, log);
@@ -923,16 +945,11 @@ static void txUnlock(struct tblock * tblk)
                                        mp->clsn = tblk->clsn;
                        } else
                                mp->clsn = tblk->clsn;
-                       LOGSYNC_UNLOCK(log);
+                       LOGSYNC_UNLOCK(log, flags);
 
                        assert(!(tlck->flag & tlckFREEPAGE));
 
-                       if (tlck->flag & tlckWRITEPAGE) {
-                               write_metapage(mp);
-                       } else {
-                               /* release page which has been forced */
-                               release_metapage(mp);
-                       }
+                       put_metapage(mp);
                }
 
                /* insert tlock, and linelock(s) of the tlock if any,
@@ -959,14 +976,13 @@ static void txUnlock(struct tblock * tblk)
         * has been inserted in logsync list at txUpdateMap())
         */
        if (tblk->lsn) {
-               LOGSYNC_LOCK(log);
+               LOGSYNC_LOCK(log, flags);
                log->count--;
                list_del(&tblk->synclist);
-               LOGSYNC_UNLOCK(log);
+               LOGSYNC_UNLOCK(log, flags);
        }
 }
 
-
 /*
  *      txMaplock()
  *
@@ -996,6 +1012,8 @@ struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
 
        /* bind the tlock and the object */
        tlck->flag = tlckINODELOCK;
+       if (S_ISDIR(ip->i_mode))
+               tlck->flag |= tlckDIRECTORY;
        tlck->ip = ip;
        tlck->mp = NULL;
 
@@ -1039,7 +1057,6 @@ struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
        return tlck;
 }
 
-
 /*
  *      txLinelock()
  *
@@ -1065,6 +1082,8 @@ struct linelock *txLinelock(struct linelock * tlock)
        linelock->flag = tlckLINELOCK;
        linelock->maxcnt = TLOCKLONG;
        linelock->index = 0;
+       if (tlck->flag & tlckDIRECTORY)
+               linelock->flag |= tlckDIRECTORY;
 
        /* append linelock after tlock */
        linelock->next = tlock->next;
@@ -1073,8 +1092,6 @@ struct linelock *txLinelock(struct linelock * tlock)
        return linelock;
 }
 
-
-
 /*
  *              transaction commit management
  *              -----------------------------
@@ -1214,10 +1231,8 @@ int txCommit(tid_t tid,          /* transaction identifier */
                 * when we don't need to worry about it at all.
                 *
                 * if ((!S_ISDIR(ip->i_mode))
-                *    && (tblk->flag & COMMIT_DELETE) == 0) {
-                *      filemap_fdatawrite(ip->i_mapping);
-                *      filemap_fdatawait(ip->i_mapping);
-                * }
+                *    && (tblk->flag & COMMIT_DELETE) == 0)
+                *      filemap_write_and_wait(ip->i_mapping);
                 */
 
                /*
@@ -1343,7 +1358,6 @@ int txCommit(tid_t tid,           /* transaction identifier */
        return rc;
 }
 
-
 /*
  * NAME:        txLog()
  *
@@ -1407,7 +1421,6 @@ static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
        return rc;
 }
 
-
 /*
  *      diLog()
  *
@@ -1435,7 +1448,6 @@ static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        if (tlck->type & tlckENTRY) {
                /* log after-image for logredo(): */
                lrd->type = cpu_to_le16(LOG_REDOPAGE);
-//              *pxd = mp->cm_pxd;
                PXDaddress(pxd, mp->index);
                PXDlength(pxd,
                          mp->logical_size >> tblk->sb->s_blocksize_bits);
@@ -1522,7 +1534,6 @@ static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        return rc;
 }
 
-
 /*
  *      dataLog()
  *
@@ -1545,16 +1556,16 @@ static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        /* log after-image for logredo(): */
        lrd->type = cpu_to_le16(LOG_REDOPAGE);
 
-       if (JFS_IP(tlck->ip)->next_index < MAX_INLINE_DIRTABLE_ENTRY) {
+       if (jfs_dirtable_inline(tlck->ip)) {
                /*
                 * The table has been truncated, we've must have deleted
                 * the last entry, so don't bother logging this
                 */
                mp->lid = 0;
-               hold_metapage(mp, 0);
-               atomic_dec(&mp->nohomeok);
+               grab_metapage(mp);
+               metapage_homeok(mp);
                discard_metapage(mp);
-               tlck->mp = 0;
+               tlck->mp = NULL;
                return 0;
        }
 
@@ -1569,7 +1580,6 @@ static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        return 0;
 }
 
-
 /*
  *      dtLog()
  *
@@ -1609,7 +1619,6 @@ static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                        lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
                else
                        lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
-//              *pxd = mp->cm_pxd;
                PXDaddress(pxd, mp->index);
                PXDlength(pxd,
                          mp->logical_size >> tblk->sb->s_blocksize_bits);
@@ -1674,7 +1683,6 @@ static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        return;
 }
 
-
 /*
  *      xtLog()
  *
@@ -1690,7 +1698,7 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        struct maplock *maplock;
        struct xdlistlock *xadlock;
        struct pxd_lock *pxdlock;
-       pxd_t *pxd;
+       pxd_t *page_pxd;
        int next, lwm, hwm;
 
        ip = tlck->ip;
@@ -1700,7 +1708,7 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
        lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
 
-       pxd = &lrd->log.redopage.pxd;
+       page_pxd = &lrd->log.redopage.pxd;
 
        if (tlck->type & tlckBTROOT) {
                lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
@@ -1730,9 +1738,8 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                 * applying the after-image to the meta-data page.
                 */
                lrd->type = cpu_to_le16(LOG_REDOPAGE);
-//              *pxd = mp->cm_pxd;
-               PXDaddress(pxd, mp->index);
-               PXDlength(pxd,
+               PXDaddress(page_pxd, mp->index);
+               PXDlength(page_pxd,
                          mp->logical_size >> tblk->sb->s_blocksize_bits);
                lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
 
@@ -1754,25 +1761,31 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                tlck->flag |= tlckUPDATEMAP;
                xadlock->flag = mlckALLOCXADLIST;
                xadlock->count = next - lwm;
-               if ((xadlock->count <= 2) && (tblk->xflag & COMMIT_LAZY)) {
+               if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
                        int i;
+                       pxd_t *pxd;
                        /*
                         * Lazy commit may allow xtree to be modified before
                         * txUpdateMap runs.  Copy xad into linelock to
                         * preserve correct data.
+                        *
+                        * We can fit twice as may pxd's as xads in the lock
                         */
-                       xadlock->xdlist = &xtlck->pxdlock;
-                       memcpy(xadlock->xdlist, &p->xad[lwm],
-                              sizeof(xad_t) * xadlock->count);
-
-                       for (i = 0; i < xadlock->count; i++)
+                       xadlock->flag = mlckALLOCPXDLIST;
+                       pxd = xadlock->xdlist = &xtlck->pxdlock;
+                       for (i = 0; i < xadlock->count; i++) {
+                               PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
+                               PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
                                p->xad[lwm + i].flag &=
                                    ~(XAD_NEW | XAD_EXTENDED);
+                               pxd++;
+                       }
                } else {
                        /*
                         * xdlist will point to into inode's xtree, ensure
                         * that transaction is not committed lazily.
                         */
+                       xadlock->flag = mlckALLOCXADLIST;
                        xadlock->xdlist = &p->xad[lwm];
                        tblk->xflag &= ~COMMIT_LAZY;
                }
@@ -1814,8 +1827,8 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                if (tblk->xflag & COMMIT_TRUNCATE) {
                        /* write NOREDOPAGE for the page */
                        lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
-                       PXDaddress(pxd, mp->index);
-                       PXDlength(pxd,
+                       PXDaddress(page_pxd, mp->index);
+                       PXDlength(page_pxd,
                                  mp->logical_size >> tblk->sb->
                                  s_blocksize_bits);
                        lrd->backchain =
@@ -1850,22 +1863,32 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                 * deleted page itself;
                 */
                tlck->flag |= tlckUPDATEMAP;
-               xadlock->flag = mlckFREEXADLIST;
                xadlock->count = hwm - XTENTRYSTART + 1;
-               if ((xadlock->count <= 2) && (tblk->xflag & COMMIT_LAZY)) {
+               if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
+                       int i;
+                       pxd_t *pxd;
                        /*
                         * Lazy commit may allow xtree to be modified before
                         * txUpdateMap runs.  Copy xad into linelock to
                         * preserve correct data.
+                        *
+                        * We can fit twice as may pxd's as xads in the lock
                         */
-                       xadlock->xdlist = &xtlck->pxdlock;
-                       memcpy(xadlock->xdlist, &p->xad[XTENTRYSTART],
-                              sizeof(xad_t) * xadlock->count);
+                       xadlock->flag = mlckFREEPXDLIST;
+                       pxd = xadlock->xdlist = &xtlck->pxdlock;
+                       for (i = 0; i < xadlock->count; i++) {
+                               PXDaddress(pxd,
+                                       addressXAD(&p->xad[XTENTRYSTART + i]));
+                               PXDlength(pxd,
+                                       lengthXAD(&p->xad[XTENTRYSTART + i]));
+                               pxd++;
+                       }
                } else {
                        /*
                         * xdlist will point to into inode's xtree, ensure
                         * that transaction is not committed lazily.
                         */
+                       xadlock->flag = mlckFREEXADLIST;
                        xadlock->xdlist = &p->xad[XTENTRYSTART];
                        tblk->xflag &= ~COMMIT_LAZY;
                }
@@ -1896,7 +1919,7 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
         * header ?
         */
        if (tlck->type & tlckTRUNCATE) {
-               pxd_t tpxd;     /* truncated extent of xad */
+               pxd_t pxd;      /* truncated extent of xad */
                int twm;
 
                /*
@@ -1925,8 +1948,9 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                 * applying the after-image to the meta-data page.
                 */
                lrd->type = cpu_to_le16(LOG_REDOPAGE);
-               PXDaddress(pxd, mp->index);
-               PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
+               PXDaddress(page_pxd, mp->index);
+               PXDlength(page_pxd,
+                         mp->logical_size >> tblk->sb->s_blocksize_bits);
                lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
 
                /*
@@ -1944,7 +1968,7 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                        lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
                        lrd->log.updatemap.nxd = cpu_to_le16(1);
                        lrd->log.updatemap.pxd = pxdlock->pxd;
-                       tpxd = pxdlock->pxd;    /* save to format maplock */
+                       pxd = pxdlock->pxd;     /* save to format maplock */
                        lrd->backchain =
                            cpu_to_le32(lmLog(log, tblk, lrd, NULL));
                }
@@ -2002,8 +2026,6 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                 * truncate entry XAD[twm == next - 1]:
                 */
                if (twm == next - 1) {
-                       struct pxd_lock *pxdlock;
-
                        /* format a maplock for txUpdateMap() to update bmap
                         * to free truncated delta extent of the truncated
                         * entry XAD[next - 1];
@@ -2013,7 +2035,7 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
                        pxdlock = (struct pxd_lock *) xadlock;
                        pxdlock->flag = mlckFREEPXD;
                        pxdlock->count = 1;
-                       pxdlock->pxd = tpxd;
+                       pxdlock->pxd = pxd;
 
                        jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
                                 "hwm:%d", ip, mp, pxdlock->count, hwm);
@@ -2046,14 +2068,13 @@ static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        return;
 }
 
-
 /*
  *      mapLog()
  *
  * function:    log from maplock of freed data extents;
  */
-void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
-           struct tlock * tlck)
+static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
+                  struct tlock * tlck)
 {
        struct pxd_lock *pxdlock;
        int i, nlock;
@@ -2133,7 +2154,6 @@ void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
        }
 }
 
-
 /*
  *      txEA()
  *
@@ -2186,14 +2206,13 @@ void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
        }
 }
 
-
 /*
  *      txForce()
  *
  * function: synchronously write pages locked by transaction
  *              after txLog() but before txUpdateMap();
  */
-void txForce(struct tblock * tblk)
+static void txForce(struct tblock * tblk)
 {
        struct tlock *tlck;
        lid_t lid, next;
@@ -2231,7 +2250,8 @@ void txForce(struct tblock * tblk)
                                tlck->flag &= ~tlckWRITEPAGE;
 
                                /* do not release page to freelist */
-
+                               force_metapage(mp);
+#if 0
                                /*
                                 * The "right" thing to do here is to
                                 * synchronously write the metadata.
@@ -2243,15 +2263,15 @@ void txForce(struct tblock * tblk)
                                 * we can get by with synchronously writing
                                 * the pages when they are released.
                                 */
-                               assert(atomic_read(&mp->nohomeok));
+                               assert(mp->nohomeok);
                                set_bit(META_dirty, &mp->flag);
                                set_bit(META_sync, &mp->flag);
+#endif
                        }
                }
        }
 }
 
-
 /*
  *      txUpdateMap()
  *
@@ -2270,7 +2290,7 @@ static void txUpdateMap(struct tblock * tblk)
        struct pxd_lock pxdlock;
        int maptype;
        int k, nlock;
-       struct metapage *mp = 0;
+       struct metapage *mp = NULL;
 
        ipimap = JFS_SBI(tblk->sb)->ipimap;
 
@@ -2305,7 +2325,7 @@ static void txUpdateMap(struct tblock * tblk)
                         */
                        mp = tlck->mp;
                        ASSERT(mp->xflag & COMMIT_PAGE);
-                       hold_metapage(mp, 0);
+                       grab_metapage(mp);
                }
 
                /*
@@ -2341,7 +2361,7 @@ static void txUpdateMap(struct tblock * tblk)
                         */
                        else {  /* (maplock->flag & mlckFREE) */
 
-                               if (S_ISDIR(tlck->ip->i_mode))
+                               if (tlck->flag & tlckDIRECTORY)
                                        txFreeMap(ipimap, maplock,
                                                  tblk, COMMIT_PWMAP);
                                else
@@ -2355,10 +2375,10 @@ static void txUpdateMap(struct tblock * tblk)
                                ASSERT(mp->lid == lid);
                                tlck->mp->lid = 0;
                        }
-                       assert(atomic_read(&mp->nohomeok) == 1);
-                       atomic_dec(&mp->nohomeok);
+                       assert(mp->nohomeok == 1);
+                       metapage_homeok(mp);
                        discard_metapage(mp);
-                       tlck->mp = 0;
+                       tlck->mp = NULL;
                }
        }
        /*
@@ -2371,8 +2391,7 @@ static void txUpdateMap(struct tblock * tblk)
         * unlock mapper/write lock
         */
        if (tblk->xflag & COMMIT_CREATE) {
-               diUpdatePMap(ipimap, tblk->ino, FALSE, tblk);
-               ipimap->i_state |= I_DIRTY;
+               diUpdatePMap(ipimap, tblk->ino, false, tblk);
                /* update persistent block allocation map
                 * for the allocation of inode extent;
                 */
@@ -2382,13 +2401,11 @@ static void txUpdateMap(struct tblock * tblk)
                txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
        } else if (tblk->xflag & COMMIT_DELETE) {
                ip = tblk->u.ip;
-               diUpdatePMap(ipimap, ip->i_ino, TRUE, tblk);
-               ipimap->i_state |= I_DIRTY;
+               diUpdatePMap(ipimap, ip->i_ino, true, tblk);
                iput(ip);
        }
 }
 
-
 /*
  *      txAllocPMap()
  *
@@ -2432,7 +2449,7 @@ static void txAllocPMap(struct inode *ip, struct maplock * maplock,
                        if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
                                xaddr = addressXAD(xad);
                                xlen = lengthXAD(xad);
-                               dbUpdatePMap(ipbmap, FALSE, xaddr,
+                               dbUpdatePMap(ipbmap, false, xaddr,
                                             (s64) xlen, tblk);
                                xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
                                jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
@@ -2443,7 +2460,7 @@ static void txAllocPMap(struct inode *ip, struct maplock * maplock,
                pxdlock = (struct pxd_lock *) maplock;
                xaddr = addressPXD(&pxdlock->pxd);
                xlen = lengthPXD(&pxdlock->pxd);
-               dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen, tblk);
+               dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
                jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
        } else {                /* (maplock->flag & mlckALLOCPXDLIST) */
 
@@ -2452,7 +2469,7 @@ static void txAllocPMap(struct inode *ip, struct maplock * maplock,
                for (n = 0; n < pxdlistlock->count; n++, pxd++) {
                        xaddr = addressPXD(pxd);
                        xlen = lengthPXD(pxd);
-                       dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen,
+                       dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
                                     tblk);
                        jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
                                 (ulong) xaddr, xlen);
@@ -2460,7 +2477,6 @@ static void txAllocPMap(struct inode *ip, struct maplock * maplock,
        }
 }
 
-
 /*
  *      txFreeMap()
  *
@@ -2495,7 +2511,7 @@ void txFreeMap(struct inode *ip,
                                if (!(xad->flag & XAD_NEW)) {
                                        xaddr = addressXAD(xad);
                                        xlen = lengthXAD(xad);
-                                       dbUpdatePMap(ipbmap, TRUE, xaddr,
+                                       dbUpdatePMap(ipbmap, true, xaddr,
                                                     (s64) xlen, tblk);
                                        jfs_info("freePMap: xaddr:0x%lx "
                                                 "xlen:%d",
@@ -2506,7 +2522,7 @@ void txFreeMap(struct inode *ip,
                        pxdlock = (struct pxd_lock *) maplock;
                        xaddr = addressPXD(&pxdlock->pxd);
                        xlen = lengthPXD(&pxdlock->pxd);
-                       dbUpdatePMap(ipbmap, TRUE, xaddr, (s64) xlen,
+                       dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
                                     tblk);
                        jfs_info("freePMap: xaddr:0x%lx xlen:%d",
                                 (ulong) xaddr, xlen);
@@ -2517,7 +2533,7 @@ void txFreeMap(struct inode *ip,
                        for (n = 0; n < pxdlistlock->count; n++, pxd++) {
                                xaddr = addressPXD(pxd);
                                xlen = lengthPXD(pxd);
-                               dbUpdatePMap(ipbmap, TRUE, xaddr,
+                               dbUpdatePMap(ipbmap, true, xaddr,
                                             (s64) xlen, tblk);
                                jfs_info("freePMap: xaddr:0x%lx xlen:%d",
                                         (ulong) xaddr, xlen);
@@ -2562,7 +2578,6 @@ void txFreeMap(struct inode *ip,
        }
 }
 
-
 /*
  *      txFreelock()
  *
@@ -2580,7 +2595,7 @@ void txFreelock(struct inode *ip)
        TXN_LOCK();
        xtlck = (struct tlock *) &jfs_ip->atlhead;
 
-       while ((lid = xtlck->next)) {
+       while ((lid = xtlck->next) != 0) {
                tlck = lid_to_tlock(lid);
                if (tlck->flag & tlckFREELOCK) {
                        xtlck->next = tlck->next;
@@ -2603,7 +2618,6 @@ void txFreelock(struct inode *ip)
        TXN_UNLOCK();
 }
 
-
 /*
  *      txAbort()
  *
@@ -2622,8 +2636,6 @@ void txAbort(tid_t tid, int dirty)
        struct tblock *tblk = tid_to_tblock(tid);
        struct tlock *tlck;
 
-       jfs_warn("txAbort: tid:%d dirty:0x%x", tid, dirty);
-
        /*
         * free tlocks of the transaction
         */
@@ -2729,12 +2741,9 @@ int jfs_lazycommit(void *arg)
        unsigned long flags;
        struct jfs_sb_info *sbi;
 
-       daemonize("jfsCommit");
-
-       complete(&jfsIOwait);
-
        do {
                LAZY_LOCK(flags);
+               jfs_commit_thread_waking = 0;   /* OK to wake another thread */
                while (!list_empty(&TxAnchor.unlock_queue)) {
                        WorkDone = 0;
                        list_for_each_entry(tblk, &TxAnchor.unlock_queue,
@@ -2775,10 +2784,12 @@ int jfs_lazycommit(void *arg)
                        if (!WorkDone)
                                break;
                }
+               /* In case a wakeup came while all threads were active */
+               jfs_commit_thread_waking = 0;
 
-               if (current->flags & PF_FREEZE) {
+               if (freezing(current)) {
                        LAZY_UNLOCK(flags);
-                       refrigerator(PF_FREEZE);
+                       refrigerator();
                } else {
                        DECLARE_WAITQUEUE(wq, current);
 
@@ -2789,13 +2800,13 @@ int jfs_lazycommit(void *arg)
                        current->state = TASK_RUNNING;
                        remove_wait_queue(&jfs_commit_thread_wait, &wq);
                }
-       } while (!jfs_stop_threads);
+       } while (!kthread_should_stop());
 
        if (!list_empty(&TxAnchor.unlock_queue))
                jfs_err("jfs_lazycommit being killed w/pending transactions!");
        else
                jfs_info("jfs_lazycommit being killed\n");
-       complete_and_exit(&jfsIOwait, 0);
+       return 0;
 }
 
 void txLazyUnlock(struct tblock * tblk)
@@ -2807,10 +2818,13 @@ void txLazyUnlock(struct tblock * tblk)
        list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
        /*
         * Don't wake up a commit thread if there is already one servicing
-        * this superblock.
+        * this superblock, or if the last one we woke up hasn't started yet.
         */
-       if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT))
+       if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
+           !jfs_commit_thread_waking) {
+               jfs_commit_thread_waking = 1;
                wake_up(&jfs_commit_thread_wait);
+       }
        LAZY_UNLOCK(flags);
 }
 
@@ -2818,24 +2832,9 @@ static void LogSyncRelease(struct metapage * mp)
 {
        struct jfs_log *log = mp->log;
 
-       assert(atomic_read(&mp->nohomeok));
+       assert(mp->nohomeok);
        assert(log);
-       atomic_dec(&mp->nohomeok);
-
-       if (atomic_read(&mp->nohomeok))
-               return;
-
-       hold_metapage(mp, 0);
-
-       LOGSYNC_LOCK(log);
-       mp->log = NULL;
-       mp->lsn = 0;
-       mp->clsn = 0;
-       log->count--;
-       list_del_init(&mp->synclist);
-       LOGSYNC_UNLOCK(log);
-
-       release_metapage(mp);
+       metapage_homeok(mp);
 }
 
 /*
@@ -2871,10 +2870,10 @@ restart:
                 */
                TXN_UNLOCK();
                tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
-               down(&jfs_ip->commit_sem);
+               mutex_lock(&jfs_ip->commit_mutex);
                txCommit(tid, 1, &ip, 0);
                txEnd(tid);
-               up(&jfs_ip->commit_sem);
+               mutex_unlock(&jfs_ip->commit_mutex);
                /*
                 * Just to be safe.  I don't know how
                 * long we can run without blocking
@@ -2927,10 +2926,6 @@ int jfs_sync(void *arg)
        int rc;
        tid_t tid;
 
-       daemonize("jfsSync");
-
-       complete(&jfsIOwait);
-
        do {
                /*
                 * write each inode on the anonymous inode list
@@ -2947,7 +2942,7 @@ int jfs_sync(void *arg)
                                 * Inode is being freed
                                 */
                                list_del_init(&jfs_ip->anon_inode_list);
-                       } else if (! down_trylock(&jfs_ip->commit_sem)) {
+                       } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
                                /*
                                 * inode will be removed from anonymous list
                                 * when it is committed
@@ -2956,7 +2951,7 @@ int jfs_sync(void *arg)
                                tid = txBegin(ip->i_sb, COMMIT_INODE);
                                rc = txCommit(tid, 1, &ip, 0);
                                txEnd(tid);
-                               up(&jfs_ip->commit_sem);
+                               mutex_unlock(&jfs_ip->commit_mutex);
 
                                iput(ip);
                                /*
@@ -2966,7 +2961,7 @@ int jfs_sync(void *arg)
                                cond_resched();
                                TXN_LOCK();
                        } else {
-                               /* We can't get the commit semaphore.  It may
+                               /* We can't get the commit mutex.  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.
@@ -2987,23 +2982,19 @@ int jfs_sync(void *arg)
                /* Add anon_list2 back to anon_list */
                list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
 
-               if (current->flags & PF_FREEZE) {
+               if (freezing(current)) {
                        TXN_UNLOCK();
-                       refrigerator(PF_FREEZE);
+                       refrigerator();
                } else {
-                       DECLARE_WAITQUEUE(wq, current);
-
-                       add_wait_queue(&jfs_sync_thread_wait, &wq);
                        set_current_state(TASK_INTERRUPTIBLE);
                        TXN_UNLOCK();
                        schedule();
                        current->state = TASK_RUNNING;
-                       remove_wait_queue(&jfs_sync_thread_wait, &wq);
                }
-       } while (!jfs_stop_threads);
+       } while (!kthread_should_stop());
 
        jfs_info("jfs_sync being killed");
-       complete_and_exit(&jfsIOwait, 0);
+       return 0;
 }
 
 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)