jfs_txnmgr.c

linux 内核源代码

/*
 *   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
 *   (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_txnmgr.c: transaction manager
 *
 * notes:
 * transaction starts with txBegin() and ends with txCommit()
 * or txAbort().
 *
 * tlock is acquired at the time of update;
 * (obviate scan at commit time for xtree and dtree)
 * tlock and mp points to each other;
 * (no hashlist for mp -> tlock).
 *
 * special cases:
 * tlock on in-memory inode:
 * in-place tlock in the in-memory inode itself;
 * converted to page lock by iWrite() at commit time.
 *
 * tlock during write()/mmap() under anonymous transaction (tid = 0):
 * transferred (?) to transaction at commit time.
 *
 * use the page itself to update allocation maps
 * (obviate intermediate replication of allocation/deallocation data)
 * hold on to mp+lock thru update of maps
 */

#include <linux/fs.h>
#include <linux/vmalloc.h>
#include <linux/completion.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"
#include "jfs_imap.h"
#include "jfs_dmap.h"
#include "jfs_superblock.h"
#include "jfs_debug.h"

/*
 *	transaction management structures
 */
static struct {
	int freetid;		/* index of a free tid structure */
	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 */
	spinlock_t LazyLock;	/* synchronize sync_queue & unlock_queue */
/*	struct tblock *sync_queue; * Transactions waiting for data sync */
	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
static struct {
	uint txBegin;
	uint txBegin_barrier;
	uint txBegin_lockslow;
	uint txBegin_freetid;
	uint txBeginAnon;
	uint txBeginAnon_barrier;
	uint txBeginAnon_lockslow;
	uint txLockAlloc;
	uint txLockAlloc_freelock;
} TxStat;
#endif

static int nTxBlock = -1;	/* number of transaction blocks */
module_param(nTxBlock, int, 0);
MODULE_PARM_DESC(nTxBlock,
		 "Number of transaction blocks (max:65536)");

static int nTxLock = -1;	/* number of transaction locks */
module_param(nTxLock, int, 0);
MODULE_PARM_DESC(nTxLock,
		 "Number of transaction locks (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 */

/*
 *	transaction management lock
 */
static DEFINE_SPINLOCK(jfsTxnLock);

#define TXN_LOCK()		spin_lock(&jfsTxnLock)
#define TXN_UNLOCK()		spin_unlock(&jfsTxnLock)

#define LAZY_LOCK_INIT()	spin_lock_init(&TxAnchor.LazyLock);
#define LAZY_LOCK(flags)	spin_lock_irqsave(&TxAnchor.LazyLock, flags)
#define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)

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.
 */
static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
{
	DECLARE_WAITQUEUE(wait, current);

	add_wait_queue(event, &wait);
	set_current_state(TASK_UNINTERRUPTIBLE);
	TXN_UNLOCK();
	io_schedule();
	__set_current_state(TASK_RUNNING);
	remove_wait_queue(event, &wait);
}

#define TXN_SLEEP(event)\
{\
	TXN_SLEEP_DROP_LOCK(event);\
	TXN_LOCK();\
}

#define TXN_WAKEUP(event) wake_up_all(event)

/*
 *	statistics
 */
static struct {
	tid_t maxtid;		/* 4: biggest tid ever used */
	lid_t maxlid;		/* 4: biggest lid ever used */
	int ntid;		/* 4: # of transactions performed */
	int nlid;		/* 4: # of tlocks acquired */
	int waitlock;		/* 4: # of tlock wait */
} stattx;

/*
 * forward references
 */
static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
		struct tlock * tlck, struct commit * cd);
static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
		struct tlock * tlck);
static void dtLog(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);
static void txAllocPMap(struct inode *ip, struct maplock * maplock,
		struct tblock * tblk);
static void txForce(struct tblock * tblk);
static int txLog(struct jfs_log * log, struct tblock * tblk,
		struct commit * cd);
static void txUpdateMap(struct tblock * tblk);
static void txRelease(struct tblock * tblk);
static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
	   struct tlock * tlck);
static void LogSyncRelease(struct metapage * mp);

/*
 *		transaction block/lock management
 *		---------------------------------
 */

/*
 * Get a transaction lock from the free list.  If the number in use is
 * greater than the high water mark, wake up the sync daemon.  This should
 * free some anonymous transaction locks.  (TXN_LOCK must be held.)
 */
static lid_t txLockAlloc(void)
{
	lid_t lid;

	INCREMENT(TxStat.txLockAlloc);
	if (!TxAnchor.freelock) {
		INCREMENT(TxStat.txLockAlloc_freelock);
	}

	while (!(lid = TxAnchor.freelock))
		TXN_SLEEP(&TxAnchor.freelockwait);
	TxAnchor.freelock = TxLock[lid].next;
	HIGHWATERMARK(stattx.maxlid, lid);
	if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
		jfs_info("txLockAlloc tlocks low");
		jfs_tlocks_low = 1;
		wake_up_process(jfsSyncThread);
	}

	return lid;
}

static void txLockFree(lid_t lid)
{
	TxLock[lid].tid = 0;
	TxLock[lid].next = TxAnchor.freelock;
	TxAnchor.freelock = lid;
	TxAnchor.tlocksInUse--;
	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);
}

/*
 * NAME:	txInit()
 *
 * FUNCTION:	initialize transaction management structures
 *
 * RETURN:
 *
 * serialization: single thread at jfs_init()
 */
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)
		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;

	printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
	       nTxBlock, nTxLock);
	/*
	 * initialize transaction block (tblock) table
	 *
	 * transaction id (tid) = tblock index
	 * tid = 0 is reserved.
	 */
	TxLockLWM = (nTxLock * 4) / 10;
	TxLockHWM = (nTxLock * 7) / 10;
	TxLockVHWM = (nTxLock * 8) / 10;

	size = sizeof(struct tblock) * nTxBlock;
	TxBlock = vmalloc(size);
	if (TxBlock == NULL)
		return -ENOMEM;

	for (k = 1; k < nTxBlock - 1; k++) {
		TxBlock[k].next = k + 1;
		init_waitqueue_head(&TxBlock[k].gcwait);
		init_waitqueue_head(&TxBlock[k].waitor);
	}
	TxBlock[k].next = 0;
	init_waitqueue_head(&TxBlock[k].gcwait);
	init_waitqueue_head(&TxBlock[k].waitor);

	TxAnchor.freetid = 1;
	init_waitqueue_head(&TxAnchor.freewait);

	stattx.maxtid = 1;	/* statistics */

	/*
	 * initialize transaction lock (tlock) table
	 *
	 * transaction lock id = tlock index
	 * tlock id = 0 is reserved.
	 */
	size = sizeof(struct tlock) * nTxLock;
	TxLock = vmalloc(size);
	if (TxLock == NULL) {
		vfree(TxBlock);
		return -ENOMEM;
	}

	/* initialize tlock table */
	for (k = 1; k < nTxLock - 1; k++)
		TxLock[k].next = k + 1;
	TxLock[k].next = 0;
	init_waitqueue_head(&TxAnchor.freelockwait);
	init_waitqueue_head(&TxAnchor.lowlockwait);

	TxAnchor.freelock = 1;
	TxAnchor.tlocksInUse = 0;
	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;
}

/*
 * NAME:	txExit()
 *
 * FUNCTION:	clean up when module is unloaded
 */
void txExit(void)
{
	vfree(TxLock);
	TxLock = NULL;
	vfree(TxBlock);
	TxBlock = NULL;
}

/*
 * NAME:	txBegin()
 *
 * FUNCTION:	start a transaction.
 *
 * PARAMETER:	sb	- superblock
 *		flag	- force for nested tx;
 *
 * RETURN:	tid	- transaction id
 *
 * note: flag force allows to start tx for nested tx
 * to prevent deadlock on logsync barrier;
 */
tid_t txBegin(struct super_block *sb, int flag)
{
	tid_t t;
	struct tblock *tblk;
	struct jfs_log *log;

	jfs_info("txBegin: flag = 0x%x", flag);
	log = JFS_SBI(sb)->log;

	TXN_LOCK();

	INCREMENT(TxStat.txBegin);

      retry:
	if (!(flag & COMMIT_FORCE)) {
		/*
		 * synchronize with logsync barrier
		 */
		if (test_bit(log_SYNCBARRIER, &log->flag) ||
		    test_bit(log_QUIESCE, &log->flag)) {
			INCREMENT(TxStat.txBegin_barrier);
			TXN_SLEEP(&log->syncwait);
			goto retry;
		}
	}
	if (flag == 0) {
		/*
		 * Don't begin transaction if we're getting starved for tlocks
		 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
		 * free tlocks)
		 */
		if (TxAnchor.tlocksInUse > TxLockVHWM) {
			INCREMENT(TxStat.txBegin_lockslow);
			TXN_SLEEP(&TxAnchor.lowlockwait);
			goto retry;
		}
	}

	/*
	 * allocate transaction id/block
	 */
	if ((t = TxAnchor.freetid) == 0) {
		jfs_info("txBegin: waiting for free tid");
		INCREMENT(TxStat.txBegin_freetid);
		TXN_SLEEP(&TxAnchor.freewait);
		goto retry;
	}

	tblk = tid_to_tblock(t);

	if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
		/* Don't let a non-forced transaction take the last tblk */
		jfs_info("txBegin: waiting for free tid");
		INCREMENT(TxStat.txBegin_freetid);
		TXN_SLEEP(&TxAnchor.freewait);
		goto retry;
	}

	TxAnchor.freetid = tblk->next;

	/*
	 * initialize transaction
	 */

	/*
	 * We can't zero the whole thing or we screw up another thread being
	 * awakened after sleeping on tblk->waitor
	 *
	 * memset(tblk, 0, sizeof(struct tblock));
	 */
	tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;

	tblk->sb = sb;
	++log->logtid;
	tblk->logtid = log->logtid;

	++log->active;

	HIGHWATERMARK(stattx.maxtid, t);	/* statistics */
	INCREMENT(stattx.ntid);	/* statistics */

	TXN_UNLOCK();

	jfs_info("txBegin: returning tid = %d", t);

	return t;
}

/*
 * NAME:	txBeginAnon()
 *
 * FUNCTION:	start an anonymous transaction.
 *		Blocks if logsync or available tlocks are low to prevent
 *		anonymous tlocks from depleting supply.
 *
 * PARAMETER:	sb	- superblock
 *
 * RETURN:	none
 */
void txBeginAnon(struct super_block *sb)
{
	struct jfs_log *log;

	log = JFS_SBI(sb)->log;

	TXN_LOCK();
	INCREMENT(TxStat.txBeginAnon);

      retry:
	/*
	 * synchronize with logsync barrier
	 */
	if (test_bit(log_SYNCBARRIER, &log->flag) ||
	    test_bit(log_QUIESCE, &log->flag)) {
		INCREMENT(TxStat.txBeginAnon_barrier);
		TXN_SLEEP(&log->syncwait);
		goto retry;
	}

	/*
	 * Don't begin transaction if we're getting starved for tlocks
	 */
	if (TxAnchor.tlocksInUse > TxLockVHWM) {
		INCREMENT(TxStat.txBeginAnon_lockslow);
		TXN_SLEEP(&TxAnchor.lowlockwait);
		goto retry;
	}
	TXN_UNLOCK();
}

/*
 *	txEnd()
 *
 * function: free specified transaction block.
 *
 *	logsync barrier processing:
 *
 * serialization:
 */
void txEnd(tid_t tid)
{
	struct tblock *tblk = tid_to_tblock(tid);
	struct jfs_log *log;

	jfs_info("txEnd: tid = %d", tid);
	TXN_LOCK();

	/*
	 * wakeup transactions waiting on the page locked
	 * by the current transaction
	 */
	TXN_WAKEUP(&tblk->waitor);

	log = JFS_SBI(tblk->sb)->log;

	/*
	 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
	 * otherwise, we would be left with a transaction that may have been
	 * reused.
	 *
	 * Lazy commit thread will turn off tblkGC_LAZY before calling this