jfs_txnmgr.c

jfs-2.4-1.1.7.tar.gz jfs 2.4－1.1.7 源码

			jfs_info("log barrier off: 0x%x", log->lsn);

			/* enable new transactions start */
			clear_bit(log_SYNCBARRIER, &log->flag);

			/* wakeup all waitors for logsync barrier */
			TXN_WAKEUP(&log->syncwait);
		}
	}

	/*
	 * wakeup all waitors for a free tblock
	 */
	TXN_WAKEUP(&TxAnchor.freewait);

	TXN_UNLOCK();
}


/*
 *      txLock()
 *
 * function: acquire a transaction lock on the specified <mp>
 *
 * parameter:
 *
 * return:      transaction lock id
 *
 * serialization:
 */
struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
		     int type)
{
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
	int dir_xtree = 0;
	lid_t lid;
	tid_t xtid;
	struct tlock *tlck;
	struct xtlock *xtlck;
	struct linelock *linelock;
	xtpage_t *p;
	struct tblock *tblk;

	assert(!test_cflag(COMMIT_Nolink, ip));

	TXN_LOCK();

	if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
	    !(mp->xflag & COMMIT_PAGE)) {
		/*
		 * Directory inode is special.  It can have both an xtree tlock
		 * and a dtree tlock associated with it.
		 */
		dir_xtree = 1;
		lid = jfs_ip->xtlid;
	} else
		lid = mp->lid;

	/* is page not locked by a transaction ? */
	if (lid == 0)
		goto allocateLock;

	jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);

	/* is page locked by the requester transaction ? */
	tlck = lid_to_tlock(lid);
	if ((xtid = tlck->tid) == tid)
		goto grantLock;

	/*
	 * is page locked by anonymous transaction/lock ?
	 *
	 * (page update without transaction (i.e., file write) is
	 * locked under anonymous transaction tid = 0:
	 * anonymous tlocks maintained on anonymous tlock list of
	 * the inode of the page and available to all anonymous
	 * transactions until txCommit() time at which point
	 * they are transferred to the transaction tlock list of
	 * the commiting transaction of the inode)
	 */
	if (xtid == 0) {
		tlck->tid = tid;
		tblk = tid_to_tblock(tid);
		/*
		 * The order of the tlocks in the transaction is important
		 * (during truncate, child xtree pages must be freed before
		 * parent's tlocks change the working map).
		 * Take tlock off anonymous list and add to tail of
		 * transaction list
		 *
		 * Note:  We really need to get rid of the tid & lid and
		 * use list_head's.  This code is getting UGLY!
		 */
		if (jfs_ip->atlhead == lid) {
			if (jfs_ip->atltail == lid) {
				/* only anonymous txn.
				 * Remove from anon_list
				 */
				list_del_init(&jfs_ip->anon_inode_list);
			}
			jfs_ip->atlhead = tlck->next;
		} else {
			lid_t last;
			for (last = jfs_ip->atlhead;
			     lid_to_tlock(last)->next != lid;
			     last = lid_to_tlock(last)->next) {
				assert(last);
			}
			lid_to_tlock(last)->next = tlck->next;
			if (jfs_ip->atltail == lid)
				jfs_ip->atltail = last;
		}

		/* insert the tlock at tail of transaction tlock list */

		if (tblk->next)
			lid_to_tlock(tblk->last)->next = lid;
		else
			tblk->next = lid;
		tlck->next = 0;
		tblk->last = lid;

		goto grantLock;
	}

	goto waitLock;

	/*
	 * allocate a tlock
	 */
      allocateLock:
	lid = txLockAlloc();
	tlck = lid_to_tlock(lid);

	/*
	 * initialize tlock
	 */
	tlck->tid = tid;

	/* 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);

		jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
			 mp, atomic_read(&mp->nohomeok), tid, tlck);

		/* if anonymous transaction, and buffer is on the group
		 * commit synclist, mark inode to show this.  This will
		 * prevent the buffer from being marked nohomeok for too
		 * long a time.
		 */
		if ((tid == 0) && mp->lsn)
			set_cflag(COMMIT_Synclist, ip);
	}
	/* mark tlock for in-memory inode */
	else
		tlck->flag = tlckINODELOCK;

	tlck->type = 0;

	/* bind the tlock and the page */
	tlck->ip = ip;
	tlck->mp = mp;
	if (dir_xtree)
		jfs_ip->xtlid = lid;
	else
		mp->lid = lid;

	/*
	 * enqueue transaction lock to transaction/inode
	 */
	/* insert the tlock at tail of transaction tlock list */
	if (tid) {
		tblk = tid_to_tblock(tid);
		if (tblk->next)
			lid_to_tlock(tblk->last)->next = lid;
		else
			tblk->next = lid;
		tlck->next = 0;
		tblk->last = lid;
	}
	/* anonymous transaction:
	 * insert the tlock at head of inode anonymous tlock list
	 */
	else {
		tlck->next = jfs_ip->atlhead;
		jfs_ip->atlhead = lid;
		if (tlck->next == 0) {
			/* This inode's first anonymous transaction */
			jfs_ip->atltail = lid;
			list_add_tail(&jfs_ip->anon_inode_list,
				      &TxAnchor.anon_list);
		}
	}

	/* initialize type dependent area for linelock */
	linelock = (struct linelock *) & tlck->lock;
	linelock->next = 0;
	linelock->flag = tlckLINELOCK;
	linelock->maxcnt = TLOCKSHORT;
	linelock->index = 0;

	switch (type & tlckTYPE) {
	case tlckDTREE:
		linelock->l2linesize = L2DTSLOTSIZE;
		break;

	case tlckXTREE:
		linelock->l2linesize = L2XTSLOTSIZE;

		xtlck = (struct xtlock *) linelock;
		xtlck->header.offset = 0;
		xtlck->header.length = 2;

		if (type & tlckNEW) {
			xtlck->lwm.offset = XTENTRYSTART;
		} else {
			if (mp->xflag & COMMIT_PAGE)
				p = (xtpage_t *) mp->data;
			else
				p = &jfs_ip->i_xtroot;
			xtlck->lwm.offset =
			    le16_to_cpu(p->header.nextindex);
		}
		xtlck->lwm.length = 0;	/* ! */
		xtlck->twm.offset = 0;
		xtlck->hwm.offset = 0;

		xtlck->index = 2;
		break;

	case tlckINODE:
		linelock->l2linesize = L2INODESLOTSIZE;
		break;

	case tlckDATA:
		linelock->l2linesize = L2DATASLOTSIZE;
		break;

	default:
		jfs_err("UFO tlock:0x%p", tlck);
	}

	/*
	 * update tlock vector
	 */
      grantLock:
	tlck->type |= type;

	TXN_UNLOCK();

	return tlck;

	/*
	 * page is being locked by another transaction:
	 */
      waitLock:
	/* Only locks on ipimap or ipaimap should reach here */
	/* assert(jfs_ip->fileset == AGGREGATE_I); */
	if (jfs_ip->fileset != AGGREGATE_I) {
		jfs_err("txLock: trying to lock locked page!");
		dump_mem("ip", ip, sizeof(struct inode));
		dump_mem("mp", mp, sizeof(struct metapage));
		dump_mem("Locker's tblk", tid_to_tblock(tid),
			 sizeof(struct tblock));
		dump_mem("Tlock", tlck, sizeof(struct tlock));
		BUG();
	}
	INCREMENT(stattx.waitlock);	/* statistics */
	release_metapage(mp);

	jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
		 tid, xtid, lid);
	TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
	jfs_info("txLock: awakened     tid = %d, lid = %d", tid, lid);

	return NULL;
}


/*
 * NAME:        txRelease()
 *
 * FUNCTION:    Release buffers associated with transaction locks, but don't
 *		mark homeok yet.  The allows other transactions to modify
 *		buffers, but won't let them go to disk until commit record
 *		actually gets written.
 *
 * PARAMETER:
 *              tblk    -
 *
 * RETURN:      Errors from subroutines.
 */
static void txRelease(struct tblock * tblk)
{
	struct metapage *mp;
	lid_t lid;
	struct tlock *tlck;

	TXN_LOCK();

	for (lid = tblk->next; lid; lid = tlck->next) {
		tlck = lid_to_tlock(lid);
		if ((mp = tlck->mp) != NULL &&
		    (tlck->type & tlckBTROOT) == 0) {
			assert(mp->xflag & COMMIT_PAGE);
			mp->lid = 0;
		}
	}

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

	TXN_UNLOCK();
}


/*
 * NAME:        txUnlock()
 *
 * FUNCTION:    Initiates pageout of pages modified by tid in journalled
 *              objects and frees their lockwords.
 */
static void txUnlock(struct tblock * tblk)
{
	struct tlock *tlck;
	struct linelock *linelock;
	lid_t lid, next, llid, k;
	struct metapage *mp;
	struct jfs_log *log;
	int difft, diffp;

	jfs_info("txUnlock: tblk = 0x%p", tblk);
	log = JFS_SBI(tblk->sb)->log;

	/*
	 * mark page under tlock homeok (its log has been written):
	 */
	for (lid = tblk->next; lid; lid = next) {
		tlck = lid_to_tlock(lid);
		next = tlck->next;

		jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);

		/* unbind page from tlock */
		if ((mp = tlck->mp) != NULL &&
		    (tlck->type & tlckBTROOT) == 0) {
			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);

			assert(atomic_read(&mp->nohomeok) > 0);
			atomic_dec(&mp->nohomeok);

			/* inherit younger/larger clsn */
			LOGSYNC_LOCK(log);
			if (mp->clsn) {
				logdiff(difft, tblk->clsn, log);
				logdiff(diffp, mp->clsn, log);
				if (difft > diffp)
					mp->clsn = tblk->clsn;
			} else
				mp->clsn = tblk->clsn;
			LOGSYNC_UNLOCK(log);

			assert(!(tlck->flag & tlckFREEPAGE));

			if (tlck->flag & tlckWRITEPAGE) {
				write_metapage(mp);
			} else {
				/* release page which has been forced */
				release_metapage(mp);
			}
		}

		/* insert tlock, and linelock(s) of the tlock if any,
		 * at head of freelist
		 */
		TXN_LOCK();

		llid = ((struct linelock *) & tlck->lock)->next;
		while (llid) {
			linelock = (struct linelock *) lid_to_tlock(llid);
			k = linelock->next;
			txLockFree(llid);
			llid = k;
		}
		txLockFree(lid);

		TXN_UNLOCK();
	}
	tblk->next = tblk->last = 0;

	/*
	 * remove tblock from logsynclist
	 * (allocation map pages inherited lsn of tblk and
	 * has been inserted in logsync list at txUpdateMap())
	 */
	if (tblk->lsn) {
		LOGSYNC_LOCK(log);
		log->count--;
		list_del(&tblk->synclist);
		LOGSYNC_UNLOCK(log);
	}
}


/*
 *      txMaplock()
 *
 * function: allocate a transaction lock for freed page/entry;
 *      for freed page, maplock is used as xtlock/dtlock type;
 */
struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
{
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
	lid_t lid;
	struct tblock *tblk;
	struct tlock *tlck;
	struct maplock *maplock;

	TXN_LOCK();

	/*
	 * allocate a tlock
	 */
	lid = txLockAlloc();
	tlck = lid_to_tlock(lid);

	/*
	 * initialize tlock
	 */
	tlck->tid = tid;

	/* bind the tlock and the object */
	tlck->flag = tlckINODELOCK;
	tlck->ip = ip;
	tlck->mp = NULL;

	tlck->type = type;

	/*
	 * enqueue transaction lock to transaction/inode
	 */
	/* insert the tlock at tail of transaction tlock list */
	if (tid) {
		tblk = tid_to_tblock(tid);
		if (tblk->next)
			lid_to_tlock(tblk->last)->next = lid;
		else
			tblk->next = lid;
		tlck->next = 0;
		tblk->last = lid;
	}
	/* anonymous transaction:
	 * insert the tlock at head of inode anonymous tlock list
	 */
	else {
		tlck->next = jfs_ip->atlhead;
		jfs_ip->atlhead = lid;
		if (tlck->next == 0) {
			/* This inode's first anonymous transaction */
			jfs_ip->atltail = lid;
			list_add_tail(&jfs_ip->anon_inode_list,
				      &TxAnchor.anon_list);
		}
	}

	TXN_UNLOCK();

	/* initialize type dependent area for maplock */
	maplock = (struct maplock *) & tlck->lock;
	maplock->next = 0;
	maplock->maxcnt = 0;
	maplock->index = 0;

	return tlck;
}


/*
 *      txLinelock()
 *
 * function: allocate a transaction lock for log vector list
 */
struct linelock *txLinelock(struct linelock * tlock)
{
	lid_t lid;
	struct tlock *tlck;
	struct linelock *linelock;

	TXN_LOCK();

	/* allocate a TxLock structure */
	lid = txLockAlloc();
	tlck = lid_to_tlock(lid);

	TXN_UNLOCK();

	/* initialize linelock */
	linelock = (struct linelock *) tlck;
	linelock->next = 0;
	linelock->flag = tlckLINELOCK;
	linelock->maxcnt = TLOCKLONG;
	linelock->index = 0;

	/* append linelock after tlock */
	linelock->next = tlock->next;
	tlock->next = lid;