jfs_txnmgr.c

linux 内核源代码

		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;
	if (tlck->flag & tlckDIRECTORY)
		linelock->flag |= tlckDIRECTORY;

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

	return linelock;
}

/*
 *		transaction commit management
 *		-----------------------------
 */

/*
 * NAME:	txCommit()
 *
 * FUNCTION:	commit the changes to the objects specified in
 *		clist.  For journalled segments only the
 *		changes of the caller are committed, ie by tid.
 *		for non-journalled segments the data are flushed to
 *		disk and then the change to the disk inode and indirect
 *		blocks committed (so blocks newly allocated to the
 *		segment will be made a part of the segment atomically).
 *
 *		all of the segments specified in clist must be in
 *		one file system. no more than 6 segments are needed
 *		to handle all unix svcs.
 *
 *		if the i_nlink field (i.e. disk inode link count)
 *		is zero, and the type of inode is a regular file or
 *		directory, or symbolic link , the inode is truncated
 *		to zero length. the truncation is committed but the
 *		VM resources are unaffected until it is closed (see
 *		iput and iclose).
 *
 * PARAMETER:
 *
 * RETURN:
 *
 * serialization:
 *		on entry the inode lock on each segment is assumed
 *		to be held.
 *
 * i/o error:
 */
int txCommit(tid_t tid,		/* transaction identifier */
	     int nip,		/* number of inodes to commit */
	     struct inode **iplist,	/* list of inode to commit */
	     int flag)
{
	int rc = 0;
	struct commit cd;
	struct jfs_log *log;
	struct tblock *tblk;
	struct lrd *lrd;
	int lsn;
	struct inode *ip;
	struct jfs_inode_info *jfs_ip;
	int k, n;
	ino_t top;
	struct super_block *sb;

	jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
	/* is read-only file system ? */
	if (isReadOnly(iplist[0])) {
		rc = -EROFS;
		goto TheEnd;
	}

	sb = cd.sb = iplist[0]->i_sb;
	cd.tid = tid;

	if (tid == 0)
		tid = txBegin(sb, 0);
	tblk = tid_to_tblock(tid);

	/*
	 * initialize commit structure
	 */
	log = JFS_SBI(sb)->log;
	cd.log = log;

	/* initialize log record descriptor in commit */
	lrd = &cd.lrd;
	lrd->logtid = cpu_to_le32(tblk->logtid);
	lrd->backchain = 0;

	tblk->xflag |= flag;

	if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
		tblk->xflag |= COMMIT_LAZY;
	/*
	 *	prepare non-journaled objects for commit
	 *
	 * flush data pages of non-journaled file
	 * to prevent the file getting non-initialized disk blocks
	 * in case of crash.
	 * (new blocks - )
	 */
	cd.iplist = iplist;
	cd.nip = nip;

	/*
	 *	acquire transaction lock on (on-disk) inodes
	 *
	 * update on-disk inode from in-memory inode
	 * acquiring transaction locks for AFTER records
	 * on the on-disk inode of file object
	 *
	 * sort the inodes array by inode number in descending order
	 * to prevent deadlock when acquiring transaction lock
	 * of on-disk inodes on multiple on-disk inode pages by
	 * multiple concurrent transactions
	 */
	for (k = 0; k < cd.nip; k++) {
		top = (cd.iplist[k])->i_ino;
		for (n = k + 1; n < cd.nip; n++) {
			ip = cd.iplist[n];
			if (ip->i_ino > top) {
				top = ip->i_ino;
				cd.iplist[n] = cd.iplist[k];
				cd.iplist[k] = ip;
			}
		}

		ip = cd.iplist[k];
		jfs_ip = JFS_IP(ip);

		/*
		 * BUGBUG - This code has temporarily been removed.  The
		 * intent is to ensure that any file data is written before
		 * the metadata is committed to the journal.  This prevents
		 * uninitialized data from appearing in a file after the
		 * journal has been replayed.  (The uninitialized data
		 * could be sensitive data removed by another user.)
		 *
		 * The problem now is that we are holding the IWRITELOCK
		 * on the inode, and calling filemap_fdatawrite on an
		 * unmapped page will cause a deadlock in jfs_get_block.
		 *
		 * The long term solution is to pare down the use of
		 * IWRITELOCK.  We are currently holding it too long.
		 * We could also be smarter about which data pages need
		 * to be written before the transaction is committed and
		 * when we don't need to worry about it at all.
		 *
		 * if ((!S_ISDIR(ip->i_mode))
		 *    && (tblk->flag & COMMIT_DELETE) == 0)
		 *	filemap_write_and_wait(ip->i_mapping);
		 */

		/*
		 * Mark inode as not dirty.  It will still be on the dirty
		 * inode list, but we'll know not to commit it again unless
		 * it gets marked dirty again
		 */
		clear_cflag(COMMIT_Dirty, ip);

		/* inherit anonymous tlock(s) of inode */
		if (jfs_ip->atlhead) {
			lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
			tblk->next = jfs_ip->atlhead;
			if (!tblk->last)
				tblk->last = jfs_ip->atltail;
			jfs_ip->atlhead = jfs_ip->atltail = 0;
			TXN_LOCK();
			list_del_init(&jfs_ip->anon_inode_list);
			TXN_UNLOCK();
		}

		/*
		 * acquire transaction lock on on-disk inode page
		 * (become first tlock of the tblk's tlock list)
		 */
		if (((rc = diWrite(tid, ip))))
			goto out;
	}

	/*
	 *	write log records from transaction locks
	 *
	 * txUpdateMap() resets XAD_NEW in XAD.
	 */
	if ((rc = txLog(log, tblk, &cd)))
		goto TheEnd;

	/*
	 * Ensure that inode isn't reused before
	 * lazy commit thread finishes processing
	 */
	if (tblk->xflag & COMMIT_DELETE) {
		atomic_inc(&tblk->u.ip->i_count);
		/*
		 * Avoid a rare deadlock
		 *
		 * If the inode is locked, we may be blocked in
		 * jfs_commit_inode.  If so, we don't want the
		 * lazy_commit thread doing the last iput() on the inode
		 * since that may block on the locked inode.  Instead,
		 * commit the transaction synchronously, so the last iput
		 * will be done by the calling thread (or later)
		 */
		/*
		 * I believe this code is no longer needed.  Splitting I_LOCK
		 * into two bits, I_LOCK and I_SYNC should prevent this
		 * deadlock as well.  But since I don't have a JFS testload
		 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
		 * Joern
		 */
		if (tblk->u.ip->i_state & I_SYNC)
			tblk->xflag &= ~COMMIT_LAZY;
	}

	ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
	       ((tblk->u.ip->i_nlink == 0) &&
		!test_cflag(COMMIT_Nolink, tblk->u.ip)));

	/*
	 *	write COMMIT log record
	 */
	lrd->type = cpu_to_le16(LOG_COMMIT);
	lrd->length = 0;
	lsn = lmLog(log, tblk, lrd, NULL);

	lmGroupCommit(log, tblk);

	/*
	 *	- transaction is now committed -
	 */

	/*
	 * force pages in careful update
	 * (imap addressing structure update)
	 */
	if (flag & COMMIT_FORCE)
		txForce(tblk);

	/*
	 *	update allocation map.
	 *
	 * update inode allocation map and inode:
	 * free pager lock on memory object of inode if any.
	 * update block allocation map.
	 *
	 * txUpdateMap() resets XAD_NEW in XAD.
	 */
	if (tblk->xflag & COMMIT_FORCE)
		txUpdateMap(tblk);

	/*
	 *	free transaction locks and pageout/free pages
	 */
	txRelease(tblk);

	if ((tblk->flag & tblkGC_LAZY) == 0)
		txUnlock(tblk);


	/*
	 *	reset in-memory object state
	 */
	for (k = 0; k < cd.nip; k++) {
		ip = cd.iplist[k];
		jfs_ip = JFS_IP(ip);

		/*
		 * reset in-memory inode state
		 */
		jfs_ip->bxflag = 0;
		jfs_ip->blid = 0;
	}

      out:
	if (rc != 0)
		txAbort(tid, 1);

      TheEnd:
	jfs_info("txCommit: tid = %d, returning %d", tid, rc);
	return rc;
}

/*
 * NAME:	txLog()
 *
 * FUNCTION:	Writes AFTER log records for all lines modified
 *		by tid for segments specified by inodes in comdata.
 *		Code assumes only WRITELOCKS are recorded in lockwords.
 *
 * PARAMETERS:
 *
 * RETURN :
 */
static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
{
	int rc = 0;
	struct inode *ip;
	lid_t lid;
	struct tlock *tlck;
	struct lrd *lrd = &cd->lrd;

	/*
	 * write log record(s) for each tlock of transaction,
	 */
	for (lid = tblk->next; lid; lid = tlck->next) {
		tlck = lid_to_tlock(lid);

		tlck->flag |= tlckLOG;

		/* initialize lrd common */
		ip = tlck->ip;
		lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
		lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
		lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);

		/* write log record of page from the tlock */
		switch (tlck->type & tlckTYPE) {
		case tlckXTREE:
			xtLog(log, tblk, lrd, tlck);
			break;

		case tlckDTREE:
			dtLog(log, tblk, lrd, tlck);
			break;

		case tlckINODE:
			diLog(log, tblk, lrd, tlck, cd);
			break;

		case tlckMAP:
			mapLog(log, tblk, lrd, tlck);
			break;

		case tlckDATA:
			dataLog(log, tblk, lrd, tlck);
			break;

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

	return rc;
}

/*
 *	diLog()
 *
 * function:	log inode tlock and format maplock to update bmap;
 */
static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
		 struct tlock * tlck, struct commit * cd)
{
	int rc = 0;
	struct metapage *mp;
	pxd_t *pxd;
	struct pxd_lock *pxdlock;

	mp = tlck->mp;

	/* initialize as REDOPAGE record format */
	lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
	lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);

	pxd = &lrd->log.redopage.pxd;

	/*
	 *	inode after image
	 */
	if (tlck->type & tlckENTRY) {
		/* log after-image for logredo(): */
		lrd->type = cpu_to_le16(LOG_REDOPAGE);
		PXDaddress(pxd, mp->index);
		PXDlength(pxd,
			  mp->logical_size >> tblk->sb->s_blocksize_bits);
		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));

		/* mark page as homeward bound */
		tlck->flag |= tlckWRITEPAGE;
	} else if (tlck->type & tlckFREE) {
		/*
		 *	free inode extent
		 *
		 * (pages of the freed inode extent have been invalidated and
		 * a maplock for free of the extent has been formatted at
		 * txLock() time);
		 *
		 * the tlock had been acquired on the inode allocation map page
		 * (iag) that specifies the freed extent, even though the map
		 * page is not itself logged, to prevent pageout of the map
		 * page before the log;
		 */

		/* log LOG_NOREDOINOEXT of the freed inode extent for
		 * logredo() to start NoRedoPage filters, and to update
		 * imap and bmap for free of the extent;
		 */
		lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
		/*
		 * For the LOG_NOREDOINOEXT record, we need
		 * to pass the IAG number and inode extent
		 * index (within that IAG) from which the
		 * the extent being released.  These have been
		 * passed to us in the iplist[1] and iplist[2].
		 */
		lrd->log.noredoinoext.iagnum =
		    cpu_to_le32((u32) (size_t) cd->iplist[1]);
		lrd->log.noredoinoext.inoext_idx =
		    cpu_to_le32((u32) (size_t) cd->iplist[2]);

		pxdlock = (struct pxd_lock *) & tlck->lock;
		*pxd = pxdlock->pxd;
		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));

		/* update bmap */
		tlck->flag |= tlckUPDATEMAP;

		/* mark page as homeward bound */
		tlck->flag |= tlckWRITEPAGE;
	} else
		jfs_err("diLog: UFO type tlck:0x%p", tlck);
#ifdef  _JFS_WIP
	/*
	 *	alloc/free external EA extent
	 *
	 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
	 * of the extent has been formatted at txLock() time;
	 */
	else {
		assert(tlck->type & tlckEA);

		/* log LOG_UPDATEMAP for logredo() to update bmap for
		 * alloc of new (and free of old) external EA extent;
		 */
		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
		pxdlock = (struct pxd_lock *) & tlck->lock;
		nlock = pxdlock->index;
		for (i = 0; i < nlock; i++, pxdlock++) {
			if (pxdlock->flag & mlckALLOCPXD)
				lrd->log.updatemap.type =
				    cpu_to_le16(LOG_ALLOCPXD);
			else
				lrd->log.updatemap.type =
				    cpu_to_le16(LOG_FREEPXD);
			lrd->log.updatemap.nxd = cpu_to_le16(1);
			lrd->log.updatemap.pxd = pxdlock->pxd;
			lrd->backchain =
			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
		}

		/* update bmap */
		tlck->flag |= tlckUPDATEMAP;
	}
#endif				/* _JFS_WIP */

	return rc;
}

/*
 *	dataLog()
 *
 * function:	log data tlock
 */
static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
	    struct tlock * tlck)
{
	struct metapage *mp;
	pxd_t *pxd;

	mp = tlck->mp;

	/* initialize as REDOPAGE record format */
	lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
	lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);