xfs_log_recover.c

linux 内核源代码

				"bad inode buffer log record (ptr = 0x%p, bp = 0x%p).  XFS trying to replay bad (0) inode di_next_unlinked field",
				item, bp);
			XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
					 XFS_ERRLEVEL_LOW, mp);
			return XFS_ERROR(EFSCORRUPTED);
		}

		buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
					      next_unlinked_offset);
		*buffer_nextp = *logged_nextp;
	}

	return 0;
}

/*
 * Perform a 'normal' buffer recovery.  Each logged region of the
 * buffer should be copied over the corresponding region in the
 * given buffer.  The bitmap in the buf log format structure indicates
 * where to place the logged data.
 */
/*ARGSUSED*/
STATIC void
xlog_recover_do_reg_buffer(
	xlog_recover_item_t	*item,
	xfs_buf_t		*bp,
	xfs_buf_log_format_t	*buf_f)
{
	int			i;
	int			bit;
	int			nbits;
	unsigned int		*data_map = NULL;
	unsigned int		map_size = 0;
	int                     error;

	switch (buf_f->blf_type) {
	case XFS_LI_BUF:
		data_map = buf_f->blf_data_map;
		map_size = buf_f->blf_map_size;
		break;
	}
	bit = 0;
	i = 1;  /* 0 is the buf format structure */
	while (1) {
		bit = xfs_next_bit(data_map, map_size, bit);
		if (bit == -1)
			break;
		nbits = xfs_contig_bits(data_map, map_size, bit);
		ASSERT(nbits > 0);
		ASSERT(item->ri_buf[i].i_addr != NULL);
		ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
		ASSERT(XFS_BUF_COUNT(bp) >=
		       ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));

		/*
		 * Do a sanity check if this is a dquot buffer. Just checking
		 * the first dquot in the buffer should do. XXXThis is
		 * probably a good thing to do for other buf types also.
		 */
		error = 0;
		if (buf_f->blf_flags &
		   (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
			error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
					       item->ri_buf[i].i_addr,
					       -1, 0, XFS_QMOPT_DOWARN,
					       "dquot_buf_recover");
		}
		if (!error)
			memcpy(xfs_buf_offset(bp,
				(uint)bit << XFS_BLI_SHIFT),	/* dest */
				item->ri_buf[i].i_addr,		/* source */
				nbits<<XFS_BLI_SHIFT);		/* length */
		i++;
		bit += nbits;
	}

	/* Shouldn't be any more regions */
	ASSERT(i == item->ri_total);
}

/*
 * Do some primitive error checking on ondisk dquot data structures.
 */
int
xfs_qm_dqcheck(
	xfs_disk_dquot_t *ddq,
	xfs_dqid_t	 id,
	uint		 type,	  /* used only when IO_dorepair is true */
	uint		 flags,
	char		 *str)
{
	xfs_dqblk_t	 *d = (xfs_dqblk_t *)ddq;
	int		errs = 0;

	/*
	 * We can encounter an uninitialized dquot buffer for 2 reasons:
	 * 1. If we crash while deleting the quotainode(s), and those blks got
	 *    used for user data. This is because we take the path of regular
	 *    file deletion; however, the size field of quotainodes is never
	 *    updated, so all the tricks that we play in itruncate_finish
	 *    don't quite matter.
	 *
	 * 2. We don't play the quota buffers when there's a quotaoff logitem.
	 *    But the allocation will be replayed so we'll end up with an
	 *    uninitialized quota block.
	 *
	 * This is all fine; things are still consistent, and we haven't lost
	 * any quota information. Just don't complain about bad dquot blks.
	 */
	if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
		if (flags & XFS_QMOPT_DOWARN)
			cmn_err(CE_ALERT,
			"%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
			str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
		errs++;
	}
	if (ddq->d_version != XFS_DQUOT_VERSION) {
		if (flags & XFS_QMOPT_DOWARN)
			cmn_err(CE_ALERT,
			"%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
			str, id, ddq->d_version, XFS_DQUOT_VERSION);
		errs++;
	}

	if (ddq->d_flags != XFS_DQ_USER &&
	    ddq->d_flags != XFS_DQ_PROJ &&
	    ddq->d_flags != XFS_DQ_GROUP) {
		if (flags & XFS_QMOPT_DOWARN)
			cmn_err(CE_ALERT,
			"%s : XFS dquot ID 0x%x, unknown flags 0x%x",
			str, id, ddq->d_flags);
		errs++;
	}

	if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
		if (flags & XFS_QMOPT_DOWARN)
			cmn_err(CE_ALERT,
			"%s : ondisk-dquot 0x%p, ID mismatch: "
			"0x%x expected, found id 0x%x",
			str, ddq, id, be32_to_cpu(ddq->d_id));
		errs++;
	}

	if (!errs && ddq->d_id) {
		if (ddq->d_blk_softlimit &&
		    be64_to_cpu(ddq->d_bcount) >=
				be64_to_cpu(ddq->d_blk_softlimit)) {
			if (!ddq->d_btimer) {
				if (flags & XFS_QMOPT_DOWARN)
					cmn_err(CE_ALERT,
					"%s : Dquot ID 0x%x (0x%p) "
					"BLK TIMER NOT STARTED",
					str, (int)be32_to_cpu(ddq->d_id), ddq);
				errs++;
			}
		}
		if (ddq->d_ino_softlimit &&
		    be64_to_cpu(ddq->d_icount) >=
				be64_to_cpu(ddq->d_ino_softlimit)) {
			if (!ddq->d_itimer) {
				if (flags & XFS_QMOPT_DOWARN)
					cmn_err(CE_ALERT,
					"%s : Dquot ID 0x%x (0x%p) "
					"INODE TIMER NOT STARTED",
					str, (int)be32_to_cpu(ddq->d_id), ddq);
				errs++;
			}
		}
		if (ddq->d_rtb_softlimit &&
		    be64_to_cpu(ddq->d_rtbcount) >=
				be64_to_cpu(ddq->d_rtb_softlimit)) {
			if (!ddq->d_rtbtimer) {
				if (flags & XFS_QMOPT_DOWARN)
					cmn_err(CE_ALERT,
					"%s : Dquot ID 0x%x (0x%p) "
					"RTBLK TIMER NOT STARTED",
					str, (int)be32_to_cpu(ddq->d_id), ddq);
				errs++;
			}
		}
	}

	if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
		return errs;

	if (flags & XFS_QMOPT_DOWARN)
		cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);

	/*
	 * Typically, a repair is only requested by quotacheck.
	 */
	ASSERT(id != -1);
	ASSERT(flags & XFS_QMOPT_DQREPAIR);
	memset(d, 0, sizeof(xfs_dqblk_t));

	d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
	d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
	d->dd_diskdq.d_flags = type;
	d->dd_diskdq.d_id = cpu_to_be32(id);

	return errs;
}

/*
 * Perform a dquot buffer recovery.
 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
 * Else, treat it as a regular buffer and do recovery.
 */
STATIC void
xlog_recover_do_dquot_buffer(
	xfs_mount_t		*mp,
	xlog_t			*log,
	xlog_recover_item_t	*item,
	xfs_buf_t		*bp,
	xfs_buf_log_format_t	*buf_f)
{
	uint			type;

	/*
	 * Filesystems are required to send in quota flags at mount time.
	 */
	if (mp->m_qflags == 0) {
		return;
	}

	type = 0;
	if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
		type |= XFS_DQ_USER;
	if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
		type |= XFS_DQ_PROJ;
	if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
		type |= XFS_DQ_GROUP;
	/*
	 * This type of quotas was turned off, so ignore this buffer
	 */
	if (log->l_quotaoffs_flag & type)
		return;

	xlog_recover_do_reg_buffer(item, bp, buf_f);
}

/*
 * This routine replays a modification made to a buffer at runtime.
 * There are actually two types of buffer, regular and inode, which
 * are handled differently.  Inode buffers are handled differently
 * in that we only recover a specific set of data from them, namely
 * the inode di_next_unlinked fields.  This is because all other inode
 * data is actually logged via inode records and any data we replay
 * here which overlaps that may be stale.
 *
 * When meta-data buffers are freed at run time we log a buffer item
 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
 * of the buffer in the log should not be replayed at recovery time.
 * This is so that if the blocks covered by the buffer are reused for
 * file data before we crash we don't end up replaying old, freed
 * meta-data into a user's file.
 *
 * To handle the cancellation of buffer log items, we make two passes
 * over the log during recovery.  During the first we build a table of
 * those buffers which have been cancelled, and during the second we
 * only replay those buffers which do not have corresponding cancel
 * records in the table.  See xlog_recover_do_buffer_pass[1,2] above
 * for more details on the implementation of the table of cancel records.
 */
STATIC int
xlog_recover_do_buffer_trans(
	xlog_t			*log,
	xlog_recover_item_t	*item,
	int			pass)
{
	xfs_buf_log_format_t	*buf_f;
	xfs_mount_t		*mp;
	xfs_buf_t		*bp;
	int			error;
	int			cancel;
	xfs_daddr_t		blkno;
	int			len;
	ushort			flags;

	buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;

	if (pass == XLOG_RECOVER_PASS1) {
		/*
		 * In this pass we're only looking for buf items
		 * with the XFS_BLI_CANCEL bit set.
		 */
		xlog_recover_do_buffer_pass1(log, buf_f);
		return 0;
	} else {
		/*
		 * In this pass we want to recover all the buffers
		 * which have not been cancelled and are not
		 * cancellation buffers themselves.  The routine
		 * we call here will tell us whether or not to
		 * continue with the replay of this buffer.
		 */
		cancel = xlog_recover_do_buffer_pass2(log, buf_f);
		if (cancel) {
			return 0;
		}
	}
	switch (buf_f->blf_type) {
	case XFS_LI_BUF:
		blkno = buf_f->blf_blkno;
		len = buf_f->blf_len;
		flags = buf_f->blf_flags;
		break;
	default:
		xfs_fs_cmn_err(CE_ALERT, log->l_mp,
			"xfs_log_recover: unknown buffer type 0x%x, logdev %s",
			buf_f->blf_type, log->l_mp->m_logname ?
			log->l_mp->m_logname : "internal");
		XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
				 XFS_ERRLEVEL_LOW, log->l_mp);
		return XFS_ERROR(EFSCORRUPTED);
	}

	mp = log->l_mp;
	if (flags & XFS_BLI_INODE_BUF) {
		bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
								XFS_BUF_LOCK);
	} else {
		bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
	}
	if (XFS_BUF_ISERROR(bp)) {
		xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
				  bp, blkno);
		error = XFS_BUF_GETERROR(bp);
		xfs_buf_relse(bp);
		return error;
	}

	error = 0;
	if (flags & XFS_BLI_INODE_BUF) {
		error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
	} else if (flags &
		  (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
		xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
	} else {
		xlog_recover_do_reg_buffer(item, bp, buf_f);
	}
	if (error)
		return XFS_ERROR(error);

	/*
	 * Perform delayed write on the buffer.  Asynchronous writes will be
	 * slower when taking into account all the buffers to be flushed.
	 *
	 * Also make sure that only inode buffers with good sizes stay in
	 * the buffer cache.  The kernel moves inodes in buffers of 1 block
	 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger.  The inode
	 * buffers in the log can be a different size if the log was generated
	 * by an older kernel using unclustered inode buffers or a newer kernel
	 * running with a different inode cluster size.  Regardless, if the
	 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
	 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
	 * the buffer out of the buffer cache so that the buffer won't
	 * overlap with future reads of those inodes.
	 */
	if (XFS_DINODE_MAGIC ==
	    INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) &&
	    (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
			(__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
		XFS_BUF_STALE(bp);
		error = xfs_bwrite(mp, bp);
	} else {
		ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
		       XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
		XFS_BUF_SET_FSPRIVATE(bp, mp);
		XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
		xfs_bdwrite(mp, bp);
	}

	return (error);
}

STATIC int
xlog_recover_do_inode_trans(
	xlog_t			*log,
	xlog_recover_item_t	*item,
	int			pass)
{
	xfs_inode_log_format_t	*in_f;
	xfs_mount_t		*mp;
	xfs_buf_t		*bp;
	xfs_imap_t		imap;
	xfs_dinode_t		*dip;
	xfs_ino_t		ino;
	int			len;
	xfs_caddr_t		src;
	xfs_caddr_t		dest;
	int			error;
	int			attr_index;
	uint			fields;
	xfs_icdinode_t		*dicp;
	int			need_free = 0;

	if (pass == XLOG_RECOVER_PASS1) {
		return 0;
	}

	if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
		in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
	} else {
		in_f = (xfs_inode_log_format_t *)kmem_alloc(
			sizeof(xfs_inode_log_format_t), KM_SLEEP);
		need_free = 1;
		error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
		if (error)
			goto error;
	}
	ino = in_f->ilf_ino;
	mp = log->l_mp;
	if (ITEM_TYPE(item) == XFS_LI_INODE) {
		imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
		imap.im_len = in_f->ilf_len;
		imap.im_boffset = in_f->ilf_boffset;
	} else {
		/*
		 * It's an old inode format record.  We don't know where
		 * its cluster is located on disk, and we can't allow
		 * xfs_imap() to figure it out because the inode btrees
		 * are not ready to be used.  Therefore do not pass the
		 * XFS_IMAP_LOOKUP flag to xfs_imap().  This will give
		 * us only the single block in which the inode lives
		 * rather than its cluster, so we must make sure to
		 * invalidate the buffer when we write it out below.
		 */
		imap.im_blkno = 0;
		xfs_imap(log->l_mp, NULL, ino, &imap, 0);
	}

	/*
	 * Inode buffers can be freed, look out for it,
	 * and do not replay the inode.
	 */
	if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
		error = 0;
		goto error;
	}

	bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
								XFS_BUF_LOCK);
	if (XFS_BUF_ISERROR(bp)) {
		xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
				  bp, imap.im_blkno);
		error = XFS_BUF_GETERROR(bp);
		xfs_buf_relse(bp);
		goto error;
	}
	error = 0;
	ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
	dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);

	/*
	 * Make sure the place we're flushing out to really looks
	 * like an inode!
	 */
	if (unlikely(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC)) {