On Tue, Mar 19, 2024 at 03:54:02PM -0700, Christoph Hellwig wrote:
> > +static bool
> > +xlog_recovery_is_dir_buf(
> > + struct xfs_buf_log_format *buf_f)
> > +{
> > + switch (xfs_blft_from_flags(buf_f)) {
> > + case XFS_BLFT_DIR_BLOCK_BUF:
> > + case XFS_BLFT_DIR_DATA_BUF:
> > + case XFS_BLFT_DIR_FREE_BUF:
> > + case XFS_BLFT_DIR_LEAF1_BUF:
> > + case XFS_BLFT_DIR_LEAFN_BUF:
> > + case XFS_BLFT_DA_NODE_BUF:
>
> XFS_BLFT_DA_NODE_BUF can also be a non-directory buffer. Maybe this
> should be named something like xlog_recover_maybe_is_partial_dabuf?
>
> > + error = bp->b_error;
> > goto out_release;
> > }
> >
> > +
>
> This adds a spurious new line.
>
> Otherwise this looks good to me, but the lack over verifiation for these
> multi-buffer recoveries really scares me..
I wondered if there was a good way to reconstruct the discontiguous
buffer, but the only things I could think of relied on guessing from the
subsequent buffer log items. Does anyone have a better idea?
--D
