Re: [PATCH 10/13] xfs: add CRC checks to the AGFL

[Christoph Hellwig <hch@infradead.org>]

On Wed, Feb 11, 2009 at 06:54:33PM +1100, Dave Chinner wrote:
> I haven't had a chance to look over any of this series in detail - just
> a quick glance really - but this popped out as looking wrong:
> 
> > +/*
> > + * Size of the AGFL.  For CRC-enabled filesystes we steal the last two slots
> > + * for location information (agno) and the crc.
> > + */
> > +#define XFS_AGFL_SIZE(mp) \
> > +	((mp)->m_sb.sb_sectsize / sizeof(xfs_agblock_t) - \
> > +	 (xfs_sb_version_hascrc(&((mp)->m_sb)) ? sizeof(struct xfs_agfl) : 0))
> 
> sb_sectsize is in bytes, sizeof(xfs_agblock_t) is in bytes, which
> means that the numerator is a count of the number of agblocks that
> will fit in the AGFL. sizeof(struct xfs_agfl) is also in bytes,
> so your subtracting a number of bytes from a count of agblocks....
> 
> I think you mean:
> 
> #define XFS_AGFL_SIZE(mp) \
> 	(((mp)->m_sb.sb_sectsize - (xfs_sb_version_hascrc(&((mp)->m_sb)) ? \
> 						sizeof(struct xfs_agfl) : 0)) \
> 	  / sizeof(xfs_agblock_t))

Yeah.  Either that, or dividing the sizeof by four.

> >  typedef struct xfs_agfl {
> > -	__be32		agfl_bno[1];	/* actually XFS_AGFL_SIZE(mp) */
> > +	__be32		agfl_magicnum;
> > +	__be32		agfl_seqno;
> > +	uuid_t		agfl_uuid;
> > +	__be32		agfl_crc;
> > +	__be32		agfl_bno[];	/* actually XFS_AGFL_SIZE(mp) */
> >  } xfs_agfl_t;
> 
> And judging by that you are stealing more than 2 slots - more like
> 8 slots.

Yeah, two slots was the very first implementation, I'll fix up the
comment too.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs