On Wed, Feb 28, 2024 at 02:08:51PM -0800, Christoph Hellwig wrote:
> On Wed, Feb 28, 2024 at 11:55:32AM -0800, Darrick J. Wong wrote:
> > static inline bool
> > xmi_can_merge_all(
> > const struct xfs_bmbt_irec *l,
> > const struct xfs_bmbt_irec *m,
> > const struct xfs_bmbt_irec *r)
> > {
> > xfs_filblks_t new_len;
> >
> > new_len = l->br_blockcount + m->br_blockcount + r->br_blockcount;
> > return new_len <= XFS_MAX_BMBT_EXTLEN;
> > }
>
> Dumb question: can the addition overflow?
No.
Both callsites of xmi_can_merge_all trigger only if both LEFT_CONTIG and
RIGHT_CONTIG have been set. Both of thse _CONTIG flags are set only if
xmi_can_merge returned true, which it only does for real mappings. Real
mappings are derived from ondisk bmbt mappings, which means they won't
be larger than 2^21 blocks in length.
Therefore, [lmr]->br_blockcount each can only be up to 2^21, and adding
them all together only requires 23 bits. The u64 here is overkill, but
it matches xfs_bmbt_irec.br_blockcount.
--D
