On Fri, Apr 15, 2022 at 04:57:58PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@xxxxxxxxxx>
>
> If rounding the log start up to the next stripe unit would cause the log
> to overrun the end of the AG, round the log size down by a stripe unit.
> We already ensured that logblocks was small enough to fit inside the AG,
> so the minor adjustment should suffice.
Self NAK, Eric pointed out more ways this can fail. New patch coming
soon.
--D
> Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx>
> ---
> mkfs/xfs_mkfs.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/mkfs/xfs_mkfs.c b/mkfs/xfs_mkfs.c
> index b932acaa..cfa38f17 100644
> --- a/mkfs/xfs_mkfs.c
> +++ b/mkfs/xfs_mkfs.c
> @@ -3219,9 +3219,19 @@ align_internal_log(
> int max_logblocks)
> {
> /* round up log start if necessary */
> - if ((cfg->logstart % sunit) != 0)
> + if ((cfg->logstart % sunit) != 0) {
> cfg->logstart = ((cfg->logstart + (sunit - 1)) / sunit) * sunit;
>
> + /*
> + * If rounding up logstart to a stripe boundary moves the end
> + * of the log past the end of the AG, reduce logblocks to get
> + * it back under EOAG.
> + */
> + if (!libxfs_verify_fsbext(mp, cfg->logstart, cfg->logblocks) &&
> + cfg->logblocks > sunit)
> + cfg->logblocks -= sunit;
> + }
> +
> /* If our log start overlaps the next AG's metadata, fail. */
> if (!libxfs_verify_fsbno(mp, cfg->logstart)) {
> fprintf(stderr,
