On Tue, 2019-12-03 at 09:29 -0500, xiubli@xxxxxxxxxx wrote:
> From: Xiubo Li <xiubli@xxxxxxxxxx>
>
> The possible max rank, it maybe larger than the m->m_num_mds,
> for example if the mds_max == 2 in the cluster, when the MDS(0)
> was laggy and being replaced by a new MDS, we will temporarily
> receive a new mds map with n_num_mds == 1 and the active MDS(1),
> and the mds rank >= m->m_num_mds.
>
> Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
> ---
> fs/ceph/mdsmap.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
> index 284d68646c40..a77e0ecb9a6b 100644
> --- a/fs/ceph/mdsmap.c
> +++ b/fs/ceph/mdsmap.c
> @@ -129,6 +129,7 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
> int err;
> u8 mdsmap_v, mdsmap_cv;
> u16 mdsmap_ev;
> + u32 possible_max_rank;
>
> m = kzalloc(sizeof(*m), GFP_NOFS);
> if (!m)
> @@ -164,6 +165,15 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
> m->m_num_mds = n = ceph_decode_32(p);
> m->m_num_active_mds = m->m_num_mds;
>
> + /*
> + * the possible max rank, it maybe larger than the m->m_num_mds,
> + * for example if the mds_max == 2 in the cluster, when the MDS(0)
> + * was laggy and being replaced by a new MDS, we will temporarily
> + * receive a new mds map with n_num_mds == 1 and the active MDS(1),
> + * and the mds rank >= m->m_num_mds.
> + */
> + possible_max_rank = max((u32)m->m_num_mds, m->m_max_mds);
> +
> m->m_info = kcalloc(m->m_num_mds, sizeof(*m->m_info), GFP_NOFS);
> if (!m->m_info)
> goto nomem;
> @@ -238,7 +248,7 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
> ceph_mds_state_name(state),
> laggy ? "(laggy)" : "");
>
> - if (mds < 0 || mds >= m->m_num_mds) {
> + if (mds < 0 || mds >= possible_max_rank) {
> pr_warn("mdsmap_decode got incorrect mds(%d)\n", mds);
> continue;
> }
Thanks, Xiubo. I'll squash this one into your earlier ceph_mdsmap_decode
patch, since it's fixing that logic up.
--
Jeff Layton <jlayton@xxxxxxxxxx>
