If refreshed inode is snapshot, we can reload entire inode and switch to working VDI immediately. It is required by the new refreshing condition introduced in the commit 8f781f02ab of sheepdog.
Cc: Teruaki Ishizaki <[email protected]> Signed-off-by: Hitoshi Mitake <[email protected]> --- usr/bs_sheepdog.c | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+) diff --git a/usr/bs_sheepdog.c b/usr/bs_sheepdog.c index 56630ac..99519ef 100644 --- a/usr/bs_sheepdog.c +++ b/usr/bs_sheepdog.c @@ -704,6 +704,34 @@ static int reload_inode(struct sheepdog_access_info *ai, int is_snapshot) ret = -1; goto ret; } + + if (!!ai->inode.snap_ctime) { + /* + * This is a case like below: + * take snapshot -> write something -> failover + * + * Because invalidated inode is readonly and latest + * working VDI can have COWed objects, we need to + * resolve VID and reload its entire inode object. + */ + memset(tag, 0, sizeof(tag)); + + ret = find_vdi_name(ai, ai->inode.name, CURRENT_VDI_ID, + tag, &vid, 1); + if (ret) { + ret = -1; + goto ret; + } + + ret = read_object(ai, (char *)&ai->inode, + vid_to_vdi_oid(vid), + ai->inode.nr_copies, SD_INODE_SIZE, 0, + &need_reload); + if (ret) { + ret = -1; + goto ret; + } + } } ai->min_dirty_data_idx = UINT32_MAX; -- 1.9.1 -- sheepdog mailing list [email protected] https://lists.wpkg.org/mailman/listinfo/sheepdog
