area: link layergood first issuestatus: needs codetype: bug
仓库指标
- 星标
- (403 个星标)
- PR 合并指标
- (30 天内没有已合并 PR)
描述
When we've stored an LLCP update, we need to apply it to the connection state after the correct connection event number (or "instant"). Currently, this is done here:
However, this code is inside the RX handler, so if we happen to miss the event we're supposed to apply the update, we never do so, resulting in the connection being lost completely. This can be seen in the following log:
2.888649s - TRACE - #18 DATA(15->22)<- Header { LLID: DataCont, NESN: 1, SN: 1, MD: false, Length: 0 }, []
2.889144s - INFO - LLCP<- ConnectionUpdateReq(ConnectionUpdateData { win_size: 2, win_offset: 17, interval: 39, latency: 0, timeout: 500, instant: 29 })
...
2.963649s - TRACE - #28 DATA(11->18)<- Header { LLID: DataCont, NESN: 1, SN: 1, MD: false, Length: 0 }, []
2.964147s - TRACE - DATA(18->25): missed conn event #29
2.972340s - TRACE - DATA(25->32): missed conn event #30
2.980533s - TRACE - DATA(32->2): missed conn event #31
2.988725s - TRACE - DATA(2->16): missed conn event #32
2.996940s - TRACE - DATA(16->16): missed conn event #33
...
Event 29 is missed presumably due to interference, we never apply the update, all future events are missed because the connection state is now desynced.