| Message ID | 20210202174754.6861-1-fw@strlen.de |
|---|---|
| Headers | show |
| Series | mptcp: netlink event fixes | expand |
On Tue, 2 Feb 2021, Florian Westphal wrote: > While testing upcoming mptcp reset option support i found a few > issues with netlink events. > > 1. When remote side is closing, the subflow close event will > be delayed until after the mptcp socket has shut down. > > Schedule the worker to do a cleanup scan to avoid this. > > 2. When the mptcp socket is closing we can avoid sending > individual 'subflow is closing' events; entire connection > is going down so might as well wait for the normal mptcp-level > close event. > Those two fixes look good for the export branch. > 3. Move the subflow close loop after dead check. > This is not a fix, it would be fine to leave things as-is > if you prefer -- its just that the extra conn_list walk > can be avoided if the msk is already dead. This also skips the MPTCP_EVENT_SUB_CLOSED events when the whole socket is closed, right? That's worthwhile. It doesn't look like delaying the ssk close(s) will change the other calls that might happen for the PM / EOF / DATA_FIN stuff, so I think the patch is ok. If there's a problem we can drop it from the export branch. Thanks! Mat > > Florian Westphal (3): > mptcp: schedule worker when subflow is closed > mptcp: skip MPTCP_EVENT_SUB_CLOSED event when mptcp socket isn't > established > mptcp: move subflow close loop after sk close check > > net/mptcp/protocol.c | 9 +++++---- > net/mptcp/subflow.c | 7 +++++++ > 2 files changed, 12 insertions(+), 4 deletions(-) > > -- > 2.26.2 -- Mat Martineau Intel
Hi Florian, Mat, On 02/02/2021 18:47, Florian Westphal wrote: > While testing upcoming mptcp reset option support i found a few > issues with netlink events. Thank you for the patches and the reviews! As discussed on IRC, I squashed patch 2/3 in "mptcp: add netlink event support": - 51e4bae2c6c9: "squashed" patch 2/3 in "mptcp: add netlink event support" - Results: f3a1a5d5ef41..eee71bc20e17 Then I added patches 1/3 and 3/3/ in our tree between "mptcp: split __mptcp_close_ssk helper" and "mptcp: pass subflow socket to a few helpers". I also added Mat's RvB tags: - 95fc52697bf7: mptcp: schedule worker when subflow is closed - 74975e28f8ec: mptcp: move subflow close loop after sk close check - Results: eee71bc20e17..7b3ea53ce791 Tests + export are in progress! Cheers, Matt
While testing upcoming mptcp reset option support i found a few issues with netlink events. 1. When remote side is closing, the subflow close event will be delayed until after the mptcp socket has shut down. Schedule the worker to do a cleanup scan to avoid this. 2. When the mptcp socket is closing we can avoid sending individual 'subflow is closing' events; entire connection is going down so might as well wait for the normal mptcp-level close event. 3. Move the subflow close loop after dead check. This is not a fix, it would be fine to leave things as-is if you prefer -- its just that the extra conn_list walk can be avoided if the msk is already dead. Florian Westphal (3): mptcp: schedule worker when subflow is closed mptcp: skip MPTCP_EVENT_SUB_CLOSED event when mptcp socket isn't established mptcp: move subflow close loop after sk close check net/mptcp/protocol.c | 9 +++++---- net/mptcp/subflow.c | 7 +++++++ 2 files changed, 12 insertions(+), 4 deletions(-)