Hello everyone, Today, we had our INT8_MAX+2 th meeting with Mat and Ossama (Intel), Christoph (Apple), Davide, Paolo and Florian (RedHat) and myself (Tessares). Thanks again for this new good meeting! Here are the minutes of the meeting: Accepted patches: - The list of accepted patches can be seen on PatchWork: https://patchwork.ozlabs.org/project/mptcp/list/?state=3 netdev (if mptcp ML is in cc) (Mat Martineau): 1414602 [net-next,9/9] mptcp: let MPTCP create max size skbs 1414603 [net-next,8/9] mptcp: pm: simplify select_local_address() 1414599 [net-next,7/9] mptcp: parse and act on incoming FASTCLOSE optio 1414600 [net-next,6/9] tcp: parse mptcp options contained in reset pack 1414601 [net-next,5/9] mptcp: hold mptcp socket before calling tcp_done 1414598 [net-next,4/9] mptcp: use MPTCPOPT_HMAC_LEN macro 1414597 [net-next,3/9] selftests: mptcp: add the flush addrs testcase 1414596 [net-next,2/9] mptcp: remove address when netlink flushes addrs 1414595 [net-next,1/9] mptcp: attach subflow socket to parent cgroup 1416290 [v2,net-next] mptcp: clear use_ack and use_map when dropping other su... our repo (by: Geliang Tang, Paolo Abeni): 1416111 [mptcp-net] mptcp: properly annotate nested lock 1416006 [mptcp-net,v2] mptcp: fix pending data accounting 1415493 [mptcp-next] Squash to "mptcp: add the address ID assignment bitmap" 1415026 [mptcp-net] mptcp: fix security context on server socket 1414708 [mptcp-next] Squash to "[PATCH v3 mptcp-next 1/6] mptcp: add the outg... 1414604 [mptcp-next] Squash-to: "mptcp: push pending frames when subflow has ... 1413273 [v3,mptcp-next,6/6] selftests: mptcp: add the MP_PRIO testcases 1413272 [v3,mptcp-next,5/6] mptcp: add the mibs for MP_PRIO 1413271 [v3,mptcp-next,4/6] selftests: mptcp: add set_flags command in 1413270 [v3,mptcp-next,3/6] mptcp: add set_flags command in PM netlink 1413269 [v3,mptcp-next,2/6] mptcp: add the incoming MP_PRIO support 1413268 [v3,mptcp-next,1/6] mptcp: add the outgoing MP_PRIO support 1413262 [v4,mptcp-next,2/2] selftests: mptcp: add testcases for setting 1413254 [v4,mptcp-next,1/2] mptcp: add the address ID assignment bitmap 1409622 [RFC] selinux: handle MPTCP consistently with TCP Pending patches: - The list of pending patches can be seen on PatchWork: https://patchwork.ozlabs.org/project/mptcp/list/?state=* netdev (if mptcp ML is in cc) (by: Paolo Abeni): 1417092 linux: handle MPTCP consistently with TCP 1417085 [net,4/4] mptcp: fix pending data accounting 1417080 [net,3/4] mptcp: push pending frames when subflow has free spac 1417083 [net,2/4] mptcp: properly annotate nested lock 1417079 [net,1/4] mptcp: fix security context on server socket: - Waiting to be applied upstream - Done just after the meeting our repo (by: Florian Westphal, Geliang Tang): 1370700: RFC: [RFC,2/4] tcp: move selected mptcp helpers to tcp.h/mptcp.h 1370702: RFC: [RFC,4/4] tcp: parse tcp options contained in reset packets 1375893: RFC: [RFC,mptpcp-next] mptcp: add ooo prune support 1395128: RFC: [1/5] tcp: make two mptcp helpers available to tcp stack 1395131: RFC: [3/5] mptcp: add mptcp reset option support 1395133: RFC: [5/5] mptcp: send fastclose if userspace closes socket with unread data: - WIP 1415495: New: [v8,mptcp-next,1/8] mptcp: create the listening socket for 1415496: New: [v8,mptcp-next,2/8] mptcp: add port number check for MP_JO 1415497: New: [v8,mptcp-next,3/8] mptcp: add port number announced check 1415498: New: [v8,mptcp-next,4/8] mptcp: enable use_port when invoke add 1415499: New: [v8,mptcp-next,5/8] mptcp: deal with MPTCP_PM_ADDR_ATTR_PO 1415500: New: [v8,mptcp-next,6/8] selftests: mptcp: add port argument fo 1415501: New: [v8,mptcp-next,7/8] mptcp: add the mibs for ADD_ADDR with 1415502: New: [v8,mptcp-next,8/8] selftests: mptcp: add testcases for A: - Waiting for reviews - Mat will have a look 1416891: Changes Requested: [v2,mptcp-next,1/2] mptcp: convert IPv4 address to IPv4-mapped 1416892: Changes Requested: [v2,mptcp-next,2/2] selftests: mptcp: add IPv4-mapped IPv6 testcases: - Changes requested by Mat - Linked to issue 122 on GH - Discussions needed: - Direction taken: modifying entries in the local address - It seems not "clean": it seems more natural not to modify stuff for this check, maybe we should change the way the check is done? - i.e. when comparing addresses, we can check for IPv4 mapped - @Geliang: would it be OK to take this direction? 1416976: New: [mptcp-next,1/3] mptcp: create subflow or signal addr for newly added address 1416977: New: [mptcp-next,2/3] selftests: mptcp: use minus value for removing address numbers 1416978: New: [mptcp-next,3/3] selftests: mptcp: add testcases for newly added addresses: - Waiting for reviews - Mat started to look - Linked to https://github.com/multipath-tcp/mptcp_net-next/issues/19 - The goal of #19 what to allow the Netlink API to use a new endpoint (e.g. new WiFi link) - (Maybe later we can add a flag to disable this behaviour) - It makes sense to do that and that's what Geliang is doing here, good! - But is it OK to do this iteration there in the code at this place? - Patch: https://patchwork.ozlabs.org/project/mptcp/patch/73d430a234a018dbef408242... - Mat might need help for the review on this one Issues on Github: https://github.com/multipath-tcp/mptcp_net-next/issues/ Recently opened (latest from last week: 127) 130 [syzkaller] WARNING in mptcp_token_destroy_request [bug] [syzkaller]: - no repro yet 129 set subflow socket consistently [bug] [enhancement]: - When looking at buf autotuning, Paolo found not nice to have different struct sockets for client/outbound subflows and server/inbound subflows. - Sharing a common structure would be nice - Also, we need to traverse the conn_list to set the NOSPACE bit on each subflow - We might want to share this structure for [sg]etsockopt 128 When the last subflow is closed without DATA_FIN and msk Established, close msk (after a timeout) [enhancement]: - We might want different behaviours for inbound and outbound sockets, e.g.: - client might want to create new subflows over new subflows even if there is no existing one - server might want not being stuck and close connections - MPTCP keep alive might be something different, more advanced behaviour - A dedicated timeout might be good for now - For the moment, we don't have dedicated MPTCP setsockopt, maybe we can also wait - sysctl should be there to configure the default value - should be fine to add new sysctl just for MPTCP if they are specific - Or Netlink API but might be more tricky for the end user, e.g. when forcing some specific app to create MPTCP sockets instead of TCP one with LD_PRELOAD Bugs (opened, flagged as "bug" and assigned) 126 [syzkaller] WARNING in mptcp_reset_timer [bug] [syzkaller] @dcaratti: - With the repro, we reproduce another issue: 115, all the time :-) - will fix this one and check this 126 happens again. - but maybe linked 123 sndmsg autotuning is disable [bug] @pabeni: - As soon as autotuning is re-enabled, some selftests are failing (multiple xmit) - Very difficult to get complete link utilisation (no longer blocked by buffers) - Trying to use mptcp.org, Paolo is interested by BLEST - Min RTT in mptcp.org works fairly well in most scenarios but in heterogeneous, we might not fully use all links - sum of the thgput of all subflow * minRTT (?) - best way to try reaching full speed: set very very large TCP rcv and snt windows, just to see if we are not blocked because of that - what Paolo sees is: unstable at the beginning and better later. So if it is transferring a few small files, it is not great - Christoph: we need early reinjection. To avoid the packets on the slow links blocking the whole connection. It is explained there: https://www.usenix.org/conference/nsdi12/technical-sessions/presentation/... - The sender is blocked somehow, best is to analyse pcaps 122 PM doesn't create subflow with IPv4 mapped IPv6 socket [bug] @geliangtang: - See patches above ↑ 94 Packetdrill: after a received DATA_FIN, no new packets can be treated [bug] [packetdrill] @dcaratti: - Davide will check if he can still reproduce it Bugs (opened and flagged as "bug" and not assigned) 130 [syzkaller] WARNING in mptcp_token_destroy_request [bug] [syzkaller] 129 set subflow socket consistently [bug] [enhancement] 127 [syzkaller] WARNING in corrupted [bug] [syzkaller] 125 [syzkaller] KASAN: wild-memory-access Write in subflow_req_destructor [bug] [syzkaller] 121 [syzkaller] divide error in __tcp_select_window [bug] [syzkaller] 120 [interop] netnext is dropping packets, causing MPTCP-level retransmissions on mptcp.org [bug] 119 [syzkaller] memory leak in __get_filter [bug] [syzkaller] 115 [syzkaller] WARNING in sk_stream_kill_queues [bug] [syzkaller] 112 sporadic failure of mptcp_join.sh selftest 13 [bug] 110 [syzkaller] memory leak in __ip_mc_join_group [bug] [syzkaller] 107 Review use of WARN_ON() / WARN_ON_ONCE() [bug] 65 clearing properly the status in listen() [bug] 56 msk connection state set without msk lock [bug] In Progress (opened and assigned) 96 Python: add support for IPPROTO_MPTCP [enhancement] @matttbe 76 [gs]etsockopt per subflow: BPF [enhancement] @matttbe 54 ADD_ADDR: ports support [enhancement] @geliangtang 51 MP_PRIO support [enhancement] @geliangtang 19 let PM netlink update live sockets on local addresses list change [enhancement] @geliangtang 60 PM: netlink: events per connection [enhancement] - Florian is working on PM to share events - the API will not be exactly the same, prefix different - Florian will try to reduce the modifications Recently closed (since last week) 124 ./mptcp_connect.sh -m mmap test blocks: regression [bug] 117 Allow user-space to set endpoint ID [enhancement] [help wanted] @geliangtang 114 packetdrill: dss tolerance issue to emit the FIN+ACK [bug] [packetdrill] 106 [syzkaller] BUG: Bad page state [bug] [syzkaller] 104 [syzkaller] general protection fault in skb_release_data [bug] [syzkaller] @geliangtang 67 `./mptcp_connect.sh -m mmap` test blocks [bug] FYI: Current Roadmap: - Bugs: https://github.com/multipath-tcp/mptcp_net-next/projects/2 - Current merge window (5.11): https://github.com/multipath-tcp/mptcp_net-next/projects/6 - For later: https://github.com/multipath-tcp/mptcp_net-next/projects/4 - TODO: 5.12 Patches to send to netdev: - net: - everything has been sent - net-next: - / (closed) Extra tests: - news about Syzkaller? (Christoph): - See above ↑ - news about interop with mptcp.org? (Christoph): - / - news about Intel's kbuild? (Mat): - Mat confirms it is the same issue he had with selftests, see the meeting notes from last week - Mat has more access now, should be easier to retrieve more details about an issue - packetdrill (Davide): - / - CI (Matth): - / Next meeting: - We cancel the two next ones? → Yes Next meeting: - We propose to have the next meeting in 3 weeks, on Thursday, the 7th of January 2021. - Usual UTC time: 16:00 UTC (8am PST, 5pm CET, Midnight CST) - Still open to everyone! - https://annuel2.framapad.org/p/mptcp_upstreaming_20210107 Feel free to comment on these points and propose new ones for the next meeting! Talk to you next year, Matt -- Tessares | Belgium | Hybrid Access Solutions www.tessares.net