Branch: refs/heads/sctp-fixes
Home: https://github.com/kronosnet/kronosnet
Commit: 42f08dda4c698b704c5b90f65374cced1e5aa893
https://github.com/kronosnet/kronosnet/commit/42f08dda4c698b704c5b90f65374c…
Author: Fabio M. Di Nitto <fdinitto(a)redhat.com>
Date: 2020-01-27 (Mon, 27 Jan 2020)
Changed paths:
M libknet/threads_rx.c
Log Message:
-----------
[rx] send reply packets only when transport is connected
Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com>
Commit: 4496bb25242ccb61ad914a9fa07672a9ffc0ca28
https://github.com/kronosnet/kronosnet/commit/4496bb25242ccb61ad914a9fa0767…
Author: Christine Caulfield <ccaulfie(a)redhat.com>
Date: 2020-01-27 (Mon, 27 Jan 2020)
Changed paths:
M libknet/transport_common.c
M libknet/transport_sctp.c
Log Message:
-----------
[rx] Don't return 512 EOF messages from _recvmmsg
If recvmsg() returns 0 for EOF then it's going to do so
until the error is rectified or read with getsockopt(). But
the _recvmmsg() wrapper keeps reading until the vector is full
thus returning a block of 512 EOF messages all of which the caller
has to plough through.
This patch causes _recvmmsg() to return as soon as it has got
the first EOF so the the caller can deal with it in good time
and not spin looking at the same thing over and over again.
I've also fixed a couple of typos in related comments
Commit: 89be99344e16416238b9769f6decd8a50cb464c6
https://github.com/kronosnet/kronosnet/commit/89be99344e16416238b9769f6decd…
Author: Christine Caulfield <ccaulfie(a)redhat.com>
Date: 2020-01-27 (Mon, 27 Jan 2020)
Changed paths:
M libknet/transport_common.c
Log Message:
-----------
[tx] Don't Clear out msghdr for all transports.
When sending a message to multiple links, if one of those links
is not connection-oriented then msg_name & msg_namelen would be cleared,
thus breaking the send to any subsequent non-connection-oriented links.
So now, if we need to clear out msg_name & msg_namelen, we take a copy of the
msghdr and edit that instead,
Commit: a0e03c376b5d5e3bd4ec2fd3c61cbf850ac53b62
https://github.com/kronosnet/kronosnet/commit/a0e03c376b5d5e3bd4ec2fd3c61cb…
Author: Fabio M. Di Nitto <fdinitto(a)redhat.com>
Date: 2020-01-27 (Mon, 27 Jan 2020)
Changed paths:
M libknet/threads_rx.c
M libknet/transport_udp.c
M libknet/transports.h
Log Message:
-----------
[rx] use defines to determine RX data types vs random numbers
also extend a bit to make ready for SCTP extra return codes
Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com>
Commit: 944f27e835cb25e260abc1ab06d79da2dc242049
https://github.com/kronosnet/kronosnet/commit/944f27e835cb25e260abc1ab06d79…
Author: Fabio M. Di Nitto <fdinitto(a)redhat.com>
Date: 2020-01-27 (Mon, 27 Jan 2020)
Changed paths:
M libknet/transport_sctp.c
Log Message:
-----------
[sctp] major surgery to use only SCTP events to determine socket status
- drop concept of on_connected_epoll to determine if socket is ready or not
- provide much better debugging output at all levels
- incorporate fix from Xin Long <lxin(a)redhat.com> to gather socket status
at the right time
- deal with a recent kernel change on SCTP socket that broke knet (from rhel7):
[net] sctp: allow delivering notifications after receiving SHUTDOWN
Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com>
Compare: https://github.com/kronosnet/kronosnet/compare/a785aa9b5b2f...944f27e835cb
Branch: refs/heads/stable1-proposed
Home: https://github.com/kronosnet/kronosnet
Commit: 4c81b969aac6afca770e31f6572c36ab761d7db6
https://github.com/kronosnet/kronosnet/commit/4c81b969aac6afca770e31f6572c3…
Author: Christine Caulfield <ccaulfie(a)redhat.com>
Date: 2020-01-24 (Fri, 24 Jan 2020)
Changed paths:
M libknet/transport_udp.c
Log Message:
-----------
[udp] don't make socket spin if a network I/F is down
UDP treats ENETUNREACH as a temporary error and just retries,
but this causes the TX thread to spin just doing sendto() therefore
blocking all other traffic.
(To reproduce this try starting corosync with 2 links configured in
corosync.conf but only one of them configured to the 'right' address
- it will spin in a tight loop and need to be killed with -9)
SCTP does not seem to suffer from this.
Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com>