[<prev] [next>] [day] [month] [year] [list]
Message-ID: <20201118074611.0df443d6@hermes.local>
Date: Wed, 18 Nov 2020 07:46:11 -0800
From: Stephen Hemminger <stephen@...workplumber.org>
To: netdev@...r.kernel.org
Subject: Fw: [Bug 210255] New: IP_UNICAST_IF has no effect on connect()ed
UDP sockets
Begin forwarded message:
Date: Wed, 18 Nov 2020 15:06:51 +0000
From: bugzilla-daemon@...zilla.kernel.org
To: stephen@...workplumber.org
Subject: [Bug 210255] New: IP_UNICAST_IF has no effect on connect()ed UDP sockets
https://bugzilla.kernel.org/show_bug.cgi?id=210255
Bug ID: 210255
Summary: IP_UNICAST_IF has no effect on connect()ed UDP sockets
Product: Networking
Version: 2.5
Kernel Version: 5.9.0-36.fc34.x86_64
Hardware: All
OS: Linux
Tree: Mainline
Status: NEW
Severity: normal
Priority: P1
Component: IPV4
Assignee: stephen@...workplumber.org
Reporter: mzxreary@...inter.de
Regression: No
The IP_UNICAST_IF sockopt has an effect for unconnected UDP sockets, and is
used as key for the route lookup. However, when using connect() on an UDP
socket, then the routing decision is already done at connect() time, and unlike
the routing decision for unconnected sockets the IP_UNICAST_IF sockopt is not
taken into consideration then.
I figure this was simply forgotten when IP_UNICAST_IF was added, but I am
pretty sure this should be corrected so that connected and unconnected UDP
sockets behave more alike.
(SO_BINDTODEVICE/SO_BINDTOINDEX actually works on both equally, but given they
do a lot more than IP_UNICAST_IF they are no replacement)
(This was noticed in context of the UDP/DNS code in systemd-resolved, see this
for further discussion:
https://github.com/systemd/systemd/issues/11935#issuecomment-618691018)
(The research on this was done by hvenev, not me, I am just propagating his
findings here.)
--
You are receiving this mail because:
You are the assignee for the bug.
Powered by blists - more mailing lists