From: "cron2 (Code Review)"
Date: Wed, 14 Jan 2026 11:51:25 +0000
In-Reply-To:
References:
Reply-To: gert@...1296..., arne-openvpn@...1227...,
openvpn-devel@lists.sourceforge.net, antonio@...2726...
Message-ID: <519b72d4eb87e0838c33d2b82dc0b4ee72cb6cbc-EmailReplacePatchSet-HTML@...2715...>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Disposition: inline
Content-Type: multipart/alternative; boundary="4k4+tkcR61o="; charset=UTF-8
Subject: [Openvpn-devel] [XS] Change in openvpn[master]: Repair interaction
between DCO and persist-tun after reconnection
List-Id:
List-Unsubscribe: ,
List-Archive:
List-Post:
List-Help:
List-Subscribe: ,
To: plaisthos
Cc: openvpn-devel
--4k4+tkcR61o=
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Hello ordex, plaisthos,
I'd like you to reexamine a change. Please visit
http://gerrit.openvpn.net/c/openvpn/+/1473?usp=3Demail
to look at the new patch set (#3).
The following approvals got outdated and were removed:
Code-Review+2 by ordex
Change subject: Repair interaction between DCO and persist-tun after reconn=
ection
......................................................................
Repair interaction between DCO and persist-tun after reconnection
When --persist-tun is active, openvpn userland on Linux and FreeBSD fails
to re-enable "poll for DCO events" after a reconnect (e.g. triggered by
a ping timeout). The reconnect will still work fine, but the *next*
DCO event notification from the kernel will not be received by OpenVPN
userland, and so the system will get into an inconsistent state (Userland
assumes "all is well", kernel DCO has disconnected the peer, connection
is broken until the next tls-renegotion and/or manual restart, *and* the
next DCO key setup might fail due to "peer id gone").
This only affects client side, --server tun is always "persistent", and
there is no "full restart" (and the code path in question is also
only used for client and p2p server).
The root cause is an incorrect check for "is this interface up?" when
calling dco_event_set() in forard.c::io_wait() - "c2.did_open_tun" is
only true if the tun interface was actually configured on this reconnect,
which it isn't if --persist-tun is active. Replace with a check for
"do we have a tuntap structure, and if yes, do we have active DCO?"
which reflects the original intent much better.
The original code also had a check for "out_socket & EVENT_READ" there,
which did to some extend avoid calling dco_event_set() for every single
UDP packet sent and received by userland - but this only worked on initial
connection, and is always true on reconnect, so this condition was removed
for simplicity. We should come back here...
v2:
- some language fixes on the commit message
- do not check ->dco.open in forward.c, as this is not available if
not on FreeBSD, or if compiled with --disable-dco.
FreeBSD DCO does the "if (!dco || !dco->open)" check in dco_event_set()
anyway, so it's not needed, and Linux DCO has "dco->nl_sock", which is
also reliably set/unset, and checked by dco_event_set() already.
Github: OpenVPN/openvpn#947
Change-Id: Idbd0a47ba4d297a833a350611a23f19fd9a797b5
Signed-off-by: Gert Doering
Acked-by: Antonio Quartulli
Gerrit URL: https://gerrit.openvpn.net/c/openvpn/+/1473
Message-Id: <20260114112403.7046-1-gert@...1296...>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg35=
239.html
Signed-off-by: Gert Doering
---
M src/openvpn/forward.c
1 file changed, 1 insertion(+), 1 deletion(-)
git pull ssh://gerrit.openvpn.net:29418/openvpn refs/changes/73/1473/3
diff --git a/src/openvpn/forward.c b/src/openvpn/forward.c
index d208c21..39ac3b3 100644
--- a/src/openvpn/forward.c
+++ b/src/openvpn/forward.c
@@ -2197,7 +2197,7 @@
multi_io_process_flags(c, c->c2.event_set, flags, &out_socket, &out_tu=
ntap);
#if defined(TARGET_LINUX) || defined(TARGET_FREEBSD)
- if (out_socket & EVENT_READ && c->c2.did_open_tun)
+ if (c->c1.tuntap)
{
dco_event_set(&c->c1.tuntap->dco, c->c2.event_set, (void *)dco_shi=
ft);
}
--
To view, visit http://gerrit.openvpn.net/c/openvpn/+/1473?usp=3Demail
To unsubscribe, or for help writing mail filters, visit http://gerrit.openv=
pn.net/settings?usp=3Demail
Gerrit-MessageType: newpatchset
Gerrit-Project: openvpn
Gerrit-Branch: master
Gerrit-Change-Id: Idbd0a47ba4d297a833a350611a23f19fd9a797b5
Gerrit-Change-Number: 1473
Gerrit-PatchSet: 3
Gerrit-Owner: cron2
Gerrit-Reviewer: ordex
Gerrit-Reviewer: plaisthos
Gerrit-CC: openvpn-devel
--4k4+tkcR61o=
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
The followi=
ng approvals got outdated and were removed:
Code-Review+2 by ordex
Repair interaction between DCO and =
persist-tun after reconnection
When --persist-tun is active, openvpn=
userland on Linux and FreeBSD fails to re-enable "poll for DCO eve=
nts" after a reconnect (e.g. triggered by a ping timeout). The rec=
onnect will still work fine, but the *next* DCO event notification from =
the kernel will not be received by OpenVPN userland, and so the system w=
ill get into an inconsistent state (Userland assumes "all is well&q=
uot;, kernel DCO has disconnected the peer, connection is broken until t=
he next tls-renegotion and/or manual restart, *and* the next DCO key set=
up might fail due to "peer id gone").
This only affects cl=
ient side, --server tun is always "persistent", and there is n=
o "full restart" (and the code path in question is also only u=
sed for client and p2p server).
The root cause is an incorrect check=
for "is this interface up?" when calling dco_event_set() in f=
orard.c::io_wait() - "c2.did_open_tun" is only true if the tun=
interface was actually configured on this reconnect, which it isn't=
if --persist-tun is active. Replace with a check for "do we have =
a tuntap structure, and if yes, do we have active DCO?" which refle=
cts the original intent much better.
The original code also had a ch=
eck for "out_socket & EVENT_READ" there, which did to some=
extend avoid calling dco_event_set() for every single UDP packet sent a=
nd received by userland - but this only worked on initial connection, an=
d is always true on reconnect, so this condition was removed for simplic=
ity. We should come back here...
v2: - some language fixes on =
the commit message - do not check ->dco.open in forward.c, as this =
is not available if not on FreeBSD, or if compiled with --disable-dc=
o. FreeBSD DCO does the "if (!dco || !dco->open)" check=
in dco_event_set() anyway, so it's not needed, and Linux DCO ha=
s "dco->nl_sock", which is also reliably set/unset, and=
checked by dco_event_set() already.
Github: OpenVPN/openvpn#947 =
Change-Id: Idbd0a47ba4d297a833a350611a23f19fd9a797b5 Signed-off-by: =
Gert Doering <gert@...1296...> Acked-by: Antonio Quartulli <ant=
onio@...2726...> Gerrit URL: https://gerrit.openvpn.net/c/openvpn/+/1=
473 Message-Id: <20260114112403.7046-1-gert@...1296...> URL: ht=
tps://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg35239.htm=
l Signed-off-by: Gert Doering <gert@...1296...> --- M src/op=
envpn/forward.c 1 file changed, 1 insertion(+), 1 deletion(-)