Message ID | 20220820140124.11325-1-gert@greenie.muc.de |
---|---|
State | Accepted |
Headers | show |
Series | [Openvpn-devel] DCO: require valid netbits setting for non-primary iroutes. | expand |
On 20 Aug 2022, at 16:01, Gert Doering wrote: > The existing DCO code had extra logic for "if this is not > MR_WITH_NETBITS, set 32/128 as address length", but only for > iroute addition. For iroute deletion, this was missing, and > subsequently iroute deletion for IPv4 host routes failed on > FreeBSD DCO (commit 3433577a99). > > Iroute handling differenciates between "primary" iroutes (coming > from anm IP pool or ccd/ifconfig-push), and "non-primary" iroutes, > coming from --iroute and --iroute-ipv6 statements in per-client config. > > "Primary" iroutes always use "-1" for their netbits, but since these > are not installed via DCO, this is of no concern here. Whether these > can and should be changed needs further study on internal route > learning and cleanup. > > Refactor options.c and multi.c to ensure that netbits is always set > for non-primary iroutes - and ASSERT() on this in the DCO path, so we can > find out if there might be other code violating this. > > Change options.c::option_iroute() to always set netbits=32 for IPv4 > host routes (options_iroute_ipv6() never differenciated). Since > netmask_to_netbits() also insists on "-1" for host routes, change > to netmask_to_netbits2(). > > Remove all the extra MR_WITH_NETBITS logic from dco.c, where it should > have never appeared. > > Signed-off-by: Gert Doering <gert@greenie.muc.de> Seems sane to me. Passes the FreeBSD automated tests. Minor style nitpicks below. Acked-by: Kristof Provost <kprovost@netgate.com> > diff --git a/src/openvpn/multi.c b/src/openvpn/multi.c > index 95414429..1f9d0201 100644 > --- a/src/openvpn/multi.c > +++ b/src/openvpn/multi.c > @@ -1241,6 +1241,7 @@ multi_learn_in_addr_t(struct multi_context *m, > /* "primary" is the VPN ifconfig address of the peer and already > * known to DCO, so only install "extra" iroutes (primary = false) > */ > + ASSERT(netbits>=0); /* DCO requires populated netbits */ I’d put spaces around the comparison, so ‘ASSERT(netbits >= 0);’. That seems to be what’s done in all conditionals anyway. > diff --git a/src/openvpn/options.c b/src/openvpn/options.c > index 2b0bb20c..21e2f69b 100644 > --- a/src/openvpn/options.c > +++ b/src/openvpn/options.c > @@ -1572,12 +1572,14 @@ option_iroute(struct options *o, > > ALLOC_OBJ_GC(ir, struct iroute, &o->gc); > ir->network = getaddr(GETADDR_HOST_ORDER, network_str, 0, NULL, NULL); > - ir->netbits = -1; > + ir->netbits = 32; /* host route if no netmask given */ > > if (netmask_str) > { > const in_addr_t netmask = getaddr(GETADDR_HOST_ORDER, netmask_str, 0, NULL, NULL); > - if (!netmask_to_netbits(ir->network, netmask, &ir->netbits)) > + ir->netbits = netmask_to_netbits2(netmask); > + > + if (ir->netbits<0) And here too. So ‘if (ir->netbits < 0)’ > { > msg(msglevel, "in --iroute %s %s : Bad network/subnet specification", > network_str, Regards, Kristof
Hi, On 20/08/2022 16:01, Gert Doering wrote: > The existing DCO code had extra logic for "if this is not > MR_WITH_NETBITS, set 32/128 as address length", but only for > iroute addition. For iroute deletion, this was missing, and > subsequently iroute deletion for IPv4 host routes failed on > FreeBSD DCO (commit 3433577a99). > > Iroute handling differenciates between "primary" iroutes (coming > from anm IP pool or ccd/ifconfig-push), and "non-primary" iroutes, > coming from --iroute and --iroute-ipv6 statements in per-client config. > > "Primary" iroutes always use "-1" for their netbits, but since these > are not installed via DCO, this is of no concern here. Whether these > can and should be changed needs further study on internal route > learning and cleanup. > > Refactor options.c and multi.c to ensure that netbits is always set > for non-primary iroutes - and ASSERT() on this in the DCO path, so we can > find out if there might be other code violating this. > > Change options.c::option_iroute() to always set netbits=32 for IPv4 > host routes (options_iroute_ipv6() never differenciated). Since > netmask_to_netbits() also insists on "-1" for host routes, change > to netmask_to_netbits2(). > > Remove all the extra MR_WITH_NETBITS logic from dco.c, where it should > have never appeared. > > Signed-off-by: Gert Doering <gert@greenie.muc.de> Thanks for this. It indeed makes sense and helps cleaning up a bit this netbits madness. I agree with Kristof that you need to have spaces around boolean operator. (How come that uncrustify did not complain? we may need to add another toggle..) sitnl does the right thing: 2022-08-24 00:50:50 net_route_v4_add: 8.8.8.8/32 via 10.10.0.3 dev tun0 table 0 metric 100 So, other than the space issue: Acked-by: Antonio Quartulli <a@unstable.cc>
Thanks for the reviewes. Dunno why I thought that "we must have no spaces around operators" :-) - so, fixed that whitespace plus 3 tabs that escaped. Patch has been applied to the master branch. commit 104e4ef1e3d49cccb3bc677bab9c24158f91b97b Author: Gert Doering Date: Sat Aug 20 16:01:24 2022 +0200 DCO: require valid netbits setting for non-primary iroutes. Signed-off-by: Gert Doering <gert@greenie.muc.de> Acked-by: Kristof Provost <kprovost@netgate.com> Acked-by: Antonio Quartulli <a@unstable.cc> Message-Id: <20220820140124.11325-1-gert@greenie.muc.de> URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg25044.html Signed-off-by: Gert Doering <gert@greenie.muc.de> -- kind regards, Gert Doering
diff --git a/src/openvpn/dco.c b/src/openvpn/dco.c index 99b544a5..cf20334c 100644 --- a/src/openvpn/dco.c +++ b/src/openvpn/dco.c @@ -603,26 +603,14 @@ dco_install_iroute(struct multi_context *m, struct multi_instance *mi, if (addrtype == MR_ADDR_IPV6) { - int netbits = 128; - if (addr->type & MR_WITH_NETBITS) - { - netbits = addr->netbits; - } - - net_route_v6_add(&m->top.net_ctx, &addr->v6.addr, netbits, + net_route_v6_add(&m->top.net_ctx, &addr->v6.addr, addr->netbits, &mi->context.c2.push_ifconfig_ipv6_local, dev, 0, DCO_IROUTE_METRIC); } else if (addrtype == MR_ADDR_IPV4) { - int netbits = 32; - if (addr->type & MR_WITH_NETBITS) - { - netbits = addr->netbits; - } - in_addr_t dest = htonl(addr->v4.addr); - net_route_v4_add(&m->top.net_ctx, &dest, netbits, + net_route_v4_add(&m->top.net_ctx, &dest, addr->netbits, &mi->context.c2.push_ifconfig_local, dev, 0, DCO_IROUTE_METRIC); } diff --git a/src/openvpn/multi.c b/src/openvpn/multi.c index 95414429..1f9d0201 100644 --- a/src/openvpn/multi.c +++ b/src/openvpn/multi.c @@ -1241,6 +1241,7 @@ multi_learn_in_addr_t(struct multi_context *m, /* "primary" is the VPN ifconfig address of the peer and already * known to DCO, so only install "extra" iroutes (primary = false) */ + ASSERT(netbits>=0); /* DCO requires populated netbits */ dco_install_iroute(m, mi, &addr); } @@ -1280,6 +1281,7 @@ multi_learn_in6_addr(struct multi_context *m, /* "primary" is the VPN ifconfig address of the peer and already * known to DCO, so only install "extra" iroutes (primary = false) */ + ASSERT(netbits>=0); /* DCO requires populated netbits */ dco_install_iroute(m, mi, &addr); } diff --git a/src/openvpn/options.c b/src/openvpn/options.c index 2b0bb20c..21e2f69b 100644 --- a/src/openvpn/options.c +++ b/src/openvpn/options.c @@ -1572,12 +1572,14 @@ option_iroute(struct options *o, ALLOC_OBJ_GC(ir, struct iroute, &o->gc); ir->network = getaddr(GETADDR_HOST_ORDER, network_str, 0, NULL, NULL); - ir->netbits = -1; + ir->netbits = 32; /* host route if no netmask given */ if (netmask_str) { const in_addr_t netmask = getaddr(GETADDR_HOST_ORDER, netmask_str, 0, NULL, NULL); - if (!netmask_to_netbits(ir->network, netmask, &ir->netbits)) + ir->netbits = netmask_to_netbits2(netmask); + + if (ir->netbits<0) { msg(msglevel, "in --iroute %s %s : Bad network/subnet specification", network_str,
The existing DCO code had extra logic for "if this is not MR_WITH_NETBITS, set 32/128 as address length", but only for iroute addition. For iroute deletion, this was missing, and subsequently iroute deletion for IPv4 host routes failed on FreeBSD DCO (commit 3433577a99). Iroute handling differenciates between "primary" iroutes (coming from anm IP pool or ccd/ifconfig-push), and "non-primary" iroutes, coming from --iroute and --iroute-ipv6 statements in per-client config. "Primary" iroutes always use "-1" for their netbits, but since these are not installed via DCO, this is of no concern here. Whether these can and should be changed needs further study on internal route learning and cleanup. Refactor options.c and multi.c to ensure that netbits is always set for non-primary iroutes - and ASSERT() on this in the DCO path, so we can find out if there might be other code violating this. Change options.c::option_iroute() to always set netbits=32 for IPv4 host routes (options_iroute_ipv6() never differenciated). Since netmask_to_netbits() also insists on "-1" for host routes, change to netmask_to_netbits2(). Remove all the extra MR_WITH_NETBITS logic from dco.c, where it should have never appeared. Signed-off-by: Gert Doering <gert@greenie.muc.de> --- src/openvpn/dco.c | 16 ++-------------- src/openvpn/multi.c | 2 ++ src/openvpn/options.c | 6 ++++-- 3 files changed, 8 insertions(+), 16 deletions(-)