The IP translation layer is using the neighbor table of the kernel to get the unicast link layer (mac) address for IP(v4|v6) addresses. The kernel can not only return unicast mac addresses to such an RTM_GETNEIGH request but also zero mac address. Such an address must be considered invalid because the global translation table may not only contain a unique client mac address entry for it. The translation from client mac to originator will therefore most likely return an unexpected originator.
Dropping these kind of (bogus) results avoids confusions while using things like batctl's ping or traceroute.
Reported-by: Andre Kasper Andre.Kasper@gmx.de Signed-off-by: Sven Eckelmann sven@narfation.org --- Cc: Andre Kasper Andre.Kasper@gmx.de
See https://www.open-mesh.org/issues/353 --- functions.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)
diff --git a/functions.c b/functions.c index cd92b60..3c340a2 100644 --- a/functions.c +++ b/functions.c @@ -571,6 +571,19 @@ static struct nla_policy neigh_policy[NDA_MAX+1] = { [NDA_PROBES] = { .type = NLA_U32 }, };
+static bool ether_addr_valid(const uint8_t *addr) +{ + /* no multicast address */ + if (addr[0] & 0x01) + return false; + + /* no zero address */ + if ((addr[0] | addr[1] | addr[2] | addr[3] | addr[4] | addr[5]) == 0) + return false; + + return true; +} + static int resolve_mac_from_parse(struct nl_msg *msg, void *arg) { struct nlattr *tb[NDA_MAX + 1]; @@ -616,6 +629,9 @@ static int resolve_mac_from_parse(struct nl_msg *msg, void *arg) mac = nla_data(tb[NDA_LLADDR]); l3addr = nla_data(tb[NDA_DST]);
+ if (!ether_addr_valid(mac)) + goto err; + if (memcmp(nl_arg->l3addr, l3addr, l3_len) == 0) { memcpy(nl_arg->mac_result, mac, ETH_ALEN); nl_arg->found = 1;
On 14/04/18 02:16, Sven Eckelmann wrote:
The IP translation layer is using the neighbor table of the kernel to get the unicast link layer (mac) address for IP(v4|v6) addresses. The kernel can not only return unicast mac addresses to such an RTM_GETNEIGH request but also zero mac address. Such an address must be considered invalid because the global translation table may not only contain a unique client mac address entry for it. The translation from client mac to originator will therefore most likely return an unexpected originator.
We already handle the case of multiple originators handling the same MAC address, no? In that case I think we pick the "best" originator.
This case sounds more like a validity check because "a zero MAC should not be in the translation table", or am I wrong?
Cheers,
On Samstag, 14. April 2018 04:34:42 CEST Antonio Quartulli wrote:
On 14/04/18 02:16, Sven Eckelmann wrote:
[...]
We already handle the case of multiple originators handling the same MAC address, no? In that case I think we pick the "best" originator.
Yes, but this doesn't make a lot of sense for multicast and zero mac addresses. The translate layer of batctl is usually used to ping/traceroute to some originator. But multicast and zero mac addresses don't represent a "client" which can be used to identify some originator. So it doesn't seem to make sense to allow them here.
Or even without the ping/traceroute stuff, the concept of calling `batctl translate` should give you an answer which you can understand. So it should tell you that batman-adv is very likely to transmit a unicast packet with this destination address to this originator. But this cannot work for multicast destination addresses because multiple answer should be given here - which is out of scope for this command. Which reminds me that I should propose a second patch which checks whether the input for translate_mac is "valid" before trying to translate it.
This case sounds more like a validity check because "a zero MAC should not be in the translation table", or am I wrong?
Partially, yes. I personally don't care (at the moment) whether there is a zero mac address in the translation table. The current translation table code (batadv_tt_local_add) doesn't check whether there is a zero mac address (is_zero_ether_addr). But Linus had some ideas when zero mac addresses can be useful - maybe he tell us whether it makes sense/problems to have them in the translation table.
Kind regards, Sven
On 14/04/18 15:10, Sven Eckelmann wrote:
On Samstag, 14. April 2018 04:34:42 CEST Antonio Quartulli wrote:
On 14/04/18 02:16, Sven Eckelmann wrote:
[...]
We already handle the case of multiple originators handling the same MAC address, no? In that case I think we pick the "best" originator.
Yes, but this doesn't make a lot of sense for multicast and zero mac addresses. The translate layer of batctl is usually used to ping/traceroute to some originator. But multicast and zero mac addresses don't represent a "client" which can be used to identify some originator. So it doesn't seem to make sense to allow them here.
Right.
Or even without the ping/traceroute stuff, the concept of calling `batctl translate` should give you an answer which you can understand. So it should tell you that batman-adv is very likely to transmit a unicast packet with this destination address to this originator. But this cannot work for multicast destination addresses because multiple answer should be given here - which is out of scope for this command. Which reminds me that I should propose a second patch which checks whether the input for translate_mac is "valid" before trying to translate it.
This case sounds more like a validity check because "a zero MAC should not be in the translation table", or am I wrong?
Partially, yes. I personally don't care (at the moment) whether there is a zero mac address in the translation table. The current translation table code (batadv_tt_local_add) doesn't check whether there is a zero mac address (is_zero_ether_addr). But Linus had some ideas when zero mac addresses can be useful - maybe he tell us whether it makes sense/problems to have them in the translation table.
Yeah, I agree that zero "may appear" in the table and therefore we have to "check" what we are getting back and whether it makes sense for us in this context.
Actually my comment was not about changing your approach, but just making it more explicit in the commit and in the error message.
An error message like like "returned invalid all-zero mac address" (or "multicast address") might help to distinguish similar "ambiguities" in the future. No?
Cheers,
Kind regards, Sven
On Samstag, 14. April 2018 10:11:28 CEST Antonio Quartulli wrote:
An error message like like "returned invalid all-zero mac address" (or "multicast address") might help to distinguish similar "ambiguities" in the future. No?
The current interface for the translation is "give me a string and I return NULL or a mac address". The resolving of the IPs for hostnames and the check of the neighbor table are done on an "I take what I get first" approach. Your suggestion would involve a change of this interface and parsing of additional information and tracking of states to make sure that the "best" result is returned (or a special error state). This is nothing which I will implement now.
Should I drop the patch for now?
Kind regards, Sven
On 14/04/18 17:20, Sven Eckelmann wrote:
On Samstag, 14. April 2018 10:11:28 CEST Antonio Quartulli wrote:
An error message like like "returned invalid all-zero mac address" (or "multicast address") might help to distinguish similar "ambiguities" in the future. No?
The current interface for the translation is "give me a string and I return NULL or a mac address". The resolving of the IPs for hostnames and the check of the neighbor table are done on an "I take what I get first" approach. Your suggestion would involve a change of this interface and parsing of additional information and tracking of states to make sure that the "best" result is returned (or a special error state). This is nothing which I will implement now.
Oh ok, I went through the code and now I better understand what you meant.
Should I drop the patch for now?
I'd suggest to still consider it for merging. Even if we can't be specific about the error, it is still better to stop the translation rather than generating unexpected results.
Cheers,
On 14/04/18 02:16, Sven Eckelmann wrote:
The IP translation layer is using the neighbor table of the kernel to get the unicast link layer (mac) address for IP(v4|v6) addresses. The kernel can not only return unicast mac addresses to such an RTM_GETNEIGH request but also zero mac address. Such an address must be considered invalid because the global translation table may not only contain a unique client mac address entry for it. The translation from client mac to originator will therefore most likely return an unexpected originator.
Dropping these kind of (bogus) results avoids confusions while using things like batctl's ping or traceroute.
Reported-by: Andre Kasper Andre.Kasper@gmx.de Signed-off-by: Sven Eckelmann sven@narfation.org
Acked-by: Antonio Quartulli a@unstable.cc
On Freitag, 13. April 2018 20:16:18 CEST Sven Eckelmann wrote:
The IP translation layer is using the neighbor table of the kernel to get the unicast link layer (mac) address for IP(v4|v6) addresses. The kernel can not only return unicast mac addresses to such an RTM_GETNEIGH request but also zero mac address. Such an address must be considered invalid because the global translation table may not only contain a unique client mac address entry for it. The translation from client mac to originator will therefore most likely return an unexpected originator.
Dropping these kind of (bogus) results avoids confusions while using things like batctl's ping or traceroute.
Reported-by: Andre Kasper Andre.Kasper@gmx.de Signed-off-by: Sven Eckelmann sven@narfation.org
Cc: Andre Kasper Andre.Kasper@gmx.de
Applied in 9f22e3d0ed2a [1]. But the ticket is not marked as resolved but it seems like the reporter wants a different behavior.
Kind regards, Sven
[1] https://git.open-mesh.org/batctl.git/commit/9f22e3d0ed2a3d83479f3db3b570f492...
b.a.t.m.a.n@lists.open-mesh.org
-
Antonio Quartulli -
Sven Eckelmann