ok, I have to retract my initial assertion that this is not a problem. Indeed for IPv4 the breakage is not visible, but IPv6 in anything other than fib_index zero is not usable. The uninitialized entries in the IPv6 fib_index_to_table_index point to a *IPv4* session table. Of course that does not work and will crash.
The fact that fib_index is passed a u8 into session_table_get_or_alloc does not help either. A invalid fib_index of ~0 is converted to 255 and the fib_index_to_table_index for IPv6 happily extendeds to 255 entries. I'll try to come up with a fix, unless someone beats me to it. Regards, Andreas Am Do., 23. Apr. 2020 um 09:07 Uhr schrieb Andreas Schultz via lists.fd.io <andreas.schultz=travelping....@lists.fd.io>: > Hi, > > While trying to troubleshoot a session problem, I've come > across session_table_get_or_alloc: > > https://gerrit.fd.io/r/gitweb?p=vpp.git;a=blob;f=src/vnet/session/session_lookup.c;h=4de6fdbe547499fcd256e387370cc5bf0c810206;hb=refs/heads/master#l158 > > To me it looks like this function could leave holes of invalid/wrong > table_indexes in the fib_index_to_table_index. > > Index 0 is initialized explicitly (fib_index = 0), but the other indexes > are only initialized with a valid session table index when the first > session in that fib_index is opened. > When the first session happens to be in e.g. fib_index 10, all indexes > from 1 up to including 9 are filled with zero (by vec_validate). The > problem is that for those fib indexes the session table index is now 0. > Fortunately, that session table index is valid, because it points to the > session table for fib_index 0. However, it also means that sessions from > the various fibs leak into each other. > > Or do I miss something? > > Regards > Andreas > > -- > > Andreas Schultz > > -- > > -- Andreas Schultz
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#16146): https://lists.fd.io/g/vpp-dev/message/16146 Mute This Topic: https://lists.fd.io/mt/73213863/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
