Hello, We are trying synchronise conntrack tables across multiple machines using conntrackd. We only want to synchronise conntrack for a limited set of IP addresses (for example: 10.132.159.60 and 10.132.159.62). I'm working on Debian stable packages : * conntrackd 1.4.7-1 * libnetfilter-conntrack3 1.0.9-6 Here is what I wrote in conntrackd.conf : Filter From Kernelspace { ... Address Accept { IPv4_address 10.132.159.60 IPv4_address 10.132.159.62 } ... } Except that does not work : After debugging, it appear that in this configuration the kernel does not send the conntrack element to conntrackd. If I understood correctly, this filter is read by conntrackd/libnetfilter-conntrack as "Accept CT that matches 10.132.159.60 AND 10.132.159.62" which won't happen. Switching this to Userspace filtering does work : CT are sent from the kernel to conntrackd and then synchronized across the other instances. This difference in user/kernel filtering sounds like a bug on one side. Userspace filtering is a bit heavy for the system, and we would really like to stay on kernel filtering. I've also tried to use one "Address Accept{}" bloc for each IPv4_address but that does not work either. Some random elements I've gathered while debugging this: * All the example I could find only used "Address Ignore {}" blocks so I guess the "Address Accept" option is not heavily used? * The code adding the IP to the filter is here : https://salsa.debian.org/pkg-netfilter-team/pkg-conntrack-tools/-/blob/master/src/read_config_yy.c#L3258 nfct_filter_add_attr(STATE(filter), NFCT_FILTER_SRC_IPV4, &filter_ipv4); Interestingly, the error messages some lines higher only mention the "ignore pool" My questions: * Has anyone pointers on how to setup this usecase (kernel filtering + accept on multiple IPs) ? * Is this a bug? (In that case, I'd push this to the bugzilla) Thanks in advance, Best regards, -- Yoann Congal Smile ECS - Tech Expert