vxlan_vnifilter_dump_dev() coalesces a contiguous run of VNIs sharing a remote into one VXLAN_VNIFILTER_ENTRY with no upper bound on the span. A device may hold the whole 24-bit space, populated by several requests, and dump it as a single entry START..END. Now that a single request is bounded to VXLAN_VNI_FILTER_MSG_MAX VNIs, replaying such an entry in one RTM_NEWTUNNEL is rejected: the kernel emits an entry it will not read back, so a dump/replay of a device's VNI configuration fails.
Clamp a merged run to VXLAN_VNI_FILTER_MSG_MAX VNIs so every entry the dump produces is one the input path accepts. The run is broken in the merge condition, by ending it once it reaches the limit even when the next VNI is contiguous and shares the remote; the two representations then agree on the same bound. Resume across netlink message boundaries is unchanged. cb->args[1] counts the VNI nodes already dumped and is advanced only when a completed entry is written; a run is still a gapless block, so its node count equals vnirange() + 1 as before, and the clamp only moves where a run ends. A device with more contiguous VNIs than fit in one skb still resumes correctly, now split into limit-sized entries rather than one. Assisted-by: LLM Signed-off-by: Ali Firas <[email protected]> --- Notes: v3: new patch. Clamp the dumped run to the request limit so its output replays. drivers/net/vxlan/vxlan_vnifilter.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/net/vxlan/vxlan_vnifilter.c b/drivers/net/vxlan/vxlan_vnifilter.c index 13f4e115701a..92ea1fc94f45 100644 --- a/drivers/net/vxlan/vxlan_vnifilter.c +++ b/drivers/net/vxlan/vxlan_vnifilter.c @@ -382,6 +382,7 @@ static int vxlan_vnifilter_dump_dev(const struct net_device *dev, continue; } if (!dump_stats && vnirange(vend, v) == 1 && + vnirange(vbegin, v) < VXLAN_VNI_FILTER_MSG_MAX && vxlan_addr_equal(&v->remote_ip, &vend->remote_ip)) { goto update_end; } else { -- 2.53.0
