Conversation
|
We've been hitting this elusive issue in production for a few weeks. Our duel-stack IPv4/IPv6 server blocking our own whitelisted IPv4 addresses. We've replicated the issue in a controlled environment. I came across this pull-request, reviewed and applied the patch to our builds of mod_evasive, and tested it against our replication process: Our IPv4 addresses are no longer blocked, and mod_evasive works normally for us now. We will watch closely over the next few weeks, and will report here if the patch gives us any issues, but for now things seem to be good. Thank you @TomCan for the patch :) |
|
Life saver... |
|
Unfortunately this commit e8c53e1 introduces another bug. The address masks from earlier whitelist directives erroneously apply to all subsequent entries inside the is_whitelisted address matching loop. The corrected patch is here: mod_evasive-afinet.patch |
|
@vlpast Can confirm, we seem to have been hit by this today, after adding an IPv6 subnet earlier in our whitelist, some of the latter IPv4 addresses started getting blocked. We are adopting your corrected patch for now. |
In dual stack environments, r->useragent_addr might return an IPv4 mapped IPv6 address instead of a regular IPv4 address. In that case, apr_sockaddr_t->family will return AF_INET6 instead of AF_INET and will not take any IPv4 entries on the whitelist into account. This can results in whitelisted IPv4 address to be blocked anyway.
This PR solves this by adding additional checks to see if this is in fact an IPv4 mapped IPv6 address, and adjusts the matching logic accordingly.