On 11/15/24 1:46 PM, Sam Edwards wrote:
Hi Hangbin,
It took me a while to grasp but the problem seems to be a confusion about what it means to set a temporary's lifetimes to 0/0:
- "The mngtmpaddrs has gone away; this temporary is slated for
deletion by addrconf_verify_rtnl()" 2) "This temporary address itself shall no longer be used, regenerate it immediately."
The existing behavior makes sense for the #2 case, but not for the #1 case. It seems sensible to me to keep the #2 behavior as-is, because userspace might be setting a 0/0 lifetime to forcibly rotate the temporary.
So it sounds like (at least) one of three fixes is in order: a) Make ipv6_create_tempaddr() verify that the `ifp` is (still) alive+mngtmpaddrs, returning with an error code if not. b) Look at the 3 callsites for ipv6_create_tempaddr() and add the above verifications before calling. c) Add a function that calls ipv6_del_addr(temp) for every temporary with a specified ifpub, and use it instead of manage_tempaddrs(..., 0, 0, false, ...) when deleting/unflagging a mngtmpaddrs.
Personally I like option C the best. What are your thoughts?
Cheers,
Off the top of my head regarding recent changes, please include Maciej:
commit 69172f0bcb6a09110c5d2a6d792627f5095a9018 Author: Maciej Żenczykowski maze@google.com Date: Thu Jul 20 09:00:22 2023 -0700
ipv6 addrconf: fix bug where deleting a mngtmpaddr can create a new temporary address
and Alex in discussions around changes to temp addresses
commit f4bcbf360ac8dc424dc4d2b384b528e69b6f34d9 Author: Alex Henrie alexhenrie24@gmail.com Date: Tue Feb 13 23:26:32 2024 -0700
net: ipv6/addrconf: clamp preferred_lft to the minimum required