Re: [PATCH 1/2] dccp: ccid: move timers to struct dccp_sock
Jakub Kicinski <[email protected]>
| Newsgroups | org.kernel.vger.dccp,org.kernel.vger.linux-kernel,org.kernel.vger.netdev |
|---|---|
| Message-ID | <20201110081655.29cbcd34@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com> |
On Tue, 10 Nov 2020 08:19:32 -0300 Thadeu Lima de Souza Cascardo wrote: > Yeah, I agree with your initial email. The patch I submitted for that fix needs > rework, which is what I tried and failed so far. I need to get back to some > testing of my latest fix and find out what needs fixing there. > > But I am also saying that simply doing a del_timer_sync on disconnect paths > won't do, because there are non-disconnect paths where there is a CCID that we > will remove and replace and that will still trigger a timer UAF. > > So I have been working on a fix that involves a refcnt on ccid itself. But I > want to test that it really fixes the problem and I have spent most of the time > finding out a way to trigger the timer in a race with the disconnect path. Sounds good, thanks a lot for working on this! > And that same test has showed me that this timer UAF will happen regardless of > commit 2677d20677314101293e6da0094ede7b5526d2b1, which led me into stating that > reverting it should be done in any case. > > I think I can find some time this week to work a little further on the fix for > the time UAF.