Re: Backport xfrm race condition fix to stable kernel version 6.6.y
Tahera Fahimi <[email protected]>
| Newsgroups | org.kernel.vger.netdev |
|---|---|
| Message-ID | <[email protected]> |
On Tue, Aug 18, 2026 at 11:32:28PM -0400, Sasha Levin wrote: > On Mon, Aug 17, 2026 at 03:10:09PM -0700, Tahera Fahimi wrote: > > I am writing to request the backport of the following patch to stable kernel versions 6.6.y, > > addressing CVE-2026-31516. > > > > Patch to Apply: > > Subject: xfrm: prevent policy_hthresh.work from racing with netns teardown > > Commit ID: 29fe3a61bcdce398ee3955101c39f89c01a8a77e Hi Sasha, Thanks for your reply. You are totally right. I noticed the issue after I sent the email. I sent this fix patch for review. https://lore.kernel.org/all/[email protected]/ Thanks, Tahera > This one is already in 7.1.y, 6.18.y and 6.12.y, but it can't be taken > as-is on 6.6 and older: the line it adds calls disable_work_sync(), which > only exists from v6.10 onwards, 86898fa6b8cd ("workqueue: Implement > disable/enable for (delayed) work items"). The cherry-pick applies > cleanly and then fails to build. > > Substituting cancel_work_sync() compiles, but it drops the re-queue > prevention the commit exists to add, and I'd rather not make that > substitution silently on a CVE fix. > > Could you send a tested backport for the pre-6.10 trees (6.6, 6.1, 5.15, > 5.10)? Netdev folks Cc'd in case they'd rather do it. > > -- > Thanks, > Sasha