net: fix __dst_negative_advice() race

[bmastbergen]

jira VULN-1303
cve CVE-2024-36971

commit-author Eric Dumazet <[email protected]>
commit 92f1655aa2b2294d0b49925f3b875a634bd3b59e
upstream-diff This change breaks the kabi.  Use the RH_KABI_REPLACE
              macro to define the negative_advice function such that
              check-kabi will still pass.  From rh_kabi.h:

              "The RH_KABI_REPLACE* macros attempt to add the ability
               to use the '_new' element while preserving size
               alignment and kabi agreement with the '_orig' element."

              Also, this version of the kernel does not have rcu
              locking in ip6_negative_advice like the upstream does
              so the rcu_read_lock/rcu_read_unlock calls are not
              in this changeset.

__dst_negative_advice() does not enforce proper RCU rules when sk->dst_cache must be cleared, leading to possible UAF.

RCU rules are that we must first clear sk->sk_dst_cache, then call dst_release(old_dst).

Note that sk_dst_reset(sk) is implementing this protocol correctly, while __dst_negative_advice() uses the wrong order.

Given that ip6_negative_advice() has special logic against RTF_CACHE, this means each of the three ->negative_advice() existing methods must perform the sk_dst_reset() themselves.

Note the check against NULL dst is centralized in
__dst_negative_advice(), there is no need to duplicate it in various callbacks.

Many thanks to Clement Lecigne for tracking this issue.

This old bug became visible after the blamed commit, using UDP sockets.

Fixes: a87cb3e48ee8 ("net: Facility to report route quality of connected sockets")
        Reported-by: Clement Lecigne <[email protected]>
Diagnosed-by: Clement Lecigne <[email protected]>
        Signed-off-by: Eric Dumazet <[email protected]>
        Cc: Tom Herbert <[email protected]>
        Reviewed-by: David Ahern <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
        Signed-off-by: Jakub Kicinski <[email protected]>
(cherry picked from commit 92f1655aa2b2294d0b49925f3b875a634bd3b59e)
        Signed-off-by: Brett Mastbergen <[email protected]>

Build log

build.log

Testing

The very limited kselftests were run before and after the change was applied:

selftests-before.log

selftests-after.log

brett@lycia ~/ciq/vuln-1303 % grep ^ok selftests-before.log | wc -l
2
brett@lycia ~/ciq/vuln-1303 % grep ^ok selftests-after.log | wc -l
2
brett@lycia ~/ciq/vuln-1303 %

[PlaidCat]

[kerneltoast]

I think a KABI break is required, otherwise an OOT module which really depends on the old KABI will introduce the CVE via its negative_advice callback.

[kerneltoast]

Does indeed break the KABI as desired, sorry for the confusion on my part.

[bmastbergen]

I think a KABI break is required, otherwise an OOT module which really depends on the old KABI will introduce the CVE via its negative_advice callback.

Just to sync up slack discussion here. THE RH_KABI_REPLACE macro does force users to use the new call. But it also fools the check-kabi tool into thinking everything is ok. In reality the kabi is broken by this change.