Thread (3 messages) 3 messages, 3 authors, 22h ago

[BUG] net/sched: act_csum can checksum beyond a short IPv6 jumbogram skb

flat view

From: shaos <hidden>
Date: 2026-10-22 14:47:41
Also in: lkml

Hello,

An unprivileged process with namespace-scoped CAP_NET_ADMIN and CAP_NET_RAW installs a csum action for UDP-Lite and sends a 300-byte IPv6 frame with a Jumbo Payload value of 0xffffffca and UDP-Lite coverage longer than the bytes present. The jumbo-length addition wraps the pull-length check, while the later checksum path still uses the unwrapped length.

HEAD commit:    c03c1a14507dfaf8b856c2608f9ac216021fc222
git tree:       30fd2ef82912c26cf3a58ad0ba6afc375fa851ac
source version: Linux 7.3.0, local master
compiler:       gcc 11.4.0, GNU ld 2.38

console output:
https://gist.githubusercontent.com/shaos26/6522612a4cac2b4db3e4cfc32ec8ea8f/raw/5b12fad1d38c90b7a8214e02bffeb7bfd0994ebe/console-output.txt
kernel config:
https://gist.githubusercontent.com/shaos26/6522612a4cac2b4db3e4cfc32ec8ea8f/raw/f6506700a588fd394fdd662c4bd91115e0cb8c3b/kernel-config.txt
reproducer sources:
https://gist.githubusercontent.com/shaos26/ef3fd748f2329565eca96a4f4e3b2c85/raw/7694fd571c683bb1f98793979b7bf2fe09559bd4/poc.c
https://gist.githubusercontent.com/shaos26/ef3fd748f2329565eca96a4f4e3b2c85/raw/200625ab70731666abbb27f0e233a42626ac7c58/run-unpriv.c
https://gist.githubusercontent.com/shaos26/ef3fd748f2329565eca96a4f4e3b2c85/raw/7c17996d150eb55421fd88cf62b4a71b8ba14207/seed-secret.c

Build the tested kernel with the attached configuration and boot a disposable QEMU guest. The linked userspace sources seed the skb allocation, drop privileges, enter fresh user and network namespaces, install the action, and capture checksum responses.

Impact and prerequisites:
The checksum oracle consumes 42 bytes beyond the skb's logical data. In two fresh boots, checksum output matched the checksum of a marker seeded by a different initial-namespace UID in 75/512 and 85/512 trials; two wrong-pattern controls had 0/512 matches. This demonstrates that data from a prior root-owned skb allocation can influence output readable by an unprivileged namespace process. The oracle is limited; no arbitrary-address read or privilege escalation is demonstrated.

Validation: Two fresh user/network-namespace positives matched the root-seeded marker checksum in 75/512 and 85/512 trials. Two wrong-pattern controls returned 0/512 matches.

Thanks,

shaos
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help