Hi Sebastian, thanks for answering.
On Wed, Feb 26 2025 at 17:31:58, Sebastian Andrzej Siewior bigeasy@linutronix.de wrote:
Just to be clear: A problem exists in v6.6 independent of my doing and 401cb7dae813 happens to fix it? The commit fecef4cd42c6 is a fixup for 401cb7dae813.
If so, can you share syzbot's reproducer and/or backtrace/ report?
Actually, it's fecef4cd42c6 the one that fixes the issue, but 401cb7dae813 is needed for it to work: it implements the bpf_net_ctx_*() api, then fecef4cd42c6 uses bpf_net_ctx_set() and bpf_net_ctx_clear() in the tun driver.
Here's the syzkaller report on stable v6.6: https://pastebin.com/yD0zVD0c
and here's the C reproducer: https://pastebin.com/d415fUDU
Cheers, Ricardo