On Fri, Sep 6, 2024 at 3:57 PM Jakub Kicinski <[email protected]> wrote: > > On Fri, 6 Sep 2024 15:41:47 -0700 Alexei Starovoitov wrote: > > The urgency is now because the situation is dire. > > The verifier assumes that skb is not null and will remove > > if (!skb) check assuming that it's a dead code. > > Meaning verifier currently isn't ready for patch 4? > Or we can crash 6.11-rc6 by attaching to a trace_tcp_send_reset() > and doing > printf("%d\n", skb->len); > ?
depends on the prog type and how it's attached, but yes :( Without Philo's patches. It was reported here: https://lore.kernel.org/bpf/[email protected]/ Jiri did the analysis. These files would need to be annotated: include/trace/events/afs.h include/trace/events/cachefiles.h include/trace/events/ext4.h include/trace/events/fib.h include/trace/events/filelock.h include/trace/events/host1x.h include/trace/events/huge_memory.h include/trace/events/kmem.h include/trace/events/netfs.h include/trace/events/power.h include/trace/events/qdisc.h include/trace/events/rxrpc.h include/trace/events/sched.h include/trace/events/sunrpc.h include/trace/events/tcp.h include/trace/events/tegra_apb_dma.h include/trace/events/timer_migration.h include/trace/events/writeback.h which is 18 out of 160. All other options are worse.
