Re: Allocation failure with subsequent kernel crash

I've been looking into it a bit today and still am. Given you've seen
this on x86_32 and also on older kernels, I presume JIT was not involved
(/proc/sys/net/core/bpf_jit_enable is 0). Do you run any specific workload
until you trigger this (e.g. fuzzer on BPF), or any specific event that
triggers at that time after ~5hrs? Or only systemd on idle machine? Have
you managed to reproduce this also elsewhere? Bisect seems indeed painful
but would help tremendously; perhaps also dumping the BPF insns that are
loaded at that point in time.

Daniel,
  I've been trying for days to bisect this, but it is hard to
reproduce. However, I did have a question.

The crash is happening when bpf_prog_load() hits an error case and
then jumps to free_used_maps(prog->aux). However, I don't see an
obvious place where the 'aux' field gets initialized in
bpf_prog_load(). So it might easily be zero/null.

Could that explain the crash due to "unable to handle kernel NULL
pointer dereference"?

- Matthew