Re: [syzbot] possible deadlock in ktime_get_coarse_ts64

[Dmitrii Banshchikov]

On Fri, Nov 05, 2021 at 06:24:30PM +0100, Thomas Gleixner wrote:
> On Fri, Nov 05 2021 at 21:03, Dmitrii Banshchikov wrote:
> > On Fri, Nov 05, 2021 at 08:53:06AM -0700, Alexei Starovoitov wrote:
> >> > Timestamps from within a tracepoint can only be taken with:
> >> >
> >> >          1) jiffies
> >> >          2) sched_clock()
> >> >          3) ktime_get_*_fast_ns()
> >> >
> >> > Those are NMI safe and can be invoked from anywhere.
> >> >
> >> > All other time getters which have to use the timekeeping seqcount
> >> > protection are prone to live locks and _cannot_ be used from
> >> > tracepoints ever.
> >> 
> >> Obviously.
> >> That helper was added for networking use cases and accidentally
> >> enabled for tracing.
> >
> > Sorry for that.
> > I'm preparing a patch that will forbid using bpf_ktime_get_coarse_ns()
> > helper in BPF_LINK_TYPE_RAW_TRACEPOINT.
> 
> It cannot be used in TRACING and PERF_EVENT either. But those contexts
> have to exclude other functions as well:
> 
>      bpf_ktime_get_ns
>      bpf_ktime_get_boot_ns
> 
> along with
> 
>     bpf_spin_lock/unlock
>     bpf_timer_*

1) bpf_ktime_get_ns and bpf_ktime_get_boot_ns use
ktime_get_{mono,boot}_fast_ns.
2) bpf_spin_lock/unlock have notrace attribute set.
3) bpf_timer_* helpers fail early if they are in NMI.

Why they have to be excluded?



-- 

Dmitrii Banshchikov