On Tue, 03 Nov 2026 03:03:02 +0000, Alexei Starovoitov wrote:
> On Sun, Nov 01, 2026 at 11:37 PM Feng Yang <[email protected]> wrote:
> > Route these paths to bpf_get_stack_sleepable() and mark the new protos
> > might_sleep, since build_id_parse() may fault in file data for
> > BPF_F_USER_STACK | BPF_F_USER_BUILD_ID.  Sleepable programs that call
> > bpf_get_stack() from inside a bpf_rcu_read_lock() region are now rejected
> > by the verifier, which is the correct behavior for a helper that can
> > sleep.
> 
> That's a new feature. It's not a fix for the splat.
> Andrii's point in v1 is about build IDs. That's a good reason for this
> patch, but the splat needs Songsen's patch anyway:
> https://lore.kernel.org/bpf/[email protected]/
> 
>                                         Songsen's patch  this patch
>   ------------------------------------  ---------------  -------------
>   splat in sleepable progs              fixed            fixed
>   splat in bpf_wq, task_work callbacks  fixed            still there
>   build ID when the page is paged out   no               yes
>   progs that load today                 all load         some rejected
> 
> Row 2: the callback is sleepable, but its prog is not.
> 
>   bpf_wq_work()                    rcu_read_lock_trace() only
>     callback_fn()                  prog->sleepable == false
>       bpf_get_task_stack()         may_fault == false
>         __bpf_get_stack()          no rcu_read_lock()
>           get_callchain_entry()    rcu_dereference(): splat
> 
> Row 4: sleepable fentry, lsm, iter, syscall and tracepoint progs that
> call bpf_get_stack() under bpf_rcu_read_lock() or bpf_preempt_disable()
> load fine today. With this patch the verifier rejects them.
> It does so even when flags don't have BPF_F_USER_BUILD_ID.
> 
> Pls respin it as a feature, with a new subject and commit log
> and a selftest.

Thanks. I will add the selftest and respin this patch as a feature change.


Reply via email to