Re: [PATCH bpf-next v2 1/2] bpf: implement bpf_send_signal_remote() kfunc

[Puranjay Mohan]

Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> writes:

> On Thu, Sep 26, 2024 at 4:53 AM Puranjay Mohan <puranjay@xxxxxxxxxx> wrote:
>>
>> Implement bpf_send_signal_remote kfunc that is similar to
>> bpf_send_signal_thread and bpf_send_signal helpers  but can be used to
>> send signals to other threads and processes. It also supports sending a
>> cookie with the signal similar to sigqueue().
>>
>> If the receiving process establishes a handler for the signal using the
>> SA_SIGINFO flag to sigaction(), then it can obtain this cookie via the
>> si_value field of the siginfo_t structure passed as the second argument
>> to the handler.
>>
>> Signed-off-by: Puranjay Mohan <puranjay@xxxxxxxxxx>
>> ---
>>  kernel/trace/bpf_trace.c | 78 +++++++++++++++++++++++++++++++++++++++-
>>  1 file changed, 77 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
>> index a582cd25ca876..51b27db1321fc 100644
>> --- a/kernel/trace/bpf_trace.c
>> +++ b/kernel/trace/bpf_trace.c
>> @@ -802,6 +802,9 @@ struct send_signal_irq_work {
>>         struct task_struct *task;
>>         u32 sig;
>>         enum pid_type type;
>> +       bool is_siginfo;
>> +       kernel_siginfo_t info;
>> +       int value;
>>  };
>>
>>  static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
>> @@ -811,7 +814,11 @@ static void do_bpf_send_signal(struct irq_work *entry)
>>         struct send_signal_irq_work *work;
>>
>>         work = container_of(entry, struct send_signal_irq_work, irq_work);
>> -       group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
>> +       if (work->is_siginfo)
>> +               group_send_sig_info(work->sig, &work->info, work->task, work->type);
>> +       else
>> +               group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
>> +
>>         put_task_struct(work->task);
>>  }
>>
>> @@ -848,6 +855,7 @@ static int bpf_send_signal_common(u32 sig, enum pid_type type)
>>                  * irq works get executed.
>>                  */
>>                 work->task = get_task_struct(current);
>> +               work->is_siginfo = false;
>>                 work->sig = sig;
>>                 work->type = type;
>>                 irq_work_queue(&work->irq_work);
>> @@ -3484,3 +3492,71 @@ static int __init bpf_kprobe_multi_kfuncs_init(void)
>>  }
>>
>>  late_initcall(bpf_kprobe_multi_kfuncs_init);
>> +
>> +__bpf_kfunc_start_defs();
>> +
>> +__bpf_kfunc int bpf_send_signal_remote(struct task_struct *task, int sig, enum pid_type type,
>> +                                      int value)
>> +{
>> +       struct send_signal_irq_work *work = NULL;
>> +       kernel_siginfo_t info;
>> +
>> +       if (type != PIDTYPE_PID && type != PIDTYPE_TGID)
>> +               return -EINVAL;
>> +       if (unlikely(task->flags & (PF_KTHREAD | PF_EXITING)))
>> +               return -EPERM;
>> +       if (unlikely(!nmi_uaccess_okay()))
>> +               return -EPERM;
>> +       /* Task should not be pid=1 to avoid kernel panic. */
>> +       if (unlikely(is_global_init(task)))
>> +               return -EPERM;
>> +
>> +       clear_siginfo(&info);
>> +       info.si_signo = sig;
>> +       info.si_errno = 0;
>> +       info.si_code = SI_KERNEL;
>> +       info.si_pid = 0;
>> +       info.si_uid = 0;
>> +       info.si_value.sival_int = value;
>
> It seems like it could be either int sival_int or `void *sival_ptr`,
> i.e., it's actually a 64-bit value on 64-bit architectures.
>
> Can we allow passing a full u64 here and assign it to sival_ptr (with a cast)?

Yes, I initially thought of allowing the kfunc to take the union itself
but turns out unions are not supported, so I will just use a cast to put
the value in sival_ptr.

>> +
>> +       if (irqs_disabled()) {
>> +               /* Do an early check on signal validity. Otherwise,
>> +                * the error is lost in deferred irq_work.
>> +                */
>> +               if (unlikely(!valid_signal(sig)))
>> +                       return -EINVAL;
>> +
>> +               work = this_cpu_ptr(&send_signal_work);
>> +               if (irq_work_is_busy(&work->irq_work))
>> +                       return -EBUSY;
>> +
>> +               work->task = get_task_struct(task);
>> +               work->is_siginfo = true;
>> +               work->info = info;
>> +               work->sig = sig;
>> +               work->type = type;
>> +               work->value = value;
>> +               irq_work_queue(&work->irq_work);
>> +               return 0;
>> +       }
>> +
>> +       return group_send_sig_info(sig, &info, task, type);
>> +}
>> +
>> +__bpf_kfunc_end_defs();
>> +
>> +BTF_KFUNCS_START(send_signal_kfunc_ids)
>> +BTF_ID_FLAGS(func, bpf_send_signal_remote, KF_TRUSTED_ARGS)
>> +BTF_KFUNCS_END(send_signal_kfunc_ids)
>> +
>> +static const struct btf_kfunc_id_set bpf_send_signal_kfunc_set = {
>> +       .owner = THIS_MODULE,
>> +       .set = &send_signal_kfunc_ids,
>> +};
>> +
>> +static int __init bpf_send_signal_kfuncs_init(void)
>> +{
>> +       return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_send_signal_kfunc_set);
>
> let's allow it for other program types (at least kprobes, tracepoints,
> raw_tp, etc, etc)? Is there any problem just allowing it for any
> program type?

I guess we can allow it for all program types.

>
>> +}
>> +
>> +late_initcall(bpf_send_signal_kfuncs_init);
>> --
>> 2.40.1
>>

Attachment: signature.asc
Description: PGP signature