Re: [patch] trace: Add user-space event tracing/injection

[Peter Zijlstra <peterz@infradead.org>]

On Wed, 2010-11-17 at 13:29 +0100, Peter Zijlstra wrote:
> > +             case PR_TASK_PERF_USER_TRACE:
> > +             {
> > +                     void __user *uevent_ptr = (void *)arg2;
> > +                     char kstring[MAX_USER_TRACE_SIZE+1];
> > +                     unsigned long uevent_len = arg3;
> > +
> > +                     if (uevent_len > MAX_USER_TRACE_SIZE)
> > +                             return -EINVAL;
> > +                     if (copy_from_user(kstring, uevent_ptr, uevent_len))
> > +                             return -EFAULT;
> > +                     kstring[uevent_len] = 0;
> > +
> > +                     trace_user(kstring);
> > +                     return 0;
> > +             } 

That implementation can be improved upon, no need to actually do the
copy_from_user() when the tracepoint isn't enabled.

Also, do we really need a max size here? I know ftrace (and the perf
tracepoint interface) are currently limited to 1 page of trace data but
that can be fixed (although not trivially).

Also, -EINVAL isn't really a helpful error code in that case, -EMSGSIZE
seems like a good candidate.

And the uevent name is already taken for something else, re-using it
here might cause confusion.