On Tue, Feb 21, 2017 at 02:00:13PM +0100, Jesper Dangaard Brouer wrote:
> On Tue, 21 Feb 2017 00:06:11 -0800
> Alexei Starovoitov <alexei.starovoi...@gmail.com> wrote:
>
> > On Mon, Feb 20, 2017 at 05:25:58PM +0100, Jesper Dangaard Brouer wrote:
> > > On Mon, 20 Feb 2017 16:57:34 +0100
> > > Daniel Borkmann <dan...@iogearbox.net> wrote:
> > >
> > > > On 02/20/2017 04:35 PM, Jesper Dangaard Brouer wrote:
> > > > > It is confusing users of samples/bpf that exceeding the resource
> > > > > limits for RLIMIT_MEMLOCK result in an "Operation not permitted"
> > > > > message. This is due to bpf limits check return -EPERM.
> > > > >
> > > > > Instead return -ENOMEM, like most other users of this API.
> > > > >
> > > > > Fixes: aaac3ba95e4c ("bpf: charge user for creation of BPF maps and
> > > > > programs")
> > > > > Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements")
> > > > > Fixes: 5ccb071e97fb ("bpf: fix overflow in prog accounting")
> > > >
> > > > Btw, last one just moves the helper so fixes doesn't really apply
> > > > there, but apart from that this is already uapi exposed behavior
> > > > like this for ~1.5yrs, so unfortunately too late to change now. I
> > > > think the original intention (arguably confusing in this context)
> > > > was that user doesn't have (rlimit) permission to allocate this
> > > > resource.
> > >
> > > This is obviously confusing end-users, thus it should be fixed IMHO.
> >
> > I don't think it's confusing and I think EPERM makes
> > the most sense as return code in such situation.
>
> Most other kernel users return ENOMEM.
the places in the kernel that check for memlock are not fully consistent.
perf does this:
lock_limit = rlimit(RLIMIT_MEMLOCK);
lock_limit >>= PAGE_SHIFT;
locked = vma->vm_mm->pinned_vm + extra;
if ((locked > lock_limit) && perf_paranoid_tracepoint_raw() &&
!capable(CAP_IPC_LOCK)) {
ret = -EPERM;
goto unlock;
}
and in bpf land we got an idea of using memlock limit from perf.
> Documented it here:
> https://prototype-kernel.readthedocs.io/en/latest/bpf/troubleshooting.html
Thanks! Looks good.
