lkml.org 
[lkml]   [2015]   [Oct]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH] xfrm: dst_entries_init() per-net dst_ops
On Wed, Oct 28, 2015 at 9:42 AM, Hannes Frederic Sowa
<hannes@stressinduktion.org> wrote:
> Hello,
>
> On Wed, Oct 28, 2015, at 14:32, Dan Streetman wrote:
>> On Tue, Oct 27, 2015 at 12:15 PM, <dan.streetman@canonical.com> wrote:
>> > From: Dan Streetman <dan.streetman@canonical.com>
>> >
>> > The ipv4 and ipv6 xfrms each create a template dst_ops object, and
>> > perform dst_entries_init() on the template objects. Then each net
>> > namespace has its net.xfrm.xfrm[46]_dst_ops field set to the template
>> > values. The problem with that is the dst_ops.pcpuc_entries field is
>> > a percpu counter and cannot be used correctly by simply copying it to
>> > another object.
>
> How hard would it be to split of the counters from the dst_ops struct?
> We could make dst_ops instances const and have normal pointers to them
> and keep the dst_entries as a small array in net namespace?

Well, the dst_ops->pcpuc_entries counter is used in dst.c which just
gets a struct dst_ops *, so it doesn't have access to the owning net
namespace. And, not all dst_ops users have a per-net-namespace
dst_ops; ipv4/route.c for example uses a global "ipv4_dst_ops" object.
So it probably does need to stay owned by dst_ops.


>
> Bye,
> Hannes


\
 
 \ /
  Last update: 2015-10-28 16:41    [W:0.079 / U:0.700 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site