Re: [PATCH 1/4] xfrm: increment genid before bumping state genids

["Timo Teräs" <timo.teras@iki.fi>]

Timo Teräs wrote:
> Herbert Xu wrote:
>> On Wed, Mar 31, 2010 at 01:17:03PM +0300, Timo Teras wrote:
>>> __xfrm_state_bump_genids() is used to update the genid of all
>>> matching xfrm_state's, so any bundle using the state would get
>>> refreshed with the newly inserted state.
>>>
>>> However, since __xfrm_state_bump_genids() is called before the
>>> __xfrm_state_insert() which actually bumps the genid counter,
>>> it is possible that the genid was not updated at all (if there
>>> was no state inserts previously).
>>>
>>> This is fixed by moving the genid incrementation to
>>> __xfrm_state_bump_genids() so the older states are guaranteed
>>> to get different genid.
>>>
>>> Signed-off-by: Timo Teras <timo.teras@iki.fi>
>>
>> It would appear that not all xfrm_state_insert calls are preceded
>> by xfrm_state_bump_genids so this patch isn't correct.
> 
> Yes, I noticed that there's one. But __xfrm_state_insert() is
> called to replace an acquire in that case. Acquires are never
> used in bundle, so this is good.
> 
> But maybe it'd be more explicit if the genid increment is done
> before each __xfrm_state_insert()?

Actually. Search for xfrm_state_genid in xfrm_state.c. It's only
used in two places: __xfrm_state_bumb_genid() and _insert().

The only case when it needs to get incremented is in the bump
function. If we are adding a state for the first time, there's
no need to bump the genid as it's per-matching state. The actual
work to invalidate states is done in the bump function, so it's
the only place that needs to increment it.

If any other xfrm_state_insert place needs to invalidate old
states it needs an additional bumping call. So the bumping function
is the right place to increment the genid.