[PATCH v2] netlink: fix NLMSG_DONE never enter family specific decoder

[JingPiao Chen]

On Mon, Jun 26, 2017 at 06:08:40PM +0300, Dmitry V. Levin wrote:
> On Mon, Jun 26, 2017 at 10:59:11PM +0800, JingPiao Chen wrote:
> > On Mon, Jun 26, 2017 at 05:48:41PM +0300, Dmitry V. Levin wrote:
> > > On Mon, Jun 26, 2017 at 10:22:10PM +0800, JingPiao Chen wrote:
> > > > On Mon, Jun 26, 2017 at 05:03:41PM +0300, Dmitry V. Levin wrote:
> > > > > On Mon, Jun 26, 2017 at 10:29:40AM +0800, JingPiao Chen wrote:
> > > > > > * netlink.c (print_nlmsghdr): Add condition
> > > > > > nlmsghdr->nlmsg_type != NLMSG_DONE.
> > > > > > (decode_nlmsg_type): Add condition type != NLMSG_DONE.
> > > > > > (decode_nlmsg_flags): Add condition type == NLMSG_DONE.
> > > > > > (decode_netlink_sock_diag): Add condition
> > > > > > nlmsghdr->nlmsg_type == NLMSG_DONE.
> > > > > 
> > > > > Now I don't follow.
> > > > > 
> > > > > What do you want to do with NLMSG_DONE messages, pass them to family
> > > > > specific decoders or not?
> > > > 
> > > > Pass them to family specific decoders.
> > > > 
> > > > > Ditto with other control messages, do you want to pass them to family
> > > > > specific decoders or not?
> > > > 
> > > > Do not pass them to family specific decoders.
> > > 
> > > In that case ...
> > > 
> > > > --- a/netlink.c
> > > > +++ b/netlink.c
> > > > @@ -177,7 +177,7 @@ decode_nlmsg_type(const uint16_t type, const unsigned int family)
> > > >  	const struct xlat *xlat = netlink_types;
> > > >  	const char *dflt = "NLMSG_???";
> > > >  
> > > > -	if (family < ARRAY_SIZE(nlmsg_types)) {
> > > > +	if (type != NLMSG_DONE && family < ARRAY_SIZE(nlmsg_types)) {
> > > >  		if (nlmsg_types[family].decoder)
> > > >  			decoder = nlmsg_types[family].decoder;
> > > >  		if (nlmsg_types[family].xlat)
> > > 
> > > ... this hunk does something different from what you want because
> > > messages of type == NLMSG_DONE are not passed to the family specific
> > > decoder.
> > 
> > I do not pass them to family specific nlmsg_types_decoder_t,
> > only pass them to family specific netlink_decoder_t.
> > Is this idea not easy to understand?
> 
> This is not quite obvious.  Would you like to describe your intention
> in the commit message, please?

Ok, I want know what do you think.

On "[PATCH] netlink: decode NLMSG_DONE message" you wrote:
> I suggest implementing a default decoder of NLMSG_DONE messages that
> would print the integer in case of len == sizeof(int) and fall back
> to  printstrn for other lengths.
> 
> When protocol specific netlink decoders are added, they could either
> decode NLMSG_DONE themselves (and return true) or just return false to
> fall back to default decoder.

Are you know the code never enter family specific decoder
or I misunderstood your intention?

If want to pass type == NLMSG_DONE to family specific netlink_decoder,
must get family when type == NLMSG_DONE. I do not pass type == NLMSG_DONE
to family specific nlmsg_types_decoder because default decoder can handle it,
and if pass them to family specific nlmsg_types_decoder need each family
write a nlmsg_types_decoder.

commit message:

While many NLMSG_DONE messages indeed have payload containing just one
integer, there are exceptions. So should pass type == NLMSG_DONE to family
specific netlink decoder.

--
JingPiao Chen