[PATCH RESEND 1/8] netlink: call get_fd_nl_family before decode nlmsghdr

[Dmitry V. Levin]

On Tue, Aug 08, 2017 at 03:07:40PM +0300, Dmitry V. Levin wrote:
> On Sun, Aug 06, 2017 at 07:52:17AM +0800, JingPiao Chen wrote:
> > On Sun, Aug 06, 2017 at 02:10:16AM +0300, Dmitry V. Levin wrote:
> > > On Tue, Aug 01, 2017 at 07:48:41AM +0800, JingPiao Chen wrote:
> > > > Prepare for NETLINK_KOBJECT_UEVENT decode. The messages
> > > > of NETLINK_KOBJECT_UEVENT do not contain nlmsghdr.
> > > [...]
> > > > (decode_nlmsghdr_with_payload): Skip family specific decoders
> > > > for type < NLMSG_MIN_TYPE && type != NLMSG_DONE.
> > > [...]
> > > > @@ -182,7 +177,7 @@ decode_nlmsg_type(const uint16_t type, const unsigned int family)
> > > >  	const struct xlat *xlat = netlink_types;
> > > >  	const char *dflt = "NLMSG_???";
> > > >  
> > > > -	if (type != NLMSG_DONE && family < ARRAY_SIZE(nlmsg_types)) {
> > > > +	if (type >= NLMSG_MIN_TYPE && family < ARRAY_SIZE(nlmsg_types)) {
> > > 
> > > Why?  How this is related to the rest of NETLINK_KOBJECT_UEVENT change?
> > 
> > Previous code when type < NLMSG_MIN_TYPE && type != NLMSG_DONE,
> > family = -2 (NL_FAMILY_DEFAULT), family < ARRAY_SIZE(nlmsg_types) filter
> > this case. Now get family ignore the nlmsg_type.
> > When type < NLMSG_MIN_TYPE && type != NLMSG_DONE, family is not
> > a negative. family < ARRAY_SIZE(nlmsg_types) can not filter this case.
> > 
> > Related commit:
> > v4.17-43-g1b63425, v4.17-44-g8700030: These commit introduce get_fd_nl_family.
> > And get family only when type >= NLMSG_MIN_TYPE.
> > 
> > v4.17-123-g54aed21: This commit changed get family when
> > type >= NLMSG_MIN_TYPE || type == NLMSG_DONE.
> 
> Now that you change things to call get_fd_nl_family from decode_netlink
> unconditionally, family is available unconditionally, too (although it
> can be -1).  This is all clear enough but doesn't answer my question:
> 
> Why do you think it means that decode_nlmsg_type should not use
> family-specific decoders for message type < NLMSG_MIN_TYPE?
> Is decoding of these message types 100% family-agnostic?
> 
> It's so far from being obvious that if it's true, there must be a comment
> explaining why it's true.

The same issue is with message type checks in decode_payload:
there has to be a comment explaining why family-specific netlink decoders
are not invoked for (nlmsg_type < 0 && nlmsg_type != NLMSG_DONE).


-- 
ldv
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: not available
URL: <http://lists.strace.io/pipermail/strace-devel/attachments/20170808/22c5cb93/attachment.bin>