Re: [PATCH 0/3] Fix ARP table parsing over netlink messages
On Fri, Aug 16, 2024 at 11:52:09PM -0400, Laine Stump wrote:
On 8/16/24 8:45 AM, Martin Kletzander wrote:
> Somehow it happened that some kernels (I noticed this with 6.10.0 and 6.10.2 on
> various machines) started sending NLMSG_DONE message (as they probably should've
> even before),
This comment shows the kernel commit that caused it:
https://bugzilla.redhat.com/2302245#c4
Oh, it would've been nice to know someone else found it as well =D My
bad, I guess.
This bug also points out that we're only reading a single
response
message, which could theoretically lead to missing the tail of a very
long arp table. But that's another battle for another day.
To be honest I did not try tracking down our usage/acceptance of
NLM_F_MULTI. It's also one of the reasons I suggested using a tiny bit
more of the high level abstraction from libnl3 where we can leave that
multipart messages and bit of error handling to it. More in 3/3.
> but our check for it could've never worked and now `virsh
> domifaddr --source arp` sometimes fails with "wrong nlmsg len". So I set
out on
> a quest to tame the netlink beast and though fierceful debugging and... Oh, I
> forgot, nobody reads cover letters. Never mind.
Oh come on!!! Just when it was getting interesting! Where is the gyring
and gimbling in the wabe?? The mome raths?? The mimsy borogroves????
Thank you for reading the commit message. You shall be known as Laine
the reader from now on ;)
(in case that makes no sense:
https://www.poetryfoundation.org/poems/42916/jabberwocky - my 3rd or 4th
favorite poem, after 2 or 3 by Dr. Seuss)
Sounds good, I just need to learn the language first =D
>
> Martin Kletzander (3):
> virarptable: Properly calculate rtattr length
> virarptable: Fix check for message length
> virarptable: End parsing earlier in case of NLMSG_DONE
>
> src/util/virarptable.c | 11 +++++------
> 1 file changed, 5 insertions(+), 6 deletions(-)
>
Attachments:
- signature.asc (application/pgp-signature — 833 bytes)