Re: [PATCH net-next v2 3/3] genetlink: fit NLMSG_DONE into same read() as families

[Jakub Kicinski <kuba@kernel.org>]

On Fri, 15 Mar 2024 12:48:08 +0100 Stefano Brivio wrote:
> > Make sure ctrl_fill_info() returns sensible error codes and
> > propagate them out to netlink core. Let netlink core decide
> > when to return skb->len and when to treat the exit as an
> > error. Netlink core does better job at it, if we always
> > return skb->len the core doesn't know when we're done
> > dumping and NLMSG_DONE ends up in a separate read().  
> 
> While this change is obviously correct, it breaks... well, broken
> applications that _wrongly_ rely on the fact that NLMSG_DONE is
> delivered in a separate datagram.
> 
> This was the (embarrassing) case for passt(1), which I just fixed:
>   https://archives.passt.top/passt-dev/20240315112432.382212-1-sbrivio@redhat.com/
> 
> but the "separate" NLMSG_DONE is such an established behaviour,
> I think, that this might raise a more general concern.
> 
> From my perspective, I'm just happy that this change revealed the
> issue, but I wanted to report this anyway in case somebody has
> similar possible breakages in mind.

Hi Stefano! I was worried this may happen :( I think we should revert
offending commits, but I'd like to take it on case by case basis. 
I'd imagine majority of netlink is only exercised by iproute2 and
libmnl-based tools. Does passt hang specifically on genetlink family
dump? Your commit also mentions RTM_GETROUTE. This is not the only
commit which removed DONE:

$ git log --since='1 month ago' --grep=NLMSG_DONE --no-merges  --oneline 

9cc4cc329d30 ipv6: use xa_array iterator to implement inet6_dump_addr()
87d381973e49 genetlink: fit NLMSG_DONE into same read() as families
4ce5dc9316de inet: switch inet_dump_fib() to RCU protection
6647b338fc5c netlink: fix netlink_diag_dump() return value