Re: [RFC iproute2-next] ss: remove /proc fallback paths

[Stephen Hemminger <stephen@networkplumber.org>]

On Sun, 18 Oct 2026 20:07:12 +0000
Gopal Malaviya <gopalmalaviya53@gmail.com> wrote:

> On Tue, 1 Sep 2026 14:38 UTC, Stephen Hemminger wrote:
> > This patch removes the old /proc/net fallback code in ss command.
> > The sock_diag netlink interfaces were all in 3.10, over 13 years ago.
> > No longer need to carry all the additional text parsing code.  
> 
> I tested this on top of iproute2-next (204a7210). It builds with no new
> warnings, testsuite/tests/ss/ssfilter.t passes, and on a WSL2 6.18
> kernel the output of ss -tan, -uan, -xan, -s and -tanpie is identical
> with and without the patch (ss -a -A netlink too, apart from the line
> for ss's own netlink socket).
> 
> One concern: the diag interfaces are in the kernel, but each one is a
> separate Kconfig option. INET_DIAG defaults to y, while UNIX_DIAG,
> PACKET_DIAG, NETLINK_DIAG, INET_UDP_DIAG and INET_RAW_DIAG are all
> "default n", and common configs leave some of them off:
> x86_64_defconfig has "# CONFIG_INET_DIAG is not set", arm64 defconfig
> sets none of the five, and Android's gki_defconfig enables only
> INET_UDP_DIAG among them.
> 
> When the interface for a socket type is missing, ss currently falls
> back to /proc/net/*. With this patch it prints only the header and
> exits 0, so those sockets silently disappear. I reproduced this for
> raw, unix and packet sockets (ss_before/ss_after are ss built
> without/with this patch). Raw sockets, on a kernel without
> INET_RAW_DIAG (the stock WSL2 kernel), with one ICMP and one ICMPv6 raw
> socket open:


I am ok with forcing use of inet_diag for this.
If users turn it off, like Android, they are in embedded envrionments
where ss doesn't make lot of sense.