On Fri, May 22, 2026 at 2:29 PM Jakub Kicinski [off-list ref] wrote:

On Fri, 22 May 2026 17:29:57 +0000 Eric Dumazet wrote:

quoted

Many shell scripts invoke iproute2 commands specifying a device by
its name.

This series improves their performance avoiding RTNL acquisition
for their (repeated) name->index conversion.

v3: insert patch 2/3 in the series (Jakub reported a KASAN splat)
v4: Addressed Sashiko's feedback.
    added 2 patches for rtnl_dump_ifinfo().

The CI looks fried, various errors:

# 0.02 [+0.02] RTNETLINK answers: File exists
# 0.02 [+0.00] Failed to create netif

# CMD: ip -d -j link show dev eth8
#   EXIT: 1
#   STDOUT: []
#   STDERR: RTNETLINK answers: Message too long
#           Cannot send link get request: Message too long

At boot we hit:

[    0.661578] ------------[ cut here ]------------
[    0.661608] WARNING: net/core/rtnetlink.c:4296 at rtnl_getlink+0x457/0x5e0, CPU#3: ip/71

Ah right, first patch went wrong, I missed that we call
nla_nest_cancel() when there was no altnames to dump.