On Wed, Apr 26, 2017 at 08:35:44AM -0400, Greg Wooledge wrote:
On Wed, Apr 26, 2017 at 01:25:18PM +0100, Darac Marjal wrote:
For interface statistics from ip, try "ip -s link [interface]".
On stretch:
wooledg:~$ ip -s link eth0
Command "eth0" is unknown, try "ip link help".
wooledg:~$ ip -s link dev eth0
Command "dev" is unknown, try "ip link help".
wooledg:~$ ip link help
[... enormous BNF dump, entirely missing -s, or any reference whatsoever
to the fact that you can stick options between "ip" and "link" ...]
wooledg:~$ ip -s link show eth0
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
state UP mode DEFAULT group default qlen 1000
link/ether a0:8c:fd:c3:89:e0 brd ff:ff:ff:ff:ff:ff
RX: bytes packets errors dropped overrun mcast
380719013 1442490 0 0 0 4731
TX: bytes packets errors dropped carrier collsns
57971257 614586 0 0 0 0
Aha!
My bad. I actually only got as far as discovering "ip -s link" on my own
system. As I was typing up the email I remembered that Richard was after
statistics for a specific interface. I should have been more diligent in
working out the correct format.
(Sadly, this is my *typical* experience with the ip command -- trying
random things until one of them works, because the documentation
is impenetrable, and the syntax barely guessable, and certainly not
predictable.)
ip *could* do a lot better, it's true. As a monolithic tool, there's not
really much excuse for the different sub-tools to parse the commands
differently. As you say, "ip address" expects the device to be expressed
as "dev eth0", so why doesn't "ip link" handle it the same way? I don't
know.