Fwd: Problem with the kernel 4.15 - cutting the band (tc)

Fwd: Problem with the kernel 4.15 - cutting the band (tc)

[Linus Torvalds]

Forwarding a report about what looks like a regression between 4.14 and 4.15.

New ENOSPC issue? I don't even knew where to start guessing where to look.

Help me, Davem-Wan Kenobi, you are my only hope.

(But adding netdev just in case somebody else goes "That's obviously Xyz")

              Linus

---------- Forwarded message ----------
From: Marcin Kabiesz <admin@hostcenter.eu>
Date: Thu, Apr 5, 2018 at 10:38 AM
Subject: Problem with the kernel 4.15 - cutting the band (tc)


Hello,
I have a problem with bandwidth cutting on kernel 4.15. On the version
up to 4.15, i.e. 4.14, this problem does not occur.

uname -a: Linux router 4.14.15 #1 SMP x86_64 Intel Xeon E3-1230 v6
command to reproduce:

tc qdisc add dev ifb0 root handle 1: htb r2q 2
tc class add dev ifb0 parent 1: classid 1:1 htb rate 10gbit ceil
10gbit quantum 16000
tc filter add dev ifb0 parent 1: prio 5 handle 1: protocol all u32 divisor 256
tc filter add dev ifb0 protocol all parent 1: prio 5 u32 ht 800::
match ip dst 0.0.0.0/0 hashkey mask 0x000000ff at 16 link 1:
tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
tc filter del dev ifb0 parent 1:0 handle 1:2c:1 prio 5 u32
tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
tc filter del dev ifb0 parent 1:0 handle 1:2c:1 prio 5 u32

This ok, no error/warnings and dmesg log.

uname -a: Linux router 4.15.8 #1 SMP x86_64 Intel Xeon E3-1230 v6 (or
4.15.14 this same effect)
command to reproduce:

tc qdisc add dev ifb0 root handle 1: htb r2q 2
tc class add dev ifb0 parent 1: classid 1:1 htb rate 10gbit ceil
10gbit quantum 16000
tc filter add dev ifb0 parent 1: prio 5 handle 1: protocol all u32 divisor 256
tc filter add dev ifb0 protocol all parent 1: prio 5 u32 ht 800::
match ip dst 0.0.0.0/0 hashkey mask 0x000000ff at 16 link 1:
tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
tc filter del dev ifb0 parent 1:0 handle 1:2c:1 prio 5 u32
tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
RTNETLINK answers: No space left on device
We have an error talking to the kernel

This not ok, on error/warnings and no dmesg log.

Best Regards
Please forgive my English
Marcin Kabiesz

Re: Problem with the kernel 4.15 - cutting the band (tc)

[Cong Wang]

On Fri, Apr 6, 2018 at 2:56 PM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
> Forwarding a report about what looks like a regression between 4.14 and 4.15.
>
> New ENOSPC issue? I don't even knew where to start guessing where to look.
>
> Help me, Davem-Wan Kenobi, you are my only hope.
>
> (But adding netdev just in case somebody else goes "That's obviously Xyz")
>
>               Linus
>
> ---------- Forwarded message ----------
> From: Marcin Kabiesz <admin@hostcenter.eu>
> Date: Thu, Apr 5, 2018 at 10:38 AM
> Subject: Problem with the kernel 4.15 - cutting the band (tc)
>
>
> Hello,
> I have a problem with bandwidth cutting on kernel 4.15. On the version
> up to 4.15, i.e. 4.14, this problem does not occur.
>
> uname -a: Linux router 4.14.15 #1 SMP x86_64 Intel Xeon E3-1230 v6
> command to reproduce:
>
> tc qdisc add dev ifb0 root handle 1: htb r2q 2
> tc class add dev ifb0 parent 1: classid 1:1 htb rate 10gbit ceil
> 10gbit quantum 16000
> tc filter add dev ifb0 parent 1: prio 5 handle 1: protocol all u32 divisor 256
> tc filter add dev ifb0 protocol all parent 1: prio 5 u32 ht 800::
> match ip dst 0.0.0.0/0 hashkey mask 0x000000ff at 16 link 1:
> tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
> ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
> tc filter del dev ifb0 parent 1:0 handle 1:2c:1 prio 5 u32
> tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
> ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
> tc filter del dev ifb0 parent 1:0 handle 1:2c:1 prio 5 u32
>
> This ok, no error/warnings and dmesg log.
>
> uname -a: Linux router 4.15.8 #1 SMP x86_64 Intel Xeon E3-1230 v6 (or
> 4.15.14 this same effect)
> command to reproduce:
>
> tc qdisc add dev ifb0 root handle 1: htb r2q 2
> tc class add dev ifb0 parent 1: classid 1:1 htb rate 10gbit ceil
> 10gbit quantum 16000
> tc filter add dev ifb0 parent 1: prio 5 handle 1: protocol all u32 divisor 256
> tc filter add dev ifb0 protocol all parent 1: prio 5 u32 ht 800::
> match ip dst 0.0.0.0/0 hashkey mask 0x000000ff at 16 link 1:
> tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
> ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
> tc filter del dev ifb0 parent 1:0 handle 1:2c:1 prio 5 u32
> tc filter add dev ifb0 parent 1:0 handle ::1 protocol all prio 5 u32
> ht 1:2c: match ip dst 192.168.3.44/32 flowid 1:2
> RTNETLINK answers: No space left on device
> We have an error talking to the kernel
>
> This not ok, on error/warnings and no dmesg log.

We forgot to call idr_remove() when deleting u32 key...

I am cooking a fix now.

Thanks!