CBQ(but probably u32 filter bug), kernel "freeze", at least from 3.2.0

Hi

Got information from friend, confirmed that it crashed at least two my 
boxes also :)
3.0.5-rc1 is working fine, 3.4.1 , 3.2.0 from ubuntu  - crashing
No watchdog fired, and didn't got yet significant debugging 
information.

Very easy to reproduce:
1)run the script
2)ping 192.168.3.234

script:
DEV_OUT=eth0
ICMP="match ip protocol 1 0xff"
U32="protocol ip u32"
DST="match ip dst"
tc qdisc add dev $DEV_OUT root handle 1: cbq avpkt 1000 bandwidth 
100mbit
tc class add dev $DEV_OUT parent 1: classid 1:1 cbq rate 512kbit allot 
1500 prio 5 bounded isolated
tc filter add dev $DEV_OUT parent 1:              prio 3 $U32 $ICMP 
$DST 192.168.3.234 flowid 1:
tc qdisc add dev $DEV_OUT parent 1:1 sfq perturb 10

---
Denys Fedoryshchenko, Network Engineer, Virtual ISP S.A.L.

On Tue, 2012-08-28 at 07:50 +0300, Denys Fedoryshchenko wrote:

Hi

Got information from friend, confirmed that it crashed at least two my 
boxes also :)
3.0.5-rc1 is working fine, 3.4.1 , 3.2.0 from ubuntu  - crashing
No watchdog fired, and didn't got yet significant debugging 
information.

Very easy to reproduce:
1)run the script
2)ping 192.168.3.234

script:
DEV_OUT=eth0
ICMP="match ip protocol 1 0xff"
U32="protocol ip u32"
DST="match ip dst"
tc qdisc add dev $DEV_OUT root handle 1: cbq avpkt 1000 bandwidth 
100mbit
tc class add dev $DEV_OUT parent 1: classid 1:1 cbq rate 512kbit allot 
1500 prio 5 bounded isolated
tc filter add dev $DEV_OUT parent 1:              prio 3 $U32 $ICMP 
$DST 192.168.3.234 flowid 1:
tc qdisc add dev $DEV_OUT parent 1:1 sfq perturb 10

Not sure what your friend expected from this buggy configuration.

It probably never worked at all.

CBQ needs at least one child class and one leaf class.

This scripts creates a loop inside CBQ, so cpu is probably looping in
cbq_enqueue() (or more exactly cbq_classify()), as instructed by the
sysadmin ;)

u32 (or sfq) seems ok.

Could you try the following patch ?

Thanks !

diff --git a/net/sched/sch_cbq.c b/net/sched/sch_cbq.c
index 6aabd77..564b9fc 100644
--- a/net/sched/sch_cbq.c
+++ b/net/sched/sch_cbq.c

@@ -250,10 +250,11 @@ cbq_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
 			else if ((cl = defmap[res.classid & TC_PRIO_MAX]) == NULL)
 				cl = defmap[TC_PRIO_BESTEFFORT];
 
-			if (cl == NULL || cl->level >= head->level)
+			if (cl == NULL)
 				goto fallback;
 		}
-
+		if (cl->level >= head->level)
+			goto fallback;
 #ifdef CONFIG_NET_CLS_ACT
 		switch (result) {
 		case TC_ACT_QUEUED:

On Tue, 2012-08-28 at 03:59 -0700, Eric Dumazet wrote:

Hi Denys

Any feedback on the suggested patch ?

Thanks !

On 2012-09-06 11:40, Eric Dumazet wrote:

On Tue, 2012-08-28 at 03:59 -0700, Eric Dumazet wrote:

quoted

On Tue, 2012-08-28 at 07:50 +0300, Denys Fedoryshchenko wrote:

quoted

Hi

Got information from friend, confirmed that it crashed at least 

two my

quoted

boxes also :)
3.0.5-rc1 is working fine, 3.4.1 , 3.2.0 from ubuntu  - crashing
No watchdog fired, and didn't got yet significant debugging
information.

Very easy to reproduce:
1)run the script
2)ping 192.168.3.234

script:
DEV_OUT=eth0
ICMP="match ip protocol 1 0xff"
U32="protocol ip u32"
DST="match ip dst"
tc qdisc add dev $DEV_OUT root handle 1: cbq avpkt 1000 bandwidth
100mbit
tc class add dev $DEV_OUT parent 1: classid 1:1 cbq rate 512kbit 

allot

quoted

1500 prio 5 bounded isolated
tc filter add dev $DEV_OUT parent 1:              prio 3 $U32 

$ICMP

quoted

$DST 192.168.3.234 flowid 1:
tc qdisc add dev $DEV_OUT parent 1:1 sfq perturb 10

Not sure what your friend expected from this buggy configuration.

It probably never worked at all.

CBQ needs at least one child class and one leaf class.

This scripts creates a loop inside CBQ, so cpu is probably looping 
in
cbq_enqueue() (or more exactly cbq_classify()), as instructed by the
sysadmin ;)

u32 (or sfq) seems ok.

Could you try the following patch ?

Thanks !

diff --git a/net/sched/sch_cbq.c b/net/sched/sch_cbq.c
index 6aabd77..564b9fc 100644
--- a/net/sched/sch_cbq.c
+++ b/net/sched/sch_cbq.c

@@ -250,10 +250,11 @@ cbq_classify(struct sk_buff *skb, struct Qdisc 

*sch, int *qerr)
 			else if ((cl = defmap[res.classid & TC_PRIO_MAX]) == NULL)
 				cl = defmap[TC_PRIO_BESTEFFORT];

-			if (cl == NULL || cl->level >= head->level)
+			if (cl == NULL)
 				goto fallback;
 		}
-
+		if (cl->level >= head->level)
+			goto fallback;
 #ifdef CONFIG_NET_CLS_ACT
 		switch (result) {
 		case TC_ACT_QUEUED:

Hi Denys

Any feedback on the suggested patch ?

Thanks !

Dear Eric

Very sorry for delay, most of time in desert, without decent internet.
I will try to test today or tomorrow.

---
Denys Fedoryshchenko, Network Engineer, Virtual ISP S.A.L.

On Thu, 2012-09-06 at 16:47 +0300, Denys Fedoryshchenko wrote:

Dear Eric

Very sorry for delay, most of time in desert, without decent internet.
I will try to test today or tomorrow.

No problem, I reproduced the bug on my dev machine, but its always
better to have bug reporter adding its own 'Tested-by:' tag ;)

Thanks