net sched: cleanup and rate limit warning
Subsystems (3)
PATCH 1/1 net sched: cleanup and rate limit warning
If the user has a bad classification configuration, and gets a packet that goes through too many steps. Chances are more packets will arrive, and the message spew will overrun syslog because it is not rate limited. And because it is not tagged with appropriate priority it can't not be screened.
Added the qdisc to the message to try and give some more context when the message does arrive.
---
Please think about this for 2.6.34 and could even be -stable material.
net/sched/sch_api.c +6 -3
@@ -1637,9 +1638,12 @@ reclassify: tp = otp;if (verd++ >=MAX_REC_LOOP ) { -printk ("rule prio %u protocol %02x reclassify loop, " -"packet dropped\n" , - tp->prio&0xffff ,ntohs (tp->protocol)); +if (net_ratelimit ()) +printk (KERN_NOTICE +"%s: packet reclassify loop" +" rule prio %u protocol %02x\n" , + tp->q->ops->id, + tp->prio &0xffff ,ntohs (tp->protocol));return TC_ACT_SHOT ; } skb->tc_verd =SET_TC_VERD (skb->tc_verd, verd);
Quoted 2 lines · Stephen Hemminger
If the user has a bad classification configuration, and gets a packet that goes through too many steps.
Can you pass me the setup which caused this to be hit?
cheers,
jamal
Quoted 8 lines · On Wed, 12 May 2010 13:17:54 -0400 jamal <[email protected]> wrote:
Quoted 2 lines · Stephen Hemminger
If the user has a bad classification configuration, and gets a packet that goes through too many steps.
Can you pass me the setup which caused this to be hit?
cheers,
jamal
The Vyatta syntax is:
traffic-limiter test-traffic-limit {
class 2048 {
bandwidth 1mbit
burst 500kbit
match onebox {
ip {
destination {
address 192.168.100.99/32
}
}
}
}}
Which generates these TC commands.
root@VC6:~# tc qdisc show dev eth0 qdisc pfifo_fast 0: root bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc ingress ffff: parent ffff:fff1 ----------------
root@VC6:~# tc filter show dev eth0 parent ffff: filter protocol all pref 20 u32 filter protocol all pref 20 u32 fh 800: ht divisor 1 filter protocol all pref 20 u32 fh 800::800 order 2048 key ht 800 bkt 0 flowid ffff:800 match c0a86463/ffffffff at 16 police 0x3 rate 1000Kbit burst 63999b mtu 2Kb action reclassify overhead 0b ref 1 bind 1
I think the bad part is the huge burst size.
Thanks for the info Stephen.
Quoted 14 lines · Stephen Hemminger
The Vyatta syntax is:
traffic-limiter test-traffic-limit { class 2048 { bandwidth 1mbit burst 500kbit match onebox { ip { destination { address 192.168.100.99/32 } } } }
Quoted 2 lines
}
;-> I guess kids these days prefer juniperism over ciscoism? Why dont they just learn linuxism?;->
Quoted 15 lines
Which generates these TC commands.
root@VC6:~# tc qdisc show dev eth0 qdisc pfifo_fast 0: root bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc ingress ffff: parent ffff:fff1 ----------------root@VC6:~# tc filter show dev eth0 parent ffff: filter protocol all pref 20 u32 filter protocol all pref 20 u32 fh 800: ht divisor 1 filter protocol all pref 20 u32 fh 800::800 order 2048 key ht 800 bkt 0 flowid ffff:800 match c0a86463/ffffffff at 16 police 0x3 rate 1000Kbit burst 63999b mtu 2Kb action reclassify overhead 0b ref 1 bind 1
Quoted 1 line
I think the bad part is the huge burst size.
That may be - but it seems your tool is the culprit.
It is generating wrong tc commands if i read the intent correctly. Basically what the tc command is saying is "if you exceed the 1Mbit upto a burst of 500kbit then reclassify".
"Reclassify" means literally that: to reuse the same classification rule again, which will find that we have exceeded 1M which will ask reclassify .... loop.... I am glad that code is there ;->
cheers,
jamal
Quoted 41 lines · On Wed, 12 May 2010 15:13:48 -0400 jamal <[email protected]> wrote:
Thanks for the info Stephen.
Quoted 14 lines · Stephen Hemminger
The Vyatta syntax is:
traffic-limiter test-traffic-limit { class 2048 { bandwidth 1mbit burst 500kbit match onebox { ip { destination { address 192.168.100.99/32 } } } }Quoted 2 lines
}
;-> I guess kids these days prefer juniperism over ciscoism? Why dont they just learn linuxism?;->
Quoted 15 lines
Which generates these TC commands.
root@VC6:~# tc qdisc show dev eth0 qdisc pfifo_fast 0: root bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc ingress ffff: parent ffff:fff1 ----------------root@VC6:~# tc filter show dev eth0 parent ffff: filter protocol all pref 20 u32 filter protocol all pref 20 u32 fh 800: ht divisor 1 filter protocol all pref 20 u32 fh 800::800 order 2048 key ht 800 bkt 0 flowid ffff:800 match c0a86463/ffffffff at 16 police 0x3 rate 1000Kbit burst 63999b mtu 2Kb action reclassify overhead 0b ref 1 bind 1Quoted 1 line
I think the bad part is the huge burst size.
Quoted 7 lines
That may be - but it seems your tool is the culprit.
It is generating wrong tc commands if i read the intent correctly. Basically what the tc command is saying is "if you exceed the 1Mbit upto a burst of 500kbit then reclassify".
"Reclassify" means literally that: to reuse the same classification rule again, which will find that we have exceeded 1M which will ask reclassify .... loop.... I am glad that code is there ;->
The tool isn't generating an action (just tc filter ... police ..) so it is getting the unfortunate default of reclassify.
Quoted 2 lines · Stephen Hemminger
The tool isn't generating an action (just tc filter ... police ..) so it is getting the unfortunate default of reclassify.
Ah ok.
My advice: you should never ever depend on defaults when you can be explicit and say "drop". Or have the users in your tool be able to specify what action to take if rate is exceeded etc (actually i think juniper does that)
I think "drop" would be the sane default for over-limit - my memory is hazy because i assumed that was the default but there may have been some reservations on that default. Patrick?
cheers,
jamal
Quoted 5 lines · jamal
Quoted 2 lines · Stephen Hemminger
The tool isn't generating an action (just tc filter ... police ..) so it is getting the unfortunate default of reclassify.
Quoted 8 lines
Ah ok.
My advice: you should never ever depend on defaults when you can be explicit and say "drop". Or have the users in your tool be able to specify what action to take if rate is exceeded etc (actually i think juniper does that)
I think "drop" would be the sane default for over-limit - my memory is hazy because i assumed that was the default but there may have been some reservations on that default. Patrick?
I don't remeber ever discussing that, the choice of "reclassify" as default precedes TC actions and is already present in the oldest iproute2 version I could find (2.2.4-ss000225).
Quoted 3 lines · Patrick McHardy
I don't remeber ever discussing that, the choice of "reclassify" as default precedes TC actions and is already present in the oldest iproute2 version I could find (2.2.4-ss000225).
I think i confused it with unrelated discussion.
If its been there from day 0 then we should leave it as is - rationale:
it will help to catch config bugs.
Stephen needs to fix his tool regardless...
cheers,
jamal
Quoted 6 lines · On Thu, 13 May 2010 12:20:26 -0400 jamal <[email protected]> wrote:
Quoted 3 lines · Patrick McHardy
I don't remeber ever discussing that, the choice of "reclassify" as default precedes TC actions and is already present in the oldest iproute2 version I could find (2.2.4-ss000225).
Quoted 9 lines
I think i confused it with unrelated discussion.
If its been there from day 0 then we should leave it as is - rationale:
it will help to catch config bugs.
Stephen needs to fix his tool regardless...cheers,
jamal
And the kernel message needs to be fixed to prevent total log overload for the next poor sop who makes the same mistake.
Please accept the patch.
jamal
Acked-byQuoted 6 lines · Stephen Hemminger
Quoted 1 line
And the kernel message needs to be fixed to prevent total log overload for the next poor sop who makes the same mistake.
Please accept the patch.
I have no problem with the patch. Dont think it needs to go to stable but i'll let Dave call it.
cheers,
jamal
From: jamal <[email protected]>
Date: Thu, 13 May 2010 12:40:08 -0400
Quoted 7 lines · Stephen Hemminger
Quoted 6 lines
Quoted 2 lines
And the kernel message needs to be fixed to prevent total log overload for the next poor sop who makes the same mistake.
Quoted 6 lines
Quoted 1 line
Please accept the patch.
I have no problem with the patch. Dont think it needs to go to stable but i'll let Dave call it.
Acked-by Jamal Hadi Salim <[email protected]>
Applied, I don't think it's -stable material personally. If we start queueing stuff like this, it will be 200 instead of 100 patches in each -stable release :-)