Skip to contentSkip to conversation

QoS and DSCP

PATCH In review Daniel Wagner 2 participants 8 messages Aug 19, 2008 mac80211

lore
Share & tools
lore
#0 lore raw

Hi,

I was playing a bit with DSCP and my wireless setup. I don't know what the expected result should be there I'd like to ask you here what is the expected result.

Server <---(ethernet)--> d-link AP <---(wireless)---> Desktop
192.168.100.1 192.168.202.198

On the server I have an icecast stream running on port 8000.

server# iptables -t mangle -A OUTPUT -p tcp --sport 8000 -j DSCP --set-dscp-class EF

The packets send from the AP to the desktop have the QoS Control Field Priority set to 5:

Frame 13 (1622 bytes on wire, 1622 bytes captured) PPI version 0, 84 bytes
IEEE 802.11 QoS Data, Flags: ......F.C
Type/Subtype: QoS Data (0x28)
Frame Control: 0x0288 (Normal)
Duration: 48
Destination address: IntelCor_9e:db:5d (00:1d:e0:9e:db:5d) BSS Id: D-Link_f0:b1:3c (00:1b:11:f0:b1:3c) Source address: D-Link_f0:b1:3c (00:1b:11:f0:b1:3c) Fragment number: 0
Sequence number: 83
Frame check sequence: 0xa06a35af [correct]
QoS Control

       Priority: 5 (Video) (Video)
       ...0 .... = EOSP: Service period
       Ack Policy: Normal Ack (0x00)
       Payload Type: MSDU
       QAP PS Buffer State: 0x0

Logical-Link Control
Internet Protocol, Src: 192.168.100.1 (192.168.100.1), Dst: 192.168.202.198 (192.168.202.198) Transmission Control Protocol, Src Port: irdmi (8000), Dst Port: 32983 (32983), Seq: 1449, Ack: 1, Len: 1448 Data (1448 bytes)

Now I'd like to prioritize frames from the desktop to the server as well.

After playing around with iptables & tc for a long time, I started to look into the code and a way I found to get my desktop to set the QoS Priority field is to use following iptables command:

desktop# iptables -t mangle -A POSTROUTING -p tcp --dport 8000 -j CLASSIFY --set-class 0:103

PPI version 0, 32 bytes
IEEE 802.11 QoS Data, Flags: .......TC
Type/Subtype: QoS Data (0x28)
Frame Control: 0x0188 (Normal)
Duration: 44
BSS Id: D-Link_f0:b1:3c (00:1b:11:f0:b1:3c)
Source address: IntelCor_9e:db:5d (00:1d:e0:9e:db:5d)
Destination address: D-Link_f0:b1:3c (00:1b:11:f0:b1:3c)
Fragment number: 0
Sequence number: 124
Frame check sequence: 0xd60fe864 [correct]
QoS Control
Priority: 3 (Excellent Effort) (Video)
Ack Policy: Normal Ack (0x00)
Payload Type: MSDU
Transmit Opportunity (TXOP) Limit Requested: 0x00
Logical-Link Control
Internet Protocol, Src: 192.168.202.198 (192.168.202.198), Dst: 192.168.100.1 (192.168.100.1) Transmission Control Protocol, Src Port: 32983 (32983), Dst Port: irdmi (8000), Seq: 1, Ack: 1449, Len: 0

The argument '--set-class 0:103' is of course not really intuitive though after reading the code it is clear why: they are magic! :)

wme.c: classify_1d

	/* skb->priority values from 256->263 are magic values to
	 * directly indicate a specific 802.1d priority.  This is used
	 * to allow 802.1d priority to be passed directly in from VLAN
	 * tags, etc.
	 */
	if (skb->priority >= 256 && skb->priority <= 263)
		return skb->priority - 256;

Trying the same trick as on the server with DSCP I had no success:

desktop# iptables -t mangle -A OUTPUT -p tcp --dport 8000 -j DSCP --set-dscp-class EF

No frame from the desktop has the QoS Priority field set. There is some code in classify_1d which does something with DSCP but I haven't figured it out how it is suppose to work.

So my question is, is my approach with iptables and DSCP supposed to work?

thanks,
daniel

maintainer
#1 lore raw
Quoted 1 line · Daniel Wagner

So my question is, is my approach with iptables and DSCP supposed to work?

The honest answer probably is: we don't know either, and it depends on the drivers on both sides and the AP in the middle...

johannes

#2 lore raw
Quoted 5 lines · Johannes Berg
Quoted 1 line · Daniel Wagner

So my question is, is my approach with iptables and DSCP supposed to work?

The honest answer probably is: we don't know either, and it depends on the drivers on both sides and the AP in the middle...

Ok. I'll try to poke a bit around in this area. Maybe I find something useful.

thanks anyway,
daniel

#3 lore raw
Quoted 1 line

Ok. I'll try to poke a bit around in this area. Maybe I find something useful.

If I use --set-dscp with 0x18 (precedence level 3) instead of my orignial --set-dscp-class EF (expedited forwarding) then I see the QoS Priority field set to 3 (as with the CLASSIFY trick).

desktop# iptables -t mangle -A OUTPUT -p tcp --dport 8000 -j DSCP --set-dscp-class EF -> no effect

desktop# iptables -t mangle -A OUTPUT -p tcp --dport 8000 -j DSCP --set-dscp 0x18
-> Frame 19 (174 bytes on wire, 174 bytes captured)
PPI version 0, 84 bytes
IEEE 802.11 QoS Data, Flags: .......TC
Type/Subtype: QoS Data (0x28)
Frame Control: 0x0188 (Normal)
Duration: 44
BSS Id: D-Link_f0:b1:3c (00:1b:11:f0:b1:3c)
Source address: IntelCor_9e:db:5d (00:1d:e0:9e:db:5d)
Destination address: D-Link_f0:b1:3c (00:1b:11:f0:b1:3c)
Fragment number: 0
Sequence number: 1224
Frame check sequence: 0x321df371 [correct]
QoS Control
Priority: 3 (Excellent Effort) (Video)
Ack Policy: Normal Ack (0x00)
Payload Type: MSDU
Transmit Opportunity (TXOP) Limit Requested: 0x00
Logical-Link Control
Internet Protocol, Src: 192.168.202.198 (192.168.202.198), Dst: 192.168.100.1 (192.168.100.1) Transmission Control Protocol, Src Port: 46310 (46310), Dst Port: irdmi (8000), Seq: 1, Ack: 1449, Len: 0

Together with some printks I see that the code in wme.c classify_1d does
exactly the thing which it should. It returns the dscp value correctly.
Using --set-dscp 0x18 classify_1d returns 3 and with --set-dscp-class EF, classify_1d returns 5. That means there is a problem with mapping the result of classify_1d to the correct QoS priority, right?

daniel

#4 lore raw
Quoted 3 lines

Together with some printks I see that the code in wme.c classify_1d does
exactly the thing which it should. It returns the dscp value correctly.
Using --set-dscp 0x18 classify_1d returns 3 and with --set-dscp-class EF, classify_1d returns 5.

Crap, that's not correct. I think I found the problem:

static unsigned int classify_1d(struct sk_buff *skb)

{
       ...

	if (dscp & 0x1c)
		return 0;

	return dscp >> 5;
}

--set-dscp 0x28 (precendence level 5) classify_1d returns the correct value 5.

Setting with --set-dscp-class EF the TOS field it will have value 0xb8 (10111000b) which maps to Bit 7-2: DSCP
Bit 1-0: ECN (expliliciet congestion notifaction)

So the question is what the 'if' is for?

stay tuned!

daniel

#5 lore raw

PATCH 1/1 Use only precedence level of DSCP field for frame classification

Bit 4-5 of DSCP should not be considered by classify_d1. The 802.11 QoS Priority field is only depending on the precedence level.

---
 net/mac80211/wme.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)
net/mac80211/wme.c +0 -2
index 4310e2f..7229e95 100644
@@ -47,8 +47,6 @@ static unsigned int classify_1d(struct sk_buff *skb)
 		return 0;
 	}
 
-	if (dscp & 0x1c)
-		return 0;
 	return dscp >> 5;
 }
 
Signature · 1 line
1.5.6.2
maintainer
#6 lore raw

Re: [PATCH] Use only precedence level of DSCP field for frame classification

Quoted 2 lines · Daniel Wagner

Bit 4-5 of DSCP should not be considered by classify_d1. The 802.11 QoS Priority field is only depending on the precedence level.

Mind explaining a bit more? I don't even know what DSCP really is, any good reference?

johannes

#7 lore raw

Re: [PATCH] Use only precedence level of DSCP field for frame classification

Quoted 6 lines · Johannes Berg
Quoted 2 lines · Daniel Wagner

Bit 4-5 of DSCP should not be considered by classify_d1. The 802.11 QoS Priority field is only depending on the precedence level.

Mind explaining a bit more? I don't even know what DSCP really is, any good reference?

Sure. Cisco has a nice summery:

http://www.cisco.com/en/US/tech/tk543/tk757/technologies_tech_note09186a00800949f2.shtml

In short DSCP supercedes the 'Type of Service' field in the IP header. It is backwarts compatible, that means prescedence field in TOS maps to the DSCP precedence level.

The current code checks if the other bits of the TOS are all zero. This ok for configuration which do only set the precedence level. But RFC 2598 introduces the Expeddited Forwarding (EF) which sets some of those lower bits. I think this test is not necessary and in the case for EF even wrong.

daniel