2

EDIT 2: This may be a bug in the kernel or module. Reported here https://bugzilla.kernel.org/show_bug.cgi?id=109471 Can anybody confirm that this phenomenon is definitely out of spec?

Original:

I'm tracking down a network phenomenon in my lab where a Wi-Fi device (possibly multiple) can connect to one of my access points (AP) with no problem, but when communicating through another AP, network flow is disrupted. The bad AP sends "oversize" ethernet frames of length 1522 (see below rx_length_errors counter). The AP where the problem does NOT exist never sends frames larger than 1514 and they are all accepted and forwarded by the gateway.

I believe the Wi-Fi client device is using an MTU of 1500 on both APs because the TCP payloads in both cases are always at/below 1500 bytes. This rules out the client being a factor. When the ethernet frame is added, one AP adds 14 bytes (resulting in 1514-byte frames), and the other AP adds 22 bytes (resulting in 1522-byte frames).

Sample oversized packet (MAC and IP modified for anonymity).

03:51:25.978066 1b:22:fe:cd:f3:77 > a8:27:db:fe:9e:51, ethertype IPv4 (0x0800), length 1522: (tos 0x0, ttl 64, id 55530, offset 0, flags [DF], proto TCP (6), length 1500)
    192.168.1.149.41744 > 54.239.17.6.443: Flags [.], cksum 0x5b39 (correct), seq 1724444568:1724446028, ack 3109213756, win 377, length 1460

`

If I connect to another of my APs, the maximum frame size ends up being 1514, and those have no problem getting through the gateway.

A comparison of the 1514 byte frame vs the 1522 byte frame reveals an incorrect four-byte frame check sequence (normal checksum offloading?) as well as four trailer bytes. Together that's 8 bytes altogether, which may account for the 8 byte difference between the working and nonworking, however I'm not 100% sure frame check sequence is absent on the working, even though wireshark doesn't show it in the packet dissector, so that may only account for 4 of the extra 8 bytes.

(Why would Linux pad an ethernet frame with a trailer when the enclosed IP packet is already 1500 bytes long?)

On the gateway, frame and length errors increase in lock-step with the 1522 byte frames coming from the AP. So this is how I reached the conclusion that the frame size is a significant factor.

/sys/class/net/eth0/statistics/rx_frame_errors:36059
/sys/class/net/eth0  /statistics/rx_length_errors:36059

Good AP is running kernel

Linux goodAP 3.18.11+ #781 PREEMPT Tue Apr 21 18:02:18 BST 2015 armv6l GNU/Linux

Bad AP is running kernel

Linux notsogoodAP 3.18.11-v7+ #781 SMP PREEMPT Tue Apr 21 18:07:59 BST 2015 armv7l GNU/Linux

So the question is, why is a Linux server (Access Point) generating overly large (1522-byte) ethernet frames with seemingly invalid frame check sequence?

EDIT: Adding full packet dump as exported from wireshark:

No.     Time            Source                Destination           Protocol Length Info
     35 21:55:21.644314 192.168.1.149        54.239.25.200         TCP      1522   [TCP Retransmission] 44053→443 [ACK] Seq=350 Ack=155 Win=88832 Len=1460 [ETHERNET FRAME CHECK SEQUENCE INCORRECT]

Frame 35: 1522 bytes on wire (12176 bits), 1522 bytes captured (12176 bits)
    Encapsulation type: Ethernet (1)
    Arrival Time: Dec 15, 2015 21:55:21.644314000 EST
    [Time shift for this packet: 0.000000000 seconds]
    Epoch Time: 1450234521.644314000 seconds
    [Time delta from previous captured frame: 0.150820000 seconds]
    [Time delta from previous displayed frame: 3.683211000 seconds]
    [Time since reference or first frame: 9.568260000 seconds]
    Frame Number: 35
    Frame Length: 1522 bytes (12176 bits)
    Capture Length: 1522 bytes (12176 bits)
    [Frame is marked: False]
    [Frame is ignored: False]
    [Protocols in frame: eth:ethertype:ip:tcp]
    [Coloring Rule Name: Bad TCP]
    [Coloring Rule String: tcp.analysis.flags && !tcp.analysis.window_update]
Ethernet II, Src: xx:xx:xx:xx:xx:xx (xx:xx:xx:xx:xx:xx), Dst: Raspberr_xx:xx:xx (xx:xx:xx:xx:xx:xx)
    Destination: Raspberr_xx:xx:xx (xx:xx:xx:xx:xx:xx)
        Address: Raspberr_xx:xx:xx (xx:xx:xx:xx:xx:xx)
        .... ..0. .... .... .... .... = LG bit: Globally unique address (factory default)
        .... ...0 .... .... .... .... = IG bit: Individual address (unicast)
    Source: xx:xx:xx:xx:xx:xx (xx:xx:xx:xx:xx:xx)
        Address: xx:xx:xx:xx:xx:xx (xx:xx:xx:xx:xx:xx)
        .... ..0. .... .... .... .... = LG bit: Globally unique address (factory default)
        .... ...0 .... .... .... .... = IG bit: Individual address (unicast)
    Type: IP (0x0800)
    Trailer: 2f8e4de5
    Frame check sequence: 0x29f37f68 [incorrect, should be 0x3b863a28]
        [FCS Good: False]
        [FCS Bad: True]
            [Expert Info (Error/Checksum): Bad checksum]
                [Bad checksum]
                [Severity level: Error]
                [Group: Checksum]
Internet Protocol Version 4, Src: 192.168.1.149 (192.168.1.149), Dst: 54.239.25.200 (54.239.25.200)
    Version: 4
    Header Length: 20 bytes
    Differentiated Services Field: 0x00 (DSCP 0x00: Default; ECN: 0x00: Not-ECT (Not ECN-Capable Transport))
        0000 00.. = Differentiated Services Codepoint: Default (0x00)
        .... ..00 = Explicit Congestion Notification: Not-ECT (Not ECN-Capable Transport) (0x00)
    Total Length: 1500
    Identification: 0xaaf3 (43763)
    Flags: 0x02 (Don't Fragment)
        0... .... = Reserved bit: Not set
        .1.. .... = Don't fragment: Set
        ..0. .... = More fragments: Not set
    Fragment offset: 0
    Time to live: 64
    Protocol: TCP (6)
    Header checksum: 0x3234 [validation disabled]
        [Good: False]
        [Bad: False]
    Source: 192.168.1.149 (192.168.1.149)
    Destination: 54.239.25.200 (54.239.25.200)
    [Source GeoIP: Unknown]
    [Destination GeoIP: Unknown]
Transmission Control Protocol, Src Port: 44053 (44053), Dst Port: 443 (443), Seq: 350, Ack: 155, Len: 1460
    Source Port: 44053 (44053)
    Destination Port: 443 (443)
    [Stream index: 1]
    [TCP Segment Len: 1460]
    Sequence number: 350    (relative sequence number)
    [Next sequence number: 1810    (relative sequence number)]
    Acknowledgment number: 155    (relative ack number)
    Header Length: 20 bytes
    .... 0000 0001 0000 = Flags: 0x010 (ACK)
        000. .... .... = Reserved: Not set
        ...0 .... .... = Nonce: Not set
        .... 0... .... = Congestion Window Reduced (CWR): Not set
        .... .0.. .... = ECN-Echo: Not set
        .... ..0. .... = Urgent: Not set
        .... ...1 .... = Acknowledgment: Set
        .... .... 0... = Push: Not set
        .... .... .0.. = Reset: Not set
        .... .... ..0. = Syn: Not set
        .... .... ...0 = Fin: Not set
    Window size value: 347
    [Calculated window size: 88832]
    [Window size scaling factor: 256]
    Checksum: 0x1c58 [validation disabled]
        [Good Checksum: False]
        [Bad Checksum: False]
    Urgent pointer: 0
    [SEQ/ACK analysis]
        [iRTT: 0.028704000 seconds]
        [Bytes in flight: 2077]
        [TCP Analysis Flags]
            [Expert Info (Note/Sequence): This frame is a (suspected) retransmission]
                [This frame is a (suspected) retransmission]
                [Severity level: Note]
                [Group: Sequence]
            [The RTO for this segment was: 7.247571000 seconds]
            [RTO based on delta from frame: 12]
    Retransmitted TCP segment data (1460 bytes)



0000  b8 27 eb fe 9e 71 1c 56 fe cd f3 75 08 00 45 00   .'...q.V...u..E.
0010  05 dc aa f4 40 00 40 06 32 33 c0 a8 46 95 36 ef   ....@.@.23..F.6.
0020  19 c8 ac 15 01 bb 71 be a7 a7 cb 7c 15 6f 50 10   ......q....|.oP.
0030  01 5b 1c 58 00 00 17 03 03 08 18 00 00 00 00 00   .[.X............
0040  00 00 02 cd 63 94 47 c0 31 81 4d ea b2 13 b6 46   ....c.G.1.M....F
0050  cb 5d 4d 64 71 21 b1 c1 3c 6d d6 1c 5e 0a 89 4b   .]Mdq!..<m..^..K
0060  a3 f8 58 f3 a9 2e 95 81 2c 08 be cc 64 a2 11 c3   ..X.....,...d...
0070  8c 7e 10 7f 51 e5 87 e7 c0 49 08 19 65 0a 7a 5b   .~..Q....I..e.z[
0080  ee 22 2e 93 31 cf 22 5d 7e 36 5f ee 3b 2e 43 f0   ."..1."]~6_.;.C.
0090  83 23 9f 69 7f ae 82 22 04 f4 02 42 fb 28 43 f3   .#.i..."...B.(C.
00a0  8f 05 80 6c fd 7f ef 47 7b 07 d2 b7 d9 e8 ab 78   ...l...G{......x
00b0  54 67 af 61 bf 55 89 33 f3 85 5d 7b a4 53 34 73   Tg.a.U.3..]{.S4s
00c0  17 75 b0 da 6a 31 d5 0a 86 7a 11 66 7f 9f 81 5a   .u..j1...z.f...Z
00d0  96 bc 64 72 0d da 32 01 8d 88 70 d3 f5 e7 70 29   ..dr..2...p...p)
00e0  94 be 35 37 10 10 41 79 fc d3 4f f1 1d a2 c3 ef   ..57..Ay..O.....
00f0  32 81 50 e2 2a ca 6e bf 95 b8 77 1b ee 76 ba fb   2.P.*.n...w..v..
0100  8d 85 25 a2 47 7a 96 fd af c3 39 98 8b bd ff 31   ..%.Gz....9....1
0110  ae ca 15 60 e1 2f 00 4f 43 c4 20 11 92 47 91 10   ...`./.OC. ..G..
0120  f6 0c a7 ea 5d 54 f1 01 eb d4 9e 7a bb ae 08 5f   ....]T.....z..._
0130  9b 31 85 af 4e b4 0b 03 ad 6b 51 11 51 e5 f8 d8   .1..N....kQ.Q...
0140  af ff 78 46 3e 24 6c 2e 40 3e aa 27 b7 87 10 c7   ..xF>$l.@>.'....
0150  5a aa d3 33 fa b6 bc 4d e9 2f 18 34 81 98 28 34   Z..3...M./.4..(4
0160  18 de e8 32 ae c0 21 6a 52 20 3b 5f 12 6c f4 df   ...2..!jR ;_.l..
0170  71 c6 e0 cc b2 d1 75 94 f2 e3 63 e5 4d a5 7c ba   q.....u...c.M.|.
0180  1c 46 e1 83 e1 ca e3 c7 dc c3 08 d7 0a 3e e2 3b   .F...........>.;
  • 1
    A regular ethernet frame can be 1518 bytes long. Using 802.1Q, you can have a four byte VLAN tag inserted in the frame, for a total of 1522 bytes. I'm guessing your AP is doing 802.1Q, but the router isn't. – Ron Maupin Dec 16 '15 at 4:26
  • Hi Ron - not using 802.1q tagging. Also note in the packet sample that there is no VLAN tag. – Brad Hein Dec 16 '15 at 14:37
  • Please provide a hex dump of the frame header (at least 18 bytes of the frame) for one of the 1522 byte frames. – Ron Maupin Dec 16 '15 at 18:11
  • OK Ron no problem. I amended my question once more, this time with the first third or so of the raw frame in hex dump form. – Brad Hein Dec 16 '15 at 19:24
1

An ethernet frame with a payload of 1500 bytes:

  • Without an 802.1Q tag is 1518 bytes
  • With an 802.1Q tag is 1522 bytes

The WAP showing you 1514 bytes is due to the fact that the frame does not have an 802.1Q tag, and the interface or driver is not giving Wireshark the FCS. From the Wireshark Wiki:

Most Ethernet interfaces also either don't supply the FCS to Wireshark or other applications, or aren't configured by their driver to do so; therefore, Wireshark will typically only be given the green fields, although on some platforms, with some interfaces, the FCS will be supplied on incoming packets.

The WAP showing you the 1522 byte frame size is apparently showing you the FCS (four bytes), so the interface or driver is giving it to Wireshark. The other four bytes are due to an 802.1Q tag.

Your router is not expecting 802.1Q tagged frames, so it is giving you errors. You need to disable 802.1Q on the WAP. WAPs can use 802.1Q on the wired side to separate traffic among multiple SSIDs.

| improve this answer | |
  • Hi Ron - actually no VLAN tag header is present. I'll amend my question with a more detailed packet dissection to help clarify. – Brad Hein Dec 16 '15 at 14:39
0

After further sleuthing, trying different USB wifi cards I found that the problem followed the USB wifi device and didn't happen on other USB wifi devices.

Pursuing the problem as a wifi card driver bug.

The device in question is "TP-Link TL-WN-722N (ath9k_htc) USB".

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.