Bugzilla – Bug 1524
Fragmentation Threshold equals to Packet Size at MAC Layer (Data + IP + UDP) crash the reception
Last modified: 2012-12-05 14:48:49 UTC
Created attachment 1462 [details] WifiRemoteStationManager patch Hello, I do a little patch for wifi-remote-station-manager. The fragmentation does work when Packet Size (Data + IP + UDP) equals to Fragmentation Threshold. In fact, the implementation does not take into account that the maximum Fragment Size has to be the Fragmentation Threshold. Example: Packet Size (Data+IP+UDP) = 3000 Number of Fragments = 2 First Fragment Size S1= FragmentationThreshold(2346) Second Fragment Size S2= 3000 % 2346 = 654 Thus, the first fragment packet has a size of FP1=(S1+26(WifiMacHeader)+4(WIFI_MAC_FCS_LENGTH)) and the second fragment packet has a size of FP2=(S2+26(WifiMacHeader)+4(WIFI_MAC_FCS_LENGTH)) Like we can see FP1 > FragmentationThreshold Now, what happens if Packet Size (Data+IP+UDP) = 2346 Number of Fragments = 2 (S1(2346)+26(WifiMacHeader)+4(WIFI_MAC_FCS_LENGTH)) and (S2(0)+26(WifiMacHeader)+4(WIFI_MAC_FCS_LENGTH)) Then, in the second segment is sent only the header. In fact, that breaks the fragmentation procedure and at application layer the client does not receive anything.
Hi Fabian, I tested your patch and it worked fine for the case when packet size = Threshold. But it failed when packet size is smaller (I tried size = threshold - 1). Additional changes are needed to correctly create fragments from the packet. Anyway, the fragmentation procedure is not working correctly. As you already mentioned that the final fragment size is still larger than the threshold. The other thing I noticed was that the length of each fragment can be an odd number, which is incorrect as the spec states that all fragments but the last shall have even length.
Created attachment 1480 [details] Modified patch for bug 1524 (fragmentation) Built on Fabian's patch. GetNFragments was modified so that it takes WifiMacHeader into account. The final fragment size (MPDU) is now at most FragmentationThreshold. This patch does _not_ addressed the issue of even length MPDU.
Fixed in changeset 9165:7b219a73b844