Messages in this thread |  | | From | "Ilpo Järvinen" <> | Subject | [PATCHv2 net-2.6 0/4 (was 0/3)]: TCP fixes | Date | Mon, 7 Apr 2008 14:55:40 +0300 |
| |
Hi Dave,
Ok, here's the v2 series which tries hard to avoid opening loopholes to if condition in tcp_mark_head_lost which again would make the added tcp_fragment code to not execute :-). The rest (patches 3-4 are identical to previous series 2-3). ...I repost the whole series to make it easy for you if you were fast enough to deleted them from your inbox already, others please bear me :-).
--- v1 intro follows ---
Here are some TCP fixes that resulted from the last weeks reports & debug. The first one is quite likely to hit but it's considerably harder to get it print an overflow warning, while I've seen two reports about the second one (one of them is for 2.6.24.y), please ignore the earlier version of the tcp_simple_retransmit patch. The FRTO patch is once again result of code review rather than some report but seems necessary to avoid detection of some not that likely cases as spurious RTOs when there were some other losses in the same window with the probe. Nevertheless, it should be safe to return non-FRTO behavior.
...So far, they're just compile tested. I'll see if I find some time to boot them on the evening.
There might still be one fackets_out miscounter awaiting detection & debug output (I hope Georgi catches it) because none of these seem an obvious reason for triggery of the !sacked_out && fackets_out trap.
All of them are also valid for stables but please note that it won't be too easy, at least for the first & second patch because of recent changes (especially if stable != 2.6.24.y). I'll try to do the adapted versions later on for the stable.
-- i.
|  |