Mirrors_Framework/nuttx
1
0
Fork 0
mirror of https://github.com/apache/nuttx.git synced 2026-06-06 08:36:24 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

While working with version 7.10 I discovered a problem in TCP stack that could be observed on high network load. Generally speaking, the problem is that RST flag is set in unnecessary case, in which between loss of some TCP packet and its proper retransmission, another packets had been successfully sent. The scenario is as follows: NuttX did not receive ACK for some sent packet, so it has been probably lost somewhere. But before its retransmission starts, NuttX is correctly issuing next TCP packets, with sequence numbers increasing properly. When the retransmission of previously lost packet finally succeeds, tcp_input receives the accumulated ACK value, which acknowledges also the packets sent in the meantime (i.e. between unsuccessful sending of lost packet and its proper retransmission). However, variable unackseq is still set to conn->isn + conn->sent, which is truth only if no further packets transmission occurred in the meantime. Because of incorrect (in such specific case) unackseq value, few lines further condition if (ackseq <= unackseq)is not met, and, as a result, we are going to reset label.

Browse Source
This commit is contained in:
Jakub Łągwa
2016-06-20 06:55:29 -06:00
committed by Gregory Nutt
parent d5b8869b10
commit 338b915008
5 changed files with 43 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ChangeLog
+18
View File
@@ -12084,3 +12084,21 @@
option. Without this configuration setting, interrupt level output
will be asynchronous. And (3) vsyslog is now a system call and is
usable with other-than-FLAT builds (2016-06-19).
* TCP Networking: While working with version 7.10 I discovered a
problem in TCP stack that could be observed on high network load.
Generally speaking, the problem is that RST flag is set in
unnecessary case, in which between loss of some TCP packet and its
proper retransmission, another packets had been successfully sent.
The scenario is as follows: NuttX did not receive ACK for some
sent packet, so it has been probably lost somewhere. But before
its retransmission starts, NuttX is correctly issuing next TCP
packets, with sequence numbers increasing properly. When the
retransmission of previously lost packet finally succeeds, tcp_input
receives the accumulated ACK value, which acknowledges also the
packets sent in the meantime (i.e. between unsuccessful sending of
lost packet and its proper retransmission). However, variable unackseq
is still set to conn->isn + conn->sent, which is truth only if no
further packets transmission occurred in the meantime. Because of
incorrect (in such specific case) unackseq value, few lines further
condition if (ackseq <= unackseq)is not met, and, as a result, we
are going to reset label. From Jakub Łągwa (2016-06-20).