[Chicken-hackers] [PATCH] TCP read/write timeout change

[Peter Bex]

Hi all,

As has been pointed out by Mario (and before, Felix, in 
http://lists.nongnu.org/archive/html/chicken-hackers/2013-05/msg00013.html),
the change in the way the TCP read and/or write timeouts
work now is slightly undesirable.

When there's a slow connection, and the read or write actions
get interrupted (EAGAIN/EINTR), the timeouts shouldn't get
reset otherwise it might never timeout at all.  However, when
a few bytes got read, there is some progress, and the timer
should probably be reset.  The exact intended semantics remain
unclear, but at least this seems most sane; it avoids getting
stuck completely, but if some progress is made the system won't
see it as "no response".

There's also another problem in the timeouts: they get computed
at socket creation time, which is completely bogus (I have no idea
why this hasn't been found at all).

Here's a patch that should make it work "correctly".

Cheers,
Peter
-- 
http://www.more-magic.net

0001-Reset-TCP-read-write-timeout-whenever-at-least-some-.patch
Description: Text document