Jeff Trawick wrote:
Brian Ruthven - Sun UK wrote:
I know this isn't telnet or rlogin, but could this be another
manifestation of 6905405 ?
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6905405
Thanks! I tried the work-around documented in that CR
ndd -set /dev/tcp tcp_max_buf 64000
and so far it works.
Viewing the CR internally, I see the statement
"This problem should only happen for a loopback connection
falling from fused state back to non-fused state."
By chance does that happen when the connection is half closed?
Personally I'm not sure. Offhand, I don't know whether connection
closing goes through unfusing, but it could be that you've found another
trigger for it. You should be able to check the fused state of the
connection (if it hangs around long enough for you to locate and inspect
it), but this may not tell you whether it was fused in the past.
Regards,
Brian
Jeff Trawick wrote:
Beginning with snv_129, an httpd testcase started failing. The
essence of the testcase is the following:
client connects to server over loopback
client sends HTTP header and blocks in another write() for 100000 bytes
server reads HTTP header
server writes response to client
server shuts down socket for write
server then goes into a loop:
wait a short time for socket to be readable
try to read 512 bytes
with snv_128a, the server is able to read the entire 100000 bytes
from the client, 512 bytes at a time
with snv_129, the server only gets 1 byte at a time from the client
and eventually gives up
if the same client program runs on snv_111b and connects to the
server on snv_129 over the e1000g0 interface, the server is able to
read the entire 100000 bytes from the client, 512 bytes at a time
if the client program runs on the server system but uses the e1000g0
interface instead of 127.0.0.1, it fails in the same way
some config details:
* snv_128a and snv_129 were running on top of VirtualBox on top of
2009.06
* the client is /usr/bin/perl (5.8.4), though as truss shows it
stuck in write(...,...,100000) during the interesting period, I
don't think that matters
* the server is not the integrated Apache but a standalone build;
the same binaries and configuration are used for both snv_128a and
snv_129
Does this ring a bell, or should I experiment further? Any hints on
what to experiment with next?
--
Brian Ruthven
Solaris Revenue Product Engineering
Sun Microsystems UK
Sparc House, Guillemont Park, Camberley, GU17 9QG
_______________________________________________
networking-discuss mailing list
[email protected]
