Something wrong with -CURRENT? (network problems)

From: Stefan Esser <se_at_freebsd.org>
Date: Thu, 12 Dec 2013 12:27:50 +0100
Since about the time of teh NEWCONS commit, freshly built kernels do not
work correctly for me. Symptoms are:

- long delays opening TCP connections (even to localhost)
- TCP transmission errors (e.g. fetchmail aborts because of short
  transfers, could be due to a timeout being triggered)

I first noticed this during multi-user startup, since some of the
daemons took tens of seconds to start (especially bad is SpamAssasin),
instead of fractions of a second.

At first I thought this could be caused by testing NEWCONS, but the
symptoms of a kernel with SCONS are identical.

Kernel configuration and other parameters are unchanged.

I do not have time to bisect the commits, but just wanted to ask whether
this is a general problem with -CURRENT, or only affects my test system
(my local file-, media-, mail-, and web-server).

(I might have time to bisect revisions during the week-end, though.)

System information:

i7-2660, N67 chipset, 24GB RAM, amd64

ZFS only configuration (ZRAID1, 4*2TB)

Kernel config is just a stripped down generic (removing all unnecessary
devices) with the following KLDs loaded:

Id Refs Address            Size     Name
 1   25 0xffffffff80200000 e0efe8   kernel
 2    1 0xffffffff8100f000 22dd88   zfs.ko
 3    2 0xffffffff8123d000 5b68     opensolaris.ko
 4    1 0xffffffff81412000 1478     fdescfs.ko
 5    1 0xffffffff81414000 c250     ipfw.ko
 6    1 0xffffffff81421000 3923     linprocfs.ko
 7    1 0xffffffff81425000 999      linsysfs.ko

Kernel based on SVN rev. 258894 works reliably with latest world.

The kernel I built after seeing the initial NEWCONS commits (still
with SCONS) worked, but shortly thereafter, when I tried NEWCONS, the
symptoms started (but persisted after going back to SCONS).

I just checked SVN rev 259250 and the problem persists (without NEWCONS
configured into the kernel).

So my questions are:

- is anybody else affected or is this a local problem?

- has there been a commit that touched the network stack or might
  cause the observed delays?

Thanks in advance fpr any insight you might be able to offer!

Regards, STefan
Received on Thu Dec 12 2013 - 10:34:10 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:45 UTC