blog'o thnet

To content | To menu | To search

Sunday 18 February 2007

DHCP Bug Reintroduced in SXCE Build 57

After live upgrade'ing to snv_57, i was unable to get network informations, normally obtained using internal DHCP server. In the system log file, i can see something like:

Sat Feb 17 11:06:18 unic /sbin/dhcpagent[1494]: [ID 91213 daemon.warning]
 configure_v4_lease: incorrect broadcast address 255.255.255.255 specified
 for nge0; ignoring
Sat Feb 17 11:06:18 unic /sbin/dhcpagent[1494]: [ID 94624 daemon.warning]
 checkaddr: expected peer address 192.168.1.255 on nge0, have 255.255.255.255

In fact, it seems an old bug was reintroduced while working on an other (related) CR. See bug report 6516629 for a detail description of the problem. Since it was already fixed in snv_58, and because there are other known issues with this build being unable to boot some systems i will stay with build 56, waiting for build 58.

Thursday 18 August 2005

Update the Notebook to FreeBSD 6.0-BETA2

I decided to take advantage of this three days week-end (at least in France) to upgrade the Notebook to FreeBSD 6.0-BETA2 and follow the RELENG_6 branch testing the upcoming FreeBSD release.

In fact, i hope this one may help me in two areas... more precisely: fix a very annoying bug resulting in a USB interrupt storm and help on the wifi side, based on the work done to incorporate the wpa_supplicant code from ports(7).

The installation went smooth and well, although there is no package yet (it is a beta release, not a release candidate) and all third party need to be compile from sources. So, i installed the minimum required. In the same time, update (one more time) the installation notes, notably regarding the DHCP part (this branch switched from the ISC DHCP client v3.x to the OpenBSD DHCP client which was based on ISC DHCP v2.x). The biggest problem was to figure how to set it up in order to be able to update its DNS records obtained from the DHCP server. The solution is very simple: don't let the client do this setting (as in RELENG_5 with ISC DHCP) but do it on the server side level. It has the other advantage not to care how to do this on multiple heterogeneous clients (UNIX, Unix-like, Windows, etc.).

On the other side, there is one problem though. Because this release needs testing, the default kernel comes with debugging options enable (INVARIANTS and WITNESS(4) in particular) which have the side effect of slowing down the machine, really. These options help debugging and lightning some problems that may appeared in the development process. One of this is called a LOR and i encountered two of them. They are already known to the developers, but we don't know yet the real impact of these: harmless or not. Here is the thread about this on current@.

Go testing. I will try to reproduce the panic i encountered before to know the code path where this happens.