Bug#488267: Should add hostap modules
(There's no need to CC me or Otavio when you reply. Just reply to the bug
report.)
On Monday 30 June 2008, Barry Tennison wrote:
> Frans Pop wrote:
> I understand the complexity of the argument.
> I would just stress that from the "simple user" (including me) point of
> view, the key things are that
I have no disagreement with that. The only issue under discussion now is
what the real source of the bug is.
> Therefore, it seems to me that the installer and the stock system have
> to be brought to the same state: if the standard (installed, lenny)
> system continues to do the renaming, then the installer must, however
> unfortunately, bend to concur, so that the newly installed system DOES
> bring up the interface.
Or: the program that is responsible for the the faulty renaming needs to
be fixed. Especially as this not only affects new installations, but is
also likely to affect upgrades (from Etch to Lenny for example).
> > Barry: a few requests.
> > 1) What was the wireless interface called in the installer?
> > You can probably tell from /var/log/installer/syslog.
>
> eth1
> You can check this from my original report (email of Fri, 27 Jun 2008
> 14:16:15 +0100 in bug 488267), and in the installer outputs I provided
> in my email of Fri, 27 Jun 2008 18:15:56 +0100 (cat /proc/net/dev and
> ifconfig while installer running)
That's not what I meant, but I probably was not clear enough.
I wanted to know what the name of the interface was when the driver loads
it. What you list here could be _after_ it is renamed by udev.
So, you have to look in the installer's syslog for lines where you see the
orinoco driver being loaded and how the interface is being referred to
there. Possibly that is indeed eth1, but it is also possible that it's
first something else and that you later see a line mentioning that udev
renames it.
> > 2) Does the workaround described above work?
>
> As I have mentioned, my own workaround, with the system exactly as
> installed, is simply to edit /etc/network/interfaces with the global
> replace of eth1 by wlan0_rename (or whatever the installed system calls
> the interface). IMHO, this is much simpler than editing udev rules
> (here be dragons!) and doesn't require a reboot.
Sure, but it does leave you with that ugly name and that name is very
simply just wrong.
> However, I am much attracted by your workaround since it avoids the
> vastly ugly interface name wlan0_rename (!), and I'm glad to report
> that it did work - but see comment after next para.
Right.
> In a little more detail,
> * I edited /etc/udev/z25_persistent-net.rules to include the
> ATTRS{type}=1 phrase
You also added the extra comma, right?
> * I rebooted - could I have done less?
Yes, unloading and reloading the hostap module should have worked too:
# modprobe -r hostap_pci
# modprobe hostap_pci
> certainly, unsurprisingly, neither udevcontrol reload_rules nor
> /etc/init.d/udev restart had any visible effect
Right. As there are no hardware events, neither actually "does" anything.
> * after reboot, the key lines of ifconfig -a read:
> > wifi0 Link encap:UNSPEC HWaddr 00-D0-59-BD-D5-C5-77-6C-[...]
> > wlan0 Link encap:Ethernet HWaddr 00:d0:59:bd:d5:c5
>
> (no mention of eth1 in ifconfig nor in /proc/net/dev nor in
> /sys/class/net/) and after an edit of /etc/network/interfaces, wlan0
> was successfully brought up.
> Do let me know if you want any more diagnostic info.
This _is_ surprising. I had expected wlan0 to be renamed to eth1.
Ah, I see the error. Typo in my last mail. It should be "ATTR{type}=1"
(and not "ATTRS"). Please try again with that.
It does explain that there was no rename at all.
> PLEASE NOTE - sorry for stating the obvious - that this single change
> will not solve the original problem, because the interface was called
> eth1 by the installer, and /etc/network/interfaces was built
> accordingly, with eth1 not wlan0. So the installer will need changing
> to use the same interface name as used by the (lenny-release) standard
> system.
Please don't worry about that. We're aware of how things fit together and
will make sure that things end up consistently.
> > 3) Could you please provide the output of
> > $ udevinfo -a -p /sys/class/net/<interface>
> > where <interface> is both of the interfaces you have in that
> > directory for your wireless.
>
> BEFORE doing the workaround as above in (2):
> (for your convenience, diff given after both outputs below)
>
> > ~$ udevinfo -a -p /sys/class/net/eth1
> >
> > looking at device '/class/net/eth1':
> > KERNEL=="eth1"
> > SUBSYSTEM=="net"
[...]
> > ATTR{type}=="801"
> > ATTR{link_mode}=="0"
> > ATTR{address}=="00:d0:59:bd:d5:c5"
> > ATTR{broadcast}=="ff:ff:ff:ff:ff:ff"
OK. This is the original wifi0 interface. Note that it has ATTR{type}
unequal to 1.
> > ~$ udevinfo -a -p /sys/class/net/wlan0_rename/
> >
> > looking at device '/class/net/wlan0_rename':
> > KERNEL=="wlan0_rename"
> > SUBSYSTEM=="net"
> > DRIVER==""
[...]
> > ATTR{type}=="1"
> > ATTR{link_mode}=="0"
> > ATTR{address}=="00:d0:59:bd:d5:c5"
And this is wlan0, which _does_ have ATTR{type} equal to 1, so the
modified rule should work.
Reply to: