Bug 110164

Created attachment 95994 [details]
Patch for /etc/sysconfig/network-scripts/network-functions

The patch changes the semantics of ethtool/mii-tool not being
installed to mean 'immediately assume the link is up'.

That's the wrong behavior.

The result is the same in the old script. The return value of 2 also
leads to the assumption the link is up.

Anyway, wathever the solution may be, I think the current script is
bugged. I don;'t wan't to wait for a minute (!) before the boot
process resumes (especially since Fedora Core 1 is partially targeted
at laptop users; laptop mode patch)

OK, new solution
Drop the modifications in check_ethtool and check_mii_tool. In the
patch for check_link_down modify
>while [ $timeout -le 10 ]; do
>  	check_mii_tool $1
>	m=$?
>	check_ethtool $1
>	e=$?
>	if [ $m -eq 1 ] && [ $e -eq 1 ] ; then
>	    return 1
>	fi
>	if [ $m -eq 2 ] && [ $e -eq 2 ] ; then
>	    return 1
>	fi
>	usleep 500000
>	timeout=$((timeout+1))
>   done

in

>while [ $timeout -le 10 ]; do
>  	check_mii_tool $1
>	m=$?
>	check_ethtool $1
>	e=$?
>	if [ $m -gt 0 ] && [ $e -gt 0 ] ; then
>	    return 1
>	fi
>	usleep 500000
>	timeout=$((timeout+1))
>   done

Which does the same as the original patch and satisfies your comments
on the original patch.

Sorry, reading back at your original comments:

        if ! LC_ALL=C ip link show dev $1 2>/dev/null| grep -q UP ; then
# Forget the above if statement. Due to hotplug the network is 
# brought up when modprobing the networking card. Thus ifup is started
# twice when starting /etc/init.d/network when no module is loaded 
# When one ifup is working, the second one will think the link is up

Hotplug is *disabled* in /etc/init.d/network.

# The above statement should use && instead of ||. 

No. That requires that any adapter support *both* methods, and that
both methods reliably report the same data (in many cases, they
don't.) If you've changed it as such, it *will* lead to longer startup
times.

 sysctl -w kernel.hotplug="/bin/true" > /dev/null 2>&1

in /etc/init.d/network before ifup is executed. Also after modprobing
the network module ifup is started.

The second solution with the if statements is better than the original
one. It does exactly the same with less statements.

'The second solution'? Which particular one are you referring to - all
the solutions I've seen so far are based on the original false premises.

I'm referring to the changes in comment #4, in which check_mii_tool en
check_ethtool are unchanged.

The if constructs in check_link_down are only based on a specific link
down detection (first and second if construct in the while loop are
mashed together). I don't know if you count that as a false premise.

(OMG. Look at the time of your response; 02:32)

I don't have the ability to reopen this bug, but I experienced the
same regression when upgrading from RH9 to FC1, and it persisted in
FC2 and FC3.  The necessary patch seems to me to be moving a single
line -- I'm not sure what all the other proposed changes by the bug
reporter were for.

(I previously blamed this issue on a change of network driver, since
the machine in question has a Broadcom 4400, and that upgrade was when
I switched from compiling the bcm4400 driver to using the provided b44
driver.)

The problem seems to be that the test in check_link_down:
  if ! LC_ALL=C ip link show dev $1 2>/dev/null| grep -q UP
leads to most of the function being skipped.  I'm not sure why the
whole function gets skipped when the link is reported as "UP", or what
the exact semantics of being "UP" are, but when doing "ifup eth0" with
my b44, it's "UP" at this point in check_link_down (even though it
wasn't before doing "ifup eth0").

I'll attach that patch shortly.

Created attachment 106686 [details]
patch against Fedora Core 3 network-functions

This is the patch mentioned in the previous comment, with appropriate
reindenting (and fixes for a bit of the indentation in the function being
off-by-one).

As a |diff -U10 -b| it looks like this (much clearer):

 check_link_down ()
 {
     if [ -x /sbin/mii-tool -o -x /sbin/ethtool ]; then
	 if ! LC_ALL=C ip link show dev $1 2>/dev/null| grep -q UP ; then
	   ip link set dev $1 up >/dev/null 2>&1
+	fi
	   timeout=0
	   while [ $timeout -le 10 ]; do
		check_mii_tool $1
		m=$?
		check_ethtool $1
		e=$?
		if [ $m -eq 1 ] || [ $e -eq 1 ] ; then
		    return 1
		fi
		if [ $m -eq 2 ] && [ $e -eq 2 ] ; then
		    return 1
		fi
		usleep 500000
		timeout=$((timeout+1))
	   done
	   return 0
-	fi
     fi
     return 1
 }

Added, will be in 7.97-1

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2005-123.html