UDEV Rules Didn't Behave Like I Thought

Dan McGhee beesnees at grm.net
Tue Mar 7 11:45:07 PST 2006

This may be long and convoluted, but please bear with me.  I was 
struggling with the age-old "I can't access my combo dvd/cdrom drive 
unless I'm root" syndrome.  The situation seemed simple:  I needed two 
different symlinks to point to, in my case, 'hdc' and establish a group 
for hdc with 'rw' access.  The solution was simple, but, along the way, 
I ran into some UDEV things that I thought I understood but apparently 
don't.  My original set up were lines in /etc/udev/rules.d.  Here are 
the relevant parts:

KERNEL="hdc", SYMLINK="dvd"

and from /etc/udev/rules.d/25-lfs.rules
BUS=="ide", KERNEL=="*[!0-9]", PROGRAM="/bin/cat /proc/ide/%k/media", 
RESULT="cdrom", NAME="%k", SYMLINK="cdrom"
KERNEL=="hd*",                      GROUP="disk"

The /dev/dvd symlink was created but the /dev/cdrom symlink was not. The 
25-lfs.rules is from UDEV-71.  In my other LFS distro I have UDEV-58 and 
both symlinks were created.  The only difference is the use of "==" in 
UDEV-71 and just "=" in UDEV-58.

Anyway I discovered this thread in the archives:

in which Andrew Benton suggested this line in a udev rule

BUS="ide", KERNEL="hdc", GROUP="video", NAME="%k", SYMLINK="dvd" MODE="0660"

I added this to my 24-dvd.rules so that it would get processed before 
anything else.  I used GROUP="cdrom" and  'SYMLINK="dvd cdrom" to get 
both.  When I ran 'udevstart,' I had the two symlinks but /dev/hdc still 
had group DISK.  I changed the appropriate lines in 25-lfs.rules to read:

KERNEL=="hdc",                      GROUP="cdrom"
KERNEL=="hd*[!c]",                  GROUP="disk"

and I got what I wanted.  Here's my question.  I knew that UDEV 
processed the first rule that used NAME and ignored others using NAME.  
But, I also thought that UDEV rules worked in lexical order--once a 
match was made, no further rules were processed for that particular 
device.  But that is apparently not the case, unless the use of "==" 
changes things.  Can some one fill me in?



More information about the blfs-support mailing list