Package: udev
Version: 0.114-2
Severity: normal
Tags: patch

Hello,
sg*|*)
        NAME=generic
        ;;

So for sg* _and_ all targets so far without a rule NAME=generic is used. 
On the specific system we have a tape-library attached and the ch.ko driver
uses ch* as device names. 
When the device was detected first the sg-driver created its /dev entries and
then the ch-driver took over and called udev for the very same 
host-id, bus, id and lun. Since scsi-devfs.sh now didn't have a rule for
ch* it gracefully overwrote the sg 
scsi/host$HOST/bus$BUS/target$TARGET/lun$LUN/generic entries with ch 
major and minors. 
Of course, the /dev/sg* still existed, but just pointed to wrong
device. The tape software trying to access the sg devices
with sg ioctls got rather confused (you have no idea how much
trouble this caused...).

Here are two patches, entirely untested though, since we solved it on
the specific system differently.


01_dev_ch.patch: Rule for media changer devices

Index: udev/scsi-devfs.sh
===================================================================
--- udev.orig/scsi-devfs.sh     2008-03-02 14:14:18.000000000 +0100
+++ udev/scsi-devfs.sh  2008-03-02 14:19:03.000000000 +0100
@@ -74,6 +74,9 @@
        NAME=$(echo "$1" | sed -e 's/st0m/mt/')
        [ $NAME = "mt0" ] && LINK=tape
        ;;
+ch*)
+       NAME=ch
+       ;;
 sg*|*)
        NAME=generic
        ;;

02_no_default_generic.patch: Abort if a device name is unknown. 
Better no devfs-style rule than a harmful rule.

Index: udev/scsi-devfs.sh
===================================================================
--- udev.orig/scsi-devfs.sh     2008-03-02 14:19:53.000000000 +0100
+++ udev/scsi-devfs.sh  2008-03-02 14:21:58.000000000 +0100
@@ -77,9 +77,13 @@
 ch*)
        NAME=ch
        ;;
-sg*|*)
+sg*)
        NAME=generic
        ;;
+*)
+       echo "Unknown device-type $1"
+       exit 1
+       ;;
 esac
 
 echo scsi/host$HOST/bus$BUS/target$TARGET/lun$LUN/$NAME $LINK



Cheers,
Bernd



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to