More on hotplug issue w/HAL

[Matthew Mastracci]

David Zeuthen wrote:

Did you happen to take a look at the SCSI UDI LUN modifications I sent 
with my last message (they may have gotten drowned out in the log 
messages :)).  Those fixed the missing block device problem I was having 
both on startup and hotplug add/remove.

I've done some more thinking to create a more complete (and probably 
more correct) solution:

I was thinking that a scsi_device needs to have the following attributes 
so that they don't errorneously get merged into a single SCSI interface 
node.

(parsed from /path/to/device/host:channel:id:lun in sysfs)
scsi_device.host = n
scsi_device.channel = n
scsi_device.id = n
scsi_device.lun = n

scsi_host should have:

(parsed from the /path/to/device/hostn path in sysfs)
scsi_host.host = n

These will uniquely identify all scsi_host and scsi_device nodes within 
a single device.  They can then be appended to the appropriate UDI for a 
scsi_host/scsi_device node.

> Ok, basically you need to check that everything works on both hald
> startup (when detecting all the devices) and on hotplug add/remove.
> 
> One thing I've been missing from your debug logs are lines like this
> 
>  [I] linux/linux_osspec.c:529 handle_udev_node_created() : udev
>  NodeCreated: /udev/sda /initrd/sys/block/sda
>  
>  [I] linux/linux_osspec.c:563 handle_udev_node_created_found_device() :
>  Setting block.device=/udev/sda for 
>  udi=/org/freedesktop/Hal/devices/temp/47
>  [I] linux/linux_osspec.c:529 handle_udev_node_created() : udev
>  NodeCreated: /udev/sda1 /initrd/sys/block/sda/sda1
>  
>  [I] linux/linux_osspec.c:563 handle_udev_node_created_found_device() :
>  Setting block.device=/udev/sda1 for 
>  udi=/org/freedesktop/Hal/devices/temp/48
> 
> on hotpluggin - the debug means that hald has received a dbus signal
> from the udev service associating a device file with a sysfs path. You
> might want to check that the file 
> 
>  /etc/dbus-1/system.d/udev_sysbus_policy.conf
> 
> exist on your system. It seems to be missing from an RPM I built from
> udev-012, but that may be my own fault.

That file is missing from my system.  I used "make install" + 
checkinstall to install HAL/dbus/udev.  I was getting a dbus permissions 
error on connection when starting udev earlier.  After moving that file 
to system.d, it looks like everything is working great!

> I'm currently working on this using polling and ioctl(2) on the cdrom
> device file (like magicdev). Unfortunately I won't have time to hack on
> this before tomorrow evening due to other commitments in real life. If
> you write code to detect media change for your devices, we should
> definitely put it in HAL.
> 
> (btw, some day we want to use the dbus events that Rob Love's kernel
> will be generating instead of polling)

I'm going to try sniffing the USB data stream to see if I get any sort 
of notification that the device media has changed.  If not, I'll 
probably have to use your polling code to perform the same tests to 
detect it.  I'll keep you posted on what I discover.

> It's the maximum amount of time, don't worry. My favourite test case is
> hotplugging this tree of devices

> and this take a fair amount of time (~20 sec) on my old laptop mostly
> due to swapping (it's a 128MB box running Fedora) and devlabel being
> invoked all the time. It does work however, even though the devices
> appear in a funny order.

It takes one heckova long time for the devices to appear in HAL for 
myself.  I'm hoping that HAL and udev might make a lot of the hotplug 
nastiness (updfstab + devlabel + pam_console_apply) go away.  ;)

Matt.