Bug 516168

Created attachment 356622 [details]
Attached traceback automatically from anaconda.

Ok,

So this is not as generic a bug as you think, something weird is going on,
from the log:

[2009-07-29 19:36:06,011]     INFO: devices to scan: ['sda1', 'sda2', 'sdb1', 'sr0', 'sda', 'sdb']
[2009-07-29 19:36:06,257]    DEBUG: scanning sda1 (/devices/pci0000:00/0000:00:07.0/host0/target0:0:0/0:0:0:0/block/sda/sda1)...
[2009-07-29 19:36:06,258]    DEBUG: looking for device 'sda1'...
[2009-07-29 19:36:06,258]    DEBUG: found None
[2009-07-29 19:36:06,259]    DEBUG: sda1 is a partition
[2009-07-29 19:36:06,262]    DEBUG: looking for device 'sda'...
[2009-07-29 19:36:06,263]    DEBUG: found None
[2009-07-29 19:36:06,264]    DEBUG: /devices/pci0000:00/0000:00:07.0/host0/target0:0:0/0:0:0:0/block/sda does not exist

So it begins with scanning sda1, then cannot find the disk in the devicetree
at it has not been scanned yet (all fine) and then it errors out with:

[2009-07-29 19:36:06,264]    DEBUG: /devices/pci0000:00/0000:00:07.0/host0/target0:0:0/0:0:0:0/block/sda does not exist

Which is the beginning of the end.

Hmm, ok, I take that back the issue does seem to be as generic as you think, digging deeper ...

Ok so the problem is that the find the disk and add it to the tree code in
addUdevPartitionDevice() passes a sysfs path without /sys/ prefixed to
udev_get_block_device()

Where as udev_get_block_device() expects a path prefixed with /sys/.

This was a pre-existing problem before the recent udev handling changes clumens did, but before these changes, udev_get_block_device() would just return
None in this case, leading to:

            if disk is None:
                # if the current device is still not in
                # the tree, something has gone wrong
                log.error("failure scanning device %s" % disk_name)
                return  

So if a partition was scanned before a disk we would ignore it. In the new udev code however we have:

    dev = udev_get_device(sysfs_path)
    if not dev or not dev.has_key("name"):
        return {"name": None, "symlinks": []}
    else:
        return dev

So  udev_get_block_device() is no longer returning None due to the wrong sysfs path, and therefor we now backtrace.

I'll send a patch to the list fixing the sysfs path issue.


p.s.

This does expose a deeper problem, just like had some places in the old storage code which expected /dev/sda and others which expected just sda, we seem to have done the same now with /sys. udev_device_get_sysfs_path() returns 
a path without the /sys prefix, where as udev_get_block_device expects a path
prefixed with /sys. This seems like something which we ought to fix.

Status upd: a patch for this has been send to the list.

This is fixed by this commit:
http://git.fedorahosted.org/git/?p=anaconda.git;a=commitdiff;h=2694e31a34d7f89ab84d59fbdfcffd3986cc1fcb

Which will be in 12.11

Confirmed fixed in rawhide.