[Bug 954452] New: udev fails to create /dev/disk/ symlinks in Xen guests
http://bugzilla.suse.com/show_bug.cgi?id=954452 Bug ID: 954452 Summary: udev fails to create /dev/disk/ symlinks in Xen guests Classification: openSUSE Product: openSUSE Tumbleweed Version: 2015* Hardware: x86-64 OS: Linux Status: NEW Severity: Normal Priority: P5 - None Component: Basesystem Assignee: bnc-team-screening@forge.provo.novell.com Reporter: ohering@suse.com QA Contact: qa-bugs@suse.de Found By: --- Blocker: --- Current Tumbleweed snapshots fail to create the /dev/disk symlinks in the initrd. As a result the guests hang during boot. This affects both PV and HVM guests. I did a fresh install. Have to check if this bug affects only the initrd or also the inst-sys/rescue system. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c1
Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c2
--- Comment #2 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c3
Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c4
--- Comment #4 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c5
--- Comment #5 from Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c6
Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c7
Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c8
Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c9
--- Comment #9 from Robert Milasan
The strings for a given block device are "free strings", that means a given block driver is free to pick any name. It just happend that IDE used to pick hd, SCSI picked "sd" and so on. In the early days the PV drivers used to claim the well known names and their major:minors to make it easier to install and run existing distros.
Not really, the device name is given based on what layer is used, for example SCSI/SATA/ATA usually can go thru SCSI or libata (I'm talking about kernel) and their naming will start with 'sd'. Xen has its own things and will be something like 'xvd', but you have 'hd' because you have a bit of a different setup, maybe not using Xen's own driver or layer or something, but the kernel and kernel sees it as an old device. I really dont know enough about Xen to comment too much. Here you are emulating an old machine, somehow.
Today the names can be still passed into the guest. For some reason I picked "hd" instead of "xvd" for my PV guests. So that part might be fixeable by adjusting both the domU.cfg and also the used devicenames within the VM. In the end I would call this a regression. "hd" per se has no relationship to IDE hardware anymore, as explained above. Please restore "hd" support, I'm sure such names are in use.
Not sure why the HVM guest fails. Here the "hd" in domU.cfg has additional meaning, it tells qemu to create an emulated IDE controller which is later driven by the SATA drivers. Have to see how to debug this.
'hd' is not really is use, because is not possible. A normal machine that is 64bit wont be using IDE, is too old. At least is what I know. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c10
--- Comment #10 from Olaf Hering
(In reply to Olaf Hering from comment #8)
The strings for a given block device are "free strings", that means a given block driver is free to pick any name. It just happend that IDE used to pick hd, SCSI picked "sd" and so on. In the early days the PV drivers used to claim the well known names and their major:minors to make it easier to install and run existing distros.
Not really, the device name is given based on what layer is used, for
It depends on what is passed to "blockdev_alloc(name, major, minor)" (siplified, whatever the exact function name is). And of course every block driver is free to pick any value of "name". If its already taken it gets -EBUSY. My point is that the "hd" code is not only connected to IDE hardware, but also to the xen frontend driver. Is it difficult to restore and maintain that removed code for upcoming releases? -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c11
--- Comment #11 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c12
--- Comment #12 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c13
Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c14
--- Comment #14 from Olaf Hering
Try:
grep ^KERNEL /usr/lib/udev/rules.d/60-persistent-storage.rules
Adding 'hd*' to that line wont help, as the links are created by the rest of the lines looking like this:
I added hd* to the list of allowed prefixes, and the PV guest starts fine now. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c15
Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c16
--- Comment #16 from Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c17
Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c18
--- Comment #18 from Robert Milasan
Is there a reason why the by-label/by-uuid are skipped for everything which is not handled? It looks like this bug could have been easily avoided by creating the symlinks for on-disk content unconditionally. There is certainly no need for by-id or by-path, which could be handled by the whitelist as it is implemented now.
Well, I don't know, thats why I was confused about this. The factory version of systemd/udev is the one that implements the new 'blacklisting/whitelisting' and didn't find yet the commit that says why this has been done. The previous way worked quite well. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c19
--- Comment #19 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c20
Franck Bui
bug#954872 is an example that the prefix "hd" is still in use, perhaps just due to old habit.
IMHO, you should simply rename your device "xvd" and we're done. IDE device support has been removed since some time now, and udev has assumed (since 2009) that no device will be named "hd*" since that's the reserved kernel name used by IDE HW. I don't know Xen at all, but it seems that "xvd" is the name for xen virtual block device and that's the name used by the driver handling this kind of virtual block device so I'm not sure why you're allowed to choose a different name. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c21
Franck Bui
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c22
Olaf Hering
I don't know Xen at all, ...
Then please dont close this bug as invalid... Instead apply the patch for the .rules file. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c23
--- Comment #23 from Franck Bui
(In reply to Franck Bui from comment #20)
I don't know Xen at all, ...
Then please dont close this bug as invalid... Instead apply the patch for the .rules file.
Then please explain why we should allow this hack whereas you can simply rename your device correctly ? I also suggest you to send your patch to upstream if you believe it as valid. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c24
Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c25
Franck Bui
Franck, how would he rename his device(s)?
"xvd", it seems to be the "official" kernel name used by the xen virtual block devices. Apparently, Olaf choose "hd" for no good reason (see his comment #8), and I don't see any reason to add one more special handling for him. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c26
--- Comment #26 from Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c27
--- Comment #27 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c28
--- Comment #28 from Franck Bui
OK, made a push request on systemd upstream, lets see what they say.
thanks, could you give the link? -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c29
--- Comment #29 from Franck Bui
libxl supports the domU.cfg files from xend. Their syntax is equal anyway, but some aspects of the "old" disk=[] syntax needed an update to handle more use cases. In the end this translates to "existing domU.cfg files must be supported". This means that even the names of the devices (such as "hda") are accepted on the libxl side.
I'm sure there are domU.cfg files with 'hd?' out there, and TW will simply breaks in those VMs. For no good reason.
Currently your the first one reporting a bug for that, and if the name you're using is considered as a bad practice, then you should rename it instead of asking for us to support one more hack forever. -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c30
--- Comment #30 from Robert Milasan
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c31
--- Comment #31 from Olaf Hering
Currently your the first one reporting a bug for that, and if the name you're using is considered as a bad practice, then you should rename it instead of asking for us to support one more hack forever.
Who said "hd?" is bad practice? -- You are receiving this mail because: You are on the CC list for the bug.
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c32
--- Comment #32 from Olaf Hering
http://bugzilla.suse.com/show_bug.cgi?id=954452
http://bugzilla.suse.com/show_bug.cgi?id=954452#c33
Olaf Hering
participants (1)
-
bugzilla_noreply@novell.com