On 10/08/2015 09:44 PM, Benjamin Marzinski wrote: > Multipath has an issue with user_friendly_names set in the initramfs. > If the bindings are in the initramfs bindings file, it will create them, > and it may use bindings that are different than the ones in the regular > file system. Once multipathd starts up in the regular file system, it > will try to register the existing bindings, but that may (and in many > cases, is likely to) fail. If it can't reanme it, will pick a new > binding. Since when multipathd starts discovering the existing devices, > it obviously doesn't know all of the existing devices yet, it may very > well pick a binding that's already in use by a device that it hasn't > discovered yet. In this case multipath will be unable to rename the > device to the new binding. Unfortunately, if it fails the rename, it > never resets the alias of the device stored in the mpvec to current > alias of the actual dm device. So multipath will have devices in the > mpvec where the alias and the wwid don't match the actual dm devices > that exist. This can cause all sorts of problems. > > This patch does two things to deal with this. First, it makes sure that > if the rename fails, the device will reset its alias to the one that it > currently has. > > Second, it adds a new option to help avoid the problem in the first > place. There actually already is a solution to this issue. If > multipathd is started with -B, it makes the bindings file read-only. If > multipathd is started this way in the initramfs, it will never make new > bindings for a device. If the device doesn't already have a binding, it > will simply be named with its wwid. The problem with this method is that > AFAICT it causes a maintenance problem with dracut. At least on RedHat, > dracut is simply copying the regular multipathd.service file into the > initramfs. I don't see a way in multipathd.service to only start > multipathd with the -B option in the initramfs (If there is a way, I'd > be happy to use that). So, the only alternative that lets us use -B is > to have a separate multipathd.service file that dracut uses for the > initramfs. This seems like more work than it's worth. > > Instead, this patch pulls some code from systemd to detect if multipathd > is running in the initramfs. If it is, and if new_bindings_in_boot is > not set, multipathd will set the bindings file to read-only, just like > the -B option does. > > Signed-off-by: Benjamin Marzinski <bmarzins@xxxxxxxxxx> Please split off this patch. The first issue is a real fix, it should really be a separate patch. For the second one it actually quite simple to inject a different service file for dracut (I did that actually in our dracut package). So I'd rather see this happening instead of second-guessing by looking at magic files. Which will fail for anyone _not_ using dracut. Cheers, Hannes -- Dr. Hannes Reinecke zSeries & Storage hare@xxxxxxx +49 911 74053 688 SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg) -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
