Hi Neil, On Tuesday 28 December 2010 07:05:42 Neil Brown wrote: > On Mon, 27 Dec 2010 22:35:45 +0100 Laurent Pinchart wrote: > > On Monday 27 December 2010 22:04:08 Neil Brown wrote: > > > On Mon, 27 Dec 2010 21:35:48 +0100 Laurent Pinchart wrote: > > > > > So something odd is happening, and more logs are needed to guess > > > > > what. > > > > > > > > I'll try to get more logs. In the meantime I've found something > > > > weird. If I run mdadm --assemble --scan twice in a row, the > > > > partitions are correctly created the second time. > > > > > > > > The next problem will be that the array is created in read-only mode. > > > > mdadm -- readwrite /dev/md126 doesn't help there, but echo active > > > > > /sys/class/block/md126/md/array_state does. > > > > > > The array is read-only because 'mdmon' is not running. When mdadm > > > assembles the array it starts mdmon to monitor the array. This is > > > needed when the array uses a metadata format that the kernel doesn't > > > understand. It is mdmon's job to switch the array to read-write > > > > > > I cannot think why mdmon is not being started, unless maybe it doesn't > > > exist on the initramfs - it should be in /sbin. > > > > Bingo! That solved part of the problem. Adding mdmon to initramfs now > > switches the array to read-write. Partitions detection still doesn't > > work though. > > Good! > > > > The logs you sent show that the partitions aren't getting detected at > > > all, so it isn't the case that they get detected and then removed. > > > > > > I don't remember exactly what triggers the partition detection. Maybe > > > just opening the array device... > > > It is entirely possible that once you get the mdmon issue sorted out, > > > the other will disappear. > > > > > > You could try > > > > > > mdmon md126 & > > > > > > and see if that makes the array read/write and creates the partitions. > > > Also check with ps to confirm that mdmon isn't running. > > > > After adding mdmon to the initramfs image, the md126 array is correctly > > switched to active. ps reports that mdmon is running of md127 (the > > container), but not md126. > > > > The partitions are still not created. If I run mdmon md126, I get > > > > mdmon: md126 is not a container - cannot monitor > > It should have been 'md127' - md127 is the container, md126 is the array > inside the container. > > But now that mdmon is on the initramfs, this shouldn't be needed. > > > And the partitions are then created. Does that sound familiar ? > > I guess it is simply anything that accesses /dev/md126 will cause the > partition detection to happen. Usually udev comes along and pokes an md > device as soon as it is created and that should be enough for the > partitions to appear. Do you have udev running?? > > Maybe I need to get mdadm to open/close the array once it is assembled so > as to trigger partition detection. > > If you are using udev, then something strange is happening. > > If you aren't using udev, then you need to have something open /dev/md/vol0 > after mdadm assembles it and before you try to mount it. I'm using mdev, the embedded udev replacement from busybox. I suppose it doesn't open the array. Would it make sense to have mdadm open the array after assembling it, or is that udev's job only ? -- Regards, Laurent Pinchart -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
