On Monday October 30, [email protected] wrote:
> Hi all,
>
> I'm running the following software-raid setup:
>
> two raid 0 with two 250GB disks each (sdd1-sdg1) named md_d2 and md_d3
> one raid 5 with three 500GB disks (sda2-sdc2) and the two raid0 as
> members named md_d5
> one raid 1 with 100MB of each of the 500GB disks (sda1-sdc1) named md_d1
>
> The only raid device that actually has a partition table is md_d5. The
> other devices are used unpartitioned, which brings me to the first
> question: Is it possible to run partitioned and unpartitioned software
> raids at the same time?
>
> Back to the topic now after this question. The resulting problem is: due
> to the raid5 layout, the partition table of md_d5 is written to where a
> partition table on md_d3 would be as well:
>
> [~]>fdisk -l /dev/md_d3
>
> Disk /dev/md_d3: 500.1 GB, 500113211392 bytes
> 2 heads, 4 sectors/track, 122097952 cylinders
> Units = cylinders of 8 * 512 = 4096 bytes
>
> Device Boot Start End Blocks Id System
> /dev/md_d3p1 1 244142 976566 83 Linux
> /dev/md_d3p2 244143 5126956 19531256 8e Linux LVM
> /dev/md_d3p3 5126957 488279488 1932610128 8e Linux LVM
>
> Note that the end of md_d3p3 is way beyond the end of the actual device.
> Now during boot udev tries to find out about the content of the devices,
> using the vol_id program. It checks the various locations for raid
> superblocks, lvm superblocks. What happens show the following strace
> excerpts:
snip
>
> The kernel above contains a lot of patches (gentoo's hardened sources),
> but the same syndrom can be seen with vanilla 2.6.18 or 2.6.19 rc3.
>
> Even if there are likely a dozend workarounds (create a partition table
> on the raid 0s one by one and resync; no not rely on raid=part for
> autodetection as the raid5 doesn't come up automatically anyway; don't
> use vol_id) this should in my oppinion not happen. The points I'd like
> to criticize are:
> - The partition table read code, which accepts to create the devices
> even though they are obviously wrong,
Yeah, I don't like that either. I tried submitting a patch to fix it
but different people have different ideas about what the 'right'
solution is (do we clip the partition to size, or make it disappear
completely?).
> - The partitioned raid device creation code, which creates subdevices
> which are larger than the containing device,
This is exactly the same as the above. The raid code doesn't create
the subdevices. The partition reading code does that.
> - The layer in the kernel that allows the read beyond end of device down
> to the raid driver,
Yes... This is generic_make_request in block/ll_rw_blk.c.
It check that the address is within the partition, but it does not
check that the address after mapping is still within the device.
It assumes partitions always fit within devices.
> - Most importantly, the raid driver for failing that bad mannered.
>
Maybe, though I would rather the block layer guaranteed never to send
a request that was out-of-range...
> I honestly didn't look into the other software raid drivers, which are
> likely to produce the same result. The attached patch for raid0.c makes
> accesses beyond the end of a device into Buffer I/O errors:
>
> xxxxxx Buffer I/O error on device md_d3p3, logical block 483152512
>
> Regards,
> Christian
> --- raid0.c.orig 2006-10-30 00:12:22.000000000 +0100
> +++ raid0.c 2006-10-30 00:14:48.000000000 +0100
> @@ -415,6 +415,10 @@
> chunksize_bits = ffz(~chunk_size);
> block = bio->bi_sector >> 1;
>
> + if (block >= mddev->array_size) {
> + bio_endio(bio, bio->bi_size, -EIO);
> + return 0;
> + }
That would work. Similar fixes needed in linear, raid10, and possible
raid5.
I'll try to push the 'fix it in the block layer' first though.
Thanks,
NeilBeown
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
