mdadm crash with lots of devs

mdadm crash with lots of devs

[Jan Engelhardt]

Using mdadm-3.2.2 from openSUSE_12.1:

# modprobe brd rd_size=1 rd_nr=4096
# mdadm -C /dev/md9 -l 10 -n 384 /dev/ram{0..383}
mdadm: Defaulting to version 1.2 metadata
Segmentation fault (core dumped)

#1  Create (st=0xb4c600, mddev=0x7fff3bda64c0 "/dev/md9", chunk=512, level=10, 
    layout=12593328, size=4, raiddisks=384, sparedisks=0, 
    name=0x7fff3bda64c7 "9", homehost=0x7fff3bda6fe0 "seven", uuid=0x0, 
    subdevs=384, devlist=0xb49030, runstop=0, verbose=0, force=0, 
    assume_clean=0, bitmap_file=0x0, bitmap_chunk=65534, write_behind=0, 
    delay=5, autof=0) at Create.c:899
899                                     fprintf(stderr,
(gdb) l
894                                                info_new.uuid, chosen_name);
895                                     me = map_by_devnum(&map, st->container_dev);
896                             }
897
898                             if (st->ss->write_init_super(st)) {
899                                     fprintf(stderr,
900                                             Name ": Failed to write metadata to %s\n",
901                                             dv->devname);
902                                     st->ss->free_super(st);
903                                     goto abort;
(gdb) p dv
$1 = <optimized out>

Re: mdadm crash with lots of devs

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 2048 bytes --]

On Mon, 9 Jul 2012 16:38:47 +0200 (CEST) Jan Engelhardt <jengelh@inai.de>
wrote:

> 
> Using mdadm-3.2.2 from openSUSE_12.1:
> 
> # modprobe brd rd_size=1 rd_nr=4096
> # mdadm -C /dev/md9 -l 10 -n 384 /dev/ram{0..383}
> mdadm: Defaulting to version 1.2 metadata
> Segmentation fault (core dumped)
> 
> #1  Create (st=0xb4c600, mddev=0x7fff3bda64c0 "/dev/md9", chunk=512, level=10, 
>     layout=12593328, size=4, raiddisks=384, sparedisks=0, 
>     name=0x7fff3bda64c7 "9", homehost=0x7fff3bda6fe0 "seven", uuid=0x0, 
>     subdevs=384, devlist=0xb49030, runstop=0, verbose=0, force=0, 
>     assume_clean=0, bitmap_file=0x0, bitmap_chunk=65534, write_behind=0, 
>     delay=5, autof=0) at Create.c:899
> 899                                     fprintf(stderr,
> (gdb) l
> 894                                                info_new.uuid, chosen_name);
> 895                                     me = map_by_devnum(&map, st->container_dev);
> 896                             }
> 897
> 898                             if (st->ss->write_init_super(st)) {
> 899                                     fprintf(stderr,
> 900                                             Name ": Failed to write metadata to %s\n",
> 901                                             dv->devname);
> 902                                     st->ss->free_super(st);
> 903                                     goto abort;
> (gdb) p dv
> $1 = <optimized out>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


This bug is fixed by 
  http://neil.brown.name/git?p=mdadm;a=commitdiff;h=401142133

It isn't specific to large numbers of devices.

If you would like an update for 12.1, I suggest opening a bug on
bugzilla.novell.com.
If you want to work with larger numbers of devices, I suggest getting the
mdadm from Factory - it has a number of fixes in this area.

Thanks,
NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: mdadm crash with lots of devs

[Jan Engelhardt]

On Monday 2012-07-09 23:50, NeilBrown wrote:
>> Using mdadm-3.2.2 from openSUSE_12.1:
>> 
>> # modprobe brd rd_size=1 rd_nr=4096
>> # mdadm -C /dev/md9 -l 10 -n 384 /dev/ram{0..383}
>> mdadm: Defaulting to version 1.2 metadata
>> Segmentation fault (core dumped)
>
>It isn't specific to large numbers of devices.

Indeed not, it seems to be related to the rather small size
of the device :)
Using an rd_size=64 (with that old mdadm version) will emit
a warning rather than crash, and rd_size=1024 seems to be
the first working size.