Bug#465556: mdadm hotadd flag does not work in its shortform

[Folken]

On Fri, 2008-04-11 at 13:40 +0200, martin f krafft wrote:
> tags 465556 unreproducible upstream
> thanks
> 
> also sprach Philipp Wollschlegel <folken at kabelsalat.ch> [2008.02.13.0946 +0100]:
> > mdadm hotadd flag is not interpreted correctly:
> > 
> > debian:/home/folken# mdadm --manage /dev/md1 -a /dev/sda2
> > 
> > the return value of the above is 0. but the raiddevice is not
> > added to the device. 
> 
> I cannot reproduce this:
> 
> lapse:/tmp/cdt.nnet6607> sudo losetup -f one                           #[1,305]
> lapse:/tmp/cdt.nnet6607> sudo losetup -f two                             #[306]
> lapse:/tmp/cdt.nnet6607> sudo mdadm -Cn2 -l1 /dev/md0 /dev/loop[01]      #[307]
> mdadm: array /dev/md0 started.
> lapse:/tmp/cdt.nnet6607> sudo mdadm --set-faulty /dev/md0 /dev/loop1     #[308]
> mdadm: set /dev/loop1 faulty in /dev/md0
> lapse:/tmp/cdt.nnet6607> cat /proc/mdstat                                #[309]
> Personalities : [raid1] 
> md0 : active raid1 loop1[2](F) loop0[0]
>       8128 blocks [2/1] [U_]
> 
> unused devices: <none>
> lapse:/tmp/cdt.nnet6607> sudo mdadm --remove /dev/md0 /dev/loop1         #[310]
> mdadm: hot removed /dev/loop1
> lapse:/tmp/cdt.nnet6607> sudo mdadm -a /dev/md0 /dev/loop1; echo $?      #[311]
> mdadm: re-added /dev/loop1
> 0
> lapse:/tmp/cdt.nnet6607> cat /proc/mdstat                                #[312]
> Personalities : [raid1] 
> md0 : active raid1 loop1[1] loop0[0]
>       8128 blocks [2/2] [UU]
> 
> unused devices: <none>

I tried it with the commands you entered and it works as long as the
second device added was once a member. (e.g. set faulty) it failes if
the md array is created initialy with missing keyword and is then added
with the manage command. Like so:

root at corsair:/tmp# for each in one two ; do dd if=/dev/zero of="${each}"
bs=1M count=50 ; done 
50+0 records in
50+0 records out
52428800 bytes (52 MB) copied, 2.41071 seconds, 21.7 MB/s
50+0 records in
50+0 records out
52428800 bytes (52 MB) copied, 4.03959 seconds, 13.0 MB/s

root at corsair:/tmp# losetup /dev/loop0 one 
root at corsair:/tmp# losetup /dev/loop1 two

root at corsair:/tmp# mdadm --create /dev/md2 -n 2 -l 1 /dev/loop0 missing
mdadm: array /dev/md2 started.

root at corsair:/tmp# mdadm --manage /dev/md2 -a /dev/loop1 ; echo $?
0

root at corsair:/tmp# mdadm --manage /dev/md2 --add /dev/loop1 ; echo $?
mdadm: added /dev/loop1
0

The hiccup seems not architecture specific, but seems to apply to 
mdadm - v2.5.6 in general.
on mdadm - v2.6.4 - 19th October 2007 i wasn't able to reproduce it.