On Oct 26, 7:25am, Neil Brown wrote:
} Subject: Re: RAID5 refuses to accept replacement drive.
Hi Neil, hope your week is going well, thanks for the reply.
> > Environment:
> > Kernel: 2.4.33.3
> > MDADM: 2.4.1/2.5.3
> > MD: Three drive RAID5 (md3)
> Old kernel, new mdadm. Not a tested combination unfortunately. I
> guess I should try booting 2.4 somewhere and try it out...
Based on what I found, its probably an old library issue as much as
anything.
More below.
> > Drives were shuffled to get the machine operational. The machine came
> > up with md3 degraded. The md3 device refuses to accept a replacement
> > partition using the following syntax:
> >
> > mdadm --manage /dev/md3 -a /dev/sde1
> >
> > No output from mdadm, nothing in the logfiles. Tail end of strace is
> > as follows:
> >
> > open("/dev/md3", O_RDWR) = 3
> > fstat64(0x3, 0xbffff8fc) = 0
> > ioctl(3, 0x800c0910, 0xbffff9f8) = 0
> Those last to lines are a called to md_get_version.
> Probably the one in open_mddev
>
> > _exit(0) = ?
>
> But I can see no way that it would exit...
>
> Are you comfortable with gdb?
> Would you be interested in single stepping around and seeing what path
> leads to the exit?
My apologies for not being quicker on the draw, I should have gone
grovelling with gdb first.
The problem appears to be due to what must be a broken implementation
of getopt_long in the version of the installed C library. Either that
or the reasonably complex.... :-) option parsing in mdadm is tripping
it up.
As I noted before the following syntax fails:
mdadm --manage /dev/md3 -a /dev/sde1
After poking around a bit and watching the option parsing in gdb I
noticed that the following syntax should work:
mdadm /dev/md3 -a /dev/sde1
I tried the latter command outside of GDB and things worked
perfectly. The drive was added to the RAID5 array and synchronization
proceeded properly.
I then failed out a drive element on one of the other MD devices on
the machine and was able to repeat the problem. The following refused
to work:
mdadm --manage /dev/md1 -a /dev/sdb2
While the following worked:
mdadm /dev/md1 -a /dev/sdb2
The getopt_long function is not picking up on the fact that -a should
have optarg set to /dev/sdb2 when the option is recognized. Instead
optarg is set to NULL and devs_found is left at 1 rather than 2. That
results in mdadm simply exiting without saying anything.
I know the 1.x version of mdadm we were using before processed the
'mdadm --manage' syntax properly. This must have been the first time
we had to add a drive element back into an MD device since we upgraded
mdadm.
I would be happy to chase this a bit more or send you a statically
linked binary if you want to see what it is up to. At the very least
it may be worthwhile to issue a warning message on exit if mdadm has
an MD device specification, a mode specification and no devices.
I remember trying to build a statically linked copy of mdadm with
dietlibc and ran into option parsing problems. The resultant binary
would always exit complaining that a device had not been specified. I
remember the dietlibc documentation noting that the GNU folks had an
inconsistent world view when it came to getopt processing
semantics... :-)
I suspect there is a common thead involved in both cases.
> NeilBrown
Hope the above is useful. Let me know if you have any
questions/issues.
Happy Halloween.
Greg
}-- End of excerpt from Neil Brown
As always,
Dr. G.W. Wettstein, Ph.D. Enjellic Systems Development, LLC.
4206 N. 19th Ave. Specializing in information infra-structure
Fargo, ND 58102 development.
PH: 701-281-1686
FAX: 701-281-3949 EMAIL: greg@xxxxxxxxxxxx
------------------------------------------------------------------------------
"Fools ignore complexity. Pragmatists suffer it. Some can avoid it.
Geniuses remove it.
-- Perliss' Programming Proverb #58
SIGPLAN National, Sept. 1982
-
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
