Bug 818087

(In reply to comment #0)
> Steps to Reproduce:
> 1. # mkfs.ext4 /dev/mapper/disk1
> 2. # tune2fs -c 0 -i 0 -m 0 /dev/mapper/disk1
> 3. Enable user quota


Can you be a little more specific on 3) ?  There must be a mount command in there somewhere, and a quota check possibly ... at what point is the mount exposed to users, and how (samba, nfs, etc?)

Thanks,
-Eric

# grep /dev/mapper/disk1 /etc/fstab
/dev/mapper/disk1 /home/stud1 ext4 defaults,usrquota,nobarrier,_netdev 0 0

File data is exposed to users by: samba3x and NFS

# rpm -qa|grep samba
samba3x-common-3.5.10-0.109.el5_8
samba3x-winbind-3.5.10-0.109.el5_8
samba3x-winbind-3.5.10-0.109.el5_8
samba3x-3.5.10-0.109.el5_8

Now when I revisit my installation notes I realize that I copied the aquota.user file from the old SAN-disk (ext3 based) the the new SAN-disk (ext4).

Can this be the cause of the problem?

# cp <src>/aquota.user <dst>/aquota.user ;copy the quota-db
# quotacheck -vu <dst>                   ;Update the quota-db on an empty fs
# quotaon -vu <dst>                      ;activate quota

after this I copied all data using rsync (with samba and NFS service stopped)

# rsync -avH --delete --exclude=/aquota.user --exclude=/lost+found /home/stud1/ /new/home/stud1/

Just so I don't have to infer anything, can you give me explicit, step-by-step details from mkfs, through mount, with any quota and/or samba/nfs operations, up to rsync, and finally whatever led to the message, as well as you can?

I don't think copying aquota.user is a problem, but I'll have to give that a little more thought.

- Create fs and enable quota

# yum install e4fsprogs
# mkfs.ext4 /dev/mapper/stud1
# tune4fs -c 0 -i 0 -m 0 /dev/mapper/stud1
# mount /new/home/stud1
# cp /home/stud1/aquota.user /new/home/stud1/
# quotacheck -vu /new/home/stud1
# quotaon -vu /new/home/stud1

then

# rsync ...

then

# /etc/init.d/smb start
# /etc/init.s/nfs start


- /etc/fstab - See above

- /etc/exports
/home/stud  some*.xx.se(rw) other*.xx.se(rw) more*.xx.se(rw,no_root_squash)
/home/stud1 some*.xx.se(rw) other*.xx.se(rw) more*.xx.se(rw,no_root_squash)


/etc/samba/smb.conf
[global]
 server string      = Student home file server
 workgroup          = XX
 security           = ADS
 realm              = XX.SE
 interfaces         = eth0 lo
 bind interfaces only = yes
 name resolve order = host
 smb ports          = 445
 socket options     = TCP_NODELAY
 log level          = 1
 log file           = /var/log/samba/%m.log
 max log size       = 300
 local master       = no
 preferred master   = no
 domain master      = no
 load printers      = no
 printcap name      = /dev/null
[homes]
 comment            = Home Folder
 browseable         = no
 writable           = yes
 create mask        = 0600
 directory mask     = 0700

Ok, there are still a couple of upstream commits that are needed to resolve this, I think - they address cases where delayed writes happen before quota is enabled.

Can either reporter confirm or deny that there is a chance for IO to happen before quota is enabled on the filesystem in their usecase?

I do not really know exactly what you are asking...

1) I mount the file system at boot from /etc/fstab (see #2 above).

2) The NFS- and SMB-services are started manually after boot.

3) The volume is accessed via iSCSI + multipath from a Compellent SAN.

No IO is performed before quota is enabled (as far as I can tell).

This request was evaluated by Red Hat Product Management for
inclusion in a Red Hat Enterprise Linux release.  Product
Management has requested further review of this request by
Red Hat Engineering, for potential inclusion in a Red Hat
Enterprise Linux release for currently deployed products.
This request is not yet committed for inclusion in a release.

Magnus, there are a couple upstream patches that should take care of this.

This can be tested/demonstrated with something like:

#!/bin/bash

DEV=/dev/null # XXX CHANGEME
MNT=/mnt/test # XXX CHANGEME

umount $DEV
mkfs.ext4 $DEV 
mount $DEV $MNT -oquota
quotacheck $MNT
umount $DEV
mount $DEV $MNT -oquota
dd if=/dev/zero of=$MNT/foo bs=1M count=1
quotaon $MNT
dd if=/dev/zero of=$MNT/foo seek=1 bs=1M count=1
sync
dmesg | tail

in the failing case you'll see the traceback from the warning

Reproduced in kernel-2.6.18-323.el5 and verified in kernel-2.6.18-335.el5.

> Please provide test package? Our partner is willing to test the fix.

I placed what didn't exceed the "quota" on my people.page from the build performed by Eric S. on 8/15. 

It's at http://people.redhat.com/ltroan/fixes/.818087/ and includes src as well as select x86 and x86_64 kernel images.

Please test and report results in your Customer Portal case or directly here in the bugzilla.

I will look at the reproducer.  However, this bug was specifically opened for the warning messages which arises from a mismatch of reserved and claimed quota space.  If that is fixed, this bug will be closed.  If you have discovered a new quota-related problem, we will need a new bug to track that.

(This behavior may be related to delayed allocation, and there is some potential mismatch in limits & actual use, but what you show above is more of an overrun than I would expect).

In any case, I suggest opening a new bug for this other problem.

Thanks,
-Eric

For what it's worth, I can't reproduce the 4% over-allocation result above on my test box.  I ran the reproducer 5 times, each with 10 loops, and did not see the error.

It would be good to know whether or not it looks like a regression on the partner's box.

-Eric

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

http://rhn.redhat.com/errata/RHBA-2013-0006.html