Skip to contentSkip to conversation

Rambling noise #2: Learning to use the v8 pquota/uquota patchset

DISCUSSION Michael L. Semon 4 participants 10 messages May 14, 2013 – May 15, 2013

lore
Share & tools
lore
#0 lore raw

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

The basic functionality works, in my opinion, and I hope nobody wastes time with a nice, educated reply. It would be mostly wasted on me and is better saved for somebody else. A reply of "do this...and this...btw, how did this test come out?" would be welcomed, though ;-)

Anyway, some vague observations as I grasp for straws...

1) The xfstests quota group tests seem to fail in different ways than the way they did before applying the patches.

2) Nothing has oopsed.

3) In testing using the xfs_quota -x command, the patches seem to work. On mount -t xfs -o gquota mounts, using the quota command from within the xfs_quota shell, the group quotas show but not the projid quotas. On mount -t xfs -o pquota mounts, the projid quotas show but not the gquota mounts. This is different than the old behavior, where the gquota numbers might be recycled into projid numbers.

4) The results of xfsquota -c print are confusing. Maybe they're showing the XFS view when they show things like 'uqnoenforce,gquota,pquota' for a mount that is gquota only. They're doubly confusing once /etc/projid and /etc/projects have been set up. The 'gqnoenforce' and 'pqnoenforce' flags show up at times for reasons that are unknown to me.

5) mount -t xfs -o gquota,pquota is not possible at this time.

6) The patches applied cleanly to a git Linux 3.10-rc1 kernel + xfs-oss, with only whitespace errors reported.

7) I question whether 'bsoft=' has a visible effect on projid quotas, whether using your patches or not. Did it ever work?

8) I had no feel on whether the filesystem had to be mounted once as gquota, then once as pquota, for the full dual functionality to work.

9) It looks like xfs_repair doesn't ruin anything, but the xfsquota -c print output looks a little different on the next mount.

That's about all that could be put together in a coherent manner. Sleep awaits.

The PC is a 32-bit Pentium 4. In addition to the kernel mentioned in (6), there are a few J. Liu and Dave Chinner patches applied as well.

Best of luck!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#1 lore raw

Hi Michael,

Have you tried this patch set with the v2 user space changes?

Looks this is the latest update from Chandra that I can searched from the mailing list:
http://oss.sgi.com/archives/xfs/2012-07/msg00293.html

Thanks,
-Jeff

Quoted 51 lines · Michael L. Semon

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

The basic functionality works, in my opinion, and I hope nobody wastes time with a nice, educated reply. It would be mostly wasted on me and is better saved for somebody else. A reply of "do this...and this...btw, how did this test come out?" would be welcomed, though ;-)

Anyway, some vague observations as I grasp for straws...

1) The xfstests quota group tests seem to fail in different ways than the way they did before applying the patches.

2) Nothing has oopsed.

3) In testing using the xfs_quota -x command, the patches seem to work. On mount -t xfs -o gquota mounts, using the quota command from within the xfs_quota shell, the group quotas show but not the projid quotas. On mount -t xfs -o pquota mounts, the projid quotas show but not the gquota mounts. This is different than the old behavior, where the gquota numbers might be recycled into projid numbers.

4) The results of xfsquota -c print are confusing. Maybe they're showing the XFS view when they show things like 'uqnoenforce,gquota,pquota' for a mount that is gquota only. They're doubly confusing once /etc/projid and /etc/projects have been set up. The 'gqnoenforce' and 'pqnoenforce' flags show up at times for reasons that are unknown to me.

5) mount -t xfs -o gquota,pquota is not possible at this time.

6) The patches applied cleanly to a git Linux 3.10-rc1 kernel + xfs-oss, with only whitespace errors reported.

7) I question whether 'bsoft=' has a visible effect on projid quotas, whether using your patches or not. Did it ever work?

8) I had no feel on whether the filesystem had to be mounted once as gquota, then once as pquota, for the full dual functionality to work.

9) It looks like xfs_repair doesn't ruin anything, but the xfsquota -c print output looks a little different on the next mount.

That's about all that could be put together in a coherent manner. Sleep awaits.

The PC is a 32-bit Pentium 4. In addition to the kernel mentioned in (6), there are a few J. Liu and Dave Chinner patches applied as well.

Quoted 9 lines

Best of luck!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#2 lore raw

No. I'll try this later on today, then post back to the group.

BTW, I'm terrible at switching between group quotas and projid quotas, and am aware that I'm too stupid to report what I see. However, because I have a projid quota to limit the size of Samba shares used by Windows 7 backup, I'm reporting this despite my stupidity. I ask for everyone's forgiveness in advance.

Thanks for your help!

Michael

Quoted 4 lines · Jeff Liu

Hi Michael,

Have you tried this patch set with the v2 user space changes?

Quoted 68 lines

Looks this is the latest update from Chandra that I can searched from the mailing list:
http://oss.sgi.com/archives/xfs/2012-07/msg00293.html

Thanks,
-Jeff

Quoted 60 lines · Michael L. Semon

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

The basic functionality works, in my opinion, and I hope nobody wastes time with a nice, educated reply. It would be mostly wasted on me and is better saved for somebody else. A reply of "do this...and this...btw, how did this test come out?" would be welcomed, though ;-)

Anyway, some vague observations as I grasp for straws...

1) The xfstests quota group tests seem to fail in different ways than the way they did before applying the patches.

2) Nothing has oopsed.

3) In testing using the xfs_quota -x command, the patches seem to work. On mount -t xfs -o gquota mounts, using the quota command from within the xfs_quota shell, the group quotas show but not the projid quotas. On mount -t xfs -o pquota mounts, the projid quotas show but not the gquota mounts. This is different than the old behavior, where the gquota numbers might be recycled into projid numbers.

4) The results of xfsquota -c print are confusing. Maybe they're showing the XFS view when they show things like 'uqnoenforce,gquota,pquota' for a mount that is gquota only. They're doubly confusing once /etc/projid and /etc/projects have been set up. The 'gqnoenforce' and 'pqnoenforce' flags show up at times for reasons that are unknown to me.

5) mount -t xfs -o gquota,pquota is not possible at this time.

6) The patches applied cleanly to a git Linux 3.10-rc1 kernel + xfs-oss, with only whitespace errors reported.

7) I question whether 'bsoft=' has a visible effect on projid quotas, whether using your patches or not. Did it ever work?

8) I had no feel on whether the filesystem had to be mounted once as gquota, then once as pquota, for the full dual functionality to work.

9) It looks like xfs_repair doesn't ruin anything, but the xfsquota -c print output looks a little different on the next mount.

That's about all that could be put together in a coherent manner. Sleep awaits.

The PC is a 32-bit Pentium 4. In addition to the kernel mentioned in (6), there are a few J. Liu and Dave Chinner patches applied as well.

Best of luck!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#3 lore raw
In reply to Jeff Liu
Quoted 10 lines · Jeff Liu

Hi Michael,

Have you tried this patch set with the v2 user space changes?

Looks this is the latest update from Chandra that I can searched from the mailing list:
http://oss.sgi.com/archives/xfs/2012-07/msg00293.html

Thanks,
-Jeff

It looks like it's not patching very well in light of Dave's CRC userspace changes, and I botched the mkfs part of the merge.

For xfsprogs include/xfs_sb.h, Patch #1 has this:

+#define XFS_SB_VERSION2_NO_OQUOTA 0x00000100 /* sep prj quota inode */

However, in the current xfsprogs include/xfs_sb.h, there is this line:

#define XFS_SB_VERSION2_CRCBIT 0x00000100 /* metadata CRCs */

This overloaded meaning of 0x00000100 did not seem right, so after my failure with the finished mkfs.xfs, it was time to go back to the current git xfsprogs. I'm sure that the userspace will be reworked before the kernel code is final. I'll show some patience, and Dave's upcoming CRC code activation will probably force me to back Chandra's patchset out anyway.

Despite my bumbling through it all, the kernel gquota/pquota patchset shows promise.

Thanks!

Michael

P.S. - I applied your xfs_growfs_data_private patch, and all seems well so far. In particular, there was no change in the test results from ./check -xfs -g growfs.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#4 lore raw
Quoted 9 lines · Michael L. Semon
Quoted 8 lines · Jeff Liu

Hi Michael,

Have you tried this patch set with the v2 user space changes?

Looks this is the latest update from Chandra that I can searched from the mailing list:
http://oss.sgi.com/archives/xfs/2012-07/msg00293.html

Quoted 5 lines
Quoted 2 lines

Thanks,
-Jeff

It looks like it's not patching very well in light of Dave's CRC userspace changes, and I botched the mkfs part of the merge.

Yes, that is based on the older version of xfsprogs with out CRC changes applied.

So we have to wait until an updated version is delivered. :)

Thanks,
-Jeff

Quoted 20 lines

For xfsprogs include/xfs_sb.h, Patch #1 has this:

+#define XFS_SB_VERSION2_NO_OQUOTA 0x00000100 /* sep prj quota
inode */

However, in the current xfsprogs include/xfs_sb.h, there is this line:

#define XFS_SB_VERSION2_CRCBIT 0x00000100 /* metadata CRCs */

This overloaded meaning of 0x00000100 did not seem right, so after my failure with the finished mkfs.xfs, it was time to go back to the current git xfsprogs. I'm sure that the userspace will be reworked before the kernel code is final. I'll show some patience, and Dave's upcoming CRC code activation will probably force me to back Chandra's patchset out anyway.

Despite my bumbling through it all, the kernel gquota/pquota patchset shows promise.

Quoted 8 lines

Thanks!

Michael

P.S. - I applied your xfs_growfs_data_private patch, and all seems well so far. In particular, there was no change in the test results from ./check -xfs -g growfs.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#6 lore raw
In reply to Michael L. Semon
Quoted 4 lines · Michael L. Semon

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

Hi Michael,

I will post the user level changes that accompanies these changes.

With those changes you can do a new mkfs and use pquota and gquota together.

Regards,

Chandra

Quoted 47 lines

The basic functionality works, in my opinion, and I hope nobody wastes time with a nice, educated reply. It would be mostly wasted on me and is better saved for somebody else. A reply of "do this...and this...btw, how did this test come out?" would be welcomed, though ;-)

Anyway, some vague observations as I grasp for straws...

1) The xfstests quota group tests seem to fail in different ways than the way they did before applying the patches.

2) Nothing has oopsed.

3) In testing using the xfs_quota -x command, the patches seem to work. On mount -t xfs -o gquota mounts, using the quota command from within the xfs_quota shell, the group quotas show but not the projid quotas. On mount -t xfs -o pquota mounts, the projid quotas show but not the gquota mounts. This is different than the old behavior, where the gquota numbers might be recycled into projid numbers.

4) The results of xfsquota -c print are confusing. Maybe they're showing the XFS view when they show things like 'uqnoenforce,gquota,pquota' for a mount that is gquota only. They're doubly confusing once /etc/projid and /etc/projects have been set up. The 'gqnoenforce' and 'pqnoenforce' flags show up at times for reasons that are unknown to me.

5) mount -t xfs -o gquota,pquota is not possible at this time.

6) The patches applied cleanly to a git Linux 3.10-rc1 kernel + xfs-oss, with only whitespace errors reported.

7) I question whether 'bsoft=' has a visible effect on projid quotas, whether using your patches or not. Did it ever work?

8) I had no feel on whether the filesystem had to be mounted once as gquota, then once as pquota, for the full dual functionality to work.

9) It looks like xfs_repair doesn't ruin anything, but the xfsquota -c print output looks a little different on the next mount.

That's about all that could be put together in a coherent manner. Sleep awaits.

The PC is a 32-bit Pentium 4. In addition to the kernel mentioned in (6), there are a few J. Liu and Dave Chinner patches applied as well.

Quoted 9 lines

Best of luck!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#9 lore raw
In reply to Michael L. Semon

On Tue, 2013-05-14 at 04:52 -0400, Michael L. Semon wrote:
Hi Michael,

Thanks for testing the code and providing feedback.

Quoted 10 lines

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

The basic functionality works, in my opinion, and I hope nobody wastes time with a nice, educated reply. It would be mostly wasted on me and is better saved for somebody else. A reply of "do this...and this...btw, how did this test come out?" would be welcomed, though ;-)

Quoted 4 lines

Anyway, some vague observations as I grasp for straws...

1) The xfstests quota group tests seem to fail in different ways than the way they did before applying the patches.

I ran the xfstests before and after these changes with the old tool chain and did not see any differences. Can you please tell me which tests failed.

Quoted 9 lines

2) Nothing has oopsed.

3) In testing using the xfs_quota -x command, the patches seem to work. On mount -t xfs -o gquota mounts, using the quota command from within the xfs_quota shell, the group quotas show but not the projid quotas. On mount -t xfs -o pquota mounts, the projid quotas show but not the gquota mounts. This is different than the old behavior, where the gquota numbers might be recycled into projid numbers.

Since we use different fields in superblock now, this should be expected.

Quoted 7 lines

4) The results of xfsquota -c print are confusing. Maybe they're showing the XFS view when they show things like 'uqnoenforce,gquota,pquota' for a mount that is gquota only. They're doubly confusing once /etc/projid and /etc/projects have been set up. The 'gqnoenforce' and 'pqnoenforce' flags show up at times for reasons that are unknown to me.

Can you please send me the exact commands you used. It will be good for me to test and add it to xfstests.

Quoted 2 lines

5) mount -t xfs -o gquota,pquota is not possible at this time.

yes. you need new xfstools to make it work.

Quoted 6 lines

6) The patches applied cleanly to a git Linux 3.10-rc1 kernel + xfs-oss, with only whitespace errors reported.

7) I question whether 'bsoft=' has a visible effect on projid quotas, whether using your patches or not. Did it ever work?

There are some xfstests that do test bsoft. Let me know if you think they are incorrect, we can revisit them.

Quoted 3 lines

8) I had no feel on whether the filesystem had to be mounted once as gquota, then once as pquota, for the full dual functionality to work.

No, you will be mounting filesystem as you did in (5) and it is supposed to work (wait for the new tools).

Quoted 3 lines

9) It looks like xfs_repair doesn't ruin anything, but the xfsquota -c print output looks a little different on the next mount.

Lets retest with tool changes.

Quoted 7 lines

That's about all that could be put together in a coherent manner. Sleep awaits.

The PC is a 32-bit Pentium 4. In addition to the kernel mentioned in (6), there are a few J. Liu and Dave Chinner patches applied as well.

Quoted 9 lines

Best of luck!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#5 lore raw
In reply to Jeff Liu
Quoted 13 lines · Jeff Liu
Quoted 12 lines · Michael L. Semon
Quoted 10 lines · Jeff Liu

Hi Michael,

Have you tried this patch set with the v2 user space changes?

Looks this is the latest update from Chandra that I can searched from the mailing list:
http://oss.sgi.com/archives/xfs/2012-07/msg00293.html

Thanks,
-Jeff

Quoted 6 lines
Quoted 2 lines

It looks like it's not patching very well in light of Dave's CRC userspace changes, and I botched the mkfs part of the merge.

Yes, that is based on the older version of xfsprogs with out CRC changes applied.

So we have to wait until an updated version is delivered. :)

There's a tarball that currently applies the CRC changes to the libxfs code in userspace. I'll post a new series out later today and you can rebase userspace on top of that.

Cheers,

Dave.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#7 lore raw
In reply to Chandra Seetharaman
Quoted 8 lines · Chandra Seetharaman
Quoted 4 lines · Michael L. Semon

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

Hi Michael,

Quoted 8 lines

I will post the user level changes that accompanies these changes.

With those changes you can do a new mkfs and use pquota and gquota together.

Regards,

Chandra

Excellent!

As for the rest of your issues, I'll have to retest very slowly using your new userspace patches, documenting every step in order. It will also require before/after runs of xfstests.

I do not know what your patches do for filesystems that already had quota on them, so I ask you to check one thing. It looked like the result of the normal quota command was having issues. On my root filesystem that was mounted uquota only, the following is the normal behavior (as an ordinary user):

mls:~$ quota
Disk quotas for user mls (uid 3001):
     Filesystem  blocks   quota   limit   grace   files   quota   limit 
  grace
      /dev/sda1  165955  262144  327680            2345    8194   16384

However, before I switched out kernels here, such a command just left a blank, as if there was no quota there. If you would at that to your list of spot-check tests, that would be great. It might work fine for you. [For the same situation, xfs_quota -c quota / worked just fine.]

You're leading a blind man in the dark, so I thank you for your patience.

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#8 lore raw

On Wed, 2013-05-15 at 17:56 -0400, Michael L. Semon wrote:
Hi Michael,

Irrespective of the user space changes, this patchset should work properly with the current tools and existing filesystems providing current behavior.

Let me know of the anomalies as the ones below.

Thanks

Chandra

Quoted 36 lines · Chandra Seetharaman
Quoted 8 lines · Michael L. Semon
Quoted 4 lines

Hi! I seem to have no luck in getting v8 of the pquota/uquota patchset working and have it pass xfstests with flying colors. Is v8 of the pquota/gquota patchset sufficient to make the new separate pquota/gquota bits work? Or is it an incremental patchset?

Hi Michael,

Quoted 8 lines

I will post the user level changes that accompanies these changes.

With those changes you can do a new mkfs and use pquota and gquota together.

Regards,

Chandra

Excellent!

As for the rest of your issues, I'll have to retest very slowly using your new userspace patches, documenting every step in order. It will also require before/after runs of xfstests.

I do not know what your patches do for filesystems that already had quota on them, so I ask you to check one thing. It looked like the result of the normal quota command was having issues. On my root filesystem that was mounted uquota only, the following is the normal behavior (as an ordinary user):

mls:~$ quota
Disk quotas for user mls (uid 3001):
      Filesystem  blocks   quota   limit   grace   files   quota   limit 
   grace
       /dev/sda1  165955  262144  327680            2345    8194   16384
Quoted 9 lines

However, before I switched out kernels here, such a command just left a blank, as if there was no quota there. If you would at that to your list of spot-check tests, that would be great. It might work fine for you. [For the same situation, xfs_quota -c quota / worked just fine.]

You're leading a blind man in the dark, so I thank you for your patience.

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs