[LU-5022] support for 3.10 rhel7 linux kernel

• Type: New Feature
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.8.0
• Affects Version/s: Lustre 2.6.0
• Labels: HB
• Environment: rhel7
• Assignee: Bob Glossman (Inactive)
• Reporter: Bob Glossman (Inactive)

Description

tracker for 3.10 kernel support in rhel7

The 3.10 kernel in the rhel7 rc2 image is quite different from the mainline upstream 3.10 in kernel.org. It includes many features from 3.12 and even 3.13 kernels. In general most of the recipe mentioned in LU-4416 for 3.12 kernels is needed in rhel7 too. The only 3.12 patch that isn't needed is http://review.whamcloud.com/9300.

Establishing a separate tracker that will be specific to rhel7 as there are some mods that will only be relevant to the rhel7 kernel, and not 3.12 or later. One example is ldiskfs support. A new patch series specific to rhel7 will be needed.

Issue Links

• is related to: LU-5626 Corruption of MDT “..” entry in non-HTree ldiskfs directories
• is related to: LU-6096 sanity test_17m: e2fsck Inode 32775, i_size is 0, should be 4096
• is related to: LU-5307 e2fsprogs build fails in Centos 7
• is related to: LU-5884 bad lnet conf causes LBUG
• is related to: LU-5310 build error in Centos 7
• is related to: LU-5014 vfs_unlink api change
• is related to: LU-5853 el7 build broken
• is related to: LU-5825 Kernel update [RHEL7 3.10.0-123.9.2.el7]
• is related to: LU-4416 support for 3.12 linux kernel

Comments

[Bob Glossman (Inactive)]

I am pretty sure all the needed changes from James' attachment are already in place in upstream quota code in el7. only 2 minor exceptions:

1) the dqptr_sem deleted by the first patch still exists. all the references to it are gone, but the data structure was left in place. I'm assuming whoever ported this into el7 did it deliberately so I'm not willing to change it.

2) the last patch adding in locking around the shrinker doesn't seem to apply or be relevant. there is no dqcache_shrink_scan() routine in el7 to be fixed up.

[James A Simmons]

Here are the quota upstream patches that the RHEL7 kernel needs. It might need some massaging to have it apply to the RHEL7 kernel.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/10249/
Subject: LU-5022 ldiskfs: enable support for RHEL7
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: eaf30e35002bf3936cdf7bf83685552d854f0f88

[Gerrit Updater]

Yang Sheng ([e-mail]) uploaded a new patch: http://review.whamcloud.com/13067
Subject: LU-5022 ldiskfs: cleanup ldiskfs_journal_dirty_metadata
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 4a0457a9e3559c872faa59c5361d84a86d8ee71c

[Gerrit Updater]

Yang Sheng ([e-mail]) uploaded a new patch: http://review.whamcloud.com/10249
Subject: LU-5022 ldiskfs: enable support for RHEL7
Project: fs/lustre-release
Branch: master
Current Patch Set: 20
Commit: d6683a2dd978dafa350e4e58c45df55cf44fe68f

[Minh Diep]

A closer look at the report, this patch hit a LBUG in https://testing.hpdd.intel.com/test_logs/16c19292-658a-11e4-8c86-5254006e85c2/show_text

[James A Simmons]

Oh I forgot but since SLES12 is now official we should make the RHEL7 ldiskfs patch use the patches from the LU-4416 SLES12 ldiskfs work like we originally did. Since RHEL7 changes far more often then SLES12 this can make sense plus this avoids the issue of SLES12 ldiskfs fixes being missed.

[James A Simmons]

Now that the client work is done we should really discuss what is needed for proper server side support. This is are chance to push for a pathless server. First we need the quota performance enhancements. Those have been merged upstream as:

commit b9ba6f94b2382ef832f97122976b73004f714714
commit 9eb6463f31cf720deaf0e810cacc403d7720b10c
commit 1ea06bec78a128adc995ca32bd906a6c9bb9cf91
commit 606cdcca04a609ed4dfbfe788942de9477da556b
commit d68aab6b8f572406aa93b45ef6483934dd3b54a6

Which should engage RedHat to have these merged into their tree.

We need to replace the dev_read_only patch with the work under LU-684. I haven't got around to it but now that the work for 2.7 is winding down I can take it up again.

The bh_lru_size patch. I saw something go upstream but it seems to have been drop. It is a simple one line patch which we should push to get merged. Once merged it will be easy for RedHat to incorporate it into their tree.

The block tunable patch. From the data posted in LU-20 I really think it is not a benefit and we should just drop it.

Lastly we have the patch under LU-3406 (raid5-mmp-unplug) which I just publish on dm-devel mailing list. Again we have it merged into the RHEL tree once it is integrated upstream.

[Bob Glossman (Inactive)]

lbuild changes for rhel 7 server
http://review.whamcloud.com/12289

[Bob Glossman (Inactive)]

minimum additional needed to build el7 servers:

http://review.whamcloud.com/#/c/8116 - LU-3373 ext4_map_block support
http://review.whamcloud.com/#/c/11088 - LU-5276 ldiskfs: Remove extents-mount-option patch
http://review.whamcloud.com/#/c/10249 - LU-5022 ldiskfs: enable support for RHEL7

e2fsprogs and lbuild changes for el7 server are still TBD.

[James A Simmons]

Yang can you post your RHEL7 test results. I like to compare them to my results from the SLES12/3.12 runs I do.