[LU-1330] split/cleanup for obdclass/fid/fld modules

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.4.0, Lustre 2.5.0
• Fix Version/s: Lustre 2.4.0, Lustre 2.6.0
• Type: Task
• Priority: Minor
• Reporter: Xuezhao Liu
• Assignee: John Hammond
• Resolution: Fixed
• Labels: llite, patch

Environment

Linux client

• Epic: client

Description

Clean up Lustre client to facilitate pushing it into upstream Linux kernel. Need to split some server code dependencies from client to avoidmaking extensive server code changes.

There are some client-server shared modules that contain both client-side and server-side codes. For pushing client-side code to upstread linux kernel, we need to do some split/cleanup work.

This ticket is to split/cleanup obdclass/fid/fld's server-side handling from client-side.

Issue Links

• is related to: LU-3346 The assignment of obt_sb was removed

Comments

[Xuezhao Liu]

seven patches submitted:
splits client-server mount routines http://review.whamcloud.com/2672
fid: splits server-side handlings from client http://review.whamcloud.com/2673
obdecho: splits metadata echo client from client-side http://review.whamcloud.com/2674
fld: splits server-side handlings from client http://review.whamcloud.com/2675
obdclass: splits server-side object stack from client http://review.whamcloud.com/2676
obdclass: splits server-side procfs handling http://review.whamcloud.com/2677
obdclass: splits obd.h http://review.whamcloud.com/2678

Those patch were generated based on LU-1214's patch and 2 LU-1347's patch

[Gregoire Pichon]

As part of the patch http://review.whamcloud.com/2678, I see you removes the llog_catinfo code. Do you plan to also remove the client part of this code: "lfs catinfo" command, lfs_catinfo() routine, ... ?

Currently, running the "lfs catinfo config ." command crashes the system. The processing of the OBD_IOC_LLOG_CATINFO in ll_dir_ioctl() expects the obd device pointed by sbi->ll_md_exp->exp_obd to be a "client" device but it is a "lmv" device, which leads to a GPF in ptlrpc_request_alloc_internal().

thanks.

[Xuezhao Liu]

Hi Greoire,

Thanks your reminding. I forgot to remove that client-side routine.
I can change my patch to remove it. But I'am afraid that patch may need some time to be merged as it depend on several other patches.

I think it make sense to format another seperate patch to remove catinfo code(both client and server side). You may format and submit it as it is your idea, I can do it if it is in-convenient for you.

Thanks.

[Gregoire Pichon]

Hi,

I have created LU-1544 "lfs catinfo command crashes". I will post in this seperate ticket a patch that removes catinfo code (both client and server side).

Grégoire.

[John Hammond]

Thanks for breaking these up in to a series, but I'm not sure that it's the best strategy for Lustre's landing process. I'm working on refactoring them into changes that I think will have an easier time going through. Here is my first patch:

http://review.whamcloud.com/#change,5721 remove osd proc handlers from lprocfs_status.c

[John Hammond]

http://review.whamcloud.com/#change,5724 prepare fid module for client server split

[John Hammond]

http://review.whamcloud.com/#change,2673 prepare FID module for client server split - merged.
http://review.whamcloud.com/#/c/2675/ prepare FLD module for client server split - merge.

Several patches still in flight.

[Christopher Morrone]

Hi guys. Some of the recent LU-1330 landings have broken "make dist" with some configure options. For instance:

65d63035c LU-1330 fid: prepare FID module for client server split

When you make compilation of a source file optional, please remember to add it to EXTRA_DIST when the option is false. I have a patch to repair that and some other spots in another ticket:

http://review.whamcloud.com/7054

[Jodi Levi (Inactive)]

Patches have landed. If more work is needed please open a new ticket.