[LU-6401] Untangle lustre userland and kernel headers

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.11.0
• Affects Version/s: Lustre 2.8.0
• Labels: patch upstream

Epic/Theme

upstream

• Assignee: Ben Evans (Inactive)
• Reporter: James A Simmons

Description

Much like the current situation with libcfs the headers for lustre being used internally for the kernel and for userland utilites is heavly intertwined. This work will look to unravel this mess and create a clear separation between lustre kernel space and lustre user land.

Issue Links

• is blocked by: LU-6245 Untangle userland and kernel space support for libcfs
• is related to: LU-6202 clean up ioctl handling
• is related to: LU-6746 move IT_* definitions to lustre_idl.h
• is related to: LU-5000 lustre_idl.h compilation regression from ORNL-10
• is related to: LU-6450 most MAY_* constants defined in lustre_idl.h are unused
• is related to: LU-9301 Ubuntu1404 client sanity test_400a/b FAIL: fatal error: libcfs/types.h: No such file or directory
• is related to: LU-5011 lustre_idl.h again does not compile in user space
• is related to: LU-9128 remove OBD_OCD_VERSION-ed out code
• is related to: LU-8945 Stop building ptlrpc/layout.c in userland
• is related to: LU-9767 FS name checking got removed
• is related to: LU-9679 Prepare lustre for adoption into the linux kernel

Comments

[Cory Spitz]

8 vs 16 length issue reported at LU-9767.

[Chris Horn]

If that change was intentional then there's a bug with mkfs.lustre rename option that restricts a rename to 8 characters.

[Chris Horn]

Was it intentional to remove the 8 character filesystem name limit in commit 424e56abd685c5eabc6a276154dec1c031cf5044 ?

 

And if so, is there an LUDOC for that change?

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/26966/
Subject: LU-6401 uapi: change lustre_cfg.h into a proper UAPI header
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 4fedbc879571305a0d14a2caa989c8c306efc54c

[Jian Yu]

Thank you James!

[James A Simmons]

Last two patches have been updated that will finish this ticket !!!!

[John Hammond]

0 lustre/include/uapi/linux/lustre_ioctl.h <global>               65 #define OBD_MAX_IOCTL_BUFFER CONFIG_LUSTRE_OBD_MAX_IOCTL_BUFFER
1 lustre/obdclass/linux/linux-module.c     obd_ioctl_getdata     167 if (hdr.ioc_len > OBD_MAX_IOCTL_BUFFER) {
2 lustre/obdclass/linux/linux-module.c     obd_ioctl_getdata     169 hdr.ioc_len, OBD_MAX_IOCTL_BUFFER);
3 lustre/utils/liblustreapi.c              cb_migrate_mdt_init  3853 char raw[OBD_MAX_IOCTL_BUFFER] = {'\0'};
4 lustre/utils/liblustreapi.c              llapi_obd_fstatfs    4167 char raw[OBD_MAX_IOCTL_BUFFER] = {'\0'};
5 lustre/utils/obd.c                       jt_obd_lov_getconfig 2219 desc.ld_tgt_count =
                                                                     ((OBD_MAX_IOCTL_BUFFER-sizeof(data)-sizeof(desc)) /

In cb_migrate_mdt_init() and llapi_obd_fstatfs() use of OBD_MAX_IOCTL_BUFFER can easily be replaced by using dynamic allocation from obd_ioctl_pack().

lctl lov_getconfig should just be deleted. But if that's not politically feasible then all of the information it prints can be gotten from lov proc.

Then there are no more uses of OBD_MAX_IOCTL_BUFFER in userspace.

[Andreas Dilger]

Can you explain what is wrong with that parameter? I honestly don't think we'll change it any time soon, so it could be moved to an internal header. That said, what is wrong with exposing this constant to userspace?

[James A Simmons]

One outstanding patch and I have the final one in the works. Almost done but I'm down to one last issue. For lustre_ioctl.h which is an UAPI headers to have CONFIG_LUSTRE_OBD_MAX_IOCTL_BUFFER is a big no no. So I see two ways this can be handled. One is make OBD_MAX_IOCTL_BUFFER a constant or make this a module parameter. Which approach should be done?

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/24325/
Subject: LU-6401 uapi: turn lustre_param.h into a proper UAPI header
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 424e56abd685c5eabc6a276154dec1c031cf5044

[Gerrit Updater]

James Simmons ([e-mail]) uploaded a new patch: https://review.whamcloud.com/26966
Subject: LU-6401 uapi: change lustre_cfg.h into a proper UAPI header
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 5c019c4dc0f8e7b281b9d6751af98158b57a50d3