[LU-10039] ioctl error after downgrade

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.10.0, Lustre 2.10.1
• Severity: 3
• Assignee: Zhenyu Xu
• Reporter: Andreas Dilger

Description

Took following steps to reproduce the issue and the issue was not reproduced.
Steps for both zfs and ldiskfs:
1. Clients and Servers with b_ieel3_0 build 222. Created Lustre file system.
2. Unmounted and upgraded all clients and Servers to 2.10.1. Mounted lustre again.
3. Unmounted Clients and Downgraded them to b_ieel3_0 build 222. Mounted Lustre again when nodes were up.
4. Unmounted MDS and downgraded it to b_ieel3_0 build 222. Mounted lustre and it worked this time.
5. Ran sanity now and the test ended abruptly after the completion of test_117 for both zfs and ldiskfs.

When I checked the states of nodes, MDT and one client were unmounted, OSTs and 2nd client were mounted.

Results:
ldiskfs - https://testing.hpdd.intel.com/test_sessions/df43fbf4-a2fb-11e7-b786-5254006e85c2
zfs - https://testing.hpdd.intel.com/test_sessions/335a5516-a2fd-11e7-bb19-5254006e85c2

Issue Links

• is related to: LU-10027 Unable to finish mount on MDS while Rolling downgrade

Comments

[Saurabh Tandan (Inactive)]

Re-ran the sanity using exact same steps as above and sanity run did not ended abruptly this time. Before running sanity I ran individual tests i.e. test_27A and test_65i which failed in previous run, but they passed when ran alone in second run. Results for second run are as follows:
ldiskfs- https://testing.hpdd.intel.com/test_sessions/d735f45c-a40a-11e7-b786-5254006e85c2
zfs- https://testing.hpdd.intel.com/test_sessions/0a221f12-a40b-11e7-b786-5254006e85c2

[Andreas Dilger]

After discussion with Saurabh, it seems that the user tools (in particular lfs) were not downgraded with the kernel modules. This is generating errors when using lfs setstripe in sanity test_27A, which is setting the default layout on the root directory:

error on ioctl 0x4008669a for '/mnt/lustre' (3): Invalid argument
error: setstripe: create stripe file '/mnt/lustre' failed
error on ioctl 0x4008669a for '/mnt/lustre' (3): Invalid argument
error: setstripe: create stripe file '/mnt/lustre' failed

The ioctl is LL_IOC_LOV_SETSTRIPE (0x4000 >> (_IOC_DIRSHIFT = 14) = _IOC_WRITE, 0x08 = sizeof(long), 0x66 = 'f', 0x9a = 154). However, it isn't clear why the kernel is complaining about the struct lov_mds_md that is being sent. Is this a bad magic, or something else? Since we aren't using composite layouts for these files, it should be compatible with existing tools.