[LU-10910] LBUG with "lfs migrate -c 1 <domfile>"

• Type: Bug
• Resolution: Fixed
• Priority: Blocker
• Fix Version/s: Lustre 2.12.0
• Affects Version/s: Lustre 2.11.0
• Severity: 3
• Assignee: Mikhail Pershin
• Reporter: Andreas Dilger

Description

I created a test DoM file with:

# mkdir /mnt/testfs/dom
# lfs setstripe -E 64k -L mdt -E 64M -c 1 -E -1 -c -1 /mnt/testfs/dom/
# dd if=/dev/zero of=/mnt/testfs/dom/64M bs=1M count=64

This seemed to work properly (lfs getstripe returned all the right info. I then tried to see what would happen if I used "lfs migrate -c 1 /mnt/testfs/dom/64M" to migrate it to a regular file layout. This produced the following output on the console:

[1460451.829926] Lustre: lt-lfs: using old ioctl(LL_IOC_LOV_GETSTRIPE) on [0x200
000402:0x4e27:0x0], use llapi_layout_get_by_path()
[1460451.843549] LustreError: 14997:0:(ldlm_resource.c:1688:ldlm_resource_dump()
) --- Resource: [0x200000402:0x4e26:0x0].0x0 (ffff8800468a29c0) refcount = 4
[1460451.843611] LustreError: 14997:0:(ldlm_resource.c:1691:ldlm_resource_dump()
) Granted locks (in reverse order):
[1460451.843663] LustreError: 14997:0:(ldlm_resource.c:1694:ldlm_resource_dump()
) ### ### ns: testfs-MDT0000-mdc-ffff880007524000 lock: ffff88003e807a80/0xdc6f0
65983eec99d lrc: 1/0,0 mode: PR/PR res: [0x200000402:0x4e26:0x0].0x0 bits 0x1b/0
x0 rrc: 5 type: IBT flags: 0x20000000000 nid: local remote: 0xdc6f065983eec9ab e
xpref: -99 pid: 12539 timeout: 0 lvb_type: 3
[1460451.843720] Pid: 14997, comm: ptlrpcd_01_01
[1460451.843721] 
Call Trace:
[1460451.843757]  [<ffffffffc051e7ae>] libcfs_call_trace+0x4e/0x60 [libcfs]
[1460451.843761]  [<ffffffffc051e7e6>] libcfs_debug_dumpstack+0x26/0x30 [libcfs]
[1460451.843778]  [<ffffffffc0bb0c2b>] mdc_get_lock_handle+0xcb/0xe0 [mdc]
[1460451.843782]  [<ffffffffc0bb1060>] mdc_req_attr_set+0x90/0x170 [mdc]
[1460451.843854]  [<ffffffffc06cfae0>] cl_req_attr_set+0x60/0x150 [obdclass]
[1460451.843874]  [<ffffffffc0b22283>] osc_build_rpc+0x483/0x1070 [osc]
[1460451.843885]  [<ffffffffc0b3c6f0>] osc_io_unplug0+0xb50/0x1920 [osc]
[1460451.843891]  [<ffffffffc0b177d3>] brw_queue_work+0x33/0xd0 [osc]
[1460451.843988]  [<ffffffffc08efa47>] work_interpreter+0x37/0xf0 [ptlrpc]
[1460451.844018]  [<ffffffffc08ec95e>] ptlrpc_check_set.part.22+0x47e/0x1da0 [ptlrpc]
[1460451.844058]  [<ffffffffc08ee2db>] ptlrpc_check_set+0x5b/0xe0 [ptlrpc]
[1460451.844090]  [<ffffffffc091b15b>] ptlrpcd_check+0x4ab/0x590 [ptlrpc]
[1460451.844119]  [<ffffffffc091b549>] ptlrpcd+0x309/0x550 [ptlrpc]
[1460451.844151]  [<ffffffff810b099f>] kthread+0xcf/0xe0
[1460451.844167] LustreError: 14997:0:(mdc_dev.c:1345:mdc_req_attr_set()) page@ffff88005fb9da00[2 ffff880115c481c0 3 2           (null)]

[1460451.844222] LustreError: 14997:0:(mdc_dev.c:1345:mdc_req_attr_set()) vvp-page@ffff88005fb9da50(0:0) vm@ffffea0001cbbc80 1fffff0008006c 3:1 0 34275811373 lru

[1460451.844276] LustreError: 14997:0:(mdc_dev.c:1345:mdc_req_attr_set()) lov-page@ffff88005fb9da90, comp index: 0, gen: 5

[1460451.844332] LustreError: 14997:0:(mdc_dev.c:1345:mdc_req_attr_set()) osc-page@ffff88005fb9dac8 0: 1< 0x845fed 257 0 + + > 2< 0 0 4096 0x7 0x109 |           (null) ffff880043b384d0 ffff880051564c80 > 3< 1 0 0 > 4< 0 0 8 4202496 - | - - - + > 5< - - - + | 0 - | 0 - ->
[1460451.844416] LustreError: 14997:0:(mdc_dev.c:1345:mdc_req_attr_set()) end page@ffff88005fb9da00
[1460451.844484] LustreError: 14997:0:(mdc_dev.c:1345:mdc_req_attr_set()) uncovered page!
[1460451.844535] LustreError: 14997:0:(mdc_dev.c:1346:mdc_req_attr_set()) LBUG
[1460451.844581] Pid: 14997, comm: ptlrpcd_01_01

For the short term, "lfs migrate" should just return -EOPNOTSUPP from the kernel if called on a file with an mdt component. That way, lfs_migrate (the script version) can fall back to "copy and rename" to do the migration in userspace, which will work at the expense of changing the inode number.

Issue Links

• is related to: LU-10177 DoM: manual migration MDT-OST
• is related to: LU-11421 DoM: manual migration OST-MDT, MDT-MDT