[LU-4279] .lustre and .lustre/fid cannot by accessed by FID

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.2
• Affects Version/s: Lustre 2.4.1
• Severity: 3
• Assignee: Bruno Faccini (Inactive)
• Reporter: Sebastien Buisson (Inactive)

Description

Hi,

We found a regression in the access by FID feature in Lustre 2.4. Indeed, stat by fid of .lustre/fid directory gets EINVAL.

This can be easily reproduced like this:

# lfs path2fid .lustre
[0x200000002:0x1:0x0]
# lfs path2fid .lustre/fid
[0x200000002:0x2:0x0]
# stat .lustre/fid/[0x200000002:0x2:0x0]
stat: cannot stat `<lustre mount point>/.lustre/fid/[0x200000002:0x2:0x0]': Invalid argument
# stat .lustre/fid/[0x200000002:0x1:0x0]
stat: cannot stat `<lustre mount point>/.lustre/fid/[0x200000002:0x1:0x0]': Invalid argument

The same commands just work fine on Lustre 2.1.6.

This is a big issue as Ganesha relies on this to work.

Thanks,
Sebastien.

Comments

[Nathaniel Clark]

backport of http://review.whamcloud.com/8360 to b2_5:
http://review.whamcloud.com/10490

[Peter Jones]

Thanks Aurelien. You are right that the fix version should have been set on this ticket. It is not uncommon for this step to be missed and we reconcile the fix version field against the got records for a given release ahead of announcing GA. This can also be considered for inclusion in 2.5.2

[Aurelien Degremont (Inactive)]

Hmm... it seems this ticket should be flagged as "Fixed in 2.6" ?

Any possibility to have it integrated in 2.5.2 also ?

[Bruno Faccini (Inactive)]

Thanks, and sure Happy New Year too, for you and your family!

[Sebastien Buisson (Inactive)]

Hi Bruno,

And Happy New Year!

Indeed it is ok now, we have backported your patch to 2.4, and this ticket can be closed now.

Cheers,
Sebastien.

[Bruno Faccini (Inactive)]

Hello Seb,
Thanks for your feedback.
So, now you tested it and patch has landed, do you agree if I close this ticket ??

[Sebastien Buisson (Inactive)]

Hi Bruno,

Thanks for the patch, I have tested it and it fixes the issue.

Cheers,
Sebastien.

[Bruno Faccini (Inactive)]

Master patch just pushed at http://review.whamcloud.com/8360.

[Bruno Faccini (Inactive)]

With latest master, error is still present :

# lfs path2fid /mnt/lustre/.lustre
[0x200000002:0x1:0x0]
# stat /mnt/lustre/.lustre/fid/[0x200000002:0x1:0x0]
stat: cannot stat `/mnt/lustre/.lustre/fid/[0x200000002:0x1:0x0]': Invalid argument
# 

when in Lustre full-debug log :

00000004:00000400:1.0:1384970440.693714:0:4765:0:(mdd_device.c:572:obf_lookup()) lustre-MDD0000: [0x200000002:0x1:0x0] is invalid, sequence should be >= 0x200000400 or within [0xc,0xffffffff].
00010000:00000010:1.1:1384970440.697923:0:4765:0:(ldlm_lock.c:453:lock_handle_free()) slab-freed 'lock': 504 at ffff880037ffec80.
00010000:00000010:1.1:1384970440.697927:0:4765:0:(ldlm_lock.c:453:lock_handle_free()) slab-freed 'lock': 504 at ffff880037ffea80.
00010000:00000010:1.1:1384970440.697928:0:4765:0:(ldlm_lock.c:453:lock_handle_free()) slab-freed 'lock': 504 at ffff880037ffe480.
00010000:00000010:1.1:1384970440.697930:0:4765:0:(ldlm_lock.c:453:lock_handle_free()) slab-freed 'lock': 504 at ffff8800377b3980.
00010000:00000010:1.1:1384970440.697931:0:4765:0:(ldlm_lock.c:453:lock_handle_free()) slab-freed 'lock': 504 at ffff88003746fdc0.
00010000:00000010:1.1:1384970440.697932:0:4765:0:(ldlm_lock.c:453:lock_handle_free()) slab-freed 'lock': 504 at ffff8800377b3180.
00000004:00000001:1.0:1384970440.697935:0:4765:0:(mdd_device.c:573:obf_lookup()) Process leaving via out (rc=18446744073709551594 : -22 : 0xffffffffffffffea)
00000004:00000001:1.0:1384970440.697937:0:4765:0:(mdt_handler.c:1353:mdt_getattr_name_lock()) Process leaving via out_parent (rc=18446744073709551594 : -22 : 0xffffffffffffffea)

[Sebastien Buisson (Inactive)]

Yes indeed, as this fix was landed for 2.4.