[LU-5564] open handle leak after setuid()

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.7.0
• Fix Version/s: Lustre 2.9.0
• Type: Bug
• Priority: Minor
• Reporter: John Hammond
• Assignee: Hongchao Zhang
• Resolution: Fixed
• Labels: mdt
• Severity: 3

Description

MDS_CLOSE RPCs are sent with the current_fs{uid,gid}(). If this id cannot be resolved on the MDT then mdt_close() fails in unpack and the open handle is never released.

# MOUNT_2=y llmount.sh
...
# id 42
id: 42: No such user
# cp /bin/sleep /mnt/lustre/sleep
# chown 42:42 /mnt/lustre/sleep
# chmod ug+s /mnt/lustre/sleep
# /mnt/lustre/sleep 0
# echo > /mnt/lustre2/sleep
-bash: /mnt/lustre2/sleep: Text file busy

[  475.032992] LustreError: 6559:0:(file.c:186:ll_close_inode_openhandle()) lustre-clilmv-ffff8802173141c8: inode [0x200000400:0x3:0x0] mdc close failed: rc = -13

Seems a bit broken this way. With the exception of HSM release why check process permissions at close? The open FD (open handle) gives us permission.

Comments

[Peter Jones]

Hongchao

Could you please look into this issue?

Thanks

Peter

[Gerrit Updater]

Hongchao Zhang ([e-mail]) uploaded a new patch: http://review.whamcloud.com/15534
Subject: LU-5564 llite: use original uid,gid to close file
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: b0ee3878dad8fa1d98b2d191983c0baa1f751b8d

[Hongchao Zhang]

status update:
the patch http://review.whamcloud.com/15534 has been updated

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/15534/
Subject: LU-5564 mdt: skip permission check for close
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: abedce670d4f396443164bb582c7b58ec78ae7b7

[Joseph Gmitter (Inactive)]

Patch has landed to master for 2.9.0

[Amit]

Dear All,

I am running into this exact same issue on our setup of this version of Lustre-2.7.20. We will be looking for upgrading to Lustre 2.10 in summer. Until then what would one recommend to avoid this issue from happening?

Is there a patch that I can apply in production?

Alternatively I was looking if I could mount the SSSD cache on the Lustre servers to a /tmpfs, in case if it helped.

Any comments are help here is greatly appreciated.

Thank you,
Amit

[Peter Jones]

Amit

Please direct that question through your support channels - this project is for matters relating to the community releases.

Peter