[LU-9086] obd_config.c:1258:class_process_config()) no device for:

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.10.0
• Fix Version/s: Lustre 2.10.0
• Type: Bug
• Priority: Minor
• Reporter: Nathan Lavender (Inactive)
• Assignee: James A Simmons
• Resolution: Fixed

Environment

CentOS 7.3

• Severity: 3

Description

Secure RPC rules are being ignored due to an issue introduced by commit 0308951e8bba1663bc9495eee2f6ae631e6bcbd0 for LU-9034. When mounting Lustre resources, clients and servers log the following messages when secure RPC rules are set:

{{LustreError: 2696:0:(obd_config.c:1258:class_process_config()) no device for: lustre-ffff880078dd9e70
Lustre: 2696:0:(obd_config.c:1311:class_process_config()) Ignoring error -22 on optional command 0xce016}}

For my tests I set the following rules:
lustre.srpc.flavor.tcp.cli2mdt=ski
lustre.srpc.flavor.tcp.cli2ost=ski

then:

lustre.srpc.flavor.tcp.cli2mdt=krb5p
lustre.srpc.flavor.tcp.cli2ost=krb5p

All resources mount fine but cli2mdt and cli2ost connections aren't secured, they default to null.

Issue Links

• is related to: LU-9073 SSK: lgss_sk generates keys with inva...
• is related to: LU-9567 sptlrpc rules are not being updated
• is related to: LU-10937 Use sysfs to fix up sptlrpc handling.
• is related to: LU-9823 LNet fails to come up when using lctl...

Comments

[James A Simmons]

This bug is preventing me for fixing LU-9073

[Gerrit Updater]

James Simmons ([e-mail]) uploaded a new patch: https://review.whamcloud.com/25959
Subject: LU-9086 gss: handle specific mgc binding
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 4fc1de3721191d30658ef8f1830bae3a982a06f7

[James A Simmons]

The patch appears to fix this problem. Have you tried it?

[Nathan Lavender (Inactive)]

krb5p_multinode_llmount.tgz

The following error is logged multiple times on each of the nodes:

mgc_request.c:2143:mgc_process_log()) MGC172.18.0.18@tcp: configuration from log 'lustre-sptlrpc' failed (-2).

 

when SEC=krb5p and executing llmount.sh. Perhaps these logs will provide some clues as to why the rules are being ignored.

[James A Simmons]

So the question is now are the sptlrpc configurations per filesystem or per obd device. John asked the question about the case if we have two MGTs each with a filesystem named 'lustre' (and the client mounts both).

[Nathan Lavender (Inactive)]

It seems that sptlrpc rules are per filesystem. Setting a rule on one MGT affects the client's connection(s) to the other filesystem named 'lustre':

 

client# lctl get_param ..srpc_* | grep -A2 mdc.lustre
mdc.lustre-MDT0000-mdc-ffff88007884d800.srpc_info=
rpc flavor: null
bulk flavor: null
–
mdc.lustre-MDT0000-mdc-ffff88007a334000.srpc_info=
rpc flavor: null
bulk flavor: null

mgs1# lctl conf_param lustre.srpc.flavor.default.cli2mdt=krb5p

client# lctl get_param ..srpc_* | grep -A2 mdc.lustre
mdc.lustre-MDT0000-mdc-ffff88007884d800.srpc_info=
rpc flavor: krb5p
bulk flavor: krb5p
–
mdc.lustre-MDT0000-mdc-ffff88007a334000.srpc_info=
rpc flavor: krb5p
bulk flavor: krb5p

 

This test was conducted with Lustre v.2.9.0

[James A Simmons]

[root@ninja11 live]# cat lustre
fsname: lustre
flags: 0x20 gen: 24
lustre-MDT0000
lustre-OST0000

Secure RPC Config Rules:
lustre.srpc.flavor.o2ib1.cli2mdt=skpi

imperative_recovery_state:
state: startup
nonir_clients: 0
nidtbl_version: 4
notify_duration_total: 0.000527674
notify_duation_max: 0.000306105
notify_count: 2

It lives!!!! Will push now.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/25959/
Subject: LU-9086 gss: handle specific mgc binding
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 1b22b5141a22d471f70339320ea156c26db0ed37

[Peter Jones]

Landed for 2.10