James, can you please take a look.
I don't think SLES12 clients are in such heavy usage that they need to get the latest netlink functionality, but at one point at least Cray was heavily based on SLES for their client distro, so at least we shouldn't break it gratuitously. If there isn't a straight forward way to fix it, I'd be fine with just configuring out the netlink functionality and always using ioctl/debugfs in this case (which isn't worse than what was available before).
What version of libnl is installed? Do we have a special Test-parameter tag for SUSE12. I suspect that the libnl library is older so its lacking proper support for NLA_S64.
"James Simmons <jsimmons@infradead.org>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49608
Subject: LU-16462 utils: handle lack of NLA_S64
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: a6780268c84f876b98f8d31a6830dcb9acf16077
This is preventing patches on e2fsprogs from passing testing, since they run with sles12sp5 clients.
https://testing.whamcloud.com/test_sessions/96f8909f-bbaa-40f2-b969-f261d4b0398f
https://testing.whamcloud.com/test_sessions/1f946f9d-86de-401d-9c45-b3b445eb10b4
Comment from Dongyang in the 49608 patch that explains the issue:
lctl dl is triggering assert inside libnl3 because we use NLA_NUL_STRING.
in the old libnl3, we don't have NLA_NUL_STRING and NLA_S8|16|32|64:enum { NLA_UNSPEC, /**< Unspecified type, binary data chunk */ NLA_U8, /**< 8 bit integer */ NLA_U16, /**< 16 bit integer */ NLA_U32, /**< 32 bit integer */ NLA_U64, /**< 64 bit integer */ NLA_STRING, /**< NUL terminated character string */ NLA_FLAG, /**< Flag */ NLA_MSECS, /**< Micro seconds (64bit) */ NLA_NESTED, /**< Nested attributes */ __NLA_TYPE_MAX, }; #define NLA_TYPE_MAX (__NLA_TYPE_MAX - 1)and if we try to use any NLA_TYPE greater than NLA_TYPE_MAX, it will trigger the assert in validate_nla().
Do we have to use NLA_NUL_STRING instead of NLA_STRING, and the signed nla types?
Hi James, any thought on how to make progress on this issue?
We have e2fsck fixes blocked from landing for a couple of months because the netlink patch has broken "lctl dl" on SLES12 clients. I don't think we need to retroactively add support for SLES12 clients to allow non-root users to run "lctl dl", so it would be fine if the netlink code was completely disabled for older clients that don't have NLA_S32 or NLA_NUL_STRING and only the ioctl fallback was used. It just needs to not break the old code.
I'm going to push a patch that disables yaml netlink usage if NLA_NUL_STRING is not defined. This works fine for "lctl dl" in my local testing, but still need to fix "lctl ping" (sanity test_217).
"Andreas Dilger <adilger@whamcloud.com>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/50610
Subject: LU-16462 utils: skip netlink for old libnl3
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 5ff26d3c3205baacefb3c0e48c2a03ff0713db39
patch https://review.whamcloud.com/49608 "LU-16462 utils: handle lack of NLA_S64" has been updated to handle the sles12sp5 libnl incompatibility, along with test and tool fixes for the netlink-unavailable fallback case so that "lctl ping" and "lctl list_nids" continue to work.
Thank you Andreas for your help 
I just ran across an old patch from Amir that is replacing usage of "lctl ping" and "lctl list_nids" with the equivalent "lnetctl" commands.
The output is clunky and needs some awk to parse it into just a NID:
$ lnetctl net show
net:
- net type: lo
local NI(s):
- nid: 0@lo
status: up
- net type: tcp
local NI(s):
- nid: 192.168.10.99@tcp
status: up
interfaces:
0: enp0s3
$ lnetctl net show | awk '/nid:/ && $3 != "0@lo" { print $3 }'
192.168.10.99@tcp
Alexey suggested in that patch to put this into a helper function on test-framework.sh instead of having it inline in multiple places. However, users would probably also want to print some of these fields outside of the testing, instead of the full YAML.
Having a command-line argument like "lnetctl net show -nid" in this case, but also able to print other fields like "... --status", "nettype", "-interfaces") would more convenient than users having to use "awk" or "yq" to extract the fields manually.
"Oleg Drokin <green@whamcloud.com>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/49608/
Subject: LU-16462 utils: handle lack of newer nla_attrs
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ae1ee11cea0a90631e14d670883528d6ac6e86b7
Landed for 2.16