On 20/07/2021 17:05, Krzysztof Kozlowski wrote:

The failure of memcg_stress_test.sh cleanup went unnoticed (except
echo message) and caused further cgroup_fj and memcg_control_test
failures because of unclean cgroups. Flow is usually like:

1. memcg_stress_test succeeds but actually the /dev/memcg was not
   cleaned up. Notice lack of any error message due to 2>/dev/null.

     memcg_stress_test 1 TINFO: Testing 150 cgroups, using 1273 MB, interval 5
     memcg_stress_test 1 TINFO: Starting cgroups
     memcg_stress_test 1 TINFO: Testing cgroups for 900s
     memcg_stress_test 1 TINFO: Killing groups
     tag=memcg_stress stime=1626770787 dur=903 exit=exited stat=2 core=no cu=19 cs=12

2. memcg_control_test has false-positive. It succeeds but actually no
   test was done due to /dev/memcg pollution from previous test:

     mkdir: cannot create directory ?/tmp/ltp-q8DjShPJeB/mnt/1?: File exists
     memcg_control    0  TINFO  :  Test #1: Checking if the memory usage limit imposed by the topmost group is enforced
     sh: echo: I/O error
     /opt/ltp/testcases/bin/memcg_control_test.sh: 86: /opt/ltp/testcases/bin/memcg_control_test.sh: cannot create /tmp/ltp-q8DjShPJeB/mnt/1/memory.memsw.limit_in_bytes:
     Permission denied
     rmdir: failed to remove 'sub': Device or resource busy
     rmdir: failed to remove '/tmp/ltp-q8DjShPJeB/mnt/1': Device or resource busy
     memcg_control    1  TPASS  :  memcg_control: passed
     tag=memcg_control stime=1626771695 dur=6 exit=exited stat=0 core=no cu=2 cs=1

3. cgroup_fj_function2_memory fails with a cryptic message of mounting a
   path with new line (because /dev/memcg was not cleaned up before):

     cgroup_fj_function2_memory 1 TINFO: Subsystem memory is mounted at /sys/fs/cgroup/memory
     /dev/memcg
     mkdir: cannot create directory ?/sys/fs/cgroup/memory?: File exists
     cgroup_fj_function2_memory 1 TBROK: mkdir /sys/fs/cgroup/memory
     /dev/memcg/ltp failed
     cgroup_fj_function2_memory 1 TINFO: Removing all ltp subgroups...
     find: ?/sys/fs/cgroup/memory\n/dev/memcg/ltp/?: No such file or directory

The actual failure was in memcg_stress_test executed before other tests,
however it went unnoticed.  Debugging such failures is difficult as
result of failing test depends on running another which did not fail.
Instead, detect unclean cgroups mounts and explicitly test it.

Signed-off-by: Krzysztof Kozlowski <redacted>
---
 .../kernel/controllers/cgroup_fj/cgroup_fj_common.sh   |  4 ++--
 .../controllers/memcg/control/memcg_control_test.sh    | 10 ++++++----
 .../controllers/memcg/stress/memcg_stress_test.sh      |  8 ++++----
 3 files changed, 12 insertions(+), 10 deletions(-)

Hi all,

Any comments here?

Best regards,
Krzysztof