[LTP] [PATCH] unshare01.sh: Setup parent mount flag before unshare testing

[Xiao Yang <yangx.jy@cn.fujitsu.com>]

On 2021/2/27 17:46, zhaogongyi wrote:
> Hi Yang,
>
>>            6) unshare_test "--mount --propagation shared" \
>>                         "mount --bind dir_A dir_B" "mounted";;
>> For example:
>> # 6) If we run with "--mount --propagation shared" and parent mount
>> namespace is shared, mount and unmount events can do propagation. "
>> # 7) If we run with "--mount --propagation shared" and parent mount
>> namespace is not shared, mount and unmount events cannot do
>> propagation. "
> 	In test 6, when we run with "--mount --propagation shared" and parent mount namespace is not shared,
> 	there are 3 steps:
> 		1) unshare --mount --propagation shared
> 		2) mount --bind dir_A dir_B
> 		3) exit the namespace
>
> 	So the test has no sence sinc step 3) has exit the namespace.
Hi Zhongyi,

Why do you think test 6 with private parent mount namespace has no 
sence?  I think the checkpoint is not related to step 3).
It indicates that mount/umount event cannot be propagated to private 
parent mount namespace even if the child mount namespace is shared.
For test 6 with shared parent mount namespace, mount/umount event has 
been propagated before the child mount namespace exits.

Best Regards,
Xiao Yang
> Thanks!
>
> =============================================================
>> On 2021/2/25 18:02, zhaogongyi wrote:
>>> Hi Yang,
>>>
>>>> I don't like the approach which enforces mountpoint to be shared in
>>>> parent mount namespace.
>>>> I think we can tune expected value by checking propagation flag in
>>>> parent mount namespace because of two reasons:
>>>> 1) Make test cover more cases.
>>>> 2) Don't depend on the fixed tmpfs.
>>> If we have no fixed parent mount namespace, the test looks like will pass
>> at any cases since we judge result by "grep -w 'dir_B' /proc/self/mountinfo
>> | grep -qw 'shared'".
>> Hi Zhongyi,
>>
>> This issue is caused by the state(e.g. share or private) of parent mount
>> namespace and both of results are expected.
>>
>>> It seems a bit tangential to our test objective as " # 6) If we run with
>> "--mount --propagation shared", mount and unmount events propagate to
>> its parent mount namespace. "
>> We also need to update the description of #6 because it is true only when
>> the parent mount namespace is shared.
>> For example:
>> # 6) If we run with "--mount --propagation shared" and parent mount
>> namespace is shared, mount and unmount events can do propagation. "
>> # 7) If we run with "--mount --propagation shared" and parent mount
>> namespace is not shared, mount and unmount events cannot do
>> propagation. "
>>
>> Best Regards,
>> Xiao Yang
>>> Thanks!
>>>
>>>
>> =========================================================
>> =
>>>> Hi Zhongyi, Petr
>>>>
>>>> I don't like the approach which enforces mountpoint to be shared in
>>>> parent mount namespace.
>>>> I think we can tune expected value by checking propagation flag in
>>>> parent mount namespace because of two reasons:
>>>> 1) Make test cover more cases.
>>>> 2) Don't depend on the fixed tmpfs.
>>>>
>>>> Zhongyi,  could you test the following patch on your enviorment?
>>>> ---------------------------------------------------------------------
>>>> ---------------------------- diff --git
>>>> a/testcases/commands/unshare/unshare01.sh
>>>> b/testcases/commands/unshare/unshare01.sh
>>>> index bf163a7f4..78ea83fc0 100755
>>>> --- a/testcases/commands/unshare/unshare01.sh
>>>> +++ b/testcases/commands/unshare/unshare01.sh
>>>> @@ -40,6 +40,17 @@
>>>> max_mntns_path="/proc/sys/user/max_mnt_namespaces"
>>>>     default_max_userns=-1
>>>>     default_max_mntns=-1
>>>>
>>>> +parse_propagation_flag()
>>>> +{
>>>> +       mount --bind dir_A dir_B
>>>> +       if grep -w 'dir_B' /proc/self/mountinfo | grep -qw 'shared';
>> then
>>>> +               echo "mounted"
>>>> +       else
>>>> +               echo "unmounted"
>>>> +       fi
>>>> +       umount dir_B
>>>> +}
>>>> +
>>>>     setup()
>>>>     {
>>>>            # On some distributions(e.g RHEL7.4), the default value of
>>>> @@
>>>> -149,7 +160,8 @@ do_test()
>>>>            4) unshare_test "--user --map-root-user" "id -g" "0";;
>>>>            5) unshare_test "--mount" "mount --bind dir_A dir_B"
>>>> "unmounted";;
>>>>            6) unshare_test "--mount --propagation shared" \
>>>> -                       "mount --bind dir_A dir_B" "mounted";;
>>>> +                       "mount --bind dir_A dir_B" \
>>>> +                       "$(parse_propagation_flag)";;
>>>>            7) unshare_test "--user --map-root-user --mount" \
>>>>                            "mount --bind dir_A dir_B"
>> "unmounted";;
>>>>            8) unshare_test "--user --map-root-user --mount
>>>> --propagation shared" \
>>>> --
>>>> ---------------------------------------------------------------------
>>>> ---------------------
>>>>
>>>> Best Regards,
>>>> Xiao Yang
>>>> On 2021/2/24 9:40, Petr Vorel wrote:
>>>>> Hi,
>>>>>
>>>>>> We need setup parent mount flag to shared before unshare testing,
>>>>>> or it will fail for system which has no systemd service since the
>>>>>> propagation flag is changed by systemd. From man 7
>>>> mount_namespaces.
>>>>> Do I understand correctly that all distros without systemd are
>>>>> affected, because systemd "automatically remounts all mount points
>>>>> as MS_SHARED on system startup" and test expect it?
>>>>>
>>>>>> Signed-off-by: Zhao Gongyi<zhaogongyi@huawei.com>
>>>>>> ---
>>>>>>     testcases/commands/unshare/unshare01.sh | 9 ++++++++-
>>>>>>     1 file changed, 8 insertions(+), 1 deletion(-) diff --git
>>>>>> a/testcases/commands/unshare/unshare01.sh
>>>>>> b/testcases/commands/unshare/unshare01.sh
>>>>>> index bf163a7f4..e1fb15035 100755
>>>>>> --- a/testcases/commands/unshare/unshare01.sh
>>>>>> +++ b/testcases/commands/unshare/unshare01.sh
>>>>>> @@ -31,7 +31,6 @@ TST_SETUP=setup
>>>>>>     TST_CLEANUP=cleanup
>>>>>>     TST_TESTFUNC=do_test
>>>>>>     TST_NEEDS_ROOT=1
>>>>>> -TST_NEEDS_TMPDIR=1
>>>>> You still need TST_NEEDS_TMPDIR=1, because you create files and
>>>> directories.
>>>>> Also your patch breaks bind test on very old systems (kernel 2.6,
>>>>> util-linux 2.17.2, glibc 2.12):
>>>>> unshare01 5 TFAIL: unshare --mount mount --bind dir_A dir_B got
>> bind
>>>>> info
>>>>>
>>>>> Any idea why (how to avoid this regression)?
>>>>>
>>>>>>     TST_NEEDS_CMDS="unshare id mount umount"
>>>>>>     . tst_test.sh
>>>>>> @@ -39,6 +38,7 @@
>>>> max_userns_path="/proc/sys/user/max_user_namespaces"
>>>>>>     max_mntns_path="/proc/sys/user/max_mnt_namespaces"
>>>>>>     default_max_userns=-1
>>>>>>     default_max_mntns=-1
>>>>>> +CURR=$(pwd)
>>>>> Instead of $CURR, cd - can be used.
>>>>>
>>>>>>     setup()
>>>>>>     {
>>>>>> @@ -55,6 +55,10 @@ setup()
>>>>>>     		echo 1024>    "${max_mntns_path}"
>>>>>>     	fi
>>>>>> +	mkdir $CURR/dir_C
>>>>> just mkdir dir_C
>>>>>> +	mount -t tmpfs none dir_C
>>>>>> +	mount --make-shared dir_C
>>>>> FYI We have tst_mount, but it'd not help much here.
>>>>>
>>>>>> +	cd dir_C
>>>>>>     	mkdir -p dir_A dir_B
>>>>>>     	touch dir_A/A dir_B/B
>>>>>>     }
>>>>>> @@ -66,6 +70,9 @@ cleanup()
>>>>>>     		echo ${default_max_userns}>    "${max_userns_path}"
>>>>>>     	[ ${default_max_mntns} -ne -1 ]&&    \
>>>>>>     		echo ${default_max_mntns}>    "${max_mntns_path}"
>>>>>> +	cd $CURR
>>>>>> +	umount dir_C
>>>>> tst_umount dir_C
>>>>>
>>>>>> +	rm -rf dir_C
>>>>> rm is not needed (cleanup is done automatically).
>>>>>>     }
>>>>>>     check_id()
>>>>> Full diff of changes I propose below.
>>>>>
>>>>> Kind regards,
>>>>> Petr
>>>>>
>>>>> diff --git testcases/commands/unshare/unshare01.sh
>>>>> testcases/commands/unshare/unshare01.sh
>>>>> index e1fb15035..0b5c56811 100755
>>>>> --- testcases/commands/unshare/unshare01.sh
>>>>> +++ testcases/commands/unshare/unshare01.sh
>>>>> @@ -31,6 +31,7 @@ TST_SETUP=setup
>>>>>     TST_CLEANUP=cleanup
>>>>>     TST_TESTFUNC=do_test
>>>>>     TST_NEEDS_ROOT=1
>>>>> +TST_NEEDS_TMPDIR=1
>>>>>     TST_NEEDS_CMDS="unshare id mount umount"
>>>>>     . tst_test.sh
>>>>>
>>>>> @@ -38,7 +39,6 @@
>>>> max_userns_path="/proc/sys/user/max_user_namespaces"
>>>>>     max_mntns_path="/proc/sys/user/max_mnt_namespaces"
>>>>>     default_max_userns=-1
>>>>>     default_max_mntns=-1
>>>>> -CURR=$(pwd)
>>>>>
>>>>>     setup()
>>>>>     {
>>>>> @@ -55,7 +55,7 @@ setup()
>>>>>     		echo 1024>    "${max_mntns_path}"
>>>>>     	fi
>>>>>
>>>>> -	mkdir $CURR/dir_C
>>>>> +	mkdir dir_C
>>>>>     	mount -t tmpfs none dir_C
>>>>>     	mount --make-shared dir_C
>>>>>     	cd dir_C
>>>>> @@ -70,9 +70,8 @@ cleanup()
>>>>>     		echo ${default_max_userns}>    "${max_userns_path}"
>>>>>     	[ ${default_max_mntns} -ne -1 ]&&    \
>>>>>     		echo ${default_max_mntns}>    "${max_mntns_path}"
>>>>> -	cd $CURR
>>>>> -	umount dir_C
>>>>> -	rm -rf dir_C
>>>>> +	cd ->/dev/null
>>>>> +	tst_umount dir_C
>>>>>     }
>>>>>
>>>>>     check_id()
>>>>>
>>>
>>> .
>>>
>>
>
>
> .
>