[LTP] [PATCHv3 3/3] network/stress/icmp: use ip xfrm for icmp4-uni-basic01 ipsec testing

[Alexey Kodanev <alexey.kodanev@oracle.com>]

Hi,
On 03/23/2016 04:30 PM, Hangbin Liu wrote:
> Hi,
> On Wed, Mar 23, 2016 at 11:21:37AM +0300, Alexey Kodanev wrote:
>> Hi,
>>>>> @@ -72,6 +65,7 @@ LINK_NUM=${LINK_NUM:-0}
>>>>>   # The version of IP
>>>>>   IP_VER=${IP_VER:-4}
>>>>> +[ $IP_VER -eq 6 ] && TST_IPV6=6
>>>> just "ipv=${TST_IPV6:-4}" instead of two lines.
>>> This is not work. Because test_net.sh will set TST_IPV6= , and icmp4-uni-basic01
>>> need source it before test. The relation looks like
>>>
>>> icmp6-uni-basic01
>>>    . icmp4-uni-basic01
>>>      . ipsec_lib.sh
>>>        . test_net.sh
>>>
>>> If we want to use like ipv=${TST_IPV6:-4}, we need source test_net.sh in all
>>> sub-testcases instead of in ipsec_lib.sh. Which will like
>>>
>>> icmp6-uni-basic01
>>>    . test_net.sh
>>>    . icmp4-uni-basic01
>>>      . ipsec_lib.sh
>> Could you add "ipv=..." to ipsec_lib.sh and we wouldn't set it in every
>> test-case?
> the IP_VER is to describe what the test run for. We shouldn't add it in
> ipsec_lib.sh. There is no relations between them.

Looking at the tests structure there... basically, one file implements 
test-cases
and the rest scripts source the first one along with changing a few 
parameters.
I think it's better to leave only one file there and remove the rest.

icmp4-uni-basic01 -> icmp-uni-basic.

May be moving them all to the upper directory, so in the end we should 
have only 3 tests:
icmp_multi_diffip.sh
icmp_multi_diffnic.sh
icmp_uni_basic.sh

Then, we could define each test-case in "runtests/", using env vars and 
parameters.

>
>> BTW, does the attached patch help in the first case (added export to
>> TST_IPV6)?
> I'm afraid not. I'd tried to use like
>
> icmp4-uni-basic01 icmp4-uni-basic01
> icmp6-uni-basic01 icmp4-uni-basic01 -6
>
> Then I find this is not only different of IP version, ipsec protocol, mode. But
> also different message sizes. So I give up to use like that.
>
>>
>>>> -# Run a client
>>>> -$LTP_RSH $RHOST "${LTPROOT}/testcases/bin/ns-echoclient -S $lhost_addr -f $IP_VER -s \"$ICMP_SIZE_ARRAY\"" &
>>>> -
>>>> -sleep $NS_DURATION
>>>> -killall_icmp_traffic
>>>> -wait
>>>> +# Make sure the connectvity
>>>> +for msg_size in $ICMP_SIZE_ARRAY; do
>>>> +    tst_ping $lhost_ifname $rhost_addr $msg_size
>>>> +    if [ $? -ne 0 ]; then
>>>> +        tst_brkm TBROK "There is no IPv$IP_VER connectivity with msg_size $msg_size"
>>>> +    else
>>>> +        tst_resm TPASS "There has IPv$IP_VER connectivity with msg_size $msg_size"
>>>> +    fi
>>>> +done
>>>>
>>>> Is it really needed to ping with different message sizes? if yes, we
>>>> couldadd this
>>>> functionality to tst_ping().
>>> Yes, we need to make sure ipsec can handle all kinds of message size. But we
>>> could not make it in tst_ping because ah/esp and tunnel/transport have
>>> different header length. e.g. when there is no ipsec, the max playload is
>>> 65507. If we test ah + transport, the max playload is 65483. etc. It's hard to
>>> make tst_ping to handle all these scenarios.
>> I thought we could add variable length parameters, something like this
>>
>> tst_ping p1 ... pN $ICMP_SIZE_ARRAY
>>
>> tst_ping()
>> {
>>      local msg_sizes=${@:N+1}
>>
>>      for size in msg_sizes; do
>>      ...
>> }
> I think tst_ping() is a unit test to check the connectivity, we do not need to
> add everything inside. Most time we only need to make sure the rhost is
> online. Let's make the test case deside what message size they want to test.
>
> How to you think?

I still think we should add it to the library. tst_ping replaces
'icmp4/6_check_connectivity'. And adding variable message size there,
will replace ns-echoclient library tool.

I don't understand why do you want to implement it in the each icmp test?

Thanks,
Alexey