[dpdk-dev] [PATCH v7 1/1] fbarray: fix duplicated fbarray file in secondary

[Yasufumi Ogawa]

Hi Anatoly,

On 2019/12/02 19:43, Burakov, Anatoly wrote:
> On 29-Nov-19 5:44 AM, Yasufumi Ogawa wrote:
>> Hi Anatoly,
>>
>> On 2019/11/27 19:26, Burakov, Anatoly wrote:
>>> On 26-Nov-19 7:40 PM, Yasufumi Ogawa wrote:
>>>> Hi David,
>>>>
>>>> Sorry for slow reply.
>>>>
>>>> On 2019/11/14 21:27, David Marchand wrote:
>>>>> On Thu, Nov 14, 2019 at 12:42 PM Yasufumi Ogawa 
>>>>> <yasufum.o at gmail.com> wrote:
>>>>>>
>>>>>> On 2019/11/14 2:01, Burakov, Anatoly wrote:
>>>>>>> On 13-Nov-19 9:43 PM, yasufum.o at gmail.com wrote:
>>>>>>>> From: Yasufumi Ogawa <ogawa.yasufumi at lab.ntt.co.jp>
>>>>>>>>
>>>>>>>> In secondary_msl_create_walk(), it creates a file for fbarrays 
>>>>>>>> with its
>>>>>>>> PID for reserving unique name among secondary processes. 
>>>>>>>> However, it
>>>>>>>> does not work if several secondaries run as app containers 
>>>>>>>> because each
>>>>>>>> of containerized secondary has PID 1, and failed to reserve 
>>>>>>>> unique name
>>>>>>>> other than first one. To reserve unique name in each of 
>>>>>>>> containers, use
>>>>>>>> hostname in addition to PID.
>>>>>>>>
>>>>>>>> Cc: stable at dpdk.org
>>>>>>>>
>>>>>>>> Signed-off-by: Yasufumi Ogawa <yasufum.o at gmail.com>
>>>>>>>> ---
>>>>>>>>    lib/librte_eal/linux/eal/eal_memalloc.c | 16 +++++++++++++---
>>>>>>>>    1 file changed, 13 insertions(+), 3 deletions(-)
>>>>>>>>
>>>>>>>> diff --git a/lib/librte_eal/linux/eal/eal_memalloc.c
>>>>>>>> b/lib/librte_eal/linux/eal/eal_memalloc.c
>>>>>>>> index af6d0d023..11de6d4d6 100644
>>>>>>>> --- a/lib/librte_eal/linux/eal/eal_memalloc.c
>>>>>>>> +++ b/lib/librte_eal/linux/eal/eal_memalloc.c
>>>>>>>> @@ -1365,6 +1365,12 @@ secondary_msl_create_walk(const struct
>>>>>>>> rte_memseg_list *msl,
>>>>>>>>        struct rte_memseg_list *primary_msl, *local_msl;
>>>>>>>>        char name[PATH_MAX];
>>>>>>>>        int msl_idx, ret;
>>>>>>>> +    char hostname[HOST_NAME_MAX+1] = { 0 };
>>>>>>>> +    /* filename of secondary's fbarray is defined such as
>>>>>>>> +     * "fbarray_memseg-1048576k-0-0_PID_HOSTNAME" and length of 
>>>>>>>> PID
>>>>>>>> +     * can be 7 digits maximumly.
>>>>>>>> +     */
>>>>>>>> +    int fbarray_sec_name_len = 32 + 7 + 1 + HOST_NAME_MAX + 1;
>>>>>>>
>>>>>>> What does 32 stand for? Maybe #define both 32 and 7 values?
>>>>>> Hi Anatoly,
>>>>>>
>>>>>> Thank you for your comments! If my understanding is correct, the 
>>>>>> prefix
>>>>>> "fbarray_memseg-1048576k-0-0_" is 28 digits and it could be larger if
>>>>>> using the size of hugepage or the number of NUMA nodes are larger
>>>>>> possibly. However, I think 32 digits is still enough.
>>>>>>
>>>>>>   > Maybe #define both 32 and 7 values?
>>>>>> Yes. I think it should be better to use #define if this values are
>>>>>> referred several times.
>>>>>
>>>>>
>>>>> We can truncate to RTE_FBARRAY_NAME_LEN in all cases.
>>>>> And iiuc, rte_fbarray_init will refuse any longer name anyway.
>>>> Could I confirm the issue? I've understood that it is failed to 
>>>> validate the name of fbarray in fully_validate() at 
>>>> "lib/librte_eal/common/eal_common_fbarray.c:697".
>>>>
>>>> static int
>>>> fully_validate(const char *name, unsigned int elt_sz, unsigned int len)
>>>> {
>>>>          if (name == NULL || elt_sz == 0 || len == 0 || len > 
>>>> INT_MAX) {
>>>>                  rte_errno = EINVAL;
>>>>                  return -1;
>>>>          }
>>>>
>>>>          if (strnlen(name, RTE_FBARRAY_NAME_LEN) == 
>>>> RTE_FBARRAY_NAME_LEN) {
>>>>                  rte_errno = ENAMETOOLONG;
>>>>                  return -1;
>>>>          }
>>>>          return 0;
>>>> }
>>>>
>>>> I should overwrite the definition of RTE_FBARRAY_NAME_LEN as 
>>>> previous patch in this case, and it causes an ABI breakage, right? 
>>>> If so, I would like to make the change and give up to update stable 
>>>> release.
>>>>
>>>> Thanks,
>>>> Yasufumi
>>>>
>>>
>>> It seems we're getting into bikeshedding...
>>>
>>> We can do this without ABI breakage. You could have just used 
>>> RTE_FBARRAY_NAME_LEN as max fbarray name length for 
>>> fbarray_sec_name_len (i.e. that would include hostname + pid + 
>>> whatever else there is). The name, as David has pointed out, would be 
>>> truncated to RTE_FBARRAY_NAME_LEN anyway (or, more precisely, it will 
>>> be refused if it's longer than that), so this is the most you can 
>>> have - so you can just use that as the maximum.
>> I sent v8 patch to change the size of RTE_FBARRAY_NAME_LEN itself to 
>> be allowed the size of secondary's fbarray over 64 bytes. I appreciate 
>> if you agree that.
>>
> 
> Why not just limit the name to RTE_FBARRAY_NAME_LEN instead of changing 
> the definition of RTE_FBARRAY_NAME_LEN?

Could I confirm my understanding? I understand that RTE_FBARRAY_NAME_LEN 
is 64 currently and it is not enough for secondary in a container if 
hostname is added to the name of secondary's fbarray.

Regards,
Yasufumi

> 
> One the other hand, technically, fbarray API is experimental. The only 
> structure that uses rte_fbarray is rte_memseg_list, but API's using 
> either rte_fbarray or rte_memseg_list are either internal (memory/VFIO 
> subsystem), or are marked as experimental (walk functions).
> 
> So i *think* we're actually OK with changing the length of 
> RTE_FBARRAY_NAME_LEN as far as ABI policy goes: nothing in the stable 
> ABI gets affected. David, thoughts?
> 
> (i think it's probably time to make experimental memory/fbarray stuff 
> stable, but that's a different conversation...)
> 
>> Thanks,
>> Yasufumi
>>
> 
>