Date:	Fri, 29 Jun 2012 15:15:12 -0400
From:	Vlad Yasevich <vyasevich@...il.com>
To:	Neil Horman <nhorman@...driver.com>
CC:	netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>
Subject: Re: [PATCH v3] sctp: be more restrictive in transport selection on
 bundled sacks

On 06/29/2012 02:43 PM, Neil Horman wrote:
> On Fri, Jun 29, 2012 at 02:29:52PM -0400, Vlad Yasevich wrote:
>> On 06/29/2012 12:34 PM, Neil Horman wrote:
>>> It was noticed recently that when we send data on a transport, its possible that
>>> we might bundle a sack that arrived on a different transport.  While this isn't
>>> a major problem, it does go against the SHOULD requirement in section 6.4 of RFC
>>> 2960:
>>>
>>>   An endpoint SHOULD transmit reply chunks (e.g., SACK, HEARTBEAT ACK,
>>>     etc.) to the same destination transport address from which it
>>>     received the DATA or control chunk to which it is replying.  This
>>>     rule should also be followed if the endpoint is bundling DATA chunks
>>>     together with the reply chunk.
>>>
>>> This patch seeks to correct that.  It restricts the bundling of sack operations
>>> to only those transports which have moved the ctsn of the association forward
>>> since the last sack.  By doing this we guarantee that we only bundle outbound
>>> saks on a transport that has received a chunk since the last sack.  This brings
>>> us into stricter compliance with the RFC.
>>>
>>> Vlad had initially suggested that we strictly allow only sack bundling on the
>>> transport that last moved the ctsn forward.  While this makes sense, I was
>>> concerned that doing so prevented us from bundling in the case where we had
>>> received chunks that moved the ctsn on multiple transports.  In those cases, the
>>> RFC allows us to select any of the transports having received chunks to bundle
>>> the sack on.  so I've modified the approach to allow for that, by adding a state
>>> variable to each transport that tracks weather it has moved the ctsn since the
>>> last sack.  This I think keeps our behavior (and performance), close enough to
>>> our current profile that I think we can do this without a sysctl knob to
>>> enable/disable it.
>>>
>>> Signed-off-by: Neil Horman<nhorman@...driver.com>
>>> CC: Vlad Yaseivch<vyasevich@...il.com>
>>> CC: David S. Miller<davem@...emloft.net>
>>> Reported-by: Michele Baldessari<michele@...hat.com>
>>> Reported-by: sorin serban<sserban@...hat.com>
>>>
>>> ---
>>> Change Notes:
>>> V2)
>>> 	* Removed unused variable as per Dave M. Request
>>> 	* Delayed rwnd adjustment until we are sure we will sack (Vlad Y.)
>>> V3)
>>> 	* Switched test to use pkt->transport rather than chunk->transport
>>> 	* Modified detection of sacka-able transport.  Instead of just setting
>>> 	  and clearning a flag, we now mark each transport and association with
>>> 	  a sack generation tag.  We increment the associations generation on
>>> 	  every sack, and assign that generation tag to every transport that
>>> 	  updates the ctsn.  This prevents us from having to iterate over a for
>>> 	  loop on every sack, which is much more scalable.
>>> ---
>>>   include/net/sctp/structs.h |    4 ++++
>>>   include/net/sctp/tsnmap.h  |    3 ++-
>>>   net/sctp/associola.c       |    1 +
>>>   net/sctp/output.c          |    9 +++++++--
>>>   net/sctp/sm_make_chunk.c   |   10 ++++++++++
>>>   net/sctp/sm_sideeffect.c   |    2 +-
>>>   net/sctp/transport.c       |    2 ++
>>>   net/sctp/tsnmap.c          |    6 +++++-
>>>   net/sctp/ulpevent.c        |    3 ++-
>>>   net/sctp/ulpqueue.c        |    2 +-
>>>   10 files changed, 35 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/include/net/sctp/structs.h b/include/net/sctp/structs.h
>>> index e4652fe..fecdf31 100644
>>> --- a/include/net/sctp/structs.h
>>> +++ b/include/net/sctp/structs.h
>>> @@ -912,6 +912,9 @@ struct sctp_transport {
>>>   		/* Is this structure kfree()able? */
>>>   		malloced:1;
>>>
>>> +	/* Has this transport moved the ctsn since we last sacked */
>>> +	__u32 sack_generation;
>>> +
>>>   	struct flowi fl;
>>>
>>>   	/* This is the peer's IP address and port. */
>>> @@ -1584,6 +1587,7 @@ struct sctp_association {
>>>   		 */
>>>   		__u8    sack_needed;     /* Do we need to sack the peer? */
>>>   		__u32	sack_cnt;
>>> +		__u32	sack_generation;
>>>
>>>   		/* These are capabilities which our peer advertised.  */
>>>   		__u8	ecn_capable:1,	    /* Can peer do ECN? */
>>> diff --git a/include/net/sctp/tsnmap.h b/include/net/sctp/tsnmap.h
>>> index e7728bc..2c5d2b4 100644
>>> --- a/include/net/sctp/tsnmap.h
>>> +++ b/include/net/sctp/tsnmap.h
>>> @@ -117,7 +117,8 @@ void sctp_tsnmap_free(struct sctp_tsnmap *map);
>>>   int sctp_tsnmap_check(const struct sctp_tsnmap *, __u32 tsn);
>>>
>>>   /* Mark this TSN as seen.  */
>>> -int sctp_tsnmap_mark(struct sctp_tsnmap *, __u32 tsn);
>>> +int sctp_tsnmap_mark(struct sctp_tsnmap *, __u32 tsn,
>>> +		     struct sctp_transport *trans);
>>>
>>>   /* Mark this TSN and all lower as seen. */
>>>   void sctp_tsnmap_skip(struct sctp_tsnmap *map, __u32 tsn);
>>> diff --git a/net/sctp/associola.c b/net/sctp/associola.c
>>> index 5bc9ab1..6c66adb 100644
>>> --- a/net/sctp/associola.c
>>> +++ b/net/sctp/associola.c
>>> @@ -271,6 +271,7 @@ static struct sctp_association *sctp_association_init(struct sctp_association *a
>>>   	 */
>>>   	asoc->peer.sack_needed = 1;
>>>   	asoc->peer.sack_cnt = 0;
>>> +	asoc->peer.sack_generation=0;
>>>
>>>   	/* Assume that the peer will tell us if he recognizes ASCONF
>>>   	 * as part of INIT exchange.
>>> diff --git a/net/sctp/output.c b/net/sctp/output.c
>>> index f1b7d4b..0de6cd5 100644
>>> --- a/net/sctp/output.c
>>> +++ b/net/sctp/output.c
>>> @@ -240,14 +240,19 @@ static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
>>>   	 */
>>>   	if (sctp_chunk_is_data(chunk)&&   !pkt->has_sack&&
>>>   	!pkt->has_cookie_echo) {
>>> -		struct sctp_association *asoc;
>>>   		struct timer_list *timer;
>>> -		asoc = pkt->transport->asoc;
>>> +		struct sctp_association *asoc = pkt->transport->asoc;
>>> +
>>>   		timer =&asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
>>>
>>>   		/* If the SACK timer is running, we have a pending SACK */
>>>   		if (timer_pending(timer)) {
>>>   			struct sctp_chunk *sack;
>>> +
>>> +			if (pkt->transport->sack_generation !=
>>> +			    pkt->transport->asoc->peer.sack_generation)
>>> +				return retval;
>>> +
>>>   			asoc->a_rwnd = asoc->rwnd;
>>>   			sack = sctp_make_sack(asoc);
>>>   			if (sack) {
>>> diff --git a/net/sctp/sm_make_chunk.c b/net/sctp/sm_make_chunk.c
>>> index a85eeeb..ffa2a8e 100644
>>> --- a/net/sctp/sm_make_chunk.c
>>> +++ b/net/sctp/sm_make_chunk.c
>>> @@ -736,6 +736,7 @@ struct sctp_chunk *sctp_make_sack(const struct sctp_association *asoc)
>>>   	__u16 num_gabs, num_dup_tsns;
>>>   	struct sctp_tsnmap *map = (struct sctp_tsnmap *)&asoc->peer.tsn_map;
>>>   	struct sctp_gap_ack_block gabs[SCTP_MAX_GABS];
>>> +	struct sctp_transport *trans;
>>>
>>>   	memset(gabs, 0, sizeof(gabs));
>>>   	ctsn = sctp_tsnmap_get_ctsn(map);
>>> @@ -805,6 +806,15 @@ struct sctp_chunk *sctp_make_sack(const struct sctp_association *asoc)
>>>   		sctp_addto_chunk(retval, sizeof(__u32) * num_dup_tsns,
>>>   				 sctp_tsnmap_get_dups(map));
>>>
>>> +	/*
>>> +	 * Once we have a sack generated, clear the moved_tsn information
>>> +	 * from all the transports
>>> +	 */
>>> +	if (!asoc->peer.sack_generation)
>>> +		list_for_each_entry(trans,&asoc->peer.transport_addr_list,
>>> +				    transports)
>>> +			trans->sack_generation = UINT_MAX;
>>> +	((struct sctp_association *)asoc)->peer.sack_generation++;
>>
>> Two points here:
>> 1) The commend no longer matches the code
> Crud, missed that, I'll fix it.
>
>> 2) Why special case the peer.sack_generations == 0 and set the
>> transport to UNIT_MAX?
>>
> To avoid wrapping problems leading to erroneous bundling errors.  Consider a
> long lived connection with two trasports (A and B).
>
> If all traffic is sent on A for a long time (generating UINT_MAX sacks), and the
> peer chooses that moment to send data on transport B, its possible that we will
> bundle a sack with that data chunk erroneously, because the associations
> sack_generation has wrapped, and now matches with the transports, even though we
> never received data on transport B.  The special casing ensures that we never
> hit that problem.
>

But you just move this condition to the UINT_MAX value instead.  If we 
use the alternate transport at the time that sack_generation == 
UINT_MAX, we may pick the wrong transport.

You may want to consider value 0 reserved as UNUSED and make 
peer.sack_generation start at 1 and wrap to 1.

-vlad
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists