Skip to contentSkip to conversation

Fast retransmit fixes

RFC PATCH Wei Yongjun 2 participants 2 messages May 14, 2008 SCTP protocol

lore
Share & tools
lore
#0 lore raw

Hi Vlad:

This patch set cannot resolve Fast Recovery problem correctly.

Test Procedure:

Endpoint A Endpoint B

DATA (TSN = 1)  ------------->
DATA (TSN = 2)  -- (lost)---->
DATA (TSN = 3)  ---(lost)---->
DATA (TSN = 4)  ------------->
DATA (TSN = 5)  ------------->
                <-------------   SACK (CTSN = 1)
DATA (TSN = 6)  ------------->
                <-------------   SACK (CTSN = 1, GAP-START = 3, GAP-ENT = 3)
                <------------    SACK (CTSN = 1, GAP-START = 3, GAP-ENT = 4)
                <------------    SACK (CTSN = 1, GAP-START = 3, GAP-ENT = 5) 
DATA (TSN = 2)  -(fast rtx)-->
                <-------------   SACK (CTSN = 2, GAP-START = 2, GAP-ENT = 4)

DATA (TSN = 3) -(can not do fast rtx)--

I am not sure is this a bug of your patch because I used this patchset manually. Can you check for this test?

Wei Yongjun

Quoted 9 lines · Vlad Yasevich

This is an update to the Fast retransmit changes.

This fixes a small bug introduced by a prior series where the fast_rtx hint in the outq structure was never cleared.

All prior comments still stand.

#1 lore raw
Quoted 19 lines · Wei Yongjun

Hi Vlad:

This patch set cannot resolve Fast Recovery problem correctly.

Test Procedure:

Endpoint A Endpoint B

DATA (TSN = 1)  ------------->
DATA (TSN = 2)  -- (lost)---->
DATA (TSN = 3)  ---(lost)---->
DATA (TSN = 4)  ------------->
DATA (TSN = 5)  ------------->
               <-------------   SACK (CTSN = 1)
DATA (TSN = 6)  ------------->
               <-------------   SACK (CTSN = 1, GAP-START = 3, GAP-ENT = 3)
               <------------    SACK (CTSN = 1, GAP-START = 3, GAP-ENT = 4)
               <------------    SACK (CTSN = 1, GAP-START = 3, GAP-ENT = 
5) DATA (TSN = 2)  -(fast rtx)-->
               <-------------   SACK (CTSN = 2, GAP-START = 2, GAP-ENT = 4)
Quoted 5 lines

DATA (TSN = 3) -(can not do fast rtx)--

I am not sure is this a bug of your patch because I used this patchset manually. Can you check for this test?

Hi Wei

Yes, I forgot to remove the call to sctp_list_dequeue() in sctp_outq_flush_rtx(). That's what was causing this error.

There was only one other small bug in the patch that I found with this scenario. These all stem from me attempting to break on 1 big patch into 3 smaller ones.

Anyway, I'll repost.

Thanks
-vlad