tcp: calc ssthresh on RTO according to RFC5681
ClosedPublic
Actions

Authored by rscheff on Feb 6 2024, 9:10 PM.

Details

Reviewers

rrs
cc
tuexen
jtl
thj
glebius
rgrimes

Group Reviewers

transport

Commits

rG32a6df57df3e: tcp: calculate ssthresh on RTO according to RFC5681

Summary

Cwnd only reflects flightsize when the first RTO
happens suddenly, but not during loss recovery.

RFC5681 states that FlightSize should be used to
adjust ssthresh to - and when an RTO happens during
FastRecovery or SACK Loss Recovery, cwnd is typically
mangled to dole out the expected retransmission or
new transmission - but not the actual FlightSize (pipe).

RFC5681 is also clear to not adjust ssthresh on
any subsequent back-to-back RTOs (only the first RTO).

This snippet of code would have to be replicated into
newreno.c, htcp.c, cubic.c, dctcp.c with the appropriate
adjustments.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 55831
Build 52720: arc lint + arc unit

Event Timeline

rscheff created this revision.Feb 6 2024, 9:10 PM

Herald added a reviewer: transport. · View Herald TranscriptFeb 6 2024, 9:10 PM

Herald added subscribers: melifaro, imp. · View Herald Transcript

rscheff requested review of this revision.Feb 6 2024, 9:10 PM

Harbormaster completed remote builds in B55831: Diff 133938.Feb 6 2024, 9:11 PM

expand rfc5681 rto ssthresh calc to all relevant places.
and also in rack

Harbormaster completed remote builds in B55843: Diff 133960.Feb 7 2024, 10:04 AM

tuexen accepted this revision.Feb 8 2024, 4:24 PM

This revision is now accepted and ready to land.Feb 8 2024, 4:24 PM

Closed by commit rG32a6df57df3e: tcp: calculate ssthresh on RTO according to RFC5681 (authored by rscheff). · Explain WhyFeb 8 2024, 6:55 PM

This revision was automatically updated to reflect the committed changes.

rscheff added a commit: rG32a6df57df3e: tcp: calculate ssthresh on RTO according to RFC5681.

Ifyou open a review and tag a person into that review you should wait at least 24 hours before you commit the code, giving that person time to respond.

Only technical comment I have is the type change on mss, do we consistently store this as uint32_t, or does this change introduce some place are using u_int for it and this code is using uint32_t? I know that for the present platforms these are the same, but the future may not forever hold that. Also, isnt mss really a 16 bit value? TCP header spec says type 2, 4 byte option, which I dont get either, as this is suppose to be derived from MTU and MTU is a 16 bit value by IP spec. Anyway already committed so Meh.

rscheff mentioned this in D43876: tcp: fix RTO ssthresh for non-6675 pipe calculation.Feb 14 2024, 12:40 PM

rscheff mentioned this in rGfcea1cc9711a: tcp: fix RTO ssthresh for non-6675 pipe calculation.Feb 14 2024, 1:58 PM

grahamperrin added a subscriber: grahamperrin.Feb 15 2024, 11:13 PM