Page MenuHomeFreeBSD
Differential D43213

tcp: retry CWR transmission if TSO doesn't support ECN
Needs ReviewPublic
Actions

Authored by rscheff on Dec 27 2023, 2:21 PM.
Tags
None
Referenced Files
Unknown Object (File)
Mon, Feb 23, 9:54 PM
Unknown Object (File)
Mon, Feb 23, 9:53 PM
Unknown Object (File)
Mon, Feb 23, 9:53 PM
Unknown Object (File)
Mon, Feb 23, 9:53 PM
Unknown Object (File)
Sun, Feb 8, 11:03 AM
Unknown Object (File)
Sat, Jan 31, 6:05 PM
Unknown Object (File)
Jan 26 2026, 11:34 PM
Unknown Object (File)
Jan 21 2026, 10:08 AM
View All Files
Subscribers
cc
imp
melifaro
pouria

Details

Reviewers
np
gallatin
tuexen
rrs
glebius
Group Reviewers
transport
Summary

If the virtio network driver is not told by the host
that proper ECN support is available with TSO, it returns
ENOTSUP. This is not dealt with other than an effective
packet discard - and subsequently, a CWR-flagged TSO
transmission attempt is made until RTO reduces the
congestion window to a single packet.

Split the CWR-flagged TSO transmission in this case,
and resend immediately with the first segment with CWR
as non-TSO.

Diff Detail

Repository
rG FreeBSD src repository
Lint
Lint Passed
Unit
No Test Coverage
Build Status
Buildable 55219
Build 52108: arc lint + arc unit

Event Timeline

rscheff created this revision.Dec 27 2023, 2:21 PM
Herald added a reviewer: transport. · View Herald TranscriptDec 27 2023, 2:21 PM
Herald added subscribers: melifaro, imp. · View Herald Transcript
rscheff requested review of this revision.Dec 27 2023, 2:21 PM
Harbormaster completed remote builds in B55136: Diff 131906.Dec 27 2023, 2:22 PM
cc added a subscriber: cc.Jan 3 2024, 2:09 PM
cc added inline comments.
sys/netinet/tcp_output.c
1708

Put this comment below and I think it looks good to me.

rscheff updated this revision to Diff 132211.Jan 3 2024, 5:55 PM
  • make ENOTSUP dedicated case
Harbormaster completed remote builds in B55219: Diff 132211.Jan 3 2024, 5:55 PM
rrs added inline comments.Mar 31 2024, 12:06 PM
sys/netinet/tcp_output.c
1694

Don't you need this same change in rack?

tuexen added inline comments.Apr 1 2024, 12:36 PM
sys/netinet/tcp_output.c
1694

I think we first should agree on how what drivers report in specific error cases... Then fix all stacks...

pouria added a subscriber: pouria.Mon, Feb 23, 7:14 PM

Is this still relevant?
If that's the case, could you please set the planned change state?

rscheff added a comment.Mon, Feb 23, 7:31 PM

Personally I would like a TCP session NOT fail because it has ECN negotiated over a virtio driver, who then throws an error when a TSO superframe with CWR is handed down...

pouria added a comment.Mon, Feb 23, 7:35 PM
In D43213#1269045, @rscheff wrote:

Personally I would like a TCP session NOT fail because it has ECN negotiated over a virtio driver, who then throws an error when a TSO superframe with CWR is handed down...

I agree.
I think we should talk about this on next transport meeting.

Revision Contents
Changeset List

PathSize
sys/
netinet/
6 lines

Diff 132211

View Options

sys/netinet/tcp_output.c

Loading...