Not changing the generic rtt base stack unit / measuring, just improviing the granularity/precision of the calculations within Cubic. since this improves cubic performance, let's put this in prior 14 branching...

tcp_compute_pipe is always controlled by mostly V_tcp_do_newsack in the output path and CC modules, and V_tcp_do_newcwv in the input path for deciding if the session is cwnd limited or not. New cwnd validation is disabled by default, newsack (formerly rfc6675_pipe) is enabled by default.

• combine assignment and if check

So when I added the capability, it happened to work incidenitally one the nic drivers i had tested this on? And this would extend the proper operation to other drivers (or skip the PCP setting, when no vlan tagging capabilities exist?

Not questioning that change per se. But what you are functionally doing is not extracting the full traffic class (tclass), but extract the DSCP bits, clearing the ECN bits and sending traffic later using the same DSCP codepoint back.

validate that new pipe calc is gated by sysctl (to retain classical non-6675 pipe calculation if need be)

These changes are already in, with D39368 - https://reviews.freebsd.org/source/src/browse/main/share/misc/committers-src.dot - not?

Yes, the entries should be sorted by mentor (1st), mentee (2nd).

• exclude limited transmit segments from prr_delivered.

Did more unit testing as promised. Found that the hardcoded CRB during dupack processing needs to be changed against the proper toggle (SACK_NEWLOSS or SACK_CHANGE) - to conform with RFC6937bis

• honor proper SACK scoreboard change

• remove fat-finger line deletion

Why would you need to remove the #ifdef TCPTIMERS when including tcp_var.h in tcp_timers.h?

Also can not really think of any sane reason why you'd require unique ephemeral outgoing ports (outside of the local host). Quite the opposite, that restricts the total number of outgoing connections globally and not on a per-destination basis. Finally, this sysctl apparently was never mentioned anywhere (searchable) to be a workaround for any observed or perceived issue - while the original intent of the referenced patch is apparently working as expected.

Thanks Gleb! While at it, I also found that most other typedef enums use a nomenclature of a _t suffix. Also fixed up all relevant variables - haven't checked gdb to verify the proper decoding though.

• properly define and use sackstatus_t type

• make the sack return status an enum for nice debugging

• rebase to main
• use static maximum value for uint8_t

I believe Cheng doesn't have the commit bit; I suggest we start the process of onboarding Cheng as another FBSD committer....

will use hardcode max_uint values in next rev.

• rebase to main

Wondering if a stronger type check can be enforced to find all the instances, where a magic number is used for pfil_return_t instead one of the literals...

• use switch statement instead of if else

• comment typo

• fix oversights

• remove tcps_ecn_sndce

• rebase
• use ect1 if session is using that

• add new flag at the bottom and next bit

• rebase to main manually

• rebase
• bump man date
• remove duplicate code

FYI - The plan is also to MFC this to Stable/13 (prior to 13.2 release) and Stable/12

• move ecn sysctls to tcp_ecn.c
• go with AccECN options from a compile time option to a global setting
• document the new sysctl option
• remove outdated comment
• add logging, and make syncache response honor most recent incoming SYN fields
• add logging
• more logging
• ensure control packets have accecn opt, if enabled
• fix packt_this_ack bug
• bump man date
• received accecn counter into tcpinfo
• add more tcpinfo counters
• make accecn option counters a struct within tcpcb, which can also be locally allocated (syncache, prior of tcpcb)
• initialize to_ae properly
• reference properly to counters
• bump man page date
• remove all logging

• remove ecn.option (part of accecn-option)

• add tcpi_delivered_ce for accecn and ecn (accounting received ECEs)
• report tcpi_delivered_ce properly

• fix packet_this_ack bug

didn't automatically close after commit.

• only reserve the accecn related counters