Add PRR 6937bis heuristic and remove prr_conservative sysctl

[rscheff]

Summary

Proportional Rate Reduction (RFC6937) is in the process
of becoming a standard. The difference to the experimental
RFC is a heuristic, which automatically chooses between
CRB and SSRB schema. Only when snd_una advances (a partial
ACK), SSRB may be used. Also, that ACK must not have
any indication of ongoing loss - using the addition of
new holes into the scoreboard as proxy for such an
event.

Now based off D29441.

MFC after: 2 weeks

[rscheff]

The code above is my current understanding of the 6927bis draft. Note that this will make PRR loss recovery slightly less aggressive especially under heavy loss.

Also, this change probably doesn't change much in the pathological instance discueed in the draft, as the base stack has no capability at the moment, to retransmit lost retransmissions, and will instead run into an RTO and slow-start thereafter.

[rscheff]

• rebase to HEAD

[rscheff]

rebase against D29441

[rscheff]

• use PRR-SSRB and PRR-CRB per -bis heuristic

[rscheff]

• update comment to include ACK split attack

[rscheff]

no magic numbers in sack-changed.

[rscheff]

• rebase without D29441
• make magic values human-readable

[glebius]

sys/netinet/tcp_var.h

507–511

Maybe make at a typedef enum? Funcion arguments will look prettier in a debugger session.

[rscheff]

• make the sack return status an enum for nice debugging

[glebius]

In D28822#868424, @rscheff wrote:

• make the sack return status an enum for nice debugging

Sorry for nit-picking, but this isn't enough for the goal of seeing the string literals in the debugger. The prototype of tcp_sack_doack() needs to be changed accordingly, as well as type of variable.

typedef enum {
        SACK_NOCHANGE = 0,
        SACK_CHANGE,
        SACK_NEWLOSS
} sack_status;

sack_status  tcp_sack_doack(struct tcpcb *, struct tcpopt *, tcp_seq);

And in tcp_do_segment():

sack_status sack_changed;

[rscheff]

• properly define and use sackstatus_t type

[rscheff]

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.

[rscheff]

• honor proper SACK scoreboard change

[rscheff]

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

[rscheff]

• exclude limited transmit segments from prr_delivered.

[rscheff]

• rebase to main

[rscheff]

• rebase to main