• cleaning up forward declarations

• update whitespace fixes

• rebase
• simplify the fastpath check if ECT should be set.
• streamline fastpath check
• update man date
• have to explicitly check for ECN SYN
• add TF2_ECN_PLUSPLUS flag to reduce cacheline churn in fastpath when checking V_ecn_generalized
• bump tcp.4 datestamp
• fix man page typos
• fix typo in man page
• update to use the common tcp_ecn.c sourcefile
• fix types

• use the correct timeout value in log output

Possibly - probably missed this over my iterations. Thanks!

Sorry for the oversight.

• fix panic with NULL socket
• update man timestamp
• don't forget on is_login_timeout

I didn't want to modify the iscsi_callout() function completely. I guess the current design was picked to keep the housekeeping overhead low. Splitting iscsi_callout into a iscsid timeout, login timeout and ping hello function is certainly possible.

In D34230#775828, @jhb wrote:

(I think this diff contains the prior change as well?) I would also suggest mstosbt() here as well.

• use mstosbt()

• make use of mstosbt()

Any objections to getting rid of this recent compiler warning?

• rename variables to match convention

• manual rebase to main

• rebase to main

• make the call centered around the 1 sec mark

the cleanup of the debug/troubleshooting/dev variable looks largely fine to me.

• fix panic with NULL socket
• directly access isc_ struct instead of handoff
• update manual entry to reference sysctls
• nicify outputs to match .conf token

• improve man wording

• cleaning up a set of KMOD_ macros, and nicifying source code

• manual rebase to main

• rebase to main, set flags in rsm_fast_output after ecn changes

fixed with 2nd commit of D34162

• move tcp_set_flag aft the added setting of the ECN flags
• remove #include from tcp_ecn.h
• shuffle #include around, so that tcp_ecn.h doesn't need any
• restore #include in tcp_ecn.h

• restore diff after revert in main
• fix type of flag variable in rack.c

• rebase to main
• set flags in rack_fsm_fast_output after ecn changes

In D32540#773132, @imp wrote:

deciseconds are too weird. We should really specify them in terms of ms, even if internally we round that to 100ms increments.I'd hate to see that get established as the user interface and not be able to change it later.

In D32540#773132, @imp wrote:

deciseconds are too weird. We should really specify them in terms of ms, even if internally we round that to 100ms increments.I'd hate to see that get established as the user interface and not be able to change it later.

In short: I would like to commit this first go with decidec granularity, mostly unchanged external (script) interfaces (.conf, sysctl - well, changes from INT to STR, but the sysctl would not care in a script context). Unless there are strong objections against this current approach.

I believe I have addressed all the concerns - but keeping the old log format, such that any tools currently interpreting log files do not get confused.

• manual rebase to main

@jhb - any thoughts on making the timeout of iscsi sessions configurable in deci-seconds? This is to facilitate more rapid tcp reconnections during network disruptions.

• add ECN handling in the 3rd output path.

In D34138#772827, @tuexen wrote:

I see that you can set socket options wildcard style across all currently existing sockets.

tcpsso can perform such a loop, but not the code in the kernel.

mostly mechanical changes, look good to me...

• clean commit against main, for ease of review

• restore "tos" variable name

• update the ip.4 man page, to discourage the use of a tos field,

• only mask out ECN codepoint for TCP, leave other protos unperturbed
• no need to parse the optval again