> Sorry but I guess I managed to misread the functions in question.
> 
> So, resuming the discussion:
> 1) Wouldn't we be breaking the standard, and maybe, adding more  
> processing in the case of  SEQ_GT th->th_ack, tp->snd_max ?

SEQ_GT(th->th_ack, tp->snd_max) mean broken peer or such, doesn't it?
is it stated in a standard?  i couldn't find it.

> 2) I think it's safer to stop/restart the retransmit timers before  
> moving on to the adjusting the congestion window.

i don't think it makes differences as we are at splsoftnet,
which blocks timers.

> 3) This is probably a matter of taste, but I prefer it this way.  
> Maybe I'm wrong, but I think it doesn't involve any performance  
> drawback this way.
> 
> What do you think?

i don't have a strong preference, so if you prefer it for some reason,
i have no trouble with it.

after all, my main concern was about names of callbacks, rather than
number of callbacks. :)  i'd suggest:
	"cwnd_inflation" -> "fast_retransmit_newack"
	"new_data_acked" -> "newack"

YAMAMOTO Takashi