Lines Matching defs:which
162 * - Matt Mathis's Rate Halving which slowly drops
165 * - Yuchung Cheng's RACK TCP (for which its named) that
2107 * All other less noisy logs here except the measure which
2465 * which is more or less a guess. Note
2792 rack_log_to_start(struct tcp_rack *rack, uint32_t cts, uint32_t to, int32_t slot, uint8_t which)
2806 log.u_bbr.flex8 = which;
3622 * 1) We have a desired BDP which is usually
3633 * have gone by which is done further below in
3704 * is effecting the last byte (which is a 50:50 chance).
4384 * of setting CA/SS to 100% always at exit (which is
4639 * for b/w with it. This will push the RTT up which
5070 * defined minimum GP (defaulting to 10 which hopefully is the
5238 * the sending rate which factors in all the
5518 * send-limit set the current time, which
5576 * ack_rcv_time which with pacing may be different.
5850 /* We place a min on the cap which defaults to 1Mbps */
5973 * which then means the rest of recovery will hopefully go
5977 * hit this as we exit recovery as well which means only
5992 * idle_snd_una value (which decreases the amount in last_amount_before_rec
6288 * have contained SACK information which we processed else
6475 * should be the last segment (which it was not).
6664 * We want the rack_rtt which is the
6666 * does not exist we fallback to the srtt (which
7304 * seconds (a bit more than 596 hours), which is probably ok
7347 * a SACK arrives (which could change things) or if
7603 /* This should *not* happen which is why there is a kassert */
7657 * we basically calculate a new offset, which is soff +
7718 * We keep the largest value, which is the newest
8510 * (most of which have hopefully been retired) that have bad VJ
8511 * header compression code which trashes TCP segments containing
8587 * are on a min-timeout (which means rrr_conf = 3)
8917 * Here we retransmitted less than the whole thing which means we
9077 * sndmbuf_noadv returned (which is
9291 * SACK'd which overrides the r_just_ret or
9383 * back, which may be a problem...
9480 * feasible timer (which is 2 ticks).
9609 * as well as the pacing delay between packets which
9666 * Now which timestamp does it match? In this block the ACK
10151 * We keep the largest value, which is the newest
10505 * We keep the largest value, which is the newest
10538 * that is acked (which was merged
10616 * which is un-acked.
10826 * "guessed" window, but then it shrinks. Now which was
10941 * the cum-ack is by the TLP before checking which is
11062 * which can happen due to reordering.
11183 * which can happen due to reordering. In this
11349 * The DSACK is because of a TLP which we don't
12155 * do use len which includes TLS overhead, but not the TCP/IP etc.
12988 * Insert segment which includes th into TCP reassembly
13892 * simultaneous SYN, which is handled later.
15582 * it to the rwnd (which the default stack does
15766 * is left (in which case rack_init() now knows how
16402 * cwnd which may be inflated).
16618 * We don't take that path here which is why we set
16928 * We would normally do drop-with-reset which would
16959 * which means we are probably in TCPS_CLOSING.
17266 * If this occurs (which it won't very often since we most likely
17290 * peer because the MSS is too small which
19169 * and acks that call the extend function which would let
19311 * from the query (which should be our b/w rate)
20721 /* Do the timers, which may override the pacer */
21601 * can delay the acks, which is why we want to make sure we
21607 * 2 MSS out in response to the acks being clocked in which
21626 * makes it impossible to transmit any options which vary per
21844 * have sent a FIN, which can speed things up for
21984 * investagate which reason we missed.
22116 * have sent a FIN, which can speed things up for
22455 * have sent a FIN, which can speed things up for
22588 * will keep happy those implementations which only give
23252 * out (with the TSO), which we should not be doing.
23874 * Now we must switch to hybrid mode as well which also