| Message ID | alpine.DEB.2.00.1010141429150.26447@wel-95.cs.helsinki.fi |
|---|---|
| State | Accepted, archived |
| Delegated to: | David Miller |
| Headers | show |
At the moment I don't have any time to spare to look into this since my master thesis has to be ready in about 2 weeks. Sorry for that! On 14.10.2010 13:42, Ilpo Järvinen wrote: > When only fast rexmit should be done, tcp_mark_head_lost marks > L too far. Also, sacked_upto below 1 is perfectly valid number, > the packets == 0 then needs to be trapped elsewhere. > > Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> > --- > I think 6830c25b7d08fbbd922959425193791bc42079f2 that added the > packets == 0 check is mostly wrong but I cc'ed Lennart if he has some > particular case I'm missing that wouldn't work after this patch. > > Dave, no particular "bad regression" fixed here, so no absolute need to > have this in net-2.6 but I leave it up to you whether there or net-next. > > ... -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
I forgot to mention that I will of course have a look at it as soon as I finished the thesis :) On 14.10.2010 15:36, Lennart Schulte wrote: > At the moment I don't have any time to spare to look into this since my > master thesis has to be ready in about 2 weeks. Sorry for that! > > On 14.10.2010 13:42, Ilpo Järvinen wrote: >> When only fast rexmit should be done, tcp_mark_head_lost marks >> L too far. Also, sacked_upto below 1 is perfectly valid number, >> the packets == 0 then needs to be trapped elsewhere. >> >> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> >> --- >> I think 6830c25b7d08fbbd922959425193791bc42079f2 that added the >> packets == 0 check is mostly wrong but I cc'ed Lennart if he has some >> particular case I'm missing that wouldn't work after this patch. >> >> Dave, no particular "bad regression" fixed here, so no absolute need to >> have this in net-2.6 but I leave it up to you whether there or net-next. >> >> ... > > -- > To unsubscribe from this list: send the line "unsubscribe netdev" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
From: "Ilpo Järvinen" <ilpo.jarvinen@helsinki.fi> Date: Thu, 14 Oct 2010 14:42:30 +0300 (EEST) > When only fast rexmit should be done, tcp_mark_head_lost marks > L too far. Also, sacked_upto below 1 is perfectly valid number, > the packets == 0 then needs to be trapped elsewhere. > > Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> Applied to net-next-2.6 -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 548d575..9924cd1 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -2504,7 +2504,7 @@ static void tcp_timeout_skbs(struct sock *sk) /* Mark head of queue up as lost. With RFC3517 SACK, the packets is * is against sacked "cnt", otherwise it's against facked "cnt" */ -static void tcp_mark_head_lost(struct sock *sk, int packets) +static void tcp_mark_head_lost(struct sock *sk, int packets, int mark_head) { struct tcp_sock *tp = tcp_sk(sk); struct sk_buff *skb; @@ -2512,13 +2512,13 @@ static void tcp_mark_head_lost(struct sock *sk, int packets) int err; unsigned int mss; - if (packets == 0) - return; - WARN_ON(packets > tp->packets_out); if (tp->lost_skb_hint) { skb = tp->lost_skb_hint; cnt = tp->lost_cnt_hint; + /* Head already handled? */ + if (mark_head && skb != tcp_write_queue_head(sk)) + return; } else { skb = tcp_write_queue_head(sk); cnt = 0; @@ -2552,6 +2552,9 @@ static void tcp_mark_head_lost(struct sock *sk, int packets) } tcp_skb_mark_lost(tp, skb); + + if (mark_head) + break; } tcp_verify_left_out(tp); } @@ -2563,17 +2566,18 @@ static void tcp_update_scoreboard(struct sock *sk, int fast_rexmit) struct tcp_sock *tp = tcp_sk(sk); if (tcp_is_reno(tp)) { - tcp_mark_head_lost(sk, 1); + tcp_mark_head_lost(sk, 1, 1); } else if (tcp_is_fack(tp)) { int lost = tp->fackets_out - tp->reordering; if (lost <= 0) lost = 1; - tcp_mark_head_lost(sk, lost); + tcp_mark_head_lost(sk, lost, 0); } else { int sacked_upto = tp->sacked_out - tp->reordering; - if (sacked_upto < fast_rexmit) - sacked_upto = fast_rexmit; - tcp_mark_head_lost(sk, sacked_upto); + if (sacked_upto >= 0) + tcp_mark_head_lost(sk, sacked_upto, 0); + else if (fast_rexmit) + tcp_mark_head_lost(sk, 1, 1); } tcp_timeout_skbs(sk); @@ -2978,7 +2982,7 @@ static void tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag) before(tp->snd_una, tp->high_seq) && icsk->icsk_ca_state != TCP_CA_Open && tp->fackets_out > tp->reordering) { - tcp_mark_head_lost(sk, tp->fackets_out - tp->reordering); + tcp_mark_head_lost(sk, tp->fackets_out - tp->reordering, 0); NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPLOSS); }
When only fast rexmit should be done, tcp_mark_head_lost marks L too far. Also, sacked_upto below 1 is perfectly valid number, the packets == 0 then needs to be trapped elsewhere. Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> --- I think 6830c25b7d08fbbd922959425193791bc42079f2 that added the packets == 0 check is mostly wrong but I cc'ed Lennart if he has some particular case I'm missing that wouldn't work after this patch. Dave, no particular "bad regression" fixed here, so no absolute need to have this in net-2.6 but I leave it up to you whether there or net-next. net/ipv4/tcp_input.c | 24 ++++++++++++++---------- 1 files changed, 14 insertions(+), 10 deletions(-)