On Wed, 5 Dec 2007, Andrew Morton wrote: > On Thu, 06 Dec 2007 17:59:37 +1100 Reuben Farrelly wrote: > > > This non fatal oops which I have just noticed may be related to this change then > > - certainly looks networking related. > > yep, but it isn't e1000. It's core TCP. > > > WARNING: at net/ipv4/tcp_input.c:2518 tcp_fastretrans_alert() > > Pid: 0, comm: swapper Not tainted 2.6.24-rc4-mm1 #1 > > Ilpo, Reuben's kernel is talking to you ;) ...Please try the patch below. Andrew, this probably fixes your problem (the packets <= tp->packets_out) as well. Dave, please include this one to net-2.6.25. -- i. -- [PATCH] [TCP]: Fix fack_count miscountings (multiple places) 1) Fack_count is set incorrectly if the highest sent skb is already sacked (the skb->prev won't return it because it's on the other list already). These manifest as fackets_out counting error later on, the second-order effects are very hard to track, so it may fix all out-standing TCP bug reports. 2) Prev == NULL check was wrong way around 3) Last skb's fack count was incorrectly skipped while() {} loop Signed-off-by: Ilpo Järvinen --- include/net/tcp.h | 22 ++++++++++++++++------ 1 files changed, 16 insertions(+), 6 deletions(-) diff --git a/include/net/tcp.h b/include/net/tcp.h index 9dbed0b..11a7e3e 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -1337,10 +1337,20 @@ static inline struct sk_buff *tcp_send_head(struct sock *sk) static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) { struct sk_buff *prev = tcp_write_queue_prev(sk, skb); + unsigned int fc = 0; + + if (prev == (struct sk_buff *)&sk->sk_write_queue) + prev = NULL; + else if (!tcp_skb_adjacent(sk, prev, skb)) + prev = NULL; - if (prev != (struct sk_buff *)&sk->sk_write_queue) - TCP_SKB_CB(skb)->fack_count = TCP_SKB_CB(prev)->fack_count + - tcp_skb_pcount(prev); + if ((prev == NULL) && !__tcp_write_queue_empty(sk, TCP_WQ_SACKED)) + prev = __tcp_write_queue_tail(sk, TCP_WQ_SACKED); + + if (prev != NULL) + fc = TCP_SKB_CB(prev)->fack_count + tcp_skb_pcount(prev); + + TCP_SKB_CB(skb)->fack_count = fc; sk->sk_send_head = tcp_write_queue_next(sk, skb); if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue) @@ -1464,7 +1474,7 @@ static inline struct sk_buff *__tcp_reset_fack_counts(struct sock *sk, { unsigned int fc = 0; - if (prev == NULL) + if (prev != NULL) fc = TCP_SKB_CB(*prev)->fack_count + tcp_skb_pcount(*prev); BUG_ON((*prev != NULL) && !tcp_skb_adjacent(sk, *prev, skb)); @@ -1521,7 +1531,7 @@ static inline void tcp_reset_fack_counts(struct sock *sk, struct sk_buff *inskb) skb[otherq] = prev->next; } - while (skb[queue] != __tcp_write_queue_tail(sk, queue)) { + do { /* Lazy find for the other queue */ if (skb[queue] == NULL) { skb[queue] = tcp_write_queue_find(sk, TCP_SKB_CB(prev)->seq, @@ -1535,7 +1545,7 @@ static inline void tcp_reset_fack_counts(struct sock *sk, struct sk_buff *inskb) break; queue ^= TCP_WQ_SACKED; - } + } while (skb[queue] != __tcp_write_queue_tail(sk, queue)); } static inline void __tcp_insert_write_queue_after(struct sk_buff *skb, -- 1.5.0.6