Host queues (Qdisc + NIC) can hold packets so long that TCP can eventually retransmit a packet before the first transmit even left the host. Its not clear right now if we could avoid this in the first place : - We could arm RTO timer not at the time we enqueue packets, but at the time we TX complete them (tcp_wfree()) - Cancel the sending of the new copy of the packet if prior one is still in queue. This patch adds instrumentation so that we can at least see how often this problem happens. TCPSpuriousRtxHostQueues SNMP counter is incremented every time we detect the fast clone is not yet freed in tcp_transmit_skb()
Unified · Split
| @@ -247,6 +247,7 @@ enum · 246 unmodified lines | |||
| 247 | LINUX_MIB_TCPFASTOPENPASSIVEFAIL, /* TCPFastOpenPassiveFail */ | 247 | LINUX_MIB_TCPFASTOPENPASSIVEFAIL, /* TCPFastOpenPassiveFail */ |
| 248 | LINUX_MIB_TCPFASTOPENLISTENOVERFLOW, /* TCPFastOpenListenOverflow */ | 248 | LINUX_MIB_TCPFASTOPENLISTENOVERFLOW, /* TCPFastOpenListenOverflow */ |
| 249 | LINUX_MIB_TCPFASTOPENCOOKIEREQD, /* TCPFastOpenCookieReqd */ | 249 | LINUX_MIB_TCPFASTOPENCOOKIEREQD, /* TCPFastOpenCookieReqd */ |
| 250 | LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES, /* TCPSpuriousRtxHostQueues */ | ||
| 250 | __LINUX_MIB_MAX | 251 | __LINUX_MIB_MAX |
| 251 | }; | 252 | }; |
| 252 | 253 | ||
| @@ -269,6 +269,7 @@ static const struct snmp_mib snmp4_net_list[] = { · 268 unmodified lines | |||
| 269 | SNMP_MIB_ITEM("TCPFastOpenPassiveFail", LINUX_MIB_TCPFASTOPENPASSIVEFAIL), | 269 | SNMP_MIB_ITEM("TCPFastOpenPassiveFail", LINUX_MIB_TCPFASTOPENPASSIVEFAIL), |
| 270 | SNMP_MIB_ITEM("TCPFastOpenListenOverflow", LINUX_MIB_TCPFASTOPENLISTENOVERFLOW), | 270 | SNMP_MIB_ITEM("TCPFastOpenListenOverflow", LINUX_MIB_TCPFASTOPENLISTENOVERFLOW), |
| 271 | SNMP_MIB_ITEM("TCPFastOpenCookieReqd", LINUX_MIB_TCPFASTOPENCOOKIEREQD), | 271 | SNMP_MIB_ITEM("TCPFastOpenCookieReqd", LINUX_MIB_TCPFASTOPENCOOKIEREQD), |
| 272 | SNMP_MIB_ITEM("TCPSpuriousRtxHostQueues", LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES), | ||
| 272 | SNMP_MIB_SENTINEL | 273 | SNMP_MIB_SENTINEL |
| 273 | }; | 274 | }; |
| 274 | 275 | ||
| @@ -846,6 +846,13 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, · 845 unmodified lines | |||
| 846 | __net_timestamp(skb); | 846 | __net_timestamp(skb); |
| 847 | 847 | ||
| 848 | if (likely(clone_it)) { | 848 | if (likely(clone_it)) { |
| 849 | const struct sk_buff *fclone = skb + 1; | ||
| 850 | |||
| 851 | if (unlikely(skb->fclone == SKB_FCLONE_ORIG && | ||
| 852 | fclone->fclone == SKB_FCLONE_CLONE)) | ||
| 853 | NET_INC_STATS_BH(sock_net(sk), | ||
| 854 | LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES); | ||
| 855 | |||
| 849 | if (unlikely(skb_cloned(skb))) | 856 | if (unlikely(skb_cloned(skb))) |
| 850 | skb = pskb_copy(skb, gfp_mask); | 857 | skb = pskb_copy(skb, gfp_mask); |
| 851 | else | 858 | else |