Author: sephe Date: Fri Aug 5 09:08:00 2016 New Revision: 303766 URL: https://svnweb.freebsd.org/changeset/base/303766
Log: tcp/lro: If timestamps mismatch or it's a FIN, force flush. This keeps the segments/ACK/FIN delivery order. Before this patch, it was observed: if A sent FIN immediately after an ACK, B would deliver FIN first to the TCP stack, then the ACK. This out-of-order delivery causes one unnecessary ACK sent from B. Reviewed by: gallatin, hps Obtained from: rrs, gallatin Sponsored by: Netflix (rrs, gallatin), Microsoft (sephe) Differential Revision: https://reviews.freebsd.org/D7415 Modified: head/sys/netinet/tcp_lro.c Modified: head/sys/netinet/tcp_lro.c ============================================================================== --- head/sys/netinet/tcp_lro.c Fri Aug 5 08:58:00 2016 (r303765) +++ head/sys/netinet/tcp_lro.c Fri Aug 5 09:08:00 2016 (r303766) @@ -603,6 +603,7 @@ tcp_lro_rx2(struct lro_ctrl *lc, struct int error, ip_len, l; uint16_t eh_type, tcp_data_len; struct lro_head *bucket; + int force_flush = 0; /* We expect a contiguous header [eh, ip, tcp]. */ @@ -669,8 +670,15 @@ tcp_lro_rx2(struct lro_ctrl *lc, struct * Check TCP header constraints. */ /* Ensure no bits set besides ACK or PSH. */ - if ((th->th_flags & ~(TH_ACK | TH_PUSH)) != 0) - return (TCP_LRO_CANNOT); + if ((th->th_flags & ~(TH_ACK | TH_PUSH)) != 0) { + if (th->th_flags & TH_SYN) + return (TCP_LRO_CANNOT); + /* + * Make sure that previously seen segements/ACKs are delivered + * before this segement, e.g. FIN. + */ + force_flush = 1; + } /* XXX-BZ We lose a ACK|PUSH flag concatenating multiple segments. */ /* XXX-BZ Ideally we'd flush on PUSH? */ @@ -686,8 +694,13 @@ tcp_lro_rx2(struct lro_ctrl *lc, struct ts_ptr = (uint32_t *)(th + 1); if (l != 0 && (__predict_false(l != TCPOLEN_TSTAMP_APPA) || (*ts_ptr != ntohl(TCPOPT_NOP<<24|TCPOPT_NOP<<16| - TCPOPT_TIMESTAMP<<8|TCPOLEN_TIMESTAMP)))) - return (TCP_LRO_CANNOT); + TCPOPT_TIMESTAMP<<8|TCPOLEN_TIMESTAMP)))) { + /* + * Make sure that previously seen segements/ACKs are delivered + * before this segement. + */ + force_flush = 1; + } /* If the driver did not pass in the checksum, set it now. */ if (csum == 0x0000) @@ -754,6 +767,13 @@ tcp_lro_rx2(struct lro_ctrl *lc, struct #endif } + if (force_flush) { + /* Timestamps mismatch; this is a FIN, etc */ + tcp_lro_active_remove(le); + tcp_lro_flush(lc, le); + return (TCP_LRO_CANNOT); + } + /* Flush now if appending will result in overflow. */ if (le->p_len > (lc->lro_length_lim - tcp_data_len)) { tcp_lro_active_remove(le); @@ -830,6 +850,14 @@ tcp_lro_rx2(struct lro_ctrl *lc, struct return (0); } + if (force_flush) { + /* + * Nothing to flush, but this segment can not be further + * aggregated/delayed. + */ + return (TCP_LRO_CANNOT); + } + /* Try to find an empty slot. */ if (LIST_EMPTY(&lc->lro_free)) return (TCP_LRO_NO_ENTRIES); _______________________________________________ svn-src-head@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
