From owner-svn-src-all@FreeBSD.ORG Fri May 25 02:17:17 2012 Return-Path: Delivered-To: svn-src-all@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [69.147.83.52]) by hub.freebsd.org (Postfix) with ESMTP id 68BC4106577F; Fri, 25 May 2012 02:17:17 +0000 (UTC) (envelope-from bz@FreeBSD.org) Received: from svn.freebsd.org (svn.freebsd.org [IPv6:2001:4f8:fff6::2c]) by mx1.freebsd.org (Postfix) with ESMTP id 490588FC16; Fri, 25 May 2012 02:17:17 +0000 (UTC) Received: from svn.freebsd.org (localhost [127.0.0.1]) by svn.freebsd.org (8.14.4/8.14.4) with ESMTP id q4P2HHTY074072; Fri, 25 May 2012 02:17:17 GMT (envelope-from bz@svn.freebsd.org) Received: (from bz@localhost) by svn.freebsd.org (8.14.4/8.14.4/Submit) id q4P2HHiG074069; Fri, 25 May 2012 02:17:17 GMT (envelope-from bz@svn.freebsd.org) Message-Id: <201205250217.q4P2HHiG074069@svn.freebsd.org> From: "Bjoern A. Zeeb" Date: Fri, 25 May 2012 02:17:17 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cc: Subject: svn commit: r235958 - head/sys/netinet6 X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 25 May 2012 02:17:17 -0000 Author: bz Date: Fri May 25 02:17:16 2012 New Revision: 235958 URL: http://svn.freebsd.org/changeset/base/235958 Log: MFp4 bz_ipv6_fast: Add support for delayed checksum calculations in the IPv6 output path. We currently cannot offload to the card if we add extension headers (which incl. fragmentation). Fix two SCTP offload support copy&paste bugs: calculate checksums if fragmenting and no need to flag IPv4 header checksums in the IPv6 forwarding path. Sponsored by: The FreeBSD Foundation Sponsored by: iXsystems Reviewed by: gnn (as part of the whole) MFC After: 3 days Modified: head/sys/netinet6/ip6_forward.c head/sys/netinet6/ip6_output.c Modified: head/sys/netinet6/ip6_forward.c ============================================================================== --- head/sys/netinet6/ip6_forward.c Fri May 25 02:07:59 2012 (r235957) +++ head/sys/netinet6/ip6_forward.c Fri May 25 02:17:16 2012 (r235958) @@ -586,8 +586,6 @@ skip_routing: CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } - m->m_pkthdr.csum_flags |= - CSUM_IP_CHECKED | CSUM_IP_VALID; #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; Modified: head/sys/netinet6/ip6_output.c ============================================================================== --- head/sys/netinet6/ip6_output.c Fri May 25 02:07:59 2012 (r235957) +++ head/sys/netinet6/ip6_output.c Fri May 25 02:17:16 2012 (r235958) @@ -83,6 +83,8 @@ __FBSDID("$FreeBSD$"); #include #include +#include + #include #include #include @@ -182,6 +184,29 @@ static int copypktopts(struct ip6_pktopt }\ } while (/*CONSTCOND*/ 0) +static void +in6_delayed_cksum(struct mbuf *m, uint32_t plen, u_short offset) +{ + u_short csum; + + csum = in_cksum_skip(m, ntohl(plen), offset); + if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) + csum = 0xffff; + offset += m->m_pkthdr.csum_data; /* checksum offset */ + + if (offset + sizeof(u_short) > m->m_len) { + printf("%s: delayed m_pullup, m->len: %d off: %d\n", + __func__, m->m_len, offset); + /* + * XXX this should not happen, but if it does, the correct + * behavior may be to insert the checksum in the appropriate + * next mbuf in the chain. + */ + return; + } + *(u_short *)(m->m_data + offset) = csum; +} + /* * IP6 output. The packet in mbuf chain m contains a skeletal IP6 * header (with pri, len, nxt, hlim, src, dst). @@ -221,9 +246,7 @@ ip6_output(struct mbuf *m0, struct ip6_p int flevalid = 0; int hdrsplit = 0; int needipsec = 0; -#ifdef SCTP - int sw_csum; -#endif + int sw_csum, tso; #ifdef IPSEC struct ipsec_output_state state; struct ip6_rthdr *rh = NULL; @@ -867,8 +890,6 @@ again: CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } - m->m_pkthdr.csum_flags |= - CSUM_IP_CHECKED | CSUM_IP_VALID; #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; @@ -891,7 +912,7 @@ again: } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP) - m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; + m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif error = netisr_queue(NETISR_IPV6, m); goto done; @@ -927,16 +948,32 @@ passout: * 4: if dontfrag == 1 && alwaysfrag == 1 * error, as we cannot handle this conflicting request */ + sw_csum = m->m_pkthdr.csum_flags; + if (!hdrsplit) { + tso = ((sw_csum & ifp->if_hwassist & CSUM_TSO) != 0) ? 1 : 0; + sw_csum &= ~ifp->if_hwassist; + } else + tso = 0; + /* + * If we added extension headers, we will not do TSO and calculate the + * checksums ourselves for now. + * XXX-BZ Need a framework to know when the NIC can handle it, even + * with ext. hdrs. + */ + if (sw_csum & CSUM_DELAY_DATA) { + sw_csum &= ~CSUM_DELAY_DATA; + in6_delayed_cksum(m, ip6->ip6_plen, sizeof(struct ip6_hdr)); + } #ifdef SCTP - sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; if (sw_csum & CSUM_SCTP) { - sctp_delayed_cksum(m, sizeof(struct ip6_hdr)); sw_csum &= ~CSUM_SCTP; + sctp_delayed_cksum(m, sizeof(struct ip6_hdr)); } #endif + m->m_pkthdr.csum_flags &= ifp->if_hwassist; tlen = m->m_pkthdr.len; - if (opt && (opt->ip6po_flags & IP6PO_DONTFRAG)) + if ((opt && (opt->ip6po_flags & IP6PO_DONTFRAG)) || tso) dontfrag = 1; else dontfrag = 0; @@ -945,7 +982,7 @@ passout: error = EMSGSIZE; goto bad; } - if (dontfrag && tlen > IN6_LINKMTU(ifp)) { /* case 2-b */ + if (dontfrag && tlen > IN6_LINKMTU(ifp) && !tso) { /* case 2-b */ /* * Even if the DONTFRAG option is specified, we cannot send the * packet when the data length is larger than the MTU of the @@ -1033,6 +1070,22 @@ passout: goto bad; } + + /* + * If the interface will not calculate checksums on + * fragmented packets, then do it here. + * XXX-BZ handle the hw offloading case. Need flags. + */ + if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { + in6_delayed_cksum(m, ip6->ip6_plen, sizeof(*ip6)); + m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; + } +#ifdef SCTP + if (m->m_pkthdr.csum_flags & CSUM_SCTP) { + sctp_delayed_cksum(m, hlen); + m->m_pkthdr.csum_flags &= ~CSUM_SCTP; + } +#endif mnext = &m->m_nextpkt; /*