1235474Sbz/*- 2235474Sbz * Copyright (c) 2007, Myricom Inc. 3235474Sbz * Copyright (c) 2008, Intel Corporation. 4235944Sbz * Copyright (c) 2012 The FreeBSD Foundation 5235474Sbz * All rights reserved. 6235474Sbz * 7235944Sbz * Portions of this software were developed by Bjoern Zeeb 8235944Sbz * under sponsorship from the FreeBSD Foundation. 9235944Sbz * 10235474Sbz * Redistribution and use in source and binary forms, with or without 11235474Sbz * modification, are permitted provided that the following conditions 12235474Sbz * are met: 13235474Sbz * 1. Redistributions of source code must retain the above copyright 14235474Sbz * notice, this list of conditions and the following disclaimer. 15235474Sbz * 2. Redistributions in binary form must reproduce the above copyright 16235474Sbz * notice, this list of conditions and the following disclaimer in the 17235474Sbz * documentation and/or other materials provided with the distribution. 18235474Sbz * 19235474Sbz * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20235474Sbz * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21235474Sbz * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22235474Sbz * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23235474Sbz * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24235474Sbz * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25235474Sbz * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26235474Sbz * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27235474Sbz * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28235474Sbz * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29235474Sbz * SUCH DAMAGE. 30235474Sbz */ 31179737Sjfv 32235944Sbz#include <sys/cdefs.h> 33235944Sbz__FBSDID("$FreeBSD$"); 34235944Sbz 35235944Sbz#include "opt_inet.h" 36235944Sbz#include "opt_inet6.h" 37235944Sbz 38179737Sjfv#include <sys/param.h> 39179737Sjfv#include <sys/systm.h> 40179737Sjfv#include <sys/mbuf.h> 41179737Sjfv#include <sys/kernel.h> 42179737Sjfv#include <sys/socket.h> 43305189Ssephe#include <sys/sysctl.h> 44179737Sjfv 45179737Sjfv#include <net/if.h> 46235944Sbz#include <net/if_var.h> 47179737Sjfv#include <net/ethernet.h> 48236394Sbz#include <net/vnet.h> 49179737Sjfv 50179737Sjfv#include <netinet/in_systm.h> 51179737Sjfv#include <netinet/in.h> 52235944Sbz#include <netinet/ip6.h> 53179737Sjfv#include <netinet/ip.h> 54235981Sbz#include <netinet/ip_var.h> 55179737Sjfv#include <netinet/tcp.h> 56179737Sjfv#include <netinet/tcp_lro.h> 57305189Ssephe#include <netinet/tcp_var.h> 58179737Sjfv 59235981Sbz#include <netinet6/ip6_var.h> 60235981Sbz 61179737Sjfv#include <machine/in_cksum.h> 62179737Sjfv 63235944Sbz#ifndef LRO_ENTRIES 64235944Sbz#define LRO_ENTRIES 8 /* # of LRO entries per RX queue. */ 65235944Sbz#endif 66179737Sjfv 67235944Sbz#define TCP_LRO_UPDATE_CSUM 1 68235944Sbz#ifndef TCP_LRO_UPDATE_CSUM 69235944Sbz#define TCP_LRO_INVALID_CSUM 0x0000 70235944Sbz#endif 71179737Sjfv 72305189SsepheSYSCTL_NODE(_net_inet_tcp, OID_AUTO, lro, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 73305189Ssephe "TCP LRO"); 74305189Ssephe 75305189Ssephestatic unsigned tcp_lro_entries = LRO_ENTRIES; 76305189SsepheSYSCTL_UINT(_net_inet_tcp_lro, OID_AUTO, entries, 77305189Ssephe CTLFLAG_RDTUN | CTLFLAG_MPSAFE, &tcp_lro_entries, 0, 78305189Ssephe "default number of LRO entries"); 79305189Ssephe 80179737Sjfvint 81235944Sbztcp_lro_init(struct lro_ctrl *lc) 82179737Sjfv{ 83235944Sbz struct lro_entry *le; 84235944Sbz int error, i; 85179737Sjfv 86235944Sbz lc->lro_bad_csum = 0; 87235944Sbz lc->lro_queued = 0; 88235944Sbz lc->lro_flushed = 0; 89235944Sbz lc->lro_cnt = 0; 90235944Sbz SLIST_INIT(&lc->lro_free); 91235944Sbz SLIST_INIT(&lc->lro_active); 92179737Sjfv 93235944Sbz error = 0; 94305189Ssephe for (i = 0; i < tcp_lro_entries; i++) { 95235944Sbz le = (struct lro_entry *)malloc(sizeof(*le), M_DEVBUF, 96235944Sbz M_NOWAIT | M_ZERO); 97235944Sbz if (le == NULL) { 98179737Sjfv if (i == 0) 99179737Sjfv error = ENOMEM; 100179737Sjfv break; 101179737Sjfv } 102235944Sbz lc->lro_cnt = i + 1; 103235944Sbz SLIST_INSERT_HEAD(&lc->lro_free, le, next); 104179737Sjfv } 105179737Sjfv 106179737Sjfv return (error); 107179737Sjfv} 108179737Sjfv 109179737Sjfvvoid 110235944Sbztcp_lro_free(struct lro_ctrl *lc) 111179737Sjfv{ 112235944Sbz struct lro_entry *le; 113179737Sjfv 114235944Sbz while (!SLIST_EMPTY(&lc->lro_free)) { 115235944Sbz le = SLIST_FIRST(&lc->lro_free); 116235944Sbz SLIST_REMOVE_HEAD(&lc->lro_free, next); 117235944Sbz free(le, M_DEVBUF); 118179737Sjfv } 119179737Sjfv} 120179737Sjfv 121235944Sbz#ifdef TCP_LRO_UPDATE_CSUM 122235944Sbzstatic uint16_t 123235944Sbztcp_lro_csum_th(struct tcphdr *th) 124235944Sbz{ 125235944Sbz uint32_t ch; 126235944Sbz uint16_t *p, l; 127235944Sbz 128235944Sbz ch = th->th_sum = 0x0000; 129235944Sbz l = th->th_off; 130235944Sbz p = (uint16_t *)th; 131235944Sbz while (l > 0) { 132235944Sbz ch += *p; 133235944Sbz p++; 134235944Sbz ch += *p; 135235944Sbz p++; 136235944Sbz l--; 137235944Sbz } 138235944Sbz while (ch > 0xffff) 139235944Sbz ch = (ch >> 16) + (ch & 0xffff); 140235944Sbz 141235944Sbz return (ch & 0xffff); 142235944Sbz} 143235944Sbz 144235944Sbzstatic uint16_t 145235944Sbztcp_lro_rx_csum_fixup(struct lro_entry *le, void *l3hdr, struct tcphdr *th, 146235944Sbz uint16_t tcp_data_len, uint16_t csum) 147235944Sbz{ 148235944Sbz uint32_t c; 149235944Sbz uint16_t cs; 150235944Sbz 151235944Sbz c = csum; 152235944Sbz 153235944Sbz /* Remove length from checksum. */ 154235944Sbz switch (le->eh_type) { 155235944Sbz#ifdef INET6 156235944Sbz case ETHERTYPE_IPV6: 157235944Sbz { 158235944Sbz struct ip6_hdr *ip6; 159235944Sbz 160235944Sbz ip6 = (struct ip6_hdr *)l3hdr; 161235944Sbz if (le->append_cnt == 0) 162235944Sbz cs = ip6->ip6_plen; 163235944Sbz else { 164235944Sbz uint32_t cx; 165235944Sbz 166235944Sbz cx = ntohs(ip6->ip6_plen); 167235944Sbz cs = in6_cksum_pseudo(ip6, cx, ip6->ip6_nxt, 0); 168235944Sbz } 169235944Sbz break; 170235944Sbz } 171235944Sbz#endif 172235944Sbz#ifdef INET 173235944Sbz case ETHERTYPE_IP: 174235944Sbz { 175235944Sbz struct ip *ip4; 176235944Sbz 177235944Sbz ip4 = (struct ip *)l3hdr; 178235944Sbz if (le->append_cnt == 0) 179235944Sbz cs = ip4->ip_len; 180235944Sbz else { 181235944Sbz cs = in_addword(ntohs(ip4->ip_len) - sizeof(*ip4), 182235944Sbz IPPROTO_TCP); 183235944Sbz cs = in_pseudo(ip4->ip_src.s_addr, ip4->ip_dst.s_addr, 184235944Sbz htons(cs)); 185235944Sbz } 186235944Sbz break; 187235944Sbz } 188235944Sbz#endif 189235944Sbz default: 190235944Sbz cs = 0; /* Keep compiler happy. */ 191235944Sbz } 192235944Sbz 193235944Sbz cs = ~cs; 194235944Sbz c += cs; 195235944Sbz 196235944Sbz /* Remove TCP header csum. */ 197235944Sbz cs = ~tcp_lro_csum_th(th); 198235944Sbz c += cs; 199235944Sbz while (c > 0xffff) 200235944Sbz c = (c >> 16) + (c & 0xffff); 201235944Sbz 202235944Sbz return (c & 0xffff); 203235944Sbz} 204235944Sbz#endif 205235944Sbz 206179737Sjfvvoid 207255010Snptcp_lro_flush_inactive(struct lro_ctrl *lc, const struct timeval *timeout) 208255010Snp{ 209255010Snp struct lro_entry *le, *le_tmp; 210255010Snp struct timeval tv; 211255010Snp 212255010Snp if (SLIST_EMPTY(&lc->lro_active)) 213255010Snp return; 214255010Snp 215255010Snp getmicrotime(&tv); 216255010Snp timevalsub(&tv, timeout); 217255010Snp SLIST_FOREACH_SAFE(le, &lc->lro_active, next, le_tmp) { 218255010Snp if (timevalcmp(&tv, &le->mtime, >=)) { 219255010Snp SLIST_REMOVE(&lc->lro_active, le, lro_entry, next); 220255010Snp tcp_lro_flush(lc, le); 221255010Snp } 222255010Snp } 223255010Snp} 224255010Snp 225255010Snpvoid 226235944Sbztcp_lro_flush(struct lro_ctrl *lc, struct lro_entry *le) 227179737Sjfv{ 228179737Sjfv 229235944Sbz if (le->append_cnt > 0) { 230235944Sbz struct tcphdr *th; 231235944Sbz uint16_t p_len; 232179737Sjfv 233235944Sbz p_len = htons(le->p_len); 234235944Sbz switch (le->eh_type) { 235235944Sbz#ifdef INET6 236235944Sbz case ETHERTYPE_IPV6: 237235944Sbz { 238235944Sbz struct ip6_hdr *ip6; 239179737Sjfv 240235944Sbz ip6 = le->le_ip6; 241235944Sbz ip6->ip6_plen = p_len; 242235944Sbz th = (struct tcphdr *)(ip6 + 1); 243235944Sbz le->m_head->m_pkthdr.csum_flags = CSUM_DATA_VALID | 244235944Sbz CSUM_PSEUDO_HDR; 245235944Sbz le->p_len += ETHER_HDR_LEN + sizeof(*ip6); 246235944Sbz break; 247235944Sbz } 248235944Sbz#endif 249235944Sbz#ifdef INET 250235944Sbz case ETHERTYPE_IP: 251235944Sbz { 252235944Sbz struct ip *ip4; 253235944Sbz#ifdef TCP_LRO_UPDATE_CSUM 254235944Sbz uint32_t cl; 255235944Sbz uint16_t c; 256235944Sbz#endif 257179737Sjfv 258235944Sbz ip4 = le->le_ip4; 259235944Sbz#ifdef TCP_LRO_UPDATE_CSUM 260235944Sbz /* Fix IP header checksum for new length. */ 261235944Sbz c = ~ip4->ip_sum; 262235944Sbz cl = c; 263235944Sbz c = ~ip4->ip_len; 264235944Sbz cl += c + p_len; 265235944Sbz while (cl > 0xffff) 266235944Sbz cl = (cl >> 16) + (cl & 0xffff); 267235944Sbz c = cl; 268235944Sbz ip4->ip_sum = ~c; 269235944Sbz#else 270235944Sbz ip4->ip_sum = TCP_LRO_INVALID_CSUM; 271235944Sbz#endif 272235944Sbz ip4->ip_len = p_len; 273235944Sbz th = (struct tcphdr *)(ip4 + 1); 274235944Sbz le->m_head->m_pkthdr.csum_flags = CSUM_DATA_VALID | 275235944Sbz CSUM_PSEUDO_HDR | CSUM_IP_CHECKED | CSUM_IP_VALID; 276235944Sbz le->p_len += ETHER_HDR_LEN; 277235944Sbz break; 278179737Sjfv } 279235944Sbz#endif 280235944Sbz default: 281235944Sbz th = NULL; /* Keep compiler happy. */ 282235944Sbz } 283235944Sbz le->m_head->m_pkthdr.csum_data = 0xffff; 284235944Sbz le->m_head->m_pkthdr.len = le->p_len; 285235944Sbz 286235944Sbz /* Incorporate the latest ACK into the TCP header. */ 287235944Sbz th->th_ack = le->ack_seq; 288235944Sbz th->th_win = le->window; 289235944Sbz /* Incorporate latest timestamp into the TCP header. */ 290235944Sbz if (le->timestamp != 0) { 291235944Sbz uint32_t *ts_ptr; 292235944Sbz 293235944Sbz ts_ptr = (uint32_t *)(th + 1); 294235944Sbz ts_ptr[1] = htonl(le->tsval); 295235944Sbz ts_ptr[2] = le->tsecr; 296235944Sbz } 297235944Sbz#ifdef TCP_LRO_UPDATE_CSUM 298235944Sbz /* Update the TCP header checksum. */ 299235944Sbz le->ulp_csum += p_len; 300235944Sbz le->ulp_csum += tcp_lro_csum_th(th); 301235944Sbz while (le->ulp_csum > 0xffff) 302235944Sbz le->ulp_csum = (le->ulp_csum >> 16) + 303235944Sbz (le->ulp_csum & 0xffff); 304235944Sbz th->th_sum = (le->ulp_csum & 0xffff); 305235944Sbz th->th_sum = ~th->th_sum; 306235944Sbz#else 307235944Sbz th->th_sum = TCP_LRO_INVALID_CSUM; 308235944Sbz#endif 309179737Sjfv } 310235944Sbz 311235944Sbz (*lc->ifp->if_input)(lc->ifp, le->m_head); 312235944Sbz lc->lro_queued += le->append_cnt + 1; 313235944Sbz lc->lro_flushed++; 314235944Sbz bzero(le, sizeof(*le)); 315235944Sbz SLIST_INSERT_HEAD(&lc->lro_free, le, next); 316179737Sjfv} 317179737Sjfv 318235944Sbz#ifdef INET6 319235944Sbzstatic int 320235944Sbztcp_lro_rx_ipv6(struct lro_ctrl *lc, struct mbuf *m, struct ip6_hdr *ip6, 321235944Sbz struct tcphdr **th) 322179737Sjfv{ 323179737Sjfv 324235944Sbz /* XXX-BZ we should check the flow-label. */ 325179737Sjfv 326235944Sbz /* XXX-BZ We do not yet support ext. hdrs. */ 327235944Sbz if (ip6->ip6_nxt != IPPROTO_TCP) 328235944Sbz return (TCP_LRO_NOT_SUPPORTED); 329179737Sjfv 330235944Sbz /* Find the TCP header. */ 331235944Sbz *th = (struct tcphdr *)(ip6 + 1); 332179737Sjfv 333235944Sbz return (0); 334235944Sbz} 335235944Sbz#endif 336235944Sbz 337235944Sbz#ifdef INET 338235944Sbzstatic int 339235944Sbztcp_lro_rx_ipv4(struct lro_ctrl *lc, struct mbuf *m, struct ip *ip4, 340235944Sbz struct tcphdr **th) 341235944Sbz{ 342235944Sbz int csum_flags; 343235944Sbz uint16_t csum; 344235944Sbz 345235944Sbz if (ip4->ip_p != IPPROTO_TCP) 346235944Sbz return (TCP_LRO_NOT_SUPPORTED); 347235944Sbz 348235944Sbz /* Ensure there are no options. */ 349235944Sbz if ((ip4->ip_hl << 2) != sizeof (*ip4)) 350235944Sbz return (TCP_LRO_CANNOT); 351235944Sbz 352235944Sbz /* .. and the packet is not fragmented. */ 353235944Sbz if (ip4->ip_off & htons(IP_MF|IP_OFFMASK)) 354235944Sbz return (TCP_LRO_CANNOT); 355235944Sbz 356235944Sbz /* Legacy IP has a header checksum that needs to be correct. */ 357235944Sbz csum_flags = m->m_pkthdr.csum_flags; 358182089Skmacy if (csum_flags & CSUM_IP_CHECKED) { 359182089Skmacy if (__predict_false((csum_flags & CSUM_IP_VALID) == 0)) { 360235944Sbz lc->lro_bad_csum++; 361235944Sbz return (TCP_LRO_CANNOT); 362182089Skmacy } 363182089Skmacy } else { 364235944Sbz csum = in_cksum_hdr(ip4); 365247104Sgallatin if (__predict_false((csum) != 0)) { 366235944Sbz lc->lro_bad_csum++; 367235944Sbz return (TCP_LRO_CANNOT); 368182089Skmacy } 369179737Sjfv } 370179737Sjfv 371235944Sbz /* Find the TCP header (we assured there are no IP options). */ 372235944Sbz *th = (struct tcphdr *)(ip4 + 1); 373179737Sjfv 374235944Sbz return (0); 375235944Sbz} 376235944Sbz#endif 377179737Sjfv 378235944Sbzint 379235944Sbztcp_lro_rx(struct lro_ctrl *lc, struct mbuf *m, uint32_t csum) 380235944Sbz{ 381235944Sbz struct lro_entry *le; 382235944Sbz struct ether_header *eh; 383235944Sbz#ifdef INET6 384235944Sbz struct ip6_hdr *ip6 = NULL; /* Keep compiler happy. */ 385235944Sbz#endif 386235944Sbz#ifdef INET 387235944Sbz struct ip *ip4 = NULL; /* Keep compiler happy. */ 388235944Sbz#endif 389235944Sbz struct tcphdr *th; 390235944Sbz void *l3hdr = NULL; /* Keep compiler happy. */ 391235944Sbz uint32_t *ts_ptr; 392235944Sbz tcp_seq seq; 393235944Sbz int error, ip_len, l; 394235944Sbz uint16_t eh_type, tcp_data_len; 395304836Ssephe int force_flush = 0; 396179737Sjfv 397235944Sbz /* We expect a contiguous header [eh, ip, tcp]. */ 398235944Sbz 399235944Sbz eh = mtod(m, struct ether_header *); 400235944Sbz eh_type = ntohs(eh->ether_type); 401235944Sbz switch (eh_type) { 402235944Sbz#ifdef INET6 403235944Sbz case ETHERTYPE_IPV6: 404236394Sbz { 405236394Sbz CURVNET_SET(lc->ifp->if_vnet); 406235981Sbz if (V_ip6_forwarding != 0) { 407235981Sbz /* XXX-BZ stats but changing lro_ctrl is a problem. */ 408236394Sbz CURVNET_RESTORE(); 409235981Sbz return (TCP_LRO_CANNOT); 410235981Sbz } 411236394Sbz CURVNET_RESTORE(); 412235944Sbz l3hdr = ip6 = (struct ip6_hdr *)(eh + 1); 413235944Sbz error = tcp_lro_rx_ipv6(lc, m, ip6, &th); 414235944Sbz if (error != 0) 415235944Sbz return (error); 416235944Sbz tcp_data_len = ntohs(ip6->ip6_plen); 417235944Sbz ip_len = sizeof(*ip6) + tcp_data_len; 418235944Sbz break; 419236394Sbz } 420235944Sbz#endif 421235944Sbz#ifdef INET 422235944Sbz case ETHERTYPE_IP: 423236394Sbz { 424236394Sbz CURVNET_SET(lc->ifp->if_vnet); 425235981Sbz if (V_ipforwarding != 0) { 426235981Sbz /* XXX-BZ stats but changing lro_ctrl is a problem. */ 427236394Sbz CURVNET_RESTORE(); 428235981Sbz return (TCP_LRO_CANNOT); 429235981Sbz } 430236394Sbz CURVNET_RESTORE(); 431235944Sbz l3hdr = ip4 = (struct ip *)(eh + 1); 432235944Sbz error = tcp_lro_rx_ipv4(lc, m, ip4, &th); 433235944Sbz if (error != 0) 434235944Sbz return (error); 435235944Sbz ip_len = ntohs(ip4->ip_len); 436235944Sbz tcp_data_len = ip_len - sizeof(*ip4); 437235944Sbz break; 438236394Sbz } 439235944Sbz#endif 440235944Sbz /* XXX-BZ what happens in case of VLAN(s)? */ 441235944Sbz default: 442235944Sbz return (TCP_LRO_NOT_SUPPORTED); 443179737Sjfv } 444179737Sjfv 445235944Sbz /* 446235944Sbz * If the frame is padded beyond the end of the IP packet, then we must 447235944Sbz * trim the extra bytes off. 448235944Sbz */ 449235944Sbz l = m->m_pkthdr.len - (ETHER_HDR_LEN + ip_len); 450235944Sbz if (l != 0) { 451235944Sbz if (l < 0) 452235944Sbz /* Truncated packet. */ 453235944Sbz return (TCP_LRO_CANNOT); 454179737Sjfv 455235944Sbz m_adj(m, -l); 456235944Sbz } 457235944Sbz 458235944Sbz /* 459235944Sbz * Check TCP header constraints. 460179737Sjfv */ 461235944Sbz /* Ensure no bits set besides ACK or PSH. */ 462304836Ssephe if ((th->th_flags & ~(TH_ACK | TH_PUSH)) != 0) { 463304836Ssephe if (th->th_flags & TH_SYN) 464304836Ssephe return (TCP_LRO_CANNOT); 465304836Ssephe /* 466304836Ssephe * Make sure that previously seen segements/ACKs are delivered 467304836Ssephe * before this segement, e.g. FIN. 468304836Ssephe */ 469304836Ssephe force_flush = 1; 470304836Ssephe } 471235944Sbz 472302051Ssephe /* XXX-BZ We lose a ACK|PUSH flag concatenating multiple segments. */ 473235944Sbz /* XXX-BZ Ideally we'd flush on PUSH? */ 474235944Sbz 475235944Sbz /* 476235944Sbz * Check for timestamps. 477235944Sbz * Since the only option we handle are timestamps, we only have to 478235944Sbz * handle the simple case of aligned timestamps. 479235944Sbz */ 480235944Sbz l = (th->th_off << 2); 481235944Sbz tcp_data_len -= l; 482235944Sbz l -= sizeof(*th); 483235944Sbz ts_ptr = (uint32_t *)(th + 1); 484235944Sbz if (l != 0 && (__predict_false(l != TCPOLEN_TSTAMP_APPA) || 485235944Sbz (*ts_ptr != ntohl(TCPOPT_NOP<<24|TCPOPT_NOP<<16| 486304836Ssephe TCPOPT_TIMESTAMP<<8|TCPOLEN_TIMESTAMP)))) { 487304836Ssephe /* 488304836Ssephe * Make sure that previously seen segements/ACKs are delivered 489304836Ssephe * before this segement. 490304836Ssephe */ 491304836Ssephe force_flush = 1; 492304836Ssephe } 493235944Sbz 494235944Sbz /* If the driver did not pass in the checksum, set it now. */ 495235944Sbz if (csum == 0x0000) 496235944Sbz csum = th->th_sum; 497235944Sbz 498235944Sbz seq = ntohl(th->th_seq); 499235944Sbz 500235944Sbz /* Try to find a matching previous segment. */ 501235944Sbz SLIST_FOREACH(le, &lc->lro_active, next) { 502235944Sbz if (le->eh_type != eh_type) 503235944Sbz continue; 504235944Sbz if (le->source_port != th->th_sport || 505235944Sbz le->dest_port != th->th_dport) 506235944Sbz continue; 507235944Sbz switch (eh_type) { 508235944Sbz#ifdef INET6 509235944Sbz case ETHERTYPE_IPV6: 510235944Sbz if (bcmp(&le->source_ip6, &ip6->ip6_src, 511235944Sbz sizeof(struct in6_addr)) != 0 || 512235944Sbz bcmp(&le->dest_ip6, &ip6->ip6_dst, 513235944Sbz sizeof(struct in6_addr)) != 0) 514235944Sbz continue; 515235944Sbz break; 516235944Sbz#endif 517235944Sbz#ifdef INET 518235944Sbz case ETHERTYPE_IP: 519235944Sbz if (le->source_ip4 != ip4->ip_src.s_addr || 520235944Sbz le->dest_ip4 != ip4->ip_dst.s_addr) 521235944Sbz continue; 522235944Sbz break; 523235944Sbz#endif 524179737Sjfv } 525179737Sjfv 526304836Ssephe if (force_flush) { 527304836Ssephe /* Timestamps mismatch; this is a FIN, etc */ 528304836Ssephe SLIST_REMOVE(&lc->lro_active, le, lro_entry, next); 529304836Ssephe tcp_lro_flush(lc, le); 530304836Ssephe return (TCP_LRO_CANNOT); 531304836Ssephe } 532304836Ssephe 533235944Sbz /* Flush now if appending will result in overflow. */ 534235944Sbz if (le->p_len > (65535 - tcp_data_len)) { 535235944Sbz SLIST_REMOVE(&lc->lro_active, le, lro_entry, next); 536235944Sbz tcp_lro_flush(lc, le); 537235944Sbz break; 538235944Sbz } 539179737Sjfv 540235944Sbz /* Try to append the new segment. */ 541235944Sbz if (__predict_false(seq != le->next_seq || 542235944Sbz (tcp_data_len == 0 && le->ack_seq == th->th_ack))) { 543235944Sbz /* Out of order packet or duplicate ACK. */ 544235944Sbz SLIST_REMOVE(&lc->lro_active, le, lro_entry, next); 545235944Sbz tcp_lro_flush(lc, le); 546235944Sbz return (TCP_LRO_CANNOT); 547235944Sbz } 548179737Sjfv 549235944Sbz if (l != 0) { 550235944Sbz uint32_t tsval = ntohl(*(ts_ptr + 1)); 551235944Sbz /* Make sure timestamp values are increasing. */ 552235944Sbz /* XXX-BZ flip and use TSTMP_GEQ macro for this? */ 553235944Sbz if (__predict_false(le->tsval > tsval || 554235944Sbz *(ts_ptr + 2) == 0)) 555235944Sbz return (TCP_LRO_CANNOT); 556235944Sbz le->tsval = tsval; 557235944Sbz le->tsecr = *(ts_ptr + 2); 558235944Sbz } 559223797Scperciva 560235944Sbz le->next_seq += tcp_data_len; 561235944Sbz le->ack_seq = th->th_ack; 562235944Sbz le->window = th->th_win; 563235944Sbz le->append_cnt++; 564179737Sjfv 565235944Sbz#ifdef TCP_LRO_UPDATE_CSUM 566235944Sbz le->ulp_csum += tcp_lro_rx_csum_fixup(le, l3hdr, th, 567235944Sbz tcp_data_len, ~csum); 568235944Sbz#endif 569179737Sjfv 570235944Sbz if (tcp_data_len == 0) { 571235944Sbz m_freem(m); 572235944Sbz return (0); 573235944Sbz } 574179737Sjfv 575235944Sbz le->p_len += tcp_data_len; 576179737Sjfv 577235944Sbz /* 578235944Sbz * Adjust the mbuf so that m_data points to the first byte of 579235944Sbz * the ULP payload. Adjust the mbuf to avoid complications and 580235944Sbz * append new segment to existing mbuf chain. 581235944Sbz */ 582235944Sbz m_adj(m, m->m_pkthdr.len - tcp_data_len); 583235944Sbz m->m_flags &= ~M_PKTHDR; 584179737Sjfv 585235944Sbz le->m_tail->m_next = m; 586235944Sbz le->m_tail = m_last(m); 587235944Sbz 588235944Sbz /* 589235944Sbz * If a possible next full length packet would cause an 590235944Sbz * overflow, pro-actively flush now. 591235944Sbz */ 592235944Sbz if (le->p_len > (65535 - lc->ifp->if_mtu)) { 593235944Sbz SLIST_REMOVE(&lc->lro_active, le, lro_entry, next); 594235944Sbz tcp_lro_flush(lc, le); 595255010Snp } else 596255010Snp getmicrotime(&le->mtime); 597235944Sbz 598235944Sbz return (0); 599179737Sjfv } 600179737Sjfv 601304836Ssephe if (force_flush) { 602304836Ssephe /* 603304836Ssephe * Nothing to flush, but this segment can not be further 604304836Ssephe * aggregated/delayed. 605304836Ssephe */ 606304836Ssephe return (TCP_LRO_CANNOT); 607304836Ssephe } 608304836Ssephe 609235944Sbz /* Try to find an empty slot. */ 610235944Sbz if (SLIST_EMPTY(&lc->lro_free)) 611301949Ssephe return (TCP_LRO_NO_ENTRIES); 612179737Sjfv 613235944Sbz /* Start a new segment chain. */ 614235944Sbz le = SLIST_FIRST(&lc->lro_free); 615235944Sbz SLIST_REMOVE_HEAD(&lc->lro_free, next); 616235944Sbz SLIST_INSERT_HEAD(&lc->lro_active, le, next); 617255010Snp getmicrotime(&le->mtime); 618179737Sjfv 619235944Sbz /* Start filling in details. */ 620235944Sbz switch (eh_type) { 621235944Sbz#ifdef INET6 622235944Sbz case ETHERTYPE_IPV6: 623235944Sbz le->le_ip6 = ip6; 624235944Sbz le->source_ip6 = ip6->ip6_src; 625235944Sbz le->dest_ip6 = ip6->ip6_dst; 626235944Sbz le->eh_type = eh_type; 627235944Sbz le->p_len = m->m_pkthdr.len - ETHER_HDR_LEN - sizeof(*ip6); 628235944Sbz break; 629235944Sbz#endif 630235944Sbz#ifdef INET 631235944Sbz case ETHERTYPE_IP: 632235944Sbz le->le_ip4 = ip4; 633235944Sbz le->source_ip4 = ip4->ip_src.s_addr; 634235944Sbz le->dest_ip4 = ip4->ip_dst.s_addr; 635235944Sbz le->eh_type = eh_type; 636235944Sbz le->p_len = m->m_pkthdr.len - ETHER_HDR_LEN; 637235944Sbz break; 638235944Sbz#endif 639235944Sbz } 640235944Sbz le->source_port = th->th_sport; 641235944Sbz le->dest_port = th->th_dport; 642235944Sbz 643235944Sbz le->next_seq = seq + tcp_data_len; 644235944Sbz le->ack_seq = th->th_ack; 645235944Sbz le->window = th->th_win; 646235944Sbz if (l != 0) { 647235944Sbz le->timestamp = 1; 648235944Sbz le->tsval = ntohl(*(ts_ptr + 1)); 649235944Sbz le->tsecr = *(ts_ptr + 2); 650235944Sbz } 651235944Sbz 652235944Sbz#ifdef TCP_LRO_UPDATE_CSUM 653235944Sbz /* 654235944Sbz * Do not touch the csum of the first packet. However save the 655235944Sbz * "adjusted" checksum of just the source and destination addresses, 656235944Sbz * the next header and the TCP payload. The length and TCP header 657235944Sbz * parts may change, so we remove those from the saved checksum and 658235944Sbz * re-add with final values on tcp_lro_flush() if needed. 659179737Sjfv */ 660235944Sbz KASSERT(le->ulp_csum == 0, ("%s: le=%p le->ulp_csum=0x%04x\n", 661235944Sbz __func__, le, le->ulp_csum)); 662235944Sbz 663235944Sbz le->ulp_csum = tcp_lro_rx_csum_fixup(le, l3hdr, th, tcp_data_len, 664235944Sbz ~csum); 665235944Sbz th->th_sum = csum; /* Restore checksum on first packet. */ 666235944Sbz#endif 667235944Sbz 668235944Sbz le->m_head = m; 669235944Sbz le->m_tail = m_last(m); 670235944Sbz 671235944Sbz return (0); 672179737Sjfv} 673235944Sbz 674235944Sbz/* end */ 675