hw/net/net_tx_pkt.c

   1 /*
   2  * QEMU TX packets abstractions
   3  *
   4  * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
   5  *
   6  * Developed by Daynix Computing LTD (http://www.daynix.com)
   7  *
   8  * Authors:
   9  * Dmitry Fleytman <dmitry@daynix.com>
  10  * Tamir Shomer <tamirs@daynix.com>
  11  * Yan Vugenfirer <yan@daynix.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  14  * See the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "qemu/osdep.h"
  19 #include "qemu/crc32c.h"
  20 #include "net/eth.h"
  21 #include "net/checksum.h"
  22 #include "net/tap.h"
  23 #include "net/net.h"
  24 #include "hw/pci/pci_device.h"
  25 #include "net_tx_pkt.h"
  26
  27 enum {
  28     NET_TX_PKT_VHDR_FRAG = 0,
  29     NET_TX_PKT_L2HDR_FRAG,
  30     NET_TX_PKT_L3HDR_FRAG,
  31     NET_TX_PKT_PL_START_FRAG
  32 };
  33
  34 /* TX packet private context */
  35 struct NetTxPkt {
  36     struct virtio_net_hdr virt_hdr;
  37
  38     struct iovec *raw;
  39     uint32_t raw_frags;
  40     uint32_t max_raw_frags;
  41
  42     struct iovec *vec;
  43
  44     struct {
  45         struct eth_header eth;
  46         struct vlan_header vlan[3];
  47     } l2_hdr;
  48     union {
  49         struct ip_header ip;
  50         struct ip6_header ip6;
  51         uint8_t octets[ETH_MAX_IP_DGRAM_LEN];
  52     } l3_hdr;
  53
  54     uint32_t payload_len;
  55
  56     uint32_t payload_frags;
  57     uint32_t max_payload_frags;
  58
  59     uint16_t hdr_len;
  60     eth_pkt_types_e packet_type;
  61     uint8_t l4proto;
  62 };
  63
  64 void net_tx_pkt_init(struct NetTxPkt **pkt, uint32_t max_frags)
  65 {
  66     struct NetTxPkt *p = g_malloc0(sizeof *p);
  67
  68     p->vec = g_new(struct iovec, max_frags + NET_TX_PKT_PL_START_FRAG);
  69
  70     p->raw = g_new(struct iovec, max_frags);
  71
  72     p->max_payload_frags = max_frags;
  73     p->max_raw_frags = max_frags;
  74     p->vec[NET_TX_PKT_VHDR_FRAG].iov_base = &p->virt_hdr;
  75     p->vec[NET_TX_PKT_VHDR_FRAG].iov_len = sizeof p->virt_hdr;
  76     p->vec[NET_TX_PKT_L2HDR_FRAG].iov_base = &p->l2_hdr;
  77     p->vec[NET_TX_PKT_L3HDR_FRAG].iov_base = &p->l3_hdr;
  78
  79     *pkt = p;
  80 }
  81
  82 void net_tx_pkt_uninit(struct NetTxPkt *pkt)
  83 {
  84     if (pkt) {
  85         g_free(pkt->vec);
  86         g_free(pkt->raw);
  87         g_free(pkt);
  88     }
  89 }
  90
  91 void net_tx_pkt_update_ip_hdr_checksum(struct NetTxPkt *pkt)
  92 {
  93     uint16_t csum;
  94     assert(pkt);
  95
  96     pkt->l3_hdr.ip.ip_len = cpu_to_be16(pkt->payload_len +
  97         pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len);
  98
  99     pkt->l3_hdr.ip.ip_sum = 0;
 100     csum = net_raw_checksum(pkt->l3_hdr.octets,
 101         pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len);
 102     pkt->l3_hdr.ip.ip_sum = cpu_to_be16(csum);
 103 }
 104
 105 void net_tx_pkt_update_ip_checksums(struct NetTxPkt *pkt)
 106 {
 107     uint16_t csum;
 108     uint32_t cntr, cso;
 109     assert(pkt);
 110     uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
 111     void *ip_hdr = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
 112
 113     if (pkt->payload_len + pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len >
 114         ETH_MAX_IP_DGRAM_LEN) {
 115         return;
 116     }
 117
 118     if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 ||
 119         gso_type == VIRTIO_NET_HDR_GSO_UDP) {
 120         /* Calculate IP header checksum */
 121         net_tx_pkt_update_ip_hdr_checksum(pkt);
 122
 123         /* Calculate IP pseudo header checksum */
 124         cntr = eth_calc_ip4_pseudo_hdr_csum(ip_hdr, pkt->payload_len, &cso);
 125         csum = cpu_to_be16(~net_checksum_finish(cntr));
 126     } else if (gso_type == VIRTIO_NET_HDR_GSO_TCPV6) {
 127         /* Calculate IP pseudo header checksum */
 128         cntr = eth_calc_ip6_pseudo_hdr_csum(ip_hdr, pkt->payload_len,
 129                                             IP_PROTO_TCP, &cso);
 130         csum = cpu_to_be16(~net_checksum_finish(cntr));
 131     } else {
 132         return;
 133     }
 134
 135     iov_from_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags,
 136                  pkt->virt_hdr.csum_offset, &csum, sizeof(csum));
 137 }
 138
 139 bool net_tx_pkt_update_sctp_checksum(struct NetTxPkt *pkt)
 140 {
 141     uint32_t csum = 0;
 142     struct iovec *pl_start_frag = pkt->vec + NET_TX_PKT_PL_START_FRAG;
 143
 144     if (iov_from_buf(pl_start_frag, pkt->payload_frags, 8, &csum, sizeof(csum)) < sizeof(csum)) {
 145         return false;
 146     }
 147
 148     csum = cpu_to_le32(iov_crc32c(0xffffffff, pl_start_frag, pkt->payload_frags));
 149     if (iov_from_buf(pl_start_frag, pkt->payload_frags, 8, &csum, sizeof(csum)) < sizeof(csum)) {
 150         return false;
 151     }
 152
 153     return true;
 154 }
 155
 156 static void net_tx_pkt_calculate_hdr_len(struct NetTxPkt *pkt)
 157 {
 158     pkt->hdr_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len +
 159         pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len;
 160 }
 161
 162 static bool net_tx_pkt_parse_headers(struct NetTxPkt *pkt)
 163 {
 164     struct iovec *l2_hdr, *l3_hdr;
 165     size_t bytes_read;
 166     size_t full_ip6hdr_len;
 167     uint16_t l3_proto;
 168
 169     assert(pkt);
 170
 171     l2_hdr = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
 172     l3_hdr = &pkt->vec[NET_TX_PKT_L3HDR_FRAG];
 173
 174     bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, 0, l2_hdr->iov_base,
 175                             ETH_MAX_L2_HDR_LEN);
 176     if (bytes_read < sizeof(struct eth_header)) {
 177         l2_hdr->iov_len = 0;
 178         return false;
 179     }
 180
 181     l2_hdr->iov_len = sizeof(struct eth_header);
 182     switch (be16_to_cpu(PKT_GET_ETH_HDR(l2_hdr->iov_base)->h_proto)) {
 183     case ETH_P_VLAN:
 184         l2_hdr->iov_len += sizeof(struct vlan_header);
 185         break;
 186     case ETH_P_DVLAN:
 187         l2_hdr->iov_len += 2 * sizeof(struct vlan_header);
 188         break;
 189     }
 190
 191     if (bytes_read < l2_hdr->iov_len) {
 192         l2_hdr->iov_len = 0;
 193         l3_hdr->iov_len = 0;
 194         pkt->packet_type = ETH_PKT_UCAST;
 195         return false;
 196     } else {
 197         l2_hdr->iov_len = ETH_MAX_L2_HDR_LEN;
 198         l2_hdr->iov_len = eth_get_l2_hdr_length(l2_hdr->iov_base);
 199         pkt->packet_type = get_eth_packet_type(l2_hdr->iov_base);
 200     }
 201
 202     l3_proto = eth_get_l3_proto(l2_hdr, 1, l2_hdr->iov_len);
 203
 204     switch (l3_proto) {
 205     case ETH_P_IP:
 206         bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
 207                                 l3_hdr->iov_base, sizeof(struct ip_header));
 208
 209         if (bytes_read < sizeof(struct ip_header)) {
 210             l3_hdr->iov_len = 0;
 211             return false;
 212         }
 213
 214         l3_hdr->iov_len = IP_HDR_GET_LEN(l3_hdr->iov_base);
 215
 216         if (l3_hdr->iov_len < sizeof(struct ip_header)) {
 217             l3_hdr->iov_len = 0;
 218             return false;
 219         }
 220
 221         pkt->l4proto = IP_HDR_GET_P(l3_hdr->iov_base);
 222
 223         if (IP_HDR_GET_LEN(l3_hdr->iov_base) != sizeof(struct ip_header)) {
 224             /* copy optional IPv4 header data if any*/
 225             bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags,
 226                                     l2_hdr->iov_len + sizeof(struct ip_header),
 227                                     l3_hdr->iov_base + sizeof(struct ip_header),
 228                                     l3_hdr->iov_len - sizeof(struct ip_header));
 229             if (bytes_read < l3_hdr->iov_len - sizeof(struct ip_header)) {
 230                 l3_hdr->iov_len = 0;
 231                 return false;
 232             }
 233         }
 234
 235         break;
 236
 237     case ETH_P_IPV6:
 238     {
 239         eth_ip6_hdr_info hdrinfo;
 240
 241         if (!eth_parse_ipv6_hdr(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
 242                                 &hdrinfo)) {
 243             l3_hdr->iov_len = 0;
 244             return false;
 245         }
 246
 247         pkt->l4proto = hdrinfo.l4proto;
 248         full_ip6hdr_len = hdrinfo.full_hdr_len;
 249
 250         if (full_ip6hdr_len > ETH_MAX_IP_DGRAM_LEN) {
 251             l3_hdr->iov_len = 0;
 252             return false;
 253         }
 254
 255         bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
 256                                 l3_hdr->iov_base, full_ip6hdr_len);
 257
 258         if (bytes_read < full_ip6hdr_len) {
 259             l3_hdr->iov_len = 0;
 260             return false;
 261         } else {
 262             l3_hdr->iov_len = full_ip6hdr_len;
 263         }
 264         break;
 265     }
 266     default:
 267         l3_hdr->iov_len = 0;
 268         break;
 269     }
 270
 271     net_tx_pkt_calculate_hdr_len(pkt);
 272     return true;
 273 }
 274
 275 static void net_tx_pkt_rebuild_payload(struct NetTxPkt *pkt)
 276 {
 277     pkt->payload_len = iov_size(pkt->raw, pkt->raw_frags) - pkt->hdr_len;
 278     pkt->payload_frags = iov_copy(&pkt->vec[NET_TX_PKT_PL_START_FRAG],
 279                                 pkt->max_payload_frags,
 280                                 pkt->raw, pkt->raw_frags,
 281                                 pkt->hdr_len, pkt->payload_len);
 282 }
 283
 284 bool net_tx_pkt_parse(struct NetTxPkt *pkt)
 285 {
 286     if (net_tx_pkt_parse_headers(pkt)) {
 287         net_tx_pkt_rebuild_payload(pkt);
 288         return true;
 289     } else {
 290         return false;
 291     }
 292 }
 293
 294 struct virtio_net_hdr *net_tx_pkt_get_vhdr(struct NetTxPkt *pkt)
 295 {
 296     assert(pkt);
 297     return &pkt->virt_hdr;
 298 }
 299
 300 static uint8_t net_tx_pkt_get_gso_type(struct NetTxPkt *pkt,
 301                                           bool tso_enable)
 302 {
 303     uint8_t rc = VIRTIO_NET_HDR_GSO_NONE;
 304     uint16_t l3_proto;
 305
 306     l3_proto = eth_get_l3_proto(&pkt->vec[NET_TX_PKT_L2HDR_FRAG], 1,
 307         pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len);
 308
 309     if (!tso_enable) {
 310         goto func_exit;
 311     }
 312
 313     rc = eth_get_gso_type(l3_proto, pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
 314                           pkt->l4proto);
 315
 316 func_exit:
 317     return rc;
 318 }
 319
 320 bool net_tx_pkt_build_vheader(struct NetTxPkt *pkt, bool tso_enable,
 321     bool csum_enable, uint32_t gso_size)
 322 {
 323     struct tcp_hdr l4hdr;
 324     size_t bytes_read;
 325     assert(pkt);
 326
 327     /* csum has to be enabled if tso is. */
 328     assert(csum_enable || !tso_enable);
 329
 330     pkt->virt_hdr.gso_type = net_tx_pkt_get_gso_type(pkt, tso_enable);
 331
 332     switch (pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 333     case VIRTIO_NET_HDR_GSO_NONE:
 334         pkt->virt_hdr.hdr_len = 0;
 335         pkt->virt_hdr.gso_size = 0;
 336         break;
 337
 338     case VIRTIO_NET_HDR_GSO_UDP:
 339         pkt->virt_hdr.gso_size = gso_size;
 340         pkt->virt_hdr.hdr_len = pkt->hdr_len + sizeof(struct udp_header);
 341         break;
 342
 343     case VIRTIO_NET_HDR_GSO_TCPV4:
 344     case VIRTIO_NET_HDR_GSO_TCPV6:
 345         bytes_read = iov_to_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG],
 346                                 pkt->payload_frags, 0, &l4hdr, sizeof(l4hdr));
 347         if (bytes_read < sizeof(l4hdr) ||
 348             l4hdr.th_off * sizeof(uint32_t) < sizeof(l4hdr)) {
 349             return false;
 350         }
 351
 352         pkt->virt_hdr.hdr_len = pkt->hdr_len + l4hdr.th_off * sizeof(uint32_t);
 353         pkt->virt_hdr.gso_size = gso_size;
 354         break;
 355
 356     default:
 357         g_assert_not_reached();
 358     }
 359
 360     if (csum_enable) {
 361         switch (pkt->l4proto) {
 362         case IP_PROTO_TCP:
 363             if (pkt->payload_len < sizeof(struct tcp_hdr)) {
 364                 return false;
 365             }
 366             pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 367             pkt->virt_hdr.csum_start = pkt->hdr_len;
 368             pkt->virt_hdr.csum_offset = offsetof(struct tcp_hdr, th_sum);
 369             break;
 370         case IP_PROTO_UDP:
 371             if (pkt->payload_len < sizeof(struct udp_hdr)) {
 372                 return false;
 373             }
 374             pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 375             pkt->virt_hdr.csum_start = pkt->hdr_len;
 376             pkt->virt_hdr.csum_offset = offsetof(struct udp_hdr, uh_sum);
 377             break;
 378         default:
 379             break;
 380         }
 381     }
 382
 383     return true;
 384 }
 385
 386 void net_tx_pkt_setup_vlan_header_ex(struct NetTxPkt *pkt,
 387     uint16_t vlan, uint16_t vlan_ethtype)
 388 {
 389     assert(pkt);
 390
 391     eth_setup_vlan_headers(pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base,
 392                            &pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len,
 393                            vlan, vlan_ethtype);
 394
 395     pkt->hdr_len += sizeof(struct vlan_header);
 396 }
 397
 398 bool net_tx_pkt_add_raw_fragment(struct NetTxPkt *pkt, void *base, size_t len)
 399 {
 400     struct iovec *ventry;
 401     assert(pkt);
 402
 403     if (pkt->raw_frags >= pkt->max_raw_frags) {
 404         return false;
 405     }
 406
 407     ventry = &pkt->raw[pkt->raw_frags];
 408     ventry->iov_base = base;
 409     ventry->iov_len = len;
 410     pkt->raw_frags++;
 411
 412     return true;
 413 }
 414
 415 bool net_tx_pkt_has_fragments(struct NetTxPkt *pkt)
 416 {
 417     return pkt->raw_frags > 0;
 418 }
 419
 420 eth_pkt_types_e net_tx_pkt_get_packet_type(struct NetTxPkt *pkt)
 421 {
 422     assert(pkt);
 423
 424     return pkt->packet_type;
 425 }
 426
 427 size_t net_tx_pkt_get_total_len(struct NetTxPkt *pkt)
 428 {
 429     assert(pkt);
 430
 431     return pkt->hdr_len + pkt->payload_len;
 432 }
 433
 434 void net_tx_pkt_dump(struct NetTxPkt *pkt)
 435 {
 436 #ifdef NET_TX_PKT_DEBUG
 437     assert(pkt);
 438
 439     printf("TX PKT: hdr_len: %d, pkt_type: 0x%X, l2hdr_len: %lu, "
 440         "l3hdr_len: %lu, payload_len: %u\n", pkt->hdr_len, pkt->packet_type,
 441         pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len,
 442         pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len, pkt->payload_len);
 443 #endif
 444 }
 445
 446 void net_tx_pkt_reset(struct NetTxPkt *pkt,
 447                       NetTxPktFreeFrag callback, void *context)
 448 {
 449     int i;
 450
 451     /* no assert, as reset can be called before tx_pkt_init */
 452     if (!pkt) {
 453         return;
 454     }
 455
 456     memset(&pkt->virt_hdr, 0, sizeof(pkt->virt_hdr));
 457
 458     assert(pkt->vec);
 459
 460     pkt->payload_len = 0;
 461     pkt->payload_frags = 0;
 462
 463     if (pkt->max_raw_frags > 0) {
 464         assert(pkt->raw);
 465         for (i = 0; i < pkt->raw_frags; i++) {
 466             assert(pkt->raw[i].iov_base);
 467             callback(context, pkt->raw[i].iov_base, pkt->raw[i].iov_len);
 468         }
 469     }
 470     pkt->raw_frags = 0;
 471
 472     pkt->hdr_len = 0;
 473     pkt->l4proto = 0;
 474 }
 475
 476 void net_tx_pkt_unmap_frag_pci(void *context, void *base, size_t len)
 477 {
 478     pci_dma_unmap(context, base, len, DMA_DIRECTION_TO_DEVICE, 0);
 479 }
 480
 481 bool net_tx_pkt_add_raw_fragment_pci(struct NetTxPkt *pkt, PCIDevice *pci_dev,
 482                                      dma_addr_t pa, size_t len)
 483 {
 484     dma_addr_t mapped_len = len;
 485     void *base = pci_dma_map(pci_dev, pa, &mapped_len, DMA_DIRECTION_TO_DEVICE);
 486     if (!base) {
 487         return false;
 488     }
 489
 490     if (mapped_len != len || !net_tx_pkt_add_raw_fragment(pkt, base, len)) {
 491         net_tx_pkt_unmap_frag_pci(pci_dev, base, mapped_len);
 492         return false;
 493     }
 494
 495     return true;
 496 }
 497
 498 static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt,
 499                                   struct iovec *iov, uint32_t iov_len,
 500                                   uint16_t csl)
 501 {
 502     uint32_t csum_cntr;
 503     uint16_t csum = 0;
 504     uint32_t cso;
 505     /* num of iovec without vhdr */
 506     size_t csum_offset = pkt->virt_hdr.csum_start + pkt->virt_hdr.csum_offset;
 507     uint16_t l3_proto = eth_get_l3_proto(iov, 1, iov->iov_len);
 508
 509     /* Put zero to checksum field */
 510     iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
 511
 512     /* Calculate L4 TCP/UDP checksum */
 513     csum_cntr = 0;
 514     cso = 0;
 515     /* add pseudo header to csum */
 516     if (l3_proto == ETH_P_IP) {
 517         csum_cntr = eth_calc_ip4_pseudo_hdr_csum(
 518                 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
 519                 csl, &cso);
 520     } else if (l3_proto == ETH_P_IPV6) {
 521         csum_cntr = eth_calc_ip6_pseudo_hdr_csum(
 522                 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
 523                 csl, pkt->l4proto, &cso);
 524     }
 525
 526     /* data checksum */
 527     csum_cntr +=
 528         net_checksum_add_iov(iov, iov_len, pkt->virt_hdr.csum_start, csl, cso);
 529
 530     /* Put the checksum obtained into the packet */
 531     csum = cpu_to_be16(net_checksum_finish_nozero(csum_cntr));
 532     iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
 533 }
 534
 535 #define NET_MAX_FRAG_SG_LIST (64)
 536
 537 static size_t net_tx_pkt_fetch_fragment(struct NetTxPkt *pkt,
 538     int *src_idx, size_t *src_offset, size_t src_len,
 539     struct iovec *dst, int *dst_idx)
 540 {
 541     size_t fetched = 0;
 542     struct iovec *src = pkt->vec;
 543
 544     while (fetched < src_len) {
 545
 546         /* no more place in fragment iov */
 547         if (*dst_idx == NET_MAX_FRAG_SG_LIST) {
 548             break;
 549         }
 550
 551         /* no more data in iovec */
 552         if (*src_idx == (pkt->payload_frags + NET_TX_PKT_PL_START_FRAG)) {
 553             break;
 554         }
 555
 556
 557         dst[*dst_idx].iov_base = src[*src_idx].iov_base + *src_offset;
 558         dst[*dst_idx].iov_len = MIN(src[*src_idx].iov_len - *src_offset,
 559             src_len - fetched);
 560
 561         *src_offset += dst[*dst_idx].iov_len;
 562         fetched += dst[*dst_idx].iov_len;
 563
 564         if (*src_offset == src[*src_idx].iov_len) {
 565             *src_offset = 0;
 566             (*src_idx)++;
 567         }
 568
 569         (*dst_idx)++;
 570     }
 571
 572     return fetched;
 573 }
 574
 575 static void net_tx_pkt_sendv(
 576     void *opaque, const struct iovec *iov, int iov_cnt,
 577     const struct iovec *virt_iov, int virt_iov_cnt)
 578 {
 579     NetClientState *nc = opaque;
 580
 581     if (qemu_get_using_vnet_hdr(nc->peer)) {
 582         qemu_sendv_packet(nc, virt_iov, virt_iov_cnt);
 583     } else {
 584         qemu_sendv_packet(nc, iov, iov_cnt);
 585     }
 586 }
 587
 588 static bool net_tx_pkt_tcp_fragment_init(struct NetTxPkt *pkt,
 589                                          struct iovec *fragment,
 590                                          int *pl_idx,
 591                                          size_t *l4hdr_len,
 592                                          int *src_idx,
 593                                          size_t *src_offset,
 594                                          size_t *src_len)
 595 {
 596     struct iovec *l4 = fragment + NET_TX_PKT_PL_START_FRAG;
 597     size_t bytes_read = 0;
 598     struct tcp_hdr *th;
 599
 600     if (!pkt->payload_frags) {
 601         return false;
 602     }
 603
 604     l4->iov_len = pkt->virt_hdr.hdr_len - pkt->hdr_len;
 605     l4->iov_base = g_malloc(l4->iov_len);
 606
 607     *src_idx = NET_TX_PKT_PL_START_FRAG;
 608     while (pkt->vec[*src_idx].iov_len < l4->iov_len - bytes_read) {
 609         memcpy((char *)l4->iov_base + bytes_read, pkt->vec[*src_idx].iov_base,
 610                pkt->vec[*src_idx].iov_len);
 611
 612         bytes_read += pkt->vec[*src_idx].iov_len;
 613
 614         (*src_idx)++;
 615         if (*src_idx >= pkt->payload_frags + NET_TX_PKT_PL_START_FRAG) {
 616             g_free(l4->iov_base);
 617             return false;
 618         }
 619     }
 620
 621     *src_offset = l4->iov_len - bytes_read;
 622     memcpy((char *)l4->iov_base + bytes_read, pkt->vec[*src_idx].iov_base,
 623            *src_offset);
 624
 625     th = l4->iov_base;
 626     th->th_flags &= ~(TH_FIN | TH_PUSH);
 627
 628     *pl_idx = NET_TX_PKT_PL_START_FRAG + 1;
 629     *l4hdr_len = l4->iov_len;
 630     *src_len = pkt->virt_hdr.gso_size;
 631
 632     return true;
 633 }
 634
 635 static void net_tx_pkt_tcp_fragment_deinit(struct iovec *fragment)
 636 {
 637     g_free(fragment[NET_TX_PKT_PL_START_FRAG].iov_base);
 638 }
 639
 640 static void net_tx_pkt_tcp_fragment_fix(struct NetTxPkt *pkt,
 641                                         struct iovec *fragment,
 642                                         size_t fragment_len,
 643                                         uint8_t gso_type)
 644 {
 645     struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG;
 646     struct iovec *l4hdr = fragment + NET_TX_PKT_PL_START_FRAG;
 647     struct ip_header *ip = l3hdr->iov_base;
 648     struct ip6_header *ip6 = l3hdr->iov_base;
 649     size_t len = l3hdr->iov_len + l4hdr->iov_len + fragment_len;
 650
 651     switch (gso_type) {
 652     case VIRTIO_NET_HDR_GSO_TCPV4:
 653         ip->ip_len = cpu_to_be16(len);
 654         eth_fix_ip4_checksum(l3hdr->iov_base, l3hdr->iov_len);
 655         break;
 656
 657     case VIRTIO_NET_HDR_GSO_TCPV6:
 658         len -= sizeof(struct ip6_header);
 659         ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = cpu_to_be16(len);
 660         break;
 661     }
 662 }
 663
 664 static void net_tx_pkt_tcp_fragment_advance(struct NetTxPkt *pkt,
 665                                             struct iovec *fragment,
 666                                             size_t fragment_len,
 667                                             uint8_t gso_type)
 668 {
 669     struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG;
 670     struct iovec *l4hdr = fragment + NET_TX_PKT_PL_START_FRAG;
 671     struct ip_header *ip = l3hdr->iov_base;
 672     struct tcp_hdr *th = l4hdr->iov_base;
 673
 674     if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4) {
 675         ip->ip_id = cpu_to_be16(be16_to_cpu(ip->ip_id) + 1);
 676     }
 677
 678     th->th_seq = cpu_to_be32(be32_to_cpu(th->th_seq) + fragment_len);
 679     th->th_flags &= ~TH_CWR;
 680 }
 681
 682 static void net_tx_pkt_udp_fragment_init(struct NetTxPkt *pkt,
 683                                          int *pl_idx,
 684                                          size_t *l4hdr_len,
 685                                          int *src_idx, size_t *src_offset,
 686                                          size_t *src_len)
 687 {
 688     *pl_idx = NET_TX_PKT_PL_START_FRAG;
 689     *l4hdr_len = 0;
 690     *src_idx = NET_TX_PKT_PL_START_FRAG;
 691     *src_offset = 0;
 692     *src_len = IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size);
 693 }
 694
 695 static void net_tx_pkt_udp_fragment_fix(struct NetTxPkt *pkt,
 696                                         struct iovec *fragment,
 697                                         size_t fragment_offset,
 698                                         size_t fragment_len)
 699 {
 700     bool more_frags = fragment_offset + fragment_len < pkt->payload_len;
 701     uint16_t orig_flags;
 702     struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG;
 703     struct ip_header *ip = l3hdr->iov_base;
 704     uint16_t frag_off_units = fragment_offset / IP_FRAG_UNIT_SIZE;
 705     uint16_t new_ip_off;
 706
 707     assert(fragment_offset % IP_FRAG_UNIT_SIZE == 0);
 708     assert((frag_off_units & ~IP_OFFMASK) == 0);
 709
 710     orig_flags = be16_to_cpu(ip->ip_off) & ~(IP_OFFMASK | IP_MF);
 711     new_ip_off = frag_off_units | orig_flags | (more_frags ? IP_MF : 0);
 712     ip->ip_off = cpu_to_be16(new_ip_off);
 713     ip->ip_len = cpu_to_be16(l3hdr->iov_len + fragment_len);
 714
 715     eth_fix_ip4_checksum(l3hdr->iov_base, l3hdr->iov_len);
 716 }
 717
 718 static bool net_tx_pkt_do_sw_fragmentation(struct NetTxPkt *pkt,
 719                                            NetTxPktSend callback,
 720                                            void *context)
 721 {
 722     uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
 723
 724     struct iovec fragment[NET_MAX_FRAG_SG_LIST];
 725     size_t fragment_len;
 726     size_t l4hdr_len;
 727     size_t src_len;
 728
 729     int src_idx, dst_idx, pl_idx;
 730     size_t src_offset;
 731     size_t fragment_offset = 0;
 732     struct virtio_net_hdr virt_hdr = {
 733         .flags = pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM ?
 734                  VIRTIO_NET_HDR_F_DATA_VALID : 0
 735     };
 736
 737     /* Copy headers */
 738     fragment[NET_TX_PKT_VHDR_FRAG].iov_base = &virt_hdr;
 739     fragment[NET_TX_PKT_VHDR_FRAG].iov_len = sizeof(virt_hdr);
 740     fragment[NET_TX_PKT_L2HDR_FRAG] = pkt->vec[NET_TX_PKT_L2HDR_FRAG];
 741     fragment[NET_TX_PKT_L3HDR_FRAG] = pkt->vec[NET_TX_PKT_L3HDR_FRAG];
 742
 743     switch (gso_type) {
 744     case VIRTIO_NET_HDR_GSO_TCPV4:
 745     case VIRTIO_NET_HDR_GSO_TCPV6:
 746         if (!net_tx_pkt_tcp_fragment_init(pkt, fragment, &pl_idx, &l4hdr_len,
 747                                           &src_idx, &src_offset, &src_len)) {
 748             return false;
 749         }
 750         break;
 751
 752     case VIRTIO_NET_HDR_GSO_UDP:
 753         net_tx_pkt_do_sw_csum(pkt, &pkt->vec[NET_TX_PKT_L2HDR_FRAG],
 754                               pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1,
 755                               pkt->payload_len);
 756         net_tx_pkt_udp_fragment_init(pkt, &pl_idx, &l4hdr_len,
 757                                      &src_idx, &src_offset, &src_len);
 758         break;
 759
 760     default:
 761         abort();
 762     }
 763
 764     /* Put as much data as possible and send */
 765     while (true) {
 766         dst_idx = pl_idx;
 767         fragment_len = net_tx_pkt_fetch_fragment(pkt,
 768             &src_idx, &src_offset, src_len, fragment, &dst_idx);
 769         if (!fragment_len) {
 770             break;
 771         }
 772
 773         switch (gso_type) {
 774         case VIRTIO_NET_HDR_GSO_TCPV4:
 775         case VIRTIO_NET_HDR_GSO_TCPV6:
 776             net_tx_pkt_tcp_fragment_fix(pkt, fragment, fragment_len, gso_type);
 777             net_tx_pkt_do_sw_csum(pkt, fragment + NET_TX_PKT_L2HDR_FRAG,
 778                                   dst_idx - NET_TX_PKT_L2HDR_FRAG,
 779                                   l4hdr_len + fragment_len);
 780             break;
 781
 782         case VIRTIO_NET_HDR_GSO_UDP:
 783             net_tx_pkt_udp_fragment_fix(pkt, fragment, fragment_offset,
 784                                         fragment_len);
 785             break;
 786         }
 787
 788         callback(context,
 789                  fragment + NET_TX_PKT_L2HDR_FRAG, dst_idx - NET_TX_PKT_L2HDR_FRAG,
 790                  fragment + NET_TX_PKT_VHDR_FRAG, dst_idx - NET_TX_PKT_VHDR_FRAG);
 791
 792         if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 ||
 793             gso_type == VIRTIO_NET_HDR_GSO_TCPV6) {
 794             net_tx_pkt_tcp_fragment_advance(pkt, fragment, fragment_len,
 795                                             gso_type);
 796         }
 797
 798         fragment_offset += fragment_len;
 799     }
 800
 801     if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 ||
 802         gso_type == VIRTIO_NET_HDR_GSO_TCPV6) {
 803         net_tx_pkt_tcp_fragment_deinit(fragment);
 804     }
 805
 806     return true;
 807 }
 808
 809 bool net_tx_pkt_send(struct NetTxPkt *pkt, NetClientState *nc)
 810 {
 811     bool offload = qemu_get_using_vnet_hdr(nc->peer);
 812     return net_tx_pkt_send_custom(pkt, offload, net_tx_pkt_sendv, nc);
 813 }
 814
 815 bool net_tx_pkt_send_custom(struct NetTxPkt *pkt, bool offload,
 816                             NetTxPktSend callback, void *context)
 817 {
 818     assert(pkt);
 819
 820     uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
 821
 822     /*
 823      * Since underlying infrastructure does not support IP datagrams longer
 824      * than 64K we should drop such packets and don't even try to send
 825      */
 826     if (VIRTIO_NET_HDR_GSO_NONE != gso_type) {
 827         if (pkt->payload_len >
 828             ETH_MAX_IP_DGRAM_LEN -
 829             pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len) {
 830             return false;
 831         }
 832     }
 833
 834     if (offload || gso_type == VIRTIO_NET_HDR_GSO_NONE) {
 835         if (!offload && pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 836             net_tx_pkt_do_sw_csum(pkt, &pkt->vec[NET_TX_PKT_L2HDR_FRAG],
 837                                   pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1,
 838                                   pkt->payload_len);
 839         }
 840
 841         net_tx_pkt_fix_ip6_payload_len(pkt);
 842         callback(context, pkt->vec + NET_TX_PKT_L2HDR_FRAG,
 843                  pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - NET_TX_PKT_L2HDR_FRAG,
 844                  pkt->vec + NET_TX_PKT_VHDR_FRAG,
 845                  pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - NET_TX_PKT_VHDR_FRAG);
 846         return true;
 847     }
 848
 849     return net_tx_pkt_do_sw_fragmentation(pkt, callback, context);
 850 }
 851
 852 void net_tx_pkt_fix_ip6_payload_len(struct NetTxPkt *pkt)
 853 {
 854     struct iovec *l2 = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
 855     if (eth_get_l3_proto(l2, 1, l2->iov_len) == ETH_P_IPV6) {
 856         /*
 857          * TODO: if qemu would support >64K packets - add jumbo option check
 858          * something like that:
 859          * 'if (ip6->ip6_plen == 0 && !has_jumbo_option(ip6)) {'
 860          */
 861         if (pkt->l3_hdr.ip6.ip6_plen == 0) {
 862             if (pkt->payload_len <= ETH_MAX_IP_DGRAM_LEN) {
 863                 pkt->l3_hdr.ip6.ip6_plen = htons(pkt->payload_len);
 864             }
 865             /*
 866              * TODO: if qemu would support >64K packets
 867              * add jumbo option for packets greater then 65,535 bytes
 868              */
 869         }
 870     }
 871 }