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_size(pl_start_frag, pkt->payload_frags) < 8 + sizeof(csum)) {
 145         return false;
 146     }
 147
 148     if (iov_from_buf(pl_start_frag, pkt->payload_frags, 8, &csum, sizeof(csum)) < sizeof(csum)) {
 149         return false;
 150     }
 151
 152     csum = cpu_to_le32(iov_crc32c(0xffffffff, pl_start_frag, pkt->payload_frags));
 153     if (iov_from_buf(pl_start_frag, pkt->payload_frags, 8, &csum, sizeof(csum)) < sizeof(csum)) {
 154         return false;
 155     }
 156
 157     return true;
 158 }
 159
 160 static void net_tx_pkt_calculate_hdr_len(struct NetTxPkt *pkt)
 161 {
 162     pkt->hdr_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len +
 163         pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len;
 164 }
 165
 166 static bool net_tx_pkt_parse_headers(struct NetTxPkt *pkt)
 167 {
 168     struct iovec *l2_hdr, *l3_hdr;
 169     size_t bytes_read;
 170     size_t full_ip6hdr_len;
 171     uint16_t l3_proto;
 172
 173     assert(pkt);
 174
 175     l2_hdr = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
 176     l3_hdr = &pkt->vec[NET_TX_PKT_L3HDR_FRAG];
 177
 178     bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, 0, l2_hdr->iov_base,
 179                             ETH_MAX_L2_HDR_LEN);
 180     if (bytes_read < sizeof(struct eth_header)) {
 181         l2_hdr->iov_len = 0;
 182         return false;
 183     }
 184
 185     l2_hdr->iov_len = sizeof(struct eth_header);
 186     switch (be16_to_cpu(PKT_GET_ETH_HDR(l2_hdr->iov_base)->h_proto)) {
 187     case ETH_P_VLAN:
 188         l2_hdr->iov_len += sizeof(struct vlan_header);
 189         break;
 190     case ETH_P_DVLAN:
 191         l2_hdr->iov_len += 2 * sizeof(struct vlan_header);
 192         break;
 193     }
 194
 195     if (bytes_read < l2_hdr->iov_len) {
 196         l2_hdr->iov_len = 0;
 197         l3_hdr->iov_len = 0;
 198         pkt->packet_type = ETH_PKT_UCAST;
 199         return false;
 200     } else {
 201         l2_hdr->iov_len = ETH_MAX_L2_HDR_LEN;
 202         l2_hdr->iov_len = eth_get_l2_hdr_length(l2_hdr->iov_base);
 203         pkt->packet_type = get_eth_packet_type(l2_hdr->iov_base);
 204     }
 205
 206     l3_proto = eth_get_l3_proto(l2_hdr, 1, l2_hdr->iov_len);
 207
 208     switch (l3_proto) {
 209     case ETH_P_IP:
 210         bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
 211                                 l3_hdr->iov_base, sizeof(struct ip_header));
 212
 213         if (bytes_read < sizeof(struct ip_header)) {
 214             l3_hdr->iov_len = 0;
 215             return false;
 216         }
 217
 218         l3_hdr->iov_len = IP_HDR_GET_LEN(l3_hdr->iov_base);
 219
 220         if (l3_hdr->iov_len < sizeof(struct ip_header)) {
 221             l3_hdr->iov_len = 0;
 222             return false;
 223         }
 224
 225         pkt->l4proto = IP_HDR_GET_P(l3_hdr->iov_base);
 226
 227         if (IP_HDR_GET_LEN(l3_hdr->iov_base) != sizeof(struct ip_header)) {
 228             /* copy optional IPv4 header data if any*/
 229             bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags,
 230                                     l2_hdr->iov_len + sizeof(struct ip_header),
 231                                     l3_hdr->iov_base + sizeof(struct ip_header),
 232                                     l3_hdr->iov_len - sizeof(struct ip_header));
 233             if (bytes_read < l3_hdr->iov_len - sizeof(struct ip_header)) {
 234                 l3_hdr->iov_len = 0;
 235                 return false;
 236             }
 237         }
 238
 239         break;
 240
 241     case ETH_P_IPV6:
 242     {
 243         eth_ip6_hdr_info hdrinfo;
 244
 245         if (!eth_parse_ipv6_hdr(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
 246                                 &hdrinfo)) {
 247             l3_hdr->iov_len = 0;
 248             return false;
 249         }
 250
 251         pkt->l4proto = hdrinfo.l4proto;
 252         full_ip6hdr_len = hdrinfo.full_hdr_len;
 253
 254         if (full_ip6hdr_len > ETH_MAX_IP_DGRAM_LEN) {
 255             l3_hdr->iov_len = 0;
 256             return false;
 257         }
 258
 259         bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
 260                                 l3_hdr->iov_base, full_ip6hdr_len);
 261
 262         if (bytes_read < full_ip6hdr_len) {
 263             l3_hdr->iov_len = 0;
 264             return false;
 265         } else {
 266             l3_hdr->iov_len = full_ip6hdr_len;
 267         }
 268         break;
 269     }
 270     default:
 271         l3_hdr->iov_len = 0;
 272         break;
 273     }
 274
 275     net_tx_pkt_calculate_hdr_len(pkt);
 276     return true;
 277 }
 278
 279 static void net_tx_pkt_rebuild_payload(struct NetTxPkt *pkt)
 280 {
 281     pkt->payload_len = iov_size(pkt->raw, pkt->raw_frags) - pkt->hdr_len;
 282     pkt->payload_frags = iov_copy(&pkt->vec[NET_TX_PKT_PL_START_FRAG],
 283                                 pkt->max_payload_frags,
 284                                 pkt->raw, pkt->raw_frags,
 285                                 pkt->hdr_len, pkt->payload_len);
 286 }
 287
 288 bool net_tx_pkt_parse(struct NetTxPkt *pkt)
 289 {
 290     if (net_tx_pkt_parse_headers(pkt)) {
 291         net_tx_pkt_rebuild_payload(pkt);
 292         return true;
 293     } else {
 294         return false;
 295     }
 296 }
 297
 298 struct virtio_net_hdr *net_tx_pkt_get_vhdr(struct NetTxPkt *pkt)
 299 {
 300     assert(pkt);
 301     return &pkt->virt_hdr;
 302 }
 303
 304 static uint8_t net_tx_pkt_get_gso_type(struct NetTxPkt *pkt,
 305                                           bool tso_enable)
 306 {
 307     uint8_t rc = VIRTIO_NET_HDR_GSO_NONE;
 308     uint16_t l3_proto;
 309
 310     l3_proto = eth_get_l3_proto(&pkt->vec[NET_TX_PKT_L2HDR_FRAG], 1,
 311         pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len);
 312
 313     if (!tso_enable) {
 314         goto func_exit;
 315     }
 316
 317     rc = eth_get_gso_type(l3_proto, pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
 318                           pkt->l4proto);
 319
 320 func_exit:
 321     return rc;
 322 }
 323
 324 bool net_tx_pkt_build_vheader(struct NetTxPkt *pkt, bool tso_enable,
 325     bool csum_enable, uint32_t gso_size)
 326 {
 327     struct tcp_hdr l4hdr;
 328     size_t bytes_read;
 329     assert(pkt);
 330
 331     /* csum has to be enabled if tso is. */
 332     assert(csum_enable || !tso_enable);
 333
 334     pkt->virt_hdr.gso_type = net_tx_pkt_get_gso_type(pkt, tso_enable);
 335
 336     switch (pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 337     case VIRTIO_NET_HDR_GSO_NONE:
 338         pkt->virt_hdr.hdr_len = 0;
 339         pkt->virt_hdr.gso_size = 0;
 340         break;
 341
 342     case VIRTIO_NET_HDR_GSO_UDP:
 343         pkt->virt_hdr.gso_size = gso_size;
 344         pkt->virt_hdr.hdr_len = pkt->hdr_len + sizeof(struct udp_header);
 345         break;
 346
 347     case VIRTIO_NET_HDR_GSO_TCPV4:
 348     case VIRTIO_NET_HDR_GSO_TCPV6:
 349         bytes_read = iov_to_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG],
 350                                 pkt->payload_frags, 0, &l4hdr, sizeof(l4hdr));
 351         if (bytes_read < sizeof(l4hdr) ||
 352             l4hdr.th_off * sizeof(uint32_t) < sizeof(l4hdr)) {
 353             return false;
 354         }
 355
 356         pkt->virt_hdr.hdr_len = pkt->hdr_len + l4hdr.th_off * sizeof(uint32_t);
 357         pkt->virt_hdr.gso_size = gso_size;
 358         break;
 359
 360     default:
 361         g_assert_not_reached();
 362     }
 363
 364     if (csum_enable) {
 365         switch (pkt->l4proto) {
 366         case IP_PROTO_TCP:
 367             if (pkt->payload_len < sizeof(struct tcp_hdr)) {
 368                 return false;
 369             }
 370             pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 371             pkt->virt_hdr.csum_start = pkt->hdr_len;
 372             pkt->virt_hdr.csum_offset = offsetof(struct tcp_hdr, th_sum);
 373             break;
 374         case IP_PROTO_UDP:
 375             if (pkt->payload_len < sizeof(struct udp_hdr)) {
 376                 return false;
 377             }
 378             pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 379             pkt->virt_hdr.csum_start = pkt->hdr_len;
 380             pkt->virt_hdr.csum_offset = offsetof(struct udp_hdr, uh_sum);
 381             break;
 382         default:
 383             break;
 384         }
 385     }
 386
 387     return true;
 388 }
 389
 390 void net_tx_pkt_setup_vlan_header_ex(struct NetTxPkt *pkt,
 391     uint16_t vlan, uint16_t vlan_ethtype)
 392 {
 393     assert(pkt);
 394
 395     eth_setup_vlan_headers(pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base,
 396                            &pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len,
 397                            vlan, vlan_ethtype);
 398
 399     pkt->hdr_len += sizeof(struct vlan_header);
 400 }
 401
 402 bool net_tx_pkt_add_raw_fragment(struct NetTxPkt *pkt, void *base, size_t len)
 403 {
 404     struct iovec *ventry;
 405     assert(pkt);
 406
 407     if (pkt->raw_frags >= pkt->max_raw_frags) {
 408         return false;
 409     }
 410
 411     ventry = &pkt->raw[pkt->raw_frags];
 412     ventry->iov_base = base;
 413     ventry->iov_len = len;
 414     pkt->raw_frags++;
 415
 416     return true;
 417 }
 418
 419 bool net_tx_pkt_has_fragments(struct NetTxPkt *pkt)
 420 {
 421     return pkt->raw_frags > 0;
 422 }
 423
 424 eth_pkt_types_e net_tx_pkt_get_packet_type(struct NetTxPkt *pkt)
 425 {
 426     assert(pkt);
 427
 428     return pkt->packet_type;
 429 }
 430
 431 size_t net_tx_pkt_get_total_len(struct NetTxPkt *pkt)
 432 {
 433     assert(pkt);
 434
 435     return pkt->hdr_len + pkt->payload_len;
 436 }
 437
 438 void net_tx_pkt_dump(struct NetTxPkt *pkt)
 439 {
 440 #ifdef NET_TX_PKT_DEBUG
 441     assert(pkt);
 442
 443     printf("TX PKT: hdr_len: %d, pkt_type: 0x%X, l2hdr_len: %lu, "
 444         "l3hdr_len: %lu, payload_len: %u\n", pkt->hdr_len, pkt->packet_type,
 445         pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len,
 446         pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len, pkt->payload_len);
 447 #endif
 448 }
 449
 450 void net_tx_pkt_reset(struct NetTxPkt *pkt,
 451                       NetTxPktFreeFrag callback, void *context)
 452 {
 453     int i;
 454
 455     /* no assert, as reset can be called before tx_pkt_init */
 456     if (!pkt) {
 457         return;
 458     }
 459
 460     memset(&pkt->virt_hdr, 0, sizeof(pkt->virt_hdr));
 461
 462     assert(pkt->vec);
 463
 464     pkt->payload_len = 0;
 465     pkt->payload_frags = 0;
 466
 467     if (pkt->max_raw_frags > 0) {
 468         assert(pkt->raw);
 469         for (i = 0; i < pkt->raw_frags; i++) {
 470             assert(pkt->raw[i].iov_base);
 471             callback(context, pkt->raw[i].iov_base, pkt->raw[i].iov_len);
 472         }
 473     }
 474     pkt->raw_frags = 0;
 475
 476     pkt->hdr_len = 0;
 477     pkt->l4proto = 0;
 478 }
 479
 480 void net_tx_pkt_unmap_frag_pci(void *context, void *base, size_t len)
 481 {
 482     pci_dma_unmap(context, base, len, DMA_DIRECTION_TO_DEVICE, 0);
 483 }
 484
 485 bool net_tx_pkt_add_raw_fragment_pci(struct NetTxPkt *pkt, PCIDevice *pci_dev,
 486                                      dma_addr_t pa, size_t len)
 487 {
 488     dma_addr_t mapped_len = len;
 489     void *base = pci_dma_map(pci_dev, pa, &mapped_len, DMA_DIRECTION_TO_DEVICE);
 490     if (!base) {
 491         return false;
 492     }
 493
 494     if (mapped_len != len || !net_tx_pkt_add_raw_fragment(pkt, base, len)) {
 495         net_tx_pkt_unmap_frag_pci(pci_dev, base, mapped_len);
 496         return false;
 497     }
 498
 499     return true;
 500 }
 501
 502 static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt,
 503                                   struct iovec *iov, uint32_t iov_len,
 504                                   uint16_t csl)
 505 {
 506     uint32_t csum_cntr;
 507     uint16_t csum = 0;
 508     uint32_t cso;
 509     /* num of iovec without vhdr */
 510     size_t csum_offset = pkt->virt_hdr.csum_start + pkt->virt_hdr.csum_offset;
 511     uint16_t l3_proto = eth_get_l3_proto(iov, 1, iov->iov_len);
 512
 513     /* Put zero to checksum field */
 514     iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
 515
 516     /* Calculate L4 TCP/UDP checksum */
 517     csum_cntr = 0;
 518     cso = 0;
 519     /* add pseudo header to csum */
 520     if (l3_proto == ETH_P_IP) {
 521         csum_cntr = eth_calc_ip4_pseudo_hdr_csum(
 522                 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
 523                 csl, &cso);
 524     } else if (l3_proto == ETH_P_IPV6) {
 525         csum_cntr = eth_calc_ip6_pseudo_hdr_csum(
 526                 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
 527                 csl, pkt->l4proto, &cso);
 528     }
 529
 530     /* data checksum */
 531     csum_cntr +=
 532         net_checksum_add_iov(iov, iov_len, pkt->virt_hdr.csum_start, csl, cso);
 533
 534     /* Put the checksum obtained into the packet */
 535     csum = cpu_to_be16(net_checksum_finish_nozero(csum_cntr));
 536     iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
 537 }
 538
 539 #define NET_MAX_FRAG_SG_LIST (64)
 540
 541 static size_t net_tx_pkt_fetch_fragment(struct NetTxPkt *pkt,
 542     int *src_idx, size_t *src_offset, size_t src_len,
 543     struct iovec *dst, int *dst_idx)
 544 {
 545     size_t fetched = 0;
 546     struct iovec *src = pkt->vec;
 547
 548     while (fetched < src_len) {
 549
 550         /* no more place in fragment iov */
 551         if (*dst_idx == NET_MAX_FRAG_SG_LIST) {
 552             break;
 553         }
 554
 555         /* no more data in iovec */
 556         if (*src_idx == (pkt->payload_frags + NET_TX_PKT_PL_START_FRAG)) {
 557             break;
 558         }
 559
 560
 561         dst[*dst_idx].iov_base = src[*src_idx].iov_base + *src_offset;
 562         dst[*dst_idx].iov_len = MIN(src[*src_idx].iov_len - *src_offset,
 563             src_len - fetched);
 564
 565         *src_offset += dst[*dst_idx].iov_len;
 566         fetched += dst[*dst_idx].iov_len;
 567
 568         if (*src_offset == src[*src_idx].iov_len) {
 569             *src_offset = 0;
 570             (*src_idx)++;
 571         }
 572
 573         (*dst_idx)++;
 574     }
 575
 576     return fetched;
 577 }
 578
 579 static void net_tx_pkt_sendv(
 580     void *opaque, const struct iovec *iov, int iov_cnt,
 581     const struct iovec *virt_iov, int virt_iov_cnt)
 582 {
 583     NetClientState *nc = opaque;
 584
 585     if (qemu_get_using_vnet_hdr(nc->peer)) {
 586         qemu_sendv_packet(nc, virt_iov, virt_iov_cnt);
 587     } else {
 588         qemu_sendv_packet(nc, iov, iov_cnt);
 589     }
 590 }
 591
 592 static bool net_tx_pkt_tcp_fragment_init(struct NetTxPkt *pkt,
 593                                          struct iovec *fragment,
 594                                          int *pl_idx,
 595                                          size_t *l4hdr_len,
 596                                          int *src_idx,
 597                                          size_t *src_offset,
 598                                          size_t *src_len)
 599 {
 600     struct iovec *l4 = fragment + NET_TX_PKT_PL_START_FRAG;
 601     size_t bytes_read = 0;
 602     struct tcp_hdr *th;
 603
 604     if (!pkt->payload_frags) {
 605         return false;
 606     }
 607
 608     l4->iov_len = pkt->virt_hdr.hdr_len - pkt->hdr_len;
 609     l4->iov_base = g_malloc(l4->iov_len);
 610
 611     *src_idx = NET_TX_PKT_PL_START_FRAG;
 612     while (pkt->vec[*src_idx].iov_len < l4->iov_len - bytes_read) {
 613         memcpy((char *)l4->iov_base + bytes_read, pkt->vec[*src_idx].iov_base,
 614                pkt->vec[*src_idx].iov_len);
 615
 616         bytes_read += pkt->vec[*src_idx].iov_len;
 617
 618         (*src_idx)++;
 619         if (*src_idx >= pkt->payload_frags + NET_TX_PKT_PL_START_FRAG) {
 620             g_free(l4->iov_base);
 621             return false;
 622         }
 623     }
 624
 625     *src_offset = l4->iov_len - bytes_read;
 626     memcpy((char *)l4->iov_base + bytes_read, pkt->vec[*src_idx].iov_base,
 627            *src_offset);
 628
 629     th = l4->iov_base;
 630     th->th_flags &= ~(TH_FIN | TH_PUSH);
 631
 632     *pl_idx = NET_TX_PKT_PL_START_FRAG + 1;
 633     *l4hdr_len = l4->iov_len;
 634     *src_len = pkt->virt_hdr.gso_size;
 635
 636     return true;
 637 }
 638
 639 static void net_tx_pkt_tcp_fragment_deinit(struct iovec *fragment)
 640 {
 641     g_free(fragment[NET_TX_PKT_PL_START_FRAG].iov_base);
 642 }
 643
 644 static void net_tx_pkt_tcp_fragment_fix(struct NetTxPkt *pkt,
 645                                         struct iovec *fragment,
 646                                         size_t fragment_len,
 647                                         uint8_t gso_type)
 648 {
 649     struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG;
 650     struct iovec *l4hdr = fragment + NET_TX_PKT_PL_START_FRAG;
 651     struct ip_header *ip = l3hdr->iov_base;
 652     struct ip6_header *ip6 = l3hdr->iov_base;
 653     size_t len = l3hdr->iov_len + l4hdr->iov_len + fragment_len;
 654
 655     switch (gso_type) {
 656     case VIRTIO_NET_HDR_GSO_TCPV4:
 657         ip->ip_len = cpu_to_be16(len);
 658         eth_fix_ip4_checksum(l3hdr->iov_base, l3hdr->iov_len);
 659         break;
 660
 661     case VIRTIO_NET_HDR_GSO_TCPV6:
 662         len -= sizeof(struct ip6_header);
 663         ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = cpu_to_be16(len);
 664         break;
 665     }
 666 }
 667
 668 static void net_tx_pkt_tcp_fragment_advance(struct NetTxPkt *pkt,
 669                                             struct iovec *fragment,
 670                                             size_t fragment_len,
 671                                             uint8_t gso_type)
 672 {
 673     struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG;
 674     struct iovec *l4hdr = fragment + NET_TX_PKT_PL_START_FRAG;
 675     struct ip_header *ip = l3hdr->iov_base;
 676     struct tcp_hdr *th = l4hdr->iov_base;
 677
 678     if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4) {
 679         ip->ip_id = cpu_to_be16(be16_to_cpu(ip->ip_id) + 1);
 680     }
 681
 682     th->th_seq = cpu_to_be32(be32_to_cpu(th->th_seq) + fragment_len);
 683     th->th_flags &= ~TH_CWR;
 684 }
 685
 686 static void net_tx_pkt_udp_fragment_init(struct NetTxPkt *pkt,
 687                                          int *pl_idx,
 688                                          size_t *l4hdr_len,
 689                                          int *src_idx, size_t *src_offset,
 690                                          size_t *src_len)
 691 {
 692     *pl_idx = NET_TX_PKT_PL_START_FRAG;
 693     *l4hdr_len = 0;
 694     *src_idx = NET_TX_PKT_PL_START_FRAG;
 695     *src_offset = 0;
 696     *src_len = IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size);
 697 }
 698
 699 static void net_tx_pkt_udp_fragment_fix(struct NetTxPkt *pkt,
 700                                         struct iovec *fragment,
 701                                         size_t fragment_offset,
 702                                         size_t fragment_len)
 703 {
 704     bool more_frags = fragment_offset + fragment_len < pkt->payload_len;
 705     uint16_t orig_flags;
 706     struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG;
 707     struct ip_header *ip = l3hdr->iov_base;
 708     uint16_t frag_off_units = fragment_offset / IP_FRAG_UNIT_SIZE;
 709     uint16_t new_ip_off;
 710
 711     assert(fragment_offset % IP_FRAG_UNIT_SIZE == 0);
 712     assert((frag_off_units & ~IP_OFFMASK) == 0);
 713
 714     orig_flags = be16_to_cpu(ip->ip_off) & ~(IP_OFFMASK | IP_MF);
 715     new_ip_off = frag_off_units | orig_flags | (more_frags ? IP_MF : 0);
 716     ip->ip_off = cpu_to_be16(new_ip_off);
 717     ip->ip_len = cpu_to_be16(l3hdr->iov_len + fragment_len);
 718
 719     eth_fix_ip4_checksum(l3hdr->iov_base, l3hdr->iov_len);
 720 }
 721
 722 static bool net_tx_pkt_do_sw_fragmentation(struct NetTxPkt *pkt,
 723                                            NetTxPktSend callback,
 724                                            void *context)
 725 {
 726     uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
 727
 728     struct iovec fragment[NET_MAX_FRAG_SG_LIST];
 729     size_t fragment_len;
 730     size_t l4hdr_len;
 731     size_t src_len;
 732
 733     int src_idx, dst_idx, pl_idx;
 734     size_t src_offset;
 735     size_t fragment_offset = 0;
 736     struct virtio_net_hdr virt_hdr = {
 737         .flags = pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM ?
 738                  VIRTIO_NET_HDR_F_DATA_VALID : 0
 739     };
 740
 741     /* Copy headers */
 742     fragment[NET_TX_PKT_VHDR_FRAG].iov_base = &virt_hdr;
 743     fragment[NET_TX_PKT_VHDR_FRAG].iov_len = sizeof(virt_hdr);
 744     fragment[NET_TX_PKT_L2HDR_FRAG] = pkt->vec[NET_TX_PKT_L2HDR_FRAG];
 745     fragment[NET_TX_PKT_L3HDR_FRAG] = pkt->vec[NET_TX_PKT_L3HDR_FRAG];
 746
 747     switch (gso_type) {
 748     case VIRTIO_NET_HDR_GSO_TCPV4:
 749     case VIRTIO_NET_HDR_GSO_TCPV6:
 750         if (!net_tx_pkt_tcp_fragment_init(pkt, fragment, &pl_idx, &l4hdr_len,
 751                                           &src_idx, &src_offset, &src_len)) {
 752             return false;
 753         }
 754         break;
 755
 756     case VIRTIO_NET_HDR_GSO_UDP:
 757         net_tx_pkt_do_sw_csum(pkt, &pkt->vec[NET_TX_PKT_L2HDR_FRAG],
 758                               pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1,
 759                               pkt->payload_len);
 760         net_tx_pkt_udp_fragment_init(pkt, &pl_idx, &l4hdr_len,
 761                                      &src_idx, &src_offset, &src_len);
 762         break;
 763
 764     default:
 765         abort();
 766     }
 767
 768     /* Put as much data as possible and send */
 769     while (true) {
 770         dst_idx = pl_idx;
 771         fragment_len = net_tx_pkt_fetch_fragment(pkt,
 772             &src_idx, &src_offset, src_len, fragment, &dst_idx);
 773         if (!fragment_len) {
 774             break;
 775         }
 776
 777         switch (gso_type) {
 778         case VIRTIO_NET_HDR_GSO_TCPV4:
 779         case VIRTIO_NET_HDR_GSO_TCPV6:
 780             net_tx_pkt_tcp_fragment_fix(pkt, fragment, fragment_len, gso_type);
 781             net_tx_pkt_do_sw_csum(pkt, fragment + NET_TX_PKT_L2HDR_FRAG,
 782                                   dst_idx - NET_TX_PKT_L2HDR_FRAG,
 783                                   l4hdr_len + fragment_len);
 784             break;
 785
 786         case VIRTIO_NET_HDR_GSO_UDP:
 787             net_tx_pkt_udp_fragment_fix(pkt, fragment, fragment_offset,
 788                                         fragment_len);
 789             break;
 790         }
 791
 792         callback(context,
 793                  fragment + NET_TX_PKT_L2HDR_FRAG, dst_idx - NET_TX_PKT_L2HDR_FRAG,
 794                  fragment + NET_TX_PKT_VHDR_FRAG, dst_idx - NET_TX_PKT_VHDR_FRAG);
 795
 796         if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 ||
 797             gso_type == VIRTIO_NET_HDR_GSO_TCPV6) {
 798             net_tx_pkt_tcp_fragment_advance(pkt, fragment, fragment_len,
 799                                             gso_type);
 800         }
 801
 802         fragment_offset += fragment_len;
 803     }
 804
 805     if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 ||
 806         gso_type == VIRTIO_NET_HDR_GSO_TCPV6) {
 807         net_tx_pkt_tcp_fragment_deinit(fragment);
 808     }
 809
 810     return true;
 811 }
 812
 813 bool net_tx_pkt_send(struct NetTxPkt *pkt, NetClientState *nc)
 814 {
 815     bool offload = qemu_get_using_vnet_hdr(nc->peer);
 816     return net_tx_pkt_send_custom(pkt, offload, net_tx_pkt_sendv, nc);
 817 }
 818
 819 bool net_tx_pkt_send_custom(struct NetTxPkt *pkt, bool offload,
 820                             NetTxPktSend callback, void *context)
 821 {
 822     assert(pkt);
 823
 824     uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
 825
 826     /*
 827      * Since underlying infrastructure does not support IP datagrams longer
 828      * than 64K we should drop such packets and don't even try to send
 829      */
 830     if (VIRTIO_NET_HDR_GSO_NONE != gso_type) {
 831         if (pkt->payload_len >
 832             ETH_MAX_IP_DGRAM_LEN -
 833             pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len) {
 834             return false;
 835         }
 836     }
 837
 838     if (offload || gso_type == VIRTIO_NET_HDR_GSO_NONE) {
 839         if (!offload && pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 840             pkt->virt_hdr.flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 841             net_tx_pkt_do_sw_csum(pkt, &pkt->vec[NET_TX_PKT_L2HDR_FRAG],
 842                                   pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1,
 843                                   pkt->payload_len);
 844         }
 845
 846         net_tx_pkt_fix_ip6_payload_len(pkt);
 847         callback(context, pkt->vec + NET_TX_PKT_L2HDR_FRAG,
 848                  pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - NET_TX_PKT_L2HDR_FRAG,
 849                  pkt->vec + NET_TX_PKT_VHDR_FRAG,
 850                  pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - NET_TX_PKT_VHDR_FRAG);
 851         return true;
 852     }
 853
 854     return net_tx_pkt_do_sw_fragmentation(pkt, callback, context);
 855 }
 856
 857 void net_tx_pkt_fix_ip6_payload_len(struct NetTxPkt *pkt)
 858 {
 859     struct iovec *l2 = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
 860     if (eth_get_l3_proto(l2, 1, l2->iov_len) == ETH_P_IPV6) {
 861         /*
 862          * TODO: if qemu would support >64K packets - add jumbo option check
 863          * something like that:
 864          * 'if (ip6->ip6_plen == 0 && !has_jumbo_option(ip6)) {'
 865          */
 866         if (pkt->l3_hdr.ip6.ip6_plen == 0) {
 867             if (pkt->payload_len <= ETH_MAX_IP_DGRAM_LEN) {
 868                 pkt->l3_hdr.ip6.ip6_plen = htons(pkt->payload_len);
 869             }
 870             /*
 871              * TODO: if qemu would support >64K packets
 872              * add jumbo option for packets greater then 65,535 bytes
 873              */
 874         }
 875     }
 876 }