hw/net/xen_nic.c

   1 /*
   2  *  xen paravirt network card backend
   3  *
   4  *  (c) Gerd Hoffmann <kraxel@redhat.com>
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; under version 2 of the License.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  You should have received a copy of the GNU General Public License along
  16  *  with this program; if not, see <http://www.gnu.org/licenses/>.
  17  *
  18  *  Contributions after 2012-01-13 are licensed under the terms of the
  19  *  GNU GPL, version 2 or (at your option) any later version.
  20  */
  21
  22 #include "qemu/osdep.h"
  23 #include <sys/socket.h>
  24 #include <sys/ioctl.h>
  25 #include <sys/wait.h>
  26
  27 #include "hw/hw.h"
  28 #include "net/net.h"
  29 #include "net/checksum.h"
  30 #include "net/util.h"
  31 #include "hw/xen/xen-legacy-backend.h"
  32
  33 #include <xen/io/netif.h>
  34
  35 /* ------------------------------------------------------------- */
  36
  37 struct XenNetDev {
  38     struct XenLegacyDevice      xendev;  /* must be first */
  39     char                  *mac;
  40     int                   tx_work;
  41     int                   tx_ring_ref;
  42     int                   rx_ring_ref;
  43     struct netif_tx_sring *txs;
  44     struct netif_rx_sring *rxs;
  45     netif_tx_back_ring_t  tx_ring;
  46     netif_rx_back_ring_t  rx_ring;
  47     NICConf               conf;
  48     NICState              *nic;
  49 };
  50
  51 /* ------------------------------------------------------------- */
  52
  53 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
  54 {
  55     RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
  56     netif_tx_response_t *resp;
  57     int notify;
  58
  59     resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
  60     resp->id     = txp->id;
  61     resp->status = st;
  62
  63 #if 0
  64     if (txp->flags & NETTXF_extra_info) {
  65         RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
  66     }
  67 #endif
  68
  69     netdev->tx_ring.rsp_prod_pvt = ++i;
  70     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  71     if (notify) {
  72         xen_pv_send_notify(&netdev->xendev);
  73     }
  74
  75     if (i == netdev->tx_ring.req_cons) {
  76         int more_to_do;
  77         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  78         if (more_to_do) {
  79             netdev->tx_work++;
  80         }
  81     }
  82 }
  83
  84 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
  85 {
  86 #if 0
  87     /*
  88      * Hmm, why netback fails everything in the ring?
  89      * Should we do that even when not supporting SG and TSO?
  90      */
  91     RING_IDX cons = netdev->tx_ring.req_cons;
  92
  93     do {
  94         make_tx_response(netif, txp, NETIF_RSP_ERROR);
  95         if (cons >= end) {
  96             break;
  97         }
  98         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
  99     } while (1);
 100     netdev->tx_ring.req_cons = cons;
 101     netif_schedule_work(netif);
 102     netif_put(netif);
 103 #else
 104     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 105 #endif
 106 }
 107
 108 static void net_tx_packets(struct XenNetDev *netdev)
 109 {
 110     netif_tx_request_t txreq;
 111     RING_IDX rc, rp;
 112     void *page;
 113     void *tmpbuf = NULL;
 114
 115     for (;;) {
 116         rc = netdev->tx_ring.req_cons;
 117         rp = netdev->tx_ring.sring->req_prod;
 118         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 119
 120         while ((rc != rp)) {
 121             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
 122                 break;
 123             }
 124             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 125             netdev->tx_ring.req_cons = ++rc;
 126
 127 #if 1
 128             /* should not happen in theory, we don't announce the *
 129              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 130             if (txreq.flags & NETTXF_extra_info) {
 131                 xen_pv_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 132                 net_tx_error(netdev, &txreq, rc);
 133                 continue;
 134             }
 135             if (txreq.flags & NETTXF_more_data) {
 136                 xen_pv_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 137                 net_tx_error(netdev, &txreq, rc);
 138                 continue;
 139             }
 140 #endif
 141
 142             if (txreq.size < 14) {
 143                 xen_pv_printf(&netdev->xendev, 0, "bad packet size: %d\n",
 144                               txreq.size);
 145                 net_tx_error(netdev, &txreq, rc);
 146                 continue;
 147             }
 148
 149             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 150                 xen_pv_printf(&netdev->xendev, 0, "error: page crossing\n");
 151                 net_tx_error(netdev, &txreq, rc);
 152                 continue;
 153             }
 154
 155             xen_pv_printf(&netdev->xendev, 3,
 156                           "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 157                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 158                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 159                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 160                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 161                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 162
 163             page = xen_be_map_grant_ref(&netdev->xendev, txreq.gref,
 164                                         PROT_READ);
 165             if (page == NULL) {
 166                 xen_pv_printf(&netdev->xendev, 0,
 167                               "error: tx gref dereference failed (%d)\n",
 168                              txreq.gref);
 169                 net_tx_error(netdev, &txreq, rc);
 170                 continue;
 171             }
 172             if (txreq.flags & NETTXF_csum_blank) {
 173                 /* have read-only mapping -> can't fill checksum in-place */
 174                 if (!tmpbuf) {
 175                     tmpbuf = g_malloc(XC_PAGE_SIZE);
 176                 }
 177                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 178                 net_checksum_calculate(tmpbuf, txreq.size);
 179                 qemu_send_packet(qemu_get_queue(netdev->nic), tmpbuf,
 180                                  txreq.size);
 181             } else {
 182                 qemu_send_packet(qemu_get_queue(netdev->nic),
 183                                  page + txreq.offset, txreq.size);
 184             }
 185             xen_be_unmap_grant_ref(&netdev->xendev, page);
 186             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 187         }
 188         if (!netdev->tx_work) {
 189             break;
 190         }
 191         netdev->tx_work = 0;
 192     }
 193     g_free(tmpbuf);
 194 }
 195
 196 /* ------------------------------------------------------------- */
 197
 198 static void net_rx_response(struct XenNetDev *netdev,
 199                             netif_rx_request_t *req, int8_t st,
 200                             uint16_t offset, uint16_t size,
 201                             uint16_t flags)
 202 {
 203     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 204     netif_rx_response_t *resp;
 205     int notify;
 206
 207     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 208     resp->offset     = offset;
 209     resp->flags      = flags;
 210     resp->id         = req->id;
 211     resp->status     = (int16_t)size;
 212     if (st < 0) {
 213         resp->status = (int16_t)st;
 214     }
 215
 216     xen_pv_printf(&netdev->xendev, 3,
 217                   "rx response: idx %d, status %d, flags 0x%x\n",
 218                   i, resp->status, resp->flags);
 219
 220     netdev->rx_ring.rsp_prod_pvt = ++i;
 221     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 222     if (notify) {
 223         xen_pv_send_notify(&netdev->xendev);
 224     }
 225 }
 226
 227 #define NET_IP_ALIGN 2
 228
 229 static ssize_t net_rx_packet(NetClientState *nc, const uint8_t *buf, size_t size)
 230 {
 231     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 232     netif_rx_request_t rxreq;
 233     RING_IDX rc, rp;
 234     void *page;
 235
 236     if (netdev->xendev.be_state != XenbusStateConnected) {
 237         return -1;
 238     }
 239
 240     rc = netdev->rx_ring.req_cons;
 241     rp = netdev->rx_ring.sring->req_prod;
 242     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 243
 244     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 245         return 0;
 246     }
 247     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 248         xen_pv_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 249                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 250         return -1;
 251     }
 252
 253     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 254     netdev->rx_ring.req_cons = ++rc;
 255
 256     page = xen_be_map_grant_ref(&netdev->xendev, rxreq.gref, PROT_WRITE);
 257     if (page == NULL) {
 258         xen_pv_printf(&netdev->xendev, 0,
 259                       "error: rx gref dereference failed (%d)\n",
 260                       rxreq.gref);
 261         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 262         return -1;
 263     }
 264     memcpy(page + NET_IP_ALIGN, buf, size);
 265     xen_be_unmap_grant_ref(&netdev->xendev, page);
 266     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 267
 268     return size;
 269 }
 270
 271 /* ------------------------------------------------------------- */
 272
 273 static NetClientInfo net_xen_info = {
 274     .type = NET_CLIENT_DRIVER_NIC,
 275     .size = sizeof(NICState),
 276     .receive = net_rx_packet,
 277 };
 278
 279 static int net_init(struct XenLegacyDevice *xendev)
 280 {
 281     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 282
 283     /* read xenstore entries */
 284     if (netdev->mac == NULL) {
 285         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 286     }
 287
 288     /* do we have all we need? */
 289     if (netdev->mac == NULL) {
 290         return -1;
 291     }
 292
 293     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 294         return -1;
 295     }
 296
 297     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 298                                "xen", NULL, netdev);
 299
 300     snprintf(qemu_get_queue(netdev->nic)->info_str,
 301              sizeof(qemu_get_queue(netdev->nic)->info_str),
 302              "nic: xenbus vif macaddr=%s", netdev->mac);
 303
 304     /* fill info */
 305     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 306     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 307
 308     return 0;
 309 }
 310
 311 static int net_connect(struct XenLegacyDevice *xendev)
 312 {
 313     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 314     int rx_copy;
 315
 316     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 317                              &netdev->tx_ring_ref) == -1) {
 318         return -1;
 319     }
 320     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 321                              &netdev->rx_ring_ref) == -1) {
 322         return 1;
 323     }
 324     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 325                              &netdev->xendev.remote_port) == -1) {
 326         return -1;
 327     }
 328
 329     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 330         rx_copy = 0;
 331     }
 332     if (rx_copy == 0) {
 333         xen_pv_printf(&netdev->xendev, 0,
 334                       "frontend doesn't support rx-copy.\n");
 335         return -1;
 336     }
 337
 338     netdev->txs = xen_be_map_grant_ref(&netdev->xendev,
 339                                        netdev->tx_ring_ref,
 340                                        PROT_READ | PROT_WRITE);
 341     if (!netdev->txs) {
 342         return -1;
 343     }
 344     netdev->rxs = xen_be_map_grant_ref(&netdev->xendev,
 345                                        netdev->rx_ring_ref,
 346                                        PROT_READ | PROT_WRITE);
 347     if (!netdev->rxs) {
 348         xen_be_unmap_grant_ref(&netdev->xendev, netdev->txs);
 349         netdev->txs = NULL;
 350         return -1;
 351     }
 352     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 353     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 354
 355     xen_be_bind_evtchn(&netdev->xendev);
 356
 357     xen_pv_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 358                   "remote port %d, local port %d\n",
 359                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 360                   netdev->xendev.remote_port, netdev->xendev.local_port);
 361
 362     net_tx_packets(netdev);
 363     return 0;
 364 }
 365
 366 static void net_disconnect(struct XenLegacyDevice *xendev)
 367 {
 368     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 369
 370     xen_pv_unbind_evtchn(&netdev->xendev);
 371
 372     if (netdev->txs) {
 373         xen_be_unmap_grant_ref(&netdev->xendev, netdev->txs);
 374         netdev->txs = NULL;
 375     }
 376     if (netdev->rxs) {
 377         xen_be_unmap_grant_ref(&netdev->xendev, netdev->rxs);
 378         netdev->rxs = NULL;
 379     }
 380 }
 381
 382 static void net_event(struct XenLegacyDevice *xendev)
 383 {
 384     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 385     net_tx_packets(netdev);
 386     qemu_flush_queued_packets(qemu_get_queue(netdev->nic));
 387 }
 388
 389 static int net_free(struct XenLegacyDevice *xendev)
 390 {
 391     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 392
 393     if (netdev->nic) {
 394         qemu_del_nic(netdev->nic);
 395         netdev->nic = NULL;
 396     }
 397     g_free(netdev->mac);
 398     netdev->mac = NULL;
 399     return 0;
 400 }
 401
 402 /* ------------------------------------------------------------- */
 403
 404 struct XenDevOps xen_netdev_ops = {
 405     .size       = sizeof(struct XenNetDev),
 406     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 407     .init       = net_init,
 408     .initialise    = net_connect,
 409     .event      = net_event,
 410     .disconnect = net_disconnect,
 411     .free       = net_free,
 412 };