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