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_backend.h"
  32
  33 #include <xen/io/netif.h>
  34
  35 /* ------------------------------------------------------------- */
  36
  37 struct XenNetDev {
  38     struct XenDevice      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 = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 164                                            netdev->xendev.dom,
 165                                            txreq.gref, PROT_READ);
 166             if (page == NULL) {
 167                 xen_pv_printf(&netdev->xendev, 0,
 168                               "error: tx gref dereference failed (%d)\n",
 169                              txreq.gref);
 170                 net_tx_error(netdev, &txreq, rc);
 171                 continue;
 172             }
 173             if (txreq.flags & NETTXF_csum_blank) {
 174                 /* have read-only mapping -> can't fill checksum in-place */
 175                 if (!tmpbuf) {
 176                     tmpbuf = g_malloc(XC_PAGE_SIZE);
 177                 }
 178                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 179                 net_checksum_calculate(tmpbuf, txreq.size);
 180                 qemu_send_packet(qemu_get_queue(netdev->nic), tmpbuf,
 181                                  txreq.size);
 182             } else {
 183                 qemu_send_packet(qemu_get_queue(netdev->nic),
 184                                  page + txreq.offset, txreq.size);
 185             }
 186             xengnttab_unmap(netdev->xendev.gnttabdev, page, 1);
 187             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 188         }
 189         if (!netdev->tx_work) {
 190             break;
 191         }
 192         netdev->tx_work = 0;
 193     }
 194     g_free(tmpbuf);
 195 }
 196
 197 /* ------------------------------------------------------------- */
 198
 199 static void net_rx_response(struct XenNetDev *netdev,
 200                             netif_rx_request_t *req, int8_t st,
 201                             uint16_t offset, uint16_t size,
 202                             uint16_t flags)
 203 {
 204     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 205     netif_rx_response_t *resp;
 206     int notify;
 207
 208     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 209     resp->offset     = offset;
 210     resp->flags      = flags;
 211     resp->id         = req->id;
 212     resp->status     = (int16_t)size;
 213     if (st < 0) {
 214         resp->status = (int16_t)st;
 215     }
 216
 217     xen_pv_printf(&netdev->xendev, 3,
 218                   "rx response: idx %d, status %d, flags 0x%x\n",
 219                   i, resp->status, resp->flags);
 220
 221     netdev->rx_ring.rsp_prod_pvt = ++i;
 222     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 223     if (notify) {
 224         xen_pv_send_notify(&netdev->xendev);
 225     }
 226 }
 227
 228 #define NET_IP_ALIGN 2
 229
 230 static ssize_t net_rx_packet(NetClientState *nc, const uint8_t *buf, size_t size)
 231 {
 232     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 233     netif_rx_request_t rxreq;
 234     RING_IDX rc, rp;
 235     void *page;
 236
 237     if (netdev->xendev.be_state != XenbusStateConnected) {
 238         return -1;
 239     }
 240
 241     rc = netdev->rx_ring.req_cons;
 242     rp = netdev->rx_ring.sring->req_prod;
 243     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 244
 245     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 246         return 0;
 247     }
 248     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 249         xen_pv_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 250                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 251         return -1;
 252     }
 253
 254     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 255     netdev->rx_ring.req_cons = ++rc;
 256
 257     page = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 258                                    netdev->xendev.dom,
 259                                    rxreq.gref, PROT_WRITE);
 260     if (page == NULL) {
 261         xen_pv_printf(&netdev->xendev, 0,
 262                       "error: rx gref dereference failed (%d)\n",
 263                       rxreq.gref);
 264         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 265         return -1;
 266     }
 267     memcpy(page + NET_IP_ALIGN, buf, size);
 268     xengnttab_unmap(netdev->xendev.gnttabdev, page, 1);
 269     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 270
 271     return size;
 272 }
 273
 274 /* ------------------------------------------------------------- */
 275
 276 static NetClientInfo net_xen_info = {
 277     .type = NET_CLIENT_DRIVER_NIC,
 278     .size = sizeof(NICState),
 279     .receive = net_rx_packet,
 280 };
 281
 282 static int net_init(struct XenDevice *xendev)
 283 {
 284     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 285
 286     /* read xenstore entries */
 287     if (netdev->mac == NULL) {
 288         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 289     }
 290
 291     /* do we have all we need? */
 292     if (netdev->mac == NULL) {
 293         return -1;
 294     }
 295
 296     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 297         return -1;
 298     }
 299
 300     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 301                                "xen", NULL, netdev);
 302
 303     snprintf(qemu_get_queue(netdev->nic)->info_str,
 304              sizeof(qemu_get_queue(netdev->nic)->info_str),
 305              "nic: xenbus vif macaddr=%s", netdev->mac);
 306
 307     /* fill info */
 308     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 309     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 310
 311     return 0;
 312 }
 313
 314 static int net_connect(struct XenDevice *xendev)
 315 {
 316     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 317     int rx_copy;
 318
 319     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 320                              &netdev->tx_ring_ref) == -1) {
 321         return -1;
 322     }
 323     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 324                              &netdev->rx_ring_ref) == -1) {
 325         return 1;
 326     }
 327     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 328                              &netdev->xendev.remote_port) == -1) {
 329         return -1;
 330     }
 331
 332     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 333         rx_copy = 0;
 334     }
 335     if (rx_copy == 0) {
 336         xen_pv_printf(&netdev->xendev, 0,
 337                       "frontend doesn't support rx-copy.\n");
 338         return -1;
 339     }
 340
 341     netdev->txs = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 342                                           netdev->xendev.dom,
 343                                           netdev->tx_ring_ref,
 344                                           PROT_READ | PROT_WRITE);
 345     if (!netdev->txs) {
 346         return -1;
 347     }
 348     netdev->rxs = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 349                                           netdev->xendev.dom,
 350                                           netdev->rx_ring_ref,
 351                                           PROT_READ | PROT_WRITE);
 352     if (!netdev->rxs) {
 353         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 354         netdev->txs = NULL;
 355         return -1;
 356     }
 357     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 358     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 359
 360     xen_be_bind_evtchn(&netdev->xendev);
 361
 362     xen_pv_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 363                   "remote port %d, local port %d\n",
 364                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 365                   netdev->xendev.remote_port, netdev->xendev.local_port);
 366
 367     net_tx_packets(netdev);
 368     return 0;
 369 }
 370
 371 static void net_disconnect(struct XenDevice *xendev)
 372 {
 373     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 374
 375     xen_pv_unbind_evtchn(&netdev->xendev);
 376
 377     if (netdev->txs) {
 378         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 379         netdev->txs = NULL;
 380     }
 381     if (netdev->rxs) {
 382         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 383         netdev->rxs = NULL;
 384     }
 385 }
 386
 387 static void net_event(struct XenDevice *xendev)
 388 {
 389     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 390     net_tx_packets(netdev);
 391     qemu_flush_queued_packets(qemu_get_queue(netdev->nic));
 392 }
 393
 394 static int net_free(struct XenDevice *xendev)
 395 {
 396     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 397
 398     if (netdev->nic) {
 399         qemu_del_nic(netdev->nic);
 400         netdev->nic = NULL;
 401     }
 402     g_free(netdev->mac);
 403     netdev->mac = NULL;
 404     return 0;
 405 }
 406
 407 /* ------------------------------------------------------------- */
 408
 409 struct XenDevOps xen_netdev_ops = {
 410     .size       = sizeof(struct XenNetDev),
 411     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 412     .init       = net_init,
 413     .initialise    = net_connect,
 414     .event      = net_event,
 415     .disconnect = net_disconnect,
 416     .free       = net_free,
 417 };