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_be_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_be_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_be_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_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 144                 net_tx_error(netdev, &txreq, rc);
 145                 continue;
 146             }
 147
 148             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 149                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 150                 net_tx_error(netdev, &txreq, rc);
 151                 continue;
 152             }
 153
 154             xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 155                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 156                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 157                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 158                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 159                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 160
 161             page = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 162                                            netdev->xendev.dom,
 163                                            txreq.gref, PROT_READ);
 164             if (page == NULL) {
 165                 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
 166                               txreq.gref);
 167                 net_tx_error(netdev, &txreq, rc);
 168                 continue;
 169             }
 170             if (txreq.flags & NETTXF_csum_blank) {
 171                 /* have read-only mapping -> can't fill checksum in-place */
 172                 if (!tmpbuf) {
 173                     tmpbuf = g_malloc(XC_PAGE_SIZE);
 174                 }
 175                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 176                 net_checksum_calculate(tmpbuf, txreq.size);
 177                 qemu_send_packet(qemu_get_queue(netdev->nic), tmpbuf,
 178                                  txreq.size);
 179             } else {
 180                 qemu_send_packet(qemu_get_queue(netdev->nic),
 181                                  page + txreq.offset, txreq.size);
 182             }
 183             xengnttab_unmap(netdev->xendev.gnttabdev, page, 1);
 184             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 185         }
 186         if (!netdev->tx_work) {
 187             break;
 188         }
 189         netdev->tx_work = 0;
 190     }
 191     g_free(tmpbuf);
 192 }
 193
 194 /* ------------------------------------------------------------- */
 195
 196 static void net_rx_response(struct XenNetDev *netdev,
 197                             netif_rx_request_t *req, int8_t st,
 198                             uint16_t offset, uint16_t size,
 199                             uint16_t flags)
 200 {
 201     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 202     netif_rx_response_t *resp;
 203     int notify;
 204
 205     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 206     resp->offset     = offset;
 207     resp->flags      = flags;
 208     resp->id         = req->id;
 209     resp->status     = (int16_t)size;
 210     if (st < 0) {
 211         resp->status = (int16_t)st;
 212     }
 213
 214     xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 215                   i, resp->status, resp->flags);
 216
 217     netdev->rx_ring.rsp_prod_pvt = ++i;
 218     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 219     if (notify) {
 220         xen_be_send_notify(&netdev->xendev);
 221     }
 222 }
 223
 224 #define NET_IP_ALIGN 2
 225
 226 static ssize_t net_rx_packet(NetClientState *nc, const uint8_t *buf, size_t size)
 227 {
 228     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 229     netif_rx_request_t rxreq;
 230     RING_IDX rc, rp;
 231     void *page;
 232
 233     if (netdev->xendev.be_state != XenbusStateConnected) {
 234         return -1;
 235     }
 236
 237     rc = netdev->rx_ring.req_cons;
 238     rp = netdev->rx_ring.sring->req_prod;
 239     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 240
 241     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 242         return 0;
 243     }
 244     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 245         xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 246                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 247         return -1;
 248     }
 249
 250     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 251     netdev->rx_ring.req_cons = ++rc;
 252
 253     page = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 254                                    netdev->xendev.dom,
 255                                    rxreq.gref, PROT_WRITE);
 256     if (page == NULL) {
 257         xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
 258                       rxreq.gref);
 259         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 260         return -1;
 261     }
 262     memcpy(page + NET_IP_ALIGN, buf, size);
 263     xengnttab_unmap(netdev->xendev.gnttabdev, page, 1);
 264     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 265
 266     return size;
 267 }
 268
 269 /* ------------------------------------------------------------- */
 270
 271 static NetClientInfo net_xen_info = {
 272     .type = NET_CLIENT_OPTIONS_KIND_NIC,
 273     .size = sizeof(NICState),
 274     .receive = net_rx_packet,
 275 };
 276
 277 static int net_init(struct XenDevice *xendev)
 278 {
 279     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 280
 281     /* read xenstore entries */
 282     if (netdev->mac == NULL) {
 283         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 284     }
 285
 286     /* do we have all we need? */
 287     if (netdev->mac == NULL) {
 288         return -1;
 289     }
 290
 291     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 292         return -1;
 293     }
 294
 295     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 296                                "xen", NULL, netdev);
 297
 298     snprintf(qemu_get_queue(netdev->nic)->info_str,
 299              sizeof(qemu_get_queue(netdev->nic)->info_str),
 300              "nic: xenbus vif macaddr=%s", netdev->mac);
 301
 302     /* fill info */
 303     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 304     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 305
 306     return 0;
 307 }
 308
 309 static int net_connect(struct XenDevice *xendev)
 310 {
 311     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 312     int rx_copy;
 313
 314     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 315                              &netdev->tx_ring_ref) == -1) {
 316         return -1;
 317     }
 318     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 319                              &netdev->rx_ring_ref) == -1) {
 320         return 1;
 321     }
 322     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 323                              &netdev->xendev.remote_port) == -1) {
 324         return -1;
 325     }
 326
 327     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 328         rx_copy = 0;
 329     }
 330     if (rx_copy == 0) {
 331         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 332         return -1;
 333     }
 334
 335     netdev->txs = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 336                                           netdev->xendev.dom,
 337                                           netdev->tx_ring_ref,
 338                                           PROT_READ | PROT_WRITE);
 339     if (!netdev->txs) {
 340         return -1;
 341     }
 342     netdev->rxs = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 343                                           netdev->xendev.dom,
 344                                           netdev->rx_ring_ref,
 345                                           PROT_READ | PROT_WRITE);
 346     if (!netdev->rxs) {
 347         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 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_be_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 XenDevice *xendev)
 366 {
 367     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 368
 369     xen_be_unbind_evtchn(&netdev->xendev);
 370
 371     if (netdev->txs) {
 372         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 373         netdev->txs = NULL;
 374     }
 375     if (netdev->rxs) {
 376         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 377         netdev->rxs = NULL;
 378     }
 379 }
 380
 381 static void net_event(struct XenDevice *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 XenDevice *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 };