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/mman.h>
  26 #include <sys/wait.h>
  27
  28 #include "hw/hw.h"
  29 #include "net/net.h"
  30 #include "net/checksum.h"
  31 #include "net/util.h"
  32 #include "hw/xen/xen_backend.h"
  33
  34 #include <xen/io/netif.h>
  35
  36 /* ------------------------------------------------------------- */
  37
  38 struct XenNetDev {
  39     struct XenDevice      xendev;  /* must be first */
  40     char                  *mac;
  41     int                   tx_work;
  42     int                   tx_ring_ref;
  43     int                   rx_ring_ref;
  44     struct netif_tx_sring *txs;
  45     struct netif_rx_sring *rxs;
  46     netif_tx_back_ring_t  tx_ring;
  47     netif_rx_back_ring_t  rx_ring;
  48     NICConf               conf;
  49     NICState              *nic;
  50 };
  51
  52 /* ------------------------------------------------------------- */
  53
  54 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
  55 {
  56     RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
  57     netif_tx_response_t *resp;
  58     int notify;
  59
  60     resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
  61     resp->id     = txp->id;
  62     resp->status = st;
  63
  64 #if 0
  65     if (txp->flags & NETTXF_extra_info) {
  66         RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
  67     }
  68 #endif
  69
  70     netdev->tx_ring.rsp_prod_pvt = ++i;
  71     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  72     if (notify) {
  73         xen_be_send_notify(&netdev->xendev);
  74     }
  75
  76     if (i == netdev->tx_ring.req_cons) {
  77         int more_to_do;
  78         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  79         if (more_to_do) {
  80             netdev->tx_work++;
  81         }
  82     }
  83 }
  84
  85 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
  86 {
  87 #if 0
  88     /*
  89      * Hmm, why netback fails everything in the ring?
  90      * Should we do that even when not supporting SG and TSO?
  91      */
  92     RING_IDX cons = netdev->tx_ring.req_cons;
  93
  94     do {
  95         make_tx_response(netif, txp, NETIF_RSP_ERROR);
  96         if (cons >= end) {
  97             break;
  98         }
  99         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
 100     } while (1);
 101     netdev->tx_ring.req_cons = cons;
 102     netif_schedule_work(netif);
 103     netif_put(netif);
 104 #else
 105     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 106 #endif
 107 }
 108
 109 static void net_tx_packets(struct XenNetDev *netdev)
 110 {
 111     netif_tx_request_t txreq;
 112     RING_IDX rc, rp;
 113     void *page;
 114     void *tmpbuf = NULL;
 115
 116     for (;;) {
 117         rc = netdev->tx_ring.req_cons;
 118         rp = netdev->tx_ring.sring->req_prod;
 119         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 120
 121         while ((rc != rp)) {
 122             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
 123                 break;
 124             }
 125             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 126             netdev->tx_ring.req_cons = ++rc;
 127
 128 #if 1
 129             /* should not happen in theory, we don't announce the *
 130              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 131             if (txreq.flags & NETTXF_extra_info) {
 132                 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 133                 net_tx_error(netdev, &txreq, rc);
 134                 continue;
 135             }
 136             if (txreq.flags & NETTXF_more_data) {
 137                 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 138                 net_tx_error(netdev, &txreq, rc);
 139                 continue;
 140             }
 141 #endif
 142
 143             if (txreq.size < 14) {
 144                 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 145                 net_tx_error(netdev, &txreq, rc);
 146                 continue;
 147             }
 148
 149             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 150                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 151                 net_tx_error(netdev, &txreq, rc);
 152                 continue;
 153             }
 154
 155             xen_be_printf(&netdev->xendev, 3, "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 = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 163                                            netdev->xendev.dom,
 164                                            txreq.gref, PROT_READ);
 165             if (page == NULL) {
 166                 xen_be_printf(&netdev->xendev, 0, "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             xengnttab_unmap(netdev->xendev.gnttabdev, page, 1);
 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_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 216                   i, resp->status, resp->flags);
 217
 218     netdev->rx_ring.rsp_prod_pvt = ++i;
 219     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 220     if (notify) {
 221         xen_be_send_notify(&netdev->xendev);
 222     }
 223 }
 224
 225 #define NET_IP_ALIGN 2
 226
 227 static ssize_t net_rx_packet(NetClientState *nc, const uint8_t *buf, size_t size)
 228 {
 229     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 230     netif_rx_request_t rxreq;
 231     RING_IDX rc, rp;
 232     void *page;
 233
 234     if (netdev->xendev.be_state != XenbusStateConnected) {
 235         return -1;
 236     }
 237
 238     rc = netdev->rx_ring.req_cons;
 239     rp = netdev->rx_ring.sring->req_prod;
 240     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 241
 242     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 243         return 0;
 244     }
 245     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 246         xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 247                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 248         return -1;
 249     }
 250
 251     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 252     netdev->rx_ring.req_cons = ++rc;
 253
 254     page = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 255                                    netdev->xendev.dom,
 256                                    rxreq.gref, PROT_WRITE);
 257     if (page == NULL) {
 258         xen_be_printf(&netdev->xendev, 0, "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     xengnttab_unmap(netdev->xendev.gnttabdev, page, 1);
 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_OPTIONS_KIND_NIC,
 274     .size = sizeof(NICState),
 275     .receive = net_rx_packet,
 276 };
 277
 278 static int net_init(struct XenDevice *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 XenDevice *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_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 333         return -1;
 334     }
 335
 336     netdev->txs = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 337                                           netdev->xendev.dom,
 338                                           netdev->tx_ring_ref,
 339                                           PROT_READ | PROT_WRITE);
 340     if (!netdev->txs) {
 341         return -1;
 342     }
 343     netdev->rxs = xengnttab_map_grant_ref(netdev->xendev.gnttabdev,
 344                                           netdev->xendev.dom,
 345                                           netdev->rx_ring_ref,
 346                                           PROT_READ | PROT_WRITE);
 347     if (!netdev->rxs) {
 348         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 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_be_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 XenDevice *xendev)
 367 {
 368     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 369
 370     xen_be_unbind_evtchn(&netdev->xendev);
 371
 372     if (netdev->txs) {
 373         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 374         netdev->txs = NULL;
 375     }
 376     if (netdev->rxs) {
 377         xengnttab_unmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 378         netdev->rxs = NULL;
 379     }
 380 }
 381
 382 static void net_event(struct XenDevice *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 XenDevice *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 };