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 <stdio.h>
  23 #include <stdlib.h>
  24 #include <stdarg.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <inttypes.h>
  28 #include <fcntl.h>
  29 #include <errno.h>
  30 #include <sys/socket.h>
  31 #include <sys/ioctl.h>
  32 #include <sys/types.h>
  33 #include <sys/stat.h>
  34 #include <sys/mman.h>
  35 #include <sys/wait.h>
  36
  37 #include "hw/hw.h"
  38 #include "net/net.h"
  39 #include "net/checksum.h"
  40 #include "net/util.h"
  41 #include "hw/xen/xen_backend.h"
  42
  43 #include <xen/io/netif.h>
  44
  45 /* ------------------------------------------------------------- */
  46
  47 struct XenNetDev {
  48     struct XenDevice      xendev;  /* must be first */
  49     char                  *mac;
  50     int                   tx_work;
  51     int                   tx_ring_ref;
  52     int                   rx_ring_ref;
  53     struct netif_tx_sring *txs;
  54     struct netif_rx_sring *rxs;
  55     netif_tx_back_ring_t  tx_ring;
  56     netif_rx_back_ring_t  rx_ring;
  57     NICConf               conf;
  58     NICState              *nic;
  59 };
  60
  61 /* ------------------------------------------------------------- */
  62
  63 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
  64 {
  65     RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
  66     netif_tx_response_t *resp;
  67     int notify;
  68
  69     resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
  70     resp->id     = txp->id;
  71     resp->status = st;
  72
  73 #if 0
  74     if (txp->flags & NETTXF_extra_info) {
  75         RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
  76     }
  77 #endif
  78
  79     netdev->tx_ring.rsp_prod_pvt = ++i;
  80     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  81     if (notify) {
  82         xen_be_send_notify(&netdev->xendev);
  83     }
  84
  85     if (i == netdev->tx_ring.req_cons) {
  86         int more_to_do;
  87         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  88         if (more_to_do) {
  89             netdev->tx_work++;
  90         }
  91     }
  92 }
  93
  94 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
  95 {
  96 #if 0
  97     /*
  98      * Hmm, why netback fails everything in the ring?
  99      * Should we do that even when not supporting SG and TSO?
 100      */
 101     RING_IDX cons = netdev->tx_ring.req_cons;
 102
 103     do {
 104         make_tx_response(netif, txp, NETIF_RSP_ERROR);
 105         if (cons >= end) {
 106             break;
 107         }
 108         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
 109     } while (1);
 110     netdev->tx_ring.req_cons = cons;
 111     netif_schedule_work(netif);
 112     netif_put(netif);
 113 #else
 114     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 115 #endif
 116 }
 117
 118 static void net_tx_packets(struct XenNetDev *netdev)
 119 {
 120     netif_tx_request_t txreq;
 121     RING_IDX rc, rp;
 122     void *page;
 123     void *tmpbuf = NULL;
 124
 125     for (;;) {
 126         rc = netdev->tx_ring.req_cons;
 127         rp = netdev->tx_ring.sring->req_prod;
 128         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 129
 130         while ((rc != rp)) {
 131             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
 132                 break;
 133             }
 134             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 135             netdev->tx_ring.req_cons = ++rc;
 136
 137 #if 1
 138             /* should not happen in theory, we don't announce the *
 139              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 140             if (txreq.flags & NETTXF_extra_info) {
 141                 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 142                 net_tx_error(netdev, &txreq, rc);
 143                 continue;
 144             }
 145             if (txreq.flags & NETTXF_more_data) {
 146                 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 147                 net_tx_error(netdev, &txreq, rc);
 148                 continue;
 149             }
 150 #endif
 151
 152             if (txreq.size < 14) {
 153                 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 154                 net_tx_error(netdev, &txreq, rc);
 155                 continue;
 156             }
 157
 158             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 159                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 160                 net_tx_error(netdev, &txreq, rc);
 161                 continue;
 162             }
 163
 164             xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 165                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 166                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 167                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 168                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 169                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 170
 171             page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 172                                            netdev->xendev.dom,
 173                                            txreq.gref, PROT_READ);
 174             if (page == NULL) {
 175                 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
 176                               txreq.gref);
 177                 net_tx_error(netdev, &txreq, rc);
 178                 continue;
 179             }
 180             if (txreq.flags & NETTXF_csum_blank) {
 181                 /* have read-only mapping -> can't fill checksum in-place */
 182                 if (!tmpbuf) {
 183                     tmpbuf = g_malloc(XC_PAGE_SIZE);
 184                 }
 185                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 186                 net_checksum_calculate(tmpbuf, txreq.size);
 187                 qemu_send_packet(qemu_get_queue(netdev->nic), tmpbuf,
 188                                  txreq.size);
 189             } else {
 190                 qemu_send_packet(qemu_get_queue(netdev->nic),
 191                                  page + txreq.offset, txreq.size);
 192             }
 193             xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 194             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 195         }
 196         if (!netdev->tx_work) {
 197             break;
 198         }
 199         netdev->tx_work = 0;
 200     }
 201     g_free(tmpbuf);
 202 }
 203
 204 /* ------------------------------------------------------------- */
 205
 206 static void net_rx_response(struct XenNetDev *netdev,
 207                             netif_rx_request_t *req, int8_t st,
 208                             uint16_t offset, uint16_t size,
 209                             uint16_t flags)
 210 {
 211     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 212     netif_rx_response_t *resp;
 213     int notify;
 214
 215     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 216     resp->offset     = offset;
 217     resp->flags      = flags;
 218     resp->id         = req->id;
 219     resp->status     = (int16_t)size;
 220     if (st < 0) {
 221         resp->status = (int16_t)st;
 222     }
 223
 224     xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 225                   i, resp->status, resp->flags);
 226
 227     netdev->rx_ring.rsp_prod_pvt = ++i;
 228     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 229     if (notify) {
 230         xen_be_send_notify(&netdev->xendev);
 231     }
 232 }
 233
 234 #define NET_IP_ALIGN 2
 235
 236 static ssize_t net_rx_packet(NetClientState *nc, const uint8_t *buf, size_t size)
 237 {
 238     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 239     netif_rx_request_t rxreq;
 240     RING_IDX rc, rp;
 241     void *page;
 242
 243     if (netdev->xendev.be_state != XenbusStateConnected) {
 244         return -1;
 245     }
 246
 247     rc = netdev->rx_ring.req_cons;
 248     rp = netdev->rx_ring.sring->req_prod;
 249     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 250
 251     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 252         return 0;
 253     }
 254     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 255         xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 256                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 257         return -1;
 258     }
 259
 260     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 261     netdev->rx_ring.req_cons = ++rc;
 262
 263     page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 264                                    netdev->xendev.dom,
 265                                    rxreq.gref, PROT_WRITE);
 266     if (page == NULL) {
 267         xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
 268                       rxreq.gref);
 269         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 270         return -1;
 271     }
 272     memcpy(page + NET_IP_ALIGN, buf, size);
 273     xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 274     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 275
 276     return size;
 277 }
 278
 279 /* ------------------------------------------------------------- */
 280
 281 static NetClientInfo net_xen_info = {
 282     .type = NET_CLIENT_OPTIONS_KIND_NIC,
 283     .size = sizeof(NICState),
 284     .receive = net_rx_packet,
 285 };
 286
 287 static int net_init(struct XenDevice *xendev)
 288 {
 289     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 290
 291     /* read xenstore entries */
 292     if (netdev->mac == NULL) {
 293         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 294     }
 295
 296     /* do we have all we need? */
 297     if (netdev->mac == NULL) {
 298         return -1;
 299     }
 300
 301     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 302         return -1;
 303     }
 304
 305     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 306                                "xen", NULL, netdev);
 307
 308     snprintf(qemu_get_queue(netdev->nic)->info_str,
 309              sizeof(qemu_get_queue(netdev->nic)->info_str),
 310              "nic: xenbus vif macaddr=%s", netdev->mac);
 311
 312     /* fill info */
 313     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 314     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 315
 316     return 0;
 317 }
 318
 319 static int net_connect(struct XenDevice *xendev)
 320 {
 321     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 322     int rx_copy;
 323
 324     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 325                              &netdev->tx_ring_ref) == -1) {
 326         return -1;
 327     }
 328     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 329                              &netdev->rx_ring_ref) == -1) {
 330         return 1;
 331     }
 332     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 333                              &netdev->xendev.remote_port) == -1) {
 334         return -1;
 335     }
 336
 337     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 338         rx_copy = 0;
 339     }
 340     if (rx_copy == 0) {
 341         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 342         return -1;
 343     }
 344
 345     netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 346                                           netdev->xendev.dom,
 347                                           netdev->tx_ring_ref,
 348                                           PROT_READ | PROT_WRITE);
 349     if (!netdev->txs) {
 350         return -1;
 351     }
 352     netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 353                                           netdev->xendev.dom,
 354                                           netdev->rx_ring_ref,
 355                                           PROT_READ | PROT_WRITE);
 356     if (!netdev->rxs) {
 357         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 358         netdev->txs = NULL;
 359         return -1;
 360     }
 361     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 362     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 363
 364     xen_be_bind_evtchn(&netdev->xendev);
 365
 366     xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 367                   "remote port %d, local port %d\n",
 368                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 369                   netdev->xendev.remote_port, netdev->xendev.local_port);
 370
 371     net_tx_packets(netdev);
 372     return 0;
 373 }
 374
 375 static void net_disconnect(struct XenDevice *xendev)
 376 {
 377     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 378
 379     xen_be_unbind_evtchn(&netdev->xendev);
 380
 381     if (netdev->txs) {
 382         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 383         netdev->txs = NULL;
 384     }
 385     if (netdev->rxs) {
 386         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 387         netdev->rxs = NULL;
 388     }
 389 }
 390
 391 static void net_event(struct XenDevice *xendev)
 392 {
 393     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 394     net_tx_packets(netdev);
 395     qemu_flush_queued_packets(qemu_get_queue(netdev->nic));
 396 }
 397
 398 static int net_free(struct XenDevice *xendev)
 399 {
 400     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 401
 402     if (netdev->nic) {
 403         qemu_del_nic(netdev->nic);
 404         netdev->nic = NULL;
 405     }
 406     g_free(netdev->mac);
 407     netdev->mac = NULL;
 408     return 0;
 409 }
 410
 411 /* ------------------------------------------------------------- */
 412
 413 struct XenDevOps xen_netdev_ops = {
 414     .size       = sizeof(struct XenNetDev),
 415     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 416     .init       = net_init,
 417     .initialise    = net_connect,
 418     .event      = net_event,
 419     .disconnect = net_disconnect,
 420     .free       = net_free,
 421 };