hw/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 <signal.h>
  28 #include <inttypes.h>
  29 #include <fcntl.h>
  30 #include <errno.h>
  31 #include <sys/socket.h>
  32 #include <sys/ioctl.h>
  33 #include <sys/types.h>
  34 #include <sys/stat.h>
  35 #include <sys/mman.h>
  36 #include <sys/wait.h>
  37
  38 #include "hw.h"
  39 #include "net/net.h"
  40 #include "net/checksum.h"
  41 #include "net/util.h"
  42 #include "xen_backend.h"
  43
  44 #include <xen/io/netif.h>
  45
  46 /* ------------------------------------------------------------- */
  47
  48 struct XenNetDev {
  49     struct XenDevice      xendev;  /* must be first */
  50     char                  *mac;
  51     int                   tx_work;
  52     int                   tx_ring_ref;
  53     int                   rx_ring_ref;
  54     struct netif_tx_sring *txs;
  55     struct netif_rx_sring *rxs;
  56     netif_tx_back_ring_t  tx_ring;
  57     netif_rx_back_ring_t  rx_ring;
  58     NICConf               conf;
  59     NICState              *nic;
  60 };
  61
  62 /* ------------------------------------------------------------- */
  63
  64 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
  65 {
  66     RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
  67     netif_tx_response_t *resp;
  68     int notify;
  69
  70     resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
  71     resp->id     = txp->id;
  72     resp->status = st;
  73
  74 #if 0
  75     if (txp->flags & NETTXF_extra_info) {
  76         RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
  77     }
  78 #endif
  79
  80     netdev->tx_ring.rsp_prod_pvt = ++i;
  81     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  82     if (notify) {
  83         xen_be_send_notify(&netdev->xendev);
  84     }
  85
  86     if (i == netdev->tx_ring.req_cons) {
  87         int more_to_do;
  88         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  89         if (more_to_do) {
  90             netdev->tx_work++;
  91         }
  92     }
  93 }
  94
  95 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
  96 {
  97 #if 0
  98     /*
  99      * Hmm, why netback fails everything in the ring?
 100      * Should we do that even when not supporting SG and TSO?
 101      */
 102     RING_IDX cons = netdev->tx_ring.req_cons;
 103
 104     do {
 105         make_tx_response(netif, txp, NETIF_RSP_ERROR);
 106         if (cons >= end) {
 107             break;
 108         }
 109         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
 110     } while (1);
 111     netdev->tx_ring.req_cons = cons;
 112     netif_schedule_work(netif);
 113     netif_put(netif);
 114 #else
 115     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 116 #endif
 117 }
 118
 119 static void net_tx_packets(struct XenNetDev *netdev)
 120 {
 121     netif_tx_request_t txreq;
 122     RING_IDX rc, rp;
 123     void *page;
 124     void *tmpbuf = NULL;
 125
 126     for (;;) {
 127         rc = netdev->tx_ring.req_cons;
 128         rp = netdev->tx_ring.sring->req_prod;
 129         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 130
 131         while ((rc != rp)) {
 132             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
 133                 break;
 134             }
 135             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 136             netdev->tx_ring.req_cons = ++rc;
 137
 138 #if 1
 139             /* should not happen in theory, we don't announce the *
 140              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 141             if (txreq.flags & NETTXF_extra_info) {
 142                 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 143                 net_tx_error(netdev, &txreq, rc);
 144                 continue;
 145             }
 146             if (txreq.flags & NETTXF_more_data) {
 147                 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 148                 net_tx_error(netdev, &txreq, rc);
 149                 continue;
 150             }
 151 #endif
 152
 153             if (txreq.size < 14) {
 154                 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 155                 net_tx_error(netdev, &txreq, rc);
 156                 continue;
 157             }
 158
 159             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 160                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 161                 net_tx_error(netdev, &txreq, rc);
 162                 continue;
 163             }
 164
 165             xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 166                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 167                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 168                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 169                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 170                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 171
 172             page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 173                                            netdev->xendev.dom,
 174                                            txreq.gref, PROT_READ);
 175             if (page == NULL) {
 176                 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
 177                               txreq.gref);
 178                 net_tx_error(netdev, &txreq, rc);
 179                 continue;
 180             }
 181             if (txreq.flags & NETTXF_csum_blank) {
 182                 /* have read-only mapping -> can't fill checksum in-place */
 183                 if (!tmpbuf) {
 184                     tmpbuf = g_malloc(XC_PAGE_SIZE);
 185                 }
 186                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 187                 net_checksum_calculate(tmpbuf, txreq.size);
 188                 qemu_send_packet(&netdev->nic->nc, tmpbuf, txreq.size);
 189             } else {
 190                 qemu_send_packet(&netdev->nic->nc, page + txreq.offset, txreq.size);
 191             }
 192             xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 193             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 194         }
 195         if (!netdev->tx_work) {
 196             break;
 197         }
 198         netdev->tx_work = 0;
 199     }
 200     g_free(tmpbuf);
 201 }
 202
 203 /* ------------------------------------------------------------- */
 204
 205 static void net_rx_response(struct XenNetDev *netdev,
 206                             netif_rx_request_t *req, int8_t st,
 207                             uint16_t offset, uint16_t size,
 208                             uint16_t flags)
 209 {
 210     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 211     netif_rx_response_t *resp;
 212     int notify;
 213
 214     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 215     resp->offset     = offset;
 216     resp->flags      = flags;
 217     resp->id         = req->id;
 218     resp->status     = (int16_t)size;
 219     if (st < 0) {
 220         resp->status = (int16_t)st;
 221     }
 222
 223     xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 224                   i, resp->status, resp->flags);
 225
 226     netdev->rx_ring.rsp_prod_pvt = ++i;
 227     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 228     if (notify) {
 229         xen_be_send_notify(&netdev->xendev);
 230     }
 231 }
 232
 233 #define NET_IP_ALIGN 2
 234
 235 static int net_rx_ok(NetClientState *nc)
 236 {
 237     struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
 238     RING_IDX rc, rp;
 239
 240     if (netdev->xendev.be_state != XenbusStateConnected) {
 241         return 0;
 242     }
 243
 244     rc = netdev->rx_ring.req_cons;
 245     rp = netdev->rx_ring.sring->req_prod;
 246     xen_rmb();
 247
 248     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 249         xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
 250                       __FUNCTION__, rc, rp);
 251         return 0;
 252     }
 253     return 1;
 254 }
 255
 256 static ssize_t net_rx_packet(NetClientState *nc, const uint8_t *buf, size_t size)
 257 {
 258     struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
 259     netif_rx_request_t rxreq;
 260     RING_IDX rc, rp;
 261     void *page;
 262
 263     if (netdev->xendev.be_state != XenbusStateConnected) {
 264         return -1;
 265     }
 266
 267     rc = netdev->rx_ring.req_cons;
 268     rp = netdev->rx_ring.sring->req_prod;
 269     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 270
 271     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 272         xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
 273         return -1;
 274     }
 275     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 276         xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 277                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 278         return -1;
 279     }
 280
 281     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 282     netdev->rx_ring.req_cons = ++rc;
 283
 284     page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 285                                    netdev->xendev.dom,
 286                                    rxreq.gref, PROT_WRITE);
 287     if (page == NULL) {
 288         xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
 289                       rxreq.gref);
 290         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 291         return -1;
 292     }
 293     memcpy(page + NET_IP_ALIGN, buf, size);
 294     xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 295     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 296
 297     return size;
 298 }
 299
 300 /* ------------------------------------------------------------- */
 301
 302 static NetClientInfo net_xen_info = {
 303     .type = NET_CLIENT_OPTIONS_KIND_NIC,
 304     .size = sizeof(NICState),
 305     .can_receive = net_rx_ok,
 306     .receive = net_rx_packet,
 307 };
 308
 309 static int net_init(struct XenDevice *xendev)
 310 {
 311     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 312
 313     /* read xenstore entries */
 314     if (netdev->mac == NULL) {
 315         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 316     }
 317
 318     /* do we have all we need? */
 319     if (netdev->mac == NULL) {
 320         return -1;
 321     }
 322
 323     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 324         return -1;
 325     }
 326
 327     netdev->conf.peer = NULL;
 328
 329     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 330                                "xen", NULL, netdev);
 331
 332     snprintf(netdev->nic->nc.info_str, sizeof(netdev->nic->nc.info_str),
 333              "nic: xenbus vif macaddr=%s", netdev->mac);
 334
 335     /* fill info */
 336     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 337     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 338
 339     return 0;
 340 }
 341
 342 static int net_connect(struct XenDevice *xendev)
 343 {
 344     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 345     int rx_copy;
 346
 347     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 348                              &netdev->tx_ring_ref) == -1) {
 349         return -1;
 350     }
 351     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 352                              &netdev->rx_ring_ref) == -1) {
 353         return 1;
 354     }
 355     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 356                              &netdev->xendev.remote_port) == -1) {
 357         return -1;
 358     }
 359
 360     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 361         rx_copy = 0;
 362     }
 363     if (rx_copy == 0) {
 364         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 365         return -1;
 366     }
 367
 368     netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 369                                           netdev->xendev.dom,
 370                                           netdev->tx_ring_ref,
 371                                           PROT_READ | PROT_WRITE);
 372     netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 373                                           netdev->xendev.dom,
 374                                           netdev->rx_ring_ref,
 375                                           PROT_READ | PROT_WRITE);
 376     if (!netdev->txs || !netdev->rxs) {
 377         return -1;
 378     }
 379     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 380     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 381
 382     xen_be_bind_evtchn(&netdev->xendev);
 383
 384     xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 385                   "remote port %d, local port %d\n",
 386                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 387                   netdev->xendev.remote_port, netdev->xendev.local_port);
 388
 389     net_tx_packets(netdev);
 390     return 0;
 391 }
 392
 393 static void net_disconnect(struct XenDevice *xendev)
 394 {
 395     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 396
 397     xen_be_unbind_evtchn(&netdev->xendev);
 398
 399     if (netdev->txs) {
 400         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 401         netdev->txs = NULL;
 402     }
 403     if (netdev->rxs) {
 404         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 405         netdev->rxs = NULL;
 406     }
 407     if (netdev->nic) {
 408         qemu_del_net_client(&netdev->nic->nc);
 409         netdev->nic = NULL;
 410     }
 411 }
 412
 413 static void net_event(struct XenDevice *xendev)
 414 {
 415     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 416     net_tx_packets(netdev);
 417     qemu_flush_queued_packets(&netdev->nic->nc);
 418 }
 419
 420 static int net_free(struct XenDevice *xendev)
 421 {
 422     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 423
 424     g_free(netdev->mac);
 425     return 0;
 426 }
 427
 428 /* ------------------------------------------------------------- */
 429
 430 struct XenDevOps xen_netdev_ops = {
 431     .size       = sizeof(struct XenNetDev),
 432     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 433     .init       = net_init,
 434     .initialise    = net_connect,
 435     .event      = net_event,
 436     .disconnect = net_disconnect,
 437     .free       = net_free,
 438 };