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 <xs.h>
  39 #include <xenctrl.h>
  40 #include <xen/io/xenbus.h>
  41 #include <xen/io/netif.h>
  42
  43 #include "hw.h"
  44 #include "net.h"
  45 #include "net/checksum.h"
  46 #include "net/util.h"
  47 #include "qemu-char.h"
  48 #include "xen_backend.h"
  49
  50 /* ------------------------------------------------------------- */
  51
  52 struct XenNetDev {
  53     struct XenDevice      xendev;  /* must be first */
  54     char                  *mac;
  55     int                   tx_work;
  56     int                   tx_ring_ref;
  57     int                   rx_ring_ref;
  58     struct netif_tx_sring *txs;
  59     struct netif_rx_sring *rxs;
  60     netif_tx_back_ring_t  tx_ring;
  61     netif_rx_back_ring_t  rx_ring;
  62     NICConf               conf;
  63     NICState              *nic;
  64 };
  65
  66 /* ------------------------------------------------------------- */
  67
  68 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
  69 {
  70     RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
  71     netif_tx_response_t *resp;
  72     int notify;
  73
  74     resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
  75     resp->id     = txp->id;
  76     resp->status = st;
  77
  78 #if 0
  79     if (txp->flags & NETTXF_extra_info) {
  80         RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
  81     }
  82 #endif
  83
  84     netdev->tx_ring.rsp_prod_pvt = ++i;
  85     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  86     if (notify) {
  87         xen_be_send_notify(&netdev->xendev);
  88     }
  89
  90     if (i == netdev->tx_ring.req_cons) {
  91         int more_to_do;
  92         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  93         if (more_to_do) {
  94             netdev->tx_work++;
  95         }
  96     }
  97 }
  98
  99 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
 100 {
 101 #if 0
 102     /*
 103      * Hmm, why netback fails everything in the ring?
 104      * Should we do that even when not supporting SG and TSO?
 105      */
 106     RING_IDX cons = netdev->tx_ring.req_cons;
 107
 108     do {
 109         make_tx_response(netif, txp, NETIF_RSP_ERROR);
 110         if (cons >= end) {
 111             break;
 112         }
 113         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
 114     } while (1);
 115     netdev->tx_ring.req_cons = cons;
 116     netif_schedule_work(netif);
 117     netif_put(netif);
 118 #else
 119     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 120 #endif
 121 }
 122
 123 static void net_tx_packets(struct XenNetDev *netdev)
 124 {
 125     netif_tx_request_t txreq;
 126     RING_IDX rc, rp;
 127     void *page;
 128     void *tmpbuf = NULL;
 129
 130     for (;;) {
 131         rc = netdev->tx_ring.req_cons;
 132         rp = netdev->tx_ring.sring->req_prod;
 133         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 134
 135         while ((rc != rp)) {
 136             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
 137                 break;
 138             }
 139             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 140             netdev->tx_ring.req_cons = ++rc;
 141
 142 #if 1
 143             /* should not happen in theory, we don't announce the *
 144              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 145             if (txreq.flags & NETTXF_extra_info) {
 146                 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 147                 net_tx_error(netdev, &txreq, rc);
 148                 continue;
 149             }
 150             if (txreq.flags & NETTXF_more_data) {
 151                 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 152                 net_tx_error(netdev, &txreq, rc);
 153                 continue;
 154             }
 155 #endif
 156
 157             if (txreq.size < 14) {
 158                 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 159                 net_tx_error(netdev, &txreq, rc);
 160                 continue;
 161             }
 162
 163             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 164                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 165                 net_tx_error(netdev, &txreq, rc);
 166                 continue;
 167             }
 168
 169             xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 170                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 171                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 172                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 173                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 174                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 175
 176             page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 177                                            netdev->xendev.dom,
 178                                            txreq.gref, PROT_READ);
 179             if (page == NULL) {
 180                 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
 181                               txreq.gref);
 182                 net_tx_error(netdev, &txreq, rc);
 183                 continue;
 184             }
 185             if (txreq.flags & NETTXF_csum_blank) {
 186                 /* have read-only mapping -> can't fill checksum in-place */
 187                 if (!tmpbuf) {
 188                     tmpbuf = g_malloc(XC_PAGE_SIZE);
 189                 }
 190                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 191                 net_checksum_calculate(tmpbuf, txreq.size);
 192                 qemu_send_packet(&netdev->nic->nc, tmpbuf, txreq.size);
 193             } else {
 194                 qemu_send_packet(&netdev->nic->nc, page + txreq.offset, txreq.size);
 195             }
 196             xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 197             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 198         }
 199         if (!netdev->tx_work) {
 200             break;
 201         }
 202         netdev->tx_work = 0;
 203     }
 204     g_free(tmpbuf);
 205 }
 206
 207 /* ------------------------------------------------------------- */
 208
 209 static void net_rx_response(struct XenNetDev *netdev,
 210                             netif_rx_request_t *req, int8_t st,
 211                             uint16_t offset, uint16_t size,
 212                             uint16_t flags)
 213 {
 214     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 215     netif_rx_response_t *resp;
 216     int notify;
 217
 218     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 219     resp->offset     = offset;
 220     resp->flags      = flags;
 221     resp->id         = req->id;
 222     resp->status     = (int16_t)size;
 223     if (st < 0) {
 224         resp->status = (int16_t)st;
 225     }
 226
 227     xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 228                   i, resp->status, resp->flags);
 229
 230     netdev->rx_ring.rsp_prod_pvt = ++i;
 231     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 232     if (notify) {
 233         xen_be_send_notify(&netdev->xendev);
 234     }
 235 }
 236
 237 #define NET_IP_ALIGN 2
 238
 239 static int net_rx_ok(VLANClientState *nc)
 240 {
 241     struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
 242     RING_IDX rc, rp;
 243
 244     if (netdev->xendev.be_state != XenbusStateConnected) {
 245         return 0;
 246     }
 247
 248     rc = netdev->rx_ring.req_cons;
 249     rp = netdev->rx_ring.sring->req_prod;
 250     xen_rmb();
 251
 252     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 253         xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
 254                       __FUNCTION__, rc, rp);
 255         return 0;
 256     }
 257     return 1;
 258 }
 259
 260 static ssize_t net_rx_packet(VLANClientState *nc, const uint8_t *buf, size_t size)
 261 {
 262     struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
 263     netif_rx_request_t rxreq;
 264     RING_IDX rc, rp;
 265     void *page;
 266
 267     if (netdev->xendev.be_state != XenbusStateConnected) {
 268         return -1;
 269     }
 270
 271     rc = netdev->rx_ring.req_cons;
 272     rp = netdev->rx_ring.sring->req_prod;
 273     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 274
 275     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 276         xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
 277         return -1;
 278     }
 279     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 280         xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 281                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 282         return -1;
 283     }
 284
 285     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 286     netdev->rx_ring.req_cons = ++rc;
 287
 288     page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 289                                    netdev->xendev.dom,
 290                                    rxreq.gref, PROT_WRITE);
 291     if (page == NULL) {
 292         xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
 293                       rxreq.gref);
 294         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 295         return -1;
 296     }
 297     memcpy(page + NET_IP_ALIGN, buf, size);
 298     xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 299     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 300
 301     return size;
 302 }
 303
 304 /* ------------------------------------------------------------- */
 305
 306 static NetClientInfo net_xen_info = {
 307     .type = NET_CLIENT_TYPE_NIC,
 308     .size = sizeof(NICState),
 309     .can_receive = net_rx_ok,
 310     .receive = net_rx_packet,
 311 };
 312
 313 static int net_init(struct XenDevice *xendev)
 314 {
 315     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 316
 317     /* read xenstore entries */
 318     if (netdev->mac == NULL) {
 319         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 320     }
 321
 322     /* do we have all we need? */
 323     if (netdev->mac == NULL) {
 324         return -1;
 325     }
 326
 327     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 328         return -1;
 329     }
 330
 331     netdev->conf.vlan = qemu_find_vlan(netdev->xendev.dev, 1);
 332     netdev->conf.peer = NULL;
 333
 334     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 335                                "xen", NULL, netdev);
 336
 337     snprintf(netdev->nic->nc.info_str, sizeof(netdev->nic->nc.info_str),
 338              "nic: xenbus vif macaddr=%s", netdev->mac);
 339
 340     /* fill info */
 341     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 342     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 343
 344     return 0;
 345 }
 346
 347 static int net_connect(struct XenDevice *xendev)
 348 {
 349     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 350     int rx_copy;
 351
 352     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 353                              &netdev->tx_ring_ref) == -1) {
 354         return -1;
 355     }
 356     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 357                              &netdev->rx_ring_ref) == -1) {
 358         return 1;
 359     }
 360     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 361                              &netdev->xendev.remote_port) == -1) {
 362         return -1;
 363     }
 364
 365     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 366         rx_copy = 0;
 367     }
 368     if (rx_copy == 0) {
 369         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 370         return -1;
 371     }
 372
 373     netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 374                                           netdev->xendev.dom,
 375                                           netdev->tx_ring_ref,
 376                                           PROT_READ | PROT_WRITE);
 377     netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 378                                           netdev->xendev.dom,
 379                                           netdev->rx_ring_ref,
 380                                           PROT_READ | PROT_WRITE);
 381     if (!netdev->txs || !netdev->rxs) {
 382         return -1;
 383     }
 384     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 385     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 386
 387     xen_be_bind_evtchn(&netdev->xendev);
 388
 389     xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 390                   "remote port %d, local port %d\n",
 391                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 392                   netdev->xendev.remote_port, netdev->xendev.local_port);
 393
 394     net_tx_packets(netdev);
 395     return 0;
 396 }
 397
 398 static void net_disconnect(struct XenDevice *xendev)
 399 {
 400     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 401
 402     xen_be_unbind_evtchn(&netdev->xendev);
 403
 404     if (netdev->txs) {
 405         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 406         netdev->txs = NULL;
 407     }
 408     if (netdev->rxs) {
 409         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 410         netdev->rxs = NULL;
 411     }
 412     if (netdev->nic) {
 413         qemu_del_vlan_client(&netdev->nic->nc);
 414         netdev->nic = NULL;
 415     }
 416 }
 417
 418 static void net_event(struct XenDevice *xendev)
 419 {
 420     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 421     net_tx_packets(netdev);
 422 }
 423
 424 static int net_free(struct XenDevice *xendev)
 425 {
 426     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 427
 428     g_free(netdev->mac);
 429     return 0;
 430 }
 431
 432 /* ------------------------------------------------------------- */
 433
 434 struct XenDevOps xen_netdev_ops = {
 435     .size       = sizeof(struct XenNetDev),
 436     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 437     .init       = net_init,
 438     .initialise    = net_connect,
 439     .event      = net_event,
 440     .disconnect = net_disconnect,
 441     .free       = net_free,
 442 };