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
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <stdarg.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <signal.h>
  25 #include <inttypes.h>
  26 #include <fcntl.h>
  27 #include <errno.h>
  28 #include <pthread.h>
  29 #include <sys/socket.h>
  30 #include <sys/ioctl.h>
  31 #include <sys/types.h>
  32 #include <sys/stat.h>
  33 #include <sys/mman.h>
  34 #include <sys/wait.h>
  35
  36 #include <xs.h>
  37 #include <xenctrl.h>
  38 #include <xen/io/xenbus.h>
  39 #include <xen/io/netif.h>
  40
  41 #include "hw.h"
  42 #include "net.h"
  43 #include "net/checksum.h"
  44 #include "net/util.h"
  45 #include "qemu-char.h"
  46 #include "xen_backend.h"
  47
  48 /* ------------------------------------------------------------- */
  49
  50 struct XenNetDev {
  51     struct XenDevice      xendev;  /* must be first */
  52     char                  *mac;
  53     int                   tx_work;
  54     int                   tx_ring_ref;
  55     int                   rx_ring_ref;
  56     struct netif_tx_sring *txs;
  57     struct netif_rx_sring *rxs;
  58     netif_tx_back_ring_t  tx_ring;
  59     netif_rx_back_ring_t  rx_ring;
  60     NICConf               conf;
  61     NICState              *nic;
  62 };
  63
  64 /* ------------------------------------------------------------- */
  65
  66 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
  67 {
  68     RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
  69     netif_tx_response_t *resp;
  70     int notify;
  71
  72     resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
  73     resp->id     = txp->id;
  74     resp->status = st;
  75
  76 #if 0
  77     if (txp->flags & NETTXF_extra_info) {
  78         RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
  79     }
  80 #endif
  81
  82     netdev->tx_ring.rsp_prod_pvt = ++i;
  83     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  84     if (notify) {
  85         xen_be_send_notify(&netdev->xendev);
  86     }
  87
  88     if (i == netdev->tx_ring.req_cons) {
  89         int more_to_do;
  90         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  91         if (more_to_do) {
  92             netdev->tx_work++;
  93         }
  94     }
  95 }
  96
  97 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
  98 {
  99 #if 0
 100     /*
 101      * Hmm, why netback fails everything in the ring?
 102      * Should we do that even when not supporting SG and TSO?
 103      */
 104     RING_IDX cons = netdev->tx_ring.req_cons;
 105
 106     do {
 107         make_tx_response(netif, txp, NETIF_RSP_ERROR);
 108         if (cons >= end) {
 109             break;
 110         }
 111         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
 112     } while (1);
 113     netdev->tx_ring.req_cons = cons;
 114     netif_schedule_work(netif);
 115     netif_put(netif);
 116 #else
 117     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 118 #endif
 119 }
 120
 121 static void net_tx_packets(struct XenNetDev *netdev)
 122 {
 123     netif_tx_request_t txreq;
 124     RING_IDX rc, rp;
 125     void *page;
 126     void *tmpbuf = NULL;
 127
 128     for (;;) {
 129         rc = netdev->tx_ring.req_cons;
 130         rp = netdev->tx_ring.sring->req_prod;
 131         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 132
 133         while ((rc != rp)) {
 134             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
 135                 break;
 136             }
 137             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 138             netdev->tx_ring.req_cons = ++rc;
 139
 140 #if 1
 141             /* should not happen in theory, we don't announce the *
 142              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 143             if (txreq.flags & NETTXF_extra_info) {
 144                 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 145                 net_tx_error(netdev, &txreq, rc);
 146                 continue;
 147             }
 148             if (txreq.flags & NETTXF_more_data) {
 149                 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 150                 net_tx_error(netdev, &txreq, rc);
 151                 continue;
 152             }
 153 #endif
 154
 155             if (txreq.size < 14) {
 156                 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 157                 net_tx_error(netdev, &txreq, rc);
 158                 continue;
 159             }
 160
 161             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 162                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 163                 net_tx_error(netdev, &txreq, rc);
 164                 continue;
 165             }
 166
 167             xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 168                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 169                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 170                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 171                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 172                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 173
 174             page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 175                                            netdev->xendev.dom,
 176                                            txreq.gref, PROT_READ);
 177             if (page == NULL) {
 178                 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
 179                               txreq.gref);
 180                 net_tx_error(netdev, &txreq, rc);
 181                 continue;
 182             }
 183             if (txreq.flags & NETTXF_csum_blank) {
 184                 /* have read-only mapping -> can't fill checksum in-place */
 185                 if (!tmpbuf) {
 186                     tmpbuf = qemu_malloc(XC_PAGE_SIZE);
 187                 }
 188                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 189                 net_checksum_calculate(tmpbuf, txreq.size);
 190                 qemu_send_packet(&netdev->nic->nc, tmpbuf, txreq.size);
 191             } else {
 192                 qemu_send_packet(&netdev->nic->nc, page + txreq.offset, txreq.size);
 193             }
 194             xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 195             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 196         }
 197         if (!netdev->tx_work) {
 198             break;
 199         }
 200         netdev->tx_work = 0;
 201     }
 202     qemu_free(tmpbuf);
 203 }
 204
 205 /* ------------------------------------------------------------- */
 206
 207 static void net_rx_response(struct XenNetDev *netdev,
 208                             netif_rx_request_t *req, int8_t st,
 209                             uint16_t offset, uint16_t size,
 210                             uint16_t flags)
 211 {
 212     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 213     netif_rx_response_t *resp;
 214     int notify;
 215
 216     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 217     resp->offset     = offset;
 218     resp->flags      = flags;
 219     resp->id         = req->id;
 220     resp->status     = (int16_t)size;
 221     if (st < 0) {
 222         resp->status = (int16_t)st;
 223     }
 224
 225     xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 226                   i, resp->status, resp->flags);
 227
 228     netdev->rx_ring.rsp_prod_pvt = ++i;
 229     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 230     if (notify) {
 231         xen_be_send_notify(&netdev->xendev);
 232     }
 233 }
 234
 235 #define NET_IP_ALIGN 2
 236
 237 static int net_rx_ok(VLANClientState *nc)
 238 {
 239     struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
 240     RING_IDX rc, rp;
 241
 242     if (netdev->xendev.be_state != XenbusStateConnected) {
 243         return 0;
 244     }
 245
 246     rc = netdev->rx_ring.req_cons;
 247     rp = netdev->rx_ring.sring->req_prod;
 248     xen_rmb();
 249
 250     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 251         xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
 252                       __FUNCTION__, rc, rp);
 253         return 0;
 254     }
 255     return 1;
 256 }
 257
 258 static ssize_t net_rx_packet(VLANClientState *nc, const uint8_t *buf, size_t size)
 259 {
 260     struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
 261     netif_rx_request_t rxreq;
 262     RING_IDX rc, rp;
 263     void *page;
 264
 265     if (netdev->xendev.be_state != XenbusStateConnected) {
 266         return -1;
 267     }
 268
 269     rc = netdev->rx_ring.req_cons;
 270     rp = netdev->rx_ring.sring->req_prod;
 271     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 272
 273     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 274         xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
 275         return -1;
 276     }
 277     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 278         xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
 279                       (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
 280         return -1;
 281     }
 282
 283     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 284     netdev->rx_ring.req_cons = ++rc;
 285
 286     page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 287                                    netdev->xendev.dom,
 288                                    rxreq.gref, PROT_WRITE);
 289     if (page == NULL) {
 290         xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
 291                       rxreq.gref);
 292         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 293         return -1;
 294     }
 295     memcpy(page + NET_IP_ALIGN, buf, size);
 296     xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 297     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 298
 299     return size;
 300 }
 301
 302 /* ------------------------------------------------------------- */
 303
 304 static NetClientInfo net_xen_info = {
 305     .type = NET_CLIENT_TYPE_NIC,
 306     .size = sizeof(NICState),
 307     .can_receive = net_rx_ok,
 308     .receive = net_rx_packet,
 309 };
 310
 311 static int net_init(struct XenDevice *xendev)
 312 {
 313     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 314
 315     /* read xenstore entries */
 316     if (netdev->mac == NULL) {
 317         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 318     }
 319
 320     /* do we have all we need? */
 321     if (netdev->mac == NULL) {
 322         return -1;
 323     }
 324
 325     if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
 326         return -1;
 327     }
 328
 329     netdev->conf.vlan = qemu_find_vlan(netdev->xendev.dev, 1);
 330     netdev->conf.peer = NULL;
 331
 332     netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 333                                "xen", NULL, netdev);
 334
 335     snprintf(netdev->nic->nc.info_str, sizeof(netdev->nic->nc.info_str),
 336              "nic: xenbus vif macaddr=%s", netdev->mac);
 337
 338     /* fill info */
 339     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 340     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 341
 342     return 0;
 343 }
 344
 345 static int net_connect(struct XenDevice *xendev)
 346 {
 347     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 348     int rx_copy;
 349
 350     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 351                              &netdev->tx_ring_ref) == -1) {
 352         return -1;
 353     }
 354     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 355                              &netdev->rx_ring_ref) == -1) {
 356         return 1;
 357     }
 358     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 359                              &netdev->xendev.remote_port) == -1) {
 360         return -1;
 361     }
 362
 363     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 364         rx_copy = 0;
 365     }
 366     if (rx_copy == 0) {
 367         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 368         return -1;
 369     }
 370
 371     netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 372                                           netdev->xendev.dom,
 373                                           netdev->tx_ring_ref,
 374                                           PROT_READ | PROT_WRITE);
 375     netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 376                                           netdev->xendev.dom,
 377                                           netdev->rx_ring_ref,
 378                                           PROT_READ | PROT_WRITE);
 379     if (!netdev->txs || !netdev->rxs) {
 380         return -1;
 381     }
 382     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 383     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 384
 385     xen_be_bind_evtchn(&netdev->xendev);
 386
 387     xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 388                   "remote port %d, local port %d\n",
 389                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 390                   netdev->xendev.remote_port, netdev->xendev.local_port);
 391
 392     net_tx_packets(netdev);
 393     return 0;
 394 }
 395
 396 static void net_disconnect(struct XenDevice *xendev)
 397 {
 398     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 399
 400     xen_be_unbind_evtchn(&netdev->xendev);
 401
 402     if (netdev->txs) {
 403         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 404         netdev->txs = NULL;
 405     }
 406     if (netdev->rxs) {
 407         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 408         netdev->rxs = NULL;
 409     }
 410     if (netdev->nic) {
 411         qemu_del_vlan_client(&netdev->nic->nc);
 412         netdev->nic = NULL;
 413     }
 414 }
 415
 416 static void net_event(struct XenDevice *xendev)
 417 {
 418     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 419     net_tx_packets(netdev);
 420 }
 421
 422 static int net_free(struct XenDevice *xendev)
 423 {
 424     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 425
 426     qemu_free(netdev->mac);
 427     return 0;
 428 }
 429
 430 /* ------------------------------------------------------------- */
 431
 432 struct XenDevOps xen_netdev_ops = {
 433     .size       = sizeof(struct XenNetDev),
 434     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 435     .init       = net_init,
 436     .connect    = net_connect,
 437     .event      = net_event,
 438     .disconnect = net_disconnect,
 439     .free       = net_free,
 440 };