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/hw.h"
  39 #include "net/net.h"
  40 #include "net/checksum.h"
  41 #include "net/util.h"
  42 #include "hw/xen/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(qemu_get_queue(netdev->nic), tmpbuf,
 189                                  txreq.size);
 190             } else {
 191                 qemu_send_packet(qemu_get_queue(netdev->nic),
 192                                  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     g_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(NetClientState *nc)
 238 {
 239     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 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(NetClientState *nc, const uint8_t *buf, size_t size)
 259 {
 260     struct XenNetDev *netdev = qemu_get_nic_opaque(nc);
 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_OPTIONS_KIND_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->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
 330                                "xen", NULL, netdev);
 331
 332     snprintf(qemu_get_queue(netdev->nic)->info_str,
 333              sizeof(qemu_get_queue(netdev->nic)->info_str),
 334              "nic: xenbus vif macaddr=%s", netdev->mac);
 335
 336     /* fill info */
 337     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 338     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 339
 340     return 0;
 341 }
 342
 343 static int net_connect(struct XenDevice *xendev)
 344 {
 345     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 346     int rx_copy;
 347
 348     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 349                              &netdev->tx_ring_ref) == -1) {
 350         return -1;
 351     }
 352     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 353                              &netdev->rx_ring_ref) == -1) {
 354         return 1;
 355     }
 356     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 357                              &netdev->xendev.remote_port) == -1) {
 358         return -1;
 359     }
 360
 361     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
 362         rx_copy = 0;
 363     }
 364     if (rx_copy == 0) {
 365         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 366         return -1;
 367     }
 368
 369     netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 370                                           netdev->xendev.dom,
 371                                           netdev->tx_ring_ref,
 372                                           PROT_READ | PROT_WRITE);
 373     netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 374                                           netdev->xendev.dom,
 375                                           netdev->rx_ring_ref,
 376                                           PROT_READ | PROT_WRITE);
 377     if (!netdev->txs || !netdev->rxs) {
 378         return -1;
 379     }
 380     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 381     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 382
 383     xen_be_bind_evtchn(&netdev->xendev);
 384
 385     xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 386                   "remote port %d, local port %d\n",
 387                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 388                   netdev->xendev.remote_port, netdev->xendev.local_port);
 389
 390     net_tx_packets(netdev);
 391     return 0;
 392 }
 393
 394 static void net_disconnect(struct XenDevice *xendev)
 395 {
 396     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 397
 398     xen_be_unbind_evtchn(&netdev->xendev);
 399
 400     if (netdev->txs) {
 401         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 402         netdev->txs = NULL;
 403     }
 404     if (netdev->rxs) {
 405         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 406         netdev->rxs = NULL;
 407     }
 408     if (netdev->nic) {
 409         qemu_del_nic(netdev->nic);
 410         netdev->nic = NULL;
 411     }
 412 }
 413
 414 static void net_event(struct XenDevice *xendev)
 415 {
 416     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 417     net_tx_packets(netdev);
 418     qemu_flush_queued_packets(qemu_get_queue(netdev->nic));
 419 }
 420
 421 static int net_free(struct XenDevice *xendev)
 422 {
 423     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 424
 425     g_free(netdev->mac);
 426     return 0;
 427 }
 428
 429 /* ------------------------------------------------------------- */
 430
 431 struct XenDevOps xen_netdev_ops = {
 432     .size       = sizeof(struct XenNetDev),
 433     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 434     .init       = net_init,
 435     .initialise    = net_connect,
 436     .event      = net_event,
 437     .disconnect = net_disconnect,
 438     .free       = net_free,
 439 };