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