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, write to the Free Software Foundation, Inc.,
  17  *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <stdarg.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25 #include <signal.h>
  26 #include <inttypes.h>
  27 #include <fcntl.h>
  28 #include <errno.h>
  29 #include <pthread.h>
  30 #include <sys/socket.h>
  31 #include <sys/ioctl.h>
  32 #include <sys/types.h>
  33 #include <sys/stat.h>
  34 #include <sys/mman.h>
  35 #include <sys/wait.h>
  36
  37 #include <xs.h>
  38 #include <xenctrl.h>
  39 #include <xen/io/xenbus.h>
  40 #include <xen/io/netif.h>
  41
  42 #include "hw.h"
  43 #include "net.h"
  44 #include "qemu-char.h"
  45 #include "xen_backend.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     VLANClientState       *vs;
  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 #endif
  78
  79     netdev->tx_ring.rsp_prod_pvt = ++i;
  80     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
  81     if (notify)
  82         xen_be_send_notify(&netdev->xendev);
  83
  84     if (i == netdev->tx_ring.req_cons) {
  85         int more_to_do;
  86         RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
  87         if (more_to_do)
  88             netdev->tx_work++;
  89     }
  90 }
  91
  92 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
  93 {
  94 #if 0
  95     /*
  96      * Hmm, why netback fails everything in the ring?
  97      * Should we do that even when not supporting SG and TSO?
  98      */
  99     RING_IDX cons = netdev->tx_ring.req_cons;
 100
 101     do {
 102         make_tx_response(netif, txp, NETIF_RSP_ERROR);
 103         if (cons >= end)
 104             break;
 105         txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
 106     } while (1);
 107     netdev->tx_ring.req_cons = cons;
 108     netif_schedule_work(netif);
 109     netif_put(netif);
 110 #else
 111     net_tx_response(netdev, txp, NETIF_RSP_ERROR);
 112 #endif
 113 }
 114
 115 static void net_tx_packets(struct XenNetDev *netdev)
 116 {
 117     netif_tx_request_t txreq;
 118     RING_IDX rc, rp;
 119     void *page;
 120     void *tmpbuf = NULL;
 121
 122     for (;;) {
 123         rc = netdev->tx_ring.req_cons;
 124         rp = netdev->tx_ring.sring->req_prod;
 125         xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 126
 127         while ((rc != rp)) {
 128             if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc))
 129                 break;
 130             memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
 131             netdev->tx_ring.req_cons = ++rc;
 132
 133 #if 1
 134             /* should not happen in theory, we don't announce the *
 135              * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
 136             if (txreq.flags & NETTXF_extra_info) {
 137                 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
 138                 net_tx_error(netdev, &txreq, rc);
 139                 continue;
 140             }
 141             if (txreq.flags & NETTXF_more_data) {
 142                 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
 143                 net_tx_error(netdev, &txreq, rc);
 144                 continue;
 145             }
 146 #endif
 147
 148             if (txreq.size < 14) {
 149                 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
 150                 net_tx_error(netdev, &txreq, rc);
 151                 continue;
 152             }
 153
 154             if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
 155                 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
 156                 net_tx_error(netdev, &txreq, rc);
 157                 continue;
 158             }
 159
 160             xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
 161                           txreq.gref, txreq.offset, txreq.size, txreq.flags,
 162                           (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
 163                           (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
 164                           (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
 165                           (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
 166
 167             page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 168                                            netdev->xendev.dom,
 169                                            txreq.gref, PROT_READ);
 170             if (page == NULL) {
 171                 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
 172                               txreq.gref);
 173                 net_tx_error(netdev, &txreq, rc);
 174                 continue;
 175             }
 176             if (txreq.flags & NETTXF_csum_blank) {
 177                 /* have read-only mapping -> can't fill checksum in-place */
 178                 if (!tmpbuf)
 179                     tmpbuf = qemu_malloc(XC_PAGE_SIZE);
 180                 memcpy(tmpbuf, page + txreq.offset, txreq.size);
 181                 net_checksum_calculate(tmpbuf, txreq.size);
 182                 qemu_send_packet(netdev->vs, tmpbuf, txreq.size);
 183             } else {
 184                 qemu_send_packet(netdev->vs, page + txreq.offset, txreq.size);
 185             }
 186             xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 187             net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
 188         }
 189         if (!netdev->tx_work)
 190             break;
 191         netdev->tx_work = 0;
 192     }
 193     qemu_free(tmpbuf);
 194 }
 195
 196 /* ------------------------------------------------------------- */
 197
 198 static void net_rx_response(struct XenNetDev *netdev,
 199                             netif_rx_request_t *req, int8_t st,
 200                             uint16_t offset, uint16_t size,
 201                             uint16_t flags)
 202 {
 203     RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
 204     netif_rx_response_t *resp;
 205     int notify;
 206
 207     resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
 208     resp->offset     = offset;
 209     resp->flags      = flags;
 210     resp->id         = req->id;
 211     resp->status     = (int16_t)size;
 212     if (st < 0)
 213         resp->status = (int16_t)st;
 214
 215     xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
 216                   i, resp->status, resp->flags);
 217
 218     netdev->rx_ring.rsp_prod_pvt = ++i;
 219     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
 220     if (notify)
 221         xen_be_send_notify(&netdev->xendev);
 222 }
 223
 224 #define NET_IP_ALIGN 2
 225
 226 static int net_rx_ok(void *opaque)
 227 {
 228     struct XenNetDev *netdev = opaque;
 229     RING_IDX rc, rp;
 230
 231     if (netdev->xendev.be_state != XenbusStateConnected)
 232         return 0;
 233
 234     rc = netdev->rx_ring.req_cons;
 235     rp = netdev->rx_ring.sring->req_prod;
 236     xen_rmb();
 237
 238     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 239         xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
 240                       __FUNCTION__, rc, rp);
 241         return 0;
 242     }
 243     return 1;
 244 }
 245
 246 static void net_rx_packet(void *opaque, const uint8_t *buf, int size)
 247 {
 248     struct XenNetDev *netdev = opaque;
 249     netif_rx_request_t rxreq;
 250     RING_IDX rc, rp;
 251     void *page;
 252
 253     if (netdev->xendev.be_state != XenbusStateConnected)
 254         return;
 255
 256     rc = netdev->rx_ring.req_cons;
 257     rp = netdev->rx_ring.sring->req_prod;
 258     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 259
 260     if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
 261         xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
 262         return;
 263     }
 264     if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
 265         xen_be_printf(&netdev->xendev, 0, "packet too big (%d > %ld)",
 266                       size, XC_PAGE_SIZE - NET_IP_ALIGN);
 267         return;
 268     }
 269
 270     memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
 271     netdev->rx_ring.req_cons = ++rc;
 272
 273     page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 274                                    netdev->xendev.dom,
 275                                    rxreq.gref, PROT_WRITE);
 276     if (page == NULL) {
 277         xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
 278                       rxreq.gref);
 279         net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
 280         return;
 281     }
 282     memcpy(page + NET_IP_ALIGN, buf, size);
 283     xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
 284     net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
 285 }
 286
 287 /* ------------------------------------------------------------- */
 288
 289 static int net_init(struct XenDevice *xendev)
 290 {
 291     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 292     VLANState *vlan;
 293
 294     /* read xenstore entries */
 295     if (netdev->mac == NULL)
 296         netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
 297
 298     /* do we have all we need? */
 299     if (netdev->mac == NULL)
 300         return -1;
 301
 302     vlan = qemu_find_vlan(netdev->xendev.dev);
 303     netdev->vs = qemu_new_vlan_client(vlan, "xen", NULL,
 304                                       net_rx_packet, net_rx_ok, NULL,
 305                                       netdev);
 306     snprintf(netdev->vs->info_str, sizeof(netdev->vs->info_str),
 307              "nic: xenbus vif macaddr=%s", netdev->mac);
 308
 309     /* fill info */
 310     xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
 311     xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
 312
 313     return 0;
 314 }
 315
 316 static int net_connect(struct XenDevice *xendev)
 317 {
 318     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 319     int rx_copy;
 320
 321     if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
 322                                    &netdev->tx_ring_ref) == -1)
 323         return -1;
 324     if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
 325                                    &netdev->rx_ring_ref) == -1)
 326         return 1;
 327     if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
 328                                    &netdev->xendev.remote_port) == -1)
 329         return -1;
 330
 331     if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1)
 332         rx_copy = 0;
 333     if (rx_copy == 0) {
 334         xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
 335         return -1;
 336     }
 337
 338     netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 339                                           netdev->xendev.dom,
 340                                           netdev->tx_ring_ref,
 341                                           PROT_READ | PROT_WRITE);
 342     netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
 343                                           netdev->xendev.dom,
 344                                           netdev->rx_ring_ref,
 345                                           PROT_READ | PROT_WRITE);
 346     if (!netdev->txs || !netdev->rxs)
 347         return -1;
 348     BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
 349     BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
 350
 351     xen_be_bind_evtchn(&netdev->xendev);
 352
 353     xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
 354                   "remote port %d, local port %d\n",
 355                   netdev->tx_ring_ref, netdev->rx_ring_ref,
 356                   netdev->xendev.remote_port, netdev->xendev.local_port);
 357     return 0;
 358 }
 359
 360 static void net_disconnect(struct XenDevice *xendev)
 361 {
 362     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 363
 364     xen_be_unbind_evtchn(&netdev->xendev);
 365
 366     if (netdev->txs) {
 367         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
 368         netdev->txs = NULL;
 369     }
 370     if (netdev->rxs) {
 371         xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
 372         netdev->rxs = NULL;
 373     }
 374     if (netdev->vs) {
 375         qemu_del_vlan_client(netdev->vs);
 376         netdev->vs = NULL;
 377     }
 378 }
 379
 380 static void net_event(struct XenDevice *xendev)
 381 {
 382     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 383     net_tx_packets(netdev);
 384 }
 385
 386 static int net_free(struct XenDevice *xendev)
 387 {
 388     struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
 389
 390     qemu_free(netdev->mac);
 391     return 0;
 392 }
 393
 394 /* ------------------------------------------------------------- */
 395
 396 struct XenDevOps xen_netdev_ops = {
 397     .size       = sizeof(struct XenNetDev),
 398     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 399     .init       = net_init,
 400     .connect    = net_connect,
 401     .event      = net_event,
 402     .disconnect = net_disconnect,
 403     .free       = net_free,
 404 };