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