net/netmap.c

   1 /*
   2  * netmap access for qemu
   3  *
   4  * Copyright (c) 2012-2013 Luigi Rizzo
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25
  26 #include "qemu/osdep.h"
  27 #include <sys/ioctl.h>
  28 #include <net/if.h>
  29 #include <sys/mman.h>
  30 #define NETMAP_WITH_LIBS
  31 #include <net/netmap.h>
  32 #include <net/netmap_user.h>
  33
  34 #include "net/net.h"
  35 #include "net/tap.h"
  36 #include "clients.h"
  37 #include "sysemu/sysemu.h"
  38 #include "qemu/error-report.h"
  39 #include "qapi/error.h"
  40 #include "qemu/iov.h"
  41 #include "qemu/cutils.h"
  42
  43 typedef struct NetmapState {
  44     NetClientState      nc;
  45     struct nm_desc      *nmd;
  46     char                ifname[IFNAMSIZ];
  47     struct netmap_ring  *tx;
  48     struct netmap_ring  *rx;
  49     bool                read_poll;
  50     bool                write_poll;
  51     struct iovec        iov[IOV_MAX];
  52     int                 vnet_hdr_len;  /* Current virtio-net header length. */
  53 } NetmapState;
  54
  55 #ifndef __FreeBSD__
  56 #define pkt_copy bcopy
  57 #else
  58 /* A fast copy routine only for multiples of 64 bytes, non overlapped. */
  59 static inline void
  60 pkt_copy(const void *_src, void *_dst, int l)
  61 {
  62     const uint64_t *src = _src;
  63     uint64_t *dst = _dst;
  64     if (unlikely(l >= 1024)) {
  65         bcopy(src, dst, l);
  66         return;
  67     }
  68     for (; l > 0; l -= 64) {
  69         *dst++ = *src++;
  70         *dst++ = *src++;
  71         *dst++ = *src++;
  72         *dst++ = *src++;
  73         *dst++ = *src++;
  74         *dst++ = *src++;
  75         *dst++ = *src++;
  76         *dst++ = *src++;
  77     }
  78 }
  79 #endif /* __FreeBSD__ */
  80
  81 /*
  82  * Open a netmap device. We assume there is only one queue
  83  * (which is the case for the VALE bridge).
  84  */
  85 static struct nm_desc *netmap_open(const NetdevNetmapOptions *nm_opts,
  86                                    Error **errp)
  87 {
  88     struct nm_desc *nmd;
  89     struct nmreq req;
  90
  91     memset(&req, 0, sizeof(req));
  92
  93     nmd = nm_open(nm_opts->ifname, &req, NETMAP_NO_TX_POLL,
  94                   NULL);
  95     if (nmd == NULL) {
  96         error_setg_errno(errp, errno, "Failed to nm_open() %s",
  97                          nm_opts->ifname);
  98         return NULL;
  99     }
 100
 101     return nmd;
 102 }
 103
 104 static void netmap_send(void *opaque);
 105 static void netmap_writable(void *opaque);
 106
 107 /* Set the event-loop handlers for the netmap backend. */
 108 static void netmap_update_fd_handler(NetmapState *s)
 109 {
 110     qemu_set_fd_handler(s->nmd->fd,
 111                         s->read_poll ? netmap_send : NULL,
 112                         s->write_poll ? netmap_writable : NULL,
 113                         s);
 114 }
 115
 116 /* Update the read handler. */
 117 static void netmap_read_poll(NetmapState *s, bool enable)
 118 {
 119     if (s->read_poll != enable) { /* Do nothing if not changed. */
 120         s->read_poll = enable;
 121         netmap_update_fd_handler(s);
 122     }
 123 }
 124
 125 /* Update the write handler. */
 126 static void netmap_write_poll(NetmapState *s, bool enable)
 127 {
 128     if (s->write_poll != enable) {
 129         s->write_poll = enable;
 130         netmap_update_fd_handler(s);
 131     }
 132 }
 133
 134 static void netmap_poll(NetClientState *nc, bool enable)
 135 {
 136     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 137
 138     if (s->read_poll != enable || s->write_poll != enable) {
 139         s->write_poll = enable;
 140         s->read_poll  = enable;
 141         netmap_update_fd_handler(s);
 142     }
 143 }
 144
 145 /*
 146  * The fd_write() callback, invoked if the fd is marked as
 147  * writable after a poll. Unregister the handler and flush any
 148  * buffered packets.
 149  */
 150 static void netmap_writable(void *opaque)
 151 {
 152     NetmapState *s = opaque;
 153
 154     netmap_write_poll(s, false);
 155     qemu_flush_queued_packets(&s->nc);
 156 }
 157
 158 static ssize_t netmap_receive(NetClientState *nc,
 159       const uint8_t *buf, size_t size)
 160 {
 161     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 162     struct netmap_ring *ring = s->tx;
 163     uint32_t i;
 164     uint32_t idx;
 165     uint8_t *dst;
 166
 167     if (unlikely(!ring)) {
 168         /* Drop. */
 169         return size;
 170     }
 171
 172     if (unlikely(size > ring->nr_buf_size)) {
 173         RD(5, "[netmap_receive] drop packet of size %d > %d\n",
 174                                     (int)size, ring->nr_buf_size);
 175         return size;
 176     }
 177
 178     if (nm_ring_empty(ring)) {
 179         /* No available slots in the netmap TX ring. */
 180         netmap_write_poll(s, true);
 181         return 0;
 182     }
 183
 184     i = ring->cur;
 185     idx = ring->slot[i].buf_idx;
 186     dst = (uint8_t *)NETMAP_BUF(ring, idx);
 187
 188     ring->slot[i].len = size;
 189     ring->slot[i].flags = 0;
 190     pkt_copy(buf, dst, size);
 191     ring->cur = ring->head = nm_ring_next(ring, i);
 192     ioctl(s->nmd->fd, NIOCTXSYNC, NULL);
 193
 194     return size;
 195 }
 196
 197 static ssize_t netmap_receive_iov(NetClientState *nc,
 198                     const struct iovec *iov, int iovcnt)
 199 {
 200     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 201     struct netmap_ring *ring = s->tx;
 202     uint32_t last;
 203     uint32_t idx;
 204     uint8_t *dst;
 205     int j;
 206     uint32_t i;
 207
 208     if (unlikely(!ring)) {
 209         /* Drop the packet. */
 210         return iov_size(iov, iovcnt);
 211     }
 212
 213     last = i = ring->cur;
 214
 215     if (nm_ring_space(ring) < iovcnt) {
 216         /* Not enough netmap slots. */
 217         netmap_write_poll(s, true);
 218         return 0;
 219     }
 220
 221     for (j = 0; j < iovcnt; j++) {
 222         int iov_frag_size = iov[j].iov_len;
 223         int offset = 0;
 224         int nm_frag_size;
 225
 226         /* Split each iovec fragment over more netmap slots, if
 227            necessary. */
 228         while (iov_frag_size) {
 229             nm_frag_size = MIN(iov_frag_size, ring->nr_buf_size);
 230
 231             if (unlikely(nm_ring_empty(ring))) {
 232                 /* We run out of netmap slots while splitting the
 233                    iovec fragments. */
 234                 netmap_write_poll(s, true);
 235                 return 0;
 236             }
 237
 238             idx = ring->slot[i].buf_idx;
 239             dst = (uint8_t *)NETMAP_BUF(ring, idx);
 240
 241             ring->slot[i].len = nm_frag_size;
 242             ring->slot[i].flags = NS_MOREFRAG;
 243             pkt_copy(iov[j].iov_base + offset, dst, nm_frag_size);
 244
 245             last = i;
 246             i = nm_ring_next(ring, i);
 247
 248             offset += nm_frag_size;
 249             iov_frag_size -= nm_frag_size;
 250         }
 251     }
 252     /* The last slot must not have NS_MOREFRAG set. */
 253     ring->slot[last].flags &= ~NS_MOREFRAG;
 254
 255     /* Now update ring->cur and ring->head. */
 256     ring->cur = ring->head = i;
 257
 258     ioctl(s->nmd->fd, NIOCTXSYNC, NULL);
 259
 260     return iov_size(iov, iovcnt);
 261 }
 262
 263 /* Complete a previous send (backend --> guest) and enable the
 264    fd_read callback. */
 265 static void netmap_send_completed(NetClientState *nc, ssize_t len)
 266 {
 267     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 268
 269     netmap_read_poll(s, true);
 270 }
 271
 272 static void netmap_send(void *opaque)
 273 {
 274     NetmapState *s = opaque;
 275     struct netmap_ring *ring = s->rx;
 276
 277     /* Keep sending while there are available packets into the netmap
 278        RX ring and the forwarding path towards the peer is open. */
 279     while (!nm_ring_empty(ring)) {
 280         uint32_t i;
 281         uint32_t idx;
 282         bool morefrag;
 283         int iovcnt = 0;
 284         int iovsize;
 285
 286         do {
 287             i = ring->cur;
 288             idx = ring->slot[i].buf_idx;
 289             morefrag = (ring->slot[i].flags & NS_MOREFRAG);
 290             s->iov[iovcnt].iov_base = (u_char *)NETMAP_BUF(ring, idx);
 291             s->iov[iovcnt].iov_len = ring->slot[i].len;
 292             iovcnt++;
 293
 294             ring->cur = ring->head = nm_ring_next(ring, i);
 295         } while (!nm_ring_empty(ring) && morefrag);
 296
 297         if (unlikely(nm_ring_empty(ring) && morefrag)) {
 298             RD(5, "[netmap_send] ran out of slots, with a pending"
 299                    "incomplete packet\n");
 300         }
 301
 302         iovsize = qemu_sendv_packet_async(&s->nc, s->iov, iovcnt,
 303                                             netmap_send_completed);
 304
 305         if (iovsize == 0) {
 306             /* The peer does not receive anymore. Packet is queued, stop
 307              * reading from the backend until netmap_send_completed()
 308              */
 309             netmap_read_poll(s, false);
 310             break;
 311         }
 312     }
 313 }
 314
 315 /* Flush and close. */
 316 static void netmap_cleanup(NetClientState *nc)
 317 {
 318     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 319
 320     qemu_purge_queued_packets(nc);
 321
 322     netmap_poll(nc, false);
 323     nm_close(s->nmd);
 324     s->nmd = NULL;
 325 }
 326
 327 /* Offloading manipulation support callbacks. */
 328 static int netmap_fd_set_vnet_hdr_len(NetmapState *s, int len)
 329 {
 330     struct nmreq req;
 331
 332     /* Issue a NETMAP_BDG_VNET_HDR command to change the virtio-net header
 333      * length for the netmap adapter associated to 's->ifname'.
 334      */
 335     memset(&req, 0, sizeof(req));
 336     pstrcpy(req.nr_name, sizeof(req.nr_name), s->ifname);
 337     req.nr_version = NETMAP_API;
 338     req.nr_cmd = NETMAP_BDG_VNET_HDR;
 339     req.nr_arg1 = len;
 340
 341     return ioctl(s->nmd->fd, NIOCREGIF, &req);
 342 }
 343
 344 static bool netmap_has_vnet_hdr_len(NetClientState *nc, int len)
 345 {
 346     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 347     int prev_len = s->vnet_hdr_len;
 348
 349     /* Check that we can set the new length. */
 350     if (netmap_fd_set_vnet_hdr_len(s, len)) {
 351         return false;
 352     }
 353
 354     /* Restore the previous length. */
 355     if (netmap_fd_set_vnet_hdr_len(s, prev_len)) {
 356         error_report("Failed to restore vnet-hdr length %d on %s: %s",
 357                      prev_len, s->ifname, strerror(errno));
 358         abort();
 359     }
 360
 361     return true;
 362 }
 363
 364 /* A netmap interface that supports virtio-net headers always
 365  * supports UFO, so we use this callback also for the has_ufo hook. */
 366 static bool netmap_has_vnet_hdr(NetClientState *nc)
 367 {
 368     return netmap_has_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 369 }
 370
 371 static void netmap_using_vnet_hdr(NetClientState *nc, bool enable)
 372 {
 373 }
 374
 375 static void netmap_set_vnet_hdr_len(NetClientState *nc, int len)
 376 {
 377     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 378     int err;
 379
 380     err = netmap_fd_set_vnet_hdr_len(s, len);
 381     if (err) {
 382         error_report("Unable to set vnet-hdr length %d on %s: %s",
 383                      len, s->ifname, strerror(errno));
 384     } else {
 385         /* Keep track of the current length. */
 386         s->vnet_hdr_len = len;
 387     }
 388 }
 389
 390 static void netmap_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
 391                                int ecn, int ufo)
 392 {
 393     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 394
 395     /* Setting a virtio-net header length greater than zero automatically
 396      * enables the offloadings. */
 397     if (!s->vnet_hdr_len) {
 398         netmap_set_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 399     }
 400 }
 401
 402 /* NetClientInfo methods */
 403 static NetClientInfo net_netmap_info = {
 404     .type = NET_CLIENT_OPTIONS_KIND_NETMAP,
 405     .size = sizeof(NetmapState),
 406     .receive = netmap_receive,
 407     .receive_iov = netmap_receive_iov,
 408     .poll = netmap_poll,
 409     .cleanup = netmap_cleanup,
 410     .has_ufo = netmap_has_vnet_hdr,
 411     .has_vnet_hdr = netmap_has_vnet_hdr,
 412     .has_vnet_hdr_len = netmap_has_vnet_hdr_len,
 413     .using_vnet_hdr = netmap_using_vnet_hdr,
 414     .set_offload = netmap_set_offload,
 415     .set_vnet_hdr_len = netmap_set_vnet_hdr_len,
 416 };
 417
 418 /* The exported init function
 419  *
 420  * ... -net netmap,ifname="..."
 421  */
 422 int net_init_netmap(const NetClientOptions *opts,
 423                     const char *name, NetClientState *peer, Error **errp)
 424 {
 425     const NetdevNetmapOptions *netmap_opts = opts->u.netmap.data;
 426     struct nm_desc *nmd;
 427     NetClientState *nc;
 428     Error *err = NULL;
 429     NetmapState *s;
 430
 431     nmd = netmap_open(netmap_opts, &err);
 432     if (err) {
 433         error_propagate(errp, err);
 434         return -1;
 435     }
 436     /* Create the object. */
 437     nc = qemu_new_net_client(&net_netmap_info, peer, "netmap", name);
 438     s = DO_UPCAST(NetmapState, nc, nc);
 439     s->nmd = nmd;
 440     s->tx = NETMAP_TXRING(nmd->nifp, 0);
 441     s->rx = NETMAP_RXRING(nmd->nifp, 0);
 442     s->vnet_hdr_len = 0;
 443     pstrcpy(s->ifname, sizeof(s->ifname), netmap_opts->ifname);
 444     netmap_read_poll(s, true); /* Initially only poll for reads. */
 445
 446     return 0;
 447 }
 448