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 #define NETMAP_WITH_LIBS
  30 #include <net/netmap.h>
  31 #include <net/netmap_user.h>
  32
  33 #include "net/net.h"
  34 #include "net/tap.h"
  35 #include "clients.h"
  36 #include "qemu/error-report.h"
  37 #include "qapi/error.h"
  38 #include "qemu/iov.h"
  39 #include "qemu/cutils.h"
  40 #include "qemu/main-loop.h"
  41
  42 typedef struct NetmapState {
  43     NetClientState      nc;
  44     struct nm_desc      *nmd;
  45     char                ifname[IFNAMSIZ];
  46     struct netmap_ring  *tx;
  47     struct netmap_ring  *rx;
  48     bool                read_poll;
  49     bool                write_poll;
  50     struct iovec        iov[IOV_MAX];
  51     int                 vnet_hdr_len;  /* Current virtio-net header length. */
  52 } NetmapState;
  53
  54 #ifndef __FreeBSD__
  55 #define pkt_copy bcopy
  56 #else
  57 /* A fast copy routine only for multiples of 64 bytes, non overlapped. */
  58 static inline void
  59 pkt_copy(const void *_src, void *_dst, int l)
  60 {
  61     const uint64_t *src = _src;
  62     uint64_t *dst = _dst;
  63     if (unlikely(l >= 1024)) {
  64         bcopy(src, dst, l);
  65         return;
  66     }
  67     for (; l > 0; l -= 64) {
  68         *dst++ = *src++;
  69         *dst++ = *src++;
  70         *dst++ = *src++;
  71         *dst++ = *src++;
  72         *dst++ = *src++;
  73         *dst++ = *src++;
  74         *dst++ = *src++;
  75         *dst++ = *src++;
  76     }
  77 }
  78 #endif /* __FreeBSD__ */
  79
  80 /*
  81  * Open a netmap device. We assume there is only one queue
  82  * (which is the case for the VALE bridge).
  83  */
  84 static struct nm_desc *netmap_open(const NetdevNetmapOptions *nm_opts,
  85                                    Error **errp)
  86 {
  87     struct nm_desc *nmd;
  88     struct nmreq req;
  89
  90     memset(&req, 0, sizeof(req));
  91
  92     nmd = nm_open(nm_opts->ifname, &req, NETMAP_NO_TX_POLL,
  93                   NULL);
  94     if (nmd == NULL) {
  95         error_setg_errno(errp, errno, "Failed to nm_open() %s",
  96                          nm_opts->ifname);
  97         return NULL;
  98     }
  99
 100     return nmd;
 101 }
 102
 103 static void netmap_send(void *opaque);
 104 static void netmap_writable(void *opaque);
 105
 106 /* Set the event-loop handlers for the netmap backend. */
 107 static void netmap_update_fd_handler(NetmapState *s)
 108 {
 109     qemu_set_fd_handler(s->nmd->fd,
 110                         s->read_poll ? netmap_send : NULL,
 111                         s->write_poll ? netmap_writable : NULL,
 112                         s);
 113 }
 114
 115 /* Update the read handler. */
 116 static void netmap_read_poll(NetmapState *s, bool enable)
 117 {
 118     if (s->read_poll != enable) { /* Do nothing if not changed. */
 119         s->read_poll = enable;
 120         netmap_update_fd_handler(s);
 121     }
 122 }
 123
 124 /* Update the write handler. */
 125 static void netmap_write_poll(NetmapState *s, bool enable)
 126 {
 127     if (s->write_poll != enable) {
 128         s->write_poll = enable;
 129         netmap_update_fd_handler(s);
 130     }
 131 }
 132
 133 static void netmap_poll(NetClientState *nc, bool enable)
 134 {
 135     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 136
 137     if (s->read_poll != enable || s->write_poll != enable) {
 138         s->write_poll = enable;
 139         s->read_poll  = enable;
 140         netmap_update_fd_handler(s);
 141     }
 142 }
 143
 144 /*
 145  * The fd_write() callback, invoked if the fd is marked as
 146  * writable after a poll. Unregister the handler and flush any
 147  * buffered packets.
 148  */
 149 static void netmap_writable(void *opaque)
 150 {
 151     NetmapState *s = opaque;
 152
 153     netmap_write_poll(s, false);
 154     qemu_flush_queued_packets(&s->nc);
 155 }
 156
 157 static ssize_t netmap_receive_iov(NetClientState *nc,
 158                     const struct iovec *iov, int iovcnt)
 159 {
 160     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 161     struct netmap_ring *ring = s->tx;
 162     unsigned int tail = ring->tail;
 163     ssize_t totlen = 0;
 164     uint32_t last;
 165     uint32_t idx;
 166     uint8_t *dst;
 167     int j;
 168     uint32_t i;
 169
 170     last = i = ring->head;
 171
 172     if (nm_ring_space(ring) < iovcnt) {
 173         /* Not enough netmap slots. Tell the kernel that we have seen the new
 174          * available slots (so that it notifies us again when it has more
 175          * ones), but without publishing any new slots to be processed
 176          * (e.g., we don't advance ring->head). */
 177         ring->cur = tail;
 178         netmap_write_poll(s, true);
 179         return 0;
 180     }
 181
 182     for (j = 0; j < iovcnt; j++) {
 183         int iov_frag_size = iov[j].iov_len;
 184         int offset = 0;
 185         int nm_frag_size;
 186
 187         totlen += iov_frag_size;
 188
 189         /* Split each iovec fragment over more netmap slots, if
 190            necessary. */
 191         while (iov_frag_size) {
 192             nm_frag_size = MIN(iov_frag_size, ring->nr_buf_size);
 193
 194             if (unlikely(i == tail)) {
 195                 /* We ran out of netmap slots while splitting the
 196                    iovec fragments. */
 197                 ring->cur = tail;
 198                 netmap_write_poll(s, true);
 199                 return 0;
 200             }
 201
 202             idx = ring->slot[i].buf_idx;
 203             dst = (uint8_t *)NETMAP_BUF(ring, idx);
 204
 205             ring->slot[i].len = nm_frag_size;
 206             ring->slot[i].flags = NS_MOREFRAG;
 207             pkt_copy(iov[j].iov_base + offset, dst, nm_frag_size);
 208
 209             last = i;
 210             i = nm_ring_next(ring, i);
 211
 212             offset += nm_frag_size;
 213             iov_frag_size -= nm_frag_size;
 214         }
 215     }
 216     /* The last slot must not have NS_MOREFRAG set. */
 217     ring->slot[last].flags &= ~NS_MOREFRAG;
 218
 219     /* Now update ring->head and ring->cur to publish the new slots and
 220      * the new wakeup point. */
 221     ring->head = ring->cur = i;
 222
 223     ioctl(s->nmd->fd, NIOCTXSYNC, NULL);
 224
 225     return totlen;
 226 }
 227
 228 static ssize_t netmap_receive(NetClientState *nc,
 229       const uint8_t *buf, size_t size)
 230 {
 231     struct iovec iov;
 232
 233     iov.iov_base = (void *)buf;
 234     iov.iov_len = size;
 235
 236     return netmap_receive_iov(nc, &iov, 1);
 237 }
 238
 239 /* Complete a previous send (backend --> guest) and enable the
 240    fd_read callback. */
 241 static void netmap_send_completed(NetClientState *nc, ssize_t len)
 242 {
 243     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 244
 245     netmap_read_poll(s, true);
 246 }
 247
 248 static void netmap_send(void *opaque)
 249 {
 250     NetmapState *s = opaque;
 251     struct netmap_ring *ring = s->rx;
 252     unsigned int tail = ring->tail;
 253
 254     /* Keep sending while there are available slots in the netmap
 255        RX ring and the forwarding path towards the peer is open. */
 256     while (ring->head != tail) {
 257         uint32_t i = ring->head;
 258         uint32_t idx;
 259         bool morefrag;
 260         int iovcnt = 0;
 261         int iovsize;
 262
 263         /* Get a (possibly multi-slot) packet. */
 264         do {
 265             idx = ring->slot[i].buf_idx;
 266             morefrag = (ring->slot[i].flags & NS_MOREFRAG);
 267             s->iov[iovcnt].iov_base = (void *)NETMAP_BUF(ring, idx);
 268             s->iov[iovcnt].iov_len = ring->slot[i].len;
 269             iovcnt++;
 270             i = nm_ring_next(ring, i);
 271         } while (i != tail && morefrag);
 272
 273         /* Advance ring->cur to tell the kernel that we have seen the slots. */
 274         ring->cur = i;
 275
 276         if (unlikely(morefrag)) {
 277             /* This is a truncated packet, so we can stop without releasing the
 278              * incomplete slots by updating ring->head. We will hopefully
 279              * re-read the complete packet the next time we are called. */
 280             break;
 281         }
 282
 283         iovsize = qemu_sendv_packet_async(&s->nc, s->iov, iovcnt,
 284                                             netmap_send_completed);
 285
 286         /* Release the slots to the kernel. */
 287         ring->head = i;
 288
 289         if (iovsize == 0) {
 290             /* The peer does not receive anymore. Packet is queued, stop
 291              * reading from the backend until netmap_send_completed(). */
 292             netmap_read_poll(s, false);
 293             break;
 294         }
 295     }
 296 }
 297
 298 /* Flush and close. */
 299 static void netmap_cleanup(NetClientState *nc)
 300 {
 301     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 302
 303     qemu_purge_queued_packets(nc);
 304
 305     netmap_poll(nc, false);
 306     nm_close(s->nmd);
 307     s->nmd = NULL;
 308 }
 309
 310 /* Offloading manipulation support callbacks. */
 311 static int netmap_fd_set_vnet_hdr_len(NetmapState *s, int len)
 312 {
 313     struct nmreq req;
 314
 315     /* Issue a NETMAP_BDG_VNET_HDR command to change the virtio-net header
 316      * length for the netmap adapter associated to 's->ifname'.
 317      */
 318     memset(&req, 0, sizeof(req));
 319     pstrcpy(req.nr_name, sizeof(req.nr_name), s->ifname);
 320     req.nr_version = NETMAP_API;
 321     req.nr_cmd = NETMAP_BDG_VNET_HDR;
 322     req.nr_arg1 = len;
 323
 324     return ioctl(s->nmd->fd, NIOCREGIF, &req);
 325 }
 326
 327 static bool netmap_has_vnet_hdr_len(NetClientState *nc, int len)
 328 {
 329     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 330     int prev_len = s->vnet_hdr_len;
 331
 332     /* Check that we can set the new length. */
 333     if (netmap_fd_set_vnet_hdr_len(s, len)) {
 334         return false;
 335     }
 336
 337     /* Restore the previous length. */
 338     if (netmap_fd_set_vnet_hdr_len(s, prev_len)) {
 339         error_report("Failed to restore vnet-hdr length %d on %s: %s",
 340                      prev_len, s->ifname, strerror(errno));
 341         abort();
 342     }
 343
 344     return true;
 345 }
 346
 347 /* A netmap interface that supports virtio-net headers always
 348  * supports UFO, so we use this callback also for the has_ufo hook. */
 349 static bool netmap_has_vnet_hdr(NetClientState *nc)
 350 {
 351     return netmap_has_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 352 }
 353
 354 static void netmap_set_vnet_hdr_len(NetClientState *nc, int len)
 355 {
 356     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 357     int err;
 358
 359     err = netmap_fd_set_vnet_hdr_len(s, len);
 360     if (err) {
 361         error_report("Unable to set vnet-hdr length %d on %s: %s",
 362                      len, s->ifname, strerror(errno));
 363     } else {
 364         /* Keep track of the current length. */
 365         s->vnet_hdr_len = len;
 366     }
 367 }
 368
 369 static void netmap_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
 370                                int ecn, int ufo, int uso4, int uso6)
 371 {
 372     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 373
 374     /* Setting a virtio-net header length greater than zero automatically
 375      * enables the offloadings. */
 376     if (!s->vnet_hdr_len) {
 377         netmap_set_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 378     }
 379 }
 380
 381 /* NetClientInfo methods */
 382 static NetClientInfo net_netmap_info = {
 383     .type = NET_CLIENT_DRIVER_NETMAP,
 384     .size = sizeof(NetmapState),
 385     .receive = netmap_receive,
 386     .receive_iov = netmap_receive_iov,
 387     .poll = netmap_poll,
 388     .cleanup = netmap_cleanup,
 389     .has_ufo = netmap_has_vnet_hdr,
 390     .has_vnet_hdr = netmap_has_vnet_hdr,
 391     .has_vnet_hdr_len = netmap_has_vnet_hdr_len,
 392     .set_offload = netmap_set_offload,
 393     .set_vnet_hdr_len = netmap_set_vnet_hdr_len,
 394 };
 395
 396 /* The exported init function
 397  *
 398  * ... -net netmap,ifname="..."
 399  */
 400 int net_init_netmap(const Netdev *netdev,
 401                     const char *name, NetClientState *peer, Error **errp)
 402 {
 403     const NetdevNetmapOptions *netmap_opts = &netdev->u.netmap;
 404     struct nm_desc *nmd;
 405     NetClientState *nc;
 406     Error *err = NULL;
 407     NetmapState *s;
 408
 409     nmd = netmap_open(netmap_opts, &err);
 410     if (err) {
 411         error_propagate(errp, err);
 412         return -1;
 413     }
 414     /* Create the object. */
 415     nc = qemu_new_net_client(&net_netmap_info, peer, "netmap", name);
 416     s = DO_UPCAST(NetmapState, nc, nc);
 417     s->nmd = nmd;
 418     s->tx = NETMAP_TXRING(nmd->nifp, 0);
 419     s->rx = NETMAP_RXRING(nmd->nifp, 0);
 420     s->vnet_hdr_len = 0;
 421     pstrcpy(s->ifname, sizeof(s->ifname), netmap_opts->ifname);
 422     netmap_read_poll(s, true); /* Initially only poll for reads. */
 423
 424     return 0;
 425 }
 426