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 "qemu/iov.h"
  40
  41 typedef struct NetmapState {
  42     NetClientState      nc;
  43     struct nm_desc      *nmd;
  44     char                ifname[IFNAMSIZ];
  45     struct netmap_ring  *tx;
  46     struct netmap_ring  *rx;
  47     bool                read_poll;
  48     bool                write_poll;
  49     struct iovec        iov[IOV_MAX];
  50     int                 vnet_hdr_len;  /* Current virtio-net header length. */
  51 } NetmapState;
  52
  53 #ifndef __FreeBSD__
  54 #define pkt_copy bcopy
  55 #else
  56 /* A fast copy routine only for multiples of 64 bytes, non overlapped. */
  57 static inline void
  58 pkt_copy(const void *_src, void *_dst, int l)
  59 {
  60     const uint64_t *src = _src;
  61     uint64_t *dst = _dst;
  62     if (unlikely(l >= 1024)) {
  63         bcopy(src, dst, l);
  64         return;
  65     }
  66     for (; l > 0; l -= 64) {
  67         *dst++ = *src++;
  68         *dst++ = *src++;
  69         *dst++ = *src++;
  70         *dst++ = *src++;
  71         *dst++ = *src++;
  72         *dst++ = *src++;
  73         *dst++ = *src++;
  74         *dst++ = *src++;
  75     }
  76 }
  77 #endif /* __FreeBSD__ */
  78
  79 /*
  80  * Open a netmap device. We assume there is only one queue
  81  * (which is the case for the VALE bridge).
  82  */
  83 static struct nm_desc *netmap_open(const NetdevNetmapOptions *nm_opts,
  84                                    Error **errp)
  85 {
  86     struct nm_desc *nmd;
  87     struct nmreq req;
  88
  89     memset(&req, 0, sizeof(req));
  90
  91     nmd = nm_open(nm_opts->ifname, &req, NETMAP_NO_TX_POLL,
  92                   NULL);
  93     if (nmd == NULL) {
  94         error_setg_errno(errp, errno, "Failed to nm_open() %s",
  95                          nm_opts->ifname);
  96         return NULL;
  97     }
  98
  99     return nmd;
 100 }
 101
 102 static void netmap_send(void *opaque);
 103 static void netmap_writable(void *opaque);
 104
 105 /* Set the event-loop handlers for the netmap backend. */
 106 static void netmap_update_fd_handler(NetmapState *s)
 107 {
 108     qemu_set_fd_handler(s->nmd->fd,
 109                         s->read_poll ? netmap_send : NULL,
 110                         s->write_poll ? netmap_writable : NULL,
 111                         s);
 112 }
 113
 114 /* Update the read handler. */
 115 static void netmap_read_poll(NetmapState *s, bool enable)
 116 {
 117     if (s->read_poll != enable) { /* Do nothing if not changed. */
 118         s->read_poll = enable;
 119         netmap_update_fd_handler(s);
 120     }
 121 }
 122
 123 /* Update the write handler. */
 124 static void netmap_write_poll(NetmapState *s, bool enable)
 125 {
 126     if (s->write_poll != enable) {
 127         s->write_poll = enable;
 128         netmap_update_fd_handler(s);
 129     }
 130 }
 131
 132 static void netmap_poll(NetClientState *nc, bool enable)
 133 {
 134     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 135
 136     if (s->read_poll != enable || s->write_poll != enable) {
 137         s->write_poll = enable;
 138         s->read_poll  = enable;
 139         netmap_update_fd_handler(s);
 140     }
 141 }
 142
 143 /*
 144  * The fd_write() callback, invoked if the fd is marked as
 145  * writable after a poll. Unregister the handler and flush any
 146  * buffered packets.
 147  */
 148 static void netmap_writable(void *opaque)
 149 {
 150     NetmapState *s = opaque;
 151
 152     netmap_write_poll(s, false);
 153     qemu_flush_queued_packets(&s->nc);
 154 }
 155
 156 static ssize_t netmap_receive(NetClientState *nc,
 157       const uint8_t *buf, size_t size)
 158 {
 159     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 160     struct netmap_ring *ring = s->tx;
 161     uint32_t i;
 162     uint32_t idx;
 163     uint8_t *dst;
 164
 165     if (unlikely(!ring)) {
 166         /* Drop. */
 167         return size;
 168     }
 169
 170     if (unlikely(size > ring->nr_buf_size)) {
 171         RD(5, "[netmap_receive] drop packet of size %d > %d\n",
 172                                     (int)size, ring->nr_buf_size);
 173         return size;
 174     }
 175
 176     if (nm_ring_empty(ring)) {
 177         /* No available slots in the netmap TX ring. */
 178         netmap_write_poll(s, true);
 179         return 0;
 180     }
 181
 182     i = ring->cur;
 183     idx = ring->slot[i].buf_idx;
 184     dst = (uint8_t *)NETMAP_BUF(ring, idx);
 185
 186     ring->slot[i].len = size;
 187     ring->slot[i].flags = 0;
 188     pkt_copy(buf, dst, size);
 189     ring->cur = ring->head = nm_ring_next(ring, i);
 190     ioctl(s->nmd->fd, NIOCTXSYNC, NULL);
 191
 192     return size;
 193 }
 194
 195 static ssize_t netmap_receive_iov(NetClientState *nc,
 196                     const struct iovec *iov, int iovcnt)
 197 {
 198     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 199     struct netmap_ring *ring = s->tx;
 200     uint32_t last;
 201     uint32_t idx;
 202     uint8_t *dst;
 203     int j;
 204     uint32_t i;
 205
 206     if (unlikely(!ring)) {
 207         /* Drop the packet. */
 208         return iov_size(iov, iovcnt);
 209     }
 210
 211     last = i = ring->cur;
 212
 213     if (nm_ring_space(ring) < iovcnt) {
 214         /* Not enough netmap slots. */
 215         netmap_write_poll(s, true);
 216         return 0;
 217     }
 218
 219     for (j = 0; j < iovcnt; j++) {
 220         int iov_frag_size = iov[j].iov_len;
 221         int offset = 0;
 222         int nm_frag_size;
 223
 224         /* Split each iovec fragment over more netmap slots, if
 225            necessary. */
 226         while (iov_frag_size) {
 227             nm_frag_size = MIN(iov_frag_size, ring->nr_buf_size);
 228
 229             if (unlikely(nm_ring_empty(ring))) {
 230                 /* We run out of netmap slots while splitting the
 231                    iovec fragments. */
 232                 netmap_write_poll(s, true);
 233                 return 0;
 234             }
 235
 236             idx = ring->slot[i].buf_idx;
 237             dst = (uint8_t *)NETMAP_BUF(ring, idx);
 238
 239             ring->slot[i].len = nm_frag_size;
 240             ring->slot[i].flags = NS_MOREFRAG;
 241             pkt_copy(iov[j].iov_base + offset, dst, nm_frag_size);
 242
 243             last = i;
 244             i = nm_ring_next(ring, i);
 245
 246             offset += nm_frag_size;
 247             iov_frag_size -= nm_frag_size;
 248         }
 249     }
 250     /* The last slot must not have NS_MOREFRAG set. */
 251     ring->slot[last].flags &= ~NS_MOREFRAG;
 252
 253     /* Now update ring->cur and ring->head. */
 254     ring->cur = ring->head = i;
 255
 256     ioctl(s->nmd->fd, NIOCTXSYNC, NULL);
 257
 258     return iov_size(iov, iovcnt);
 259 }
 260
 261 /* Complete a previous send (backend --> guest) and enable the
 262    fd_read callback. */
 263 static void netmap_send_completed(NetClientState *nc, ssize_t len)
 264 {
 265     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 266
 267     netmap_read_poll(s, true);
 268 }
 269
 270 static void netmap_send(void *opaque)
 271 {
 272     NetmapState *s = opaque;
 273     struct netmap_ring *ring = s->rx;
 274
 275     /* Keep sending while there are available packets into the netmap
 276        RX ring and the forwarding path towards the peer is open. */
 277     while (!nm_ring_empty(ring)) {
 278         uint32_t i;
 279         uint32_t idx;
 280         bool morefrag;
 281         int iovcnt = 0;
 282         int iovsize;
 283
 284         do {
 285             i = ring->cur;
 286             idx = ring->slot[i].buf_idx;
 287             morefrag = (ring->slot[i].flags & NS_MOREFRAG);
 288             s->iov[iovcnt].iov_base = (u_char *)NETMAP_BUF(ring, idx);
 289             s->iov[iovcnt].iov_len = ring->slot[i].len;
 290             iovcnt++;
 291
 292             ring->cur = ring->head = nm_ring_next(ring, i);
 293         } while (!nm_ring_empty(ring) && morefrag);
 294
 295         if (unlikely(nm_ring_empty(ring) && morefrag)) {
 296             RD(5, "[netmap_send] ran out of slots, with a pending"
 297                    "incomplete packet\n");
 298         }
 299
 300         iovsize = qemu_sendv_packet_async(&s->nc, s->iov, iovcnt,
 301                                             netmap_send_completed);
 302
 303         if (iovsize == 0) {
 304             /* The peer does not receive anymore. Packet is queued, stop
 305              * reading from the backend until netmap_send_completed()
 306              */
 307             netmap_read_poll(s, false);
 308             break;
 309         }
 310     }
 311 }
 312
 313 /* Flush and close. */
 314 static void netmap_cleanup(NetClientState *nc)
 315 {
 316     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 317
 318     qemu_purge_queued_packets(nc);
 319
 320     netmap_poll(nc, false);
 321     nm_close(s->nmd);
 322     s->nmd = NULL;
 323 }
 324
 325 /* Offloading manipulation support callbacks. */
 326 static bool netmap_has_ufo(NetClientState *nc)
 327 {
 328     return true;
 329 }
 330
 331 static bool netmap_has_vnet_hdr(NetClientState *nc)
 332 {
 333     return true;
 334 }
 335
 336 static bool netmap_has_vnet_hdr_len(NetClientState *nc, int len)
 337 {
 338     return len == 0 || len == sizeof(struct virtio_net_hdr) ||
 339                 len == sizeof(struct virtio_net_hdr_mrg_rxbuf);
 340 }
 341
 342 static void netmap_using_vnet_hdr(NetClientState *nc, bool enable)
 343 {
 344 }
 345
 346 static void netmap_set_vnet_hdr_len(NetClientState *nc, int len)
 347 {
 348     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 349     int err;
 350     struct nmreq req;
 351
 352     /* Issue a NETMAP_BDG_VNET_HDR command to change the virtio-net header
 353      * length for the netmap adapter associated to 's->ifname'.
 354      */
 355     memset(&req, 0, sizeof(req));
 356     pstrcpy(req.nr_name, sizeof(req.nr_name), s->ifname);
 357     req.nr_version = NETMAP_API;
 358     req.nr_cmd = NETMAP_BDG_VNET_HDR;
 359     req.nr_arg1 = len;
 360     err = ioctl(s->nmd->fd, NIOCREGIF, &req);
 361     if (err) {
 362         error_report("Unable to execute NETMAP_BDG_VNET_HDR on %s: %s",
 363                      s->ifname, strerror(errno));
 364     } else {
 365         /* Keep track of the current length. */
 366         s->vnet_hdr_len = len;
 367     }
 368 }
 369
 370 static void netmap_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
 371                                int ecn, int ufo)
 372 {
 373     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 374
 375     /* Setting a virtio-net header length greater than zero automatically
 376      * enables the offloadings.
 377      */
 378     if (!s->vnet_hdr_len) {
 379         netmap_set_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 380     }
 381 }
 382
 383 /* NetClientInfo methods */
 384 static NetClientInfo net_netmap_info = {
 385     .type = NET_CLIENT_OPTIONS_KIND_NETMAP,
 386     .size = sizeof(NetmapState),
 387     .receive = netmap_receive,
 388     .receive_iov = netmap_receive_iov,
 389     .poll = netmap_poll,
 390     .cleanup = netmap_cleanup,
 391     .has_ufo = netmap_has_ufo,
 392     .has_vnet_hdr = netmap_has_vnet_hdr,
 393     .has_vnet_hdr_len = netmap_has_vnet_hdr_len,
 394     .using_vnet_hdr = netmap_using_vnet_hdr,
 395     .set_offload = netmap_set_offload,
 396     .set_vnet_hdr_len = netmap_set_vnet_hdr_len,
 397 };
 398
 399 /* The exported init function
 400  *
 401  * ... -net netmap,ifname="..."
 402  */
 403 int net_init_netmap(const NetClientOptions *opts,
 404                     const char *name, NetClientState *peer, Error **errp)
 405 {
 406     const NetdevNetmapOptions *netmap_opts = opts->u.netmap;
 407     struct nm_desc *nmd;
 408     NetClientState *nc;
 409     Error *err = NULL;
 410     NetmapState *s;
 411
 412     nmd = netmap_open(netmap_opts, &err);
 413     if (err) {
 414         error_propagate(errp, err);
 415         return -1;
 416     }
 417     /* Create the object. */
 418     nc = qemu_new_net_client(&net_netmap_info, peer, "netmap", name);
 419     s = DO_UPCAST(NetmapState, nc, nc);
 420     s->nmd = nmd;
 421     s->tx = NETMAP_TXRING(nmd->nifp, 0);
 422     s->rx = NETMAP_RXRING(nmd->nifp, 0);
 423     s->vnet_hdr_len = 0;
 424     pstrcpy(s->ifname, sizeof(s->ifname), netmap_opts->ifname);
 425     netmap_read_poll(s, true); /* Initially only poll for reads. */
 426
 427     return 0;
 428 }
 429