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 <sys/ioctl.h>
  27 #include <net/if.h>
  28 #include <sys/mman.h>
  29 #include <stdint.h>
  30 #include <stdio.h>
  31 #define NETMAP_WITH_LIBS
  32 #include <net/netmap.h>
  33 #include <net/netmap_user.h>
  34
  35 #include "net/net.h"
  36 #include "net/tap.h"
  37 #include "clients.h"
  38 #include "sysemu/sysemu.h"
  39 #include "qemu/error-report.h"
  40 #include "qemu/iov.h"
  41
  42 /* Private netmap device info. */
  43 typedef struct NetmapPriv {
  44     int                 fd;
  45     size_t              memsize;
  46     void                *mem;
  47     struct netmap_if    *nifp;
  48     struct netmap_ring  *rx;
  49     struct netmap_ring  *tx;
  50     char                fdname[PATH_MAX];        /* Normally "/dev/netmap". */
  51     char                ifname[IFNAMSIZ];
  52 } NetmapPriv;
  53
  54 typedef struct NetmapState {
  55     NetClientState      nc;
  56     NetmapPriv          me;
  57     bool                read_poll;
  58     bool                write_poll;
  59     struct iovec        iov[IOV_MAX];
  60     int                 vnet_hdr_len;  /* Current virtio-net header length. */
  61 } NetmapState;
  62
  63 #ifndef __FreeBSD__
  64 #define pkt_copy bcopy
  65 #else
  66 /* A fast copy routine only for multiples of 64 bytes, non overlapped. */
  67 static inline void
  68 pkt_copy(const void *_src, void *_dst, int l)
  69 {
  70     const uint64_t *src = _src;
  71     uint64_t *dst = _dst;
  72     if (unlikely(l >= 1024)) {
  73         bcopy(src, dst, l);
  74         return;
  75     }
  76     for (; l > 0; l -= 64) {
  77         *dst++ = *src++;
  78         *dst++ = *src++;
  79         *dst++ = *src++;
  80         *dst++ = *src++;
  81         *dst++ = *src++;
  82         *dst++ = *src++;
  83         *dst++ = *src++;
  84         *dst++ = *src++;
  85     }
  86 }
  87 #endif /* __FreeBSD__ */
  88
  89 /*
  90  * Open a netmap device. We assume there is only one queue
  91  * (which is the case for the VALE bridge).
  92  */
  93 static int netmap_open(NetmapPriv *me)
  94 {
  95     int fd;
  96     int err;
  97     size_t l;
  98     struct nmreq req;
  99
 100     me->fd = fd = open(me->fdname, O_RDWR);
 101     if (fd < 0) {
 102         error_report("Unable to open netmap device '%s' (%s)",
 103                         me->fdname, strerror(errno));
 104         return -1;
 105     }
 106     memset(&req, 0, sizeof(req));
 107     pstrcpy(req.nr_name, sizeof(req.nr_name), me->ifname);
 108     req.nr_ringid = NETMAP_NO_TX_POLL;
 109     req.nr_version = NETMAP_API;
 110     err = ioctl(fd, NIOCREGIF, &req);
 111     if (err) {
 112         error_report("Unable to register %s: %s", me->ifname, strerror(errno));
 113         goto error;
 114     }
 115     l = me->memsize = req.nr_memsize;
 116
 117     me->mem = mmap(0, l, PROT_WRITE | PROT_READ, MAP_SHARED, fd, 0);
 118     if (me->mem == MAP_FAILED) {
 119         error_report("Unable to mmap netmap shared memory: %s",
 120                         strerror(errno));
 121         me->mem = NULL;
 122         goto error;
 123     }
 124
 125     me->nifp = NETMAP_IF(me->mem, req.nr_offset);
 126     me->tx = NETMAP_TXRING(me->nifp, 0);
 127     me->rx = NETMAP_RXRING(me->nifp, 0);
 128     return 0;
 129
 130 error:
 131     close(me->fd);
 132     return -1;
 133 }
 134
 135 static void netmap_send(void *opaque);
 136 static void netmap_writable(void *opaque);
 137
 138 /* Set the event-loop handlers for the netmap backend. */
 139 static void netmap_update_fd_handler(NetmapState *s)
 140 {
 141     qemu_set_fd_handler(s->me.fd,
 142                         s->read_poll ? netmap_send : NULL,
 143                         s->write_poll ? netmap_writable : NULL,
 144                         s);
 145 }
 146
 147 /* Update the read handler. */
 148 static void netmap_read_poll(NetmapState *s, bool enable)
 149 {
 150     if (s->read_poll != enable) { /* Do nothing if not changed. */
 151         s->read_poll = enable;
 152         netmap_update_fd_handler(s);
 153     }
 154 }
 155
 156 /* Update the write handler. */
 157 static void netmap_write_poll(NetmapState *s, bool enable)
 158 {
 159     if (s->write_poll != enable) {
 160         s->write_poll = enable;
 161         netmap_update_fd_handler(s);
 162     }
 163 }
 164
 165 static void netmap_poll(NetClientState *nc, bool enable)
 166 {
 167     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 168
 169     if (s->read_poll != enable || s->write_poll != enable) {
 170         s->write_poll = enable;
 171         s->read_poll  = enable;
 172         netmap_update_fd_handler(s);
 173     }
 174 }
 175
 176 /*
 177  * The fd_write() callback, invoked if the fd is marked as
 178  * writable after a poll. Unregister the handler and flush any
 179  * buffered packets.
 180  */
 181 static void netmap_writable(void *opaque)
 182 {
 183     NetmapState *s = opaque;
 184
 185     netmap_write_poll(s, false);
 186     qemu_flush_queued_packets(&s->nc);
 187 }
 188
 189 static ssize_t netmap_receive(NetClientState *nc,
 190       const uint8_t *buf, size_t size)
 191 {
 192     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 193     struct netmap_ring *ring = s->me.tx;
 194     uint32_t i;
 195     uint32_t idx;
 196     uint8_t *dst;
 197
 198     if (unlikely(!ring)) {
 199         /* Drop. */
 200         return size;
 201     }
 202
 203     if (unlikely(size > ring->nr_buf_size)) {
 204         RD(5, "[netmap_receive] drop packet of size %d > %d\n",
 205                                     (int)size, ring->nr_buf_size);
 206         return size;
 207     }
 208
 209     if (nm_ring_empty(ring)) {
 210         /* No available slots in the netmap TX ring. */
 211         netmap_write_poll(s, true);
 212         return 0;
 213     }
 214
 215     i = ring->cur;
 216     idx = ring->slot[i].buf_idx;
 217     dst = (uint8_t *)NETMAP_BUF(ring, idx);
 218
 219     ring->slot[i].len = size;
 220     ring->slot[i].flags = 0;
 221     pkt_copy(buf, dst, size);
 222     ring->cur = ring->head = nm_ring_next(ring, i);
 223     ioctl(s->me.fd, NIOCTXSYNC, NULL);
 224
 225     return size;
 226 }
 227
 228 static ssize_t netmap_receive_iov(NetClientState *nc,
 229                     const struct iovec *iov, int iovcnt)
 230 {
 231     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 232     struct netmap_ring *ring = s->me.tx;
 233     uint32_t last;
 234     uint32_t idx;
 235     uint8_t *dst;
 236     int j;
 237     uint32_t i;
 238
 239     if (unlikely(!ring)) {
 240         /* Drop the packet. */
 241         return iov_size(iov, iovcnt);
 242     }
 243
 244     last = i = ring->cur;
 245
 246     if (nm_ring_space(ring) < iovcnt) {
 247         /* Not enough netmap slots. */
 248         netmap_write_poll(s, true);
 249         return 0;
 250     }
 251
 252     for (j = 0; j < iovcnt; j++) {
 253         int iov_frag_size = iov[j].iov_len;
 254         int offset = 0;
 255         int nm_frag_size;
 256
 257         /* Split each iovec fragment over more netmap slots, if
 258            necessary. */
 259         while (iov_frag_size) {
 260             nm_frag_size = MIN(iov_frag_size, ring->nr_buf_size);
 261
 262             if (unlikely(nm_ring_empty(ring))) {
 263                 /* We run out of netmap slots while splitting the
 264                    iovec fragments. */
 265                 netmap_write_poll(s, true);
 266                 return 0;
 267             }
 268
 269             idx = ring->slot[i].buf_idx;
 270             dst = (uint8_t *)NETMAP_BUF(ring, idx);
 271
 272             ring->slot[i].len = nm_frag_size;
 273             ring->slot[i].flags = NS_MOREFRAG;
 274             pkt_copy(iov[j].iov_base + offset, dst, nm_frag_size);
 275
 276             last = i;
 277             i = nm_ring_next(ring, i);
 278
 279             offset += nm_frag_size;
 280             iov_frag_size -= nm_frag_size;
 281         }
 282     }
 283     /* The last slot must not have NS_MOREFRAG set. */
 284     ring->slot[last].flags &= ~NS_MOREFRAG;
 285
 286     /* Now update ring->cur and ring->head. */
 287     ring->cur = ring->head = i;
 288
 289     ioctl(s->me.fd, NIOCTXSYNC, NULL);
 290
 291     return iov_size(iov, iovcnt);
 292 }
 293
 294 /* Complete a previous send (backend --> guest) and enable the
 295    fd_read callback. */
 296 static void netmap_send_completed(NetClientState *nc, ssize_t len)
 297 {
 298     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 299
 300     netmap_read_poll(s, true);
 301 }
 302
 303 static void netmap_send(void *opaque)
 304 {
 305     NetmapState *s = opaque;
 306     struct netmap_ring *ring = s->me.rx;
 307
 308     /* Keep sending while there are available packets into the netmap
 309        RX ring and the forwarding path towards the peer is open. */
 310     while (!nm_ring_empty(ring)) {
 311         uint32_t i;
 312         uint32_t idx;
 313         bool morefrag;
 314         int iovcnt = 0;
 315         int iovsize;
 316
 317         do {
 318             i = ring->cur;
 319             idx = ring->slot[i].buf_idx;
 320             morefrag = (ring->slot[i].flags & NS_MOREFRAG);
 321             s->iov[iovcnt].iov_base = (u_char *)NETMAP_BUF(ring, idx);
 322             s->iov[iovcnt].iov_len = ring->slot[i].len;
 323             iovcnt++;
 324
 325             ring->cur = ring->head = nm_ring_next(ring, i);
 326         } while (!nm_ring_empty(ring) && morefrag);
 327
 328         if (unlikely(nm_ring_empty(ring) && morefrag)) {
 329             RD(5, "[netmap_send] ran out of slots, with a pending"
 330                    "incomplete packet\n");
 331         }
 332
 333         iovsize = qemu_sendv_packet_async(&s->nc, s->iov, iovcnt,
 334                                             netmap_send_completed);
 335
 336         if (iovsize == 0) {
 337             /* The peer does not receive anymore. Packet is queued, stop
 338              * reading from the backend until netmap_send_completed()
 339              */
 340             netmap_read_poll(s, false);
 341             break;
 342         }
 343     }
 344 }
 345
 346 /* Flush and close. */
 347 static void netmap_cleanup(NetClientState *nc)
 348 {
 349     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 350
 351     qemu_purge_queued_packets(nc);
 352
 353     netmap_poll(nc, false);
 354     munmap(s->me.mem, s->me.memsize);
 355     close(s->me.fd);
 356
 357     s->me.fd = -1;
 358 }
 359
 360 /* Offloading manipulation support callbacks. */
 361 static bool netmap_has_ufo(NetClientState *nc)
 362 {
 363     return true;
 364 }
 365
 366 static bool netmap_has_vnet_hdr(NetClientState *nc)
 367 {
 368     return true;
 369 }
 370
 371 static bool netmap_has_vnet_hdr_len(NetClientState *nc, int len)
 372 {
 373     return len == 0 || len == sizeof(struct virtio_net_hdr) ||
 374                 len == sizeof(struct virtio_net_hdr_mrg_rxbuf);
 375 }
 376
 377 static void netmap_using_vnet_hdr(NetClientState *nc, bool enable)
 378 {
 379 }
 380
 381 static void netmap_set_vnet_hdr_len(NetClientState *nc, int len)
 382 {
 383     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 384     int err;
 385     struct nmreq req;
 386
 387     /* Issue a NETMAP_BDG_VNET_HDR command to change the virtio-net header
 388      * length for the netmap adapter associated to 'me->ifname'.
 389      */
 390     memset(&req, 0, sizeof(req));
 391     pstrcpy(req.nr_name, sizeof(req.nr_name), s->me.ifname);
 392     req.nr_version = NETMAP_API;
 393     req.nr_cmd = NETMAP_BDG_VNET_HDR;
 394     req.nr_arg1 = len;
 395     err = ioctl(s->me.fd, NIOCREGIF, &req);
 396     if (err) {
 397         error_report("Unable to execute NETMAP_BDG_VNET_HDR on %s: %s",
 398                      s->me.ifname, strerror(errno));
 399     } else {
 400         /* Keep track of the current length. */
 401         s->vnet_hdr_len = len;
 402     }
 403 }
 404
 405 static void netmap_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
 406                                int ecn, int ufo)
 407 {
 408     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 409
 410     /* Setting a virtio-net header length greater than zero automatically
 411      * enables the offloadings.
 412      */
 413     if (!s->vnet_hdr_len) {
 414         netmap_set_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 415     }
 416 }
 417
 418 /* NetClientInfo methods */
 419 static NetClientInfo net_netmap_info = {
 420     .type = NET_CLIENT_OPTIONS_KIND_NETMAP,
 421     .size = sizeof(NetmapState),
 422     .receive = netmap_receive,
 423     .receive_iov = netmap_receive_iov,
 424     .poll = netmap_poll,
 425     .cleanup = netmap_cleanup,
 426     .has_ufo = netmap_has_ufo,
 427     .has_vnet_hdr = netmap_has_vnet_hdr,
 428     .has_vnet_hdr_len = netmap_has_vnet_hdr_len,
 429     .using_vnet_hdr = netmap_using_vnet_hdr,
 430     .set_offload = netmap_set_offload,
 431     .set_vnet_hdr_len = netmap_set_vnet_hdr_len,
 432 };
 433
 434 /* The exported init function
 435  *
 436  * ... -net netmap,ifname="..."
 437  */
 438 int net_init_netmap(const NetClientOptions *opts,
 439                     const char *name, NetClientState *peer, Error **errp)
 440 {
 441     /* FIXME error_setg(errp, ...) on failure */
 442     const NetdevNetmapOptions *netmap_opts = opts->netmap;
 443     NetClientState *nc;
 444     NetmapPriv me;
 445     NetmapState *s;
 446
 447     pstrcpy(me.fdname, sizeof(me.fdname),
 448         netmap_opts->has_devname ? netmap_opts->devname : "/dev/netmap");
 449     /* Set default name for the port if not supplied. */
 450     pstrcpy(me.ifname, sizeof(me.ifname), netmap_opts->ifname);
 451     if (netmap_open(&me)) {
 452         return -1;
 453     }
 454     /* Create the object. */
 455     nc = qemu_new_net_client(&net_netmap_info, peer, "netmap", name);
 456     s = DO_UPCAST(NetmapState, nc, nc);
 457     s->me = me;
 458     s->vnet_hdr_len = 0;
 459     netmap_read_poll(s, true); /* Initially only poll for reads. */
 460
 461     return 0;
 462 }
 463