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 <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 "sysemu/sysemu.h"
  37 #include "qemu/error-report.h"
  38 #include "qemu/iov.h"
  39
  40 /* Private netmap device info. */
  41 typedef struct NetmapPriv {
  42     int                 fd;
  43     size_t              memsize;
  44     void                *mem;
  45     struct netmap_if    *nifp;
  46     struct netmap_ring  *rx;
  47     struct netmap_ring  *tx;
  48     char                fdname[PATH_MAX];        /* Normally "/dev/netmap". */
  49     char                ifname[IFNAMSIZ];
  50 } NetmapPriv;
  51
  52 typedef struct NetmapState {
  53     NetClientState      nc;
  54     NetmapPriv          me;
  55     bool                read_poll;
  56     bool                write_poll;
  57     struct iovec        iov[IOV_MAX];
  58     int                 vnet_hdr_len;  /* Current virtio-net header length. */
  59 } NetmapState;
  60
  61 #define D(format, ...)                                          \
  62     do {                                                        \
  63         struct timeval __xxts;                                  \
  64         gettimeofday(&__xxts, NULL);                            \
  65         printf("%03d.%06d %s [%d] " format "\n",                \
  66                 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \
  67                 __func__, __LINE__, ##__VA_ARGS__);         \
  68     } while (0)
  69
  70 /* Rate limited version of "D", lps indicates how many per second */
  71 #define RD(lps, format, ...)                                    \
  72     do {                                                        \
  73         static int t0, __cnt;                                   \
  74         struct timeval __xxts;                                  \
  75         gettimeofday(&__xxts, NULL);                            \
  76         if (t0 != __xxts.tv_sec) {                              \
  77             t0 = __xxts.tv_sec;                                 \
  78             __cnt = 0;                                          \
  79         }                                                       \
  80         if (__cnt++ < lps) {                                    \
  81             D(format, ##__VA_ARGS__);                           \
  82         }                                                       \
  83     } while (0)
  84
  85
  86 #ifndef __FreeBSD__
  87 #define pkt_copy bcopy
  88 #else
  89 /* A fast copy routine only for multiples of 64 bytes, non overlapped. */
  90 static inline void
  91 pkt_copy(const void *_src, void *_dst, int l)
  92 {
  93     const uint64_t *src = _src;
  94     uint64_t *dst = _dst;
  95     if (unlikely(l >= 1024)) {
  96         bcopy(src, dst, l);
  97         return;
  98     }
  99     for (; l > 0; l -= 64) {
 100         *dst++ = *src++;
 101         *dst++ = *src++;
 102         *dst++ = *src++;
 103         *dst++ = *src++;
 104         *dst++ = *src++;
 105         *dst++ = *src++;
 106         *dst++ = *src++;
 107         *dst++ = *src++;
 108     }
 109 }
 110 #endif /* __FreeBSD__ */
 111
 112 /*
 113  * Open a netmap device. We assume there is only one queue
 114  * (which is the case for the VALE bridge).
 115  */
 116 static int netmap_open(NetmapPriv *me)
 117 {
 118     int fd;
 119     int err;
 120     size_t l;
 121     struct nmreq req;
 122
 123     me->fd = fd = open(me->fdname, O_RDWR);
 124     if (fd < 0) {
 125         error_report("Unable to open netmap device '%s' (%s)",
 126                         me->fdname, strerror(errno));
 127         return -1;
 128     }
 129     memset(&req, 0, sizeof(req));
 130     pstrcpy(req.nr_name, sizeof(req.nr_name), me->ifname);
 131     req.nr_ringid = NETMAP_NO_TX_POLL;
 132     req.nr_version = NETMAP_API;
 133     err = ioctl(fd, NIOCREGIF, &req);
 134     if (err) {
 135         error_report("Unable to register %s: %s", me->ifname, strerror(errno));
 136         goto error;
 137     }
 138     l = me->memsize = req.nr_memsize;
 139
 140     me->mem = mmap(0, l, PROT_WRITE | PROT_READ, MAP_SHARED, fd, 0);
 141     if (me->mem == MAP_FAILED) {
 142         error_report("Unable to mmap netmap shared memory: %s",
 143                         strerror(errno));
 144         me->mem = NULL;
 145         goto error;
 146     }
 147
 148     me->nifp = NETMAP_IF(me->mem, req.nr_offset);
 149     me->tx = NETMAP_TXRING(me->nifp, 0);
 150     me->rx = NETMAP_RXRING(me->nifp, 0);
 151     return 0;
 152
 153 error:
 154     close(me->fd);
 155     return -1;
 156 }
 157
 158 /* Tell the event-loop if the netmap backend can send packets
 159    to the frontend. */
 160 static int netmap_can_send(void *opaque)
 161 {
 162     NetmapState *s = opaque;
 163
 164     return qemu_can_send_packet(&s->nc);
 165 }
 166
 167 static void netmap_send(void *opaque);
 168 static void netmap_writable(void *opaque);
 169
 170 /* Set the event-loop handlers for the netmap backend. */
 171 static void netmap_update_fd_handler(NetmapState *s)
 172 {
 173     qemu_set_fd_handler2(s->me.fd,
 174                          s->read_poll  ? netmap_can_send : NULL,
 175                          s->read_poll  ? netmap_send     : NULL,
 176                          s->write_poll ? netmap_writable : NULL,
 177                          s);
 178 }
 179
 180 /* Update the read handler. */
 181 static void netmap_read_poll(NetmapState *s, bool enable)
 182 {
 183     if (s->read_poll != enable) { /* Do nothing if not changed. */
 184         s->read_poll = enable;
 185         netmap_update_fd_handler(s);
 186     }
 187 }
 188
 189 /* Update the write handler. */
 190 static void netmap_write_poll(NetmapState *s, bool enable)
 191 {
 192     if (s->write_poll != enable) {
 193         s->write_poll = enable;
 194         netmap_update_fd_handler(s);
 195     }
 196 }
 197
 198 static void netmap_poll(NetClientState *nc, bool enable)
 199 {
 200     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 201
 202     if (s->read_poll != enable || s->write_poll != enable) {
 203         s->read_poll = enable;
 204         s->read_poll = enable;
 205         netmap_update_fd_handler(s);
 206     }
 207 }
 208
 209 /*
 210  * The fd_write() callback, invoked if the fd is marked as
 211  * writable after a poll. Unregister the handler and flush any
 212  * buffered packets.
 213  */
 214 static void netmap_writable(void *opaque)
 215 {
 216     NetmapState *s = opaque;
 217
 218     netmap_write_poll(s, false);
 219     qemu_flush_queued_packets(&s->nc);
 220 }
 221
 222 static ssize_t netmap_receive(NetClientState *nc,
 223       const uint8_t *buf, size_t size)
 224 {
 225     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 226     struct netmap_ring *ring = s->me.tx;
 227     uint32_t i;
 228     uint32_t idx;
 229     uint8_t *dst;
 230
 231     if (unlikely(!ring)) {
 232         /* Drop. */
 233         return size;
 234     }
 235
 236     if (unlikely(size > ring->nr_buf_size)) {
 237         RD(5, "[netmap_receive] drop packet of size %d > %d\n",
 238                                     (int)size, ring->nr_buf_size);
 239         return size;
 240     }
 241
 242     if (ring->avail == 0) {
 243         /* No available slots in the netmap TX ring. */
 244         netmap_write_poll(s, true);
 245         return 0;
 246     }
 247
 248     i = ring->cur;
 249     idx = ring->slot[i].buf_idx;
 250     dst = (uint8_t *)NETMAP_BUF(ring, idx);
 251
 252     ring->slot[i].len = size;
 253     ring->slot[i].flags = 0;
 254     pkt_copy(buf, dst, size);
 255     ring->cur = NETMAP_RING_NEXT(ring, i);
 256     ring->avail--;
 257     ioctl(s->me.fd, NIOCTXSYNC, NULL);
 258
 259     return size;
 260 }
 261
 262 static ssize_t netmap_receive_iov(NetClientState *nc,
 263                     const struct iovec *iov, int iovcnt)
 264 {
 265     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 266     struct netmap_ring *ring = s->me.tx;
 267     uint32_t last;
 268     uint32_t idx;
 269     uint8_t *dst;
 270     int j;
 271     uint32_t i;
 272     uint32_t avail;
 273
 274     if (unlikely(!ring)) {
 275         /* Drop the packet. */
 276         return iov_size(iov, iovcnt);
 277     }
 278
 279     last = i = ring->cur;
 280     avail = ring->avail;
 281
 282     if (avail < iovcnt) {
 283         /* Not enough netmap slots. */
 284         netmap_write_poll(s, true);
 285         return 0;
 286     }
 287
 288     for (j = 0; j < iovcnt; j++) {
 289         int iov_frag_size = iov[j].iov_len;
 290         int offset = 0;
 291         int nm_frag_size;
 292
 293         /* Split each iovec fragment over more netmap slots, if
 294            necessary. */
 295         while (iov_frag_size) {
 296             nm_frag_size = MIN(iov_frag_size, ring->nr_buf_size);
 297
 298             if (unlikely(avail == 0)) {
 299                 /* We run out of netmap slots while splitting the
 300                    iovec fragments. */
 301                 netmap_write_poll(s, true);
 302                 return 0;
 303             }
 304
 305             idx = ring->slot[i].buf_idx;
 306             dst = (uint8_t *)NETMAP_BUF(ring, idx);
 307
 308             ring->slot[i].len = nm_frag_size;
 309             ring->slot[i].flags = NS_MOREFRAG;
 310             pkt_copy(iov[j].iov_base + offset, dst, nm_frag_size);
 311
 312             last = i;
 313             i = NETMAP_RING_NEXT(ring, i);
 314             avail--;
 315
 316             offset += nm_frag_size;
 317             iov_frag_size -= nm_frag_size;
 318         }
 319     }
 320     /* The last slot must not have NS_MOREFRAG set. */
 321     ring->slot[last].flags &= ~NS_MOREFRAG;
 322
 323     /* Now update ring->cur and ring->avail. */
 324     ring->cur = i;
 325     ring->avail = avail;
 326
 327     ioctl(s->me.fd, NIOCTXSYNC, NULL);
 328
 329     return iov_size(iov, iovcnt);
 330 }
 331
 332 /* Complete a previous send (backend --> guest) and enable the
 333    fd_read callback. */
 334 static void netmap_send_completed(NetClientState *nc, ssize_t len)
 335 {
 336     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 337
 338     netmap_read_poll(s, true);
 339 }
 340
 341 static void netmap_send(void *opaque)
 342 {
 343     NetmapState *s = opaque;
 344     struct netmap_ring *ring = s->me.rx;
 345
 346     /* Keep sending while there are available packets into the netmap
 347        RX ring and the forwarding path towards the peer is open. */
 348     while (ring->avail > 0 && qemu_can_send_packet(&s->nc)) {
 349         uint32_t i;
 350         uint32_t idx;
 351         bool morefrag;
 352         int iovcnt = 0;
 353         int iovsize;
 354
 355         do {
 356             i = ring->cur;
 357             idx = ring->slot[i].buf_idx;
 358             morefrag = (ring->slot[i].flags & NS_MOREFRAG);
 359             s->iov[iovcnt].iov_base = (u_char *)NETMAP_BUF(ring, idx);
 360             s->iov[iovcnt].iov_len = ring->slot[i].len;
 361             iovcnt++;
 362
 363             ring->cur = NETMAP_RING_NEXT(ring, i);
 364             ring->avail--;
 365         } while (ring->avail && morefrag);
 366
 367         if (unlikely(!ring->avail && morefrag)) {
 368             RD(5, "[netmap_send] ran out of slots, with a pending"
 369                    "incomplete packet\n");
 370         }
 371
 372         iovsize = qemu_sendv_packet_async(&s->nc, s->iov, iovcnt,
 373                                             netmap_send_completed);
 374
 375         if (iovsize == 0) {
 376             /* The peer does not receive anymore. Packet is queued, stop
 377              * reading from the backend until netmap_send_completed()
 378              */
 379             netmap_read_poll(s, false);
 380             break;
 381         }
 382     }
 383 }
 384
 385 /* Flush and close. */
 386 static void netmap_cleanup(NetClientState *nc)
 387 {
 388     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 389
 390     qemu_purge_queued_packets(nc);
 391
 392     netmap_poll(nc, false);
 393     munmap(s->me.mem, s->me.memsize);
 394     close(s->me.fd);
 395
 396     s->me.fd = -1;
 397 }
 398
 399 /* Offloading manipulation support callbacks. */
 400 static bool netmap_has_ufo(NetClientState *nc)
 401 {
 402     return true;
 403 }
 404
 405 static bool netmap_has_vnet_hdr(NetClientState *nc)
 406 {
 407     return true;
 408 }
 409
 410 static bool netmap_has_vnet_hdr_len(NetClientState *nc, int len)
 411 {
 412     return len == 0 || len == sizeof(struct virtio_net_hdr) ||
 413                 len == sizeof(struct virtio_net_hdr_mrg_rxbuf);
 414 }
 415
 416 static void netmap_using_vnet_hdr(NetClientState *nc, bool enable)
 417 {
 418 }
 419
 420 static void netmap_set_vnet_hdr_len(NetClientState *nc, int len)
 421 {
 422     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 423     int err;
 424     struct nmreq req;
 425
 426     /* Issue a NETMAP_BDG_VNET_HDR command to change the virtio-net header
 427      * length for the netmap adapter associated to 'me->ifname'.
 428      */
 429     memset(&req, 0, sizeof(req));
 430     pstrcpy(req.nr_name, sizeof(req.nr_name), s->me.ifname);
 431     req.nr_version = NETMAP_API;
 432     req.nr_cmd = NETMAP_BDG_VNET_HDR;
 433     req.nr_arg1 = len;
 434     err = ioctl(s->me.fd, NIOCREGIF, &req);
 435     if (err) {
 436         error_report("Unable to execute NETMAP_BDG_VNET_HDR on %s: %s",
 437                      s->me.ifname, strerror(errno));
 438     } else {
 439         /* Keep track of the current length. */
 440         s->vnet_hdr_len = len;
 441     }
 442 }
 443
 444 static void netmap_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
 445                                int ecn, int ufo)
 446 {
 447     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 448
 449     /* Setting a virtio-net header length greater than zero automatically
 450      * enables the offloadings.
 451      */
 452     if (!s->vnet_hdr_len) {
 453         netmap_set_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 454     }
 455 }
 456
 457 /* NetClientInfo methods */
 458 static NetClientInfo net_netmap_info = {
 459     .type = NET_CLIENT_OPTIONS_KIND_NETMAP,
 460     .size = sizeof(NetmapState),
 461     .receive = netmap_receive,
 462     .receive_iov = netmap_receive_iov,
 463     .poll = netmap_poll,
 464     .cleanup = netmap_cleanup,
 465     .has_ufo = netmap_has_ufo,
 466     .has_vnet_hdr = netmap_has_vnet_hdr,
 467     .has_vnet_hdr_len = netmap_has_vnet_hdr_len,
 468     .using_vnet_hdr = netmap_using_vnet_hdr,
 469     .set_offload = netmap_set_offload,
 470     .set_vnet_hdr_len = netmap_set_vnet_hdr_len,
 471 };
 472
 473 /* The exported init function
 474  *
 475  * ... -net netmap,ifname="..."
 476  */
 477 int net_init_netmap(const NetClientOptions *opts,
 478         const char *name, NetClientState *peer)
 479 {
 480     const NetdevNetmapOptions *netmap_opts = opts->netmap;
 481     NetClientState *nc;
 482     NetmapPriv me;
 483     NetmapState *s;
 484
 485     pstrcpy(me.fdname, sizeof(me.fdname),
 486         netmap_opts->has_devname ? netmap_opts->devname : "/dev/netmap");
 487     /* Set default name for the port if not supplied. */
 488     pstrcpy(me.ifname, sizeof(me.ifname), netmap_opts->ifname);
 489     if (netmap_open(&me)) {
 490         return -1;
 491     }
 492     /* Create the object. */
 493     nc = qemu_new_net_client(&net_netmap_info, peer, "netmap", name);
 494     s = DO_UPCAST(NetmapState, nc, nc);
 495     s->me = me;
 496     s->vnet_hdr_len = 0;
 497     netmap_read_poll(s, true); /* Initially only poll for reads. */
 498
 499     return 0;
 500 }
 501