sys/dev/virtual/net/if_vke.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Sepherosa Ziehau <sepherosa@gmail.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/dev/virtual/net/if_vke.c,v 1.5 2007/02/10 11:32:39 sephe Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/endian.h>
  39 #include <sys/kernel.h>
  40 #include <sys/malloc.h>
  41 #include <sys/proc.h>
  42 #include <sys/serialize.h>
  43 #include <sys/socket.h>
  44 #include <sys/sockio.h>
  45 #include <sys/sysctl.h>
  46
  47 #include <machine/md_var.h>
  48
  49 #include <net/ethernet.h>
  50 #include <net/if.h>
  51 #include <net/bpf.h>
  52 #include <net/if_arp.h>
  53 #include <net/ifq_var.h>
  54
  55 #include <netinet/in_var.h>
  56
  57 #include <sys/stat.h>
  58 #include <sys/ioccom.h>
  59 #include <net/tap/if_tap.h>
  60 #include <errno.h>
  61 #include <stdio.h>
  62 #include <string.h>
  63 #include <unistd.h>
  64
  65 #define VKE_DEVNAME             "vke"
  66
  67 struct vke_softc {
  68         struct arpcom           arpcom;
  69         int                     sc_fd;
  70         int                     sc_unit;
  71
  72         struct kqueue_info      *sc_kqueue;
  73
  74         void                    *sc_txbuf;
  75         struct mbuf             *sc_rx_mbuf;
  76
  77         struct sysctl_ctx_list  sc_sysctl_ctx;
  78         struct sysctl_oid       *sc_sysctl_tree;
  79
  80         int                     sc_tap_unit;    /* unit of backend tap(4) */
  81         in_addr_t               sc_addr;        /* address */
  82         in_addr_t               sc_mask;        /* netmask */
  83 };
  84
  85 static void     vke_start(struct ifnet *);
  86 static void     vke_init(void *);
  87 static int      vke_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
  88
  89 static int      vke_attach(const struct vknetif_info *, int);
  90 static void     vke_intr(void *, struct intrframe *);
  91 static int      vke_stop(struct vke_softc *);
  92 static int      vke_rxeof(struct vke_softc *);
  93 static int      vke_init_addr(struct ifnet *, in_addr_t, in_addr_t);
  94
  95 static void
  96 vke_sysinit(void *arg __unused)
  97 {
  98         int i, unit;
  99
 100         KASSERT(NetifNum <= VKNETIF_MAX, ("too many netifs: %d\n", NetifNum));
 101
 102         unit = 0;
 103         for (i = 0; i < NetifNum; ++i) {
 104                 if (vke_attach(&NetifInfo[i], unit) == 0)
 105                         ++unit;
 106         }
 107 }
 108 SYSINIT(vke, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, vke_sysinit, NULL);
 109
 110 static void
 111 vke_init(void *xsc)
 112 {
 113         struct vke_softc *sc = xsc;
 114         struct ifnet *ifp = &sc->arpcom.ac_if;
 115
 116         ASSERT_SERIALIZED(ifp->if_serializer);
 117
 118         vke_stop(sc);
 119
 120         KKASSERT(sc->sc_kqueue == NULL);
 121         sc->sc_kqueue = kqueue_add(sc->sc_fd, vke_intr, sc);
 122         KKASSERT(sc->sc_kqueue != NULL);
 123
 124         ifp->if_flags |= IFF_RUNNING;
 125         ifp->if_flags &= ~IFF_OACTIVE;
 126
 127         if (sc->sc_addr != 0) {
 128                 in_addr_t addr, mask;
 129
 130                 addr = sc->sc_addr;
 131                 mask = sc->sc_mask;
 132
 133                 /*
 134                  * Make sure vkernel assigned
 135                  * address will not be added
 136                  * again.
 137                  */
 138                 sc->sc_addr = 0;
 139                 sc->sc_mask = 0;
 140
 141                 vke_init_addr(ifp, addr, mask);
 142         }
 143
 144         ifp->if_start(ifp);
 145 }
 146
 147 static void
 148 vke_start(struct ifnet *ifp)
 149 {
 150         struct vke_softc *sc = ifp->if_softc;
 151         struct mbuf *m;
 152
 153         ASSERT_SERIALIZED(ifp->if_serializer);
 154
 155         if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
 156                 return;
 157
 158         while ((m = ifq_dequeue(&ifp->if_snd, NULL)) != NULL) {
 159                 m_copydata(m, 0, m->m_pkthdr.len, sc->sc_txbuf);
 160                 BPF_MTAP(ifp, m);
 161                 write(sc->sc_fd, sc->sc_txbuf, m->m_pkthdr.len);
 162                 m_freem(m);
 163                 ifp->if_opackets++;
 164         }
 165 }
 166
 167 static int
 168 vke_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
 169 {
 170         struct vke_softc *sc = ifp->if_softc;
 171         int error = 0;
 172
 173         ASSERT_SERIALIZED(ifp->if_serializer);
 174
 175         switch (cmd) {
 176         case SIOCSIFFLAGS:
 177                 if (ifp->if_flags & IFF_UP) {
 178                         if ((ifp->if_flags & IFF_RUNNING) == 0)
 179                                 vke_init(sc);
 180                 } else {
 181                         if (ifp->if_flags & IFF_RUNNING)
 182                                 vke_stop(sc);
 183                 }
 184                 break;
 185         case SIOCGIFMEDIA:
 186         case SIOCSIFMEDIA:
 187                 error = EOPNOTSUPP;
 188                 /* TODO */
 189                 break;
 190         case SIOCGIFSTATUS: {
 191                 struct ifstat *ifs = (struct ifstat *)data;
 192                 int len;
 193
 194                 len = strlen(ifs->ascii);
 195                 if (len < sizeof(ifs->ascii)) {
 196                         ksnprintf(ifs->ascii + len, sizeof(ifs->ascii) - len,
 197                                   "\tBacked by tap%d\n", sc->sc_tap_unit);
 198                 }
 199                 break;
 200         }
 201         case SIOCSIFADDR:
 202                 if (((struct ifaddr *)data)->ifa_addr->sa_family == AF_INET) {
 203                         /*
 204                          * If we are explicitly requested to change address,
 205                          * we should invalidate address/netmask passed in
 206                          * from vkernel command line.
 207                          */
 208                         sc->sc_addr = 0;
 209                         sc->sc_mask = 0;
 210                 }
 211                 /* FALL THROUGH */
 212         default:
 213                 error = ether_ioctl(ifp, cmd, data);
 214                 break;
 215         }
 216         return error;
 217 }
 218
 219 static int
 220 vke_stop(struct vke_softc *sc)
 221 {
 222         struct ifnet *ifp = &sc->arpcom.ac_if;
 223
 224         ASSERT_SERIALIZED(ifp->if_serializer);
 225
 226         ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
 227         if (sc->sc_kqueue) {
 228                 kqueue_del(sc->sc_kqueue);
 229                 sc->sc_kqueue = NULL;
 230         }
 231
 232         if (sc->sc_rx_mbuf != NULL) {
 233                 m_freem(sc->sc_rx_mbuf);
 234                 sc->sc_rx_mbuf = NULL;
 235         }
 236         return 0;
 237 }
 238
 239 static void
 240 vke_intr(void *xsc, struct intrframe *frame __unused)
 241 {
 242         struct vke_softc *sc = xsc;
 243         struct ifnet *ifp = &sc->arpcom.ac_if;
 244
 245         lwkt_serialize_enter(ifp->if_serializer);
 246
 247         if ((ifp->if_flags & IFF_RUNNING) == 0)
 248                 goto back;
 249
 250         vke_rxeof(sc);
 251
 252         ifp->if_start(ifp);
 253
 254 back:
 255         lwkt_serialize_exit(ifp->if_serializer);
 256 }
 257
 258 static int
 259 vke_rxeof(struct vke_softc *sc)
 260 {
 261         struct ifnet *ifp = &sc->arpcom.ac_if;
 262         struct mbuf *m;
 263
 264         ASSERT_SERIALIZED(ifp->if_serializer);
 265
 266         for (;;) {
 267                 int n;
 268
 269                 if (sc->sc_rx_mbuf != NULL) {
 270                         m = sc->sc_rx_mbuf;
 271                         sc->sc_rx_mbuf = NULL;
 272                 } else {
 273                         m = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
 274                         if (m == NULL)
 275                                 return ENOBUFS;
 276                 }
 277
 278                 n = read(sc->sc_fd, mtod(m, void *), MCLBYTES);
 279                 if (n < 0) {
 280                         sc->sc_rx_mbuf = m;     /* We can use it next time */
 281
 282                         /* TODO: handle fatal error */
 283                         break;
 284                 }
 285                 ifp->if_ipackets++;
 286                 m->m_pkthdr.rcvif = ifp;
 287                 m->m_pkthdr.len = m->m_len = n;
 288                 ifp->if_input(ifp, m);
 289         }
 290         return 0;
 291 }
 292
 293 static int
 294 vke_attach(const struct vknetif_info *info, int unit)
 295 {
 296         struct vke_softc *sc;
 297         struct ifnet *ifp;
 298         struct tapinfo tapinfo;
 299         uint8_t enaddr[ETHER_ADDR_LEN];
 300         int fd;
 301
 302         KKASSERT(info->tap_fd >= 0);
 303         fd = info->tap_fd;
 304
 305         if (ioctl(fd, TAPGIFINFO, &tapinfo) < 0) {
 306                 kprintf(VKE_DEVNAME "%d: ioctl(TAPGIFINFO) failed: %s\n",
 307                         unit, strerror(errno));
 308                 return ENXIO;
 309         }
 310
 311         if (ioctl(fd, SIOCGIFADDR, enaddr) < 0) {
 312                 kprintf(VKE_DEVNAME "%d: ioctl(SIOCGIFADDR) failed: %s\n",
 313                         unit, strerror(errno));
 314                 return ENXIO;
 315         }
 316         enaddr[1] += 1;
 317
 318         sc = kmalloc(sizeof(*sc), M_DEVBUF, M_WAITOK | M_ZERO);
 319
 320         sc->sc_txbuf = kmalloc(MCLBYTES, M_DEVBUF, M_WAITOK);
 321         sc->sc_fd = fd;
 322         sc->sc_unit = unit;
 323         sc->sc_tap_unit = info->tap_unit;
 324         sc->sc_addr = info->netif_addr;
 325         sc->sc_mask = info->netif_mask;
 326
 327         ifp = &sc->arpcom.ac_if;
 328         if_initname(ifp, VKE_DEVNAME, sc->sc_unit);
 329
 330         /* NB: after if_initname() */
 331         sysctl_ctx_init(&sc->sc_sysctl_ctx);
 332         sc->sc_sysctl_tree = SYSCTL_ADD_NODE(&sc->sc_sysctl_ctx,
 333                                              SYSCTL_STATIC_CHILDREN(_hw),
 334                                              OID_AUTO, ifp->if_xname,
 335                                              CTLFLAG_RD, 0, "");
 336         if (sc->sc_sysctl_tree == NULL) {
 337                 kprintf(VKE_DEVNAME "%d: can't add sysctl node\n", unit);
 338         } else {
 339                 SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
 340                                SYSCTL_CHILDREN(sc->sc_sysctl_tree),
 341                                OID_AUTO, "tap_unit",
 342                                CTLFLAG_RW, &sc->sc_tap_unit, 0,
 343                                "Backend tap(4) unit");
 344         }
 345
 346         ifp->if_softc = sc;
 347         ifp->if_ioctl = vke_ioctl;
 348         ifp->if_start = vke_start;
 349         ifp->if_init = vke_init;
 350         ifp->if_mtu = tapinfo.mtu;
 351         ifp->if_baudrate = tapinfo.baudrate;
 352         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
 353         ifq_set_maxlen(&ifp->if_snd, IFQ_MAXLEN);
 354         ifq_set_ready(&ifp->if_snd);
 355
 356         /* TODO: if_media */
 357
 358         ether_ifattach(ifp, enaddr, NULL);
 359
 360         if (bootverbose && sc->sc_addr != 0) {
 361                 if_printf(ifp, "pre-configured "
 362                           "address 0x%08x, netmask 0x%08x\n",
 363                           ntohl(sc->sc_addr), ntohl(sc->sc_mask));
 364         }
 365
 366         return 0;
 367 }
 368
 369 static int
 370 vke_init_addr(struct ifnet *ifp, in_addr_t addr, in_addr_t mask)
 371 {
 372         struct ifaliasreq ifra;
 373         struct sockaddr_in *sin;
 374         int ret;
 375
 376         ASSERT_SERIALIZED(ifp->if_serializer);
 377
 378         if (bootverbose) {
 379                 if_printf(ifp, "add pre-configured "
 380                           "address 0x%08x, netmask 0x%08x\n",
 381                           ntohl(addr), ntohl(mask));
 382         }
 383
 384         bzero(&ifra, sizeof(ifra));
 385
 386         /* NB: no need to set ifaliasreq.ifra_name */
 387
 388         sin = (struct sockaddr_in *)&ifra.ifra_addr;
 389         sin->sin_family = AF_INET;
 390         sin->sin_len = sizeof(*sin);
 391         sin->sin_addr.s_addr = addr;
 392
 393         if (mask != 0) {
 394                 sin = (struct sockaddr_in *)&ifra.ifra_mask;
 395                 sin->sin_len = sizeof(*sin);
 396                 sin->sin_addr.s_addr = mask;
 397         }
 398
 399         /*
 400          * Temporarily release serializer, in_control() will hold
 401          * it again before calling ifnet.if_ioctl().
 402          */
 403         lwkt_serialize_exit(ifp->if_serializer);
 404         ret = in_control(NULL, SIOCAIFADDR, (caddr_t)&ifra, ifp, NULL);
 405         lwkt_serialize_enter(ifp->if_serializer);
 406
 407         return ret;
 408 }