kernel/net/ip/ip2mac.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 /*
  28  * Functions to implement IP address -> link layer address (PSARC 2006/482)
  29  */
  30 #include <inet/ip2mac.h>
  31 #include <inet/ip2mac_impl.h>
  32 #include <sys/zone.h>
  33 #include <inet/ip_ndp.h>
  34 #include <inet/ip_if.h>
  35 #include <inet/ip6.h>
  36
  37 /*
  38  * dispatch pending callbacks.
  39  */
  40 void
  41 ncec_cb_dispatch(ncec_t *ncec)
  42 {
  43         ncec_cb_t *ncec_cb;
  44         ip2mac_t ip2m;
  45
  46         mutex_enter(&ncec->ncec_lock);
  47         if (list_is_empty(&ncec->ncec_cb)) {
  48                 mutex_exit(&ncec->ncec_lock);
  49                 return;
  50         }
  51         ncec_ip2mac_response(&ip2m, ncec);
  52         ncec_cb_refhold_locked(ncec);
  53         /*
  54          * IP does not hold internal locks like nce_lock across calls to
  55          * other subsystems for fear of recursive lock entry and lock
  56          * hierarchy violation. The caller may be holding locks across
  57          * the call to IP. (It would be ideal if no subsystem holds locks
  58          * across calls into another subsystem, especially if calls can
  59          * happen in either direction).
  60          */
  61         ncec_cb = list_head(&ncec->ncec_cb);
  62         for (; ncec_cb != NULL; ncec_cb = list_next(&ncec->ncec_cb, ncec_cb)) {
  63                 if (ncec_cb->ncec_cb_flags & NCE_CB_DISPATCHED)
  64                         continue;
  65                 ncec_cb->ncec_cb_flags |= NCE_CB_DISPATCHED;
  66                 mutex_exit(&ncec->ncec_lock);
  67                 (*ncec_cb->ncec_cb_func)(&ip2m, ncec_cb->ncec_cb_arg);
  68                 mutex_enter(&ncec->ncec_lock);
  69         }
  70         ncec_cb_refrele(ncec);
  71         mutex_exit(&ncec->ncec_lock);
  72 }
  73
  74 /*
  75  * fill up the ip2m response fields with inforamation from the nce.
  76  */
  77 void
  78 ncec_ip2mac_response(ip2mac_t *ip2m, ncec_t *ncec)
  79 {
  80         boolean_t isv6 = (ncec->ncec_ipversion == IPV6_VERSION);
  81         sin_t   *sin;
  82         sin6_t  *sin6;
  83         struct sockaddr_dl *sdl;
  84
  85         ASSERT(MUTEX_HELD(&ncec->ncec_lock));
  86         bzero(ip2m, sizeof (*ip2m));
  87         if (NCE_ISREACHABLE(ncec) && !NCE_ISCONDEMNED(ncec))
  88                 ip2m->ip2mac_err = 0;
  89         else
  90                 ip2m->ip2mac_err = ESRCH;
  91         if (isv6) {
  92                 sin6 = (sin6_t *)&ip2m->ip2mac_pa;
  93                 sin6->sin6_family = AF_INET6;
  94                 sin6->sin6_addr = ncec->ncec_addr;
  95         } else {
  96                 sin = (sin_t *)&ip2m->ip2mac_pa;
  97                 sin->sin_family = AF_INET;
  98                 IN6_V4MAPPED_TO_INADDR(&ncec->ncec_addr, &sin->sin_addr);
  99         }
 100         if (ip2m->ip2mac_err == 0) {
 101                 sdl = &ip2m->ip2mac_ha;
 102                 sdl->sdl_family = AF_LINK;
 103                 sdl->sdl_type = ncec->ncec_ill->ill_type;
 104                 /*
 105                  * should we put ncec_ill->ill_name in there? why?
 106                  * likewise for the sdl_index
 107                  */
 108                 sdl->sdl_nlen = 0;
 109                 sdl->sdl_alen = ncec->ncec_ill->ill_phys_addr_length;
 110                 if (ncec->ncec_lladdr != NULL)
 111                         bcopy(ncec->ncec_lladdr, LLADDR(sdl), sdl->sdl_alen);
 112         }
 113 }
 114
 115 void
 116 ncec_cb_refhold_locked(ncec_t *ncec)
 117 {
 118         ASSERT(MUTEX_HELD(&ncec->ncec_lock));
 119         ncec->ncec_cb_walker_cnt++;
 120 }
 121
 122 void
 123 ncec_cb_refrele(ncec_t *ncec)
 124 {
 125         ncec_cb_t *ncec_cb, *ncec_cb_next = NULL;
 126
 127         ASSERT(MUTEX_HELD(&ncec->ncec_lock));
 128         if (--ncec->ncec_cb_walker_cnt == 0) {
 129                 for (ncec_cb = list_head(&ncec->ncec_cb); ncec_cb != NULL;
 130                     ncec_cb = ncec_cb_next) {
 131
 132                         ncec_cb_next = list_next(&ncec->ncec_cb, ncec_cb);
 133                         if ((ncec_cb->ncec_cb_flags & NCE_CB_DISPATCHED) == 0)
 134                                 continue;
 135                         list_remove(&ncec->ncec_cb, ncec_cb);
 136                         kmem_free(ncec_cb, sizeof (*ncec_cb));
 137                 }
 138         }
 139 }
 140
 141 /*
 142  * add a callback to the nce, so that the callback can be invoked
 143  * after address resolution succeeds/fails.
 144  */
 145 static ip2mac_id_t
 146 ncec_add_cb(ncec_t *ncec, ip2mac_callback_t *cb, void *cbarg)
 147 {
 148         ncec_cb_t       *nce_cb;
 149         ip2mac_id_t     ip2mid = NULL;
 150
 151         ASSERT(MUTEX_HELD(&ncec->ncec_lock));
 152         if ((nce_cb = kmem_zalloc(sizeof (*nce_cb), KM_NOSLEEP)) == NULL)
 153                 return (ip2mid);
 154         nce_cb->ncec_cb_func = cb;
 155         nce_cb->ncec_cb_arg = cbarg;
 156         /*
 157          * We identify the ncec_cb_t during cancellation by the address
 158          * of the nce_cb_t itself, and, as a short-cut for eliminating
 159          * clear mismatches, only look in the callback list of ncec's
 160          * whose address is equal to the nce_cb_id.
 161          */
 162         nce_cb->ncec_cb_id = ncec; /* no refs! just an address */
 163         list_insert_tail(&ncec->ncec_cb, nce_cb);
 164         ip2mid = ncec;  /* this is the id to be used in ip2mac_cancel */
 165
 166         return (nce_cb);
 167 }
 168
 169 /*
 170  * Resolve an IP address to a link-layer address using the data-structures
 171  * defined in PSARC 2006/482. If the current link-layer address for the
 172  * IP address is not known, the state-machine for resolving the resolution
 173  * will be triggered, and the callback function (*cb) will be invoked after
 174  * the resolution completes.
 175  */
 176 ip2mac_id_t
 177 ip2mac(uint_t op, ip2mac_t *ip2m, ip2mac_callback_t *cb, void *cbarg,
 178     zoneid_t zoneid)
 179 {
 180         ncec_t          *ncec;
 181         nce_t           *nce = NULL;
 182         boolean_t       isv6;
 183         ill_t           *ill;
 184         netstack_t      *ns;
 185         ip_stack_t      *ipst;
 186         ip2mac_id_t     ip2mid = NULL;
 187         sin_t           *sin;
 188         sin6_t          *sin6;
 189         int             err;
 190         uint64_t        delta;
 191         boolean_t       need_resolve = B_FALSE;
 192
 193         isv6 = (ip2m->ip2mac_pa.ss_family == AF_INET6);
 194
 195         ns = netstack_find_by_zoneid(zoneid);
 196         if (ns == NULL) {
 197                 ip2m->ip2mac_err = EINVAL;
 198                 return (NULL);
 199         }
 200         /*
 201          * For exclusive stacks we reset the zoneid to zero
 202          * since IP uses the global zoneid in the exclusive stacks.
 203          */
 204         if (ns->netstack_stackid != GLOBAL_NETSTACKID)
 205                 zoneid = GLOBAL_ZONEID;
 206         ipst = ns->netstack_ip;
 207         /*
 208          * find the ill from the ip2m->ip2mac_ifindex
 209          */
 210         ill = ill_lookup_on_ifindex(ip2m->ip2mac_ifindex, isv6, ipst);
 211         if (ill == NULL) {
 212                 ip2m->ip2mac_err = ENXIO;
 213                 netstack_rele(ns);
 214                 return (NULL);
 215         }
 216         if (isv6) {
 217                 sin6 = (sin6_t *)&ip2m->ip2mac_pa;
 218                 if (op == IP2MAC_LOOKUP) {
 219                         nce = nce_lookup_v6(ill, &sin6->sin6_addr);
 220                 } else {
 221                         err = nce_lookup_then_add_v6(ill, NULL,
 222                             ill->ill_phys_addr_length,
 223                             &sin6->sin6_addr, 0, ND_UNCHANGED, &nce);
 224                 }
 225         } else  {
 226                 sin = (sin_t *)&ip2m->ip2mac_pa;
 227                 if (op == IP2MAC_LOOKUP) {
 228                         nce = nce_lookup_v4(ill, &sin->sin_addr.s_addr);
 229                 } else {
 230                         err = nce_lookup_then_add_v4(ill, NULL,
 231                             ill->ill_phys_addr_length,
 232                             &sin->sin_addr.s_addr, 0, ND_UNCHANGED, &nce);
 233                 }
 234         }
 235         if (op == IP2MAC_LOOKUP) {
 236                 if (nce == NULL) {
 237                         ip2m->ip2mac_err = ESRCH;
 238                         goto done;
 239                 }
 240                 ncec = nce->nce_common;
 241                 delta = TICK_TO_MSEC(ddi_get_lbolt64()) - ncec->ncec_last;
 242                 mutex_enter(&ncec->ncec_lock);
 243                 if (NCE_ISREACHABLE(ncec) &&
 244                     delta < (uint64_t)ill->ill_reachable_time) {
 245                         ncec_ip2mac_response(ip2m, ncec);
 246                         ip2m->ip2mac_err = 0;
 247                 } else {
 248                         ip2m->ip2mac_err = ESRCH;
 249                 }
 250                 mutex_exit(&ncec->ncec_lock);
 251                 goto done;
 252         } else {
 253                 if (err != 0 && err != EEXIST) {
 254                         ip2m->ip2mac_err = err;
 255                         goto done;
 256                 }
 257         }
 258         ncec = nce->nce_common;
 259         delta = TICK_TO_MSEC(ddi_get_lbolt64()) - ncec->ncec_last;
 260         mutex_enter(&ncec->ncec_lock);
 261         if (NCE_ISCONDEMNED(ncec)) {
 262                 ip2m->ip2mac_err = ESRCH;
 263         } else {
 264                 if (NCE_ISREACHABLE(ncec)) {
 265                         if (NCE_MYADDR(ncec) ||
 266                             delta < (uint64_t)ill->ill_reachable_time) {
 267                                 ncec_ip2mac_response(ip2m, ncec);
 268                                 ip2m->ip2mac_err = 0;
 269                                 mutex_exit(&ncec->ncec_lock);
 270                                 goto done;
 271                         }
 272                         /*
 273                          * Since we do not control the packet output
 274                          * path for ip2mac() callers, we need to verify
 275                          * if the existing information in the nce is
 276                          * very old, and retrigger resolution if necessary.
 277                          * We will not return the existing stale
 278                          * information until it is verified through a
 279                          * resolver request/response exchange.
 280                          *
 281                          * In the future, we may want to support extensions
 282                          * that do additional callbacks on link-layer updates,
 283                          * so that we can return the stale information but
 284                          * also update the caller if the lladdr changes.
 285                          */
 286                         ncec->ncec_rcnt = ill->ill_xmit_count;
 287                         ncec->ncec_state = ND_PROBE;
 288                         need_resolve = B_TRUE; /* reachable but very old nce */
 289                 } else if (ncec->ncec_state == ND_INITIAL) {
 290                         need_resolve = B_TRUE; /* ND_INITIAL nce */
 291                         ncec->ncec_state = ND_INCOMPLETE;
 292                 }
 293                 /*
 294                  * NCE not known to be reachable in the recent past. We must
 295                  * reconfirm the information before returning it to the caller
 296                  */
 297                 if (ncec->ncec_rcnt > 0) {
 298                         /*
 299                          * Still resolving this ncec, so we can queue the
 300                          * callback information in ncec->ncec_cb
 301                          */
 302                         ip2mid = ncec_add_cb(ncec, cb, cbarg);
 303                         ip2m->ip2mac_err = EINPROGRESS;
 304                 } else {
 305                         /*
 306                          * No more retransmits allowed -- resolution failed.
 307                          */
 308                         ip2m->ip2mac_err = ESRCH;
 309                 }
 310         }
 311         mutex_exit(&ncec->ncec_lock);
 312 done:
 313         /*
 314          * if NCE_ISREACHABLE(ncec) but very old, or if it is ND_INITIAL,
 315          * trigger resolve.
 316          */
 317         if (need_resolve)
 318                 ip_ndp_resolve(ncec);
 319         if (nce != NULL)
 320                 nce_refrele(nce);
 321         netstack_rele(ns);
 322         ill_refrele(ill);
 323         return (ip2mid);
 324 }
 325
 326 /*
 327  * data passed to ncec_walk for canceling outstanding callbacks.
 328  */
 329 typedef struct ip2mac_cancel_data_s {
 330         ip2mac_id_t ip2m_cancel_id;
 331         int     ip2m_cancel_err;
 332 } ip2mac_cancel_data_t;
 333
 334 /*
 335  * callback invoked for each active ncec. If the ip2mac_id_t corresponds
 336  * to an active nce_cb_t in the ncec's callback list, we want to remove
 337  * the callback (if there are no walkers) or return EBUSY to the caller
 338  */
 339 static void
 340 ip2mac_cancel_callback(ncec_t *ncec, void *arg)
 341 {
 342         ip2mac_cancel_data_t *ip2m_wdata = arg;
 343         ncec_cb_t *ip2m_nce_cb = ip2m_wdata->ip2m_cancel_id;
 344         ncec_cb_t *ncec_cb;
 345
 346         if (ip2m_nce_cb->ncec_cb_id != ncec)
 347                 return;
 348
 349         mutex_enter(&ncec->ncec_lock);
 350         if (list_is_empty(&ncec->ncec_cb)) {
 351                 mutex_exit(&ncec->ncec_lock);
 352                 return;
 353         }
 354         /*
 355          * IP does not hold internal locks like nce_lock across calls to
 356          * other subsystems for fear of recursive lock entry and lock
 357          * hierarchy violation. The caller may be holding locks across
 358          * the call to IP. (It would be ideal if no subsystem holds locks
 359          * across calls into another subsystem, especially if calls can
 360          * happen in either direction).
 361          */
 362         ncec_cb = list_head(&ncec->ncec_cb);
 363         for (; ncec_cb != NULL; ncec_cb = list_next(&ncec->ncec_cb, ncec_cb)) {
 364                 if (ncec_cb != ip2m_nce_cb)
 365                         continue;
 366                 /*
 367                  * If there are no walkers we can remove the nce_cb.
 368                  * Otherwise the exiting walker will clean up.
 369                  */
 370                 if (ncec->ncec_cb_walker_cnt == 0) {
 371                         list_remove(&ncec->ncec_cb, ncec_cb);
 372                 } else {
 373                         ip2m_wdata->ip2m_cancel_err = EBUSY;
 374                 }
 375                 break;
 376         }
 377         mutex_exit(&ncec->ncec_lock);
 378 }
 379
 380 /*
 381  * cancel an outstanding timeout set up via ip2mac
 382  */
 383 int
 384 ip2mac_cancel(ip2mac_id_t ip2mid, zoneid_t zoneid)
 385 {
 386         netstack_t      *ns;
 387         ip_stack_t      *ipst;
 388         ip2mac_cancel_data_t ip2m_wdata;
 389
 390         ns = netstack_find_by_zoneid(zoneid);
 391         if (ns == NULL) {
 392                 ip2m_wdata.ip2m_cancel_err = EINVAL;
 393                 return (ip2m_wdata.ip2m_cancel_err);
 394         }
 395         /*
 396          * For exclusive stacks we reset the zoneid to zero
 397          * since IP uses the global zoneid in the exclusive stacks.
 398          */
 399         if (ns->netstack_stackid != GLOBAL_NETSTACKID)
 400                 zoneid = GLOBAL_ZONEID;
 401         ipst = ns->netstack_ip;
 402
 403         ip2m_wdata.ip2m_cancel_id = ip2mid;
 404         ip2m_wdata.ip2m_cancel_err = 0;
 405         ncec_walk(NULL, ip2mac_cancel_callback, &ip2m_wdata, ipst);
 406         /*
 407          * We may return EBUSY if a walk to dispatch callbacks is
 408          * in progress, in which case the caller needs to synchronize
 409          * with the registered callback function to make sure the
 410          * module does not exit when there is a callback pending.
 411          */
 412         netstack_rele(ns);
 413         return (ip2m_wdata.ip2m_cancel_err);
 414 }