lib/libdmsg/msg_lnk.c

   1 /*
   2  * Copyright (c) 2012 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@dragonflybsd.org>
   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 /*
  35  * LNK_SPAN PROTOCOL SUPPORT FUNCTIONS
  36  *
  37  * This code supports the LNK_SPAN protocol.  Essentially all PFS's
  38  * clients and services rendezvous with the userland hammer2 service and
  39  * open LNK_SPAN transactions using a message header linkid of 0,
  40  * registering any PFS's they have connectivity to with us.
  41  *
  42  * --
  43  *
  44  * Each registration maintains its own open LNK_SPAN message transaction.
  45  * The SPANs are collected, aggregated, and retransmitted over available
  46  * connections through the maintainance of additional LNK_SPAN message
  47  * transactions on each link.
  48  *
  49  * The msgid for each active LNK_SPAN transaction we receive allows us to
  50  * send a message to the target PFS (which might be one of many belonging
  51  * to the same cluster), by specifying that msgid as the linkid in any
  52  * message we send to the target PFS.
  53  *
  54  * Similarly the msgid we allocate for any LNK_SPAN transaction we transmit
  55  * (and remember we will maintain multiple open LNK_SPAN transactions on
  56  * each connection representing the topology span, so every node sees every
  57  * other node as a separate open transaction).  So, similarly the msgid for
  58  * these active transactions which we initiated can be used by the other
  59  * end to route messages through us to another node, ultimately winding up
  60  * at the identified hammer2 PFS.  We have to adjust the spanid in the message
  61  * header at each hop to be representative of the outgoing LNK_SPAN we
  62  * are forwarding the message through.
  63  *
  64  * --
  65  *
  66  * If we were to retransmit every LNK_SPAN transaction we receive it would
  67  * create a huge mess, so we have to aggregate all received LNK_SPAN
  68  * transactions, sort them by the fsid (the cluster) and sub-sort them by
  69  * the pfs_fsid (individual nodes in the cluster), and only retransmit
  70  * (create outgoing transactions) for a subset of the nearest distance-hops
  71  * for each individual node.
  72  *
  73  * The higher level protocols can then issue transactions to the nodes making
  74  * up a cluster to perform all actions required.
  75  *
  76  * --
  77  *
  78  * Since this is a large topology and a spanning tree protocol, links can
  79  * go up and down all the time.  Any time a link goes down its transaction
  80  * is closed.  The transaction has to be closed on both ends before we can
  81  * delete (and potentially reuse) the related spanid.  The LNK_SPAN being
  82  * closed may have been propagated out to other connections and those related
  83  * LNK_SPANs are also closed.  Ultimately all routes via the lost LNK_SPAN
  84  * go away, ultimately reaching all sources and all targets.
  85  *
  86  * Any messages in-transit using a route that goes away will be thrown away.
  87  * Open transactions are only tracked at the two end-points.  When a link
  88  * failure propagates to an end-point the related open transactions lose
  89  * their spanid and are automatically aborted.
  90  *
  91  * It is important to note that internal route nodes cannot just associate
  92  * a lost LNK_SPAN transaction with another route to the same destination.
  93  * Message transactions MUST be serialized and MUST be ordered.  All messages
  94  * for a transaction must run over the same route.  So if the route used by
  95  * an active transaction is lost, the related messages will be fully aborted
  96  * and the higher protocol levels will retry as appropriate.
  97  *
  98  * FULLY ABORTING A ROUTED MESSAGE is handled via link-failure propagation
  99  * back to the originator.  Only the originator keeps tracks of a message.
 100  * Routers just pass it through.  If a route is lost during transit the
 101  * message is simply thrown away.
 102  *
 103  * It is also important to note that several paths to the same PFS can be
 104  * propagated along the same link, which allows concurrency and even
 105  * redundancy over several network interfaces or via different routes through
 106  * the topology.  Any given transaction will use only a single route but busy
 107  * servers will often have hundreds of transactions active simultaniously,
 108  * so having multiple active paths through the network topology for A<->B
 109  * will improve performance.
 110  *
 111  * --
 112  *
 113  * Most protocols consolidate operations rather than simply relaying them.
 114  * This is particularly true of LEAF protocols (such as strict HAMMER2
 115  * clients), of which there can be millions connecting into the cluster at
 116  * various points.  The SPAN protocol is not used for these LEAF elements.
 117  *
 118  * Instead the primary service they connect to implements a proxy for the
 119  * client protocols so the core topology only has to propagate a couple of
 120  * LNK_SPANs and not millions.  LNK_SPANs are meant to be used only for
 121  * core master nodes and satellite slaves and cache nodes.
 122  */
 123
 124 #include "dmsg_local.h"
 125
 126 /*
 127  * Maximum spanning tree distance.  This has the practical effect of
 128  * stopping tail-chasing closed loops when a feeder span is lost.
 129  */
 130 #define DMSG_SPAN_MAXDIST       16
 131
 132 /*
 133  * RED-BLACK TREE DEFINITIONS
 134  *
 135  * We need to track:
 136  *
 137  * (1) shared fsid's (a cluster).
 138  * (2) unique fsid's (a node in a cluster) <--- LNK_SPAN transactions.
 139  *
 140  * We need to aggegate all active LNK_SPANs, aggregate, and create our own
 141  * outgoing LNK_SPAN transactions on each of our connections representing
 142  * the aggregated state.
 143  *
 144  * h2span_conn          - list of iocom connections who wish to receive SPAN
 145  *                        propagation from other connections.  Might contain
 146  *                        a filter string.  Only iocom's with an open
 147  *                        LNK_CONN transactions are applicable for SPAN
 148  *                        propagation.
 149  *
 150  * h2span_relay         - List of links relayed (via SPAN).  Essentially
 151  *                        each relay structure represents a LNK_SPAN
 152  *                        transaction that we initiated, verses h2span_link
 153  *                        which is a LNK_SPAN transaction that we received.
 154  *
 155  * --
 156  *
 157  * h2span_cluster       - Organizes the shared fsid's.  One structure for
 158  *                        each cluster.
 159  *
 160  * h2span_node          - Organizes the nodes in a cluster.  One structure
 161  *                        for each unique {cluster,node}, aka {fsid, pfs_fsid}.
 162  *
 163  * h2span_link          - Organizes all incoming and outgoing LNK_SPAN message
 164  *                        transactions related to a node.
 165  *
 166  *                        One h2span_link structure for each incoming LNK_SPAN
 167  *                        transaction.  Links selected for propagation back
 168  *                        out are also where the outgoing LNK_SPAN messages
 169  *                        are indexed into (so we can propagate changes).
 170  *
 171  *                        The h2span_link's use a red-black tree to sort the
 172  *                        distance hop metric for the incoming LNK_SPAN.  We
 173  *                        then select the top N for outgoing.  When the
 174  *                        topology changes the top N may also change and cause
 175  *                        new outgoing LNK_SPAN transactions to be opened
 176  *                        and less desireable ones to be closed, causing
 177  *                        transactional aborts within the message flow in
 178  *                        the process.
 179  *
 180  * Also note            - All outgoing LNK_SPAN message transactions are also
 181  *                        entered into a red-black tree for use by the routing
 182  *                        function.  This is handled by msg.c in the state
 183  *                        code, not here.
 184  */
 185
 186 struct h2span_link;
 187 struct h2span_relay;
 188 TAILQ_HEAD(h2span_media_queue, h2span_media);
 189 TAILQ_HEAD(h2span_conn_queue, h2span_conn);
 190 TAILQ_HEAD(h2span_relay_queue, h2span_relay);
 191
 192 RB_HEAD(h2span_cluster_tree, h2span_cluster);
 193 RB_HEAD(h2span_node_tree, h2span_node);
 194 RB_HEAD(h2span_link_tree, h2span_link);
 195 RB_HEAD(h2span_relay_tree, h2span_relay);
 196
 197 /*
 198  * This represents a media
 199  */
 200 struct h2span_media {
 201         TAILQ_ENTRY(h2span_media) entry;
 202         uuid_t  mediaid;
 203         int     refs;
 204         struct h2span_media_config {
 205                 dmsg_vol_data_t         copy_run;
 206                 dmsg_vol_data_t         copy_pend;
 207                 pthread_t               thread;
 208                 pthread_cond_t          cond;
 209                 int                     ctl;
 210                 int                     fd;
 211                 dmsg_iocom_t            iocom;
 212                 pthread_t               iocom_thread;
 213                 enum { H2MC_STOPPED, H2MC_CONNECT, H2MC_RUNNING } state;
 214         } config[DMSG_COPYID_COUNT];
 215 };
 216
 217 typedef struct h2span_media_config h2span_media_config_t;
 218
 219 #define H2CONFCTL_STOP          0x00000001
 220 #define H2CONFCTL_UPDATE        0x00000002
 221
 222 /*
 223  * Received LNK_CONN transaction enables SPAN protocol over connection.
 224  * (may contain filter).  Typically one for each mount and several may
 225  * share the same media.
 226  */
 227 struct h2span_conn {
 228         TAILQ_ENTRY(h2span_conn) entry;
 229         struct h2span_relay_tree tree;
 230         struct h2span_media *media;
 231         dmsg_state_t *state;
 232 };
 233
 234 /*
 235  * All received LNK_SPANs are organized by cluster (pfs_clid),
 236  * node (pfs_fsid), and link (received LNK_SPAN transaction).
 237  */
 238 struct h2span_cluster {
 239         RB_ENTRY(h2span_cluster) rbnode;
 240         struct h2span_node_tree tree;
 241         uuid_t  pfs_clid;               /* shared fsid */
 242         int     refs;                   /* prevents destruction */
 243 };
 244
 245 struct h2span_node {
 246         RB_ENTRY(h2span_node) rbnode;
 247         struct h2span_link_tree tree;
 248         struct h2span_cluster *cls;
 249         uint8_t peer_type;
 250         uuid_t  pfs_fsid;               /* unique fsid */
 251         char    label[64];
 252 };
 253
 254 struct h2span_link {
 255         RB_ENTRY(h2span_link) rbnode;
 256         dmsg_state_t    *state;         /* state<->link */
 257         struct h2span_node *node;       /* related node */
 258         int32_t dist;
 259         struct h2span_relay_queue relayq; /* relay out */
 260         struct dmsg_router *router;     /* route out this link */
 261 };
 262
 263 /*
 264  * Any LNK_SPAN transactions we receive which are relayed out other
 265  * connections utilize this structure to track the LNK_SPAN transaction
 266  * we initiate on the other connections, if selected for relay.
 267  *
 268  * In many respects this is the core of the protocol... actually figuring
 269  * out what LNK_SPANs to relay.  The spanid used for relaying is the
 270  * address of the 'state' structure, which is why h2span_relay has to
 271  * be entered into a RB-TREE based at h2span_conn (so we can look
 272  * up the spanid to validate it).
 273  *
 274  * NOTE: Messages can be received via the LNK_SPAN transaction the
 275  *       relay maintains, and can be replied via relay->router, but
 276  *       messages are NOT initiated via a relay.  Messages are initiated
 277  *       via incoming links (h2span_link's).
 278  *
 279  *       relay->link represents the link being relayed, NOT the LNK_SPAN
 280  *       transaction the relay is holding open.
 281  */
 282 struct h2span_relay {
 283         RB_ENTRY(h2span_relay) rbnode;  /* from h2span_conn */
 284         TAILQ_ENTRY(h2span_relay) entry; /* from link */
 285         struct h2span_conn *conn;
 286         dmsg_state_t    *state;         /* transmitted LNK_SPAN */
 287         struct h2span_link *link;       /* LNK_SPAN being relayed */
 288         struct dmsg_router      *router;/* route out this relay */
 289 };
 290
 291
 292 typedef struct h2span_media h2span_media_t;
 293 typedef struct h2span_conn h2span_conn_t;
 294 typedef struct h2span_cluster h2span_cluster_t;
 295 typedef struct h2span_node h2span_node_t;
 296 typedef struct h2span_link h2span_link_t;
 297 typedef struct h2span_relay h2span_relay_t;
 298
 299 static
 300 int
 301 h2span_cluster_cmp(h2span_cluster_t *cls1, h2span_cluster_t *cls2)
 302 {
 303         return(uuid_compare(&cls1->pfs_clid, &cls2->pfs_clid, NULL));
 304 }
 305
 306 static
 307 int
 308 h2span_node_cmp(h2span_node_t *node1, h2span_node_t *node2)
 309 {
 310         int r;
 311
 312         if (node1->peer_type < node2->peer_type)
 313                 return(-1);
 314         if (node1->peer_type > node2->peer_type)
 315                 return(1);
 316         r = uuid_compare(&node1->pfs_fsid, &node2->pfs_fsid, NULL);
 317         if (r == 0 && node1->peer_type == DMSG_PEER_BLOCK)
 318                 r = strcmp(node1->label, node2->label);
 319         return (r);
 320 }
 321
 322 /*
 323  * Sort/subsort must match h2span_relay_cmp() under any given node
 324  * to make the aggregation algorithm easier, so the best links are
 325  * in the same sorted order as the best relays.
 326  *
 327  * NOTE: We cannot use link*->state->msgid because this msgid is created
 328  *       by each remote host and thus might wind up being the same.
 329  */
 330 static
 331 int
 332 h2span_link_cmp(h2span_link_t *link1, h2span_link_t *link2)
 333 {
 334         if (link1->dist < link2->dist)
 335                 return(-1);
 336         if (link1->dist > link2->dist)
 337                 return(1);
 338 #if 1
 339         if ((uintptr_t)link1->state < (uintptr_t)link2->state)
 340                 return(-1);
 341         if ((uintptr_t)link1->state > (uintptr_t)link2->state)
 342                 return(1);
 343 #else
 344         if (link1->state->msgid < link2->state->msgid)
 345                 return(-1);
 346         if (link1->state->msgid > link2->state->msgid)
 347                 return(1);
 348 #endif
 349         return(0);
 350 }
 351
 352 /*
 353  * Relay entries are sorted by node, subsorted by distance and link
 354  * address (so we can match up the conn->tree relay topology with
 355  * a node's link topology).
 356  */
 357 static
 358 int
 359 h2span_relay_cmp(h2span_relay_t *relay1, h2span_relay_t *relay2)
 360 {
 361         h2span_link_t *link1 = relay1->link;
 362         h2span_link_t *link2 = relay2->link;
 363
 364         if ((intptr_t)link1->node < (intptr_t)link2->node)
 365                 return(-1);
 366         if ((intptr_t)link1->node > (intptr_t)link2->node)
 367                 return(1);
 368         if (link1->dist < link2->dist)
 369                 return(-1);
 370         if (link1->dist > link2->dist)
 371                 return(1);
 372 #if 1
 373         if ((uintptr_t)link1->state < (uintptr_t)link2->state)
 374                 return(-1);
 375         if ((uintptr_t)link1->state > (uintptr_t)link2->state)
 376                 return(1);
 377 #else
 378         if (link1->state->msgid < link2->state->msgid)
 379                 return(-1);
 380         if (link1->state->msgid > link2->state->msgid)
 381                 return(1);
 382 #endif
 383         return(0);
 384 }
 385
 386 RB_PROTOTYPE_STATIC(h2span_cluster_tree, h2span_cluster,
 387              rbnode, h2span_cluster_cmp);
 388 RB_PROTOTYPE_STATIC(h2span_node_tree, h2span_node,
 389              rbnode, h2span_node_cmp);
 390 RB_PROTOTYPE_STATIC(h2span_link_tree, h2span_link,
 391              rbnode, h2span_link_cmp);
 392 RB_PROTOTYPE_STATIC(h2span_relay_tree, h2span_relay,
 393              rbnode, h2span_relay_cmp);
 394
 395 RB_GENERATE_STATIC(h2span_cluster_tree, h2span_cluster,
 396              rbnode, h2span_cluster_cmp);
 397 RB_GENERATE_STATIC(h2span_node_tree, h2span_node,
 398              rbnode, h2span_node_cmp);
 399 RB_GENERATE_STATIC(h2span_link_tree, h2span_link,
 400              rbnode, h2span_link_cmp);
 401 RB_GENERATE_STATIC(h2span_relay_tree, h2span_relay,
 402              rbnode, h2span_relay_cmp);
 403
 404 /*
 405  * Global mutex protects cluster_tree lookups, connq, mediaq.
 406  */
 407 static pthread_mutex_t cluster_mtx;
 408 static struct h2span_cluster_tree cluster_tree = RB_INITIALIZER(cluster_tree);
 409 static struct h2span_conn_queue connq = TAILQ_HEAD_INITIALIZER(connq);
 410 static struct h2span_media_queue mediaq = TAILQ_HEAD_INITIALIZER(mediaq);
 411
 412 static void dmsg_lnk_span(dmsg_msg_t *msg);
 413 static void dmsg_lnk_conn(dmsg_msg_t *msg);
 414 static void dmsg_lnk_relay(dmsg_msg_t *msg);
 415 static void dmsg_relay_scan(h2span_conn_t *conn, h2span_node_t *node);
 416 static void dmsg_relay_delete(h2span_relay_t *relay);
 417
 418 static void *dmsg_volconf_thread(void *info);
 419 static void dmsg_volconf_stop(h2span_media_config_t *conf);
 420 static void dmsg_volconf_start(h2span_media_config_t *conf,
 421                                 const char *hostname);
 422
 423 void
 424 dmsg_msg_lnk_signal(dmsg_router_t *router __unused)
 425 {
 426         pthread_mutex_lock(&cluster_mtx);
 427         dmsg_relay_scan(NULL, NULL);
 428         pthread_mutex_unlock(&cluster_mtx);
 429 }
 430
 431 /*
 432  * Receive a DMSG_PROTO_LNK message.  This only called for
 433  * one-way and opening-transactions since state->func will be assigned
 434  * in all other cases.
 435  */
 436 void
 437 dmsg_msg_lnk(dmsg_msg_t *msg)
 438 {
 439         switch(msg->any.head.cmd & DMSGF_BASECMDMASK) {
 440         case DMSG_LNK_CONN:
 441                 dmsg_lnk_conn(msg);
 442                 break;
 443         case DMSG_LNK_SPAN:
 444                 dmsg_lnk_span(msg);
 445                 break;
 446         default:
 447                 fprintf(stderr,
 448                         "MSG_PROTO_LNK: Unknown msg %08x\n", msg->any.head.cmd);
 449                 dmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
 450                 /* state invalid after reply */
 451                 break;
 452         }
 453 }
 454
 455 void
 456 dmsg_lnk_conn(dmsg_msg_t *msg)
 457 {
 458         dmsg_state_t *state = msg->state;
 459         h2span_media_t *media;
 460         h2span_media_config_t *conf;
 461         h2span_conn_t *conn;
 462         h2span_relay_t *relay;
 463         char *alloc = NULL;
 464         int i;
 465
 466         pthread_mutex_lock(&cluster_mtx);
 467
 468         switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
 469         case DMSG_LNK_CONN | DMSGF_CREATE:
 470         case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_DELETE:
 471                 /*
 472                  * On transaction start we allocate a new h2span_conn and
 473                  * acknowledge the request, leaving the transaction open.
 474                  * We then relay priority-selected SPANs.
 475                  */
 476                 fprintf(stderr, "LNK_CONN(%08x): %s/%s\n",
 477                         (uint32_t)msg->any.head.msgid,
 478                         dmsg_uuid_to_str(&msg->any.lnk_conn.pfs_clid,
 479                                             &alloc),
 480                         msg->any.lnk_conn.label);
 481                 free(alloc);
 482
 483                 conn = dmsg_alloc(sizeof(*conn));
 484
 485                 RB_INIT(&conn->tree);
 486                 conn->state = state;
 487                 state->func = dmsg_lnk_conn;
 488                 state->any.conn = conn;
 489                 TAILQ_INSERT_TAIL(&connq, conn, entry);
 490
 491                 /*
 492                  * Set up media
 493                  */
 494                 TAILQ_FOREACH(media, &mediaq, entry) {
 495                         if (uuid_compare(&msg->any.lnk_conn.mediaid,
 496                                          &media->mediaid, NULL) == 0) {
 497                                 break;
 498                         }
 499                 }
 500                 if (media == NULL) {
 501                         media = dmsg_alloc(sizeof(*media));
 502                         media->mediaid = msg->any.lnk_conn.mediaid;
 503                         TAILQ_INSERT_TAIL(&mediaq, media, entry);
 504                 }
 505                 conn->media = media;
 506                 ++media->refs;
 507
 508                 if ((msg->any.head.cmd & DMSGF_DELETE) == 0) {
 509                         dmsg_msg_result(msg, 0);
 510                         dmsg_router_signal(msg->router);
 511                         break;
 512                 }
 513                 /* FALL THROUGH */
 514         case DMSG_LNK_CONN | DMSGF_DELETE:
 515         case DMSG_LNK_ERROR | DMSGF_DELETE:
 516 deleteconn:
 517                 /*
 518                  * On transaction terminate we clean out our h2span_conn
 519                  * and acknowledge the request, closing the transaction.
 520                  */
 521                 fprintf(stderr, "LNK_CONN: Terminated\n");
 522                 conn = state->any.conn;
 523                 assert(conn);
 524
 525                 /*
 526                  * Clean out the media structure. If refs drops to zero we
 527                  * also clean out the media config threads.  These threads
 528                  * maintain span connections to other hammer2 service daemons.
 529                  */
 530                 media = conn->media;
 531                 if (--media->refs == 0) {
 532                         fprintf(stderr, "Shutting down media spans\n");
 533                         for (i = 0; i < DMSG_COPYID_COUNT; ++i) {
 534                                 conf = &media->config[i];
 535
 536                                 if (conf->thread == NULL)
 537                                         continue;
 538                                 conf->ctl = H2CONFCTL_STOP;
 539                                 pthread_cond_signal(&conf->cond);
 540                         }
 541                         for (i = 0; i < DMSG_COPYID_COUNT; ++i) {
 542                                 conf = &media->config[i];
 543
 544                                 if (conf->thread == NULL)
 545                                         continue;
 546                                 pthread_mutex_unlock(&cluster_mtx);
 547                                 pthread_join(conf->thread, NULL);
 548                                 pthread_mutex_lock(&cluster_mtx);
 549                                 conf->thread = NULL;
 550                                 pthread_cond_destroy(&conf->cond);
 551                         }
 552                         fprintf(stderr, "Media shutdown complete\n");
 553                         TAILQ_REMOVE(&mediaq, media, entry);
 554                         dmsg_free(media);
 555                 }
 556
 557                 /*
 558                  * Clean out all relays.  This requires terminating each
 559                  * relay transaction.
 560                  */
 561                 while ((relay = RB_ROOT(&conn->tree)) != NULL) {
 562                         dmsg_relay_delete(relay);
 563                 }
 564
 565                 /*
 566                  * Clean out conn
 567                  */
 568                 conn->media = NULL;
 569                 conn->state = NULL;
 570                 msg->state->any.conn = NULL;
 571                 TAILQ_REMOVE(&connq, conn, entry);
 572                 dmsg_free(conn);
 573
 574                 dmsg_msg_reply(msg, 0);
 575                 /* state invalid after reply */
 576                 break;
 577         case DMSG_LNK_VOLCONF:
 578                 /*
 579                  * One-way volume-configuration message is transmitted
 580                  * over the open LNK_CONN transaction.
 581                  */
 582                 fprintf(stderr, "RECEIVED VOLCONF\n");
 583                 if (msg->any.lnk_volconf.index < 0 ||
 584                     msg->any.lnk_volconf.index >= DMSG_COPYID_COUNT) {
 585                         fprintf(stderr, "VOLCONF: ILLEGAL INDEX %d\n",
 586                                 msg->any.lnk_volconf.index);
 587                         break;
 588                 }
 589                 if (msg->any.lnk_volconf.copy.path[sizeof(msg->any.lnk_volconf.copy.path) - 1] != 0 ||
 590                     msg->any.lnk_volconf.copy.path[0] == 0) {
 591                         fprintf(stderr, "VOLCONF: ILLEGAL PATH %d\n",
 592                                 msg->any.lnk_volconf.index);
 593                         break;
 594                 }
 595                 conn = msg->state->any.conn;
 596                 if (conn == NULL) {
 597                         fprintf(stderr, "VOLCONF: LNK_CONN is missing\n");
 598                         break;
 599                 }
 600                 conf = &conn->media->config[msg->any.lnk_volconf.index];
 601                 conf->copy_pend = msg->any.lnk_volconf.copy;
 602                 conf->ctl |= H2CONFCTL_UPDATE;
 603                 if (conf->thread == NULL) {
 604                         fprintf(stderr, "VOLCONF THREAD STARTED\n");
 605                         pthread_cond_init(&conf->cond, NULL);
 606                         pthread_create(&conf->thread, NULL,
 607                                        dmsg_volconf_thread, (void *)conf);
 608                 }
 609                 pthread_cond_signal(&conf->cond);
 610                 break;
 611         default:
 612                 /*
 613                  * Failsafe
 614                  */
 615                 if (msg->any.head.cmd & DMSGF_DELETE)
 616                         goto deleteconn;
 617                 dmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
 618                 break;
 619         }
 620         pthread_mutex_unlock(&cluster_mtx);
 621 }
 622
 623 void
 624 dmsg_lnk_span(dmsg_msg_t *msg)
 625 {
 626         dmsg_state_t *state = msg->state;
 627         h2span_cluster_t dummy_cls;
 628         h2span_node_t dummy_node;
 629         h2span_cluster_t *cls;
 630         h2span_node_t *node;
 631         h2span_link_t *slink;
 632         h2span_relay_t *relay;
 633         char *alloc = NULL;
 634
 635         assert((msg->any.head.cmd & DMSGF_REPLY) == 0);
 636
 637         pthread_mutex_lock(&cluster_mtx);
 638
 639         /*
 640          * On transaction start we initialize the tracking infrastructure
 641          */
 642         if (msg->any.head.cmd & DMSGF_CREATE) {
 643                 assert(state->func == NULL);
 644                 state->func = dmsg_lnk_span;
 645
 646                 msg->any.lnk_span.label[sizeof(msg->any.lnk_span.label)-1] = 0;
 647
 648                 /*
 649                  * Find the cluster
 650                  */
 651                 dummy_cls.pfs_clid = msg->any.lnk_span.pfs_clid;
 652                 cls = RB_FIND(h2span_cluster_tree, &cluster_tree, &dummy_cls);
 653                 if (cls == NULL) {
 654                         cls = dmsg_alloc(sizeof(*cls));
 655                         cls->pfs_clid = msg->any.lnk_span.pfs_clid;
 656                         RB_INIT(&cls->tree);
 657                         RB_INSERT(h2span_cluster_tree, &cluster_tree, cls);
 658                 }
 659
 660                 /*
 661                  * Find the node
 662                  */
 663                 dummy_node.pfs_fsid = msg->any.lnk_span.pfs_fsid;
 664                 dummy_node.peer_type = msg->any.lnk_span.peer_type;
 665                 snprintf(dummy_node.label, sizeof(dummy_node.label),
 666                          "%s", msg->any.lnk_span.label);
 667                 node = RB_FIND(h2span_node_tree, &cls->tree, &dummy_node);
 668                 if (node == NULL) {
 669                         node = dmsg_alloc(sizeof(*node));
 670                         node->pfs_fsid = msg->any.lnk_span.pfs_fsid;
 671                         node->peer_type = msg->any.lnk_span.peer_type;
 672                         snprintf(node->label, sizeof(node->label),
 673                                  "%s", msg->any.lnk_span.label);
 674                         node->cls = cls;
 675                         RB_INIT(&node->tree);
 676                         RB_INSERT(h2span_node_tree, &cls->tree, node);
 677                 }
 678
 679                 /*
 680                  * Create the link
 681                  */
 682                 assert(state->any.link == NULL);
 683                 slink = dmsg_alloc(sizeof(*slink));
 684                 TAILQ_INIT(&slink->relayq);
 685                 slink->node = node;
 686                 slink->dist = msg->any.lnk_span.dist;
 687                 slink->state = state;
 688                 state->any.link = slink;
 689
 690                 /*
 691                  * Embedded router structure in link for message forwarding.
 692                  *
 693                  * The spanning id for the router is the message id of
 694                  * the SPAN link it is embedded in, allowing messages to
 695                  * be routed via &slink->router.
 696                  */
 697                 slink->router = dmsg_router_alloc();
 698                 slink->router->iocom = state->iocom;
 699                 slink->router->link = slink;
 700                 slink->router->target = state->msgid;
 701                 dmsg_router_connect(slink->router);
 702
 703                 RB_INSERT(h2span_link_tree, &node->tree, slink);
 704
 705                 fprintf(stderr, "LNK_SPAN(thr %p): %p %s/%s dist=%d\n",
 706                         msg->router->iocom,
 707                         slink,
 708                         dmsg_uuid_to_str(&msg->any.lnk_span.pfs_clid,
 709                                             &alloc),
 710                         msg->any.lnk_span.label,
 711                         msg->any.lnk_span.dist);
 712                 free(alloc);
 713 #if 0
 714                 dmsg_relay_scan(NULL, node);
 715 #endif
 716                 dmsg_router_signal(msg->router);
 717         }
 718
 719         /*
 720          * On transaction terminate we remove the tracking infrastructure.
 721          */
 722         if (msg->any.head.cmd & DMSGF_DELETE) {
 723                 slink = state->any.link;
 724                 assert(slink != NULL);
 725                 node = slink->node;
 726                 cls = node->cls;
 727
 728                 fprintf(stderr, "LNK_DELE(thr %p): %p %s/%s dist=%d\n",
 729                         msg->router->iocom,
 730                         slink,
 731                         dmsg_uuid_to_str(&cls->pfs_clid, &alloc),
 732                         state->msg->any.lnk_span.label,
 733                         state->msg->any.lnk_span.dist);
 734                 free(alloc);
 735
 736                 /*
 737                  * Remove the router from consideration
 738                  */
 739                 dmsg_router_disconnect(&slink->router);
 740
 741                 /*
 742                  * Clean out all relays.  This requires terminating each
 743                  * relay transaction.
 744                  */
 745                 while ((relay = TAILQ_FIRST(&slink->relayq)) != NULL) {
 746                         dmsg_relay_delete(relay);
 747                 }
 748
 749                 /*
 750                  * Clean out the topology
 751                  */
 752                 RB_REMOVE(h2span_link_tree, &node->tree, slink);
 753                 if (RB_EMPTY(&node->tree)) {
 754                         RB_REMOVE(h2span_node_tree, &cls->tree, node);
 755                         if (RB_EMPTY(&cls->tree) && cls->refs == 0) {
 756                                 RB_REMOVE(h2span_cluster_tree,
 757                                           &cluster_tree, cls);
 758                                 dmsg_free(cls);
 759                         }
 760                         node->cls = NULL;
 761                         dmsg_free(node);
 762                         node = NULL;
 763                 }
 764                 state->any.link = NULL;
 765                 slink->state = NULL;
 766                 slink->node = NULL;
 767                 dmsg_free(slink);
 768
 769                 /*
 770                  * We have to terminate the transaction
 771                  */
 772                 dmsg_state_reply(state, 0);
 773                 /* state invalid after reply */
 774
 775                 /*
 776                  * If the node still exists issue any required updates.  If
 777                  * it doesn't then all related relays have already been
 778                  * removed and there's nothing left to do.
 779                  */
 780 #if 0
 781                 if (node)
 782                         dmsg_relay_scan(NULL, node);
 783 #endif
 784                 if (node)
 785                         dmsg_router_signal(msg->router);
 786         }
 787
 788         pthread_mutex_unlock(&cluster_mtx);
 789 }
 790
 791 /*
 792  * Messages received on relay SPANs.  These are open transactions so it is
 793  * in fact possible for the other end to close the transaction.
 794  *
 795  * XXX MPRACE on state structure
 796  */
 797 static void
 798 dmsg_lnk_relay(dmsg_msg_t *msg)
 799 {
 800         dmsg_state_t *state = msg->state;
 801         h2span_relay_t *relay;
 802
 803         assert(msg->any.head.cmd & DMSGF_REPLY);
 804
 805         if (msg->any.head.cmd & DMSGF_DELETE) {
 806                 pthread_mutex_lock(&cluster_mtx);
 807                 if ((relay = state->any.relay) != NULL) {
 808                         dmsg_relay_delete(relay);
 809                 } else {
 810                         dmsg_state_reply(state, 0);
 811                 }
 812                 pthread_mutex_unlock(&cluster_mtx);
 813         }
 814 }
 815
 816 /*
 817  * Update relay transactions for SPANs.
 818  *
 819  * Called with cluster_mtx held.
 820  */
 821 static void dmsg_relay_scan_specific(h2span_node_t *node,
 822                                         h2span_conn_t *conn);
 823
 824 static void
 825 dmsg_relay_scan(h2span_conn_t *conn, h2span_node_t *node)
 826 {
 827         h2span_cluster_t *cls;
 828
 829         if (node) {
 830                 /*
 831                  * Iterate specific node
 832                  */
 833                 TAILQ_FOREACH(conn, &connq, entry)
 834                         dmsg_relay_scan_specific(node, conn);
 835         } else {
 836                 /*
 837                  * Full iteration.
 838                  *
 839                  * Iterate cluster ids, nodes, and either a specific connection
 840                  * or all connections.
 841                  */
 842                 RB_FOREACH(cls, h2span_cluster_tree, &cluster_tree) {
 843                         /*
 844                          * Iterate node ids
 845                          */
 846                         RB_FOREACH(node, h2span_node_tree, &cls->tree) {
 847                                 /*
 848                                  * Synchronize the node's link (received SPANs)
 849                                  * with each connection's relays.
 850                                  */
 851                                 if (conn) {
 852                                         dmsg_relay_scan_specific(node, conn);
 853                                 } else {
 854                                         TAILQ_FOREACH(conn, &connq, entry) {
 855                                             dmsg_relay_scan_specific(node,
 856                                                                         conn);
 857                                         }
 858                                         assert(conn == NULL);
 859                                 }
 860                         }
 861                 }
 862         }
 863 }
 864
 865 /*
 866  * Update the relay'd SPANs for this (node, conn).
 867  *
 868  * Iterate links and adjust relays to match.  We only propagate the top link
 869  * for now (XXX we want to propagate the top two).
 870  *
 871  * The dmsg_relay_scan_cmp() function locates the first relay element
 872  * for any given node.  The relay elements will be sub-sorted by dist.
 873  */
 874 struct relay_scan_info {
 875         h2span_node_t *node;
 876         h2span_relay_t *relay;
 877 };
 878
 879 static int
 880 dmsg_relay_scan_cmp(h2span_relay_t *relay, void *arg)
 881 {
 882         struct relay_scan_info *info = arg;
 883
 884         if ((intptr_t)relay->link->node < (intptr_t)info->node)
 885                 return(-1);
 886         if ((intptr_t)relay->link->node > (intptr_t)info->node)
 887                 return(1);
 888         return(0);
 889 }
 890
 891 static int
 892 dmsg_relay_scan_callback(h2span_relay_t *relay, void *arg)
 893 {
 894         struct relay_scan_info *info = arg;
 895
 896         info->relay = relay;
 897         return(-1);
 898 }
 899
 900 static void
 901 dmsg_relay_scan_specific(h2span_node_t *node, h2span_conn_t *conn)
 902 {
 903         struct relay_scan_info info;
 904         h2span_relay_t *relay;
 905         h2span_relay_t *next_relay;
 906         h2span_link_t *slink;
 907         dmsg_lnk_conn_t *lconn;
 908         dmsg_msg_t *msg;
 909         int count = 2;
 910         uint8_t peer_type;
 911
 912         info.node = node;
 913         info.relay = NULL;
 914
 915         /*
 916          * Locate the first related relay for the node on this connection.
 917          * relay will be NULL if there were none.
 918          */
 919         RB_SCAN(h2span_relay_tree, &conn->tree,
 920                 dmsg_relay_scan_cmp, dmsg_relay_scan_callback, &info);
 921         relay = info.relay;
 922         info.relay = NULL;
 923         if (relay)
 924                 assert(relay->link->node == node);
 925
 926         if (DMsgDebugOpt > 8)
 927                 fprintf(stderr, "relay scan for connection %p\n", conn);
 928
 929         /*
 930          * Iterate the node's links (received SPANs) in distance order,
 931          * lowest (best) dist first.
 932          *
 933          * PROPAGATE THE BEST LINKS OVER THE SPECIFIED CONNECTION.
 934          *
 935          * Track relays while iterating the best links and construct
 936          * missing relays when necessary.
 937          *
 938          * (If some prior better link was removed it would have also
 939          *  removed the relay, so the relay can only match exactly or
 940          *  be worse).
 941          */
 942         RB_FOREACH(slink, h2span_link_tree, &node->tree) {
 943                 /*
 944                  * Match, relay already in-place, get the next
 945                  * relay to match against the next slink.
 946                  */
 947                 if (relay && relay->link == slink) {
 948                         relay = RB_NEXT(h2span_relay_tree, &conn->tree, relay);
 949                         if (--count == 0)
 950                                 break;
 951                         continue;
 952                 }
 953
 954                 /*
 955                  * We might want this SLINK, if it passes our filters.
 956                  *
 957                  * The spanning tree can cause closed loops so we have
 958                  * to limit slink->dist.
 959                  */
 960                 if (slink->dist > DMSG_SPAN_MAXDIST)
 961                         break;
 962
 963                 /*
 964                  * Don't bother transmitting a LNK_SPAN out the same
 965                  * connection it came in on.  Trivial optimization.
 966                  */
 967                 if (slink->state->iocom == conn->state->iocom)
 968                         break;
 969
 970                 /*
 971                  * NOTE ON FILTERS: The protocol spec allows non-requested
 972                  * SPANs to be transmitted, the other end is expected to
 973                  * leave their transactions open but otherwise ignore them.
 974                  *
 975                  * Don't bother transmitting if the remote connection
 976                  * is not accepting this SPAN's peer_type.
 977                  */
 978                 peer_type = slink->state->msg->any.lnk_span.peer_type;
 979                 lconn = &conn->state->msg->any.lnk_conn;
 980                 if (((1LLU << peer_type) & lconn->peer_mask) == 0)
 981                         break;
 982
 983                 /*
 984                  * Filter based on pfs_clid or label (XXX).  This typically
 985                  * reduces the amount of SPAN traffic that a mount end-point
 986                  * sees by only passing along SPANs related to the cluster id
 987                  * (that is, it will see all PFS's associated with the
 988                  * particular cluster it represents).
 989                  */
 990                 if (peer_type == lconn->peer_type &&
 991                     peer_type == DMSG_PEER_HAMMER2) {
 992                         if (!uuid_is_nil(&slink->node->cls->pfs_clid, NULL) &&
 993                             uuid_compare(&slink->node->cls->pfs_clid,
 994                                          &lconn->pfs_clid, NULL) != 0) {
 995                                 break;
 996                         }
 997                 }
 998
 999                 /*
1000                  * Ok, we've accepted this SPAN for relaying.
1001                  */
1002                 assert(relay == NULL ||
1003                        relay->link->node != slink->node ||
1004                        relay->link->dist >= slink->dist);
1005                 relay = dmsg_alloc(sizeof(*relay));
1006                 relay->conn = conn;
1007                 relay->link = slink;
1008
1009                 msg = dmsg_msg_alloc(conn->state->iocom->router, 0,
1010                                         DMSG_LNK_SPAN |
1011                                         DMSGF_CREATE,
1012                                         dmsg_lnk_relay, relay);
1013                 relay->state = msg->state;
1014                 relay->router = dmsg_router_alloc();
1015                 relay->router->iocom = relay->state->iocom;
1016                 relay->router->relay = relay;
1017                 relay->router->target = relay->state->msgid;
1018
1019                 msg->any.lnk_span = slink->state->msg->any.lnk_span;
1020                 msg->any.lnk_span.dist = slink->dist + 1;
1021
1022                 dmsg_router_connect(relay->router);
1023
1024                 RB_INSERT(h2span_relay_tree, &conn->tree, relay);
1025                 TAILQ_INSERT_TAIL(&slink->relayq, relay, entry);
1026
1027                 dmsg_msg_write(msg);
1028
1029                 fprintf(stderr,
1030                         "RELAY SPAN %p RELAY %p ON CLS=%p NODE=%p DIST=%d "
1031                         "FD %d state %p\n",
1032                         slink,
1033                         relay,
1034                         node->cls, node, slink->dist,
1035                         conn->state->iocom->sock_fd, relay->state);
1036
1037                 /*
1038                  * Match (created new relay), get the next relay to
1039                  * match against the next slink.
1040                  */
1041                 relay = RB_NEXT(h2span_relay_tree, &conn->tree, relay);
1042                 if (--count == 0)
1043                         break;
1044         }
1045
1046         /*
1047          * Any remaining relay's belonging to this connection which match
1048          * the node are in excess of the current aggregate spanning state
1049          * and should be removed.
1050          */
1051         while (relay && relay->link->node == node) {
1052                 next_relay = RB_NEXT(h2span_relay_tree, &conn->tree, relay);
1053                 dmsg_relay_delete(relay);
1054                 relay = next_relay;
1055         }
1056 }
1057
1058 static
1059 void
1060 dmsg_relay_delete(h2span_relay_t *relay)
1061 {
1062         fprintf(stderr,
1063                 "RELAY DELETE %p RELAY %p ON CLS=%p NODE=%p DIST=%d FD %d STATE %p\n",
1064                 relay->link,
1065                 relay,
1066                 relay->link->node->cls, relay->link->node,
1067                 relay->link->dist,
1068                 relay->conn->state->iocom->sock_fd, relay->state);
1069
1070         dmsg_router_disconnect(&relay->router);
1071
1072         RB_REMOVE(h2span_relay_tree, &relay->conn->tree, relay);
1073         TAILQ_REMOVE(&relay->link->relayq, relay, entry);
1074
1075         if (relay->state) {
1076                 relay->state->any.relay = NULL;
1077                 dmsg_state_reply(relay->state, 0);
1078                 /* state invalid after reply */
1079                 relay->state = NULL;
1080         }
1081         relay->conn = NULL;
1082         relay->link = NULL;
1083         dmsg_free(relay);
1084 }
1085
1086 static void *
1087 dmsg_volconf_thread(void *info)
1088 {
1089         h2span_media_config_t *conf = info;
1090
1091         pthread_mutex_lock(&cluster_mtx);
1092         while ((conf->ctl & H2CONFCTL_STOP) == 0) {
1093                 if (conf->ctl & H2CONFCTL_UPDATE) {
1094                         fprintf(stderr, "VOLCONF UPDATE\n");
1095                         conf->ctl &= ~H2CONFCTL_UPDATE;
1096                         if (bcmp(&conf->copy_run, &conf->copy_pend,
1097                                  sizeof(conf->copy_run)) == 0) {
1098                                 fprintf(stderr, "VOLCONF: no changes\n");
1099                                 continue;
1100                         }
1101                         /*
1102                          * XXX TODO - auto reconnect on lookup failure or
1103                          *              connect failure or stream failure.
1104                          */
1105
1106                         pthread_mutex_unlock(&cluster_mtx);
1107                         dmsg_volconf_stop(conf);
1108                         conf->copy_run = conf->copy_pend;
1109                         if (conf->copy_run.copyid != 0 &&
1110                             strncmp(conf->copy_run.path, "span:", 5) == 0) {
1111                                 dmsg_volconf_start(conf,
1112                                                       conf->copy_run.path + 5);
1113                         }
1114                         pthread_mutex_lock(&cluster_mtx);
1115                         fprintf(stderr, "VOLCONF UPDATE DONE state %d\n", conf->state);
1116                 }
1117                 if (conf->state == H2MC_CONNECT) {
1118                         dmsg_volconf_start(conf, conf->copy_run.path + 5);
1119                         pthread_mutex_unlock(&cluster_mtx);
1120                         sleep(5);
1121                         pthread_mutex_lock(&cluster_mtx);
1122                 } else {
1123                         pthread_cond_wait(&conf->cond, &cluster_mtx);
1124                 }
1125         }
1126         pthread_mutex_unlock(&cluster_mtx);
1127         dmsg_volconf_stop(conf);
1128         return(NULL);
1129 }
1130
1131 static
1132 void
1133 dmsg_volconf_stop(h2span_media_config_t *conf)
1134 {
1135         switch(conf->state) {
1136         case H2MC_STOPPED:
1137                 break;
1138         case H2MC_CONNECT:
1139                 conf->state = H2MC_STOPPED;
1140                 break;
1141         case H2MC_RUNNING:
1142                 shutdown(conf->fd, SHUT_WR);
1143                 pthread_join(conf->iocom_thread, NULL);
1144                 conf->iocom_thread = NULL;
1145                 break;
1146         }
1147 }
1148
1149 static
1150 void
1151 dmsg_volconf_start(h2span_media_config_t *conf, const char *hostname)
1152 {
1153         dmsg_master_service_info_t *info;
1154
1155         switch(conf->state) {
1156         case H2MC_STOPPED:
1157         case H2MC_CONNECT:
1158                 conf->fd = dmsg_connect(hostname);
1159                 if (conf->fd < 0) {
1160                         fprintf(stderr, "Unable to connect to %s\n", hostname);
1161                         conf->state = H2MC_CONNECT;
1162                 } else {
1163                         info = malloc(sizeof(*info));
1164                         bzero(info, sizeof(*info));
1165                         info->fd = conf->fd;
1166                         info->detachme = 0;
1167                         conf->state = H2MC_RUNNING;
1168                         pthread_create(&conf->iocom_thread, NULL,
1169                                        dmsg_master_service, info);
1170                 }
1171                 break;
1172         case H2MC_RUNNING:
1173                 break;
1174         }
1175 }
1176
1177 /************************************************************************
1178  *                      ROUTER AND MESSAGING HANDLES                    *
1179  ************************************************************************
1180  *
1181  * Basically the idea here is to provide a stable data structure which
1182  * can be localized to the caller for higher level protocols to work with.
1183  * Depends on the context, these dmsg_handle's can be pooled by use-case
1184  * and remain persistent through a client (or mount point's) life.
1185  */
1186
1187 #if 0
1188 /*
1189  * Obtain a stable handle on a cluster given its uuid.  This ties directly
1190  * into the global cluster topology, creating the structure if necessary
1191  * (even if the uuid does not exist or does not exist yet), and preventing
1192  * the structure from getting ripped out from under us while we hold a
1193  * pointer to it.
1194  */
1195 h2span_cluster_t *
1196 dmsg_cluster_get(uuid_t *pfs_clid)
1197 {
1198         h2span_cluster_t dummy_cls;
1199         h2span_cluster_t *cls;
1200
1201         dummy_cls.pfs_clid = *pfs_clid;
1202         pthread_mutex_lock(&cluster_mtx);
1203         cls = RB_FIND(h2span_cluster_tree, &cluster_tree, &dummy_cls);
1204         if (cls)
1205                 ++cls->refs;
1206         pthread_mutex_unlock(&cluster_mtx);
1207         return (cls);
1208 }
1209
1210 void
1211 dmsg_cluster_put(h2span_cluster_t *cls)
1212 {
1213         pthread_mutex_lock(&cluster_mtx);
1214         assert(cls->refs > 0);
1215         --cls->refs;
1216         if (RB_EMPTY(&cls->tree) && cls->refs == 0) {
1217                 RB_REMOVE(h2span_cluster_tree,
1218                           &cluster_tree, cls);
1219                 dmsg_free(cls);
1220         }
1221         pthread_mutex_unlock(&cluster_mtx);
1222 }
1223
1224 /*
1225  * Obtain a stable handle to a specific cluster node given its uuid.
1226  * This handle does NOT lock in the route to the node and is typically
1227  * used as part of the dmsg_handle_*() API to obtain a set of
1228  * stable nodes.
1229  */
1230 h2span_node_t *
1231 dmsg_node_get(h2span_cluster_t *cls, uuid_t *pfs_fsid)
1232 {
1233 }
1234
1235 #endif
1236
1237 #if 0
1238 /*
1239  * Acquire a persistent router structure given the cluster and node ids.
1240  * Messages can be transacted via this structure while held.  If the route
1241  * is lost messages will return failure.
1242  */
1243 dmsg_router_t *
1244 dmsg_router_get(uuid_t *pfs_clid, uuid_t *pfs_fsid)
1245 {
1246 }
1247
1248 /*
1249  * Release previously acquired router.
1250  */
1251 void
1252 dmsg_router_put(dmsg_router_t *router)
1253 {
1254 }
1255 #endif
1256
1257 /*
1258  * Dumps the spanning tree
1259  */
1260 void
1261 dmsg_shell_tree(dmsg_router_t *router, char *cmdbuf __unused)
1262 {
1263         h2span_cluster_t *cls;
1264         h2span_node_t *node;
1265         h2span_link_t *slink;
1266         char *uustr = NULL;
1267
1268         pthread_mutex_lock(&cluster_mtx);
1269         RB_FOREACH(cls, h2span_cluster_tree, &cluster_tree) {
1270                 dmsg_router_printf(router, "Cluster %s\n",
1271                                    dmsg_uuid_to_str(&cls->pfs_clid, &uustr));
1272                 RB_FOREACH(node, h2span_node_tree, &cls->tree) {
1273                         dmsg_router_printf(router, "    Node %s (%s)\n",
1274                                 dmsg_uuid_to_str(&node->pfs_fsid, &uustr),
1275                                 node->label);
1276                         RB_FOREACH(slink, h2span_link_tree, &node->tree) {
1277                                 dmsg_router_printf(router,
1278                                             "\tLink dist=%d via %d\n",
1279                                             slink->dist,
1280                                             slink->state->iocom->sock_fd);
1281                         }
1282                 }
1283         }
1284         pthread_mutex_unlock(&cluster_mtx);
1285         if (uustr)
1286                 free(uustr);
1287 #if 0
1288         TAILQ_FOREACH(conn, &connq, entry) {
1289         }
1290 #endif
1291 }