1 /*
   2  * This file and its contents are supplied under the terms of the
   3  * Common Development and Distribution License ("CDDL"), version 1.0.
   4  * You may only use this file in accordance with the terms of version
   5  * 1.0 of the CDDL.
   6  *
   7  * A full copy of the text of the CDDL should have accompanied this
   8  * source.  A copy of the CDDL is also available via the Internet at
   9  * http://www.illumos.org/license/CDDL.
  10  */
  11 
  12 /*
  13  * Copyright 2018, Joyent, Inc.
  14  */
  15 
  16 /*
  17  * This plugin implements the SDC VXLAN Protocol (SVP).
  18  *
  19  * This plugin is designed to work with a broader distributed system that
  20  * mainains a database of mappings and provides a means of looking up data and
  21  * provides a stream of updates. While it is named after VXLAN, there isn't
  22  * anything specific to VXLAN baked into the protocol at this time, other than
  23  * that it requires both an IP address and a port; however, if there's a good
  24  * reason to support others here, we can modify that.
  25  *
  26  * -----------
  27  * Terminology
  28  * -----------
  29  *
  30  * Throughout this module we refer to a few different kinds of addresses:
  31  *
  32  *    VL3
  33  *
  34  *      A VL3 address, or virtual layer 3, refers to the layer three addreses
  35  *      that are used by entities on an overlay network. As far as we're
  36  *      concerned that means that this is the IP address of an interface on an
  37  *      overlay network.
  38  *
  39  *    VL2
  40  *
  41  *      A VL2 address, or a virtual layer 2, referes to the link-layer addresses
  42  *      that are used by entities on an overlay network. As far as we're
  43  *      concerned that means that this is the MAC addresses of an interface on
  44  *      an overlay network.
  45  *
  46  *    UL3
  47  *
  48  *      A UL3, or underlay layer 3, refers to the layer three (IP) address on
  49  *      the underlay network.
  50  *
  51  * The svp plugin provides lookups from VL3->VL2, eg. the equivalent of an ARP
  52  * or NDP query, and then also provides VL2->UL3 lookups.
  53  *
  54  * -------------------
  55  * Protocol Operations
  56  * -------------------
  57  *
  58  * The svp protocol is defined in lib/varpd/svp/common/libvarpd_svp_prot.h. It
  59  * defines the basic TCP protocol that we use to communicate to hosts. At this
  60  * time, it is not quite 100% implemented in both this plug-in and our primary
  61  * server, sdc-portolan (see https://github.com/joyent/sdc-portolan).
  62  *
  63  * At this time, we don't quite support everything that we need to. Including
  64  * the SVP_R_BULK_REQ and SVP_R_SHOOTDOWN.
  65  *
  66  * ---------------------------------
  67  * General Design and Considerations
  68  * ---------------------------------
  69  *
  70  * Every instance of the svp plugin requires the hostname and port of a server
  71  * to contact. Though, we have co-opted the port 1296 (the year of the oldest
  72  * extant portolan) as our default port.
  73  *
  74  * Each of the different instance of the plugins has a corresponding remote
  75  * backend. The remote backend represents the tuple of the [ host, port ].
  76  * Different instances that share the same host and port tuple will use the same
  77  * backend.
  78  *
  79  * The backend is actually in charge of performing lookups, resolving and
  80  * updating the set of remote hosts based on the DNS resolution we've been
  81  * provided, and taking care of things like shootdowns.
  82  *
  83  * The whole plugin itself maintains an event loop and a number of threads to
  84  * service that event loop. On top of that event loop, we have a simple timer
  85  * backend that ticks at one second intervals and performs various callbacks,
  86  * such as idle query timers, DNS resolution, connection backoff, etc. Each of
  87  * the remote hosts that we obtain is wrapped up in an svp_conn_t, which manages
  88  * the connection state, reconnecting, etc.
  89  *
  90  * All in all, the general way that this all looks like is:
  91  *
  92  *  +----------------------------+
  93  *  | Plugin Instance            |
  94  *  | svp_t                      |
  95  *  |                            |
  96  *  | varpd_provider_handle_t * -+-> varpd handle
  97  *  | uint64_t               ----+-> varpd ID
  98  *  | char *                 ----+-> remote host
  99  *  | uint16_t               ----+-> remote port
 100  *  | svp_remote_t *   ---+------+-> remote backend
 101  *  +---------------------+------+
 102  *                        |
 103  *                        v
 104  *   +----------------------+                   +----------------+
 105  *   | Remote backend       |------------------>| Remove Backend |---> ...
 106  *   | svp_remote_t         |                   | svp_remote_t   |
 107  *   |                      |                   +----------------+
 108  *   | svp_remote_state_t --+-> state flags
 109  *   | svp_degrade_state_t -+-> degraded reason
 110  *   | struct addrinfo *  --+-> resolved hosts
 111  *   | uint_t            ---+-> active hosts
 112  *   | uint_t            ---+-> DNS generation
 113  *   | uint_t            ---+-> Reference count
 114  *   | uint_t            ---+-> active conns
 115  *   | uint_t            ---+-> degraded conns
 116  *   | list_t        ---+---+-> connection list
 117  *   +------------------+---+
 118  *                      |
 119  *                      +------------------------------+-----------------+
 120  *                      |                              |                 |
 121  *                      v                              v                 v
 122  *   +-------------------+                       +----------------
 123  *   | SVP Connection    |                       | SVP connection |     ...
 124  *   | svp_conn_t        |                       | svp_conn_t     |
 125  *   |                   |                       +----------------+
 126  *   | svp_event_t   ----+-> event loop handle
 127  *   | svp_timer_t   ----+-> backoff timer
 128  *   | svp_timer_t   ----+-> query timer
 129  *   | int           ----+-> socket fd
 130  *   | uint_t        ----+-> generation
 131  *   | uint_t        ----+-> current backoff
 132  *   | svp_conn_flags_t -+-> connection flags
 133  *   | svp_conn_state_t -+-> connection state
 134  *   | svp_conn_error_t -+-> connection error
 135  *   | int            ---+-> last errrno
 136  *   | hrtime_t       ---+-> activity timestamp
 137  *   | svp_conn_out_t ---+-> outgoing data state
 138  *   | svp_conn_in_t  ---+-> incoming data state
 139  *   | list_t      ---+--+-> active queries
 140  *   +----------------+--+
 141  *                    |
 142  *                    +----------------------------------+-----------------+
 143  *                    |                                  |                 |
 144  *                    v                                  v                 v
 145  *   +--------------------+                       +-------------+
 146  *   | SVP Query          |                       | SVP Query   |         ...
 147  *   | svp_query_t        |                       | svp_query_t |
 148  *   |                    |                       +-------------+
 149  *   | svp_query_f     ---+-> callback function
 150  *   | void *          ---+-> callback arg
 151  *   | svp_query_state_t -+-> state flags
 152  *   | svp_req_t       ---+-> svp prot. header
 153  *   | svp_query_data_t --+-> read data
 154  *   | svp_query_data_t --+-> write data
 155  *   | svp_status_t    ---+-> request status
 156  *   +--------------------+
 157  *
 158  * The svp_t is the instance that we assoicate with varpd. The instance itself
 159  * maintains properties and then when it's started associates with an
 160  * svp_remote_t, which is the remote backend. The remote backend itself,
 161  * maintains the DNS state and spins up and downs connections based on the
 162  * results from DNS. By default, we query DNS every 30 seconds. For more on the
 163  * connection life cycle, see the next section.
 164  *
 165  * By default, each connection maintains its own back off timer and list of
 166  * queries it's servicing. Only one request is generally outstanding at a time
 167  * and requests are round robined across the various connections.
 168  *
 169  * The query itself represents the svp request that's going on and keep track of
 170  * its state and is a place for data that's read and written to as part of the
 171  * request.
 172  *
 173  * Connections maintain a query timer such that if we have not received data on
 174  * a socket for a certain amount of time, we kill that socket and begin a
 175  * reconnection cycle with backoff.
 176  *
 177  * ------------------------
 178  * Connection State Machine
 179  * ------------------------
 180  *
 181  * We have a connection pool that's built upon DNS records. DNS describes the
 182  * membership of the set of remote peers that make up our pool and we maintain
 183  * one connection to each of them.  In addition, we maintain an exponential
 184  * backoff for each peer and will attempt to reconect immediately before backing
 185  * off. The following are the valid states that a connection can be in:
 186  *
 187  *      SVP_CS_ERROR            An OS error has occurred on this connection,
 188  *                              such as failure to create a socket or associate
 189  *                              the socket with an event port. We also
 190  *                              transition all connections to this state before
 191  *                              we destroy them.
 192  *
 193  *      SVP_CS_INITIAL          This is the initial state of a connection, all
 194  *                              that should exist is an unbound socket.
 195  *
 196  *      SVP_CS_CONNECTING       A call to connect has been made and we are
 197  *                              polling for it to complete.
 198  *
 199  *      SVP_CS_BACKOFF          A connect attempt has failed and we are
 200  *                              currently backing off, waiting to try again.
 201  *
 202  *      SVP_CS_ACTIVE           We have successfully connected to the remote
 203  *                              system.
 204  *
 205  *      SVP_CS_WINDDOWN         This connection is going to valhalla. In other
 206  *                              words, a previously active connection is no
 207  *                              longer valid in DNS, so we should curb our use
 208  *                              of it, and reap it as soon as we have other
 209  *                              active connections.
 210  *
 211  * The following diagram attempts to describe our state transition scheme, and
 212  * when we transition from one state to the next.
 213  *
 214  *                               |
 215  *                               * New remote IP from DNS resolution,
 216  *                               | not currently active in the system.
 217  *                               |
 218  *                               v                                Socket Error,
 219  *                       +----------------+                       still in DNS
 220  *  +----------------<---| SVP_CS_INITIAL |<----------------------*--------+
 221  *  |                    +----------------+                                |
 222  *  |                            System  |                                 |
 223  *  | Connection . . . . .       success *     Successful                  |
 224  *  | failed             .               |     connect()                   |
 225  *  |                    .               |        +-------------------+    |
 226  *  |               +----*---------+     |    +-*>| SVP_CS_VERSIONING +    |
 227  *  |               |              |     |    |   +-------------------+    |
 228  *  |               |              |     |    |          V   V Set version |
 229  *  |               |              |     |    |          |   * based on    |
 230  *  |               |              |     |    |          |   | SVP_R_PONG  |
 231  *  |               V              ^     v    ^          |   V             ^
 232  *  |  +----------------+         +-------------------+  |  +---------------+
 233  *  +<-| SVP_CS_BACKOFF |         | SVP_CS_CONNECTING |  |  | SVP_CS_ACTIVE |
 234  *  |  +----------------+         +-------------------+  |  +---------------+
 235  *  |               V              ^  V                  |    V  V
 236  *  | Backoff wait  *              |  |                  |    |  * Removed
 237  *  v interval      +--------------+  +-----------------<+----+  | from DNS
 238  *  | finished                        |                          |
 239  *  |                                 V                          |
 240  *  |                                 |                          V
 241  *  |                                 |            +-----------------+
 242  *  +----------------+----------<-----+-------<----| SVP_CS_WINDDOWN |
 243  *                   |                             +-----------------+
 244  *                   * . . .   Fatal system, not
 245  *                   |         socket error or
 246  *                   V         quiesced after
 247  *           +--------------+  removal from DNS
 248  *           | SVP_CS_ERROR |
 249  *           +--------------+
 250  *                   |
 251  *                   * . . . Removed from DNS
 252  *                   v
 253  *            +------------+
 254  *            | Connection |
 255  *            | Destroyed  |
 256  *            +------------+
 257  *
 258  * --------------------------
 259  * Connection Event Injection
 260  * --------------------------
 261  *
 262  * For each connection that exists in the system, we have a timer in place that
 263  * is in charge of performing timeout activity. It fires once every thirty
 264  * seconds or so for a given connection and checks to ensure that we have had
 265  * activity for the most recent query on the connection. If not, it terminates
 266  * the connection. This is important as if we have sent all our data and are
 267  * waiting for the remote end to reply, without enabling something like TCP
 268  * keep-alive, we will not be notified that anything that has happened to the
 269  * remote connection, for example a panic. In addition, this also protects
 270  * against a server that is up, but a portolan that is not making forward
 271  * progress.
 272  *
 273  * When a timeout occurs, we first try to disassociate any active events, which
 274  * by definition must exist. Once that's done, we inject a port source user
 275  * event. Now, there is a small gotcha. Let's assume for a moment that we have a
 276  * pathological portolan. That means that it knows to inject activity right at
 277  * the time out window. That means, that the event may be disassociated before
 278  * we could get to it. If that's the case, we must _not_ inject the user event
 279  * and instead, we'll let the pending event take care of it. We know that the
 280  * pending event hasn't hit the main part of the loop yet, otherwise, it would
 281  * have released the lock protecting our state and associated the event.
 282  *
 283  * ------------
 284  * Notes on DNS
 285  * ------------
 286  *
 287  * Unfortunately, doing host name resolution in a way that allows us to leverage
 288  * the system's resolvers and the system's caching, require us to make blocking
 289  * calls in libc via getaddrinfo(3SOCKET). If we can't reach a given server,
 290  * that will tie up a thread for quite some time. To work around that fact,
 291  * we're going to create a fixed number of threads and we'll use them to service
 292  * our DNS requests. While this isn't ideal, until we have a sane means of
 293  * integrating a DNS resolution into an event loop with say portfs, it's not
 294  * going to be a fun day no matter what we do.
 295  *
 296  * ------
 297  * Timers
 298  * ------
 299  *
 300  * We maintain a single timer based on CLOCK_REALTIME. It's designed to fire
 301  * every second. While we'd rather use CLOCK_HIGHRES just to alleviate ourselves
 302  * from timer drift; however, as zones may not actually have CLOCK_HIGHRES
 303  * access, we don't want them to end up in there. The timer itself is just a
 304  * simple avl tree sorted by expiration time, which is stored as a tick in the
 305  * future, a tick is just one second.
 306  *
 307  * ----------
 308  * Shootdowns
 309  * ----------
 310  *
 311  * As part of the protocol, we need to be able to handle shootdowns that inform
 312  * us some of the information in the system is out of date. This information
 313  * needs to be processed promptly; however, the information is hopefully going
 314  * to be relatively infrequent relative to the normal flow of information.
 315  *
 316  * The shoot down information needs to be done on a per-backend basis. The
 317  * general design is that we'll have a single query for this which can fire on a
 318  * 5-10s period, we randmoize the latter part to give us a bit more load
 319  * spreading. If we complete because there's no work to do, then we wait the
 320  * normal period. If we complete, but there's still work to do, we'll go again
 321  * after a second.
 322  *
 323  * A shootdown has a few different parts. We first receive a list of items to
 324  * shootdown. After performing all of those, we need to acknowledge them. When
 325  * that's been done successfully, we can move onto the next part. From a
 326  * protocol perspective, we make a SVP_R_LOG_REQ, we get a reply, and then after
 327  * processing them, send an SVP_R_LOG_RM. Only once that's been acked do we
 328  * continue.
 329  *
 330  * However, one of the challenges that we have is that these invalidations are
 331  * just that, an invalidation. For a virtual layer two request, that's fine,
 332  * because the kernel supports that. However, for virtual layer three
 333  * invalidations, we have a bit more work to do. These protocols, ARP and NDP,
 334  * don't really support a notion of just an invalidation, instead you have to
 335  * inject the new data in a gratuitous fashion.
 336  *
 337  * To that end, what we instead do is when we receive a VL3 invalidation, we
 338  * turn that info a VL3 request. We hold the general request as outstanding
 339  * until we receive all of the callbacks for the VL3 invalidations, at which
 340  * point we go through and do the log removal request.
 341  */
 342 
 343 #include <umem.h>
 344 #include <errno.h>
 345 #include <stdlib.h>
 346 #include <sys/types.h>
 347 #include <sys/socket.h>
 348 #include <netinet/in.h>
 349 #include <arpa/inet.h>
 350 #include <libnvpair.h>
 351 #include <strings.h>
 352 #include <string.h>
 353 #include <assert.h>
 354 #include <unistd.h>
 355 
 356 #include <libvarpd_provider.h>
 357 #include "libvarpd_svp.h"
 358 
 359 bunyan_logger_t *svp_bunyan;
 360 static int svp_defport = 1296;
 361 static int svp_defuport = 1339;
 362 static umem_cache_t *svp_lookup_cache;
 363 
 364 typedef enum svp_lookup_type {
 365         SVP_L_UNKNOWN   = 0x0,
 366         SVP_L_VL2       = 0x1,
 367         SVP_L_VL3       = 0x2,
 368         SVP_L_ROUTE     = 0x3
 369 } svp_lookup_type_t;
 370 
 371 typedef struct svp_lookup {
 372         int svl_type;
 373         union {
 374                 struct svl_lookup_vl2 {
 375                         varpd_query_handle_t    *svl_handle;
 376                         overlay_target_point_t  *svl_point;
 377                 } svl_vl2;
 378                 struct svl_lookup_vl3 {
 379                         varpd_arp_handle_t      *svl_vah;
 380                         uint8_t                 *svl_out;
 381                 } svl_vl3;
 382                 struct svl_lookup_route {
 383                         varpd_query_handle_t    *svl_handle;
 384                         overlay_target_point_t  *svl_point;
 385                         overlay_target_route_t  *svl_route;
 386                         overlay_target_mac_t    *svl_mac;
 387                 } svl_route;
 388         } svl_u;
 389         svp_query_t                             svl_query;
 390 } svp_lookup_t;
 391 
 392 static const char *varpd_svp_props[] = {
 393         "svp/host",
 394         "svp/port",
 395         "svp/underlay_ip",
 396         "svp/underlay_port",
 397         "svp/dcid",
 398         "svp/router_oui"
 399 };
 400 
 401 static const uint8_t svp_bcast[6] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
 402 
 403 int
 404 svp_comparator(const void *l, const void *r)
 405 {
 406         const svp_t *ls = l;
 407         const svp_t *rs = r;
 408 
 409         if (ls->svp_vid > rs->svp_vid)
 410                 return (1);
 411         if (ls->svp_vid < rs->svp_vid)
 412                 return (-1);
 413         return (0);
 414 }
 415 
 416 static void
 417 svp_vl2_lookup_cb(svp_t *svp, svp_status_t status, const struct in6_addr *uip,
 418     const uint16_t uport, void *arg)
 419 {
 420         svp_lookup_t *svl = arg;
 421         overlay_target_point_t *otp;
 422 
 423         assert(svp != NULL);
 424         assert(arg != NULL);
 425 
 426         if (status != SVP_S_OK) {
 427                 libvarpd_plugin_query_reply(svl->svl_u.svl_vl2.svl_handle,
 428                     VARPD_LOOKUP_DROP);
 429                 umem_cache_free(svp_lookup_cache, svl);
 430                 return;
 431         }
 432 
 433         otp = svl->svl_u.svl_vl2.svl_point;
 434         bcopy(uip, &otp->otp_ip, sizeof (struct in6_addr));
 435         otp->otp_port = uport;
 436         libvarpd_plugin_query_reply(svl->svl_u.svl_vl2.svl_handle,
 437             VARPD_LOOKUP_OK);
 438         umem_cache_free(svp_lookup_cache, svl);
 439 }
 440 
 441 static void
 442 svp_vl3_lookup_cb(svp_t *svp, svp_status_t status, const uint8_t *vl2mac,
 443     const struct in6_addr *uip, const uint16_t uport, void *arg)
 444 {
 445         /* Initialize address-holders to 0 for comparisons-to-zeroes later. */
 446         overlay_target_point_t point = { 0 };
 447         svp_lookup_t *svl = arg;
 448         uint8_t nexthop_mac[6] = { 0, 0, 0, 0, 0, 0 };
 449 
 450         assert(svp != NULL);
 451         assert(svl != NULL);
 452 
 453         if (status != SVP_S_OK) {
 454                 libvarpd_plugin_arp_reply(svl->svl_u.svl_vl3.svl_vah,
 455                     VARPD_LOOKUP_DROP);
 456                 umem_cache_free(svp_lookup_cache, svl);
 457                 return;
 458         }
 459 
 460         /* Inject the L2 mapping before the L3 */
 461         bcopy(uip, &point.otp_ip, sizeof (struct in6_addr));
 462         point.otp_port = uport;
 463         libvarpd_inject_varp(svp->svp_hdl, vl2mac, &point);
 464 
 465         bcopy(vl2mac, svl->svl_u.svl_vl3.svl_out, ETHERADDRL);
 466         libvarpd_plugin_arp_reply(svl->svl_u.svl_vl3.svl_vah,
 467             VARPD_LOOKUP_OK);
 468         umem_cache_free(svp_lookup_cache, svl);
 469 }
 470 
 471 static void
 472 svp_vl2_invalidate_cb(svp_t *svp, const uint8_t *vl2mac)
 473 {
 474         libvarpd_inject_varp(svp->svp_hdl, vl2mac, NULL);
 475 }
 476 
 477 static void
 478 svp_vl3_inject_cb(svp_t *svp, const uint16_t vlan, const struct in6_addr *vl3ip,
 479     const uint8_t *vl2mac, const uint8_t *targmac)
 480 {
 481         struct in_addr v4;
 482 
 483         /*
 484          * At the moment we don't support any IPv6 related log entries, this
 485          * will change soon as we develop a bit more of the IPv6 related
 486          * infrastructure so we can properly test the injection.
 487          */
 488         if (IN6_IS_ADDR_V4MAPPED(vl3ip) == 0) {
 489                 return;
 490         } else {
 491                 IN6_V4MAPPED_TO_INADDR(vl3ip, &v4);
 492                 if (targmac == NULL)
 493                         targmac = svp_bcast;
 494                 libvarpd_inject_arp(svp->svp_hdl, vlan, vl2mac, &v4, targmac);
 495         }
 496 }
 497 
 498 /* ARGSUSED */
 499 static void
 500 svp_shootdown_cb(svp_t *svp, const uint8_t *vl2mac, const struct in6_addr *uip,
 501     const uint16_t uport)
 502 {
 503         /*
 504          * We should probably do a conditional invalidation here.
 505          */
 506         libvarpd_inject_varp(svp->svp_hdl, vl2mac, NULL);
 507 }
 508 
 509 static void
 510 svp_route_lookup_cb(svp_t *svp, svp_status_t status, uint32_t dcid,
 511     uint32_t vnetid, uint16_t vlan, uint8_t *srcmac, uint8_t *dstmac,
 512     uint16_t ul3_port, uint8_t *ul3_addr, uint8_t srcpfx, uint8_t dstpfx,
 513     void *arg)
 514 {
 515         svp_lookup_t *svl = arg;
 516         overlay_target_point_t *otp;
 517         overlay_target_route_t *otr;
 518         overlay_target_mac_t *otm;
 519 
 520         if (status != SVP_S_OK) {
 521                 libvarpd_plugin_query_reply(svl->svl_u.svl_route.svl_handle,
 522                     VARPD_LOOKUP_DROP);
 523                 umem_cache_free(svp_lookup_cache, svl);
 524                 return;
 525         }
 526 
 527         otp = svl->svl_u.svl_route.svl_point;
 528         bcopy(ul3_addr, &otp->otp_ip, sizeof (struct in6_addr));
 529         otp->otp_port = ul3_port;
 530 
 531         otr = svl->svl_u.svl_route.svl_route;
 532         otr->otr_vnet = vnetid;
 533         otr->otr_vlan = vlan;
 534         bcopy(srcmac, otr->otr_srcmac, ETHERADDRL);
 535 
 536         otm = svl->svl_u.svl_route.svl_mac;
 537         otm->otm_dcid = dcid;
 538         bcopy(dstmac, otm->otm_mac, ETHERADDRL);
 539 
 540         libvarpd_plugin_query_reply(svl->svl_u.svl_route.svl_handle,
 541             VARPD_LOOKUP_OK);
 542         umem_cache_free(svp_lookup_cache, svl);
 543 }
 544 
 545 static svp_cb_t svp_defops = {
 546         svp_vl2_lookup_cb,
 547         svp_vl3_lookup_cb,
 548         svp_vl2_invalidate_cb,
 549         svp_vl3_inject_cb,
 550         svp_shootdown_cb,
 551         svp_route_lookup_cb,
 552 };
 553 
 554 static boolean_t
 555 varpd_svp_valid_dest(overlay_plugin_dest_t dest)
 556 {
 557         if (dest != (OVERLAY_PLUGIN_D_IP | OVERLAY_PLUGIN_D_PORT))
 558                 return (B_FALSE);
 559 
 560         return (B_TRUE);
 561 }
 562 
 563 static int
 564 varpd_svp_create(varpd_provider_handle_t *hdl, void **outp,
 565     overlay_plugin_dest_t dest)
 566 {
 567         int ret;
 568         svp_t *svp;
 569 
 570         if (varpd_svp_valid_dest(dest) == B_FALSE)
 571                 return (ENOTSUP);
 572 
 573         svp = umem_zalloc(sizeof (svp_t), UMEM_DEFAULT);
 574         if (svp == NULL)
 575                 return (ENOMEM);
 576 
 577         if ((ret = mutex_init(&svp->svp_lock, USYNC_THREAD | LOCK_ERRORCHECK,
 578             NULL)) != 0) {
 579                 umem_free(svp, sizeof (svp_t));
 580                 return (ret);
 581         }
 582 
 583         svp->svp_port = svp_defport;
 584         svp->svp_uport = svp_defuport;
 585         svp->svp_cb = svp_defops;
 586         svp->svp_hdl = hdl;
 587         svp->svp_vid = libvarpd_plugin_vnetid(svp->svp_hdl);
 588         *outp = svp;
 589         return (0);
 590 }
 591 
 592 static int
 593 varpd_svp_start(void *arg)
 594 {
 595         int ret;
 596         svp_remote_t *srp;
 597         svp_t *svp = arg;
 598 
 599         mutex_enter(&svp->svp_lock);
 600         if (svp->svp_host == NULL || svp->svp_port == 0 ||
 601             svp->svp_huip == B_FALSE || svp->svp_uport == 0) {
 602                 mutex_exit(&svp->svp_lock);
 603                 return (EAGAIN);
 604         }
 605         mutex_exit(&svp->svp_lock);
 606 
 607         if ((ret = svp_remote_find(svp->svp_host, svp->svp_port, &svp->svp_uip,
 608             &srp)) != 0)
 609                 return (ret);
 610 
 611         if ((ret = svp_remote_attach(srp, svp)) != 0) {
 612                 svp_remote_release(srp);
 613                 return (ret);
 614         }
 615 
 616         return (0);
 617 }
 618 
 619 static void
 620 varpd_svp_stop(void *arg)
 621 {
 622         svp_t *svp = arg;
 623 
 624         svp_remote_detach(svp);
 625 }
 626 
 627 static void
 628 varpd_svp_destroy(void *arg)
 629 {
 630         svp_t *svp = arg;
 631 
 632         if (svp->svp_host != NULL)
 633                 umem_free(svp->svp_host, strlen(svp->svp_host) + 1);
 634 
 635         if (mutex_destroy(&svp->svp_lock) != 0)
 636                 libvarpd_panic("failed to destroy svp_t`svp_lock");
 637 
 638         umem_free(svp, sizeof (svp_t));
 639 }
 640 
 641 static void
 642 varpd_svp_lookup_l3(svp_t *svp, varpd_query_handle_t *vqh,
 643     const overlay_targ_lookup_t *otl, overlay_target_point_t *otp,
 644     overlay_target_route_t *otr, overlay_target_mac_t *otm)
 645 {
 646         svp_lookup_t *slp;
 647         uint32_t type;
 648         const struct in6_addr *src = &otl->otl_addru.otlu_l3.otl3_srcip,
 649             *dst = &otl->otl_addru.otlu_l3.otl3_dstip;
 650 
 651         /*
 652          * otl is an L3 request, so we have src/dst IPs for the inner packet.
 653          * We also have the vlan.
 654          *
 655          * Assume kernel's overlay module is caching well, so we are directly
 656          * going to query (i.e. no caching up here of actual destinations).
 657          *
 658          * Our existing remote sever (svp_remote), but with the new message
 659          * SVP_R_ROUTE_REQ.
 660          */
 661 
 662         if (IN6_IS_ADDR_V4MAPPED(src)) {
 663                 if (!IN6_IS_ADDR_V4MAPPED(dst)) {
 664                         libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
 665                         return;
 666                 }
 667                 type = SVP_VL3_IP;
 668         } else {
 669                 if (IN6_IS_ADDR_V4MAPPED(dst)) {
 670                         libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
 671                         return;
 672                 }
 673                 type = SVP_VL3_IPV6;
 674         }
 675 
 676         slp = umem_cache_alloc(svp_lookup_cache, UMEM_DEFAULT);
 677         if (slp == NULL) {
 678                 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
 679                 return;
 680         }
 681 
 682         slp->svl_type = SVP_L_ROUTE;
 683         slp->svl_u.svl_route.svl_handle = vqh;
 684         slp->svl_u.svl_route.svl_point = otp;
 685         slp->svl_u.svl_route.svl_route = otr;
 686         slp->svl_u.svl_route.svl_mac = otm;
 687 
 688         svp_remote_route_lookup(svp, &slp->svl_query, src, dst,
 689             otl->otl_vnetid, (uint16_t)otl->otl_vlan, slp);
 690 }
 691 
 692 static void
 693 varpd_svp_lookup(void *arg, varpd_query_handle_t *vqh,
 694     const overlay_targ_lookup_t *otl, overlay_target_point_t *otp,
 695     overlay_target_route_t *otr, overlay_target_mac_t *otm)
 696 {
 697         svp_lookup_t *slp;
 698         svp_t *svp = arg;
 699 
 700         /*
 701          * Shuffle off L3 lookups to their own codepath.
 702          */
 703         if (otl->otl_l3req) {
 704                 varpd_svp_lookup_l3(svp, vqh, otl, otp, otr, otm);
 705                 return;
 706         }
 707         /*
 708          * At this point, the traditional overlay_target_point_t is all that
 709          * needs filling in.  Zero-out the otr for safety.
 710          */
 711         bzero(otr, sizeof (*otr));
 712 
 713 
 714         /*
 715          * Check if this is something that we need to proxy, eg. arp or ndp.
 716          */
 717         if (otl->otl_addru.otlu_l2.otl2_sap == ETHERTYPE_ARP) {
 718                 libvarpd_plugin_proxy_arp(svp->svp_hdl, vqh, otl);
 719                 return;
 720         }
 721 
 722         if (otl->otl_addru.otlu_l2.otl2_dstaddr[0] == 0x33 &&
 723             otl->otl_addru.otlu_l2.otl2_dstaddr[1] == 0x33) {
 724                 if (otl->otl_addru.otlu_l2.otl2_sap == ETHERTYPE_IPV6) {
 725                         libvarpd_plugin_proxy_ndp(svp->svp_hdl, vqh, otl);
 726                 } else {
 727                         libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
 728                 }
 729                 return;
 730         }
 731 
 732         /*
 733          * Watch out for various multicast and broadcast addresses. We've
 734          * already taken care of the IPv6 range above. Now we just need to
 735          * handle broadcast and if the multicast bit is set, lowest bit of the
 736          * first octet of the MAC, then we drop it now.
 737          */
 738         if (bcmp(otl->otl_addru.otlu_l2.otl2_dstaddr, svp_bcast,
 739             ETHERADDRL) == 0 ||
 740             (otl->otl_addru.otlu_l2.otl2_dstaddr[0] & 0x01) == 0x01) {
 741                 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
 742                 return;
 743         }
 744 
 745         /*
 746          * If we have a failure to allocate memory for this, that's not good.
 747          * However, telling the kernel to just drop this packet is much better
 748          * than the alternative at this moment. At least we'll try again and we
 749          * may have something more available to us in a little bit.
 750          */
 751         slp = umem_cache_alloc(svp_lookup_cache, UMEM_DEFAULT);
 752         if (slp == NULL) {
 753                 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
 754                 return;
 755         }
 756 
 757         slp->svl_type = SVP_L_VL2;
 758         slp->svl_u.svl_vl2.svl_handle = vqh;
 759         slp->svl_u.svl_vl2.svl_point = otp;
 760 
 761         svp_remote_vl2_lookup(svp, &slp->svl_query,
 762             otl->otl_addru.otlu_l2.otl2_dstaddr, slp);
 763 }
 764 
 765 /* ARGSUSED */
 766 static int
 767 varpd_svp_nprops(void *arg, uint_t *nprops)
 768 {
 769         *nprops = sizeof (varpd_svp_props) / sizeof (char *);
 770         return (0);
 771 }
 772 
 773 /* ARGSUSED */
 774 static int
 775 varpd_svp_propinfo(void *arg, uint_t propid, varpd_prop_handle_t *vph)
 776 {
 777         switch (propid) {
 778         case 0:
 779                 /* svp/host */
 780                 libvarpd_prop_set_name(vph, varpd_svp_props[0]);
 781                 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
 782                 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_STRING);
 783                 libvarpd_prop_set_nodefault(vph);
 784                 break;
 785         case 1:
 786                 /* svp/port */
 787                 libvarpd_prop_set_name(vph, varpd_svp_props[1]);
 788                 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
 789                 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_UINT);
 790                 (void) libvarpd_prop_set_default(vph, &svp_defport,
 791                     sizeof (svp_defport));
 792                 libvarpd_prop_set_range_uint32(vph, 1, UINT16_MAX);
 793                 break;
 794         case 2:
 795                 /* svp/underlay_ip */
 796                 libvarpd_prop_set_name(vph, varpd_svp_props[2]);
 797                 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
 798                 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_IP);
 799                 libvarpd_prop_set_nodefault(vph);
 800                 break;
 801         case 3:
 802                 /* svp/underlay_port */
 803                 libvarpd_prop_set_name(vph, varpd_svp_props[3]);
 804                 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
 805                 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_UINT);
 806                 (void) libvarpd_prop_set_default(vph, &svp_defuport,
 807                     sizeof (svp_defuport));
 808                 libvarpd_prop_set_range_uint32(vph, 1, UINT16_MAX);
 809                 break;
 810         case 4:
 811                 /* svp/dcid */
 812                 libvarpd_prop_set_name(vph, varpd_svp_props[4]);
 813                 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
 814                 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_UINT);
 815                 libvarpd_prop_set_nodefault(vph);
 816                 libvarpd_prop_set_range_uint32(vph, 1, UINT32_MAX - 1);
 817                 break;
 818         case 5:
 819                 /* svp/router_oui */
 820                 libvarpd_prop_set_name(vph, varpd_svp_props[5]);
 821                 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
 822                 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_ETHER);
 823                 libvarpd_prop_set_nodefault(vph);
 824                 break;
 825         default:
 826                 return (EINVAL);
 827         }
 828         return (0);
 829 }
 830 
 831 static int
 832 varpd_svp_getprop(void *arg, const char *pname, void *buf, uint32_t *sizep)
 833 {
 834         svp_t *svp = arg;
 835 
 836         /* svp/host */
 837         if (strcmp(pname, varpd_svp_props[0]) == 0) {
 838                 size_t len;
 839 
 840                 mutex_enter(&svp->svp_lock);
 841                 if (svp->svp_host == NULL) {
 842                         *sizep = 0;
 843                 } else {
 844                         len = strlen(svp->svp_host) + 1;
 845                         if (*sizep < len) {
 846                                 mutex_exit(&svp->svp_lock);
 847                                 return (EOVERFLOW);
 848                         }
 849                         *sizep = len;
 850                         (void) strlcpy(buf, svp->svp_host, *sizep);
 851                 }
 852                 mutex_exit(&svp->svp_lock);
 853                 return (0);
 854         }
 855 
 856         /* svp/port */
 857         if (strcmp(pname, varpd_svp_props[1]) == 0) {
 858                 uint64_t val;
 859 
 860                 if (*sizep < sizeof (uint64_t))
 861                         return (EOVERFLOW);
 862 
 863                 mutex_enter(&svp->svp_lock);
 864                 if (svp->svp_port == 0) {
 865                         *sizep = 0;
 866                 } else {
 867                         val = svp->svp_port;
 868                         bcopy(&val, buf, sizeof (uint64_t));
 869                         *sizep = sizeof (uint64_t);
 870                 }
 871                 mutex_exit(&svp->svp_lock);
 872                 return (0);
 873         }
 874 
 875         /* svp/underlay_ip */
 876         if (strcmp(pname, varpd_svp_props[2]) == 0) {
 877                 if (*sizep < sizeof (struct in6_addr))
 878                         return (EOVERFLOW);
 879                 mutex_enter(&svp->svp_lock);
 880                 if (svp->svp_huip == B_FALSE) {
 881                         *sizep = 0;
 882                 } else {
 883                         bcopy(&svp->svp_uip, buf, sizeof (struct in6_addr));
 884                         *sizep = sizeof (struct in6_addr);
 885                 }
 886                 mutex_exit(&svp->svp_lock);
 887                 return (0);
 888         }
 889 
 890         /* svp/underlay_port */
 891         if (strcmp(pname, varpd_svp_props[3]) == 0) {
 892                 uint64_t val;
 893 
 894                 if (*sizep < sizeof (uint64_t))
 895                         return (EOVERFLOW);
 896 
 897                 mutex_enter(&svp->svp_lock);
 898                 if (svp->svp_uport == 0) {
 899                         *sizep = 0;
 900                 } else {
 901                         val = svp->svp_uport;
 902                         bcopy(&val, buf, sizeof (uint64_t));
 903                         *sizep = sizeof (uint64_t);
 904                 }
 905 
 906                 mutex_exit(&svp->svp_lock);
 907                 return (0);
 908         }
 909 
 910         /* svp/dcid */
 911         if (strcmp(pname, varpd_svp_props[4]) == 0) {
 912                 uint64_t val;
 913 
 914                 if (*sizep < sizeof (uint64_t))
 915                         return (EOVERFLOW);
 916 
 917                 mutex_enter(&svp->svp_lock);
 918                 if (svp->svp_uport == 0) {
 919                         *sizep = 0;
 920                 } else {
 921                         val = svp->svp_dcid;
 922                         bcopy(&val, buf, sizeof (uint64_t));
 923                         *sizep = sizeof (uint64_t);
 924                 }
 925 
 926                 mutex_exit(&svp->svp_lock);
 927                 return (0);
 928         }
 929 
 930         /* svp/router_oui */
 931         if (strcmp(pname, varpd_svp_props[5]) == 0) {
 932                 if (*sizep < ETHERADDRL)
 933                         return (EOVERFLOW);
 934                 mutex_enter(&svp->svp_lock);
 935 
 936                 if (ether_is_zero(&svp->svp_router_oui)) {
 937                         *sizep = 0;
 938                 } else {
 939                         bcopy(&svp->svp_router_oui, buf, ETHERADDRL);
 940                         *sizep = ETHERADDRL;
 941                 }
 942 
 943                 mutex_exit(&svp->svp_lock);
 944                 return (0);
 945         }
 946         return (EINVAL);
 947 }
 948 
 949 static int
 950 varpd_svp_setprop(void *arg, const char *pname, const void *buf,
 951     const uint32_t size)
 952 {
 953         svp_t *svp = arg;
 954 
 955         /* svp/host */
 956         if (strcmp(pname, varpd_svp_props[0]) == 0) {
 957                 char *dup;
 958                 dup = umem_alloc(size, UMEM_DEFAULT);
 959                 (void) strlcpy(dup, buf, size);
 960                 if (dup == NULL)
 961                         return (ENOMEM);
 962                 mutex_enter(&svp->svp_lock);
 963                 if (svp->svp_host != NULL)
 964                         umem_free(svp->svp_host, strlen(svp->svp_host) + 1);
 965                 svp->svp_host = dup;
 966                 mutex_exit(&svp->svp_lock);
 967                 return (0);
 968         }
 969 
 970         /* svp/port */
 971         if (strcmp(pname, varpd_svp_props[1]) == 0) {
 972                 const uint64_t *valp = buf;
 973                 if (size < sizeof (uint64_t))
 974                         return (EOVERFLOW);
 975 
 976                 if (*valp == 0 || *valp > UINT16_MAX)
 977                         return (EINVAL);
 978 
 979                 mutex_enter(&svp->svp_lock);
 980                 svp->svp_port = (uint16_t)*valp;
 981                 mutex_exit(&svp->svp_lock);
 982                 return (0);
 983         }
 984 
 985         /* svp/underlay_ip */
 986         if (strcmp(pname, varpd_svp_props[2]) == 0) {
 987                 const struct in6_addr *ipv6 = buf;
 988 
 989                 if (size < sizeof (struct in6_addr))
 990                         return (EOVERFLOW);
 991 
 992                 if (IN6_IS_ADDR_V4COMPAT(ipv6))
 993                         return (EINVAL);
 994 
 995                 if (IN6_IS_ADDR_MULTICAST(ipv6))
 996                         return (EINVAL);
 997 
 998                 if (IN6_IS_ADDR_6TO4(ipv6))
 999                         return (EINVAL);
1000 
1001                 if (IN6_IS_ADDR_V4MAPPED(ipv6)) {
1002                         ipaddr_t v4;
1003                         IN6_V4MAPPED_TO_IPADDR(ipv6, v4);
1004                         if (IN_MULTICAST(v4))
1005                                 return (EINVAL);
1006                 }
1007 
1008                 mutex_enter(&svp->svp_lock);
1009                 bcopy(buf, &svp->svp_uip, sizeof (struct in6_addr));
1010                 svp->svp_huip = B_TRUE;
1011                 mutex_exit(&svp->svp_lock);
1012                 return (0);
1013         }
1014 
1015         /* svp/underlay_port */
1016         if (strcmp(pname, varpd_svp_props[3]) == 0) {
1017                 const uint64_t *valp = buf;
1018                 if (size < sizeof (uint64_t))
1019                         return (EOVERFLOW);
1020 
1021                 if (*valp == 0 || *valp > UINT16_MAX)
1022                         return (EINVAL);
1023 
1024                 mutex_enter(&svp->svp_lock);
1025                 svp->svp_uport = (uint16_t)*valp;
1026                 mutex_exit(&svp->svp_lock);
1027 
1028                 return (0);
1029         }
1030 
1031         /* svp/dcid */
1032         if (strcmp(pname, varpd_svp_props[4]) == 0) {
1033                 const uint64_t *valp = buf;
1034                 if (size < sizeof (uint64_t))
1035                         return (EOVERFLOW);
1036 
1037                 if (*valp == 0 || *valp > UINT32_MAX - 1)
1038                         return (EINVAL);
1039 
1040                 mutex_enter(&svp->svp_lock);
1041                 svp->svp_dcid = (uint32_t)*valp;
1042                 mutex_exit(&svp->svp_lock);
1043 
1044                 return (0);
1045         }
1046 
1047         /* svp/router_oui */
1048         if (strcmp(pname, varpd_svp_props[5]) == 0) {
1049                 if (size < ETHERADDRL)
1050                         return (EOVERFLOW);
1051                 mutex_enter(&svp->svp_lock);
1052                 bcopy(buf, &svp->svp_router_oui, ETHERADDRL);
1053                 /* Zero-out the low three bytes. */
1054                 svp->svp_router_oui[3] = 0;
1055                 svp->svp_router_oui[4] = 0;
1056                 svp->svp_router_oui[5] = 0;
1057                 mutex_exit(&svp->svp_lock);
1058                 return (0);
1059         }
1060 
1061         return (EINVAL);
1062 }
1063 
1064 static int
1065 varpd_svp_save(void *arg, nvlist_t *nvp)
1066 {
1067         int ret;
1068         svp_t *svp = arg;
1069 
1070         mutex_enter(&svp->svp_lock);
1071         /* svp/host */
1072         if (svp->svp_host != NULL) {
1073                 if ((ret = nvlist_add_string(nvp, varpd_svp_props[0],
1074                     svp->svp_host)) != 0) {
1075                         mutex_exit(&svp->svp_lock);
1076                         return (ret);
1077                 }
1078         }
1079 
1080         /* svp/port */
1081         if (svp->svp_port != 0) {
1082                 if ((ret = nvlist_add_uint16(nvp, varpd_svp_props[1],
1083                     svp->svp_port)) != 0) {
1084                         mutex_exit(&svp->svp_lock);
1085                         return (ret);
1086                 }
1087         }
1088 
1089         /* svp/underlay_ip */
1090         if (svp->svp_huip == B_TRUE) {
1091                 char buf[INET6_ADDRSTRLEN];
1092 
1093                 if (inet_ntop(AF_INET6, &svp->svp_uip, buf, sizeof (buf)) ==
1094                     NULL)
1095                         libvarpd_panic("unexpected inet_ntop failure: %d",
1096                             errno);
1097 
1098                 if ((ret = nvlist_add_string(nvp, varpd_svp_props[2],
1099                     buf)) != 0) {
1100                         mutex_exit(&svp->svp_lock);
1101                         return (ret);
1102                 }
1103         }
1104 
1105         /* svp/underlay_port */
1106         if (svp->svp_uport != 0) {
1107                 if ((ret = nvlist_add_uint16(nvp, varpd_svp_props[3],
1108                     svp->svp_uport)) != 0) {
1109                         mutex_exit(&svp->svp_lock);
1110                         return (ret);
1111                 }
1112         }
1113 
1114         /* svp/dcid */
1115         if (svp->svp_dcid != 0) {
1116                 if ((ret = nvlist_add_uint32(nvp, varpd_svp_props[4],
1117                     svp->svp_dcid)) != 0) {
1118                         mutex_exit(&svp->svp_lock);
1119                         return (ret);
1120                 }
1121         }
1122 
1123         /* svp/router_oui */
1124         if (!ether_is_zero(&svp->svp_router_oui)) {
1125                 char buf[ETHERADDRSTRL];
1126 
1127                 if (ether_ntoa_r((struct ether_addr *)&svp->svp_router_oui,
1128                     buf) == NULL) {
1129                         libvarpd_panic("unexpected ether_ntoa_r failure: %d",
1130                             errno);
1131                 }
1132 
1133                 if ((ret = nvlist_add_string(nvp, varpd_svp_props[5],
1134                     buf)) != 0) {
1135                         mutex_exit(&svp->svp_lock);
1136                         return (ret);
1137                 }
1138         }
1139 
1140         mutex_exit(&svp->svp_lock);
1141         return (0);
1142 }
1143 
1144 static int
1145 varpd_svp_restore(nvlist_t *nvp, varpd_provider_handle_t *hdl,
1146     overlay_plugin_dest_t dest, void **outp)
1147 {
1148         int ret;
1149         svp_t *svp;
1150         char *ipstr, *hstr, *etherstr;
1151 
1152         if (varpd_svp_valid_dest(dest) == B_FALSE)
1153                 return (ENOTSUP);
1154 
1155         if ((ret = varpd_svp_create(hdl, (void **)&svp, dest)) != 0)
1156                 return (ret);
1157 
1158         /* svp/host */
1159         if ((ret = nvlist_lookup_string(nvp, varpd_svp_props[0],
1160             &hstr)) != 0) {
1161                 if (ret != ENOENT) {
1162                         varpd_svp_destroy(svp);
1163                         return (ret);
1164                 }
1165                 svp->svp_host = NULL;
1166         } else {
1167                 size_t blen = strlen(hstr) + 1;
1168                 svp->svp_host = umem_alloc(blen, UMEM_DEFAULT);
1169                 (void) strlcpy(svp->svp_host, hstr, blen);
1170         }
1171 
1172         /* svp/port */
1173         if ((ret = nvlist_lookup_uint16(nvp, varpd_svp_props[1],
1174             &svp->svp_port)) != 0) {
1175                 if (ret != ENOENT) {
1176                         varpd_svp_destroy(svp);
1177                         return (ret);
1178                 }
1179                 svp->svp_port = 0;
1180         }
1181 
1182         /* svp/underlay_ip */
1183         if ((ret = nvlist_lookup_string(nvp, varpd_svp_props[2],
1184             &ipstr)) != 0) {
1185                 if (ret != ENOENT) {
1186                         varpd_svp_destroy(svp);
1187                         return (ret);
1188                 }
1189                 svp->svp_huip = B_FALSE;
1190         } else {
1191                 ret = inet_pton(AF_INET6, ipstr, &svp->svp_uip);
1192                 if (ret == -1) {
1193                         assert(errno == EAFNOSUPPORT);
1194                         libvarpd_panic("unexpected inet_pton failure: %d",
1195                             errno);
1196                 }
1197 
1198                 if (ret == 0) {
1199                         varpd_svp_destroy(svp);
1200                         return (EINVAL);
1201                 }
1202                 svp->svp_huip = B_TRUE;
1203         }
1204 
1205         /* svp/underlay_port */
1206         if ((ret = nvlist_lookup_uint16(nvp, varpd_svp_props[3],
1207             &svp->svp_uport)) != 0) {
1208                 if (ret != ENOENT) {
1209                         varpd_svp_destroy(svp);
1210                         return (ret);
1211                 }
1212                 svp->svp_uport = 0;
1213         }
1214 
1215         /* svp/dcid */
1216         if ((ret = nvlist_lookup_uint32(nvp, varpd_svp_props[4],
1217             &svp->svp_dcid)) != 0) {
1218                 if (ret != ENOENT) {
1219                         varpd_svp_destroy(svp);
1220                         return (ret);
1221                 }
1222                 svp->svp_dcid = 0;
1223         }
1224 
1225         /* svp/router_oui */
1226         if ((ret = nvlist_lookup_string(nvp, varpd_svp_props[5],
1227             ðerstr)) != 0) {
1228                 if (ret != ENOENT) {
1229                         varpd_svp_destroy(svp);
1230                         return (ret);
1231                 }
1232                 bzero(&svp->svp_router_oui, ETHERADDRL);
1233         } else if (ether_aton_r(etherstr,
1234             (struct ether_addr *)&svp->svp_router_oui) == NULL) {
1235                 libvarpd_panic("unexpected ether_aton_r failure: %d", errno);
1236         }
1237 
1238         svp->svp_hdl = hdl;
1239         *outp = svp;
1240         return (0);
1241 }
1242 
1243 static void
1244 varpd_svp_arp(void *arg, varpd_arp_handle_t *vah, int type,
1245     const struct sockaddr *sock, uint16_t vlan __unused, uint8_t *out)
1246 {
1247         svp_t *svp = arg;
1248         svp_lookup_t *svl;
1249 
1250         if (type != VARPD_QTYPE_ETHERNET) {
1251                 libvarpd_plugin_arp_reply(vah, VARPD_LOOKUP_DROP);
1252                 return;
1253         }
1254 
1255         svl = umem_cache_alloc(svp_lookup_cache, UMEM_DEFAULT);
1256         if (svl == NULL) {
1257                 libvarpd_plugin_arp_reply(vah, VARPD_LOOKUP_DROP);
1258                 return;
1259         }
1260 
1261         svl->svl_type = SVP_L_VL3;
1262         svl->svl_u.svl_vl3.svl_vah = vah;
1263         svl->svl_u.svl_vl3.svl_out = out;
1264         svp_remote_vl3_lookup(svp, &svl->svl_query, sock, svl);
1265 }
1266 
1267 static const varpd_plugin_ops_t varpd_svp_ops = {
1268         0,
1269         varpd_svp_create,
1270         varpd_svp_start,
1271         varpd_svp_stop,
1272         varpd_svp_destroy,
1273         NULL,
1274         varpd_svp_lookup,
1275         varpd_svp_nprops,
1276         varpd_svp_propinfo,
1277         varpd_svp_getprop,
1278         varpd_svp_setprop,
1279         varpd_svp_save,
1280         varpd_svp_restore,
1281         varpd_svp_arp,
1282         NULL
1283 };
1284 
1285 static int
1286 svp_bunyan_init(void)
1287 {
1288         int ret;
1289 
1290         if ((ret = bunyan_init("svp", &svp_bunyan)) != 0)
1291                 return (ret);
1292         ret = bunyan_stream_add(svp_bunyan, "stderr", BUNYAN_L_INFO,
1293             bunyan_stream_fd, (void *)STDERR_FILENO);
1294         if (ret != 0)
1295                 bunyan_fini(svp_bunyan);
1296         return (ret);
1297 }
1298 
1299 static void
1300 svp_bunyan_fini(void)
1301 {
1302         if (svp_bunyan != NULL)
1303                 bunyan_fini(svp_bunyan);
1304 }
1305 
1306 #pragma init(varpd_svp_init)
1307 static void
1308 varpd_svp_init(void)
1309 {
1310         int err;
1311         varpd_plugin_register_t *vpr;
1312 
1313         if (svp_bunyan_init() != 0)
1314                 return;
1315 
1316         if ((err = svp_host_init()) != 0) {
1317                 (void) bunyan_error(svp_bunyan, "failed to init host subsystem",
1318                     BUNYAN_T_INT32, "error", err,
1319                     BUNYAN_T_END);
1320                 svp_bunyan_fini();
1321                 return;
1322         }
1323 
1324         svp_lookup_cache = umem_cache_create("svp_lookup",
1325             sizeof (svp_lookup_t),  0, NULL, NULL, NULL, NULL, NULL, 0);
1326         if (svp_lookup_cache == NULL) {
1327                 (void) bunyan_error(svp_bunyan,
1328                     "failed to create svp_lookup cache",
1329                     BUNYAN_T_INT32, "error", errno,
1330                     BUNYAN_T_END);
1331                 svp_bunyan_fini();
1332                 return;
1333         }
1334 
1335         if ((err = svp_event_init()) != 0) {
1336                 (void) bunyan_error(svp_bunyan,
1337                     "failed to init event subsystem",
1338                     BUNYAN_T_INT32, "error", err,
1339                     BUNYAN_T_END);
1340                 svp_bunyan_fini();
1341                 umem_cache_destroy(svp_lookup_cache);
1342                 return;
1343         }
1344 
1345         if ((err = svp_timer_init()) != 0) {
1346                 (void) bunyan_error(svp_bunyan,
1347                     "failed to init timer subsystem",
1348                     BUNYAN_T_INT32, "error", err,
1349                     BUNYAN_T_END);
1350                 svp_event_fini();
1351                 umem_cache_destroy(svp_lookup_cache);
1352                 svp_bunyan_fini();
1353                 return;
1354         }
1355 
1356         if ((err = svp_remote_init()) != 0) {
1357                 (void) bunyan_error(svp_bunyan,
1358                     "failed to init remote subsystem",
1359                     BUNYAN_T_INT32, "error", err,
1360                     BUNYAN_T_END);
1361                 svp_event_fini();
1362                 umem_cache_destroy(svp_lookup_cache);
1363                 svp_bunyan_fini();
1364                 return;
1365         }
1366 
1367         vpr = libvarpd_plugin_alloc(VARPD_CURRENT_VERSION, &err);
1368         if (vpr == NULL) {
1369                 (void) bunyan_error(svp_bunyan,
1370                     "failed to alloc varpd plugin",
1371                     BUNYAN_T_INT32, "error", err,
1372                     BUNYAN_T_END);
1373                 svp_remote_fini();
1374                 svp_event_fini();
1375                 umem_cache_destroy(svp_lookup_cache);
1376                 svp_bunyan_fini();
1377                 return;
1378         }
1379 
1380         vpr->vpr_mode = OVERLAY_TARGET_DYNAMIC;
1381         vpr->vpr_name = "svp";
1382         vpr->vpr_ops = &varpd_svp_ops;
1383 
1384         if ((err = libvarpd_plugin_register(vpr)) != 0) {
1385                 (void) bunyan_error(svp_bunyan,
1386                     "failed to register varpd plugin",
1387                     BUNYAN_T_INT32, "error", err,
1388                     BUNYAN_T_END);
1389                 svp_remote_fini();
1390                 svp_event_fini();
1391                 umem_cache_destroy(svp_lookup_cache);
1392                 svp_bunyan_fini();
1393 
1394         }
1395         libvarpd_plugin_free(vpr);
1396 }