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 * | | | | | | |
227 * | V ^ v ^ V ^
228 * | +----------------+ +-------------------+ +---------------+
229 * +<-| SVP_CS_BACKOFF | | SVP_CS_CONNECTING | | SVP_CS_ACTIVE |
230 * | +----------------+ +-------------------+ +---------------+
231 * | V ^ V V V
232 * | Backoff wait * | | | * Removed
233 * v interval +--------------+ +-----------------<-----+ | from DNS
234 * | finished | |
235 * | V |
236 * | | V
237 * | | +-----------------+
238 * +----------------+----------<-----+-------<----| SVP_CS_WINDDOWN |
239 * | +-----------------+
240 * * . . . Fatal system, not
241 * | socket error or
242 * V quiesced after
243 * +--------------+ removal from DNS
244 * | SVP_CS_ERROR |
245 * +--------------+
246 * |
247 * * . . . Removed from DNS
248 * v
249 * +------------+
250 * | Connection |
251 * | Destroyed |
252 * +------------+
253 *
254 * --------------------------
255 * Connection Event Injection
256 * --------------------------
257 *
258 * For each connection that exists in the system, we have a timer in place that
259 * is in charge of performing timeout activity. It fires once every thirty
260 * seconds or so for a given connection and checks to ensure that we have had
261 * activity for the most recent query on the connection. If not, it terminates
262 * the connection. This is important as if we have sent all our data and are
263 * waiting for the remote end to reply, without enabling something like TCP
264 * keep-alive, we will not be notified that anything that has happened to the
265 * remote connection, for example a panic. In addition, this also protects
266 * against a server that is up, but a portolan that is not making forward
267 * progress.
268 *
269 * When a timeout occurs, we first try to disassociate any active events, which
270 * by definition must exist. Once that's done, we inject a port source user
271 * event. Now, there is a small gotcha. Let's assume for a moment that we have a
272 * pathological portolan. That means that it knows to inject activity right at
273 * the time out window. That means, that the event may be disassociated before
274 * we could get to it. If that's the case, we must _not_ inject the user event
275 * and instead, we'll let the pending event take care of it. We know that the
276 * pending event hasn't hit the main part of the loop yet, otherwise, it would
277 * have released the lock protecting our state and associated the event.
278 *
279 * ------------
280 * Notes on DNS
281 * ------------
282 *
283 * Unfortunately, doing host name resolution in a way that allows us to leverage
284 * the system's resolvers and the system's caching, require us to make blocking
285 * calls in libc via getaddrinfo(3SOCKET). If we can't reach a given server,
286 * that will tie up a thread for quite some time. To work around that fact,
287 * we're going to create a fixed number of threads and we'll use them to service
288 * our DNS requests. While this isn't ideal, until we have a sane means of
289 * integrating a DNS resolution into an event loop with say portfs, it's not
290 * going to be a fun day no matter what we do.
291 *
292 * ------
293 * Timers
294 * ------
295 *
296 * We maintain a single timer based on CLOCK_REALTIME. It's designed to fire
297 * every second. While we'd rather use CLOCK_HIGHRES just to alleviate ourselves
298 * from timer drift; however, as zones may not actually have CLOCK_HIGHRES
299 * access, we don't want them to end up in there. The timer itself is just a
300 * simple avl tree sorted by expiration time, which is stored as a tick in the
301 * future, a tick is just one second.
302 *
303 * ----------
304 * Shootdowns
305 * ----------
306 *
307 * As part of the protocol, we need to be able to handle shootdowns that inform
308 * us some of the information in the system is out of date. This information
309 * needs to be processed promptly; however, the information is hopefully going
310 * to be relatively infrequent relative to the normal flow of information.
311 *
312 * The shoot down information needs to be done on a per-backend basis. The
313 * general design is that we'll have a single query for this which can fire on a
314 * 5-10s period, we randmoize the latter part to give us a bit more load
315 * spreading. If we complete because there's no work to do, then we wait the
316 * normal period. If we complete, but there's still work to do, we'll go again
317 * after a second.
318 *
319 * A shootdown has a few different parts. We first receive a list of items to
320 * shootdown. After performing all of those, we need to acknowledge them. When
321 * that's been done successfully, we can move onto the next part. From a
322 * protocol perspective, we make a SVP_R_LOG_REQ, we get a reply, and then after
323 * processing them, send an SVP_R_LOG_RM. Only once that's been acked do we
324 * continue.
325 *
326 * However, one of the challenges that we have is that these invalidations are
327 * just that, an invalidation. For a virtual layer two request, that's fine,
328 * because the kernel supports that. However, for virtual layer three
329 * invalidations, we have a bit more work to do. These protocols, ARP and NDP,
330 * don't really support a notion of just an invalidation, instead you have to
331 * inject the new data in a gratuitous fashion.
332 *
333 * To that end, what we instead do is when we receive a VL3 invalidation, we
334 * turn that info a VL3 request. We hold the general request as outstanding
335 * until we receive all of the callbacks for the VL3 invalidations, at which
336 * point we go through and do the log removal request.
337 */
338
339 #include <umem.h>
340 #include <errno.h>
341 #include <stdlib.h>
342 #include <sys/types.h>
343 #include <sys/socket.h>
344 #include <netinet/in.h>
345 #include <arpa/inet.h>
346 #include <libnvpair.h>
347 #include <strings.h>
348 #include <string.h>
349 #include <assert.h>
350 #include <unistd.h>
351
352 #include <libvarpd_provider.h>
353 #include "libvarpd_svp.h"
354
355 bunyan_logger_t *svp_bunyan;
356 static int svp_defport = 1296;
357 static int svp_defuport = 1339;
358 static umem_cache_t *svp_lookup_cache;
359
360 typedef enum svp_lookup_type {
361 SVP_L_UNKNOWN = 0x0,
362 SVP_L_VL2 = 0x1,
363 SVP_L_VL3 = 0x2,
364 SVP_L_ROUTE = 0x3
365 } svp_lookup_type_t;
366
367 typedef struct svp_lookup {
368 int svl_type;
369 union {
370 struct svl_lookup_vl2 {
371 varpd_query_handle_t *svl_handle;
372 overlay_target_point_t *svl_point;
373 } svl_vl2;
374 struct svl_lookup_vl3 {
375 varpd_arp_handle_t *svl_vah;
376 uint8_t *svl_out;
377 } svl_vl3;
378 struct svl_lookup_route {
379 varpd_query_handle_t *svl_handle;
380 overlay_target_point_t *svl_point;
381 overlay_target_route_t *svl_route;
382 overlay_target_mac_t *svl_mac;
383 } svl_route;
384 } svl_u;
385 svp_query_t svl_query;
386 } svp_lookup_t;
387
388 static const char *varpd_svp_props[] = {
389 "svp/host",
390 "svp/port",
391 "svp/underlay_ip",
392 "svp/underlay_port",
393 "svp/dcid",
394 "svp/router_oui"
395 };
396
397 static const uint8_t svp_bcast[6] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
398
399 int
400 svp_comparator(const void *l, const void *r)
401 {
402 const svp_t *ls = l;
403 const svp_t *rs = r;
404
405 if (ls->svp_vid > rs->svp_vid)
406 return (1);
407 if (ls->svp_vid < rs->svp_vid)
408 return (-1);
409 return (0);
410 }
411
412 static void
413 svp_vl2_lookup_cb(svp_t *svp, svp_status_t status, const struct in6_addr *uip,
414 const uint16_t uport, void *arg)
415 {
416 svp_lookup_t *svl = arg;
417 overlay_target_point_t *otp;
418
419 assert(svp != NULL);
420 assert(arg != NULL);
421
422 if (status != SVP_S_OK) {
423 libvarpd_plugin_query_reply(svl->svl_u.svl_vl2.svl_handle,
424 VARPD_LOOKUP_DROP);
425 umem_cache_free(svp_lookup_cache, svl);
426 return;
427 }
428
429 otp = svl->svl_u.svl_vl2.svl_point;
430 bcopy(uip, &otp->otp_ip, sizeof (struct in6_addr));
431 otp->otp_port = uport;
432 libvarpd_plugin_query_reply(svl->svl_u.svl_vl2.svl_handle,
433 VARPD_LOOKUP_OK);
434 umem_cache_free(svp_lookup_cache, svl);
435 }
436
437 static void
438 svp_vl3_lookup_cb(svp_t *svp, svp_status_t status, const uint8_t *vl2mac,
439 const struct in6_addr *uip, const uint16_t uport, void *arg)
440 {
441 /* Initialize address-holders to 0 for comparisons-to-zeroes later. */
442 overlay_target_point_t point = { 0 };
443 svp_lookup_t *svl = arg;
444 uint8_t nexthop_mac[6] = { 0, 0, 0, 0, 0, 0 };
445
446 assert(svp != NULL);
447 assert(svl != NULL);
448
449 if (status != SVP_S_OK) {
450 libvarpd_plugin_arp_reply(svl->svl_u.svl_vl3.svl_vah,
451 VARPD_LOOKUP_DROP);
452 umem_cache_free(svp_lookup_cache, svl);
453 return;
454 }
455
456 /* Inject the L2 mapping before the L3 */
457 if (uport != 0 &&
458 bcmp(uip, &point.otp_ip, sizeof (struct in6_addr)) != 0) {
459 /* Normal L3 lookup result... */
460 bcopy(uip, &point.otp_ip, sizeof (struct in6_addr));
461 point.otp_port = uport;
462 libvarpd_inject_varp(svp->svp_hdl, vl2mac, &point);
463 } else {
464 /*
465 * Oh my, we have a next-hop router IP.
466 * Set the MAC to the ouid+vid concatenated
467 * special-router-MAC. Overlay down below will know
468 * that uport == 0 means the MAC is a special one.
469 */
470 if (bcmp(svp->svp_router_oui, nexthop_mac, ETHERADDRL) == 0) {
471 /*
472 * We don't have a router_oui, so we can't support
473 * special-router-MAC. Drop it.
474 */
475 libvarpd_plugin_arp_reply(svl->svl_u.svl_vl3.svl_vah,
476 VARPD_LOOKUP_DROP);
477 umem_cache_free(svp_lookup_cache, svl);
478 return;
479 }
480 bcopy(svp->svp_router_oui, nexthop_mac, 3);
481 nexthop_mac[3] = (svp->svp_vid >> 16) & 0xff;
482 nexthop_mac[4] = (svp->svp_vid >> 8) & 0xff;
483 nexthop_mac[5] = svp->svp_vid & 0xff;
484 vl2mac = nexthop_mac;
485 }
486
487 bcopy(vl2mac, svl->svl_u.svl_vl3.svl_out, ETHERADDRL);
488 libvarpd_plugin_arp_reply(svl->svl_u.svl_vl3.svl_vah,
489 VARPD_LOOKUP_OK);
490 umem_cache_free(svp_lookup_cache, svl);
491 }
492
493 static void
494 svp_vl2_invalidate_cb(svp_t *svp, const uint8_t *vl2mac)
495 {
496 libvarpd_inject_varp(svp->svp_hdl, vl2mac, NULL);
497 }
498
499 static void
500 svp_vl3_inject_cb(svp_t *svp, const uint16_t vlan, const struct in6_addr *vl3ip,
501 const uint8_t *vl2mac, const uint8_t *targmac)
502 {
503 struct in_addr v4;
504
505 /*
506 * At the moment we don't support any IPv6 related log entries, this
507 * will change soon as we develop a bit more of the IPv6 related
508 * infrastructure so we can properly test the injection.
509 */
510 if (IN6_IS_ADDR_V4MAPPED(vl3ip) == 0) {
511 return;
512 } else {
513 IN6_V4MAPPED_TO_INADDR(vl3ip, &v4);
514 if (targmac == NULL)
515 targmac = svp_bcast;
516 libvarpd_inject_arp(svp->svp_hdl, vlan, vl2mac, &v4, targmac);
517 }
518 }
519
520 /* ARGSUSED */
521 static void
522 svp_shootdown_cb(svp_t *svp, const uint8_t *vl2mac, const struct in6_addr *uip,
523 const uint16_t uport)
524 {
525 /*
526 * We should probably do a conditional invalidation here.
527 */
528 libvarpd_inject_varp(svp->svp_hdl, vl2mac, NULL);
529 }
530
531 static void
532 svp_route_lookup_cb(svp_t *svp, svp_status_t status, uint32_t dcid,
533 uint32_t vnetid, uint16_t vlan, uint8_t *srcmac, uint8_t *dstmac,
534 uint16_t ul3_port, uint8_t *ul3_addr, uint8_t srcpfx, uint8_t dstpfx,
535 void *arg)
536 {
537 svp_lookup_t *svl = arg;
538 overlay_target_point_t *otp;
539 overlay_target_route_t *otr;
540 overlay_target_mac_t *otm;
541
542 if (status != SVP_S_OK) {
543 libvarpd_plugin_query_reply(svl->svl_u.svl_route.svl_handle,
544 VARPD_LOOKUP_DROP);
545 umem_cache_free(svp_lookup_cache, svl);
546 return;
547 }
548
549 otp = svl->svl_u.svl_route.svl_point;
550 bcopy(ul3_addr, &otp->otp_ip, sizeof (struct in6_addr));
551 otp->otp_port = ul3_port;
552
553 otr = svl->svl_u.svl_route.svl_route;
554 otr->otr_vnet = vnetid;
555 otr->otr_vlan = vlan;
556 bcopy(srcmac, otr->otr_srcmac, ETHERADDRL);
557
558 otm = svl->svl_u.svl_route.svl_mac;
559 otm->otm_dcid = dcid;
560 bcopy(dstmac, otm->otm_mac, ETHERADDRL);
561
562 libvarpd_plugin_query_reply(svl->svl_u.svl_route.svl_handle,
563 VARPD_LOOKUP_OK);
564 umem_cache_free(svp_lookup_cache, svl);
565 }
566
567 static svp_cb_t svp_defops = {
568 svp_vl2_lookup_cb,
569 svp_vl3_lookup_cb,
570 svp_vl2_invalidate_cb,
571 svp_vl3_inject_cb,
572 svp_shootdown_cb,
573 svp_route_lookup_cb,
574 };
575
576 static boolean_t
577 varpd_svp_valid_dest(overlay_plugin_dest_t dest)
578 {
579 if (dest != (OVERLAY_PLUGIN_D_IP | OVERLAY_PLUGIN_D_PORT))
580 return (B_FALSE);
581
582 return (B_TRUE);
583 }
584
585 static int
586 varpd_svp_create(varpd_provider_handle_t *hdl, void **outp,
587 overlay_plugin_dest_t dest)
588 {
589 int ret;
590 svp_t *svp;
591
592 if (varpd_svp_valid_dest(dest) == B_FALSE)
593 return (ENOTSUP);
594
595 svp = umem_zalloc(sizeof (svp_t), UMEM_DEFAULT);
596 if (svp == NULL)
597 return (ENOMEM);
598
599 if ((ret = mutex_init(&svp->svp_lock, USYNC_THREAD | LOCK_ERRORCHECK,
600 NULL)) != 0) {
601 umem_free(svp, sizeof (svp_t));
602 return (ret);
603 }
604
605 svp->svp_port = svp_defport;
606 svp->svp_uport = svp_defuport;
607 svp->svp_cb = svp_defops;
608 svp->svp_hdl = hdl;
609 svp->svp_vid = libvarpd_plugin_vnetid(svp->svp_hdl);
610 *outp = svp;
611 return (0);
612 }
613
614 static int
615 varpd_svp_start(void *arg)
616 {
617 int ret;
618 svp_remote_t *srp;
619 svp_t *svp = arg;
620
621 mutex_enter(&svp->svp_lock);
622 if (svp->svp_host == NULL || svp->svp_port == 0 ||
623 svp->svp_huip == B_FALSE || svp->svp_uport == 0) {
624 mutex_exit(&svp->svp_lock);
625 return (EAGAIN);
626 }
627 mutex_exit(&svp->svp_lock);
628
629 if ((ret = svp_remote_find(svp->svp_host, svp->svp_port, &svp->svp_uip,
630 &srp)) != 0)
631 return (ret);
632
633 if ((ret = svp_remote_attach(srp, svp)) != 0) {
634 svp_remote_release(srp);
635 return (ret);
636 }
637
638 return (0);
639 }
640
641 static void
642 varpd_svp_stop(void *arg)
643 {
644 svp_t *svp = arg;
645
646 svp_remote_detach(svp);
647 }
648
649 static void
650 varpd_svp_destroy(void *arg)
651 {
652 svp_t *svp = arg;
653
654 if (svp->svp_host != NULL)
655 umem_free(svp->svp_host, strlen(svp->svp_host) + 1);
656
657 if (mutex_destroy(&svp->svp_lock) != 0)
658 libvarpd_panic("failed to destroy svp_t`svp_lock");
659
660 umem_free(svp, sizeof (svp_t));
661 }
662
663 static void
664 varpd_svp_lookup_l3(svp_t *svp, varpd_query_handle_t *vqh,
665 const overlay_targ_lookup_t *otl, overlay_target_point_t *otp,
666 overlay_target_route_t *otr, overlay_target_mac_t *otm)
667 {
668 svp_lookup_t *slp;
669 uint32_t type;
670 const struct in6_addr *src = &otl->otl_addru.otlu_l3.otl3_srcip,
671 *dst = &otl->otl_addru.otlu_l3.otl3_dstip;
672
673 /*
674 * otl is an L3 request, so we have src/dst IPs for the inner packet.
675 * We also have the vlan.
676 *
677 * Assume kernel's overlay module is caching well, so we are directly
678 * going to query (i.e. no caching up here of actual destinations).
679 *
680 * Our existing remote sever (svp_remote), but with the new message
681 * SVP_R_ROUTE_REQ.
682 */
683
684 if (IN6_IS_ADDR_V4MAPPED(src)) {
685 if (!IN6_IS_ADDR_V4MAPPED(dst)) {
686 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
687 return;
688 }
689 type = SVP_VL3_IP;
690 } else {
691 if (IN6_IS_ADDR_V4MAPPED(dst)) {
692 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
693 return;
694 }
695 type = SVP_VL3_IPV6;
696 }
697
698 slp = umem_cache_alloc(svp_lookup_cache, UMEM_DEFAULT);
699 if (slp == NULL) {
700 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
701 return;
702 }
703
704 slp->svl_type = SVP_L_ROUTE;
705 slp->svl_u.svl_route.svl_handle = vqh;
706 slp->svl_u.svl_route.svl_point = otp;
707 slp->svl_u.svl_route.svl_route = otr;
708 slp->svl_u.svl_route.svl_mac = otm;
709
710 svp_remote_route_lookup(svp, &slp->svl_query, src, dst,
711 otl->otl_vnetid, (uint16_t)otl->otl_vlan, slp);
712 }
713
714 static void
715 varpd_svp_lookup(void *arg, varpd_query_handle_t *vqh,
716 const overlay_targ_lookup_t *otl, overlay_target_point_t *otp,
717 overlay_target_route_t *otr, overlay_target_mac_t *otm)
718 {
719 svp_lookup_t *slp;
720 svp_t *svp = arg;
721
722 /*
723 * Shuffle off L3 lookups to their own codepath.
724 */
725 if (otl->otl_l3req) {
726 varpd_svp_lookup_l3(svp, vqh, otl, otp, otr, otm);
727 return;
728 }
729 /*
730 * At this point, the traditional overlay_target_point_t is all that
731 * needs filling in. Zero-out the otr for safety.
732 */
733 bzero(otr, sizeof (*otr));
734
735
736 /*
737 * Check if this is something that we need to proxy, eg. arp or ndp.
738 */
739 if (otl->otl_addru.otlu_l2.otl2_sap == ETHERTYPE_ARP) {
740 libvarpd_plugin_proxy_arp(svp->svp_hdl, vqh, otl);
741 return;
742 }
743
744 if (otl->otl_addru.otlu_l2.otl2_dstaddr[0] == 0x33 &&
745 otl->otl_addru.otlu_l2.otl2_dstaddr[1] == 0x33) {
746 if (otl->otl_addru.otlu_l2.otl2_sap == ETHERTYPE_IPV6) {
747 libvarpd_plugin_proxy_ndp(svp->svp_hdl, vqh, otl);
748 } else {
749 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
750 }
751 return;
752 }
753
754 /*
755 * Watch out for various multicast and broadcast addresses. We've
756 * already taken care of the IPv6 range above. Now we just need to
757 * handle broadcast and if the multicast bit is set, lowest bit of the
758 * first octet of the MAC, then we drop it now.
759 */
760 if (bcmp(otl->otl_addru.otlu_l2.otl2_dstaddr, svp_bcast,
761 ETHERADDRL) == 0 ||
762 (otl->otl_addru.otlu_l2.otl2_dstaddr[0] & 0x01) == 0x01) {
763 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
764 return;
765 }
766
767 /*
768 * If we have a failure to allocate memory for this, that's not good.
769 * However, telling the kernel to just drop this packet is much better
770 * than the alternative at this moment. At least we'll try again and we
771 * may have something more available to us in a little bit.
772 */
773 slp = umem_cache_alloc(svp_lookup_cache, UMEM_DEFAULT);
774 if (slp == NULL) {
775 libvarpd_plugin_query_reply(vqh, VARPD_LOOKUP_DROP);
776 return;
777 }
778
779 slp->svl_type = SVP_L_VL2;
780 slp->svl_u.svl_vl2.svl_handle = vqh;
781 slp->svl_u.svl_vl2.svl_point = otp;
782
783 svp_remote_vl2_lookup(svp, &slp->svl_query,
784 otl->otl_addru.otlu_l2.otl2_dstaddr, slp);
785 }
786
787 /* ARGSUSED */
788 static int
789 varpd_svp_nprops(void *arg, uint_t *nprops)
790 {
791 *nprops = sizeof (varpd_svp_props) / sizeof (char *);
792 return (0);
793 }
794
795 /* ARGSUSED */
796 static int
797 varpd_svp_propinfo(void *arg, uint_t propid, varpd_prop_handle_t *vph)
798 {
799 switch (propid) {
800 case 0:
801 /* svp/host */
802 libvarpd_prop_set_name(vph, varpd_svp_props[0]);
803 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
804 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_STRING);
805 libvarpd_prop_set_nodefault(vph);
806 break;
807 case 1:
808 /* svp/port */
809 libvarpd_prop_set_name(vph, varpd_svp_props[1]);
810 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
811 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_UINT);
812 (void) libvarpd_prop_set_default(vph, &svp_defport,
813 sizeof (svp_defport));
814 libvarpd_prop_set_range_uint32(vph, 1, UINT16_MAX);
815 break;
816 case 2:
817 /* svp/underlay_ip */
818 libvarpd_prop_set_name(vph, varpd_svp_props[2]);
819 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
820 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_IP);
821 libvarpd_prop_set_nodefault(vph);
822 break;
823 case 3:
824 /* svp/underlay_port */
825 libvarpd_prop_set_name(vph, varpd_svp_props[3]);
826 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
827 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_UINT);
828 (void) libvarpd_prop_set_default(vph, &svp_defuport,
829 sizeof (svp_defuport));
830 libvarpd_prop_set_range_uint32(vph, 1, UINT16_MAX);
831 break;
832 case 4:
833 /* svp/dcid */
834 libvarpd_prop_set_name(vph, varpd_svp_props[4]);
835 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
836 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_UINT);
837 libvarpd_prop_set_nodefault(vph);
838 libvarpd_prop_set_range_uint32(vph, 1, UINT32_MAX - 1);
839 break;
840 case 5:
841 /* svp/router_oui */
842 libvarpd_prop_set_name(vph, varpd_svp_props[5]);
843 libvarpd_prop_set_prot(vph, OVERLAY_PROP_PERM_RRW);
844 libvarpd_prop_set_type(vph, OVERLAY_PROP_T_ETHER);
845 libvarpd_prop_set_nodefault(vph);
846 break;
847 default:
848 return (EINVAL);
849 }
850 return (0);
851 }
852
853 static int
854 varpd_svp_getprop(void *arg, const char *pname, void *buf, uint32_t *sizep)
855 {
856 svp_t *svp = arg;
857
858 /* svp/host */
859 if (strcmp(pname, varpd_svp_props[0]) == 0) {
860 size_t len;
861
862 mutex_enter(&svp->svp_lock);
863 if (svp->svp_host == NULL) {
864 *sizep = 0;
865 } else {
866 len = strlen(svp->svp_host) + 1;
867 if (*sizep < len) {
868 mutex_exit(&svp->svp_lock);
869 return (EOVERFLOW);
870 }
871 *sizep = len;
872 (void) strlcpy(buf, svp->svp_host, *sizep);
873 }
874 mutex_exit(&svp->svp_lock);
875 return (0);
876 }
877
878 /* svp/port */
879 if (strcmp(pname, varpd_svp_props[1]) == 0) {
880 uint64_t val;
881
882 if (*sizep < sizeof (uint64_t))
883 return (EOVERFLOW);
884
885 mutex_enter(&svp->svp_lock);
886 if (svp->svp_port == 0) {
887 *sizep = 0;
888 } else {
889 val = svp->svp_port;
890 bcopy(&val, buf, sizeof (uint64_t));
891 *sizep = sizeof (uint64_t);
892 }
893 mutex_exit(&svp->svp_lock);
894 return (0);
895 }
896
897 /* svp/underlay_ip */
898 if (strcmp(pname, varpd_svp_props[2]) == 0) {
899 if (*sizep < sizeof (struct in6_addr))
900 return (EOVERFLOW);
901 mutex_enter(&svp->svp_lock);
902 if (svp->svp_huip == B_FALSE) {
903 *sizep = 0;
904 } else {
905 bcopy(&svp->svp_uip, buf, sizeof (struct in6_addr));
906 *sizep = sizeof (struct in6_addr);
907 }
908 mutex_exit(&svp->svp_lock);
909 return (0);
910 }
911
912 /* svp/underlay_port */
913 if (strcmp(pname, varpd_svp_props[3]) == 0) {
914 uint64_t val;
915
916 if (*sizep < sizeof (uint64_t))
917 return (EOVERFLOW);
918
919 mutex_enter(&svp->svp_lock);
920 if (svp->svp_uport == 0) {
921 *sizep = 0;
922 } else {
923 val = svp->svp_uport;
924 bcopy(&val, buf, sizeof (uint64_t));
925 *sizep = sizeof (uint64_t);
926 }
927
928 mutex_exit(&svp->svp_lock);
929 return (0);
930 }
931
932 /* svp/dcid */
933 if (strcmp(pname, varpd_svp_props[4]) == 0) {
934 uint64_t val;
935
936 if (*sizep < sizeof (uint64_t))
937 return (EOVERFLOW);
938
939 mutex_enter(&svp->svp_lock);
940 if (svp->svp_uport == 0) {
941 *sizep = 0;
942 } else {
943 val = svp->svp_dcid;
944 bcopy(&val, buf, sizeof (uint64_t));
945 *sizep = sizeof (uint64_t);
946 }
947
948 mutex_exit(&svp->svp_lock);
949 return (0);
950 }
951
952 /* svp/router_oui */
953 if (strcmp(pname, varpd_svp_props[5]) == 0) {
954 if (*sizep < ETHERADDRL)
955 return (EOVERFLOW);
956 mutex_enter(&svp->svp_lock);
957
958 if (ether_is_zero(&svp->svp_router_oui)) {
959 *sizep = 0;
960 } else {
961 bcopy(&svp->svp_router_oui, buf, ETHERADDRL);
962 *sizep = ETHERADDRL;
963 }
964
965 mutex_exit(&svp->svp_lock);
966 return (0);
967 }
968 return (EINVAL);
969 }
970
971 static int
972 varpd_svp_setprop(void *arg, const char *pname, const void *buf,
973 const uint32_t size)
974 {
975 svp_t *svp = arg;
976
977 /* svp/host */
978 if (strcmp(pname, varpd_svp_props[0]) == 0) {
979 char *dup;
980 dup = umem_alloc(size, UMEM_DEFAULT);
981 (void) strlcpy(dup, buf, size);
982 if (dup == NULL)
983 return (ENOMEM);
984 mutex_enter(&svp->svp_lock);
985 if (svp->svp_host != NULL)
986 umem_free(svp->svp_host, strlen(svp->svp_host) + 1);
987 svp->svp_host = dup;
988 mutex_exit(&svp->svp_lock);
989 return (0);
990 }
991
992 /* svp/port */
993 if (strcmp(pname, varpd_svp_props[1]) == 0) {
994 const uint64_t *valp = buf;
995 if (size < sizeof (uint64_t))
996 return (EOVERFLOW);
997
998 if (*valp == 0 || *valp > UINT16_MAX)
999 return (EINVAL);
1000
1001 mutex_enter(&svp->svp_lock);
1002 svp->svp_port = (uint16_t)*valp;
1003 mutex_exit(&svp->svp_lock);
1004 return (0);
1005 }
1006
1007 /* svp/underlay_ip */
1008 if (strcmp(pname, varpd_svp_props[2]) == 0) {
1009 const struct in6_addr *ipv6 = buf;
1010
1011 if (size < sizeof (struct in6_addr))
1012 return (EOVERFLOW);
1013
1014 if (IN6_IS_ADDR_V4COMPAT(ipv6))
1015 return (EINVAL);
1016
1017 if (IN6_IS_ADDR_MULTICAST(ipv6))
1018 return (EINVAL);
1019
1020 if (IN6_IS_ADDR_6TO4(ipv6))
1021 return (EINVAL);
1022
1023 if (IN6_IS_ADDR_V4MAPPED(ipv6)) {
1024 ipaddr_t v4;
1025 IN6_V4MAPPED_TO_IPADDR(ipv6, v4);
1026 if (IN_MULTICAST(v4))
1027 return (EINVAL);
1028 }
1029
1030 mutex_enter(&svp->svp_lock);
1031 bcopy(buf, &svp->svp_uip, sizeof (struct in6_addr));
1032 svp->svp_huip = B_TRUE;
1033 mutex_exit(&svp->svp_lock);
1034 return (0);
1035 }
1036
1037 /* svp/underlay_port */
1038 if (strcmp(pname, varpd_svp_props[3]) == 0) {
1039 const uint64_t *valp = buf;
1040 if (size < sizeof (uint64_t))
1041 return (EOVERFLOW);
1042
1043 if (*valp == 0 || *valp > UINT16_MAX)
1044 return (EINVAL);
1045
1046 mutex_enter(&svp->svp_lock);
1047 svp->svp_uport = (uint16_t)*valp;
1048 mutex_exit(&svp->svp_lock);
1049
1050 return (0);
1051 }
1052
1053 /* svp/dcid */
1054 if (strcmp(pname, varpd_svp_props[4]) == 0) {
1055 const uint64_t *valp = buf;
1056 if (size < sizeof (uint64_t))
1057 return (EOVERFLOW);
1058
1059 if (*valp == 0 || *valp > UINT32_MAX - 1)
1060 return (EINVAL);
1061
1062 mutex_enter(&svp->svp_lock);
1063 svp->svp_dcid = (uint32_t)*valp;
1064 mutex_exit(&svp->svp_lock);
1065
1066 return (0);
1067 }
1068
1069 /* svp/router_oui */
1070 if (strcmp(pname, varpd_svp_props[5]) == 0) {
1071 if (size < ETHERADDRL)
1072 return (EOVERFLOW);
1073 mutex_enter(&svp->svp_lock);
1074 bcopy(buf, &svp->svp_router_oui, ETHERADDRL);
1075 /* Zero-out the low three bytes. */
1076 svp->svp_router_oui[3] = 0;
1077 svp->svp_router_oui[4] = 0;
1078 svp->svp_router_oui[5] = 0;
1079 mutex_exit(&svp->svp_lock);
1080 return (0);
1081 }
1082
1083 return (EINVAL);
1084 }
1085
1086 static int
1087 varpd_svp_save(void *arg, nvlist_t *nvp)
1088 {
1089 int ret;
1090 svp_t *svp = arg;
1091
1092 mutex_enter(&svp->svp_lock);
1093 /* svp/host */
1094 if (svp->svp_host != NULL) {
1095 if ((ret = nvlist_add_string(nvp, varpd_svp_props[0],
1096 svp->svp_host)) != 0) {
1097 mutex_exit(&svp->svp_lock);
1098 return (ret);
1099 }
1100 }
1101
1102 /* svp/port */
1103 if (svp->svp_port != 0) {
1104 if ((ret = nvlist_add_uint16(nvp, varpd_svp_props[1],
1105 svp->svp_port)) != 0) {
1106 mutex_exit(&svp->svp_lock);
1107 return (ret);
1108 }
1109 }
1110
1111 /* svp/underlay_ip */
1112 if (svp->svp_huip == B_TRUE) {
1113 char buf[INET6_ADDRSTRLEN];
1114
1115 if (inet_ntop(AF_INET6, &svp->svp_uip, buf, sizeof (buf)) ==
1116 NULL)
1117 libvarpd_panic("unexpected inet_ntop failure: %d",
1118 errno);
1119
1120 if ((ret = nvlist_add_string(nvp, varpd_svp_props[2],
1121 buf)) != 0) {
1122 mutex_exit(&svp->svp_lock);
1123 return (ret);
1124 }
1125 }
1126
1127 /* svp/underlay_port */
1128 if (svp->svp_uport != 0) {
1129 if ((ret = nvlist_add_uint16(nvp, varpd_svp_props[3],
1130 svp->svp_uport)) != 0) {
1131 mutex_exit(&svp->svp_lock);
1132 return (ret);
1133 }
1134 }
1135
1136 /* svp/dcid */
1137 if (svp->svp_dcid != 0) {
1138 if ((ret = nvlist_add_uint32(nvp, varpd_svp_props[4],
1139 svp->svp_dcid)) != 0) {
1140 mutex_exit(&svp->svp_lock);
1141 return (ret);
1142 }
1143 }
1144
1145 /* svp/router_oui */
1146 if (!ether_is_zero(&svp->svp_router_oui)) {
1147 char buf[ETHERADDRSTRL];
1148
1149 if (ether_ntoa_r((struct ether_addr *)&svp->svp_router_oui,
1150 buf) == NULL) {
1151 libvarpd_panic("unexpected ether_ntoa_r failure: %d",
1152 errno);
1153 }
1154
1155 if ((ret = nvlist_add_string(nvp, varpd_svp_props[5],
1156 buf)) != 0) {
1157 mutex_exit(&svp->svp_lock);
1158 return (ret);
1159 }
1160 }
1161
1162 mutex_exit(&svp->svp_lock);
1163 return (0);
1164 }
1165
1166 static int
1167 varpd_svp_restore(nvlist_t *nvp, varpd_provider_handle_t *hdl,
1168 overlay_plugin_dest_t dest, void **outp)
1169 {
1170 int ret;
1171 svp_t *svp;
1172 char *ipstr, *hstr, *etherstr;
1173
1174 if (varpd_svp_valid_dest(dest) == B_FALSE)
1175 return (ENOTSUP);
1176
1177 if ((ret = varpd_svp_create(hdl, (void **)&svp, dest)) != 0)
1178 return (ret);
1179
1180 /* svp/host */
1181 if ((ret = nvlist_lookup_string(nvp, varpd_svp_props[0],
1182 &hstr)) != 0) {
1183 if (ret != ENOENT) {
1184 varpd_svp_destroy(svp);
1185 return (ret);
1186 }
1187 svp->svp_host = NULL;
1188 } else {
1189 size_t blen = strlen(hstr) + 1;
1190 svp->svp_host = umem_alloc(blen, UMEM_DEFAULT);
1191 (void) strlcpy(svp->svp_host, hstr, blen);
1192 }
1193
1194 /* svp/port */
1195 if ((ret = nvlist_lookup_uint16(nvp, varpd_svp_props[1],
1196 &svp->svp_port)) != 0) {
1197 if (ret != ENOENT) {
1198 varpd_svp_destroy(svp);
1199 return (ret);
1200 }
1201 svp->svp_port = 0;
1202 }
1203
1204 /* svp/underlay_ip */
1205 if ((ret = nvlist_lookup_string(nvp, varpd_svp_props[2],
1206 &ipstr)) != 0) {
1207 if (ret != ENOENT) {
1208 varpd_svp_destroy(svp);
1209 return (ret);
1210 }
1211 svp->svp_huip = B_FALSE;
1212 } else {
1213 ret = inet_pton(AF_INET6, ipstr, &svp->svp_uip);
1214 if (ret == -1) {
1215 assert(errno == EAFNOSUPPORT);
1216 libvarpd_panic("unexpected inet_pton failure: %d",
1217 errno);
1218 }
1219
1220 if (ret == 0) {
1221 varpd_svp_destroy(svp);
1222 return (EINVAL);
1223 }
1224 svp->svp_huip = B_TRUE;
1225 }
1226
1227 /* svp/underlay_port */
1228 if ((ret = nvlist_lookup_uint16(nvp, varpd_svp_props[3],
1229 &svp->svp_uport)) != 0) {
1230 if (ret != ENOENT) {
1231 varpd_svp_destroy(svp);
1232 return (ret);
1233 }
1234 svp->svp_uport = 0;
1235 }
1236
1237 /* svp/dcid */
1238 if ((ret = nvlist_lookup_uint32(nvp, varpd_svp_props[4],
1239 &svp->svp_dcid)) != 0) {
1240 if (ret != ENOENT) {
1241 varpd_svp_destroy(svp);
1242 return (ret);
1243 }
1244 svp->svp_dcid = 0;
1245 }
1246
1247 /* svp/router_oui */
1248 if ((ret = nvlist_lookup_string(nvp, varpd_svp_props[5],
1249 ðerstr)) != 0) {
1250 if (ret != ENOENT) {
1251 varpd_svp_destroy(svp);
1252 return (ret);
1253 }
1254 bzero(&svp->svp_router_oui, ETHERADDRL);
1255 } else if (ether_aton_r(etherstr,
1256 (struct ether_addr *)&svp->svp_router_oui) == NULL) {
1257 libvarpd_panic("unexpected ether_aton_r failure: %d", errno);
1258 }
1259
1260 svp->svp_hdl = hdl;
1261 *outp = svp;
1262 return (0);
1263 }
1264
1265 static void
1266 varpd_svp_arp(void *arg, varpd_arp_handle_t *vah, int type,
1267 const struct sockaddr *sock, uint16_t vlan __unused, uint8_t *out)
1268 {
1269 svp_t *svp = arg;
1270 svp_lookup_t *svl;
1271
1272 if (type != VARPD_QTYPE_ETHERNET) {
1273 libvarpd_plugin_arp_reply(vah, VARPD_LOOKUP_DROP);
1274 return;
1275 }
1276
1277 svl = umem_cache_alloc(svp_lookup_cache, UMEM_DEFAULT);
1278 if (svl == NULL) {
1279 libvarpd_plugin_arp_reply(vah, VARPD_LOOKUP_DROP);
1280 return;
1281 }
1282
1283 svl->svl_type = SVP_L_VL3;
1284 svl->svl_u.svl_vl3.svl_vah = vah;
1285 svl->svl_u.svl_vl3.svl_out = out;
1286 svp_remote_vl3_lookup(svp, &svl->svl_query, sock, svl);
1287 }
1288
1289 static const varpd_plugin_ops_t varpd_svp_ops = {
1290 0,
1291 varpd_svp_create,
1292 varpd_svp_start,
1293 varpd_svp_stop,
1294 varpd_svp_destroy,
1295 NULL,
1296 varpd_svp_lookup,
1297 varpd_svp_nprops,
1298 varpd_svp_propinfo,
1299 varpd_svp_getprop,
1300 varpd_svp_setprop,
1301 varpd_svp_save,
1302 varpd_svp_restore,
1303 varpd_svp_arp,
1304 NULL
1305 };
1306
1307 static int
1308 svp_bunyan_init(void)
1309 {
1310 int ret;
1311
1312 if ((ret = bunyan_init("svp", &svp_bunyan)) != 0)
1313 return (ret);
1314 ret = bunyan_stream_add(svp_bunyan, "stderr", BUNYAN_L_INFO,
1315 bunyan_stream_fd, (void *)STDERR_FILENO);
1316 if (ret != 0)
1317 bunyan_fini(svp_bunyan);
1318 return (ret);
1319 }
1320
1321 static void
1322 svp_bunyan_fini(void)
1323 {
1324 if (svp_bunyan != NULL)
1325 bunyan_fini(svp_bunyan);
1326 }
1327
1328 #pragma init(varpd_svp_init)
1329 static void
1330 varpd_svp_init(void)
1331 {
1332 int err;
1333 varpd_plugin_register_t *vpr;
1334
1335 if (svp_bunyan_init() != 0)
1336 return;
1337
1338 if ((err = svp_host_init()) != 0) {
1339 (void) bunyan_error(svp_bunyan, "failed to init host subsystem",
1340 BUNYAN_T_INT32, "error", err,
1341 BUNYAN_T_END);
1342 svp_bunyan_fini();
1343 return;
1344 }
1345
1346 svp_lookup_cache = umem_cache_create("svp_lookup",
1347 sizeof (svp_lookup_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
1348 if (svp_lookup_cache == NULL) {
1349 (void) bunyan_error(svp_bunyan,
1350 "failed to create svp_lookup cache",
1351 BUNYAN_T_INT32, "error", errno,
1352 BUNYAN_T_END);
1353 svp_bunyan_fini();
1354 return;
1355 }
1356
1357 if ((err = svp_event_init()) != 0) {
1358 (void) bunyan_error(svp_bunyan,
1359 "failed to init event subsystem",
1360 BUNYAN_T_INT32, "error", err,
1361 BUNYAN_T_END);
1362 svp_bunyan_fini();
1363 umem_cache_destroy(svp_lookup_cache);
1364 return;
1365 }
1366
1367 if ((err = svp_timer_init()) != 0) {
1368 (void) bunyan_error(svp_bunyan,
1369 "failed to init timer subsystem",
1370 BUNYAN_T_INT32, "error", err,
1371 BUNYAN_T_END);
1372 svp_event_fini();
1373 umem_cache_destroy(svp_lookup_cache);
1374 svp_bunyan_fini();
1375 return;
1376 }
1377
1378 if ((err = svp_remote_init()) != 0) {
1379 (void) bunyan_error(svp_bunyan,
1380 "failed to init remote subsystem",
1381 BUNYAN_T_INT32, "error", err,
1382 BUNYAN_T_END);
1383 svp_event_fini();
1384 umem_cache_destroy(svp_lookup_cache);
1385 svp_bunyan_fini();
1386 return;
1387 }
1388
1389 vpr = libvarpd_plugin_alloc(VARPD_CURRENT_VERSION, &err);
1390 if (vpr == NULL) {
1391 (void) bunyan_error(svp_bunyan,
1392 "failed to alloc varpd plugin",
1393 BUNYAN_T_INT32, "error", err,
1394 BUNYAN_T_END);
1395 svp_remote_fini();
1396 svp_event_fini();
1397 umem_cache_destroy(svp_lookup_cache);
1398 svp_bunyan_fini();
1399 return;
1400 }
1401
1402 vpr->vpr_mode = OVERLAY_TARGET_DYNAMIC;
1403 vpr->vpr_name = "svp";
1404 vpr->vpr_ops = &varpd_svp_ops;
1405
1406 if ((err = libvarpd_plugin_register(vpr)) != 0) {
1407 (void) bunyan_error(svp_bunyan,
1408 "failed to register varpd plugin",
1409 BUNYAN_T_INT32, "error", err,
1410 BUNYAN_T_END);
1411 svp_remote_fini();
1412 svp_event_fini();
1413 umem_cache_destroy(svp_lookup_cache);
1414 svp_bunyan_fini();
1415
1416 }
1417 libvarpd_plugin_free(vpr);
1418 }