xref: /illumos-gate/usr/src/uts/common/io/ib/clients/iser/iser_cm.c (revision dbed73cbda2229fd1aa6dc5743993cae7f0a7ee9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/ddi.h>
28 #include <sys/types.h>
29 #include <sys/socket.h>
30 #include <netinet/in.h>
31 #include <sys/sunddi.h>
32 #include <sys/ib/ibtl/ibti.h>
33 #include <sys/ib/ibtl/ibtl_types.h>
34 
35 #include <sys/ib/clients/iser/iser.h>
36 
37 extern idm_transport_ops_t	iser_transport_ops;
38 
39 /*
40  * iser_cm.c
41  *    InfiniBand Communication Manager routines for iSER
42  */
43 static ibt_cm_status_t iser_ib_handle_cm_req(idm_svc_t *svc_hdl,
44     ibt_cm_event_t *evp, ibt_cm_return_args_t *rargsp, void *rcmp,
45     ibt_priv_data_len_t rcmp_len);
46 
47 static ibt_cm_status_t iser_ib_handle_cm_rep(iser_state_t *statep,
48     ibt_cm_event_t *evp, ibt_cm_return_args_t *rargsp, void *rcmp,
49     ibt_priv_data_len_t rcmp_len);
50 
51 static ibt_cm_status_t iser_handle_cm_conn_est(ibt_cm_event_t *evp);
52 static ibt_cm_status_t iser_handle_cm_conn_closed(ibt_cm_event_t *evp);
53 static ibt_cm_status_t iser_handle_cm_event_failure(ibt_cm_event_t *evp);
54 
55 /*
56  * iser_ib_cm_handler()
57  */
58 ibt_cm_status_t
59 iser_ib_cm_handler(void *cm_private, ibt_cm_event_t *eventp,
60     ibt_cm_return_args_t *ret_args, void *ret_priv_data,
61     ibt_priv_data_len_t ret_len_max)
62 {
63 	ibt_cm_status_t	ret = IBT_CM_REJECT;
64 
65 	switch (eventp->cm_type) {
66 
67 	case IBT_CM_EVENT_REQ_RCV:
68 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: IBT_CM_EVENT_REQ_RCV");
69 		ret = iser_ib_handle_cm_req((idm_svc_t *)cm_private, eventp,
70 		    ret_args, ret_priv_data, ret_len_max);
71 		break;
72 
73 	case IBT_CM_EVENT_REP_RCV:
74 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: IBT_CM_EVENT_REP_RCV");
75 		ret = iser_ib_handle_cm_rep((iser_state_t *)cm_private,
76 		    eventp, ret_args, ret_priv_data, ret_len_max);
77 		break;
78 
79 	case IBT_CM_EVENT_CONN_EST:
80 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: IBT_CM_EVENT_CONN_EST");
81 		ret = iser_handle_cm_conn_est(eventp);
82 		break;
83 
84 	case IBT_CM_EVENT_CONN_CLOSED:
85 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: "
86 		    "IBT_CM_EVENT_CONN_CLOSED");
87 		ret = iser_handle_cm_conn_closed(eventp);
88 		break;
89 
90 	case IBT_CM_EVENT_FAILURE:
91 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler:  Event failure");
92 		ret = iser_handle_cm_event_failure(eventp);
93 		break;
94 
95 	case IBT_CM_EVENT_MRA_RCV:
96 		/* Not supported */
97 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler:  MRA message received");
98 		break;
99 
100 	case IBT_CM_EVENT_LAP_RCV:
101 		/* Not supported */
102 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: LAP message received");
103 		break;
104 
105 	case IBT_CM_EVENT_APR_RCV:
106 		/* Not supported */
107 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: APR message received");
108 		break;
109 
110 	default:
111 		ISER_LOG(CE_NOTE, "iser_ib_cm_handler: unknown event (0x%x)",
112 		    eventp->cm_type);
113 		break;
114 	}
115 
116 	return (ret);
117 }
118 
119 /* ARGSUSED */
120 static ibt_cm_status_t
121 iser_ib_handle_cm_req(idm_svc_t *svc_hdl, ibt_cm_event_t *evp,
122     ibt_cm_return_args_t *rargsp, void *rcmp, ibt_priv_data_len_t rcmp_len)
123 {
124 
125 	iser_private_data_t	iser_priv_data;
126 	ibt_ip_cm_info_t	ipcm_info;
127 	iser_chan_t		*chan;
128 	iser_conn_t		*iser_conn;
129 	int			status;
130 
131 	/*
132 	 * CM private data brings IP information
133 	 * Private data received is a stream of bytes and may not be properly
134 	 * aligned. So, bcopy the data onto the stack before accessing it.
135 	 */
136 	bcopy((uint8_t *)evp->cm_priv_data, &iser_priv_data,
137 	    sizeof (iser_private_data_t));
138 
139 	/* extract the CM IP info */
140 	status = ibt_get_ip_data(evp->cm_priv_data_len, evp->cm_priv_data,
141 	    &ipcm_info);
142 	if (status != IBT_SUCCESS) {
143 		return (IBT_CM_REJECT);
144 	}
145 
146 	ISER_LOG(CE_NOTE, "iser_ib_handle_cm_req: ipcm_info (0x%p): src IP "
147 	    "(0x%08x) src port (0x%04x) dst IP: (0x%08x)", (void *)&ipcm_info,
148 	    ipcm_info.src_addr.un.ip4addr, ipcm_info.src_port,
149 	    ipcm_info.dst_addr.un.ip4addr);
150 
151 	/* Allocate a channel to establish the new connection */
152 	chan = iser_ib_alloc_channel_nopathlookup(
153 	    evp->cm_event.req.req_hca_guid,
154 	    evp->cm_event.req.req_prim_hca_port);
155 	if (chan == NULL) {
156 		ISER_LOG(CE_NOTE, "iser_ib_handle_cm_req: failed to allocate "
157 		    "a channel from src IP (0x%08x) src port (0x%04x) "
158 		    "to dst IP: (0x%08x) on hca(%llx %d)",
159 		    ipcm_info.src_addr.un.ip4addr, ipcm_info.src_port,
160 		    ipcm_info.dst_addr.un.ip4addr,
161 		    (longlong_t)evp->cm_event.req.req_hca_guid,
162 		    evp->cm_event.req.req_prim_hca_port);
163 		return (IBT_CM_REJECT);
164 	}
165 
166 	/* Set the local and remote ip */
167 	chan->ic_localip = ipcm_info.dst_addr;
168 	chan->ic_remoteip = ipcm_info.src_addr;
169 
170 	/* Set the local and remote port numbers on the channel handle */
171 	chan->ic_lport = svc_hdl->is_svc_req.sr_port;
172 	chan->ic_rport = ipcm_info.src_port;
173 
174 	/* Allocate the iser_conn_t for the IDM svc binding */
175 	iser_conn = kmem_zalloc(sizeof (iser_conn_t), KM_SLEEP);
176 
177 	/* Set up the iser_conn attributes */
178 	mutex_init(&iser_conn->ic_lock, NULL, MUTEX_DRIVER, NULL);
179 	cv_init(&iser_conn->ic_stage_cv, NULL, CV_DEFAULT, NULL);
180 	iser_conn->ic_type = ISER_CONN_TYPE_TGT;
181 	iser_conn->ic_chan = chan;
182 	iser_conn->ic_stage = ISER_CONN_STAGE_ALLOCATED;
183 
184 	/* Hold a reference to the iSER service handle */
185 	iser_tgt_svc_hold((iser_svc_t *)svc_hdl->is_iser_svc);
186 
187 	iser_conn->ic_idms = svc_hdl;
188 
189 	/*
190 	 * Now set a pointer to the iser_conn in the iser_chan for
191 	 * access during CM event handling
192 	 */
193 	chan->ic_conn = iser_conn;
194 
195 	rargsp->cm_ret.rep.cm_channel = chan->ic_chanhdl;
196 
197 	return (IBT_CM_ACCEPT);
198 }
199 
200 /* ARGSUSED */
201 static ibt_cm_status_t
202 iser_ib_handle_cm_rep(iser_state_t *statep, ibt_cm_event_t *evp,
203     ibt_cm_return_args_t *rargsp, void *rcmp, ibt_priv_data_len_t rcmp_len)
204 {
205 	/* pre-post work requests into the receive queue */
206 	iser_ib_post_recv(evp->cm_channel);
207 
208 	/* It looks like the RTU need not be send specifically */
209 	return (IBT_CM_ACCEPT);
210 }
211 
212 static ibt_cm_status_t
213 iser_handle_cm_conn_est(ibt_cm_event_t *evp)
214 {
215 	iser_chan_t	*iser_chan;
216 	iser_conn_t	*iser_conn;
217 	iser_svc_t	*iser_svc;
218 	idm_status_t	status;
219 	idm_conn_t	*ic;
220 
221 	iser_chan = (iser_chan_t *)ibt_get_chan_private(evp->cm_channel);
222 
223 	/*
224 	 * An ibt_open_rc_channel() comes in as a IBT_CM_EVENT_REQ_RCV on the
225 	 * iSER-IB target, upon which the target sends a Response, accepting
226 	 * the request. This comes in as a IBT_CM_EVENT_REP_RCV on the iSER-IB
227 	 * initiator, which then sends an RTU. Upon getting this RTU from the
228 	 * iSER-IB initiator, the IBT_CM_EVENT_CONN_EST event is generated on
229 	 * the target. Then subsequently an IBT_CM_EVENT_CONN_EST event is
230 	 * generated on the initiator.
231 	 *
232 	 * Our new connection has been established on the target. If we are
233 	 * receiving this event on the target side, the iser_channel can be
234 	 * used as it is already populated. On the target side, an IDM
235 	 * connection is then allocated and the IDM layer is notified.
236 	 * If we are on the initiator we needn't do anything, since we
237 	 * already have the IDM linkage in place for this connection.
238 	 */
239 	if (iser_chan->ic_conn->ic_type == ISER_CONN_TYPE_TGT) {
240 
241 		iser_conn = iser_chan->ic_conn;
242 		iser_svc  = (iser_svc_t *)iser_conn->ic_idms->is_iser_svc;
243 
244 		mutex_enter(&iser_conn->ic_lock);
245 
246 		status = idm_svc_conn_create(iser_conn->ic_idms,
247 		    IDM_TRANSPORT_TYPE_ISER, &ic);
248 		if (status != IDM_STATUS_SUCCESS) {
249 			/*
250 			 * No IDM rsrcs or something equally Bad.
251 			 * Return non-SUCCESS to IBCM. He'll give
252 			 * us a CONN_CLOSED, which we'll handle
253 			 * below.
254 			 */
255 			ISER_LOG(CE_NOTE, "iser_handle_cm_conn_est: "
256 			    "idm_svc_conn_create_failed");
257 			mutex_exit(&iser_conn->ic_lock);
258 			return (IBT_CM_NO_RESOURCE);
259 		}
260 
261 		/* We no longer need the hold on the iSER service handle */
262 		iser_tgt_svc_rele(iser_svc);
263 
264 		/* Hold a reference on the IDM connection handle */
265 		idm_conn_hold(ic);
266 
267 		/* Set the transport ops and conn on the idm_conn handle */
268 		ic->ic_transport_ops = &iser_transport_ops;
269 		ic->ic_transport_private = (void *)iser_conn;
270 		ic->ic_transport_hdrlen = ISER_HEADER_LENGTH;
271 		iser_conn->ic_idmc = ic;
272 
273 		/*
274 		 * Set the local and remote addresses in the idm conn handle.
275 		 */
276 		iser_ib_conv_ibtaddr2sockaddr(&ic->ic_laddr,
277 		    &iser_conn->ic_chan->ic_localip, iser_chan->ic_lport);
278 		iser_ib_conv_ibtaddr2sockaddr(&ic->ic_raddr,
279 		    &iser_conn->ic_chan->ic_remoteip, iser_chan->ic_rport);
280 
281 		/*
282 		 * Kick the state machine.  At CS_S3_XPT_UP the state machine
283 		 * will notify the client (target) about the new connection.
284 		 */
285 		idm_conn_event(ic, CE_CONNECT_ACCEPT, NULL);
286 		iser_conn->ic_stage = ISER_CONN_STAGE_IC_CONNECTED;
287 		mutex_exit(&iser_conn->ic_lock);
288 
289 		/*
290 		 * Post work requests on the receive queue
291 		 */
292 		iser_ib_post_recv(iser_chan->ic_chanhdl);
293 
294 	}
295 
296 	return (IBT_CM_ACCEPT);
297 }
298 
299 static ibt_cm_status_t
300 iser_handle_cm_conn_closed(ibt_cm_event_t *evp)
301 {
302 
303 	iser_chan_t	*chan;
304 
305 	chan = (iser_chan_t *)ibt_get_chan_private(evp->cm_channel);
306 
307 	ISER_LOG(CE_NOTE, "iser_handle_cm_conn_closed: chan (0x%p) "
308 	    "reason (0x%x)", (void *)chan, evp->cm_event.closed);
309 
310 	switch (evp->cm_event.closed) {
311 	case IBT_CM_CLOSED_DREP_RCVD:	/* we requested a disconnect */
312 	case IBT_CM_CLOSED_ALREADY:	/* duplicate close */
313 		/* ignore these */
314 		return (IBT_CM_ACCEPT);
315 
316 	case IBT_CM_CLOSED_DREQ_RCVD:	/* request to close the channel */
317 	case IBT_CM_CLOSED_REJ_RCVD:	/* reject after conn establishment */
318 	case IBT_CM_CLOSED_DREQ_TIMEOUT: /* our close request timed out */
319 	case IBT_CM_CLOSED_DUP:		/* duplicate close request */
320 	case IBT_CM_CLOSED_ABORT:	/* aborted connection establishment */
321 	case IBT_CM_CLOSED_STALE:	/* stale / unref connection */
322 		/* handle these depending upon our connection state */
323 		mutex_enter(&chan->ic_conn->ic_lock);
324 		switch (chan->ic_conn->ic_stage) {
325 		case ISER_CONN_STAGE_UNDEFINED:
326 		case ISER_CONN_STAGE_CLOSED:
327 			/* do nothing, just drop the lock */
328 			mutex_exit(&chan->ic_conn->ic_lock);
329 			break;
330 
331 		case ISER_CONN_STAGE_ALLOCATED:
332 			/*
333 			 * We blew up or were offlined during connection
334 			 * establishment. Teardown the iSER conn and chan
335 			 * handles.
336 			 */
337 			mutex_exit(&chan->ic_conn->ic_lock);
338 			iser_internal_conn_destroy(chan->ic_conn);
339 			break;
340 
341 		case ISER_CONN_STAGE_IC_DISCONNECTED:
342 		case ISER_CONN_STAGE_IC_FREED:
343 		case ISER_CONN_STAGE_CLOSING:
344 			/* we're down, set CLOSED */
345 			chan->ic_conn->ic_stage = ISER_CONN_STAGE_CLOSED;
346 			mutex_exit(&chan->ic_conn->ic_lock);
347 			break;
348 
349 		case ISER_CONN_STAGE_IC_CONNECTED:
350 		case ISER_CONN_STAGE_HELLO_SENT:
351 		case ISER_CONN_STAGE_HELLO_SENT_FAIL:
352 		case ISER_CONN_STAGE_HELLO_WAIT:
353 		case ISER_CONN_STAGE_HELLO_RCV:
354 		case ISER_CONN_STAGE_HELLO_RCV_FAIL:
355 		case ISER_CONN_STAGE_HELLOREPLY_SENT:
356 		case ISER_CONN_STAGE_HELLOREPLY_SENT_FAIL:
357 		case ISER_CONN_STAGE_HELLOREPLY_RCV:
358 		case ISER_CONN_STAGE_HELLOREPLY_RCV_FAIL:
359 		case ISER_CONN_STAGE_LOGGED_IN:
360 			/* for all other stages, fail the transport */
361 			idm_conn_event(chan->ic_conn->ic_idmc,
362 			    CE_TRANSPORT_FAIL, IDM_STATUS_FAIL);
363 			chan->ic_conn->ic_stage = ISER_CONN_STAGE_CLOSING;
364 			mutex_exit(&chan->ic_conn->ic_lock);
365 			break;
366 
367 		default:
368 			mutex_exit(&chan->ic_conn->ic_lock);
369 			ASSERT(0);
370 
371 		}
372 
373 		/* accept the event */
374 		return (IBT_CM_ACCEPT);
375 
376 	default:
377 		/* unknown event */
378 		ISER_LOG(CE_NOTE, "iser_handle_cm_conn_closed: unknown closed "
379 		    "event: (0x%x)", evp->cm_event.closed);
380 		return (IBT_CM_REJECT);
381 	}
382 }
383 
384 /*
385  * Handle EVENT FAILURE
386  */
387 static ibt_cm_status_t
388 iser_handle_cm_event_failure(ibt_cm_event_t *evp)
389 {
390 	iser_chan_t	*chan;
391 
392 	chan = (iser_chan_t *)ibt_get_chan_private(evp->cm_channel);
393 
394 	ISER_LOG(CE_NOTE, "iser_handle_cm_event_failure: chan (0x%p): "
395 	    "code: %d msg: %d reason: %d", (void *)chan,
396 	    evp->cm_event.failed.cf_code, evp->cm_event.failed.cf_msg,
397 	    evp->cm_event.failed.cf_reason);
398 
399 	if ((evp->cm_channel == NULL) || (chan == NULL)) {
400 		/* channel not established yet */
401 		return (IBT_CM_ACCEPT);
402 	}
403 
404 	if ((evp->cm_event.failed.cf_code != IBT_CM_FAILURE_STALE) &&
405 	    (evp->cm_event.failed.cf_msg == IBT_CM_FAILURE_REQ)) {
406 		/*
407 		 * This end is active, just ignore, ibt_open_rc_channel()
408 		 * caller will take care of cleanup.
409 		 */
410 		return (IBT_CM_ACCEPT);
411 	}
412 
413 	/* handle depending upon our connection state */
414 	mutex_enter(&chan->ic_conn->ic_lock);
415 	switch (chan->ic_conn->ic_stage) {
416 	case ISER_CONN_STAGE_UNDEFINED:
417 	case ISER_CONN_STAGE_CLOSED:
418 		/* do nothing, just drop the lock */
419 		mutex_exit(&chan->ic_conn->ic_lock);
420 		break;
421 
422 	case ISER_CONN_STAGE_ALLOCATED:
423 		/*
424 		 * We blew up or were offlined during connection
425 		 * establishment. Teardown the iSER conn and chan
426 		 * handles.
427 		 */
428 		mutex_exit(&chan->ic_conn->ic_lock);
429 		iser_internal_conn_destroy(chan->ic_conn);
430 		break;
431 
432 	case ISER_CONN_STAGE_IC_DISCONNECTED:
433 	case ISER_CONN_STAGE_IC_FREED:
434 	case ISER_CONN_STAGE_CLOSING:
435 		/* update to CLOSED, then drop the lock */
436 		chan->ic_conn->ic_stage = ISER_CONN_STAGE_CLOSED;
437 		mutex_exit(&chan->ic_conn->ic_lock);
438 		break;
439 
440 	case ISER_CONN_STAGE_IC_CONNECTED:
441 	case ISER_CONN_STAGE_HELLO_SENT:
442 	case ISER_CONN_STAGE_HELLO_SENT_FAIL:
443 	case ISER_CONN_STAGE_HELLO_WAIT:
444 	case ISER_CONN_STAGE_HELLO_RCV:
445 	case ISER_CONN_STAGE_HELLO_RCV_FAIL:
446 	case ISER_CONN_STAGE_HELLOREPLY_SENT:
447 	case ISER_CONN_STAGE_HELLOREPLY_SENT_FAIL:
448 	case ISER_CONN_STAGE_HELLOREPLY_RCV:
449 	case ISER_CONN_STAGE_HELLOREPLY_RCV_FAIL:
450 	case ISER_CONN_STAGE_LOGGED_IN:
451 		/* fail the transport and move the conn to CLOSING */
452 		idm_conn_event(chan->ic_conn->ic_idmc, CE_TRANSPORT_FAIL,
453 		    IDM_STATUS_FAIL);
454 		chan->ic_conn->ic_stage = ISER_CONN_STAGE_CLOSING;
455 		mutex_exit(&chan->ic_conn->ic_lock);
456 		break;
457 
458 	default:
459 		mutex_exit(&chan->ic_conn->ic_lock);
460 		ASSERT(0);
461 	}
462 
463 	/* accept the event */
464 	return (IBT_CM_ACCEPT);
465 }
466