xref: /freebsd/sys/dev/iscsi/iscsi.c (revision a0ee8cc636cd5c2374ec44ca71226564ea0bca95)
1 /*-
2  * Copyright (c) 2012 The FreeBSD Foundation
3  * All rights reserved.
4  *
5  * This software was developed by Edward Tomasz Napierala under sponsorship
6  * from the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/condvar.h>
36 #include <sys/conf.h>
37 #include <sys/endian.h>
38 #include <sys/eventhandler.h>
39 #include <sys/file.h>
40 #include <sys/kernel.h>
41 #include <sys/kthread.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
45 #include <sys/module.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/sx.h>
49 #include <vm/uma.h>
50 
51 #include <cam/cam.h>
52 #include <cam/cam_ccb.h>
53 #include <cam/cam_xpt.h>
54 #include <cam/cam_debug.h>
55 #include <cam/cam_sim.h>
56 #include <cam/cam_xpt_sim.h>
57 #include <cam/cam_xpt_periph.h>
58 #include <cam/cam_periph.h>
59 #include <cam/scsi/scsi_all.h>
60 #include <cam/scsi/scsi_message.h>
61 
62 #include <dev/iscsi/icl.h>
63 #include <dev/iscsi/icl_wrappers.h>
64 #include <dev/iscsi/iscsi_ioctl.h>
65 #include <dev/iscsi/iscsi_proto.h>
66 #include <dev/iscsi/iscsi.h>
67 
68 #ifdef ICL_KERNEL_PROXY
69 #include <sys/socketvar.h>
70 #endif
71 
72 #ifdef ICL_KERNEL_PROXY
73 FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
74 #endif
75 
76 /*
77  * XXX: This is global so the iscsi_unload() can access it.
78  * 	Think about how to do this properly.
79  */
80 static struct iscsi_softc	*sc;
81 
82 SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD, 0, "iSCSI initiator");
83 static int debug = 1;
84 SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
85     &debug, 0, "Enable debug messages");
86 static int ping_timeout = 5;
87 SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
88     0, "Timeout for ping (NOP-Out) requests, in seconds");
89 static int iscsid_timeout = 60;
90 SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
91     0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
92 static int login_timeout = 60;
93 SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
94     0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
95 static int maxtags = 255;
96 SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
97     0, "Max number of IO requests queued");
98 static int fail_on_disconnection = 0;
99 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
100     &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
101 static int fail_on_shutdown = 1;
102 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_shutdown, CTLFLAG_RWTUN,
103     &fail_on_shutdown, 0, "Fail disconnected sessions on shutdown");
104 
105 static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
106 static uma_zone_t iscsi_outstanding_zone;
107 
108 #define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
109 #define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
110 
111 #define	ISCSI_DEBUG(X, ...)						\
112 	do {								\
113 		if (debug > 1) 						\
114 			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
115 	} while (0)
116 
117 #define	ISCSI_WARN(X, ...)						\
118 	do {								\
119 		if (debug > 0) {					\
120 			printf("WARNING: %s: " X "\n",			\
121 			    __func__, ## __VA_ARGS__);			\
122 		}							\
123 	} while (0)
124 
125 #define	ISCSI_SESSION_DEBUG(S, X, ...)					\
126 	do {								\
127 		if (debug > 1) {					\
128 			printf("%s: %s (%s): " X "\n",			\
129 			    __func__, S->is_conf.isc_target_addr,	\
130 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
131 		}							\
132 	} while (0)
133 
134 #define	ISCSI_SESSION_WARN(S, X, ...)					\
135 	do {								\
136 		if (debug > 0) {					\
137 			printf("WARNING: %s (%s): " X "\n",		\
138 			    S->is_conf.isc_target_addr,			\
139 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
140 		}							\
141 	} while (0)
142 
143 #define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
144 #define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
145 #define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
146 #define ISCSI_SESSION_LOCK_ASSERT_NOT(X) mtx_assert(&X->is_lock, MA_NOTOWNED)
147 
148 static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
149 		    int mode, struct thread *td);
150 
151 static struct cdevsw iscsi_cdevsw = {
152      .d_version = D_VERSION,
153      .d_ioctl   = iscsi_ioctl,
154      .d_name    = "iscsi",
155 };
156 
157 static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
158 static void	iscsi_pdu_queue(struct icl_pdu *request);
159 static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
160 static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
161 static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
162 static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
163 static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
164 static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
165 static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
166 static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
167 static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
168 static void	iscsi_session_reconnect(struct iscsi_session *is);
169 static void	iscsi_session_terminate(struct iscsi_session *is);
170 static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
171 static void	iscsi_poll(struct cam_sim *sim);
172 static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
173 		    uint32_t initiator_task_tag);
174 static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
175 		    union ccb *ccb, uint32_t *initiator_task_tagp);
176 static void	iscsi_outstanding_remove(struct iscsi_session *is,
177 		    struct iscsi_outstanding *io);
178 
179 static bool
180 iscsi_pdu_prepare(struct icl_pdu *request)
181 {
182 	struct iscsi_session *is;
183 	struct iscsi_bhs_scsi_command *bhssc;
184 
185 	is = PDU_SESSION(request);
186 
187 	ISCSI_SESSION_LOCK_ASSERT(is);
188 
189 	/*
190 	 * We're only using fields common for all the request
191 	 * (initiator -> target) PDUs.
192 	 */
193 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
194 
195 	/*
196 	 * Data-Out PDU does not contain CmdSN.
197 	 */
198 	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
199 		if (ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn) &&
200 		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
201 			/*
202 			 * Current MaxCmdSN prevents us from sending any more
203 			 * SCSI Command PDUs to the target; postpone the PDU.
204 			 * It will get resent by either iscsi_pdu_queue(),
205 			 * or by maintenance thread.
206 			 */
207 #if 0
208 			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %u, "
209 			    "ExpCmdSN %u, MaxCmdSN %u, opcode 0x%x",
210 			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn,
211 			    bhssc->bhssc_opcode);
212 #endif
213 			return (true);
214 		}
215 		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
216 		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
217 			is->is_cmdsn++;
218 	}
219 	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
220 
221 	return (false);
222 }
223 
224 static void
225 iscsi_session_send_postponed(struct iscsi_session *is)
226 {
227 	struct icl_pdu *request;
228 	bool postpone;
229 
230 	ISCSI_SESSION_LOCK_ASSERT(is);
231 
232 	while (!STAILQ_EMPTY(&is->is_postponed)) {
233 		request = STAILQ_FIRST(&is->is_postponed);
234 		postpone = iscsi_pdu_prepare(request);
235 		if (postpone)
236 			break;
237 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
238 		icl_pdu_queue(request);
239 	}
240 }
241 
242 static void
243 iscsi_pdu_queue_locked(struct icl_pdu *request)
244 {
245 	struct iscsi_session *is;
246 	bool postpone;
247 
248 	is = PDU_SESSION(request);
249 	ISCSI_SESSION_LOCK_ASSERT(is);
250 	iscsi_session_send_postponed(is);
251 	postpone = iscsi_pdu_prepare(request);
252 	if (postpone) {
253 		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
254 		return;
255 	}
256 	icl_pdu_queue(request);
257 }
258 
259 static void
260 iscsi_pdu_queue(struct icl_pdu *request)
261 {
262 	struct iscsi_session *is;
263 
264 	is = PDU_SESSION(request);
265 	ISCSI_SESSION_LOCK(is);
266 	iscsi_pdu_queue_locked(request);
267 	ISCSI_SESSION_UNLOCK(is);
268 }
269 
270 static void
271 iscsi_session_logout(struct iscsi_session *is)
272 {
273 	struct icl_pdu *request;
274 	struct iscsi_bhs_logout_request *bhslr;
275 
276 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
277 	if (request == NULL)
278 		return;
279 
280 	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
281 	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
282 	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
283 	iscsi_pdu_queue_locked(request);
284 }
285 
286 static void
287 iscsi_session_terminate_task(struct iscsi_session *is,
288     struct iscsi_outstanding *io, bool requeue)
289 {
290 
291 	if (io->io_ccb != NULL) {
292 		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
293 		if (requeue)
294 			io->io_ccb->ccb_h.status |= CAM_REQUEUE_REQ;
295 		else
296 			io->io_ccb->ccb_h.status |= CAM_REQ_ABORTED;
297 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
298 			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
299 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
300 			ISCSI_SESSION_DEBUG(is, "freezing devq");
301 		}
302 		xpt_done(io->io_ccb);
303 	}
304 	iscsi_outstanding_remove(is, io);
305 }
306 
307 static void
308 iscsi_session_terminate_tasks(struct iscsi_session *is, bool requeue)
309 {
310 	struct iscsi_outstanding *io, *tmp;
311 
312 	ISCSI_SESSION_LOCK_ASSERT(is);
313 
314 	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
315 		iscsi_session_terminate_task(is, io, requeue);
316 	}
317 }
318 
319 static void
320 iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
321 {
322 	struct icl_pdu *pdu;
323 
324 	ISCSI_SESSION_LOCK_ASSERT(is);
325 
326 	/*
327 	 * Don't queue any new PDUs.
328 	 */
329 	if (is->is_sim != NULL && is->is_simq_frozen == false) {
330 		ISCSI_SESSION_DEBUG(is, "freezing");
331 		xpt_freeze_simq(is->is_sim, 1);
332 		is->is_simq_frozen = true;
333 	}
334 
335 	/*
336 	 * Remove postponed PDUs.
337 	 */
338 	while (!STAILQ_EMPTY(&is->is_postponed)) {
339 		pdu = STAILQ_FIRST(&is->is_postponed);
340 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
341 		icl_pdu_free(pdu);
342 	}
343 
344 	if (destroy_sim == false) {
345 		/*
346 		 * Terminate SCSI tasks, asking CAM to requeue them.
347 		 */
348 		iscsi_session_terminate_tasks(is, true);
349 		return;
350 	}
351 
352 	iscsi_session_terminate_tasks(is, false);
353 
354 	if (is->is_sim == NULL)
355 		return;
356 
357 	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
358 	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
359 
360 	if (is->is_simq_frozen) {
361 		xpt_release_simq(is->is_sim, 1);
362 		is->is_simq_frozen = false;
363 	}
364 
365 	xpt_free_path(is->is_path);
366 	is->is_path = NULL;
367 	xpt_bus_deregister(cam_sim_path(is->is_sim));
368 	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
369 	is->is_sim = NULL;
370 	is->is_devq = NULL;
371 }
372 
373 static void
374 iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
375 {
376 
377 	icl_conn_close(is->is_conn);
378 
379 	ISCSI_SESSION_LOCK(is);
380 
381 	is->is_connected = false;
382 	is->is_reconnecting = false;
383 	is->is_login_phase = false;
384 
385 #ifdef ICL_KERNEL_PROXY
386 	if (is->is_login_pdu != NULL) {
387 		icl_pdu_free(is->is_login_pdu);
388 		is->is_login_pdu = NULL;
389 	}
390 	cv_signal(&is->is_login_cv);
391 #endif
392 
393 	if (fail_on_disconnection) {
394 		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
395 		iscsi_session_cleanup(is, true);
396 	} else {
397 		iscsi_session_cleanup(is, false);
398 	}
399 
400 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
401 	    ("destroying session with active tasks"));
402 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
403 	    ("destroying session with postponed PDUs"));
404 
405 	/*
406 	 * Request immediate reconnection from iscsid(8).
407 	 */
408 	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
409 	is->is_waiting_for_iscsid = true;
410 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
411 	is->is_timeout = 0;
412 	ISCSI_SESSION_UNLOCK(is);
413 	cv_signal(&is->is_softc->sc_cv);
414 }
415 
416 static void
417 iscsi_maintenance_thread_terminate(struct iscsi_session *is)
418 {
419 	struct iscsi_softc *sc;
420 
421 	sc = is->is_softc;
422 	sx_xlock(&sc->sc_lock);
423 
424 	icl_conn_close(is->is_conn);
425 	callout_drain(&is->is_callout);
426 
427 	ISCSI_SESSION_LOCK(is);
428 
429 	KASSERT(is->is_terminating, ("is_terminating == false"));
430 
431 #ifdef ICL_KERNEL_PROXY
432 	if (is->is_login_pdu != NULL) {
433 		icl_pdu_free(is->is_login_pdu);
434 		is->is_login_pdu = NULL;
435 	}
436 	cv_signal(&is->is_login_cv);
437 #endif
438 
439 	iscsi_session_cleanup(is, true);
440 
441 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
442 	    ("destroying session with active tasks"));
443 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
444 	    ("destroying session with postponed PDUs"));
445 
446 	ISCSI_SESSION_UNLOCK(is);
447 
448 	icl_conn_free(is->is_conn);
449 	mtx_destroy(&is->is_lock);
450 	cv_destroy(&is->is_maintenance_cv);
451 #ifdef ICL_KERNEL_PROXY
452 	cv_destroy(&is->is_login_cv);
453 #endif
454 	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
455 	sx_xunlock(&sc->sc_lock);
456 
457 	ISCSI_SESSION_DEBUG(is, "terminated");
458 	free(is, M_ISCSI);
459 
460 	/*
461 	 * The iscsi_unload() routine might be waiting.
462 	 */
463 	cv_signal(&sc->sc_cv);
464 }
465 
466 static void
467 iscsi_maintenance_thread(void *arg)
468 {
469 	struct iscsi_session *is;
470 
471 	is = arg;
472 
473 	for (;;) {
474 		ISCSI_SESSION_LOCK(is);
475 		if (is->is_reconnecting == false &&
476 		    is->is_terminating == false &&
477 		    STAILQ_EMPTY(&is->is_postponed))
478 			cv_wait(&is->is_maintenance_cv, &is->is_lock);
479 
480 		/* Terminate supersedes reconnect. */
481 		if (is->is_terminating) {
482 			ISCSI_SESSION_UNLOCK(is);
483 			iscsi_maintenance_thread_terminate(is);
484 			kthread_exit();
485 			return;
486 		}
487 
488 		if (is->is_reconnecting) {
489 			ISCSI_SESSION_UNLOCK(is);
490 			iscsi_maintenance_thread_reconnect(is);
491 			continue;
492 		}
493 
494 		iscsi_session_send_postponed(is);
495 		ISCSI_SESSION_UNLOCK(is);
496 	}
497 }
498 
499 static void
500 iscsi_session_reconnect(struct iscsi_session *is)
501 {
502 
503 	/*
504 	 * XXX: We can't use locking here, because
505 	 * 	it's being called from various contexts.
506 	 * 	Hope it doesn't break anything.
507 	 */
508 	if (is->is_reconnecting)
509 		return;
510 
511 	is->is_reconnecting = true;
512 	cv_signal(&is->is_maintenance_cv);
513 }
514 
515 static void
516 iscsi_session_terminate(struct iscsi_session *is)
517 {
518 
519 	if (is->is_terminating)
520 		return;
521 
522 	is->is_terminating = true;
523 
524 #if 0
525 	iscsi_session_logout(is);
526 #endif
527 	cv_signal(&is->is_maintenance_cv);
528 }
529 
530 static void
531 iscsi_callout(void *context)
532 {
533 	struct icl_pdu *request;
534 	struct iscsi_bhs_nop_out *bhsno;
535 	struct iscsi_session *is;
536 	bool reconnect_needed = false;
537 
538 	is = context;
539 
540 	ISCSI_SESSION_LOCK(is);
541 	if (is->is_terminating) {
542 		ISCSI_SESSION_UNLOCK(is);
543 		return;
544 	}
545 
546 	callout_schedule(&is->is_callout, 1 * hz);
547 
548 	is->is_timeout++;
549 
550 	if (is->is_waiting_for_iscsid) {
551 		if (iscsid_timeout > 0 && is->is_timeout > iscsid_timeout) {
552 			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
553 			    "for %d seconds; reconnecting",
554 			    is->is_timeout);
555 			reconnect_needed = true;
556 		}
557 		goto out;
558 	}
559 
560 	if (is->is_login_phase) {
561 		if (login_timeout > 0 && is->is_timeout > login_timeout) {
562 			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
563 			    "reconnecting", is->is_timeout);
564 			reconnect_needed = true;
565 		}
566 		goto out;
567 	}
568 
569 	if (ping_timeout <= 0) {
570 		/*
571 		 * Pings are disabled.  Don't send NOP-Out in this case.
572 		 * Reset the timeout, to avoid triggering reconnection,
573 		 * should the user decide to reenable them.
574 		 */
575 		is->is_timeout = 0;
576 		goto out;
577 	}
578 
579 	if (is->is_timeout >= ping_timeout) {
580 		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
581 		    "reconnecting", ping_timeout);
582 		reconnect_needed = true;
583 		goto out;
584 	}
585 
586 	ISCSI_SESSION_UNLOCK(is);
587 
588 	/*
589 	 * If the ping was reset less than one second ago - which means
590 	 * that we've received some PDU during the last second - assume
591 	 * the traffic flows correctly and don't bother sending a NOP-Out.
592 	 *
593 	 * (It's 2 - one for one second, and one for incrementing is_timeout
594 	 * earlier in this routine.)
595 	 */
596 	if (is->is_timeout < 2)
597 		return;
598 
599 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
600 	if (request == NULL) {
601 		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
602 		return;
603 	}
604 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
605 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
606 	    ISCSI_BHS_OPCODE_IMMEDIATE;
607 	bhsno->bhsno_flags = 0x80;
608 	bhsno->bhsno_target_transfer_tag = 0xffffffff;
609 	iscsi_pdu_queue(request);
610 	return;
611 
612 out:
613 	if (is->is_terminating) {
614 		ISCSI_SESSION_UNLOCK(is);
615 		return;
616 	}
617 
618 	ISCSI_SESSION_UNLOCK(is);
619 
620 	if (reconnect_needed)
621 		iscsi_session_reconnect(is);
622 }
623 
624 static void
625 iscsi_pdu_update_statsn(const struct icl_pdu *response)
626 {
627 	const struct iscsi_bhs_data_in *bhsdi;
628 	struct iscsi_session *is;
629 	uint32_t expcmdsn, maxcmdsn, statsn;
630 
631 	is = PDU_SESSION(response);
632 
633 	ISCSI_SESSION_LOCK_ASSERT(is);
634 
635 	/*
636 	 * We're only using fields common for all the response
637 	 * (target -> initiator) PDUs.
638 	 */
639 	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
640 	/*
641 	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
642 	 * and Residual Count only have meaningful content if the S bit
643 	 * is set to 1", so we also need to check the bit specific for
644 	 * Data-In PDU.
645 	 */
646 	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
647 	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
648 		statsn = ntohl(bhsdi->bhsdi_statsn);
649 		if (statsn != is->is_statsn && statsn != (is->is_statsn + 1)) {
650 			/* XXX: This is normal situation for MCS */
651 			ISCSI_SESSION_WARN(is, "PDU 0x%x StatSN %u != "
652 			    "session ExpStatSN %u (or + 1); reconnecting",
653 			    bhsdi->bhsdi_opcode, statsn, is->is_statsn);
654 			iscsi_session_reconnect(is);
655 		}
656 		if (ISCSI_SNGT(statsn, is->is_statsn))
657 			is->is_statsn = statsn;
658 	}
659 
660 	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
661 	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
662 
663 	if (ISCSI_SNLT(maxcmdsn + 1, expcmdsn)) {
664 		ISCSI_SESSION_DEBUG(is,
665 		    "PDU MaxCmdSN %u + 1 < PDU ExpCmdSN %u; ignoring",
666 		    maxcmdsn, expcmdsn);
667 	} else {
668 		if (ISCSI_SNGT(maxcmdsn, is->is_maxcmdsn)) {
669 			is->is_maxcmdsn = maxcmdsn;
670 
671 			/*
672 			 * Command window increased; kick the maintanance thread
673 			 * to send out postponed commands.
674 			 */
675 			if (!STAILQ_EMPTY(&is->is_postponed))
676 				cv_signal(&is->is_maintenance_cv);
677 		} else if (ISCSI_SNLT(maxcmdsn, is->is_maxcmdsn)) {
678 			/* XXX: This is normal situation for MCS */
679 			ISCSI_SESSION_DEBUG(is,
680 			    "PDU MaxCmdSN %u < session MaxCmdSN %u; ignoring",
681 			    maxcmdsn, is->is_maxcmdsn);
682 		}
683 
684 		if (ISCSI_SNGT(expcmdsn, is->is_expcmdsn)) {
685 			is->is_expcmdsn = expcmdsn;
686 		} else if (ISCSI_SNLT(expcmdsn, is->is_expcmdsn)) {
687 			/* XXX: This is normal situation for MCS */
688 			ISCSI_SESSION_DEBUG(is,
689 			    "PDU ExpCmdSN %u < session ExpCmdSN %u; ignoring",
690 			    expcmdsn, is->is_expcmdsn);
691 		}
692 	}
693 
694 	/*
695 	 * Every incoming PDU - not just NOP-In - resets the ping timer.
696 	 * The purpose of the timeout is to reset the connection when it stalls;
697 	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
698 	 * in some queue.
699 	 */
700 	is->is_timeout = 0;
701 }
702 
703 static void
704 iscsi_receive_callback(struct icl_pdu *response)
705 {
706 	struct iscsi_session *is;
707 
708 	is = PDU_SESSION(response);
709 
710 	ISCSI_SESSION_LOCK(is);
711 
712 #ifdef ICL_KERNEL_PROXY
713 	if (is->is_login_phase) {
714 		if (is->is_login_pdu == NULL)
715 			is->is_login_pdu = response;
716 		else
717 			icl_pdu_free(response);
718 		ISCSI_SESSION_UNLOCK(is);
719 		cv_signal(&is->is_login_cv);
720 		return;
721 	}
722 #endif
723 
724 	iscsi_pdu_update_statsn(response);
725 
726 	/*
727 	 * The handling routine is responsible for freeing the PDU
728 	 * when it's no longer needed.
729 	 */
730 	switch (response->ip_bhs->bhs_opcode) {
731 	case ISCSI_BHS_OPCODE_NOP_IN:
732 		iscsi_pdu_handle_nop_in(response);
733 		ISCSI_SESSION_UNLOCK(is);
734 		break;
735 	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
736 		iscsi_pdu_handle_scsi_response(response);
737 		/* Session lock dropped inside. */
738 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
739 		break;
740 	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
741 		iscsi_pdu_handle_task_response(response);
742 		ISCSI_SESSION_UNLOCK(is);
743 		break;
744 	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
745 		iscsi_pdu_handle_data_in(response);
746 		/* Session lock dropped inside. */
747 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
748 		break;
749 	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
750 		iscsi_pdu_handle_logout_response(response);
751 		ISCSI_SESSION_UNLOCK(is);
752 		break;
753 	case ISCSI_BHS_OPCODE_R2T:
754 		iscsi_pdu_handle_r2t(response);
755 		ISCSI_SESSION_UNLOCK(is);
756 		break;
757 	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
758 		iscsi_pdu_handle_async_message(response);
759 		ISCSI_SESSION_UNLOCK(is);
760 		break;
761 	case ISCSI_BHS_OPCODE_REJECT:
762 		iscsi_pdu_handle_reject(response);
763 		ISCSI_SESSION_UNLOCK(is);
764 		break;
765 	default:
766 		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
767 		    "opcode 0x%x; reconnecting",
768 		    response->ip_bhs->bhs_opcode);
769 		iscsi_session_reconnect(is);
770 		ISCSI_SESSION_UNLOCK(is);
771 		icl_pdu_free(response);
772 	}
773 }
774 
775 static void
776 iscsi_error_callback(struct icl_conn *ic)
777 {
778 	struct iscsi_session *is;
779 
780 	is = CONN_SESSION(ic);
781 
782 	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
783 	iscsi_session_reconnect(is);
784 }
785 
786 static void
787 iscsi_pdu_handle_nop_in(struct icl_pdu *response)
788 {
789 	struct iscsi_session *is;
790 	struct iscsi_bhs_nop_out *bhsno;
791 	struct iscsi_bhs_nop_in *bhsni;
792 	struct icl_pdu *request;
793 	void *data = NULL;
794 	size_t datasize;
795 	int error;
796 
797 	is = PDU_SESSION(response);
798 	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
799 
800 	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
801 		/*
802 		 * Nothing to do; iscsi_pdu_update_statsn() already
803 		 * zeroed the timeout.
804 		 */
805 		icl_pdu_free(response);
806 		return;
807 	}
808 
809 	datasize = icl_pdu_data_segment_length(response);
810 	if (datasize > 0) {
811 		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
812 		if (data == NULL) {
813 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
814 			    "reconnecting");
815 			icl_pdu_free(response);
816 			iscsi_session_reconnect(is);
817 			return;
818 		}
819 		icl_pdu_get_data(response, 0, data, datasize);
820 	}
821 
822 	request = icl_pdu_new(response->ip_conn, M_NOWAIT);
823 	if (request == NULL) {
824 		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
825 		    "reconnecting");
826 		free(data, M_ISCSI);
827 		icl_pdu_free(response);
828 		iscsi_session_reconnect(is);
829 		return;
830 	}
831 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
832 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
833 	    ISCSI_BHS_OPCODE_IMMEDIATE;
834 	bhsno->bhsno_flags = 0x80;
835 	bhsno->bhsno_initiator_task_tag = 0xffffffff;
836 	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
837 	if (datasize > 0) {
838 		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
839 		if (error != 0) {
840 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
841 			    "reconnecting");
842 			free(data, M_ISCSI);
843 			icl_pdu_free(request);
844 			icl_pdu_free(response);
845 			iscsi_session_reconnect(is);
846 			return;
847 		}
848 		free(data, M_ISCSI);
849 	}
850 
851 	icl_pdu_free(response);
852 	iscsi_pdu_queue_locked(request);
853 }
854 
855 static void
856 iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
857 {
858 	struct iscsi_bhs_scsi_response *bhssr;
859 	struct iscsi_outstanding *io;
860 	struct iscsi_session *is;
861 	union ccb *ccb;
862 	struct ccb_scsiio *csio;
863 	size_t data_segment_len, received;
864 	uint16_t sense_len;
865 
866 	is = PDU_SESSION(response);
867 
868 	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
869 	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
870 	if (io == NULL || io->io_ccb == NULL) {
871 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
872 		icl_pdu_free(response);
873 		iscsi_session_reconnect(is);
874 		ISCSI_SESSION_UNLOCK(is);
875 		return;
876 	}
877 
878 	ccb = io->io_ccb;
879 	received = io->io_received;
880 	iscsi_outstanding_remove(is, io);
881 	ISCSI_SESSION_UNLOCK(is);
882 
883 	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
884 		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
885  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
886  			xpt_freeze_devq(ccb->ccb_h.path, 1);
887 			ISCSI_SESSION_DEBUG(is, "freezing devq");
888 		}
889  		ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
890 	} else if (bhssr->bhssr_status == 0) {
891 		ccb->ccb_h.status = CAM_REQ_CMP;
892 	} else {
893  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
894  			xpt_freeze_devq(ccb->ccb_h.path, 1);
895 			ISCSI_SESSION_DEBUG(is, "freezing devq");
896 		}
897  		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
898 		ccb->csio.scsi_status = bhssr->bhssr_status;
899 	}
900 
901 	csio = &ccb->csio;
902 	data_segment_len = icl_pdu_data_segment_length(response);
903 	if (data_segment_len > 0) {
904 		if (data_segment_len < sizeof(sense_len)) {
905 			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
906 			    data_segment_len);
907 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
908 				xpt_freeze_devq(ccb->ccb_h.path, 1);
909 				ISCSI_SESSION_DEBUG(is, "freezing devq");
910 			}
911 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
912 			goto out;
913 		}
914 		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
915 		sense_len = ntohs(sense_len);
916 #if 0
917 		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
918 		    sense_len, data_segment_len);
919 #endif
920 		if (sizeof(sense_len) + sense_len > data_segment_len) {
921 			ISCSI_SESSION_WARN(is, "truncated data segment "
922 			    "(%zd bytes, should be %zd)",
923 			    data_segment_len, sizeof(sense_len) + sense_len);
924 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
925 				xpt_freeze_devq(ccb->ccb_h.path, 1);
926 				ISCSI_SESSION_DEBUG(is, "freezing devq");
927 			}
928 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
929 			goto out;
930 		} else if (sizeof(sense_len) + sense_len < data_segment_len)
931 			ISCSI_SESSION_WARN(is, "oversize data segment "
932 			    "(%zd bytes, should be %zd)",
933 			    data_segment_len, sizeof(sense_len) + sense_len);
934 		if (sense_len > csio->sense_len) {
935 			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
936 			    sense_len, csio->sense_len);
937 			sense_len = csio->sense_len;
938 		}
939 		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
940 		csio->sense_resid = csio->sense_len - sense_len;
941 		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
942 	}
943 
944 out:
945 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
946 		csio->resid = ntohl(bhssr->bhssr_residual_count);
947 
948 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
949 		KASSERT(received <= csio->dxfer_len,
950 		    ("received > csio->dxfer_len"));
951 		if (received < csio->dxfer_len) {
952 			if (csio->resid != csio->dxfer_len - received) {
953 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
954 				    "target indicates %d, we calculated %zd",
955 				    csio->resid, csio->dxfer_len - received);
956 			}
957 			csio->resid = csio->dxfer_len - received;
958 		}
959 	}
960 
961 	xpt_done(ccb);
962 	icl_pdu_free(response);
963 }
964 
965 static void
966 iscsi_pdu_handle_task_response(struct icl_pdu *response)
967 {
968 	struct iscsi_bhs_task_management_response *bhstmr;
969 	struct iscsi_outstanding *io, *aio;
970 	struct iscsi_session *is;
971 
972 	is = PDU_SESSION(response);
973 
974 	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
975 	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
976 	if (io == NULL || io->io_ccb != NULL) {
977 		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
978 		    bhstmr->bhstmr_initiator_task_tag);
979 		icl_pdu_free(response);
980 		iscsi_session_reconnect(is);
981 		return;
982 	}
983 
984 	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
985 		ISCSI_SESSION_WARN(is, "task response 0x%x",
986 		    bhstmr->bhstmr_response);
987 	} else {
988 		aio = iscsi_outstanding_find(is, io->io_datasn);
989 		if (aio != NULL && aio->io_ccb != NULL)
990 			iscsi_session_terminate_task(is, aio, false);
991 	}
992 
993 	iscsi_outstanding_remove(is, io);
994 	icl_pdu_free(response);
995 }
996 
997 static void
998 iscsi_pdu_handle_data_in(struct icl_pdu *response)
999 {
1000 	struct iscsi_bhs_data_in *bhsdi;
1001 	struct iscsi_outstanding *io;
1002 	struct iscsi_session *is;
1003 	union ccb *ccb;
1004 	struct ccb_scsiio *csio;
1005 	size_t data_segment_len, received, oreceived;
1006 
1007 	is = PDU_SESSION(response);
1008 	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
1009 	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
1010 	if (io == NULL || io->io_ccb == NULL) {
1011 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
1012 		icl_pdu_free(response);
1013 		iscsi_session_reconnect(is);
1014 		ISCSI_SESSION_UNLOCK(is);
1015 		return;
1016 	}
1017 
1018 	data_segment_len = icl_pdu_data_segment_length(response);
1019 	if (data_segment_len == 0) {
1020 		/*
1021 		 * "The sending of 0 length data segments should be avoided,
1022 		 * but initiators and targets MUST be able to properly receive
1023 		 * 0 length data segments."
1024 		 */
1025 		ISCSI_SESSION_UNLOCK(is);
1026 		icl_pdu_free(response);
1027 		return;
1028 	}
1029 
1030 	/*
1031 	 * We need to track this for security reasons - without it, malicious target
1032 	 * could respond to SCSI READ without sending Data-In PDUs, which would result
1033 	 * in read operation on the initiator side returning random kernel data.
1034 	 */
1035 	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1036 		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1037 		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1038 		icl_pdu_free(response);
1039 		iscsi_session_reconnect(is);
1040 		ISCSI_SESSION_UNLOCK(is);
1041 		return;
1042 	}
1043 
1044 	ccb = io->io_ccb;
1045 	csio = &ccb->csio;
1046 
1047 	if (io->io_received + data_segment_len > csio->dxfer_len) {
1048 		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1049 		    "at offset %zd, buffer is %d)",
1050 		    data_segment_len, io->io_received, csio->dxfer_len);
1051 		icl_pdu_free(response);
1052 		iscsi_session_reconnect(is);
1053 		ISCSI_SESSION_UNLOCK(is);
1054 		return;
1055 	}
1056 
1057 	oreceived = io->io_received;
1058 	io->io_received += data_segment_len;
1059 	received = io->io_received;
1060 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0)
1061 		iscsi_outstanding_remove(is, io);
1062 	ISCSI_SESSION_UNLOCK(is);
1063 
1064 	icl_pdu_get_data(response, 0, csio->data_ptr + oreceived, data_segment_len);
1065 
1066 	/*
1067 	 * XXX: Check DataSN.
1068 	 * XXX: Check F.
1069 	 */
1070 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1071 		/*
1072 		 * Nothing more to do.
1073 		 */
1074 		icl_pdu_free(response);
1075 		return;
1076 	}
1077 
1078 	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1079 	if (bhsdi->bhsdi_status == 0) {
1080 		ccb->ccb_h.status = CAM_REQ_CMP;
1081 	} else {
1082 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1083 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1084 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1085 		}
1086 		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1087 		csio->scsi_status = bhsdi->bhsdi_status;
1088 	}
1089 
1090 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1091 		KASSERT(received <= csio->dxfer_len,
1092 		    ("received > csio->dxfer_len"));
1093 		if (received < csio->dxfer_len) {
1094 			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1095 			if (csio->resid != csio->dxfer_len - received) {
1096 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1097 				    "target indicates %d, we calculated %zd",
1098 				    csio->resid, csio->dxfer_len - received);
1099 			}
1100 			csio->resid = csio->dxfer_len - received;
1101 		}
1102 	}
1103 
1104 	xpt_done(ccb);
1105 	icl_pdu_free(response);
1106 }
1107 
1108 static void
1109 iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1110 {
1111 
1112 	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1113 	icl_pdu_free(response);
1114 }
1115 
1116 static void
1117 iscsi_pdu_handle_r2t(struct icl_pdu *response)
1118 {
1119 	struct icl_pdu *request;
1120 	struct iscsi_session *is;
1121 	struct iscsi_bhs_r2t *bhsr2t;
1122 	struct iscsi_bhs_data_out *bhsdo;
1123 	struct iscsi_outstanding *io;
1124 	struct ccb_scsiio *csio;
1125 	size_t off, len, total_len;
1126 	int error;
1127 
1128 	is = PDU_SESSION(response);
1129 
1130 	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1131 	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1132 	if (io == NULL || io->io_ccb == NULL) {
1133 		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1134 		    bhsr2t->bhsr2t_initiator_task_tag);
1135 		icl_pdu_free(response);
1136 		iscsi_session_reconnect(is);
1137 		return;
1138 	}
1139 
1140 	csio = &io->io_ccb->csio;
1141 
1142 	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1143 		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1144 		icl_pdu_free(response);
1145 		iscsi_session_reconnect(is);
1146 		return;
1147 	}
1148 
1149 	/*
1150 	 * XXX: Verify R2TSN.
1151 	 */
1152 
1153 	io->io_datasn = 0;
1154 
1155 	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1156 	if (off > csio->dxfer_len) {
1157 		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1158 		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1159 		icl_pdu_free(response);
1160 		iscsi_session_reconnect(is);
1161 		return;
1162 	}
1163 
1164 	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1165 	if (total_len == 0 || total_len > csio->dxfer_len) {
1166 		ISCSI_SESSION_WARN(is, "target requested invalid length "
1167 		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1168 		icl_pdu_free(response);
1169 		iscsi_session_reconnect(is);
1170 		return;
1171 	}
1172 
1173 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1174 
1175 	for (;;) {
1176 		len = total_len;
1177 
1178 		if (len > is->is_max_data_segment_length)
1179 			len = is->is_max_data_segment_length;
1180 
1181 		if (off + len > csio->dxfer_len) {
1182 			ISCSI_SESSION_WARN(is, "target requested invalid "
1183 			    "length/offset %zd, buffer is %d; reconnecting",
1184 			    off + len, csio->dxfer_len);
1185 			icl_pdu_free(response);
1186 			iscsi_session_reconnect(is);
1187 			return;
1188 		}
1189 
1190 		request = icl_pdu_new(response->ip_conn, M_NOWAIT);
1191 		if (request == NULL) {
1192 			icl_pdu_free(response);
1193 			iscsi_session_reconnect(is);
1194 			return;
1195 		}
1196 
1197 		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1198 		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1199 		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1200 		bhsdo->bhsdo_initiator_task_tag =
1201 		    bhsr2t->bhsr2t_initiator_task_tag;
1202 		bhsdo->bhsdo_target_transfer_tag =
1203 		    bhsr2t->bhsr2t_target_transfer_tag;
1204 		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1205 		bhsdo->bhsdo_buffer_offset = htonl(off);
1206 		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1207 		    M_NOWAIT);
1208 		if (error != 0) {
1209 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1210 			    "reconnecting");
1211 			icl_pdu_free(request);
1212 			icl_pdu_free(response);
1213 			iscsi_session_reconnect(is);
1214 			return;
1215 		}
1216 
1217 		off += len;
1218 		total_len -= len;
1219 
1220 		if (total_len == 0) {
1221 			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1222 			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1223 		} else {
1224 			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1225 		}
1226 
1227 		iscsi_pdu_queue_locked(request);
1228 
1229 		if (total_len == 0)
1230 			break;
1231 	}
1232 
1233 	icl_pdu_free(response);
1234 }
1235 
1236 static void
1237 iscsi_pdu_handle_async_message(struct icl_pdu *response)
1238 {
1239 	struct iscsi_bhs_asynchronous_message *bhsam;
1240 	struct iscsi_session *is;
1241 
1242 	is = PDU_SESSION(response);
1243 	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1244 	switch (bhsam->bhsam_async_event) {
1245 	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1246 		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1247 		iscsi_session_logout(is);
1248 		iscsi_session_terminate(is);
1249 		break;
1250 	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1251 		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the connection");
1252 		break;
1253 	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1254 		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the session");
1255 		break;
1256 	default:
1257 		/*
1258 		 * XXX: Technically, we're obligated to also handle
1259 		 * 	parameter renegotiation.
1260 		 */
1261 		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1262 		break;
1263 	}
1264 
1265 	icl_pdu_free(response);
1266 }
1267 
1268 static void
1269 iscsi_pdu_handle_reject(struct icl_pdu *response)
1270 {
1271 	struct iscsi_bhs_reject *bhsr;
1272 	struct iscsi_session *is;
1273 
1274 	is = PDU_SESSION(response);
1275 	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1276 	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1277 	    bhsr->bhsr_reason);
1278 
1279 	icl_pdu_free(response);
1280 }
1281 
1282 static int
1283 iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1284     struct iscsi_daemon_request *request)
1285 {
1286 	struct iscsi_session *is;
1287 	int error;
1288 
1289 	sx_slock(&sc->sc_lock);
1290 	for (;;) {
1291 		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1292 			ISCSI_SESSION_LOCK(is);
1293 			if (is->is_waiting_for_iscsid)
1294 				break;
1295 			ISCSI_SESSION_UNLOCK(is);
1296 		}
1297 
1298 		if (is == NULL) {
1299 			/*
1300 			 * No session requires attention from iscsid(8); wait.
1301 			 */
1302 			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1303 			if (error != 0) {
1304 				sx_sunlock(&sc->sc_lock);
1305 				return (error);
1306 			}
1307 			continue;
1308 		}
1309 
1310 		is->is_waiting_for_iscsid = false;
1311 		is->is_login_phase = true;
1312 		is->is_reason[0] = '\0';
1313 		ISCSI_SESSION_UNLOCK(is);
1314 
1315 		request->idr_session_id = is->is_id;
1316 		memcpy(&request->idr_isid, &is->is_isid,
1317 		    sizeof(request->idr_isid));
1318 		request->idr_tsih = 0;	/* New or reinstated session. */
1319 		memcpy(&request->idr_conf, &is->is_conf,
1320 		    sizeof(request->idr_conf));
1321 
1322 		error = icl_limits(is->is_conf.isc_offload,
1323 		    &request->idr_limits.isl_max_data_segment_length);
1324 		if (error != 0) {
1325 			ISCSI_SESSION_WARN(is, "icl_limits for offload \"%s\" "
1326 			    "failed with error %d", is->is_conf.isc_offload,
1327 			    error);
1328 			sx_sunlock(&sc->sc_lock);
1329 			return (error);
1330 		}
1331 
1332 		sx_sunlock(&sc->sc_lock);
1333 		return (0);
1334 	}
1335 }
1336 
1337 static int
1338 iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1339     struct iscsi_daemon_handoff *handoff)
1340 {
1341 	struct iscsi_session *is;
1342 	struct icl_conn *ic;
1343 	int error;
1344 
1345 	sx_slock(&sc->sc_lock);
1346 
1347 	/*
1348 	 * Find the session to hand off socket to.
1349 	 */
1350 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1351 		if (is->is_id == handoff->idh_session_id)
1352 			break;
1353 	}
1354 	if (is == NULL) {
1355 		sx_sunlock(&sc->sc_lock);
1356 		return (ESRCH);
1357 	}
1358 	ISCSI_SESSION_LOCK(is);
1359 	ic = is->is_conn;
1360 	if (is->is_conf.isc_discovery || is->is_terminating) {
1361 		ISCSI_SESSION_UNLOCK(is);
1362 		sx_sunlock(&sc->sc_lock);
1363 		return (EINVAL);
1364 	}
1365 	if (is->is_connected) {
1366 		/*
1367 		 * This might have happened because another iscsid(8)
1368 		 * instance handed off the connection in the meantime.
1369 		 * Just return.
1370 		 */
1371 		ISCSI_SESSION_WARN(is, "handoff on already connected "
1372 		    "session");
1373 		ISCSI_SESSION_UNLOCK(is);
1374 		sx_sunlock(&sc->sc_lock);
1375 		return (EBUSY);
1376 	}
1377 
1378 	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1379 	    sizeof(is->is_target_alias));
1380 	is->is_tsih = handoff->idh_tsih;
1381 	is->is_statsn = handoff->idh_statsn;
1382 	is->is_initial_r2t = handoff->idh_initial_r2t;
1383 	is->is_immediate_data = handoff->idh_immediate_data;
1384 
1385 	/*
1386 	 * Cap MaxRecvDataSegmentLength obtained from the target to the maximum
1387 	 * size supported by our ICL module.
1388 	 */
1389 	is->is_max_data_segment_length = min(ic->ic_max_data_segment_length,
1390 	    handoff->idh_max_data_segment_length);
1391 	is->is_max_burst_length = handoff->idh_max_burst_length;
1392 	is->is_first_burst_length = handoff->idh_first_burst_length;
1393 
1394 	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1395 		ic->ic_header_crc32c = true;
1396 	else
1397 		ic->ic_header_crc32c = false;
1398 	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1399 		ic->ic_data_crc32c = true;
1400 	else
1401 		ic->ic_data_crc32c = false;
1402 
1403 	is->is_cmdsn = 0;
1404 	is->is_expcmdsn = 0;
1405 	is->is_maxcmdsn = 0;
1406 	is->is_waiting_for_iscsid = false;
1407 	is->is_login_phase = false;
1408 	is->is_timeout = 0;
1409 	is->is_connected = true;
1410 	is->is_reason[0] = '\0';
1411 
1412 	ISCSI_SESSION_UNLOCK(is);
1413 
1414 #ifdef ICL_KERNEL_PROXY
1415 	if (handoff->idh_socket != 0) {
1416 #endif
1417 		/*
1418 		 * Handoff without using ICL proxy.
1419 		 */
1420 		error = icl_conn_handoff(ic, handoff->idh_socket);
1421 		if (error != 0) {
1422 			sx_sunlock(&sc->sc_lock);
1423 			iscsi_session_terminate(is);
1424 			return (error);
1425 		}
1426 #ifdef ICL_KERNEL_PROXY
1427 	}
1428 #endif
1429 
1430 	sx_sunlock(&sc->sc_lock);
1431 
1432 	if (is->is_sim != NULL) {
1433 		/*
1434 		 * When reconnecting, there already is SIM allocated for the session.
1435 		 */
1436 		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1437 		ISCSI_SESSION_LOCK(is);
1438 		ISCSI_SESSION_DEBUG(is, "releasing");
1439 		xpt_release_simq(is->is_sim, 1);
1440 		is->is_simq_frozen = false;
1441 		ISCSI_SESSION_UNLOCK(is);
1442 
1443 	} else {
1444 		ISCSI_SESSION_LOCK(is);
1445 		is->is_devq = cam_simq_alloc(maxtags);
1446 		if (is->is_devq == NULL) {
1447 			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1448 			iscsi_session_terminate(is);
1449 			return (ENOMEM);
1450 		}
1451 
1452 		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1453 		    is, is->is_id /* unit */, &is->is_lock,
1454 		    1, maxtags, is->is_devq);
1455 		if (is->is_sim == NULL) {
1456 			ISCSI_SESSION_UNLOCK(is);
1457 			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1458 			cam_simq_free(is->is_devq);
1459 			iscsi_session_terminate(is);
1460 			return (ENOMEM);
1461 		}
1462 
1463 		error = xpt_bus_register(is->is_sim, NULL, 0);
1464 		if (error != 0) {
1465 			ISCSI_SESSION_UNLOCK(is);
1466 			ISCSI_SESSION_WARN(is, "failed to register bus");
1467 			iscsi_session_terminate(is);
1468 			return (ENOMEM);
1469 		}
1470 
1471 		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1472 		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1473 		    CAM_LUN_WILDCARD);
1474 		if (error != CAM_REQ_CMP) {
1475 			ISCSI_SESSION_UNLOCK(is);
1476 			ISCSI_SESSION_WARN(is, "failed to create path");
1477 			iscsi_session_terminate(is);
1478 			return (ENOMEM);
1479 		}
1480 		ISCSI_SESSION_UNLOCK(is);
1481 	}
1482 
1483 	return (0);
1484 }
1485 
1486 static int
1487 iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1488     struct iscsi_daemon_fail *fail)
1489 {
1490 	struct iscsi_session *is;
1491 
1492 	sx_slock(&sc->sc_lock);
1493 
1494 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1495 		if (is->is_id == fail->idf_session_id)
1496 			break;
1497 	}
1498 	if (is == NULL) {
1499 		sx_sunlock(&sc->sc_lock);
1500 		return (ESRCH);
1501 	}
1502 	ISCSI_SESSION_LOCK(is);
1503 	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1504 	    fail->idf_reason);
1505 	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1506 	//is->is_waiting_for_iscsid = false;
1507 	//is->is_login_phase = true;
1508 	//iscsi_session_reconnect(is);
1509 	ISCSI_SESSION_UNLOCK(is);
1510 	sx_sunlock(&sc->sc_lock);
1511 
1512 	return (0);
1513 }
1514 
1515 #ifdef ICL_KERNEL_PROXY
1516 static int
1517 iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1518     struct iscsi_daemon_connect *idc)
1519 {
1520 	struct iscsi_session *is;
1521 	struct sockaddr *from_sa, *to_sa;
1522 	int error;
1523 
1524 	sx_slock(&sc->sc_lock);
1525 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1526 		if (is->is_id == idc->idc_session_id)
1527 			break;
1528 	}
1529 	if (is == NULL) {
1530 		sx_sunlock(&sc->sc_lock);
1531 		return (ESRCH);
1532 	}
1533 	sx_sunlock(&sc->sc_lock);
1534 
1535 	if (idc->idc_from_addrlen > 0) {
1536 		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1537 		if (error != 0) {
1538 			ISCSI_SESSION_WARN(is,
1539 			    "getsockaddr failed with error %d", error);
1540 			return (error);
1541 		}
1542 	} else {
1543 		from_sa = NULL;
1544 	}
1545 	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1546 	if (error != 0) {
1547 		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1548 		    error);
1549 		free(from_sa, M_SONAME);
1550 		return (error);
1551 	}
1552 
1553 	ISCSI_SESSION_LOCK(is);
1554 	is->is_waiting_for_iscsid = false;
1555 	is->is_login_phase = true;
1556 	is->is_timeout = 0;
1557 	ISCSI_SESSION_UNLOCK(is);
1558 
1559 	error = icl_conn_connect(is->is_conn, idc->idc_iser, idc->idc_domain,
1560 	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1561 	free(from_sa, M_SONAME);
1562 	free(to_sa, M_SONAME);
1563 
1564 	/*
1565 	 * Digests are always disabled during login phase.
1566 	 */
1567 	is->is_conn->ic_header_crc32c = false;
1568 	is->is_conn->ic_data_crc32c = false;
1569 
1570 	return (error);
1571 }
1572 
1573 static int
1574 iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1575     struct iscsi_daemon_send *ids)
1576 {
1577 	struct iscsi_session *is;
1578 	struct icl_pdu *ip;
1579 	size_t datalen;
1580 	void *data;
1581 	int error;
1582 
1583 	sx_slock(&sc->sc_lock);
1584 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1585 		if (is->is_id == ids->ids_session_id)
1586 			break;
1587 	}
1588 	if (is == NULL) {
1589 		sx_sunlock(&sc->sc_lock);
1590 		return (ESRCH);
1591 	}
1592 	sx_sunlock(&sc->sc_lock);
1593 
1594 	if (is->is_login_phase == false)
1595 		return (EBUSY);
1596 
1597 	if (is->is_terminating || is->is_reconnecting)
1598 		return (EIO);
1599 
1600 	datalen = ids->ids_data_segment_len;
1601 	if (datalen > ISCSI_MAX_DATA_SEGMENT_LENGTH)
1602 		return (EINVAL);
1603 	if (datalen > 0) {
1604 		data = malloc(datalen, M_ISCSI, M_WAITOK);
1605 		error = copyin(ids->ids_data_segment, data, datalen);
1606 		if (error != 0) {
1607 			free(data, M_ISCSI);
1608 			return (error);
1609 		}
1610 	}
1611 
1612 	ip = icl_pdu_new(is->is_conn, M_WAITOK);
1613 	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1614 	if (datalen > 0) {
1615 		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1616 		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1617 		free(data, M_ISCSI);
1618 	}
1619 	icl_pdu_queue(ip);
1620 
1621 	return (0);
1622 }
1623 
1624 static int
1625 iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1626     struct iscsi_daemon_receive *idr)
1627 {
1628 	struct iscsi_session *is;
1629 	struct icl_pdu *ip;
1630 	void *data;
1631 
1632 	sx_slock(&sc->sc_lock);
1633 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1634 		if (is->is_id == idr->idr_session_id)
1635 			break;
1636 	}
1637 	if (is == NULL) {
1638 		sx_sunlock(&sc->sc_lock);
1639 		return (ESRCH);
1640 	}
1641 	sx_sunlock(&sc->sc_lock);
1642 
1643 	if (is->is_login_phase == false)
1644 		return (EBUSY);
1645 
1646 	ISCSI_SESSION_LOCK(is);
1647 	while (is->is_login_pdu == NULL &&
1648 	    is->is_terminating == false &&
1649 	    is->is_reconnecting == false)
1650 		cv_wait(&is->is_login_cv, &is->is_lock);
1651 	if (is->is_terminating || is->is_reconnecting) {
1652 		ISCSI_SESSION_UNLOCK(is);
1653 		return (EIO);
1654 	}
1655 	ip = is->is_login_pdu;
1656 	is->is_login_pdu = NULL;
1657 	ISCSI_SESSION_UNLOCK(is);
1658 
1659 	if (ip->ip_data_len > idr->idr_data_segment_len) {
1660 		icl_pdu_free(ip);
1661 		return (EMSGSIZE);
1662 	}
1663 
1664 	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1665 	if (ip->ip_data_len > 0) {
1666 		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1667 		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1668 		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1669 		free(data, M_ISCSI);
1670 	}
1671 
1672 	icl_pdu_free(ip);
1673 
1674 	return (0);
1675 }
1676 #endif /* ICL_KERNEL_PROXY */
1677 
1678 static void
1679 iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1680 {
1681 	/*
1682 	 * Just make sure all the fields are null-terminated.
1683 	 *
1684 	 * XXX: This is not particularly secure.  We should
1685 	 * 	create our own conf and then copy in relevant
1686 	 * 	fields.
1687 	 */
1688 	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1689 	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1690 	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1691 	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1692 	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1693 	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1694 	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1695 	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1696 	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1697 }
1698 
1699 static bool
1700 iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1701 {
1702 
1703 	if (isc->isc_initiator[0] == '\0') {
1704 		ISCSI_DEBUG("empty isc_initiator");
1705 		return (false);
1706 	}
1707 
1708 	if (isc->isc_target_addr[0] == '\0') {
1709 		ISCSI_DEBUG("empty isc_target_addr");
1710 		return (false);
1711 	}
1712 
1713 	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1714 		ISCSI_DEBUG("non-empty isc_target for discovery session");
1715 		return (false);
1716 	}
1717 
1718 	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1719 		ISCSI_DEBUG("empty isc_target for non-discovery session");
1720 		return (false);
1721 	}
1722 
1723 	return (true);
1724 }
1725 
1726 static int
1727 iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1728 {
1729 	struct iscsi_session *is;
1730 	const struct iscsi_session *is2;
1731 	int error;
1732 
1733 	iscsi_sanitize_session_conf(&isa->isa_conf);
1734 	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1735 		return (EINVAL);
1736 
1737 	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1738 	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1739 
1740 	sx_xlock(&sc->sc_lock);
1741 
1742 	/*
1743 	 * Prevent duplicates.
1744 	 */
1745 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1746 		if (!!is->is_conf.isc_discovery !=
1747 		    !!is2->is_conf.isc_discovery)
1748 			continue;
1749 
1750 		if (strcmp(is->is_conf.isc_target_addr,
1751 		    is2->is_conf.isc_target_addr) != 0)
1752 			continue;
1753 
1754 		if (is->is_conf.isc_discovery == 0 &&
1755 		    strcmp(is->is_conf.isc_target,
1756 		    is2->is_conf.isc_target) != 0)
1757 			continue;
1758 
1759 		sx_xunlock(&sc->sc_lock);
1760 		free(is, M_ISCSI);
1761 		return (EBUSY);
1762 	}
1763 
1764 	is->is_conn = icl_new_conn(is->is_conf.isc_offload,
1765 	    "iscsi", &is->is_lock);
1766 	if (is->is_conn == NULL) {
1767 		sx_xunlock(&sc->sc_lock);
1768 		free(is, M_ISCSI);
1769 		return (EINVAL);
1770 	}
1771 	is->is_conn->ic_receive = iscsi_receive_callback;
1772 	is->is_conn->ic_error = iscsi_error_callback;
1773 	is->is_conn->ic_prv0 = is;
1774 	TAILQ_INIT(&is->is_outstanding);
1775 	STAILQ_INIT(&is->is_postponed);
1776 	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1777 	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1778 #ifdef ICL_KERNEL_PROXY
1779 	cv_init(&is->is_login_cv, "iscsi_login");
1780 #endif
1781 
1782 	is->is_softc = sc;
1783 	sc->sc_last_session_id++;
1784 	is->is_id = sc->sc_last_session_id;
1785 	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1786 	arc4rand(&is->is_isid[1], 5, 0);
1787 	is->is_tsih = 0;
1788 	callout_init(&is->is_callout, 1);
1789 
1790 	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1791 	if (error != 0) {
1792 		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1793 		sx_xunlock(&sc->sc_lock);
1794 		return (error);
1795 	}
1796 
1797 	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1798 	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1799 
1800 	/*
1801 	 * Trigger immediate reconnection.
1802 	 */
1803 	ISCSI_SESSION_LOCK(is);
1804 	is->is_waiting_for_iscsid = true;
1805 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1806 	ISCSI_SESSION_UNLOCK(is);
1807 	cv_signal(&sc->sc_cv);
1808 
1809 	sx_xunlock(&sc->sc_lock);
1810 
1811 	return (0);
1812 }
1813 
1814 static bool
1815 iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1816     unsigned int id2, const struct iscsi_session_conf *c2)
1817 {
1818 
1819 	if (id2 != 0 && id2 != id1)
1820 		return (false);
1821 	if (c2->isc_target[0] != '\0' &&
1822 	    strcmp(c1->isc_target, c2->isc_target) != 0)
1823 		return (false);
1824 	if (c2->isc_target_addr[0] != '\0' &&
1825 	    strcmp(c1->isc_target_addr, c2->isc_target_addr) != 0)
1826 		return (false);
1827 	return (true);
1828 }
1829 
1830 static int
1831 iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1832     struct iscsi_session_remove *isr)
1833 {
1834 	struct iscsi_session *is, *tmp;
1835 	bool found = false;
1836 
1837 	iscsi_sanitize_session_conf(&isr->isr_conf);
1838 
1839 	sx_xlock(&sc->sc_lock);
1840 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1841 		ISCSI_SESSION_LOCK(is);
1842 		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1843 		    isr->isr_session_id, &isr->isr_conf)) {
1844 			found = true;
1845 			iscsi_session_logout(is);
1846 			iscsi_session_terminate(is);
1847 		}
1848 		ISCSI_SESSION_UNLOCK(is);
1849 	}
1850 	sx_xunlock(&sc->sc_lock);
1851 
1852 	if (!found)
1853 		return (ESRCH);
1854 
1855 	return (0);
1856 }
1857 
1858 static int
1859 iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1860 {
1861 	int error;
1862 	unsigned int i = 0;
1863 	struct iscsi_session *is;
1864 	struct iscsi_session_state iss;
1865 
1866 	sx_slock(&sc->sc_lock);
1867 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1868 		if (i >= isl->isl_nentries) {
1869 			sx_sunlock(&sc->sc_lock);
1870 			return (EMSGSIZE);
1871 		}
1872 		memset(&iss, 0, sizeof(iss));
1873 		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1874 		iss.iss_id = is->is_id;
1875 		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1876 		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1877 		strlcpy(iss.iss_offload, is->is_conn->ic_offload, sizeof(iss.iss_offload));
1878 
1879 		if (is->is_conn->ic_header_crc32c)
1880 			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1881 		else
1882 			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1883 
1884 		if (is->is_conn->ic_data_crc32c)
1885 			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1886 		else
1887 			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1888 
1889 		iss.iss_max_data_segment_length = is->is_max_data_segment_length;
1890 		iss.iss_immediate_data = is->is_immediate_data;
1891 		iss.iss_connected = is->is_connected;
1892 
1893 		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1894 		if (error != 0) {
1895 			sx_sunlock(&sc->sc_lock);
1896 			return (error);
1897 		}
1898 		i++;
1899 	}
1900 	sx_sunlock(&sc->sc_lock);
1901 
1902 	isl->isl_nentries = i;
1903 
1904 	return (0);
1905 }
1906 
1907 static int
1908 iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1909     struct iscsi_session_modify *ism)
1910 {
1911 	struct iscsi_session *is;
1912 
1913 	iscsi_sanitize_session_conf(&ism->ism_conf);
1914 	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1915 		return (EINVAL);
1916 
1917 	sx_xlock(&sc->sc_lock);
1918 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1919 		ISCSI_SESSION_LOCK(is);
1920 		if (is->is_id == ism->ism_session_id)
1921 			break;
1922 		ISCSI_SESSION_UNLOCK(is);
1923 	}
1924 	if (is == NULL) {
1925 		sx_xunlock(&sc->sc_lock);
1926 		return (ESRCH);
1927 	}
1928 	sx_xunlock(&sc->sc_lock);
1929 
1930 	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
1931 	ISCSI_SESSION_UNLOCK(is);
1932 
1933 	iscsi_session_reconnect(is);
1934 
1935 	return (0);
1936 }
1937 
1938 static int
1939 iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
1940     struct thread *td)
1941 {
1942 	struct iscsi_softc *sc;
1943 
1944 	sc = dev->si_drv1;
1945 
1946 	switch (cmd) {
1947 	case ISCSIDWAIT:
1948 		return (iscsi_ioctl_daemon_wait(sc,
1949 		    (struct iscsi_daemon_request *)arg));
1950 	case ISCSIDHANDOFF:
1951 		return (iscsi_ioctl_daemon_handoff(sc,
1952 		    (struct iscsi_daemon_handoff *)arg));
1953 	case ISCSIDFAIL:
1954 		return (iscsi_ioctl_daemon_fail(sc,
1955 		    (struct iscsi_daemon_fail *)arg));
1956 #ifdef ICL_KERNEL_PROXY
1957 	case ISCSIDCONNECT:
1958 		return (iscsi_ioctl_daemon_connect(sc,
1959 		    (struct iscsi_daemon_connect *)arg));
1960 	case ISCSIDSEND:
1961 		return (iscsi_ioctl_daemon_send(sc,
1962 		    (struct iscsi_daemon_send *)arg));
1963 	case ISCSIDRECEIVE:
1964 		return (iscsi_ioctl_daemon_receive(sc,
1965 		    (struct iscsi_daemon_receive *)arg));
1966 #endif /* ICL_KERNEL_PROXY */
1967 	case ISCSISADD:
1968 		return (iscsi_ioctl_session_add(sc,
1969 		    (struct iscsi_session_add *)arg));
1970 	case ISCSISREMOVE:
1971 		return (iscsi_ioctl_session_remove(sc,
1972 		    (struct iscsi_session_remove *)arg));
1973 	case ISCSISLIST:
1974 		return (iscsi_ioctl_session_list(sc,
1975 		    (struct iscsi_session_list *)arg));
1976 	case ISCSISMODIFY:
1977 		return (iscsi_ioctl_session_modify(sc,
1978 		    (struct iscsi_session_modify *)arg));
1979 	default:
1980 		return (EINVAL);
1981 	}
1982 }
1983 
1984 static struct iscsi_outstanding *
1985 iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
1986 {
1987 	struct iscsi_outstanding *io;
1988 
1989 	ISCSI_SESSION_LOCK_ASSERT(is);
1990 
1991 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1992 		if (io->io_initiator_task_tag == initiator_task_tag)
1993 			return (io);
1994 	}
1995 	return (NULL);
1996 }
1997 
1998 static struct iscsi_outstanding *
1999 iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
2000 {
2001 	struct iscsi_outstanding *io;
2002 
2003 	ISCSI_SESSION_LOCK_ASSERT(is);
2004 
2005 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2006 		if (io->io_ccb == ccb)
2007 			return (io);
2008 	}
2009 	return (NULL);
2010 }
2011 
2012 static struct iscsi_outstanding *
2013 iscsi_outstanding_add(struct iscsi_session *is,
2014     union ccb *ccb, uint32_t *initiator_task_tagp)
2015 {
2016 	struct iscsi_outstanding *io;
2017 	int error;
2018 
2019 	ISCSI_SESSION_LOCK_ASSERT(is);
2020 
2021 	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
2022 	if (io == NULL) {
2023 		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes",
2024 		    sizeof(*io));
2025 		return (NULL);
2026 	}
2027 
2028 	error = icl_conn_task_setup(is->is_conn, &ccb->csio,
2029 	    initiator_task_tagp, &io->io_icl_prv);
2030 	if (error != 0) {
2031 		ISCSI_SESSION_WARN(is,
2032 		    "icl_conn_task_setup() failed with error %d", error);
2033 		uma_zfree(iscsi_outstanding_zone, io);
2034 		return (NULL);
2035 	}
2036 
2037 	KASSERT(iscsi_outstanding_find(is, *initiator_task_tagp) == NULL,
2038 	    ("initiator_task_tag 0x%x already added", *initiator_task_tagp));
2039 
2040 	io->io_initiator_task_tag = *initiator_task_tagp;
2041 	io->io_ccb = ccb;
2042 	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
2043 	return (io);
2044 }
2045 
2046 static void
2047 iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2048 {
2049 
2050 	ISCSI_SESSION_LOCK_ASSERT(is);
2051 
2052 	icl_conn_task_done(is->is_conn, io->io_icl_prv);
2053 	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2054 	uma_zfree(iscsi_outstanding_zone, io);
2055 }
2056 
2057 static void
2058 iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2059 {
2060 	struct icl_pdu *request;
2061 	struct iscsi_bhs_task_management_request *bhstmr;
2062 	struct ccb_abort *cab = &ccb->cab;
2063 	struct iscsi_outstanding *io, *aio;
2064 	uint32_t initiator_task_tag;
2065 
2066 	ISCSI_SESSION_LOCK_ASSERT(is);
2067 
2068 #if 0
2069 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2070 #else
2071 	if (is->is_login_phase) {
2072 		ccb->ccb_h.status = CAM_REQ_ABORTED;
2073 		xpt_done(ccb);
2074 		return;
2075 	}
2076 #endif
2077 
2078 	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2079 	if (aio == NULL) {
2080 		ccb->ccb_h.status = CAM_REQ_CMP;
2081 		xpt_done(ccb);
2082 		return;
2083 	}
2084 
2085 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2086 	if (request == NULL) {
2087 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2088 		xpt_done(ccb);
2089 		return;
2090 	}
2091 
2092 	initiator_task_tag = is->is_initiator_task_tag++;
2093 
2094 	io = iscsi_outstanding_add(is, NULL, &initiator_task_tag);
2095 	if (io == NULL) {
2096 		icl_pdu_free(request);
2097 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2098 		xpt_done(ccb);
2099 		return;
2100 	}
2101 	io->io_datasn = aio->io_initiator_task_tag;
2102 
2103 	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2104 	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2105 	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2106 	bhstmr->bhstmr_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2107 	bhstmr->bhstmr_initiator_task_tag = initiator_task_tag;
2108 	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2109 
2110 	iscsi_pdu_queue_locked(request);
2111 }
2112 
2113 static void
2114 iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2115 {
2116 	struct icl_pdu *request;
2117 	struct iscsi_bhs_scsi_command *bhssc;
2118 	struct ccb_scsiio *csio;
2119 	struct iscsi_outstanding *io;
2120 	size_t len;
2121 	uint32_t initiator_task_tag;
2122 	int error;
2123 
2124 	ISCSI_SESSION_LOCK_ASSERT(is);
2125 
2126 #if 0
2127 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2128 #else
2129 	if (is->is_login_phase) {
2130 		ISCSI_SESSION_DEBUG(is, "called during login phase");
2131 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2132 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2133 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2134 		}
2135 		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2136 		xpt_done(ccb);
2137 		return;
2138 	}
2139 #endif
2140 
2141 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2142 	if (request == NULL) {
2143 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2144 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2145 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2146 		}
2147 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2148 		xpt_done(ccb);
2149 		return;
2150 	}
2151 
2152 	initiator_task_tag = is->is_initiator_task_tag++;
2153 	io = iscsi_outstanding_add(is, ccb, &initiator_task_tag);
2154 	if (io == NULL) {
2155 		icl_pdu_free(request);
2156 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2157 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2158 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2159 		}
2160 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2161 		xpt_done(ccb);
2162 		return;
2163 	}
2164 
2165 	csio = &ccb->csio;
2166 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2167 	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2168 	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2169 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2170 	case CAM_DIR_IN:
2171 		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2172 		break;
2173 	case CAM_DIR_OUT:
2174 		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2175 		break;
2176 	}
2177 
2178 	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2179 		switch (csio->tag_action) {
2180 		case MSG_HEAD_OF_Q_TAG:
2181 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2182 			break;
2183 		case MSG_ORDERED_Q_TAG:
2184 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2185 			break;
2186 		case MSG_ACA_TASK:
2187 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2188 			break;
2189 		case MSG_SIMPLE_Q_TAG:
2190 		default:
2191 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2192 			break;
2193 		}
2194 	} else
2195 		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2196 
2197 	bhssc->bhssc_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2198 	bhssc->bhssc_initiator_task_tag = initiator_task_tag;
2199 	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2200 	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2201 	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2202 
2203 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2204 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2205 	else
2206 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2207 
2208 	if (is->is_immediate_data &&
2209 	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2210 		len = csio->dxfer_len;
2211 		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2212 		if (len > is->is_first_burst_length) {
2213 			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_first_burst_length);
2214 			len = is->is_first_burst_length;
2215 		}
2216 		if (len > is->is_max_data_segment_length) {
2217 			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_max_data_segment_length);
2218 			len = is->is_max_data_segment_length;
2219 		}
2220 
2221 		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2222 		if (error != 0) {
2223 			icl_pdu_free(request);
2224 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2225 				xpt_freeze_devq(ccb->ccb_h.path, 1);
2226 				ISCSI_SESSION_DEBUG(is, "freezing devq");
2227 			}
2228 			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2229 			xpt_done(ccb);
2230 			return;
2231 		}
2232 	}
2233 	iscsi_pdu_queue_locked(request);
2234 }
2235 
2236 static void
2237 iscsi_action(struct cam_sim *sim, union ccb *ccb)
2238 {
2239 	struct iscsi_session *is;
2240 
2241 	is = cam_sim_softc(sim);
2242 
2243 	ISCSI_SESSION_LOCK_ASSERT(is);
2244 
2245 	if (is->is_terminating ||
2246 	    (is->is_connected == false && fail_on_disconnection)) {
2247 		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2248 		xpt_done(ccb);
2249 		return;
2250 	}
2251 
2252 	switch (ccb->ccb_h.func_code) {
2253 	case XPT_PATH_INQ:
2254 	{
2255 		struct ccb_pathinq *cpi = &ccb->cpi;
2256 
2257 		cpi->version_num = 1;
2258 		cpi->hba_inquiry = PI_TAG_ABLE;
2259 		cpi->target_sprt = 0;
2260 		cpi->hba_misc = PIM_EXTLUNS;
2261 		cpi->hba_eng_cnt = 0;
2262 		cpi->max_target = 0;
2263 		/*
2264 		 * Note that the variable below is only relevant for targets
2265 		 * that don't claim compliance with anything above SPC2, which
2266 		 * means they don't support REPORT_LUNS.
2267 		 */
2268 		cpi->max_lun = 255;
2269 		cpi->initiator_id = ~0;
2270 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2271 		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2272 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2273 		cpi->unit_number = cam_sim_unit(sim);
2274 		cpi->bus_id = cam_sim_bus(sim);
2275 		cpi->base_transfer_speed = 150000; /* XXX */
2276 		cpi->transport = XPORT_ISCSI;
2277 		cpi->transport_version = 0;
2278 		cpi->protocol = PROTO_SCSI;
2279 		cpi->protocol_version = SCSI_REV_SPC3;
2280 		cpi->maxio = MAXPHYS;
2281 		cpi->ccb_h.status = CAM_REQ_CMP;
2282 		break;
2283 	}
2284 	case XPT_GET_TRAN_SETTINGS:
2285 	{
2286 		struct ccb_trans_settings	*cts;
2287 		struct ccb_trans_settings_scsi	*scsi;
2288 
2289 		cts = &ccb->cts;
2290 		scsi = &cts->proto_specific.scsi;
2291 
2292 		cts->protocol = PROTO_SCSI;
2293 		cts->protocol_version = SCSI_REV_SPC3;
2294 		cts->transport = XPORT_ISCSI;
2295 		cts->transport_version = 0;
2296 		scsi->valid = CTS_SCSI_VALID_TQ;
2297 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2298 		cts->ccb_h.status = CAM_REQ_CMP;
2299 		break;
2300 	}
2301 	case XPT_CALC_GEOMETRY:
2302 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2303 		ccb->ccb_h.status = CAM_REQ_CMP;
2304 		break;
2305 #if 0
2306 	/*
2307 	 * XXX: What's the point?
2308 	 */
2309 	case XPT_RESET_BUS:
2310 	case XPT_TERM_IO:
2311 		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2312 		ccb->ccb_h.status = CAM_REQ_CMP;
2313 		break;
2314 #endif
2315 	case XPT_ABORT:
2316 		iscsi_action_abort(is, ccb);
2317 		return;
2318 	case XPT_SCSI_IO:
2319 		iscsi_action_scsiio(is, ccb);
2320 		return;
2321 	default:
2322 #if 0
2323 		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2324 #endif
2325 		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2326 		break;
2327 	}
2328 	xpt_done(ccb);
2329 }
2330 
2331 static void
2332 iscsi_poll(struct cam_sim *sim)
2333 {
2334 
2335 	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2336 }
2337 
2338 static void
2339 iscsi_terminate_sessions(struct iscsi_softc *sc)
2340 {
2341 	struct iscsi_session *is;
2342 
2343 	sx_slock(&sc->sc_lock);
2344 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2345 		iscsi_session_terminate(is);
2346 	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2347 		ISCSI_DEBUG("waiting for sessions to terminate");
2348 		cv_wait(&sc->sc_cv, &sc->sc_lock);
2349 	}
2350 	ISCSI_DEBUG("all sessions terminated");
2351 	sx_sunlock(&sc->sc_lock);
2352 }
2353 
2354 static void
2355 iscsi_shutdown_pre(struct iscsi_softc *sc)
2356 {
2357 	struct iscsi_session *is;
2358 
2359 	if (!fail_on_shutdown)
2360 		return;
2361 
2362 	/*
2363 	 * If we have any sessions waiting for reconnection, request
2364 	 * maintenance thread to fail them immediately instead of waiting
2365 	 * for reconnect timeout.
2366 	 *
2367 	 * This prevents LUNs with mounted filesystems that are supported
2368 	 * by disconnected iSCSI sessions from hanging, however it will
2369 	 * fail all queued BIOs.
2370 	 */
2371 	ISCSI_DEBUG("forcing failing all disconnected sessions due to shutdown");
2372 
2373 	fail_on_disconnection = 1;
2374 
2375 	sx_slock(&sc->sc_lock);
2376 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2377 		ISCSI_SESSION_LOCK(is);
2378 		if (!is->is_connected) {
2379 			ISCSI_SESSION_DEBUG(is, "force failing disconnected session early");
2380 			iscsi_session_reconnect(is);
2381 		}
2382 		ISCSI_SESSION_UNLOCK(is);
2383 	}
2384 	sx_sunlock(&sc->sc_lock);
2385 }
2386 
2387 static void
2388 iscsi_shutdown_post(struct iscsi_softc *sc)
2389 {
2390 
2391 	ISCSI_DEBUG("removing all sessions due to shutdown");
2392 	iscsi_terminate_sessions(sc);
2393 }
2394 
2395 static int
2396 iscsi_load(void)
2397 {
2398 	int error;
2399 
2400 	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2401 	sx_init(&sc->sc_lock, "iscsi");
2402 	TAILQ_INIT(&sc->sc_sessions);
2403 	cv_init(&sc->sc_cv, "iscsi_cv");
2404 
2405 	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2406 	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2407 	    UMA_ALIGN_PTR, 0);
2408 
2409 	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2410 	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2411 	if (error != 0) {
2412 		ISCSI_WARN("failed to create device node, error %d", error);
2413 		return (error);
2414 	}
2415 	sc->sc_cdev->si_drv1 = sc;
2416 
2417 	sc->sc_shutdown_pre_eh = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2418 	    iscsi_shutdown_pre, sc, SHUTDOWN_PRI_FIRST);
2419 	/*
2420 	 * shutdown_post_sync needs to run after filesystem shutdown and before
2421 	 * CAM shutdown - otherwise when rebooting with an iSCSI session that is
2422 	 * disconnected but has outstanding requests, dashutdown() will hang on
2423 	 * cam_periph_runccb().
2424 	 */
2425 	sc->sc_shutdown_post_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2426 	    iscsi_shutdown_post, sc, SHUTDOWN_PRI_DEFAULT - 1);
2427 
2428 	return (0);
2429 }
2430 
2431 static int
2432 iscsi_unload(void)
2433 {
2434 
2435 	if (sc->sc_cdev != NULL) {
2436 		ISCSI_DEBUG("removing device node");
2437 		destroy_dev(sc->sc_cdev);
2438 		ISCSI_DEBUG("device node removed");
2439 	}
2440 
2441 	if (sc->sc_shutdown_pre_eh != NULL)
2442 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, sc->sc_shutdown_pre_eh);
2443 	if (sc->sc_shutdown_post_eh != NULL)
2444 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_post_eh);
2445 
2446 	iscsi_terminate_sessions(sc);
2447 
2448 	uma_zdestroy(iscsi_outstanding_zone);
2449 	sx_destroy(&sc->sc_lock);
2450 	cv_destroy(&sc->sc_cv);
2451 	free(sc, M_ISCSI);
2452 	return (0);
2453 }
2454 
2455 static int
2456 iscsi_quiesce(void)
2457 {
2458 	sx_slock(&sc->sc_lock);
2459 	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2460 		sx_sunlock(&sc->sc_lock);
2461 		return (EBUSY);
2462 	}
2463 	sx_sunlock(&sc->sc_lock);
2464 	return (0);
2465 }
2466 
2467 static int
2468 iscsi_modevent(module_t mod, int what, void *arg)
2469 {
2470 	int error;
2471 
2472 	switch (what) {
2473 	case MOD_LOAD:
2474 		error = iscsi_load();
2475 		break;
2476 	case MOD_UNLOAD:
2477 		error = iscsi_unload();
2478 		break;
2479 	case MOD_QUIESCE:
2480 		error = iscsi_quiesce();
2481 		break;
2482 	default:
2483 		error = EINVAL;
2484 		break;
2485 	}
2486 	return (error);
2487 }
2488 
2489 moduledata_t iscsi_data = {
2490 	"iscsi",
2491 	iscsi_modevent,
2492 	0
2493 };
2494 
2495 DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2496 MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2497 MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2498