xref: /freebsd/sys/dev/iscsi/iscsi.c (revision 243e928310d073338c5ec089f0dce238a80b9866)
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/socket.h>
47 #include <sys/sysctl.h>
48 #include <sys/systm.h>
49 #include <sys/sx.h>
50 #include <vm/uma.h>
51 
52 #include <cam/cam.h>
53 #include <cam/cam_ccb.h>
54 #include <cam/cam_xpt.h>
55 #include <cam/cam_debug.h>
56 #include <cam/cam_sim.h>
57 #include <cam/cam_xpt_sim.h>
58 #include <cam/cam_xpt_periph.h>
59 #include <cam/cam_periph.h>
60 #include <cam/scsi/scsi_all.h>
61 #include <cam/scsi/scsi_message.h>
62 
63 #include <dev/iscsi/icl.h>
64 #include <dev/iscsi/icl_wrappers.h>
65 #include <dev/iscsi/iscsi_ioctl.h>
66 #include <dev/iscsi/iscsi_proto.h>
67 #include <dev/iscsi/iscsi.h>
68 
69 #ifdef ICL_KERNEL_PROXY
70 #include <sys/socketvar.h>
71 #endif
72 
73 #ifdef ICL_KERNEL_PROXY
74 FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
75 #endif
76 
77 /*
78  * XXX: This is global so the iscsi_unload() can access it.
79  * 	Think about how to do this properly.
80  */
81 static struct iscsi_softc	*sc;
82 
83 SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD, 0, "iSCSI initiator");
84 static int debug = 1;
85 SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
86     &debug, 0, "Enable debug messages");
87 static int ping_timeout = 5;
88 SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
89     0, "Timeout for ping (NOP-Out) requests, in seconds");
90 static int iscsid_timeout = 60;
91 SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
92     0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
93 static int login_timeout = 60;
94 SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
95     0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
96 static int maxtags = 255;
97 SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
98     0, "Max number of IO requests queued");
99 static int fail_on_disconnection = 0;
100 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
101     &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
102 static int fail_on_shutdown = 1;
103 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_shutdown, CTLFLAG_RWTUN,
104     &fail_on_shutdown, 0, "Fail disconnected sessions on shutdown");
105 
106 static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
107 static uma_zone_t iscsi_outstanding_zone;
108 
109 #define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
110 #define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
111 
112 #define	ISCSI_DEBUG(X, ...)						\
113 	do {								\
114 		if (debug > 1) 						\
115 			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
116 	} while (0)
117 
118 #define	ISCSI_WARN(X, ...)						\
119 	do {								\
120 		if (debug > 0) {					\
121 			printf("WARNING: %s: " X "\n",			\
122 			    __func__, ## __VA_ARGS__);			\
123 		}							\
124 	} while (0)
125 
126 #define	ISCSI_SESSION_DEBUG(S, X, ...)					\
127 	do {								\
128 		if (debug > 1) {					\
129 			printf("%s: %s (%s): " X "\n",			\
130 			    __func__, S->is_conf.isc_target_addr,	\
131 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
132 		}							\
133 	} while (0)
134 
135 #define	ISCSI_SESSION_WARN(S, X, ...)					\
136 	do {								\
137 		if (debug > 0) {					\
138 			printf("WARNING: %s (%s): " X "\n",		\
139 			    S->is_conf.isc_target_addr,			\
140 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
141 		}							\
142 	} while (0)
143 
144 #define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
145 #define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
146 #define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
147 #define ISCSI_SESSION_LOCK_ASSERT_NOT(X) mtx_assert(&X->is_lock, MA_NOTOWNED)
148 
149 static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
150 		    int mode, struct thread *td);
151 
152 static struct cdevsw iscsi_cdevsw = {
153      .d_version = D_VERSION,
154      .d_ioctl   = iscsi_ioctl,
155      .d_name    = "iscsi",
156 };
157 
158 static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
159 static void	iscsi_pdu_queue(struct icl_pdu *request);
160 static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
161 static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
162 static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
163 static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
164 static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
165 static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
166 static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
167 static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
168 static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
169 static void	iscsi_session_reconnect(struct iscsi_session *is);
170 static void	iscsi_session_terminate(struct iscsi_session *is);
171 static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
172 static void	iscsi_poll(struct cam_sim *sim);
173 static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
174 		    uint32_t initiator_task_tag);
175 static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
176 		    struct icl_pdu *request, union ccb *ccb,
177 		    uint32_t *initiator_task_tagp);
178 static void	iscsi_outstanding_remove(struct iscsi_session *is,
179 		    struct iscsi_outstanding *io);
180 
181 static bool
182 iscsi_pdu_prepare(struct icl_pdu *request)
183 {
184 	struct iscsi_session *is;
185 	struct iscsi_bhs_scsi_command *bhssc;
186 
187 	is = PDU_SESSION(request);
188 
189 	ISCSI_SESSION_LOCK_ASSERT(is);
190 
191 	/*
192 	 * We're only using fields common for all the request
193 	 * (initiator -> target) PDUs.
194 	 */
195 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
196 
197 	/*
198 	 * Data-Out PDU does not contain CmdSN.
199 	 */
200 	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
201 		if (ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn) &&
202 		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
203 			/*
204 			 * Current MaxCmdSN prevents us from sending any more
205 			 * SCSI Command PDUs to the target; postpone the PDU.
206 			 * It will get resent by either iscsi_pdu_queue(),
207 			 * or by maintenance thread.
208 			 */
209 #if 0
210 			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %u, "
211 			    "ExpCmdSN %u, MaxCmdSN %u, opcode 0x%x",
212 			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn,
213 			    bhssc->bhssc_opcode);
214 #endif
215 			return (true);
216 		}
217 		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
218 		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
219 			is->is_cmdsn++;
220 	}
221 	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
222 
223 	return (false);
224 }
225 
226 static void
227 iscsi_session_send_postponed(struct iscsi_session *is)
228 {
229 	struct icl_pdu *request;
230 	bool postpone;
231 
232 	ISCSI_SESSION_LOCK_ASSERT(is);
233 
234 	while (!STAILQ_EMPTY(&is->is_postponed)) {
235 		request = STAILQ_FIRST(&is->is_postponed);
236 		postpone = iscsi_pdu_prepare(request);
237 		if (postpone)
238 			break;
239 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
240 		icl_pdu_queue(request);
241 	}
242 }
243 
244 static void
245 iscsi_pdu_queue_locked(struct icl_pdu *request)
246 {
247 	struct iscsi_session *is;
248 	bool postpone;
249 
250 	is = PDU_SESSION(request);
251 	ISCSI_SESSION_LOCK_ASSERT(is);
252 	iscsi_session_send_postponed(is);
253 	postpone = iscsi_pdu_prepare(request);
254 	if (postpone) {
255 		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
256 		return;
257 	}
258 	icl_pdu_queue(request);
259 }
260 
261 static void
262 iscsi_pdu_queue(struct icl_pdu *request)
263 {
264 	struct iscsi_session *is;
265 
266 	is = PDU_SESSION(request);
267 	ISCSI_SESSION_LOCK(is);
268 	iscsi_pdu_queue_locked(request);
269 	ISCSI_SESSION_UNLOCK(is);
270 }
271 
272 static void
273 iscsi_session_logout(struct iscsi_session *is)
274 {
275 	struct icl_pdu *request;
276 	struct iscsi_bhs_logout_request *bhslr;
277 
278 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
279 	if (request == NULL)
280 		return;
281 
282 	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
283 	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
284 	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
285 	iscsi_pdu_queue_locked(request);
286 }
287 
288 static void
289 iscsi_session_terminate_task(struct iscsi_session *is,
290     struct iscsi_outstanding *io, bool requeue)
291 {
292 
293 	ISCSI_SESSION_LOCK_ASSERT(is);
294 
295 	if (io->io_ccb != NULL) {
296 		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
297 		if (requeue)
298 			io->io_ccb->ccb_h.status |= CAM_REQUEUE_REQ;
299 		else
300 			io->io_ccb->ccb_h.status |= CAM_REQ_ABORTED;
301 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
302 			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
303 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
304 			ISCSI_SESSION_DEBUG(is, "freezing devq");
305 		}
306 		xpt_done(io->io_ccb);
307 	}
308 	iscsi_outstanding_remove(is, io);
309 }
310 
311 static void
312 iscsi_session_terminate_tasks(struct iscsi_session *is, bool requeue)
313 {
314 	struct iscsi_outstanding *io, *tmp;
315 
316 	ISCSI_SESSION_LOCK_ASSERT(is);
317 
318 	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
319 		iscsi_session_terminate_task(is, io, requeue);
320 	}
321 }
322 
323 static void
324 iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
325 {
326 	struct icl_pdu *pdu;
327 
328 	ISCSI_SESSION_LOCK_ASSERT(is);
329 
330 	/*
331 	 * Don't queue any new PDUs.
332 	 */
333 	if (is->is_sim != NULL && is->is_simq_frozen == false) {
334 		ISCSI_SESSION_DEBUG(is, "freezing");
335 		xpt_freeze_simq(is->is_sim, 1);
336 		is->is_simq_frozen = true;
337 	}
338 
339 	/*
340 	 * Remove postponed PDUs.
341 	 */
342 	while (!STAILQ_EMPTY(&is->is_postponed)) {
343 		pdu = STAILQ_FIRST(&is->is_postponed);
344 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
345 		icl_pdu_free(pdu);
346 	}
347 
348 	if (destroy_sim == false) {
349 		/*
350 		 * Terminate SCSI tasks, asking CAM to requeue them.
351 		 */
352 		iscsi_session_terminate_tasks(is, true);
353 		return;
354 	}
355 
356 	iscsi_session_terminate_tasks(is, false);
357 
358 	if (is->is_sim == NULL)
359 		return;
360 
361 	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
362 	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
363 
364 	if (is->is_simq_frozen) {
365 		xpt_release_simq(is->is_sim, 1);
366 		is->is_simq_frozen = false;
367 	}
368 
369 	xpt_free_path(is->is_path);
370 	is->is_path = NULL;
371 	xpt_bus_deregister(cam_sim_path(is->is_sim));
372 	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
373 	is->is_sim = NULL;
374 	is->is_devq = NULL;
375 }
376 
377 static void
378 iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
379 {
380 
381 	icl_conn_close(is->is_conn);
382 
383 	ISCSI_SESSION_LOCK(is);
384 
385 	is->is_connected = false;
386 	is->is_reconnecting = false;
387 	is->is_login_phase = false;
388 
389 #ifdef ICL_KERNEL_PROXY
390 	if (is->is_login_pdu != NULL) {
391 		icl_pdu_free(is->is_login_pdu);
392 		is->is_login_pdu = NULL;
393 	}
394 	cv_signal(&is->is_login_cv);
395 #endif
396 
397 	if (fail_on_disconnection) {
398 		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
399 		iscsi_session_cleanup(is, true);
400 	} else {
401 		iscsi_session_cleanup(is, false);
402 	}
403 
404 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
405 	    ("destroying session with active tasks"));
406 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
407 	    ("destroying session with postponed PDUs"));
408 
409 	/*
410 	 * Request immediate reconnection from iscsid(8).
411 	 */
412 	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
413 	is->is_waiting_for_iscsid = true;
414 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
415 	is->is_timeout = 0;
416 	ISCSI_SESSION_UNLOCK(is);
417 	cv_signal(&is->is_softc->sc_cv);
418 }
419 
420 static void
421 iscsi_maintenance_thread_terminate(struct iscsi_session *is)
422 {
423 	struct iscsi_softc *sc;
424 
425 	sc = is->is_softc;
426 	sx_xlock(&sc->sc_lock);
427 
428 	icl_conn_close(is->is_conn);
429 	callout_drain(&is->is_callout);
430 
431 	ISCSI_SESSION_LOCK(is);
432 
433 	KASSERT(is->is_terminating, ("is_terminating == false"));
434 
435 #ifdef ICL_KERNEL_PROXY
436 	if (is->is_login_pdu != NULL) {
437 		icl_pdu_free(is->is_login_pdu);
438 		is->is_login_pdu = NULL;
439 	}
440 	cv_signal(&is->is_login_cv);
441 #endif
442 
443 	iscsi_session_cleanup(is, true);
444 
445 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
446 	    ("destroying session with active tasks"));
447 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
448 	    ("destroying session with postponed PDUs"));
449 
450 	ISCSI_SESSION_UNLOCK(is);
451 
452 	icl_conn_free(is->is_conn);
453 	mtx_destroy(&is->is_lock);
454 	cv_destroy(&is->is_maintenance_cv);
455 #ifdef ICL_KERNEL_PROXY
456 	cv_destroy(&is->is_login_cv);
457 #endif
458 	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
459 	sx_xunlock(&sc->sc_lock);
460 
461 	ISCSI_SESSION_DEBUG(is, "terminated");
462 	free(is, M_ISCSI);
463 
464 	/*
465 	 * The iscsi_unload() routine might be waiting.
466 	 */
467 	cv_signal(&sc->sc_cv);
468 }
469 
470 static void
471 iscsi_maintenance_thread(void *arg)
472 {
473 	struct iscsi_session *is;
474 
475 	is = arg;
476 
477 	for (;;) {
478 		ISCSI_SESSION_LOCK(is);
479 		if (is->is_reconnecting == false &&
480 		    is->is_terminating == false &&
481 		    STAILQ_EMPTY(&is->is_postponed))
482 			cv_wait(&is->is_maintenance_cv, &is->is_lock);
483 
484 		/* Terminate supersedes reconnect. */
485 		if (is->is_terminating) {
486 			ISCSI_SESSION_UNLOCK(is);
487 			iscsi_maintenance_thread_terminate(is);
488 			kthread_exit();
489 			return;
490 		}
491 
492 		if (is->is_reconnecting) {
493 			ISCSI_SESSION_UNLOCK(is);
494 			iscsi_maintenance_thread_reconnect(is);
495 			continue;
496 		}
497 
498 		iscsi_session_send_postponed(is);
499 		ISCSI_SESSION_UNLOCK(is);
500 	}
501 }
502 
503 static void
504 iscsi_session_reconnect(struct iscsi_session *is)
505 {
506 
507 	/*
508 	 * XXX: We can't use locking here, because
509 	 * 	it's being called from various contexts.
510 	 * 	Hope it doesn't break anything.
511 	 */
512 	if (is->is_reconnecting)
513 		return;
514 
515 	is->is_reconnecting = true;
516 	cv_signal(&is->is_maintenance_cv);
517 }
518 
519 static void
520 iscsi_session_terminate(struct iscsi_session *is)
521 {
522 
523 	if (is->is_terminating)
524 		return;
525 
526 	is->is_terminating = true;
527 
528 #if 0
529 	iscsi_session_logout(is);
530 #endif
531 	cv_signal(&is->is_maintenance_cv);
532 }
533 
534 static void
535 iscsi_callout(void *context)
536 {
537 	struct icl_pdu *request;
538 	struct iscsi_bhs_nop_out *bhsno;
539 	struct iscsi_session *is;
540 	bool reconnect_needed = false;
541 
542 	is = context;
543 
544 	ISCSI_SESSION_LOCK(is);
545 	if (is->is_terminating) {
546 		ISCSI_SESSION_UNLOCK(is);
547 		return;
548 	}
549 
550 	callout_schedule(&is->is_callout, 1 * hz);
551 
552 	is->is_timeout++;
553 
554 	if (is->is_waiting_for_iscsid) {
555 		if (iscsid_timeout > 0 && is->is_timeout > iscsid_timeout) {
556 			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
557 			    "for %d seconds; reconnecting",
558 			    is->is_timeout);
559 			reconnect_needed = true;
560 		}
561 		goto out;
562 	}
563 
564 	if (is->is_login_phase) {
565 		if (login_timeout > 0 && is->is_timeout > login_timeout) {
566 			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
567 			    "reconnecting", is->is_timeout);
568 			reconnect_needed = true;
569 		}
570 		goto out;
571 	}
572 
573 	if (ping_timeout <= 0) {
574 		/*
575 		 * Pings are disabled.  Don't send NOP-Out in this case.
576 		 * Reset the timeout, to avoid triggering reconnection,
577 		 * should the user decide to reenable them.
578 		 */
579 		is->is_timeout = 0;
580 		goto out;
581 	}
582 
583 	if (is->is_timeout >= ping_timeout) {
584 		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
585 		    "reconnecting", ping_timeout);
586 		reconnect_needed = true;
587 		goto out;
588 	}
589 
590 	ISCSI_SESSION_UNLOCK(is);
591 
592 	/*
593 	 * If the ping was reset less than one second ago - which means
594 	 * that we've received some PDU during the last second - assume
595 	 * the traffic flows correctly and don't bother sending a NOP-Out.
596 	 *
597 	 * (It's 2 - one for one second, and one for incrementing is_timeout
598 	 * earlier in this routine.)
599 	 */
600 	if (is->is_timeout < 2)
601 		return;
602 
603 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
604 	if (request == NULL) {
605 		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
606 		return;
607 	}
608 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
609 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
610 	    ISCSI_BHS_OPCODE_IMMEDIATE;
611 	bhsno->bhsno_flags = 0x80;
612 	bhsno->bhsno_target_transfer_tag = 0xffffffff;
613 	iscsi_pdu_queue(request);
614 	return;
615 
616 out:
617 	if (is->is_terminating) {
618 		ISCSI_SESSION_UNLOCK(is);
619 		return;
620 	}
621 
622 	ISCSI_SESSION_UNLOCK(is);
623 
624 	if (reconnect_needed)
625 		iscsi_session_reconnect(is);
626 }
627 
628 static void
629 iscsi_pdu_update_statsn(const struct icl_pdu *response)
630 {
631 	const struct iscsi_bhs_data_in *bhsdi;
632 	struct iscsi_session *is;
633 	uint32_t expcmdsn, maxcmdsn, statsn;
634 
635 	is = PDU_SESSION(response);
636 
637 	ISCSI_SESSION_LOCK_ASSERT(is);
638 
639 	/*
640 	 * We're only using fields common for all the response
641 	 * (target -> initiator) PDUs.
642 	 */
643 	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
644 	/*
645 	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
646 	 * and Residual Count only have meaningful content if the S bit
647 	 * is set to 1", so we also need to check the bit specific for
648 	 * Data-In PDU.
649 	 */
650 	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
651 	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
652 		statsn = ntohl(bhsdi->bhsdi_statsn);
653 		if (statsn != is->is_statsn && statsn != (is->is_statsn + 1)) {
654 			/* XXX: This is normal situation for MCS */
655 			ISCSI_SESSION_WARN(is, "PDU 0x%x StatSN %u != "
656 			    "session ExpStatSN %u (or + 1); reconnecting",
657 			    bhsdi->bhsdi_opcode, statsn, is->is_statsn);
658 			iscsi_session_reconnect(is);
659 		}
660 		if (ISCSI_SNGT(statsn, is->is_statsn))
661 			is->is_statsn = statsn;
662 	}
663 
664 	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
665 	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
666 
667 	if (ISCSI_SNLT(maxcmdsn + 1, expcmdsn)) {
668 		ISCSI_SESSION_DEBUG(is,
669 		    "PDU MaxCmdSN %u + 1 < PDU ExpCmdSN %u; ignoring",
670 		    maxcmdsn, expcmdsn);
671 	} else {
672 		if (ISCSI_SNGT(maxcmdsn, is->is_maxcmdsn)) {
673 			is->is_maxcmdsn = maxcmdsn;
674 
675 			/*
676 			 * Command window increased; kick the maintanance thread
677 			 * to send out postponed commands.
678 			 */
679 			if (!STAILQ_EMPTY(&is->is_postponed))
680 				cv_signal(&is->is_maintenance_cv);
681 		} else if (ISCSI_SNLT(maxcmdsn, is->is_maxcmdsn)) {
682 			/* XXX: This is normal situation for MCS */
683 			ISCSI_SESSION_DEBUG(is,
684 			    "PDU MaxCmdSN %u < session MaxCmdSN %u; ignoring",
685 			    maxcmdsn, is->is_maxcmdsn);
686 		}
687 
688 		if (ISCSI_SNGT(expcmdsn, is->is_expcmdsn)) {
689 			is->is_expcmdsn = expcmdsn;
690 		} else if (ISCSI_SNLT(expcmdsn, is->is_expcmdsn)) {
691 			/* XXX: This is normal situation for MCS */
692 			ISCSI_SESSION_DEBUG(is,
693 			    "PDU ExpCmdSN %u < session ExpCmdSN %u; ignoring",
694 			    expcmdsn, is->is_expcmdsn);
695 		}
696 	}
697 
698 	/*
699 	 * Every incoming PDU - not just NOP-In - resets the ping timer.
700 	 * The purpose of the timeout is to reset the connection when it stalls;
701 	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
702 	 * in some queue.
703 	 */
704 	is->is_timeout = 0;
705 }
706 
707 static void
708 iscsi_receive_callback(struct icl_pdu *response)
709 {
710 	struct iscsi_session *is;
711 
712 	is = PDU_SESSION(response);
713 
714 	ISCSI_SESSION_LOCK(is);
715 
716 	iscsi_pdu_update_statsn(response);
717 
718 #ifdef ICL_KERNEL_PROXY
719 	if (is->is_login_phase) {
720 		if (is->is_login_pdu == NULL)
721 			is->is_login_pdu = response;
722 		else
723 			icl_pdu_free(response);
724 		ISCSI_SESSION_UNLOCK(is);
725 		cv_signal(&is->is_login_cv);
726 		return;
727 	}
728 #endif
729 
730 	/*
731 	 * The handling routine is responsible for freeing the PDU
732 	 * when it's no longer needed.
733 	 */
734 	switch (response->ip_bhs->bhs_opcode) {
735 	case ISCSI_BHS_OPCODE_NOP_IN:
736 		iscsi_pdu_handle_nop_in(response);
737 		ISCSI_SESSION_UNLOCK(is);
738 		break;
739 	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
740 		iscsi_pdu_handle_scsi_response(response);
741 		/* Session lock dropped inside. */
742 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
743 		break;
744 	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
745 		iscsi_pdu_handle_task_response(response);
746 		ISCSI_SESSION_UNLOCK(is);
747 		break;
748 	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
749 		iscsi_pdu_handle_data_in(response);
750 		/* Session lock dropped inside. */
751 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
752 		break;
753 	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
754 		iscsi_pdu_handle_logout_response(response);
755 		ISCSI_SESSION_UNLOCK(is);
756 		break;
757 	case ISCSI_BHS_OPCODE_R2T:
758 		iscsi_pdu_handle_r2t(response);
759 		ISCSI_SESSION_UNLOCK(is);
760 		break;
761 	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
762 		iscsi_pdu_handle_async_message(response);
763 		ISCSI_SESSION_UNLOCK(is);
764 		break;
765 	case ISCSI_BHS_OPCODE_REJECT:
766 		iscsi_pdu_handle_reject(response);
767 		ISCSI_SESSION_UNLOCK(is);
768 		break;
769 	default:
770 		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
771 		    "opcode 0x%x; reconnecting",
772 		    response->ip_bhs->bhs_opcode);
773 		iscsi_session_reconnect(is);
774 		ISCSI_SESSION_UNLOCK(is);
775 		icl_pdu_free(response);
776 	}
777 }
778 
779 static void
780 iscsi_error_callback(struct icl_conn *ic)
781 {
782 	struct iscsi_session *is;
783 
784 	is = CONN_SESSION(ic);
785 
786 	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
787 	iscsi_session_reconnect(is);
788 }
789 
790 static void
791 iscsi_pdu_handle_nop_in(struct icl_pdu *response)
792 {
793 	struct iscsi_session *is;
794 	struct iscsi_bhs_nop_out *bhsno;
795 	struct iscsi_bhs_nop_in *bhsni;
796 	struct icl_pdu *request;
797 	void *data = NULL;
798 	size_t datasize;
799 	int error;
800 
801 	is = PDU_SESSION(response);
802 	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
803 
804 	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
805 		/*
806 		 * Nothing to do; iscsi_pdu_update_statsn() already
807 		 * zeroed the timeout.
808 		 */
809 		icl_pdu_free(response);
810 		return;
811 	}
812 
813 	datasize = icl_pdu_data_segment_length(response);
814 	if (datasize > 0) {
815 		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
816 		if (data == NULL) {
817 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
818 			    "reconnecting");
819 			icl_pdu_free(response);
820 			iscsi_session_reconnect(is);
821 			return;
822 		}
823 		icl_pdu_get_data(response, 0, data, datasize);
824 	}
825 
826 	request = icl_pdu_new(response->ip_conn, M_NOWAIT);
827 	if (request == NULL) {
828 		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
829 		    "reconnecting");
830 		free(data, M_ISCSI);
831 		icl_pdu_free(response);
832 		iscsi_session_reconnect(is);
833 		return;
834 	}
835 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
836 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
837 	    ISCSI_BHS_OPCODE_IMMEDIATE;
838 	bhsno->bhsno_flags = 0x80;
839 	bhsno->bhsno_initiator_task_tag = 0xffffffff;
840 	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
841 	if (datasize > 0) {
842 		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
843 		if (error != 0) {
844 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
845 			    "reconnecting");
846 			free(data, M_ISCSI);
847 			icl_pdu_free(request);
848 			icl_pdu_free(response);
849 			iscsi_session_reconnect(is);
850 			return;
851 		}
852 		free(data, M_ISCSI);
853 	}
854 
855 	icl_pdu_free(response);
856 	iscsi_pdu_queue_locked(request);
857 }
858 
859 static void
860 iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
861 {
862 	struct iscsi_bhs_scsi_response *bhssr;
863 	struct iscsi_outstanding *io;
864 	struct iscsi_session *is;
865 	union ccb *ccb;
866 	struct ccb_scsiio *csio;
867 	size_t data_segment_len, received;
868 	uint16_t sense_len;
869 
870 	is = PDU_SESSION(response);
871 
872 	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
873 	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
874 	if (io == NULL || io->io_ccb == NULL) {
875 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
876 		icl_pdu_free(response);
877 		iscsi_session_reconnect(is);
878 		ISCSI_SESSION_UNLOCK(is);
879 		return;
880 	}
881 
882 	ccb = io->io_ccb;
883 	received = io->io_received;
884 	iscsi_outstanding_remove(is, io);
885 	ISCSI_SESSION_UNLOCK(is);
886 
887 	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
888 		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
889  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
890  			xpt_freeze_devq(ccb->ccb_h.path, 1);
891 			ISCSI_SESSION_DEBUG(is, "freezing devq");
892 		}
893  		ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
894 	} else if (bhssr->bhssr_status == 0) {
895 		ccb->ccb_h.status = CAM_REQ_CMP;
896 	} else {
897  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
898  			xpt_freeze_devq(ccb->ccb_h.path, 1);
899 			ISCSI_SESSION_DEBUG(is, "freezing devq");
900 		}
901  		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
902 		ccb->csio.scsi_status = bhssr->bhssr_status;
903 	}
904 
905 	csio = &ccb->csio;
906 	data_segment_len = icl_pdu_data_segment_length(response);
907 	if (data_segment_len > 0) {
908 		if (data_segment_len < sizeof(sense_len)) {
909 			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
910 			    data_segment_len);
911 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
912 				xpt_freeze_devq(ccb->ccb_h.path, 1);
913 				ISCSI_SESSION_DEBUG(is, "freezing devq");
914 			}
915 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
916 			goto out;
917 		}
918 		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
919 		sense_len = ntohs(sense_len);
920 #if 0
921 		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
922 		    sense_len, data_segment_len);
923 #endif
924 		if (sizeof(sense_len) + sense_len > data_segment_len) {
925 			ISCSI_SESSION_WARN(is, "truncated data segment "
926 			    "(%zd bytes, should be %zd)",
927 			    data_segment_len, sizeof(sense_len) + sense_len);
928 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
929 				xpt_freeze_devq(ccb->ccb_h.path, 1);
930 				ISCSI_SESSION_DEBUG(is, "freezing devq");
931 			}
932 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
933 			goto out;
934 		} else if (sizeof(sense_len) + sense_len < data_segment_len)
935 			ISCSI_SESSION_WARN(is, "oversize data segment "
936 			    "(%zd bytes, should be %zd)",
937 			    data_segment_len, sizeof(sense_len) + sense_len);
938 		if (sense_len > csio->sense_len) {
939 			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
940 			    sense_len, csio->sense_len);
941 			sense_len = csio->sense_len;
942 		}
943 		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
944 		csio->sense_resid = csio->sense_len - sense_len;
945 		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
946 	}
947 
948 out:
949 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
950 		csio->resid = ntohl(bhssr->bhssr_residual_count);
951 
952 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
953 		KASSERT(received <= csio->dxfer_len,
954 		    ("received > csio->dxfer_len"));
955 		if (received < csio->dxfer_len) {
956 			if (csio->resid != csio->dxfer_len - received) {
957 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
958 				    "target indicates %d, we calculated %zd",
959 				    csio->resid, csio->dxfer_len - received);
960 			}
961 			csio->resid = csio->dxfer_len - received;
962 		}
963 	}
964 
965 	xpt_done(ccb);
966 	icl_pdu_free(response);
967 }
968 
969 static void
970 iscsi_pdu_handle_task_response(struct icl_pdu *response)
971 {
972 	struct iscsi_bhs_task_management_response *bhstmr;
973 	struct iscsi_outstanding *io, *aio;
974 	struct iscsi_session *is;
975 
976 	is = PDU_SESSION(response);
977 
978 	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
979 	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
980 	if (io == NULL || io->io_ccb != NULL) {
981 		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
982 		    bhstmr->bhstmr_initiator_task_tag);
983 		icl_pdu_free(response);
984 		iscsi_session_reconnect(is);
985 		return;
986 	}
987 
988 	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
989 		ISCSI_SESSION_WARN(is, "task response 0x%x",
990 		    bhstmr->bhstmr_response);
991 	} else {
992 		aio = iscsi_outstanding_find(is, io->io_datasn);
993 		if (aio != NULL && aio->io_ccb != NULL)
994 			iscsi_session_terminate_task(is, aio, false);
995 	}
996 
997 	iscsi_outstanding_remove(is, io);
998 	icl_pdu_free(response);
999 }
1000 
1001 static void
1002 iscsi_pdu_handle_data_in(struct icl_pdu *response)
1003 {
1004 	struct iscsi_bhs_data_in *bhsdi;
1005 	struct iscsi_outstanding *io;
1006 	struct iscsi_session *is;
1007 	union ccb *ccb;
1008 	struct ccb_scsiio *csio;
1009 	size_t data_segment_len, received, oreceived;
1010 
1011 	is = PDU_SESSION(response);
1012 	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
1013 	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
1014 	if (io == NULL || io->io_ccb == NULL) {
1015 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
1016 		icl_pdu_free(response);
1017 		iscsi_session_reconnect(is);
1018 		ISCSI_SESSION_UNLOCK(is);
1019 		return;
1020 	}
1021 
1022 	data_segment_len = icl_pdu_data_segment_length(response);
1023 	if (data_segment_len == 0) {
1024 		/*
1025 		 * "The sending of 0 length data segments should be avoided,
1026 		 * but initiators and targets MUST be able to properly receive
1027 		 * 0 length data segments."
1028 		 */
1029 		ISCSI_SESSION_UNLOCK(is);
1030 		icl_pdu_free(response);
1031 		return;
1032 	}
1033 
1034 	/*
1035 	 * We need to track this for security reasons - without it, malicious target
1036 	 * could respond to SCSI READ without sending Data-In PDUs, which would result
1037 	 * in read operation on the initiator side returning random kernel data.
1038 	 */
1039 	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1040 		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1041 		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1042 		icl_pdu_free(response);
1043 		iscsi_session_reconnect(is);
1044 		ISCSI_SESSION_UNLOCK(is);
1045 		return;
1046 	}
1047 
1048 	ccb = io->io_ccb;
1049 	csio = &ccb->csio;
1050 
1051 	if (io->io_received + data_segment_len > csio->dxfer_len) {
1052 		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1053 		    "at offset %zd, buffer is %d)",
1054 		    data_segment_len, io->io_received, csio->dxfer_len);
1055 		icl_pdu_free(response);
1056 		iscsi_session_reconnect(is);
1057 		ISCSI_SESSION_UNLOCK(is);
1058 		return;
1059 	}
1060 
1061 	oreceived = io->io_received;
1062 	io->io_received += data_segment_len;
1063 	received = io->io_received;
1064 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0)
1065 		iscsi_outstanding_remove(is, io);
1066 	ISCSI_SESSION_UNLOCK(is);
1067 
1068 	icl_pdu_get_data(response, 0, csio->data_ptr + oreceived, data_segment_len);
1069 
1070 	/*
1071 	 * XXX: Check DataSN.
1072 	 * XXX: Check F.
1073 	 */
1074 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1075 		/*
1076 		 * Nothing more to do.
1077 		 */
1078 		icl_pdu_free(response);
1079 		return;
1080 	}
1081 
1082 	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1083 	if (bhsdi->bhsdi_status == 0) {
1084 		ccb->ccb_h.status = CAM_REQ_CMP;
1085 	} else {
1086 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1087 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1088 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1089 		}
1090 		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1091 		csio->scsi_status = bhsdi->bhsdi_status;
1092 	}
1093 
1094 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1095 		KASSERT(received <= csio->dxfer_len,
1096 		    ("received > csio->dxfer_len"));
1097 		if (received < csio->dxfer_len) {
1098 			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1099 			if (csio->resid != csio->dxfer_len - received) {
1100 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1101 				    "target indicates %d, we calculated %zd",
1102 				    csio->resid, csio->dxfer_len - received);
1103 			}
1104 			csio->resid = csio->dxfer_len - received;
1105 		}
1106 	}
1107 
1108 	xpt_done(ccb);
1109 	icl_pdu_free(response);
1110 }
1111 
1112 static void
1113 iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1114 {
1115 
1116 	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1117 	icl_pdu_free(response);
1118 }
1119 
1120 static void
1121 iscsi_pdu_handle_r2t(struct icl_pdu *response)
1122 {
1123 	struct icl_pdu *request;
1124 	struct iscsi_session *is;
1125 	struct iscsi_bhs_r2t *bhsr2t;
1126 	struct iscsi_bhs_data_out *bhsdo;
1127 	struct iscsi_outstanding *io;
1128 	struct ccb_scsiio *csio;
1129 	size_t off, len, total_len;
1130 	int error;
1131 
1132 	is = PDU_SESSION(response);
1133 
1134 	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1135 	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1136 	if (io == NULL || io->io_ccb == NULL) {
1137 		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1138 		    bhsr2t->bhsr2t_initiator_task_tag);
1139 		icl_pdu_free(response);
1140 		iscsi_session_reconnect(is);
1141 		return;
1142 	}
1143 
1144 	csio = &io->io_ccb->csio;
1145 
1146 	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1147 		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1148 		icl_pdu_free(response);
1149 		iscsi_session_reconnect(is);
1150 		return;
1151 	}
1152 
1153 	/*
1154 	 * XXX: Verify R2TSN.
1155 	 */
1156 
1157 	io->io_datasn = 0;
1158 
1159 	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1160 	if (off > csio->dxfer_len) {
1161 		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1162 		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1163 		icl_pdu_free(response);
1164 		iscsi_session_reconnect(is);
1165 		return;
1166 	}
1167 
1168 	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1169 	if (total_len == 0 || total_len > csio->dxfer_len) {
1170 		ISCSI_SESSION_WARN(is, "target requested invalid length "
1171 		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1172 		icl_pdu_free(response);
1173 		iscsi_session_reconnect(is);
1174 		return;
1175 	}
1176 
1177 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1178 
1179 	for (;;) {
1180 		len = total_len;
1181 
1182 		if (len > is->is_max_data_segment_length)
1183 			len = is->is_max_data_segment_length;
1184 
1185 		if (off + len > csio->dxfer_len) {
1186 			ISCSI_SESSION_WARN(is, "target requested invalid "
1187 			    "length/offset %zd, buffer is %d; reconnecting",
1188 			    off + len, csio->dxfer_len);
1189 			icl_pdu_free(response);
1190 			iscsi_session_reconnect(is);
1191 			return;
1192 		}
1193 
1194 		request = icl_pdu_new(response->ip_conn, M_NOWAIT);
1195 		if (request == NULL) {
1196 			icl_pdu_free(response);
1197 			iscsi_session_reconnect(is);
1198 			return;
1199 		}
1200 
1201 		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1202 		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1203 		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1204 		bhsdo->bhsdo_initiator_task_tag =
1205 		    bhsr2t->bhsr2t_initiator_task_tag;
1206 		bhsdo->bhsdo_target_transfer_tag =
1207 		    bhsr2t->bhsr2t_target_transfer_tag;
1208 		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1209 		bhsdo->bhsdo_buffer_offset = htonl(off);
1210 		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1211 		    M_NOWAIT);
1212 		if (error != 0) {
1213 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1214 			    "reconnecting");
1215 			icl_pdu_free(request);
1216 			icl_pdu_free(response);
1217 			iscsi_session_reconnect(is);
1218 			return;
1219 		}
1220 
1221 		off += len;
1222 		total_len -= len;
1223 
1224 		if (total_len == 0) {
1225 			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1226 			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1227 		} else {
1228 			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1229 		}
1230 
1231 		iscsi_pdu_queue_locked(request);
1232 
1233 		if (total_len == 0)
1234 			break;
1235 	}
1236 
1237 	icl_pdu_free(response);
1238 }
1239 
1240 static void
1241 iscsi_pdu_handle_async_message(struct icl_pdu *response)
1242 {
1243 	struct iscsi_bhs_asynchronous_message *bhsam;
1244 	struct iscsi_session *is;
1245 
1246 	is = PDU_SESSION(response);
1247 	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1248 	switch (bhsam->bhsam_async_event) {
1249 	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1250 		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1251 		iscsi_session_logout(is);
1252 		iscsi_session_terminate(is);
1253 		break;
1254 	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1255 		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the connection");
1256 		break;
1257 	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1258 		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the session");
1259 		break;
1260 	default:
1261 		/*
1262 		 * XXX: Technically, we're obligated to also handle
1263 		 * 	parameter renegotiation.
1264 		 */
1265 		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1266 		break;
1267 	}
1268 
1269 	icl_pdu_free(response);
1270 }
1271 
1272 static void
1273 iscsi_pdu_handle_reject(struct icl_pdu *response)
1274 {
1275 	struct iscsi_bhs_reject *bhsr;
1276 	struct iscsi_session *is;
1277 
1278 	is = PDU_SESSION(response);
1279 	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1280 	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1281 	    bhsr->bhsr_reason);
1282 
1283 	icl_pdu_free(response);
1284 }
1285 
1286 static int
1287 iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1288     struct iscsi_daemon_request *request)
1289 {
1290 	struct iscsi_session *is;
1291 	int error;
1292 
1293 	sx_slock(&sc->sc_lock);
1294 	for (;;) {
1295 		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1296 			ISCSI_SESSION_LOCK(is);
1297 			if (is->is_waiting_for_iscsid)
1298 				break;
1299 			ISCSI_SESSION_UNLOCK(is);
1300 		}
1301 
1302 		if (is == NULL) {
1303 			/*
1304 			 * No session requires attention from iscsid(8); wait.
1305 			 */
1306 			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1307 			if (error != 0) {
1308 				sx_sunlock(&sc->sc_lock);
1309 				return (error);
1310 			}
1311 			continue;
1312 		}
1313 
1314 		is->is_waiting_for_iscsid = false;
1315 		is->is_login_phase = true;
1316 		is->is_reason[0] = '\0';
1317 		ISCSI_SESSION_UNLOCK(is);
1318 
1319 		request->idr_session_id = is->is_id;
1320 		memcpy(&request->idr_isid, &is->is_isid,
1321 		    sizeof(request->idr_isid));
1322 		request->idr_tsih = 0;	/* New or reinstated session. */
1323 		memcpy(&request->idr_conf, &is->is_conf,
1324 		    sizeof(request->idr_conf));
1325 
1326 		error = icl_limits(is->is_conf.isc_offload,
1327 		    is->is_conf.isc_iser,
1328 		    &request->idr_limits.isl_max_data_segment_length);
1329 		if (error != 0) {
1330 			ISCSI_SESSION_WARN(is, "icl_limits for offload \"%s\" "
1331 			    "failed with error %d", is->is_conf.isc_offload,
1332 			    error);
1333 			sx_sunlock(&sc->sc_lock);
1334 			return (error);
1335 		}
1336 
1337 		sx_sunlock(&sc->sc_lock);
1338 		return (0);
1339 	}
1340 }
1341 
1342 static int
1343 iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1344     struct iscsi_daemon_handoff *handoff)
1345 {
1346 	struct iscsi_session *is;
1347 	struct icl_conn *ic;
1348 	int error;
1349 
1350 	sx_slock(&sc->sc_lock);
1351 
1352 	/*
1353 	 * Find the session to hand off socket to.
1354 	 */
1355 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1356 		if (is->is_id == handoff->idh_session_id)
1357 			break;
1358 	}
1359 	if (is == NULL) {
1360 		sx_sunlock(&sc->sc_lock);
1361 		return (ESRCH);
1362 	}
1363 	ISCSI_SESSION_LOCK(is);
1364 	ic = is->is_conn;
1365 	if (is->is_conf.isc_discovery || is->is_terminating) {
1366 		ISCSI_SESSION_UNLOCK(is);
1367 		sx_sunlock(&sc->sc_lock);
1368 		return (EINVAL);
1369 	}
1370 	if (is->is_connected) {
1371 		/*
1372 		 * This might have happened because another iscsid(8)
1373 		 * instance handed off the connection in the meantime.
1374 		 * Just return.
1375 		 */
1376 		ISCSI_SESSION_WARN(is, "handoff on already connected "
1377 		    "session");
1378 		ISCSI_SESSION_UNLOCK(is);
1379 		sx_sunlock(&sc->sc_lock);
1380 		return (EBUSY);
1381 	}
1382 
1383 	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1384 	    sizeof(is->is_target_alias));
1385 	is->is_tsih = handoff->idh_tsih;
1386 	is->is_statsn = handoff->idh_statsn;
1387 	is->is_initial_r2t = handoff->idh_initial_r2t;
1388 	is->is_immediate_data = handoff->idh_immediate_data;
1389 
1390 	/*
1391 	 * Cap MaxRecvDataSegmentLength obtained from the target to the maximum
1392 	 * size supported by our ICL module.
1393 	 */
1394 	is->is_max_data_segment_length = min(ic->ic_max_data_segment_length,
1395 	    handoff->idh_max_data_segment_length);
1396 	is->is_max_burst_length = handoff->idh_max_burst_length;
1397 	is->is_first_burst_length = handoff->idh_first_burst_length;
1398 
1399 	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1400 		ic->ic_header_crc32c = true;
1401 	else
1402 		ic->ic_header_crc32c = false;
1403 	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1404 		ic->ic_data_crc32c = true;
1405 	else
1406 		ic->ic_data_crc32c = false;
1407 	ic->ic_maxtags = maxtags;
1408 
1409 	is->is_cmdsn = 0;
1410 	is->is_expcmdsn = 0;
1411 	is->is_maxcmdsn = 0;
1412 	is->is_waiting_for_iscsid = false;
1413 	is->is_login_phase = false;
1414 	is->is_timeout = 0;
1415 	is->is_connected = true;
1416 	is->is_reason[0] = '\0';
1417 
1418 	ISCSI_SESSION_UNLOCK(is);
1419 
1420 	/*
1421 	 * If we're going through the proxy, the idh_socket will be 0,
1422 	 * and the ICL module can simply ignore this call.  It can also
1423 	 * use it to determine it's no longer in the Login phase.
1424 	 */
1425 	error = icl_conn_handoff(ic, handoff->idh_socket);
1426 	if (error != 0) {
1427 		sx_sunlock(&sc->sc_lock);
1428 		iscsi_session_terminate(is);
1429 		return (error);
1430 	}
1431 
1432 	sx_sunlock(&sc->sc_lock);
1433 
1434 	if (is->is_sim != NULL) {
1435 		/*
1436 		 * When reconnecting, there already is SIM allocated for the session.
1437 		 */
1438 		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1439 		ISCSI_SESSION_LOCK(is);
1440 		ISCSI_SESSION_DEBUG(is, "releasing");
1441 		xpt_release_simq(is->is_sim, 1);
1442 		is->is_simq_frozen = false;
1443 		ISCSI_SESSION_UNLOCK(is);
1444 
1445 	} else {
1446 		ISCSI_SESSION_LOCK(is);
1447 		is->is_devq = cam_simq_alloc(ic->ic_maxtags);
1448 		if (is->is_devq == NULL) {
1449 			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1450 			iscsi_session_terminate(is);
1451 			return (ENOMEM);
1452 		}
1453 
1454 		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1455 		    is, is->is_id /* unit */, &is->is_lock,
1456 		    1, ic->ic_maxtags, is->is_devq);
1457 		if (is->is_sim == NULL) {
1458 			ISCSI_SESSION_UNLOCK(is);
1459 			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1460 			cam_simq_free(is->is_devq);
1461 			iscsi_session_terminate(is);
1462 			return (ENOMEM);
1463 		}
1464 
1465 		error = xpt_bus_register(is->is_sim, NULL, 0);
1466 		if (error != 0) {
1467 			ISCSI_SESSION_UNLOCK(is);
1468 			ISCSI_SESSION_WARN(is, "failed to register bus");
1469 			iscsi_session_terminate(is);
1470 			return (ENOMEM);
1471 		}
1472 
1473 		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1474 		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1475 		    CAM_LUN_WILDCARD);
1476 		if (error != CAM_REQ_CMP) {
1477 			ISCSI_SESSION_UNLOCK(is);
1478 			ISCSI_SESSION_WARN(is, "failed to create path");
1479 			iscsi_session_terminate(is);
1480 			return (ENOMEM);
1481 		}
1482 		ISCSI_SESSION_UNLOCK(is);
1483 	}
1484 
1485 	return (0);
1486 }
1487 
1488 static int
1489 iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1490     struct iscsi_daemon_fail *fail)
1491 {
1492 	struct iscsi_session *is;
1493 
1494 	sx_slock(&sc->sc_lock);
1495 
1496 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1497 		if (is->is_id == fail->idf_session_id)
1498 			break;
1499 	}
1500 	if (is == NULL) {
1501 		sx_sunlock(&sc->sc_lock);
1502 		return (ESRCH);
1503 	}
1504 	ISCSI_SESSION_LOCK(is);
1505 	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1506 	    fail->idf_reason);
1507 	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1508 	//is->is_waiting_for_iscsid = false;
1509 	//is->is_login_phase = true;
1510 	//iscsi_session_reconnect(is);
1511 	ISCSI_SESSION_UNLOCK(is);
1512 	sx_sunlock(&sc->sc_lock);
1513 
1514 	return (0);
1515 }
1516 
1517 #ifdef ICL_KERNEL_PROXY
1518 static int
1519 iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1520     struct iscsi_daemon_connect *idc)
1521 {
1522 	struct iscsi_session *is;
1523 	struct sockaddr *from_sa, *to_sa;
1524 	int error;
1525 
1526 	sx_slock(&sc->sc_lock);
1527 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1528 		if (is->is_id == idc->idc_session_id)
1529 			break;
1530 	}
1531 	if (is == NULL) {
1532 		sx_sunlock(&sc->sc_lock);
1533 		return (ESRCH);
1534 	}
1535 	sx_sunlock(&sc->sc_lock);
1536 
1537 	if (idc->idc_from_addrlen > 0) {
1538 		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1539 		if (error != 0) {
1540 			ISCSI_SESSION_WARN(is,
1541 			    "getsockaddr failed with error %d", error);
1542 			return (error);
1543 		}
1544 	} else {
1545 		from_sa = NULL;
1546 	}
1547 	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1548 	if (error != 0) {
1549 		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1550 		    error);
1551 		free(from_sa, M_SONAME);
1552 		return (error);
1553 	}
1554 
1555 	ISCSI_SESSION_LOCK(is);
1556 	is->is_statsn = 0;
1557 	is->is_cmdsn = 0;
1558 	is->is_expcmdsn = 0;
1559 	is->is_maxcmdsn = 0;
1560 	is->is_waiting_for_iscsid = false;
1561 	is->is_login_phase = true;
1562 	is->is_timeout = 0;
1563 	ISCSI_SESSION_UNLOCK(is);
1564 
1565 	error = icl_conn_connect(is->is_conn, idc->idc_domain,
1566 	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1567 	free(from_sa, M_SONAME);
1568 	free(to_sa, M_SONAME);
1569 
1570 	/*
1571 	 * Digests are always disabled during login phase.
1572 	 */
1573 	is->is_conn->ic_header_crc32c = false;
1574 	is->is_conn->ic_data_crc32c = false;
1575 
1576 	return (error);
1577 }
1578 
1579 static int
1580 iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1581     struct iscsi_daemon_send *ids)
1582 {
1583 	struct iscsi_session *is;
1584 	struct icl_pdu *ip;
1585 	size_t datalen;
1586 	void *data;
1587 	int error;
1588 
1589 	sx_slock(&sc->sc_lock);
1590 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1591 		if (is->is_id == ids->ids_session_id)
1592 			break;
1593 	}
1594 	if (is == NULL) {
1595 		sx_sunlock(&sc->sc_lock);
1596 		return (ESRCH);
1597 	}
1598 	sx_sunlock(&sc->sc_lock);
1599 
1600 	if (is->is_login_phase == false)
1601 		return (EBUSY);
1602 
1603 	if (is->is_terminating || is->is_reconnecting)
1604 		return (EIO);
1605 
1606 	datalen = ids->ids_data_segment_len;
1607 	if (datalen > ISCSI_MAX_DATA_SEGMENT_LENGTH)
1608 		return (EINVAL);
1609 	if (datalen > 0) {
1610 		data = malloc(datalen, M_ISCSI, M_WAITOK);
1611 		error = copyin(ids->ids_data_segment, data, datalen);
1612 		if (error != 0) {
1613 			free(data, M_ISCSI);
1614 			return (error);
1615 		}
1616 	}
1617 
1618 	ip = icl_pdu_new(is->is_conn, M_WAITOK);
1619 	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1620 	if (datalen > 0) {
1621 		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1622 		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1623 		free(data, M_ISCSI);
1624 	}
1625 	iscsi_pdu_queue(ip);
1626 
1627 	return (0);
1628 }
1629 
1630 static int
1631 iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1632     struct iscsi_daemon_receive *idr)
1633 {
1634 	struct iscsi_session *is;
1635 	struct icl_pdu *ip;
1636 	void *data;
1637 	int error;
1638 
1639 	sx_slock(&sc->sc_lock);
1640 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1641 		if (is->is_id == idr->idr_session_id)
1642 			break;
1643 	}
1644 	if (is == NULL) {
1645 		sx_sunlock(&sc->sc_lock);
1646 		return (ESRCH);
1647 	}
1648 	sx_sunlock(&sc->sc_lock);
1649 
1650 	if (is->is_login_phase == false)
1651 		return (EBUSY);
1652 
1653 	ISCSI_SESSION_LOCK(is);
1654 	while (is->is_login_pdu == NULL &&
1655 	    is->is_terminating == false &&
1656 	    is->is_reconnecting == false) {
1657 		error = cv_wait_sig(&is->is_login_cv, &is->is_lock);
1658 		if (error != 0) {
1659 			ISCSI_SESSION_UNLOCK(is);
1660 			return (error);
1661 		}
1662 	}
1663 	if (is->is_terminating || is->is_reconnecting) {
1664 		ISCSI_SESSION_UNLOCK(is);
1665 		return (EIO);
1666 	}
1667 	ip = is->is_login_pdu;
1668 	is->is_login_pdu = NULL;
1669 	ISCSI_SESSION_UNLOCK(is);
1670 
1671 	if (ip->ip_data_len > idr->idr_data_segment_len) {
1672 		icl_pdu_free(ip);
1673 		return (EMSGSIZE);
1674 	}
1675 
1676 	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1677 	if (ip->ip_data_len > 0) {
1678 		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1679 		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1680 		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1681 		free(data, M_ISCSI);
1682 	}
1683 
1684 	icl_pdu_free(ip);
1685 
1686 	return (0);
1687 }
1688 #endif /* ICL_KERNEL_PROXY */
1689 
1690 static void
1691 iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1692 {
1693 	/*
1694 	 * Just make sure all the fields are null-terminated.
1695 	 *
1696 	 * XXX: This is not particularly secure.  We should
1697 	 * 	create our own conf and then copy in relevant
1698 	 * 	fields.
1699 	 */
1700 	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1701 	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1702 	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1703 	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1704 	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1705 	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1706 	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1707 	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1708 	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1709 }
1710 
1711 static bool
1712 iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1713 {
1714 
1715 	if (isc->isc_initiator[0] == '\0') {
1716 		ISCSI_DEBUG("empty isc_initiator");
1717 		return (false);
1718 	}
1719 
1720 	if (isc->isc_target_addr[0] == '\0') {
1721 		ISCSI_DEBUG("empty isc_target_addr");
1722 		return (false);
1723 	}
1724 
1725 	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1726 		ISCSI_DEBUG("non-empty isc_target for discovery session");
1727 		return (false);
1728 	}
1729 
1730 	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1731 		ISCSI_DEBUG("empty isc_target for non-discovery session");
1732 		return (false);
1733 	}
1734 
1735 	return (true);
1736 }
1737 
1738 static int
1739 iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1740 {
1741 	struct iscsi_session *is;
1742 	const struct iscsi_session *is2;
1743 	int error;
1744 
1745 	iscsi_sanitize_session_conf(&isa->isa_conf);
1746 	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1747 		return (EINVAL);
1748 
1749 	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1750 	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1751 
1752 	sx_xlock(&sc->sc_lock);
1753 
1754 	/*
1755 	 * Prevent duplicates.
1756 	 */
1757 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1758 		if (!!is->is_conf.isc_discovery !=
1759 		    !!is2->is_conf.isc_discovery)
1760 			continue;
1761 
1762 		if (strcmp(is->is_conf.isc_target_addr,
1763 		    is2->is_conf.isc_target_addr) != 0)
1764 			continue;
1765 
1766 		if (is->is_conf.isc_discovery == 0 &&
1767 		    strcmp(is->is_conf.isc_target,
1768 		    is2->is_conf.isc_target) != 0)
1769 			continue;
1770 
1771 		sx_xunlock(&sc->sc_lock);
1772 		free(is, M_ISCSI);
1773 		return (EBUSY);
1774 	}
1775 
1776 	is->is_conn = icl_new_conn(is->is_conf.isc_offload,
1777 	    is->is_conf.isc_iser, "iscsi", &is->is_lock);
1778 	if (is->is_conn == NULL) {
1779 		sx_xunlock(&sc->sc_lock);
1780 		free(is, M_ISCSI);
1781 		return (EINVAL);
1782 	}
1783 	is->is_conn->ic_receive = iscsi_receive_callback;
1784 	is->is_conn->ic_error = iscsi_error_callback;
1785 	is->is_conn->ic_prv0 = is;
1786 	TAILQ_INIT(&is->is_outstanding);
1787 	STAILQ_INIT(&is->is_postponed);
1788 	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1789 	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1790 #ifdef ICL_KERNEL_PROXY
1791 	cv_init(&is->is_login_cv, "iscsi_login");
1792 #endif
1793 
1794 	is->is_softc = sc;
1795 	sc->sc_last_session_id++;
1796 	is->is_id = sc->sc_last_session_id;
1797 	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1798 	arc4rand(&is->is_isid[1], 5, 0);
1799 	is->is_tsih = 0;
1800 	callout_init(&is->is_callout, 1);
1801 
1802 	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1803 	if (error != 0) {
1804 		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1805 		sx_xunlock(&sc->sc_lock);
1806 		return (error);
1807 	}
1808 
1809 	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1810 	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1811 
1812 	/*
1813 	 * Trigger immediate reconnection.
1814 	 */
1815 	ISCSI_SESSION_LOCK(is);
1816 	is->is_waiting_for_iscsid = true;
1817 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1818 	ISCSI_SESSION_UNLOCK(is);
1819 	cv_signal(&sc->sc_cv);
1820 
1821 	sx_xunlock(&sc->sc_lock);
1822 
1823 	return (0);
1824 }
1825 
1826 static bool
1827 iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1828     unsigned int id2, const struct iscsi_session_conf *c2)
1829 {
1830 
1831 	if (id2 != 0 && id2 != id1)
1832 		return (false);
1833 	if (c2->isc_target[0] != '\0' &&
1834 	    strcmp(c1->isc_target, c2->isc_target) != 0)
1835 		return (false);
1836 	if (c2->isc_target_addr[0] != '\0' &&
1837 	    strcmp(c1->isc_target_addr, c2->isc_target_addr) != 0)
1838 		return (false);
1839 	return (true);
1840 }
1841 
1842 static int
1843 iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1844     struct iscsi_session_remove *isr)
1845 {
1846 	struct iscsi_session *is, *tmp;
1847 	bool found = false;
1848 
1849 	iscsi_sanitize_session_conf(&isr->isr_conf);
1850 
1851 	sx_xlock(&sc->sc_lock);
1852 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1853 		ISCSI_SESSION_LOCK(is);
1854 		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1855 		    isr->isr_session_id, &isr->isr_conf)) {
1856 			found = true;
1857 			iscsi_session_logout(is);
1858 			iscsi_session_terminate(is);
1859 		}
1860 		ISCSI_SESSION_UNLOCK(is);
1861 	}
1862 	sx_xunlock(&sc->sc_lock);
1863 
1864 	if (!found)
1865 		return (ESRCH);
1866 
1867 	return (0);
1868 }
1869 
1870 static int
1871 iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1872 {
1873 	int error;
1874 	unsigned int i = 0;
1875 	struct iscsi_session *is;
1876 	struct iscsi_session_state iss;
1877 
1878 	sx_slock(&sc->sc_lock);
1879 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1880 		if (i >= isl->isl_nentries) {
1881 			sx_sunlock(&sc->sc_lock);
1882 			return (EMSGSIZE);
1883 		}
1884 		memset(&iss, 0, sizeof(iss));
1885 		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1886 		iss.iss_id = is->is_id;
1887 		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1888 		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1889 		strlcpy(iss.iss_offload, is->is_conn->ic_offload, sizeof(iss.iss_offload));
1890 
1891 		if (is->is_conn->ic_header_crc32c)
1892 			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1893 		else
1894 			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1895 
1896 		if (is->is_conn->ic_data_crc32c)
1897 			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1898 		else
1899 			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1900 
1901 		iss.iss_max_data_segment_length = is->is_max_data_segment_length;
1902 		iss.iss_immediate_data = is->is_immediate_data;
1903 		iss.iss_connected = is->is_connected;
1904 
1905 		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1906 		if (error != 0) {
1907 			sx_sunlock(&sc->sc_lock);
1908 			return (error);
1909 		}
1910 		i++;
1911 	}
1912 	sx_sunlock(&sc->sc_lock);
1913 
1914 	isl->isl_nentries = i;
1915 
1916 	return (0);
1917 }
1918 
1919 static int
1920 iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1921     struct iscsi_session_modify *ism)
1922 {
1923 	struct iscsi_session *is;
1924 
1925 	iscsi_sanitize_session_conf(&ism->ism_conf);
1926 	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1927 		return (EINVAL);
1928 
1929 	sx_xlock(&sc->sc_lock);
1930 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1931 		ISCSI_SESSION_LOCK(is);
1932 		if (is->is_id == ism->ism_session_id)
1933 			break;
1934 		ISCSI_SESSION_UNLOCK(is);
1935 	}
1936 	if (is == NULL) {
1937 		sx_xunlock(&sc->sc_lock);
1938 		return (ESRCH);
1939 	}
1940 	sx_xunlock(&sc->sc_lock);
1941 
1942 	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
1943 	ISCSI_SESSION_UNLOCK(is);
1944 
1945 	iscsi_session_reconnect(is);
1946 
1947 	return (0);
1948 }
1949 
1950 static int
1951 iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
1952     struct thread *td)
1953 {
1954 	struct iscsi_softc *sc;
1955 
1956 	sc = dev->si_drv1;
1957 
1958 	switch (cmd) {
1959 	case ISCSIDWAIT:
1960 		return (iscsi_ioctl_daemon_wait(sc,
1961 		    (struct iscsi_daemon_request *)arg));
1962 	case ISCSIDHANDOFF:
1963 		return (iscsi_ioctl_daemon_handoff(sc,
1964 		    (struct iscsi_daemon_handoff *)arg));
1965 	case ISCSIDFAIL:
1966 		return (iscsi_ioctl_daemon_fail(sc,
1967 		    (struct iscsi_daemon_fail *)arg));
1968 #ifdef ICL_KERNEL_PROXY
1969 	case ISCSIDCONNECT:
1970 		return (iscsi_ioctl_daemon_connect(sc,
1971 		    (struct iscsi_daemon_connect *)arg));
1972 	case ISCSIDSEND:
1973 		return (iscsi_ioctl_daemon_send(sc,
1974 		    (struct iscsi_daemon_send *)arg));
1975 	case ISCSIDRECEIVE:
1976 		return (iscsi_ioctl_daemon_receive(sc,
1977 		    (struct iscsi_daemon_receive *)arg));
1978 #endif /* ICL_KERNEL_PROXY */
1979 	case ISCSISADD:
1980 		return (iscsi_ioctl_session_add(sc,
1981 		    (struct iscsi_session_add *)arg));
1982 	case ISCSISREMOVE:
1983 		return (iscsi_ioctl_session_remove(sc,
1984 		    (struct iscsi_session_remove *)arg));
1985 	case ISCSISLIST:
1986 		return (iscsi_ioctl_session_list(sc,
1987 		    (struct iscsi_session_list *)arg));
1988 	case ISCSISMODIFY:
1989 		return (iscsi_ioctl_session_modify(sc,
1990 		    (struct iscsi_session_modify *)arg));
1991 	default:
1992 		return (EINVAL);
1993 	}
1994 }
1995 
1996 static struct iscsi_outstanding *
1997 iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
1998 {
1999 	struct iscsi_outstanding *io;
2000 
2001 	ISCSI_SESSION_LOCK_ASSERT(is);
2002 
2003 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2004 		if (io->io_initiator_task_tag == initiator_task_tag)
2005 			return (io);
2006 	}
2007 	return (NULL);
2008 }
2009 
2010 static struct iscsi_outstanding *
2011 iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
2012 {
2013 	struct iscsi_outstanding *io;
2014 
2015 	ISCSI_SESSION_LOCK_ASSERT(is);
2016 
2017 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2018 		if (io->io_ccb == ccb)
2019 			return (io);
2020 	}
2021 	return (NULL);
2022 }
2023 
2024 static struct iscsi_outstanding *
2025 iscsi_outstanding_add(struct iscsi_session *is, struct icl_pdu *request,
2026     union ccb *ccb, uint32_t *initiator_task_tagp)
2027 {
2028 	struct iscsi_outstanding *io;
2029 	int error;
2030 
2031 	ISCSI_SESSION_LOCK_ASSERT(is);
2032 
2033 	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
2034 	if (io == NULL) {
2035 		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes",
2036 		    sizeof(*io));
2037 		return (NULL);
2038 	}
2039 
2040 	error = icl_conn_task_setup(is->is_conn, request, &ccb->csio,
2041 	    initiator_task_tagp, &io->io_icl_prv);
2042 	if (error != 0) {
2043 		ISCSI_SESSION_WARN(is,
2044 		    "icl_conn_task_setup() failed with error %d", error);
2045 		uma_zfree(iscsi_outstanding_zone, io);
2046 		return (NULL);
2047 	}
2048 
2049 	KASSERT(iscsi_outstanding_find(is, *initiator_task_tagp) == NULL,
2050 	    ("initiator_task_tag 0x%x already added", *initiator_task_tagp));
2051 
2052 	io->io_initiator_task_tag = *initiator_task_tagp;
2053 	io->io_ccb = ccb;
2054 	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
2055 	return (io);
2056 }
2057 
2058 static void
2059 iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2060 {
2061 
2062 	ISCSI_SESSION_LOCK_ASSERT(is);
2063 
2064 	icl_conn_task_done(is->is_conn, io->io_icl_prv);
2065 	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2066 	uma_zfree(iscsi_outstanding_zone, io);
2067 }
2068 
2069 static void
2070 iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2071 {
2072 	struct icl_pdu *request;
2073 	struct iscsi_bhs_task_management_request *bhstmr;
2074 	struct ccb_abort *cab = &ccb->cab;
2075 	struct iscsi_outstanding *io, *aio;
2076 	uint32_t initiator_task_tag;
2077 
2078 	ISCSI_SESSION_LOCK_ASSERT(is);
2079 
2080 #if 0
2081 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2082 #else
2083 	if (is->is_login_phase) {
2084 		ccb->ccb_h.status = CAM_REQ_ABORTED;
2085 		xpt_done(ccb);
2086 		return;
2087 	}
2088 #endif
2089 
2090 	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2091 	if (aio == NULL) {
2092 		ccb->ccb_h.status = CAM_REQ_CMP;
2093 		xpt_done(ccb);
2094 		return;
2095 	}
2096 
2097 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2098 	if (request == NULL) {
2099 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2100 		xpt_done(ccb);
2101 		return;
2102 	}
2103 
2104 	initiator_task_tag = is->is_initiator_task_tag++;
2105 
2106 	io = iscsi_outstanding_add(is, request, NULL, &initiator_task_tag);
2107 	if (io == NULL) {
2108 		icl_pdu_free(request);
2109 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2110 		xpt_done(ccb);
2111 		return;
2112 	}
2113 	io->io_datasn = aio->io_initiator_task_tag;
2114 
2115 	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2116 	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2117 	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2118 	bhstmr->bhstmr_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2119 	bhstmr->bhstmr_initiator_task_tag = initiator_task_tag;
2120 	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2121 
2122 	iscsi_pdu_queue_locked(request);
2123 }
2124 
2125 static void
2126 iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2127 {
2128 	struct icl_pdu *request;
2129 	struct iscsi_bhs_scsi_command *bhssc;
2130 	struct ccb_scsiio *csio;
2131 	struct iscsi_outstanding *io;
2132 	size_t len;
2133 	uint32_t initiator_task_tag;
2134 	int error;
2135 
2136 	ISCSI_SESSION_LOCK_ASSERT(is);
2137 
2138 #if 0
2139 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2140 #else
2141 	if (is->is_login_phase) {
2142 		ISCSI_SESSION_DEBUG(is, "called during login phase");
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_REQ_ABORTED | CAM_DEV_QFRZN;
2148 		xpt_done(ccb);
2149 		return;
2150 	}
2151 #endif
2152 
2153 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2154 	if (request == NULL) {
2155 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2156 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2157 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2158 		}
2159 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2160 		xpt_done(ccb);
2161 		return;
2162 	}
2163 
2164 	initiator_task_tag = is->is_initiator_task_tag++;
2165 	io = iscsi_outstanding_add(is, request, ccb, &initiator_task_tag);
2166 	if (io == NULL) {
2167 		icl_pdu_free(request);
2168 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2169 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2170 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2171 		}
2172 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2173 		xpt_done(ccb);
2174 		return;
2175 	}
2176 
2177 	csio = &ccb->csio;
2178 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2179 	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2180 	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2181 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2182 	case CAM_DIR_IN:
2183 		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2184 		break;
2185 	case CAM_DIR_OUT:
2186 		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2187 		break;
2188 	}
2189 
2190 	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2191 		switch (csio->tag_action) {
2192 		case MSG_HEAD_OF_Q_TAG:
2193 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2194 			break;
2195 		case MSG_ORDERED_Q_TAG:
2196 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2197 			break;
2198 		case MSG_ACA_TASK:
2199 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2200 			break;
2201 		case MSG_SIMPLE_Q_TAG:
2202 		default:
2203 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2204 			break;
2205 		}
2206 	} else
2207 		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2208 
2209 	bhssc->bhssc_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2210 	bhssc->bhssc_initiator_task_tag = initiator_task_tag;
2211 	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2212 	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2213 	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2214 
2215 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2216 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2217 	else
2218 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2219 
2220 	if (is->is_immediate_data &&
2221 	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2222 		len = csio->dxfer_len;
2223 		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2224 		if (len > is->is_first_burst_length) {
2225 			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_first_burst_length);
2226 			len = is->is_first_burst_length;
2227 		}
2228 		if (len > is->is_max_data_segment_length) {
2229 			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_max_data_segment_length);
2230 			len = is->is_max_data_segment_length;
2231 		}
2232 
2233 		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2234 		if (error != 0) {
2235 			iscsi_outstanding_remove(is, io);
2236 			icl_pdu_free(request);
2237 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2238 				xpt_freeze_devq(ccb->ccb_h.path, 1);
2239 				ISCSI_SESSION_DEBUG(is, "freezing devq");
2240 			}
2241 			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2242 			xpt_done(ccb);
2243 			return;
2244 		}
2245 	}
2246 	iscsi_pdu_queue_locked(request);
2247 }
2248 
2249 static void
2250 iscsi_action(struct cam_sim *sim, union ccb *ccb)
2251 {
2252 	struct iscsi_session *is;
2253 
2254 	is = cam_sim_softc(sim);
2255 
2256 	ISCSI_SESSION_LOCK_ASSERT(is);
2257 
2258 	if (is->is_terminating ||
2259 	    (is->is_connected == false && fail_on_disconnection)) {
2260 		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2261 		xpt_done(ccb);
2262 		return;
2263 	}
2264 
2265 	switch (ccb->ccb_h.func_code) {
2266 	case XPT_PATH_INQ:
2267 	{
2268 		struct ccb_pathinq *cpi = &ccb->cpi;
2269 
2270 		cpi->version_num = 1;
2271 		cpi->hba_inquiry = PI_TAG_ABLE;
2272 		cpi->target_sprt = 0;
2273 		cpi->hba_misc = PIM_EXTLUNS;
2274 		/*
2275 		 * XXX: It shouldn't ever be NULL; this could be turned
2276 		 *      into a KASSERT eventually.
2277 		 */
2278 		if (is->is_conn == NULL)
2279 			ISCSI_WARN("NULL conn");
2280 		else if (is->is_conn->ic_unmapped)
2281 			cpi->hba_misc |= PIM_UNMAPPED;
2282 		cpi->hba_eng_cnt = 0;
2283 		cpi->max_target = 0;
2284 		/*
2285 		 * Note that the variable below is only relevant for targets
2286 		 * that don't claim compliance with anything above SPC2, which
2287 		 * means they don't support REPORT_LUNS.
2288 		 */
2289 		cpi->max_lun = 255;
2290 		cpi->initiator_id = ~0;
2291 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2292 		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2293 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2294 		cpi->unit_number = cam_sim_unit(sim);
2295 		cpi->bus_id = cam_sim_bus(sim);
2296 		cpi->base_transfer_speed = 150000; /* XXX */
2297 		cpi->transport = XPORT_ISCSI;
2298 		cpi->transport_version = 0;
2299 		cpi->protocol = PROTO_SCSI;
2300 		cpi->protocol_version = SCSI_REV_SPC3;
2301 		cpi->maxio = MAXPHYS;
2302 		cpi->ccb_h.status = CAM_REQ_CMP;
2303 		break;
2304 	}
2305 	case XPT_GET_TRAN_SETTINGS:
2306 	{
2307 		struct ccb_trans_settings	*cts;
2308 		struct ccb_trans_settings_scsi	*scsi;
2309 
2310 		cts = &ccb->cts;
2311 		scsi = &cts->proto_specific.scsi;
2312 
2313 		cts->protocol = PROTO_SCSI;
2314 		cts->protocol_version = SCSI_REV_SPC3;
2315 		cts->transport = XPORT_ISCSI;
2316 		cts->transport_version = 0;
2317 		scsi->valid = CTS_SCSI_VALID_TQ;
2318 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2319 		cts->ccb_h.status = CAM_REQ_CMP;
2320 		break;
2321 	}
2322 	case XPT_CALC_GEOMETRY:
2323 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2324 		ccb->ccb_h.status = CAM_REQ_CMP;
2325 		break;
2326 #if 0
2327 	/*
2328 	 * XXX: What's the point?
2329 	 */
2330 	case XPT_RESET_BUS:
2331 	case XPT_TERM_IO:
2332 		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2333 		ccb->ccb_h.status = CAM_REQ_CMP;
2334 		break;
2335 #endif
2336 	case XPT_ABORT:
2337 		iscsi_action_abort(is, ccb);
2338 		return;
2339 	case XPT_SCSI_IO:
2340 		iscsi_action_scsiio(is, ccb);
2341 		return;
2342 	default:
2343 #if 0
2344 		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2345 #endif
2346 		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2347 		break;
2348 	}
2349 	xpt_done(ccb);
2350 }
2351 
2352 static void
2353 iscsi_poll(struct cam_sim *sim)
2354 {
2355 
2356 	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2357 }
2358 
2359 static void
2360 iscsi_terminate_sessions(struct iscsi_softc *sc)
2361 {
2362 	struct iscsi_session *is;
2363 
2364 	sx_slock(&sc->sc_lock);
2365 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2366 		iscsi_session_terminate(is);
2367 	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2368 		ISCSI_DEBUG("waiting for sessions to terminate");
2369 		cv_wait(&sc->sc_cv, &sc->sc_lock);
2370 	}
2371 	ISCSI_DEBUG("all sessions terminated");
2372 	sx_sunlock(&sc->sc_lock);
2373 }
2374 
2375 static void
2376 iscsi_shutdown_pre(struct iscsi_softc *sc)
2377 {
2378 	struct iscsi_session *is;
2379 
2380 	if (!fail_on_shutdown)
2381 		return;
2382 
2383 	/*
2384 	 * If we have any sessions waiting for reconnection, request
2385 	 * maintenance thread to fail them immediately instead of waiting
2386 	 * for reconnect timeout.
2387 	 *
2388 	 * This prevents LUNs with mounted filesystems that are supported
2389 	 * by disconnected iSCSI sessions from hanging, however it will
2390 	 * fail all queued BIOs.
2391 	 */
2392 	ISCSI_DEBUG("forcing failing all disconnected sessions due to shutdown");
2393 
2394 	fail_on_disconnection = 1;
2395 
2396 	sx_slock(&sc->sc_lock);
2397 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2398 		ISCSI_SESSION_LOCK(is);
2399 		if (!is->is_connected) {
2400 			ISCSI_SESSION_DEBUG(is, "force failing disconnected session early");
2401 			iscsi_session_reconnect(is);
2402 		}
2403 		ISCSI_SESSION_UNLOCK(is);
2404 	}
2405 	sx_sunlock(&sc->sc_lock);
2406 }
2407 
2408 static void
2409 iscsi_shutdown_post(struct iscsi_softc *sc)
2410 {
2411 
2412 	ISCSI_DEBUG("removing all sessions due to shutdown");
2413 	iscsi_terminate_sessions(sc);
2414 }
2415 
2416 static int
2417 iscsi_load(void)
2418 {
2419 	int error;
2420 
2421 	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2422 	sx_init(&sc->sc_lock, "iscsi");
2423 	TAILQ_INIT(&sc->sc_sessions);
2424 	cv_init(&sc->sc_cv, "iscsi_cv");
2425 
2426 	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2427 	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2428 	    UMA_ALIGN_PTR, 0);
2429 
2430 	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2431 	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2432 	if (error != 0) {
2433 		ISCSI_WARN("failed to create device node, error %d", error);
2434 		return (error);
2435 	}
2436 	sc->sc_cdev->si_drv1 = sc;
2437 
2438 	sc->sc_shutdown_pre_eh = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2439 	    iscsi_shutdown_pre, sc, SHUTDOWN_PRI_FIRST);
2440 	/*
2441 	 * shutdown_post_sync needs to run after filesystem shutdown and before
2442 	 * CAM shutdown - otherwise when rebooting with an iSCSI session that is
2443 	 * disconnected but has outstanding requests, dashutdown() will hang on
2444 	 * cam_periph_runccb().
2445 	 */
2446 	sc->sc_shutdown_post_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2447 	    iscsi_shutdown_post, sc, SHUTDOWN_PRI_DEFAULT - 1);
2448 
2449 	return (0);
2450 }
2451 
2452 static int
2453 iscsi_unload(void)
2454 {
2455 
2456 	if (sc->sc_cdev != NULL) {
2457 		ISCSI_DEBUG("removing device node");
2458 		destroy_dev(sc->sc_cdev);
2459 		ISCSI_DEBUG("device node removed");
2460 	}
2461 
2462 	if (sc->sc_shutdown_pre_eh != NULL)
2463 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, sc->sc_shutdown_pre_eh);
2464 	if (sc->sc_shutdown_post_eh != NULL)
2465 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_post_eh);
2466 
2467 	iscsi_terminate_sessions(sc);
2468 
2469 	uma_zdestroy(iscsi_outstanding_zone);
2470 	sx_destroy(&sc->sc_lock);
2471 	cv_destroy(&sc->sc_cv);
2472 	free(sc, M_ISCSI);
2473 	return (0);
2474 }
2475 
2476 static int
2477 iscsi_quiesce(void)
2478 {
2479 	sx_slock(&sc->sc_lock);
2480 	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2481 		sx_sunlock(&sc->sc_lock);
2482 		return (EBUSY);
2483 	}
2484 	sx_sunlock(&sc->sc_lock);
2485 	return (0);
2486 }
2487 
2488 static int
2489 iscsi_modevent(module_t mod, int what, void *arg)
2490 {
2491 	int error;
2492 
2493 	switch (what) {
2494 	case MOD_LOAD:
2495 		error = iscsi_load();
2496 		break;
2497 	case MOD_UNLOAD:
2498 		error = iscsi_unload();
2499 		break;
2500 	case MOD_QUIESCE:
2501 		error = iscsi_quiesce();
2502 		break;
2503 	default:
2504 		error = EINVAL;
2505 		break;
2506 	}
2507 	return (error);
2508 }
2509 
2510 moduledata_t iscsi_data = {
2511 	"iscsi",
2512 	iscsi_modevent,
2513 	0
2514 };
2515 
2516 DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2517 MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2518 MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2519