xref: /freebsd/sys/dev/iscsi/iscsi.c (revision 25fb30bd9abc492359ad1f66901a06cb8cd08370)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2012 The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * This software was developed by Edward Tomasz Napierala under sponsorship
8  * from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include <sys/param.h>
37 #include <sys/condvar.h>
38 #include <sys/conf.h>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/file.h>
42 #include <sys/kernel.h>
43 #include <sys/kthread.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mutex.h>
47 #include <sys/module.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/sx.h>
52 #include <vm/uma.h>
53 
54 #include <cam/cam.h>
55 #include <cam/cam_ccb.h>
56 #include <cam/cam_xpt.h>
57 #include <cam/cam_debug.h>
58 #include <cam/cam_sim.h>
59 #include <cam/cam_xpt_sim.h>
60 #include <cam/cam_xpt_periph.h>
61 #include <cam/cam_periph.h>
62 #include <cam/scsi/scsi_all.h>
63 #include <cam/scsi/scsi_message.h>
64 
65 #include <dev/iscsi/icl.h>
66 #include <dev/iscsi/icl_wrappers.h>
67 #include <dev/iscsi/iscsi_ioctl.h>
68 #include <dev/iscsi/iscsi_proto.h>
69 #include <dev/iscsi/iscsi.h>
70 
71 #ifdef ICL_KERNEL_PROXY
72 #include <sys/socketvar.h>
73 #endif
74 
75 #ifdef ICL_KERNEL_PROXY
76 FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
77 #endif
78 
79 /*
80  * XXX: This is global so the iscsi_unload() can access it.
81  * 	Think about how to do this properly.
82  */
83 static struct iscsi_softc	*sc;
84 
85 SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
86     "iSCSI initiator");
87 static int debug = 1;
88 SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
89     &debug, 0, "Enable debug messages");
90 static int ping_timeout = 5;
91 SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
92     0, "Timeout for ping (NOP-Out) requests, in seconds");
93 static int iscsid_timeout = 60;
94 SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
95     0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
96 static int login_timeout = 60;
97 SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
98     0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
99 static int maxtags = 255;
100 SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
101     0, "Max number of IO requests queued");
102 static int fail_on_disconnection = 0;
103 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
104     &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
105 static int fail_on_shutdown = 1;
106 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_shutdown, CTLFLAG_RWTUN,
107     &fail_on_shutdown, 0, "Fail disconnected sessions on shutdown");
108 
109 static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
110 static uma_zone_t iscsi_outstanding_zone;
111 
112 #define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
113 #define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
114 
115 #define	ISCSI_DEBUG(X, ...)						\
116 	do {								\
117 		if (debug > 1) 						\
118 			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
119 	} while (0)
120 
121 #define	ISCSI_WARN(X, ...)						\
122 	do {								\
123 		if (debug > 0) {					\
124 			printf("WARNING: %s: " X "\n",			\
125 			    __func__, ## __VA_ARGS__);			\
126 		}							\
127 	} while (0)
128 
129 #define	ISCSI_SESSION_DEBUG(S, X, ...)					\
130 	do {								\
131 		if (debug > 1) {					\
132 			printf("%s: %s (%s): " X "\n",			\
133 			    __func__, S->is_conf.isc_target_addr,	\
134 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
135 		}							\
136 	} while (0)
137 
138 #define	ISCSI_SESSION_WARN(S, X, ...)					\
139 	do {								\
140 		if (debug > 0) {					\
141 			printf("WARNING: %s (%s): " X "\n",		\
142 			    S->is_conf.isc_target_addr,			\
143 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
144 		}							\
145 	} while (0)
146 
147 #define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
148 #define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
149 #define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
150 #define ISCSI_SESSION_LOCK_ASSERT_NOT(X) mtx_assert(&X->is_lock, MA_NOTOWNED)
151 
152 static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
153 		    int mode, struct thread *td);
154 
155 static struct cdevsw iscsi_cdevsw = {
156      .d_version = D_VERSION,
157      .d_ioctl   = iscsi_ioctl,
158      .d_name    = "iscsi",
159 };
160 
161 static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
162 static void	iscsi_pdu_queue(struct icl_pdu *request);
163 static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
164 static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
165 static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
166 static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
167 static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
168 static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
169 static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
170 static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
171 static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
172 static void	iscsi_session_reconnect(struct iscsi_session *is);
173 static void	iscsi_session_terminate(struct iscsi_session *is);
174 static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
175 static void	iscsi_poll(struct cam_sim *sim);
176 static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
177 		    uint32_t initiator_task_tag);
178 static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
179 		    struct icl_pdu *request, union ccb *ccb,
180 		    uint32_t *initiator_task_tagp);
181 static void	iscsi_outstanding_remove(struct iscsi_session *is,
182 		    struct iscsi_outstanding *io);
183 
184 static bool
185 iscsi_pdu_prepare(struct icl_pdu *request)
186 {
187 	struct iscsi_session *is;
188 	struct iscsi_bhs_scsi_command *bhssc;
189 
190 	is = PDU_SESSION(request);
191 
192 	ISCSI_SESSION_LOCK_ASSERT(is);
193 
194 	/*
195 	 * We're only using fields common for all the request
196 	 * (initiator -> target) PDUs.
197 	 */
198 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
199 
200 	/*
201 	 * Data-Out PDU does not contain CmdSN.
202 	 */
203 	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
204 		if (ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn) &&
205 		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
206 			/*
207 			 * Current MaxCmdSN prevents us from sending any more
208 			 * SCSI Command PDUs to the target; postpone the PDU.
209 			 * It will get resent by either iscsi_pdu_queue(),
210 			 * or by maintenance thread.
211 			 */
212 #if 0
213 			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %u, "
214 			    "ExpCmdSN %u, MaxCmdSN %u, opcode 0x%x",
215 			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn,
216 			    bhssc->bhssc_opcode);
217 #endif
218 			return (true);
219 		}
220 		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
221 		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
222 			is->is_cmdsn++;
223 	}
224 	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
225 
226 	return (false);
227 }
228 
229 static void
230 iscsi_session_send_postponed(struct iscsi_session *is)
231 {
232 	struct icl_pdu *request;
233 	bool postpone;
234 
235 	ISCSI_SESSION_LOCK_ASSERT(is);
236 
237 	if (STAILQ_EMPTY(&is->is_postponed))
238 		return;
239 	while ((request = STAILQ_FIRST(&is->is_postponed)) != NULL) {
240 		postpone = iscsi_pdu_prepare(request);
241 		if (postpone)
242 			return;
243 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
244 		icl_pdu_queue(request);
245 	}
246 	xpt_release_simq(is->is_sim, 1);
247 }
248 
249 static void
250 iscsi_pdu_queue_locked(struct icl_pdu *request)
251 {
252 	struct iscsi_session *is;
253 	bool postpone;
254 
255 	is = PDU_SESSION(request);
256 	ISCSI_SESSION_LOCK_ASSERT(is);
257 	iscsi_session_send_postponed(is);
258 	postpone = iscsi_pdu_prepare(request);
259 	if (postpone) {
260 		if (STAILQ_EMPTY(&is->is_postponed))
261 			xpt_freeze_simq(is->is_sim, 1);
262 		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
263 		return;
264 	}
265 	icl_pdu_queue(request);
266 }
267 
268 static void
269 iscsi_pdu_queue(struct icl_pdu *request)
270 {
271 	struct iscsi_session *is;
272 
273 	is = PDU_SESSION(request);
274 	ISCSI_SESSION_LOCK(is);
275 	iscsi_pdu_queue_locked(request);
276 	ISCSI_SESSION_UNLOCK(is);
277 }
278 
279 static void
280 iscsi_session_logout(struct iscsi_session *is)
281 {
282 	struct icl_pdu *request;
283 	struct iscsi_bhs_logout_request *bhslr;
284 
285 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
286 	if (request == NULL)
287 		return;
288 
289 	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
290 	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
291 	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
292 	iscsi_pdu_queue_locked(request);
293 }
294 
295 static void
296 iscsi_session_terminate_task(struct iscsi_session *is,
297     struct iscsi_outstanding *io, cam_status status)
298 {
299 
300 	ISCSI_SESSION_LOCK_ASSERT(is);
301 
302 	if (io->io_ccb != NULL) {
303 		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
304 		io->io_ccb->ccb_h.status |= status;
305 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
306 			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
307 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
308 			ISCSI_SESSION_DEBUG(is, "freezing devq");
309 		}
310 		xpt_done(io->io_ccb);
311 	}
312 	iscsi_outstanding_remove(is, io);
313 }
314 
315 static void
316 iscsi_session_terminate_tasks(struct iscsi_session *is, cam_status status)
317 {
318 	struct iscsi_outstanding *io, *tmp;
319 
320 	ISCSI_SESSION_LOCK_ASSERT(is);
321 
322 	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
323 		iscsi_session_terminate_task(is, io, status);
324 	}
325 }
326 
327 static void
328 iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
329 {
330 	struct icl_pdu *pdu;
331 
332 	ISCSI_SESSION_LOCK_ASSERT(is);
333 
334 	/*
335 	 * Don't queue any new PDUs.
336 	 */
337 	if (is->is_sim != NULL && is->is_simq_frozen == false) {
338 		ISCSI_SESSION_DEBUG(is, "freezing");
339 		xpt_freeze_simq(is->is_sim, 1);
340 		is->is_simq_frozen = true;
341 	}
342 
343 	/*
344 	 * Remove postponed PDUs.
345 	 */
346 	if (!STAILQ_EMPTY(&is->is_postponed))
347 		xpt_release_simq(is->is_sim, 1);
348 	while ((pdu = STAILQ_FIRST(&is->is_postponed)) != NULL) {
349 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
350 		icl_pdu_free(pdu);
351 	}
352 
353 	if (destroy_sim == false) {
354 		/*
355 		 * Terminate SCSI tasks, asking CAM to requeue them.
356 		 */
357 		iscsi_session_terminate_tasks(is, CAM_REQUEUE_REQ);
358 		return;
359 	}
360 
361 	iscsi_session_terminate_tasks(is, CAM_DEV_NOT_THERE);
362 
363 	if (is->is_sim == NULL)
364 		return;
365 
366 	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
367 	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
368 
369 	if (is->is_simq_frozen) {
370 		is->is_simq_frozen = false;
371 		xpt_release_simq(is->is_sim, 1);
372 	}
373 
374 	xpt_free_path(is->is_path);
375 	is->is_path = NULL;
376 	xpt_bus_deregister(cam_sim_path(is->is_sim));
377 	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
378 	is->is_sim = NULL;
379 	is->is_devq = NULL;
380 }
381 
382 static void
383 iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
384 {
385 
386 	icl_conn_close(is->is_conn);
387 
388 	ISCSI_SESSION_LOCK(is);
389 
390 	is->is_connected = false;
391 	is->is_reconnecting = false;
392 	is->is_login_phase = false;
393 
394 #ifdef ICL_KERNEL_PROXY
395 	if (is->is_login_pdu != NULL) {
396 		icl_pdu_free(is->is_login_pdu);
397 		is->is_login_pdu = NULL;
398 	}
399 	cv_signal(&is->is_login_cv);
400 #endif
401 
402 	if (fail_on_disconnection) {
403 		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
404 		iscsi_session_cleanup(is, true);
405 	} else {
406 		iscsi_session_cleanup(is, false);
407 	}
408 
409 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
410 	    ("destroying session with active tasks"));
411 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
412 	    ("destroying session with postponed PDUs"));
413 
414 	if (is->is_conf.isc_enable == 0 && is->is_conf.isc_discovery == 0) {
415 		ISCSI_SESSION_UNLOCK(is);
416 		return;
417 	}
418 
419 	/*
420 	 * Request immediate reconnection from iscsid(8).
421 	 */
422 	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
423 	is->is_waiting_for_iscsid = true;
424 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
425 	is->is_timeout = 0;
426 	ISCSI_SESSION_UNLOCK(is);
427 	cv_signal(&is->is_softc->sc_cv);
428 }
429 
430 static void
431 iscsi_maintenance_thread_terminate(struct iscsi_session *is)
432 {
433 	struct iscsi_softc *sc;
434 
435 	sc = is->is_softc;
436 	sx_xlock(&sc->sc_lock);
437 	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
438 	sx_xunlock(&sc->sc_lock);
439 
440 	icl_conn_close(is->is_conn);
441 	callout_drain(&is->is_callout);
442 
443 	ISCSI_SESSION_LOCK(is);
444 
445 	KASSERT(is->is_terminating, ("is_terminating == false"));
446 
447 #ifdef ICL_KERNEL_PROXY
448 	if (is->is_login_pdu != NULL) {
449 		icl_pdu_free(is->is_login_pdu);
450 		is->is_login_pdu = NULL;
451 	}
452 	cv_signal(&is->is_login_cv);
453 #endif
454 
455 	iscsi_session_cleanup(is, true);
456 
457 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
458 	    ("destroying session with active tasks"));
459 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
460 	    ("destroying session with postponed PDUs"));
461 
462 	ISCSI_SESSION_UNLOCK(is);
463 
464 	icl_conn_free(is->is_conn);
465 	mtx_destroy(&is->is_lock);
466 	cv_destroy(&is->is_maintenance_cv);
467 #ifdef ICL_KERNEL_PROXY
468 	cv_destroy(&is->is_login_cv);
469 #endif
470 
471 	ISCSI_SESSION_DEBUG(is, "terminated");
472 	free(is, M_ISCSI);
473 
474 	/*
475 	 * The iscsi_unload() routine might be waiting.
476 	 */
477 	cv_signal(&sc->sc_cv);
478 }
479 
480 static void
481 iscsi_maintenance_thread(void *arg)
482 {
483 	struct iscsi_session *is = arg;
484 
485 	ISCSI_SESSION_LOCK(is);
486 	for (;;) {
487 		if (is->is_reconnecting == false &&
488 		    is->is_terminating == false &&
489 		    (STAILQ_EMPTY(&is->is_postponed) ||
490 		     ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn)))
491 			cv_wait(&is->is_maintenance_cv, &is->is_lock);
492 
493 		/* Terminate supersedes reconnect. */
494 		if (is->is_terminating) {
495 			ISCSI_SESSION_UNLOCK(is);
496 			iscsi_maintenance_thread_terminate(is);
497 			kthread_exit();
498 			return;
499 		}
500 
501 		if (is->is_reconnecting) {
502 			ISCSI_SESSION_UNLOCK(is);
503 			iscsi_maintenance_thread_reconnect(is);
504 			ISCSI_SESSION_LOCK(is);
505 			continue;
506 		}
507 
508 		iscsi_session_send_postponed(is);
509 	}
510 	ISCSI_SESSION_UNLOCK(is);
511 }
512 
513 static void
514 iscsi_session_reconnect(struct iscsi_session *is)
515 {
516 
517 	/*
518 	 * XXX: We can't use locking here, because
519 	 * 	it's being called from various contexts.
520 	 * 	Hope it doesn't break anything.
521 	 */
522 	if (is->is_reconnecting)
523 		return;
524 
525 	is->is_reconnecting = true;
526 	cv_signal(&is->is_maintenance_cv);
527 }
528 
529 static void
530 iscsi_session_terminate(struct iscsi_session *is)
531 {
532 
533 	if (is->is_terminating)
534 		return;
535 
536 	is->is_terminating = true;
537 
538 #if 0
539 	iscsi_session_logout(is);
540 #endif
541 	cv_signal(&is->is_maintenance_cv);
542 }
543 
544 static void
545 iscsi_callout(void *context)
546 {
547 	struct icl_pdu *request;
548 	struct iscsi_bhs_nop_out *bhsno;
549 	struct iscsi_session *is;
550 	bool reconnect_needed = false;
551 
552 	is = context;
553 
554 	ISCSI_SESSION_LOCK(is);
555 	if (is->is_terminating) {
556 		ISCSI_SESSION_UNLOCK(is);
557 		return;
558 	}
559 
560 	callout_schedule(&is->is_callout, 1 * hz);
561 
562 	if (is->is_conf.isc_enable == 0)
563 		goto out;
564 
565 	is->is_timeout++;
566 
567 	if (is->is_waiting_for_iscsid) {
568 		if (iscsid_timeout > 0 && is->is_timeout > iscsid_timeout) {
569 			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
570 			    "for %d seconds; reconnecting",
571 			    is->is_timeout);
572 			reconnect_needed = true;
573 		}
574 		goto out;
575 	}
576 
577 	if (is->is_login_phase) {
578 		if (login_timeout > 0 && is->is_timeout > login_timeout) {
579 			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
580 			    "reconnecting", is->is_timeout);
581 			reconnect_needed = true;
582 		}
583 		goto out;
584 	}
585 
586 	if (ping_timeout <= 0) {
587 		/*
588 		 * Pings are disabled.  Don't send NOP-Out in this case.
589 		 * Reset the timeout, to avoid triggering reconnection,
590 		 * should the user decide to reenable them.
591 		 */
592 		is->is_timeout = 0;
593 		goto out;
594 	}
595 
596 	if (is->is_timeout >= ping_timeout) {
597 		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
598 		    "reconnecting", ping_timeout);
599 		reconnect_needed = true;
600 		goto out;
601 	}
602 
603 	ISCSI_SESSION_UNLOCK(is);
604 
605 	/*
606 	 * If the ping was reset less than one second ago - which means
607 	 * that we've received some PDU during the last second - assume
608 	 * the traffic flows correctly and don't bother sending a NOP-Out.
609 	 *
610 	 * (It's 2 - one for one second, and one for incrementing is_timeout
611 	 * earlier in this routine.)
612 	 */
613 	if (is->is_timeout < 2)
614 		return;
615 
616 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
617 	if (request == NULL) {
618 		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
619 		return;
620 	}
621 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
622 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
623 	    ISCSI_BHS_OPCODE_IMMEDIATE;
624 	bhsno->bhsno_flags = 0x80;
625 	bhsno->bhsno_target_transfer_tag = 0xffffffff;
626 	iscsi_pdu_queue(request);
627 	return;
628 
629 out:
630 	if (is->is_terminating) {
631 		ISCSI_SESSION_UNLOCK(is);
632 		return;
633 	}
634 
635 	ISCSI_SESSION_UNLOCK(is);
636 
637 	if (reconnect_needed)
638 		iscsi_session_reconnect(is);
639 }
640 
641 static void
642 iscsi_pdu_update_statsn(const struct icl_pdu *response)
643 {
644 	const struct iscsi_bhs_data_in *bhsdi;
645 	struct iscsi_session *is;
646 	uint32_t expcmdsn, maxcmdsn, statsn;
647 
648 	is = PDU_SESSION(response);
649 
650 	ISCSI_SESSION_LOCK_ASSERT(is);
651 
652 	/*
653 	 * We're only using fields common for all the response
654 	 * (target -> initiator) PDUs.
655 	 */
656 	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
657 	/*
658 	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
659 	 * and Residual Count only have meaningful content if the S bit
660 	 * is set to 1", so we also need to check the bit specific for
661 	 * Data-In PDU.
662 	 */
663 	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
664 	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
665 		statsn = ntohl(bhsdi->bhsdi_statsn);
666 		if (statsn != is->is_statsn && statsn != (is->is_statsn + 1)) {
667 			/* XXX: This is normal situation for MCS */
668 			ISCSI_SESSION_WARN(is, "PDU 0x%x StatSN %u != "
669 			    "session ExpStatSN %u (or + 1); reconnecting",
670 			    bhsdi->bhsdi_opcode, statsn, is->is_statsn);
671 			iscsi_session_reconnect(is);
672 		}
673 		if (ISCSI_SNGT(statsn, is->is_statsn))
674 			is->is_statsn = statsn;
675 	}
676 
677 	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
678 	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
679 
680 	if (ISCSI_SNLT(maxcmdsn + 1, expcmdsn)) {
681 		ISCSI_SESSION_DEBUG(is,
682 		    "PDU MaxCmdSN %u + 1 < PDU ExpCmdSN %u; ignoring",
683 		    maxcmdsn, expcmdsn);
684 	} else {
685 		if (ISCSI_SNGT(maxcmdsn, is->is_maxcmdsn)) {
686 			is->is_maxcmdsn = maxcmdsn;
687 
688 			/*
689 			 * Command window increased; kick the maintanance thread
690 			 * to send out postponed commands.
691 			 */
692 			if (!STAILQ_EMPTY(&is->is_postponed))
693 				cv_signal(&is->is_maintenance_cv);
694 		} else if (ISCSI_SNLT(maxcmdsn, is->is_maxcmdsn)) {
695 			/* XXX: This is normal situation for MCS */
696 			ISCSI_SESSION_DEBUG(is,
697 			    "PDU MaxCmdSN %u < session MaxCmdSN %u; ignoring",
698 			    maxcmdsn, is->is_maxcmdsn);
699 		}
700 
701 		if (ISCSI_SNGT(expcmdsn, is->is_expcmdsn)) {
702 			is->is_expcmdsn = expcmdsn;
703 		} else if (ISCSI_SNLT(expcmdsn, is->is_expcmdsn)) {
704 			/* XXX: This is normal situation for MCS */
705 			ISCSI_SESSION_DEBUG(is,
706 			    "PDU ExpCmdSN %u < session ExpCmdSN %u; ignoring",
707 			    expcmdsn, is->is_expcmdsn);
708 		}
709 	}
710 
711 	/*
712 	 * Every incoming PDU - not just NOP-In - resets the ping timer.
713 	 * The purpose of the timeout is to reset the connection when it stalls;
714 	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
715 	 * in some queue.
716 	 */
717 	is->is_timeout = 0;
718 }
719 
720 static void
721 iscsi_receive_callback(struct icl_pdu *response)
722 {
723 	struct iscsi_session *is;
724 
725 	is = PDU_SESSION(response);
726 
727 	ISCSI_SESSION_LOCK(is);
728 
729 	iscsi_pdu_update_statsn(response);
730 
731 #ifdef ICL_KERNEL_PROXY
732 	if (is->is_login_phase) {
733 		if (is->is_login_pdu == NULL)
734 			is->is_login_pdu = response;
735 		else
736 			icl_pdu_free(response);
737 		ISCSI_SESSION_UNLOCK(is);
738 		cv_signal(&is->is_login_cv);
739 		return;
740 	}
741 #endif
742 
743 	/*
744 	 * The handling routine is responsible for freeing the PDU
745 	 * when it's no longer needed.
746 	 */
747 	switch (response->ip_bhs->bhs_opcode) {
748 	case ISCSI_BHS_OPCODE_NOP_IN:
749 		iscsi_pdu_handle_nop_in(response);
750 		ISCSI_SESSION_UNLOCK(is);
751 		break;
752 	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
753 		iscsi_pdu_handle_scsi_response(response);
754 		/* Session lock dropped inside. */
755 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
756 		break;
757 	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
758 		iscsi_pdu_handle_task_response(response);
759 		ISCSI_SESSION_UNLOCK(is);
760 		break;
761 	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
762 		iscsi_pdu_handle_data_in(response);
763 		/* Session lock dropped inside. */
764 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
765 		break;
766 	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
767 		iscsi_pdu_handle_logout_response(response);
768 		ISCSI_SESSION_UNLOCK(is);
769 		break;
770 	case ISCSI_BHS_OPCODE_R2T:
771 		iscsi_pdu_handle_r2t(response);
772 		ISCSI_SESSION_UNLOCK(is);
773 		break;
774 	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
775 		iscsi_pdu_handle_async_message(response);
776 		ISCSI_SESSION_UNLOCK(is);
777 		break;
778 	case ISCSI_BHS_OPCODE_REJECT:
779 		iscsi_pdu_handle_reject(response);
780 		ISCSI_SESSION_UNLOCK(is);
781 		break;
782 	default:
783 		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
784 		    "opcode 0x%x; reconnecting",
785 		    response->ip_bhs->bhs_opcode);
786 		iscsi_session_reconnect(is);
787 		ISCSI_SESSION_UNLOCK(is);
788 		icl_pdu_free(response);
789 	}
790 }
791 
792 static void
793 iscsi_error_callback(struct icl_conn *ic)
794 {
795 	struct iscsi_session *is;
796 
797 	is = CONN_SESSION(ic);
798 
799 	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
800 	iscsi_session_reconnect(is);
801 }
802 
803 static void
804 iscsi_pdu_handle_nop_in(struct icl_pdu *response)
805 {
806 	struct iscsi_session *is;
807 	struct iscsi_bhs_nop_out *bhsno;
808 	struct iscsi_bhs_nop_in *bhsni;
809 	struct icl_pdu *request;
810 	void *data = NULL;
811 	size_t datasize;
812 	int error;
813 
814 	is = PDU_SESSION(response);
815 	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
816 
817 	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
818 		/*
819 		 * Nothing to do; iscsi_pdu_update_statsn() already
820 		 * zeroed the timeout.
821 		 */
822 		icl_pdu_free(response);
823 		return;
824 	}
825 
826 	datasize = icl_pdu_data_segment_length(response);
827 	if (datasize > 0) {
828 		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
829 		if (data == NULL) {
830 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
831 			    "reconnecting");
832 			icl_pdu_free(response);
833 			iscsi_session_reconnect(is);
834 			return;
835 		}
836 		icl_pdu_get_data(response, 0, data, datasize);
837 	}
838 
839 	request = icl_pdu_new(response->ip_conn, M_NOWAIT);
840 	if (request == NULL) {
841 		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
842 		    "reconnecting");
843 		free(data, M_ISCSI);
844 		icl_pdu_free(response);
845 		iscsi_session_reconnect(is);
846 		return;
847 	}
848 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
849 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
850 	    ISCSI_BHS_OPCODE_IMMEDIATE;
851 	bhsno->bhsno_flags = 0x80;
852 	bhsno->bhsno_initiator_task_tag = 0xffffffff;
853 	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
854 	if (datasize > 0) {
855 		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
856 		if (error != 0) {
857 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
858 			    "reconnecting");
859 			free(data, M_ISCSI);
860 			icl_pdu_free(request);
861 			icl_pdu_free(response);
862 			iscsi_session_reconnect(is);
863 			return;
864 		}
865 		free(data, M_ISCSI);
866 	}
867 
868 	icl_pdu_free(response);
869 	iscsi_pdu_queue_locked(request);
870 }
871 
872 static void
873 iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
874 {
875 	struct iscsi_bhs_scsi_response *bhssr;
876 	struct iscsi_outstanding *io;
877 	struct iscsi_session *is;
878 	union ccb *ccb;
879 	struct ccb_scsiio *csio;
880 	size_t data_segment_len, received;
881 	uint16_t sense_len;
882 	uint32_t resid;
883 
884 	is = PDU_SESSION(response);
885 
886 	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
887 	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
888 	if (io == NULL || io->io_ccb == NULL) {
889 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
890 		icl_pdu_free(response);
891 		iscsi_session_reconnect(is);
892 		ISCSI_SESSION_UNLOCK(is);
893 		return;
894 	}
895 
896 	ccb = io->io_ccb;
897 
898 	/*
899 	 * With iSER, after getting good response we can be sure
900 	 * that all the data has been successfully transferred.
901 	 */
902 	if (is->is_conn->ic_iser) {
903 		resid = ntohl(bhssr->bhssr_residual_count);
904 		if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW) {
905 			io->io_received = ccb->csio.dxfer_len - resid;
906 		} else if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_OVERFLOW) {
907 			ISCSI_SESSION_WARN(is, "overflow: target indicates %d", resid);
908 		} else {
909 			io->io_received = ccb->csio.dxfer_len;
910 		}
911 	}
912 
913 	received = io->io_received;
914 	iscsi_outstanding_remove(is, io);
915 	ISCSI_SESSION_UNLOCK(is);
916 
917 	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
918 		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
919  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
920  			xpt_freeze_devq(ccb->ccb_h.path, 1);
921 			ISCSI_SESSION_DEBUG(is, "freezing devq");
922 		}
923  		ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
924 	} else if (bhssr->bhssr_status == 0) {
925 		ccb->ccb_h.status = CAM_REQ_CMP;
926 	} else {
927  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
928  			xpt_freeze_devq(ccb->ccb_h.path, 1);
929 			ISCSI_SESSION_DEBUG(is, "freezing devq");
930 		}
931  		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
932 		ccb->csio.scsi_status = bhssr->bhssr_status;
933 	}
934 
935 	csio = &ccb->csio;
936 	data_segment_len = icl_pdu_data_segment_length(response);
937 	if (data_segment_len > 0) {
938 		if (data_segment_len < sizeof(sense_len)) {
939 			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
940 			    data_segment_len);
941 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
942 				xpt_freeze_devq(ccb->ccb_h.path, 1);
943 				ISCSI_SESSION_DEBUG(is, "freezing devq");
944 			}
945 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
946 			goto out;
947 		}
948 		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
949 		sense_len = ntohs(sense_len);
950 #if 0
951 		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
952 		    sense_len, data_segment_len);
953 #endif
954 		if (sizeof(sense_len) + sense_len > data_segment_len) {
955 			ISCSI_SESSION_WARN(is, "truncated data segment "
956 			    "(%zd bytes, should be %zd)",
957 			    data_segment_len, sizeof(sense_len) + sense_len);
958 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
959 				xpt_freeze_devq(ccb->ccb_h.path, 1);
960 				ISCSI_SESSION_DEBUG(is, "freezing devq");
961 			}
962 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
963 			goto out;
964 		} else if (sizeof(sense_len) + sense_len < data_segment_len)
965 			ISCSI_SESSION_WARN(is, "oversize data segment "
966 			    "(%zd bytes, should be %zd)",
967 			    data_segment_len, sizeof(sense_len) + sense_len);
968 		if (sense_len > csio->sense_len) {
969 			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
970 			    sense_len, csio->sense_len);
971 			sense_len = csio->sense_len;
972 		}
973 		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
974 		csio->sense_resid = csio->sense_len - sense_len;
975 		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
976 	}
977 
978 out:
979 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
980 		csio->resid = ntohl(bhssr->bhssr_residual_count);
981 
982 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
983 		KASSERT(received <= csio->dxfer_len,
984 		    ("received > csio->dxfer_len"));
985 		if (received < csio->dxfer_len) {
986 			if (csio->resid != csio->dxfer_len - received) {
987 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
988 				    "target indicates %d, we calculated %zd",
989 				    csio->resid, csio->dxfer_len - received);
990 			}
991 			csio->resid = csio->dxfer_len - received;
992 		}
993 	}
994 
995 	xpt_done(ccb);
996 	icl_pdu_free(response);
997 }
998 
999 static void
1000 iscsi_pdu_handle_task_response(struct icl_pdu *response)
1001 {
1002 	struct iscsi_bhs_task_management_response *bhstmr;
1003 	struct iscsi_outstanding *io, *aio;
1004 	struct iscsi_session *is;
1005 
1006 	is = PDU_SESSION(response);
1007 
1008 	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
1009 	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
1010 	if (io == NULL || io->io_ccb != NULL) {
1011 		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
1012 		    bhstmr->bhstmr_initiator_task_tag);
1013 		icl_pdu_free(response);
1014 		iscsi_session_reconnect(is);
1015 		return;
1016 	}
1017 
1018 	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
1019 		ISCSI_SESSION_WARN(is, "task response 0x%x",
1020 		    bhstmr->bhstmr_response);
1021 	} else {
1022 		aio = iscsi_outstanding_find(is, io->io_datasn);
1023 		if (aio != NULL && aio->io_ccb != NULL)
1024 			iscsi_session_terminate_task(is, aio, CAM_REQ_ABORTED);
1025 	}
1026 
1027 	iscsi_outstanding_remove(is, io);
1028 	icl_pdu_free(response);
1029 }
1030 
1031 static void
1032 iscsi_pdu_handle_data_in(struct icl_pdu *response)
1033 {
1034 	struct iscsi_bhs_data_in *bhsdi;
1035 	struct iscsi_outstanding *io;
1036 	struct iscsi_session *is;
1037 	union ccb *ccb;
1038 	struct ccb_scsiio *csio;
1039 	size_t data_segment_len, received, oreceived;
1040 
1041 	is = PDU_SESSION(response);
1042 	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
1043 	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
1044 	if (io == NULL || io->io_ccb == NULL) {
1045 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
1046 		icl_pdu_free(response);
1047 		iscsi_session_reconnect(is);
1048 		ISCSI_SESSION_UNLOCK(is);
1049 		return;
1050 	}
1051 
1052 	data_segment_len = icl_pdu_data_segment_length(response);
1053 	if (data_segment_len == 0) {
1054 		/*
1055 		 * "The sending of 0 length data segments should be avoided,
1056 		 * but initiators and targets MUST be able to properly receive
1057 		 * 0 length data segments."
1058 		 */
1059 		ISCSI_SESSION_UNLOCK(is);
1060 		icl_pdu_free(response);
1061 		return;
1062 	}
1063 
1064 	/*
1065 	 * We need to track this for security reasons - without it, malicious target
1066 	 * could respond to SCSI READ without sending Data-In PDUs, which would result
1067 	 * in read operation on the initiator side returning random kernel data.
1068 	 */
1069 	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1070 		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1071 		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1072 		icl_pdu_free(response);
1073 		iscsi_session_reconnect(is);
1074 		ISCSI_SESSION_UNLOCK(is);
1075 		return;
1076 	}
1077 
1078 	ccb = io->io_ccb;
1079 	csio = &ccb->csio;
1080 
1081 	if (io->io_received + data_segment_len > csio->dxfer_len) {
1082 		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1083 		    "at offset %zd, buffer is %d)",
1084 		    data_segment_len, io->io_received, csio->dxfer_len);
1085 		icl_pdu_free(response);
1086 		iscsi_session_reconnect(is);
1087 		ISCSI_SESSION_UNLOCK(is);
1088 		return;
1089 	}
1090 
1091 	oreceived = io->io_received;
1092 	io->io_received += data_segment_len;
1093 	received = io->io_received;
1094 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0)
1095 		iscsi_outstanding_remove(is, io);
1096 	ISCSI_SESSION_UNLOCK(is);
1097 
1098 	icl_pdu_get_data(response, 0, csio->data_ptr + oreceived, data_segment_len);
1099 
1100 	/*
1101 	 * XXX: Check DataSN.
1102 	 * XXX: Check F.
1103 	 */
1104 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1105 		/*
1106 		 * Nothing more to do.
1107 		 */
1108 		icl_pdu_free(response);
1109 		return;
1110 	}
1111 
1112 	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1113 	if (bhsdi->bhsdi_status == 0) {
1114 		ccb->ccb_h.status = CAM_REQ_CMP;
1115 	} else {
1116 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1117 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1118 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1119 		}
1120 		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1121 		csio->scsi_status = bhsdi->bhsdi_status;
1122 	}
1123 
1124 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1125 		KASSERT(received <= csio->dxfer_len,
1126 		    ("received > csio->dxfer_len"));
1127 		if (received < csio->dxfer_len) {
1128 			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1129 			if (csio->resid != csio->dxfer_len - received) {
1130 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1131 				    "target indicates %d, we calculated %zd",
1132 				    csio->resid, csio->dxfer_len - received);
1133 			}
1134 			csio->resid = csio->dxfer_len - received;
1135 		}
1136 	}
1137 
1138 	xpt_done(ccb);
1139 	icl_pdu_free(response);
1140 }
1141 
1142 static void
1143 iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1144 {
1145 
1146 	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1147 	icl_pdu_free(response);
1148 }
1149 
1150 static void
1151 iscsi_pdu_handle_r2t(struct icl_pdu *response)
1152 {
1153 	struct icl_pdu *request;
1154 	struct iscsi_session *is;
1155 	struct iscsi_bhs_r2t *bhsr2t;
1156 	struct iscsi_bhs_data_out *bhsdo;
1157 	struct iscsi_outstanding *io;
1158 	struct ccb_scsiio *csio;
1159 	size_t off, len, total_len;
1160 	int error;
1161 
1162 	is = PDU_SESSION(response);
1163 
1164 	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1165 	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1166 	if (io == NULL || io->io_ccb == NULL) {
1167 		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1168 		    bhsr2t->bhsr2t_initiator_task_tag);
1169 		icl_pdu_free(response);
1170 		iscsi_session_reconnect(is);
1171 		return;
1172 	}
1173 
1174 	csio = &io->io_ccb->csio;
1175 
1176 	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1177 		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1178 		icl_pdu_free(response);
1179 		iscsi_session_reconnect(is);
1180 		return;
1181 	}
1182 
1183 	/*
1184 	 * XXX: Verify R2TSN.
1185 	 */
1186 
1187 	io->io_datasn = 0;
1188 
1189 	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1190 	if (off > csio->dxfer_len) {
1191 		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1192 		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1193 		icl_pdu_free(response);
1194 		iscsi_session_reconnect(is);
1195 		return;
1196 	}
1197 
1198 	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1199 	if (total_len == 0 || total_len > csio->dxfer_len) {
1200 		ISCSI_SESSION_WARN(is, "target requested invalid length "
1201 		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1202 		icl_pdu_free(response);
1203 		iscsi_session_reconnect(is);
1204 		return;
1205 	}
1206 
1207 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1208 
1209 	for (;;) {
1210 		len = total_len;
1211 
1212 		if (len > is->is_max_send_data_segment_length)
1213 			len = is->is_max_send_data_segment_length;
1214 
1215 		if (off + len > csio->dxfer_len) {
1216 			ISCSI_SESSION_WARN(is, "target requested invalid "
1217 			    "length/offset %zd, buffer is %d; reconnecting",
1218 			    off + len, csio->dxfer_len);
1219 			icl_pdu_free(response);
1220 			iscsi_session_reconnect(is);
1221 			return;
1222 		}
1223 
1224 		request = icl_pdu_new(response->ip_conn, M_NOWAIT);
1225 		if (request == NULL) {
1226 			icl_pdu_free(response);
1227 			iscsi_session_reconnect(is);
1228 			return;
1229 		}
1230 
1231 		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1232 		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1233 		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1234 		bhsdo->bhsdo_initiator_task_tag =
1235 		    bhsr2t->bhsr2t_initiator_task_tag;
1236 		bhsdo->bhsdo_target_transfer_tag =
1237 		    bhsr2t->bhsr2t_target_transfer_tag;
1238 		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1239 		bhsdo->bhsdo_buffer_offset = htonl(off);
1240 		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1241 		    M_NOWAIT);
1242 		if (error != 0) {
1243 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1244 			    "reconnecting");
1245 			icl_pdu_free(request);
1246 			icl_pdu_free(response);
1247 			iscsi_session_reconnect(is);
1248 			return;
1249 		}
1250 
1251 		off += len;
1252 		total_len -= len;
1253 
1254 		if (total_len == 0) {
1255 			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1256 			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1257 		} else {
1258 			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1259 		}
1260 
1261 		iscsi_pdu_queue_locked(request);
1262 
1263 		if (total_len == 0)
1264 			break;
1265 	}
1266 
1267 	icl_pdu_free(response);
1268 }
1269 
1270 static void
1271 iscsi_pdu_handle_async_message(struct icl_pdu *response)
1272 {
1273 	struct iscsi_bhs_asynchronous_message *bhsam;
1274 	struct iscsi_session *is;
1275 
1276 	is = PDU_SESSION(response);
1277 	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1278 	switch (bhsam->bhsam_async_event) {
1279 	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1280 		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1281 		iscsi_session_logout(is);
1282 		iscsi_session_terminate(is);
1283 		break;
1284 	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1285 		ISCSI_SESSION_WARN(is, "target indicates it will drop the connection");
1286 		break;
1287 	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1288 		ISCSI_SESSION_WARN(is, "target indicates it will drop the session");
1289 		break;
1290 	default:
1291 		/*
1292 		 * XXX: Technically, we're obligated to also handle
1293 		 * 	parameter renegotiation.
1294 		 */
1295 		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1296 		break;
1297 	}
1298 
1299 	icl_pdu_free(response);
1300 }
1301 
1302 static void
1303 iscsi_pdu_handle_reject(struct icl_pdu *response)
1304 {
1305 	struct iscsi_bhs_reject *bhsr;
1306 	struct iscsi_session *is;
1307 
1308 	is = PDU_SESSION(response);
1309 	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1310 	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1311 	    bhsr->bhsr_reason);
1312 
1313 	icl_pdu_free(response);
1314 }
1315 
1316 static int
1317 iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1318     struct iscsi_daemon_request *request)
1319 {
1320 	struct iscsi_session *is;
1321 	struct icl_drv_limits idl;
1322 	int error;
1323 
1324 	sx_slock(&sc->sc_lock);
1325 	for (;;) {
1326 		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1327 			ISCSI_SESSION_LOCK(is);
1328 			if (is->is_conf.isc_enable == 0 &&
1329 			    is->is_conf.isc_discovery == 0) {
1330 				ISCSI_SESSION_UNLOCK(is);
1331 				continue;
1332 			}
1333 			if (is->is_waiting_for_iscsid)
1334 				break;
1335 			ISCSI_SESSION_UNLOCK(is);
1336 		}
1337 
1338 		if (is == NULL) {
1339 			/*
1340 			 * No session requires attention from iscsid(8); wait.
1341 			 */
1342 			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1343 			if (error != 0) {
1344 				sx_sunlock(&sc->sc_lock);
1345 				return (error);
1346 			}
1347 			continue;
1348 		}
1349 
1350 		is->is_waiting_for_iscsid = false;
1351 		is->is_login_phase = true;
1352 		is->is_reason[0] = '\0';
1353 		ISCSI_SESSION_UNLOCK(is);
1354 
1355 		request->idr_session_id = is->is_id;
1356 		memcpy(&request->idr_isid, &is->is_isid,
1357 		    sizeof(request->idr_isid));
1358 		request->idr_tsih = 0;	/* New or reinstated session. */
1359 		memcpy(&request->idr_conf, &is->is_conf,
1360 		    sizeof(request->idr_conf));
1361 
1362 		error = icl_limits(is->is_conf.isc_offload,
1363 		    is->is_conf.isc_iser, &idl);
1364 		if (error != 0) {
1365 			ISCSI_SESSION_WARN(is, "icl_limits for offload \"%s\" "
1366 			    "failed with error %d", is->is_conf.isc_offload,
1367 			    error);
1368 			sx_sunlock(&sc->sc_lock);
1369 			return (error);
1370 		}
1371 		request->idr_limits.isl_max_recv_data_segment_length =
1372 		    idl.idl_max_recv_data_segment_length;
1373 		request->idr_limits.isl_max_send_data_segment_length =
1374 		    idl.idl_max_send_data_segment_length;
1375 		request->idr_limits.isl_max_burst_length =
1376 		    idl.idl_max_burst_length;
1377 		request->idr_limits.isl_first_burst_length =
1378 		    idl.idl_first_burst_length;
1379 
1380 		sx_sunlock(&sc->sc_lock);
1381 		return (0);
1382 	}
1383 }
1384 
1385 static int
1386 iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1387     struct iscsi_daemon_handoff *handoff)
1388 {
1389 	struct iscsi_session *is;
1390 	struct icl_conn *ic;
1391 	int error;
1392 
1393 	sx_slock(&sc->sc_lock);
1394 
1395 	/*
1396 	 * Find the session to hand off socket to.
1397 	 */
1398 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1399 		if (is->is_id == handoff->idh_session_id)
1400 			break;
1401 	}
1402 	if (is == NULL) {
1403 		sx_sunlock(&sc->sc_lock);
1404 		return (ESRCH);
1405 	}
1406 	ISCSI_SESSION_LOCK(is);
1407 	ic = is->is_conn;
1408 	if (is->is_conf.isc_discovery || is->is_terminating) {
1409 		ISCSI_SESSION_UNLOCK(is);
1410 		sx_sunlock(&sc->sc_lock);
1411 		return (EINVAL);
1412 	}
1413 	if (is->is_connected) {
1414 		/*
1415 		 * This might have happened because another iscsid(8)
1416 		 * instance handed off the connection in the meantime.
1417 		 * Just return.
1418 		 */
1419 		ISCSI_SESSION_WARN(is, "handoff on already connected "
1420 		    "session");
1421 		ISCSI_SESSION_UNLOCK(is);
1422 		sx_sunlock(&sc->sc_lock);
1423 		return (EBUSY);
1424 	}
1425 
1426 	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1427 	    sizeof(is->is_target_alias));
1428 	is->is_tsih = handoff->idh_tsih;
1429 	is->is_statsn = handoff->idh_statsn;
1430 	is->is_protocol_level = handoff->idh_protocol_level;
1431 	is->is_initial_r2t = handoff->idh_initial_r2t;
1432 	is->is_immediate_data = handoff->idh_immediate_data;
1433 
1434 	is->is_max_recv_data_segment_length =
1435 	    handoff->idh_max_recv_data_segment_length;
1436 	is->is_max_send_data_segment_length =
1437 	    handoff->idh_max_send_data_segment_length;
1438 	is->is_max_burst_length = handoff->idh_max_burst_length;
1439 	is->is_first_burst_length = handoff->idh_first_burst_length;
1440 
1441 	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1442 		ic->ic_header_crc32c = true;
1443 	else
1444 		ic->ic_header_crc32c = false;
1445 	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1446 		ic->ic_data_crc32c = true;
1447 	else
1448 		ic->ic_data_crc32c = false;
1449 	ic->ic_maxtags = maxtags;
1450 
1451 	is->is_cmdsn = 0;
1452 	is->is_expcmdsn = 0;
1453 	is->is_maxcmdsn = 0;
1454 	is->is_waiting_for_iscsid = false;
1455 	is->is_login_phase = false;
1456 	is->is_timeout = 0;
1457 	is->is_connected = true;
1458 	is->is_reason[0] = '\0';
1459 
1460 	ISCSI_SESSION_UNLOCK(is);
1461 
1462 	/*
1463 	 * If we're going through the proxy, the idh_socket will be 0,
1464 	 * and the ICL module can simply ignore this call.  It can also
1465 	 * use it to determine it's no longer in the Login phase.
1466 	 */
1467 	error = icl_conn_handoff(ic, handoff->idh_socket);
1468 	if (error != 0) {
1469 		sx_sunlock(&sc->sc_lock);
1470 		iscsi_session_terminate(is);
1471 		return (error);
1472 	}
1473 
1474 	sx_sunlock(&sc->sc_lock);
1475 
1476 	if (is->is_sim != NULL) {
1477 		/*
1478 		 * When reconnecting, there already is SIM allocated for the session.
1479 		 */
1480 		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1481 		ISCSI_SESSION_LOCK(is);
1482 		ISCSI_SESSION_DEBUG(is, "releasing");
1483 		is->is_simq_frozen = false;
1484 		xpt_release_simq(is->is_sim, 1);
1485 		ISCSI_SESSION_UNLOCK(is);
1486 
1487 	} else {
1488 		ISCSI_SESSION_LOCK(is);
1489 		is->is_devq = cam_simq_alloc(ic->ic_maxtags);
1490 		if (is->is_devq == NULL) {
1491 			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1492 			iscsi_session_terminate(is);
1493 			return (ENOMEM);
1494 		}
1495 
1496 		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1497 		    is, is->is_id /* unit */, &is->is_lock,
1498 		    1, ic->ic_maxtags, is->is_devq);
1499 		if (is->is_sim == NULL) {
1500 			ISCSI_SESSION_UNLOCK(is);
1501 			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1502 			cam_simq_free(is->is_devq);
1503 			iscsi_session_terminate(is);
1504 			return (ENOMEM);
1505 		}
1506 
1507 		error = xpt_bus_register(is->is_sim, NULL, 0);
1508 		if (error != 0) {
1509 			ISCSI_SESSION_UNLOCK(is);
1510 			ISCSI_SESSION_WARN(is, "failed to register bus");
1511 			iscsi_session_terminate(is);
1512 			return (ENOMEM);
1513 		}
1514 
1515 		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1516 		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1517 		    CAM_LUN_WILDCARD);
1518 		if (error != CAM_REQ_CMP) {
1519 			ISCSI_SESSION_UNLOCK(is);
1520 			ISCSI_SESSION_WARN(is, "failed to create path");
1521 			iscsi_session_terminate(is);
1522 			return (ENOMEM);
1523 		}
1524 		ISCSI_SESSION_UNLOCK(is);
1525 	}
1526 
1527 	return (0);
1528 }
1529 
1530 static int
1531 iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1532     struct iscsi_daemon_fail *fail)
1533 {
1534 	struct iscsi_session *is;
1535 
1536 	sx_slock(&sc->sc_lock);
1537 
1538 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1539 		if (is->is_id == fail->idf_session_id)
1540 			break;
1541 	}
1542 	if (is == NULL) {
1543 		sx_sunlock(&sc->sc_lock);
1544 		return (ESRCH);
1545 	}
1546 	ISCSI_SESSION_LOCK(is);
1547 	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1548 	    fail->idf_reason);
1549 	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1550 	//is->is_waiting_for_iscsid = false;
1551 	//is->is_login_phase = true;
1552 	//iscsi_session_reconnect(is);
1553 	ISCSI_SESSION_UNLOCK(is);
1554 	sx_sunlock(&sc->sc_lock);
1555 
1556 	return (0);
1557 }
1558 
1559 #ifdef ICL_KERNEL_PROXY
1560 static int
1561 iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1562     struct iscsi_daemon_connect *idc)
1563 {
1564 	struct iscsi_session *is;
1565 	struct sockaddr *from_sa, *to_sa;
1566 	int error;
1567 
1568 	sx_slock(&sc->sc_lock);
1569 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1570 		if (is->is_id == idc->idc_session_id)
1571 			break;
1572 	}
1573 	if (is == NULL) {
1574 		sx_sunlock(&sc->sc_lock);
1575 		return (ESRCH);
1576 	}
1577 	sx_sunlock(&sc->sc_lock);
1578 
1579 	if (idc->idc_from_addrlen > 0) {
1580 		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1581 		if (error != 0) {
1582 			ISCSI_SESSION_WARN(is,
1583 			    "getsockaddr failed with error %d", error);
1584 			return (error);
1585 		}
1586 	} else {
1587 		from_sa = NULL;
1588 	}
1589 	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1590 	if (error != 0) {
1591 		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1592 		    error);
1593 		free(from_sa, M_SONAME);
1594 		return (error);
1595 	}
1596 
1597 	ISCSI_SESSION_LOCK(is);
1598 	is->is_statsn = 0;
1599 	is->is_cmdsn = 0;
1600 	is->is_expcmdsn = 0;
1601 	is->is_maxcmdsn = 0;
1602 	is->is_waiting_for_iscsid = false;
1603 	is->is_login_phase = true;
1604 	is->is_timeout = 0;
1605 	ISCSI_SESSION_UNLOCK(is);
1606 
1607 	error = icl_conn_connect(is->is_conn, idc->idc_domain,
1608 	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1609 	free(from_sa, M_SONAME);
1610 	free(to_sa, M_SONAME);
1611 
1612 	/*
1613 	 * Digests are always disabled during login phase.
1614 	 */
1615 	is->is_conn->ic_header_crc32c = false;
1616 	is->is_conn->ic_data_crc32c = false;
1617 
1618 	return (error);
1619 }
1620 
1621 static int
1622 iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1623     struct iscsi_daemon_send *ids)
1624 {
1625 	struct iscsi_session *is;
1626 	struct icl_pdu *ip;
1627 	size_t datalen;
1628 	void *data;
1629 	int error;
1630 
1631 	sx_slock(&sc->sc_lock);
1632 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1633 		if (is->is_id == ids->ids_session_id)
1634 			break;
1635 	}
1636 	if (is == NULL) {
1637 		sx_sunlock(&sc->sc_lock);
1638 		return (ESRCH);
1639 	}
1640 	sx_sunlock(&sc->sc_lock);
1641 
1642 	if (is->is_login_phase == false)
1643 		return (EBUSY);
1644 
1645 	if (is->is_terminating || is->is_reconnecting)
1646 		return (EIO);
1647 
1648 	datalen = ids->ids_data_segment_len;
1649 	if (datalen > is->is_max_send_data_segment_length)
1650 		return (EINVAL);
1651 	if (datalen > 0) {
1652 		data = malloc(datalen, M_ISCSI, M_WAITOK);
1653 		error = copyin(ids->ids_data_segment, data, datalen);
1654 		if (error != 0) {
1655 			free(data, M_ISCSI);
1656 			return (error);
1657 		}
1658 	}
1659 
1660 	ip = icl_pdu_new(is->is_conn, M_WAITOK);
1661 	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1662 	if (datalen > 0) {
1663 		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1664 		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1665 		free(data, M_ISCSI);
1666 	}
1667 	iscsi_pdu_queue(ip);
1668 
1669 	return (0);
1670 }
1671 
1672 static int
1673 iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1674     struct iscsi_daemon_receive *idr)
1675 {
1676 	struct iscsi_session *is;
1677 	struct icl_pdu *ip;
1678 	void *data;
1679 	int error;
1680 
1681 	sx_slock(&sc->sc_lock);
1682 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1683 		if (is->is_id == idr->idr_session_id)
1684 			break;
1685 	}
1686 	if (is == NULL) {
1687 		sx_sunlock(&sc->sc_lock);
1688 		return (ESRCH);
1689 	}
1690 	sx_sunlock(&sc->sc_lock);
1691 
1692 	if (is->is_login_phase == false)
1693 		return (EBUSY);
1694 
1695 	ISCSI_SESSION_LOCK(is);
1696 	while (is->is_login_pdu == NULL &&
1697 	    is->is_terminating == false &&
1698 	    is->is_reconnecting == false) {
1699 		error = cv_wait_sig(&is->is_login_cv, &is->is_lock);
1700 		if (error != 0) {
1701 			ISCSI_SESSION_UNLOCK(is);
1702 			return (error);
1703 		}
1704 	}
1705 	if (is->is_terminating || is->is_reconnecting) {
1706 		ISCSI_SESSION_UNLOCK(is);
1707 		return (EIO);
1708 	}
1709 	ip = is->is_login_pdu;
1710 	is->is_login_pdu = NULL;
1711 	ISCSI_SESSION_UNLOCK(is);
1712 
1713 	if (ip->ip_data_len > idr->idr_data_segment_len) {
1714 		icl_pdu_free(ip);
1715 		return (EMSGSIZE);
1716 	}
1717 
1718 	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1719 	if (ip->ip_data_len > 0) {
1720 		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1721 		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1722 		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1723 		free(data, M_ISCSI);
1724 	}
1725 
1726 	icl_pdu_free(ip);
1727 
1728 	return (0);
1729 }
1730 #endif /* ICL_KERNEL_PROXY */
1731 
1732 static void
1733 iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1734 {
1735 	/*
1736 	 * Just make sure all the fields are null-terminated.
1737 	 *
1738 	 * XXX: This is not particularly secure.  We should
1739 	 * 	create our own conf and then copy in relevant
1740 	 * 	fields.
1741 	 */
1742 	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1743 	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1744 	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1745 	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1746 	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1747 	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1748 	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1749 	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1750 	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1751 }
1752 
1753 static bool
1754 iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1755 {
1756 
1757 	if (isc->isc_initiator[0] == '\0') {
1758 		ISCSI_DEBUG("empty isc_initiator");
1759 		return (false);
1760 	}
1761 
1762 	if (isc->isc_target_addr[0] == '\0') {
1763 		ISCSI_DEBUG("empty isc_target_addr");
1764 		return (false);
1765 	}
1766 
1767 	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1768 		ISCSI_DEBUG("non-empty isc_target for discovery session");
1769 		return (false);
1770 	}
1771 
1772 	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1773 		ISCSI_DEBUG("empty isc_target for non-discovery session");
1774 		return (false);
1775 	}
1776 
1777 	return (true);
1778 }
1779 
1780 static int
1781 iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1782 {
1783 	struct iscsi_session *is;
1784 	const struct iscsi_session *is2;
1785 	int error;
1786 
1787 	iscsi_sanitize_session_conf(&isa->isa_conf);
1788 	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1789 		return (EINVAL);
1790 
1791 	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1792 	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1793 
1794 	/*
1795 	 * Set some default values, from RFC 3720, section 12.
1796 	 *
1797 	 * These values are updated by the handoff IOCTL, but are
1798 	 * needed prior to the handoff to support sending the ISER
1799 	 * login PDU.
1800 	 */
1801 	is->is_max_recv_data_segment_length = 8192;
1802 	is->is_max_send_data_segment_length = 8192;
1803 	is->is_max_burst_length = 262144;
1804 	is->is_first_burst_length = 65536;
1805 
1806 	sx_xlock(&sc->sc_lock);
1807 
1808 	/*
1809 	 * Prevent duplicates.
1810 	 */
1811 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1812 		if (!!is->is_conf.isc_discovery !=
1813 		    !!is2->is_conf.isc_discovery)
1814 			continue;
1815 
1816 		if (strcmp(is->is_conf.isc_target_addr,
1817 		    is2->is_conf.isc_target_addr) != 0)
1818 			continue;
1819 
1820 		if (is->is_conf.isc_discovery == 0 &&
1821 		    strcmp(is->is_conf.isc_target,
1822 		    is2->is_conf.isc_target) != 0)
1823 			continue;
1824 
1825 		sx_xunlock(&sc->sc_lock);
1826 		free(is, M_ISCSI);
1827 		return (EBUSY);
1828 	}
1829 
1830 	is->is_conn = icl_new_conn(is->is_conf.isc_offload,
1831 	    is->is_conf.isc_iser, "iscsi", &is->is_lock);
1832 	if (is->is_conn == NULL) {
1833 		sx_xunlock(&sc->sc_lock);
1834 		free(is, M_ISCSI);
1835 		return (EINVAL);
1836 	}
1837 	is->is_conn->ic_receive = iscsi_receive_callback;
1838 	is->is_conn->ic_error = iscsi_error_callback;
1839 	is->is_conn->ic_prv0 = is;
1840 	TAILQ_INIT(&is->is_outstanding);
1841 	STAILQ_INIT(&is->is_postponed);
1842 	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1843 	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1844 #ifdef ICL_KERNEL_PROXY
1845 	cv_init(&is->is_login_cv, "iscsi_login");
1846 #endif
1847 
1848 	is->is_softc = sc;
1849 	sc->sc_last_session_id++;
1850 	is->is_id = sc->sc_last_session_id;
1851 	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1852 	arc4rand(&is->is_isid[1], 5, 0);
1853 	is->is_tsih = 0;
1854 	callout_init(&is->is_callout, 1);
1855 
1856 	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1857 	if (error != 0) {
1858 		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1859 		sx_xunlock(&sc->sc_lock);
1860 		return (error);
1861 	}
1862 
1863 	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1864 	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1865 
1866 	ISCSI_SESSION_LOCK(is);
1867 	/*
1868 	 * Don't notify iscsid(8) if the session is disabled and it's not
1869 	 * a discovery session,
1870 	 */
1871 	if (is->is_conf.isc_enable == 0 && is->is_conf.isc_discovery == 0) {
1872 		ISCSI_SESSION_UNLOCK(is);
1873 		sx_xunlock(&sc->sc_lock);
1874 		return (0);
1875 	}
1876 
1877 	is->is_waiting_for_iscsid = true;
1878 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1879 	ISCSI_SESSION_UNLOCK(is);
1880 	cv_signal(&sc->sc_cv);
1881 	sx_xunlock(&sc->sc_lock);
1882 	return (0);
1883 }
1884 
1885 static bool
1886 iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1887     unsigned int id2, const struct iscsi_session_conf *c2)
1888 {
1889 
1890 	if (id2 != 0 && id2 != id1)
1891 		return (false);
1892 	if (c2->isc_target[0] != '\0' &&
1893 	    strcmp(c1->isc_target, c2->isc_target) != 0)
1894 		return (false);
1895 	if (c2->isc_target_addr[0] != '\0' &&
1896 	    strcmp(c1->isc_target_addr, c2->isc_target_addr) != 0)
1897 		return (false);
1898 	return (true);
1899 }
1900 
1901 static int
1902 iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1903     struct iscsi_session_remove *isr)
1904 {
1905 	struct iscsi_session *is, *tmp;
1906 	bool found = false;
1907 
1908 	iscsi_sanitize_session_conf(&isr->isr_conf);
1909 
1910 	sx_xlock(&sc->sc_lock);
1911 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1912 		ISCSI_SESSION_LOCK(is);
1913 		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1914 		    isr->isr_session_id, &isr->isr_conf)) {
1915 			found = true;
1916 			iscsi_session_logout(is);
1917 			iscsi_session_terminate(is);
1918 		}
1919 		ISCSI_SESSION_UNLOCK(is);
1920 	}
1921 	sx_xunlock(&sc->sc_lock);
1922 
1923 	if (!found)
1924 		return (ESRCH);
1925 
1926 	return (0);
1927 }
1928 
1929 static int
1930 iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1931 {
1932 	int error;
1933 	unsigned int i = 0;
1934 	struct iscsi_session *is;
1935 	struct iscsi_session_state iss;
1936 
1937 	sx_slock(&sc->sc_lock);
1938 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1939 		if (i >= isl->isl_nentries) {
1940 			sx_sunlock(&sc->sc_lock);
1941 			return (EMSGSIZE);
1942 		}
1943 		memset(&iss, 0, sizeof(iss));
1944 		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1945 		iss.iss_id = is->is_id;
1946 		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1947 		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1948 		strlcpy(iss.iss_offload, is->is_conn->ic_offload, sizeof(iss.iss_offload));
1949 
1950 		if (is->is_conn->ic_header_crc32c)
1951 			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1952 		else
1953 			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1954 
1955 		if (is->is_conn->ic_data_crc32c)
1956 			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1957 		else
1958 			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1959 
1960 		iss.iss_max_send_data_segment_length =
1961 		    is->is_max_send_data_segment_length;
1962 		iss.iss_max_recv_data_segment_length =
1963 		    is->is_max_recv_data_segment_length;
1964 		iss.iss_max_burst_length = is->is_max_burst_length;
1965 		iss.iss_first_burst_length = is->is_first_burst_length;
1966 		iss.iss_immediate_data = is->is_immediate_data;
1967 		iss.iss_connected = is->is_connected;
1968 
1969 		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1970 		if (error != 0) {
1971 			sx_sunlock(&sc->sc_lock);
1972 			return (error);
1973 		}
1974 		i++;
1975 	}
1976 	sx_sunlock(&sc->sc_lock);
1977 
1978 	isl->isl_nentries = i;
1979 
1980 	return (0);
1981 }
1982 
1983 static int
1984 iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1985     struct iscsi_session_modify *ism)
1986 {
1987 	struct iscsi_session *is;
1988 	const struct iscsi_session *is2;
1989 
1990 	iscsi_sanitize_session_conf(&ism->ism_conf);
1991 	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1992 		return (EINVAL);
1993 
1994 	sx_xlock(&sc->sc_lock);
1995 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1996 		ISCSI_SESSION_LOCK(is);
1997 		if (is->is_id == ism->ism_session_id) {
1998 			/* Note that the session remains locked. */
1999 			break;
2000 		}
2001 		ISCSI_SESSION_UNLOCK(is);
2002 	}
2003 	if (is == NULL) {
2004 		sx_xunlock(&sc->sc_lock);
2005 		return (ESRCH);
2006 	}
2007 
2008 	/*
2009 	 * Prevent duplicates.
2010 	 */
2011 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
2012 		if (is == is2)
2013 			continue;
2014 
2015 		if (!!ism->ism_conf.isc_discovery !=
2016 		    !!is2->is_conf.isc_discovery)
2017 			continue;
2018 
2019 		if (strcmp(ism->ism_conf.isc_target_addr,
2020 		    is2->is_conf.isc_target_addr) != 0)
2021 			continue;
2022 
2023 		if (ism->ism_conf.isc_discovery == 0 &&
2024 		    strcmp(ism->ism_conf.isc_target,
2025 		    is2->is_conf.isc_target) != 0)
2026 			continue;
2027 
2028 		ISCSI_SESSION_UNLOCK(is);
2029 		sx_xunlock(&sc->sc_lock);
2030 		return (EBUSY);
2031 	}
2032 
2033 	sx_xunlock(&sc->sc_lock);
2034 
2035 	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
2036 	ISCSI_SESSION_UNLOCK(is);
2037 
2038 	iscsi_session_reconnect(is);
2039 
2040 	return (0);
2041 }
2042 
2043 static int
2044 iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
2045     struct thread *td)
2046 {
2047 	struct iscsi_softc *sc;
2048 
2049 	sc = dev->si_drv1;
2050 
2051 	switch (cmd) {
2052 	case ISCSIDWAIT:
2053 		return (iscsi_ioctl_daemon_wait(sc,
2054 		    (struct iscsi_daemon_request *)arg));
2055 	case ISCSIDHANDOFF:
2056 		return (iscsi_ioctl_daemon_handoff(sc,
2057 		    (struct iscsi_daemon_handoff *)arg));
2058 	case ISCSIDFAIL:
2059 		return (iscsi_ioctl_daemon_fail(sc,
2060 		    (struct iscsi_daemon_fail *)arg));
2061 #ifdef ICL_KERNEL_PROXY
2062 	case ISCSIDCONNECT:
2063 		return (iscsi_ioctl_daemon_connect(sc,
2064 		    (struct iscsi_daemon_connect *)arg));
2065 	case ISCSIDSEND:
2066 		return (iscsi_ioctl_daemon_send(sc,
2067 		    (struct iscsi_daemon_send *)arg));
2068 	case ISCSIDRECEIVE:
2069 		return (iscsi_ioctl_daemon_receive(sc,
2070 		    (struct iscsi_daemon_receive *)arg));
2071 #endif /* ICL_KERNEL_PROXY */
2072 	case ISCSISADD:
2073 		return (iscsi_ioctl_session_add(sc,
2074 		    (struct iscsi_session_add *)arg));
2075 	case ISCSISREMOVE:
2076 		return (iscsi_ioctl_session_remove(sc,
2077 		    (struct iscsi_session_remove *)arg));
2078 	case ISCSISLIST:
2079 		return (iscsi_ioctl_session_list(sc,
2080 		    (struct iscsi_session_list *)arg));
2081 	case ISCSISMODIFY:
2082 		return (iscsi_ioctl_session_modify(sc,
2083 		    (struct iscsi_session_modify *)arg));
2084 	default:
2085 		return (EINVAL);
2086 	}
2087 }
2088 
2089 static struct iscsi_outstanding *
2090 iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
2091 {
2092 	struct iscsi_outstanding *io;
2093 
2094 	ISCSI_SESSION_LOCK_ASSERT(is);
2095 
2096 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2097 		if (io->io_initiator_task_tag == initiator_task_tag)
2098 			return (io);
2099 	}
2100 	return (NULL);
2101 }
2102 
2103 static struct iscsi_outstanding *
2104 iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
2105 {
2106 	struct iscsi_outstanding *io;
2107 
2108 	ISCSI_SESSION_LOCK_ASSERT(is);
2109 
2110 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2111 		if (io->io_ccb == ccb)
2112 			return (io);
2113 	}
2114 	return (NULL);
2115 }
2116 
2117 static struct iscsi_outstanding *
2118 iscsi_outstanding_add(struct iscsi_session *is, struct icl_pdu *request,
2119     union ccb *ccb, uint32_t *initiator_task_tagp)
2120 {
2121 	struct iscsi_outstanding *io;
2122 	int error;
2123 
2124 	ISCSI_SESSION_LOCK_ASSERT(is);
2125 
2126 	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
2127 	if (io == NULL) {
2128 		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes",
2129 		    sizeof(*io));
2130 		return (NULL);
2131 	}
2132 
2133 	error = icl_conn_task_setup(is->is_conn, request, &ccb->csio,
2134 	    initiator_task_tagp, &io->io_icl_prv);
2135 	if (error != 0) {
2136 		ISCSI_SESSION_WARN(is,
2137 		    "icl_conn_task_setup() failed with error %d", error);
2138 		uma_zfree(iscsi_outstanding_zone, io);
2139 		return (NULL);
2140 	}
2141 
2142 	KASSERT(iscsi_outstanding_find(is, *initiator_task_tagp) == NULL,
2143 	    ("initiator_task_tag 0x%x already added", *initiator_task_tagp));
2144 
2145 	io->io_initiator_task_tag = *initiator_task_tagp;
2146 	io->io_ccb = ccb;
2147 	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
2148 	return (io);
2149 }
2150 
2151 static void
2152 iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2153 {
2154 
2155 	ISCSI_SESSION_LOCK_ASSERT(is);
2156 
2157 	icl_conn_task_done(is->is_conn, io->io_icl_prv);
2158 	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2159 	uma_zfree(iscsi_outstanding_zone, io);
2160 }
2161 
2162 static void
2163 iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2164 {
2165 	struct icl_pdu *request;
2166 	struct iscsi_bhs_task_management_request *bhstmr;
2167 	struct ccb_abort *cab = &ccb->cab;
2168 	struct iscsi_outstanding *io, *aio;
2169 	uint32_t initiator_task_tag;
2170 
2171 	ISCSI_SESSION_LOCK_ASSERT(is);
2172 
2173 #if 0
2174 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2175 #else
2176 	if (is->is_login_phase) {
2177 		ccb->ccb_h.status = CAM_REQ_ABORTED;
2178 		xpt_done(ccb);
2179 		return;
2180 	}
2181 #endif
2182 
2183 	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2184 	if (aio == NULL) {
2185 		ccb->ccb_h.status = CAM_REQ_CMP;
2186 		xpt_done(ccb);
2187 		return;
2188 	}
2189 
2190 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2191 	if (request == NULL) {
2192 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2193 		xpt_done(ccb);
2194 		return;
2195 	}
2196 
2197 	initiator_task_tag = is->is_initiator_task_tag++;
2198 
2199 	io = iscsi_outstanding_add(is, request, NULL, &initiator_task_tag);
2200 	if (io == NULL) {
2201 		icl_pdu_free(request);
2202 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2203 		xpt_done(ccb);
2204 		return;
2205 	}
2206 	io->io_datasn = aio->io_initiator_task_tag;
2207 
2208 	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2209 	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2210 	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2211 	bhstmr->bhstmr_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2212 	bhstmr->bhstmr_initiator_task_tag = initiator_task_tag;
2213 	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2214 
2215 	iscsi_pdu_queue_locked(request);
2216 }
2217 
2218 static void
2219 iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2220 {
2221 	struct icl_pdu *request;
2222 	struct iscsi_bhs_scsi_command *bhssc;
2223 	struct ccb_scsiio *csio;
2224 	struct iscsi_outstanding *io;
2225 	size_t len;
2226 	uint32_t initiator_task_tag;
2227 	int error;
2228 
2229 	ISCSI_SESSION_LOCK_ASSERT(is);
2230 
2231 #if 0
2232 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2233 #else
2234 	if (is->is_login_phase) {
2235 		ISCSI_SESSION_DEBUG(is, "called during login phase");
2236 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2237 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2238 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2239 		}
2240 		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2241 		xpt_done(ccb);
2242 		return;
2243 	}
2244 #endif
2245 
2246 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2247 	if (request == NULL) {
2248 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2249 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2250 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2251 		}
2252 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2253 		xpt_done(ccb);
2254 		return;
2255 	}
2256 
2257 	initiator_task_tag = is->is_initiator_task_tag++;
2258 	io = iscsi_outstanding_add(is, request, ccb, &initiator_task_tag);
2259 	if (io == NULL) {
2260 		icl_pdu_free(request);
2261 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2262 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2263 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2264 		}
2265 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2266 		xpt_done(ccb);
2267 		return;
2268 	}
2269 
2270 	csio = &ccb->csio;
2271 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2272 	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2273 	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2274 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2275 	case CAM_DIR_IN:
2276 		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2277 		break;
2278 	case CAM_DIR_OUT:
2279 		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2280 		break;
2281 	}
2282 
2283 	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2284 		switch (csio->tag_action) {
2285 		case MSG_HEAD_OF_Q_TAG:
2286 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2287 			break;
2288 		case MSG_ORDERED_Q_TAG:
2289 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2290 			break;
2291 		case MSG_ACA_TASK:
2292 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2293 			break;
2294 		case MSG_SIMPLE_Q_TAG:
2295 		default:
2296 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2297 			break;
2298 		}
2299 	} else
2300 		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2301 
2302 	bhssc->bhssc_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2303 	bhssc->bhssc_initiator_task_tag = initiator_task_tag;
2304 	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2305 	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2306 	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2307 
2308 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2309 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2310 	else
2311 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2312 
2313 	if (is->is_immediate_data &&
2314 	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2315 		len = csio->dxfer_len;
2316 		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2317 		if (len > is->is_first_burst_length) {
2318 			ISCSI_SESSION_DEBUG(is, "len %zd -> %d", len, is->is_first_burst_length);
2319 			len = is->is_first_burst_length;
2320 		}
2321 		if (len > is->is_max_send_data_segment_length) {
2322 			ISCSI_SESSION_DEBUG(is, "len %zd -> %d", len,
2323 			    is->is_max_send_data_segment_length);
2324 			len = is->is_max_send_data_segment_length;
2325 		}
2326 
2327 		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2328 		if (error != 0) {
2329 			iscsi_outstanding_remove(is, io);
2330 			icl_pdu_free(request);
2331 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2332 				xpt_freeze_devq(ccb->ccb_h.path, 1);
2333 				ISCSI_SESSION_DEBUG(is, "freezing devq");
2334 			}
2335 			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2336 			xpt_done(ccb);
2337 			return;
2338 		}
2339 	}
2340 	iscsi_pdu_queue_locked(request);
2341 }
2342 
2343 static void
2344 iscsi_action(struct cam_sim *sim, union ccb *ccb)
2345 {
2346 	struct iscsi_session *is;
2347 
2348 	is = cam_sim_softc(sim);
2349 
2350 	ISCSI_SESSION_LOCK_ASSERT(is);
2351 
2352 	if (is->is_terminating ||
2353 	    (is->is_connected == false && fail_on_disconnection)) {
2354 		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2355 		xpt_done(ccb);
2356 		return;
2357 	}
2358 
2359 	/*
2360 	 * Make sure CAM doesn't sneak in a CCB just after freezing the queue.
2361 	 */
2362 	if (is->is_simq_frozen == true) {
2363 		ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
2364 		ccb->ccb_h.status |= CAM_REQUEUE_REQ;
2365 		/* Don't freeze the devq - the SIM queue is already frozen. */
2366 		xpt_done(ccb);
2367 		return;
2368 	}
2369 
2370 	switch (ccb->ccb_h.func_code) {
2371 	case XPT_PATH_INQ:
2372 	{
2373 		struct ccb_pathinq *cpi = &ccb->cpi;
2374 
2375 		cpi->version_num = 1;
2376 		cpi->hba_inquiry = PI_TAG_ABLE;
2377 		cpi->target_sprt = 0;
2378 		cpi->hba_misc = PIM_EXTLUNS;
2379 		/*
2380 		 * XXX: It shouldn't ever be NULL; this could be turned
2381 		 *      into a KASSERT eventually.
2382 		 */
2383 		if (is->is_conn == NULL)
2384 			ISCSI_WARN("NULL conn");
2385 		else if (is->is_conn->ic_unmapped)
2386 			cpi->hba_misc |= PIM_UNMAPPED;
2387 		cpi->hba_eng_cnt = 0;
2388 		cpi->max_target = 0;
2389 		/*
2390 		 * Note that the variable below is only relevant for targets
2391 		 * that don't claim compliance with anything above SPC2, which
2392 		 * means they don't support REPORT_LUNS.
2393 		 */
2394 		cpi->max_lun = 255;
2395 		cpi->initiator_id = ~0;
2396 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2397 		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2398 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2399 		cpi->unit_number = cam_sim_unit(sim);
2400 		cpi->bus_id = cam_sim_bus(sim);
2401 		cpi->base_transfer_speed = 150000; /* XXX */
2402 		cpi->transport = XPORT_ISCSI;
2403 		cpi->transport_version = 0;
2404 		cpi->protocol = PROTO_SCSI;
2405 		cpi->protocol_version = SCSI_REV_SPC3;
2406 		cpi->maxio = MAXPHYS;
2407 		cpi->ccb_h.status = CAM_REQ_CMP;
2408 		break;
2409 	}
2410 	case XPT_GET_TRAN_SETTINGS:
2411 	{
2412 		struct ccb_trans_settings	*cts;
2413 		struct ccb_trans_settings_scsi	*scsi;
2414 
2415 		cts = &ccb->cts;
2416 		scsi = &cts->proto_specific.scsi;
2417 
2418 		cts->protocol = PROTO_SCSI;
2419 		cts->protocol_version = SCSI_REV_SPC3;
2420 		cts->transport = XPORT_ISCSI;
2421 		cts->transport_version = 0;
2422 		scsi->valid = CTS_SCSI_VALID_TQ;
2423 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2424 		cts->ccb_h.status = CAM_REQ_CMP;
2425 		break;
2426 	}
2427 	case XPT_CALC_GEOMETRY:
2428 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2429 		ccb->ccb_h.status = CAM_REQ_CMP;
2430 		break;
2431 #if 0
2432 	/*
2433 	 * XXX: What's the point?
2434 	 */
2435 	case XPT_RESET_BUS:
2436 	case XPT_TERM_IO:
2437 		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2438 		ccb->ccb_h.status = CAM_REQ_CMP;
2439 		break;
2440 #endif
2441 	case XPT_ABORT:
2442 		iscsi_action_abort(is, ccb);
2443 		return;
2444 	case XPT_SCSI_IO:
2445 		iscsi_action_scsiio(is, ccb);
2446 		return;
2447 	default:
2448 #if 0
2449 		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2450 #endif
2451 		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2452 		break;
2453 	}
2454 	xpt_done(ccb);
2455 }
2456 
2457 static void
2458 iscsi_poll(struct cam_sim *sim)
2459 {
2460 
2461 	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2462 }
2463 
2464 static void
2465 iscsi_terminate_sessions(struct iscsi_softc *sc)
2466 {
2467 	struct iscsi_session *is;
2468 
2469 	sx_slock(&sc->sc_lock);
2470 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2471 		iscsi_session_terminate(is);
2472 	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2473 		ISCSI_DEBUG("waiting for sessions to terminate");
2474 		cv_wait(&sc->sc_cv, &sc->sc_lock);
2475 	}
2476 	ISCSI_DEBUG("all sessions terminated");
2477 	sx_sunlock(&sc->sc_lock);
2478 }
2479 
2480 static void
2481 iscsi_shutdown_pre(struct iscsi_softc *sc)
2482 {
2483 	struct iscsi_session *is;
2484 
2485 	if (!fail_on_shutdown)
2486 		return;
2487 
2488 	/*
2489 	 * If we have any sessions waiting for reconnection, request
2490 	 * maintenance thread to fail them immediately instead of waiting
2491 	 * for reconnect timeout.
2492 	 *
2493 	 * This prevents LUNs with mounted filesystems that are supported
2494 	 * by disconnected iSCSI sessions from hanging, however it will
2495 	 * fail all queued BIOs.
2496 	 */
2497 	ISCSI_DEBUG("forcing failing all disconnected sessions due to shutdown");
2498 
2499 	fail_on_disconnection = 1;
2500 
2501 	sx_slock(&sc->sc_lock);
2502 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2503 		ISCSI_SESSION_LOCK(is);
2504 		if (!is->is_connected) {
2505 			ISCSI_SESSION_DEBUG(is, "force failing disconnected session early");
2506 			iscsi_session_reconnect(is);
2507 		}
2508 		ISCSI_SESSION_UNLOCK(is);
2509 	}
2510 	sx_sunlock(&sc->sc_lock);
2511 }
2512 
2513 static void
2514 iscsi_shutdown_post(struct iscsi_softc *sc)
2515 {
2516 
2517 	if (!KERNEL_PANICKED()) {
2518 		ISCSI_DEBUG("removing all sessions due to shutdown");
2519 		iscsi_terminate_sessions(sc);
2520 	}
2521 }
2522 
2523 static int
2524 iscsi_load(void)
2525 {
2526 	int error;
2527 
2528 	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2529 	sx_init(&sc->sc_lock, "iscsi");
2530 	TAILQ_INIT(&sc->sc_sessions);
2531 	cv_init(&sc->sc_cv, "iscsi_cv");
2532 
2533 	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2534 	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2535 	    UMA_ALIGN_PTR, 0);
2536 
2537 	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2538 	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2539 	if (error != 0) {
2540 		ISCSI_WARN("failed to create device node, error %d", error);
2541 		return (error);
2542 	}
2543 	sc->sc_cdev->si_drv1 = sc;
2544 
2545 	sc->sc_shutdown_pre_eh = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2546 	    iscsi_shutdown_pre, sc, SHUTDOWN_PRI_FIRST);
2547 	/*
2548 	 * shutdown_post_sync needs to run after filesystem shutdown and before
2549 	 * CAM shutdown - otherwise when rebooting with an iSCSI session that is
2550 	 * disconnected but has outstanding requests, dashutdown() will hang on
2551 	 * cam_periph_runccb().
2552 	 */
2553 	sc->sc_shutdown_post_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2554 	    iscsi_shutdown_post, sc, SHUTDOWN_PRI_DEFAULT - 1);
2555 
2556 	return (0);
2557 }
2558 
2559 static int
2560 iscsi_unload(void)
2561 {
2562 
2563 	if (sc->sc_cdev != NULL) {
2564 		ISCSI_DEBUG("removing device node");
2565 		destroy_dev(sc->sc_cdev);
2566 		ISCSI_DEBUG("device node removed");
2567 	}
2568 
2569 	if (sc->sc_shutdown_pre_eh != NULL)
2570 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, sc->sc_shutdown_pre_eh);
2571 	if (sc->sc_shutdown_post_eh != NULL)
2572 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_post_eh);
2573 
2574 	iscsi_terminate_sessions(sc);
2575 
2576 	uma_zdestroy(iscsi_outstanding_zone);
2577 	sx_destroy(&sc->sc_lock);
2578 	cv_destroy(&sc->sc_cv);
2579 	free(sc, M_ISCSI);
2580 	return (0);
2581 }
2582 
2583 static int
2584 iscsi_quiesce(void)
2585 {
2586 	sx_slock(&sc->sc_lock);
2587 	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2588 		sx_sunlock(&sc->sc_lock);
2589 		return (EBUSY);
2590 	}
2591 	sx_sunlock(&sc->sc_lock);
2592 	return (0);
2593 }
2594 
2595 static int
2596 iscsi_modevent(module_t mod, int what, void *arg)
2597 {
2598 	int error;
2599 
2600 	switch (what) {
2601 	case MOD_LOAD:
2602 		error = iscsi_load();
2603 		break;
2604 	case MOD_UNLOAD:
2605 		error = iscsi_unload();
2606 		break;
2607 	case MOD_QUIESCE:
2608 		error = iscsi_quiesce();
2609 		break;
2610 	default:
2611 		error = EINVAL;
2612 		break;
2613 	}
2614 	return (error);
2615 }
2616 
2617 moduledata_t iscsi_data = {
2618 	"iscsi",
2619 	iscsi_modevent,
2620 	0
2621 };
2622 
2623 DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2624 MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2625 MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2626