xref: /freebsd/sys/dev/iscsi/iscsi.c (revision 413a368c90ea8621329ec702d46501ac7a4ffb4b)
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  * $FreeBSD$
30  */
31 
32 #include <sys/param.h>
33 #include <sys/condvar.h>
34 #include <sys/conf.h>
35 #include <sys/endian.h>
36 #include <sys/eventhandler.h>
37 #include <sys/file.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mutex.h>
43 #include <sys/module.h>
44 #include <sys/sysctl.h>
45 #include <sys/systm.h>
46 #include <sys/sx.h>
47 #include <vm/uma.h>
48 
49 #include <cam/cam.h>
50 #include <cam/cam_ccb.h>
51 #include <cam/cam_xpt.h>
52 #include <cam/cam_debug.h>
53 #include <cam/cam_sim.h>
54 #include <cam/cam_xpt_sim.h>
55 #include <cam/cam_xpt_periph.h>
56 #include <cam/cam_periph.h>
57 #include <cam/scsi/scsi_all.h>
58 #include <cam/scsi/scsi_message.h>
59 
60 #include "iscsi_ioctl.h"
61 #include "iscsi.h"
62 #include "icl.h"
63 #include "iscsi_proto.h"
64 
65 #ifdef ICL_KERNEL_PROXY
66 #include <sys/socketvar.h>
67 #endif
68 
69 #ifdef ICL_KERNEL_PROXY
70 FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
71 #endif
72 
73 /*
74  * XXX: This is global so the iscsi_unload() can access it.
75  * 	Think about how to do this properly.
76  */
77 static struct iscsi_softc	*sc;
78 
79 SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD, 0, "iSCSI initiator");
80 static int debug = 1;
81 SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
82     &debug, 0, "Enable debug messages");
83 static int ping_timeout = 5;
84 SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
85     0, "Timeout for ping (NOP-Out) requests, in seconds");
86 static int iscsid_timeout = 60;
87 SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
88     0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
89 static int login_timeout = 60;
90 SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
91     0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
92 static int maxtags = 255;
93 SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
94     0, "Max number of IO requests queued");
95 static int fail_on_disconnection = 0;
96 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
97     &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
98 
99 static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
100 static uma_zone_t iscsi_outstanding_zone;
101 
102 #define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
103 #define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
104 
105 #define	ISCSI_DEBUG(X, ...)						\
106 	do {								\
107 		if (debug > 1) 						\
108 			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
109 	} while (0)
110 
111 #define	ISCSI_WARN(X, ...)						\
112 	do {								\
113 		if (debug > 0) {					\
114 			printf("WARNING: %s: " X "\n",			\
115 			    __func__, ## __VA_ARGS__);			\
116 		}							\
117 	} while (0)
118 
119 #define	ISCSI_SESSION_DEBUG(S, X, ...)					\
120 	do {								\
121 		if (debug > 1) {					\
122 			printf("%s: %s (%s): " X "\n",			\
123 			    __func__, S->is_conf.isc_target_addr,	\
124 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
125 		}							\
126 	} while (0)
127 
128 #define	ISCSI_SESSION_WARN(S, X, ...)					\
129 	do {								\
130 		if (debug > 0) {					\
131 			printf("WARNING: %s (%s): " X "\n",		\
132 			    S->is_conf.isc_target_addr,			\
133 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
134 		}							\
135 	} while (0)
136 
137 #define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
138 #define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
139 #define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
140 
141 static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
142 		    int mode, struct thread *td);
143 
144 static struct cdevsw iscsi_cdevsw = {
145      .d_version = D_VERSION,
146      .d_ioctl   = iscsi_ioctl,
147      .d_name    = "iscsi",
148 };
149 
150 static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
151 static void	iscsi_pdu_queue(struct icl_pdu *request);
152 static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
153 static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
154 static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
155 static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
156 static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
157 static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
158 static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
159 static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
160 static void	iscsi_session_reconnect(struct iscsi_session *is);
161 static void	iscsi_session_terminate(struct iscsi_session *is);
162 static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
163 static void	iscsi_poll(struct cam_sim *sim);
164 static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
165 		    uint32_t initiator_task_tag);
166 static int	iscsi_outstanding_add(struct iscsi_session *is,
167 		    uint32_t initiator_task_tag, union ccb *ccb);
168 static void	iscsi_outstanding_remove(struct iscsi_session *is,
169 		    struct iscsi_outstanding *io);
170 
171 static bool
172 iscsi_pdu_prepare(struct icl_pdu *request)
173 {
174 	struct iscsi_session *is;
175 	struct iscsi_bhs_scsi_command *bhssc;
176 
177 	is = PDU_SESSION(request);
178 
179 	ISCSI_SESSION_LOCK_ASSERT(is);
180 
181 	/*
182 	 * We're only using fields common for all the request
183 	 * (initiator -> target) PDUs.
184 	 */
185 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
186 
187 	/*
188 	 * Data-Out PDU does not contain CmdSN.
189 	 */
190 	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
191 		if (is->is_cmdsn > is->is_maxcmdsn &&
192 		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
193 			/*
194 			 * Current MaxCmdSN prevents us from sending any more
195 			 * SCSI Command PDUs to the target; postpone the PDU.
196 			 * It will get resent by either iscsi_pdu_queue(),
197 			 * or by maintenance thread.
198 			 */
199 #if 0
200 			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %d, ExpCmdSN %d, MaxCmdSN %d, opcode 0x%x",
201 			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn, bhssc->bhssc_opcode);
202 #endif
203 			return (true);
204 		}
205 		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
206 		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
207 			is->is_cmdsn++;
208 	}
209 	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
210 
211 	return (false);
212 }
213 
214 static void
215 iscsi_session_send_postponed(struct iscsi_session *is)
216 {
217 	struct icl_pdu *request;
218 	bool postpone;
219 
220 	ISCSI_SESSION_LOCK_ASSERT(is);
221 
222 	while (!STAILQ_EMPTY(&is->is_postponed)) {
223 		request = STAILQ_FIRST(&is->is_postponed);
224 		postpone = iscsi_pdu_prepare(request);
225 		if (postpone)
226 			break;
227 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
228 		icl_pdu_queue(request);
229 	}
230 }
231 
232 static void
233 iscsi_pdu_queue_locked(struct icl_pdu *request)
234 {
235 	struct iscsi_session *is;
236 	bool postpone;
237 
238 	is = PDU_SESSION(request);
239 	ISCSI_SESSION_LOCK_ASSERT(is);
240 	iscsi_session_send_postponed(is);
241 	postpone = iscsi_pdu_prepare(request);
242 	if (postpone) {
243 		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
244 		return;
245 	}
246 	icl_pdu_queue(request);
247 }
248 
249 static void
250 iscsi_pdu_queue(struct icl_pdu *request)
251 {
252 	struct iscsi_session *is;
253 
254 	is = PDU_SESSION(request);
255 	ISCSI_SESSION_LOCK(is);
256 	iscsi_pdu_queue_locked(request);
257 	ISCSI_SESSION_UNLOCK(is);
258 }
259 
260 static void
261 iscsi_session_logout(struct iscsi_session *is)
262 {
263 	struct icl_pdu *request;
264 	struct iscsi_bhs_logout_request *bhslr;
265 
266 	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
267 	if (request == NULL)
268 		return;
269 
270 	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
271 	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
272 	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
273 	iscsi_pdu_queue_locked(request);
274 }
275 
276 static void
277 iscsi_session_terminate_tasks(struct iscsi_session *is, bool requeue)
278 {
279 	struct iscsi_outstanding *io, *tmp;
280 
281 	ISCSI_SESSION_LOCK_ASSERT(is);
282 
283 	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
284 		if (requeue) {
285 			io->io_ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
286 			io->io_ccb->ccb_h.status |= CAM_REQUEUE_REQ;
287 		} else {
288 			io->io_ccb->ccb_h.status = CAM_REQ_ABORTED;
289 		}
290 
291 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
292 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
293 			ISCSI_SESSION_DEBUG(is, "freezing devq");
294 		}
295 		io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
296 		xpt_done(io->io_ccb);
297 		iscsi_outstanding_remove(is, io);
298 	}
299 }
300 
301 static void
302 iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
303 {
304 	struct icl_pdu *pdu;
305 
306 	ISCSI_SESSION_LOCK_ASSERT(is);
307 
308 	/*
309 	 * Don't queue any new PDUs.
310 	 */
311 	if (is->is_sim != NULL && is->is_simq_frozen == false) {
312 		ISCSI_SESSION_DEBUG(is, "freezing");
313 		xpt_freeze_simq(is->is_sim, 1);
314 		is->is_simq_frozen = true;
315 	}
316 
317 	/*
318 	 * Remove postponed PDUs.
319 	 */
320 	while (!STAILQ_EMPTY(&is->is_postponed)) {
321 		pdu = STAILQ_FIRST(&is->is_postponed);
322 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
323 		icl_pdu_free(pdu);
324 	}
325 
326 	if (destroy_sim == false) {
327 		/*
328 		 * Terminate SCSI tasks, asking CAM to requeue them.
329 		 */
330 		iscsi_session_terminate_tasks(is, true);
331 		return;
332 	}
333 
334 	iscsi_session_terminate_tasks(is, false);
335 
336 	if (is->is_sim == NULL)
337 		return;
338 
339 	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
340 	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
341 
342 	if (is->is_simq_frozen) {
343 		xpt_release_simq(is->is_sim, 1);
344 		is->is_simq_frozen = false;
345 	}
346 
347 	xpt_free_path(is->is_path);
348 	is->is_path = NULL;
349 	xpt_bus_deregister(cam_sim_path(is->is_sim));
350 	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
351 	is->is_sim = NULL;
352 	is->is_devq = NULL;
353 }
354 
355 static void
356 iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
357 {
358 
359 	icl_conn_shutdown(is->is_conn);
360 	icl_conn_close(is->is_conn);
361 
362 	ISCSI_SESSION_LOCK(is);
363 
364 	is->is_connected = false;
365 	is->is_reconnecting = false;
366 	is->is_login_phase = false;
367 
368 #ifdef ICL_KERNEL_PROXY
369 	if (is->is_login_pdu != NULL) {
370 		icl_pdu_free(is->is_login_pdu);
371 		is->is_login_pdu = NULL;
372 	}
373 	cv_signal(&is->is_login_cv);
374 #endif
375 
376 	if (fail_on_disconnection) {
377 		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
378 		iscsi_session_cleanup(is, true);
379 	} else {
380 		iscsi_session_cleanup(is, false);
381 	}
382 
383 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
384 	    ("destroying session with active tasks"));
385 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
386 	    ("destroying session with postponed PDUs"));
387 
388 	/*
389 	 * Request immediate reconnection from iscsid(8).
390 	 */
391 	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
392 	is->is_waiting_for_iscsid = true;
393 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
394 	is->is_timeout = 0;
395 	ISCSI_SESSION_UNLOCK(is);
396 	cv_signal(&is->is_softc->sc_cv);
397 }
398 
399 static void
400 iscsi_maintenance_thread_terminate(struct iscsi_session *is)
401 {
402 	struct iscsi_softc *sc;
403 
404 	sc = is->is_softc;
405 	sx_xlock(&sc->sc_lock);
406 	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
407 	sx_xunlock(&sc->sc_lock);
408 
409 	icl_conn_close(is->is_conn);
410 
411 	ISCSI_SESSION_LOCK(is);
412 
413 	KASSERT(is->is_terminating, ("is_terminating == false"));
414 
415 #ifdef ICL_KERNEL_PROXY
416 	if (is->is_login_pdu != NULL) {
417 		icl_pdu_free(is->is_login_pdu);
418 		is->is_login_pdu = NULL;
419 	}
420 	cv_signal(&is->is_login_cv);
421 #endif
422 
423 	callout_drain(&is->is_callout);
424 
425 	iscsi_session_cleanup(is, true);
426 
427 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
428 	    ("destroying session with active tasks"));
429 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
430 	    ("destroying session with postponed PDUs"));
431 
432 	ISCSI_SESSION_UNLOCK(is);
433 
434 	icl_conn_free(is->is_conn);
435 	mtx_destroy(&is->is_lock);
436 	cv_destroy(&is->is_maintenance_cv);
437 #ifdef ICL_KERNEL_PROXY
438 	cv_destroy(&is->is_login_cv);
439 #endif
440 	ISCSI_SESSION_DEBUG(is, "terminated");
441 	free(is, M_ISCSI);
442 
443 	/*
444 	 * The iscsi_unload() routine might be waiting.
445 	 */
446 	cv_signal(&sc->sc_cv);
447 }
448 
449 static void
450 iscsi_maintenance_thread(void *arg)
451 {
452 	struct iscsi_session *is;
453 
454 	is = arg;
455 
456 	for (;;) {
457 		ISCSI_SESSION_LOCK(is);
458 		if (is->is_reconnecting == false &&
459 		    is->is_terminating == false &&
460 		    STAILQ_EMPTY(&is->is_postponed))
461 			cv_wait(&is->is_maintenance_cv, &is->is_lock);
462 
463 		if (is->is_reconnecting) {
464 			ISCSI_SESSION_UNLOCK(is);
465 			iscsi_maintenance_thread_reconnect(is);
466 			continue;
467 		}
468 
469 		if (is->is_terminating) {
470 			ISCSI_SESSION_UNLOCK(is);
471 			iscsi_maintenance_thread_terminate(is);
472 			kthread_exit();
473 			return;
474 		}
475 
476 		iscsi_session_send_postponed(is);
477 		ISCSI_SESSION_UNLOCK(is);
478 	}
479 }
480 
481 static void
482 iscsi_session_reconnect(struct iscsi_session *is)
483 {
484 
485 	/*
486 	 * XXX: We can't use locking here, because
487 	 * 	it's being called from various contexts.
488 	 * 	Hope it doesn't break anything.
489 	 */
490 	if (is->is_reconnecting)
491 		return;
492 
493 	is->is_reconnecting = true;
494 	cv_signal(&is->is_maintenance_cv);
495 }
496 
497 static void
498 iscsi_session_terminate(struct iscsi_session *is)
499 {
500 	if (is->is_terminating)
501 		return;
502 
503 	is->is_terminating = true;
504 
505 #if 0
506 	iscsi_session_logout(is);
507 #endif
508 	cv_signal(&is->is_maintenance_cv);
509 }
510 
511 static void
512 iscsi_callout(void *context)
513 {
514 	struct icl_pdu *request;
515 	struct iscsi_bhs_nop_out *bhsno;
516 	struct iscsi_session *is;
517 	bool reconnect_needed = false;
518 
519 	is = context;
520 
521 	if (is->is_terminating)
522 		return;
523 
524 	callout_schedule(&is->is_callout, 1 * hz);
525 
526 	ISCSI_SESSION_LOCK(is);
527 	is->is_timeout++;
528 
529 	if (is->is_waiting_for_iscsid) {
530 		if (is->is_timeout > iscsid_timeout) {
531 			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
532 			    "for %d seconds; reconnecting",
533 			    is->is_timeout);
534 			reconnect_needed = true;
535 		}
536 		goto out;
537 	}
538 
539 	if (is->is_login_phase) {
540 		if (is->is_timeout > login_timeout) {
541 			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
542 			    "reconnecting", is->is_timeout);
543 			reconnect_needed = true;
544 		}
545 		goto out;
546 	}
547 
548 	if (is->is_timeout >= ping_timeout) {
549 		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
550 		    "reconnecting", ping_timeout);
551 		reconnect_needed = true;
552 		goto out;
553 	}
554 
555 	ISCSI_SESSION_UNLOCK(is);
556 
557 	/*
558 	 * If the ping was reset less than one second ago - which means
559 	 * that we've received some PDU during the last second - assume
560 	 * the traffic flows correctly and don't bother sending a NOP-Out.
561 	 *
562 	 * (It's 2 - one for one second, and one for incrementing is_timeout
563 	 * earlier in this routine.)
564 	 */
565 	if (is->is_timeout < 2)
566 		return;
567 
568 	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
569 	if (request == NULL) {
570 		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
571 		return;
572 	}
573 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
574 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
575 	    ISCSI_BHS_OPCODE_IMMEDIATE;
576 	bhsno->bhsno_flags = 0x80;
577 	bhsno->bhsno_target_transfer_tag = 0xffffffff;
578 	iscsi_pdu_queue(request);
579 	return;
580 
581 out:
582 	ISCSI_SESSION_UNLOCK(is);
583 
584 	if (reconnect_needed)
585 		iscsi_session_reconnect(is);
586 }
587 
588 static void
589 iscsi_pdu_update_statsn(const struct icl_pdu *response)
590 {
591 	const struct iscsi_bhs_data_in *bhsdi;
592 	struct iscsi_session *is;
593 	uint32_t expcmdsn, maxcmdsn;
594 
595 	is = PDU_SESSION(response);
596 
597 	ISCSI_SESSION_LOCK_ASSERT(is);
598 
599 	/*
600 	 * We're only using fields common for all the response
601 	 * (target -> initiator) PDUs.
602 	 */
603 	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
604 	/*
605 	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
606 	 * and Residual Count only have meaningful content if the S bit
607 	 * is set to 1", so we also need to check the bit specific for
608 	 * Data-In PDU.
609 	 */
610 	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
611 	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
612 		if (ntohl(bhsdi->bhsdi_statsn) < is->is_statsn) {
613 			ISCSI_SESSION_WARN(is,
614 			    "PDU StatSN %d >= session StatSN %d, opcode 0x%x",
615 			    is->is_statsn, ntohl(bhsdi->bhsdi_statsn),
616 			    bhsdi->bhsdi_opcode);
617 		}
618 		is->is_statsn = ntohl(bhsdi->bhsdi_statsn);
619 	}
620 
621 	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
622 	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
623 
624 	/*
625 	 * XXX: Compare using Serial Arithmetic Sense.
626 	 */
627 	if (maxcmdsn + 1 < expcmdsn) {
628 		ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d + 1 < PDU ExpCmdSN %d; ignoring",
629 		    maxcmdsn, expcmdsn);
630 	} else {
631 		if (maxcmdsn > is->is_maxcmdsn) {
632 			is->is_maxcmdsn = maxcmdsn;
633 
634 			/*
635 			 * Command window increased; kick the maintanance thread
636 			 * to send out postponed commands.
637 			 */
638 			if (!STAILQ_EMPTY(&is->is_postponed))
639 				cv_signal(&is->is_maintenance_cv);
640 		} else if (maxcmdsn < is->is_maxcmdsn) {
641 			ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d < session MaxCmdSN %d; ignoring",
642 			    maxcmdsn, is->is_maxcmdsn);
643 		}
644 
645 		if (expcmdsn > is->is_expcmdsn) {
646 			is->is_expcmdsn = expcmdsn;
647 		} else if (expcmdsn < is->is_expcmdsn) {
648 			ISCSI_SESSION_DEBUG(is, "PDU ExpCmdSN %d < session ExpCmdSN %d; ignoring",
649 			    expcmdsn, is->is_expcmdsn);
650 		}
651 	}
652 
653 	/*
654 	 * Every incoming PDU - not just NOP-In - resets the ping timer.
655 	 * The purpose of the timeout is to reset the connection when it stalls;
656 	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
657 	 * in some queue.
658 	 */
659 	is->is_timeout = 0;
660 }
661 
662 static void
663 iscsi_receive_callback(struct icl_pdu *response)
664 {
665 	struct iscsi_session *is;
666 
667 	is = PDU_SESSION(response);
668 
669 	ISCSI_SESSION_LOCK(is);
670 
671 #ifdef ICL_KERNEL_PROXY
672 	if (is->is_login_phase) {
673 		if (is->is_login_pdu == NULL)
674 			is->is_login_pdu = response;
675 		else
676 			icl_pdu_free(response);
677 		ISCSI_SESSION_UNLOCK(is);
678 		cv_signal(&is->is_login_cv);
679 		return;
680 	}
681 #endif
682 
683 	iscsi_pdu_update_statsn(response);
684 
685 	/*
686 	 * The handling routine is responsible for freeing the PDU
687 	 * when it's no longer needed.
688 	 */
689 	switch (response->ip_bhs->bhs_opcode) {
690 	case ISCSI_BHS_OPCODE_NOP_IN:
691 		iscsi_pdu_handle_nop_in(response);
692 		break;
693 	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
694 		iscsi_pdu_handle_scsi_response(response);
695 		break;
696 	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
697 		iscsi_pdu_handle_data_in(response);
698 		break;
699 	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
700 		iscsi_pdu_handle_logout_response(response);
701 		break;
702 	case ISCSI_BHS_OPCODE_R2T:
703 		iscsi_pdu_handle_r2t(response);
704 		break;
705 	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
706 		iscsi_pdu_handle_async_message(response);
707 		break;
708 	case ISCSI_BHS_OPCODE_REJECT:
709 		iscsi_pdu_handle_reject(response);
710 		break;
711 	default:
712 		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
713 		    "opcode 0x%x; reconnecting",
714 		    response->ip_bhs->bhs_opcode);
715 		iscsi_session_reconnect(is);
716 		icl_pdu_free(response);
717 	}
718 
719 	ISCSI_SESSION_UNLOCK(is);
720 }
721 
722 static void
723 iscsi_error_callback(struct icl_conn *ic)
724 {
725 	struct iscsi_session *is;
726 
727 	is = CONN_SESSION(ic);
728 
729 	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
730 	iscsi_session_reconnect(is);
731 }
732 
733 static void
734 iscsi_pdu_handle_nop_in(struct icl_pdu *response)
735 {
736 	struct iscsi_session *is;
737 	struct iscsi_bhs_nop_out *bhsno;
738 	struct iscsi_bhs_nop_in *bhsni;
739 	struct icl_pdu *request;
740 	void *data = NULL;
741 	size_t datasize;
742 	int error;
743 
744 	is = PDU_SESSION(response);
745 	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
746 
747 	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
748 		/*
749 		 * Nothing to do; iscsi_pdu_update_statsn() already
750 		 * zeroed the timeout.
751 		 */
752 		icl_pdu_free(response);
753 		return;
754 	}
755 
756 	datasize = icl_pdu_data_segment_length(response);
757 	if (datasize > 0) {
758 		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
759 		if (data == NULL) {
760 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
761 			    "reconnecting");
762 			icl_pdu_free(response);
763 			iscsi_session_reconnect(is);
764 			return;
765 		}
766 		icl_pdu_get_data(response, 0, data, datasize);
767 	}
768 
769 	request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
770 	if (request == NULL) {
771 		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
772 		    "reconnecting");
773 		free(data, M_ISCSI);
774 		icl_pdu_free(response);
775 		iscsi_session_reconnect(is);
776 		return;
777 	}
778 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
779 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
780 	    ISCSI_BHS_OPCODE_IMMEDIATE;
781 	bhsno->bhsno_flags = 0x80;
782 	bhsno->bhsno_initiator_task_tag = 0xffffffff;
783 	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
784 	if (datasize > 0) {
785 		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
786 		if (error != 0) {
787 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
788 			    "reconnecting");
789 			free(data, M_ISCSI);
790 			icl_pdu_free(request);
791 			icl_pdu_free(response);
792 			iscsi_session_reconnect(is);
793 			return;
794 		}
795 		free(data, M_ISCSI);
796 	}
797 
798 	icl_pdu_free(response);
799 	iscsi_pdu_queue_locked(request);
800 }
801 
802 static void
803 iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
804 {
805 	struct iscsi_bhs_scsi_response *bhssr;
806 	struct iscsi_outstanding *io;
807 	struct iscsi_session *is;
808 	struct ccb_scsiio *csio;
809 	size_t data_segment_len;
810 	uint16_t sense_len;
811 
812 	is = PDU_SESSION(response);
813 
814 	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
815 	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
816 	if (io == NULL) {
817 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
818 		icl_pdu_free(response);
819 		iscsi_session_reconnect(is);
820 		return;
821 	}
822 
823 	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
824 		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
825  		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
826  			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
827 			ISCSI_SESSION_DEBUG(is, "freezing devq");
828 		}
829  		io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
830 	} else if (bhssr->bhssr_status == 0) {
831 		io->io_ccb->ccb_h.status = CAM_REQ_CMP;
832 	} else {
833  		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
834  			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
835 			ISCSI_SESSION_DEBUG(is, "freezing devq");
836 		}
837  		io->io_ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
838 		io->io_ccb->csio.scsi_status = bhssr->bhssr_status;
839 	}
840 
841 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_OVERFLOW) {
842 		ISCSI_SESSION_WARN(is, "target indicated residual overflow");
843 		icl_pdu_free(response);
844 		iscsi_session_reconnect(is);
845 		return;
846 	}
847 
848 	csio = &io->io_ccb->csio;
849 
850 	data_segment_len = icl_pdu_data_segment_length(response);
851 	if (data_segment_len > 0) {
852 		if (data_segment_len < sizeof(sense_len)) {
853 			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
854 			    data_segment_len);
855 			if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
856 				xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
857 				ISCSI_SESSION_DEBUG(is, "freezing devq");
858 			}
859 			io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
860 			goto out;
861 		}
862 		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
863 		sense_len = ntohs(sense_len);
864 #if 0
865 		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
866 		    sense_len, data_segment_len);
867 #endif
868 		if (sizeof(sense_len) + sense_len > data_segment_len) {
869 			ISCSI_SESSION_WARN(is, "truncated data segment "
870 			    "(%zd bytes, should be %zd)",
871 			    data_segment_len, sizeof(sense_len) + sense_len);
872 			if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
873 				xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
874 				ISCSI_SESSION_DEBUG(is, "freezing devq");
875 			}
876 			io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
877 			goto out;
878 		} else if (sizeof(sense_len) + sense_len < data_segment_len)
879 			ISCSI_SESSION_WARN(is, "oversize data segment "
880 			    "(%zd bytes, should be %zd)",
881 			    data_segment_len, sizeof(sense_len) + sense_len);
882 		if (sense_len > csio->sense_len) {
883 			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
884 			    sense_len, csio->sense_len);
885 			sense_len = csio->sense_len;
886 		}
887 		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
888 		csio->sense_resid = csio->sense_len - sense_len;
889 		io->io_ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
890 	}
891 
892 out:
893 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
894 		csio->resid = ntohl(bhssr->bhssr_residual_count);
895 
896 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
897 		KASSERT(io->io_received <= csio->dxfer_len,
898 		    ("io->io_received > csio->dxfer_len"));
899 		if (io->io_received < csio->dxfer_len) {
900 			if (csio->resid != csio->dxfer_len - io->io_received) {
901 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
902 				    "target indicates %d, we calculated %zd",
903 				    csio->resid,
904 				    csio->dxfer_len - io->io_received);
905 			}
906 			csio->resid = csio->dxfer_len - io->io_received;
907 		}
908 	}
909 
910 	xpt_done(io->io_ccb);
911 	iscsi_outstanding_remove(is, io);
912 	icl_pdu_free(response);
913 }
914 
915 static void
916 iscsi_pdu_handle_data_in(struct icl_pdu *response)
917 {
918 	struct iscsi_bhs_data_in *bhsdi;
919 	struct iscsi_outstanding *io;
920 	struct iscsi_session *is;
921 	struct ccb_scsiio *csio;
922 	size_t data_segment_len;
923 
924 	is = PDU_SESSION(response);
925 	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
926 	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
927 	if (io == NULL) {
928 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
929 		icl_pdu_free(response);
930 		iscsi_session_reconnect(is);
931 		return;
932 	}
933 
934 	data_segment_len = icl_pdu_data_segment_length(response);
935 	if (data_segment_len == 0) {
936 		/*
937 		 * "The sending of 0 length data segments should be avoided,
938 		 * but initiators and targets MUST be able to properly receive
939 		 * 0 length data segments."
940 		 */
941 		icl_pdu_free(response);
942 		return;
943 	}
944 
945 	/*
946 	 * We need to track this for security reasons - without it, malicious target
947 	 * could respond to SCSI READ without sending Data-In PDUs, which would result
948 	 * in read operation on the initiator side returning random kernel data.
949 	 */
950 	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
951 		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
952 		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
953 		icl_pdu_free(response);
954 		iscsi_session_reconnect(is);
955 		return;
956 	}
957 
958 	csio = &io->io_ccb->csio;
959 
960 	if (io->io_received + data_segment_len > csio->dxfer_len) {
961 		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
962 		    "at offset %zd, buffer is %d)",
963 		    data_segment_len, io->io_received, csio->dxfer_len);
964 		icl_pdu_free(response);
965 		iscsi_session_reconnect(is);
966 		return;
967 	}
968 
969 	icl_pdu_get_data(response, 0, csio->data_ptr + io->io_received, data_segment_len);
970 	io->io_received += data_segment_len;
971 
972 	/*
973 	 * XXX: Check DataSN.
974 	 * XXX: Check F.
975 	 */
976 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
977 		/*
978 		 * Nothing more to do.
979 		 */
980 		icl_pdu_free(response);
981 		return;
982 	}
983 
984 	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
985 	if (bhsdi->bhsdi_status == 0) {
986 		io->io_ccb->ccb_h.status = CAM_REQ_CMP;
987 	} else {
988 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
989 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
990 			ISCSI_SESSION_DEBUG(is, "freezing devq");
991 		}
992 		io->io_ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
993 		csio->scsi_status = bhsdi->bhsdi_status;
994 	}
995 
996 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
997 		KASSERT(io->io_received <= csio->dxfer_len,
998 		    ("io->io_received > csio->dxfer_len"));
999 		if (io->io_received < csio->dxfer_len) {
1000 			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1001 			if (csio->resid != csio->dxfer_len - io->io_received) {
1002 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1003 				    "target indicates %d, we calculated %zd",
1004 				    csio->resid,
1005 				    csio->dxfer_len - io->io_received);
1006 			}
1007 			csio->resid = csio->dxfer_len - io->io_received;
1008 		}
1009 	}
1010 
1011 	xpt_done(io->io_ccb);
1012 	iscsi_outstanding_remove(is, io);
1013 	icl_pdu_free(response);
1014 }
1015 
1016 static void
1017 iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1018 {
1019 
1020 	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1021 	icl_pdu_free(response);
1022 }
1023 
1024 static void
1025 iscsi_pdu_handle_r2t(struct icl_pdu *response)
1026 {
1027 	struct icl_pdu *request;
1028 	struct iscsi_session *is;
1029 	struct iscsi_bhs_r2t *bhsr2t;
1030 	struct iscsi_bhs_data_out *bhsdo;
1031 	struct iscsi_outstanding *io;
1032 	struct ccb_scsiio *csio;
1033 	size_t off, len, total_len;
1034 	int error;
1035 
1036 	is = PDU_SESSION(response);
1037 
1038 	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1039 	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1040 	if (io == NULL) {
1041 		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1042 		    bhsr2t->bhsr2t_initiator_task_tag);
1043 		icl_pdu_free(response);
1044 		iscsi_session_reconnect(is);
1045 		return;
1046 	}
1047 
1048 	csio = &io->io_ccb->csio;
1049 
1050 	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1051 		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1052 		icl_pdu_free(response);
1053 		iscsi_session_reconnect(is);
1054 		return;
1055 	}
1056 
1057 	/*
1058 	 * XXX: Verify R2TSN.
1059 	 */
1060 
1061 	io->io_datasn = 0;
1062 
1063 	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1064 	if (off > csio->dxfer_len) {
1065 		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1066 		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1067 		icl_pdu_free(response);
1068 		iscsi_session_reconnect(is);
1069 		return;
1070 	}
1071 
1072 	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1073 	if (total_len == 0 || total_len > csio->dxfer_len) {
1074 		ISCSI_SESSION_WARN(is, "target requested invalid length "
1075 		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1076 		icl_pdu_free(response);
1077 		iscsi_session_reconnect(is);
1078 		return;
1079 	}
1080 
1081 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1082 
1083 	for (;;) {
1084 		len = total_len;
1085 
1086 		if (len > is->is_max_data_segment_length)
1087 			len = is->is_max_data_segment_length;
1088 
1089 		if (off + len > csio->dxfer_len) {
1090 			ISCSI_SESSION_WARN(is, "target requested invalid "
1091 			    "length/offset %zd, buffer is %d; reconnecting",
1092 			    off + len, csio->dxfer_len);
1093 			icl_pdu_free(response);
1094 			iscsi_session_reconnect(is);
1095 			return;
1096 		}
1097 
1098 		request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
1099 		if (request == NULL) {
1100 			icl_pdu_free(response);
1101 			iscsi_session_reconnect(is);
1102 			return;
1103 		}
1104 
1105 		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1106 		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1107 		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1108 		bhsdo->bhsdo_initiator_task_tag =
1109 		    bhsr2t->bhsr2t_initiator_task_tag;
1110 		bhsdo->bhsdo_target_transfer_tag =
1111 		    bhsr2t->bhsr2t_target_transfer_tag;
1112 		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1113 		bhsdo->bhsdo_buffer_offset = htonl(off);
1114 		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1115 		    M_NOWAIT);
1116 		if (error != 0) {
1117 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1118 			    "reconnecting");
1119 			icl_pdu_free(request);
1120 			icl_pdu_free(response);
1121 			iscsi_session_reconnect(is);
1122 			return;
1123 		}
1124 
1125 		off += len;
1126 		total_len -= len;
1127 
1128 		if (total_len == 0) {
1129 			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1130 			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1131 		} else {
1132 			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1133 		}
1134 
1135 		iscsi_pdu_queue_locked(request);
1136 
1137 		if (total_len == 0)
1138 			break;
1139 	}
1140 
1141 	icl_pdu_free(response);
1142 }
1143 
1144 static void
1145 iscsi_pdu_handle_async_message(struct icl_pdu *response)
1146 {
1147 	struct iscsi_bhs_asynchronous_message *bhsam;
1148 	struct iscsi_session *is;
1149 
1150 	is = PDU_SESSION(response);
1151 	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1152 	switch (bhsam->bhsam_async_event) {
1153 	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1154 		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1155 		iscsi_session_logout(is);
1156 		iscsi_session_terminate(is);
1157 		break;
1158 	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1159 		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the connection");
1160 		break;
1161 	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1162 		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the session");
1163 		break;
1164 	default:
1165 		/*
1166 		 * XXX: Technically, we're obligated to also handle
1167 		 * 	parameter renegotiation.
1168 		 */
1169 		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1170 		break;
1171 	}
1172 
1173 	icl_pdu_free(response);
1174 }
1175 
1176 static void
1177 iscsi_pdu_handle_reject(struct icl_pdu *response)
1178 {
1179 	struct iscsi_bhs_reject *bhsr;
1180 	struct iscsi_session *is;
1181 
1182 	is = PDU_SESSION(response);
1183 	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1184 	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1185 	    bhsr->bhsr_reason);
1186 
1187 	icl_pdu_free(response);
1188 }
1189 
1190 static int
1191 iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1192     struct iscsi_daemon_request *request)
1193 {
1194 	struct iscsi_session *is;
1195 	int error;
1196 
1197 	sx_slock(&sc->sc_lock);
1198 	for (;;) {
1199 		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1200 			ISCSI_SESSION_LOCK(is);
1201 			if (is->is_waiting_for_iscsid)
1202 				break;
1203 			ISCSI_SESSION_UNLOCK(is);
1204 		}
1205 
1206 		if (is == NULL) {
1207 			/*
1208 			 * No session requires attention from iscsid(8); wait.
1209 			 */
1210 			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1211 			if (error != 0) {
1212 				sx_sunlock(&sc->sc_lock);
1213 				return (error);
1214 			}
1215 			continue;
1216 		}
1217 
1218 		is->is_waiting_for_iscsid = false;
1219 		is->is_login_phase = true;
1220 		is->is_reason[0] = '\0';
1221 		ISCSI_SESSION_UNLOCK(is);
1222 
1223 		request->idr_session_id = is->is_id;
1224 		memcpy(&request->idr_isid, &is->is_isid,
1225 		    sizeof(request->idr_isid));
1226 		request->idr_tsih = 0;	/* New or reinstated session. */
1227 		memcpy(&request->idr_conf, &is->is_conf,
1228 		    sizeof(request->idr_conf));
1229 
1230 		sx_sunlock(&sc->sc_lock);
1231 		return (0);
1232 	}
1233 }
1234 
1235 static int
1236 iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1237     struct iscsi_daemon_handoff *handoff)
1238 {
1239 	struct iscsi_session *is;
1240 	int error;
1241 
1242 	sx_slock(&sc->sc_lock);
1243 
1244 	/*
1245 	 * Find the session to hand off socket to.
1246 	 */
1247 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1248 		if (is->is_id == handoff->idh_session_id)
1249 			break;
1250 	}
1251 	if (is == NULL) {
1252 		sx_sunlock(&sc->sc_lock);
1253 		return (ESRCH);
1254 	}
1255 	ISCSI_SESSION_LOCK(is);
1256 	if (is->is_conf.isc_discovery || is->is_terminating) {
1257 		ISCSI_SESSION_UNLOCK(is);
1258 		sx_sunlock(&sc->sc_lock);
1259 		return (EINVAL);
1260 	}
1261 	if (is->is_connected) {
1262 		/*
1263 		 * This might have happened because another iscsid(8)
1264 		 * instance handed off the connection in the meantime.
1265 		 * Just return.
1266 		 */
1267 		ISCSI_SESSION_WARN(is, "handoff on already connected "
1268 		    "session");
1269 		ISCSI_SESSION_UNLOCK(is);
1270 		sx_sunlock(&sc->sc_lock);
1271 		return (EBUSY);
1272 	}
1273 
1274 	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1275 	    sizeof(is->is_target_alias));
1276 	is->is_tsih = handoff->idh_tsih;
1277 	is->is_statsn = handoff->idh_statsn;
1278 	is->is_initial_r2t = handoff->idh_initial_r2t;
1279 	is->is_immediate_data = handoff->idh_immediate_data;
1280 	is->is_max_data_segment_length = handoff->idh_max_data_segment_length;
1281 	is->is_max_burst_length = handoff->idh_max_burst_length;
1282 	is->is_first_burst_length = handoff->idh_first_burst_length;
1283 
1284 	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1285 		is->is_conn->ic_header_crc32c = true;
1286 	else
1287 		is->is_conn->ic_header_crc32c = false;
1288 	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1289 		is->is_conn->ic_data_crc32c = true;
1290 	else
1291 		is->is_conn->ic_data_crc32c = false;
1292 
1293 	is->is_cmdsn = 0;
1294 	is->is_expcmdsn = 0;
1295 	is->is_maxcmdsn = 0;
1296 	is->is_waiting_for_iscsid = false;
1297 	is->is_login_phase = false;
1298 	is->is_timeout = 0;
1299 	is->is_connected = true;
1300 	is->is_reason[0] = '\0';
1301 
1302 	ISCSI_SESSION_UNLOCK(is);
1303 
1304 #ifdef ICL_KERNEL_PROXY
1305 	if (handoff->idh_socket != 0) {
1306 #endif
1307 		/*
1308 		 * Handoff without using ICL proxy.
1309 		 */
1310 		error = icl_conn_handoff(is->is_conn, handoff->idh_socket);
1311 		if (error != 0) {
1312 			sx_sunlock(&sc->sc_lock);
1313 			iscsi_session_terminate(is);
1314 			return (error);
1315 		}
1316 #ifdef ICL_KERNEL_PROXY
1317 	}
1318 #endif
1319 
1320 	sx_sunlock(&sc->sc_lock);
1321 
1322 	if (is->is_sim != NULL) {
1323 		/*
1324 		 * When reconnecting, there already is SIM allocated for the session.
1325 		 */
1326 		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1327 		ISCSI_SESSION_LOCK(is);
1328 		ISCSI_SESSION_DEBUG(is, "releasing");
1329 		xpt_release_simq(is->is_sim, 1);
1330 		is->is_simq_frozen = false;
1331 		ISCSI_SESSION_UNLOCK(is);
1332 
1333 	} else {
1334 		ISCSI_SESSION_LOCK(is);
1335 		is->is_devq = cam_simq_alloc(maxtags);
1336 		if (is->is_devq == NULL) {
1337 			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1338 			iscsi_session_terminate(is);
1339 			return (ENOMEM);
1340 		}
1341 
1342 		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1343 		    is, is->is_id /* unit */, &is->is_lock,
1344 		    1, maxtags, is->is_devq);
1345 		if (is->is_sim == NULL) {
1346 			ISCSI_SESSION_UNLOCK(is);
1347 			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1348 			cam_simq_free(is->is_devq);
1349 			iscsi_session_terminate(is);
1350 			return (ENOMEM);
1351 		}
1352 
1353 		error = xpt_bus_register(is->is_sim, NULL, 0);
1354 		if (error != 0) {
1355 			ISCSI_SESSION_UNLOCK(is);
1356 			ISCSI_SESSION_WARN(is, "failed to register bus");
1357 			iscsi_session_terminate(is);
1358 			return (ENOMEM);
1359 		}
1360 
1361 		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1362 		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1363 		    CAM_LUN_WILDCARD);
1364 		if (error != CAM_REQ_CMP) {
1365 			ISCSI_SESSION_UNLOCK(is);
1366 			ISCSI_SESSION_WARN(is, "failed to create path");
1367 			iscsi_session_terminate(is);
1368 			return (ENOMEM);
1369 		}
1370 		ISCSI_SESSION_UNLOCK(is);
1371 	}
1372 
1373 	return (0);
1374 }
1375 
1376 static int
1377 iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1378     struct iscsi_daemon_fail *fail)
1379 {
1380 	struct iscsi_session *is;
1381 
1382 	sx_slock(&sc->sc_lock);
1383 
1384 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1385 		if (is->is_id == fail->idf_session_id)
1386 			break;
1387 	}
1388 	if (is == NULL) {
1389 		sx_sunlock(&sc->sc_lock);
1390 		return (ESRCH);
1391 	}
1392 	ISCSI_SESSION_LOCK(is);
1393 	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1394 	    fail->idf_reason);
1395 	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1396 	//is->is_waiting_for_iscsid = false;
1397 	//is->is_login_phase = true;
1398 	//iscsi_session_reconnect(is);
1399 	ISCSI_SESSION_UNLOCK(is);
1400 	sx_sunlock(&sc->sc_lock);
1401 
1402 	return (0);
1403 }
1404 
1405 #ifdef ICL_KERNEL_PROXY
1406 static int
1407 iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1408     struct iscsi_daemon_connect *idc)
1409 {
1410 	struct iscsi_session *is;
1411 	struct sockaddr *from_sa, *to_sa;
1412 	int error;
1413 
1414 	sx_slock(&sc->sc_lock);
1415 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1416 		if (is->is_id == idc->idc_session_id)
1417 			break;
1418 	}
1419 	if (is == NULL) {
1420 		sx_sunlock(&sc->sc_lock);
1421 		return (ESRCH);
1422 	}
1423 	sx_sunlock(&sc->sc_lock);
1424 
1425 	if (idc->idc_from_addrlen > 0) {
1426 		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1427 		if (error != 0) {
1428 			ISCSI_SESSION_WARN(is,
1429 			    "getsockaddr failed with error %d", error);
1430 			return (error);
1431 		}
1432 	} else {
1433 		from_sa = NULL;
1434 	}
1435 	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1436 	if (error != 0) {
1437 		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1438 		    error);
1439 		free(from_sa, M_SONAME);
1440 		return (error);
1441 	}
1442 
1443 	ISCSI_SESSION_LOCK(is);
1444 	is->is_waiting_for_iscsid = false;
1445 	is->is_login_phase = true;
1446 	is->is_timeout = 0;
1447 	ISCSI_SESSION_UNLOCK(is);
1448 
1449 	error = icl_conn_connect(is->is_conn, idc->idc_iser, idc->idc_domain,
1450 	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1451 	free(from_sa, M_SONAME);
1452 	free(to_sa, M_SONAME);
1453 
1454 	/*
1455 	 * Digests are always disabled during login phase.
1456 	 */
1457 	is->is_conn->ic_header_crc32c = false;
1458 	is->is_conn->ic_data_crc32c = false;
1459 
1460 	return (error);
1461 }
1462 
1463 static int
1464 iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1465     struct iscsi_daemon_send *ids)
1466 {
1467 	struct iscsi_session *is;
1468 	struct icl_pdu *ip;
1469 	size_t datalen;
1470 	void *data;
1471 	int error;
1472 
1473 	sx_slock(&sc->sc_lock);
1474 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1475 		if (is->is_id == ids->ids_session_id)
1476 			break;
1477 	}
1478 	if (is == NULL) {
1479 		sx_sunlock(&sc->sc_lock);
1480 		return (ESRCH);
1481 	}
1482 	sx_sunlock(&sc->sc_lock);
1483 
1484 	if (is->is_login_phase == false)
1485 		return (EBUSY);
1486 
1487 	if (is->is_terminating || is->is_reconnecting)
1488 		return (EIO);
1489 
1490 	datalen = ids->ids_data_segment_len;
1491 	if (datalen > ISCSI_MAX_DATA_SEGMENT_LENGTH)
1492 		return (EINVAL);
1493 	if (datalen > 0) {
1494 		data = malloc(datalen, M_ISCSI, M_WAITOK);
1495 		error = copyin(ids->ids_data_segment, data, datalen);
1496 		if (error != 0) {
1497 			free(data, M_ISCSI);
1498 			return (error);
1499 		}
1500 	}
1501 
1502 	ip = icl_pdu_new_bhs(is->is_conn, M_WAITOK);
1503 	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1504 	if (datalen > 0) {
1505 		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1506 		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1507 		free(data, M_ISCSI);
1508 	}
1509 	icl_pdu_queue(ip);
1510 
1511 	return (0);
1512 }
1513 
1514 static int
1515 iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1516     struct iscsi_daemon_receive *idr)
1517 {
1518 	struct iscsi_session *is;
1519 	struct icl_pdu *ip;
1520 	void *data;
1521 
1522 	sx_slock(&sc->sc_lock);
1523 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1524 		if (is->is_id == idr->idr_session_id)
1525 			break;
1526 	}
1527 	if (is == NULL) {
1528 		sx_sunlock(&sc->sc_lock);
1529 		return (ESRCH);
1530 	}
1531 	sx_sunlock(&sc->sc_lock);
1532 
1533 	if (is->is_login_phase == false)
1534 		return (EBUSY);
1535 
1536 	ISCSI_SESSION_LOCK(is);
1537 	while (is->is_login_pdu == NULL &&
1538 	    is->is_terminating == false &&
1539 	    is->is_reconnecting == false)
1540 		cv_wait(&is->is_login_cv, &is->is_lock);
1541 	if (is->is_terminating || is->is_reconnecting) {
1542 		ISCSI_SESSION_UNLOCK(is);
1543 		return (EIO);
1544 	}
1545 	ip = is->is_login_pdu;
1546 	is->is_login_pdu = NULL;
1547 	ISCSI_SESSION_UNLOCK(is);
1548 
1549 	if (ip->ip_data_len > idr->idr_data_segment_len) {
1550 		icl_pdu_free(ip);
1551 		return (EMSGSIZE);
1552 	}
1553 
1554 	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1555 	if (ip->ip_data_len > 0) {
1556 		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1557 		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1558 		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1559 		free(data, M_ISCSI);
1560 	}
1561 
1562 	icl_pdu_free(ip);
1563 
1564 	return (0);
1565 }
1566 #endif /* ICL_KERNEL_PROXY */
1567 
1568 static void
1569 iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1570 {
1571 	/*
1572 	 * Just make sure all the fields are null-terminated.
1573 	 *
1574 	 * XXX: This is not particularly secure.  We should
1575 	 * 	create our own conf and then copy in relevant
1576 	 * 	fields.
1577 	 */
1578 	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1579 	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1580 	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1581 	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1582 	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1583 	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1584 	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1585 	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1586 	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1587 }
1588 
1589 static bool
1590 iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1591 {
1592 
1593 	if (isc->isc_initiator[0] == '\0') {
1594 		ISCSI_DEBUG("empty isc_initiator");
1595 		return (false);
1596 	}
1597 
1598 	if (isc->isc_target_addr[0] == '\0') {
1599 		ISCSI_DEBUG("empty isc_target_addr");
1600 		return (false);
1601 	}
1602 
1603 	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1604 		ISCSI_DEBUG("non-empty isc_target for discovery session");
1605 		return (false);
1606 	}
1607 
1608 	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1609 		ISCSI_DEBUG("empty isc_target for non-discovery session");
1610 		return (false);
1611 	}
1612 
1613 	return (true);
1614 }
1615 
1616 static int
1617 iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1618 {
1619 	struct iscsi_session *is;
1620 	const struct iscsi_session *is2;
1621 	int error;
1622 
1623 	iscsi_sanitize_session_conf(&isa->isa_conf);
1624 	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1625 		return (EINVAL);
1626 
1627 	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1628 	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1629 
1630 	sx_xlock(&sc->sc_lock);
1631 
1632 	/*
1633 	 * Prevent duplicates.
1634 	 */
1635 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1636 		if (!!is->is_conf.isc_discovery !=
1637 		    !!is2->is_conf.isc_discovery)
1638 			continue;
1639 
1640 		if (strcmp(is->is_conf.isc_target_addr,
1641 		    is2->is_conf.isc_target_addr) != 0)
1642 			continue;
1643 
1644 		if (is->is_conf.isc_discovery == 0 &&
1645 		    strcmp(is->is_conf.isc_target,
1646 		    is2->is_conf.isc_target) != 0)
1647 			continue;
1648 
1649 		sx_xunlock(&sc->sc_lock);
1650 		free(is, M_ISCSI);
1651 		return (EBUSY);
1652 	}
1653 
1654 	is->is_conn = icl_conn_new("iscsi", &is->is_lock);
1655 	is->is_conn->ic_receive = iscsi_receive_callback;
1656 	is->is_conn->ic_error = iscsi_error_callback;
1657 	is->is_conn->ic_prv0 = is;
1658 	TAILQ_INIT(&is->is_outstanding);
1659 	STAILQ_INIT(&is->is_postponed);
1660 	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1661 	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1662 #ifdef ICL_KERNEL_PROXY
1663 	cv_init(&is->is_login_cv, "iscsi_login");
1664 #endif
1665 
1666 	is->is_softc = sc;
1667 	sc->sc_last_session_id++;
1668 	is->is_id = sc->sc_last_session_id;
1669 	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1670 	arc4rand(&is->is_isid[1], 5, 0);
1671 	is->is_tsih = 0;
1672 	callout_init(&is->is_callout, 1);
1673 	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1674 	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1675 
1676 	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1677 	if (error != 0) {
1678 		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1679 		return (error);
1680 	}
1681 
1682 	/*
1683 	 * Trigger immediate reconnection.
1684 	 */
1685 	ISCSI_SESSION_LOCK(is);
1686 	is->is_waiting_for_iscsid = true;
1687 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1688 	ISCSI_SESSION_UNLOCK(is);
1689 	cv_signal(&sc->sc_cv);
1690 
1691 	sx_xunlock(&sc->sc_lock);
1692 
1693 	return (0);
1694 }
1695 
1696 static bool
1697 iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1698     unsigned int id2, const struct iscsi_session_conf *c2)
1699 {
1700 	if (id2 == 0 && c2->isc_target[0] == '\0' &&
1701 	    c2->isc_target_addr[0] == '\0')
1702 		return (true);
1703 	if (id2 != 0 && id2 == id1)
1704 		return (true);
1705 	if (c2->isc_target[0] != '\0' &&
1706 	    strcmp(c1->isc_target, c2->isc_target) == 0)
1707 		return (true);
1708 	if (c2->isc_target_addr[0] != '\0' &&
1709 	    strcmp(c1->isc_target_addr, c2->isc_target_addr) == 0)
1710 		return (true);
1711 	return (false);
1712 }
1713 
1714 static int
1715 iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1716     struct iscsi_session_remove *isr)
1717 {
1718 	struct iscsi_session *is, *tmp;
1719 	bool found = false;
1720 
1721 	iscsi_sanitize_session_conf(&isr->isr_conf);
1722 
1723 	sx_xlock(&sc->sc_lock);
1724 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1725 		ISCSI_SESSION_LOCK(is);
1726 		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1727 		    isr->isr_session_id, &isr->isr_conf)) {
1728 			found = true;
1729 			iscsi_session_logout(is);
1730 			iscsi_session_terminate(is);
1731 		}
1732 		ISCSI_SESSION_UNLOCK(is);
1733 	}
1734 	sx_xunlock(&sc->sc_lock);
1735 
1736 	if (!found)
1737 		return (ESRCH);
1738 
1739 	return (0);
1740 }
1741 
1742 static int
1743 iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1744 {
1745 	int error;
1746 	unsigned int i = 0;
1747 	struct iscsi_session *is;
1748 	struct iscsi_session_state iss;
1749 
1750 	sx_slock(&sc->sc_lock);
1751 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1752 		if (i >= isl->isl_nentries) {
1753 			sx_sunlock(&sc->sc_lock);
1754 			return (EMSGSIZE);
1755 		}
1756 		memset(&iss, 0, sizeof(iss));
1757 		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1758 		iss.iss_id = is->is_id;
1759 		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1760 		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1761 
1762 		if (is->is_conn->ic_header_crc32c)
1763 			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1764 		else
1765 			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1766 
1767 		if (is->is_conn->ic_data_crc32c)
1768 			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1769 		else
1770 			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1771 
1772 		iss.iss_max_data_segment_length = is->is_max_data_segment_length;
1773 		iss.iss_immediate_data = is->is_immediate_data;
1774 		iss.iss_connected = is->is_connected;
1775 
1776 		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1777 		if (error != 0) {
1778 			sx_sunlock(&sc->sc_lock);
1779 			return (error);
1780 		}
1781 		i++;
1782 	}
1783 	sx_sunlock(&sc->sc_lock);
1784 
1785 	isl->isl_nentries = i;
1786 
1787 	return (0);
1788 }
1789 
1790 static int
1791 iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1792     struct iscsi_session_modify *ism)
1793 {
1794 	struct iscsi_session *is;
1795 
1796 	iscsi_sanitize_session_conf(&ism->ism_conf);
1797 	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1798 		return (EINVAL);
1799 
1800 	sx_xlock(&sc->sc_lock);
1801 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1802 		ISCSI_SESSION_LOCK(is);
1803 		if (is->is_id == ism->ism_session_id)
1804 			break;
1805 		ISCSI_SESSION_UNLOCK(is);
1806 	}
1807 	if (is == NULL) {
1808 		sx_xunlock(&sc->sc_lock);
1809 		return (ESRCH);
1810 	}
1811 	sx_xunlock(&sc->sc_lock);
1812 
1813 	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
1814 	ISCSI_SESSION_UNLOCK(is);
1815 
1816 	iscsi_session_reconnect(is);
1817 
1818 	return (0);
1819 }
1820 
1821 static int
1822 iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
1823     struct thread *td)
1824 {
1825 	struct iscsi_softc *sc;
1826 
1827 	sc = dev->si_drv1;
1828 
1829 	switch (cmd) {
1830 	case ISCSIDWAIT:
1831 		return (iscsi_ioctl_daemon_wait(sc,
1832 		    (struct iscsi_daemon_request *)arg));
1833 	case ISCSIDHANDOFF:
1834 		return (iscsi_ioctl_daemon_handoff(sc,
1835 		    (struct iscsi_daemon_handoff *)arg));
1836 	case ISCSIDFAIL:
1837 		return (iscsi_ioctl_daemon_fail(sc,
1838 		    (struct iscsi_daemon_fail *)arg));
1839 #ifdef ICL_KERNEL_PROXY
1840 	case ISCSIDCONNECT:
1841 		return (iscsi_ioctl_daemon_connect(sc,
1842 		    (struct iscsi_daemon_connect *)arg));
1843 	case ISCSIDSEND:
1844 		return (iscsi_ioctl_daemon_send(sc,
1845 		    (struct iscsi_daemon_send *)arg));
1846 	case ISCSIDRECEIVE:
1847 		return (iscsi_ioctl_daemon_receive(sc,
1848 		    (struct iscsi_daemon_receive *)arg));
1849 #endif /* ICL_KERNEL_PROXY */
1850 	case ISCSISADD:
1851 		return (iscsi_ioctl_session_add(sc,
1852 		    (struct iscsi_session_add *)arg));
1853 	case ISCSISREMOVE:
1854 		return (iscsi_ioctl_session_remove(sc,
1855 		    (struct iscsi_session_remove *)arg));
1856 	case ISCSISLIST:
1857 		return (iscsi_ioctl_session_list(sc,
1858 		    (struct iscsi_session_list *)arg));
1859 	case ISCSISMODIFY:
1860 		return (iscsi_ioctl_session_modify(sc,
1861 		    (struct iscsi_session_modify *)arg));
1862 	default:
1863 		return (EINVAL);
1864 	}
1865 }
1866 
1867 static struct iscsi_outstanding *
1868 iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
1869 {
1870 	struct iscsi_outstanding *io;
1871 
1872 	ISCSI_SESSION_LOCK_ASSERT(is);
1873 
1874 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1875 		if (io->io_initiator_task_tag == initiator_task_tag)
1876 			return (io);
1877 	}
1878 	return (NULL);
1879 }
1880 
1881 static int
1882 iscsi_outstanding_add(struct iscsi_session *is,
1883     uint32_t initiator_task_tag, union ccb *ccb)
1884 {
1885 	struct iscsi_outstanding *io;
1886 
1887 	ISCSI_SESSION_LOCK_ASSERT(is);
1888 
1889 	KASSERT(iscsi_outstanding_find(is, initiator_task_tag) == NULL,
1890 	    ("initiator_task_tag 0x%x already added", initiator_task_tag));
1891 
1892 	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
1893 	if (io == NULL) {
1894 		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes", sizeof(*io));
1895 		return (ENOMEM);
1896 	}
1897 	io->io_initiator_task_tag = initiator_task_tag;
1898 	io->io_ccb = ccb;
1899 	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
1900 	return (0);
1901 }
1902 
1903 static void
1904 iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
1905 {
1906 
1907 	ISCSI_SESSION_LOCK_ASSERT(is);
1908 
1909 	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
1910 	uma_zfree(iscsi_outstanding_zone, io);
1911 }
1912 
1913 static void
1914 iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
1915 {
1916 	struct icl_pdu *request;
1917 	struct iscsi_bhs_scsi_command *bhssc;
1918 	struct ccb_scsiio *csio;
1919 	size_t len;
1920 	int error;
1921 
1922 	ISCSI_SESSION_LOCK_ASSERT(is);
1923 
1924 #if 0
1925 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
1926 #else
1927 	if (is->is_login_phase) {
1928 		ISCSI_SESSION_DEBUG(is, "called during login phase");
1929 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1930 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1931 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1932 		}
1933 		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
1934 		xpt_done(ccb);
1935 		return;
1936 	}
1937 #endif
1938 
1939 	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
1940 	if (request == NULL) {
1941 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1942 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1943 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1944 		}
1945 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
1946 		xpt_done(ccb);
1947 		return;
1948 	}
1949 
1950 	csio = &ccb->csio;
1951 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
1952 	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
1953 	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
1954 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
1955 	case CAM_DIR_IN:
1956 		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
1957 		break;
1958 	case CAM_DIR_OUT:
1959 		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
1960 		break;
1961 	}
1962 
1963 	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
1964 		switch (csio->tag_action) {
1965 		case MSG_HEAD_OF_Q_TAG:
1966 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
1967 			break;
1968 		case MSG_ORDERED_Q_TAG:
1969 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
1970 			break;
1971 		case MSG_ACA_TASK:
1972 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
1973 			break;
1974 		case MSG_SIMPLE_Q_TAG:
1975 		default:
1976 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
1977 			break;
1978 		}
1979 	} else
1980 		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
1981 
1982 	bhssc->bhssc_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
1983 	bhssc->bhssc_initiator_task_tag = is->is_initiator_task_tag;
1984 	is->is_initiator_task_tag++;
1985 	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
1986 	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
1987 	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
1988 
1989 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
1990 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
1991 	else
1992 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
1993 
1994 	error = iscsi_outstanding_add(is, bhssc->bhssc_initiator_task_tag, ccb);
1995 	if (error != 0) {
1996 		icl_pdu_free(request);
1997 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1998 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1999 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2000 		}
2001 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2002 		xpt_done(ccb);
2003 		return;
2004 	}
2005 
2006 	if (is->is_immediate_data &&
2007 	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2008 		len = csio->dxfer_len;
2009 		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2010 		if (len > is->is_first_burst_length) {
2011 			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_first_burst_length);
2012 			len = is->is_first_burst_length;
2013 		}
2014 
2015 		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2016 		if (error != 0) {
2017 			icl_pdu_free(request);
2018 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2019 				xpt_freeze_devq(ccb->ccb_h.path, 1);
2020 				ISCSI_SESSION_DEBUG(is, "freezing devq");
2021 			}
2022 			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2023 			xpt_done(ccb);
2024 			return;
2025 		}
2026 	}
2027 	iscsi_pdu_queue_locked(request);
2028 }
2029 
2030 static void
2031 iscsi_action(struct cam_sim *sim, union ccb *ccb)
2032 {
2033 	struct iscsi_session *is;
2034 
2035 	is = cam_sim_softc(sim);
2036 
2037 	ISCSI_SESSION_LOCK_ASSERT(is);
2038 
2039 	if (is->is_terminating ||
2040 	    (is->is_connected == false && fail_on_disconnection)) {
2041 		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2042 		xpt_done(ccb);
2043 		return;
2044 	}
2045 
2046 	switch (ccb->ccb_h.func_code) {
2047 	case XPT_PATH_INQ:
2048 	{
2049 		struct ccb_pathinq *cpi = &ccb->cpi;
2050 
2051 		cpi->version_num = 1;
2052 		cpi->hba_inquiry = PI_TAG_ABLE;
2053 		cpi->target_sprt = 0;
2054 		cpi->hba_misc = PIM_EXTLUNS;
2055 		cpi->hba_eng_cnt = 0;
2056 		cpi->max_target = 0;
2057 		cpi->max_lun = 0;
2058 		cpi->initiator_id = ~0;
2059 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2060 		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2061 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2062 		cpi->unit_number = cam_sim_unit(sim);
2063 		cpi->bus_id = cam_sim_bus(sim);
2064 		cpi->base_transfer_speed = 150000; /* XXX */
2065 		cpi->transport = XPORT_ISCSI;
2066 		cpi->transport_version = 0;
2067 		cpi->protocol = PROTO_SCSI;
2068 		cpi->protocol_version = SCSI_REV_SPC3;
2069 		cpi->maxio = MAXPHYS;
2070 		cpi->ccb_h.status = CAM_REQ_CMP;
2071 		break;
2072 	}
2073 	case XPT_GET_TRAN_SETTINGS:
2074 	{
2075 		struct ccb_trans_settings	*cts;
2076 		struct ccb_trans_settings_scsi	*scsi;
2077 
2078 		cts = &ccb->cts;
2079 		scsi = &cts->proto_specific.scsi;
2080 
2081 		cts->protocol = PROTO_SCSI;
2082 		cts->protocol_version = SCSI_REV_SPC3;
2083 		cts->transport = XPORT_ISCSI;
2084 		cts->transport_version = 0;
2085 		scsi->valid = CTS_SCSI_VALID_TQ;
2086 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2087 		cts->ccb_h.status = CAM_REQ_CMP;
2088 		break;
2089 	}
2090 	case XPT_CALC_GEOMETRY:
2091 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2092 		ccb->ccb_h.status = CAM_REQ_CMP;
2093 		break;
2094 #if 0
2095 	/*
2096 	 * XXX: What's the point?
2097 	 */
2098 	case XPT_RESET_BUS:
2099 	case XPT_ABORT:
2100 	case XPT_TERM_IO:
2101 		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2102 		ccb->ccb_h.status = CAM_REQ_CMP;
2103 		break;
2104 #endif
2105 	case XPT_SCSI_IO:
2106 		iscsi_action_scsiio(is, ccb);
2107 		return;
2108 	default:
2109 #if 0
2110 		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2111 #endif
2112 		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2113 		break;
2114 	}
2115 	xpt_done(ccb);
2116 }
2117 
2118 static void
2119 iscsi_poll(struct cam_sim *sim)
2120 {
2121 
2122 	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2123 }
2124 
2125 static void
2126 iscsi_shutdown(struct iscsi_softc *sc)
2127 {
2128 	struct iscsi_session *is;
2129 
2130 	ISCSI_DEBUG("removing all sessions due to shutdown");
2131 
2132 	sx_slock(&sc->sc_lock);
2133 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2134 		iscsi_session_terminate(is);
2135 	sx_sunlock(&sc->sc_lock);
2136 }
2137 
2138 static int
2139 iscsi_load(void)
2140 {
2141 	int error;
2142 
2143 	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2144 	sx_init(&sc->sc_lock, "iscsi");
2145 	TAILQ_INIT(&sc->sc_sessions);
2146 	cv_init(&sc->sc_cv, "iscsi_cv");
2147 
2148 	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2149 	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2150 	    UMA_ALIGN_PTR, 0);
2151 
2152 	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2153 	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2154 	if (error != 0) {
2155 		ISCSI_WARN("failed to create device node, error %d", error);
2156 		return (error);
2157 	}
2158 	sc->sc_cdev->si_drv1 = sc;
2159 
2160 	/*
2161 	 * Note that this needs to get run before dashutdown().  Otherwise,
2162 	 * when rebooting with iSCSI session with outstanding requests,
2163 	 * but disconnected, dashutdown() will hang on cam_periph_runccb().
2164 	 */
2165 	sc->sc_shutdown_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2166 	    iscsi_shutdown, sc, SHUTDOWN_PRI_FIRST);
2167 
2168 	return (0);
2169 }
2170 
2171 static int
2172 iscsi_unload(void)
2173 {
2174 	struct iscsi_session *is, *tmp;
2175 
2176 	if (sc->sc_cdev != NULL) {
2177 		ISCSI_DEBUG("removing device node");
2178 		destroy_dev(sc->sc_cdev);
2179 		ISCSI_DEBUG("device node removed");
2180 	}
2181 
2182 	if (sc->sc_shutdown_eh != NULL)
2183 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_eh);
2184 
2185 	sx_slock(&sc->sc_lock);
2186 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp)
2187 		iscsi_session_terminate(is);
2188 	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2189 		ISCSI_DEBUG("waiting for sessions to terminate");
2190 		cv_wait(&sc->sc_cv, &sc->sc_lock);
2191 	}
2192 	ISCSI_DEBUG("all sessions terminated");
2193 	sx_sunlock(&sc->sc_lock);
2194 
2195 	uma_zdestroy(iscsi_outstanding_zone);
2196 	sx_destroy(&sc->sc_lock);
2197 	cv_destroy(&sc->sc_cv);
2198 	free(sc, M_ISCSI);
2199 	return (0);
2200 }
2201 
2202 static int
2203 iscsi_quiesce(void)
2204 {
2205 	sx_slock(&sc->sc_lock);
2206 	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2207 		sx_sunlock(&sc->sc_lock);
2208 		return (EBUSY);
2209 	}
2210 	sx_sunlock(&sc->sc_lock);
2211 	return (0);
2212 }
2213 
2214 static int
2215 iscsi_modevent(module_t mod, int what, void *arg)
2216 {
2217 	int error;
2218 
2219 	switch (what) {
2220 	case MOD_LOAD:
2221 		error = iscsi_load();
2222 		break;
2223 	case MOD_UNLOAD:
2224 		error = iscsi_unload();
2225 		break;
2226 	case MOD_QUIESCE:
2227 		error = iscsi_quiesce();
2228 		break;
2229 	default:
2230 		error = EINVAL;
2231 		break;
2232 	}
2233 	return (error);
2234 }
2235 
2236 moduledata_t iscsi_data = {
2237 	"iscsi",
2238 	iscsi_modevent,
2239 	0
2240 };
2241 
2242 DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2243 MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2244 MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2245