xref: /freebsd/sys/dev/iscsi/iscsi.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2012 The FreeBSD Foundation
5  *
6  * This software was developed by Edward Tomasz Napierala under sponsorship
7  * from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  */
31 
32 #include <sys/param.h>
33 #include <sys/bio.h>
34 #include <sys/condvar.h>
35 #include <sys/conf.h>
36 #include <sys/endian.h>
37 #include <sys/eventhandler.h>
38 #include <sys/file.h>
39 #include <sys/kernel.h>
40 #include <sys/kthread.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.h>
45 #include <sys/module.h>
46 #include <sys/proc.h>
47 #include <sys/reboot.h>
48 #include <sys/socket.h>
49 #include <sys/sockopt.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/sx.h>
53 
54 #include <vm/uma.h>
55 
56 #include <cam/cam.h>
57 #include <cam/cam_ccb.h>
58 #include <cam/cam_xpt.h>
59 #include <cam/cam_debug.h>
60 #include <cam/cam_sim.h>
61 #include <cam/cam_xpt_sim.h>
62 #include <cam/cam_xpt_periph.h>
63 #include <cam/cam_periph.h>
64 #include <cam/scsi/scsi_all.h>
65 #include <cam/scsi/scsi_message.h>
66 
67 #include <dev/iscsi/icl.h>
68 #include <dev/iscsi/icl_wrappers.h>
69 #include <dev/iscsi/iscsi_ioctl.h>
70 #include <dev/iscsi/iscsi_proto.h>
71 #include <dev/iscsi/iscsi.h>
72 
73 #ifdef ICL_KERNEL_PROXY
74 #include <sys/socketvar.h>
75 #endif
76 
77 #ifdef ICL_KERNEL_PROXY
78 FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
79 #endif
80 
81 #ifdef COMPAT_FREEBSD13
82 struct iscsi_daemon_request13 {
83 	unsigned int			idr_session_id;
84 	struct iscsi_session_conf	idr_conf;
85 	uint8_t				idr_isid[6];
86 	uint16_t			idr_tsih;
87 	uint16_t			idr_spare_cid;
88 	struct iscsi_session_limits	idr_limits;
89 	int				idr_spare[4];
90 };
91 
92 #define	ISCSIDWAIT13	_IOR('I', 0x01, struct iscsi_daemon_request13)
93 #endif
94 
95 /*
96  * XXX: This is global so the iscsi_unload() can access it.
97  * 	Think about how to do this properly.
98  */
99 static struct iscsi_softc	*sc;
100 
101 SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
102     "iSCSI initiator");
103 static int debug = 1;
104 SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
105     &debug, 0, "Enable debug messages");
106 
107 static int ping_timeout = 5;
108 SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
109     0, "Timeout for ping (NOP-Out) requests, in seconds");
110 static int iscsid_timeout = 60;
111 SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
112     0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
113 static int login_timeout = 60;
114 SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
115     0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
116 static int maxtags = 255;
117 SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
118     0, "Max number of IO requests queued");
119 static int fail_on_disconnection = 0;
120 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
121     &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
122 static int fail_on_shutdown = 1;
123 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_shutdown, CTLFLAG_RWTUN,
124     &fail_on_shutdown, 0, "Fail disconnected sessions on shutdown");
125 
126 static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
127 static uma_zone_t iscsi_outstanding_zone;
128 
129 #define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
130 #define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
131 
132 #define	ISCSI_DEBUG(X, ...)						\
133 	do {								\
134 		if (debug > 1) 						\
135 			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
136 	} while (0)
137 
138 #define	ISCSI_WARN(X, ...)						\
139 	do {								\
140 		if (debug > 0) {					\
141 			printf("WARNING: %s: " X "\n",			\
142 			    __func__, ## __VA_ARGS__);			\
143 		}							\
144 	} while (0)
145 
146 #define	ISCSI_SESSION_DEBUG(S, X, ...)					\
147 	do {								\
148 		if (debug > 1) {					\
149 			printf("%s: %s (%s): " X "\n",			\
150 			    __func__, S->is_conf.isc_target_addr,	\
151 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
152 		}							\
153 	} while (0)
154 
155 #define	ISCSI_SESSION_WARN(S, X, ...)					\
156 	do {								\
157 		if (debug > 0) {					\
158 			printf("WARNING: %s (%s): " X "\n",		\
159 			    S->is_conf.isc_target_addr,			\
160 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
161 		}							\
162 	} while (0)
163 
164 #define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
165 #define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
166 #define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
167 #define ISCSI_SESSION_LOCK_ASSERT_NOT(X) mtx_assert(&X->is_lock, MA_NOTOWNED)
168 
169 static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
170 		    int mode, struct thread *td);
171 
172 static struct cdevsw iscsi_cdevsw = {
173      .d_version = D_VERSION,
174      .d_ioctl   = iscsi_ioctl,
175      .d_name    = "iscsi",
176 };
177 
178 static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
179 static void	iscsi_pdu_queue(struct icl_pdu *request);
180 static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
181 static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
182 static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
183 static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
184 static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
185 static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
186 static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
187 static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
188 static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
189 static void	iscsi_session_reconnect(struct iscsi_session *is);
190 static void	iscsi_session_terminate(struct iscsi_session *is);
191 static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
192 static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
193 		    uint32_t initiator_task_tag);
194 static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
195 		    struct icl_pdu *request, union ccb *ccb,
196 		    uint32_t *initiator_task_tagp);
197 static void	iscsi_outstanding_remove(struct iscsi_session *is,
198 		    struct iscsi_outstanding *io);
199 
200 static bool
201 iscsi_pdu_prepare(struct icl_pdu *request)
202 {
203 	struct iscsi_session *is;
204 	struct iscsi_bhs_scsi_command *bhssc;
205 
206 	is = PDU_SESSION(request);
207 
208 	ISCSI_SESSION_LOCK_ASSERT(is);
209 
210 	/*
211 	 * We're only using fields common for all the request
212 	 * (initiator -> target) PDUs.
213 	 */
214 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
215 
216 	/*
217 	 * Data-Out PDU does not contain CmdSN.
218 	 */
219 	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
220 		if (ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn) &&
221 		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
222 			/*
223 			 * Current MaxCmdSN prevents us from sending any more
224 			 * SCSI Command PDUs to the target; postpone the PDU.
225 			 * It will get resent by either iscsi_pdu_queue(),
226 			 * or by maintenance thread.
227 			 */
228 #if 0
229 			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %u, "
230 			    "ExpCmdSN %u, MaxCmdSN %u, opcode 0x%x",
231 			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn,
232 			    bhssc->bhssc_opcode);
233 #endif
234 			return (true);
235 		}
236 		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
237 		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
238 			is->is_cmdsn++;
239 	}
240 	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
241 
242 	return (false);
243 }
244 
245 static void
246 iscsi_session_send_postponed(struct iscsi_session *is)
247 {
248 	struct icl_pdu *request;
249 	bool postpone;
250 
251 	ISCSI_SESSION_LOCK_ASSERT(is);
252 
253 	if (STAILQ_EMPTY(&is->is_postponed))
254 		return;
255 	while ((request = STAILQ_FIRST(&is->is_postponed)) != NULL) {
256 		postpone = iscsi_pdu_prepare(request);
257 		if (postpone)
258 			return;
259 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
260 		icl_pdu_queue(request);
261 	}
262 	xpt_release_simq(is->is_sim, 1);
263 }
264 
265 static void
266 iscsi_pdu_queue_locked(struct icl_pdu *request)
267 {
268 	struct iscsi_session *is;
269 	bool postpone;
270 
271 	is = PDU_SESSION(request);
272 	ISCSI_SESSION_LOCK_ASSERT(is);
273 	iscsi_session_send_postponed(is);
274 	postpone = iscsi_pdu_prepare(request);
275 	if (postpone) {
276 		if (STAILQ_EMPTY(&is->is_postponed))
277 			xpt_freeze_simq(is->is_sim, 1);
278 		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
279 		return;
280 	}
281 	icl_pdu_queue(request);
282 }
283 
284 static void
285 iscsi_pdu_queue(struct icl_pdu *request)
286 {
287 	struct iscsi_session *is;
288 
289 	is = PDU_SESSION(request);
290 	ISCSI_SESSION_LOCK(is);
291 	iscsi_pdu_queue_locked(request);
292 	ISCSI_SESSION_UNLOCK(is);
293 }
294 
295 static void
296 iscsi_session_logout(struct iscsi_session *is)
297 {
298 	struct icl_pdu *request;
299 	struct iscsi_bhs_logout_request *bhslr;
300 
301 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
302 	if (request == NULL)
303 		return;
304 
305 	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
306 	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
307 	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
308 	iscsi_pdu_queue_locked(request);
309 }
310 
311 static void
312 iscsi_session_terminate_task(struct iscsi_session *is,
313     struct iscsi_outstanding *io, cam_status status)
314 {
315 
316 	ISCSI_SESSION_LOCK_ASSERT(is);
317 
318 	if (io->io_ccb != NULL) {
319 		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
320 		io->io_ccb->ccb_h.status |= status;
321 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
322 			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
323 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
324 			ISCSI_SESSION_DEBUG(is, "freezing devq");
325 		}
326 		xpt_done(io->io_ccb);
327 	}
328 	iscsi_outstanding_remove(is, io);
329 }
330 
331 static void
332 iscsi_session_terminate_tasks(struct iscsi_session *is, cam_status status)
333 {
334 	struct iscsi_outstanding *io, *tmp;
335 
336 	ISCSI_SESSION_LOCK_ASSERT(is);
337 
338 	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
339 		iscsi_session_terminate_task(is, io, status);
340 	}
341 }
342 
343 static void
344 iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
345 {
346 	struct icl_pdu *pdu;
347 
348 	ISCSI_SESSION_LOCK_ASSERT(is);
349 
350 	/*
351 	 * Don't queue any new PDUs.
352 	 */
353 	if (is->is_sim != NULL && is->is_simq_frozen == false) {
354 		ISCSI_SESSION_DEBUG(is, "freezing");
355 		xpt_freeze_simq(is->is_sim, 1);
356 		is->is_simq_frozen = true;
357 	}
358 
359 	/*
360 	 * Remove postponed PDUs.
361 	 */
362 	if (!STAILQ_EMPTY(&is->is_postponed))
363 		xpt_release_simq(is->is_sim, 1);
364 	while ((pdu = STAILQ_FIRST(&is->is_postponed)) != NULL) {
365 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
366 		icl_pdu_free(pdu);
367 	}
368 
369 	if (destroy_sim == false) {
370 		/*
371 		 * Terminate SCSI tasks, asking CAM to requeue them.
372 		 */
373 		iscsi_session_terminate_tasks(is, CAM_REQUEUE_REQ);
374 		return;
375 	}
376 
377 	iscsi_session_terminate_tasks(is, CAM_DEV_NOT_THERE);
378 
379 	if (is->is_sim == NULL)
380 		return;
381 
382 	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
383 	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
384 
385 	if (is->is_simq_frozen) {
386 		is->is_simq_frozen = false;
387 		xpt_release_simq(is->is_sim, 1);
388 	}
389 
390 	xpt_free_path(is->is_path);
391 	is->is_path = NULL;
392 	xpt_bus_deregister(cam_sim_path(is->is_sim));
393 	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
394 	is->is_sim = NULL;
395 	is->is_devq = NULL;
396 }
397 
398 static void
399 iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
400 {
401 	/*
402 	 * As we will be reconnecting shortly,
403 	 * discard outstanding data immediately on
404 	 * close(), also notify peer via RST if
405 	 * any packets come in.
406 	 */
407 	struct socket *so;
408 	so = is->is_conn->ic_socket;
409 	if (so != NULL) {
410 		struct sockopt sopt;
411 		struct linger sl;
412 		sopt.sopt_dir     = SOPT_SET;
413 		sopt.sopt_level   = SOL_SOCKET;
414 		sopt.sopt_name    = SO_LINGER;
415 		sopt.sopt_val     = &sl;
416 		sopt.sopt_valsize = sizeof(sl);
417 		sl.l_onoff        = 1;	/* non-zero value enables linger option in kernel */
418 		sl.l_linger       = 0;	/* timeout interval in seconds */
419 		sosetopt(is->is_conn->ic_socket, &sopt);
420 	}
421 
422 	icl_conn_close(is->is_conn);
423 
424 	ISCSI_SESSION_LOCK(is);
425 
426 	is->is_connected = false;
427 	is->is_reconnecting = false;
428 	is->is_login_phase = false;
429 
430 #ifdef ICL_KERNEL_PROXY
431 	if (is->is_login_pdu != NULL) {
432 		icl_pdu_free(is->is_login_pdu);
433 		is->is_login_pdu = NULL;
434 	}
435 	cv_signal(&is->is_login_cv);
436 #endif
437 
438 	if (fail_on_disconnection) {
439 		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
440 		iscsi_session_cleanup(is, true);
441 	} else {
442 		iscsi_session_cleanup(is, false);
443 	}
444 
445 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
446 	    ("destroying session with active tasks"));
447 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
448 	    ("destroying session with postponed PDUs"));
449 
450 	if (is->is_conf.isc_enable == 0 && is->is_conf.isc_discovery == 0) {
451 		ISCSI_SESSION_UNLOCK(is);
452 		return;
453 	}
454 
455 	/*
456 	 * Request immediate reconnection from iscsid(8).
457 	 */
458 	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
459 	is->is_waiting_for_iscsid = true;
460 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
461 	is->is_timeout = 0;
462 	ISCSI_SESSION_UNLOCK(is);
463 	cv_signal(&is->is_softc->sc_cv);
464 }
465 
466 static void
467 iscsi_maintenance_thread_terminate(struct iscsi_session *is)
468 {
469 	struct iscsi_softc *sc;
470 
471 	sc = is->is_softc;
472 	sx_xlock(&sc->sc_lock);
473 	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
474 	sx_xunlock(&sc->sc_lock);
475 
476 	icl_conn_close(is->is_conn);
477 	callout_drain(&is->is_callout);
478 
479 	ISCSI_SESSION_LOCK(is);
480 
481 	KASSERT(is->is_terminating, ("is_terminating == false"));
482 
483 #ifdef ICL_KERNEL_PROXY
484 	if (is->is_login_pdu != NULL) {
485 		icl_pdu_free(is->is_login_pdu);
486 		is->is_login_pdu = NULL;
487 	}
488 	cv_signal(&is->is_login_cv);
489 #endif
490 
491 	iscsi_session_cleanup(is, true);
492 
493 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
494 	    ("destroying session with active tasks"));
495 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
496 	    ("destroying session with postponed PDUs"));
497 
498 	ISCSI_SESSION_UNLOCK(is);
499 
500 	icl_conn_free(is->is_conn);
501 	mtx_destroy(&is->is_lock);
502 	cv_destroy(&is->is_maintenance_cv);
503 #ifdef ICL_KERNEL_PROXY
504 	cv_destroy(&is->is_login_cv);
505 #endif
506 
507 	ISCSI_SESSION_DEBUG(is, "terminated");
508 	free(is, M_ISCSI);
509 
510 	/*
511 	 * The iscsi_unload() routine might be waiting.
512 	 */
513 	cv_signal(&sc->sc_cv);
514 }
515 
516 static void
517 iscsi_maintenance_thread(void *arg)
518 {
519 	struct iscsi_session *is = arg;
520 
521 	ISCSI_SESSION_LOCK(is);
522 	for (;;) {
523 		if (is->is_reconnecting == false &&
524 		    is->is_terminating == false &&
525 		    (STAILQ_EMPTY(&is->is_postponed) ||
526 		     ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn)))
527 			cv_wait(&is->is_maintenance_cv, &is->is_lock);
528 
529 		/* Terminate supersedes reconnect. */
530 		if (is->is_terminating) {
531 			ISCSI_SESSION_UNLOCK(is);
532 			iscsi_maintenance_thread_terminate(is);
533 			kthread_exit();
534 			return;
535 		}
536 
537 		if (is->is_reconnecting) {
538 			ISCSI_SESSION_UNLOCK(is);
539 			iscsi_maintenance_thread_reconnect(is);
540 			ISCSI_SESSION_LOCK(is);
541 			continue;
542 		}
543 
544 		iscsi_session_send_postponed(is);
545 	}
546 	ISCSI_SESSION_UNLOCK(is);
547 }
548 
549 static void
550 iscsi_session_reconnect(struct iscsi_session *is)
551 {
552 
553 	/*
554 	 * XXX: We can't use locking here, because
555 	 * 	it's being called from various contexts.
556 	 * 	Hope it doesn't break anything.
557 	 */
558 	if (is->is_reconnecting)
559 		return;
560 
561 	is->is_reconnecting = true;
562 	cv_signal(&is->is_maintenance_cv);
563 }
564 
565 static void
566 iscsi_session_terminate(struct iscsi_session *is)
567 {
568 
569 	if (is->is_terminating)
570 		return;
571 
572 	is->is_terminating = true;
573 
574 #if 0
575 	iscsi_session_logout(is);
576 #endif
577 	cv_signal(&is->is_maintenance_cv);
578 }
579 
580 static void
581 iscsi_callout(void *context)
582 {
583 	struct icl_pdu *request;
584 	struct iscsi_bhs_nop_out *bhsno;
585 	struct iscsi_session *is;
586 	bool reconnect_needed = false;
587 	sbintime_t sbt, pr;
588 
589 	is = context;
590 
591 	ISCSI_SESSION_LOCK(is);
592 	if (is->is_terminating) {
593 		ISCSI_SESSION_UNLOCK(is);
594 		return;
595 	}
596 
597 	sbt = mstosbt(995);
598 	pr  = mstosbt(10);
599 	callout_schedule_sbt(&is->is_callout, sbt, pr, 0);
600 
601 	if (is->is_conf.isc_enable == 0)
602 		goto out;
603 
604 	is->is_timeout++;
605 
606 	if (is->is_waiting_for_iscsid) {
607 		if (iscsid_timeout > 0 && is->is_timeout > iscsid_timeout) {
608 			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
609 			    "for %d seconds; reconnecting",
610 			    is->is_timeout);
611 			reconnect_needed = true;
612 		}
613 		goto out;
614 	}
615 
616 	if (is->is_login_phase) {
617 		if (is->is_login_timeout > 0 && is->is_timeout > is->is_login_timeout) {
618 			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
619 			    "reconnecting", is->is_timeout);
620 			reconnect_needed = true;
621 		}
622 		goto out;
623 	}
624 
625 	if (is->is_ping_timeout <= 0) {
626 		/*
627 		 * Pings are disabled.  Don't send NOP-Out in this case.
628 		 * Reset the timeout, to avoid triggering reconnection,
629 		 * should the user decide to reenable them.
630 		 */
631 		is->is_timeout = 0;
632 		goto out;
633 	}
634 
635 	if (is->is_timeout >= is->is_ping_timeout) {
636 		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
637 		    "reconnecting", is->is_ping_timeout);
638 		reconnect_needed = true;
639 		goto out;
640 	}
641 
642 	ISCSI_SESSION_UNLOCK(is);
643 
644 	/*
645 	 * If the ping was reset less than one second ago - which means
646 	 * that we've received some PDU during the last second - assume
647 	 * the traffic flows correctly and don't bother sending a NOP-Out.
648 	 *
649 	 * (It's 2 - one for one second, and one for incrementing is_timeout
650 	 * earlier in this routine.)
651 	 */
652 	if (is->is_timeout < 2)
653 		return;
654 
655 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
656 	if (request == NULL) {
657 		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
658 		return;
659 	}
660 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
661 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
662 	    ISCSI_BHS_OPCODE_IMMEDIATE;
663 	bhsno->bhsno_flags = 0x80;
664 	bhsno->bhsno_target_transfer_tag = 0xffffffff;
665 	iscsi_pdu_queue(request);
666 	return;
667 
668 out:
669 	if (is->is_terminating) {
670 		ISCSI_SESSION_UNLOCK(is);
671 		return;
672 	}
673 
674 	ISCSI_SESSION_UNLOCK(is);
675 
676 	if (reconnect_needed)
677 		iscsi_session_reconnect(is);
678 }
679 
680 static void
681 iscsi_pdu_update_statsn(const struct icl_pdu *response)
682 {
683 	const struct iscsi_bhs_data_in *bhsdi;
684 	struct iscsi_session *is;
685 	uint32_t expcmdsn, maxcmdsn, statsn;
686 
687 	is = PDU_SESSION(response);
688 
689 	ISCSI_SESSION_LOCK_ASSERT(is);
690 
691 	/*
692 	 * We're only using fields common for all the response
693 	 * (target -> initiator) PDUs.
694 	 */
695 	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
696 	/*
697 	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
698 	 * and Residual Count only have meaningful content if the S bit
699 	 * is set to 1", so we also need to check the bit specific for
700 	 * Data-In PDU.
701 	 */
702 	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
703 	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
704 		statsn = ntohl(bhsdi->bhsdi_statsn);
705 		if (statsn != is->is_statsn && statsn != (is->is_statsn + 1)) {
706 			/* XXX: This is normal situation for MCS */
707 			ISCSI_SESSION_WARN(is, "PDU 0x%x StatSN %u != "
708 			    "session ExpStatSN %u (or + 1); reconnecting",
709 			    bhsdi->bhsdi_opcode, statsn, is->is_statsn);
710 			iscsi_session_reconnect(is);
711 		}
712 		if (ISCSI_SNGT(statsn, is->is_statsn))
713 			is->is_statsn = statsn;
714 	}
715 
716 	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
717 	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
718 
719 	if (ISCSI_SNLT(maxcmdsn + 1, expcmdsn)) {
720 		ISCSI_SESSION_DEBUG(is,
721 		    "PDU MaxCmdSN %u + 1 < PDU ExpCmdSN %u; ignoring",
722 		    maxcmdsn, expcmdsn);
723 	} else {
724 		if (ISCSI_SNGT(maxcmdsn, is->is_maxcmdsn)) {
725 			is->is_maxcmdsn = maxcmdsn;
726 
727 			/*
728 			 * Command window increased; kick the maintanance thread
729 			 * to send out postponed commands.
730 			 */
731 			if (!STAILQ_EMPTY(&is->is_postponed))
732 				cv_signal(&is->is_maintenance_cv);
733 		} else if (ISCSI_SNLT(maxcmdsn, is->is_maxcmdsn)) {
734 			/* XXX: This is normal situation for MCS */
735 			ISCSI_SESSION_DEBUG(is,
736 			    "PDU MaxCmdSN %u < session MaxCmdSN %u; ignoring",
737 			    maxcmdsn, is->is_maxcmdsn);
738 		}
739 
740 		if (ISCSI_SNGT(expcmdsn, is->is_expcmdsn)) {
741 			is->is_expcmdsn = expcmdsn;
742 		} else if (ISCSI_SNLT(expcmdsn, is->is_expcmdsn)) {
743 			/* XXX: This is normal situation for MCS */
744 			ISCSI_SESSION_DEBUG(is,
745 			    "PDU ExpCmdSN %u < session ExpCmdSN %u; ignoring",
746 			    expcmdsn, is->is_expcmdsn);
747 		}
748 	}
749 
750 	/*
751 	 * Every incoming PDU - not just NOP-In - resets the ping timer.
752 	 * The purpose of the timeout is to reset the connection when it stalls;
753 	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
754 	 * in some queue.
755 	 */
756 	is->is_timeout = 0;
757 }
758 
759 static void
760 iscsi_receive_callback(struct icl_pdu *response)
761 {
762 	struct iscsi_session *is;
763 
764 	is = PDU_SESSION(response);
765 
766 	ISCSI_SESSION_LOCK(is);
767 
768 	iscsi_pdu_update_statsn(response);
769 
770 #ifdef ICL_KERNEL_PROXY
771 	if (is->is_login_phase) {
772 		if (is->is_login_pdu == NULL)
773 			is->is_login_pdu = response;
774 		else
775 			icl_pdu_free(response);
776 		ISCSI_SESSION_UNLOCK(is);
777 		cv_signal(&is->is_login_cv);
778 		return;
779 	}
780 #endif
781 
782 	/*
783 	 * The handling routine is responsible for freeing the PDU
784 	 * when it's no longer needed.
785 	 */
786 	switch (response->ip_bhs->bhs_opcode) {
787 	case ISCSI_BHS_OPCODE_NOP_IN:
788 		iscsi_pdu_handle_nop_in(response);
789 		ISCSI_SESSION_UNLOCK(is);
790 		break;
791 	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
792 		iscsi_pdu_handle_scsi_response(response);
793 		/* Session lock dropped inside. */
794 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
795 		break;
796 	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
797 		iscsi_pdu_handle_task_response(response);
798 		ISCSI_SESSION_UNLOCK(is);
799 		break;
800 	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
801 		iscsi_pdu_handle_data_in(response);
802 		/* Session lock dropped inside. */
803 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
804 		break;
805 	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
806 		iscsi_pdu_handle_logout_response(response);
807 		ISCSI_SESSION_UNLOCK(is);
808 		break;
809 	case ISCSI_BHS_OPCODE_R2T:
810 		iscsi_pdu_handle_r2t(response);
811 		ISCSI_SESSION_UNLOCK(is);
812 		break;
813 	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
814 		iscsi_pdu_handle_async_message(response);
815 		ISCSI_SESSION_UNLOCK(is);
816 		break;
817 	case ISCSI_BHS_OPCODE_REJECT:
818 		iscsi_pdu_handle_reject(response);
819 		ISCSI_SESSION_UNLOCK(is);
820 		break;
821 	default:
822 		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
823 		    "opcode 0x%x; reconnecting",
824 		    response->ip_bhs->bhs_opcode);
825 		iscsi_session_reconnect(is);
826 		ISCSI_SESSION_UNLOCK(is);
827 		icl_pdu_free(response);
828 	}
829 }
830 
831 static void
832 iscsi_error_callback(struct icl_conn *ic)
833 {
834 	struct iscsi_session *is;
835 
836 	is = CONN_SESSION(ic);
837 
838 	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
839 	iscsi_session_reconnect(is);
840 }
841 
842 static void
843 iscsi_pdu_handle_nop_in(struct icl_pdu *response)
844 {
845 	struct iscsi_session *is;
846 	struct iscsi_bhs_nop_out *bhsno;
847 	struct iscsi_bhs_nop_in *bhsni;
848 	struct icl_pdu *request;
849 	void *data = NULL;
850 	size_t datasize;
851 	int error;
852 
853 	is = PDU_SESSION(response);
854 	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
855 
856 	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
857 		/*
858 		 * Nothing to do; iscsi_pdu_update_statsn() already
859 		 * zeroed the timeout.
860 		 */
861 		icl_pdu_free(response);
862 		return;
863 	}
864 
865 	datasize = icl_pdu_data_segment_length(response);
866 	if (datasize > 0) {
867 		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
868 		if (data == NULL) {
869 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
870 			    "reconnecting");
871 			icl_pdu_free(response);
872 			iscsi_session_reconnect(is);
873 			return;
874 		}
875 		icl_pdu_get_data(response, 0, data, datasize);
876 	}
877 
878 	request = icl_pdu_new(response->ip_conn, M_NOWAIT);
879 	if (request == NULL) {
880 		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
881 		    "reconnecting");
882 		free(data, M_ISCSI);
883 		icl_pdu_free(response);
884 		iscsi_session_reconnect(is);
885 		return;
886 	}
887 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
888 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
889 	    ISCSI_BHS_OPCODE_IMMEDIATE;
890 	bhsno->bhsno_flags = 0x80;
891 	bhsno->bhsno_initiator_task_tag = 0xffffffff;
892 	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
893 	if (datasize > 0) {
894 		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
895 		if (error != 0) {
896 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
897 			    "reconnecting");
898 			free(data, M_ISCSI);
899 			icl_pdu_free(request);
900 			icl_pdu_free(response);
901 			iscsi_session_reconnect(is);
902 			return;
903 		}
904 		free(data, M_ISCSI);
905 	}
906 
907 	icl_pdu_free(response);
908 	iscsi_pdu_queue_locked(request);
909 }
910 
911 static void
912 iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
913 {
914 	struct iscsi_bhs_scsi_response *bhssr;
915 	struct iscsi_outstanding *io;
916 	struct iscsi_session *is;
917 	union ccb *ccb;
918 	struct ccb_scsiio *csio;
919 	size_t data_segment_len, received;
920 	uint16_t sense_len;
921 	uint32_t resid;
922 
923 	is = PDU_SESSION(response);
924 
925 	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
926 	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
927 	if (io == NULL || io->io_ccb == NULL) {
928 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
929 		icl_pdu_free(response);
930 		iscsi_session_reconnect(is);
931 		ISCSI_SESSION_UNLOCK(is);
932 		return;
933 	}
934 
935 	ccb = io->io_ccb;
936 	if (bhssr->bhssr_response == BHSSR_RESPONSE_COMMAND_COMPLETED) {
937 		if (ntohl(bhssr->bhssr_expdatasn) != io->io_datasn) {
938 			ISCSI_SESSION_WARN(is,
939 			    "ExpDataSN mismatch in SCSI Response (%u vs %u)",
940 			    ntohl(bhssr->bhssr_expdatasn), io->io_datasn);
941 
942 			/*
943 			 * XXX: Permit an ExpDataSN of zero for errors.
944 			 *
945 			 * This doesn't conform to RFC 7143, but some
946 			 * targets seem to do this.
947 			 */
948 			if (bhssr->bhssr_status != 0 &&
949 			    bhssr->bhssr_expdatasn == htonl(0))
950 				goto skip_expdatasn;
951 
952 			icl_pdu_free(response);
953 			iscsi_session_reconnect(is);
954 			ISCSI_SESSION_UNLOCK(is);
955 			return;
956 		}
957 	} else {
958 		if (bhssr->bhssr_expdatasn != htonl(0)) {
959 			ISCSI_SESSION_WARN(is,
960 			    "ExpDataSN mismatch in SCSI Response (%u vs 0)",
961 			    ntohl(bhssr->bhssr_expdatasn));
962 			icl_pdu_free(response);
963 			iscsi_session_reconnect(is);
964 			ISCSI_SESSION_UNLOCK(is);
965 			return;
966 		}
967 	}
968 skip_expdatasn:
969 
970 	/*
971 	 * With iSER, after getting good response we can be sure
972 	 * that all the data has been successfully transferred.
973 	 */
974 	if (is->is_conn->ic_iser) {
975 		resid = ntohl(bhssr->bhssr_residual_count);
976 		if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW) {
977 			io->io_received = ccb->csio.dxfer_len - resid;
978 		} else if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_OVERFLOW) {
979 			ISCSI_SESSION_WARN(is, "overflow: target indicates %d", resid);
980 		} else {
981 			io->io_received = ccb->csio.dxfer_len;
982 		}
983 	}
984 
985 	received = io->io_received;
986 	iscsi_outstanding_remove(is, io);
987 	ISCSI_SESSION_UNLOCK(is);
988 
989 	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
990 		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
991  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
992  			xpt_freeze_devq(ccb->ccb_h.path, 1);
993 			ISCSI_SESSION_DEBUG(is, "freezing devq");
994 		}
995  		ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
996 	} else if (bhssr->bhssr_status == 0) {
997 		ccb->ccb_h.status = CAM_REQ_CMP;
998 	} else {
999  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1000  			xpt_freeze_devq(ccb->ccb_h.path, 1);
1001 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1002 		}
1003  		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1004 		ccb->csio.scsi_status = bhssr->bhssr_status;
1005 	}
1006 
1007 	csio = &ccb->csio;
1008 	data_segment_len = icl_pdu_data_segment_length(response);
1009 	if (data_segment_len > 0) {
1010 		if (data_segment_len < sizeof(sense_len)) {
1011 			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
1012 			    data_segment_len);
1013 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1014 				xpt_freeze_devq(ccb->ccb_h.path, 1);
1015 				ISCSI_SESSION_DEBUG(is, "freezing devq");
1016 			}
1017 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
1018 			goto out;
1019 		}
1020 		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
1021 		sense_len = ntohs(sense_len);
1022 #if 0
1023 		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
1024 		    sense_len, data_segment_len);
1025 #endif
1026 		if (sizeof(sense_len) + sense_len > data_segment_len) {
1027 			ISCSI_SESSION_WARN(is, "truncated data segment "
1028 			    "(%zd bytes, should be %zd)",
1029 			    data_segment_len, sizeof(sense_len) + sense_len);
1030 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1031 				xpt_freeze_devq(ccb->ccb_h.path, 1);
1032 				ISCSI_SESSION_DEBUG(is, "freezing devq");
1033 			}
1034 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
1035 			goto out;
1036 		} else if (sizeof(sense_len) + sense_len < data_segment_len)
1037 			ISCSI_SESSION_WARN(is, "oversize data segment "
1038 			    "(%zd bytes, should be %zd)",
1039 			    data_segment_len, sizeof(sense_len) + sense_len);
1040 		if (sense_len > csio->sense_len) {
1041 			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
1042 			    sense_len, csio->sense_len);
1043 			sense_len = csio->sense_len;
1044 		}
1045 		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
1046 		csio->sense_resid = csio->sense_len - sense_len;
1047 		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
1048 	}
1049 
1050 out:
1051 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
1052 		csio->resid = ntohl(bhssr->bhssr_residual_count);
1053 
1054 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1055 		KASSERT(received <= csio->dxfer_len,
1056 		    ("received > csio->dxfer_len"));
1057 		if (received < csio->dxfer_len) {
1058 			if (csio->resid != csio->dxfer_len - received) {
1059 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1060 				    "target indicates %d, we calculated %zd",
1061 				    csio->resid, csio->dxfer_len - received);
1062 			}
1063 			csio->resid = csio->dxfer_len - received;
1064 		}
1065 	}
1066 
1067 	xpt_done(ccb);
1068 	icl_pdu_free(response);
1069 }
1070 
1071 static void
1072 iscsi_pdu_handle_task_response(struct icl_pdu *response)
1073 {
1074 	struct iscsi_bhs_task_management_response *bhstmr;
1075 	struct iscsi_outstanding *io, *aio;
1076 	struct iscsi_session *is;
1077 
1078 	is = PDU_SESSION(response);
1079 
1080 	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
1081 	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
1082 	if (io == NULL || io->io_ccb != NULL) {
1083 		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
1084 		    bhstmr->bhstmr_initiator_task_tag);
1085 		icl_pdu_free(response);
1086 		iscsi_session_reconnect(is);
1087 		return;
1088 	}
1089 
1090 	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
1091 		ISCSI_SESSION_WARN(is, "task response 0x%x",
1092 		    bhstmr->bhstmr_response);
1093 	} else {
1094 		aio = iscsi_outstanding_find(is, io->io_referenced_task_tag);
1095 		if (aio != NULL && aio->io_ccb != NULL)
1096 			iscsi_session_terminate_task(is, aio, CAM_REQ_ABORTED);
1097 	}
1098 
1099 	iscsi_outstanding_remove(is, io);
1100 	icl_pdu_free(response);
1101 }
1102 
1103 static void
1104 iscsi_pdu_get_data_csio(struct icl_pdu *response, size_t pdu_offset,
1105     struct ccb_scsiio *csio, size_t oreceived, size_t data_segment_len)
1106 {
1107 	switch (csio->ccb_h.flags & CAM_DATA_MASK) {
1108 	case CAM_DATA_BIO:
1109 		icl_pdu_get_bio(response, pdu_offset,
1110 		    (struct bio *)csio->data_ptr, oreceived, data_segment_len);
1111 		break;
1112 	case CAM_DATA_VADDR:
1113 		icl_pdu_get_data(response, pdu_offset,
1114 		    csio->data_ptr + oreceived, data_segment_len);
1115 		break;
1116 	default:
1117 		__assert_unreachable();
1118 	}
1119 }
1120 
1121 static void
1122 iscsi_pdu_handle_data_in(struct icl_pdu *response)
1123 {
1124 	struct iscsi_bhs_data_in *bhsdi;
1125 	struct iscsi_outstanding *io;
1126 	struct iscsi_session *is;
1127 	union ccb *ccb;
1128 	struct ccb_scsiio *csio;
1129 	size_t data_segment_len, received, oreceived;
1130 
1131 	is = PDU_SESSION(response);
1132 	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
1133 	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
1134 	if (io == NULL || io->io_ccb == NULL) {
1135 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
1136 		icl_pdu_free(response);
1137 		iscsi_session_reconnect(is);
1138 		ISCSI_SESSION_UNLOCK(is);
1139 		return;
1140 	}
1141 
1142 	if (io->io_datasn != ntohl(bhsdi->bhsdi_datasn)) {
1143 		ISCSI_SESSION_WARN(is, "received Data-In PDU with "
1144 		    "DataSN %u, while expected %u; dropping connection",
1145 		    ntohl(bhsdi->bhsdi_datasn), io->io_datasn);
1146 		icl_pdu_free(response);
1147 		iscsi_session_reconnect(is);
1148 		ISCSI_SESSION_UNLOCK(is);
1149 		return;
1150 	}
1151 	io->io_datasn += response->ip_additional_pdus + 1;
1152 
1153 	data_segment_len = icl_pdu_data_segment_length(response);
1154 	if (data_segment_len == 0) {
1155 		/*
1156 		 * "The sending of 0 length data segments should be avoided,
1157 		 * but initiators and targets MUST be able to properly receive
1158 		 * 0 length data segments."
1159 		 */
1160 		ISCSI_SESSION_UNLOCK(is);
1161 		icl_pdu_free(response);
1162 		return;
1163 	}
1164 
1165 	/*
1166 	 * We need to track this for security reasons - without it, malicious target
1167 	 * could respond to SCSI READ without sending Data-In PDUs, which would result
1168 	 * in read operation on the initiator side returning random kernel data.
1169 	 */
1170 	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1171 		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1172 		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1173 		icl_pdu_free(response);
1174 		iscsi_session_reconnect(is);
1175 		ISCSI_SESSION_UNLOCK(is);
1176 		return;
1177 	}
1178 
1179 	ccb = io->io_ccb;
1180 	csio = &ccb->csio;
1181 
1182 	if (io->io_received + data_segment_len > csio->dxfer_len) {
1183 		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1184 		    "at offset %zd, buffer is %d)",
1185 		    data_segment_len, io->io_received, csio->dxfer_len);
1186 		icl_pdu_free(response);
1187 		iscsi_session_reconnect(is);
1188 		ISCSI_SESSION_UNLOCK(is);
1189 		return;
1190 	}
1191 
1192 	oreceived = io->io_received;
1193 	io->io_received += data_segment_len;
1194 	received = io->io_received;
1195 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0)
1196 		iscsi_outstanding_remove(is, io);
1197 	ISCSI_SESSION_UNLOCK(is);
1198 
1199 	iscsi_pdu_get_data_csio(response, 0, csio, oreceived, data_segment_len);
1200 
1201 	/*
1202 	 * XXX: Check F.
1203 	 */
1204 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1205 		/*
1206 		 * Nothing more to do.
1207 		 */
1208 		icl_pdu_free(response);
1209 		return;
1210 	}
1211 
1212 	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1213 	if (bhsdi->bhsdi_status == 0) {
1214 		ccb->ccb_h.status = CAM_REQ_CMP;
1215 	} else {
1216 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1217 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1218 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1219 		}
1220 		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1221 		csio->scsi_status = bhsdi->bhsdi_status;
1222 	}
1223 
1224 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1225 		KASSERT(received <= csio->dxfer_len,
1226 		    ("received > csio->dxfer_len"));
1227 		if (received < csio->dxfer_len) {
1228 			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1229 			if (csio->resid != csio->dxfer_len - received) {
1230 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1231 				    "target indicates %d, we calculated %zd",
1232 				    csio->resid, csio->dxfer_len - received);
1233 			}
1234 			csio->resid = csio->dxfer_len - received;
1235 		}
1236 	}
1237 
1238 	xpt_done(ccb);
1239 	icl_pdu_free(response);
1240 }
1241 
1242 static void
1243 iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1244 {
1245 
1246 	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1247 	icl_pdu_free(response);
1248 }
1249 
1250 static int
1251 iscsi_pdu_append_data_csio(struct icl_pdu *request, struct ccb_scsiio *csio,
1252     size_t off, size_t len, int how)
1253 {
1254 	switch (csio->ccb_h.flags & CAM_DATA_MASK) {
1255 	case CAM_DATA_BIO:
1256 		return (icl_pdu_append_bio(request,
1257 			(struct bio *)csio->data_ptr, off, len, how));
1258 	case CAM_DATA_VADDR:
1259 		return (icl_pdu_append_data(request, csio->data_ptr + off, len,
1260 		    how));
1261 	default:
1262 		__assert_unreachable();
1263 	}
1264 }
1265 
1266 static void
1267 iscsi_pdu_handle_r2t(struct icl_pdu *response)
1268 {
1269 	struct icl_pdu *request;
1270 	struct iscsi_session *is;
1271 	struct iscsi_bhs_r2t *bhsr2t;
1272 	struct iscsi_bhs_data_out *bhsdo;
1273 	struct iscsi_outstanding *io;
1274 	struct ccb_scsiio *csio;
1275 	size_t off, len, max_send_data_segment_length, total_len;
1276 	int error;
1277 	uint32_t datasn = 0;
1278 
1279 	is = PDU_SESSION(response);
1280 
1281 	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1282 	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1283 	if (io == NULL || io->io_ccb == NULL) {
1284 		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1285 		    bhsr2t->bhsr2t_initiator_task_tag);
1286 		icl_pdu_free(response);
1287 		iscsi_session_reconnect(is);
1288 		return;
1289 	}
1290 
1291 	csio = &io->io_ccb->csio;
1292 
1293 	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1294 		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1295 		icl_pdu_free(response);
1296 		iscsi_session_reconnect(is);
1297 		return;
1298 	}
1299 
1300 	/*
1301 	 * XXX: Verify R2TSN.
1302 	 */
1303 
1304 	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1305 	if (off > csio->dxfer_len) {
1306 		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1307 		    "%zd, buffer is %d; reconnecting", off, csio->dxfer_len);
1308 		icl_pdu_free(response);
1309 		iscsi_session_reconnect(is);
1310 		return;
1311 	}
1312 
1313 	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1314 	if (total_len == 0 || total_len > csio->dxfer_len) {
1315 		ISCSI_SESSION_WARN(is, "target requested invalid length "
1316 		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1317 		icl_pdu_free(response);
1318 		iscsi_session_reconnect(is);
1319 		return;
1320 	}
1321 
1322 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1323 
1324 	if (is->is_conn->ic_hw_isomax != 0)
1325 		max_send_data_segment_length = is->is_conn->ic_hw_isomax;
1326 	else
1327 		max_send_data_segment_length =
1328 		    is->is_conn->ic_max_send_data_segment_length;
1329 	for (;;) {
1330 		len = total_len;
1331 
1332 		if (len > max_send_data_segment_length)
1333 			len = max_send_data_segment_length;
1334 
1335 		if (off + len > csio->dxfer_len) {
1336 			ISCSI_SESSION_WARN(is, "target requested invalid "
1337 			    "length/offset %zd, buffer is %d; reconnecting",
1338 			    off + len, csio->dxfer_len);
1339 			icl_pdu_free(response);
1340 			iscsi_session_reconnect(is);
1341 			return;
1342 		}
1343 
1344 		request = icl_pdu_new(response->ip_conn, M_NOWAIT);
1345 		if (request == NULL) {
1346 			icl_pdu_free(response);
1347 			iscsi_session_reconnect(is);
1348 			return;
1349 		}
1350 
1351 		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1352 		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1353 		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1354 		bhsdo->bhsdo_initiator_task_tag =
1355 		    bhsr2t->bhsr2t_initiator_task_tag;
1356 		bhsdo->bhsdo_target_transfer_tag =
1357 		    bhsr2t->bhsr2t_target_transfer_tag;
1358 		bhsdo->bhsdo_datasn = htonl(datasn);
1359 		bhsdo->bhsdo_buffer_offset = htonl(off);
1360 		error = iscsi_pdu_append_data_csio(request, csio, off, len,
1361 		    M_NOWAIT | ICL_NOCOPY);
1362 		if (error != 0) {
1363 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1364 			    "reconnecting");
1365 			icl_pdu_free(request);
1366 			icl_pdu_free(response);
1367 			iscsi_session_reconnect(is);
1368 			return;
1369 		}
1370 
1371 		datasn += howmany(len,
1372 		    is->is_conn->ic_max_send_data_segment_length);
1373 		off += len;
1374 		total_len -= len;
1375 
1376 		if (total_len == 0) {
1377 			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1378 			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1379 		} else {
1380 			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1381 		}
1382 
1383 		iscsi_pdu_queue_locked(request);
1384 
1385 		if (total_len == 0)
1386 			break;
1387 	}
1388 
1389 	icl_pdu_free(response);
1390 }
1391 
1392 static void
1393 iscsi_pdu_handle_async_message(struct icl_pdu *response)
1394 {
1395 	struct iscsi_bhs_asynchronous_message *bhsam;
1396 	struct iscsi_session *is;
1397 
1398 	is = PDU_SESSION(response);
1399 	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1400 	switch (bhsam->bhsam_async_event) {
1401 	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1402 		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1403 		iscsi_session_logout(is);
1404 		iscsi_session_terminate(is);
1405 		break;
1406 	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1407 		ISCSI_SESSION_WARN(is, "target indicates it will drop the connection");
1408 		break;
1409 	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1410 		ISCSI_SESSION_WARN(is, "target indicates it will drop the session");
1411 		break;
1412 	default:
1413 		/*
1414 		 * XXX: Technically, we're obligated to also handle
1415 		 * 	parameter renegotiation.
1416 		 */
1417 		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1418 		break;
1419 	}
1420 
1421 	icl_pdu_free(response);
1422 }
1423 
1424 static void
1425 iscsi_pdu_handle_reject(struct icl_pdu *response)
1426 {
1427 	struct iscsi_bhs_reject *bhsr;
1428 	struct iscsi_session *is;
1429 
1430 	is = PDU_SESSION(response);
1431 	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1432 	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1433 	    bhsr->bhsr_reason);
1434 
1435 	icl_pdu_free(response);
1436 }
1437 
1438 static int
1439 iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1440     struct iscsi_daemon_request *request, bool freebsd13)
1441 {
1442 	struct iscsi_session *is;
1443 	int error;
1444 
1445 	sx_slock(&sc->sc_lock);
1446 	for (;;) {
1447 		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1448 			ISCSI_SESSION_LOCK(is);
1449 			if (is->is_conf.isc_enable == 0 &&
1450 			    is->is_conf.isc_discovery == 0) {
1451 				ISCSI_SESSION_UNLOCK(is);
1452 				continue;
1453 			}
1454 			if (is->is_waiting_for_iscsid)
1455 				break;
1456 			ISCSI_SESSION_UNLOCK(is);
1457 		}
1458 
1459 		if (is == NULL) {
1460 			if (sc->sc_unloading) {
1461 				sx_sunlock(&sc->sc_lock);
1462 				return (ENXIO);
1463 			}
1464 
1465 			/*
1466 			 * No session requires attention from iscsid(8); wait.
1467 			 */
1468 			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1469 			if (error != 0) {
1470 				sx_sunlock(&sc->sc_lock);
1471 				return (error);
1472 			}
1473 			continue;
1474 		}
1475 
1476 		is->is_waiting_for_iscsid = false;
1477 		is->is_login_phase = true;
1478 		is->is_reason[0] = '\0';
1479 		ISCSI_SESSION_UNLOCK(is);
1480 
1481 		request->idr_session_id = is->is_id;
1482 		memcpy(&request->idr_isid, &is->is_isid,
1483 		    sizeof(request->idr_isid));
1484 		request->idr_tsih = 0;	/* New or reinstated session. */
1485 		memcpy(&request->idr_conf, &is->is_conf,
1486 		    sizeof(request->idr_conf));
1487 
1488 #ifdef COMPAT_FREEBSD13
1489 		if (freebsd13) {
1490 			struct icl_drv_limits idl;
1491 			struct iscsi_daemon_request13 *request13;
1492 
1493 			error = icl_limits(is->is_conf.isc_offload,
1494 			    is->is_conf.isc_iser, 0, &idl);
1495 			if (error != 0) {
1496 				ISCSI_SESSION_WARN(is, "icl_limits for "
1497 				    "offload \"%s\" failed with error %d",
1498 				    is->is_conf.isc_offload, error);
1499 				sx_sunlock(&sc->sc_lock);
1500 				return (error);
1501 			}
1502 			request13 = (struct iscsi_daemon_request13 *)request;
1503 			request13->idr_limits.isl_max_recv_data_segment_length =
1504 			    idl.idl_max_recv_data_segment_length;
1505 			request13->idr_limits.isl_max_send_data_segment_length =
1506 			    idl.idl_max_send_data_segment_length;
1507 			request13->idr_limits.isl_max_burst_length =
1508 			    idl.idl_max_burst_length;
1509 			request13->idr_limits.isl_first_burst_length =
1510 			    idl.idl_first_burst_length;
1511 		}
1512 #endif
1513 		sx_sunlock(&sc->sc_lock);
1514 		return (0);
1515 	}
1516 }
1517 
1518 static int
1519 iscsi_ioctl_daemon_limits(struct iscsi_softc *sc,
1520     struct iscsi_daemon_limits *limits)
1521 {
1522 	struct icl_drv_limits idl;
1523 	struct iscsi_session *is;
1524 	int error;
1525 
1526 	sx_slock(&sc->sc_lock);
1527 
1528 	/*
1529 	 * Find the session to fetch limits for.
1530 	 */
1531 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1532 		if (is->is_id == limits->idl_session_id)
1533 			break;
1534 	}
1535 	if (is == NULL) {
1536 		sx_sunlock(&sc->sc_lock);
1537 		return (ESRCH);
1538 	}
1539 
1540 	error = icl_limits(is->is_conf.isc_offload, is->is_conf.isc_iser,
1541 	    limits->idl_socket, &idl);
1542 	sx_sunlock(&sc->sc_lock);
1543 	if (error != 0) {
1544 		ISCSI_SESSION_WARN(is, "icl_limits for offload \"%s\" "
1545 		    "failed with error %d", is->is_conf.isc_offload, error);
1546 		return (error);
1547 	}
1548 	limits->idl_limits.isl_max_recv_data_segment_length =
1549 	    idl.idl_max_recv_data_segment_length;
1550 	limits->idl_limits.isl_max_send_data_segment_length =
1551 	    idl.idl_max_send_data_segment_length;
1552 	limits->idl_limits.isl_max_burst_length =
1553 	    idl.idl_max_burst_length;
1554 	limits->idl_limits.isl_first_burst_length =
1555 	    idl.idl_first_burst_length;
1556 
1557 	return (0);
1558 }
1559 
1560 static int
1561 iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1562     struct iscsi_daemon_handoff *handoff)
1563 {
1564 	struct iscsi_session *is;
1565 	struct icl_conn *ic;
1566 	int error;
1567 
1568 	sx_slock(&sc->sc_lock);
1569 
1570 	/*
1571 	 * Find the session to hand off socket to.
1572 	 */
1573 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1574 		if (is->is_id == handoff->idh_session_id)
1575 			break;
1576 	}
1577 	if (is == NULL) {
1578 		sx_sunlock(&sc->sc_lock);
1579 		return (ESRCH);
1580 	}
1581 	ISCSI_SESSION_LOCK(is);
1582 	ic = is->is_conn;
1583 	if (is->is_conf.isc_discovery || is->is_terminating) {
1584 		ISCSI_SESSION_UNLOCK(is);
1585 		sx_sunlock(&sc->sc_lock);
1586 		return (EINVAL);
1587 	}
1588 	if (is->is_connected) {
1589 		/*
1590 		 * This might have happened because another iscsid(8)
1591 		 * instance handed off the connection in the meantime.
1592 		 * Just return.
1593 		 */
1594 		ISCSI_SESSION_WARN(is, "handoff on already connected "
1595 		    "session");
1596 		ISCSI_SESSION_UNLOCK(is);
1597 		sx_sunlock(&sc->sc_lock);
1598 		return (EBUSY);
1599 	}
1600 
1601 	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1602 	    sizeof(is->is_target_alias));
1603 	is->is_tsih = handoff->idh_tsih;
1604 	is->is_statsn = handoff->idh_statsn;
1605 	is->is_protocol_level = handoff->idh_protocol_level;
1606 	is->is_initial_r2t = handoff->idh_initial_r2t;
1607 	is->is_immediate_data = handoff->idh_immediate_data;
1608 
1609 	ic->ic_max_recv_data_segment_length =
1610 	    handoff->idh_max_recv_data_segment_length;
1611 	ic->ic_max_send_data_segment_length =
1612 	    handoff->idh_max_send_data_segment_length;
1613 	is->is_max_burst_length = handoff->idh_max_burst_length;
1614 	is->is_first_burst_length = handoff->idh_first_burst_length;
1615 
1616 	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1617 		ic->ic_header_crc32c = true;
1618 	else
1619 		ic->ic_header_crc32c = false;
1620 	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1621 		ic->ic_data_crc32c = true;
1622 	else
1623 		ic->ic_data_crc32c = false;
1624 	ic->ic_maxtags = maxtags;
1625 
1626 	is->is_cmdsn = 0;
1627 	is->is_expcmdsn = 0;
1628 	is->is_maxcmdsn = 0;
1629 	is->is_waiting_for_iscsid = false;
1630 	is->is_login_phase = false;
1631 	is->is_timeout = 0;
1632 	is->is_ping_timeout = is->is_conf.isc_ping_timeout;
1633 	if (is->is_ping_timeout < 0)
1634 		is->is_ping_timeout = ping_timeout;
1635 	is->is_login_timeout = is->is_conf.isc_login_timeout;
1636 	if (is->is_login_timeout < 0)
1637 		is->is_login_timeout = login_timeout;
1638 	is->is_connected = true;
1639 	is->is_reason[0] = '\0';
1640 
1641 	ISCSI_SESSION_UNLOCK(is);
1642 
1643 	/*
1644 	 * If we're going through the proxy, the idh_socket will be 0,
1645 	 * and the ICL module can simply ignore this call.  It can also
1646 	 * use it to determine it's no longer in the Login phase.
1647 	 */
1648 	error = icl_conn_handoff(ic, handoff->idh_socket);
1649 	if (error != 0) {
1650 		sx_sunlock(&sc->sc_lock);
1651 		iscsi_session_terminate(is);
1652 		return (error);
1653 	}
1654 
1655 	sx_sunlock(&sc->sc_lock);
1656 
1657 	if (is->is_sim != NULL) {
1658 		/*
1659 		 * When reconnecting, there already is SIM allocated for the session.
1660 		 */
1661 		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1662 		ISCSI_SESSION_LOCK(is);
1663 		ISCSI_SESSION_DEBUG(is, "releasing");
1664 		is->is_simq_frozen = false;
1665 		xpt_release_simq(is->is_sim, 1);
1666 		ISCSI_SESSION_UNLOCK(is);
1667 
1668 	} else {
1669 		ISCSI_SESSION_LOCK(is);
1670 		is->is_devq = cam_simq_alloc(ic->ic_maxtags);
1671 		if (is->is_devq == NULL) {
1672 			ISCSI_SESSION_UNLOCK(is);
1673 			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1674 			iscsi_session_terminate(is);
1675 			return (ENOMEM);
1676 		}
1677 
1678 		is->is_sim = cam_sim_alloc(iscsi_action, NULL, "iscsi",
1679 		    is, is->is_id /* unit */, &is->is_lock,
1680 		    1, ic->ic_maxtags, is->is_devq);
1681 		if (is->is_sim == NULL) {
1682 			ISCSI_SESSION_UNLOCK(is);
1683 			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1684 			cam_simq_free(is->is_devq);
1685 			iscsi_session_terminate(is);
1686 			return (ENOMEM);
1687 		}
1688 
1689 		if (xpt_bus_register(is->is_sim, NULL, 0) != 0) {
1690 			ISCSI_SESSION_UNLOCK(is);
1691 			ISCSI_SESSION_WARN(is, "failed to register bus");
1692 			iscsi_session_terminate(is);
1693 			return (ENOMEM);
1694 		}
1695 
1696 		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1697 		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1698 		    CAM_LUN_WILDCARD);
1699 		if (error != CAM_REQ_CMP) {
1700 			ISCSI_SESSION_UNLOCK(is);
1701 			ISCSI_SESSION_WARN(is, "failed to create path");
1702 			iscsi_session_terminate(is);
1703 			return (ENOMEM);
1704 		}
1705 		ISCSI_SESSION_UNLOCK(is);
1706 	}
1707 
1708 	return (0);
1709 }
1710 
1711 static int
1712 iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1713     struct iscsi_daemon_fail *fail)
1714 {
1715 	struct iscsi_session *is;
1716 
1717 	sx_slock(&sc->sc_lock);
1718 
1719 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1720 		if (is->is_id == fail->idf_session_id)
1721 			break;
1722 	}
1723 	if (is == NULL) {
1724 		sx_sunlock(&sc->sc_lock);
1725 		return (ESRCH);
1726 	}
1727 	ISCSI_SESSION_LOCK(is);
1728 	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1729 	    fail->idf_reason);
1730 	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1731 	//is->is_waiting_for_iscsid = false;
1732 	//is->is_login_phase = true;
1733 	//iscsi_session_reconnect(is);
1734 	ISCSI_SESSION_UNLOCK(is);
1735 	sx_sunlock(&sc->sc_lock);
1736 
1737 	return (0);
1738 }
1739 
1740 #ifdef ICL_KERNEL_PROXY
1741 static int
1742 iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1743     struct iscsi_daemon_connect *idc)
1744 {
1745 	struct iscsi_session *is;
1746 	struct sockaddr *from_sa, *to_sa;
1747 	int error;
1748 
1749 	sx_slock(&sc->sc_lock);
1750 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1751 		if (is->is_id == idc->idc_session_id)
1752 			break;
1753 	}
1754 	if (is == NULL) {
1755 		sx_sunlock(&sc->sc_lock);
1756 		return (ESRCH);
1757 	}
1758 	sx_sunlock(&sc->sc_lock);
1759 
1760 	if (idc->idc_from_addrlen > 0) {
1761 		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1762 		if (error != 0) {
1763 			ISCSI_SESSION_WARN(is,
1764 			    "getsockaddr failed with error %d", error);
1765 			return (error);
1766 		}
1767 	} else {
1768 		from_sa = NULL;
1769 	}
1770 	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1771 	if (error != 0) {
1772 		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1773 		    error);
1774 		free(from_sa, M_SONAME);
1775 		return (error);
1776 	}
1777 
1778 	ISCSI_SESSION_LOCK(is);
1779 	is->is_statsn = 0;
1780 	is->is_cmdsn = 0;
1781 	is->is_expcmdsn = 0;
1782 	is->is_maxcmdsn = 0;
1783 	is->is_waiting_for_iscsid = false;
1784 	is->is_login_phase = true;
1785 	is->is_timeout = 0;
1786 	ISCSI_SESSION_UNLOCK(is);
1787 
1788 	error = icl_conn_connect(is->is_conn, idc->idc_domain,
1789 	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1790 	free(from_sa, M_SONAME);
1791 	free(to_sa, M_SONAME);
1792 
1793 	/*
1794 	 * Digests are always disabled during login phase.
1795 	 */
1796 	is->is_conn->ic_header_crc32c = false;
1797 	is->is_conn->ic_data_crc32c = false;
1798 
1799 	return (error);
1800 }
1801 
1802 static int
1803 iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1804     struct iscsi_daemon_send *ids)
1805 {
1806 	struct iscsi_session *is;
1807 	struct icl_pdu *ip;
1808 	size_t datalen;
1809 	void *data;
1810 	int error;
1811 
1812 	sx_slock(&sc->sc_lock);
1813 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1814 		if (is->is_id == ids->ids_session_id)
1815 			break;
1816 	}
1817 	if (is == NULL) {
1818 		sx_sunlock(&sc->sc_lock);
1819 		return (ESRCH);
1820 	}
1821 	sx_sunlock(&sc->sc_lock);
1822 
1823 	if (is->is_login_phase == false)
1824 		return (EBUSY);
1825 
1826 	if (is->is_terminating || is->is_reconnecting)
1827 		return (EIO);
1828 
1829 	datalen = ids->ids_data_segment_len;
1830 	if (datalen > is->is_conn->ic_max_send_data_segment_length)
1831 		return (EINVAL);
1832 	if (datalen > 0) {
1833 		data = malloc(datalen, M_ISCSI, M_WAITOK);
1834 		error = copyin(ids->ids_data_segment, data, datalen);
1835 		if (error != 0) {
1836 			free(data, M_ISCSI);
1837 			return (error);
1838 		}
1839 	}
1840 
1841 	ip = icl_pdu_new(is->is_conn, M_WAITOK);
1842 	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1843 	if (datalen > 0) {
1844 		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1845 		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1846 		free(data, M_ISCSI);
1847 	}
1848 	iscsi_pdu_queue(ip);
1849 
1850 	return (0);
1851 }
1852 
1853 static int
1854 iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1855     struct iscsi_daemon_receive *idr)
1856 {
1857 	struct iscsi_session *is;
1858 	struct icl_pdu *ip;
1859 	void *data;
1860 	int error;
1861 
1862 	sx_slock(&sc->sc_lock);
1863 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1864 		if (is->is_id == idr->idr_session_id)
1865 			break;
1866 	}
1867 	if (is == NULL) {
1868 		sx_sunlock(&sc->sc_lock);
1869 		return (ESRCH);
1870 	}
1871 	sx_sunlock(&sc->sc_lock);
1872 
1873 	if (is->is_login_phase == false)
1874 		return (EBUSY);
1875 
1876 	ISCSI_SESSION_LOCK(is);
1877 	while (is->is_login_pdu == NULL &&
1878 	    is->is_terminating == false &&
1879 	    is->is_reconnecting == false) {
1880 		error = cv_wait_sig(&is->is_login_cv, &is->is_lock);
1881 		if (error != 0) {
1882 			ISCSI_SESSION_UNLOCK(is);
1883 			return (error);
1884 		}
1885 	}
1886 	if (is->is_terminating || is->is_reconnecting) {
1887 		ISCSI_SESSION_UNLOCK(is);
1888 		return (EIO);
1889 	}
1890 	ip = is->is_login_pdu;
1891 	is->is_login_pdu = NULL;
1892 	ISCSI_SESSION_UNLOCK(is);
1893 
1894 	if (ip->ip_data_len > idr->idr_data_segment_len) {
1895 		icl_pdu_free(ip);
1896 		return (EMSGSIZE);
1897 	}
1898 
1899 	error = copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1900 	if (error == 0 && ip->ip_data_len > 0) {
1901 		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1902 		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1903 		error = copyout(data, idr->idr_data_segment, ip->ip_data_len);
1904 		free(data, M_ISCSI);
1905 	}
1906 
1907 	icl_pdu_free(ip);
1908 
1909 	return (error);
1910 }
1911 #endif /* ICL_KERNEL_PROXY */
1912 
1913 static void
1914 iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1915 {
1916 	/*
1917 	 * Just make sure all the fields are null-terminated.
1918 	 *
1919 	 * XXX: This is not particularly secure.  We should
1920 	 * 	create our own conf and then copy in relevant
1921 	 * 	fields.
1922 	 */
1923 	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1924 	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1925 	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1926 	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1927 	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1928 	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1929 	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1930 	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1931 	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1932 }
1933 
1934 static bool
1935 iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1936 {
1937 
1938 	if (isc->isc_initiator[0] == '\0') {
1939 		ISCSI_DEBUG("empty isc_initiator");
1940 		return (false);
1941 	}
1942 
1943 	if (isc->isc_target_addr[0] == '\0') {
1944 		ISCSI_DEBUG("empty isc_target_addr");
1945 		return (false);
1946 	}
1947 
1948 	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1949 		ISCSI_DEBUG("non-empty isc_target for discovery session");
1950 		return (false);
1951 	}
1952 
1953 	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1954 		ISCSI_DEBUG("empty isc_target for non-discovery session");
1955 		return (false);
1956 	}
1957 
1958 	return (true);
1959 }
1960 
1961 static int
1962 iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1963 {
1964 	struct iscsi_session *is;
1965 	const struct iscsi_session *is2;
1966 	int error;
1967 	sbintime_t sbt, pr;
1968 
1969 	iscsi_sanitize_session_conf(&isa->isa_conf);
1970 	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1971 		return (EINVAL);
1972 
1973 	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1974 	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1975 
1976 	sx_xlock(&sc->sc_lock);
1977 
1978 	/*
1979 	 * Prevent duplicates.
1980 	 */
1981 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1982 		if (!!is->is_conf.isc_discovery !=
1983 		    !!is2->is_conf.isc_discovery)
1984 			continue;
1985 
1986 		if (strcmp(is->is_conf.isc_target_addr,
1987 		    is2->is_conf.isc_target_addr) != 0)
1988 			continue;
1989 
1990 		if (is->is_conf.isc_discovery == 0 &&
1991 		    strcmp(is->is_conf.isc_target,
1992 		    is2->is_conf.isc_target) != 0)
1993 			continue;
1994 
1995 		sx_xunlock(&sc->sc_lock);
1996 		free(is, M_ISCSI);
1997 		return (EBUSY);
1998 	}
1999 
2000 	is->is_conn = icl_new_conn(is->is_conf.isc_offload,
2001 	    is->is_conf.isc_iser, "iscsi", &is->is_lock);
2002 	if (is->is_conn == NULL) {
2003 		sx_xunlock(&sc->sc_lock);
2004 		free(is, M_ISCSI);
2005 		return (EINVAL);
2006 	}
2007 	is->is_conn->ic_receive = iscsi_receive_callback;
2008 	is->is_conn->ic_error = iscsi_error_callback;
2009 	is->is_conn->ic_prv0 = is;
2010 	TAILQ_INIT(&is->is_outstanding);
2011 	STAILQ_INIT(&is->is_postponed);
2012 	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
2013 	cv_init(&is->is_maintenance_cv, "iscsi_mt");
2014 #ifdef ICL_KERNEL_PROXY
2015 	cv_init(&is->is_login_cv, "iscsi_login");
2016 #endif
2017 
2018 	/*
2019 	 * Set some default values, from RFC 3720, section 12.
2020 	 *
2021 	 * These values are updated by the handoff IOCTL, but are
2022 	 * needed prior to the handoff to support sending the ISER
2023 	 * login PDU.
2024 	 */
2025 	is->is_conn->ic_max_recv_data_segment_length = 8192;
2026 	is->is_conn->ic_max_send_data_segment_length = 8192;
2027 	is->is_max_burst_length = 262144;
2028 	is->is_first_burst_length = 65536;
2029 
2030 	is->is_softc = sc;
2031 	sc->sc_last_session_id++;
2032 	is->is_id = sc->sc_last_session_id;
2033 	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
2034 	arc4rand(&is->is_isid[1], 5, 0);
2035 	is->is_tsih = 0;
2036 	callout_init(&is->is_callout, 1);
2037 
2038 	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
2039 	if (error != 0) {
2040 		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
2041 		sx_xunlock(&sc->sc_lock);
2042 		return (error);
2043 	}
2044 	is->is_ping_timeout = is->is_conf.isc_ping_timeout;
2045 	if (is->is_ping_timeout < 0)
2046 		is->is_ping_timeout = ping_timeout;
2047 	is->is_login_timeout = is->is_conf.isc_login_timeout;
2048 	if (is->is_login_timeout < 0)
2049 		is->is_login_timeout = login_timeout;
2050 
2051 	sbt = mstosbt(995);
2052 	pr = mstosbt(10);
2053 	callout_reset_sbt(&is->is_callout, sbt, pr, iscsi_callout, is, 0);
2054 	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
2055 
2056 	ISCSI_SESSION_LOCK(is);
2057 	/*
2058 	 * Don't notify iscsid(8) if the session is disabled and it's not
2059 	 * a discovery session,
2060 	 */
2061 	if (is->is_conf.isc_enable == 0 && is->is_conf.isc_discovery == 0) {
2062 		ISCSI_SESSION_UNLOCK(is);
2063 		sx_xunlock(&sc->sc_lock);
2064 		return (0);
2065 	}
2066 
2067 	is->is_waiting_for_iscsid = true;
2068 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
2069 	ISCSI_SESSION_UNLOCK(is);
2070 	cv_signal(&sc->sc_cv);
2071 	sx_xunlock(&sc->sc_lock);
2072 	return (0);
2073 }
2074 
2075 static bool
2076 iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
2077     unsigned int id2, const struct iscsi_session_conf *c2)
2078 {
2079 
2080 	if (id2 != 0 && id2 != id1)
2081 		return (false);
2082 	if (c2->isc_target[0] != '\0' &&
2083 	    strcmp(c1->isc_target, c2->isc_target) != 0)
2084 		return (false);
2085 	if (c2->isc_target_addr[0] != '\0' &&
2086 	    strcmp(c1->isc_target_addr, c2->isc_target_addr) != 0)
2087 		return (false);
2088 	return (true);
2089 }
2090 
2091 static int
2092 iscsi_ioctl_session_remove(struct iscsi_softc *sc,
2093     struct iscsi_session_remove *isr)
2094 {
2095 	struct iscsi_session *is, *tmp;
2096 	bool found = false;
2097 
2098 	iscsi_sanitize_session_conf(&isr->isr_conf);
2099 
2100 	sx_xlock(&sc->sc_lock);
2101 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
2102 		ISCSI_SESSION_LOCK(is);
2103 		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
2104 		    isr->isr_session_id, &isr->isr_conf)) {
2105 			found = true;
2106 			iscsi_session_logout(is);
2107 			iscsi_session_terminate(is);
2108 		}
2109 		ISCSI_SESSION_UNLOCK(is);
2110 	}
2111 	sx_xunlock(&sc->sc_lock);
2112 
2113 	if (!found)
2114 		return (ESRCH);
2115 
2116 	return (0);
2117 }
2118 
2119 static int
2120 iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
2121 {
2122 	int error;
2123 	unsigned int i = 0;
2124 	struct iscsi_session *is;
2125 	struct iscsi_session_state iss;
2126 
2127 	sx_slock(&sc->sc_lock);
2128 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2129 		if (i >= isl->isl_nentries) {
2130 			sx_sunlock(&sc->sc_lock);
2131 			return (EMSGSIZE);
2132 		}
2133 		memset(&iss, 0, sizeof(iss));
2134 		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
2135 		iss.iss_id = is->is_id;
2136 		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
2137 		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
2138 		strlcpy(iss.iss_offload, is->is_conn->ic_offload, sizeof(iss.iss_offload));
2139 
2140 		if (is->is_conn->ic_header_crc32c)
2141 			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
2142 		else
2143 			iss.iss_header_digest = ISCSI_DIGEST_NONE;
2144 
2145 		if (is->is_conn->ic_data_crc32c)
2146 			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
2147 		else
2148 			iss.iss_data_digest = ISCSI_DIGEST_NONE;
2149 
2150 		iss.iss_max_send_data_segment_length =
2151 		    is->is_conn->ic_max_send_data_segment_length;
2152 		iss.iss_max_recv_data_segment_length =
2153 		    is->is_conn->ic_max_recv_data_segment_length;
2154 		iss.iss_max_burst_length = is->is_max_burst_length;
2155 		iss.iss_first_burst_length = is->is_first_burst_length;
2156 		iss.iss_immediate_data = is->is_immediate_data;
2157 		iss.iss_connected = is->is_connected;
2158 
2159 		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
2160 		if (error != 0) {
2161 			sx_sunlock(&sc->sc_lock);
2162 			return (error);
2163 		}
2164 		i++;
2165 	}
2166 	sx_sunlock(&sc->sc_lock);
2167 
2168 	isl->isl_nentries = i;
2169 
2170 	return (0);
2171 }
2172 
2173 static int
2174 iscsi_ioctl_session_modify(struct iscsi_softc *sc,
2175     struct iscsi_session_modify *ism)
2176 {
2177 	struct iscsi_session *is;
2178 	const struct iscsi_session *is2;
2179 
2180 	iscsi_sanitize_session_conf(&ism->ism_conf);
2181 	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
2182 		return (EINVAL);
2183 
2184 	sx_xlock(&sc->sc_lock);
2185 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2186 		ISCSI_SESSION_LOCK(is);
2187 		if (is->is_id == ism->ism_session_id) {
2188 			/* Note that the session remains locked. */
2189 			break;
2190 		}
2191 		ISCSI_SESSION_UNLOCK(is);
2192 	}
2193 	if (is == NULL) {
2194 		sx_xunlock(&sc->sc_lock);
2195 		return (ESRCH);
2196 	}
2197 
2198 	/*
2199 	 * Prevent duplicates.
2200 	 */
2201 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
2202 		if (is == is2)
2203 			continue;
2204 
2205 		if (!!ism->ism_conf.isc_discovery !=
2206 		    !!is2->is_conf.isc_discovery)
2207 			continue;
2208 
2209 		if (strcmp(ism->ism_conf.isc_target_addr,
2210 		    is2->is_conf.isc_target_addr) != 0)
2211 			continue;
2212 
2213 		if (ism->ism_conf.isc_discovery == 0 &&
2214 		    strcmp(ism->ism_conf.isc_target,
2215 		    is2->is_conf.isc_target) != 0)
2216 			continue;
2217 
2218 		ISCSI_SESSION_UNLOCK(is);
2219 		sx_xunlock(&sc->sc_lock);
2220 		return (EBUSY);
2221 	}
2222 
2223 	sx_xunlock(&sc->sc_lock);
2224 
2225 	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
2226 	ISCSI_SESSION_UNLOCK(is);
2227 
2228 	iscsi_session_reconnect(is);
2229 
2230 	return (0);
2231 }
2232 
2233 static int
2234 iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
2235     struct thread *td)
2236 {
2237 	struct iscsi_softc *sc;
2238 
2239 	sc = dev->si_drv1;
2240 
2241 	switch (cmd) {
2242 	case ISCSIDWAIT:
2243 		return (iscsi_ioctl_daemon_wait(sc,
2244 		    (struct iscsi_daemon_request *)arg, false));
2245 #ifdef COMPAT_FREEBSD13
2246 	case ISCSIDWAIT13:
2247 		return (iscsi_ioctl_daemon_wait(sc,
2248 		    (struct iscsi_daemon_request *)arg, true));
2249 #endif
2250 	case ISCSIDLIMITS:
2251 		return (iscsi_ioctl_daemon_limits(sc,
2252 		    (struct iscsi_daemon_limits *)arg));
2253 	case ISCSIDHANDOFF:
2254 		return (iscsi_ioctl_daemon_handoff(sc,
2255 		    (struct iscsi_daemon_handoff *)arg));
2256 	case ISCSIDFAIL:
2257 		return (iscsi_ioctl_daemon_fail(sc,
2258 		    (struct iscsi_daemon_fail *)arg));
2259 #ifdef ICL_KERNEL_PROXY
2260 	case ISCSIDCONNECT:
2261 		return (iscsi_ioctl_daemon_connect(sc,
2262 		    (struct iscsi_daemon_connect *)arg));
2263 	case ISCSIDSEND:
2264 		return (iscsi_ioctl_daemon_send(sc,
2265 		    (struct iscsi_daemon_send *)arg));
2266 	case ISCSIDRECEIVE:
2267 		return (iscsi_ioctl_daemon_receive(sc,
2268 		    (struct iscsi_daemon_receive *)arg));
2269 #endif /* ICL_KERNEL_PROXY */
2270 	case ISCSISADD:
2271 		return (iscsi_ioctl_session_add(sc,
2272 		    (struct iscsi_session_add *)arg));
2273 	case ISCSISREMOVE:
2274 		return (iscsi_ioctl_session_remove(sc,
2275 		    (struct iscsi_session_remove *)arg));
2276 	case ISCSISLIST:
2277 		return (iscsi_ioctl_session_list(sc,
2278 		    (struct iscsi_session_list *)arg));
2279 	case ISCSISMODIFY:
2280 		return (iscsi_ioctl_session_modify(sc,
2281 		    (struct iscsi_session_modify *)arg));
2282 	default:
2283 		return (EINVAL);
2284 	}
2285 }
2286 
2287 static struct iscsi_outstanding *
2288 iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
2289 {
2290 	struct iscsi_outstanding *io;
2291 
2292 	ISCSI_SESSION_LOCK_ASSERT(is);
2293 
2294 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2295 		if (io->io_initiator_task_tag == initiator_task_tag)
2296 			return (io);
2297 	}
2298 	return (NULL);
2299 }
2300 
2301 static struct iscsi_outstanding *
2302 iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
2303 {
2304 	struct iscsi_outstanding *io;
2305 
2306 	ISCSI_SESSION_LOCK_ASSERT(is);
2307 
2308 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2309 		if (io->io_ccb == ccb)
2310 			return (io);
2311 	}
2312 	return (NULL);
2313 }
2314 
2315 static struct iscsi_outstanding *
2316 iscsi_outstanding_add(struct iscsi_session *is, struct icl_pdu *request,
2317     union ccb *ccb, uint32_t *initiator_task_tagp)
2318 {
2319 	struct iscsi_outstanding *io;
2320 	int error;
2321 
2322 	ISCSI_SESSION_LOCK_ASSERT(is);
2323 
2324 	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
2325 	if (io == NULL) {
2326 		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes",
2327 		    sizeof(*io));
2328 		return (NULL);
2329 	}
2330 
2331 	error = icl_conn_task_setup(is->is_conn, request, &ccb->csio,
2332 	    initiator_task_tagp, &io->io_icl_prv);
2333 	if (error != 0) {
2334 		ISCSI_SESSION_WARN(is,
2335 		    "icl_conn_task_setup() failed with error %d", error);
2336 		uma_zfree(iscsi_outstanding_zone, io);
2337 		return (NULL);
2338 	}
2339 
2340 	KASSERT(iscsi_outstanding_find(is, *initiator_task_tagp) == NULL,
2341 	    ("initiator_task_tag 0x%x already added", *initiator_task_tagp));
2342 
2343 	io->io_initiator_task_tag = *initiator_task_tagp;
2344 	io->io_ccb = ccb;
2345 	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
2346 	return (io);
2347 }
2348 
2349 static void
2350 iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2351 {
2352 
2353 	ISCSI_SESSION_LOCK_ASSERT(is);
2354 
2355 	icl_conn_task_done(is->is_conn, io->io_icl_prv);
2356 	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2357 	uma_zfree(iscsi_outstanding_zone, io);
2358 }
2359 
2360 static void
2361 iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2362 {
2363 	struct icl_pdu *request;
2364 	struct iscsi_bhs_task_management_request *bhstmr;
2365 	struct ccb_abort *cab = &ccb->cab;
2366 	struct iscsi_outstanding *io, *aio;
2367 	uint32_t initiator_task_tag;
2368 
2369 	ISCSI_SESSION_LOCK_ASSERT(is);
2370 
2371 #if 0
2372 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2373 #else
2374 	if (is->is_login_phase) {
2375 		ccb->ccb_h.status = CAM_REQ_ABORTED;
2376 		xpt_done(ccb);
2377 		return;
2378 	}
2379 #endif
2380 
2381 	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2382 	if (aio == NULL) {
2383 		ccb->ccb_h.status = CAM_REQ_CMP;
2384 		xpt_done(ccb);
2385 		return;
2386 	}
2387 
2388 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2389 	if (request == NULL) {
2390 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2391 		xpt_done(ccb);
2392 		return;
2393 	}
2394 
2395 	initiator_task_tag = is->is_initiator_task_tag++;
2396 	if (initiator_task_tag == 0xffffffff)
2397 		initiator_task_tag = is->is_initiator_task_tag++;
2398 
2399 	io = iscsi_outstanding_add(is, request, NULL, &initiator_task_tag);
2400 	if (io == NULL) {
2401 		icl_pdu_free(request);
2402 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2403 		xpt_done(ccb);
2404 		return;
2405 	}
2406 	io->io_referenced_task_tag = aio->io_initiator_task_tag;
2407 
2408 	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2409 	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2410 	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2411 	bhstmr->bhstmr_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2412 	bhstmr->bhstmr_initiator_task_tag = initiator_task_tag;
2413 	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2414 
2415 	iscsi_pdu_queue_locked(request);
2416 }
2417 
2418 static void
2419 iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2420 {
2421 	struct icl_pdu *request;
2422 	struct iscsi_bhs_scsi_command *bhssc;
2423 	struct ccb_scsiio *csio;
2424 	struct iscsi_outstanding *io;
2425 	size_t len;
2426 	uint32_t initiator_task_tag;
2427 	int error;
2428 
2429 	ISCSI_SESSION_LOCK_ASSERT(is);
2430 
2431 #if 0
2432 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2433 #else
2434 	if (is->is_login_phase) {
2435 		ISCSI_SESSION_DEBUG(is, "called during login phase");
2436 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2437 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2438 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2439 		}
2440 		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2441 		xpt_done(ccb);
2442 		return;
2443 	}
2444 #endif
2445 
2446 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2447 	if (request == NULL) {
2448 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2449 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2450 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2451 		}
2452 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2453 		xpt_done(ccb);
2454 		return;
2455 	}
2456 
2457 	initiator_task_tag = is->is_initiator_task_tag++;
2458 	if (initiator_task_tag == 0xffffffff)
2459 		initiator_task_tag = is->is_initiator_task_tag++;
2460 
2461 	io = iscsi_outstanding_add(is, request, ccb, &initiator_task_tag);
2462 	if (io == NULL) {
2463 		icl_pdu_free(request);
2464 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2465 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2466 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2467 		}
2468 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2469 		xpt_done(ccb);
2470 		return;
2471 	}
2472 
2473 	csio = &ccb->csio;
2474 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2475 	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2476 	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2477 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2478 	case CAM_DIR_IN:
2479 		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2480 		break;
2481 	case CAM_DIR_OUT:
2482 		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2483 		break;
2484 	}
2485 
2486 	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2487 		switch (csio->tag_action) {
2488 		case MSG_HEAD_OF_Q_TAG:
2489 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2490 			break;
2491 		case MSG_ORDERED_Q_TAG:
2492 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2493 			break;
2494 		case MSG_ACA_TASK:
2495 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2496 			break;
2497 		case MSG_SIMPLE_Q_TAG:
2498 		default:
2499 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2500 			break;
2501 		}
2502 	} else
2503 		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2504 
2505 	if (is->is_protocol_level >= 2) {
2506 		bhssc->bhssc_pri = (csio->priority << BHSSC_PRI_SHIFT) &
2507 		    BHSSC_PRI_MASK;
2508 	}
2509 
2510 	bhssc->bhssc_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2511 	bhssc->bhssc_initiator_task_tag = initiator_task_tag;
2512 	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2513 	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2514 	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2515 
2516 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2517 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2518 	else
2519 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2520 
2521 	if (is->is_immediate_data &&
2522 	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2523 		len = csio->dxfer_len;
2524 		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2525 		if (len > is->is_first_burst_length) {
2526 			ISCSI_SESSION_DEBUG(is, "len %zd -> %d", len, is->is_first_burst_length);
2527 			len = is->is_first_burst_length;
2528 		}
2529 		if (len > is->is_conn->ic_max_send_data_segment_length) {
2530 			ISCSI_SESSION_DEBUG(is, "len %zd -> %d", len,
2531 			    is->is_conn->ic_max_send_data_segment_length);
2532 			len = is->is_conn->ic_max_send_data_segment_length;
2533 		}
2534 
2535 		error = iscsi_pdu_append_data_csio(request, csio, 0, len,
2536 		    M_NOWAIT | ICL_NOCOPY);
2537 		if (error != 0) {
2538 			iscsi_outstanding_remove(is, io);
2539 			icl_pdu_free(request);
2540 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2541 				xpt_freeze_devq(ccb->ccb_h.path, 1);
2542 				ISCSI_SESSION_DEBUG(is, "freezing devq");
2543 			}
2544 			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2545 			xpt_done(ccb);
2546 			return;
2547 		}
2548 	}
2549 	iscsi_pdu_queue_locked(request);
2550 }
2551 
2552 static void
2553 iscsi_action(struct cam_sim *sim, union ccb *ccb)
2554 {
2555 	struct iscsi_session *is;
2556 
2557 	is = cam_sim_softc(sim);
2558 
2559 	ISCSI_SESSION_LOCK_ASSERT(is);
2560 
2561 	if (is->is_terminating ||
2562 	    (is->is_connected == false && fail_on_disconnection)) {
2563 		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2564 		xpt_done(ccb);
2565 		return;
2566 	}
2567 
2568 	/*
2569 	 * Make sure CAM doesn't sneak in a CCB just after freezing the queue.
2570 	 */
2571 	if (is->is_simq_frozen == true) {
2572 		ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
2573 		ccb->ccb_h.status |= CAM_REQUEUE_REQ;
2574 		/* Don't freeze the devq - the SIM queue is already frozen. */
2575 		xpt_done(ccb);
2576 		return;
2577 	}
2578 
2579 	switch (ccb->ccb_h.func_code) {
2580 	case XPT_PATH_INQ:
2581 	{
2582 		struct ccb_pathinq *cpi = &ccb->cpi;
2583 
2584 		cpi->version_num = 1;
2585 		cpi->hba_inquiry = PI_TAG_ABLE;
2586 		cpi->target_sprt = 0;
2587 		cpi->hba_misc = PIM_EXTLUNS;
2588 		/*
2589 		 * XXX: It shouldn't ever be NULL; this could be turned
2590 		 *      into a KASSERT eventually.
2591 		 */
2592 		if (is->is_conn == NULL)
2593 			ISCSI_WARN("NULL conn");
2594 		else if (is->is_conn->ic_unmapped)
2595 			cpi->hba_misc |= PIM_UNMAPPED;
2596 		cpi->hba_eng_cnt = 0;
2597 		cpi->max_target = 0;
2598 		/*
2599 		 * Note that the variable below is only relevant for targets
2600 		 * that don't claim compliance with anything above SPC2, which
2601 		 * means they don't support REPORT_LUNS.
2602 		 */
2603 		cpi->max_lun = 255;
2604 		cpi->initiator_id = ~0;
2605 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2606 		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2607 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2608 		cpi->unit_number = cam_sim_unit(sim);
2609 		cpi->bus_id = cam_sim_bus(sim);
2610 		cpi->base_transfer_speed = 150000; /* XXX */
2611 		cpi->transport = XPORT_ISCSI;
2612 		cpi->transport_version = 0;
2613 		cpi->protocol = PROTO_SCSI;
2614 		cpi->protocol_version = SCSI_REV_SPC3;
2615 		cpi->maxio = maxphys;
2616 		cpi->ccb_h.status = CAM_REQ_CMP;
2617 		break;
2618 	}
2619 	case XPT_GET_TRAN_SETTINGS:
2620 	{
2621 		struct ccb_trans_settings	*cts;
2622 		struct ccb_trans_settings_scsi	*scsi;
2623 
2624 		cts = &ccb->cts;
2625 		scsi = &cts->proto_specific.scsi;
2626 
2627 		cts->protocol = PROTO_SCSI;
2628 		cts->protocol_version = SCSI_REV_SPC3;
2629 		cts->transport = XPORT_ISCSI;
2630 		cts->transport_version = 0;
2631 		scsi->valid = CTS_SCSI_VALID_TQ;
2632 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2633 		cts->ccb_h.status = CAM_REQ_CMP;
2634 		break;
2635 	}
2636 	case XPT_CALC_GEOMETRY:
2637 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2638 		ccb->ccb_h.status = CAM_REQ_CMP;
2639 		break;
2640 #if 0
2641 	/*
2642 	 * XXX: What's the point?
2643 	 */
2644 	case XPT_RESET_BUS:
2645 	case XPT_TERM_IO:
2646 		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2647 		ccb->ccb_h.status = CAM_REQ_CMP;
2648 		break;
2649 #endif
2650 	case XPT_ABORT:
2651 		iscsi_action_abort(is, ccb);
2652 		return;
2653 	case XPT_SCSI_IO:
2654 		iscsi_action_scsiio(is, ccb);
2655 		return;
2656 	default:
2657 #if 0
2658 		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2659 #endif
2660 		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2661 		break;
2662 	}
2663 	xpt_done(ccb);
2664 }
2665 
2666 static void
2667 iscsi_terminate_sessions(struct iscsi_softc *sc)
2668 {
2669 	struct iscsi_session *is;
2670 
2671 	sx_slock(&sc->sc_lock);
2672 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2673 		iscsi_session_terminate(is);
2674 	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2675 		ISCSI_DEBUG("waiting for sessions to terminate");
2676 		cv_wait(&sc->sc_cv, &sc->sc_lock);
2677 	}
2678 	ISCSI_DEBUG("all sessions terminated");
2679 	sx_sunlock(&sc->sc_lock);
2680 }
2681 
2682 static void
2683 iscsi_shutdown_pre(struct iscsi_softc *sc, int howto)
2684 {
2685 	struct iscsi_session *is;
2686 
2687 	if (!fail_on_shutdown || (howto & RB_NOSYNC) != 0 ||
2688 	    SCHEDULER_STOPPED())
2689 		return;
2690 
2691 	/*
2692 	 * If we have any sessions waiting for reconnection, request
2693 	 * maintenance thread to fail them immediately instead of waiting
2694 	 * for reconnect timeout.
2695 	 *
2696 	 * This prevents LUNs with mounted filesystems that are supported
2697 	 * by disconnected iSCSI sessions from hanging, however it will
2698 	 * fail all queued BIOs.
2699 	 */
2700 	ISCSI_DEBUG("forcing failing all disconnected sessions due to shutdown");
2701 
2702 	fail_on_disconnection = 1;
2703 
2704 	sx_slock(&sc->sc_lock);
2705 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2706 		ISCSI_SESSION_LOCK(is);
2707 		if (!is->is_connected) {
2708 			ISCSI_SESSION_DEBUG(is, "force failing disconnected session early");
2709 			iscsi_session_reconnect(is);
2710 		}
2711 		ISCSI_SESSION_UNLOCK(is);
2712 	}
2713 	sx_sunlock(&sc->sc_lock);
2714 }
2715 
2716 static void
2717 iscsi_shutdown_post_sync(struct iscsi_softc *sc, int howto)
2718 {
2719 
2720 	if ((howto & RB_NOSYNC) == 0) {
2721 		ISCSI_DEBUG("removing all sessions due to shutdown");
2722 		iscsi_terminate_sessions(sc);
2723 	}
2724 }
2725 
2726 static int
2727 iscsi_load(void)
2728 {
2729 	int error;
2730 
2731 	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2732 	sx_init(&sc->sc_lock, "iscsi");
2733 	TAILQ_INIT(&sc->sc_sessions);
2734 	cv_init(&sc->sc_cv, "iscsi_cv");
2735 
2736 	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2737 	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2738 	    UMA_ALIGN_PTR, 0);
2739 
2740 	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2741 	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2742 	if (error != 0) {
2743 		ISCSI_WARN("failed to create device node, error %d", error);
2744 		return (error);
2745 	}
2746 	sc->sc_cdev->si_drv1 = sc;
2747 
2748 	sc->sc_shutdown_pre_eh = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2749 	    iscsi_shutdown_pre, sc, SHUTDOWN_PRI_FIRST);
2750 	/*
2751 	 * shutdown_post_sync needs to run after filesystem shutdown and before
2752 	 * CAM shutdown - otherwise when rebooting with an iSCSI session that is
2753 	 * disconnected but has outstanding requests, dashutdown() will hang on
2754 	 * cam_periph_runccb().
2755 	 */
2756 	sc->sc_shutdown_post_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2757 	    iscsi_shutdown_post_sync, sc, SHUTDOWN_PRI_DEFAULT - 1);
2758 
2759 	return (0);
2760 }
2761 
2762 static int
2763 iscsi_unload(void)
2764 {
2765 
2766 	/* Awaken any threads asleep in iscsi_ioctl(). */
2767 	sx_xlock(&sc->sc_lock);
2768 	sc->sc_unloading = true;
2769 	cv_signal(&sc->sc_cv);
2770 	sx_xunlock(&sc->sc_lock);
2771 
2772 	if (sc->sc_cdev != NULL) {
2773 		ISCSI_DEBUG("removing device node");
2774 		destroy_dev(sc->sc_cdev);
2775 		ISCSI_DEBUG("device node removed");
2776 	}
2777 
2778 	if (sc->sc_shutdown_pre_eh != NULL)
2779 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, sc->sc_shutdown_pre_eh);
2780 	if (sc->sc_shutdown_post_eh != NULL)
2781 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_post_eh);
2782 
2783 	iscsi_terminate_sessions(sc);
2784 
2785 	uma_zdestroy(iscsi_outstanding_zone);
2786 	sx_destroy(&sc->sc_lock);
2787 	cv_destroy(&sc->sc_cv);
2788 	free(sc, M_ISCSI);
2789 	return (0);
2790 }
2791 
2792 static int
2793 iscsi_quiesce(void)
2794 {
2795 	sx_slock(&sc->sc_lock);
2796 	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2797 		sx_sunlock(&sc->sc_lock);
2798 		return (EBUSY);
2799 	}
2800 	sx_sunlock(&sc->sc_lock);
2801 	return (0);
2802 }
2803 
2804 static int
2805 iscsi_modevent(module_t mod, int what, void *arg)
2806 {
2807 	int error;
2808 
2809 	switch (what) {
2810 	case MOD_LOAD:
2811 		error = iscsi_load();
2812 		break;
2813 	case MOD_UNLOAD:
2814 		error = iscsi_unload();
2815 		break;
2816 	case MOD_QUIESCE:
2817 		error = iscsi_quiesce();
2818 		break;
2819 	default:
2820 		error = EINVAL;
2821 		break;
2822 	}
2823 	return (error);
2824 }
2825 
2826 moduledata_t iscsi_data = {
2827 	"iscsi",
2828 	iscsi_modevent,
2829 	0
2830 };
2831 
2832 DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2833 MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2834 MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2835