iscsi.c revision 269065
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: stable/10/sys/dev/iscsi/iscsi.c 269065 2014-07-24 15:31:45Z mav $
30 */
31
32#include <sys/param.h>
33#include <sys/condvar.h>
34#include <sys/conf.h>
35#include <sys/eventhandler.h>
36#include <sys/file.h>
37#include <sys/kernel.h>
38#include <sys/kthread.h>
39#include <sys/lock.h>
40#include <sys/malloc.h>
41#include <sys/mutex.h>
42#include <sys/module.h>
43#include <sys/sysctl.h>
44#include <sys/systm.h>
45#include <sys/sx.h>
46#include <vm/uma.h>
47
48#include <cam/cam.h>
49#include <cam/cam_ccb.h>
50#include <cam/cam_xpt.h>
51#include <cam/cam_debug.h>
52#include <cam/cam_sim.h>
53#include <cam/cam_xpt_sim.h>
54#include <cam/cam_xpt_periph.h>
55#include <cam/cam_periph.h>
56#include <cam/scsi/scsi_all.h>
57#include <cam/scsi/scsi_message.h>
58
59#include "iscsi_ioctl.h"
60#include "iscsi.h"
61#include "icl.h"
62#include "iscsi_proto.h"
63
64#ifdef ICL_KERNEL_PROXY
65#include <sys/socketvar.h>
66#endif
67
68#ifdef ICL_KERNEL_PROXY
69FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
70#endif
71
72/*
73 * XXX: This is global so the iscsi_unload() can access it.
74 * 	Think about how to do this properly.
75 */
76static struct iscsi_softc	*sc;
77
78SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD, 0, "iSCSI initiator");
79static int debug = 1;
80TUNABLE_INT("kern.iscsi.debug", &debug);
81SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
82    &debug, 0, "Enable debug messages");
83static int ping_timeout = 5;
84TUNABLE_INT("kern.iscsi.ping_timeout", &ping_timeout);
85SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
86    0, "Timeout for ping (NOP-Out) requests, in seconds");
87static int iscsid_timeout = 60;
88TUNABLE_INT("kern.iscsi.iscsid_timeout", &iscsid_timeout);
89SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
90    0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
91static int login_timeout = 60;
92TUNABLE_INT("kern.iscsi.login_timeout", &login_timeout);
93SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
94    0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
95static int maxtags = 255;
96TUNABLE_INT("kern.iscsi.maxtags", &maxtags);
97SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
98    0, "Max number of IO requests queued");
99static int fail_on_disconnection = 0;
100TUNABLE_INT("kern.iscsi.fail_on_disconnection", &fail_on_disconnection);
101SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
102    &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
103
104static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
105static uma_zone_t iscsi_outstanding_zone;
106
107#define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
108#define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
109
110#define	ISCSI_DEBUG(X, ...)						\
111	do {								\
112		if (debug > 1) 						\
113			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
114	} while (0)
115
116#define	ISCSI_WARN(X, ...)						\
117	do {								\
118		if (debug > 0) {					\
119			printf("WARNING: %s: " X "\n",			\
120			    __func__, ## __VA_ARGS__);			\
121		}							\
122	} while (0)
123
124#define	ISCSI_SESSION_DEBUG(S, X, ...)					\
125	do {								\
126		if (debug > 1) {					\
127			printf("%s: %s (%s): " X "\n",			\
128			    __func__, S->is_conf.isc_target_addr,	\
129			    S->is_conf.isc_target, ## __VA_ARGS__);	\
130		}							\
131	} while (0)
132
133#define	ISCSI_SESSION_WARN(S, X, ...)					\
134	do {								\
135		if (debug > 0) {					\
136			printf("WARNING: %s (%s): " X "\n",		\
137			    S->is_conf.isc_target_addr,			\
138			    S->is_conf.isc_target, ## __VA_ARGS__);	\
139		}							\
140	} while (0)
141
142#define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
143#define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
144#define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
145
146static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
147		    int mode, struct thread *td);
148
149static struct cdevsw iscsi_cdevsw = {
150     .d_version = D_VERSION,
151     .d_ioctl   = iscsi_ioctl,
152     .d_name    = "iscsi",
153};
154
155static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
156static void	iscsi_pdu_queue(struct icl_pdu *request);
157static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
158static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
159static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
160static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
161static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
162static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
163static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
164static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
165static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
166static void	iscsi_session_reconnect(struct iscsi_session *is);
167static void	iscsi_session_terminate(struct iscsi_session *is);
168static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
169static void	iscsi_poll(struct cam_sim *sim);
170static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
171		    uint32_t initiator_task_tag);
172static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
173		    uint32_t initiator_task_tag, union ccb *ccb);
174static void	iscsi_outstanding_remove(struct iscsi_session *is,
175		    struct iscsi_outstanding *io);
176
177static bool
178iscsi_pdu_prepare(struct icl_pdu *request)
179{
180	struct iscsi_session *is;
181	struct iscsi_bhs_scsi_command *bhssc;
182
183	is = PDU_SESSION(request);
184
185	ISCSI_SESSION_LOCK_ASSERT(is);
186
187	/*
188	 * We're only using fields common for all the request
189	 * (initiator -> target) PDUs.
190	 */
191	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
192
193	/*
194	 * Data-Out PDU does not contain CmdSN.
195	 */
196	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
197		if (is->is_cmdsn > is->is_maxcmdsn &&
198		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
199			/*
200			 * Current MaxCmdSN prevents us from sending any more
201			 * SCSI Command PDUs to the target; postpone the PDU.
202			 * It will get resent by either iscsi_pdu_queue(),
203			 * or by maintenance thread.
204			 */
205#if 0
206			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %d, ExpCmdSN %d, MaxCmdSN %d, opcode 0x%x",
207			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn, bhssc->bhssc_opcode);
208#endif
209			return (true);
210		}
211		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
212		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
213			is->is_cmdsn++;
214	}
215	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
216
217	return (false);
218}
219
220static void
221iscsi_session_send_postponed(struct iscsi_session *is)
222{
223	struct icl_pdu *request;
224	bool postpone;
225
226	ISCSI_SESSION_LOCK_ASSERT(is);
227
228	while (!STAILQ_EMPTY(&is->is_postponed)) {
229		request = STAILQ_FIRST(&is->is_postponed);
230		postpone = iscsi_pdu_prepare(request);
231		if (postpone)
232			break;
233		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
234		icl_pdu_queue(request);
235	}
236}
237
238static void
239iscsi_pdu_queue_locked(struct icl_pdu *request)
240{
241	struct iscsi_session *is;
242	bool postpone;
243
244	is = PDU_SESSION(request);
245	ISCSI_SESSION_LOCK_ASSERT(is);
246	iscsi_session_send_postponed(is);
247	postpone = iscsi_pdu_prepare(request);
248	if (postpone) {
249		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
250		return;
251	}
252	icl_pdu_queue(request);
253}
254
255static void
256iscsi_pdu_queue(struct icl_pdu *request)
257{
258	struct iscsi_session *is;
259
260	is = PDU_SESSION(request);
261	ISCSI_SESSION_LOCK(is);
262	iscsi_pdu_queue_locked(request);
263	ISCSI_SESSION_UNLOCK(is);
264}
265
266static void
267iscsi_session_logout(struct iscsi_session *is)
268{
269	struct icl_pdu *request;
270	struct iscsi_bhs_logout_request *bhslr;
271
272	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
273	if (request == NULL)
274		return;
275
276	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
277	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
278	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
279	iscsi_pdu_queue_locked(request);
280}
281
282static void
283iscsi_session_terminate_task(struct iscsi_session *is,
284    struct iscsi_outstanding *io, bool requeue)
285{
286
287	if (io->io_ccb != NULL) {
288		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
289		if (requeue)
290			io->io_ccb->ccb_h.status |= CAM_REQUEUE_REQ;
291		else
292			io->io_ccb->ccb_h.status |= CAM_REQ_ABORTED;
293		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
294			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
295			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
296			ISCSI_SESSION_DEBUG(is, "freezing devq");
297		}
298		xpt_done(io->io_ccb);
299	}
300	iscsi_outstanding_remove(is, io);
301}
302
303static void
304iscsi_session_terminate_tasks(struct iscsi_session *is, bool requeue)
305{
306	struct iscsi_outstanding *io, *tmp;
307
308	ISCSI_SESSION_LOCK_ASSERT(is);
309
310	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
311		iscsi_session_terminate_task(is, io, requeue);
312	}
313}
314
315static void
316iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
317{
318	struct icl_pdu *pdu;
319
320	ISCSI_SESSION_LOCK_ASSERT(is);
321
322	/*
323	 * Don't queue any new PDUs.
324	 */
325	if (is->is_sim != NULL && is->is_simq_frozen == false) {
326		ISCSI_SESSION_DEBUG(is, "freezing");
327		xpt_freeze_simq(is->is_sim, 1);
328		is->is_simq_frozen = true;
329	}
330
331	/*
332	 * Remove postponed PDUs.
333	 */
334	while (!STAILQ_EMPTY(&is->is_postponed)) {
335		pdu = STAILQ_FIRST(&is->is_postponed);
336		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
337		icl_pdu_free(pdu);
338	}
339
340	if (destroy_sim == false) {
341		/*
342		 * Terminate SCSI tasks, asking CAM to requeue them.
343		 */
344		iscsi_session_terminate_tasks(is, true);
345		return;
346	}
347
348	iscsi_session_terminate_tasks(is, false);
349
350	if (is->is_sim == NULL)
351		return;
352
353	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
354	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
355
356	if (is->is_simq_frozen) {
357		xpt_release_simq(is->is_sim, 1);
358		is->is_simq_frozen = false;
359	}
360
361	xpt_free_path(is->is_path);
362	is->is_path = NULL;
363	xpt_bus_deregister(cam_sim_path(is->is_sim));
364	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
365	is->is_sim = NULL;
366	is->is_devq = NULL;
367}
368
369static void
370iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
371{
372
373	icl_conn_shutdown(is->is_conn);
374	icl_conn_close(is->is_conn);
375
376	ISCSI_SESSION_LOCK(is);
377
378	is->is_connected = false;
379	is->is_reconnecting = false;
380	is->is_login_phase = false;
381
382#ifdef ICL_KERNEL_PROXY
383	if (is->is_login_pdu != NULL) {
384		icl_pdu_free(is->is_login_pdu);
385		is->is_login_pdu = NULL;
386	}
387	cv_signal(&is->is_login_cv);
388#endif
389
390	if (fail_on_disconnection) {
391		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
392		iscsi_session_cleanup(is, true);
393	} else {
394		iscsi_session_cleanup(is, false);
395	}
396
397	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
398	    ("destroying session with active tasks"));
399	KASSERT(STAILQ_EMPTY(&is->is_postponed),
400	    ("destroying session with postponed PDUs"));
401
402	/*
403	 * Request immediate reconnection from iscsid(8).
404	 */
405	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
406	is->is_waiting_for_iscsid = true;
407	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
408	is->is_timeout = 0;
409	ISCSI_SESSION_UNLOCK(is);
410	cv_signal(&is->is_softc->sc_cv);
411}
412
413static void
414iscsi_maintenance_thread_terminate(struct iscsi_session *is)
415{
416	struct iscsi_softc *sc;
417
418	sc = is->is_softc;
419	sx_xlock(&sc->sc_lock);
420	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
421	sx_xunlock(&sc->sc_lock);
422
423	icl_conn_close(is->is_conn);
424
425	ISCSI_SESSION_LOCK(is);
426
427	KASSERT(is->is_terminating, ("is_terminating == false"));
428
429#ifdef ICL_KERNEL_PROXY
430	if (is->is_login_pdu != NULL) {
431		icl_pdu_free(is->is_login_pdu);
432		is->is_login_pdu = NULL;
433	}
434	cv_signal(&is->is_login_cv);
435#endif
436
437	callout_drain(&is->is_callout);
438
439	iscsi_session_cleanup(is, true);
440
441	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
442	    ("destroying session with active tasks"));
443	KASSERT(STAILQ_EMPTY(&is->is_postponed),
444	    ("destroying session with postponed PDUs"));
445
446	ISCSI_SESSION_UNLOCK(is);
447
448	icl_conn_free(is->is_conn);
449	mtx_destroy(&is->is_lock);
450	cv_destroy(&is->is_maintenance_cv);
451#ifdef ICL_KERNEL_PROXY
452	cv_destroy(&is->is_login_cv);
453#endif
454	ISCSI_SESSION_DEBUG(is, "terminated");
455	free(is, M_ISCSI);
456
457	/*
458	 * The iscsi_unload() routine might be waiting.
459	 */
460	cv_signal(&sc->sc_cv);
461}
462
463static void
464iscsi_maintenance_thread(void *arg)
465{
466	struct iscsi_session *is;
467
468	is = arg;
469
470	for (;;) {
471		ISCSI_SESSION_LOCK(is);
472		if (is->is_reconnecting == false &&
473		    is->is_terminating == false &&
474		    STAILQ_EMPTY(&is->is_postponed))
475			cv_wait(&is->is_maintenance_cv, &is->is_lock);
476
477		if (is->is_reconnecting) {
478			ISCSI_SESSION_UNLOCK(is);
479			iscsi_maintenance_thread_reconnect(is);
480			continue;
481		}
482
483		if (is->is_terminating) {
484			ISCSI_SESSION_UNLOCK(is);
485			iscsi_maintenance_thread_terminate(is);
486			kthread_exit();
487			return;
488		}
489
490		iscsi_session_send_postponed(is);
491		ISCSI_SESSION_UNLOCK(is);
492	}
493}
494
495static void
496iscsi_session_reconnect(struct iscsi_session *is)
497{
498
499	/*
500	 * XXX: We can't use locking here, because
501	 * 	it's being called from various contexts.
502	 * 	Hope it doesn't break anything.
503	 */
504	if (is->is_reconnecting)
505		return;
506
507	is->is_reconnecting = true;
508	cv_signal(&is->is_maintenance_cv);
509}
510
511static void
512iscsi_session_terminate(struct iscsi_session *is)
513{
514	if (is->is_terminating)
515		return;
516
517	is->is_terminating = true;
518
519#if 0
520	iscsi_session_logout(is);
521#endif
522	cv_signal(&is->is_maintenance_cv);
523}
524
525static void
526iscsi_callout(void *context)
527{
528	struct icl_pdu *request;
529	struct iscsi_bhs_nop_out *bhsno;
530	struct iscsi_session *is;
531	bool reconnect_needed = false;
532
533	is = context;
534
535	if (is->is_terminating)
536		return;
537
538	callout_schedule(&is->is_callout, 1 * hz);
539
540	ISCSI_SESSION_LOCK(is);
541	is->is_timeout++;
542
543	if (is->is_waiting_for_iscsid) {
544		if (is->is_timeout > iscsid_timeout) {
545			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
546			    "for %d seconds; reconnecting",
547			    is->is_timeout);
548			reconnect_needed = true;
549		}
550		goto out;
551	}
552
553	if (is->is_login_phase) {
554		if (is->is_timeout > login_timeout) {
555			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
556			    "reconnecting", is->is_timeout);
557			reconnect_needed = true;
558		}
559		goto out;
560	}
561
562	if (is->is_timeout >= ping_timeout) {
563		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
564		    "reconnecting", ping_timeout);
565		reconnect_needed = true;
566		goto out;
567	}
568
569	ISCSI_SESSION_UNLOCK(is);
570
571	/*
572	 * If the ping was reset less than one second ago - which means
573	 * that we've received some PDU during the last second - assume
574	 * the traffic flows correctly and don't bother sending a NOP-Out.
575	 *
576	 * (It's 2 - one for one second, and one for incrementing is_timeout
577	 * earlier in this routine.)
578	 */
579	if (is->is_timeout < 2)
580		return;
581
582	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
583	if (request == NULL) {
584		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
585		return;
586	}
587	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
588	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
589	    ISCSI_BHS_OPCODE_IMMEDIATE;
590	bhsno->bhsno_flags = 0x80;
591	bhsno->bhsno_target_transfer_tag = 0xffffffff;
592	iscsi_pdu_queue(request);
593	return;
594
595out:
596	ISCSI_SESSION_UNLOCK(is);
597
598	if (reconnect_needed)
599		iscsi_session_reconnect(is);
600}
601
602static void
603iscsi_pdu_update_statsn(const struct icl_pdu *response)
604{
605	const struct iscsi_bhs_data_in *bhsdi;
606	struct iscsi_session *is;
607	uint32_t expcmdsn, maxcmdsn;
608
609	is = PDU_SESSION(response);
610
611	ISCSI_SESSION_LOCK_ASSERT(is);
612
613	/*
614	 * We're only using fields common for all the response
615	 * (target -> initiator) PDUs.
616	 */
617	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
618	/*
619	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
620	 * and Residual Count only have meaningful content if the S bit
621	 * is set to 1", so we also need to check the bit specific for
622	 * Data-In PDU.
623	 */
624	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
625	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
626		if (ntohl(bhsdi->bhsdi_statsn) < is->is_statsn) {
627			ISCSI_SESSION_WARN(is,
628			    "PDU StatSN %d >= session StatSN %d, opcode 0x%x",
629			    is->is_statsn, ntohl(bhsdi->bhsdi_statsn),
630			    bhsdi->bhsdi_opcode);
631		}
632		is->is_statsn = ntohl(bhsdi->bhsdi_statsn);
633	}
634
635	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
636	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
637
638	/*
639	 * XXX: Compare using Serial Arithmetic Sense.
640	 */
641	if (maxcmdsn + 1 < expcmdsn) {
642		ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d + 1 < PDU ExpCmdSN %d; ignoring",
643		    maxcmdsn, expcmdsn);
644	} else {
645		if (maxcmdsn > is->is_maxcmdsn) {
646			is->is_maxcmdsn = maxcmdsn;
647
648			/*
649			 * Command window increased; kick the maintanance thread
650			 * to send out postponed commands.
651			 */
652			if (!STAILQ_EMPTY(&is->is_postponed))
653				cv_signal(&is->is_maintenance_cv);
654		} else if (maxcmdsn < is->is_maxcmdsn) {
655			ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d < session MaxCmdSN %d; ignoring",
656			    maxcmdsn, is->is_maxcmdsn);
657		}
658
659		if (expcmdsn > is->is_expcmdsn) {
660			is->is_expcmdsn = expcmdsn;
661		} else if (expcmdsn < is->is_expcmdsn) {
662			ISCSI_SESSION_DEBUG(is, "PDU ExpCmdSN %d < session ExpCmdSN %d; ignoring",
663			    expcmdsn, is->is_expcmdsn);
664		}
665	}
666
667	/*
668	 * Every incoming PDU - not just NOP-In - resets the ping timer.
669	 * The purpose of the timeout is to reset the connection when it stalls;
670	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
671	 * in some queue.
672	 */
673	is->is_timeout = 0;
674}
675
676static void
677iscsi_receive_callback(struct icl_pdu *response)
678{
679	struct iscsi_session *is;
680
681	is = PDU_SESSION(response);
682
683	ISCSI_SESSION_LOCK(is);
684
685#ifdef ICL_KERNEL_PROXY
686	if (is->is_login_phase) {
687		if (is->is_login_pdu == NULL)
688			is->is_login_pdu = response;
689		else
690			icl_pdu_free(response);
691		ISCSI_SESSION_UNLOCK(is);
692		cv_signal(&is->is_login_cv);
693		return;
694	}
695#endif
696
697	iscsi_pdu_update_statsn(response);
698
699	/*
700	 * The handling routine is responsible for freeing the PDU
701	 * when it's no longer needed.
702	 */
703	switch (response->ip_bhs->bhs_opcode) {
704	case ISCSI_BHS_OPCODE_NOP_IN:
705		iscsi_pdu_handle_nop_in(response);
706		break;
707	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
708		iscsi_pdu_handle_scsi_response(response);
709		break;
710	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
711		iscsi_pdu_handle_task_response(response);
712		break;
713	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
714		iscsi_pdu_handle_data_in(response);
715		break;
716	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
717		iscsi_pdu_handle_logout_response(response);
718		break;
719	case ISCSI_BHS_OPCODE_R2T:
720		iscsi_pdu_handle_r2t(response);
721		break;
722	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
723		iscsi_pdu_handle_async_message(response);
724		break;
725	case ISCSI_BHS_OPCODE_REJECT:
726		iscsi_pdu_handle_reject(response);
727		break;
728	default:
729		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
730		    "opcode 0x%x; reconnecting",
731		    response->ip_bhs->bhs_opcode);
732		iscsi_session_reconnect(is);
733		icl_pdu_free(response);
734	}
735
736	ISCSI_SESSION_UNLOCK(is);
737}
738
739static void
740iscsi_error_callback(struct icl_conn *ic)
741{
742	struct iscsi_session *is;
743
744	is = CONN_SESSION(ic);
745
746	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
747	iscsi_session_reconnect(is);
748}
749
750static void
751iscsi_pdu_handle_nop_in(struct icl_pdu *response)
752{
753	struct iscsi_session *is;
754	struct iscsi_bhs_nop_out *bhsno;
755	struct iscsi_bhs_nop_in *bhsni;
756	struct icl_pdu *request;
757	void *data = NULL;
758	size_t datasize;
759	int error;
760
761	is = PDU_SESSION(response);
762	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
763
764	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
765		/*
766		 * Nothing to do; iscsi_pdu_update_statsn() already
767		 * zeroed the timeout.
768		 */
769		icl_pdu_free(response);
770		return;
771	}
772
773	datasize = icl_pdu_data_segment_length(response);
774	if (datasize > 0) {
775		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
776		if (data == NULL) {
777			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
778			    "reconnecting");
779			icl_pdu_free(response);
780			iscsi_session_reconnect(is);
781			return;
782		}
783		icl_pdu_get_data(response, 0, data, datasize);
784	}
785
786	request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
787	if (request == NULL) {
788		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
789		    "reconnecting");
790		free(data, M_ISCSI);
791		icl_pdu_free(response);
792		iscsi_session_reconnect(is);
793		return;
794	}
795	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
796	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
797	    ISCSI_BHS_OPCODE_IMMEDIATE;
798	bhsno->bhsno_flags = 0x80;
799	bhsno->bhsno_initiator_task_tag = 0xffffffff;
800	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
801	if (datasize > 0) {
802		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
803		if (error != 0) {
804			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
805			    "reconnecting");
806			free(data, M_ISCSI);
807			icl_pdu_free(request);
808			icl_pdu_free(response);
809			iscsi_session_reconnect(is);
810			return;
811		}
812		free(data, M_ISCSI);
813	}
814
815	icl_pdu_free(response);
816	iscsi_pdu_queue_locked(request);
817}
818
819static void
820iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
821{
822	struct iscsi_bhs_scsi_response *bhssr;
823	struct iscsi_outstanding *io;
824	struct iscsi_session *is;
825	struct ccb_scsiio *csio;
826	size_t data_segment_len;
827	uint16_t sense_len;
828
829	is = PDU_SESSION(response);
830
831	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
832	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
833	if (io == NULL || io->io_ccb == NULL) {
834		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
835		icl_pdu_free(response);
836		iscsi_session_reconnect(is);
837		return;
838	}
839
840	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
841		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
842 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
843 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
844			ISCSI_SESSION_DEBUG(is, "freezing devq");
845		}
846 		io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
847	} else if (bhssr->bhssr_status == 0) {
848		io->io_ccb->ccb_h.status = CAM_REQ_CMP;
849	} else {
850 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
851 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
852			ISCSI_SESSION_DEBUG(is, "freezing devq");
853		}
854 		io->io_ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
855		io->io_ccb->csio.scsi_status = bhssr->bhssr_status;
856	}
857
858	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_OVERFLOW) {
859		ISCSI_SESSION_WARN(is, "target indicated residual overflow");
860		icl_pdu_free(response);
861		iscsi_session_reconnect(is);
862		return;
863	}
864
865	csio = &io->io_ccb->csio;
866
867	data_segment_len = icl_pdu_data_segment_length(response);
868	if (data_segment_len > 0) {
869		if (data_segment_len < sizeof(sense_len)) {
870			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
871			    data_segment_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		}
879		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
880		sense_len = ntohs(sense_len);
881#if 0
882		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
883		    sense_len, data_segment_len);
884#endif
885		if (sizeof(sense_len) + sense_len > data_segment_len) {
886			ISCSI_SESSION_WARN(is, "truncated data segment "
887			    "(%zd bytes, should be %zd)",
888			    data_segment_len, sizeof(sense_len) + sense_len);
889			if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
890				xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
891				ISCSI_SESSION_DEBUG(is, "freezing devq");
892			}
893			io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
894			goto out;
895		} else if (sizeof(sense_len) + sense_len < data_segment_len)
896			ISCSI_SESSION_WARN(is, "oversize data segment "
897			    "(%zd bytes, should be %zd)",
898			    data_segment_len, sizeof(sense_len) + sense_len);
899		if (sense_len > csio->sense_len) {
900			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
901			    sense_len, csio->sense_len);
902			sense_len = csio->sense_len;
903		}
904		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
905		csio->sense_resid = csio->sense_len - sense_len;
906		io->io_ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
907	}
908
909out:
910	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
911		csio->resid = ntohl(bhssr->bhssr_residual_count);
912
913	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
914		KASSERT(io->io_received <= csio->dxfer_len,
915		    ("io->io_received > csio->dxfer_len"));
916		if (io->io_received < csio->dxfer_len) {
917			if (csio->resid != csio->dxfer_len - io->io_received) {
918				ISCSI_SESSION_WARN(is, "underflow mismatch: "
919				    "target indicates %d, we calculated %zd",
920				    csio->resid,
921				    csio->dxfer_len - io->io_received);
922			}
923			csio->resid = csio->dxfer_len - io->io_received;
924		}
925	}
926
927	xpt_done(io->io_ccb);
928	iscsi_outstanding_remove(is, io);
929	icl_pdu_free(response);
930}
931
932static void
933iscsi_pdu_handle_task_response(struct icl_pdu *response)
934{
935	struct iscsi_bhs_task_management_response *bhstmr;
936	struct iscsi_outstanding *io, *aio;
937	struct iscsi_session *is;
938
939	is = PDU_SESSION(response);
940
941	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
942	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
943	if (io == NULL || io->io_ccb != NULL) {
944		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
945		    bhstmr->bhstmr_initiator_task_tag);
946		icl_pdu_free(response);
947		iscsi_session_reconnect(is);
948		return;
949	}
950
951	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
952		ISCSI_SESSION_WARN(is, "task response 0x%x",
953		    bhstmr->bhstmr_response);
954	} else {
955		aio = iscsi_outstanding_find(is, io->io_datasn);
956		if (aio != NULL && aio->io_ccb != NULL)
957			iscsi_session_terminate_task(is, aio, false);
958	}
959
960	iscsi_outstanding_remove(is, io);
961	icl_pdu_free(response);
962}
963
964static void
965iscsi_pdu_handle_data_in(struct icl_pdu *response)
966{
967	struct iscsi_bhs_data_in *bhsdi;
968	struct iscsi_outstanding *io;
969	struct iscsi_session *is;
970	struct ccb_scsiio *csio;
971	size_t data_segment_len;
972
973	is = PDU_SESSION(response);
974	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
975	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
976	if (io == NULL || io->io_ccb == NULL) {
977		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
978		icl_pdu_free(response);
979		iscsi_session_reconnect(is);
980		return;
981	}
982
983	data_segment_len = icl_pdu_data_segment_length(response);
984	if (data_segment_len == 0) {
985		/*
986		 * "The sending of 0 length data segments should be avoided,
987		 * but initiators and targets MUST be able to properly receive
988		 * 0 length data segments."
989		 */
990		icl_pdu_free(response);
991		return;
992	}
993
994	/*
995	 * We need to track this for security reasons - without it, malicious target
996	 * could respond to SCSI READ without sending Data-In PDUs, which would result
997	 * in read operation on the initiator side returning random kernel data.
998	 */
999	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1000		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1001		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1002		icl_pdu_free(response);
1003		iscsi_session_reconnect(is);
1004		return;
1005	}
1006
1007	csio = &io->io_ccb->csio;
1008
1009	if (io->io_received + data_segment_len > csio->dxfer_len) {
1010		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1011		    "at offset %zd, buffer is %d)",
1012		    data_segment_len, io->io_received, csio->dxfer_len);
1013		icl_pdu_free(response);
1014		iscsi_session_reconnect(is);
1015		return;
1016	}
1017
1018	icl_pdu_get_data(response, 0, csio->data_ptr + io->io_received, data_segment_len);
1019	io->io_received += data_segment_len;
1020
1021	/*
1022	 * XXX: Check DataSN.
1023	 * XXX: Check F.
1024	 */
1025	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1026		/*
1027		 * Nothing more to do.
1028		 */
1029		icl_pdu_free(response);
1030		return;
1031	}
1032
1033	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1034	if (bhsdi->bhsdi_status == 0) {
1035		io->io_ccb->ccb_h.status = CAM_REQ_CMP;
1036	} else {
1037		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1038			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
1039			ISCSI_SESSION_DEBUG(is, "freezing devq");
1040		}
1041		io->io_ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1042		csio->scsi_status = bhsdi->bhsdi_status;
1043	}
1044
1045	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1046		KASSERT(io->io_received <= csio->dxfer_len,
1047		    ("io->io_received > csio->dxfer_len"));
1048		if (io->io_received < csio->dxfer_len) {
1049			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1050			if (csio->resid != csio->dxfer_len - io->io_received) {
1051				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1052				    "target indicates %d, we calculated %zd",
1053				    csio->resid,
1054				    csio->dxfer_len - io->io_received);
1055			}
1056			csio->resid = csio->dxfer_len - io->io_received;
1057		}
1058	}
1059
1060	xpt_done(io->io_ccb);
1061	iscsi_outstanding_remove(is, io);
1062	icl_pdu_free(response);
1063}
1064
1065static void
1066iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1067{
1068
1069	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1070	icl_pdu_free(response);
1071}
1072
1073static void
1074iscsi_pdu_handle_r2t(struct icl_pdu *response)
1075{
1076	struct icl_pdu *request;
1077	struct iscsi_session *is;
1078	struct iscsi_bhs_r2t *bhsr2t;
1079	struct iscsi_bhs_data_out *bhsdo;
1080	struct iscsi_outstanding *io;
1081	struct ccb_scsiio *csio;
1082	size_t off, len, total_len;
1083	int error;
1084
1085	is = PDU_SESSION(response);
1086
1087	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1088	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1089	if (io == NULL || io->io_ccb == NULL) {
1090		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1091		    bhsr2t->bhsr2t_initiator_task_tag);
1092		icl_pdu_free(response);
1093		iscsi_session_reconnect(is);
1094		return;
1095	}
1096
1097	csio = &io->io_ccb->csio;
1098
1099	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1100		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1101		icl_pdu_free(response);
1102		iscsi_session_reconnect(is);
1103		return;
1104	}
1105
1106	/*
1107	 * XXX: Verify R2TSN.
1108	 */
1109
1110	io->io_datasn = 0;
1111
1112	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1113	if (off > csio->dxfer_len) {
1114		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1115		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1116		icl_pdu_free(response);
1117		iscsi_session_reconnect(is);
1118		return;
1119	}
1120
1121	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1122	if (total_len == 0 || total_len > csio->dxfer_len) {
1123		ISCSI_SESSION_WARN(is, "target requested invalid length "
1124		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1125		icl_pdu_free(response);
1126		iscsi_session_reconnect(is);
1127		return;
1128	}
1129
1130	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1131
1132	for (;;) {
1133		len = total_len;
1134
1135		if (len > is->is_max_data_segment_length)
1136			len = is->is_max_data_segment_length;
1137
1138		if (off + len > csio->dxfer_len) {
1139			ISCSI_SESSION_WARN(is, "target requested invalid "
1140			    "length/offset %zd, buffer is %d; reconnecting",
1141			    off + len, csio->dxfer_len);
1142			icl_pdu_free(response);
1143			iscsi_session_reconnect(is);
1144			return;
1145		}
1146
1147		request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
1148		if (request == NULL) {
1149			icl_pdu_free(response);
1150			iscsi_session_reconnect(is);
1151			return;
1152		}
1153
1154		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1155		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1156		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1157		bhsdo->bhsdo_initiator_task_tag =
1158		    bhsr2t->bhsr2t_initiator_task_tag;
1159		bhsdo->bhsdo_target_transfer_tag =
1160		    bhsr2t->bhsr2t_target_transfer_tag;
1161		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1162		bhsdo->bhsdo_buffer_offset = htonl(off);
1163		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1164		    M_NOWAIT);
1165		if (error != 0) {
1166			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1167			    "reconnecting");
1168			icl_pdu_free(request);
1169			icl_pdu_free(response);
1170			iscsi_session_reconnect(is);
1171			return;
1172		}
1173
1174		off += len;
1175		total_len -= len;
1176
1177		if (total_len == 0) {
1178			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1179			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1180		} else {
1181			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1182		}
1183
1184		iscsi_pdu_queue_locked(request);
1185
1186		if (total_len == 0)
1187			break;
1188	}
1189
1190	icl_pdu_free(response);
1191}
1192
1193static void
1194iscsi_pdu_handle_async_message(struct icl_pdu *response)
1195{
1196	struct iscsi_bhs_asynchronous_message *bhsam;
1197	struct iscsi_session *is;
1198
1199	is = PDU_SESSION(response);
1200	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1201	switch (bhsam->bhsam_async_event) {
1202	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1203		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1204		iscsi_session_logout(is);
1205		iscsi_session_terminate(is);
1206		break;
1207	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1208		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the connection");
1209		break;
1210	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1211		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the session");
1212		break;
1213	default:
1214		/*
1215		 * XXX: Technically, we're obligated to also handle
1216		 * 	parameter renegotiation.
1217		 */
1218		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1219		break;
1220	}
1221
1222	icl_pdu_free(response);
1223}
1224
1225static void
1226iscsi_pdu_handle_reject(struct icl_pdu *response)
1227{
1228	struct iscsi_bhs_reject *bhsr;
1229	struct iscsi_session *is;
1230
1231	is = PDU_SESSION(response);
1232	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1233	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1234	    bhsr->bhsr_reason);
1235
1236	icl_pdu_free(response);
1237}
1238
1239static int
1240iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1241    struct iscsi_daemon_request *request)
1242{
1243	struct iscsi_session *is;
1244	int error;
1245
1246	sx_slock(&sc->sc_lock);
1247	for (;;) {
1248		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1249			ISCSI_SESSION_LOCK(is);
1250			if (is->is_waiting_for_iscsid)
1251				break;
1252			ISCSI_SESSION_UNLOCK(is);
1253		}
1254
1255		if (is == NULL) {
1256			/*
1257			 * No session requires attention from iscsid(8); wait.
1258			 */
1259			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1260			if (error != 0) {
1261				sx_sunlock(&sc->sc_lock);
1262				return (error);
1263			}
1264			continue;
1265		}
1266
1267		is->is_waiting_for_iscsid = false;
1268		is->is_login_phase = true;
1269		is->is_reason[0] = '\0';
1270		ISCSI_SESSION_UNLOCK(is);
1271
1272		request->idr_session_id = is->is_id;
1273		memcpy(&request->idr_isid, &is->is_isid,
1274		    sizeof(request->idr_isid));
1275		request->idr_tsih = 0;	/* New or reinstated session. */
1276		memcpy(&request->idr_conf, &is->is_conf,
1277		    sizeof(request->idr_conf));
1278
1279		sx_sunlock(&sc->sc_lock);
1280		return (0);
1281	}
1282}
1283
1284static int
1285iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1286    struct iscsi_daemon_handoff *handoff)
1287{
1288	struct iscsi_session *is;
1289	int error;
1290
1291	sx_slock(&sc->sc_lock);
1292
1293	/*
1294	 * Find the session to hand off socket to.
1295	 */
1296	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1297		if (is->is_id == handoff->idh_session_id)
1298			break;
1299	}
1300	if (is == NULL) {
1301		sx_sunlock(&sc->sc_lock);
1302		return (ESRCH);
1303	}
1304	ISCSI_SESSION_LOCK(is);
1305	if (is->is_conf.isc_discovery || is->is_terminating) {
1306		ISCSI_SESSION_UNLOCK(is);
1307		sx_sunlock(&sc->sc_lock);
1308		return (EINVAL);
1309	}
1310	if (is->is_connected) {
1311		/*
1312		 * This might have happened because another iscsid(8)
1313		 * instance handed off the connection in the meantime.
1314		 * Just return.
1315		 */
1316		ISCSI_SESSION_WARN(is, "handoff on already connected "
1317		    "session");
1318		ISCSI_SESSION_UNLOCK(is);
1319		sx_sunlock(&sc->sc_lock);
1320		return (EBUSY);
1321	}
1322
1323	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1324	    sizeof(is->is_target_alias));
1325	is->is_tsih = handoff->idh_tsih;
1326	is->is_statsn = handoff->idh_statsn;
1327	is->is_initial_r2t = handoff->idh_initial_r2t;
1328	is->is_immediate_data = handoff->idh_immediate_data;
1329	is->is_max_data_segment_length = handoff->idh_max_data_segment_length;
1330	is->is_max_burst_length = handoff->idh_max_burst_length;
1331	is->is_first_burst_length = handoff->idh_first_burst_length;
1332
1333	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1334		is->is_conn->ic_header_crc32c = true;
1335	else
1336		is->is_conn->ic_header_crc32c = false;
1337	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1338		is->is_conn->ic_data_crc32c = true;
1339	else
1340		is->is_conn->ic_data_crc32c = false;
1341
1342	is->is_cmdsn = 0;
1343	is->is_expcmdsn = 0;
1344	is->is_maxcmdsn = 0;
1345	is->is_waiting_for_iscsid = false;
1346	is->is_login_phase = false;
1347	is->is_timeout = 0;
1348	is->is_connected = true;
1349	is->is_reason[0] = '\0';
1350
1351	ISCSI_SESSION_UNLOCK(is);
1352
1353#ifdef ICL_KERNEL_PROXY
1354	if (handoff->idh_socket != 0) {
1355#endif
1356		/*
1357		 * Handoff without using ICL proxy.
1358		 */
1359		error = icl_conn_handoff(is->is_conn, handoff->idh_socket);
1360		if (error != 0) {
1361			sx_sunlock(&sc->sc_lock);
1362			iscsi_session_terminate(is);
1363			return (error);
1364		}
1365#ifdef ICL_KERNEL_PROXY
1366	}
1367#endif
1368
1369	sx_sunlock(&sc->sc_lock);
1370
1371	if (is->is_sim != NULL) {
1372		/*
1373		 * When reconnecting, there already is SIM allocated for the session.
1374		 */
1375		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1376		ISCSI_SESSION_LOCK(is);
1377		ISCSI_SESSION_DEBUG(is, "releasing");
1378		xpt_release_simq(is->is_sim, 1);
1379		is->is_simq_frozen = false;
1380		ISCSI_SESSION_UNLOCK(is);
1381
1382	} else {
1383		ISCSI_SESSION_LOCK(is);
1384		is->is_devq = cam_simq_alloc(maxtags);
1385		if (is->is_devq == NULL) {
1386			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1387			iscsi_session_terminate(is);
1388			return (ENOMEM);
1389		}
1390
1391		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1392		    is, is->is_id /* unit */, &is->is_lock,
1393		    1, maxtags, is->is_devq);
1394		if (is->is_sim == NULL) {
1395			ISCSI_SESSION_UNLOCK(is);
1396			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1397			cam_simq_free(is->is_devq);
1398			iscsi_session_terminate(is);
1399			return (ENOMEM);
1400		}
1401
1402		error = xpt_bus_register(is->is_sim, NULL, 0);
1403		if (error != 0) {
1404			ISCSI_SESSION_UNLOCK(is);
1405			ISCSI_SESSION_WARN(is, "failed to register bus");
1406			iscsi_session_terminate(is);
1407			return (ENOMEM);
1408		}
1409
1410		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1411		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1412		    CAM_LUN_WILDCARD);
1413		if (error != CAM_REQ_CMP) {
1414			ISCSI_SESSION_UNLOCK(is);
1415			ISCSI_SESSION_WARN(is, "failed to create path");
1416			iscsi_session_terminate(is);
1417			return (ENOMEM);
1418		}
1419		ISCSI_SESSION_UNLOCK(is);
1420	}
1421
1422	return (0);
1423}
1424
1425static int
1426iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1427    struct iscsi_daemon_fail *fail)
1428{
1429	struct iscsi_session *is;
1430
1431	sx_slock(&sc->sc_lock);
1432
1433	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1434		if (is->is_id == fail->idf_session_id)
1435			break;
1436	}
1437	if (is == NULL) {
1438		sx_sunlock(&sc->sc_lock);
1439		return (ESRCH);
1440	}
1441	ISCSI_SESSION_LOCK(is);
1442	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1443	    fail->idf_reason);
1444	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1445	//is->is_waiting_for_iscsid = false;
1446	//is->is_login_phase = true;
1447	//iscsi_session_reconnect(is);
1448	ISCSI_SESSION_UNLOCK(is);
1449	sx_sunlock(&sc->sc_lock);
1450
1451	return (0);
1452}
1453
1454#ifdef ICL_KERNEL_PROXY
1455static int
1456iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1457    struct iscsi_daemon_connect *idc)
1458{
1459	struct iscsi_session *is;
1460	struct sockaddr *from_sa, *to_sa;
1461	int error;
1462
1463	sx_slock(&sc->sc_lock);
1464	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1465		if (is->is_id == idc->idc_session_id)
1466			break;
1467	}
1468	if (is == NULL) {
1469		sx_sunlock(&sc->sc_lock);
1470		return (ESRCH);
1471	}
1472	sx_sunlock(&sc->sc_lock);
1473
1474	if (idc->idc_from_addrlen > 0) {
1475		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1476		if (error != 0) {
1477			ISCSI_SESSION_WARN(is,
1478			    "getsockaddr failed with error %d", error);
1479			return (error);
1480		}
1481	} else {
1482		from_sa = NULL;
1483	}
1484	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1485	if (error != 0) {
1486		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1487		    error);
1488		free(from_sa, M_SONAME);
1489		return (error);
1490	}
1491
1492	ISCSI_SESSION_LOCK(is);
1493	is->is_waiting_for_iscsid = false;
1494	is->is_login_phase = true;
1495	is->is_timeout = 0;
1496	ISCSI_SESSION_UNLOCK(is);
1497
1498	error = icl_conn_connect(is->is_conn, idc->idc_iser, idc->idc_domain,
1499	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1500	free(from_sa, M_SONAME);
1501	free(to_sa, M_SONAME);
1502
1503	/*
1504	 * Digests are always disabled during login phase.
1505	 */
1506	is->is_conn->ic_header_crc32c = false;
1507	is->is_conn->ic_data_crc32c = false;
1508
1509	return (error);
1510}
1511
1512static int
1513iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1514    struct iscsi_daemon_send *ids)
1515{
1516	struct iscsi_session *is;
1517	struct icl_pdu *ip;
1518	size_t datalen;
1519	void *data;
1520	int error;
1521
1522	sx_slock(&sc->sc_lock);
1523	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1524		if (is->is_id == ids->ids_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	if (is->is_terminating || is->is_reconnecting)
1537		return (EIO);
1538
1539	datalen = ids->ids_data_segment_len;
1540	if (datalen > ISCSI_MAX_DATA_SEGMENT_LENGTH)
1541		return (EINVAL);
1542	if (datalen > 0) {
1543		data = malloc(datalen, M_ISCSI, M_WAITOK);
1544		error = copyin(ids->ids_data_segment, data, datalen);
1545		if (error != 0) {
1546			free(data, M_ISCSI);
1547			return (error);
1548		}
1549	}
1550
1551	ip = icl_pdu_new_bhs(is->is_conn, M_WAITOK);
1552	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1553	if (datalen > 0) {
1554		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1555		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1556		free(data, M_ISCSI);
1557	}
1558	icl_pdu_queue(ip);
1559
1560	return (0);
1561}
1562
1563static int
1564iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1565    struct iscsi_daemon_receive *idr)
1566{
1567	struct iscsi_session *is;
1568	struct icl_pdu *ip;
1569	void *data;
1570
1571	sx_slock(&sc->sc_lock);
1572	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1573		if (is->is_id == idr->idr_session_id)
1574			break;
1575	}
1576	if (is == NULL) {
1577		sx_sunlock(&sc->sc_lock);
1578		return (ESRCH);
1579	}
1580	sx_sunlock(&sc->sc_lock);
1581
1582	if (is->is_login_phase == false)
1583		return (EBUSY);
1584
1585	ISCSI_SESSION_LOCK(is);
1586	while (is->is_login_pdu == NULL &&
1587	    is->is_terminating == false &&
1588	    is->is_reconnecting == false)
1589		cv_wait(&is->is_login_cv, &is->is_lock);
1590	if (is->is_terminating || is->is_reconnecting) {
1591		ISCSI_SESSION_UNLOCK(is);
1592		return (EIO);
1593	}
1594	ip = is->is_login_pdu;
1595	is->is_login_pdu = NULL;
1596	ISCSI_SESSION_UNLOCK(is);
1597
1598	if (ip->ip_data_len > idr->idr_data_segment_len) {
1599		icl_pdu_free(ip);
1600		return (EMSGSIZE);
1601	}
1602
1603	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1604	if (ip->ip_data_len > 0) {
1605		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1606		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1607		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1608		free(data, M_ISCSI);
1609	}
1610
1611	icl_pdu_free(ip);
1612
1613	return (0);
1614}
1615#endif /* ICL_KERNEL_PROXY */
1616
1617static void
1618iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1619{
1620	/*
1621	 * Just make sure all the fields are null-terminated.
1622	 *
1623	 * XXX: This is not particularly secure.  We should
1624	 * 	create our own conf and then copy in relevant
1625	 * 	fields.
1626	 */
1627	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1628	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1629	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1630	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1631	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1632	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1633	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1634	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1635	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1636}
1637
1638static bool
1639iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1640{
1641
1642	if (isc->isc_initiator[0] == '\0') {
1643		ISCSI_DEBUG("empty isc_initiator");
1644		return (false);
1645	}
1646
1647	if (isc->isc_target_addr[0] == '\0') {
1648		ISCSI_DEBUG("empty isc_target_addr");
1649		return (false);
1650	}
1651
1652	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1653		ISCSI_DEBUG("non-empty isc_target for discovery session");
1654		return (false);
1655	}
1656
1657	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1658		ISCSI_DEBUG("empty isc_target for non-discovery session");
1659		return (false);
1660	}
1661
1662	return (true);
1663}
1664
1665static int
1666iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1667{
1668	struct iscsi_session *is;
1669	const struct iscsi_session *is2;
1670	int error;
1671
1672	iscsi_sanitize_session_conf(&isa->isa_conf);
1673	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1674		return (EINVAL);
1675
1676	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1677	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1678
1679	sx_xlock(&sc->sc_lock);
1680
1681	/*
1682	 * Prevent duplicates.
1683	 */
1684	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1685		if (!!is->is_conf.isc_discovery !=
1686		    !!is2->is_conf.isc_discovery)
1687			continue;
1688
1689		if (strcmp(is->is_conf.isc_target_addr,
1690		    is2->is_conf.isc_target_addr) != 0)
1691			continue;
1692
1693		if (is->is_conf.isc_discovery == 0 &&
1694		    strcmp(is->is_conf.isc_target,
1695		    is2->is_conf.isc_target) != 0)
1696			continue;
1697
1698		sx_xunlock(&sc->sc_lock);
1699		free(is, M_ISCSI);
1700		return (EBUSY);
1701	}
1702
1703	is->is_conn = icl_conn_new("iscsi", &is->is_lock);
1704	is->is_conn->ic_receive = iscsi_receive_callback;
1705	is->is_conn->ic_error = iscsi_error_callback;
1706	is->is_conn->ic_prv0 = is;
1707	TAILQ_INIT(&is->is_outstanding);
1708	STAILQ_INIT(&is->is_postponed);
1709	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1710	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1711#ifdef ICL_KERNEL_PROXY
1712	cv_init(&is->is_login_cv, "iscsi_login");
1713#endif
1714
1715	is->is_softc = sc;
1716	sc->sc_last_session_id++;
1717	is->is_id = sc->sc_last_session_id;
1718	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1719	arc4rand(&is->is_isid[1], 5, 0);
1720	is->is_tsih = 0;
1721	callout_init(&is->is_callout, 1);
1722	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1723	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1724
1725	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1726	if (error != 0) {
1727		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1728		return (error);
1729	}
1730
1731	/*
1732	 * Trigger immediate reconnection.
1733	 */
1734	ISCSI_SESSION_LOCK(is);
1735	is->is_waiting_for_iscsid = true;
1736	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1737	ISCSI_SESSION_UNLOCK(is);
1738	cv_signal(&sc->sc_cv);
1739
1740	sx_xunlock(&sc->sc_lock);
1741
1742	return (0);
1743}
1744
1745static bool
1746iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1747    unsigned int id2, const struct iscsi_session_conf *c2)
1748{
1749	if (id2 == 0 && c2->isc_target[0] == '\0' &&
1750	    c2->isc_target_addr[0] == '\0')
1751		return (true);
1752	if (id2 != 0 && id2 == id1)
1753		return (true);
1754	if (c2->isc_target[0] != '\0' &&
1755	    strcmp(c1->isc_target, c2->isc_target) == 0)
1756		return (true);
1757	if (c2->isc_target_addr[0] != '\0' &&
1758	    strcmp(c1->isc_target_addr, c2->isc_target_addr) == 0)
1759		return (true);
1760	return (false);
1761}
1762
1763static int
1764iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1765    struct iscsi_session_remove *isr)
1766{
1767	struct iscsi_session *is, *tmp;
1768	bool found = false;
1769
1770	iscsi_sanitize_session_conf(&isr->isr_conf);
1771
1772	sx_xlock(&sc->sc_lock);
1773	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1774		ISCSI_SESSION_LOCK(is);
1775		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1776		    isr->isr_session_id, &isr->isr_conf)) {
1777			found = true;
1778			iscsi_session_logout(is);
1779			iscsi_session_terminate(is);
1780		}
1781		ISCSI_SESSION_UNLOCK(is);
1782	}
1783	sx_xunlock(&sc->sc_lock);
1784
1785	if (!found)
1786		return (ESRCH);
1787
1788	return (0);
1789}
1790
1791static int
1792iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1793{
1794	int error;
1795	unsigned int i = 0;
1796	struct iscsi_session *is;
1797	struct iscsi_session_state iss;
1798
1799	sx_slock(&sc->sc_lock);
1800	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1801		if (i >= isl->isl_nentries) {
1802			sx_sunlock(&sc->sc_lock);
1803			return (EMSGSIZE);
1804		}
1805		memset(&iss, 0, sizeof(iss));
1806		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1807		iss.iss_id = is->is_id;
1808		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1809		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1810
1811		if (is->is_conn->ic_header_crc32c)
1812			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1813		else
1814			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1815
1816		if (is->is_conn->ic_data_crc32c)
1817			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1818		else
1819			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1820
1821		iss.iss_max_data_segment_length = is->is_max_data_segment_length;
1822		iss.iss_immediate_data = is->is_immediate_data;
1823		iss.iss_connected = is->is_connected;
1824
1825		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1826		if (error != 0) {
1827			sx_sunlock(&sc->sc_lock);
1828			return (error);
1829		}
1830		i++;
1831	}
1832	sx_sunlock(&sc->sc_lock);
1833
1834	isl->isl_nentries = i;
1835
1836	return (0);
1837}
1838
1839static int
1840iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1841    struct iscsi_session_modify *ism)
1842{
1843	struct iscsi_session *is;
1844
1845	iscsi_sanitize_session_conf(&ism->ism_conf);
1846	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1847		return (EINVAL);
1848
1849	sx_xlock(&sc->sc_lock);
1850	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1851		ISCSI_SESSION_LOCK(is);
1852		if (is->is_id == ism->ism_session_id)
1853			break;
1854		ISCSI_SESSION_UNLOCK(is);
1855	}
1856	if (is == NULL) {
1857		sx_xunlock(&sc->sc_lock);
1858		return (ESRCH);
1859	}
1860	sx_xunlock(&sc->sc_lock);
1861
1862	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
1863	ISCSI_SESSION_UNLOCK(is);
1864
1865	iscsi_session_reconnect(is);
1866
1867	return (0);
1868}
1869
1870static int
1871iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
1872    struct thread *td)
1873{
1874	struct iscsi_softc *sc;
1875
1876	sc = dev->si_drv1;
1877
1878	switch (cmd) {
1879	case ISCSIDWAIT:
1880		return (iscsi_ioctl_daemon_wait(sc,
1881		    (struct iscsi_daemon_request *)arg));
1882	case ISCSIDHANDOFF:
1883		return (iscsi_ioctl_daemon_handoff(sc,
1884		    (struct iscsi_daemon_handoff *)arg));
1885	case ISCSIDFAIL:
1886		return (iscsi_ioctl_daemon_fail(sc,
1887		    (struct iscsi_daemon_fail *)arg));
1888#ifdef ICL_KERNEL_PROXY
1889	case ISCSIDCONNECT:
1890		return (iscsi_ioctl_daemon_connect(sc,
1891		    (struct iscsi_daemon_connect *)arg));
1892	case ISCSIDSEND:
1893		return (iscsi_ioctl_daemon_send(sc,
1894		    (struct iscsi_daemon_send *)arg));
1895	case ISCSIDRECEIVE:
1896		return (iscsi_ioctl_daemon_receive(sc,
1897		    (struct iscsi_daemon_receive *)arg));
1898#endif /* ICL_KERNEL_PROXY */
1899	case ISCSISADD:
1900		return (iscsi_ioctl_session_add(sc,
1901		    (struct iscsi_session_add *)arg));
1902	case ISCSISREMOVE:
1903		return (iscsi_ioctl_session_remove(sc,
1904		    (struct iscsi_session_remove *)arg));
1905	case ISCSISLIST:
1906		return (iscsi_ioctl_session_list(sc,
1907		    (struct iscsi_session_list *)arg));
1908	case ISCSISMODIFY:
1909		return (iscsi_ioctl_session_modify(sc,
1910		    (struct iscsi_session_modify *)arg));
1911	default:
1912		return (EINVAL);
1913	}
1914}
1915
1916static uint64_t
1917iscsi_encode_lun(uint32_t lun)
1918{
1919	uint8_t encoded[8];
1920	uint64_t result;
1921
1922	memset(encoded, 0, sizeof(encoded));
1923
1924	if (lun < 256) {
1925		/*
1926		 * Peripheral device addressing.
1927		 */
1928		encoded[1] = lun;
1929	} else if (lun < 16384) {
1930		/*
1931		 * Flat space addressing.
1932		 */
1933		encoded[0] = 0x40;
1934		encoded[0] |= (lun >> 8) & 0x3f;
1935		encoded[1] = lun & 0xff;
1936	} else {
1937		/*
1938		 * Extended flat space addressing.
1939		 */
1940		encoded[0] = 0xd2;
1941		encoded[1] = lun >> 16;
1942		encoded[2] = lun >> 8;
1943		encoded[3] = lun;
1944	}
1945
1946	memcpy(&result, encoded, sizeof(result));
1947	return (result);
1948}
1949
1950static struct iscsi_outstanding *
1951iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
1952{
1953	struct iscsi_outstanding *io;
1954
1955	ISCSI_SESSION_LOCK_ASSERT(is);
1956
1957	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1958		if (io->io_initiator_task_tag == initiator_task_tag)
1959			return (io);
1960	}
1961	return (NULL);
1962}
1963
1964static struct iscsi_outstanding *
1965iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
1966{
1967	struct iscsi_outstanding *io;
1968
1969	ISCSI_SESSION_LOCK_ASSERT(is);
1970
1971	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1972		if (io->io_ccb == ccb)
1973			return (io);
1974	}
1975	return (NULL);
1976}
1977
1978static struct iscsi_outstanding *
1979iscsi_outstanding_add(struct iscsi_session *is,
1980    uint32_t initiator_task_tag, union ccb *ccb)
1981{
1982	struct iscsi_outstanding *io;
1983
1984	ISCSI_SESSION_LOCK_ASSERT(is);
1985
1986	KASSERT(iscsi_outstanding_find(is, initiator_task_tag) == NULL,
1987	    ("initiator_task_tag 0x%x already added", initiator_task_tag));
1988
1989	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
1990	if (io == NULL) {
1991		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes", sizeof(*io));
1992		return (NULL);
1993	}
1994	io->io_initiator_task_tag = initiator_task_tag;
1995	io->io_ccb = ccb;
1996	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
1997	return (io);
1998}
1999
2000static void
2001iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2002{
2003
2004	ISCSI_SESSION_LOCK_ASSERT(is);
2005
2006	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2007	uma_zfree(iscsi_outstanding_zone, io);
2008}
2009
2010static void
2011iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2012{
2013	struct icl_pdu *request;
2014	struct iscsi_bhs_task_management_request *bhstmr;
2015	struct ccb_abort *cab = &ccb->cab;
2016	struct iscsi_outstanding *io, *aio;
2017
2018	ISCSI_SESSION_LOCK_ASSERT(is);
2019
2020#if 0
2021	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2022#else
2023	if (is->is_login_phase) {
2024		ccb->ccb_h.status = CAM_REQ_ABORTED;
2025		xpt_done(ccb);
2026		return;
2027	}
2028#endif
2029
2030	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2031	if (aio == NULL) {
2032		ccb->ccb_h.status = CAM_REQ_CMP;
2033		xpt_done(ccb);
2034		return;
2035	}
2036
2037	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
2038	if (request == NULL) {
2039		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2040		xpt_done(ccb);
2041		return;
2042	}
2043
2044	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2045	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2046	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2047
2048	bhstmr->bhstmr_lun = iscsi_encode_lun(ccb->ccb_h.target_lun);
2049	bhstmr->bhstmr_initiator_task_tag = is->is_initiator_task_tag;
2050	is->is_initiator_task_tag++;
2051	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2052
2053	io = iscsi_outstanding_add(is, bhstmr->bhstmr_initiator_task_tag, NULL);
2054	if (io == NULL) {
2055		icl_pdu_free(request);
2056		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2057		xpt_done(ccb);
2058		return;
2059	}
2060	io->io_datasn = aio->io_initiator_task_tag;
2061	iscsi_pdu_queue_locked(request);
2062}
2063
2064static void
2065iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2066{
2067	struct icl_pdu *request;
2068	struct iscsi_bhs_scsi_command *bhssc;
2069	struct ccb_scsiio *csio;
2070	struct iscsi_outstanding *io;
2071	size_t len;
2072	int error;
2073
2074	ISCSI_SESSION_LOCK_ASSERT(is);
2075
2076#if 0
2077	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2078#else
2079	if (is->is_login_phase) {
2080		ISCSI_SESSION_DEBUG(is, "called during login phase");
2081		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2082			xpt_freeze_devq(ccb->ccb_h.path, 1);
2083			ISCSI_SESSION_DEBUG(is, "freezing devq");
2084		}
2085		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2086		xpt_done(ccb);
2087		return;
2088	}
2089#endif
2090
2091	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
2092	if (request == NULL) {
2093		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2094			xpt_freeze_devq(ccb->ccb_h.path, 1);
2095			ISCSI_SESSION_DEBUG(is, "freezing devq");
2096		}
2097		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2098		xpt_done(ccb);
2099		return;
2100	}
2101
2102	csio = &ccb->csio;
2103	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2104	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2105	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2106	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2107	case CAM_DIR_IN:
2108		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2109		break;
2110	case CAM_DIR_OUT:
2111		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2112		break;
2113	}
2114
2115	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2116		switch (csio->tag_action) {
2117		case MSG_HEAD_OF_Q_TAG:
2118			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2119			break;
2120		case MSG_ORDERED_Q_TAG:
2121			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2122			break;
2123		case MSG_ACA_TASK:
2124			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2125			break;
2126		case MSG_SIMPLE_Q_TAG:
2127		default:
2128			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2129			break;
2130		}
2131	} else
2132		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2133
2134	bhssc->bhssc_lun = iscsi_encode_lun(csio->ccb_h.target_lun);
2135	bhssc->bhssc_initiator_task_tag = is->is_initiator_task_tag;
2136	is->is_initiator_task_tag++;
2137	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2138	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2139	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2140
2141	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2142		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2143	else
2144		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2145
2146	io = iscsi_outstanding_add(is, bhssc->bhssc_initiator_task_tag, ccb);
2147	if (io == NULL) {
2148		icl_pdu_free(request);
2149		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2150			xpt_freeze_devq(ccb->ccb_h.path, 1);
2151			ISCSI_SESSION_DEBUG(is, "freezing devq");
2152		}
2153		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2154		xpt_done(ccb);
2155		return;
2156	}
2157
2158	if (is->is_immediate_data &&
2159	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2160		len = csio->dxfer_len;
2161		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2162		if (len > is->is_first_burst_length) {
2163			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_first_burst_length);
2164			len = is->is_first_burst_length;
2165		}
2166
2167		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2168		if (error != 0) {
2169			icl_pdu_free(request);
2170			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2171				xpt_freeze_devq(ccb->ccb_h.path, 1);
2172				ISCSI_SESSION_DEBUG(is, "freezing devq");
2173			}
2174			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2175			xpt_done(ccb);
2176			return;
2177		}
2178	}
2179	iscsi_pdu_queue_locked(request);
2180}
2181
2182static void
2183iscsi_action(struct cam_sim *sim, union ccb *ccb)
2184{
2185	struct iscsi_session *is;
2186
2187	is = cam_sim_softc(sim);
2188
2189	ISCSI_SESSION_LOCK_ASSERT(is);
2190
2191	if (is->is_terminating ||
2192	    (is->is_connected == false && fail_on_disconnection)) {
2193		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2194		xpt_done(ccb);
2195		return;
2196	}
2197
2198	switch (ccb->ccb_h.func_code) {
2199	case XPT_PATH_INQ:
2200	{
2201		struct ccb_pathinq *cpi = &ccb->cpi;
2202
2203		cpi->version_num = 1;
2204		cpi->hba_inquiry = PI_TAG_ABLE;
2205		cpi->target_sprt = 0;
2206		//cpi->hba_misc = PIM_NOBUSRESET;
2207		cpi->hba_misc = 0;
2208		cpi->hba_eng_cnt = 0;
2209		cpi->max_target = 0;
2210		cpi->max_lun = 255;
2211		//cpi->initiator_id = 0; /* XXX */
2212		cpi->initiator_id = 64; /* XXX */
2213		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2214		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2215		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2216		cpi->unit_number = cam_sim_unit(sim);
2217		cpi->bus_id = cam_sim_bus(sim);
2218		cpi->base_transfer_speed = 150000; /* XXX */
2219		cpi->transport = XPORT_ISCSI;
2220		cpi->transport_version = 0;
2221		cpi->protocol = PROTO_SCSI;
2222		cpi->protocol_version = SCSI_REV_SPC3;
2223		cpi->maxio = MAXPHYS;
2224		cpi->ccb_h.status = CAM_REQ_CMP;
2225		break;
2226	}
2227	case XPT_GET_TRAN_SETTINGS:
2228	{
2229		struct ccb_trans_settings	*cts;
2230		struct ccb_trans_settings_scsi	*scsi;
2231
2232		cts = &ccb->cts;
2233		scsi = &cts->proto_specific.scsi;
2234
2235		cts->protocol = PROTO_SCSI;
2236		cts->protocol_version = SCSI_REV_SPC3;
2237		cts->transport = XPORT_ISCSI;
2238		cts->transport_version = 0;
2239		scsi->valid = CTS_SCSI_VALID_TQ;
2240		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2241		cts->ccb_h.status = CAM_REQ_CMP;
2242		break;
2243	}
2244	case XPT_CALC_GEOMETRY:
2245		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2246		ccb->ccb_h.status = CAM_REQ_CMP;
2247		break;
2248#if 0
2249	/*
2250	 * XXX: What's the point?
2251	 */
2252	case XPT_RESET_BUS:
2253	case XPT_TERM_IO:
2254		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2255		ccb->ccb_h.status = CAM_REQ_CMP;
2256		break;
2257#endif
2258	case XPT_ABORT:
2259		iscsi_action_abort(is, ccb);
2260		return;
2261	case XPT_SCSI_IO:
2262		iscsi_action_scsiio(is, ccb);
2263		return;
2264	default:
2265#if 0
2266		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2267#endif
2268		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2269		break;
2270	}
2271	xpt_done(ccb);
2272}
2273
2274static void
2275iscsi_poll(struct cam_sim *sim)
2276{
2277
2278	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2279}
2280
2281static void
2282iscsi_shutdown(struct iscsi_softc *sc)
2283{
2284	struct iscsi_session *is;
2285
2286	ISCSI_DEBUG("removing all sessions due to shutdown");
2287
2288	sx_slock(&sc->sc_lock);
2289	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2290		iscsi_session_terminate(is);
2291	sx_sunlock(&sc->sc_lock);
2292}
2293
2294static int
2295iscsi_load(void)
2296{
2297	int error;
2298
2299	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2300	sx_init(&sc->sc_lock, "iscsi");
2301	TAILQ_INIT(&sc->sc_sessions);
2302	cv_init(&sc->sc_cv, "iscsi_cv");
2303
2304	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2305	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2306	    UMA_ALIGN_PTR, 0);
2307
2308	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2309	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2310	if (error != 0) {
2311		ISCSI_WARN("failed to create device node, error %d", error);
2312		return (error);
2313	}
2314	sc->sc_cdev->si_drv1 = sc;
2315
2316	/*
2317	 * Note that this needs to get run before dashutdown().  Otherwise,
2318	 * when rebooting with iSCSI session with outstanding requests,
2319	 * but disconnected, dashutdown() will hang on cam_periph_runccb().
2320	 */
2321	sc->sc_shutdown_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2322	    iscsi_shutdown, sc, SHUTDOWN_PRI_FIRST);
2323
2324	return (0);
2325}
2326
2327static int
2328iscsi_unload(void)
2329{
2330	struct iscsi_session *is, *tmp;
2331
2332	if (sc->sc_cdev != NULL) {
2333		ISCSI_DEBUG("removing device node");
2334		destroy_dev(sc->sc_cdev);
2335		ISCSI_DEBUG("device node removed");
2336	}
2337
2338	if (sc->sc_shutdown_eh != NULL)
2339		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_eh);
2340
2341	sx_slock(&sc->sc_lock);
2342	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp)
2343		iscsi_session_terminate(is);
2344	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2345		ISCSI_DEBUG("waiting for sessions to terminate");
2346		cv_wait(&sc->sc_cv, &sc->sc_lock);
2347	}
2348	ISCSI_DEBUG("all sessions terminated");
2349	sx_sunlock(&sc->sc_lock);
2350
2351	uma_zdestroy(iscsi_outstanding_zone);
2352	sx_destroy(&sc->sc_lock);
2353	cv_destroy(&sc->sc_cv);
2354	free(sc, M_ISCSI);
2355	return (0);
2356}
2357
2358static int
2359iscsi_quiesce(void)
2360{
2361	sx_slock(&sc->sc_lock);
2362	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2363		sx_sunlock(&sc->sc_lock);
2364		return (EBUSY);
2365	}
2366	sx_sunlock(&sc->sc_lock);
2367	return (0);
2368}
2369
2370static int
2371iscsi_modevent(module_t mod, int what, void *arg)
2372{
2373	int error;
2374
2375	switch (what) {
2376	case MOD_LOAD:
2377		error = iscsi_load();
2378		break;
2379	case MOD_UNLOAD:
2380		error = iscsi_unload();
2381		break;
2382	case MOD_QUIESCE:
2383		error = iscsi_quiesce();
2384		break;
2385	default:
2386		error = EINVAL;
2387		break;
2388	}
2389	return (error);
2390}
2391
2392moduledata_t iscsi_data = {
2393	"iscsi",
2394	iscsi_modevent,
2395	0
2396};
2397
2398DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2399MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2400MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2401