iscsi.c revision 270888
1/*-
2 * Copyright (c) 2012 The FreeBSD Foundation
3 * All rights reserved.
4 *
5 * This software was developed by Edward Tomasz Napierala under sponsorship
6 * from the FreeBSD Foundation.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD: stable/10/sys/dev/iscsi/iscsi.c 270888 2014-08-31 20:21:08Z trasz $");
33
34#include <sys/param.h>
35#include <sys/condvar.h>
36#include <sys/conf.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/mutex.h>
44#include <sys/module.h>
45#include <sys/sysctl.h>
46#include <sys/systm.h>
47#include <sys/sx.h>
48#include <vm/uma.h>
49
50#include <cam/cam.h>
51#include <cam/cam_ccb.h>
52#include <cam/cam_xpt.h>
53#include <cam/cam_debug.h>
54#include <cam/cam_sim.h>
55#include <cam/cam_xpt_sim.h>
56#include <cam/cam_xpt_periph.h>
57#include <cam/cam_periph.h>
58#include <cam/scsi/scsi_all.h>
59#include <cam/scsi/scsi_message.h>
60
61#include "iscsi_ioctl.h"
62#include "iscsi.h"
63#include "icl.h"
64#include "iscsi_proto.h"
65
66#ifdef ICL_KERNEL_PROXY
67#include <sys/socketvar.h>
68#endif
69
70#ifdef ICL_KERNEL_PROXY
71FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
72#endif
73
74/*
75 * XXX: This is global so the iscsi_unload() can access it.
76 * 	Think about how to do this properly.
77 */
78static struct iscsi_softc	*sc;
79
80SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD, 0, "iSCSI initiator");
81static int debug = 1;
82TUNABLE_INT("kern.iscsi.debug", &debug);
83SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
84    &debug, 0, "Enable debug messages");
85static int ping_timeout = 5;
86TUNABLE_INT("kern.iscsi.ping_timeout", &ping_timeout);
87SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
88    0, "Timeout for ping (NOP-Out) requests, in seconds");
89static int iscsid_timeout = 60;
90TUNABLE_INT("kern.iscsi.iscsid_timeout", &iscsid_timeout);
91SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
92    0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
93static int login_timeout = 60;
94TUNABLE_INT("kern.iscsi.login_timeout", &login_timeout);
95SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
96    0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
97static int maxtags = 255;
98TUNABLE_INT("kern.iscsi.maxtags", &maxtags);
99SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
100    0, "Max number of IO requests queued");
101static int fail_on_disconnection = 0;
102TUNABLE_INT("kern.iscsi.fail_on_disconnection", &fail_on_disconnection);
103SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
104    &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
105
106static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
107static uma_zone_t iscsi_outstanding_zone;
108
109#define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
110#define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
111
112#define	ISCSI_DEBUG(X, ...)						\
113	do {								\
114		if (debug > 1) 						\
115			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
116	} while (0)
117
118#define	ISCSI_WARN(X, ...)						\
119	do {								\
120		if (debug > 0) {					\
121			printf("WARNING: %s: " X "\n",			\
122			    __func__, ## __VA_ARGS__);			\
123		}							\
124	} while (0)
125
126#define	ISCSI_SESSION_DEBUG(S, X, ...)					\
127	do {								\
128		if (debug > 1) {					\
129			printf("%s: %s (%s): " X "\n",			\
130			    __func__, S->is_conf.isc_target_addr,	\
131			    S->is_conf.isc_target, ## __VA_ARGS__);	\
132		}							\
133	} while (0)
134
135#define	ISCSI_SESSION_WARN(S, X, ...)					\
136	do {								\
137		if (debug > 0) {					\
138			printf("WARNING: %s (%s): " X "\n",		\
139			    S->is_conf.isc_target_addr,			\
140			    S->is_conf.isc_target, ## __VA_ARGS__);	\
141		}							\
142	} while (0)
143
144#define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
145#define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
146#define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
147
148static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
149		    int mode, struct thread *td);
150
151static struct cdevsw iscsi_cdevsw = {
152     .d_version = D_VERSION,
153     .d_ioctl   = iscsi_ioctl,
154     .d_name    = "iscsi",
155};
156
157static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
158static void	iscsi_pdu_queue(struct icl_pdu *request);
159static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
160static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
161static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
162static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
163static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
164static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
165static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
166static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
167static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
168static void	iscsi_session_reconnect(struct iscsi_session *is);
169static void	iscsi_session_terminate(struct iscsi_session *is);
170static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
171static void	iscsi_poll(struct cam_sim *sim);
172static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
173		    uint32_t initiator_task_tag);
174static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
175		    uint32_t initiator_task_tag, union ccb *ccb);
176static void	iscsi_outstanding_remove(struct iscsi_session *is,
177		    struct iscsi_outstanding *io);
178
179static bool
180iscsi_pdu_prepare(struct icl_pdu *request)
181{
182	struct iscsi_session *is;
183	struct iscsi_bhs_scsi_command *bhssc;
184
185	is = PDU_SESSION(request);
186
187	ISCSI_SESSION_LOCK_ASSERT(is);
188
189	/*
190	 * We're only using fields common for all the request
191	 * (initiator -> target) PDUs.
192	 */
193	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
194
195	/*
196	 * Data-Out PDU does not contain CmdSN.
197	 */
198	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
199		if (is->is_cmdsn > is->is_maxcmdsn &&
200		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
201			/*
202			 * Current MaxCmdSN prevents us from sending any more
203			 * SCSI Command PDUs to the target; postpone the PDU.
204			 * It will get resent by either iscsi_pdu_queue(),
205			 * or by maintenance thread.
206			 */
207#if 0
208			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %d, ExpCmdSN %d, MaxCmdSN %d, opcode 0x%x",
209			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn, bhssc->bhssc_opcode);
210#endif
211			return (true);
212		}
213		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
214		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
215			is->is_cmdsn++;
216	}
217	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
218
219	return (false);
220}
221
222static void
223iscsi_session_send_postponed(struct iscsi_session *is)
224{
225	struct icl_pdu *request;
226	bool postpone;
227
228	ISCSI_SESSION_LOCK_ASSERT(is);
229
230	while (!STAILQ_EMPTY(&is->is_postponed)) {
231		request = STAILQ_FIRST(&is->is_postponed);
232		postpone = iscsi_pdu_prepare(request);
233		if (postpone)
234			break;
235		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
236		icl_pdu_queue(request);
237	}
238}
239
240static void
241iscsi_pdu_queue_locked(struct icl_pdu *request)
242{
243	struct iscsi_session *is;
244	bool postpone;
245
246	is = PDU_SESSION(request);
247	ISCSI_SESSION_LOCK_ASSERT(is);
248	iscsi_session_send_postponed(is);
249	postpone = iscsi_pdu_prepare(request);
250	if (postpone) {
251		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
252		return;
253	}
254	icl_pdu_queue(request);
255}
256
257static void
258iscsi_pdu_queue(struct icl_pdu *request)
259{
260	struct iscsi_session *is;
261
262	is = PDU_SESSION(request);
263	ISCSI_SESSION_LOCK(is);
264	iscsi_pdu_queue_locked(request);
265	ISCSI_SESSION_UNLOCK(is);
266}
267
268static void
269iscsi_session_logout(struct iscsi_session *is)
270{
271	struct icl_pdu *request;
272	struct iscsi_bhs_logout_request *bhslr;
273
274	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
275	if (request == NULL)
276		return;
277
278	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
279	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
280	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
281	iscsi_pdu_queue_locked(request);
282}
283
284static void
285iscsi_session_terminate_task(struct iscsi_session *is,
286    struct iscsi_outstanding *io, bool requeue)
287{
288
289	if (io->io_ccb != NULL) {
290		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
291		if (requeue)
292			io->io_ccb->ccb_h.status |= CAM_REQUEUE_REQ;
293		else
294			io->io_ccb->ccb_h.status |= CAM_REQ_ABORTED;
295		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
296			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
297			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
298			ISCSI_SESSION_DEBUG(is, "freezing devq");
299		}
300		xpt_done(io->io_ccb);
301	}
302	iscsi_outstanding_remove(is, io);
303}
304
305static void
306iscsi_session_terminate_tasks(struct iscsi_session *is, bool requeue)
307{
308	struct iscsi_outstanding *io, *tmp;
309
310	ISCSI_SESSION_LOCK_ASSERT(is);
311
312	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
313		iscsi_session_terminate_task(is, io, requeue);
314	}
315}
316
317static void
318iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
319{
320	struct icl_pdu *pdu;
321
322	ISCSI_SESSION_LOCK_ASSERT(is);
323
324	/*
325	 * Don't queue any new PDUs.
326	 */
327	if (is->is_sim != NULL && is->is_simq_frozen == false) {
328		ISCSI_SESSION_DEBUG(is, "freezing");
329		xpt_freeze_simq(is->is_sim, 1);
330		is->is_simq_frozen = true;
331	}
332
333	/*
334	 * Remove postponed PDUs.
335	 */
336	while (!STAILQ_EMPTY(&is->is_postponed)) {
337		pdu = STAILQ_FIRST(&is->is_postponed);
338		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
339		icl_pdu_free(pdu);
340	}
341
342	if (destroy_sim == false) {
343		/*
344		 * Terminate SCSI tasks, asking CAM to requeue them.
345		 */
346		iscsi_session_terminate_tasks(is, true);
347		return;
348	}
349
350	iscsi_session_terminate_tasks(is, false);
351
352	if (is->is_sim == NULL)
353		return;
354
355	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
356	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
357
358	if (is->is_simq_frozen) {
359		xpt_release_simq(is->is_sim, 1);
360		is->is_simq_frozen = false;
361	}
362
363	xpt_free_path(is->is_path);
364	is->is_path = NULL;
365	xpt_bus_deregister(cam_sim_path(is->is_sim));
366	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
367	is->is_sim = NULL;
368	is->is_devq = NULL;
369}
370
371static void
372iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
373{
374
375	icl_conn_shutdown(is->is_conn);
376	icl_conn_close(is->is_conn);
377
378	ISCSI_SESSION_LOCK(is);
379
380	is->is_connected = false;
381	is->is_reconnecting = false;
382	is->is_login_phase = false;
383
384#ifdef ICL_KERNEL_PROXY
385	if (is->is_login_pdu != NULL) {
386		icl_pdu_free(is->is_login_pdu);
387		is->is_login_pdu = NULL;
388	}
389	cv_signal(&is->is_login_cv);
390#endif
391
392	if (fail_on_disconnection) {
393		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
394		iscsi_session_cleanup(is, true);
395	} else {
396		iscsi_session_cleanup(is, false);
397	}
398
399	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
400	    ("destroying session with active tasks"));
401	KASSERT(STAILQ_EMPTY(&is->is_postponed),
402	    ("destroying session with postponed PDUs"));
403
404	/*
405	 * Request immediate reconnection from iscsid(8).
406	 */
407	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
408	is->is_waiting_for_iscsid = true;
409	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
410	is->is_timeout = 0;
411	ISCSI_SESSION_UNLOCK(is);
412	cv_signal(&is->is_softc->sc_cv);
413}
414
415static void
416iscsi_maintenance_thread_terminate(struct iscsi_session *is)
417{
418	struct iscsi_softc *sc;
419
420	sc = is->is_softc;
421	sx_xlock(&sc->sc_lock);
422	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
423	sx_xunlock(&sc->sc_lock);
424
425	icl_conn_close(is->is_conn);
426
427	ISCSI_SESSION_LOCK(is);
428
429	KASSERT(is->is_terminating, ("is_terminating == false"));
430
431#ifdef ICL_KERNEL_PROXY
432	if (is->is_login_pdu != NULL) {
433		icl_pdu_free(is->is_login_pdu);
434		is->is_login_pdu = NULL;
435	}
436	cv_signal(&is->is_login_cv);
437#endif
438
439	callout_drain(&is->is_callout);
440
441	iscsi_session_cleanup(is, true);
442
443	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
444	    ("destroying session with active tasks"));
445	KASSERT(STAILQ_EMPTY(&is->is_postponed),
446	    ("destroying session with postponed PDUs"));
447
448	ISCSI_SESSION_UNLOCK(is);
449
450	icl_conn_free(is->is_conn);
451	mtx_destroy(&is->is_lock);
452	cv_destroy(&is->is_maintenance_cv);
453#ifdef ICL_KERNEL_PROXY
454	cv_destroy(&is->is_login_cv);
455#endif
456	ISCSI_SESSION_DEBUG(is, "terminated");
457	free(is, M_ISCSI);
458
459	/*
460	 * The iscsi_unload() routine might be waiting.
461	 */
462	cv_signal(&sc->sc_cv);
463}
464
465static void
466iscsi_maintenance_thread(void *arg)
467{
468	struct iscsi_session *is;
469
470	is = arg;
471
472	for (;;) {
473		ISCSI_SESSION_LOCK(is);
474		if (is->is_reconnecting == false &&
475		    is->is_terminating == false &&
476		    STAILQ_EMPTY(&is->is_postponed))
477			cv_wait(&is->is_maintenance_cv, &is->is_lock);
478
479		if (is->is_reconnecting) {
480			ISCSI_SESSION_UNLOCK(is);
481			iscsi_maintenance_thread_reconnect(is);
482			continue;
483		}
484
485		if (is->is_terminating) {
486			ISCSI_SESSION_UNLOCK(is);
487			iscsi_maintenance_thread_terminate(is);
488			kthread_exit();
489			return;
490		}
491
492		iscsi_session_send_postponed(is);
493		ISCSI_SESSION_UNLOCK(is);
494	}
495}
496
497static void
498iscsi_session_reconnect(struct iscsi_session *is)
499{
500
501	/*
502	 * XXX: We can't use locking here, because
503	 * 	it's being called from various contexts.
504	 * 	Hope it doesn't break anything.
505	 */
506	if (is->is_reconnecting)
507		return;
508
509	is->is_reconnecting = true;
510	cv_signal(&is->is_maintenance_cv);
511}
512
513static void
514iscsi_session_terminate(struct iscsi_session *is)
515{
516	if (is->is_terminating)
517		return;
518
519	is->is_terminating = true;
520
521#if 0
522	iscsi_session_logout(is);
523#endif
524	cv_signal(&is->is_maintenance_cv);
525}
526
527static void
528iscsi_callout(void *context)
529{
530	struct icl_pdu *request;
531	struct iscsi_bhs_nop_out *bhsno;
532	struct iscsi_session *is;
533	bool reconnect_needed = false;
534
535	is = context;
536
537	if (is->is_terminating)
538		return;
539
540	callout_schedule(&is->is_callout, 1 * hz);
541
542	ISCSI_SESSION_LOCK(is);
543	is->is_timeout++;
544
545	if (is->is_waiting_for_iscsid) {
546		if (is->is_timeout > iscsid_timeout) {
547			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
548			    "for %d seconds; reconnecting",
549			    is->is_timeout);
550			reconnect_needed = true;
551		}
552		goto out;
553	}
554
555	if (is->is_login_phase) {
556		if (is->is_timeout > login_timeout) {
557			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
558			    "reconnecting", is->is_timeout);
559			reconnect_needed = true;
560		}
561		goto out;
562	}
563
564	if (is->is_timeout >= ping_timeout) {
565		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
566		    "reconnecting", ping_timeout);
567		reconnect_needed = true;
568		goto out;
569	}
570
571	ISCSI_SESSION_UNLOCK(is);
572
573	/*
574	 * If the ping was reset less than one second ago - which means
575	 * that we've received some PDU during the last second - assume
576	 * the traffic flows correctly and don't bother sending a NOP-Out.
577	 *
578	 * (It's 2 - one for one second, and one for incrementing is_timeout
579	 * earlier in this routine.)
580	 */
581	if (is->is_timeout < 2)
582		return;
583
584	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
585	if (request == NULL) {
586		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
587		return;
588	}
589	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
590	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
591	    ISCSI_BHS_OPCODE_IMMEDIATE;
592	bhsno->bhsno_flags = 0x80;
593	bhsno->bhsno_target_transfer_tag = 0xffffffff;
594	iscsi_pdu_queue(request);
595	return;
596
597out:
598	ISCSI_SESSION_UNLOCK(is);
599
600	if (reconnect_needed)
601		iscsi_session_reconnect(is);
602}
603
604static void
605iscsi_pdu_update_statsn(const struct icl_pdu *response)
606{
607	const struct iscsi_bhs_data_in *bhsdi;
608	struct iscsi_session *is;
609	uint32_t expcmdsn, maxcmdsn;
610
611	is = PDU_SESSION(response);
612
613	ISCSI_SESSION_LOCK_ASSERT(is);
614
615	/*
616	 * We're only using fields common for all the response
617	 * (target -> initiator) PDUs.
618	 */
619	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
620	/*
621	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
622	 * and Residual Count only have meaningful content if the S bit
623	 * is set to 1", so we also need to check the bit specific for
624	 * Data-In PDU.
625	 */
626	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
627	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
628		if (ntohl(bhsdi->bhsdi_statsn) < is->is_statsn) {
629			ISCSI_SESSION_WARN(is,
630			    "PDU StatSN %d >= session StatSN %d, opcode 0x%x",
631			    is->is_statsn, ntohl(bhsdi->bhsdi_statsn),
632			    bhsdi->bhsdi_opcode);
633		}
634		is->is_statsn = ntohl(bhsdi->bhsdi_statsn);
635	}
636
637	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
638	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
639
640	/*
641	 * XXX: Compare using Serial Arithmetic Sense.
642	 */
643	if (maxcmdsn + 1 < expcmdsn) {
644		ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d + 1 < PDU ExpCmdSN %d; ignoring",
645		    maxcmdsn, expcmdsn);
646	} else {
647		if (maxcmdsn > is->is_maxcmdsn) {
648			is->is_maxcmdsn = maxcmdsn;
649
650			/*
651			 * Command window increased; kick the maintanance thread
652			 * to send out postponed commands.
653			 */
654			if (!STAILQ_EMPTY(&is->is_postponed))
655				cv_signal(&is->is_maintenance_cv);
656		} else if (maxcmdsn < is->is_maxcmdsn) {
657			ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d < session MaxCmdSN %d; ignoring",
658			    maxcmdsn, is->is_maxcmdsn);
659		}
660
661		if (expcmdsn > is->is_expcmdsn) {
662			is->is_expcmdsn = expcmdsn;
663		} else if (expcmdsn < is->is_expcmdsn) {
664			ISCSI_SESSION_DEBUG(is, "PDU ExpCmdSN %d < session ExpCmdSN %d; ignoring",
665			    expcmdsn, is->is_expcmdsn);
666		}
667	}
668
669	/*
670	 * Every incoming PDU - not just NOP-In - resets the ping timer.
671	 * The purpose of the timeout is to reset the connection when it stalls;
672	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
673	 * in some queue.
674	 */
675	is->is_timeout = 0;
676}
677
678static void
679iscsi_receive_callback(struct icl_pdu *response)
680{
681	struct iscsi_session *is;
682
683	is = PDU_SESSION(response);
684
685	ISCSI_SESSION_LOCK(is);
686
687#ifdef ICL_KERNEL_PROXY
688	if (is->is_login_phase) {
689		if (is->is_login_pdu == NULL)
690			is->is_login_pdu = response;
691		else
692			icl_pdu_free(response);
693		ISCSI_SESSION_UNLOCK(is);
694		cv_signal(&is->is_login_cv);
695		return;
696	}
697#endif
698
699	iscsi_pdu_update_statsn(response);
700
701	/*
702	 * The handling routine is responsible for freeing the PDU
703	 * when it's no longer needed.
704	 */
705	switch (response->ip_bhs->bhs_opcode) {
706	case ISCSI_BHS_OPCODE_NOP_IN:
707		iscsi_pdu_handle_nop_in(response);
708		break;
709	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
710		iscsi_pdu_handle_scsi_response(response);
711		break;
712	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
713		iscsi_pdu_handle_task_response(response);
714		break;
715	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
716		iscsi_pdu_handle_data_in(response);
717		break;
718	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
719		iscsi_pdu_handle_logout_response(response);
720		break;
721	case ISCSI_BHS_OPCODE_R2T:
722		iscsi_pdu_handle_r2t(response);
723		break;
724	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
725		iscsi_pdu_handle_async_message(response);
726		break;
727	case ISCSI_BHS_OPCODE_REJECT:
728		iscsi_pdu_handle_reject(response);
729		break;
730	default:
731		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
732		    "opcode 0x%x; reconnecting",
733		    response->ip_bhs->bhs_opcode);
734		iscsi_session_reconnect(is);
735		icl_pdu_free(response);
736	}
737
738	ISCSI_SESSION_UNLOCK(is);
739}
740
741static void
742iscsi_error_callback(struct icl_conn *ic)
743{
744	struct iscsi_session *is;
745
746	is = CONN_SESSION(ic);
747
748	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
749	iscsi_session_reconnect(is);
750}
751
752static void
753iscsi_pdu_handle_nop_in(struct icl_pdu *response)
754{
755	struct iscsi_session *is;
756	struct iscsi_bhs_nop_out *bhsno;
757	struct iscsi_bhs_nop_in *bhsni;
758	struct icl_pdu *request;
759	void *data = NULL;
760	size_t datasize;
761	int error;
762
763	is = PDU_SESSION(response);
764	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
765
766	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
767		/*
768		 * Nothing to do; iscsi_pdu_update_statsn() already
769		 * zeroed the timeout.
770		 */
771		icl_pdu_free(response);
772		return;
773	}
774
775	datasize = icl_pdu_data_segment_length(response);
776	if (datasize > 0) {
777		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
778		if (data == NULL) {
779			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
780			    "reconnecting");
781			icl_pdu_free(response);
782			iscsi_session_reconnect(is);
783			return;
784		}
785		icl_pdu_get_data(response, 0, data, datasize);
786	}
787
788	request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
789	if (request == NULL) {
790		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
791		    "reconnecting");
792		free(data, M_ISCSI);
793		icl_pdu_free(response);
794		iscsi_session_reconnect(is);
795		return;
796	}
797	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
798	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
799	    ISCSI_BHS_OPCODE_IMMEDIATE;
800	bhsno->bhsno_flags = 0x80;
801	bhsno->bhsno_initiator_task_tag = 0xffffffff;
802	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
803	if (datasize > 0) {
804		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
805		if (error != 0) {
806			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
807			    "reconnecting");
808			free(data, M_ISCSI);
809			icl_pdu_free(request);
810			icl_pdu_free(response);
811			iscsi_session_reconnect(is);
812			return;
813		}
814		free(data, M_ISCSI);
815	}
816
817	icl_pdu_free(response);
818	iscsi_pdu_queue_locked(request);
819}
820
821static void
822iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
823{
824	struct iscsi_bhs_scsi_response *bhssr;
825	struct iscsi_outstanding *io;
826	struct iscsi_session *is;
827	struct ccb_scsiio *csio;
828	size_t data_segment_len;
829	uint16_t sense_len;
830
831	is = PDU_SESSION(response);
832
833	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
834	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
835	if (io == NULL || io->io_ccb == NULL) {
836		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
837		icl_pdu_free(response);
838		iscsi_session_reconnect(is);
839		return;
840	}
841
842	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
843		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
844 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
845 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
846			ISCSI_SESSION_DEBUG(is, "freezing devq");
847		}
848 		io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
849	} else if (bhssr->bhssr_status == 0) {
850		io->io_ccb->ccb_h.status = CAM_REQ_CMP;
851	} else {
852 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
853 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
854			ISCSI_SESSION_DEBUG(is, "freezing devq");
855		}
856 		io->io_ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
857		io->io_ccb->csio.scsi_status = bhssr->bhssr_status;
858	}
859
860	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_OVERFLOW) {
861		ISCSI_SESSION_WARN(is, "target indicated residual overflow");
862		icl_pdu_free(response);
863		iscsi_session_reconnect(is);
864		return;
865	}
866
867	csio = &io->io_ccb->csio;
868
869	data_segment_len = icl_pdu_data_segment_length(response);
870	if (data_segment_len > 0) {
871		if (data_segment_len < sizeof(sense_len)) {
872			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
873			    data_segment_len);
874			if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
875				xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
876				ISCSI_SESSION_DEBUG(is, "freezing devq");
877			}
878			io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
879			goto out;
880		}
881		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
882		sense_len = ntohs(sense_len);
883#if 0
884		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
885		    sense_len, data_segment_len);
886#endif
887		if (sizeof(sense_len) + sense_len > data_segment_len) {
888			ISCSI_SESSION_WARN(is, "truncated data segment "
889			    "(%zd bytes, should be %zd)",
890			    data_segment_len, sizeof(sense_len) + sense_len);
891			if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
892				xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
893				ISCSI_SESSION_DEBUG(is, "freezing devq");
894			}
895			io->io_ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
896			goto out;
897		} else if (sizeof(sense_len) + sense_len < data_segment_len)
898			ISCSI_SESSION_WARN(is, "oversize data segment "
899			    "(%zd bytes, should be %zd)",
900			    data_segment_len, sizeof(sense_len) + sense_len);
901		if (sense_len > csio->sense_len) {
902			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
903			    sense_len, csio->sense_len);
904			sense_len = csio->sense_len;
905		}
906		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
907		csio->sense_resid = csio->sense_len - sense_len;
908		io->io_ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
909	}
910
911out:
912	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
913		csio->resid = ntohl(bhssr->bhssr_residual_count);
914
915	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
916		KASSERT(io->io_received <= csio->dxfer_len,
917		    ("io->io_received > csio->dxfer_len"));
918		if (io->io_received < csio->dxfer_len) {
919			if (csio->resid != csio->dxfer_len - io->io_received) {
920				ISCSI_SESSION_WARN(is, "underflow mismatch: "
921				    "target indicates %d, we calculated %zd",
922				    csio->resid,
923				    csio->dxfer_len - io->io_received);
924			}
925			csio->resid = csio->dxfer_len - io->io_received;
926		}
927	}
928
929	xpt_done(io->io_ccb);
930	iscsi_outstanding_remove(is, io);
931	icl_pdu_free(response);
932}
933
934static void
935iscsi_pdu_handle_task_response(struct icl_pdu *response)
936{
937	struct iscsi_bhs_task_management_response *bhstmr;
938	struct iscsi_outstanding *io, *aio;
939	struct iscsi_session *is;
940
941	is = PDU_SESSION(response);
942
943	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
944	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
945	if (io == NULL || io->io_ccb != NULL) {
946		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
947		    bhstmr->bhstmr_initiator_task_tag);
948		icl_pdu_free(response);
949		iscsi_session_reconnect(is);
950		return;
951	}
952
953	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
954		ISCSI_SESSION_WARN(is, "task response 0x%x",
955		    bhstmr->bhstmr_response);
956	} else {
957		aio = iscsi_outstanding_find(is, io->io_datasn);
958		if (aio != NULL && aio->io_ccb != NULL)
959			iscsi_session_terminate_task(is, aio, false);
960	}
961
962	iscsi_outstanding_remove(is, io);
963	icl_pdu_free(response);
964}
965
966static void
967iscsi_pdu_handle_data_in(struct icl_pdu *response)
968{
969	struct iscsi_bhs_data_in *bhsdi;
970	struct iscsi_outstanding *io;
971	struct iscsi_session *is;
972	struct ccb_scsiio *csio;
973	size_t data_segment_len;
974
975	is = PDU_SESSION(response);
976	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
977	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
978	if (io == NULL || io->io_ccb == NULL) {
979		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
980		icl_pdu_free(response);
981		iscsi_session_reconnect(is);
982		return;
983	}
984
985	data_segment_len = icl_pdu_data_segment_length(response);
986	if (data_segment_len == 0) {
987		/*
988		 * "The sending of 0 length data segments should be avoided,
989		 * but initiators and targets MUST be able to properly receive
990		 * 0 length data segments."
991		 */
992		icl_pdu_free(response);
993		return;
994	}
995
996	/*
997	 * We need to track this for security reasons - without it, malicious target
998	 * could respond to SCSI READ without sending Data-In PDUs, which would result
999	 * in read operation on the initiator side returning random kernel data.
1000	 */
1001	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1002		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1003		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1004		icl_pdu_free(response);
1005		iscsi_session_reconnect(is);
1006		return;
1007	}
1008
1009	csio = &io->io_ccb->csio;
1010
1011	if (io->io_received + data_segment_len > csio->dxfer_len) {
1012		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1013		    "at offset %zd, buffer is %d)",
1014		    data_segment_len, io->io_received, csio->dxfer_len);
1015		icl_pdu_free(response);
1016		iscsi_session_reconnect(is);
1017		return;
1018	}
1019
1020	icl_pdu_get_data(response, 0, csio->data_ptr + io->io_received, data_segment_len);
1021	io->io_received += data_segment_len;
1022
1023	/*
1024	 * XXX: Check DataSN.
1025	 * XXX: Check F.
1026	 */
1027	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1028		/*
1029		 * Nothing more to do.
1030		 */
1031		icl_pdu_free(response);
1032		return;
1033	}
1034
1035	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1036	if (bhsdi->bhsdi_status == 0) {
1037		io->io_ccb->ccb_h.status = CAM_REQ_CMP;
1038	} else {
1039		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1040			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
1041			ISCSI_SESSION_DEBUG(is, "freezing devq");
1042		}
1043		io->io_ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1044		csio->scsi_status = bhsdi->bhsdi_status;
1045	}
1046
1047	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1048		KASSERT(io->io_received <= csio->dxfer_len,
1049		    ("io->io_received > csio->dxfer_len"));
1050		if (io->io_received < csio->dxfer_len) {
1051			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1052			if (csio->resid != csio->dxfer_len - io->io_received) {
1053				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1054				    "target indicates %d, we calculated %zd",
1055				    csio->resid,
1056				    csio->dxfer_len - io->io_received);
1057			}
1058			csio->resid = csio->dxfer_len - io->io_received;
1059		}
1060	}
1061
1062	xpt_done(io->io_ccb);
1063	iscsi_outstanding_remove(is, io);
1064	icl_pdu_free(response);
1065}
1066
1067static void
1068iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1069{
1070
1071	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1072	icl_pdu_free(response);
1073}
1074
1075static void
1076iscsi_pdu_handle_r2t(struct icl_pdu *response)
1077{
1078	struct icl_pdu *request;
1079	struct iscsi_session *is;
1080	struct iscsi_bhs_r2t *bhsr2t;
1081	struct iscsi_bhs_data_out *bhsdo;
1082	struct iscsi_outstanding *io;
1083	struct ccb_scsiio *csio;
1084	size_t off, len, total_len;
1085	int error;
1086
1087	is = PDU_SESSION(response);
1088
1089	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1090	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1091	if (io == NULL || io->io_ccb == NULL) {
1092		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1093		    bhsr2t->bhsr2t_initiator_task_tag);
1094		icl_pdu_free(response);
1095		iscsi_session_reconnect(is);
1096		return;
1097	}
1098
1099	csio = &io->io_ccb->csio;
1100
1101	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1102		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1103		icl_pdu_free(response);
1104		iscsi_session_reconnect(is);
1105		return;
1106	}
1107
1108	/*
1109	 * XXX: Verify R2TSN.
1110	 */
1111
1112	io->io_datasn = 0;
1113
1114	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1115	if (off > csio->dxfer_len) {
1116		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1117		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1118		icl_pdu_free(response);
1119		iscsi_session_reconnect(is);
1120		return;
1121	}
1122
1123	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1124	if (total_len == 0 || total_len > csio->dxfer_len) {
1125		ISCSI_SESSION_WARN(is, "target requested invalid length "
1126		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1127		icl_pdu_free(response);
1128		iscsi_session_reconnect(is);
1129		return;
1130	}
1131
1132	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1133
1134	for (;;) {
1135		len = total_len;
1136
1137		if (len > is->is_max_data_segment_length)
1138			len = is->is_max_data_segment_length;
1139
1140		if (off + len > csio->dxfer_len) {
1141			ISCSI_SESSION_WARN(is, "target requested invalid "
1142			    "length/offset %zd, buffer is %d; reconnecting",
1143			    off + len, csio->dxfer_len);
1144			icl_pdu_free(response);
1145			iscsi_session_reconnect(is);
1146			return;
1147		}
1148
1149		request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
1150		if (request == NULL) {
1151			icl_pdu_free(response);
1152			iscsi_session_reconnect(is);
1153			return;
1154		}
1155
1156		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1157		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1158		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1159		bhsdo->bhsdo_initiator_task_tag =
1160		    bhsr2t->bhsr2t_initiator_task_tag;
1161		bhsdo->bhsdo_target_transfer_tag =
1162		    bhsr2t->bhsr2t_target_transfer_tag;
1163		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1164		bhsdo->bhsdo_buffer_offset = htonl(off);
1165		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1166		    M_NOWAIT);
1167		if (error != 0) {
1168			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1169			    "reconnecting");
1170			icl_pdu_free(request);
1171			icl_pdu_free(response);
1172			iscsi_session_reconnect(is);
1173			return;
1174		}
1175
1176		off += len;
1177		total_len -= len;
1178
1179		if (total_len == 0) {
1180			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1181			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1182		} else {
1183			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1184		}
1185
1186		iscsi_pdu_queue_locked(request);
1187
1188		if (total_len == 0)
1189			break;
1190	}
1191
1192	icl_pdu_free(response);
1193}
1194
1195static void
1196iscsi_pdu_handle_async_message(struct icl_pdu *response)
1197{
1198	struct iscsi_bhs_asynchronous_message *bhsam;
1199	struct iscsi_session *is;
1200
1201	is = PDU_SESSION(response);
1202	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1203	switch (bhsam->bhsam_async_event) {
1204	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1205		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1206		iscsi_session_logout(is);
1207		iscsi_session_terminate(is);
1208		break;
1209	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1210		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the connection");
1211		break;
1212	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1213		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the session");
1214		break;
1215	default:
1216		/*
1217		 * XXX: Technically, we're obligated to also handle
1218		 * 	parameter renegotiation.
1219		 */
1220		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1221		break;
1222	}
1223
1224	icl_pdu_free(response);
1225}
1226
1227static void
1228iscsi_pdu_handle_reject(struct icl_pdu *response)
1229{
1230	struct iscsi_bhs_reject *bhsr;
1231	struct iscsi_session *is;
1232
1233	is = PDU_SESSION(response);
1234	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1235	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1236	    bhsr->bhsr_reason);
1237
1238	icl_pdu_free(response);
1239}
1240
1241static int
1242iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1243    struct iscsi_daemon_request *request)
1244{
1245	struct iscsi_session *is;
1246	int error;
1247
1248	sx_slock(&sc->sc_lock);
1249	for (;;) {
1250		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1251			ISCSI_SESSION_LOCK(is);
1252			if (is->is_waiting_for_iscsid)
1253				break;
1254			ISCSI_SESSION_UNLOCK(is);
1255		}
1256
1257		if (is == NULL) {
1258			/*
1259			 * No session requires attention from iscsid(8); wait.
1260			 */
1261			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1262			if (error != 0) {
1263				sx_sunlock(&sc->sc_lock);
1264				return (error);
1265			}
1266			continue;
1267		}
1268
1269		is->is_waiting_for_iscsid = false;
1270		is->is_login_phase = true;
1271		is->is_reason[0] = '\0';
1272		ISCSI_SESSION_UNLOCK(is);
1273
1274		request->idr_session_id = is->is_id;
1275		memcpy(&request->idr_isid, &is->is_isid,
1276		    sizeof(request->idr_isid));
1277		request->idr_tsih = 0;	/* New or reinstated session. */
1278		memcpy(&request->idr_conf, &is->is_conf,
1279		    sizeof(request->idr_conf));
1280
1281		sx_sunlock(&sc->sc_lock);
1282		return (0);
1283	}
1284}
1285
1286static int
1287iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1288    struct iscsi_daemon_handoff *handoff)
1289{
1290	struct iscsi_session *is;
1291	int error;
1292
1293	sx_slock(&sc->sc_lock);
1294
1295	/*
1296	 * Find the session to hand off socket to.
1297	 */
1298	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1299		if (is->is_id == handoff->idh_session_id)
1300			break;
1301	}
1302	if (is == NULL) {
1303		sx_sunlock(&sc->sc_lock);
1304		return (ESRCH);
1305	}
1306	ISCSI_SESSION_LOCK(is);
1307	if (is->is_conf.isc_discovery || is->is_terminating) {
1308		ISCSI_SESSION_UNLOCK(is);
1309		sx_sunlock(&sc->sc_lock);
1310		return (EINVAL);
1311	}
1312	if (is->is_connected) {
1313		/*
1314		 * This might have happened because another iscsid(8)
1315		 * instance handed off the connection in the meantime.
1316		 * Just return.
1317		 */
1318		ISCSI_SESSION_WARN(is, "handoff on already connected "
1319		    "session");
1320		ISCSI_SESSION_UNLOCK(is);
1321		sx_sunlock(&sc->sc_lock);
1322		return (EBUSY);
1323	}
1324
1325	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1326	    sizeof(is->is_target_alias));
1327	is->is_tsih = handoff->idh_tsih;
1328	is->is_statsn = handoff->idh_statsn;
1329	is->is_initial_r2t = handoff->idh_initial_r2t;
1330	is->is_immediate_data = handoff->idh_immediate_data;
1331	is->is_max_data_segment_length = handoff->idh_max_data_segment_length;
1332	is->is_max_burst_length = handoff->idh_max_burst_length;
1333	is->is_first_burst_length = handoff->idh_first_burst_length;
1334
1335	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1336		is->is_conn->ic_header_crc32c = true;
1337	else
1338		is->is_conn->ic_header_crc32c = false;
1339	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1340		is->is_conn->ic_data_crc32c = true;
1341	else
1342		is->is_conn->ic_data_crc32c = false;
1343
1344	is->is_cmdsn = 0;
1345	is->is_expcmdsn = 0;
1346	is->is_maxcmdsn = 0;
1347	is->is_waiting_for_iscsid = false;
1348	is->is_login_phase = false;
1349	is->is_timeout = 0;
1350	is->is_connected = true;
1351	is->is_reason[0] = '\0';
1352
1353	ISCSI_SESSION_UNLOCK(is);
1354
1355#ifdef ICL_KERNEL_PROXY
1356	if (handoff->idh_socket != 0) {
1357#endif
1358		/*
1359		 * Handoff without using ICL proxy.
1360		 */
1361		error = icl_conn_handoff(is->is_conn, handoff->idh_socket);
1362		if (error != 0) {
1363			sx_sunlock(&sc->sc_lock);
1364			iscsi_session_terminate(is);
1365			return (error);
1366		}
1367#ifdef ICL_KERNEL_PROXY
1368	}
1369#endif
1370
1371	sx_sunlock(&sc->sc_lock);
1372
1373	if (is->is_sim != NULL) {
1374		/*
1375		 * When reconnecting, there already is SIM allocated for the session.
1376		 */
1377		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1378		ISCSI_SESSION_LOCK(is);
1379		ISCSI_SESSION_DEBUG(is, "releasing");
1380		xpt_release_simq(is->is_sim, 1);
1381		is->is_simq_frozen = false;
1382		ISCSI_SESSION_UNLOCK(is);
1383
1384	} else {
1385		ISCSI_SESSION_LOCK(is);
1386		is->is_devq = cam_simq_alloc(maxtags);
1387		if (is->is_devq == NULL) {
1388			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1389			iscsi_session_terminate(is);
1390			return (ENOMEM);
1391		}
1392
1393		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1394		    is, is->is_id /* unit */, &is->is_lock,
1395		    1, maxtags, is->is_devq);
1396		if (is->is_sim == NULL) {
1397			ISCSI_SESSION_UNLOCK(is);
1398			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1399			cam_simq_free(is->is_devq);
1400			iscsi_session_terminate(is);
1401			return (ENOMEM);
1402		}
1403
1404		error = xpt_bus_register(is->is_sim, NULL, 0);
1405		if (error != 0) {
1406			ISCSI_SESSION_UNLOCK(is);
1407			ISCSI_SESSION_WARN(is, "failed to register bus");
1408			iscsi_session_terminate(is);
1409			return (ENOMEM);
1410		}
1411
1412		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1413		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1414		    CAM_LUN_WILDCARD);
1415		if (error != CAM_REQ_CMP) {
1416			ISCSI_SESSION_UNLOCK(is);
1417			ISCSI_SESSION_WARN(is, "failed to create path");
1418			iscsi_session_terminate(is);
1419			return (ENOMEM);
1420		}
1421		ISCSI_SESSION_UNLOCK(is);
1422	}
1423
1424	return (0);
1425}
1426
1427static int
1428iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1429    struct iscsi_daemon_fail *fail)
1430{
1431	struct iscsi_session *is;
1432
1433	sx_slock(&sc->sc_lock);
1434
1435	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1436		if (is->is_id == fail->idf_session_id)
1437			break;
1438	}
1439	if (is == NULL) {
1440		sx_sunlock(&sc->sc_lock);
1441		return (ESRCH);
1442	}
1443	ISCSI_SESSION_LOCK(is);
1444	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1445	    fail->idf_reason);
1446	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1447	//is->is_waiting_for_iscsid = false;
1448	//is->is_login_phase = true;
1449	//iscsi_session_reconnect(is);
1450	ISCSI_SESSION_UNLOCK(is);
1451	sx_sunlock(&sc->sc_lock);
1452
1453	return (0);
1454}
1455
1456#ifdef ICL_KERNEL_PROXY
1457static int
1458iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1459    struct iscsi_daemon_connect *idc)
1460{
1461	struct iscsi_session *is;
1462	struct sockaddr *from_sa, *to_sa;
1463	int error;
1464
1465	sx_slock(&sc->sc_lock);
1466	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1467		if (is->is_id == idc->idc_session_id)
1468			break;
1469	}
1470	if (is == NULL) {
1471		sx_sunlock(&sc->sc_lock);
1472		return (ESRCH);
1473	}
1474	sx_sunlock(&sc->sc_lock);
1475
1476	if (idc->idc_from_addrlen > 0) {
1477		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1478		if (error != 0) {
1479			ISCSI_SESSION_WARN(is,
1480			    "getsockaddr failed with error %d", error);
1481			return (error);
1482		}
1483	} else {
1484		from_sa = NULL;
1485	}
1486	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1487	if (error != 0) {
1488		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1489		    error);
1490		free(from_sa, M_SONAME);
1491		return (error);
1492	}
1493
1494	ISCSI_SESSION_LOCK(is);
1495	is->is_waiting_for_iscsid = false;
1496	is->is_login_phase = true;
1497	is->is_timeout = 0;
1498	ISCSI_SESSION_UNLOCK(is);
1499
1500	error = icl_conn_connect(is->is_conn, idc->idc_iser, idc->idc_domain,
1501	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1502	free(from_sa, M_SONAME);
1503	free(to_sa, M_SONAME);
1504
1505	/*
1506	 * Digests are always disabled during login phase.
1507	 */
1508	is->is_conn->ic_header_crc32c = false;
1509	is->is_conn->ic_data_crc32c = false;
1510
1511	return (error);
1512}
1513
1514static int
1515iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1516    struct iscsi_daemon_send *ids)
1517{
1518	struct iscsi_session *is;
1519	struct icl_pdu *ip;
1520	size_t datalen;
1521	void *data;
1522	int error;
1523
1524	sx_slock(&sc->sc_lock);
1525	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1526		if (is->is_id == ids->ids_session_id)
1527			break;
1528	}
1529	if (is == NULL) {
1530		sx_sunlock(&sc->sc_lock);
1531		return (ESRCH);
1532	}
1533	sx_sunlock(&sc->sc_lock);
1534
1535	if (is->is_login_phase == false)
1536		return (EBUSY);
1537
1538	if (is->is_terminating || is->is_reconnecting)
1539		return (EIO);
1540
1541	datalen = ids->ids_data_segment_len;
1542	if (datalen > ISCSI_MAX_DATA_SEGMENT_LENGTH)
1543		return (EINVAL);
1544	if (datalen > 0) {
1545		data = malloc(datalen, M_ISCSI, M_WAITOK);
1546		error = copyin(ids->ids_data_segment, data, datalen);
1547		if (error != 0) {
1548			free(data, M_ISCSI);
1549			return (error);
1550		}
1551	}
1552
1553	ip = icl_pdu_new_bhs(is->is_conn, M_WAITOK);
1554	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1555	if (datalen > 0) {
1556		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1557		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1558		free(data, M_ISCSI);
1559	}
1560	icl_pdu_queue(ip);
1561
1562	return (0);
1563}
1564
1565static int
1566iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1567    struct iscsi_daemon_receive *idr)
1568{
1569	struct iscsi_session *is;
1570	struct icl_pdu *ip;
1571	void *data;
1572
1573	sx_slock(&sc->sc_lock);
1574	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1575		if (is->is_id == idr->idr_session_id)
1576			break;
1577	}
1578	if (is == NULL) {
1579		sx_sunlock(&sc->sc_lock);
1580		return (ESRCH);
1581	}
1582	sx_sunlock(&sc->sc_lock);
1583
1584	if (is->is_login_phase == false)
1585		return (EBUSY);
1586
1587	ISCSI_SESSION_LOCK(is);
1588	while (is->is_login_pdu == NULL &&
1589	    is->is_terminating == false &&
1590	    is->is_reconnecting == false)
1591		cv_wait(&is->is_login_cv, &is->is_lock);
1592	if (is->is_terminating || is->is_reconnecting) {
1593		ISCSI_SESSION_UNLOCK(is);
1594		return (EIO);
1595	}
1596	ip = is->is_login_pdu;
1597	is->is_login_pdu = NULL;
1598	ISCSI_SESSION_UNLOCK(is);
1599
1600	if (ip->ip_data_len > idr->idr_data_segment_len) {
1601		icl_pdu_free(ip);
1602		return (EMSGSIZE);
1603	}
1604
1605	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1606	if (ip->ip_data_len > 0) {
1607		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1608		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1609		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1610		free(data, M_ISCSI);
1611	}
1612
1613	icl_pdu_free(ip);
1614
1615	return (0);
1616}
1617#endif /* ICL_KERNEL_PROXY */
1618
1619static void
1620iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1621{
1622	/*
1623	 * Just make sure all the fields are null-terminated.
1624	 *
1625	 * XXX: This is not particularly secure.  We should
1626	 * 	create our own conf and then copy in relevant
1627	 * 	fields.
1628	 */
1629	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1630	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1631	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1632	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1633	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1634	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1635	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1636	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1637	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1638}
1639
1640static bool
1641iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1642{
1643
1644	if (isc->isc_initiator[0] == '\0') {
1645		ISCSI_DEBUG("empty isc_initiator");
1646		return (false);
1647	}
1648
1649	if (isc->isc_target_addr[0] == '\0') {
1650		ISCSI_DEBUG("empty isc_target_addr");
1651		return (false);
1652	}
1653
1654	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1655		ISCSI_DEBUG("non-empty isc_target for discovery session");
1656		return (false);
1657	}
1658
1659	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1660		ISCSI_DEBUG("empty isc_target for non-discovery session");
1661		return (false);
1662	}
1663
1664	return (true);
1665}
1666
1667static int
1668iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1669{
1670	struct iscsi_session *is;
1671	const struct iscsi_session *is2;
1672	int error;
1673
1674	iscsi_sanitize_session_conf(&isa->isa_conf);
1675	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1676		return (EINVAL);
1677
1678	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1679	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1680
1681	sx_xlock(&sc->sc_lock);
1682
1683	/*
1684	 * Prevent duplicates.
1685	 */
1686	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1687		if (!!is->is_conf.isc_discovery !=
1688		    !!is2->is_conf.isc_discovery)
1689			continue;
1690
1691		if (strcmp(is->is_conf.isc_target_addr,
1692		    is2->is_conf.isc_target_addr) != 0)
1693			continue;
1694
1695		if (is->is_conf.isc_discovery == 0 &&
1696		    strcmp(is->is_conf.isc_target,
1697		    is2->is_conf.isc_target) != 0)
1698			continue;
1699
1700		sx_xunlock(&sc->sc_lock);
1701		free(is, M_ISCSI);
1702		return (EBUSY);
1703	}
1704
1705	is->is_conn = icl_conn_new("iscsi", &is->is_lock);
1706	is->is_conn->ic_receive = iscsi_receive_callback;
1707	is->is_conn->ic_error = iscsi_error_callback;
1708	is->is_conn->ic_prv0 = is;
1709	TAILQ_INIT(&is->is_outstanding);
1710	STAILQ_INIT(&is->is_postponed);
1711	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1712	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1713#ifdef ICL_KERNEL_PROXY
1714	cv_init(&is->is_login_cv, "iscsi_login");
1715#endif
1716
1717	is->is_softc = sc;
1718	sc->sc_last_session_id++;
1719	is->is_id = sc->sc_last_session_id;
1720	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1721	arc4rand(&is->is_isid[1], 5, 0);
1722	is->is_tsih = 0;
1723	callout_init(&is->is_callout, 1);
1724	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1725	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1726
1727	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1728	if (error != 0) {
1729		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1730		return (error);
1731	}
1732
1733	/*
1734	 * Trigger immediate reconnection.
1735	 */
1736	ISCSI_SESSION_LOCK(is);
1737	is->is_waiting_for_iscsid = true;
1738	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1739	ISCSI_SESSION_UNLOCK(is);
1740	cv_signal(&sc->sc_cv);
1741
1742	sx_xunlock(&sc->sc_lock);
1743
1744	return (0);
1745}
1746
1747static bool
1748iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1749    unsigned int id2, const struct iscsi_session_conf *c2)
1750{
1751	if (id2 == 0 && c2->isc_target[0] == '\0' &&
1752	    c2->isc_target_addr[0] == '\0')
1753		return (true);
1754	if (id2 != 0 && id2 == id1)
1755		return (true);
1756	if (c2->isc_target[0] != '\0' &&
1757	    strcmp(c1->isc_target, c2->isc_target) == 0)
1758		return (true);
1759	if (c2->isc_target_addr[0] != '\0' &&
1760	    strcmp(c1->isc_target_addr, c2->isc_target_addr) == 0)
1761		return (true);
1762	return (false);
1763}
1764
1765static int
1766iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1767    struct iscsi_session_remove *isr)
1768{
1769	struct iscsi_session *is, *tmp;
1770	bool found = false;
1771
1772	iscsi_sanitize_session_conf(&isr->isr_conf);
1773
1774	sx_xlock(&sc->sc_lock);
1775	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1776		ISCSI_SESSION_LOCK(is);
1777		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1778		    isr->isr_session_id, &isr->isr_conf)) {
1779			found = true;
1780			iscsi_session_logout(is);
1781			iscsi_session_terminate(is);
1782		}
1783		ISCSI_SESSION_UNLOCK(is);
1784	}
1785	sx_xunlock(&sc->sc_lock);
1786
1787	if (!found)
1788		return (ESRCH);
1789
1790	return (0);
1791}
1792
1793static int
1794iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1795{
1796	int error;
1797	unsigned int i = 0;
1798	struct iscsi_session *is;
1799	struct iscsi_session_state iss;
1800
1801	sx_slock(&sc->sc_lock);
1802	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1803		if (i >= isl->isl_nentries) {
1804			sx_sunlock(&sc->sc_lock);
1805			return (EMSGSIZE);
1806		}
1807		memset(&iss, 0, sizeof(iss));
1808		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1809		iss.iss_id = is->is_id;
1810		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1811		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1812
1813		if (is->is_conn->ic_header_crc32c)
1814			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1815		else
1816			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1817
1818		if (is->is_conn->ic_data_crc32c)
1819			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1820		else
1821			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1822
1823		iss.iss_max_data_segment_length = is->is_max_data_segment_length;
1824		iss.iss_immediate_data = is->is_immediate_data;
1825		iss.iss_connected = is->is_connected;
1826
1827		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1828		if (error != 0) {
1829			sx_sunlock(&sc->sc_lock);
1830			return (error);
1831		}
1832		i++;
1833	}
1834	sx_sunlock(&sc->sc_lock);
1835
1836	isl->isl_nentries = i;
1837
1838	return (0);
1839}
1840
1841static int
1842iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1843    struct iscsi_session_modify *ism)
1844{
1845	struct iscsi_session *is;
1846
1847	iscsi_sanitize_session_conf(&ism->ism_conf);
1848	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1849		return (EINVAL);
1850
1851	sx_xlock(&sc->sc_lock);
1852	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1853		ISCSI_SESSION_LOCK(is);
1854		if (is->is_id == ism->ism_session_id)
1855			break;
1856		ISCSI_SESSION_UNLOCK(is);
1857	}
1858	if (is == NULL) {
1859		sx_xunlock(&sc->sc_lock);
1860		return (ESRCH);
1861	}
1862	sx_xunlock(&sc->sc_lock);
1863
1864	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
1865	ISCSI_SESSION_UNLOCK(is);
1866
1867	iscsi_session_reconnect(is);
1868
1869	return (0);
1870}
1871
1872static int
1873iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
1874    struct thread *td)
1875{
1876	struct iscsi_softc *sc;
1877
1878	sc = dev->si_drv1;
1879
1880	switch (cmd) {
1881	case ISCSIDWAIT:
1882		return (iscsi_ioctl_daemon_wait(sc,
1883		    (struct iscsi_daemon_request *)arg));
1884	case ISCSIDHANDOFF:
1885		return (iscsi_ioctl_daemon_handoff(sc,
1886		    (struct iscsi_daemon_handoff *)arg));
1887	case ISCSIDFAIL:
1888		return (iscsi_ioctl_daemon_fail(sc,
1889		    (struct iscsi_daemon_fail *)arg));
1890#ifdef ICL_KERNEL_PROXY
1891	case ISCSIDCONNECT:
1892		return (iscsi_ioctl_daemon_connect(sc,
1893		    (struct iscsi_daemon_connect *)arg));
1894	case ISCSIDSEND:
1895		return (iscsi_ioctl_daemon_send(sc,
1896		    (struct iscsi_daemon_send *)arg));
1897	case ISCSIDRECEIVE:
1898		return (iscsi_ioctl_daemon_receive(sc,
1899		    (struct iscsi_daemon_receive *)arg));
1900#endif /* ICL_KERNEL_PROXY */
1901	case ISCSISADD:
1902		return (iscsi_ioctl_session_add(sc,
1903		    (struct iscsi_session_add *)arg));
1904	case ISCSISREMOVE:
1905		return (iscsi_ioctl_session_remove(sc,
1906		    (struct iscsi_session_remove *)arg));
1907	case ISCSISLIST:
1908		return (iscsi_ioctl_session_list(sc,
1909		    (struct iscsi_session_list *)arg));
1910	case ISCSISMODIFY:
1911		return (iscsi_ioctl_session_modify(sc,
1912		    (struct iscsi_session_modify *)arg));
1913	default:
1914		return (EINVAL);
1915	}
1916}
1917
1918static uint64_t
1919iscsi_encode_lun(uint32_t lun)
1920{
1921	uint8_t encoded[8];
1922	uint64_t result;
1923
1924	memset(encoded, 0, sizeof(encoded));
1925
1926	if (lun < 256) {
1927		/*
1928		 * Peripheral device addressing.
1929		 */
1930		encoded[1] = lun;
1931	} else if (lun < 16384) {
1932		/*
1933		 * Flat space addressing.
1934		 */
1935		encoded[0] = 0x40;
1936		encoded[0] |= (lun >> 8) & 0x3f;
1937		encoded[1] = lun & 0xff;
1938	} else {
1939		/*
1940		 * Extended flat space addressing.
1941		 */
1942		encoded[0] = 0xd2;
1943		encoded[1] = lun >> 16;
1944		encoded[2] = lun >> 8;
1945		encoded[3] = lun;
1946	}
1947
1948	memcpy(&result, encoded, sizeof(result));
1949	return (result);
1950}
1951
1952static struct iscsi_outstanding *
1953iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
1954{
1955	struct iscsi_outstanding *io;
1956
1957	ISCSI_SESSION_LOCK_ASSERT(is);
1958
1959	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1960		if (io->io_initiator_task_tag == initiator_task_tag)
1961			return (io);
1962	}
1963	return (NULL);
1964}
1965
1966static struct iscsi_outstanding *
1967iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
1968{
1969	struct iscsi_outstanding *io;
1970
1971	ISCSI_SESSION_LOCK_ASSERT(is);
1972
1973	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1974		if (io->io_ccb == ccb)
1975			return (io);
1976	}
1977	return (NULL);
1978}
1979
1980static struct iscsi_outstanding *
1981iscsi_outstanding_add(struct iscsi_session *is,
1982    uint32_t initiator_task_tag, union ccb *ccb)
1983{
1984	struct iscsi_outstanding *io;
1985
1986	ISCSI_SESSION_LOCK_ASSERT(is);
1987
1988	KASSERT(iscsi_outstanding_find(is, initiator_task_tag) == NULL,
1989	    ("initiator_task_tag 0x%x already added", initiator_task_tag));
1990
1991	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
1992	if (io == NULL) {
1993		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes", sizeof(*io));
1994		return (NULL);
1995	}
1996	io->io_initiator_task_tag = initiator_task_tag;
1997	io->io_ccb = ccb;
1998	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
1999	return (io);
2000}
2001
2002static void
2003iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2004{
2005
2006	ISCSI_SESSION_LOCK_ASSERT(is);
2007
2008	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2009	uma_zfree(iscsi_outstanding_zone, io);
2010}
2011
2012static void
2013iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2014{
2015	struct icl_pdu *request;
2016	struct iscsi_bhs_task_management_request *bhstmr;
2017	struct ccb_abort *cab = &ccb->cab;
2018	struct iscsi_outstanding *io, *aio;
2019
2020	ISCSI_SESSION_LOCK_ASSERT(is);
2021
2022#if 0
2023	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2024#else
2025	if (is->is_login_phase) {
2026		ccb->ccb_h.status = CAM_REQ_ABORTED;
2027		xpt_done(ccb);
2028		return;
2029	}
2030#endif
2031
2032	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2033	if (aio == NULL) {
2034		ccb->ccb_h.status = CAM_REQ_CMP;
2035		xpt_done(ccb);
2036		return;
2037	}
2038
2039	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
2040	if (request == NULL) {
2041		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2042		xpt_done(ccb);
2043		return;
2044	}
2045
2046	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2047	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2048	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2049
2050	bhstmr->bhstmr_lun = iscsi_encode_lun(ccb->ccb_h.target_lun);
2051	bhstmr->bhstmr_initiator_task_tag = is->is_initiator_task_tag;
2052	is->is_initiator_task_tag++;
2053	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2054
2055	io = iscsi_outstanding_add(is, bhstmr->bhstmr_initiator_task_tag, NULL);
2056	if (io == NULL) {
2057		icl_pdu_free(request);
2058		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2059		xpt_done(ccb);
2060		return;
2061	}
2062	io->io_datasn = aio->io_initiator_task_tag;
2063	iscsi_pdu_queue_locked(request);
2064}
2065
2066static void
2067iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2068{
2069	struct icl_pdu *request;
2070	struct iscsi_bhs_scsi_command *bhssc;
2071	struct ccb_scsiio *csio;
2072	struct iscsi_outstanding *io;
2073	size_t len;
2074	int error;
2075
2076	ISCSI_SESSION_LOCK_ASSERT(is);
2077
2078#if 0
2079	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2080#else
2081	if (is->is_login_phase) {
2082		ISCSI_SESSION_DEBUG(is, "called during login phase");
2083		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2084			xpt_freeze_devq(ccb->ccb_h.path, 1);
2085			ISCSI_SESSION_DEBUG(is, "freezing devq");
2086		}
2087		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2088		xpt_done(ccb);
2089		return;
2090	}
2091#endif
2092
2093	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
2094	if (request == NULL) {
2095		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2096			xpt_freeze_devq(ccb->ccb_h.path, 1);
2097			ISCSI_SESSION_DEBUG(is, "freezing devq");
2098		}
2099		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2100		xpt_done(ccb);
2101		return;
2102	}
2103
2104	csio = &ccb->csio;
2105	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2106	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2107	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2108	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2109	case CAM_DIR_IN:
2110		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2111		break;
2112	case CAM_DIR_OUT:
2113		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2114		break;
2115	}
2116
2117	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2118		switch (csio->tag_action) {
2119		case MSG_HEAD_OF_Q_TAG:
2120			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2121			break;
2122		case MSG_ORDERED_Q_TAG:
2123			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2124			break;
2125		case MSG_ACA_TASK:
2126			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2127			break;
2128		case MSG_SIMPLE_Q_TAG:
2129		default:
2130			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2131			break;
2132		}
2133	} else
2134		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2135
2136	bhssc->bhssc_lun = iscsi_encode_lun(csio->ccb_h.target_lun);
2137	bhssc->bhssc_initiator_task_tag = is->is_initiator_task_tag;
2138	is->is_initiator_task_tag++;
2139	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2140	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2141	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2142
2143	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2144		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2145	else
2146		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2147
2148	io = iscsi_outstanding_add(is, bhssc->bhssc_initiator_task_tag, ccb);
2149	if (io == NULL) {
2150		icl_pdu_free(request);
2151		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2152			xpt_freeze_devq(ccb->ccb_h.path, 1);
2153			ISCSI_SESSION_DEBUG(is, "freezing devq");
2154		}
2155		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2156		xpt_done(ccb);
2157		return;
2158	}
2159
2160	if (is->is_immediate_data &&
2161	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2162		len = csio->dxfer_len;
2163		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2164		if (len > is->is_first_burst_length) {
2165			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_first_burst_length);
2166			len = is->is_first_burst_length;
2167		}
2168
2169		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2170		if (error != 0) {
2171			icl_pdu_free(request);
2172			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2173				xpt_freeze_devq(ccb->ccb_h.path, 1);
2174				ISCSI_SESSION_DEBUG(is, "freezing devq");
2175			}
2176			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2177			xpt_done(ccb);
2178			return;
2179		}
2180	}
2181	iscsi_pdu_queue_locked(request);
2182}
2183
2184static void
2185iscsi_action(struct cam_sim *sim, union ccb *ccb)
2186{
2187	struct iscsi_session *is;
2188
2189	is = cam_sim_softc(sim);
2190
2191	ISCSI_SESSION_LOCK_ASSERT(is);
2192
2193	if (is->is_terminating ||
2194	    (is->is_connected == false && fail_on_disconnection)) {
2195		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2196		xpt_done(ccb);
2197		return;
2198	}
2199
2200	switch (ccb->ccb_h.func_code) {
2201	case XPT_PATH_INQ:
2202	{
2203		struct ccb_pathinq *cpi = &ccb->cpi;
2204
2205		cpi->version_num = 1;
2206		cpi->hba_inquiry = PI_TAG_ABLE;
2207		cpi->target_sprt = 0;
2208		//cpi->hba_misc = PIM_NOBUSRESET;
2209		cpi->hba_misc = 0;
2210		cpi->hba_eng_cnt = 0;
2211		cpi->max_target = 0;
2212		cpi->max_lun = 255;
2213		//cpi->initiator_id = 0; /* XXX */
2214		cpi->initiator_id = 64; /* XXX */
2215		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2216		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2217		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2218		cpi->unit_number = cam_sim_unit(sim);
2219		cpi->bus_id = cam_sim_bus(sim);
2220		cpi->base_transfer_speed = 150000; /* XXX */
2221		cpi->transport = XPORT_ISCSI;
2222		cpi->transport_version = 0;
2223		cpi->protocol = PROTO_SCSI;
2224		cpi->protocol_version = SCSI_REV_SPC3;
2225		cpi->maxio = MAXPHYS;
2226		cpi->ccb_h.status = CAM_REQ_CMP;
2227		break;
2228	}
2229	case XPT_GET_TRAN_SETTINGS:
2230	{
2231		struct ccb_trans_settings	*cts;
2232		struct ccb_trans_settings_scsi	*scsi;
2233
2234		cts = &ccb->cts;
2235		scsi = &cts->proto_specific.scsi;
2236
2237		cts->protocol = PROTO_SCSI;
2238		cts->protocol_version = SCSI_REV_SPC3;
2239		cts->transport = XPORT_ISCSI;
2240		cts->transport_version = 0;
2241		scsi->valid = CTS_SCSI_VALID_TQ;
2242		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2243		cts->ccb_h.status = CAM_REQ_CMP;
2244		break;
2245	}
2246	case XPT_CALC_GEOMETRY:
2247		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2248		ccb->ccb_h.status = CAM_REQ_CMP;
2249		break;
2250#if 0
2251	/*
2252	 * XXX: What's the point?
2253	 */
2254	case XPT_RESET_BUS:
2255	case XPT_TERM_IO:
2256		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2257		ccb->ccb_h.status = CAM_REQ_CMP;
2258		break;
2259#endif
2260	case XPT_ABORT:
2261		iscsi_action_abort(is, ccb);
2262		return;
2263	case XPT_SCSI_IO:
2264		iscsi_action_scsiio(is, ccb);
2265		return;
2266	default:
2267#if 0
2268		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2269#endif
2270		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2271		break;
2272	}
2273	xpt_done(ccb);
2274}
2275
2276static void
2277iscsi_poll(struct cam_sim *sim)
2278{
2279
2280	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2281}
2282
2283static void
2284iscsi_shutdown(struct iscsi_softc *sc)
2285{
2286	struct iscsi_session *is;
2287
2288	ISCSI_DEBUG("removing all sessions due to shutdown");
2289
2290	sx_slock(&sc->sc_lock);
2291	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2292		iscsi_session_terminate(is);
2293	sx_sunlock(&sc->sc_lock);
2294}
2295
2296static int
2297iscsi_load(void)
2298{
2299	int error;
2300
2301	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2302	sx_init(&sc->sc_lock, "iscsi");
2303	TAILQ_INIT(&sc->sc_sessions);
2304	cv_init(&sc->sc_cv, "iscsi_cv");
2305
2306	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2307	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2308	    UMA_ALIGN_PTR, 0);
2309
2310	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2311	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2312	if (error != 0) {
2313		ISCSI_WARN("failed to create device node, error %d", error);
2314		return (error);
2315	}
2316	sc->sc_cdev->si_drv1 = sc;
2317
2318	/*
2319	 * Note that this needs to get run before dashutdown().  Otherwise,
2320	 * when rebooting with iSCSI session with outstanding requests,
2321	 * but disconnected, dashutdown() will hang on cam_periph_runccb().
2322	 */
2323	sc->sc_shutdown_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2324	    iscsi_shutdown, sc, SHUTDOWN_PRI_FIRST);
2325
2326	return (0);
2327}
2328
2329static int
2330iscsi_unload(void)
2331{
2332	struct iscsi_session *is, *tmp;
2333
2334	if (sc->sc_cdev != NULL) {
2335		ISCSI_DEBUG("removing device node");
2336		destroy_dev(sc->sc_cdev);
2337		ISCSI_DEBUG("device node removed");
2338	}
2339
2340	if (sc->sc_shutdown_eh != NULL)
2341		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_eh);
2342
2343	sx_slock(&sc->sc_lock);
2344	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp)
2345		iscsi_session_terminate(is);
2346	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2347		ISCSI_DEBUG("waiting for sessions to terminate");
2348		cv_wait(&sc->sc_cv, &sc->sc_lock);
2349	}
2350	ISCSI_DEBUG("all sessions terminated");
2351	sx_sunlock(&sc->sc_lock);
2352
2353	uma_zdestroy(iscsi_outstanding_zone);
2354	sx_destroy(&sc->sc_lock);
2355	cv_destroy(&sc->sc_cv);
2356	free(sc, M_ISCSI);
2357	return (0);
2358}
2359
2360static int
2361iscsi_quiesce(void)
2362{
2363	sx_slock(&sc->sc_lock);
2364	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2365		sx_sunlock(&sc->sc_lock);
2366		return (EBUSY);
2367	}
2368	sx_sunlock(&sc->sc_lock);
2369	return (0);
2370}
2371
2372static int
2373iscsi_modevent(module_t mod, int what, void *arg)
2374{
2375	int error;
2376
2377	switch (what) {
2378	case MOD_LOAD:
2379		error = iscsi_load();
2380		break;
2381	case MOD_UNLOAD:
2382		error = iscsi_unload();
2383		break;
2384	case MOD_QUIESCE:
2385		error = iscsi_quiesce();
2386		break;
2387	default:
2388		error = EINVAL;
2389		break;
2390	}
2391	return (error);
2392}
2393
2394moduledata_t iscsi_data = {
2395	"iscsi",
2396	iscsi_modevent,
2397	0
2398};
2399
2400DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2401MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2402MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2403