iscsi.c revision 276234
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 276234 2014-12-26 09:12:19Z mav $");
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 <dev/iscsi/icl.h>
62#include <dev/iscsi/iscsi_ioctl.h>
63#include <dev/iscsi/iscsi_proto.h>
64#include <dev/iscsi/iscsi.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#define ISCSI_SESSION_LOCK_ASSERT_NOT(X) mtx_assert(&X->is_lock, MA_NOTOWNED)
148
149static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
150		    int mode, struct thread *td);
151
152static struct cdevsw iscsi_cdevsw = {
153     .d_version = D_VERSION,
154     .d_ioctl   = iscsi_ioctl,
155     .d_name    = "iscsi",
156};
157
158static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
159static void	iscsi_pdu_queue(struct icl_pdu *request);
160static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
161static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
162static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
163static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
164static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
165static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
166static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
167static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
168static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
169static void	iscsi_session_reconnect(struct iscsi_session *is);
170static void	iscsi_session_terminate(struct iscsi_session *is);
171static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
172static void	iscsi_poll(struct cam_sim *sim);
173static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
174		    uint32_t initiator_task_tag);
175static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
176		    uint32_t initiator_task_tag, union ccb *ccb);
177static void	iscsi_outstanding_remove(struct iscsi_session *is,
178		    struct iscsi_outstanding *io);
179
180static bool
181iscsi_pdu_prepare(struct icl_pdu *request)
182{
183	struct iscsi_session *is;
184	struct iscsi_bhs_scsi_command *bhssc;
185
186	is = PDU_SESSION(request);
187
188	ISCSI_SESSION_LOCK_ASSERT(is);
189
190	/*
191	 * We're only using fields common for all the request
192	 * (initiator -> target) PDUs.
193	 */
194	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
195
196	/*
197	 * Data-Out PDU does not contain CmdSN.
198	 */
199	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
200		if (is->is_cmdsn > is->is_maxcmdsn &&
201		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
202			/*
203			 * Current MaxCmdSN prevents us from sending any more
204			 * SCSI Command PDUs to the target; postpone the PDU.
205			 * It will get resent by either iscsi_pdu_queue(),
206			 * or by maintenance thread.
207			 */
208#if 0
209			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %d, ExpCmdSN %d, MaxCmdSN %d, opcode 0x%x",
210			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn, bhssc->bhssc_opcode);
211#endif
212			return (true);
213		}
214		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
215		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
216			is->is_cmdsn++;
217	}
218	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
219
220	return (false);
221}
222
223static void
224iscsi_session_send_postponed(struct iscsi_session *is)
225{
226	struct icl_pdu *request;
227	bool postpone;
228
229	ISCSI_SESSION_LOCK_ASSERT(is);
230
231	while (!STAILQ_EMPTY(&is->is_postponed)) {
232		request = STAILQ_FIRST(&is->is_postponed);
233		postpone = iscsi_pdu_prepare(request);
234		if (postpone)
235			break;
236		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
237		icl_pdu_queue(request);
238	}
239}
240
241static void
242iscsi_pdu_queue_locked(struct icl_pdu *request)
243{
244	struct iscsi_session *is;
245	bool postpone;
246
247	is = PDU_SESSION(request);
248	ISCSI_SESSION_LOCK_ASSERT(is);
249	iscsi_session_send_postponed(is);
250	postpone = iscsi_pdu_prepare(request);
251	if (postpone) {
252		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
253		return;
254	}
255	icl_pdu_queue(request);
256}
257
258static void
259iscsi_pdu_queue(struct icl_pdu *request)
260{
261	struct iscsi_session *is;
262
263	is = PDU_SESSION(request);
264	ISCSI_SESSION_LOCK(is);
265	iscsi_pdu_queue_locked(request);
266	ISCSI_SESSION_UNLOCK(is);
267}
268
269static void
270iscsi_session_logout(struct iscsi_session *is)
271{
272	struct icl_pdu *request;
273	struct iscsi_bhs_logout_request *bhslr;
274
275	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
276	if (request == NULL)
277		return;
278
279	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
280	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
281	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
282	iscsi_pdu_queue_locked(request);
283}
284
285static void
286iscsi_session_terminate_task(struct iscsi_session *is,
287    struct iscsi_outstanding *io, bool requeue)
288{
289
290	if (io->io_ccb != NULL) {
291		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
292		if (requeue)
293			io->io_ccb->ccb_h.status |= CAM_REQUEUE_REQ;
294		else
295			io->io_ccb->ccb_h.status |= CAM_REQ_ABORTED;
296		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
297			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
298			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
299			ISCSI_SESSION_DEBUG(is, "freezing devq");
300		}
301		xpt_done(io->io_ccb);
302	}
303	iscsi_outstanding_remove(is, io);
304}
305
306static void
307iscsi_session_terminate_tasks(struct iscsi_session *is, bool requeue)
308{
309	struct iscsi_outstanding *io, *tmp;
310
311	ISCSI_SESSION_LOCK_ASSERT(is);
312
313	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
314		iscsi_session_terminate_task(is, io, requeue);
315	}
316}
317
318static void
319iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
320{
321	struct icl_pdu *pdu;
322
323	ISCSI_SESSION_LOCK_ASSERT(is);
324
325	/*
326	 * Don't queue any new PDUs.
327	 */
328	if (is->is_sim != NULL && is->is_simq_frozen == false) {
329		ISCSI_SESSION_DEBUG(is, "freezing");
330		xpt_freeze_simq(is->is_sim, 1);
331		is->is_simq_frozen = true;
332	}
333
334	/*
335	 * Remove postponed PDUs.
336	 */
337	while (!STAILQ_EMPTY(&is->is_postponed)) {
338		pdu = STAILQ_FIRST(&is->is_postponed);
339		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
340		icl_pdu_free(pdu);
341	}
342
343	if (destroy_sim == false) {
344		/*
345		 * Terminate SCSI tasks, asking CAM to requeue them.
346		 */
347		iscsi_session_terminate_tasks(is, true);
348		return;
349	}
350
351	iscsi_session_terminate_tasks(is, false);
352
353	if (is->is_sim == NULL)
354		return;
355
356	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
357	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
358
359	if (is->is_simq_frozen) {
360		xpt_release_simq(is->is_sim, 1);
361		is->is_simq_frozen = false;
362	}
363
364	xpt_free_path(is->is_path);
365	is->is_path = NULL;
366	xpt_bus_deregister(cam_sim_path(is->is_sim));
367	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
368	is->is_sim = NULL;
369	is->is_devq = NULL;
370}
371
372static void
373iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
374{
375
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 (iscsid_timeout > 0 && 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 (login_timeout > 0 && 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 (ping_timeout <= 0) {
565		/*
566		 * Pings are disabled.  Don't send NOP-Out in this case.
567		 * Reset the timeout, to avoid triggering reconnection,
568		 * should the user decide to reenable them.
569		 */
570		is->is_timeout = 0;
571		goto out;
572	}
573
574	if (is->is_timeout >= ping_timeout) {
575		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
576		    "reconnecting", ping_timeout);
577		reconnect_needed = true;
578		goto out;
579	}
580
581	ISCSI_SESSION_UNLOCK(is);
582
583	/*
584	 * If the ping was reset less than one second ago - which means
585	 * that we've received some PDU during the last second - assume
586	 * the traffic flows correctly and don't bother sending a NOP-Out.
587	 *
588	 * (It's 2 - one for one second, and one for incrementing is_timeout
589	 * earlier in this routine.)
590	 */
591	if (is->is_timeout < 2)
592		return;
593
594	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
595	if (request == NULL) {
596		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
597		return;
598	}
599	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
600	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
601	    ISCSI_BHS_OPCODE_IMMEDIATE;
602	bhsno->bhsno_flags = 0x80;
603	bhsno->bhsno_target_transfer_tag = 0xffffffff;
604	iscsi_pdu_queue(request);
605	return;
606
607out:
608	ISCSI_SESSION_UNLOCK(is);
609
610	if (reconnect_needed)
611		iscsi_session_reconnect(is);
612}
613
614static void
615iscsi_pdu_update_statsn(const struct icl_pdu *response)
616{
617	const struct iscsi_bhs_data_in *bhsdi;
618	struct iscsi_session *is;
619	uint32_t expcmdsn, maxcmdsn;
620
621	is = PDU_SESSION(response);
622
623	ISCSI_SESSION_LOCK_ASSERT(is);
624
625	/*
626	 * We're only using fields common for all the response
627	 * (target -> initiator) PDUs.
628	 */
629	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
630	/*
631	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
632	 * and Residual Count only have meaningful content if the S bit
633	 * is set to 1", so we also need to check the bit specific for
634	 * Data-In PDU.
635	 */
636	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
637	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
638		if (ntohl(bhsdi->bhsdi_statsn) < is->is_statsn) {
639			ISCSI_SESSION_WARN(is,
640			    "PDU StatSN %d >= session StatSN %d, opcode 0x%x",
641			    is->is_statsn, ntohl(bhsdi->bhsdi_statsn),
642			    bhsdi->bhsdi_opcode);
643		}
644		is->is_statsn = ntohl(bhsdi->bhsdi_statsn);
645	}
646
647	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
648	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
649
650	/*
651	 * XXX: Compare using Serial Arithmetic Sense.
652	 */
653	if (maxcmdsn + 1 < expcmdsn) {
654		ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d + 1 < PDU ExpCmdSN %d; ignoring",
655		    maxcmdsn, expcmdsn);
656	} else {
657		if (maxcmdsn > is->is_maxcmdsn) {
658			is->is_maxcmdsn = maxcmdsn;
659
660			/*
661			 * Command window increased; kick the maintanance thread
662			 * to send out postponed commands.
663			 */
664			if (!STAILQ_EMPTY(&is->is_postponed))
665				cv_signal(&is->is_maintenance_cv);
666		} else if (maxcmdsn < is->is_maxcmdsn) {
667			ISCSI_SESSION_DEBUG(is, "PDU MaxCmdSN %d < session MaxCmdSN %d; ignoring",
668			    maxcmdsn, is->is_maxcmdsn);
669		}
670
671		if (expcmdsn > is->is_expcmdsn) {
672			is->is_expcmdsn = expcmdsn;
673		} else if (expcmdsn < is->is_expcmdsn) {
674			ISCSI_SESSION_DEBUG(is, "PDU ExpCmdSN %d < session ExpCmdSN %d; ignoring",
675			    expcmdsn, is->is_expcmdsn);
676		}
677	}
678
679	/*
680	 * Every incoming PDU - not just NOP-In - resets the ping timer.
681	 * The purpose of the timeout is to reset the connection when it stalls;
682	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
683	 * in some queue.
684	 */
685	is->is_timeout = 0;
686}
687
688static void
689iscsi_receive_callback(struct icl_pdu *response)
690{
691	struct iscsi_session *is;
692
693	is = PDU_SESSION(response);
694
695	ISCSI_SESSION_LOCK(is);
696
697#ifdef ICL_KERNEL_PROXY
698	if (is->is_login_phase) {
699		if (is->is_login_pdu == NULL)
700			is->is_login_pdu = response;
701		else
702			icl_pdu_free(response);
703		ISCSI_SESSION_UNLOCK(is);
704		cv_signal(&is->is_login_cv);
705		return;
706	}
707#endif
708
709	iscsi_pdu_update_statsn(response);
710
711	/*
712	 * The handling routine is responsible for freeing the PDU
713	 * when it's no longer needed.
714	 */
715	switch (response->ip_bhs->bhs_opcode) {
716	case ISCSI_BHS_OPCODE_NOP_IN:
717		iscsi_pdu_handle_nop_in(response);
718		ISCSI_SESSION_UNLOCK(is);
719		break;
720	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
721		iscsi_pdu_handle_scsi_response(response);
722		/* Session lock dropped inside. */
723		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
724		break;
725	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
726		iscsi_pdu_handle_task_response(response);
727		ISCSI_SESSION_UNLOCK(is);
728		break;
729	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
730		iscsi_pdu_handle_data_in(response);
731		/* Session lock dropped inside. */
732		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
733		break;
734	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
735		iscsi_pdu_handle_logout_response(response);
736		ISCSI_SESSION_UNLOCK(is);
737		break;
738	case ISCSI_BHS_OPCODE_R2T:
739		iscsi_pdu_handle_r2t(response);
740		ISCSI_SESSION_UNLOCK(is);
741		break;
742	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
743		iscsi_pdu_handle_async_message(response);
744		ISCSI_SESSION_UNLOCK(is);
745		break;
746	case ISCSI_BHS_OPCODE_REJECT:
747		iscsi_pdu_handle_reject(response);
748		ISCSI_SESSION_UNLOCK(is);
749		break;
750	default:
751		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
752		    "opcode 0x%x; reconnecting",
753		    response->ip_bhs->bhs_opcode);
754		iscsi_session_reconnect(is);
755		ISCSI_SESSION_UNLOCK(is);
756		icl_pdu_free(response);
757	}
758}
759
760static void
761iscsi_error_callback(struct icl_conn *ic)
762{
763	struct iscsi_session *is;
764
765	is = CONN_SESSION(ic);
766
767	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
768	iscsi_session_reconnect(is);
769}
770
771static void
772iscsi_pdu_handle_nop_in(struct icl_pdu *response)
773{
774	struct iscsi_session *is;
775	struct iscsi_bhs_nop_out *bhsno;
776	struct iscsi_bhs_nop_in *bhsni;
777	struct icl_pdu *request;
778	void *data = NULL;
779	size_t datasize;
780	int error;
781
782	is = PDU_SESSION(response);
783	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
784
785	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
786		/*
787		 * Nothing to do; iscsi_pdu_update_statsn() already
788		 * zeroed the timeout.
789		 */
790		icl_pdu_free(response);
791		return;
792	}
793
794	datasize = icl_pdu_data_segment_length(response);
795	if (datasize > 0) {
796		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
797		if (data == NULL) {
798			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
799			    "reconnecting");
800			icl_pdu_free(response);
801			iscsi_session_reconnect(is);
802			return;
803		}
804		icl_pdu_get_data(response, 0, data, datasize);
805	}
806
807	request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
808	if (request == NULL) {
809		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
810		    "reconnecting");
811		free(data, M_ISCSI);
812		icl_pdu_free(response);
813		iscsi_session_reconnect(is);
814		return;
815	}
816	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
817	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
818	    ISCSI_BHS_OPCODE_IMMEDIATE;
819	bhsno->bhsno_flags = 0x80;
820	bhsno->bhsno_initiator_task_tag = 0xffffffff;
821	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
822	if (datasize > 0) {
823		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
824		if (error != 0) {
825			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
826			    "reconnecting");
827			free(data, M_ISCSI);
828			icl_pdu_free(request);
829			icl_pdu_free(response);
830			iscsi_session_reconnect(is);
831			return;
832		}
833		free(data, M_ISCSI);
834	}
835
836	icl_pdu_free(response);
837	iscsi_pdu_queue_locked(request);
838}
839
840static void
841iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
842{
843	struct iscsi_bhs_scsi_response *bhssr;
844	struct iscsi_outstanding *io;
845	struct iscsi_session *is;
846	union ccb *ccb;
847	struct ccb_scsiio *csio;
848	size_t data_segment_len, received;
849	uint16_t sense_len;
850
851	is = PDU_SESSION(response);
852
853	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
854	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
855	if (io == NULL || io->io_ccb == NULL) {
856		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
857		icl_pdu_free(response);
858		iscsi_session_reconnect(is);
859		ISCSI_SESSION_UNLOCK(is);
860		return;
861	}
862
863	ccb = io->io_ccb;
864	received = io->io_received;
865	iscsi_outstanding_remove(is, io);
866	ISCSI_SESSION_UNLOCK(is);
867
868	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
869		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
870 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
871 			xpt_freeze_devq(ccb->ccb_h.path, 1);
872			ISCSI_SESSION_DEBUG(is, "freezing devq");
873		}
874 		ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
875	} else if (bhssr->bhssr_status == 0) {
876		ccb->ccb_h.status = CAM_REQ_CMP;
877	} else {
878 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
879 			xpt_freeze_devq(ccb->ccb_h.path, 1);
880			ISCSI_SESSION_DEBUG(is, "freezing devq");
881		}
882 		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
883		ccb->csio.scsi_status = bhssr->bhssr_status;
884	}
885
886	csio = &ccb->csio;
887	data_segment_len = icl_pdu_data_segment_length(response);
888	if (data_segment_len > 0) {
889		if (data_segment_len < sizeof(sense_len)) {
890			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
891			    data_segment_len);
892			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
893				xpt_freeze_devq(ccb->ccb_h.path, 1);
894				ISCSI_SESSION_DEBUG(is, "freezing devq");
895			}
896			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
897			goto out;
898		}
899		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
900		sense_len = ntohs(sense_len);
901#if 0
902		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
903		    sense_len, data_segment_len);
904#endif
905		if (sizeof(sense_len) + sense_len > data_segment_len) {
906			ISCSI_SESSION_WARN(is, "truncated data segment "
907			    "(%zd bytes, should be %zd)",
908			    data_segment_len, sizeof(sense_len) + sense_len);
909			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
910				xpt_freeze_devq(ccb->ccb_h.path, 1);
911				ISCSI_SESSION_DEBUG(is, "freezing devq");
912			}
913			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
914			goto out;
915		} else if (sizeof(sense_len) + sense_len < data_segment_len)
916			ISCSI_SESSION_WARN(is, "oversize data segment "
917			    "(%zd bytes, should be %zd)",
918			    data_segment_len, sizeof(sense_len) + sense_len);
919		if (sense_len > csio->sense_len) {
920			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
921			    sense_len, csio->sense_len);
922			sense_len = csio->sense_len;
923		}
924		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
925		csio->sense_resid = csio->sense_len - sense_len;
926		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
927	}
928
929out:
930	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
931		csio->resid = ntohl(bhssr->bhssr_residual_count);
932
933	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
934		KASSERT(received <= csio->dxfer_len,
935		    ("received > csio->dxfer_len"));
936		if (received < csio->dxfer_len) {
937			if (csio->resid != csio->dxfer_len - received) {
938				ISCSI_SESSION_WARN(is, "underflow mismatch: "
939				    "target indicates %d, we calculated %zd",
940				    csio->resid, csio->dxfer_len - received);
941			}
942			csio->resid = csio->dxfer_len - received;
943		}
944	}
945
946	xpt_done(ccb);
947	icl_pdu_free(response);
948}
949
950static void
951iscsi_pdu_handle_task_response(struct icl_pdu *response)
952{
953	struct iscsi_bhs_task_management_response *bhstmr;
954	struct iscsi_outstanding *io, *aio;
955	struct iscsi_session *is;
956
957	is = PDU_SESSION(response);
958
959	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
960	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
961	if (io == NULL || io->io_ccb != NULL) {
962		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
963		    bhstmr->bhstmr_initiator_task_tag);
964		icl_pdu_free(response);
965		iscsi_session_reconnect(is);
966		return;
967	}
968
969	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
970		ISCSI_SESSION_WARN(is, "task response 0x%x",
971		    bhstmr->bhstmr_response);
972	} else {
973		aio = iscsi_outstanding_find(is, io->io_datasn);
974		if (aio != NULL && aio->io_ccb != NULL)
975			iscsi_session_terminate_task(is, aio, false);
976	}
977
978	iscsi_outstanding_remove(is, io);
979	icl_pdu_free(response);
980}
981
982static void
983iscsi_pdu_handle_data_in(struct icl_pdu *response)
984{
985	struct iscsi_bhs_data_in *bhsdi;
986	struct iscsi_outstanding *io;
987	struct iscsi_session *is;
988	union ccb *ccb;
989	struct ccb_scsiio *csio;
990	size_t data_segment_len, received, oreceived;
991
992	is = PDU_SESSION(response);
993	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
994	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
995	if (io == NULL || io->io_ccb == NULL) {
996		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
997		icl_pdu_free(response);
998		iscsi_session_reconnect(is);
999		ISCSI_SESSION_UNLOCK(is);
1000		return;
1001	}
1002
1003	data_segment_len = icl_pdu_data_segment_length(response);
1004	if (data_segment_len == 0) {
1005		/*
1006		 * "The sending of 0 length data segments should be avoided,
1007		 * but initiators and targets MUST be able to properly receive
1008		 * 0 length data segments."
1009		 */
1010		ISCSI_SESSION_UNLOCK(is);
1011		icl_pdu_free(response);
1012		return;
1013	}
1014
1015	/*
1016	 * We need to track this for security reasons - without it, malicious target
1017	 * could respond to SCSI READ without sending Data-In PDUs, which would result
1018	 * in read operation on the initiator side returning random kernel data.
1019	 */
1020	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1021		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1022		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1023		icl_pdu_free(response);
1024		iscsi_session_reconnect(is);
1025		ISCSI_SESSION_UNLOCK(is);
1026		return;
1027	}
1028
1029	ccb = io->io_ccb;
1030	csio = &ccb->csio;
1031
1032	if (io->io_received + data_segment_len > csio->dxfer_len) {
1033		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1034		    "at offset %zd, buffer is %d)",
1035		    data_segment_len, io->io_received, csio->dxfer_len);
1036		icl_pdu_free(response);
1037		iscsi_session_reconnect(is);
1038		ISCSI_SESSION_UNLOCK(is);
1039		return;
1040	}
1041
1042	oreceived = io->io_received;
1043	io->io_received += data_segment_len;
1044	received = io->io_received;
1045	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0)
1046		iscsi_outstanding_remove(is, io);
1047	ISCSI_SESSION_UNLOCK(is);
1048
1049	icl_pdu_get_data(response, 0, csio->data_ptr + oreceived, data_segment_len);
1050
1051	/*
1052	 * XXX: Check DataSN.
1053	 * XXX: Check F.
1054	 */
1055	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1056		/*
1057		 * Nothing more to do.
1058		 */
1059		icl_pdu_free(response);
1060		return;
1061	}
1062
1063	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1064	if (bhsdi->bhsdi_status == 0) {
1065		ccb->ccb_h.status = CAM_REQ_CMP;
1066	} else {
1067		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1068			xpt_freeze_devq(ccb->ccb_h.path, 1);
1069			ISCSI_SESSION_DEBUG(is, "freezing devq");
1070		}
1071		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1072		csio->scsi_status = bhsdi->bhsdi_status;
1073	}
1074
1075	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1076		KASSERT(received <= csio->dxfer_len,
1077		    ("received > csio->dxfer_len"));
1078		if (received < csio->dxfer_len) {
1079			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1080			if (csio->resid != csio->dxfer_len - received) {
1081				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1082				    "target indicates %d, we calculated %zd",
1083				    csio->resid, csio->dxfer_len - received);
1084			}
1085			csio->resid = csio->dxfer_len - received;
1086		}
1087	}
1088
1089	xpt_done(ccb);
1090	icl_pdu_free(response);
1091}
1092
1093static void
1094iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1095{
1096
1097	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1098	icl_pdu_free(response);
1099}
1100
1101static void
1102iscsi_pdu_handle_r2t(struct icl_pdu *response)
1103{
1104	struct icl_pdu *request;
1105	struct iscsi_session *is;
1106	struct iscsi_bhs_r2t *bhsr2t;
1107	struct iscsi_bhs_data_out *bhsdo;
1108	struct iscsi_outstanding *io;
1109	struct ccb_scsiio *csio;
1110	size_t off, len, total_len;
1111	int error;
1112
1113	is = PDU_SESSION(response);
1114
1115	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1116	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1117	if (io == NULL || io->io_ccb == NULL) {
1118		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1119		    bhsr2t->bhsr2t_initiator_task_tag);
1120		icl_pdu_free(response);
1121		iscsi_session_reconnect(is);
1122		return;
1123	}
1124
1125	csio = &io->io_ccb->csio;
1126
1127	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1128		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1129		icl_pdu_free(response);
1130		iscsi_session_reconnect(is);
1131		return;
1132	}
1133
1134	/*
1135	 * XXX: Verify R2TSN.
1136	 */
1137
1138	io->io_datasn = 0;
1139
1140	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1141	if (off > csio->dxfer_len) {
1142		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1143		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1144		icl_pdu_free(response);
1145		iscsi_session_reconnect(is);
1146		return;
1147	}
1148
1149	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1150	if (total_len == 0 || total_len > csio->dxfer_len) {
1151		ISCSI_SESSION_WARN(is, "target requested invalid length "
1152		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1153		icl_pdu_free(response);
1154		iscsi_session_reconnect(is);
1155		return;
1156	}
1157
1158	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1159
1160	for (;;) {
1161		len = total_len;
1162
1163		if (len > is->is_max_data_segment_length)
1164			len = is->is_max_data_segment_length;
1165
1166		if (off + len > csio->dxfer_len) {
1167			ISCSI_SESSION_WARN(is, "target requested invalid "
1168			    "length/offset %zd, buffer is %d; reconnecting",
1169			    off + len, csio->dxfer_len);
1170			icl_pdu_free(response);
1171			iscsi_session_reconnect(is);
1172			return;
1173		}
1174
1175		request = icl_pdu_new_bhs(response->ip_conn, M_NOWAIT);
1176		if (request == NULL) {
1177			icl_pdu_free(response);
1178			iscsi_session_reconnect(is);
1179			return;
1180		}
1181
1182		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1183		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1184		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1185		bhsdo->bhsdo_initiator_task_tag =
1186		    bhsr2t->bhsr2t_initiator_task_tag;
1187		bhsdo->bhsdo_target_transfer_tag =
1188		    bhsr2t->bhsr2t_target_transfer_tag;
1189		bhsdo->bhsdo_datasn = htonl(io->io_datasn++);
1190		bhsdo->bhsdo_buffer_offset = htonl(off);
1191		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1192		    M_NOWAIT);
1193		if (error != 0) {
1194			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1195			    "reconnecting");
1196			icl_pdu_free(request);
1197			icl_pdu_free(response);
1198			iscsi_session_reconnect(is);
1199			return;
1200		}
1201
1202		off += len;
1203		total_len -= len;
1204
1205		if (total_len == 0) {
1206			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1207			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1208		} else {
1209			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1210		}
1211
1212		iscsi_pdu_queue_locked(request);
1213
1214		if (total_len == 0)
1215			break;
1216	}
1217
1218	icl_pdu_free(response);
1219}
1220
1221static void
1222iscsi_pdu_handle_async_message(struct icl_pdu *response)
1223{
1224	struct iscsi_bhs_asynchronous_message *bhsam;
1225	struct iscsi_session *is;
1226
1227	is = PDU_SESSION(response);
1228	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1229	switch (bhsam->bhsam_async_event) {
1230	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1231		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1232		iscsi_session_logout(is);
1233		iscsi_session_terminate(is);
1234		break;
1235	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1236		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the connection");
1237		break;
1238	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1239		ISCSI_SESSION_WARN(is, "target indicates it will drop drop the session");
1240		break;
1241	default:
1242		/*
1243		 * XXX: Technically, we're obligated to also handle
1244		 * 	parameter renegotiation.
1245		 */
1246		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1247		break;
1248	}
1249
1250	icl_pdu_free(response);
1251}
1252
1253static void
1254iscsi_pdu_handle_reject(struct icl_pdu *response)
1255{
1256	struct iscsi_bhs_reject *bhsr;
1257	struct iscsi_session *is;
1258
1259	is = PDU_SESSION(response);
1260	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1261	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1262	    bhsr->bhsr_reason);
1263
1264	icl_pdu_free(response);
1265}
1266
1267static int
1268iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1269    struct iscsi_daemon_request *request)
1270{
1271	struct iscsi_session *is;
1272	int error;
1273
1274	sx_slock(&sc->sc_lock);
1275	for (;;) {
1276		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1277			ISCSI_SESSION_LOCK(is);
1278			if (is->is_waiting_for_iscsid)
1279				break;
1280			ISCSI_SESSION_UNLOCK(is);
1281		}
1282
1283		if (is == NULL) {
1284			/*
1285			 * No session requires attention from iscsid(8); wait.
1286			 */
1287			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1288			if (error != 0) {
1289				sx_sunlock(&sc->sc_lock);
1290				return (error);
1291			}
1292			continue;
1293		}
1294
1295		is->is_waiting_for_iscsid = false;
1296		is->is_login_phase = true;
1297		is->is_reason[0] = '\0';
1298		ISCSI_SESSION_UNLOCK(is);
1299
1300		request->idr_session_id = is->is_id;
1301		memcpy(&request->idr_isid, &is->is_isid,
1302		    sizeof(request->idr_isid));
1303		request->idr_tsih = 0;	/* New or reinstated session. */
1304		memcpy(&request->idr_conf, &is->is_conf,
1305		    sizeof(request->idr_conf));
1306
1307		sx_sunlock(&sc->sc_lock);
1308		return (0);
1309	}
1310}
1311
1312static int
1313iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1314    struct iscsi_daemon_handoff *handoff)
1315{
1316	struct iscsi_session *is;
1317	int error;
1318
1319	sx_slock(&sc->sc_lock);
1320
1321	/*
1322	 * Find the session to hand off socket to.
1323	 */
1324	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1325		if (is->is_id == handoff->idh_session_id)
1326			break;
1327	}
1328	if (is == NULL) {
1329		sx_sunlock(&sc->sc_lock);
1330		return (ESRCH);
1331	}
1332	ISCSI_SESSION_LOCK(is);
1333	if (is->is_conf.isc_discovery || is->is_terminating) {
1334		ISCSI_SESSION_UNLOCK(is);
1335		sx_sunlock(&sc->sc_lock);
1336		return (EINVAL);
1337	}
1338	if (is->is_connected) {
1339		/*
1340		 * This might have happened because another iscsid(8)
1341		 * instance handed off the connection in the meantime.
1342		 * Just return.
1343		 */
1344		ISCSI_SESSION_WARN(is, "handoff on already connected "
1345		    "session");
1346		ISCSI_SESSION_UNLOCK(is);
1347		sx_sunlock(&sc->sc_lock);
1348		return (EBUSY);
1349	}
1350
1351	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1352	    sizeof(is->is_target_alias));
1353	is->is_tsih = handoff->idh_tsih;
1354	is->is_statsn = handoff->idh_statsn;
1355	is->is_initial_r2t = handoff->idh_initial_r2t;
1356	is->is_immediate_data = handoff->idh_immediate_data;
1357	is->is_max_data_segment_length = handoff->idh_max_data_segment_length;
1358	is->is_max_burst_length = handoff->idh_max_burst_length;
1359	is->is_first_burst_length = handoff->idh_first_burst_length;
1360
1361	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1362		is->is_conn->ic_header_crc32c = true;
1363	else
1364		is->is_conn->ic_header_crc32c = false;
1365	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1366		is->is_conn->ic_data_crc32c = true;
1367	else
1368		is->is_conn->ic_data_crc32c = false;
1369
1370	is->is_cmdsn = 0;
1371	is->is_expcmdsn = 0;
1372	is->is_maxcmdsn = 0;
1373	is->is_waiting_for_iscsid = false;
1374	is->is_login_phase = false;
1375	is->is_timeout = 0;
1376	is->is_connected = true;
1377	is->is_reason[0] = '\0';
1378
1379	ISCSI_SESSION_UNLOCK(is);
1380
1381#ifdef ICL_KERNEL_PROXY
1382	if (handoff->idh_socket != 0) {
1383#endif
1384		/*
1385		 * Handoff without using ICL proxy.
1386		 */
1387		error = icl_conn_handoff(is->is_conn, handoff->idh_socket);
1388		if (error != 0) {
1389			sx_sunlock(&sc->sc_lock);
1390			iscsi_session_terminate(is);
1391			return (error);
1392		}
1393#ifdef ICL_KERNEL_PROXY
1394	}
1395#endif
1396
1397	sx_sunlock(&sc->sc_lock);
1398
1399	if (is->is_sim != NULL) {
1400		/*
1401		 * When reconnecting, there already is SIM allocated for the session.
1402		 */
1403		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1404		ISCSI_SESSION_LOCK(is);
1405		ISCSI_SESSION_DEBUG(is, "releasing");
1406		xpt_release_simq(is->is_sim, 1);
1407		is->is_simq_frozen = false;
1408		ISCSI_SESSION_UNLOCK(is);
1409
1410	} else {
1411		ISCSI_SESSION_LOCK(is);
1412		is->is_devq = cam_simq_alloc(maxtags);
1413		if (is->is_devq == NULL) {
1414			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1415			iscsi_session_terminate(is);
1416			return (ENOMEM);
1417		}
1418
1419		is->is_sim = cam_sim_alloc(iscsi_action, iscsi_poll, "iscsi",
1420		    is, is->is_id /* unit */, &is->is_lock,
1421		    1, maxtags, is->is_devq);
1422		if (is->is_sim == NULL) {
1423			ISCSI_SESSION_UNLOCK(is);
1424			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1425			cam_simq_free(is->is_devq);
1426			iscsi_session_terminate(is);
1427			return (ENOMEM);
1428		}
1429
1430		error = xpt_bus_register(is->is_sim, NULL, 0);
1431		if (error != 0) {
1432			ISCSI_SESSION_UNLOCK(is);
1433			ISCSI_SESSION_WARN(is, "failed to register bus");
1434			iscsi_session_terminate(is);
1435			return (ENOMEM);
1436		}
1437
1438		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1439		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1440		    CAM_LUN_WILDCARD);
1441		if (error != CAM_REQ_CMP) {
1442			ISCSI_SESSION_UNLOCK(is);
1443			ISCSI_SESSION_WARN(is, "failed to create path");
1444			iscsi_session_terminate(is);
1445			return (ENOMEM);
1446		}
1447		ISCSI_SESSION_UNLOCK(is);
1448	}
1449
1450	return (0);
1451}
1452
1453static int
1454iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1455    struct iscsi_daemon_fail *fail)
1456{
1457	struct iscsi_session *is;
1458
1459	sx_slock(&sc->sc_lock);
1460
1461	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1462		if (is->is_id == fail->idf_session_id)
1463			break;
1464	}
1465	if (is == NULL) {
1466		sx_sunlock(&sc->sc_lock);
1467		return (ESRCH);
1468	}
1469	ISCSI_SESSION_LOCK(is);
1470	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1471	    fail->idf_reason);
1472	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1473	//is->is_waiting_for_iscsid = false;
1474	//is->is_login_phase = true;
1475	//iscsi_session_reconnect(is);
1476	ISCSI_SESSION_UNLOCK(is);
1477	sx_sunlock(&sc->sc_lock);
1478
1479	return (0);
1480}
1481
1482#ifdef ICL_KERNEL_PROXY
1483static int
1484iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1485    struct iscsi_daemon_connect *idc)
1486{
1487	struct iscsi_session *is;
1488	struct sockaddr *from_sa, *to_sa;
1489	int error;
1490
1491	sx_slock(&sc->sc_lock);
1492	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1493		if (is->is_id == idc->idc_session_id)
1494			break;
1495	}
1496	if (is == NULL) {
1497		sx_sunlock(&sc->sc_lock);
1498		return (ESRCH);
1499	}
1500	sx_sunlock(&sc->sc_lock);
1501
1502	if (idc->idc_from_addrlen > 0) {
1503		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1504		if (error != 0) {
1505			ISCSI_SESSION_WARN(is,
1506			    "getsockaddr failed with error %d", error);
1507			return (error);
1508		}
1509	} else {
1510		from_sa = NULL;
1511	}
1512	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1513	if (error != 0) {
1514		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1515		    error);
1516		free(from_sa, M_SONAME);
1517		return (error);
1518	}
1519
1520	ISCSI_SESSION_LOCK(is);
1521	is->is_waiting_for_iscsid = false;
1522	is->is_login_phase = true;
1523	is->is_timeout = 0;
1524	ISCSI_SESSION_UNLOCK(is);
1525
1526	error = icl_conn_connect(is->is_conn, idc->idc_iser, idc->idc_domain,
1527	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1528	free(from_sa, M_SONAME);
1529	free(to_sa, M_SONAME);
1530
1531	/*
1532	 * Digests are always disabled during login phase.
1533	 */
1534	is->is_conn->ic_header_crc32c = false;
1535	is->is_conn->ic_data_crc32c = false;
1536
1537	return (error);
1538}
1539
1540static int
1541iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1542    struct iscsi_daemon_send *ids)
1543{
1544	struct iscsi_session *is;
1545	struct icl_pdu *ip;
1546	size_t datalen;
1547	void *data;
1548	int error;
1549
1550	sx_slock(&sc->sc_lock);
1551	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1552		if (is->is_id == ids->ids_session_id)
1553			break;
1554	}
1555	if (is == NULL) {
1556		sx_sunlock(&sc->sc_lock);
1557		return (ESRCH);
1558	}
1559	sx_sunlock(&sc->sc_lock);
1560
1561	if (is->is_login_phase == false)
1562		return (EBUSY);
1563
1564	if (is->is_terminating || is->is_reconnecting)
1565		return (EIO);
1566
1567	datalen = ids->ids_data_segment_len;
1568	if (datalen > ISCSI_MAX_DATA_SEGMENT_LENGTH)
1569		return (EINVAL);
1570	if (datalen > 0) {
1571		data = malloc(datalen, M_ISCSI, M_WAITOK);
1572		error = copyin(ids->ids_data_segment, data, datalen);
1573		if (error != 0) {
1574			free(data, M_ISCSI);
1575			return (error);
1576		}
1577	}
1578
1579	ip = icl_pdu_new_bhs(is->is_conn, M_WAITOK);
1580	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1581	if (datalen > 0) {
1582		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1583		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1584		free(data, M_ISCSI);
1585	}
1586	icl_pdu_queue(ip);
1587
1588	return (0);
1589}
1590
1591static int
1592iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1593    struct iscsi_daemon_receive *idr)
1594{
1595	struct iscsi_session *is;
1596	struct icl_pdu *ip;
1597	void *data;
1598
1599	sx_slock(&sc->sc_lock);
1600	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1601		if (is->is_id == idr->idr_session_id)
1602			break;
1603	}
1604	if (is == NULL) {
1605		sx_sunlock(&sc->sc_lock);
1606		return (ESRCH);
1607	}
1608	sx_sunlock(&sc->sc_lock);
1609
1610	if (is->is_login_phase == false)
1611		return (EBUSY);
1612
1613	ISCSI_SESSION_LOCK(is);
1614	while (is->is_login_pdu == NULL &&
1615	    is->is_terminating == false &&
1616	    is->is_reconnecting == false)
1617		cv_wait(&is->is_login_cv, &is->is_lock);
1618	if (is->is_terminating || is->is_reconnecting) {
1619		ISCSI_SESSION_UNLOCK(is);
1620		return (EIO);
1621	}
1622	ip = is->is_login_pdu;
1623	is->is_login_pdu = NULL;
1624	ISCSI_SESSION_UNLOCK(is);
1625
1626	if (ip->ip_data_len > idr->idr_data_segment_len) {
1627		icl_pdu_free(ip);
1628		return (EMSGSIZE);
1629	}
1630
1631	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1632	if (ip->ip_data_len > 0) {
1633		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1634		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1635		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1636		free(data, M_ISCSI);
1637	}
1638
1639	icl_pdu_free(ip);
1640
1641	return (0);
1642}
1643#endif /* ICL_KERNEL_PROXY */
1644
1645static void
1646iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1647{
1648	/*
1649	 * Just make sure all the fields are null-terminated.
1650	 *
1651	 * XXX: This is not particularly secure.  We should
1652	 * 	create our own conf and then copy in relevant
1653	 * 	fields.
1654	 */
1655	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1656	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1657	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1658	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1659	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1660	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1661	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1662	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1663	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1664}
1665
1666static bool
1667iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1668{
1669
1670	if (isc->isc_initiator[0] == '\0') {
1671		ISCSI_DEBUG("empty isc_initiator");
1672		return (false);
1673	}
1674
1675	if (isc->isc_target_addr[0] == '\0') {
1676		ISCSI_DEBUG("empty isc_target_addr");
1677		return (false);
1678	}
1679
1680	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1681		ISCSI_DEBUG("non-empty isc_target for discovery session");
1682		return (false);
1683	}
1684
1685	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1686		ISCSI_DEBUG("empty isc_target for non-discovery session");
1687		return (false);
1688	}
1689
1690	return (true);
1691}
1692
1693static int
1694iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1695{
1696	struct iscsi_session *is;
1697	const struct iscsi_session *is2;
1698	int error;
1699
1700	iscsi_sanitize_session_conf(&isa->isa_conf);
1701	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1702		return (EINVAL);
1703
1704	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1705	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1706
1707	sx_xlock(&sc->sc_lock);
1708
1709	/*
1710	 * Prevent duplicates.
1711	 */
1712	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1713		if (!!is->is_conf.isc_discovery !=
1714		    !!is2->is_conf.isc_discovery)
1715			continue;
1716
1717		if (strcmp(is->is_conf.isc_target_addr,
1718		    is2->is_conf.isc_target_addr) != 0)
1719			continue;
1720
1721		if (is->is_conf.isc_discovery == 0 &&
1722		    strcmp(is->is_conf.isc_target,
1723		    is2->is_conf.isc_target) != 0)
1724			continue;
1725
1726		sx_xunlock(&sc->sc_lock);
1727		free(is, M_ISCSI);
1728		return (EBUSY);
1729	}
1730
1731	is->is_conn = icl_conn_new("iscsi", &is->is_lock);
1732	is->is_conn->ic_receive = iscsi_receive_callback;
1733	is->is_conn->ic_error = iscsi_error_callback;
1734	is->is_conn->ic_prv0 = is;
1735	TAILQ_INIT(&is->is_outstanding);
1736	STAILQ_INIT(&is->is_postponed);
1737	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1738	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1739#ifdef ICL_KERNEL_PROXY
1740	cv_init(&is->is_login_cv, "iscsi_login");
1741#endif
1742
1743	is->is_softc = sc;
1744	sc->sc_last_session_id++;
1745	is->is_id = sc->sc_last_session_id;
1746	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1747	arc4rand(&is->is_isid[1], 5, 0);
1748	is->is_tsih = 0;
1749	callout_init(&is->is_callout, 1);
1750	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1751	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1752
1753	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1754	if (error != 0) {
1755		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1756		return (error);
1757	}
1758
1759	/*
1760	 * Trigger immediate reconnection.
1761	 */
1762	ISCSI_SESSION_LOCK(is);
1763	is->is_waiting_for_iscsid = true;
1764	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1765	ISCSI_SESSION_UNLOCK(is);
1766	cv_signal(&sc->sc_cv);
1767
1768	sx_xunlock(&sc->sc_lock);
1769
1770	return (0);
1771}
1772
1773static bool
1774iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1775    unsigned int id2, const struct iscsi_session_conf *c2)
1776{
1777
1778	if (id2 != 0 && id2 != id1)
1779		return (false);
1780	if (c2->isc_target[0] != '\0' &&
1781	    strcmp(c1->isc_target, c2->isc_target) != 0)
1782		return (false);
1783	if (c2->isc_target_addr[0] != '\0' &&
1784	    strcmp(c1->isc_target_addr, c2->isc_target_addr) != 0)
1785		return (false);
1786	return (true);
1787}
1788
1789static int
1790iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1791    struct iscsi_session_remove *isr)
1792{
1793	struct iscsi_session *is, *tmp;
1794	bool found = false;
1795
1796	iscsi_sanitize_session_conf(&isr->isr_conf);
1797
1798	sx_xlock(&sc->sc_lock);
1799	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1800		ISCSI_SESSION_LOCK(is);
1801		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1802		    isr->isr_session_id, &isr->isr_conf)) {
1803			found = true;
1804			iscsi_session_logout(is);
1805			iscsi_session_terminate(is);
1806		}
1807		ISCSI_SESSION_UNLOCK(is);
1808	}
1809	sx_xunlock(&sc->sc_lock);
1810
1811	if (!found)
1812		return (ESRCH);
1813
1814	return (0);
1815}
1816
1817static int
1818iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1819{
1820	int error;
1821	unsigned int i = 0;
1822	struct iscsi_session *is;
1823	struct iscsi_session_state iss;
1824
1825	sx_slock(&sc->sc_lock);
1826	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1827		if (i >= isl->isl_nentries) {
1828			sx_sunlock(&sc->sc_lock);
1829			return (EMSGSIZE);
1830		}
1831		memset(&iss, 0, sizeof(iss));
1832		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1833		iss.iss_id = is->is_id;
1834		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1835		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1836
1837		if (is->is_conn->ic_header_crc32c)
1838			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1839		else
1840			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1841
1842		if (is->is_conn->ic_data_crc32c)
1843			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1844		else
1845			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1846
1847		iss.iss_max_data_segment_length = is->is_max_data_segment_length;
1848		iss.iss_immediate_data = is->is_immediate_data;
1849		iss.iss_connected = is->is_connected;
1850
1851		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1852		if (error != 0) {
1853			sx_sunlock(&sc->sc_lock);
1854			return (error);
1855		}
1856		i++;
1857	}
1858	sx_sunlock(&sc->sc_lock);
1859
1860	isl->isl_nentries = i;
1861
1862	return (0);
1863}
1864
1865static int
1866iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1867    struct iscsi_session_modify *ism)
1868{
1869	struct iscsi_session *is;
1870
1871	iscsi_sanitize_session_conf(&ism->ism_conf);
1872	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
1873		return (EINVAL);
1874
1875	sx_xlock(&sc->sc_lock);
1876	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1877		ISCSI_SESSION_LOCK(is);
1878		if (is->is_id == ism->ism_session_id)
1879			break;
1880		ISCSI_SESSION_UNLOCK(is);
1881	}
1882	if (is == NULL) {
1883		sx_xunlock(&sc->sc_lock);
1884		return (ESRCH);
1885	}
1886	sx_xunlock(&sc->sc_lock);
1887
1888	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
1889	ISCSI_SESSION_UNLOCK(is);
1890
1891	iscsi_session_reconnect(is);
1892
1893	return (0);
1894}
1895
1896static int
1897iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
1898    struct thread *td)
1899{
1900	struct iscsi_softc *sc;
1901
1902	sc = dev->si_drv1;
1903
1904	switch (cmd) {
1905	case ISCSIDWAIT:
1906		return (iscsi_ioctl_daemon_wait(sc,
1907		    (struct iscsi_daemon_request *)arg));
1908	case ISCSIDHANDOFF:
1909		return (iscsi_ioctl_daemon_handoff(sc,
1910		    (struct iscsi_daemon_handoff *)arg));
1911	case ISCSIDFAIL:
1912		return (iscsi_ioctl_daemon_fail(sc,
1913		    (struct iscsi_daemon_fail *)arg));
1914#ifdef ICL_KERNEL_PROXY
1915	case ISCSIDCONNECT:
1916		return (iscsi_ioctl_daemon_connect(sc,
1917		    (struct iscsi_daemon_connect *)arg));
1918	case ISCSIDSEND:
1919		return (iscsi_ioctl_daemon_send(sc,
1920		    (struct iscsi_daemon_send *)arg));
1921	case ISCSIDRECEIVE:
1922		return (iscsi_ioctl_daemon_receive(sc,
1923		    (struct iscsi_daemon_receive *)arg));
1924#endif /* ICL_KERNEL_PROXY */
1925	case ISCSISADD:
1926		return (iscsi_ioctl_session_add(sc,
1927		    (struct iscsi_session_add *)arg));
1928	case ISCSISREMOVE:
1929		return (iscsi_ioctl_session_remove(sc,
1930		    (struct iscsi_session_remove *)arg));
1931	case ISCSISLIST:
1932		return (iscsi_ioctl_session_list(sc,
1933		    (struct iscsi_session_list *)arg));
1934	case ISCSISMODIFY:
1935		return (iscsi_ioctl_session_modify(sc,
1936		    (struct iscsi_session_modify *)arg));
1937	default:
1938		return (EINVAL);
1939	}
1940}
1941
1942static uint64_t
1943iscsi_encode_lun(uint32_t lun)
1944{
1945	uint8_t encoded[8];
1946	uint64_t result;
1947
1948	memset(encoded, 0, sizeof(encoded));
1949
1950	if (lun < 256) {
1951		/*
1952		 * Peripheral device addressing.
1953		 */
1954		encoded[1] = lun;
1955	} else if (lun < 16384) {
1956		/*
1957		 * Flat space addressing.
1958		 */
1959		encoded[0] = 0x40;
1960		encoded[0] |= (lun >> 8) & 0x3f;
1961		encoded[1] = lun & 0xff;
1962	} else {
1963		/*
1964		 * Extended flat space addressing.
1965		 */
1966		encoded[0] = 0xd2;
1967		encoded[1] = lun >> 16;
1968		encoded[2] = lun >> 8;
1969		encoded[3] = lun;
1970	}
1971
1972	memcpy(&result, encoded, sizeof(result));
1973	return (result);
1974}
1975
1976static struct iscsi_outstanding *
1977iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
1978{
1979	struct iscsi_outstanding *io;
1980
1981	ISCSI_SESSION_LOCK_ASSERT(is);
1982
1983	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1984		if (io->io_initiator_task_tag == initiator_task_tag)
1985			return (io);
1986	}
1987	return (NULL);
1988}
1989
1990static struct iscsi_outstanding *
1991iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
1992{
1993	struct iscsi_outstanding *io;
1994
1995	ISCSI_SESSION_LOCK_ASSERT(is);
1996
1997	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
1998		if (io->io_ccb == ccb)
1999			return (io);
2000	}
2001	return (NULL);
2002}
2003
2004static struct iscsi_outstanding *
2005iscsi_outstanding_add(struct iscsi_session *is,
2006    uint32_t initiator_task_tag, union ccb *ccb)
2007{
2008	struct iscsi_outstanding *io;
2009
2010	ISCSI_SESSION_LOCK_ASSERT(is);
2011
2012	KASSERT(iscsi_outstanding_find(is, initiator_task_tag) == NULL,
2013	    ("initiator_task_tag 0x%x already added", initiator_task_tag));
2014
2015	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
2016	if (io == NULL) {
2017		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes", sizeof(*io));
2018		return (NULL);
2019	}
2020	io->io_initiator_task_tag = initiator_task_tag;
2021	io->io_ccb = ccb;
2022	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
2023	return (io);
2024}
2025
2026static void
2027iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2028{
2029
2030	ISCSI_SESSION_LOCK_ASSERT(is);
2031
2032	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2033	uma_zfree(iscsi_outstanding_zone, io);
2034}
2035
2036static void
2037iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2038{
2039	struct icl_pdu *request;
2040	struct iscsi_bhs_task_management_request *bhstmr;
2041	struct ccb_abort *cab = &ccb->cab;
2042	struct iscsi_outstanding *io, *aio;
2043
2044	ISCSI_SESSION_LOCK_ASSERT(is);
2045
2046#if 0
2047	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2048#else
2049	if (is->is_login_phase) {
2050		ccb->ccb_h.status = CAM_REQ_ABORTED;
2051		xpt_done(ccb);
2052		return;
2053	}
2054#endif
2055
2056	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2057	if (aio == NULL) {
2058		ccb->ccb_h.status = CAM_REQ_CMP;
2059		xpt_done(ccb);
2060		return;
2061	}
2062
2063	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
2064	if (request == NULL) {
2065		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2066		xpt_done(ccb);
2067		return;
2068	}
2069
2070	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2071	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2072	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2073
2074	bhstmr->bhstmr_lun = iscsi_encode_lun(ccb->ccb_h.target_lun);
2075	bhstmr->bhstmr_initiator_task_tag = is->is_initiator_task_tag;
2076	is->is_initiator_task_tag++;
2077	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2078
2079	io = iscsi_outstanding_add(is, bhstmr->bhstmr_initiator_task_tag, NULL);
2080	if (io == NULL) {
2081		icl_pdu_free(request);
2082		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2083		xpt_done(ccb);
2084		return;
2085	}
2086	io->io_datasn = aio->io_initiator_task_tag;
2087	iscsi_pdu_queue_locked(request);
2088}
2089
2090static void
2091iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2092{
2093	struct icl_pdu *request;
2094	struct iscsi_bhs_scsi_command *bhssc;
2095	struct ccb_scsiio *csio;
2096	struct iscsi_outstanding *io;
2097	size_t len;
2098	int error;
2099
2100	ISCSI_SESSION_LOCK_ASSERT(is);
2101
2102#if 0
2103	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2104#else
2105	if (is->is_login_phase) {
2106		ISCSI_SESSION_DEBUG(is, "called during login phase");
2107		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2108			xpt_freeze_devq(ccb->ccb_h.path, 1);
2109			ISCSI_SESSION_DEBUG(is, "freezing devq");
2110		}
2111		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2112		xpt_done(ccb);
2113		return;
2114	}
2115#endif
2116
2117	request = icl_pdu_new_bhs(is->is_conn, M_NOWAIT);
2118	if (request == NULL) {
2119		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2120			xpt_freeze_devq(ccb->ccb_h.path, 1);
2121			ISCSI_SESSION_DEBUG(is, "freezing devq");
2122		}
2123		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2124		xpt_done(ccb);
2125		return;
2126	}
2127
2128	csio = &ccb->csio;
2129	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2130	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2131	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2132	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2133	case CAM_DIR_IN:
2134		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2135		break;
2136	case CAM_DIR_OUT:
2137		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2138		break;
2139	}
2140
2141	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2142		switch (csio->tag_action) {
2143		case MSG_HEAD_OF_Q_TAG:
2144			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2145			break;
2146		case MSG_ORDERED_Q_TAG:
2147			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2148			break;
2149		case MSG_ACA_TASK:
2150			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2151			break;
2152		case MSG_SIMPLE_Q_TAG:
2153		default:
2154			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2155			break;
2156		}
2157	} else
2158		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2159
2160	bhssc->bhssc_lun = iscsi_encode_lun(csio->ccb_h.target_lun);
2161	bhssc->bhssc_initiator_task_tag = is->is_initiator_task_tag;
2162	is->is_initiator_task_tag++;
2163	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2164	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2165	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2166
2167	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2168		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2169	else
2170		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2171
2172	io = iscsi_outstanding_add(is, bhssc->bhssc_initiator_task_tag, ccb);
2173	if (io == NULL) {
2174		icl_pdu_free(request);
2175		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2176			xpt_freeze_devq(ccb->ccb_h.path, 1);
2177			ISCSI_SESSION_DEBUG(is, "freezing devq");
2178		}
2179		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2180		xpt_done(ccb);
2181		return;
2182	}
2183
2184	if (is->is_immediate_data &&
2185	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2186		len = csio->dxfer_len;
2187		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2188		if (len > is->is_first_burst_length) {
2189			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_first_burst_length);
2190			len = is->is_first_burst_length;
2191		}
2192		if (len > is->is_max_data_segment_length) {
2193			ISCSI_SESSION_DEBUG(is, "len %zd -> %zd", len, is->is_max_data_segment_length);
2194			len = is->is_max_data_segment_length;
2195		}
2196
2197		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2198		if (error != 0) {
2199			icl_pdu_free(request);
2200			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2201				xpt_freeze_devq(ccb->ccb_h.path, 1);
2202				ISCSI_SESSION_DEBUG(is, "freezing devq");
2203			}
2204			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2205			xpt_done(ccb);
2206			return;
2207		}
2208	}
2209	iscsi_pdu_queue_locked(request);
2210}
2211
2212static void
2213iscsi_action(struct cam_sim *sim, union ccb *ccb)
2214{
2215	struct iscsi_session *is;
2216
2217	is = cam_sim_softc(sim);
2218
2219	ISCSI_SESSION_LOCK_ASSERT(is);
2220
2221	if (is->is_terminating ||
2222	    (is->is_connected == false && fail_on_disconnection)) {
2223		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2224		xpt_done(ccb);
2225		return;
2226	}
2227
2228	switch (ccb->ccb_h.func_code) {
2229	case XPT_PATH_INQ:
2230	{
2231		struct ccb_pathinq *cpi = &ccb->cpi;
2232
2233		cpi->version_num = 1;
2234		cpi->hba_inquiry = PI_TAG_ABLE;
2235		cpi->target_sprt = 0;
2236		//cpi->hba_misc = PIM_NOBUSRESET;
2237		cpi->hba_misc = 0;
2238		cpi->hba_eng_cnt = 0;
2239		cpi->max_target = 0;
2240		cpi->max_lun = 255;
2241		//cpi->initiator_id = 0; /* XXX */
2242		cpi->initiator_id = 64; /* XXX */
2243		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2244		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2245		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2246		cpi->unit_number = cam_sim_unit(sim);
2247		cpi->bus_id = cam_sim_bus(sim);
2248		cpi->base_transfer_speed = 150000; /* XXX */
2249		cpi->transport = XPORT_ISCSI;
2250		cpi->transport_version = 0;
2251		cpi->protocol = PROTO_SCSI;
2252		cpi->protocol_version = SCSI_REV_SPC3;
2253		cpi->maxio = MAXPHYS;
2254		cpi->ccb_h.status = CAM_REQ_CMP;
2255		break;
2256	}
2257	case XPT_GET_TRAN_SETTINGS:
2258	{
2259		struct ccb_trans_settings	*cts;
2260		struct ccb_trans_settings_scsi	*scsi;
2261
2262		cts = &ccb->cts;
2263		scsi = &cts->proto_specific.scsi;
2264
2265		cts->protocol = PROTO_SCSI;
2266		cts->protocol_version = SCSI_REV_SPC3;
2267		cts->transport = XPORT_ISCSI;
2268		cts->transport_version = 0;
2269		scsi->valid = CTS_SCSI_VALID_TQ;
2270		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2271		cts->ccb_h.status = CAM_REQ_CMP;
2272		break;
2273	}
2274	case XPT_CALC_GEOMETRY:
2275		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2276		ccb->ccb_h.status = CAM_REQ_CMP;
2277		break;
2278#if 0
2279	/*
2280	 * XXX: What's the point?
2281	 */
2282	case XPT_RESET_BUS:
2283	case XPT_TERM_IO:
2284		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2285		ccb->ccb_h.status = CAM_REQ_CMP;
2286		break;
2287#endif
2288	case XPT_ABORT:
2289		iscsi_action_abort(is, ccb);
2290		return;
2291	case XPT_SCSI_IO:
2292		iscsi_action_scsiio(is, ccb);
2293		return;
2294	default:
2295#if 0
2296		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2297#endif
2298		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2299		break;
2300	}
2301	xpt_done(ccb);
2302}
2303
2304static void
2305iscsi_poll(struct cam_sim *sim)
2306{
2307
2308	KASSERT(0, ("%s: you're not supposed to be here", __func__));
2309}
2310
2311static void
2312iscsi_shutdown(struct iscsi_softc *sc)
2313{
2314	struct iscsi_session *is;
2315
2316	ISCSI_DEBUG("removing all sessions due to shutdown");
2317
2318	sx_slock(&sc->sc_lock);
2319	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2320		iscsi_session_terminate(is);
2321	sx_sunlock(&sc->sc_lock);
2322}
2323
2324static int
2325iscsi_load(void)
2326{
2327	int error;
2328
2329	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2330	sx_init(&sc->sc_lock, "iscsi");
2331	TAILQ_INIT(&sc->sc_sessions);
2332	cv_init(&sc->sc_cv, "iscsi_cv");
2333
2334	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2335	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2336	    UMA_ALIGN_PTR, 0);
2337
2338	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2339	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2340	if (error != 0) {
2341		ISCSI_WARN("failed to create device node, error %d", error);
2342		return (error);
2343	}
2344	sc->sc_cdev->si_drv1 = sc;
2345
2346	/*
2347	 * Note that this needs to get run before dashutdown().  Otherwise,
2348	 * when rebooting with iSCSI session with outstanding requests,
2349	 * but disconnected, dashutdown() will hang on cam_periph_runccb().
2350	 */
2351	sc->sc_shutdown_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2352	    iscsi_shutdown, sc, SHUTDOWN_PRI_FIRST);
2353
2354	return (0);
2355}
2356
2357static int
2358iscsi_unload(void)
2359{
2360	struct iscsi_session *is, *tmp;
2361
2362	if (sc->sc_cdev != NULL) {
2363		ISCSI_DEBUG("removing device node");
2364		destroy_dev(sc->sc_cdev);
2365		ISCSI_DEBUG("device node removed");
2366	}
2367
2368	if (sc->sc_shutdown_eh != NULL)
2369		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_eh);
2370
2371	sx_slock(&sc->sc_lock);
2372	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp)
2373		iscsi_session_terminate(is);
2374	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2375		ISCSI_DEBUG("waiting for sessions to terminate");
2376		cv_wait(&sc->sc_cv, &sc->sc_lock);
2377	}
2378	ISCSI_DEBUG("all sessions terminated");
2379	sx_sunlock(&sc->sc_lock);
2380
2381	uma_zdestroy(iscsi_outstanding_zone);
2382	sx_destroy(&sc->sc_lock);
2383	cv_destroy(&sc->sc_cv);
2384	free(sc, M_ISCSI);
2385	return (0);
2386}
2387
2388static int
2389iscsi_quiesce(void)
2390{
2391	sx_slock(&sc->sc_lock);
2392	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2393		sx_sunlock(&sc->sc_lock);
2394		return (EBUSY);
2395	}
2396	sx_sunlock(&sc->sc_lock);
2397	return (0);
2398}
2399
2400static int
2401iscsi_modevent(module_t mod, int what, void *arg)
2402{
2403	int error;
2404
2405	switch (what) {
2406	case MOD_LOAD:
2407		error = iscsi_load();
2408		break;
2409	case MOD_UNLOAD:
2410		error = iscsi_unload();
2411		break;
2412	case MOD_QUIESCE:
2413		error = iscsi_quiesce();
2414		break;
2415	default:
2416		error = EINVAL;
2417		break;
2418	}
2419	return (error);
2420}
2421
2422moduledata_t iscsi_data = {
2423	"iscsi",
2424	iscsi_modevent,
2425	0
2426};
2427
2428DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2429MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2430MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2431