xref: /freebsd/sys/dev/iscsi/iscsi.c (revision 1f88aa09417f1cfb3929fd37531b1ab51213c2d6)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2012 The FreeBSD Foundation
5  *
6  * This software was developed by Edward Tomasz Napierala under sponsorship
7  * from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/condvar.h>
37 #include <sys/conf.h>
38 #include <sys/endian.h>
39 #include <sys/eventhandler.h>
40 #include <sys/file.h>
41 #include <sys/kernel.h>
42 #include <sys/kthread.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mutex.h>
46 #include <sys/module.h>
47 #include <sys/socket.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 #include <sys/sx.h>
51 #include <vm/uma.h>
52 
53 #include <cam/cam.h>
54 #include <cam/cam_ccb.h>
55 #include <cam/cam_xpt.h>
56 #include <cam/cam_debug.h>
57 #include <cam/cam_sim.h>
58 #include <cam/cam_xpt_sim.h>
59 #include <cam/cam_xpt_periph.h>
60 #include <cam/cam_periph.h>
61 #include <cam/scsi/scsi_all.h>
62 #include <cam/scsi/scsi_message.h>
63 
64 #include <dev/iscsi/icl.h>
65 #include <dev/iscsi/icl_wrappers.h>
66 #include <dev/iscsi/iscsi_ioctl.h>
67 #include <dev/iscsi/iscsi_proto.h>
68 #include <dev/iscsi/iscsi.h>
69 
70 #ifdef ICL_KERNEL_PROXY
71 #include <sys/socketvar.h>
72 #endif
73 
74 #ifdef ICL_KERNEL_PROXY
75 FEATURE(iscsi_kernel_proxy, "iSCSI initiator built with ICL_KERNEL_PROXY");
76 #endif
77 
78 /*
79  * XXX: This is global so the iscsi_unload() can access it.
80  * 	Think about how to do this properly.
81  */
82 static struct iscsi_softc	*sc;
83 
84 SYSCTL_NODE(_kern, OID_AUTO, iscsi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
85     "iSCSI initiator");
86 static int debug = 1;
87 SYSCTL_INT(_kern_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN,
88     &debug, 0, "Enable debug messages");
89 static int ping_timeout = 5;
90 SYSCTL_INT(_kern_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout,
91     0, "Timeout for ping (NOP-Out) requests, in seconds");
92 static int iscsid_timeout = 60;
93 SYSCTL_INT(_kern_iscsi, OID_AUTO, iscsid_timeout, CTLFLAG_RWTUN, &iscsid_timeout,
94     0, "Time to wait for iscsid(8) to handle reconnection, in seconds");
95 static int login_timeout = 60;
96 SYSCTL_INT(_kern_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout,
97     0, "Time to wait for iscsid(8) to finish Login Phase, in seconds");
98 static int maxtags = 255;
99 SYSCTL_INT(_kern_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags,
100     0, "Max number of IO requests queued");
101 static int fail_on_disconnection = 0;
102 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_disconnection, CTLFLAG_RWTUN,
103     &fail_on_disconnection, 0, "Destroy CAM SIM on connection failure");
104 static int fail_on_shutdown = 1;
105 SYSCTL_INT(_kern_iscsi, OID_AUTO, fail_on_shutdown, CTLFLAG_RWTUN,
106     &fail_on_shutdown, 0, "Fail disconnected sessions on shutdown");
107 
108 static MALLOC_DEFINE(M_ISCSI, "iSCSI", "iSCSI initiator");
109 static uma_zone_t iscsi_outstanding_zone;
110 
111 #define	CONN_SESSION(X)	((struct iscsi_session *)X->ic_prv0)
112 #define	PDU_SESSION(X)	(CONN_SESSION(X->ip_conn))
113 
114 #define	ISCSI_DEBUG(X, ...)						\
115 	do {								\
116 		if (debug > 1) 						\
117 			printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
118 	} while (0)
119 
120 #define	ISCSI_WARN(X, ...)						\
121 	do {								\
122 		if (debug > 0) {					\
123 			printf("WARNING: %s: " X "\n",			\
124 			    __func__, ## __VA_ARGS__);			\
125 		}							\
126 	} while (0)
127 
128 #define	ISCSI_SESSION_DEBUG(S, X, ...)					\
129 	do {								\
130 		if (debug > 1) {					\
131 			printf("%s: %s (%s): " X "\n",			\
132 			    __func__, S->is_conf.isc_target_addr,	\
133 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
134 		}							\
135 	} while (0)
136 
137 #define	ISCSI_SESSION_WARN(S, X, ...)					\
138 	do {								\
139 		if (debug > 0) {					\
140 			printf("WARNING: %s (%s): " X "\n",		\
141 			    S->is_conf.isc_target_addr,			\
142 			    S->is_conf.isc_target, ## __VA_ARGS__);	\
143 		}							\
144 	} while (0)
145 
146 #define ISCSI_SESSION_LOCK(X)		mtx_lock(&X->is_lock)
147 #define ISCSI_SESSION_UNLOCK(X)		mtx_unlock(&X->is_lock)
148 #define ISCSI_SESSION_LOCK_ASSERT(X)	mtx_assert(&X->is_lock, MA_OWNED)
149 #define ISCSI_SESSION_LOCK_ASSERT_NOT(X) mtx_assert(&X->is_lock, MA_NOTOWNED)
150 
151 static int	iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg,
152 		    int mode, struct thread *td);
153 
154 static struct cdevsw iscsi_cdevsw = {
155      .d_version = D_VERSION,
156      .d_ioctl   = iscsi_ioctl,
157      .d_name    = "iscsi",
158 };
159 
160 static void	iscsi_pdu_queue_locked(struct icl_pdu *request);
161 static void	iscsi_pdu_queue(struct icl_pdu *request);
162 static void	iscsi_pdu_update_statsn(const struct icl_pdu *response);
163 static void	iscsi_pdu_handle_nop_in(struct icl_pdu *response);
164 static void	iscsi_pdu_handle_scsi_response(struct icl_pdu *response);
165 static void	iscsi_pdu_handle_task_response(struct icl_pdu *response);
166 static void	iscsi_pdu_handle_data_in(struct icl_pdu *response);
167 static void	iscsi_pdu_handle_logout_response(struct icl_pdu *response);
168 static void	iscsi_pdu_handle_r2t(struct icl_pdu *response);
169 static void	iscsi_pdu_handle_async_message(struct icl_pdu *response);
170 static void	iscsi_pdu_handle_reject(struct icl_pdu *response);
171 static void	iscsi_session_reconnect(struct iscsi_session *is);
172 static void	iscsi_session_terminate(struct iscsi_session *is);
173 static void	iscsi_action(struct cam_sim *sim, union ccb *ccb);
174 static struct iscsi_outstanding	*iscsi_outstanding_find(struct iscsi_session *is,
175 		    uint32_t initiator_task_tag);
176 static struct iscsi_outstanding	*iscsi_outstanding_add(struct iscsi_session *is,
177 		    struct icl_pdu *request, union ccb *ccb,
178 		    uint32_t *initiator_task_tagp);
179 static void	iscsi_outstanding_remove(struct iscsi_session *is,
180 		    struct iscsi_outstanding *io);
181 
182 static bool
183 iscsi_pdu_prepare(struct icl_pdu *request)
184 {
185 	struct iscsi_session *is;
186 	struct iscsi_bhs_scsi_command *bhssc;
187 
188 	is = PDU_SESSION(request);
189 
190 	ISCSI_SESSION_LOCK_ASSERT(is);
191 
192 	/*
193 	 * We're only using fields common for all the request
194 	 * (initiator -> target) PDUs.
195 	 */
196 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
197 
198 	/*
199 	 * Data-Out PDU does not contain CmdSN.
200 	 */
201 	if (bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_OUT) {
202 		if (ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn) &&
203 		    (bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) {
204 			/*
205 			 * Current MaxCmdSN prevents us from sending any more
206 			 * SCSI Command PDUs to the target; postpone the PDU.
207 			 * It will get resent by either iscsi_pdu_queue(),
208 			 * or by maintenance thread.
209 			 */
210 #if 0
211 			ISCSI_SESSION_DEBUG(is, "postponing send, CmdSN %u, "
212 			    "ExpCmdSN %u, MaxCmdSN %u, opcode 0x%x",
213 			    is->is_cmdsn, is->is_expcmdsn, is->is_maxcmdsn,
214 			    bhssc->bhssc_opcode);
215 #endif
216 			return (true);
217 		}
218 		bhssc->bhssc_cmdsn = htonl(is->is_cmdsn);
219 		if ((bhssc->bhssc_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0)
220 			is->is_cmdsn++;
221 	}
222 	bhssc->bhssc_expstatsn = htonl(is->is_statsn + 1);
223 
224 	return (false);
225 }
226 
227 static void
228 iscsi_session_send_postponed(struct iscsi_session *is)
229 {
230 	struct icl_pdu *request;
231 	bool postpone;
232 
233 	ISCSI_SESSION_LOCK_ASSERT(is);
234 
235 	if (STAILQ_EMPTY(&is->is_postponed))
236 		return;
237 	while ((request = STAILQ_FIRST(&is->is_postponed)) != NULL) {
238 		postpone = iscsi_pdu_prepare(request);
239 		if (postpone)
240 			return;
241 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
242 		icl_pdu_queue(request);
243 	}
244 	xpt_release_simq(is->is_sim, 1);
245 }
246 
247 static void
248 iscsi_pdu_queue_locked(struct icl_pdu *request)
249 {
250 	struct iscsi_session *is;
251 	bool postpone;
252 
253 	is = PDU_SESSION(request);
254 	ISCSI_SESSION_LOCK_ASSERT(is);
255 	iscsi_session_send_postponed(is);
256 	postpone = iscsi_pdu_prepare(request);
257 	if (postpone) {
258 		if (STAILQ_EMPTY(&is->is_postponed))
259 			xpt_freeze_simq(is->is_sim, 1);
260 		STAILQ_INSERT_TAIL(&is->is_postponed, request, ip_next);
261 		return;
262 	}
263 	icl_pdu_queue(request);
264 }
265 
266 static void
267 iscsi_pdu_queue(struct icl_pdu *request)
268 {
269 	struct iscsi_session *is;
270 
271 	is = PDU_SESSION(request);
272 	ISCSI_SESSION_LOCK(is);
273 	iscsi_pdu_queue_locked(request);
274 	ISCSI_SESSION_UNLOCK(is);
275 }
276 
277 static void
278 iscsi_session_logout(struct iscsi_session *is)
279 {
280 	struct icl_pdu *request;
281 	struct iscsi_bhs_logout_request *bhslr;
282 
283 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
284 	if (request == NULL)
285 		return;
286 
287 	bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs;
288 	bhslr->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_REQUEST;
289 	bhslr->bhslr_reason = BHSLR_REASON_CLOSE_SESSION;
290 	iscsi_pdu_queue_locked(request);
291 }
292 
293 static void
294 iscsi_session_terminate_task(struct iscsi_session *is,
295     struct iscsi_outstanding *io, cam_status status)
296 {
297 
298 	ISCSI_SESSION_LOCK_ASSERT(is);
299 
300 	if (io->io_ccb != NULL) {
301 		io->io_ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
302 		io->io_ccb->ccb_h.status |= status;
303 		if ((io->io_ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
304 			io->io_ccb->ccb_h.status |= CAM_DEV_QFRZN;
305 			xpt_freeze_devq(io->io_ccb->ccb_h.path, 1);
306 			ISCSI_SESSION_DEBUG(is, "freezing devq");
307 		}
308 		xpt_done(io->io_ccb);
309 	}
310 	iscsi_outstanding_remove(is, io);
311 }
312 
313 static void
314 iscsi_session_terminate_tasks(struct iscsi_session *is, cam_status status)
315 {
316 	struct iscsi_outstanding *io, *tmp;
317 
318 	ISCSI_SESSION_LOCK_ASSERT(is);
319 
320 	TAILQ_FOREACH_SAFE(io, &is->is_outstanding, io_next, tmp) {
321 		iscsi_session_terminate_task(is, io, status);
322 	}
323 }
324 
325 static void
326 iscsi_session_cleanup(struct iscsi_session *is, bool destroy_sim)
327 {
328 	struct icl_pdu *pdu;
329 
330 	ISCSI_SESSION_LOCK_ASSERT(is);
331 
332 	/*
333 	 * Don't queue any new PDUs.
334 	 */
335 	if (is->is_sim != NULL && is->is_simq_frozen == false) {
336 		ISCSI_SESSION_DEBUG(is, "freezing");
337 		xpt_freeze_simq(is->is_sim, 1);
338 		is->is_simq_frozen = true;
339 	}
340 
341 	/*
342 	 * Remove postponed PDUs.
343 	 */
344 	if (!STAILQ_EMPTY(&is->is_postponed))
345 		xpt_release_simq(is->is_sim, 1);
346 	while ((pdu = STAILQ_FIRST(&is->is_postponed)) != NULL) {
347 		STAILQ_REMOVE_HEAD(&is->is_postponed, ip_next);
348 		icl_pdu_free(pdu);
349 	}
350 
351 	if (destroy_sim == false) {
352 		/*
353 		 * Terminate SCSI tasks, asking CAM to requeue them.
354 		 */
355 		iscsi_session_terminate_tasks(is, CAM_REQUEUE_REQ);
356 		return;
357 	}
358 
359 	iscsi_session_terminate_tasks(is, CAM_DEV_NOT_THERE);
360 
361 	if (is->is_sim == NULL)
362 		return;
363 
364 	ISCSI_SESSION_DEBUG(is, "deregistering SIM");
365 	xpt_async(AC_LOST_DEVICE, is->is_path, NULL);
366 
367 	if (is->is_simq_frozen) {
368 		is->is_simq_frozen = false;
369 		xpt_release_simq(is->is_sim, 1);
370 	}
371 
372 	xpt_free_path(is->is_path);
373 	is->is_path = NULL;
374 	xpt_bus_deregister(cam_sim_path(is->is_sim));
375 	cam_sim_free(is->is_sim, TRUE /*free_devq*/);
376 	is->is_sim = NULL;
377 	is->is_devq = NULL;
378 }
379 
380 static void
381 iscsi_maintenance_thread_reconnect(struct iscsi_session *is)
382 {
383 
384 	icl_conn_close(is->is_conn);
385 
386 	ISCSI_SESSION_LOCK(is);
387 
388 	is->is_connected = false;
389 	is->is_reconnecting = false;
390 	is->is_login_phase = false;
391 
392 #ifdef ICL_KERNEL_PROXY
393 	if (is->is_login_pdu != NULL) {
394 		icl_pdu_free(is->is_login_pdu);
395 		is->is_login_pdu = NULL;
396 	}
397 	cv_signal(&is->is_login_cv);
398 #endif
399 
400 	if (fail_on_disconnection) {
401 		ISCSI_SESSION_DEBUG(is, "connection failed, destroying devices");
402 		iscsi_session_cleanup(is, true);
403 	} else {
404 		iscsi_session_cleanup(is, false);
405 	}
406 
407 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
408 	    ("destroying session with active tasks"));
409 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
410 	    ("destroying session with postponed PDUs"));
411 
412 	if (is->is_conf.isc_enable == 0 && is->is_conf.isc_discovery == 0) {
413 		ISCSI_SESSION_UNLOCK(is);
414 		return;
415 	}
416 
417 	/*
418 	 * Request immediate reconnection from iscsid(8).
419 	 */
420 	//ISCSI_SESSION_DEBUG(is, "waking up iscsid(8)");
421 	is->is_waiting_for_iscsid = true;
422 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
423 	is->is_timeout = 0;
424 	ISCSI_SESSION_UNLOCK(is);
425 	cv_signal(&is->is_softc->sc_cv);
426 }
427 
428 static void
429 iscsi_maintenance_thread_terminate(struct iscsi_session *is)
430 {
431 	struct iscsi_softc *sc;
432 
433 	sc = is->is_softc;
434 	sx_xlock(&sc->sc_lock);
435 	TAILQ_REMOVE(&sc->sc_sessions, is, is_next);
436 	sx_xunlock(&sc->sc_lock);
437 
438 	icl_conn_close(is->is_conn);
439 	callout_drain(&is->is_callout);
440 
441 	ISCSI_SESSION_LOCK(is);
442 
443 	KASSERT(is->is_terminating, ("is_terminating == false"));
444 
445 #ifdef ICL_KERNEL_PROXY
446 	if (is->is_login_pdu != NULL) {
447 		icl_pdu_free(is->is_login_pdu);
448 		is->is_login_pdu = NULL;
449 	}
450 	cv_signal(&is->is_login_cv);
451 #endif
452 
453 	iscsi_session_cleanup(is, true);
454 
455 	KASSERT(TAILQ_EMPTY(&is->is_outstanding),
456 	    ("destroying session with active tasks"));
457 	KASSERT(STAILQ_EMPTY(&is->is_postponed),
458 	    ("destroying session with postponed PDUs"));
459 
460 	ISCSI_SESSION_UNLOCK(is);
461 
462 	icl_conn_free(is->is_conn);
463 	mtx_destroy(&is->is_lock);
464 	cv_destroy(&is->is_maintenance_cv);
465 #ifdef ICL_KERNEL_PROXY
466 	cv_destroy(&is->is_login_cv);
467 #endif
468 
469 	ISCSI_SESSION_DEBUG(is, "terminated");
470 	free(is, M_ISCSI);
471 
472 	/*
473 	 * The iscsi_unload() routine might be waiting.
474 	 */
475 	cv_signal(&sc->sc_cv);
476 }
477 
478 static void
479 iscsi_maintenance_thread(void *arg)
480 {
481 	struct iscsi_session *is = arg;
482 
483 	ISCSI_SESSION_LOCK(is);
484 	for (;;) {
485 		if (is->is_reconnecting == false &&
486 		    is->is_terminating == false &&
487 		    (STAILQ_EMPTY(&is->is_postponed) ||
488 		     ISCSI_SNGT(is->is_cmdsn, is->is_maxcmdsn)))
489 			cv_wait(&is->is_maintenance_cv, &is->is_lock);
490 
491 		/* Terminate supersedes reconnect. */
492 		if (is->is_terminating) {
493 			ISCSI_SESSION_UNLOCK(is);
494 			iscsi_maintenance_thread_terminate(is);
495 			kthread_exit();
496 			return;
497 		}
498 
499 		if (is->is_reconnecting) {
500 			ISCSI_SESSION_UNLOCK(is);
501 			iscsi_maintenance_thread_reconnect(is);
502 			ISCSI_SESSION_LOCK(is);
503 			continue;
504 		}
505 
506 		iscsi_session_send_postponed(is);
507 	}
508 	ISCSI_SESSION_UNLOCK(is);
509 }
510 
511 static void
512 iscsi_session_reconnect(struct iscsi_session *is)
513 {
514 
515 	/*
516 	 * XXX: We can't use locking here, because
517 	 * 	it's being called from various contexts.
518 	 * 	Hope it doesn't break anything.
519 	 */
520 	if (is->is_reconnecting)
521 		return;
522 
523 	is->is_reconnecting = true;
524 	cv_signal(&is->is_maintenance_cv);
525 }
526 
527 static void
528 iscsi_session_terminate(struct iscsi_session *is)
529 {
530 
531 	if (is->is_terminating)
532 		return;
533 
534 	is->is_terminating = true;
535 
536 #if 0
537 	iscsi_session_logout(is);
538 #endif
539 	cv_signal(&is->is_maintenance_cv);
540 }
541 
542 static void
543 iscsi_callout(void *context)
544 {
545 	struct icl_pdu *request;
546 	struct iscsi_bhs_nop_out *bhsno;
547 	struct iscsi_session *is;
548 	bool reconnect_needed = false;
549 
550 	is = context;
551 
552 	ISCSI_SESSION_LOCK(is);
553 	if (is->is_terminating) {
554 		ISCSI_SESSION_UNLOCK(is);
555 		return;
556 	}
557 
558 	callout_schedule(&is->is_callout, 1 * hz);
559 
560 	if (is->is_conf.isc_enable == 0)
561 		goto out;
562 
563 	is->is_timeout++;
564 
565 	if (is->is_waiting_for_iscsid) {
566 		if (iscsid_timeout > 0 && is->is_timeout > iscsid_timeout) {
567 			ISCSI_SESSION_WARN(is, "timed out waiting for iscsid(8) "
568 			    "for %d seconds; reconnecting",
569 			    is->is_timeout);
570 			reconnect_needed = true;
571 		}
572 		goto out;
573 	}
574 
575 	if (is->is_login_phase) {
576 		if (login_timeout > 0 && is->is_timeout > login_timeout) {
577 			ISCSI_SESSION_WARN(is, "login timed out after %d seconds; "
578 			    "reconnecting", is->is_timeout);
579 			reconnect_needed = true;
580 		}
581 		goto out;
582 	}
583 
584 	if (ping_timeout <= 0) {
585 		/*
586 		 * Pings are disabled.  Don't send NOP-Out in this case.
587 		 * Reset the timeout, to avoid triggering reconnection,
588 		 * should the user decide to reenable them.
589 		 */
590 		is->is_timeout = 0;
591 		goto out;
592 	}
593 
594 	if (is->is_timeout >= ping_timeout) {
595 		ISCSI_SESSION_WARN(is, "no ping reply (NOP-In) after %d seconds; "
596 		    "reconnecting", ping_timeout);
597 		reconnect_needed = true;
598 		goto out;
599 	}
600 
601 	ISCSI_SESSION_UNLOCK(is);
602 
603 	/*
604 	 * If the ping was reset less than one second ago - which means
605 	 * that we've received some PDU during the last second - assume
606 	 * the traffic flows correctly and don't bother sending a NOP-Out.
607 	 *
608 	 * (It's 2 - one for one second, and one for incrementing is_timeout
609 	 * earlier in this routine.)
610 	 */
611 	if (is->is_timeout < 2)
612 		return;
613 
614 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
615 	if (request == NULL) {
616 		ISCSI_SESSION_WARN(is, "failed to allocate PDU");
617 		return;
618 	}
619 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
620 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
621 	    ISCSI_BHS_OPCODE_IMMEDIATE;
622 	bhsno->bhsno_flags = 0x80;
623 	bhsno->bhsno_target_transfer_tag = 0xffffffff;
624 	iscsi_pdu_queue(request);
625 	return;
626 
627 out:
628 	if (is->is_terminating) {
629 		ISCSI_SESSION_UNLOCK(is);
630 		return;
631 	}
632 
633 	ISCSI_SESSION_UNLOCK(is);
634 
635 	if (reconnect_needed)
636 		iscsi_session_reconnect(is);
637 }
638 
639 static void
640 iscsi_pdu_update_statsn(const struct icl_pdu *response)
641 {
642 	const struct iscsi_bhs_data_in *bhsdi;
643 	struct iscsi_session *is;
644 	uint32_t expcmdsn, maxcmdsn, statsn;
645 
646 	is = PDU_SESSION(response);
647 
648 	ISCSI_SESSION_LOCK_ASSERT(is);
649 
650 	/*
651 	 * We're only using fields common for all the response
652 	 * (target -> initiator) PDUs.
653 	 */
654 	bhsdi = (const struct iscsi_bhs_data_in *)response->ip_bhs;
655 	/*
656 	 * Ok, I lied.  In case of Data-In, "The fields StatSN, Status,
657 	 * and Residual Count only have meaningful content if the S bit
658 	 * is set to 1", so we also need to check the bit specific for
659 	 * Data-In PDU.
660 	 */
661 	if (bhsdi->bhsdi_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN ||
662 	    (bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0) {
663 		statsn = ntohl(bhsdi->bhsdi_statsn);
664 		if (statsn != is->is_statsn && statsn != (is->is_statsn + 1)) {
665 			/* XXX: This is normal situation for MCS */
666 			ISCSI_SESSION_WARN(is, "PDU 0x%x StatSN %u != "
667 			    "session ExpStatSN %u (or + 1); reconnecting",
668 			    bhsdi->bhsdi_opcode, statsn, is->is_statsn);
669 			iscsi_session_reconnect(is);
670 		}
671 		if (ISCSI_SNGT(statsn, is->is_statsn))
672 			is->is_statsn = statsn;
673 	}
674 
675 	expcmdsn = ntohl(bhsdi->bhsdi_expcmdsn);
676 	maxcmdsn = ntohl(bhsdi->bhsdi_maxcmdsn);
677 
678 	if (ISCSI_SNLT(maxcmdsn + 1, expcmdsn)) {
679 		ISCSI_SESSION_DEBUG(is,
680 		    "PDU MaxCmdSN %u + 1 < PDU ExpCmdSN %u; ignoring",
681 		    maxcmdsn, expcmdsn);
682 	} else {
683 		if (ISCSI_SNGT(maxcmdsn, is->is_maxcmdsn)) {
684 			is->is_maxcmdsn = maxcmdsn;
685 
686 			/*
687 			 * Command window increased; kick the maintanance thread
688 			 * to send out postponed commands.
689 			 */
690 			if (!STAILQ_EMPTY(&is->is_postponed))
691 				cv_signal(&is->is_maintenance_cv);
692 		} else if (ISCSI_SNLT(maxcmdsn, is->is_maxcmdsn)) {
693 			/* XXX: This is normal situation for MCS */
694 			ISCSI_SESSION_DEBUG(is,
695 			    "PDU MaxCmdSN %u < session MaxCmdSN %u; ignoring",
696 			    maxcmdsn, is->is_maxcmdsn);
697 		}
698 
699 		if (ISCSI_SNGT(expcmdsn, is->is_expcmdsn)) {
700 			is->is_expcmdsn = expcmdsn;
701 		} else if (ISCSI_SNLT(expcmdsn, is->is_expcmdsn)) {
702 			/* XXX: This is normal situation for MCS */
703 			ISCSI_SESSION_DEBUG(is,
704 			    "PDU ExpCmdSN %u < session ExpCmdSN %u; ignoring",
705 			    expcmdsn, is->is_expcmdsn);
706 		}
707 	}
708 
709 	/*
710 	 * Every incoming PDU - not just NOP-In - resets the ping timer.
711 	 * The purpose of the timeout is to reset the connection when it stalls;
712 	 * we don't want this to happen when NOP-In or NOP-Out ends up delayed
713 	 * in some queue.
714 	 */
715 	is->is_timeout = 0;
716 }
717 
718 static void
719 iscsi_receive_callback(struct icl_pdu *response)
720 {
721 	struct iscsi_session *is;
722 
723 	is = PDU_SESSION(response);
724 
725 	ISCSI_SESSION_LOCK(is);
726 
727 	iscsi_pdu_update_statsn(response);
728 
729 #ifdef ICL_KERNEL_PROXY
730 	if (is->is_login_phase) {
731 		if (is->is_login_pdu == NULL)
732 			is->is_login_pdu = response;
733 		else
734 			icl_pdu_free(response);
735 		ISCSI_SESSION_UNLOCK(is);
736 		cv_signal(&is->is_login_cv);
737 		return;
738 	}
739 #endif
740 
741 	/*
742 	 * The handling routine is responsible for freeing the PDU
743 	 * when it's no longer needed.
744 	 */
745 	switch (response->ip_bhs->bhs_opcode) {
746 	case ISCSI_BHS_OPCODE_NOP_IN:
747 		iscsi_pdu_handle_nop_in(response);
748 		ISCSI_SESSION_UNLOCK(is);
749 		break;
750 	case ISCSI_BHS_OPCODE_SCSI_RESPONSE:
751 		iscsi_pdu_handle_scsi_response(response);
752 		/* Session lock dropped inside. */
753 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
754 		break;
755 	case ISCSI_BHS_OPCODE_TASK_RESPONSE:
756 		iscsi_pdu_handle_task_response(response);
757 		ISCSI_SESSION_UNLOCK(is);
758 		break;
759 	case ISCSI_BHS_OPCODE_SCSI_DATA_IN:
760 		iscsi_pdu_handle_data_in(response);
761 		/* Session lock dropped inside. */
762 		ISCSI_SESSION_LOCK_ASSERT_NOT(is);
763 		break;
764 	case ISCSI_BHS_OPCODE_LOGOUT_RESPONSE:
765 		iscsi_pdu_handle_logout_response(response);
766 		ISCSI_SESSION_UNLOCK(is);
767 		break;
768 	case ISCSI_BHS_OPCODE_R2T:
769 		iscsi_pdu_handle_r2t(response);
770 		ISCSI_SESSION_UNLOCK(is);
771 		break;
772 	case ISCSI_BHS_OPCODE_ASYNC_MESSAGE:
773 		iscsi_pdu_handle_async_message(response);
774 		ISCSI_SESSION_UNLOCK(is);
775 		break;
776 	case ISCSI_BHS_OPCODE_REJECT:
777 		iscsi_pdu_handle_reject(response);
778 		ISCSI_SESSION_UNLOCK(is);
779 		break;
780 	default:
781 		ISCSI_SESSION_WARN(is, "received PDU with unsupported "
782 		    "opcode 0x%x; reconnecting",
783 		    response->ip_bhs->bhs_opcode);
784 		iscsi_session_reconnect(is);
785 		ISCSI_SESSION_UNLOCK(is);
786 		icl_pdu_free(response);
787 	}
788 }
789 
790 static void
791 iscsi_error_callback(struct icl_conn *ic)
792 {
793 	struct iscsi_session *is;
794 
795 	is = CONN_SESSION(ic);
796 
797 	ISCSI_SESSION_WARN(is, "connection error; reconnecting");
798 	iscsi_session_reconnect(is);
799 }
800 
801 static void
802 iscsi_pdu_handle_nop_in(struct icl_pdu *response)
803 {
804 	struct iscsi_session *is;
805 	struct iscsi_bhs_nop_out *bhsno;
806 	struct iscsi_bhs_nop_in *bhsni;
807 	struct icl_pdu *request;
808 	void *data = NULL;
809 	size_t datasize;
810 	int error;
811 
812 	is = PDU_SESSION(response);
813 	bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs;
814 
815 	if (bhsni->bhsni_target_transfer_tag == 0xffffffff) {
816 		/*
817 		 * Nothing to do; iscsi_pdu_update_statsn() already
818 		 * zeroed the timeout.
819 		 */
820 		icl_pdu_free(response);
821 		return;
822 	}
823 
824 	datasize = icl_pdu_data_segment_length(response);
825 	if (datasize > 0) {
826 		data = malloc(datasize, M_ISCSI, M_NOWAIT | M_ZERO);
827 		if (data == NULL) {
828 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
829 			    "reconnecting");
830 			icl_pdu_free(response);
831 			iscsi_session_reconnect(is);
832 			return;
833 		}
834 		icl_pdu_get_data(response, 0, data, datasize);
835 	}
836 
837 	request = icl_pdu_new(response->ip_conn, M_NOWAIT);
838 	if (request == NULL) {
839 		ISCSI_SESSION_WARN(is, "failed to allocate memory; "
840 		    "reconnecting");
841 		free(data, M_ISCSI);
842 		icl_pdu_free(response);
843 		iscsi_session_reconnect(is);
844 		return;
845 	}
846 	bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs;
847 	bhsno->bhsno_opcode = ISCSI_BHS_OPCODE_NOP_OUT |
848 	    ISCSI_BHS_OPCODE_IMMEDIATE;
849 	bhsno->bhsno_flags = 0x80;
850 	bhsno->bhsno_initiator_task_tag = 0xffffffff;
851 	bhsno->bhsno_target_transfer_tag = bhsni->bhsni_target_transfer_tag;
852 	if (datasize > 0) {
853 		error = icl_pdu_append_data(request, data, datasize, M_NOWAIT);
854 		if (error != 0) {
855 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
856 			    "reconnecting");
857 			free(data, M_ISCSI);
858 			icl_pdu_free(request);
859 			icl_pdu_free(response);
860 			iscsi_session_reconnect(is);
861 			return;
862 		}
863 		free(data, M_ISCSI);
864 	}
865 
866 	icl_pdu_free(response);
867 	iscsi_pdu_queue_locked(request);
868 }
869 
870 static void
871 iscsi_pdu_handle_scsi_response(struct icl_pdu *response)
872 {
873 	struct iscsi_bhs_scsi_response *bhssr;
874 	struct iscsi_outstanding *io;
875 	struct iscsi_session *is;
876 	union ccb *ccb;
877 	struct ccb_scsiio *csio;
878 	size_t data_segment_len, received;
879 	uint16_t sense_len;
880 	uint32_t resid;
881 
882 	is = PDU_SESSION(response);
883 
884 	bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs;
885 	io = iscsi_outstanding_find(is, bhssr->bhssr_initiator_task_tag);
886 	if (io == NULL || io->io_ccb == NULL) {
887 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhssr->bhssr_initiator_task_tag);
888 		icl_pdu_free(response);
889 		iscsi_session_reconnect(is);
890 		ISCSI_SESSION_UNLOCK(is);
891 		return;
892 	}
893 
894 	ccb = io->io_ccb;
895 
896 	/*
897 	 * With iSER, after getting good response we can be sure
898 	 * that all the data has been successfully transferred.
899 	 */
900 	if (is->is_conn->ic_iser) {
901 		resid = ntohl(bhssr->bhssr_residual_count);
902 		if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW) {
903 			io->io_received = ccb->csio.dxfer_len - resid;
904 		} else if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_OVERFLOW) {
905 			ISCSI_SESSION_WARN(is, "overflow: target indicates %d", resid);
906 		} else {
907 			io->io_received = ccb->csio.dxfer_len;
908 		}
909 	}
910 
911 	received = io->io_received;
912 	iscsi_outstanding_remove(is, io);
913 	ISCSI_SESSION_UNLOCK(is);
914 
915 	if (bhssr->bhssr_response != BHSSR_RESPONSE_COMMAND_COMPLETED) {
916 		ISCSI_SESSION_WARN(is, "service response 0x%x", bhssr->bhssr_response);
917  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
918  			xpt_freeze_devq(ccb->ccb_h.path, 1);
919 			ISCSI_SESSION_DEBUG(is, "freezing devq");
920 		}
921  		ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
922 	} else if (bhssr->bhssr_status == 0) {
923 		ccb->ccb_h.status = CAM_REQ_CMP;
924 	} else {
925  		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
926  			xpt_freeze_devq(ccb->ccb_h.path, 1);
927 			ISCSI_SESSION_DEBUG(is, "freezing devq");
928 		}
929  		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
930 		ccb->csio.scsi_status = bhssr->bhssr_status;
931 	}
932 
933 	csio = &ccb->csio;
934 	data_segment_len = icl_pdu_data_segment_length(response);
935 	if (data_segment_len > 0) {
936 		if (data_segment_len < sizeof(sense_len)) {
937 			ISCSI_SESSION_WARN(is, "truncated data segment (%zd bytes)",
938 			    data_segment_len);
939 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
940 				xpt_freeze_devq(ccb->ccb_h.path, 1);
941 				ISCSI_SESSION_DEBUG(is, "freezing devq");
942 			}
943 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
944 			goto out;
945 		}
946 		icl_pdu_get_data(response, 0, &sense_len, sizeof(sense_len));
947 		sense_len = ntohs(sense_len);
948 #if 0
949 		ISCSI_SESSION_DEBUG(is, "sense_len %d, data len %zd",
950 		    sense_len, data_segment_len);
951 #endif
952 		if (sizeof(sense_len) + sense_len > data_segment_len) {
953 			ISCSI_SESSION_WARN(is, "truncated data segment "
954 			    "(%zd bytes, should be %zd)",
955 			    data_segment_len, sizeof(sense_len) + sense_len);
956 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
957 				xpt_freeze_devq(ccb->ccb_h.path, 1);
958 				ISCSI_SESSION_DEBUG(is, "freezing devq");
959 			}
960 			ccb->ccb_h.status = CAM_REQ_CMP_ERR | CAM_DEV_QFRZN;
961 			goto out;
962 		} else if (sizeof(sense_len) + sense_len < data_segment_len)
963 			ISCSI_SESSION_WARN(is, "oversize data segment "
964 			    "(%zd bytes, should be %zd)",
965 			    data_segment_len, sizeof(sense_len) + sense_len);
966 		if (sense_len > csio->sense_len) {
967 			ISCSI_SESSION_DEBUG(is, "truncating sense from %d to %d",
968 			    sense_len, csio->sense_len);
969 			sense_len = csio->sense_len;
970 		}
971 		icl_pdu_get_data(response, sizeof(sense_len), &csio->sense_data, sense_len);
972 		csio->sense_resid = csio->sense_len - sense_len;
973 		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
974 	}
975 
976 out:
977 	if (bhssr->bhssr_flags & BHSSR_FLAGS_RESIDUAL_UNDERFLOW)
978 		csio->resid = ntohl(bhssr->bhssr_residual_count);
979 
980 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
981 		KASSERT(received <= csio->dxfer_len,
982 		    ("received > csio->dxfer_len"));
983 		if (received < csio->dxfer_len) {
984 			if (csio->resid != csio->dxfer_len - received) {
985 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
986 				    "target indicates %d, we calculated %zd",
987 				    csio->resid, csio->dxfer_len - received);
988 			}
989 			csio->resid = csio->dxfer_len - received;
990 		}
991 	}
992 
993 	xpt_done(ccb);
994 	icl_pdu_free(response);
995 }
996 
997 static void
998 iscsi_pdu_handle_task_response(struct icl_pdu *response)
999 {
1000 	struct iscsi_bhs_task_management_response *bhstmr;
1001 	struct iscsi_outstanding *io, *aio;
1002 	struct iscsi_session *is;
1003 
1004 	is = PDU_SESSION(response);
1005 
1006 	bhstmr = (struct iscsi_bhs_task_management_response *)response->ip_bhs;
1007 	io = iscsi_outstanding_find(is, bhstmr->bhstmr_initiator_task_tag);
1008 	if (io == NULL || io->io_ccb != NULL) {
1009 		ISCSI_SESSION_WARN(is, "bad itt 0x%x",
1010 		    bhstmr->bhstmr_initiator_task_tag);
1011 		icl_pdu_free(response);
1012 		iscsi_session_reconnect(is);
1013 		return;
1014 	}
1015 
1016 	if (bhstmr->bhstmr_response != BHSTMR_RESPONSE_FUNCTION_COMPLETE) {
1017 		ISCSI_SESSION_WARN(is, "task response 0x%x",
1018 		    bhstmr->bhstmr_response);
1019 	} else {
1020 		aio = iscsi_outstanding_find(is, io->io_referenced_task_tag);
1021 		if (aio != NULL && aio->io_ccb != NULL)
1022 			iscsi_session_terminate_task(is, aio, CAM_REQ_ABORTED);
1023 	}
1024 
1025 	iscsi_outstanding_remove(is, io);
1026 	icl_pdu_free(response);
1027 }
1028 
1029 static void
1030 iscsi_pdu_handle_data_in(struct icl_pdu *response)
1031 {
1032 	struct iscsi_bhs_data_in *bhsdi;
1033 	struct iscsi_outstanding *io;
1034 	struct iscsi_session *is;
1035 	union ccb *ccb;
1036 	struct ccb_scsiio *csio;
1037 	size_t data_segment_len, received, oreceived;
1038 
1039 	is = PDU_SESSION(response);
1040 	bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs;
1041 	io = iscsi_outstanding_find(is, bhsdi->bhsdi_initiator_task_tag);
1042 	if (io == NULL || io->io_ccb == NULL) {
1043 		ISCSI_SESSION_WARN(is, "bad itt 0x%x", bhsdi->bhsdi_initiator_task_tag);
1044 		icl_pdu_free(response);
1045 		iscsi_session_reconnect(is);
1046 		ISCSI_SESSION_UNLOCK(is);
1047 		return;
1048 	}
1049 
1050 	data_segment_len = icl_pdu_data_segment_length(response);
1051 	if (data_segment_len == 0) {
1052 		/*
1053 		 * "The sending of 0 length data segments should be avoided,
1054 		 * but initiators and targets MUST be able to properly receive
1055 		 * 0 length data segments."
1056 		 */
1057 		ISCSI_SESSION_UNLOCK(is);
1058 		icl_pdu_free(response);
1059 		return;
1060 	}
1061 
1062 	/*
1063 	 * We need to track this for security reasons - without it, malicious target
1064 	 * could respond to SCSI READ without sending Data-In PDUs, which would result
1065 	 * in read operation on the initiator side returning random kernel data.
1066 	 */
1067 	if (ntohl(bhsdi->bhsdi_buffer_offset) != io->io_received) {
1068 		ISCSI_SESSION_WARN(is, "data out of order; expected offset %zd, got %zd",
1069 		    io->io_received, (size_t)ntohl(bhsdi->bhsdi_buffer_offset));
1070 		icl_pdu_free(response);
1071 		iscsi_session_reconnect(is);
1072 		ISCSI_SESSION_UNLOCK(is);
1073 		return;
1074 	}
1075 
1076 	ccb = io->io_ccb;
1077 	csio = &ccb->csio;
1078 
1079 	if (io->io_received + data_segment_len > csio->dxfer_len) {
1080 		ISCSI_SESSION_WARN(is, "oversize data segment (%zd bytes "
1081 		    "at offset %zd, buffer is %d)",
1082 		    data_segment_len, io->io_received, csio->dxfer_len);
1083 		icl_pdu_free(response);
1084 		iscsi_session_reconnect(is);
1085 		ISCSI_SESSION_UNLOCK(is);
1086 		return;
1087 	}
1088 
1089 	oreceived = io->io_received;
1090 	io->io_received += data_segment_len;
1091 	received = io->io_received;
1092 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) != 0)
1093 		iscsi_outstanding_remove(is, io);
1094 	ISCSI_SESSION_UNLOCK(is);
1095 
1096 	icl_pdu_get_data(response, 0, csio->data_ptr + oreceived, data_segment_len);
1097 
1098 	/*
1099 	 * XXX: Check DataSN.
1100 	 * XXX: Check F.
1101 	 */
1102 	if ((bhsdi->bhsdi_flags & BHSDI_FLAGS_S) == 0) {
1103 		/*
1104 		 * Nothing more to do.
1105 		 */
1106 		icl_pdu_free(response);
1107 		return;
1108 	}
1109 
1110 	//ISCSI_SESSION_DEBUG(is, "got S flag; status 0x%x", bhsdi->bhsdi_status);
1111 	if (bhsdi->bhsdi_status == 0) {
1112 		ccb->ccb_h.status = CAM_REQ_CMP;
1113 	} else {
1114 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
1115 			xpt_freeze_devq(ccb->ccb_h.path, 1);
1116 			ISCSI_SESSION_DEBUG(is, "freezing devq");
1117 		}
1118 		ccb->ccb_h.status = CAM_SCSI_STATUS_ERROR | CAM_DEV_QFRZN;
1119 		csio->scsi_status = bhsdi->bhsdi_status;
1120 	}
1121 
1122 	if ((csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
1123 		KASSERT(received <= csio->dxfer_len,
1124 		    ("received > csio->dxfer_len"));
1125 		if (received < csio->dxfer_len) {
1126 			csio->resid = ntohl(bhsdi->bhsdi_residual_count);
1127 			if (csio->resid != csio->dxfer_len - received) {
1128 				ISCSI_SESSION_WARN(is, "underflow mismatch: "
1129 				    "target indicates %d, we calculated %zd",
1130 				    csio->resid, csio->dxfer_len - received);
1131 			}
1132 			csio->resid = csio->dxfer_len - received;
1133 		}
1134 	}
1135 
1136 	xpt_done(ccb);
1137 	icl_pdu_free(response);
1138 }
1139 
1140 static void
1141 iscsi_pdu_handle_logout_response(struct icl_pdu *response)
1142 {
1143 
1144 	ISCSI_SESSION_DEBUG(PDU_SESSION(response), "logout response");
1145 	icl_pdu_free(response);
1146 }
1147 
1148 static void
1149 iscsi_pdu_handle_r2t(struct icl_pdu *response)
1150 {
1151 	struct icl_pdu *request;
1152 	struct iscsi_session *is;
1153 	struct iscsi_bhs_r2t *bhsr2t;
1154 	struct iscsi_bhs_data_out *bhsdo;
1155 	struct iscsi_outstanding *io;
1156 	struct ccb_scsiio *csio;
1157 	size_t off, len, max_send_data_segment_length, total_len;
1158 	int error;
1159 	uint32_t datasn = 0;
1160 
1161 	is = PDU_SESSION(response);
1162 
1163 	bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs;
1164 	io = iscsi_outstanding_find(is, bhsr2t->bhsr2t_initiator_task_tag);
1165 	if (io == NULL || io->io_ccb == NULL) {
1166 		ISCSI_SESSION_WARN(is, "bad itt 0x%x; reconnecting",
1167 		    bhsr2t->bhsr2t_initiator_task_tag);
1168 		icl_pdu_free(response);
1169 		iscsi_session_reconnect(is);
1170 		return;
1171 	}
1172 
1173 	csio = &io->io_ccb->csio;
1174 
1175 	if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_OUT) {
1176 		ISCSI_SESSION_WARN(is, "received R2T for read command; reconnecting");
1177 		icl_pdu_free(response);
1178 		iscsi_session_reconnect(is);
1179 		return;
1180 	}
1181 
1182 	/*
1183 	 * XXX: Verify R2TSN.
1184 	 */
1185 
1186 	off = ntohl(bhsr2t->bhsr2t_buffer_offset);
1187 	if (off > csio->dxfer_len) {
1188 		ISCSI_SESSION_WARN(is, "target requested invalid offset "
1189 		    "%zd, buffer is is %d; reconnecting", off, csio->dxfer_len);
1190 		icl_pdu_free(response);
1191 		iscsi_session_reconnect(is);
1192 		return;
1193 	}
1194 
1195 	total_len = ntohl(bhsr2t->bhsr2t_desired_data_transfer_length);
1196 	if (total_len == 0 || total_len > csio->dxfer_len) {
1197 		ISCSI_SESSION_WARN(is, "target requested invalid length "
1198 		    "%zd, buffer is %d; reconnecting", total_len, csio->dxfer_len);
1199 		icl_pdu_free(response);
1200 		iscsi_session_reconnect(is);
1201 		return;
1202 	}
1203 
1204 	//ISCSI_SESSION_DEBUG(is, "r2t; off %zd, len %zd", off, total_len);
1205 
1206 	if (is->is_conn->ic_hw_isomax != 0)
1207 		max_send_data_segment_length = is->is_conn->ic_hw_isomax;
1208 	else
1209 		max_send_data_segment_length =
1210 		    is->is_conn->ic_max_send_data_segment_length;
1211 	for (;;) {
1212 		len = total_len;
1213 
1214 		if (len > max_send_data_segment_length)
1215 			len = max_send_data_segment_length;
1216 
1217 		if (off + len > csio->dxfer_len) {
1218 			ISCSI_SESSION_WARN(is, "target requested invalid "
1219 			    "length/offset %zd, buffer is %d; reconnecting",
1220 			    off + len, csio->dxfer_len);
1221 			icl_pdu_free(response);
1222 			iscsi_session_reconnect(is);
1223 			return;
1224 		}
1225 
1226 		request = icl_pdu_new(response->ip_conn, M_NOWAIT);
1227 		if (request == NULL) {
1228 			icl_pdu_free(response);
1229 			iscsi_session_reconnect(is);
1230 			return;
1231 		}
1232 
1233 		bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs;
1234 		bhsdo->bhsdo_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_OUT;
1235 		bhsdo->bhsdo_lun = bhsr2t->bhsr2t_lun;
1236 		bhsdo->bhsdo_initiator_task_tag =
1237 		    bhsr2t->bhsr2t_initiator_task_tag;
1238 		bhsdo->bhsdo_target_transfer_tag =
1239 		    bhsr2t->bhsr2t_target_transfer_tag;
1240 		bhsdo->bhsdo_datasn = htonl(datasn);
1241 		bhsdo->bhsdo_buffer_offset = htonl(off);
1242 		error = icl_pdu_append_data(request, csio->data_ptr + off, len,
1243 		    M_NOWAIT);
1244 		if (error != 0) {
1245 			ISCSI_SESSION_WARN(is, "failed to allocate memory; "
1246 			    "reconnecting");
1247 			icl_pdu_free(request);
1248 			icl_pdu_free(response);
1249 			iscsi_session_reconnect(is);
1250 			return;
1251 		}
1252 
1253 		datasn += howmany(len,
1254 		    is->is_conn->ic_max_send_data_segment_length);
1255 		off += len;
1256 		total_len -= len;
1257 
1258 		if (total_len == 0) {
1259 			bhsdo->bhsdo_flags |= BHSDO_FLAGS_F;
1260 			//ISCSI_SESSION_DEBUG(is, "setting F, off %zd", off);
1261 		} else {
1262 			//ISCSI_SESSION_DEBUG(is, "not finished, off %zd", off);
1263 		}
1264 
1265 		iscsi_pdu_queue_locked(request);
1266 
1267 		if (total_len == 0)
1268 			break;
1269 	}
1270 
1271 	icl_pdu_free(response);
1272 }
1273 
1274 static void
1275 iscsi_pdu_handle_async_message(struct icl_pdu *response)
1276 {
1277 	struct iscsi_bhs_asynchronous_message *bhsam;
1278 	struct iscsi_session *is;
1279 
1280 	is = PDU_SESSION(response);
1281 	bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs;
1282 	switch (bhsam->bhsam_async_event) {
1283 	case BHSAM_EVENT_TARGET_REQUESTS_LOGOUT:
1284 		ISCSI_SESSION_WARN(is, "target requests logout; removing session");
1285 		iscsi_session_logout(is);
1286 		iscsi_session_terminate(is);
1287 		break;
1288 	case BHSAM_EVENT_TARGET_TERMINATES_CONNECTION:
1289 		ISCSI_SESSION_WARN(is, "target indicates it will drop the connection");
1290 		break;
1291 	case BHSAM_EVENT_TARGET_TERMINATES_SESSION:
1292 		ISCSI_SESSION_WARN(is, "target indicates it will drop the session");
1293 		break;
1294 	default:
1295 		/*
1296 		 * XXX: Technically, we're obligated to also handle
1297 		 * 	parameter renegotiation.
1298 		 */
1299 		ISCSI_SESSION_WARN(is, "ignoring AsyncEvent %d", bhsam->bhsam_async_event);
1300 		break;
1301 	}
1302 
1303 	icl_pdu_free(response);
1304 }
1305 
1306 static void
1307 iscsi_pdu_handle_reject(struct icl_pdu *response)
1308 {
1309 	struct iscsi_bhs_reject *bhsr;
1310 	struct iscsi_session *is;
1311 
1312 	is = PDU_SESSION(response);
1313 	bhsr = (struct iscsi_bhs_reject *)response->ip_bhs;
1314 	ISCSI_SESSION_WARN(is, "received Reject PDU, reason 0x%x; protocol error?",
1315 	    bhsr->bhsr_reason);
1316 
1317 	icl_pdu_free(response);
1318 }
1319 
1320 static int
1321 iscsi_ioctl_daemon_wait(struct iscsi_softc *sc,
1322     struct iscsi_daemon_request *request)
1323 {
1324 	struct iscsi_session *is;
1325 	struct icl_drv_limits idl;
1326 	int error;
1327 
1328 	sx_slock(&sc->sc_lock);
1329 	for (;;) {
1330 		TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1331 			ISCSI_SESSION_LOCK(is);
1332 			if (is->is_conf.isc_enable == 0 &&
1333 			    is->is_conf.isc_discovery == 0) {
1334 				ISCSI_SESSION_UNLOCK(is);
1335 				continue;
1336 			}
1337 			if (is->is_waiting_for_iscsid)
1338 				break;
1339 			ISCSI_SESSION_UNLOCK(is);
1340 		}
1341 
1342 		if (is == NULL) {
1343 			if (sc->sc_unloading) {
1344 				sx_sunlock(&sc->sc_lock);
1345 				return (ENXIO);
1346 			}
1347 
1348 			/*
1349 			 * No session requires attention from iscsid(8); wait.
1350 			 */
1351 			error = cv_wait_sig(&sc->sc_cv, &sc->sc_lock);
1352 			if (error != 0) {
1353 				sx_sunlock(&sc->sc_lock);
1354 				return (error);
1355 			}
1356 			continue;
1357 		}
1358 
1359 		is->is_waiting_for_iscsid = false;
1360 		is->is_login_phase = true;
1361 		is->is_reason[0] = '\0';
1362 		ISCSI_SESSION_UNLOCK(is);
1363 
1364 		request->idr_session_id = is->is_id;
1365 		memcpy(&request->idr_isid, &is->is_isid,
1366 		    sizeof(request->idr_isid));
1367 		request->idr_tsih = 0;	/* New or reinstated session. */
1368 		memcpy(&request->idr_conf, &is->is_conf,
1369 		    sizeof(request->idr_conf));
1370 
1371 		error = icl_limits(is->is_conf.isc_offload,
1372 		    is->is_conf.isc_iser, &idl);
1373 		if (error != 0) {
1374 			ISCSI_SESSION_WARN(is, "icl_limits for offload \"%s\" "
1375 			    "failed with error %d", is->is_conf.isc_offload,
1376 			    error);
1377 			sx_sunlock(&sc->sc_lock);
1378 			return (error);
1379 		}
1380 		request->idr_limits.isl_max_recv_data_segment_length =
1381 		    idl.idl_max_recv_data_segment_length;
1382 		request->idr_limits.isl_max_send_data_segment_length =
1383 		    idl.idl_max_send_data_segment_length;
1384 		request->idr_limits.isl_max_burst_length =
1385 		    idl.idl_max_burst_length;
1386 		request->idr_limits.isl_first_burst_length =
1387 		    idl.idl_first_burst_length;
1388 
1389 		sx_sunlock(&sc->sc_lock);
1390 		return (0);
1391 	}
1392 }
1393 
1394 static int
1395 iscsi_ioctl_daemon_handoff(struct iscsi_softc *sc,
1396     struct iscsi_daemon_handoff *handoff)
1397 {
1398 	struct iscsi_session *is;
1399 	struct icl_conn *ic;
1400 	int error;
1401 
1402 	sx_slock(&sc->sc_lock);
1403 
1404 	/*
1405 	 * Find the session to hand off socket to.
1406 	 */
1407 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1408 		if (is->is_id == handoff->idh_session_id)
1409 			break;
1410 	}
1411 	if (is == NULL) {
1412 		sx_sunlock(&sc->sc_lock);
1413 		return (ESRCH);
1414 	}
1415 	ISCSI_SESSION_LOCK(is);
1416 	ic = is->is_conn;
1417 	if (is->is_conf.isc_discovery || is->is_terminating) {
1418 		ISCSI_SESSION_UNLOCK(is);
1419 		sx_sunlock(&sc->sc_lock);
1420 		return (EINVAL);
1421 	}
1422 	if (is->is_connected) {
1423 		/*
1424 		 * This might have happened because another iscsid(8)
1425 		 * instance handed off the connection in the meantime.
1426 		 * Just return.
1427 		 */
1428 		ISCSI_SESSION_WARN(is, "handoff on already connected "
1429 		    "session");
1430 		ISCSI_SESSION_UNLOCK(is);
1431 		sx_sunlock(&sc->sc_lock);
1432 		return (EBUSY);
1433 	}
1434 
1435 	strlcpy(is->is_target_alias, handoff->idh_target_alias,
1436 	    sizeof(is->is_target_alias));
1437 	is->is_tsih = handoff->idh_tsih;
1438 	is->is_statsn = handoff->idh_statsn;
1439 	is->is_protocol_level = handoff->idh_protocol_level;
1440 	is->is_initial_r2t = handoff->idh_initial_r2t;
1441 	is->is_immediate_data = handoff->idh_immediate_data;
1442 
1443 	ic->ic_max_recv_data_segment_length =
1444 	    handoff->idh_max_recv_data_segment_length;
1445 	ic->ic_max_send_data_segment_length =
1446 	    handoff->idh_max_send_data_segment_length;
1447 	is->is_max_burst_length = handoff->idh_max_burst_length;
1448 	is->is_first_burst_length = handoff->idh_first_burst_length;
1449 
1450 	if (handoff->idh_header_digest == ISCSI_DIGEST_CRC32C)
1451 		ic->ic_header_crc32c = true;
1452 	else
1453 		ic->ic_header_crc32c = false;
1454 	if (handoff->idh_data_digest == ISCSI_DIGEST_CRC32C)
1455 		ic->ic_data_crc32c = true;
1456 	else
1457 		ic->ic_data_crc32c = false;
1458 	ic->ic_maxtags = maxtags;
1459 
1460 	is->is_cmdsn = 0;
1461 	is->is_expcmdsn = 0;
1462 	is->is_maxcmdsn = 0;
1463 	is->is_waiting_for_iscsid = false;
1464 	is->is_login_phase = false;
1465 	is->is_timeout = 0;
1466 	is->is_connected = true;
1467 	is->is_reason[0] = '\0';
1468 
1469 	ISCSI_SESSION_UNLOCK(is);
1470 
1471 	/*
1472 	 * If we're going through the proxy, the idh_socket will be 0,
1473 	 * and the ICL module can simply ignore this call.  It can also
1474 	 * use it to determine it's no longer in the Login phase.
1475 	 */
1476 	error = icl_conn_handoff(ic, handoff->idh_socket);
1477 	if (error != 0) {
1478 		sx_sunlock(&sc->sc_lock);
1479 		iscsi_session_terminate(is);
1480 		return (error);
1481 	}
1482 
1483 	sx_sunlock(&sc->sc_lock);
1484 
1485 	if (is->is_sim != NULL) {
1486 		/*
1487 		 * When reconnecting, there already is SIM allocated for the session.
1488 		 */
1489 		KASSERT(is->is_simq_frozen, ("reconnect without frozen simq"));
1490 		ISCSI_SESSION_LOCK(is);
1491 		ISCSI_SESSION_DEBUG(is, "releasing");
1492 		is->is_simq_frozen = false;
1493 		xpt_release_simq(is->is_sim, 1);
1494 		ISCSI_SESSION_UNLOCK(is);
1495 
1496 	} else {
1497 		ISCSI_SESSION_LOCK(is);
1498 		is->is_devq = cam_simq_alloc(ic->ic_maxtags);
1499 		if (is->is_devq == NULL) {
1500 			ISCSI_SESSION_WARN(is, "failed to allocate simq");
1501 			iscsi_session_terminate(is);
1502 			return (ENOMEM);
1503 		}
1504 
1505 		is->is_sim = cam_sim_alloc(iscsi_action, NULL, "iscsi",
1506 		    is, is->is_id /* unit */, &is->is_lock,
1507 		    1, ic->ic_maxtags, is->is_devq);
1508 		if (is->is_sim == NULL) {
1509 			ISCSI_SESSION_UNLOCK(is);
1510 			ISCSI_SESSION_WARN(is, "failed to allocate SIM");
1511 			cam_simq_free(is->is_devq);
1512 			iscsi_session_terminate(is);
1513 			return (ENOMEM);
1514 		}
1515 
1516 		if (xpt_bus_register(is->is_sim, NULL, 0) != 0) {
1517 			ISCSI_SESSION_UNLOCK(is);
1518 			ISCSI_SESSION_WARN(is, "failed to register bus");
1519 			iscsi_session_terminate(is);
1520 			return (ENOMEM);
1521 		}
1522 
1523 		error = xpt_create_path(&is->is_path, /*periph*/NULL,
1524 		    cam_sim_path(is->is_sim), CAM_TARGET_WILDCARD,
1525 		    CAM_LUN_WILDCARD);
1526 		if (error != CAM_REQ_CMP) {
1527 			ISCSI_SESSION_UNLOCK(is);
1528 			ISCSI_SESSION_WARN(is, "failed to create path");
1529 			iscsi_session_terminate(is);
1530 			return (ENOMEM);
1531 		}
1532 		ISCSI_SESSION_UNLOCK(is);
1533 	}
1534 
1535 	return (0);
1536 }
1537 
1538 static int
1539 iscsi_ioctl_daemon_fail(struct iscsi_softc *sc,
1540     struct iscsi_daemon_fail *fail)
1541 {
1542 	struct iscsi_session *is;
1543 
1544 	sx_slock(&sc->sc_lock);
1545 
1546 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1547 		if (is->is_id == fail->idf_session_id)
1548 			break;
1549 	}
1550 	if (is == NULL) {
1551 		sx_sunlock(&sc->sc_lock);
1552 		return (ESRCH);
1553 	}
1554 	ISCSI_SESSION_LOCK(is);
1555 	ISCSI_SESSION_DEBUG(is, "iscsid(8) failed: %s",
1556 	    fail->idf_reason);
1557 	strlcpy(is->is_reason, fail->idf_reason, sizeof(is->is_reason));
1558 	//is->is_waiting_for_iscsid = false;
1559 	//is->is_login_phase = true;
1560 	//iscsi_session_reconnect(is);
1561 	ISCSI_SESSION_UNLOCK(is);
1562 	sx_sunlock(&sc->sc_lock);
1563 
1564 	return (0);
1565 }
1566 
1567 #ifdef ICL_KERNEL_PROXY
1568 static int
1569 iscsi_ioctl_daemon_connect(struct iscsi_softc *sc,
1570     struct iscsi_daemon_connect *idc)
1571 {
1572 	struct iscsi_session *is;
1573 	struct sockaddr *from_sa, *to_sa;
1574 	int error;
1575 
1576 	sx_slock(&sc->sc_lock);
1577 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1578 		if (is->is_id == idc->idc_session_id)
1579 			break;
1580 	}
1581 	if (is == NULL) {
1582 		sx_sunlock(&sc->sc_lock);
1583 		return (ESRCH);
1584 	}
1585 	sx_sunlock(&sc->sc_lock);
1586 
1587 	if (idc->idc_from_addrlen > 0) {
1588 		error = getsockaddr(&from_sa, (void *)idc->idc_from_addr, idc->idc_from_addrlen);
1589 		if (error != 0) {
1590 			ISCSI_SESSION_WARN(is,
1591 			    "getsockaddr failed with error %d", error);
1592 			return (error);
1593 		}
1594 	} else {
1595 		from_sa = NULL;
1596 	}
1597 	error = getsockaddr(&to_sa, (void *)idc->idc_to_addr, idc->idc_to_addrlen);
1598 	if (error != 0) {
1599 		ISCSI_SESSION_WARN(is, "getsockaddr failed with error %d",
1600 		    error);
1601 		free(from_sa, M_SONAME);
1602 		return (error);
1603 	}
1604 
1605 	ISCSI_SESSION_LOCK(is);
1606 	is->is_statsn = 0;
1607 	is->is_cmdsn = 0;
1608 	is->is_expcmdsn = 0;
1609 	is->is_maxcmdsn = 0;
1610 	is->is_waiting_for_iscsid = false;
1611 	is->is_login_phase = true;
1612 	is->is_timeout = 0;
1613 	ISCSI_SESSION_UNLOCK(is);
1614 
1615 	error = icl_conn_connect(is->is_conn, idc->idc_domain,
1616 	    idc->idc_socktype, idc->idc_protocol, from_sa, to_sa);
1617 	free(from_sa, M_SONAME);
1618 	free(to_sa, M_SONAME);
1619 
1620 	/*
1621 	 * Digests are always disabled during login phase.
1622 	 */
1623 	is->is_conn->ic_header_crc32c = false;
1624 	is->is_conn->ic_data_crc32c = false;
1625 
1626 	return (error);
1627 }
1628 
1629 static int
1630 iscsi_ioctl_daemon_send(struct iscsi_softc *sc,
1631     struct iscsi_daemon_send *ids)
1632 {
1633 	struct iscsi_session *is;
1634 	struct icl_pdu *ip;
1635 	size_t datalen;
1636 	void *data;
1637 	int error;
1638 
1639 	sx_slock(&sc->sc_lock);
1640 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1641 		if (is->is_id == ids->ids_session_id)
1642 			break;
1643 	}
1644 	if (is == NULL) {
1645 		sx_sunlock(&sc->sc_lock);
1646 		return (ESRCH);
1647 	}
1648 	sx_sunlock(&sc->sc_lock);
1649 
1650 	if (is->is_login_phase == false)
1651 		return (EBUSY);
1652 
1653 	if (is->is_terminating || is->is_reconnecting)
1654 		return (EIO);
1655 
1656 	datalen = ids->ids_data_segment_len;
1657 	if (datalen > is->is_conn->ic_max_send_data_segment_length)
1658 		return (EINVAL);
1659 	if (datalen > 0) {
1660 		data = malloc(datalen, M_ISCSI, M_WAITOK);
1661 		error = copyin(ids->ids_data_segment, data, datalen);
1662 		if (error != 0) {
1663 			free(data, M_ISCSI);
1664 			return (error);
1665 		}
1666 	}
1667 
1668 	ip = icl_pdu_new(is->is_conn, M_WAITOK);
1669 	memcpy(ip->ip_bhs, ids->ids_bhs, sizeof(*ip->ip_bhs));
1670 	if (datalen > 0) {
1671 		error = icl_pdu_append_data(ip, data, datalen, M_WAITOK);
1672 		KASSERT(error == 0, ("icl_pdu_append_data(..., M_WAITOK) failed"));
1673 		free(data, M_ISCSI);
1674 	}
1675 	iscsi_pdu_queue(ip);
1676 
1677 	return (0);
1678 }
1679 
1680 static int
1681 iscsi_ioctl_daemon_receive(struct iscsi_softc *sc,
1682     struct iscsi_daemon_receive *idr)
1683 {
1684 	struct iscsi_session *is;
1685 	struct icl_pdu *ip;
1686 	void *data;
1687 	int error;
1688 
1689 	sx_slock(&sc->sc_lock);
1690 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1691 		if (is->is_id == idr->idr_session_id)
1692 			break;
1693 	}
1694 	if (is == NULL) {
1695 		sx_sunlock(&sc->sc_lock);
1696 		return (ESRCH);
1697 	}
1698 	sx_sunlock(&sc->sc_lock);
1699 
1700 	if (is->is_login_phase == false)
1701 		return (EBUSY);
1702 
1703 	ISCSI_SESSION_LOCK(is);
1704 	while (is->is_login_pdu == NULL &&
1705 	    is->is_terminating == false &&
1706 	    is->is_reconnecting == false) {
1707 		error = cv_wait_sig(&is->is_login_cv, &is->is_lock);
1708 		if (error != 0) {
1709 			ISCSI_SESSION_UNLOCK(is);
1710 			return (error);
1711 		}
1712 	}
1713 	if (is->is_terminating || is->is_reconnecting) {
1714 		ISCSI_SESSION_UNLOCK(is);
1715 		return (EIO);
1716 	}
1717 	ip = is->is_login_pdu;
1718 	is->is_login_pdu = NULL;
1719 	ISCSI_SESSION_UNLOCK(is);
1720 
1721 	if (ip->ip_data_len > idr->idr_data_segment_len) {
1722 		icl_pdu_free(ip);
1723 		return (EMSGSIZE);
1724 	}
1725 
1726 	copyout(ip->ip_bhs, idr->idr_bhs, sizeof(*ip->ip_bhs));
1727 	if (ip->ip_data_len > 0) {
1728 		data = malloc(ip->ip_data_len, M_ISCSI, M_WAITOK);
1729 		icl_pdu_get_data(ip, 0, data, ip->ip_data_len);
1730 		copyout(data, idr->idr_data_segment, ip->ip_data_len);
1731 		free(data, M_ISCSI);
1732 	}
1733 
1734 	icl_pdu_free(ip);
1735 
1736 	return (0);
1737 }
1738 #endif /* ICL_KERNEL_PROXY */
1739 
1740 static void
1741 iscsi_sanitize_session_conf(struct iscsi_session_conf *isc)
1742 {
1743 	/*
1744 	 * Just make sure all the fields are null-terminated.
1745 	 *
1746 	 * XXX: This is not particularly secure.  We should
1747 	 * 	create our own conf and then copy in relevant
1748 	 * 	fields.
1749 	 */
1750 	isc->isc_initiator[ISCSI_NAME_LEN - 1] = '\0';
1751 	isc->isc_initiator_addr[ISCSI_ADDR_LEN - 1] = '\0';
1752 	isc->isc_initiator_alias[ISCSI_ALIAS_LEN - 1] = '\0';
1753 	isc->isc_target[ISCSI_NAME_LEN - 1] = '\0';
1754 	isc->isc_target_addr[ISCSI_ADDR_LEN - 1] = '\0';
1755 	isc->isc_user[ISCSI_NAME_LEN - 1] = '\0';
1756 	isc->isc_secret[ISCSI_SECRET_LEN - 1] = '\0';
1757 	isc->isc_mutual_user[ISCSI_NAME_LEN - 1] = '\0';
1758 	isc->isc_mutual_secret[ISCSI_SECRET_LEN - 1] = '\0';
1759 }
1760 
1761 static bool
1762 iscsi_valid_session_conf(const struct iscsi_session_conf *isc)
1763 {
1764 
1765 	if (isc->isc_initiator[0] == '\0') {
1766 		ISCSI_DEBUG("empty isc_initiator");
1767 		return (false);
1768 	}
1769 
1770 	if (isc->isc_target_addr[0] == '\0') {
1771 		ISCSI_DEBUG("empty isc_target_addr");
1772 		return (false);
1773 	}
1774 
1775 	if (isc->isc_discovery != 0 && isc->isc_target[0] != 0) {
1776 		ISCSI_DEBUG("non-empty isc_target for discovery session");
1777 		return (false);
1778 	}
1779 
1780 	if (isc->isc_discovery == 0 && isc->isc_target[0] == 0) {
1781 		ISCSI_DEBUG("empty isc_target for non-discovery session");
1782 		return (false);
1783 	}
1784 
1785 	return (true);
1786 }
1787 
1788 static int
1789 iscsi_ioctl_session_add(struct iscsi_softc *sc, struct iscsi_session_add *isa)
1790 {
1791 	struct iscsi_session *is;
1792 	const struct iscsi_session *is2;
1793 	int error;
1794 
1795 	iscsi_sanitize_session_conf(&isa->isa_conf);
1796 	if (iscsi_valid_session_conf(&isa->isa_conf) == false)
1797 		return (EINVAL);
1798 
1799 	is = malloc(sizeof(*is), M_ISCSI, M_ZERO | M_WAITOK);
1800 	memcpy(&is->is_conf, &isa->isa_conf, sizeof(is->is_conf));
1801 
1802 	sx_xlock(&sc->sc_lock);
1803 
1804 	/*
1805 	 * Prevent duplicates.
1806 	 */
1807 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
1808 		if (!!is->is_conf.isc_discovery !=
1809 		    !!is2->is_conf.isc_discovery)
1810 			continue;
1811 
1812 		if (strcmp(is->is_conf.isc_target_addr,
1813 		    is2->is_conf.isc_target_addr) != 0)
1814 			continue;
1815 
1816 		if (is->is_conf.isc_discovery == 0 &&
1817 		    strcmp(is->is_conf.isc_target,
1818 		    is2->is_conf.isc_target) != 0)
1819 			continue;
1820 
1821 		sx_xunlock(&sc->sc_lock);
1822 		free(is, M_ISCSI);
1823 		return (EBUSY);
1824 	}
1825 
1826 	is->is_conn = icl_new_conn(is->is_conf.isc_offload,
1827 	    is->is_conf.isc_iser, "iscsi", &is->is_lock);
1828 	if (is->is_conn == NULL) {
1829 		sx_xunlock(&sc->sc_lock);
1830 		free(is, M_ISCSI);
1831 		return (EINVAL);
1832 	}
1833 	is->is_conn->ic_receive = iscsi_receive_callback;
1834 	is->is_conn->ic_error = iscsi_error_callback;
1835 	is->is_conn->ic_prv0 = is;
1836 	TAILQ_INIT(&is->is_outstanding);
1837 	STAILQ_INIT(&is->is_postponed);
1838 	mtx_init(&is->is_lock, "iscsi_lock", NULL, MTX_DEF);
1839 	cv_init(&is->is_maintenance_cv, "iscsi_mt");
1840 #ifdef ICL_KERNEL_PROXY
1841 	cv_init(&is->is_login_cv, "iscsi_login");
1842 #endif
1843 
1844 	/*
1845 	 * Set some default values, from RFC 3720, section 12.
1846 	 *
1847 	 * These values are updated by the handoff IOCTL, but are
1848 	 * needed prior to the handoff to support sending the ISER
1849 	 * login PDU.
1850 	 */
1851 	is->is_conn->ic_max_recv_data_segment_length = 8192;
1852 	is->is_conn->ic_max_send_data_segment_length = 8192;
1853 	is->is_max_burst_length = 262144;
1854 	is->is_first_burst_length = 65536;
1855 
1856 	is->is_softc = sc;
1857 	sc->sc_last_session_id++;
1858 	is->is_id = sc->sc_last_session_id;
1859 	is->is_isid[0] = 0x80; /* RFC 3720, 10.12.5: 10b, "Random" ISID. */
1860 	arc4rand(&is->is_isid[1], 5, 0);
1861 	is->is_tsih = 0;
1862 	callout_init(&is->is_callout, 1);
1863 
1864 	error = kthread_add(iscsi_maintenance_thread, is, NULL, NULL, 0, 0, "iscsimt");
1865 	if (error != 0) {
1866 		ISCSI_SESSION_WARN(is, "kthread_add(9) failed with error %d", error);
1867 		sx_xunlock(&sc->sc_lock);
1868 		return (error);
1869 	}
1870 
1871 	callout_reset(&is->is_callout, 1 * hz, iscsi_callout, is);
1872 	TAILQ_INSERT_TAIL(&sc->sc_sessions, is, is_next);
1873 
1874 	ISCSI_SESSION_LOCK(is);
1875 	/*
1876 	 * Don't notify iscsid(8) if the session is disabled and it's not
1877 	 * a discovery session,
1878 	 */
1879 	if (is->is_conf.isc_enable == 0 && is->is_conf.isc_discovery == 0) {
1880 		ISCSI_SESSION_UNLOCK(is);
1881 		sx_xunlock(&sc->sc_lock);
1882 		return (0);
1883 	}
1884 
1885 	is->is_waiting_for_iscsid = true;
1886 	strlcpy(is->is_reason, "Waiting for iscsid(8)", sizeof(is->is_reason));
1887 	ISCSI_SESSION_UNLOCK(is);
1888 	cv_signal(&sc->sc_cv);
1889 	sx_xunlock(&sc->sc_lock);
1890 	return (0);
1891 }
1892 
1893 static bool
1894 iscsi_session_conf_matches(unsigned int id1, const struct iscsi_session_conf *c1,
1895     unsigned int id2, const struct iscsi_session_conf *c2)
1896 {
1897 
1898 	if (id2 != 0 && id2 != id1)
1899 		return (false);
1900 	if (c2->isc_target[0] != '\0' &&
1901 	    strcmp(c1->isc_target, c2->isc_target) != 0)
1902 		return (false);
1903 	if (c2->isc_target_addr[0] != '\0' &&
1904 	    strcmp(c1->isc_target_addr, c2->isc_target_addr) != 0)
1905 		return (false);
1906 	return (true);
1907 }
1908 
1909 static int
1910 iscsi_ioctl_session_remove(struct iscsi_softc *sc,
1911     struct iscsi_session_remove *isr)
1912 {
1913 	struct iscsi_session *is, *tmp;
1914 	bool found = false;
1915 
1916 	iscsi_sanitize_session_conf(&isr->isr_conf);
1917 
1918 	sx_xlock(&sc->sc_lock);
1919 	TAILQ_FOREACH_SAFE(is, &sc->sc_sessions, is_next, tmp) {
1920 		ISCSI_SESSION_LOCK(is);
1921 		if (iscsi_session_conf_matches(is->is_id, &is->is_conf,
1922 		    isr->isr_session_id, &isr->isr_conf)) {
1923 			found = true;
1924 			iscsi_session_logout(is);
1925 			iscsi_session_terminate(is);
1926 		}
1927 		ISCSI_SESSION_UNLOCK(is);
1928 	}
1929 	sx_xunlock(&sc->sc_lock);
1930 
1931 	if (!found)
1932 		return (ESRCH);
1933 
1934 	return (0);
1935 }
1936 
1937 static int
1938 iscsi_ioctl_session_list(struct iscsi_softc *sc, struct iscsi_session_list *isl)
1939 {
1940 	int error;
1941 	unsigned int i = 0;
1942 	struct iscsi_session *is;
1943 	struct iscsi_session_state iss;
1944 
1945 	sx_slock(&sc->sc_lock);
1946 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
1947 		if (i >= isl->isl_nentries) {
1948 			sx_sunlock(&sc->sc_lock);
1949 			return (EMSGSIZE);
1950 		}
1951 		memset(&iss, 0, sizeof(iss));
1952 		memcpy(&iss.iss_conf, &is->is_conf, sizeof(iss.iss_conf));
1953 		iss.iss_id = is->is_id;
1954 		strlcpy(iss.iss_target_alias, is->is_target_alias, sizeof(iss.iss_target_alias));
1955 		strlcpy(iss.iss_reason, is->is_reason, sizeof(iss.iss_reason));
1956 		strlcpy(iss.iss_offload, is->is_conn->ic_offload, sizeof(iss.iss_offload));
1957 
1958 		if (is->is_conn->ic_header_crc32c)
1959 			iss.iss_header_digest = ISCSI_DIGEST_CRC32C;
1960 		else
1961 			iss.iss_header_digest = ISCSI_DIGEST_NONE;
1962 
1963 		if (is->is_conn->ic_data_crc32c)
1964 			iss.iss_data_digest = ISCSI_DIGEST_CRC32C;
1965 		else
1966 			iss.iss_data_digest = ISCSI_DIGEST_NONE;
1967 
1968 		iss.iss_max_send_data_segment_length =
1969 		    is->is_conn->ic_max_send_data_segment_length;
1970 		iss.iss_max_recv_data_segment_length =
1971 		    is->is_conn->ic_max_recv_data_segment_length;
1972 		iss.iss_max_burst_length = is->is_max_burst_length;
1973 		iss.iss_first_burst_length = is->is_first_burst_length;
1974 		iss.iss_immediate_data = is->is_immediate_data;
1975 		iss.iss_connected = is->is_connected;
1976 
1977 		error = copyout(&iss, isl->isl_pstates + i, sizeof(iss));
1978 		if (error != 0) {
1979 			sx_sunlock(&sc->sc_lock);
1980 			return (error);
1981 		}
1982 		i++;
1983 	}
1984 	sx_sunlock(&sc->sc_lock);
1985 
1986 	isl->isl_nentries = i;
1987 
1988 	return (0);
1989 }
1990 
1991 static int
1992 iscsi_ioctl_session_modify(struct iscsi_softc *sc,
1993     struct iscsi_session_modify *ism)
1994 {
1995 	struct iscsi_session *is;
1996 	const struct iscsi_session *is2;
1997 
1998 	iscsi_sanitize_session_conf(&ism->ism_conf);
1999 	if (iscsi_valid_session_conf(&ism->ism_conf) == false)
2000 		return (EINVAL);
2001 
2002 	sx_xlock(&sc->sc_lock);
2003 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2004 		ISCSI_SESSION_LOCK(is);
2005 		if (is->is_id == ism->ism_session_id) {
2006 			/* Note that the session remains locked. */
2007 			break;
2008 		}
2009 		ISCSI_SESSION_UNLOCK(is);
2010 	}
2011 	if (is == NULL) {
2012 		sx_xunlock(&sc->sc_lock);
2013 		return (ESRCH);
2014 	}
2015 
2016 	/*
2017 	 * Prevent duplicates.
2018 	 */
2019 	TAILQ_FOREACH(is2, &sc->sc_sessions, is_next) {
2020 		if (is == is2)
2021 			continue;
2022 
2023 		if (!!ism->ism_conf.isc_discovery !=
2024 		    !!is2->is_conf.isc_discovery)
2025 			continue;
2026 
2027 		if (strcmp(ism->ism_conf.isc_target_addr,
2028 		    is2->is_conf.isc_target_addr) != 0)
2029 			continue;
2030 
2031 		if (ism->ism_conf.isc_discovery == 0 &&
2032 		    strcmp(ism->ism_conf.isc_target,
2033 		    is2->is_conf.isc_target) != 0)
2034 			continue;
2035 
2036 		ISCSI_SESSION_UNLOCK(is);
2037 		sx_xunlock(&sc->sc_lock);
2038 		return (EBUSY);
2039 	}
2040 
2041 	sx_xunlock(&sc->sc_lock);
2042 
2043 	memcpy(&is->is_conf, &ism->ism_conf, sizeof(is->is_conf));
2044 	ISCSI_SESSION_UNLOCK(is);
2045 
2046 	iscsi_session_reconnect(is);
2047 
2048 	return (0);
2049 }
2050 
2051 static int
2052 iscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int mode,
2053     struct thread *td)
2054 {
2055 	struct iscsi_softc *sc;
2056 
2057 	sc = dev->si_drv1;
2058 
2059 	switch (cmd) {
2060 	case ISCSIDWAIT:
2061 		return (iscsi_ioctl_daemon_wait(sc,
2062 		    (struct iscsi_daemon_request *)arg));
2063 	case ISCSIDHANDOFF:
2064 		return (iscsi_ioctl_daemon_handoff(sc,
2065 		    (struct iscsi_daemon_handoff *)arg));
2066 	case ISCSIDFAIL:
2067 		return (iscsi_ioctl_daemon_fail(sc,
2068 		    (struct iscsi_daemon_fail *)arg));
2069 #ifdef ICL_KERNEL_PROXY
2070 	case ISCSIDCONNECT:
2071 		return (iscsi_ioctl_daemon_connect(sc,
2072 		    (struct iscsi_daemon_connect *)arg));
2073 	case ISCSIDSEND:
2074 		return (iscsi_ioctl_daemon_send(sc,
2075 		    (struct iscsi_daemon_send *)arg));
2076 	case ISCSIDRECEIVE:
2077 		return (iscsi_ioctl_daemon_receive(sc,
2078 		    (struct iscsi_daemon_receive *)arg));
2079 #endif /* ICL_KERNEL_PROXY */
2080 	case ISCSISADD:
2081 		return (iscsi_ioctl_session_add(sc,
2082 		    (struct iscsi_session_add *)arg));
2083 	case ISCSISREMOVE:
2084 		return (iscsi_ioctl_session_remove(sc,
2085 		    (struct iscsi_session_remove *)arg));
2086 	case ISCSISLIST:
2087 		return (iscsi_ioctl_session_list(sc,
2088 		    (struct iscsi_session_list *)arg));
2089 	case ISCSISMODIFY:
2090 		return (iscsi_ioctl_session_modify(sc,
2091 		    (struct iscsi_session_modify *)arg));
2092 	default:
2093 		return (EINVAL);
2094 	}
2095 }
2096 
2097 static struct iscsi_outstanding *
2098 iscsi_outstanding_find(struct iscsi_session *is, uint32_t initiator_task_tag)
2099 {
2100 	struct iscsi_outstanding *io;
2101 
2102 	ISCSI_SESSION_LOCK_ASSERT(is);
2103 
2104 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2105 		if (io->io_initiator_task_tag == initiator_task_tag)
2106 			return (io);
2107 	}
2108 	return (NULL);
2109 }
2110 
2111 static struct iscsi_outstanding *
2112 iscsi_outstanding_find_ccb(struct iscsi_session *is, union ccb *ccb)
2113 {
2114 	struct iscsi_outstanding *io;
2115 
2116 	ISCSI_SESSION_LOCK_ASSERT(is);
2117 
2118 	TAILQ_FOREACH(io, &is->is_outstanding, io_next) {
2119 		if (io->io_ccb == ccb)
2120 			return (io);
2121 	}
2122 	return (NULL);
2123 }
2124 
2125 static struct iscsi_outstanding *
2126 iscsi_outstanding_add(struct iscsi_session *is, struct icl_pdu *request,
2127     union ccb *ccb, uint32_t *initiator_task_tagp)
2128 {
2129 	struct iscsi_outstanding *io;
2130 	int error;
2131 
2132 	ISCSI_SESSION_LOCK_ASSERT(is);
2133 
2134 	io = uma_zalloc(iscsi_outstanding_zone, M_NOWAIT | M_ZERO);
2135 	if (io == NULL) {
2136 		ISCSI_SESSION_WARN(is, "failed to allocate %zd bytes",
2137 		    sizeof(*io));
2138 		return (NULL);
2139 	}
2140 
2141 	error = icl_conn_task_setup(is->is_conn, request, &ccb->csio,
2142 	    initiator_task_tagp, &io->io_icl_prv);
2143 	if (error != 0) {
2144 		ISCSI_SESSION_WARN(is,
2145 		    "icl_conn_task_setup() failed with error %d", error);
2146 		uma_zfree(iscsi_outstanding_zone, io);
2147 		return (NULL);
2148 	}
2149 
2150 	KASSERT(iscsi_outstanding_find(is, *initiator_task_tagp) == NULL,
2151 	    ("initiator_task_tag 0x%x already added", *initiator_task_tagp));
2152 
2153 	io->io_initiator_task_tag = *initiator_task_tagp;
2154 	io->io_ccb = ccb;
2155 	TAILQ_INSERT_TAIL(&is->is_outstanding, io, io_next);
2156 	return (io);
2157 }
2158 
2159 static void
2160 iscsi_outstanding_remove(struct iscsi_session *is, struct iscsi_outstanding *io)
2161 {
2162 
2163 	ISCSI_SESSION_LOCK_ASSERT(is);
2164 
2165 	icl_conn_task_done(is->is_conn, io->io_icl_prv);
2166 	TAILQ_REMOVE(&is->is_outstanding, io, io_next);
2167 	uma_zfree(iscsi_outstanding_zone, io);
2168 }
2169 
2170 static void
2171 iscsi_action_abort(struct iscsi_session *is, union ccb *ccb)
2172 {
2173 	struct icl_pdu *request;
2174 	struct iscsi_bhs_task_management_request *bhstmr;
2175 	struct ccb_abort *cab = &ccb->cab;
2176 	struct iscsi_outstanding *io, *aio;
2177 	uint32_t initiator_task_tag;
2178 
2179 	ISCSI_SESSION_LOCK_ASSERT(is);
2180 
2181 #if 0
2182 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2183 #else
2184 	if (is->is_login_phase) {
2185 		ccb->ccb_h.status = CAM_REQ_ABORTED;
2186 		xpt_done(ccb);
2187 		return;
2188 	}
2189 #endif
2190 
2191 	aio = iscsi_outstanding_find_ccb(is, cab->abort_ccb);
2192 	if (aio == NULL) {
2193 		ccb->ccb_h.status = CAM_REQ_CMP;
2194 		xpt_done(ccb);
2195 		return;
2196 	}
2197 
2198 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2199 	if (request == NULL) {
2200 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2201 		xpt_done(ccb);
2202 		return;
2203 	}
2204 
2205 	initiator_task_tag = is->is_initiator_task_tag++;
2206 	if (initiator_task_tag == 0xffffffff)
2207 		initiator_task_tag = is->is_initiator_task_tag++;
2208 
2209 	io = iscsi_outstanding_add(is, request, NULL, &initiator_task_tag);
2210 	if (io == NULL) {
2211 		icl_pdu_free(request);
2212 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL;
2213 		xpt_done(ccb);
2214 		return;
2215 	}
2216 	io->io_referenced_task_tag = aio->io_initiator_task_tag;
2217 
2218 	bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs;
2219 	bhstmr->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_REQUEST;
2220 	bhstmr->bhstmr_function = 0x80 | BHSTMR_FUNCTION_ABORT_TASK;
2221 	bhstmr->bhstmr_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2222 	bhstmr->bhstmr_initiator_task_tag = initiator_task_tag;
2223 	bhstmr->bhstmr_referenced_task_tag = aio->io_initiator_task_tag;
2224 
2225 	iscsi_pdu_queue_locked(request);
2226 }
2227 
2228 static void
2229 iscsi_action_scsiio(struct iscsi_session *is, union ccb *ccb)
2230 {
2231 	struct icl_pdu *request;
2232 	struct iscsi_bhs_scsi_command *bhssc;
2233 	struct ccb_scsiio *csio;
2234 	struct iscsi_outstanding *io;
2235 	size_t len;
2236 	uint32_t initiator_task_tag;
2237 	int error;
2238 
2239 	ISCSI_SESSION_LOCK_ASSERT(is);
2240 
2241 #if 0
2242 	KASSERT(is->is_login_phase == false, ("%s called during Login Phase", __func__));
2243 #else
2244 	if (is->is_login_phase) {
2245 		ISCSI_SESSION_DEBUG(is, "called during login phase");
2246 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2247 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2248 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2249 		}
2250 		ccb->ccb_h.status = CAM_REQ_ABORTED | CAM_DEV_QFRZN;
2251 		xpt_done(ccb);
2252 		return;
2253 	}
2254 #endif
2255 
2256 	request = icl_pdu_new(is->is_conn, M_NOWAIT);
2257 	if (request == NULL) {
2258 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2259 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2260 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2261 		}
2262 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2263 		xpt_done(ccb);
2264 		return;
2265 	}
2266 
2267 	initiator_task_tag = is->is_initiator_task_tag++;
2268 	if (initiator_task_tag == 0xffffffff)
2269 		initiator_task_tag = is->is_initiator_task_tag++;
2270 
2271 	io = iscsi_outstanding_add(is, request, ccb, &initiator_task_tag);
2272 	if (io == NULL) {
2273 		icl_pdu_free(request);
2274 		if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2275 			xpt_freeze_devq(ccb->ccb_h.path, 1);
2276 			ISCSI_SESSION_DEBUG(is, "freezing devq");
2277 		}
2278 		ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2279 		xpt_done(ccb);
2280 		return;
2281 	}
2282 
2283 	csio = &ccb->csio;
2284 	bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs;
2285 	bhssc->bhssc_opcode = ISCSI_BHS_OPCODE_SCSI_COMMAND;
2286 	bhssc->bhssc_flags |= BHSSC_FLAGS_F;
2287 	switch (csio->ccb_h.flags & CAM_DIR_MASK) {
2288 	case CAM_DIR_IN:
2289 		bhssc->bhssc_flags |= BHSSC_FLAGS_R;
2290 		break;
2291 	case CAM_DIR_OUT:
2292 		bhssc->bhssc_flags |= BHSSC_FLAGS_W;
2293 		break;
2294 	}
2295 
2296 	if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
2297 		switch (csio->tag_action) {
2298 		case MSG_HEAD_OF_Q_TAG:
2299 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_HOQ;
2300 			break;
2301 		case MSG_ORDERED_Q_TAG:
2302 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ORDERED;
2303 			break;
2304 		case MSG_ACA_TASK:
2305 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_ACA;
2306 			break;
2307 		case MSG_SIMPLE_Q_TAG:
2308 		default:
2309 			bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_SIMPLE;
2310 			break;
2311 		}
2312 	} else
2313 		bhssc->bhssc_flags |= BHSSC_FLAGS_ATTR_UNTAGGED;
2314 
2315 	if (is->is_protocol_level >= 2) {
2316 		bhssc->bhssc_pri = (csio->priority << BHSSC_PRI_SHIFT) &
2317 		    BHSSC_PRI_MASK;
2318 	}
2319 
2320 	bhssc->bhssc_lun = htobe64(CAM_EXTLUN_BYTE_SWIZZLE(ccb->ccb_h.target_lun));
2321 	bhssc->bhssc_initiator_task_tag = initiator_task_tag;
2322 	bhssc->bhssc_expected_data_transfer_length = htonl(csio->dxfer_len);
2323 	KASSERT(csio->cdb_len <= sizeof(bhssc->bhssc_cdb),
2324 	    ("unsupported CDB size %zd", (size_t)csio->cdb_len));
2325 
2326 	if (csio->ccb_h.flags & CAM_CDB_POINTER)
2327 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_ptr, csio->cdb_len);
2328 	else
2329 		memcpy(&bhssc->bhssc_cdb, csio->cdb_io.cdb_bytes, csio->cdb_len);
2330 
2331 	if (is->is_immediate_data &&
2332 	    (csio->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
2333 		len = csio->dxfer_len;
2334 		//ISCSI_SESSION_DEBUG(is, "adding %zd of immediate data", len);
2335 		if (len > is->is_first_burst_length) {
2336 			ISCSI_SESSION_DEBUG(is, "len %zd -> %d", len, is->is_first_burst_length);
2337 			len = is->is_first_burst_length;
2338 		}
2339 		if (len > is->is_conn->ic_max_send_data_segment_length) {
2340 			ISCSI_SESSION_DEBUG(is, "len %zd -> %d", len,
2341 			    is->is_conn->ic_max_send_data_segment_length);
2342 			len = is->is_conn->ic_max_send_data_segment_length;
2343 		}
2344 
2345 		error = icl_pdu_append_data(request, csio->data_ptr, len, M_NOWAIT);
2346 		if (error != 0) {
2347 			iscsi_outstanding_remove(is, io);
2348 			icl_pdu_free(request);
2349 			if ((ccb->ccb_h.status & CAM_DEV_QFRZN) == 0) {
2350 				xpt_freeze_devq(ccb->ccb_h.path, 1);
2351 				ISCSI_SESSION_DEBUG(is, "freezing devq");
2352 			}
2353 			ccb->ccb_h.status = CAM_RESRC_UNAVAIL | CAM_DEV_QFRZN;
2354 			xpt_done(ccb);
2355 			return;
2356 		}
2357 	}
2358 	iscsi_pdu_queue_locked(request);
2359 }
2360 
2361 static void
2362 iscsi_action(struct cam_sim *sim, union ccb *ccb)
2363 {
2364 	struct iscsi_session *is;
2365 
2366 	is = cam_sim_softc(sim);
2367 
2368 	ISCSI_SESSION_LOCK_ASSERT(is);
2369 
2370 	if (is->is_terminating ||
2371 	    (is->is_connected == false && fail_on_disconnection)) {
2372 		ccb->ccb_h.status = CAM_DEV_NOT_THERE;
2373 		xpt_done(ccb);
2374 		return;
2375 	}
2376 
2377 	/*
2378 	 * Make sure CAM doesn't sneak in a CCB just after freezing the queue.
2379 	 */
2380 	if (is->is_simq_frozen == true) {
2381 		ccb->ccb_h.status &= ~(CAM_SIM_QUEUED | CAM_STATUS_MASK);
2382 		ccb->ccb_h.status |= CAM_REQUEUE_REQ;
2383 		/* Don't freeze the devq - the SIM queue is already frozen. */
2384 		xpt_done(ccb);
2385 		return;
2386 	}
2387 
2388 	switch (ccb->ccb_h.func_code) {
2389 	case XPT_PATH_INQ:
2390 	{
2391 		struct ccb_pathinq *cpi = &ccb->cpi;
2392 
2393 		cpi->version_num = 1;
2394 		cpi->hba_inquiry = PI_TAG_ABLE;
2395 		cpi->target_sprt = 0;
2396 		cpi->hba_misc = PIM_EXTLUNS;
2397 		/*
2398 		 * XXX: It shouldn't ever be NULL; this could be turned
2399 		 *      into a KASSERT eventually.
2400 		 */
2401 		if (is->is_conn == NULL)
2402 			ISCSI_WARN("NULL conn");
2403 		else if (is->is_conn->ic_unmapped)
2404 			cpi->hba_misc |= PIM_UNMAPPED;
2405 		cpi->hba_eng_cnt = 0;
2406 		cpi->max_target = 0;
2407 		/*
2408 		 * Note that the variable below is only relevant for targets
2409 		 * that don't claim compliance with anything above SPC2, which
2410 		 * means they don't support REPORT_LUNS.
2411 		 */
2412 		cpi->max_lun = 255;
2413 		cpi->initiator_id = ~0;
2414 		strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
2415 		strlcpy(cpi->hba_vid, "iSCSI", HBA_IDLEN);
2416 		strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
2417 		cpi->unit_number = cam_sim_unit(sim);
2418 		cpi->bus_id = cam_sim_bus(sim);
2419 		cpi->base_transfer_speed = 150000; /* XXX */
2420 		cpi->transport = XPORT_ISCSI;
2421 		cpi->transport_version = 0;
2422 		cpi->protocol = PROTO_SCSI;
2423 		cpi->protocol_version = SCSI_REV_SPC3;
2424 		cpi->maxio = maxphys;
2425 		cpi->ccb_h.status = CAM_REQ_CMP;
2426 		break;
2427 	}
2428 	case XPT_GET_TRAN_SETTINGS:
2429 	{
2430 		struct ccb_trans_settings	*cts;
2431 		struct ccb_trans_settings_scsi	*scsi;
2432 
2433 		cts = &ccb->cts;
2434 		scsi = &cts->proto_specific.scsi;
2435 
2436 		cts->protocol = PROTO_SCSI;
2437 		cts->protocol_version = SCSI_REV_SPC3;
2438 		cts->transport = XPORT_ISCSI;
2439 		cts->transport_version = 0;
2440 		scsi->valid = CTS_SCSI_VALID_TQ;
2441 		scsi->flags = CTS_SCSI_FLAGS_TAG_ENB;
2442 		cts->ccb_h.status = CAM_REQ_CMP;
2443 		break;
2444 	}
2445 	case XPT_CALC_GEOMETRY:
2446 		cam_calc_geometry(&ccb->ccg, /*extended*/1);
2447 		ccb->ccb_h.status = CAM_REQ_CMP;
2448 		break;
2449 #if 0
2450 	/*
2451 	 * XXX: What's the point?
2452 	 */
2453 	case XPT_RESET_BUS:
2454 	case XPT_TERM_IO:
2455 		ISCSI_SESSION_DEBUG(is, "faking success for reset, abort, or term_io");
2456 		ccb->ccb_h.status = CAM_REQ_CMP;
2457 		break;
2458 #endif
2459 	case XPT_ABORT:
2460 		iscsi_action_abort(is, ccb);
2461 		return;
2462 	case XPT_SCSI_IO:
2463 		iscsi_action_scsiio(is, ccb);
2464 		return;
2465 	default:
2466 #if 0
2467 		ISCSI_SESSION_DEBUG(is, "got unsupported code 0x%x", ccb->ccb_h.func_code);
2468 #endif
2469 		ccb->ccb_h.status = CAM_FUNC_NOTAVAIL;
2470 		break;
2471 	}
2472 	xpt_done(ccb);
2473 }
2474 
2475 static void
2476 iscsi_terminate_sessions(struct iscsi_softc *sc)
2477 {
2478 	struct iscsi_session *is;
2479 
2480 	sx_slock(&sc->sc_lock);
2481 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next)
2482 		iscsi_session_terminate(is);
2483 	while(!TAILQ_EMPTY(&sc->sc_sessions)) {
2484 		ISCSI_DEBUG("waiting for sessions to terminate");
2485 		cv_wait(&sc->sc_cv, &sc->sc_lock);
2486 	}
2487 	ISCSI_DEBUG("all sessions terminated");
2488 	sx_sunlock(&sc->sc_lock);
2489 }
2490 
2491 static void
2492 iscsi_shutdown_pre(struct iscsi_softc *sc)
2493 {
2494 	struct iscsi_session *is;
2495 
2496 	if (!fail_on_shutdown)
2497 		return;
2498 
2499 	/*
2500 	 * If we have any sessions waiting for reconnection, request
2501 	 * maintenance thread to fail them immediately instead of waiting
2502 	 * for reconnect timeout.
2503 	 *
2504 	 * This prevents LUNs with mounted filesystems that are supported
2505 	 * by disconnected iSCSI sessions from hanging, however it will
2506 	 * fail all queued BIOs.
2507 	 */
2508 	ISCSI_DEBUG("forcing failing all disconnected sessions due to shutdown");
2509 
2510 	fail_on_disconnection = 1;
2511 
2512 	sx_slock(&sc->sc_lock);
2513 	TAILQ_FOREACH(is, &sc->sc_sessions, is_next) {
2514 		ISCSI_SESSION_LOCK(is);
2515 		if (!is->is_connected) {
2516 			ISCSI_SESSION_DEBUG(is, "force failing disconnected session early");
2517 			iscsi_session_reconnect(is);
2518 		}
2519 		ISCSI_SESSION_UNLOCK(is);
2520 	}
2521 	sx_sunlock(&sc->sc_lock);
2522 }
2523 
2524 static void
2525 iscsi_shutdown_post(struct iscsi_softc *sc)
2526 {
2527 
2528 	if (!KERNEL_PANICKED()) {
2529 		ISCSI_DEBUG("removing all sessions due to shutdown");
2530 		iscsi_terminate_sessions(sc);
2531 	}
2532 }
2533 
2534 static int
2535 iscsi_load(void)
2536 {
2537 	int error;
2538 
2539 	sc = malloc(sizeof(*sc), M_ISCSI, M_ZERO | M_WAITOK);
2540 	sx_init(&sc->sc_lock, "iscsi");
2541 	TAILQ_INIT(&sc->sc_sessions);
2542 	cv_init(&sc->sc_cv, "iscsi_cv");
2543 
2544 	iscsi_outstanding_zone = uma_zcreate("iscsi_outstanding",
2545 	    sizeof(struct iscsi_outstanding), NULL, NULL, NULL, NULL,
2546 	    UMA_ALIGN_PTR, 0);
2547 
2548 	error = make_dev_p(MAKEDEV_CHECKNAME, &sc->sc_cdev, &iscsi_cdevsw,
2549 	    NULL, UID_ROOT, GID_WHEEL, 0600, "iscsi");
2550 	if (error != 0) {
2551 		ISCSI_WARN("failed to create device node, error %d", error);
2552 		return (error);
2553 	}
2554 	sc->sc_cdev->si_drv1 = sc;
2555 
2556 	sc->sc_shutdown_pre_eh = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2557 	    iscsi_shutdown_pre, sc, SHUTDOWN_PRI_FIRST);
2558 	/*
2559 	 * shutdown_post_sync needs to run after filesystem shutdown and before
2560 	 * CAM shutdown - otherwise when rebooting with an iSCSI session that is
2561 	 * disconnected but has outstanding requests, dashutdown() will hang on
2562 	 * cam_periph_runccb().
2563 	 */
2564 	sc->sc_shutdown_post_eh = EVENTHANDLER_REGISTER(shutdown_post_sync,
2565 	    iscsi_shutdown_post, sc, SHUTDOWN_PRI_DEFAULT - 1);
2566 
2567 	return (0);
2568 }
2569 
2570 static int
2571 iscsi_unload(void)
2572 {
2573 
2574 	/* Awaken any threads asleep in iscsi_ioctl(). */
2575 	sx_xlock(&sc->sc_lock);
2576 	sc->sc_unloading = true;
2577 	cv_signal(&sc->sc_cv);
2578 	sx_xunlock(&sc->sc_lock);
2579 
2580 	if (sc->sc_cdev != NULL) {
2581 		ISCSI_DEBUG("removing device node");
2582 		destroy_dev(sc->sc_cdev);
2583 		ISCSI_DEBUG("device node removed");
2584 	}
2585 
2586 	if (sc->sc_shutdown_pre_eh != NULL)
2587 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, sc->sc_shutdown_pre_eh);
2588 	if (sc->sc_shutdown_post_eh != NULL)
2589 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, sc->sc_shutdown_post_eh);
2590 
2591 	iscsi_terminate_sessions(sc);
2592 
2593 	uma_zdestroy(iscsi_outstanding_zone);
2594 	sx_destroy(&sc->sc_lock);
2595 	cv_destroy(&sc->sc_cv);
2596 	free(sc, M_ISCSI);
2597 	return (0);
2598 }
2599 
2600 static int
2601 iscsi_quiesce(void)
2602 {
2603 	sx_slock(&sc->sc_lock);
2604 	if (!TAILQ_EMPTY(&sc->sc_sessions)) {
2605 		sx_sunlock(&sc->sc_lock);
2606 		return (EBUSY);
2607 	}
2608 	sx_sunlock(&sc->sc_lock);
2609 	return (0);
2610 }
2611 
2612 static int
2613 iscsi_modevent(module_t mod, int what, void *arg)
2614 {
2615 	int error;
2616 
2617 	switch (what) {
2618 	case MOD_LOAD:
2619 		error = iscsi_load();
2620 		break;
2621 	case MOD_UNLOAD:
2622 		error = iscsi_unload();
2623 		break;
2624 	case MOD_QUIESCE:
2625 		error = iscsi_quiesce();
2626 		break;
2627 	default:
2628 		error = EINVAL;
2629 		break;
2630 	}
2631 	return (error);
2632 }
2633 
2634 moduledata_t iscsi_data = {
2635 	"iscsi",
2636 	iscsi_modevent,
2637 	0
2638 };
2639 
2640 DECLARE_MODULE(iscsi, iscsi_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
2641 MODULE_DEPEND(iscsi, cam, 1, 1, 1);
2642 MODULE_DEPEND(iscsi, icl, 1, 1, 1);
2643