1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 /*
25  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
26  */
27 
28 #include <sys/cpuvar.h>
29 #include <sys/types.h>
30 #include <sys/conf.h>
31 #include <sys/stat.h>
32 #include <sys/file.h>
33 #include <sys/ddi.h>
34 #include <sys/sunddi.h>
35 #include <sys/modctl.h>
36 #include <sys/sysmacros.h>
37 #include <sys/socket.h>
38 #include <sys/strsubr.h>
39 #include <sys/nvpair.h>
40 
41 #include <sys/stmf.h>
42 #include <sys/stmf_ioctl.h>
43 #include <sys/portif.h>
44 #include <sys/idm/idm.h>
45 #include <sys/idm/idm_conn_sm.h>
46 
47 #include "iscsit_isns.h"
48 #include "iscsit.h"
49 
50 #define	ISCSIT_VERSION		BUILD_DATE "-1.18dev"
51 #define	ISCSIT_NAME_VERSION	"COMSTAR ISCSIT v" ISCSIT_VERSION
52 
53 /*
54  * DDI entry points.
55  */
56 static int iscsit_drv_attach(dev_info_t *, ddi_attach_cmd_t);
57 static int iscsit_drv_detach(dev_info_t *, ddi_detach_cmd_t);
58 static int iscsit_drv_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
59 static int iscsit_drv_open(dev_t *, int, int, cred_t *);
60 static int iscsit_drv_close(dev_t, int, int, cred_t *);
61 static boolean_t iscsit_drv_busy(void);
62 static int iscsit_drv_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
63 
64 extern struct mod_ops mod_miscops;
65 
66 
67 static struct cb_ops iscsit_cb_ops = {
68 	iscsit_drv_open,	/* cb_open */
69 	iscsit_drv_close,	/* cb_close */
70 	nodev,			/* cb_strategy */
71 	nodev,			/* cb_print */
72 	nodev,			/* cb_dump */
73 	nodev,			/* cb_read */
74 	nodev,			/* cb_write */
75 	iscsit_drv_ioctl,	/* cb_ioctl */
76 	nodev,			/* cb_devmap */
77 	nodev,			/* cb_mmap */
78 	nodev,			/* cb_segmap */
79 	nochpoll,		/* cb_chpoll */
80 	ddi_prop_op,		/* cb_prop_op */
81 	NULL,			/* cb_streamtab */
82 	D_MP,			/* cb_flag */
83 	CB_REV,			/* cb_rev */
84 	nodev,			/* cb_aread */
85 	nodev,			/* cb_awrite */
86 };
87 
88 static struct dev_ops iscsit_dev_ops = {
89 	DEVO_REV,		/* devo_rev */
90 	0,			/* devo_refcnt */
91 	iscsit_drv_getinfo,	/* devo_getinfo */
92 	nulldev,		/* devo_identify */
93 	nulldev,		/* devo_probe */
94 	iscsit_drv_attach,	/* devo_attach */
95 	iscsit_drv_detach,	/* devo_detach */
96 	nodev,			/* devo_reset */
97 	&iscsit_cb_ops,		/* devo_cb_ops */
98 	NULL,			/* devo_bus_ops */
99 	NULL,			/* devo_power */
100 	ddi_quiesce_not_needed,	/* quiesce */
101 };
102 
103 static struct modldrv modldrv = {
104 	&mod_driverops,
105 	"iSCSI Target",
106 	&iscsit_dev_ops,
107 };
108 
109 static struct modlinkage modlinkage = {
110 	MODREV_1,
111 	&modldrv,
112 	NULL,
113 };
114 
115 
116 iscsit_global_t iscsit_global;
117 
118 kmem_cache_t	*iscsit_status_pdu_cache;
119 
120 boolean_t	iscsit_sm_logging = B_FALSE;
121 
122 kmutex_t	login_sm_session_mutex;
123 
124 static idm_status_t iscsit_init(dev_info_t *dip);
125 static idm_status_t iscsit_enable_svc(iscsit_hostinfo_t *hostinfo);
126 static void iscsit_disable_svc(void);
127 
128 static int
129 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu);
130 
131 static void
132 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu);
133 
134 static idm_pdu_t *
135 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn);
136 
137 static void
138 iscsit_process_pdu_in_queue(iscsit_sess_t *ist);
139 
140 static void
141 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist);
142 
143 static void
144 iscsit_rxpdu_queue_monitor(void *arg);
145 
146 static void
147 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu);
148 
149 static void
150 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu);
151 
152 static void
153 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
154 
155 static void
156 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
157 
158 static void
159 iscsit_pdu_op_login_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
160 
161 void
162 iscsit_pdu_op_text_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
163 
164 static void
165 iscsit_pdu_op_logout_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
166 
167 int iscsit_cmd_window();
168 
169 static  int
170 iscsit_sna_lt(uint32_t sn1, uint32_t sn2);
171 
172 void
173 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
174 
175 static void
176 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu);
177 
178 static void
179 iscsit_deferred(void *rx_pdu_void);
180 
181 static idm_status_t
182 iscsit_conn_accept(idm_conn_t *ic);
183 
184 static idm_status_t
185 iscsit_ffp_enabled(idm_conn_t *ic);
186 
187 static idm_status_t
188 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class);
189 
190 static idm_status_t
191 iscsit_conn_lost(idm_conn_t *ic);
192 
193 static idm_status_t
194 iscsit_conn_destroy(idm_conn_t *ic);
195 
196 static stmf_data_buf_t *
197 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
198     uint32_t flags);
199 
200 static void
201 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf);
202 
203 static void
204 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status);
205 
206 static void
207 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status);
208 
209 static void
210 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status);
211 
212 static stmf_status_t
213 iscsit_idm_to_stmf(idm_status_t idmrc);
214 
215 static iscsit_task_t *
216 iscsit_task_alloc(iscsit_conn_t *ict);
217 
218 static void
219 iscsit_task_free(iscsit_task_t *itask);
220 
221 static iscsit_task_t *
222 iscsit_tm_task_alloc(iscsit_conn_t *ict);
223 
224 static void
225 iscsit_tm_task_free(iscsit_task_t *itask);
226 
227 static idm_status_t
228 iscsit_task_start(iscsit_task_t *itask);
229 
230 static void
231 iscsit_task_done(iscsit_task_t *itask);
232 
233 static int
234 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags);
235 
236 static void
237 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags);
238 
239 static it_cfg_status_t
240 iscsit_config_merge(it_config_t *cfg);
241 
242 static idm_status_t
243 iscsit_login_fail(idm_conn_t *ic);
244 
245 static boolean_t iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn);
246 static void iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
247     uint8_t response, uint8_t cmd_status);
248 static void iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu,
249     uint8_t tm_status);
250 
251 /*
252  * MC/S: Out-of-order commands are staged on a session-wide wait
253  * queue until a system-tunable threshold is reached. A separate
254  * thread is used to scan the staging queue on all the session,
255  * If a delayed PDU does not arrive within a timeout, the target
256  * will advance to the staged PDU that is next in sequence, skipping
257  * over the missing PDU(s) to go past a hole in the sequence.
258  */
259 volatile int rxpdu_queue_threshold = ISCSIT_RXPDU_QUEUE_THRESHOLD;
260 
261 static kmutex_t		iscsit_rxpdu_queue_monitor_mutex;
262 kthread_t		*iscsit_rxpdu_queue_monitor_thr_id;
263 static kt_did_t		iscsit_rxpdu_queue_monitor_thr_did;
264 static boolean_t	iscsit_rxpdu_queue_monitor_thr_running;
265 static kcondvar_t	iscsit_rxpdu_queue_monitor_cv;
266 
267 int
268 _init(void)
269 {
270 	int rc;
271 
272 	rw_init(&iscsit_global.global_rwlock, NULL, RW_DRIVER, NULL);
273 	mutex_init(&iscsit_global.global_state_mutex, NULL,
274 	    MUTEX_DRIVER, NULL);
275 	iscsit_global.global_svc_state = ISE_DETACHED;
276 
277 	mutex_init(&iscsit_rxpdu_queue_monitor_mutex, NULL,
278 	    MUTEX_DRIVER, NULL);
279 	mutex_init(&login_sm_session_mutex, NULL, MUTEX_DRIVER, NULL);
280 	iscsit_rxpdu_queue_monitor_thr_id = NULL;
281 	iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
282 	cv_init(&iscsit_rxpdu_queue_monitor_cv, NULL, CV_DEFAULT, NULL);
283 
284 	if ((rc = mod_install(&modlinkage)) != 0) {
285 		mutex_destroy(&iscsit_global.global_state_mutex);
286 		rw_destroy(&iscsit_global.global_rwlock);
287 		return (rc);
288 	}
289 
290 	return (rc);
291 }
292 
293 int
294 _info(struct modinfo *modinfop)
295 {
296 	return (mod_info(&modlinkage, modinfop));
297 }
298 
299 int
300 _fini(void)
301 {
302 	int rc;
303 
304 	rc = mod_remove(&modlinkage);
305 
306 	if (rc == 0) {
307 		mutex_destroy(&iscsit_rxpdu_queue_monitor_mutex);
308 		mutex_destroy(&login_sm_session_mutex);
309 		cv_destroy(&iscsit_rxpdu_queue_monitor_cv);
310 		mutex_destroy(&iscsit_global.global_state_mutex);
311 		rw_destroy(&iscsit_global.global_rwlock);
312 	}
313 
314 	return (rc);
315 }
316 
317 /*
318  * DDI entry points.
319  */
320 
321 /* ARGSUSED */
322 static int
323 iscsit_drv_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
324     void **result)
325 {
326 	ulong_t instance = getminor((dev_t)arg);
327 
328 	switch (cmd) {
329 	case DDI_INFO_DEVT2DEVINFO:
330 		*result = iscsit_global.global_dip;
331 		return (DDI_SUCCESS);
332 
333 	case DDI_INFO_DEVT2INSTANCE:
334 		*result = (void *)instance;
335 		return (DDI_SUCCESS);
336 
337 	default:
338 		break;
339 	}
340 
341 	return (DDI_FAILURE);
342 }
343 
344 static int
345 iscsit_drv_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
346 {
347 	if (cmd != DDI_ATTACH) {
348 		return (DDI_FAILURE);
349 	}
350 
351 	if (ddi_get_instance(dip) != 0) {
352 		/* we only allow instance 0 to attach */
353 		return (DDI_FAILURE);
354 	}
355 
356 	/* create the minor node */
357 	if (ddi_create_minor_node(dip, ISCSIT_MODNAME, S_IFCHR, 0,
358 	    DDI_PSEUDO, 0) != DDI_SUCCESS) {
359 		cmn_err(CE_WARN, "iscsit_drv_attach: "
360 		    "failed creating minor node");
361 		return (DDI_FAILURE);
362 	}
363 
364 	if (iscsit_init(dip) != IDM_STATUS_SUCCESS) {
365 		cmn_err(CE_WARN, "iscsit_drv_attach: "
366 		    "failed to initialize");
367 		ddi_remove_minor_node(dip, NULL);
368 		return (DDI_FAILURE);
369 	}
370 
371 	iscsit_global.global_svc_state = ISE_DISABLED;
372 	iscsit_global.global_dip = dip;
373 
374 	return (DDI_SUCCESS);
375 }
376 
377 /*ARGSUSED*/
378 static int
379 iscsit_drv_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
380 {
381 	if (cmd != DDI_DETACH)
382 		return (DDI_FAILURE);
383 
384 	/*
385 	 * drv_detach is called in a context that owns the
386 	 * device node for the /dev/pseudo device.  If this thread blocks
387 	 * for any resource, other threads that need the /dev/pseudo device
388 	 * may end up in a deadlock with this thread.Hence, we use a
389 	 * separate lock just for the structures that drv_detach needs
390 	 * to access.
391 	 */
392 	mutex_enter(&iscsit_global.global_state_mutex);
393 	if (iscsit_drv_busy()) {
394 		mutex_exit(&iscsit_global.global_state_mutex);
395 		return (EBUSY);
396 	}
397 
398 	iscsit_global.global_dip = NULL;
399 	ddi_remove_minor_node(dip, NULL);
400 
401 	ldi_ident_release(iscsit_global.global_li);
402 	iscsit_global.global_svc_state = ISE_DETACHED;
403 
404 	mutex_exit(&iscsit_global.global_state_mutex);
405 
406 	return (DDI_SUCCESS);
407 }
408 
409 /*ARGSUSED*/
410 static int
411 iscsit_drv_open(dev_t *devp, int flag, int otyp, cred_t *credp)
412 {
413 	return (0);
414 }
415 
416 /* ARGSUSED */
417 static int
418 iscsit_drv_close(dev_t dev, int flag, int otyp, cred_t *credp)
419 {
420 	return (0);
421 }
422 
423 static boolean_t
424 iscsit_drv_busy(void)
425 {
426 	ASSERT(MUTEX_HELD(&iscsit_global.global_state_mutex));
427 
428 	switch (iscsit_global.global_svc_state) {
429 	case ISE_DISABLED:
430 	case ISE_DETACHED:
431 		return (B_FALSE);
432 	default:
433 		return (B_TRUE);
434 	}
435 	/* NOTREACHED */
436 }
437 
438 /* ARGSUSED */
439 static int
440 iscsit_drv_ioctl(dev_t drv, int cmd, intptr_t argp, int flag, cred_t *cred,
441     int *retval)
442 {
443 	iscsit_ioc_set_config_t		setcfg;
444 	iscsit_ioc_set_config32_t	setcfg32;
445 	char				*cfg_pnvlist = NULL;
446 	nvlist_t			*cfg_nvlist = NULL;
447 	it_config_t			*cfg = NULL;
448 	idm_status_t			idmrc;
449 	int				rc = 0;
450 
451 	if (drv_priv(cred) != 0) {
452 		return (EPERM);
453 	}
454 
455 	mutex_enter(&iscsit_global.global_state_mutex);
456 
457 	/*
458 	 * Validate ioctl requests against global service state
459 	 */
460 	switch (iscsit_global.global_svc_state) {
461 	case ISE_ENABLED:
462 		if (cmd == ISCSIT_IOC_DISABLE_SVC) {
463 			iscsit_global.global_svc_state = ISE_DISABLING;
464 		} else if (cmd == ISCSIT_IOC_ENABLE_SVC) {
465 			/* Already enabled */
466 			mutex_exit(&iscsit_global.global_state_mutex);
467 			return (0);
468 		} else {
469 			iscsit_global.global_svc_state = ISE_BUSY;
470 		}
471 		break;
472 	case ISE_DISABLED:
473 		if (cmd == ISCSIT_IOC_ENABLE_SVC) {
474 			iscsit_global.global_svc_state = ISE_ENABLING;
475 		} else if (cmd == ISCSIT_IOC_DISABLE_SVC) {
476 			/* Already disabled */
477 			mutex_exit(&iscsit_global.global_state_mutex);
478 			return (0);
479 		} else {
480 			rc = EFAULT;
481 		}
482 		break;
483 	case ISE_BUSY:
484 	case ISE_ENABLING:
485 	case ISE_DISABLING:
486 		rc = EAGAIN;
487 		break;
488 	case ISE_DETACHED:
489 	default:
490 		rc = EFAULT;
491 		break;
492 	}
493 
494 	mutex_exit(&iscsit_global.global_state_mutex);
495 	if (rc != 0)
496 		return (rc);
497 
498 	/* Handle ioctl request (enable/disable have already been handled) */
499 	switch (cmd) {
500 	case ISCSIT_IOC_SET_CONFIG:
501 		/* Any errors must set state back to ISE_ENABLED */
502 		switch (ddi_model_convert_from(flag & FMODELS)) {
503 		case DDI_MODEL_ILP32:
504 			if (ddi_copyin((void *)argp, &setcfg32,
505 			    sizeof (iscsit_ioc_set_config32_t), flag) != 0) {
506 				rc = EFAULT;
507 				goto cleanup;
508 			}
509 
510 			setcfg.set_cfg_pnvlist =
511 			    (char *)((uintptr_t)setcfg32.set_cfg_pnvlist);
512 			setcfg.set_cfg_vers = setcfg32.set_cfg_vers;
513 			setcfg.set_cfg_pnvlist_len =
514 			    setcfg32.set_cfg_pnvlist_len;
515 			break;
516 		case DDI_MODEL_NONE:
517 			if (ddi_copyin((void *)argp, &setcfg,
518 			    sizeof (iscsit_ioc_set_config_t), flag) != 0) {
519 				rc = EFAULT;
520 				goto cleanup;
521 			}
522 			break;
523 		default:
524 			rc = EFAULT;
525 			goto cleanup;
526 		}
527 
528 		/* Check API version */
529 		if (setcfg.set_cfg_vers != ISCSIT_API_VERS0) {
530 			rc = EINVAL;
531 			goto cleanup;
532 		}
533 
534 		/* Config is in packed nvlist format so unpack it */
535 		cfg_pnvlist = kmem_alloc(setcfg.set_cfg_pnvlist_len,
536 		    KM_SLEEP);
537 		ASSERT(cfg_pnvlist != NULL);
538 
539 		if (ddi_copyin(setcfg.set_cfg_pnvlist, cfg_pnvlist,
540 		    setcfg.set_cfg_pnvlist_len, flag) != 0) {
541 			rc = EFAULT;
542 			goto cleanup;
543 		}
544 
545 		rc = nvlist_unpack(cfg_pnvlist, setcfg.set_cfg_pnvlist_len,
546 		    &cfg_nvlist, KM_SLEEP);
547 		if (rc != 0) {
548 			goto cleanup;
549 		}
550 
551 		/* Translate nvlist */
552 		rc = it_nv_to_config(cfg_nvlist, &cfg);
553 		if (rc != 0) {
554 			cmn_err(CE_WARN, "Configuration is invalid");
555 			goto cleanup;
556 		}
557 
558 		/* Update config */
559 		rc = iscsit_config_merge(cfg);
560 		/* FALLTHROUGH */
561 
562 cleanup:
563 		if (cfg)
564 			it_config_free_cmn(cfg);
565 		if (cfg_pnvlist)
566 			kmem_free(cfg_pnvlist, setcfg.set_cfg_pnvlist_len);
567 		if (cfg_nvlist)
568 			nvlist_free(cfg_nvlist);
569 
570 		/*
571 		 * Now that the reconfig is complete set our state back to
572 		 * enabled.
573 		 */
574 		mutex_enter(&iscsit_global.global_state_mutex);
575 		iscsit_global.global_svc_state = ISE_ENABLED;
576 		mutex_exit(&iscsit_global.global_state_mutex);
577 		break;
578 	case ISCSIT_IOC_ENABLE_SVC: {
579 		iscsit_hostinfo_t hostinfo;
580 
581 		if (ddi_copyin((void *)argp, &hostinfo.length,
582 		    sizeof (hostinfo.length), flag) != 0) {
583 			mutex_enter(&iscsit_global.global_state_mutex);
584 			iscsit_global.global_svc_state = ISE_DISABLED;
585 			mutex_exit(&iscsit_global.global_state_mutex);
586 			return (EFAULT);
587 		}
588 
589 		if (hostinfo.length > sizeof (hostinfo.fqhn))
590 			hostinfo.length = sizeof (hostinfo.fqhn);
591 
592 		if (ddi_copyin((void *)((caddr_t)argp +
593 		    sizeof (hostinfo.length)), &hostinfo.fqhn,
594 		    hostinfo.length, flag) != 0) {
595 			mutex_enter(&iscsit_global.global_state_mutex);
596 			iscsit_global.global_svc_state = ISE_DISABLED;
597 			mutex_exit(&iscsit_global.global_state_mutex);
598 			return (EFAULT);
599 		}
600 
601 		idmrc = iscsit_enable_svc(&hostinfo);
602 		mutex_enter(&iscsit_global.global_state_mutex);
603 		if (idmrc == IDM_STATUS_SUCCESS) {
604 			iscsit_global.global_svc_state = ISE_ENABLED;
605 		} else {
606 			rc = EIO;
607 			iscsit_global.global_svc_state = ISE_DISABLED;
608 		}
609 		mutex_exit(&iscsit_global.global_state_mutex);
610 		break;
611 	}
612 	case ISCSIT_IOC_DISABLE_SVC:
613 		iscsit_disable_svc();
614 		mutex_enter(&iscsit_global.global_state_mutex);
615 		iscsit_global.global_svc_state = ISE_DISABLED;
616 		mutex_exit(&iscsit_global.global_state_mutex);
617 		break;
618 
619 	default:
620 		rc = EINVAL;
621 		mutex_enter(&iscsit_global.global_state_mutex);
622 		iscsit_global.global_svc_state = ISE_ENABLED;
623 		mutex_exit(&iscsit_global.global_state_mutex);
624 	}
625 
626 	return (rc);
627 }
628 
629 static idm_status_t
630 iscsit_init(dev_info_t *dip)
631 {
632 	int			rc;
633 
634 	rc = ldi_ident_from_dip(dip, &iscsit_global.global_li);
635 	ASSERT(rc == 0);  /* Failure indicates invalid argument */
636 
637 	iscsit_global.global_svc_state = ISE_DISABLED;
638 
639 	return (IDM_STATUS_SUCCESS);
640 }
641 
642 /*
643  * iscsit_enable_svc
644  *
645  * registers all the configured targets and target portals with STMF
646  */
647 static idm_status_t
648 iscsit_enable_svc(iscsit_hostinfo_t *hostinfo)
649 {
650 	stmf_port_provider_t	*pp;
651 	stmf_dbuf_store_t	*dbuf_store;
652 	boolean_t		did_iscsit_isns_init;
653 	idm_status_t		retval = IDM_STATUS_SUCCESS;
654 
655 	ASSERT(iscsit_global.global_svc_state == ISE_ENABLING);
656 
657 	/*
658 	 * Make sure that can tell if we have partially allocated
659 	 * in case we need to exit and tear down anything allocated.
660 	 */
661 	iscsit_global.global_tsih_pool = NULL;
662 	iscsit_global.global_dbuf_store = NULL;
663 	iscsit_status_pdu_cache = NULL;
664 	pp = NULL;
665 	iscsit_global.global_pp = NULL;
666 	iscsit_global.global_default_tpg = NULL;
667 	did_iscsit_isns_init = B_FALSE;
668 	iscsit_global.global_dispatch_taskq = NULL;
669 
670 	/* Setup remaining fields in iscsit_global_t */
671 	idm_refcnt_init(&iscsit_global.global_refcnt,
672 	    &iscsit_global);
673 
674 	avl_create(&iscsit_global.global_discovery_sessions,
675 	    iscsit_sess_avl_compare, sizeof (iscsit_sess_t),
676 	    offsetof(iscsit_sess_t, ist_tgt_ln));
677 
678 	avl_create(&iscsit_global.global_target_list,
679 	    iscsit_tgt_avl_compare, sizeof (iscsit_tgt_t),
680 	    offsetof(iscsit_tgt_t, target_global_ln));
681 
682 	list_create(&iscsit_global.global_deleted_target_list,
683 	    sizeof (iscsit_tgt_t),
684 	    offsetof(iscsit_tgt_t, target_global_deleted_ln));
685 
686 	avl_create(&iscsit_global.global_tpg_list,
687 	    iscsit_tpg_avl_compare, sizeof (iscsit_tpg_t),
688 	    offsetof(iscsit_tpg_t, tpg_global_ln));
689 
690 	avl_create(&iscsit_global.global_ini_list,
691 	    iscsit_ini_avl_compare, sizeof (iscsit_ini_t),
692 	    offsetof(iscsit_ini_t, ini_global_ln));
693 
694 	iscsit_global.global_tsih_pool = vmem_create("iscsit_tsih_pool",
695 	    (void *)1, ISCSI_MAX_TSIH, 1, NULL, NULL, NULL, 0,
696 	    VM_SLEEP | VMC_IDENTIFIER);
697 
698 	/*
699 	 * Setup STMF dbuf store.  Our buffers are bound to a specific
700 	 * connection so we really can't let STMF cache buffers for us.
701 	 * Consequently we'll just allocate one global buffer store.
702 	 */
703 	dbuf_store = stmf_alloc(STMF_STRUCT_DBUF_STORE, 0, 0);
704 	if (dbuf_store == NULL) {
705 		retval = IDM_STATUS_FAIL;
706 		goto tear_down_and_return;
707 	}
708 	dbuf_store->ds_alloc_data_buf = iscsit_dbuf_alloc;
709 	dbuf_store->ds_free_data_buf = iscsit_dbuf_free;
710 	dbuf_store->ds_port_private = NULL;
711 	iscsit_global.global_dbuf_store = dbuf_store;
712 
713 	/* Status PDU cache */
714 	iscsit_status_pdu_cache = kmem_cache_create("iscsit_status_pdu_cache",
715 	    sizeof (idm_pdu_t) + sizeof (iscsi_scsi_rsp_hdr_t), 8,
716 	    &iscsit_status_pdu_constructor,
717 	    NULL, NULL, NULL, NULL, KM_SLEEP);
718 
719 	/* Default TPG and portal */
720 	iscsit_global.global_default_tpg = iscsit_tpg_createdefault();
721 	if (iscsit_global.global_default_tpg == NULL) {
722 		retval = IDM_STATUS_FAIL;
723 		goto tear_down_and_return;
724 	}
725 
726 	/* initialize isns client */
727 	(void) iscsit_isns_init(hostinfo);
728 	did_iscsit_isns_init = B_TRUE;
729 
730 	/* Register port provider */
731 	pp = stmf_alloc(STMF_STRUCT_PORT_PROVIDER, 0, 0);
732 	if (pp == NULL) {
733 		retval = IDM_STATUS_FAIL;
734 		goto tear_down_and_return;
735 	}
736 
737 	pp->pp_portif_rev = PORTIF_REV_1;
738 	pp->pp_instance = 0;
739 	pp->pp_name = ISCSIT_MODNAME;
740 	pp->pp_cb = iscsit_pp_cb;
741 
742 	iscsit_global.global_pp = pp;
743 
744 
745 	if (stmf_register_port_provider(pp) != STMF_SUCCESS) {
746 		retval = IDM_STATUS_FAIL;
747 		goto tear_down_and_return;
748 	}
749 
750 	iscsit_global.global_dispatch_taskq = taskq_create("iscsit_dispatch",
751 	    1, minclsyspri, 16, 16, TASKQ_PREPOPULATE);
752 
753 	/* Scan staged PDUs, meaningful in MC/S situations */
754 	iscsit_rxpdu_queue_monitor_start();
755 
756 	return (IDM_STATUS_SUCCESS);
757 
758 tear_down_and_return:
759 
760 	if (iscsit_global.global_dispatch_taskq) {
761 		taskq_destroy(iscsit_global.global_dispatch_taskq);
762 		iscsit_global.global_dispatch_taskq = NULL;
763 	}
764 
765 	if (did_iscsit_isns_init)
766 		iscsit_isns_fini();
767 
768 	if (iscsit_global.global_default_tpg) {
769 		iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
770 		iscsit_global.global_default_tpg = NULL;
771 	}
772 
773 	if (iscsit_global.global_pp)
774 		iscsit_global.global_pp = NULL;
775 
776 	if (pp)
777 		stmf_free(pp);
778 
779 	if (iscsit_status_pdu_cache) {
780 		kmem_cache_destroy(iscsit_status_pdu_cache);
781 		iscsit_status_pdu_cache = NULL;
782 	}
783 
784 	if (iscsit_global.global_dbuf_store) {
785 		stmf_free(iscsit_global.global_dbuf_store);
786 		iscsit_global.global_dbuf_store = NULL;
787 	}
788 
789 	if (iscsit_global.global_tsih_pool) {
790 		vmem_destroy(iscsit_global.global_tsih_pool);
791 		iscsit_global.global_tsih_pool = NULL;
792 	}
793 
794 	avl_destroy(&iscsit_global.global_ini_list);
795 	avl_destroy(&iscsit_global.global_tpg_list);
796 	list_destroy(&iscsit_global.global_deleted_target_list);
797 	avl_destroy(&iscsit_global.global_target_list);
798 	avl_destroy(&iscsit_global.global_discovery_sessions);
799 
800 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
801 
802 	return (retval);
803 }
804 
805 /*
806  * iscsit_disable_svc
807  *
808  * clean up all existing connections and deregister targets from STMF
809  */
810 static void
811 iscsit_disable_svc(void)
812 {
813 	iscsit_sess_t	*sess;
814 
815 	ASSERT(iscsit_global.global_svc_state == ISE_DISABLING);
816 
817 	iscsit_rxpdu_queue_monitor_stop();
818 
819 	/* tear down discovery sessions */
820 	for (sess = avl_first(&iscsit_global.global_discovery_sessions);
821 	    sess != NULL;
822 	    sess = AVL_NEXT(&iscsit_global.global_discovery_sessions, sess))
823 		iscsit_sess_close(sess);
824 
825 	/*
826 	 * Passing NULL to iscsit_config_merge tells it to go to an empty
827 	 * config.
828 	 */
829 	(void) iscsit_config_merge(NULL);
830 
831 	/*
832 	 * Wait until there are no more global references
833 	 */
834 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
835 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
836 
837 	/*
838 	 * Default TPG must be destroyed after global_refcnt is 0.
839 	 */
840 	iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
841 
842 	avl_destroy(&iscsit_global.global_discovery_sessions);
843 	list_destroy(&iscsit_global.global_deleted_target_list);
844 	avl_destroy(&iscsit_global.global_target_list);
845 	avl_destroy(&iscsit_global.global_tpg_list);
846 	avl_destroy(&iscsit_global.global_ini_list);
847 
848 	taskq_destroy(iscsit_global.global_dispatch_taskq);
849 
850 	iscsit_isns_fini();
851 
852 	stmf_free(iscsit_global.global_dbuf_store);
853 	iscsit_global.global_dbuf_store = NULL;
854 
855 	(void) stmf_deregister_port_provider(iscsit_global.global_pp);
856 	stmf_free(iscsit_global.global_pp);
857 	iscsit_global.global_pp = NULL;
858 
859 	kmem_cache_destroy(iscsit_status_pdu_cache);
860 	iscsit_status_pdu_cache = NULL;
861 
862 	vmem_destroy(iscsit_global.global_tsih_pool);
863 	iscsit_global.global_tsih_pool = NULL;
864 }
865 
866 void
867 iscsit_global_hold()
868 {
869 	/*
870 	 * To take out a global hold, we must either own the global
871 	 * state mutex or we must be running inside of an ioctl that
872 	 * has set the global state to ISE_BUSY, ISE_DISABLING, or
873 	 * ISE_ENABLING.  We don't track the "owner" for these flags,
874 	 * so just checking if they are set is enough for now.
875 	 */
876 	ASSERT((iscsit_global.global_svc_state == ISE_ENABLING) ||
877 	    (iscsit_global.global_svc_state == ISE_DISABLING) ||
878 	    (iscsit_global.global_svc_state == ISE_BUSY) ||
879 	    MUTEX_HELD(&iscsit_global.global_state_mutex));
880 
881 	idm_refcnt_hold(&iscsit_global.global_refcnt);
882 }
883 
884 void
885 iscsit_global_rele()
886 {
887 	idm_refcnt_rele(&iscsit_global.global_refcnt);
888 }
889 
890 void
891 iscsit_global_wait_ref()
892 {
893 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
894 }
895 
896 /*
897  * IDM callbacks
898  */
899 
900 /*ARGSUSED*/
901 void
902 iscsit_rx_pdu(idm_conn_t *ic, idm_pdu_t *rx_pdu)
903 {
904 	iscsit_conn_t *ict = ic->ic_handle;
905 	switch (IDM_PDU_OPCODE(rx_pdu)) {
906 	case ISCSI_OP_SCSI_CMD:
907 		ASSERT(0); /* Shouldn't happen */
908 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
909 		break;
910 	case ISCSI_OP_SNACK_CMD:
911 		/*
912 		 * We'll need to handle this when we support ERL1/2.  For
913 		 * now we treat it as a protocol error.
914 		 */
915 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
916 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
917 		break;
918 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
919 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
920 			iscsit_set_cmdsn(ict, rx_pdu);
921 			iscsit_op_scsi_task_mgmt(ict, rx_pdu);
922 		}
923 		break;
924 	case ISCSI_OP_NOOP_OUT:
925 	case ISCSI_OP_LOGIN_CMD:
926 	case ISCSI_OP_TEXT_CMD:
927 	case ISCSI_OP_LOGOUT_CMD:
928 		/*
929 		 * If/when we switch to userland processing these PDU's
930 		 * will be handled by iscsitd.
931 		 */
932 		iscsit_deferred_dispatch(rx_pdu);
933 		break;
934 	default:
935 		/* Protocol error */
936 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
937 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
938 		break;
939 	}
940 }
941 
942 /*ARGSUSED*/
943 void
944 iscsit_rx_pdu_error(idm_conn_t *ic, idm_pdu_t *rx_pdu, idm_status_t status)
945 {
946 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
947 }
948 
949 void
950 iscsit_task_aborted(idm_task_t *idt, idm_status_t status)
951 {
952 	iscsit_task_t *itask = idt->idt_private;
953 
954 	switch (status) {
955 	case IDM_STATUS_SUSPENDED:
956 		break;
957 	case IDM_STATUS_ABORTED:
958 		mutex_enter(&itask->it_mutex);
959 		itask->it_aborted = B_TRUE;
960 		/*
961 		 * We rely on the fact that STMF tracks outstanding
962 		 * buffer transfers and will free all of our buffers
963 		 * before freeing the task so we don't need to
964 		 * explicitly free the buffers from iscsit/idm
965 		 */
966 		if (itask->it_stmf_abort) {
967 			mutex_exit(&itask->it_mutex);
968 			/*
969 			 * Task is no longer active
970 			 */
971 			iscsit_task_done(itask);
972 
973 			/*
974 			 * STMF has already asked for this task to be aborted
975 			 *
976 			 * STMF specification is wrong... says to return
977 			 * STMF_ABORTED, the code actually looks for
978 			 * STMF_ABORT_SUCCESS.
979 			 */
980 			stmf_task_lport_aborted(itask->it_stmf_task,
981 			    STMF_ABORT_SUCCESS, STMF_IOF_LPORT_DONE);
982 			return;
983 		} else {
984 			mutex_exit(&itask->it_mutex);
985 			/*
986 			 * Tell STMF to stop processing the task.
987 			 */
988 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
989 			    STMF_ABORTED, NULL);
990 			return;
991 		}
992 		/*NOTREACHED*/
993 	default:
994 		ASSERT(0);
995 	}
996 }
997 
998 /*ARGSUSED*/
999 idm_status_t
1000 iscsit_client_notify(idm_conn_t *ic, idm_client_notify_t icn,
1001     uintptr_t data)
1002 {
1003 	idm_status_t rc = IDM_STATUS_SUCCESS;
1004 
1005 	/*
1006 	 * IDM client notifications will never occur at interrupt level
1007 	 * since they are generated from the connection state machine which
1008 	 * running on taskq threads.
1009 	 *
1010 	 */
1011 	switch (icn) {
1012 	case CN_CONNECT_ACCEPT:
1013 		rc = iscsit_conn_accept(ic); /* No data */
1014 		break;
1015 	case CN_FFP_ENABLED:
1016 		rc = iscsit_ffp_enabled(ic); /* No data */
1017 		break;
1018 	case CN_FFP_DISABLED:
1019 		/*
1020 		 * Data indicates whether this was the result of an
1021 		 * explicit logout request.
1022 		 */
1023 		rc = iscsit_ffp_disabled(ic, (idm_ffp_disable_t)data);
1024 		break;
1025 	case CN_CONNECT_LOST:
1026 		rc = iscsit_conn_lost(ic);
1027 		break;
1028 	case CN_CONNECT_DESTROY:
1029 		rc = iscsit_conn_destroy(ic);
1030 		break;
1031 	case CN_LOGIN_FAIL:
1032 		/*
1033 		 * Force the login state machine to completion
1034 		 */
1035 		rc = iscsit_login_fail(ic);
1036 		break;
1037 	default:
1038 		rc = IDM_STATUS_REJECT;
1039 		break;
1040 	}
1041 
1042 	return (rc);
1043 }
1044 
1045 /*
1046  * iscsit_update_statsn is invoked for all the PDUs which have the StatSN
1047  * field in the header. The StatSN is incremented if the IDM_PDU_ADVANCE_STATSN
1048  * flag is set in the pdu flags field. The StatSN is connection-wide and is
1049  * protected by the mutex ict_statsn_mutex. For Data-In PDUs, if the flag
1050  * IDM_TASK_PHASECOLLAPSE_REQ is set, the status (phase-collapse) is also filled
1051  */
1052 void
1053 iscsit_update_statsn(idm_task_t *idm_task, idm_pdu_t *pdu)
1054 {
1055 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1056 	iscsit_conn_t *ict = (iscsit_conn_t *)pdu->isp_ic->ic_handle;
1057 	iscsit_task_t *itask = NULL;
1058 	scsi_task_t *task = NULL;
1059 
1060 	mutex_enter(&ict->ict_statsn_mutex);
1061 	rsp->statsn = htonl(ict->ict_statsn);
1062 	if (pdu->isp_flags & IDM_PDU_ADVANCE_STATSN)
1063 		ict->ict_statsn++;
1064 	mutex_exit(&ict->ict_statsn_mutex);
1065 
1066 	/*
1067 	 * The last SCSI Data PDU passed for a command may also contain the
1068 	 * status if the status indicates termination with no expections, i.e.
1069 	 * no sense data or response involved. If the command completes with
1070 	 * an error, then the response and sense data will be sent in a
1071 	 * separate iSCSI Response PDU.
1072 	 */
1073 	if ((idm_task) && (idm_task->idt_flags & IDM_TASK_PHASECOLLAPSE_REQ)) {
1074 		itask = idm_task->idt_private;
1075 		task = itask->it_stmf_task;
1076 
1077 		rsp->cmd_status = task->task_scsi_status;
1078 		rsp->flags	|= ISCSI_FLAG_DATA_STATUS;
1079 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1080 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1081 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1082 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1083 		}
1084 		rsp->residual_count = htonl(task->task_resid);
1085 
1086 		/*
1087 		 * Removing the task from the session task list
1088 		 * just before the status is sent in the last
1089 		 * Data PDU transfer
1090 		 */
1091 		iscsit_task_done(itask);
1092 	}
1093 }
1094 
1095 void
1096 iscsit_build_hdr(idm_task_t *idm_task, idm_pdu_t *pdu, uint8_t opcode)
1097 {
1098 	iscsit_task_t *itask = idm_task->idt_private;
1099 	iscsi_data_rsp_hdr_t *dh = (iscsi_data_rsp_hdr_t *)pdu->isp_hdr;
1100 
1101 	/*
1102 	 * We acquired iscsit_sess_t.ist_sn_mutex in iscsit_xfer_scsi_data
1103 	 */
1104 	ASSERT(MUTEX_HELD(&itask->it_ict->ict_sess->ist_sn_mutex));
1105 	/*
1106 	 * On incoming data, the target transfer tag and Lun is only
1107 	 * provided by the target if the A bit is set, Since the target
1108 	 * does not currently support Error Recovery Level 1, the A
1109 	 * bit is never set.
1110 	 */
1111 	dh->opcode = opcode;
1112 	dh->itt = itask->it_itt;
1113 	dh->ttt = ((opcode & ISCSI_OPCODE_MASK) == ISCSI_OP_SCSI_DATA_RSP) ?
1114 	    ISCSI_RSVD_TASK_TAG : itask->it_ttt;
1115 
1116 	dh->expcmdsn = htonl(itask->it_ict->ict_sess->ist_expcmdsn);
1117 	dh->maxcmdsn = htonl(itask->it_ict->ict_sess->ist_maxcmdsn);
1118 
1119 	/*
1120 	 * IDM must set:
1121 	 *
1122 	 * data.flags and rtt.flags
1123 	 * data.dlength
1124 	 * data.datasn
1125 	 * data.offset
1126 	 * statsn, residual_count and cmd_status (for phase collapse)
1127 	 * rtt.rttsn
1128 	 * rtt.data_offset
1129 	 * rtt.data_length
1130 	 */
1131 }
1132 
1133 void
1134 iscsit_keepalive(idm_conn_t *ic)
1135 {
1136 	idm_pdu_t		*nop_in_pdu;
1137 	iscsi_nop_in_hdr_t	*nop_in;
1138 	iscsit_conn_t		*ict = ic->ic_handle;
1139 
1140 	/*
1141 	 * IDM noticed the connection has been idle for too long so it's
1142 	 * time to provoke some activity.  Build and transmit an iSCSI
1143 	 * nop-in PDU -- when the initiator responds it will be counted
1144 	 * as "activity" and keep the connection alive.
1145 	 *
1146 	 * We don't actually care about the response here at the iscsit level
1147 	 * so we will just throw it away without looking at it when it arrives.
1148 	 */
1149 	nop_in_pdu = idm_pdu_alloc(sizeof (*nop_in), 0);
1150 	idm_pdu_init(nop_in_pdu, ic, NULL, NULL);
1151 	nop_in = (iscsi_nop_in_hdr_t *)nop_in_pdu->isp_hdr;
1152 	bzero(nop_in, sizeof (*nop_in));
1153 	nop_in->opcode = ISCSI_OP_NOOP_IN;
1154 	nop_in->flags = ISCSI_FLAG_FINAL;
1155 	nop_in->itt = ISCSI_RSVD_TASK_TAG;
1156 	/*
1157 	 * When the target sends a NOP-In as a Ping, the target transfer tag
1158 	 * is set to a valid (not reserved) value and the initiator task tag
1159 	 * is set to ISCSI_RSVD_TASK_TAG (0xffffffff). In this case the StatSN
1160 	 * will always contain the next sequence number but the StatSN for the
1161 	 * connection is not advanced after this PDU is sent.
1162 	 */
1163 	nop_in_pdu->isp_flags |= IDM_PDU_SET_STATSN;
1164 	/*
1165 	 * This works because we don't currently allocate ttt's anywhere else
1166 	 * in iscsit so as long as we stay out of IDM's range we are safe.
1167 	 * If we need to allocate ttt's for other PDU's in the future this will
1168 	 * need to be improved.
1169 	 */
1170 	mutex_enter(&ict->ict_mutex);
1171 	nop_in->ttt = ict->ict_keepalive_ttt;
1172 	ict->ict_keepalive_ttt++;
1173 	if (ict->ict_keepalive_ttt == ISCSI_RSVD_TASK_TAG)
1174 		ict->ict_keepalive_ttt = IDM_TASKIDS_MAX;
1175 	mutex_exit(&ict->ict_mutex);
1176 
1177 	iscsit_pdu_tx(nop_in_pdu);
1178 }
1179 
1180 static idm_status_t
1181 iscsit_conn_accept(idm_conn_t *ic)
1182 {
1183 	iscsit_conn_t *ict;
1184 
1185 	/*
1186 	 * We need to get a global hold here to ensure that the service
1187 	 * doesn't get shutdown prior to establishing a session. This
1188 	 * gets released in iscsit_conn_destroy().
1189 	 */
1190 	mutex_enter(&iscsit_global.global_state_mutex);
1191 	if (iscsit_global.global_svc_state != ISE_ENABLED) {
1192 		mutex_exit(&iscsit_global.global_state_mutex);
1193 		return (IDM_STATUS_FAIL);
1194 	}
1195 	iscsit_global_hold();
1196 	mutex_exit(&iscsit_global.global_state_mutex);
1197 
1198 	/*
1199 	 * Allocate an associated iscsit structure to represent this
1200 	 * connection.  We shouldn't really create a session until we
1201 	 * get the first login PDU.
1202 	 */
1203 	ict = kmem_zalloc(sizeof (*ict), KM_SLEEP);
1204 
1205 	ict->ict_ic = ic;
1206 	ict->ict_statsn = 1;
1207 	ict->ict_keepalive_ttt = IDM_TASKIDS_MAX; /* Avoid IDM TT range */
1208 	ic->ic_handle = ict;
1209 	mutex_init(&ict->ict_mutex, NULL, MUTEX_DRIVER, NULL);
1210 	mutex_init(&ict->ict_statsn_mutex, NULL, MUTEX_DRIVER, NULL);
1211 	idm_refcnt_init(&ict->ict_refcnt, ict);
1212 
1213 	/*
1214 	 * Initialize login state machine
1215 	 */
1216 	if (iscsit_login_sm_init(ict) != IDM_STATUS_SUCCESS) {
1217 		iscsit_global_rele();
1218 		/*
1219 		 * Cleanup the ict after idm notifies us about this failure
1220 		 */
1221 		return (IDM_STATUS_FAIL);
1222 	}
1223 
1224 	return (IDM_STATUS_SUCCESS);
1225 }
1226 
1227 idm_status_t
1228 iscsit_conn_reinstate(iscsit_conn_t *reinstate_ict, iscsit_conn_t *new_ict)
1229 {
1230 	idm_status_t	result;
1231 
1232 	/*
1233 	 * Note in new connection state that this connection is
1234 	 * reinstating an existing connection.
1235 	 */
1236 	new_ict->ict_reinstating = B_TRUE;
1237 	new_ict->ict_reinstate_conn = reinstate_ict;
1238 	new_ict->ict_statsn = reinstate_ict->ict_statsn;
1239 
1240 	/*
1241 	 * Now generate connection state machine event to existing connection
1242 	 * so that it starts the cleanup process.
1243 	 */
1244 	result = idm_conn_reinstate_event(reinstate_ict->ict_ic,
1245 	    new_ict->ict_ic);
1246 
1247 	return (result);
1248 }
1249 
1250 void
1251 iscsit_conn_hold(iscsit_conn_t *ict)
1252 {
1253 	idm_refcnt_hold(&ict->ict_refcnt);
1254 }
1255 
1256 void
1257 iscsit_conn_rele(iscsit_conn_t *ict)
1258 {
1259 	idm_refcnt_rele(&ict->ict_refcnt);
1260 }
1261 
1262 void
1263 iscsit_conn_dispatch_hold(iscsit_conn_t *ict)
1264 {
1265 	idm_refcnt_hold(&ict->ict_dispatch_refcnt);
1266 }
1267 
1268 void
1269 iscsit_conn_dispatch_rele(iscsit_conn_t *ict)
1270 {
1271 	idm_refcnt_rele(&ict->ict_dispatch_refcnt);
1272 }
1273 
1274 static idm_status_t
1275 iscsit_login_fail(idm_conn_t *ic)
1276 {
1277 	iscsit_conn_t *ict = ic->ic_handle;
1278 
1279 	/* Generate login state machine event */
1280 	iscsit_login_sm_event(ict, ILE_LOGIN_CONN_ERROR, NULL);
1281 
1282 	return (IDM_STATUS_SUCCESS);
1283 }
1284 
1285 static idm_status_t
1286 iscsit_ffp_enabled(idm_conn_t *ic)
1287 {
1288 	iscsit_conn_t *ict = ic->ic_handle;
1289 
1290 	/* Generate session state machine event */
1291 	iscsit_sess_sm_event(ict->ict_sess, SE_CONN_LOGGED_IN, ict);
1292 
1293 	return (IDM_STATUS_SUCCESS);
1294 }
1295 
1296 static idm_status_t
1297 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class)
1298 {
1299 	iscsit_conn_t *ict = ic->ic_handle;
1300 
1301 	/* Generate session state machine event */
1302 	switch (disable_class) {
1303 	case FD_CONN_FAIL:
1304 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_FAIL, ict);
1305 		break;
1306 	case FD_CONN_LOGOUT:
1307 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_DISABLE, ict);
1308 		break;
1309 	case FD_SESS_LOGOUT:
1310 		iscsit_sess_sm_event(ict->ict_sess, SE_SESSION_CLOSE, ict);
1311 		break;
1312 	default:
1313 		ASSERT(0);
1314 	}
1315 
1316 	return (IDM_STATUS_SUCCESS);
1317 }
1318 
1319 static idm_status_t
1320 iscsit_conn_lost(idm_conn_t *ic)
1321 {
1322 	iscsit_conn_t	*ict	= ic->ic_handle;
1323 	iscsit_sess_t	*ist	= ict->ict_sess;
1324 	iscsit_cbuf_t	*cbuf;
1325 	idm_pdu_t	*rx_pdu;
1326 	int i;
1327 
1328 	mutex_enter(&ict->ict_mutex);
1329 	ict->ict_lost = B_TRUE;
1330 	mutex_exit(&ict->ict_mutex);
1331 	/*
1332 	 * scrub the staging queue for all PDUs on this connection
1333 	 */
1334 	if (ist != NULL) {
1335 		mutex_enter(&ist->ist_sn_mutex);
1336 		for (cbuf = ist->ist_rxpdu_queue, i = 0;
1337 		    ((cbuf->cb_num_elems > 0) && (i < ISCSIT_RXPDU_QUEUE_LEN));
1338 		    i++) {
1339 			if (((rx_pdu = cbuf->cb_buffer[i]) != NULL) &&
1340 			    (rx_pdu->isp_ic == ic)) {
1341 				/* conn is lost, drop the pdu */
1342 				DTRACE_PROBE3(scrubbing__staging__queue,
1343 				    iscsit_sess_t *, ist, idm_conn_t *, ic,
1344 				    idm_pdu_t *, rx_pdu);
1345 				idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
1346 				cbuf->cb_buffer[i] = NULL;
1347 				cbuf->cb_num_elems--;
1348 				iscsit_conn_dispatch_rele(ict);
1349 			}
1350 		}
1351 		mutex_exit(&ist->ist_sn_mutex);
1352 	}
1353 	/*
1354 	 * Make sure there aren't any PDU's transitioning from the receive
1355 	 * handler to the dispatch taskq.
1356 	 */
1357 	idm_refcnt_wait_ref(&ict->ict_dispatch_refcnt);
1358 
1359 	return (IDM_STATUS_SUCCESS);
1360 }
1361 
1362 static idm_status_t
1363 iscsit_conn_destroy(idm_conn_t *ic)
1364 {
1365 	iscsit_conn_t *ict = ic->ic_handle;
1366 
1367 	mutex_enter(&ict->ict_mutex);
1368 	ict->ict_destroyed = B_TRUE;
1369 	mutex_exit(&ict->ict_mutex);
1370 
1371 	/* Generate session state machine event */
1372 	if (ict->ict_sess != NULL) {
1373 		/*
1374 		 * Session state machine will call iscsit_conn_destroy_done()
1375 		 * when it has removed references to this connection.
1376 		 */
1377 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FAIL, ict);
1378 	}
1379 
1380 	idm_refcnt_wait_ref(&ict->ict_refcnt);
1381 	/*
1382 	 * The session state machine does not need to post
1383 	 * events to IDM any longer, so it is safe to set
1384 	 * the idm connection reference to NULL
1385 	 */
1386 	ict->ict_ic = NULL;
1387 
1388 	/* Reap the login state machine */
1389 	iscsit_login_sm_fini(ict);
1390 
1391 	/* Clean up any text command remnants */
1392 	iscsit_text_cmd_fini(ict);
1393 
1394 	mutex_destroy(&ict->ict_mutex);
1395 	idm_refcnt_destroy(&ict->ict_refcnt);
1396 	kmem_free(ict, sizeof (*ict));
1397 
1398 	iscsit_global_rele();
1399 
1400 	return (IDM_STATUS_SUCCESS);
1401 }
1402 
1403 void
1404 iscsit_conn_logout(iscsit_conn_t *ict)
1405 {
1406 	/*
1407 	 * If the iscsi connection is active, then
1408 	 * logout the IDM connection by sending a
1409 	 * CE_LOGOUT_SESSION_SUCCESS, else, no action
1410 	 * needs to be taken because the connection
1411 	 * is already in the teardown process.
1412 	 */
1413 	mutex_enter(&ict->ict_mutex);
1414 	if (ict->ict_lost == B_FALSE && ict->ict_destroyed == B_FALSE) {
1415 		idm_conn_event(ict->ict_ic, CE_LOGOUT_SESSION_SUCCESS, NULL);
1416 	}
1417 	mutex_exit(&ict->ict_mutex);
1418 }
1419 
1420 /*
1421  * STMF-related functions
1422  *
1423  * iSCSI to STMF mapping
1424  *
1425  * Session == ?
1426  * Connection == bound to local port but not itself a local port
1427  * Target
1428  * Target portal (group?) == local port (really but we're not going to do this)
1429  *	iscsit needs to map connections to local ports (whatever we decide
1430  * 	they are)
1431  * Target == ?
1432  */
1433 
1434 /*ARGSUSED*/
1435 static stmf_data_buf_t *
1436 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
1437     uint32_t flags)
1438 {
1439 	iscsit_task_t *itask = task->task_port_private;
1440 	idm_buf_t *idm_buffer;
1441 	iscsit_buf_t	*ibuf;
1442 	stmf_data_buf_t *result;
1443 	uint32_t	bsize;
1444 
1445 	/*
1446 	 * If the requested size is larger than MaxBurstLength and the
1447 	 * given pminsize is also larger than MaxBurstLength, then the
1448 	 * allocation fails (dbuf = NULL) and pminsize is modified to
1449 	 * be equal to MaxBurstLength. stmf/sbd then should re-invoke
1450 	 * this function with the corrected values for transfer.
1451 	 */
1452 	ASSERT(pminsize);
1453 	if (size <= itask->it_ict->ict_op.op_max_burst_length) {
1454 		bsize = size;
1455 	} else if (*pminsize <= itask->it_ict->ict_op.op_max_burst_length) {
1456 		bsize = itask->it_ict->ict_op.op_max_burst_length;
1457 	} else {
1458 		*pminsize = itask->it_ict->ict_op.op_max_burst_length;
1459 		return (NULL);
1460 	}
1461 
1462 	/* Alloc buffer */
1463 	idm_buffer = idm_buf_alloc(itask->it_ict->ict_ic, NULL, bsize);
1464 	if (idm_buffer != NULL) {
1465 		result = stmf_alloc(STMF_STRUCT_DATA_BUF,
1466 		    sizeof (iscsit_buf_t), 0);
1467 		if (result != NULL) {
1468 			/* Fill in stmf_data_buf_t */
1469 			ibuf = result->db_port_private;
1470 			ibuf->ibuf_idm_buf = idm_buffer;
1471 			ibuf->ibuf_stmf_buf = result;
1472 			ibuf->ibuf_is_immed = B_FALSE;
1473 			result->db_flags = DB_DONT_CACHE;
1474 			result->db_buf_size = bsize;
1475 			result->db_data_size = bsize;
1476 			result->db_sglist_length = 1;
1477 			result->db_sglist[0].seg_addr = idm_buffer->idb_buf;
1478 			result->db_sglist[0].seg_length =
1479 			    idm_buffer->idb_buflen;
1480 			return (result);
1481 		}
1482 
1483 		/* Couldn't get the stmf_data_buf_t so free the buffer */
1484 		idm_buf_free(idm_buffer);
1485 	}
1486 
1487 	return (NULL);
1488 }
1489 
1490 /*ARGSUSED*/
1491 static void
1492 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf)
1493 {
1494 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1495 
1496 	if (ibuf->ibuf_is_immed) {
1497 		/*
1498 		 * The iscsit_buf_t structure itself will be freed with its
1499 		 * associated task.  Here we just need to free the PDU that
1500 		 * held the immediate data.
1501 		 */
1502 		idm_pdu_complete(ibuf->ibuf_immed_data_pdu, IDM_STATUS_SUCCESS);
1503 		ibuf->ibuf_immed_data_pdu = 0;
1504 	} else {
1505 		idm_buf_free(ibuf->ibuf_idm_buf);
1506 		stmf_free(dbuf);
1507 	}
1508 }
1509 
1510 /*ARGSUSED*/
1511 stmf_status_t
1512 iscsit_xfer_scsi_data(scsi_task_t *task, stmf_data_buf_t *dbuf,
1513     uint32_t ioflags)
1514 {
1515 	iscsit_task_t *iscsit_task = task->task_port_private;
1516 	iscsit_sess_t *ict_sess = iscsit_task->it_ict->ict_sess;
1517 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1518 	int idm_rc;
1519 
1520 	/*
1521 	 * If we are aborting then we can ignore this request
1522 	 */
1523 	if (iscsit_task->it_stmf_abort) {
1524 		return (STMF_SUCCESS);
1525 	}
1526 
1527 	/*
1528 	 * If it's not immediate data then start the transfer
1529 	 */
1530 	ASSERT(ibuf->ibuf_is_immed == B_FALSE);
1531 	if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
1532 		/*
1533 		 * The DB_SEND_STATUS_GOOD flag in the STMF data buffer allows
1534 		 * the port provider to phase-collapse, i.e. send the status
1535 		 * along with the final data PDU for the command. The port
1536 		 * provider passes this request to the transport layer by
1537 		 * setting a flag IDM_TASK_PHASECOLLAPSE_REQ in the task.
1538 		 */
1539 		if (dbuf->db_flags & DB_SEND_STATUS_GOOD)
1540 			iscsit_task->it_idm_task->idt_flags |=
1541 			    IDM_TASK_PHASECOLLAPSE_REQ;
1542 		/*
1543 		 * IDM will call iscsit_build_hdr so lock now to serialize
1544 		 * access to the SN values.  We need to lock here to enforce
1545 		 * lock ordering
1546 		 */
1547 		mutex_enter(&ict_sess->ist_sn_mutex);
1548 		idm_rc = idm_buf_tx_to_ini(iscsit_task->it_idm_task,
1549 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1550 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1551 		mutex_exit(&ict_sess->ist_sn_mutex);
1552 
1553 		return (iscsit_idm_to_stmf(idm_rc));
1554 	} else if (dbuf->db_flags & DB_DIRECTION_FROM_RPORT) {
1555 		/* Grab the SN lock (see comment above) */
1556 		mutex_enter(&ict_sess->ist_sn_mutex);
1557 		idm_rc = idm_buf_rx_from_ini(iscsit_task->it_idm_task,
1558 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1559 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1560 		mutex_exit(&ict_sess->ist_sn_mutex);
1561 
1562 		return (iscsit_idm_to_stmf(idm_rc));
1563 	}
1564 
1565 	/* What are we supposed to do if there is no direction? */
1566 	return (STMF_INVALID_ARG);
1567 }
1568 
1569 static void
1570 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status)
1571 {
1572 	iscsit_task_t *itask = idb->idb_task_binding->idt_private;
1573 	stmf_data_buf_t *dbuf = idb->idb_cb_arg;
1574 
1575 	dbuf->db_xfer_status = iscsit_idm_to_stmf(status);
1576 
1577 	/*
1578 	 * If the task has been aborted then we don't need to call STMF
1579 	 */
1580 	if (itask->it_stmf_abort) {
1581 		return;
1582 	}
1583 
1584 	/*
1585 	 * For ISCSI over TCP (not iSER), the last SCSI Data PDU passed
1586 	 * for a successful command contains the status as requested by
1587 	 * by COMSTAR (via the DB_SEND_STATUS_GOOD flag). But the iSER
1588 	 * transport does not support phase-collapse. So pretend we are
1589 	 * COMSTAR and send the status in a separate PDU now.
1590 	 */
1591 	if (idb->idb_task_binding->idt_flags & IDM_TASK_PHASECOLLAPSE_SUCCESS) {
1592 		/*
1593 		 * Mark task complete and notify COMSTAR
1594 		 * that the status has been sent.
1595 		 */
1596 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1597 		stmf_send_status_done(itask->it_stmf_task,
1598 		    iscsit_idm_to_stmf(status), STMF_IOF_LPORT_DONE);
1599 	} else if ((dbuf->db_flags & DB_SEND_STATUS_GOOD) &&
1600 	    status == IDM_STATUS_SUCCESS) {
1601 
1602 		/*
1603 		 * The iscsi target port provider - for iSER, emulates the
1604 		 * DB_SEND_STATUS_GOOD optimization if requested by STMF;
1605 		 * it sends the status in a separate PDU after the data
1606 		 * transfer. In this case the port provider should first
1607 		 * call stmf_data_xfer_done() to mark the transfer complete
1608 		 * and then send the status. Although STMF will free the
1609 		 * buffer at the time the task is freed, even if the transfer
1610 		 * is not marked complete, this behavior makes statistics
1611 		 * gathering and task state tracking more difficult than it
1612 		 * needs to be.
1613 		 */
1614 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1615 		if (iscsit_send_scsi_status(itask->it_stmf_task, 0)
1616 		    != STMF_SUCCESS) {
1617 			stmf_send_status_done(itask->it_stmf_task,
1618 			    STMF_FAILURE, STMF_IOF_LPORT_DONE);
1619 		}
1620 	} else {
1621 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1622 		/* don't touch dbuf after stmf_data_xfer_done */
1623 	}
1624 }
1625 
1626 
1627 /*ARGSUSED*/
1628 stmf_status_t
1629 iscsit_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
1630 {
1631 	iscsit_task_t *itask = task->task_port_private;
1632 	iscsi_scsi_rsp_hdr_t *rsp;
1633 	idm_pdu_t *pdu;
1634 	int resp_datalen;
1635 
1636 	/*
1637 	 * If this task is aborted then we don't need to respond.
1638 	 */
1639 	if (itask->it_stmf_abort) {
1640 		return (STMF_SUCCESS);
1641 	}
1642 
1643 	/*
1644 	 * If this is a task management status, handle it elsewhere.
1645 	 */
1646 	if (task->task_mgmt_function != TM_NONE) {
1647 		/*
1648 		 * Don't wait for the PDU completion to tell STMF
1649 		 * the task is done -- it doesn't really matter and
1650 		 * it makes life complicated if STMF later asks us to
1651 		 * abort the request and we don't know whether the
1652 		 * status has been sent or not.
1653 		 */
1654 		itask->it_tm_responded = B_TRUE;
1655 		iscsit_send_task_mgmt_resp(itask->it_tm_pdu,
1656 		    (task->task_completion_status == STMF_SUCCESS) ?
1657 		    SCSI_TCP_TM_RESP_COMPLETE : SCSI_TCP_TM_RESP_FUNC_NOT_SUPP);
1658 		stmf_send_status_done(task, STMF_SUCCESS,
1659 		    STMF_IOF_LPORT_DONE);
1660 		return (STMF_SUCCESS);
1661 	}
1662 
1663 	/*
1664 	 * Remove the task from the session task list
1665 	 */
1666 	iscsit_task_done(itask);
1667 
1668 	/*
1669 	 * Send status
1670 	 */
1671 	mutex_enter(&itask->it_idm_task->idt_mutex);
1672 	if ((itask->it_idm_task->idt_state == TASK_ACTIVE) &&
1673 	    (task->task_completion_status == STMF_SUCCESS) &&
1674 	    (task->task_sense_length == 0) &&
1675 	    (task->task_resid == 0)) {
1676 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1677 		/* PDU callback releases task hold */
1678 		idm_task_hold(itask->it_idm_task);
1679 		mutex_exit(&itask->it_idm_task->idt_mutex);
1680 		/*
1681 		 * Fast path.  Cached status PDU's are already
1682 		 * initialized.  We just need to fill in
1683 		 * connection and task information. StatSN is
1684 		 * incremented by 1 for every status sent a
1685 		 * connection.
1686 		 */
1687 		pdu = kmem_cache_alloc(iscsit_status_pdu_cache, KM_SLEEP);
1688 		pdu->isp_ic = itask->it_ict->ict_ic;
1689 		pdu->isp_private = itask;
1690 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1691 
1692 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1693 		rsp->itt = itask->it_itt;
1694 		/*
1695 		 * ExpDataSN is the number of R2T and Data-In (read)
1696 		 * PDUs the target has sent for the SCSI command.
1697 		 *
1698 		 * Since there is no support for bidirectional transfer
1699 		 * yet, either idt_exp_datasn or idt_exp_rttsn, but not
1700 		 * both is valid at any time
1701 		 */
1702 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1703 		    htonl(itask->it_idm_task->idt_exp_datasn):
1704 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1705 		rsp->cmd_status = task->task_scsi_status;
1706 		iscsit_pdu_tx(pdu);
1707 		return (STMF_SUCCESS);
1708 	} else {
1709 		if (itask->it_idm_task->idt_state != TASK_ACTIVE) {
1710 			mutex_exit(&itask->it_idm_task->idt_mutex);
1711 			return (STMF_FAILURE);
1712 		}
1713 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1714 		/* PDU callback releases task hold */
1715 		idm_task_hold(itask->it_idm_task);
1716 		mutex_exit(&itask->it_idm_task->idt_mutex);
1717 
1718 		resp_datalen = (task->task_sense_length == 0) ? 0 :
1719 		    (task->task_sense_length + sizeof (uint16_t));
1720 
1721 		pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
1722 		idm_pdu_init(pdu, itask->it_ict->ict_ic, itask,
1723 		    iscsit_send_status_done);
1724 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1725 
1726 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1727 		bzero(rsp, sizeof (*rsp));
1728 		rsp->opcode = ISCSI_OP_SCSI_RSP;
1729 
1730 		rsp->flags = ISCSI_FLAG_FINAL;
1731 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1732 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1733 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1734 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1735 		}
1736 
1737 		rsp->bi_residual_count = 0;
1738 		rsp->residual_count = htonl(task->task_resid);
1739 		rsp->itt = itask->it_itt;
1740 		rsp->response = ISCSI_STATUS_CMD_COMPLETED;
1741 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1742 		    htonl(itask->it_idm_task->idt_exp_datasn):
1743 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1744 		rsp->cmd_status = task->task_scsi_status;
1745 		if (task->task_sense_length != 0) {
1746 			/*
1747 			 * Add a byte to provide the sense length in
1748 			 * the response
1749 			 */
1750 			*(uint16_t *)((void *)pdu->isp_data) =
1751 			    htons(task->task_sense_length);
1752 			bcopy(task->task_sense_data,
1753 			    (uint8_t *)pdu->isp_data +
1754 			    sizeof (uint16_t),
1755 			    task->task_sense_length);
1756 			hton24(rsp->dlength, resp_datalen);
1757 		}
1758 
1759 		DTRACE_PROBE5(iscsi__scsi__response,
1760 		    iscsit_conn_t *, itask->it_ict,
1761 		    uint8_t, rsp->response,
1762 		    uint8_t, rsp->cmd_status,
1763 		    idm_pdu_t *, pdu,
1764 		    scsi_task_t *, task);
1765 
1766 		iscsit_pdu_tx(pdu);
1767 
1768 		return (STMF_SUCCESS);
1769 	}
1770 }
1771 
1772 /*ARGSUSED*/
1773 static void
1774 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status)
1775 {
1776 	iscsit_task_t	*itask;
1777 	boolean_t	aborted;
1778 
1779 	itask = pdu->isp_private;
1780 	aborted = itask->it_stmf_abort;
1781 
1782 	/*
1783 	 * After releasing the hold the task may be freed at any time so
1784 	 * don't touch it.
1785 	 */
1786 	idm_task_rele(itask->it_idm_task);
1787 	if (!aborted) {
1788 		stmf_send_status_done(itask->it_stmf_task,
1789 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1790 	}
1791 	kmem_cache_free(iscsit_status_pdu_cache, pdu);
1792 }
1793 
1794 /*ARGSUSED*/
1795 static void
1796 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status)
1797 {
1798 	iscsit_task_t	 *itask;
1799 	boolean_t	aborted;
1800 
1801 	itask = pdu->isp_private;
1802 	aborted = itask->it_stmf_abort;
1803 
1804 	/*
1805 	 * After releasing the hold the task may be freed at any time so
1806 	 * don't touch it.
1807 	 */
1808 	idm_task_rele(itask->it_idm_task);
1809 	if (!aborted) {
1810 		stmf_send_status_done(itask->it_stmf_task,
1811 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1812 	}
1813 	idm_pdu_free(pdu);
1814 }
1815 
1816 
1817 void
1818 iscsit_lport_task_free(scsi_task_t *task)
1819 {
1820 	iscsit_task_t *itask = task->task_port_private;
1821 
1822 	/* We only call idm_task_start for regular tasks, not task management */
1823 	if (task->task_mgmt_function == TM_NONE) {
1824 		idm_task_done(itask->it_idm_task);
1825 		iscsit_task_free(itask);
1826 		return;
1827 	} else {
1828 		iscsit_tm_task_free(itask);
1829 	}
1830 }
1831 
1832 /*ARGSUSED*/
1833 stmf_status_t
1834 iscsit_abort(stmf_local_port_t *lport, int abort_cmd, void *arg, uint32_t flags)
1835 {
1836 	scsi_task_t	*st = (scsi_task_t *)arg;
1837 	iscsit_task_t	*iscsit_task;
1838 	idm_task_t	*idt;
1839 
1840 	/*
1841 	 * If this is a task management request then there's really not much to
1842 	 * do.
1843 	 */
1844 	if (st->task_mgmt_function != TM_NONE) {
1845 		return (STMF_ABORT_SUCCESS);
1846 	}
1847 
1848 	/*
1849 	 * Regular task, start cleaning up
1850 	 */
1851 	iscsit_task = st->task_port_private;
1852 	idt = iscsit_task->it_idm_task;
1853 	mutex_enter(&iscsit_task->it_mutex);
1854 	iscsit_task->it_stmf_abort = B_TRUE;
1855 	if (iscsit_task->it_aborted) {
1856 		mutex_exit(&iscsit_task->it_mutex);
1857 		/*
1858 		 * Task is no longer active
1859 		 */
1860 		iscsit_task_done(iscsit_task);
1861 
1862 		/*
1863 		 * STMF specification is wrong... says to return
1864 		 * STMF_ABORTED, the code actually looks for
1865 		 * STMF_ABORT_SUCCESS.
1866 		 */
1867 		return (STMF_ABORT_SUCCESS);
1868 	} else {
1869 		mutex_exit(&iscsit_task->it_mutex);
1870 		/*
1871 		 * Call IDM to abort the task.  Due to a variety of
1872 		 * circumstances the task may already be in the process of
1873 		 * aborting.
1874 		 * We'll let IDM worry about rationalizing all that except
1875 		 * for one particular instance.  If the state of the task
1876 		 * is TASK_COMPLETE, we need to indicate to the framework
1877 		 * that we are in fact done.  This typically happens with
1878 		 * framework-initiated task management type requests
1879 		 * (e.g. abort task).
1880 		 */
1881 		if (idt->idt_state == TASK_COMPLETE) {
1882 			idm_refcnt_wait_ref(&idt->idt_refcnt);
1883 			return (STMF_ABORT_SUCCESS);
1884 		} else {
1885 			idm_task_abort(idt->idt_ic, idt, AT_TASK_MGMT_ABORT);
1886 			return (STMF_SUCCESS);
1887 		}
1888 	}
1889 
1890 	/*NOTREACHED*/
1891 }
1892 
1893 /*ARGSUSED*/
1894 void
1895 iscsit_ctl(stmf_local_port_t *lport, int cmd, void *arg)
1896 {
1897 	iscsit_tgt_t		*iscsit_tgt;
1898 
1899 	ASSERT((cmd == STMF_CMD_LPORT_ONLINE) ||
1900 	    (cmd == STMF_ACK_LPORT_ONLINE_COMPLETE) ||
1901 	    (cmd == STMF_CMD_LPORT_OFFLINE) ||
1902 	    (cmd == STMF_ACK_LPORT_OFFLINE_COMPLETE));
1903 
1904 	iscsit_tgt = (iscsit_tgt_t *)lport->lport_port_private;
1905 
1906 	switch (cmd) {
1907 	case STMF_CMD_LPORT_ONLINE:
1908 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_REQ);
1909 		break;
1910 	case STMF_CMD_LPORT_OFFLINE:
1911 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_REQ);
1912 		break;
1913 	case STMF_ACK_LPORT_ONLINE_COMPLETE:
1914 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_COMPLETE_ACK);
1915 		break;
1916 	case STMF_ACK_LPORT_OFFLINE_COMPLETE:
1917 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_COMPLETE_ACK);
1918 		break;
1919 
1920 	default:
1921 		break;
1922 	}
1923 }
1924 
1925 static stmf_status_t
1926 iscsit_idm_to_stmf(idm_status_t idmrc)
1927 {
1928 	switch (idmrc) {
1929 	case IDM_STATUS_SUCCESS:
1930 		return (STMF_SUCCESS);
1931 	default:
1932 		return (STMF_FAILURE);
1933 	}
1934 	/*NOTREACHED*/
1935 }
1936 
1937 void
1938 iscsit_op_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1939 {
1940 	iscsit_conn_t		*ict = ic->ic_handle;
1941 
1942 	if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
1943 		iscsit_post_scsi_cmd(ic, rx_pdu);
1944 	}
1945 	iscsit_process_pdu_in_queue(ict->ict_sess);
1946 }
1947 
1948 /*
1949  * ISCSI protocol
1950  */
1951 
1952 void
1953 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1954 {
1955 	iscsit_conn_t		*ict;
1956 	iscsit_task_t		*itask;
1957 	scsi_task_t		*task;
1958 	iscsit_buf_t		*ibuf;
1959 	iscsi_scsi_cmd_hdr_t	*iscsi_scsi =
1960 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
1961 	iscsi_addl_hdr_t	*ahs_hdr;
1962 	uint16_t		addl_cdb_len = 0;
1963 
1964 	ict = ic->ic_handle;
1965 
1966 	itask = iscsit_task_alloc(ict);
1967 	if (itask == NULL) {
1968 		/* Finish processing request */
1969 		iscsit_set_cmdsn(ict, rx_pdu);
1970 
1971 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
1972 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
1973 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
1974 		return;
1975 	}
1976 
1977 	/*
1978 	 * Note CmdSN and ITT in task.  IDM will have already validated this
1979 	 * request against the connection state so we don't need to check
1980 	 * that (the connection may have changed state in the meantime but
1981 	 * we will catch that when we try to send a response)
1982 	 */
1983 	itask->it_cmdsn = ntohl(iscsi_scsi->cmdsn);
1984 	itask->it_itt = iscsi_scsi->itt;
1985 
1986 	/*
1987 	 * Check for extended CDB AHS
1988 	 */
1989 	if (iscsi_scsi->hlength > 0) {
1990 		ahs_hdr = (iscsi_addl_hdr_t *)iscsi_scsi;
1991 		addl_cdb_len = ((ahs_hdr->ahs_hlen_hi << 8) |
1992 		    ahs_hdr->ahs_hlen_lo) - 1; /* Adjust for reserved byte */
1993 		if (((addl_cdb_len + 4) / sizeof (uint32_t)) >
1994 		    iscsi_scsi->hlength) {
1995 			/* Mangled header info, drop it */
1996 			idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
1997 			return;
1998 		}
1999 	}
2000 
2001 	ict = rx_pdu->isp_ic->ic_handle; /* IDM client private */
2002 
2003 	/*
2004 	 * Add task to session list.  This function will also check to
2005 	 * ensure that the task does not already exist.
2006 	 */
2007 	if (iscsit_task_start(itask) != IDM_STATUS_SUCCESS) {
2008 		/*
2009 		 * Task exists, free all resources and reject.  Don't
2010 		 * update expcmdsn in this case because RFC 3720 says
2011 		 * "The CmdSN of the rejected command PDU (if it is a
2012 		 * non-immediate command) MUST NOT be considered received
2013 		 * by the target (i.e., a command sequence gap must be
2014 		 * assumed for the CmdSN), even though the CmdSN of the
2015 		 * rejected command PDU may be reliably ascertained.  Upon
2016 		 * receiving the Reject, the initiator MUST plug the CmdSN
2017 		 * gap in order to continue to use the session.  The gap
2018 		 * may be plugged either by transmitting a command PDU
2019 		 * with the same CmdSN, or by aborting the task (see section
2020 		 * 6.9 on how an abort may plug a CmdSN gap)." (Section 6.3)
2021 		 */
2022 		iscsit_task_free(itask);
2023 		iscsit_send_reject(ict, rx_pdu, ISCSI_REJECT_TASK_IN_PROGRESS);
2024 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2025 		return;
2026 	}
2027 
2028 	/* Update sequence numbers */
2029 	iscsit_set_cmdsn(ict, rx_pdu);
2030 
2031 	/*
2032 	 * Allocate STMF task
2033 	 */
2034 	itask->it_stmf_task = stmf_task_alloc(
2035 	    itask->it_ict->ict_sess->ist_lport,
2036 	    itask->it_ict->ict_sess->ist_stmf_sess, iscsi_scsi->lun,
2037 	    16 + addl_cdb_len, 0);
2038 	if (itask->it_stmf_task == NULL) {
2039 		/*
2040 		 * Either stmf really couldn't get memory for a task or,
2041 		 * more likely, the LU is currently in reset.  Either way
2042 		 * we have no choice but to fail the request.
2043 		 */
2044 		iscsit_task_done(itask);
2045 		iscsit_task_free(itask);
2046 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
2047 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
2048 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2049 		return;
2050 	}
2051 
2052 	task = itask->it_stmf_task;
2053 	task->task_port_private = itask;
2054 
2055 	bcopy(iscsi_scsi->lun, task->task_lun_no, sizeof (task->task_lun_no));
2056 
2057 	/*
2058 	 * iSCSI and Comstar use the same values.  Should we rely on this
2059 	 * or translate them bit-wise?
2060 	 */
2061 
2062 	task->task_flags =
2063 	    (((iscsi_scsi->flags & ISCSI_FLAG_CMD_READ) ? TF_READ_DATA : 0) |
2064 	    ((iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ? TF_WRITE_DATA : 0) |
2065 	    ((rx_pdu->isp_datalen == 0) ? 0 : TF_INITIAL_BURST));
2066 
2067 	switch (iscsi_scsi->flags & ISCSI_FLAG_CMD_ATTR_MASK) {
2068 	case ISCSI_ATTR_UNTAGGED:
2069 		break;
2070 	case ISCSI_ATTR_SIMPLE:
2071 		task->task_additional_flags |= TF_ATTR_SIMPLE_QUEUE;
2072 		break;
2073 	case ISCSI_ATTR_ORDERED:
2074 		task->task_additional_flags |= TF_ATTR_ORDERED_QUEUE;
2075 		break;
2076 	case ISCSI_ATTR_HEAD_OF_QUEUE:
2077 		task->task_additional_flags |= TF_ATTR_HEAD_OF_QUEUE;
2078 		break;
2079 	case ISCSI_ATTR_ACA:
2080 		task->task_additional_flags |= TF_ATTR_ACA;
2081 		break;
2082 	default:
2083 		/* Protocol error but just take it, treat as untagged */
2084 		break;
2085 	}
2086 
2087 
2088 	task->task_additional_flags = 0;
2089 	task->task_priority = 0;
2090 	task->task_mgmt_function = TM_NONE;
2091 
2092 	/*
2093 	 * This "task_max_nbufs" doesn't map well to BIDI.  We probably need
2094 	 * parameter for each direction.  "MaxOutstandingR2T" may very well
2095 	 * be set to one which could prevent us from doing simultaneous
2096 	 * transfers in each direction.
2097 	 */
2098 	task->task_max_nbufs = (iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ?
2099 	    ict->ict_op.op_max_outstanding_r2t : STMF_BUFS_MAX;
2100 	task->task_cmd_seq_no = ntohl(iscsi_scsi->itt);
2101 	task->task_expected_xfer_length = ntohl(iscsi_scsi->data_length);
2102 
2103 	/* Copy CDB */
2104 	bcopy(iscsi_scsi->scb, task->task_cdb, 16);
2105 	if (addl_cdb_len > 0) {
2106 		bcopy(ahs_hdr->ahs_extscb, task->task_cdb + 16, addl_cdb_len);
2107 	}
2108 
2109 	DTRACE_ISCSI_3(scsi__command, idm_conn_t *, ic,
2110 	    iscsi_scsi_cmd_hdr_t *, (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr,
2111 	    scsi_task_t *, task);
2112 
2113 	/*
2114 	 * Copy the transport header into the task handle from the PDU
2115 	 * handle. The transport header describes this task's remote tagged
2116 	 * buffer.
2117 	 */
2118 	if (rx_pdu->isp_transport_hdrlen != 0) {
2119 		bcopy(rx_pdu->isp_transport_hdr,
2120 		    itask->it_idm_task->idt_transport_hdr,
2121 		    rx_pdu->isp_transport_hdrlen);
2122 	}
2123 
2124 	/*
2125 	 * Tell IDM about our new active task
2126 	 */
2127 	idm_task_start(itask->it_idm_task, (uintptr_t)itask->it_itt);
2128 
2129 	/*
2130 	 * If we have any immediate data then setup the immediate buffer
2131 	 * context that comes with the task
2132 	 */
2133 	if (rx_pdu->isp_datalen) {
2134 		ibuf = itask->it_immed_data;
2135 		ibuf->ibuf_immed_data_pdu = rx_pdu;
2136 		ibuf->ibuf_stmf_buf->db_data_size = rx_pdu->isp_datalen;
2137 		ibuf->ibuf_stmf_buf->db_buf_size = rx_pdu->isp_datalen;
2138 		ibuf->ibuf_stmf_buf->db_relative_offset = 0;
2139 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_length =
2140 		    rx_pdu->isp_datalen;
2141 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr = rx_pdu->isp_data;
2142 
2143 		DTRACE_ISCSI_8(xfer__start, idm_conn_t *, ic,
2144 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2145 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2146 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2147 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2148 
2149 		/*
2150 		 * For immediate data transfer, there is no callback from
2151 		 * stmf to indicate that the initial burst of data is
2152 		 * transferred successfully. In some cases, the task can
2153 		 * get freed before execution returns from stmf_post_task.
2154 		 * Although this xfer-start/done probe accurately tracks
2155 		 * the size of the transfer, it does only provide a best
2156 		 * effort on the timing of the transfer.
2157 		 */
2158 		DTRACE_ISCSI_8(xfer__done, idm_conn_t *, ic,
2159 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2160 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2161 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2162 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2163 		stmf_post_task(task, ibuf->ibuf_stmf_buf);
2164 	} else {
2165 
2166 		stmf_post_task(task, NULL);
2167 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2168 	}
2169 }
2170 
2171 void
2172 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu)
2173 {
2174 	iscsit_conn_t *ict = rx_pdu->isp_ic->ic_handle;
2175 
2176 	/*
2177 	 * If the connection has been lost then ignore new PDU's
2178 	 */
2179 	mutex_enter(&ict->ict_mutex);
2180 	if (ict->ict_lost) {
2181 		mutex_exit(&ict->ict_mutex);
2182 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2183 		return;
2184 	}
2185 
2186 	/*
2187 	 * Grab a hold on the connection to prevent it from going away
2188 	 * between now and when the taskq function is called.
2189 	 */
2190 	iscsit_conn_dispatch_hold(ict);
2191 	mutex_exit(&ict->ict_mutex);
2192 
2193 	taskq_dispatch_ent(iscsit_global.global_dispatch_taskq,
2194 	    iscsit_deferred, rx_pdu, 0, &rx_pdu->isp_tqent);
2195 }
2196 
2197 static void
2198 iscsit_deferred(void *rx_pdu_void)
2199 {
2200 	idm_pdu_t		*rx_pdu = rx_pdu_void;
2201 	idm_conn_t		*ic = rx_pdu->isp_ic;
2202 	iscsit_conn_t		*ict = ic->ic_handle;
2203 
2204 	/*
2205 	 * NOP and Task Management Commands can be marked for immediate
2206 	 * delivery. Commands marked as 'Immediate' are to be considered
2207 	 * for execution as soon as they arrive on the target. So these
2208 	 * should not be checked for sequence order and put in a queue.
2209 	 * The CmdSN is not advanced for Immediate Commands.
2210 	 */
2211 	switch (IDM_PDU_OPCODE(rx_pdu)) {
2212 	case ISCSI_OP_NOOP_OUT:
2213 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2214 			iscsit_set_cmdsn(ict, rx_pdu);
2215 			iscsit_pdu_op_noop(ict, rx_pdu);
2216 		}
2217 		break;
2218 	case ISCSI_OP_LOGIN_CMD:
2219 		iscsit_pdu_op_login_cmd(ict, rx_pdu);
2220 		iscsit_conn_dispatch_rele(ict);
2221 		return;
2222 	case ISCSI_OP_TEXT_CMD:
2223 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2224 			iscsit_set_cmdsn(ict, rx_pdu);
2225 			iscsit_pdu_op_text_cmd(ict, rx_pdu);
2226 		}
2227 		break;
2228 	case ISCSI_OP_LOGOUT_CMD:
2229 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2230 			iscsit_set_cmdsn(ict, rx_pdu);
2231 			iscsit_pdu_op_logout_cmd(ict, rx_pdu);
2232 		}
2233 		break;
2234 	default:
2235 		/* Protocol error.  IDM should have caught this */
2236 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2237 		ASSERT(0);
2238 		break;
2239 	}
2240 	/*
2241 	 * Check if there are other PDUs in the session staging queue
2242 	 * waiting to be posted to SCSI layer.
2243 	 */
2244 	iscsit_process_pdu_in_queue(ict->ict_sess);
2245 
2246 	iscsit_conn_dispatch_rele(ict);
2247 }
2248 
2249 static void
2250 iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
2251     uint8_t response, uint8_t cmd_status)
2252 {
2253 	idm_pdu_t			*rsp_pdu;
2254 	idm_conn_t			*ic;
2255 	iscsi_scsi_rsp_hdr_t		*resp;
2256 	iscsi_scsi_cmd_hdr_t		*req =
2257 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2258 
2259 	ic = ict->ict_ic;
2260 
2261 	rsp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_rsp_hdr_t), 0);
2262 	idm_pdu_init(rsp_pdu, ic, NULL, NULL);
2263 	/*
2264 	 * StatSN is incremented by 1 for every response sent on
2265 	 * a connection except for responses sent as a result of
2266 	 * a retry or SNACK
2267 	 */
2268 	rsp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2269 
2270 	resp = (iscsi_scsi_rsp_hdr_t *)rsp_pdu->isp_hdr;
2271 
2272 	resp->opcode = ISCSI_OP_SCSI_RSP;
2273 	resp->flags = ISCSI_FLAG_FINAL;
2274 	resp->response = response;
2275 	resp->cmd_status = cmd_status;
2276 	resp->itt = req->itt;
2277 	if ((response == ISCSI_STATUS_CMD_COMPLETED) &&
2278 	    (req->data_length != 0) &&
2279 	    ((req->flags & ISCSI_FLAG_CMD_READ) ||
2280 	    (req->flags & ISCSI_FLAG_CMD_WRITE))) {
2281 		resp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
2282 		resp->residual_count = req->data_length;
2283 	}
2284 
2285 	DTRACE_PROBE4(iscsi__scsi__direct__response,
2286 	    iscsit_conn_t *, ict,
2287 	    uint8_t, resp->response,
2288 	    uint8_t, resp->cmd_status,
2289 	    idm_pdu_t *, rsp_pdu);
2290 
2291 	iscsit_pdu_tx(rsp_pdu);
2292 }
2293 
2294 void
2295 iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu, uint8_t tm_status)
2296 {
2297 	iscsi_scsi_task_mgt_rsp_hdr_t	*tm_resp;
2298 
2299 	/*
2300 	 * The target must take note of the last-sent StatSN.
2301 	 * The StatSN is to be incremented after sending a
2302 	 * task management response. Digest recovery can only
2303 	 * work if StatSN is incremented.
2304 	 */
2305 	tm_resp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2306 	tm_resp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2307 	tm_resp->response = tm_status;
2308 
2309 	DTRACE_PROBE3(iscsi__scsi__tm__response,
2310 	    iscsit_conn_t *, tm_resp_pdu->isp_ic->ic_handle,
2311 	    uint8_t, tm_resp->response,
2312 	    idm_pdu_t *, tm_resp_pdu);
2313 	iscsit_pdu_tx(tm_resp_pdu);
2314 }
2315 
2316 void
2317 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2318 {
2319 	idm_pdu_t			*tm_resp_pdu;
2320 	iscsit_task_t			*itask;
2321 	iscsit_task_t			*tm_itask;
2322 	scsi_task_t			*task;
2323 	iscsi_scsi_task_mgt_hdr_t 	*iscsi_tm =
2324 	    (iscsi_scsi_task_mgt_hdr_t *)rx_pdu->isp_hdr;
2325 	iscsi_scsi_task_mgt_rsp_hdr_t 	*iscsi_tm_rsp =
2326 	    (iscsi_scsi_task_mgt_rsp_hdr_t *)rx_pdu->isp_hdr;
2327 	uint32_t			rtt, cmdsn, refcmdsn;
2328 	uint8_t				tm_func;
2329 
2330 	/*
2331 	 * Setup response PDU (response field will get filled in later)
2332 	 */
2333 	tm_resp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_task_mgt_rsp_hdr_t), 0);
2334 	if (tm_resp_pdu == NULL) {
2335 		/* Can't respond, just drop it */
2336 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2337 		return;
2338 	}
2339 	idm_pdu_init(tm_resp_pdu, ict->ict_ic, NULL, NULL);
2340 	iscsi_tm_rsp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2341 	bzero(iscsi_tm_rsp, sizeof (iscsi_scsi_task_mgt_rsp_hdr_t));
2342 	iscsi_tm_rsp->opcode = ISCSI_OP_SCSI_TASK_MGT_RSP;
2343 	iscsi_tm_rsp->flags = ISCSI_FLAG_FINAL;
2344 	iscsi_tm_rsp->itt = rx_pdu->isp_hdr->itt;
2345 
2346 	/*
2347 	 * Figure out what we're being asked to do.
2348 	 */
2349 	DTRACE_PROBE4(iscsi__scsi__tm__request,
2350 	    iscsit_conn_t *, ict,
2351 	    uint8_t, (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK),
2352 	    uint32_t, iscsi_tm->rtt,
2353 	    idm_pdu_t *, rx_pdu);
2354 	switch (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK) {
2355 	case ISCSI_TM_FUNC_ABORT_TASK:
2356 		/*
2357 		 * STMF doesn't currently support the "abort task" task
2358 		 * management command although it does support aborting
2359 		 * an individual task.  We'll get STMF to abort the task
2360 		 * for us but handle the details of the task management
2361 		 * command ourselves.
2362 		 *
2363 		 * Find the task associated with the referenced task tag.
2364 		 */
2365 		rtt = iscsi_tm->rtt;
2366 		itask = (iscsit_task_t *)idm_task_find_by_handle(ict->ict_ic,
2367 		    (uintptr_t)rtt);
2368 
2369 		if (itask == NULL) {
2370 			cmdsn = ntohl(iscsi_tm->cmdsn);
2371 			refcmdsn = ntohl(iscsi_tm->refcmdsn);
2372 
2373 			/*
2374 			 * Task was not found. But the SCSI command could be
2375 			 * on the rxpdu wait queue. If RefCmdSN is within
2376 			 * the CmdSN window and less than CmdSN of the TM
2377 			 * function, return "Function Complete". Otherwise,
2378 			 * return "Task Does Not Exist".
2379 			 */
2380 
2381 			if (iscsit_cmdsn_in_window(ict, refcmdsn) &&
2382 			    iscsit_sna_lt(refcmdsn, cmdsn)) {
2383 				mutex_enter(&ict->ict_sess->ist_sn_mutex);
2384 				(void) iscsit_remove_pdu_from_queue(
2385 				    ict->ict_sess, refcmdsn);
2386 				iscsit_conn_dispatch_rele(ict);
2387 				mutex_exit(&ict->ict_sess->ist_sn_mutex);
2388 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2389 				    SCSI_TCP_TM_RESP_COMPLETE);
2390 			} else {
2391 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2392 				    SCSI_TCP_TM_RESP_NO_TASK);
2393 			}
2394 		} else {
2395 
2396 			/*
2397 			 * Tell STMF to abort the task.  This will do no harm
2398 			 * if the task is already complete.
2399 			 */
2400 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
2401 			    STMF_ABORTED, NULL);
2402 
2403 			/*
2404 			 * Make sure the task hasn't already completed
2405 			 */
2406 			mutex_enter(&itask->it_idm_task->idt_mutex);
2407 			if ((itask->it_idm_task->idt_state == TASK_COMPLETE) ||
2408 			    (itask->it_idm_task->idt_state == TASK_IDLE)) {
2409 				/*
2410 				 * Task is complete, return "Task Does Not
2411 				 * Exist"
2412 				 */
2413 				mutex_exit(&itask->it_idm_task->idt_mutex);
2414 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2415 				    SCSI_TCP_TM_RESP_NO_TASK);
2416 			} else {
2417 				/*
2418 				 * STMF is now aborting the task, return
2419 				 * "Function Complete"
2420 				 */
2421 				mutex_exit(&itask->it_idm_task->idt_mutex);
2422 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2423 				    SCSI_TCP_TM_RESP_COMPLETE);
2424 			}
2425 			idm_task_rele(itask->it_idm_task);
2426 		}
2427 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2428 		return;
2429 
2430 	case ISCSI_TM_FUNC_ABORT_TASK_SET:
2431 		tm_func = TM_ABORT_TASK_SET;
2432 		break;
2433 
2434 	case ISCSI_TM_FUNC_CLEAR_ACA:
2435 		tm_func = TM_CLEAR_ACA;
2436 		break;
2437 
2438 	case ISCSI_TM_FUNC_CLEAR_TASK_SET:
2439 		tm_func = TM_CLEAR_TASK_SET;
2440 		break;
2441 
2442 	case ISCSI_TM_FUNC_LOGICAL_UNIT_RESET:
2443 		tm_func = TM_LUN_RESET;
2444 		break;
2445 
2446 	case ISCSI_TM_FUNC_TARGET_WARM_RESET:
2447 		tm_func = TM_TARGET_WARM_RESET;
2448 		break;
2449 
2450 	case ISCSI_TM_FUNC_TARGET_COLD_RESET:
2451 		tm_func = TM_TARGET_COLD_RESET;
2452 		break;
2453 
2454 	case ISCSI_TM_FUNC_TASK_REASSIGN:
2455 		/*
2456 		 * We do not currently support allegiance reassignment.  When
2457 		 * we start supporting ERL1+, we will need to.
2458 		 */
2459 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2460 		    SCSI_TCP_TM_RESP_NO_ALLG_REASSN);
2461 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2462 		return;
2463 
2464 	default:
2465 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2466 		    SCSI_TCP_TM_RESP_REJECTED);
2467 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2468 		return;
2469 	}
2470 
2471 	tm_itask = iscsit_tm_task_alloc(ict);
2472 	if (tm_itask == NULL) {
2473 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2474 		    SCSI_TCP_TM_RESP_REJECTED);
2475 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2476 		return;
2477 	}
2478 
2479 
2480 	task = stmf_task_alloc(ict->ict_sess->ist_lport,
2481 	    ict->ict_sess->ist_stmf_sess, iscsi_tm->lun,
2482 	    0, STMF_TASK_EXT_NONE);
2483 	if (task == NULL) {
2484 		/*
2485 		 * If this happens, either the LU is in reset, couldn't
2486 		 * get memory, or some other condition in which we simply
2487 		 * can't complete this request.  It would be nice to return
2488 		 * an error code like "busy" but the closest we have is
2489 		 * "rejected".
2490 		 */
2491 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2492 		    SCSI_TCP_TM_RESP_REJECTED);
2493 		iscsit_tm_task_free(tm_itask);
2494 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2495 		return;
2496 	}
2497 
2498 	tm_itask->it_tm_pdu = tm_resp_pdu;
2499 	tm_itask->it_stmf_task = task;
2500 	task->task_port_private = tm_itask;
2501 	task->task_mgmt_function = tm_func;
2502 	task->task_additional_flags = TASK_AF_NO_EXPECTED_XFER_LENGTH;
2503 	task->task_priority = 0;
2504 	task->task_max_nbufs = STMF_BUFS_MAX;
2505 	task->task_cmd_seq_no = iscsi_tm->itt;
2506 	task->task_expected_xfer_length = 0;
2507 
2508 	stmf_post_task(task, NULL);
2509 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2510 }
2511 
2512 static void
2513 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2514 {
2515 	iscsi_nop_out_hdr_t *out = (iscsi_nop_out_hdr_t *)rx_pdu->isp_hdr;
2516 	iscsi_nop_in_hdr_t *in;
2517 	int resp_datalen;
2518 	idm_pdu_t *resp;
2519 
2520 	/* Ignore the response from initiator */
2521 	if ((out->itt == ISCSI_RSVD_TASK_TAG) ||
2522 	    (out->ttt != ISCSI_RSVD_TASK_TAG)) {
2523 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2524 		return;
2525 	}
2526 
2527 	/* Allocate a PDU to respond */
2528 	resp_datalen = ntoh24(out->dlength);
2529 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
2530 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2531 	if (resp_datalen > 0) {
2532 		bcopy(rx_pdu->isp_data, resp->isp_data, resp_datalen);
2533 	}
2534 
2535 	/*
2536 	 * When sending a NOP-In as a response to a NOP-Out from the initiator,
2537 	 * the target must respond with the same initiator task tag that was
2538 	 * provided in the NOP-Out request, the target transfer tag must be
2539 	 * ISCSI_RSVD_TASK_TAG (0xffffffff) and StatSN will contain the next
2540 	 * status sequence number. The StatSN for the connection is advanced
2541 	 * after this PDU is sent.
2542 	 */
2543 	in = (iscsi_nop_in_hdr_t *)resp->isp_hdr;
2544 	bzero(in, sizeof (*in));
2545 	in->opcode = ISCSI_OP_NOOP_IN;
2546 	in->flags = ISCSI_FLAG_FINAL;
2547 	bcopy(out->lun, in->lun, 8);
2548 	in->itt		= out->itt;
2549 	in->ttt		= ISCSI_RSVD_TASK_TAG;
2550 	hton24(in->dlength, resp_datalen);
2551 	resp->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2552 	/* Any other field in resp to be set? */
2553 	iscsit_pdu_tx(resp);
2554 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2555 }
2556 
2557 static void
2558 iscsit_pdu_op_login_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2559 {
2560 
2561 	/*
2562 	 * Submit PDU to login state machine.  State machine will free the
2563 	 * PDU.
2564 	 */
2565 	iscsit_login_sm_event(ict, ILE_LOGIN_RCV, rx_pdu);
2566 }
2567 
2568 void
2569 iscsit_pdu_op_logout_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2570 {
2571 	iscsi_logout_hdr_t 	*logout_req =
2572 	    (iscsi_logout_hdr_t *)rx_pdu->isp_hdr;
2573 	iscsi_logout_rsp_hdr_t	*logout_rsp;
2574 	idm_pdu_t *resp;
2575 
2576 	/* Allocate a PDU to respond */
2577 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2578 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2579 	/*
2580 	 * The StatSN is to be sent to the initiator,
2581 	 * it is not required to increment the number
2582 	 * as the connection is terminating.
2583 	 */
2584 	resp->isp_flags |= IDM_PDU_SET_STATSN;
2585 	/*
2586 	 * Logout results in the immediate termination of all tasks except
2587 	 * if the logout reason is ISCSI_LOGOUT_REASON_RECOVERY.  The
2588 	 * connection state machine will drive this task cleanup automatically
2589 	 * so we don't need to handle that here.
2590 	 */
2591 	logout_rsp = (iscsi_logout_rsp_hdr_t *)resp->isp_hdr;
2592 	bzero(logout_rsp, sizeof (*logout_rsp));
2593 	logout_rsp->opcode = ISCSI_OP_LOGOUT_RSP;
2594 	logout_rsp->flags = ISCSI_FLAG_FINAL;
2595 	logout_rsp->itt = logout_req->itt;
2596 	if ((logout_req->flags & ISCSI_FLAG_LOGOUT_REASON_MASK) >
2597 	    ISCSI_LOGOUT_REASON_RECOVERY) {
2598 		logout_rsp->response = ISCSI_LOGOUT_RECOVERY_UNSUPPORTED;
2599 	} else {
2600 		logout_rsp->response = ISCSI_LOGOUT_SUCCESS;
2601 	}
2602 
2603 	iscsit_pdu_tx(resp);
2604 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2605 }
2606 
2607 /*
2608  * Calculate the number of outstanding commands we can process
2609  */
2610 int
2611 iscsit_cmd_window()
2612 {
2613 	/*
2614 	 * Instead of using a pre-defined constant for the command window,
2615 	 * it should be made confiurable and dynamic. With MC/S, sequence
2616 	 * numbers will be used up at a much faster rate than with SC/S.
2617 	 */
2618 	return	(ISCSIT_MAX_WINDOW);
2619 }
2620 
2621 /*
2622  * Set local registers based on incoming PDU
2623  */
2624 void
2625 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2626 {
2627 	iscsit_sess_t *ist;
2628 	iscsi_scsi_cmd_hdr_t *req;
2629 
2630 	ist = ict->ict_sess;
2631 
2632 	req = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2633 	if (req->opcode & ISCSI_OP_IMMEDIATE) {
2634 		/* no cmdsn increment for immediate PDUs */
2635 		return;
2636 	}
2637 
2638 	/* Ensure that the ExpCmdSN advances in an orderly manner */
2639 	mutex_enter(&ist->ist_sn_mutex);
2640 	ist->ist_expcmdsn = ntohl(req->cmdsn) + 1;
2641 	ist->ist_maxcmdsn = ntohl(req->cmdsn) + iscsit_cmd_window();
2642 	mutex_exit(&ist->ist_sn_mutex);
2643 }
2644 
2645 /*
2646  * Wrapper funtion, calls iscsi_calc_rspsn and idm_pdu_tx
2647  */
2648 void
2649 iscsit_pdu_tx(idm_pdu_t *pdu)
2650 {
2651 	iscsit_conn_t *ict = pdu->isp_ic->ic_handle;
2652 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2653 	iscsit_sess_t *ist = ict->ict_sess;
2654 
2655 	/*
2656 	 * The command sequence numbers are session-wide and must stay
2657 	 * consistent across the transfer, so protect the cmdsn with a
2658 	 * mutex lock on the session. The status sequence number will
2659 	 * be updated just before the transport layer transmits the PDU.
2660 	 */
2661 
2662 	mutex_enter(&ict->ict_sess->ist_sn_mutex);
2663 	/* Set ExpCmdSN and MaxCmdSN */
2664 	rsp->maxcmdsn = htonl(ist->ist_maxcmdsn);
2665 	rsp->expcmdsn = htonl(ist->ist_expcmdsn);
2666 	idm_pdu_tx(pdu);
2667 	mutex_exit(&ict->ict_sess->ist_sn_mutex);
2668 }
2669 
2670 /*
2671  * Internal functions
2672  */
2673 
2674 void
2675 iscsit_send_async_event(iscsit_conn_t *ict, uint8_t event)
2676 {
2677 	idm_pdu_t		*abt;
2678 	iscsi_async_evt_hdr_t	*async_abt;
2679 
2680 	/*
2681 	 * Get a PDU to build the abort request.
2682 	 */
2683 	abt = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2684 	if (abt == NULL) {
2685 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2686 		return;
2687 	}
2688 
2689 	/*
2690 	 * A asynchronous message is sent by the target to request a logout.
2691 	 * The StatSN for the connection is advanced after the PDU is sent
2692 	 * to allow for initiator and target state synchronization.
2693 	 */
2694 	idm_pdu_init(abt, ict->ict_ic, NULL, NULL);
2695 	abt->isp_datalen = 0;
2696 	abt->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2697 
2698 	async_abt = (iscsi_async_evt_hdr_t *)abt->isp_hdr;
2699 	bzero(async_abt, sizeof (*async_abt));
2700 	async_abt->opcode = ISCSI_OP_ASYNC_EVENT;
2701 	async_abt->async_event = event;
2702 	async_abt->flags = ISCSI_FLAG_FINAL;
2703 	async_abt->rsvd4[0] = 0xff;
2704 	async_abt->rsvd4[1] = 0xff;
2705 	async_abt->rsvd4[2] = 0xff;
2706 	async_abt->rsvd4[3] = 0xff;
2707 
2708 	switch (event) {
2709 	case ISCSI_ASYNC_EVENT_REQUEST_LOGOUT:
2710 		async_abt->param3 = htons(IDM_LOGOUT_SECONDS);
2711 		break;
2712 	case ISCSI_ASYNC_EVENT_SCSI_EVENT:
2713 	case ISCSI_ASYNC_EVENT_DROPPING_CONNECTION:
2714 	case ISCSI_ASYNC_EVENT_DROPPING_ALL_CONNECTIONS:
2715 	case ISCSI_ASYNC_EVENT_PARAM_NEGOTIATION:
2716 	default:
2717 		ASSERT(0);
2718 	}
2719 
2720 	iscsit_pdu_tx(abt);
2721 }
2722 
2723 void
2724 iscsit_send_reject(iscsit_conn_t *ict, idm_pdu_t *rejected_pdu, uint8_t reason)
2725 {
2726 	idm_pdu_t		*reject_pdu;
2727 	iscsi_reject_rsp_hdr_t	*reject;
2728 
2729 	/*
2730 	 * Get a PDU to build the abort request.
2731 	 */
2732 	reject_pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t),
2733 	    rejected_pdu->isp_hdrlen);
2734 	if (reject_pdu == NULL) {
2735 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2736 		return;
2737 	}
2738 	idm_pdu_init(reject_pdu, ict->ict_ic, NULL, NULL);
2739 	/* StatSN is advanced after a Reject PDU */
2740 	reject_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2741 	reject_pdu->isp_datalen = rejected_pdu->isp_hdrlen;
2742 	bcopy(rejected_pdu->isp_hdr, reject_pdu->isp_data,
2743 	    rejected_pdu->isp_hdrlen);
2744 
2745 	reject = (iscsi_reject_rsp_hdr_t *)reject_pdu->isp_hdr;
2746 	bzero(reject, sizeof (*reject));
2747 	reject->opcode = ISCSI_OP_REJECT_MSG;
2748 	reject->reason = reason;
2749 	reject->flags = ISCSI_FLAG_FINAL;
2750 	hton24(reject->dlength, rejected_pdu->isp_hdrlen);
2751 	reject->must_be_ff[0] = 0xff;
2752 	reject->must_be_ff[1] = 0xff;
2753 	reject->must_be_ff[2] = 0xff;
2754 	reject->must_be_ff[3] = 0xff;
2755 
2756 	iscsit_pdu_tx(reject_pdu);
2757 }
2758 
2759 
2760 static iscsit_task_t *
2761 iscsit_task_alloc(iscsit_conn_t *ict)
2762 {
2763 	iscsit_task_t *itask;
2764 	iscsit_buf_t *immed_ibuf;
2765 
2766 	/*
2767 	 * Possible items to pre-alloc if we cache iscsit_task_t's:
2768 	 *
2769 	 * Status PDU w/ sense buffer
2770 	 * stmf_data_buf_t for immediate data
2771 	 */
2772 	itask = kmem_alloc(sizeof (iscsit_task_t) + sizeof (iscsit_buf_t) +
2773 	    sizeof (stmf_data_buf_t), KM_NOSLEEP);
2774 	if (itask != NULL) {
2775 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2776 		itask->it_aborted = itask->it_stmf_abort =
2777 		    itask->it_tm_task = 0;
2778 
2779 		immed_ibuf = (iscsit_buf_t *)(itask + 1);
2780 		bzero(immed_ibuf, sizeof (*immed_ibuf));
2781 		immed_ibuf->ibuf_is_immed = B_TRUE;
2782 		immed_ibuf->ibuf_stmf_buf = (stmf_data_buf_t *)(immed_ibuf + 1);
2783 
2784 		bzero(immed_ibuf->ibuf_stmf_buf, sizeof (stmf_data_buf_t));
2785 		immed_ibuf->ibuf_stmf_buf->db_port_private = immed_ibuf;
2786 		immed_ibuf->ibuf_stmf_buf->db_sglist_length = 1;
2787 		immed_ibuf->ibuf_stmf_buf->db_flags = DB_DIRECTION_FROM_RPORT |
2788 		    DB_DONT_CACHE;
2789 		itask->it_immed_data = immed_ibuf;
2790 		itask->it_idm_task = idm_task_alloc(ict->ict_ic);
2791 		if (itask->it_idm_task != NULL) {
2792 			itask->it_idm_task->idt_private = itask;
2793 			itask->it_ict = ict;
2794 			itask->it_ttt = itask->it_idm_task->idt_tt;
2795 			return (itask);
2796 		} else {
2797 			kmem_free(itask, sizeof (iscsit_task_t) +
2798 			    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2799 		}
2800 	}
2801 
2802 	return (NULL);
2803 }
2804 
2805 static void
2806 iscsit_task_free(iscsit_task_t *itask)
2807 {
2808 	idm_task_free(itask->it_idm_task);
2809 	mutex_destroy(&itask->it_mutex);
2810 	kmem_free(itask, sizeof (iscsit_task_t) +
2811 	    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2812 }
2813 
2814 static iscsit_task_t *
2815 iscsit_tm_task_alloc(iscsit_conn_t *ict)
2816 {
2817 	iscsit_task_t *itask;
2818 
2819 	itask = kmem_zalloc(sizeof (iscsit_task_t), KM_NOSLEEP);
2820 	if (itask != NULL) {
2821 		idm_conn_hold(ict->ict_ic);
2822 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2823 		itask->it_aborted = itask->it_stmf_abort =
2824 		    itask->it_tm_responded = 0;
2825 		itask->it_tm_pdu = NULL;
2826 		itask->it_tm_task = 1;
2827 		itask->it_ict = ict;
2828 	}
2829 
2830 	return (itask);
2831 }
2832 
2833 static void
2834 iscsit_tm_task_free(iscsit_task_t *itask)
2835 {
2836 	/*
2837 	 * If we responded then the call to idm_pdu_complete will free the
2838 	 * PDU.  Otherwise we got aborted before the TM function could
2839 	 * complete and we need to free the PDU explicitly.
2840 	 */
2841 	if (itask->it_tm_pdu != NULL && !itask->it_tm_responded)
2842 		idm_pdu_free(itask->it_tm_pdu);
2843 	idm_conn_rele(itask->it_ict->ict_ic);
2844 	mutex_destroy(&itask->it_mutex);
2845 	kmem_free(itask, sizeof (iscsit_task_t));
2846 }
2847 
2848 static idm_status_t
2849 iscsit_task_start(iscsit_task_t *itask)
2850 {
2851 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2852 	avl_index_t		where;
2853 
2854 	/*
2855 	 * Sanity check the ITT and ensure that this task does not already
2856 	 * exist.  If not then add the task to the session task list.
2857 	 */
2858 	mutex_enter(&ist->ist_mutex);
2859 	mutex_enter(&itask->it_mutex);
2860 	itask->it_active = 1;
2861 	if (avl_find(&ist->ist_task_list, itask, &where) == NULL) {
2862 		/* New task, add to AVL */
2863 		avl_insert(&ist->ist_task_list, itask, where);
2864 		mutex_exit(&itask->it_mutex);
2865 		mutex_exit(&ist->ist_mutex);
2866 		return (IDM_STATUS_SUCCESS);
2867 	}
2868 	mutex_exit(&itask->it_mutex);
2869 	mutex_exit(&ist->ist_mutex);
2870 
2871 	return (IDM_STATUS_REJECT);
2872 }
2873 
2874 static void
2875 iscsit_task_done(iscsit_task_t *itask)
2876 {
2877 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2878 
2879 	mutex_enter(&ist->ist_mutex);
2880 	mutex_enter(&itask->it_mutex);
2881 	if (itask->it_active) {
2882 		avl_remove(&ist->ist_task_list, itask);
2883 		itask->it_active = 0;
2884 	}
2885 	mutex_exit(&itask->it_mutex);
2886 	mutex_exit(&ist->ist_mutex);
2887 }
2888 
2889 /*
2890  * iscsit status PDU cache
2891  */
2892 
2893 /*ARGSUSED*/
2894 static int
2895 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags)
2896 {
2897 	idm_pdu_t *pdu = pdu_void;
2898 	iscsi_scsi_rsp_hdr_t *rsp;
2899 
2900 	bzero(pdu, sizeof (idm_pdu_t));
2901 	pdu->isp_callback = iscsit_send_good_status_done;
2902 	pdu->isp_magic = IDM_PDU_MAGIC;
2903 	pdu->isp_hdr = (iscsi_hdr_t *)(pdu + 1); /* Ptr arithmetic */
2904 	pdu->isp_hdrlen = sizeof (iscsi_hdr_t);
2905 
2906 	/* Setup status response */
2907 	rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2908 	bzero(rsp, sizeof (*rsp));
2909 	rsp->opcode = ISCSI_OP_SCSI_RSP;
2910 	rsp->flags = ISCSI_FLAG_FINAL;
2911 	rsp->response = ISCSI_STATUS_CMD_COMPLETED;
2912 
2913 	return (0);
2914 }
2915 
2916 /*
2917  * iscsit private data handler
2918  */
2919 
2920 /*ARGSUSED*/
2921 static void
2922 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags)
2923 {
2924 	it_config_t		*cfg;
2925 	nvlist_t		*nvl;
2926 	iscsit_service_enabled_t	old_state;
2927 
2928 	if ((cmd != STMF_PROVIDER_DATA_UPDATED) || (arg == NULL)) {
2929 		return;
2930 	}
2931 
2932 	nvl = (nvlist_t *)arg;
2933 
2934 	/* Translate nvlist */
2935 	if (it_nv_to_config(nvl, &cfg) != 0) {
2936 		cmn_err(CE_WARN, "Configuration is invalid");
2937 		return;
2938 	}
2939 
2940 	/* Check that no iSCSI ioctl is currently running */
2941 	mutex_enter(&iscsit_global.global_state_mutex);
2942 	old_state = iscsit_global.global_svc_state;
2943 	switch (iscsit_global.global_svc_state) {
2944 	case ISE_ENABLED:
2945 	case ISE_DISABLED:
2946 		iscsit_global.global_svc_state = ISE_BUSY;
2947 		break;
2948 	case ISE_ENABLING:
2949 		/*
2950 		 * It is OK for the iscsit_pp_cb to be called from inside of
2951 		 * an iSCSI ioctl only if we are currently executing inside
2952 		 * of stmf_register_port_provider.
2953 		 */
2954 		ASSERT((flags & STMF_PCB_PREG_COMPLETE) != 0);
2955 		break;
2956 	default:
2957 		cmn_err(CE_WARN, "iscsit_pp_cb called when global_svc_state"
2958 		    " is not ENABLED(0x%x) -- ignoring",
2959 		    iscsit_global.global_svc_state);
2960 		mutex_exit(&iscsit_global.global_state_mutex);
2961 		it_config_free_cmn(cfg);
2962 		return;
2963 	}
2964 	mutex_exit(&iscsit_global.global_state_mutex);
2965 
2966 	/* Update config */
2967 	(void) iscsit_config_merge(cfg);
2968 
2969 	it_config_free_cmn(cfg);
2970 
2971 	/* Restore old iSCSI driver global state */
2972 	mutex_enter(&iscsit_global.global_state_mutex);
2973 	ASSERT(iscsit_global.global_svc_state == ISE_BUSY ||
2974 	    iscsit_global.global_svc_state == ISE_ENABLING);
2975 	iscsit_global.global_svc_state = old_state;
2976 	mutex_exit(&iscsit_global.global_state_mutex);
2977 }
2978 
2979 
2980 static it_cfg_status_t
2981 iscsit_config_merge(it_config_t *in_cfg)
2982 {
2983 	it_cfg_status_t	status;
2984 	it_config_t	*cfg;
2985 	it_config_t	tmp_cfg;
2986 	list_t		tpg_del_list;
2987 
2988 	if (in_cfg) {
2989 		cfg = in_cfg;
2990 	} else {
2991 		/* Make empty config */
2992 		bzero(&tmp_cfg, sizeof (tmp_cfg));
2993 		cfg = &tmp_cfg;
2994 	}
2995 
2996 	list_create(&tpg_del_list,  sizeof (iscsit_tpg_t),
2997 	    offsetof(iscsit_tpg_t, tpg_delete_ln));
2998 
2999 	/*
3000 	 * Update targets, initiator contexts, target portal groups,
3001 	 * and iSNS client
3002 	 */
3003 	ISCSIT_GLOBAL_LOCK(RW_WRITER);
3004 	if (((status = iscsit_config_merge_tpg(cfg, &tpg_del_list))
3005 	    != 0) ||
3006 	    ((status = iscsit_config_merge_tgt(cfg)) != 0) ||
3007 	    ((status = iscsit_config_merge_ini(cfg)) != 0) ||
3008 	    ((status = isnst_config_merge(cfg)) != 0)) {
3009 		ISCSIT_GLOBAL_UNLOCK();
3010 		return (status);
3011 	}
3012 
3013 	/* Update other global config parameters */
3014 	if (iscsit_global.global_props) {
3015 		nvlist_free(iscsit_global.global_props);
3016 		iscsit_global.global_props = NULL;
3017 	}
3018 	if (in_cfg) {
3019 		(void) nvlist_dup(cfg->config_global_properties,
3020 		    &iscsit_global.global_props, KM_SLEEP);
3021 	}
3022 	ISCSIT_GLOBAL_UNLOCK();
3023 
3024 	iscsit_config_destroy_tpgs(&tpg_del_list);
3025 
3026 	list_destroy(&tpg_del_list);
3027 
3028 	return (ITCFG_SUCCESS);
3029 }
3030 
3031 /*
3032  * iscsit_sna_lt[e]
3033  *
3034  * Compare serial numbers using serial number arithmetic as defined in
3035  * RFC 1982.
3036  *
3037  * NOTE: This code is duplicated in the isns server. It ought to be common.
3038  */
3039 
3040 static int
3041 iscsit_sna_lt(uint32_t sn1, uint32_t sn2)
3042 {
3043 	return ((sn1 != sn2) &&
3044 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3045 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3046 }
3047 
3048 static int
3049 iscsit_sna_lte(uint32_t sn1, uint32_t sn2)
3050 {
3051 	return ((sn1 == sn2) ||
3052 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3053 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3054 }
3055 
3056 
3057 static boolean_t
3058 iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn)
3059 {
3060 	iscsit_sess_t	*ist = ict->ict_sess;
3061 	int		rval = B_TRUE;
3062 
3063 	ist = ict->ict_sess;
3064 
3065 	mutex_enter(&ist->ist_sn_mutex);
3066 
3067 	/*
3068 	 * If cmdsn is less than ist_expcmdsn - iscsit_cmd_window() or
3069 	 * greater than ist_expcmdsn, it's not in the window.
3070 	 */
3071 
3072 	if (iscsit_sna_lt(cmdsn, (ist->ist_expcmdsn - iscsit_cmd_window())) ||
3073 	    !iscsit_sna_lte(cmdsn, ist->ist_expcmdsn)) {
3074 		rval = B_FALSE;
3075 	}
3076 
3077 	mutex_exit(&ist->ist_sn_mutex);
3078 
3079 	return (rval);
3080 }
3081 
3082 /*
3083  * iscsit_check_cmdsn_and_queue
3084  *
3085  * Independent of the order in which the iSCSI target receives non-immediate
3086  * command PDU across the entire session and any multiple connections within
3087  * the session, the target must deliver the commands to the SCSI layer in
3088  * CmdSN order. So out-of-order non-immediate commands are queued up on a
3089  * session-wide wait queue. Duplicate commands are ignored.
3090  *
3091  */
3092 static int
3093 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu)
3094 {
3095 	idm_conn_t		*ic = rx_pdu->isp_ic;
3096 	iscsit_conn_t		*ict = ic->ic_handle;
3097 	iscsit_sess_t		*ist = ict->ict_sess;
3098 	iscsi_scsi_cmd_hdr_t	*hdr = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
3099 
3100 	mutex_enter(&ist->ist_sn_mutex);
3101 	if (hdr->opcode & ISCSI_OP_IMMEDIATE) {
3102 		/* do not queue, handle it immediately */
3103 		DTRACE_PROBE2(immediate__cmd, iscsit_sess_t *, ist,
3104 		    idm_pdu_t *, rx_pdu);
3105 		mutex_exit(&ist->ist_sn_mutex);
3106 		return (ISCSIT_CMDSN_EQ_EXPCMDSN);
3107 	}
3108 	if (iscsit_sna_lt(ist->ist_expcmdsn, ntohl(hdr->cmdsn))) {
3109 		/*
3110 		 * Out-of-order commands (cmdSN higher than ExpCmdSN)
3111 		 * are staged on a fixed-size circular buffer until
3112 		 * the missing command is delivered to the SCSI layer.
3113 		 * Irrespective of the order of insertion into the
3114 		 * staging queue, the commands are processed out of the
3115 		 * queue in cmdSN order only.
3116 		 */
3117 		rx_pdu->isp_queue_time = ddi_get_time();
3118 		iscsit_add_pdu_to_queue(ist, rx_pdu);
3119 		mutex_exit(&ist->ist_sn_mutex);
3120 		return (ISCSIT_CMDSN_GT_EXPCMDSN);
3121 	} else if (iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_expcmdsn)) {
3122 		DTRACE_PROBE3(cmdsn__lt__expcmdsn, iscsit_sess_t *, ist,
3123 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
3124 		mutex_exit(&ist->ist_sn_mutex);
3125 		return (ISCSIT_CMDSN_LT_EXPCMDSN);
3126 	} else {
3127 		mutex_exit(&ist->ist_sn_mutex);
3128 		return (ISCSIT_CMDSN_EQ_EXPCMDSN);
3129 	}
3130 }
3131 
3132 /*
3133  * iscsit_add_pdu_to_queue() adds PDUs into the array indexed by
3134  * their cmdsn value. The length of the array is kept above the
3135  * maximum window size. The window keeps the cmdsn within a range
3136  * such that there are no collisons. e.g. the assumption is that
3137  * the windowing checks make it impossible to receive PDUs that
3138  * index into the same location in the array.
3139  */
3140 static void
3141 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu)
3142 {
3143 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3144 	iscsit_conn_t	*ict 	= rx_pdu->isp_ic->ic_handle;
3145 	uint32_t	cmdsn	=
3146 	    ((iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr)->cmdsn;
3147 	uint32_t	index;
3148 
3149 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3150 	/*
3151 	 * If the connection is being torn down, then
3152 	 * don't add the PDU to the staging queue
3153 	 */
3154 	mutex_enter(&ict->ict_mutex);
3155 	if (ict->ict_lost) {
3156 		mutex_exit(&ict->ict_mutex);
3157 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3158 		return;
3159 	}
3160 	iscsit_conn_dispatch_hold(ict);
3161 	mutex_exit(&ict->ict_mutex);
3162 
3163 	index = ntohl(cmdsn) % ISCSIT_RXPDU_QUEUE_LEN;
3164 	ASSERT(cbuf->cb_buffer[index] == NULL);
3165 	cbuf->cb_buffer[index] = rx_pdu;
3166 	cbuf->cb_num_elems++;
3167 }
3168 
3169 static idm_pdu_t *
3170 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn)
3171 {
3172 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3173 	idm_pdu_t	*pdu	= NULL;
3174 	uint32_t	index;
3175 
3176 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3177 	index = cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3178 	if ((pdu = cbuf->cb_buffer[index]) != NULL) {
3179 		ASSERT(cmdsn ==
3180 		    ntohl(((iscsi_scsi_cmd_hdr_t *)pdu->isp_hdr)->cmdsn));
3181 		cbuf->cb_buffer[index] = NULL;
3182 		cbuf->cb_num_elems--;
3183 		return (pdu);
3184 	}
3185 	return (NULL);
3186 }
3187 
3188 /*
3189  * iscsit_process_pdu_in_queue() finds the next pdu in sequence
3190  * and posts it to the SCSI layer
3191  */
3192 static void
3193 iscsit_process_pdu_in_queue(iscsit_sess_t *ist)
3194 {
3195 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3196 	idm_pdu_t	*pdu = NULL;
3197 	uint32_t	expcmdsn;
3198 
3199 	for (;;) {
3200 		mutex_enter(&ist->ist_sn_mutex);
3201 		if (cbuf->cb_num_elems == 0) {
3202 			mutex_exit(&ist->ist_sn_mutex);
3203 			break;
3204 		}
3205 		expcmdsn = ist->ist_expcmdsn;
3206 		if ((pdu = iscsit_remove_pdu_from_queue(ist, expcmdsn))
3207 		    == NULL) {
3208 			mutex_exit(&ist->ist_sn_mutex);
3209 			break;
3210 		}
3211 		mutex_exit(&ist->ist_sn_mutex);
3212 		iscsit_post_staged_pdu(pdu);
3213 	}
3214 }
3215 
3216 static void
3217 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu)
3218 {
3219 	iscsit_conn_t	*ict	= rx_pdu->isp_ic->ic_handle;
3220 
3221 	/* Post the PDU to the SCSI layer */
3222 	switch (IDM_PDU_OPCODE(rx_pdu)) {
3223 	case ISCSI_OP_NOOP_OUT:
3224 		iscsit_set_cmdsn(ict, rx_pdu);
3225 		iscsit_pdu_op_noop(ict, rx_pdu);
3226 		break;
3227 	case ISCSI_OP_TEXT_CMD:
3228 		iscsit_set_cmdsn(ict, rx_pdu);
3229 		iscsit_pdu_op_text_cmd(ict, rx_pdu);
3230 		break;
3231 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
3232 		iscsit_set_cmdsn(ict, rx_pdu);
3233 		iscsit_op_scsi_task_mgmt(ict, rx_pdu);
3234 		break;
3235 	case ISCSI_OP_SCSI_CMD:
3236 		/* cmdSN will be incremented after creating itask */
3237 		iscsit_post_scsi_cmd(rx_pdu->isp_ic, rx_pdu);
3238 		break;
3239 	case ISCSI_OP_LOGOUT_CMD:
3240 		iscsit_set_cmdsn(ict, rx_pdu);
3241 		iscsit_pdu_op_logout_cmd(ict, rx_pdu);
3242 		break;
3243 	default:
3244 		/* No other PDUs should be placed on the queue */
3245 		ASSERT(0);
3246 	}
3247 	iscsit_conn_dispatch_rele(ict); /* release hold on the conn */
3248 }
3249 
3250 /* ARGSUSED */
3251 void
3252 iscsit_rxpdu_queue_monitor_start(void)
3253 {
3254 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3255 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3256 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3257 		return;
3258 	}
3259 	iscsit_rxpdu_queue_monitor_thr_id =
3260 	    thread_create(NULL, 0, iscsit_rxpdu_queue_monitor, NULL,
3261 	    0, &p0, TS_RUN, minclsyspri);
3262 	while (!iscsit_rxpdu_queue_monitor_thr_running) {
3263 		cv_wait(&iscsit_rxpdu_queue_monitor_cv,
3264 		    &iscsit_rxpdu_queue_monitor_mutex);
3265 	}
3266 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3267 
3268 }
3269 
3270 /* ARGSUSED */
3271 void
3272 iscsit_rxpdu_queue_monitor_stop(void)
3273 {
3274 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3275 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3276 		iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
3277 		cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3278 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3279 
3280 		thread_join(iscsit_rxpdu_queue_monitor_thr_did);
3281 		return;
3282 	}
3283 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3284 }
3285 
3286 /*
3287  * A separate thread is used to scan the staging queue on all the
3288  * sessions, If a delayed PDU does not arrive within a timeout, the
3289  * target will advance to the staged PDU that is next in sequence
3290  * and exceeded the threshold wait time. It is up to the initiator
3291  * to note that the target has not acknowledged a particular cmdsn
3292  * and take appropriate action.
3293  */
3294 /* ARGSUSED */
3295 static void
3296 iscsit_rxpdu_queue_monitor(void *arg)
3297 {
3298 	iscsit_tgt_t	*tgt;
3299 	iscsit_sess_t	*ist;
3300 
3301 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3302 	iscsit_rxpdu_queue_monitor_thr_did = curthread->t_did;
3303 	iscsit_rxpdu_queue_monitor_thr_running = B_TRUE;
3304 	cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3305 
3306 	while (iscsit_rxpdu_queue_monitor_thr_running) {
3307 		ISCSIT_GLOBAL_LOCK(RW_READER);
3308 		for (tgt = avl_first(&iscsit_global.global_target_list);
3309 		    tgt != NULL;
3310 		    tgt = AVL_NEXT(&iscsit_global.global_target_list, tgt)) {
3311 			mutex_enter(&tgt->target_mutex);
3312 			for (ist = avl_first(&tgt->target_sess_list);
3313 			    ist != NULL;
3314 			    ist = AVL_NEXT(&tgt->target_sess_list, ist)) {
3315 
3316 				iscsit_rxpdu_queue_monitor_session(ist);
3317 			}
3318 			mutex_exit(&tgt->target_mutex);
3319 		}
3320 		ISCSIT_GLOBAL_UNLOCK();
3321 		if (iscsit_rxpdu_queue_monitor_thr_running == B_FALSE) {
3322 			break;
3323 		}
3324 		(void) cv_reltimedwait(&iscsit_rxpdu_queue_monitor_cv,
3325 		    &iscsit_rxpdu_queue_monitor_mutex,
3326 		    ISCSIT_RXPDU_QUEUE_MONITOR_INTERVAL * drv_usectohz(1000000),
3327 		    TR_CLOCK_TICK);
3328 	}
3329 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3330 	thread_exit();
3331 }
3332 
3333 static void
3334 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist)
3335 {
3336 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3337 	idm_pdu_t	*next_pdu = NULL;
3338 	uint32_t	index, next_cmdsn, i;
3339 
3340 	/*
3341 	 * Assume that all PDUs in the staging queue have a cmdsn >= expcmdsn.
3342 	 * Starting with the expcmdsn, iterate over the staged PDUs to find
3343 	 * the next PDU with a wait time greater than the threshold. If found
3344 	 * advance the staged PDU to the SCSI layer, skipping over the missing
3345 	 * PDU(s) to get past the hole in the command sequence. It is up to
3346 	 * the initiator to note that the target has not acknowledged a cmdsn
3347 	 * and take appropriate action.
3348 	 *
3349 	 * Since the PDU(s) arrive in any random order, it is possible that
3350 	 * that the actual wait time for a particular PDU is much longer than
3351 	 * the defined threshold. e.g. Consider a case where commands are sent
3352 	 * over 4 different connections, and cmdsn = 1004 arrives first, then
3353 	 * 1003, and 1002 and 1001 are lost due to a connection failure.
3354 	 * So now 1003 is waiting for 1002 to be delivered, and although the
3355 	 * wait time of 1004 > wait time of 1003, only 1003 will be considered
3356 	 * by the monitor thread. 1004 will be automatically processed by
3357 	 * iscsit_process_pdu_in_queue() once the scan is complete and the
3358 	 * expcmdsn becomes current.
3359 	 */
3360 	mutex_enter(&ist->ist_sn_mutex);
3361 	cbuf = ist->ist_rxpdu_queue;
3362 	if (cbuf->cb_num_elems == 0) {
3363 		mutex_exit(&ist->ist_sn_mutex);
3364 		return;
3365 	}
3366 	for (next_pdu = NULL, i = 0; ; i++) {
3367 		next_cmdsn = ist->ist_expcmdsn + i; /* start at expcmdsn */
3368 		index = next_cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3369 		if ((next_pdu = cbuf->cb_buffer[index]) != NULL) {
3370 			/*
3371 			 * If the PDU wait time has not exceeded threshold
3372 			 * stop scanning the staging queue until the timer
3373 			 * fires again
3374 			 */
3375 			if ((ddi_get_time() - next_pdu->isp_queue_time)
3376 			    < rxpdu_queue_threshold) {
3377 				mutex_exit(&ist->ist_sn_mutex);
3378 				return;
3379 			}
3380 			/*
3381 			 * Remove the next PDU from the queue and post it
3382 			 * to the SCSI layer, skipping over the missing
3383 			 * PDU. Stop scanning the staging queue until
3384 			 * the monitor timer fires again
3385 			 */
3386 			(void) iscsit_remove_pdu_from_queue(ist, next_cmdsn);
3387 			mutex_exit(&ist->ist_sn_mutex);
3388 			DTRACE_PROBE3(advanced__to__blocked__cmdsn,
3389 			    iscsit_sess_t *, ist, idm_pdu_t *, next_pdu,
3390 			    uint32_t, next_cmdsn);
3391 			iscsit_post_staged_pdu(next_pdu);
3392 			/* Deliver any subsequent PDUs immediately */
3393 			iscsit_process_pdu_in_queue(ist);
3394 			return;
3395 		}
3396 		/*
3397 		 * Skipping over i PDUs, e.g. a case where commands 1001 and
3398 		 * 1002 are lost in the network, skip over both and post 1003
3399 		 * expcmdsn then becomes 1004 at the end of the scan.
3400 		 */
3401 		DTRACE_PROBE2(skipping__over__cmdsn, iscsit_sess_t *, ist,
3402 		    uint32_t, next_cmdsn);
3403 	}
3404 	/*
3405 	 * following the assumption, staged cmdsn >= expcmdsn, this statement
3406 	 * is never reached.
3407 	 */
3408 }
3409