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