xref: /illumos-gate/usr/src/uts/common/io/comstar/port/iscsit/iscsit.c (revision 5626beece2e5dedec7197ecf325cfaa1854a6c2e)
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  * Copyright 2014, 2015 Nexenta Systems, Inc. All rights reserved.
25  */
26 
27 #include <sys/cpuvar.h>
28 #include <sys/types.h>
29 #include <sys/conf.h>
30 #include <sys/stat.h>
31 #include <sys/file.h>
32 #include <sys/ddi.h>
33 #include <sys/sunddi.h>
34 #include <sys/modctl.h>
35 #include <sys/sysmacros.h>
36 #include <sys/socket.h>
37 #include <sys/strsubr.h>
38 #include <sys/nvpair.h>
39 
40 #include <sys/stmf.h>
41 #include <sys/stmf_ioctl.h>
42 #include <sys/portif.h>
43 #include <sys/idm/idm.h>
44 #include <sys/idm/idm_conn_sm.h>
45 
46 #include "iscsit_isns.h"
47 #include "iscsit.h"
48 
49 #define	ISCSIT_VERSION		BUILD_DATE "-1.18dev"
50 #define	ISCSIT_NAME_VERSION	"COMSTAR ISCSIT v" ISCSIT_VERSION
51 
52 /*
53  * DDI entry points.
54  */
55 static int iscsit_drv_attach(dev_info_t *, ddi_attach_cmd_t);
56 static int iscsit_drv_detach(dev_info_t *, ddi_detach_cmd_t);
57 static int iscsit_drv_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
58 static int iscsit_drv_open(dev_t *, int, int, cred_t *);
59 static int iscsit_drv_close(dev_t, int, int, cred_t *);
60 static boolean_t iscsit_drv_busy(void);
61 static int iscsit_drv_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
62 
63 extern struct mod_ops mod_miscops;
64 
65 
66 static struct cb_ops iscsit_cb_ops = {
67 	iscsit_drv_open,	/* cb_open */
68 	iscsit_drv_close,	/* cb_close */
69 	nodev,			/* cb_strategy */
70 	nodev,			/* cb_print */
71 	nodev,			/* cb_dump */
72 	nodev,			/* cb_read */
73 	nodev,			/* cb_write */
74 	iscsit_drv_ioctl,	/* cb_ioctl */
75 	nodev,			/* cb_devmap */
76 	nodev,			/* cb_mmap */
77 	nodev,			/* cb_segmap */
78 	nochpoll,		/* cb_chpoll */
79 	ddi_prop_op,		/* cb_prop_op */
80 	NULL,			/* cb_streamtab */
81 	D_MP,			/* cb_flag */
82 	CB_REV,			/* cb_rev */
83 	nodev,			/* cb_aread */
84 	nodev,			/* cb_awrite */
85 };
86 
87 static struct dev_ops iscsit_dev_ops = {
88 	DEVO_REV,		/* devo_rev */
89 	0,			/* devo_refcnt */
90 	iscsit_drv_getinfo,	/* devo_getinfo */
91 	nulldev,		/* devo_identify */
92 	nulldev,		/* devo_probe */
93 	iscsit_drv_attach,	/* devo_attach */
94 	iscsit_drv_detach,	/* devo_detach */
95 	nodev,			/* devo_reset */
96 	&iscsit_cb_ops,		/* devo_cb_ops */
97 	NULL,			/* devo_bus_ops */
98 	NULL,			/* devo_power */
99 	ddi_quiesce_not_needed,	/* quiesce */
100 };
101 
102 static struct modldrv modldrv = {
103 	&mod_driverops,
104 	"iSCSI Target",
105 	&iscsit_dev_ops,
106 };
107 
108 static struct modlinkage modlinkage = {
109 	MODREV_1,
110 	&modldrv,
111 	NULL,
112 };
113 
114 
115 iscsit_global_t iscsit_global;
116 
117 kmem_cache_t	*iscsit_status_pdu_cache;
118 
119 boolean_t	iscsit_sm_logging = B_FALSE;
120 
121 kmutex_t	login_sm_session_mutex;
122 
123 static idm_status_t iscsit_init(dev_info_t *dip);
124 static idm_status_t iscsit_enable_svc(iscsit_hostinfo_t *hostinfo);
125 static void iscsit_disable_svc(void);
126 
127 static int
128 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu);
129 
130 static void
131 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu);
132 
133 static idm_pdu_t *
134 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn);
135 
136 static void
137 iscsit_process_pdu_in_queue(iscsit_sess_t *ist);
138 
139 static void
140 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist);
141 
142 static void
143 iscsit_rxpdu_queue_monitor(void *arg);
144 
145 static void
146 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu);
147 
148 static void
149 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu);
150 
151 static void
152 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
153 
154 static void
155 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
156 
157 static void
158 iscsit_pdu_op_login_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
159 
160 void
161 iscsit_pdu_op_text_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
162 
163 static void
164 iscsit_pdu_op_logout_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
165 
166 int iscsit_cmd_window();
167 
168 static  int
169 iscsit_sna_lt(uint32_t sn1, uint32_t sn2);
170 
171 void
172 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
173 
174 static void
175 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu);
176 
177 static void
178 iscsit_deferred(void *rx_pdu_void);
179 
180 static idm_status_t
181 iscsit_conn_accept(idm_conn_t *ic);
182 
183 static idm_status_t
184 iscsit_ffp_enabled(idm_conn_t *ic);
185 
186 static idm_status_t
187 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class);
188 
189 static idm_status_t
190 iscsit_conn_lost(idm_conn_t *ic);
191 
192 static idm_status_t
193 iscsit_conn_destroy(idm_conn_t *ic);
194 
195 static stmf_data_buf_t *
196 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
197     uint32_t flags);
198 
199 static void
200 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf);
201 
202 static void
203 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status);
204 
205 static void
206 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status);
207 
208 static void
209 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status);
210 
211 static stmf_status_t
212 iscsit_idm_to_stmf(idm_status_t idmrc);
213 
214 static iscsit_task_t *
215 iscsit_task_alloc(iscsit_conn_t *ict);
216 
217 static void
218 iscsit_task_free(iscsit_task_t *itask);
219 
220 static iscsit_task_t *
221 iscsit_tm_task_alloc(iscsit_conn_t *ict);
222 
223 static void
224 iscsit_tm_task_free(iscsit_task_t *itask);
225 
226 static idm_status_t
227 iscsit_task_start(iscsit_task_t *itask);
228 
229 static void
230 iscsit_task_done(iscsit_task_t *itask);
231 
232 static int
233 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags);
234 
235 static void
236 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags);
237 
238 static it_cfg_status_t
239 iscsit_config_merge(it_config_t *cfg);
240 
241 static idm_status_t
242 iscsit_login_fail(idm_conn_t *ic);
243 
244 static boolean_t iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn);
245 static void iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
246     uint8_t response, uint8_t cmd_status);
247 static void iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu,
248     uint8_t tm_status);
249 
250 /*
251  * MC/S: Out-of-order commands are staged on a session-wide wait
252  * queue until a system-tunable threshold is reached. A separate
253  * thread is used to scan the staging queue on all the session,
254  * If a delayed PDU does not arrive within a timeout, the target
255  * will advance to the staged PDU that is next in sequence, skipping
256  * over the missing PDU(s) to go past a hole in the sequence.
257  */
258 volatile int rxpdu_queue_threshold = ISCSIT_RXPDU_QUEUE_THRESHOLD;
259 
260 static kmutex_t		iscsit_rxpdu_queue_monitor_mutex;
261 kthread_t		*iscsit_rxpdu_queue_monitor_thr_id;
262 static kt_did_t		iscsit_rxpdu_queue_monitor_thr_did;
263 static boolean_t	iscsit_rxpdu_queue_monitor_thr_running;
264 static kcondvar_t	iscsit_rxpdu_queue_monitor_cv;
265 
266 int
267 _init(void)
268 {
269 	int rc;
270 
271 	rw_init(&iscsit_global.global_rwlock, NULL, RW_DRIVER, NULL);
272 	mutex_init(&iscsit_global.global_state_mutex, NULL,
273 	    MUTEX_DRIVER, NULL);
274 	iscsit_global.global_svc_state = ISE_DETACHED;
275 
276 	mutex_init(&iscsit_rxpdu_queue_monitor_mutex, NULL,
277 	    MUTEX_DRIVER, NULL);
278 	mutex_init(&login_sm_session_mutex, NULL, MUTEX_DRIVER, NULL);
279 	iscsit_rxpdu_queue_monitor_thr_id = NULL;
280 	iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
281 	cv_init(&iscsit_rxpdu_queue_monitor_cv, NULL, CV_DEFAULT, NULL);
282 
283 	if ((rc = mod_install(&modlinkage)) != 0) {
284 		mutex_destroy(&iscsit_global.global_state_mutex);
285 		rw_destroy(&iscsit_global.global_rwlock);
286 		return (rc);
287 	}
288 
289 	return (rc);
290 }
291 
292 int
293 _info(struct modinfo *modinfop)
294 {
295 	return (mod_info(&modlinkage, modinfop));
296 }
297 
298 int
299 _fini(void)
300 {
301 	int rc;
302 
303 	rc = mod_remove(&modlinkage);
304 
305 	if (rc == 0) {
306 		mutex_destroy(&iscsit_rxpdu_queue_monitor_mutex);
307 		mutex_destroy(&login_sm_session_mutex);
308 		cv_destroy(&iscsit_rxpdu_queue_monitor_cv);
309 		mutex_destroy(&iscsit_global.global_state_mutex);
310 		rw_destroy(&iscsit_global.global_rwlock);
311 	}
312 
313 	return (rc);
314 }
315 
316 /*
317  * DDI entry points.
318  */
319 
320 /* ARGSUSED */
321 static int
322 iscsit_drv_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
323     void **result)
324 {
325 	ulong_t instance = getminor((dev_t)arg);
326 
327 	switch (cmd) {
328 	case DDI_INFO_DEVT2DEVINFO:
329 		*result = iscsit_global.global_dip;
330 		return (DDI_SUCCESS);
331 
332 	case DDI_INFO_DEVT2INSTANCE:
333 		*result = (void *)instance;
334 		return (DDI_SUCCESS);
335 
336 	default:
337 		break;
338 	}
339 
340 	return (DDI_FAILURE);
341 }
342 
343 static int
344 iscsit_drv_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
345 {
346 	if (cmd != DDI_ATTACH) {
347 		return (DDI_FAILURE);
348 	}
349 
350 	if (ddi_get_instance(dip) != 0) {
351 		/* we only allow instance 0 to attach */
352 		return (DDI_FAILURE);
353 	}
354 
355 	/* create the minor node */
356 	if (ddi_create_minor_node(dip, ISCSIT_MODNAME, S_IFCHR, 0,
357 	    DDI_PSEUDO, 0) != DDI_SUCCESS) {
358 		cmn_err(CE_WARN, "iscsit_drv_attach: "
359 		    "failed creating minor node");
360 		return (DDI_FAILURE);
361 	}
362 
363 	if (iscsit_init(dip) != IDM_STATUS_SUCCESS) {
364 		cmn_err(CE_WARN, "iscsit_drv_attach: "
365 		    "failed to initialize");
366 		ddi_remove_minor_node(dip, NULL);
367 		return (DDI_FAILURE);
368 	}
369 
370 	iscsit_global.global_svc_state = ISE_DISABLED;
371 	iscsit_global.global_dip = dip;
372 
373 	return (DDI_SUCCESS);
374 }
375 
376 /*ARGSUSED*/
377 static int
378 iscsit_drv_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
379 {
380 	if (cmd != DDI_DETACH)
381 		return (DDI_FAILURE);
382 
383 	/*
384 	 * drv_detach is called in a context that owns the
385 	 * device node for the /dev/pseudo device.  If this thread blocks
386 	 * for any resource, other threads that need the /dev/pseudo device
387 	 * may end up in a deadlock with this thread.Hence, we use a
388 	 * separate lock just for the structures that drv_detach needs
389 	 * to access.
390 	 */
391 	mutex_enter(&iscsit_global.global_state_mutex);
392 	if (iscsit_drv_busy()) {
393 		mutex_exit(&iscsit_global.global_state_mutex);
394 		return (EBUSY);
395 	}
396 
397 	iscsit_global.global_dip = NULL;
398 	ddi_remove_minor_node(dip, NULL);
399 
400 	ldi_ident_release(iscsit_global.global_li);
401 	iscsit_global.global_svc_state = ISE_DETACHED;
402 
403 	mutex_exit(&iscsit_global.global_state_mutex);
404 
405 	return (DDI_SUCCESS);
406 }
407 
408 /*ARGSUSED*/
409 static int
410 iscsit_drv_open(dev_t *devp, int flag, int otyp, cred_t *credp)
411 {
412 	return (0);
413 }
414 
415 /* ARGSUSED */
416 static int
417 iscsit_drv_close(dev_t dev, int flag, int otyp, cred_t *credp)
418 {
419 	return (0);
420 }
421 
422 static boolean_t
423 iscsit_drv_busy(void)
424 {
425 	ASSERT(MUTEX_HELD(&iscsit_global.global_state_mutex));
426 
427 	switch (iscsit_global.global_svc_state) {
428 	case ISE_DISABLED:
429 	case ISE_DETACHED:
430 		return (B_FALSE);
431 	default:
432 		return (B_TRUE);
433 	}
434 	/* NOTREACHED */
435 }
436 
437 /* ARGSUSED */
438 static int
439 iscsit_drv_ioctl(dev_t drv, int cmd, intptr_t argp, int flag, cred_t *cred,
440     int *retval)
441 {
442 	iscsit_ioc_set_config_t		setcfg;
443 	iscsit_ioc_set_config32_t	setcfg32;
444 	char				*cfg_pnvlist = NULL;
445 	nvlist_t			*cfg_nvlist = NULL;
446 	it_config_t			*cfg = NULL;
447 	idm_status_t			idmrc;
448 	int				rc = 0;
449 
450 	if (drv_priv(cred) != 0) {
451 		return (EPERM);
452 	}
453 
454 	mutex_enter(&iscsit_global.global_state_mutex);
455 
456 	/*
457 	 * Validate ioctl requests against global service state
458 	 */
459 	switch (iscsit_global.global_svc_state) {
460 	case ISE_ENABLED:
461 		if (cmd == ISCSIT_IOC_DISABLE_SVC) {
462 			iscsit_global.global_svc_state = ISE_DISABLING;
463 		} else if (cmd == ISCSIT_IOC_ENABLE_SVC) {
464 			/* Already enabled */
465 			mutex_exit(&iscsit_global.global_state_mutex);
466 			return (0);
467 		} else {
468 			iscsit_global.global_svc_state = ISE_BUSY;
469 		}
470 		break;
471 	case ISE_DISABLED:
472 		if (cmd == ISCSIT_IOC_ENABLE_SVC) {
473 			iscsit_global.global_svc_state = ISE_ENABLING;
474 		} else if (cmd == ISCSIT_IOC_DISABLE_SVC) {
475 			/* Already disabled */
476 			mutex_exit(&iscsit_global.global_state_mutex);
477 			return (0);
478 		} else {
479 			rc = EFAULT;
480 		}
481 		break;
482 	case ISE_BUSY:
483 	case ISE_ENABLING:
484 	case ISE_DISABLING:
485 		rc = EAGAIN;
486 		break;
487 	case ISE_DETACHED:
488 	default:
489 		rc = EFAULT;
490 		break;
491 	}
492 
493 	mutex_exit(&iscsit_global.global_state_mutex);
494 	if (rc != 0)
495 		return (rc);
496 
497 	/* Handle ioctl request (enable/disable have already been handled) */
498 	switch (cmd) {
499 	case ISCSIT_IOC_SET_CONFIG:
500 		/* Any errors must set state back to ISE_ENABLED */
501 		switch (ddi_model_convert_from(flag & FMODELS)) {
502 		case DDI_MODEL_ILP32:
503 			if (ddi_copyin((void *)argp, &setcfg32,
504 			    sizeof (iscsit_ioc_set_config32_t), flag) != 0) {
505 				rc = EFAULT;
506 				goto cleanup;
507 			}
508 
509 			setcfg.set_cfg_pnvlist =
510 			    (char *)((uintptr_t)setcfg32.set_cfg_pnvlist);
511 			setcfg.set_cfg_vers = setcfg32.set_cfg_vers;
512 			setcfg.set_cfg_pnvlist_len =
513 			    setcfg32.set_cfg_pnvlist_len;
514 			break;
515 		case DDI_MODEL_NONE:
516 			if (ddi_copyin((void *)argp, &setcfg,
517 			    sizeof (iscsit_ioc_set_config_t), flag) != 0) {
518 				rc = EFAULT;
519 				goto cleanup;
520 			}
521 			break;
522 		default:
523 			rc = EFAULT;
524 			goto cleanup;
525 		}
526 
527 		/* Check API version */
528 		if (setcfg.set_cfg_vers != ISCSIT_API_VERS0) {
529 			rc = EINVAL;
530 			goto cleanup;
531 		}
532 
533 		/* Config is in packed nvlist format so unpack it */
534 		cfg_pnvlist = kmem_alloc(setcfg.set_cfg_pnvlist_len,
535 		    KM_SLEEP);
536 		ASSERT(cfg_pnvlist != NULL);
537 
538 		if (ddi_copyin(setcfg.set_cfg_pnvlist, cfg_pnvlist,
539 		    setcfg.set_cfg_pnvlist_len, flag) != 0) {
540 			rc = EFAULT;
541 			goto cleanup;
542 		}
543 
544 		rc = nvlist_unpack(cfg_pnvlist, setcfg.set_cfg_pnvlist_len,
545 		    &cfg_nvlist, KM_SLEEP);
546 		if (rc != 0) {
547 			goto cleanup;
548 		}
549 
550 		/* Translate nvlist */
551 		rc = it_nv_to_config(cfg_nvlist, &cfg);
552 		if (rc != 0) {
553 			cmn_err(CE_WARN, "Configuration is invalid");
554 			goto cleanup;
555 		}
556 
557 		/* Update config */
558 		rc = iscsit_config_merge(cfg);
559 		/* FALLTHROUGH */
560 
561 cleanup:
562 		if (cfg)
563 			it_config_free_cmn(cfg);
564 		if (cfg_pnvlist)
565 			kmem_free(cfg_pnvlist, setcfg.set_cfg_pnvlist_len);
566 		nvlist_free(cfg_nvlist);
567 
568 		/*
569 		 * Now that the reconfig is complete set our state back to
570 		 * enabled.
571 		 */
572 		mutex_enter(&iscsit_global.global_state_mutex);
573 		iscsit_global.global_svc_state = ISE_ENABLED;
574 		mutex_exit(&iscsit_global.global_state_mutex);
575 		break;
576 	case ISCSIT_IOC_ENABLE_SVC: {
577 		iscsit_hostinfo_t hostinfo;
578 
579 		if (ddi_copyin((void *)argp, &hostinfo.length,
580 		    sizeof (hostinfo.length), flag) != 0) {
581 			mutex_enter(&iscsit_global.global_state_mutex);
582 			iscsit_global.global_svc_state = ISE_DISABLED;
583 			mutex_exit(&iscsit_global.global_state_mutex);
584 			return (EFAULT);
585 		}
586 
587 		if (hostinfo.length > sizeof (hostinfo.fqhn))
588 			hostinfo.length = sizeof (hostinfo.fqhn);
589 
590 		if (ddi_copyin((void *)((caddr_t)argp +
591 		    sizeof (hostinfo.length)), &hostinfo.fqhn,
592 		    hostinfo.length, flag) != 0) {
593 			mutex_enter(&iscsit_global.global_state_mutex);
594 			iscsit_global.global_svc_state = ISE_DISABLED;
595 			mutex_exit(&iscsit_global.global_state_mutex);
596 			return (EFAULT);
597 		}
598 
599 		idmrc = iscsit_enable_svc(&hostinfo);
600 		mutex_enter(&iscsit_global.global_state_mutex);
601 		if (idmrc == IDM_STATUS_SUCCESS) {
602 			iscsit_global.global_svc_state = ISE_ENABLED;
603 		} else {
604 			rc = EIO;
605 			iscsit_global.global_svc_state = ISE_DISABLED;
606 		}
607 		mutex_exit(&iscsit_global.global_state_mutex);
608 		break;
609 	}
610 	case ISCSIT_IOC_DISABLE_SVC:
611 		iscsit_disable_svc();
612 		mutex_enter(&iscsit_global.global_state_mutex);
613 		iscsit_global.global_svc_state = ISE_DISABLED;
614 		mutex_exit(&iscsit_global.global_state_mutex);
615 		break;
616 
617 	default:
618 		rc = EINVAL;
619 		mutex_enter(&iscsit_global.global_state_mutex);
620 		iscsit_global.global_svc_state = ISE_ENABLED;
621 		mutex_exit(&iscsit_global.global_state_mutex);
622 	}
623 
624 	return (rc);
625 }
626 
627 static idm_status_t
628 iscsit_init(dev_info_t *dip)
629 {
630 	int			rc;
631 
632 	rc = ldi_ident_from_dip(dip, &iscsit_global.global_li);
633 	ASSERT(rc == 0);  /* Failure indicates invalid argument */
634 
635 	iscsit_global.global_svc_state = ISE_DISABLED;
636 
637 	return (IDM_STATUS_SUCCESS);
638 }
639 
640 /*
641  * iscsit_enable_svc
642  *
643  * registers all the configured targets and target portals with STMF
644  */
645 static idm_status_t
646 iscsit_enable_svc(iscsit_hostinfo_t *hostinfo)
647 {
648 	stmf_port_provider_t	*pp;
649 	stmf_dbuf_store_t	*dbuf_store;
650 	boolean_t		did_iscsit_isns_init;
651 	idm_status_t		retval = IDM_STATUS_SUCCESS;
652 
653 	ASSERT(iscsit_global.global_svc_state == ISE_ENABLING);
654 
655 	/*
656 	 * Make sure that can tell if we have partially allocated
657 	 * in case we need to exit and tear down anything allocated.
658 	 */
659 	iscsit_global.global_tsih_pool = NULL;
660 	iscsit_global.global_dbuf_store = NULL;
661 	iscsit_status_pdu_cache = NULL;
662 	pp = NULL;
663 	iscsit_global.global_pp = NULL;
664 	iscsit_global.global_default_tpg = NULL;
665 	did_iscsit_isns_init = B_FALSE;
666 	iscsit_global.global_dispatch_taskq = NULL;
667 
668 	/* Setup remaining fields in iscsit_global_t */
669 	idm_refcnt_init(&iscsit_global.global_refcnt,
670 	    &iscsit_global);
671 
672 	avl_create(&iscsit_global.global_discovery_sessions,
673 	    iscsit_sess_avl_compare, sizeof (iscsit_sess_t),
674 	    offsetof(iscsit_sess_t, ist_tgt_ln));
675 
676 	avl_create(&iscsit_global.global_target_list,
677 	    iscsit_tgt_avl_compare, sizeof (iscsit_tgt_t),
678 	    offsetof(iscsit_tgt_t, target_global_ln));
679 
680 	list_create(&iscsit_global.global_deleted_target_list,
681 	    sizeof (iscsit_tgt_t),
682 	    offsetof(iscsit_tgt_t, target_global_deleted_ln));
683 
684 	avl_create(&iscsit_global.global_tpg_list,
685 	    iscsit_tpg_avl_compare, sizeof (iscsit_tpg_t),
686 	    offsetof(iscsit_tpg_t, tpg_global_ln));
687 
688 	avl_create(&iscsit_global.global_ini_list,
689 	    iscsit_ini_avl_compare, sizeof (iscsit_ini_t),
690 	    offsetof(iscsit_ini_t, ini_global_ln));
691 
692 	iscsit_global.global_tsih_pool = vmem_create("iscsit_tsih_pool",
693 	    (void *)1, ISCSI_MAX_TSIH, 1, NULL, NULL, NULL, 0,
694 	    VM_SLEEP | VMC_IDENTIFIER);
695 
696 	/*
697 	 * Setup STMF dbuf store.  Our buffers are bound to a specific
698 	 * connection so we really can't let STMF cache buffers for us.
699 	 * Consequently we'll just allocate one global buffer store.
700 	 */
701 	dbuf_store = stmf_alloc(STMF_STRUCT_DBUF_STORE, 0, 0);
702 	if (dbuf_store == NULL) {
703 		retval = IDM_STATUS_FAIL;
704 		goto tear_down_and_return;
705 	}
706 	dbuf_store->ds_alloc_data_buf = iscsit_dbuf_alloc;
707 	dbuf_store->ds_free_data_buf = iscsit_dbuf_free;
708 	dbuf_store->ds_port_private = NULL;
709 	iscsit_global.global_dbuf_store = dbuf_store;
710 
711 	/* Status PDU cache */
712 	iscsit_status_pdu_cache = kmem_cache_create("iscsit_status_pdu_cache",
713 	    sizeof (idm_pdu_t) + sizeof (iscsi_scsi_rsp_hdr_t), 8,
714 	    &iscsit_status_pdu_constructor,
715 	    NULL, NULL, NULL, NULL, KM_SLEEP);
716 
717 	/* Default TPG and portal */
718 	iscsit_global.global_default_tpg = iscsit_tpg_createdefault();
719 	if (iscsit_global.global_default_tpg == NULL) {
720 		retval = IDM_STATUS_FAIL;
721 		goto tear_down_and_return;
722 	}
723 
724 	/* initialize isns client */
725 	(void) iscsit_isns_init(hostinfo);
726 	did_iscsit_isns_init = B_TRUE;
727 
728 	/* Register port provider */
729 	pp = stmf_alloc(STMF_STRUCT_PORT_PROVIDER, 0, 0);
730 	if (pp == NULL) {
731 		retval = IDM_STATUS_FAIL;
732 		goto tear_down_and_return;
733 	}
734 
735 	pp->pp_portif_rev = PORTIF_REV_1;
736 	pp->pp_instance = 0;
737 	pp->pp_name = ISCSIT_MODNAME;
738 	pp->pp_cb = iscsit_pp_cb;
739 
740 	iscsit_global.global_pp = pp;
741 
742 
743 	if (stmf_register_port_provider(pp) != STMF_SUCCESS) {
744 		retval = IDM_STATUS_FAIL;
745 		goto tear_down_and_return;
746 	}
747 
748 	iscsit_global.global_dispatch_taskq = taskq_create("iscsit_dispatch",
749 	    1, minclsyspri, 16, 16, TASKQ_PREPOPULATE);
750 
751 	/* Scan staged PDUs, meaningful in MC/S situations */
752 	iscsit_rxpdu_queue_monitor_start();
753 
754 	return (IDM_STATUS_SUCCESS);
755 
756 tear_down_and_return:
757 
758 	if (iscsit_global.global_dispatch_taskq) {
759 		taskq_destroy(iscsit_global.global_dispatch_taskq);
760 		iscsit_global.global_dispatch_taskq = NULL;
761 	}
762 
763 	if (did_iscsit_isns_init)
764 		iscsit_isns_fini();
765 
766 	if (iscsit_global.global_default_tpg) {
767 		iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
768 		iscsit_global.global_default_tpg = NULL;
769 	}
770 
771 	if (iscsit_global.global_pp)
772 		iscsit_global.global_pp = NULL;
773 
774 	if (pp)
775 		stmf_free(pp);
776 
777 	if (iscsit_status_pdu_cache) {
778 		kmem_cache_destroy(iscsit_status_pdu_cache);
779 		iscsit_status_pdu_cache = NULL;
780 	}
781 
782 	if (iscsit_global.global_dbuf_store) {
783 		stmf_free(iscsit_global.global_dbuf_store);
784 		iscsit_global.global_dbuf_store = NULL;
785 	}
786 
787 	if (iscsit_global.global_tsih_pool) {
788 		vmem_destroy(iscsit_global.global_tsih_pool);
789 		iscsit_global.global_tsih_pool = NULL;
790 	}
791 
792 	avl_destroy(&iscsit_global.global_ini_list);
793 	avl_destroy(&iscsit_global.global_tpg_list);
794 	list_destroy(&iscsit_global.global_deleted_target_list);
795 	avl_destroy(&iscsit_global.global_target_list);
796 	avl_destroy(&iscsit_global.global_discovery_sessions);
797 
798 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
799 
800 	return (retval);
801 }
802 
803 /*
804  * iscsit_disable_svc
805  *
806  * clean up all existing connections and deregister targets from STMF
807  */
808 static void
809 iscsit_disable_svc(void)
810 {
811 	iscsit_sess_t	*sess;
812 
813 	ASSERT(iscsit_global.global_svc_state == ISE_DISABLING);
814 
815 	iscsit_rxpdu_queue_monitor_stop();
816 
817 	/* tear down discovery sessions */
818 	for (sess = avl_first(&iscsit_global.global_discovery_sessions);
819 	    sess != NULL;
820 	    sess = AVL_NEXT(&iscsit_global.global_discovery_sessions, sess))
821 		iscsit_sess_close(sess);
822 
823 	/*
824 	 * Passing NULL to iscsit_config_merge tells it to go to an empty
825 	 * config.
826 	 */
827 	(void) iscsit_config_merge(NULL);
828 
829 	/*
830 	 * Wait until there are no more global references
831 	 */
832 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
833 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
834 
835 	/*
836 	 * Default TPG must be destroyed after global_refcnt is 0.
837 	 */
838 	iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
839 
840 	avl_destroy(&iscsit_global.global_discovery_sessions);
841 	list_destroy(&iscsit_global.global_deleted_target_list);
842 	avl_destroy(&iscsit_global.global_target_list);
843 	avl_destroy(&iscsit_global.global_tpg_list);
844 	avl_destroy(&iscsit_global.global_ini_list);
845 
846 	taskq_destroy(iscsit_global.global_dispatch_taskq);
847 
848 	iscsit_isns_fini();
849 
850 	stmf_free(iscsit_global.global_dbuf_store);
851 	iscsit_global.global_dbuf_store = NULL;
852 
853 	(void) stmf_deregister_port_provider(iscsit_global.global_pp);
854 	stmf_free(iscsit_global.global_pp);
855 	iscsit_global.global_pp = NULL;
856 
857 	kmem_cache_destroy(iscsit_status_pdu_cache);
858 	iscsit_status_pdu_cache = NULL;
859 
860 	vmem_destroy(iscsit_global.global_tsih_pool);
861 	iscsit_global.global_tsih_pool = NULL;
862 }
863 
864 void
865 iscsit_global_hold()
866 {
867 	/*
868 	 * To take out a global hold, we must either own the global
869 	 * state mutex or we must be running inside of an ioctl that
870 	 * has set the global state to ISE_BUSY, ISE_DISABLING, or
871 	 * ISE_ENABLING.  We don't track the "owner" for these flags,
872 	 * so just checking if they are set is enough for now.
873 	 */
874 	ASSERT((iscsit_global.global_svc_state == ISE_ENABLING) ||
875 	    (iscsit_global.global_svc_state == ISE_DISABLING) ||
876 	    (iscsit_global.global_svc_state == ISE_BUSY) ||
877 	    MUTEX_HELD(&iscsit_global.global_state_mutex));
878 
879 	idm_refcnt_hold(&iscsit_global.global_refcnt);
880 }
881 
882 void
883 iscsit_global_rele()
884 {
885 	idm_refcnt_rele(&iscsit_global.global_refcnt);
886 }
887 
888 void
889 iscsit_global_wait_ref()
890 {
891 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
892 }
893 
894 /*
895  * IDM callbacks
896  */
897 
898 /*ARGSUSED*/
899 void
900 iscsit_rx_pdu(idm_conn_t *ic, idm_pdu_t *rx_pdu)
901 {
902 	iscsit_conn_t *ict = ic->ic_handle;
903 	switch (IDM_PDU_OPCODE(rx_pdu)) {
904 	case ISCSI_OP_SCSI_CMD:
905 		ASSERT(0); /* Shouldn't happen */
906 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
907 		break;
908 	case ISCSI_OP_SNACK_CMD:
909 		/*
910 		 * We'll need to handle this when we support ERL1/2.  For
911 		 * now we treat it as a protocol error.
912 		 */
913 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
914 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
915 		break;
916 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
917 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
918 			iscsit_set_cmdsn(ict, rx_pdu);
919 			iscsit_op_scsi_task_mgmt(ict, rx_pdu);
920 		}
921 		break;
922 	case ISCSI_OP_NOOP_OUT:
923 	case ISCSI_OP_LOGIN_CMD:
924 	case ISCSI_OP_TEXT_CMD:
925 	case ISCSI_OP_LOGOUT_CMD:
926 		/*
927 		 * If/when we switch to userland processing these PDU's
928 		 * will be handled by iscsitd.
929 		 */
930 		iscsit_deferred_dispatch(rx_pdu);
931 		break;
932 	default:
933 		/* Protocol error */
934 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
935 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
936 		break;
937 	}
938 }
939 
940 /*ARGSUSED*/
941 void
942 iscsit_rx_pdu_error(idm_conn_t *ic, idm_pdu_t *rx_pdu, idm_status_t status)
943 {
944 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
945 }
946 
947 /*
948  * iscsit_rx_scsi_rsp -- cause the connection to be closed if response rx'd
949  *
950  * A target sends an SCSI Response PDU, it should never receive one.
951  * This has been seen when running the Codemonicon suite of tests which
952  * does negative testing of the protocol. If such a condition occurs using
953  * a normal initiator it most likely means there's data corruption in the
954  * header and that's grounds for dropping the connection as well.
955  */
956 void
957 iscsit_rx_scsi_rsp(idm_conn_t *ic, idm_pdu_t *rx_pdu)
958 {
959 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
960 	idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
961 }
962 
963 void
964 iscsit_task_aborted(idm_task_t *idt, idm_status_t status)
965 {
966 	iscsit_task_t *itask = idt->idt_private;
967 
968 	switch (status) {
969 	case IDM_STATUS_SUSPENDED:
970 		break;
971 	case IDM_STATUS_ABORTED:
972 		mutex_enter(&itask->it_mutex);
973 		itask->it_aborted = B_TRUE;
974 		/*
975 		 * We rely on the fact that STMF tracks outstanding
976 		 * buffer transfers and will free all of our buffers
977 		 * before freeing the task so we don't need to
978 		 * explicitly free the buffers from iscsit/idm
979 		 */
980 		if (itask->it_stmf_abort) {
981 			mutex_exit(&itask->it_mutex);
982 			/*
983 			 * Task is no longer active
984 			 */
985 			iscsit_task_done(itask);
986 
987 			/*
988 			 * STMF has already asked for this task to be aborted
989 			 *
990 			 * STMF specification is wrong... says to return
991 			 * STMF_ABORTED, the code actually looks for
992 			 * STMF_ABORT_SUCCESS.
993 			 */
994 			stmf_task_lport_aborted(itask->it_stmf_task,
995 			    STMF_ABORT_SUCCESS, STMF_IOF_LPORT_DONE);
996 			return;
997 		} else {
998 			mutex_exit(&itask->it_mutex);
999 			/*
1000 			 * Tell STMF to stop processing the task.
1001 			 */
1002 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
1003 			    STMF_ABORTED, NULL);
1004 			return;
1005 		}
1006 		/*NOTREACHED*/
1007 	default:
1008 		ASSERT(0);
1009 	}
1010 }
1011 
1012 /*ARGSUSED*/
1013 idm_status_t
1014 iscsit_client_notify(idm_conn_t *ic, idm_client_notify_t icn,
1015     uintptr_t data)
1016 {
1017 	idm_status_t rc = IDM_STATUS_SUCCESS;
1018 
1019 	/*
1020 	 * IDM client notifications will never occur at interrupt level
1021 	 * since they are generated from the connection state machine which
1022 	 * running on taskq threads.
1023 	 *
1024 	 */
1025 	switch (icn) {
1026 	case CN_CONNECT_ACCEPT:
1027 		rc = iscsit_conn_accept(ic); /* No data */
1028 		break;
1029 	case CN_FFP_ENABLED:
1030 		rc = iscsit_ffp_enabled(ic); /* No data */
1031 		break;
1032 	case CN_FFP_DISABLED:
1033 		/*
1034 		 * Data indicates whether this was the result of an
1035 		 * explicit logout request.
1036 		 */
1037 		rc = iscsit_ffp_disabled(ic, (idm_ffp_disable_t)data);
1038 		break;
1039 	case CN_CONNECT_LOST:
1040 		rc = iscsit_conn_lost(ic);
1041 		break;
1042 	case CN_CONNECT_DESTROY:
1043 		rc = iscsit_conn_destroy(ic);
1044 		break;
1045 	case CN_LOGIN_FAIL:
1046 		/*
1047 		 * Force the login state machine to completion
1048 		 */
1049 		rc = iscsit_login_fail(ic);
1050 		break;
1051 	default:
1052 		rc = IDM_STATUS_REJECT;
1053 		break;
1054 	}
1055 
1056 	return (rc);
1057 }
1058 
1059 /*
1060  * iscsit_update_statsn is invoked for all the PDUs which have the StatSN
1061  * field in the header. The StatSN is incremented if the IDM_PDU_ADVANCE_STATSN
1062  * flag is set in the pdu flags field. The StatSN is connection-wide and is
1063  * protected by the mutex ict_statsn_mutex. For Data-In PDUs, if the flag
1064  * IDM_TASK_PHASECOLLAPSE_REQ is set, the status (phase-collapse) is also filled
1065  */
1066 void
1067 iscsit_update_statsn(idm_task_t *idm_task, idm_pdu_t *pdu)
1068 {
1069 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1070 	iscsit_conn_t *ict = (iscsit_conn_t *)pdu->isp_ic->ic_handle;
1071 	iscsit_task_t *itask = NULL;
1072 	scsi_task_t *task = NULL;
1073 
1074 	mutex_enter(&ict->ict_statsn_mutex);
1075 	rsp->statsn = htonl(ict->ict_statsn);
1076 	if (pdu->isp_flags & IDM_PDU_ADVANCE_STATSN)
1077 		ict->ict_statsn++;
1078 	mutex_exit(&ict->ict_statsn_mutex);
1079 
1080 	/*
1081 	 * The last SCSI Data PDU passed for a command may also contain the
1082 	 * status if the status indicates termination with no expections, i.e.
1083 	 * no sense data or response involved. If the command completes with
1084 	 * an error, then the response and sense data will be sent in a
1085 	 * separate iSCSI Response PDU.
1086 	 */
1087 	if ((idm_task) && (idm_task->idt_flags & IDM_TASK_PHASECOLLAPSE_REQ)) {
1088 		itask = idm_task->idt_private;
1089 		task = itask->it_stmf_task;
1090 
1091 		rsp->cmd_status = task->task_scsi_status;
1092 		rsp->flags	|= ISCSI_FLAG_DATA_STATUS;
1093 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1094 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1095 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1096 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1097 		}
1098 		rsp->residual_count = htonl(task->task_resid);
1099 
1100 		/*
1101 		 * Removing the task from the session task list
1102 		 * just before the status is sent in the last
1103 		 * Data PDU transfer
1104 		 */
1105 		iscsit_task_done(itask);
1106 	}
1107 }
1108 
1109 void
1110 iscsit_build_hdr(idm_task_t *idm_task, idm_pdu_t *pdu, uint8_t opcode)
1111 {
1112 	iscsit_task_t *itask = idm_task->idt_private;
1113 	iscsi_data_rsp_hdr_t *dh = (iscsi_data_rsp_hdr_t *)pdu->isp_hdr;
1114 
1115 	/*
1116 	 * We acquired iscsit_sess_t.ist_sn_mutex in iscsit_xfer_scsi_data
1117 	 */
1118 	ASSERT(MUTEX_HELD(&itask->it_ict->ict_sess->ist_sn_mutex));
1119 	/*
1120 	 * On incoming data, the target transfer tag and Lun is only
1121 	 * provided by the target if the A bit is set, Since the target
1122 	 * does not currently support Error Recovery Level 1, the A
1123 	 * bit is never set.
1124 	 */
1125 	dh->opcode = opcode;
1126 	dh->itt = itask->it_itt;
1127 	dh->ttt = ((opcode & ISCSI_OPCODE_MASK) == ISCSI_OP_SCSI_DATA_RSP) ?
1128 	    ISCSI_RSVD_TASK_TAG : itask->it_ttt;
1129 
1130 	dh->expcmdsn = htonl(itask->it_ict->ict_sess->ist_expcmdsn);
1131 	dh->maxcmdsn = htonl(itask->it_ict->ict_sess->ist_maxcmdsn);
1132 
1133 	/*
1134 	 * IDM must set:
1135 	 *
1136 	 * data.flags and rtt.flags
1137 	 * data.dlength
1138 	 * data.datasn
1139 	 * data.offset
1140 	 * statsn, residual_count and cmd_status (for phase collapse)
1141 	 * rtt.rttsn
1142 	 * rtt.data_offset
1143 	 * rtt.data_length
1144 	 */
1145 }
1146 
1147 void
1148 iscsit_keepalive(idm_conn_t *ic)
1149 {
1150 	idm_pdu_t		*nop_in_pdu;
1151 	iscsi_nop_in_hdr_t	*nop_in;
1152 	iscsit_conn_t		*ict = ic->ic_handle;
1153 
1154 	/*
1155 	 * IDM noticed the connection has been idle for too long so it's
1156 	 * time to provoke some activity.  Build and transmit an iSCSI
1157 	 * nop-in PDU -- when the initiator responds it will be counted
1158 	 * as "activity" and keep the connection alive.
1159 	 *
1160 	 * We don't actually care about the response here at the iscsit level
1161 	 * so we will just throw it away without looking at it when it arrives.
1162 	 */
1163 	nop_in_pdu = idm_pdu_alloc(sizeof (*nop_in), 0);
1164 	idm_pdu_init(nop_in_pdu, ic, NULL, NULL);
1165 	nop_in = (iscsi_nop_in_hdr_t *)nop_in_pdu->isp_hdr;
1166 	bzero(nop_in, sizeof (*nop_in));
1167 	nop_in->opcode = ISCSI_OP_NOOP_IN;
1168 	nop_in->flags = ISCSI_FLAG_FINAL;
1169 	nop_in->itt = ISCSI_RSVD_TASK_TAG;
1170 	/*
1171 	 * When the target sends a NOP-In as a Ping, the target transfer tag
1172 	 * is set to a valid (not reserved) value and the initiator task tag
1173 	 * is set to ISCSI_RSVD_TASK_TAG (0xffffffff). In this case the StatSN
1174 	 * will always contain the next sequence number but the StatSN for the
1175 	 * connection is not advanced after this PDU is sent.
1176 	 */
1177 	nop_in_pdu->isp_flags |= IDM_PDU_SET_STATSN;
1178 	/*
1179 	 * This works because we don't currently allocate ttt's anywhere else
1180 	 * in iscsit so as long as we stay out of IDM's range we are safe.
1181 	 * If we need to allocate ttt's for other PDU's in the future this will
1182 	 * need to be improved.
1183 	 */
1184 	mutex_enter(&ict->ict_mutex);
1185 	nop_in->ttt = ict->ict_keepalive_ttt;
1186 	ict->ict_keepalive_ttt++;
1187 	if (ict->ict_keepalive_ttt == ISCSI_RSVD_TASK_TAG)
1188 		ict->ict_keepalive_ttt = IDM_TASKIDS_MAX;
1189 	mutex_exit(&ict->ict_mutex);
1190 
1191 	iscsit_pdu_tx(nop_in_pdu);
1192 }
1193 
1194 static idm_status_t
1195 iscsit_conn_accept(idm_conn_t *ic)
1196 {
1197 	iscsit_conn_t *ict;
1198 
1199 	/*
1200 	 * We need to get a global hold here to ensure that the service
1201 	 * doesn't get shutdown prior to establishing a session. This
1202 	 * gets released in iscsit_conn_destroy().
1203 	 */
1204 	mutex_enter(&iscsit_global.global_state_mutex);
1205 	if (iscsit_global.global_svc_state != ISE_ENABLED) {
1206 		mutex_exit(&iscsit_global.global_state_mutex);
1207 		return (IDM_STATUS_FAIL);
1208 	}
1209 	iscsit_global_hold();
1210 	mutex_exit(&iscsit_global.global_state_mutex);
1211 
1212 	/*
1213 	 * Allocate an associated iscsit structure to represent this
1214 	 * connection.  We shouldn't really create a session until we
1215 	 * get the first login PDU.
1216 	 */
1217 	ict = kmem_zalloc(sizeof (*ict), KM_SLEEP);
1218 
1219 	ict->ict_ic = ic;
1220 	ict->ict_statsn = 1;
1221 	ict->ict_keepalive_ttt = IDM_TASKIDS_MAX; /* Avoid IDM TT range */
1222 	ic->ic_handle = ict;
1223 	mutex_init(&ict->ict_mutex, NULL, MUTEX_DRIVER, NULL);
1224 	mutex_init(&ict->ict_statsn_mutex, NULL, MUTEX_DRIVER, NULL);
1225 	idm_refcnt_init(&ict->ict_refcnt, ict);
1226 
1227 	/*
1228 	 * Initialize login state machine
1229 	 */
1230 	if (iscsit_login_sm_init(ict) != IDM_STATUS_SUCCESS) {
1231 		iscsit_global_rele();
1232 		/*
1233 		 * Cleanup the ict after idm notifies us about this failure
1234 		 */
1235 		return (IDM_STATUS_FAIL);
1236 	}
1237 
1238 	return (IDM_STATUS_SUCCESS);
1239 }
1240 
1241 idm_status_t
1242 iscsit_conn_reinstate(iscsit_conn_t *reinstate_ict, iscsit_conn_t *new_ict)
1243 {
1244 	idm_status_t	result;
1245 
1246 	/*
1247 	 * Note in new connection state that this connection is
1248 	 * reinstating an existing connection.
1249 	 */
1250 	new_ict->ict_reinstating = B_TRUE;
1251 	new_ict->ict_reinstate_conn = reinstate_ict;
1252 	new_ict->ict_statsn = reinstate_ict->ict_statsn;
1253 
1254 	/*
1255 	 * Now generate connection state machine event to existing connection
1256 	 * so that it starts the cleanup process.
1257 	 */
1258 	result = idm_conn_reinstate_event(reinstate_ict->ict_ic,
1259 	    new_ict->ict_ic);
1260 
1261 	return (result);
1262 }
1263 
1264 void
1265 iscsit_conn_hold(iscsit_conn_t *ict)
1266 {
1267 	idm_refcnt_hold(&ict->ict_refcnt);
1268 }
1269 
1270 void
1271 iscsit_conn_rele(iscsit_conn_t *ict)
1272 {
1273 	idm_refcnt_rele(&ict->ict_refcnt);
1274 }
1275 
1276 void
1277 iscsit_conn_dispatch_hold(iscsit_conn_t *ict)
1278 {
1279 	idm_refcnt_hold(&ict->ict_dispatch_refcnt);
1280 }
1281 
1282 void
1283 iscsit_conn_dispatch_rele(iscsit_conn_t *ict)
1284 {
1285 	idm_refcnt_rele(&ict->ict_dispatch_refcnt);
1286 }
1287 
1288 static idm_status_t
1289 iscsit_login_fail(idm_conn_t *ic)
1290 {
1291 	iscsit_conn_t *ict = ic->ic_handle;
1292 
1293 	/* Generate login state machine event */
1294 	iscsit_login_sm_event(ict, ILE_LOGIN_CONN_ERROR, NULL);
1295 
1296 	return (IDM_STATUS_SUCCESS);
1297 }
1298 
1299 static idm_status_t
1300 iscsit_ffp_enabled(idm_conn_t *ic)
1301 {
1302 	iscsit_conn_t *ict = ic->ic_handle;
1303 
1304 	/* Generate session state machine event */
1305 	iscsit_sess_sm_event(ict->ict_sess, SE_CONN_LOGGED_IN, ict);
1306 
1307 	return (IDM_STATUS_SUCCESS);
1308 }
1309 
1310 static idm_status_t
1311 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class)
1312 {
1313 	iscsit_conn_t *ict = ic->ic_handle;
1314 
1315 	/* Generate session state machine event */
1316 	switch (disable_class) {
1317 	case FD_CONN_FAIL:
1318 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_FAIL, ict);
1319 		break;
1320 	case FD_CONN_LOGOUT:
1321 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_DISABLE, ict);
1322 		break;
1323 	case FD_SESS_LOGOUT:
1324 		iscsit_sess_sm_event(ict->ict_sess, SE_SESSION_CLOSE, ict);
1325 		break;
1326 	default:
1327 		ASSERT(0);
1328 	}
1329 
1330 	return (IDM_STATUS_SUCCESS);
1331 }
1332 
1333 static idm_status_t
1334 iscsit_conn_lost(idm_conn_t *ic)
1335 {
1336 	iscsit_conn_t	*ict	= ic->ic_handle;
1337 	iscsit_sess_t	*ist	= ict->ict_sess;
1338 	iscsit_cbuf_t	*cbuf;
1339 	idm_pdu_t	*rx_pdu;
1340 	int i;
1341 
1342 	mutex_enter(&ict->ict_mutex);
1343 	ict->ict_lost = B_TRUE;
1344 	mutex_exit(&ict->ict_mutex);
1345 	/*
1346 	 * scrub the staging queue for all PDUs on this connection
1347 	 */
1348 	if (ist != NULL) {
1349 		mutex_enter(&ist->ist_sn_mutex);
1350 		for (cbuf = ist->ist_rxpdu_queue, i = 0;
1351 		    ((cbuf->cb_num_elems > 0) && (i < ISCSIT_RXPDU_QUEUE_LEN));
1352 		    i++) {
1353 			if (((rx_pdu = cbuf->cb_buffer[i]) != NULL) &&
1354 			    (rx_pdu->isp_ic == ic)) {
1355 				/* conn is lost, drop the pdu */
1356 				DTRACE_PROBE3(scrubbing__staging__queue,
1357 				    iscsit_sess_t *, ist, idm_conn_t *, ic,
1358 				    idm_pdu_t *, rx_pdu);
1359 				idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
1360 				cbuf->cb_buffer[i] = NULL;
1361 				cbuf->cb_num_elems--;
1362 				iscsit_conn_dispatch_rele(ict);
1363 			}
1364 		}
1365 		mutex_exit(&ist->ist_sn_mutex);
1366 	}
1367 	/*
1368 	 * Make sure there aren't any PDU's transitioning from the receive
1369 	 * handler to the dispatch taskq.
1370 	 */
1371 	idm_refcnt_wait_ref(&ict->ict_dispatch_refcnt);
1372 
1373 	return (IDM_STATUS_SUCCESS);
1374 }
1375 
1376 static idm_status_t
1377 iscsit_conn_destroy(idm_conn_t *ic)
1378 {
1379 	iscsit_conn_t *ict = ic->ic_handle;
1380 
1381 	mutex_enter(&ict->ict_mutex);
1382 	ict->ict_destroyed = B_TRUE;
1383 	mutex_exit(&ict->ict_mutex);
1384 
1385 	/* Generate session state machine event */
1386 	if (ict->ict_sess != NULL) {
1387 		/*
1388 		 * Session state machine will call iscsit_conn_destroy_done()
1389 		 * when it has removed references to this connection.
1390 		 */
1391 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FAIL, ict);
1392 	}
1393 
1394 	idm_refcnt_wait_ref(&ict->ict_refcnt);
1395 	/*
1396 	 * The session state machine does not need to post
1397 	 * events to IDM any longer, so it is safe to set
1398 	 * the idm connection reference to NULL
1399 	 */
1400 	ict->ict_ic = NULL;
1401 
1402 	/* Reap the login state machine */
1403 	iscsit_login_sm_fini(ict);
1404 
1405 	/* Clean up any text command remnants */
1406 	iscsit_text_cmd_fini(ict);
1407 
1408 	mutex_destroy(&ict->ict_mutex);
1409 	idm_refcnt_destroy(&ict->ict_refcnt);
1410 	kmem_free(ict, sizeof (*ict));
1411 
1412 	iscsit_global_rele();
1413 
1414 	return (IDM_STATUS_SUCCESS);
1415 }
1416 
1417 void
1418 iscsit_conn_logout(iscsit_conn_t *ict)
1419 {
1420 	/*
1421 	 * If the iscsi connection is active, then
1422 	 * logout the IDM connection by sending a
1423 	 * CE_LOGOUT_SESSION_SUCCESS, else, no action
1424 	 * needs to be taken because the connection
1425 	 * is already in the teardown process.
1426 	 */
1427 	mutex_enter(&ict->ict_mutex);
1428 	if (ict->ict_lost == B_FALSE && ict->ict_destroyed == B_FALSE) {
1429 		idm_conn_event(ict->ict_ic, CE_LOGOUT_SESSION_SUCCESS, NULL);
1430 	}
1431 	mutex_exit(&ict->ict_mutex);
1432 }
1433 
1434 /*
1435  * STMF-related functions
1436  *
1437  * iSCSI to STMF mapping
1438  *
1439  * Session == ?
1440  * Connection == bound to local port but not itself a local port
1441  * Target
1442  * Target portal (group?) == local port (really but we're not going to do this)
1443  *	iscsit needs to map connections to local ports (whatever we decide
1444  * 	they are)
1445  * Target == ?
1446  */
1447 
1448 /*ARGSUSED*/
1449 static stmf_data_buf_t *
1450 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
1451     uint32_t flags)
1452 {
1453 	iscsit_task_t *itask = task->task_port_private;
1454 	idm_buf_t *idm_buffer;
1455 	iscsit_buf_t	*ibuf;
1456 	stmf_data_buf_t *result;
1457 	uint32_t	bsize;
1458 
1459 	/*
1460 	 * If the requested size is larger than MaxBurstLength and the
1461 	 * given pminsize is also larger than MaxBurstLength, then the
1462 	 * allocation fails (dbuf = NULL) and pminsize is modified to
1463 	 * be equal to MaxBurstLength. stmf/sbd then should re-invoke
1464 	 * this function with the corrected values for transfer.
1465 	 */
1466 	ASSERT(pminsize);
1467 	if (size <= itask->it_ict->ict_op.op_max_burst_length) {
1468 		bsize = size;
1469 	} else if (*pminsize <= itask->it_ict->ict_op.op_max_burst_length) {
1470 		bsize = itask->it_ict->ict_op.op_max_burst_length;
1471 	} else {
1472 		*pminsize = itask->it_ict->ict_op.op_max_burst_length;
1473 		return (NULL);
1474 	}
1475 
1476 	/* Alloc buffer */
1477 	idm_buffer = idm_buf_alloc(itask->it_ict->ict_ic, NULL, bsize);
1478 	if (idm_buffer != NULL) {
1479 		result = stmf_alloc(STMF_STRUCT_DATA_BUF,
1480 		    sizeof (iscsit_buf_t), 0);
1481 		if (result != NULL) {
1482 			/* Fill in stmf_data_buf_t */
1483 			ibuf = result->db_port_private;
1484 			ibuf->ibuf_idm_buf = idm_buffer;
1485 			ibuf->ibuf_stmf_buf = result;
1486 			ibuf->ibuf_is_immed = B_FALSE;
1487 			result->db_flags = DB_DONT_CACHE;
1488 			result->db_buf_size = bsize;
1489 			result->db_data_size = bsize;
1490 			result->db_sglist_length = 1;
1491 			result->db_sglist[0].seg_addr = idm_buffer->idb_buf;
1492 			result->db_sglist[0].seg_length =
1493 			    idm_buffer->idb_buflen;
1494 			return (result);
1495 		}
1496 
1497 		/* Couldn't get the stmf_data_buf_t so free the buffer */
1498 		idm_buf_free(idm_buffer);
1499 	}
1500 
1501 	return (NULL);
1502 }
1503 
1504 /*ARGSUSED*/
1505 static void
1506 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf)
1507 {
1508 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1509 
1510 	if (ibuf->ibuf_is_immed) {
1511 		/*
1512 		 * The iscsit_buf_t structure itself will be freed with its
1513 		 * associated task.  Here we just need to free the PDU that
1514 		 * held the immediate data.
1515 		 */
1516 		idm_pdu_complete(ibuf->ibuf_immed_data_pdu, IDM_STATUS_SUCCESS);
1517 		ibuf->ibuf_immed_data_pdu = 0;
1518 	} else {
1519 		idm_buf_free(ibuf->ibuf_idm_buf);
1520 		stmf_free(dbuf);
1521 	}
1522 }
1523 
1524 /*ARGSUSED*/
1525 stmf_status_t
1526 iscsit_xfer_scsi_data(scsi_task_t *task, stmf_data_buf_t *dbuf,
1527     uint32_t ioflags)
1528 {
1529 	iscsit_task_t *iscsit_task = task->task_port_private;
1530 	iscsit_sess_t *ict_sess = iscsit_task->it_ict->ict_sess;
1531 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1532 	int idm_rc;
1533 
1534 	/*
1535 	 * If we are aborting then we can ignore this request
1536 	 */
1537 	if (iscsit_task->it_stmf_abort) {
1538 		return (STMF_SUCCESS);
1539 	}
1540 
1541 	/*
1542 	 * If it's not immediate data then start the transfer
1543 	 */
1544 	ASSERT(ibuf->ibuf_is_immed == B_FALSE);
1545 	if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
1546 		/*
1547 		 * The DB_SEND_STATUS_GOOD flag in the STMF data buffer allows
1548 		 * the port provider to phase-collapse, i.e. send the status
1549 		 * along with the final data PDU for the command. The port
1550 		 * provider passes this request to the transport layer by
1551 		 * setting a flag IDM_TASK_PHASECOLLAPSE_REQ in the task.
1552 		 */
1553 		if (dbuf->db_flags & DB_SEND_STATUS_GOOD)
1554 			iscsit_task->it_idm_task->idt_flags |=
1555 			    IDM_TASK_PHASECOLLAPSE_REQ;
1556 		/*
1557 		 * IDM will call iscsit_build_hdr so lock now to serialize
1558 		 * access to the SN values.  We need to lock here to enforce
1559 		 * lock ordering
1560 		 */
1561 		mutex_enter(&ict_sess->ist_sn_mutex);
1562 		idm_rc = idm_buf_tx_to_ini(iscsit_task->it_idm_task,
1563 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1564 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1565 		mutex_exit(&ict_sess->ist_sn_mutex);
1566 
1567 		return (iscsit_idm_to_stmf(idm_rc));
1568 	} else if (dbuf->db_flags & DB_DIRECTION_FROM_RPORT) {
1569 		/* Grab the SN lock (see comment above) */
1570 		mutex_enter(&ict_sess->ist_sn_mutex);
1571 		idm_rc = idm_buf_rx_from_ini(iscsit_task->it_idm_task,
1572 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1573 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1574 		mutex_exit(&ict_sess->ist_sn_mutex);
1575 
1576 		return (iscsit_idm_to_stmf(idm_rc));
1577 	}
1578 
1579 	/* What are we supposed to do if there is no direction? */
1580 	return (STMF_INVALID_ARG);
1581 }
1582 
1583 static void
1584 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status)
1585 {
1586 	iscsit_task_t *itask = idb->idb_task_binding->idt_private;
1587 	stmf_data_buf_t *dbuf = idb->idb_cb_arg;
1588 
1589 	dbuf->db_xfer_status = iscsit_idm_to_stmf(status);
1590 
1591 	/*
1592 	 * If the task has been aborted then we don't need to call STMF
1593 	 */
1594 	if (itask->it_stmf_abort) {
1595 		return;
1596 	}
1597 
1598 	/*
1599 	 * For ISCSI over TCP (not iSER), the last SCSI Data PDU passed
1600 	 * for a successful command contains the status as requested by
1601 	 * by COMSTAR (via the DB_SEND_STATUS_GOOD flag). But the iSER
1602 	 * transport does not support phase-collapse. So pretend we are
1603 	 * COMSTAR and send the status in a separate PDU now.
1604 	 */
1605 	if (idb->idb_task_binding->idt_flags & IDM_TASK_PHASECOLLAPSE_SUCCESS) {
1606 		/*
1607 		 * Mark task complete and notify COMSTAR
1608 		 * that the status has been sent.
1609 		 */
1610 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1611 		stmf_send_status_done(itask->it_stmf_task,
1612 		    iscsit_idm_to_stmf(status), STMF_IOF_LPORT_DONE);
1613 	} else if ((dbuf->db_flags & DB_SEND_STATUS_GOOD) &&
1614 	    status == IDM_STATUS_SUCCESS) {
1615 
1616 		/*
1617 		 * The iscsi target port provider - for iSER, emulates the
1618 		 * DB_SEND_STATUS_GOOD optimization if requested by STMF;
1619 		 * it sends the status in a separate PDU after the data
1620 		 * transfer. In this case the port provider should first
1621 		 * call stmf_data_xfer_done() to mark the transfer complete
1622 		 * and then send the status. Although STMF will free the
1623 		 * buffer at the time the task is freed, even if the transfer
1624 		 * is not marked complete, this behavior makes statistics
1625 		 * gathering and task state tracking more difficult than it
1626 		 * needs to be.
1627 		 */
1628 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1629 		if (iscsit_send_scsi_status(itask->it_stmf_task, 0)
1630 		    != STMF_SUCCESS) {
1631 			stmf_send_status_done(itask->it_stmf_task,
1632 			    STMF_FAILURE, STMF_IOF_LPORT_DONE);
1633 		}
1634 	} else {
1635 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1636 		/* don't touch dbuf after stmf_data_xfer_done */
1637 	}
1638 }
1639 
1640 
1641 /*ARGSUSED*/
1642 stmf_status_t
1643 iscsit_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
1644 {
1645 	iscsit_task_t *itask = task->task_port_private;
1646 	iscsi_scsi_rsp_hdr_t *rsp;
1647 	idm_pdu_t *pdu;
1648 	int resp_datalen;
1649 
1650 	/*
1651 	 * If this task is aborted then we don't need to respond.
1652 	 */
1653 	if (itask->it_stmf_abort) {
1654 		return (STMF_SUCCESS);
1655 	}
1656 
1657 	/*
1658 	 * If this is a task management status, handle it elsewhere.
1659 	 */
1660 	if (task->task_mgmt_function != TM_NONE) {
1661 		/*
1662 		 * Don't wait for the PDU completion to tell STMF
1663 		 * the task is done -- it doesn't really matter and
1664 		 * it makes life complicated if STMF later asks us to
1665 		 * abort the request and we don't know whether the
1666 		 * status has been sent or not.
1667 		 */
1668 		itask->it_tm_responded = B_TRUE;
1669 		iscsit_send_task_mgmt_resp(itask->it_tm_pdu,
1670 		    (task->task_completion_status == STMF_SUCCESS) ?
1671 		    SCSI_TCP_TM_RESP_COMPLETE : SCSI_TCP_TM_RESP_FUNC_NOT_SUPP);
1672 		stmf_send_status_done(task, STMF_SUCCESS,
1673 		    STMF_IOF_LPORT_DONE);
1674 		return (STMF_SUCCESS);
1675 	}
1676 
1677 	/*
1678 	 * Remove the task from the session task list
1679 	 */
1680 	iscsit_task_done(itask);
1681 
1682 	/*
1683 	 * Send status
1684 	 */
1685 	mutex_enter(&itask->it_idm_task->idt_mutex);
1686 	if ((itask->it_idm_task->idt_state == TASK_ACTIVE) &&
1687 	    (task->task_completion_status == STMF_SUCCESS) &&
1688 	    (task->task_sense_length == 0) &&
1689 	    (task->task_resid == 0)) {
1690 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1691 		/* PDU callback releases task hold */
1692 		idm_task_hold(itask->it_idm_task);
1693 		mutex_exit(&itask->it_idm_task->idt_mutex);
1694 		/*
1695 		 * Fast path.  Cached status PDU's are already
1696 		 * initialized.  We just need to fill in
1697 		 * connection and task information. StatSN is
1698 		 * incremented by 1 for every status sent a
1699 		 * connection.
1700 		 */
1701 		pdu = kmem_cache_alloc(iscsit_status_pdu_cache, KM_SLEEP);
1702 		pdu->isp_ic = itask->it_ict->ict_ic;
1703 		pdu->isp_private = itask;
1704 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1705 
1706 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1707 		rsp->itt = itask->it_itt;
1708 		/*
1709 		 * ExpDataSN is the number of R2T and Data-In (read)
1710 		 * PDUs the target has sent for the SCSI command.
1711 		 *
1712 		 * Since there is no support for bidirectional transfer
1713 		 * yet, either idt_exp_datasn or idt_exp_rttsn, but not
1714 		 * both is valid at any time
1715 		 */
1716 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1717 		    htonl(itask->it_idm_task->idt_exp_datasn):
1718 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1719 		rsp->cmd_status = task->task_scsi_status;
1720 		iscsit_pdu_tx(pdu);
1721 		return (STMF_SUCCESS);
1722 	} else {
1723 		if (itask->it_idm_task->idt_state != TASK_ACTIVE) {
1724 			mutex_exit(&itask->it_idm_task->idt_mutex);
1725 			return (STMF_FAILURE);
1726 		}
1727 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1728 		/* PDU callback releases task hold */
1729 		idm_task_hold(itask->it_idm_task);
1730 		mutex_exit(&itask->it_idm_task->idt_mutex);
1731 
1732 		resp_datalen = (task->task_sense_length == 0) ? 0 :
1733 		    (task->task_sense_length + sizeof (uint16_t));
1734 
1735 		pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
1736 		idm_pdu_init(pdu, itask->it_ict->ict_ic, itask,
1737 		    iscsit_send_status_done);
1738 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1739 
1740 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1741 		bzero(rsp, sizeof (*rsp));
1742 		rsp->opcode = ISCSI_OP_SCSI_RSP;
1743 
1744 		rsp->flags = ISCSI_FLAG_FINAL;
1745 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1746 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1747 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1748 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1749 		}
1750 
1751 		rsp->bi_residual_count = 0;
1752 		rsp->residual_count = htonl(task->task_resid);
1753 		rsp->itt = itask->it_itt;
1754 		rsp->response = ISCSI_STATUS_CMD_COMPLETED;
1755 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1756 		    htonl(itask->it_idm_task->idt_exp_datasn):
1757 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1758 		rsp->cmd_status = task->task_scsi_status;
1759 		if (task->task_sense_length != 0) {
1760 			/*
1761 			 * Add a byte to provide the sense length in
1762 			 * the response
1763 			 */
1764 			*(uint16_t *)((void *)pdu->isp_data) =
1765 			    htons(task->task_sense_length);
1766 			bcopy(task->task_sense_data,
1767 			    (uint8_t *)pdu->isp_data +
1768 			    sizeof (uint16_t),
1769 			    task->task_sense_length);
1770 			hton24(rsp->dlength, resp_datalen);
1771 		}
1772 
1773 		DTRACE_PROBE5(iscsi__scsi__response,
1774 		    iscsit_conn_t *, itask->it_ict,
1775 		    uint8_t, rsp->response,
1776 		    uint8_t, rsp->cmd_status,
1777 		    idm_pdu_t *, pdu,
1778 		    scsi_task_t *, task);
1779 
1780 		iscsit_pdu_tx(pdu);
1781 
1782 		return (STMF_SUCCESS);
1783 	}
1784 }
1785 
1786 /*ARGSUSED*/
1787 static void
1788 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status)
1789 {
1790 	iscsit_task_t	*itask;
1791 	boolean_t	aborted;
1792 
1793 	itask = pdu->isp_private;
1794 	aborted = itask->it_stmf_abort;
1795 
1796 	/*
1797 	 * After releasing the hold the task may be freed at any time so
1798 	 * don't touch it.
1799 	 */
1800 	idm_task_rele(itask->it_idm_task);
1801 	if (!aborted) {
1802 		stmf_send_status_done(itask->it_stmf_task,
1803 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1804 	}
1805 	kmem_cache_free(iscsit_status_pdu_cache, pdu);
1806 }
1807 
1808 /*ARGSUSED*/
1809 static void
1810 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status)
1811 {
1812 	iscsit_task_t	 *itask;
1813 	boolean_t	aborted;
1814 
1815 	itask = pdu->isp_private;
1816 	aborted = itask->it_stmf_abort;
1817 
1818 	/*
1819 	 * After releasing the hold the task may be freed at any time so
1820 	 * don't touch it.
1821 	 */
1822 	idm_task_rele(itask->it_idm_task);
1823 	if (!aborted) {
1824 		stmf_send_status_done(itask->it_stmf_task,
1825 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1826 	}
1827 	idm_pdu_free(pdu);
1828 }
1829 
1830 
1831 void
1832 iscsit_lport_task_free(scsi_task_t *task)
1833 {
1834 	iscsit_task_t *itask = task->task_port_private;
1835 
1836 	/* We only call idm_task_start for regular tasks, not task management */
1837 	if (task->task_mgmt_function == TM_NONE) {
1838 		idm_task_done(itask->it_idm_task);
1839 		iscsit_task_free(itask);
1840 		return;
1841 	} else {
1842 		iscsit_tm_task_free(itask);
1843 	}
1844 }
1845 
1846 /*ARGSUSED*/
1847 stmf_status_t
1848 iscsit_abort(stmf_local_port_t *lport, int abort_cmd, void *arg, uint32_t flags)
1849 {
1850 	scsi_task_t	*st = (scsi_task_t *)arg;
1851 	iscsit_task_t	*iscsit_task;
1852 	idm_task_t	*idt;
1853 
1854 	/*
1855 	 * If this is a task management request then there's really not much to
1856 	 * do.
1857 	 */
1858 	if (st->task_mgmt_function != TM_NONE) {
1859 		return (STMF_ABORT_SUCCESS);
1860 	}
1861 
1862 	/*
1863 	 * Regular task, start cleaning up
1864 	 */
1865 	iscsit_task = st->task_port_private;
1866 	idt = iscsit_task->it_idm_task;
1867 	mutex_enter(&iscsit_task->it_mutex);
1868 	iscsit_task->it_stmf_abort = B_TRUE;
1869 	if (iscsit_task->it_aborted) {
1870 		mutex_exit(&iscsit_task->it_mutex);
1871 		/*
1872 		 * Task is no longer active
1873 		 */
1874 		iscsit_task_done(iscsit_task);
1875 
1876 		/*
1877 		 * STMF specification is wrong... says to return
1878 		 * STMF_ABORTED, the code actually looks for
1879 		 * STMF_ABORT_SUCCESS.
1880 		 */
1881 		return (STMF_ABORT_SUCCESS);
1882 	} else {
1883 		mutex_exit(&iscsit_task->it_mutex);
1884 		/*
1885 		 * Call IDM to abort the task.  Due to a variety of
1886 		 * circumstances the task may already be in the process of
1887 		 * aborting.
1888 		 * We'll let IDM worry about rationalizing all that except
1889 		 * for one particular instance.  If the state of the task
1890 		 * is TASK_COMPLETE, we need to indicate to the framework
1891 		 * that we are in fact done.  This typically happens with
1892 		 * framework-initiated task management type requests
1893 		 * (e.g. abort task).
1894 		 */
1895 		if (idt->idt_state == TASK_COMPLETE) {
1896 			idm_refcnt_wait_ref(&idt->idt_refcnt);
1897 			return (STMF_ABORT_SUCCESS);
1898 		} else {
1899 			idm_task_abort(idt->idt_ic, idt, AT_TASK_MGMT_ABORT);
1900 			return (STMF_SUCCESS);
1901 		}
1902 	}
1903 
1904 	/*NOTREACHED*/
1905 }
1906 
1907 /*ARGSUSED*/
1908 void
1909 iscsit_ctl(stmf_local_port_t *lport, int cmd, void *arg)
1910 {
1911 	iscsit_tgt_t		*iscsit_tgt;
1912 
1913 	ASSERT((cmd == STMF_CMD_LPORT_ONLINE) ||
1914 	    (cmd == STMF_ACK_LPORT_ONLINE_COMPLETE) ||
1915 	    (cmd == STMF_CMD_LPORT_OFFLINE) ||
1916 	    (cmd == STMF_ACK_LPORT_OFFLINE_COMPLETE));
1917 
1918 	iscsit_tgt = (iscsit_tgt_t *)lport->lport_port_private;
1919 
1920 	switch (cmd) {
1921 	case STMF_CMD_LPORT_ONLINE:
1922 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_REQ);
1923 		break;
1924 	case STMF_CMD_LPORT_OFFLINE:
1925 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_REQ);
1926 		break;
1927 	case STMF_ACK_LPORT_ONLINE_COMPLETE:
1928 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_COMPLETE_ACK);
1929 		break;
1930 	case STMF_ACK_LPORT_OFFLINE_COMPLETE:
1931 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_COMPLETE_ACK);
1932 		break;
1933 
1934 	default:
1935 		break;
1936 	}
1937 }
1938 
1939 static stmf_status_t
1940 iscsit_idm_to_stmf(idm_status_t idmrc)
1941 {
1942 	switch (idmrc) {
1943 	case IDM_STATUS_SUCCESS:
1944 		return (STMF_SUCCESS);
1945 	default:
1946 		return (STMF_FAILURE);
1947 	}
1948 	/*NOTREACHED*/
1949 }
1950 
1951 void
1952 iscsit_op_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1953 {
1954 	iscsit_conn_t		*ict = ic->ic_handle;
1955 
1956 	if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
1957 		iscsit_post_scsi_cmd(ic, rx_pdu);
1958 	}
1959 	iscsit_process_pdu_in_queue(ict->ict_sess);
1960 }
1961 
1962 /*
1963  * ISCSI protocol
1964  */
1965 
1966 void
1967 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1968 {
1969 	iscsit_conn_t		*ict;
1970 	iscsit_task_t		*itask;
1971 	scsi_task_t		*task;
1972 	iscsit_buf_t		*ibuf;
1973 	iscsi_scsi_cmd_hdr_t	*iscsi_scsi =
1974 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
1975 	iscsi_addl_hdr_t	*ahs_hdr;
1976 	uint16_t		addl_cdb_len = 0;
1977 
1978 	ict = ic->ic_handle;
1979 
1980 	itask = iscsit_task_alloc(ict);
1981 	if (itask == NULL) {
1982 		/* Finish processing request */
1983 		iscsit_set_cmdsn(ict, rx_pdu);
1984 
1985 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
1986 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
1987 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
1988 		return;
1989 	}
1990 
1991 	/*
1992 	 * Note CmdSN and ITT in task.  IDM will have already validated this
1993 	 * request against the connection state so we don't need to check
1994 	 * that (the connection may have changed state in the meantime but
1995 	 * we will catch that when we try to send a response)
1996 	 */
1997 	itask->it_cmdsn = ntohl(iscsi_scsi->cmdsn);
1998 	itask->it_itt = iscsi_scsi->itt;
1999 
2000 	/*
2001 	 * Check for extended CDB AHS
2002 	 */
2003 	if (iscsi_scsi->hlength > 0) {
2004 		ahs_hdr = (iscsi_addl_hdr_t *)iscsi_scsi;
2005 		addl_cdb_len = ((ahs_hdr->ahs_hlen_hi << 8) |
2006 		    ahs_hdr->ahs_hlen_lo) - 1; /* Adjust for reserved byte */
2007 		if (((addl_cdb_len + 4) / sizeof (uint32_t)) >
2008 		    iscsi_scsi->hlength) {
2009 			/* Mangled header info, drop it */
2010 			idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2011 			return;
2012 		}
2013 	}
2014 
2015 	ict = rx_pdu->isp_ic->ic_handle; /* IDM client private */
2016 
2017 	/*
2018 	 * Add task to session list.  This function will also check to
2019 	 * ensure that the task does not already exist.
2020 	 */
2021 	if (iscsit_task_start(itask) != IDM_STATUS_SUCCESS) {
2022 		/*
2023 		 * Task exists, free all resources and reject.  Don't
2024 		 * update expcmdsn in this case because RFC 3720 says
2025 		 * "The CmdSN of the rejected command PDU (if it is a
2026 		 * non-immediate command) MUST NOT be considered received
2027 		 * by the target (i.e., a command sequence gap must be
2028 		 * assumed for the CmdSN), even though the CmdSN of the
2029 		 * rejected command PDU may be reliably ascertained.  Upon
2030 		 * receiving the Reject, the initiator MUST plug the CmdSN
2031 		 * gap in order to continue to use the session.  The gap
2032 		 * may be plugged either by transmitting a command PDU
2033 		 * with the same CmdSN, or by aborting the task (see section
2034 		 * 6.9 on how an abort may plug a CmdSN gap)." (Section 6.3)
2035 		 */
2036 		iscsit_task_free(itask);
2037 		iscsit_send_reject(ict, rx_pdu, ISCSI_REJECT_TASK_IN_PROGRESS);
2038 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2039 		return;
2040 	}
2041 
2042 	/* Update sequence numbers */
2043 	iscsit_set_cmdsn(ict, rx_pdu);
2044 
2045 	/*
2046 	 * Allocate STMF task
2047 	 */
2048 	itask->it_stmf_task = stmf_task_alloc(
2049 	    itask->it_ict->ict_sess->ist_lport,
2050 	    itask->it_ict->ict_sess->ist_stmf_sess, iscsi_scsi->lun,
2051 	    16 + addl_cdb_len, 0);
2052 	if (itask->it_stmf_task == NULL) {
2053 		/*
2054 		 * Either stmf really couldn't get memory for a task or,
2055 		 * more likely, the LU is currently in reset.  Either way
2056 		 * we have no choice but to fail the request.
2057 		 */
2058 		iscsit_task_done(itask);
2059 		iscsit_task_free(itask);
2060 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
2061 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
2062 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2063 		return;
2064 	}
2065 
2066 	task = itask->it_stmf_task;
2067 	task->task_port_private = itask;
2068 
2069 	bcopy(iscsi_scsi->lun, task->task_lun_no, sizeof (task->task_lun_no));
2070 
2071 	/*
2072 	 * iSCSI and Comstar use the same values.  Should we rely on this
2073 	 * or translate them bit-wise?
2074 	 */
2075 
2076 	task->task_flags =
2077 	    (((iscsi_scsi->flags & ISCSI_FLAG_CMD_READ) ? TF_READ_DATA : 0) |
2078 	    ((iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ? TF_WRITE_DATA : 0) |
2079 	    ((rx_pdu->isp_datalen == 0) ? 0 : TF_INITIAL_BURST));
2080 
2081 	switch (iscsi_scsi->flags & ISCSI_FLAG_CMD_ATTR_MASK) {
2082 	case ISCSI_ATTR_UNTAGGED:
2083 		break;
2084 	case ISCSI_ATTR_SIMPLE:
2085 		task->task_additional_flags |= TF_ATTR_SIMPLE_QUEUE;
2086 		break;
2087 	case ISCSI_ATTR_ORDERED:
2088 		task->task_additional_flags |= TF_ATTR_ORDERED_QUEUE;
2089 		break;
2090 	case ISCSI_ATTR_HEAD_OF_QUEUE:
2091 		task->task_additional_flags |= TF_ATTR_HEAD_OF_QUEUE;
2092 		break;
2093 	case ISCSI_ATTR_ACA:
2094 		task->task_additional_flags |= TF_ATTR_ACA;
2095 		break;
2096 	default:
2097 		/* Protocol error but just take it, treat as untagged */
2098 		break;
2099 	}
2100 
2101 
2102 	task->task_additional_flags = 0;
2103 	task->task_priority = 0;
2104 	task->task_mgmt_function = TM_NONE;
2105 
2106 	/*
2107 	 * This "task_max_nbufs" doesn't map well to BIDI.  We probably need
2108 	 * parameter for each direction.  "MaxOutstandingR2T" may very well
2109 	 * be set to one which could prevent us from doing simultaneous
2110 	 * transfers in each direction.
2111 	 */
2112 	task->task_max_nbufs = (iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ?
2113 	    ict->ict_op.op_max_outstanding_r2t : STMF_BUFS_MAX;
2114 	task->task_cmd_seq_no = ntohl(iscsi_scsi->itt);
2115 	task->task_expected_xfer_length = ntohl(iscsi_scsi->data_length);
2116 
2117 	/* Copy CDB */
2118 	bcopy(iscsi_scsi->scb, task->task_cdb, 16);
2119 	if (addl_cdb_len > 0) {
2120 		bcopy(ahs_hdr->ahs_extscb, task->task_cdb + 16, addl_cdb_len);
2121 	}
2122 
2123 	DTRACE_ISCSI_3(scsi__command, idm_conn_t *, ic,
2124 	    iscsi_scsi_cmd_hdr_t *, (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr,
2125 	    scsi_task_t *, task);
2126 
2127 	/*
2128 	 * Copy the transport header into the task handle from the PDU
2129 	 * handle. The transport header describes this task's remote tagged
2130 	 * buffer.
2131 	 */
2132 	if (rx_pdu->isp_transport_hdrlen != 0) {
2133 		bcopy(rx_pdu->isp_transport_hdr,
2134 		    itask->it_idm_task->idt_transport_hdr,
2135 		    rx_pdu->isp_transport_hdrlen);
2136 	}
2137 
2138 	/*
2139 	 * Tell IDM about our new active task
2140 	 */
2141 	idm_task_start(itask->it_idm_task, (uintptr_t)itask->it_itt);
2142 
2143 	/*
2144 	 * If we have any immediate data then setup the immediate buffer
2145 	 * context that comes with the task
2146 	 */
2147 	if (rx_pdu->isp_datalen) {
2148 		ibuf = itask->it_immed_data;
2149 		ibuf->ibuf_immed_data_pdu = rx_pdu;
2150 		ibuf->ibuf_stmf_buf->db_data_size = rx_pdu->isp_datalen;
2151 		ibuf->ibuf_stmf_buf->db_buf_size = rx_pdu->isp_datalen;
2152 		ibuf->ibuf_stmf_buf->db_relative_offset = 0;
2153 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_length =
2154 		    rx_pdu->isp_datalen;
2155 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr = rx_pdu->isp_data;
2156 
2157 		DTRACE_ISCSI_8(xfer__start, idm_conn_t *, ic,
2158 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2159 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2160 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2161 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2162 
2163 		/*
2164 		 * For immediate data transfer, there is no callback from
2165 		 * stmf to indicate that the initial burst of data is
2166 		 * transferred successfully. In some cases, the task can
2167 		 * get freed before execution returns from stmf_post_task.
2168 		 * Although this xfer-start/done probe accurately tracks
2169 		 * the size of the transfer, it does only provide a best
2170 		 * effort on the timing of the transfer.
2171 		 */
2172 		DTRACE_ISCSI_8(xfer__done, idm_conn_t *, ic,
2173 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2174 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2175 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2176 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2177 		stmf_post_task(task, ibuf->ibuf_stmf_buf);
2178 	} else {
2179 
2180 		stmf_post_task(task, NULL);
2181 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2182 	}
2183 }
2184 
2185 void
2186 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu)
2187 {
2188 	iscsit_conn_t *ict = rx_pdu->isp_ic->ic_handle;
2189 
2190 	/*
2191 	 * If this isn't a login packet, we need a session.  Otherwise
2192 	 * this is a protocol error (perhaps one IDM should've caught?).
2193 	 */
2194 	if (IDM_PDU_OPCODE(rx_pdu) != ISCSI_OP_LOGIN_CMD &&
2195 	    ict->ict_sess == NULL) {
2196 		DTRACE_PROBE2(iscsi__idm__deferred__no__session,
2197 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
2198 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2199 		return;
2200 	}
2201 
2202 	/*
2203 	 * If the connection has been lost then ignore new PDU's
2204 	 */
2205 	mutex_enter(&ict->ict_mutex);
2206 	if (ict->ict_lost) {
2207 		mutex_exit(&ict->ict_mutex);
2208 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2209 		return;
2210 	}
2211 
2212 	/*
2213 	 * Grab a hold on the connection to prevent it from going away
2214 	 * between now and when the taskq function is called.
2215 	 */
2216 	iscsit_conn_dispatch_hold(ict);
2217 	mutex_exit(&ict->ict_mutex);
2218 
2219 	taskq_dispatch_ent(iscsit_global.global_dispatch_taskq,
2220 	    iscsit_deferred, rx_pdu, 0, &rx_pdu->isp_tqent);
2221 }
2222 
2223 static void
2224 iscsit_deferred(void *rx_pdu_void)
2225 {
2226 	idm_pdu_t		*rx_pdu = rx_pdu_void;
2227 	idm_conn_t		*ic = rx_pdu->isp_ic;
2228 	iscsit_conn_t		*ict = ic->ic_handle;
2229 
2230 	/*
2231 	 * NOP and Task Management Commands can be marked for immediate
2232 	 * delivery. Commands marked as 'Immediate' are to be considered
2233 	 * for execution as soon as they arrive on the target. So these
2234 	 * should not be checked for sequence order and put in a queue.
2235 	 * The CmdSN is not advanced for Immediate Commands.
2236 	 */
2237 	switch (IDM_PDU_OPCODE(rx_pdu)) {
2238 	case ISCSI_OP_NOOP_OUT:
2239 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2240 			iscsit_set_cmdsn(ict, rx_pdu);
2241 			iscsit_pdu_op_noop(ict, rx_pdu);
2242 		}
2243 		break;
2244 	case ISCSI_OP_LOGIN_CMD:
2245 		iscsit_pdu_op_login_cmd(ict, rx_pdu);
2246 		iscsit_conn_dispatch_rele(ict);
2247 		return;
2248 	case ISCSI_OP_TEXT_CMD:
2249 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2250 			iscsit_set_cmdsn(ict, rx_pdu);
2251 			iscsit_pdu_op_text_cmd(ict, rx_pdu);
2252 		}
2253 		break;
2254 	case ISCSI_OP_LOGOUT_CMD:
2255 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2256 			iscsit_set_cmdsn(ict, rx_pdu);
2257 			iscsit_pdu_op_logout_cmd(ict, rx_pdu);
2258 		}
2259 		break;
2260 	default:
2261 		/* Protocol error.  IDM should have caught this */
2262 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2263 		ASSERT(0);
2264 		break;
2265 	}
2266 	/*
2267 	 * Check if there are other PDUs in the session staging queue
2268 	 * waiting to be posted to SCSI layer.
2269 	 */
2270 	iscsit_process_pdu_in_queue(ict->ict_sess);
2271 
2272 	iscsit_conn_dispatch_rele(ict);
2273 }
2274 
2275 static void
2276 iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
2277     uint8_t response, uint8_t cmd_status)
2278 {
2279 	idm_pdu_t			*rsp_pdu;
2280 	idm_conn_t			*ic;
2281 	iscsi_scsi_rsp_hdr_t		*resp;
2282 	iscsi_scsi_cmd_hdr_t		*req =
2283 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2284 
2285 	ic = ict->ict_ic;
2286 
2287 	rsp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_rsp_hdr_t), 0);
2288 	idm_pdu_init(rsp_pdu, ic, NULL, NULL);
2289 	/*
2290 	 * StatSN is incremented by 1 for every response sent on
2291 	 * a connection except for responses sent as a result of
2292 	 * a retry or SNACK
2293 	 */
2294 	rsp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2295 
2296 	resp = (iscsi_scsi_rsp_hdr_t *)rsp_pdu->isp_hdr;
2297 
2298 	resp->opcode = ISCSI_OP_SCSI_RSP;
2299 	resp->flags = ISCSI_FLAG_FINAL;
2300 	resp->response = response;
2301 	resp->cmd_status = cmd_status;
2302 	resp->itt = req->itt;
2303 	if ((response == ISCSI_STATUS_CMD_COMPLETED) &&
2304 	    (req->data_length != 0) &&
2305 	    ((req->flags & ISCSI_FLAG_CMD_READ) ||
2306 	    (req->flags & ISCSI_FLAG_CMD_WRITE))) {
2307 		resp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
2308 		resp->residual_count = req->data_length;
2309 	}
2310 
2311 	DTRACE_PROBE4(iscsi__scsi__direct__response,
2312 	    iscsit_conn_t *, ict,
2313 	    uint8_t, resp->response,
2314 	    uint8_t, resp->cmd_status,
2315 	    idm_pdu_t *, rsp_pdu);
2316 
2317 	iscsit_pdu_tx(rsp_pdu);
2318 }
2319 
2320 void
2321 iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu, uint8_t tm_status)
2322 {
2323 	iscsi_scsi_task_mgt_rsp_hdr_t	*tm_resp;
2324 
2325 	/*
2326 	 * The target must take note of the last-sent StatSN.
2327 	 * The StatSN is to be incremented after sending a
2328 	 * task management response. Digest recovery can only
2329 	 * work if StatSN is incremented.
2330 	 */
2331 	tm_resp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2332 	tm_resp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2333 	tm_resp->response = tm_status;
2334 
2335 	DTRACE_PROBE3(iscsi__scsi__tm__response,
2336 	    iscsit_conn_t *, tm_resp_pdu->isp_ic->ic_handle,
2337 	    uint8_t, tm_resp->response,
2338 	    idm_pdu_t *, tm_resp_pdu);
2339 	iscsit_pdu_tx(tm_resp_pdu);
2340 }
2341 
2342 void
2343 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2344 {
2345 	idm_pdu_t			*tm_resp_pdu;
2346 	iscsit_task_t			*itask;
2347 	iscsit_task_t			*tm_itask;
2348 	scsi_task_t			*task;
2349 	iscsi_scsi_task_mgt_hdr_t 	*iscsi_tm =
2350 	    (iscsi_scsi_task_mgt_hdr_t *)rx_pdu->isp_hdr;
2351 	iscsi_scsi_task_mgt_rsp_hdr_t 	*iscsi_tm_rsp =
2352 	    (iscsi_scsi_task_mgt_rsp_hdr_t *)rx_pdu->isp_hdr;
2353 	uint32_t			rtt, cmdsn, refcmdsn;
2354 	uint8_t				tm_func;
2355 
2356 	/*
2357 	 * Setup response PDU (response field will get filled in later)
2358 	 */
2359 	tm_resp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_task_mgt_rsp_hdr_t), 0);
2360 	if (tm_resp_pdu == NULL) {
2361 		/* Can't respond, just drop it */
2362 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2363 		return;
2364 	}
2365 	idm_pdu_init(tm_resp_pdu, ict->ict_ic, NULL, NULL);
2366 	iscsi_tm_rsp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2367 	bzero(iscsi_tm_rsp, sizeof (iscsi_scsi_task_mgt_rsp_hdr_t));
2368 	iscsi_tm_rsp->opcode = ISCSI_OP_SCSI_TASK_MGT_RSP;
2369 	iscsi_tm_rsp->flags = ISCSI_FLAG_FINAL;
2370 	iscsi_tm_rsp->itt = rx_pdu->isp_hdr->itt;
2371 
2372 	/*
2373 	 * Figure out what we're being asked to do.
2374 	 */
2375 	DTRACE_PROBE4(iscsi__scsi__tm__request,
2376 	    iscsit_conn_t *, ict,
2377 	    uint8_t, (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK),
2378 	    uint32_t, iscsi_tm->rtt,
2379 	    idm_pdu_t *, rx_pdu);
2380 	switch (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK) {
2381 	case ISCSI_TM_FUNC_ABORT_TASK:
2382 		/*
2383 		 * STMF doesn't currently support the "abort task" task
2384 		 * management command although it does support aborting
2385 		 * an individual task.  We'll get STMF to abort the task
2386 		 * for us but handle the details of the task management
2387 		 * command ourselves.
2388 		 *
2389 		 * Find the task associated with the referenced task tag.
2390 		 */
2391 		rtt = iscsi_tm->rtt;
2392 		itask = (iscsit_task_t *)idm_task_find_by_handle(ict->ict_ic,
2393 		    (uintptr_t)rtt);
2394 
2395 		if (itask == NULL) {
2396 			cmdsn = ntohl(iscsi_tm->cmdsn);
2397 			refcmdsn = ntohl(iscsi_tm->refcmdsn);
2398 
2399 			/*
2400 			 * Task was not found. But the SCSI command could be
2401 			 * on the rxpdu wait queue. If RefCmdSN is within
2402 			 * the CmdSN window and less than CmdSN of the TM
2403 			 * function, return "Function Complete". Otherwise,
2404 			 * return "Task Does Not Exist".
2405 			 */
2406 
2407 			if (iscsit_cmdsn_in_window(ict, refcmdsn) &&
2408 			    iscsit_sna_lt(refcmdsn, cmdsn)) {
2409 				mutex_enter(&ict->ict_sess->ist_sn_mutex);
2410 				if (iscsit_remove_pdu_from_queue(
2411 				    ict->ict_sess, refcmdsn)) {
2412 					iscsit_conn_dispatch_rele(ict);
2413 				}
2414 				mutex_exit(&ict->ict_sess->ist_sn_mutex);
2415 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2416 				    SCSI_TCP_TM_RESP_COMPLETE);
2417 			} else {
2418 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2419 				    SCSI_TCP_TM_RESP_NO_TASK);
2420 			}
2421 		} else {
2422 
2423 			/*
2424 			 * Tell STMF to abort the task.  This will do no harm
2425 			 * if the task is already complete.
2426 			 */
2427 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
2428 			    STMF_ABORTED, NULL);
2429 
2430 			/*
2431 			 * Make sure the task hasn't already completed
2432 			 */
2433 			mutex_enter(&itask->it_idm_task->idt_mutex);
2434 			if ((itask->it_idm_task->idt_state == TASK_COMPLETE) ||
2435 			    (itask->it_idm_task->idt_state == TASK_IDLE)) {
2436 				/*
2437 				 * Task is complete, return "Task Does Not
2438 				 * Exist"
2439 				 */
2440 				mutex_exit(&itask->it_idm_task->idt_mutex);
2441 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2442 				    SCSI_TCP_TM_RESP_NO_TASK);
2443 			} else {
2444 				/*
2445 				 * STMF is now aborting the task, return
2446 				 * "Function Complete"
2447 				 */
2448 				mutex_exit(&itask->it_idm_task->idt_mutex);
2449 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2450 				    SCSI_TCP_TM_RESP_COMPLETE);
2451 			}
2452 			idm_task_rele(itask->it_idm_task);
2453 		}
2454 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2455 		return;
2456 
2457 	case ISCSI_TM_FUNC_ABORT_TASK_SET:
2458 		tm_func = TM_ABORT_TASK_SET;
2459 		break;
2460 
2461 	case ISCSI_TM_FUNC_CLEAR_ACA:
2462 		tm_func = TM_CLEAR_ACA;
2463 		break;
2464 
2465 	case ISCSI_TM_FUNC_CLEAR_TASK_SET:
2466 		tm_func = TM_CLEAR_TASK_SET;
2467 		break;
2468 
2469 	case ISCSI_TM_FUNC_LOGICAL_UNIT_RESET:
2470 		tm_func = TM_LUN_RESET;
2471 		break;
2472 
2473 	case ISCSI_TM_FUNC_TARGET_WARM_RESET:
2474 		tm_func = TM_TARGET_WARM_RESET;
2475 		break;
2476 
2477 	case ISCSI_TM_FUNC_TARGET_COLD_RESET:
2478 		tm_func = TM_TARGET_COLD_RESET;
2479 		break;
2480 
2481 	case ISCSI_TM_FUNC_TASK_REASSIGN:
2482 		/*
2483 		 * We do not currently support allegiance reassignment.  When
2484 		 * we start supporting ERL1+, we will need to.
2485 		 */
2486 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2487 		    SCSI_TCP_TM_RESP_NO_ALLG_REASSN);
2488 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2489 		return;
2490 
2491 	default:
2492 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2493 		    SCSI_TCP_TM_RESP_REJECTED);
2494 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2495 		return;
2496 	}
2497 
2498 	tm_itask = iscsit_tm_task_alloc(ict);
2499 	if (tm_itask == NULL) {
2500 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2501 		    SCSI_TCP_TM_RESP_REJECTED);
2502 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2503 		return;
2504 	}
2505 
2506 
2507 	task = stmf_task_alloc(ict->ict_sess->ist_lport,
2508 	    ict->ict_sess->ist_stmf_sess, iscsi_tm->lun,
2509 	    0, STMF_TASK_EXT_NONE);
2510 	if (task == NULL) {
2511 		/*
2512 		 * If this happens, either the LU is in reset, couldn't
2513 		 * get memory, or some other condition in which we simply
2514 		 * can't complete this request.  It would be nice to return
2515 		 * an error code like "busy" but the closest we have is
2516 		 * "rejected".
2517 		 */
2518 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2519 		    SCSI_TCP_TM_RESP_REJECTED);
2520 		iscsit_tm_task_free(tm_itask);
2521 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2522 		return;
2523 	}
2524 
2525 	tm_itask->it_tm_pdu = tm_resp_pdu;
2526 	tm_itask->it_stmf_task = task;
2527 	task->task_port_private = tm_itask;
2528 	task->task_mgmt_function = tm_func;
2529 	task->task_additional_flags = TASK_AF_NO_EXPECTED_XFER_LENGTH;
2530 	task->task_priority = 0;
2531 	task->task_max_nbufs = STMF_BUFS_MAX;
2532 	task->task_cmd_seq_no = iscsi_tm->itt;
2533 	task->task_expected_xfer_length = 0;
2534 
2535 	stmf_post_task(task, NULL);
2536 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2537 }
2538 
2539 static void
2540 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2541 {
2542 	iscsi_nop_out_hdr_t *out = (iscsi_nop_out_hdr_t *)rx_pdu->isp_hdr;
2543 	iscsi_nop_in_hdr_t *in;
2544 	int resp_datalen;
2545 	idm_pdu_t *resp;
2546 
2547 	/* Ignore the response from initiator */
2548 	if ((out->itt == ISCSI_RSVD_TASK_TAG) ||
2549 	    (out->ttt != ISCSI_RSVD_TASK_TAG)) {
2550 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2551 		return;
2552 	}
2553 
2554 	/* Allocate a PDU to respond */
2555 	resp_datalen = ntoh24(out->dlength);
2556 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
2557 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2558 	if (resp_datalen > 0) {
2559 		bcopy(rx_pdu->isp_data, resp->isp_data, resp_datalen);
2560 	}
2561 
2562 	/*
2563 	 * When sending a NOP-In as a response to a NOP-Out from the initiator,
2564 	 * the target must respond with the same initiator task tag that was
2565 	 * provided in the NOP-Out request, the target transfer tag must be
2566 	 * ISCSI_RSVD_TASK_TAG (0xffffffff) and StatSN will contain the next
2567 	 * status sequence number. The StatSN for the connection is advanced
2568 	 * after this PDU is sent.
2569 	 */
2570 	in = (iscsi_nop_in_hdr_t *)resp->isp_hdr;
2571 	bzero(in, sizeof (*in));
2572 	in->opcode = ISCSI_OP_NOOP_IN;
2573 	in->flags = ISCSI_FLAG_FINAL;
2574 	bcopy(out->lun, in->lun, 8);
2575 	in->itt		= out->itt;
2576 	in->ttt		= ISCSI_RSVD_TASK_TAG;
2577 	hton24(in->dlength, resp_datalen);
2578 	resp->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2579 	/* Any other field in resp to be set? */
2580 	iscsit_pdu_tx(resp);
2581 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2582 }
2583 
2584 static void
2585 iscsit_pdu_op_login_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2586 {
2587 
2588 	/*
2589 	 * Submit PDU to login state machine.  State machine will free the
2590 	 * PDU.
2591 	 */
2592 	iscsit_login_sm_event(ict, ILE_LOGIN_RCV, rx_pdu);
2593 }
2594 
2595 void
2596 iscsit_pdu_op_logout_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2597 {
2598 	iscsi_logout_hdr_t 	*logout_req =
2599 	    (iscsi_logout_hdr_t *)rx_pdu->isp_hdr;
2600 	iscsi_logout_rsp_hdr_t	*logout_rsp;
2601 	idm_pdu_t *resp;
2602 
2603 	/* Allocate a PDU to respond */
2604 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2605 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2606 	/*
2607 	 * The StatSN is to be sent to the initiator,
2608 	 * it is not required to increment the number
2609 	 * as the connection is terminating.
2610 	 */
2611 	resp->isp_flags |= IDM_PDU_SET_STATSN;
2612 	/*
2613 	 * Logout results in the immediate termination of all tasks except
2614 	 * if the logout reason is ISCSI_LOGOUT_REASON_RECOVERY.  The
2615 	 * connection state machine will drive this task cleanup automatically
2616 	 * so we don't need to handle that here.
2617 	 */
2618 	logout_rsp = (iscsi_logout_rsp_hdr_t *)resp->isp_hdr;
2619 	bzero(logout_rsp, sizeof (*logout_rsp));
2620 	logout_rsp->opcode = ISCSI_OP_LOGOUT_RSP;
2621 	logout_rsp->flags = ISCSI_FLAG_FINAL;
2622 	logout_rsp->itt = logout_req->itt;
2623 	if ((logout_req->flags & ISCSI_FLAG_LOGOUT_REASON_MASK) >
2624 	    ISCSI_LOGOUT_REASON_RECOVERY) {
2625 		logout_rsp->response = ISCSI_LOGOUT_RECOVERY_UNSUPPORTED;
2626 	} else {
2627 		logout_rsp->response = ISCSI_LOGOUT_SUCCESS;
2628 	}
2629 
2630 	iscsit_pdu_tx(resp);
2631 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2632 }
2633 
2634 /*
2635  * Calculate the number of outstanding commands we can process
2636  */
2637 int
2638 iscsit_cmd_window()
2639 {
2640 	/*
2641 	 * Instead of using a pre-defined constant for the command window,
2642 	 * it should be made confiurable and dynamic. With MC/S, sequence
2643 	 * numbers will be used up at a much faster rate than with SC/S.
2644 	 */
2645 	return	(ISCSIT_MAX_WINDOW);
2646 }
2647 
2648 /*
2649  * Set local registers based on incoming PDU
2650  */
2651 void
2652 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2653 {
2654 	iscsit_sess_t *ist;
2655 	iscsi_scsi_cmd_hdr_t *req;
2656 
2657 	ist = ict->ict_sess;
2658 
2659 	req = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2660 	if (req->opcode & ISCSI_OP_IMMEDIATE) {
2661 		/* no cmdsn increment for immediate PDUs */
2662 		return;
2663 	}
2664 
2665 	/* Ensure that the ExpCmdSN advances in an orderly manner */
2666 	mutex_enter(&ist->ist_sn_mutex);
2667 	ist->ist_expcmdsn = ntohl(req->cmdsn) + 1;
2668 	ist->ist_maxcmdsn = ntohl(req->cmdsn) + iscsit_cmd_window();
2669 	mutex_exit(&ist->ist_sn_mutex);
2670 }
2671 
2672 /*
2673  * Wrapper funtion, calls iscsi_calc_rspsn and idm_pdu_tx
2674  */
2675 void
2676 iscsit_pdu_tx(idm_pdu_t *pdu)
2677 {
2678 	iscsit_conn_t *ict = pdu->isp_ic->ic_handle;
2679 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2680 	iscsit_sess_t *ist = ict->ict_sess;
2681 
2682 	/*
2683 	 * The command sequence numbers are session-wide and must stay
2684 	 * consistent across the transfer, so protect the cmdsn with a
2685 	 * mutex lock on the session. The status sequence number will
2686 	 * be updated just before the transport layer transmits the PDU.
2687 	 */
2688 
2689 	mutex_enter(&ict->ict_sess->ist_sn_mutex);
2690 	/* Set ExpCmdSN and MaxCmdSN */
2691 	rsp->maxcmdsn = htonl(ist->ist_maxcmdsn);
2692 	rsp->expcmdsn = htonl(ist->ist_expcmdsn);
2693 	idm_pdu_tx(pdu);
2694 	mutex_exit(&ict->ict_sess->ist_sn_mutex);
2695 }
2696 
2697 /*
2698  * Internal functions
2699  */
2700 
2701 void
2702 iscsit_send_async_event(iscsit_conn_t *ict, uint8_t event)
2703 {
2704 	idm_pdu_t		*abt;
2705 	iscsi_async_evt_hdr_t	*async_abt;
2706 
2707 	/*
2708 	 * Get a PDU to build the abort request.
2709 	 */
2710 	abt = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2711 	if (abt == NULL) {
2712 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2713 		return;
2714 	}
2715 
2716 	/*
2717 	 * A asynchronous message is sent by the target to request a logout.
2718 	 * The StatSN for the connection is advanced after the PDU is sent
2719 	 * to allow for initiator and target state synchronization.
2720 	 */
2721 	idm_pdu_init(abt, ict->ict_ic, NULL, NULL);
2722 	abt->isp_datalen = 0;
2723 	abt->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2724 
2725 	async_abt = (iscsi_async_evt_hdr_t *)abt->isp_hdr;
2726 	bzero(async_abt, sizeof (*async_abt));
2727 	async_abt->opcode = ISCSI_OP_ASYNC_EVENT;
2728 	async_abt->async_event = event;
2729 	async_abt->flags = ISCSI_FLAG_FINAL;
2730 	async_abt->rsvd4[0] = 0xff;
2731 	async_abt->rsvd4[1] = 0xff;
2732 	async_abt->rsvd4[2] = 0xff;
2733 	async_abt->rsvd4[3] = 0xff;
2734 
2735 	switch (event) {
2736 	case ISCSI_ASYNC_EVENT_REQUEST_LOGOUT:
2737 		async_abt->param3 = htons(IDM_LOGOUT_SECONDS);
2738 		break;
2739 	case ISCSI_ASYNC_EVENT_SCSI_EVENT:
2740 	case ISCSI_ASYNC_EVENT_DROPPING_CONNECTION:
2741 	case ISCSI_ASYNC_EVENT_DROPPING_ALL_CONNECTIONS:
2742 	case ISCSI_ASYNC_EVENT_PARAM_NEGOTIATION:
2743 	default:
2744 		ASSERT(0);
2745 	}
2746 
2747 	iscsit_pdu_tx(abt);
2748 }
2749 
2750 void
2751 iscsit_send_reject(iscsit_conn_t *ict, idm_pdu_t *rejected_pdu, uint8_t reason)
2752 {
2753 	idm_pdu_t		*reject_pdu;
2754 	iscsi_reject_rsp_hdr_t	*reject;
2755 
2756 	/*
2757 	 * Get a PDU to build the abort request.
2758 	 */
2759 	reject_pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t),
2760 	    rejected_pdu->isp_hdrlen);
2761 	if (reject_pdu == NULL) {
2762 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2763 		return;
2764 	}
2765 	idm_pdu_init(reject_pdu, ict->ict_ic, NULL, NULL);
2766 	/* StatSN is advanced after a Reject PDU */
2767 	reject_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2768 	reject_pdu->isp_datalen = rejected_pdu->isp_hdrlen;
2769 	bcopy(rejected_pdu->isp_hdr, reject_pdu->isp_data,
2770 	    rejected_pdu->isp_hdrlen);
2771 
2772 	reject = (iscsi_reject_rsp_hdr_t *)reject_pdu->isp_hdr;
2773 	bzero(reject, sizeof (*reject));
2774 	reject->opcode = ISCSI_OP_REJECT_MSG;
2775 	reject->reason = reason;
2776 	reject->flags = ISCSI_FLAG_FINAL;
2777 	hton24(reject->dlength, rejected_pdu->isp_hdrlen);
2778 	reject->must_be_ff[0] = 0xff;
2779 	reject->must_be_ff[1] = 0xff;
2780 	reject->must_be_ff[2] = 0xff;
2781 	reject->must_be_ff[3] = 0xff;
2782 
2783 	iscsit_pdu_tx(reject_pdu);
2784 }
2785 
2786 
2787 static iscsit_task_t *
2788 iscsit_task_alloc(iscsit_conn_t *ict)
2789 {
2790 	iscsit_task_t *itask;
2791 	iscsit_buf_t *immed_ibuf;
2792 
2793 	/*
2794 	 * Possible items to pre-alloc if we cache iscsit_task_t's:
2795 	 *
2796 	 * Status PDU w/ sense buffer
2797 	 * stmf_data_buf_t for immediate data
2798 	 */
2799 	itask = kmem_alloc(sizeof (iscsit_task_t) + sizeof (iscsit_buf_t) +
2800 	    sizeof (stmf_data_buf_t), KM_NOSLEEP);
2801 	if (itask != NULL) {
2802 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2803 		itask->it_aborted = itask->it_stmf_abort =
2804 		    itask->it_tm_task = 0;
2805 
2806 		immed_ibuf = (iscsit_buf_t *)(itask + 1);
2807 		bzero(immed_ibuf, sizeof (*immed_ibuf));
2808 		immed_ibuf->ibuf_is_immed = B_TRUE;
2809 		immed_ibuf->ibuf_stmf_buf = (stmf_data_buf_t *)(immed_ibuf + 1);
2810 
2811 		bzero(immed_ibuf->ibuf_stmf_buf, sizeof (stmf_data_buf_t));
2812 		immed_ibuf->ibuf_stmf_buf->db_port_private = immed_ibuf;
2813 		immed_ibuf->ibuf_stmf_buf->db_sglist_length = 1;
2814 		immed_ibuf->ibuf_stmf_buf->db_flags = DB_DIRECTION_FROM_RPORT |
2815 		    DB_DONT_CACHE;
2816 		itask->it_immed_data = immed_ibuf;
2817 		itask->it_idm_task = idm_task_alloc(ict->ict_ic);
2818 		if (itask->it_idm_task != NULL) {
2819 			itask->it_idm_task->idt_private = itask;
2820 			itask->it_ict = ict;
2821 			itask->it_ttt = itask->it_idm_task->idt_tt;
2822 			return (itask);
2823 		} else {
2824 			kmem_free(itask, sizeof (iscsit_task_t) +
2825 			    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2826 		}
2827 	}
2828 
2829 	return (NULL);
2830 }
2831 
2832 static void
2833 iscsit_task_free(iscsit_task_t *itask)
2834 {
2835 	idm_task_free(itask->it_idm_task);
2836 	mutex_destroy(&itask->it_mutex);
2837 	kmem_free(itask, sizeof (iscsit_task_t) +
2838 	    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2839 }
2840 
2841 static iscsit_task_t *
2842 iscsit_tm_task_alloc(iscsit_conn_t *ict)
2843 {
2844 	iscsit_task_t *itask;
2845 
2846 	itask = kmem_zalloc(sizeof (iscsit_task_t), KM_NOSLEEP);
2847 	if (itask != NULL) {
2848 		idm_conn_hold(ict->ict_ic);
2849 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2850 		itask->it_aborted = itask->it_stmf_abort =
2851 		    itask->it_tm_responded = 0;
2852 		itask->it_tm_pdu = NULL;
2853 		itask->it_tm_task = 1;
2854 		itask->it_ict = ict;
2855 	}
2856 
2857 	return (itask);
2858 }
2859 
2860 static void
2861 iscsit_tm_task_free(iscsit_task_t *itask)
2862 {
2863 	/*
2864 	 * If we responded then the call to idm_pdu_complete will free the
2865 	 * PDU.  Otherwise we got aborted before the TM function could
2866 	 * complete and we need to free the PDU explicitly.
2867 	 */
2868 	if (itask->it_tm_pdu != NULL && !itask->it_tm_responded)
2869 		idm_pdu_free(itask->it_tm_pdu);
2870 	idm_conn_rele(itask->it_ict->ict_ic);
2871 	mutex_destroy(&itask->it_mutex);
2872 	kmem_free(itask, sizeof (iscsit_task_t));
2873 }
2874 
2875 static idm_status_t
2876 iscsit_task_start(iscsit_task_t *itask)
2877 {
2878 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2879 	avl_index_t		where;
2880 
2881 	/*
2882 	 * Sanity check the ITT and ensure that this task does not already
2883 	 * exist.  If not then add the task to the session task list.
2884 	 */
2885 	mutex_enter(&ist->ist_mutex);
2886 	mutex_enter(&itask->it_mutex);
2887 	itask->it_active = 1;
2888 	if (avl_find(&ist->ist_task_list, itask, &where) == NULL) {
2889 		/* New task, add to AVL */
2890 		avl_insert(&ist->ist_task_list, itask, where);
2891 		mutex_exit(&itask->it_mutex);
2892 		mutex_exit(&ist->ist_mutex);
2893 		return (IDM_STATUS_SUCCESS);
2894 	}
2895 	mutex_exit(&itask->it_mutex);
2896 	mutex_exit(&ist->ist_mutex);
2897 
2898 	return (IDM_STATUS_REJECT);
2899 }
2900 
2901 static void
2902 iscsit_task_done(iscsit_task_t *itask)
2903 {
2904 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2905 
2906 	mutex_enter(&ist->ist_mutex);
2907 	mutex_enter(&itask->it_mutex);
2908 	if (itask->it_active) {
2909 		avl_remove(&ist->ist_task_list, itask);
2910 		itask->it_active = 0;
2911 	}
2912 	mutex_exit(&itask->it_mutex);
2913 	mutex_exit(&ist->ist_mutex);
2914 }
2915 
2916 /*
2917  * iscsit status PDU cache
2918  */
2919 
2920 /*ARGSUSED*/
2921 static int
2922 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags)
2923 {
2924 	idm_pdu_t *pdu = pdu_void;
2925 	iscsi_scsi_rsp_hdr_t *rsp;
2926 
2927 	bzero(pdu, sizeof (idm_pdu_t));
2928 	pdu->isp_callback = iscsit_send_good_status_done;
2929 	pdu->isp_magic = IDM_PDU_MAGIC;
2930 	pdu->isp_hdr = (iscsi_hdr_t *)(pdu + 1); /* Ptr arithmetic */
2931 	pdu->isp_hdrlen = sizeof (iscsi_hdr_t);
2932 
2933 	/* Setup status response */
2934 	rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2935 	bzero(rsp, sizeof (*rsp));
2936 	rsp->opcode = ISCSI_OP_SCSI_RSP;
2937 	rsp->flags = ISCSI_FLAG_FINAL;
2938 	rsp->response = ISCSI_STATUS_CMD_COMPLETED;
2939 
2940 	return (0);
2941 }
2942 
2943 /*
2944  * iscsit private data handler
2945  */
2946 
2947 /*ARGSUSED*/
2948 static void
2949 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags)
2950 {
2951 	it_config_t		*cfg;
2952 	nvlist_t		*nvl;
2953 	iscsit_service_enabled_t	old_state;
2954 
2955 	if ((cmd != STMF_PROVIDER_DATA_UPDATED) || (arg == NULL)) {
2956 		return;
2957 	}
2958 
2959 	nvl = (nvlist_t *)arg;
2960 
2961 	/* Translate nvlist */
2962 	if (it_nv_to_config(nvl, &cfg) != 0) {
2963 		cmn_err(CE_WARN, "Configuration is invalid");
2964 		return;
2965 	}
2966 
2967 	/* Check that no iSCSI ioctl is currently running */
2968 	mutex_enter(&iscsit_global.global_state_mutex);
2969 	old_state = iscsit_global.global_svc_state;
2970 	switch (iscsit_global.global_svc_state) {
2971 	case ISE_ENABLED:
2972 	case ISE_DISABLED:
2973 		iscsit_global.global_svc_state = ISE_BUSY;
2974 		break;
2975 	case ISE_ENABLING:
2976 		/*
2977 		 * It is OK for the iscsit_pp_cb to be called from inside of
2978 		 * an iSCSI ioctl only if we are currently executing inside
2979 		 * of stmf_register_port_provider.
2980 		 */
2981 		ASSERT((flags & STMF_PCB_PREG_COMPLETE) != 0);
2982 		break;
2983 	default:
2984 		cmn_err(CE_WARN, "iscsit_pp_cb called when global_svc_state"
2985 		    " is not ENABLED(0x%x) -- ignoring",
2986 		    iscsit_global.global_svc_state);
2987 		mutex_exit(&iscsit_global.global_state_mutex);
2988 		it_config_free_cmn(cfg);
2989 		return;
2990 	}
2991 	mutex_exit(&iscsit_global.global_state_mutex);
2992 
2993 	/* Update config */
2994 	(void) iscsit_config_merge(cfg);
2995 
2996 	it_config_free_cmn(cfg);
2997 
2998 	/* Restore old iSCSI driver global state */
2999 	mutex_enter(&iscsit_global.global_state_mutex);
3000 	ASSERT(iscsit_global.global_svc_state == ISE_BUSY ||
3001 	    iscsit_global.global_svc_state == ISE_ENABLING);
3002 	iscsit_global.global_svc_state = old_state;
3003 	mutex_exit(&iscsit_global.global_state_mutex);
3004 }
3005 
3006 
3007 static it_cfg_status_t
3008 iscsit_config_merge(it_config_t *in_cfg)
3009 {
3010 	it_cfg_status_t	status;
3011 	it_config_t	*cfg;
3012 	it_config_t	tmp_cfg;
3013 	list_t		tpg_del_list;
3014 
3015 	if (in_cfg) {
3016 		cfg = in_cfg;
3017 	} else {
3018 		/* Make empty config */
3019 		bzero(&tmp_cfg, sizeof (tmp_cfg));
3020 		cfg = &tmp_cfg;
3021 	}
3022 
3023 	list_create(&tpg_del_list,  sizeof (iscsit_tpg_t),
3024 	    offsetof(iscsit_tpg_t, tpg_delete_ln));
3025 
3026 	/*
3027 	 * Update targets, initiator contexts, target portal groups,
3028 	 * and iSNS client
3029 	 */
3030 	ISCSIT_GLOBAL_LOCK(RW_WRITER);
3031 	if (((status = iscsit_config_merge_tpg(cfg, &tpg_del_list))
3032 	    != 0) ||
3033 	    ((status = iscsit_config_merge_tgt(cfg)) != 0) ||
3034 	    ((status = iscsit_config_merge_ini(cfg)) != 0) ||
3035 	    ((status = isnst_config_merge(cfg)) != 0)) {
3036 		ISCSIT_GLOBAL_UNLOCK();
3037 		return (status);
3038 	}
3039 
3040 	/* Update other global config parameters */
3041 	if (iscsit_global.global_props) {
3042 		nvlist_free(iscsit_global.global_props);
3043 		iscsit_global.global_props = NULL;
3044 	}
3045 	if (in_cfg) {
3046 		(void) nvlist_dup(cfg->config_global_properties,
3047 		    &iscsit_global.global_props, KM_SLEEP);
3048 	}
3049 	ISCSIT_GLOBAL_UNLOCK();
3050 
3051 	iscsit_config_destroy_tpgs(&tpg_del_list);
3052 
3053 	list_destroy(&tpg_del_list);
3054 
3055 	return (ITCFG_SUCCESS);
3056 }
3057 
3058 /*
3059  * iscsit_sna_lt[e]
3060  *
3061  * Compare serial numbers using serial number arithmetic as defined in
3062  * RFC 1982.
3063  *
3064  * NOTE: This code is duplicated in the isns server. It ought to be common.
3065  */
3066 
3067 static int
3068 iscsit_sna_lt(uint32_t sn1, uint32_t sn2)
3069 {
3070 	return ((sn1 != sn2) &&
3071 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3072 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3073 }
3074 
3075 static int
3076 iscsit_sna_lte(uint32_t sn1, uint32_t sn2)
3077 {
3078 	return ((sn1 == sn2) ||
3079 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3080 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3081 }
3082 
3083 
3084 static boolean_t
3085 iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn)
3086 {
3087 	iscsit_sess_t	*ist = ict->ict_sess;
3088 	int		rval = B_TRUE;
3089 
3090 	ist = ict->ict_sess;
3091 
3092 	mutex_enter(&ist->ist_sn_mutex);
3093 
3094 	/*
3095 	 * If cmdsn is less than ist_expcmdsn - iscsit_cmd_window() or
3096 	 * greater than ist_expcmdsn, it's not in the window.
3097 	 */
3098 
3099 	if (iscsit_sna_lt(cmdsn, (ist->ist_expcmdsn - iscsit_cmd_window())) ||
3100 	    !iscsit_sna_lte(cmdsn, ist->ist_expcmdsn)) {
3101 		rval = B_FALSE;
3102 	}
3103 
3104 	mutex_exit(&ist->ist_sn_mutex);
3105 
3106 	return (rval);
3107 }
3108 
3109 /*
3110  * iscsit_check_cmdsn_and_queue
3111  *
3112  * Independent of the order in which the iSCSI target receives non-immediate
3113  * command PDU across the entire session and any multiple connections within
3114  * the session, the target must deliver the commands to the SCSI layer in
3115  * CmdSN order. So out-of-order non-immediate commands are queued up on a
3116  * session-wide wait queue. Duplicate commands are ignored.
3117  *
3118  */
3119 static int
3120 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu)
3121 {
3122 	idm_conn_t		*ic = rx_pdu->isp_ic;
3123 	iscsit_conn_t		*ict = ic->ic_handle;
3124 	iscsit_sess_t		*ist = ict->ict_sess;
3125 	iscsi_scsi_cmd_hdr_t	*hdr = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
3126 
3127 	mutex_enter(&ist->ist_sn_mutex);
3128 	if (hdr->opcode & ISCSI_OP_IMMEDIATE) {
3129 		/* do not queue, handle it immediately */
3130 		DTRACE_PROBE2(immediate__cmd, iscsit_sess_t *, ist,
3131 		    idm_pdu_t *, rx_pdu);
3132 		mutex_exit(&ist->ist_sn_mutex);
3133 		return (ISCSIT_CMDSN_EQ_EXPCMDSN);
3134 	}
3135 	if (iscsit_sna_lt(ist->ist_expcmdsn, ntohl(hdr->cmdsn))) {
3136 		/*
3137 		 * Out-of-order commands (cmdSN higher than ExpCmdSN)
3138 		 * are staged on a fixed-size circular buffer until
3139 		 * the missing command is delivered to the SCSI layer.
3140 		 * Irrespective of the order of insertion into the
3141 		 * staging queue, the commands are processed out of the
3142 		 * queue in cmdSN order only.
3143 		 */
3144 		rx_pdu->isp_queue_time = gethrtime();
3145 		iscsit_add_pdu_to_queue(ist, rx_pdu);
3146 		mutex_exit(&ist->ist_sn_mutex);
3147 		return (ISCSIT_CMDSN_GT_EXPCMDSN);
3148 	} else if (iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_expcmdsn)) {
3149 		DTRACE_PROBE3(cmdsn__lt__expcmdsn, iscsit_sess_t *, ist,
3150 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
3151 		mutex_exit(&ist->ist_sn_mutex);
3152 		return (ISCSIT_CMDSN_LT_EXPCMDSN);
3153 	} else {
3154 		mutex_exit(&ist->ist_sn_mutex);
3155 		return (ISCSIT_CMDSN_EQ_EXPCMDSN);
3156 	}
3157 }
3158 
3159 /*
3160  * iscsit_add_pdu_to_queue() adds PDUs into the array indexed by
3161  * their cmdsn value. The length of the array is kept above the
3162  * maximum window size. The window keeps the cmdsn within a range
3163  * such that there are no collisons. e.g. the assumption is that
3164  * the windowing checks make it impossible to receive PDUs that
3165  * index into the same location in the array.
3166  */
3167 static void
3168 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu)
3169 {
3170 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3171 	iscsit_conn_t	*ict 	= rx_pdu->isp_ic->ic_handle;
3172 	uint32_t	cmdsn	=
3173 	    ((iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr)->cmdsn;
3174 	uint32_t	index;
3175 
3176 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3177 	/*
3178 	 * If the connection is being torn down, then
3179 	 * don't add the PDU to the staging queue
3180 	 */
3181 	mutex_enter(&ict->ict_mutex);
3182 	if (ict->ict_lost) {
3183 		mutex_exit(&ict->ict_mutex);
3184 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3185 		return;
3186 	}
3187 	iscsit_conn_dispatch_hold(ict);
3188 	mutex_exit(&ict->ict_mutex);
3189 
3190 	index = ntohl(cmdsn) % ISCSIT_RXPDU_QUEUE_LEN;
3191 	/*
3192 	 * In the normal case, assuming that the Initiator is not
3193 	 * buggy and that we don't have packet duplication occuring,
3194 	 * the entry in the array will be NULL.  However, we may have
3195 	 * received a duplicate PDU with cmdsn > expsn , and in that
3196 	 * case we just ignore this PDU -- the previously received one
3197 	 * remains queued for processing.  We need to be careful not
3198 	 * to leak this one however.
3199 	 */
3200 	if (cbuf->cb_buffer[index] != NULL) {
3201 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3202 	} else {
3203 		cbuf->cb_buffer[index] = rx_pdu;
3204 		cbuf->cb_num_elems++;
3205 	}
3206 }
3207 
3208 static idm_pdu_t *
3209 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn)
3210 {
3211 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3212 	idm_pdu_t	*pdu	= NULL;
3213 	uint32_t	index;
3214 
3215 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3216 	index = cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3217 	if ((pdu = cbuf->cb_buffer[index]) != NULL) {
3218 		ASSERT(cmdsn ==
3219 		    ntohl(((iscsi_scsi_cmd_hdr_t *)pdu->isp_hdr)->cmdsn));
3220 		cbuf->cb_buffer[index] = NULL;
3221 		cbuf->cb_num_elems--;
3222 		return (pdu);
3223 	}
3224 	return (NULL);
3225 }
3226 
3227 /*
3228  * iscsit_process_pdu_in_queue() finds the next pdu in sequence
3229  * and posts it to the SCSI layer
3230  */
3231 static void
3232 iscsit_process_pdu_in_queue(iscsit_sess_t *ist)
3233 {
3234 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3235 	idm_pdu_t	*pdu = NULL;
3236 	uint32_t	expcmdsn;
3237 
3238 	for (;;) {
3239 		mutex_enter(&ist->ist_sn_mutex);
3240 		if (cbuf->cb_num_elems == 0) {
3241 			mutex_exit(&ist->ist_sn_mutex);
3242 			break;
3243 		}
3244 		expcmdsn = ist->ist_expcmdsn;
3245 		if ((pdu = iscsit_remove_pdu_from_queue(ist, expcmdsn))
3246 		    == NULL) {
3247 			mutex_exit(&ist->ist_sn_mutex);
3248 			break;
3249 		}
3250 		mutex_exit(&ist->ist_sn_mutex);
3251 		iscsit_post_staged_pdu(pdu);
3252 	}
3253 }
3254 
3255 static void
3256 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu)
3257 {
3258 	iscsit_conn_t	*ict	= rx_pdu->isp_ic->ic_handle;
3259 
3260 	/* Post the PDU to the SCSI layer */
3261 	switch (IDM_PDU_OPCODE(rx_pdu)) {
3262 	case ISCSI_OP_NOOP_OUT:
3263 		iscsit_set_cmdsn(ict, rx_pdu);
3264 		iscsit_pdu_op_noop(ict, rx_pdu);
3265 		break;
3266 	case ISCSI_OP_TEXT_CMD:
3267 		iscsit_set_cmdsn(ict, rx_pdu);
3268 		iscsit_pdu_op_text_cmd(ict, rx_pdu);
3269 		break;
3270 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
3271 		iscsit_set_cmdsn(ict, rx_pdu);
3272 		iscsit_op_scsi_task_mgmt(ict, rx_pdu);
3273 		break;
3274 	case ISCSI_OP_SCSI_CMD:
3275 		/* cmdSN will be incremented after creating itask */
3276 		iscsit_post_scsi_cmd(rx_pdu->isp_ic, rx_pdu);
3277 		break;
3278 	case ISCSI_OP_LOGOUT_CMD:
3279 		iscsit_set_cmdsn(ict, rx_pdu);
3280 		iscsit_pdu_op_logout_cmd(ict, rx_pdu);
3281 		break;
3282 	default:
3283 		/* No other PDUs should be placed on the queue */
3284 		ASSERT(0);
3285 	}
3286 	iscsit_conn_dispatch_rele(ict); /* release hold on the conn */
3287 }
3288 
3289 /* ARGSUSED */
3290 void
3291 iscsit_rxpdu_queue_monitor_start(void)
3292 {
3293 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3294 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3295 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3296 		return;
3297 	}
3298 	iscsit_rxpdu_queue_monitor_thr_id =
3299 	    thread_create(NULL, 0, iscsit_rxpdu_queue_monitor, NULL,
3300 	    0, &p0, TS_RUN, minclsyspri);
3301 	while (!iscsit_rxpdu_queue_monitor_thr_running) {
3302 		cv_wait(&iscsit_rxpdu_queue_monitor_cv,
3303 		    &iscsit_rxpdu_queue_monitor_mutex);
3304 	}
3305 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3306 
3307 }
3308 
3309 /* ARGSUSED */
3310 void
3311 iscsit_rxpdu_queue_monitor_stop(void)
3312 {
3313 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3314 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3315 		iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
3316 		cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3317 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3318 
3319 		thread_join(iscsit_rxpdu_queue_monitor_thr_did);
3320 		return;
3321 	}
3322 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3323 }
3324 
3325 /*
3326  * A separate thread is used to scan the staging queue on all the
3327  * sessions, If a delayed PDU does not arrive within a timeout, the
3328  * target will advance to the staged PDU that is next in sequence
3329  * and exceeded the threshold wait time. It is up to the initiator
3330  * to note that the target has not acknowledged a particular cmdsn
3331  * and take appropriate action.
3332  */
3333 /* ARGSUSED */
3334 static void
3335 iscsit_rxpdu_queue_monitor(void *arg)
3336 {
3337 	iscsit_tgt_t	*tgt;
3338 	iscsit_sess_t	*ist;
3339 
3340 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3341 	iscsit_rxpdu_queue_monitor_thr_did = curthread->t_did;
3342 	iscsit_rxpdu_queue_monitor_thr_running = B_TRUE;
3343 	cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3344 
3345 	while (iscsit_rxpdu_queue_monitor_thr_running) {
3346 		ISCSIT_GLOBAL_LOCK(RW_READER);
3347 		for (tgt = avl_first(&iscsit_global.global_target_list);
3348 		    tgt != NULL;
3349 		    tgt = AVL_NEXT(&iscsit_global.global_target_list, tgt)) {
3350 			mutex_enter(&tgt->target_mutex);
3351 			for (ist = avl_first(&tgt->target_sess_list);
3352 			    ist != NULL;
3353 			    ist = AVL_NEXT(&tgt->target_sess_list, ist)) {
3354 
3355 				iscsit_rxpdu_queue_monitor_session(ist);
3356 			}
3357 			mutex_exit(&tgt->target_mutex);
3358 		}
3359 		ISCSIT_GLOBAL_UNLOCK();
3360 		if (iscsit_rxpdu_queue_monitor_thr_running == B_FALSE) {
3361 			break;
3362 		}
3363 		(void) cv_reltimedwait(&iscsit_rxpdu_queue_monitor_cv,
3364 		    &iscsit_rxpdu_queue_monitor_mutex,
3365 		    ISCSIT_RXPDU_QUEUE_MONITOR_INTERVAL * drv_usectohz(1000000),
3366 		    TR_CLOCK_TICK);
3367 	}
3368 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3369 	thread_exit();
3370 }
3371 
3372 static void
3373 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist)
3374 {
3375 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3376 	idm_pdu_t	*next_pdu = NULL;
3377 	uint32_t	index, next_cmdsn, i;
3378 
3379 	/*
3380 	 * Assume that all PDUs in the staging queue have a cmdsn >= expcmdsn.
3381 	 * Starting with the expcmdsn, iterate over the staged PDUs to find
3382 	 * the next PDU with a wait time greater than the threshold. If found
3383 	 * advance the staged PDU to the SCSI layer, skipping over the missing
3384 	 * PDU(s) to get past the hole in the command sequence. It is up to
3385 	 * the initiator to note that the target has not acknowledged a cmdsn
3386 	 * and take appropriate action.
3387 	 *
3388 	 * Since the PDU(s) arrive in any random order, it is possible that
3389 	 * that the actual wait time for a particular PDU is much longer than
3390 	 * the defined threshold. e.g. Consider a case where commands are sent
3391 	 * over 4 different connections, and cmdsn = 1004 arrives first, then
3392 	 * 1003, and 1002 and 1001 are lost due to a connection failure.
3393 	 * So now 1003 is waiting for 1002 to be delivered, and although the
3394 	 * wait time of 1004 > wait time of 1003, only 1003 will be considered
3395 	 * by the monitor thread. 1004 will be automatically processed by
3396 	 * iscsit_process_pdu_in_queue() once the scan is complete and the
3397 	 * expcmdsn becomes current.
3398 	 */
3399 	mutex_enter(&ist->ist_sn_mutex);
3400 	cbuf = ist->ist_rxpdu_queue;
3401 	if (cbuf->cb_num_elems == 0) {
3402 		mutex_exit(&ist->ist_sn_mutex);
3403 		return;
3404 	}
3405 	for (next_pdu = NULL, i = 0; ; i++) {
3406 		next_cmdsn = ist->ist_expcmdsn + i; /* start at expcmdsn */
3407 		index = next_cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3408 		if ((next_pdu = cbuf->cb_buffer[index]) != NULL) {
3409 			/*
3410 			 * If the PDU wait time has not exceeded threshold
3411 			 * stop scanning the staging queue until the timer
3412 			 * fires again
3413 			 */
3414 			if ((gethrtime() - next_pdu->isp_queue_time)
3415 			    < (rxpdu_queue_threshold * NANOSEC)) {
3416 				mutex_exit(&ist->ist_sn_mutex);
3417 				return;
3418 			}
3419 			/*
3420 			 * Remove the next PDU from the queue and post it
3421 			 * to the SCSI layer, skipping over the missing
3422 			 * PDU. Stop scanning the staging queue until
3423 			 * the monitor timer fires again
3424 			 */
3425 			(void) iscsit_remove_pdu_from_queue(ist, next_cmdsn);
3426 			mutex_exit(&ist->ist_sn_mutex);
3427 			DTRACE_PROBE3(advanced__to__blocked__cmdsn,
3428 			    iscsit_sess_t *, ist, idm_pdu_t *, next_pdu,
3429 			    uint32_t, next_cmdsn);
3430 			iscsit_post_staged_pdu(next_pdu);
3431 			/* Deliver any subsequent PDUs immediately */
3432 			iscsit_process_pdu_in_queue(ist);
3433 			return;
3434 		}
3435 		/*
3436 		 * Skipping over i PDUs, e.g. a case where commands 1001 and
3437 		 * 1002 are lost in the network, skip over both and post 1003
3438 		 * expcmdsn then becomes 1004 at the end of the scan.
3439 		 */
3440 		DTRACE_PROBE2(skipping__over__cmdsn, iscsit_sess_t *, ist,
3441 		    uint32_t, next_cmdsn);
3442 	}
3443 	/*
3444 	 * following the assumption, staged cmdsn >= expcmdsn, this statement
3445 	 * is never reached.
3446 	 */
3447 }
3448