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  * Copyright (c) 2017, Joyent, Inc.  All rights reserved.
26  */
27 
28 #include <sys/cpuvar.h>
29 #include <sys/types.h>
30 #include <sys/conf.h>
31 #include <sys/stat.h>
32 #include <sys/file.h>
33 #include <sys/ddi.h>
34 #include <sys/sunddi.h>
35 #include <sys/modctl.h>
36 #include <sys/sysmacros.h>
37 #include <sys/socket.h>
38 #include <sys/strsubr.h>
39 #include <sys/nvpair.h>
40 
41 #include <sys/stmf.h>
42 #include <sys/stmf_ioctl.h>
43 #include <sys/portif.h>
44 #include <sys/idm/idm.h>
45 #include <sys/idm/idm_conn_sm.h>
46 
47 #include "iscsit_isns.h"
48 #include "iscsit.h"
49 
50 #define	ISCSIT_VERSION		BUILD_DATE "-1.18dev"
51 #define	ISCSIT_NAME_VERSION	"COMSTAR ISCSIT v" ISCSIT_VERSION
52 
53 /*
54  * DDI entry points.
55  */
56 static int iscsit_drv_attach(dev_info_t *, ddi_attach_cmd_t);
57 static int iscsit_drv_detach(dev_info_t *, ddi_detach_cmd_t);
58 static int iscsit_drv_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
59 static int iscsit_drv_open(dev_t *, int, int, cred_t *);
60 static int iscsit_drv_close(dev_t, int, int, cred_t *);
61 static boolean_t iscsit_drv_busy(void);
62 static int iscsit_drv_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
63 
64 extern struct mod_ops mod_miscops;
65 
66 
67 static struct cb_ops iscsit_cb_ops = {
68 	iscsit_drv_open,	/* cb_open */
69 	iscsit_drv_close,	/* cb_close */
70 	nodev,			/* cb_strategy */
71 	nodev,			/* cb_print */
72 	nodev,			/* cb_dump */
73 	nodev,			/* cb_read */
74 	nodev,			/* cb_write */
75 	iscsit_drv_ioctl,	/* cb_ioctl */
76 	nodev,			/* cb_devmap */
77 	nodev,			/* cb_mmap */
78 	nodev,			/* cb_segmap */
79 	nochpoll,		/* cb_chpoll */
80 	ddi_prop_op,		/* cb_prop_op */
81 	NULL,			/* cb_streamtab */
82 	D_MP,			/* cb_flag */
83 	CB_REV,			/* cb_rev */
84 	nodev,			/* cb_aread */
85 	nodev,			/* cb_awrite */
86 };
87 
88 static struct dev_ops iscsit_dev_ops = {
89 	DEVO_REV,		/* devo_rev */
90 	0,			/* devo_refcnt */
91 	iscsit_drv_getinfo,	/* devo_getinfo */
92 	nulldev,		/* devo_identify */
93 	nulldev,		/* devo_probe */
94 	iscsit_drv_attach,	/* devo_attach */
95 	iscsit_drv_detach,	/* devo_detach */
96 	nodev,			/* devo_reset */
97 	&iscsit_cb_ops,		/* devo_cb_ops */
98 	NULL,			/* devo_bus_ops */
99 	NULL,			/* devo_power */
100 	ddi_quiesce_not_needed,	/* quiesce */
101 };
102 
103 static struct modldrv modldrv = {
104 	&mod_driverops,
105 	"iSCSI Target",
106 	&iscsit_dev_ops,
107 };
108 
109 static struct modlinkage modlinkage = {
110 	MODREV_1,
111 	&modldrv,
112 	NULL,
113 };
114 
115 
116 iscsit_global_t iscsit_global;
117 
118 kmem_cache_t	*iscsit_status_pdu_cache;
119 
120 boolean_t	iscsit_sm_logging = B_FALSE;
121 
122 kmutex_t	login_sm_session_mutex;
123 
124 static idm_status_t iscsit_init(dev_info_t *dip);
125 static idm_status_t iscsit_enable_svc(iscsit_hostinfo_t *hostinfo);
126 static void iscsit_disable_svc(void);
127 
128 static boolean_t
129 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu);
130 
131 static void
132 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu);
133 
134 static idm_pdu_t *
135 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn);
136 
137 static void
138 iscsit_process_pdu_in_queue(iscsit_sess_t *ist);
139 
140 static void
141 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist);
142 
143 static void
144 iscsit_rxpdu_queue_monitor(void *arg);
145 
146 static void
147 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu);
148 
149 static void
150 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu);
151 
152 static void
153 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
154 
155 static void
156 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
157 
158 static void
159 iscsit_pdu_op_login_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
160 
161 void
162 iscsit_pdu_op_text_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
163 
164 static void
165 iscsit_pdu_op_logout_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
166 
167 int iscsit_cmd_window();
168 
169 static  int
170 iscsit_sna_lt(uint32_t sn1, uint32_t sn2);
171 
172 void
173 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
174 
175 static void
176 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu);
177 
178 static void
179 iscsit_deferred(void *rx_pdu_void);
180 
181 static idm_status_t
182 iscsit_conn_accept(idm_conn_t *ic);
183 
184 static idm_status_t
185 iscsit_ffp_enabled(idm_conn_t *ic);
186 
187 static idm_status_t
188 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class);
189 
190 static idm_status_t
191 iscsit_conn_lost(idm_conn_t *ic);
192 
193 static idm_status_t
194 iscsit_conn_destroy(idm_conn_t *ic);
195 
196 static stmf_data_buf_t *
197 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
198     uint32_t flags);
199 
200 static void
201 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf);
202 
203 static void
204 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status);
205 
206 static void
207 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status);
208 
209 static void
210 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status);
211 
212 static stmf_status_t
213 iscsit_idm_to_stmf(idm_status_t idmrc);
214 
215 static iscsit_task_t *
216 iscsit_task_alloc(iscsit_conn_t *ict);
217 
218 static void
219 iscsit_task_free(iscsit_task_t *itask);
220 
221 static iscsit_task_t *
222 iscsit_tm_task_alloc(iscsit_conn_t *ict);
223 
224 static void
225 iscsit_tm_task_free(iscsit_task_t *itask);
226 
227 static idm_status_t
228 iscsit_task_start(iscsit_task_t *itask);
229 
230 static void
231 iscsit_task_done(iscsit_task_t *itask);
232 
233 static int
234 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags);
235 
236 static void
237 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags);
238 
239 static it_cfg_status_t
240 iscsit_config_merge(it_config_t *cfg);
241 
242 static idm_status_t
243 iscsit_login_fail(idm_conn_t *ic);
244 
245 static boolean_t iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn);
246 static void iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
247     uint8_t response, uint8_t cmd_status);
248 static void iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu,
249     uint8_t tm_status);
250 
251 /*
252  * MC/S: Out-of-order commands are staged on a session-wide wait
253  * queue until a system-tunable threshold is reached. A separate
254  * thread is used to scan the staging queue on all the session,
255  * If a delayed PDU does not arrive within a timeout, the target
256  * will advance to the staged PDU that is next in sequence, skipping
257  * over the missing PDU(s) to go past a hole in the sequence.
258  */
259 volatile int rxpdu_queue_threshold = ISCSIT_RXPDU_QUEUE_THRESHOLD;
260 
261 static kmutex_t		iscsit_rxpdu_queue_monitor_mutex;
262 kthread_t		*iscsit_rxpdu_queue_monitor_thr_id;
263 static kt_did_t		iscsit_rxpdu_queue_monitor_thr_did;
264 static boolean_t	iscsit_rxpdu_queue_monitor_thr_running;
265 static kcondvar_t	iscsit_rxpdu_queue_monitor_cv;
266 
267 int
268 _init(void)
269 {
270 	int rc;
271 
272 	rw_init(&iscsit_global.global_rwlock, NULL, RW_DRIVER, NULL);
273 	mutex_init(&iscsit_global.global_state_mutex, NULL,
274 	    MUTEX_DRIVER, NULL);
275 	iscsit_global.global_svc_state = ISE_DETACHED;
276 
277 	mutex_init(&iscsit_rxpdu_queue_monitor_mutex, NULL,
278 	    MUTEX_DRIVER, NULL);
279 	mutex_init(&login_sm_session_mutex, NULL, MUTEX_DRIVER, NULL);
280 	iscsit_rxpdu_queue_monitor_thr_id = NULL;
281 	iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
282 	cv_init(&iscsit_rxpdu_queue_monitor_cv, NULL, CV_DEFAULT, NULL);
283 
284 	if ((rc = mod_install(&modlinkage)) != 0) {
285 		mutex_destroy(&iscsit_global.global_state_mutex);
286 		rw_destroy(&iscsit_global.global_rwlock);
287 		return (rc);
288 	}
289 
290 	return (rc);
291 }
292 
293 int
294 _info(struct modinfo *modinfop)
295 {
296 	return (mod_info(&modlinkage, modinfop));
297 }
298 
299 int
300 _fini(void)
301 {
302 	int rc;
303 
304 	rc = mod_remove(&modlinkage);
305 
306 	if (rc == 0) {
307 		mutex_destroy(&iscsit_rxpdu_queue_monitor_mutex);
308 		mutex_destroy(&login_sm_session_mutex);
309 		cv_destroy(&iscsit_rxpdu_queue_monitor_cv);
310 		mutex_destroy(&iscsit_global.global_state_mutex);
311 		rw_destroy(&iscsit_global.global_rwlock);
312 	}
313 
314 	return (rc);
315 }
316 
317 /*
318  * DDI entry points.
319  */
320 
321 /* ARGSUSED */
322 static int
323 iscsit_drv_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
324     void **result)
325 {
326 	ulong_t instance = getminor((dev_t)arg);
327 
328 	switch (cmd) {
329 	case DDI_INFO_DEVT2DEVINFO:
330 		*result = iscsit_global.global_dip;
331 		return (DDI_SUCCESS);
332 
333 	case DDI_INFO_DEVT2INSTANCE:
334 		*result = (void *)instance;
335 		return (DDI_SUCCESS);
336 
337 	default:
338 		break;
339 	}
340 
341 	return (DDI_FAILURE);
342 }
343 
344 static int
345 iscsit_drv_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
346 {
347 	if (cmd != DDI_ATTACH) {
348 		return (DDI_FAILURE);
349 	}
350 
351 	if (ddi_get_instance(dip) != 0) {
352 		/* we only allow instance 0 to attach */
353 		return (DDI_FAILURE);
354 	}
355 
356 	/* create the minor node */
357 	if (ddi_create_minor_node(dip, ISCSIT_MODNAME, S_IFCHR, 0,
358 	    DDI_PSEUDO, 0) != DDI_SUCCESS) {
359 		cmn_err(CE_WARN, "iscsit_drv_attach: "
360 		    "failed creating minor node");
361 		return (DDI_FAILURE);
362 	}
363 
364 	if (iscsit_init(dip) != IDM_STATUS_SUCCESS) {
365 		cmn_err(CE_WARN, "iscsit_drv_attach: "
366 		    "failed to initialize");
367 		ddi_remove_minor_node(dip, NULL);
368 		return (DDI_FAILURE);
369 	}
370 
371 	iscsit_global.global_svc_state = ISE_DISABLED;
372 	iscsit_global.global_dip = dip;
373 
374 	return (DDI_SUCCESS);
375 }
376 
377 /*ARGSUSED*/
378 static int
379 iscsit_drv_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
380 {
381 	if (cmd != DDI_DETACH)
382 		return (DDI_FAILURE);
383 
384 	/*
385 	 * drv_detach is called in a context that owns the
386 	 * device node for the /dev/pseudo device.  If this thread blocks
387 	 * for any resource, other threads that need the /dev/pseudo device
388 	 * may end up in a deadlock with this thread.Hence, we use a
389 	 * separate lock just for the structures that drv_detach needs
390 	 * to access.
391 	 */
392 	mutex_enter(&iscsit_global.global_state_mutex);
393 	if (iscsit_drv_busy()) {
394 		mutex_exit(&iscsit_global.global_state_mutex);
395 		return (EBUSY);
396 	}
397 
398 	iscsit_global.global_dip = NULL;
399 	ddi_remove_minor_node(dip, NULL);
400 
401 	ldi_ident_release(iscsit_global.global_li);
402 	iscsit_global.global_svc_state = ISE_DETACHED;
403 
404 	mutex_exit(&iscsit_global.global_state_mutex);
405 
406 	return (DDI_SUCCESS);
407 }
408 
409 /*ARGSUSED*/
410 static int
411 iscsit_drv_open(dev_t *devp, int flag, int otyp, cred_t *credp)
412 {
413 	return (0);
414 }
415 
416 /* ARGSUSED */
417 static int
418 iscsit_drv_close(dev_t dev, int flag, int otyp, cred_t *credp)
419 {
420 	return (0);
421 }
422 
423 static boolean_t
424 iscsit_drv_busy(void)
425 {
426 	ASSERT(MUTEX_HELD(&iscsit_global.global_state_mutex));
427 
428 	switch (iscsit_global.global_svc_state) {
429 	case ISE_DISABLED:
430 	case ISE_DETACHED:
431 		return (B_FALSE);
432 	default:
433 		return (B_TRUE);
434 	}
435 	/* NOTREACHED */
436 }
437 
438 /* ARGSUSED */
439 static int
440 iscsit_drv_ioctl(dev_t drv, int cmd, intptr_t argp, int flag, cred_t *cred,
441     int *retval)
442 {
443 	iscsit_ioc_set_config_t		setcfg;
444 	iscsit_ioc_set_config32_t	setcfg32;
445 	char				*cfg_pnvlist = NULL;
446 	nvlist_t			*cfg_nvlist = NULL;
447 	it_config_t			*cfg = NULL;
448 	idm_status_t			idmrc;
449 	int				rc = 0;
450 
451 	if (drv_priv(cred) != 0) {
452 		return (EPERM);
453 	}
454 
455 	mutex_enter(&iscsit_global.global_state_mutex);
456 
457 	/*
458 	 * Validate ioctl requests against global service state
459 	 */
460 	switch (iscsit_global.global_svc_state) {
461 	case ISE_ENABLED:
462 		if (cmd == ISCSIT_IOC_DISABLE_SVC) {
463 			iscsit_global.global_svc_state = ISE_DISABLING;
464 		} else if (cmd == ISCSIT_IOC_ENABLE_SVC) {
465 			/* Already enabled */
466 			mutex_exit(&iscsit_global.global_state_mutex);
467 			return (0);
468 		} else {
469 			iscsit_global.global_svc_state = ISE_BUSY;
470 		}
471 		break;
472 	case ISE_DISABLED:
473 		if (cmd == ISCSIT_IOC_ENABLE_SVC) {
474 			iscsit_global.global_svc_state = ISE_ENABLING;
475 		} else if (cmd == ISCSIT_IOC_DISABLE_SVC) {
476 			/* Already disabled */
477 			mutex_exit(&iscsit_global.global_state_mutex);
478 			return (0);
479 		} else {
480 			rc = EFAULT;
481 		}
482 		break;
483 	case ISE_BUSY:
484 	case ISE_ENABLING:
485 	case ISE_DISABLING:
486 		rc = EAGAIN;
487 		break;
488 	case ISE_DETACHED:
489 	default:
490 		rc = EFAULT;
491 		break;
492 	}
493 
494 	mutex_exit(&iscsit_global.global_state_mutex);
495 	if (rc != 0)
496 		return (rc);
497 
498 	/* Handle ioctl request (enable/disable have already been handled) */
499 	switch (cmd) {
500 	case ISCSIT_IOC_SET_CONFIG:
501 		/* Any errors must set state back to ISE_ENABLED */
502 		switch (ddi_model_convert_from(flag & FMODELS)) {
503 		case DDI_MODEL_ILP32:
504 			if (ddi_copyin((void *)argp, &setcfg32,
505 			    sizeof (iscsit_ioc_set_config32_t), flag) != 0) {
506 				rc = EFAULT;
507 				goto cleanup;
508 			}
509 
510 			setcfg.set_cfg_pnvlist =
511 			    (char *)((uintptr_t)setcfg32.set_cfg_pnvlist);
512 			setcfg.set_cfg_vers = setcfg32.set_cfg_vers;
513 			setcfg.set_cfg_pnvlist_len =
514 			    setcfg32.set_cfg_pnvlist_len;
515 			break;
516 		case DDI_MODEL_NONE:
517 			if (ddi_copyin((void *)argp, &setcfg,
518 			    sizeof (iscsit_ioc_set_config_t), flag) != 0) {
519 				rc = EFAULT;
520 				goto cleanup;
521 			}
522 			break;
523 		default:
524 			rc = EFAULT;
525 			goto cleanup;
526 		}
527 
528 		/* Check API version */
529 		if (setcfg.set_cfg_vers != ISCSIT_API_VERS0) {
530 			rc = EINVAL;
531 			goto cleanup;
532 		}
533 
534 		/* Config is in packed nvlist format so unpack it */
535 		cfg_pnvlist = kmem_alloc(setcfg.set_cfg_pnvlist_len,
536 		    KM_SLEEP);
537 		ASSERT(cfg_pnvlist != NULL);
538 
539 		if (ddi_copyin(setcfg.set_cfg_pnvlist, cfg_pnvlist,
540 		    setcfg.set_cfg_pnvlist_len, flag) != 0) {
541 			rc = EFAULT;
542 			goto cleanup;
543 		}
544 
545 		rc = nvlist_unpack(cfg_pnvlist, setcfg.set_cfg_pnvlist_len,
546 		    &cfg_nvlist, KM_SLEEP);
547 		if (rc != 0) {
548 			goto cleanup;
549 		}
550 
551 		/* Translate nvlist */
552 		rc = it_nv_to_config(cfg_nvlist, &cfg);
553 		if (rc != 0) {
554 			cmn_err(CE_WARN, "Configuration is invalid");
555 			goto cleanup;
556 		}
557 
558 		/* Update config */
559 		rc = iscsit_config_merge(cfg);
560 		/* FALLTHROUGH */
561 
562 cleanup:
563 		if (cfg)
564 			it_config_free_cmn(cfg);
565 		if (cfg_pnvlist)
566 			kmem_free(cfg_pnvlist, setcfg.set_cfg_pnvlist_len);
567 		nvlist_free(cfg_nvlist);
568 
569 		/*
570 		 * Now that the reconfig is complete set our state back to
571 		 * enabled.
572 		 */
573 		mutex_enter(&iscsit_global.global_state_mutex);
574 		iscsit_global.global_svc_state = ISE_ENABLED;
575 		mutex_exit(&iscsit_global.global_state_mutex);
576 		break;
577 	case ISCSIT_IOC_ENABLE_SVC: {
578 		iscsit_hostinfo_t hostinfo;
579 
580 		if (ddi_copyin((void *)argp, &hostinfo.length,
581 		    sizeof (hostinfo.length), flag) != 0) {
582 			mutex_enter(&iscsit_global.global_state_mutex);
583 			iscsit_global.global_svc_state = ISE_DISABLED;
584 			mutex_exit(&iscsit_global.global_state_mutex);
585 			return (EFAULT);
586 		}
587 
588 		if (hostinfo.length > sizeof (hostinfo.fqhn))
589 			hostinfo.length = sizeof (hostinfo.fqhn);
590 
591 		if (ddi_copyin((void *)((caddr_t)argp +
592 		    sizeof (hostinfo.length)), &hostinfo.fqhn,
593 		    hostinfo.length, flag) != 0) {
594 			mutex_enter(&iscsit_global.global_state_mutex);
595 			iscsit_global.global_svc_state = ISE_DISABLED;
596 			mutex_exit(&iscsit_global.global_state_mutex);
597 			return (EFAULT);
598 		}
599 
600 		idmrc = iscsit_enable_svc(&hostinfo);
601 		mutex_enter(&iscsit_global.global_state_mutex);
602 		if (idmrc == IDM_STATUS_SUCCESS) {
603 			iscsit_global.global_svc_state = ISE_ENABLED;
604 		} else {
605 			rc = EIO;
606 			iscsit_global.global_svc_state = ISE_DISABLED;
607 		}
608 		mutex_exit(&iscsit_global.global_state_mutex);
609 		break;
610 	}
611 	case ISCSIT_IOC_DISABLE_SVC:
612 		iscsit_disable_svc();
613 		mutex_enter(&iscsit_global.global_state_mutex);
614 		iscsit_global.global_svc_state = ISE_DISABLED;
615 		mutex_exit(&iscsit_global.global_state_mutex);
616 		break;
617 
618 	default:
619 		rc = EINVAL;
620 		mutex_enter(&iscsit_global.global_state_mutex);
621 		iscsit_global.global_svc_state = ISE_ENABLED;
622 		mutex_exit(&iscsit_global.global_state_mutex);
623 	}
624 
625 	return (rc);
626 }
627 
628 static idm_status_t
629 iscsit_init(dev_info_t *dip)
630 {
631 	int			rc;
632 
633 	rc = ldi_ident_from_dip(dip, &iscsit_global.global_li);
634 	ASSERT(rc == 0);  /* Failure indicates invalid argument */
635 
636 	iscsit_global.global_svc_state = ISE_DISABLED;
637 
638 	return (IDM_STATUS_SUCCESS);
639 }
640 
641 /*
642  * iscsit_enable_svc
643  *
644  * registers all the configured targets and target portals with STMF
645  */
646 static idm_status_t
647 iscsit_enable_svc(iscsit_hostinfo_t *hostinfo)
648 {
649 	stmf_port_provider_t	*pp;
650 	stmf_dbuf_store_t	*dbuf_store;
651 	boolean_t		did_iscsit_isns_init;
652 	idm_status_t		retval = IDM_STATUS_SUCCESS;
653 
654 	ASSERT(iscsit_global.global_svc_state == ISE_ENABLING);
655 
656 	/*
657 	 * Make sure that can tell if we have partially allocated
658 	 * in case we need to exit and tear down anything allocated.
659 	 */
660 	iscsit_global.global_tsih_pool = NULL;
661 	iscsit_global.global_dbuf_store = NULL;
662 	iscsit_status_pdu_cache = NULL;
663 	pp = NULL;
664 	iscsit_global.global_pp = NULL;
665 	iscsit_global.global_default_tpg = NULL;
666 	did_iscsit_isns_init = B_FALSE;
667 	iscsit_global.global_dispatch_taskq = NULL;
668 
669 	/* Setup remaining fields in iscsit_global_t */
670 	idm_refcnt_init(&iscsit_global.global_refcnt,
671 	    &iscsit_global);
672 
673 	avl_create(&iscsit_global.global_discovery_sessions,
674 	    iscsit_sess_avl_compare, sizeof (iscsit_sess_t),
675 	    offsetof(iscsit_sess_t, ist_tgt_ln));
676 
677 	avl_create(&iscsit_global.global_target_list,
678 	    iscsit_tgt_avl_compare, sizeof (iscsit_tgt_t),
679 	    offsetof(iscsit_tgt_t, target_global_ln));
680 
681 	list_create(&iscsit_global.global_deleted_target_list,
682 	    sizeof (iscsit_tgt_t),
683 	    offsetof(iscsit_tgt_t, target_global_deleted_ln));
684 
685 	avl_create(&iscsit_global.global_tpg_list,
686 	    iscsit_tpg_avl_compare, sizeof (iscsit_tpg_t),
687 	    offsetof(iscsit_tpg_t, tpg_global_ln));
688 
689 	avl_create(&iscsit_global.global_ini_list,
690 	    iscsit_ini_avl_compare, sizeof (iscsit_ini_t),
691 	    offsetof(iscsit_ini_t, ini_global_ln));
692 
693 	iscsit_global.global_tsih_pool = vmem_create("iscsit_tsih_pool",
694 	    (void *)1, ISCSI_MAX_TSIH, 1, NULL, NULL, NULL, 0,
695 	    VM_SLEEP | VMC_IDENTIFIER);
696 
697 	/*
698 	 * Setup STMF dbuf store.  Our buffers are bound to a specific
699 	 * connection so we really can't let STMF cache buffers for us.
700 	 * Consequently we'll just allocate one global buffer store.
701 	 */
702 	dbuf_store = stmf_alloc(STMF_STRUCT_DBUF_STORE, 0, 0);
703 	if (dbuf_store == NULL) {
704 		retval = IDM_STATUS_FAIL;
705 		goto tear_down_and_return;
706 	}
707 	dbuf_store->ds_alloc_data_buf = iscsit_dbuf_alloc;
708 	dbuf_store->ds_free_data_buf = iscsit_dbuf_free;
709 	dbuf_store->ds_port_private = NULL;
710 	iscsit_global.global_dbuf_store = dbuf_store;
711 
712 	/* Status PDU cache */
713 	iscsit_status_pdu_cache = kmem_cache_create("iscsit_status_pdu_cache",
714 	    sizeof (idm_pdu_t) + sizeof (iscsi_scsi_rsp_hdr_t), 8,
715 	    &iscsit_status_pdu_constructor,
716 	    NULL, NULL, NULL, NULL, KM_SLEEP);
717 
718 	/* Default TPG and portal */
719 	iscsit_global.global_default_tpg = iscsit_tpg_createdefault();
720 	if (iscsit_global.global_default_tpg == NULL) {
721 		retval = IDM_STATUS_FAIL;
722 		goto tear_down_and_return;
723 	}
724 
725 	/* initialize isns client */
726 	(void) iscsit_isns_init(hostinfo);
727 	did_iscsit_isns_init = B_TRUE;
728 
729 	/* Register port provider */
730 	pp = stmf_alloc(STMF_STRUCT_PORT_PROVIDER, 0, 0);
731 	if (pp == NULL) {
732 		retval = IDM_STATUS_FAIL;
733 		goto tear_down_and_return;
734 	}
735 
736 	pp->pp_portif_rev = PORTIF_REV_1;
737 	pp->pp_instance = 0;
738 	pp->pp_name = ISCSIT_MODNAME;
739 	pp->pp_cb = iscsit_pp_cb;
740 
741 	iscsit_global.global_pp = pp;
742 
743 
744 	if (stmf_register_port_provider(pp) != STMF_SUCCESS) {
745 		retval = IDM_STATUS_FAIL;
746 		goto tear_down_and_return;
747 	}
748 
749 	iscsit_global.global_dispatch_taskq = taskq_create("iscsit_dispatch",
750 	    1, minclsyspri, 16, 16, TASKQ_PREPOPULATE);
751 
752 	/* Scan staged PDUs, meaningful in MC/S situations */
753 	iscsit_rxpdu_queue_monitor_start();
754 
755 	return (IDM_STATUS_SUCCESS);
756 
757 tear_down_and_return:
758 
759 	if (iscsit_global.global_dispatch_taskq) {
760 		taskq_destroy(iscsit_global.global_dispatch_taskq);
761 		iscsit_global.global_dispatch_taskq = NULL;
762 	}
763 
764 	if (did_iscsit_isns_init)
765 		iscsit_isns_fini();
766 
767 	if (iscsit_global.global_default_tpg) {
768 		iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
769 		iscsit_global.global_default_tpg = NULL;
770 	}
771 
772 	if (iscsit_global.global_pp)
773 		iscsit_global.global_pp = NULL;
774 
775 	if (pp)
776 		stmf_free(pp);
777 
778 	if (iscsit_status_pdu_cache) {
779 		kmem_cache_destroy(iscsit_status_pdu_cache);
780 		iscsit_status_pdu_cache = NULL;
781 	}
782 
783 	if (iscsit_global.global_dbuf_store) {
784 		stmf_free(iscsit_global.global_dbuf_store);
785 		iscsit_global.global_dbuf_store = NULL;
786 	}
787 
788 	if (iscsit_global.global_tsih_pool) {
789 		vmem_destroy(iscsit_global.global_tsih_pool);
790 		iscsit_global.global_tsih_pool = NULL;
791 	}
792 
793 	avl_destroy(&iscsit_global.global_ini_list);
794 	avl_destroy(&iscsit_global.global_tpg_list);
795 	list_destroy(&iscsit_global.global_deleted_target_list);
796 	avl_destroy(&iscsit_global.global_target_list);
797 	avl_destroy(&iscsit_global.global_discovery_sessions);
798 
799 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
800 
801 	return (retval);
802 }
803 
804 /*
805  * iscsit_disable_svc
806  *
807  * clean up all existing connections and deregister targets from STMF
808  */
809 static void
810 iscsit_disable_svc(void)
811 {
812 	iscsit_sess_t	*sess;
813 
814 	ASSERT(iscsit_global.global_svc_state == ISE_DISABLING);
815 
816 	iscsit_rxpdu_queue_monitor_stop();
817 
818 	/* tear down discovery sessions */
819 	for (sess = avl_first(&iscsit_global.global_discovery_sessions);
820 	    sess != NULL;
821 	    sess = AVL_NEXT(&iscsit_global.global_discovery_sessions, sess))
822 		iscsit_sess_close(sess);
823 
824 	/*
825 	 * Passing NULL to iscsit_config_merge tells it to go to an empty
826 	 * config.
827 	 */
828 	(void) iscsit_config_merge(NULL);
829 
830 	/*
831 	 * Wait until there are no more global references
832 	 */
833 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
834 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
835 
836 	/*
837 	 * Default TPG must be destroyed after global_refcnt is 0.
838 	 */
839 	iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
840 
841 	avl_destroy(&iscsit_global.global_discovery_sessions);
842 	list_destroy(&iscsit_global.global_deleted_target_list);
843 	avl_destroy(&iscsit_global.global_target_list);
844 	avl_destroy(&iscsit_global.global_tpg_list);
845 	avl_destroy(&iscsit_global.global_ini_list);
846 
847 	taskq_destroy(iscsit_global.global_dispatch_taskq);
848 
849 	iscsit_isns_fini();
850 
851 	stmf_free(iscsit_global.global_dbuf_store);
852 	iscsit_global.global_dbuf_store = NULL;
853 
854 	(void) stmf_deregister_port_provider(iscsit_global.global_pp);
855 	stmf_free(iscsit_global.global_pp);
856 	iscsit_global.global_pp = NULL;
857 
858 	kmem_cache_destroy(iscsit_status_pdu_cache);
859 	iscsit_status_pdu_cache = NULL;
860 
861 	vmem_destroy(iscsit_global.global_tsih_pool);
862 	iscsit_global.global_tsih_pool = NULL;
863 }
864 
865 void
866 iscsit_global_hold()
867 {
868 	/*
869 	 * To take out a global hold, we must either own the global
870 	 * state mutex or we must be running inside of an ioctl that
871 	 * has set the global state to ISE_BUSY, ISE_DISABLING, or
872 	 * ISE_ENABLING.  We don't track the "owner" for these flags,
873 	 * so just checking if they are set is enough for now.
874 	 */
875 	ASSERT((iscsit_global.global_svc_state == ISE_ENABLING) ||
876 	    (iscsit_global.global_svc_state == ISE_DISABLING) ||
877 	    (iscsit_global.global_svc_state == ISE_BUSY) ||
878 	    MUTEX_HELD(&iscsit_global.global_state_mutex));
879 
880 	idm_refcnt_hold(&iscsit_global.global_refcnt);
881 }
882 
883 void
884 iscsit_global_rele()
885 {
886 	idm_refcnt_rele(&iscsit_global.global_refcnt);
887 }
888 
889 void
890 iscsit_global_wait_ref()
891 {
892 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
893 }
894 
895 /*
896  * IDM callbacks
897  */
898 
899 /*ARGSUSED*/
900 void
901 iscsit_rx_pdu(idm_conn_t *ic, idm_pdu_t *rx_pdu)
902 {
903 	iscsit_conn_t *ict = ic->ic_handle;
904 	switch (IDM_PDU_OPCODE(rx_pdu)) {
905 	case ISCSI_OP_SCSI_CMD:
906 		ASSERT(0); /* Shouldn't happen */
907 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
908 		break;
909 	case ISCSI_OP_SNACK_CMD:
910 		/*
911 		 * We'll need to handle this when we support ERL1/2.  For
912 		 * now we treat it as a protocol error.
913 		 */
914 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
915 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
916 		break;
917 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
918 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
919 			iscsit_set_cmdsn(ict, rx_pdu);
920 			iscsit_op_scsi_task_mgmt(ict, rx_pdu);
921 		}
922 		break;
923 	case ISCSI_OP_NOOP_OUT:
924 	case ISCSI_OP_LOGIN_CMD:
925 	case ISCSI_OP_TEXT_CMD:
926 	case ISCSI_OP_LOGOUT_CMD:
927 		/*
928 		 * If/when we switch to userland processing these PDU's
929 		 * will be handled by iscsitd.
930 		 */
931 		iscsit_deferred_dispatch(rx_pdu);
932 		break;
933 	default:
934 		/* Protocol error */
935 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
936 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
937 		break;
938 	}
939 }
940 
941 /*ARGSUSED*/
942 void
943 iscsit_rx_pdu_error(idm_conn_t *ic, idm_pdu_t *rx_pdu, idm_status_t status)
944 {
945 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
946 }
947 
948 /*
949  * iscsit_rx_scsi_rsp -- cause the connection to be closed if response rx'd
950  *
951  * A target sends an SCSI Response PDU, it should never receive one.
952  * This has been seen when running the Codemonicon suite of tests which
953  * does negative testing of the protocol. If such a condition occurs using
954  * a normal initiator it most likely means there's data corruption in the
955  * header and that's grounds for dropping the connection as well.
956  */
957 void
958 iscsit_rx_scsi_rsp(idm_conn_t *ic, idm_pdu_t *rx_pdu)
959 {
960 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
961 	idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
962 }
963 
964 void
965 iscsit_task_aborted(idm_task_t *idt, idm_status_t status)
966 {
967 	iscsit_task_t *itask = idt->idt_private;
968 
969 	switch (status) {
970 	case IDM_STATUS_SUSPENDED:
971 		break;
972 	case IDM_STATUS_ABORTED:
973 		mutex_enter(&itask->it_mutex);
974 		itask->it_aborted = B_TRUE;
975 		/*
976 		 * We rely on the fact that STMF tracks outstanding
977 		 * buffer transfers and will free all of our buffers
978 		 * before freeing the task so we don't need to
979 		 * explicitly free the buffers from iscsit/idm
980 		 */
981 		if (itask->it_stmf_abort) {
982 			mutex_exit(&itask->it_mutex);
983 			/*
984 			 * Task is no longer active
985 			 */
986 			iscsit_task_done(itask);
987 
988 			/*
989 			 * STMF has already asked for this task to be aborted
990 			 *
991 			 * STMF specification is wrong... says to return
992 			 * STMF_ABORTED, the code actually looks for
993 			 * STMF_ABORT_SUCCESS.
994 			 */
995 			stmf_task_lport_aborted(itask->it_stmf_task,
996 			    STMF_ABORT_SUCCESS, STMF_IOF_LPORT_DONE);
997 			return;
998 		} else {
999 			mutex_exit(&itask->it_mutex);
1000 			/*
1001 			 * Tell STMF to stop processing the task.
1002 			 */
1003 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
1004 			    STMF_ABORTED, NULL);
1005 			return;
1006 		}
1007 		/*NOTREACHED*/
1008 	default:
1009 		ASSERT(0);
1010 	}
1011 }
1012 
1013 /*ARGSUSED*/
1014 idm_status_t
1015 iscsit_client_notify(idm_conn_t *ic, idm_client_notify_t icn,
1016     uintptr_t data)
1017 {
1018 	idm_status_t rc = IDM_STATUS_SUCCESS;
1019 
1020 	/*
1021 	 * IDM client notifications will never occur at interrupt level
1022 	 * since they are generated from the connection state machine which
1023 	 * running on taskq threads.
1024 	 *
1025 	 */
1026 	switch (icn) {
1027 	case CN_CONNECT_ACCEPT:
1028 		rc = iscsit_conn_accept(ic); /* No data */
1029 		break;
1030 	case CN_FFP_ENABLED:
1031 		rc = iscsit_ffp_enabled(ic); /* No data */
1032 		break;
1033 	case CN_FFP_DISABLED:
1034 		/*
1035 		 * Data indicates whether this was the result of an
1036 		 * explicit logout request.
1037 		 */
1038 		rc = iscsit_ffp_disabled(ic, (idm_ffp_disable_t)data);
1039 		break;
1040 	case CN_CONNECT_LOST:
1041 		rc = iscsit_conn_lost(ic);
1042 		break;
1043 	case CN_CONNECT_DESTROY:
1044 		rc = iscsit_conn_destroy(ic);
1045 		break;
1046 	case CN_LOGIN_FAIL:
1047 		/*
1048 		 * Force the login state machine to completion
1049 		 */
1050 		rc = iscsit_login_fail(ic);
1051 		break;
1052 	default:
1053 		rc = IDM_STATUS_REJECT;
1054 		break;
1055 	}
1056 
1057 	return (rc);
1058 }
1059 
1060 /*
1061  * iscsit_update_statsn is invoked for all the PDUs which have the StatSN
1062  * field in the header. The StatSN is incremented if the IDM_PDU_ADVANCE_STATSN
1063  * flag is set in the pdu flags field. The StatSN is connection-wide and is
1064  * protected by the mutex ict_statsn_mutex. For Data-In PDUs, if the flag
1065  * IDM_TASK_PHASECOLLAPSE_REQ is set, the status (phase-collapse) is also filled
1066  */
1067 void
1068 iscsit_update_statsn(idm_task_t *idm_task, idm_pdu_t *pdu)
1069 {
1070 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1071 	iscsit_conn_t *ict = (iscsit_conn_t *)pdu->isp_ic->ic_handle;
1072 	iscsit_task_t *itask = NULL;
1073 	scsi_task_t *task = NULL;
1074 
1075 	mutex_enter(&ict->ict_statsn_mutex);
1076 	rsp->statsn = htonl(ict->ict_statsn);
1077 	if (pdu->isp_flags & IDM_PDU_ADVANCE_STATSN)
1078 		ict->ict_statsn++;
1079 	mutex_exit(&ict->ict_statsn_mutex);
1080 
1081 	/*
1082 	 * The last SCSI Data PDU passed for a command may also contain the
1083 	 * status if the status indicates termination with no expections, i.e.
1084 	 * no sense data or response involved. If the command completes with
1085 	 * an error, then the response and sense data will be sent in a
1086 	 * separate iSCSI Response PDU.
1087 	 */
1088 	if ((idm_task) && (idm_task->idt_flags & IDM_TASK_PHASECOLLAPSE_REQ)) {
1089 		itask = idm_task->idt_private;
1090 		task = itask->it_stmf_task;
1091 
1092 		rsp->cmd_status = task->task_scsi_status;
1093 		rsp->flags	|= ISCSI_FLAG_DATA_STATUS;
1094 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1095 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1096 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1097 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1098 		}
1099 		rsp->residual_count = htonl(task->task_resid);
1100 
1101 		/*
1102 		 * Removing the task from the session task list
1103 		 * just before the status is sent in the last
1104 		 * Data PDU transfer
1105 		 */
1106 		iscsit_task_done(itask);
1107 	}
1108 }
1109 
1110 void
1111 iscsit_build_hdr(idm_task_t *idm_task, idm_pdu_t *pdu, uint8_t opcode)
1112 {
1113 	iscsit_task_t *itask = idm_task->idt_private;
1114 	iscsi_data_rsp_hdr_t *dh = (iscsi_data_rsp_hdr_t *)pdu->isp_hdr;
1115 
1116 	/*
1117 	 * We acquired iscsit_sess_t.ist_sn_mutex in iscsit_xfer_scsi_data
1118 	 */
1119 	ASSERT(MUTEX_HELD(&itask->it_ict->ict_sess->ist_sn_mutex));
1120 	/*
1121 	 * On incoming data, the target transfer tag and Lun is only
1122 	 * provided by the target if the A bit is set, Since the target
1123 	 * does not currently support Error Recovery Level 1, the A
1124 	 * bit is never set.
1125 	 */
1126 	dh->opcode = opcode;
1127 	dh->itt = itask->it_itt;
1128 	dh->ttt = ((opcode & ISCSI_OPCODE_MASK) == ISCSI_OP_SCSI_DATA_RSP) ?
1129 	    ISCSI_RSVD_TASK_TAG : itask->it_ttt;
1130 
1131 	dh->expcmdsn = htonl(itask->it_ict->ict_sess->ist_expcmdsn);
1132 	dh->maxcmdsn = htonl(itask->it_ict->ict_sess->ist_maxcmdsn);
1133 
1134 	/*
1135 	 * IDM must set:
1136 	 *
1137 	 * data.flags and rtt.flags
1138 	 * data.dlength
1139 	 * data.datasn
1140 	 * data.offset
1141 	 * statsn, residual_count and cmd_status (for phase collapse)
1142 	 * rtt.rttsn
1143 	 * rtt.data_offset
1144 	 * rtt.data_length
1145 	 */
1146 }
1147 
1148 void
1149 iscsit_keepalive(idm_conn_t *ic)
1150 {
1151 	idm_pdu_t		*nop_in_pdu;
1152 	iscsi_nop_in_hdr_t	*nop_in;
1153 	iscsit_conn_t		*ict = ic->ic_handle;
1154 
1155 	/*
1156 	 * IDM noticed the connection has been idle for too long so it's
1157 	 * time to provoke some activity.  Build and transmit an iSCSI
1158 	 * nop-in PDU -- when the initiator responds it will be counted
1159 	 * as "activity" and keep the connection alive.
1160 	 *
1161 	 * We don't actually care about the response here at the iscsit level
1162 	 * so we will just throw it away without looking at it when it arrives.
1163 	 */
1164 	nop_in_pdu = idm_pdu_alloc(sizeof (*nop_in), 0);
1165 	idm_pdu_init(nop_in_pdu, ic, NULL, NULL);
1166 	nop_in = (iscsi_nop_in_hdr_t *)nop_in_pdu->isp_hdr;
1167 	bzero(nop_in, sizeof (*nop_in));
1168 	nop_in->opcode = ISCSI_OP_NOOP_IN;
1169 	nop_in->flags = ISCSI_FLAG_FINAL;
1170 	nop_in->itt = ISCSI_RSVD_TASK_TAG;
1171 	/*
1172 	 * When the target sends a NOP-In as a Ping, the target transfer tag
1173 	 * is set to a valid (not reserved) value and the initiator task tag
1174 	 * is set to ISCSI_RSVD_TASK_TAG (0xffffffff). In this case the StatSN
1175 	 * will always contain the next sequence number but the StatSN for the
1176 	 * connection is not advanced after this PDU is sent.
1177 	 */
1178 	nop_in_pdu->isp_flags |= IDM_PDU_SET_STATSN;
1179 	/*
1180 	 * This works because we don't currently allocate ttt's anywhere else
1181 	 * in iscsit so as long as we stay out of IDM's range we are safe.
1182 	 * If we need to allocate ttt's for other PDU's in the future this will
1183 	 * need to be improved.
1184 	 */
1185 	mutex_enter(&ict->ict_mutex);
1186 	nop_in->ttt = ict->ict_keepalive_ttt;
1187 	ict->ict_keepalive_ttt++;
1188 	if (ict->ict_keepalive_ttt == ISCSI_RSVD_TASK_TAG)
1189 		ict->ict_keepalive_ttt = IDM_TASKIDS_MAX;
1190 	mutex_exit(&ict->ict_mutex);
1191 
1192 	iscsit_pdu_tx(nop_in_pdu);
1193 }
1194 
1195 static idm_status_t
1196 iscsit_conn_accept(idm_conn_t *ic)
1197 {
1198 	iscsit_conn_t *ict;
1199 
1200 	/*
1201 	 * We need to get a global hold here to ensure that the service
1202 	 * doesn't get shutdown prior to establishing a session. This
1203 	 * gets released in iscsit_conn_destroy().
1204 	 */
1205 	mutex_enter(&iscsit_global.global_state_mutex);
1206 	if (iscsit_global.global_svc_state != ISE_ENABLED) {
1207 		mutex_exit(&iscsit_global.global_state_mutex);
1208 		return (IDM_STATUS_FAIL);
1209 	}
1210 	iscsit_global_hold();
1211 	mutex_exit(&iscsit_global.global_state_mutex);
1212 
1213 	/*
1214 	 * Allocate an associated iscsit structure to represent this
1215 	 * connection.  We shouldn't really create a session until we
1216 	 * get the first login PDU.
1217 	 */
1218 	ict = kmem_zalloc(sizeof (*ict), KM_SLEEP);
1219 
1220 	ict->ict_ic = ic;
1221 	ict->ict_statsn = 1;
1222 	ict->ict_keepalive_ttt = IDM_TASKIDS_MAX; /* Avoid IDM TT range */
1223 	ic->ic_handle = ict;
1224 	mutex_init(&ict->ict_mutex, NULL, MUTEX_DRIVER, NULL);
1225 	mutex_init(&ict->ict_statsn_mutex, NULL, MUTEX_DRIVER, NULL);
1226 	idm_refcnt_init(&ict->ict_refcnt, ict);
1227 
1228 	/*
1229 	 * Initialize login state machine
1230 	 */
1231 	if (iscsit_login_sm_init(ict) != IDM_STATUS_SUCCESS) {
1232 		iscsit_global_rele();
1233 		/*
1234 		 * Cleanup the ict after idm notifies us about this failure
1235 		 */
1236 		return (IDM_STATUS_FAIL);
1237 	}
1238 
1239 	return (IDM_STATUS_SUCCESS);
1240 }
1241 
1242 idm_status_t
1243 iscsit_conn_reinstate(iscsit_conn_t *reinstate_ict, iscsit_conn_t *new_ict)
1244 {
1245 	idm_status_t	result;
1246 
1247 	/*
1248 	 * Note in new connection state that this connection is
1249 	 * reinstating an existing connection.
1250 	 */
1251 	new_ict->ict_reinstating = B_TRUE;
1252 	new_ict->ict_reinstate_conn = reinstate_ict;
1253 	new_ict->ict_statsn = reinstate_ict->ict_statsn;
1254 
1255 	/*
1256 	 * Now generate connection state machine event to existing connection
1257 	 * so that it starts the cleanup process.
1258 	 */
1259 	result = idm_conn_reinstate_event(reinstate_ict->ict_ic,
1260 	    new_ict->ict_ic);
1261 
1262 	return (result);
1263 }
1264 
1265 void
1266 iscsit_conn_hold(iscsit_conn_t *ict)
1267 {
1268 	idm_refcnt_hold(&ict->ict_refcnt);
1269 }
1270 
1271 void
1272 iscsit_conn_rele(iscsit_conn_t *ict)
1273 {
1274 	idm_refcnt_rele(&ict->ict_refcnt);
1275 }
1276 
1277 void
1278 iscsit_conn_dispatch_hold(iscsit_conn_t *ict)
1279 {
1280 	idm_refcnt_hold(&ict->ict_dispatch_refcnt);
1281 }
1282 
1283 void
1284 iscsit_conn_dispatch_rele(iscsit_conn_t *ict)
1285 {
1286 	idm_refcnt_rele(&ict->ict_dispatch_refcnt);
1287 }
1288 
1289 static idm_status_t
1290 iscsit_login_fail(idm_conn_t *ic)
1291 {
1292 	iscsit_conn_t *ict = ic->ic_handle;
1293 
1294 	/* Generate login state machine event */
1295 	iscsit_login_sm_event(ict, ILE_LOGIN_CONN_ERROR, NULL);
1296 
1297 	return (IDM_STATUS_SUCCESS);
1298 }
1299 
1300 static idm_status_t
1301 iscsit_ffp_enabled(idm_conn_t *ic)
1302 {
1303 	iscsit_conn_t *ict = ic->ic_handle;
1304 
1305 	/* Generate session state machine event */
1306 	iscsit_sess_sm_event(ict->ict_sess, SE_CONN_LOGGED_IN, ict);
1307 
1308 	return (IDM_STATUS_SUCCESS);
1309 }
1310 
1311 static idm_status_t
1312 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class)
1313 {
1314 	iscsit_conn_t *ict = ic->ic_handle;
1315 
1316 	/* Generate session state machine event */
1317 	switch (disable_class) {
1318 	case FD_CONN_FAIL:
1319 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_FAIL, ict);
1320 		break;
1321 	case FD_CONN_LOGOUT:
1322 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_DISABLE, ict);
1323 		break;
1324 	case FD_SESS_LOGOUT:
1325 		iscsit_sess_sm_event(ict->ict_sess, SE_SESSION_CLOSE, ict);
1326 		break;
1327 	default:
1328 		ASSERT(0);
1329 	}
1330 
1331 	return (IDM_STATUS_SUCCESS);
1332 }
1333 
1334 static idm_status_t
1335 iscsit_conn_lost(idm_conn_t *ic)
1336 {
1337 	iscsit_conn_t	*ict	= ic->ic_handle;
1338 	iscsit_sess_t	*ist	= ict->ict_sess;
1339 	iscsit_cbuf_t	*cbuf;
1340 	idm_pdu_t	*rx_pdu;
1341 	int i;
1342 
1343 	mutex_enter(&ict->ict_mutex);
1344 	ict->ict_lost = B_TRUE;
1345 	mutex_exit(&ict->ict_mutex);
1346 	/*
1347 	 * scrub the staging queue for all PDUs on this connection
1348 	 */
1349 	if (ist != NULL) {
1350 		mutex_enter(&ist->ist_sn_mutex);
1351 		for (cbuf = ist->ist_rxpdu_queue, i = 0;
1352 		    ((cbuf->cb_num_elems > 0) && (i < ISCSIT_RXPDU_QUEUE_LEN));
1353 		    i++) {
1354 			if (((rx_pdu = cbuf->cb_buffer[i]) != NULL) &&
1355 			    (rx_pdu->isp_ic == ic)) {
1356 				/* conn is lost, drop the pdu */
1357 				DTRACE_PROBE3(scrubbing__staging__queue,
1358 				    iscsit_sess_t *, ist, idm_conn_t *, ic,
1359 				    idm_pdu_t *, rx_pdu);
1360 				idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
1361 				cbuf->cb_buffer[i] = NULL;
1362 				cbuf->cb_num_elems--;
1363 				iscsit_conn_dispatch_rele(ict);
1364 			}
1365 		}
1366 		mutex_exit(&ist->ist_sn_mutex);
1367 	}
1368 	/*
1369 	 * Make sure there aren't any PDU's transitioning from the receive
1370 	 * handler to the dispatch taskq.
1371 	 */
1372 	idm_refcnt_wait_ref(&ict->ict_dispatch_refcnt);
1373 
1374 	return (IDM_STATUS_SUCCESS);
1375 }
1376 
1377 static idm_status_t
1378 iscsit_conn_destroy(idm_conn_t *ic)
1379 {
1380 	iscsit_conn_t *ict = ic->ic_handle;
1381 
1382 	mutex_enter(&ict->ict_mutex);
1383 	ict->ict_destroyed = B_TRUE;
1384 	mutex_exit(&ict->ict_mutex);
1385 
1386 	/* Generate session state machine event */
1387 	if (ict->ict_sess != NULL) {
1388 		/*
1389 		 * Session state machine will call iscsit_conn_destroy_done()
1390 		 * when it has removed references to this connection.
1391 		 */
1392 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FAIL, ict);
1393 	}
1394 
1395 	idm_refcnt_wait_ref(&ict->ict_refcnt);
1396 	/*
1397 	 * The session state machine does not need to post
1398 	 * events to IDM any longer, so it is safe to set
1399 	 * the idm connection reference to NULL
1400 	 */
1401 	ict->ict_ic = NULL;
1402 
1403 	/* Reap the login state machine */
1404 	iscsit_login_sm_fini(ict);
1405 
1406 	/* Clean up any text command remnants */
1407 	iscsit_text_cmd_fini(ict);
1408 
1409 	mutex_destroy(&ict->ict_mutex);
1410 	idm_refcnt_destroy(&ict->ict_refcnt);
1411 	kmem_free(ict, sizeof (*ict));
1412 
1413 	iscsit_global_rele();
1414 
1415 	return (IDM_STATUS_SUCCESS);
1416 }
1417 
1418 void
1419 iscsit_conn_logout(iscsit_conn_t *ict)
1420 {
1421 	/*
1422 	 * If the iscsi connection is active, then
1423 	 * logout the IDM connection by sending a
1424 	 * CE_LOGOUT_SESSION_SUCCESS, else, no action
1425 	 * needs to be taken because the connection
1426 	 * is already in the teardown process.
1427 	 */
1428 	mutex_enter(&ict->ict_mutex);
1429 	if (ict->ict_lost == B_FALSE && ict->ict_destroyed == B_FALSE) {
1430 		idm_conn_event(ict->ict_ic, CE_LOGOUT_SESSION_SUCCESS, NULL);
1431 	}
1432 	mutex_exit(&ict->ict_mutex);
1433 }
1434 
1435 /*
1436  * STMF-related functions
1437  *
1438  * iSCSI to STMF mapping
1439  *
1440  * Session == ?
1441  * Connection == bound to local port but not itself a local port
1442  * Target
1443  * Target portal (group?) == local port (really but we're not going to do this)
1444  *	iscsit needs to map connections to local ports (whatever we decide
1445  * 	they are)
1446  * Target == ?
1447  */
1448 
1449 /*ARGSUSED*/
1450 static stmf_data_buf_t *
1451 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
1452     uint32_t flags)
1453 {
1454 	iscsit_task_t *itask = task->task_port_private;
1455 	idm_buf_t *idm_buffer;
1456 	iscsit_buf_t	*ibuf;
1457 	stmf_data_buf_t *result;
1458 	uint32_t	bsize;
1459 
1460 	/*
1461 	 * If the requested size is larger than MaxBurstLength and the
1462 	 * given pminsize is also larger than MaxBurstLength, then the
1463 	 * allocation fails (dbuf = NULL) and pminsize is modified to
1464 	 * be equal to MaxBurstLength. stmf/sbd then should re-invoke
1465 	 * this function with the corrected values for transfer.
1466 	 */
1467 	ASSERT(pminsize);
1468 	if (size <= itask->it_ict->ict_op.op_max_burst_length) {
1469 		bsize = size;
1470 	} else if (*pminsize <= itask->it_ict->ict_op.op_max_burst_length) {
1471 		bsize = itask->it_ict->ict_op.op_max_burst_length;
1472 	} else {
1473 		*pminsize = itask->it_ict->ict_op.op_max_burst_length;
1474 		return (NULL);
1475 	}
1476 
1477 	/* Alloc buffer */
1478 	idm_buffer = idm_buf_alloc(itask->it_ict->ict_ic, NULL, bsize);
1479 	if (idm_buffer != NULL) {
1480 		result = stmf_alloc(STMF_STRUCT_DATA_BUF,
1481 		    sizeof (iscsit_buf_t), 0);
1482 		if (result != NULL) {
1483 			/* Fill in stmf_data_buf_t */
1484 			ibuf = result->db_port_private;
1485 			ibuf->ibuf_idm_buf = idm_buffer;
1486 			ibuf->ibuf_stmf_buf = result;
1487 			ibuf->ibuf_is_immed = B_FALSE;
1488 			result->db_flags = DB_DONT_CACHE;
1489 			result->db_buf_size = bsize;
1490 			result->db_data_size = bsize;
1491 			result->db_sglist_length = 1;
1492 			result->db_sglist[0].seg_addr = idm_buffer->idb_buf;
1493 			result->db_sglist[0].seg_length =
1494 			    idm_buffer->idb_buflen;
1495 			return (result);
1496 		}
1497 
1498 		/* Couldn't get the stmf_data_buf_t so free the buffer */
1499 		idm_buf_free(idm_buffer);
1500 	}
1501 
1502 	return (NULL);
1503 }
1504 
1505 /*ARGSUSED*/
1506 static void
1507 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf)
1508 {
1509 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1510 
1511 	if (ibuf->ibuf_is_immed) {
1512 		/*
1513 		 * The iscsit_buf_t structure itself will be freed with its
1514 		 * associated task.  Here we just need to free the PDU that
1515 		 * held the immediate data.
1516 		 */
1517 		idm_pdu_complete(ibuf->ibuf_immed_data_pdu, IDM_STATUS_SUCCESS);
1518 		ibuf->ibuf_immed_data_pdu = 0;
1519 	} else {
1520 		idm_buf_free(ibuf->ibuf_idm_buf);
1521 		stmf_free(dbuf);
1522 	}
1523 }
1524 
1525 /*ARGSUSED*/
1526 stmf_status_t
1527 iscsit_xfer_scsi_data(scsi_task_t *task, stmf_data_buf_t *dbuf,
1528     uint32_t ioflags)
1529 {
1530 	iscsit_task_t *iscsit_task = task->task_port_private;
1531 	iscsit_sess_t *ict_sess = iscsit_task->it_ict->ict_sess;
1532 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1533 	int idm_rc;
1534 
1535 	/*
1536 	 * If we are aborting then we can ignore this request
1537 	 */
1538 	if (iscsit_task->it_stmf_abort) {
1539 		return (STMF_SUCCESS);
1540 	}
1541 
1542 	/*
1543 	 * If it's not immediate data then start the transfer
1544 	 */
1545 	if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
1546 		if (ibuf->ibuf_is_immed)
1547 			return (iscsit_idm_to_stmf(IDM_STATUS_SUCCESS));
1548 		/*
1549 		 * The DB_SEND_STATUS_GOOD flag in the STMF data buffer allows
1550 		 * the port provider to phase-collapse, i.e. send the status
1551 		 * along with the final data PDU for the command. The port
1552 		 * provider passes this request to the transport layer by
1553 		 * setting a flag IDM_TASK_PHASECOLLAPSE_REQ in the task.
1554 		 */
1555 		if (dbuf->db_flags & DB_SEND_STATUS_GOOD)
1556 			iscsit_task->it_idm_task->idt_flags |=
1557 			    IDM_TASK_PHASECOLLAPSE_REQ;
1558 		/*
1559 		 * IDM will call iscsit_build_hdr so lock now to serialize
1560 		 * access to the SN values.  We need to lock here to enforce
1561 		 * lock ordering
1562 		 */
1563 		mutex_enter(&ict_sess->ist_sn_mutex);
1564 		idm_rc = idm_buf_tx_to_ini(iscsit_task->it_idm_task,
1565 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1566 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1567 		mutex_exit(&ict_sess->ist_sn_mutex);
1568 
1569 		return (iscsit_idm_to_stmf(idm_rc));
1570 	} else if (dbuf->db_flags & DB_DIRECTION_FROM_RPORT) {
1571 		ASSERT(ibuf->ibuf_is_immed == B_FALSE);
1572 		/* Grab the SN lock (see comment above) */
1573 		mutex_enter(&ict_sess->ist_sn_mutex);
1574 		idm_rc = idm_buf_rx_from_ini(iscsit_task->it_idm_task,
1575 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1576 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1577 		mutex_exit(&ict_sess->ist_sn_mutex);
1578 
1579 		return (iscsit_idm_to_stmf(idm_rc));
1580 	}
1581 
1582 	/* What are we supposed to do if there is no direction? */
1583 	return (STMF_INVALID_ARG);
1584 }
1585 
1586 static void
1587 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status)
1588 {
1589 	iscsit_task_t *itask = idb->idb_task_binding->idt_private;
1590 	stmf_data_buf_t *dbuf = idb->idb_cb_arg;
1591 
1592 	dbuf->db_xfer_status = iscsit_idm_to_stmf(status);
1593 
1594 	/*
1595 	 * If the task has been aborted then we don't need to call STMF
1596 	 */
1597 	if (itask->it_stmf_abort) {
1598 		return;
1599 	}
1600 
1601 	/*
1602 	 * For ISCSI over TCP (not iSER), the last SCSI Data PDU passed
1603 	 * for a successful command contains the status as requested by
1604 	 * by COMSTAR (via the DB_SEND_STATUS_GOOD flag). But the iSER
1605 	 * transport does not support phase-collapse. So pretend we are
1606 	 * COMSTAR and send the status in a separate PDU now.
1607 	 */
1608 	if (idb->idb_task_binding->idt_flags & IDM_TASK_PHASECOLLAPSE_SUCCESS) {
1609 		/*
1610 		 * Mark task complete and notify COMSTAR
1611 		 * that the status has been sent.
1612 		 */
1613 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1614 		stmf_send_status_done(itask->it_stmf_task,
1615 		    iscsit_idm_to_stmf(status), STMF_IOF_LPORT_DONE);
1616 	} else if ((dbuf->db_flags & DB_SEND_STATUS_GOOD) &&
1617 	    status == IDM_STATUS_SUCCESS) {
1618 
1619 		/*
1620 		 * The iscsi target port provider - for iSER, emulates the
1621 		 * DB_SEND_STATUS_GOOD optimization if requested by STMF;
1622 		 * it sends the status in a separate PDU after the data
1623 		 * transfer. In this case the port provider should first
1624 		 * call stmf_data_xfer_done() to mark the transfer complete
1625 		 * and then send the status. Although STMF will free the
1626 		 * buffer at the time the task is freed, even if the transfer
1627 		 * is not marked complete, this behavior makes statistics
1628 		 * gathering and task state tracking more difficult than it
1629 		 * needs to be.
1630 		 */
1631 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1632 		if (iscsit_send_scsi_status(itask->it_stmf_task, 0)
1633 		    != STMF_SUCCESS) {
1634 			stmf_send_status_done(itask->it_stmf_task,
1635 			    STMF_FAILURE, STMF_IOF_LPORT_DONE);
1636 		}
1637 	} else {
1638 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1639 		/* don't touch dbuf after stmf_data_xfer_done */
1640 	}
1641 }
1642 
1643 
1644 /*ARGSUSED*/
1645 stmf_status_t
1646 iscsit_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
1647 {
1648 	iscsit_task_t *itask = task->task_port_private;
1649 	iscsi_scsi_rsp_hdr_t *rsp;
1650 	idm_pdu_t *pdu;
1651 	int resp_datalen;
1652 
1653 	/*
1654 	 * If this task is aborted then we don't need to respond.
1655 	 */
1656 	if (itask->it_stmf_abort) {
1657 		return (STMF_SUCCESS);
1658 	}
1659 
1660 	/*
1661 	 * If this is a task management status, handle it elsewhere.
1662 	 */
1663 	if (task->task_mgmt_function != TM_NONE) {
1664 		/*
1665 		 * Don't wait for the PDU completion to tell STMF
1666 		 * the task is done -- it doesn't really matter and
1667 		 * it makes life complicated if STMF later asks us to
1668 		 * abort the request and we don't know whether the
1669 		 * status has been sent or not.
1670 		 */
1671 		itask->it_tm_responded = B_TRUE;
1672 		iscsit_send_task_mgmt_resp(itask->it_tm_pdu,
1673 		    (task->task_completion_status == STMF_SUCCESS) ?
1674 		    SCSI_TCP_TM_RESP_COMPLETE : SCSI_TCP_TM_RESP_FUNC_NOT_SUPP);
1675 		stmf_send_status_done(task, STMF_SUCCESS,
1676 		    STMF_IOF_LPORT_DONE);
1677 		return (STMF_SUCCESS);
1678 	}
1679 
1680 	/*
1681 	 * Remove the task from the session task list
1682 	 */
1683 	iscsit_task_done(itask);
1684 
1685 	/*
1686 	 * Send status
1687 	 */
1688 	mutex_enter(&itask->it_idm_task->idt_mutex);
1689 	if ((itask->it_idm_task->idt_state == TASK_ACTIVE) &&
1690 	    (task->task_completion_status == STMF_SUCCESS) &&
1691 	    (task->task_sense_length == 0) &&
1692 	    (task->task_resid == 0)) {
1693 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1694 		/* PDU callback releases task hold */
1695 		idm_task_hold(itask->it_idm_task);
1696 		mutex_exit(&itask->it_idm_task->idt_mutex);
1697 		/*
1698 		 * Fast path.  Cached status PDU's are already
1699 		 * initialized.  We just need to fill in
1700 		 * connection and task information. StatSN is
1701 		 * incremented by 1 for every status sent a
1702 		 * connection.
1703 		 */
1704 		pdu = kmem_cache_alloc(iscsit_status_pdu_cache, KM_SLEEP);
1705 		pdu->isp_ic = itask->it_ict->ict_ic;
1706 		pdu->isp_private = itask;
1707 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1708 
1709 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1710 		rsp->itt = itask->it_itt;
1711 		/*
1712 		 * ExpDataSN is the number of R2T and Data-In (read)
1713 		 * PDUs the target has sent for the SCSI command.
1714 		 *
1715 		 * Since there is no support for bidirectional transfer
1716 		 * yet, either idt_exp_datasn or idt_exp_rttsn, but not
1717 		 * both is valid at any time
1718 		 */
1719 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1720 		    htonl(itask->it_idm_task->idt_exp_datasn):
1721 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1722 		rsp->cmd_status = task->task_scsi_status;
1723 		iscsit_pdu_tx(pdu);
1724 		return (STMF_SUCCESS);
1725 	} else {
1726 		if (itask->it_idm_task->idt_state != TASK_ACTIVE) {
1727 			mutex_exit(&itask->it_idm_task->idt_mutex);
1728 			return (STMF_FAILURE);
1729 		}
1730 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1731 		/* PDU callback releases task hold */
1732 		idm_task_hold(itask->it_idm_task);
1733 		mutex_exit(&itask->it_idm_task->idt_mutex);
1734 
1735 		resp_datalen = (task->task_sense_length == 0) ? 0 :
1736 		    (task->task_sense_length + sizeof (uint16_t));
1737 
1738 		pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
1739 		idm_pdu_init(pdu, itask->it_ict->ict_ic, itask,
1740 		    iscsit_send_status_done);
1741 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1742 
1743 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1744 		bzero(rsp, sizeof (*rsp));
1745 		rsp->opcode = ISCSI_OP_SCSI_RSP;
1746 
1747 		rsp->flags = ISCSI_FLAG_FINAL;
1748 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1749 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1750 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1751 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1752 		}
1753 
1754 		rsp->bi_residual_count = 0;
1755 		rsp->residual_count = htonl(task->task_resid);
1756 		rsp->itt = itask->it_itt;
1757 		rsp->response = ISCSI_STATUS_CMD_COMPLETED;
1758 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1759 		    htonl(itask->it_idm_task->idt_exp_datasn):
1760 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1761 		rsp->cmd_status = task->task_scsi_status;
1762 		if (task->task_sense_length != 0) {
1763 			/*
1764 			 * Add a byte to provide the sense length in
1765 			 * the response
1766 			 */
1767 			*(uint16_t *)((void *)pdu->isp_data) =
1768 			    htons(task->task_sense_length);
1769 			bcopy(task->task_sense_data,
1770 			    (uint8_t *)pdu->isp_data +
1771 			    sizeof (uint16_t),
1772 			    task->task_sense_length);
1773 			hton24(rsp->dlength, resp_datalen);
1774 		}
1775 
1776 		DTRACE_PROBE5(iscsi__scsi__response,
1777 		    iscsit_conn_t *, itask->it_ict,
1778 		    uint8_t, rsp->response,
1779 		    uint8_t, rsp->cmd_status,
1780 		    idm_pdu_t *, pdu,
1781 		    scsi_task_t *, task);
1782 
1783 		iscsit_pdu_tx(pdu);
1784 
1785 		return (STMF_SUCCESS);
1786 	}
1787 }
1788 
1789 /*ARGSUSED*/
1790 static void
1791 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status)
1792 {
1793 	iscsit_task_t	*itask;
1794 	boolean_t	aborted;
1795 
1796 	itask = pdu->isp_private;
1797 	aborted = itask->it_stmf_abort;
1798 
1799 	/*
1800 	 * After releasing the hold the task may be freed at any time so
1801 	 * don't touch it.
1802 	 */
1803 	idm_task_rele(itask->it_idm_task);
1804 	if (!aborted) {
1805 		stmf_send_status_done(itask->it_stmf_task,
1806 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1807 	}
1808 	kmem_cache_free(iscsit_status_pdu_cache, pdu);
1809 }
1810 
1811 /*ARGSUSED*/
1812 static void
1813 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status)
1814 {
1815 	iscsit_task_t	 *itask;
1816 	boolean_t	aborted;
1817 
1818 	itask = pdu->isp_private;
1819 	aborted = itask->it_stmf_abort;
1820 
1821 	/*
1822 	 * After releasing the hold the task may be freed at any time so
1823 	 * don't touch it.
1824 	 */
1825 	idm_task_rele(itask->it_idm_task);
1826 	if (!aborted) {
1827 		stmf_send_status_done(itask->it_stmf_task,
1828 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1829 	}
1830 	idm_pdu_free(pdu);
1831 }
1832 
1833 
1834 void
1835 iscsit_lport_task_free(scsi_task_t *task)
1836 {
1837 	iscsit_task_t *itask = task->task_port_private;
1838 
1839 	/* We only call idm_task_start for regular tasks, not task management */
1840 	if (task->task_mgmt_function == TM_NONE) {
1841 		idm_task_done(itask->it_idm_task);
1842 		iscsit_task_free(itask);
1843 		return;
1844 	} else {
1845 		iscsit_tm_task_free(itask);
1846 	}
1847 }
1848 
1849 /*ARGSUSED*/
1850 stmf_status_t
1851 iscsit_abort(stmf_local_port_t *lport, int abort_cmd, void *arg, uint32_t flags)
1852 {
1853 	scsi_task_t	*st = (scsi_task_t *)arg;
1854 	iscsit_task_t	*iscsit_task;
1855 	idm_task_t	*idt;
1856 
1857 	/*
1858 	 * If this is a task management request then there's really not much to
1859 	 * do.
1860 	 */
1861 	if (st->task_mgmt_function != TM_NONE) {
1862 		return (STMF_ABORT_SUCCESS);
1863 	}
1864 
1865 	/*
1866 	 * Regular task, start cleaning up
1867 	 */
1868 	iscsit_task = st->task_port_private;
1869 	idt = iscsit_task->it_idm_task;
1870 	mutex_enter(&iscsit_task->it_mutex);
1871 	iscsit_task->it_stmf_abort = B_TRUE;
1872 	if (iscsit_task->it_aborted) {
1873 		mutex_exit(&iscsit_task->it_mutex);
1874 		/*
1875 		 * Task is no longer active
1876 		 */
1877 		iscsit_task_done(iscsit_task);
1878 
1879 		/*
1880 		 * STMF specification is wrong... says to return
1881 		 * STMF_ABORTED, the code actually looks for
1882 		 * STMF_ABORT_SUCCESS.
1883 		 */
1884 		return (STMF_ABORT_SUCCESS);
1885 	} else {
1886 		mutex_exit(&iscsit_task->it_mutex);
1887 		/*
1888 		 * Call IDM to abort the task.  Due to a variety of
1889 		 * circumstances the task may already be in the process of
1890 		 * aborting.
1891 		 * We'll let IDM worry about rationalizing all that except
1892 		 * for one particular instance.  If the state of the task
1893 		 * is TASK_COMPLETE, we need to indicate to the framework
1894 		 * that we are in fact done.  This typically happens with
1895 		 * framework-initiated task management type requests
1896 		 * (e.g. abort task).
1897 		 */
1898 		if (idt->idt_state == TASK_COMPLETE) {
1899 			idm_refcnt_wait_ref(&idt->idt_refcnt);
1900 			return (STMF_ABORT_SUCCESS);
1901 		} else {
1902 			idm_task_abort(idt->idt_ic, idt, AT_TASK_MGMT_ABORT);
1903 			return (STMF_SUCCESS);
1904 		}
1905 	}
1906 
1907 	/*NOTREACHED*/
1908 }
1909 
1910 /*ARGSUSED*/
1911 void
1912 iscsit_ctl(stmf_local_port_t *lport, int cmd, void *arg)
1913 {
1914 	iscsit_tgt_t		*iscsit_tgt;
1915 
1916 	ASSERT((cmd == STMF_CMD_LPORT_ONLINE) ||
1917 	    (cmd == STMF_ACK_LPORT_ONLINE_COMPLETE) ||
1918 	    (cmd == STMF_CMD_LPORT_OFFLINE) ||
1919 	    (cmd == STMF_ACK_LPORT_OFFLINE_COMPLETE));
1920 
1921 	iscsit_tgt = (iscsit_tgt_t *)lport->lport_port_private;
1922 
1923 	switch (cmd) {
1924 	case STMF_CMD_LPORT_ONLINE:
1925 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_REQ);
1926 		break;
1927 	case STMF_CMD_LPORT_OFFLINE:
1928 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_REQ);
1929 		break;
1930 	case STMF_ACK_LPORT_ONLINE_COMPLETE:
1931 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_COMPLETE_ACK);
1932 		break;
1933 	case STMF_ACK_LPORT_OFFLINE_COMPLETE:
1934 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_COMPLETE_ACK);
1935 		break;
1936 
1937 	default:
1938 		break;
1939 	}
1940 }
1941 
1942 static stmf_status_t
1943 iscsit_idm_to_stmf(idm_status_t idmrc)
1944 {
1945 	switch (idmrc) {
1946 	case IDM_STATUS_SUCCESS:
1947 		return (STMF_SUCCESS);
1948 	default:
1949 		return (STMF_FAILURE);
1950 	}
1951 	/*NOTREACHED*/
1952 }
1953 
1954 void
1955 iscsit_op_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1956 {
1957 	iscsit_conn_t		*ict = ic->ic_handle;
1958 
1959 	if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
1960 		iscsit_post_scsi_cmd(ic, rx_pdu);
1961 	}
1962 	iscsit_process_pdu_in_queue(ict->ict_sess);
1963 }
1964 
1965 /*
1966  * ISCSI protocol
1967  */
1968 
1969 void
1970 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1971 {
1972 	iscsit_conn_t		*ict;
1973 	iscsit_task_t		*itask;
1974 	scsi_task_t		*task;
1975 	iscsit_buf_t		*ibuf;
1976 	iscsi_scsi_cmd_hdr_t	*iscsi_scsi =
1977 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
1978 	iscsi_addl_hdr_t	*ahs_hdr;
1979 	uint16_t		addl_cdb_len = 0;
1980 
1981 	ict = ic->ic_handle;
1982 
1983 	itask = iscsit_task_alloc(ict);
1984 	if (itask == NULL) {
1985 		/* Finish processing request */
1986 		iscsit_set_cmdsn(ict, rx_pdu);
1987 
1988 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
1989 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
1990 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
1991 		return;
1992 	}
1993 
1994 	/*
1995 	 * Note CmdSN and ITT in task.  IDM will have already validated this
1996 	 * request against the connection state so we don't need to check
1997 	 * that (the connection may have changed state in the meantime but
1998 	 * we will catch that when we try to send a response)
1999 	 */
2000 	itask->it_cmdsn = ntohl(iscsi_scsi->cmdsn);
2001 	itask->it_itt = iscsi_scsi->itt;
2002 
2003 	/*
2004 	 * Check for extended CDB AHS
2005 	 */
2006 	if (iscsi_scsi->hlength > 0) {
2007 		ahs_hdr = (iscsi_addl_hdr_t *)iscsi_scsi;
2008 		addl_cdb_len = ((ahs_hdr->ahs_hlen_hi << 8) |
2009 		    ahs_hdr->ahs_hlen_lo) - 1; /* Adjust for reserved byte */
2010 		if (((addl_cdb_len + 4) / sizeof (uint32_t)) >
2011 		    iscsi_scsi->hlength) {
2012 			/* Mangled header info, drop it */
2013 			idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2014 			return;
2015 		}
2016 	}
2017 
2018 	ict = rx_pdu->isp_ic->ic_handle; /* IDM client private */
2019 
2020 	/*
2021 	 * Add task to session list.  This function will also check to
2022 	 * ensure that the task does not already exist.
2023 	 */
2024 	if (iscsit_task_start(itask) != IDM_STATUS_SUCCESS) {
2025 		/*
2026 		 * Task exists, free all resources and reject.  Don't
2027 		 * update expcmdsn in this case because RFC 3720 says
2028 		 * "The CmdSN of the rejected command PDU (if it is a
2029 		 * non-immediate command) MUST NOT be considered received
2030 		 * by the target (i.e., a command sequence gap must be
2031 		 * assumed for the CmdSN), even though the CmdSN of the
2032 		 * rejected command PDU may be reliably ascertained.  Upon
2033 		 * receiving the Reject, the initiator MUST plug the CmdSN
2034 		 * gap in order to continue to use the session.  The gap
2035 		 * may be plugged either by transmitting a command PDU
2036 		 * with the same CmdSN, or by aborting the task (see section
2037 		 * 6.9 on how an abort may plug a CmdSN gap)." (Section 6.3)
2038 		 */
2039 		iscsit_task_free(itask);
2040 		iscsit_send_reject(ict, rx_pdu, ISCSI_REJECT_TASK_IN_PROGRESS);
2041 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2042 		return;
2043 	}
2044 
2045 	/* Update sequence numbers */
2046 	iscsit_set_cmdsn(ict, rx_pdu);
2047 
2048 	/*
2049 	 * Allocate STMF task
2050 	 */
2051 	itask->it_stmf_task = stmf_task_alloc(
2052 	    itask->it_ict->ict_sess->ist_lport,
2053 	    itask->it_ict->ict_sess->ist_stmf_sess, iscsi_scsi->lun,
2054 	    16 + addl_cdb_len, 0);
2055 	if (itask->it_stmf_task == NULL) {
2056 		/*
2057 		 * Either stmf really couldn't get memory for a task or,
2058 		 * more likely, the LU is currently in reset.  Either way
2059 		 * we have no choice but to fail the request.
2060 		 */
2061 		iscsit_task_done(itask);
2062 		iscsit_task_free(itask);
2063 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
2064 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
2065 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2066 		return;
2067 	}
2068 
2069 	task = itask->it_stmf_task;
2070 	task->task_port_private = itask;
2071 
2072 	bcopy(iscsi_scsi->lun, task->task_lun_no, sizeof (task->task_lun_no));
2073 
2074 	/*
2075 	 * iSCSI and Comstar use the same values.  Should we rely on this
2076 	 * or translate them bit-wise?
2077 	 */
2078 
2079 	task->task_flags =
2080 	    (((iscsi_scsi->flags & ISCSI_FLAG_CMD_READ) ? TF_READ_DATA : 0) |
2081 	    ((iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ? TF_WRITE_DATA : 0) |
2082 	    ((rx_pdu->isp_datalen == 0) ? 0 : TF_INITIAL_BURST));
2083 
2084 	switch (iscsi_scsi->flags & ISCSI_FLAG_CMD_ATTR_MASK) {
2085 	case ISCSI_ATTR_UNTAGGED:
2086 		break;
2087 	case ISCSI_ATTR_SIMPLE:
2088 		task->task_additional_flags |= TF_ATTR_SIMPLE_QUEUE;
2089 		break;
2090 	case ISCSI_ATTR_ORDERED:
2091 		task->task_additional_flags |= TF_ATTR_ORDERED_QUEUE;
2092 		break;
2093 	case ISCSI_ATTR_HEAD_OF_QUEUE:
2094 		task->task_additional_flags |= TF_ATTR_HEAD_OF_QUEUE;
2095 		break;
2096 	case ISCSI_ATTR_ACA:
2097 		task->task_additional_flags |= TF_ATTR_ACA;
2098 		break;
2099 	default:
2100 		/* Protocol error but just take it, treat as untagged */
2101 		break;
2102 	}
2103 
2104 
2105 	task->task_additional_flags = 0;
2106 	task->task_priority = 0;
2107 	task->task_mgmt_function = TM_NONE;
2108 
2109 	/*
2110 	 * This "task_max_nbufs" doesn't map well to BIDI.  We probably need
2111 	 * parameter for each direction.  "MaxOutstandingR2T" may very well
2112 	 * be set to one which could prevent us from doing simultaneous
2113 	 * transfers in each direction.
2114 	 */
2115 	task->task_max_nbufs = (iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ?
2116 	    ict->ict_op.op_max_outstanding_r2t : STMF_BUFS_MAX;
2117 	task->task_cmd_seq_no = ntohl(iscsi_scsi->itt);
2118 	task->task_expected_xfer_length = ntohl(iscsi_scsi->data_length);
2119 
2120 	/* Copy CDB */
2121 	bcopy(iscsi_scsi->scb, task->task_cdb, 16);
2122 	if (addl_cdb_len > 0) {
2123 		bcopy(ahs_hdr->ahs_extscb, task->task_cdb + 16, addl_cdb_len);
2124 	}
2125 
2126 	DTRACE_ISCSI_3(scsi__command, idm_conn_t *, ic,
2127 	    iscsi_scsi_cmd_hdr_t *, (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr,
2128 	    scsi_task_t *, task);
2129 
2130 	/*
2131 	 * Copy the transport header into the task handle from the PDU
2132 	 * handle. The transport header describes this task's remote tagged
2133 	 * buffer.
2134 	 */
2135 	if (rx_pdu->isp_transport_hdrlen != 0) {
2136 		bcopy(rx_pdu->isp_transport_hdr,
2137 		    itask->it_idm_task->idt_transport_hdr,
2138 		    rx_pdu->isp_transport_hdrlen);
2139 	}
2140 
2141 	/*
2142 	 * Tell IDM about our new active task
2143 	 */
2144 	idm_task_start(itask->it_idm_task, (uintptr_t)itask->it_itt);
2145 
2146 	/*
2147 	 * If we have any immediate data then setup the immediate buffer
2148 	 * context that comes with the task
2149 	 */
2150 	if (rx_pdu->isp_datalen) {
2151 		ibuf = itask->it_immed_data;
2152 		ibuf->ibuf_immed_data_pdu = rx_pdu;
2153 		ibuf->ibuf_stmf_buf->db_data_size = rx_pdu->isp_datalen;
2154 		ibuf->ibuf_stmf_buf->db_buf_size = rx_pdu->isp_datalen;
2155 		ibuf->ibuf_stmf_buf->db_relative_offset = 0;
2156 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_length =
2157 		    rx_pdu->isp_datalen;
2158 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr = rx_pdu->isp_data;
2159 
2160 		DTRACE_ISCSI_8(xfer__start, idm_conn_t *, ic,
2161 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2162 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2163 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2164 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2165 
2166 		/*
2167 		 * For immediate data transfer, there is no callback from
2168 		 * stmf to indicate that the initial burst of data is
2169 		 * transferred successfully. In some cases, the task can
2170 		 * get freed before execution returns from stmf_post_task.
2171 		 * Although this xfer-start/done probe accurately tracks
2172 		 * the size of the transfer, it does only provide a best
2173 		 * effort on the timing of the transfer.
2174 		 */
2175 		DTRACE_ISCSI_8(xfer__done, idm_conn_t *, ic,
2176 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2177 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2178 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2179 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2180 		stmf_post_task(task, ibuf->ibuf_stmf_buf);
2181 	} else {
2182 
2183 		stmf_post_task(task, NULL);
2184 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2185 	}
2186 }
2187 
2188 void
2189 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu)
2190 {
2191 	iscsit_conn_t *ict = rx_pdu->isp_ic->ic_handle;
2192 
2193 	/*
2194 	 * If this isn't a login packet, we need a session.  Otherwise
2195 	 * this is a protocol error (perhaps one IDM should've caught?).
2196 	 */
2197 	if (IDM_PDU_OPCODE(rx_pdu) != ISCSI_OP_LOGIN_CMD &&
2198 	    ict->ict_sess == NULL) {
2199 		DTRACE_PROBE2(iscsi__idm__deferred__no__session,
2200 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
2201 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2202 		return;
2203 	}
2204 
2205 	/*
2206 	 * If the connection has been lost then ignore new PDU's
2207 	 */
2208 	mutex_enter(&ict->ict_mutex);
2209 	if (ict->ict_lost) {
2210 		mutex_exit(&ict->ict_mutex);
2211 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2212 		return;
2213 	}
2214 
2215 	/*
2216 	 * Grab a hold on the connection to prevent it from going away
2217 	 * between now and when the taskq function is called.
2218 	 */
2219 	iscsit_conn_dispatch_hold(ict);
2220 	mutex_exit(&ict->ict_mutex);
2221 
2222 	taskq_dispatch_ent(iscsit_global.global_dispatch_taskq,
2223 	    iscsit_deferred, rx_pdu, 0, &rx_pdu->isp_tqent);
2224 }
2225 
2226 static void
2227 iscsit_deferred(void *rx_pdu_void)
2228 {
2229 	idm_pdu_t		*rx_pdu = rx_pdu_void;
2230 	idm_conn_t		*ic = rx_pdu->isp_ic;
2231 	iscsit_conn_t		*ict = ic->ic_handle;
2232 
2233 	/*
2234 	 * NOP and Task Management Commands can be marked for immediate
2235 	 * delivery. Commands marked as 'Immediate' are to be considered
2236 	 * for execution as soon as they arrive on the target. So these
2237 	 * should not be checked for sequence order and put in a queue.
2238 	 * The CmdSN is not advanced for Immediate Commands.
2239 	 */
2240 	switch (IDM_PDU_OPCODE(rx_pdu)) {
2241 	case ISCSI_OP_NOOP_OUT:
2242 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2243 			iscsit_set_cmdsn(ict, rx_pdu);
2244 			iscsit_pdu_op_noop(ict, rx_pdu);
2245 		}
2246 		break;
2247 	case ISCSI_OP_LOGIN_CMD:
2248 		iscsit_pdu_op_login_cmd(ict, rx_pdu);
2249 		iscsit_conn_dispatch_rele(ict);
2250 		return;
2251 	case ISCSI_OP_TEXT_CMD:
2252 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2253 			iscsit_set_cmdsn(ict, rx_pdu);
2254 			iscsit_pdu_op_text_cmd(ict, rx_pdu);
2255 		}
2256 		break;
2257 	case ISCSI_OP_LOGOUT_CMD:
2258 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2259 			iscsit_set_cmdsn(ict, rx_pdu);
2260 			iscsit_pdu_op_logout_cmd(ict, rx_pdu);
2261 		}
2262 		break;
2263 	default:
2264 		/* Protocol error.  IDM should have caught this */
2265 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2266 		ASSERT(0);
2267 		break;
2268 	}
2269 	/*
2270 	 * Check if there are other PDUs in the session staging queue
2271 	 * waiting to be posted to SCSI layer.
2272 	 */
2273 	iscsit_process_pdu_in_queue(ict->ict_sess);
2274 
2275 	iscsit_conn_dispatch_rele(ict);
2276 }
2277 
2278 static void
2279 iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
2280     uint8_t response, uint8_t cmd_status)
2281 {
2282 	idm_pdu_t			*rsp_pdu;
2283 	idm_conn_t			*ic;
2284 	iscsi_scsi_rsp_hdr_t		*resp;
2285 	iscsi_scsi_cmd_hdr_t		*req =
2286 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2287 
2288 	ic = ict->ict_ic;
2289 
2290 	rsp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_rsp_hdr_t), 0);
2291 	idm_pdu_init(rsp_pdu, ic, NULL, NULL);
2292 	/*
2293 	 * StatSN is incremented by 1 for every response sent on
2294 	 * a connection except for responses sent as a result of
2295 	 * a retry or SNACK
2296 	 */
2297 	rsp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2298 
2299 	resp = (iscsi_scsi_rsp_hdr_t *)rsp_pdu->isp_hdr;
2300 
2301 	resp->opcode = ISCSI_OP_SCSI_RSP;
2302 	resp->flags = ISCSI_FLAG_FINAL;
2303 	resp->response = response;
2304 	resp->cmd_status = cmd_status;
2305 	resp->itt = req->itt;
2306 	if ((response == ISCSI_STATUS_CMD_COMPLETED) &&
2307 	    (req->data_length != 0) &&
2308 	    ((req->flags & ISCSI_FLAG_CMD_READ) ||
2309 	    (req->flags & ISCSI_FLAG_CMD_WRITE))) {
2310 		resp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
2311 		resp->residual_count = req->data_length;
2312 	}
2313 
2314 	DTRACE_PROBE4(iscsi__scsi__direct__response,
2315 	    iscsit_conn_t *, ict,
2316 	    uint8_t, resp->response,
2317 	    uint8_t, resp->cmd_status,
2318 	    idm_pdu_t *, rsp_pdu);
2319 
2320 	iscsit_pdu_tx(rsp_pdu);
2321 }
2322 
2323 void
2324 iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu, uint8_t tm_status)
2325 {
2326 	iscsi_scsi_task_mgt_rsp_hdr_t	*tm_resp;
2327 
2328 	/*
2329 	 * The target must take note of the last-sent StatSN.
2330 	 * The StatSN is to be incremented after sending a
2331 	 * task management response. Digest recovery can only
2332 	 * work if StatSN is incremented.
2333 	 */
2334 	tm_resp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2335 	tm_resp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2336 	tm_resp->response = tm_status;
2337 
2338 	DTRACE_PROBE3(iscsi__scsi__tm__response,
2339 	    iscsit_conn_t *, tm_resp_pdu->isp_ic->ic_handle,
2340 	    uint8_t, tm_resp->response,
2341 	    idm_pdu_t *, tm_resp_pdu);
2342 	iscsit_pdu_tx(tm_resp_pdu);
2343 }
2344 
2345 void
2346 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2347 {
2348 	idm_pdu_t			*tm_resp_pdu;
2349 	iscsit_task_t			*itask;
2350 	iscsit_task_t			*tm_itask;
2351 	scsi_task_t			*task;
2352 	iscsi_scsi_task_mgt_hdr_t 	*iscsi_tm =
2353 	    (iscsi_scsi_task_mgt_hdr_t *)rx_pdu->isp_hdr;
2354 	iscsi_scsi_task_mgt_rsp_hdr_t 	*iscsi_tm_rsp =
2355 	    (iscsi_scsi_task_mgt_rsp_hdr_t *)rx_pdu->isp_hdr;
2356 	uint32_t			rtt, cmdsn, refcmdsn;
2357 	uint8_t				tm_func;
2358 
2359 	/*
2360 	 * Setup response PDU (response field will get filled in later)
2361 	 */
2362 	tm_resp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_task_mgt_rsp_hdr_t), 0);
2363 	if (tm_resp_pdu == NULL) {
2364 		/* Can't respond, just drop it */
2365 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2366 		return;
2367 	}
2368 	idm_pdu_init(tm_resp_pdu, ict->ict_ic, NULL, NULL);
2369 	iscsi_tm_rsp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2370 	bzero(iscsi_tm_rsp, sizeof (iscsi_scsi_task_mgt_rsp_hdr_t));
2371 	iscsi_tm_rsp->opcode = ISCSI_OP_SCSI_TASK_MGT_RSP;
2372 	iscsi_tm_rsp->flags = ISCSI_FLAG_FINAL;
2373 	iscsi_tm_rsp->itt = rx_pdu->isp_hdr->itt;
2374 
2375 	/*
2376 	 * Figure out what we're being asked to do.
2377 	 */
2378 	DTRACE_PROBE4(iscsi__scsi__tm__request,
2379 	    iscsit_conn_t *, ict,
2380 	    uint8_t, (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK),
2381 	    uint32_t, iscsi_tm->rtt,
2382 	    idm_pdu_t *, rx_pdu);
2383 	switch (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK) {
2384 	case ISCSI_TM_FUNC_ABORT_TASK:
2385 		/*
2386 		 * STMF doesn't currently support the "abort task" task
2387 		 * management command although it does support aborting
2388 		 * an individual task.  We'll get STMF to abort the task
2389 		 * for us but handle the details of the task management
2390 		 * command ourselves.
2391 		 *
2392 		 * Find the task associated with the referenced task tag.
2393 		 */
2394 		rtt = iscsi_tm->rtt;
2395 		itask = (iscsit_task_t *)idm_task_find_by_handle(ict->ict_ic,
2396 		    (uintptr_t)rtt);
2397 
2398 		if (itask == NULL) {
2399 			cmdsn = ntohl(iscsi_tm->cmdsn);
2400 			refcmdsn = ntohl(iscsi_tm->refcmdsn);
2401 
2402 			/*
2403 			 * Task was not found. But the SCSI command could be
2404 			 * on the rxpdu wait queue. If RefCmdSN is within
2405 			 * the CmdSN window and less than CmdSN of the TM
2406 			 * function, return "Function Complete". Otherwise,
2407 			 * return "Task Does Not Exist".
2408 			 */
2409 
2410 			if (iscsit_cmdsn_in_window(ict, refcmdsn) &&
2411 			    iscsit_sna_lt(refcmdsn, cmdsn)) {
2412 				mutex_enter(&ict->ict_sess->ist_sn_mutex);
2413 				if (iscsit_remove_pdu_from_queue(
2414 				    ict->ict_sess, refcmdsn)) {
2415 					iscsit_conn_dispatch_rele(ict);
2416 				}
2417 				mutex_exit(&ict->ict_sess->ist_sn_mutex);
2418 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2419 				    SCSI_TCP_TM_RESP_COMPLETE);
2420 			} else {
2421 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2422 				    SCSI_TCP_TM_RESP_NO_TASK);
2423 			}
2424 		} else {
2425 
2426 			/*
2427 			 * Tell STMF to abort the task.  This will do no harm
2428 			 * if the task is already complete.
2429 			 */
2430 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
2431 			    STMF_ABORTED, NULL);
2432 
2433 			/*
2434 			 * Make sure the task hasn't already completed
2435 			 */
2436 			mutex_enter(&itask->it_idm_task->idt_mutex);
2437 			if ((itask->it_idm_task->idt_state == TASK_COMPLETE) ||
2438 			    (itask->it_idm_task->idt_state == TASK_IDLE)) {
2439 				/*
2440 				 * Task is complete, return "Task Does Not
2441 				 * Exist"
2442 				 */
2443 				mutex_exit(&itask->it_idm_task->idt_mutex);
2444 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2445 				    SCSI_TCP_TM_RESP_NO_TASK);
2446 			} else {
2447 				/*
2448 				 * STMF is now aborting the task, return
2449 				 * "Function Complete"
2450 				 */
2451 				mutex_exit(&itask->it_idm_task->idt_mutex);
2452 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2453 				    SCSI_TCP_TM_RESP_COMPLETE);
2454 			}
2455 			idm_task_rele(itask->it_idm_task);
2456 		}
2457 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2458 		return;
2459 
2460 	case ISCSI_TM_FUNC_ABORT_TASK_SET:
2461 		tm_func = TM_ABORT_TASK_SET;
2462 		break;
2463 
2464 	case ISCSI_TM_FUNC_CLEAR_ACA:
2465 		tm_func = TM_CLEAR_ACA;
2466 		break;
2467 
2468 	case ISCSI_TM_FUNC_CLEAR_TASK_SET:
2469 		tm_func = TM_CLEAR_TASK_SET;
2470 		break;
2471 
2472 	case ISCSI_TM_FUNC_LOGICAL_UNIT_RESET:
2473 		tm_func = TM_LUN_RESET;
2474 		break;
2475 
2476 	case ISCSI_TM_FUNC_TARGET_WARM_RESET:
2477 		tm_func = TM_TARGET_WARM_RESET;
2478 		break;
2479 
2480 	case ISCSI_TM_FUNC_TARGET_COLD_RESET:
2481 		tm_func = TM_TARGET_COLD_RESET;
2482 		break;
2483 
2484 	case ISCSI_TM_FUNC_TASK_REASSIGN:
2485 		/*
2486 		 * We do not currently support allegiance reassignment.  When
2487 		 * we start supporting ERL1+, we will need to.
2488 		 */
2489 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2490 		    SCSI_TCP_TM_RESP_NO_ALLG_REASSN);
2491 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2492 		return;
2493 
2494 	default:
2495 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2496 		    SCSI_TCP_TM_RESP_REJECTED);
2497 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2498 		return;
2499 	}
2500 
2501 	tm_itask = iscsit_tm_task_alloc(ict);
2502 	if (tm_itask == NULL) {
2503 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2504 		    SCSI_TCP_TM_RESP_REJECTED);
2505 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2506 		return;
2507 	}
2508 
2509 
2510 	task = stmf_task_alloc(ict->ict_sess->ist_lport,
2511 	    ict->ict_sess->ist_stmf_sess, iscsi_tm->lun,
2512 	    0, STMF_TASK_EXT_NONE);
2513 	if (task == NULL) {
2514 		/*
2515 		 * If this happens, either the LU is in reset, couldn't
2516 		 * get memory, or some other condition in which we simply
2517 		 * can't complete this request.  It would be nice to return
2518 		 * an error code like "busy" but the closest we have is
2519 		 * "rejected".
2520 		 */
2521 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2522 		    SCSI_TCP_TM_RESP_REJECTED);
2523 		iscsit_tm_task_free(tm_itask);
2524 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2525 		return;
2526 	}
2527 
2528 	tm_itask->it_tm_pdu = tm_resp_pdu;
2529 	tm_itask->it_stmf_task = task;
2530 	task->task_port_private = tm_itask;
2531 	task->task_mgmt_function = tm_func;
2532 	task->task_additional_flags = TASK_AF_NO_EXPECTED_XFER_LENGTH;
2533 	task->task_priority = 0;
2534 	task->task_max_nbufs = STMF_BUFS_MAX;
2535 	task->task_cmd_seq_no = iscsi_tm->itt;
2536 	task->task_expected_xfer_length = 0;
2537 
2538 	stmf_post_task(task, NULL);
2539 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2540 }
2541 
2542 static void
2543 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2544 {
2545 	iscsi_nop_out_hdr_t *out = (iscsi_nop_out_hdr_t *)rx_pdu->isp_hdr;
2546 	iscsi_nop_in_hdr_t *in;
2547 	int resp_datalen;
2548 	idm_pdu_t *resp;
2549 
2550 	/* Ignore the response from initiator */
2551 	if ((out->itt == ISCSI_RSVD_TASK_TAG) ||
2552 	    (out->ttt != ISCSI_RSVD_TASK_TAG)) {
2553 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2554 		return;
2555 	}
2556 
2557 	/* Allocate a PDU to respond */
2558 	resp_datalen = ntoh24(out->dlength);
2559 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
2560 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2561 	if (resp_datalen > 0) {
2562 		bcopy(rx_pdu->isp_data, resp->isp_data, resp_datalen);
2563 	}
2564 
2565 	/*
2566 	 * When sending a NOP-In as a response to a NOP-Out from the initiator,
2567 	 * the target must respond with the same initiator task tag that was
2568 	 * provided in the NOP-Out request, the target transfer tag must be
2569 	 * ISCSI_RSVD_TASK_TAG (0xffffffff) and StatSN will contain the next
2570 	 * status sequence number. The StatSN for the connection is advanced
2571 	 * after this PDU is sent.
2572 	 */
2573 	in = (iscsi_nop_in_hdr_t *)resp->isp_hdr;
2574 	bzero(in, sizeof (*in));
2575 	in->opcode = ISCSI_OP_NOOP_IN;
2576 	in->flags = ISCSI_FLAG_FINAL;
2577 	bcopy(out->lun, in->lun, 8);
2578 	in->itt		= out->itt;
2579 	in->ttt		= ISCSI_RSVD_TASK_TAG;
2580 	hton24(in->dlength, resp_datalen);
2581 	resp->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2582 	/* Any other field in resp to be set? */
2583 	iscsit_pdu_tx(resp);
2584 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2585 }
2586 
2587 static void
2588 iscsit_pdu_op_login_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2589 {
2590 
2591 	/*
2592 	 * Submit PDU to login state machine.  State machine will free the
2593 	 * PDU.
2594 	 */
2595 	iscsit_login_sm_event(ict, ILE_LOGIN_RCV, rx_pdu);
2596 }
2597 
2598 void
2599 iscsit_pdu_op_logout_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2600 {
2601 	iscsi_logout_hdr_t 	*logout_req =
2602 	    (iscsi_logout_hdr_t *)rx_pdu->isp_hdr;
2603 	iscsi_logout_rsp_hdr_t	*logout_rsp;
2604 	idm_pdu_t *resp;
2605 
2606 	/* Allocate a PDU to respond */
2607 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2608 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2609 	/*
2610 	 * The StatSN is to be sent to the initiator,
2611 	 * it is not required to increment the number
2612 	 * as the connection is terminating.
2613 	 */
2614 	resp->isp_flags |= IDM_PDU_SET_STATSN;
2615 	/*
2616 	 * Logout results in the immediate termination of all tasks except
2617 	 * if the logout reason is ISCSI_LOGOUT_REASON_RECOVERY.  The
2618 	 * connection state machine will drive this task cleanup automatically
2619 	 * so we don't need to handle that here.
2620 	 */
2621 	logout_rsp = (iscsi_logout_rsp_hdr_t *)resp->isp_hdr;
2622 	bzero(logout_rsp, sizeof (*logout_rsp));
2623 	logout_rsp->opcode = ISCSI_OP_LOGOUT_RSP;
2624 	logout_rsp->flags = ISCSI_FLAG_FINAL;
2625 	logout_rsp->itt = logout_req->itt;
2626 	if ((logout_req->flags & ISCSI_FLAG_LOGOUT_REASON_MASK) >
2627 	    ISCSI_LOGOUT_REASON_RECOVERY) {
2628 		logout_rsp->response = ISCSI_LOGOUT_RECOVERY_UNSUPPORTED;
2629 	} else {
2630 		logout_rsp->response = ISCSI_LOGOUT_SUCCESS;
2631 	}
2632 
2633 	iscsit_pdu_tx(resp);
2634 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2635 }
2636 
2637 /*
2638  * Calculate the number of outstanding commands we can process
2639  */
2640 int
2641 iscsit_cmd_window()
2642 {
2643 	/*
2644 	 * Instead of using a pre-defined constant for the command window,
2645 	 * it should be made confiurable and dynamic. With MC/S, sequence
2646 	 * numbers will be used up at a much faster rate than with SC/S.
2647 	 */
2648 	return	(ISCSIT_MAX_WINDOW);
2649 }
2650 
2651 /*
2652  * Set local registers based on incoming PDU
2653  */
2654 void
2655 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2656 {
2657 	iscsit_sess_t *ist;
2658 	iscsi_scsi_cmd_hdr_t *req;
2659 
2660 	ist = ict->ict_sess;
2661 
2662 	req = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2663 	if (req->opcode & ISCSI_OP_IMMEDIATE) {
2664 		/* no cmdsn increment for immediate PDUs */
2665 		return;
2666 	}
2667 
2668 	/* Ensure that the ExpCmdSN advances in an orderly manner */
2669 	mutex_enter(&ist->ist_sn_mutex);
2670 	ist->ist_expcmdsn = ntohl(req->cmdsn) + 1;
2671 	ist->ist_maxcmdsn = ntohl(req->cmdsn) + iscsit_cmd_window();
2672 	mutex_exit(&ist->ist_sn_mutex);
2673 }
2674 
2675 /*
2676  * Wrapper funtion, calls iscsi_calc_rspsn and idm_pdu_tx
2677  */
2678 void
2679 iscsit_pdu_tx(idm_pdu_t *pdu)
2680 {
2681 	iscsit_conn_t *ict = pdu->isp_ic->ic_handle;
2682 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2683 	iscsit_sess_t *ist = ict->ict_sess;
2684 
2685 	/*
2686 	 * The command sequence numbers are session-wide and must stay
2687 	 * consistent across the transfer, so protect the cmdsn with a
2688 	 * mutex lock on the session. The status sequence number will
2689 	 * be updated just before the transport layer transmits the PDU.
2690 	 */
2691 
2692 	mutex_enter(&ict->ict_sess->ist_sn_mutex);
2693 	/* Set ExpCmdSN and MaxCmdSN */
2694 	rsp->maxcmdsn = htonl(ist->ist_maxcmdsn);
2695 	rsp->expcmdsn = htonl(ist->ist_expcmdsn);
2696 	idm_pdu_tx(pdu);
2697 	mutex_exit(&ict->ict_sess->ist_sn_mutex);
2698 }
2699 
2700 /*
2701  * Internal functions
2702  */
2703 
2704 void
2705 iscsit_send_async_event(iscsit_conn_t *ict, uint8_t event)
2706 {
2707 	idm_pdu_t		*abt;
2708 	iscsi_async_evt_hdr_t	*async_abt;
2709 
2710 	/*
2711 	 * Get a PDU to build the abort request.
2712 	 */
2713 	abt = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2714 	if (abt == NULL) {
2715 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2716 		return;
2717 	}
2718 
2719 	/*
2720 	 * A asynchronous message is sent by the target to request a logout.
2721 	 * The StatSN for the connection is advanced after the PDU is sent
2722 	 * to allow for initiator and target state synchronization.
2723 	 */
2724 	idm_pdu_init(abt, ict->ict_ic, NULL, NULL);
2725 	abt->isp_datalen = 0;
2726 	abt->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2727 
2728 	async_abt = (iscsi_async_evt_hdr_t *)abt->isp_hdr;
2729 	bzero(async_abt, sizeof (*async_abt));
2730 	async_abt->opcode = ISCSI_OP_ASYNC_EVENT;
2731 	async_abt->async_event = event;
2732 	async_abt->flags = ISCSI_FLAG_FINAL;
2733 	async_abt->rsvd4[0] = 0xff;
2734 	async_abt->rsvd4[1] = 0xff;
2735 	async_abt->rsvd4[2] = 0xff;
2736 	async_abt->rsvd4[3] = 0xff;
2737 
2738 	switch (event) {
2739 	case ISCSI_ASYNC_EVENT_REQUEST_LOGOUT:
2740 		async_abt->param3 = htons(IDM_LOGOUT_SECONDS);
2741 		break;
2742 	case ISCSI_ASYNC_EVENT_SCSI_EVENT:
2743 	case ISCSI_ASYNC_EVENT_DROPPING_CONNECTION:
2744 	case ISCSI_ASYNC_EVENT_DROPPING_ALL_CONNECTIONS:
2745 	case ISCSI_ASYNC_EVENT_PARAM_NEGOTIATION:
2746 	default:
2747 		ASSERT(0);
2748 	}
2749 
2750 	iscsit_pdu_tx(abt);
2751 }
2752 
2753 void
2754 iscsit_send_reject(iscsit_conn_t *ict, idm_pdu_t *rejected_pdu, uint8_t reason)
2755 {
2756 	idm_pdu_t		*reject_pdu;
2757 	iscsi_reject_rsp_hdr_t	*reject;
2758 
2759 	/*
2760 	 * Get a PDU to build the abort request.
2761 	 */
2762 	reject_pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t),
2763 	    rejected_pdu->isp_hdrlen);
2764 	if (reject_pdu == NULL) {
2765 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2766 		return;
2767 	}
2768 	idm_pdu_init(reject_pdu, ict->ict_ic, NULL, NULL);
2769 	/* StatSN is advanced after a Reject PDU */
2770 	reject_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2771 	reject_pdu->isp_datalen = rejected_pdu->isp_hdrlen;
2772 	bcopy(rejected_pdu->isp_hdr, reject_pdu->isp_data,
2773 	    rejected_pdu->isp_hdrlen);
2774 
2775 	reject = (iscsi_reject_rsp_hdr_t *)reject_pdu->isp_hdr;
2776 	bzero(reject, sizeof (*reject));
2777 	reject->opcode = ISCSI_OP_REJECT_MSG;
2778 	reject->reason = reason;
2779 	reject->flags = ISCSI_FLAG_FINAL;
2780 	hton24(reject->dlength, rejected_pdu->isp_hdrlen);
2781 	reject->must_be_ff[0] = 0xff;
2782 	reject->must_be_ff[1] = 0xff;
2783 	reject->must_be_ff[2] = 0xff;
2784 	reject->must_be_ff[3] = 0xff;
2785 
2786 	iscsit_pdu_tx(reject_pdu);
2787 }
2788 
2789 
2790 static iscsit_task_t *
2791 iscsit_task_alloc(iscsit_conn_t *ict)
2792 {
2793 	iscsit_task_t *itask;
2794 	iscsit_buf_t *immed_ibuf;
2795 
2796 	/*
2797 	 * Possible items to pre-alloc if we cache iscsit_task_t's:
2798 	 *
2799 	 * Status PDU w/ sense buffer
2800 	 * stmf_data_buf_t for immediate data
2801 	 */
2802 	itask = kmem_alloc(sizeof (iscsit_task_t) + sizeof (iscsit_buf_t) +
2803 	    sizeof (stmf_data_buf_t), KM_NOSLEEP);
2804 	if (itask != NULL) {
2805 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2806 		itask->it_aborted = itask->it_stmf_abort =
2807 		    itask->it_tm_task = 0;
2808 
2809 		immed_ibuf = (iscsit_buf_t *)(itask + 1);
2810 		bzero(immed_ibuf, sizeof (*immed_ibuf));
2811 		immed_ibuf->ibuf_is_immed = B_TRUE;
2812 		immed_ibuf->ibuf_stmf_buf = (stmf_data_buf_t *)(immed_ibuf + 1);
2813 
2814 		bzero(immed_ibuf->ibuf_stmf_buf, sizeof (stmf_data_buf_t));
2815 		immed_ibuf->ibuf_stmf_buf->db_port_private = immed_ibuf;
2816 		immed_ibuf->ibuf_stmf_buf->db_sglist_length = 1;
2817 		immed_ibuf->ibuf_stmf_buf->db_flags = DB_DIRECTION_FROM_RPORT |
2818 		    DB_DONT_CACHE;
2819 		itask->it_immed_data = immed_ibuf;
2820 		itask->it_idm_task = idm_task_alloc(ict->ict_ic);
2821 		if (itask->it_idm_task != NULL) {
2822 			itask->it_idm_task->idt_private = itask;
2823 			itask->it_ict = ict;
2824 			itask->it_ttt = itask->it_idm_task->idt_tt;
2825 			return (itask);
2826 		} else {
2827 			kmem_free(itask, sizeof (iscsit_task_t) +
2828 			    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2829 		}
2830 	}
2831 
2832 	return (NULL);
2833 }
2834 
2835 static void
2836 iscsit_task_free(iscsit_task_t *itask)
2837 {
2838 	idm_task_free(itask->it_idm_task);
2839 	mutex_destroy(&itask->it_mutex);
2840 	kmem_free(itask, sizeof (iscsit_task_t) +
2841 	    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2842 }
2843 
2844 static iscsit_task_t *
2845 iscsit_tm_task_alloc(iscsit_conn_t *ict)
2846 {
2847 	iscsit_task_t *itask;
2848 
2849 	itask = kmem_zalloc(sizeof (iscsit_task_t), KM_NOSLEEP);
2850 	if (itask != NULL) {
2851 		idm_conn_hold(ict->ict_ic);
2852 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2853 		itask->it_aborted = itask->it_stmf_abort =
2854 		    itask->it_tm_responded = 0;
2855 		itask->it_tm_pdu = NULL;
2856 		itask->it_tm_task = 1;
2857 		itask->it_ict = ict;
2858 	}
2859 
2860 	return (itask);
2861 }
2862 
2863 static void
2864 iscsit_tm_task_free(iscsit_task_t *itask)
2865 {
2866 	/*
2867 	 * If we responded then the call to idm_pdu_complete will free the
2868 	 * PDU.  Otherwise we got aborted before the TM function could
2869 	 * complete and we need to free the PDU explicitly.
2870 	 */
2871 	if (itask->it_tm_pdu != NULL && !itask->it_tm_responded)
2872 		idm_pdu_free(itask->it_tm_pdu);
2873 	idm_conn_rele(itask->it_ict->ict_ic);
2874 	mutex_destroy(&itask->it_mutex);
2875 	kmem_free(itask, sizeof (iscsit_task_t));
2876 }
2877 
2878 static idm_status_t
2879 iscsit_task_start(iscsit_task_t *itask)
2880 {
2881 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2882 	avl_index_t		where;
2883 
2884 	/*
2885 	 * Sanity check the ITT and ensure that this task does not already
2886 	 * exist.  If not then add the task to the session task list.
2887 	 */
2888 	mutex_enter(&ist->ist_mutex);
2889 	mutex_enter(&itask->it_mutex);
2890 	itask->it_active = 1;
2891 	if (avl_find(&ist->ist_task_list, itask, &where) == NULL) {
2892 		/* New task, add to AVL */
2893 		avl_insert(&ist->ist_task_list, itask, where);
2894 		mutex_exit(&itask->it_mutex);
2895 		mutex_exit(&ist->ist_mutex);
2896 		return (IDM_STATUS_SUCCESS);
2897 	}
2898 	mutex_exit(&itask->it_mutex);
2899 	mutex_exit(&ist->ist_mutex);
2900 
2901 	return (IDM_STATUS_REJECT);
2902 }
2903 
2904 static void
2905 iscsit_task_done(iscsit_task_t *itask)
2906 {
2907 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2908 
2909 	mutex_enter(&ist->ist_mutex);
2910 	mutex_enter(&itask->it_mutex);
2911 	if (itask->it_active) {
2912 		avl_remove(&ist->ist_task_list, itask);
2913 		itask->it_active = 0;
2914 	}
2915 	mutex_exit(&itask->it_mutex);
2916 	mutex_exit(&ist->ist_mutex);
2917 }
2918 
2919 /*
2920  * iscsit status PDU cache
2921  */
2922 
2923 /*ARGSUSED*/
2924 static int
2925 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags)
2926 {
2927 	idm_pdu_t *pdu = pdu_void;
2928 	iscsi_scsi_rsp_hdr_t *rsp;
2929 
2930 	bzero(pdu, sizeof (idm_pdu_t));
2931 	pdu->isp_callback = iscsit_send_good_status_done;
2932 	pdu->isp_magic = IDM_PDU_MAGIC;
2933 	pdu->isp_hdr = (iscsi_hdr_t *)(pdu + 1); /* Ptr arithmetic */
2934 	pdu->isp_hdrlen = sizeof (iscsi_hdr_t);
2935 
2936 	/* Setup status response */
2937 	rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2938 	bzero(rsp, sizeof (*rsp));
2939 	rsp->opcode = ISCSI_OP_SCSI_RSP;
2940 	rsp->flags = ISCSI_FLAG_FINAL;
2941 	rsp->response = ISCSI_STATUS_CMD_COMPLETED;
2942 
2943 	return (0);
2944 }
2945 
2946 /*
2947  * iscsit private data handler
2948  */
2949 
2950 /*ARGSUSED*/
2951 static void
2952 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags)
2953 {
2954 	it_config_t		*cfg;
2955 	nvlist_t		*nvl;
2956 	iscsit_service_enabled_t	old_state;
2957 
2958 	if ((cmd != STMF_PROVIDER_DATA_UPDATED) || (arg == NULL)) {
2959 		return;
2960 	}
2961 
2962 	nvl = (nvlist_t *)arg;
2963 
2964 	/* Translate nvlist */
2965 	if (it_nv_to_config(nvl, &cfg) != 0) {
2966 		cmn_err(CE_WARN, "Configuration is invalid");
2967 		return;
2968 	}
2969 
2970 	/* Check that no iSCSI ioctl is currently running */
2971 	mutex_enter(&iscsit_global.global_state_mutex);
2972 	old_state = iscsit_global.global_svc_state;
2973 	switch (iscsit_global.global_svc_state) {
2974 	case ISE_ENABLED:
2975 	case ISE_DISABLED:
2976 		iscsit_global.global_svc_state = ISE_BUSY;
2977 		break;
2978 	case ISE_ENABLING:
2979 		/*
2980 		 * It is OK for the iscsit_pp_cb to be called from inside of
2981 		 * an iSCSI ioctl only if we are currently executing inside
2982 		 * of stmf_register_port_provider.
2983 		 */
2984 		ASSERT((flags & STMF_PCB_PREG_COMPLETE) != 0);
2985 		break;
2986 	default:
2987 		cmn_err(CE_WARN, "iscsit_pp_cb called when global_svc_state"
2988 		    " is not ENABLED(0x%x) -- ignoring",
2989 		    iscsit_global.global_svc_state);
2990 		mutex_exit(&iscsit_global.global_state_mutex);
2991 		it_config_free_cmn(cfg);
2992 		return;
2993 	}
2994 	mutex_exit(&iscsit_global.global_state_mutex);
2995 
2996 	/* Update config */
2997 	(void) iscsit_config_merge(cfg);
2998 
2999 	it_config_free_cmn(cfg);
3000 
3001 	/* Restore old iSCSI driver global state */
3002 	mutex_enter(&iscsit_global.global_state_mutex);
3003 	ASSERT(iscsit_global.global_svc_state == ISE_BUSY ||
3004 	    iscsit_global.global_svc_state == ISE_ENABLING);
3005 	iscsit_global.global_svc_state = old_state;
3006 	mutex_exit(&iscsit_global.global_state_mutex);
3007 }
3008 
3009 
3010 static it_cfg_status_t
3011 iscsit_config_merge(it_config_t *in_cfg)
3012 {
3013 	it_cfg_status_t	status;
3014 	it_config_t	*cfg;
3015 	it_config_t	tmp_cfg;
3016 	list_t		tpg_del_list;
3017 
3018 	if (in_cfg) {
3019 		cfg = in_cfg;
3020 	} else {
3021 		/* Make empty config */
3022 		bzero(&tmp_cfg, sizeof (tmp_cfg));
3023 		cfg = &tmp_cfg;
3024 	}
3025 
3026 	list_create(&tpg_del_list,  sizeof (iscsit_tpg_t),
3027 	    offsetof(iscsit_tpg_t, tpg_delete_ln));
3028 
3029 	/*
3030 	 * Update targets, initiator contexts, target portal groups,
3031 	 * and iSNS client
3032 	 */
3033 	ISCSIT_GLOBAL_LOCK(RW_WRITER);
3034 	if (((status = iscsit_config_merge_tpg(cfg, &tpg_del_list))
3035 	    != 0) ||
3036 	    ((status = iscsit_config_merge_tgt(cfg)) != 0) ||
3037 	    ((status = iscsit_config_merge_ini(cfg)) != 0) ||
3038 	    ((status = isnst_config_merge(cfg)) != 0)) {
3039 		ISCSIT_GLOBAL_UNLOCK();
3040 		return (status);
3041 	}
3042 
3043 	/* Update other global config parameters */
3044 	if (iscsit_global.global_props) {
3045 		nvlist_free(iscsit_global.global_props);
3046 		iscsit_global.global_props = NULL;
3047 	}
3048 	if (in_cfg) {
3049 		(void) nvlist_dup(cfg->config_global_properties,
3050 		    &iscsit_global.global_props, KM_SLEEP);
3051 	}
3052 	ISCSIT_GLOBAL_UNLOCK();
3053 
3054 	iscsit_config_destroy_tpgs(&tpg_del_list);
3055 
3056 	list_destroy(&tpg_del_list);
3057 
3058 	return (ITCFG_SUCCESS);
3059 }
3060 
3061 /*
3062  * iscsit_sna_lt[e]
3063  *
3064  * Compare serial numbers using serial number arithmetic as defined in
3065  * RFC 1982.
3066  *
3067  * NOTE: This code is duplicated in the isns server. It ought to be common.
3068  */
3069 
3070 static int
3071 iscsit_sna_lt(uint32_t sn1, uint32_t sn2)
3072 {
3073 	return ((sn1 != sn2) &&
3074 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3075 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3076 }
3077 
3078 static int
3079 iscsit_sna_lte(uint32_t sn1, uint32_t sn2)
3080 {
3081 	return ((sn1 == sn2) ||
3082 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3083 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3084 }
3085 
3086 
3087 static boolean_t
3088 iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn)
3089 {
3090 	iscsit_sess_t	*ist = ict->ict_sess;
3091 	int		rval = B_TRUE;
3092 
3093 	ist = ict->ict_sess;
3094 
3095 	mutex_enter(&ist->ist_sn_mutex);
3096 
3097 	/*
3098 	 * If cmdsn is less than ist_expcmdsn - iscsit_cmd_window() or
3099 	 * greater than ist_expcmdsn, it's not in the window.
3100 	 */
3101 
3102 	if (iscsit_sna_lt(cmdsn, (ist->ist_expcmdsn - iscsit_cmd_window())) ||
3103 	    !iscsit_sna_lte(cmdsn, ist->ist_expcmdsn)) {
3104 		rval = B_FALSE;
3105 	}
3106 
3107 	mutex_exit(&ist->ist_sn_mutex);
3108 
3109 	return (rval);
3110 }
3111 
3112 /*
3113  * iscsit_check_cmdsn_and_queue
3114  *
3115  * Independent of the order in which the iSCSI target receives non-immediate
3116  * command PDU across the entire session and any multiple connections within
3117  * the session, the target must deliver the commands to the SCSI layer in
3118  * CmdSN order. So out-of-order non-immediate commands are queued up on a
3119  * session-wide wait queue. Duplicate commands are ignored.
3120  *
3121  * returns B_TRUE for commands which can be executed immediately (are
3122  * non-deferred), B_FALSE for cases where a command was deferred or invalid.
3123  */
3124 static boolean_t
3125 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu)
3126 {
3127 	idm_conn_t		*ic = rx_pdu->isp_ic;
3128 	iscsit_conn_t		*ict = ic->ic_handle;
3129 	iscsit_sess_t		*ist = ict->ict_sess;
3130 	iscsi_scsi_cmd_hdr_t	*hdr = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
3131 
3132 	mutex_enter(&ist->ist_sn_mutex);
3133 	if (hdr->opcode & ISCSI_OP_IMMEDIATE) {
3134 		/* do not queue, handle it immediately */
3135 		DTRACE_PROBE2(immediate__cmd, iscsit_sess_t *, ist,
3136 		    idm_pdu_t *, rx_pdu);
3137 		mutex_exit(&ist->ist_sn_mutex);
3138 		return (B_TRUE);
3139 	}
3140 	/*
3141 	 * See RFC3270 3.1.1.2: non-immediate commands outside of the
3142 	 * expected window (from expcmdsn to maxcmdsn, inclusive)
3143 	 * should be silently ignored.
3144 	 */
3145 	if (iscsit_sna_lt(ist->ist_expcmdsn, ntohl(hdr->cmdsn)) &&
3146 	    iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_maxcmdsn)) {
3147 		/*
3148 		 * Out-of-order commands (cmdSN higher than ExpCmdSN)
3149 		 * are staged on a fixed-size circular buffer until
3150 		 * the missing command is delivered to the SCSI layer.
3151 		 * Irrespective of the order of insertion into the
3152 		 * staging queue, the commands are processed out of the
3153 		 * queue in cmdSN order only.
3154 		 */
3155 		rx_pdu->isp_queue_time = gethrtime();
3156 		iscsit_add_pdu_to_queue(ist, rx_pdu);
3157 		mutex_exit(&ist->ist_sn_mutex);
3158 		return (B_FALSE);
3159 	} else if (iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_expcmdsn) ||
3160 	    iscsit_sna_lt(ist->ist_maxcmdsn, ntohl(hdr->cmdsn))) {
3161 		/*
3162 		 * See above, this command is outside of our acceptable
3163 		 * window, we need to discard/complete.
3164 		 */
3165 		DTRACE_PROBE3(cmdsn__lt__expcmdsn, iscsit_sess_t *, ist,
3166 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
3167 		mutex_exit(&ist->ist_sn_mutex);
3168 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
3169 		/*
3170 		 * tell our callers that the PDU "finished."
3171 		 */
3172 		return (B_FALSE);
3173 	} else {
3174 		mutex_exit(&ist->ist_sn_mutex);
3175 		return (B_TRUE);
3176 	}
3177 }
3178 
3179 /*
3180  * iscsit_add_pdu_to_queue() adds PDUs into the array indexed by
3181  * their cmdsn value. The length of the array is kept above the
3182  * maximum window size. The window keeps the cmdsn within a range
3183  * such that there are no collisons. e.g. the assumption is that
3184  * the windowing checks make it impossible to receive PDUs that
3185  * index into the same location in the array.
3186  */
3187 static void
3188 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu)
3189 {
3190 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3191 	iscsit_conn_t	*ict 	= rx_pdu->isp_ic->ic_handle;
3192 	uint32_t	cmdsn	=
3193 	    ((iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr)->cmdsn;
3194 	uint32_t	index;
3195 
3196 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3197 	/*
3198 	 * If the connection is being torn down, then
3199 	 * don't add the PDU to the staging queue
3200 	 */
3201 	mutex_enter(&ict->ict_mutex);
3202 	if (ict->ict_lost) {
3203 		mutex_exit(&ict->ict_mutex);
3204 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3205 		return;
3206 	}
3207 	iscsit_conn_dispatch_hold(ict);
3208 	mutex_exit(&ict->ict_mutex);
3209 
3210 	index = ntohl(cmdsn) % ISCSIT_RXPDU_QUEUE_LEN;
3211 	/*
3212 	 * In the normal case, assuming that the Initiator is not
3213 	 * buggy and that we don't have packet duplication occuring,
3214 	 * the entry in the array will be NULL.  However, we may have
3215 	 * received a duplicate PDU with cmdsn > expsn , and in that
3216 	 * case we just ignore this PDU -- the previously received one
3217 	 * remains queued for processing.  We need to be careful not
3218 	 * to leak this one however.
3219 	 */
3220 	if (cbuf->cb_buffer[index] != NULL) {
3221 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3222 	} else {
3223 		cbuf->cb_buffer[index] = rx_pdu;
3224 		cbuf->cb_num_elems++;
3225 	}
3226 }
3227 
3228 static idm_pdu_t *
3229 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn)
3230 {
3231 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3232 	idm_pdu_t	*pdu	= NULL;
3233 	uint32_t	index;
3234 
3235 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3236 	index = cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3237 	if ((pdu = cbuf->cb_buffer[index]) != NULL) {
3238 		ASSERT(cmdsn ==
3239 		    ntohl(((iscsi_scsi_cmd_hdr_t *)pdu->isp_hdr)->cmdsn));
3240 		cbuf->cb_buffer[index] = NULL;
3241 		cbuf->cb_num_elems--;
3242 		return (pdu);
3243 	}
3244 	return (NULL);
3245 }
3246 
3247 /*
3248  * iscsit_process_pdu_in_queue() finds the next pdu in sequence
3249  * and posts it to the SCSI layer
3250  */
3251 static void
3252 iscsit_process_pdu_in_queue(iscsit_sess_t *ist)
3253 {
3254 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3255 	idm_pdu_t	*pdu = NULL;
3256 	uint32_t	expcmdsn;
3257 
3258 	for (;;) {
3259 		mutex_enter(&ist->ist_sn_mutex);
3260 		if (cbuf->cb_num_elems == 0) {
3261 			mutex_exit(&ist->ist_sn_mutex);
3262 			break;
3263 		}
3264 		expcmdsn = ist->ist_expcmdsn;
3265 		if ((pdu = iscsit_remove_pdu_from_queue(ist, expcmdsn))
3266 		    == NULL) {
3267 			mutex_exit(&ist->ist_sn_mutex);
3268 			break;
3269 		}
3270 		mutex_exit(&ist->ist_sn_mutex);
3271 		iscsit_post_staged_pdu(pdu);
3272 	}
3273 }
3274 
3275 static void
3276 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu)
3277 {
3278 	iscsit_conn_t	*ict	= rx_pdu->isp_ic->ic_handle;
3279 
3280 	/* Post the PDU to the SCSI layer */
3281 	switch (IDM_PDU_OPCODE(rx_pdu)) {
3282 	case ISCSI_OP_NOOP_OUT:
3283 		iscsit_set_cmdsn(ict, rx_pdu);
3284 		iscsit_pdu_op_noop(ict, rx_pdu);
3285 		break;
3286 	case ISCSI_OP_TEXT_CMD:
3287 		iscsit_set_cmdsn(ict, rx_pdu);
3288 		iscsit_pdu_op_text_cmd(ict, rx_pdu);
3289 		break;
3290 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
3291 		iscsit_set_cmdsn(ict, rx_pdu);
3292 		iscsit_op_scsi_task_mgmt(ict, rx_pdu);
3293 		break;
3294 	case ISCSI_OP_SCSI_CMD:
3295 		/* cmdSN will be incremented after creating itask */
3296 		iscsit_post_scsi_cmd(rx_pdu->isp_ic, rx_pdu);
3297 		break;
3298 	case ISCSI_OP_LOGOUT_CMD:
3299 		iscsit_set_cmdsn(ict, rx_pdu);
3300 		iscsit_pdu_op_logout_cmd(ict, rx_pdu);
3301 		break;
3302 	default:
3303 		/* No other PDUs should be placed on the queue */
3304 		ASSERT(0);
3305 	}
3306 	iscsit_conn_dispatch_rele(ict); /* release hold on the conn */
3307 }
3308 
3309 /* ARGSUSED */
3310 void
3311 iscsit_rxpdu_queue_monitor_start(void)
3312 {
3313 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3314 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3315 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3316 		return;
3317 	}
3318 	iscsit_rxpdu_queue_monitor_thr_id =
3319 	    thread_create(NULL, 0, iscsit_rxpdu_queue_monitor, NULL,
3320 	    0, &p0, TS_RUN, minclsyspri);
3321 	while (!iscsit_rxpdu_queue_monitor_thr_running) {
3322 		cv_wait(&iscsit_rxpdu_queue_monitor_cv,
3323 		    &iscsit_rxpdu_queue_monitor_mutex);
3324 	}
3325 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3326 
3327 }
3328 
3329 /* ARGSUSED */
3330 void
3331 iscsit_rxpdu_queue_monitor_stop(void)
3332 {
3333 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3334 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3335 		iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
3336 		cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3337 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3338 
3339 		thread_join(iscsit_rxpdu_queue_monitor_thr_did);
3340 		return;
3341 	}
3342 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3343 }
3344 
3345 /*
3346  * A separate thread is used to scan the staging queue on all the
3347  * sessions, If a delayed PDU does not arrive within a timeout, the
3348  * target will advance to the staged PDU that is next in sequence
3349  * and exceeded the threshold wait time. It is up to the initiator
3350  * to note that the target has not acknowledged a particular cmdsn
3351  * and take appropriate action.
3352  */
3353 /* ARGSUSED */
3354 static void
3355 iscsit_rxpdu_queue_monitor(void *arg)
3356 {
3357 	iscsit_tgt_t	*tgt;
3358 	iscsit_sess_t	*ist;
3359 
3360 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3361 	iscsit_rxpdu_queue_monitor_thr_did = curthread->t_did;
3362 	iscsit_rxpdu_queue_monitor_thr_running = B_TRUE;
3363 	cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3364 
3365 	while (iscsit_rxpdu_queue_monitor_thr_running) {
3366 		ISCSIT_GLOBAL_LOCK(RW_READER);
3367 		for (tgt = avl_first(&iscsit_global.global_target_list);
3368 		    tgt != NULL;
3369 		    tgt = AVL_NEXT(&iscsit_global.global_target_list, tgt)) {
3370 			mutex_enter(&tgt->target_mutex);
3371 			for (ist = avl_first(&tgt->target_sess_list);
3372 			    ist != NULL;
3373 			    ist = AVL_NEXT(&tgt->target_sess_list, ist)) {
3374 
3375 				iscsit_rxpdu_queue_monitor_session(ist);
3376 			}
3377 			mutex_exit(&tgt->target_mutex);
3378 		}
3379 		ISCSIT_GLOBAL_UNLOCK();
3380 		if (iscsit_rxpdu_queue_monitor_thr_running == B_FALSE) {
3381 			break;
3382 		}
3383 		(void) cv_reltimedwait(&iscsit_rxpdu_queue_monitor_cv,
3384 		    &iscsit_rxpdu_queue_monitor_mutex,
3385 		    ISCSIT_RXPDU_QUEUE_MONITOR_INTERVAL * drv_usectohz(1000000),
3386 		    TR_CLOCK_TICK);
3387 	}
3388 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3389 	thread_exit();
3390 }
3391 
3392 static void
3393 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist)
3394 {
3395 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3396 	idm_pdu_t	*next_pdu = NULL;
3397 	uint32_t	index, next_cmdsn, i;
3398 
3399 	/*
3400 	 * Assume that all PDUs in the staging queue have a cmdsn >= expcmdsn.
3401 	 * Starting with the expcmdsn, iterate over the staged PDUs to find
3402 	 * the next PDU with a wait time greater than the threshold. If found
3403 	 * advance the staged PDU to the SCSI layer, skipping over the missing
3404 	 * PDU(s) to get past the hole in the command sequence. It is up to
3405 	 * the initiator to note that the target has not acknowledged a cmdsn
3406 	 * and take appropriate action.
3407 	 *
3408 	 * Since the PDU(s) arrive in any random order, it is possible that
3409 	 * that the actual wait time for a particular PDU is much longer than
3410 	 * the defined threshold. e.g. Consider a case where commands are sent
3411 	 * over 4 different connections, and cmdsn = 1004 arrives first, then
3412 	 * 1003, and 1002 and 1001 are lost due to a connection failure.
3413 	 * So now 1003 is waiting for 1002 to be delivered, and although the
3414 	 * wait time of 1004 > wait time of 1003, only 1003 will be considered
3415 	 * by the monitor thread. 1004 will be automatically processed by
3416 	 * iscsit_process_pdu_in_queue() once the scan is complete and the
3417 	 * expcmdsn becomes current.
3418 	 */
3419 	mutex_enter(&ist->ist_sn_mutex);
3420 	cbuf = ist->ist_rxpdu_queue;
3421 	if (cbuf->cb_num_elems == 0) {
3422 		mutex_exit(&ist->ist_sn_mutex);
3423 		return;
3424 	}
3425 	for (next_pdu = NULL, i = 0; ; i++) {
3426 		next_cmdsn = ist->ist_expcmdsn + i; /* start at expcmdsn */
3427 		index = next_cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3428 		if ((next_pdu = cbuf->cb_buffer[index]) != NULL) {
3429 			/*
3430 			 * If the PDU wait time has not exceeded threshold
3431 			 * stop scanning the staging queue until the timer
3432 			 * fires again
3433 			 */
3434 			if ((gethrtime() - next_pdu->isp_queue_time)
3435 			    < (rxpdu_queue_threshold * NANOSEC)) {
3436 				mutex_exit(&ist->ist_sn_mutex);
3437 				return;
3438 			}
3439 			/*
3440 			 * Remove the next PDU from the queue and post it
3441 			 * to the SCSI layer, skipping over the missing
3442 			 * PDU. Stop scanning the staging queue until
3443 			 * the monitor timer fires again
3444 			 */
3445 			(void) iscsit_remove_pdu_from_queue(ist, next_cmdsn);
3446 			mutex_exit(&ist->ist_sn_mutex);
3447 			DTRACE_PROBE3(advanced__to__blocked__cmdsn,
3448 			    iscsit_sess_t *, ist, idm_pdu_t *, next_pdu,
3449 			    uint32_t, next_cmdsn);
3450 			iscsit_post_staged_pdu(next_pdu);
3451 			/* Deliver any subsequent PDUs immediately */
3452 			iscsit_process_pdu_in_queue(ist);
3453 			return;
3454 		}
3455 		/*
3456 		 * Skipping over i PDUs, e.g. a case where commands 1001 and
3457 		 * 1002 are lost in the network, skip over both and post 1003
3458 		 * expcmdsn then becomes 1004 at the end of the scan.
3459 		 */
3460 		DTRACE_PROBE2(skipping__over__cmdsn, iscsit_sess_t *, ist,
3461 		    uint32_t, next_cmdsn);
3462 	}
3463 	/*
3464 	 * following the assumption, staged cmdsn >= expcmdsn, this statement
3465 	 * is never reached.
3466 	 */
3467 }
3468