xref: /titanic_51/usr/src/uts/common/io/comstar/port/iscsit/iscsit.c (revision aab83bb83be7342f6cfccaed8d5fe0b2f404855d)
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 Nexenta Systems, Inc. All rights reserved.
25  */
26 
27 #include <sys/cpuvar.h>
28 #include <sys/types.h>
29 #include <sys/conf.h>
30 #include <sys/stat.h>
31 #include <sys/file.h>
32 #include <sys/ddi.h>
33 #include <sys/sunddi.h>
34 #include <sys/modctl.h>
35 #include <sys/sysmacros.h>
36 #include <sys/socket.h>
37 #include <sys/strsubr.h>
38 #include <sys/nvpair.h>
39 
40 #include <sys/stmf.h>
41 #include <sys/stmf_ioctl.h>
42 #include <sys/portif.h>
43 #include <sys/idm/idm.h>
44 #include <sys/idm/idm_conn_sm.h>
45 
46 #include "iscsit_isns.h"
47 #include "iscsit.h"
48 
49 #define	ISCSIT_VERSION		BUILD_DATE "-1.18dev"
50 #define	ISCSIT_NAME_VERSION	"COMSTAR ISCSIT v" ISCSIT_VERSION
51 
52 /*
53  * DDI entry points.
54  */
55 static int iscsit_drv_attach(dev_info_t *, ddi_attach_cmd_t);
56 static int iscsit_drv_detach(dev_info_t *, ddi_detach_cmd_t);
57 static int iscsit_drv_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
58 static int iscsit_drv_open(dev_t *, int, int, cred_t *);
59 static int iscsit_drv_close(dev_t, int, int, cred_t *);
60 static boolean_t iscsit_drv_busy(void);
61 static int iscsit_drv_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
62 
63 extern struct mod_ops mod_miscops;
64 
65 
66 static struct cb_ops iscsit_cb_ops = {
67 	iscsit_drv_open,	/* cb_open */
68 	iscsit_drv_close,	/* cb_close */
69 	nodev,			/* cb_strategy */
70 	nodev,			/* cb_print */
71 	nodev,			/* cb_dump */
72 	nodev,			/* cb_read */
73 	nodev,			/* cb_write */
74 	iscsit_drv_ioctl,	/* cb_ioctl */
75 	nodev,			/* cb_devmap */
76 	nodev,			/* cb_mmap */
77 	nodev,			/* cb_segmap */
78 	nochpoll,		/* cb_chpoll */
79 	ddi_prop_op,		/* cb_prop_op */
80 	NULL,			/* cb_streamtab */
81 	D_MP,			/* cb_flag */
82 	CB_REV,			/* cb_rev */
83 	nodev,			/* cb_aread */
84 	nodev,			/* cb_awrite */
85 };
86 
87 static struct dev_ops iscsit_dev_ops = {
88 	DEVO_REV,		/* devo_rev */
89 	0,			/* devo_refcnt */
90 	iscsit_drv_getinfo,	/* devo_getinfo */
91 	nulldev,		/* devo_identify */
92 	nulldev,		/* devo_probe */
93 	iscsit_drv_attach,	/* devo_attach */
94 	iscsit_drv_detach,	/* devo_detach */
95 	nodev,			/* devo_reset */
96 	&iscsit_cb_ops,		/* devo_cb_ops */
97 	NULL,			/* devo_bus_ops */
98 	NULL,			/* devo_power */
99 	ddi_quiesce_not_needed,	/* quiesce */
100 };
101 
102 static struct modldrv modldrv = {
103 	&mod_driverops,
104 	"iSCSI Target",
105 	&iscsit_dev_ops,
106 };
107 
108 static struct modlinkage modlinkage = {
109 	MODREV_1,
110 	&modldrv,
111 	NULL,
112 };
113 
114 
115 iscsit_global_t iscsit_global;
116 
117 kmem_cache_t	*iscsit_status_pdu_cache;
118 
119 boolean_t	iscsit_sm_logging = B_FALSE;
120 
121 kmutex_t	login_sm_session_mutex;
122 
123 static idm_status_t iscsit_init(dev_info_t *dip);
124 static idm_status_t iscsit_enable_svc(iscsit_hostinfo_t *hostinfo);
125 static void iscsit_disable_svc(void);
126 
127 static int
128 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu);
129 
130 static void
131 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu);
132 
133 static idm_pdu_t *
134 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn);
135 
136 static void
137 iscsit_process_pdu_in_queue(iscsit_sess_t *ist);
138 
139 static void
140 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist);
141 
142 static void
143 iscsit_rxpdu_queue_monitor(void *arg);
144 
145 static void
146 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu);
147 
148 static void
149 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu);
150 
151 static void
152 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
153 
154 static void
155 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
156 
157 static void
158 iscsit_pdu_op_login_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
159 
160 void
161 iscsit_pdu_op_text_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
162 
163 static void
164 iscsit_pdu_op_logout_cmd(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
165 
166 int iscsit_cmd_window();
167 
168 static  int
169 iscsit_sna_lt(uint32_t sn1, uint32_t sn2);
170 
171 void
172 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu);
173 
174 static void
175 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu);
176 
177 static void
178 iscsit_deferred(void *rx_pdu_void);
179 
180 static idm_status_t
181 iscsit_conn_accept(idm_conn_t *ic);
182 
183 static idm_status_t
184 iscsit_ffp_enabled(idm_conn_t *ic);
185 
186 static idm_status_t
187 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class);
188 
189 static idm_status_t
190 iscsit_conn_lost(idm_conn_t *ic);
191 
192 static idm_status_t
193 iscsit_conn_destroy(idm_conn_t *ic);
194 
195 static stmf_data_buf_t *
196 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
197     uint32_t flags);
198 
199 static void
200 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf);
201 
202 static void
203 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status);
204 
205 static void
206 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status);
207 
208 static void
209 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status);
210 
211 static stmf_status_t
212 iscsit_idm_to_stmf(idm_status_t idmrc);
213 
214 static iscsit_task_t *
215 iscsit_task_alloc(iscsit_conn_t *ict);
216 
217 static void
218 iscsit_task_free(iscsit_task_t *itask);
219 
220 static iscsit_task_t *
221 iscsit_tm_task_alloc(iscsit_conn_t *ict);
222 
223 static void
224 iscsit_tm_task_free(iscsit_task_t *itask);
225 
226 static idm_status_t
227 iscsit_task_start(iscsit_task_t *itask);
228 
229 static void
230 iscsit_task_done(iscsit_task_t *itask);
231 
232 static int
233 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags);
234 
235 static void
236 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags);
237 
238 static it_cfg_status_t
239 iscsit_config_merge(it_config_t *cfg);
240 
241 static idm_status_t
242 iscsit_login_fail(idm_conn_t *ic);
243 
244 static boolean_t iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn);
245 static void iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
246     uint8_t response, uint8_t cmd_status);
247 static void iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu,
248     uint8_t tm_status);
249 
250 /*
251  * MC/S: Out-of-order commands are staged on a session-wide wait
252  * queue until a system-tunable threshold is reached. A separate
253  * thread is used to scan the staging queue on all the session,
254  * If a delayed PDU does not arrive within a timeout, the target
255  * will advance to the staged PDU that is next in sequence, skipping
256  * over the missing PDU(s) to go past a hole in the sequence.
257  */
258 volatile int rxpdu_queue_threshold = ISCSIT_RXPDU_QUEUE_THRESHOLD;
259 
260 static kmutex_t		iscsit_rxpdu_queue_monitor_mutex;
261 kthread_t		*iscsit_rxpdu_queue_monitor_thr_id;
262 static kt_did_t		iscsit_rxpdu_queue_monitor_thr_did;
263 static boolean_t	iscsit_rxpdu_queue_monitor_thr_running;
264 static kcondvar_t	iscsit_rxpdu_queue_monitor_cv;
265 
266 int
267 _init(void)
268 {
269 	int rc;
270 
271 	rw_init(&iscsit_global.global_rwlock, NULL, RW_DRIVER, NULL);
272 	mutex_init(&iscsit_global.global_state_mutex, NULL,
273 	    MUTEX_DRIVER, NULL);
274 	iscsit_global.global_svc_state = ISE_DETACHED;
275 
276 	mutex_init(&iscsit_rxpdu_queue_monitor_mutex, NULL,
277 	    MUTEX_DRIVER, NULL);
278 	mutex_init(&login_sm_session_mutex, NULL, MUTEX_DRIVER, NULL);
279 	iscsit_rxpdu_queue_monitor_thr_id = NULL;
280 	iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
281 	cv_init(&iscsit_rxpdu_queue_monitor_cv, NULL, CV_DEFAULT, NULL);
282 
283 	if ((rc = mod_install(&modlinkage)) != 0) {
284 		mutex_destroy(&iscsit_global.global_state_mutex);
285 		rw_destroy(&iscsit_global.global_rwlock);
286 		return (rc);
287 	}
288 
289 	return (rc);
290 }
291 
292 int
293 _info(struct modinfo *modinfop)
294 {
295 	return (mod_info(&modlinkage, modinfop));
296 }
297 
298 int
299 _fini(void)
300 {
301 	int rc;
302 
303 	rc = mod_remove(&modlinkage);
304 
305 	if (rc == 0) {
306 		mutex_destroy(&iscsit_rxpdu_queue_monitor_mutex);
307 		mutex_destroy(&login_sm_session_mutex);
308 		cv_destroy(&iscsit_rxpdu_queue_monitor_cv);
309 		mutex_destroy(&iscsit_global.global_state_mutex);
310 		rw_destroy(&iscsit_global.global_rwlock);
311 	}
312 
313 	return (rc);
314 }
315 
316 /*
317  * DDI entry points.
318  */
319 
320 /* ARGSUSED */
321 static int
322 iscsit_drv_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
323     void **result)
324 {
325 	ulong_t instance = getminor((dev_t)arg);
326 
327 	switch (cmd) {
328 	case DDI_INFO_DEVT2DEVINFO:
329 		*result = iscsit_global.global_dip;
330 		return (DDI_SUCCESS);
331 
332 	case DDI_INFO_DEVT2INSTANCE:
333 		*result = (void *)instance;
334 		return (DDI_SUCCESS);
335 
336 	default:
337 		break;
338 	}
339 
340 	return (DDI_FAILURE);
341 }
342 
343 static int
344 iscsit_drv_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
345 {
346 	if (cmd != DDI_ATTACH) {
347 		return (DDI_FAILURE);
348 	}
349 
350 	if (ddi_get_instance(dip) != 0) {
351 		/* we only allow instance 0 to attach */
352 		return (DDI_FAILURE);
353 	}
354 
355 	/* create the minor node */
356 	if (ddi_create_minor_node(dip, ISCSIT_MODNAME, S_IFCHR, 0,
357 	    DDI_PSEUDO, 0) != DDI_SUCCESS) {
358 		cmn_err(CE_WARN, "iscsit_drv_attach: "
359 		    "failed creating minor node");
360 		return (DDI_FAILURE);
361 	}
362 
363 	if (iscsit_init(dip) != IDM_STATUS_SUCCESS) {
364 		cmn_err(CE_WARN, "iscsit_drv_attach: "
365 		    "failed to initialize");
366 		ddi_remove_minor_node(dip, NULL);
367 		return (DDI_FAILURE);
368 	}
369 
370 	iscsit_global.global_svc_state = ISE_DISABLED;
371 	iscsit_global.global_dip = dip;
372 
373 	return (DDI_SUCCESS);
374 }
375 
376 /*ARGSUSED*/
377 static int
378 iscsit_drv_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
379 {
380 	if (cmd != DDI_DETACH)
381 		return (DDI_FAILURE);
382 
383 	/*
384 	 * drv_detach is called in a context that owns the
385 	 * device node for the /dev/pseudo device.  If this thread blocks
386 	 * for any resource, other threads that need the /dev/pseudo device
387 	 * may end up in a deadlock with this thread.Hence, we use a
388 	 * separate lock just for the structures that drv_detach needs
389 	 * to access.
390 	 */
391 	mutex_enter(&iscsit_global.global_state_mutex);
392 	if (iscsit_drv_busy()) {
393 		mutex_exit(&iscsit_global.global_state_mutex);
394 		return (EBUSY);
395 	}
396 
397 	iscsit_global.global_dip = NULL;
398 	ddi_remove_minor_node(dip, NULL);
399 
400 	ldi_ident_release(iscsit_global.global_li);
401 	iscsit_global.global_svc_state = ISE_DETACHED;
402 
403 	mutex_exit(&iscsit_global.global_state_mutex);
404 
405 	return (DDI_SUCCESS);
406 }
407 
408 /*ARGSUSED*/
409 static int
410 iscsit_drv_open(dev_t *devp, int flag, int otyp, cred_t *credp)
411 {
412 	return (0);
413 }
414 
415 /* ARGSUSED */
416 static int
417 iscsit_drv_close(dev_t dev, int flag, int otyp, cred_t *credp)
418 {
419 	return (0);
420 }
421 
422 static boolean_t
423 iscsit_drv_busy(void)
424 {
425 	ASSERT(MUTEX_HELD(&iscsit_global.global_state_mutex));
426 
427 	switch (iscsit_global.global_svc_state) {
428 	case ISE_DISABLED:
429 	case ISE_DETACHED:
430 		return (B_FALSE);
431 	default:
432 		return (B_TRUE);
433 	}
434 	/* NOTREACHED */
435 }
436 
437 /* ARGSUSED */
438 static int
439 iscsit_drv_ioctl(dev_t drv, int cmd, intptr_t argp, int flag, cred_t *cred,
440     int *retval)
441 {
442 	iscsit_ioc_set_config_t		setcfg;
443 	iscsit_ioc_set_config32_t	setcfg32;
444 	char				*cfg_pnvlist = NULL;
445 	nvlist_t			*cfg_nvlist = NULL;
446 	it_config_t			*cfg = NULL;
447 	idm_status_t			idmrc;
448 	int				rc = 0;
449 
450 	if (drv_priv(cred) != 0) {
451 		return (EPERM);
452 	}
453 
454 	mutex_enter(&iscsit_global.global_state_mutex);
455 
456 	/*
457 	 * Validate ioctl requests against global service state
458 	 */
459 	switch (iscsit_global.global_svc_state) {
460 	case ISE_ENABLED:
461 		if (cmd == ISCSIT_IOC_DISABLE_SVC) {
462 			iscsit_global.global_svc_state = ISE_DISABLING;
463 		} else if (cmd == ISCSIT_IOC_ENABLE_SVC) {
464 			/* Already enabled */
465 			mutex_exit(&iscsit_global.global_state_mutex);
466 			return (0);
467 		} else {
468 			iscsit_global.global_svc_state = ISE_BUSY;
469 		}
470 		break;
471 	case ISE_DISABLED:
472 		if (cmd == ISCSIT_IOC_ENABLE_SVC) {
473 			iscsit_global.global_svc_state = ISE_ENABLING;
474 		} else if (cmd == ISCSIT_IOC_DISABLE_SVC) {
475 			/* Already disabled */
476 			mutex_exit(&iscsit_global.global_state_mutex);
477 			return (0);
478 		} else {
479 			rc = EFAULT;
480 		}
481 		break;
482 	case ISE_BUSY:
483 	case ISE_ENABLING:
484 	case ISE_DISABLING:
485 		rc = EAGAIN;
486 		break;
487 	case ISE_DETACHED:
488 	default:
489 		rc = EFAULT;
490 		break;
491 	}
492 
493 	mutex_exit(&iscsit_global.global_state_mutex);
494 	if (rc != 0)
495 		return (rc);
496 
497 	/* Handle ioctl request (enable/disable have already been handled) */
498 	switch (cmd) {
499 	case ISCSIT_IOC_SET_CONFIG:
500 		/* Any errors must set state back to ISE_ENABLED */
501 		switch (ddi_model_convert_from(flag & FMODELS)) {
502 		case DDI_MODEL_ILP32:
503 			if (ddi_copyin((void *)argp, &setcfg32,
504 			    sizeof (iscsit_ioc_set_config32_t), flag) != 0) {
505 				rc = EFAULT;
506 				goto cleanup;
507 			}
508 
509 			setcfg.set_cfg_pnvlist =
510 			    (char *)((uintptr_t)setcfg32.set_cfg_pnvlist);
511 			setcfg.set_cfg_vers = setcfg32.set_cfg_vers;
512 			setcfg.set_cfg_pnvlist_len =
513 			    setcfg32.set_cfg_pnvlist_len;
514 			break;
515 		case DDI_MODEL_NONE:
516 			if (ddi_copyin((void *)argp, &setcfg,
517 			    sizeof (iscsit_ioc_set_config_t), flag) != 0) {
518 				rc = EFAULT;
519 				goto cleanup;
520 			}
521 			break;
522 		default:
523 			rc = EFAULT;
524 			goto cleanup;
525 		}
526 
527 		/* Check API version */
528 		if (setcfg.set_cfg_vers != ISCSIT_API_VERS0) {
529 			rc = EINVAL;
530 			goto cleanup;
531 		}
532 
533 		/* Config is in packed nvlist format so unpack it */
534 		cfg_pnvlist = kmem_alloc(setcfg.set_cfg_pnvlist_len,
535 		    KM_SLEEP);
536 		ASSERT(cfg_pnvlist != NULL);
537 
538 		if (ddi_copyin(setcfg.set_cfg_pnvlist, cfg_pnvlist,
539 		    setcfg.set_cfg_pnvlist_len, flag) != 0) {
540 			rc = EFAULT;
541 			goto cleanup;
542 		}
543 
544 		rc = nvlist_unpack(cfg_pnvlist, setcfg.set_cfg_pnvlist_len,
545 		    &cfg_nvlist, KM_SLEEP);
546 		if (rc != 0) {
547 			goto cleanup;
548 		}
549 
550 		/* Translate nvlist */
551 		rc = it_nv_to_config(cfg_nvlist, &cfg);
552 		if (rc != 0) {
553 			cmn_err(CE_WARN, "Configuration is invalid");
554 			goto cleanup;
555 		}
556 
557 		/* Update config */
558 		rc = iscsit_config_merge(cfg);
559 		/* FALLTHROUGH */
560 
561 cleanup:
562 		if (cfg)
563 			it_config_free_cmn(cfg);
564 		if (cfg_pnvlist)
565 			kmem_free(cfg_pnvlist, setcfg.set_cfg_pnvlist_len);
566 		nvlist_free(cfg_nvlist);
567 
568 		/*
569 		 * Now that the reconfig is complete set our state back to
570 		 * enabled.
571 		 */
572 		mutex_enter(&iscsit_global.global_state_mutex);
573 		iscsit_global.global_svc_state = ISE_ENABLED;
574 		mutex_exit(&iscsit_global.global_state_mutex);
575 		break;
576 	case ISCSIT_IOC_ENABLE_SVC: {
577 		iscsit_hostinfo_t hostinfo;
578 
579 		if (ddi_copyin((void *)argp, &hostinfo.length,
580 		    sizeof (hostinfo.length), flag) != 0) {
581 			mutex_enter(&iscsit_global.global_state_mutex);
582 			iscsit_global.global_svc_state = ISE_DISABLED;
583 			mutex_exit(&iscsit_global.global_state_mutex);
584 			return (EFAULT);
585 		}
586 
587 		if (hostinfo.length > sizeof (hostinfo.fqhn))
588 			hostinfo.length = sizeof (hostinfo.fqhn);
589 
590 		if (ddi_copyin((void *)((caddr_t)argp +
591 		    sizeof (hostinfo.length)), &hostinfo.fqhn,
592 		    hostinfo.length, flag) != 0) {
593 			mutex_enter(&iscsit_global.global_state_mutex);
594 			iscsit_global.global_svc_state = ISE_DISABLED;
595 			mutex_exit(&iscsit_global.global_state_mutex);
596 			return (EFAULT);
597 		}
598 
599 		idmrc = iscsit_enable_svc(&hostinfo);
600 		mutex_enter(&iscsit_global.global_state_mutex);
601 		if (idmrc == IDM_STATUS_SUCCESS) {
602 			iscsit_global.global_svc_state = ISE_ENABLED;
603 		} else {
604 			rc = EIO;
605 			iscsit_global.global_svc_state = ISE_DISABLED;
606 		}
607 		mutex_exit(&iscsit_global.global_state_mutex);
608 		break;
609 	}
610 	case ISCSIT_IOC_DISABLE_SVC:
611 		iscsit_disable_svc();
612 		mutex_enter(&iscsit_global.global_state_mutex);
613 		iscsit_global.global_svc_state = ISE_DISABLED;
614 		mutex_exit(&iscsit_global.global_state_mutex);
615 		break;
616 
617 	default:
618 		rc = EINVAL;
619 		mutex_enter(&iscsit_global.global_state_mutex);
620 		iscsit_global.global_svc_state = ISE_ENABLED;
621 		mutex_exit(&iscsit_global.global_state_mutex);
622 	}
623 
624 	return (rc);
625 }
626 
627 static idm_status_t
628 iscsit_init(dev_info_t *dip)
629 {
630 	int			rc;
631 
632 	rc = ldi_ident_from_dip(dip, &iscsit_global.global_li);
633 	ASSERT(rc == 0);  /* Failure indicates invalid argument */
634 
635 	iscsit_global.global_svc_state = ISE_DISABLED;
636 
637 	return (IDM_STATUS_SUCCESS);
638 }
639 
640 /*
641  * iscsit_enable_svc
642  *
643  * registers all the configured targets and target portals with STMF
644  */
645 static idm_status_t
646 iscsit_enable_svc(iscsit_hostinfo_t *hostinfo)
647 {
648 	stmf_port_provider_t	*pp;
649 	stmf_dbuf_store_t	*dbuf_store;
650 	boolean_t		did_iscsit_isns_init;
651 	idm_status_t		retval = IDM_STATUS_SUCCESS;
652 
653 	ASSERT(iscsit_global.global_svc_state == ISE_ENABLING);
654 
655 	/*
656 	 * Make sure that can tell if we have partially allocated
657 	 * in case we need to exit and tear down anything allocated.
658 	 */
659 	iscsit_global.global_tsih_pool = NULL;
660 	iscsit_global.global_dbuf_store = NULL;
661 	iscsit_status_pdu_cache = NULL;
662 	pp = NULL;
663 	iscsit_global.global_pp = NULL;
664 	iscsit_global.global_default_tpg = NULL;
665 	did_iscsit_isns_init = B_FALSE;
666 	iscsit_global.global_dispatch_taskq = NULL;
667 
668 	/* Setup remaining fields in iscsit_global_t */
669 	idm_refcnt_init(&iscsit_global.global_refcnt,
670 	    &iscsit_global);
671 
672 	avl_create(&iscsit_global.global_discovery_sessions,
673 	    iscsit_sess_avl_compare, sizeof (iscsit_sess_t),
674 	    offsetof(iscsit_sess_t, ist_tgt_ln));
675 
676 	avl_create(&iscsit_global.global_target_list,
677 	    iscsit_tgt_avl_compare, sizeof (iscsit_tgt_t),
678 	    offsetof(iscsit_tgt_t, target_global_ln));
679 
680 	list_create(&iscsit_global.global_deleted_target_list,
681 	    sizeof (iscsit_tgt_t),
682 	    offsetof(iscsit_tgt_t, target_global_deleted_ln));
683 
684 	avl_create(&iscsit_global.global_tpg_list,
685 	    iscsit_tpg_avl_compare, sizeof (iscsit_tpg_t),
686 	    offsetof(iscsit_tpg_t, tpg_global_ln));
687 
688 	avl_create(&iscsit_global.global_ini_list,
689 	    iscsit_ini_avl_compare, sizeof (iscsit_ini_t),
690 	    offsetof(iscsit_ini_t, ini_global_ln));
691 
692 	iscsit_global.global_tsih_pool = vmem_create("iscsit_tsih_pool",
693 	    (void *)1, ISCSI_MAX_TSIH, 1, NULL, NULL, NULL, 0,
694 	    VM_SLEEP | VMC_IDENTIFIER);
695 
696 	/*
697 	 * Setup STMF dbuf store.  Our buffers are bound to a specific
698 	 * connection so we really can't let STMF cache buffers for us.
699 	 * Consequently we'll just allocate one global buffer store.
700 	 */
701 	dbuf_store = stmf_alloc(STMF_STRUCT_DBUF_STORE, 0, 0);
702 	if (dbuf_store == NULL) {
703 		retval = IDM_STATUS_FAIL;
704 		goto tear_down_and_return;
705 	}
706 	dbuf_store->ds_alloc_data_buf = iscsit_dbuf_alloc;
707 	dbuf_store->ds_free_data_buf = iscsit_dbuf_free;
708 	dbuf_store->ds_port_private = NULL;
709 	iscsit_global.global_dbuf_store = dbuf_store;
710 
711 	/* Status PDU cache */
712 	iscsit_status_pdu_cache = kmem_cache_create("iscsit_status_pdu_cache",
713 	    sizeof (idm_pdu_t) + sizeof (iscsi_scsi_rsp_hdr_t), 8,
714 	    &iscsit_status_pdu_constructor,
715 	    NULL, NULL, NULL, NULL, KM_SLEEP);
716 
717 	/* Default TPG and portal */
718 	iscsit_global.global_default_tpg = iscsit_tpg_createdefault();
719 	if (iscsit_global.global_default_tpg == NULL) {
720 		retval = IDM_STATUS_FAIL;
721 		goto tear_down_and_return;
722 	}
723 
724 	/* initialize isns client */
725 	(void) iscsit_isns_init(hostinfo);
726 	did_iscsit_isns_init = B_TRUE;
727 
728 	/* Register port provider */
729 	pp = stmf_alloc(STMF_STRUCT_PORT_PROVIDER, 0, 0);
730 	if (pp == NULL) {
731 		retval = IDM_STATUS_FAIL;
732 		goto tear_down_and_return;
733 	}
734 
735 	pp->pp_portif_rev = PORTIF_REV_1;
736 	pp->pp_instance = 0;
737 	pp->pp_name = ISCSIT_MODNAME;
738 	pp->pp_cb = iscsit_pp_cb;
739 
740 	iscsit_global.global_pp = pp;
741 
742 
743 	if (stmf_register_port_provider(pp) != STMF_SUCCESS) {
744 		retval = IDM_STATUS_FAIL;
745 		goto tear_down_and_return;
746 	}
747 
748 	iscsit_global.global_dispatch_taskq = taskq_create("iscsit_dispatch",
749 	    1, minclsyspri, 16, 16, TASKQ_PREPOPULATE);
750 
751 	/* Scan staged PDUs, meaningful in MC/S situations */
752 	iscsit_rxpdu_queue_monitor_start();
753 
754 	return (IDM_STATUS_SUCCESS);
755 
756 tear_down_and_return:
757 
758 	if (iscsit_global.global_dispatch_taskq) {
759 		taskq_destroy(iscsit_global.global_dispatch_taskq);
760 		iscsit_global.global_dispatch_taskq = NULL;
761 	}
762 
763 	if (did_iscsit_isns_init)
764 		iscsit_isns_fini();
765 
766 	if (iscsit_global.global_default_tpg) {
767 		iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
768 		iscsit_global.global_default_tpg = NULL;
769 	}
770 
771 	if (iscsit_global.global_pp)
772 		iscsit_global.global_pp = NULL;
773 
774 	if (pp)
775 		stmf_free(pp);
776 
777 	if (iscsit_status_pdu_cache) {
778 		kmem_cache_destroy(iscsit_status_pdu_cache);
779 		iscsit_status_pdu_cache = NULL;
780 	}
781 
782 	if (iscsit_global.global_dbuf_store) {
783 		stmf_free(iscsit_global.global_dbuf_store);
784 		iscsit_global.global_dbuf_store = NULL;
785 	}
786 
787 	if (iscsit_global.global_tsih_pool) {
788 		vmem_destroy(iscsit_global.global_tsih_pool);
789 		iscsit_global.global_tsih_pool = NULL;
790 	}
791 
792 	avl_destroy(&iscsit_global.global_ini_list);
793 	avl_destroy(&iscsit_global.global_tpg_list);
794 	list_destroy(&iscsit_global.global_deleted_target_list);
795 	avl_destroy(&iscsit_global.global_target_list);
796 	avl_destroy(&iscsit_global.global_discovery_sessions);
797 
798 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
799 
800 	return (retval);
801 }
802 
803 /*
804  * iscsit_disable_svc
805  *
806  * clean up all existing connections and deregister targets from STMF
807  */
808 static void
809 iscsit_disable_svc(void)
810 {
811 	iscsit_sess_t	*sess;
812 
813 	ASSERT(iscsit_global.global_svc_state == ISE_DISABLING);
814 
815 	iscsit_rxpdu_queue_monitor_stop();
816 
817 	/* tear down discovery sessions */
818 	for (sess = avl_first(&iscsit_global.global_discovery_sessions);
819 	    sess != NULL;
820 	    sess = AVL_NEXT(&iscsit_global.global_discovery_sessions, sess))
821 		iscsit_sess_close(sess);
822 
823 	/*
824 	 * Passing NULL to iscsit_config_merge tells it to go to an empty
825 	 * config.
826 	 */
827 	(void) iscsit_config_merge(NULL);
828 
829 	/*
830 	 * Wait until there are no more global references
831 	 */
832 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
833 	idm_refcnt_destroy(&iscsit_global.global_refcnt);
834 
835 	/*
836 	 * Default TPG must be destroyed after global_refcnt is 0.
837 	 */
838 	iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
839 
840 	avl_destroy(&iscsit_global.global_discovery_sessions);
841 	list_destroy(&iscsit_global.global_deleted_target_list);
842 	avl_destroy(&iscsit_global.global_target_list);
843 	avl_destroy(&iscsit_global.global_tpg_list);
844 	avl_destroy(&iscsit_global.global_ini_list);
845 
846 	taskq_destroy(iscsit_global.global_dispatch_taskq);
847 
848 	iscsit_isns_fini();
849 
850 	stmf_free(iscsit_global.global_dbuf_store);
851 	iscsit_global.global_dbuf_store = NULL;
852 
853 	(void) stmf_deregister_port_provider(iscsit_global.global_pp);
854 	stmf_free(iscsit_global.global_pp);
855 	iscsit_global.global_pp = NULL;
856 
857 	kmem_cache_destroy(iscsit_status_pdu_cache);
858 	iscsit_status_pdu_cache = NULL;
859 
860 	vmem_destroy(iscsit_global.global_tsih_pool);
861 	iscsit_global.global_tsih_pool = NULL;
862 }
863 
864 void
865 iscsit_global_hold()
866 {
867 	/*
868 	 * To take out a global hold, we must either own the global
869 	 * state mutex or we must be running inside of an ioctl that
870 	 * has set the global state to ISE_BUSY, ISE_DISABLING, or
871 	 * ISE_ENABLING.  We don't track the "owner" for these flags,
872 	 * so just checking if they are set is enough for now.
873 	 */
874 	ASSERT((iscsit_global.global_svc_state == ISE_ENABLING) ||
875 	    (iscsit_global.global_svc_state == ISE_DISABLING) ||
876 	    (iscsit_global.global_svc_state == ISE_BUSY) ||
877 	    MUTEX_HELD(&iscsit_global.global_state_mutex));
878 
879 	idm_refcnt_hold(&iscsit_global.global_refcnt);
880 }
881 
882 void
883 iscsit_global_rele()
884 {
885 	idm_refcnt_rele(&iscsit_global.global_refcnt);
886 }
887 
888 void
889 iscsit_global_wait_ref()
890 {
891 	idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
892 }
893 
894 /*
895  * IDM callbacks
896  */
897 
898 /*ARGSUSED*/
899 void
900 iscsit_rx_pdu(idm_conn_t *ic, idm_pdu_t *rx_pdu)
901 {
902 	iscsit_conn_t *ict = ic->ic_handle;
903 	switch (IDM_PDU_OPCODE(rx_pdu)) {
904 	case ISCSI_OP_SCSI_CMD:
905 		ASSERT(0); /* Shouldn't happen */
906 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
907 		break;
908 	case ISCSI_OP_SNACK_CMD:
909 		/*
910 		 * We'll need to handle this when we support ERL1/2.  For
911 		 * now we treat it as a protocol error.
912 		 */
913 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
914 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
915 		break;
916 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
917 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
918 			iscsit_set_cmdsn(ict, rx_pdu);
919 			iscsit_op_scsi_task_mgmt(ict, rx_pdu);
920 		}
921 		break;
922 	case ISCSI_OP_NOOP_OUT:
923 	case ISCSI_OP_LOGIN_CMD:
924 	case ISCSI_OP_TEXT_CMD:
925 	case ISCSI_OP_LOGOUT_CMD:
926 		/*
927 		 * If/when we switch to userland processing these PDU's
928 		 * will be handled by iscsitd.
929 		 */
930 		iscsit_deferred_dispatch(rx_pdu);
931 		break;
932 	default:
933 		/* Protocol error */
934 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
935 		idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
936 		break;
937 	}
938 }
939 
940 /*ARGSUSED*/
941 void
942 iscsit_rx_pdu_error(idm_conn_t *ic, idm_pdu_t *rx_pdu, idm_status_t status)
943 {
944 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
945 }
946 
947 void
948 iscsit_task_aborted(idm_task_t *idt, idm_status_t status)
949 {
950 	iscsit_task_t *itask = idt->idt_private;
951 
952 	switch (status) {
953 	case IDM_STATUS_SUSPENDED:
954 		break;
955 	case IDM_STATUS_ABORTED:
956 		mutex_enter(&itask->it_mutex);
957 		itask->it_aborted = B_TRUE;
958 		/*
959 		 * We rely on the fact that STMF tracks outstanding
960 		 * buffer transfers and will free all of our buffers
961 		 * before freeing the task so we don't need to
962 		 * explicitly free the buffers from iscsit/idm
963 		 */
964 		if (itask->it_stmf_abort) {
965 			mutex_exit(&itask->it_mutex);
966 			/*
967 			 * Task is no longer active
968 			 */
969 			iscsit_task_done(itask);
970 
971 			/*
972 			 * STMF has already asked for this task to be aborted
973 			 *
974 			 * STMF specification is wrong... says to return
975 			 * STMF_ABORTED, the code actually looks for
976 			 * STMF_ABORT_SUCCESS.
977 			 */
978 			stmf_task_lport_aborted(itask->it_stmf_task,
979 			    STMF_ABORT_SUCCESS, STMF_IOF_LPORT_DONE);
980 			return;
981 		} else {
982 			mutex_exit(&itask->it_mutex);
983 			/*
984 			 * Tell STMF to stop processing the task.
985 			 */
986 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
987 			    STMF_ABORTED, NULL);
988 			return;
989 		}
990 		/*NOTREACHED*/
991 	default:
992 		ASSERT(0);
993 	}
994 }
995 
996 /*ARGSUSED*/
997 idm_status_t
998 iscsit_client_notify(idm_conn_t *ic, idm_client_notify_t icn,
999     uintptr_t data)
1000 {
1001 	idm_status_t rc = IDM_STATUS_SUCCESS;
1002 
1003 	/*
1004 	 * IDM client notifications will never occur at interrupt level
1005 	 * since they are generated from the connection state machine which
1006 	 * running on taskq threads.
1007 	 *
1008 	 */
1009 	switch (icn) {
1010 	case CN_CONNECT_ACCEPT:
1011 		rc = iscsit_conn_accept(ic); /* No data */
1012 		break;
1013 	case CN_FFP_ENABLED:
1014 		rc = iscsit_ffp_enabled(ic); /* No data */
1015 		break;
1016 	case CN_FFP_DISABLED:
1017 		/*
1018 		 * Data indicates whether this was the result of an
1019 		 * explicit logout request.
1020 		 */
1021 		rc = iscsit_ffp_disabled(ic, (idm_ffp_disable_t)data);
1022 		break;
1023 	case CN_CONNECT_LOST:
1024 		rc = iscsit_conn_lost(ic);
1025 		break;
1026 	case CN_CONNECT_DESTROY:
1027 		rc = iscsit_conn_destroy(ic);
1028 		break;
1029 	case CN_LOGIN_FAIL:
1030 		/*
1031 		 * Force the login state machine to completion
1032 		 */
1033 		rc = iscsit_login_fail(ic);
1034 		break;
1035 	default:
1036 		rc = IDM_STATUS_REJECT;
1037 		break;
1038 	}
1039 
1040 	return (rc);
1041 }
1042 
1043 /*
1044  * iscsit_update_statsn is invoked for all the PDUs which have the StatSN
1045  * field in the header. The StatSN is incremented if the IDM_PDU_ADVANCE_STATSN
1046  * flag is set in the pdu flags field. The StatSN is connection-wide and is
1047  * protected by the mutex ict_statsn_mutex. For Data-In PDUs, if the flag
1048  * IDM_TASK_PHASECOLLAPSE_REQ is set, the status (phase-collapse) is also filled
1049  */
1050 void
1051 iscsit_update_statsn(idm_task_t *idm_task, idm_pdu_t *pdu)
1052 {
1053 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1054 	iscsit_conn_t *ict = (iscsit_conn_t *)pdu->isp_ic->ic_handle;
1055 	iscsit_task_t *itask = NULL;
1056 	scsi_task_t *task = NULL;
1057 
1058 	mutex_enter(&ict->ict_statsn_mutex);
1059 	rsp->statsn = htonl(ict->ict_statsn);
1060 	if (pdu->isp_flags & IDM_PDU_ADVANCE_STATSN)
1061 		ict->ict_statsn++;
1062 	mutex_exit(&ict->ict_statsn_mutex);
1063 
1064 	/*
1065 	 * The last SCSI Data PDU passed for a command may also contain the
1066 	 * status if the status indicates termination with no expections, i.e.
1067 	 * no sense data or response involved. If the command completes with
1068 	 * an error, then the response and sense data will be sent in a
1069 	 * separate iSCSI Response PDU.
1070 	 */
1071 	if ((idm_task) && (idm_task->idt_flags & IDM_TASK_PHASECOLLAPSE_REQ)) {
1072 		itask = idm_task->idt_private;
1073 		task = itask->it_stmf_task;
1074 
1075 		rsp->cmd_status = task->task_scsi_status;
1076 		rsp->flags	|= ISCSI_FLAG_DATA_STATUS;
1077 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1078 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1079 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1080 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1081 		}
1082 		rsp->residual_count = htonl(task->task_resid);
1083 
1084 		/*
1085 		 * Removing the task from the session task list
1086 		 * just before the status is sent in the last
1087 		 * Data PDU transfer
1088 		 */
1089 		iscsit_task_done(itask);
1090 	}
1091 }
1092 
1093 void
1094 iscsit_build_hdr(idm_task_t *idm_task, idm_pdu_t *pdu, uint8_t opcode)
1095 {
1096 	iscsit_task_t *itask = idm_task->idt_private;
1097 	iscsi_data_rsp_hdr_t *dh = (iscsi_data_rsp_hdr_t *)pdu->isp_hdr;
1098 
1099 	/*
1100 	 * We acquired iscsit_sess_t.ist_sn_mutex in iscsit_xfer_scsi_data
1101 	 */
1102 	ASSERT(MUTEX_HELD(&itask->it_ict->ict_sess->ist_sn_mutex));
1103 	/*
1104 	 * On incoming data, the target transfer tag and Lun is only
1105 	 * provided by the target if the A bit is set, Since the target
1106 	 * does not currently support Error Recovery Level 1, the A
1107 	 * bit is never set.
1108 	 */
1109 	dh->opcode = opcode;
1110 	dh->itt = itask->it_itt;
1111 	dh->ttt = ((opcode & ISCSI_OPCODE_MASK) == ISCSI_OP_SCSI_DATA_RSP) ?
1112 	    ISCSI_RSVD_TASK_TAG : itask->it_ttt;
1113 
1114 	dh->expcmdsn = htonl(itask->it_ict->ict_sess->ist_expcmdsn);
1115 	dh->maxcmdsn = htonl(itask->it_ict->ict_sess->ist_maxcmdsn);
1116 
1117 	/*
1118 	 * IDM must set:
1119 	 *
1120 	 * data.flags and rtt.flags
1121 	 * data.dlength
1122 	 * data.datasn
1123 	 * data.offset
1124 	 * statsn, residual_count and cmd_status (for phase collapse)
1125 	 * rtt.rttsn
1126 	 * rtt.data_offset
1127 	 * rtt.data_length
1128 	 */
1129 }
1130 
1131 void
1132 iscsit_keepalive(idm_conn_t *ic)
1133 {
1134 	idm_pdu_t		*nop_in_pdu;
1135 	iscsi_nop_in_hdr_t	*nop_in;
1136 	iscsit_conn_t		*ict = ic->ic_handle;
1137 
1138 	/*
1139 	 * IDM noticed the connection has been idle for too long so it's
1140 	 * time to provoke some activity.  Build and transmit an iSCSI
1141 	 * nop-in PDU -- when the initiator responds it will be counted
1142 	 * as "activity" and keep the connection alive.
1143 	 *
1144 	 * We don't actually care about the response here at the iscsit level
1145 	 * so we will just throw it away without looking at it when it arrives.
1146 	 */
1147 	nop_in_pdu = idm_pdu_alloc(sizeof (*nop_in), 0);
1148 	idm_pdu_init(nop_in_pdu, ic, NULL, NULL);
1149 	nop_in = (iscsi_nop_in_hdr_t *)nop_in_pdu->isp_hdr;
1150 	bzero(nop_in, sizeof (*nop_in));
1151 	nop_in->opcode = ISCSI_OP_NOOP_IN;
1152 	nop_in->flags = ISCSI_FLAG_FINAL;
1153 	nop_in->itt = ISCSI_RSVD_TASK_TAG;
1154 	/*
1155 	 * When the target sends a NOP-In as a Ping, the target transfer tag
1156 	 * is set to a valid (not reserved) value and the initiator task tag
1157 	 * is set to ISCSI_RSVD_TASK_TAG (0xffffffff). In this case the StatSN
1158 	 * will always contain the next sequence number but the StatSN for the
1159 	 * connection is not advanced after this PDU is sent.
1160 	 */
1161 	nop_in_pdu->isp_flags |= IDM_PDU_SET_STATSN;
1162 	/*
1163 	 * This works because we don't currently allocate ttt's anywhere else
1164 	 * in iscsit so as long as we stay out of IDM's range we are safe.
1165 	 * If we need to allocate ttt's for other PDU's in the future this will
1166 	 * need to be improved.
1167 	 */
1168 	mutex_enter(&ict->ict_mutex);
1169 	nop_in->ttt = ict->ict_keepalive_ttt;
1170 	ict->ict_keepalive_ttt++;
1171 	if (ict->ict_keepalive_ttt == ISCSI_RSVD_TASK_TAG)
1172 		ict->ict_keepalive_ttt = IDM_TASKIDS_MAX;
1173 	mutex_exit(&ict->ict_mutex);
1174 
1175 	iscsit_pdu_tx(nop_in_pdu);
1176 }
1177 
1178 static idm_status_t
1179 iscsit_conn_accept(idm_conn_t *ic)
1180 {
1181 	iscsit_conn_t *ict;
1182 
1183 	/*
1184 	 * We need to get a global hold here to ensure that the service
1185 	 * doesn't get shutdown prior to establishing a session. This
1186 	 * gets released in iscsit_conn_destroy().
1187 	 */
1188 	mutex_enter(&iscsit_global.global_state_mutex);
1189 	if (iscsit_global.global_svc_state != ISE_ENABLED) {
1190 		mutex_exit(&iscsit_global.global_state_mutex);
1191 		return (IDM_STATUS_FAIL);
1192 	}
1193 	iscsit_global_hold();
1194 	mutex_exit(&iscsit_global.global_state_mutex);
1195 
1196 	/*
1197 	 * Allocate an associated iscsit structure to represent this
1198 	 * connection.  We shouldn't really create a session until we
1199 	 * get the first login PDU.
1200 	 */
1201 	ict = kmem_zalloc(sizeof (*ict), KM_SLEEP);
1202 
1203 	ict->ict_ic = ic;
1204 	ict->ict_statsn = 1;
1205 	ict->ict_keepalive_ttt = IDM_TASKIDS_MAX; /* Avoid IDM TT range */
1206 	ic->ic_handle = ict;
1207 	mutex_init(&ict->ict_mutex, NULL, MUTEX_DRIVER, NULL);
1208 	mutex_init(&ict->ict_statsn_mutex, NULL, MUTEX_DRIVER, NULL);
1209 	idm_refcnt_init(&ict->ict_refcnt, ict);
1210 
1211 	/*
1212 	 * Initialize login state machine
1213 	 */
1214 	if (iscsit_login_sm_init(ict) != IDM_STATUS_SUCCESS) {
1215 		iscsit_global_rele();
1216 		/*
1217 		 * Cleanup the ict after idm notifies us about this failure
1218 		 */
1219 		return (IDM_STATUS_FAIL);
1220 	}
1221 
1222 	return (IDM_STATUS_SUCCESS);
1223 }
1224 
1225 idm_status_t
1226 iscsit_conn_reinstate(iscsit_conn_t *reinstate_ict, iscsit_conn_t *new_ict)
1227 {
1228 	idm_status_t	result;
1229 
1230 	/*
1231 	 * Note in new connection state that this connection is
1232 	 * reinstating an existing connection.
1233 	 */
1234 	new_ict->ict_reinstating = B_TRUE;
1235 	new_ict->ict_reinstate_conn = reinstate_ict;
1236 	new_ict->ict_statsn = reinstate_ict->ict_statsn;
1237 
1238 	/*
1239 	 * Now generate connection state machine event to existing connection
1240 	 * so that it starts the cleanup process.
1241 	 */
1242 	result = idm_conn_reinstate_event(reinstate_ict->ict_ic,
1243 	    new_ict->ict_ic);
1244 
1245 	return (result);
1246 }
1247 
1248 void
1249 iscsit_conn_hold(iscsit_conn_t *ict)
1250 {
1251 	idm_refcnt_hold(&ict->ict_refcnt);
1252 }
1253 
1254 void
1255 iscsit_conn_rele(iscsit_conn_t *ict)
1256 {
1257 	idm_refcnt_rele(&ict->ict_refcnt);
1258 }
1259 
1260 void
1261 iscsit_conn_dispatch_hold(iscsit_conn_t *ict)
1262 {
1263 	idm_refcnt_hold(&ict->ict_dispatch_refcnt);
1264 }
1265 
1266 void
1267 iscsit_conn_dispatch_rele(iscsit_conn_t *ict)
1268 {
1269 	idm_refcnt_rele(&ict->ict_dispatch_refcnt);
1270 }
1271 
1272 static idm_status_t
1273 iscsit_login_fail(idm_conn_t *ic)
1274 {
1275 	iscsit_conn_t *ict = ic->ic_handle;
1276 
1277 	/* Generate login state machine event */
1278 	iscsit_login_sm_event(ict, ILE_LOGIN_CONN_ERROR, NULL);
1279 
1280 	return (IDM_STATUS_SUCCESS);
1281 }
1282 
1283 static idm_status_t
1284 iscsit_ffp_enabled(idm_conn_t *ic)
1285 {
1286 	iscsit_conn_t *ict = ic->ic_handle;
1287 
1288 	/* Generate session state machine event */
1289 	iscsit_sess_sm_event(ict->ict_sess, SE_CONN_LOGGED_IN, ict);
1290 
1291 	return (IDM_STATUS_SUCCESS);
1292 }
1293 
1294 static idm_status_t
1295 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class)
1296 {
1297 	iscsit_conn_t *ict = ic->ic_handle;
1298 
1299 	/* Generate session state machine event */
1300 	switch (disable_class) {
1301 	case FD_CONN_FAIL:
1302 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_FAIL, ict);
1303 		break;
1304 	case FD_CONN_LOGOUT:
1305 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_DISABLE, ict);
1306 		break;
1307 	case FD_SESS_LOGOUT:
1308 		iscsit_sess_sm_event(ict->ict_sess, SE_SESSION_CLOSE, ict);
1309 		break;
1310 	default:
1311 		ASSERT(0);
1312 	}
1313 
1314 	return (IDM_STATUS_SUCCESS);
1315 }
1316 
1317 static idm_status_t
1318 iscsit_conn_lost(idm_conn_t *ic)
1319 {
1320 	iscsit_conn_t	*ict	= ic->ic_handle;
1321 	iscsit_sess_t	*ist	= ict->ict_sess;
1322 	iscsit_cbuf_t	*cbuf;
1323 	idm_pdu_t	*rx_pdu;
1324 	int i;
1325 
1326 	mutex_enter(&ict->ict_mutex);
1327 	ict->ict_lost = B_TRUE;
1328 	mutex_exit(&ict->ict_mutex);
1329 	/*
1330 	 * scrub the staging queue for all PDUs on this connection
1331 	 */
1332 	if (ist != NULL) {
1333 		mutex_enter(&ist->ist_sn_mutex);
1334 		for (cbuf = ist->ist_rxpdu_queue, i = 0;
1335 		    ((cbuf->cb_num_elems > 0) && (i < ISCSIT_RXPDU_QUEUE_LEN));
1336 		    i++) {
1337 			if (((rx_pdu = cbuf->cb_buffer[i]) != NULL) &&
1338 			    (rx_pdu->isp_ic == ic)) {
1339 				/* conn is lost, drop the pdu */
1340 				DTRACE_PROBE3(scrubbing__staging__queue,
1341 				    iscsit_sess_t *, ist, idm_conn_t *, ic,
1342 				    idm_pdu_t *, rx_pdu);
1343 				idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
1344 				cbuf->cb_buffer[i] = NULL;
1345 				cbuf->cb_num_elems--;
1346 				iscsit_conn_dispatch_rele(ict);
1347 			}
1348 		}
1349 		mutex_exit(&ist->ist_sn_mutex);
1350 	}
1351 	/*
1352 	 * Make sure there aren't any PDU's transitioning from the receive
1353 	 * handler to the dispatch taskq.
1354 	 */
1355 	idm_refcnt_wait_ref(&ict->ict_dispatch_refcnt);
1356 
1357 	return (IDM_STATUS_SUCCESS);
1358 }
1359 
1360 static idm_status_t
1361 iscsit_conn_destroy(idm_conn_t *ic)
1362 {
1363 	iscsit_conn_t *ict = ic->ic_handle;
1364 
1365 	mutex_enter(&ict->ict_mutex);
1366 	ict->ict_destroyed = B_TRUE;
1367 	mutex_exit(&ict->ict_mutex);
1368 
1369 	/* Generate session state machine event */
1370 	if (ict->ict_sess != NULL) {
1371 		/*
1372 		 * Session state machine will call iscsit_conn_destroy_done()
1373 		 * when it has removed references to this connection.
1374 		 */
1375 		iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FAIL, ict);
1376 	}
1377 
1378 	idm_refcnt_wait_ref(&ict->ict_refcnt);
1379 	/*
1380 	 * The session state machine does not need to post
1381 	 * events to IDM any longer, so it is safe to set
1382 	 * the idm connection reference to NULL
1383 	 */
1384 	ict->ict_ic = NULL;
1385 
1386 	/* Reap the login state machine */
1387 	iscsit_login_sm_fini(ict);
1388 
1389 	/* Clean up any text command remnants */
1390 	iscsit_text_cmd_fini(ict);
1391 
1392 	mutex_destroy(&ict->ict_mutex);
1393 	idm_refcnt_destroy(&ict->ict_refcnt);
1394 	kmem_free(ict, sizeof (*ict));
1395 
1396 	iscsit_global_rele();
1397 
1398 	return (IDM_STATUS_SUCCESS);
1399 }
1400 
1401 void
1402 iscsit_conn_logout(iscsit_conn_t *ict)
1403 {
1404 	/*
1405 	 * If the iscsi connection is active, then
1406 	 * logout the IDM connection by sending a
1407 	 * CE_LOGOUT_SESSION_SUCCESS, else, no action
1408 	 * needs to be taken because the connection
1409 	 * is already in the teardown process.
1410 	 */
1411 	mutex_enter(&ict->ict_mutex);
1412 	if (ict->ict_lost == B_FALSE && ict->ict_destroyed == B_FALSE) {
1413 		idm_conn_event(ict->ict_ic, CE_LOGOUT_SESSION_SUCCESS, NULL);
1414 	}
1415 	mutex_exit(&ict->ict_mutex);
1416 }
1417 
1418 /*
1419  * STMF-related functions
1420  *
1421  * iSCSI to STMF mapping
1422  *
1423  * Session == ?
1424  * Connection == bound to local port but not itself a local port
1425  * Target
1426  * Target portal (group?) == local port (really but we're not going to do this)
1427  *	iscsit needs to map connections to local ports (whatever we decide
1428  * 	they are)
1429  * Target == ?
1430  */
1431 
1432 /*ARGSUSED*/
1433 static stmf_data_buf_t *
1434 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
1435     uint32_t flags)
1436 {
1437 	iscsit_task_t *itask = task->task_port_private;
1438 	idm_buf_t *idm_buffer;
1439 	iscsit_buf_t	*ibuf;
1440 	stmf_data_buf_t *result;
1441 	uint32_t	bsize;
1442 
1443 	/*
1444 	 * If the requested size is larger than MaxBurstLength and the
1445 	 * given pminsize is also larger than MaxBurstLength, then the
1446 	 * allocation fails (dbuf = NULL) and pminsize is modified to
1447 	 * be equal to MaxBurstLength. stmf/sbd then should re-invoke
1448 	 * this function with the corrected values for transfer.
1449 	 */
1450 	ASSERT(pminsize);
1451 	if (size <= itask->it_ict->ict_op.op_max_burst_length) {
1452 		bsize = size;
1453 	} else if (*pminsize <= itask->it_ict->ict_op.op_max_burst_length) {
1454 		bsize = itask->it_ict->ict_op.op_max_burst_length;
1455 	} else {
1456 		*pminsize = itask->it_ict->ict_op.op_max_burst_length;
1457 		return (NULL);
1458 	}
1459 
1460 	/* Alloc buffer */
1461 	idm_buffer = idm_buf_alloc(itask->it_ict->ict_ic, NULL, bsize);
1462 	if (idm_buffer != NULL) {
1463 		result = stmf_alloc(STMF_STRUCT_DATA_BUF,
1464 		    sizeof (iscsit_buf_t), 0);
1465 		if (result != NULL) {
1466 			/* Fill in stmf_data_buf_t */
1467 			ibuf = result->db_port_private;
1468 			ibuf->ibuf_idm_buf = idm_buffer;
1469 			ibuf->ibuf_stmf_buf = result;
1470 			ibuf->ibuf_is_immed = B_FALSE;
1471 			result->db_flags = DB_DONT_CACHE;
1472 			result->db_buf_size = bsize;
1473 			result->db_data_size = bsize;
1474 			result->db_sglist_length = 1;
1475 			result->db_sglist[0].seg_addr = idm_buffer->idb_buf;
1476 			result->db_sglist[0].seg_length =
1477 			    idm_buffer->idb_buflen;
1478 			return (result);
1479 		}
1480 
1481 		/* Couldn't get the stmf_data_buf_t so free the buffer */
1482 		idm_buf_free(idm_buffer);
1483 	}
1484 
1485 	return (NULL);
1486 }
1487 
1488 /*ARGSUSED*/
1489 static void
1490 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf)
1491 {
1492 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1493 
1494 	if (ibuf->ibuf_is_immed) {
1495 		/*
1496 		 * The iscsit_buf_t structure itself will be freed with its
1497 		 * associated task.  Here we just need to free the PDU that
1498 		 * held the immediate data.
1499 		 */
1500 		idm_pdu_complete(ibuf->ibuf_immed_data_pdu, IDM_STATUS_SUCCESS);
1501 		ibuf->ibuf_immed_data_pdu = 0;
1502 	} else {
1503 		idm_buf_free(ibuf->ibuf_idm_buf);
1504 		stmf_free(dbuf);
1505 	}
1506 }
1507 
1508 /*ARGSUSED*/
1509 stmf_status_t
1510 iscsit_xfer_scsi_data(scsi_task_t *task, stmf_data_buf_t *dbuf,
1511     uint32_t ioflags)
1512 {
1513 	iscsit_task_t *iscsit_task = task->task_port_private;
1514 	iscsit_sess_t *ict_sess = iscsit_task->it_ict->ict_sess;
1515 	iscsit_buf_t *ibuf = dbuf->db_port_private;
1516 	int idm_rc;
1517 
1518 	/*
1519 	 * If we are aborting then we can ignore this request
1520 	 */
1521 	if (iscsit_task->it_stmf_abort) {
1522 		return (STMF_SUCCESS);
1523 	}
1524 
1525 	/*
1526 	 * If it's not immediate data then start the transfer
1527 	 */
1528 	ASSERT(ibuf->ibuf_is_immed == B_FALSE);
1529 	if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
1530 		/*
1531 		 * The DB_SEND_STATUS_GOOD flag in the STMF data buffer allows
1532 		 * the port provider to phase-collapse, i.e. send the status
1533 		 * along with the final data PDU for the command. The port
1534 		 * provider passes this request to the transport layer by
1535 		 * setting a flag IDM_TASK_PHASECOLLAPSE_REQ in the task.
1536 		 */
1537 		if (dbuf->db_flags & DB_SEND_STATUS_GOOD)
1538 			iscsit_task->it_idm_task->idt_flags |=
1539 			    IDM_TASK_PHASECOLLAPSE_REQ;
1540 		/*
1541 		 * IDM will call iscsit_build_hdr so lock now to serialize
1542 		 * access to the SN values.  We need to lock here to enforce
1543 		 * lock ordering
1544 		 */
1545 		mutex_enter(&ict_sess->ist_sn_mutex);
1546 		idm_rc = idm_buf_tx_to_ini(iscsit_task->it_idm_task,
1547 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1548 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1549 		mutex_exit(&ict_sess->ist_sn_mutex);
1550 
1551 		return (iscsit_idm_to_stmf(idm_rc));
1552 	} else if (dbuf->db_flags & DB_DIRECTION_FROM_RPORT) {
1553 		/* Grab the SN lock (see comment above) */
1554 		mutex_enter(&ict_sess->ist_sn_mutex);
1555 		idm_rc = idm_buf_rx_from_ini(iscsit_task->it_idm_task,
1556 		    ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1557 		    dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1558 		mutex_exit(&ict_sess->ist_sn_mutex);
1559 
1560 		return (iscsit_idm_to_stmf(idm_rc));
1561 	}
1562 
1563 	/* What are we supposed to do if there is no direction? */
1564 	return (STMF_INVALID_ARG);
1565 }
1566 
1567 static void
1568 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status)
1569 {
1570 	iscsit_task_t *itask = idb->idb_task_binding->idt_private;
1571 	stmf_data_buf_t *dbuf = idb->idb_cb_arg;
1572 
1573 	dbuf->db_xfer_status = iscsit_idm_to_stmf(status);
1574 
1575 	/*
1576 	 * If the task has been aborted then we don't need to call STMF
1577 	 */
1578 	if (itask->it_stmf_abort) {
1579 		return;
1580 	}
1581 
1582 	/*
1583 	 * For ISCSI over TCP (not iSER), the last SCSI Data PDU passed
1584 	 * for a successful command contains the status as requested by
1585 	 * by COMSTAR (via the DB_SEND_STATUS_GOOD flag). But the iSER
1586 	 * transport does not support phase-collapse. So pretend we are
1587 	 * COMSTAR and send the status in a separate PDU now.
1588 	 */
1589 	if (idb->idb_task_binding->idt_flags & IDM_TASK_PHASECOLLAPSE_SUCCESS) {
1590 		/*
1591 		 * Mark task complete and notify COMSTAR
1592 		 * that the status has been sent.
1593 		 */
1594 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1595 		stmf_send_status_done(itask->it_stmf_task,
1596 		    iscsit_idm_to_stmf(status), STMF_IOF_LPORT_DONE);
1597 	} else if ((dbuf->db_flags & DB_SEND_STATUS_GOOD) &&
1598 	    status == IDM_STATUS_SUCCESS) {
1599 
1600 		/*
1601 		 * The iscsi target port provider - for iSER, emulates the
1602 		 * DB_SEND_STATUS_GOOD optimization if requested by STMF;
1603 		 * it sends the status in a separate PDU after the data
1604 		 * transfer. In this case the port provider should first
1605 		 * call stmf_data_xfer_done() to mark the transfer complete
1606 		 * and then send the status. Although STMF will free the
1607 		 * buffer at the time the task is freed, even if the transfer
1608 		 * is not marked complete, this behavior makes statistics
1609 		 * gathering and task state tracking more difficult than it
1610 		 * needs to be.
1611 		 */
1612 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1613 		if (iscsit_send_scsi_status(itask->it_stmf_task, 0)
1614 		    != STMF_SUCCESS) {
1615 			stmf_send_status_done(itask->it_stmf_task,
1616 			    STMF_FAILURE, STMF_IOF_LPORT_DONE);
1617 		}
1618 	} else {
1619 		stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1620 		/* don't touch dbuf after stmf_data_xfer_done */
1621 	}
1622 }
1623 
1624 
1625 /*ARGSUSED*/
1626 stmf_status_t
1627 iscsit_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
1628 {
1629 	iscsit_task_t *itask = task->task_port_private;
1630 	iscsi_scsi_rsp_hdr_t *rsp;
1631 	idm_pdu_t *pdu;
1632 	int resp_datalen;
1633 
1634 	/*
1635 	 * If this task is aborted then we don't need to respond.
1636 	 */
1637 	if (itask->it_stmf_abort) {
1638 		return (STMF_SUCCESS);
1639 	}
1640 
1641 	/*
1642 	 * If this is a task management status, handle it elsewhere.
1643 	 */
1644 	if (task->task_mgmt_function != TM_NONE) {
1645 		/*
1646 		 * Don't wait for the PDU completion to tell STMF
1647 		 * the task is done -- it doesn't really matter and
1648 		 * it makes life complicated if STMF later asks us to
1649 		 * abort the request and we don't know whether the
1650 		 * status has been sent or not.
1651 		 */
1652 		itask->it_tm_responded = B_TRUE;
1653 		iscsit_send_task_mgmt_resp(itask->it_tm_pdu,
1654 		    (task->task_completion_status == STMF_SUCCESS) ?
1655 		    SCSI_TCP_TM_RESP_COMPLETE : SCSI_TCP_TM_RESP_FUNC_NOT_SUPP);
1656 		stmf_send_status_done(task, STMF_SUCCESS,
1657 		    STMF_IOF_LPORT_DONE);
1658 		return (STMF_SUCCESS);
1659 	}
1660 
1661 	/*
1662 	 * Remove the task from the session task list
1663 	 */
1664 	iscsit_task_done(itask);
1665 
1666 	/*
1667 	 * Send status
1668 	 */
1669 	mutex_enter(&itask->it_idm_task->idt_mutex);
1670 	if ((itask->it_idm_task->idt_state == TASK_ACTIVE) &&
1671 	    (task->task_completion_status == STMF_SUCCESS) &&
1672 	    (task->task_sense_length == 0) &&
1673 	    (task->task_resid == 0)) {
1674 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1675 		/* PDU callback releases task hold */
1676 		idm_task_hold(itask->it_idm_task);
1677 		mutex_exit(&itask->it_idm_task->idt_mutex);
1678 		/*
1679 		 * Fast path.  Cached status PDU's are already
1680 		 * initialized.  We just need to fill in
1681 		 * connection and task information. StatSN is
1682 		 * incremented by 1 for every status sent a
1683 		 * connection.
1684 		 */
1685 		pdu = kmem_cache_alloc(iscsit_status_pdu_cache, KM_SLEEP);
1686 		pdu->isp_ic = itask->it_ict->ict_ic;
1687 		pdu->isp_private = itask;
1688 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1689 
1690 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1691 		rsp->itt = itask->it_itt;
1692 		/*
1693 		 * ExpDataSN is the number of R2T and Data-In (read)
1694 		 * PDUs the target has sent for the SCSI command.
1695 		 *
1696 		 * Since there is no support for bidirectional transfer
1697 		 * yet, either idt_exp_datasn or idt_exp_rttsn, but not
1698 		 * both is valid at any time
1699 		 */
1700 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1701 		    htonl(itask->it_idm_task->idt_exp_datasn):
1702 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1703 		rsp->cmd_status = task->task_scsi_status;
1704 		iscsit_pdu_tx(pdu);
1705 		return (STMF_SUCCESS);
1706 	} else {
1707 		if (itask->it_idm_task->idt_state != TASK_ACTIVE) {
1708 			mutex_exit(&itask->it_idm_task->idt_mutex);
1709 			return (STMF_FAILURE);
1710 		}
1711 		itask->it_idm_task->idt_state = TASK_COMPLETE;
1712 		/* PDU callback releases task hold */
1713 		idm_task_hold(itask->it_idm_task);
1714 		mutex_exit(&itask->it_idm_task->idt_mutex);
1715 
1716 		resp_datalen = (task->task_sense_length == 0) ? 0 :
1717 		    (task->task_sense_length + sizeof (uint16_t));
1718 
1719 		pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
1720 		idm_pdu_init(pdu, itask->it_ict->ict_ic, itask,
1721 		    iscsit_send_status_done);
1722 		pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1723 
1724 		rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1725 		bzero(rsp, sizeof (*rsp));
1726 		rsp->opcode = ISCSI_OP_SCSI_RSP;
1727 
1728 		rsp->flags = ISCSI_FLAG_FINAL;
1729 		if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1730 			rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1731 		} else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1732 			rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1733 		}
1734 
1735 		rsp->bi_residual_count = 0;
1736 		rsp->residual_count = htonl(task->task_resid);
1737 		rsp->itt = itask->it_itt;
1738 		rsp->response = ISCSI_STATUS_CMD_COMPLETED;
1739 		rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1740 		    htonl(itask->it_idm_task->idt_exp_datasn):
1741 		    htonl(itask->it_idm_task->idt_exp_rttsn);
1742 		rsp->cmd_status = task->task_scsi_status;
1743 		if (task->task_sense_length != 0) {
1744 			/*
1745 			 * Add a byte to provide the sense length in
1746 			 * the response
1747 			 */
1748 			*(uint16_t *)((void *)pdu->isp_data) =
1749 			    htons(task->task_sense_length);
1750 			bcopy(task->task_sense_data,
1751 			    (uint8_t *)pdu->isp_data +
1752 			    sizeof (uint16_t),
1753 			    task->task_sense_length);
1754 			hton24(rsp->dlength, resp_datalen);
1755 		}
1756 
1757 		DTRACE_PROBE5(iscsi__scsi__response,
1758 		    iscsit_conn_t *, itask->it_ict,
1759 		    uint8_t, rsp->response,
1760 		    uint8_t, rsp->cmd_status,
1761 		    idm_pdu_t *, pdu,
1762 		    scsi_task_t *, task);
1763 
1764 		iscsit_pdu_tx(pdu);
1765 
1766 		return (STMF_SUCCESS);
1767 	}
1768 }
1769 
1770 /*ARGSUSED*/
1771 static void
1772 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status)
1773 {
1774 	iscsit_task_t	*itask;
1775 	boolean_t	aborted;
1776 
1777 	itask = pdu->isp_private;
1778 	aborted = itask->it_stmf_abort;
1779 
1780 	/*
1781 	 * After releasing the hold the task may be freed at any time so
1782 	 * don't touch it.
1783 	 */
1784 	idm_task_rele(itask->it_idm_task);
1785 	if (!aborted) {
1786 		stmf_send_status_done(itask->it_stmf_task,
1787 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1788 	}
1789 	kmem_cache_free(iscsit_status_pdu_cache, pdu);
1790 }
1791 
1792 /*ARGSUSED*/
1793 static void
1794 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status)
1795 {
1796 	iscsit_task_t	 *itask;
1797 	boolean_t	aborted;
1798 
1799 	itask = pdu->isp_private;
1800 	aborted = itask->it_stmf_abort;
1801 
1802 	/*
1803 	 * After releasing the hold the task may be freed at any time so
1804 	 * don't touch it.
1805 	 */
1806 	idm_task_rele(itask->it_idm_task);
1807 	if (!aborted) {
1808 		stmf_send_status_done(itask->it_stmf_task,
1809 		    iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1810 	}
1811 	idm_pdu_free(pdu);
1812 }
1813 
1814 
1815 void
1816 iscsit_lport_task_free(scsi_task_t *task)
1817 {
1818 	iscsit_task_t *itask = task->task_port_private;
1819 
1820 	/* We only call idm_task_start for regular tasks, not task management */
1821 	if (task->task_mgmt_function == TM_NONE) {
1822 		idm_task_done(itask->it_idm_task);
1823 		iscsit_task_free(itask);
1824 		return;
1825 	} else {
1826 		iscsit_tm_task_free(itask);
1827 	}
1828 }
1829 
1830 /*ARGSUSED*/
1831 stmf_status_t
1832 iscsit_abort(stmf_local_port_t *lport, int abort_cmd, void *arg, uint32_t flags)
1833 {
1834 	scsi_task_t	*st = (scsi_task_t *)arg;
1835 	iscsit_task_t	*iscsit_task;
1836 	idm_task_t	*idt;
1837 
1838 	/*
1839 	 * If this is a task management request then there's really not much to
1840 	 * do.
1841 	 */
1842 	if (st->task_mgmt_function != TM_NONE) {
1843 		return (STMF_ABORT_SUCCESS);
1844 	}
1845 
1846 	/*
1847 	 * Regular task, start cleaning up
1848 	 */
1849 	iscsit_task = st->task_port_private;
1850 	idt = iscsit_task->it_idm_task;
1851 	mutex_enter(&iscsit_task->it_mutex);
1852 	iscsit_task->it_stmf_abort = B_TRUE;
1853 	if (iscsit_task->it_aborted) {
1854 		mutex_exit(&iscsit_task->it_mutex);
1855 		/*
1856 		 * Task is no longer active
1857 		 */
1858 		iscsit_task_done(iscsit_task);
1859 
1860 		/*
1861 		 * STMF specification is wrong... says to return
1862 		 * STMF_ABORTED, the code actually looks for
1863 		 * STMF_ABORT_SUCCESS.
1864 		 */
1865 		return (STMF_ABORT_SUCCESS);
1866 	} else {
1867 		mutex_exit(&iscsit_task->it_mutex);
1868 		/*
1869 		 * Call IDM to abort the task.  Due to a variety of
1870 		 * circumstances the task may already be in the process of
1871 		 * aborting.
1872 		 * We'll let IDM worry about rationalizing all that except
1873 		 * for one particular instance.  If the state of the task
1874 		 * is TASK_COMPLETE, we need to indicate to the framework
1875 		 * that we are in fact done.  This typically happens with
1876 		 * framework-initiated task management type requests
1877 		 * (e.g. abort task).
1878 		 */
1879 		if (idt->idt_state == TASK_COMPLETE) {
1880 			idm_refcnt_wait_ref(&idt->idt_refcnt);
1881 			return (STMF_ABORT_SUCCESS);
1882 		} else {
1883 			idm_task_abort(idt->idt_ic, idt, AT_TASK_MGMT_ABORT);
1884 			return (STMF_SUCCESS);
1885 		}
1886 	}
1887 
1888 	/*NOTREACHED*/
1889 }
1890 
1891 /*ARGSUSED*/
1892 void
1893 iscsit_ctl(stmf_local_port_t *lport, int cmd, void *arg)
1894 {
1895 	iscsit_tgt_t		*iscsit_tgt;
1896 
1897 	ASSERT((cmd == STMF_CMD_LPORT_ONLINE) ||
1898 	    (cmd == STMF_ACK_LPORT_ONLINE_COMPLETE) ||
1899 	    (cmd == STMF_CMD_LPORT_OFFLINE) ||
1900 	    (cmd == STMF_ACK_LPORT_OFFLINE_COMPLETE));
1901 
1902 	iscsit_tgt = (iscsit_tgt_t *)lport->lport_port_private;
1903 
1904 	switch (cmd) {
1905 	case STMF_CMD_LPORT_ONLINE:
1906 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_REQ);
1907 		break;
1908 	case STMF_CMD_LPORT_OFFLINE:
1909 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_REQ);
1910 		break;
1911 	case STMF_ACK_LPORT_ONLINE_COMPLETE:
1912 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_COMPLETE_ACK);
1913 		break;
1914 	case STMF_ACK_LPORT_OFFLINE_COMPLETE:
1915 		iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_COMPLETE_ACK);
1916 		break;
1917 
1918 	default:
1919 		break;
1920 	}
1921 }
1922 
1923 static stmf_status_t
1924 iscsit_idm_to_stmf(idm_status_t idmrc)
1925 {
1926 	switch (idmrc) {
1927 	case IDM_STATUS_SUCCESS:
1928 		return (STMF_SUCCESS);
1929 	default:
1930 		return (STMF_FAILURE);
1931 	}
1932 	/*NOTREACHED*/
1933 }
1934 
1935 void
1936 iscsit_op_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1937 {
1938 	iscsit_conn_t		*ict = ic->ic_handle;
1939 
1940 	if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
1941 		iscsit_post_scsi_cmd(ic, rx_pdu);
1942 	}
1943 	iscsit_process_pdu_in_queue(ict->ict_sess);
1944 }
1945 
1946 /*
1947  * ISCSI protocol
1948  */
1949 
1950 void
1951 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1952 {
1953 	iscsit_conn_t		*ict;
1954 	iscsit_task_t		*itask;
1955 	scsi_task_t		*task;
1956 	iscsit_buf_t		*ibuf;
1957 	iscsi_scsi_cmd_hdr_t	*iscsi_scsi =
1958 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
1959 	iscsi_addl_hdr_t	*ahs_hdr;
1960 	uint16_t		addl_cdb_len = 0;
1961 
1962 	ict = ic->ic_handle;
1963 
1964 	itask = iscsit_task_alloc(ict);
1965 	if (itask == NULL) {
1966 		/* Finish processing request */
1967 		iscsit_set_cmdsn(ict, rx_pdu);
1968 
1969 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
1970 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
1971 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
1972 		return;
1973 	}
1974 
1975 	/*
1976 	 * Note CmdSN and ITT in task.  IDM will have already validated this
1977 	 * request against the connection state so we don't need to check
1978 	 * that (the connection may have changed state in the meantime but
1979 	 * we will catch that when we try to send a response)
1980 	 */
1981 	itask->it_cmdsn = ntohl(iscsi_scsi->cmdsn);
1982 	itask->it_itt = iscsi_scsi->itt;
1983 
1984 	/*
1985 	 * Check for extended CDB AHS
1986 	 */
1987 	if (iscsi_scsi->hlength > 0) {
1988 		ahs_hdr = (iscsi_addl_hdr_t *)iscsi_scsi;
1989 		addl_cdb_len = ((ahs_hdr->ahs_hlen_hi << 8) |
1990 		    ahs_hdr->ahs_hlen_lo) - 1; /* Adjust for reserved byte */
1991 		if (((addl_cdb_len + 4) / sizeof (uint32_t)) >
1992 		    iscsi_scsi->hlength) {
1993 			/* Mangled header info, drop it */
1994 			idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
1995 			return;
1996 		}
1997 	}
1998 
1999 	ict = rx_pdu->isp_ic->ic_handle; /* IDM client private */
2000 
2001 	/*
2002 	 * Add task to session list.  This function will also check to
2003 	 * ensure that the task does not already exist.
2004 	 */
2005 	if (iscsit_task_start(itask) != IDM_STATUS_SUCCESS) {
2006 		/*
2007 		 * Task exists, free all resources and reject.  Don't
2008 		 * update expcmdsn in this case because RFC 3720 says
2009 		 * "The CmdSN of the rejected command PDU (if it is a
2010 		 * non-immediate command) MUST NOT be considered received
2011 		 * by the target (i.e., a command sequence gap must be
2012 		 * assumed for the CmdSN), even though the CmdSN of the
2013 		 * rejected command PDU may be reliably ascertained.  Upon
2014 		 * receiving the Reject, the initiator MUST plug the CmdSN
2015 		 * gap in order to continue to use the session.  The gap
2016 		 * may be plugged either by transmitting a command PDU
2017 		 * with the same CmdSN, or by aborting the task (see section
2018 		 * 6.9 on how an abort may plug a CmdSN gap)." (Section 6.3)
2019 		 */
2020 		iscsit_task_free(itask);
2021 		iscsit_send_reject(ict, rx_pdu, ISCSI_REJECT_TASK_IN_PROGRESS);
2022 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2023 		return;
2024 	}
2025 
2026 	/* Update sequence numbers */
2027 	iscsit_set_cmdsn(ict, rx_pdu);
2028 
2029 	/*
2030 	 * Allocate STMF task
2031 	 */
2032 	itask->it_stmf_task = stmf_task_alloc(
2033 	    itask->it_ict->ict_sess->ist_lport,
2034 	    itask->it_ict->ict_sess->ist_stmf_sess, iscsi_scsi->lun,
2035 	    16 + addl_cdb_len, 0);
2036 	if (itask->it_stmf_task == NULL) {
2037 		/*
2038 		 * Either stmf really couldn't get memory for a task or,
2039 		 * more likely, the LU is currently in reset.  Either way
2040 		 * we have no choice but to fail the request.
2041 		 */
2042 		iscsit_task_done(itask);
2043 		iscsit_task_free(itask);
2044 		iscsit_send_direct_scsi_resp(ict, rx_pdu,
2045 		    ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
2046 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2047 		return;
2048 	}
2049 
2050 	task = itask->it_stmf_task;
2051 	task->task_port_private = itask;
2052 
2053 	bcopy(iscsi_scsi->lun, task->task_lun_no, sizeof (task->task_lun_no));
2054 
2055 	/*
2056 	 * iSCSI and Comstar use the same values.  Should we rely on this
2057 	 * or translate them bit-wise?
2058 	 */
2059 
2060 	task->task_flags =
2061 	    (((iscsi_scsi->flags & ISCSI_FLAG_CMD_READ) ? TF_READ_DATA : 0) |
2062 	    ((iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ? TF_WRITE_DATA : 0) |
2063 	    ((rx_pdu->isp_datalen == 0) ? 0 : TF_INITIAL_BURST));
2064 
2065 	switch (iscsi_scsi->flags & ISCSI_FLAG_CMD_ATTR_MASK) {
2066 	case ISCSI_ATTR_UNTAGGED:
2067 		break;
2068 	case ISCSI_ATTR_SIMPLE:
2069 		task->task_additional_flags |= TF_ATTR_SIMPLE_QUEUE;
2070 		break;
2071 	case ISCSI_ATTR_ORDERED:
2072 		task->task_additional_flags |= TF_ATTR_ORDERED_QUEUE;
2073 		break;
2074 	case ISCSI_ATTR_HEAD_OF_QUEUE:
2075 		task->task_additional_flags |= TF_ATTR_HEAD_OF_QUEUE;
2076 		break;
2077 	case ISCSI_ATTR_ACA:
2078 		task->task_additional_flags |= TF_ATTR_ACA;
2079 		break;
2080 	default:
2081 		/* Protocol error but just take it, treat as untagged */
2082 		break;
2083 	}
2084 
2085 
2086 	task->task_additional_flags = 0;
2087 	task->task_priority = 0;
2088 	task->task_mgmt_function = TM_NONE;
2089 
2090 	/*
2091 	 * This "task_max_nbufs" doesn't map well to BIDI.  We probably need
2092 	 * parameter for each direction.  "MaxOutstandingR2T" may very well
2093 	 * be set to one which could prevent us from doing simultaneous
2094 	 * transfers in each direction.
2095 	 */
2096 	task->task_max_nbufs = (iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ?
2097 	    ict->ict_op.op_max_outstanding_r2t : STMF_BUFS_MAX;
2098 	task->task_cmd_seq_no = ntohl(iscsi_scsi->itt);
2099 	task->task_expected_xfer_length = ntohl(iscsi_scsi->data_length);
2100 
2101 	/* Copy CDB */
2102 	bcopy(iscsi_scsi->scb, task->task_cdb, 16);
2103 	if (addl_cdb_len > 0) {
2104 		bcopy(ahs_hdr->ahs_extscb, task->task_cdb + 16, addl_cdb_len);
2105 	}
2106 
2107 	DTRACE_ISCSI_3(scsi__command, idm_conn_t *, ic,
2108 	    iscsi_scsi_cmd_hdr_t *, (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr,
2109 	    scsi_task_t *, task);
2110 
2111 	/*
2112 	 * Copy the transport header into the task handle from the PDU
2113 	 * handle. The transport header describes this task's remote tagged
2114 	 * buffer.
2115 	 */
2116 	if (rx_pdu->isp_transport_hdrlen != 0) {
2117 		bcopy(rx_pdu->isp_transport_hdr,
2118 		    itask->it_idm_task->idt_transport_hdr,
2119 		    rx_pdu->isp_transport_hdrlen);
2120 	}
2121 
2122 	/*
2123 	 * Tell IDM about our new active task
2124 	 */
2125 	idm_task_start(itask->it_idm_task, (uintptr_t)itask->it_itt);
2126 
2127 	/*
2128 	 * If we have any immediate data then setup the immediate buffer
2129 	 * context that comes with the task
2130 	 */
2131 	if (rx_pdu->isp_datalen) {
2132 		ibuf = itask->it_immed_data;
2133 		ibuf->ibuf_immed_data_pdu = rx_pdu;
2134 		ibuf->ibuf_stmf_buf->db_data_size = rx_pdu->isp_datalen;
2135 		ibuf->ibuf_stmf_buf->db_buf_size = rx_pdu->isp_datalen;
2136 		ibuf->ibuf_stmf_buf->db_relative_offset = 0;
2137 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_length =
2138 		    rx_pdu->isp_datalen;
2139 		ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr = rx_pdu->isp_data;
2140 
2141 		DTRACE_ISCSI_8(xfer__start, idm_conn_t *, ic,
2142 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2143 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2144 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2145 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2146 
2147 		/*
2148 		 * For immediate data transfer, there is no callback from
2149 		 * stmf to indicate that the initial burst of data is
2150 		 * transferred successfully. In some cases, the task can
2151 		 * get freed before execution returns from stmf_post_task.
2152 		 * Although this xfer-start/done probe accurately tracks
2153 		 * the size of the transfer, it does only provide a best
2154 		 * effort on the timing of the transfer.
2155 		 */
2156 		DTRACE_ISCSI_8(xfer__done, idm_conn_t *, ic,
2157 		    uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2158 		    uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2159 		    uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2160 		    uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2161 		stmf_post_task(task, ibuf->ibuf_stmf_buf);
2162 	} else {
2163 
2164 		stmf_post_task(task, NULL);
2165 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2166 	}
2167 }
2168 
2169 void
2170 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu)
2171 {
2172 	iscsit_conn_t *ict = rx_pdu->isp_ic->ic_handle;
2173 
2174 	/*
2175 	 * If this isn't a login packet, we need a session.  Otherwise
2176 	 * this is a protocol error (perhaps one IDM should've caught?).
2177 	 */
2178 	if (IDM_PDU_OPCODE(rx_pdu) != ISCSI_OP_LOGIN_CMD &&
2179 	    ict->ict_sess == NULL) {
2180 		DTRACE_PROBE2(iscsi__idm__deferred__no__session,
2181 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
2182 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2183 		return;
2184 	}
2185 
2186 	/*
2187 	 * If the connection has been lost then ignore new PDU's
2188 	 */
2189 	mutex_enter(&ict->ict_mutex);
2190 	if (ict->ict_lost) {
2191 		mutex_exit(&ict->ict_mutex);
2192 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2193 		return;
2194 	}
2195 
2196 	/*
2197 	 * Grab a hold on the connection to prevent it from going away
2198 	 * between now and when the taskq function is called.
2199 	 */
2200 	iscsit_conn_dispatch_hold(ict);
2201 	mutex_exit(&ict->ict_mutex);
2202 
2203 	taskq_dispatch_ent(iscsit_global.global_dispatch_taskq,
2204 	    iscsit_deferred, rx_pdu, 0, &rx_pdu->isp_tqent);
2205 }
2206 
2207 static void
2208 iscsit_deferred(void *rx_pdu_void)
2209 {
2210 	idm_pdu_t		*rx_pdu = rx_pdu_void;
2211 	idm_conn_t		*ic = rx_pdu->isp_ic;
2212 	iscsit_conn_t		*ict = ic->ic_handle;
2213 
2214 	/*
2215 	 * NOP and Task Management Commands can be marked for immediate
2216 	 * delivery. Commands marked as 'Immediate' are to be considered
2217 	 * for execution as soon as they arrive on the target. So these
2218 	 * should not be checked for sequence order and put in a queue.
2219 	 * The CmdSN is not advanced for Immediate Commands.
2220 	 */
2221 	switch (IDM_PDU_OPCODE(rx_pdu)) {
2222 	case ISCSI_OP_NOOP_OUT:
2223 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2224 			iscsit_set_cmdsn(ict, rx_pdu);
2225 			iscsit_pdu_op_noop(ict, rx_pdu);
2226 		}
2227 		break;
2228 	case ISCSI_OP_LOGIN_CMD:
2229 		iscsit_pdu_op_login_cmd(ict, rx_pdu);
2230 		iscsit_conn_dispatch_rele(ict);
2231 		return;
2232 	case ISCSI_OP_TEXT_CMD:
2233 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2234 			iscsit_set_cmdsn(ict, rx_pdu);
2235 			iscsit_pdu_op_text_cmd(ict, rx_pdu);
2236 		}
2237 		break;
2238 	case ISCSI_OP_LOGOUT_CMD:
2239 		if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2240 			iscsit_set_cmdsn(ict, rx_pdu);
2241 			iscsit_pdu_op_logout_cmd(ict, rx_pdu);
2242 		}
2243 		break;
2244 	default:
2245 		/* Protocol error.  IDM should have caught this */
2246 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2247 		ASSERT(0);
2248 		break;
2249 	}
2250 	/*
2251 	 * Check if there are other PDUs in the session staging queue
2252 	 * waiting to be posted to SCSI layer.
2253 	 */
2254 	iscsit_process_pdu_in_queue(ict->ict_sess);
2255 
2256 	iscsit_conn_dispatch_rele(ict);
2257 }
2258 
2259 static void
2260 iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
2261     uint8_t response, uint8_t cmd_status)
2262 {
2263 	idm_pdu_t			*rsp_pdu;
2264 	idm_conn_t			*ic;
2265 	iscsi_scsi_rsp_hdr_t		*resp;
2266 	iscsi_scsi_cmd_hdr_t		*req =
2267 	    (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2268 
2269 	ic = ict->ict_ic;
2270 
2271 	rsp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_rsp_hdr_t), 0);
2272 	idm_pdu_init(rsp_pdu, ic, NULL, NULL);
2273 	/*
2274 	 * StatSN is incremented by 1 for every response sent on
2275 	 * a connection except for responses sent as a result of
2276 	 * a retry or SNACK
2277 	 */
2278 	rsp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2279 
2280 	resp = (iscsi_scsi_rsp_hdr_t *)rsp_pdu->isp_hdr;
2281 
2282 	resp->opcode = ISCSI_OP_SCSI_RSP;
2283 	resp->flags = ISCSI_FLAG_FINAL;
2284 	resp->response = response;
2285 	resp->cmd_status = cmd_status;
2286 	resp->itt = req->itt;
2287 	if ((response == ISCSI_STATUS_CMD_COMPLETED) &&
2288 	    (req->data_length != 0) &&
2289 	    ((req->flags & ISCSI_FLAG_CMD_READ) ||
2290 	    (req->flags & ISCSI_FLAG_CMD_WRITE))) {
2291 		resp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
2292 		resp->residual_count = req->data_length;
2293 	}
2294 
2295 	DTRACE_PROBE4(iscsi__scsi__direct__response,
2296 	    iscsit_conn_t *, ict,
2297 	    uint8_t, resp->response,
2298 	    uint8_t, resp->cmd_status,
2299 	    idm_pdu_t *, rsp_pdu);
2300 
2301 	iscsit_pdu_tx(rsp_pdu);
2302 }
2303 
2304 void
2305 iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu, uint8_t tm_status)
2306 {
2307 	iscsi_scsi_task_mgt_rsp_hdr_t	*tm_resp;
2308 
2309 	/*
2310 	 * The target must take note of the last-sent StatSN.
2311 	 * The StatSN is to be incremented after sending a
2312 	 * task management response. Digest recovery can only
2313 	 * work if StatSN is incremented.
2314 	 */
2315 	tm_resp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2316 	tm_resp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2317 	tm_resp->response = tm_status;
2318 
2319 	DTRACE_PROBE3(iscsi__scsi__tm__response,
2320 	    iscsit_conn_t *, tm_resp_pdu->isp_ic->ic_handle,
2321 	    uint8_t, tm_resp->response,
2322 	    idm_pdu_t *, tm_resp_pdu);
2323 	iscsit_pdu_tx(tm_resp_pdu);
2324 }
2325 
2326 void
2327 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2328 {
2329 	idm_pdu_t			*tm_resp_pdu;
2330 	iscsit_task_t			*itask;
2331 	iscsit_task_t			*tm_itask;
2332 	scsi_task_t			*task;
2333 	iscsi_scsi_task_mgt_hdr_t 	*iscsi_tm =
2334 	    (iscsi_scsi_task_mgt_hdr_t *)rx_pdu->isp_hdr;
2335 	iscsi_scsi_task_mgt_rsp_hdr_t 	*iscsi_tm_rsp =
2336 	    (iscsi_scsi_task_mgt_rsp_hdr_t *)rx_pdu->isp_hdr;
2337 	uint32_t			rtt, cmdsn, refcmdsn;
2338 	uint8_t				tm_func;
2339 
2340 	/*
2341 	 * Setup response PDU (response field will get filled in later)
2342 	 */
2343 	tm_resp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_task_mgt_rsp_hdr_t), 0);
2344 	if (tm_resp_pdu == NULL) {
2345 		/* Can't respond, just drop it */
2346 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2347 		return;
2348 	}
2349 	idm_pdu_init(tm_resp_pdu, ict->ict_ic, NULL, NULL);
2350 	iscsi_tm_rsp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2351 	bzero(iscsi_tm_rsp, sizeof (iscsi_scsi_task_mgt_rsp_hdr_t));
2352 	iscsi_tm_rsp->opcode = ISCSI_OP_SCSI_TASK_MGT_RSP;
2353 	iscsi_tm_rsp->flags = ISCSI_FLAG_FINAL;
2354 	iscsi_tm_rsp->itt = rx_pdu->isp_hdr->itt;
2355 
2356 	/*
2357 	 * Figure out what we're being asked to do.
2358 	 */
2359 	DTRACE_PROBE4(iscsi__scsi__tm__request,
2360 	    iscsit_conn_t *, ict,
2361 	    uint8_t, (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK),
2362 	    uint32_t, iscsi_tm->rtt,
2363 	    idm_pdu_t *, rx_pdu);
2364 	switch (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK) {
2365 	case ISCSI_TM_FUNC_ABORT_TASK:
2366 		/*
2367 		 * STMF doesn't currently support the "abort task" task
2368 		 * management command although it does support aborting
2369 		 * an individual task.  We'll get STMF to abort the task
2370 		 * for us but handle the details of the task management
2371 		 * command ourselves.
2372 		 *
2373 		 * Find the task associated with the referenced task tag.
2374 		 */
2375 		rtt = iscsi_tm->rtt;
2376 		itask = (iscsit_task_t *)idm_task_find_by_handle(ict->ict_ic,
2377 		    (uintptr_t)rtt);
2378 
2379 		if (itask == NULL) {
2380 			cmdsn = ntohl(iscsi_tm->cmdsn);
2381 			refcmdsn = ntohl(iscsi_tm->refcmdsn);
2382 
2383 			/*
2384 			 * Task was not found. But the SCSI command could be
2385 			 * on the rxpdu wait queue. If RefCmdSN is within
2386 			 * the CmdSN window and less than CmdSN of the TM
2387 			 * function, return "Function Complete". Otherwise,
2388 			 * return "Task Does Not Exist".
2389 			 */
2390 
2391 			if (iscsit_cmdsn_in_window(ict, refcmdsn) &&
2392 			    iscsit_sna_lt(refcmdsn, cmdsn)) {
2393 				mutex_enter(&ict->ict_sess->ist_sn_mutex);
2394 				(void) iscsit_remove_pdu_from_queue(
2395 				    ict->ict_sess, refcmdsn);
2396 				iscsit_conn_dispatch_rele(ict);
2397 				mutex_exit(&ict->ict_sess->ist_sn_mutex);
2398 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2399 				    SCSI_TCP_TM_RESP_COMPLETE);
2400 			} else {
2401 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2402 				    SCSI_TCP_TM_RESP_NO_TASK);
2403 			}
2404 		} else {
2405 
2406 			/*
2407 			 * Tell STMF to abort the task.  This will do no harm
2408 			 * if the task is already complete.
2409 			 */
2410 			stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
2411 			    STMF_ABORTED, NULL);
2412 
2413 			/*
2414 			 * Make sure the task hasn't already completed
2415 			 */
2416 			mutex_enter(&itask->it_idm_task->idt_mutex);
2417 			if ((itask->it_idm_task->idt_state == TASK_COMPLETE) ||
2418 			    (itask->it_idm_task->idt_state == TASK_IDLE)) {
2419 				/*
2420 				 * Task is complete, return "Task Does Not
2421 				 * Exist"
2422 				 */
2423 				mutex_exit(&itask->it_idm_task->idt_mutex);
2424 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2425 				    SCSI_TCP_TM_RESP_NO_TASK);
2426 			} else {
2427 				/*
2428 				 * STMF is now aborting the task, return
2429 				 * "Function Complete"
2430 				 */
2431 				mutex_exit(&itask->it_idm_task->idt_mutex);
2432 				iscsit_send_task_mgmt_resp(tm_resp_pdu,
2433 				    SCSI_TCP_TM_RESP_COMPLETE);
2434 			}
2435 			idm_task_rele(itask->it_idm_task);
2436 		}
2437 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2438 		return;
2439 
2440 	case ISCSI_TM_FUNC_ABORT_TASK_SET:
2441 		tm_func = TM_ABORT_TASK_SET;
2442 		break;
2443 
2444 	case ISCSI_TM_FUNC_CLEAR_ACA:
2445 		tm_func = TM_CLEAR_ACA;
2446 		break;
2447 
2448 	case ISCSI_TM_FUNC_CLEAR_TASK_SET:
2449 		tm_func = TM_CLEAR_TASK_SET;
2450 		break;
2451 
2452 	case ISCSI_TM_FUNC_LOGICAL_UNIT_RESET:
2453 		tm_func = TM_LUN_RESET;
2454 		break;
2455 
2456 	case ISCSI_TM_FUNC_TARGET_WARM_RESET:
2457 		tm_func = TM_TARGET_WARM_RESET;
2458 		break;
2459 
2460 	case ISCSI_TM_FUNC_TARGET_COLD_RESET:
2461 		tm_func = TM_TARGET_COLD_RESET;
2462 		break;
2463 
2464 	case ISCSI_TM_FUNC_TASK_REASSIGN:
2465 		/*
2466 		 * We do not currently support allegiance reassignment.  When
2467 		 * we start supporting ERL1+, we will need to.
2468 		 */
2469 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2470 		    SCSI_TCP_TM_RESP_NO_ALLG_REASSN);
2471 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2472 		return;
2473 
2474 	default:
2475 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2476 		    SCSI_TCP_TM_RESP_REJECTED);
2477 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2478 		return;
2479 	}
2480 
2481 	tm_itask = iscsit_tm_task_alloc(ict);
2482 	if (tm_itask == NULL) {
2483 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2484 		    SCSI_TCP_TM_RESP_REJECTED);
2485 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2486 		return;
2487 	}
2488 
2489 
2490 	task = stmf_task_alloc(ict->ict_sess->ist_lport,
2491 	    ict->ict_sess->ist_stmf_sess, iscsi_tm->lun,
2492 	    0, STMF_TASK_EXT_NONE);
2493 	if (task == NULL) {
2494 		/*
2495 		 * If this happens, either the LU is in reset, couldn't
2496 		 * get memory, or some other condition in which we simply
2497 		 * can't complete this request.  It would be nice to return
2498 		 * an error code like "busy" but the closest we have is
2499 		 * "rejected".
2500 		 */
2501 		iscsit_send_task_mgmt_resp(tm_resp_pdu,
2502 		    SCSI_TCP_TM_RESP_REJECTED);
2503 		iscsit_tm_task_free(tm_itask);
2504 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2505 		return;
2506 	}
2507 
2508 	tm_itask->it_tm_pdu = tm_resp_pdu;
2509 	tm_itask->it_stmf_task = task;
2510 	task->task_port_private = tm_itask;
2511 	task->task_mgmt_function = tm_func;
2512 	task->task_additional_flags = TASK_AF_NO_EXPECTED_XFER_LENGTH;
2513 	task->task_priority = 0;
2514 	task->task_max_nbufs = STMF_BUFS_MAX;
2515 	task->task_cmd_seq_no = iscsi_tm->itt;
2516 	task->task_expected_xfer_length = 0;
2517 
2518 	stmf_post_task(task, NULL);
2519 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2520 }
2521 
2522 static void
2523 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2524 {
2525 	iscsi_nop_out_hdr_t *out = (iscsi_nop_out_hdr_t *)rx_pdu->isp_hdr;
2526 	iscsi_nop_in_hdr_t *in;
2527 	int resp_datalen;
2528 	idm_pdu_t *resp;
2529 
2530 	/* Ignore the response from initiator */
2531 	if ((out->itt == ISCSI_RSVD_TASK_TAG) ||
2532 	    (out->ttt != ISCSI_RSVD_TASK_TAG)) {
2533 		idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2534 		return;
2535 	}
2536 
2537 	/* Allocate a PDU to respond */
2538 	resp_datalen = ntoh24(out->dlength);
2539 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
2540 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2541 	if (resp_datalen > 0) {
2542 		bcopy(rx_pdu->isp_data, resp->isp_data, resp_datalen);
2543 	}
2544 
2545 	/*
2546 	 * When sending a NOP-In as a response to a NOP-Out from the initiator,
2547 	 * the target must respond with the same initiator task tag that was
2548 	 * provided in the NOP-Out request, the target transfer tag must be
2549 	 * ISCSI_RSVD_TASK_TAG (0xffffffff) and StatSN will contain the next
2550 	 * status sequence number. The StatSN for the connection is advanced
2551 	 * after this PDU is sent.
2552 	 */
2553 	in = (iscsi_nop_in_hdr_t *)resp->isp_hdr;
2554 	bzero(in, sizeof (*in));
2555 	in->opcode = ISCSI_OP_NOOP_IN;
2556 	in->flags = ISCSI_FLAG_FINAL;
2557 	bcopy(out->lun, in->lun, 8);
2558 	in->itt		= out->itt;
2559 	in->ttt		= ISCSI_RSVD_TASK_TAG;
2560 	hton24(in->dlength, resp_datalen);
2561 	resp->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2562 	/* Any other field in resp to be set? */
2563 	iscsit_pdu_tx(resp);
2564 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2565 }
2566 
2567 static void
2568 iscsit_pdu_op_login_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2569 {
2570 
2571 	/*
2572 	 * Submit PDU to login state machine.  State machine will free the
2573 	 * PDU.
2574 	 */
2575 	iscsit_login_sm_event(ict, ILE_LOGIN_RCV, rx_pdu);
2576 }
2577 
2578 void
2579 iscsit_pdu_op_logout_cmd(iscsit_conn_t	*ict, idm_pdu_t *rx_pdu)
2580 {
2581 	iscsi_logout_hdr_t 	*logout_req =
2582 	    (iscsi_logout_hdr_t *)rx_pdu->isp_hdr;
2583 	iscsi_logout_rsp_hdr_t	*logout_rsp;
2584 	idm_pdu_t *resp;
2585 
2586 	/* Allocate a PDU to respond */
2587 	resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2588 	idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2589 	/*
2590 	 * The StatSN is to be sent to the initiator,
2591 	 * it is not required to increment the number
2592 	 * as the connection is terminating.
2593 	 */
2594 	resp->isp_flags |= IDM_PDU_SET_STATSN;
2595 	/*
2596 	 * Logout results in the immediate termination of all tasks except
2597 	 * if the logout reason is ISCSI_LOGOUT_REASON_RECOVERY.  The
2598 	 * connection state machine will drive this task cleanup automatically
2599 	 * so we don't need to handle that here.
2600 	 */
2601 	logout_rsp = (iscsi_logout_rsp_hdr_t *)resp->isp_hdr;
2602 	bzero(logout_rsp, sizeof (*logout_rsp));
2603 	logout_rsp->opcode = ISCSI_OP_LOGOUT_RSP;
2604 	logout_rsp->flags = ISCSI_FLAG_FINAL;
2605 	logout_rsp->itt = logout_req->itt;
2606 	if ((logout_req->flags & ISCSI_FLAG_LOGOUT_REASON_MASK) >
2607 	    ISCSI_LOGOUT_REASON_RECOVERY) {
2608 		logout_rsp->response = ISCSI_LOGOUT_RECOVERY_UNSUPPORTED;
2609 	} else {
2610 		logout_rsp->response = ISCSI_LOGOUT_SUCCESS;
2611 	}
2612 
2613 	iscsit_pdu_tx(resp);
2614 	idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2615 }
2616 
2617 /*
2618  * Calculate the number of outstanding commands we can process
2619  */
2620 int
2621 iscsit_cmd_window()
2622 {
2623 	/*
2624 	 * Instead of using a pre-defined constant for the command window,
2625 	 * it should be made confiurable and dynamic. With MC/S, sequence
2626 	 * numbers will be used up at a much faster rate than with SC/S.
2627 	 */
2628 	return	(ISCSIT_MAX_WINDOW);
2629 }
2630 
2631 /*
2632  * Set local registers based on incoming PDU
2633  */
2634 void
2635 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2636 {
2637 	iscsit_sess_t *ist;
2638 	iscsi_scsi_cmd_hdr_t *req;
2639 
2640 	ist = ict->ict_sess;
2641 
2642 	req = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2643 	if (req->opcode & ISCSI_OP_IMMEDIATE) {
2644 		/* no cmdsn increment for immediate PDUs */
2645 		return;
2646 	}
2647 
2648 	/* Ensure that the ExpCmdSN advances in an orderly manner */
2649 	mutex_enter(&ist->ist_sn_mutex);
2650 	ist->ist_expcmdsn = ntohl(req->cmdsn) + 1;
2651 	ist->ist_maxcmdsn = ntohl(req->cmdsn) + iscsit_cmd_window();
2652 	mutex_exit(&ist->ist_sn_mutex);
2653 }
2654 
2655 /*
2656  * Wrapper funtion, calls iscsi_calc_rspsn and idm_pdu_tx
2657  */
2658 void
2659 iscsit_pdu_tx(idm_pdu_t *pdu)
2660 {
2661 	iscsit_conn_t *ict = pdu->isp_ic->ic_handle;
2662 	iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2663 	iscsit_sess_t *ist = ict->ict_sess;
2664 
2665 	/*
2666 	 * The command sequence numbers are session-wide and must stay
2667 	 * consistent across the transfer, so protect the cmdsn with a
2668 	 * mutex lock on the session. The status sequence number will
2669 	 * be updated just before the transport layer transmits the PDU.
2670 	 */
2671 
2672 	mutex_enter(&ict->ict_sess->ist_sn_mutex);
2673 	/* Set ExpCmdSN and MaxCmdSN */
2674 	rsp->maxcmdsn = htonl(ist->ist_maxcmdsn);
2675 	rsp->expcmdsn = htonl(ist->ist_expcmdsn);
2676 	idm_pdu_tx(pdu);
2677 	mutex_exit(&ict->ict_sess->ist_sn_mutex);
2678 }
2679 
2680 /*
2681  * Internal functions
2682  */
2683 
2684 void
2685 iscsit_send_async_event(iscsit_conn_t *ict, uint8_t event)
2686 {
2687 	idm_pdu_t		*abt;
2688 	iscsi_async_evt_hdr_t	*async_abt;
2689 
2690 	/*
2691 	 * Get a PDU to build the abort request.
2692 	 */
2693 	abt = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2694 	if (abt == NULL) {
2695 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2696 		return;
2697 	}
2698 
2699 	/*
2700 	 * A asynchronous message is sent by the target to request a logout.
2701 	 * The StatSN for the connection is advanced after the PDU is sent
2702 	 * to allow for initiator and target state synchronization.
2703 	 */
2704 	idm_pdu_init(abt, ict->ict_ic, NULL, NULL);
2705 	abt->isp_datalen = 0;
2706 	abt->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2707 
2708 	async_abt = (iscsi_async_evt_hdr_t *)abt->isp_hdr;
2709 	bzero(async_abt, sizeof (*async_abt));
2710 	async_abt->opcode = ISCSI_OP_ASYNC_EVENT;
2711 	async_abt->async_event = event;
2712 	async_abt->flags = ISCSI_FLAG_FINAL;
2713 	async_abt->rsvd4[0] = 0xff;
2714 	async_abt->rsvd4[1] = 0xff;
2715 	async_abt->rsvd4[2] = 0xff;
2716 	async_abt->rsvd4[3] = 0xff;
2717 
2718 	switch (event) {
2719 	case ISCSI_ASYNC_EVENT_REQUEST_LOGOUT:
2720 		async_abt->param3 = htons(IDM_LOGOUT_SECONDS);
2721 		break;
2722 	case ISCSI_ASYNC_EVENT_SCSI_EVENT:
2723 	case ISCSI_ASYNC_EVENT_DROPPING_CONNECTION:
2724 	case ISCSI_ASYNC_EVENT_DROPPING_ALL_CONNECTIONS:
2725 	case ISCSI_ASYNC_EVENT_PARAM_NEGOTIATION:
2726 	default:
2727 		ASSERT(0);
2728 	}
2729 
2730 	iscsit_pdu_tx(abt);
2731 }
2732 
2733 void
2734 iscsit_send_reject(iscsit_conn_t *ict, idm_pdu_t *rejected_pdu, uint8_t reason)
2735 {
2736 	idm_pdu_t		*reject_pdu;
2737 	iscsi_reject_rsp_hdr_t	*reject;
2738 
2739 	/*
2740 	 * Get a PDU to build the abort request.
2741 	 */
2742 	reject_pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t),
2743 	    rejected_pdu->isp_hdrlen);
2744 	if (reject_pdu == NULL) {
2745 		idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2746 		return;
2747 	}
2748 	idm_pdu_init(reject_pdu, ict->ict_ic, NULL, NULL);
2749 	/* StatSN is advanced after a Reject PDU */
2750 	reject_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2751 	reject_pdu->isp_datalen = rejected_pdu->isp_hdrlen;
2752 	bcopy(rejected_pdu->isp_hdr, reject_pdu->isp_data,
2753 	    rejected_pdu->isp_hdrlen);
2754 
2755 	reject = (iscsi_reject_rsp_hdr_t *)reject_pdu->isp_hdr;
2756 	bzero(reject, sizeof (*reject));
2757 	reject->opcode = ISCSI_OP_REJECT_MSG;
2758 	reject->reason = reason;
2759 	reject->flags = ISCSI_FLAG_FINAL;
2760 	hton24(reject->dlength, rejected_pdu->isp_hdrlen);
2761 	reject->must_be_ff[0] = 0xff;
2762 	reject->must_be_ff[1] = 0xff;
2763 	reject->must_be_ff[2] = 0xff;
2764 	reject->must_be_ff[3] = 0xff;
2765 
2766 	iscsit_pdu_tx(reject_pdu);
2767 }
2768 
2769 
2770 static iscsit_task_t *
2771 iscsit_task_alloc(iscsit_conn_t *ict)
2772 {
2773 	iscsit_task_t *itask;
2774 	iscsit_buf_t *immed_ibuf;
2775 
2776 	/*
2777 	 * Possible items to pre-alloc if we cache iscsit_task_t's:
2778 	 *
2779 	 * Status PDU w/ sense buffer
2780 	 * stmf_data_buf_t for immediate data
2781 	 */
2782 	itask = kmem_alloc(sizeof (iscsit_task_t) + sizeof (iscsit_buf_t) +
2783 	    sizeof (stmf_data_buf_t), KM_NOSLEEP);
2784 	if (itask != NULL) {
2785 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2786 		itask->it_aborted = itask->it_stmf_abort =
2787 		    itask->it_tm_task = 0;
2788 
2789 		immed_ibuf = (iscsit_buf_t *)(itask + 1);
2790 		bzero(immed_ibuf, sizeof (*immed_ibuf));
2791 		immed_ibuf->ibuf_is_immed = B_TRUE;
2792 		immed_ibuf->ibuf_stmf_buf = (stmf_data_buf_t *)(immed_ibuf + 1);
2793 
2794 		bzero(immed_ibuf->ibuf_stmf_buf, sizeof (stmf_data_buf_t));
2795 		immed_ibuf->ibuf_stmf_buf->db_port_private = immed_ibuf;
2796 		immed_ibuf->ibuf_stmf_buf->db_sglist_length = 1;
2797 		immed_ibuf->ibuf_stmf_buf->db_flags = DB_DIRECTION_FROM_RPORT |
2798 		    DB_DONT_CACHE;
2799 		itask->it_immed_data = immed_ibuf;
2800 		itask->it_idm_task = idm_task_alloc(ict->ict_ic);
2801 		if (itask->it_idm_task != NULL) {
2802 			itask->it_idm_task->idt_private = itask;
2803 			itask->it_ict = ict;
2804 			itask->it_ttt = itask->it_idm_task->idt_tt;
2805 			return (itask);
2806 		} else {
2807 			kmem_free(itask, sizeof (iscsit_task_t) +
2808 			    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2809 		}
2810 	}
2811 
2812 	return (NULL);
2813 }
2814 
2815 static void
2816 iscsit_task_free(iscsit_task_t *itask)
2817 {
2818 	idm_task_free(itask->it_idm_task);
2819 	mutex_destroy(&itask->it_mutex);
2820 	kmem_free(itask, sizeof (iscsit_task_t) +
2821 	    sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2822 }
2823 
2824 static iscsit_task_t *
2825 iscsit_tm_task_alloc(iscsit_conn_t *ict)
2826 {
2827 	iscsit_task_t *itask;
2828 
2829 	itask = kmem_zalloc(sizeof (iscsit_task_t), KM_NOSLEEP);
2830 	if (itask != NULL) {
2831 		idm_conn_hold(ict->ict_ic);
2832 		mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2833 		itask->it_aborted = itask->it_stmf_abort =
2834 		    itask->it_tm_responded = 0;
2835 		itask->it_tm_pdu = NULL;
2836 		itask->it_tm_task = 1;
2837 		itask->it_ict = ict;
2838 	}
2839 
2840 	return (itask);
2841 }
2842 
2843 static void
2844 iscsit_tm_task_free(iscsit_task_t *itask)
2845 {
2846 	/*
2847 	 * If we responded then the call to idm_pdu_complete will free the
2848 	 * PDU.  Otherwise we got aborted before the TM function could
2849 	 * complete and we need to free the PDU explicitly.
2850 	 */
2851 	if (itask->it_tm_pdu != NULL && !itask->it_tm_responded)
2852 		idm_pdu_free(itask->it_tm_pdu);
2853 	idm_conn_rele(itask->it_ict->ict_ic);
2854 	mutex_destroy(&itask->it_mutex);
2855 	kmem_free(itask, sizeof (iscsit_task_t));
2856 }
2857 
2858 static idm_status_t
2859 iscsit_task_start(iscsit_task_t *itask)
2860 {
2861 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2862 	avl_index_t		where;
2863 
2864 	/*
2865 	 * Sanity check the ITT and ensure that this task does not already
2866 	 * exist.  If not then add the task to the session task list.
2867 	 */
2868 	mutex_enter(&ist->ist_mutex);
2869 	mutex_enter(&itask->it_mutex);
2870 	itask->it_active = 1;
2871 	if (avl_find(&ist->ist_task_list, itask, &where) == NULL) {
2872 		/* New task, add to AVL */
2873 		avl_insert(&ist->ist_task_list, itask, where);
2874 		mutex_exit(&itask->it_mutex);
2875 		mutex_exit(&ist->ist_mutex);
2876 		return (IDM_STATUS_SUCCESS);
2877 	}
2878 	mutex_exit(&itask->it_mutex);
2879 	mutex_exit(&ist->ist_mutex);
2880 
2881 	return (IDM_STATUS_REJECT);
2882 }
2883 
2884 static void
2885 iscsit_task_done(iscsit_task_t *itask)
2886 {
2887 	iscsit_sess_t *ist = itask->it_ict->ict_sess;
2888 
2889 	mutex_enter(&ist->ist_mutex);
2890 	mutex_enter(&itask->it_mutex);
2891 	if (itask->it_active) {
2892 		avl_remove(&ist->ist_task_list, itask);
2893 		itask->it_active = 0;
2894 	}
2895 	mutex_exit(&itask->it_mutex);
2896 	mutex_exit(&ist->ist_mutex);
2897 }
2898 
2899 /*
2900  * iscsit status PDU cache
2901  */
2902 
2903 /*ARGSUSED*/
2904 static int
2905 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags)
2906 {
2907 	idm_pdu_t *pdu = pdu_void;
2908 	iscsi_scsi_rsp_hdr_t *rsp;
2909 
2910 	bzero(pdu, sizeof (idm_pdu_t));
2911 	pdu->isp_callback = iscsit_send_good_status_done;
2912 	pdu->isp_magic = IDM_PDU_MAGIC;
2913 	pdu->isp_hdr = (iscsi_hdr_t *)(pdu + 1); /* Ptr arithmetic */
2914 	pdu->isp_hdrlen = sizeof (iscsi_hdr_t);
2915 
2916 	/* Setup status response */
2917 	rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2918 	bzero(rsp, sizeof (*rsp));
2919 	rsp->opcode = ISCSI_OP_SCSI_RSP;
2920 	rsp->flags = ISCSI_FLAG_FINAL;
2921 	rsp->response = ISCSI_STATUS_CMD_COMPLETED;
2922 
2923 	return (0);
2924 }
2925 
2926 /*
2927  * iscsit private data handler
2928  */
2929 
2930 /*ARGSUSED*/
2931 static void
2932 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags)
2933 {
2934 	it_config_t		*cfg;
2935 	nvlist_t		*nvl;
2936 	iscsit_service_enabled_t	old_state;
2937 
2938 	if ((cmd != STMF_PROVIDER_DATA_UPDATED) || (arg == NULL)) {
2939 		return;
2940 	}
2941 
2942 	nvl = (nvlist_t *)arg;
2943 
2944 	/* Translate nvlist */
2945 	if (it_nv_to_config(nvl, &cfg) != 0) {
2946 		cmn_err(CE_WARN, "Configuration is invalid");
2947 		return;
2948 	}
2949 
2950 	/* Check that no iSCSI ioctl is currently running */
2951 	mutex_enter(&iscsit_global.global_state_mutex);
2952 	old_state = iscsit_global.global_svc_state;
2953 	switch (iscsit_global.global_svc_state) {
2954 	case ISE_ENABLED:
2955 	case ISE_DISABLED:
2956 		iscsit_global.global_svc_state = ISE_BUSY;
2957 		break;
2958 	case ISE_ENABLING:
2959 		/*
2960 		 * It is OK for the iscsit_pp_cb to be called from inside of
2961 		 * an iSCSI ioctl only if we are currently executing inside
2962 		 * of stmf_register_port_provider.
2963 		 */
2964 		ASSERT((flags & STMF_PCB_PREG_COMPLETE) != 0);
2965 		break;
2966 	default:
2967 		cmn_err(CE_WARN, "iscsit_pp_cb called when global_svc_state"
2968 		    " is not ENABLED(0x%x) -- ignoring",
2969 		    iscsit_global.global_svc_state);
2970 		mutex_exit(&iscsit_global.global_state_mutex);
2971 		it_config_free_cmn(cfg);
2972 		return;
2973 	}
2974 	mutex_exit(&iscsit_global.global_state_mutex);
2975 
2976 	/* Update config */
2977 	(void) iscsit_config_merge(cfg);
2978 
2979 	it_config_free_cmn(cfg);
2980 
2981 	/* Restore old iSCSI driver global state */
2982 	mutex_enter(&iscsit_global.global_state_mutex);
2983 	ASSERT(iscsit_global.global_svc_state == ISE_BUSY ||
2984 	    iscsit_global.global_svc_state == ISE_ENABLING);
2985 	iscsit_global.global_svc_state = old_state;
2986 	mutex_exit(&iscsit_global.global_state_mutex);
2987 }
2988 
2989 
2990 static it_cfg_status_t
2991 iscsit_config_merge(it_config_t *in_cfg)
2992 {
2993 	it_cfg_status_t	status;
2994 	it_config_t	*cfg;
2995 	it_config_t	tmp_cfg;
2996 	list_t		tpg_del_list;
2997 
2998 	if (in_cfg) {
2999 		cfg = in_cfg;
3000 	} else {
3001 		/* Make empty config */
3002 		bzero(&tmp_cfg, sizeof (tmp_cfg));
3003 		cfg = &tmp_cfg;
3004 	}
3005 
3006 	list_create(&tpg_del_list,  sizeof (iscsit_tpg_t),
3007 	    offsetof(iscsit_tpg_t, tpg_delete_ln));
3008 
3009 	/*
3010 	 * Update targets, initiator contexts, target portal groups,
3011 	 * and iSNS client
3012 	 */
3013 	ISCSIT_GLOBAL_LOCK(RW_WRITER);
3014 	if (((status = iscsit_config_merge_tpg(cfg, &tpg_del_list))
3015 	    != 0) ||
3016 	    ((status = iscsit_config_merge_tgt(cfg)) != 0) ||
3017 	    ((status = iscsit_config_merge_ini(cfg)) != 0) ||
3018 	    ((status = isnst_config_merge(cfg)) != 0)) {
3019 		ISCSIT_GLOBAL_UNLOCK();
3020 		return (status);
3021 	}
3022 
3023 	/* Update other global config parameters */
3024 	if (iscsit_global.global_props) {
3025 		nvlist_free(iscsit_global.global_props);
3026 		iscsit_global.global_props = NULL;
3027 	}
3028 	if (in_cfg) {
3029 		(void) nvlist_dup(cfg->config_global_properties,
3030 		    &iscsit_global.global_props, KM_SLEEP);
3031 	}
3032 	ISCSIT_GLOBAL_UNLOCK();
3033 
3034 	iscsit_config_destroy_tpgs(&tpg_del_list);
3035 
3036 	list_destroy(&tpg_del_list);
3037 
3038 	return (ITCFG_SUCCESS);
3039 }
3040 
3041 /*
3042  * iscsit_sna_lt[e]
3043  *
3044  * Compare serial numbers using serial number arithmetic as defined in
3045  * RFC 1982.
3046  *
3047  * NOTE: This code is duplicated in the isns server. It ought to be common.
3048  */
3049 
3050 static int
3051 iscsit_sna_lt(uint32_t sn1, uint32_t sn2)
3052 {
3053 	return ((sn1 != sn2) &&
3054 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3055 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3056 }
3057 
3058 static int
3059 iscsit_sna_lte(uint32_t sn1, uint32_t sn2)
3060 {
3061 	return ((sn1 == sn2) ||
3062 	    (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3063 	    ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3064 }
3065 
3066 
3067 static boolean_t
3068 iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn)
3069 {
3070 	iscsit_sess_t	*ist = ict->ict_sess;
3071 	int		rval = B_TRUE;
3072 
3073 	ist = ict->ict_sess;
3074 
3075 	mutex_enter(&ist->ist_sn_mutex);
3076 
3077 	/*
3078 	 * If cmdsn is less than ist_expcmdsn - iscsit_cmd_window() or
3079 	 * greater than ist_expcmdsn, it's not in the window.
3080 	 */
3081 
3082 	if (iscsit_sna_lt(cmdsn, (ist->ist_expcmdsn - iscsit_cmd_window())) ||
3083 	    !iscsit_sna_lte(cmdsn, ist->ist_expcmdsn)) {
3084 		rval = B_FALSE;
3085 	}
3086 
3087 	mutex_exit(&ist->ist_sn_mutex);
3088 
3089 	return (rval);
3090 }
3091 
3092 /*
3093  * iscsit_check_cmdsn_and_queue
3094  *
3095  * Independent of the order in which the iSCSI target receives non-immediate
3096  * command PDU across the entire session and any multiple connections within
3097  * the session, the target must deliver the commands to the SCSI layer in
3098  * CmdSN order. So out-of-order non-immediate commands are queued up on a
3099  * session-wide wait queue. Duplicate commands are ignored.
3100  *
3101  */
3102 static int
3103 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu)
3104 {
3105 	idm_conn_t		*ic = rx_pdu->isp_ic;
3106 	iscsit_conn_t		*ict = ic->ic_handle;
3107 	iscsit_sess_t		*ist = ict->ict_sess;
3108 	iscsi_scsi_cmd_hdr_t	*hdr = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
3109 
3110 	mutex_enter(&ist->ist_sn_mutex);
3111 	if (hdr->opcode & ISCSI_OP_IMMEDIATE) {
3112 		/* do not queue, handle it immediately */
3113 		DTRACE_PROBE2(immediate__cmd, iscsit_sess_t *, ist,
3114 		    idm_pdu_t *, rx_pdu);
3115 		mutex_exit(&ist->ist_sn_mutex);
3116 		return (ISCSIT_CMDSN_EQ_EXPCMDSN);
3117 	}
3118 	if (iscsit_sna_lt(ist->ist_expcmdsn, ntohl(hdr->cmdsn))) {
3119 		/*
3120 		 * Out-of-order commands (cmdSN higher than ExpCmdSN)
3121 		 * are staged on a fixed-size circular buffer until
3122 		 * the missing command is delivered to the SCSI layer.
3123 		 * Irrespective of the order of insertion into the
3124 		 * staging queue, the commands are processed out of the
3125 		 * queue in cmdSN order only.
3126 		 */
3127 		rx_pdu->isp_queue_time = gethrtime();
3128 		iscsit_add_pdu_to_queue(ist, rx_pdu);
3129 		mutex_exit(&ist->ist_sn_mutex);
3130 		return (ISCSIT_CMDSN_GT_EXPCMDSN);
3131 	} else if (iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_expcmdsn)) {
3132 		DTRACE_PROBE3(cmdsn__lt__expcmdsn, iscsit_sess_t *, ist,
3133 		    iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
3134 		mutex_exit(&ist->ist_sn_mutex);
3135 		return (ISCSIT_CMDSN_LT_EXPCMDSN);
3136 	} else {
3137 		mutex_exit(&ist->ist_sn_mutex);
3138 		return (ISCSIT_CMDSN_EQ_EXPCMDSN);
3139 	}
3140 }
3141 
3142 /*
3143  * iscsit_add_pdu_to_queue() adds PDUs into the array indexed by
3144  * their cmdsn value. The length of the array is kept above the
3145  * maximum window size. The window keeps the cmdsn within a range
3146  * such that there are no collisons. e.g. the assumption is that
3147  * the windowing checks make it impossible to receive PDUs that
3148  * index into the same location in the array.
3149  */
3150 static void
3151 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu)
3152 {
3153 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3154 	iscsit_conn_t	*ict 	= rx_pdu->isp_ic->ic_handle;
3155 	uint32_t	cmdsn	=
3156 	    ((iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr)->cmdsn;
3157 	uint32_t	index;
3158 
3159 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3160 	/*
3161 	 * If the connection is being torn down, then
3162 	 * don't add the PDU to the staging queue
3163 	 */
3164 	mutex_enter(&ict->ict_mutex);
3165 	if (ict->ict_lost) {
3166 		mutex_exit(&ict->ict_mutex);
3167 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3168 		return;
3169 	}
3170 	iscsit_conn_dispatch_hold(ict);
3171 	mutex_exit(&ict->ict_mutex);
3172 
3173 	index = ntohl(cmdsn) % ISCSIT_RXPDU_QUEUE_LEN;
3174 	/*
3175 	 * In the normal case, assuming that the Initiator is not
3176 	 * buggy and that we don't have packet duplication occuring,
3177 	 * the entry in the array will be NULL.  However, we may have
3178 	 * received a duplicate PDU with cmdsn > expsn , and in that
3179 	 * case we just ignore this PDU -- the previously received one
3180 	 * remains queued for processing.  We need to be careful not
3181 	 * to leak this one however.
3182 	 */
3183 	if (cbuf->cb_buffer[index] != NULL) {
3184 		idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3185 	} else {
3186 		cbuf->cb_buffer[index] = rx_pdu;
3187 		cbuf->cb_num_elems++;
3188 	}
3189 }
3190 
3191 static idm_pdu_t *
3192 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn)
3193 {
3194 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3195 	idm_pdu_t	*pdu	= NULL;
3196 	uint32_t	index;
3197 
3198 	ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3199 	index = cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3200 	if ((pdu = cbuf->cb_buffer[index]) != NULL) {
3201 		ASSERT(cmdsn ==
3202 		    ntohl(((iscsi_scsi_cmd_hdr_t *)pdu->isp_hdr)->cmdsn));
3203 		cbuf->cb_buffer[index] = NULL;
3204 		cbuf->cb_num_elems--;
3205 		return (pdu);
3206 	}
3207 	return (NULL);
3208 }
3209 
3210 /*
3211  * iscsit_process_pdu_in_queue() finds the next pdu in sequence
3212  * and posts it to the SCSI layer
3213  */
3214 static void
3215 iscsit_process_pdu_in_queue(iscsit_sess_t *ist)
3216 {
3217 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3218 	idm_pdu_t	*pdu = NULL;
3219 	uint32_t	expcmdsn;
3220 
3221 	for (;;) {
3222 		mutex_enter(&ist->ist_sn_mutex);
3223 		if (cbuf->cb_num_elems == 0) {
3224 			mutex_exit(&ist->ist_sn_mutex);
3225 			break;
3226 		}
3227 		expcmdsn = ist->ist_expcmdsn;
3228 		if ((pdu = iscsit_remove_pdu_from_queue(ist, expcmdsn))
3229 		    == NULL) {
3230 			mutex_exit(&ist->ist_sn_mutex);
3231 			break;
3232 		}
3233 		mutex_exit(&ist->ist_sn_mutex);
3234 		iscsit_post_staged_pdu(pdu);
3235 	}
3236 }
3237 
3238 static void
3239 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu)
3240 {
3241 	iscsit_conn_t	*ict	= rx_pdu->isp_ic->ic_handle;
3242 
3243 	/* Post the PDU to the SCSI layer */
3244 	switch (IDM_PDU_OPCODE(rx_pdu)) {
3245 	case ISCSI_OP_NOOP_OUT:
3246 		iscsit_set_cmdsn(ict, rx_pdu);
3247 		iscsit_pdu_op_noop(ict, rx_pdu);
3248 		break;
3249 	case ISCSI_OP_TEXT_CMD:
3250 		iscsit_set_cmdsn(ict, rx_pdu);
3251 		iscsit_pdu_op_text_cmd(ict, rx_pdu);
3252 		break;
3253 	case ISCSI_OP_SCSI_TASK_MGT_MSG:
3254 		iscsit_set_cmdsn(ict, rx_pdu);
3255 		iscsit_op_scsi_task_mgmt(ict, rx_pdu);
3256 		break;
3257 	case ISCSI_OP_SCSI_CMD:
3258 		/* cmdSN will be incremented after creating itask */
3259 		iscsit_post_scsi_cmd(rx_pdu->isp_ic, rx_pdu);
3260 		break;
3261 	case ISCSI_OP_LOGOUT_CMD:
3262 		iscsit_set_cmdsn(ict, rx_pdu);
3263 		iscsit_pdu_op_logout_cmd(ict, rx_pdu);
3264 		break;
3265 	default:
3266 		/* No other PDUs should be placed on the queue */
3267 		ASSERT(0);
3268 	}
3269 	iscsit_conn_dispatch_rele(ict); /* release hold on the conn */
3270 }
3271 
3272 /* ARGSUSED */
3273 void
3274 iscsit_rxpdu_queue_monitor_start(void)
3275 {
3276 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3277 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3278 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3279 		return;
3280 	}
3281 	iscsit_rxpdu_queue_monitor_thr_id =
3282 	    thread_create(NULL, 0, iscsit_rxpdu_queue_monitor, NULL,
3283 	    0, &p0, TS_RUN, minclsyspri);
3284 	while (!iscsit_rxpdu_queue_monitor_thr_running) {
3285 		cv_wait(&iscsit_rxpdu_queue_monitor_cv,
3286 		    &iscsit_rxpdu_queue_monitor_mutex);
3287 	}
3288 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3289 
3290 }
3291 
3292 /* ARGSUSED */
3293 void
3294 iscsit_rxpdu_queue_monitor_stop(void)
3295 {
3296 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3297 	if (iscsit_rxpdu_queue_monitor_thr_running) {
3298 		iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
3299 		cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3300 		mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3301 
3302 		thread_join(iscsit_rxpdu_queue_monitor_thr_did);
3303 		return;
3304 	}
3305 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3306 }
3307 
3308 /*
3309  * A separate thread is used to scan the staging queue on all the
3310  * sessions, If a delayed PDU does not arrive within a timeout, the
3311  * target will advance to the staged PDU that is next in sequence
3312  * and exceeded the threshold wait time. It is up to the initiator
3313  * to note that the target has not acknowledged a particular cmdsn
3314  * and take appropriate action.
3315  */
3316 /* ARGSUSED */
3317 static void
3318 iscsit_rxpdu_queue_monitor(void *arg)
3319 {
3320 	iscsit_tgt_t	*tgt;
3321 	iscsit_sess_t	*ist;
3322 
3323 	mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3324 	iscsit_rxpdu_queue_monitor_thr_did = curthread->t_did;
3325 	iscsit_rxpdu_queue_monitor_thr_running = B_TRUE;
3326 	cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3327 
3328 	while (iscsit_rxpdu_queue_monitor_thr_running) {
3329 		ISCSIT_GLOBAL_LOCK(RW_READER);
3330 		for (tgt = avl_first(&iscsit_global.global_target_list);
3331 		    tgt != NULL;
3332 		    tgt = AVL_NEXT(&iscsit_global.global_target_list, tgt)) {
3333 			mutex_enter(&tgt->target_mutex);
3334 			for (ist = avl_first(&tgt->target_sess_list);
3335 			    ist != NULL;
3336 			    ist = AVL_NEXT(&tgt->target_sess_list, ist)) {
3337 
3338 				iscsit_rxpdu_queue_monitor_session(ist);
3339 			}
3340 			mutex_exit(&tgt->target_mutex);
3341 		}
3342 		ISCSIT_GLOBAL_UNLOCK();
3343 		if (iscsit_rxpdu_queue_monitor_thr_running == B_FALSE) {
3344 			break;
3345 		}
3346 		(void) cv_reltimedwait(&iscsit_rxpdu_queue_monitor_cv,
3347 		    &iscsit_rxpdu_queue_monitor_mutex,
3348 		    ISCSIT_RXPDU_QUEUE_MONITOR_INTERVAL * drv_usectohz(1000000),
3349 		    TR_CLOCK_TICK);
3350 	}
3351 	mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3352 	thread_exit();
3353 }
3354 
3355 static void
3356 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist)
3357 {
3358 	iscsit_cbuf_t	*cbuf	= ist->ist_rxpdu_queue;
3359 	idm_pdu_t	*next_pdu = NULL;
3360 	uint32_t	index, next_cmdsn, i;
3361 
3362 	/*
3363 	 * Assume that all PDUs in the staging queue have a cmdsn >= expcmdsn.
3364 	 * Starting with the expcmdsn, iterate over the staged PDUs to find
3365 	 * the next PDU with a wait time greater than the threshold. If found
3366 	 * advance the staged PDU to the SCSI layer, skipping over the missing
3367 	 * PDU(s) to get past the hole in the command sequence. It is up to
3368 	 * the initiator to note that the target has not acknowledged a cmdsn
3369 	 * and take appropriate action.
3370 	 *
3371 	 * Since the PDU(s) arrive in any random order, it is possible that
3372 	 * that the actual wait time for a particular PDU is much longer than
3373 	 * the defined threshold. e.g. Consider a case where commands are sent
3374 	 * over 4 different connections, and cmdsn = 1004 arrives first, then
3375 	 * 1003, and 1002 and 1001 are lost due to a connection failure.
3376 	 * So now 1003 is waiting for 1002 to be delivered, and although the
3377 	 * wait time of 1004 > wait time of 1003, only 1003 will be considered
3378 	 * by the monitor thread. 1004 will be automatically processed by
3379 	 * iscsit_process_pdu_in_queue() once the scan is complete and the
3380 	 * expcmdsn becomes current.
3381 	 */
3382 	mutex_enter(&ist->ist_sn_mutex);
3383 	cbuf = ist->ist_rxpdu_queue;
3384 	if (cbuf->cb_num_elems == 0) {
3385 		mutex_exit(&ist->ist_sn_mutex);
3386 		return;
3387 	}
3388 	for (next_pdu = NULL, i = 0; ; i++) {
3389 		next_cmdsn = ist->ist_expcmdsn + i; /* start at expcmdsn */
3390 		index = next_cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3391 		if ((next_pdu = cbuf->cb_buffer[index]) != NULL) {
3392 			/*
3393 			 * If the PDU wait time has not exceeded threshold
3394 			 * stop scanning the staging queue until the timer
3395 			 * fires again
3396 			 */
3397 			if ((gethrtime() - next_pdu->isp_queue_time)
3398 			    < (rxpdu_queue_threshold * NANOSEC)) {
3399 				mutex_exit(&ist->ist_sn_mutex);
3400 				return;
3401 			}
3402 			/*
3403 			 * Remove the next PDU from the queue and post it
3404 			 * to the SCSI layer, skipping over the missing
3405 			 * PDU. Stop scanning the staging queue until
3406 			 * the monitor timer fires again
3407 			 */
3408 			(void) iscsit_remove_pdu_from_queue(ist, next_cmdsn);
3409 			mutex_exit(&ist->ist_sn_mutex);
3410 			DTRACE_PROBE3(advanced__to__blocked__cmdsn,
3411 			    iscsit_sess_t *, ist, idm_pdu_t *, next_pdu,
3412 			    uint32_t, next_cmdsn);
3413 			iscsit_post_staged_pdu(next_pdu);
3414 			/* Deliver any subsequent PDUs immediately */
3415 			iscsit_process_pdu_in_queue(ist);
3416 			return;
3417 		}
3418 		/*
3419 		 * Skipping over i PDUs, e.g. a case where commands 1001 and
3420 		 * 1002 are lost in the network, skip over both and post 1003
3421 		 * expcmdsn then becomes 1004 at the end of the scan.
3422 		 */
3423 		DTRACE_PROBE2(skipping__over__cmdsn, iscsit_sess_t *, ist,
3424 		    uint32_t, next_cmdsn);
3425 	}
3426 	/*
3427 	 * following the assumption, staged cmdsn >= expcmdsn, this statement
3428 	 * is never reached.
3429 	 */
3430 }
3431