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  * Copyright 2017 Nexenta Systems, Inc.
  24  * Copyright (c) 2017, Joyent, 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 boolean_t
 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                 if (cfg_nvlist)
 567                         nvlist_free(cfg_nvlist);
 568 
 569                 /*
 570                  * Now that the reconfig is complete set our state back to
 571                  * enabled.
 572                  */
 573                 mutex_enter(&iscsit_global.global_state_mutex);
 574                 iscsit_global.global_svc_state = ISE_ENABLED;
 575                 mutex_exit(&iscsit_global.global_state_mutex);
 576                 break;
 577         case ISCSIT_IOC_ENABLE_SVC: {
 578                 iscsit_hostinfo_t hostinfo;
 579 
 580                 if (ddi_copyin((void *)argp, &hostinfo.length,
 581                     sizeof (hostinfo.length), flag) != 0) {
 582                         mutex_enter(&iscsit_global.global_state_mutex);
 583                         iscsit_global.global_svc_state = ISE_DISABLED;
 584                         mutex_exit(&iscsit_global.global_state_mutex);
 585                         return (EFAULT);
 586                 }
 587 
 588                 if (hostinfo.length > sizeof (hostinfo.fqhn))
 589                         hostinfo.length = sizeof (hostinfo.fqhn);
 590 
 591                 if (ddi_copyin((void *)((caddr_t)argp +
 592                     sizeof (hostinfo.length)), &hostinfo.fqhn,
 593                     hostinfo.length, flag) != 0) {
 594                         mutex_enter(&iscsit_global.global_state_mutex);
 595                         iscsit_global.global_svc_state = ISE_DISABLED;
 596                         mutex_exit(&iscsit_global.global_state_mutex);
 597                         return (EFAULT);
 598                 }
 599 
 600                 idmrc = iscsit_enable_svc(&hostinfo);
 601                 mutex_enter(&iscsit_global.global_state_mutex);
 602                 if (idmrc == IDM_STATUS_SUCCESS) {
 603                         iscsit_global.global_svc_state = ISE_ENABLED;
 604                 } else {
 605                         rc = EIO;
 606                         iscsit_global.global_svc_state = ISE_DISABLED;
 607                 }
 608                 mutex_exit(&iscsit_global.global_state_mutex);
 609                 break;
 610         }
 611         case ISCSIT_IOC_DISABLE_SVC:
 612                 iscsit_disable_svc();
 613                 mutex_enter(&iscsit_global.global_state_mutex);
 614                 iscsit_global.global_svc_state = ISE_DISABLED;
 615                 mutex_exit(&iscsit_global.global_state_mutex);
 616                 break;
 617 
 618         default:
 619                 rc = EINVAL;
 620                 mutex_enter(&iscsit_global.global_state_mutex);
 621                 iscsit_global.global_svc_state = ISE_ENABLED;
 622                 mutex_exit(&iscsit_global.global_state_mutex);
 623         }
 624 
 625         return (rc);
 626 }
 627 
 628 static idm_status_t
 629 iscsit_init(dev_info_t *dip)
 630 {
 631         int                     rc;
 632 
 633         rc = ldi_ident_from_dip(dip, &iscsit_global.global_li);
 634         ASSERT(rc == 0);  /* Failure indicates invalid argument */
 635 
 636         iscsit_global.global_svc_state = ISE_DISABLED;
 637 
 638         return (IDM_STATUS_SUCCESS);
 639 }
 640 
 641 /*
 642  * iscsit_enable_svc
 643  *
 644  * registers all the configured targets and target portals with STMF
 645  */
 646 static idm_status_t
 647 iscsit_enable_svc(iscsit_hostinfo_t *hostinfo)
 648 {
 649         stmf_port_provider_t    *pp;
 650         stmf_dbuf_store_t       *dbuf_store;
 651         boolean_t               did_iscsit_isns_init;
 652         idm_status_t            retval = IDM_STATUS_SUCCESS;
 653 
 654         ASSERT(iscsit_global.global_svc_state == ISE_ENABLING);
 655 
 656         /*
 657          * Make sure that can tell if we have partially allocated
 658          * in case we need to exit and tear down anything allocated.
 659          */
 660         iscsit_global.global_tsih_pool = NULL;
 661         iscsit_global.global_dbuf_store = NULL;
 662         iscsit_status_pdu_cache = NULL;
 663         pp = NULL;
 664         iscsit_global.global_pp = NULL;
 665         iscsit_global.global_default_tpg = NULL;
 666         did_iscsit_isns_init = B_FALSE;
 667         iscsit_global.global_dispatch_taskq = NULL;
 668 
 669         /* Setup remaining fields in iscsit_global_t */
 670         idm_refcnt_init(&iscsit_global.global_refcnt,
 671             &iscsit_global);
 672 
 673         avl_create(&iscsit_global.global_discovery_sessions,
 674             iscsit_sess_avl_compare, sizeof (iscsit_sess_t),
 675             offsetof(iscsit_sess_t, ist_tgt_ln));
 676 
 677         avl_create(&iscsit_global.global_target_list,
 678             iscsit_tgt_avl_compare, sizeof (iscsit_tgt_t),
 679             offsetof(iscsit_tgt_t, target_global_ln));
 680 
 681         list_create(&iscsit_global.global_deleted_target_list,
 682             sizeof (iscsit_tgt_t),
 683             offsetof(iscsit_tgt_t, target_global_deleted_ln));
 684 
 685         avl_create(&iscsit_global.global_tpg_list,
 686             iscsit_tpg_avl_compare, sizeof (iscsit_tpg_t),
 687             offsetof(iscsit_tpg_t, tpg_global_ln));
 688 
 689         avl_create(&iscsit_global.global_ini_list,
 690             iscsit_ini_avl_compare, sizeof (iscsit_ini_t),
 691             offsetof(iscsit_ini_t, ini_global_ln));
 692 
 693         iscsit_global.global_tsih_pool = vmem_create("iscsit_tsih_pool",
 694             (void *)1, ISCSI_MAX_TSIH, 1, NULL, NULL, NULL, 0,
 695             VM_SLEEP | VMC_IDENTIFIER);
 696 
 697         /*
 698          * Setup STMF dbuf store.  Our buffers are bound to a specific
 699          * connection so we really can't let STMF cache buffers for us.
 700          * Consequently we'll just allocate one global buffer store.
 701          */
 702         dbuf_store = stmf_alloc(STMF_STRUCT_DBUF_STORE, 0, 0);
 703         if (dbuf_store == NULL) {
 704                 retval = IDM_STATUS_FAIL;
 705                 goto tear_down_and_return;
 706         }
 707         dbuf_store->ds_alloc_data_buf = iscsit_dbuf_alloc;
 708         dbuf_store->ds_free_data_buf = iscsit_dbuf_free;
 709         dbuf_store->ds_port_private = NULL;
 710         iscsit_global.global_dbuf_store = dbuf_store;
 711 
 712         /* Status PDU cache */
 713         iscsit_status_pdu_cache = kmem_cache_create("iscsit_status_pdu_cache",
 714             sizeof (idm_pdu_t) + sizeof (iscsi_scsi_rsp_hdr_t), 8,
 715             &iscsit_status_pdu_constructor,
 716             NULL, NULL, NULL, NULL, KM_SLEEP);
 717 
 718         /* Default TPG and portal */
 719         iscsit_global.global_default_tpg = iscsit_tpg_createdefault();
 720         if (iscsit_global.global_default_tpg == NULL) {
 721                 retval = IDM_STATUS_FAIL;
 722                 goto tear_down_and_return;
 723         }
 724 
 725         /* initialize isns client */
 726         (void) iscsit_isns_init(hostinfo);
 727         did_iscsit_isns_init = B_TRUE;
 728 
 729         /* Register port provider */
 730         pp = stmf_alloc(STMF_STRUCT_PORT_PROVIDER, 0, 0);
 731         if (pp == NULL) {
 732                 retval = IDM_STATUS_FAIL;
 733                 goto tear_down_and_return;
 734         }
 735 
 736         pp->pp_portif_rev = PORTIF_REV_1;
 737         pp->pp_instance = 0;
 738         pp->pp_name = ISCSIT_MODNAME;
 739         pp->pp_cb = iscsit_pp_cb;
 740 
 741         iscsit_global.global_pp = pp;
 742 
 743 
 744         if (stmf_register_port_provider(pp) != STMF_SUCCESS) {
 745                 retval = IDM_STATUS_FAIL;
 746                 goto tear_down_and_return;
 747         }
 748 
 749         iscsit_global.global_dispatch_taskq = taskq_create("iscsit_dispatch",
 750             1, minclsyspri, 16, 16, TASKQ_PREPOPULATE);
 751 
 752         /* Scan staged PDUs, meaningful in MC/S situations */
 753         iscsit_rxpdu_queue_monitor_start();
 754 
 755         return (IDM_STATUS_SUCCESS);
 756 
 757 tear_down_and_return:
 758 
 759         if (iscsit_global.global_dispatch_taskq) {
 760                 taskq_destroy(iscsit_global.global_dispatch_taskq);
 761                 iscsit_global.global_dispatch_taskq = NULL;
 762         }
 763 
 764         if (did_iscsit_isns_init)
 765                 iscsit_isns_fini();
 766 
 767         if (iscsit_global.global_default_tpg) {
 768                 iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
 769                 iscsit_global.global_default_tpg = NULL;
 770         }
 771 
 772         if (iscsit_global.global_pp)
 773                 iscsit_global.global_pp = NULL;
 774 
 775         if (pp)
 776                 stmf_free(pp);
 777 
 778         if (iscsit_status_pdu_cache) {
 779                 kmem_cache_destroy(iscsit_status_pdu_cache);
 780                 iscsit_status_pdu_cache = NULL;
 781         }
 782 
 783         if (iscsit_global.global_dbuf_store) {
 784                 stmf_free(iscsit_global.global_dbuf_store);
 785                 iscsit_global.global_dbuf_store = NULL;
 786         }
 787 
 788         if (iscsit_global.global_tsih_pool) {
 789                 vmem_destroy(iscsit_global.global_tsih_pool);
 790                 iscsit_global.global_tsih_pool = NULL;
 791         }
 792 
 793         avl_destroy(&iscsit_global.global_ini_list);
 794         avl_destroy(&iscsit_global.global_tpg_list);
 795         list_destroy(&iscsit_global.global_deleted_target_list);
 796         avl_destroy(&iscsit_global.global_target_list);
 797         avl_destroy(&iscsit_global.global_discovery_sessions);
 798 
 799         idm_refcnt_destroy(&iscsit_global.global_refcnt);
 800 
 801         return (retval);
 802 }
 803 
 804 /*
 805  * iscsit_disable_svc
 806  *
 807  * clean up all existing connections and deregister targets from STMF
 808  */
 809 static void
 810 iscsit_disable_svc(void)
 811 {
 812         iscsit_sess_t   *sess;
 813 
 814         ASSERT(iscsit_global.global_svc_state == ISE_DISABLING);
 815 
 816         iscsit_rxpdu_queue_monitor_stop();
 817 
 818         /* tear down discovery sessions */
 819         for (sess = avl_first(&iscsit_global.global_discovery_sessions);
 820             sess != NULL;
 821             sess = AVL_NEXT(&iscsit_global.global_discovery_sessions, sess))
 822                 iscsit_sess_close(sess);
 823 
 824         /*
 825          * Passing NULL to iscsit_config_merge tells it to go to an empty
 826          * config.
 827          */
 828         (void) iscsit_config_merge(NULL);
 829 
 830         /*
 831          * Wait until there are no more global references
 832          */
 833         idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
 834         idm_refcnt_destroy(&iscsit_global.global_refcnt);
 835 
 836         /*
 837          * Default TPG must be destroyed after global_refcnt is 0.
 838          */
 839         iscsit_tpg_destroydefault(iscsit_global.global_default_tpg);
 840 
 841         avl_destroy(&iscsit_global.global_discovery_sessions);
 842         list_destroy(&iscsit_global.global_deleted_target_list);
 843         avl_destroy(&iscsit_global.global_target_list);
 844         avl_destroy(&iscsit_global.global_tpg_list);
 845         avl_destroy(&iscsit_global.global_ini_list);
 846 
 847         taskq_destroy(iscsit_global.global_dispatch_taskq);
 848 
 849         iscsit_isns_fini();
 850 
 851         stmf_free(iscsit_global.global_dbuf_store);
 852         iscsit_global.global_dbuf_store = NULL;
 853 
 854         (void) stmf_deregister_port_provider(iscsit_global.global_pp);
 855         stmf_free(iscsit_global.global_pp);
 856         iscsit_global.global_pp = NULL;
 857 
 858         kmem_cache_destroy(iscsit_status_pdu_cache);
 859         iscsit_status_pdu_cache = NULL;
 860 
 861         vmem_destroy(iscsit_global.global_tsih_pool);
 862         iscsit_global.global_tsih_pool = NULL;
 863 }
 864 
 865 void
 866 iscsit_global_hold()
 867 {
 868         /*
 869          * To take out a global hold, we must either own the global
 870          * state mutex or we must be running inside of an ioctl that
 871          * has set the global state to ISE_BUSY, ISE_DISABLING, or
 872          * ISE_ENABLING.  We don't track the "owner" for these flags,
 873          * so just checking if they are set is enough for now.
 874          */
 875         ASSERT((iscsit_global.global_svc_state == ISE_ENABLING) ||
 876             (iscsit_global.global_svc_state == ISE_DISABLING) ||
 877             (iscsit_global.global_svc_state == ISE_BUSY) ||
 878             MUTEX_HELD(&iscsit_global.global_state_mutex));
 879 
 880         idm_refcnt_hold(&iscsit_global.global_refcnt);
 881 }
 882 
 883 void
 884 iscsit_global_rele()
 885 {
 886         idm_refcnt_rele(&iscsit_global.global_refcnt);
 887 }
 888 
 889 void
 890 iscsit_global_wait_ref()
 891 {
 892         idm_refcnt_wait_ref(&iscsit_global.global_refcnt);
 893 }
 894 
 895 /*
 896  * IDM callbacks
 897  */
 898 
 899 /*ARGSUSED*/
 900 void
 901 iscsit_rx_pdu(idm_conn_t *ic, idm_pdu_t *rx_pdu)
 902 {
 903         iscsit_conn_t *ict = ic->ic_handle;
 904         switch (IDM_PDU_OPCODE(rx_pdu)) {
 905         case ISCSI_OP_SCSI_CMD:
 906                 ASSERT(0); /* Shouldn't happen */
 907                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
 908                 break;
 909         case ISCSI_OP_SNACK_CMD:
 910                 /*
 911                  * We'll need to handle this when we support ERL1/2.  For
 912                  * now we treat it as a protocol error.
 913                  */
 914                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
 915                 idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
 916                 break;
 917         case ISCSI_OP_SCSI_TASK_MGT_MSG:
 918                 if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
 919                         iscsit_set_cmdsn(ict, rx_pdu);
 920                         iscsit_op_scsi_task_mgmt(ict, rx_pdu);
 921                 }
 922                 break;
 923         case ISCSI_OP_NOOP_OUT:
 924         case ISCSI_OP_LOGIN_CMD:
 925         case ISCSI_OP_TEXT_CMD:
 926         case ISCSI_OP_LOGOUT_CMD:
 927                 /*
 928                  * If/when we switch to userland processing these PDU's
 929                  * will be handled by iscsitd.
 930                  */
 931                 iscsit_deferred_dispatch(rx_pdu);
 932                 break;
 933         default:
 934                 /* Protocol error */
 935                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
 936                 idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
 937                 break;
 938         }
 939 }
 940 
 941 /*ARGSUSED*/
 942 void
 943 iscsit_rx_pdu_error(idm_conn_t *ic, idm_pdu_t *rx_pdu, idm_status_t status)
 944 {
 945         idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
 946 }
 947 
 948 /*
 949  * iscsit_rx_scsi_rsp -- cause the connection to be closed if response rx'd
 950  *
 951  * A target sends an SCSI Response PDU, it should never receive one.
 952  * This has been seen when running the Codemonicon suite of tests which
 953  * does negative testing of the protocol. If such a condition occurs using
 954  * a normal initiator it most likely means there's data corruption in the
 955  * header and that's grounds for dropping the connection as well.
 956  */
 957 void
 958 iscsit_rx_scsi_rsp(idm_conn_t *ic, idm_pdu_t *rx_pdu)
 959 {
 960         idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
 961         idm_conn_event(ic, CE_TRANSPORT_FAIL, NULL);
 962 }
 963 
 964 void
 965 iscsit_task_aborted(idm_task_t *idt, idm_status_t status)
 966 {
 967         iscsit_task_t *itask = idt->idt_private;
 968 
 969         switch (status) {
 970         case IDM_STATUS_SUSPENDED:
 971                 break;
 972         case IDM_STATUS_ABORTED:
 973                 mutex_enter(&itask->it_mutex);
 974                 itask->it_aborted = B_TRUE;
 975                 /*
 976                  * We rely on the fact that STMF tracks outstanding
 977                  * buffer transfers and will free all of our buffers
 978                  * before freeing the task so we don't need to
 979                  * explicitly free the buffers from iscsit/idm
 980                  */
 981                 if (itask->it_stmf_abort) {
 982                         mutex_exit(&itask->it_mutex);
 983                         /*
 984                          * Task is no longer active
 985                          */
 986                         iscsit_task_done(itask);
 987 
 988                         /*
 989                          * STMF has already asked for this task to be aborted
 990                          *
 991                          * STMF specification is wrong... says to return
 992                          * STMF_ABORTED, the code actually looks for
 993                          * STMF_ABORT_SUCCESS.
 994                          */
 995                         stmf_task_lport_aborted_unlocked(itask->it_stmf_task,
 996                             STMF_ABORT_SUCCESS, STMF_IOF_LPORT_DONE);
 997                         return;
 998                 } else {
 999                         mutex_exit(&itask->it_mutex);
1000                         /*
1001                          * Tell STMF to stop processing the task.
1002                          */
1003                         stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
1004                             STMF_ABORTED, NULL);
1005                         return;
1006                 }
1007                 /*NOTREACHED*/
1008         default:
1009                 ASSERT(0);
1010         }
1011 }
1012 
1013 /*ARGSUSED*/
1014 idm_status_t
1015 iscsit_client_notify(idm_conn_t *ic, idm_client_notify_t icn,
1016     uintptr_t data)
1017 {
1018         idm_status_t rc = IDM_STATUS_SUCCESS;
1019 
1020         /*
1021          * IDM client notifications will never occur at interrupt level
1022          * since they are generated from the connection state machine which
1023          * running on taskq threads.
1024          *
1025          */
1026         switch (icn) {
1027         case CN_CONNECT_ACCEPT:
1028                 rc = iscsit_conn_accept(ic); /* No data */
1029                 break;
1030         case CN_FFP_ENABLED:
1031                 rc = iscsit_ffp_enabled(ic); /* No data */
1032                 break;
1033         case CN_FFP_DISABLED:
1034                 /*
1035                  * Data indicates whether this was the result of an
1036                  * explicit logout request.
1037                  */
1038                 rc = iscsit_ffp_disabled(ic, (idm_ffp_disable_t)data);
1039                 break;
1040         case CN_CONNECT_LOST:
1041                 rc = iscsit_conn_lost(ic);
1042                 break;
1043         case CN_CONNECT_DESTROY:
1044                 rc = iscsit_conn_destroy(ic);
1045                 break;
1046         case CN_LOGIN_FAIL:
1047                 /*
1048                  * Force the login state machine to completion
1049                  */
1050                 rc = iscsit_login_fail(ic);
1051                 break;
1052         default:
1053                 rc = IDM_STATUS_REJECT;
1054                 break;
1055         }
1056 
1057         return (rc);
1058 }
1059 
1060 /*
1061  * iscsit_update_statsn is invoked for all the PDUs which have the StatSN
1062  * field in the header. The StatSN is incremented if the IDM_PDU_ADVANCE_STATSN
1063  * flag is set in the pdu flags field. The StatSN is connection-wide and is
1064  * protected by the mutex ict_statsn_mutex. For Data-In PDUs, if the flag
1065  * IDM_TASK_PHASECOLLAPSE_REQ is set, the status (phase-collapse) is also filled
1066  */
1067 void
1068 iscsit_update_statsn(idm_task_t *idm_task, idm_pdu_t *pdu)
1069 {
1070         iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1071         iscsit_conn_t *ict = (iscsit_conn_t *)pdu->isp_ic->ic_handle;
1072         iscsit_task_t *itask = NULL;
1073         scsi_task_t *task = NULL;
1074 
1075         mutex_enter(&ict->ict_statsn_mutex);
1076         rsp->statsn = htonl(ict->ict_statsn);
1077         if (pdu->isp_flags & IDM_PDU_ADVANCE_STATSN)
1078                 ict->ict_statsn++;
1079         mutex_exit(&ict->ict_statsn_mutex);
1080 
1081         /*
1082          * The last SCSI Data PDU passed for a command may also contain the
1083          * status if the status indicates termination with no expections, i.e.
1084          * no sense data or response involved. If the command completes with
1085          * an error, then the response and sense data will be sent in a
1086          * separate iSCSI Response PDU.
1087          */
1088         if ((idm_task) && (idm_task->idt_flags & IDM_TASK_PHASECOLLAPSE_REQ)) {
1089                 itask = idm_task->idt_private;
1090                 task = itask->it_stmf_task;
1091 
1092                 rsp->cmd_status = task->task_scsi_status;
1093                 rsp->flags   |= ISCSI_FLAG_DATA_STATUS;
1094                 if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1095                         rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1096                 } else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1097                         rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1098                 }
1099                 rsp->residual_count = htonl(task->task_resid);
1100 
1101                 /*
1102                  * Removing the task from the session task list
1103                  * just before the status is sent in the last
1104                  * Data PDU transfer
1105                  */
1106                 iscsit_task_done(itask);
1107         }
1108 }
1109 
1110 void
1111 iscsit_build_hdr(idm_task_t *idm_task, idm_pdu_t *pdu, uint8_t opcode)
1112 {
1113         iscsit_task_t *itask = idm_task->idt_private;
1114         iscsi_data_rsp_hdr_t *dh = (iscsi_data_rsp_hdr_t *)pdu->isp_hdr;
1115 
1116         /*
1117          * We acquired iscsit_sess_t.ist_sn_mutex in iscsit_xfer_scsi_data
1118          */
1119         ASSERT(MUTEX_HELD(&itask->it_ict->ict_sess->ist_sn_mutex));
1120         /*
1121          * On incoming data, the target transfer tag and Lun is only
1122          * provided by the target if the A bit is set, Since the target
1123          * does not currently support Error Recovery Level 1, the A
1124          * bit is never set.
1125          */
1126         dh->opcode = opcode;
1127         dh->itt = itask->it_itt;
1128         dh->ttt = ((opcode & ISCSI_OPCODE_MASK) == ISCSI_OP_SCSI_DATA_RSP) ?
1129             ISCSI_RSVD_TASK_TAG : itask->it_ttt;
1130 
1131         dh->expcmdsn = htonl(itask->it_ict->ict_sess->ist_expcmdsn);
1132         dh->maxcmdsn = htonl(itask->it_ict->ict_sess->ist_maxcmdsn);
1133 
1134         /*
1135          * IDM must set:
1136          *
1137          * data.flags and rtt.flags
1138          * data.dlength
1139          * data.datasn
1140          * data.offset
1141          * statsn, residual_count and cmd_status (for phase collapse)
1142          * rtt.rttsn
1143          * rtt.data_offset
1144          * rtt.data_length
1145          */
1146 }
1147 
1148 void
1149 iscsit_keepalive(idm_conn_t *ic)
1150 {
1151         idm_pdu_t               *nop_in_pdu;
1152         iscsi_nop_in_hdr_t      *nop_in;
1153         iscsit_conn_t           *ict = ic->ic_handle;
1154 
1155         /*
1156          * IDM noticed the connection has been idle for too long so it's
1157          * time to provoke some activity.  Build and transmit an iSCSI
1158          * nop-in PDU -- when the initiator responds it will be counted
1159          * as "activity" and keep the connection alive.
1160          *
1161          * We don't actually care about the response here at the iscsit level
1162          * so we will just throw it away without looking at it when it arrives.
1163          */
1164         nop_in_pdu = idm_pdu_alloc(sizeof (*nop_in), 0);
1165         idm_pdu_init(nop_in_pdu, ic, NULL, NULL);
1166         nop_in = (iscsi_nop_in_hdr_t *)nop_in_pdu->isp_hdr;
1167         bzero(nop_in, sizeof (*nop_in));
1168         nop_in->opcode = ISCSI_OP_NOOP_IN;
1169         nop_in->flags = ISCSI_FLAG_FINAL;
1170         nop_in->itt = ISCSI_RSVD_TASK_TAG;
1171         /*
1172          * When the target sends a NOP-In as a Ping, the target transfer tag
1173          * is set to a valid (not reserved) value and the initiator task tag
1174          * is set to ISCSI_RSVD_TASK_TAG (0xffffffff). In this case the StatSN
1175          * will always contain the next sequence number but the StatSN for the
1176          * connection is not advanced after this PDU is sent.
1177          */
1178         nop_in_pdu->isp_flags |= IDM_PDU_SET_STATSN;
1179         /*
1180          * This works because we don't currently allocate ttt's anywhere else
1181          * in iscsit so as long as we stay out of IDM's range we are safe.
1182          * If we need to allocate ttt's for other PDU's in the future this will
1183          * need to be improved.
1184          */
1185         mutex_enter(&ict->ict_mutex);
1186         nop_in->ttt = ict->ict_keepalive_ttt;
1187         ict->ict_keepalive_ttt++;
1188         if (ict->ict_keepalive_ttt == ISCSI_RSVD_TASK_TAG)
1189                 ict->ict_keepalive_ttt = IDM_TASKIDS_MAX;
1190         mutex_exit(&ict->ict_mutex);
1191 
1192         iscsit_pdu_tx(nop_in_pdu);
1193 }
1194 
1195 static idm_status_t
1196 iscsit_conn_accept(idm_conn_t *ic)
1197 {
1198         iscsit_conn_t *ict;
1199 
1200         /*
1201          * We need to get a global hold here to ensure that the service
1202          * doesn't get shutdown prior to establishing a session. This
1203          * gets released in iscsit_conn_destroy().
1204          */
1205         mutex_enter(&iscsit_global.global_state_mutex);
1206         if (iscsit_global.global_svc_state != ISE_ENABLED) {
1207                 mutex_exit(&iscsit_global.global_state_mutex);
1208                 return (IDM_STATUS_FAIL);
1209         }
1210         iscsit_global_hold();
1211         mutex_exit(&iscsit_global.global_state_mutex);
1212 
1213         /*
1214          * Allocate an associated iscsit structure to represent this
1215          * connection.  We shouldn't really create a session until we
1216          * get the first login PDU.
1217          */
1218         ict = kmem_zalloc(sizeof (*ict), KM_SLEEP);
1219 
1220         ict->ict_ic = ic;
1221         ict->ict_statsn = 1;
1222         ict->ict_keepalive_ttt = IDM_TASKIDS_MAX; /* Avoid IDM TT range */
1223         ic->ic_handle = ict;
1224         mutex_init(&ict->ict_mutex, NULL, MUTEX_DRIVER, NULL);
1225         mutex_init(&ict->ict_statsn_mutex, NULL, MUTEX_DRIVER, NULL);
1226         idm_refcnt_init(&ict->ict_refcnt, ict);
1227         idm_refcnt_init(&ict->ict_dispatch_refcnt, ict);
1228 
1229         /*
1230          * Initialize login state machine
1231          */
1232         if (iscsit_login_sm_init(ict) != IDM_STATUS_SUCCESS) {
1233                 iscsit_global_rele();
1234                 /*
1235                  * Cleanup the ict after idm notifies us about this failure
1236                  */
1237                 return (IDM_STATUS_FAIL);
1238         }
1239 
1240         return (IDM_STATUS_SUCCESS);
1241 }
1242 
1243 idm_status_t
1244 iscsit_conn_reinstate(iscsit_conn_t *reinstate_ict, iscsit_conn_t *new_ict)
1245 {
1246         idm_status_t    result;
1247 
1248         /*
1249          * Note in new connection state that this connection is
1250          * reinstating an existing connection.
1251          */
1252         new_ict->ict_reinstating = B_TRUE;
1253         new_ict->ict_reinstate_conn = reinstate_ict;
1254         new_ict->ict_statsn = reinstate_ict->ict_statsn;
1255 
1256         /*
1257          * Now generate connection state machine event to existing connection
1258          * so that it starts the cleanup process.
1259          */
1260         result = idm_conn_reinstate_event(reinstate_ict->ict_ic,
1261             new_ict->ict_ic);
1262 
1263         return (result);
1264 }
1265 
1266 void
1267 iscsit_conn_hold(iscsit_conn_t *ict)
1268 {
1269         idm_refcnt_hold(&ict->ict_refcnt);
1270 }
1271 
1272 void
1273 iscsit_conn_rele(iscsit_conn_t *ict)
1274 {
1275         idm_refcnt_rele(&ict->ict_refcnt);
1276 }
1277 
1278 void
1279 iscsit_conn_dispatch_hold(iscsit_conn_t *ict)
1280 {
1281         idm_refcnt_hold(&ict->ict_dispatch_refcnt);
1282 }
1283 
1284 void
1285 iscsit_conn_dispatch_rele(iscsit_conn_t *ict)
1286 {
1287         idm_refcnt_rele(&ict->ict_dispatch_refcnt);
1288 }
1289 
1290 static idm_status_t
1291 iscsit_login_fail(idm_conn_t *ic)
1292 {
1293         iscsit_conn_t *ict = ic->ic_handle;
1294 
1295         /* Generate login state machine event */
1296         iscsit_login_sm_event(ict, ILE_LOGIN_CONN_ERROR, NULL);
1297 
1298         return (IDM_STATUS_SUCCESS);
1299 }
1300 
1301 static idm_status_t
1302 iscsit_ffp_enabled(idm_conn_t *ic)
1303 {
1304         iscsit_conn_t *ict = ic->ic_handle;
1305 
1306         /* Generate session state machine event */
1307         iscsit_sess_sm_event(ict->ict_sess, SE_CONN_LOGGED_IN, ict);
1308 
1309         return (IDM_STATUS_SUCCESS);
1310 }
1311 
1312 static idm_status_t
1313 iscsit_ffp_disabled(idm_conn_t *ic, idm_ffp_disable_t disable_class)
1314 {
1315         iscsit_conn_t *ict = ic->ic_handle;
1316 
1317         /* Generate session state machine event */
1318         switch (disable_class) {
1319         case FD_CONN_FAIL:
1320                 iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_FAIL, ict);
1321                 break;
1322         case FD_CONN_LOGOUT:
1323                 iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FFP_DISABLE, ict);
1324                 break;
1325         case FD_SESS_LOGOUT:
1326                 iscsit_sess_sm_event(ict->ict_sess, SE_SESSION_CLOSE, ict);
1327                 break;
1328         default:
1329                 ASSERT(0);
1330         }
1331 
1332         return (IDM_STATUS_SUCCESS);
1333 }
1334 
1335 static idm_status_t
1336 iscsit_conn_lost(idm_conn_t *ic)
1337 {
1338         iscsit_conn_t   *ict    = ic->ic_handle;
1339         iscsit_sess_t   *ist    = ict->ict_sess;
1340         iscsit_cbuf_t   *cbuf;
1341         idm_pdu_t       *rx_pdu;
1342         int i;
1343 
1344         mutex_enter(&ict->ict_mutex);
1345         ict->ict_lost = B_TRUE;
1346         mutex_exit(&ict->ict_mutex);
1347         /*
1348          * scrub the staging queue for all PDUs on this connection
1349          */
1350         if (ist != NULL) {
1351                 mutex_enter(&ist->ist_sn_mutex);
1352                 for (cbuf = ist->ist_rxpdu_queue, i = 0;
1353                     ((cbuf->cb_num_elems > 0) && (i < ISCSIT_RXPDU_QUEUE_LEN));
1354                     i++) {
1355                         if (((rx_pdu = cbuf->cb_buffer[i]) != NULL) &&
1356                             (rx_pdu->isp_ic == ic)) {
1357                                 /* conn is lost, drop the pdu */
1358                                 DTRACE_PROBE3(scrubbing__staging__queue,
1359                                     iscsit_sess_t *, ist, idm_conn_t *, ic,
1360                                     idm_pdu_t *, rx_pdu);
1361                                 idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
1362                                 cbuf->cb_buffer[i] = NULL;
1363                                 cbuf->cb_num_elems--;
1364                                 iscsit_conn_dispatch_rele(ict);
1365                         }
1366                 }
1367                 mutex_exit(&ist->ist_sn_mutex);
1368         }
1369         /*
1370          * Make sure there aren't any PDU's transitioning from the receive
1371          * handler to the dispatch taskq.
1372          */
1373         if (idm_refcnt_is_held(&ict->ict_dispatch_refcnt) < 0) {
1374                 cmn_err(CE_WARN, "Possible hang in iscsit_conn_lost");
1375         }
1376         idm_refcnt_wait_ref(&ict->ict_dispatch_refcnt);
1377 
1378         return (IDM_STATUS_SUCCESS);
1379 }
1380 
1381 static idm_status_t
1382 iscsit_conn_destroy(idm_conn_t *ic)
1383 {
1384         iscsit_conn_t *ict = ic->ic_handle;
1385 
1386         mutex_enter(&ict->ict_mutex);
1387         ict->ict_destroyed = B_TRUE;
1388         mutex_exit(&ict->ict_mutex);
1389 
1390         /* Generate session state machine event */
1391         if (ict->ict_sess != NULL) {
1392                 iscsit_sess_sm_event(ict->ict_sess, SE_CONN_FAIL, ict);
1393         }
1394 
1395         idm_refcnt_wait_ref(&ict->ict_dispatch_refcnt);
1396         idm_refcnt_wait_ref(&ict->ict_refcnt);
1397         /*
1398          * The session state machine does not need to post
1399          * events to IDM any longer, so it is safe to set
1400          * the idm connection reference to NULL
1401          */
1402         ict->ict_ic = NULL;
1403 
1404         /* Reap the login state machine */
1405         iscsit_login_sm_fini(ict);
1406 
1407         /* Clean up any text command remnants */
1408         iscsit_text_cmd_fini(ict);
1409 
1410         mutex_destroy(&ict->ict_mutex);
1411         idm_refcnt_destroy(&ict->ict_dispatch_refcnt);
1412         idm_refcnt_destroy(&ict->ict_refcnt);
1413         kmem_free(ict, sizeof (*ict));
1414 
1415         iscsit_global_rele();
1416 
1417         return (IDM_STATUS_SUCCESS);
1418 }
1419 
1420 void
1421 iscsit_conn_logout(iscsit_conn_t *ict)
1422 {
1423         /*
1424          * If the iscsi connection is active, then
1425          * logout the IDM connection by sending a
1426          * CE_LOGOUT_SESSION_SUCCESS, else, no action
1427          * needs to be taken because the connection
1428          * is already in the teardown process.
1429          */
1430         mutex_enter(&ict->ict_mutex);
1431         if (ict->ict_lost == B_FALSE && ict->ict_destroyed == B_FALSE) {
1432                 idm_conn_event(ict->ict_ic, CE_LOGOUT_SESSION_SUCCESS, NULL);
1433         }
1434         mutex_exit(&ict->ict_mutex);
1435 }
1436 
1437 /*
1438  * STMF-related functions
1439  *
1440  * iSCSI to STMF mapping
1441  *
1442  * Session == ?
1443  * Connection == bound to local port but not itself a local port
1444  * Target
1445  * Target portal (group?) == local port (really but we're not going to do this)
1446  *      iscsit needs to map connections to local ports (whatever we decide
1447  *      they are)
1448  * Target == ?
1449  */
1450 
1451 /*ARGSUSED*/
1452 static stmf_data_buf_t *
1453 iscsit_dbuf_alloc(scsi_task_t *task, uint32_t size, uint32_t *pminsize,
1454     uint32_t flags)
1455 {
1456         iscsit_task_t *itask = task->task_port_private;
1457         idm_buf_t *idm_buffer;
1458         iscsit_buf_t    *ibuf;
1459         stmf_data_buf_t *result;
1460         uint32_t        bsize;
1461 
1462         /*
1463          * If the requested size is larger than MaxBurstLength and the
1464          * given pminsize is also larger than MaxBurstLength, then the
1465          * allocation fails (dbuf = NULL) and pminsize is modified to
1466          * be equal to MaxBurstLength. stmf/sbd then should re-invoke
1467          * this function with the corrected values for transfer.
1468          */
1469         ASSERT(pminsize);
1470         if (size <= itask->it_ict->ict_op.op_max_burst_length) {
1471                 bsize = size;
1472         } else if (*pminsize <= itask->it_ict->ict_op.op_max_burst_length) {
1473                 bsize = itask->it_ict->ict_op.op_max_burst_length;
1474         } else {
1475                 *pminsize = itask->it_ict->ict_op.op_max_burst_length;
1476                 return (NULL);
1477         }
1478 
1479         /* Alloc buffer */
1480         idm_buffer = idm_buf_alloc(itask->it_ict->ict_ic, NULL, bsize);
1481         if (idm_buffer != NULL) {
1482                 result = stmf_alloc(STMF_STRUCT_DATA_BUF,
1483                     sizeof (iscsit_buf_t), 0);
1484                 if (result != NULL) {
1485                         /* Fill in stmf_data_buf_t */
1486                         ibuf = result->db_port_private;
1487                         ibuf->ibuf_idm_buf = idm_buffer;
1488                         ibuf->ibuf_stmf_buf = result;
1489                         ibuf->ibuf_is_immed = B_FALSE;
1490                         result->db_flags = DB_DONT_CACHE;
1491                         result->db_buf_size = bsize;
1492                         result->db_data_size = bsize;
1493                         result->db_sglist_length = 1;
1494                         result->db_sglist[0].seg_addr = idm_buffer->idb_buf;
1495                         result->db_sglist[0].seg_length =
1496                             idm_buffer->idb_buflen;
1497                         return (result);
1498                 }
1499 
1500                 /* Couldn't get the stmf_data_buf_t so free the buffer */
1501                 idm_buf_free(idm_buffer);
1502         }
1503 
1504         return (NULL);
1505 }
1506 
1507 /*ARGSUSED*/
1508 static void
1509 iscsit_dbuf_free(stmf_dbuf_store_t *ds, stmf_data_buf_t *dbuf)
1510 {
1511         iscsit_buf_t *ibuf = dbuf->db_port_private;
1512 
1513         if (ibuf->ibuf_is_immed) {
1514                 /*
1515                  * The iscsit_buf_t structure itself will be freed with its
1516                  * associated task.  Here we just need to free the PDU that
1517                  * held the immediate data.
1518                  */
1519                 idm_pdu_complete(ibuf->ibuf_immed_data_pdu, IDM_STATUS_SUCCESS);
1520                 ibuf->ibuf_immed_data_pdu = 0;
1521         } else {
1522                 idm_buf_free(ibuf->ibuf_idm_buf);
1523                 stmf_free(dbuf);
1524         }
1525 }
1526 
1527 /*ARGSUSED*/
1528 stmf_status_t
1529 iscsit_xfer_scsi_data(scsi_task_t *task, stmf_data_buf_t *dbuf,
1530     uint32_t ioflags)
1531 {
1532         iscsit_task_t *iscsit_task = task->task_port_private;
1533         iscsit_sess_t *ict_sess = iscsit_task->it_ict->ict_sess;
1534         iscsit_buf_t *ibuf = dbuf->db_port_private;
1535         int idm_rc;
1536 
1537         /*
1538          * If we are aborting then we can ignore this request
1539          */
1540         if (iscsit_task->it_stmf_abort) {
1541                 return (STMF_SUCCESS);
1542         }
1543 
1544         /*
1545          * If it's not immediate data then start the transfer
1546          */
1547         if (dbuf->db_flags & DB_DIRECTION_TO_RPORT) {
1548                 if (ibuf->ibuf_is_immed)
1549                         return (iscsit_idm_to_stmf(IDM_STATUS_SUCCESS));
1550                 /*
1551                  * The DB_SEND_STATUS_GOOD flag in the STMF data buffer allows
1552                  * the port provider to phase-collapse, i.e. send the status
1553                  * along with the final data PDU for the command. The port
1554                  * provider passes this request to the transport layer by
1555                  * setting a flag IDM_TASK_PHASECOLLAPSE_REQ in the task.
1556                  */
1557                 if (dbuf->db_flags & DB_SEND_STATUS_GOOD)
1558                         iscsit_task->it_idm_task->idt_flags |=
1559                             IDM_TASK_PHASECOLLAPSE_REQ;
1560                 /*
1561                  * IDM will call iscsit_build_hdr so lock now to serialize
1562                  * access to the SN values.  We need to lock here to enforce
1563                  * lock ordering
1564                  */
1565                 mutex_enter(&ict_sess->ist_sn_mutex);
1566                 idm_rc = idm_buf_tx_to_ini(iscsit_task->it_idm_task,
1567                     ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1568                     dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1569                 mutex_exit(&ict_sess->ist_sn_mutex);
1570 
1571                 return (iscsit_idm_to_stmf(idm_rc));
1572         } else if (dbuf->db_flags & DB_DIRECTION_FROM_RPORT) {
1573                 ASSERT(ibuf->ibuf_is_immed == B_FALSE);
1574                 /* Grab the SN lock (see comment above) */
1575                 mutex_enter(&ict_sess->ist_sn_mutex);
1576                 idm_rc = idm_buf_rx_from_ini(iscsit_task->it_idm_task,
1577                     ibuf->ibuf_idm_buf, dbuf->db_relative_offset,
1578                     dbuf->db_data_size, &iscsit_buf_xfer_cb, dbuf);
1579                 mutex_exit(&ict_sess->ist_sn_mutex);
1580 
1581                 return (iscsit_idm_to_stmf(idm_rc));
1582         }
1583 
1584         /* What are we supposed to do if there is no direction? */
1585         return (STMF_INVALID_ARG);
1586 }
1587 
1588 static void
1589 iscsit_buf_xfer_cb(idm_buf_t *idb, idm_status_t status)
1590 {
1591         iscsit_task_t *itask = idb->idb_task_binding->idt_private;
1592         stmf_data_buf_t *dbuf = idb->idb_cb_arg;
1593 
1594         dbuf->db_xfer_status = iscsit_idm_to_stmf(status);
1595 
1596         /*
1597          * If the task has been aborted then we don't need to call STMF
1598          */
1599         if (itask->it_stmf_abort) {
1600                 return;
1601         }
1602 
1603         /*
1604          * For ISCSI over TCP (not iSER), the last SCSI Data PDU passed
1605          * for a successful command contains the status as requested by
1606          * by COMSTAR (via the DB_SEND_STATUS_GOOD flag). But the iSER
1607          * transport does not support phase-collapse. So pretend we are
1608          * COMSTAR and send the status in a separate PDU now.
1609          */
1610         if (idb->idb_task_binding->idt_flags & IDM_TASK_PHASECOLLAPSE_SUCCESS) {
1611                 /*
1612                  * Mark task complete and notify COMSTAR
1613                  * that the status has been sent.
1614                  */
1615                 itask->it_idm_task->idt_state = TASK_COMPLETE;
1616                 stmf_send_status_done(itask->it_stmf_task,
1617                     iscsit_idm_to_stmf(status), STMF_IOF_LPORT_DONE);
1618         } else if ((dbuf->db_flags & DB_SEND_STATUS_GOOD) &&
1619             status == IDM_STATUS_SUCCESS) {
1620 
1621                 /*
1622                  * The iscsi target port provider - for iSER, emulates the
1623                  * DB_SEND_STATUS_GOOD optimization if requested by STMF;
1624                  * it sends the status in a separate PDU after the data
1625                  * transfer. In this case the port provider should first
1626                  * call stmf_data_xfer_done() to mark the transfer complete
1627                  * and then send the status. Although STMF will free the
1628                  * buffer at the time the task is freed, even if the transfer
1629                  * is not marked complete, this behavior makes statistics
1630                  * gathering and task state tracking more difficult than it
1631                  * needs to be.
1632                  */
1633                 stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1634                 if (iscsit_send_scsi_status(itask->it_stmf_task, 0)
1635                     != STMF_SUCCESS) {
1636                         stmf_send_status_done(itask->it_stmf_task,
1637                             STMF_FAILURE, STMF_IOF_LPORT_DONE);
1638                 }
1639         } else {
1640                 stmf_data_xfer_done(itask->it_stmf_task, dbuf, 0);
1641                 /* don't touch dbuf after stmf_data_xfer_done */
1642         }
1643 }
1644 
1645 
1646 /*ARGSUSED*/
1647 stmf_status_t
1648 iscsit_send_scsi_status(scsi_task_t *task, uint32_t ioflags)
1649 {
1650         iscsit_task_t *itask = task->task_port_private;
1651         iscsi_scsi_rsp_hdr_t *rsp;
1652         idm_pdu_t *pdu;
1653         int resp_datalen;
1654 
1655         /*
1656          * If this task is aborted then we don't need to respond.
1657          */
1658         if (itask->it_stmf_abort) {
1659                 return (STMF_SUCCESS);
1660         }
1661 
1662         /*
1663          * If this is a task management status, handle it elsewhere.
1664          */
1665         if (task->task_mgmt_function != TM_NONE) {
1666                 /*
1667                  * Don't wait for the PDU completion to tell STMF
1668                  * the task is done -- it doesn't really matter and
1669                  * it makes life complicated if STMF later asks us to
1670                  * abort the request and we don't know whether the
1671                  * status has been sent or not.
1672                  */
1673                 itask->it_tm_responded = B_TRUE;
1674                 iscsit_send_task_mgmt_resp(itask->it_tm_pdu,
1675                     (task->task_completion_status == STMF_SUCCESS) ?
1676                     SCSI_TCP_TM_RESP_COMPLETE : SCSI_TCP_TM_RESP_FUNC_NOT_SUPP);
1677                 stmf_send_status_done(task, STMF_SUCCESS,
1678                     STMF_IOF_LPORT_DONE);
1679                 return (STMF_SUCCESS);
1680         }
1681 
1682         /*
1683          * Remove the task from the session task list
1684          */
1685         iscsit_task_done(itask);
1686 
1687         /*
1688          * Send status
1689          */
1690         mutex_enter(&itask->it_idm_task->idt_mutex);
1691         if ((itask->it_idm_task->idt_state == TASK_ACTIVE) &&
1692             (task->task_completion_status == STMF_SUCCESS) &&
1693             (task->task_sense_length == 0) &&
1694             (task->task_resid == 0)) {
1695                 itask->it_idm_task->idt_state = TASK_COMPLETE;
1696                 /* PDU callback releases task hold */
1697                 idm_task_hold(itask->it_idm_task);
1698                 mutex_exit(&itask->it_idm_task->idt_mutex);
1699                 /*
1700                  * Fast path.  Cached status PDU's are already
1701                  * initialized.  We just need to fill in
1702                  * connection and task information. StatSN is
1703                  * incremented by 1 for every status sent a
1704                  * connection.
1705                  */
1706                 pdu = kmem_cache_alloc(iscsit_status_pdu_cache, KM_SLEEP);
1707                 pdu->isp_ic = itask->it_ict->ict_ic;
1708                 pdu->isp_private = itask;
1709                 pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1710 
1711                 rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1712                 rsp->itt = itask->it_itt;
1713                 /*
1714                  * ExpDataSN is the number of R2T and Data-In (read)
1715                  * PDUs the target has sent for the SCSI command.
1716                  *
1717                  * Since there is no support for bidirectional transfer
1718                  * yet, either idt_exp_datasn or idt_exp_rttsn, but not
1719                  * both is valid at any time
1720                  */
1721                 rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1722                     htonl(itask->it_idm_task->idt_exp_datasn):
1723                     htonl(itask->it_idm_task->idt_exp_rttsn);
1724                 rsp->cmd_status = task->task_scsi_status;
1725                 iscsit_pdu_tx(pdu);
1726                 return (STMF_SUCCESS);
1727         } else {
1728                 if (itask->it_idm_task->idt_state != TASK_ACTIVE) {
1729                         mutex_exit(&itask->it_idm_task->idt_mutex);
1730                         return (STMF_FAILURE);
1731                 }
1732                 itask->it_idm_task->idt_state = TASK_COMPLETE;
1733                 /* PDU callback releases task hold */
1734                 idm_task_hold(itask->it_idm_task);
1735                 mutex_exit(&itask->it_idm_task->idt_mutex);
1736 
1737                 resp_datalen = (task->task_sense_length == 0) ? 0 :
1738                     (task->task_sense_length + sizeof (uint16_t));
1739 
1740                 pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
1741                 idm_pdu_init(pdu, itask->it_ict->ict_ic, itask,
1742                     iscsit_send_status_done);
1743                 pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
1744 
1745                 rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
1746                 bzero(rsp, sizeof (*rsp));
1747                 rsp->opcode = ISCSI_OP_SCSI_RSP;
1748 
1749                 rsp->flags = ISCSI_FLAG_FINAL;
1750                 if (task->task_status_ctrl & TASK_SCTRL_OVER) {
1751                         rsp->flags |= ISCSI_FLAG_CMD_OVERFLOW;
1752                 } else if (task->task_status_ctrl & TASK_SCTRL_UNDER) {
1753                         rsp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
1754                 }
1755 
1756                 rsp->bi_residual_count = 0;
1757                 rsp->residual_count = htonl(task->task_resid);
1758                 rsp->itt = itask->it_itt;
1759                 rsp->response = ISCSI_STATUS_CMD_COMPLETED;
1760                 rsp->expdatasn = (itask->it_idm_task->idt_exp_datasn != 0) ?
1761                     htonl(itask->it_idm_task->idt_exp_datasn):
1762                     htonl(itask->it_idm_task->idt_exp_rttsn);
1763                 rsp->cmd_status = task->task_scsi_status;
1764                 if (task->task_sense_length != 0) {
1765                         /*
1766                          * Add a byte to provide the sense length in
1767                          * the response
1768                          */
1769                         *(uint16_t *)((void *)pdu->isp_data) =
1770                             htons(task->task_sense_length);
1771                         bcopy(task->task_sense_data,
1772                             (uint8_t *)pdu->isp_data +
1773                             sizeof (uint16_t),
1774                             task->task_sense_length);
1775                         hton24(rsp->dlength, resp_datalen);
1776                 }
1777 
1778                 DTRACE_PROBE5(iscsi__scsi__response,
1779                     iscsit_conn_t *, itask->it_ict,
1780                     uint8_t, rsp->response,
1781                     uint8_t, rsp->cmd_status,
1782                     idm_pdu_t *, pdu,
1783                     scsi_task_t *, task);
1784 
1785                 iscsit_pdu_tx(pdu);
1786 
1787                 return (STMF_SUCCESS);
1788         }
1789 }
1790 
1791 /*ARGSUSED*/
1792 static void
1793 iscsit_send_good_status_done(idm_pdu_t *pdu, idm_status_t status)
1794 {
1795         iscsit_task_t   *itask;
1796         boolean_t       aborted;
1797 
1798         itask = pdu->isp_private;
1799         aborted = itask->it_stmf_abort;
1800 
1801         /*
1802          * After releasing the hold the task may be freed at any time so
1803          * don't touch it.
1804          */
1805         idm_task_rele(itask->it_idm_task);
1806         if (!aborted) {
1807                 stmf_send_status_done(itask->it_stmf_task,
1808                     iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1809         }
1810         kmem_cache_free(iscsit_status_pdu_cache, pdu);
1811 }
1812 
1813 /*ARGSUSED*/
1814 static void
1815 iscsit_send_status_done(idm_pdu_t *pdu, idm_status_t status)
1816 {
1817         iscsit_task_t    *itask;
1818         boolean_t       aborted;
1819 
1820         itask = pdu->isp_private;
1821         aborted = itask->it_stmf_abort;
1822 
1823         /*
1824          * After releasing the hold the task may be freed at any time so
1825          * don't touch it.
1826          */
1827         idm_task_rele(itask->it_idm_task);
1828         if (!aborted) {
1829                 stmf_send_status_done(itask->it_stmf_task,
1830                     iscsit_idm_to_stmf(pdu->isp_status), STMF_IOF_LPORT_DONE);
1831         }
1832         idm_pdu_free(pdu);
1833 }
1834 
1835 
1836 void
1837 iscsit_lport_task_free(scsi_task_t *task)
1838 {
1839         iscsit_task_t *itask = task->task_port_private;
1840 
1841         /* We only call idm_task_start for regular tasks, not task management */
1842         if (task->task_mgmt_function == TM_NONE) {
1843                 idm_task_done(itask->it_idm_task);
1844                 iscsit_task_free(itask);
1845                 return;
1846         } else {
1847                 iscsit_tm_task_free(itask);
1848         }
1849 }
1850 
1851 /*ARGSUSED*/
1852 stmf_status_t
1853 iscsit_abort(stmf_local_port_t *lport, int abort_cmd, void *arg, uint32_t flags)
1854 {
1855         scsi_task_t     *st = (scsi_task_t *)arg;
1856         iscsit_task_t   *iscsit_task;
1857         idm_task_t      *idt;
1858 
1859         /*
1860          * If this is a task management request then there's really not much to
1861          * do.
1862          */
1863         if (st->task_mgmt_function != TM_NONE) {
1864                 return (STMF_ABORT_SUCCESS);
1865         }
1866 
1867         /*
1868          * Regular task, start cleaning up
1869          */
1870         iscsit_task = st->task_port_private;
1871         idt = iscsit_task->it_idm_task;
1872         mutex_enter(&iscsit_task->it_mutex);
1873         iscsit_task->it_stmf_abort = B_TRUE;
1874         if (iscsit_task->it_aborted) {
1875                 mutex_exit(&iscsit_task->it_mutex);
1876                 /*
1877                  * Task is no longer active
1878                  */
1879                 iscsit_task_done(iscsit_task);
1880 
1881                 /*
1882                  * STMF specification is wrong... says to return
1883                  * STMF_ABORTED, the code actually looks for
1884                  * STMF_ABORT_SUCCESS.
1885                  */
1886                 return (STMF_ABORT_SUCCESS);
1887         } else {
1888                 mutex_exit(&iscsit_task->it_mutex);
1889                 /*
1890                  * Call IDM to abort the task.  Due to a variety of
1891                  * circumstances the task may already be in the process of
1892                  * aborting.
1893                  */
1894                 return (idm_task_abort(idt->idt_ic, idt, AT_TASK_MGMT_ABORT));
1895         }
1896 
1897         /*NOTREACHED*/
1898 }
1899 
1900 /*ARGSUSED*/
1901 void
1902 iscsit_ctl(stmf_local_port_t *lport, int cmd, void *arg)
1903 {
1904         iscsit_tgt_t            *iscsit_tgt;
1905 
1906         ASSERT((cmd == STMF_CMD_LPORT_ONLINE) ||
1907             (cmd == STMF_ACK_LPORT_ONLINE_COMPLETE) ||
1908             (cmd == STMF_CMD_LPORT_OFFLINE) ||
1909             (cmd == STMF_ACK_LPORT_OFFLINE_COMPLETE));
1910 
1911         iscsit_tgt = (iscsit_tgt_t *)lport->lport_port_private;
1912 
1913         switch (cmd) {
1914         case STMF_CMD_LPORT_ONLINE:
1915                 iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_REQ);
1916                 break;
1917         case STMF_CMD_LPORT_OFFLINE:
1918                 iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_REQ);
1919                 break;
1920         case STMF_ACK_LPORT_ONLINE_COMPLETE:
1921                 iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_ONLINE_COMPLETE_ACK);
1922                 break;
1923         case STMF_ACK_LPORT_OFFLINE_COMPLETE:
1924                 iscsit_tgt_sm_event(iscsit_tgt, TE_STMF_OFFLINE_COMPLETE_ACK);
1925                 break;
1926 
1927         default:
1928                 break;
1929         }
1930 }
1931 
1932 static stmf_status_t
1933 iscsit_idm_to_stmf(idm_status_t idmrc)
1934 {
1935         switch (idmrc) {
1936         case IDM_STATUS_SUCCESS:
1937                 return (STMF_SUCCESS);
1938         default:
1939                 return (STMF_FAILURE);
1940         }
1941         /*NOTREACHED*/
1942 }
1943 
1944 void
1945 iscsit_op_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1946 {
1947         iscsit_conn_t           *ict = ic->ic_handle;
1948 
1949         if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
1950                 iscsit_post_scsi_cmd(ic, rx_pdu);
1951         }
1952         iscsit_process_pdu_in_queue(ict->ict_sess);
1953 }
1954 
1955 static int
1956 iscsit_validate_idm_pdu(idm_pdu_t *rx_pdu)
1957 {
1958         iscsi_scsi_cmd_hdr_t    *iscsi_scsi =
1959             (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
1960 
1961         if ((iscsi_scsi->scb[0] == SCMD_READ) ||
1962             (iscsi_scsi->scb[0] == SCMD_READ_G1) ||
1963             (iscsi_scsi->scb[0] == SCMD_READ_G4)) {
1964                 if (iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE)
1965                         return (IDM_STATUS_FAIL);
1966         }
1967         return (IDM_STATUS_SUCCESS);
1968 }
1969 
1970 /*
1971  * ISCSI protocol
1972  */
1973 
1974 void
1975 iscsit_post_scsi_cmd(idm_conn_t *ic, idm_pdu_t *rx_pdu)
1976 {
1977         iscsit_conn_t           *ict;
1978         iscsit_task_t           *itask;
1979         scsi_task_t             *task;
1980         iscsit_buf_t            *ibuf;
1981         iscsi_scsi_cmd_hdr_t    *iscsi_scsi =
1982             (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
1983         iscsi_addl_hdr_t        *ahs_hdr;
1984         uint16_t                addl_cdb_len = 0;
1985 
1986         ict = ic->ic_handle;
1987         if (iscsit_validate_idm_pdu(rx_pdu) != IDM_STATUS_SUCCESS) {
1988                 /* Finish processing request */
1989                 iscsit_set_cmdsn(ict, rx_pdu);
1990 
1991                 iscsit_send_direct_scsi_resp(ict, rx_pdu,
1992                     ISCSI_STATUS_CMD_COMPLETED, STATUS_CHECK);
1993                 idm_pdu_complete(rx_pdu, IDM_STATUS_PROTOCOL_ERROR);
1994                 return;
1995         }
1996 
1997         itask = iscsit_task_alloc(ict);
1998         if (itask == NULL) {
1999                 /* Finish processing request */
2000                 iscsit_set_cmdsn(ict, rx_pdu);
2001 
2002                 iscsit_send_direct_scsi_resp(ict, rx_pdu,
2003                     ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
2004                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2005                 return;
2006         }
2007 
2008         /*
2009          * Note CmdSN and ITT in task.  IDM will have already validated this
2010          * request against the connection state so we don't need to check
2011          * that (the connection may have changed state in the meantime but
2012          * we will catch that when we try to send a response)
2013          */
2014         itask->it_cmdsn = ntohl(iscsi_scsi->cmdsn);
2015         itask->it_itt = iscsi_scsi->itt;
2016 
2017         /*
2018          * Check for extended CDB AHS
2019          */
2020         if (iscsi_scsi->hlength > 0) {
2021                 ahs_hdr = (iscsi_addl_hdr_t *)iscsi_scsi;
2022                 addl_cdb_len = ((ahs_hdr->ahs_hlen_hi << 8) |
2023                     ahs_hdr->ahs_hlen_lo) - 1; /* Adjust for reserved byte */
2024                 if (((addl_cdb_len + 4) / sizeof (uint32_t)) >
2025                     iscsi_scsi->hlength) {
2026                         /* Mangled header info, drop it */
2027                         idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2028                         return;
2029                 }
2030         }
2031 
2032         ict = rx_pdu->isp_ic->ic_handle; /* IDM client private */
2033 
2034         /*
2035          * Add task to session list.  This function will also check to
2036          * ensure that the task does not already exist.
2037          */
2038         if (iscsit_task_start(itask) != IDM_STATUS_SUCCESS) {
2039                 /*
2040                  * Task exists, free all resources and reject.  Don't
2041                  * update expcmdsn in this case because RFC 3720 says
2042                  * "The CmdSN of the rejected command PDU (if it is a
2043                  * non-immediate command) MUST NOT be considered received
2044                  * by the target (i.e., a command sequence gap must be
2045                  * assumed for the CmdSN), even though the CmdSN of the
2046                  * rejected command PDU may be reliably ascertained.  Upon
2047                  * receiving the Reject, the initiator MUST plug the CmdSN
2048                  * gap in order to continue to use the session.  The gap
2049                  * may be plugged either by transmitting a command PDU
2050                  * with the same CmdSN, or by aborting the task (see section
2051                  * 6.9 on how an abort may plug a CmdSN gap)." (Section 6.3)
2052                  */
2053                 iscsit_task_free(itask);
2054                 iscsit_send_reject(ict, rx_pdu, ISCSI_REJECT_TASK_IN_PROGRESS);
2055                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2056                 return;
2057         }
2058 
2059         /* Update sequence numbers */
2060         iscsit_set_cmdsn(ict, rx_pdu);
2061 
2062         /*
2063          * Allocate STMF task
2064          */
2065         itask->it_stmf_task = stmf_task_alloc(
2066             itask->it_ict->ict_sess->ist_lport,
2067             itask->it_ict->ict_sess->ist_stmf_sess, iscsi_scsi->lun,
2068             16 + addl_cdb_len, 0);
2069         if (itask->it_stmf_task == NULL) {
2070                 /*
2071                  * Either stmf really couldn't get memory for a task or,
2072                  * more likely, the LU is currently in reset.  Either way
2073                  * we have no choice but to fail the request.
2074                  */
2075                 iscsit_task_done(itask);
2076                 iscsit_task_free(itask);
2077                 iscsit_send_direct_scsi_resp(ict, rx_pdu,
2078                     ISCSI_STATUS_CMD_COMPLETED, STATUS_BUSY);
2079                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2080                 return;
2081         }
2082 
2083         task = itask->it_stmf_task;
2084         task->task_port_private = itask;
2085 
2086         bcopy(iscsi_scsi->lun, task->task_lun_no, sizeof (task->task_lun_no));
2087 
2088         /*
2089          * iSCSI and Comstar use the same values.  Should we rely on this
2090          * or translate them bit-wise?
2091          */
2092 
2093         task->task_flags =
2094             (((iscsi_scsi->flags & ISCSI_FLAG_CMD_READ) ? TF_READ_DATA : 0) |
2095             ((iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ? TF_WRITE_DATA : 0) |
2096             ((rx_pdu->isp_datalen == 0) ? 0 : TF_INITIAL_BURST));
2097 
2098         switch (iscsi_scsi->flags & ISCSI_FLAG_CMD_ATTR_MASK) {
2099         case ISCSI_ATTR_UNTAGGED:
2100                 break;
2101         case ISCSI_ATTR_SIMPLE:
2102                 task->task_additional_flags |= TF_ATTR_SIMPLE_QUEUE;
2103                 break;
2104         case ISCSI_ATTR_ORDERED:
2105                 task->task_additional_flags |= TF_ATTR_ORDERED_QUEUE;
2106                 break;
2107         case ISCSI_ATTR_HEAD_OF_QUEUE:
2108                 task->task_additional_flags |= TF_ATTR_HEAD_OF_QUEUE;
2109                 break;
2110         case ISCSI_ATTR_ACA:
2111                 task->task_additional_flags |= TF_ATTR_ACA;
2112                 break;
2113         default:
2114                 /* Protocol error but just take it, treat as untagged */
2115                 break;
2116         }
2117 
2118 
2119         task->task_additional_flags = 0;
2120         task->task_priority = 0;
2121         task->task_mgmt_function = TM_NONE;
2122 
2123         /*
2124          * This "task_max_nbufs" doesn't map well to BIDI.  We probably need
2125          * parameter for each direction.  "MaxOutstandingR2T" may very well
2126          * be set to one which could prevent us from doing simultaneous
2127          * transfers in each direction.
2128          */
2129         task->task_max_nbufs = (iscsi_scsi->flags & ISCSI_FLAG_CMD_WRITE) ?
2130             ict->ict_op.op_max_outstanding_r2t : STMF_BUFS_MAX;
2131         task->task_cmd_seq_no = ntohl(iscsi_scsi->itt);
2132         task->task_expected_xfer_length = ntohl(iscsi_scsi->data_length);
2133 
2134         /* Copy CDB */
2135         bcopy(iscsi_scsi->scb, task->task_cdb, 16);
2136         if (addl_cdb_len > 0) {
2137                 bcopy(ahs_hdr->ahs_extscb, task->task_cdb + 16, addl_cdb_len);
2138         }
2139 
2140         DTRACE_ISCSI_3(scsi__command, idm_conn_t *, ic,
2141             iscsi_scsi_cmd_hdr_t *, (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr,
2142             scsi_task_t *, task);
2143 
2144         /*
2145          * Copy the transport header into the task handle from the PDU
2146          * handle. The transport header describes this task's remote tagged
2147          * buffer.
2148          */
2149         if (rx_pdu->isp_transport_hdrlen != 0) {
2150                 bcopy(rx_pdu->isp_transport_hdr,
2151                     itask->it_idm_task->idt_transport_hdr,
2152                     rx_pdu->isp_transport_hdrlen);
2153         }
2154 
2155         /*
2156          * Tell IDM about our new active task
2157          */
2158         idm_task_start(itask->it_idm_task, (uintptr_t)itask->it_itt);
2159 
2160         /*
2161          * If we have any immediate data then setup the immediate buffer
2162          * context that comes with the task
2163          */
2164         if (rx_pdu->isp_datalen) {
2165                 ibuf = itask->it_immed_data;
2166                 ibuf->ibuf_immed_data_pdu = rx_pdu;
2167                 ibuf->ibuf_stmf_buf->db_data_size = rx_pdu->isp_datalen;
2168                 ibuf->ibuf_stmf_buf->db_buf_size = rx_pdu->isp_datalen;
2169                 ibuf->ibuf_stmf_buf->db_relative_offset = 0;
2170                 ibuf->ibuf_stmf_buf->db_sglist[0].seg_length =
2171                     rx_pdu->isp_datalen;
2172                 ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr = rx_pdu->isp_data;
2173 
2174                 DTRACE_ISCSI_8(xfer__start, idm_conn_t *, ic,
2175                     uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2176                     uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2177                     uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2178                     uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2179 
2180                 /*
2181                  * For immediate data transfer, there is no callback from
2182                  * stmf to indicate that the initial burst of data is
2183                  * transferred successfully. In some cases, the task can
2184                  * get freed before execution returns from stmf_post_task.
2185                  * Although this xfer-start/done probe accurately tracks
2186                  * the size of the transfer, it does only provide a best
2187                  * effort on the timing of the transfer.
2188                  */
2189                 DTRACE_ISCSI_8(xfer__done, idm_conn_t *, ic,
2190                     uintptr_t, ibuf->ibuf_stmf_buf->db_sglist[0].seg_addr,
2191                     uint32_t, ibuf->ibuf_stmf_buf->db_relative_offset,
2192                     uint64_t, 0, uint32_t, 0, uint32_t, 0, /* no raddr */
2193                     uint32_t, rx_pdu->isp_datalen, int, XFER_BUF_TX_TO_INI);
2194                 stmf_post_task(task, ibuf->ibuf_stmf_buf);
2195         } else {
2196 
2197                 stmf_post_task(task, NULL);
2198                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2199         }
2200 }
2201 
2202 void
2203 iscsit_deferred_dispatch(idm_pdu_t *rx_pdu)
2204 {
2205         iscsit_conn_t *ict = rx_pdu->isp_ic->ic_handle;
2206 
2207         /*
2208          * If this isn't a login packet, we need a session.  Otherwise
2209          * this is a protocol error (perhaps one IDM should've caught?).
2210          */
2211         if (IDM_PDU_OPCODE(rx_pdu) != ISCSI_OP_LOGIN_CMD &&
2212             ict->ict_sess == NULL) {
2213                 DTRACE_PROBE2(iscsi__idm__deferred__no__session,
2214                     iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
2215                 idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2216                 return;
2217         }
2218 
2219         /*
2220          * If the connection has been lost then ignore new PDU's
2221          */
2222         mutex_enter(&ict->ict_mutex);
2223         if (ict->ict_lost) {
2224                 mutex_exit(&ict->ict_mutex);
2225                 idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2226                 return;
2227         }
2228 
2229         /*
2230          * Grab a hold on the connection to prevent it from going away
2231          * between now and when the taskq function is called.
2232          */
2233         iscsit_conn_dispatch_hold(ict);
2234         mutex_exit(&ict->ict_mutex);
2235 
2236         taskq_dispatch_ent(iscsit_global.global_dispatch_taskq,
2237             iscsit_deferred, rx_pdu, 0, &rx_pdu->isp_tqent);
2238 }
2239 
2240 static void
2241 iscsit_deferred(void *rx_pdu_void)
2242 {
2243         idm_pdu_t               *rx_pdu = rx_pdu_void;
2244         idm_conn_t              *ic = rx_pdu->isp_ic;
2245         iscsit_conn_t           *ict = ic->ic_handle;
2246 
2247         /*
2248          * NOP and Task Management Commands can be marked for immediate
2249          * delivery. Commands marked as 'Immediate' are to be considered
2250          * for execution as soon as they arrive on the target. So these
2251          * should not be checked for sequence order and put in a queue.
2252          * The CmdSN is not advanced for Immediate Commands.
2253          */
2254         switch (IDM_PDU_OPCODE(rx_pdu)) {
2255         case ISCSI_OP_NOOP_OUT:
2256                 if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2257                         iscsit_set_cmdsn(ict, rx_pdu);
2258                         iscsit_pdu_op_noop(ict, rx_pdu);
2259                 }
2260                 break;
2261         case ISCSI_OP_LOGIN_CMD:
2262                 iscsit_pdu_op_login_cmd(ict, rx_pdu);
2263                 iscsit_conn_dispatch_rele(ict);
2264                 return;
2265         case ISCSI_OP_TEXT_CMD:
2266                 if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2267                         iscsit_set_cmdsn(ict, rx_pdu);
2268                         iscsit_pdu_op_text_cmd(ict, rx_pdu);
2269                 }
2270                 break;
2271         case ISCSI_OP_LOGOUT_CMD:
2272                 if (iscsit_check_cmdsn_and_queue(rx_pdu)) {
2273                         iscsit_set_cmdsn(ict, rx_pdu);
2274                         iscsit_pdu_op_logout_cmd(ict, rx_pdu);
2275                 }
2276                 break;
2277         default:
2278                 /* Protocol error.  IDM should have caught this */
2279                 idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
2280                 ASSERT(0);
2281                 break;
2282         }
2283         /*
2284          * Check if there are other PDUs in the session staging queue
2285          * waiting to be posted to SCSI layer.
2286          */
2287         iscsit_process_pdu_in_queue(ict->ict_sess);
2288 
2289         iscsit_conn_dispatch_rele(ict);
2290 }
2291 
2292 static void
2293 iscsit_send_direct_scsi_resp(iscsit_conn_t *ict, idm_pdu_t *rx_pdu,
2294     uint8_t response, uint8_t cmd_status)
2295 {
2296         idm_pdu_t                       *rsp_pdu;
2297         idm_conn_t                      *ic;
2298         iscsi_scsi_rsp_hdr_t            *resp;
2299         iscsi_scsi_cmd_hdr_t            *req =
2300             (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2301 
2302         ic = ict->ict_ic;
2303 
2304         rsp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_rsp_hdr_t), 0);
2305         idm_pdu_init(rsp_pdu, ic, NULL, NULL);
2306         /*
2307          * StatSN is incremented by 1 for every response sent on
2308          * a connection except for responses sent as a result of
2309          * a retry or SNACK
2310          */
2311         rsp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2312 
2313         resp = (iscsi_scsi_rsp_hdr_t *)rsp_pdu->isp_hdr;
2314 
2315         resp->opcode = ISCSI_OP_SCSI_RSP;
2316         resp->flags = ISCSI_FLAG_FINAL;
2317         resp->response = response;
2318         resp->cmd_status = cmd_status;
2319         resp->itt = req->itt;
2320         if ((response == ISCSI_STATUS_CMD_COMPLETED) &&
2321             (req->data_length != 0) &&
2322             ((req->flags & ISCSI_FLAG_CMD_READ) ||
2323             (req->flags & ISCSI_FLAG_CMD_WRITE))) {
2324                 resp->flags |= ISCSI_FLAG_CMD_UNDERFLOW;
2325                 resp->residual_count = req->data_length;
2326         }
2327 
2328         DTRACE_PROBE4(iscsi__scsi__direct__response,
2329             iscsit_conn_t *, ict,
2330             uint8_t, resp->response,
2331             uint8_t, resp->cmd_status,
2332             idm_pdu_t *, rsp_pdu);
2333 
2334         iscsit_pdu_tx(rsp_pdu);
2335 }
2336 
2337 void
2338 iscsit_send_task_mgmt_resp(idm_pdu_t *tm_resp_pdu, uint8_t tm_status)
2339 {
2340         iscsi_scsi_task_mgt_rsp_hdr_t   *tm_resp;
2341 
2342         /*
2343          * The target must take note of the last-sent StatSN.
2344          * The StatSN is to be incremented after sending a
2345          * task management response. Digest recovery can only
2346          * work if StatSN is incremented.
2347          */
2348         tm_resp_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2349         tm_resp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2350         tm_resp->response = tm_status;
2351 
2352         DTRACE_PROBE3(iscsi__scsi__tm__response,
2353             iscsit_conn_t *, tm_resp_pdu->isp_ic->ic_handle,
2354             uint8_t, tm_resp->response,
2355             idm_pdu_t *, tm_resp_pdu);
2356         iscsit_pdu_tx(tm_resp_pdu);
2357 }
2358 
2359 void
2360 iscsit_op_scsi_task_mgmt(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2361 {
2362         idm_pdu_t                       *tm_resp_pdu;
2363         iscsit_task_t                   *itask;
2364         iscsit_task_t                   *tm_itask;
2365         scsi_task_t                     *task;
2366         iscsi_scsi_task_mgt_hdr_t       *iscsi_tm =
2367             (iscsi_scsi_task_mgt_hdr_t *)rx_pdu->isp_hdr;
2368         iscsi_scsi_task_mgt_rsp_hdr_t   *iscsi_tm_rsp =
2369             (iscsi_scsi_task_mgt_rsp_hdr_t *)rx_pdu->isp_hdr;
2370         uint32_t                        rtt, cmdsn, refcmdsn;
2371         uint8_t                         tm_func;
2372 
2373         /*
2374          * Setup response PDU (response field will get filled in later)
2375          */
2376         tm_resp_pdu = idm_pdu_alloc(sizeof (iscsi_scsi_task_mgt_rsp_hdr_t), 0);
2377         if (tm_resp_pdu == NULL) {
2378                 /* Can't respond, just drop it */
2379                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2380                 return;
2381         }
2382         idm_pdu_init(tm_resp_pdu, ict->ict_ic, NULL, NULL);
2383         iscsi_tm_rsp = (iscsi_scsi_task_mgt_rsp_hdr_t *)tm_resp_pdu->isp_hdr;
2384         bzero(iscsi_tm_rsp, sizeof (iscsi_scsi_task_mgt_rsp_hdr_t));
2385         iscsi_tm_rsp->opcode = ISCSI_OP_SCSI_TASK_MGT_RSP;
2386         iscsi_tm_rsp->flags = ISCSI_FLAG_FINAL;
2387         iscsi_tm_rsp->itt = rx_pdu->isp_hdr->itt;
2388 
2389         /*
2390          * Figure out what we're being asked to do.
2391          */
2392         DTRACE_PROBE4(iscsi__scsi__tm__request,
2393             iscsit_conn_t *, ict,
2394             uint8_t, (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK),
2395             uint32_t, iscsi_tm->rtt,
2396             idm_pdu_t *, rx_pdu);
2397         switch (iscsi_tm->function & ISCSI_FLAG_TASK_MGMT_FUNCTION_MASK) {
2398         case ISCSI_TM_FUNC_ABORT_TASK:
2399                 /*
2400                  * STMF doesn't currently support the "abort task" task
2401                  * management command although it does support aborting
2402                  * an individual task.  We'll get STMF to abort the task
2403                  * for us but handle the details of the task management
2404                  * command ourselves.
2405                  *
2406                  * Find the task associated with the referenced task tag.
2407                  */
2408                 rtt = iscsi_tm->rtt;
2409                 itask = (iscsit_task_t *)idm_task_find_by_handle(ict->ict_ic,
2410                     (uintptr_t)rtt);
2411 
2412                 if (itask == NULL) {
2413                         cmdsn = ntohl(iscsi_tm->cmdsn);
2414                         refcmdsn = ntohl(iscsi_tm->refcmdsn);
2415 
2416                         /*
2417                          * Task was not found. But the SCSI command could be
2418                          * on the rxpdu wait queue. If RefCmdSN is within
2419                          * the CmdSN window and less than CmdSN of the TM
2420                          * function, return "Function Complete". Otherwise,
2421                          * return "Task Does Not Exist".
2422                          */
2423 
2424                         if (iscsit_cmdsn_in_window(ict, refcmdsn) &&
2425                             iscsit_sna_lt(refcmdsn, cmdsn)) {
2426                                 mutex_enter(&ict->ict_sess->ist_sn_mutex);
2427                                 if (iscsit_remove_pdu_from_queue(
2428                                     ict->ict_sess, refcmdsn)) {
2429                                         iscsit_conn_dispatch_rele(ict);
2430                                 }
2431                                 mutex_exit(&ict->ict_sess->ist_sn_mutex);
2432                                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2433                                     SCSI_TCP_TM_RESP_COMPLETE);
2434                         } else {
2435                                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2436                                     SCSI_TCP_TM_RESP_NO_TASK);
2437                         }
2438                 } else {
2439 
2440                         /*
2441                          * Tell STMF to abort the task.  This will do no harm
2442                          * if the task is already complete.
2443                          */
2444                         stmf_abort(STMF_QUEUE_TASK_ABORT, itask->it_stmf_task,
2445                             STMF_ABORTED, NULL);
2446 
2447                         /*
2448                          * Make sure the task hasn't already completed
2449                          */
2450                         mutex_enter(&itask->it_idm_task->idt_mutex);
2451                         if ((itask->it_idm_task->idt_state == TASK_COMPLETE) ||
2452                             (itask->it_idm_task->idt_state == TASK_IDLE)) {
2453                                 /*
2454                                  * Task is complete, return "Task Does Not
2455                                  * Exist"
2456                                  */
2457                                 mutex_exit(&itask->it_idm_task->idt_mutex);
2458                                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2459                                     SCSI_TCP_TM_RESP_NO_TASK);
2460                         } else {
2461                                 /*
2462                                  * STMF is now aborting the task, return
2463                                  * "Function Complete"
2464                                  */
2465                                 mutex_exit(&itask->it_idm_task->idt_mutex);
2466                                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2467                                     SCSI_TCP_TM_RESP_COMPLETE);
2468                         }
2469                         idm_task_rele(itask->it_idm_task);
2470                 }
2471                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2472                 return;
2473 
2474         case ISCSI_TM_FUNC_ABORT_TASK_SET:
2475                 tm_func = TM_ABORT_TASK_SET;
2476                 break;
2477 
2478         case ISCSI_TM_FUNC_CLEAR_ACA:
2479                 tm_func = TM_CLEAR_ACA;
2480                 break;
2481 
2482         case ISCSI_TM_FUNC_CLEAR_TASK_SET:
2483                 tm_func = TM_CLEAR_TASK_SET;
2484                 break;
2485 
2486         case ISCSI_TM_FUNC_LOGICAL_UNIT_RESET:
2487                 tm_func = TM_LUN_RESET;
2488                 break;
2489 
2490         case ISCSI_TM_FUNC_TARGET_WARM_RESET:
2491                 tm_func = TM_TARGET_WARM_RESET;
2492                 break;
2493 
2494         case ISCSI_TM_FUNC_TARGET_COLD_RESET:
2495                 tm_func = TM_TARGET_COLD_RESET;
2496                 break;
2497 
2498         case ISCSI_TM_FUNC_TASK_REASSIGN:
2499                 /*
2500                  * We do not currently support allegiance reassignment.  When
2501                  * we start supporting ERL1+, we will need to.
2502                  */
2503                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2504                     SCSI_TCP_TM_RESP_NO_ALLG_REASSN);
2505                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2506                 return;
2507 
2508         default:
2509                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2510                     SCSI_TCP_TM_RESP_REJECTED);
2511                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2512                 return;
2513         }
2514 
2515         tm_itask = iscsit_tm_task_alloc(ict);
2516         if (tm_itask == NULL) {
2517                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2518                     SCSI_TCP_TM_RESP_REJECTED);
2519                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2520                 return;
2521         }
2522 
2523 
2524         task = stmf_task_alloc(ict->ict_sess->ist_lport,
2525             ict->ict_sess->ist_stmf_sess, iscsi_tm->lun,
2526             0, STMF_TASK_EXT_NONE);
2527         if (task == NULL) {
2528                 /*
2529                  * If this happens, either the LU is in reset, couldn't
2530                  * get memory, or some other condition in which we simply
2531                  * can't complete this request.  It would be nice to return
2532                  * an error code like "busy" but the closest we have is
2533                  * "rejected".
2534                  */
2535                 iscsit_send_task_mgmt_resp(tm_resp_pdu,
2536                     SCSI_TCP_TM_RESP_REJECTED);
2537                 iscsit_tm_task_free(tm_itask);
2538                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2539                 return;
2540         }
2541 
2542         tm_itask->it_tm_pdu = tm_resp_pdu;
2543         tm_itask->it_stmf_task = task;
2544         task->task_port_private = tm_itask;
2545         task->task_mgmt_function = tm_func;
2546         task->task_additional_flags = TASK_AF_NO_EXPECTED_XFER_LENGTH;
2547         task->task_priority = 0;
2548         task->task_max_nbufs = STMF_BUFS_MAX;
2549         task->task_cmd_seq_no = iscsi_tm->itt;
2550         task->task_expected_xfer_length = 0;
2551 
2552         stmf_post_task(task, NULL);
2553         idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2554 }
2555 
2556 static void
2557 iscsit_pdu_op_noop(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2558 {
2559         iscsi_nop_out_hdr_t *out = (iscsi_nop_out_hdr_t *)rx_pdu->isp_hdr;
2560         iscsi_nop_in_hdr_t *in;
2561         int resp_datalen;
2562         idm_pdu_t *resp;
2563 
2564         /* Ignore the response from initiator */
2565         if ((out->itt == ISCSI_RSVD_TASK_TAG) ||
2566             (out->ttt != ISCSI_RSVD_TASK_TAG)) {
2567                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2568                 return;
2569         }
2570 
2571         /* Allocate a PDU to respond */
2572         resp_datalen = ntoh24(out->dlength);
2573         resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), resp_datalen);
2574         idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2575         if (resp_datalen > 0) {
2576                 bcopy(rx_pdu->isp_data, resp->isp_data, resp_datalen);
2577         }
2578 
2579         /*
2580          * When sending a NOP-In as a response to a NOP-Out from the initiator,
2581          * the target must respond with the same initiator task tag that was
2582          * provided in the NOP-Out request, the target transfer tag must be
2583          * ISCSI_RSVD_TASK_TAG (0xffffffff) and StatSN will contain the next
2584          * status sequence number. The StatSN for the connection is advanced
2585          * after this PDU is sent.
2586          */
2587         in = (iscsi_nop_in_hdr_t *)resp->isp_hdr;
2588         bzero(in, sizeof (*in));
2589         in->opcode = ISCSI_OP_NOOP_IN;
2590         in->flags = ISCSI_FLAG_FINAL;
2591         bcopy(out->lun, in->lun, 8);
2592         in->itt              = out->itt;
2593         in->ttt              = ISCSI_RSVD_TASK_TAG;
2594         hton24(in->dlength, resp_datalen);
2595         resp->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2596         /* Any other field in resp to be set? */
2597         iscsit_pdu_tx(resp);
2598         idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2599 }
2600 
2601 static void
2602 iscsit_pdu_op_login_cmd(iscsit_conn_t   *ict, idm_pdu_t *rx_pdu)
2603 {
2604 
2605         /*
2606          * Submit PDU to login state machine.  State machine will free the
2607          * PDU.
2608          */
2609         iscsit_login_sm_event(ict, ILE_LOGIN_RCV, rx_pdu);
2610 }
2611 
2612 void
2613 iscsit_pdu_op_logout_cmd(iscsit_conn_t  *ict, idm_pdu_t *rx_pdu)
2614 {
2615         iscsi_logout_hdr_t      *logout_req =
2616             (iscsi_logout_hdr_t *)rx_pdu->isp_hdr;
2617         iscsi_logout_rsp_hdr_t  *logout_rsp;
2618         idm_pdu_t *resp;
2619 
2620         /* Allocate a PDU to respond */
2621         resp = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2622         idm_pdu_init(resp, ict->ict_ic, NULL, NULL);
2623         /*
2624          * The StatSN is to be sent to the initiator,
2625          * it is not required to increment the number
2626          * as the connection is terminating.
2627          */
2628         resp->isp_flags |= IDM_PDU_SET_STATSN;
2629         /*
2630          * Logout results in the immediate termination of all tasks except
2631          * if the logout reason is ISCSI_LOGOUT_REASON_RECOVERY.  The
2632          * connection state machine will drive this task cleanup automatically
2633          * so we don't need to handle that here.
2634          */
2635         logout_rsp = (iscsi_logout_rsp_hdr_t *)resp->isp_hdr;
2636         bzero(logout_rsp, sizeof (*logout_rsp));
2637         logout_rsp->opcode = ISCSI_OP_LOGOUT_RSP;
2638         logout_rsp->flags = ISCSI_FLAG_FINAL;
2639         logout_rsp->itt = logout_req->itt;
2640         if ((logout_req->flags & ISCSI_FLAG_LOGOUT_REASON_MASK) >
2641             ISCSI_LOGOUT_REASON_RECOVERY) {
2642                 logout_rsp->response = ISCSI_LOGOUT_RECOVERY_UNSUPPORTED;
2643         } else {
2644                 logout_rsp->response = ISCSI_LOGOUT_SUCCESS;
2645         }
2646 
2647         iscsit_pdu_tx(resp);
2648         idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
2649 }
2650 
2651 /*
2652  * Calculate the number of outstanding commands we can process
2653  */
2654 int
2655 iscsit_cmd_window()
2656 {
2657         /*
2658          * Instead of using a pre-defined constant for the command window,
2659          * it should be made confiurable and dynamic. With MC/S, sequence
2660          * numbers will be used up at a much faster rate than with SC/S.
2661          */
2662         return  (ISCSIT_MAX_WINDOW);
2663 }
2664 
2665 /*
2666  * Set local registers based on incoming PDU
2667  */
2668 void
2669 iscsit_set_cmdsn(iscsit_conn_t *ict, idm_pdu_t *rx_pdu)
2670 {
2671         iscsit_sess_t *ist;
2672         iscsi_scsi_cmd_hdr_t *req;
2673 
2674         ist = ict->ict_sess;
2675 
2676         req = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
2677         if (req->opcode & ISCSI_OP_IMMEDIATE) {
2678                 /* no cmdsn increment for immediate PDUs */
2679                 return;
2680         }
2681 
2682         /* Ensure that the ExpCmdSN advances in an orderly manner */
2683         mutex_enter(&ist->ist_sn_mutex);
2684         ist->ist_expcmdsn = ntohl(req->cmdsn) + 1;
2685         ist->ist_maxcmdsn = ntohl(req->cmdsn) + iscsit_cmd_window();
2686         mutex_exit(&ist->ist_sn_mutex);
2687 }
2688 
2689 /*
2690  * Wrapper funtion, calls iscsi_calc_rspsn and idm_pdu_tx
2691  */
2692 void
2693 iscsit_pdu_tx(idm_pdu_t *pdu)
2694 {
2695         iscsit_conn_t *ict = pdu->isp_ic->ic_handle;
2696         iscsi_scsi_rsp_hdr_t *rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2697         iscsit_sess_t *ist = ict->ict_sess;
2698 
2699         /*
2700          * The command sequence numbers are session-wide and must stay
2701          * consistent across the transfer, so protect the cmdsn with a
2702          * mutex lock on the session. The status sequence number will
2703          * be updated just before the transport layer transmits the PDU.
2704          */
2705 
2706         mutex_enter(&ict->ict_sess->ist_sn_mutex);
2707         /* Set ExpCmdSN and MaxCmdSN */
2708         rsp->maxcmdsn = htonl(ist->ist_maxcmdsn);
2709         rsp->expcmdsn = htonl(ist->ist_expcmdsn);
2710         idm_pdu_tx(pdu);
2711         mutex_exit(&ict->ict_sess->ist_sn_mutex);
2712 }
2713 
2714 /*
2715  * Internal functions
2716  */
2717 
2718 void
2719 iscsit_send_async_event(iscsit_conn_t *ict, uint8_t event)
2720 {
2721         idm_pdu_t               *abt;
2722         iscsi_async_evt_hdr_t   *async_abt;
2723 
2724         /*
2725          * Get a PDU to build the abort request.
2726          */
2727         abt = idm_pdu_alloc(sizeof (iscsi_hdr_t), 0);
2728         if (abt == NULL) {
2729                 idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2730                 return;
2731         }
2732 
2733         /*
2734          * A asynchronous message is sent by the target to request a logout.
2735          * The StatSN for the connection is advanced after the PDU is sent
2736          * to allow for initiator and target state synchronization.
2737          */
2738         idm_pdu_init(abt, ict->ict_ic, NULL, NULL);
2739         abt->isp_datalen = 0;
2740         abt->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2741 
2742         async_abt = (iscsi_async_evt_hdr_t *)abt->isp_hdr;
2743         bzero(async_abt, sizeof (*async_abt));
2744         async_abt->opcode = ISCSI_OP_ASYNC_EVENT;
2745         async_abt->async_event = event;
2746         async_abt->flags = ISCSI_FLAG_FINAL;
2747         async_abt->rsvd4[0] = 0xff;
2748         async_abt->rsvd4[1] = 0xff;
2749         async_abt->rsvd4[2] = 0xff;
2750         async_abt->rsvd4[3] = 0xff;
2751 
2752         switch (event) {
2753         case ISCSI_ASYNC_EVENT_REQUEST_LOGOUT:
2754                 async_abt->param3 = htons(IDM_LOGOUT_SECONDS);
2755                 break;
2756         case ISCSI_ASYNC_EVENT_SCSI_EVENT:
2757         case ISCSI_ASYNC_EVENT_DROPPING_CONNECTION:
2758         case ISCSI_ASYNC_EVENT_DROPPING_ALL_CONNECTIONS:
2759         case ISCSI_ASYNC_EVENT_PARAM_NEGOTIATION:
2760         default:
2761                 ASSERT(0);
2762         }
2763 
2764         iscsit_pdu_tx(abt);
2765 }
2766 
2767 void
2768 iscsit_send_reject(iscsit_conn_t *ict, idm_pdu_t *rejected_pdu, uint8_t reason)
2769 {
2770         idm_pdu_t               *reject_pdu;
2771         iscsi_reject_rsp_hdr_t  *reject;
2772 
2773         /*
2774          * Get a PDU to build the abort request.
2775          */
2776         reject_pdu = idm_pdu_alloc(sizeof (iscsi_hdr_t),
2777             rejected_pdu->isp_hdrlen);
2778         if (reject_pdu == NULL) {
2779                 idm_conn_event(ict->ict_ic, CE_TRANSPORT_FAIL, NULL);
2780                 return;
2781         }
2782         idm_pdu_init(reject_pdu, ict->ict_ic, NULL, NULL);
2783         /* StatSN is advanced after a Reject PDU */
2784         reject_pdu->isp_flags |= IDM_PDU_SET_STATSN | IDM_PDU_ADVANCE_STATSN;
2785         reject_pdu->isp_datalen = rejected_pdu->isp_hdrlen;
2786         bcopy(rejected_pdu->isp_hdr, reject_pdu->isp_data,
2787             rejected_pdu->isp_hdrlen);
2788 
2789         reject = (iscsi_reject_rsp_hdr_t *)reject_pdu->isp_hdr;
2790         bzero(reject, sizeof (*reject));
2791         reject->opcode = ISCSI_OP_REJECT_MSG;
2792         reject->reason = reason;
2793         reject->flags = ISCSI_FLAG_FINAL;
2794         hton24(reject->dlength, rejected_pdu->isp_hdrlen);
2795         reject->must_be_ff[0] = 0xff;
2796         reject->must_be_ff[1] = 0xff;
2797         reject->must_be_ff[2] = 0xff;
2798         reject->must_be_ff[3] = 0xff;
2799 
2800         iscsit_pdu_tx(reject_pdu);
2801 }
2802 
2803 
2804 static iscsit_task_t *
2805 iscsit_task_alloc(iscsit_conn_t *ict)
2806 {
2807         iscsit_task_t *itask;
2808         iscsit_buf_t *immed_ibuf;
2809 
2810         /*
2811          * Possible items to pre-alloc if we cache iscsit_task_t's:
2812          *
2813          * Status PDU w/ sense buffer
2814          * stmf_data_buf_t for immediate data
2815          */
2816         itask = kmem_alloc(sizeof (iscsit_task_t) + sizeof (iscsit_buf_t) +
2817             sizeof (stmf_data_buf_t), KM_NOSLEEP);
2818         if (itask != NULL) {
2819                 mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2820                 itask->it_aborted = itask->it_stmf_abort =
2821                     itask->it_tm_task = 0;
2822 
2823                 immed_ibuf = (iscsit_buf_t *)(itask + 1);
2824                 bzero(immed_ibuf, sizeof (*immed_ibuf));
2825                 immed_ibuf->ibuf_is_immed = B_TRUE;
2826                 immed_ibuf->ibuf_stmf_buf = (stmf_data_buf_t *)(immed_ibuf + 1);
2827 
2828                 bzero(immed_ibuf->ibuf_stmf_buf, sizeof (stmf_data_buf_t));
2829                 immed_ibuf->ibuf_stmf_buf->db_port_private = immed_ibuf;
2830                 immed_ibuf->ibuf_stmf_buf->db_sglist_length = 1;
2831                 immed_ibuf->ibuf_stmf_buf->db_flags = DB_DIRECTION_FROM_RPORT |
2832                     DB_DONT_CACHE;
2833                 itask->it_immed_data = immed_ibuf;
2834                 itask->it_idm_task = idm_task_alloc(ict->ict_ic);
2835                 if (itask->it_idm_task != NULL) {
2836                         itask->it_idm_task->idt_private = itask;
2837                         itask->it_ict = ict;
2838                         itask->it_ttt = itask->it_idm_task->idt_tt;
2839                         return (itask);
2840                 } else {
2841                         kmem_free(itask, sizeof (iscsit_task_t) +
2842                             sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2843                 }
2844         }
2845 
2846         return (NULL);
2847 }
2848 
2849 static void
2850 iscsit_task_free(iscsit_task_t *itask)
2851 {
2852         idm_task_free(itask->it_idm_task);
2853         mutex_destroy(&itask->it_mutex);
2854         kmem_free(itask, sizeof (iscsit_task_t) +
2855             sizeof (iscsit_buf_t) + sizeof (stmf_data_buf_t));
2856 }
2857 
2858 static iscsit_task_t *
2859 iscsit_tm_task_alloc(iscsit_conn_t *ict)
2860 {
2861         iscsit_task_t *itask;
2862 
2863         itask = kmem_zalloc(sizeof (iscsit_task_t), KM_NOSLEEP);
2864         if (itask != NULL) {
2865                 idm_conn_hold(ict->ict_ic);
2866                 mutex_init(&itask->it_mutex, NULL, MUTEX_DRIVER, NULL);
2867                 itask->it_aborted = itask->it_stmf_abort =
2868                     itask->it_tm_responded = 0;
2869                 itask->it_tm_pdu = NULL;
2870                 itask->it_tm_task = 1;
2871                 itask->it_ict = ict;
2872         }
2873 
2874         return (itask);
2875 }
2876 
2877 static void
2878 iscsit_tm_task_free(iscsit_task_t *itask)
2879 {
2880         /*
2881          * If we responded then the call to idm_pdu_complete will free the
2882          * PDU.  Otherwise we got aborted before the TM function could
2883          * complete and we need to free the PDU explicitly.
2884          */
2885         if (itask->it_tm_pdu != NULL && !itask->it_tm_responded)
2886                 idm_pdu_free(itask->it_tm_pdu);
2887         idm_conn_rele(itask->it_ict->ict_ic);
2888         mutex_destroy(&itask->it_mutex);
2889         kmem_free(itask, sizeof (iscsit_task_t));
2890 }
2891 
2892 static idm_status_t
2893 iscsit_task_start(iscsit_task_t *itask)
2894 {
2895         iscsit_sess_t *ist = itask->it_ict->ict_sess;
2896         avl_index_t             where;
2897 
2898         /*
2899          * Sanity check the ITT and ensure that this task does not already
2900          * exist.  If not then add the task to the session task list.
2901          */
2902         mutex_enter(&ist->ist_mutex);
2903         mutex_enter(&itask->it_mutex);
2904         itask->it_active = 1;
2905         if (avl_find(&ist->ist_task_list, itask, &where) == NULL) {
2906                 /* New task, add to AVL */
2907                 avl_insert(&ist->ist_task_list, itask, where);
2908                 mutex_exit(&itask->it_mutex);
2909                 mutex_exit(&ist->ist_mutex);
2910                 return (IDM_STATUS_SUCCESS);
2911         }
2912         mutex_exit(&itask->it_mutex);
2913         mutex_exit(&ist->ist_mutex);
2914 
2915         return (IDM_STATUS_REJECT);
2916 }
2917 
2918 static void
2919 iscsit_task_done(iscsit_task_t *itask)
2920 {
2921         iscsit_sess_t *ist = itask->it_ict->ict_sess;
2922 
2923         mutex_enter(&ist->ist_mutex);
2924         mutex_enter(&itask->it_mutex);
2925         if (itask->it_active) {
2926                 avl_remove(&ist->ist_task_list, itask);
2927                 itask->it_active = 0;
2928         }
2929         mutex_exit(&itask->it_mutex);
2930         mutex_exit(&ist->ist_mutex);
2931 }
2932 
2933 /*
2934  * iscsit status PDU cache
2935  */
2936 
2937 /*ARGSUSED*/
2938 static int
2939 iscsit_status_pdu_constructor(void *pdu_void, void *arg, int flags)
2940 {
2941         idm_pdu_t *pdu = pdu_void;
2942         iscsi_scsi_rsp_hdr_t *rsp;
2943 
2944         bzero(pdu, sizeof (idm_pdu_t));
2945         pdu->isp_callback = iscsit_send_good_status_done;
2946         pdu->isp_magic = IDM_PDU_MAGIC;
2947         pdu->isp_hdr = (iscsi_hdr_t *)(pdu + 1); /* Ptr arithmetic */
2948         pdu->isp_hdrlen = sizeof (iscsi_hdr_t);
2949 
2950         /* Setup status response */
2951         rsp = (iscsi_scsi_rsp_hdr_t *)pdu->isp_hdr;
2952         bzero(rsp, sizeof (*rsp));
2953         rsp->opcode = ISCSI_OP_SCSI_RSP;
2954         rsp->flags = ISCSI_FLAG_FINAL;
2955         rsp->response = ISCSI_STATUS_CMD_COMPLETED;
2956 
2957         return (0);
2958 }
2959 
2960 /*
2961  * iscsit private data handler
2962  */
2963 
2964 /*ARGSUSED*/
2965 static void
2966 iscsit_pp_cb(struct stmf_port_provider *pp, int cmd, void *arg, uint32_t flags)
2967 {
2968         it_config_t             *cfg;
2969         nvlist_t                *nvl;
2970         iscsit_service_enabled_t        old_state;
2971 
2972         if ((cmd != STMF_PROVIDER_DATA_UPDATED) || (arg == NULL)) {
2973                 return;
2974         }
2975 
2976         nvl = (nvlist_t *)arg;
2977 
2978         /* Translate nvlist */
2979         if (it_nv_to_config(nvl, &cfg) != 0) {
2980                 cmn_err(CE_WARN, "Configuration is invalid");
2981                 return;
2982         }
2983 
2984         /* Check that no iSCSI ioctl is currently running */
2985         mutex_enter(&iscsit_global.global_state_mutex);
2986         old_state = iscsit_global.global_svc_state;
2987         switch (iscsit_global.global_svc_state) {
2988         case ISE_ENABLED:
2989         case ISE_DISABLED:
2990                 iscsit_global.global_svc_state = ISE_BUSY;
2991                 break;
2992         case ISE_ENABLING:
2993                 /*
2994                  * It is OK for the iscsit_pp_cb to be called from inside of
2995                  * an iSCSI ioctl only if we are currently executing inside
2996                  * of stmf_register_port_provider.
2997                  */
2998                 ASSERT((flags & STMF_PCB_PREG_COMPLETE) != 0);
2999                 break;
3000         default:
3001                 cmn_err(CE_WARN, "iscsit_pp_cb called when global_svc_state"
3002                     " is not ENABLED(0x%x) -- ignoring",
3003                     iscsit_global.global_svc_state);
3004                 mutex_exit(&iscsit_global.global_state_mutex);
3005                 it_config_free_cmn(cfg);
3006                 return;
3007         }
3008         mutex_exit(&iscsit_global.global_state_mutex);
3009 
3010         /* Update config */
3011         (void) iscsit_config_merge(cfg);
3012 
3013         it_config_free_cmn(cfg);
3014 
3015         /* Restore old iSCSI driver global state */
3016         mutex_enter(&iscsit_global.global_state_mutex);
3017         ASSERT(iscsit_global.global_svc_state == ISE_BUSY ||
3018             iscsit_global.global_svc_state == ISE_ENABLING);
3019         iscsit_global.global_svc_state = old_state;
3020         mutex_exit(&iscsit_global.global_state_mutex);
3021 }
3022 
3023 
3024 static it_cfg_status_t
3025 iscsit_config_merge(it_config_t *in_cfg)
3026 {
3027         it_cfg_status_t status;
3028         it_config_t     *cfg;
3029         it_config_t     tmp_cfg;
3030         list_t          tpg_del_list;
3031 
3032         if (in_cfg) {
3033                 cfg = in_cfg;
3034         } else {
3035                 /* Make empty config */
3036                 bzero(&tmp_cfg, sizeof (tmp_cfg));
3037                 cfg = &tmp_cfg;
3038         }
3039 
3040         list_create(&tpg_del_list,  sizeof (iscsit_tpg_t),
3041             offsetof(iscsit_tpg_t, tpg_delete_ln));
3042 
3043         /*
3044          * Update targets, initiator contexts, target portal groups,
3045          * and iSNS client
3046          */
3047         ISCSIT_GLOBAL_LOCK(RW_WRITER);
3048         if (((status = iscsit_config_merge_tpg(cfg, &tpg_del_list))
3049             != 0) ||
3050             ((status = iscsit_config_merge_tgt(cfg)) != 0) ||
3051             ((status = iscsit_config_merge_ini(cfg)) != 0) ||
3052             ((status = isnst_config_merge(cfg)) != 0)) {
3053                 ISCSIT_GLOBAL_UNLOCK();
3054                 return (status);
3055         }
3056 
3057         /* Update other global config parameters */
3058         if (iscsit_global.global_props) {
3059                 nvlist_free(iscsit_global.global_props);
3060                 iscsit_global.global_props = NULL;
3061         }
3062         if (in_cfg) {
3063                 (void) nvlist_dup(cfg->config_global_properties,
3064                     &iscsit_global.global_props, KM_SLEEP);
3065         }
3066         ISCSIT_GLOBAL_UNLOCK();
3067 
3068         iscsit_config_destroy_tpgs(&tpg_del_list);
3069 
3070         list_destroy(&tpg_del_list);
3071 
3072         return (ITCFG_SUCCESS);
3073 }
3074 
3075 /*
3076  * iscsit_sna_lt[e]
3077  *
3078  * Compare serial numbers using serial number arithmetic as defined in
3079  * RFC 1982.
3080  *
3081  * NOTE: This code is duplicated in the isns server. It ought to be common.
3082  */
3083 
3084 static int
3085 iscsit_sna_lt(uint32_t sn1, uint32_t sn2)
3086 {
3087         return ((sn1 != sn2) &&
3088             (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3089             ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3090 }
3091 
3092 static int
3093 iscsit_sna_lte(uint32_t sn1, uint32_t sn2)
3094 {
3095         return ((sn1 == sn2) ||
3096             (((sn1 < sn2) && ((sn2 - sn1) < ISCSIT_SNA32_CHECK)) ||
3097             ((sn1 > sn2) && ((sn1 - sn2) > ISCSIT_SNA32_CHECK))));
3098 }
3099 
3100 
3101 static boolean_t
3102 iscsit_cmdsn_in_window(iscsit_conn_t *ict, uint32_t cmdsn)
3103 {
3104         iscsit_sess_t   *ist = ict->ict_sess;
3105         int             rval = B_TRUE;
3106 
3107         ist = ict->ict_sess;
3108 
3109         mutex_enter(&ist->ist_sn_mutex);
3110 
3111         /*
3112          * If cmdsn is less than ist_expcmdsn - iscsit_cmd_window() or
3113          * greater than ist_expcmdsn, it's not in the window.
3114          */
3115 
3116         if (iscsit_sna_lt(cmdsn, (ist->ist_expcmdsn - iscsit_cmd_window())) ||
3117             !iscsit_sna_lte(cmdsn, ist->ist_expcmdsn)) {
3118                 rval = B_FALSE;
3119         }
3120 
3121         mutex_exit(&ist->ist_sn_mutex);
3122 
3123         return (rval);
3124 }
3125 
3126 /*
3127  * iscsit_check_cmdsn_and_queue
3128  *
3129  * Independent of the order in which the iSCSI target receives non-immediate
3130  * command PDU across the entire session and any multiple connections within
3131  * the session, the target must deliver the commands to the SCSI layer in
3132  * CmdSN order. So out-of-order non-immediate commands are queued up on a
3133  * session-wide wait queue. Duplicate commands are ignored.
3134  *
3135  * returns B_TRUE for commands which can be executed immediately (are
3136  * non-deferred), B_FALSE for cases where a command was deferred or invalid.
3137  */
3138 static boolean_t
3139 iscsit_check_cmdsn_and_queue(idm_pdu_t *rx_pdu)
3140 {
3141         idm_conn_t              *ic = rx_pdu->isp_ic;
3142         iscsit_conn_t           *ict = ic->ic_handle;
3143         iscsit_sess_t           *ist = ict->ict_sess;
3144         iscsi_scsi_cmd_hdr_t    *hdr = (iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr;
3145 
3146         mutex_enter(&ist->ist_sn_mutex);
3147         if (hdr->opcode & ISCSI_OP_IMMEDIATE) {
3148                 /* do not queue, handle it immediately */
3149                 DTRACE_PROBE2(immediate__cmd, iscsit_sess_t *, ist,
3150                     idm_pdu_t *, rx_pdu);
3151                 mutex_exit(&ist->ist_sn_mutex);
3152                 return (B_TRUE);
3153         }
3154         /*
3155          * See RFC3270 3.1.1.2: non-immediate commands outside of the
3156          * expected window (from expcmdsn to maxcmdsn, inclusive)
3157          * should be silently ignored.
3158          */
3159         if (iscsit_sna_lt(ist->ist_expcmdsn, ntohl(hdr->cmdsn)) &&
3160             iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_maxcmdsn)) {
3161                 /*
3162                  * Out-of-order commands (cmdSN higher than ExpCmdSN)
3163                  * are staged on a fixed-size circular buffer until
3164                  * the missing command is delivered to the SCSI layer.
3165                  * Irrespective of the order of insertion into the
3166                  * staging queue, the commands are processed out of the
3167                  * queue in cmdSN order only.
3168                  */
3169                 rx_pdu->isp_queue_time = gethrtime();
3170                 iscsit_add_pdu_to_queue(ist, rx_pdu);
3171                 mutex_exit(&ist->ist_sn_mutex);
3172                 return (B_FALSE);
3173         } else if (iscsit_sna_lt(ntohl(hdr->cmdsn), ist->ist_expcmdsn) ||
3174             iscsit_sna_lt(ist->ist_maxcmdsn, ntohl(hdr->cmdsn))) {
3175                 /*
3176                  * See above, this command is outside of our acceptable
3177                  * window, we need to discard/complete.
3178                  */
3179                 DTRACE_PROBE3(cmdsn__lt__expcmdsn, iscsit_sess_t *, ist,
3180                     iscsit_conn_t *, ict, idm_pdu_t *, rx_pdu);
3181                 mutex_exit(&ist->ist_sn_mutex);
3182                 idm_pdu_complete(rx_pdu, IDM_STATUS_SUCCESS);
3183                 /*
3184                  * tell our callers that the PDU "finished."
3185                  */
3186                 return (B_FALSE);
3187         } else {
3188                 mutex_exit(&ist->ist_sn_mutex);
3189                 return (B_TRUE);
3190         }
3191 }
3192 
3193 /*
3194  * iscsit_add_pdu_to_queue() adds PDUs into the array indexed by
3195  * their cmdsn value. The length of the array is kept above the
3196  * maximum window size. The window keeps the cmdsn within a range
3197  * such that there are no collisons. e.g. the assumption is that
3198  * the windowing checks make it impossible to receive PDUs that
3199  * index into the same location in the array.
3200  */
3201 static void
3202 iscsit_add_pdu_to_queue(iscsit_sess_t *ist, idm_pdu_t *rx_pdu)
3203 {
3204         iscsit_cbuf_t   *cbuf   = ist->ist_rxpdu_queue;
3205         iscsit_conn_t   *ict    = rx_pdu->isp_ic->ic_handle;
3206         uint32_t        cmdsn   =
3207             ((iscsi_scsi_cmd_hdr_t *)rx_pdu->isp_hdr)->cmdsn;
3208         uint32_t        index;
3209 
3210         ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3211         /*
3212          * If the connection is being torn down, then
3213          * don't add the PDU to the staging queue
3214          */
3215         mutex_enter(&ict->ict_mutex);
3216         if (ict->ict_lost) {
3217                 mutex_exit(&ict->ict_mutex);
3218                 idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3219                 return;
3220         }
3221         iscsit_conn_dispatch_hold(ict);
3222         mutex_exit(&ict->ict_mutex);
3223 
3224         index = ntohl(cmdsn) % ISCSIT_RXPDU_QUEUE_LEN;
3225         /*
3226          * In the normal case, assuming that the Initiator is not
3227          * buggy and that we don't have packet duplication occuring,
3228          * the entry in the array will be NULL.  However, we may have
3229          * received a duplicate PDU with cmdsn > expsn , and in that
3230          * case we just ignore this PDU -- the previously received one
3231          * remains queued for processing.  We need to be careful not
3232          * to leak this one however.
3233          */
3234         if (cbuf->cb_buffer[index] != NULL) {
3235                 idm_pdu_complete(rx_pdu, IDM_STATUS_FAIL);
3236         } else {
3237                 cbuf->cb_buffer[index] = rx_pdu;
3238                 cbuf->cb_num_elems++;
3239         }
3240 }
3241 
3242 static idm_pdu_t *
3243 iscsit_remove_pdu_from_queue(iscsit_sess_t *ist, uint32_t cmdsn)
3244 {
3245         iscsit_cbuf_t   *cbuf   = ist->ist_rxpdu_queue;
3246         idm_pdu_t       *pdu    = NULL;
3247         uint32_t        index;
3248 
3249         ASSERT(MUTEX_HELD(&ist->ist_sn_mutex));
3250         index = cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3251         if ((pdu = cbuf->cb_buffer[index]) != NULL) {
3252                 ASSERT(cmdsn ==
3253                     ntohl(((iscsi_scsi_cmd_hdr_t *)pdu->isp_hdr)->cmdsn));
3254                 cbuf->cb_buffer[index] = NULL;
3255                 cbuf->cb_num_elems--;
3256                 return (pdu);
3257         }
3258         return (NULL);
3259 }
3260 
3261 /*
3262  * iscsit_process_pdu_in_queue() finds the next pdu in sequence
3263  * and posts it to the SCSI layer
3264  */
3265 static void
3266 iscsit_process_pdu_in_queue(iscsit_sess_t *ist)
3267 {
3268         iscsit_cbuf_t   *cbuf   = ist->ist_rxpdu_queue;
3269         idm_pdu_t       *pdu = NULL;
3270         uint32_t        expcmdsn;
3271 
3272         for (;;) {
3273                 mutex_enter(&ist->ist_sn_mutex);
3274                 if (cbuf->cb_num_elems == 0) {
3275                         mutex_exit(&ist->ist_sn_mutex);
3276                         break;
3277                 }
3278                 expcmdsn = ist->ist_expcmdsn;
3279                 if ((pdu = iscsit_remove_pdu_from_queue(ist, expcmdsn))
3280                     == NULL) {
3281                         mutex_exit(&ist->ist_sn_mutex);
3282                         break;
3283                 }
3284                 mutex_exit(&ist->ist_sn_mutex);
3285                 iscsit_post_staged_pdu(pdu);
3286         }
3287 }
3288 
3289 static void
3290 iscsit_post_staged_pdu(idm_pdu_t *rx_pdu)
3291 {
3292         iscsit_conn_t   *ict    = rx_pdu->isp_ic->ic_handle;
3293 
3294         /* Post the PDU to the SCSI layer */
3295         switch (IDM_PDU_OPCODE(rx_pdu)) {
3296         case ISCSI_OP_NOOP_OUT:
3297                 iscsit_set_cmdsn(ict, rx_pdu);
3298                 iscsit_pdu_op_noop(ict, rx_pdu);
3299                 break;
3300         case ISCSI_OP_TEXT_CMD:
3301                 iscsit_set_cmdsn(ict, rx_pdu);
3302                 iscsit_pdu_op_text_cmd(ict, rx_pdu);
3303                 break;
3304         case ISCSI_OP_SCSI_TASK_MGT_MSG:
3305                 iscsit_set_cmdsn(ict, rx_pdu);
3306                 iscsit_op_scsi_task_mgmt(ict, rx_pdu);
3307                 break;
3308         case ISCSI_OP_SCSI_CMD:
3309                 /* cmdSN will be incremented after creating itask */
3310                 iscsit_post_scsi_cmd(rx_pdu->isp_ic, rx_pdu);
3311                 break;
3312         case ISCSI_OP_LOGOUT_CMD:
3313                 iscsit_set_cmdsn(ict, rx_pdu);
3314                 iscsit_pdu_op_logout_cmd(ict, rx_pdu);
3315                 break;
3316         default:
3317                 /* No other PDUs should be placed on the queue */
3318                 ASSERT(0);
3319         }
3320         iscsit_conn_dispatch_rele(ict); /* release hold on the conn */
3321 }
3322 
3323 /* ARGSUSED */
3324 void
3325 iscsit_rxpdu_queue_monitor_start(void)
3326 {
3327         mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3328         if (iscsit_rxpdu_queue_monitor_thr_running) {
3329                 mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3330                 return;
3331         }
3332         iscsit_rxpdu_queue_monitor_thr_id =
3333             thread_create(NULL, 0, iscsit_rxpdu_queue_monitor, NULL,
3334             0, &p0, TS_RUN, minclsyspri);
3335         while (!iscsit_rxpdu_queue_monitor_thr_running) {
3336                 cv_wait(&iscsit_rxpdu_queue_monitor_cv,
3337                     &iscsit_rxpdu_queue_monitor_mutex);
3338         }
3339         mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3340 
3341 }
3342 
3343 /* ARGSUSED */
3344 void
3345 iscsit_rxpdu_queue_monitor_stop(void)
3346 {
3347         mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3348         if (iscsit_rxpdu_queue_monitor_thr_running) {
3349                 iscsit_rxpdu_queue_monitor_thr_running = B_FALSE;
3350                 cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3351                 mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3352 
3353                 thread_join(iscsit_rxpdu_queue_monitor_thr_did);
3354                 return;
3355         }
3356         mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3357 }
3358 
3359 /*
3360  * A separate thread is used to scan the staging queue on all the
3361  * sessions, If a delayed PDU does not arrive within a timeout, the
3362  * target will advance to the staged PDU that is next in sequence
3363  * and exceeded the threshold wait time. It is up to the initiator
3364  * to note that the target has not acknowledged a particular cmdsn
3365  * and take appropriate action.
3366  */
3367 /* ARGSUSED */
3368 static void
3369 iscsit_rxpdu_queue_monitor(void *arg)
3370 {
3371         iscsit_tgt_t    *tgt;
3372         iscsit_sess_t   *ist;
3373 
3374         mutex_enter(&iscsit_rxpdu_queue_monitor_mutex);
3375         iscsit_rxpdu_queue_monitor_thr_did = curthread->t_did;
3376         iscsit_rxpdu_queue_monitor_thr_running = B_TRUE;
3377         cv_signal(&iscsit_rxpdu_queue_monitor_cv);
3378 
3379         while (iscsit_rxpdu_queue_monitor_thr_running) {
3380                 ISCSIT_GLOBAL_LOCK(RW_READER);
3381                 for (tgt = avl_first(&iscsit_global.global_target_list);
3382                     tgt != NULL;
3383                     tgt = AVL_NEXT(&iscsit_global.global_target_list, tgt)) {
3384                         mutex_enter(&tgt->target_mutex);
3385                         for (ist = avl_first(&tgt->target_sess_list);
3386                             ist != NULL;
3387                             ist = AVL_NEXT(&tgt->target_sess_list, ist)) {
3388 
3389                                 iscsit_rxpdu_queue_monitor_session(ist);
3390                         }
3391                         mutex_exit(&tgt->target_mutex);
3392                 }
3393                 ISCSIT_GLOBAL_UNLOCK();
3394                 if (iscsit_rxpdu_queue_monitor_thr_running == B_FALSE) {
3395                         break;
3396                 }
3397                 (void) cv_reltimedwait(&iscsit_rxpdu_queue_monitor_cv,
3398                     &iscsit_rxpdu_queue_monitor_mutex,
3399                     ISCSIT_RXPDU_QUEUE_MONITOR_INTERVAL * drv_usectohz(1000000),
3400                     TR_CLOCK_TICK);
3401         }
3402         mutex_exit(&iscsit_rxpdu_queue_monitor_mutex);
3403         thread_exit();
3404 }
3405 
3406 static void
3407 iscsit_rxpdu_queue_monitor_session(iscsit_sess_t *ist)
3408 {
3409         iscsit_cbuf_t   *cbuf   = ist->ist_rxpdu_queue;
3410         idm_pdu_t       *next_pdu = NULL;
3411         uint32_t        index, next_cmdsn, i;
3412 
3413         /*
3414          * Assume that all PDUs in the staging queue have a cmdsn >= expcmdsn.
3415          * Starting with the expcmdsn, iterate over the staged PDUs to find
3416          * the next PDU with a wait time greater than the threshold. If found
3417          * advance the staged PDU to the SCSI layer, skipping over the missing
3418          * PDU(s) to get past the hole in the command sequence. It is up to
3419          * the initiator to note that the target has not acknowledged a cmdsn
3420          * and take appropriate action.
3421          *
3422          * Since the PDU(s) arrive in any random order, it is possible that
3423          * that the actual wait time for a particular PDU is much longer than
3424          * the defined threshold. e.g. Consider a case where commands are sent
3425          * over 4 different connections, and cmdsn = 1004 arrives first, then
3426          * 1003, and 1002 and 1001 are lost due to a connection failure.
3427          * So now 1003 is waiting for 1002 to be delivered, and although the
3428          * wait time of 1004 > wait time of 1003, only 1003 will be considered
3429          * by the monitor thread. 1004 will be automatically processed by
3430          * iscsit_process_pdu_in_queue() once the scan is complete and the
3431          * expcmdsn becomes current.
3432          */
3433         mutex_enter(&ist->ist_sn_mutex);
3434         cbuf = ist->ist_rxpdu_queue;
3435         if (cbuf->cb_num_elems == 0) {
3436                 mutex_exit(&ist->ist_sn_mutex);
3437                 return;
3438         }
3439         for (next_pdu = NULL, i = 0; ; i++) {
3440                 next_cmdsn = ist->ist_expcmdsn + i; /* start at expcmdsn */
3441                 index = next_cmdsn % ISCSIT_RXPDU_QUEUE_LEN;
3442                 if ((next_pdu = cbuf->cb_buffer[index]) != NULL) {
3443                         /*
3444                          * If the PDU wait time has not exceeded threshold
3445                          * stop scanning the staging queue until the timer
3446                          * fires again
3447                          */
3448                         if ((gethrtime() - next_pdu->isp_queue_time)
3449                             < (rxpdu_queue_threshold * NANOSEC)) {
3450                                 mutex_exit(&ist->ist_sn_mutex);
3451                                 return;
3452                         }
3453                         /*
3454                          * Remove the next PDU from the queue and post it
3455                          * to the SCSI layer, skipping over the missing
3456                          * PDU. Stop scanning the staging queue until
3457                          * the monitor timer fires again
3458                          */
3459                         (void) iscsit_remove_pdu_from_queue(ist, next_cmdsn);
3460                         mutex_exit(&ist->ist_sn_mutex);
3461                         DTRACE_PROBE3(advanced__to__blocked__cmdsn,
3462                             iscsit_sess_t *, ist, idm_pdu_t *, next_pdu,
3463                             uint32_t, next_cmdsn);
3464                         iscsit_post_staged_pdu(next_pdu);
3465                         /* Deliver any subsequent PDUs immediately */
3466                         iscsit_process_pdu_in_queue(ist);
3467                         return;
3468                 }
3469                 /*
3470                  * Skipping over i PDUs, e.g. a case where commands 1001 and
3471                  * 1002 are lost in the network, skip over both and post 1003
3472                  * expcmdsn then becomes 1004 at the end of the scan.
3473                  */
3474                 DTRACE_PROBE2(skipping__over__cmdsn, iscsit_sess_t *, ist,
3475                     uint32_t, next_cmdsn);
3476         }
3477         /*
3478          * following the assumption, staged cmdsn >= expcmdsn, this statement
3479          * is never reached.
3480          */
3481 }