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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 #pragma ident "%Z%%M% %I% %E% SMI" 26 27 /* 28 * Multiplexed I/O SCSI vHCI implementation 29 */ 30 31 #include <sys/conf.h> 32 #include <sys/file.h> 33 #include <sys/ddi.h> 34 #include <sys/sunddi.h> 35 #include <sys/scsi/scsi.h> 36 #include <sys/scsi/impl/scsi_reset_notify.h> 37 #include <sys/sunmdi.h> 38 #include <sys/mdi_impldefs.h> 39 #include <sys/scsi/adapters/scsi_vhci.h> 40 #include <sys/disp.h> 41 #include <sys/byteorder.h> 42 43 extern uintptr_t scsi_callback_id; 44 extern ddi_dma_attr_t scsi_alloc_attr; 45 46 #ifdef DEBUG 47 int vhci_debug = VHCI_DEBUG_DEFAULT_VAL; 48 #endif 49 50 /* retry for the vhci_do_prout command when a not ready is returned */ 51 int vhci_prout_not_ready_retry = 180; 52 53 /* 54 * These values are defined to support the internal retry of 55 * SCSI packets for better sense code handling. 56 */ 57 #define VHCI_CMD_CMPLT 0 58 #define VHCI_CMD_RETRY 1 59 #define VHCI_CMD_ERROR -1 60 61 #define PROPFLAGS (DDI_PROP_DONTPASS | DDI_PROP_NOTPROM) 62 #define VHCI_SCSI_PERR 0x47 63 #define VHCI_PGR_ILLEGALOP -2 64 #define VHCI_NUM_UPDATE_TASKQ 8 65 /* changed to 132 to accomodate HDS */ 66 #define VHCI_STD_INQ_SIZE 132 67 68 /* 69 * Version Macros 70 */ 71 #define VHCI_NAME_VERSION "SCSI VHCI Driver %I%" 72 char vhci_version_name[] = VHCI_NAME_VERSION; 73 74 int vhci_first_time = 0; 75 clock_t vhci_to_ticks = 0; 76 int vhci_init_wait_timeout = VHCI_INIT_WAIT_TIMEOUT; 77 kcondvar_t vhci_cv; 78 kmutex_t vhci_global_mutex; 79 void *vhci_softstate = NULL; /* for soft state */ 80 81 /* 82 * Flag to delay the retry of the reserve command 83 */ 84 int vhci_reserve_delay = 100000; 85 static int vhci_path_quiesce_timeout = 60; 86 static uchar_t zero_key[MHIOC_RESV_KEY_SIZE]; 87 88 /* uscsi delay for a TRAN_BUSY */ 89 static int vhci_uscsi_delay = 100000; 90 static int vhci_uscsi_retry_count = 180; 91 /* uscsi_restart_sense timeout id in case it needs to get canceled */ 92 static timeout_id_t vhci_restart_timeid = 0; 93 94 /* 95 * Bidirectional map of 'target-port' to port id <pid> for support of 96 * iostat(1M) '-Xx' and '-Yx' output. 97 */ 98 static kmutex_t vhci_targetmap_mutex; 99 static uint_t vhci_targetmap_pid = 1; 100 static mod_hash_t *vhci_targetmap_bypid; /* <pid> -> 'target-port' */ 101 static mod_hash_t *vhci_targetmap_byport; /* 'target-port' -> <pid> */ 102 103 /* 104 * functions exported by scsi_vhci struct cb_ops 105 */ 106 static int vhci_open(dev_t *, int, int, cred_t *); 107 static int vhci_close(dev_t, int, int, cred_t *); 108 static int vhci_ioctl(dev_t, int, intptr_t, int, cred_t *, int *); 109 110 /* 111 * functions exported by scsi_vhci struct dev_ops 112 */ 113 static int vhci_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **); 114 static int vhci_attach(dev_info_t *, ddi_attach_cmd_t); 115 static int vhci_detach(dev_info_t *, ddi_detach_cmd_t); 116 117 /* 118 * functions exported by scsi_vhci scsi_hba_tran_t transport table 119 */ 120 static int vhci_scsi_tgt_init(dev_info_t *, dev_info_t *, 121 scsi_hba_tran_t *, struct scsi_device *); 122 static void vhci_scsi_tgt_free(dev_info_t *, dev_info_t *, scsi_hba_tran_t *, 123 struct scsi_device *); 124 static int vhci_pgr_register_start(scsi_vhci_lun_t *, struct scsi_pkt *); 125 static int vhci_scsi_start(struct scsi_address *, struct scsi_pkt *); 126 static int vhci_scsi_abort(struct scsi_address *, struct scsi_pkt *); 127 static int vhci_scsi_reset(struct scsi_address *, int); 128 static int vhci_scsi_reset_target(struct scsi_address *, int level, 129 uint8_t select_path); 130 static int vhci_scsi_reset_bus(struct scsi_address *); 131 static int vhci_scsi_getcap(struct scsi_address *, char *, int); 132 static int vhci_scsi_setcap(struct scsi_address *, char *, int, int); 133 static int vhci_commoncap(struct scsi_address *, char *, int, int, int); 134 static int vhci_pHCI_cap(struct scsi_address *ap, char *cap, int val, int whom, 135 mdi_pathinfo_t *pip); 136 static struct scsi_pkt *vhci_scsi_init_pkt(struct scsi_address *, 137 struct scsi_pkt *, struct buf *, int, int, int, int, int (*)(), caddr_t); 138 static void vhci_scsi_destroy_pkt(struct scsi_address *, struct scsi_pkt *); 139 static void vhci_scsi_dmafree(struct scsi_address *, struct scsi_pkt *); 140 static void vhci_scsi_sync_pkt(struct scsi_address *, struct scsi_pkt *); 141 static int vhci_scsi_reset_notify(struct scsi_address *, int, void (*)(caddr_t), 142 caddr_t); 143 static int vhci_scsi_get_bus_addr(struct scsi_device *, char *, int); 144 static int vhci_scsi_get_name(struct scsi_device *, char *, int); 145 static int vhci_scsi_bus_power(dev_info_t *, void *, pm_bus_power_op_t, 146 void *, void *); 147 static int vhci_scsi_bus_config(dev_info_t *, uint_t, ddi_bus_config_op_t, 148 void *, dev_info_t **); 149 150 /* 151 * functions registered with the mpxio framework via mdi_vhci_ops_t 152 */ 153 static int vhci_pathinfo_init(dev_info_t *, mdi_pathinfo_t *, int); 154 static int vhci_pathinfo_uninit(dev_info_t *, mdi_pathinfo_t *, int); 155 static int vhci_pathinfo_state_change(dev_info_t *, mdi_pathinfo_t *, 156 mdi_pathinfo_state_t, uint32_t, int); 157 static int vhci_pathinfo_online(dev_info_t *, mdi_pathinfo_t *, int); 158 static int vhci_pathinfo_offline(dev_info_t *, mdi_pathinfo_t *, int); 159 static int vhci_failover(dev_info_t *, dev_info_t *, int); 160 static void vhci_client_attached(dev_info_t *); 161 162 static int vhci_ctl(dev_t, int, intptr_t, int, cred_t *, int *); 163 static int vhci_devctl(dev_t, int, intptr_t, int, cred_t *, int *); 164 static int vhci_ioc_get_phci_path(sv_iocdata_t *, caddr_t, int, caddr_t); 165 static int vhci_ioc_get_client_path(sv_iocdata_t *, caddr_t, int, caddr_t); 166 static int vhci_ioc_get_paddr(sv_iocdata_t *, caddr_t, int, caddr_t); 167 static int vhci_ioc_send_client_path(caddr_t, sv_iocdata_t *, int, caddr_t); 168 static void vhci_ioc_devi_to_path(dev_info_t *, caddr_t); 169 static int vhci_get_phci_path_list(dev_info_t *, sv_path_info_t *, uint_t); 170 static int vhci_get_client_path_list(dev_info_t *, sv_path_info_t *, uint_t); 171 static int vhci_get_iocdata(const void *, sv_iocdata_t *, int, caddr_t); 172 static int vhci_get_iocswitchdata(const void *, sv_switch_to_cntlr_iocdata_t *, 173 int, caddr_t); 174 static int vhci_ioc_alloc_pathinfo(sv_path_info_t **, sv_path_info_t **, 175 uint_t, sv_iocdata_t *, int, caddr_t); 176 static void vhci_ioc_free_pathinfo(sv_path_info_t *, sv_path_info_t *, uint_t); 177 static int vhci_ioc_send_pathinfo(sv_path_info_t *, sv_path_info_t *, uint_t, 178 sv_iocdata_t *, int, caddr_t); 179 static int vhci_handle_ext_fo(struct scsi_pkt *, int); 180 static int vhci_efo_watch_cb(caddr_t, struct scsi_watch_result *); 181 static int vhci_quiesce_lun(struct scsi_vhci_lun *); 182 static int vhci_pgr_validate_and_register(scsi_vhci_priv_t *); 183 static void vhci_dispatch_scsi_start(void *); 184 static void vhci_efo_done(void *); 185 static void vhci_initiate_auto_failback(void *); 186 static void vhci_update_pHCI_pkt(struct vhci_pkt *, struct scsi_pkt *); 187 static int vhci_update_pathinfo(struct scsi_device *, mdi_pathinfo_t *, 188 struct scsi_failover_ops *, scsi_vhci_lun_t *, struct scsi_vhci *); 189 static void vhci_kstat_create_pathinfo(mdi_pathinfo_t *); 190 static int vhci_quiesce_paths(dev_info_t *, dev_info_t *, 191 scsi_vhci_lun_t *, char *, char *); 192 193 static char *vhci_devnm_to_guid(char *); 194 static int vhci_bind_transport(struct scsi_address *, struct vhci_pkt *, 195 int, int (*func)(caddr_t)); 196 static void vhci_intr(struct scsi_pkt *); 197 static int vhci_do_prout(scsi_vhci_priv_t *); 198 static void vhci_run_cmd(void *); 199 static int vhci_do_prin(struct vhci_pkt **); 200 static struct scsi_pkt *vhci_create_retry_pkt(struct vhci_pkt *); 201 static struct vhci_pkt *vhci_sync_retry_pkt(struct vhci_pkt *); 202 static struct scsi_vhci_lun *vhci_lun_lookup(dev_info_t *); 203 static struct scsi_vhci_lun *vhci_lun_lookup_alloc(dev_info_t *, char *, int *); 204 static void vhci_lun_free(dev_info_t *); 205 static int vhci_recovery_reset(scsi_vhci_lun_t *, struct scsi_address *, 206 uint8_t, uint8_t); 207 void vhci_update_pathstates(void *); 208 209 #ifdef DEBUG 210 static void vhci_print_prin_keys(vhci_prin_readkeys_t *, int); 211 #endif 212 static void vhci_print_prout_keys(scsi_vhci_lun_t *, char *); 213 static void vhci_uscsi_iodone(struct scsi_pkt *pkt); 214 215 /* 216 * MP-API related functions 217 */ 218 extern int vhci_mpapi_init(struct scsi_vhci *); 219 extern void vhci_mpapi_add_dev_prod(struct scsi_vhci *, char *); 220 extern int vhci_mpapi_ctl(dev_t, int, intptr_t, int, cred_t *, int *); 221 extern void vhci_update_mpapi_data(struct scsi_vhci *, 222 scsi_vhci_lun_t *, mdi_pathinfo_t *); 223 extern void* vhci_get_mpapi_item(struct scsi_vhci *, mpapi_list_header_t *, 224 uint8_t, void*); 225 extern void vhci_mpapi_set_path_state(dev_info_t *, mdi_pathinfo_t *, int); 226 extern int vhci_mpapi_update_tpg_acc_state_for_lu(struct scsi_vhci *, 227 scsi_vhci_lun_t *); 228 229 /* Special export to MP-API of tpgs non-'fops' entry point */ 230 int (*tpgs_set_target_groups)(struct scsi_address *, int, int); 231 232 #define VHCI_DMA_MAX_XFER_CAP 0xffffffffULL 233 234 #define VHCI_MAX_PGR_RETRIES 3 235 236 /* 237 * Macros for the device-type mpxio options 238 */ 239 #define LOAD_BALANCE_OPTIONS "load-balance-options" 240 #define LOGICAL_BLOCK_REGION_SIZE "region-size" 241 #define MPXIO_OPTIONS_LIST "device-type-mpxio-options-list" 242 #define DEVICE_TYPE_STR "device-type" 243 #define isdigit(ch) ((ch) >= '0' && (ch) <= '9') 244 245 static struct cb_ops vhci_cb_ops = { 246 vhci_open, /* open */ 247 vhci_close, /* close */ 248 nodev, /* strategy */ 249 nodev, /* print */ 250 nodev, /* dump */ 251 nodev, /* read */ 252 nodev, /* write */ 253 vhci_ioctl, /* ioctl */ 254 nodev, /* devmap */ 255 nodev, /* mmap */ 256 nodev, /* segmap */ 257 nochpoll, /* chpoll */ 258 ddi_prop_op, /* cb_prop_op */ 259 0, /* streamtab */ 260 D_NEW | D_MP, /* cb_flag */ 261 CB_REV, /* rev */ 262 nodev, /* aread */ 263 nodev /* awrite */ 264 }; 265 266 static struct dev_ops vhci_ops = { 267 DEVO_REV, 268 0, 269 vhci_getinfo, 270 nulldev, /* identify */ 271 nulldev, /* probe */ 272 vhci_attach, /* attach and detach are mandatory */ 273 vhci_detach, 274 nodev, /* reset */ 275 &vhci_cb_ops, /* cb_ops */ 276 NULL, /* bus_ops */ 277 NULL, /* power */ 278 }; 279 280 extern struct mod_ops mod_driverops; 281 282 static struct modldrv modldrv = { 283 &mod_driverops, 284 vhci_version_name, /* module name */ 285 &vhci_ops 286 }; 287 288 static struct modlinkage modlinkage = { 289 MODREV_1, 290 &modldrv, 291 NULL 292 }; 293 294 static mdi_vhci_ops_t vhci_opinfo = { 295 MDI_VHCI_OPS_REV, 296 vhci_pathinfo_init, /* Pathinfo node init callback */ 297 vhci_pathinfo_uninit, /* Pathinfo uninit callback */ 298 vhci_pathinfo_state_change, /* Pathinfo node state change */ 299 vhci_failover, /* failover callback */ 300 vhci_client_attached /* client attached callback */ 301 }; 302 303 /* 304 * The scsi_failover table defines an ordered set of 'fops' modules supported 305 * by scsi_vhci. Currently, initialize this table from the 'ddi-forceload' 306 * property specified in scsi_vhci.conf. 307 */ 308 struct scsi_failover { 309 ddi_modhandle_t sf_mod; 310 struct scsi_failover_ops *sf_sfo; 311 } *scsi_failover_table; 312 uint_t scsi_nfailover; 313 314 int 315 _init(void) 316 { 317 int rval; 318 319 /* 320 * Allocate soft state and prepare to do ddi_soft_state_zalloc() 321 * before registering with the transport first. 322 */ 323 if ((rval = ddi_soft_state_init(&vhci_softstate, 324 sizeof (struct scsi_vhci), 1)) != 0) { 325 VHCI_DEBUG(1, (CE_NOTE, NULL, 326 "!_init:soft state init failed\n")); 327 return (rval); 328 } 329 330 if ((rval = scsi_hba_init(&modlinkage)) != 0) { 331 VHCI_DEBUG(1, (CE_NOTE, NULL, 332 "!_init: scsi hba init failed\n")); 333 ddi_soft_state_fini(&vhci_softstate); 334 return (rval); 335 } 336 337 mutex_init(&vhci_global_mutex, NULL, MUTEX_DRIVER, NULL); 338 cv_init(&vhci_cv, NULL, CV_DRIVER, NULL); 339 340 mutex_init(&vhci_targetmap_mutex, NULL, MUTEX_DRIVER, NULL); 341 vhci_targetmap_byport = mod_hash_create_strhash( 342 "vhci_targetmap_byport", 256, mod_hash_null_valdtor); 343 vhci_targetmap_bypid = mod_hash_create_idhash( 344 "vhci_targetmap_bypid", 256, mod_hash_null_valdtor); 345 346 if ((rval = mod_install(&modlinkage)) != 0) { 347 VHCI_DEBUG(1, (CE_NOTE, NULL, "!_init: mod_install failed\n")); 348 if (vhci_targetmap_bypid) 349 mod_hash_destroy_idhash(vhci_targetmap_bypid); 350 if (vhci_targetmap_byport) 351 mod_hash_destroy_strhash(vhci_targetmap_byport); 352 mutex_destroy(&vhci_targetmap_mutex); 353 cv_destroy(&vhci_cv); 354 mutex_destroy(&vhci_global_mutex); 355 scsi_hba_fini(&modlinkage); 356 ddi_soft_state_fini(&vhci_softstate); 357 } 358 return (rval); 359 } 360 361 362 /* 363 * the system is done with us as a driver, so clean up 364 */ 365 int 366 _fini(void) 367 { 368 int rval; 369 370 /* 371 * don't start cleaning up until we know that the module remove 372 * has worked -- if this works, then we know that each instance 373 * has successfully been DDI_DETACHed 374 */ 375 if ((rval = mod_remove(&modlinkage)) != 0) { 376 VHCI_DEBUG(4, (CE_NOTE, NULL, "!_fini: mod_remove failed\n")); 377 return (rval); 378 } 379 380 if (vhci_targetmap_bypid) 381 mod_hash_destroy_idhash(vhci_targetmap_bypid); 382 if (vhci_targetmap_byport) 383 mod_hash_destroy_strhash(vhci_targetmap_byport); 384 mutex_destroy(&vhci_targetmap_mutex); 385 cv_destroy(&vhci_cv); 386 mutex_destroy(&vhci_global_mutex); 387 scsi_hba_fini(&modlinkage); 388 ddi_soft_state_fini(&vhci_softstate); 389 390 return (rval); 391 } 392 393 int 394 _info(struct modinfo *modinfop) 395 { 396 return (mod_info(&modlinkage, modinfop)); 397 } 398 399 /* 400 * Lookup scsi_failover by "short name" of failover module. 401 */ 402 struct scsi_failover_ops * 403 vhci_failover_ops_by_name(char *name) 404 { 405 struct scsi_failover *sf; 406 407 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 408 if (sf->sf_sfo == NULL) 409 continue; 410 if (strcmp(sf->sf_sfo->sfo_name, name) == 0) 411 return (sf->sf_sfo); 412 } 413 return (NULL); 414 } 415 416 /* 417 * Load all scsi_failover_ops 'fops' modules. 418 */ 419 static void 420 vhci_failover_modopen(struct scsi_vhci *vhci) 421 { 422 char **module; 423 int i; 424 struct scsi_failover *sf; 425 char **dt; 426 int e; 427 428 if (scsi_failover_table) 429 return; 430 431 /* Get the list of modules from scsi_vhci.conf */ 432 if (ddi_prop_lookup_string_array(DDI_DEV_T_ANY, 433 vhci->vhci_dip, DDI_PROP_DONTPASS, "ddi-forceload", 434 &module, &scsi_nfailover) != DDI_PROP_SUCCESS) { 435 cmn_err(CE_WARN, "scsi_vhci: " 436 "scsi_vhci.conf is missing 'ddi-forceload'"); 437 return; 438 } 439 if (scsi_nfailover == 0) { 440 cmn_err(CE_WARN, "scsi_vhci: " 441 "scsi_vhci.conf has empty 'ddi-forceload'"); 442 ddi_prop_free(module); 443 return; 444 } 445 446 /* allocate failover table based on number of modules */ 447 scsi_failover_table = (struct scsi_failover *) 448 kmem_zalloc(sizeof (struct scsi_failover) * (scsi_nfailover + 1), 449 KM_SLEEP); 450 451 /* loop over modules specified in scsi_vhci.conf and open each module */ 452 for (i = 0, sf = scsi_failover_table; i < scsi_nfailover; i++) { 453 if (module[i] == NULL) 454 continue; 455 456 sf->sf_mod = ddi_modopen(module[i], KRTLD_MODE_FIRST, &e); 457 if (sf->sf_mod == NULL) { 458 /* 459 * A module returns EEXIST if other software is 460 * supporting the intended function: for example 461 * the scsi_vhci_f_sum_emc module returns EEXIST 462 * from _init if EMC powerpath software is installed. 463 */ 464 if (e != EEXIST) 465 cmn_err(CE_WARN, "scsi_vhci: unable to open " 466 "module '%s', error %d", module[i], e); 467 continue; 468 } 469 sf->sf_sfo = ddi_modsym(sf->sf_mod, 470 "scsi_vhci_failover_ops", &e); 471 if (sf->sf_sfo == NULL) { 472 cmn_err(CE_WARN, "scsi_vhci: " 473 "unable to import 'scsi_failover_ops' from '%s', " 474 "error %d", module[i], e); 475 (void) ddi_modclose(sf->sf_mod); 476 sf->sf_mod = NULL; 477 continue; 478 } 479 480 /* register vid/pid of devices supported with mpapi */ 481 for (dt = sf->sf_sfo->sfo_devices; *dt; dt++) 482 vhci_mpapi_add_dev_prod(vhci, *dt); 483 484 /* 485 * Special processing for SFO_NAME_TPGS module, which contains 486 * the `tpgs_set_target_groups` implementation needed by the 487 * MP-API code. 488 */ 489 if (strcmp(sf->sf_sfo->sfo_name, SFO_NAME_TPGS) == 0) { 490 tpgs_set_target_groups = 491 (int (*)(struct scsi_address *, int, int)) 492 ddi_modsym(sf->sf_mod, "std_set_target_groups", &e); 493 if (tpgs_set_target_groups == NULL) { 494 cmn_err(CE_WARN, "scsi_vhci: " 495 "unable to import 'std_set_target_groups' " 496 "from '%s', error %d", module[i], e); 497 } 498 } 499 500 sf++; 501 } 502 503 /* verify that at least the "well-known" modules were there */ 504 if (vhci_failover_ops_by_name(SFO_NAME_SYM) == NULL) 505 cmn_err(CE_WARN, "scsi_vhci: well-known module \"" 506 SFO_NAME_SYM "\" not defined in scsi_vhci.conf's " 507 "'ddi-forceload'"); 508 if (vhci_failover_ops_by_name(SFO_NAME_TPGS) == NULL) 509 cmn_err(CE_WARN, "scsi_vhci: well-known module \"" 510 SFO_NAME_TPGS "\" not defined in scsi_vhci.conf's " 511 "'ddi-forceload'"); 512 513 /* call sfo_init for modules that need it */ 514 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 515 if (sf->sf_sfo && sf->sf_sfo->sfo_init) 516 (*sf->sf_sfo->sfo_init)(); 517 } 518 519 ddi_prop_free(module); 520 } 521 522 /* 523 * unload all loaded scsi_failover_ops modules 524 */ 525 static void 526 vhci_failover_modclose() 527 { 528 struct scsi_failover *sf; 529 530 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 531 if ((sf->sf_mod == NULL) || (sf->sf_sfo == NULL)) 532 continue; 533 (void) ddi_modclose(sf->sf_mod); 534 sf->sf_mod = NULL; 535 sf->sf_sfo = NULL; 536 } 537 538 if (scsi_failover_table && scsi_nfailover) 539 kmem_free(scsi_failover_table, 540 sizeof (struct scsi_failover) * (scsi_nfailover + 1)); 541 scsi_failover_table = NULL; 542 scsi_nfailover = 0; 543 } 544 545 /* ARGSUSED */ 546 static int 547 vhci_open(dev_t *devp, int flag, int otype, cred_t *credp) 548 { 549 struct scsi_vhci *vhci; 550 551 if (otype != OTYP_CHR) { 552 return (EINVAL); 553 } 554 555 vhci = ddi_get_soft_state(vhci_softstate, MINOR2INST(getminor(*devp))); 556 if (vhci == NULL) { 557 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_open: failed ENXIO\n")); 558 return (ENXIO); 559 } 560 561 mutex_enter(&vhci->vhci_mutex); 562 if ((flag & FEXCL) && (vhci->vhci_state & VHCI_STATE_OPEN)) { 563 mutex_exit(&vhci->vhci_mutex); 564 vhci_log(CE_NOTE, vhci->vhci_dip, 565 "!vhci%d: Already open\n", getminor(*devp)); 566 return (EBUSY); 567 } 568 569 vhci->vhci_state |= VHCI_STATE_OPEN; 570 mutex_exit(&vhci->vhci_mutex); 571 return (0); 572 } 573 574 575 /* ARGSUSED */ 576 static int 577 vhci_close(dev_t dev, int flag, int otype, cred_t *credp) 578 { 579 struct scsi_vhci *vhci; 580 581 if (otype != OTYP_CHR) { 582 return (EINVAL); 583 } 584 585 vhci = ddi_get_soft_state(vhci_softstate, MINOR2INST(getminor(dev))); 586 if (vhci == NULL) { 587 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_close: failed ENXIO\n")); 588 return (ENXIO); 589 } 590 591 mutex_enter(&vhci->vhci_mutex); 592 vhci->vhci_state &= ~VHCI_STATE_OPEN; 593 mutex_exit(&vhci->vhci_mutex); 594 595 return (0); 596 } 597 598 /* ARGSUSED */ 599 static int 600 vhci_ioctl(dev_t dev, int cmd, intptr_t data, int mode, 601 cred_t *credp, int *rval) 602 { 603 if (IS_DEVCTL(cmd)) { 604 return (vhci_devctl(dev, cmd, data, mode, credp, rval)); 605 } else if (cmd == MP_CMD) { 606 return (vhci_mpapi_ctl(dev, cmd, data, mode, credp, rval)); 607 } else { 608 return (vhci_ctl(dev, cmd, data, mode, credp, rval)); 609 } 610 } 611 612 /* 613 * attach the module 614 */ 615 static int 616 vhci_attach(dev_info_t *dip, ddi_attach_cmd_t cmd) 617 { 618 int rval = DDI_FAILURE; 619 int scsi_hba_attached = 0; 620 int vhci_attached = 0; 621 int mutex_initted = 0; 622 int instance; 623 struct scsi_vhci *vhci; 624 scsi_hba_tran_t *tran; 625 char cache_name_buf[64]; 626 char *data; 627 628 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_attach: cmd=0x%x\n", cmd)); 629 630 instance = ddi_get_instance(dip); 631 632 switch (cmd) { 633 case DDI_ATTACH: 634 break; 635 636 case DDI_RESUME: 637 case DDI_PM_RESUME: 638 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_attach: resume not yet" 639 "implemented\n")); 640 return (rval); 641 642 default: 643 VHCI_DEBUG(1, (CE_NOTE, NULL, 644 "!vhci_attach: unknown ddi command\n")); 645 return (rval); 646 } 647 648 /* 649 * Allocate vhci data structure. 650 */ 651 if (ddi_soft_state_zalloc(vhci_softstate, instance) != DDI_SUCCESS) { 652 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 653 "soft state alloc failed\n")); 654 return (DDI_FAILURE); 655 } 656 657 if ((vhci = ddi_get_soft_state(vhci_softstate, instance)) == NULL) { 658 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 659 "bad soft state\n")); 660 ddi_soft_state_free(vhci_softstate, instance); 661 return (DDI_FAILURE); 662 } 663 664 /* Allocate packet cache */ 665 (void) snprintf(cache_name_buf, sizeof (cache_name_buf), 666 "vhci%d_cache", instance); 667 668 mutex_init(&vhci->vhci_mutex, NULL, MUTEX_DRIVER, NULL); 669 mutex_initted++; 670 671 /* 672 * Allocate a transport structure 673 */ 674 tran = scsi_hba_tran_alloc(dip, SCSI_HBA_CANSLEEP); 675 ASSERT(tran != NULL); 676 677 vhci->vhci_tran = tran; 678 vhci->vhci_dip = dip; 679 vhci->vhci_instance = instance; 680 681 tran->tran_hba_private = vhci; 682 tran->tran_tgt_private = NULL; 683 tran->tran_tgt_init = vhci_scsi_tgt_init; 684 tran->tran_tgt_probe = NULL; 685 tran->tran_tgt_free = vhci_scsi_tgt_free; 686 687 tran->tran_start = vhci_scsi_start; 688 tran->tran_abort = vhci_scsi_abort; 689 tran->tran_reset = vhci_scsi_reset; 690 tran->tran_getcap = vhci_scsi_getcap; 691 tran->tran_setcap = vhci_scsi_setcap; 692 tran->tran_init_pkt = vhci_scsi_init_pkt; 693 tran->tran_destroy_pkt = vhci_scsi_destroy_pkt; 694 tran->tran_dmafree = vhci_scsi_dmafree; 695 tran->tran_sync_pkt = vhci_scsi_sync_pkt; 696 tran->tran_reset_notify = vhci_scsi_reset_notify; 697 698 tran->tran_get_bus_addr = vhci_scsi_get_bus_addr; 699 tran->tran_get_name = vhci_scsi_get_name; 700 tran->tran_bus_reset = NULL; 701 tran->tran_quiesce = NULL; 702 tran->tran_unquiesce = NULL; 703 704 /* 705 * register event notification routines with scsa 706 */ 707 tran->tran_get_eventcookie = NULL; 708 tran->tran_add_eventcall = NULL; 709 tran->tran_remove_eventcall = NULL; 710 tran->tran_post_event = NULL; 711 712 tran->tran_bus_power = vhci_scsi_bus_power; 713 714 tran->tran_bus_config = vhci_scsi_bus_config; 715 716 /* 717 * Attach this instance with the mpxio framework 718 */ 719 if (mdi_vhci_register(MDI_HCI_CLASS_SCSI, dip, &vhci_opinfo, 0) 720 != MDI_SUCCESS) { 721 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 722 "mdi_vhci_register failed\n")); 723 goto attach_fail; 724 } 725 vhci_attached++; 726 727 /* 728 * Attach this instance of the hba. 729 * 730 * Regarding dma attributes: Since scsi_vhci is a virtual scsi HBA 731 * driver, it has nothing to do with DMA. However, when calling 732 * scsi_hba_attach_setup() we need to pass something valid in the 733 * dma attributes parameter. So we just use scsi_alloc_attr. 734 * SCSA itself seems to care only for dma_attr_minxfer and 735 * dma_attr_burstsizes fields of dma attributes structure. 736 * It expects those fileds to be non-zero. 737 */ 738 if (scsi_hba_attach_setup(dip, &scsi_alloc_attr, tran, 739 SCSI_HBA_TRAN_CLONE) != DDI_SUCCESS) { 740 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 741 "hba attach failed\n")); 742 goto attach_fail; 743 } 744 scsi_hba_attached++; 745 746 if (ddi_create_minor_node(dip, "devctl", S_IFCHR, 747 INST2DEVCTL(instance), DDI_NT_SCSI_NEXUS, 0) != DDI_SUCCESS) { 748 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 749 " ddi_create_minor_node failed\n")); 750 goto attach_fail; 751 } 752 753 /* 754 * Set pm-want-child-notification property for 755 * power management of the phci and client 756 */ 757 if (ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP, 758 "pm-want-child-notification?", NULL, NULL) != DDI_PROP_SUCCESS) { 759 cmn_err(CE_WARN, 760 "%s%d fail to create pm-want-child-notification? prop", 761 ddi_driver_name(dip), ddi_get_instance(dip)); 762 goto attach_fail; 763 } 764 765 vhci->vhci_taskq = taskq_create("vhci_taskq", 1, MINCLSYSPRI, 1, 4, 0); 766 vhci->vhci_update_pathstates_taskq = 767 taskq_create("vhci_update_pathstates", VHCI_NUM_UPDATE_TASKQ, 768 MINCLSYSPRI, 1, 4, 0); 769 ASSERT(vhci->vhci_taskq); 770 ASSERT(vhci->vhci_update_pathstates_taskq); 771 772 /* 773 * Set appropriate configuration flags based on options set in 774 * conf file. 775 */ 776 vhci->vhci_conf_flags = 0; 777 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, dip, PROPFLAGS, 778 "auto-failback", &data) == DDI_SUCCESS) { 779 if (strcmp(data, "enable") == 0) 780 vhci->vhci_conf_flags |= VHCI_CONF_FLAGS_AUTO_FAILBACK; 781 ddi_prop_free(data); 782 } 783 784 if (!(vhci->vhci_conf_flags & VHCI_CONF_FLAGS_AUTO_FAILBACK)) 785 vhci_log(CE_NOTE, dip, "!Auto-failback capability " 786 "disabled through scsi_vhci.conf file."); 787 788 /* 789 * Allocate an mpapi private structure 790 */ 791 vhci->mp_priv = kmem_zalloc(sizeof (mpapi_priv_t), KM_SLEEP); 792 if (vhci_mpapi_init(vhci) != 0) { 793 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_attach: " 794 "vhci_mpapi_init() failed")); 795 } 796 797 vhci_failover_modopen(vhci); /* load failover modules */ 798 799 ddi_report_dev(dip); 800 return (DDI_SUCCESS); 801 802 attach_fail: 803 if (vhci_attached) 804 (void) mdi_vhci_unregister(dip, 0); 805 806 if (scsi_hba_attached) 807 (void) scsi_hba_detach(dip); 808 809 if (vhci->vhci_tran) 810 scsi_hba_tran_free(vhci->vhci_tran); 811 812 if (mutex_initted) { 813 mutex_destroy(&vhci->vhci_mutex); 814 } 815 816 ddi_soft_state_free(vhci_softstate, instance); 817 return (DDI_FAILURE); 818 } 819 820 821 /*ARGSUSED*/ 822 static int 823 vhci_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 824 { 825 int instance = ddi_get_instance(dip); 826 scsi_hba_tran_t *tran; 827 struct scsi_vhci *vhci; 828 829 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_detach: cmd=0x%x\n", cmd)); 830 831 if ((tran = ddi_get_driver_private(dip)) == NULL) 832 return (DDI_FAILURE); 833 834 vhci = TRAN2HBAPRIVATE(tran); 835 if (!vhci) { 836 return (DDI_FAILURE); 837 } 838 839 switch (cmd) { 840 case DDI_DETACH: 841 break; 842 843 case DDI_SUSPEND: 844 case DDI_PM_SUSPEND: 845 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_detach: suspend/pm not yet" 846 "implemented\n")); 847 return (DDI_FAILURE); 848 849 default: 850 VHCI_DEBUG(1, (CE_NOTE, NULL, 851 "!vhci_detach: unknown ddi command\n")); 852 return (DDI_FAILURE); 853 } 854 855 (void) mdi_vhci_unregister(dip, 0); 856 (void) scsi_hba_detach(dip); 857 scsi_hba_tran_free(tran); 858 859 if (ddi_prop_remove(DDI_DEV_T_NONE, dip, 860 "pm-want-child-notification?") != DDI_PROP_SUCCESS) { 861 cmn_err(CE_WARN, 862 "%s%d unable to remove prop pm-want_child_notification?", 863 ddi_driver_name(dip), ddi_get_instance(dip)); 864 } 865 if (vhci_restart_timeid != 0) { 866 (void) untimeout(vhci_restart_timeid); 867 } 868 vhci_restart_timeid = 0; 869 870 mutex_destroy(&vhci->vhci_mutex); 871 vhci->vhci_dip = NULL; 872 vhci->vhci_tran = NULL; 873 taskq_destroy(vhci->vhci_taskq); 874 taskq_destroy(vhci->vhci_update_pathstates_taskq); 875 ddi_remove_minor_node(dip, NULL); 876 ddi_soft_state_free(vhci_softstate, instance); 877 878 vhci_failover_modclose(); /* unload failover modules */ 879 return (DDI_SUCCESS); 880 } 881 882 /* 883 * vhci_getinfo() 884 * Given the device number, return the devinfo pointer or the 885 * instance number. 886 * Note: always succeed DDI_INFO_DEVT2INSTANCE, even before attach. 887 */ 888 889 /*ARGSUSED*/ 890 static int 891 vhci_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result) 892 { 893 struct scsi_vhci *vhcip; 894 int instance = MINOR2INST(getminor((dev_t)arg)); 895 896 switch (cmd) { 897 case DDI_INFO_DEVT2DEVINFO: 898 vhcip = ddi_get_soft_state(vhci_softstate, instance); 899 if (vhcip != NULL) 900 *result = vhcip->vhci_dip; 901 else { 902 *result = NULL; 903 return (DDI_FAILURE); 904 } 905 break; 906 907 case DDI_INFO_DEVT2INSTANCE: 908 *result = (void *)(uintptr_t)instance; 909 break; 910 911 default: 912 return (DDI_FAILURE); 913 } 914 915 return (DDI_SUCCESS); 916 } 917 918 /*ARGSUSED*/ 919 static int 920 vhci_scsi_tgt_init(dev_info_t *hba_dip, dev_info_t *tgt_dip, 921 scsi_hba_tran_t *hba_tran, struct scsi_device *sd) 922 { 923 char *guid; 924 scsi_vhci_lun_t *vlun; 925 struct scsi_vhci *vhci; 926 clock_t from_ticks; 927 mdi_pathinfo_t *pip; 928 int rval; 929 930 ASSERT(hba_dip != NULL); 931 ASSERT(tgt_dip != NULL); 932 933 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, PROPFLAGS, 934 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 935 /* 936 * This must be the .conf node without GUID property. 937 * The node under fp already inserts a delay, so we 938 * just return from here. We rely on this delay to have 939 * all dips be posted to the ndi hotplug thread's newdev 940 * list. This is necessary for the deferred attach 941 * mechanism to work and opens() done soon after boot to 942 * succeed. 943 */ 944 VHCI_DEBUG(4, (CE_WARN, hba_dip, "tgt_init: lun guid " 945 "property failed")); 946 return (DDI_NOT_WELL_FORMED); 947 } 948 949 if (ndi_dev_is_persistent_node(tgt_dip) == 0) { 950 /* 951 * This must be .conf node with the GUID property. We don't 952 * merge property by ndi_merge_node() here because the 953 * devi_addr_buf of .conf node is "" always according the 954 * implementation of vhci_scsi_get_name_bus_addr(). 955 */ 956 ddi_set_name_addr(tgt_dip, NULL); 957 return (DDI_FAILURE); 958 } 959 960 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(hba_dip)); 961 ASSERT(vhci != NULL); 962 963 VHCI_DEBUG(4, (CE_NOTE, hba_dip, 964 "!tgt_init: called for %s (instance %d)\n", 965 ddi_driver_name(tgt_dip), ddi_get_instance(tgt_dip))); 966 967 vlun = vhci_lun_lookup(tgt_dip); 968 969 mutex_enter(&vhci_global_mutex); 970 971 from_ticks = ddi_get_lbolt(); 972 if (vhci_to_ticks == 0) { 973 vhci_to_ticks = from_ticks + 974 drv_usectohz(vhci_init_wait_timeout); 975 } 976 977 #if DEBUG 978 if (vlun) { 979 VHCI_DEBUG(1, (CE_WARN, hba_dip, "tgt_init: " 980 "vhci_scsi_tgt_init: guid %s : found vlun 0x%p " 981 "from_ticks %lx to_ticks %lx", 982 guid, (void *)vlun, from_ticks, vhci_to_ticks)); 983 } else { 984 VHCI_DEBUG(1, (CE_WARN, hba_dip, "tgt_init: " 985 "vhci_scsi_tgt_init: guid %s : vlun not found " 986 "from_ticks %lx to_ticks %lx", guid, from_ticks, 987 vhci_to_ticks)); 988 } 989 #endif 990 991 rval = mdi_select_path(tgt_dip, NULL, 992 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), NULL, &pip); 993 if (rval == MDI_SUCCESS) { 994 mdi_rele_path(pip); 995 } 996 997 /* 998 * Wait for the following conditions : 999 * 1. no vlun available yet 1000 * 2. no path established 1001 * 3. timer did not expire 1002 */ 1003 while ((vlun == NULL) || (mdi_client_get_path_count(tgt_dip) == 0) || 1004 (rval != MDI_SUCCESS)) { 1005 if (vlun && vlun->svl_not_supported) { 1006 VHCI_DEBUG(1, (CE_WARN, hba_dip, "tgt_init: " 1007 "vlun 0x%p lun guid %s not supported!", 1008 (void *)vlun, guid)); 1009 mutex_exit(&vhci_global_mutex); 1010 ddi_prop_free(guid); 1011 return (DDI_NOT_WELL_FORMED); 1012 } 1013 if ((vhci_first_time == 0) && (from_ticks >= vhci_to_ticks)) { 1014 vhci_first_time = 1; 1015 } 1016 if (vhci_first_time == 1) { 1017 VHCI_DEBUG(1, (CE_WARN, hba_dip, "vhci_scsi_tgt_init: " 1018 "no wait for %s. from_tick %lx, to_tick %lx", 1019 guid, from_ticks, vhci_to_ticks)); 1020 mutex_exit(&vhci_global_mutex); 1021 ddi_prop_free(guid); 1022 return (DDI_NOT_WELL_FORMED); 1023 } 1024 1025 if (cv_timedwait(&vhci_cv, 1026 &vhci_global_mutex, vhci_to_ticks) == -1) { 1027 /* Timed out */ 1028 #ifdef DEBUG 1029 if (vlun == NULL) { 1030 VHCI_DEBUG(1, (CE_WARN, hba_dip, 1031 "tgt_init: no vlun for %s!", guid)); 1032 } else if (mdi_client_get_path_count(tgt_dip) == 0) { 1033 VHCI_DEBUG(1, (CE_WARN, hba_dip, 1034 "tgt_init: client path count is " 1035 "zero for %s!", guid)); 1036 } else { 1037 VHCI_DEBUG(1, (CE_WARN, hba_dip, 1038 "tgt_init: client path not " 1039 "available yet for %s!", guid)); 1040 } 1041 #endif /* DEBUG */ 1042 mutex_exit(&vhci_global_mutex); 1043 ddi_prop_free(guid); 1044 return (DDI_NOT_WELL_FORMED); 1045 } 1046 vlun = vhci_lun_lookup(tgt_dip); 1047 rval = mdi_select_path(tgt_dip, NULL, 1048 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 1049 NULL, &pip); 1050 if (rval == MDI_SUCCESS) { 1051 mdi_rele_path(pip); 1052 } 1053 from_ticks = ddi_get_lbolt(); 1054 } 1055 mutex_exit(&vhci_global_mutex); 1056 1057 ASSERT(vlun != NULL); 1058 ddi_prop_free(guid); 1059 hba_tran->tran_tgt_private = vlun; 1060 1061 return (DDI_SUCCESS); 1062 } 1063 1064 /*ARGSUSED*/ 1065 static void 1066 vhci_scsi_tgt_free(dev_info_t *hba_dip, dev_info_t *tgt_dip, 1067 scsi_hba_tran_t *hba_tran, struct scsi_device *sd) 1068 { 1069 } 1070 1071 /* 1072 * a PGR register command has started; copy the info we need 1073 */ 1074 int 1075 vhci_pgr_register_start(scsi_vhci_lun_t *vlun, struct scsi_pkt *pkt) 1076 { 1077 struct vhci_pkt *vpkt = TGTPKT2VHCIPKT(pkt); 1078 void *addr; 1079 1080 if (!vpkt->vpkt_tgt_init_bp) 1081 return (TRAN_BADPKT); 1082 1083 addr = bp_mapin_common(vpkt->vpkt_tgt_init_bp, 1084 (vpkt->vpkt_flags & CFLAG_NOWAIT) ? VM_NOSLEEP : VM_SLEEP); 1085 if (addr == NULL) 1086 return (TRAN_BUSY); 1087 1088 mutex_enter(&vlun->svl_mutex); 1089 1090 vhci_print_prout_keys(vlun, "v_pgr_reg_start: before bcopy:"); 1091 1092 bcopy(addr, &vlun->svl_prout, sizeof (vhci_prout_t) - 1093 (2 * MHIOC_RESV_KEY_SIZE*sizeof (char))); 1094 bcopy(pkt->pkt_cdbp, vlun->svl_cdb, sizeof (vlun->svl_cdb)); 1095 1096 vhci_print_prout_keys(vlun, "v_pgr_reg_start: after bcopy:"); 1097 1098 vlun->svl_time = pkt->pkt_time; 1099 vlun->svl_bcount = vpkt->vpkt_tgt_init_bp->b_bcount; 1100 vlun->svl_first_path = vpkt->vpkt_path; 1101 mutex_exit(&vlun->svl_mutex); 1102 return (0); 1103 } 1104 1105 /* 1106 * Function name : vhci_scsi_start() 1107 * 1108 * Return Values : TRAN_FATAL_ERROR - vhci has been shutdown 1109 * or other fatal failure 1110 * preventing packet transportation 1111 * TRAN_BUSY - request queue is full 1112 * TRAN_ACCEPT - pkt has been submitted to phci 1113 * (or is held in the waitQ) 1114 * Description : Implements SCSA's tran_start() entry point for 1115 * packet transport 1116 * 1117 */ 1118 static int 1119 vhci_scsi_start(struct scsi_address *ap, struct scsi_pkt *pkt) 1120 { 1121 int rval = TRAN_ACCEPT; 1122 int instance, held; 1123 struct scsi_vhci *vhci = ADDR2VHCI(ap); 1124 struct scsi_vhci_lun *vlun = ADDR2VLUN(ap); 1125 struct vhci_pkt *vpkt = TGTPKT2VHCIPKT(pkt); 1126 int flags = 0; 1127 scsi_vhci_priv_t *svp; 1128 dev_info_t *cdip; 1129 client_lb_t lbp; 1130 int restore_lbp = 0; 1131 /* set if pkt is SCSI-II RESERVE cmd */ 1132 int pkt_reserve_cmd = 0; 1133 int reserve_failed = 0; 1134 1135 ASSERT(vhci != NULL); 1136 ASSERT(vpkt != NULL); 1137 ASSERT(vpkt->vpkt_state != VHCI_PKT_ISSUED); 1138 cdip = ADDR2DIP(ap); 1139 1140 /* 1141 * Block IOs if LUN is held or QUIESCED for IOs. 1142 */ 1143 if ((VHCI_LUN_IS_HELD(vlun)) || 1144 ((vlun->svl_flags & VLUN_QUIESCED_FLG) == VLUN_QUIESCED_FLG)) { 1145 return (TRAN_BUSY); 1146 } 1147 1148 /* 1149 * vhci_lun needs to be quiesced before SCSI-II RESERVE command 1150 * can be issued. This may require a cv_timedwait, which is 1151 * dangerous to perform in an interrupt context. So if this 1152 * is a RESERVE command a taskq is dispatched to service it. 1153 * This taskq shall again call vhci_scsi_start, but we shall be 1154 * sure its not in an interrupt context. 1155 */ 1156 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 1157 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 1158 if (!(vpkt->vpkt_state & VHCI_PKT_THRU_TASKQ)) { 1159 if (taskq_dispatch(vhci->vhci_taskq, 1160 vhci_dispatch_scsi_start, (void *) vpkt, 1161 KM_NOSLEEP)) { 1162 return (TRAN_ACCEPT); 1163 } else { 1164 return (TRAN_BUSY); 1165 } 1166 } 1167 1168 /* 1169 * Here we ensure that simultaneous SCSI-II RESERVE cmds don't 1170 * get serviced for a lun. 1171 */ 1172 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 1173 if (!held) { 1174 return (TRAN_BUSY); 1175 } else if ((vlun->svl_flags & VLUN_QUIESCED_FLG) == 1176 VLUN_QUIESCED_FLG) { 1177 VHCI_RELEASE_LUN(vlun); 1178 return (TRAN_BUSY); 1179 } 1180 1181 /* 1182 * To ensure that no IOs occur for this LUN for the duration 1183 * of this pkt set the VLUN_QUIESCED_FLG. 1184 * In case this routine needs to exit on error make sure that 1185 * this flag is cleared. 1186 */ 1187 vlun->svl_flags |= VLUN_QUIESCED_FLG; 1188 pkt_reserve_cmd = 1; 1189 1190 /* 1191 * if this is a SCSI-II RESERVE command, set load balancing 1192 * policy to be ALTERNATE PATH to ensure that all subsequent 1193 * IOs are routed on the same path. This is because if commands 1194 * are routed across multiple paths then IOs on paths other than 1195 * the one on which the RESERVE was executed will get a 1196 * RESERVATION CONFLICT 1197 */ 1198 lbp = mdi_get_lb_policy(cdip); 1199 if (lbp != LOAD_BALANCE_NONE) { 1200 if (vhci_quiesce_lun(vlun) != 1) { 1201 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1202 VHCI_RELEASE_LUN(vlun); 1203 return (TRAN_FATAL_ERROR); 1204 } 1205 vlun->svl_lb_policy_save = lbp; 1206 if (mdi_set_lb_policy(cdip, LOAD_BALANCE_NONE) != 1207 MDI_SUCCESS) { 1208 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1209 VHCI_RELEASE_LUN(vlun); 1210 return (TRAN_FATAL_ERROR); 1211 } 1212 restore_lbp = 1; 1213 } 1214 /* 1215 * See comments for VLUN_RESERVE_ACTIVE_FLG in scsi_vhci.h 1216 * To narrow this window where a reserve command may be sent 1217 * down an inactive path the path states first need to be 1218 * updated. Before calling vhci_update_pathstates reset 1219 * VLUN_RESERVE_ACTIVE_FLG, just in case it was already set 1220 * for this lun. This shall prevent an unnecessary reset 1221 * from being sent out. 1222 */ 1223 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 1224 vhci_update_pathstates((void *)vlun); 1225 } 1226 1227 instance = ddi_get_instance(vhci->vhci_dip); 1228 1229 /* 1230 * If the command is PRIN with action of zero, then the cmd 1231 * is reading PR keys which requires filtering on completion. 1232 * Data cache sync must be guaranteed. 1233 */ 1234 if ((pkt->pkt_cdbp[0] == SCMD_PRIN) && 1235 (pkt->pkt_cdbp[1] == 0) && 1236 (vpkt->vpkt_org_vpkt == NULL)) { 1237 vpkt->vpkt_tgt_init_pkt_flags |= PKT_CONSISTENT; 1238 } 1239 1240 /* 1241 * Do not defer bind for PKT_DMA_PARTIAL 1242 */ 1243 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1244 1245 /* This is a non pkt_dma_partial case */ 1246 if ((rval = vhci_bind_transport( 1247 ap, vpkt, vpkt->vpkt_tgt_init_pkt_flags, NULL_FUNC)) 1248 != TRAN_ACCEPT) { 1249 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1250 "!vhci%d %x: failed to bind transport: " 1251 "vlun 0x%p pkt_reserved %x restore_lbp %x," 1252 "lbp %x", instance, rval, (void *)vlun, 1253 pkt_reserve_cmd, restore_lbp, lbp)); 1254 if (restore_lbp) 1255 (void) mdi_set_lb_policy(cdip, lbp); 1256 if (pkt_reserve_cmd) 1257 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1258 return (rval); 1259 } 1260 VHCI_DEBUG(8, (CE_NOTE, NULL, 1261 "vhci_scsi_start: v_b_t called 0x%p\n", (void *)vpkt)); 1262 } 1263 ASSERT(vpkt->vpkt_hba_pkt != NULL); 1264 ASSERT(vpkt->vpkt_path != NULL); 1265 1266 /* 1267 * This is the chance to adjust the pHCI's pkt and other information 1268 * from target driver's pkt. 1269 */ 1270 VHCI_DEBUG(8, (CE_NOTE, vhci->vhci_dip, "vhci_scsi_start vpkt %p\n", 1271 (void *)vpkt)); 1272 vhci_update_pHCI_pkt(vpkt, pkt); 1273 1274 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 1275 if (vpkt->vpkt_path != vlun->svl_resrv_pip) { 1276 VHCI_DEBUG(1, (CE_WARN, vhci->vhci_dip, 1277 "!vhci_bind: reserve flag set for vlun 0x%p, but, " 1278 "pktpath 0x%p resrv path 0x%p differ. lb_policy %x", 1279 (void *)vlun, (void *)vpkt->vpkt_path, 1280 (void *)vlun->svl_resrv_pip, 1281 mdi_get_lb_policy(cdip))); 1282 reserve_failed = 1; 1283 } 1284 } 1285 1286 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private( 1287 vpkt->vpkt_path); 1288 if (svp == NULL || reserve_failed) { 1289 if (pkt_reserve_cmd) { 1290 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1291 "!vhci_bind returned null svp vlun 0x%p", 1292 (void *)vlun)); 1293 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1294 if (restore_lbp) 1295 (void) mdi_set_lb_policy(cdip, lbp); 1296 } 1297 pkt_cleanup: 1298 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1299 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1300 vpkt->vpkt_hba_pkt = NULL; 1301 if (vpkt->vpkt_path) { 1302 mdi_rele_path(vpkt->vpkt_path); 1303 vpkt->vpkt_path = NULL; 1304 } 1305 } 1306 if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 1307 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 1308 ((pkt->pkt_cdbp[1] & 0x1f) == 1309 VHCI_PROUT_R_AND_IGNORE))) { 1310 sema_v(&vlun->svl_pgr_sema); 1311 } 1312 return (TRAN_BUSY); 1313 } 1314 1315 VHCI_INCR_PATH_CMDCOUNT(svp); 1316 1317 /* 1318 * Ensure that no other IOs raced ahead, while a RESERVE cmd was 1319 * QUIESCING the same lun. 1320 */ 1321 if ((!pkt_reserve_cmd) && 1322 ((vlun->svl_flags & VLUN_QUIESCED_FLG) == VLUN_QUIESCED_FLG)) { 1323 VHCI_DECR_PATH_CMDCOUNT(svp); 1324 goto pkt_cleanup; 1325 } 1326 1327 if ((pkt->pkt_cdbp[0] == SCMD_PRIN) || 1328 (pkt->pkt_cdbp[0] == SCMD_PROUT)) { 1329 /* 1330 * currently this thread only handles running PGR 1331 * commands, so don't bother creating it unless 1332 * something interesting is going to happen (like 1333 * either a PGR out, or a PGR in with enough space 1334 * to hold the keys that are getting returned) 1335 */ 1336 mutex_enter(&vlun->svl_mutex); 1337 if (((vlun->svl_flags & VLUN_TASK_D_ALIVE_FLG) == 0) && 1338 (pkt->pkt_cdbp[0] == SCMD_PROUT)) { 1339 vlun->svl_taskq = taskq_create("vlun_pgr_task_daemon", 1340 1, MINCLSYSPRI, 1, 4, 0); 1341 vlun->svl_flags |= VLUN_TASK_D_ALIVE_FLG; 1342 } 1343 mutex_exit(&vlun->svl_mutex); 1344 if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 1345 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 1346 ((pkt->pkt_cdbp[1] & 0x1f) == 1347 VHCI_PROUT_R_AND_IGNORE))) { 1348 if (rval = vhci_pgr_register_start(vlun, pkt)) { 1349 /* an error */ 1350 sema_v(&vlun->svl_pgr_sema); 1351 return (rval); 1352 } 1353 } 1354 } 1355 1356 /* 1357 * SCSI-II RESERVE cmd is not expected in polled mode. 1358 * If this changes it needs to be handled for the polled scenario. 1359 */ 1360 flags = vpkt->vpkt_hba_pkt->pkt_flags; 1361 1362 /* 1363 * Set the path_instance *before* sending the scsi_pkt down the path 1364 * to mpxio's pHCI so that additional path abstractions at a pHCI 1365 * level (like maybe iSCSI at some point in the future) can update 1366 * the path_instance. 1367 */ 1368 if (scsi_pkt_allocated_correctly(vpkt->vpkt_hba_pkt)) 1369 vpkt->vpkt_hba_pkt->pkt_path_instance = 1370 mdi_pi_get_path_instance(vpkt->vpkt_path); 1371 1372 rval = scsi_transport(vpkt->vpkt_hba_pkt); 1373 if (rval == TRAN_ACCEPT) { 1374 if (flags & FLAG_NOINTR) { 1375 struct scsi_pkt *tpkt = vpkt->vpkt_tgt_pkt; 1376 struct scsi_pkt *pkt = vpkt->vpkt_hba_pkt; 1377 1378 ASSERT(tpkt != NULL); 1379 *(tpkt->pkt_scbp) = *(pkt->pkt_scbp); 1380 tpkt->pkt_resid = pkt->pkt_resid; 1381 tpkt->pkt_state = pkt->pkt_state; 1382 tpkt->pkt_statistics = pkt->pkt_statistics; 1383 tpkt->pkt_reason = pkt->pkt_reason; 1384 1385 if ((*(pkt->pkt_scbp) == STATUS_CHECK) && 1386 (pkt->pkt_state & STATE_ARQ_DONE)) { 1387 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 1388 vpkt->vpkt_tgt_init_scblen); 1389 } 1390 1391 VHCI_DECR_PATH_CMDCOUNT(svp); 1392 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1393 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1394 vpkt->vpkt_hba_pkt = NULL; 1395 if (vpkt->vpkt_path) { 1396 mdi_rele_path(vpkt->vpkt_path); 1397 vpkt->vpkt_path = NULL; 1398 } 1399 } 1400 /* 1401 * This path will not automatically retry pkts 1402 * internally, therefore, vpkt_org_vpkt should 1403 * never be set. 1404 */ 1405 ASSERT(vpkt->vpkt_org_vpkt == NULL); 1406 if (tpkt->pkt_comp) { 1407 (*tpkt->pkt_comp)(tpkt); 1408 } 1409 } 1410 return (rval); 1411 } else if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 1412 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 1413 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE))) { 1414 /* the command exited with bad status */ 1415 sema_v(&vlun->svl_pgr_sema); 1416 } else if (vpkt->vpkt_tgt_pkt->pkt_cdbp[0] == SCMD_PRIN) { 1417 /* the command exited with bad status */ 1418 sema_v(&vlun->svl_pgr_sema); 1419 } else if (pkt_reserve_cmd) { 1420 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1421 "!vhci_scsi_start: reserve failed vlun 0x%p", 1422 (void *)vlun)); 1423 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1424 if (restore_lbp) 1425 (void) mdi_set_lb_policy(cdip, lbp); 1426 } 1427 1428 ASSERT(vpkt->vpkt_hba_pkt != NULL); 1429 VHCI_DECR_PATH_CMDCOUNT(svp); 1430 1431 /* Do not destroy phci packet information for PKT_DMA_PARTIAL */ 1432 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1433 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1434 vpkt->vpkt_hba_pkt = NULL; 1435 if (vpkt->vpkt_path) { 1436 MDI_PI_ERRSTAT(vpkt->vpkt_path, MDI_PI_TRANSERR); 1437 mdi_rele_path(vpkt->vpkt_path); 1438 vpkt->vpkt_path = NULL; 1439 } 1440 } 1441 return (TRAN_BUSY); 1442 } 1443 1444 /* 1445 * Function name : vhci_scsi_reset() 1446 * 1447 * Return Values : 0 - reset failed 1448 * 1 - reset succeeded 1449 */ 1450 1451 /* ARGSUSED */ 1452 static int 1453 vhci_scsi_reset(struct scsi_address *ap, int level) 1454 { 1455 int rval = 0; 1456 1457 cmn_err(CE_WARN, "!vhci_scsi_reset 0x%x", level); 1458 if ((level == RESET_TARGET) || (level == RESET_LUN)) { 1459 return (vhci_scsi_reset_target(ap, level, TRUE)); 1460 } else if (level == RESET_ALL) { 1461 return (vhci_scsi_reset_bus(ap)); 1462 } 1463 1464 return (rval); 1465 } 1466 1467 /* 1468 * vhci_recovery_reset: 1469 * Issues reset to the device 1470 * Input: 1471 * vlun - vhci lun pointer of the device 1472 * ap - address of the device 1473 * select_path: 1474 * If select_path is FALSE, then the address specified in ap is 1475 * the path on which reset will be issued. 1476 * If select_path is TRUE, then path is obtained by calling 1477 * mdi_select_path. 1478 * 1479 * recovery_depth: 1480 * Caller can specify the level of reset. 1481 * VHCI_DEPTH_LUN - 1482 * Issues LUN RESET if device supports lun reset. 1483 * VHCI_DEPTH_TARGET - 1484 * If Lun Reset fails or the device does not support 1485 * Lun Reset, issues TARGET RESET 1486 * VHCI_DEPTH_ALL - 1487 * If Lun Reset fails or the device does not support 1488 * Lun Reset, issues TARGET RESET. 1489 * If TARGET RESET does not succeed, issues Bus Reset. 1490 */ 1491 1492 static int 1493 vhci_recovery_reset(scsi_vhci_lun_t *vlun, struct scsi_address *ap, 1494 uint8_t select_path, uint8_t recovery_depth) 1495 { 1496 int ret = 0; 1497 1498 ASSERT(ap != NULL); 1499 1500 if (vlun && vlun->svl_support_lun_reset == 1) { 1501 ret = vhci_scsi_reset_target(ap, RESET_LUN, 1502 select_path); 1503 } 1504 1505 recovery_depth--; 1506 1507 if ((ret == 0) && recovery_depth) { 1508 ret = vhci_scsi_reset_target(ap, RESET_TARGET, 1509 select_path); 1510 recovery_depth--; 1511 } 1512 1513 if ((ret == 0) && recovery_depth) { 1514 (void) scsi_reset(ap, RESET_ALL); 1515 } 1516 1517 return (ret); 1518 } 1519 1520 /* 1521 * Note: The scsi_address passed to this routine could be the scsi_address 1522 * for the virtual device or the physical device. No assumptions should be 1523 * made in this routine about the ap structure and a_hba_tran->tran_tgt_private 1524 * field of ap can not be assumed to be the vhci structure. 1525 * Further note that the child dip would be the dip of the ssd node irrespective 1526 * of the scsi_address passed. 1527 */ 1528 1529 static int 1530 vhci_scsi_reset_target(struct scsi_address *ap, int level, uint8_t select_path) 1531 { 1532 dev_info_t *vdip, *pdip, *cdip = ADDR2DIP(ap); 1533 mdi_pathinfo_t *pip = NULL; 1534 mdi_pathinfo_t *npip = NULL; 1535 int rval = -1; 1536 scsi_vhci_priv_t *svp = NULL; 1537 struct scsi_address *pap = NULL; 1538 scsi_hba_tran_t *hba = NULL; 1539 int sps; 1540 struct scsi_vhci *vhci = NULL; 1541 1542 if (select_path != TRUE) { 1543 ASSERT(ap != NULL); 1544 if (level == RESET_LUN) { 1545 hba = ap->a_hba_tran; 1546 ASSERT(hba != NULL); 1547 return ((*hba->tran_reset)(ap, RESET_LUN)); 1548 } 1549 return (scsi_reset(ap, level)); 1550 } 1551 1552 ASSERT(cdip != NULL); 1553 vdip = ddi_get_parent(cdip); 1554 ASSERT(vdip != NULL); 1555 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 1556 ASSERT(vhci != NULL); 1557 1558 rval = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH, NULL, &pip); 1559 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 1560 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_scsi_reset_target: " 1561 "Unable to get a path, dip 0x%p", (void *)cdip)); 1562 return (0); 1563 } 1564 again: 1565 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 1566 if (svp == NULL) { 1567 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_scsi_reset_target: " 1568 "priv is NULL, pip 0x%p", (void *)pip)); 1569 mdi_rele_path(pip); 1570 return (0); 1571 } 1572 1573 if (svp->svp_psd == NULL) { 1574 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_scsi_reset_target: " 1575 "psd is NULL, pip 0x%p, svp 0x%p", 1576 (void *)pip, (void *)svp)); 1577 mdi_rele_path(pip); 1578 return (0); 1579 } 1580 1581 pap = &svp->svp_psd->sd_address; 1582 hba = pap->a_hba_tran; 1583 1584 ASSERT(pap != NULL); 1585 ASSERT(hba != NULL); 1586 1587 if (hba->tran_reset != NULL) { 1588 if ((*hba->tran_reset)(pap, level) == 0) { 1589 pdip = mdi_pi_get_phci(pip); 1590 vhci_log(CE_WARN, vdip, "!(%s%d):" 1591 " path (%s%d), reset %d failed", 1592 ddi_driver_name(cdip), ddi_get_instance(cdip), 1593 ddi_driver_name(pdip), ddi_get_instance(pdip), 1594 level); 1595 1596 /* 1597 * Select next path and issue the reset, repeat 1598 * until all paths are exhausted 1599 */ 1600 sps = mdi_select_path(cdip, NULL, 1601 MDI_SELECT_ONLINE_PATH, pip, &npip); 1602 if ((sps != MDI_SUCCESS) || (npip == NULL)) { 1603 mdi_rele_path(pip); 1604 return (0); 1605 } 1606 mdi_rele_path(pip); 1607 pip = npip; 1608 goto again; 1609 } 1610 mdi_rele_path(pip); 1611 mutex_enter(&vhci->vhci_mutex); 1612 scsi_hba_reset_notify_callback(&vhci->vhci_mutex, 1613 &vhci->vhci_reset_notify_listf); 1614 mutex_exit(&vhci->vhci_mutex); 1615 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_scsi_reset_target: " 1616 "reset %d sent down pip:%p for cdip:%p\n", level, 1617 (void *)pip, (void *)cdip)); 1618 return (1); 1619 } 1620 mdi_rele_path(pip); 1621 return (0); 1622 } 1623 1624 1625 /* ARGSUSED */ 1626 static int 1627 vhci_scsi_reset_bus(struct scsi_address *ap) 1628 { 1629 return (1); 1630 } 1631 1632 1633 /* 1634 * called by vhci_getcap and vhci_setcap to get and set (respectively) 1635 * SCSI capabilities 1636 */ 1637 /* ARGSUSED */ 1638 static int 1639 vhci_commoncap(struct scsi_address *ap, char *cap, 1640 int val, int tgtonly, int doset) 1641 { 1642 struct scsi_vhci *vhci = ADDR2VHCI(ap); 1643 struct scsi_vhci_lun *vlun = ADDR2VLUN(ap); 1644 int cidx; 1645 int rval = 0; 1646 1647 if (cap == (char *)0) { 1648 VHCI_DEBUG(3, (CE_WARN, vhci->vhci_dip, 1649 "!vhci_commoncap: invalid arg")); 1650 return (rval); 1651 } 1652 1653 if (vlun == NULL) { 1654 VHCI_DEBUG(3, (CE_WARN, vhci->vhci_dip, 1655 "!vhci_commoncap: vlun is null")); 1656 return (rval); 1657 } 1658 1659 if ((cidx = scsi_hba_lookup_capstr(cap)) == -1) { 1660 return (UNDEFINED); 1661 } 1662 1663 /* 1664 * Process setcap request. 1665 */ 1666 if (doset) { 1667 /* 1668 * At present, we can only set binary (0/1) values 1669 */ 1670 switch (cidx) { 1671 case SCSI_CAP_ARQ: 1672 if (val == 0) { 1673 rval = 0; 1674 } else { 1675 rval = 1; 1676 } 1677 break; 1678 1679 case SCSI_CAP_LUN_RESET: 1680 if (tgtonly == 0) { 1681 VHCI_DEBUG(1, (CE_WARN, vhci->vhci_dip, 1682 "scsi_vhci_setcap: " 1683 "Returning error since whom = 0")); 1684 rval = -1; 1685 break; 1686 } 1687 /* 1688 * Set the capability accordingly. 1689 */ 1690 mutex_enter(&vlun->svl_mutex); 1691 vlun->svl_support_lun_reset = val; 1692 rval = val; 1693 mutex_exit(&vlun->svl_mutex); 1694 break; 1695 1696 case SCSI_CAP_SECTOR_SIZE: 1697 mutex_enter(&vlun->svl_mutex); 1698 vlun->svl_sector_size = val; 1699 vlun->svl_setcap_done = 1; 1700 mutex_exit(&vlun->svl_mutex); 1701 (void) vhci_pHCI_cap(ap, cap, val, tgtonly, NULL); 1702 1703 /* Always return success */ 1704 rval = 1; 1705 break; 1706 1707 default: 1708 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1709 "!vhci_setcap: unsupported %d", cidx)); 1710 rval = UNDEFINED; 1711 break; 1712 } 1713 1714 VHCI_DEBUG(6, (CE_NOTE, vhci->vhci_dip, 1715 "!set cap: cap=%s, val/tgtonly/doset/rval = " 1716 "0x%x/0x%x/0x%x/%d\n", 1717 cap, val, tgtonly, doset, rval)); 1718 1719 } else { 1720 /* 1721 * Process getcap request. 1722 */ 1723 switch (cidx) { 1724 case SCSI_CAP_DMA_MAX: 1725 rval = (int)VHCI_DMA_MAX_XFER_CAP; 1726 break; 1727 1728 case SCSI_CAP_INITIATOR_ID: 1729 rval = 0x00; 1730 break; 1731 1732 case SCSI_CAP_ARQ: 1733 case SCSI_CAP_RESET_NOTIFICATION: 1734 case SCSI_CAP_TAGGED_QING: 1735 rval = 1; 1736 break; 1737 1738 case SCSI_CAP_SCSI_VERSION: 1739 rval = 3; 1740 break; 1741 1742 case SCSI_CAP_INTERCONNECT_TYPE: 1743 rval = INTERCONNECT_FABRIC; 1744 break; 1745 1746 case SCSI_CAP_LUN_RESET: 1747 /* 1748 * scsi_vhci will always return success for LUN reset. 1749 * When request for doing LUN reset comes 1750 * through scsi_reset entry point, at that time attempt 1751 * will be made to do reset through all the possible 1752 * paths. 1753 */ 1754 mutex_enter(&vlun->svl_mutex); 1755 rval = vlun->svl_support_lun_reset; 1756 mutex_exit(&vlun->svl_mutex); 1757 VHCI_DEBUG(4, (CE_WARN, vhci->vhci_dip, 1758 "scsi_vhci_getcap:" 1759 "Getting the Lun reset capability %d", rval)); 1760 break; 1761 1762 case SCSI_CAP_SECTOR_SIZE: 1763 mutex_enter(&vlun->svl_mutex); 1764 rval = vlun->svl_sector_size; 1765 mutex_exit(&vlun->svl_mutex); 1766 break; 1767 1768 default: 1769 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1770 "!vhci_getcap: unsupported %d", cidx)); 1771 rval = UNDEFINED; 1772 break; 1773 } 1774 1775 VHCI_DEBUG(6, (CE_NOTE, vhci->vhci_dip, 1776 "!get cap: cap=%s, val/tgtonly/doset/rval = " 1777 "0x%x/0x%x/0x%x/%d\n", 1778 cap, val, tgtonly, doset, rval)); 1779 } 1780 return (rval); 1781 } 1782 1783 1784 /* 1785 * Function name : vhci_scsi_getcap() 1786 * 1787 */ 1788 static int 1789 vhci_scsi_getcap(struct scsi_address *ap, char *cap, int whom) 1790 { 1791 return (vhci_commoncap(ap, cap, 0, whom, 0)); 1792 } 1793 1794 static int 1795 vhci_scsi_setcap(struct scsi_address *ap, char *cap, int value, int whom) 1796 { 1797 return (vhci_commoncap(ap, cap, value, whom, 1)); 1798 } 1799 1800 /* 1801 * Function name : vhci_scsi_abort() 1802 */ 1803 /* ARGSUSED */ 1804 static int 1805 vhci_scsi_abort(struct scsi_address *ap, struct scsi_pkt *pkt) 1806 { 1807 return (0); 1808 } 1809 1810 /* 1811 * Function name : vhci_scsi_init_pkt 1812 * 1813 * Return Values : pointer to scsi_pkt, or NULL 1814 */ 1815 /* ARGSUSED */ 1816 static struct scsi_pkt * 1817 vhci_scsi_init_pkt(struct scsi_address *ap, struct scsi_pkt *pkt, 1818 struct buf *bp, int cmdlen, int statuslen, int tgtlen, 1819 int flags, int (*callback)(caddr_t), caddr_t arg) 1820 { 1821 struct scsi_vhci *vhci = ADDR2VHCI(ap); 1822 struct vhci_pkt *vpkt; 1823 int rval; 1824 int newpkt = 0; 1825 struct scsi_pkt *pktp; 1826 1827 1828 if (pkt == NULL) { 1829 if (cmdlen > VHCI_SCSI_CDB_SIZE) { 1830 VHCI_DEBUG(1, (CE_NOTE, NULL, 1831 "!init pkt: cdb size not supported\n")); 1832 return (NULL); 1833 } 1834 1835 pktp = scsi_hba_pkt_alloc(vhci->vhci_dip, 1836 ap, cmdlen, statuslen, tgtlen, sizeof (*vpkt), callback, 1837 arg); 1838 1839 if (pktp == NULL) { 1840 return (NULL); 1841 } 1842 1843 /* Get the vhci's private structure */ 1844 vpkt = (struct vhci_pkt *)(pktp->pkt_ha_private); 1845 ASSERT(vpkt); 1846 1847 /* Save the target driver's packet */ 1848 vpkt->vpkt_tgt_pkt = pktp; 1849 1850 /* 1851 * Save pkt_tgt_init_pkt fields if deferred binding 1852 * is needed or for other purposes. 1853 */ 1854 vpkt->vpkt_tgt_init_pkt_flags = flags; 1855 vpkt->vpkt_flags = (callback == NULL_FUNC) ? CFLAG_NOWAIT : 0; 1856 vpkt->vpkt_state = VHCI_PKT_IDLE; 1857 vpkt->vpkt_tgt_init_cdblen = cmdlen; 1858 vpkt->vpkt_tgt_init_scblen = statuslen; 1859 newpkt = 1; 1860 } else { /* pkt not NULL */ 1861 vpkt = pkt->pkt_ha_private; 1862 } 1863 1864 VHCI_DEBUG(8, (CE_NOTE, NULL, "vhci_scsi_init_pkt " 1865 "vpkt %p flags %x\n", (void *)vpkt, flags)); 1866 1867 /* Clear any stale error flags */ 1868 if (bp) { 1869 bioerror(bp, 0); 1870 } 1871 1872 vpkt->vpkt_tgt_init_bp = bp; 1873 1874 if (flags & PKT_DMA_PARTIAL) { 1875 1876 /* 1877 * Immediate binding is needed. 1878 * Target driver may not set this flag in next invocation. 1879 * vhci has to remember this flag was set during first 1880 * invocation of vhci_scsi_init_pkt. 1881 */ 1882 vpkt->vpkt_flags |= CFLAG_DMA_PARTIAL; 1883 } 1884 1885 if (vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) { 1886 1887 /* 1888 * Re-initialize some of the target driver packet state 1889 * information. 1890 */ 1891 vpkt->vpkt_tgt_pkt->pkt_state = 0; 1892 vpkt->vpkt_tgt_pkt->pkt_statistics = 0; 1893 vpkt->vpkt_tgt_pkt->pkt_reason = 0; 1894 1895 /* 1896 * Binding a vpkt->vpkt_path for this IO at init_time. 1897 * If an IO error happens later, target driver will clear 1898 * this vpkt->vpkt_path binding before re-init IO again. 1899 */ 1900 VHCI_DEBUG(8, (CE_NOTE, NULL, 1901 "vhci_scsi_init_pkt: calling v_b_t %p, newpkt %d\n", 1902 (void *)vpkt, newpkt)); 1903 if (pkt && vpkt->vpkt_hba_pkt) { 1904 VHCI_DEBUG(4, (CE_NOTE, NULL, 1905 "v_s_i_p calling update_pHCI_pkt resid %ld\n", 1906 pkt->pkt_resid)); 1907 vhci_update_pHCI_pkt(vpkt, pkt); 1908 } 1909 if (callback == SLEEP_FUNC) { 1910 rval = vhci_bind_transport( 1911 ap, vpkt, flags, callback); 1912 } else { 1913 rval = vhci_bind_transport( 1914 ap, vpkt, flags, NULL_FUNC); 1915 } 1916 VHCI_DEBUG(8, (CE_NOTE, NULL, 1917 "vhci_scsi_init_pkt: v_b_t called 0x%p rval 0x%x\n", 1918 (void *)vpkt, rval)); 1919 if (bp) { 1920 if (rval == TRAN_FATAL_ERROR) { 1921 /* 1922 * No paths available. Could not bind 1923 * any pHCI. Setting EFAULT as a way 1924 * to indicate no DMA is mapped. 1925 */ 1926 bioerror(bp, EFAULT); 1927 } else { 1928 /* 1929 * Do not indicate any pHCI errors to 1930 * target driver otherwise. 1931 */ 1932 bioerror(bp, 0); 1933 } 1934 } 1935 if (rval != TRAN_ACCEPT) { 1936 VHCI_DEBUG(8, (CE_NOTE, NULL, 1937 "vhci_scsi_init_pkt: " 1938 "v_b_t failed 0x%p newpkt %x\n", 1939 (void *)vpkt, newpkt)); 1940 if (newpkt) { 1941 scsi_hba_pkt_free(ap, 1942 vpkt->vpkt_tgt_pkt); 1943 } 1944 return (NULL); 1945 } 1946 ASSERT(vpkt->vpkt_hba_pkt != NULL); 1947 ASSERT(vpkt->vpkt_path != NULL); 1948 1949 /* Update the resid for the target driver */ 1950 vpkt->vpkt_tgt_pkt->pkt_resid = 1951 vpkt->vpkt_hba_pkt->pkt_resid; 1952 } 1953 1954 return (vpkt->vpkt_tgt_pkt); 1955 } 1956 1957 /* 1958 * Function name : vhci_scsi_destroy_pkt 1959 * 1960 * Return Values : none 1961 */ 1962 static void 1963 vhci_scsi_destroy_pkt(struct scsi_address *ap, struct scsi_pkt *pkt) 1964 { 1965 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_ha_private; 1966 1967 VHCI_DEBUG(8, (CE_NOTE, NULL, 1968 "vhci_scsi_destroy_pkt: vpkt 0x%p\n", (void *)vpkt)); 1969 1970 vpkt->vpkt_tgt_init_pkt_flags = 0; 1971 if (vpkt->vpkt_hba_pkt) { 1972 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1973 vpkt->vpkt_hba_pkt = NULL; 1974 } 1975 if (vpkt->vpkt_path) { 1976 mdi_rele_path(vpkt->vpkt_path); 1977 vpkt->vpkt_path = NULL; 1978 } 1979 1980 ASSERT(vpkt->vpkt_state != VHCI_PKT_ISSUED); 1981 scsi_hba_pkt_free(ap, vpkt->vpkt_tgt_pkt); 1982 } 1983 1984 /* 1985 * Function name : vhci_scsi_dmafree() 1986 * 1987 * Return Values : none 1988 */ 1989 /*ARGSUSED*/ 1990 static void 1991 vhci_scsi_dmafree(struct scsi_address *ap, struct scsi_pkt *pkt) 1992 { 1993 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_ha_private; 1994 1995 VHCI_DEBUG(6, (CE_NOTE, NULL, 1996 "vhci_scsi_dmafree: vpkt 0x%p\n", (void *)vpkt)); 1997 1998 ASSERT(vpkt != NULL); 1999 if (vpkt->vpkt_hba_pkt) { 2000 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 2001 vpkt->vpkt_hba_pkt = NULL; 2002 } 2003 if (vpkt->vpkt_path) { 2004 mdi_rele_path(vpkt->vpkt_path); 2005 vpkt->vpkt_path = NULL; 2006 } 2007 } 2008 2009 /* 2010 * Function name : vhci_scsi_sync_pkt() 2011 * 2012 * Return Values : none 2013 */ 2014 /*ARGSUSED*/ 2015 static void 2016 vhci_scsi_sync_pkt(struct scsi_address *ap, struct scsi_pkt *pkt) 2017 { 2018 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_ha_private; 2019 2020 ASSERT(vpkt != NULL); 2021 if (vpkt->vpkt_hba_pkt) { 2022 scsi_sync_pkt(vpkt->vpkt_hba_pkt); 2023 } 2024 } 2025 2026 /* 2027 * routine for reset notification setup, to register or cancel. 2028 */ 2029 static int 2030 vhci_scsi_reset_notify(struct scsi_address *ap, int flag, 2031 void (*callback)(caddr_t), caddr_t arg) 2032 { 2033 struct scsi_vhci *vhci = ADDR2VHCI(ap); 2034 return (scsi_hba_reset_notify_setup(ap, flag, callback, arg, 2035 &vhci->vhci_mutex, &vhci->vhci_reset_notify_listf)); 2036 } 2037 2038 static int 2039 vhci_scsi_get_name_bus_addr(struct scsi_device *sd, 2040 char *name, int len, int bus_addr) 2041 { 2042 dev_info_t *cdip; 2043 char *guid; 2044 scsi_vhci_lun_t *vlun; 2045 2046 ASSERT(sd != NULL); 2047 ASSERT(name != NULL); 2048 2049 cdip = sd->sd_dev; 2050 2051 ASSERT(cdip != NULL); 2052 2053 if (mdi_component_is_client(cdip, NULL) != MDI_SUCCESS) { 2054 name[0] = '\0'; 2055 return (1); 2056 } 2057 2058 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, cdip, PROPFLAGS, 2059 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 2060 name[0] = '\0'; 2061 return (1); 2062 } 2063 2064 vlun = ADDR2VLUN(&sd->sd_address); 2065 if (bus_addr && vlun && vlun->svl_fops_name) { 2066 /* report the guid and the name of the failover module */ 2067 (void) snprintf(name, len, "g%s %s", guid, vlun->svl_fops_name); 2068 } else { 2069 /* report the guid */ 2070 (void) snprintf(name, len, "g%s", guid); 2071 } 2072 2073 ddi_prop_free(guid); 2074 return (1); 2075 } 2076 2077 static int 2078 vhci_scsi_get_bus_addr(struct scsi_device *sd, char *name, int len) 2079 { 2080 return (vhci_scsi_get_name_bus_addr(sd, name, len, 1)); 2081 } 2082 2083 static int 2084 vhci_scsi_get_name(struct scsi_device *sd, char *name, int len) 2085 { 2086 return (vhci_scsi_get_name_bus_addr(sd, name, len, 0)); 2087 } 2088 2089 /* 2090 * Return a pointer to the guid part of the devnm. 2091 * devnm format is "nodename@busaddr", busaddr format is "gGUID". 2092 */ 2093 static char * 2094 vhci_devnm_to_guid(char *devnm) 2095 { 2096 char *cp = devnm; 2097 2098 if (devnm == NULL) 2099 return (NULL); 2100 2101 while (*cp != '\0' && *cp != '@') 2102 cp++; 2103 if (*cp == '@' && *(cp + 1) == 'g') 2104 return (cp + 2); 2105 return (NULL); 2106 } 2107 2108 static int 2109 vhci_bind_transport(struct scsi_address *ap, struct vhci_pkt *vpkt, int flags, 2110 int (*func)(caddr_t)) 2111 { 2112 struct scsi_vhci *vhci = ADDR2VHCI(ap); 2113 dev_info_t *cdip = ADDR2DIP(ap); 2114 mdi_pathinfo_t *pip = NULL; 2115 mdi_pathinfo_t *npip = NULL; 2116 scsi_vhci_priv_t *svp = NULL; 2117 struct scsi_device *psd = NULL; 2118 struct scsi_address *address = NULL; 2119 struct scsi_pkt *pkt = NULL; 2120 int rval = -1; 2121 int pgr_sema_held = 0; 2122 int held; 2123 int mps_flag = MDI_SELECT_ONLINE_PATH; 2124 struct scsi_vhci_lun *vlun; 2125 time_t tnow; 2126 int path_instance; 2127 2128 vlun = ADDR2VLUN(ap); 2129 ASSERT(vlun != 0); 2130 2131 if ((vpkt->vpkt_tgt_pkt->pkt_cdbp[0] == SCMD_PROUT) && 2132 (((vpkt->vpkt_tgt_pkt->pkt_cdbp[1] & 0x1f) == 2133 VHCI_PROUT_REGISTER) || 2134 ((vpkt->vpkt_tgt_pkt->pkt_cdbp[1] & 0x1f) == 2135 VHCI_PROUT_R_AND_IGNORE))) { 2136 if (!sema_tryp(&vlun->svl_pgr_sema)) 2137 return (TRAN_BUSY); 2138 pgr_sema_held = 1; 2139 if (vlun->svl_first_path != NULL) { 2140 rval = mdi_select_path(cdip, NULL, 2141 MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH, 2142 NULL, &pip); 2143 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 2144 VHCI_DEBUG(4, (CE_NOTE, NULL, 2145 "vhci_bind_transport: path select fail\n")); 2146 } else { 2147 npip = pip; 2148 do { 2149 if (npip == vlun->svl_first_path) { 2150 VHCI_DEBUG(4, (CE_NOTE, NULL, 2151 "vhci_bind_transport: " 2152 "valid first path 0x%p\n", 2153 (void *) 2154 vlun->svl_first_path)); 2155 pip = vlun->svl_first_path; 2156 goto bind_path; 2157 } 2158 pip = npip; 2159 rval = mdi_select_path(cdip, NULL, 2160 MDI_SELECT_ONLINE_PATH | 2161 MDI_SELECT_STANDBY_PATH, 2162 pip, &npip); 2163 mdi_rele_path(pip); 2164 } while ((rval == MDI_SUCCESS) && 2165 (npip != NULL)); 2166 } 2167 } 2168 2169 if (vlun->svl_first_path) { 2170 VHCI_DEBUG(4, (CE_NOTE, NULL, 2171 "vhci_bind_transport: invalid first path 0x%p\n", 2172 (void *)vlun->svl_first_path)); 2173 vlun->svl_first_path = NULL; 2174 } 2175 } else if (vpkt->vpkt_tgt_pkt->pkt_cdbp[0] == SCMD_PRIN) { 2176 if ((vpkt->vpkt_state & VHCI_PKT_THRU_TASKQ) == 0) { 2177 if (!sema_tryp(&vlun->svl_pgr_sema)) 2178 return (TRAN_BUSY); 2179 } 2180 pgr_sema_held = 1; 2181 } 2182 2183 /* 2184 * If the path is already bound for PKT_PARTIAL_DMA case, 2185 * try to use the same path. 2186 */ 2187 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) && vpkt->vpkt_path) { 2188 VHCI_DEBUG(4, (CE_NOTE, NULL, 2189 "vhci_bind_transport: PKT_PARTIAL_DMA " 2190 "vpkt 0x%p, path 0x%p\n", 2191 (void *)vpkt, (void *)vpkt->vpkt_path)); 2192 pip = vpkt->vpkt_path; 2193 goto bind_path; 2194 } 2195 2196 /* 2197 * Get path_instance. Non-zero indicates that mdi_select_path should 2198 * be called to select a specific instance. 2199 * 2200 * NB: Condition pkt_path_instance reference on proper allocation. 2201 */ 2202 if (scsi_pkt_allocated_correctly(vpkt->vpkt_tgt_pkt)) 2203 path_instance = vpkt->vpkt_tgt_pkt->pkt_path_instance; 2204 else 2205 path_instance = 0; 2206 2207 /* 2208 * If reservation is active bind the transport directly to the pip 2209 * with the reservation. 2210 */ 2211 if (vpkt->vpkt_hba_pkt == NULL) { 2212 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 2213 if (MDI_PI_IS_ONLINE(vlun->svl_resrv_pip)) { 2214 pip = vlun->svl_resrv_pip; 2215 mdi_hold_path(pip); 2216 vlun->svl_waiting_for_activepath = 0; 2217 rval = MDI_SUCCESS; 2218 goto bind_path; 2219 } else { 2220 if (pgr_sema_held) { 2221 sema_v(&vlun->svl_pgr_sema); 2222 } 2223 return (TRAN_BUSY); 2224 } 2225 } 2226 try_again: 2227 rval = mdi_select_path(cdip, vpkt->vpkt_tgt_init_bp, 2228 path_instance ? MDI_SELECT_PATH_INSTANCE : 0, 2229 (void *)(intptr_t)path_instance, &pip); 2230 if (rval == MDI_BUSY) { 2231 if (pgr_sema_held) { 2232 sema_v(&vlun->svl_pgr_sema); 2233 } 2234 return (TRAN_BUSY); 2235 } else if (rval == MDI_DEVI_ONLINING) { 2236 /* 2237 * if we are here then we are in the midst of 2238 * an attach/probe of the client device. 2239 * We attempt to bind to ONLINE path if available, 2240 * else it is OK to bind to a STANDBY path (instead 2241 * of triggering a failover) because IO associated 2242 * with attach/probe (eg. INQUIRY, block 0 read) 2243 * are completed by targets even on passive paths 2244 * If no ONLINE paths available, it is important 2245 * to set svl_waiting_for_activepath for two 2246 * reasons: (1) avoid sense analysis in the 2247 * "external failure detection" codepath in 2248 * vhci_intr(). Failure to do so will result in 2249 * infinite loop (unless an ONLINE path becomes 2250 * available at some point) (2) avoid 2251 * unnecessary failover (see "---Waiting For Active 2252 * Path---" comment below). 2253 */ 2254 VHCI_DEBUG(1, (CE_NOTE, NULL, "!%p in onlining " 2255 "state\n", (void *)cdip)); 2256 pip = NULL; 2257 rval = mdi_select_path(cdip, vpkt->vpkt_tgt_init_bp, 2258 mps_flag, NULL, &pip); 2259 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 2260 if (vlun->svl_waiting_for_activepath == 0) { 2261 vlun->svl_waiting_for_activepath = 1; 2262 vlun->svl_wfa_time = ddi_get_time(); 2263 } 2264 mps_flag |= MDI_SELECT_STANDBY_PATH; 2265 rval = mdi_select_path(cdip, 2266 vpkt->vpkt_tgt_init_bp, 2267 mps_flag, NULL, &pip); 2268 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 2269 if (pgr_sema_held) { 2270 sema_v(&vlun->svl_pgr_sema); 2271 } 2272 return (TRAN_FATAL_ERROR); 2273 } 2274 goto bind_path; 2275 } 2276 } else if (rval == MDI_FAILURE) { 2277 if (pgr_sema_held) { 2278 sema_v(&vlun->svl_pgr_sema); 2279 } 2280 return (TRAN_FATAL_ERROR); 2281 } 2282 2283 if ((pip == NULL) || (rval == MDI_NOPATH)) { 2284 while (vlun->svl_waiting_for_activepath) { 2285 /* 2286 * ---Waiting For Active Path--- 2287 * This device was discovered across a 2288 * passive path; lets wait for a little 2289 * bit, hopefully an active path will 2290 * show up obviating the need for a 2291 * failover 2292 */ 2293 tnow = ddi_get_time(); 2294 if (tnow - vlun->svl_wfa_time >= 60) { 2295 vlun->svl_waiting_for_activepath = 0; 2296 } else { 2297 drv_usecwait(1000); 2298 if (vlun->svl_waiting_for_activepath 2299 == 0) { 2300 /* 2301 * an active path has come 2302 * online! 2303 */ 2304 goto try_again; 2305 } 2306 } 2307 } 2308 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 2309 if (!held) { 2310 VHCI_DEBUG(4, (CE_NOTE, NULL, 2311 "!Lun not held\n")); 2312 if (pgr_sema_held) { 2313 sema_v(&vlun->svl_pgr_sema); 2314 } 2315 return (TRAN_BUSY); 2316 } 2317 /* 2318 * now that the LUN is stable, one last check 2319 * to make sure no other changes sneaked in 2320 * (like a path coming online or a 2321 * failover initiated by another thread) 2322 */ 2323 pip = NULL; 2324 rval = mdi_select_path(cdip, vpkt->vpkt_tgt_init_bp, 2325 0, NULL, &pip); 2326 if (pip != NULL) { 2327 VHCI_RELEASE_LUN(vlun); 2328 vlun->svl_waiting_for_activepath = 0; 2329 goto bind_path; 2330 } 2331 2332 /* 2333 * Check if there is an ONLINE path OR a STANDBY path 2334 * available. If none is available, do not attempt 2335 * to do a failover, just return a fatal error at this 2336 * point. 2337 */ 2338 npip = NULL; 2339 rval = mdi_select_path(cdip, NULL, 2340 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 2341 NULL, &npip); 2342 if ((npip == NULL) || (rval != MDI_SUCCESS)) { 2343 /* 2344 * No paths available, jus return FATAL error. 2345 */ 2346 VHCI_RELEASE_LUN(vlun); 2347 if (pgr_sema_held) { 2348 sema_v(&vlun->svl_pgr_sema); 2349 } 2350 return (TRAN_FATAL_ERROR); 2351 } 2352 mdi_rele_path(npip); 2353 VHCI_DEBUG(1, (CE_NOTE, NULL, "!invoking " 2354 "mdi_failover\n")); 2355 rval = mdi_failover(vhci->vhci_dip, cdip, 2356 MDI_FAILOVER_ASYNC); 2357 if (rval == MDI_FAILURE) { 2358 VHCI_RELEASE_LUN(vlun); 2359 if (pgr_sema_held) { 2360 sema_v(&vlun->svl_pgr_sema); 2361 } 2362 return (TRAN_FATAL_ERROR); 2363 } else if (rval == MDI_BUSY) { 2364 VHCI_RELEASE_LUN(vlun); 2365 if (pgr_sema_held) { 2366 sema_v(&vlun->svl_pgr_sema); 2367 } 2368 return (TRAN_BUSY); 2369 } else { 2370 if (pgr_sema_held) { 2371 sema_v(&vlun->svl_pgr_sema); 2372 } 2373 return (TRAN_BUSY); 2374 } 2375 } 2376 vlun->svl_waiting_for_activepath = 0; 2377 bind_path: 2378 vpkt->vpkt_path = pip; 2379 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 2380 ASSERT(svp != NULL); 2381 2382 psd = svp->svp_psd; 2383 ASSERT(psd != NULL); 2384 address = &psd->sd_address; 2385 } else { 2386 pkt = vpkt->vpkt_hba_pkt; 2387 address = &pkt->pkt_address; 2388 } 2389 2390 /* Verify match of specified path_instance and selected path_instance */ 2391 ASSERT((path_instance == 0) || 2392 (path_instance == mdi_pi_get_path_instance(vpkt->vpkt_path))); 2393 2394 /* 2395 * For PKT_PARTIAL_DMA case, call pHCI's scsi_init_pkt whenever 2396 * target driver calls vhci_scsi_init_pkt. 2397 */ 2398 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) && 2399 vpkt->vpkt_path && vpkt->vpkt_hba_pkt) { 2400 VHCI_DEBUG(4, (CE_NOTE, NULL, 2401 "vhci_bind_transport: PKT_PARTIAL_DMA " 2402 "vpkt 0x%p, path 0x%p hba_pkt 0x%p\n", 2403 (void *)vpkt, (void *)vpkt->vpkt_path, (void *)pkt)); 2404 pkt = vpkt->vpkt_hba_pkt; 2405 address = &pkt->pkt_address; 2406 } 2407 2408 if (pkt == NULL || (vpkt->vpkt_flags & CFLAG_DMA_PARTIAL)) { 2409 pkt = scsi_init_pkt(address, pkt, 2410 vpkt->vpkt_tgt_init_bp, vpkt->vpkt_tgt_init_cdblen, 2411 vpkt->vpkt_tgt_init_scblen, 2412 0, flags, func, NULL); 2413 2414 if (pkt == NULL) { 2415 VHCI_DEBUG(4, (CE_NOTE, NULL, 2416 "!bind transport: 0x%p 0x%p 0x%p\n", 2417 (void *)vhci, (void *)psd, (void *)vpkt)); 2418 if ((vpkt->vpkt_hba_pkt == NULL) && vpkt->vpkt_path) { 2419 MDI_PI_ERRSTAT(vpkt->vpkt_path, 2420 MDI_PI_TRANSERR); 2421 mdi_rele_path(vpkt->vpkt_path); 2422 vpkt->vpkt_path = NULL; 2423 } 2424 if (pgr_sema_held) { 2425 sema_v(&vlun->svl_pgr_sema); 2426 } 2427 /* 2428 * Looks like a fatal error. 2429 * May be device disappeared underneath. 2430 * Give another chance to target driver for a retry to 2431 * get another path. 2432 */ 2433 return (TRAN_BUSY); 2434 } 2435 } 2436 2437 pkt->pkt_private = vpkt; 2438 vpkt->vpkt_hba_pkt = pkt; 2439 return (TRAN_ACCEPT); 2440 } 2441 2442 2443 /*PRINTFLIKE3*/ 2444 void 2445 vhci_log(int level, dev_info_t *dip, const char *fmt, ...) 2446 { 2447 char buf[256]; 2448 va_list ap; 2449 2450 va_start(ap, fmt); 2451 (void) vsprintf(buf, fmt, ap); 2452 va_end(ap); 2453 2454 scsi_log(dip, "scsi_vhci", level, buf); 2455 } 2456 2457 /* do a PGR out with the information we've saved away */ 2458 static int 2459 vhci_do_prout(scsi_vhci_priv_t *svp) 2460 { 2461 2462 struct scsi_pkt *new_pkt; 2463 struct buf *bp; 2464 scsi_vhci_lun_t *vlun; 2465 int rval, retry, nr_retry, ua_retry; 2466 struct scsi_extended_sense *sns; 2467 2468 bp = getrbuf(KM_SLEEP); 2469 bp->b_flags = B_WRITE; 2470 bp->b_resid = 0; 2471 2472 VHCI_INCR_PATH_CMDCOUNT(svp); 2473 vlun = svp->svp_svl; 2474 2475 new_pkt = scsi_init_pkt(&svp->svp_psd->sd_address, NULL, bp, 2476 CDB_GROUP1, sizeof (struct scsi_arq_status), 0, 0, 2477 SLEEP_FUNC, NULL); 2478 if (new_pkt == NULL) { 2479 VHCI_DECR_PATH_CMDCOUNT(svp); 2480 freerbuf(bp); 2481 cmn_err(CE_WARN, "!vhci_do_prout: scsi_init_pkt failed"); 2482 return (0); 2483 } 2484 mutex_enter(&vlun->svl_mutex); 2485 bp->b_un.b_addr = (caddr_t)&vlun->svl_prout; 2486 bp->b_bcount = vlun->svl_bcount; 2487 bcopy(vlun->svl_cdb, new_pkt->pkt_cdbp, 2488 sizeof (vlun->svl_cdb)); 2489 new_pkt->pkt_time = vlun->svl_time; 2490 mutex_exit(&vlun->svl_mutex); 2491 new_pkt->pkt_flags = FLAG_NOINTR; 2492 2493 ua_retry = nr_retry = retry = 0; 2494 again: 2495 rval = vhci_do_scsi_cmd(new_pkt); 2496 if (rval != 1) { 2497 if ((new_pkt->pkt_reason == CMD_CMPLT) && 2498 (SCBP_C(new_pkt) == STATUS_CHECK) && 2499 (new_pkt->pkt_state & STATE_ARQ_DONE)) { 2500 sns = &(((struct scsi_arq_status *)(uintptr_t) 2501 (new_pkt->pkt_scbp))->sts_sensedata); 2502 if ((sns->es_key == KEY_UNIT_ATTENTION) || 2503 (sns->es_key == KEY_NOT_READY)) { 2504 int max_retry; 2505 struct scsi_failover_ops *fops; 2506 fops = vlun->svl_fops; 2507 rval = (*fops->sfo_analyze_sense) 2508 (svp->svp_psd, sns, 2509 vlun->svl_fops_ctpriv); 2510 if (rval == SCSI_SENSE_NOT_READY) { 2511 max_retry = vhci_prout_not_ready_retry; 2512 retry = nr_retry++; 2513 delay(1*drv_usectohz(1000000)); 2514 } else { 2515 /* chk for state change and update */ 2516 if (rval == SCSI_SENSE_STATE_CHANGED) { 2517 int held; 2518 VHCI_HOLD_LUN(vlun, 2519 VH_NOSLEEP, held); 2520 if (!held) { 2521 rval = TRAN_BUSY; 2522 } else { 2523 /* chk for alua first */ 2524 vhci_update_pathstates( 2525 (void *)vlun); 2526 } 2527 } 2528 retry = ua_retry++; 2529 max_retry = VHCI_MAX_PGR_RETRIES; 2530 } 2531 if (retry < max_retry) { 2532 VHCI_DEBUG(4, (CE_WARN, NULL, 2533 "!vhci_do_prout retry 0x%x " 2534 "(0x%x 0x%x 0x%x)", 2535 SCBP_C(new_pkt), 2536 new_pkt->pkt_cdbp[0], 2537 new_pkt->pkt_cdbp[1], 2538 new_pkt->pkt_cdbp[2])); 2539 goto again; 2540 } 2541 rval = 0; 2542 VHCI_DEBUG(4, (CE_WARN, NULL, 2543 "!vhci_do_prout 0x%x " 2544 "(0x%x 0x%x 0x%x)", 2545 SCBP_C(new_pkt), 2546 new_pkt->pkt_cdbp[0], 2547 new_pkt->pkt_cdbp[1], 2548 new_pkt->pkt_cdbp[2])); 2549 } else if (sns->es_key == KEY_ILLEGAL_REQUEST) 2550 rval = VHCI_PGR_ILLEGALOP; 2551 } 2552 } else { 2553 rval = 1; 2554 } 2555 scsi_destroy_pkt(new_pkt); 2556 VHCI_DECR_PATH_CMDCOUNT(svp); 2557 freerbuf(bp); 2558 return (rval); 2559 } 2560 2561 static void 2562 vhci_run_cmd(void *arg) 2563 { 2564 struct scsi_pkt *pkt = (struct scsi_pkt *)arg; 2565 struct scsi_pkt *tpkt; 2566 scsi_vhci_priv_t *svp; 2567 mdi_pathinfo_t *pip, *npip; 2568 scsi_vhci_lun_t *vlun; 2569 dev_info_t *cdip; 2570 scsi_vhci_priv_t *nsvp; 2571 int fail = 0; 2572 int rval; 2573 struct vhci_pkt *vpkt; 2574 uchar_t cdb_1; 2575 vhci_prout_t *prout; 2576 2577 vpkt = (struct vhci_pkt *)pkt->pkt_private; 2578 tpkt = vpkt->vpkt_tgt_pkt; 2579 pip = vpkt->vpkt_path; 2580 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 2581 if (svp == NULL) { 2582 tpkt->pkt_reason = CMD_TRAN_ERR; 2583 tpkt->pkt_statistics = STAT_ABORTED; 2584 goto done; 2585 } 2586 vlun = svp->svp_svl; 2587 prout = &vlun->svl_prout; 2588 if (SCBP_C(pkt) != STATUS_GOOD) 2589 fail++; 2590 cdip = vlun->svl_dip; 2591 pip = npip = NULL; 2592 rval = mdi_select_path(cdip, NULL, 2593 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, NULL, &npip); 2594 if ((rval != MDI_SUCCESS) || (npip == NULL)) { 2595 VHCI_DEBUG(4, (CE_NOTE, NULL, 2596 "vhci_run_cmd: no path! 0x%p\n", (void *)svp)); 2597 tpkt->pkt_reason = CMD_TRAN_ERR; 2598 tpkt->pkt_statistics = STAT_ABORTED; 2599 goto done; 2600 } 2601 2602 cdb_1 = vlun->svl_cdb[1]; 2603 vlun->svl_cdb[1] &= 0xe0; 2604 vlun->svl_cdb[1] |= VHCI_PROUT_R_AND_IGNORE; 2605 2606 do { 2607 nsvp = (scsi_vhci_priv_t *) 2608 mdi_pi_get_vhci_private(npip); 2609 if (nsvp == NULL) { 2610 VHCI_DEBUG(4, (CE_NOTE, NULL, 2611 "vhci_run_cmd: no " 2612 "client priv! 0x%p offlined?\n", 2613 (void *)npip)); 2614 goto next_path; 2615 } 2616 if (vlun->svl_first_path == npip) { 2617 goto next_path; 2618 } else { 2619 if (vhci_do_prout(nsvp) != 1) 2620 fail++; 2621 } 2622 next_path: 2623 pip = npip; 2624 rval = mdi_select_path(cdip, NULL, 2625 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, 2626 pip, &npip); 2627 mdi_rele_path(pip); 2628 } while ((rval == MDI_SUCCESS) && (npip != NULL)); 2629 2630 vlun->svl_cdb[1] = cdb_1; 2631 2632 if (fail) { 2633 VHCI_DEBUG(4, (CE_WARN, NULL, "%s%d: key registration failed, " 2634 "couldn't be replicated on all paths", 2635 ddi_driver_name(cdip), ddi_get_instance(cdip))); 2636 vhci_print_prout_keys(vlun, "vhci_run_cmd: "); 2637 2638 if (SCBP_C(pkt) != STATUS_GOOD) { 2639 tpkt->pkt_reason = CMD_TRAN_ERR; 2640 tpkt->pkt_statistics = STAT_ABORTED; 2641 } 2642 } else { 2643 vlun->svl_pgr_active = 1; 2644 vhci_print_prout_keys(vlun, "vhci_run_cmd: before bcopy:"); 2645 2646 bcopy((const void *)prout->service_key, 2647 (void *)prout->active_service_key, MHIOC_RESV_KEY_SIZE); 2648 bcopy((const void *)prout->res_key, 2649 (void *)prout->active_res_key, MHIOC_RESV_KEY_SIZE); 2650 2651 vhci_print_prout_keys(vlun, "vhci_run_cmd: after bcopy:"); 2652 } 2653 done: 2654 if (SCBP_C(pkt) == STATUS_GOOD) 2655 vlun->svl_first_path = NULL; 2656 2657 if (svp) 2658 VHCI_DECR_PATH_CMDCOUNT(svp); 2659 2660 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 2661 scsi_destroy_pkt(pkt); 2662 vpkt->vpkt_hba_pkt = NULL; 2663 if (vpkt->vpkt_path) { 2664 mdi_rele_path(vpkt->vpkt_path); 2665 vpkt->vpkt_path = NULL; 2666 } 2667 } 2668 2669 sema_v(&vlun->svl_pgr_sema); 2670 /* 2671 * The PROUT commands are not included in the automatic retry 2672 * mechanism, therefore, vpkt_org_vpkt should never be set here. 2673 */ 2674 ASSERT(vpkt->vpkt_org_vpkt == NULL); 2675 if (tpkt->pkt_comp) 2676 (*tpkt->pkt_comp)(tpkt); 2677 2678 } 2679 2680 /* 2681 * Get the keys registered with this target. Since we will have 2682 * registered the same key with multiple initiators, strip out 2683 * any duplicate keys. 2684 * 2685 * The pointers which will be used to filter the registered keys from 2686 * the device will be stored in filter_prin and filter_pkt. If the 2687 * allocation length of the buffer was sufficient for the number of 2688 * parameter data bytes available to be returned by the device then the 2689 * key filtering will use the keylist returned from the original 2690 * request. If the allocation length of the buffer was not sufficient, 2691 * then the filtering will use the keylist returned from the request 2692 * that is resent below. 2693 * 2694 * If the device returns an additional length field that is greater than 2695 * the allocation length of the buffer, then allocate a new buffer which 2696 * can accommodate the number of parameter data bytes available to be 2697 * returned. Resend the scsi PRIN command, filter out the duplicate 2698 * keys and return as many of the unique keys found that was originally 2699 * requested and set the additional length field equal to the data bytes 2700 * of unique reservation keys available to be returned. 2701 * 2702 * If the device returns an additional length field that is less than or 2703 * equal to the allocation length of the buffer, then all the available 2704 * keys registered were returned by the device. Filter out the 2705 * duplicate keys and return all of the unique keys found and set the 2706 * additional length field equal to the data bytes of the reservation 2707 * keys to be returned. 2708 */ 2709 static int 2710 vhci_do_prin(struct vhci_pkt **vpkt) 2711 { 2712 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 2713 mdi_pi_get_vhci_private((*vpkt)->vpkt_path); 2714 vhci_prin_readkeys_t *prin; 2715 scsi_vhci_lun_t *vlun = svp->svp_svl; 2716 struct scsi_vhci *vhci = 2717 ADDR2VHCI(&((*vpkt)->vpkt_tgt_pkt->pkt_address)); 2718 2719 struct buf *new_bp = NULL; 2720 struct scsi_pkt *new_pkt = NULL; 2721 struct vhci_pkt *new_vpkt = NULL; 2722 int hdr_len = 0; 2723 int rval = VHCI_CMD_CMPLT; 2724 uint32_t prin_length = 0; 2725 uint32_t svl_prin_length = 0; 2726 2727 prin = (vhci_prin_readkeys_t *) 2728 bp_mapin_common((*vpkt)->vpkt_tgt_init_bp, VM_NOSLEEP); 2729 2730 if (prin != NULL) { 2731 prin_length = BE_32(prin->length); 2732 } 2733 2734 if (prin == NULL) { 2735 VHCI_DEBUG(5, (CE_WARN, NULL, 2736 "vhci_do_prin: bp_mapin_common failed.")); 2737 rval = VHCI_CMD_ERROR; 2738 } else { 2739 /* 2740 * According to SPC-3r22, sec 4.3.4.6: "If the amount of 2741 * information to be transferred exceeds the maximum value 2742 * that the ALLOCATION LENGTH field is capable of specifying, 2743 * the device server shall...terminate the command with CHECK 2744 * CONDITION status". The ALLOCATION LENGTH field of the 2745 * PERSISTENT RESERVE IN command is 2 bytes. We should never 2746 * get here with an ADDITIONAL LENGTH greater than 0xFFFF 2747 * so if we do, then it is an error! 2748 */ 2749 2750 hdr_len = sizeof (prin->length) + sizeof (prin->generation); 2751 2752 if ((prin_length + hdr_len) > 0xFFFF) { 2753 VHCI_DEBUG(5, (CE_NOTE, NULL, 2754 "vhci_do_prin: Device returned invalid " 2755 "length 0x%x\n", prin_length)); 2756 rval = VHCI_CMD_ERROR; 2757 } 2758 } 2759 2760 /* 2761 * If prin->length is greater than the byte count allocated in the 2762 * original buffer, then resend the request with enough buffer 2763 * allocated to get all of the available registered keys. 2764 */ 2765 if (rval != VHCI_CMD_ERROR) { 2766 if (((*vpkt)->vpkt_tgt_init_bp->b_bcount - hdr_len) < 2767 prin_length) { 2768 if ((*vpkt)->vpkt_org_vpkt == NULL) { 2769 new_pkt = vhci_create_retry_pkt(*vpkt); 2770 if (new_pkt != NULL) { 2771 new_vpkt = TGTPKT2VHCIPKT(new_pkt); 2772 2773 /* 2774 * This is the buf with buffer pointer 2775 * where the prin readkeys will be 2776 * returned from the device 2777 */ 2778 new_bp = scsi_alloc_consistent_buf( 2779 &svp->svp_psd->sd_address, 2780 NULL, (prin_length + hdr_len), 2781 ((*vpkt)->vpkt_tgt_init_bp-> 2782 b_flags & (B_READ | B_WRITE)), 2783 NULL_FUNC, NULL); 2784 if (new_bp != NULL) { 2785 if (new_bp->b_un.b_addr != 2786 NULL) { 2787 2788 new_bp->b_bcount = 2789 prin_length + 2790 hdr_len; 2791 2792 new_pkt->pkt_cdbp[7] = 2793 (uchar_t)(new_bp-> 2794 b_bcount >> 8); 2795 new_pkt->pkt_cdbp[8] = 2796 (uchar_t)new_bp-> 2797 b_bcount; 2798 2799 rval = VHCI_CMD_RETRY; 2800 } else { 2801 rval = VHCI_CMD_ERROR; 2802 } 2803 } else { 2804 rval = VHCI_CMD_ERROR; 2805 } 2806 } else { 2807 rval = VHCI_CMD_ERROR; 2808 } 2809 } else { 2810 rval = VHCI_CMD_ERROR; 2811 } 2812 } 2813 } 2814 2815 if (rval == VHCI_CMD_RETRY) { 2816 new_vpkt->vpkt_tgt_init_bp = new_bp; 2817 2818 /* 2819 * Release the old path because it does not matter which path 2820 * this command is sent down. This allows the normal bind 2821 * transport mechanism to be used. 2822 */ 2823 if ((*vpkt)->vpkt_path != NULL) { 2824 mdi_rele_path((*vpkt)->vpkt_path); 2825 (*vpkt)->vpkt_path = NULL; 2826 } 2827 2828 /* 2829 * Dispatch the retry command 2830 */ 2831 if (taskq_dispatch(vhci->vhci_taskq, vhci_dispatch_scsi_start, 2832 (void *) new_vpkt, KM_NOSLEEP) == NULL) { 2833 rval = VHCI_CMD_ERROR; 2834 } else { 2835 /* 2836 * If we return VHCI_CMD_RETRY, that means the caller 2837 * is going to bail and wait for the reissued command 2838 * to complete. In that case, we need to decrement 2839 * the path command count right now. In any other 2840 * case, it'll be decremented by the caller. 2841 */ 2842 VHCI_DECR_PATH_CMDCOUNT(svp); 2843 } 2844 } 2845 2846 if ((rval != VHCI_CMD_ERROR) && (rval != VHCI_CMD_RETRY)) { 2847 int new, old; 2848 int data_len = 0; 2849 2850 data_len = prin_length / MHIOC_RESV_KEY_SIZE; 2851 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_do_prin: %d keys read\n", 2852 data_len)); 2853 2854 #ifdef DEBUG 2855 VHCI_DEBUG(5, (CE_NOTE, NULL, "vhci_do_prin: from storage\n")); 2856 if (vhci_debug == 5) 2857 vhci_print_prin_keys(prin, data_len); 2858 VHCI_DEBUG(5, (CE_NOTE, NULL, 2859 "vhci_do_prin: MPxIO old keys:\n")); 2860 if (vhci_debug == 5) 2861 vhci_print_prin_keys(&vlun->svl_prin, data_len); 2862 #endif 2863 2864 /* 2865 * Filter out all duplicate keys returned from the device 2866 * We know that we use a different key for every host, so we 2867 * can simply strip out duplicates. Otherwise we would need to 2868 * do more bookkeeping to figure out which keys to strip out. 2869 */ 2870 2871 new = 0; 2872 2873 if (data_len > 0) { 2874 vlun->svl_prin.keylist[0] = prin->keylist[0]; 2875 new++; 2876 } 2877 2878 for (old = 1; old < data_len; old++) { 2879 int j; 2880 int match = 0; 2881 for (j = 0; j < new; j++) { 2882 if (bcmp(&prin->keylist[old], 2883 &vlun->svl_prin.keylist[j], 2884 sizeof (mhioc_resv_key_t)) == 0) { 2885 match = 1; 2886 break; 2887 } 2888 } 2889 if (!match) { 2890 vlun->svl_prin.keylist[new] = 2891 prin->keylist[old]; 2892 new++; 2893 } 2894 } 2895 2896 vlun->svl_prin.generation = prin->generation; 2897 svl_prin_length = new * MHIOC_RESV_KEY_SIZE; 2898 vlun->svl_prin.length = BE_32(svl_prin_length); 2899 2900 /* 2901 * If we arrived at this point after issuing a retry, make sure 2902 * that we put everything back the way it originally was so 2903 * that the target driver can complete the command correctly. 2904 */ 2905 if ((*vpkt)->vpkt_org_vpkt != NULL) { 2906 new_bp = (*vpkt)->vpkt_tgt_init_bp; 2907 2908 scsi_free_consistent_buf(new_bp); 2909 2910 *vpkt = vhci_sync_retry_pkt(*vpkt); 2911 2912 /* 2913 * Make sure the original buffer is mapped into kernel 2914 * space before we try to copy the filtered keys into 2915 * it. 2916 */ 2917 prin = (vhci_prin_readkeys_t *)bp_mapin_common( 2918 (*vpkt)->vpkt_tgt_init_bp, VM_NOSLEEP); 2919 } 2920 2921 /* 2922 * Now copy the desired number of prin keys into the original 2923 * target buffer. 2924 */ 2925 if (svl_prin_length <= 2926 ((*vpkt)->vpkt_tgt_init_bp->b_bcount - hdr_len)) { 2927 /* 2928 * It is safe to return all of the available unique 2929 * keys 2930 */ 2931 bcopy(&vlun->svl_prin, prin, svl_prin_length + hdr_len); 2932 } else { 2933 /* 2934 * Not all of the available keys were requested by the 2935 * original command. 2936 */ 2937 bcopy(&vlun->svl_prin, prin, 2938 (*vpkt)->vpkt_tgt_init_bp->b_bcount); 2939 } 2940 #ifdef DEBUG 2941 VHCI_DEBUG(5, (CE_NOTE, NULL, 2942 "vhci_do_prin: To Application:\n")); 2943 if (vhci_debug == 5) 2944 vhci_print_prin_keys(prin, new); 2945 VHCI_DEBUG(5, (CE_NOTE, NULL, 2946 "vhci_do_prin: MPxIO new keys:\n")); 2947 if (vhci_debug == 5) 2948 vhci_print_prin_keys(&vlun->svl_prin, new); 2949 #endif 2950 } 2951 2952 if (rval == VHCI_CMD_ERROR) { 2953 /* 2954 * If we arrived at this point after issuing a 2955 * retry, make sure that we put everything back 2956 * the way it originally was so that ssd can 2957 * complete the command correctly. 2958 */ 2959 2960 if ((*vpkt)->vpkt_org_vpkt != NULL) { 2961 new_bp = (*vpkt)->vpkt_tgt_init_bp; 2962 if (new_bp != NULL) { 2963 scsi_free_consistent_buf(new_bp); 2964 } 2965 2966 new_vpkt = *vpkt; 2967 *vpkt = (*vpkt)->vpkt_org_vpkt; 2968 2969 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 2970 new_vpkt->vpkt_tgt_pkt); 2971 } 2972 2973 /* 2974 * Mark this command completion as having an error so that 2975 * ssd will retry the command. 2976 */ 2977 2978 (*vpkt)->vpkt_tgt_pkt->pkt_reason = CMD_ABORTED; 2979 (*vpkt)->vpkt_tgt_pkt->pkt_statistics |= STAT_ABORTED; 2980 2981 rval = VHCI_CMD_CMPLT; 2982 } 2983 2984 /* 2985 * Make sure that the semaphore is only released once. 2986 */ 2987 if (rval == VHCI_CMD_CMPLT) { 2988 sema_v(&vlun->svl_pgr_sema); 2989 } 2990 2991 return (rval); 2992 } 2993 2994 static void 2995 vhci_intr(struct scsi_pkt *pkt) 2996 { 2997 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_private; 2998 struct scsi_pkt *tpkt; 2999 scsi_vhci_priv_t *svp; 3000 scsi_vhci_lun_t *vlun; 3001 int rval, held; 3002 struct scsi_failover_ops *fops; 3003 struct scsi_extended_sense *sns; 3004 mdi_pathinfo_t *lpath; 3005 static char *timeout_err = "Command Timeout"; 3006 static char *parity_err = "Parity Error"; 3007 char *err_str = NULL; 3008 dev_info_t *vdip, *cdip, *pdip; 3009 char *cpath, *dpath; 3010 3011 ASSERT(vpkt != NULL); 3012 tpkt = vpkt->vpkt_tgt_pkt; 3013 ASSERT(tpkt != NULL); 3014 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(vpkt->vpkt_path); 3015 ASSERT(svp != NULL); 3016 vlun = svp->svp_svl; 3017 ASSERT(vlun != NULL); 3018 lpath = vpkt->vpkt_path; 3019 3020 /* 3021 * sync up the target driver's pkt with the pkt that 3022 * we actually used 3023 */ 3024 *(tpkt->pkt_scbp) = *(pkt->pkt_scbp); 3025 tpkt->pkt_resid = pkt->pkt_resid; 3026 tpkt->pkt_state = pkt->pkt_state; 3027 tpkt->pkt_statistics = pkt->pkt_statistics; 3028 tpkt->pkt_reason = pkt->pkt_reason; 3029 3030 /* Return path_instance information back to the target driver. */ 3031 if (scsi_pkt_allocated_correctly(tpkt)) { 3032 if (scsi_pkt_allocated_correctly(pkt)) { 3033 /* 3034 * If both packets were correctly allocated, 3035 * return path returned by pHCI. 3036 */ 3037 tpkt->pkt_path_instance = pkt->pkt_path_instance; 3038 } else { 3039 /* Otherwise return path of pHCI we used */ 3040 tpkt->pkt_path_instance = 3041 mdi_pi_get_path_instance(lpath); 3042 } 3043 } 3044 3045 if (pkt->pkt_cdbp[0] == SCMD_PROUT && 3046 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 3047 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE)) { 3048 if ((SCBP_C(pkt) != STATUS_GOOD) || 3049 (pkt->pkt_reason != CMD_CMPLT)) { 3050 sema_v(&vlun->svl_pgr_sema); 3051 } 3052 } else if (pkt->pkt_cdbp[0] == SCMD_PRIN) { 3053 if (pkt->pkt_reason != CMD_CMPLT || 3054 (SCBP_C(pkt) != STATUS_GOOD)) { 3055 sema_v(&vlun->svl_pgr_sema); 3056 } 3057 } 3058 3059 switch (pkt->pkt_reason) { 3060 case CMD_CMPLT: 3061 /* 3062 * cmd completed successfully, check for scsi errors 3063 */ 3064 switch (*(pkt->pkt_scbp)) { 3065 case STATUS_CHECK: 3066 if (pkt->pkt_state & STATE_ARQ_DONE) { 3067 sns = &(((struct scsi_arq_status *)(uintptr_t) 3068 (pkt->pkt_scbp))->sts_sensedata); 3069 fops = vlun->svl_fops; 3070 ASSERT(fops != NULL); 3071 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_intr: " 3072 "Received sns key %x esc %x escq %x\n", 3073 sns->es_key, sns->es_add_code, 3074 sns->es_qual_code)); 3075 3076 if (vlun->svl_waiting_for_activepath == 1) { 3077 /* 3078 * if we are here it means we are 3079 * in the midst of a probe/attach 3080 * through a passive path; this 3081 * case is exempt from sense analysis 3082 * for detection of ext. failover 3083 * because that would unnecessarily 3084 * increase attach time. 3085 */ 3086 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 3087 vpkt->vpkt_tgt_init_scblen); 3088 break; 3089 } 3090 if (sns->es_add_code == VHCI_SCSI_PERR) { 3091 /* 3092 * parity error 3093 */ 3094 err_str = parity_err; 3095 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 3096 vpkt->vpkt_tgt_init_scblen); 3097 break; 3098 } 3099 rval = (*fops->sfo_analyze_sense) 3100 (svp->svp_psd, sns, vlun->svl_fops_ctpriv); 3101 if ((rval == SCSI_SENSE_NOFAILOVER) || 3102 (rval == SCSI_SENSE_UNKNOWN) || 3103 (rval == SCSI_SENSE_NOT_READY)) { 3104 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 3105 vpkt->vpkt_tgt_init_scblen); 3106 break; 3107 } else if (rval == SCSI_SENSE_STATE_CHANGED) { 3108 struct scsi_vhci *vhci; 3109 vhci = ADDR2VHCI(&tpkt->pkt_address); 3110 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 3111 if (!held) { 3112 /* 3113 * looks like some other thread 3114 * has already detected this 3115 * condition 3116 */ 3117 tpkt->pkt_state &= 3118 ~STATE_ARQ_DONE; 3119 *(tpkt->pkt_scbp) = 3120 STATUS_BUSY; 3121 break; 3122 } 3123 (void) taskq_dispatch( 3124 vhci->vhci_update_pathstates_taskq, 3125 vhci_update_pathstates, 3126 (void *)vlun, KM_SLEEP); 3127 } else { 3128 /* 3129 * externally initiated failover 3130 * has occurred or is in progress 3131 */ 3132 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 3133 if (!held) { 3134 /* 3135 * looks like some other thread 3136 * has already detected this 3137 * condition 3138 */ 3139 tpkt->pkt_state &= 3140 ~STATE_ARQ_DONE; 3141 *(tpkt->pkt_scbp) = 3142 STATUS_BUSY; 3143 break; 3144 } else { 3145 rval = vhci_handle_ext_fo 3146 (pkt, rval); 3147 if (rval == BUSY_RETURN) { 3148 tpkt->pkt_state &= 3149 ~STATE_ARQ_DONE; 3150 *(tpkt->pkt_scbp) = 3151 STATUS_BUSY; 3152 break; 3153 } 3154 bcopy(pkt->pkt_scbp, 3155 tpkt->pkt_scbp, 3156 vpkt->vpkt_tgt_init_scblen); 3157 break; 3158 } 3159 } 3160 } 3161 break; 3162 3163 /* 3164 * If this is a good SCSI-II RELEASE cmd completion then restore 3165 * the load balancing policy and reset VLUN_RESERVE_ACTIVE_FLG. 3166 * If this is a good SCSI-II RESERVE cmd completion then set 3167 * VLUN_RESERVE_ACTIVE_FLG. 3168 */ 3169 case STATUS_GOOD: 3170 if ((pkt->pkt_cdbp[0] == SCMD_RELEASE) || 3171 (pkt->pkt_cdbp[0] == SCMD_RELEASE_G1)) { 3172 (void) mdi_set_lb_policy(vlun->svl_dip, 3173 vlun->svl_lb_policy_save); 3174 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 3175 VHCI_DEBUG(1, (CE_WARN, NULL, 3176 "!vhci_intr: vlun 0x%p release path 0x%p", 3177 (void *)vlun, (void *)vpkt->vpkt_path)); 3178 } 3179 3180 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 3181 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 3182 vlun->svl_flags |= VLUN_RESERVE_ACTIVE_FLG; 3183 vlun->svl_resrv_pip = vpkt->vpkt_path; 3184 VHCI_DEBUG(1, (CE_WARN, NULL, 3185 "!vhci_intr: vlun 0x%p reserved path 0x%p", 3186 (void *)vlun, (void *)vpkt->vpkt_path)); 3187 } 3188 break; 3189 3190 case STATUS_RESERVATION_CONFLICT: 3191 VHCI_DEBUG(1, (CE_WARN, NULL, 3192 "!vhci_intr: vlun 0x%p " 3193 "reserve conflict on path 0x%p", 3194 (void *)vlun, (void *)vpkt->vpkt_path)); 3195 /* FALLTHROUGH */ 3196 default: 3197 break; 3198 } 3199 3200 /* 3201 * Update I/O completion statistics for the path 3202 */ 3203 mdi_pi_kstat_iosupdate(vpkt->vpkt_path, vpkt->vpkt_tgt_init_bp); 3204 3205 /* 3206 * Command completed successfully, release the dma binding and 3207 * destroy the transport side of the packet. 3208 */ 3209 if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 3210 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 3211 ((pkt->pkt_cdbp[1] & 0x1f) == 3212 VHCI_PROUT_R_AND_IGNORE))) { 3213 if (SCBP_C(pkt) == STATUS_GOOD) { 3214 ASSERT(vlun->svl_taskq); 3215 svp->svp_last_pkt_reason = pkt->pkt_reason; 3216 (void) taskq_dispatch(vlun->svl_taskq, 3217 vhci_run_cmd, pkt, KM_SLEEP); 3218 return; 3219 } 3220 } 3221 if ((SCBP_C(pkt) == STATUS_GOOD) && 3222 (pkt->pkt_cdbp[0] == SCMD_PRIN) && 3223 vpkt->vpkt_tgt_init_bp) { 3224 /* 3225 * If the action (value in byte 1 of the cdb) is zero, 3226 * we're reading keys, and that's the only condition 3227 * where we need to be concerned with filtering keys 3228 * and potential retries. Otherwise, we simply signal 3229 * the semaphore and move on. 3230 */ 3231 if (pkt->pkt_cdbp[1] == 0) { 3232 /* 3233 * If this is the completion of an internal 3234 * retry then we need to make sure that the 3235 * pkt and tpkt pointers are readjusted so 3236 * the calls to scsi_destroy_pkt and pkt_comp 3237 * below work * correctly. 3238 */ 3239 if (vpkt->vpkt_org_vpkt != NULL) { 3240 pkt = vpkt->vpkt_org_vpkt->vpkt_hba_pkt; 3241 tpkt = vpkt->vpkt_org_vpkt-> 3242 vpkt_tgt_pkt; 3243 3244 /* 3245 * If this command was issued through 3246 * the taskq then we need to clear 3247 * this flag for proper processing in 3248 * the case of a retry from the target 3249 * driver. 3250 */ 3251 vpkt->vpkt_state &= 3252 ~VHCI_PKT_THRU_TASKQ; 3253 } 3254 3255 /* 3256 * if vhci_do_prin returns VHCI_CMD_CMPLT then 3257 * vpkt will contain the address of the 3258 * original vpkt 3259 */ 3260 if (vhci_do_prin(&vpkt) == VHCI_CMD_RETRY) { 3261 /* 3262 * The command has been resent to get 3263 * all the keys from the device. Don't 3264 * complete the command with ssd until 3265 * the retry completes. 3266 */ 3267 return; 3268 } 3269 } else { 3270 sema_v(&vlun->svl_pgr_sema); 3271 } 3272 } 3273 3274 break; 3275 3276 case CMD_TIMEOUT: 3277 if ((pkt->pkt_statistics & 3278 (STAT_BUS_RESET|STAT_DEV_RESET|STAT_ABORTED)) == 0) { 3279 3280 VHCI_DEBUG(1, (CE_NOTE, NULL, 3281 "!scsi vhci timeout invoked\n")); 3282 3283 (void) vhci_recovery_reset(vlun, &pkt->pkt_address, 3284 FALSE, VHCI_DEPTH_ALL); 3285 } 3286 MDI_PI_ERRSTAT(lpath, MDI_PI_TRANSERR); 3287 tpkt->pkt_statistics |= STAT_ABORTED; 3288 err_str = timeout_err; 3289 break; 3290 3291 case CMD_TRAN_ERR: 3292 /* 3293 * This status is returned if the transport has sent the cmd 3294 * down the link to the target and then some error occurs. 3295 * In case of SCSI-II RESERVE cmd, we don't know if the 3296 * reservation been accepted by the target or not, so we need 3297 * to clear the reservation. 3298 */ 3299 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 3300 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 3301 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_intr received" 3302 " cmd_tran_err for scsi-2 reserve cmd\n")); 3303 if (!vhci_recovery_reset(vlun, &pkt->pkt_address, 3304 TRUE, VHCI_DEPTH_TARGET)) { 3305 VHCI_DEBUG(1, (CE_WARN, NULL, 3306 "!vhci_intr cmd_tran_err reset failed!")); 3307 } 3308 } 3309 break; 3310 3311 case CMD_DEV_GONE: 3312 tpkt->pkt_reason = CMD_CMPLT; 3313 tpkt->pkt_state = STATE_GOT_BUS | 3314 STATE_GOT_TARGET | STATE_SENT_CMD | 3315 STATE_GOT_STATUS; 3316 *(tpkt->pkt_scbp) = STATUS_BUSY; 3317 break; 3318 3319 default: 3320 break; 3321 } 3322 3323 /* 3324 * SCSI-II RESERVE cmd has been serviced by the lower layers clear 3325 * the flag so the lun is not QUIESCED any longer. 3326 * Also clear the VHCI_PKT_THRU_TASKQ flag, to ensure that if this pkt 3327 * is retried, a taskq shall again be dispatched to service it. Else 3328 * it may lead to a system hang if the retry is within interrupt 3329 * context. 3330 */ 3331 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 3332 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 3333 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 3334 vpkt->vpkt_state &= ~VHCI_PKT_THRU_TASKQ; 3335 } 3336 3337 /* 3338 * vpkt_org_vpkt should always be NULL here if the retry command 3339 * has been successfully processed. If vpkt_org_vpkt != NULL at 3340 * this point, it is an error so restore the original vpkt and 3341 * return an error to the target driver so it can retry the 3342 * command as appropriate. 3343 */ 3344 if (vpkt->vpkt_org_vpkt != NULL) { 3345 struct vhci_pkt *new_vpkt = vpkt; 3346 vpkt = vpkt->vpkt_org_vpkt; 3347 3348 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 3349 new_vpkt->vpkt_tgt_pkt); 3350 3351 /* 3352 * Mark this command completion as having an error so that 3353 * ssd will retry the command. 3354 */ 3355 vpkt->vpkt_tgt_pkt->pkt_reason = CMD_ABORTED; 3356 vpkt->vpkt_tgt_pkt->pkt_statistics |= STAT_ABORTED; 3357 3358 pkt = vpkt->vpkt_hba_pkt; 3359 tpkt = vpkt->vpkt_tgt_pkt; 3360 } 3361 3362 if ((err_str != NULL) && (pkt->pkt_reason != 3363 svp->svp_last_pkt_reason)) { 3364 cdip = vlun->svl_dip; 3365 pdip = mdi_pi_get_phci(vpkt->vpkt_path); 3366 vdip = ddi_get_parent(cdip); 3367 cpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3368 dpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3369 vhci_log(CE_WARN, vdip, "!%s (%s%d): %s on path %s (%s%d)", 3370 ddi_pathname(cdip, cpath), ddi_driver_name(cdip), 3371 ddi_get_instance(cdip), err_str, 3372 ddi_pathname(pdip, dpath), ddi_driver_name(pdip), 3373 ddi_get_instance(pdip)); 3374 kmem_free(cpath, MAXPATHLEN); 3375 kmem_free(dpath, MAXPATHLEN); 3376 } 3377 svp->svp_last_pkt_reason = pkt->pkt_reason; 3378 VHCI_DECR_PATH_CMDCOUNT(svp); 3379 3380 /* 3381 * For PARTIAL_DMA, vhci should not free the path. 3382 * Target driver will call into vhci_scsi_dmafree or 3383 * destroy pkt to release this path. 3384 */ 3385 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 3386 scsi_destroy_pkt(pkt); 3387 vpkt->vpkt_hba_pkt = NULL; 3388 if (vpkt->vpkt_path) { 3389 mdi_rele_path(vpkt->vpkt_path); 3390 vpkt->vpkt_path = NULL; 3391 } 3392 } 3393 3394 if (tpkt->pkt_comp) { 3395 (*tpkt->pkt_comp)(tpkt); 3396 } 3397 } 3398 3399 /* 3400 * two possibilities: (1) failover has completed 3401 * or (2) is in progress; update our path states for 3402 * the former case; for the latter case, 3403 * initiate a scsi_watch request to 3404 * determine when failover completes - vlun is HELD 3405 * until failover completes; BUSY is returned to upper 3406 * layer in both the cases 3407 */ 3408 static int 3409 vhci_handle_ext_fo(struct scsi_pkt *pkt, int fostat) 3410 { 3411 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_private; 3412 struct scsi_pkt *tpkt; 3413 scsi_vhci_priv_t *svp; 3414 scsi_vhci_lun_t *vlun; 3415 struct scsi_vhci *vhci; 3416 scsi_vhci_swarg_t *swarg; 3417 char *path; 3418 3419 ASSERT(vpkt != NULL); 3420 tpkt = vpkt->vpkt_tgt_pkt; 3421 ASSERT(tpkt != NULL); 3422 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(vpkt->vpkt_path); 3423 ASSERT(svp != NULL); 3424 vlun = svp->svp_svl; 3425 ASSERT(vlun != NULL); 3426 ASSERT(VHCI_LUN_IS_HELD(vlun)); 3427 3428 vhci = ADDR2VHCI(&tpkt->pkt_address); 3429 3430 if (fostat == SCSI_SENSE_INACTIVE) { 3431 VHCI_DEBUG(1, (CE_NOTE, NULL, "!Failover " 3432 "detected for %s; updating path states...\n", 3433 vlun->svl_lun_wwn)); 3434 /* 3435 * set the vlun flag to indicate to the task that the target 3436 * port group needs updating 3437 */ 3438 vlun->svl_flags |= VLUN_UPDATE_TPG; 3439 (void) taskq_dispatch(vhci->vhci_update_pathstates_taskq, 3440 vhci_update_pathstates, (void *)vlun, KM_SLEEP); 3441 } else { 3442 path = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3443 vhci_log(CE_NOTE, ddi_get_parent(vlun->svl_dip), 3444 "!%s (%s%d): Waiting for externally initiated failover " 3445 "to complete", ddi_pathname(vlun->svl_dip, path), 3446 ddi_driver_name(vlun->svl_dip), 3447 ddi_get_instance(vlun->svl_dip)); 3448 kmem_free(path, MAXPATHLEN); 3449 swarg = kmem_alloc(sizeof (*swarg), KM_NOSLEEP); 3450 if (swarg == NULL) { 3451 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_handle_ext_fo: " 3452 "request packet allocation for %s failed....\n", 3453 vlun->svl_lun_wwn)); 3454 VHCI_RELEASE_LUN(vlun); 3455 return (PKT_RETURN); 3456 } 3457 swarg->svs_svp = svp; 3458 swarg->svs_tos = ddi_get_time(); 3459 swarg->svs_pi = vpkt->vpkt_path; 3460 swarg->svs_release_lun = 0; 3461 swarg->svs_done = 0; 3462 /* 3463 * place a hold on the path...we don't want it to 3464 * vanish while scsi_watch is in progress 3465 */ 3466 mdi_hold_path(vpkt->vpkt_path); 3467 svp->svp_sw_token = scsi_watch_request_submit(svp->svp_psd, 3468 VHCI_FOWATCH_INTERVAL, SENSE_LENGTH, vhci_efo_watch_cb, 3469 (caddr_t)swarg); 3470 } 3471 return (BUSY_RETURN); 3472 } 3473 3474 /* 3475 * vhci_efo_watch_cb: 3476 * Callback from scsi_watch request to check the failover status. 3477 * Completion is either due to successful failover or timeout. 3478 * Upon successful completion, vhci_update_path_states is called. 3479 * For timeout condition, vhci_efo_done is called. 3480 * Always returns 0 to scsi_watch to keep retrying till vhci_efo_done 3481 * terminates this request properly in a separate thread. 3482 */ 3483 3484 static int 3485 vhci_efo_watch_cb(caddr_t arg, struct scsi_watch_result *resultp) 3486 { 3487 struct scsi_status *statusp = resultp->statusp; 3488 struct scsi_extended_sense *sensep = resultp->sensep; 3489 struct scsi_pkt *pkt = resultp->pkt; 3490 scsi_vhci_swarg_t *swarg; 3491 scsi_vhci_priv_t *svp; 3492 scsi_vhci_lun_t *vlun; 3493 struct scsi_vhci *vhci; 3494 dev_info_t *vdip; 3495 int rval, updt_paths; 3496 3497 swarg = (scsi_vhci_swarg_t *)(uintptr_t)arg; 3498 svp = swarg->svs_svp; 3499 if (swarg->svs_done) { 3500 /* 3501 * Already completed failover or timedout. 3502 * Waiting for vhci_efo_done to terminate this scsi_watch. 3503 */ 3504 return (0); 3505 } 3506 3507 ASSERT(svp != NULL); 3508 vlun = svp->svp_svl; 3509 ASSERT(vlun != NULL); 3510 ASSERT(VHCI_LUN_IS_HELD(vlun)); 3511 vlun->svl_efo_update_path = 0; 3512 vdip = ddi_get_parent(vlun->svl_dip); 3513 vhci = ddi_get_soft_state(vhci_softstate, 3514 ddi_get_instance(vdip)); 3515 3516 updt_paths = 0; 3517 3518 if (pkt->pkt_reason != CMD_CMPLT) { 3519 if ((ddi_get_time() - swarg->svs_tos) >= VHCI_EXTFO_TIMEOUT) { 3520 swarg->svs_release_lun = 1; 3521 goto done; 3522 } 3523 return (0); 3524 } 3525 if (*((unsigned char *)statusp) == STATUS_CHECK) { 3526 rval = (*(vlun->svl_fops->sfo_analyze_sense)) 3527 (svp->svp_psd, sensep, vlun->svl_fops_ctpriv); 3528 switch (rval) { 3529 /* 3530 * Only update path states in case path is definitely 3531 * inactive, or no failover occurred. For all other 3532 * check conditions continue pinging. A unexpected 3533 * check condition shouldn't cause pinging to complete 3534 * prematurely. 3535 */ 3536 case SCSI_SENSE_INACTIVE: 3537 case SCSI_SENSE_NOFAILOVER: 3538 updt_paths = 1; 3539 break; 3540 default: 3541 if ((ddi_get_time() - swarg->svs_tos) 3542 >= VHCI_EXTFO_TIMEOUT) { 3543 swarg->svs_release_lun = 1; 3544 goto done; 3545 } 3546 return (0); 3547 } 3548 } else if (*((unsigned char *)statusp) == 3549 STATUS_RESERVATION_CONFLICT) { 3550 updt_paths = 1; 3551 } else if ((*((unsigned char *)statusp)) & 3552 (STATUS_BUSY | STATUS_QFULL)) { 3553 return (0); 3554 } 3555 if ((*((unsigned char *)statusp) == STATUS_GOOD) || 3556 (updt_paths == 1)) { 3557 /* 3558 * we got here because we had detected an 3559 * externally initiated failover; things 3560 * have settled down now, so let's 3561 * start up a task to update the 3562 * path states and target port group 3563 */ 3564 vlun->svl_efo_update_path = 1; 3565 swarg->svs_done = 1; 3566 vlun->svl_swarg = swarg; 3567 vlun->svl_flags |= VLUN_UPDATE_TPG; 3568 (void) taskq_dispatch(vhci->vhci_update_pathstates_taskq, 3569 vhci_update_pathstates, (void *)vlun, 3570 KM_SLEEP); 3571 return (0); 3572 } 3573 if ((ddi_get_time() - swarg->svs_tos) >= VHCI_EXTFO_TIMEOUT) { 3574 swarg->svs_release_lun = 1; 3575 goto done; 3576 } 3577 return (0); 3578 done: 3579 swarg->svs_done = 1; 3580 (void) taskq_dispatch(vhci->vhci_taskq, 3581 vhci_efo_done, (void *)swarg, KM_SLEEP); 3582 return (0); 3583 } 3584 3585 /* 3586 * vhci_efo_done: 3587 * cleanly terminates scsi_watch and free up resources. 3588 * Called as taskq function in vhci_efo_watch_cb for EFO timeout condition 3589 * or by vhci_update_path_states invoked during external initiated 3590 * failover completion. 3591 */ 3592 static void 3593 vhci_efo_done(void *arg) 3594 { 3595 scsi_vhci_lun_t *vlun; 3596 scsi_vhci_swarg_t *swarg = (scsi_vhci_swarg_t *)arg; 3597 scsi_vhci_priv_t *svp = swarg->svs_svp; 3598 ASSERT(svp); 3599 3600 vlun = svp->svp_svl; 3601 ASSERT(vlun); 3602 3603 /* Wait for clean termination of scsi_watch */ 3604 (void) scsi_watch_request_terminate(svp->svp_sw_token, 3605 SCSI_WATCH_TERMINATE_WAIT); 3606 svp->svp_sw_token = NULL; 3607 3608 /* release path and freeup resources to indicate failover completion */ 3609 mdi_rele_path(swarg->svs_pi); 3610 if (swarg->svs_release_lun) { 3611 VHCI_RELEASE_LUN(vlun); 3612 } 3613 kmem_free((void *)swarg, sizeof (*swarg)); 3614 } 3615 3616 /* 3617 * Update the path states 3618 * vlun should be HELD when this is invoked. 3619 * Calls vhci_efo_done to cleanup resources allocated for EFO. 3620 */ 3621 void 3622 vhci_update_pathstates(void *arg) 3623 { 3624 mdi_pathinfo_t *pip, *npip; 3625 dev_info_t *dip, *pdip; 3626 struct scsi_failover_ops *fo; 3627 struct scsi_vhci_priv *svp; 3628 struct scsi_device *psd; 3629 struct scsi_path_opinfo opinfo; 3630 char *pclass, *tptr; 3631 struct scsi_vhci_lun *vlun = (struct scsi_vhci_lun *)arg; 3632 int sps; /* mdi_select_path() status */ 3633 char *cpath, *dpath; 3634 struct scsi_vhci *vhci; 3635 struct scsi_pkt *pkt; 3636 struct buf *bp; 3637 int reserve_conflict = 0; 3638 3639 ASSERT(VHCI_LUN_IS_HELD(vlun)); 3640 dip = vlun->svl_dip; 3641 pip = npip = NULL; 3642 3643 vhci = ddi_get_soft_state(vhci_softstate, 3644 ddi_get_instance(ddi_get_parent(dip))); 3645 3646 sps = mdi_select_path(dip, NULL, (MDI_SELECT_ONLINE_PATH | 3647 MDI_SELECT_STANDBY_PATH), NULL, &npip); 3648 if ((npip == NULL) || (sps != MDI_SUCCESS)) { 3649 goto done; 3650 } 3651 3652 fo = vlun->svl_fops; 3653 do { 3654 pip = npip; 3655 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 3656 psd = svp->svp_psd; 3657 if ((*fo->sfo_path_get_opinfo)(psd, &opinfo, 3658 vlun->svl_fops_ctpriv) != 0) { 3659 sps = mdi_select_path(dip, NULL, 3660 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 3661 pip, &npip); 3662 mdi_rele_path(pip); 3663 continue; 3664 } 3665 3666 if (mdi_prop_lookup_string(pip, "path-class", &pclass) != 3667 MDI_SUCCESS) { 3668 VHCI_DEBUG(1, (CE_NOTE, NULL, 3669 "!vhci_update_pathstates: prop lookup failed for " 3670 "path 0x%p\n", (void *)pip)); 3671 sps = mdi_select_path(dip, NULL, 3672 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 3673 pip, &npip); 3674 mdi_rele_path(pip); 3675 continue; 3676 } 3677 3678 /* 3679 * Need to update the "path-class" property 3680 * value in the device tree if different 3681 * from the existing value. 3682 */ 3683 if (strcmp(pclass, opinfo.opinfo_path_attr) != 0) { 3684 (void) mdi_prop_update_string(pip, "path-class", 3685 opinfo.opinfo_path_attr); 3686 } 3687 3688 /* 3689 * Only change the state if needed. i.e. Don't call 3690 * mdi_pi_set_state to ONLINE a path if its already 3691 * ONLINE. Same for STANDBY paths. 3692 */ 3693 3694 if ((opinfo.opinfo_path_state == SCSI_PATH_ACTIVE || 3695 opinfo.opinfo_path_state == SCSI_PATH_ACTIVE_NONOPT)) { 3696 if (!(MDI_PI_IS_ONLINE(pip))) { 3697 VHCI_DEBUG(1, (CE_NOTE, NULL, 3698 "!vhci_update_pathstates: marking path" 3699 " 0x%p as ONLINE\n", (void *)pip)); 3700 pdip = mdi_pi_get_phci(pip); 3701 cpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3702 dpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3703 vhci_log(CE_NOTE, ddi_get_parent(dip), "!%s" 3704 " (%s%d): path %s (%s%d) target address %s" 3705 " is now ONLINE because of" 3706 " an externally initiated failover", 3707 ddi_pathname(dip, cpath), 3708 ddi_driver_name(dip), 3709 ddi_get_instance(dip), 3710 ddi_pathname(pdip, dpath), 3711 ddi_driver_name(pdip), 3712 ddi_get_instance(pdip), 3713 mdi_pi_get_addr(pip)); 3714 kmem_free(cpath, MAXPATHLEN); 3715 kmem_free(dpath, MAXPATHLEN); 3716 mdi_pi_set_state(pip, 3717 MDI_PATHINFO_STATE_ONLINE); 3718 mdi_pi_set_preferred(pip, 3719 opinfo.opinfo_preferred); 3720 tptr = kmem_alloc(strlen 3721 (opinfo.opinfo_path_attr)+1, KM_SLEEP); 3722 (void) strlcpy(tptr, opinfo.opinfo_path_attr, 3723 (strlen(opinfo.opinfo_path_attr)+1)); 3724 mutex_enter(&vlun->svl_mutex); 3725 if (vlun->svl_active_pclass != NULL) { 3726 kmem_free(vlun->svl_active_pclass, 3727 strlen(vlun->svl_active_pclass)+1); 3728 } 3729 vlun->svl_active_pclass = tptr; 3730 if (vlun->svl_waiting_for_activepath) { 3731 vlun->svl_waiting_for_activepath = 0; 3732 } 3733 mutex_exit(&vlun->svl_mutex); 3734 /* Check for Reservation Conflict */ 3735 bp = scsi_alloc_consistent_buf( 3736 &svp->svp_psd->sd_address, 3737 (struct buf *)NULL, DEV_BSIZE, B_READ, 3738 NULL, NULL); 3739 if (!bp) { 3740 VHCI_DEBUG(1, (CE_NOTE, NULL, 3741 "vhci_update_pathstates: " 3742 "!No resources (buf)\n")); 3743 mdi_rele_path(pip); 3744 goto done; 3745 } 3746 pkt = scsi_init_pkt(&svp->svp_psd->sd_address, 3747 NULL, bp, CDB_GROUP1, 3748 sizeof (struct scsi_arq_status), 0, 3749 PKT_CONSISTENT, NULL, NULL); 3750 if (pkt) { 3751 (void) scsi_setup_cdb((union scsi_cdb *) 3752 (uintptr_t)pkt->pkt_cdbp, 3753 SCMD_READ, 1, 1, 0); 3754 pkt->pkt_time = 3*30; 3755 pkt->pkt_flags = FLAG_NOINTR; 3756 pkt->pkt_path_instance = 3757 mdi_pi_get_path_instance(pip); 3758 3759 if ((scsi_transport(pkt) == 3760 TRAN_ACCEPT) && (pkt->pkt_reason 3761 == CMD_CMPLT) && (SCBP_C(pkt) == 3762 STATUS_RESERVATION_CONFLICT)) { 3763 reserve_conflict = 1; 3764 } 3765 scsi_destroy_pkt(pkt); 3766 } 3767 scsi_free_consistent_buf(bp); 3768 } else if (MDI_PI_IS_ONLINE(pip)) { 3769 if (strcmp(pclass, opinfo.opinfo_path_attr) 3770 != 0) { 3771 mdi_pi_set_preferred(pip, 3772 opinfo.opinfo_preferred); 3773 mutex_enter(&vlun->svl_mutex); 3774 if (vlun->svl_active_pclass == NULL || 3775 strcmp(opinfo.opinfo_path_attr, 3776 vlun->svl_active_pclass) != 0) { 3777 mutex_exit(&vlun->svl_mutex); 3778 tptr = kmem_alloc(strlen 3779 (opinfo.opinfo_path_attr)+1, 3780 KM_SLEEP); 3781 (void) strlcpy(tptr, 3782 opinfo.opinfo_path_attr, 3783 (strlen 3784 (opinfo.opinfo_path_attr) 3785 +1)); 3786 mutex_enter(&vlun->svl_mutex); 3787 } else { 3788 /* 3789 * No need to update 3790 * svl_active_pclass 3791 */ 3792 tptr = NULL; 3793 mutex_exit(&vlun->svl_mutex); 3794 } 3795 if (tptr) { 3796 if (vlun->svl_active_pclass 3797 != NULL) { 3798 kmem_free(vlun-> 3799 svl_active_pclass, 3800 strlen(vlun-> 3801 svl_active_pclass) 3802 +1); 3803 } 3804 vlun->svl_active_pclass = tptr; 3805 mutex_exit(&vlun->svl_mutex); 3806 } 3807 } 3808 } 3809 } else if ((opinfo.opinfo_path_state == SCSI_PATH_INACTIVE) && 3810 !(MDI_PI_IS_STANDBY(pip))) { 3811 VHCI_DEBUG(1, (CE_NOTE, NULL, 3812 "!vhci_update_pathstates: marking path" 3813 " 0x%p as STANDBY\n", (void *)pip)); 3814 pdip = mdi_pi_get_phci(pip); 3815 cpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3816 dpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3817 vhci_log(CE_NOTE, ddi_get_parent(dip), "!%s" 3818 " (%s%d): path %s (%s%d) target address %s" 3819 " is now STANDBY because of" 3820 " an externally initiated failover", 3821 ddi_pathname(dip, cpath), 3822 ddi_driver_name(dip), 3823 ddi_get_instance(dip), 3824 ddi_pathname(pdip, dpath), 3825 ddi_driver_name(pdip), 3826 ddi_get_instance(pdip), 3827 mdi_pi_get_addr(pip)); 3828 kmem_free(cpath, MAXPATHLEN); 3829 kmem_free(dpath, MAXPATHLEN); 3830 mdi_pi_set_state(pip, 3831 MDI_PATHINFO_STATE_STANDBY); 3832 mdi_pi_set_preferred(pip, 3833 opinfo.opinfo_preferred); 3834 mutex_enter(&vlun->svl_mutex); 3835 if (vlun->svl_active_pclass != NULL) { 3836 if (strcmp(vlun->svl_active_pclass, 3837 opinfo.opinfo_path_attr) == 0) { 3838 kmem_free(vlun-> 3839 svl_active_pclass, 3840 strlen(vlun-> 3841 svl_active_pclass)+1); 3842 vlun->svl_active_pclass = NULL; 3843 } 3844 } 3845 mutex_exit(&vlun->svl_mutex); 3846 } 3847 (void) mdi_prop_free(pclass); 3848 sps = mdi_select_path(dip, NULL, 3849 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 3850 pip, &npip); 3851 mdi_rele_path(pip); 3852 3853 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 3854 3855 /* 3856 * Check to see if this vlun has an active SCSI-II RESERVE. If so 3857 * clear the reservation by sending a reset, so the host doesn't 3858 * receive a reservation conflict. 3859 * Reset VLUN_RESERVE_ACTIVE_FLG for this vlun. Also notify ssd 3860 * of the reset, explicitly. 3861 */ 3862 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 3863 if (reserve_conflict && (vlun->svl_xlf_capable == 0)) { 3864 (void) vhci_recovery_reset(vlun, 3865 &svp->svp_psd->sd_address, FALSE, 3866 VHCI_DEPTH_TARGET); 3867 } 3868 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 3869 mutex_enter(&vhci->vhci_mutex); 3870 scsi_hba_reset_notify_callback(&vhci->vhci_mutex, 3871 &vhci->vhci_reset_notify_listf); 3872 mutex_exit(&vhci->vhci_mutex); 3873 } 3874 if (vlun->svl_flags & VLUN_UPDATE_TPG) { 3875 /* 3876 * Update the AccessState of related MP-API TPGs 3877 */ 3878 (void) vhci_mpapi_update_tpg_acc_state_for_lu(vhci, vlun); 3879 vlun->svl_flags &= ~VLUN_UPDATE_TPG; 3880 } 3881 done: 3882 if (vlun->svl_efo_update_path) { 3883 vlun->svl_efo_update_path = 0; 3884 vhci_efo_done(vlun->svl_swarg); 3885 vlun->svl_swarg = 0; 3886 } 3887 VHCI_RELEASE_LUN(vlun); 3888 } 3889 3890 /* ARGSUSED */ 3891 static int 3892 vhci_pathinfo_init(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 3893 { 3894 scsi_hba_tran_t *hba = NULL; 3895 struct scsi_device *psd = NULL; 3896 scsi_vhci_lun_t *vlun = NULL; 3897 dev_info_t *pdip = NULL; 3898 dev_info_t *tgt_dip; 3899 struct scsi_vhci *vhci; 3900 char *guid; 3901 scsi_vhci_priv_t *svp = NULL; 3902 int rval = MDI_FAILURE; 3903 int vlun_alloced = 0; 3904 3905 ASSERT(vdip != NULL); 3906 ASSERT(pip != NULL); 3907 3908 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 3909 ASSERT(vhci != NULL); 3910 3911 pdip = mdi_pi_get_phci(pip); 3912 ASSERT(pdip != NULL); 3913 3914 hba = ddi_get_driver_private(pdip); 3915 ASSERT(hba != NULL); 3916 3917 tgt_dip = mdi_pi_get_client(pip); 3918 ASSERT(tgt_dip != NULL); 3919 3920 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, PROPFLAGS, 3921 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 3922 VHCI_DEBUG(1, (CE_WARN, NULL, 3923 "vhci_pathinfo_init: lun guid property failed")); 3924 goto failure; 3925 } 3926 3927 vlun = vhci_lun_lookup_alloc(tgt_dip, guid, &vlun_alloced); 3928 ddi_prop_free(guid); 3929 3930 vlun->svl_dip = tgt_dip; 3931 3932 svp = kmem_zalloc(sizeof (*svp), KM_SLEEP); 3933 svp->svp_svl = vlun; 3934 3935 vlun->svl_lb_policy_save = mdi_get_lb_policy(tgt_dip); 3936 mutex_init(&svp->svp_mutex, NULL, MUTEX_DRIVER, NULL); 3937 cv_init(&svp->svp_cv, NULL, CV_DRIVER, NULL); 3938 3939 psd = kmem_zalloc(sizeof (*psd), KM_SLEEP); 3940 mutex_init(&psd->sd_mutex, NULL, MUTEX_DRIVER, NULL); 3941 3942 /* 3943 * Clone transport structure if requested, so 3944 * Self enumerating HBAs always need to use cloning 3945 */ 3946 3947 if (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) { 3948 scsi_hba_tran_t *clone = 3949 kmem_alloc(sizeof (scsi_hba_tran_t), KM_SLEEP); 3950 bcopy(hba, clone, sizeof (scsi_hba_tran_t)); 3951 hba = clone; 3952 hba->tran_sd = psd; 3953 } else { 3954 ASSERT(hba->tran_sd == NULL); 3955 } 3956 psd->sd_dev = tgt_dip; 3957 psd->sd_address.a_hba_tran = hba; 3958 psd->sd_private = (caddr_t)pip; 3959 svp->svp_psd = psd; 3960 mdi_pi_set_vhci_private(pip, (caddr_t)svp); 3961 3962 /* 3963 * call hba's target init entry point if it exists 3964 */ 3965 if (hba->tran_tgt_init != NULL) { 3966 if ((rval = (*hba->tran_tgt_init)(pdip, tgt_dip, 3967 hba, psd)) != DDI_SUCCESS) { 3968 VHCI_DEBUG(1, (CE_WARN, pdip, 3969 "!vhci_pathinfo_init: tran_tgt_init failed for " 3970 "path=0x%p rval=%x", (void *)pip, rval)); 3971 goto failure; 3972 } 3973 } 3974 3975 svp->svp_new_path = 1; 3976 3977 psd->sd_inq = NULL; 3978 3979 VHCI_DEBUG(4, (CE_NOTE, NULL, "!vhci_pathinfo_init: path:%p\n", 3980 (void *)pip)); 3981 return (MDI_SUCCESS); 3982 3983 failure: 3984 if (psd) { 3985 mutex_destroy(&psd->sd_mutex); 3986 kmem_free(psd, sizeof (*psd)); 3987 } 3988 if (svp) { 3989 mdi_pi_set_vhci_private(pip, NULL); 3990 mutex_destroy(&svp->svp_mutex); 3991 cv_destroy(&svp->svp_cv); 3992 kmem_free(svp, sizeof (*svp)); 3993 } 3994 if (hba && hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) 3995 kmem_free(hba, sizeof (scsi_hba_tran_t)); 3996 3997 if (vlun_alloced) 3998 vhci_lun_free(tgt_dip); 3999 4000 return (rval); 4001 } 4002 4003 /* ARGSUSED */ 4004 static int 4005 vhci_pathinfo_uninit(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 4006 { 4007 scsi_hba_tran_t *hba = NULL; 4008 struct scsi_device *psd = NULL; 4009 dev_info_t *pdip = NULL; 4010 dev_info_t *cdip = NULL; 4011 scsi_vhci_priv_t *svp = NULL; 4012 4013 ASSERT(vdip != NULL); 4014 ASSERT(pip != NULL); 4015 4016 pdip = mdi_pi_get_phci(pip); 4017 ASSERT(pdip != NULL); 4018 4019 cdip = mdi_pi_get_client(pip); 4020 ASSERT(cdip != NULL); 4021 4022 hba = ddi_get_driver_private(pdip); 4023 ASSERT(hba != NULL); 4024 4025 vhci_mpapi_set_path_state(vdip, pip, MP_DRVR_PATH_STATE_REMOVED); 4026 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 4027 if (svp == NULL) { 4028 /* path already freed. Nothing to do. */ 4029 return (MDI_SUCCESS); 4030 } 4031 4032 psd = svp->svp_psd; 4033 ASSERT(psd != NULL); 4034 4035 if (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) { 4036 hba = psd->sd_address.a_hba_tran; 4037 ASSERT(hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE); 4038 ASSERT(hba->tran_sd == psd); 4039 } else { 4040 ASSERT(hba->tran_sd == NULL); 4041 } 4042 4043 if (hba->tran_tgt_free != NULL) { 4044 (*hba->tran_tgt_free) (pdip, cdip, hba, psd); 4045 } 4046 mutex_destroy(&psd->sd_mutex); 4047 if (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) { 4048 kmem_free(hba, sizeof (*hba)); 4049 } 4050 4051 mdi_pi_set_vhci_private(pip, NULL); 4052 kmem_free((caddr_t)psd, sizeof (*psd)); 4053 4054 mutex_destroy(&svp->svp_mutex); 4055 cv_destroy(&svp->svp_cv); 4056 kmem_free((caddr_t)svp, sizeof (*svp)); 4057 4058 /* 4059 * If this is the last path to the client, 4060 * then free up the vlun as well. 4061 */ 4062 if (mdi_client_get_path_count(cdip) == 1) { 4063 vhci_lun_free(cdip); 4064 } 4065 4066 VHCI_DEBUG(4, (CE_NOTE, NULL, "!vhci_pathinfo_uninit: path=0x%p\n", 4067 (void *)pip)); 4068 return (MDI_SUCCESS); 4069 } 4070 4071 /* ARGSUSED */ 4072 static int 4073 vhci_pathinfo_state_change(dev_info_t *vdip, mdi_pathinfo_t *pip, 4074 mdi_pathinfo_state_t state, uint32_t ext_state, int flags) 4075 { 4076 int rval = MDI_SUCCESS; 4077 scsi_vhci_priv_t *svp; 4078 scsi_vhci_lun_t *vlun; 4079 int held; 4080 int op = (flags & 0xf00) >> 8; 4081 struct scsi_vhci *vhci; 4082 4083 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 4084 4085 if (flags & MDI_EXT_STATE_CHANGE) { 4086 /* 4087 * We do not want to issue any commands down the path in case 4088 * sync flag is set. Lower layers might not be ready to accept 4089 * any I/O commands. 4090 */ 4091 if (op == DRIVER_DISABLE) 4092 return (MDI_SUCCESS); 4093 4094 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 4095 if (svp == NULL) { 4096 return (MDI_FAILURE); 4097 } 4098 vlun = svp->svp_svl; 4099 4100 if (flags & MDI_BEFORE_STATE_CHANGE) { 4101 /* 4102 * Hold the LUN. 4103 */ 4104 VHCI_HOLD_LUN(vlun, VH_SLEEP, held); 4105 if (flags & MDI_DISABLE_OP) { 4106 /* 4107 * Issue scsi reset if it happens to be 4108 * reserved path. 4109 */ 4110 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 4111 /* 4112 * if reservation pending on 4113 * this path, dont' mark the 4114 * path busy 4115 */ 4116 if (op == DRIVER_DISABLE_TRANSIENT) { 4117 VHCI_DEBUG(1, (CE_NOTE, NULL, 4118 "!vhci_pathinfo" 4119 "_state_change (pip:%p): " 4120 " reservation: fail busy\n", 4121 (void *)pip)); 4122 return (MDI_FAILURE); 4123 } 4124 if (pip == vlun->svl_resrv_pip) { 4125 if (vhci_recovery_reset( 4126 svp->svp_svl, 4127 &svp->svp_psd->sd_address, 4128 TRUE, 4129 VHCI_DEPTH_TARGET) == 0) { 4130 VHCI_DEBUG(1, 4131 (CE_NOTE, NULL, 4132 "!vhci_pathinfo" 4133 "_state_change " 4134 " (pip:%p): " 4135 "reset failed, " 4136 "give up!\n", 4137 (void *)pip)); 4138 } 4139 vlun->svl_flags &= 4140 ~VLUN_RESERVE_ACTIVE_FLG; 4141 } 4142 } 4143 } else if (flags & MDI_ENABLE_OP) { 4144 if (((vhci->vhci_conf_flags & 4145 VHCI_CONF_FLAGS_AUTO_FAILBACK) == 4146 VHCI_CONF_FLAGS_AUTO_FAILBACK) && 4147 MDI_PI_IS_USER_DISABLE(pip) && 4148 MDI_PI_IS_STANDBY(pip)) { 4149 struct scsi_failover_ops *fo; 4150 char *best_pclass, *pclass = NULL; 4151 int best_class, rv; 4152 /* 4153 * Failback if enabling a standby path 4154 * and it is the primary class or 4155 * preferred class 4156 */ 4157 best_class = mdi_pi_get_preferred(pip); 4158 if (best_class == 0) { 4159 /* 4160 * if not preferred - compare 4161 * path-class with class 4162 */ 4163 fo = vlun->svl_fops; 4164 (*fo->sfo_pathclass_next)(NULL, 4165 &best_pclass, 4166 vlun->svl_fops_ctpriv); 4167 pclass = NULL; 4168 rv = mdi_prop_lookup_string(pip, 4169 "path-class", &pclass); 4170 if (rv != MDI_SUCCESS || 4171 pclass == NULL) { 4172 vhci_log(CE_NOTE, vdip, 4173 "!path-class " 4174 " lookup " 4175 "failed. rv: %d" 4176 "class: %p", rv, 4177 (void *)pclass); 4178 } else if (strncmp(pclass, 4179 best_pclass, 4180 strlen(best_pclass)) == 0) { 4181 best_class = 1; 4182 } 4183 if (rv == MDI_SUCCESS && 4184 pclass != NULL) { 4185 rv = mdi_prop_free( 4186 pclass); 4187 if (rv != 4188 DDI_PROP_SUCCESS) { 4189 vhci_log( 4190 CE_NOTE, 4191 vdip, 4192 "!path-" 4193 "class" 4194 " free" 4195 " failed" 4196 " rv: %d" 4197 " class: " 4198 "%p", 4199 rv, 4200 (void *) 4201 pclass); 4202 } 4203 } 4204 } 4205 if (best_class == 1) { 4206 VHCI_DEBUG(1, (CE_NOTE, NULL, 4207 "preferred path: %p " 4208 "USER_DISABLE->USER_ENABLE " 4209 "transition for lun %s\n", 4210 (void *)pip, 4211 vlun->svl_lun_wwn)); 4212 (void) taskq_dispatch( 4213 vhci->vhci_taskq, 4214 vhci_initiate_auto_failback, 4215 (void *) vlun, KM_SLEEP); 4216 } 4217 } 4218 /* 4219 * if PGR is active, revalidate key and 4220 * register on this path also, if key is 4221 * still valid 4222 */ 4223 sema_p(&vlun->svl_pgr_sema); 4224 if (vlun->svl_pgr_active) 4225 (void) 4226 vhci_pgr_validate_and_register(svp); 4227 sema_v(&vlun->svl_pgr_sema); 4228 /* 4229 * Inform target driver about any 4230 * reservations to be reinstated if target 4231 * has dropped reservation during the busy 4232 * period. 4233 */ 4234 mutex_enter(&vhci->vhci_mutex); 4235 scsi_hba_reset_notify_callback( 4236 &vhci->vhci_mutex, 4237 &vhci->vhci_reset_notify_listf); 4238 mutex_exit(&vhci->vhci_mutex); 4239 } 4240 } 4241 if (flags & MDI_AFTER_STATE_CHANGE) { 4242 if (flags & MDI_ENABLE_OP) { 4243 mutex_enter(&vhci_global_mutex); 4244 cv_broadcast(&vhci_cv); 4245 mutex_exit(&vhci_global_mutex); 4246 } 4247 if (vlun->svl_setcap_done) { 4248 (void) vhci_pHCI_cap(&svp->svp_psd->sd_address, 4249 "sector-size", vlun->svl_sector_size, 4250 1, pip); 4251 } 4252 4253 /* 4254 * Release the LUN 4255 */ 4256 VHCI_RELEASE_LUN(vlun); 4257 4258 /* 4259 * Path transition is complete. 4260 * Run callback to indicate target driver to 4261 * retry to prevent IO starvation. 4262 */ 4263 if (scsi_callback_id != 0) { 4264 ddi_run_callback(&scsi_callback_id); 4265 } 4266 } 4267 } else { 4268 switch (state) { 4269 case MDI_PATHINFO_STATE_ONLINE: 4270 rval = vhci_pathinfo_online(vdip, pip, flags); 4271 break; 4272 4273 case MDI_PATHINFO_STATE_OFFLINE: 4274 rval = vhci_pathinfo_offline(vdip, pip, flags); 4275 break; 4276 4277 default: 4278 break; 4279 } 4280 /* 4281 * Path transition is complete. 4282 * Run callback to indicate target driver to 4283 * retry to prevent IO starvation. 4284 */ 4285 if ((rval == MDI_SUCCESS) && (scsi_callback_id != 0)) { 4286 ddi_run_callback(&scsi_callback_id); 4287 } 4288 return (rval); 4289 } 4290 4291 return (MDI_SUCCESS); 4292 } 4293 4294 /* 4295 * Parse the mpxio load balancing options. The datanameptr 4296 * will point to a string containing the load-balance-options value. 4297 * The load-balance-options value will be a property that 4298 * defines the load-balance algorithm and any arguments to that 4299 * algorithm. 4300 * For example: 4301 * device-type-mpxio-options-list= 4302 * "device-type=SUN SENA", "load-balance-options=logical-block-options" 4303 * "device-type=SUN SE6920", "round-robin-options"; 4304 * logical-block-options="load-balance=logical-block", "region-size=15"; 4305 * round-robin-options="load-balance=round-robin"; 4306 * 4307 * If the load-balance is not defined the load balance algorithm will 4308 * default to the global setting. There will be default values assigned 4309 * to the arguments (region-size=18) and if an argument is one 4310 * that is not known, it will be ignored. 4311 */ 4312 static void 4313 vhci_parse_mpxio_lb_options(dev_info_t *dip, dev_info_t *cdip, 4314 caddr_t datanameptr) 4315 { 4316 char *dataptr, *next_entry; 4317 caddr_t config_list = NULL; 4318 int config_list_len = 0, list_len = 0; 4319 int region_size = -1; 4320 client_lb_t load_balance; 4321 4322 if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, datanameptr, 4323 (caddr_t)&config_list, &config_list_len) != DDI_PROP_SUCCESS) { 4324 return; 4325 } 4326 4327 list_len = config_list_len; 4328 next_entry = config_list; 4329 while (config_list_len > 0) { 4330 dataptr = next_entry; 4331 4332 if (strncmp(mdi_load_balance, dataptr, 4333 strlen(mdi_load_balance)) == 0) { 4334 /* get the load-balance scheme */ 4335 dataptr += strlen(mdi_load_balance) + 1; 4336 if (strcmp(dataptr, LOAD_BALANCE_PROP_RR) == 0) { 4337 (void) mdi_set_lb_policy(cdip, LOAD_BALANCE_RR); 4338 load_balance = LOAD_BALANCE_RR; 4339 } else if (strcmp(dataptr, 4340 LOAD_BALANCE_PROP_LBA) == 0) { 4341 (void) mdi_set_lb_policy(cdip, 4342 LOAD_BALANCE_LBA); 4343 load_balance = LOAD_BALANCE_LBA; 4344 } else if (strcmp(dataptr, 4345 LOAD_BALANCE_PROP_NONE) == 0) { 4346 (void) mdi_set_lb_policy(cdip, 4347 LOAD_BALANCE_NONE); 4348 load_balance = LOAD_BALANCE_NONE; 4349 } 4350 } else if (strncmp(dataptr, LOGICAL_BLOCK_REGION_SIZE, 4351 strlen(LOGICAL_BLOCK_REGION_SIZE)) == 0) { 4352 int i = 0; 4353 char *ptr; 4354 char *tmp; 4355 4356 tmp = dataptr + (strlen(LOGICAL_BLOCK_REGION_SIZE) + 1); 4357 /* check for numeric value */ 4358 for (ptr = tmp; i < strlen(tmp); i++, ptr++) { 4359 if (!isdigit(*ptr)) { 4360 cmn_err(CE_WARN, 4361 "Illegal region size: %s." 4362 " Setting to default value: %d", 4363 tmp, 4364 LOAD_BALANCE_DEFAULT_REGION_SIZE); 4365 region_size = 4366 LOAD_BALANCE_DEFAULT_REGION_SIZE; 4367 break; 4368 } 4369 } 4370 if (i >= strlen(tmp)) { 4371 region_size = stoi(&tmp); 4372 } 4373 (void) mdi_set_lb_region_size(cdip, region_size); 4374 } 4375 config_list_len -= (strlen(next_entry) + 1); 4376 next_entry += strlen(next_entry) + 1; 4377 } 4378 #ifdef DEBUG 4379 if ((region_size >= 0) && (load_balance != LOAD_BALANCE_LBA)) { 4380 VHCI_DEBUG(1, (CE_NOTE, dip, 4381 "!vhci_parse_mpxio_lb_options: region-size: %d" 4382 "only valid for load-balance=logical-block\n", 4383 region_size)); 4384 } 4385 #endif 4386 if ((region_size == -1) && (load_balance == LOAD_BALANCE_LBA)) { 4387 VHCI_DEBUG(1, (CE_NOTE, dip, 4388 "!vhci_parse_mpxio_lb_options: No region-size" 4389 " defined load-balance=logical-block." 4390 " Default to: %d\n", LOAD_BALANCE_DEFAULT_REGION_SIZE)); 4391 (void) mdi_set_lb_region_size(cdip, 4392 LOAD_BALANCE_DEFAULT_REGION_SIZE); 4393 } 4394 if (list_len > 0) { 4395 kmem_free(config_list, list_len); 4396 } 4397 } 4398 4399 /* 4400 * Parse the device-type-mpxio-options-list looking for the key of 4401 * "load-balance-options". If found, parse the load balancing options. 4402 * Check the comment of the vhci_get_device_type_mpxio_options() 4403 * for the device-type-mpxio-options-list. 4404 */ 4405 static void 4406 vhci_parse_mpxio_options(dev_info_t *dip, dev_info_t *cdip, 4407 caddr_t datanameptr, int list_len) 4408 { 4409 char *dataptr; 4410 int len; 4411 4412 /* 4413 * get the data list 4414 */ 4415 dataptr = datanameptr; 4416 len = 0; 4417 while (len < list_len && 4418 strncmp(dataptr, DEVICE_TYPE_STR, strlen(DEVICE_TYPE_STR)) 4419 != 0) { 4420 if (strncmp(dataptr, LOAD_BALANCE_OPTIONS, 4421 strlen(LOAD_BALANCE_OPTIONS)) == 0) { 4422 len += strlen(LOAD_BALANCE_OPTIONS) + 1; 4423 dataptr += strlen(LOAD_BALANCE_OPTIONS) + 1; 4424 vhci_parse_mpxio_lb_options(dip, cdip, dataptr); 4425 } 4426 len += strlen(dataptr) + 1; 4427 dataptr += strlen(dataptr) + 1; 4428 } 4429 } 4430 4431 /* 4432 * Check the inquriy string returned from the device wiith the device-type 4433 * Check for the existence of the device-type-mpxio-options-list and 4434 * if found parse the list checking for a match with the device-type 4435 * value and the inquiry string returned from the device. If a match 4436 * is found, parse the mpxio options list. The format of the 4437 * device-type-mpxio-options-list is: 4438 * device-type-mpxio-options-list= 4439 * "device-type=SUN SENA", "load-balance-options=logical-block-options" 4440 * "device-type=SUN SE6920", "round-robin-options"; 4441 * logical-block-options="load-balance=logical-block", "region-size=15"; 4442 * round-robin-options="load-balance=round-robin"; 4443 */ 4444 void 4445 vhci_get_device_type_mpxio_options(dev_info_t *dip, dev_info_t *cdip, 4446 struct scsi_device *devp) 4447 { 4448 4449 caddr_t config_list = NULL; 4450 caddr_t vidptr, datanameptr; 4451 int vidlen, dupletlen = 0; 4452 int config_list_len = 0, len; 4453 struct scsi_inquiry *inq = devp->sd_inq; 4454 4455 /* 4456 * look up the device-type-mpxio-options-list and walk thru 4457 * the list compare the vendor ids of the earlier inquiry command and 4458 * with those vids in the list if there is a match, lookup 4459 * the mpxio-options value 4460 */ 4461 if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, 4462 MPXIO_OPTIONS_LIST, 4463 (caddr_t)&config_list, &config_list_len) == DDI_PROP_SUCCESS) { 4464 4465 /* 4466 * Compare vids in each duplet - if it matches, 4467 * parse the mpxio options list. 4468 */ 4469 for (len = config_list_len, vidptr = config_list; len > 0; 4470 len -= dupletlen) { 4471 4472 dupletlen = 0; 4473 4474 if (strlen(vidptr) != 0 && 4475 strncmp(vidptr, DEVICE_TYPE_STR, 4476 strlen(DEVICE_TYPE_STR)) == 0) { 4477 /* point to next duplet */ 4478 datanameptr = vidptr + strlen(vidptr) + 1; 4479 /* add len of this duplet */ 4480 dupletlen += strlen(vidptr) + 1; 4481 /* get to device type */ 4482 vidptr += strlen(DEVICE_TYPE_STR) + 1; 4483 vidlen = strlen(vidptr); 4484 if ((vidlen != 0) && 4485 bcmp(inq->inq_vid, vidptr, vidlen) == 0) { 4486 vhci_parse_mpxio_options(dip, cdip, 4487 datanameptr, len - dupletlen); 4488 break; 4489 } 4490 /* get to next duplet */ 4491 vidptr += strlen(vidptr) + 1; 4492 } 4493 /* get to the next device-type */ 4494 while (len - dupletlen > 0 && 4495 strlen(vidptr) != 0 && 4496 strncmp(vidptr, DEVICE_TYPE_STR, 4497 strlen(DEVICE_TYPE_STR)) != 0) { 4498 dupletlen += strlen(vidptr) + 1; 4499 vidptr += strlen(vidptr) + 1; 4500 } 4501 } 4502 if (config_list_len > 0) { 4503 kmem_free(config_list, config_list_len); 4504 } 4505 } 4506 } 4507 4508 static int 4509 vhci_update_pathinfo(struct scsi_device *psd, mdi_pathinfo_t *pip, 4510 struct scsi_failover_ops *fo, 4511 scsi_vhci_lun_t *vlun, 4512 struct scsi_vhci *vhci) 4513 { 4514 struct scsi_path_opinfo opinfo; 4515 char *pclass, *best_pclass; 4516 4517 if ((*fo->sfo_path_get_opinfo)(psd, &opinfo, 4518 vlun->svl_fops_ctpriv) != 0) { 4519 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_update_pathinfo: " 4520 "Failed to get operation info for path:%p\n", (void *)pip)); 4521 return (MDI_FAILURE); 4522 } 4523 /* set the xlf capable flag in the vlun for future use */ 4524 vlun->svl_xlf_capable = opinfo.opinfo_xlf_capable; 4525 (void) mdi_prop_update_string(pip, "path-class", 4526 opinfo.opinfo_path_attr); 4527 4528 pclass = opinfo.opinfo_path_attr; 4529 if (opinfo.opinfo_path_state == SCSI_PATH_ACTIVE) { 4530 mutex_enter(&vlun->svl_mutex); 4531 if (vlun->svl_active_pclass != NULL) { 4532 if (strcmp(vlun->svl_active_pclass, pclass) != 0) { 4533 mutex_exit(&vlun->svl_mutex); 4534 /* 4535 * Externally initiated failover has happened; 4536 * force the path state to be STANDBY/ONLINE, 4537 * next IO will trigger failover and thus 4538 * sync-up the pathstates. Reason we don't 4539 * sync-up immediately by invoking 4540 * vhci_update_pathstates() is because it 4541 * needs a VHCI_HOLD_LUN() and we don't 4542 * want to block here. 4543 * 4544 * Further, if the device is an ALUA device, 4545 * then failure to exactly match 'pclass' and 4546 * 'svl_active_pclass'(as is the case here) 4547 * indicates that the currently active path 4548 * is a 'non-optimized' path - which means 4549 * that 'svl_active_pclass' needs to be 4550 * replaced with opinfo.opinfo_path_state 4551 * value. 4552 */ 4553 4554 if (SCSI_FAILOVER_IS_TPGS(vlun->svl_fops)) { 4555 char *tptr; 4556 4557 /* 4558 * The device is ALUA compliant. The 4559 * state need to be changed to online 4560 * rather than standby state which is 4561 * done typically for a asymmetric 4562 * device that is non ALUA compliant. 4563 */ 4564 mdi_pi_set_state(pip, 4565 MDI_PATHINFO_STATE_ONLINE); 4566 tptr = kmem_alloc(strlen 4567 (opinfo.opinfo_path_attr)+1, 4568 KM_SLEEP); 4569 (void) strlcpy(tptr, 4570 opinfo.opinfo_path_attr, 4571 (strlen(opinfo.opinfo_path_attr) 4572 +1)); 4573 mutex_enter(&vlun->svl_mutex); 4574 kmem_free(vlun->svl_active_pclass, 4575 strlen(vlun->svl_active_pclass)+1); 4576 vlun->svl_active_pclass = tptr; 4577 mutex_exit(&vlun->svl_mutex); 4578 } else { 4579 /* 4580 * Non ALUA device case. 4581 */ 4582 mdi_pi_set_state(pip, 4583 MDI_PATHINFO_STATE_STANDBY); 4584 } 4585 vlun->svl_fo_support = opinfo.opinfo_mode; 4586 mdi_pi_set_preferred(pip, 4587 opinfo.opinfo_preferred); 4588 return (MDI_SUCCESS); 4589 } 4590 } else { 4591 char *tptr; 4592 4593 /* 4594 * lets release the mutex before we try to 4595 * allocate since the potential to sleep is 4596 * possible. 4597 */ 4598 mutex_exit(&vlun->svl_mutex); 4599 tptr = kmem_alloc(strlen(pclass)+1, KM_SLEEP); 4600 (void) strlcpy(tptr, pclass, (strlen(pclass)+1)); 4601 mutex_enter(&vlun->svl_mutex); 4602 vlun->svl_active_pclass = tptr; 4603 } 4604 mutex_exit(&vlun->svl_mutex); 4605 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_ONLINE); 4606 vlun->svl_waiting_for_activepath = 0; 4607 } else if (opinfo.opinfo_path_state == SCSI_PATH_ACTIVE_NONOPT) { 4608 mutex_enter(&vlun->svl_mutex); 4609 if (vlun->svl_active_pclass == NULL) { 4610 char *tptr; 4611 4612 mutex_exit(&vlun->svl_mutex); 4613 tptr = kmem_alloc(strlen(pclass)+1, KM_SLEEP); 4614 (void) strlcpy(tptr, pclass, (strlen(pclass)+1)); 4615 mutex_enter(&vlun->svl_mutex); 4616 vlun->svl_active_pclass = tptr; 4617 } 4618 mutex_exit(&vlun->svl_mutex); 4619 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_ONLINE); 4620 vlun->svl_waiting_for_activepath = 0; 4621 } else if (opinfo.opinfo_path_state == SCSI_PATH_INACTIVE) { 4622 mutex_enter(&vlun->svl_mutex); 4623 if (vlun->svl_active_pclass != NULL) { 4624 if (strcmp(vlun->svl_active_pclass, pclass) == 0) { 4625 mutex_exit(&vlun->svl_mutex); 4626 /* 4627 * externally initiated failover has happened; 4628 * force state to ONLINE (see comment above) 4629 */ 4630 mdi_pi_set_state(pip, 4631 MDI_PATHINFO_STATE_ONLINE); 4632 vlun->svl_fo_support = opinfo.opinfo_mode; 4633 mdi_pi_set_preferred(pip, 4634 opinfo.opinfo_preferred); 4635 return (MDI_SUCCESS); 4636 } 4637 } 4638 mutex_exit(&vlun->svl_mutex); 4639 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_STANDBY); 4640 4641 /* 4642 * Initiate auto-failback, if enabled, for path if path-state 4643 * is transitioning from OFFLINE->STANDBY and pathclass is the 4644 * prefered pathclass for this storage. 4645 * NOTE: In case where opinfo_path_state is SCSI_PATH_ACTIVE 4646 * (above), where the pi state is set to STANDBY, we don't 4647 * initiate auto-failback as the next IO shall take care of. 4648 * this. See comment above. 4649 */ 4650 (*fo->sfo_pathclass_next)(NULL, &best_pclass, 4651 vlun->svl_fops_ctpriv); 4652 if (((vhci->vhci_conf_flags & VHCI_CONF_FLAGS_AUTO_FAILBACK) == 4653 VHCI_CONF_FLAGS_AUTO_FAILBACK) && 4654 ((strcmp(pclass, best_pclass) == 0) || 4655 mdi_pi_get_preferred(pip) == 1) && 4656 ((MDI_PI_OLD_STATE(pip) == MDI_PATHINFO_STATE_OFFLINE)|| 4657 (MDI_PI_OLD_STATE(pip) == MDI_PATHINFO_STATE_INIT))) { 4658 VHCI_DEBUG(1, (CE_NOTE, NULL, "%s pathclass path: %p" 4659 " OFFLINE->STANDBY transition for lun %s\n", 4660 best_pclass, (void *)pip, vlun->svl_lun_wwn)); 4661 (void) taskq_dispatch(vhci->vhci_taskq, 4662 vhci_initiate_auto_failback, (void *) vlun, 4663 KM_SLEEP); 4664 } 4665 } 4666 vlun->svl_fo_support = opinfo.opinfo_mode; 4667 mdi_pi_set_preferred(pip, opinfo.opinfo_preferred); 4668 4669 VHCI_DEBUG(8, (CE_NOTE, NULL, "vhci_update_pathinfo: opinfo_rev = %x," 4670 " opinfo_path_state = %x opinfo_preferred = %x, opinfo_mode = %x\n", 4671 opinfo.opinfo_rev, opinfo.opinfo_path_state, 4672 opinfo.opinfo_preferred, opinfo.opinfo_mode)); 4673 4674 return (MDI_SUCCESS); 4675 } 4676 4677 /* 4678 * Form the kstat name and and call mdi_pi_kstat_create() 4679 */ 4680 void 4681 vhci_kstat_create_pathinfo(mdi_pathinfo_t *pip) 4682 { 4683 dev_info_t *tgt_dip; 4684 dev_info_t *pdip; 4685 char *guid; 4686 char *target_port, *target_port_dup; 4687 char ks_name[KSTAT_STRLEN]; 4688 uint_t pid; 4689 int by_id; 4690 mod_hash_val_t hv; 4691 4692 4693 /* return if we have already allocated kstats */ 4694 if (mdi_pi_kstat_exists(pip)) 4695 return; 4696 4697 /* 4698 * We need instance numbers to create a kstat name, return if we don't 4699 * have instance numbers assigned yet. 4700 */ 4701 tgt_dip = mdi_pi_get_client(pip); 4702 pdip = mdi_pi_get_phci(pip); 4703 if ((ddi_get_instance(tgt_dip) == -1) || (ddi_get_instance(pdip) == -1)) 4704 return; 4705 4706 /* 4707 * A path oriented kstat has a ks_name of the form: 4708 * 4709 * <client-driver><instance>.t<pid>.<pHCI-driver><instance> 4710 * 4711 * We maintain a bidirectional 'target-port' to <pid> map, 4712 * called targetmap. All pathinfo nodes with the same 4713 * 'target-port' map to the same <pid>. The iostat(1M) code, 4714 * when parsing a path oriented kstat name, uses the <pid> as 4715 * a SCSI_VHCI_GET_TARGET_LONGNAME ioctl argument in order 4716 * to get the 'target-port'. For KSTAT_FLAG_PERSISTENT kstats, 4717 * this ioctl needs to translate a <pid> to a 'target-port' 4718 * even after all pathinfo nodes associated with the 4719 * 'target-port' have been destroyed. This is needed to support 4720 * consistent first-iteration activity-since-boot iostat(1M) 4721 * output. Because of this requirement, the mapping can't be 4722 * based on pathinfo information in a devinfo snapshot. 4723 */ 4724 4725 /* determine 'target-port' */ 4726 if (mdi_prop_lookup_string(pip, 4727 "target-port", &target_port) == MDI_SUCCESS) { 4728 target_port_dup = i_ddi_strdup(target_port, KM_SLEEP); 4729 (void) mdi_prop_free(target_port); 4730 by_id = 1; 4731 } else { 4732 /* 4733 * If the pHCI did not set up 'target-port' on this 4734 * pathinfo node, assume that our client is the only 4735 * one with paths to the device by using the guid 4736 * value as the 'target-port'. Since no other client 4737 * will have the same guid, no other client will use 4738 * the same <pid>. NOTE: a client with an instance 4739 * number always has a guid. 4740 */ 4741 (void) ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, 4742 PROPFLAGS, MDI_CLIENT_GUID_PROP, &guid); 4743 target_port_dup = i_ddi_strdup(guid, KM_SLEEP); 4744 ddi_prop_free(guid); 4745 4746 /* 4747 * For this type of mapping we don't want the 4748 * <id> -> 'target-port' mapping to be made. This 4749 * will cause the SCSI_VHCI_GET_TARGET_LONGNAME ioctl 4750 * to fail, and the iostat(1M) long '-n' output will 4751 * still use the <pid>. We do this because we just 4752 * made up the 'target-port' using the guid, and we 4753 * don't want to expose that fact in iostat output. 4754 */ 4755 by_id = 0; 4756 } 4757 4758 /* find/establish <pid> given 'target-port' */ 4759 mutex_enter(&vhci_targetmap_mutex); 4760 if (mod_hash_find(vhci_targetmap_byport, 4761 (mod_hash_key_t)target_port_dup, &hv) == 0) { 4762 pid = (int)(intptr_t)hv; /* mapping exists */ 4763 } else { 4764 pid = vhci_targetmap_pid++; /* new mapping */ 4765 4766 (void) mod_hash_insert(vhci_targetmap_byport, 4767 (mod_hash_key_t)target_port_dup, 4768 (mod_hash_val_t)(intptr_t)pid); 4769 if (by_id) { 4770 (void) mod_hash_insert(vhci_targetmap_bypid, 4771 (mod_hash_key_t)(uintptr_t)pid, 4772 (mod_hash_val_t)(uintptr_t)target_port_dup); 4773 } 4774 target_port_dup = NULL; /* owned by hash */ 4775 } 4776 mutex_exit(&vhci_targetmap_mutex); 4777 4778 /* form kstat name */ 4779 (void) snprintf(ks_name, KSTAT_STRLEN, "%s%d.t%d.%s%d", 4780 ddi_driver_name(tgt_dip), ddi_get_instance(tgt_dip), 4781 pid, ddi_driver_name(pdip), ddi_get_instance(pdip)); 4782 4783 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_path_online: path:%p " 4784 "kstat %s: pid %x <-> port %s\n", (void *)pip, 4785 ks_name, pid, target_port_dup)); 4786 if (target_port_dup) 4787 kmem_free(target_port_dup, strlen(target_port_dup) + 1); 4788 4789 /* call mdi to create kstats with the name we built */ 4790 (void) mdi_pi_kstat_create(pip, ks_name); 4791 } 4792 4793 /* ARGSUSED */ 4794 static int 4795 vhci_pathinfo_online(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 4796 { 4797 scsi_hba_tran_t *hba = NULL; 4798 struct scsi_device *psd = NULL; 4799 scsi_vhci_lun_t *vlun = NULL; 4800 dev_info_t *pdip = NULL; 4801 dev_info_t *tgt_dip; 4802 struct scsi_vhci *vhci; 4803 char *guid; 4804 struct scsi_failover *sf; 4805 struct scsi_failover_ops *sfo; 4806 char *override; 4807 scsi_vhci_priv_t *svp = NULL; 4808 struct buf *bp; 4809 struct scsi_address *ap; 4810 struct scsi_pkt *pkt; 4811 int rval = MDI_FAILURE; 4812 uint_t inq_size = VHCI_STD_INQ_SIZE; 4813 mpapi_item_list_t *list_ptr; 4814 mpapi_lu_data_t *ld; 4815 4816 ASSERT(vdip != NULL); 4817 ASSERT(pip != NULL); 4818 4819 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 4820 ASSERT(vhci != NULL); 4821 4822 pdip = mdi_pi_get_phci(pip); 4823 hba = ddi_get_driver_private(pdip); 4824 ASSERT(hba != NULL); 4825 4826 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 4827 ASSERT(svp != NULL); 4828 4829 tgt_dip = mdi_pi_get_client(pip); 4830 ASSERT(tgt_dip != NULL); 4831 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, PROPFLAGS, 4832 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 4833 VHCI_DEBUG(1, (CE_WARN, NULL, "vhci_path_online: lun guid " 4834 "property failed")); 4835 goto failure; 4836 } 4837 4838 vlun = vhci_lun_lookup(tgt_dip); 4839 ASSERT(vlun != NULL); 4840 4841 ddi_prop_free(guid); 4842 4843 vlun->svl_dip = mdi_pi_get_client(pip); 4844 ASSERT(vlun->svl_dip != NULL); 4845 4846 psd = svp->svp_psd; 4847 ASSERT(psd != NULL); 4848 4849 /* 4850 * For INQUIRY response buffer size, we use VHCI_STD_INQ_SIZE(132bytes) 4851 * instead of SUN_INQSIZE(48bytes) which is used in sd layer. This is 4852 * because we could get the Vendor specific parameters(present 97th 4853 * byte onwards) which are required to process Vendor specific data 4854 * based on array type. 4855 * This INQUIRY buffer is freed in vhci_pathinfo_offline but NEVER 4856 * in a different layer like sd/phci transport. In other words, vhci 4857 * maintains its own copy of scsi_device and scsi_inquiry data on a 4858 * per-path basis. 4859 */ 4860 if (psd->sd_inq == NULL) { 4861 psd->sd_inq = (struct scsi_inquiry *) 4862 kmem_zalloc(inq_size, KM_SLEEP); 4863 } 4864 4865 tgt_dip = psd->sd_dev; 4866 ASSERT(tgt_dip != NULL); 4867 4868 /* 4869 * do inquiry to pass into probe routine; this 4870 * will avoid each probe routine doing scsi inquiry 4871 */ 4872 bp = getrbuf(KM_SLEEP); 4873 bp->b_un.b_addr = (caddr_t)psd->sd_inq; 4874 bp->b_flags = B_READ; 4875 bp->b_bcount = inq_size; 4876 bp->b_resid = 0; 4877 4878 ap = &psd->sd_address; 4879 pkt = scsi_init_pkt(ap, NULL, bp, CDB_GROUP0, 4880 sizeof (struct scsi_arq_status), 0, 0, SLEEP_FUNC, NULL); 4881 if (pkt == NULL) { 4882 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_pathinfo_online: " 4883 "Inquiry init_pkt failed :%p\n", (void *)pip)); 4884 rval = MDI_FAILURE; 4885 goto failure; 4886 } 4887 pkt->pkt_cdbp[0] = SCMD_INQUIRY; 4888 pkt->pkt_cdbp[4] = (uchar_t)inq_size; 4889 pkt->pkt_time = 60; 4890 4891 rval = vhci_do_scsi_cmd(pkt); 4892 scsi_destroy_pkt(pkt); 4893 freerbuf(bp); 4894 if (rval == 0) { 4895 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_pathinfo_online: " 4896 "Failover Inquiry failed path:%p rval:%x\n", 4897 (void *)pip, rval)); 4898 rval = MDI_FAILURE; 4899 goto failure; 4900 } 4901 4902 /* 4903 * Determine if device is supported under scsi_vhci, and select 4904 * failover module. 4905 * 4906 * See if there is a scsi_vhci.conf file override for this devices's 4907 * VID/PID. The following values can be returned: 4908 * 4909 * NULL If the NULL is returned then there is no scsi_vhci.conf 4910 * override. For NULL, we determine the failover_ops for 4911 * this device by checking the sfo_device_probe entry 4912 * point for each 'fops' module, in order. 4913 * 4914 * NOTE: Correct operation may depend on module ordering 4915 * of 'specific' (failover modules that are completely 4916 * VID/PID table based) to 'generic' (failover modules 4917 * that based on T10 standards like TPGS). Currently, 4918 * the value of 'ddi-forceload' in scsi_vhci.conf is used 4919 * to establish the module list and probe order. 4920 * 4921 * "NONE" If value "NONE" is returned then there is a 4922 * scsi_vhci.conf VID/PID override to indicate the device 4923 * should not be supported under scsi_vhci (even if there 4924 * is an 'fops' module supporting the device). 4925 * 4926 * "<other>" If another value is returned then that value is the 4927 * name of the 'fops' module that should be used. 4928 */ 4929 sfo = NULL; /* "NONE" */ 4930 override = scsi_get_device_type_string( 4931 "scsi-vhci-failover-override", vdip, psd); 4932 4933 if (override == NULL) { 4934 /* NULL: default: select based on sfo_device_probe results */ 4935 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 4936 if ((sf->sf_sfo == NULL) || 4937 ((*sf->sf_sfo->sfo_device_probe) (psd, 4938 psd->sd_inq, &vlun->svl_fops_ctpriv) == 4939 SFO_DEVICE_PROBE_PHCI)) 4940 continue; 4941 4942 /* found failover module, supported under scsi_vhci */ 4943 sfo = sf->sf_sfo; 4944 vlun->svl_fops_name = 4945 i_ddi_strdup(sfo->sfo_name, KM_SLEEP); 4946 break; 4947 } 4948 } else if (strcmp(override, "NONE") && strcmp(override, "none")) { 4949 /* !"NONE": select based on driver.conf specified name */ 4950 for (sf = scsi_failover_table, sfo = NULL; sf->sf_mod; sf++) { 4951 if ((sf->sf_sfo == NULL) || 4952 (sf->sf_sfo->sfo_name == NULL) || 4953 strcmp(override, sf->sf_sfo->sfo_name)) 4954 continue; 4955 4956 /* found failover module, supported under scsi_vhci */ 4957 sfo = sf->sf_sfo; 4958 vlun->svl_fops_name = kmem_alloc(strlen("conf ") + 4959 strlen(sfo->sfo_name) + 1, KM_SLEEP); 4960 (void) sprintf(vlun->svl_fops_name, "conf %s", 4961 sfo->sfo_name); 4962 break; 4963 } 4964 } 4965 if (override) 4966 kmem_free(override, strlen(override) + 1); 4967 4968 if (sfo == NULL) { 4969 /* no failover module - device not supported */ 4970 VHCI_DEBUG(1, (CE_NOTE, vhci->vhci_dip, 4971 "!vhci_pathinfo_online: dev (path 0x%p) not " 4972 "supported\n", (void *)pip)); 4973 vlun->svl_not_supported = 1; 4974 rval = MDI_NOT_SUPPORTED; 4975 goto done; 4976 } 4977 4978 /* failover supported for device - save failover_ops in vlun */ 4979 vlun->svl_fops = sfo; 4980 4981 /* 4982 * Obtain the device-type based mpxio options as specified in 4983 * scsi_vhci.conf file. 4984 * 4985 * NOTE: currently, the end result is a call to 4986 * mdi_set_lb_region_size(). 4987 */ 4988 vhci_get_device_type_mpxio_options(vdip, tgt_dip, psd); 4989 4990 /* 4991 * if PGR is active, revalidate key and register on this path also, 4992 * if key is still valid 4993 */ 4994 sema_p(&vlun->svl_pgr_sema); 4995 if (vlun->svl_pgr_active) { 4996 rval = vhci_pgr_validate_and_register(svp); 4997 if (rval != 1) { 4998 rval = MDI_FAILURE; 4999 sema_v(&vlun->svl_pgr_sema); 5000 goto failure; 5001 } 5002 } 5003 sema_v(&vlun->svl_pgr_sema); 5004 5005 if (svp->svp_new_path) { 5006 /* 5007 * Last chance to perform any cleanup operations on this 5008 * new path before making this path completely online. 5009 */ 5010 svp->svp_new_path = 0; 5011 5012 /* 5013 * If scsi_vhci knows the lun is alread RESERVE'd, 5014 * then skip the issue of RELEASE on new path. 5015 */ 5016 if ((vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) == 0) { 5017 /* 5018 * Issue SCSI-2 RELEASE only for the first time on 5019 * a new path just in case the host rebooted and 5020 * a reservation is still pending on this path. 5021 * IBM Shark storage does not clear RESERVE upon 5022 * host reboot. 5023 */ 5024 ap = &psd->sd_address; 5025 pkt = scsi_init_pkt(ap, NULL, NULL, CDB_GROUP0, 5026 sizeof (struct scsi_arq_status), 0, 0, 5027 SLEEP_FUNC, NULL); 5028 if (pkt == NULL) { 5029 VHCI_DEBUG(1, (CE_NOTE, NULL, 5030 "!vhci_pathinfo_online: " 5031 "Release init_pkt failed :%p\n", 5032 (void *)pip)); 5033 rval = MDI_FAILURE; 5034 goto failure; 5035 } 5036 pkt->pkt_cdbp[0] = SCMD_RELEASE; 5037 pkt->pkt_time = 60; 5038 5039 VHCI_DEBUG(1, (CE_NOTE, NULL, 5040 "!vhci_path_online: path:%p " 5041 "Issued SCSI-2 RELEASE\n", (void *)pip)); 5042 5043 /* Ignore the return value */ 5044 (void) vhci_do_scsi_cmd(pkt); 5045 scsi_destroy_pkt(pkt); 5046 } 5047 } 5048 5049 rval = vhci_update_pathinfo(psd, pip, sfo, vlun, vhci); 5050 if (rval == MDI_FAILURE) { 5051 goto failure; 5052 } 5053 5054 /* Initialize MP-API data */ 5055 vhci_update_mpapi_data(vhci, vlun, pip); 5056 5057 /* 5058 * MP-API also needs the Inquiry data to be maintained in the 5059 * mp_vendor_prop_t structure, so find the lun and update its 5060 * structure with this data. 5061 */ 5062 list_ptr = (mpapi_item_list_t *)vhci_get_mpapi_item(vhci, NULL, 5063 MP_OBJECT_TYPE_MULTIPATH_LU, (void *)vlun); 5064 ld = (mpapi_lu_data_t *)list_ptr->item->idata; 5065 if (ld != NULL) { 5066 bcopy(psd->sd_inq->inq_vid, ld->prop.prodInfo.vendor, 8); 5067 bcopy(psd->sd_inq->inq_pid, ld->prop.prodInfo.product, 16); 5068 bcopy(psd->sd_inq->inq_revision, ld->prop.prodInfo.revision, 4); 5069 } else { 5070 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_pathinfo_online: " 5071 "mpapi_lu_data_t is NULL")); 5072 } 5073 5074 /* create kstats for path */ 5075 vhci_kstat_create_pathinfo(pip); 5076 5077 done: 5078 mutex_enter(&vhci_global_mutex); 5079 cv_broadcast(&vhci_cv); 5080 mutex_exit(&vhci_global_mutex); 5081 5082 if (vlun->svl_setcap_done) { 5083 (void) vhci_pHCI_cap(ap, "sector-size", 5084 vlun->svl_sector_size, 1, pip); 5085 } 5086 5087 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_path_online: path:%p\n", 5088 (void *)pip)); 5089 5090 failure: 5091 if ((rval != MDI_SUCCESS) && psd->sd_inq) { 5092 kmem_free((caddr_t)psd->sd_inq, inq_size); 5093 psd->sd_inq = (struct scsi_inquiry *)NULL; 5094 } 5095 return (rval); 5096 } 5097 5098 /* 5099 * path offline handler. Release all bindings that will not be 5100 * released by the normal packet transport/completion code path. 5101 * Since we don't (presently) keep any bindings alive outside of 5102 * the in-transport packets (which will be released on completion) 5103 * there is not much to do here. 5104 */ 5105 /* ARGSUSED */ 5106 static int 5107 vhci_pathinfo_offline(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 5108 { 5109 scsi_hba_tran_t *hba = NULL; 5110 struct scsi_device *psd = NULL; 5111 dev_info_t *pdip = NULL; 5112 dev_info_t *cdip = NULL; 5113 scsi_vhci_priv_t *svp = NULL; 5114 uint_t inq_size = VHCI_STD_INQ_SIZE; 5115 5116 ASSERT(vdip != NULL); 5117 ASSERT(pip != NULL); 5118 5119 pdip = mdi_pi_get_phci(pip); 5120 ASSERT(pdip != NULL); 5121 if (pdip == NULL) { 5122 VHCI_DEBUG(1, (CE_WARN, vdip, "Invalid path 0x%p: NULL " 5123 "phci dip", (void *)pip)); 5124 return (MDI_FAILURE); 5125 } 5126 5127 cdip = mdi_pi_get_client(pip); 5128 ASSERT(cdip != NULL); 5129 if (cdip == NULL) { 5130 VHCI_DEBUG(1, (CE_WARN, vdip, "Invalid path 0x%p: NULL " 5131 "client dip", (void *)pip)); 5132 return (MDI_FAILURE); 5133 } 5134 5135 hba = ddi_get_driver_private(pdip); 5136 ASSERT(hba != NULL); 5137 5138 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 5139 if (svp == NULL) { 5140 /* 5141 * mdi_pathinfo node in INIT state can have vHCI private 5142 * information set to null 5143 */ 5144 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5145 "svp is NULL for pip 0x%p\n", (void *)pip)); 5146 return (MDI_SUCCESS); 5147 } 5148 5149 psd = svp->svp_psd; 5150 ASSERT(psd != NULL); 5151 5152 mutex_enter(&svp->svp_mutex); 5153 5154 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5155 "%d cmds pending on path: 0x%p\n", svp->svp_cmds, (void *)pip)); 5156 while (svp->svp_cmds != 0) { 5157 if (cv_timedwait(&svp->svp_cv, &svp->svp_mutex, 5158 ddi_get_lbolt() + 5159 drv_usectohz(vhci_path_quiesce_timeout * 1000000)) == -1) { 5160 /* 5161 * The timeout time reached without the condition 5162 * being signaled. 5163 */ 5164 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5165 "Timeout reached on path 0x%p without the cond\n", 5166 (void *)pip)); 5167 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5168 "%d cmds still pending on path: 0x%p\n", 5169 svp->svp_cmds, (void *)pip)); 5170 break; 5171 } 5172 } 5173 mutex_exit(&svp->svp_mutex); 5174 5175 /* 5176 * Check to see if this vlun has an active SCSI-II RESERVE. And this 5177 * is the pip for the path that has been reserved. 5178 * If so clear the reservation by sending a reset, so the host will not 5179 * get a reservation conflict. Reset the flag VLUN_RESERVE_ACTIVE_FLG 5180 * for this lun. Also a reset notify is sent to the target driver 5181 * just in case the POR check condition is cleared by some other layer 5182 * in the stack. 5183 */ 5184 if (svp->svp_svl->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 5185 if (pip == svp->svp_svl->svl_resrv_pip) { 5186 if (vhci_recovery_reset(svp->svp_svl, 5187 &svp->svp_psd->sd_address, TRUE, 5188 VHCI_DEPTH_TARGET) == 0) { 5189 VHCI_DEBUG(1, (CE_NOTE, NULL, 5190 "!vhci_pathinfo_offline (pip:%p):" 5191 "reset failed, retrying\n", (void *)pip)); 5192 delay(1*drv_usectohz(1000000)); 5193 if (vhci_recovery_reset(svp->svp_svl, 5194 &svp->svp_psd->sd_address, TRUE, 5195 VHCI_DEPTH_TARGET) == 0) { 5196 VHCI_DEBUG(1, (CE_NOTE, NULL, 5197 "!vhci_pathinfo_offline " 5198 "(pip:%p): reset failed, " 5199 "giving up!\n", (void *)pip)); 5200 } 5201 } 5202 svp->svp_svl->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 5203 } 5204 } 5205 5206 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_OFFLINE); 5207 if (psd->sd_inq) { 5208 kmem_free((caddr_t)psd->sd_inq, inq_size); 5209 psd->sd_inq = (struct scsi_inquiry *)NULL; 5210 } 5211 vhci_mpapi_set_path_state(vdip, pip, MP_DRVR_PATH_STATE_REMOVED); 5212 5213 VHCI_DEBUG(1, (CE_NOTE, NULL, 5214 "!vhci_pathinfo_offline: offlined path 0x%p\n", (void *)pip)); 5215 return (MDI_SUCCESS); 5216 } 5217 5218 5219 /* 5220 * routine for SCSI VHCI IOCTL implementation. 5221 */ 5222 /* ARGSUSED */ 5223 static int 5224 vhci_ctl(dev_t dev, int cmd, intptr_t data, int mode, cred_t *credp, int *rval) 5225 { 5226 struct scsi_vhci *vhci; 5227 dev_info_t *vdip; 5228 mdi_pathinfo_t *pip; 5229 int instance, held; 5230 int retval = 0; 5231 caddr_t phci_path = NULL, client_path = NULL; 5232 caddr_t paddr = NULL; 5233 sv_iocdata_t ioc; 5234 sv_iocdata_t *pioc = &ioc; 5235 sv_switch_to_cntlr_iocdata_t iocsc; 5236 sv_switch_to_cntlr_iocdata_t *piocsc = &iocsc; 5237 caddr_t s; 5238 scsi_vhci_lun_t *vlun; 5239 struct scsi_failover_ops *fo; 5240 char *pclass; 5241 5242 /* Check for validity of vhci structure */ 5243 vhci = ddi_get_soft_state(vhci_softstate, MINOR2INST(getminor(dev))); 5244 if (vhci == NULL) { 5245 return (ENXIO); 5246 } 5247 5248 mutex_enter(&vhci->vhci_mutex); 5249 if ((vhci->vhci_state & VHCI_STATE_OPEN) == 0) { 5250 mutex_exit(&vhci->vhci_mutex); 5251 return (ENXIO); 5252 } 5253 mutex_exit(&vhci->vhci_mutex); 5254 5255 /* Get the vhci dip */ 5256 vdip = vhci->vhci_dip; 5257 ASSERT(vdip != NULL); 5258 instance = ddi_get_instance(vdip); 5259 5260 /* Allocate memory for getting parameters from userland */ 5261 phci_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5262 client_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5263 paddr = kmem_zalloc(MAXNAMELEN, KM_SLEEP); 5264 5265 /* 5266 * Set a local variable indicating the ioctl name. Used for 5267 * printing debug strings. 5268 */ 5269 switch (cmd) { 5270 case SCSI_VHCI_GET_CLIENT_MULTIPATH_INFO: 5271 s = "GET_CLIENT_MULTIPATH_INFO"; 5272 break; 5273 5274 case SCSI_VHCI_GET_PHCI_MULTIPATH_INFO: 5275 s = "GET_PHCI_MULTIPATH_INFO"; 5276 break; 5277 5278 case SCSI_VHCI_GET_CLIENT_NAME: 5279 s = "GET_CLIENT_NAME"; 5280 break; 5281 5282 case SCSI_VHCI_PATH_ONLINE: 5283 s = "PATH_ONLINE"; 5284 break; 5285 5286 case SCSI_VHCI_PATH_OFFLINE: 5287 s = "PATH_OFFLINE"; 5288 break; 5289 5290 case SCSI_VHCI_PATH_STANDBY: 5291 s = "PATH_STANDBY"; 5292 break; 5293 5294 case SCSI_VHCI_PATH_TEST: 5295 s = "PATH_TEST"; 5296 break; 5297 5298 case SCSI_VHCI_SWITCH_TO_CNTLR: 5299 s = "SWITCH_TO_CNTLR"; 5300 break; 5301 case SCSI_VHCI_PATH_DISABLE: 5302 s = "PATH_DISABLE"; 5303 break; 5304 case SCSI_VHCI_PATH_ENABLE: 5305 s = "PATH_ENABLE"; 5306 break; 5307 5308 case SCSI_VHCI_GET_TARGET_LONGNAME: 5309 s = "GET_TARGET_LONGNAME"; 5310 break; 5311 5312 #ifdef DEBUG 5313 case SCSI_VHCI_CONFIGURE_PHCI: 5314 s = "CONFIGURE_PHCI"; 5315 break; 5316 5317 case SCSI_VHCI_UNCONFIGURE_PHCI: 5318 s = "UNCONFIGURE_PHCI"; 5319 break; 5320 #endif 5321 5322 default: 5323 s = "Unknown"; 5324 vhci_log(CE_NOTE, vdip, 5325 "!vhci%d: ioctl %x (unsupported ioctl)", instance, cmd); 5326 retval = ENOTSUP; 5327 break; 5328 } 5329 if (retval != 0) { 5330 goto end; 5331 } 5332 5333 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci%d: ioctl <%s>", instance, s)); 5334 5335 /* 5336 * Get IOCTL parameters from userland 5337 */ 5338 switch (cmd) { 5339 case SCSI_VHCI_GET_CLIENT_MULTIPATH_INFO: 5340 case SCSI_VHCI_GET_PHCI_MULTIPATH_INFO: 5341 case SCSI_VHCI_GET_CLIENT_NAME: 5342 case SCSI_VHCI_PATH_ONLINE: 5343 case SCSI_VHCI_PATH_OFFLINE: 5344 case SCSI_VHCI_PATH_STANDBY: 5345 case SCSI_VHCI_PATH_TEST: 5346 case SCSI_VHCI_PATH_DISABLE: 5347 case SCSI_VHCI_PATH_ENABLE: 5348 case SCSI_VHCI_GET_TARGET_LONGNAME: 5349 #ifdef DEBUG 5350 case SCSI_VHCI_CONFIGURE_PHCI: 5351 case SCSI_VHCI_UNCONFIGURE_PHCI: 5352 #endif 5353 retval = vhci_get_iocdata((const void *)data, pioc, mode, s); 5354 break; 5355 5356 case SCSI_VHCI_SWITCH_TO_CNTLR: 5357 retval = vhci_get_iocswitchdata((const void *)data, piocsc, 5358 mode, s); 5359 break; 5360 } 5361 if (retval != 0) { 5362 goto end; 5363 } 5364 5365 5366 /* 5367 * Process the IOCTL 5368 */ 5369 switch (cmd) { 5370 case SCSI_VHCI_GET_CLIENT_MULTIPATH_INFO: 5371 { 5372 uint_t num_paths; /* Num paths to client dev */ 5373 sv_path_info_t *upibuf = NULL; /* To keep userland values */ 5374 sv_path_info_t *kpibuf = NULL; /* Kernel data for ioctls */ 5375 dev_info_t *cdip; /* Client device dip */ 5376 5377 if (pioc->ret_elem == NULL) { 5378 retval = EINVAL; 5379 break; 5380 } 5381 5382 /* Get client device path from user land */ 5383 if (vhci_ioc_get_client_path(pioc, client_path, mode, s)) { 5384 retval = EFAULT; 5385 break; 5386 } 5387 5388 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5389 "client <%s>", s, client_path)); 5390 5391 /* Get number of paths to this client device */ 5392 if ((cdip = mdi_client_path2devinfo(vdip, client_path)) 5393 == NULL) { 5394 retval = ENXIO; 5395 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5396 "client dip doesn't exist. invalid path <%s>", 5397 s, client_path)); 5398 break; 5399 } 5400 num_paths = mdi_client_get_path_count(cdip); 5401 5402 if (ddi_copyout(&num_paths, pioc->ret_elem, 5403 sizeof (num_paths), mode)) { 5404 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5405 "num_paths copyout failed", s)); 5406 retval = EFAULT; 5407 break; 5408 } 5409 5410 /* If user just wanted num_paths, then return */ 5411 if (pioc->buf_elem == 0 || pioc->ret_buf == NULL || 5412 num_paths == 0) { 5413 break; 5414 } 5415 5416 /* Set num_paths to value as much as can be sent to userland */ 5417 if (num_paths > pioc->buf_elem) { 5418 num_paths = pioc->buf_elem; 5419 } 5420 5421 /* Allocate memory and get userland pointers */ 5422 if (vhci_ioc_alloc_pathinfo(&upibuf, &kpibuf, num_paths, 5423 pioc, mode, s) != 0) { 5424 retval = EFAULT; 5425 break; 5426 } 5427 ASSERT(upibuf != NULL); 5428 ASSERT(kpibuf != NULL); 5429 5430 /* 5431 * Get the path information and send it to userland. 5432 */ 5433 if (vhci_get_client_path_list(cdip, kpibuf, num_paths) 5434 != MDI_SUCCESS) { 5435 retval = ENXIO; 5436 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5437 break; 5438 } 5439 5440 if (vhci_ioc_send_pathinfo(upibuf, kpibuf, num_paths, 5441 pioc, mode, s)) { 5442 retval = EFAULT; 5443 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5444 break; 5445 } 5446 5447 /* Free the memory allocated for path information */ 5448 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5449 break; 5450 } 5451 5452 case SCSI_VHCI_GET_PHCI_MULTIPATH_INFO: 5453 { 5454 uint_t num_paths; /* Num paths to client dev */ 5455 sv_path_info_t *upibuf = NULL; /* To keep userland values */ 5456 sv_path_info_t *kpibuf = NULL; /* Kernel data for ioctls */ 5457 dev_info_t *pdip; /* PHCI device dip */ 5458 5459 if (pioc->ret_elem == NULL) { 5460 retval = EINVAL; 5461 break; 5462 } 5463 5464 /* Get PHCI device path from user land */ 5465 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 5466 retval = EFAULT; 5467 break; 5468 } 5469 5470 VHCI_DEBUG(6, (CE_WARN, vdip, 5471 "!vhci_ioctl: ioctl <%s> phci <%s>", s, phci_path)); 5472 5473 /* Get number of devices associated with this PHCI device */ 5474 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5475 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5476 "phci dip doesn't exist. invalid path <%s>", 5477 s, phci_path)); 5478 retval = ENXIO; 5479 break; 5480 } 5481 5482 num_paths = mdi_phci_get_path_count(pdip); 5483 5484 if (ddi_copyout(&num_paths, pioc->ret_elem, 5485 sizeof (num_paths), mode)) { 5486 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5487 "num_paths copyout failed", s)); 5488 retval = EFAULT; 5489 break; 5490 } 5491 5492 /* If user just wanted num_paths, then return */ 5493 if (pioc->buf_elem == 0 || pioc->ret_buf == NULL || 5494 num_paths == 0) { 5495 break; 5496 } 5497 5498 /* Set num_paths to value as much as can be sent to userland */ 5499 if (num_paths > pioc->buf_elem) { 5500 num_paths = pioc->buf_elem; 5501 } 5502 5503 /* Allocate memory and get userland pointers */ 5504 if (vhci_ioc_alloc_pathinfo(&upibuf, &kpibuf, num_paths, 5505 pioc, mode, s) != 0) { 5506 retval = EFAULT; 5507 break; 5508 } 5509 ASSERT(upibuf != NULL); 5510 ASSERT(kpibuf != NULL); 5511 5512 /* 5513 * Get the path information and send it to userland. 5514 */ 5515 if (vhci_get_phci_path_list(pdip, kpibuf, num_paths) 5516 != MDI_SUCCESS) { 5517 retval = ENXIO; 5518 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5519 break; 5520 } 5521 5522 if (vhci_ioc_send_pathinfo(upibuf, kpibuf, num_paths, 5523 pioc, mode, s)) { 5524 retval = EFAULT; 5525 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5526 break; 5527 } 5528 5529 /* Free the memory allocated for path information */ 5530 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5531 break; 5532 } 5533 5534 case SCSI_VHCI_GET_CLIENT_NAME: 5535 { 5536 dev_info_t *cdip, *pdip; 5537 5538 /* Get PHCI path and device address from user land */ 5539 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s) || 5540 vhci_ioc_get_paddr(pioc, paddr, mode, s)) { 5541 retval = EFAULT; 5542 break; 5543 } 5544 5545 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5546 "phci <%s>, paddr <%s>", s, phci_path, paddr)); 5547 5548 /* Get the PHCI dip */ 5549 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5550 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5551 "phci dip doesn't exist. invalid path <%s>", 5552 s, phci_path)); 5553 retval = ENXIO; 5554 break; 5555 } 5556 5557 if ((pip = mdi_pi_find(pdip, NULL, paddr)) == NULL) { 5558 VHCI_DEBUG(1, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5559 "pathinfo doesn't exist. invalid device addr", s)); 5560 retval = ENXIO; 5561 break; 5562 } 5563 5564 /* Get the client device pathname and send to userland */ 5565 cdip = mdi_pi_get_client(pip); 5566 vhci_ioc_devi_to_path(cdip, client_path); 5567 5568 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5569 "client <%s>", s, client_path)); 5570 5571 if (vhci_ioc_send_client_path(client_path, pioc, mode, s)) { 5572 retval = EFAULT; 5573 break; 5574 } 5575 break; 5576 } 5577 5578 case SCSI_VHCI_PATH_ONLINE: 5579 case SCSI_VHCI_PATH_OFFLINE: 5580 case SCSI_VHCI_PATH_STANDBY: 5581 case SCSI_VHCI_PATH_TEST: 5582 { 5583 dev_info_t *pdip; /* PHCI dip */ 5584 5585 /* Get PHCI path and device address from user land */ 5586 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s) || 5587 vhci_ioc_get_paddr(pioc, paddr, mode, s)) { 5588 retval = EFAULT; 5589 break; 5590 } 5591 5592 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5593 "phci <%s>, paddr <%s>", s, phci_path, paddr)); 5594 5595 /* Get the PHCI dip */ 5596 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5597 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5598 "phci dip doesn't exist. invalid path <%s>", 5599 s, phci_path)); 5600 retval = ENXIO; 5601 break; 5602 } 5603 5604 if ((pip = mdi_pi_find(pdip, NULL, paddr)) == NULL) { 5605 VHCI_DEBUG(1, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5606 "pathinfo doesn't exist. invalid device addr", s)); 5607 retval = ENXIO; 5608 break; 5609 } 5610 5611 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5612 "Calling MDI function to change device state", s)); 5613 5614 switch (cmd) { 5615 case SCSI_VHCI_PATH_ONLINE: 5616 retval = mdi_pi_online(pip, 0); 5617 break; 5618 5619 case SCSI_VHCI_PATH_OFFLINE: 5620 retval = mdi_pi_offline(pip, 0); 5621 break; 5622 5623 case SCSI_VHCI_PATH_STANDBY: 5624 retval = mdi_pi_standby(pip, 0); 5625 break; 5626 5627 case SCSI_VHCI_PATH_TEST: 5628 break; 5629 } 5630 break; 5631 } 5632 5633 case SCSI_VHCI_SWITCH_TO_CNTLR: 5634 { 5635 dev_info_t *cdip; 5636 struct scsi_device *devp; 5637 5638 /* Get the client device pathname */ 5639 if (ddi_copyin(piocsc->client, client_path, 5640 MAXPATHLEN, mode)) { 5641 VHCI_DEBUG(2, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5642 "client_path copyin failed", s)); 5643 retval = EFAULT; 5644 break; 5645 } 5646 5647 /* Get the path class to which user wants to switch */ 5648 if (ddi_copyin(piocsc->class, paddr, MAXNAMELEN, mode)) { 5649 VHCI_DEBUG(2, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5650 "controller_class copyin failed", s)); 5651 retval = EFAULT; 5652 break; 5653 } 5654 5655 /* Perform validity checks */ 5656 if ((cdip = mdi_client_path2devinfo(vdip, 5657 client_path)) == NULL) { 5658 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5659 "client dip doesn't exist. invalid path <%s>", 5660 s, client_path)); 5661 retval = ENXIO; 5662 break; 5663 } 5664 5665 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: Calling MDI func " 5666 "to switch controller")); 5667 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: client <%s> " 5668 "class <%s>", client_path, paddr)); 5669 5670 if (strcmp(paddr, PCLASS_PRIMARY) && 5671 strcmp(paddr, PCLASS_SECONDARY)) { 5672 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5673 "invalid path class <%s>", s, paddr)); 5674 retval = ENXIO; 5675 break; 5676 } 5677 5678 devp = ddi_get_driver_private(cdip); 5679 if (devp == NULL) { 5680 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5681 "invalid scsi device <%s>", s, client_path)); 5682 retval = ENXIO; 5683 break; 5684 } 5685 vlun = ADDR2VLUN(&devp->sd_address); 5686 ASSERT(vlun); 5687 5688 /* 5689 * Checking to see if device has only one pclass, PRIMARY. 5690 * If so this device doesn't support failovers. Assumed 5691 * that the devices with one pclass is PRIMARY, as thats the 5692 * case today. If this is not true and in future other 5693 * symmetric devices are supported with other pclass, this 5694 * IOCTL shall have to be overhauled anyways as now the only 5695 * arguments it accepts are PRIMARY and SECONDARY. 5696 */ 5697 fo = vlun->svl_fops; 5698 if ((*fo->sfo_pathclass_next)(PCLASS_PRIMARY, &pclass, 5699 vlun->svl_fops_ctpriv)) { 5700 retval = ENOTSUP; 5701 break; 5702 } 5703 5704 VHCI_HOLD_LUN(vlun, VH_SLEEP, held); 5705 mutex_enter(&vlun->svl_mutex); 5706 if (vlun->svl_active_pclass != NULL) { 5707 if (strcmp(vlun->svl_active_pclass, paddr) == 0) { 5708 mutex_exit(&vlun->svl_mutex); 5709 retval = EALREADY; 5710 VHCI_RELEASE_LUN(vlun); 5711 break; 5712 } 5713 } 5714 mutex_exit(&vlun->svl_mutex); 5715 /* Call mdi function to cause a switch over */ 5716 retval = mdi_failover(vdip, cdip, MDI_FAILOVER_SYNC); 5717 if (retval == MDI_SUCCESS) { 5718 retval = 0; 5719 } else if (retval == MDI_BUSY) { 5720 retval = EBUSY; 5721 } else { 5722 retval = EIO; 5723 } 5724 VHCI_RELEASE_LUN(vlun); 5725 break; 5726 } 5727 5728 case SCSI_VHCI_PATH_ENABLE: 5729 case SCSI_VHCI_PATH_DISABLE: 5730 { 5731 dev_info_t *cdip, *pdip; 5732 5733 /* 5734 * Get client device path from user land 5735 */ 5736 if (vhci_ioc_get_client_path(pioc, client_path, mode, s)) { 5737 retval = EFAULT; 5738 break; 5739 } 5740 5741 /* 5742 * Get Phci device path from user land 5743 */ 5744 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 5745 retval = EFAULT; 5746 break; 5747 } 5748 5749 /* 5750 * Get the devinfo for the Phci. 5751 */ 5752 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5753 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5754 "phci dip doesn't exist. invalid path <%s>", 5755 s, phci_path)); 5756 retval = ENXIO; 5757 break; 5758 } 5759 5760 /* 5761 * If the client path is set to /scsi_vhci then we need 5762 * to do the operation on all the clients so set cdip to NULL. 5763 * Else, try to get the client dip. 5764 */ 5765 if (strcmp(client_path, "/scsi_vhci") == 0) { 5766 cdip = NULL; 5767 } else { 5768 if ((cdip = mdi_client_path2devinfo(vdip, 5769 client_path)) == NULL) { 5770 retval = ENXIO; 5771 VHCI_DEBUG(1, (CE_WARN, NULL, 5772 "!vhci_ioctl: ioctl <%s> client dip " 5773 "doesn't exist. invalid path <%s>", 5774 s, client_path)); 5775 break; 5776 } 5777 } 5778 5779 if (cmd == SCSI_VHCI_PATH_ENABLE) 5780 retval = mdi_pi_enable(cdip, pdip, USER_DISABLE); 5781 else 5782 retval = mdi_pi_disable(cdip, pdip, USER_DISABLE); 5783 5784 break; 5785 } 5786 5787 case SCSI_VHCI_GET_TARGET_LONGNAME: 5788 { 5789 uint_t pid = pioc->buf_elem; 5790 char *target_port; 5791 mod_hash_val_t hv; 5792 5793 /* targetmap lookup of 'target-port' by <pid> */ 5794 if (mod_hash_find(vhci_targetmap_bypid, 5795 (mod_hash_key_t)(uintptr_t)pid, &hv) != 0) { 5796 /* 5797 * NOTE: failure to find the mapping is OK for guid 5798 * based 'target-port' values. 5799 */ 5800 VHCI_DEBUG(3, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5801 "targetport mapping doesn't exist: pid %d", 5802 s, pid)); 5803 retval = ENXIO; 5804 break; 5805 } 5806 5807 /* copyout 'target-port' result */ 5808 target_port = (char *)hv; 5809 if (copyoutstr(target_port, pioc->addr, MAXNAMELEN, NULL)) { 5810 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5811 "targetport copyout failed: len: %d", 5812 s, (int)strlen(target_port))); 5813 retval = EFAULT; 5814 } 5815 break; 5816 } 5817 5818 #ifdef DEBUG 5819 case SCSI_VHCI_CONFIGURE_PHCI: 5820 { 5821 dev_info_t *pdip; 5822 5823 /* Get PHCI path and device address from user land */ 5824 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 5825 retval = EFAULT; 5826 break; 5827 } 5828 5829 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5830 "phci <%s>", s, phci_path)); 5831 5832 /* Get the PHCI dip */ 5833 if ((pdip = e_ddi_hold_devi_by_path(phci_path, 0)) == NULL) { 5834 VHCI_DEBUG(3, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5835 "phci dip doesn't exist. invalid path <%s>", 5836 s, phci_path)); 5837 retval = ENXIO; 5838 break; 5839 } 5840 5841 if (ndi_devi_config(pdip, 5842 NDI_DEVFS_CLEAN|NDI_DEVI_PERSIST) != NDI_SUCCESS) { 5843 retval = EIO; 5844 } 5845 5846 ddi_release_devi(pdip); 5847 break; 5848 } 5849 5850 case SCSI_VHCI_UNCONFIGURE_PHCI: 5851 { 5852 dev_info_t *pdip; 5853 5854 /* Get PHCI path and device address from user land */ 5855 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 5856 retval = EFAULT; 5857 break; 5858 } 5859 5860 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5861 "phci <%s>", s, phci_path)); 5862 5863 /* Get the PHCI dip */ 5864 if ((pdip = e_ddi_hold_devi_by_path(phci_path, 0)) == NULL) { 5865 VHCI_DEBUG(3, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5866 "phci dip doesn't exist. invalid path <%s>", 5867 s, phci_path)); 5868 retval = ENXIO; 5869 break; 5870 } 5871 5872 if (ndi_devi_unconfig(pdip, 5873 NDI_DEVI_REMOVE|NDI_DEVFS_CLEAN) != NDI_SUCCESS) { 5874 retval = EBUSY; 5875 } 5876 5877 ddi_release_devi(pdip); 5878 break; 5879 } 5880 #endif 5881 } 5882 5883 end: 5884 /* Free the memory allocated above */ 5885 if (phci_path != NULL) { 5886 kmem_free(phci_path, MAXPATHLEN); 5887 } 5888 if (client_path != NULL) { 5889 kmem_free(client_path, MAXPATHLEN); 5890 } 5891 if (paddr != NULL) { 5892 kmem_free(paddr, MAXNAMELEN); 5893 } 5894 return (retval); 5895 } 5896 5897 /* 5898 * devctl IOCTL support for client device DR 5899 */ 5900 /* ARGSUSED */ 5901 int 5902 vhci_devctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, 5903 int *rvalp) 5904 { 5905 dev_info_t *self; 5906 dev_info_t *child; 5907 scsi_hba_tran_t *hba; 5908 struct devctl_iocdata *dcp; 5909 struct scsi_vhci *vhci; 5910 int rv = 0; 5911 int retval = 0; 5912 scsi_vhci_priv_t *svp; 5913 mdi_pathinfo_t *pip; 5914 5915 if ((vhci = ddi_get_soft_state(vhci_softstate, 5916 MINOR2INST(getminor(dev)))) == NULL) 5917 return (ENXIO); 5918 5919 /* 5920 * check if :devctl minor device has been opened 5921 */ 5922 mutex_enter(&vhci->vhci_mutex); 5923 if ((vhci->vhci_state & VHCI_STATE_OPEN) == 0) { 5924 mutex_exit(&vhci->vhci_mutex); 5925 return (ENXIO); 5926 } 5927 mutex_exit(&vhci->vhci_mutex); 5928 5929 self = vhci->vhci_dip; 5930 hba = ddi_get_driver_private(self); 5931 if (hba == NULL) 5932 return (ENXIO); 5933 5934 /* 5935 * We can use the generic implementation for these ioctls 5936 */ 5937 switch (cmd) { 5938 case DEVCTL_DEVICE_GETSTATE: 5939 case DEVCTL_DEVICE_ONLINE: 5940 case DEVCTL_DEVICE_OFFLINE: 5941 case DEVCTL_DEVICE_REMOVE: 5942 case DEVCTL_BUS_GETSTATE: 5943 return (ndi_devctl_ioctl(self, cmd, arg, mode, 0)); 5944 } 5945 5946 /* 5947 * read devctl ioctl data 5948 */ 5949 if (ndi_dc_allochdl((void *)arg, &dcp) != NDI_SUCCESS) 5950 return (EFAULT); 5951 5952 switch (cmd) { 5953 5954 case DEVCTL_DEVICE_RESET: 5955 /* 5956 * lookup and hold child device 5957 */ 5958 if ((child = ndi_devi_find(self, ndi_dc_getname(dcp), 5959 ndi_dc_getaddr(dcp))) == NULL) { 5960 rv = ENXIO; 5961 break; 5962 } 5963 retval = mdi_select_path(child, NULL, 5964 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 5965 NULL, &pip); 5966 if ((retval != MDI_SUCCESS) || (pip == NULL)) { 5967 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl:" 5968 "Unable to get a path, dip 0x%p", (void *)child)); 5969 rv = ENXIO; 5970 break; 5971 } 5972 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 5973 if (vhci_recovery_reset(svp->svp_svl, 5974 &svp->svp_psd->sd_address, TRUE, 5975 VHCI_DEPTH_TARGET) == 0) { 5976 VHCI_DEBUG(1, (CE_NOTE, NULL, 5977 "!vhci_ioctl(pip:%p): " 5978 "reset failed\n", (void *)pip)); 5979 rv = ENXIO; 5980 } 5981 mdi_rele_path(pip); 5982 break; 5983 5984 case DEVCTL_BUS_QUIESCE: 5985 case DEVCTL_BUS_UNQUIESCE: 5986 case DEVCTL_BUS_RESET: 5987 case DEVCTL_BUS_RESETALL: 5988 #ifdef DEBUG 5989 case DEVCTL_BUS_CONFIGURE: 5990 case DEVCTL_BUS_UNCONFIGURE: 5991 #endif 5992 rv = ENOTSUP; 5993 break; 5994 5995 default: 5996 rv = ENOTTY; 5997 } /* end of outer switch */ 5998 5999 ndi_dc_freehdl(dcp); 6000 return (rv); 6001 } 6002 6003 /* 6004 * Routine to get the PHCI pathname from ioctl structures in userland 6005 */ 6006 /* ARGSUSED */ 6007 static int 6008 vhci_ioc_get_phci_path(sv_iocdata_t *pioc, caddr_t phci_path, 6009 int mode, caddr_t s) 6010 { 6011 int retval = 0; 6012 6013 if (ddi_copyin(pioc->phci, phci_path, MAXPATHLEN, mode)) { 6014 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_get_phci: ioctl <%s> " 6015 "phci_path copyin failed", s)); 6016 retval = EFAULT; 6017 } 6018 return (retval); 6019 6020 } 6021 6022 6023 /* 6024 * Routine to get the Client device pathname from ioctl structures in userland 6025 */ 6026 /* ARGSUSED */ 6027 static int 6028 vhci_ioc_get_client_path(sv_iocdata_t *pioc, caddr_t client_path, 6029 int mode, caddr_t s) 6030 { 6031 int retval = 0; 6032 6033 if (ddi_copyin(pioc->client, client_path, MAXPATHLEN, mode)) { 6034 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_get_client: " 6035 "ioctl <%s> client_path copyin failed", s)); 6036 retval = EFAULT; 6037 } 6038 return (retval); 6039 } 6040 6041 6042 /* 6043 * Routine to get physical device address from ioctl structure in userland 6044 */ 6045 /* ARGSUSED */ 6046 static int 6047 vhci_ioc_get_paddr(sv_iocdata_t *pioc, caddr_t paddr, int mode, caddr_t s) 6048 { 6049 int retval = 0; 6050 6051 if (ddi_copyin(pioc->addr, paddr, MAXNAMELEN, mode)) { 6052 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_get_paddr: " 6053 "ioctl <%s> device addr copyin failed", s)); 6054 retval = EFAULT; 6055 } 6056 return (retval); 6057 } 6058 6059 6060 /* 6061 * Routine to send client device pathname to userland. 6062 */ 6063 /* ARGSUSED */ 6064 static int 6065 vhci_ioc_send_client_path(caddr_t client_path, sv_iocdata_t *pioc, 6066 int mode, caddr_t s) 6067 { 6068 int retval = 0; 6069 6070 if (ddi_copyout(client_path, pioc->client, MAXPATHLEN, mode)) { 6071 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_send_client: " 6072 "ioctl <%s> client_path copyout failed", s)); 6073 retval = EFAULT; 6074 } 6075 return (retval); 6076 } 6077 6078 6079 /* 6080 * Routine to translated dev_info pointer (dip) to device pathname. 6081 */ 6082 static void 6083 vhci_ioc_devi_to_path(dev_info_t *dip, caddr_t path) 6084 { 6085 (void) ddi_pathname(dip, path); 6086 } 6087 6088 6089 /* 6090 * vhci_get_phci_path_list: 6091 * get information about devices associated with a 6092 * given PHCI device. 6093 * 6094 * Return Values: 6095 * path information elements 6096 */ 6097 int 6098 vhci_get_phci_path_list(dev_info_t *pdip, sv_path_info_t *pibuf, 6099 uint_t num_elems) 6100 { 6101 uint_t count, done; 6102 mdi_pathinfo_t *pip; 6103 sv_path_info_t *ret_pip; 6104 int status; 6105 size_t prop_size; 6106 int circular; 6107 6108 /* 6109 * Get the PHCI structure and retrieve the path information 6110 * from the GUID hash table. 6111 */ 6112 6113 ret_pip = pibuf; 6114 count = 0; 6115 6116 ndi_devi_enter(pdip, &circular); 6117 6118 done = (count >= num_elems); 6119 pip = mdi_get_next_client_path(pdip, NULL); 6120 while (pip && !done) { 6121 mdi_pi_lock(pip); 6122 (void) ddi_pathname(mdi_pi_get_phci(pip), 6123 ret_pip->device.ret_phci); 6124 (void) strcpy(ret_pip->ret_addr, mdi_pi_get_addr(pip)); 6125 (void) mdi_pi_get_state2(pip, &ret_pip->ret_state, 6126 &ret_pip->ret_ext_state); 6127 6128 status = mdi_prop_size(pip, &prop_size); 6129 if (status == MDI_SUCCESS && ret_pip->ret_prop.ret_buf_size) { 6130 *ret_pip->ret_prop.ret_buf_size = (uint_t)prop_size; 6131 } 6132 6133 #ifdef DEBUG 6134 if (status != MDI_SUCCESS) { 6135 VHCI_DEBUG(2, (CE_WARN, NULL, 6136 "!vhci_get_phci_path_list: " 6137 "phci <%s>, prop size failure 0x%x", 6138 ret_pip->device.ret_phci, status)); 6139 } 6140 #endif /* DEBUG */ 6141 6142 6143 if (status == MDI_SUCCESS && ret_pip->ret_prop.buf && 6144 prop_size && ret_pip->ret_prop.buf_size >= prop_size) { 6145 status = mdi_prop_pack(pip, 6146 &ret_pip->ret_prop.buf, 6147 ret_pip->ret_prop.buf_size); 6148 6149 #ifdef DEBUG 6150 if (status != MDI_SUCCESS) { 6151 VHCI_DEBUG(2, (CE_WARN, NULL, 6152 "!vhci_get_phci_path_list: " 6153 "phci <%s>, prop pack failure 0x%x", 6154 ret_pip->device.ret_phci, status)); 6155 } 6156 #endif /* DEBUG */ 6157 } 6158 6159 mdi_pi_unlock(pip); 6160 pip = mdi_get_next_client_path(pdip, pip); 6161 ret_pip++; 6162 count++; 6163 done = (count >= num_elems); 6164 } 6165 6166 ndi_devi_exit(pdip, circular); 6167 6168 return (MDI_SUCCESS); 6169 } 6170 6171 6172 /* 6173 * vhci_get_client_path_list: 6174 * get information about various paths associated with a 6175 * given client device. 6176 * 6177 * Return Values: 6178 * path information elements 6179 */ 6180 int 6181 vhci_get_client_path_list(dev_info_t *cdip, sv_path_info_t *pibuf, 6182 uint_t num_elems) 6183 { 6184 uint_t count, done; 6185 mdi_pathinfo_t *pip; 6186 sv_path_info_t *ret_pip; 6187 int status; 6188 size_t prop_size; 6189 int circular; 6190 6191 ret_pip = pibuf; 6192 count = 0; 6193 6194 ndi_devi_enter(cdip, &circular); 6195 6196 done = (count >= num_elems); 6197 pip = mdi_get_next_phci_path(cdip, NULL); 6198 while (pip && !done) { 6199 mdi_pi_lock(pip); 6200 (void) ddi_pathname(mdi_pi_get_phci(pip), 6201 ret_pip->device.ret_phci); 6202 (void) strcpy(ret_pip->ret_addr, mdi_pi_get_addr(pip)); 6203 (void) mdi_pi_get_state2(pip, &ret_pip->ret_state, 6204 &ret_pip->ret_ext_state); 6205 6206 status = mdi_prop_size(pip, &prop_size); 6207 if (status == MDI_SUCCESS && ret_pip->ret_prop.ret_buf_size) { 6208 *ret_pip->ret_prop.ret_buf_size = (uint_t)prop_size; 6209 } 6210 6211 #ifdef DEBUG 6212 if (status != MDI_SUCCESS) { 6213 VHCI_DEBUG(2, (CE_WARN, NULL, 6214 "!vhci_get_client_path_list: " 6215 "phci <%s>, prop size failure 0x%x", 6216 ret_pip->device.ret_phci, status)); 6217 } 6218 #endif /* DEBUG */ 6219 6220 6221 if (status == MDI_SUCCESS && ret_pip->ret_prop.buf && 6222 prop_size && ret_pip->ret_prop.buf_size >= prop_size) { 6223 status = mdi_prop_pack(pip, 6224 &ret_pip->ret_prop.buf, 6225 ret_pip->ret_prop.buf_size); 6226 6227 #ifdef DEBUG 6228 if (status != MDI_SUCCESS) { 6229 VHCI_DEBUG(2, (CE_WARN, NULL, 6230 "!vhci_get_client_path_list: " 6231 "phci <%s>, prop pack failure 0x%x", 6232 ret_pip->device.ret_phci, status)); 6233 } 6234 #endif /* DEBUG */ 6235 } 6236 6237 mdi_pi_unlock(pip); 6238 pip = mdi_get_next_phci_path(cdip, pip); 6239 ret_pip++; 6240 count++; 6241 done = (count >= num_elems); 6242 } 6243 6244 ndi_devi_exit(cdip, circular); 6245 6246 return (MDI_SUCCESS); 6247 } 6248 6249 6250 /* 6251 * Routine to get ioctl argument structure from userland. 6252 */ 6253 /* ARGSUSED */ 6254 static int 6255 vhci_get_iocdata(const void *data, sv_iocdata_t *pioc, int mode, caddr_t s) 6256 { 6257 int retval = 0; 6258 6259 #ifdef _MULTI_DATAMODEL 6260 switch (ddi_model_convert_from(mode & FMODELS)) { 6261 case DDI_MODEL_ILP32: 6262 { 6263 sv_iocdata32_t ioc32; 6264 6265 if (ddi_copyin(data, &ioc32, sizeof (ioc32), mode)) { 6266 retval = EFAULT; 6267 break; 6268 } 6269 pioc->client = (caddr_t)(uintptr_t)ioc32.client; 6270 pioc->phci = (caddr_t)(uintptr_t)ioc32.phci; 6271 pioc->addr = (caddr_t)(uintptr_t)ioc32.addr; 6272 pioc->buf_elem = (uint_t)ioc32.buf_elem; 6273 pioc->ret_buf = (sv_path_info_t *)(uintptr_t)ioc32.ret_buf; 6274 pioc->ret_elem = (uint_t *)(uintptr_t)ioc32.ret_elem; 6275 break; 6276 } 6277 6278 case DDI_MODEL_NONE: 6279 if (ddi_copyin(data, pioc, sizeof (*pioc), mode)) { 6280 retval = EFAULT; 6281 break; 6282 } 6283 break; 6284 } 6285 #else /* _MULTI_DATAMODEL */ 6286 if (ddi_copyin(data, pioc, sizeof (*pioc), mode)) { 6287 retval = EFAULT; 6288 } 6289 #endif /* _MULTI_DATAMODEL */ 6290 6291 #ifdef DEBUG 6292 if (retval) { 6293 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: cmd <%s> " 6294 "iocdata copyin failed", s)); 6295 } 6296 #endif 6297 6298 return (retval); 6299 } 6300 6301 6302 /* 6303 * Routine to get the ioctl argument for ioctl causing controller switchover. 6304 */ 6305 /* ARGSUSED */ 6306 static int 6307 vhci_get_iocswitchdata(const void *data, sv_switch_to_cntlr_iocdata_t *piocsc, 6308 int mode, caddr_t s) 6309 { 6310 int retval = 0; 6311 6312 #ifdef _MULTI_DATAMODEL 6313 switch (ddi_model_convert_from(mode & FMODELS)) { 6314 case DDI_MODEL_ILP32: 6315 { 6316 sv_switch_to_cntlr_iocdata32_t ioc32; 6317 6318 if (ddi_copyin(data, &ioc32, sizeof (ioc32), mode)) { 6319 retval = EFAULT; 6320 break; 6321 } 6322 piocsc->client = (caddr_t)(uintptr_t)ioc32.client; 6323 piocsc->class = (caddr_t)(uintptr_t)ioc32.class; 6324 break; 6325 } 6326 6327 case DDI_MODEL_NONE: 6328 if (ddi_copyin(data, piocsc, sizeof (*piocsc), mode)) { 6329 retval = EFAULT; 6330 } 6331 break; 6332 } 6333 #else /* _MULTI_DATAMODEL */ 6334 if (ddi_copyin(data, piocsc, sizeof (*piocsc), mode)) { 6335 retval = EFAULT; 6336 } 6337 #endif /* _MULTI_DATAMODEL */ 6338 6339 #ifdef DEBUG 6340 if (retval) { 6341 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: cmd <%s> " 6342 "switch_to_cntlr_iocdata copyin failed", s)); 6343 } 6344 #endif 6345 6346 return (retval); 6347 } 6348 6349 6350 /* 6351 * Routine to allocate memory for the path information structures. 6352 * It allocates two chunks of memory - one for keeping userland 6353 * pointers/values for path information and path properties, second for 6354 * keeping allocating kernel memory for path properties. These path 6355 * properties are finally copied to userland. 6356 */ 6357 /* ARGSUSED */ 6358 static int 6359 vhci_ioc_alloc_pathinfo(sv_path_info_t **upibuf, sv_path_info_t **kpibuf, 6360 uint_t num_paths, sv_iocdata_t *pioc, int mode, caddr_t s) 6361 { 6362 sv_path_info_t *pi; 6363 uint_t bufsize; 6364 int retval = 0; 6365 int index; 6366 6367 /* Allocate memory */ 6368 *upibuf = (sv_path_info_t *) 6369 kmem_zalloc(sizeof (sv_path_info_t) * num_paths, KM_SLEEP); 6370 ASSERT(*upibuf != NULL); 6371 *kpibuf = (sv_path_info_t *) 6372 kmem_zalloc(sizeof (sv_path_info_t) * num_paths, KM_SLEEP); 6373 ASSERT(*kpibuf != NULL); 6374 6375 /* 6376 * Get the path info structure from the user space. 6377 * We are interested in the following fields: 6378 * - user size of buffer for per path properties. 6379 * - user address of buffer for path info properties. 6380 * - user pointer for returning actual buffer size 6381 * Keep these fields in the 'upibuf' structures. 6382 * Allocate buffer for per path info properties in kernel 6383 * structure ('kpibuf'). 6384 * Size of these buffers will be equal to the size of buffers 6385 * in the user space. 6386 */ 6387 #ifdef _MULTI_DATAMODEL 6388 switch (ddi_model_convert_from(mode & FMODELS)) { 6389 case DDI_MODEL_ILP32: 6390 { 6391 sv_path_info32_t *src; 6392 sv_path_info32_t pi32; 6393 6394 src = (sv_path_info32_t *)pioc->ret_buf; 6395 pi = (sv_path_info_t *)*upibuf; 6396 for (index = 0; index < num_paths; index++, src++, pi++) { 6397 if (ddi_copyin(src, &pi32, sizeof (pi32), mode)) { 6398 retval = EFAULT; 6399 break; 6400 } 6401 6402 pi->ret_prop.buf_size = 6403 (uint_t)pi32.ret_prop.buf_size; 6404 pi->ret_prop.ret_buf_size = 6405 (uint_t *)(uintptr_t)pi32.ret_prop.ret_buf_size; 6406 pi->ret_prop.buf = 6407 (caddr_t)(uintptr_t)pi32.ret_prop.buf; 6408 } 6409 break; 6410 } 6411 6412 case DDI_MODEL_NONE: 6413 if (ddi_copyin(pioc->ret_buf, *upibuf, 6414 sizeof (sv_path_info_t) * num_paths, mode)) { 6415 retval = EFAULT; 6416 } 6417 break; 6418 } 6419 #else /* _MULTI_DATAMODEL */ 6420 if (ddi_copyin(pioc->ret_buf, *upibuf, 6421 sizeof (sv_path_info_t) * num_paths, mode)) { 6422 retval = EFAULT; 6423 } 6424 #endif /* _MULTI_DATAMODEL */ 6425 6426 if (retval != 0) { 6427 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_alloc_path_info: " 6428 "ioctl <%s> normal: path_info copyin failed", s)); 6429 kmem_free(*upibuf, sizeof (sv_path_info_t) * num_paths); 6430 kmem_free(*kpibuf, sizeof (sv_path_info_t) * num_paths); 6431 *upibuf = NULL; 6432 *kpibuf = NULL; 6433 return (retval); 6434 } 6435 6436 /* 6437 * Allocate memory for per path properties. 6438 */ 6439 for (index = 0, pi = *kpibuf; index < num_paths; index++, pi++) { 6440 bufsize = (*upibuf)[index].ret_prop.buf_size; 6441 6442 if (bufsize && bufsize <= SV_PROP_MAX_BUF_SIZE) { 6443 pi->ret_prop.buf_size = bufsize; 6444 pi->ret_prop.buf = (caddr_t) 6445 kmem_zalloc(bufsize, KM_SLEEP); 6446 ASSERT(pi->ret_prop.buf != NULL); 6447 } else { 6448 pi->ret_prop.buf_size = 0; 6449 pi->ret_prop.buf = NULL; 6450 } 6451 6452 if ((*upibuf)[index].ret_prop.ret_buf_size != NULL) { 6453 pi->ret_prop.ret_buf_size = (uint_t *)kmem_zalloc( 6454 sizeof (*pi->ret_prop.ret_buf_size), KM_SLEEP); 6455 ASSERT(pi->ret_prop.ret_buf_size != NULL); 6456 } else { 6457 pi->ret_prop.ret_buf_size = NULL; 6458 } 6459 } 6460 6461 return (0); 6462 } 6463 6464 6465 /* 6466 * Routine to free memory for the path information structures. 6467 * This is the memory which was allocated earlier. 6468 */ 6469 /* ARGSUSED */ 6470 static void 6471 vhci_ioc_free_pathinfo(sv_path_info_t *upibuf, sv_path_info_t *kpibuf, 6472 uint_t num_paths) 6473 { 6474 sv_path_info_t *pi; 6475 int index; 6476 6477 /* Free memory for per path properties */ 6478 for (index = 0, pi = kpibuf; index < num_paths; index++, pi++) { 6479 if (pi->ret_prop.ret_buf_size != NULL) { 6480 kmem_free(pi->ret_prop.ret_buf_size, 6481 sizeof (*pi->ret_prop.ret_buf_size)); 6482 } 6483 6484 if (pi->ret_prop.buf != NULL) { 6485 kmem_free(pi->ret_prop.buf, pi->ret_prop.buf_size); 6486 } 6487 } 6488 6489 /* Free memory for path info structures */ 6490 kmem_free(upibuf, sizeof (sv_path_info_t) * num_paths); 6491 kmem_free(kpibuf, sizeof (sv_path_info_t) * num_paths); 6492 } 6493 6494 6495 /* 6496 * Routine to copy path information and path properties to userland. 6497 */ 6498 /* ARGSUSED */ 6499 static int 6500 vhci_ioc_send_pathinfo(sv_path_info_t *upibuf, sv_path_info_t *kpibuf, 6501 uint_t num_paths, sv_iocdata_t *pioc, int mode, caddr_t s) 6502 { 6503 int retval = 0, index; 6504 sv_path_info_t *upi_ptr; 6505 sv_path_info32_t *upi32_ptr; 6506 6507 #ifdef _MULTI_DATAMODEL 6508 switch (ddi_model_convert_from(mode & FMODELS)) { 6509 case DDI_MODEL_ILP32: 6510 goto copy_32bit; 6511 6512 case DDI_MODEL_NONE: 6513 goto copy_normal; 6514 } 6515 #else /* _MULTI_DATAMODEL */ 6516 6517 goto copy_normal; 6518 6519 #endif /* _MULTI_DATAMODEL */ 6520 6521 copy_normal: 6522 6523 /* 6524 * Copy path information and path properties to user land. 6525 * Pointer fields inside the path property structure were 6526 * saved in the 'upibuf' structure earlier. 6527 */ 6528 upi_ptr = pioc->ret_buf; 6529 for (index = 0; index < num_paths; index++) { 6530 if (ddi_copyout(kpibuf[index].device.ret_ct, 6531 upi_ptr[index].device.ret_ct, MAXPATHLEN, mode)) { 6532 retval = EFAULT; 6533 break; 6534 } 6535 6536 if (ddi_copyout(kpibuf[index].ret_addr, 6537 upi_ptr[index].ret_addr, MAXNAMELEN, mode)) { 6538 retval = EFAULT; 6539 break; 6540 } 6541 6542 if (ddi_copyout(&kpibuf[index].ret_state, 6543 &upi_ptr[index].ret_state, sizeof (kpibuf[index].ret_state), 6544 mode)) { 6545 retval = EFAULT; 6546 break; 6547 } 6548 6549 if (ddi_copyout(&kpibuf[index].ret_ext_state, 6550 &upi_ptr[index].ret_ext_state, 6551 sizeof (kpibuf[index].ret_ext_state), mode)) { 6552 retval = EFAULT; 6553 break; 6554 } 6555 6556 if ((kpibuf[index].ret_prop.ret_buf_size != NULL) && 6557 ddi_copyout(kpibuf[index].ret_prop.ret_buf_size, 6558 upibuf[index].ret_prop.ret_buf_size, 6559 sizeof (*upibuf[index].ret_prop.ret_buf_size), mode)) { 6560 retval = EFAULT; 6561 break; 6562 } 6563 6564 if ((kpibuf[index].ret_prop.buf != NULL) && 6565 ddi_copyout(kpibuf[index].ret_prop.buf, 6566 upibuf[index].ret_prop.buf, 6567 upibuf[index].ret_prop.buf_size, mode)) { 6568 retval = EFAULT; 6569 break; 6570 } 6571 } 6572 6573 #ifdef DEBUG 6574 if (retval) { 6575 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: ioctl <%s> " 6576 "normal: path_info copyout failed", s)); 6577 } 6578 #endif 6579 6580 return (retval); 6581 6582 copy_32bit: 6583 /* 6584 * Copy path information and path properties to user land. 6585 * Pointer fields inside the path property structure were 6586 * saved in the 'upibuf' structure earlier. 6587 */ 6588 upi32_ptr = (sv_path_info32_t *)pioc->ret_buf; 6589 for (index = 0; index < num_paths; index++) { 6590 if (ddi_copyout(kpibuf[index].device.ret_ct, 6591 upi32_ptr[index].device.ret_ct, MAXPATHLEN, mode)) { 6592 retval = EFAULT; 6593 break; 6594 } 6595 6596 if (ddi_copyout(kpibuf[index].ret_addr, 6597 upi32_ptr[index].ret_addr, MAXNAMELEN, mode)) { 6598 retval = EFAULT; 6599 break; 6600 } 6601 6602 if (ddi_copyout(&kpibuf[index].ret_state, 6603 &upi32_ptr[index].ret_state, 6604 sizeof (kpibuf[index].ret_state), mode)) { 6605 retval = EFAULT; 6606 break; 6607 } 6608 6609 if (ddi_copyout(&kpibuf[index].ret_ext_state, 6610 &upi32_ptr[index].ret_ext_state, 6611 sizeof (kpibuf[index].ret_ext_state), mode)) { 6612 retval = EFAULT; 6613 break; 6614 } 6615 if ((kpibuf[index].ret_prop.ret_buf_size != NULL) && 6616 ddi_copyout(kpibuf[index].ret_prop.ret_buf_size, 6617 upibuf[index].ret_prop.ret_buf_size, 6618 sizeof (*upibuf[index].ret_prop.ret_buf_size), mode)) { 6619 retval = EFAULT; 6620 break; 6621 } 6622 6623 if ((kpibuf[index].ret_prop.buf != NULL) && 6624 ddi_copyout(kpibuf[index].ret_prop.buf, 6625 upibuf[index].ret_prop.buf, 6626 upibuf[index].ret_prop.buf_size, mode)) { 6627 retval = EFAULT; 6628 break; 6629 } 6630 } 6631 6632 #ifdef DEBUG 6633 if (retval) { 6634 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: ioctl <%s> " 6635 "normal: path_info copyout failed", s)); 6636 } 6637 #endif 6638 6639 return (retval); 6640 } 6641 6642 6643 /* 6644 * vhci_failover() 6645 * This routine expects VHCI_HOLD_LUN before being invoked. It can be invoked 6646 * as MDI_FAILOVER_ASYNC or MDI_FAILOVER_SYNC. For Asynchronous failovers 6647 * this routine shall VHCI_RELEASE_LUN on exiting. For synchronous failovers 6648 * it is the callers responsibility to release lun. 6649 */ 6650 6651 /* ARGSUSED */ 6652 static int 6653 vhci_failover(dev_info_t *vdip, dev_info_t *cdip, int flags) 6654 { 6655 char *guid; 6656 scsi_vhci_lun_t *vlun = NULL; 6657 struct scsi_vhci *vhci; 6658 mdi_pathinfo_t *pip, *npip; 6659 char *s_pclass, *pclass1, *pclass2, *pclass; 6660 char active_pclass_copy[255], *active_pclass_ptr; 6661 char *ptr1, *ptr2; 6662 mdi_pathinfo_state_t pi_state; 6663 uint32_t pi_ext_state; 6664 scsi_vhci_priv_t *svp; 6665 struct scsi_device *sd; 6666 struct scsi_failover_ops *sfo; 6667 int sps; /* mdi_select_path() status */ 6668 int activation_done = 0; 6669 int rval, retval = MDI_FAILURE; 6670 int reserve_pending, check_condition, UA_condition; 6671 struct scsi_pkt *pkt; 6672 struct buf *bp; 6673 6674 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 6675 sd = ddi_get_driver_private(cdip); 6676 vlun = ADDR2VLUN(&sd->sd_address); 6677 ASSERT(vlun != 0); 6678 ASSERT(VHCI_LUN_IS_HELD(vlun)); 6679 guid = vlun->svl_lun_wwn; 6680 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(1): guid %s\n", guid)); 6681 vhci_log(CE_NOTE, vdip, "!Initiating failover for device %s " 6682 "(GUID %s)", ddi_node_name(cdip), guid); 6683 6684 /* 6685 * Lets maintain a local copy of the vlun->svl_active_pclass 6686 * for the rest of the processing. Accessing the field 6687 * directly in the loop below causes loop logic to break 6688 * especially when the field gets updated by other threads 6689 * update path status etc and causes 'paths are not currently 6690 * available' condition to be declared prematurely. 6691 */ 6692 mutex_enter(&vlun->svl_mutex); 6693 if (vlun->svl_active_pclass != NULL) { 6694 (void) strlcpy(active_pclass_copy, vlun->svl_active_pclass, 6695 sizeof (active_pclass_copy)); 6696 active_pclass_ptr = &active_pclass_copy[0]; 6697 mutex_exit(&vlun->svl_mutex); 6698 if (vhci_quiesce_paths(vdip, cdip, vlun, guid, 6699 active_pclass_ptr) != 0) { 6700 retval = MDI_FAILURE; 6701 } 6702 } else { 6703 /* 6704 * can happen only when the available path to device 6705 * discovered is a STANDBY path. 6706 */ 6707 mutex_exit(&vlun->svl_mutex); 6708 active_pclass_copy[0] = '\0'; 6709 active_pclass_ptr = NULL; 6710 } 6711 6712 sfo = vlun->svl_fops; 6713 ASSERT(sfo != NULL); 6714 pclass1 = s_pclass = active_pclass_ptr; 6715 VHCI_DEBUG(1, (CE_NOTE, NULL, "!(%s)failing over from %s\n", guid, 6716 (s_pclass == NULL ? "<none>" : s_pclass))); 6717 6718 next_pathclass: 6719 6720 rval = (*sfo->sfo_pathclass_next)(pclass1, &pclass2, 6721 vlun->svl_fops_ctpriv); 6722 if (rval == ENOENT) { 6723 if (s_pclass == NULL) { 6724 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(4)(%s): " 6725 "failed, no more pathclasses\n", guid)); 6726 goto done; 6727 } else { 6728 (*sfo->sfo_pathclass_next)(NULL, &pclass2, 6729 vlun->svl_fops_ctpriv); 6730 } 6731 } else if (rval == EINVAL) { 6732 vhci_log(CE_NOTE, vdip, "!Failover operation failed for " 6733 "device %s (GUID %s): Invalid path-class %s", 6734 ddi_node_name(cdip), guid, 6735 ((pclass1 == NULL) ? "<none>" : pclass1)); 6736 goto done; 6737 } 6738 if ((s_pclass != NULL) && (strcmp(pclass2, s_pclass) == 0)) { 6739 /* 6740 * paths are not currently available 6741 */ 6742 vhci_log(CE_NOTE, vdip, "!Failover path currently unavailable" 6743 " for device %s (GUID %s)", 6744 ddi_node_name(cdip), guid); 6745 goto done; 6746 } 6747 pip = npip = NULL; 6748 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(5.2)(%s): considering " 6749 "%s as failover destination\n", guid, pclass2)); 6750 sps = mdi_select_path(cdip, NULL, MDI_SELECT_STANDBY_PATH, NULL, &npip); 6751 if ((npip == NULL) || (sps != MDI_SUCCESS)) { 6752 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(%s): no " 6753 "STANDBY paths found (status:%x)!\n", guid, sps)); 6754 pclass1 = pclass2; 6755 goto next_pathclass; 6756 } 6757 do { 6758 pclass = NULL; 6759 if ((mdi_prop_lookup_string(npip, "path-class", 6760 &pclass) != MDI_SUCCESS) || (strcmp(pclass2, 6761 pclass) != 0)) { 6762 VHCI_DEBUG(1, (CE_NOTE, NULL, 6763 "!vhci_failover(5.5)(%s): skipping path " 6764 "%p(%s)...\n", guid, (void *)npip, pclass)); 6765 pip = npip; 6766 sps = mdi_select_path(cdip, NULL, 6767 MDI_SELECT_STANDBY_PATH, pip, &npip); 6768 mdi_rele_path(pip); 6769 (void) mdi_prop_free(pclass); 6770 continue; 6771 } 6772 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(npip); 6773 6774 /* 6775 * Issue READ at non-zer block on this STANDBY path. 6776 * Purple returns 6777 * 1. RESERVATION_CONFLICT if reservation is pending 6778 * 2. POR check condition if it reset happened. 6779 * 2. failover Check Conditions if one is already in progress. 6780 */ 6781 reserve_pending = 0; 6782 check_condition = 0; 6783 UA_condition = 0; 6784 6785 bp = scsi_alloc_consistent_buf(&svp->svp_psd->sd_address, 6786 (struct buf *)NULL, DEV_BSIZE, B_READ, NULL, NULL); 6787 if (!bp) { 6788 VHCI_DEBUG(1, (CE_NOTE, NULL, 6789 "vhci_failover !No resources (buf)\n")); 6790 mdi_rele_path(npip); 6791 goto done; 6792 } 6793 pkt = scsi_init_pkt(&svp->svp_psd->sd_address, NULL, bp, 6794 CDB_GROUP1, sizeof (struct scsi_arq_status), 0, 6795 PKT_CONSISTENT, NULL, NULL); 6796 if (pkt) { 6797 (void) scsi_setup_cdb((union scsi_cdb *)(uintptr_t) 6798 pkt->pkt_cdbp, SCMD_READ, 1, 1, 0); 6799 pkt->pkt_flags = FLAG_NOINTR; 6800 check_path_again: 6801 pkt->pkt_path_instance = mdi_pi_get_path_instance(npip); 6802 pkt->pkt_time = 3*30; 6803 6804 if (scsi_transport(pkt) == TRAN_ACCEPT) { 6805 switch (pkt->pkt_reason) { 6806 case CMD_CMPLT: 6807 switch (SCBP_C(pkt)) { 6808 case STATUS_GOOD: 6809 /* Already failed over */ 6810 activation_done = 1; 6811 break; 6812 case STATUS_RESERVATION_CONFLICT: 6813 reserve_pending = 1; 6814 break; 6815 case STATUS_CHECK: 6816 check_condition = 1; 6817 break; 6818 } 6819 } 6820 } 6821 if (check_condition && 6822 (pkt->pkt_state & STATE_ARQ_DONE)) { 6823 struct scsi_extended_sense *sns = 6824 &(((struct scsi_arq_status *)(uintptr_t) 6825 (pkt->pkt_scbp))->sts_sensedata); 6826 if (sns->es_key == KEY_UNIT_ATTENTION && 6827 sns->es_add_code == 0x29) { 6828 /* Already failed over */ 6829 VHCI_DEBUG(1, (CE_NOTE, NULL, 6830 "!vhci_failover(7)(%s): " 6831 "path 0x%p POR UA condition\n", 6832 guid, (void *)npip)); 6833 if (UA_condition == 0) { 6834 UA_condition = 1; 6835 goto check_path_again; 6836 } 6837 } else { 6838 activation_done = 0; 6839 VHCI_DEBUG(1, (CE_NOTE, NULL, 6840 "!vhci_failover(%s): path 0x%p " 6841 "unhandled chkcond %x %x %x\n", 6842 guid, (void *)npip, sns->es_key, 6843 sns->es_add_code, 6844 sns->es_qual_code)); 6845 } 6846 } 6847 scsi_destroy_pkt(pkt); 6848 } 6849 scsi_free_consistent_buf(bp); 6850 6851 if (activation_done) { 6852 mdi_rele_path(npip); 6853 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(7)(%s): " 6854 "path 0x%p already failedover\n", guid, 6855 (void *)npip)); 6856 break; 6857 } 6858 if (reserve_pending && (vlun->svl_xlf_capable == 0)) { 6859 (void) vhci_recovery_reset(vlun, 6860 &svp->svp_psd->sd_address, 6861 FALSE, VHCI_DEPTH_ALL); 6862 } 6863 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(6)(%s): " 6864 "activating path 0x%p(psd:%p)\n", guid, (void *)npip, 6865 (void *)svp->svp_psd)); 6866 if ((*sfo->sfo_path_activate)(svp->svp_psd, pclass2, 6867 vlun->svl_fops_ctpriv) == 0) { 6868 activation_done = 1; 6869 mdi_rele_path(npip); 6870 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(7)(%s): " 6871 "path 0x%p successfully activated\n", guid, 6872 (void *)npip)); 6873 break; 6874 } 6875 pip = npip; 6876 sps = mdi_select_path(cdip, NULL, MDI_SELECT_STANDBY_PATH, 6877 pip, &npip); 6878 mdi_rele_path(pip); 6879 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 6880 if (activation_done == 0) { 6881 pclass1 = pclass2; 6882 goto next_pathclass; 6883 } 6884 6885 /* 6886 * if we are here, we have succeeded in activating path npip of 6887 * pathclass pclass2; let us validate all paths of pclass2 by 6888 * "ping"-ing each one and mark the good ones ONLINE 6889 * Also, set the state of the paths belonging to the previously 6890 * active pathclass to STANDBY 6891 */ 6892 pip = npip = NULL; 6893 sps = mdi_select_path(cdip, NULL, (MDI_SELECT_ONLINE_PATH | 6894 MDI_SELECT_STANDBY_PATH | MDI_SELECT_USER_DISABLE_PATH), 6895 NULL, &npip); 6896 if (npip == NULL || sps != MDI_SUCCESS) { 6897 VHCI_DEBUG(1, (CE_NOTE, NULL, "!Failover operation failed for " 6898 "device %s (GUID %s): paths may be busy\n", 6899 ddi_node_name(cdip), guid)); 6900 goto done; 6901 } 6902 do { 6903 (void) mdi_pi_get_state2(npip, &pi_state, &pi_ext_state); 6904 if (mdi_prop_lookup_string(npip, "path-class", &pclass) 6905 != MDI_SUCCESS) { 6906 pip = npip; 6907 sps = mdi_select_path(cdip, NULL, 6908 (MDI_SELECT_ONLINE_PATH | 6909 MDI_SELECT_STANDBY_PATH | 6910 MDI_SELECT_USER_DISABLE_PATH), 6911 pip, &npip); 6912 mdi_rele_path(pip); 6913 continue; 6914 } 6915 if (strcmp(pclass, pclass2) == 0) { 6916 if (pi_state == MDI_PATHINFO_STATE_STANDBY) { 6917 svp = (scsi_vhci_priv_t *) 6918 mdi_pi_get_vhci_private(npip); 6919 VHCI_DEBUG(1, (CE_NOTE, NULL, 6920 "!vhci_failover(8)(%s): " 6921 "pinging path 0x%p\n", 6922 guid, (void *)npip)); 6923 if ((*sfo->sfo_path_ping)(svp->svp_psd, 6924 vlun->svl_fops_ctpriv) == 1) { 6925 mdi_pi_set_state(npip, 6926 MDI_PATHINFO_STATE_ONLINE); 6927 VHCI_DEBUG(1, (CE_NOTE, NULL, 6928 "!vhci_failover(9)(%s): " 6929 "path 0x%p ping successful, " 6930 "marked online\n", guid, 6931 (void *)npip)); 6932 MDI_PI_ERRSTAT(npip, MDI_PI_FAILTO); 6933 } 6934 } 6935 } else if ((s_pclass != NULL) && (strcmp(pclass, s_pclass) 6936 == 0)) { 6937 if (pi_state == MDI_PATHINFO_STATE_ONLINE) { 6938 mdi_pi_set_state(npip, 6939 MDI_PATHINFO_STATE_STANDBY); 6940 VHCI_DEBUG(1, (CE_NOTE, NULL, 6941 "!vhci_failover(10)(%s): path 0x%p marked " 6942 "STANDBY\n", guid, (void *)npip)); 6943 MDI_PI_ERRSTAT(npip, MDI_PI_FAILFROM); 6944 } 6945 } 6946 (void) mdi_prop_free(pclass); 6947 pip = npip; 6948 sps = mdi_select_path(cdip, NULL, (MDI_SELECT_ONLINE_PATH | 6949 MDI_SELECT_STANDBY_PATH|MDI_SELECT_USER_DISABLE_PATH), 6950 pip, &npip); 6951 mdi_rele_path(pip); 6952 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 6953 6954 /* 6955 * Update the AccessState of related MP-API TPGs 6956 */ 6957 (void) vhci_mpapi_update_tpg_acc_state_for_lu(vhci, vlun); 6958 6959 vhci_log(CE_NOTE, vdip, "!Failover operation completed successfully " 6960 "for device %s (GUID %s): failed over from %s to %s", 6961 ddi_node_name(cdip), guid, ((s_pclass == NULL) ? "<none>" : 6962 s_pclass), pclass2); 6963 ptr1 = kmem_alloc(strlen(pclass2)+1, KM_SLEEP); 6964 (void) strlcpy(ptr1, pclass2, (strlen(pclass2)+1)); 6965 mutex_enter(&vlun->svl_mutex); 6966 ptr2 = vlun->svl_active_pclass; 6967 vlun->svl_active_pclass = ptr1; 6968 mutex_exit(&vlun->svl_mutex); 6969 if (ptr2) { 6970 kmem_free(ptr2, strlen(ptr2)+1); 6971 } 6972 mutex_enter(&vhci->vhci_mutex); 6973 scsi_hba_reset_notify_callback(&vhci->vhci_mutex, 6974 &vhci->vhci_reset_notify_listf); 6975 /* All reservations are cleared upon these resets. */ 6976 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 6977 mutex_exit(&vhci->vhci_mutex); 6978 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(11): DONE! Active " 6979 "pathclass for %s is now %s\n", guid, pclass2)); 6980 retval = MDI_SUCCESS; 6981 6982 done: 6983 if (flags == MDI_FAILOVER_ASYNC) { 6984 VHCI_RELEASE_LUN(vlun); 6985 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_failover(12): DONE! " 6986 "releasing lun, as failover was ASYNC\n")); 6987 } else { 6988 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_failover(12): DONE! " 6989 "NOT releasing lun, as failover was SYNC\n")); 6990 } 6991 return (retval); 6992 } 6993 6994 /* 6995 * vhci_client_attached is called after the successful attach of a 6996 * client devinfo node. 6997 */ 6998 static void 6999 vhci_client_attached(dev_info_t *cdip) 7000 { 7001 mdi_pathinfo_t *pip; 7002 int circular; 7003 7004 /* 7005 * At this point the client has attached and it's instance number is 7006 * valid, so we can set up kstats. We need to do this here because it 7007 * is possible for paths to go online prior to client attach, in which 7008 * case the call to vhci_kstat_create_pathinfo in vhci_pathinfo_online 7009 * was a noop. 7010 */ 7011 ndi_devi_enter(cdip, &circular); 7012 for (pip = mdi_get_next_phci_path(cdip, NULL); pip; 7013 pip = mdi_get_next_phci_path(cdip, pip)) 7014 vhci_kstat_create_pathinfo(pip); 7015 ndi_devi_exit(cdip, circular); 7016 } 7017 7018 /* 7019 * quiesce all of the online paths 7020 */ 7021 static int 7022 vhci_quiesce_paths(dev_info_t *vdip, dev_info_t *cdip, scsi_vhci_lun_t *vlun, 7023 char *guid, char *active_pclass_ptr) 7024 { 7025 scsi_vhci_priv_t *svp; 7026 char *s_pclass = NULL; 7027 mdi_pathinfo_t *npip, *pip; 7028 int sps; 7029 7030 /* quiesce currently active paths */ 7031 s_pclass = NULL; 7032 pip = npip = NULL; 7033 sps = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH, NULL, &npip); 7034 if ((npip == NULL) || (sps != MDI_SUCCESS)) { 7035 return (1); 7036 } 7037 do { 7038 if (mdi_prop_lookup_string(npip, "path-class", 7039 &s_pclass) != MDI_SUCCESS) { 7040 mdi_rele_path(npip); 7041 vhci_log(CE_NOTE, vdip, "!Failover operation failed " 7042 "for device %s (GUID %s) due to an internal " 7043 "error", ddi_node_name(cdip), guid); 7044 return (1); 7045 } 7046 if (strcmp(s_pclass, active_pclass_ptr) == 0) { 7047 /* 7048 * quiesce path. Free s_pclass since 7049 * we don't need it anymore 7050 */ 7051 VHCI_DEBUG(1, (CE_NOTE, NULL, 7052 "!vhci_failover(2)(%s): failing over " 7053 "from %s; quiescing path %p\n", 7054 guid, s_pclass, (void *)npip)); 7055 (void) mdi_prop_free(s_pclass); 7056 svp = (scsi_vhci_priv_t *) 7057 mdi_pi_get_vhci_private(npip); 7058 if (svp == NULL) { 7059 VHCI_DEBUG(1, (CE_NOTE, NULL, 7060 "!vhci_failover(2.5)(%s): no " 7061 "client priv! %p offlined?\n", 7062 guid, (void *)npip)); 7063 pip = npip; 7064 sps = mdi_select_path(cdip, NULL, 7065 MDI_SELECT_ONLINE_PATH, pip, &npip); 7066 mdi_rele_path(pip); 7067 continue; 7068 } 7069 if (scsi_abort(&svp->svp_psd->sd_address, NULL) 7070 == 0) { 7071 (void) vhci_recovery_reset(vlun, 7072 &svp->svp_psd->sd_address, FALSE, 7073 VHCI_DEPTH_TARGET); 7074 } 7075 mutex_enter(&svp->svp_mutex); 7076 if (svp->svp_cmds == 0) { 7077 VHCI_DEBUG(1, (CE_NOTE, NULL, 7078 "!vhci_failover(3)(%s):" 7079 "quiesced path %p\n", guid, (void *)npip)); 7080 } else { 7081 while (svp->svp_cmds != 0) { 7082 cv_wait(&svp->svp_cv, &svp->svp_mutex); 7083 VHCI_DEBUG(1, (CE_NOTE, NULL, 7084 "!vhci_failover(3.cv)(%s):" 7085 "quiesced path %p\n", guid, 7086 (void *)npip)); 7087 } 7088 } 7089 mutex_exit(&svp->svp_mutex); 7090 } else { 7091 /* 7092 * make sure we freeup the memory 7093 */ 7094 (void) mdi_prop_free(s_pclass); 7095 } 7096 pip = npip; 7097 sps = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH, 7098 pip, &npip); 7099 mdi_rele_path(pip); 7100 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 7101 return (0); 7102 } 7103 7104 static struct scsi_vhci_lun * 7105 vhci_lun_lookup(dev_info_t *tgt_dip) 7106 { 7107 return ((struct scsi_vhci_lun *) 7108 mdi_client_get_vhci_private(tgt_dip)); 7109 } 7110 7111 static struct scsi_vhci_lun * 7112 vhci_lun_lookup_alloc(dev_info_t *tgt_dip, char *guid, int *didalloc) 7113 { 7114 struct scsi_vhci_lun *svl; 7115 7116 if (svl = vhci_lun_lookup(tgt_dip)) { 7117 return (svl); 7118 } 7119 7120 svl = kmem_zalloc(sizeof (*svl), KM_SLEEP); 7121 svl->svl_lun_wwn = kmem_zalloc(strlen(guid)+1, KM_SLEEP); 7122 (void) strcpy(svl->svl_lun_wwn, guid); 7123 mutex_init(&svl->svl_mutex, NULL, MUTEX_DRIVER, NULL); 7124 cv_init(&svl->svl_cv, NULL, CV_DRIVER, NULL); 7125 sema_init(&svl->svl_pgr_sema, 1, NULL, SEMA_DRIVER, NULL); 7126 svl->svl_waiting_for_activepath = 1; 7127 svl->svl_sector_size = 1; 7128 mdi_client_set_vhci_private(tgt_dip, svl); 7129 *didalloc = 1; 7130 VHCI_DEBUG(1, (CE_NOTE, NULL, 7131 "vhci_lun_lookup_alloc: guid %s vlun 0x%p\n", 7132 guid, (void *)svl)); 7133 return (svl); 7134 } 7135 7136 static void 7137 vhci_lun_free(dev_info_t *tgt_dip) 7138 { 7139 struct scsi_vhci_lun *dvlp; 7140 char *guid; 7141 struct scsi_device *sd; 7142 7143 /* 7144 * The scsi_device was set to driver private during child node 7145 * initialization in the scsi_hba_bus_ctl(). 7146 */ 7147 sd = (struct scsi_device *)ddi_get_driver_private(tgt_dip); 7148 7149 dvlp = (struct scsi_vhci_lun *) 7150 mdi_client_get_vhci_private(tgt_dip); 7151 ASSERT(dvlp != NULL); 7152 7153 mdi_client_set_vhci_private(tgt_dip, NULL); 7154 7155 guid = dvlp->svl_lun_wwn; 7156 ASSERT(guid != NULL); 7157 VHCI_DEBUG(4, (CE_NOTE, NULL, "!vhci_lun_free: %s\n", guid)); 7158 7159 mutex_enter(&dvlp->svl_mutex); 7160 if (dvlp->svl_active_pclass != NULL) { 7161 kmem_free(dvlp->svl_active_pclass, 7162 strlen(dvlp->svl_active_pclass)+1); 7163 } 7164 dvlp->svl_active_pclass = NULL; 7165 mutex_exit(&dvlp->svl_mutex); 7166 7167 if (dvlp->svl_lun_wwn != NULL) { 7168 kmem_free(dvlp->svl_lun_wwn, strlen(dvlp->svl_lun_wwn)+1); 7169 } 7170 dvlp->svl_lun_wwn = NULL; 7171 7172 if (dvlp->svl_fops_name) { 7173 kmem_free(dvlp->svl_fops_name, strlen(dvlp->svl_fops_name)+1); 7174 } 7175 dvlp->svl_fops_name = NULL; 7176 7177 if (dvlp->svl_fops_ctpriv != NULL) { 7178 dvlp->svl_fops->sfo_device_unprobe(sd, dvlp->svl_fops_ctpriv); 7179 } 7180 7181 if (dvlp->svl_flags & VLUN_TASK_D_ALIVE_FLG) 7182 taskq_destroy(dvlp->svl_taskq); 7183 7184 mutex_destroy(&dvlp->svl_mutex); 7185 cv_destroy(&dvlp->svl_cv); 7186 sema_destroy(&dvlp->svl_pgr_sema); 7187 kmem_free(dvlp, sizeof (*dvlp)); 7188 /* 7189 * vhci_lun_free may be called before the tgt_dip 7190 * initialization so check if the sd is NULL. 7191 */ 7192 if (sd != NULL) 7193 sd->sd_address.a_hba_tran->tran_tgt_private = NULL; 7194 } 7195 7196 7197 int 7198 vhci_do_scsi_cmd(struct scsi_pkt *pkt) 7199 { 7200 int err = 0; 7201 int retry_cnt = 0; 7202 struct scsi_extended_sense *sns; 7203 7204 retry: 7205 err = scsi_poll(pkt); 7206 if (err) { 7207 if (pkt->pkt_cdbp[0] == SCMD_RELEASE) { 7208 if (SCBP_C(pkt) == STATUS_RESERVATION_CONFLICT) { 7209 VHCI_DEBUG(1, (CE_NOTE, NULL, 7210 "!v_s_do_s_c: RELEASE conflict\n")); 7211 return (0); 7212 } 7213 } 7214 if (retry_cnt++ < 3) { 7215 VHCI_DEBUG(1, (CE_WARN, NULL, 7216 "!v_s_do_s_c:retry packet 0x%p " 7217 "status 0x%x reason %s", 7218 (void *)pkt, SCBP_C(pkt), 7219 scsi_rname(pkt->pkt_reason))); 7220 if ((pkt->pkt_reason == CMD_CMPLT) && 7221 (SCBP_C(pkt) == STATUS_CHECK) && 7222 (pkt->pkt_state & STATE_ARQ_DONE)) { 7223 sns = &(((struct scsi_arq_status *)(uintptr_t) 7224 (pkt->pkt_scbp))->sts_sensedata); 7225 VHCI_DEBUG(1, (CE_WARN, NULL, 7226 "!v_s_do_s_c:retry " 7227 "packet 0x%p sense data %s", (void *)pkt, 7228 scsi_sname(sns->es_key))); 7229 } 7230 goto retry; 7231 } 7232 VHCI_DEBUG(1, (CE_WARN, NULL, 7233 "!v_s_do_s_c: failed transport 0x%p 0x%x", 7234 (void *)pkt, SCBP_C(pkt))); 7235 return (0); 7236 } 7237 7238 switch (pkt->pkt_reason) { 7239 case CMD_TIMEOUT: 7240 VHCI_DEBUG(1, (CE_WARN, NULL, "!pkt timed " 7241 "out (pkt 0x%p)", (void *)pkt)); 7242 return (0); 7243 case CMD_CMPLT: 7244 switch (SCBP_C(pkt)) { 7245 case STATUS_GOOD: 7246 break; 7247 case STATUS_CHECK: 7248 if (pkt->pkt_state & STATE_ARQ_DONE) { 7249 sns = &((( 7250 struct scsi_arq_status *) 7251 (uintptr_t) 7252 (pkt->pkt_scbp))-> 7253 sts_sensedata); 7254 if ((sns->es_key == 7255 KEY_UNIT_ATTENTION) || 7256 (sns->es_key == 7257 KEY_NOT_READY)) { 7258 /* 7259 * clear unit attn. 7260 */ 7261 7262 VHCI_DEBUG(1, 7263 (CE_WARN, NULL, 7264 "!v_s_do_s_c: " 7265 "retry " 7266 "packet 0x%p sense " 7267 "data %s", 7268 (void *)pkt, 7269 scsi_sname 7270 (sns->es_key))); 7271 goto retry; 7272 } 7273 VHCI_DEBUG(4, (CE_WARN, NULL, 7274 "!ARQ while " 7275 "transporting " 7276 "(pkt 0x%p)", 7277 (void *)pkt)); 7278 return (0); 7279 } 7280 return (0); 7281 default: 7282 VHCI_DEBUG(1, (CE_WARN, NULL, 7283 "!Bad status returned " 7284 "(pkt 0x%p, status %x)", 7285 (void *)pkt, SCBP_C(pkt))); 7286 return (0); 7287 } 7288 break; 7289 case CMD_INCOMPLETE: 7290 case CMD_RESET: 7291 case CMD_ABORTED: 7292 case CMD_TRAN_ERR: 7293 if (retry_cnt++ < 1) { 7294 VHCI_DEBUG(1, (CE_WARN, NULL, 7295 "!v_s_do_s_c: retry packet 0x%p %s", 7296 (void *)pkt, scsi_rname(pkt->pkt_reason))); 7297 goto retry; 7298 } 7299 /* FALLTHROUGH */ 7300 default: 7301 VHCI_DEBUG(1, (CE_WARN, NULL, "!pkt did not " 7302 "complete successfully (pkt 0x%p," 7303 "reason %x)", (void *)pkt, pkt->pkt_reason)); 7304 return (0); 7305 } 7306 return (1); 7307 } 7308 7309 static int 7310 vhci_quiesce_lun(struct scsi_vhci_lun *vlun) 7311 { 7312 mdi_pathinfo_t *pip, *spip; 7313 dev_info_t *cdip; 7314 struct scsi_vhci_priv *svp; 7315 mdi_pathinfo_state_t pstate; 7316 uint32_t p_ext_state; 7317 int circular; 7318 7319 cdip = vlun->svl_dip; 7320 pip = spip = NULL; 7321 ndi_devi_enter(cdip, &circular); 7322 pip = mdi_get_next_phci_path(cdip, NULL); 7323 while (pip != NULL) { 7324 (void) mdi_pi_get_state2(pip, &pstate, &p_ext_state); 7325 if (pstate != MDI_PATHINFO_STATE_ONLINE) { 7326 spip = pip; 7327 pip = mdi_get_next_phci_path(cdip, spip); 7328 continue; 7329 } 7330 mdi_hold_path(pip); 7331 ndi_devi_exit(cdip, circular); 7332 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 7333 mutex_enter(&svp->svp_mutex); 7334 while (svp->svp_cmds != 0) { 7335 if (cv_timedwait(&svp->svp_cv, &svp->svp_mutex, 7336 ddi_get_lbolt() + drv_usectohz 7337 (vhci_path_quiesce_timeout * 1000000)) == -1) { 7338 mutex_exit(&svp->svp_mutex); 7339 mdi_rele_path(pip); 7340 VHCI_DEBUG(1, (CE_WARN, NULL, 7341 "Quiesce of lun is not successful " 7342 "vlun: 0x%p.", (void *)vlun)); 7343 return (0); 7344 } 7345 } 7346 mutex_exit(&svp->svp_mutex); 7347 ndi_devi_enter(cdip, &circular); 7348 spip = pip; 7349 pip = mdi_get_next_phci_path(cdip, spip); 7350 mdi_rele_path(spip); 7351 } 7352 ndi_devi_exit(cdip, circular); 7353 return (1); 7354 } 7355 7356 static int 7357 vhci_pgr_validate_and_register(scsi_vhci_priv_t *svp) 7358 { 7359 scsi_vhci_lun_t *vlun; 7360 vhci_prout_t *prout; 7361 int rval, success; 7362 mdi_pathinfo_t *pip, *npip; 7363 scsi_vhci_priv_t *osvp; 7364 dev_info_t *cdip; 7365 uchar_t cdb_1; 7366 uchar_t temp_res_key[MHIOC_RESV_KEY_SIZE]; 7367 7368 7369 /* 7370 * see if there are any other paths available; if none, 7371 * then there is nothing to do. 7372 */ 7373 cdip = svp->svp_svl->svl_dip; 7374 rval = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH | 7375 MDI_SELECT_STANDBY_PATH, NULL, &pip); 7376 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 7377 VHCI_DEBUG(4, (CE_NOTE, NULL, 7378 "%s%d: vhci_pgr_validate_and_register: first path\n", 7379 ddi_driver_name(cdip), ddi_get_instance(cdip))); 7380 return (1); 7381 } 7382 7383 vlun = svp->svp_svl; 7384 prout = &vlun->svl_prout; 7385 ASSERT(vlun->svl_pgr_active != 0); 7386 7387 /* 7388 * When the path was busy/offlined, some other host might have 7389 * cleared this key. Validate key on some other path first. 7390 * If it fails, return failure. 7391 */ 7392 7393 npip = pip; 7394 pip = NULL; 7395 success = 0; 7396 7397 /* Save the res key */ 7398 bcopy((const void *)prout->res_key, 7399 (void *)temp_res_key, MHIOC_RESV_KEY_SIZE); 7400 7401 /* 7402 * Sometimes CDB from application can be a Register_And_Ignore. 7403 * Instead of validation, this cdb would result in force registration. 7404 * Convert it to normal cdb for validation. 7405 * After that be sure to restore the cdb. 7406 */ 7407 cdb_1 = vlun->svl_cdb[1]; 7408 vlun->svl_cdb[1] &= 0xe0; 7409 7410 do { 7411 osvp = (scsi_vhci_priv_t *) 7412 mdi_pi_get_vhci_private(npip); 7413 if (osvp == NULL) { 7414 VHCI_DEBUG(4, (CE_NOTE, NULL, 7415 "vhci_pgr_validate_and_register: no " 7416 "client priv! 0x%p offlined?\n", 7417 (void *)npip)); 7418 goto next_path_1; 7419 } 7420 7421 if (osvp == svp) { 7422 VHCI_DEBUG(4, (CE_NOTE, NULL, 7423 "vhci_pgr_validate_and_register: same svp 0x%p" 7424 " npip 0x%p vlun 0x%p\n", 7425 (void *)svp, (void *)npip, (void *)vlun)); 7426 goto next_path_1; 7427 } 7428 7429 VHCI_DEBUG(4, (CE_NOTE, NULL, 7430 "vhci_pgr_validate_and_register: First validate on" 7431 " osvp 0x%p being done. vlun 0x%p thread 0x%p Before bcopy" 7432 " cdb1 %x\n", (void *)osvp, (void *)vlun, 7433 (void *)curthread, vlun->svl_cdb[1])); 7434 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy:"); 7435 7436 bcopy((const void *)prout->service_key, 7437 (void *)prout->res_key, MHIOC_RESV_KEY_SIZE); 7438 7439 VHCI_DEBUG(4, (CE_WARN, NULL, "vlun 0x%p After bcopy", 7440 (void *)vlun)); 7441 vhci_print_prout_keys(vlun, "v_pgr_val_reg: after bcopy: "); 7442 7443 rval = vhci_do_prout(osvp); 7444 if (rval == 1) { 7445 VHCI_DEBUG(4, (CE_NOTE, NULL, 7446 "%s%d: vhci_pgr_validate_and_register: key" 7447 " validated thread 0x%p\n", ddi_driver_name(cdip), 7448 ddi_get_instance(cdip), (void *)curthread)); 7449 pip = npip; 7450 success = 1; 7451 break; 7452 } else { 7453 VHCI_DEBUG(4, (CE_NOTE, NULL, 7454 "vhci_pgr_validate_and_register: First validation" 7455 " on osvp 0x%p failed %x\n", (void *)osvp, rval)); 7456 vhci_print_prout_keys(vlun, "v_pgr_val_reg: failed:"); 7457 } 7458 7459 /* 7460 * Try other paths 7461 */ 7462 next_path_1: 7463 pip = npip; 7464 rval = mdi_select_path(cdip, NULL, 7465 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, 7466 pip, &npip); 7467 mdi_rele_path(pip); 7468 } while ((rval == MDI_SUCCESS) && (npip != NULL)); 7469 7470 7471 /* Be sure to restore original cdb */ 7472 vlun->svl_cdb[1] = cdb_1; 7473 7474 /* Restore the res_key */ 7475 bcopy((const void *)temp_res_key, 7476 (void *)prout->res_key, MHIOC_RESV_KEY_SIZE); 7477 7478 /* 7479 * If key could not be registered on any path for the first time, 7480 * return success as online should still continue. 7481 */ 7482 if (success == 0) { 7483 return (1); 7484 } 7485 7486 ASSERT(pip != NULL); 7487 7488 /* 7489 * Force register on new path 7490 */ 7491 cdb_1 = vlun->svl_cdb[1]; /* store the cdb */ 7492 7493 vlun->svl_cdb[1] &= 0xe0; 7494 vlun->svl_cdb[1] |= VHCI_PROUT_R_AND_IGNORE; 7495 7496 vhci_print_prout_keys(vlun, "v_pgr_val_reg: keys before bcopy: "); 7497 7498 bcopy((const void *)prout->active_service_key, 7499 (void *)prout->service_key, MHIOC_RESV_KEY_SIZE); 7500 bcopy((const void *)prout->active_res_key, 7501 (void *)prout->res_key, MHIOC_RESV_KEY_SIZE); 7502 7503 vhci_print_prout_keys(vlun, "v_pgr_val_reg:keys after bcopy: "); 7504 7505 rval = vhci_do_prout(svp); 7506 vlun->svl_cdb[1] = cdb_1; /* restore the cdb */ 7507 if (rval != 1) { 7508 VHCI_DEBUG(4, (CE_NOTE, NULL, 7509 "vhci_pgr_validate_and_register: register on new" 7510 " path 0x%p svp 0x%p failed %x\n", 7511 (void *)pip, (void *)svp, rval)); 7512 vhci_print_prout_keys(vlun, "v_pgr_val_reg: reg failed: "); 7513 mdi_rele_path(pip); 7514 return (0); 7515 } 7516 7517 if (bcmp(prout->service_key, zero_key, MHIOC_RESV_KEY_SIZE) == 0) { 7518 VHCI_DEBUG(4, (CE_NOTE, NULL, 7519 "vhci_pgr_validate_and_register: zero service key\n")); 7520 mdi_rele_path(pip); 7521 return (rval); 7522 } 7523 7524 /* 7525 * While the key was force registered, some other host might have 7526 * cleared the key. Re-validate key on another pre-existing path 7527 * before declaring success. 7528 */ 7529 npip = pip; 7530 pip = NULL; 7531 7532 /* 7533 * Sometimes CDB from application can be Register and Ignore. 7534 * Instead of validation, it would result in force registration. 7535 * Convert it to normal cdb for validation. 7536 * After that be sure to restore the cdb. 7537 */ 7538 cdb_1 = vlun->svl_cdb[1]; 7539 vlun->svl_cdb[1] &= 0xe0; 7540 success = 0; 7541 7542 do { 7543 osvp = (scsi_vhci_priv_t *) 7544 mdi_pi_get_vhci_private(npip); 7545 if (osvp == NULL) { 7546 VHCI_DEBUG(4, (CE_NOTE, NULL, 7547 "vhci_pgr_validate_and_register: no " 7548 "client priv! 0x%p offlined?\n", 7549 (void *)npip)); 7550 goto next_path_2; 7551 } 7552 7553 if (osvp == svp) { 7554 VHCI_DEBUG(4, (CE_NOTE, NULL, 7555 "vhci_pgr_validate_and_register: same osvp 0x%p" 7556 " npip 0x%p vlun 0x%p\n", 7557 (void *)svp, (void *)npip, (void *)vlun)); 7558 goto next_path_2; 7559 } 7560 7561 VHCI_DEBUG(4, (CE_NOTE, NULL, 7562 "vhci_pgr_validate_and_register: Re-validation on" 7563 " osvp 0x%p being done. vlun 0x%p Before bcopy cdb1 %x\n", 7564 (void *)osvp, (void *)vlun, vlun->svl_cdb[1])); 7565 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy: "); 7566 7567 bcopy((const void *)prout->service_key, 7568 (void *)prout->res_key, MHIOC_RESV_KEY_SIZE); 7569 7570 vhci_print_prout_keys(vlun, "v_pgr_val_reg: after bcopy: "); 7571 7572 rval = vhci_do_prout(osvp); 7573 if (rval == 1) { 7574 VHCI_DEBUG(4, (CE_NOTE, NULL, 7575 "%s%d: vhci_pgr_validate_and_register: key" 7576 " validated thread 0x%p\n", ddi_driver_name(cdip), 7577 ddi_get_instance(cdip), (void *)curthread)); 7578 pip = npip; 7579 success = 1; 7580 break; 7581 } else { 7582 VHCI_DEBUG(4, (CE_NOTE, NULL, 7583 "vhci_pgr_validate_and_register: Re-validation on" 7584 " osvp 0x%p failed %x\n", (void *)osvp, rval)); 7585 vhci_print_prout_keys(vlun, 7586 "v_pgr_val_reg: reval failed: "); 7587 } 7588 7589 /* 7590 * Try other paths 7591 */ 7592 next_path_2: 7593 pip = npip; 7594 rval = mdi_select_path(cdip, NULL, 7595 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, 7596 pip, &npip); 7597 mdi_rele_path(pip); 7598 } while ((rval == MDI_SUCCESS) && (npip != NULL)); 7599 7600 /* Be sure to restore original cdb */ 7601 vlun->svl_cdb[1] = cdb_1; 7602 7603 if (success == 1) { 7604 /* Successfully validated registration */ 7605 mdi_rele_path(pip); 7606 return (1); 7607 } 7608 7609 VHCI_DEBUG(4, (CE_WARN, NULL, "key validation failed")); 7610 7611 /* 7612 * key invalid, back out by registering key value of 0 7613 */ 7614 VHCI_DEBUG(4, (CE_NOTE, NULL, 7615 "vhci_pgr_validate_and_register: backout on" 7616 " svp 0x%p being done\n", (void *)svp)); 7617 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy: "); 7618 7619 bcopy((const void *)prout->service_key, (void *)prout->res_key, 7620 MHIOC_RESV_KEY_SIZE); 7621 bzero((void *)prout->service_key, MHIOC_RESV_KEY_SIZE); 7622 7623 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy: "); 7624 7625 /* 7626 * Get a new path 7627 */ 7628 rval = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH | 7629 MDI_SELECT_STANDBY_PATH, NULL, &pip); 7630 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 7631 VHCI_DEBUG(4, (CE_NOTE, NULL, 7632 "%s%d: vhci_pgr_validate_and_register: no valid pip\n", 7633 ddi_driver_name(cdip), ddi_get_instance(cdip))); 7634 return (0); 7635 } 7636 7637 if ((rval = vhci_do_prout(svp)) != 1) { 7638 VHCI_DEBUG(4, (CE_NOTE, NULL, 7639 "vhci_pgr_validate_and_register: backout on" 7640 " svp 0x%p failed\n", (void *)svp)); 7641 vhci_print_prout_keys(vlun, "backout failed"); 7642 7643 VHCI_DEBUG(4, (CE_WARN, NULL, 7644 "%s%d: vhci_pgr_validate_and_register: key" 7645 " validation and backout failed", ddi_driver_name(cdip), 7646 ddi_get_instance(cdip))); 7647 if (rval == VHCI_PGR_ILLEGALOP) { 7648 VHCI_DEBUG(4, (CE_WARN, NULL, 7649 "%s%d: vhci_pgr_validate_and_register: key" 7650 " already cleared", ddi_driver_name(cdip), 7651 ddi_get_instance(cdip))); 7652 rval = 1; 7653 } else 7654 rval = 0; 7655 } else { 7656 VHCI_DEBUG(4, (CE_NOTE, NULL, 7657 "%s%d: vhci_pgr_validate_and_register: key" 7658 " validation failed, key backed out\n", 7659 ddi_driver_name(cdip), ddi_get_instance(cdip))); 7660 vhci_print_prout_keys(vlun, "v_pgr_val_reg: key backed out: "); 7661 } 7662 mdi_rele_path(pip); 7663 7664 return (rval); 7665 } 7666 7667 /* 7668 * taskq routine to dispatch a scsi cmd to vhci_scsi_start. This ensures 7669 * that vhci_scsi_start is not called in interrupt context. 7670 * As the upper layer gets TRAN_ACCEPT when the command is dispatched, we 7671 * need to complete the command if something goes wrong. 7672 */ 7673 static void 7674 vhci_dispatch_scsi_start(void *arg) 7675 { 7676 struct vhci_pkt *vpkt = (struct vhci_pkt *)arg; 7677 struct scsi_pkt *tpkt = vpkt->vpkt_tgt_pkt; 7678 int rval = TRAN_BUSY; 7679 7680 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_dispatch_scsi_start: sending" 7681 " scsi-2 reserve for 0x%p\n", 7682 (void *)ADDR2DIP(&(vpkt->vpkt_tgt_pkt->pkt_address)))); 7683 7684 /* 7685 * To prevent the taskq from being called recursively we set the 7686 * the VHCI_PKT_THRU_TASKQ bit in the vhci_pkt_states. 7687 */ 7688 vpkt->vpkt_state |= VHCI_PKT_THRU_TASKQ; 7689 7690 /* 7691 * Wait for the transport to get ready to send packets 7692 * and if it times out, it will return something other than 7693 * TRAN_BUSY. The vhci_reserve_delay may want to 7694 * get tuned for other transports and is therefore a global. 7695 * Using delay since this routine is called by taskq dispatch 7696 * and not called during interrupt context. 7697 */ 7698 while ((rval = vhci_scsi_start(&(vpkt->vpkt_tgt_pkt->pkt_address), 7699 vpkt->vpkt_tgt_pkt)) == TRAN_BUSY) { 7700 delay(drv_usectohz(vhci_reserve_delay)); 7701 } 7702 7703 switch (rval) { 7704 case TRAN_ACCEPT: 7705 return; 7706 7707 default: 7708 /* 7709 * This pkt shall be retried, and to ensure another taskq 7710 * is dispatched for it, clear the VHCI_PKT_THRU_TASKQ 7711 * flag. 7712 */ 7713 vpkt->vpkt_state &= ~VHCI_PKT_THRU_TASKQ; 7714 7715 /* Ensure that the pkt is retried without a reset */ 7716 tpkt->pkt_reason = CMD_ABORTED; 7717 tpkt->pkt_statistics |= STAT_ABORTED; 7718 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_dispatch_scsi_start: " 7719 "TRAN_rval %d returned for dip 0x%p", rval, 7720 (void *)ADDR2DIP(&(vpkt->vpkt_tgt_pkt->pkt_address)))); 7721 break; 7722 } 7723 7724 /* 7725 * vpkt_org_vpkt should always be NULL here if the retry command 7726 * has been successfully dispatched. If vpkt_org_vpkt != NULL at 7727 * this point, it is an error so restore the original vpkt and 7728 * return an error to the target driver so it can retry the 7729 * command as appropriate. 7730 */ 7731 if (vpkt->vpkt_org_vpkt != NULL) { 7732 struct vhci_pkt *new_vpkt = vpkt; 7733 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 7734 mdi_pi_get_vhci_private(vpkt->vpkt_path); 7735 7736 vpkt = vpkt->vpkt_org_vpkt; 7737 7738 vpkt->vpkt_tgt_pkt->pkt_reason = tpkt->pkt_reason; 7739 vpkt->vpkt_tgt_pkt->pkt_statistics = tpkt->pkt_statistics; 7740 7741 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 7742 new_vpkt->vpkt_tgt_pkt); 7743 7744 tpkt = vpkt->vpkt_tgt_pkt; 7745 } 7746 7747 if (tpkt->pkt_comp) { 7748 (*tpkt->pkt_comp)(tpkt); 7749 } 7750 } 7751 7752 static void 7753 vhci_initiate_auto_failback(void *arg) 7754 { 7755 struct scsi_vhci_lun *vlun = (struct scsi_vhci_lun *)arg; 7756 dev_info_t *vdip, *cdip; 7757 int held; 7758 7759 cdip = vlun->svl_dip; 7760 vdip = ddi_get_parent(cdip); 7761 7762 VHCI_HOLD_LUN(vlun, VH_SLEEP, held); 7763 7764 /* 7765 * Perform a final check to see if the active path class is indeed 7766 * not the prefered path class. As in the time the auto failback 7767 * was dispatched, an external failover could have been detected. 7768 * [Some other host could have detected this condition and triggered 7769 * the auto failback before]. 7770 * In such a case if we go ahead with failover we will be negating the 7771 * whole purpose of auto failback. 7772 */ 7773 mutex_enter(&vlun->svl_mutex); 7774 if (vlun->svl_active_pclass != NULL) { 7775 char *best_pclass; 7776 struct scsi_failover_ops *fo; 7777 7778 fo = vlun->svl_fops; 7779 7780 (*fo->sfo_pathclass_next)(NULL, &best_pclass, 7781 vlun->svl_fops_ctpriv); 7782 if (strcmp(vlun->svl_active_pclass, best_pclass) == 0) { 7783 mutex_exit(&vlun->svl_mutex); 7784 VHCI_RELEASE_LUN(vlun); 7785 VHCI_DEBUG(1, (CE_NOTE, NULL, "Not initiating " 7786 "auto failback for %s as %s pathclass already " 7787 "active.\n", vlun->svl_lun_wwn, best_pclass)); 7788 return; 7789 } 7790 } 7791 mutex_exit(&vlun->svl_mutex); 7792 if (mdi_failover(vdip, vlun->svl_dip, MDI_FAILOVER_SYNC) 7793 == MDI_SUCCESS) { 7794 vhci_log(CE_NOTE, vdip, "!Auto failback operation " 7795 "succeeded for device %s (GUID %s)", 7796 ddi_node_name(cdip), vlun->svl_lun_wwn); 7797 } else { 7798 vhci_log(CE_NOTE, vdip, "!Auto failback operation " 7799 "failed for device %s (GUID %s)", 7800 ddi_node_name(cdip), vlun->svl_lun_wwn); 7801 } 7802 VHCI_RELEASE_LUN(vlun); 7803 } 7804 7805 #ifdef DEBUG 7806 static void 7807 vhci_print_prin_keys(vhci_prin_readkeys_t *prin, int numkeys) 7808 { 7809 uchar_t index = 0; 7810 char buf[100]; 7811 7812 VHCI_DEBUG(5, (CE_NOTE, NULL, "num keys %d\n", numkeys)); 7813 7814 while (index < numkeys) { 7815 bcopy(&prin->keylist[index], buf, MHIOC_RESV_KEY_SIZE); 7816 VHCI_DEBUG(5, (CE_NOTE, NULL, 7817 "%02x%02x%02x%02x%02x%02x%02x%02x\t", 7818 buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6], 7819 buf[7])); 7820 index++; 7821 } 7822 } 7823 #endif 7824 7825 static void 7826 vhci_print_prout_keys(scsi_vhci_lun_t *vlun, char *msg) 7827 { 7828 int i; 7829 vhci_prout_t *prout; 7830 char buf1[4*MHIOC_RESV_KEY_SIZE + 1]; 7831 char buf2[4*MHIOC_RESV_KEY_SIZE + 1]; 7832 char buf3[4*MHIOC_RESV_KEY_SIZE + 1]; 7833 char buf4[4*MHIOC_RESV_KEY_SIZE + 1]; 7834 7835 prout = &vlun->svl_prout; 7836 7837 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7838 (void) sprintf(&buf1[4*i], "[%02x]", prout->res_key[i]); 7839 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7840 (void) sprintf(&buf2[(4*i)], "[%02x]", prout->service_key[i]); 7841 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7842 (void) sprintf(&buf3[4*i], "[%02x]", prout->active_res_key[i]); 7843 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7844 (void) sprintf(&buf4[4*i], "[%02x]", 7845 prout->active_service_key[i]); 7846 7847 /* Printing all in one go. Otherwise it will jumble up */ 7848 VHCI_DEBUG(5, (CE_CONT, NULL, "%s vlun 0x%p, thread 0x%p\n" 7849 "res_key: : %s\n" 7850 "service_key : %s\n" 7851 "active_res_key : %s\n" 7852 "active_service_key: %s\n", 7853 msg, (void *)vlun, (void *)curthread, buf1, buf2, buf3, buf4)); 7854 } 7855 7856 /* 7857 * Called from vhci_scsi_start to update the pHCI pkt with target packet. 7858 */ 7859 static void 7860 vhci_update_pHCI_pkt(struct vhci_pkt *vpkt, struct scsi_pkt *pkt) 7861 { 7862 7863 ASSERT(vpkt->vpkt_hba_pkt); 7864 7865 vpkt->vpkt_hba_pkt->pkt_flags = pkt->pkt_flags; 7866 vpkt->vpkt_hba_pkt->pkt_flags |= FLAG_NOQUEUE; 7867 7868 if ((vpkt->vpkt_hba_pkt->pkt_flags & FLAG_NOINTR) || 7869 MDI_PI_IS_SUSPENDED(vpkt->vpkt_path)) { 7870 /* 7871 * Polled Command is requested or HBA is in 7872 * suspended state 7873 */ 7874 vpkt->vpkt_hba_pkt->pkt_flags |= FLAG_NOINTR; 7875 vpkt->vpkt_hba_pkt->pkt_comp = NULL; 7876 } else { 7877 vpkt->vpkt_hba_pkt->pkt_comp = vhci_intr; 7878 } 7879 vpkt->vpkt_hba_pkt->pkt_time = pkt->pkt_time; 7880 bcopy(pkt->pkt_cdbp, vpkt->vpkt_hba_pkt->pkt_cdbp, 7881 vpkt->vpkt_tgt_init_cdblen); 7882 vpkt->vpkt_hba_pkt->pkt_resid = pkt->pkt_resid; 7883 7884 /* Re-initialize the following pHCI packet state information */ 7885 vpkt->vpkt_hba_pkt->pkt_state = 0; 7886 vpkt->vpkt_hba_pkt->pkt_statistics = 0; 7887 vpkt->vpkt_hba_pkt->pkt_reason = 0; 7888 } 7889 7890 static int 7891 vhci_scsi_bus_power(dev_info_t *parent, void *impl_arg, pm_bus_power_op_t op, 7892 void *arg, void *result) 7893 { 7894 int ret = DDI_SUCCESS; 7895 7896 /* 7897 * Generic processing in MPxIO framework 7898 */ 7899 ret = mdi_bus_power(parent, impl_arg, op, arg, result); 7900 7901 switch (ret) { 7902 case MDI_SUCCESS: 7903 ret = DDI_SUCCESS; 7904 break; 7905 case MDI_FAILURE: 7906 ret = DDI_FAILURE; 7907 break; 7908 default: 7909 break; 7910 } 7911 7912 return (ret); 7913 } 7914 7915 static int 7916 vhci_pHCI_cap(struct scsi_address *ap, char *cap, int val, int whom, 7917 mdi_pathinfo_t *pip) 7918 { 7919 dev_info_t *cdip; 7920 mdi_pathinfo_t *npip = NULL; 7921 scsi_vhci_priv_t *svp = NULL; 7922 struct scsi_address *pap = NULL; 7923 scsi_hba_tran_t *hba = NULL; 7924 int sps; 7925 int mps_flag; 7926 int rval = 0; 7927 7928 mps_flag = (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH); 7929 if (pip) { 7930 /* 7931 * If the call is from vhci_pathinfo_state_change, 7932 * then this path was busy and is becoming ready to accept IO. 7933 */ 7934 ASSERT(ap != NULL); 7935 hba = ap->a_hba_tran; 7936 ASSERT(hba != NULL); 7937 rval = scsi_ifsetcap(ap, cap, val, whom); 7938 7939 VHCI_DEBUG(2, (CE_NOTE, NULL, 7940 "!vhci_pHCI_cap: only on path %p, ap %p, rval %x\n", 7941 (void *)pip, (void *)ap, rval)); 7942 7943 return (rval); 7944 } 7945 7946 /* 7947 * Set capability on all the pHCIs. 7948 * If any path is busy, then the capability would be set by 7949 * vhci_pathinfo_state_change. 7950 */ 7951 7952 cdip = ADDR2DIP(ap); 7953 ASSERT(cdip != NULL); 7954 sps = mdi_select_path(cdip, NULL, mps_flag, NULL, &pip); 7955 if ((sps != MDI_SUCCESS) || (pip == NULL)) { 7956 VHCI_DEBUG(2, (CE_WARN, NULL, 7957 "!vhci_pHCI_cap: Unable to get a path, dip 0x%p", 7958 (void *)cdip)); 7959 return (0); 7960 } 7961 7962 again: 7963 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 7964 if (svp == NULL) { 7965 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_pHCI_cap: " 7966 "priv is NULL, pip 0x%p", (void *)pip)); 7967 mdi_rele_path(pip); 7968 return (rval); 7969 } 7970 7971 if (svp->svp_psd == NULL) { 7972 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_pHCI_cap: " 7973 "psd is NULL, pip 0x%p, svp 0x%p", 7974 (void *)pip, (void *)svp)); 7975 mdi_rele_path(pip); 7976 return (rval); 7977 } 7978 7979 pap = &svp->svp_psd->sd_address; 7980 ASSERT(pap != NULL); 7981 hba = pap->a_hba_tran; 7982 ASSERT(hba != NULL); 7983 7984 if (hba->tran_setcap != NULL) { 7985 rval = scsi_ifsetcap(pap, cap, val, whom); 7986 7987 VHCI_DEBUG(2, (CE_NOTE, NULL, 7988 "!vhci_pHCI_cap: path %p, ap %p, rval %x\n", 7989 (void *)pip, (void *)ap, rval)); 7990 7991 /* 7992 * Select next path and issue the setcap, repeat 7993 * until all paths are exhausted 7994 */ 7995 sps = mdi_select_path(cdip, NULL, mps_flag, pip, &npip); 7996 if ((sps != MDI_SUCCESS) || (npip == NULL)) { 7997 mdi_rele_path(pip); 7998 return (1); 7999 } 8000 mdi_rele_path(pip); 8001 pip = npip; 8002 goto again; 8003 } 8004 mdi_rele_path(pip); 8005 return (rval); 8006 } 8007 8008 static int 8009 vhci_scsi_bus_config(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op, 8010 void *arg, dev_info_t **child) 8011 { 8012 char *guid; 8013 8014 if (op == BUS_CONFIG_ONE || op == BUS_UNCONFIG_ONE) 8015 guid = vhci_devnm_to_guid((char *)arg); 8016 else 8017 guid = NULL; 8018 8019 if (mdi_vhci_bus_config(pdip, flags, op, arg, child, guid) 8020 == MDI_SUCCESS) 8021 return (NDI_SUCCESS); 8022 else 8023 return (NDI_FAILURE); 8024 } 8025 8026 /* 8027 * Take the original vhci_pkt, create a duplicate of the pkt for resending 8028 * as though it originated in ssd. 8029 */ 8030 static struct scsi_pkt * 8031 vhci_create_retry_pkt(struct vhci_pkt *vpkt) 8032 { 8033 struct vhci_pkt *new_vpkt = NULL; 8034 struct scsi_pkt *pkt = NULL; 8035 8036 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 8037 mdi_pi_get_vhci_private(vpkt->vpkt_path); 8038 8039 /* 8040 * Ensure consistent data at completion time by setting PKT_CONSISTENT 8041 */ 8042 pkt = vhci_scsi_init_pkt(&svp->svp_psd->sd_address, pkt, 8043 vpkt->vpkt_tgt_init_bp, vpkt->vpkt_tgt_init_cdblen, 8044 vpkt->vpkt_tgt_init_scblen, 0, 8045 PKT_CONSISTENT, 8046 NULL_FUNC, NULL); 8047 if (pkt != NULL) { 8048 new_vpkt = TGTPKT2VHCIPKT(pkt); 8049 8050 pkt->pkt_address = vpkt->vpkt_tgt_pkt->pkt_address; 8051 pkt->pkt_flags = vpkt->vpkt_tgt_pkt->pkt_flags; 8052 pkt->pkt_time = vpkt->vpkt_tgt_pkt->pkt_time; 8053 pkt->pkt_comp = vpkt->vpkt_tgt_pkt->pkt_comp; 8054 8055 pkt->pkt_resid = 0; 8056 pkt->pkt_statistics = 0; 8057 pkt->pkt_reason = 0; 8058 8059 bcopy(vpkt->vpkt_tgt_pkt->pkt_cdbp, 8060 pkt->pkt_cdbp, vpkt->vpkt_tgt_init_cdblen); 8061 8062 /* 8063 * Save a pointer to the original vhci_pkt 8064 */ 8065 new_vpkt->vpkt_org_vpkt = vpkt; 8066 } 8067 8068 return (pkt); 8069 } 8070 8071 /* 8072 * Copy the successful completion information from the hba packet into 8073 * the original target pkt from the upper layer. Returns the original 8074 * vpkt and destroys the new vpkt from the internal retry. 8075 */ 8076 static struct vhci_pkt * 8077 vhci_sync_retry_pkt(struct vhci_pkt *vpkt) 8078 { 8079 struct vhci_pkt *ret_vpkt = NULL; 8080 struct scsi_pkt *tpkt = NULL; 8081 struct scsi_pkt *hba_pkt = NULL; 8082 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 8083 mdi_pi_get_vhci_private(vpkt->vpkt_path); 8084 8085 ASSERT(vpkt->vpkt_org_vpkt != NULL); 8086 VHCI_DEBUG(0, (CE_NOTE, NULL, "vhci_sync_retry_pkt: Retry pkt " 8087 "completed successfully!\n")); 8088 8089 ret_vpkt = vpkt->vpkt_org_vpkt; 8090 tpkt = ret_vpkt->vpkt_tgt_pkt; 8091 hba_pkt = vpkt->vpkt_hba_pkt; 8092 8093 /* 8094 * Copy the good status into the target driver's packet 8095 */ 8096 *(tpkt->pkt_scbp) = *(hba_pkt->pkt_scbp); 8097 tpkt->pkt_resid = hba_pkt->pkt_resid; 8098 tpkt->pkt_state = hba_pkt->pkt_state; 8099 tpkt->pkt_statistics = hba_pkt->pkt_statistics; 8100 tpkt->pkt_reason = hba_pkt->pkt_reason; 8101 8102 /* 8103 * Destroy the internally created vpkt for the retry 8104 */ 8105 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 8106 vpkt->vpkt_tgt_pkt); 8107 8108 return (ret_vpkt); 8109 } 8110 8111 /* restart the request sense request */ 8112 static void 8113 vhci_uscsi_restart_sense(void *arg) 8114 { 8115 struct buf *rqbp; 8116 struct buf *bp; 8117 struct scsi_pkt *rqpkt = (struct scsi_pkt *)arg; 8118 mp_uscsi_cmd_t *mp_uscmdp; 8119 8120 VHCI_DEBUG(4, (CE_WARN, NULL, 8121 "vhci_uscsi_restart_sense: enter: rqpkt: %p", (void *)rqpkt)); 8122 8123 if (scsi_transport(rqpkt) != TRAN_ACCEPT) { 8124 /* if it fails - need to wakeup the original command */ 8125 mp_uscmdp = rqpkt->pkt_private; 8126 bp = mp_uscmdp->cmdbp; 8127 rqbp = mp_uscmdp->rqbp; 8128 ASSERT(mp_uscmdp && bp && rqbp); 8129 scsi_free_consistent_buf(rqbp); 8130 scsi_destroy_pkt(rqpkt); 8131 bp->b_resid = bp->b_bcount; 8132 bioerror(bp, EIO); 8133 biodone(bp); 8134 } 8135 } 8136 8137 /* 8138 * auto-rqsense is not enabled so we have to retrieve the request sense 8139 * manually. 8140 */ 8141 static int 8142 vhci_uscsi_send_sense(struct scsi_pkt *pkt, mp_uscsi_cmd_t *mp_uscmdp) 8143 { 8144 struct buf *rqbp, *cmdbp; 8145 struct scsi_pkt *rqpkt; 8146 int rval = 0; 8147 8148 cmdbp = mp_uscmdp->cmdbp; 8149 ASSERT(cmdbp != NULL); 8150 8151 VHCI_DEBUG(4, (CE_WARN, NULL, 8152 "vhci_uscsi_send_sense: enter: bp: %p pkt: %p scmd: %p", 8153 (void *)cmdbp, (void *)pkt, (void *)mp_uscmdp)); 8154 /* set up the packet information and cdb */ 8155 if ((rqbp = scsi_alloc_consistent_buf(mp_uscmdp->ap, NULL, 8156 SENSE_LENGTH, B_READ, NULL, NULL)) == NULL) { 8157 return (-1); 8158 } 8159 8160 if ((rqpkt = scsi_init_pkt(mp_uscmdp->ap, NULL, rqbp, 8161 CDB_GROUP0, 1, 0, PKT_CONSISTENT, NULL, NULL)) == NULL) { 8162 scsi_free_consistent_buf(rqbp); 8163 return (-1); 8164 } 8165 8166 (void) scsi_setup_cdb((union scsi_cdb *)(intptr_t)rqpkt->pkt_cdbp, 8167 SCMD_REQUEST_SENSE, 0, SENSE_LENGTH, 0); 8168 8169 mp_uscmdp->rqbp = rqbp; 8170 rqbp->b_private = mp_uscmdp; 8171 rqpkt->pkt_flags |= FLAG_SENSING; 8172 rqpkt->pkt_time = 60; 8173 rqpkt->pkt_comp = vhci_uscsi_iodone; 8174 rqpkt->pkt_private = mp_uscmdp; 8175 8176 /* 8177 * NOTE: This code path is related to MPAPI uscsi(7I), so path 8178 * selection is not based on path_instance. 8179 */ 8180 if (scsi_pkt_allocated_correctly(rqpkt)) 8181 rqpkt->pkt_path_instance = 0; 8182 8183 /* get her done */ 8184 switch (scsi_transport(rqpkt)) { 8185 case TRAN_ACCEPT: 8186 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_uscsi_send_sense: " 8187 "transport accepted.")); 8188 break; 8189 case TRAN_BUSY: 8190 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_uscsi_send_sense: " 8191 "transport busy, setting timeout.")); 8192 vhci_restart_timeid = timeout(vhci_uscsi_restart_sense, rqpkt, 8193 (drv_usectohz(5 * 1000000))); 8194 break; 8195 default: 8196 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_uscsi_send_sense: " 8197 "transport failed")); 8198 scsi_free_consistent_buf(rqbp); 8199 scsi_destroy_pkt(rqpkt); 8200 rval = -1; 8201 } 8202 8203 return (rval); 8204 } 8205 8206 /* 8207 * done routine for the mpapi uscsi command - this is behaving as though 8208 * FLAG_DIAGNOSE is set meaning there are no retries except for a manual 8209 * request sense. 8210 */ 8211 void 8212 vhci_uscsi_iodone(struct scsi_pkt *pkt) 8213 { 8214 struct buf *bp; 8215 mp_uscsi_cmd_t *mp_uscmdp; 8216 struct uscsi_cmd *uscmdp; 8217 struct scsi_arq_status *arqstat; 8218 int err; 8219 8220 mp_uscmdp = (mp_uscsi_cmd_t *)pkt->pkt_private; 8221 uscmdp = mp_uscmdp->uscmdp; 8222 bp = mp_uscmdp->cmdbp; 8223 ASSERT(bp != NULL); 8224 VHCI_DEBUG(4, (CE_WARN, NULL, 8225 "vhci_uscsi_iodone: enter: bp: %p pkt: %p scmd: %p", 8226 (void *)bp, (void *)pkt, (void *)mp_uscmdp)); 8227 /* Save the status and the residual into the uscsi_cmd struct */ 8228 uscmdp->uscsi_status = ((*(pkt)->pkt_scbp) & STATUS_MASK); 8229 uscmdp->uscsi_resid = bp->b_resid; 8230 8231 /* return on a very successful command */ 8232 if (pkt->pkt_reason == CMD_CMPLT && 8233 SCBP_C(pkt) == 0 && ((pkt->pkt_flags & FLAG_SENSING) == 0) && 8234 pkt->pkt_resid == 0) { 8235 mdi_pi_kstat_iosupdate(mp_uscmdp->pip, bp); 8236 scsi_destroy_pkt(pkt); 8237 biodone(bp); 8238 return; 8239 } 8240 VHCI_DEBUG(4, (CE_NOTE, NULL, "iodone: reason=0x%x " 8241 " pkt_resid=%ld pkt_state: 0x%x b_count: %ld b_resid: %ld", 8242 pkt->pkt_reason, pkt->pkt_resid, 8243 pkt->pkt_state, bp->b_bcount, bp->b_resid)); 8244 8245 err = EIO; 8246 8247 arqstat = (struct scsi_arq_status *)(intptr_t)(pkt->pkt_scbp); 8248 if (pkt->pkt_reason != CMD_CMPLT) { 8249 /* 8250 * The command did not complete. 8251 */ 8252 VHCI_DEBUG(4, (CE_NOTE, NULL, 8253 "vhci_uscsi_iodone: command did not complete." 8254 " reason: %x flag: %x", pkt->pkt_reason, pkt->pkt_flags)); 8255 if (pkt->pkt_flags & FLAG_SENSING) { 8256 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_TRANSERR); 8257 } else if (pkt->pkt_reason == CMD_TIMEOUT) { 8258 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_HARDERR); 8259 err = ETIMEDOUT; 8260 } 8261 } else if (pkt->pkt_state & STATE_ARQ_DONE && mp_uscmdp->arq_enabled) { 8262 /* 8263 * The auto-rqsense happened, and the packet has a filled-in 8264 * scsi_arq_status structure, pointed to by pkt_scbp. 8265 */ 8266 VHCI_DEBUG(4, (CE_NOTE, NULL, 8267 "vhci_uscsi_iodone: received auto-requested sense")); 8268 if (uscmdp->uscsi_flags & USCSI_RQENABLE) { 8269 /* get the amount of data to copy into rqbuf */ 8270 int rqlen = SENSE_LENGTH - arqstat->sts_rqpkt_resid; 8271 rqlen = min(((int)uscmdp->uscsi_rqlen), rqlen); 8272 uscmdp->uscsi_rqresid = uscmdp->uscsi_rqlen - rqlen; 8273 uscmdp->uscsi_rqstatus = 8274 *((char *)&arqstat->sts_rqpkt_status); 8275 if (uscmdp->uscsi_rqbuf && uscmdp->uscsi_rqlen && 8276 rqlen != 0) { 8277 bcopy(&(arqstat->sts_sensedata), 8278 uscmdp->uscsi_rqbuf, rqlen); 8279 } 8280 mdi_pi_kstat_iosupdate(mp_uscmdp->pip, bp); 8281 VHCI_DEBUG(4, (CE_NOTE, NULL, 8282 "vhci_uscsi_iodone: ARQ " 8283 "uscsi_rqstatus=0x%x uscsi_rqresid=%d rqlen: %d " 8284 "xfer: %d rqpkt_resid: %d\n", 8285 uscmdp->uscsi_rqstatus, uscmdp->uscsi_rqresid, 8286 uscmdp->uscsi_rqlen, rqlen, 8287 arqstat->sts_rqpkt_resid)); 8288 } 8289 } else if (pkt->pkt_flags & FLAG_SENSING) { 8290 struct buf *rqbp; 8291 struct scsi_status *rqstatus; 8292 8293 rqstatus = (struct scsi_status *)pkt->pkt_scbp; 8294 /* a manual request sense was done - get the information */ 8295 if (uscmdp->uscsi_flags & USCSI_RQENABLE) { 8296 int rqlen = SENSE_LENGTH - pkt->pkt_resid; 8297 8298 rqbp = mp_uscmdp->rqbp; 8299 /* get the amount of data to copy into rqbuf */ 8300 rqlen = min(((int)uscmdp->uscsi_rqlen), rqlen); 8301 uscmdp->uscsi_rqresid = uscmdp->uscsi_rqlen - rqlen; 8302 uscmdp->uscsi_rqstatus = *((char *)rqstatus); 8303 if (uscmdp->uscsi_rqlen && uscmdp->uscsi_rqbuf) { 8304 bcopy(rqbp->b_un.b_addr, uscmdp->uscsi_rqbuf, 8305 rqlen); 8306 } 8307 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_TRANSERR); 8308 scsi_free_consistent_buf(rqbp); 8309 } 8310 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_uscsi_iodone: FLAG_SENSING" 8311 "uscsi_rqstatus=0x%x uscsi_rqresid=%d\n", 8312 uscmdp->uscsi_rqstatus, uscmdp->uscsi_rqresid)); 8313 } else { 8314 struct scsi_status *status = 8315 (struct scsi_status *)pkt->pkt_scbp; 8316 /* 8317 * Command completed and we're not getting sense. Check for 8318 * errors and decide what to do next. 8319 */ 8320 VHCI_DEBUG(4, (CE_NOTE, NULL, 8321 "vhci_uscsi_iodone: command appears complete: reason: %x", 8322 pkt->pkt_reason)); 8323 if (status->sts_chk) { 8324 /* need to manually get the request sense */ 8325 if (vhci_uscsi_send_sense(pkt, mp_uscmdp) == 0) { 8326 scsi_destroy_pkt(pkt); 8327 return; 8328 } 8329 } else { 8330 VHCI_DEBUG(4, (CE_NOTE, NULL, 8331 "vhci_chk_err: appears complete")); 8332 err = 0; 8333 mdi_pi_kstat_iosupdate(mp_uscmdp->pip, bp); 8334 if (pkt->pkt_resid) { 8335 bp->b_resid += pkt->pkt_resid; 8336 } 8337 } 8338 } 8339 8340 if (err) { 8341 if (bp->b_resid == 0) 8342 bp->b_resid = bp->b_bcount; 8343 bioerror(bp, err); 8344 bp->b_flags |= B_ERROR; 8345 } 8346 8347 scsi_destroy_pkt(pkt); 8348 biodone(bp); 8349 8350 VHCI_DEBUG(4, (CE_WARN, NULL, "vhci_uscsi_iodone: exit")); 8351 } 8352 8353 /* 8354 * start routine for the mpapi uscsi command 8355 */ 8356 int 8357 vhci_uscsi_iostart(struct buf *bp) 8358 { 8359 struct scsi_pkt *pkt; 8360 struct uscsi_cmd *uscmdp; 8361 mp_uscsi_cmd_t *mp_uscmdp; 8362 int stat_size, rval; 8363 int retry = 0; 8364 8365 ASSERT(bp->b_private != NULL); 8366 8367 mp_uscmdp = (mp_uscsi_cmd_t *)bp->b_private; 8368 uscmdp = mp_uscmdp->uscmdp; 8369 if (uscmdp->uscsi_flags & USCSI_RQENABLE) { 8370 stat_size = SENSE_LENGTH; 8371 } else { 8372 stat_size = 1; 8373 } 8374 8375 pkt = scsi_init_pkt(mp_uscmdp->ap, NULL, bp, uscmdp->uscsi_cdblen, 8376 stat_size, 0, 0, SLEEP_FUNC, NULL); 8377 if (pkt == NULL) { 8378 VHCI_DEBUG(4, (CE_NOTE, NULL, 8379 "vhci_uscsi_iostart: rval: EINVAL")); 8380 bp->b_resid = bp->b_bcount; 8381 uscmdp->uscsi_resid = bp->b_bcount; 8382 bioerror(bp, EINVAL); 8383 biodone(bp); 8384 return (EINVAL); 8385 } 8386 8387 pkt->pkt_time = uscmdp->uscsi_timeout; 8388 bcopy(uscmdp->uscsi_cdb, pkt->pkt_cdbp, (size_t)uscmdp->uscsi_cdblen); 8389 pkt->pkt_comp = vhci_uscsi_iodone; 8390 pkt->pkt_private = mp_uscmdp; 8391 if (uscmdp->uscsi_flags & USCSI_SILENT) 8392 pkt->pkt_flags |= FLAG_SILENT; 8393 if (uscmdp->uscsi_flags & USCSI_ISOLATE) 8394 pkt->pkt_flags |= FLAG_ISOLATE; 8395 if (uscmdp->uscsi_flags & USCSI_DIAGNOSE) 8396 pkt->pkt_flags |= FLAG_DIAGNOSE; 8397 if (uscmdp->uscsi_flags & USCSI_RENEGOT) { 8398 pkt->pkt_flags |= FLAG_RENEGOTIATE_WIDE_SYNC; 8399 } 8400 VHCI_DEBUG(4, (CE_WARN, NULL, 8401 "vhci_uscsi_iostart: ap: %p pkt: %p pcdbp: %p uscmdp: %p" 8402 " ucdbp: %p pcdblen: %d bp: %p count: %ld pip: %p" 8403 " stat_size: %d", 8404 (void *)mp_uscmdp->ap, (void *)pkt, (void *)pkt->pkt_cdbp, 8405 (void *)uscmdp, (void *)uscmdp->uscsi_cdb, pkt->pkt_cdblen, 8406 (void *)bp, bp->b_bcount, (void *)mp_uscmdp->pip, stat_size)); 8407 8408 /* 8409 * NOTE: This code path is related to MPAPI uscsi(7I), so path 8410 * selection is not based on path_instance. 8411 */ 8412 if (scsi_pkt_allocated_correctly(pkt)) 8413 pkt->pkt_path_instance = 0; 8414 8415 while (((rval = scsi_transport(pkt)) == TRAN_BUSY) && 8416 retry < vhci_uscsi_retry_count) { 8417 delay(drv_usectohz(vhci_uscsi_delay)); 8418 retry++; 8419 } 8420 if (retry >= vhci_uscsi_retry_count) { 8421 VHCI_DEBUG(4, (CE_NOTE, NULL, 8422 "vhci_uscsi_iostart: tran_busy - retry: %d", retry)); 8423 } 8424 switch (rval) { 8425 case TRAN_ACCEPT: 8426 rval = 0; 8427 break; 8428 8429 default: 8430 VHCI_DEBUG(4, (CE_NOTE, NULL, 8431 "vhci_uscsi_iostart: rval: %d count: %ld res: %ld", 8432 rval, bp->b_bcount, bp->b_resid)); 8433 bp->b_resid = bp->b_bcount; 8434 uscmdp->uscsi_resid = bp->b_bcount; 8435 bioerror(bp, EIO); 8436 scsi_destroy_pkt(pkt); 8437 biodone(bp); 8438 rval = EIO; 8439 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_TRANSERR); 8440 break; 8441 } 8442 VHCI_DEBUG(4, (CE_NOTE, NULL, 8443 "vhci_uscsi_iostart: exit: rval: %d", rval)); 8444 return (rval); 8445 } 8446