1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 /* 25 * Copyright 2014 Nexenta Systems, Inc. All rights reserved. 26 */ 27 28 /* 29 * Multiplexed I/O SCSI vHCI implementation 30 */ 31 32 #include <sys/conf.h> 33 #include <sys/file.h> 34 #include <sys/ddi.h> 35 #include <sys/sunddi.h> 36 #include <sys/scsi/scsi.h> 37 #include <sys/scsi/impl/scsi_reset_notify.h> 38 #include <sys/scsi/impl/services.h> 39 #include <sys/sunmdi.h> 40 #include <sys/mdi_impldefs.h> 41 #include <sys/scsi/adapters/scsi_vhci.h> 42 #include <sys/disp.h> 43 #include <sys/byteorder.h> 44 45 extern uintptr_t scsi_callback_id; 46 extern ddi_dma_attr_t scsi_alloc_attr; 47 48 #ifdef DEBUG 49 int vhci_debug = VHCI_DEBUG_DEFAULT_VAL; 50 #endif 51 52 /* retry for the vhci_do_prout command when a not ready is returned */ 53 int vhci_prout_not_ready_retry = 180; 54 55 /* 56 * These values are defined to support the internal retry of 57 * SCSI packets for better sense code handling. 58 */ 59 #define VHCI_CMD_CMPLT 0 60 #define VHCI_CMD_RETRY 1 61 #define VHCI_CMD_ERROR -1 62 63 #define PROPFLAGS (DDI_PROP_DONTPASS | DDI_PROP_NOTPROM) 64 #define VHCI_SCSI_PERR 0x47 65 #define VHCI_PGR_ILLEGALOP -2 66 #define VHCI_NUM_UPDATE_TASKQ 8 67 /* changed to 132 to accomodate HDS */ 68 69 /* 70 * Version Macros 71 */ 72 #define VHCI_NAME_VERSION "SCSI VHCI Driver" 73 char vhci_version_name[] = VHCI_NAME_VERSION; 74 75 int vhci_first_time = 0; 76 clock_t vhci_to_ticks = 0; 77 int vhci_init_wait_timeout = VHCI_INIT_WAIT_TIMEOUT; 78 kcondvar_t vhci_cv; 79 kmutex_t vhci_global_mutex; 80 void *vhci_softstate = NULL; /* for soft state */ 81 82 /* 83 * Flag to delay the retry of the reserve command 84 */ 85 int vhci_reserve_delay = 100000; 86 static int vhci_path_quiesce_timeout = 60; 87 static uchar_t zero_key[MHIOC_RESV_KEY_SIZE]; 88 89 /* uscsi delay for a TRAN_BUSY */ 90 static int vhci_uscsi_delay = 100000; 91 static int vhci_uscsi_retry_count = 180; 92 /* uscsi_restart_sense timeout id in case it needs to get canceled */ 93 static timeout_id_t vhci_restart_timeid = 0; 94 95 static int vhci_bus_config_debug = 0; 96 97 /* 98 * Bidirectional map of 'target-port' to port id <pid> for support of 99 * iostat(1M) '-Xx' and '-Yx' output. 100 */ 101 static kmutex_t vhci_targetmap_mutex; 102 static uint_t vhci_targetmap_pid = 1; 103 static mod_hash_t *vhci_targetmap_bypid; /* <pid> -> 'target-port' */ 104 static mod_hash_t *vhci_targetmap_byport; /* 'target-port' -> <pid> */ 105 106 /* 107 * functions exported by scsi_vhci struct cb_ops 108 */ 109 static int vhci_open(dev_t *, int, int, cred_t *); 110 static int vhci_close(dev_t, int, int, cred_t *); 111 static int vhci_ioctl(dev_t, int, intptr_t, int, cred_t *, int *); 112 113 /* 114 * functions exported by scsi_vhci struct dev_ops 115 */ 116 static int vhci_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **); 117 static int vhci_attach(dev_info_t *, ddi_attach_cmd_t); 118 static int vhci_detach(dev_info_t *, ddi_detach_cmd_t); 119 120 /* 121 * functions exported by scsi_vhci scsi_hba_tran_t transport table 122 */ 123 static int vhci_scsi_tgt_init(dev_info_t *, dev_info_t *, 124 scsi_hba_tran_t *, struct scsi_device *); 125 static void vhci_scsi_tgt_free(dev_info_t *, dev_info_t *, scsi_hba_tran_t *, 126 struct scsi_device *); 127 static int vhci_pgr_register_start(scsi_vhci_lun_t *, struct scsi_pkt *); 128 static int vhci_scsi_start(struct scsi_address *, struct scsi_pkt *); 129 static int vhci_scsi_abort(struct scsi_address *, struct scsi_pkt *); 130 static int vhci_scsi_reset(struct scsi_address *, int); 131 static int vhci_scsi_reset_target(struct scsi_address *, int level, 132 uint8_t select_path); 133 static int vhci_scsi_reset_bus(struct scsi_address *); 134 static int vhci_scsi_getcap(struct scsi_address *, char *, int); 135 static int vhci_scsi_setcap(struct scsi_address *, char *, int, int); 136 static int vhci_commoncap(struct scsi_address *, char *, int, int, int); 137 static int vhci_pHCI_cap(struct scsi_address *ap, char *cap, int val, int whom, 138 mdi_pathinfo_t *pip); 139 static struct scsi_pkt *vhci_scsi_init_pkt(struct scsi_address *, 140 struct scsi_pkt *, struct buf *, int, int, int, int, int (*)(), caddr_t); 141 static void vhci_scsi_destroy_pkt(struct scsi_address *, struct scsi_pkt *); 142 static void vhci_scsi_dmafree(struct scsi_address *, struct scsi_pkt *); 143 static void vhci_scsi_sync_pkt(struct scsi_address *, struct scsi_pkt *); 144 static int vhci_scsi_reset_notify(struct scsi_address *, int, void (*)(caddr_t), 145 caddr_t); 146 static int vhci_scsi_get_bus_addr(struct scsi_device *, char *, int); 147 static int vhci_scsi_get_name(struct scsi_device *, char *, int); 148 static int vhci_scsi_bus_power(dev_info_t *, void *, pm_bus_power_op_t, 149 void *, void *); 150 static int vhci_scsi_bus_config(dev_info_t *, uint_t, ddi_bus_config_op_t, 151 void *, dev_info_t **); 152 static int vhci_scsi_bus_unconfig(dev_info_t *, uint_t, ddi_bus_config_op_t, 153 void *); 154 static struct scsi_failover_ops *vhci_dev_fo(dev_info_t *, struct scsi_device *, 155 void **, char **); 156 157 /* 158 * functions registered with the mpxio framework via mdi_vhci_ops_t 159 */ 160 static int vhci_pathinfo_init(dev_info_t *, mdi_pathinfo_t *, int); 161 static int vhci_pathinfo_uninit(dev_info_t *, mdi_pathinfo_t *, int); 162 static int vhci_pathinfo_state_change(dev_info_t *, mdi_pathinfo_t *, 163 mdi_pathinfo_state_t, uint32_t, int); 164 static int vhci_pathinfo_online(dev_info_t *, mdi_pathinfo_t *, int); 165 static int vhci_pathinfo_offline(dev_info_t *, mdi_pathinfo_t *, int); 166 static int vhci_failover(dev_info_t *, dev_info_t *, int); 167 static void vhci_client_attached(dev_info_t *); 168 static int vhci_is_dev_supported(dev_info_t *, dev_info_t *, void *); 169 170 static int vhci_ctl(dev_t, int, intptr_t, int, cred_t *, int *); 171 static int vhci_devctl(dev_t, int, intptr_t, int, cred_t *, int *); 172 static int vhci_ioc_get_phci_path(sv_iocdata_t *, caddr_t, int, caddr_t); 173 static int vhci_ioc_get_client_path(sv_iocdata_t *, caddr_t, int, caddr_t); 174 static int vhci_ioc_get_paddr(sv_iocdata_t *, caddr_t, int, caddr_t); 175 static int vhci_ioc_send_client_path(caddr_t, sv_iocdata_t *, int, caddr_t); 176 static void vhci_ioc_devi_to_path(dev_info_t *, caddr_t); 177 static int vhci_get_phci_path_list(dev_info_t *, sv_path_info_t *, uint_t); 178 static int vhci_get_client_path_list(dev_info_t *, sv_path_info_t *, uint_t); 179 static int vhci_get_iocdata(const void *, sv_iocdata_t *, int, caddr_t); 180 static int vhci_get_iocswitchdata(const void *, sv_switch_to_cntlr_iocdata_t *, 181 int, caddr_t); 182 static int vhci_ioc_alloc_pathinfo(sv_path_info_t **, sv_path_info_t **, 183 uint_t, sv_iocdata_t *, int, caddr_t); 184 static void vhci_ioc_free_pathinfo(sv_path_info_t *, sv_path_info_t *, uint_t); 185 static int vhci_ioc_send_pathinfo(sv_path_info_t *, sv_path_info_t *, uint_t, 186 sv_iocdata_t *, int, caddr_t); 187 static int vhci_handle_ext_fo(struct scsi_pkt *, int); 188 static int vhci_efo_watch_cb(caddr_t, struct scsi_watch_result *); 189 static int vhci_quiesce_lun(struct scsi_vhci_lun *); 190 static int vhci_pgr_validate_and_register(scsi_vhci_priv_t *); 191 static void vhci_dispatch_scsi_start(void *); 192 static void vhci_efo_done(void *); 193 static void vhci_initiate_auto_failback(void *); 194 static void vhci_update_pHCI_pkt(struct vhci_pkt *, struct scsi_pkt *); 195 static int vhci_update_pathinfo(struct scsi_device *, mdi_pathinfo_t *, 196 struct scsi_failover_ops *, scsi_vhci_lun_t *, struct scsi_vhci *); 197 static void vhci_kstat_create_pathinfo(mdi_pathinfo_t *); 198 static int vhci_quiesce_paths(dev_info_t *, dev_info_t *, 199 scsi_vhci_lun_t *, char *, char *); 200 201 static char *vhci_devnm_to_guid(char *); 202 static int vhci_bind_transport(struct scsi_address *, struct vhci_pkt *, 203 int, int (*func)(caddr_t)); 204 static void vhci_intr(struct scsi_pkt *); 205 static int vhci_do_prout(scsi_vhci_priv_t *); 206 static void vhci_run_cmd(void *); 207 static int vhci_do_prin(struct vhci_pkt **); 208 static struct scsi_pkt *vhci_create_retry_pkt(struct vhci_pkt *); 209 static struct vhci_pkt *vhci_sync_retry_pkt(struct vhci_pkt *); 210 static struct scsi_vhci_lun *vhci_lun_lookup(dev_info_t *); 211 static struct scsi_vhci_lun *vhci_lun_lookup_alloc(dev_info_t *, char *, int *); 212 static void vhci_lun_free(struct scsi_vhci_lun *dvlp, struct scsi_device *sd); 213 static int vhci_recovery_reset(scsi_vhci_lun_t *, struct scsi_address *, 214 uint8_t, uint8_t); 215 void vhci_update_pathstates(void *); 216 217 #ifdef DEBUG 218 static void vhci_print_prin_keys(vhci_prin_readkeys_t *, int); 219 static void vhci_print_cdb(dev_info_t *dip, uint_t level, 220 char *title, uchar_t *cdb); 221 static void vhci_clean_print(dev_info_t *dev, uint_t level, 222 char *title, uchar_t *data, int len); 223 #endif 224 static void vhci_print_prout_keys(scsi_vhci_lun_t *, char *); 225 static void vhci_uscsi_iodone(struct scsi_pkt *pkt); 226 static void vhci_invalidate_mpapi_lu(struct scsi_vhci *, scsi_vhci_lun_t *); 227 228 /* 229 * MP-API related functions 230 */ 231 extern int vhci_mpapi_init(struct scsi_vhci *); 232 extern void vhci_mpapi_add_dev_prod(struct scsi_vhci *, char *); 233 extern int vhci_mpapi_ctl(dev_t, int, intptr_t, int, cred_t *, int *); 234 extern void vhci_update_mpapi_data(struct scsi_vhci *, 235 scsi_vhci_lun_t *, mdi_pathinfo_t *); 236 extern void* vhci_get_mpapi_item(struct scsi_vhci *, mpapi_list_header_t *, 237 uint8_t, void*); 238 extern void vhci_mpapi_set_path_state(dev_info_t *, mdi_pathinfo_t *, int); 239 extern int vhci_mpapi_update_tpg_acc_state_for_lu(struct scsi_vhci *, 240 scsi_vhci_lun_t *); 241 242 #define VHCI_DMA_MAX_XFER_CAP INT_MAX 243 244 #define VHCI_MAX_PGR_RETRIES 3 245 246 /* 247 * Macros for the device-type mpxio options 248 */ 249 #define LOAD_BALANCE_OPTIONS "load-balance-options" 250 #define LOGICAL_BLOCK_REGION_SIZE "region-size" 251 #define MPXIO_OPTIONS_LIST "device-type-mpxio-options-list" 252 #define DEVICE_TYPE_STR "device-type" 253 #define isdigit(ch) ((ch) >= '0' && (ch) <= '9') 254 255 static struct cb_ops vhci_cb_ops = { 256 vhci_open, /* open */ 257 vhci_close, /* close */ 258 nodev, /* strategy */ 259 nodev, /* print */ 260 nodev, /* dump */ 261 nodev, /* read */ 262 nodev, /* write */ 263 vhci_ioctl, /* ioctl */ 264 nodev, /* devmap */ 265 nodev, /* mmap */ 266 nodev, /* segmap */ 267 nochpoll, /* chpoll */ 268 ddi_prop_op, /* cb_prop_op */ 269 0, /* streamtab */ 270 D_NEW | D_MP, /* cb_flag */ 271 CB_REV, /* rev */ 272 nodev, /* aread */ 273 nodev /* awrite */ 274 }; 275 276 static struct dev_ops vhci_ops = { 277 DEVO_REV, 278 0, 279 vhci_getinfo, 280 nulldev, /* identify */ 281 nulldev, /* probe */ 282 vhci_attach, /* attach and detach are mandatory */ 283 vhci_detach, 284 nodev, /* reset */ 285 &vhci_cb_ops, /* cb_ops */ 286 NULL, /* bus_ops */ 287 NULL, /* power */ 288 ddi_quiesce_not_needed, /* quiesce */ 289 }; 290 291 extern struct mod_ops mod_driverops; 292 293 static struct modldrv modldrv = { 294 &mod_driverops, 295 vhci_version_name, /* module name */ 296 &vhci_ops 297 }; 298 299 static struct modlinkage modlinkage = { 300 MODREV_1, 301 { &modldrv, NULL } 302 }; 303 304 static mdi_vhci_ops_t vhci_opinfo = { 305 MDI_VHCI_OPS_REV, 306 vhci_pathinfo_init, /* Pathinfo node init callback */ 307 vhci_pathinfo_uninit, /* Pathinfo uninit callback */ 308 vhci_pathinfo_state_change, /* Pathinfo node state change */ 309 vhci_failover, /* failover callback */ 310 vhci_client_attached, /* client attached callback */ 311 vhci_is_dev_supported /* is device supported by mdi */ 312 }; 313 314 /* 315 * The scsi_failover table defines an ordered set of 'fops' modules supported 316 * by scsi_vhci. Currently, initialize this table from the 'ddi-forceload' 317 * property specified in scsi_vhci.conf. 318 */ 319 static struct scsi_failover { 320 ddi_modhandle_t sf_mod; 321 struct scsi_failover_ops *sf_sfo; 322 } *scsi_failover_table; 323 static uint_t scsi_nfailover; 324 325 int 326 _init(void) 327 { 328 int rval; 329 330 /* 331 * Allocate soft state and prepare to do ddi_soft_state_zalloc() 332 * before registering with the transport first. 333 */ 334 if ((rval = ddi_soft_state_init(&vhci_softstate, 335 sizeof (struct scsi_vhci), 1)) != 0) { 336 VHCI_DEBUG(1, (CE_NOTE, NULL, 337 "!_init:soft state init failed\n")); 338 return (rval); 339 } 340 341 if ((rval = scsi_hba_init(&modlinkage)) != 0) { 342 VHCI_DEBUG(1, (CE_NOTE, NULL, 343 "!_init: scsi hba init failed\n")); 344 ddi_soft_state_fini(&vhci_softstate); 345 return (rval); 346 } 347 348 mutex_init(&vhci_global_mutex, NULL, MUTEX_DRIVER, NULL); 349 cv_init(&vhci_cv, NULL, CV_DRIVER, NULL); 350 351 mutex_init(&vhci_targetmap_mutex, NULL, MUTEX_DRIVER, NULL); 352 vhci_targetmap_byport = mod_hash_create_strhash( 353 "vhci_targetmap_byport", 256, mod_hash_null_valdtor); 354 vhci_targetmap_bypid = mod_hash_create_idhash( 355 "vhci_targetmap_bypid", 256, mod_hash_null_valdtor); 356 357 if ((rval = mod_install(&modlinkage)) != 0) { 358 VHCI_DEBUG(1, (CE_NOTE, NULL, "!_init: mod_install failed\n")); 359 if (vhci_targetmap_bypid) 360 mod_hash_destroy_idhash(vhci_targetmap_bypid); 361 if (vhci_targetmap_byport) 362 mod_hash_destroy_strhash(vhci_targetmap_byport); 363 mutex_destroy(&vhci_targetmap_mutex); 364 cv_destroy(&vhci_cv); 365 mutex_destroy(&vhci_global_mutex); 366 scsi_hba_fini(&modlinkage); 367 ddi_soft_state_fini(&vhci_softstate); 368 } 369 return (rval); 370 } 371 372 373 /* 374 * the system is done with us as a driver, so clean up 375 */ 376 int 377 _fini(void) 378 { 379 int rval; 380 381 /* 382 * don't start cleaning up until we know that the module remove 383 * has worked -- if this works, then we know that each instance 384 * has successfully been DDI_DETACHed 385 */ 386 if ((rval = mod_remove(&modlinkage)) != 0) { 387 VHCI_DEBUG(4, (CE_NOTE, NULL, "!_fini: mod_remove failed\n")); 388 return (rval); 389 } 390 391 if (vhci_targetmap_bypid) 392 mod_hash_destroy_idhash(vhci_targetmap_bypid); 393 if (vhci_targetmap_byport) 394 mod_hash_destroy_strhash(vhci_targetmap_byport); 395 mutex_destroy(&vhci_targetmap_mutex); 396 cv_destroy(&vhci_cv); 397 mutex_destroy(&vhci_global_mutex); 398 scsi_hba_fini(&modlinkage); 399 ddi_soft_state_fini(&vhci_softstate); 400 401 return (rval); 402 } 403 404 int 405 _info(struct modinfo *modinfop) 406 { 407 return (mod_info(&modlinkage, modinfop)); 408 } 409 410 /* 411 * Lookup scsi_failover by "short name" of failover module. 412 */ 413 struct scsi_failover_ops * 414 vhci_failover_ops_by_name(char *name) 415 { 416 struct scsi_failover *sf; 417 418 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 419 if (sf->sf_sfo == NULL) 420 continue; 421 if (strcmp(sf->sf_sfo->sfo_name, name) == 0) 422 return (sf->sf_sfo); 423 } 424 return (NULL); 425 } 426 427 /* 428 * Load all scsi_failover_ops 'fops' modules. 429 */ 430 static void 431 vhci_failover_modopen(struct scsi_vhci *vhci) 432 { 433 char **module; 434 int i; 435 struct scsi_failover *sf; 436 char **dt; 437 int e; 438 439 if (scsi_failover_table) 440 return; 441 442 /* Get the list of modules from scsi_vhci.conf */ 443 if (ddi_prop_lookup_string_array(DDI_DEV_T_ANY, 444 vhci->vhci_dip, DDI_PROP_DONTPASS, "ddi-forceload", 445 &module, &scsi_nfailover) != DDI_PROP_SUCCESS) { 446 cmn_err(CE_WARN, "scsi_vhci: " 447 "scsi_vhci.conf is missing 'ddi-forceload'"); 448 return; 449 } 450 if (scsi_nfailover == 0) { 451 cmn_err(CE_WARN, "scsi_vhci: " 452 "scsi_vhci.conf has empty 'ddi-forceload'"); 453 ddi_prop_free(module); 454 return; 455 } 456 457 /* allocate failover table based on number of modules */ 458 scsi_failover_table = (struct scsi_failover *) 459 kmem_zalloc(sizeof (struct scsi_failover) * (scsi_nfailover + 1), 460 KM_SLEEP); 461 462 /* loop over modules specified in scsi_vhci.conf and open each module */ 463 for (i = 0, sf = scsi_failover_table; i < scsi_nfailover; i++) { 464 if (module[i] == NULL) 465 continue; 466 467 sf->sf_mod = ddi_modopen(module[i], KRTLD_MODE_FIRST, &e); 468 if (sf->sf_mod == NULL) { 469 /* 470 * A module returns EEXIST if other software is 471 * supporting the intended function: for example 472 * the scsi_vhci_f_sum_emc module returns EEXIST 473 * from _init if EMC powerpath software is installed. 474 */ 475 if (e != EEXIST) 476 cmn_err(CE_WARN, "scsi_vhci: unable to open " 477 "module '%s', error %d", module[i], e); 478 continue; 479 } 480 sf->sf_sfo = ddi_modsym(sf->sf_mod, 481 "scsi_vhci_failover_ops", &e); 482 if (sf->sf_sfo == NULL) { 483 cmn_err(CE_WARN, "scsi_vhci: " 484 "unable to import 'scsi_failover_ops' from '%s', " 485 "error %d", module[i], e); 486 (void) ddi_modclose(sf->sf_mod); 487 sf->sf_mod = NULL; 488 continue; 489 } 490 491 /* register vid/pid of devices supported with mpapi */ 492 for (dt = sf->sf_sfo->sfo_devices; *dt; dt++) 493 vhci_mpapi_add_dev_prod(vhci, *dt); 494 sf++; 495 } 496 497 /* verify that at least the "well-known" modules were there */ 498 if (vhci_failover_ops_by_name(SFO_NAME_SYM) == NULL) 499 cmn_err(CE_WARN, "scsi_vhci: well-known module \"" 500 SFO_NAME_SYM "\" not defined in scsi_vhci.conf's " 501 "'ddi-forceload'"); 502 if (vhci_failover_ops_by_name(SFO_NAME_TPGS) == NULL) 503 cmn_err(CE_WARN, "scsi_vhci: well-known module \"" 504 SFO_NAME_TPGS "\" not defined in scsi_vhci.conf's " 505 "'ddi-forceload'"); 506 507 /* call sfo_init for modules that need it */ 508 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 509 if (sf->sf_sfo && sf->sf_sfo->sfo_init) 510 sf->sf_sfo->sfo_init(); 511 } 512 513 ddi_prop_free(module); 514 } 515 516 /* 517 * unload all loaded scsi_failover_ops modules 518 */ 519 static void 520 vhci_failover_modclose() 521 { 522 struct scsi_failover *sf; 523 524 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 525 if ((sf->sf_mod == NULL) || (sf->sf_sfo == NULL)) 526 continue; 527 (void) ddi_modclose(sf->sf_mod); 528 sf->sf_mod = NULL; 529 sf->sf_sfo = NULL; 530 } 531 532 if (scsi_failover_table && scsi_nfailover) 533 kmem_free(scsi_failover_table, 534 sizeof (struct scsi_failover) * (scsi_nfailover + 1)); 535 scsi_failover_table = NULL; 536 scsi_nfailover = 0; 537 } 538 539 /* ARGSUSED */ 540 static int 541 vhci_open(dev_t *devp, int flag, int otype, cred_t *credp) 542 { 543 struct scsi_vhci *vhci; 544 545 if (otype != OTYP_CHR) { 546 return (EINVAL); 547 } 548 549 vhci = ddi_get_soft_state(vhci_softstate, MINOR2INST(getminor(*devp))); 550 if (vhci == NULL) { 551 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_open: failed ENXIO\n")); 552 return (ENXIO); 553 } 554 555 mutex_enter(&vhci->vhci_mutex); 556 if ((flag & FEXCL) && (vhci->vhci_state & VHCI_STATE_OPEN)) { 557 mutex_exit(&vhci->vhci_mutex); 558 vhci_log(CE_NOTE, vhci->vhci_dip, 559 "!vhci%d: Already open\n", getminor(*devp)); 560 return (EBUSY); 561 } 562 563 vhci->vhci_state |= VHCI_STATE_OPEN; 564 mutex_exit(&vhci->vhci_mutex); 565 return (0); 566 } 567 568 569 /* ARGSUSED */ 570 static int 571 vhci_close(dev_t dev, int flag, int otype, cred_t *credp) 572 { 573 struct scsi_vhci *vhci; 574 575 if (otype != OTYP_CHR) { 576 return (EINVAL); 577 } 578 579 vhci = ddi_get_soft_state(vhci_softstate, MINOR2INST(getminor(dev))); 580 if (vhci == NULL) { 581 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_close: failed ENXIO\n")); 582 return (ENXIO); 583 } 584 585 mutex_enter(&vhci->vhci_mutex); 586 vhci->vhci_state &= ~VHCI_STATE_OPEN; 587 mutex_exit(&vhci->vhci_mutex); 588 589 return (0); 590 } 591 592 /* ARGSUSED */ 593 static int 594 vhci_ioctl(dev_t dev, int cmd, intptr_t data, int mode, 595 cred_t *credp, int *rval) 596 { 597 if (IS_DEVCTL(cmd)) { 598 return (vhci_devctl(dev, cmd, data, mode, credp, rval)); 599 } else if (cmd == MP_CMD) { 600 return (vhci_mpapi_ctl(dev, cmd, data, mode, credp, rval)); 601 } else { 602 return (vhci_ctl(dev, cmd, data, mode, credp, rval)); 603 } 604 } 605 606 /* 607 * attach the module 608 */ 609 static int 610 vhci_attach(dev_info_t *dip, ddi_attach_cmd_t cmd) 611 { 612 int rval = DDI_FAILURE; 613 int scsi_hba_attached = 0; 614 int vhci_attached = 0; 615 int mutex_initted = 0; 616 int instance; 617 struct scsi_vhci *vhci; 618 scsi_hba_tran_t *tran; 619 char cache_name_buf[64]; 620 char *data; 621 622 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_attach: cmd=0x%x\n", cmd)); 623 624 instance = ddi_get_instance(dip); 625 626 switch (cmd) { 627 case DDI_ATTACH: 628 break; 629 630 case DDI_RESUME: 631 case DDI_PM_RESUME: 632 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_attach: resume not yet" 633 "implemented\n")); 634 return (rval); 635 636 default: 637 VHCI_DEBUG(1, (CE_NOTE, NULL, 638 "!vhci_attach: unknown ddi command\n")); 639 return (rval); 640 } 641 642 /* 643 * Allocate vhci data structure. 644 */ 645 if (ddi_soft_state_zalloc(vhci_softstate, instance) != DDI_SUCCESS) { 646 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 647 "soft state alloc failed\n")); 648 return (DDI_FAILURE); 649 } 650 651 if ((vhci = ddi_get_soft_state(vhci_softstate, instance)) == NULL) { 652 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 653 "bad soft state\n")); 654 ddi_soft_state_free(vhci_softstate, instance); 655 return (DDI_FAILURE); 656 } 657 658 /* Allocate packet cache */ 659 (void) snprintf(cache_name_buf, sizeof (cache_name_buf), 660 "vhci%d_cache", instance); 661 662 mutex_init(&vhci->vhci_mutex, NULL, MUTEX_DRIVER, NULL); 663 mutex_initted++; 664 665 /* 666 * Allocate a transport structure 667 */ 668 tran = scsi_hba_tran_alloc(dip, SCSI_HBA_CANSLEEP); 669 ASSERT(tran != NULL); 670 671 vhci->vhci_tran = tran; 672 vhci->vhci_dip = dip; 673 vhci->vhci_instance = instance; 674 675 tran->tran_hba_private = vhci; 676 tran->tran_tgt_init = vhci_scsi_tgt_init; 677 tran->tran_tgt_probe = NULL; 678 tran->tran_tgt_free = vhci_scsi_tgt_free; 679 680 tran->tran_start = vhci_scsi_start; 681 tran->tran_abort = vhci_scsi_abort; 682 tran->tran_reset = vhci_scsi_reset; 683 tran->tran_getcap = vhci_scsi_getcap; 684 tran->tran_setcap = vhci_scsi_setcap; 685 tran->tran_init_pkt = vhci_scsi_init_pkt; 686 tran->tran_destroy_pkt = vhci_scsi_destroy_pkt; 687 tran->tran_dmafree = vhci_scsi_dmafree; 688 tran->tran_sync_pkt = vhci_scsi_sync_pkt; 689 tran->tran_reset_notify = vhci_scsi_reset_notify; 690 691 tran->tran_get_bus_addr = vhci_scsi_get_bus_addr; 692 tran->tran_get_name = vhci_scsi_get_name; 693 tran->tran_bus_reset = NULL; 694 tran->tran_quiesce = NULL; 695 tran->tran_unquiesce = NULL; 696 697 /* 698 * register event notification routines with scsa 699 */ 700 tran->tran_get_eventcookie = NULL; 701 tran->tran_add_eventcall = NULL; 702 tran->tran_remove_eventcall = NULL; 703 tran->tran_post_event = NULL; 704 705 tran->tran_bus_power = vhci_scsi_bus_power; 706 707 tran->tran_bus_config = vhci_scsi_bus_config; 708 tran->tran_bus_unconfig = vhci_scsi_bus_unconfig; 709 710 /* 711 * Attach this instance with the mpxio framework 712 */ 713 if (mdi_vhci_register(MDI_HCI_CLASS_SCSI, dip, &vhci_opinfo, 0) 714 != MDI_SUCCESS) { 715 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 716 "mdi_vhci_register failed\n")); 717 goto attach_fail; 718 } 719 vhci_attached++; 720 721 /* 722 * Attach this instance of the hba. 723 * 724 * Regarding dma attributes: Since scsi_vhci is a virtual scsi HBA 725 * driver, it has nothing to do with DMA. However, when calling 726 * scsi_hba_attach_setup() we need to pass something valid in the 727 * dma attributes parameter. So we just use scsi_alloc_attr. 728 * SCSA itself seems to care only for dma_attr_minxfer and 729 * dma_attr_burstsizes fields of dma attributes structure. 730 * It expects those fileds to be non-zero. 731 */ 732 if (scsi_hba_attach_setup(dip, &scsi_alloc_attr, tran, 733 SCSI_HBA_ADDR_COMPLEX) != DDI_SUCCESS) { 734 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 735 "hba attach failed\n")); 736 goto attach_fail; 737 } 738 scsi_hba_attached++; 739 740 if (ddi_create_minor_node(dip, "devctl", S_IFCHR, 741 INST2DEVCTL(instance), DDI_NT_SCSI_NEXUS, 0) != DDI_SUCCESS) { 742 VHCI_DEBUG(1, (CE_NOTE, dip, "!vhci_attach:" 743 " ddi_create_minor_node failed\n")); 744 goto attach_fail; 745 } 746 747 /* 748 * Set pm-want-child-notification property for 749 * power management of the phci and client 750 */ 751 if (ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP, 752 "pm-want-child-notification?", NULL, NULL) != DDI_PROP_SUCCESS) { 753 cmn_err(CE_WARN, 754 "%s%d fail to create pm-want-child-notification? prop", 755 ddi_driver_name(dip), ddi_get_instance(dip)); 756 goto attach_fail; 757 } 758 759 vhci->vhci_taskq = taskq_create("vhci_taskq", 1, MINCLSYSPRI, 1, 4, 0); 760 vhci->vhci_update_pathstates_taskq = 761 taskq_create("vhci_update_pathstates", VHCI_NUM_UPDATE_TASKQ, 762 MINCLSYSPRI, 1, 4, 0); 763 ASSERT(vhci->vhci_taskq); 764 ASSERT(vhci->vhci_update_pathstates_taskq); 765 766 /* 767 * Set appropriate configuration flags based on options set in 768 * conf file. 769 */ 770 vhci->vhci_conf_flags = 0; 771 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, dip, PROPFLAGS, 772 "auto-failback", &data) == DDI_SUCCESS) { 773 if (strcmp(data, "enable") == 0) 774 vhci->vhci_conf_flags |= VHCI_CONF_FLAGS_AUTO_FAILBACK; 775 ddi_prop_free(data); 776 } 777 778 if (!(vhci->vhci_conf_flags & VHCI_CONF_FLAGS_AUTO_FAILBACK)) 779 vhci_log(CE_NOTE, dip, "!Auto-failback capability " 780 "disabled through scsi_vhci.conf file."); 781 782 /* 783 * Allocate an mpapi private structure 784 */ 785 vhci->mp_priv = kmem_zalloc(sizeof (mpapi_priv_t), KM_SLEEP); 786 if (vhci_mpapi_init(vhci) != 0) { 787 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_attach: " 788 "vhci_mpapi_init() failed")); 789 } 790 791 vhci_failover_modopen(vhci); /* load failover modules */ 792 793 ddi_report_dev(dip); 794 return (DDI_SUCCESS); 795 796 attach_fail: 797 if (vhci_attached) 798 (void) mdi_vhci_unregister(dip, 0); 799 800 if (scsi_hba_attached) 801 (void) scsi_hba_detach(dip); 802 803 if (vhci->vhci_tran) 804 scsi_hba_tran_free(vhci->vhci_tran); 805 806 if (mutex_initted) { 807 mutex_destroy(&vhci->vhci_mutex); 808 } 809 810 ddi_soft_state_free(vhci_softstate, instance); 811 return (DDI_FAILURE); 812 } 813 814 815 /*ARGSUSED*/ 816 static int 817 vhci_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 818 { 819 int instance = ddi_get_instance(dip); 820 scsi_hba_tran_t *tran; 821 struct scsi_vhci *vhci; 822 823 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_detach: cmd=0x%x\n", cmd)); 824 825 if ((tran = ddi_get_driver_private(dip)) == NULL) 826 return (DDI_FAILURE); 827 828 vhci = TRAN2HBAPRIVATE(tran); 829 if (!vhci) { 830 return (DDI_FAILURE); 831 } 832 833 switch (cmd) { 834 case DDI_DETACH: 835 break; 836 837 case DDI_SUSPEND: 838 case DDI_PM_SUSPEND: 839 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_detach: suspend/pm not yet" 840 "implemented\n")); 841 return (DDI_FAILURE); 842 843 default: 844 VHCI_DEBUG(1, (CE_NOTE, NULL, 845 "!vhci_detach: unknown ddi command\n")); 846 return (DDI_FAILURE); 847 } 848 849 (void) mdi_vhci_unregister(dip, 0); 850 (void) scsi_hba_detach(dip); 851 scsi_hba_tran_free(tran); 852 853 if (ddi_prop_remove(DDI_DEV_T_NONE, dip, 854 "pm-want-child-notification?") != DDI_PROP_SUCCESS) { 855 cmn_err(CE_WARN, 856 "%s%d unable to remove prop pm-want_child_notification?", 857 ddi_driver_name(dip), ddi_get_instance(dip)); 858 } 859 if (vhci_restart_timeid != 0) { 860 (void) untimeout(vhci_restart_timeid); 861 } 862 vhci_restart_timeid = 0; 863 864 mutex_destroy(&vhci->vhci_mutex); 865 vhci->vhci_dip = NULL; 866 vhci->vhci_tran = NULL; 867 taskq_destroy(vhci->vhci_taskq); 868 taskq_destroy(vhci->vhci_update_pathstates_taskq); 869 ddi_remove_minor_node(dip, NULL); 870 ddi_soft_state_free(vhci_softstate, instance); 871 872 vhci_failover_modclose(); /* unload failover modules */ 873 return (DDI_SUCCESS); 874 } 875 876 /* 877 * vhci_getinfo() 878 * Given the device number, return the devinfo pointer or the 879 * instance number. 880 * Note: always succeed DDI_INFO_DEVT2INSTANCE, even before attach. 881 */ 882 883 /*ARGSUSED*/ 884 static int 885 vhci_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result) 886 { 887 struct scsi_vhci *vhcip; 888 int instance = MINOR2INST(getminor((dev_t)arg)); 889 890 switch (cmd) { 891 case DDI_INFO_DEVT2DEVINFO: 892 vhcip = ddi_get_soft_state(vhci_softstate, instance); 893 if (vhcip != NULL) 894 *result = vhcip->vhci_dip; 895 else { 896 *result = NULL; 897 return (DDI_FAILURE); 898 } 899 break; 900 901 case DDI_INFO_DEVT2INSTANCE: 902 *result = (void *)(uintptr_t)instance; 903 break; 904 905 default: 906 return (DDI_FAILURE); 907 } 908 909 return (DDI_SUCCESS); 910 } 911 912 /*ARGSUSED*/ 913 static int 914 vhci_scsi_tgt_init(dev_info_t *hba_dip, dev_info_t *tgt_dip, 915 scsi_hba_tran_t *hba_tran, struct scsi_device *sd) 916 { 917 char *guid; 918 scsi_vhci_lun_t *vlun; 919 struct scsi_vhci *vhci; 920 clock_t from_ticks; 921 mdi_pathinfo_t *pip; 922 int rval; 923 924 ASSERT(hba_dip != NULL); 925 ASSERT(tgt_dip != NULL); 926 927 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, PROPFLAGS, 928 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 929 /* 930 * This must be the .conf node without GUID property. 931 * The node under fp already inserts a delay, so we 932 * just return from here. We rely on this delay to have 933 * all dips be posted to the ndi hotplug thread's newdev 934 * list. This is necessary for the deferred attach 935 * mechanism to work and opens() done soon after boot to 936 * succeed. 937 */ 938 VHCI_DEBUG(4, (CE_WARN, hba_dip, "tgt_init: lun guid " 939 "property failed")); 940 return (DDI_NOT_WELL_FORMED); 941 } 942 943 if (ndi_dev_is_persistent_node(tgt_dip) == 0) { 944 /* 945 * This must be .conf node with the GUID property. We don't 946 * merge property by ndi_merge_node() here because the 947 * devi_addr_buf of .conf node is "" always according the 948 * implementation of vhci_scsi_get_name_bus_addr(). 949 */ 950 ddi_set_name_addr(tgt_dip, NULL); 951 return (DDI_FAILURE); 952 } 953 954 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(hba_dip)); 955 ASSERT(vhci != NULL); 956 957 VHCI_DEBUG(4, (CE_NOTE, hba_dip, 958 "!tgt_init: called for %s (instance %d)\n", 959 ddi_driver_name(tgt_dip), ddi_get_instance(tgt_dip))); 960 961 vlun = vhci_lun_lookup(tgt_dip); 962 963 mutex_enter(&vhci_global_mutex); 964 965 from_ticks = ddi_get_lbolt(); 966 if (vhci_to_ticks == 0) { 967 vhci_to_ticks = from_ticks + 968 drv_usectohz(vhci_init_wait_timeout); 969 } 970 971 #if DEBUG 972 if (vlun) { 973 VHCI_DEBUG(1, (CE_WARN, hba_dip, "tgt_init: " 974 "vhci_scsi_tgt_init: guid %s : found vlun 0x%p " 975 "from_ticks %lx to_ticks %lx", 976 guid, (void *)vlun, from_ticks, vhci_to_ticks)); 977 } else { 978 VHCI_DEBUG(1, (CE_WARN, hba_dip, "tgt_init: " 979 "vhci_scsi_tgt_init: guid %s : vlun not found " 980 "from_ticks %lx to_ticks %lx", guid, from_ticks, 981 vhci_to_ticks)); 982 } 983 #endif 984 985 rval = mdi_select_path(tgt_dip, NULL, 986 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), NULL, &pip); 987 if (rval == MDI_SUCCESS) { 988 mdi_rele_path(pip); 989 } 990 991 /* 992 * Wait for the following conditions : 993 * 1. no vlun available yet 994 * 2. no path established 995 * 3. timer did not expire 996 */ 997 while ((vlun == NULL) || (mdi_client_get_path_count(tgt_dip) == 0) || 998 (rval != MDI_SUCCESS)) { 999 if (vlun && vlun->svl_not_supported) { 1000 VHCI_DEBUG(1, (CE_WARN, hba_dip, "tgt_init: " 1001 "vlun 0x%p lun guid %s not supported!", 1002 (void *)vlun, guid)); 1003 mutex_exit(&vhci_global_mutex); 1004 ddi_prop_free(guid); 1005 return (DDI_NOT_WELL_FORMED); 1006 } 1007 if ((vhci_first_time == 0) && (from_ticks >= vhci_to_ticks)) { 1008 vhci_first_time = 1; 1009 } 1010 if (vhci_first_time == 1) { 1011 VHCI_DEBUG(1, (CE_WARN, hba_dip, "vhci_scsi_tgt_init: " 1012 "no wait for %s. from_tick %lx, to_tick %lx", 1013 guid, from_ticks, vhci_to_ticks)); 1014 mutex_exit(&vhci_global_mutex); 1015 ddi_prop_free(guid); 1016 return (DDI_NOT_WELL_FORMED); 1017 } 1018 1019 if (cv_timedwait(&vhci_cv, 1020 &vhci_global_mutex, vhci_to_ticks) == -1) { 1021 /* Timed out */ 1022 #ifdef DEBUG 1023 if (vlun == NULL) { 1024 VHCI_DEBUG(1, (CE_WARN, hba_dip, 1025 "tgt_init: no vlun for %s!", guid)); 1026 } else if (mdi_client_get_path_count(tgt_dip) == 0) { 1027 VHCI_DEBUG(1, (CE_WARN, hba_dip, 1028 "tgt_init: client path count is " 1029 "zero for %s!", guid)); 1030 } else { 1031 VHCI_DEBUG(1, (CE_WARN, hba_dip, 1032 "tgt_init: client path not " 1033 "available yet for %s!", guid)); 1034 } 1035 #endif /* DEBUG */ 1036 mutex_exit(&vhci_global_mutex); 1037 ddi_prop_free(guid); 1038 return (DDI_NOT_WELL_FORMED); 1039 } 1040 vlun = vhci_lun_lookup(tgt_dip); 1041 rval = mdi_select_path(tgt_dip, NULL, 1042 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 1043 NULL, &pip); 1044 if (rval == MDI_SUCCESS) { 1045 mdi_rele_path(pip); 1046 } 1047 from_ticks = ddi_get_lbolt(); 1048 } 1049 mutex_exit(&vhci_global_mutex); 1050 1051 ASSERT(vlun != NULL); 1052 ddi_prop_free(guid); 1053 1054 scsi_device_hba_private_set(sd, vlun); 1055 1056 return (DDI_SUCCESS); 1057 } 1058 1059 /*ARGSUSED*/ 1060 static void 1061 vhci_scsi_tgt_free(dev_info_t *hba_dip, dev_info_t *tgt_dip, 1062 scsi_hba_tran_t *hba_tran, struct scsi_device *sd) 1063 { 1064 struct scsi_vhci_lun *dvlp; 1065 ASSERT(mdi_client_get_path_count(tgt_dip) <= 0); 1066 dvlp = (struct scsi_vhci_lun *)scsi_device_hba_private_get(sd); 1067 ASSERT(dvlp != NULL); 1068 1069 vhci_lun_free(dvlp, sd); 1070 } 1071 1072 /* 1073 * a PGR register command has started; copy the info we need 1074 */ 1075 int 1076 vhci_pgr_register_start(scsi_vhci_lun_t *vlun, struct scsi_pkt *pkt) 1077 { 1078 struct vhci_pkt *vpkt = TGTPKT2VHCIPKT(pkt); 1079 void *addr; 1080 1081 if (!vpkt->vpkt_tgt_init_bp) 1082 return (TRAN_BADPKT); 1083 1084 addr = bp_mapin_common(vpkt->vpkt_tgt_init_bp, 1085 (vpkt->vpkt_flags & CFLAG_NOWAIT) ? VM_NOSLEEP : VM_SLEEP); 1086 if (addr == NULL) 1087 return (TRAN_BUSY); 1088 1089 mutex_enter(&vlun->svl_mutex); 1090 1091 vhci_print_prout_keys(vlun, "v_pgr_reg_start: before bcopy:"); 1092 1093 bcopy(addr, &vlun->svl_prout, sizeof (vhci_prout_t) - 1094 (2 * MHIOC_RESV_KEY_SIZE*sizeof (char))); 1095 bcopy(pkt->pkt_cdbp, vlun->svl_cdb, sizeof (vlun->svl_cdb)); 1096 1097 vhci_print_prout_keys(vlun, "v_pgr_reg_start: after bcopy:"); 1098 1099 vlun->svl_time = pkt->pkt_time; 1100 vlun->svl_bcount = vpkt->vpkt_tgt_init_bp->b_bcount; 1101 vlun->svl_first_path = vpkt->vpkt_path; 1102 mutex_exit(&vlun->svl_mutex); 1103 return (0); 1104 } 1105 1106 /* 1107 * Function name : vhci_scsi_start() 1108 * 1109 * Return Values : TRAN_FATAL_ERROR - vhci has been shutdown 1110 * or other fatal failure 1111 * preventing packet transportation 1112 * TRAN_BUSY - request queue is full 1113 * TRAN_ACCEPT - pkt has been submitted to phci 1114 * (or is held in the waitQ) 1115 * Description : Implements SCSA's tran_start() entry point for 1116 * packet transport 1117 * 1118 */ 1119 static int 1120 vhci_scsi_start(struct scsi_address *ap, struct scsi_pkt *pkt) 1121 { 1122 int rval = TRAN_ACCEPT; 1123 int instance, held; 1124 struct scsi_vhci *vhci = ADDR2VHCI(ap); 1125 struct scsi_vhci_lun *vlun = ADDR2VLUN(ap); 1126 struct vhci_pkt *vpkt = TGTPKT2VHCIPKT(pkt); 1127 int flags = 0; 1128 scsi_vhci_priv_t *svp, *svp_resrv; 1129 dev_info_t *cdip; 1130 client_lb_t lbp; 1131 int restore_lbp = 0; 1132 /* set if pkt is SCSI-II RESERVE cmd */ 1133 int pkt_reserve_cmd = 0; 1134 int reserve_failed = 0; 1135 int resrv_instance = 0; 1136 mdi_pathinfo_t *pip; 1137 struct scsi_pkt *rel_pkt; 1138 1139 ASSERT(vhci != NULL); 1140 ASSERT(vpkt != NULL); 1141 ASSERT(vpkt->vpkt_state != VHCI_PKT_ISSUED); 1142 cdip = ADDR2DIP(ap); 1143 1144 /* 1145 * Block IOs if LUN is held or QUIESCED for IOs. 1146 */ 1147 if ((VHCI_LUN_IS_HELD(vlun)) || 1148 ((vlun->svl_flags & VLUN_QUIESCED_FLG) == VLUN_QUIESCED_FLG)) { 1149 return (TRAN_BUSY); 1150 } 1151 1152 /* 1153 * vhci_lun needs to be quiesced before SCSI-II RESERVE command 1154 * can be issued. This may require a cv_timedwait, which is 1155 * dangerous to perform in an interrupt context. So if this 1156 * is a RESERVE command a taskq is dispatched to service it. 1157 * This taskq shall again call vhci_scsi_start, but we shall be 1158 * sure its not in an interrupt context. 1159 */ 1160 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 1161 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 1162 if (!(vpkt->vpkt_state & VHCI_PKT_THRU_TASKQ)) { 1163 if (taskq_dispatch(vhci->vhci_taskq, 1164 vhci_dispatch_scsi_start, (void *) vpkt, 1165 KM_NOSLEEP)) { 1166 return (TRAN_ACCEPT); 1167 } else { 1168 return (TRAN_BUSY); 1169 } 1170 } 1171 1172 /* 1173 * Here we ensure that simultaneous SCSI-II RESERVE cmds don't 1174 * get serviced for a lun. 1175 */ 1176 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 1177 if (!held) { 1178 return (TRAN_BUSY); 1179 } else if ((vlun->svl_flags & VLUN_QUIESCED_FLG) == 1180 VLUN_QUIESCED_FLG) { 1181 VHCI_RELEASE_LUN(vlun); 1182 return (TRAN_BUSY); 1183 } 1184 1185 /* 1186 * To ensure that no IOs occur for this LUN for the duration 1187 * of this pkt set the VLUN_QUIESCED_FLG. 1188 * In case this routine needs to exit on error make sure that 1189 * this flag is cleared. 1190 */ 1191 vlun->svl_flags |= VLUN_QUIESCED_FLG; 1192 pkt_reserve_cmd = 1; 1193 1194 /* 1195 * if this is a SCSI-II RESERVE command, set load balancing 1196 * policy to be ALTERNATE PATH to ensure that all subsequent 1197 * IOs are routed on the same path. This is because if commands 1198 * are routed across multiple paths then IOs on paths other than 1199 * the one on which the RESERVE was executed will get a 1200 * RESERVATION CONFLICT 1201 */ 1202 lbp = mdi_get_lb_policy(cdip); 1203 if (lbp != LOAD_BALANCE_NONE) { 1204 if (vhci_quiesce_lun(vlun) != 1) { 1205 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1206 VHCI_RELEASE_LUN(vlun); 1207 return (TRAN_FATAL_ERROR); 1208 } 1209 vlun->svl_lb_policy_save = lbp; 1210 if (mdi_set_lb_policy(cdip, LOAD_BALANCE_NONE) != 1211 MDI_SUCCESS) { 1212 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1213 VHCI_RELEASE_LUN(vlun); 1214 return (TRAN_FATAL_ERROR); 1215 } 1216 restore_lbp = 1; 1217 } 1218 1219 VHCI_DEBUG(2, (CE_NOTE, vhci->vhci_dip, 1220 "!vhci_scsi_start: sending SCSI-2 RESERVE, vlun 0x%p, " 1221 "svl_resrv_pip 0x%p, svl_flags: %x, lb_policy %x", 1222 (void *)vlun, (void *)vlun->svl_resrv_pip, vlun->svl_flags, 1223 mdi_get_lb_policy(cdip))); 1224 1225 /* 1226 * See comments for VLUN_RESERVE_ACTIVE_FLG in scsi_vhci.h 1227 * To narrow this window where a reserve command may be sent 1228 * down an inactive path the path states first need to be 1229 * updated. Before calling vhci_update_pathstates reset 1230 * VLUN_RESERVE_ACTIVE_FLG, just in case it was already set 1231 * for this lun. This shall prevent an unnecessary reset 1232 * from being sent out. Also remember currently reserved path 1233 * just for a case the new reservation will go to another path. 1234 */ 1235 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 1236 resrv_instance = mdi_pi_get_path_instance( 1237 vlun->svl_resrv_pip); 1238 } 1239 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 1240 vhci_update_pathstates((void *)vlun); 1241 } 1242 1243 instance = ddi_get_instance(vhci->vhci_dip); 1244 1245 /* 1246 * If the command is PRIN with action of zero, then the cmd 1247 * is reading PR keys which requires filtering on completion. 1248 * Data cache sync must be guaranteed. 1249 */ 1250 if ((pkt->pkt_cdbp[0] == SCMD_PRIN) && (pkt->pkt_cdbp[1] == 0) && 1251 (vpkt->vpkt_org_vpkt == NULL)) { 1252 vpkt->vpkt_tgt_init_pkt_flags |= PKT_CONSISTENT; 1253 } 1254 1255 /* 1256 * Do not defer bind for PKT_DMA_PARTIAL 1257 */ 1258 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1259 1260 /* This is a non pkt_dma_partial case */ 1261 if ((rval = vhci_bind_transport( 1262 ap, vpkt, vpkt->vpkt_tgt_init_pkt_flags, NULL_FUNC)) 1263 != TRAN_ACCEPT) { 1264 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1265 "!vhci%d %x: failed to bind transport: " 1266 "vlun 0x%p pkt_reserved %x restore_lbp %x," 1267 "lbp %x", instance, rval, (void *)vlun, 1268 pkt_reserve_cmd, restore_lbp, lbp)); 1269 if (restore_lbp) 1270 (void) mdi_set_lb_policy(cdip, lbp); 1271 if (pkt_reserve_cmd) 1272 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1273 return (rval); 1274 } 1275 VHCI_DEBUG(8, (CE_NOTE, NULL, 1276 "vhci_scsi_start: v_b_t called 0x%p\n", (void *)vpkt)); 1277 } 1278 ASSERT(vpkt->vpkt_hba_pkt != NULL); 1279 ASSERT(vpkt->vpkt_path != NULL); 1280 1281 /* 1282 * This is the chance to adjust the pHCI's pkt and other information 1283 * from target driver's pkt. 1284 */ 1285 VHCI_DEBUG(8, (CE_NOTE, vhci->vhci_dip, "vhci_scsi_start vpkt %p\n", 1286 (void *)vpkt)); 1287 vhci_update_pHCI_pkt(vpkt, pkt); 1288 1289 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 1290 if (vpkt->vpkt_path != vlun->svl_resrv_pip) { 1291 VHCI_DEBUG(1, (CE_WARN, vhci->vhci_dip, 1292 "!vhci_bind: reserve flag set for vlun 0x%p, but, " 1293 "pktpath 0x%p resrv path 0x%p differ. lb_policy %x", 1294 (void *)vlun, (void *)vpkt->vpkt_path, 1295 (void *)vlun->svl_resrv_pip, 1296 mdi_get_lb_policy(cdip))); 1297 reserve_failed = 1; 1298 } 1299 } 1300 1301 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(vpkt->vpkt_path); 1302 if (svp == NULL || reserve_failed) { 1303 if (pkt_reserve_cmd) { 1304 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1305 "!vhci_bind returned null svp vlun 0x%p", 1306 (void *)vlun)); 1307 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1308 if (restore_lbp) 1309 (void) mdi_set_lb_policy(cdip, lbp); 1310 } 1311 pkt_cleanup: 1312 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1313 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1314 vpkt->vpkt_hba_pkt = NULL; 1315 if (vpkt->vpkt_path) { 1316 mdi_rele_path(vpkt->vpkt_path); 1317 vpkt->vpkt_path = NULL; 1318 } 1319 } 1320 if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 1321 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 1322 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE))) { 1323 sema_v(&vlun->svl_pgr_sema); 1324 } 1325 return (TRAN_BUSY); 1326 } 1327 1328 if ((resrv_instance != 0) && (resrv_instance != 1329 mdi_pi_get_path_instance(vpkt->vpkt_path))) { 1330 /* 1331 * This is an attempt to reserve vpkt->vpkt_path. But the 1332 * previously reserved path referred by resrv_instance might 1333 * still be reserved. Hence we will send a release command 1334 * there in order to avoid a reservation conflict. 1335 */ 1336 VHCI_DEBUG(1, (CE_NOTE, vhci->vhci_dip, "!vhci_scsi_start: " 1337 "conflicting reservation on another path, vlun 0x%p, " 1338 "reserved instance %d, new instance: %d, pip: 0x%p", 1339 (void *)vlun, resrv_instance, 1340 mdi_pi_get_path_instance(vpkt->vpkt_path), 1341 (void *)vpkt->vpkt_path)); 1342 1343 /* 1344 * In rare cases, the path referred by resrv_instance could 1345 * disappear in the meantime. Calling mdi_select_path() below 1346 * is an attempt to find out if the path still exists. It also 1347 * ensures that the path will be held when the release is sent. 1348 */ 1349 rval = mdi_select_path(cdip, NULL, MDI_SELECT_PATH_INSTANCE, 1350 (void *)(intptr_t)resrv_instance, &pip); 1351 1352 if ((rval == MDI_SUCCESS) && (pip != NULL)) { 1353 svp_resrv = (scsi_vhci_priv_t *) 1354 mdi_pi_get_vhci_private(pip); 1355 rel_pkt = scsi_init_pkt(&svp_resrv->svp_psd->sd_address, 1356 NULL, NULL, CDB_GROUP0, 1357 sizeof (struct scsi_arq_status), 0, 0, SLEEP_FUNC, 1358 NULL); 1359 1360 if (rel_pkt == NULL) { 1361 char *p_path; 1362 1363 /* 1364 * This is very unlikely. 1365 * scsi_init_pkt(SLEEP_FUNC) does not fail 1366 * because of resources. But in theory it could 1367 * fail for some other reason. There is not an 1368 * easy way how to recover though. Log a warning 1369 * and return. 1370 */ 1371 p_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 1372 vhci_log(CE_WARN, vhci->vhci_dip, "!Sending " 1373 "RELEASE(6) to %s failed, a potential " 1374 "reservation conflict ahead.", 1375 ddi_pathname(mdi_pi_get_phci(pip), p_path)); 1376 kmem_free(p_path, MAXPATHLEN); 1377 1378 if (restore_lbp) 1379 (void) mdi_set_lb_policy(cdip, lbp); 1380 1381 /* no need to check pkt_reserve_cmd here */ 1382 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1383 return (TRAN_FATAL_ERROR); 1384 } 1385 1386 rel_pkt->pkt_cdbp[0] = SCMD_RELEASE; 1387 rel_pkt->pkt_time = 60; 1388 1389 /* 1390 * Ignore the return value. If it will fail 1391 * then most likely it is no longer reserved 1392 * anyway. 1393 */ 1394 (void) vhci_do_scsi_cmd(rel_pkt); 1395 VHCI_DEBUG(1, (CE_NOTE, NULL, 1396 "!vhci_scsi_start: path 0x%p, issued SCSI-2" 1397 " RELEASE\n", (void *)pip)); 1398 scsi_destroy_pkt(rel_pkt); 1399 mdi_rele_path(pip); 1400 } 1401 } 1402 1403 VHCI_INCR_PATH_CMDCOUNT(svp); 1404 1405 /* 1406 * Ensure that no other IOs raced ahead, while a RESERVE cmd was 1407 * QUIESCING the same lun. 1408 */ 1409 if ((!pkt_reserve_cmd) && 1410 ((vlun->svl_flags & VLUN_QUIESCED_FLG) == VLUN_QUIESCED_FLG)) { 1411 VHCI_DECR_PATH_CMDCOUNT(svp); 1412 goto pkt_cleanup; 1413 } 1414 1415 if ((pkt->pkt_cdbp[0] == SCMD_PRIN) || 1416 (pkt->pkt_cdbp[0] == SCMD_PROUT)) { 1417 /* 1418 * currently this thread only handles running PGR 1419 * commands, so don't bother creating it unless 1420 * something interesting is going to happen (like 1421 * either a PGR out, or a PGR in with enough space 1422 * to hold the keys that are getting returned) 1423 */ 1424 mutex_enter(&vlun->svl_mutex); 1425 if (((vlun->svl_flags & VLUN_TASK_D_ALIVE_FLG) == 0) && 1426 (pkt->pkt_cdbp[0] == SCMD_PROUT)) { 1427 vlun->svl_taskq = taskq_create("vlun_pgr_task_daemon", 1428 1, MINCLSYSPRI, 1, 4, 0); 1429 vlun->svl_flags |= VLUN_TASK_D_ALIVE_FLG; 1430 } 1431 mutex_exit(&vlun->svl_mutex); 1432 if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 1433 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 1434 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE))) { 1435 if (rval = vhci_pgr_register_start(vlun, pkt)) { 1436 /* an error */ 1437 sema_v(&vlun->svl_pgr_sema); 1438 return (rval); 1439 } 1440 } 1441 } 1442 1443 /* 1444 * SCSI-II RESERVE cmd is not expected in polled mode. 1445 * If this changes it needs to be handled for the polled scenario. 1446 */ 1447 flags = vpkt->vpkt_hba_pkt->pkt_flags; 1448 1449 /* 1450 * Set the path_instance *before* sending the scsi_pkt down the path 1451 * to mpxio's pHCI so that additional path abstractions at a pHCI 1452 * level (like maybe iSCSI at some point in the future) can update 1453 * the path_instance. 1454 */ 1455 if (scsi_pkt_allocated_correctly(vpkt->vpkt_hba_pkt)) 1456 vpkt->vpkt_hba_pkt->pkt_path_instance = 1457 mdi_pi_get_path_instance(vpkt->vpkt_path); 1458 1459 rval = scsi_transport(vpkt->vpkt_hba_pkt); 1460 if (rval == TRAN_ACCEPT) { 1461 if (flags & FLAG_NOINTR) { 1462 struct scsi_pkt *tpkt = vpkt->vpkt_tgt_pkt; 1463 struct scsi_pkt *pkt = vpkt->vpkt_hba_pkt; 1464 1465 ASSERT(tpkt != NULL); 1466 *(tpkt->pkt_scbp) = *(pkt->pkt_scbp); 1467 tpkt->pkt_resid = pkt->pkt_resid; 1468 tpkt->pkt_state = pkt->pkt_state; 1469 tpkt->pkt_statistics = pkt->pkt_statistics; 1470 tpkt->pkt_reason = pkt->pkt_reason; 1471 1472 if ((*(pkt->pkt_scbp) == STATUS_CHECK) && 1473 (pkt->pkt_state & STATE_ARQ_DONE)) { 1474 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 1475 vpkt->vpkt_tgt_init_scblen); 1476 } 1477 1478 VHCI_DECR_PATH_CMDCOUNT(svp); 1479 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1480 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1481 vpkt->vpkt_hba_pkt = NULL; 1482 if (vpkt->vpkt_path) { 1483 mdi_rele_path(vpkt->vpkt_path); 1484 vpkt->vpkt_path = NULL; 1485 } 1486 } 1487 /* 1488 * This path will not automatically retry pkts 1489 * internally, therefore, vpkt_org_vpkt should 1490 * never be set. 1491 */ 1492 ASSERT(vpkt->vpkt_org_vpkt == NULL); 1493 scsi_hba_pkt_comp(tpkt); 1494 } 1495 return (rval); 1496 } else if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 1497 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 1498 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE))) { 1499 /* the command exited with bad status */ 1500 sema_v(&vlun->svl_pgr_sema); 1501 } else if (vpkt->vpkt_tgt_pkt->pkt_cdbp[0] == SCMD_PRIN) { 1502 /* the command exited with bad status */ 1503 sema_v(&vlun->svl_pgr_sema); 1504 } else if (pkt_reserve_cmd) { 1505 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1506 "!vhci_scsi_start: reserve failed vlun 0x%p", 1507 (void *)vlun)); 1508 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 1509 if (restore_lbp) 1510 (void) mdi_set_lb_policy(cdip, lbp); 1511 } 1512 1513 ASSERT(vpkt->vpkt_hba_pkt != NULL); 1514 VHCI_DECR_PATH_CMDCOUNT(svp); 1515 1516 /* Do not destroy phci packet information for PKT_DMA_PARTIAL */ 1517 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 1518 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 1519 vpkt->vpkt_hba_pkt = NULL; 1520 if (vpkt->vpkt_path) { 1521 MDI_PI_ERRSTAT(vpkt->vpkt_path, MDI_PI_TRANSERR); 1522 mdi_rele_path(vpkt->vpkt_path); 1523 vpkt->vpkt_path = NULL; 1524 } 1525 } 1526 return (TRAN_BUSY); 1527 } 1528 1529 /* 1530 * Function name : vhci_scsi_reset() 1531 * 1532 * Return Values : 0 - reset failed 1533 * 1 - reset succeeded 1534 */ 1535 1536 /* ARGSUSED */ 1537 static int 1538 vhci_scsi_reset(struct scsi_address *ap, int level) 1539 { 1540 int rval = 0; 1541 1542 cmn_err(CE_WARN, "!vhci_scsi_reset 0x%x", level); 1543 if ((level == RESET_TARGET) || (level == RESET_LUN)) { 1544 return (vhci_scsi_reset_target(ap, level, TRUE)); 1545 } else if (level == RESET_ALL) { 1546 return (vhci_scsi_reset_bus(ap)); 1547 } 1548 1549 return (rval); 1550 } 1551 1552 /* 1553 * vhci_recovery_reset: 1554 * Issues reset to the device 1555 * Input: 1556 * vlun - vhci lun pointer of the device 1557 * ap - address of the device 1558 * select_path: 1559 * If select_path is FALSE, then the address specified in ap is 1560 * the path on which reset will be issued. 1561 * If select_path is TRUE, then path is obtained by calling 1562 * mdi_select_path. 1563 * 1564 * recovery_depth: 1565 * Caller can specify the level of reset. 1566 * VHCI_DEPTH_LUN - 1567 * Issues LUN RESET if device supports lun reset. 1568 * VHCI_DEPTH_TARGET - 1569 * If Lun Reset fails or the device does not support 1570 * Lun Reset, issues TARGET RESET 1571 * VHCI_DEPTH_ALL - 1572 * If Lun Reset fails or the device does not support 1573 * Lun Reset, issues TARGET RESET. 1574 * If TARGET RESET does not succeed, issues Bus Reset. 1575 */ 1576 1577 static int 1578 vhci_recovery_reset(scsi_vhci_lun_t *vlun, struct scsi_address *ap, 1579 uint8_t select_path, uint8_t recovery_depth) 1580 { 1581 int ret = 0; 1582 1583 ASSERT(ap != NULL); 1584 1585 if (vlun && vlun->svl_support_lun_reset == 1) { 1586 ret = vhci_scsi_reset_target(ap, RESET_LUN, 1587 select_path); 1588 } 1589 1590 recovery_depth--; 1591 1592 if ((ret == 0) && recovery_depth) { 1593 ret = vhci_scsi_reset_target(ap, RESET_TARGET, 1594 select_path); 1595 recovery_depth--; 1596 } 1597 1598 if ((ret == 0) && recovery_depth) { 1599 (void) scsi_reset(ap, RESET_ALL); 1600 } 1601 1602 return (ret); 1603 } 1604 1605 /* 1606 * Note: The scsi_address passed to this routine could be the scsi_address 1607 * for the virtual device or the physical device. No assumptions should be 1608 * made in this routine about the contents of the ap structure. 1609 * Further, note that the child dip would be the dip of the ssd node regardless 1610 * of the scsi_address passed in. 1611 */ 1612 static int 1613 vhci_scsi_reset_target(struct scsi_address *ap, int level, uint8_t select_path) 1614 { 1615 dev_info_t *vdip, *cdip; 1616 mdi_pathinfo_t *pip = NULL; 1617 mdi_pathinfo_t *npip = NULL; 1618 int rval = -1; 1619 scsi_vhci_priv_t *svp = NULL; 1620 struct scsi_address *pap = NULL; 1621 scsi_hba_tran_t *hba = NULL; 1622 int sps; 1623 struct scsi_vhci *vhci = NULL; 1624 1625 if (select_path != TRUE) { 1626 ASSERT(ap != NULL); 1627 if (level == RESET_LUN) { 1628 hba = ap->a_hba_tran; 1629 ASSERT(hba != NULL); 1630 return (hba->tran_reset(ap, RESET_LUN)); 1631 } 1632 return (scsi_reset(ap, level)); 1633 } 1634 1635 cdip = ADDR2DIP(ap); 1636 ASSERT(cdip != NULL); 1637 vdip = ddi_get_parent(cdip); 1638 ASSERT(vdip != NULL); 1639 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 1640 ASSERT(vhci != NULL); 1641 1642 rval = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH, NULL, &pip); 1643 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 1644 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_scsi_reset_target: " 1645 "Unable to get a path, dip 0x%p", (void *)cdip)); 1646 return (0); 1647 } 1648 again: 1649 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 1650 if (svp == NULL) { 1651 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_scsi_reset_target: " 1652 "priv is NULL, pip 0x%p", (void *)pip)); 1653 mdi_rele_path(pip); 1654 return (0); 1655 } 1656 1657 if (svp->svp_psd == NULL) { 1658 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_scsi_reset_target: " 1659 "psd is NULL, pip 0x%p, svp 0x%p", 1660 (void *)pip, (void *)svp)); 1661 mdi_rele_path(pip); 1662 return (0); 1663 } 1664 1665 pap = &svp->svp_psd->sd_address; 1666 hba = pap->a_hba_tran; 1667 1668 ASSERT(pap != NULL); 1669 ASSERT(hba != NULL); 1670 1671 if (hba->tran_reset != NULL) { 1672 if (hba->tran_reset(pap, level) == 0) { 1673 vhci_log(CE_WARN, vdip, "!%s%d: " 1674 "path %s, reset %d failed", 1675 ddi_driver_name(cdip), ddi_get_instance(cdip), 1676 mdi_pi_spathname(pip), level); 1677 1678 /* 1679 * Select next path and issue the reset, repeat 1680 * until all paths are exhausted 1681 */ 1682 sps = mdi_select_path(cdip, NULL, 1683 MDI_SELECT_ONLINE_PATH, pip, &npip); 1684 if ((sps != MDI_SUCCESS) || (npip == NULL)) { 1685 mdi_rele_path(pip); 1686 return (0); 1687 } 1688 mdi_rele_path(pip); 1689 pip = npip; 1690 goto again; 1691 } 1692 mdi_rele_path(pip); 1693 mutex_enter(&vhci->vhci_mutex); 1694 scsi_hba_reset_notify_callback(&vhci->vhci_mutex, 1695 &vhci->vhci_reset_notify_listf); 1696 mutex_exit(&vhci->vhci_mutex); 1697 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_scsi_reset_target: " 1698 "reset %d sent down pip:%p for cdip:%p\n", level, 1699 (void *)pip, (void *)cdip)); 1700 return (1); 1701 } 1702 mdi_rele_path(pip); 1703 return (0); 1704 } 1705 1706 1707 /* ARGSUSED */ 1708 static int 1709 vhci_scsi_reset_bus(struct scsi_address *ap) 1710 { 1711 return (1); 1712 } 1713 1714 1715 /* 1716 * called by vhci_getcap and vhci_setcap to get and set (respectively) 1717 * SCSI capabilities 1718 */ 1719 /* ARGSUSED */ 1720 static int 1721 vhci_commoncap(struct scsi_address *ap, char *cap, 1722 int val, int tgtonly, int doset) 1723 { 1724 struct scsi_vhci *vhci = ADDR2VHCI(ap); 1725 struct scsi_vhci_lun *vlun = ADDR2VLUN(ap); 1726 int cidx; 1727 int rval = 0; 1728 1729 if (cap == (char *)0) { 1730 VHCI_DEBUG(3, (CE_WARN, vhci->vhci_dip, 1731 "!vhci_commoncap: invalid arg")); 1732 return (rval); 1733 } 1734 1735 if (vlun == NULL) { 1736 VHCI_DEBUG(3, (CE_WARN, vhci->vhci_dip, 1737 "!vhci_commoncap: vlun is null")); 1738 return (rval); 1739 } 1740 1741 if ((cidx = scsi_hba_lookup_capstr(cap)) == -1) { 1742 return (UNDEFINED); 1743 } 1744 1745 /* 1746 * Process setcap request. 1747 */ 1748 if (doset) { 1749 /* 1750 * At present, we can only set binary (0/1) values 1751 */ 1752 switch (cidx) { 1753 case SCSI_CAP_ARQ: 1754 if (val == 0) { 1755 rval = 0; 1756 } else { 1757 rval = 1; 1758 } 1759 break; 1760 1761 case SCSI_CAP_LUN_RESET: 1762 if (tgtonly == 0) { 1763 VHCI_DEBUG(1, (CE_WARN, vhci->vhci_dip, 1764 "scsi_vhci_setcap: " 1765 "Returning error since whom = 0")); 1766 rval = -1; 1767 break; 1768 } 1769 /* 1770 * Set the capability accordingly. 1771 */ 1772 mutex_enter(&vlun->svl_mutex); 1773 vlun->svl_support_lun_reset = val; 1774 rval = val; 1775 mutex_exit(&vlun->svl_mutex); 1776 break; 1777 1778 case SCSI_CAP_SECTOR_SIZE: 1779 mutex_enter(&vlun->svl_mutex); 1780 vlun->svl_sector_size = val; 1781 vlun->svl_setcap_done = 1; 1782 mutex_exit(&vlun->svl_mutex); 1783 (void) vhci_pHCI_cap(ap, cap, val, tgtonly, NULL); 1784 1785 /* Always return success */ 1786 rval = 1; 1787 break; 1788 1789 default: 1790 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1791 "!vhci_setcap: unsupported %d", cidx)); 1792 rval = UNDEFINED; 1793 break; 1794 } 1795 1796 VHCI_DEBUG(6, (CE_NOTE, vhci->vhci_dip, 1797 "!set cap: cap=%s, val/tgtonly/doset/rval = " 1798 "0x%x/0x%x/0x%x/%d\n", 1799 cap, val, tgtonly, doset, rval)); 1800 1801 } else { 1802 /* 1803 * Process getcap request. 1804 */ 1805 switch (cidx) { 1806 case SCSI_CAP_DMA_MAX: 1807 /* 1808 * For X86 this capability is caught in scsi_ifgetcap(). 1809 * XXX Should this be getting the value from the pHCI? 1810 */ 1811 rval = (int)VHCI_DMA_MAX_XFER_CAP; 1812 break; 1813 1814 case SCSI_CAP_INITIATOR_ID: 1815 rval = 0x00; 1816 break; 1817 1818 case SCSI_CAP_ARQ: 1819 case SCSI_CAP_RESET_NOTIFICATION: 1820 case SCSI_CAP_TAGGED_QING: 1821 rval = 1; 1822 break; 1823 1824 case SCSI_CAP_SCSI_VERSION: 1825 rval = 3; 1826 break; 1827 1828 case SCSI_CAP_INTERCONNECT_TYPE: 1829 rval = INTERCONNECT_FABRIC; 1830 break; 1831 1832 case SCSI_CAP_LUN_RESET: 1833 /* 1834 * scsi_vhci will always return success for LUN reset. 1835 * When request for doing LUN reset comes 1836 * through scsi_reset entry point, at that time attempt 1837 * will be made to do reset through all the possible 1838 * paths. 1839 */ 1840 mutex_enter(&vlun->svl_mutex); 1841 rval = vlun->svl_support_lun_reset; 1842 mutex_exit(&vlun->svl_mutex); 1843 VHCI_DEBUG(4, (CE_WARN, vhci->vhci_dip, 1844 "scsi_vhci_getcap:" 1845 "Getting the Lun reset capability %d", rval)); 1846 break; 1847 1848 case SCSI_CAP_SECTOR_SIZE: 1849 mutex_enter(&vlun->svl_mutex); 1850 rval = vlun->svl_sector_size; 1851 mutex_exit(&vlun->svl_mutex); 1852 break; 1853 1854 case SCSI_CAP_CDB_LEN: 1855 rval = VHCI_SCSI_CDB_SIZE; 1856 break; 1857 1858 case SCSI_CAP_DMA_MAX_ARCH: 1859 /* 1860 * For X86 this capability is caught in scsi_ifgetcap(). 1861 * XXX Should this be getting the value from the pHCI? 1862 */ 1863 rval = 0; 1864 break; 1865 1866 default: 1867 VHCI_DEBUG(6, (CE_WARN, vhci->vhci_dip, 1868 "!vhci_getcap: unsupported %d", cidx)); 1869 rval = UNDEFINED; 1870 break; 1871 } 1872 1873 VHCI_DEBUG(6, (CE_NOTE, vhci->vhci_dip, 1874 "!get cap: cap=%s, val/tgtonly/doset/rval = " 1875 "0x%x/0x%x/0x%x/%d\n", 1876 cap, val, tgtonly, doset, rval)); 1877 } 1878 return (rval); 1879 } 1880 1881 1882 /* 1883 * Function name : vhci_scsi_getcap() 1884 * 1885 */ 1886 static int 1887 vhci_scsi_getcap(struct scsi_address *ap, char *cap, int whom) 1888 { 1889 return (vhci_commoncap(ap, cap, 0, whom, 0)); 1890 } 1891 1892 static int 1893 vhci_scsi_setcap(struct scsi_address *ap, char *cap, int value, int whom) 1894 { 1895 return (vhci_commoncap(ap, cap, value, whom, 1)); 1896 } 1897 1898 /* 1899 * Function name : vhci_scsi_abort() 1900 */ 1901 /* ARGSUSED */ 1902 static int 1903 vhci_scsi_abort(struct scsi_address *ap, struct scsi_pkt *pkt) 1904 { 1905 return (0); 1906 } 1907 1908 /* 1909 * Function name : vhci_scsi_init_pkt 1910 * 1911 * Return Values : pointer to scsi_pkt, or NULL 1912 */ 1913 /* ARGSUSED */ 1914 static struct scsi_pkt * 1915 vhci_scsi_init_pkt(struct scsi_address *ap, struct scsi_pkt *pkt, 1916 struct buf *bp, int cmdlen, int statuslen, int tgtlen, 1917 int flags, int (*callback)(caddr_t), caddr_t arg) 1918 { 1919 struct scsi_vhci *vhci = ADDR2VHCI(ap); 1920 struct vhci_pkt *vpkt; 1921 int rval; 1922 int newpkt = 0; 1923 struct scsi_pkt *pktp; 1924 1925 1926 if (pkt == NULL) { 1927 if (cmdlen > VHCI_SCSI_CDB_SIZE) { 1928 if ((cmdlen != VHCI_SCSI_OSD_CDB_SIZE) || 1929 ((flags & VHCI_SCSI_OSD_PKT_FLAGS) != 1930 VHCI_SCSI_OSD_PKT_FLAGS)) { 1931 VHCI_DEBUG(1, (CE_NOTE, NULL, 1932 "!init pkt: cdb size not supported\n")); 1933 return (NULL); 1934 } 1935 } 1936 1937 pktp = scsi_hba_pkt_alloc(vhci->vhci_dip, 1938 ap, cmdlen, statuslen, tgtlen, sizeof (*vpkt), callback, 1939 arg); 1940 1941 if (pktp == NULL) { 1942 return (NULL); 1943 } 1944 1945 /* Get the vhci's private structure */ 1946 vpkt = (struct vhci_pkt *)(pktp->pkt_ha_private); 1947 ASSERT(vpkt); 1948 1949 /* Save the target driver's packet */ 1950 vpkt->vpkt_tgt_pkt = pktp; 1951 1952 /* 1953 * Save pkt_tgt_init_pkt fields if deferred binding 1954 * is needed or for other purposes. 1955 */ 1956 vpkt->vpkt_tgt_init_pkt_flags = flags; 1957 vpkt->vpkt_flags = (callback == NULL_FUNC) ? CFLAG_NOWAIT : 0; 1958 vpkt->vpkt_state = VHCI_PKT_IDLE; 1959 vpkt->vpkt_tgt_init_cdblen = cmdlen; 1960 vpkt->vpkt_tgt_init_scblen = statuslen; 1961 newpkt = 1; 1962 } else { /* pkt not NULL */ 1963 vpkt = pkt->pkt_ha_private; 1964 } 1965 1966 VHCI_DEBUG(8, (CE_NOTE, NULL, "vhci_scsi_init_pkt " 1967 "vpkt %p flags %x\n", (void *)vpkt, flags)); 1968 1969 /* Clear any stale error flags */ 1970 if (bp) { 1971 bioerror(bp, 0); 1972 } 1973 1974 vpkt->vpkt_tgt_init_bp = bp; 1975 1976 if (flags & PKT_DMA_PARTIAL) { 1977 1978 /* 1979 * Immediate binding is needed. 1980 * Target driver may not set this flag in next invocation. 1981 * vhci has to remember this flag was set during first 1982 * invocation of vhci_scsi_init_pkt. 1983 */ 1984 vpkt->vpkt_flags |= CFLAG_DMA_PARTIAL; 1985 } 1986 1987 if (vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) { 1988 1989 /* 1990 * Re-initialize some of the target driver packet state 1991 * information. 1992 */ 1993 vpkt->vpkt_tgt_pkt->pkt_state = 0; 1994 vpkt->vpkt_tgt_pkt->pkt_statistics = 0; 1995 vpkt->vpkt_tgt_pkt->pkt_reason = 0; 1996 1997 /* 1998 * Binding a vpkt->vpkt_path for this IO at init_time. 1999 * If an IO error happens later, target driver will clear 2000 * this vpkt->vpkt_path binding before re-init IO again. 2001 */ 2002 VHCI_DEBUG(8, (CE_NOTE, NULL, 2003 "vhci_scsi_init_pkt: calling v_b_t %p, newpkt %d\n", 2004 (void *)vpkt, newpkt)); 2005 if (pkt && vpkt->vpkt_hba_pkt) { 2006 VHCI_DEBUG(4, (CE_NOTE, NULL, 2007 "v_s_i_p calling update_pHCI_pkt resid %ld\n", 2008 pkt->pkt_resid)); 2009 vhci_update_pHCI_pkt(vpkt, pkt); 2010 } 2011 if (callback == SLEEP_FUNC) { 2012 rval = vhci_bind_transport( 2013 ap, vpkt, flags, callback); 2014 } else { 2015 rval = vhci_bind_transport( 2016 ap, vpkt, flags, NULL_FUNC); 2017 } 2018 VHCI_DEBUG(8, (CE_NOTE, NULL, 2019 "vhci_scsi_init_pkt: v_b_t called 0x%p rval 0x%x\n", 2020 (void *)vpkt, rval)); 2021 if (bp) { 2022 if (rval == TRAN_FATAL_ERROR) { 2023 /* 2024 * No paths available. Could not bind 2025 * any pHCI. Setting EFAULT as a way 2026 * to indicate no DMA is mapped. 2027 */ 2028 bioerror(bp, EFAULT); 2029 } else { 2030 /* 2031 * Do not indicate any pHCI errors to 2032 * target driver otherwise. 2033 */ 2034 bioerror(bp, 0); 2035 } 2036 } 2037 if (rval != TRAN_ACCEPT) { 2038 VHCI_DEBUG(8, (CE_NOTE, NULL, 2039 "vhci_scsi_init_pkt: " 2040 "v_b_t failed 0x%p newpkt %x\n", 2041 (void *)vpkt, newpkt)); 2042 if (newpkt) { 2043 scsi_hba_pkt_free(ap, 2044 vpkt->vpkt_tgt_pkt); 2045 } 2046 return (NULL); 2047 } 2048 ASSERT(vpkt->vpkt_hba_pkt != NULL); 2049 ASSERT(vpkt->vpkt_path != NULL); 2050 2051 /* Update the resid for the target driver */ 2052 vpkt->vpkt_tgt_pkt->pkt_resid = 2053 vpkt->vpkt_hba_pkt->pkt_resid; 2054 } 2055 2056 return (vpkt->vpkt_tgt_pkt); 2057 } 2058 2059 /* 2060 * Function name : vhci_scsi_destroy_pkt 2061 * 2062 * Return Values : none 2063 */ 2064 static void 2065 vhci_scsi_destroy_pkt(struct scsi_address *ap, struct scsi_pkt *pkt) 2066 { 2067 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_ha_private; 2068 2069 VHCI_DEBUG(8, (CE_NOTE, NULL, 2070 "vhci_scsi_destroy_pkt: vpkt 0x%p\n", (void *)vpkt)); 2071 2072 vpkt->vpkt_tgt_init_pkt_flags = 0; 2073 if (vpkt->vpkt_hba_pkt) { 2074 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 2075 vpkt->vpkt_hba_pkt = NULL; 2076 } 2077 if (vpkt->vpkt_path) { 2078 mdi_rele_path(vpkt->vpkt_path); 2079 vpkt->vpkt_path = NULL; 2080 } 2081 2082 ASSERT(vpkt->vpkt_state != VHCI_PKT_ISSUED); 2083 scsi_hba_pkt_free(ap, vpkt->vpkt_tgt_pkt); 2084 } 2085 2086 /* 2087 * Function name : vhci_scsi_dmafree() 2088 * 2089 * Return Values : none 2090 */ 2091 /*ARGSUSED*/ 2092 static void 2093 vhci_scsi_dmafree(struct scsi_address *ap, struct scsi_pkt *pkt) 2094 { 2095 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_ha_private; 2096 2097 VHCI_DEBUG(6, (CE_NOTE, NULL, 2098 "vhci_scsi_dmafree: vpkt 0x%p\n", (void *)vpkt)); 2099 2100 ASSERT(vpkt != NULL); 2101 if (vpkt->vpkt_hba_pkt) { 2102 scsi_destroy_pkt(vpkt->vpkt_hba_pkt); 2103 vpkt->vpkt_hba_pkt = NULL; 2104 } 2105 if (vpkt->vpkt_path) { 2106 mdi_rele_path(vpkt->vpkt_path); 2107 vpkt->vpkt_path = NULL; 2108 } 2109 } 2110 2111 /* 2112 * Function name : vhci_scsi_sync_pkt() 2113 * 2114 * Return Values : none 2115 */ 2116 /*ARGSUSED*/ 2117 static void 2118 vhci_scsi_sync_pkt(struct scsi_address *ap, struct scsi_pkt *pkt) 2119 { 2120 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_ha_private; 2121 2122 ASSERT(vpkt != NULL); 2123 if (vpkt->vpkt_hba_pkt) { 2124 scsi_sync_pkt(vpkt->vpkt_hba_pkt); 2125 } 2126 } 2127 2128 /* 2129 * routine for reset notification setup, to register or cancel. 2130 */ 2131 static int 2132 vhci_scsi_reset_notify(struct scsi_address *ap, int flag, 2133 void (*callback)(caddr_t), caddr_t arg) 2134 { 2135 struct scsi_vhci *vhci = ADDR2VHCI(ap); 2136 return (scsi_hba_reset_notify_setup(ap, flag, callback, arg, 2137 &vhci->vhci_mutex, &vhci->vhci_reset_notify_listf)); 2138 } 2139 2140 static int 2141 vhci_scsi_get_name_bus_addr(struct scsi_device *sd, 2142 char *name, int len, int bus_addr) 2143 { 2144 dev_info_t *cdip; 2145 char *guid; 2146 scsi_vhci_lun_t *vlun; 2147 2148 ASSERT(sd != NULL); 2149 ASSERT(name != NULL); 2150 2151 *name = 0; 2152 cdip = sd->sd_dev; 2153 2154 ASSERT(cdip != NULL); 2155 2156 if (mdi_component_is_client(cdip, NULL) != MDI_SUCCESS) 2157 return (1); 2158 2159 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, cdip, PROPFLAGS, 2160 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) 2161 return (1); 2162 2163 /* 2164 * Message is "sd# at scsi_vhci0: unit-address <guid>: <bus_addr>". 2165 * <guid> bus_addr argument == 0 2166 * <bus_addr> bus_addr argument != 0 2167 * Since the <guid> is already provided with unit-address, we just 2168 * provide failover module in <bus_addr> to keep output shorter. 2169 */ 2170 vlun = ADDR2VLUN(&sd->sd_address); 2171 if (bus_addr == 0) { 2172 /* report the guid: */ 2173 (void) snprintf(name, len, "g%s", guid); 2174 } else if (vlun && vlun->svl_fops_name) { 2175 /* report the name of the failover module */ 2176 (void) snprintf(name, len, "%s", vlun->svl_fops_name); 2177 } 2178 2179 ddi_prop_free(guid); 2180 return (1); 2181 } 2182 2183 static int 2184 vhci_scsi_get_bus_addr(struct scsi_device *sd, char *name, int len) 2185 { 2186 return (vhci_scsi_get_name_bus_addr(sd, name, len, 1)); 2187 } 2188 2189 static int 2190 vhci_scsi_get_name(struct scsi_device *sd, char *name, int len) 2191 { 2192 return (vhci_scsi_get_name_bus_addr(sd, name, len, 0)); 2193 } 2194 2195 /* 2196 * Return a pointer to the guid part of the devnm. 2197 * devnm format is "nodename@busaddr", busaddr format is "gGUID". 2198 */ 2199 static char * 2200 vhci_devnm_to_guid(char *devnm) 2201 { 2202 char *cp = devnm; 2203 2204 if (devnm == NULL) 2205 return (NULL); 2206 2207 while (*cp != '\0' && *cp != '@') 2208 cp++; 2209 if (*cp == '@' && *(cp + 1) == 'g') 2210 return (cp + 2); 2211 return (NULL); 2212 } 2213 2214 static int 2215 vhci_bind_transport(struct scsi_address *ap, struct vhci_pkt *vpkt, int flags, 2216 int (*func)(caddr_t)) 2217 { 2218 struct scsi_vhci *vhci = ADDR2VHCI(ap); 2219 dev_info_t *cdip = ADDR2DIP(ap); 2220 mdi_pathinfo_t *pip = NULL; 2221 mdi_pathinfo_t *npip = NULL; 2222 scsi_vhci_priv_t *svp = NULL; 2223 struct scsi_device *psd = NULL; 2224 struct scsi_address *address = NULL; 2225 struct scsi_pkt *pkt = NULL; 2226 int rval = -1; 2227 int pgr_sema_held = 0; 2228 int held; 2229 int mps_flag = MDI_SELECT_ONLINE_PATH; 2230 struct scsi_vhci_lun *vlun; 2231 int path_instance = 0; 2232 2233 vlun = ADDR2VLUN(ap); 2234 ASSERT(vlun != 0); 2235 2236 if ((vpkt->vpkt_tgt_pkt->pkt_cdbp[0] == SCMD_PROUT) && 2237 (((vpkt->vpkt_tgt_pkt->pkt_cdbp[1] & 0x1f) == 2238 VHCI_PROUT_REGISTER) || 2239 ((vpkt->vpkt_tgt_pkt->pkt_cdbp[1] & 0x1f) == 2240 VHCI_PROUT_R_AND_IGNORE))) { 2241 if (!sema_tryp(&vlun->svl_pgr_sema)) 2242 return (TRAN_BUSY); 2243 pgr_sema_held = 1; 2244 if (vlun->svl_first_path != NULL) { 2245 rval = mdi_select_path(cdip, NULL, 2246 MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH, 2247 NULL, &pip); 2248 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 2249 VHCI_DEBUG(4, (CE_NOTE, NULL, 2250 "vhci_bind_transport: path select fail\n")); 2251 } else { 2252 npip = pip; 2253 do { 2254 if (npip == vlun->svl_first_path) { 2255 VHCI_DEBUG(4, (CE_NOTE, NULL, 2256 "vhci_bind_transport: " 2257 "valid first path 0x%p\n", 2258 (void *) 2259 vlun->svl_first_path)); 2260 pip = vlun->svl_first_path; 2261 goto bind_path; 2262 } 2263 pip = npip; 2264 rval = mdi_select_path(cdip, NULL, 2265 MDI_SELECT_ONLINE_PATH | 2266 MDI_SELECT_STANDBY_PATH, 2267 pip, &npip); 2268 mdi_rele_path(pip); 2269 } while ((rval == MDI_SUCCESS) && 2270 (npip != NULL)); 2271 } 2272 } 2273 2274 if (vlun->svl_first_path) { 2275 VHCI_DEBUG(4, (CE_NOTE, NULL, 2276 "vhci_bind_transport: invalid first path 0x%p\n", 2277 (void *)vlun->svl_first_path)); 2278 vlun->svl_first_path = NULL; 2279 } 2280 } else if (vpkt->vpkt_tgt_pkt->pkt_cdbp[0] == SCMD_PRIN) { 2281 if ((vpkt->vpkt_state & VHCI_PKT_THRU_TASKQ) == 0) { 2282 if (!sema_tryp(&vlun->svl_pgr_sema)) 2283 return (TRAN_BUSY); 2284 } 2285 pgr_sema_held = 1; 2286 } 2287 2288 /* 2289 * If the path is already bound for PKT_PARTIAL_DMA case, 2290 * try to use the same path. 2291 */ 2292 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) && vpkt->vpkt_path) { 2293 VHCI_DEBUG(4, (CE_NOTE, NULL, 2294 "vhci_bind_transport: PKT_PARTIAL_DMA " 2295 "vpkt 0x%p, path 0x%p\n", 2296 (void *)vpkt, (void *)vpkt->vpkt_path)); 2297 pip = vpkt->vpkt_path; 2298 goto bind_path; 2299 } 2300 2301 /* 2302 * Get path_instance. Non-zero with FLAG_PKT_PATH_INSTANCE set 2303 * indicates that mdi_select_path should be called to select a 2304 * specific instance. 2305 * 2306 * NB: Condition pkt_path_instance reference on proper allocation. 2307 */ 2308 if ((vpkt->vpkt_tgt_pkt->pkt_flags & FLAG_PKT_PATH_INSTANCE) && 2309 scsi_pkt_allocated_correctly(vpkt->vpkt_tgt_pkt)) { 2310 path_instance = vpkt->vpkt_tgt_pkt->pkt_path_instance; 2311 } 2312 2313 /* 2314 * If reservation is active bind the transport directly to the pip 2315 * with the reservation. 2316 */ 2317 if (vpkt->vpkt_hba_pkt == NULL) { 2318 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 2319 if (MDI_PI_IS_ONLINE(vlun->svl_resrv_pip)) { 2320 pip = vlun->svl_resrv_pip; 2321 mdi_hold_path(pip); 2322 vlun->svl_waiting_for_activepath = 0; 2323 rval = MDI_SUCCESS; 2324 goto bind_path; 2325 } else { 2326 if (pgr_sema_held) { 2327 sema_v(&vlun->svl_pgr_sema); 2328 } 2329 return (TRAN_BUSY); 2330 } 2331 } 2332 try_again: 2333 rval = mdi_select_path(cdip, vpkt->vpkt_tgt_init_bp, 2334 path_instance ? MDI_SELECT_PATH_INSTANCE : 0, 2335 (void *)(intptr_t)path_instance, &pip); 2336 if (rval == MDI_BUSY) { 2337 if (pgr_sema_held) { 2338 sema_v(&vlun->svl_pgr_sema); 2339 } 2340 return (TRAN_BUSY); 2341 } else if (rval == MDI_DEVI_ONLINING) { 2342 /* 2343 * if we are here then we are in the midst of 2344 * an attach/probe of the client device. 2345 * We attempt to bind to ONLINE path if available, 2346 * else it is OK to bind to a STANDBY path (instead 2347 * of triggering a failover) because IO associated 2348 * with attach/probe (eg. INQUIRY, block 0 read) 2349 * are completed by targets even on passive paths 2350 * If no ONLINE paths available, it is important 2351 * to set svl_waiting_for_activepath for two 2352 * reasons: (1) avoid sense analysis in the 2353 * "external failure detection" codepath in 2354 * vhci_intr(). Failure to do so will result in 2355 * infinite loop (unless an ONLINE path becomes 2356 * available at some point) (2) avoid 2357 * unnecessary failover (see "---Waiting For Active 2358 * Path---" comment below). 2359 */ 2360 VHCI_DEBUG(1, (CE_NOTE, NULL, "!%p in onlining " 2361 "state\n", (void *)cdip)); 2362 pip = NULL; 2363 rval = mdi_select_path(cdip, vpkt->vpkt_tgt_init_bp, 2364 mps_flag, NULL, &pip); 2365 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 2366 if (vlun->svl_waiting_for_activepath == 0) { 2367 vlun->svl_waiting_for_activepath = 1; 2368 vlun->svl_wfa_time = gethrtime(); 2369 } 2370 mps_flag |= MDI_SELECT_STANDBY_PATH; 2371 rval = mdi_select_path(cdip, 2372 vpkt->vpkt_tgt_init_bp, 2373 mps_flag, NULL, &pip); 2374 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 2375 if (pgr_sema_held) { 2376 sema_v(&vlun->svl_pgr_sema); 2377 } 2378 return (TRAN_FATAL_ERROR); 2379 } 2380 goto bind_path; 2381 } 2382 } else if ((rval == MDI_FAILURE) || 2383 ((rval == MDI_NOPATH) && (path_instance))) { 2384 if (pgr_sema_held) { 2385 sema_v(&vlun->svl_pgr_sema); 2386 } 2387 return (TRAN_FATAL_ERROR); 2388 } 2389 2390 if ((pip == NULL) || (rval == MDI_NOPATH)) { 2391 while (vlun->svl_waiting_for_activepath) { 2392 /* 2393 * ---Waiting For Active Path--- 2394 * This device was discovered across a 2395 * passive path; lets wait for a little 2396 * bit, hopefully an active path will 2397 * show up obviating the need for a 2398 * failover 2399 */ 2400 if ((gethrtime() - vlun->svl_wfa_time) >= 2401 (60 * NANOSEC)) { 2402 vlun->svl_waiting_for_activepath = 0; 2403 } else { 2404 drv_usecwait(1000); 2405 if (vlun->svl_waiting_for_activepath 2406 == 0) { 2407 /* 2408 * an active path has come 2409 * online! 2410 */ 2411 goto try_again; 2412 } 2413 } 2414 } 2415 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 2416 if (!held) { 2417 VHCI_DEBUG(4, (CE_NOTE, NULL, 2418 "!Lun not held\n")); 2419 if (pgr_sema_held) { 2420 sema_v(&vlun->svl_pgr_sema); 2421 } 2422 return (TRAN_BUSY); 2423 } 2424 /* 2425 * now that the LUN is stable, one last check 2426 * to make sure no other changes sneaked in 2427 * (like a path coming online or a 2428 * failover initiated by another thread) 2429 */ 2430 pip = NULL; 2431 rval = mdi_select_path(cdip, vpkt->vpkt_tgt_init_bp, 2432 0, NULL, &pip); 2433 if (pip != NULL) { 2434 VHCI_RELEASE_LUN(vlun); 2435 vlun->svl_waiting_for_activepath = 0; 2436 goto bind_path; 2437 } 2438 2439 /* 2440 * Check if there is an ONLINE path OR a STANDBY path 2441 * available. If none is available, do not attempt 2442 * to do a failover, just return a fatal error at this 2443 * point. 2444 */ 2445 npip = NULL; 2446 rval = mdi_select_path(cdip, NULL, 2447 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 2448 NULL, &npip); 2449 if ((npip == NULL) || (rval != MDI_SUCCESS)) { 2450 /* 2451 * No paths available, jus return FATAL error. 2452 */ 2453 VHCI_RELEASE_LUN(vlun); 2454 if (pgr_sema_held) { 2455 sema_v(&vlun->svl_pgr_sema); 2456 } 2457 return (TRAN_FATAL_ERROR); 2458 } 2459 mdi_rele_path(npip); 2460 if (!(vpkt->vpkt_state & VHCI_PKT_IN_FAILOVER)) { 2461 VHCI_DEBUG(1, (CE_NOTE, NULL, "!invoking " 2462 "mdi_failover\n")); 2463 rval = mdi_failover(vhci->vhci_dip, cdip, 2464 MDI_FAILOVER_ASYNC); 2465 } else { 2466 rval = vlun->svl_failover_status; 2467 } 2468 if (rval == MDI_FAILURE) { 2469 VHCI_RELEASE_LUN(vlun); 2470 if (pgr_sema_held) { 2471 sema_v(&vlun->svl_pgr_sema); 2472 } 2473 return (TRAN_FATAL_ERROR); 2474 } else if (rval == MDI_BUSY) { 2475 VHCI_RELEASE_LUN(vlun); 2476 if (pgr_sema_held) { 2477 sema_v(&vlun->svl_pgr_sema); 2478 } 2479 return (TRAN_BUSY); 2480 } else { 2481 if (pgr_sema_held) { 2482 sema_v(&vlun->svl_pgr_sema); 2483 } 2484 vpkt->vpkt_state |= VHCI_PKT_IN_FAILOVER; 2485 return (TRAN_BUSY); 2486 } 2487 } 2488 vlun->svl_waiting_for_activepath = 0; 2489 bind_path: 2490 vpkt->vpkt_path = pip; 2491 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 2492 ASSERT(svp != NULL); 2493 2494 psd = svp->svp_psd; 2495 ASSERT(psd != NULL); 2496 address = &psd->sd_address; 2497 } else { 2498 pkt = vpkt->vpkt_hba_pkt; 2499 address = &pkt->pkt_address; 2500 } 2501 2502 /* Verify match of specified path_instance and selected path_instance */ 2503 ASSERT((path_instance == 0) || 2504 (path_instance == mdi_pi_get_path_instance(vpkt->vpkt_path))); 2505 2506 /* 2507 * For PKT_PARTIAL_DMA case, call pHCI's scsi_init_pkt whenever 2508 * target driver calls vhci_scsi_init_pkt. 2509 */ 2510 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) && 2511 vpkt->vpkt_path && vpkt->vpkt_hba_pkt) { 2512 VHCI_DEBUG(4, (CE_NOTE, NULL, 2513 "vhci_bind_transport: PKT_PARTIAL_DMA " 2514 "vpkt 0x%p, path 0x%p hba_pkt 0x%p\n", 2515 (void *)vpkt, (void *)vpkt->vpkt_path, (void *)pkt)); 2516 pkt = vpkt->vpkt_hba_pkt; 2517 address = &pkt->pkt_address; 2518 } 2519 2520 if (pkt == NULL || (vpkt->vpkt_flags & CFLAG_DMA_PARTIAL)) { 2521 pkt = scsi_init_pkt(address, pkt, 2522 vpkt->vpkt_tgt_init_bp, vpkt->vpkt_tgt_init_cdblen, 2523 vpkt->vpkt_tgt_init_scblen, 0, flags, func, NULL); 2524 2525 if (pkt == NULL) { 2526 VHCI_DEBUG(4, (CE_NOTE, NULL, 2527 "!bind transport: 0x%p 0x%p 0x%p\n", 2528 (void *)vhci, (void *)psd, (void *)vpkt)); 2529 if ((vpkt->vpkt_hba_pkt == NULL) && vpkt->vpkt_path) { 2530 MDI_PI_ERRSTAT(vpkt->vpkt_path, 2531 MDI_PI_TRANSERR); 2532 mdi_rele_path(vpkt->vpkt_path); 2533 vpkt->vpkt_path = NULL; 2534 } 2535 if (pgr_sema_held) { 2536 sema_v(&vlun->svl_pgr_sema); 2537 } 2538 /* 2539 * Consider it a fatal error if b_error is 2540 * set as a result of DMA binding failure 2541 * vs. a condition of being temporarily out of 2542 * some resource 2543 */ 2544 if (vpkt->vpkt_tgt_init_bp == NULL || 2545 geterror(vpkt->vpkt_tgt_init_bp)) 2546 return (TRAN_FATAL_ERROR); 2547 else 2548 return (TRAN_BUSY); 2549 } 2550 } 2551 2552 pkt->pkt_private = vpkt; 2553 vpkt->vpkt_hba_pkt = pkt; 2554 return (TRAN_ACCEPT); 2555 } 2556 2557 2558 /*PRINTFLIKE3*/ 2559 void 2560 vhci_log(int level, dev_info_t *dip, const char *fmt, ...) 2561 { 2562 char buf[256]; 2563 va_list ap; 2564 2565 va_start(ap, fmt); 2566 (void) vsprintf(buf, fmt, ap); 2567 va_end(ap); 2568 2569 scsi_log(dip, "scsi_vhci", level, buf); 2570 } 2571 2572 /* do a PGR out with the information we've saved away */ 2573 static int 2574 vhci_do_prout(scsi_vhci_priv_t *svp) 2575 { 2576 2577 struct scsi_pkt *new_pkt; 2578 struct buf *bp; 2579 scsi_vhci_lun_t *vlun = svp->svp_svl; 2580 int rval, retry, nr_retry, ua_retry; 2581 uint8_t *sns, skey; 2582 2583 bp = getrbuf(KM_SLEEP); 2584 bp->b_flags = B_WRITE; 2585 bp->b_resid = 0; 2586 bp->b_un.b_addr = (caddr_t)&vlun->svl_prout; 2587 bp->b_bcount = vlun->svl_bcount; 2588 2589 VHCI_INCR_PATH_CMDCOUNT(svp); 2590 2591 new_pkt = scsi_init_pkt(&svp->svp_psd->sd_address, NULL, bp, 2592 CDB_GROUP1, sizeof (struct scsi_arq_status), 0, 0, 2593 SLEEP_FUNC, NULL); 2594 if (new_pkt == NULL) { 2595 VHCI_DECR_PATH_CMDCOUNT(svp); 2596 freerbuf(bp); 2597 cmn_err(CE_WARN, "!vhci_do_prout: scsi_init_pkt failed"); 2598 return (0); 2599 } 2600 mutex_enter(&vlun->svl_mutex); 2601 bp->b_un.b_addr = (caddr_t)&vlun->svl_prout; 2602 bp->b_bcount = vlun->svl_bcount; 2603 bcopy(vlun->svl_cdb, new_pkt->pkt_cdbp, 2604 sizeof (vlun->svl_cdb)); 2605 new_pkt->pkt_time = vlun->svl_time; 2606 mutex_exit(&vlun->svl_mutex); 2607 new_pkt->pkt_flags = FLAG_NOINTR; 2608 2609 ua_retry = nr_retry = retry = 0; 2610 again: 2611 rval = vhci_do_scsi_cmd(new_pkt); 2612 if (rval != 1) { 2613 if ((new_pkt->pkt_reason == CMD_CMPLT) && 2614 (SCBP_C(new_pkt) == STATUS_CHECK) && 2615 (new_pkt->pkt_state & STATE_ARQ_DONE)) { 2616 sns = (uint8_t *) 2617 &(((struct scsi_arq_status *)(uintptr_t) 2618 (new_pkt->pkt_scbp))->sts_sensedata); 2619 skey = scsi_sense_key(sns); 2620 if ((skey == KEY_UNIT_ATTENTION) || 2621 (skey == KEY_NOT_READY)) { 2622 int max_retry; 2623 struct scsi_failover_ops *fops; 2624 fops = vlun->svl_fops; 2625 rval = fops->sfo_analyze_sense(svp->svp_psd, 2626 sns, vlun->svl_fops_ctpriv); 2627 if (rval == SCSI_SENSE_NOT_READY) { 2628 max_retry = vhci_prout_not_ready_retry; 2629 retry = nr_retry++; 2630 delay(1*drv_usectohz(1000000)); 2631 } else { 2632 /* chk for state change and update */ 2633 if (rval == SCSI_SENSE_STATE_CHANGED) { 2634 int held; 2635 VHCI_HOLD_LUN(vlun, 2636 VH_NOSLEEP, held); 2637 if (!held) { 2638 rval = TRAN_BUSY; 2639 } else { 2640 /* chk for alua first */ 2641 vhci_update_pathstates( 2642 (void *)vlun); 2643 } 2644 } 2645 retry = ua_retry++; 2646 max_retry = VHCI_MAX_PGR_RETRIES; 2647 } 2648 if (retry < max_retry) { 2649 VHCI_DEBUG(4, (CE_WARN, NULL, 2650 "!vhci_do_prout retry 0x%x " 2651 "(0x%x 0x%x 0x%x)", 2652 SCBP_C(new_pkt), 2653 new_pkt->pkt_cdbp[0], 2654 new_pkt->pkt_cdbp[1], 2655 new_pkt->pkt_cdbp[2])); 2656 goto again; 2657 } 2658 rval = 0; 2659 VHCI_DEBUG(4, (CE_WARN, NULL, 2660 "!vhci_do_prout 0x%x " 2661 "(0x%x 0x%x 0x%x)", 2662 SCBP_C(new_pkt), 2663 new_pkt->pkt_cdbp[0], 2664 new_pkt->pkt_cdbp[1], 2665 new_pkt->pkt_cdbp[2])); 2666 } else if (skey == KEY_ILLEGAL_REQUEST) 2667 rval = VHCI_PGR_ILLEGALOP; 2668 } 2669 } else { 2670 rval = 1; 2671 } 2672 scsi_destroy_pkt(new_pkt); 2673 VHCI_DECR_PATH_CMDCOUNT(svp); 2674 freerbuf(bp); 2675 return (rval); 2676 } 2677 2678 static void 2679 vhci_run_cmd(void *arg) 2680 { 2681 struct scsi_pkt *pkt = (struct scsi_pkt *)arg; 2682 struct scsi_pkt *tpkt; 2683 scsi_vhci_priv_t *svp; 2684 mdi_pathinfo_t *pip, *npip; 2685 scsi_vhci_lun_t *vlun; 2686 dev_info_t *cdip; 2687 scsi_vhci_priv_t *nsvp; 2688 int fail = 0; 2689 int rval; 2690 struct vhci_pkt *vpkt; 2691 uchar_t cdb_1; 2692 vhci_prout_t *prout; 2693 2694 vpkt = (struct vhci_pkt *)pkt->pkt_private; 2695 tpkt = vpkt->vpkt_tgt_pkt; 2696 pip = vpkt->vpkt_path; 2697 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 2698 if (svp == NULL) { 2699 tpkt->pkt_reason = CMD_TRAN_ERR; 2700 tpkt->pkt_statistics = STAT_ABORTED; 2701 goto done; 2702 } 2703 vlun = svp->svp_svl; 2704 prout = &vlun->svl_prout; 2705 if (SCBP_C(pkt) != STATUS_GOOD) 2706 fail++; 2707 cdip = vlun->svl_dip; 2708 pip = npip = NULL; 2709 rval = mdi_select_path(cdip, NULL, 2710 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, NULL, &npip); 2711 if ((rval != MDI_SUCCESS) || (npip == NULL)) { 2712 VHCI_DEBUG(4, (CE_NOTE, NULL, 2713 "vhci_run_cmd: no path! 0x%p\n", (void *)svp)); 2714 tpkt->pkt_reason = CMD_TRAN_ERR; 2715 tpkt->pkt_statistics = STAT_ABORTED; 2716 goto done; 2717 } 2718 2719 cdb_1 = vlun->svl_cdb[1]; 2720 vlun->svl_cdb[1] &= 0xe0; 2721 vlun->svl_cdb[1] |= VHCI_PROUT_R_AND_IGNORE; 2722 2723 do { 2724 nsvp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(npip); 2725 if (nsvp == NULL) { 2726 VHCI_DEBUG(4, (CE_NOTE, NULL, 2727 "vhci_run_cmd: no " 2728 "client priv! 0x%p offlined?\n", 2729 (void *)npip)); 2730 goto next_path; 2731 } 2732 if (vlun->svl_first_path == npip) { 2733 goto next_path; 2734 } else { 2735 if (vhci_do_prout(nsvp) != 1) 2736 fail++; 2737 } 2738 next_path: 2739 pip = npip; 2740 rval = mdi_select_path(cdip, NULL, 2741 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, 2742 pip, &npip); 2743 mdi_rele_path(pip); 2744 } while ((rval == MDI_SUCCESS) && (npip != NULL)); 2745 2746 vlun->svl_cdb[1] = cdb_1; 2747 2748 if (fail) { 2749 VHCI_DEBUG(4, (CE_WARN, NULL, "%s%d: key registration failed, " 2750 "couldn't be replicated on all paths", 2751 ddi_driver_name(cdip), ddi_get_instance(cdip))); 2752 vhci_print_prout_keys(vlun, "vhci_run_cmd: "); 2753 2754 if (SCBP_C(pkt) != STATUS_GOOD) { 2755 tpkt->pkt_reason = CMD_TRAN_ERR; 2756 tpkt->pkt_statistics = STAT_ABORTED; 2757 } 2758 } else { 2759 vlun->svl_pgr_active = 1; 2760 vhci_print_prout_keys(vlun, "vhci_run_cmd: before bcopy:"); 2761 2762 bcopy((const void *)prout->service_key, 2763 (void *)prout->active_service_key, MHIOC_RESV_KEY_SIZE); 2764 bcopy((const void *)prout->res_key, 2765 (void *)prout->active_res_key, MHIOC_RESV_KEY_SIZE); 2766 2767 vhci_print_prout_keys(vlun, "vhci_run_cmd: after bcopy:"); 2768 } 2769 done: 2770 if (SCBP_C(pkt) == STATUS_GOOD) 2771 vlun->svl_first_path = NULL; 2772 2773 if (svp) 2774 VHCI_DECR_PATH_CMDCOUNT(svp); 2775 2776 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 2777 scsi_destroy_pkt(pkt); 2778 vpkt->vpkt_hba_pkt = NULL; 2779 if (vpkt->vpkt_path) { 2780 mdi_rele_path(vpkt->vpkt_path); 2781 vpkt->vpkt_path = NULL; 2782 } 2783 } 2784 2785 sema_v(&vlun->svl_pgr_sema); 2786 /* 2787 * The PROUT commands are not included in the automatic retry 2788 * mechanism, therefore, vpkt_org_vpkt should never be set here. 2789 */ 2790 ASSERT(vpkt->vpkt_org_vpkt == NULL); 2791 scsi_hba_pkt_comp(tpkt); 2792 } 2793 2794 /* 2795 * Get the keys registered with this target. Since we will have 2796 * registered the same key with multiple initiators, strip out 2797 * any duplicate keys. 2798 * 2799 * The pointers which will be used to filter the registered keys from 2800 * the device will be stored in filter_prin and filter_pkt. If the 2801 * allocation length of the buffer was sufficient for the number of 2802 * parameter data bytes available to be returned by the device then the 2803 * key filtering will use the keylist returned from the original 2804 * request. If the allocation length of the buffer was not sufficient, 2805 * then the filtering will use the keylist returned from the request 2806 * that is resent below. 2807 * 2808 * If the device returns an additional length field that is greater than 2809 * the allocation length of the buffer, then allocate a new buffer which 2810 * can accommodate the number of parameter data bytes available to be 2811 * returned. Resend the scsi PRIN command, filter out the duplicate 2812 * keys and return as many of the unique keys found that was originally 2813 * requested and set the additional length field equal to the data bytes 2814 * of unique reservation keys available to be returned. 2815 * 2816 * If the device returns an additional length field that is less than or 2817 * equal to the allocation length of the buffer, then all the available 2818 * keys registered were returned by the device. Filter out the 2819 * duplicate keys and return all of the unique keys found and set the 2820 * additional length field equal to the data bytes of the reservation 2821 * keys to be returned. 2822 */ 2823 2824 #define VHCI_PRIN_HEADER_SZ (sizeof (prin->length) + sizeof (prin->generation)) 2825 2826 static int 2827 vhci_do_prin(struct vhci_pkt **intr_vpkt) 2828 { 2829 scsi_vhci_priv_t *svp; 2830 struct vhci_pkt *vpkt = *intr_vpkt; 2831 vhci_prin_readkeys_t *prin; 2832 scsi_vhci_lun_t *vlun; 2833 struct scsi_vhci *vhci = ADDR2VHCI(&vpkt->vpkt_tgt_pkt->pkt_address); 2834 2835 struct buf *new_bp = NULL; 2836 struct scsi_pkt *new_pkt = NULL; 2837 struct vhci_pkt *new_vpkt = NULL; 2838 uint32_t needed_length; 2839 int rval = VHCI_CMD_CMPLT; 2840 uint32_t prin_length = 0; 2841 uint32_t svl_prin_length = 0; 2842 2843 ASSERT(vpkt->vpkt_path); 2844 svp = mdi_pi_get_vhci_private(vpkt->vpkt_path); 2845 ASSERT(svp); 2846 vlun = svp->svp_svl; 2847 ASSERT(vlun); 2848 2849 /* 2850 * If the caller only asked for an amount of data that would not 2851 * be enough to include any key data it is likely that they will 2852 * send the next command with a buffer size based on the information 2853 * from this header. Doing recovery on this would be a duplication 2854 * of efforts. 2855 */ 2856 if (vpkt->vpkt_tgt_init_bp->b_bcount <= VHCI_PRIN_HEADER_SZ) { 2857 rval = VHCI_CMD_CMPLT; 2858 goto exit; 2859 } 2860 2861 if (vpkt->vpkt_org_vpkt == NULL) { 2862 /* 2863 * Can fail as sleep is not allowed. 2864 */ 2865 prin = (vhci_prin_readkeys_t *) 2866 bp_mapin_common(vpkt->vpkt_tgt_init_bp, VM_NOSLEEP); 2867 } else { 2868 /* 2869 * The retry buf doesn't need to be mapped in. 2870 */ 2871 prin = (vhci_prin_readkeys_t *) 2872 vpkt->vpkt_tgt_init_bp->b_un.b_daddr; 2873 } 2874 2875 if (prin == NULL) { 2876 VHCI_DEBUG(5, (CE_WARN, NULL, 2877 "vhci_do_prin: bp_mapin_common failed.")); 2878 rval = VHCI_CMD_ERROR; 2879 goto fail; 2880 } 2881 2882 prin_length = BE_32(prin->length); 2883 2884 /* 2885 * According to SPC-3r22, sec 4.3.4.6: "If the amount of 2886 * information to be transferred exceeds the maximum value 2887 * that the ALLOCATION LENGTH field is capable of specifying, 2888 * the device server shall...terminate the command with CHECK 2889 * CONDITION status". The ALLOCATION LENGTH field of the 2890 * PERSISTENT RESERVE IN command is 2 bytes. We should never 2891 * get here with an ADDITIONAL LENGTH greater than 0xFFFF 2892 * so if we do, then it is an error! 2893 */ 2894 2895 2896 if ((prin_length + VHCI_PRIN_HEADER_SZ) > 0xFFFF) { 2897 VHCI_DEBUG(5, (CE_NOTE, NULL, 2898 "vhci_do_prin: Device returned invalid " 2899 "length 0x%x\n", prin_length)); 2900 rval = VHCI_CMD_ERROR; 2901 goto fail; 2902 } 2903 needed_length = prin_length + VHCI_PRIN_HEADER_SZ; 2904 2905 /* 2906 * If prin->length is greater than the byte count allocated in the 2907 * original buffer, then resend the request with enough buffer 2908 * allocated to get all of the available registered keys. 2909 */ 2910 if ((vpkt->vpkt_tgt_init_bp->b_bcount < needed_length) && 2911 (vpkt->vpkt_org_vpkt == NULL)) { 2912 2913 new_pkt = vhci_create_retry_pkt(vpkt); 2914 if (new_pkt == NULL) { 2915 rval = VHCI_CMD_ERROR; 2916 goto fail; 2917 } 2918 new_vpkt = TGTPKT2VHCIPKT(new_pkt); 2919 2920 /* 2921 * This is the buf with buffer pointer 2922 * where the prin readkeys will be 2923 * returned from the device 2924 */ 2925 new_bp = scsi_alloc_consistent_buf(&svp->svp_psd->sd_address, 2926 NULL, needed_length, B_READ, NULL_FUNC, NULL); 2927 if ((new_bp == NULL) || (new_bp->b_un.b_addr == NULL)) { 2928 if (new_bp) { 2929 scsi_free_consistent_buf(new_bp); 2930 } 2931 vhci_scsi_destroy_pkt(&new_pkt->pkt_address, new_pkt); 2932 rval = VHCI_CMD_ERROR; 2933 goto fail; 2934 } 2935 new_bp->b_bcount = needed_length; 2936 new_pkt->pkt_cdbp[7] = (uchar_t)(needed_length >> 8); 2937 new_pkt->pkt_cdbp[8] = (uchar_t)needed_length; 2938 2939 rval = VHCI_CMD_RETRY; 2940 2941 new_vpkt->vpkt_tgt_init_bp = new_bp; 2942 } 2943 2944 if (rval == VHCI_CMD_RETRY) { 2945 2946 /* 2947 * There were more keys then the original request asked for. 2948 */ 2949 mdi_pathinfo_t *path_holder = vpkt->vpkt_path; 2950 2951 /* 2952 * Release the old path because it does not matter which path 2953 * this command is sent down. This allows the normal bind 2954 * transport mechanism to be used. 2955 */ 2956 if (vpkt->vpkt_path != NULL) { 2957 mdi_rele_path(vpkt->vpkt_path); 2958 vpkt->vpkt_path = NULL; 2959 } 2960 2961 /* 2962 * Dispatch the retry command 2963 */ 2964 if (taskq_dispatch(vhci->vhci_taskq, vhci_dispatch_scsi_start, 2965 (void *) new_vpkt, KM_NOSLEEP) == NULL) { 2966 if (path_holder) { 2967 vpkt->vpkt_path = path_holder; 2968 mdi_hold_path(path_holder); 2969 } 2970 scsi_free_consistent_buf(new_bp); 2971 vhci_scsi_destroy_pkt(&new_pkt->pkt_address, new_pkt); 2972 rval = VHCI_CMD_ERROR; 2973 goto fail; 2974 } 2975 2976 /* 2977 * If we return VHCI_CMD_RETRY, that means the caller 2978 * is going to bail and wait for the reissued command 2979 * to complete. In that case, we need to decrement 2980 * the path command count right now. In any other 2981 * case, it'll be decremented by the caller. 2982 */ 2983 VHCI_DECR_PATH_CMDCOUNT(svp); 2984 goto exit; 2985 2986 } 2987 2988 if (rval == VHCI_CMD_CMPLT) { 2989 /* 2990 * The original request got all of the keys or the recovery 2991 * packet returns. 2992 */ 2993 int new; 2994 int old; 2995 int num_keys = prin_length / MHIOC_RESV_KEY_SIZE; 2996 2997 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_do_prin: %d keys read\n", 2998 num_keys)); 2999 3000 #ifdef DEBUG 3001 VHCI_DEBUG(5, (CE_NOTE, NULL, "vhci_do_prin: from storage\n")); 3002 if (vhci_debug == 5) 3003 vhci_print_prin_keys(prin, num_keys); 3004 VHCI_DEBUG(5, (CE_NOTE, NULL, 3005 "vhci_do_prin: MPxIO old keys:\n")); 3006 if (vhci_debug == 5) 3007 vhci_print_prin_keys(&vlun->svl_prin, num_keys); 3008 #endif 3009 3010 /* 3011 * Filter out all duplicate keys returned from the device 3012 * We know that we use a different key for every host, so we 3013 * can simply strip out duplicates. Otherwise we would need to 3014 * do more bookkeeping to figure out which keys to strip out. 3015 */ 3016 3017 new = 0; 3018 3019 /* 3020 * If we got at least 1 key copy it. 3021 */ 3022 if (num_keys > 0) { 3023 vlun->svl_prin.keylist[0] = prin->keylist[0]; 3024 new++; 3025 } 3026 3027 /* 3028 * find next unique key. 3029 */ 3030 for (old = 1; old < num_keys; old++) { 3031 int j; 3032 int match = 0; 3033 3034 if (new >= VHCI_NUM_RESV_KEYS) 3035 break; 3036 for (j = 0; j < new; j++) { 3037 if (bcmp(&prin->keylist[old], 3038 &vlun->svl_prin.keylist[j], 3039 sizeof (mhioc_resv_key_t)) == 0) { 3040 match = 1; 3041 break; 3042 } 3043 } 3044 if (!match) { 3045 vlun->svl_prin.keylist[new] = 3046 prin->keylist[old]; 3047 new++; 3048 } 3049 } 3050 3051 /* Stored Big Endian */ 3052 vlun->svl_prin.generation = prin->generation; 3053 svl_prin_length = new * sizeof (mhioc_resv_key_t); 3054 /* Stored Big Endian */ 3055 vlun->svl_prin.length = BE_32(svl_prin_length); 3056 svl_prin_length += VHCI_PRIN_HEADER_SZ; 3057 3058 /* 3059 * If we arrived at this point after issuing a retry, make sure 3060 * that we put everything back the way it originally was so 3061 * that the target driver can complete the command correctly. 3062 */ 3063 if (vpkt->vpkt_org_vpkt != NULL) { 3064 new_bp = vpkt->vpkt_tgt_init_bp; 3065 3066 scsi_free_consistent_buf(new_bp); 3067 3068 vpkt = vhci_sync_retry_pkt(vpkt); 3069 *intr_vpkt = vpkt; 3070 3071 /* 3072 * Make sure the original buffer is mapped into kernel 3073 * space before we try to copy the filtered keys into 3074 * it. 3075 */ 3076 prin = (vhci_prin_readkeys_t *)bp_mapin_common( 3077 vpkt->vpkt_tgt_init_bp, VM_NOSLEEP); 3078 } 3079 3080 /* 3081 * Now copy the desired number of prin keys into the original 3082 * target buffer. 3083 */ 3084 if (svl_prin_length <= vpkt->vpkt_tgt_init_bp->b_bcount) { 3085 /* 3086 * It is safe to return all of the available unique 3087 * keys 3088 */ 3089 bcopy(&vlun->svl_prin, prin, svl_prin_length); 3090 } else { 3091 /* 3092 * Not all of the available keys were requested by the 3093 * original command. 3094 */ 3095 bcopy(&vlun->svl_prin, prin, 3096 vpkt->vpkt_tgt_init_bp->b_bcount); 3097 } 3098 #ifdef DEBUG 3099 VHCI_DEBUG(5, (CE_NOTE, NULL, 3100 "vhci_do_prin: To Application:\n")); 3101 if (vhci_debug == 5) 3102 vhci_print_prin_keys(prin, new); 3103 VHCI_DEBUG(5, (CE_NOTE, NULL, 3104 "vhci_do_prin: MPxIO new keys:\n")); 3105 if (vhci_debug == 5) 3106 vhci_print_prin_keys(&vlun->svl_prin, new); 3107 #endif 3108 } 3109 fail: 3110 if (rval == VHCI_CMD_ERROR) { 3111 /* 3112 * If we arrived at this point after issuing a 3113 * retry, make sure that we put everything back 3114 * the way it originally was so that ssd can 3115 * complete the command correctly. 3116 */ 3117 3118 if (vpkt->vpkt_org_vpkt != NULL) { 3119 new_bp = vpkt->vpkt_tgt_init_bp; 3120 if (new_bp != NULL) { 3121 scsi_free_consistent_buf(new_bp); 3122 } 3123 3124 new_vpkt = vpkt; 3125 vpkt = vpkt->vpkt_org_vpkt; 3126 3127 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 3128 new_vpkt->vpkt_tgt_pkt); 3129 } 3130 3131 /* 3132 * Mark this command completion as having an error so that 3133 * ssd will retry the command. 3134 */ 3135 3136 vpkt->vpkt_tgt_pkt->pkt_reason = CMD_ABORTED; 3137 vpkt->vpkt_tgt_pkt->pkt_statistics |= STAT_ABORTED; 3138 3139 rval = VHCI_CMD_CMPLT; 3140 } 3141 exit: 3142 /* 3143 * Make sure that the semaphore is only released once. 3144 */ 3145 if (rval == VHCI_CMD_CMPLT) { 3146 sema_v(&vlun->svl_pgr_sema); 3147 } 3148 3149 return (rval); 3150 } 3151 3152 static void 3153 vhci_intr(struct scsi_pkt *pkt) 3154 { 3155 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_private; 3156 struct scsi_pkt *tpkt; 3157 scsi_vhci_priv_t *svp; 3158 scsi_vhci_lun_t *vlun; 3159 int rval, held; 3160 struct scsi_failover_ops *fops; 3161 uint8_t *sns, skey, asc, ascq; 3162 mdi_pathinfo_t *lpath; 3163 static char *timeout_err = "Command Timeout"; 3164 static char *parity_err = "Parity Error"; 3165 char *err_str = NULL; 3166 dev_info_t *vdip, *cdip; 3167 char *cpath; 3168 3169 ASSERT(vpkt != NULL); 3170 tpkt = vpkt->vpkt_tgt_pkt; 3171 ASSERT(tpkt != NULL); 3172 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(vpkt->vpkt_path); 3173 ASSERT(svp != NULL); 3174 vlun = svp->svp_svl; 3175 ASSERT(vlun != NULL); 3176 lpath = vpkt->vpkt_path; 3177 3178 /* 3179 * sync up the target driver's pkt with the pkt that 3180 * we actually used 3181 */ 3182 *(tpkt->pkt_scbp) = *(pkt->pkt_scbp); 3183 tpkt->pkt_resid = pkt->pkt_resid; 3184 tpkt->pkt_state = pkt->pkt_state; 3185 tpkt->pkt_statistics = pkt->pkt_statistics; 3186 tpkt->pkt_reason = pkt->pkt_reason; 3187 3188 /* Return path_instance information back to the target driver. */ 3189 if (scsi_pkt_allocated_correctly(tpkt)) { 3190 if (scsi_pkt_allocated_correctly(pkt)) { 3191 /* 3192 * If both packets were correctly allocated, 3193 * return path returned by pHCI. 3194 */ 3195 tpkt->pkt_path_instance = pkt->pkt_path_instance; 3196 } else { 3197 /* Otherwise return path of pHCI we used */ 3198 tpkt->pkt_path_instance = 3199 mdi_pi_get_path_instance(lpath); 3200 } 3201 } 3202 3203 if (pkt->pkt_cdbp[0] == SCMD_PROUT && 3204 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 3205 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE)) { 3206 if ((SCBP_C(pkt) != STATUS_GOOD) || 3207 (pkt->pkt_reason != CMD_CMPLT)) { 3208 sema_v(&vlun->svl_pgr_sema); 3209 } 3210 } else if (pkt->pkt_cdbp[0] == SCMD_PRIN) { 3211 if (pkt->pkt_reason != CMD_CMPLT || 3212 (SCBP_C(pkt) != STATUS_GOOD)) { 3213 sema_v(&vlun->svl_pgr_sema); 3214 } 3215 } 3216 3217 switch (pkt->pkt_reason) { 3218 case CMD_CMPLT: 3219 /* 3220 * cmd completed successfully, check for scsi errors 3221 */ 3222 switch (*(pkt->pkt_scbp)) { 3223 case STATUS_CHECK: 3224 if (pkt->pkt_state & STATE_ARQ_DONE) { 3225 sns = (uint8_t *) 3226 &(((struct scsi_arq_status *)(uintptr_t) 3227 (pkt->pkt_scbp))->sts_sensedata); 3228 skey = scsi_sense_key(sns); 3229 asc = scsi_sense_asc(sns); 3230 ascq = scsi_sense_ascq(sns); 3231 fops = vlun->svl_fops; 3232 ASSERT(fops != NULL); 3233 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_intr: " 3234 "Received sns key %x esc %x escq %x\n", 3235 skey, asc, ascq)); 3236 3237 if (vlun->svl_waiting_for_activepath == 1) { 3238 /* 3239 * if we are here it means we are 3240 * in the midst of a probe/attach 3241 * through a passive path; this 3242 * case is exempt from sense analysis 3243 * for detection of ext. failover 3244 * because that would unnecessarily 3245 * increase attach time. 3246 */ 3247 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 3248 vpkt->vpkt_tgt_init_scblen); 3249 break; 3250 } 3251 if (asc == VHCI_SCSI_PERR) { 3252 /* 3253 * parity error 3254 */ 3255 err_str = parity_err; 3256 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 3257 vpkt->vpkt_tgt_init_scblen); 3258 break; 3259 } 3260 rval = fops->sfo_analyze_sense(svp->svp_psd, 3261 sns, vlun->svl_fops_ctpriv); 3262 if ((rval == SCSI_SENSE_NOFAILOVER) || 3263 (rval == SCSI_SENSE_UNKNOWN) || 3264 (rval == SCSI_SENSE_NOT_READY)) { 3265 bcopy(pkt->pkt_scbp, tpkt->pkt_scbp, 3266 vpkt->vpkt_tgt_init_scblen); 3267 break; 3268 } else if (rval == SCSI_SENSE_STATE_CHANGED) { 3269 struct scsi_vhci *vhci; 3270 vhci = ADDR2VHCI(&tpkt->pkt_address); 3271 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 3272 if (!held) { 3273 /* 3274 * looks like some other thread 3275 * has already detected this 3276 * condition 3277 */ 3278 tpkt->pkt_state &= 3279 ~STATE_ARQ_DONE; 3280 *(tpkt->pkt_scbp) = 3281 STATUS_BUSY; 3282 break; 3283 } 3284 (void) taskq_dispatch( 3285 vhci->vhci_update_pathstates_taskq, 3286 vhci_update_pathstates, 3287 (void *)vlun, KM_SLEEP); 3288 } else { 3289 /* 3290 * externally initiated failover 3291 * has occurred or is in progress 3292 */ 3293 VHCI_HOLD_LUN(vlun, VH_NOSLEEP, held); 3294 if (!held) { 3295 /* 3296 * looks like some other thread 3297 * has already detected this 3298 * condition 3299 */ 3300 tpkt->pkt_state &= 3301 ~STATE_ARQ_DONE; 3302 *(tpkt->pkt_scbp) = 3303 STATUS_BUSY; 3304 break; 3305 } else { 3306 rval = vhci_handle_ext_fo 3307 (pkt, rval); 3308 if (rval == BUSY_RETURN) { 3309 tpkt->pkt_state &= 3310 ~STATE_ARQ_DONE; 3311 *(tpkt->pkt_scbp) = 3312 STATUS_BUSY; 3313 break; 3314 } 3315 bcopy(pkt->pkt_scbp, 3316 tpkt->pkt_scbp, 3317 vpkt->vpkt_tgt_init_scblen); 3318 break; 3319 } 3320 } 3321 } 3322 break; 3323 3324 /* 3325 * If this is a good SCSI-II RELEASE cmd completion then restore 3326 * the load balancing policy and reset VLUN_RESERVE_ACTIVE_FLG. 3327 * If this is a good SCSI-II RESERVE cmd completion then set 3328 * VLUN_RESERVE_ACTIVE_FLG. 3329 */ 3330 case STATUS_GOOD: 3331 if ((pkt->pkt_cdbp[0] == SCMD_RELEASE) || 3332 (pkt->pkt_cdbp[0] == SCMD_RELEASE_G1)) { 3333 (void) mdi_set_lb_policy(vlun->svl_dip, 3334 vlun->svl_lb_policy_save); 3335 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 3336 VHCI_DEBUG(1, (CE_WARN, NULL, 3337 "!vhci_intr: vlun 0x%p release path 0x%p", 3338 (void *)vlun, (void *)vpkt->vpkt_path)); 3339 } 3340 3341 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 3342 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 3343 vlun->svl_flags |= VLUN_RESERVE_ACTIVE_FLG; 3344 vlun->svl_resrv_pip = vpkt->vpkt_path; 3345 VHCI_DEBUG(1, (CE_WARN, NULL, 3346 "!vhci_intr: vlun 0x%p reserved path 0x%p", 3347 (void *)vlun, (void *)vpkt->vpkt_path)); 3348 } 3349 break; 3350 3351 case STATUS_RESERVATION_CONFLICT: 3352 VHCI_DEBUG(1, (CE_WARN, NULL, 3353 "!vhci_intr: vlun 0x%p " 3354 "reserve conflict on path 0x%p", 3355 (void *)vlun, (void *)vpkt->vpkt_path)); 3356 /* FALLTHROUGH */ 3357 default: 3358 break; 3359 } 3360 3361 /* 3362 * Update I/O completion statistics for the path 3363 */ 3364 mdi_pi_kstat_iosupdate(vpkt->vpkt_path, vpkt->vpkt_tgt_init_bp); 3365 3366 /* 3367 * Command completed successfully, release the dma binding and 3368 * destroy the transport side of the packet. 3369 */ 3370 if ((pkt->pkt_cdbp[0] == SCMD_PROUT) && 3371 (((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_REGISTER) || 3372 ((pkt->pkt_cdbp[1] & 0x1f) == VHCI_PROUT_R_AND_IGNORE))) { 3373 if (SCBP_C(pkt) == STATUS_GOOD) { 3374 ASSERT(vlun->svl_taskq); 3375 svp->svp_last_pkt_reason = pkt->pkt_reason; 3376 (void) taskq_dispatch(vlun->svl_taskq, 3377 vhci_run_cmd, pkt, KM_SLEEP); 3378 return; 3379 } 3380 } 3381 if ((SCBP_C(pkt) == STATUS_GOOD) && 3382 (pkt->pkt_cdbp[0] == SCMD_PRIN) && vpkt->vpkt_tgt_init_bp) { 3383 /* 3384 * If the action (value in byte 1 of the cdb) is zero, 3385 * we're reading keys, and that's the only condition 3386 * where we need to be concerned with filtering keys 3387 * and potential retries. Otherwise, we simply signal 3388 * the semaphore and move on. 3389 */ 3390 if (pkt->pkt_cdbp[1] == 0) { 3391 /* 3392 * If this is the completion of an internal 3393 * retry then we need to make sure that the 3394 * pkt and tpkt pointers are readjusted so 3395 * the calls to scsi_destroy_pkt and pkt_comp 3396 * below work * correctly. 3397 */ 3398 if (vpkt->vpkt_org_vpkt != NULL) { 3399 pkt = vpkt->vpkt_org_vpkt->vpkt_hba_pkt; 3400 tpkt = vpkt->vpkt_org_vpkt-> 3401 vpkt_tgt_pkt; 3402 3403 /* 3404 * If this command was issued through 3405 * the taskq then we need to clear 3406 * this flag for proper processing in 3407 * the case of a retry from the target 3408 * driver. 3409 */ 3410 vpkt->vpkt_state &= 3411 ~VHCI_PKT_THRU_TASKQ; 3412 } 3413 3414 /* 3415 * if vhci_do_prin returns VHCI_CMD_CMPLT then 3416 * vpkt will contain the address of the 3417 * original vpkt 3418 */ 3419 if (vhci_do_prin(&vpkt) == VHCI_CMD_RETRY) { 3420 /* 3421 * The command has been resent to get 3422 * all the keys from the device. Don't 3423 * complete the command with ssd until 3424 * the retry completes. 3425 */ 3426 return; 3427 } 3428 } else { 3429 sema_v(&vlun->svl_pgr_sema); 3430 } 3431 } 3432 3433 break; 3434 3435 case CMD_TIMEOUT: 3436 if ((pkt->pkt_statistics & 3437 (STAT_BUS_RESET|STAT_DEV_RESET|STAT_ABORTED)) == 0) { 3438 3439 VHCI_DEBUG(1, (CE_NOTE, NULL, 3440 "!scsi vhci timeout invoked\n")); 3441 3442 (void) vhci_recovery_reset(vlun, &pkt->pkt_address, 3443 FALSE, VHCI_DEPTH_ALL); 3444 } 3445 MDI_PI_ERRSTAT(lpath, MDI_PI_TRANSERR); 3446 tpkt->pkt_statistics |= STAT_ABORTED; 3447 err_str = timeout_err; 3448 break; 3449 3450 case CMD_TRAN_ERR: 3451 /* 3452 * This status is returned if the transport has sent the cmd 3453 * down the link to the target and then some error occurs. 3454 * In case of SCSI-II RESERVE cmd, we don't know if the 3455 * reservation been accepted by the target or not, so we need 3456 * to clear the reservation. 3457 */ 3458 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 3459 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 3460 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_intr received" 3461 " cmd_tran_err for scsi-2 reserve cmd\n")); 3462 if (!vhci_recovery_reset(vlun, &pkt->pkt_address, 3463 TRUE, VHCI_DEPTH_TARGET)) { 3464 VHCI_DEBUG(1, (CE_WARN, NULL, 3465 "!vhci_intr cmd_tran_err reset failed!")); 3466 } 3467 } 3468 break; 3469 3470 case CMD_DEV_GONE: 3471 /* 3472 * If this is the last path then report CMD_DEV_GONE to the 3473 * target driver, otherwise report BUSY to triggger retry. 3474 */ 3475 if (vlun->svl_dip && 3476 (mdi_client_get_path_count(vlun->svl_dip) <= 1)) { 3477 struct scsi_vhci *vhci; 3478 vhci = ADDR2VHCI(&tpkt->pkt_address); 3479 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_intr received " 3480 "cmd_dev_gone on last path\n")); 3481 (void) vhci_invalidate_mpapi_lu(vhci, vlun); 3482 break; 3483 } 3484 3485 /* Report CMD_CMPLT-with-BUSY to cause retry. */ 3486 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_intr received " 3487 "cmd_dev_gone\n")); 3488 tpkt->pkt_reason = CMD_CMPLT; 3489 tpkt->pkt_state = STATE_GOT_BUS | 3490 STATE_GOT_TARGET | STATE_SENT_CMD | 3491 STATE_GOT_STATUS; 3492 *(tpkt->pkt_scbp) = STATUS_BUSY; 3493 break; 3494 3495 default: 3496 break; 3497 } 3498 3499 /* 3500 * SCSI-II RESERVE cmd has been serviced by the lower layers clear 3501 * the flag so the lun is not QUIESCED any longer. 3502 * Also clear the VHCI_PKT_THRU_TASKQ flag, to ensure that if this pkt 3503 * is retried, a taskq shall again be dispatched to service it. Else 3504 * it may lead to a system hang if the retry is within interrupt 3505 * context. 3506 */ 3507 if ((pkt->pkt_cdbp[0] == SCMD_RESERVE) || 3508 (pkt->pkt_cdbp[0] == SCMD_RESERVE_G1)) { 3509 vlun->svl_flags &= ~VLUN_QUIESCED_FLG; 3510 vpkt->vpkt_state &= ~VHCI_PKT_THRU_TASKQ; 3511 } 3512 3513 /* 3514 * vpkt_org_vpkt should always be NULL here if the retry command 3515 * has been successfully processed. If vpkt_org_vpkt != NULL at 3516 * this point, it is an error so restore the original vpkt and 3517 * return an error to the target driver so it can retry the 3518 * command as appropriate. 3519 */ 3520 if (vpkt->vpkt_org_vpkt != NULL) { 3521 struct vhci_pkt *new_vpkt = vpkt; 3522 vpkt = vpkt->vpkt_org_vpkt; 3523 3524 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 3525 new_vpkt->vpkt_tgt_pkt); 3526 3527 /* 3528 * Mark this command completion as having an error so that 3529 * ssd will retry the command. 3530 */ 3531 vpkt->vpkt_tgt_pkt->pkt_reason = CMD_ABORTED; 3532 vpkt->vpkt_tgt_pkt->pkt_statistics |= STAT_ABORTED; 3533 3534 pkt = vpkt->vpkt_hba_pkt; 3535 tpkt = vpkt->vpkt_tgt_pkt; 3536 } 3537 3538 if ((err_str != NULL) && (pkt->pkt_reason != 3539 svp->svp_last_pkt_reason)) { 3540 cdip = vlun->svl_dip; 3541 vdip = ddi_get_parent(cdip); 3542 cpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3543 vhci_log(CE_WARN, vdip, "!%s (%s%d): %s on path %s", 3544 ddi_pathname(cdip, cpath), ddi_driver_name(cdip), 3545 ddi_get_instance(cdip), err_str, 3546 mdi_pi_spathname(vpkt->vpkt_path)); 3547 kmem_free(cpath, MAXPATHLEN); 3548 } 3549 svp->svp_last_pkt_reason = pkt->pkt_reason; 3550 VHCI_DECR_PATH_CMDCOUNT(svp); 3551 3552 /* 3553 * For PARTIAL_DMA, vhci should not free the path. 3554 * Target driver will call into vhci_scsi_dmafree or 3555 * destroy pkt to release this path. 3556 */ 3557 if ((vpkt->vpkt_flags & CFLAG_DMA_PARTIAL) == 0) { 3558 scsi_destroy_pkt(pkt); 3559 vpkt->vpkt_hba_pkt = NULL; 3560 if (vpkt->vpkt_path) { 3561 mdi_rele_path(vpkt->vpkt_path); 3562 vpkt->vpkt_path = NULL; 3563 } 3564 } 3565 3566 scsi_hba_pkt_comp(tpkt); 3567 } 3568 3569 /* 3570 * two possibilities: (1) failover has completed 3571 * or (2) is in progress; update our path states for 3572 * the former case; for the latter case, 3573 * initiate a scsi_watch request to 3574 * determine when failover completes - vlun is HELD 3575 * until failover completes; BUSY is returned to upper 3576 * layer in both the cases 3577 */ 3578 static int 3579 vhci_handle_ext_fo(struct scsi_pkt *pkt, int fostat) 3580 { 3581 struct vhci_pkt *vpkt = (struct vhci_pkt *)pkt->pkt_private; 3582 struct scsi_pkt *tpkt; 3583 scsi_vhci_priv_t *svp; 3584 scsi_vhci_lun_t *vlun; 3585 struct scsi_vhci *vhci; 3586 scsi_vhci_swarg_t *swarg; 3587 char *path; 3588 3589 ASSERT(vpkt != NULL); 3590 tpkt = vpkt->vpkt_tgt_pkt; 3591 ASSERT(tpkt != NULL); 3592 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(vpkt->vpkt_path); 3593 ASSERT(svp != NULL); 3594 vlun = svp->svp_svl; 3595 ASSERT(vlun != NULL); 3596 ASSERT(VHCI_LUN_IS_HELD(vlun)); 3597 3598 vhci = ADDR2VHCI(&tpkt->pkt_address); 3599 3600 if (fostat == SCSI_SENSE_INACTIVE) { 3601 VHCI_DEBUG(1, (CE_NOTE, NULL, "!Failover " 3602 "detected for %s; updating path states...\n", 3603 vlun->svl_lun_wwn)); 3604 /* 3605 * set the vlun flag to indicate to the task that the target 3606 * port group needs updating 3607 */ 3608 vlun->svl_flags |= VLUN_UPDATE_TPG; 3609 (void) taskq_dispatch(vhci->vhci_update_pathstates_taskq, 3610 vhci_update_pathstates, (void *)vlun, KM_SLEEP); 3611 } else { 3612 path = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3613 vhci_log(CE_NOTE, ddi_get_parent(vlun->svl_dip), 3614 "!%s (%s%d): Waiting for externally initiated failover " 3615 "to complete", ddi_pathname(vlun->svl_dip, path), 3616 ddi_driver_name(vlun->svl_dip), 3617 ddi_get_instance(vlun->svl_dip)); 3618 kmem_free(path, MAXPATHLEN); 3619 swarg = kmem_alloc(sizeof (*swarg), KM_NOSLEEP); 3620 if (swarg == NULL) { 3621 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_handle_ext_fo: " 3622 "request packet allocation for %s failed....\n", 3623 vlun->svl_lun_wwn)); 3624 VHCI_RELEASE_LUN(vlun); 3625 return (PKT_RETURN); 3626 } 3627 swarg->svs_svp = svp; 3628 swarg->svs_tos = gethrtime(); 3629 swarg->svs_pi = vpkt->vpkt_path; 3630 swarg->svs_release_lun = 0; 3631 swarg->svs_done = 0; 3632 /* 3633 * place a hold on the path...we don't want it to 3634 * vanish while scsi_watch is in progress 3635 */ 3636 mdi_hold_path(vpkt->vpkt_path); 3637 svp->svp_sw_token = scsi_watch_request_submit(svp->svp_psd, 3638 VHCI_FOWATCH_INTERVAL, SENSE_LENGTH, vhci_efo_watch_cb, 3639 (caddr_t)swarg); 3640 } 3641 return (BUSY_RETURN); 3642 } 3643 3644 /* 3645 * vhci_efo_watch_cb: 3646 * Callback from scsi_watch request to check the failover status. 3647 * Completion is either due to successful failover or timeout. 3648 * Upon successful completion, vhci_update_path_states is called. 3649 * For timeout condition, vhci_efo_done is called. 3650 * Always returns 0 to scsi_watch to keep retrying till vhci_efo_done 3651 * terminates this request properly in a separate thread. 3652 */ 3653 3654 static int 3655 vhci_efo_watch_cb(caddr_t arg, struct scsi_watch_result *resultp) 3656 { 3657 struct scsi_status *statusp = resultp->statusp; 3658 uint8_t *sensep = (uint8_t *)resultp->sensep; 3659 struct scsi_pkt *pkt = resultp->pkt; 3660 scsi_vhci_swarg_t *swarg; 3661 scsi_vhci_priv_t *svp; 3662 scsi_vhci_lun_t *vlun; 3663 struct scsi_vhci *vhci; 3664 dev_info_t *vdip; 3665 int rval, updt_paths; 3666 3667 swarg = (scsi_vhci_swarg_t *)(uintptr_t)arg; 3668 svp = swarg->svs_svp; 3669 if (swarg->svs_done) { 3670 /* 3671 * Already completed failover or timedout. 3672 * Waiting for vhci_efo_done to terminate this scsi_watch. 3673 */ 3674 return (0); 3675 } 3676 3677 ASSERT(svp != NULL); 3678 vlun = svp->svp_svl; 3679 ASSERT(vlun != NULL); 3680 ASSERT(VHCI_LUN_IS_HELD(vlun)); 3681 vlun->svl_efo_update_path = 0; 3682 vdip = ddi_get_parent(vlun->svl_dip); 3683 vhci = ddi_get_soft_state(vhci_softstate, 3684 ddi_get_instance(vdip)); 3685 3686 updt_paths = 0; 3687 3688 if (pkt->pkt_reason != CMD_CMPLT) { 3689 if ((gethrtime() - swarg->svs_tos) >= VHCI_EXTFO_TIMEOUT) { 3690 swarg->svs_release_lun = 1; 3691 goto done; 3692 } 3693 return (0); 3694 } 3695 if (*((unsigned char *)statusp) == STATUS_CHECK) { 3696 rval = vlun->svl_fops->sfo_analyze_sense(svp->svp_psd, sensep, 3697 vlun->svl_fops_ctpriv); 3698 switch (rval) { 3699 /* 3700 * Only update path states in case path is definitely 3701 * inactive, or no failover occurred. For all other 3702 * check conditions continue pinging. A unexpected 3703 * check condition shouldn't cause pinging to complete 3704 * prematurely. 3705 */ 3706 case SCSI_SENSE_INACTIVE: 3707 case SCSI_SENSE_NOFAILOVER: 3708 updt_paths = 1; 3709 break; 3710 default: 3711 if ((gethrtime() - swarg->svs_tos) 3712 >= VHCI_EXTFO_TIMEOUT) { 3713 swarg->svs_release_lun = 1; 3714 goto done; 3715 } 3716 return (0); 3717 } 3718 } else if (*((unsigned char *)statusp) == 3719 STATUS_RESERVATION_CONFLICT) { 3720 updt_paths = 1; 3721 } else if ((*((unsigned char *)statusp)) & 3722 (STATUS_BUSY | STATUS_QFULL)) { 3723 return (0); 3724 } 3725 if ((*((unsigned char *)statusp) == STATUS_GOOD) || 3726 (updt_paths == 1)) { 3727 /* 3728 * we got here because we had detected an 3729 * externally initiated failover; things 3730 * have settled down now, so let's 3731 * start up a task to update the 3732 * path states and target port group 3733 */ 3734 vlun->svl_efo_update_path = 1; 3735 swarg->svs_done = 1; 3736 vlun->svl_swarg = swarg; 3737 vlun->svl_flags |= VLUN_UPDATE_TPG; 3738 (void) taskq_dispatch(vhci->vhci_update_pathstates_taskq, 3739 vhci_update_pathstates, (void *)vlun, 3740 KM_SLEEP); 3741 return (0); 3742 } 3743 if ((gethrtime() - swarg->svs_tos) >= VHCI_EXTFO_TIMEOUT) { 3744 swarg->svs_release_lun = 1; 3745 goto done; 3746 } 3747 return (0); 3748 done: 3749 swarg->svs_done = 1; 3750 (void) taskq_dispatch(vhci->vhci_taskq, 3751 vhci_efo_done, (void *)swarg, KM_SLEEP); 3752 return (0); 3753 } 3754 3755 /* 3756 * vhci_efo_done: 3757 * cleanly terminates scsi_watch and free up resources. 3758 * Called as taskq function in vhci_efo_watch_cb for EFO timeout condition 3759 * or by vhci_update_path_states invoked during external initiated 3760 * failover completion. 3761 */ 3762 static void 3763 vhci_efo_done(void *arg) 3764 { 3765 scsi_vhci_lun_t *vlun; 3766 scsi_vhci_swarg_t *swarg = (scsi_vhci_swarg_t *)arg; 3767 scsi_vhci_priv_t *svp = swarg->svs_svp; 3768 ASSERT(svp); 3769 3770 vlun = svp->svp_svl; 3771 ASSERT(vlun); 3772 3773 /* Wait for clean termination of scsi_watch */ 3774 (void) scsi_watch_request_terminate(svp->svp_sw_token, 3775 SCSI_WATCH_TERMINATE_ALL_WAIT); 3776 svp->svp_sw_token = NULL; 3777 3778 /* release path and freeup resources to indicate failover completion */ 3779 mdi_rele_path(swarg->svs_pi); 3780 if (swarg->svs_release_lun) { 3781 VHCI_RELEASE_LUN(vlun); 3782 } 3783 kmem_free((void *)swarg, sizeof (*swarg)); 3784 } 3785 3786 /* 3787 * Update the path states 3788 * vlun should be HELD when this is invoked. 3789 * Calls vhci_efo_done to cleanup resources allocated for EFO. 3790 */ 3791 void 3792 vhci_update_pathstates(void *arg) 3793 { 3794 mdi_pathinfo_t *pip, *npip; 3795 dev_info_t *dip; 3796 struct scsi_failover_ops *fo; 3797 struct scsi_vhci_priv *svp; 3798 struct scsi_device *psd; 3799 struct scsi_path_opinfo opinfo; 3800 char *pclass, *tptr; 3801 struct scsi_vhci_lun *vlun = (struct scsi_vhci_lun *)arg; 3802 int sps; /* mdi_select_path() status */ 3803 char *cpath; 3804 struct scsi_vhci *vhci; 3805 struct scsi_pkt *pkt; 3806 struct buf *bp; 3807 struct scsi_vhci_priv *svp_conflict = NULL; 3808 3809 ASSERT(VHCI_LUN_IS_HELD(vlun)); 3810 dip = vlun->svl_dip; 3811 pip = npip = NULL; 3812 3813 vhci = ddi_get_soft_state(vhci_softstate, 3814 ddi_get_instance(ddi_get_parent(dip))); 3815 3816 sps = mdi_select_path(dip, NULL, (MDI_SELECT_ONLINE_PATH | 3817 MDI_SELECT_STANDBY_PATH | MDI_SELECT_NO_PREFERRED), NULL, &npip); 3818 if ((npip == NULL) || (sps != MDI_SUCCESS)) { 3819 goto done; 3820 } 3821 3822 fo = vlun->svl_fops; 3823 do { 3824 pip = npip; 3825 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 3826 psd = svp->svp_psd; 3827 if (fo->sfo_path_get_opinfo(psd, &opinfo, 3828 vlun->svl_fops_ctpriv) != 0) { 3829 sps = mdi_select_path(dip, NULL, 3830 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH | 3831 MDI_SELECT_NO_PREFERRED), pip, &npip); 3832 mdi_rele_path(pip); 3833 continue; 3834 } 3835 3836 if (mdi_prop_lookup_string(pip, "path-class", &pclass) != 3837 MDI_SUCCESS) { 3838 VHCI_DEBUG(1, (CE_NOTE, NULL, 3839 "!vhci_update_pathstates: prop lookup failed for " 3840 "path 0x%p\n", (void *)pip)); 3841 sps = mdi_select_path(dip, NULL, 3842 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH | 3843 MDI_SELECT_NO_PREFERRED), pip, &npip); 3844 mdi_rele_path(pip); 3845 continue; 3846 } 3847 3848 /* 3849 * Need to update the "path-class" property 3850 * value in the device tree if different 3851 * from the existing value. 3852 */ 3853 if (strcmp(pclass, opinfo.opinfo_path_attr) != 0) { 3854 (void) mdi_prop_update_string(pip, "path-class", 3855 opinfo.opinfo_path_attr); 3856 } 3857 3858 /* 3859 * Only change the state if needed. i.e. Don't call 3860 * mdi_pi_set_state to ONLINE a path if its already 3861 * ONLINE. Same for STANDBY paths. 3862 */ 3863 3864 if ((opinfo.opinfo_path_state == SCSI_PATH_ACTIVE || 3865 opinfo.opinfo_path_state == SCSI_PATH_ACTIVE_NONOPT)) { 3866 if (!(MDI_PI_IS_ONLINE(pip))) { 3867 VHCI_DEBUG(1, (CE_NOTE, NULL, 3868 "!vhci_update_pathstates: marking path" 3869 " 0x%p as ONLINE\n", (void *)pip)); 3870 cpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3871 vhci_log(CE_NOTE, ddi_get_parent(dip), "!%s " 3872 "(%s%d): path %s " 3873 "is now ONLINE because of " 3874 "an externally initiated failover", 3875 ddi_pathname(dip, cpath), 3876 ddi_driver_name(dip), 3877 ddi_get_instance(dip), 3878 mdi_pi_spathname(pip)); 3879 kmem_free(cpath, MAXPATHLEN); 3880 mdi_pi_set_state(pip, 3881 MDI_PATHINFO_STATE_ONLINE); 3882 mdi_pi_set_preferred(pip, 3883 opinfo.opinfo_preferred); 3884 tptr = kmem_alloc(strlen 3885 (opinfo.opinfo_path_attr)+1, KM_SLEEP); 3886 (void) strlcpy(tptr, opinfo.opinfo_path_attr, 3887 (strlen(opinfo.opinfo_path_attr)+1)); 3888 mutex_enter(&vlun->svl_mutex); 3889 if (vlun->svl_active_pclass != NULL) { 3890 kmem_free(vlun->svl_active_pclass, 3891 strlen(vlun->svl_active_pclass)+1); 3892 } 3893 vlun->svl_active_pclass = tptr; 3894 if (vlun->svl_waiting_for_activepath) { 3895 vlun->svl_waiting_for_activepath = 0; 3896 } 3897 mutex_exit(&vlun->svl_mutex); 3898 } else if (MDI_PI_IS_ONLINE(pip)) { 3899 if (strcmp(pclass, opinfo.opinfo_path_attr) 3900 != 0) { 3901 mdi_pi_set_preferred(pip, 3902 opinfo.opinfo_preferred); 3903 mutex_enter(&vlun->svl_mutex); 3904 if (vlun->svl_active_pclass == NULL || 3905 strcmp(opinfo.opinfo_path_attr, 3906 vlun->svl_active_pclass) != 0) { 3907 mutex_exit(&vlun->svl_mutex); 3908 tptr = kmem_alloc(strlen 3909 (opinfo.opinfo_path_attr)+1, 3910 KM_SLEEP); 3911 (void) strlcpy(tptr, 3912 opinfo.opinfo_path_attr, 3913 (strlen 3914 (opinfo.opinfo_path_attr) 3915 +1)); 3916 mutex_enter(&vlun->svl_mutex); 3917 } else { 3918 /* 3919 * No need to update 3920 * svl_active_pclass 3921 */ 3922 tptr = NULL; 3923 mutex_exit(&vlun->svl_mutex); 3924 } 3925 if (tptr) { 3926 if (vlun->svl_active_pclass 3927 != NULL) { 3928 kmem_free(vlun-> 3929 svl_active_pclass, 3930 strlen(vlun-> 3931 svl_active_pclass) 3932 +1); 3933 } 3934 vlun->svl_active_pclass = tptr; 3935 mutex_exit(&vlun->svl_mutex); 3936 } 3937 } 3938 } 3939 3940 /* Check for Reservation Conflict */ 3941 bp = scsi_alloc_consistent_buf( 3942 &svp->svp_psd->sd_address, (struct buf *)NULL, 3943 DEV_BSIZE, B_READ, NULL, NULL); 3944 if (!bp) { 3945 VHCI_DEBUG(1, (CE_NOTE, NULL, 3946 "!vhci_update_pathstates: No resources " 3947 "(buf)\n")); 3948 mdi_rele_path(pip); 3949 goto done; 3950 } 3951 pkt = scsi_init_pkt(&svp->svp_psd->sd_address, NULL, bp, 3952 CDB_GROUP1, sizeof (struct scsi_arq_status), 0, 3953 PKT_CONSISTENT, NULL, NULL); 3954 if (pkt) { 3955 (void) scsi_setup_cdb((union scsi_cdb *) 3956 (uintptr_t)pkt->pkt_cdbp, SCMD_READ, 1, 1, 3957 0); 3958 pkt->pkt_time = 3*30; 3959 pkt->pkt_flags = FLAG_NOINTR; 3960 pkt->pkt_path_instance = 3961 mdi_pi_get_path_instance(pip); 3962 3963 if ((scsi_transport(pkt) == TRAN_ACCEPT) && 3964 (pkt->pkt_reason == CMD_CMPLT) && 3965 (SCBP_C(pkt) == 3966 STATUS_RESERVATION_CONFLICT)) { 3967 VHCI_DEBUG(1, (CE_NOTE, NULL, 3968 "!vhci_update_pathstates: reserv. " 3969 "conflict to be resolved on 0x%p\n", 3970 (void *)pip)); 3971 svp_conflict = svp; 3972 } 3973 scsi_destroy_pkt(pkt); 3974 } 3975 scsi_free_consistent_buf(bp); 3976 } else if ((opinfo.opinfo_path_state == SCSI_PATH_INACTIVE) && 3977 !(MDI_PI_IS_STANDBY(pip))) { 3978 VHCI_DEBUG(1, (CE_NOTE, NULL, 3979 "!vhci_update_pathstates: marking path" 3980 " 0x%p as STANDBY\n", (void *)pip)); 3981 cpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3982 vhci_log(CE_NOTE, ddi_get_parent(dip), "!%s " 3983 "(%s%d): path %s " 3984 "is now STANDBY because of " 3985 "an externally initiated failover", 3986 ddi_pathname(dip, cpath), 3987 ddi_driver_name(dip), 3988 ddi_get_instance(dip), 3989 mdi_pi_spathname(pip)); 3990 kmem_free(cpath, MAXPATHLEN); 3991 mdi_pi_set_state(pip, 3992 MDI_PATHINFO_STATE_STANDBY); 3993 mdi_pi_set_preferred(pip, 3994 opinfo.opinfo_preferred); 3995 mutex_enter(&vlun->svl_mutex); 3996 if (vlun->svl_active_pclass != NULL) { 3997 if (strcmp(vlun->svl_active_pclass, 3998 opinfo.opinfo_path_attr) == 0) { 3999 kmem_free(vlun-> 4000 svl_active_pclass, 4001 strlen(vlun-> 4002 svl_active_pclass)+1); 4003 vlun->svl_active_pclass = NULL; 4004 } 4005 } 4006 mutex_exit(&vlun->svl_mutex); 4007 } 4008 (void) mdi_prop_free(pclass); 4009 sps = mdi_select_path(dip, NULL, 4010 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH | 4011 MDI_SELECT_NO_PREFERRED), pip, &npip); 4012 mdi_rele_path(pip); 4013 4014 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 4015 4016 /* 4017 * Check to see if this vlun has an active SCSI-II RESERVE. If so 4018 * clear the reservation by sending a reset, so the host doesn't 4019 * receive a reservation conflict. The reset has to be sent via a 4020 * working path. Let's use a path referred to by svp_conflict as it 4021 * should be working. 4022 * Reset VLUN_RESERVE_ACTIVE_FLG for this vlun. Also notify ssd 4023 * of the reset, explicitly. 4024 */ 4025 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 4026 if (svp_conflict && (vlun->svl_xlf_capable == 0)) { 4027 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_update_pathstates:" 4028 " sending recovery reset on 0x%p, path_state: %x", 4029 svp_conflict->svp_psd->sd_private, 4030 mdi_pi_get_state((mdi_pathinfo_t *) 4031 svp_conflict->svp_psd->sd_private))); 4032 4033 (void) vhci_recovery_reset(vlun, 4034 &svp_conflict->svp_psd->sd_address, FALSE, 4035 VHCI_DEPTH_TARGET); 4036 } 4037 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 4038 mutex_enter(&vhci->vhci_mutex); 4039 scsi_hba_reset_notify_callback(&vhci->vhci_mutex, 4040 &vhci->vhci_reset_notify_listf); 4041 mutex_exit(&vhci->vhci_mutex); 4042 } 4043 if (vlun->svl_flags & VLUN_UPDATE_TPG) { 4044 /* 4045 * Update the AccessState of related MP-API TPGs 4046 */ 4047 (void) vhci_mpapi_update_tpg_acc_state_for_lu(vhci, vlun); 4048 vlun->svl_flags &= ~VLUN_UPDATE_TPG; 4049 } 4050 done: 4051 if (vlun->svl_efo_update_path) { 4052 vlun->svl_efo_update_path = 0; 4053 vhci_efo_done(vlun->svl_swarg); 4054 vlun->svl_swarg = 0; 4055 } 4056 VHCI_RELEASE_LUN(vlun); 4057 } 4058 4059 /* ARGSUSED */ 4060 static int 4061 vhci_pathinfo_init(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 4062 { 4063 scsi_hba_tran_t *hba = NULL; 4064 struct scsi_device *psd = NULL; 4065 scsi_vhci_lun_t *vlun = NULL; 4066 dev_info_t *pdip = NULL; 4067 dev_info_t *tgt_dip; 4068 struct scsi_vhci *vhci; 4069 char *guid; 4070 scsi_vhci_priv_t *svp = NULL; 4071 int rval = MDI_FAILURE; 4072 int vlun_alloced = 0; 4073 4074 ASSERT(vdip != NULL); 4075 ASSERT(pip != NULL); 4076 4077 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 4078 ASSERT(vhci != NULL); 4079 4080 pdip = mdi_pi_get_phci(pip); 4081 ASSERT(pdip != NULL); 4082 4083 hba = ddi_get_driver_private(pdip); 4084 ASSERT(hba != NULL); 4085 4086 tgt_dip = mdi_pi_get_client(pip); 4087 ASSERT(tgt_dip != NULL); 4088 4089 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, PROPFLAGS, 4090 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 4091 VHCI_DEBUG(1, (CE_WARN, NULL, 4092 "vhci_pathinfo_init: lun guid property failed")); 4093 goto failure; 4094 } 4095 4096 vlun = vhci_lun_lookup_alloc(tgt_dip, guid, &vlun_alloced); 4097 ddi_prop_free(guid); 4098 4099 vlun->svl_dip = tgt_dip; 4100 4101 svp = kmem_zalloc(sizeof (*svp), KM_SLEEP); 4102 svp->svp_svl = vlun; 4103 4104 /* 4105 * Initialize svl_lb_policy_save only for newly allocated vlun. Writing 4106 * to svl_lb_policy_save later could accidentally overwrite saved lb 4107 * policy. 4108 */ 4109 if (vlun_alloced) { 4110 vlun->svl_lb_policy_save = mdi_get_lb_policy(tgt_dip); 4111 } 4112 4113 mutex_init(&svp->svp_mutex, NULL, MUTEX_DRIVER, NULL); 4114 cv_init(&svp->svp_cv, NULL, CV_DRIVER, NULL); 4115 4116 psd = kmem_zalloc(sizeof (*psd), KM_SLEEP); 4117 mutex_init(&psd->sd_mutex, NULL, MUTEX_DRIVER, NULL); 4118 4119 if (hba->tran_hba_flags & SCSI_HBA_ADDR_COMPLEX) { 4120 /* 4121 * For a SCSI_HBA_ADDR_COMPLEX transport we store a pointer to 4122 * scsi_device in the scsi_address structure. This allows an 4123 * an HBA driver to find its scsi_device(9S) and 4124 * per-scsi_device(9S) HBA private data given a 4125 * scsi_address(9S) by using scsi_address_device(9F) and 4126 * scsi_device_hba_private_get(9F)). 4127 */ 4128 psd->sd_address.a.a_sd = psd; 4129 } else if (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) { 4130 /* 4131 * Clone transport structure if requested, so 4132 * Self enumerating HBAs always need to use cloning 4133 */ 4134 scsi_hba_tran_t *clone = 4135 kmem_alloc(sizeof (scsi_hba_tran_t), KM_SLEEP); 4136 bcopy(hba, clone, sizeof (scsi_hba_tran_t)); 4137 hba = clone; 4138 hba->tran_sd = psd; 4139 } else { 4140 /* 4141 * SPI pHCI unit-address. If we ever need to support this 4142 * we could set a.spi.a_target/a.spi.a_lun based on pathinfo 4143 * node unit-address properties. For now we fail... 4144 */ 4145 goto failure; 4146 } 4147 4148 psd->sd_dev = tgt_dip; 4149 psd->sd_address.a_hba_tran = hba; 4150 4151 /* 4152 * Mark scsi_device as being associated with a pathinfo node. For 4153 * a scsi_device structure associated with a devinfo node, 4154 * scsi_ctlops_initchild sets this field to NULL. 4155 */ 4156 psd->sd_pathinfo = pip; 4157 4158 /* 4159 * LEGACY: sd_private: set for older mpxio-capable pHCI drivers with 4160 * too much scsi_vhci/mdi/ndi knowledge. Remove this code when all 4161 * mpxio-capable pHCI drivers use SCSA enumeration services (or at 4162 * least have been changed to use sd_pathinfo instead). 4163 */ 4164 psd->sd_private = (caddr_t)pip; 4165 4166 /* See scsi_hba.c for info on sd_tran_safe kludge */ 4167 psd->sd_tran_safe = hba; 4168 4169 svp->svp_psd = psd; 4170 mdi_pi_set_vhci_private(pip, (caddr_t)svp); 4171 4172 /* 4173 * call hba's target init entry point if it exists 4174 */ 4175 if (hba->tran_tgt_init != NULL) { 4176 psd->sd_tran_tgt_free_done = 0; 4177 if ((rval = (*hba->tran_tgt_init)(pdip, tgt_dip, 4178 hba, psd)) != DDI_SUCCESS) { 4179 VHCI_DEBUG(1, (CE_WARN, pdip, 4180 "!vhci_pathinfo_init: tran_tgt_init failed for " 4181 "path=0x%p rval=%x", (void *)pip, rval)); 4182 goto failure; 4183 } 4184 } 4185 4186 svp->svp_new_path = 1; 4187 4188 VHCI_DEBUG(4, (CE_NOTE, NULL, "!vhci_pathinfo_init: path:%p\n", 4189 (void *)pip)); 4190 return (MDI_SUCCESS); 4191 4192 failure: 4193 if (psd) { 4194 mutex_destroy(&psd->sd_mutex); 4195 kmem_free(psd, sizeof (*psd)); 4196 } 4197 if (svp) { 4198 mdi_pi_set_vhci_private(pip, NULL); 4199 mutex_destroy(&svp->svp_mutex); 4200 cv_destroy(&svp->svp_cv); 4201 kmem_free(svp, sizeof (*svp)); 4202 } 4203 if (hba && (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE)) 4204 kmem_free(hba, sizeof (scsi_hba_tran_t)); 4205 4206 if (vlun_alloced) 4207 vhci_lun_free(vlun, NULL); 4208 4209 return (rval); 4210 } 4211 4212 /* ARGSUSED */ 4213 static int 4214 vhci_pathinfo_uninit(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 4215 { 4216 scsi_hba_tran_t *hba = NULL; 4217 struct scsi_device *psd = NULL; 4218 dev_info_t *pdip = NULL; 4219 dev_info_t *cdip = NULL; 4220 scsi_vhci_priv_t *svp = NULL; 4221 4222 ASSERT(vdip != NULL); 4223 ASSERT(pip != NULL); 4224 4225 pdip = mdi_pi_get_phci(pip); 4226 ASSERT(pdip != NULL); 4227 4228 cdip = mdi_pi_get_client(pip); 4229 ASSERT(cdip != NULL); 4230 4231 hba = ddi_get_driver_private(pdip); 4232 ASSERT(hba != NULL); 4233 4234 vhci_mpapi_set_path_state(vdip, pip, MP_DRVR_PATH_STATE_UNINIT); 4235 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 4236 if (svp == NULL) { 4237 /* path already freed. Nothing to do. */ 4238 return (MDI_SUCCESS); 4239 } 4240 4241 psd = svp->svp_psd; 4242 ASSERT(psd != NULL); 4243 4244 if (hba->tran_hba_flags & SCSI_HBA_ADDR_COMPLEX) { 4245 /* Verify plumbing */ 4246 ASSERT(psd->sd_address.a_hba_tran == hba); 4247 ASSERT(psd->sd_address.a.a_sd == psd); 4248 } else if (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) { 4249 /* Switch to cloned scsi_hba_tran(9S) structure */ 4250 hba = psd->sd_address.a_hba_tran; 4251 ASSERT(hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE); 4252 ASSERT(hba->tran_sd == psd); 4253 } 4254 4255 if ((hba->tran_tgt_free != NULL) && !psd->sd_tran_tgt_free_done) { 4256 (*hba->tran_tgt_free) (pdip, cdip, hba, psd); 4257 psd->sd_tran_tgt_free_done = 1; 4258 } 4259 mutex_destroy(&psd->sd_mutex); 4260 if (hba->tran_hba_flags & SCSI_HBA_TRAN_CLONE) { 4261 kmem_free(hba, sizeof (*hba)); 4262 } 4263 4264 mdi_pi_set_vhci_private(pip, NULL); 4265 4266 /* 4267 * Free the pathinfo related scsi_device inquiry data. Note that this 4268 * matches what happens for scsi_hba.c devinfo case at uninitchild time. 4269 */ 4270 if (psd->sd_inq) 4271 kmem_free((caddr_t)psd->sd_inq, sizeof (struct scsi_inquiry)); 4272 kmem_free((caddr_t)psd, sizeof (*psd)); 4273 4274 mutex_destroy(&svp->svp_mutex); 4275 cv_destroy(&svp->svp_cv); 4276 kmem_free((caddr_t)svp, sizeof (*svp)); 4277 4278 VHCI_DEBUG(4, (CE_NOTE, NULL, "!vhci_pathinfo_uninit: path=0x%p\n", 4279 (void *)pip)); 4280 return (MDI_SUCCESS); 4281 } 4282 4283 /* ARGSUSED */ 4284 static int 4285 vhci_pathinfo_state_change(dev_info_t *vdip, mdi_pathinfo_t *pip, 4286 mdi_pathinfo_state_t state, uint32_t ext_state, int flags) 4287 { 4288 int rval = MDI_SUCCESS; 4289 scsi_vhci_priv_t *svp; 4290 scsi_vhci_lun_t *vlun; 4291 int held; 4292 int op = (flags & 0xf00) >> 8; 4293 struct scsi_vhci *vhci; 4294 4295 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 4296 4297 if (flags & MDI_EXT_STATE_CHANGE) { 4298 /* 4299 * We do not want to issue any commands down the path in case 4300 * sync flag is set. Lower layers might not be ready to accept 4301 * any I/O commands. 4302 */ 4303 if (op == DRIVER_DISABLE) 4304 return (MDI_SUCCESS); 4305 4306 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 4307 if (svp == NULL) { 4308 return (MDI_FAILURE); 4309 } 4310 vlun = svp->svp_svl; 4311 4312 if (flags & MDI_BEFORE_STATE_CHANGE) { 4313 /* 4314 * Hold the LUN. 4315 */ 4316 VHCI_HOLD_LUN(vlun, VH_SLEEP, held); 4317 if (flags & MDI_DISABLE_OP) { 4318 /* 4319 * Issue scsi reset if it happens to be 4320 * reserved path. 4321 */ 4322 if (vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 4323 /* 4324 * if reservation pending on 4325 * this path, dont' mark the 4326 * path busy 4327 */ 4328 if (op == DRIVER_DISABLE_TRANSIENT) { 4329 VHCI_DEBUG(1, (CE_NOTE, NULL, 4330 "!vhci_pathinfo" 4331 "_state_change (pip:%p): " 4332 " reservation: fail busy\n", 4333 (void *)pip)); 4334 return (MDI_FAILURE); 4335 } 4336 if (pip == vlun->svl_resrv_pip) { 4337 if (vhci_recovery_reset( 4338 svp->svp_svl, 4339 &svp->svp_psd->sd_address, 4340 TRUE, 4341 VHCI_DEPTH_TARGET) == 0) { 4342 VHCI_DEBUG(1, 4343 (CE_NOTE, NULL, 4344 "!vhci_pathinfo" 4345 "_state_change " 4346 " (pip:%p): " 4347 "reset failed, " 4348 "give up!\n", 4349 (void *)pip)); 4350 } 4351 vlun->svl_flags &= 4352 ~VLUN_RESERVE_ACTIVE_FLG; 4353 } 4354 } 4355 } else if (flags & MDI_ENABLE_OP) { 4356 if (((vhci->vhci_conf_flags & 4357 VHCI_CONF_FLAGS_AUTO_FAILBACK) == 4358 VHCI_CONF_FLAGS_AUTO_FAILBACK) && 4359 MDI_PI_IS_USER_DISABLE(pip) && 4360 MDI_PI_IS_STANDBY(pip)) { 4361 struct scsi_failover_ops *fo; 4362 char *best_pclass, *pclass = NULL; 4363 int best_class, rv; 4364 /* 4365 * Failback if enabling a standby path 4366 * and it is the primary class or 4367 * preferred class 4368 */ 4369 best_class = mdi_pi_get_preferred(pip); 4370 if (best_class == 0) { 4371 /* 4372 * if not preferred - compare 4373 * path-class with class 4374 */ 4375 fo = vlun->svl_fops; 4376 (void) fo->sfo_pathclass_next( 4377 NULL, &best_pclass, 4378 vlun->svl_fops_ctpriv); 4379 pclass = NULL; 4380 rv = mdi_prop_lookup_string(pip, 4381 "path-class", &pclass); 4382 if (rv != MDI_SUCCESS || 4383 pclass == NULL) { 4384 vhci_log(CE_NOTE, vdip, 4385 "!path-class " 4386 " lookup " 4387 "failed. rv: %d" 4388 "class: %p", rv, 4389 (void *)pclass); 4390 } else if (strncmp(pclass, 4391 best_pclass, 4392 strlen(best_pclass)) == 0) { 4393 best_class = 1; 4394 } 4395 if (rv == MDI_SUCCESS && 4396 pclass != NULL) { 4397 rv = mdi_prop_free( 4398 pclass); 4399 if (rv != 4400 DDI_PROP_SUCCESS) { 4401 vhci_log( 4402 CE_NOTE, 4403 vdip, 4404 "!path-" 4405 "class" 4406 " free" 4407 " failed" 4408 " rv: %d" 4409 " class: " 4410 "%p", 4411 rv, 4412 (void *) 4413 pclass); 4414 } 4415 } 4416 } 4417 if (best_class == 1) { 4418 VHCI_DEBUG(1, (CE_NOTE, NULL, 4419 "preferred path: %p " 4420 "USER_DISABLE->USER_ENABLE " 4421 "transition for lun %s\n", 4422 (void *)pip, 4423 vlun->svl_lun_wwn)); 4424 (void) taskq_dispatch( 4425 vhci->vhci_taskq, 4426 vhci_initiate_auto_failback, 4427 (void *) vlun, KM_SLEEP); 4428 } 4429 } 4430 /* 4431 * if PGR is active, revalidate key and 4432 * register on this path also, if key is 4433 * still valid 4434 */ 4435 sema_p(&vlun->svl_pgr_sema); 4436 if (vlun->svl_pgr_active) 4437 (void) 4438 vhci_pgr_validate_and_register(svp); 4439 sema_v(&vlun->svl_pgr_sema); 4440 /* 4441 * Inform target driver about any 4442 * reservations to be reinstated if target 4443 * has dropped reservation during the busy 4444 * period. 4445 */ 4446 mutex_enter(&vhci->vhci_mutex); 4447 scsi_hba_reset_notify_callback( 4448 &vhci->vhci_mutex, 4449 &vhci->vhci_reset_notify_listf); 4450 mutex_exit(&vhci->vhci_mutex); 4451 } 4452 } 4453 if (flags & MDI_AFTER_STATE_CHANGE) { 4454 if (flags & MDI_ENABLE_OP) { 4455 mutex_enter(&vhci_global_mutex); 4456 cv_broadcast(&vhci_cv); 4457 mutex_exit(&vhci_global_mutex); 4458 } 4459 if (vlun->svl_setcap_done) { 4460 (void) vhci_pHCI_cap(&svp->svp_psd->sd_address, 4461 "sector-size", vlun->svl_sector_size, 4462 1, pip); 4463 } 4464 4465 /* 4466 * Release the LUN 4467 */ 4468 VHCI_RELEASE_LUN(vlun); 4469 4470 /* 4471 * Path transition is complete. 4472 * Run callback to indicate target driver to 4473 * retry to prevent IO starvation. 4474 */ 4475 if (scsi_callback_id != 0) { 4476 ddi_run_callback(&scsi_callback_id); 4477 } 4478 } 4479 } else { 4480 switch (state) { 4481 case MDI_PATHINFO_STATE_ONLINE: 4482 rval = vhci_pathinfo_online(vdip, pip, flags); 4483 break; 4484 4485 case MDI_PATHINFO_STATE_OFFLINE: 4486 rval = vhci_pathinfo_offline(vdip, pip, flags); 4487 break; 4488 4489 default: 4490 break; 4491 } 4492 /* 4493 * Path transition is complete. 4494 * Run callback to indicate target driver to 4495 * retry to prevent IO starvation. 4496 */ 4497 if ((rval == MDI_SUCCESS) && (scsi_callback_id != 0)) { 4498 ddi_run_callback(&scsi_callback_id); 4499 } 4500 return (rval); 4501 } 4502 4503 return (MDI_SUCCESS); 4504 } 4505 4506 /* 4507 * Parse the mpxio load balancing options. The datanameptr 4508 * will point to a string containing the load-balance-options value. 4509 * The load-balance-options value will be a property that 4510 * defines the load-balance algorithm and any arguments to that 4511 * algorithm. 4512 * For example: 4513 * device-type-mpxio-options-list= 4514 * "device-type=SUN SENA", "load-balance-options=logical-block-options" 4515 * "device-type=SUN SE6920", "round-robin-options"; 4516 * logical-block-options="load-balance=logical-block", "region-size=15"; 4517 * round-robin-options="load-balance=round-robin"; 4518 * 4519 * If the load-balance is not defined the load balance algorithm will 4520 * default to the global setting. There will be default values assigned 4521 * to the arguments (region-size=18) and if an argument is one 4522 * that is not known, it will be ignored. 4523 */ 4524 static void 4525 vhci_parse_mpxio_lb_options(dev_info_t *dip, dev_info_t *cdip, 4526 caddr_t datanameptr) 4527 { 4528 char *dataptr, *next_entry; 4529 caddr_t config_list = NULL; 4530 int config_list_len = 0, list_len = 0; 4531 int region_size = -1; 4532 client_lb_t load_balance; 4533 4534 if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, datanameptr, 4535 (caddr_t)&config_list, &config_list_len) != DDI_PROP_SUCCESS) { 4536 return; 4537 } 4538 4539 list_len = config_list_len; 4540 next_entry = config_list; 4541 while (config_list_len > 0) { 4542 dataptr = next_entry; 4543 4544 if (strncmp(mdi_load_balance, dataptr, 4545 strlen(mdi_load_balance)) == 0) { 4546 /* get the load-balance scheme */ 4547 dataptr += strlen(mdi_load_balance) + 1; 4548 if (strcmp(dataptr, LOAD_BALANCE_PROP_RR) == 0) { 4549 (void) mdi_set_lb_policy(cdip, LOAD_BALANCE_RR); 4550 load_balance = LOAD_BALANCE_RR; 4551 } else if (strcmp(dataptr, 4552 LOAD_BALANCE_PROP_LBA) == 0) { 4553 (void) mdi_set_lb_policy(cdip, 4554 LOAD_BALANCE_LBA); 4555 load_balance = LOAD_BALANCE_LBA; 4556 } else if (strcmp(dataptr, 4557 LOAD_BALANCE_PROP_NONE) == 0) { 4558 (void) mdi_set_lb_policy(cdip, 4559 LOAD_BALANCE_NONE); 4560 load_balance = LOAD_BALANCE_NONE; 4561 } 4562 } else if (strncmp(dataptr, LOGICAL_BLOCK_REGION_SIZE, 4563 strlen(LOGICAL_BLOCK_REGION_SIZE)) == 0) { 4564 int i = 0; 4565 char *ptr; 4566 char *tmp; 4567 4568 tmp = dataptr + (strlen(LOGICAL_BLOCK_REGION_SIZE) + 1); 4569 /* check for numeric value */ 4570 for (ptr = tmp; i < strlen(tmp); i++, ptr++) { 4571 if (!isdigit(*ptr)) { 4572 cmn_err(CE_WARN, 4573 "Illegal region size: %s." 4574 " Setting to default value: %d", 4575 tmp, 4576 LOAD_BALANCE_DEFAULT_REGION_SIZE); 4577 region_size = 4578 LOAD_BALANCE_DEFAULT_REGION_SIZE; 4579 break; 4580 } 4581 } 4582 if (i >= strlen(tmp)) { 4583 region_size = stoi(&tmp); 4584 } 4585 (void) mdi_set_lb_region_size(cdip, region_size); 4586 } 4587 config_list_len -= (strlen(next_entry) + 1); 4588 next_entry += strlen(next_entry) + 1; 4589 } 4590 #ifdef DEBUG 4591 if ((region_size >= 0) && (load_balance != LOAD_BALANCE_LBA)) { 4592 VHCI_DEBUG(1, (CE_NOTE, dip, 4593 "!vhci_parse_mpxio_lb_options: region-size: %d" 4594 "only valid for load-balance=logical-block\n", 4595 region_size)); 4596 } 4597 #endif 4598 if ((region_size == -1) && (load_balance == LOAD_BALANCE_LBA)) { 4599 VHCI_DEBUG(1, (CE_NOTE, dip, 4600 "!vhci_parse_mpxio_lb_options: No region-size" 4601 " defined load-balance=logical-block." 4602 " Default to: %d\n", LOAD_BALANCE_DEFAULT_REGION_SIZE)); 4603 (void) mdi_set_lb_region_size(cdip, 4604 LOAD_BALANCE_DEFAULT_REGION_SIZE); 4605 } 4606 if (list_len > 0) { 4607 kmem_free(config_list, list_len); 4608 } 4609 } 4610 4611 /* 4612 * Parse the device-type-mpxio-options-list looking for the key of 4613 * "load-balance-options". If found, parse the load balancing options. 4614 * Check the comment of the vhci_get_device_type_mpxio_options() 4615 * for the device-type-mpxio-options-list. 4616 */ 4617 static void 4618 vhci_parse_mpxio_options(dev_info_t *dip, dev_info_t *cdip, 4619 caddr_t datanameptr, int list_len) 4620 { 4621 char *dataptr; 4622 int len; 4623 4624 /* 4625 * get the data list 4626 */ 4627 dataptr = datanameptr; 4628 len = 0; 4629 while (len < list_len && 4630 strncmp(dataptr, DEVICE_TYPE_STR, strlen(DEVICE_TYPE_STR)) 4631 != 0) { 4632 if (strncmp(dataptr, LOAD_BALANCE_OPTIONS, 4633 strlen(LOAD_BALANCE_OPTIONS)) == 0) { 4634 len += strlen(LOAD_BALANCE_OPTIONS) + 1; 4635 dataptr += strlen(LOAD_BALANCE_OPTIONS) + 1; 4636 vhci_parse_mpxio_lb_options(dip, cdip, dataptr); 4637 } 4638 len += strlen(dataptr) + 1; 4639 dataptr += strlen(dataptr) + 1; 4640 } 4641 } 4642 4643 /* 4644 * Check the inquriy string returned from the device with the device-type 4645 * Check for the existence of the device-type-mpxio-options-list and 4646 * if found parse the list checking for a match with the device-type 4647 * value and the inquiry string returned from the device. If a match 4648 * is found, parse the mpxio options list. The format of the 4649 * device-type-mpxio-options-list is: 4650 * device-type-mpxio-options-list= 4651 * "device-type=SUN SENA", "load-balance-options=logical-block-options" 4652 * "device-type=SUN SE6920", "round-robin-options"; 4653 * logical-block-options="load-balance=logical-block", "region-size=15"; 4654 * round-robin-options="load-balance=round-robin"; 4655 */ 4656 void 4657 vhci_get_device_type_mpxio_options(dev_info_t *dip, dev_info_t *cdip, 4658 struct scsi_device *devp) 4659 { 4660 4661 caddr_t config_list = NULL; 4662 caddr_t vidptr, datanameptr; 4663 int vidlen, dupletlen = 0; 4664 int config_list_len = 0, len; 4665 struct scsi_inquiry *inq = devp->sd_inq; 4666 4667 /* 4668 * look up the device-type-mpxio-options-list and walk thru 4669 * the list compare the vendor ids of the earlier inquiry command and 4670 * with those vids in the list if there is a match, lookup 4671 * the mpxio-options value 4672 */ 4673 if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, 4674 MPXIO_OPTIONS_LIST, 4675 (caddr_t)&config_list, &config_list_len) == DDI_PROP_SUCCESS) { 4676 4677 /* 4678 * Compare vids in each duplet - if it matches, 4679 * parse the mpxio options list. 4680 */ 4681 for (len = config_list_len, vidptr = config_list; len > 0; 4682 len -= dupletlen) { 4683 4684 dupletlen = 0; 4685 4686 if (strlen(vidptr) != 0 && 4687 strncmp(vidptr, DEVICE_TYPE_STR, 4688 strlen(DEVICE_TYPE_STR)) == 0) { 4689 /* point to next duplet */ 4690 datanameptr = vidptr + strlen(vidptr) + 1; 4691 /* add len of this duplet */ 4692 dupletlen += strlen(vidptr) + 1; 4693 /* get to device type */ 4694 vidptr += strlen(DEVICE_TYPE_STR) + 1; 4695 vidlen = strlen(vidptr); 4696 if ((vidlen != 0) && 4697 bcmp(inq->inq_vid, vidptr, vidlen) == 0) { 4698 vhci_parse_mpxio_options(dip, cdip, 4699 datanameptr, len - dupletlen); 4700 break; 4701 } 4702 /* get to next duplet */ 4703 vidptr += strlen(vidptr) + 1; 4704 } 4705 /* get to the next device-type */ 4706 while (len - dupletlen > 0 && 4707 strlen(vidptr) != 0 && 4708 strncmp(vidptr, DEVICE_TYPE_STR, 4709 strlen(DEVICE_TYPE_STR)) != 0) { 4710 dupletlen += strlen(vidptr) + 1; 4711 vidptr += strlen(vidptr) + 1; 4712 } 4713 } 4714 if (config_list_len > 0) { 4715 kmem_free(config_list, config_list_len); 4716 } 4717 } 4718 } 4719 4720 static int 4721 vhci_update_pathinfo(struct scsi_device *psd, mdi_pathinfo_t *pip, 4722 struct scsi_failover_ops *fo, 4723 scsi_vhci_lun_t *vlun, 4724 struct scsi_vhci *vhci) 4725 { 4726 struct scsi_path_opinfo opinfo; 4727 char *pclass, *best_pclass; 4728 char *resrv_pclass = NULL; 4729 int force_rereserve = 0; 4730 int update_pathinfo_done = 0; 4731 4732 if (fo->sfo_path_get_opinfo(psd, &opinfo, vlun->svl_fops_ctpriv) != 0) { 4733 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_update_pathinfo: " 4734 "Failed to get operation info for path:%p\n", (void *)pip)); 4735 return (MDI_FAILURE); 4736 } 4737 /* set the xlf capable flag in the vlun for future use */ 4738 vlun->svl_xlf_capable = opinfo.opinfo_xlf_capable; 4739 (void) mdi_prop_update_string(pip, "path-class", 4740 opinfo.opinfo_path_attr); 4741 4742 pclass = opinfo.opinfo_path_attr; 4743 if (opinfo.opinfo_path_state == SCSI_PATH_ACTIVE) { 4744 mutex_enter(&vlun->svl_mutex); 4745 if (vlun->svl_active_pclass != NULL) { 4746 if (strcmp(vlun->svl_active_pclass, pclass) != 0) { 4747 mutex_exit(&vlun->svl_mutex); 4748 /* 4749 * Externally initiated failover has happened; 4750 * force the path state to be STANDBY/ONLINE, 4751 * next IO will trigger failover and thus 4752 * sync-up the pathstates. Reason we don't 4753 * sync-up immediately by invoking 4754 * vhci_update_pathstates() is because it 4755 * needs a VHCI_HOLD_LUN() and we don't 4756 * want to block here. 4757 * 4758 * Further, if the device is an ALUA device, 4759 * then failure to exactly match 'pclass' and 4760 * 'svl_active_pclass'(as is the case here) 4761 * indicates that the currently active path 4762 * is a 'non-optimized' path - which means 4763 * that 'svl_active_pclass' needs to be 4764 * replaced with opinfo.opinfo_path_state 4765 * value. 4766 */ 4767 4768 if (SCSI_FAILOVER_IS_TPGS(vlun->svl_fops)) { 4769 char *tptr; 4770 4771 /* 4772 * The device is ALUA compliant. The 4773 * state need to be changed to online 4774 * rather than standby state which is 4775 * done typically for a asymmetric 4776 * device that is non ALUA compliant. 4777 */ 4778 mdi_pi_set_state(pip, 4779 MDI_PATHINFO_STATE_ONLINE); 4780 tptr = kmem_alloc(strlen 4781 (opinfo.opinfo_path_attr)+1, 4782 KM_SLEEP); 4783 (void) strlcpy(tptr, 4784 opinfo.opinfo_path_attr, 4785 (strlen(opinfo.opinfo_path_attr) 4786 +1)); 4787 mutex_enter(&vlun->svl_mutex); 4788 kmem_free(vlun->svl_active_pclass, 4789 strlen(vlun->svl_active_pclass)+1); 4790 vlun->svl_active_pclass = tptr; 4791 mutex_exit(&vlun->svl_mutex); 4792 } else { 4793 /* 4794 * Non ALUA device case. 4795 */ 4796 mdi_pi_set_state(pip, 4797 MDI_PATHINFO_STATE_STANDBY); 4798 } 4799 vlun->svl_fo_support = opinfo.opinfo_mode; 4800 mdi_pi_set_preferred(pip, 4801 opinfo.opinfo_preferred); 4802 update_pathinfo_done = 1; 4803 } 4804 4805 /* 4806 * Find out a class of currently reserved path if there 4807 * is any. 4808 */ 4809 if ((vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) && 4810 mdi_prop_lookup_string(vlun->svl_resrv_pip, 4811 "path-class", &resrv_pclass) != MDI_SUCCESS) { 4812 VHCI_DEBUG(1, (CE_NOTE, NULL, 4813 "!vhci_update_pathinfo: prop lookup " 4814 "failed for path 0x%p\n", 4815 (void *)vlun->svl_resrv_pip)); 4816 /* 4817 * Something is wrong with the reserved path. 4818 * We can't do much with that right here. Just 4819 * force re-reservation to another path. 4820 */ 4821 force_rereserve = 1; 4822 } 4823 4824 (void) fo->sfo_pathclass_next(NULL, &best_pclass, 4825 vlun->svl_fops_ctpriv); 4826 if ((force_rereserve == 1) || ((resrv_pclass != NULL) && 4827 (strcmp(pclass, best_pclass) == 0) && 4828 (strcmp(resrv_pclass, best_pclass) != 0))) { 4829 /* 4830 * Inform target driver that a reservation 4831 * should be reinstated because the reserved 4832 * path is not the most preferred one. 4833 */ 4834 mutex_enter(&vhci->vhci_mutex); 4835 scsi_hba_reset_notify_callback( 4836 &vhci->vhci_mutex, 4837 &vhci->vhci_reset_notify_listf); 4838 mutex_exit(&vhci->vhci_mutex); 4839 } 4840 4841 if (update_pathinfo_done == 1) { 4842 return (MDI_SUCCESS); 4843 } 4844 } else { 4845 char *tptr; 4846 4847 /* 4848 * lets release the mutex before we try to 4849 * allocate since the potential to sleep is 4850 * possible. 4851 */ 4852 mutex_exit(&vlun->svl_mutex); 4853 tptr = kmem_alloc(strlen(pclass)+1, KM_SLEEP); 4854 (void) strlcpy(tptr, pclass, (strlen(pclass)+1)); 4855 mutex_enter(&vlun->svl_mutex); 4856 vlun->svl_active_pclass = tptr; 4857 } 4858 mutex_exit(&vlun->svl_mutex); 4859 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_ONLINE); 4860 vlun->svl_waiting_for_activepath = 0; 4861 } else if (opinfo.opinfo_path_state == SCSI_PATH_ACTIVE_NONOPT) { 4862 mutex_enter(&vlun->svl_mutex); 4863 if (vlun->svl_active_pclass == NULL) { 4864 char *tptr; 4865 4866 mutex_exit(&vlun->svl_mutex); 4867 tptr = kmem_alloc(strlen(pclass)+1, KM_SLEEP); 4868 (void) strlcpy(tptr, pclass, (strlen(pclass)+1)); 4869 mutex_enter(&vlun->svl_mutex); 4870 vlun->svl_active_pclass = tptr; 4871 } 4872 mutex_exit(&vlun->svl_mutex); 4873 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_ONLINE); 4874 vlun->svl_waiting_for_activepath = 0; 4875 } else if (opinfo.opinfo_path_state == SCSI_PATH_INACTIVE) { 4876 mutex_enter(&vlun->svl_mutex); 4877 if (vlun->svl_active_pclass != NULL) { 4878 if (strcmp(vlun->svl_active_pclass, pclass) == 0) { 4879 mutex_exit(&vlun->svl_mutex); 4880 /* 4881 * externally initiated failover has happened; 4882 * force state to ONLINE (see comment above) 4883 */ 4884 mdi_pi_set_state(pip, 4885 MDI_PATHINFO_STATE_ONLINE); 4886 vlun->svl_fo_support = opinfo.opinfo_mode; 4887 mdi_pi_set_preferred(pip, 4888 opinfo.opinfo_preferred); 4889 return (MDI_SUCCESS); 4890 } 4891 } 4892 mutex_exit(&vlun->svl_mutex); 4893 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_STANDBY); 4894 4895 /* 4896 * Initiate auto-failback, if enabled, for path if path-state 4897 * is transitioning from OFFLINE->STANDBY and pathclass is the 4898 * preferred pathclass for this storage. 4899 * NOTE: In case where opinfo_path_state is SCSI_PATH_ACTIVE 4900 * (above), where the pi state is set to STANDBY, we don't 4901 * initiate auto-failback as the next IO shall take care of. 4902 * this. See comment above. 4903 */ 4904 (void) fo->sfo_pathclass_next(NULL, &best_pclass, 4905 vlun->svl_fops_ctpriv); 4906 if (((vhci->vhci_conf_flags & VHCI_CONF_FLAGS_AUTO_FAILBACK) == 4907 VHCI_CONF_FLAGS_AUTO_FAILBACK) && 4908 (strcmp(pclass, best_pclass) == 0) && 4909 ((MDI_PI_OLD_STATE(pip) == MDI_PATHINFO_STATE_OFFLINE)|| 4910 (MDI_PI_OLD_STATE(pip) == MDI_PATHINFO_STATE_INIT))) { 4911 VHCI_DEBUG(1, (CE_NOTE, NULL, "%s pathclass path: %p" 4912 " OFFLINE->STANDBY transition for lun %s\n", 4913 best_pclass, (void *)pip, vlun->svl_lun_wwn)); 4914 (void) taskq_dispatch(vhci->vhci_taskq, 4915 vhci_initiate_auto_failback, (void *) vlun, 4916 KM_SLEEP); 4917 } 4918 } 4919 vlun->svl_fo_support = opinfo.opinfo_mode; 4920 mdi_pi_set_preferred(pip, opinfo.opinfo_preferred); 4921 4922 VHCI_DEBUG(8, (CE_NOTE, NULL, "vhci_update_pathinfo: opinfo_rev = %x," 4923 " opinfo_path_state = %x opinfo_preferred = %x, opinfo_mode = %x\n", 4924 opinfo.opinfo_rev, opinfo.opinfo_path_state, 4925 opinfo.opinfo_preferred, opinfo.opinfo_mode)); 4926 4927 return (MDI_SUCCESS); 4928 } 4929 4930 /* 4931 * Form the kstat name and and call mdi_pi_kstat_create() 4932 */ 4933 void 4934 vhci_kstat_create_pathinfo(mdi_pathinfo_t *pip) 4935 { 4936 dev_info_t *tgt_dip; 4937 dev_info_t *pdip; 4938 char *guid; 4939 char *target_port, *target_port_dup; 4940 char ks_name[KSTAT_STRLEN]; 4941 uint_t pid; 4942 int by_id; 4943 mod_hash_val_t hv; 4944 4945 4946 /* return if we have already allocated kstats */ 4947 if (mdi_pi_kstat_exists(pip)) 4948 return; 4949 4950 /* 4951 * We need instance numbers to create a kstat name, return if we don't 4952 * have instance numbers assigned yet. 4953 */ 4954 tgt_dip = mdi_pi_get_client(pip); 4955 pdip = mdi_pi_get_phci(pip); 4956 if ((ddi_get_instance(tgt_dip) == -1) || (ddi_get_instance(pdip) == -1)) 4957 return; 4958 4959 /* 4960 * A path oriented kstat has a ks_name of the form: 4961 * 4962 * <client-driver><instance>.t<pid>.<pHCI-driver><instance> 4963 * 4964 * We maintain a bidirectional 'target-port' to <pid> map, 4965 * called targetmap. All pathinfo nodes with the same 4966 * 'target-port' map to the same <pid>. The iostat(1M) code, 4967 * when parsing a path oriented kstat name, uses the <pid> as 4968 * a SCSI_VHCI_GET_TARGET_LONGNAME ioctl argument in order 4969 * to get the 'target-port'. For KSTAT_FLAG_PERSISTENT kstats, 4970 * this ioctl needs to translate a <pid> to a 'target-port' 4971 * even after all pathinfo nodes associated with the 4972 * 'target-port' have been destroyed. This is needed to support 4973 * consistent first-iteration activity-since-boot iostat(1M) 4974 * output. Because of this requirement, the mapping can't be 4975 * based on pathinfo information in a devinfo snapshot. 4976 */ 4977 4978 /* determine 'target-port' */ 4979 if (mdi_prop_lookup_string(pip, 4980 SCSI_ADDR_PROP_TARGET_PORT, &target_port) == MDI_SUCCESS) { 4981 target_port_dup = i_ddi_strdup(target_port, KM_SLEEP); 4982 (void) mdi_prop_free(target_port); 4983 by_id = 1; 4984 } else { 4985 /* 4986 * If the pHCI did not set up 'target-port' on this 4987 * pathinfo node, assume that our client is the only 4988 * one with paths to the device by using the guid 4989 * value as the 'target-port'. Since no other client 4990 * will have the same guid, no other client will use 4991 * the same <pid>. NOTE: a client with an instance 4992 * number always has a guid. 4993 */ 4994 (void) ddi_prop_lookup_string(DDI_DEV_T_ANY, tgt_dip, 4995 PROPFLAGS, MDI_CLIENT_GUID_PROP, &guid); 4996 target_port_dup = i_ddi_strdup(guid, KM_SLEEP); 4997 ddi_prop_free(guid); 4998 4999 /* 5000 * For this type of mapping we don't want the 5001 * <id> -> 'target-port' mapping to be made. This 5002 * will cause the SCSI_VHCI_GET_TARGET_LONGNAME ioctl 5003 * to fail, and the iostat(1M) long '-n' output will 5004 * still use the <pid>. We do this because we just 5005 * made up the 'target-port' using the guid, and we 5006 * don't want to expose that fact in iostat output. 5007 */ 5008 by_id = 0; 5009 } 5010 5011 /* find/establish <pid> given 'target-port' */ 5012 mutex_enter(&vhci_targetmap_mutex); 5013 if (mod_hash_find(vhci_targetmap_byport, 5014 (mod_hash_key_t)target_port_dup, &hv) == 0) { 5015 pid = (int)(intptr_t)hv; /* mapping exists */ 5016 } else { 5017 pid = vhci_targetmap_pid++; /* new mapping */ 5018 5019 (void) mod_hash_insert(vhci_targetmap_byport, 5020 (mod_hash_key_t)target_port_dup, 5021 (mod_hash_val_t)(intptr_t)pid); 5022 if (by_id) { 5023 (void) mod_hash_insert(vhci_targetmap_bypid, 5024 (mod_hash_key_t)(uintptr_t)pid, 5025 (mod_hash_val_t)(uintptr_t)target_port_dup); 5026 } 5027 target_port_dup = NULL; /* owned by hash */ 5028 } 5029 mutex_exit(&vhci_targetmap_mutex); 5030 5031 /* form kstat name */ 5032 (void) snprintf(ks_name, KSTAT_STRLEN, "%s%d.t%d.%s%d", 5033 ddi_driver_name(tgt_dip), ddi_get_instance(tgt_dip), 5034 pid, ddi_driver_name(pdip), ddi_get_instance(pdip)); 5035 5036 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_path_online: path:%p " 5037 "kstat %s: pid %x <-> port %s\n", (void *)pip, 5038 ks_name, pid, target_port_dup)); 5039 if (target_port_dup) 5040 kmem_free(target_port_dup, strlen(target_port_dup) + 1); 5041 5042 /* call mdi to create kstats with the name we built */ 5043 (void) mdi_pi_kstat_create(pip, ks_name); 5044 } 5045 5046 /* ARGSUSED */ 5047 static int 5048 vhci_pathinfo_online(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 5049 { 5050 scsi_hba_tran_t *hba = NULL; 5051 struct scsi_device *psd = NULL; 5052 scsi_vhci_lun_t *vlun = NULL; 5053 dev_info_t *pdip = NULL; 5054 dev_info_t *cdip; 5055 dev_info_t *tgt_dip; 5056 struct scsi_vhci *vhci; 5057 char *guid; 5058 struct scsi_failover_ops *sfo; 5059 scsi_vhci_priv_t *svp = NULL; 5060 struct scsi_address *ap; 5061 struct scsi_pkt *pkt; 5062 int rval = MDI_FAILURE; 5063 mpapi_item_list_t *list_ptr; 5064 mpapi_lu_data_t *ld; 5065 5066 ASSERT(vdip != NULL); 5067 ASSERT(pip != NULL); 5068 5069 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 5070 ASSERT(vhci != NULL); 5071 5072 pdip = mdi_pi_get_phci(pip); 5073 hba = ddi_get_driver_private(pdip); 5074 ASSERT(hba != NULL); 5075 5076 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 5077 ASSERT(svp != NULL); 5078 5079 cdip = mdi_pi_get_client(pip); 5080 ASSERT(cdip != NULL); 5081 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, cdip, PROPFLAGS, 5082 MDI_CLIENT_GUID_PROP, &guid) != DDI_SUCCESS) { 5083 VHCI_DEBUG(1, (CE_WARN, NULL, "vhci_path_online: lun guid " 5084 "property failed")); 5085 goto failure; 5086 } 5087 5088 vlun = vhci_lun_lookup(cdip); 5089 ASSERT(vlun != NULL); 5090 5091 ddi_prop_free(guid); 5092 5093 vlun->svl_dip = mdi_pi_get_client(pip); 5094 ASSERT(vlun->svl_dip != NULL); 5095 5096 psd = svp->svp_psd; 5097 ASSERT(psd != NULL); 5098 5099 ap = &psd->sd_address; 5100 5101 /* 5102 * Get inquiry data into pathinfo related scsi_device structure. 5103 * Free sq_inq when pathinfo related scsi_device structure is destroyed 5104 * by vhci_pathinfo_uninit(). In other words, vhci maintains its own 5105 * copy of scsi_device and scsi_inquiry data on a per-path basis. 5106 */ 5107 if (scsi_probe(psd, SLEEP_FUNC) != SCSIPROBE_EXISTS) { 5108 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_pathinfo_online: " 5109 "scsi_probe failed path:%p rval:%x\n", (void *)pip, rval)); 5110 rval = MDI_FAILURE; 5111 goto failure; 5112 } 5113 5114 /* 5115 * See if we have a failover module to support the device. 5116 * 5117 * We re-probe to determine the failover ops for each path. This 5118 * is done in case there are any path-specific side-effects associated 5119 * with the sfo_device_probe implementation. 5120 * 5121 * Give the first successfull sfo_device_probe the opportunity to 5122 * establish 'ctpriv', vlun/client private data. The ctpriv will 5123 * then be passed into the failover module on all other sfo_device_*() 5124 * operations (and must be freed by sfo_device_unprobe implementation). 5125 * 5126 * NOTE: While sfo_device_probe is done once per path, 5127 * sfo_device_unprobe only occurs once - when the vlun is destroyed. 5128 * 5129 * NOTE: We don't currently support per-path fops private data 5130 * mechanism. 5131 */ 5132 sfo = vhci_dev_fo(vdip, psd, 5133 &vlun->svl_fops_ctpriv, &vlun->svl_fops_name); 5134 5135 /* check path configuration result with current vlun state */ 5136 if (((sfo && vlun->svl_fops) && (sfo != vlun->svl_fops)) || 5137 (sfo && vlun->svl_not_supported) || 5138 ((sfo == NULL) && vlun->svl_fops)) { 5139 /* Getting different results for different paths. */ 5140 VHCI_DEBUG(1, (CE_NOTE, vhci->vhci_dip, 5141 "!vhci_pathinfo_online: dev (path 0x%p) contradiction\n", 5142 (void *)pip)); 5143 cmn_err(CE_WARN, "scsi_vhci: failover contradiction: " 5144 "'%s'.vs.'%s': path %s\n", 5145 vlun->svl_fops ? vlun->svl_fops->sfo_name : "NULL", 5146 sfo ? sfo->sfo_name : "NULL", mdi_pi_pathname(pip)); 5147 vlun->svl_not_supported = 1; 5148 rval = MDI_NOT_SUPPORTED; 5149 goto done; 5150 } else if (sfo == NULL) { 5151 /* No failover module - device not supported under vHCI. */ 5152 VHCI_DEBUG(1, (CE_NOTE, vhci->vhci_dip, 5153 "!vhci_pathinfo_online: dev (path 0x%p) not " 5154 "supported\n", (void *)pip)); 5155 5156 /* XXX does this contradict vhci_is_dev_supported ? */ 5157 vlun->svl_not_supported = 1; 5158 rval = MDI_NOT_SUPPORTED; 5159 goto done; 5160 } 5161 5162 /* failover supported for device - save failover_ops in vlun */ 5163 vlun->svl_fops = sfo; 5164 ASSERT(vlun->svl_fops_name != NULL); 5165 5166 /* 5167 * Obtain the device-type based mpxio options as specified in 5168 * scsi_vhci.conf file. 5169 * 5170 * NOTE: currently, the end result is a call to 5171 * mdi_set_lb_region_size(). 5172 */ 5173 tgt_dip = psd->sd_dev; 5174 ASSERT(tgt_dip != NULL); 5175 vhci_get_device_type_mpxio_options(vdip, tgt_dip, psd); 5176 5177 /* 5178 * if PGR is active, revalidate key and register on this path also, 5179 * if key is still valid 5180 */ 5181 sema_p(&vlun->svl_pgr_sema); 5182 if (vlun->svl_pgr_active) { 5183 rval = vhci_pgr_validate_and_register(svp); 5184 if (rval != 1) { 5185 rval = MDI_FAILURE; 5186 sema_v(&vlun->svl_pgr_sema); 5187 goto failure; 5188 } 5189 } 5190 sema_v(&vlun->svl_pgr_sema); 5191 5192 if (svp->svp_new_path) { 5193 /* 5194 * Last chance to perform any cleanup operations on this 5195 * new path before making this path completely online. 5196 */ 5197 svp->svp_new_path = 0; 5198 5199 /* 5200 * If scsi_vhci knows the lun is alread RESERVE'd, 5201 * then skip the issue of RELEASE on new path. 5202 */ 5203 if ((vlun->svl_flags & VLUN_RESERVE_ACTIVE_FLG) == 0) { 5204 /* 5205 * Issue SCSI-2 RELEASE only for the first time on 5206 * a new path just in case the host rebooted and 5207 * a reservation is still pending on this path. 5208 * IBM Shark storage does not clear RESERVE upon 5209 * host reboot. 5210 */ 5211 pkt = scsi_init_pkt(ap, NULL, NULL, CDB_GROUP0, 5212 sizeof (struct scsi_arq_status), 0, 0, 5213 SLEEP_FUNC, NULL); 5214 if (pkt == NULL) { 5215 VHCI_DEBUG(1, (CE_NOTE, NULL, 5216 "!vhci_pathinfo_online: " 5217 "Release init_pkt failed :%p\n", 5218 (void *)pip)); 5219 rval = MDI_FAILURE; 5220 goto failure; 5221 } 5222 pkt->pkt_cdbp[0] = SCMD_RELEASE; 5223 pkt->pkt_time = 60; 5224 5225 VHCI_DEBUG(1, (CE_NOTE, NULL, 5226 "!vhci_path_online: path:%p " 5227 "Issued SCSI-2 RELEASE\n", (void *)pip)); 5228 5229 /* Ignore the return value */ 5230 (void) vhci_do_scsi_cmd(pkt); 5231 scsi_destroy_pkt(pkt); 5232 } 5233 } 5234 5235 rval = vhci_update_pathinfo(psd, pip, sfo, vlun, vhci); 5236 if (rval == MDI_FAILURE) { 5237 goto failure; 5238 } 5239 5240 /* Initialize MP-API data */ 5241 vhci_update_mpapi_data(vhci, vlun, pip); 5242 5243 /* 5244 * MP-API also needs the Inquiry data to be maintained in the 5245 * mp_vendor_prop_t structure, so find the lun and update its 5246 * structure with this data. 5247 */ 5248 list_ptr = (mpapi_item_list_t *)vhci_get_mpapi_item(vhci, NULL, 5249 MP_OBJECT_TYPE_MULTIPATH_LU, (void *)vlun); 5250 ld = (mpapi_lu_data_t *)list_ptr->item->idata; 5251 if (ld != NULL) { 5252 bcopy(psd->sd_inq->inq_vid, ld->prop.prodInfo.vendor, 8); 5253 bcopy(psd->sd_inq->inq_pid, ld->prop.prodInfo.product, 16); 5254 bcopy(psd->sd_inq->inq_revision, ld->prop.prodInfo.revision, 4); 5255 } else { 5256 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_pathinfo_online: " 5257 "mpapi_lu_data_t is NULL")); 5258 } 5259 5260 /* create kstats for path */ 5261 vhci_kstat_create_pathinfo(pip); 5262 5263 done: 5264 mutex_enter(&vhci_global_mutex); 5265 cv_broadcast(&vhci_cv); 5266 mutex_exit(&vhci_global_mutex); 5267 5268 if (vlun->svl_setcap_done) { 5269 (void) vhci_pHCI_cap(ap, "sector-size", 5270 vlun->svl_sector_size, 1, pip); 5271 } 5272 5273 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_path_online: path:%p\n", 5274 (void *)pip)); 5275 5276 failure: 5277 return (rval); 5278 } 5279 5280 /* 5281 * path offline handler. Release all bindings that will not be 5282 * released by the normal packet transport/completion code path. 5283 * Since we don't (presently) keep any bindings alive outside of 5284 * the in-transport packets (which will be released on completion) 5285 * there is not much to do here. 5286 */ 5287 /* ARGSUSED */ 5288 static int 5289 vhci_pathinfo_offline(dev_info_t *vdip, mdi_pathinfo_t *pip, int flags) 5290 { 5291 scsi_hba_tran_t *hba = NULL; 5292 struct scsi_device *psd = NULL; 5293 dev_info_t *pdip = NULL; 5294 dev_info_t *cdip = NULL; 5295 scsi_vhci_priv_t *svp = NULL; 5296 5297 ASSERT(vdip != NULL); 5298 ASSERT(pip != NULL); 5299 5300 pdip = mdi_pi_get_phci(pip); 5301 ASSERT(pdip != NULL); 5302 if (pdip == NULL) { 5303 VHCI_DEBUG(1, (CE_WARN, vdip, "Invalid path 0x%p: NULL " 5304 "phci dip", (void *)pip)); 5305 return (MDI_FAILURE); 5306 } 5307 5308 cdip = mdi_pi_get_client(pip); 5309 ASSERT(cdip != NULL); 5310 if (cdip == NULL) { 5311 VHCI_DEBUG(1, (CE_WARN, vdip, "Invalid path 0x%p: NULL " 5312 "client dip", (void *)pip)); 5313 return (MDI_FAILURE); 5314 } 5315 5316 hba = ddi_get_driver_private(pdip); 5317 ASSERT(hba != NULL); 5318 5319 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 5320 if (svp == NULL) { 5321 /* 5322 * mdi_pathinfo node in INIT state can have vHCI private 5323 * information set to null 5324 */ 5325 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5326 "svp is NULL for pip 0x%p\n", (void *)pip)); 5327 return (MDI_SUCCESS); 5328 } 5329 5330 psd = svp->svp_psd; 5331 ASSERT(psd != NULL); 5332 5333 mutex_enter(&svp->svp_mutex); 5334 5335 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5336 "%d cmds pending on path: 0x%p\n", svp->svp_cmds, (void *)pip)); 5337 while (svp->svp_cmds != 0) { 5338 if (cv_reltimedwait(&svp->svp_cv, &svp->svp_mutex, 5339 drv_usectohz(vhci_path_quiesce_timeout * 1000000), 5340 TR_CLOCK_TICK) == -1) { 5341 /* 5342 * The timeout time reached without the condition 5343 * being signaled. 5344 */ 5345 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5346 "Timeout reached on path 0x%p without the cond\n", 5347 (void *)pip)); 5348 VHCI_DEBUG(1, (CE_NOTE, vdip, "!vhci_pathinfo_offline: " 5349 "%d cmds still pending on path: 0x%p\n", 5350 svp->svp_cmds, (void *)pip)); 5351 break; 5352 } 5353 } 5354 mutex_exit(&svp->svp_mutex); 5355 5356 /* 5357 * Check to see if this vlun has an active SCSI-II RESERVE. And this 5358 * is the pip for the path that has been reserved. 5359 * If so clear the reservation by sending a reset, so the host will not 5360 * get a reservation conflict. Reset the flag VLUN_RESERVE_ACTIVE_FLG 5361 * for this lun. Also a reset notify is sent to the target driver 5362 * just in case the POR check condition is cleared by some other layer 5363 * in the stack. 5364 */ 5365 if (svp->svp_svl->svl_flags & VLUN_RESERVE_ACTIVE_FLG) { 5366 if (pip == svp->svp_svl->svl_resrv_pip) { 5367 if (vhci_recovery_reset(svp->svp_svl, 5368 &svp->svp_psd->sd_address, TRUE, 5369 VHCI_DEPTH_TARGET) == 0) { 5370 VHCI_DEBUG(1, (CE_NOTE, NULL, 5371 "!vhci_pathinfo_offline (pip:%p):" 5372 "reset failed, retrying\n", (void *)pip)); 5373 delay(1*drv_usectohz(1000000)); 5374 if (vhci_recovery_reset(svp->svp_svl, 5375 &svp->svp_psd->sd_address, TRUE, 5376 VHCI_DEPTH_TARGET) == 0) { 5377 VHCI_DEBUG(1, (CE_NOTE, NULL, 5378 "!vhci_pathinfo_offline " 5379 "(pip:%p): reset failed, " 5380 "giving up!\n", (void *)pip)); 5381 } 5382 } 5383 svp->svp_svl->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 5384 } 5385 } 5386 5387 mdi_pi_set_state(pip, MDI_PATHINFO_STATE_OFFLINE); 5388 vhci_mpapi_set_path_state(vdip, pip, MP_DRVR_PATH_STATE_REMOVED); 5389 5390 VHCI_DEBUG(1, (CE_NOTE, NULL, 5391 "!vhci_pathinfo_offline: offlined path 0x%p\n", (void *)pip)); 5392 return (MDI_SUCCESS); 5393 } 5394 5395 5396 /* 5397 * routine for SCSI VHCI IOCTL implementation. 5398 */ 5399 /* ARGSUSED */ 5400 static int 5401 vhci_ctl(dev_t dev, int cmd, intptr_t data, int mode, cred_t *credp, int *rval) 5402 { 5403 struct scsi_vhci *vhci; 5404 dev_info_t *vdip; 5405 mdi_pathinfo_t *pip; 5406 int instance, held; 5407 int retval = 0; 5408 caddr_t phci_path = NULL, client_path = NULL; 5409 caddr_t paddr = NULL; 5410 sv_iocdata_t ioc; 5411 sv_iocdata_t *pioc = &ioc; 5412 sv_switch_to_cntlr_iocdata_t iocsc; 5413 sv_switch_to_cntlr_iocdata_t *piocsc = &iocsc; 5414 caddr_t s; 5415 scsi_vhci_lun_t *vlun; 5416 struct scsi_failover_ops *fo; 5417 char *pclass; 5418 5419 /* Check for validity of vhci structure */ 5420 vhci = ddi_get_soft_state(vhci_softstate, MINOR2INST(getminor(dev))); 5421 if (vhci == NULL) { 5422 return (ENXIO); 5423 } 5424 5425 mutex_enter(&vhci->vhci_mutex); 5426 if ((vhci->vhci_state & VHCI_STATE_OPEN) == 0) { 5427 mutex_exit(&vhci->vhci_mutex); 5428 return (ENXIO); 5429 } 5430 mutex_exit(&vhci->vhci_mutex); 5431 5432 /* Get the vhci dip */ 5433 vdip = vhci->vhci_dip; 5434 ASSERT(vdip != NULL); 5435 instance = ddi_get_instance(vdip); 5436 5437 /* Allocate memory for getting parameters from userland */ 5438 phci_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5439 client_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5440 paddr = kmem_zalloc(MAXNAMELEN, KM_SLEEP); 5441 5442 /* 5443 * Set a local variable indicating the ioctl name. Used for 5444 * printing debug strings. 5445 */ 5446 switch (cmd) { 5447 case SCSI_VHCI_GET_CLIENT_MULTIPATH_INFO: 5448 s = "GET_CLIENT_MULTIPATH_INFO"; 5449 break; 5450 5451 case SCSI_VHCI_GET_PHCI_MULTIPATH_INFO: 5452 s = "GET_PHCI_MULTIPATH_INFO"; 5453 break; 5454 5455 case SCSI_VHCI_GET_CLIENT_NAME: 5456 s = "GET_CLIENT_NAME"; 5457 break; 5458 5459 case SCSI_VHCI_PATH_ONLINE: 5460 s = "PATH_ONLINE"; 5461 break; 5462 5463 case SCSI_VHCI_PATH_OFFLINE: 5464 s = "PATH_OFFLINE"; 5465 break; 5466 5467 case SCSI_VHCI_PATH_STANDBY: 5468 s = "PATH_STANDBY"; 5469 break; 5470 5471 case SCSI_VHCI_PATH_TEST: 5472 s = "PATH_TEST"; 5473 break; 5474 5475 case SCSI_VHCI_SWITCH_TO_CNTLR: 5476 s = "SWITCH_TO_CNTLR"; 5477 break; 5478 case SCSI_VHCI_PATH_DISABLE: 5479 s = "PATH_DISABLE"; 5480 break; 5481 case SCSI_VHCI_PATH_ENABLE: 5482 s = "PATH_ENABLE"; 5483 break; 5484 5485 case SCSI_VHCI_GET_TARGET_LONGNAME: 5486 s = "GET_TARGET_LONGNAME"; 5487 break; 5488 5489 #ifdef DEBUG 5490 case SCSI_VHCI_CONFIGURE_PHCI: 5491 s = "CONFIGURE_PHCI"; 5492 break; 5493 5494 case SCSI_VHCI_UNCONFIGURE_PHCI: 5495 s = "UNCONFIGURE_PHCI"; 5496 break; 5497 #endif 5498 5499 default: 5500 s = "Unknown"; 5501 vhci_log(CE_NOTE, vdip, 5502 "!vhci%d: ioctl %x (unsupported ioctl)", instance, cmd); 5503 retval = ENOTSUP; 5504 break; 5505 } 5506 if (retval != 0) { 5507 goto end; 5508 } 5509 5510 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci%d: ioctl <%s>", instance, s)); 5511 5512 /* 5513 * Get IOCTL parameters from userland 5514 */ 5515 switch (cmd) { 5516 case SCSI_VHCI_GET_CLIENT_MULTIPATH_INFO: 5517 case SCSI_VHCI_GET_PHCI_MULTIPATH_INFO: 5518 case SCSI_VHCI_GET_CLIENT_NAME: 5519 case SCSI_VHCI_PATH_ONLINE: 5520 case SCSI_VHCI_PATH_OFFLINE: 5521 case SCSI_VHCI_PATH_STANDBY: 5522 case SCSI_VHCI_PATH_TEST: 5523 case SCSI_VHCI_PATH_DISABLE: 5524 case SCSI_VHCI_PATH_ENABLE: 5525 case SCSI_VHCI_GET_TARGET_LONGNAME: 5526 #ifdef DEBUG 5527 case SCSI_VHCI_CONFIGURE_PHCI: 5528 case SCSI_VHCI_UNCONFIGURE_PHCI: 5529 #endif 5530 retval = vhci_get_iocdata((const void *)data, pioc, mode, s); 5531 break; 5532 5533 case SCSI_VHCI_SWITCH_TO_CNTLR: 5534 retval = vhci_get_iocswitchdata((const void *)data, piocsc, 5535 mode, s); 5536 break; 5537 } 5538 if (retval != 0) { 5539 goto end; 5540 } 5541 5542 5543 /* 5544 * Process the IOCTL 5545 */ 5546 switch (cmd) { 5547 case SCSI_VHCI_GET_CLIENT_MULTIPATH_INFO: 5548 { 5549 uint_t num_paths; /* Num paths to client dev */ 5550 sv_path_info_t *upibuf = NULL; /* To keep userland values */ 5551 sv_path_info_t *kpibuf = NULL; /* Kernel data for ioctls */ 5552 dev_info_t *cdip; /* Client device dip */ 5553 5554 if (pioc->ret_elem == NULL) { 5555 retval = EINVAL; 5556 break; 5557 } 5558 5559 /* Get client device path from user land */ 5560 if (vhci_ioc_get_client_path(pioc, client_path, mode, s)) { 5561 retval = EFAULT; 5562 break; 5563 } 5564 5565 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5566 "client <%s>", s, client_path)); 5567 5568 /* Get number of paths to this client device */ 5569 if ((cdip = mdi_client_path2devinfo(vdip, client_path)) 5570 == NULL) { 5571 retval = ENXIO; 5572 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5573 "client dip doesn't exist. invalid path <%s>", 5574 s, client_path)); 5575 break; 5576 } 5577 num_paths = mdi_client_get_path_count(cdip); 5578 5579 if (ddi_copyout(&num_paths, pioc->ret_elem, 5580 sizeof (num_paths), mode)) { 5581 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5582 "num_paths copyout failed", s)); 5583 retval = EFAULT; 5584 break; 5585 } 5586 5587 /* If user just wanted num_paths, then return */ 5588 if (pioc->buf_elem == 0 || pioc->ret_buf == NULL || 5589 num_paths == 0) { 5590 break; 5591 } 5592 5593 /* Set num_paths to value as much as can be sent to userland */ 5594 if (num_paths > pioc->buf_elem) { 5595 num_paths = pioc->buf_elem; 5596 } 5597 5598 /* Allocate memory and get userland pointers */ 5599 if (vhci_ioc_alloc_pathinfo(&upibuf, &kpibuf, num_paths, 5600 pioc, mode, s) != 0) { 5601 retval = EFAULT; 5602 break; 5603 } 5604 ASSERT(upibuf != NULL); 5605 ASSERT(kpibuf != NULL); 5606 5607 /* 5608 * Get the path information and send it to userland. 5609 */ 5610 if (vhci_get_client_path_list(cdip, kpibuf, num_paths) 5611 != MDI_SUCCESS) { 5612 retval = ENXIO; 5613 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5614 break; 5615 } 5616 5617 if (vhci_ioc_send_pathinfo(upibuf, kpibuf, num_paths, 5618 pioc, mode, s)) { 5619 retval = EFAULT; 5620 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5621 break; 5622 } 5623 5624 /* Free the memory allocated for path information */ 5625 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5626 break; 5627 } 5628 5629 case SCSI_VHCI_GET_PHCI_MULTIPATH_INFO: 5630 { 5631 uint_t num_paths; /* Num paths to client dev */ 5632 sv_path_info_t *upibuf = NULL; /* To keep userland values */ 5633 sv_path_info_t *kpibuf = NULL; /* Kernel data for ioctls */ 5634 dev_info_t *pdip; /* PHCI device dip */ 5635 5636 if (pioc->ret_elem == NULL) { 5637 retval = EINVAL; 5638 break; 5639 } 5640 5641 /* Get PHCI device path from user land */ 5642 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 5643 retval = EFAULT; 5644 break; 5645 } 5646 5647 VHCI_DEBUG(6, (CE_WARN, vdip, 5648 "!vhci_ioctl: ioctl <%s> phci <%s>", s, phci_path)); 5649 5650 /* Get number of devices associated with this PHCI device */ 5651 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5652 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5653 "phci dip doesn't exist. invalid path <%s>", 5654 s, phci_path)); 5655 retval = ENXIO; 5656 break; 5657 } 5658 5659 num_paths = mdi_phci_get_path_count(pdip); 5660 5661 if (ddi_copyout(&num_paths, pioc->ret_elem, 5662 sizeof (num_paths), mode)) { 5663 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5664 "num_paths copyout failed", s)); 5665 retval = EFAULT; 5666 break; 5667 } 5668 5669 /* If user just wanted num_paths, then return */ 5670 if (pioc->buf_elem == 0 || pioc->ret_buf == NULL || 5671 num_paths == 0) { 5672 break; 5673 } 5674 5675 /* Set num_paths to value as much as can be sent to userland */ 5676 if (num_paths > pioc->buf_elem) { 5677 num_paths = pioc->buf_elem; 5678 } 5679 5680 /* Allocate memory and get userland pointers */ 5681 if (vhci_ioc_alloc_pathinfo(&upibuf, &kpibuf, num_paths, 5682 pioc, mode, s) != 0) { 5683 retval = EFAULT; 5684 break; 5685 } 5686 ASSERT(upibuf != NULL); 5687 ASSERT(kpibuf != NULL); 5688 5689 /* 5690 * Get the path information and send it to userland. 5691 */ 5692 if (vhci_get_phci_path_list(pdip, kpibuf, num_paths) 5693 != MDI_SUCCESS) { 5694 retval = ENXIO; 5695 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5696 break; 5697 } 5698 5699 if (vhci_ioc_send_pathinfo(upibuf, kpibuf, num_paths, 5700 pioc, mode, s)) { 5701 retval = EFAULT; 5702 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5703 break; 5704 } 5705 5706 /* Free the memory allocated for path information */ 5707 vhci_ioc_free_pathinfo(upibuf, kpibuf, num_paths); 5708 break; 5709 } 5710 5711 case SCSI_VHCI_GET_CLIENT_NAME: 5712 { 5713 dev_info_t *cdip, *pdip; 5714 5715 /* Get PHCI path and device address from user land */ 5716 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s) || 5717 vhci_ioc_get_paddr(pioc, paddr, mode, s)) { 5718 retval = EFAULT; 5719 break; 5720 } 5721 5722 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5723 "phci <%s>, paddr <%s>", s, phci_path, paddr)); 5724 5725 /* Get the PHCI dip */ 5726 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5727 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5728 "phci dip doesn't exist. invalid path <%s>", 5729 s, phci_path)); 5730 retval = ENXIO; 5731 break; 5732 } 5733 5734 if ((pip = mdi_pi_find(pdip, NULL, paddr)) == NULL) { 5735 VHCI_DEBUG(1, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5736 "pathinfo doesn't exist. invalid device addr", s)); 5737 retval = ENXIO; 5738 break; 5739 } 5740 5741 /* Get the client device pathname and send to userland */ 5742 cdip = mdi_pi_get_client(pip); 5743 vhci_ioc_devi_to_path(cdip, client_path); 5744 5745 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5746 "client <%s>", s, client_path)); 5747 5748 if (vhci_ioc_send_client_path(client_path, pioc, mode, s)) { 5749 retval = EFAULT; 5750 break; 5751 } 5752 break; 5753 } 5754 5755 case SCSI_VHCI_PATH_ONLINE: 5756 case SCSI_VHCI_PATH_OFFLINE: 5757 case SCSI_VHCI_PATH_STANDBY: 5758 case SCSI_VHCI_PATH_TEST: 5759 { 5760 dev_info_t *pdip; /* PHCI dip */ 5761 5762 /* Get PHCI path and device address from user land */ 5763 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s) || 5764 vhci_ioc_get_paddr(pioc, paddr, mode, s)) { 5765 retval = EFAULT; 5766 break; 5767 } 5768 5769 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5770 "phci <%s>, paddr <%s>", s, phci_path, paddr)); 5771 5772 /* Get the PHCI dip */ 5773 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5774 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5775 "phci dip doesn't exist. invalid path <%s>", 5776 s, phci_path)); 5777 retval = ENXIO; 5778 break; 5779 } 5780 5781 if ((pip = mdi_pi_find(pdip, NULL, paddr)) == NULL) { 5782 VHCI_DEBUG(1, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5783 "pathinfo doesn't exist. invalid device addr", s)); 5784 retval = ENXIO; 5785 break; 5786 } 5787 5788 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5789 "Calling MDI function to change device state", s)); 5790 5791 switch (cmd) { 5792 case SCSI_VHCI_PATH_ONLINE: 5793 retval = mdi_pi_online(pip, 0); 5794 break; 5795 5796 case SCSI_VHCI_PATH_OFFLINE: 5797 retval = mdi_pi_offline(pip, 0); 5798 break; 5799 5800 case SCSI_VHCI_PATH_STANDBY: 5801 retval = mdi_pi_standby(pip, 0); 5802 break; 5803 5804 case SCSI_VHCI_PATH_TEST: 5805 break; 5806 } 5807 break; 5808 } 5809 5810 case SCSI_VHCI_SWITCH_TO_CNTLR: 5811 { 5812 dev_info_t *cdip; 5813 struct scsi_device *devp; 5814 5815 /* Get the client device pathname */ 5816 if (ddi_copyin(piocsc->client, client_path, 5817 MAXPATHLEN, mode)) { 5818 VHCI_DEBUG(2, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5819 "client_path copyin failed", s)); 5820 retval = EFAULT; 5821 break; 5822 } 5823 5824 /* Get the path class to which user wants to switch */ 5825 if (ddi_copyin(piocsc->class, paddr, MAXNAMELEN, mode)) { 5826 VHCI_DEBUG(2, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 5827 "controller_class copyin failed", s)); 5828 retval = EFAULT; 5829 break; 5830 } 5831 5832 /* Perform validity checks */ 5833 if ((cdip = mdi_client_path2devinfo(vdip, 5834 client_path)) == NULL) { 5835 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5836 "client dip doesn't exist. invalid path <%s>", 5837 s, client_path)); 5838 retval = ENXIO; 5839 break; 5840 } 5841 5842 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: Calling MDI func " 5843 "to switch controller")); 5844 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: client <%s> " 5845 "class <%s>", client_path, paddr)); 5846 5847 if (strcmp(paddr, PCLASS_PRIMARY) && 5848 strcmp(paddr, PCLASS_SECONDARY)) { 5849 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5850 "invalid path class <%s>", s, paddr)); 5851 retval = ENXIO; 5852 break; 5853 } 5854 5855 devp = ddi_get_driver_private(cdip); 5856 if (devp == NULL) { 5857 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5858 "invalid scsi device <%s>", s, client_path)); 5859 retval = ENXIO; 5860 break; 5861 } 5862 vlun = ADDR2VLUN(&devp->sd_address); 5863 ASSERT(vlun); 5864 5865 /* 5866 * Checking to see if device has only one pclass, PRIMARY. 5867 * If so this device doesn't support failovers. Assumed 5868 * that the devices with one pclass is PRIMARY, as thats the 5869 * case today. If this is not true and in future other 5870 * symmetric devices are supported with other pclass, this 5871 * IOCTL shall have to be overhauled anyways as now the only 5872 * arguments it accepts are PRIMARY and SECONDARY. 5873 */ 5874 fo = vlun->svl_fops; 5875 if (fo->sfo_pathclass_next(PCLASS_PRIMARY, &pclass, 5876 vlun->svl_fops_ctpriv)) { 5877 retval = ENOTSUP; 5878 break; 5879 } 5880 5881 VHCI_HOLD_LUN(vlun, VH_SLEEP, held); 5882 mutex_enter(&vlun->svl_mutex); 5883 if (vlun->svl_active_pclass != NULL) { 5884 if (strcmp(vlun->svl_active_pclass, paddr) == 0) { 5885 mutex_exit(&vlun->svl_mutex); 5886 retval = EALREADY; 5887 VHCI_RELEASE_LUN(vlun); 5888 break; 5889 } 5890 } 5891 mutex_exit(&vlun->svl_mutex); 5892 /* Call mdi function to cause a switch over */ 5893 retval = mdi_failover(vdip, cdip, MDI_FAILOVER_SYNC); 5894 if (retval == MDI_SUCCESS) { 5895 retval = 0; 5896 } else if (retval == MDI_BUSY) { 5897 retval = EBUSY; 5898 } else { 5899 retval = EIO; 5900 } 5901 VHCI_RELEASE_LUN(vlun); 5902 break; 5903 } 5904 5905 case SCSI_VHCI_PATH_ENABLE: 5906 case SCSI_VHCI_PATH_DISABLE: 5907 { 5908 dev_info_t *cdip, *pdip; 5909 5910 /* 5911 * Get client device path from user land 5912 */ 5913 if (vhci_ioc_get_client_path(pioc, client_path, mode, s)) { 5914 retval = EFAULT; 5915 break; 5916 } 5917 5918 /* 5919 * Get Phci device path from user land 5920 */ 5921 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 5922 retval = EFAULT; 5923 break; 5924 } 5925 5926 /* 5927 * Get the devinfo for the Phci. 5928 */ 5929 if ((pdip = mdi_phci_path2devinfo(vdip, phci_path)) == NULL) { 5930 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5931 "phci dip doesn't exist. invalid path <%s>", 5932 s, phci_path)); 5933 retval = ENXIO; 5934 break; 5935 } 5936 5937 /* 5938 * If the client path is set to /scsi_vhci then we need 5939 * to do the operation on all the clients so set cdip to NULL. 5940 * Else, try to get the client dip. 5941 */ 5942 if (strcmp(client_path, "/scsi_vhci") == 0) { 5943 cdip = NULL; 5944 } else { 5945 if ((cdip = mdi_client_path2devinfo(vdip, 5946 client_path)) == NULL) { 5947 retval = ENXIO; 5948 VHCI_DEBUG(1, (CE_WARN, NULL, 5949 "!vhci_ioctl: ioctl <%s> client dip " 5950 "doesn't exist. invalid path <%s>", 5951 s, client_path)); 5952 break; 5953 } 5954 } 5955 5956 if (cmd == SCSI_VHCI_PATH_ENABLE) 5957 retval = mdi_pi_enable(cdip, pdip, USER_DISABLE); 5958 else 5959 retval = mdi_pi_disable(cdip, pdip, USER_DISABLE); 5960 5961 break; 5962 } 5963 5964 case SCSI_VHCI_GET_TARGET_LONGNAME: 5965 { 5966 uint_t pid = pioc->buf_elem; 5967 char *target_port; 5968 mod_hash_val_t hv; 5969 5970 /* targetmap lookup of 'target-port' by <pid> */ 5971 if (mod_hash_find(vhci_targetmap_bypid, 5972 (mod_hash_key_t)(uintptr_t)pid, &hv) != 0) { 5973 /* 5974 * NOTE: failure to find the mapping is OK for guid 5975 * based 'target-port' values. 5976 */ 5977 VHCI_DEBUG(3, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5978 "targetport mapping doesn't exist: pid %d", 5979 s, pid)); 5980 retval = ENXIO; 5981 break; 5982 } 5983 5984 /* copyout 'target-port' result */ 5985 target_port = (char *)hv; 5986 if (copyoutstr(target_port, pioc->addr, MAXNAMELEN, NULL)) { 5987 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 5988 "targetport copyout failed: len: %d", 5989 s, (int)strlen(target_port))); 5990 retval = EFAULT; 5991 } 5992 break; 5993 } 5994 5995 #ifdef DEBUG 5996 case SCSI_VHCI_CONFIGURE_PHCI: 5997 { 5998 dev_info_t *pdip; 5999 6000 /* Get PHCI path and device address from user land */ 6001 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 6002 retval = EFAULT; 6003 break; 6004 } 6005 6006 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 6007 "phci <%s>", s, phci_path)); 6008 6009 /* Get the PHCI dip */ 6010 if ((pdip = e_ddi_hold_devi_by_path(phci_path, 0)) == NULL) { 6011 VHCI_DEBUG(3, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 6012 "phci dip doesn't exist. invalid path <%s>", 6013 s, phci_path)); 6014 retval = ENXIO; 6015 break; 6016 } 6017 6018 if (ndi_devi_config(pdip, 6019 NDI_DEVFS_CLEAN|NDI_DEVI_PERSIST) != NDI_SUCCESS) { 6020 retval = EIO; 6021 } 6022 6023 ddi_release_devi(pdip); 6024 break; 6025 } 6026 6027 case SCSI_VHCI_UNCONFIGURE_PHCI: 6028 { 6029 dev_info_t *pdip; 6030 6031 /* Get PHCI path and device address from user land */ 6032 if (vhci_ioc_get_phci_path(pioc, phci_path, mode, s)) { 6033 retval = EFAULT; 6034 break; 6035 } 6036 6037 VHCI_DEBUG(6, (CE_WARN, vdip, "!vhci_ioctl: ioctl <%s> " 6038 "phci <%s>", s, phci_path)); 6039 6040 /* Get the PHCI dip */ 6041 if ((pdip = e_ddi_hold_devi_by_path(phci_path, 0)) == NULL) { 6042 VHCI_DEBUG(3, (CE_WARN, NULL, "!vhci_ioctl: ioctl <%s> " 6043 "phci dip doesn't exist. invalid path <%s>", 6044 s, phci_path)); 6045 retval = ENXIO; 6046 break; 6047 } 6048 6049 if (ndi_devi_unconfig(pdip, 6050 NDI_DEVI_REMOVE|NDI_DEVFS_CLEAN) != NDI_SUCCESS) { 6051 retval = EBUSY; 6052 } 6053 6054 ddi_release_devi(pdip); 6055 break; 6056 } 6057 #endif 6058 } 6059 6060 end: 6061 /* Free the memory allocated above */ 6062 if (phci_path != NULL) { 6063 kmem_free(phci_path, MAXPATHLEN); 6064 } 6065 if (client_path != NULL) { 6066 kmem_free(client_path, MAXPATHLEN); 6067 } 6068 if (paddr != NULL) { 6069 kmem_free(paddr, MAXNAMELEN); 6070 } 6071 return (retval); 6072 } 6073 6074 /* 6075 * devctl IOCTL support for client device DR 6076 */ 6077 /* ARGSUSED */ 6078 int 6079 vhci_devctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, 6080 int *rvalp) 6081 { 6082 dev_info_t *self; 6083 dev_info_t *child; 6084 scsi_hba_tran_t *hba; 6085 struct devctl_iocdata *dcp; 6086 struct scsi_vhci *vhci; 6087 int rv = 0; 6088 int retval = 0; 6089 scsi_vhci_priv_t *svp; 6090 mdi_pathinfo_t *pip; 6091 6092 if ((vhci = ddi_get_soft_state(vhci_softstate, 6093 MINOR2INST(getminor(dev)))) == NULL) 6094 return (ENXIO); 6095 6096 /* 6097 * check if :devctl minor device has been opened 6098 */ 6099 mutex_enter(&vhci->vhci_mutex); 6100 if ((vhci->vhci_state & VHCI_STATE_OPEN) == 0) { 6101 mutex_exit(&vhci->vhci_mutex); 6102 return (ENXIO); 6103 } 6104 mutex_exit(&vhci->vhci_mutex); 6105 6106 self = vhci->vhci_dip; 6107 hba = ddi_get_driver_private(self); 6108 if (hba == NULL) 6109 return (ENXIO); 6110 6111 /* 6112 * We can use the generic implementation for these ioctls 6113 */ 6114 switch (cmd) { 6115 case DEVCTL_DEVICE_GETSTATE: 6116 case DEVCTL_DEVICE_ONLINE: 6117 case DEVCTL_DEVICE_OFFLINE: 6118 case DEVCTL_DEVICE_REMOVE: 6119 case DEVCTL_BUS_GETSTATE: 6120 return (ndi_devctl_ioctl(self, cmd, arg, mode, 0)); 6121 } 6122 6123 /* 6124 * read devctl ioctl data 6125 */ 6126 if (ndi_dc_allochdl((void *)arg, &dcp) != NDI_SUCCESS) 6127 return (EFAULT); 6128 6129 switch (cmd) { 6130 6131 case DEVCTL_DEVICE_RESET: 6132 /* 6133 * lookup and hold child device 6134 */ 6135 if ((child = ndi_devi_find(self, ndi_dc_getname(dcp), 6136 ndi_dc_getaddr(dcp))) == NULL) { 6137 rv = ENXIO; 6138 break; 6139 } 6140 retval = mdi_select_path(child, NULL, 6141 (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH), 6142 NULL, &pip); 6143 if ((retval != MDI_SUCCESS) || (pip == NULL)) { 6144 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioctl:" 6145 "Unable to get a path, dip 0x%p", (void *)child)); 6146 rv = ENXIO; 6147 break; 6148 } 6149 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 6150 if (vhci_recovery_reset(svp->svp_svl, 6151 &svp->svp_psd->sd_address, TRUE, 6152 VHCI_DEPTH_TARGET) == 0) { 6153 VHCI_DEBUG(1, (CE_NOTE, NULL, 6154 "!vhci_ioctl(pip:%p): " 6155 "reset failed\n", (void *)pip)); 6156 rv = ENXIO; 6157 } 6158 mdi_rele_path(pip); 6159 break; 6160 6161 case DEVCTL_BUS_QUIESCE: 6162 case DEVCTL_BUS_UNQUIESCE: 6163 case DEVCTL_BUS_RESET: 6164 case DEVCTL_BUS_RESETALL: 6165 #ifdef DEBUG 6166 case DEVCTL_BUS_CONFIGURE: 6167 case DEVCTL_BUS_UNCONFIGURE: 6168 #endif 6169 rv = ENOTSUP; 6170 break; 6171 6172 default: 6173 rv = ENOTTY; 6174 } /* end of outer switch */ 6175 6176 ndi_dc_freehdl(dcp); 6177 return (rv); 6178 } 6179 6180 /* 6181 * Routine to get the PHCI pathname from ioctl structures in userland 6182 */ 6183 /* ARGSUSED */ 6184 static int 6185 vhci_ioc_get_phci_path(sv_iocdata_t *pioc, caddr_t phci_path, 6186 int mode, caddr_t s) 6187 { 6188 int retval = 0; 6189 6190 if (ddi_copyin(pioc->phci, phci_path, MAXPATHLEN, mode)) { 6191 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_get_phci: ioctl <%s> " 6192 "phci_path copyin failed", s)); 6193 retval = EFAULT; 6194 } 6195 return (retval); 6196 6197 } 6198 6199 6200 /* 6201 * Routine to get the Client device pathname from ioctl structures in userland 6202 */ 6203 /* ARGSUSED */ 6204 static int 6205 vhci_ioc_get_client_path(sv_iocdata_t *pioc, caddr_t client_path, 6206 int mode, caddr_t s) 6207 { 6208 int retval = 0; 6209 6210 if (ddi_copyin(pioc->client, client_path, MAXPATHLEN, mode)) { 6211 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_get_client: " 6212 "ioctl <%s> client_path copyin failed", s)); 6213 retval = EFAULT; 6214 } 6215 return (retval); 6216 } 6217 6218 6219 /* 6220 * Routine to get physical device address from ioctl structure in userland 6221 */ 6222 /* ARGSUSED */ 6223 static int 6224 vhci_ioc_get_paddr(sv_iocdata_t *pioc, caddr_t paddr, int mode, caddr_t s) 6225 { 6226 int retval = 0; 6227 6228 if (ddi_copyin(pioc->addr, paddr, MAXNAMELEN, mode)) { 6229 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_get_paddr: " 6230 "ioctl <%s> device addr copyin failed", s)); 6231 retval = EFAULT; 6232 } 6233 return (retval); 6234 } 6235 6236 6237 /* 6238 * Routine to send client device pathname to userland. 6239 */ 6240 /* ARGSUSED */ 6241 static int 6242 vhci_ioc_send_client_path(caddr_t client_path, sv_iocdata_t *pioc, 6243 int mode, caddr_t s) 6244 { 6245 int retval = 0; 6246 6247 if (ddi_copyout(client_path, pioc->client, MAXPATHLEN, mode)) { 6248 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_ioc_send_client: " 6249 "ioctl <%s> client_path copyout failed", s)); 6250 retval = EFAULT; 6251 } 6252 return (retval); 6253 } 6254 6255 6256 /* 6257 * Routine to translated dev_info pointer (dip) to device pathname. 6258 */ 6259 static void 6260 vhci_ioc_devi_to_path(dev_info_t *dip, caddr_t path) 6261 { 6262 (void) ddi_pathname(dip, path); 6263 } 6264 6265 6266 /* 6267 * vhci_get_phci_path_list: 6268 * get information about devices associated with a 6269 * given PHCI device. 6270 * 6271 * Return Values: 6272 * path information elements 6273 */ 6274 int 6275 vhci_get_phci_path_list(dev_info_t *pdip, sv_path_info_t *pibuf, 6276 uint_t num_elems) 6277 { 6278 uint_t count, done; 6279 mdi_pathinfo_t *pip; 6280 sv_path_info_t *ret_pip; 6281 int status; 6282 size_t prop_size; 6283 int circular; 6284 6285 /* 6286 * Get the PHCI structure and retrieve the path information 6287 * from the GUID hash table. 6288 */ 6289 6290 ret_pip = pibuf; 6291 count = 0; 6292 6293 ndi_devi_enter(pdip, &circular); 6294 6295 done = (count >= num_elems); 6296 pip = mdi_get_next_client_path(pdip, NULL); 6297 while (pip && !done) { 6298 mdi_pi_lock(pip); 6299 (void) ddi_pathname(mdi_pi_get_phci(pip), 6300 ret_pip->device.ret_phci); 6301 (void) strcpy(ret_pip->ret_addr, mdi_pi_get_addr(pip)); 6302 (void) mdi_pi_get_state2(pip, &ret_pip->ret_state, 6303 &ret_pip->ret_ext_state); 6304 6305 status = mdi_prop_size(pip, &prop_size); 6306 if (status == MDI_SUCCESS && ret_pip->ret_prop.ret_buf_size) { 6307 *ret_pip->ret_prop.ret_buf_size = (uint_t)prop_size; 6308 } 6309 6310 #ifdef DEBUG 6311 if (status != MDI_SUCCESS) { 6312 VHCI_DEBUG(2, (CE_WARN, NULL, 6313 "!vhci_get_phci_path_list: " 6314 "phci <%s>, prop size failure 0x%x", 6315 ret_pip->device.ret_phci, status)); 6316 } 6317 #endif /* DEBUG */ 6318 6319 6320 if (status == MDI_SUCCESS && ret_pip->ret_prop.buf && 6321 prop_size && ret_pip->ret_prop.buf_size >= prop_size) { 6322 status = mdi_prop_pack(pip, 6323 &ret_pip->ret_prop.buf, 6324 ret_pip->ret_prop.buf_size); 6325 6326 #ifdef DEBUG 6327 if (status != MDI_SUCCESS) { 6328 VHCI_DEBUG(2, (CE_WARN, NULL, 6329 "!vhci_get_phci_path_list: " 6330 "phci <%s>, prop pack failure 0x%x", 6331 ret_pip->device.ret_phci, status)); 6332 } 6333 #endif /* DEBUG */ 6334 } 6335 6336 mdi_pi_unlock(pip); 6337 pip = mdi_get_next_client_path(pdip, pip); 6338 ret_pip++; 6339 count++; 6340 done = (count >= num_elems); 6341 } 6342 6343 ndi_devi_exit(pdip, circular); 6344 6345 return (MDI_SUCCESS); 6346 } 6347 6348 6349 /* 6350 * vhci_get_client_path_list: 6351 * get information about various paths associated with a 6352 * given client device. 6353 * 6354 * Return Values: 6355 * path information elements 6356 */ 6357 int 6358 vhci_get_client_path_list(dev_info_t *cdip, sv_path_info_t *pibuf, 6359 uint_t num_elems) 6360 { 6361 uint_t count, done; 6362 mdi_pathinfo_t *pip; 6363 sv_path_info_t *ret_pip; 6364 int status; 6365 size_t prop_size; 6366 int circular; 6367 6368 ret_pip = pibuf; 6369 count = 0; 6370 6371 ndi_devi_enter(cdip, &circular); 6372 6373 done = (count >= num_elems); 6374 pip = mdi_get_next_phci_path(cdip, NULL); 6375 while (pip && !done) { 6376 mdi_pi_lock(pip); 6377 (void) ddi_pathname(mdi_pi_get_phci(pip), 6378 ret_pip->device.ret_phci); 6379 (void) strcpy(ret_pip->ret_addr, mdi_pi_get_addr(pip)); 6380 (void) mdi_pi_get_state2(pip, &ret_pip->ret_state, 6381 &ret_pip->ret_ext_state); 6382 6383 status = mdi_prop_size(pip, &prop_size); 6384 if (status == MDI_SUCCESS && ret_pip->ret_prop.ret_buf_size) { 6385 *ret_pip->ret_prop.ret_buf_size = (uint_t)prop_size; 6386 } 6387 6388 #ifdef DEBUG 6389 if (status != MDI_SUCCESS) { 6390 VHCI_DEBUG(2, (CE_WARN, NULL, 6391 "!vhci_get_client_path_list: " 6392 "phci <%s>, prop size failure 0x%x", 6393 ret_pip->device.ret_phci, status)); 6394 } 6395 #endif /* DEBUG */ 6396 6397 6398 if (status == MDI_SUCCESS && ret_pip->ret_prop.buf && 6399 prop_size && ret_pip->ret_prop.buf_size >= prop_size) { 6400 status = mdi_prop_pack(pip, 6401 &ret_pip->ret_prop.buf, 6402 ret_pip->ret_prop.buf_size); 6403 6404 #ifdef DEBUG 6405 if (status != MDI_SUCCESS) { 6406 VHCI_DEBUG(2, (CE_WARN, NULL, 6407 "!vhci_get_client_path_list: " 6408 "phci <%s>, prop pack failure 0x%x", 6409 ret_pip->device.ret_phci, status)); 6410 } 6411 #endif /* DEBUG */ 6412 } 6413 6414 mdi_pi_unlock(pip); 6415 pip = mdi_get_next_phci_path(cdip, pip); 6416 ret_pip++; 6417 count++; 6418 done = (count >= num_elems); 6419 } 6420 6421 ndi_devi_exit(cdip, circular); 6422 6423 return (MDI_SUCCESS); 6424 } 6425 6426 6427 /* 6428 * Routine to get ioctl argument structure from userland. 6429 */ 6430 /* ARGSUSED */ 6431 static int 6432 vhci_get_iocdata(const void *data, sv_iocdata_t *pioc, int mode, caddr_t s) 6433 { 6434 int retval = 0; 6435 6436 #ifdef _MULTI_DATAMODEL 6437 switch (ddi_model_convert_from(mode & FMODELS)) { 6438 case DDI_MODEL_ILP32: 6439 { 6440 sv_iocdata32_t ioc32; 6441 6442 if (ddi_copyin(data, &ioc32, sizeof (ioc32), mode)) { 6443 retval = EFAULT; 6444 break; 6445 } 6446 pioc->client = (caddr_t)(uintptr_t)ioc32.client; 6447 pioc->phci = (caddr_t)(uintptr_t)ioc32.phci; 6448 pioc->addr = (caddr_t)(uintptr_t)ioc32.addr; 6449 pioc->buf_elem = (uint_t)ioc32.buf_elem; 6450 pioc->ret_buf = (sv_path_info_t *)(uintptr_t)ioc32.ret_buf; 6451 pioc->ret_elem = (uint_t *)(uintptr_t)ioc32.ret_elem; 6452 break; 6453 } 6454 6455 case DDI_MODEL_NONE: 6456 if (ddi_copyin(data, pioc, sizeof (*pioc), mode)) { 6457 retval = EFAULT; 6458 break; 6459 } 6460 break; 6461 } 6462 #else /* _MULTI_DATAMODEL */ 6463 if (ddi_copyin(data, pioc, sizeof (*pioc), mode)) { 6464 retval = EFAULT; 6465 } 6466 #endif /* _MULTI_DATAMODEL */ 6467 6468 #ifdef DEBUG 6469 if (retval) { 6470 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: cmd <%s> " 6471 "iocdata copyin failed", s)); 6472 } 6473 #endif 6474 6475 return (retval); 6476 } 6477 6478 6479 /* 6480 * Routine to get the ioctl argument for ioctl causing controller switchover. 6481 */ 6482 /* ARGSUSED */ 6483 static int 6484 vhci_get_iocswitchdata(const void *data, sv_switch_to_cntlr_iocdata_t *piocsc, 6485 int mode, caddr_t s) 6486 { 6487 int retval = 0; 6488 6489 #ifdef _MULTI_DATAMODEL 6490 switch (ddi_model_convert_from(mode & FMODELS)) { 6491 case DDI_MODEL_ILP32: 6492 { 6493 sv_switch_to_cntlr_iocdata32_t ioc32; 6494 6495 if (ddi_copyin(data, &ioc32, sizeof (ioc32), mode)) { 6496 retval = EFAULT; 6497 break; 6498 } 6499 piocsc->client = (caddr_t)(uintptr_t)ioc32.client; 6500 piocsc->class = (caddr_t)(uintptr_t)ioc32.class; 6501 break; 6502 } 6503 6504 case DDI_MODEL_NONE: 6505 if (ddi_copyin(data, piocsc, sizeof (*piocsc), mode)) { 6506 retval = EFAULT; 6507 } 6508 break; 6509 } 6510 #else /* _MULTI_DATAMODEL */ 6511 if (ddi_copyin(data, piocsc, sizeof (*piocsc), mode)) { 6512 retval = EFAULT; 6513 } 6514 #endif /* _MULTI_DATAMODEL */ 6515 6516 #ifdef DEBUG 6517 if (retval) { 6518 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: cmd <%s> " 6519 "switch_to_cntlr_iocdata copyin failed", s)); 6520 } 6521 #endif 6522 6523 return (retval); 6524 } 6525 6526 6527 /* 6528 * Routine to allocate memory for the path information structures. 6529 * It allocates two chunks of memory - one for keeping userland 6530 * pointers/values for path information and path properties, second for 6531 * keeping allocating kernel memory for path properties. These path 6532 * properties are finally copied to userland. 6533 */ 6534 /* ARGSUSED */ 6535 static int 6536 vhci_ioc_alloc_pathinfo(sv_path_info_t **upibuf, sv_path_info_t **kpibuf, 6537 uint_t num_paths, sv_iocdata_t *pioc, int mode, caddr_t s) 6538 { 6539 sv_path_info_t *pi; 6540 uint_t bufsize; 6541 int retval = 0; 6542 int index; 6543 6544 /* Allocate memory */ 6545 *upibuf = (sv_path_info_t *) 6546 kmem_zalloc(sizeof (sv_path_info_t) * num_paths, KM_SLEEP); 6547 ASSERT(*upibuf != NULL); 6548 *kpibuf = (sv_path_info_t *) 6549 kmem_zalloc(sizeof (sv_path_info_t) * num_paths, KM_SLEEP); 6550 ASSERT(*kpibuf != NULL); 6551 6552 /* 6553 * Get the path info structure from the user space. 6554 * We are interested in the following fields: 6555 * - user size of buffer for per path properties. 6556 * - user address of buffer for path info properties. 6557 * - user pointer for returning actual buffer size 6558 * Keep these fields in the 'upibuf' structures. 6559 * Allocate buffer for per path info properties in kernel 6560 * structure ('kpibuf'). 6561 * Size of these buffers will be equal to the size of buffers 6562 * in the user space. 6563 */ 6564 #ifdef _MULTI_DATAMODEL 6565 switch (ddi_model_convert_from(mode & FMODELS)) { 6566 case DDI_MODEL_ILP32: 6567 { 6568 sv_path_info32_t *src; 6569 sv_path_info32_t pi32; 6570 6571 src = (sv_path_info32_t *)pioc->ret_buf; 6572 pi = (sv_path_info_t *)*upibuf; 6573 for (index = 0; index < num_paths; index++, src++, pi++) { 6574 if (ddi_copyin(src, &pi32, sizeof (pi32), mode)) { 6575 retval = EFAULT; 6576 break; 6577 } 6578 6579 pi->ret_prop.buf_size = 6580 (uint_t)pi32.ret_prop.buf_size; 6581 pi->ret_prop.ret_buf_size = 6582 (uint_t *)(uintptr_t)pi32.ret_prop.ret_buf_size; 6583 pi->ret_prop.buf = 6584 (caddr_t)(uintptr_t)pi32.ret_prop.buf; 6585 } 6586 break; 6587 } 6588 6589 case DDI_MODEL_NONE: 6590 if (ddi_copyin(pioc->ret_buf, *upibuf, 6591 sizeof (sv_path_info_t) * num_paths, mode)) { 6592 retval = EFAULT; 6593 } 6594 break; 6595 } 6596 #else /* _MULTI_DATAMODEL */ 6597 if (ddi_copyin(pioc->ret_buf, *upibuf, 6598 sizeof (sv_path_info_t) * num_paths, mode)) { 6599 retval = EFAULT; 6600 } 6601 #endif /* _MULTI_DATAMODEL */ 6602 6603 if (retval != 0) { 6604 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_alloc_path_info: " 6605 "ioctl <%s> normal: path_info copyin failed", s)); 6606 kmem_free(*upibuf, sizeof (sv_path_info_t) * num_paths); 6607 kmem_free(*kpibuf, sizeof (sv_path_info_t) * num_paths); 6608 *upibuf = NULL; 6609 *kpibuf = NULL; 6610 return (retval); 6611 } 6612 6613 /* 6614 * Allocate memory for per path properties. 6615 */ 6616 for (index = 0, pi = *kpibuf; index < num_paths; index++, pi++) { 6617 bufsize = (*upibuf)[index].ret_prop.buf_size; 6618 6619 if (bufsize && bufsize <= SV_PROP_MAX_BUF_SIZE) { 6620 pi->ret_prop.buf_size = bufsize; 6621 pi->ret_prop.buf = (caddr_t) 6622 kmem_zalloc(bufsize, KM_SLEEP); 6623 ASSERT(pi->ret_prop.buf != NULL); 6624 } else { 6625 pi->ret_prop.buf_size = 0; 6626 pi->ret_prop.buf = NULL; 6627 } 6628 6629 if ((*upibuf)[index].ret_prop.ret_buf_size != NULL) { 6630 pi->ret_prop.ret_buf_size = (uint_t *)kmem_zalloc( 6631 sizeof (*pi->ret_prop.ret_buf_size), KM_SLEEP); 6632 ASSERT(pi->ret_prop.ret_buf_size != NULL); 6633 } else { 6634 pi->ret_prop.ret_buf_size = NULL; 6635 } 6636 } 6637 6638 return (0); 6639 } 6640 6641 6642 /* 6643 * Routine to free memory for the path information structures. 6644 * This is the memory which was allocated earlier. 6645 */ 6646 /* ARGSUSED */ 6647 static void 6648 vhci_ioc_free_pathinfo(sv_path_info_t *upibuf, sv_path_info_t *kpibuf, 6649 uint_t num_paths) 6650 { 6651 sv_path_info_t *pi; 6652 int index; 6653 6654 /* Free memory for per path properties */ 6655 for (index = 0, pi = kpibuf; index < num_paths; index++, pi++) { 6656 if (pi->ret_prop.ret_buf_size != NULL) { 6657 kmem_free(pi->ret_prop.ret_buf_size, 6658 sizeof (*pi->ret_prop.ret_buf_size)); 6659 } 6660 6661 if (pi->ret_prop.buf != NULL) { 6662 kmem_free(pi->ret_prop.buf, pi->ret_prop.buf_size); 6663 } 6664 } 6665 6666 /* Free memory for path info structures */ 6667 kmem_free(upibuf, sizeof (sv_path_info_t) * num_paths); 6668 kmem_free(kpibuf, sizeof (sv_path_info_t) * num_paths); 6669 } 6670 6671 6672 /* 6673 * Routine to copy path information and path properties to userland. 6674 */ 6675 /* ARGSUSED */ 6676 static int 6677 vhci_ioc_send_pathinfo(sv_path_info_t *upibuf, sv_path_info_t *kpibuf, 6678 uint_t num_paths, sv_iocdata_t *pioc, int mode, caddr_t s) 6679 { 6680 int retval = 0, index; 6681 sv_path_info_t *upi_ptr; 6682 sv_path_info32_t *upi32_ptr; 6683 6684 #ifdef _MULTI_DATAMODEL 6685 switch (ddi_model_convert_from(mode & FMODELS)) { 6686 case DDI_MODEL_ILP32: 6687 goto copy_32bit; 6688 6689 case DDI_MODEL_NONE: 6690 goto copy_normal; 6691 } 6692 #else /* _MULTI_DATAMODEL */ 6693 6694 goto copy_normal; 6695 6696 #endif /* _MULTI_DATAMODEL */ 6697 6698 copy_normal: 6699 6700 /* 6701 * Copy path information and path properties to user land. 6702 * Pointer fields inside the path property structure were 6703 * saved in the 'upibuf' structure earlier. 6704 */ 6705 upi_ptr = pioc->ret_buf; 6706 for (index = 0; index < num_paths; index++) { 6707 if (ddi_copyout(kpibuf[index].device.ret_ct, 6708 upi_ptr[index].device.ret_ct, MAXPATHLEN, mode)) { 6709 retval = EFAULT; 6710 break; 6711 } 6712 6713 if (ddi_copyout(kpibuf[index].ret_addr, 6714 upi_ptr[index].ret_addr, MAXNAMELEN, mode)) { 6715 retval = EFAULT; 6716 break; 6717 } 6718 6719 if (ddi_copyout(&kpibuf[index].ret_state, 6720 &upi_ptr[index].ret_state, sizeof (kpibuf[index].ret_state), 6721 mode)) { 6722 retval = EFAULT; 6723 break; 6724 } 6725 6726 if (ddi_copyout(&kpibuf[index].ret_ext_state, 6727 &upi_ptr[index].ret_ext_state, 6728 sizeof (kpibuf[index].ret_ext_state), mode)) { 6729 retval = EFAULT; 6730 break; 6731 } 6732 6733 if ((kpibuf[index].ret_prop.ret_buf_size != NULL) && 6734 ddi_copyout(kpibuf[index].ret_prop.ret_buf_size, 6735 upibuf[index].ret_prop.ret_buf_size, 6736 sizeof (*upibuf[index].ret_prop.ret_buf_size), mode)) { 6737 retval = EFAULT; 6738 break; 6739 } 6740 6741 if ((kpibuf[index].ret_prop.buf != NULL) && 6742 ddi_copyout(kpibuf[index].ret_prop.buf, 6743 upibuf[index].ret_prop.buf, 6744 upibuf[index].ret_prop.buf_size, mode)) { 6745 retval = EFAULT; 6746 break; 6747 } 6748 } 6749 6750 #ifdef DEBUG 6751 if (retval) { 6752 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: ioctl <%s> " 6753 "normal: path_info copyout failed", s)); 6754 } 6755 #endif 6756 6757 return (retval); 6758 6759 copy_32bit: 6760 /* 6761 * Copy path information and path properties to user land. 6762 * Pointer fields inside the path property structure were 6763 * saved in the 'upibuf' structure earlier. 6764 */ 6765 upi32_ptr = (sv_path_info32_t *)pioc->ret_buf; 6766 for (index = 0; index < num_paths; index++) { 6767 if (ddi_copyout(kpibuf[index].device.ret_ct, 6768 upi32_ptr[index].device.ret_ct, MAXPATHLEN, mode)) { 6769 retval = EFAULT; 6770 break; 6771 } 6772 6773 if (ddi_copyout(kpibuf[index].ret_addr, 6774 upi32_ptr[index].ret_addr, MAXNAMELEN, mode)) { 6775 retval = EFAULT; 6776 break; 6777 } 6778 6779 if (ddi_copyout(&kpibuf[index].ret_state, 6780 &upi32_ptr[index].ret_state, 6781 sizeof (kpibuf[index].ret_state), mode)) { 6782 retval = EFAULT; 6783 break; 6784 } 6785 6786 if (ddi_copyout(&kpibuf[index].ret_ext_state, 6787 &upi32_ptr[index].ret_ext_state, 6788 sizeof (kpibuf[index].ret_ext_state), mode)) { 6789 retval = EFAULT; 6790 break; 6791 } 6792 if ((kpibuf[index].ret_prop.ret_buf_size != NULL) && 6793 ddi_copyout(kpibuf[index].ret_prop.ret_buf_size, 6794 upibuf[index].ret_prop.ret_buf_size, 6795 sizeof (*upibuf[index].ret_prop.ret_buf_size), mode)) { 6796 retval = EFAULT; 6797 break; 6798 } 6799 6800 if ((kpibuf[index].ret_prop.buf != NULL) && 6801 ddi_copyout(kpibuf[index].ret_prop.buf, 6802 upibuf[index].ret_prop.buf, 6803 upibuf[index].ret_prop.buf_size, mode)) { 6804 retval = EFAULT; 6805 break; 6806 } 6807 } 6808 6809 #ifdef DEBUG 6810 if (retval) { 6811 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_get_ioc: ioctl <%s> " 6812 "normal: path_info copyout failed", s)); 6813 } 6814 #endif 6815 6816 return (retval); 6817 } 6818 6819 6820 /* 6821 * vhci_failover() 6822 * This routine expects VHCI_HOLD_LUN before being invoked. It can be invoked 6823 * as MDI_FAILOVER_ASYNC or MDI_FAILOVER_SYNC. For Asynchronous failovers 6824 * this routine shall VHCI_RELEASE_LUN on exiting. For synchronous failovers 6825 * it is the callers responsibility to release lun. 6826 */ 6827 6828 /* ARGSUSED */ 6829 static int 6830 vhci_failover(dev_info_t *vdip, dev_info_t *cdip, int flags) 6831 { 6832 char *guid; 6833 scsi_vhci_lun_t *vlun = NULL; 6834 struct scsi_vhci *vhci; 6835 mdi_pathinfo_t *pip, *npip; 6836 char *s_pclass, *pclass1, *pclass2, *pclass; 6837 char active_pclass_copy[255], *active_pclass_ptr; 6838 char *ptr1, *ptr2; 6839 mdi_pathinfo_state_t pi_state; 6840 uint32_t pi_ext_state; 6841 scsi_vhci_priv_t *svp; 6842 struct scsi_device *sd; 6843 struct scsi_failover_ops *sfo; 6844 int sps; /* mdi_select_path() status */ 6845 int activation_done = 0; 6846 int rval, retval = MDI_FAILURE; 6847 int reserve_pending, check_condition, UA_condition; 6848 struct scsi_pkt *pkt; 6849 struct buf *bp; 6850 6851 vhci = ddi_get_soft_state(vhci_softstate, ddi_get_instance(vdip)); 6852 sd = ddi_get_driver_private(cdip); 6853 vlun = ADDR2VLUN(&sd->sd_address); 6854 ASSERT(vlun != 0); 6855 ASSERT(VHCI_LUN_IS_HELD(vlun)); 6856 guid = vlun->svl_lun_wwn; 6857 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(1): guid %s\n", guid)); 6858 vhci_log(CE_NOTE, vdip, "!Initiating failover for device %s " 6859 "(GUID %s)", ddi_node_name(cdip), guid); 6860 6861 /* 6862 * Lets maintain a local copy of the vlun->svl_active_pclass 6863 * for the rest of the processing. Accessing the field 6864 * directly in the loop below causes loop logic to break 6865 * especially when the field gets updated by other threads 6866 * update path status etc and causes 'paths are not currently 6867 * available' condition to be declared prematurely. 6868 */ 6869 mutex_enter(&vlun->svl_mutex); 6870 if (vlun->svl_active_pclass != NULL) { 6871 (void) strlcpy(active_pclass_copy, vlun->svl_active_pclass, 6872 sizeof (active_pclass_copy)); 6873 active_pclass_ptr = &active_pclass_copy[0]; 6874 mutex_exit(&vlun->svl_mutex); 6875 if (vhci_quiesce_paths(vdip, cdip, vlun, guid, 6876 active_pclass_ptr) != 0) { 6877 retval = MDI_FAILURE; 6878 } 6879 } else { 6880 /* 6881 * can happen only when the available path to device 6882 * discovered is a STANDBY path. 6883 */ 6884 mutex_exit(&vlun->svl_mutex); 6885 active_pclass_copy[0] = '\0'; 6886 active_pclass_ptr = NULL; 6887 } 6888 6889 sfo = vlun->svl_fops; 6890 ASSERT(sfo != NULL); 6891 pclass1 = s_pclass = active_pclass_ptr; 6892 VHCI_DEBUG(1, (CE_NOTE, NULL, "!(%s)failing over from %s\n", guid, 6893 (s_pclass == NULL ? "<none>" : s_pclass))); 6894 6895 next_pathclass: 6896 6897 rval = sfo->sfo_pathclass_next(pclass1, &pclass2, 6898 vlun->svl_fops_ctpriv); 6899 if (rval == ENOENT) { 6900 if (s_pclass == NULL) { 6901 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(4)(%s): " 6902 "failed, no more pathclasses\n", guid)); 6903 goto done; 6904 } else { 6905 (void) sfo->sfo_pathclass_next(NULL, &pclass2, 6906 vlun->svl_fops_ctpriv); 6907 } 6908 } else if (rval == EINVAL) { 6909 vhci_log(CE_NOTE, vdip, "!Failover operation failed for " 6910 "device %s (GUID %s): Invalid path-class %s", 6911 ddi_node_name(cdip), guid, 6912 ((pclass1 == NULL) ? "<none>" : pclass1)); 6913 goto done; 6914 } 6915 if ((s_pclass != NULL) && (strcmp(pclass2, s_pclass) == 0)) { 6916 /* 6917 * paths are not currently available 6918 */ 6919 vhci_log(CE_NOTE, vdip, "!Failover path currently unavailable" 6920 " for device %s (GUID %s)", 6921 ddi_node_name(cdip), guid); 6922 goto done; 6923 } 6924 pip = npip = NULL; 6925 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(5.2)(%s): considering " 6926 "%s as failover destination\n", guid, pclass2)); 6927 sps = mdi_select_path(cdip, NULL, MDI_SELECT_STANDBY_PATH, NULL, &npip); 6928 if ((npip == NULL) || (sps != MDI_SUCCESS)) { 6929 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(%s): no " 6930 "STANDBY paths found (status:%x)!\n", guid, sps)); 6931 pclass1 = pclass2; 6932 goto next_pathclass; 6933 } 6934 do { 6935 pclass = NULL; 6936 if ((mdi_prop_lookup_string(npip, "path-class", 6937 &pclass) != MDI_SUCCESS) || (strcmp(pclass2, 6938 pclass) != 0)) { 6939 VHCI_DEBUG(1, (CE_NOTE, NULL, 6940 "!vhci_failover(5.5)(%s): skipping path " 6941 "%p(%s)...\n", guid, (void *)npip, pclass)); 6942 pip = npip; 6943 sps = mdi_select_path(cdip, NULL, 6944 MDI_SELECT_STANDBY_PATH, pip, &npip); 6945 mdi_rele_path(pip); 6946 (void) mdi_prop_free(pclass); 6947 continue; 6948 } 6949 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(npip); 6950 6951 /* 6952 * Issue READ at non-zer block on this STANDBY path. 6953 * Purple returns 6954 * 1. RESERVATION_CONFLICT if reservation is pending 6955 * 2. POR check condition if it reset happened. 6956 * 2. failover Check Conditions if one is already in progress. 6957 */ 6958 reserve_pending = 0; 6959 check_condition = 0; 6960 UA_condition = 0; 6961 6962 bp = scsi_alloc_consistent_buf(&svp->svp_psd->sd_address, 6963 (struct buf *)NULL, DEV_BSIZE, B_READ, NULL, NULL); 6964 if (!bp) { 6965 VHCI_DEBUG(1, (CE_NOTE, NULL, 6966 "vhci_failover !No resources (buf)\n")); 6967 mdi_rele_path(npip); 6968 goto done; 6969 } 6970 pkt = scsi_init_pkt(&svp->svp_psd->sd_address, NULL, bp, 6971 CDB_GROUP1, sizeof (struct scsi_arq_status), 0, 6972 PKT_CONSISTENT, NULL, NULL); 6973 if (pkt) { 6974 (void) scsi_setup_cdb((union scsi_cdb *)(uintptr_t) 6975 pkt->pkt_cdbp, SCMD_READ, 1, 1, 0); 6976 pkt->pkt_flags = FLAG_NOINTR; 6977 check_path_again: 6978 pkt->pkt_path_instance = mdi_pi_get_path_instance(npip); 6979 pkt->pkt_time = 3*30; 6980 6981 if (scsi_transport(pkt) == TRAN_ACCEPT) { 6982 switch (pkt->pkt_reason) { 6983 case CMD_CMPLT: 6984 switch (SCBP_C(pkt)) { 6985 case STATUS_GOOD: 6986 /* Already failed over */ 6987 activation_done = 1; 6988 break; 6989 case STATUS_RESERVATION_CONFLICT: 6990 reserve_pending = 1; 6991 break; 6992 case STATUS_CHECK: 6993 check_condition = 1; 6994 break; 6995 } 6996 } 6997 } 6998 if (check_condition && 6999 (pkt->pkt_state & STATE_ARQ_DONE)) { 7000 uint8_t *sns, skey, asc, ascq; 7001 sns = (uint8_t *) 7002 &(((struct scsi_arq_status *)(uintptr_t) 7003 (pkt->pkt_scbp))->sts_sensedata); 7004 skey = scsi_sense_key(sns); 7005 asc = scsi_sense_asc(sns); 7006 ascq = scsi_sense_ascq(sns); 7007 if (skey == KEY_UNIT_ATTENTION && 7008 asc == 0x29) { 7009 /* Already failed over */ 7010 VHCI_DEBUG(1, (CE_NOTE, NULL, 7011 "!vhci_failover(7)(%s): " 7012 "path 0x%p POR UA condition\n", 7013 guid, (void *)npip)); 7014 if (UA_condition == 0) { 7015 UA_condition = 1; 7016 goto check_path_again; 7017 } 7018 } else { 7019 activation_done = 0; 7020 VHCI_DEBUG(1, (CE_NOTE, NULL, 7021 "!vhci_failover(%s): path 0x%p " 7022 "unhandled chkcond %x %x %x\n", 7023 guid, (void *)npip, skey, 7024 asc, ascq)); 7025 } 7026 } 7027 scsi_destroy_pkt(pkt); 7028 } 7029 scsi_free_consistent_buf(bp); 7030 7031 if (activation_done) { 7032 mdi_rele_path(npip); 7033 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(7)(%s): " 7034 "path 0x%p already failedover\n", guid, 7035 (void *)npip)); 7036 break; 7037 } 7038 if (reserve_pending && (vlun->svl_xlf_capable == 0)) { 7039 (void) vhci_recovery_reset(vlun, 7040 &svp->svp_psd->sd_address, 7041 FALSE, VHCI_DEPTH_ALL); 7042 } 7043 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(6)(%s): " 7044 "activating path 0x%p(psd:%p)\n", guid, (void *)npip, 7045 (void *)svp->svp_psd)); 7046 if (sfo->sfo_path_activate(svp->svp_psd, pclass2, 7047 vlun->svl_fops_ctpriv) == 0) { 7048 activation_done = 1; 7049 mdi_rele_path(npip); 7050 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(7)(%s): " 7051 "path 0x%p successfully activated\n", guid, 7052 (void *)npip)); 7053 break; 7054 } 7055 pip = npip; 7056 sps = mdi_select_path(cdip, NULL, MDI_SELECT_STANDBY_PATH, 7057 pip, &npip); 7058 mdi_rele_path(pip); 7059 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 7060 if (activation_done == 0) { 7061 pclass1 = pclass2; 7062 goto next_pathclass; 7063 } 7064 7065 /* 7066 * if we are here, we have succeeded in activating path npip of 7067 * pathclass pclass2; let us validate all paths of pclass2 by 7068 * "ping"-ing each one and mark the good ones ONLINE 7069 * Also, set the state of the paths belonging to the previously 7070 * active pathclass to STANDBY 7071 */ 7072 pip = npip = NULL; 7073 sps = mdi_select_path(cdip, NULL, (MDI_SELECT_ONLINE_PATH | 7074 MDI_SELECT_STANDBY_PATH | MDI_SELECT_USER_DISABLE_PATH), 7075 NULL, &npip); 7076 if (npip == NULL || sps != MDI_SUCCESS) { 7077 VHCI_DEBUG(1, (CE_NOTE, NULL, "!Failover operation failed for " 7078 "device %s (GUID %s): paths may be busy\n", 7079 ddi_node_name(cdip), guid)); 7080 goto done; 7081 } 7082 do { 7083 (void) mdi_pi_get_state2(npip, &pi_state, &pi_ext_state); 7084 if (mdi_prop_lookup_string(npip, "path-class", &pclass) 7085 != MDI_SUCCESS) { 7086 pip = npip; 7087 sps = mdi_select_path(cdip, NULL, 7088 (MDI_SELECT_ONLINE_PATH | 7089 MDI_SELECT_STANDBY_PATH | 7090 MDI_SELECT_USER_DISABLE_PATH), 7091 pip, &npip); 7092 mdi_rele_path(pip); 7093 continue; 7094 } 7095 if (strcmp(pclass, pclass2) == 0) { 7096 if (pi_state == MDI_PATHINFO_STATE_STANDBY) { 7097 svp = (scsi_vhci_priv_t *) 7098 mdi_pi_get_vhci_private(npip); 7099 VHCI_DEBUG(1, (CE_NOTE, NULL, 7100 "!vhci_failover(8)(%s): " 7101 "pinging path 0x%p\n", 7102 guid, (void *)npip)); 7103 if (sfo->sfo_path_ping(svp->svp_psd, 7104 vlun->svl_fops_ctpriv) == 1) { 7105 mdi_pi_set_state(npip, 7106 MDI_PATHINFO_STATE_ONLINE); 7107 VHCI_DEBUG(1, (CE_NOTE, NULL, 7108 "!vhci_failover(9)(%s): " 7109 "path 0x%p ping successful, " 7110 "marked online\n", guid, 7111 (void *)npip)); 7112 MDI_PI_ERRSTAT(npip, MDI_PI_FAILTO); 7113 } 7114 } 7115 } else if ((s_pclass != NULL) && (strcmp(pclass, s_pclass) 7116 == 0)) { 7117 if (pi_state == MDI_PATHINFO_STATE_ONLINE) { 7118 mdi_pi_set_state(npip, 7119 MDI_PATHINFO_STATE_STANDBY); 7120 VHCI_DEBUG(1, (CE_NOTE, NULL, 7121 "!vhci_failover(10)(%s): path 0x%p marked " 7122 "STANDBY\n", guid, (void *)npip)); 7123 MDI_PI_ERRSTAT(npip, MDI_PI_FAILFROM); 7124 } 7125 } 7126 (void) mdi_prop_free(pclass); 7127 pip = npip; 7128 sps = mdi_select_path(cdip, NULL, (MDI_SELECT_ONLINE_PATH | 7129 MDI_SELECT_STANDBY_PATH|MDI_SELECT_USER_DISABLE_PATH), 7130 pip, &npip); 7131 mdi_rele_path(pip); 7132 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 7133 7134 /* 7135 * Update the AccessState of related MP-API TPGs 7136 */ 7137 (void) vhci_mpapi_update_tpg_acc_state_for_lu(vhci, vlun); 7138 7139 vhci_log(CE_NOTE, vdip, "!Failover operation completed successfully " 7140 "for device %s (GUID %s): failed over from %s to %s", 7141 ddi_node_name(cdip), guid, ((s_pclass == NULL) ? "<none>" : 7142 s_pclass), pclass2); 7143 ptr1 = kmem_alloc(strlen(pclass2)+1, KM_SLEEP); 7144 (void) strlcpy(ptr1, pclass2, (strlen(pclass2)+1)); 7145 mutex_enter(&vlun->svl_mutex); 7146 ptr2 = vlun->svl_active_pclass; 7147 vlun->svl_active_pclass = ptr1; 7148 mutex_exit(&vlun->svl_mutex); 7149 if (ptr2) { 7150 kmem_free(ptr2, strlen(ptr2)+1); 7151 } 7152 mutex_enter(&vhci->vhci_mutex); 7153 scsi_hba_reset_notify_callback(&vhci->vhci_mutex, 7154 &vhci->vhci_reset_notify_listf); 7155 /* All reservations are cleared upon these resets. */ 7156 vlun->svl_flags &= ~VLUN_RESERVE_ACTIVE_FLG; 7157 mutex_exit(&vhci->vhci_mutex); 7158 VHCI_DEBUG(1, (CE_NOTE, NULL, "!vhci_failover(11): DONE! Active " 7159 "pathclass for %s is now %s\n", guid, pclass2)); 7160 retval = MDI_SUCCESS; 7161 7162 done: 7163 vlun->svl_failover_status = retval; 7164 if (flags == MDI_FAILOVER_ASYNC) { 7165 VHCI_RELEASE_LUN(vlun); 7166 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_failover(12): DONE! " 7167 "releasing lun, as failover was ASYNC\n")); 7168 } else { 7169 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_failover(12): DONE! " 7170 "NOT releasing lun, as failover was SYNC\n")); 7171 } 7172 return (retval); 7173 } 7174 7175 /* 7176 * vhci_client_attached is called after the successful attach of a 7177 * client devinfo node. 7178 */ 7179 static void 7180 vhci_client_attached(dev_info_t *cdip) 7181 { 7182 mdi_pathinfo_t *pip; 7183 int circular; 7184 7185 /* 7186 * At this point the client has attached and it's instance number is 7187 * valid, so we can set up kstats. We need to do this here because it 7188 * is possible for paths to go online prior to client attach, in which 7189 * case the call to vhci_kstat_create_pathinfo in vhci_pathinfo_online 7190 * was a noop. 7191 */ 7192 ndi_devi_enter(cdip, &circular); 7193 for (pip = mdi_get_next_phci_path(cdip, NULL); pip; 7194 pip = mdi_get_next_phci_path(cdip, pip)) 7195 vhci_kstat_create_pathinfo(pip); 7196 ndi_devi_exit(cdip, circular); 7197 } 7198 7199 /* 7200 * quiesce all of the online paths 7201 */ 7202 static int 7203 vhci_quiesce_paths(dev_info_t *vdip, dev_info_t *cdip, scsi_vhci_lun_t *vlun, 7204 char *guid, char *active_pclass_ptr) 7205 { 7206 scsi_vhci_priv_t *svp; 7207 char *s_pclass = NULL; 7208 mdi_pathinfo_t *npip, *pip; 7209 int sps; 7210 7211 /* quiesce currently active paths */ 7212 s_pclass = NULL; 7213 pip = npip = NULL; 7214 sps = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH, NULL, &npip); 7215 if ((npip == NULL) || (sps != MDI_SUCCESS)) { 7216 return (1); 7217 } 7218 do { 7219 if (mdi_prop_lookup_string(npip, "path-class", 7220 &s_pclass) != MDI_SUCCESS) { 7221 mdi_rele_path(npip); 7222 vhci_log(CE_NOTE, vdip, "!Failover operation failed " 7223 "for device %s (GUID %s) due to an internal " 7224 "error", ddi_node_name(cdip), guid); 7225 return (1); 7226 } 7227 if (strcmp(s_pclass, active_pclass_ptr) == 0) { 7228 /* 7229 * quiesce path. Free s_pclass since 7230 * we don't need it anymore 7231 */ 7232 VHCI_DEBUG(1, (CE_NOTE, NULL, 7233 "!vhci_failover(2)(%s): failing over " 7234 "from %s; quiescing path %p\n", 7235 guid, s_pclass, (void *)npip)); 7236 (void) mdi_prop_free(s_pclass); 7237 svp = (scsi_vhci_priv_t *) 7238 mdi_pi_get_vhci_private(npip); 7239 if (svp == NULL) { 7240 VHCI_DEBUG(1, (CE_NOTE, NULL, 7241 "!vhci_failover(2.5)(%s): no " 7242 "client priv! %p offlined?\n", 7243 guid, (void *)npip)); 7244 pip = npip; 7245 sps = mdi_select_path(cdip, NULL, 7246 MDI_SELECT_ONLINE_PATH, pip, &npip); 7247 mdi_rele_path(pip); 7248 continue; 7249 } 7250 if (scsi_abort(&svp->svp_psd->sd_address, NULL) 7251 == 0) { 7252 (void) vhci_recovery_reset(vlun, 7253 &svp->svp_psd->sd_address, FALSE, 7254 VHCI_DEPTH_TARGET); 7255 } 7256 mutex_enter(&svp->svp_mutex); 7257 if (svp->svp_cmds == 0) { 7258 VHCI_DEBUG(1, (CE_NOTE, NULL, 7259 "!vhci_failover(3)(%s):" 7260 "quiesced path %p\n", guid, (void *)npip)); 7261 } else { 7262 while (svp->svp_cmds != 0) { 7263 cv_wait(&svp->svp_cv, &svp->svp_mutex); 7264 VHCI_DEBUG(1, (CE_NOTE, NULL, 7265 "!vhci_failover(3.cv)(%s):" 7266 "quiesced path %p\n", guid, 7267 (void *)npip)); 7268 } 7269 } 7270 mutex_exit(&svp->svp_mutex); 7271 } else { 7272 /* 7273 * make sure we freeup the memory 7274 */ 7275 (void) mdi_prop_free(s_pclass); 7276 } 7277 pip = npip; 7278 sps = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH, 7279 pip, &npip); 7280 mdi_rele_path(pip); 7281 } while ((npip != NULL) && (sps == MDI_SUCCESS)); 7282 return (0); 7283 } 7284 7285 static struct scsi_vhci_lun * 7286 vhci_lun_lookup(dev_info_t *tgt_dip) 7287 { 7288 return ((struct scsi_vhci_lun *) 7289 mdi_client_get_vhci_private(tgt_dip)); 7290 } 7291 7292 static struct scsi_vhci_lun * 7293 vhci_lun_lookup_alloc(dev_info_t *tgt_dip, char *guid, int *didalloc) 7294 { 7295 struct scsi_vhci_lun *svl; 7296 7297 if (svl = vhci_lun_lookup(tgt_dip)) { 7298 return (svl); 7299 } 7300 7301 svl = kmem_zalloc(sizeof (*svl), KM_SLEEP); 7302 svl->svl_lun_wwn = kmem_zalloc(strlen(guid)+1, KM_SLEEP); 7303 (void) strcpy(svl->svl_lun_wwn, guid); 7304 mutex_init(&svl->svl_mutex, NULL, MUTEX_DRIVER, NULL); 7305 cv_init(&svl->svl_cv, NULL, CV_DRIVER, NULL); 7306 sema_init(&svl->svl_pgr_sema, 1, NULL, SEMA_DRIVER, NULL); 7307 svl->svl_waiting_for_activepath = 1; 7308 svl->svl_sector_size = 1; 7309 mdi_client_set_vhci_private(tgt_dip, svl); 7310 *didalloc = 1; 7311 VHCI_DEBUG(1, (CE_NOTE, NULL, 7312 "vhci_lun_lookup_alloc: guid %s vlun 0x%p\n", 7313 guid, (void *)svl)); 7314 return (svl); 7315 } 7316 7317 static void 7318 vhci_lun_free(struct scsi_vhci_lun *dvlp, struct scsi_device *sd) 7319 { 7320 char *guid; 7321 7322 guid = dvlp->svl_lun_wwn; 7323 ASSERT(guid != NULL); 7324 VHCI_DEBUG(4, (CE_NOTE, NULL, "!vhci_lun_free: %s\n", guid)); 7325 7326 mutex_enter(&dvlp->svl_mutex); 7327 if (dvlp->svl_active_pclass != NULL) { 7328 kmem_free(dvlp->svl_active_pclass, 7329 strlen(dvlp->svl_active_pclass)+1); 7330 } 7331 dvlp->svl_active_pclass = NULL; 7332 mutex_exit(&dvlp->svl_mutex); 7333 7334 if (dvlp->svl_lun_wwn != NULL) { 7335 kmem_free(dvlp->svl_lun_wwn, strlen(dvlp->svl_lun_wwn)+1); 7336 } 7337 dvlp->svl_lun_wwn = NULL; 7338 7339 if (dvlp->svl_fops_name) { 7340 kmem_free(dvlp->svl_fops_name, strlen(dvlp->svl_fops_name)+1); 7341 } 7342 dvlp->svl_fops_name = NULL; 7343 7344 if (dvlp->svl_fops_ctpriv != NULL && 7345 dvlp->svl_fops != NULL) { 7346 dvlp->svl_fops->sfo_device_unprobe(sd, dvlp->svl_fops_ctpriv); 7347 } 7348 7349 if (dvlp->svl_flags & VLUN_TASK_D_ALIVE_FLG) 7350 taskq_destroy(dvlp->svl_taskq); 7351 7352 mutex_destroy(&dvlp->svl_mutex); 7353 cv_destroy(&dvlp->svl_cv); 7354 sema_destroy(&dvlp->svl_pgr_sema); 7355 kmem_free(dvlp, sizeof (*dvlp)); 7356 /* 7357 * vhci_lun_free may be called before the tgt_dip 7358 * initialization so check if the sd is NULL. 7359 */ 7360 if (sd != NULL) 7361 scsi_device_hba_private_set(sd, NULL); 7362 } 7363 7364 int 7365 vhci_do_scsi_cmd(struct scsi_pkt *pkt) 7366 { 7367 int err = 0; 7368 int retry_cnt = 0; 7369 uint8_t *sns, skey; 7370 7371 #ifdef DEBUG 7372 if (vhci_debug > 5) { 7373 vhci_print_cdb(pkt->pkt_address.a_hba_tran->tran_hba_dip, 7374 CE_WARN, "Vhci command", pkt->pkt_cdbp); 7375 } 7376 #endif 7377 7378 retry: 7379 err = scsi_poll(pkt); 7380 if (err) { 7381 if (pkt->pkt_cdbp[0] == SCMD_RELEASE) { 7382 if (SCBP_C(pkt) == STATUS_RESERVATION_CONFLICT) { 7383 VHCI_DEBUG(1, (CE_NOTE, NULL, 7384 "!v_s_do_s_c: RELEASE conflict\n")); 7385 return (0); 7386 } 7387 } 7388 if (retry_cnt++ < 6) { 7389 VHCI_DEBUG(1, (CE_WARN, NULL, 7390 "!v_s_do_s_c:retry packet 0x%p " 7391 "status 0x%x reason %s", 7392 (void *)pkt, SCBP_C(pkt), 7393 scsi_rname(pkt->pkt_reason))); 7394 if ((pkt->pkt_reason == CMD_CMPLT) && 7395 (SCBP_C(pkt) == STATUS_CHECK) && 7396 (pkt->pkt_state & STATE_ARQ_DONE)) { 7397 sns = (uint8_t *) 7398 &(((struct scsi_arq_status *)(uintptr_t) 7399 (pkt->pkt_scbp))->sts_sensedata); 7400 skey = scsi_sense_key(sns); 7401 VHCI_DEBUG(1, (CE_WARN, NULL, 7402 "!v_s_do_s_c:retry " 7403 "packet 0x%p sense data %s", (void *)pkt, 7404 scsi_sname(skey))); 7405 } 7406 goto retry; 7407 } 7408 VHCI_DEBUG(1, (CE_WARN, NULL, 7409 "!v_s_do_s_c: failed transport 0x%p 0x%x", 7410 (void *)pkt, SCBP_C(pkt))); 7411 return (0); 7412 } 7413 7414 switch (pkt->pkt_reason) { 7415 case CMD_TIMEOUT: 7416 VHCI_DEBUG(1, (CE_WARN, NULL, "!pkt timed " 7417 "out (pkt 0x%p)", (void *)pkt)); 7418 return (0); 7419 case CMD_CMPLT: 7420 switch (SCBP_C(pkt)) { 7421 case STATUS_GOOD: 7422 break; 7423 case STATUS_CHECK: 7424 if (pkt->pkt_state & STATE_ARQ_DONE) { 7425 sns = (uint8_t *)&((( 7426 struct scsi_arq_status *) 7427 (uintptr_t) 7428 (pkt->pkt_scbp))-> 7429 sts_sensedata); 7430 skey = scsi_sense_key(sns); 7431 if ((skey == 7432 KEY_UNIT_ATTENTION) || 7433 (skey == 7434 KEY_NOT_READY)) { 7435 /* 7436 * clear unit attn. 7437 */ 7438 7439 VHCI_DEBUG(1, 7440 (CE_WARN, NULL, 7441 "!v_s_do_s_c: " 7442 "retry " 7443 "packet 0x%p sense " 7444 "data %s", 7445 (void *)pkt, 7446 scsi_sname 7447 (skey))); 7448 goto retry; 7449 } 7450 VHCI_DEBUG(4, (CE_WARN, NULL, 7451 "!ARQ while " 7452 "transporting " 7453 "(pkt 0x%p)", 7454 (void *)pkt)); 7455 return (0); 7456 } 7457 return (0); 7458 default: 7459 VHCI_DEBUG(1, (CE_WARN, NULL, 7460 "!Bad status returned " 7461 "(pkt 0x%p, status %x)", 7462 (void *)pkt, SCBP_C(pkt))); 7463 return (0); 7464 } 7465 break; 7466 case CMD_INCOMPLETE: 7467 case CMD_RESET: 7468 case CMD_ABORTED: 7469 case CMD_TRAN_ERR: 7470 if (retry_cnt++ < 1) { 7471 VHCI_DEBUG(1, (CE_WARN, NULL, 7472 "!v_s_do_s_c: retry packet 0x%p %s", 7473 (void *)pkt, scsi_rname(pkt->pkt_reason))); 7474 goto retry; 7475 } 7476 /* FALLTHROUGH */ 7477 default: 7478 VHCI_DEBUG(1, (CE_WARN, NULL, "!pkt did not " 7479 "complete successfully (pkt 0x%p," 7480 "reason %x)", (void *)pkt, pkt->pkt_reason)); 7481 return (0); 7482 } 7483 return (1); 7484 } 7485 7486 static int 7487 vhci_quiesce_lun(struct scsi_vhci_lun *vlun) 7488 { 7489 mdi_pathinfo_t *pip, *spip; 7490 dev_info_t *cdip; 7491 struct scsi_vhci_priv *svp; 7492 mdi_pathinfo_state_t pstate; 7493 uint32_t p_ext_state; 7494 int circular; 7495 7496 cdip = vlun->svl_dip; 7497 pip = spip = NULL; 7498 ndi_devi_enter(cdip, &circular); 7499 pip = mdi_get_next_phci_path(cdip, NULL); 7500 while (pip != NULL) { 7501 (void) mdi_pi_get_state2(pip, &pstate, &p_ext_state); 7502 if (pstate != MDI_PATHINFO_STATE_ONLINE) { 7503 spip = pip; 7504 pip = mdi_get_next_phci_path(cdip, spip); 7505 continue; 7506 } 7507 mdi_hold_path(pip); 7508 ndi_devi_exit(cdip, circular); 7509 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 7510 mutex_enter(&svp->svp_mutex); 7511 while (svp->svp_cmds != 0) { 7512 if (cv_reltimedwait(&svp->svp_cv, &svp->svp_mutex, 7513 drv_usectohz(vhci_path_quiesce_timeout * 1000000), 7514 TR_CLOCK_TICK) == -1) { 7515 mutex_exit(&svp->svp_mutex); 7516 mdi_rele_path(pip); 7517 VHCI_DEBUG(1, (CE_WARN, NULL, 7518 "Quiesce of lun is not successful " 7519 "vlun: 0x%p.", (void *)vlun)); 7520 return (0); 7521 } 7522 } 7523 mutex_exit(&svp->svp_mutex); 7524 ndi_devi_enter(cdip, &circular); 7525 spip = pip; 7526 pip = mdi_get_next_phci_path(cdip, spip); 7527 mdi_rele_path(spip); 7528 } 7529 ndi_devi_exit(cdip, circular); 7530 return (1); 7531 } 7532 7533 static int 7534 vhci_pgr_validate_and_register(scsi_vhci_priv_t *svp) 7535 { 7536 scsi_vhci_lun_t *vlun; 7537 vhci_prout_t *prout; 7538 int rval, success; 7539 mdi_pathinfo_t *pip, *npip; 7540 scsi_vhci_priv_t *osvp; 7541 dev_info_t *cdip; 7542 uchar_t cdb_1; 7543 uchar_t temp_res_key[MHIOC_RESV_KEY_SIZE]; 7544 7545 7546 /* 7547 * see if there are any other paths available; if none, 7548 * then there is nothing to do. 7549 */ 7550 cdip = svp->svp_svl->svl_dip; 7551 rval = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH | 7552 MDI_SELECT_STANDBY_PATH, NULL, &pip); 7553 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 7554 VHCI_DEBUG(4, (CE_NOTE, NULL, 7555 "%s%d: vhci_pgr_validate_and_register: first path\n", 7556 ddi_driver_name(cdip), ddi_get_instance(cdip))); 7557 return (1); 7558 } 7559 7560 vlun = svp->svp_svl; 7561 prout = &vlun->svl_prout; 7562 ASSERT(vlun->svl_pgr_active != 0); 7563 7564 /* 7565 * When the path was busy/offlined, some other host might have 7566 * cleared this key. Validate key on some other path first. 7567 * If it fails, return failure. 7568 */ 7569 7570 npip = pip; 7571 pip = NULL; 7572 success = 0; 7573 7574 /* Save the res key */ 7575 bcopy(prout->res_key, temp_res_key, MHIOC_RESV_KEY_SIZE); 7576 7577 /* 7578 * Sometimes CDB from application can be a Register_And_Ignore. 7579 * Instead of validation, this cdb would result in force registration. 7580 * Convert it to normal cdb for validation. 7581 * After that be sure to restore the cdb. 7582 */ 7583 cdb_1 = vlun->svl_cdb[1]; 7584 vlun->svl_cdb[1] &= 0xe0; 7585 7586 do { 7587 osvp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(npip); 7588 if (osvp == NULL) { 7589 VHCI_DEBUG(4, (CE_NOTE, NULL, 7590 "vhci_pgr_validate_and_register: no " 7591 "client priv! 0x%p offlined?\n", 7592 (void *)npip)); 7593 goto next_path_1; 7594 } 7595 7596 if (osvp == svp) { 7597 VHCI_DEBUG(4, (CE_NOTE, NULL, 7598 "vhci_pgr_validate_and_register: same svp 0x%p" 7599 " npip 0x%p vlun 0x%p\n", 7600 (void *)svp, (void *)npip, (void *)vlun)); 7601 goto next_path_1; 7602 } 7603 7604 VHCI_DEBUG(4, (CE_NOTE, NULL, 7605 "vhci_pgr_validate_and_register: First validate on" 7606 " osvp 0x%p being done. vlun 0x%p thread 0x%p Before bcopy" 7607 " cdb1 %x\n", (void *)osvp, (void *)vlun, 7608 (void *)curthread, vlun->svl_cdb[1])); 7609 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy:"); 7610 7611 bcopy(prout->service_key, prout->res_key, MHIOC_RESV_KEY_SIZE); 7612 7613 VHCI_DEBUG(4, (CE_WARN, NULL, "vlun 0x%p After bcopy", 7614 (void *)vlun)); 7615 vhci_print_prout_keys(vlun, "v_pgr_val_reg: after bcopy: "); 7616 7617 rval = vhci_do_prout(osvp); 7618 if (rval == 1) { 7619 VHCI_DEBUG(4, (CE_NOTE, NULL, 7620 "%s%d: vhci_pgr_validate_and_register: key" 7621 " validated thread 0x%p\n", ddi_driver_name(cdip), 7622 ddi_get_instance(cdip), (void *)curthread)); 7623 pip = npip; 7624 success = 1; 7625 break; 7626 } else { 7627 VHCI_DEBUG(4, (CE_NOTE, NULL, 7628 "vhci_pgr_validate_and_register: First validation" 7629 " on osvp 0x%p failed %x\n", (void *)osvp, rval)); 7630 vhci_print_prout_keys(vlun, "v_pgr_val_reg: failed:"); 7631 } 7632 7633 /* 7634 * Try other paths 7635 */ 7636 next_path_1: 7637 pip = npip; 7638 rval = mdi_select_path(cdip, NULL, 7639 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, 7640 pip, &npip); 7641 mdi_rele_path(pip); 7642 } while ((rval == MDI_SUCCESS) && (npip != NULL)); 7643 7644 7645 /* Be sure to restore original cdb */ 7646 vlun->svl_cdb[1] = cdb_1; 7647 7648 /* Restore the res_key */ 7649 bcopy(temp_res_key, prout->res_key, MHIOC_RESV_KEY_SIZE); 7650 7651 /* 7652 * If key could not be registered on any path for the first time, 7653 * return success as online should still continue. 7654 */ 7655 if (success == 0) { 7656 return (1); 7657 } 7658 7659 ASSERT(pip != NULL); 7660 7661 /* 7662 * Force register on new path 7663 */ 7664 cdb_1 = vlun->svl_cdb[1]; /* store the cdb */ 7665 7666 vlun->svl_cdb[1] &= 0xe0; 7667 vlun->svl_cdb[1] |= VHCI_PROUT_R_AND_IGNORE; 7668 7669 vhci_print_prout_keys(vlun, "v_pgr_val_reg: keys before bcopy: "); 7670 7671 bcopy(prout->active_service_key, prout->service_key, 7672 MHIOC_RESV_KEY_SIZE); 7673 bcopy(prout->active_res_key, prout->res_key, MHIOC_RESV_KEY_SIZE); 7674 7675 vhci_print_prout_keys(vlun, "v_pgr_val_reg:keys after bcopy: "); 7676 7677 rval = vhci_do_prout(svp); 7678 vlun->svl_cdb[1] = cdb_1; /* restore the cdb */ 7679 if (rval != 1) { 7680 VHCI_DEBUG(4, (CE_NOTE, NULL, 7681 "vhci_pgr_validate_and_register: register on new" 7682 " path 0x%p svp 0x%p failed %x\n", 7683 (void *)pip, (void *)svp, rval)); 7684 vhci_print_prout_keys(vlun, "v_pgr_val_reg: reg failed: "); 7685 mdi_rele_path(pip); 7686 return (0); 7687 } 7688 7689 if (bcmp(prout->service_key, zero_key, MHIOC_RESV_KEY_SIZE) == 0) { 7690 VHCI_DEBUG(4, (CE_NOTE, NULL, 7691 "vhci_pgr_validate_and_register: zero service key\n")); 7692 mdi_rele_path(pip); 7693 return (rval); 7694 } 7695 7696 /* 7697 * While the key was force registered, some other host might have 7698 * cleared the key. Re-validate key on another pre-existing path 7699 * before declaring success. 7700 */ 7701 npip = pip; 7702 pip = NULL; 7703 7704 /* 7705 * Sometimes CDB from application can be Register and Ignore. 7706 * Instead of validation, it would result in force registration. 7707 * Convert it to normal cdb for validation. 7708 * After that be sure to restore the cdb. 7709 */ 7710 cdb_1 = vlun->svl_cdb[1]; 7711 vlun->svl_cdb[1] &= 0xe0; 7712 success = 0; 7713 7714 do { 7715 osvp = (scsi_vhci_priv_t *) 7716 mdi_pi_get_vhci_private(npip); 7717 if (osvp == NULL) { 7718 VHCI_DEBUG(4, (CE_NOTE, NULL, 7719 "vhci_pgr_validate_and_register: no " 7720 "client priv! 0x%p offlined?\n", 7721 (void *)npip)); 7722 goto next_path_2; 7723 } 7724 7725 if (osvp == svp) { 7726 VHCI_DEBUG(4, (CE_NOTE, NULL, 7727 "vhci_pgr_validate_and_register: same osvp 0x%p" 7728 " npip 0x%p vlun 0x%p\n", 7729 (void *)svp, (void *)npip, (void *)vlun)); 7730 goto next_path_2; 7731 } 7732 7733 VHCI_DEBUG(4, (CE_NOTE, NULL, 7734 "vhci_pgr_validate_and_register: Re-validation on" 7735 " osvp 0x%p being done. vlun 0x%p Before bcopy cdb1 %x\n", 7736 (void *)osvp, (void *)vlun, vlun->svl_cdb[1])); 7737 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy: "); 7738 7739 bcopy(prout->service_key, prout->res_key, MHIOC_RESV_KEY_SIZE); 7740 7741 vhci_print_prout_keys(vlun, "v_pgr_val_reg: after bcopy: "); 7742 7743 rval = vhci_do_prout(osvp); 7744 if (rval == 1) { 7745 VHCI_DEBUG(4, (CE_NOTE, NULL, 7746 "%s%d: vhci_pgr_validate_and_register: key" 7747 " validated thread 0x%p\n", ddi_driver_name(cdip), 7748 ddi_get_instance(cdip), (void *)curthread)); 7749 pip = npip; 7750 success = 1; 7751 break; 7752 } else { 7753 VHCI_DEBUG(4, (CE_NOTE, NULL, 7754 "vhci_pgr_validate_and_register: Re-validation on" 7755 " osvp 0x%p failed %x\n", (void *)osvp, rval)); 7756 vhci_print_prout_keys(vlun, 7757 "v_pgr_val_reg: reval failed: "); 7758 } 7759 7760 /* 7761 * Try other paths 7762 */ 7763 next_path_2: 7764 pip = npip; 7765 rval = mdi_select_path(cdip, NULL, 7766 MDI_SELECT_ONLINE_PATH|MDI_SELECT_STANDBY_PATH, 7767 pip, &npip); 7768 mdi_rele_path(pip); 7769 } while ((rval == MDI_SUCCESS) && (npip != NULL)); 7770 7771 /* Be sure to restore original cdb */ 7772 vlun->svl_cdb[1] = cdb_1; 7773 7774 if (success == 1) { 7775 /* Successfully validated registration */ 7776 mdi_rele_path(pip); 7777 return (1); 7778 } 7779 7780 VHCI_DEBUG(4, (CE_WARN, NULL, "key validation failed")); 7781 7782 /* 7783 * key invalid, back out by registering key value of 0 7784 */ 7785 VHCI_DEBUG(4, (CE_NOTE, NULL, 7786 "vhci_pgr_validate_and_register: backout on" 7787 " svp 0x%p being done\n", (void *)svp)); 7788 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy: "); 7789 7790 bcopy(prout->service_key, prout->res_key, MHIOC_RESV_KEY_SIZE); 7791 bzero(prout->service_key, MHIOC_RESV_KEY_SIZE); 7792 7793 vhci_print_prout_keys(vlun, "v_pgr_val_reg: before bcopy: "); 7794 7795 /* 7796 * Get a new path 7797 */ 7798 rval = mdi_select_path(cdip, NULL, MDI_SELECT_ONLINE_PATH | 7799 MDI_SELECT_STANDBY_PATH, NULL, &pip); 7800 if ((rval != MDI_SUCCESS) || (pip == NULL)) { 7801 VHCI_DEBUG(4, (CE_NOTE, NULL, 7802 "%s%d: vhci_pgr_validate_and_register: no valid pip\n", 7803 ddi_driver_name(cdip), ddi_get_instance(cdip))); 7804 return (0); 7805 } 7806 7807 if ((rval = vhci_do_prout(svp)) != 1) { 7808 VHCI_DEBUG(4, (CE_NOTE, NULL, 7809 "vhci_pgr_validate_and_register: backout on" 7810 " svp 0x%p failed\n", (void *)svp)); 7811 vhci_print_prout_keys(vlun, "backout failed"); 7812 7813 VHCI_DEBUG(4, (CE_WARN, NULL, 7814 "%s%d: vhci_pgr_validate_and_register: key" 7815 " validation and backout failed", ddi_driver_name(cdip), 7816 ddi_get_instance(cdip))); 7817 if (rval == VHCI_PGR_ILLEGALOP) { 7818 VHCI_DEBUG(4, (CE_WARN, NULL, 7819 "%s%d: vhci_pgr_validate_and_register: key" 7820 " already cleared", ddi_driver_name(cdip), 7821 ddi_get_instance(cdip))); 7822 rval = 1; 7823 } else 7824 rval = 0; 7825 } else { 7826 VHCI_DEBUG(4, (CE_NOTE, NULL, 7827 "%s%d: vhci_pgr_validate_and_register: key" 7828 " validation failed, key backed out\n", 7829 ddi_driver_name(cdip), ddi_get_instance(cdip))); 7830 vhci_print_prout_keys(vlun, "v_pgr_val_reg: key backed out: "); 7831 } 7832 mdi_rele_path(pip); 7833 7834 return (rval); 7835 } 7836 7837 /* 7838 * taskq routine to dispatch a scsi cmd to vhci_scsi_start. This ensures 7839 * that vhci_scsi_start is not called in interrupt context. 7840 * As the upper layer gets TRAN_ACCEPT when the command is dispatched, we 7841 * need to complete the command if something goes wrong. 7842 */ 7843 static void 7844 vhci_dispatch_scsi_start(void *arg) 7845 { 7846 struct vhci_pkt *vpkt = (struct vhci_pkt *)arg; 7847 struct scsi_pkt *tpkt = vpkt->vpkt_tgt_pkt; 7848 int rval = TRAN_BUSY; 7849 7850 VHCI_DEBUG(6, (CE_NOTE, NULL, "!vhci_dispatch_scsi_start: sending" 7851 " scsi-2 reserve for 0x%p\n", 7852 (void *)ADDR2DIP(&(vpkt->vpkt_tgt_pkt->pkt_address)))); 7853 7854 /* 7855 * To prevent the taskq from being called recursively we set the 7856 * the VHCI_PKT_THRU_TASKQ bit in the vhci_pkt_states. 7857 */ 7858 vpkt->vpkt_state |= VHCI_PKT_THRU_TASKQ; 7859 7860 /* 7861 * Wait for the transport to get ready to send packets 7862 * and if it times out, it will return something other than 7863 * TRAN_BUSY. The vhci_reserve_delay may want to 7864 * get tuned for other transports and is therefore a global. 7865 * Using delay since this routine is called by taskq dispatch 7866 * and not called during interrupt context. 7867 */ 7868 while ((rval = vhci_scsi_start(&(vpkt->vpkt_tgt_pkt->pkt_address), 7869 vpkt->vpkt_tgt_pkt)) == TRAN_BUSY) { 7870 delay(drv_usectohz(vhci_reserve_delay)); 7871 } 7872 7873 switch (rval) { 7874 case TRAN_ACCEPT: 7875 return; 7876 7877 default: 7878 /* 7879 * This pkt shall be retried, and to ensure another taskq 7880 * is dispatched for it, clear the VHCI_PKT_THRU_TASKQ 7881 * flag. 7882 */ 7883 vpkt->vpkt_state &= ~VHCI_PKT_THRU_TASKQ; 7884 7885 /* Ensure that the pkt is retried without a reset */ 7886 tpkt->pkt_reason = CMD_ABORTED; 7887 tpkt->pkt_statistics |= STAT_ABORTED; 7888 VHCI_DEBUG(1, (CE_WARN, NULL, "!vhci_dispatch_scsi_start: " 7889 "TRAN_rval %d returned for dip 0x%p", rval, 7890 (void *)ADDR2DIP(&(vpkt->vpkt_tgt_pkt->pkt_address)))); 7891 break; 7892 } 7893 7894 /* 7895 * vpkt_org_vpkt should always be NULL here if the retry command 7896 * has been successfully dispatched. If vpkt_org_vpkt != NULL at 7897 * this point, it is an error so restore the original vpkt and 7898 * return an error to the target driver so it can retry the 7899 * command as appropriate. 7900 */ 7901 if (vpkt->vpkt_org_vpkt != NULL) { 7902 struct vhci_pkt *new_vpkt = vpkt; 7903 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 7904 mdi_pi_get_vhci_private(vpkt->vpkt_path); 7905 7906 vpkt = vpkt->vpkt_org_vpkt; 7907 7908 vpkt->vpkt_tgt_pkt->pkt_reason = tpkt->pkt_reason; 7909 vpkt->vpkt_tgt_pkt->pkt_statistics = tpkt->pkt_statistics; 7910 7911 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 7912 new_vpkt->vpkt_tgt_pkt); 7913 7914 tpkt = vpkt->vpkt_tgt_pkt; 7915 } 7916 7917 scsi_hba_pkt_comp(tpkt); 7918 } 7919 7920 static void 7921 vhci_initiate_auto_failback(void *arg) 7922 { 7923 struct scsi_vhci_lun *vlun = (struct scsi_vhci_lun *)arg; 7924 dev_info_t *vdip, *cdip; 7925 int held; 7926 7927 cdip = vlun->svl_dip; 7928 vdip = ddi_get_parent(cdip); 7929 7930 VHCI_HOLD_LUN(vlun, VH_SLEEP, held); 7931 7932 /* 7933 * Perform a final check to see if the active path class is indeed 7934 * not the preferred path class. As in the time the auto failback 7935 * was dispatched, an external failover could have been detected. 7936 * [Some other host could have detected this condition and triggered 7937 * the auto failback before]. 7938 * In such a case if we go ahead with failover we will be negating the 7939 * whole purpose of auto failback. 7940 */ 7941 mutex_enter(&vlun->svl_mutex); 7942 if (vlun->svl_active_pclass != NULL) { 7943 char *best_pclass; 7944 struct scsi_failover_ops *fo; 7945 7946 fo = vlun->svl_fops; 7947 7948 (void) fo->sfo_pathclass_next(NULL, &best_pclass, 7949 vlun->svl_fops_ctpriv); 7950 if (strcmp(vlun->svl_active_pclass, best_pclass) == 0) { 7951 mutex_exit(&vlun->svl_mutex); 7952 VHCI_RELEASE_LUN(vlun); 7953 VHCI_DEBUG(1, (CE_NOTE, NULL, "Not initiating " 7954 "auto failback for %s as %s pathclass already " 7955 "active.\n", vlun->svl_lun_wwn, best_pclass)); 7956 return; 7957 } 7958 } 7959 mutex_exit(&vlun->svl_mutex); 7960 if (mdi_failover(vdip, vlun->svl_dip, MDI_FAILOVER_SYNC) 7961 == MDI_SUCCESS) { 7962 vhci_log(CE_NOTE, vdip, "!Auto failback operation " 7963 "succeeded for device %s (GUID %s)", 7964 ddi_node_name(cdip), vlun->svl_lun_wwn); 7965 } else { 7966 vhci_log(CE_NOTE, vdip, "!Auto failback operation " 7967 "failed for device %s (GUID %s)", 7968 ddi_node_name(cdip), vlun->svl_lun_wwn); 7969 } 7970 VHCI_RELEASE_LUN(vlun); 7971 } 7972 7973 #ifdef DEBUG 7974 static void 7975 vhci_print_prin_keys(vhci_prin_readkeys_t *prin, int numkeys) 7976 { 7977 vhci_clean_print(NULL, 5, "Current PGR Keys", 7978 (uchar_t *)prin, numkeys * 8); 7979 } 7980 #endif 7981 7982 static void 7983 vhci_print_prout_keys(scsi_vhci_lun_t *vlun, char *msg) 7984 { 7985 int i; 7986 vhci_prout_t *prout; 7987 char buf1[4*MHIOC_RESV_KEY_SIZE + 1]; 7988 char buf2[4*MHIOC_RESV_KEY_SIZE + 1]; 7989 char buf3[4*MHIOC_RESV_KEY_SIZE + 1]; 7990 char buf4[4*MHIOC_RESV_KEY_SIZE + 1]; 7991 7992 prout = &vlun->svl_prout; 7993 7994 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7995 (void) sprintf(&buf1[4*i], "[%02x]", prout->res_key[i]); 7996 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7997 (void) sprintf(&buf2[(4*i)], "[%02x]", prout->service_key[i]); 7998 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 7999 (void) sprintf(&buf3[4*i], "[%02x]", prout->active_res_key[i]); 8000 for (i = 0; i < MHIOC_RESV_KEY_SIZE; i++) 8001 (void) sprintf(&buf4[4*i], "[%02x]", 8002 prout->active_service_key[i]); 8003 8004 /* Printing all in one go. Otherwise it will jumble up */ 8005 VHCI_DEBUG(5, (CE_CONT, NULL, "%s vlun 0x%p, thread 0x%p\n" 8006 "res_key: : %s\n" 8007 "service_key : %s\n" 8008 "active_res_key : %s\n" 8009 "active_service_key: %s\n", 8010 msg, (void *)vlun, (void *)curthread, buf1, buf2, buf3, buf4)); 8011 } 8012 8013 /* 8014 * Called from vhci_scsi_start to update the pHCI pkt with target packet. 8015 */ 8016 static void 8017 vhci_update_pHCI_pkt(struct vhci_pkt *vpkt, struct scsi_pkt *pkt) 8018 { 8019 8020 ASSERT(vpkt->vpkt_hba_pkt); 8021 8022 vpkt->vpkt_hba_pkt->pkt_flags = pkt->pkt_flags; 8023 vpkt->vpkt_hba_pkt->pkt_flags |= FLAG_NOQUEUE; 8024 8025 if ((vpkt->vpkt_hba_pkt->pkt_flags & FLAG_NOINTR) || 8026 MDI_PI_IS_SUSPENDED(vpkt->vpkt_path)) { 8027 /* 8028 * Polled Command is requested or HBA is in 8029 * suspended state 8030 */ 8031 vpkt->vpkt_hba_pkt->pkt_flags |= FLAG_NOINTR; 8032 vpkt->vpkt_hba_pkt->pkt_comp = NULL; 8033 } else { 8034 vpkt->vpkt_hba_pkt->pkt_comp = vhci_intr; 8035 } 8036 vpkt->vpkt_hba_pkt->pkt_time = pkt->pkt_time; 8037 bcopy(pkt->pkt_cdbp, vpkt->vpkt_hba_pkt->pkt_cdbp, 8038 vpkt->vpkt_tgt_init_cdblen); 8039 vpkt->vpkt_hba_pkt->pkt_resid = pkt->pkt_resid; 8040 8041 /* Re-initialize the following pHCI packet state information */ 8042 vpkt->vpkt_hba_pkt->pkt_state = 0; 8043 vpkt->vpkt_hba_pkt->pkt_statistics = 0; 8044 vpkt->vpkt_hba_pkt->pkt_reason = 0; 8045 } 8046 8047 static int 8048 vhci_scsi_bus_power(dev_info_t *parent, void *impl_arg, pm_bus_power_op_t op, 8049 void *arg, void *result) 8050 { 8051 int ret = DDI_SUCCESS; 8052 8053 /* 8054 * Generic processing in MPxIO framework 8055 */ 8056 ret = mdi_bus_power(parent, impl_arg, op, arg, result); 8057 8058 switch (ret) { 8059 case MDI_SUCCESS: 8060 ret = DDI_SUCCESS; 8061 break; 8062 case MDI_FAILURE: 8063 ret = DDI_FAILURE; 8064 break; 8065 default: 8066 break; 8067 } 8068 8069 return (ret); 8070 } 8071 8072 static int 8073 vhci_pHCI_cap(struct scsi_address *ap, char *cap, int val, int whom, 8074 mdi_pathinfo_t *pip) 8075 { 8076 dev_info_t *cdip; 8077 mdi_pathinfo_t *npip = NULL; 8078 scsi_vhci_priv_t *svp = NULL; 8079 struct scsi_address *pap = NULL; 8080 scsi_hba_tran_t *hba = NULL; 8081 int sps; 8082 int mps_flag; 8083 int rval = 0; 8084 8085 mps_flag = (MDI_SELECT_ONLINE_PATH | MDI_SELECT_STANDBY_PATH); 8086 if (pip) { 8087 /* 8088 * If the call is from vhci_pathinfo_state_change, 8089 * then this path was busy and is becoming ready to accept IO. 8090 */ 8091 ASSERT(ap != NULL); 8092 hba = ap->a_hba_tran; 8093 ASSERT(hba != NULL); 8094 rval = scsi_ifsetcap(ap, cap, val, whom); 8095 8096 VHCI_DEBUG(2, (CE_NOTE, NULL, 8097 "!vhci_pHCI_cap: only on path %p, ap %p, rval %x\n", 8098 (void *)pip, (void *)ap, rval)); 8099 8100 return (rval); 8101 } 8102 8103 /* 8104 * Set capability on all the pHCIs. 8105 * If any path is busy, then the capability would be set by 8106 * vhci_pathinfo_state_change. 8107 */ 8108 8109 cdip = ADDR2DIP(ap); 8110 ASSERT(cdip != NULL); 8111 sps = mdi_select_path(cdip, NULL, mps_flag, NULL, &pip); 8112 if ((sps != MDI_SUCCESS) || (pip == NULL)) { 8113 VHCI_DEBUG(2, (CE_WARN, NULL, 8114 "!vhci_pHCI_cap: Unable to get a path, dip 0x%p", 8115 (void *)cdip)); 8116 return (0); 8117 } 8118 8119 again: 8120 svp = (scsi_vhci_priv_t *)mdi_pi_get_vhci_private(pip); 8121 if (svp == NULL) { 8122 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_pHCI_cap: " 8123 "priv is NULL, pip 0x%p", (void *)pip)); 8124 mdi_rele_path(pip); 8125 return (rval); 8126 } 8127 8128 if (svp->svp_psd == NULL) { 8129 VHCI_DEBUG(2, (CE_WARN, NULL, "!vhci_pHCI_cap: " 8130 "psd is NULL, pip 0x%p, svp 0x%p", 8131 (void *)pip, (void *)svp)); 8132 mdi_rele_path(pip); 8133 return (rval); 8134 } 8135 8136 pap = &svp->svp_psd->sd_address; 8137 ASSERT(pap != NULL); 8138 hba = pap->a_hba_tran; 8139 ASSERT(hba != NULL); 8140 8141 if (hba->tran_setcap != NULL) { 8142 rval = scsi_ifsetcap(pap, cap, val, whom); 8143 8144 VHCI_DEBUG(2, (CE_NOTE, NULL, 8145 "!vhci_pHCI_cap: path %p, ap %p, rval %x\n", 8146 (void *)pip, (void *)ap, rval)); 8147 8148 /* 8149 * Select next path and issue the setcap, repeat 8150 * until all paths are exhausted 8151 */ 8152 sps = mdi_select_path(cdip, NULL, mps_flag, pip, &npip); 8153 if ((sps != MDI_SUCCESS) || (npip == NULL)) { 8154 mdi_rele_path(pip); 8155 return (1); 8156 } 8157 mdi_rele_path(pip); 8158 pip = npip; 8159 goto again; 8160 } 8161 mdi_rele_path(pip); 8162 return (rval); 8163 } 8164 8165 static int 8166 vhci_scsi_bus_config(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op, 8167 void *arg, dev_info_t **child) 8168 { 8169 char *guid; 8170 8171 if (vhci_bus_config_debug) 8172 flags |= NDI_DEVI_DEBUG; 8173 8174 if (op == BUS_CONFIG_ONE || op == BUS_UNCONFIG_ONE) 8175 guid = vhci_devnm_to_guid((char *)arg); 8176 else 8177 guid = NULL; 8178 8179 if (mdi_vhci_bus_config(pdip, flags, op, arg, child, guid) 8180 == MDI_SUCCESS) 8181 return (NDI_SUCCESS); 8182 else 8183 return (NDI_FAILURE); 8184 } 8185 8186 static int 8187 vhci_scsi_bus_unconfig(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op, 8188 void *arg) 8189 { 8190 if (vhci_bus_config_debug) 8191 flags |= NDI_DEVI_DEBUG; 8192 8193 return (ndi_busop_bus_unconfig(pdip, flags, op, arg)); 8194 } 8195 8196 /* 8197 * Take the original vhci_pkt, create a duplicate of the pkt for resending 8198 * as though it originated in ssd. 8199 */ 8200 static struct scsi_pkt * 8201 vhci_create_retry_pkt(struct vhci_pkt *vpkt) 8202 { 8203 struct vhci_pkt *new_vpkt = NULL; 8204 struct scsi_pkt *pkt = NULL; 8205 8206 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 8207 mdi_pi_get_vhci_private(vpkt->vpkt_path); 8208 8209 /* 8210 * Ensure consistent data at completion time by setting PKT_CONSISTENT 8211 */ 8212 pkt = vhci_scsi_init_pkt(&svp->svp_psd->sd_address, pkt, 8213 vpkt->vpkt_tgt_init_bp, vpkt->vpkt_tgt_init_cdblen, 8214 vpkt->vpkt_tgt_init_scblen, 0, PKT_CONSISTENT, NULL_FUNC, NULL); 8215 if (pkt != NULL) { 8216 new_vpkt = TGTPKT2VHCIPKT(pkt); 8217 8218 pkt->pkt_address = vpkt->vpkt_tgt_pkt->pkt_address; 8219 pkt->pkt_flags = vpkt->vpkt_tgt_pkt->pkt_flags; 8220 pkt->pkt_time = vpkt->vpkt_tgt_pkt->pkt_time; 8221 pkt->pkt_comp = vpkt->vpkt_tgt_pkt->pkt_comp; 8222 8223 pkt->pkt_resid = 0; 8224 pkt->pkt_statistics = 0; 8225 pkt->pkt_reason = 0; 8226 8227 bcopy(vpkt->vpkt_tgt_pkt->pkt_cdbp, 8228 pkt->pkt_cdbp, vpkt->vpkt_tgt_init_cdblen); 8229 8230 /* 8231 * Save a pointer to the original vhci_pkt 8232 */ 8233 new_vpkt->vpkt_org_vpkt = vpkt; 8234 } 8235 8236 return (pkt); 8237 } 8238 8239 /* 8240 * Copy the successful completion information from the hba packet into 8241 * the original target pkt from the upper layer. Returns the original 8242 * vpkt and destroys the new vpkt from the internal retry. 8243 */ 8244 static struct vhci_pkt * 8245 vhci_sync_retry_pkt(struct vhci_pkt *vpkt) 8246 { 8247 struct vhci_pkt *ret_vpkt = NULL; 8248 struct scsi_pkt *tpkt = NULL; 8249 struct scsi_pkt *hba_pkt = NULL; 8250 scsi_vhci_priv_t *svp = (scsi_vhci_priv_t *) 8251 mdi_pi_get_vhci_private(vpkt->vpkt_path); 8252 8253 ASSERT(vpkt->vpkt_org_vpkt != NULL); 8254 VHCI_DEBUG(0, (CE_NOTE, NULL, "vhci_sync_retry_pkt: Retry pkt " 8255 "completed successfully!\n")); 8256 8257 ret_vpkt = vpkt->vpkt_org_vpkt; 8258 tpkt = ret_vpkt->vpkt_tgt_pkt; 8259 hba_pkt = vpkt->vpkt_hba_pkt; 8260 8261 /* 8262 * Copy the good status into the target driver's packet 8263 */ 8264 *(tpkt->pkt_scbp) = *(hba_pkt->pkt_scbp); 8265 tpkt->pkt_resid = hba_pkt->pkt_resid; 8266 tpkt->pkt_state = hba_pkt->pkt_state; 8267 tpkt->pkt_statistics = hba_pkt->pkt_statistics; 8268 tpkt->pkt_reason = hba_pkt->pkt_reason; 8269 8270 /* 8271 * Destroy the internally created vpkt for the retry 8272 */ 8273 vhci_scsi_destroy_pkt(&svp->svp_psd->sd_address, 8274 vpkt->vpkt_tgt_pkt); 8275 8276 return (ret_vpkt); 8277 } 8278 8279 /* restart the request sense request */ 8280 static void 8281 vhci_uscsi_restart_sense(void *arg) 8282 { 8283 struct buf *rqbp; 8284 struct buf *bp; 8285 struct scsi_pkt *rqpkt = (struct scsi_pkt *)arg; 8286 mp_uscsi_cmd_t *mp_uscmdp; 8287 8288 VHCI_DEBUG(4, (CE_WARN, NULL, 8289 "vhci_uscsi_restart_sense: enter: rqpkt: %p", (void *)rqpkt)); 8290 8291 if (scsi_transport(rqpkt) != TRAN_ACCEPT) { 8292 /* if it fails - need to wakeup the original command */ 8293 mp_uscmdp = rqpkt->pkt_private; 8294 bp = mp_uscmdp->cmdbp; 8295 rqbp = mp_uscmdp->rqbp; 8296 ASSERT(mp_uscmdp && bp && rqbp); 8297 scsi_free_consistent_buf(rqbp); 8298 scsi_destroy_pkt(rqpkt); 8299 bp->b_resid = bp->b_bcount; 8300 bioerror(bp, EIO); 8301 biodone(bp); 8302 } 8303 } 8304 8305 /* 8306 * auto-rqsense is not enabled so we have to retrieve the request sense 8307 * manually. 8308 */ 8309 static int 8310 vhci_uscsi_send_sense(struct scsi_pkt *pkt, mp_uscsi_cmd_t *mp_uscmdp) 8311 { 8312 struct buf *rqbp, *cmdbp; 8313 struct scsi_pkt *rqpkt; 8314 int rval = 0; 8315 8316 cmdbp = mp_uscmdp->cmdbp; 8317 ASSERT(cmdbp != NULL); 8318 8319 VHCI_DEBUG(4, (CE_WARN, NULL, 8320 "vhci_uscsi_send_sense: enter: bp: %p pkt: %p scmd: %p", 8321 (void *)cmdbp, (void *)pkt, (void *)mp_uscmdp)); 8322 /* set up the packet information and cdb */ 8323 if ((rqbp = scsi_alloc_consistent_buf(mp_uscmdp->ap, NULL, 8324 SENSE_LENGTH, B_READ, NULL, NULL)) == NULL) { 8325 return (-1); 8326 } 8327 8328 if ((rqpkt = scsi_init_pkt(mp_uscmdp->ap, NULL, rqbp, 8329 CDB_GROUP0, 1, 0, PKT_CONSISTENT, NULL, NULL)) == NULL) { 8330 scsi_free_consistent_buf(rqbp); 8331 return (-1); 8332 } 8333 8334 (void) scsi_setup_cdb((union scsi_cdb *)(intptr_t)rqpkt->pkt_cdbp, 8335 SCMD_REQUEST_SENSE, 0, SENSE_LENGTH, 0); 8336 8337 mp_uscmdp->rqbp = rqbp; 8338 rqbp->b_private = mp_uscmdp; 8339 rqpkt->pkt_flags |= FLAG_SENSING; 8340 rqpkt->pkt_time = 60; 8341 rqpkt->pkt_comp = vhci_uscsi_iodone; 8342 rqpkt->pkt_private = mp_uscmdp; 8343 8344 /* 8345 * NOTE: This code path is related to MPAPI uscsi(7I), so path 8346 * selection is not based on path_instance. 8347 */ 8348 if (scsi_pkt_allocated_correctly(rqpkt)) 8349 rqpkt->pkt_path_instance = 0; 8350 8351 /* get her done */ 8352 switch (scsi_transport(rqpkt)) { 8353 case TRAN_ACCEPT: 8354 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_uscsi_send_sense: " 8355 "transport accepted.")); 8356 break; 8357 case TRAN_BUSY: 8358 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_uscsi_send_sense: " 8359 "transport busy, setting timeout.")); 8360 vhci_restart_timeid = timeout(vhci_uscsi_restart_sense, rqpkt, 8361 (drv_usectohz(5 * 1000000))); 8362 break; 8363 default: 8364 VHCI_DEBUG(1, (CE_NOTE, NULL, "vhci_uscsi_send_sense: " 8365 "transport failed")); 8366 scsi_free_consistent_buf(rqbp); 8367 scsi_destroy_pkt(rqpkt); 8368 rval = -1; 8369 } 8370 8371 return (rval); 8372 } 8373 8374 /* 8375 * done routine for the mpapi uscsi command - this is behaving as though 8376 * FLAG_DIAGNOSE is set meaning there are no retries except for a manual 8377 * request sense. 8378 */ 8379 void 8380 vhci_uscsi_iodone(struct scsi_pkt *pkt) 8381 { 8382 struct buf *bp; 8383 mp_uscsi_cmd_t *mp_uscmdp; 8384 struct uscsi_cmd *uscmdp; 8385 struct scsi_arq_status *arqstat; 8386 int err; 8387 8388 mp_uscmdp = (mp_uscsi_cmd_t *)pkt->pkt_private; 8389 uscmdp = mp_uscmdp->uscmdp; 8390 bp = mp_uscmdp->cmdbp; 8391 ASSERT(bp != NULL); 8392 VHCI_DEBUG(4, (CE_WARN, NULL, 8393 "vhci_uscsi_iodone: enter: bp: %p pkt: %p scmd: %p", 8394 (void *)bp, (void *)pkt, (void *)mp_uscmdp)); 8395 /* Save the status and the residual into the uscsi_cmd struct */ 8396 uscmdp->uscsi_status = ((*(pkt)->pkt_scbp) & STATUS_MASK); 8397 uscmdp->uscsi_resid = bp->b_resid; 8398 8399 /* return on a very successful command */ 8400 if (pkt->pkt_reason == CMD_CMPLT && 8401 SCBP_C(pkt) == 0 && ((pkt->pkt_flags & FLAG_SENSING) == 0) && 8402 pkt->pkt_resid == 0) { 8403 mdi_pi_kstat_iosupdate(mp_uscmdp->pip, bp); 8404 scsi_destroy_pkt(pkt); 8405 biodone(bp); 8406 return; 8407 } 8408 VHCI_DEBUG(4, (CE_NOTE, NULL, "iodone: reason=0x%x " 8409 " pkt_resid=%ld pkt_state: 0x%x b_count: %ld b_resid: %ld", 8410 pkt->pkt_reason, pkt->pkt_resid, 8411 pkt->pkt_state, bp->b_bcount, bp->b_resid)); 8412 8413 err = EIO; 8414 8415 arqstat = (struct scsi_arq_status *)(intptr_t)(pkt->pkt_scbp); 8416 if (pkt->pkt_reason != CMD_CMPLT) { 8417 /* 8418 * The command did not complete. 8419 */ 8420 VHCI_DEBUG(4, (CE_NOTE, NULL, 8421 "vhci_uscsi_iodone: command did not complete." 8422 " reason: %x flag: %x", pkt->pkt_reason, pkt->pkt_flags)); 8423 if (pkt->pkt_flags & FLAG_SENSING) { 8424 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_TRANSERR); 8425 } else if (pkt->pkt_reason == CMD_TIMEOUT) { 8426 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_HARDERR); 8427 err = ETIMEDOUT; 8428 } 8429 } else if (pkt->pkt_state & STATE_ARQ_DONE && mp_uscmdp->arq_enabled) { 8430 /* 8431 * The auto-rqsense happened, and the packet has a filled-in 8432 * scsi_arq_status structure, pointed to by pkt_scbp. 8433 */ 8434 VHCI_DEBUG(4, (CE_NOTE, NULL, 8435 "vhci_uscsi_iodone: received auto-requested sense")); 8436 if (uscmdp->uscsi_flags & USCSI_RQENABLE) { 8437 /* get the amount of data to copy into rqbuf */ 8438 int rqlen = SENSE_LENGTH - arqstat->sts_rqpkt_resid; 8439 rqlen = min(((int)uscmdp->uscsi_rqlen), rqlen); 8440 uscmdp->uscsi_rqresid = uscmdp->uscsi_rqlen - rqlen; 8441 uscmdp->uscsi_rqstatus = 8442 *((char *)&arqstat->sts_rqpkt_status); 8443 if (uscmdp->uscsi_rqbuf && uscmdp->uscsi_rqlen && 8444 rqlen != 0) { 8445 bcopy(&(arqstat->sts_sensedata), 8446 uscmdp->uscsi_rqbuf, rqlen); 8447 } 8448 mdi_pi_kstat_iosupdate(mp_uscmdp->pip, bp); 8449 VHCI_DEBUG(4, (CE_NOTE, NULL, 8450 "vhci_uscsi_iodone: ARQ " 8451 "uscsi_rqstatus=0x%x uscsi_rqresid=%d rqlen: %d " 8452 "xfer: %d rqpkt_resid: %d\n", 8453 uscmdp->uscsi_rqstatus, uscmdp->uscsi_rqresid, 8454 uscmdp->uscsi_rqlen, rqlen, 8455 arqstat->sts_rqpkt_resid)); 8456 } 8457 } else if (pkt->pkt_flags & FLAG_SENSING) { 8458 struct buf *rqbp; 8459 struct scsi_status *rqstatus; 8460 8461 rqstatus = (struct scsi_status *)pkt->pkt_scbp; 8462 /* a manual request sense was done - get the information */ 8463 if (uscmdp->uscsi_flags & USCSI_RQENABLE) { 8464 int rqlen = SENSE_LENGTH - pkt->pkt_resid; 8465 8466 rqbp = mp_uscmdp->rqbp; 8467 /* get the amount of data to copy into rqbuf */ 8468 rqlen = min(((int)uscmdp->uscsi_rqlen), rqlen); 8469 uscmdp->uscsi_rqresid = uscmdp->uscsi_rqlen - rqlen; 8470 uscmdp->uscsi_rqstatus = *((char *)rqstatus); 8471 if (uscmdp->uscsi_rqlen && uscmdp->uscsi_rqbuf) { 8472 bcopy(rqbp->b_un.b_addr, uscmdp->uscsi_rqbuf, 8473 rqlen); 8474 } 8475 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_TRANSERR); 8476 scsi_free_consistent_buf(rqbp); 8477 } 8478 VHCI_DEBUG(4, (CE_NOTE, NULL, "vhci_uscsi_iodone: FLAG_SENSING" 8479 "uscsi_rqstatus=0x%x uscsi_rqresid=%d\n", 8480 uscmdp->uscsi_rqstatus, uscmdp->uscsi_rqresid)); 8481 } else { 8482 struct scsi_status *status = 8483 (struct scsi_status *)pkt->pkt_scbp; 8484 /* 8485 * Command completed and we're not getting sense. Check for 8486 * errors and decide what to do next. 8487 */ 8488 VHCI_DEBUG(4, (CE_NOTE, NULL, 8489 "vhci_uscsi_iodone: command appears complete: reason: %x", 8490 pkt->pkt_reason)); 8491 if (status->sts_chk) { 8492 /* need to manually get the request sense */ 8493 if (vhci_uscsi_send_sense(pkt, mp_uscmdp) == 0) { 8494 scsi_destroy_pkt(pkt); 8495 return; 8496 } 8497 } else { 8498 VHCI_DEBUG(4, (CE_NOTE, NULL, 8499 "vhci_chk_err: appears complete")); 8500 err = 0; 8501 mdi_pi_kstat_iosupdate(mp_uscmdp->pip, bp); 8502 if (pkt->pkt_resid) { 8503 bp->b_resid += pkt->pkt_resid; 8504 } 8505 } 8506 } 8507 8508 if (err) { 8509 if (bp->b_resid == 0) 8510 bp->b_resid = bp->b_bcount; 8511 bioerror(bp, err); 8512 bp->b_flags |= B_ERROR; 8513 } 8514 8515 scsi_destroy_pkt(pkt); 8516 biodone(bp); 8517 8518 VHCI_DEBUG(4, (CE_WARN, NULL, "vhci_uscsi_iodone: exit")); 8519 } 8520 8521 /* 8522 * start routine for the mpapi uscsi command 8523 */ 8524 int 8525 vhci_uscsi_iostart(struct buf *bp) 8526 { 8527 struct scsi_pkt *pkt; 8528 struct uscsi_cmd *uscmdp; 8529 mp_uscsi_cmd_t *mp_uscmdp; 8530 int stat_size, rval; 8531 int retry = 0; 8532 8533 ASSERT(bp->b_private != NULL); 8534 8535 mp_uscmdp = (mp_uscsi_cmd_t *)bp->b_private; 8536 uscmdp = mp_uscmdp->uscmdp; 8537 if (uscmdp->uscsi_flags & USCSI_RQENABLE) { 8538 stat_size = SENSE_LENGTH; 8539 } else { 8540 stat_size = 1; 8541 } 8542 8543 pkt = scsi_init_pkt(mp_uscmdp->ap, NULL, bp, uscmdp->uscsi_cdblen, 8544 stat_size, 0, 0, SLEEP_FUNC, NULL); 8545 if (pkt == NULL) { 8546 VHCI_DEBUG(4, (CE_NOTE, NULL, 8547 "vhci_uscsi_iostart: rval: EINVAL")); 8548 bp->b_resid = bp->b_bcount; 8549 uscmdp->uscsi_resid = bp->b_bcount; 8550 bioerror(bp, EINVAL); 8551 biodone(bp); 8552 return (EINVAL); 8553 } 8554 8555 pkt->pkt_time = uscmdp->uscsi_timeout; 8556 bcopy(uscmdp->uscsi_cdb, pkt->pkt_cdbp, (size_t)uscmdp->uscsi_cdblen); 8557 pkt->pkt_comp = vhci_uscsi_iodone; 8558 pkt->pkt_private = mp_uscmdp; 8559 if (uscmdp->uscsi_flags & USCSI_SILENT) 8560 pkt->pkt_flags |= FLAG_SILENT; 8561 if (uscmdp->uscsi_flags & USCSI_ISOLATE) 8562 pkt->pkt_flags |= FLAG_ISOLATE; 8563 if (uscmdp->uscsi_flags & USCSI_DIAGNOSE) 8564 pkt->pkt_flags |= FLAG_DIAGNOSE; 8565 if (uscmdp->uscsi_flags & USCSI_RENEGOT) { 8566 pkt->pkt_flags |= FLAG_RENEGOTIATE_WIDE_SYNC; 8567 } 8568 VHCI_DEBUG(4, (CE_WARN, NULL, 8569 "vhci_uscsi_iostart: ap: %p pkt: %p pcdbp: %p uscmdp: %p" 8570 " ucdbp: %p pcdblen: %d bp: %p count: %ld pip: %p" 8571 " stat_size: %d", 8572 (void *)mp_uscmdp->ap, (void *)pkt, (void *)pkt->pkt_cdbp, 8573 (void *)uscmdp, (void *)uscmdp->uscsi_cdb, pkt->pkt_cdblen, 8574 (void *)bp, bp->b_bcount, (void *)mp_uscmdp->pip, stat_size)); 8575 8576 /* 8577 * NOTE: This code path is related to MPAPI uscsi(7I), so path 8578 * selection is not based on path_instance. 8579 */ 8580 if (scsi_pkt_allocated_correctly(pkt)) 8581 pkt->pkt_path_instance = 0; 8582 8583 while (((rval = scsi_transport(pkt)) == TRAN_BUSY) && 8584 retry < vhci_uscsi_retry_count) { 8585 delay(drv_usectohz(vhci_uscsi_delay)); 8586 retry++; 8587 } 8588 if (retry >= vhci_uscsi_retry_count) { 8589 VHCI_DEBUG(4, (CE_NOTE, NULL, 8590 "vhci_uscsi_iostart: tran_busy - retry: %d", retry)); 8591 } 8592 switch (rval) { 8593 case TRAN_ACCEPT: 8594 rval = 0; 8595 break; 8596 8597 default: 8598 VHCI_DEBUG(4, (CE_NOTE, NULL, 8599 "vhci_uscsi_iostart: rval: %d count: %ld res: %ld", 8600 rval, bp->b_bcount, bp->b_resid)); 8601 bp->b_resid = bp->b_bcount; 8602 uscmdp->uscsi_resid = bp->b_bcount; 8603 bioerror(bp, EIO); 8604 scsi_destroy_pkt(pkt); 8605 biodone(bp); 8606 rval = EIO; 8607 MDI_PI_ERRSTAT(mp_uscmdp->pip, MDI_PI_TRANSERR); 8608 break; 8609 } 8610 VHCI_DEBUG(4, (CE_NOTE, NULL, 8611 "vhci_uscsi_iostart: exit: rval: %d", rval)); 8612 return (rval); 8613 } 8614 8615 /* ARGSUSED */ 8616 static struct scsi_failover_ops * 8617 vhci_dev_fo(dev_info_t *vdip, struct scsi_device *psd, 8618 void **ctprivp, char **fo_namep) 8619 { 8620 struct scsi_failover_ops *sfo; 8621 char *sfo_name; 8622 char *override; 8623 struct scsi_failover *sf; 8624 8625 ASSERT(psd && psd->sd_inq); 8626 if ((psd == NULL) || (psd->sd_inq == NULL)) { 8627 VHCI_DEBUG(1, (CE_NOTE, NULL, 8628 "!vhci_dev_fo:return NULL no scsi_device or inquiry")); 8629 return (NULL); 8630 } 8631 8632 /* 8633 * Determine if device is supported under scsi_vhci, and select 8634 * failover module. 8635 * 8636 * See if there is a scsi_vhci.conf file override for this devices's 8637 * VID/PID. The following values can be returned: 8638 * 8639 * NULL If the NULL is returned then there is no scsi_vhci.conf 8640 * override. For NULL, we determine the failover_ops for 8641 * this device by checking the sfo_device_probe entry 8642 * point for each 'fops' module, in order. 8643 * 8644 * NOTE: Correct operation may depend on module ordering 8645 * of 'specific' (failover modules that are completely 8646 * VID/PID table based) to 'generic' (failover modules 8647 * that based on T10 standards like TPGS). Currently, 8648 * the value of 'ddi-forceload' in scsi_vhci.conf is used 8649 * to establish the module list and probe order. 8650 * 8651 * "NONE" If value "NONE" is returned then there is a 8652 * scsi_vhci.conf VID/PID override to indicate the device 8653 * should not be supported under scsi_vhci (even if there 8654 * is an 'fops' module supporting the device). 8655 * 8656 * "<other>" If another value is returned then that value is the 8657 * name of the 'fops' module that should be used. 8658 */ 8659 sfo = NULL; /* "NONE" */ 8660 override = scsi_get_device_type_string( 8661 "scsi-vhci-failover-override", vdip, psd); 8662 if (override == NULL) { 8663 /* NULL: default: select based on sfo_device_probe results */ 8664 for (sf = scsi_failover_table; sf->sf_mod; sf++) { 8665 if ((sf->sf_sfo == NULL) || 8666 sf->sf_sfo->sfo_device_probe(psd, psd->sd_inq, 8667 ctprivp) == SFO_DEVICE_PROBE_PHCI) 8668 continue; 8669 8670 /* found failover module, supported under scsi_vhci */ 8671 sfo = sf->sf_sfo; 8672 if (fo_namep && (*fo_namep == NULL)) { 8673 sfo_name = i_ddi_strdup(sfo->sfo_name, 8674 KM_SLEEP); 8675 *fo_namep = sfo_name; 8676 } 8677 break; 8678 } 8679 } else if (strcasecmp(override, "NONE")) { 8680 /* !"NONE": select based on driver.conf specified name */ 8681 for (sf = scsi_failover_table, sfo = NULL; sf->sf_mod; sf++) { 8682 if ((sf->sf_sfo == NULL) || 8683 (sf->sf_sfo->sfo_name == NULL) || 8684 strcmp(override, sf->sf_sfo->sfo_name)) 8685 continue; 8686 8687 /* 8688 * NOTE: If sfo_device_probe() has side-effects, 8689 * including setting *ctprivp, these are not going 8690 * to occur with override config. 8691 */ 8692 8693 /* found failover module, supported under scsi_vhci */ 8694 sfo = sf->sf_sfo; 8695 if (fo_namep && (*fo_namep == NULL)) { 8696 sfo_name = kmem_alloc(strlen("conf ") + 8697 strlen(sfo->sfo_name) + 1, KM_SLEEP); 8698 (void) sprintf(sfo_name, "conf %s", 8699 sfo->sfo_name); 8700 *fo_namep = sfo_name; 8701 } 8702 break; 8703 } 8704 } 8705 if (override) 8706 kmem_free(override, strlen(override) + 1); 8707 return (sfo); 8708 } 8709 8710 /* 8711 * Determine the device described by cinfo should be enumerated under 8712 * the vHCI or the pHCI - if there is a failover ops then device is 8713 * supported under vHCI. By agreement with SCSA cinfo is a pointer 8714 * to a scsi_device structure associated with a decorated pHCI probe node. 8715 */ 8716 /* ARGSUSED */ 8717 int 8718 vhci_is_dev_supported(dev_info_t *vdip, dev_info_t *pdip, void *cinfo) 8719 { 8720 struct scsi_device *psd = (struct scsi_device *)cinfo; 8721 8722 return (vhci_dev_fo(vdip, psd, NULL, NULL) ? MDI_SUCCESS : MDI_FAILURE); 8723 } 8724 8725 8726 #ifdef DEBUG 8727 extern struct scsi_key_strings scsi_cmds[]; 8728 8729 static char * 8730 vhci_print_scsi_cmd(char cmd) 8731 { 8732 char tmp[64]; 8733 char *cpnt; 8734 8735 cpnt = scsi_cmd_name(cmd, scsi_cmds, tmp); 8736 /* tmp goes out of scope on return and caller sees garbage */ 8737 if (cpnt == tmp) { 8738 cpnt = "Unknown Command"; 8739 } 8740 return (cpnt); 8741 } 8742 8743 extern uchar_t scsi_cdb_size[]; 8744 8745 static void 8746 vhci_print_cdb(dev_info_t *dip, uint_t level, char *title, uchar_t *cdb) 8747 { 8748 int len = scsi_cdb_size[CDB_GROUPID(cdb[0])]; 8749 char buf[256]; 8750 8751 if (level == CE_NOTE) { 8752 vhci_log(level, dip, "path cmd %s\n", 8753 vhci_print_scsi_cmd(*cdb)); 8754 return; 8755 } 8756 8757 (void) sprintf(buf, "%s for cmd(%s)", title, vhci_print_scsi_cmd(*cdb)); 8758 vhci_clean_print(dip, level, buf, cdb, len); 8759 } 8760 8761 static void 8762 vhci_clean_print(dev_info_t *dev, uint_t level, char *title, uchar_t *data, 8763 int len) 8764 { 8765 int i; 8766 int c; 8767 char *format; 8768 char buf[256]; 8769 uchar_t byte; 8770 8771 (void) sprintf(buf, "%s:\n", title); 8772 vhci_log(level, dev, "%s", buf); 8773 level = CE_CONT; 8774 for (i = 0; i < len; ) { 8775 buf[0] = 0; 8776 for (c = 0; c < 8 && i < len; c++, i++) { 8777 byte = (uchar_t)data[i]; 8778 if (byte < 0x10) 8779 format = "0x0%x "; 8780 else 8781 format = "0x%x "; 8782 (void) sprintf(&buf[(int)strlen(buf)], format, byte); 8783 } 8784 (void) sprintf(&buf[(int)strlen(buf)], "\n"); 8785 8786 vhci_log(level, dev, "%s\n", buf); 8787 } 8788 } 8789 #endif 8790 static void 8791 vhci_invalidate_mpapi_lu(struct scsi_vhci *vhci, scsi_vhci_lun_t *vlun) 8792 { 8793 char *svl_wwn; 8794 mpapi_item_list_t *ilist; 8795 mpapi_lu_data_t *ld; 8796 8797 if (vlun == NULL) { 8798 return; 8799 } else { 8800 svl_wwn = vlun->svl_lun_wwn; 8801 } 8802 8803 ilist = vhci->mp_priv->obj_hdr_list[MP_OBJECT_TYPE_MULTIPATH_LU]->head; 8804 8805 while (ilist != NULL) { 8806 ld = (mpapi_lu_data_t *)(ilist->item->idata); 8807 if ((ld != NULL) && (strncmp(ld->prop.name, svl_wwn, 8808 strlen(svl_wwn)) == 0)) { 8809 ld->valid = 0; 8810 VHCI_DEBUG(6, (CE_WARN, NULL, 8811 "vhci_invalidate_mpapi_lu: " 8812 "Invalidated LU(%s)", svl_wwn)); 8813 return; 8814 } 8815 ilist = ilist->next; 8816 } 8817 VHCI_DEBUG(6, (CE_WARN, NULL, "vhci_invalidate_mpapi_lu: " 8818 "Could not find LU(%s) to invalidate.", svl_wwn)); 8819 }