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