1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Generic SCSI-3 ALUA SCSI Device Handler 4 * 5 * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH. 6 * All rights reserved. 7 */ 8 #include <linux/slab.h> 9 #include <linux/delay.h> 10 #include <linux/module.h> 11 #include <asm/unaligned.h> 12 #include <scsi/scsi.h> 13 #include <scsi/scsi_proto.h> 14 #include <scsi/scsi_dbg.h> 15 #include <scsi/scsi_eh.h> 16 #include <scsi/scsi_dh.h> 17 18 #define ALUA_DH_NAME "alua" 19 #define ALUA_DH_VER "2.0" 20 21 #define TPGS_SUPPORT_NONE 0x00 22 #define TPGS_SUPPORT_OPTIMIZED 0x01 23 #define TPGS_SUPPORT_NONOPTIMIZED 0x02 24 #define TPGS_SUPPORT_STANDBY 0x04 25 #define TPGS_SUPPORT_UNAVAILABLE 0x08 26 #define TPGS_SUPPORT_LBA_DEPENDENT 0x10 27 #define TPGS_SUPPORT_OFFLINE 0x40 28 #define TPGS_SUPPORT_TRANSITION 0x80 29 #define TPGS_SUPPORT_ALL 0xdf 30 31 #define RTPG_FMT_MASK 0x70 32 #define RTPG_FMT_EXT_HDR 0x10 33 34 #define TPGS_MODE_UNINITIALIZED -1 35 #define TPGS_MODE_NONE 0x0 36 #define TPGS_MODE_IMPLICIT 0x1 37 #define TPGS_MODE_EXPLICIT 0x2 38 39 #define ALUA_RTPG_SIZE 128 40 #define ALUA_FAILOVER_TIMEOUT 60 41 #define ALUA_FAILOVER_RETRIES 5 42 #define ALUA_RTPG_DELAY_MSECS 5 43 #define ALUA_RTPG_RETRY_DELAY 2 44 45 /* device handler flags */ 46 #define ALUA_OPTIMIZE_STPG 0x01 47 #define ALUA_RTPG_EXT_HDR_UNSUPP 0x02 48 /* State machine flags */ 49 #define ALUA_PG_RUN_RTPG 0x10 50 #define ALUA_PG_RUN_STPG 0x20 51 #define ALUA_PG_RUNNING 0x40 52 53 static uint optimize_stpg; 54 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR); 55 MODULE_PARM_DESC(optimize_stpg, "Allow use of a non-optimized path, rather than sending a STPG, when implicit TPGS is supported (0=No,1=Yes). Default is 0."); 56 57 static LIST_HEAD(port_group_list); 58 static DEFINE_SPINLOCK(port_group_lock); 59 static struct workqueue_struct *kaluad_wq; 60 61 struct alua_port_group { 62 struct kref kref; 63 struct rcu_head rcu; 64 struct list_head node; 65 struct list_head dh_list; 66 unsigned char device_id_str[256]; 67 int device_id_len; 68 int group_id; 69 int tpgs; 70 int state; 71 int pref; 72 int valid_states; 73 unsigned flags; /* used for optimizing STPG */ 74 unsigned char transition_tmo; 75 unsigned long expiry; 76 unsigned long interval; 77 struct delayed_work rtpg_work; 78 spinlock_t lock; 79 struct list_head rtpg_list; 80 struct scsi_device *rtpg_sdev; 81 }; 82 83 struct alua_dh_data { 84 struct list_head node; 85 struct alua_port_group __rcu *pg; 86 int group_id; 87 spinlock_t pg_lock; 88 struct scsi_device *sdev; 89 int init_error; 90 struct mutex init_mutex; 91 }; 92 93 struct alua_queue_data { 94 struct list_head entry; 95 activate_complete callback_fn; 96 void *callback_data; 97 }; 98 99 #define ALUA_POLICY_SWITCH_CURRENT 0 100 #define ALUA_POLICY_SWITCH_ALL 1 101 102 static void alua_rtpg_work(struct work_struct *work); 103 static bool alua_rtpg_queue(struct alua_port_group *pg, 104 struct scsi_device *sdev, 105 struct alua_queue_data *qdata, bool force); 106 static void alua_check(struct scsi_device *sdev, bool force); 107 108 static void release_port_group(struct kref *kref) 109 { 110 struct alua_port_group *pg; 111 112 pg = container_of(kref, struct alua_port_group, kref); 113 if (pg->rtpg_sdev) 114 flush_delayed_work(&pg->rtpg_work); 115 spin_lock(&port_group_lock); 116 list_del(&pg->node); 117 spin_unlock(&port_group_lock); 118 kfree_rcu(pg, rcu); 119 } 120 121 /* 122 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command 123 * @sdev: sdev the command should be sent to 124 */ 125 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff, 126 int bufflen, struct scsi_sense_hdr *sshdr, int flags) 127 { 128 u8 cdb[MAX_COMMAND_SIZE]; 129 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | 130 REQ_FAILFAST_DRIVER; 131 132 /* Prepare the command. */ 133 memset(cdb, 0x0, MAX_COMMAND_SIZE); 134 cdb[0] = MAINTENANCE_IN; 135 if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP)) 136 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT; 137 else 138 cdb[1] = MI_REPORT_TARGET_PGS; 139 put_unaligned_be32(bufflen, &cdb[6]); 140 141 return scsi_execute(sdev, cdb, DMA_FROM_DEVICE, buff, bufflen, NULL, 142 sshdr, ALUA_FAILOVER_TIMEOUT * HZ, 143 ALUA_FAILOVER_RETRIES, req_flags, 0, NULL); 144 } 145 146 /* 147 * submit_stpg - Issue a SET TARGET PORT GROUP command 148 * 149 * Currently we're only setting the current target port group state 150 * to 'active/optimized' and let the array firmware figure out 151 * the states of the remaining groups. 152 */ 153 static int submit_stpg(struct scsi_device *sdev, int group_id, 154 struct scsi_sense_hdr *sshdr) 155 { 156 u8 cdb[MAX_COMMAND_SIZE]; 157 unsigned char stpg_data[8]; 158 int stpg_len = 8; 159 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | 160 REQ_FAILFAST_DRIVER; 161 162 /* Prepare the data buffer */ 163 memset(stpg_data, 0, stpg_len); 164 stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL; 165 put_unaligned_be16(group_id, &stpg_data[6]); 166 167 /* Prepare the command. */ 168 memset(cdb, 0x0, MAX_COMMAND_SIZE); 169 cdb[0] = MAINTENANCE_OUT; 170 cdb[1] = MO_SET_TARGET_PGS; 171 put_unaligned_be32(stpg_len, &cdb[6]); 172 173 return scsi_execute(sdev, cdb, DMA_TO_DEVICE, stpg_data, stpg_len, NULL, 174 sshdr, ALUA_FAILOVER_TIMEOUT * HZ, 175 ALUA_FAILOVER_RETRIES, req_flags, 0, NULL); 176 } 177 178 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size, 179 int group_id) 180 { 181 struct alua_port_group *pg; 182 183 if (!id_str || !id_size || !strlen(id_str)) 184 return NULL; 185 186 list_for_each_entry(pg, &port_group_list, node) { 187 if (pg->group_id != group_id) 188 continue; 189 if (!pg->device_id_len || pg->device_id_len != id_size) 190 continue; 191 if (strncmp(pg->device_id_str, id_str, id_size)) 192 continue; 193 if (!kref_get_unless_zero(&pg->kref)) 194 continue; 195 return pg; 196 } 197 198 return NULL; 199 } 200 201 /* 202 * alua_alloc_pg - Allocate a new port_group structure 203 * @sdev: scsi device 204 * @group_id: port group id 205 * @tpgs: target port group settings 206 * 207 * Allocate a new port_group structure for a given 208 * device. 209 */ 210 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev, 211 int group_id, int tpgs) 212 { 213 struct alua_port_group *pg, *tmp_pg; 214 215 pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL); 216 if (!pg) 217 return ERR_PTR(-ENOMEM); 218 219 pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str, 220 sizeof(pg->device_id_str)); 221 if (pg->device_id_len <= 0) { 222 /* 223 * TPGS supported but no device identification found. 224 * Generate private device identification. 225 */ 226 sdev_printk(KERN_INFO, sdev, 227 "%s: No device descriptors found\n", 228 ALUA_DH_NAME); 229 pg->device_id_str[0] = '\0'; 230 pg->device_id_len = 0; 231 } 232 pg->group_id = group_id; 233 pg->tpgs = tpgs; 234 pg->state = SCSI_ACCESS_STATE_OPTIMAL; 235 pg->valid_states = TPGS_SUPPORT_ALL; 236 if (optimize_stpg) 237 pg->flags |= ALUA_OPTIMIZE_STPG; 238 kref_init(&pg->kref); 239 INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work); 240 INIT_LIST_HEAD(&pg->rtpg_list); 241 INIT_LIST_HEAD(&pg->node); 242 INIT_LIST_HEAD(&pg->dh_list); 243 spin_lock_init(&pg->lock); 244 245 spin_lock(&port_group_lock); 246 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len, 247 group_id); 248 if (tmp_pg) { 249 spin_unlock(&port_group_lock); 250 kfree(pg); 251 return tmp_pg; 252 } 253 254 list_add(&pg->node, &port_group_list); 255 spin_unlock(&port_group_lock); 256 257 return pg; 258 } 259 260 /* 261 * alua_check_tpgs - Evaluate TPGS setting 262 * @sdev: device to be checked 263 * 264 * Examine the TPGS setting of the sdev to find out if ALUA 265 * is supported. 266 */ 267 static int alua_check_tpgs(struct scsi_device *sdev) 268 { 269 int tpgs = TPGS_MODE_NONE; 270 271 /* 272 * ALUA support for non-disk devices is fraught with 273 * difficulties, so disable it for now. 274 */ 275 if (sdev->type != TYPE_DISK) { 276 sdev_printk(KERN_INFO, sdev, 277 "%s: disable for non-disk devices\n", 278 ALUA_DH_NAME); 279 return tpgs; 280 } 281 282 tpgs = scsi_device_tpgs(sdev); 283 switch (tpgs) { 284 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT: 285 sdev_printk(KERN_INFO, sdev, 286 "%s: supports implicit and explicit TPGS\n", 287 ALUA_DH_NAME); 288 break; 289 case TPGS_MODE_EXPLICIT: 290 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n", 291 ALUA_DH_NAME); 292 break; 293 case TPGS_MODE_IMPLICIT: 294 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n", 295 ALUA_DH_NAME); 296 break; 297 case TPGS_MODE_NONE: 298 sdev_printk(KERN_INFO, sdev, "%s: not supported\n", 299 ALUA_DH_NAME); 300 break; 301 default: 302 sdev_printk(KERN_INFO, sdev, 303 "%s: unsupported TPGS setting %d\n", 304 ALUA_DH_NAME, tpgs); 305 tpgs = TPGS_MODE_NONE; 306 break; 307 } 308 309 return tpgs; 310 } 311 312 /* 313 * alua_check_vpd - Evaluate INQUIRY vpd page 0x83 314 * @sdev: device to be checked 315 * 316 * Extract the relative target port and the target port group 317 * descriptor from the list of identificators. 318 */ 319 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h, 320 int tpgs) 321 { 322 int rel_port = -1, group_id; 323 struct alua_port_group *pg, *old_pg = NULL; 324 bool pg_updated = false; 325 unsigned long flags; 326 327 group_id = scsi_vpd_tpg_id(sdev, &rel_port); 328 if (group_id < 0) { 329 /* 330 * Internal error; TPGS supported but required 331 * VPD identification descriptors not present. 332 * Disable ALUA support 333 */ 334 sdev_printk(KERN_INFO, sdev, 335 "%s: No target port descriptors found\n", 336 ALUA_DH_NAME); 337 return SCSI_DH_DEV_UNSUPP; 338 } 339 340 pg = alua_alloc_pg(sdev, group_id, tpgs); 341 if (IS_ERR(pg)) { 342 if (PTR_ERR(pg) == -ENOMEM) 343 return SCSI_DH_NOMEM; 344 return SCSI_DH_DEV_UNSUPP; 345 } 346 if (pg->device_id_len) 347 sdev_printk(KERN_INFO, sdev, 348 "%s: device %s port group %x rel port %x\n", 349 ALUA_DH_NAME, pg->device_id_str, 350 group_id, rel_port); 351 else 352 sdev_printk(KERN_INFO, sdev, 353 "%s: port group %x rel port %x\n", 354 ALUA_DH_NAME, group_id, rel_port); 355 356 /* Check for existing port group references */ 357 spin_lock(&h->pg_lock); 358 old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock)); 359 if (old_pg != pg) { 360 /* port group has changed. Update to new port group */ 361 if (h->pg) { 362 spin_lock_irqsave(&old_pg->lock, flags); 363 list_del_rcu(&h->node); 364 spin_unlock_irqrestore(&old_pg->lock, flags); 365 } 366 rcu_assign_pointer(h->pg, pg); 367 pg_updated = true; 368 } 369 370 spin_lock_irqsave(&pg->lock, flags); 371 if (pg_updated) 372 list_add_rcu(&h->node, &pg->dh_list); 373 spin_unlock_irqrestore(&pg->lock, flags); 374 375 alua_rtpg_queue(rcu_dereference_protected(h->pg, 376 lockdep_is_held(&h->pg_lock)), 377 sdev, NULL, true); 378 spin_unlock(&h->pg_lock); 379 380 if (old_pg) 381 kref_put(&old_pg->kref, release_port_group); 382 383 return SCSI_DH_OK; 384 } 385 386 static char print_alua_state(unsigned char state) 387 { 388 switch (state) { 389 case SCSI_ACCESS_STATE_OPTIMAL: 390 return 'A'; 391 case SCSI_ACCESS_STATE_ACTIVE: 392 return 'N'; 393 case SCSI_ACCESS_STATE_STANDBY: 394 return 'S'; 395 case SCSI_ACCESS_STATE_UNAVAILABLE: 396 return 'U'; 397 case SCSI_ACCESS_STATE_LBA: 398 return 'L'; 399 case SCSI_ACCESS_STATE_OFFLINE: 400 return 'O'; 401 case SCSI_ACCESS_STATE_TRANSITIONING: 402 return 'T'; 403 default: 404 return 'X'; 405 } 406 } 407 408 static int alua_check_sense(struct scsi_device *sdev, 409 struct scsi_sense_hdr *sense_hdr) 410 { 411 switch (sense_hdr->sense_key) { 412 case NOT_READY: 413 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) { 414 /* 415 * LUN Not Accessible - ALUA state transition 416 */ 417 alua_check(sdev, false); 418 return NEEDS_RETRY; 419 } 420 break; 421 case UNIT_ATTENTION: 422 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) { 423 /* 424 * Power On, Reset, or Bus Device Reset. 425 * Might have obscured a state transition, 426 * so schedule a recheck. 427 */ 428 alua_check(sdev, true); 429 return ADD_TO_MLQUEUE; 430 } 431 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04) 432 /* 433 * Device internal reset 434 */ 435 return ADD_TO_MLQUEUE; 436 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01) 437 /* 438 * Mode Parameters Changed 439 */ 440 return ADD_TO_MLQUEUE; 441 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) { 442 /* 443 * ALUA state changed 444 */ 445 alua_check(sdev, true); 446 return ADD_TO_MLQUEUE; 447 } 448 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) { 449 /* 450 * Implicit ALUA state transition failed 451 */ 452 alua_check(sdev, true); 453 return ADD_TO_MLQUEUE; 454 } 455 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03) 456 /* 457 * Inquiry data has changed 458 */ 459 return ADD_TO_MLQUEUE; 460 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e) 461 /* 462 * REPORTED_LUNS_DATA_HAS_CHANGED is reported 463 * when switching controllers on targets like 464 * Intel Multi-Flex. We can just retry. 465 */ 466 return ADD_TO_MLQUEUE; 467 break; 468 } 469 470 return SCSI_RETURN_NOT_HANDLED; 471 } 472 473 /* 474 * alua_tur - Send a TEST UNIT READY 475 * @sdev: device to which the TEST UNIT READY command should be send 476 * 477 * Send a TEST UNIT READY to @sdev to figure out the device state 478 * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING, 479 * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise. 480 */ 481 static int alua_tur(struct scsi_device *sdev) 482 { 483 struct scsi_sense_hdr sense_hdr; 484 int retval; 485 486 retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ, 487 ALUA_FAILOVER_RETRIES, &sense_hdr); 488 if (sense_hdr.sense_key == NOT_READY && 489 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) 490 return SCSI_DH_RETRY; 491 else if (retval) 492 return SCSI_DH_IO; 493 else 494 return SCSI_DH_OK; 495 } 496 497 /* 498 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES 499 * @sdev: the device to be evaluated. 500 * 501 * Evaluate the Target Port Group State. 502 * Returns SCSI_DH_DEV_OFFLINED if the path is 503 * found to be unusable. 504 */ 505 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg) 506 { 507 struct scsi_sense_hdr sense_hdr; 508 struct alua_port_group *tmp_pg; 509 int len, k, off, bufflen = ALUA_RTPG_SIZE; 510 unsigned char *desc, *buff; 511 unsigned err, retval; 512 unsigned int tpg_desc_tbl_off; 513 unsigned char orig_transition_tmo; 514 unsigned long flags; 515 bool transitioning_sense = false; 516 517 if (!pg->expiry) { 518 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ; 519 520 if (pg->transition_tmo) 521 transition_tmo = pg->transition_tmo * HZ; 522 523 pg->expiry = round_jiffies_up(jiffies + transition_tmo); 524 } 525 526 buff = kzalloc(bufflen, GFP_KERNEL); 527 if (!buff) 528 return SCSI_DH_DEV_TEMP_BUSY; 529 530 retry: 531 err = 0; 532 retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags); 533 534 if (retval) { 535 /* 536 * Some (broken) implementations have a habit of returning 537 * an error during things like firmware update etc. 538 * But if the target only supports active/optimized there's 539 * not much we can do; it's not that we can switch paths 540 * or anything. 541 * So ignore any errors to avoid spurious failures during 542 * path failover. 543 */ 544 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) { 545 sdev_printk(KERN_INFO, sdev, 546 "%s: ignoring rtpg result %d\n", 547 ALUA_DH_NAME, retval); 548 kfree(buff); 549 return SCSI_DH_OK; 550 } 551 if (!scsi_sense_valid(&sense_hdr)) { 552 sdev_printk(KERN_INFO, sdev, 553 "%s: rtpg failed, result %d\n", 554 ALUA_DH_NAME, retval); 555 kfree(buff); 556 if (driver_byte(retval) == DRIVER_ERROR) 557 return SCSI_DH_DEV_TEMP_BUSY; 558 return SCSI_DH_IO; 559 } 560 561 /* 562 * submit_rtpg() has failed on existing arrays 563 * when requesting extended header info, and 564 * the array doesn't support extended headers, 565 * even though it shouldn't according to T10. 566 * The retry without rtpg_ext_hdr_req set 567 * handles this. 568 */ 569 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) && 570 sense_hdr.sense_key == ILLEGAL_REQUEST && 571 sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) { 572 pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP; 573 goto retry; 574 } 575 /* 576 * If the array returns with 'ALUA state transition' 577 * sense code here it cannot return RTPG data during 578 * transition. So set the state to 'transitioning' directly. 579 */ 580 if (sense_hdr.sense_key == NOT_READY && 581 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) { 582 transitioning_sense = true; 583 goto skip_rtpg; 584 } 585 /* 586 * Retry on any other UNIT ATTENTION occurred. 587 */ 588 if (sense_hdr.sense_key == UNIT_ATTENTION) 589 err = SCSI_DH_RETRY; 590 if (err == SCSI_DH_RETRY && 591 pg->expiry != 0 && time_before(jiffies, pg->expiry)) { 592 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n", 593 ALUA_DH_NAME); 594 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr); 595 kfree(buff); 596 return err; 597 } 598 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n", 599 ALUA_DH_NAME); 600 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr); 601 kfree(buff); 602 pg->expiry = 0; 603 return SCSI_DH_IO; 604 } 605 606 len = get_unaligned_be32(&buff[0]) + 4; 607 608 if (len > bufflen) { 609 /* Resubmit with the correct length */ 610 kfree(buff); 611 bufflen = len; 612 buff = kmalloc(bufflen, GFP_KERNEL); 613 if (!buff) { 614 sdev_printk(KERN_WARNING, sdev, 615 "%s: kmalloc buffer failed\n",__func__); 616 /* Temporary failure, bypass */ 617 pg->expiry = 0; 618 return SCSI_DH_DEV_TEMP_BUSY; 619 } 620 goto retry; 621 } 622 623 orig_transition_tmo = pg->transition_tmo; 624 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0) 625 pg->transition_tmo = buff[5]; 626 else 627 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT; 628 629 if (orig_transition_tmo != pg->transition_tmo) { 630 sdev_printk(KERN_INFO, sdev, 631 "%s: transition timeout set to %d seconds\n", 632 ALUA_DH_NAME, pg->transition_tmo); 633 pg->expiry = jiffies + pg->transition_tmo * HZ; 634 } 635 636 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR) 637 tpg_desc_tbl_off = 8; 638 else 639 tpg_desc_tbl_off = 4; 640 641 for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off; 642 k < len; 643 k += off, desc += off) { 644 u16 group_id = get_unaligned_be16(&desc[2]); 645 646 spin_lock_irqsave(&port_group_lock, flags); 647 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len, 648 group_id); 649 spin_unlock_irqrestore(&port_group_lock, flags); 650 if (tmp_pg) { 651 if (spin_trylock_irqsave(&tmp_pg->lock, flags)) { 652 if ((tmp_pg == pg) || 653 !(tmp_pg->flags & ALUA_PG_RUNNING)) { 654 struct alua_dh_data *h; 655 656 tmp_pg->state = desc[0] & 0x0f; 657 tmp_pg->pref = desc[0] >> 7; 658 rcu_read_lock(); 659 list_for_each_entry_rcu(h, 660 &tmp_pg->dh_list, node) { 661 if (!h->sdev) 662 continue; 663 h->sdev->access_state = desc[0]; 664 } 665 rcu_read_unlock(); 666 } 667 if (tmp_pg == pg) 668 tmp_pg->valid_states = desc[1]; 669 spin_unlock_irqrestore(&tmp_pg->lock, flags); 670 } 671 kref_put(&tmp_pg->kref, release_port_group); 672 } 673 off = 8 + (desc[7] * 4); 674 } 675 676 skip_rtpg: 677 spin_lock_irqsave(&pg->lock, flags); 678 if (transitioning_sense) 679 pg->state = SCSI_ACCESS_STATE_TRANSITIONING; 680 681 sdev_printk(KERN_INFO, sdev, 682 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n", 683 ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state), 684 pg->pref ? "preferred" : "non-preferred", 685 pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t', 686 pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o', 687 pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l', 688 pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u', 689 pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s', 690 pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n', 691 pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a'); 692 693 switch (pg->state) { 694 case SCSI_ACCESS_STATE_TRANSITIONING: 695 if (time_before(jiffies, pg->expiry)) { 696 /* State transition, retry */ 697 pg->interval = ALUA_RTPG_RETRY_DELAY; 698 err = SCSI_DH_RETRY; 699 } else { 700 struct alua_dh_data *h; 701 702 /* Transitioning time exceeded, set port to standby */ 703 err = SCSI_DH_IO; 704 pg->state = SCSI_ACCESS_STATE_STANDBY; 705 pg->expiry = 0; 706 rcu_read_lock(); 707 list_for_each_entry_rcu(h, &pg->dh_list, node) { 708 if (!h->sdev) 709 continue; 710 h->sdev->access_state = 711 (pg->state & SCSI_ACCESS_STATE_MASK); 712 if (pg->pref) 713 h->sdev->access_state |= 714 SCSI_ACCESS_STATE_PREFERRED; 715 } 716 rcu_read_unlock(); 717 } 718 break; 719 case SCSI_ACCESS_STATE_OFFLINE: 720 /* Path unusable */ 721 err = SCSI_DH_DEV_OFFLINED; 722 pg->expiry = 0; 723 break; 724 default: 725 /* Useable path if active */ 726 err = SCSI_DH_OK; 727 pg->expiry = 0; 728 break; 729 } 730 spin_unlock_irqrestore(&pg->lock, flags); 731 kfree(buff); 732 return err; 733 } 734 735 /* 736 * alua_stpg - Issue a SET TARGET PORT GROUP command 737 * 738 * Issue a SET TARGET PORT GROUP command and evaluate the 739 * response. Returns SCSI_DH_RETRY per default to trigger 740 * a re-evaluation of the target group state or SCSI_DH_OK 741 * if no further action needs to be taken. 742 */ 743 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg) 744 { 745 int retval; 746 struct scsi_sense_hdr sense_hdr; 747 748 if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) { 749 /* Only implicit ALUA supported, retry */ 750 return SCSI_DH_RETRY; 751 } 752 switch (pg->state) { 753 case SCSI_ACCESS_STATE_OPTIMAL: 754 return SCSI_DH_OK; 755 case SCSI_ACCESS_STATE_ACTIVE: 756 if ((pg->flags & ALUA_OPTIMIZE_STPG) && 757 !pg->pref && 758 (pg->tpgs & TPGS_MODE_IMPLICIT)) 759 return SCSI_DH_OK; 760 break; 761 case SCSI_ACCESS_STATE_STANDBY: 762 case SCSI_ACCESS_STATE_UNAVAILABLE: 763 break; 764 case SCSI_ACCESS_STATE_OFFLINE: 765 return SCSI_DH_IO; 766 case SCSI_ACCESS_STATE_TRANSITIONING: 767 break; 768 default: 769 sdev_printk(KERN_INFO, sdev, 770 "%s: stpg failed, unhandled TPGS state %d", 771 ALUA_DH_NAME, pg->state); 772 return SCSI_DH_NOSYS; 773 } 774 retval = submit_stpg(sdev, pg->group_id, &sense_hdr); 775 776 if (retval) { 777 if (!scsi_sense_valid(&sense_hdr)) { 778 sdev_printk(KERN_INFO, sdev, 779 "%s: stpg failed, result %d", 780 ALUA_DH_NAME, retval); 781 if (driver_byte(retval) == DRIVER_ERROR) 782 return SCSI_DH_DEV_TEMP_BUSY; 783 } else { 784 sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n", 785 ALUA_DH_NAME); 786 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr); 787 } 788 } 789 /* Retry RTPG */ 790 return SCSI_DH_RETRY; 791 } 792 793 static void alua_rtpg_work(struct work_struct *work) 794 { 795 struct alua_port_group *pg = 796 container_of(work, struct alua_port_group, rtpg_work.work); 797 struct scsi_device *sdev; 798 LIST_HEAD(qdata_list); 799 int err = SCSI_DH_OK; 800 struct alua_queue_data *qdata, *tmp; 801 unsigned long flags; 802 803 spin_lock_irqsave(&pg->lock, flags); 804 sdev = pg->rtpg_sdev; 805 if (!sdev) { 806 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG); 807 WARN_ON(pg->flags & ALUA_PG_RUN_STPG); 808 spin_unlock_irqrestore(&pg->lock, flags); 809 kref_put(&pg->kref, release_port_group); 810 return; 811 } 812 pg->flags |= ALUA_PG_RUNNING; 813 if (pg->flags & ALUA_PG_RUN_RTPG) { 814 int state = pg->state; 815 816 pg->flags &= ~ALUA_PG_RUN_RTPG; 817 spin_unlock_irqrestore(&pg->lock, flags); 818 if (state == SCSI_ACCESS_STATE_TRANSITIONING) { 819 if (alua_tur(sdev) == SCSI_DH_RETRY) { 820 spin_lock_irqsave(&pg->lock, flags); 821 pg->flags &= ~ALUA_PG_RUNNING; 822 pg->flags |= ALUA_PG_RUN_RTPG; 823 if (!pg->interval) 824 pg->interval = ALUA_RTPG_RETRY_DELAY; 825 spin_unlock_irqrestore(&pg->lock, flags); 826 queue_delayed_work(kaluad_wq, &pg->rtpg_work, 827 pg->interval * HZ); 828 return; 829 } 830 /* Send RTPG on failure or if TUR indicates SUCCESS */ 831 } 832 err = alua_rtpg(sdev, pg); 833 spin_lock_irqsave(&pg->lock, flags); 834 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) { 835 pg->flags &= ~ALUA_PG_RUNNING; 836 if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG)) 837 pg->interval = ALUA_RTPG_RETRY_DELAY; 838 pg->flags |= ALUA_PG_RUN_RTPG; 839 spin_unlock_irqrestore(&pg->lock, flags); 840 queue_delayed_work(kaluad_wq, &pg->rtpg_work, 841 pg->interval * HZ); 842 return; 843 } 844 if (err != SCSI_DH_OK) 845 pg->flags &= ~ALUA_PG_RUN_STPG; 846 } 847 if (pg->flags & ALUA_PG_RUN_STPG) { 848 pg->flags &= ~ALUA_PG_RUN_STPG; 849 spin_unlock_irqrestore(&pg->lock, flags); 850 err = alua_stpg(sdev, pg); 851 spin_lock_irqsave(&pg->lock, flags); 852 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) { 853 pg->flags |= ALUA_PG_RUN_RTPG; 854 pg->interval = 0; 855 pg->flags &= ~ALUA_PG_RUNNING; 856 spin_unlock_irqrestore(&pg->lock, flags); 857 queue_delayed_work(kaluad_wq, &pg->rtpg_work, 858 pg->interval * HZ); 859 return; 860 } 861 } 862 863 list_splice_init(&pg->rtpg_list, &qdata_list); 864 pg->rtpg_sdev = NULL; 865 spin_unlock_irqrestore(&pg->lock, flags); 866 867 list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) { 868 list_del(&qdata->entry); 869 if (qdata->callback_fn) 870 qdata->callback_fn(qdata->callback_data, err); 871 kfree(qdata); 872 } 873 spin_lock_irqsave(&pg->lock, flags); 874 pg->flags &= ~ALUA_PG_RUNNING; 875 spin_unlock_irqrestore(&pg->lock, flags); 876 scsi_device_put(sdev); 877 kref_put(&pg->kref, release_port_group); 878 } 879 880 /** 881 * alua_rtpg_queue() - cause RTPG to be submitted asynchronously 882 * @pg: ALUA port group associated with @sdev. 883 * @sdev: SCSI device for which to submit an RTPG. 884 * @qdata: Information about the callback to invoke after the RTPG. 885 * @force: Whether or not to submit an RTPG if a work item that will submit an 886 * RTPG already has been scheduled. 887 * 888 * Returns true if and only if alua_rtpg_work() will be called asynchronously. 889 * That function is responsible for calling @qdata->fn(). 890 */ 891 static bool alua_rtpg_queue(struct alua_port_group *pg, 892 struct scsi_device *sdev, 893 struct alua_queue_data *qdata, bool force) 894 { 895 int start_queue = 0; 896 unsigned long flags; 897 if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev)) 898 return false; 899 900 spin_lock_irqsave(&pg->lock, flags); 901 if (qdata) { 902 list_add_tail(&qdata->entry, &pg->rtpg_list); 903 pg->flags |= ALUA_PG_RUN_STPG; 904 force = true; 905 } 906 if (pg->rtpg_sdev == NULL) { 907 pg->interval = 0; 908 pg->flags |= ALUA_PG_RUN_RTPG; 909 kref_get(&pg->kref); 910 pg->rtpg_sdev = sdev; 911 start_queue = 1; 912 } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) { 913 pg->flags |= ALUA_PG_RUN_RTPG; 914 /* Do not queue if the worker is already running */ 915 if (!(pg->flags & ALUA_PG_RUNNING)) { 916 kref_get(&pg->kref); 917 start_queue = 1; 918 } 919 } 920 921 spin_unlock_irqrestore(&pg->lock, flags); 922 923 if (start_queue) { 924 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work, 925 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS))) 926 sdev = NULL; 927 else 928 kref_put(&pg->kref, release_port_group); 929 } 930 if (sdev) 931 scsi_device_put(sdev); 932 933 return true; 934 } 935 936 /* 937 * alua_initialize - Initialize ALUA state 938 * @sdev: the device to be initialized 939 * 940 * For the prep_fn to work correctly we have 941 * to initialize the ALUA state for the device. 942 */ 943 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h) 944 { 945 int err = SCSI_DH_DEV_UNSUPP, tpgs; 946 947 mutex_lock(&h->init_mutex); 948 tpgs = alua_check_tpgs(sdev); 949 if (tpgs != TPGS_MODE_NONE) 950 err = alua_check_vpd(sdev, h, tpgs); 951 h->init_error = err; 952 mutex_unlock(&h->init_mutex); 953 return err; 954 } 955 /* 956 * alua_set_params - set/unset the optimize flag 957 * @sdev: device on the path to be activated 958 * params - parameters in the following format 959 * "no_of_params\0param1\0param2\0param3\0...\0" 960 * For example, to set the flag pass the following parameters 961 * from multipath.conf 962 * hardware_handler "2 alua 1" 963 */ 964 static int alua_set_params(struct scsi_device *sdev, const char *params) 965 { 966 struct alua_dh_data *h = sdev->handler_data; 967 struct alua_port_group *pg = NULL; 968 unsigned int optimize = 0, argc; 969 const char *p = params; 970 int result = SCSI_DH_OK; 971 unsigned long flags; 972 973 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1)) 974 return -EINVAL; 975 976 while (*p++) 977 ; 978 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1)) 979 return -EINVAL; 980 981 rcu_read_lock(); 982 pg = rcu_dereference(h->pg); 983 if (!pg) { 984 rcu_read_unlock(); 985 return -ENXIO; 986 } 987 spin_lock_irqsave(&pg->lock, flags); 988 if (optimize) 989 pg->flags |= ALUA_OPTIMIZE_STPG; 990 else 991 pg->flags &= ~ALUA_OPTIMIZE_STPG; 992 spin_unlock_irqrestore(&pg->lock, flags); 993 rcu_read_unlock(); 994 995 return result; 996 } 997 998 /* 999 * alua_activate - activate a path 1000 * @sdev: device on the path to be activated 1001 * 1002 * We're currently switching the port group to be activated only and 1003 * let the array figure out the rest. 1004 * There may be other arrays which require us to switch all port groups 1005 * based on a certain policy. But until we actually encounter them it 1006 * should be okay. 1007 */ 1008 static int alua_activate(struct scsi_device *sdev, 1009 activate_complete fn, void *data) 1010 { 1011 struct alua_dh_data *h = sdev->handler_data; 1012 int err = SCSI_DH_OK; 1013 struct alua_queue_data *qdata; 1014 struct alua_port_group *pg; 1015 1016 qdata = kzalloc(sizeof(*qdata), GFP_KERNEL); 1017 if (!qdata) { 1018 err = SCSI_DH_RES_TEMP_UNAVAIL; 1019 goto out; 1020 } 1021 qdata->callback_fn = fn; 1022 qdata->callback_data = data; 1023 1024 mutex_lock(&h->init_mutex); 1025 rcu_read_lock(); 1026 pg = rcu_dereference(h->pg); 1027 if (!pg || !kref_get_unless_zero(&pg->kref)) { 1028 rcu_read_unlock(); 1029 kfree(qdata); 1030 err = h->init_error; 1031 mutex_unlock(&h->init_mutex); 1032 goto out; 1033 } 1034 rcu_read_unlock(); 1035 mutex_unlock(&h->init_mutex); 1036 1037 if (alua_rtpg_queue(pg, sdev, qdata, true)) 1038 fn = NULL; 1039 else 1040 err = SCSI_DH_DEV_OFFLINED; 1041 kref_put(&pg->kref, release_port_group); 1042 out: 1043 if (fn) 1044 fn(data, err); 1045 return 0; 1046 } 1047 1048 /* 1049 * alua_check - check path status 1050 * @sdev: device on the path to be checked 1051 * 1052 * Check the device status 1053 */ 1054 static void alua_check(struct scsi_device *sdev, bool force) 1055 { 1056 struct alua_dh_data *h = sdev->handler_data; 1057 struct alua_port_group *pg; 1058 1059 rcu_read_lock(); 1060 pg = rcu_dereference(h->pg); 1061 if (!pg || !kref_get_unless_zero(&pg->kref)) { 1062 rcu_read_unlock(); 1063 return; 1064 } 1065 rcu_read_unlock(); 1066 1067 alua_rtpg_queue(pg, sdev, NULL, force); 1068 kref_put(&pg->kref, release_port_group); 1069 } 1070 1071 /* 1072 * alua_prep_fn - request callback 1073 * 1074 * Fail I/O to all paths not in state 1075 * active/optimized or active/non-optimized. 1076 */ 1077 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req) 1078 { 1079 struct alua_dh_data *h = sdev->handler_data; 1080 struct alua_port_group *pg; 1081 unsigned char state = SCSI_ACCESS_STATE_OPTIMAL; 1082 1083 rcu_read_lock(); 1084 pg = rcu_dereference(h->pg); 1085 if (pg) 1086 state = pg->state; 1087 rcu_read_unlock(); 1088 1089 switch (state) { 1090 case SCSI_ACCESS_STATE_OPTIMAL: 1091 case SCSI_ACCESS_STATE_ACTIVE: 1092 case SCSI_ACCESS_STATE_LBA: 1093 return BLK_STS_OK; 1094 case SCSI_ACCESS_STATE_TRANSITIONING: 1095 return BLK_STS_RESOURCE; 1096 default: 1097 req->rq_flags |= RQF_QUIET; 1098 return BLK_STS_IOERR; 1099 } 1100 } 1101 1102 static void alua_rescan(struct scsi_device *sdev) 1103 { 1104 struct alua_dh_data *h = sdev->handler_data; 1105 1106 alua_initialize(sdev, h); 1107 } 1108 1109 /* 1110 * alua_bus_attach - Attach device handler 1111 * @sdev: device to be attached to 1112 */ 1113 static int alua_bus_attach(struct scsi_device *sdev) 1114 { 1115 struct alua_dh_data *h; 1116 int err; 1117 1118 h = kzalloc(sizeof(*h) , GFP_KERNEL); 1119 if (!h) 1120 return SCSI_DH_NOMEM; 1121 spin_lock_init(&h->pg_lock); 1122 rcu_assign_pointer(h->pg, NULL); 1123 h->init_error = SCSI_DH_OK; 1124 h->sdev = sdev; 1125 INIT_LIST_HEAD(&h->node); 1126 1127 mutex_init(&h->init_mutex); 1128 err = alua_initialize(sdev, h); 1129 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED) 1130 goto failed; 1131 1132 sdev->handler_data = h; 1133 return SCSI_DH_OK; 1134 failed: 1135 kfree(h); 1136 return err; 1137 } 1138 1139 /* 1140 * alua_bus_detach - Detach device handler 1141 * @sdev: device to be detached from 1142 */ 1143 static void alua_bus_detach(struct scsi_device *sdev) 1144 { 1145 struct alua_dh_data *h = sdev->handler_data; 1146 struct alua_port_group *pg; 1147 1148 spin_lock(&h->pg_lock); 1149 pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock)); 1150 rcu_assign_pointer(h->pg, NULL); 1151 spin_unlock(&h->pg_lock); 1152 if (pg) { 1153 spin_lock_irq(&pg->lock); 1154 list_del_rcu(&h->node); 1155 spin_unlock_irq(&pg->lock); 1156 kref_put(&pg->kref, release_port_group); 1157 } 1158 sdev->handler_data = NULL; 1159 synchronize_rcu(); 1160 kfree(h); 1161 } 1162 1163 static struct scsi_device_handler alua_dh = { 1164 .name = ALUA_DH_NAME, 1165 .module = THIS_MODULE, 1166 .attach = alua_bus_attach, 1167 .detach = alua_bus_detach, 1168 .prep_fn = alua_prep_fn, 1169 .check_sense = alua_check_sense, 1170 .activate = alua_activate, 1171 .rescan = alua_rescan, 1172 .set_params = alua_set_params, 1173 }; 1174 1175 static int __init alua_init(void) 1176 { 1177 int r; 1178 1179 kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0); 1180 if (!kaluad_wq) 1181 return -ENOMEM; 1182 1183 r = scsi_register_device_handler(&alua_dh); 1184 if (r != 0) { 1185 printk(KERN_ERR "%s: Failed to register scsi device handler", 1186 ALUA_DH_NAME); 1187 destroy_workqueue(kaluad_wq); 1188 } 1189 return r; 1190 } 1191 1192 static void __exit alua_exit(void) 1193 { 1194 scsi_unregister_device_handler(&alua_dh); 1195 destroy_workqueue(kaluad_wq); 1196 } 1197 1198 module_init(alua_init); 1199 module_exit(alua_exit); 1200 1201 MODULE_DESCRIPTION("DM Multipath ALUA support"); 1202 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>"); 1203 MODULE_LICENSE("GPL"); 1204 MODULE_VERSION(ALUA_DH_VER); 1205