drivers/scsi/device_handler/scsi_dh_alua.c

   1 /*
   2  * Generic SCSI-3 ALUA SCSI Device Handler
   3  *
   4  * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
   5  * All rights reserved.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  20  *
  21  */
  22 #include <linux/slab.h>
  23 #include <linux/delay.h>
  24 #include <linux/module.h>
  25 #include <asm/unaligned.h>
  26 #include <scsi/scsi.h>
  27 #include <scsi/scsi_dbg.h>
  28 #include <scsi/scsi_eh.h>
  29 #include <scsi/scsi_dh.h>
  30
  31 #define ALUA_DH_NAME "alua"
  32 #define ALUA_DH_VER "1.3"
  33
  34 #define TPGS_STATE_OPTIMIZED            0x0
  35 #define TPGS_STATE_NONOPTIMIZED         0x1
  36 #define TPGS_STATE_STANDBY              0x2
  37 #define TPGS_STATE_UNAVAILABLE          0x3
  38 #define TPGS_STATE_LBA_DEPENDENT        0x4
  39 #define TPGS_STATE_OFFLINE              0xe
  40 #define TPGS_STATE_TRANSITIONING        0xf
  41
  42 #define TPGS_SUPPORT_NONE               0x00
  43 #define TPGS_SUPPORT_OPTIMIZED          0x01
  44 #define TPGS_SUPPORT_NONOPTIMIZED       0x02
  45 #define TPGS_SUPPORT_STANDBY            0x04
  46 #define TPGS_SUPPORT_UNAVAILABLE        0x08
  47 #define TPGS_SUPPORT_LBA_DEPENDENT      0x10
  48 #define TPGS_SUPPORT_OFFLINE            0x40
  49 #define TPGS_SUPPORT_TRANSITION         0x80
  50
  51 #define RTPG_FMT_MASK                   0x70
  52 #define RTPG_FMT_EXT_HDR                0x10
  53
  54 #define TPGS_MODE_UNINITIALIZED          -1
  55 #define TPGS_MODE_NONE                  0x0
  56 #define TPGS_MODE_IMPLICIT              0x1
  57 #define TPGS_MODE_EXPLICIT              0x2
  58
  59 #define ALUA_RTPG_SIZE                  128
  60 #define ALUA_FAILOVER_TIMEOUT           60
  61 #define ALUA_FAILOVER_RETRIES           5
  62 #define ALUA_RTPG_DELAY_MSECS           5
  63
  64 /* device handler flags */
  65 #define ALUA_OPTIMIZE_STPG              0x01
  66 #define ALUA_RTPG_EXT_HDR_UNSUPP        0x02
  67 #define ALUA_SYNC_STPG                  0x04
  68 /* State machine flags */
  69 #define ALUA_PG_RUN_RTPG                0x10
  70 #define ALUA_PG_RUN_STPG                0x20
  71 #define ALUA_PG_RUNNING                 0x40
  72
  73 static uint optimize_stpg;
  74 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
  75 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.");
  76
  77 static LIST_HEAD(port_group_list);
  78 static DEFINE_SPINLOCK(port_group_lock);
  79 static struct workqueue_struct *kaluad_wq;
  80 static struct workqueue_struct *kaluad_sync_wq;
  81
  82 struct alua_port_group {
  83         struct kref             kref;
  84         struct rcu_head         rcu;
  85         struct list_head        node;
  86         unsigned char           device_id_str[256];
  87         int                     device_id_len;
  88         int                     group_id;
  89         int                     tpgs;
  90         int                     state;
  91         int                     pref;
  92         unsigned                flags; /* used for optimizing STPG */
  93         unsigned char           transition_tmo;
  94         unsigned long           expiry;
  95         unsigned long           interval;
  96         struct delayed_work     rtpg_work;
  97         spinlock_t              lock;
  98         struct list_head        rtpg_list;
  99         struct scsi_device      *rtpg_sdev;
 100 };
 101
 102 struct alua_dh_data {
 103         struct alua_port_group  *pg;
 104         int                     group_id;
 105         spinlock_t              pg_lock;
 106         struct scsi_device      *sdev;
 107         int                     init_error;
 108         struct mutex            init_mutex;
 109 };
 110
 111 struct alua_queue_data {
 112         struct list_head        entry;
 113         activate_complete       callback_fn;
 114         void                    *callback_data;
 115 };
 116
 117 #define ALUA_POLICY_SWITCH_CURRENT      0
 118 #define ALUA_POLICY_SWITCH_ALL          1
 119
 120 static void alua_rtpg_work(struct work_struct *work);
 121 static void alua_rtpg_queue(struct alua_port_group *pg,
 122                             struct scsi_device *sdev,
 123                             struct alua_queue_data *qdata);
 124
 125 static void release_port_group(struct kref *kref)
 126 {
 127         struct alua_port_group *pg;
 128
 129         pg = container_of(kref, struct alua_port_group, kref);
 130         if (pg->rtpg_sdev)
 131                 flush_delayed_work(&pg->rtpg_work);
 132         spin_lock(&port_group_lock);
 133         list_del(&pg->node);
 134         spin_unlock(&port_group_lock);
 135         kfree_rcu(pg, rcu);
 136 }
 137
 138 /*
 139  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
 140  * @sdev: sdev the command should be sent to
 141  */
 142 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
 143                        int bufflen, struct scsi_sense_hdr *sshdr, int flags)
 144 {
 145         u8 cdb[COMMAND_SIZE(MAINTENANCE_IN)];
 146         int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
 147                 REQ_FAILFAST_DRIVER;
 148
 149         /* Prepare the command. */
 150         memset(cdb, 0x0, COMMAND_SIZE(MAINTENANCE_IN));
 151         cdb[0] = MAINTENANCE_IN;
 152         if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
 153                 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
 154         else
 155                 cdb[1] = MI_REPORT_TARGET_PGS;
 156         put_unaligned_be32(bufflen, &cdb[6]);
 157
 158         return scsi_execute_req_flags(sdev, cdb, DMA_FROM_DEVICE,
 159                                       buff, bufflen, sshdr,
 160                                       ALUA_FAILOVER_TIMEOUT * HZ,
 161                                       ALUA_FAILOVER_RETRIES, NULL, req_flags);
 162 }
 163
 164 /*
 165  * submit_stpg - Issue a SET TARGET PORT GROUP command
 166  *
 167  * Currently we're only setting the current target port group state
 168  * to 'active/optimized' and let the array firmware figure out
 169  * the states of the remaining groups.
 170  */
 171 static int submit_stpg(struct scsi_device *sdev, int group_id,
 172                        struct scsi_sense_hdr *sshdr)
 173 {
 174         u8 cdb[COMMAND_SIZE(MAINTENANCE_OUT)];
 175         unsigned char stpg_data[8];
 176         int stpg_len = 8;
 177         int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
 178                 REQ_FAILFAST_DRIVER;
 179
 180         /* Prepare the data buffer */
 181         memset(stpg_data, 0, stpg_len);
 182         stpg_data[4] = TPGS_STATE_OPTIMIZED & 0x0f;
 183         put_unaligned_be16(group_id, &stpg_data[6]);
 184
 185         /* Prepare the command. */
 186         memset(cdb, 0x0, COMMAND_SIZE(MAINTENANCE_OUT));
 187         cdb[0] = MAINTENANCE_OUT;
 188         cdb[1] = MO_SET_TARGET_PGS;
 189         put_unaligned_be32(stpg_len, &cdb[6]);
 190
 191         return scsi_execute_req_flags(sdev, cdb, DMA_TO_DEVICE,
 192                                       stpg_data, stpg_len,
 193                                       sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
 194                                       ALUA_FAILOVER_RETRIES, NULL, req_flags);
 195 }
 196
 197 struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
 198                                          int group_id)
 199 {
 200         struct alua_port_group *pg;
 201
 202         list_for_each_entry(pg, &port_group_list, node) {
 203                 if (pg->group_id != group_id)
 204                         continue;
 205                 if (pg->device_id_len != id_size)
 206                         continue;
 207                 if (strncmp(pg->device_id_str, id_str, id_size))
 208                         continue;
 209                 if (!kref_get_unless_zero(&pg->kref))
 210                         continue;
 211                 return pg;
 212         }
 213
 214         return NULL;
 215 }
 216
 217 /*
 218  * alua_alloc_pg - Allocate a new port_group structure
 219  * @sdev: scsi device
 220  * @h: alua device_handler data
 221  * @group_id: port group id
 222  *
 223  * Allocate a new port_group structure for a given
 224  * device.
 225  */
 226 struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
 227                                       int group_id, int tpgs)
 228 {
 229         struct alua_port_group *pg, *tmp_pg;
 230
 231         pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
 232         if (!pg)
 233                 return ERR_PTR(-ENOMEM);
 234
 235         pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
 236                                             sizeof(pg->device_id_str));
 237         if (pg->device_id_len <= 0) {
 238                 /*
 239                  * Internal error: TPGS supported but no device
 240                  * identifcation found. Disable ALUA support.
 241                  */
 242                 kfree(pg);
 243                 sdev_printk(KERN_INFO, sdev,
 244                             "%s: No device descriptors found\n",
 245                             ALUA_DH_NAME);
 246                 return ERR_PTR(-ENXIO);
 247         }
 248         pg->group_id = group_id;
 249         pg->tpgs = tpgs;
 250         pg->state = TPGS_STATE_OPTIMIZED;
 251         if (optimize_stpg)
 252                 pg->flags |= ALUA_OPTIMIZE_STPG;
 253         kref_init(&pg->kref);
 254         INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
 255         INIT_LIST_HEAD(&pg->rtpg_list);
 256         INIT_LIST_HEAD(&pg->node);
 257         spin_lock_init(&pg->lock);
 258
 259         spin_lock(&port_group_lock);
 260         tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
 261                                   group_id);
 262         if (tmp_pg) {
 263                 spin_unlock(&port_group_lock);
 264                 kfree(pg);
 265                 return tmp_pg;
 266         }
 267
 268         list_add(&pg->node, &port_group_list);
 269         spin_unlock(&port_group_lock);
 270
 271         return pg;
 272 }
 273
 274 /*
 275  * alua_check_tpgs - Evaluate TPGS setting
 276  * @sdev: device to be checked
 277  *
 278  * Examine the TPGS setting of the sdev to find out if ALUA
 279  * is supported.
 280  */
 281 static int alua_check_tpgs(struct scsi_device *sdev)
 282 {
 283         int tpgs = TPGS_MODE_NONE;
 284
 285         /*
 286          * ALUA support for non-disk devices is fraught with
 287          * difficulties, so disable it for now.
 288          */
 289         if (sdev->type != TYPE_DISK) {
 290                 sdev_printk(KERN_INFO, sdev,
 291                             "%s: disable for non-disk devices\n",
 292                             ALUA_DH_NAME);
 293                 return tpgs;
 294         }
 295
 296         tpgs = scsi_device_tpgs(sdev);
 297         switch (tpgs) {
 298         case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
 299                 sdev_printk(KERN_INFO, sdev,
 300                             "%s: supports implicit and explicit TPGS\n",
 301                             ALUA_DH_NAME);
 302                 break;
 303         case TPGS_MODE_EXPLICIT:
 304                 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
 305                             ALUA_DH_NAME);
 306                 break;
 307         case TPGS_MODE_IMPLICIT:
 308                 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
 309                             ALUA_DH_NAME);
 310                 break;
 311         case TPGS_MODE_NONE:
 312                 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
 313                             ALUA_DH_NAME);
 314                 break;
 315         default:
 316                 sdev_printk(KERN_INFO, sdev,
 317                             "%s: unsupported TPGS setting %d\n",
 318                             ALUA_DH_NAME, tpgs);
 319                 tpgs = TPGS_MODE_NONE;
 320                 break;
 321         }
 322
 323         return tpgs;
 324 }
 325
 326 /*
 327  * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
 328  * @sdev: device to be checked
 329  *
 330  * Extract the relative target port and the target port group
 331  * descriptor from the list of identificators.
 332  */
 333 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
 334                           int tpgs)
 335 {
 336         int rel_port = -1, group_id;
 337         struct alua_port_group *pg, *old_pg = NULL;
 338
 339         group_id = scsi_vpd_tpg_id(sdev, &rel_port);
 340         if (group_id < 0) {
 341                 /*
 342                  * Internal error; TPGS supported but required
 343                  * VPD identification descriptors not present.
 344                  * Disable ALUA support
 345                  */
 346                 sdev_printk(KERN_INFO, sdev,
 347                             "%s: No target port descriptors found\n",
 348                             ALUA_DH_NAME);
 349                 return SCSI_DH_DEV_UNSUPP;
 350         }
 351
 352         pg = alua_alloc_pg(sdev, group_id, tpgs);
 353         if (IS_ERR(pg)) {
 354                 if (PTR_ERR(pg) == -ENOMEM)
 355                         return SCSI_DH_NOMEM;
 356                 return SCSI_DH_DEV_UNSUPP;
 357         }
 358         sdev_printk(KERN_INFO, sdev,
 359                     "%s: device %s port group %x rel port %x\n",
 360                     ALUA_DH_NAME, pg->device_id_str, group_id, rel_port);
 361
 362         /* Check for existing port group references */
 363         spin_lock(&h->pg_lock);
 364         old_pg = h->pg;
 365         if (old_pg != pg) {
 366                 /* port group has changed. Update to new port group */
 367                 rcu_assign_pointer(h->pg, pg);
 368         }
 369         alua_rtpg_queue(h->pg, sdev, NULL);
 370         spin_unlock(&h->pg_lock);
 371
 372         if (old_pg)
 373                 kref_put(&old_pg->kref, release_port_group);
 374
 375         return SCSI_DH_OK;
 376 }
 377
 378 static char print_alua_state(int state)
 379 {
 380         switch (state) {
 381         case TPGS_STATE_OPTIMIZED:
 382                 return 'A';
 383         case TPGS_STATE_NONOPTIMIZED:
 384                 return 'N';
 385         case TPGS_STATE_STANDBY:
 386                 return 'S';
 387         case TPGS_STATE_UNAVAILABLE:
 388                 return 'U';
 389         case TPGS_STATE_LBA_DEPENDENT:
 390                 return 'L';
 391         case TPGS_STATE_OFFLINE:
 392                 return 'O';
 393         case TPGS_STATE_TRANSITIONING:
 394                 return 'T';
 395         default:
 396                 return 'X';
 397         }
 398 }
 399
 400 static int alua_check_sense(struct scsi_device *sdev,
 401                             struct scsi_sense_hdr *sense_hdr)
 402 {
 403         switch (sense_hdr->sense_key) {
 404         case NOT_READY:
 405                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
 406                         /*
 407                          * LUN Not Accessible - ALUA state transition
 408                          */
 409                         return ADD_TO_MLQUEUE;
 410                 break;
 411         case UNIT_ATTENTION:
 412                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
 413                         /*
 414                          * Power On, Reset, or Bus Device Reset, just retry.
 415                          */
 416                         return ADD_TO_MLQUEUE;
 417                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
 418                         /*
 419                          * Device internal reset
 420                          */
 421                         return ADD_TO_MLQUEUE;
 422                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
 423                         /*
 424                          * Mode Parameters Changed
 425                          */
 426                         return ADD_TO_MLQUEUE;
 427                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06)
 428                         /*
 429                          * ALUA state changed
 430                          */
 431                         return ADD_TO_MLQUEUE;
 432                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07)
 433                         /*
 434                          * Implicit ALUA state transition failed
 435                          */
 436                         return ADD_TO_MLQUEUE;
 437                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
 438                         /*
 439                          * Inquiry data has changed
 440                          */
 441                         return ADD_TO_MLQUEUE;
 442                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
 443                         /*
 444                          * REPORTED_LUNS_DATA_HAS_CHANGED is reported
 445                          * when switching controllers on targets like
 446                          * Intel Multi-Flex. We can just retry.
 447                          */
 448                         return ADD_TO_MLQUEUE;
 449                 break;
 450         }
 451
 452         return SCSI_RETURN_NOT_HANDLED;
 453 }
 454
 455 /*
 456  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
 457  * @sdev: the device to be evaluated.
 458  *
 459  * Evaluate the Target Port Group State.
 460  * Returns SCSI_DH_DEV_OFFLINED if the path is
 461  * found to be unusable.
 462  */
 463 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
 464 {
 465         struct scsi_sense_hdr sense_hdr;
 466         int len, k, off, valid_states = 0, bufflen = ALUA_RTPG_SIZE;
 467         unsigned char *ucp, *buff;
 468         unsigned err, retval;
 469         unsigned int tpg_desc_tbl_off;
 470         unsigned char orig_transition_tmo;
 471
 472         if (!pg->expiry) {
 473                 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
 474
 475                 if (pg->transition_tmo)
 476                         transition_tmo = pg->transition_tmo * HZ;
 477
 478                 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
 479         }
 480
 481         buff = kzalloc(bufflen, GFP_KERNEL);
 482         if (!buff)
 483                 return SCSI_DH_DEV_TEMP_BUSY;
 484
 485  retry:
 486         retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
 487
 488         if (retval) {
 489                 if (!scsi_sense_valid(&sense_hdr)) {
 490                         sdev_printk(KERN_INFO, sdev,
 491                                     "%s: rtpg failed, result %d\n",
 492                                     ALUA_DH_NAME, retval);
 493                         kfree(buff);
 494                         if (driver_byte(retval) == DRIVER_ERROR)
 495                                 return SCSI_DH_DEV_TEMP_BUSY;
 496                         return SCSI_DH_IO;
 497                 }
 498
 499                 /*
 500                  * submit_rtpg() has failed on existing arrays
 501                  * when requesting extended header info, and
 502                  * the array doesn't support extended headers,
 503                  * even though it shouldn't according to T10.
 504                  * The retry without rtpg_ext_hdr_req set
 505                  * handles this.
 506                  */
 507                 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
 508                     sense_hdr.sense_key == ILLEGAL_REQUEST &&
 509                     sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
 510                         pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
 511                         goto retry;
 512                 }
 513                 /*
 514                  * Retry on ALUA state transition or if any
 515                  * UNIT ATTENTION occurred.
 516                  */
 517                 if (sense_hdr.sense_key == NOT_READY &&
 518                     sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
 519                         err = SCSI_DH_RETRY;
 520                 else if (sense_hdr.sense_key == UNIT_ATTENTION)
 521                         err = SCSI_DH_RETRY;
 522                 if (err == SCSI_DH_RETRY &&
 523                     pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
 524                         sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
 525                                     ALUA_DH_NAME);
 526                         scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
 527                         return err;
 528                 }
 529                 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
 530                             ALUA_DH_NAME);
 531                 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
 532                 kfree(buff);
 533                 pg->expiry = 0;
 534                 return SCSI_DH_IO;
 535         }
 536
 537         len = get_unaligned_be32(&buff[0]) + 4;
 538
 539         if (len > bufflen) {
 540                 /* Resubmit with the correct length */
 541                 kfree(buff);
 542                 bufflen = len;
 543                 buff = kmalloc(bufflen, GFP_KERNEL);
 544                 if (!buff) {
 545                         sdev_printk(KERN_WARNING, sdev,
 546                                     "%s: kmalloc buffer failed\n",__func__);
 547                         /* Temporary failure, bypass */
 548                         pg->expiry = 0;
 549                         return SCSI_DH_DEV_TEMP_BUSY;
 550                 }
 551                 goto retry;
 552         }
 553
 554         orig_transition_tmo = pg->transition_tmo;
 555         if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
 556                 pg->transition_tmo = buff[5];
 557         else
 558                 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
 559
 560         if (orig_transition_tmo != pg->transition_tmo) {
 561                 sdev_printk(KERN_INFO, sdev,
 562                             "%s: transition timeout set to %d seconds\n",
 563                             ALUA_DH_NAME, pg->transition_tmo);
 564                 pg->expiry = jiffies + pg->transition_tmo * HZ;
 565         }
 566
 567         if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
 568                 tpg_desc_tbl_off = 8;
 569         else
 570                 tpg_desc_tbl_off = 4;
 571
 572         for (k = tpg_desc_tbl_off, ucp = buff + tpg_desc_tbl_off;
 573              k < len;
 574              k += off, ucp += off) {
 575
 576                 if (pg->group_id == get_unaligned_be16(&ucp[2])) {
 577                         pg->state = ucp[0] & 0x0f;
 578                         pg->pref = ucp[0] >> 7;
 579                         valid_states = ucp[1];
 580                 }
 581                 off = 8 + (ucp[7] * 4);
 582         }
 583
 584         sdev_printk(KERN_INFO, sdev,
 585                     "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
 586                     ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
 587                     pg->pref ? "preferred" : "non-preferred",
 588                     valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
 589                     valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
 590                     valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
 591                     valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
 592                     valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
 593                     valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
 594                     valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
 595
 596         switch (pg->state) {
 597         case TPGS_STATE_TRANSITIONING:
 598                 if (time_before(jiffies, pg->expiry)) {
 599                         /* State transition, retry */
 600                         pg->interval = 2;
 601                         err = SCSI_DH_RETRY;
 602                 } else {
 603                         /* Transitioning time exceeded, set port to standby */
 604                         err = SCSI_DH_IO;
 605                         pg->state = TPGS_STATE_STANDBY;
 606                         pg->expiry = 0;
 607                 }
 608                 break;
 609         case TPGS_STATE_OFFLINE:
 610                 /* Path unusable */
 611                 err = SCSI_DH_DEV_OFFLINED;
 612                 pg->expiry = 0;
 613                 break;
 614         default:
 615                 /* Useable path if active */
 616                 err = SCSI_DH_OK;
 617                 pg->expiry = 0;
 618                 break;
 619         }
 620         kfree(buff);
 621         return err;
 622 }
 623
 624 /*
 625  * alua_stpg - Issue a SET TARGET PORT GROUP command
 626  *
 627  * Issue a SET TARGET PORT GROUP command and evaluate the
 628  * response. Returns SCSI_DH_RETRY per default to trigger
 629  * a re-evaluation of the target group state or SCSI_DH_OK
 630  * if no further action needs to be taken.
 631  */
 632 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
 633 {
 634         int retval;
 635         struct scsi_sense_hdr sense_hdr;
 636
 637         if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
 638                 /* Only implicit ALUA supported, retry */
 639                 return SCSI_DH_RETRY;
 640         }
 641         switch (pg->state) {
 642         case TPGS_STATE_OPTIMIZED:
 643                 return SCSI_DH_OK;
 644         case TPGS_STATE_NONOPTIMIZED:
 645                 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
 646                     !pg->pref &&
 647                     (pg->tpgs & TPGS_MODE_IMPLICIT))
 648                         return SCSI_DH_OK;
 649                 break;
 650         case TPGS_STATE_STANDBY:
 651         case TPGS_STATE_UNAVAILABLE:
 652                 break;
 653         case TPGS_STATE_OFFLINE:
 654                 return SCSI_DH_IO;
 655         case TPGS_STATE_TRANSITIONING:
 656                 break;
 657         default:
 658                 sdev_printk(KERN_INFO, sdev,
 659                             "%s: stpg failed, unhandled TPGS state %d",
 660                             ALUA_DH_NAME, pg->state);
 661                 return SCSI_DH_NOSYS;
 662         }
 663         retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
 664
 665         if (retval) {
 666                 if (!scsi_sense_valid(&sense_hdr)) {
 667                         sdev_printk(KERN_INFO, sdev,
 668                                     "%s: stpg failed, result %d",
 669                                     ALUA_DH_NAME, retval);
 670                         if (driver_byte(retval) == DRIVER_ERROR)
 671                                 return SCSI_DH_DEV_TEMP_BUSY;
 672                 } else {
 673                         sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
 674                                     ALUA_DH_NAME);
 675                         scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
 676                 }
 677         }
 678         /* Retry RTPG */
 679         return SCSI_DH_RETRY;
 680 }
 681
 682 static void alua_rtpg_work(struct work_struct *work)
 683 {
 684         struct alua_port_group *pg =
 685                 container_of(work, struct alua_port_group, rtpg_work.work);
 686         struct scsi_device *sdev;
 687         LIST_HEAD(qdata_list);
 688         int err = SCSI_DH_OK;
 689         struct alua_queue_data *qdata, *tmp;
 690         unsigned long flags;
 691         struct workqueue_struct *alua_wq = kaluad_wq;
 692
 693         spin_lock_irqsave(&pg->lock, flags);
 694         sdev = pg->rtpg_sdev;
 695         if (!sdev) {
 696                 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
 697                 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
 698                 spin_unlock_irqrestore(&pg->lock, flags);
 699                 return;
 700         }
 701         if (pg->flags & ALUA_SYNC_STPG)
 702                 alua_wq = kaluad_sync_wq;
 703         pg->flags |= ALUA_PG_RUNNING;
 704         if (pg->flags & ALUA_PG_RUN_RTPG) {
 705                 pg->flags &= ~ALUA_PG_RUN_RTPG;
 706                 spin_unlock_irqrestore(&pg->lock, flags);
 707                 err = alua_rtpg(sdev, pg);
 708                 spin_lock_irqsave(&pg->lock, flags);
 709                 if (err == SCSI_DH_RETRY) {
 710                         pg->flags &= ~ALUA_PG_RUNNING;
 711                         pg->flags |= ALUA_PG_RUN_RTPG;
 712                         spin_unlock_irqrestore(&pg->lock, flags);
 713                         queue_delayed_work(alua_wq, &pg->rtpg_work,
 714                                            pg->interval * HZ);
 715                         return;
 716                 }
 717                 if (err != SCSI_DH_OK)
 718                         pg->flags &= ~ALUA_PG_RUN_STPG;
 719         }
 720         if (pg->flags & ALUA_PG_RUN_STPG) {
 721                 pg->flags &= ~ALUA_PG_RUN_STPG;
 722                 spin_unlock_irqrestore(&pg->lock, flags);
 723                 err = alua_stpg(sdev, pg);
 724                 spin_lock_irqsave(&pg->lock, flags);
 725                 if (err == SCSI_DH_RETRY) {
 726                         pg->flags |= ALUA_PG_RUN_RTPG;
 727                         pg->interval = 0;
 728                         pg->flags &= ~ALUA_PG_RUNNING;
 729                         spin_unlock_irqrestore(&pg->lock, flags);
 730                         queue_delayed_work(alua_wq, &pg->rtpg_work,
 731                                            pg->interval * HZ);
 732                         return;
 733                 }
 734         }
 735
 736         list_splice_init(&pg->rtpg_list, &qdata_list);
 737         pg->rtpg_sdev = NULL;
 738         spin_unlock_irqrestore(&pg->lock, flags);
 739
 740         list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
 741                 list_del(&qdata->entry);
 742                 if (qdata->callback_fn)
 743                         qdata->callback_fn(qdata->callback_data, err);
 744                 kfree(qdata);
 745         }
 746         spin_lock_irqsave(&pg->lock, flags);
 747         pg->flags &= ~ALUA_PG_RUNNING;
 748         spin_unlock_irqrestore(&pg->lock, flags);
 749         scsi_device_put(sdev);
 750         kref_put(&pg->kref, release_port_group);
 751 }
 752
 753 static void alua_rtpg_queue(struct alua_port_group *pg,
 754                             struct scsi_device *sdev,
 755                             struct alua_queue_data *qdata)
 756 {
 757         int start_queue = 0;
 758         unsigned long flags;
 759         struct workqueue_struct *alua_wq = kaluad_wq;
 760
 761         if (!pg)
 762                 return;
 763
 764         spin_lock_irqsave(&pg->lock, flags);
 765         if (qdata) {
 766                 list_add_tail(&qdata->entry, &pg->rtpg_list);
 767                 pg->flags |= ALUA_PG_RUN_STPG;
 768         }
 769         if (pg->rtpg_sdev == NULL) {
 770                 pg->interval = 0;
 771                 pg->flags |= ALUA_PG_RUN_RTPG;
 772                 kref_get(&pg->kref);
 773                 pg->rtpg_sdev = sdev;
 774                 scsi_device_get(sdev);
 775                 start_queue = 1;
 776         }
 777         if (pg->flags & ALUA_SYNC_STPG)
 778                 alua_wq = kaluad_sync_wq;
 779         spin_unlock_irqrestore(&pg->lock, flags);
 780
 781         if (start_queue &&
 782             !queue_delayed_work(alua_wq, &pg->rtpg_work,
 783                                 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS))) {
 784                 scsi_device_put(sdev);
 785                 kref_put(&pg->kref, release_port_group);
 786         }
 787 }
 788
 789 /*
 790  * alua_initialize - Initialize ALUA state
 791  * @sdev: the device to be initialized
 792  *
 793  * For the prep_fn to work correctly we have
 794  * to initialize the ALUA state for the device.
 795  */
 796 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
 797 {
 798         int err = SCSI_DH_DEV_UNSUPP, tpgs;
 799
 800         mutex_lock(&h->init_mutex);
 801         tpgs = alua_check_tpgs(sdev);
 802         if (tpgs != TPGS_MODE_NONE)
 803                 err = alua_check_vpd(sdev, h, tpgs);
 804         h->init_error = err;
 805         mutex_unlock(&h->init_mutex);
 806         return err;
 807 }
 808 /*
 809  * alua_set_params - set/unset the optimize flag
 810  * @sdev: device on the path to be activated
 811  * params - parameters in the following format
 812  *      "no_of_params\0param1\0param2\0param3\0...\0"
 813  * For example, to set the flag pass the following parameters
 814  * from multipath.conf
 815  *     hardware_handler        "2 alua 1"
 816  */
 817 static int alua_set_params(struct scsi_device *sdev, const char *params)
 818 {
 819         struct alua_dh_data *h = sdev->handler_data;
 820         struct alua_port_group __rcu *pg = NULL;
 821         unsigned int optimize = 0, argc;
 822         const char *p = params;
 823         int result = SCSI_DH_OK;
 824         unsigned long flags;
 825
 826         if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
 827                 return -EINVAL;
 828
 829         while (*p++)
 830                 ;
 831         if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
 832                 return -EINVAL;
 833
 834         rcu_read_lock();
 835         pg = rcu_dereference(h->pg);
 836         if (!pg) {
 837                 rcu_read_unlock();
 838                 return -ENXIO;
 839         }
 840         spin_lock_irqsave(&pg->lock, flags);
 841         if (optimize)
 842                 pg->flags |= ALUA_OPTIMIZE_STPG;
 843         else
 844                 pg->flags &= ~ALUA_OPTIMIZE_STPG;
 845         spin_unlock_irqrestore(&pg->lock, flags);
 846         rcu_read_unlock();
 847
 848         return result;
 849 }
 850
 851 /*
 852  * alua_activate - activate a path
 853  * @sdev: device on the path to be activated
 854  *
 855  * We're currently switching the port group to be activated only and
 856  * let the array figure out the rest.
 857  * There may be other arrays which require us to switch all port groups
 858  * based on a certain policy. But until we actually encounter them it
 859  * should be okay.
 860  */
 861 static int alua_activate(struct scsi_device *sdev,
 862                         activate_complete fn, void *data)
 863 {
 864         struct alua_dh_data *h = sdev->handler_data;
 865         int err = SCSI_DH_OK;
 866         struct alua_queue_data *qdata;
 867         struct alua_port_group __rcu *pg;
 868
 869         qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
 870         if (!qdata) {
 871                 err = SCSI_DH_RES_TEMP_UNAVAIL;
 872                 goto out;
 873         }
 874         qdata->callback_fn = fn;
 875         qdata->callback_data = data;
 876
 877         mutex_lock(&h->init_mutex);
 878         rcu_read_lock();
 879         pg = rcu_dereference(h->pg);
 880         if (!pg || !kref_get_unless_zero(&pg->kref)) {
 881                 rcu_read_unlock();
 882                 kfree(qdata);
 883                 err = h->init_error;
 884                 mutex_unlock(&h->init_mutex);
 885                 goto out;
 886         }
 887         fn = NULL;
 888         rcu_read_unlock();
 889         mutex_unlock(&h->init_mutex);
 890
 891         alua_rtpg_queue(pg, sdev, qdata);
 892         kref_put(&pg->kref, release_port_group);
 893 out:
 894         if (fn)
 895                 fn(data, err);
 896         return 0;
 897 }
 898
 899 /*
 900  * alua_prep_fn - request callback
 901  *
 902  * Fail I/O to all paths not in state
 903  * active/optimized or active/non-optimized.
 904  */
 905 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
 906 {
 907         struct alua_dh_data *h = sdev->handler_data;
 908         struct alua_port_group __rcu *pg;
 909         int state = TPGS_STATE_OPTIMIZED;
 910         int ret = BLKPREP_OK;
 911
 912         rcu_read_lock();
 913         pg = rcu_dereference(h->pg);
 914         if (pg)
 915                 state = pg->state;
 916         rcu_read_unlock();
 917         if (state == TPGS_STATE_TRANSITIONING)
 918                 ret = BLKPREP_DEFER;
 919         else if (state != TPGS_STATE_OPTIMIZED &&
 920                  state != TPGS_STATE_NONOPTIMIZED &&
 921                  state != TPGS_STATE_LBA_DEPENDENT) {
 922                 ret = BLKPREP_KILL;
 923                 req->cmd_flags |= REQ_QUIET;
 924         }
 925         return ret;
 926
 927 }
 928
 929 /*
 930  * alua_bus_attach - Attach device handler
 931  * @sdev: device to be attached to
 932  */
 933 static int alua_bus_attach(struct scsi_device *sdev)
 934 {
 935         struct alua_dh_data *h;
 936         int err, ret = -EINVAL;
 937
 938         h = kzalloc(sizeof(*h) , GFP_KERNEL);
 939         if (!h)
 940                 return -ENOMEM;
 941         spin_lock_init(&h->pg_lock);
 942         rcu_assign_pointer(h->pg, NULL);
 943         h->init_error = SCSI_DH_OK;
 944         h->sdev = sdev;
 945
 946         mutex_init(&h->init_mutex);
 947         err = alua_initialize(sdev, h);
 948         if (err == SCSI_DH_NOMEM)
 949                 ret = -ENOMEM;
 950         if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
 951                 goto failed;
 952
 953         sdev->handler_data = h;
 954         return 0;
 955 failed:
 956         kfree(h);
 957         return ret;
 958 }
 959
 960 /*
 961  * alua_bus_detach - Detach device handler
 962  * @sdev: device to be detached from
 963  */
 964 static void alua_bus_detach(struct scsi_device *sdev)
 965 {
 966         struct alua_dh_data *h = sdev->handler_data;
 967         struct alua_port_group *pg;
 968
 969         spin_lock(&h->pg_lock);
 970         pg = h->pg;
 971         rcu_assign_pointer(h->pg, NULL);
 972         h->sdev = NULL;
 973         spin_unlock(&h->pg_lock);
 974         if (pg)
 975                 kref_put(&pg->kref, release_port_group);
 976
 977         sdev->handler_data = NULL;
 978         kfree(h);
 979 }
 980
 981 static struct scsi_device_handler alua_dh = {
 982         .name = ALUA_DH_NAME,
 983         .module = THIS_MODULE,
 984         .attach = alua_bus_attach,
 985         .detach = alua_bus_detach,
 986         .prep_fn = alua_prep_fn,
 987         .check_sense = alua_check_sense,
 988         .activate = alua_activate,
 989         .set_params = alua_set_params,
 990 };
 991
 992 static int __init alua_init(void)
 993 {
 994         int r;
 995
 996         kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
 997         if (!kaluad_wq) {
 998                 /* Temporary failure, bypass */
 999                 return SCSI_DH_DEV_TEMP_BUSY;
1000         }
1001         kaluad_sync_wq = create_workqueue("kaluad_sync");
1002         if (!kaluad_sync_wq) {
1003                 destroy_workqueue(kaluad_wq);
1004                 return SCSI_DH_DEV_TEMP_BUSY;
1005         }
1006         r = scsi_register_device_handler(&alua_dh);
1007         if (r != 0) {
1008                 printk(KERN_ERR "%s: Failed to register scsi device handler",
1009                         ALUA_DH_NAME);
1010                 destroy_workqueue(kaluad_sync_wq);
1011                 destroy_workqueue(kaluad_wq);
1012         }
1013         return r;
1014 }
1015
1016 static void __exit alua_exit(void)
1017 {
1018         scsi_unregister_device_handler(&alua_dh);
1019         destroy_workqueue(kaluad_sync_wq);
1020         destroy_workqueue(kaluad_wq);
1021 }
1022
1023 module_init(alua_init);
1024 module_exit(alua_exit);
1025
1026 MODULE_DESCRIPTION("DM Multipath ALUA support");
1027 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1028 MODULE_LICENSE("GPL");
1029 MODULE_VERSION(ALUA_DH_VER);