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 <scsi/scsi.h>
  26 #include <scsi/scsi_eh.h>
  27 #include <scsi/scsi_dh.h>
  28
  29 #define ALUA_DH_NAME "alua"
  30 #define ALUA_DH_VER "1.3"
  31
  32 #define TPGS_STATE_OPTIMIZED            0x0
  33 #define TPGS_STATE_NONOPTIMIZED         0x1
  34 #define TPGS_STATE_STANDBY              0x2
  35 #define TPGS_STATE_UNAVAILABLE          0x3
  36 #define TPGS_STATE_LBA_DEPENDENT        0x4
  37 #define TPGS_STATE_OFFLINE              0xe
  38 #define TPGS_STATE_TRANSITIONING        0xf
  39
  40 #define TPGS_SUPPORT_NONE               0x00
  41 #define TPGS_SUPPORT_OPTIMIZED          0x01
  42 #define TPGS_SUPPORT_NONOPTIMIZED       0x02
  43 #define TPGS_SUPPORT_STANDBY            0x04
  44 #define TPGS_SUPPORT_UNAVAILABLE        0x08
  45 #define TPGS_SUPPORT_LBA_DEPENDENT      0x10
  46 #define TPGS_SUPPORT_OFFLINE            0x40
  47 #define TPGS_SUPPORT_TRANSITION         0x80
  48
  49 #define RTPG_FMT_MASK                   0x70
  50 #define RTPG_FMT_EXT_HDR                0x10
  51
  52 #define TPGS_MODE_UNINITIALIZED          -1
  53 #define TPGS_MODE_NONE                  0x0
  54 #define TPGS_MODE_IMPLICIT              0x1
  55 #define TPGS_MODE_EXPLICIT              0x2
  56
  57 #define ALUA_INQUIRY_SIZE               36
  58 #define ALUA_FAILOVER_TIMEOUT           60
  59 #define ALUA_FAILOVER_RETRIES           5
  60
  61 /* flags passed from user level */
  62 #define ALUA_OPTIMIZE_STPG              1
  63
  64 struct alua_dh_data {
  65         int                     group_id;
  66         int                     rel_port;
  67         int                     tpgs;
  68         int                     state;
  69         int                     pref;
  70         unsigned                flags; /* used for optimizing STPG */
  71         unsigned char           inq[ALUA_INQUIRY_SIZE];
  72         unsigned char           *buff;
  73         int                     bufflen;
  74         unsigned char           transition_tmo;
  75         unsigned char           sense[SCSI_SENSE_BUFFERSIZE];
  76         int                     senselen;
  77         struct scsi_device      *sdev;
  78         activate_complete       callback_fn;
  79         void                    *callback_data;
  80 };
  81
  82 #define ALUA_POLICY_SWITCH_CURRENT      0
  83 #define ALUA_POLICY_SWITCH_ALL          1
  84
  85 static char print_alua_state(int);
  86 static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
  87
  88 static int realloc_buffer(struct alua_dh_data *h, unsigned len)
  89 {
  90         if (h->buff && h->buff != h->inq)
  91                 kfree(h->buff);
  92
  93         h->buff = kmalloc(len, GFP_NOIO);
  94         if (!h->buff) {
  95                 h->buff = h->inq;
  96                 h->bufflen = ALUA_INQUIRY_SIZE;
  97                 return 1;
  98         }
  99         h->bufflen = len;
 100         return 0;
 101 }
 102
 103 static struct request *get_alua_req(struct scsi_device *sdev,
 104                                     void *buffer, unsigned buflen, int rw)
 105 {
 106         struct request *rq;
 107         struct request_queue *q = sdev->request_queue;
 108
 109         rq = blk_get_request(q, rw, GFP_NOIO);
 110
 111         if (IS_ERR(rq)) {
 112                 sdev_printk(KERN_INFO, sdev,
 113                             "%s: blk_get_request failed\n", __func__);
 114                 return NULL;
 115         }
 116         blk_rq_set_block_pc(rq);
 117
 118         if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
 119                 blk_put_request(rq);
 120                 sdev_printk(KERN_INFO, sdev,
 121                             "%s: blk_rq_map_kern failed\n", __func__);
 122                 return NULL;
 123         }
 124
 125         rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
 126                          REQ_FAILFAST_DRIVER;
 127         rq->retries = ALUA_FAILOVER_RETRIES;
 128         rq->timeout = ALUA_FAILOVER_TIMEOUT * HZ;
 129
 130         return rq;
 131 }
 132
 133 /*
 134  * submit_vpd_inquiry - Issue an INQUIRY VPD page 0x83 command
 135  * @sdev: sdev the command should be sent to
 136  */
 137 static int submit_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
 138 {
 139         struct request *rq;
 140         int err = SCSI_DH_RES_TEMP_UNAVAIL;
 141
 142         rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
 143         if (!rq)
 144                 goto done;
 145
 146         /* Prepare the command. */
 147         rq->cmd[0] = INQUIRY;
 148         rq->cmd[1] = 1;
 149         rq->cmd[2] = 0x83;
 150         rq->cmd[4] = h->bufflen;
 151         rq->cmd_len = COMMAND_SIZE(INQUIRY);
 152
 153         rq->sense = h->sense;
 154         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 155         rq->sense_len = h->senselen = 0;
 156
 157         err = blk_execute_rq(rq->q, NULL, rq, 1);
 158         if (err == -EIO) {
 159                 sdev_printk(KERN_INFO, sdev,
 160                             "%s: evpd inquiry failed with %x\n",
 161                             ALUA_DH_NAME, rq->errors);
 162                 h->senselen = rq->sense_len;
 163                 err = SCSI_DH_IO;
 164         }
 165         blk_put_request(rq);
 166 done:
 167         return err;
 168 }
 169
 170 /*
 171  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
 172  * @sdev: sdev the command should be sent to
 173  */
 174 static unsigned submit_rtpg(struct scsi_device *sdev, struct alua_dh_data *h,
 175                             bool rtpg_ext_hdr_req)
 176 {
 177         struct request *rq;
 178         int err = SCSI_DH_RES_TEMP_UNAVAIL;
 179
 180         rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
 181         if (!rq)
 182                 goto done;
 183
 184         /* Prepare the command. */
 185         rq->cmd[0] = MAINTENANCE_IN;
 186         if (rtpg_ext_hdr_req)
 187                 rq->cmd[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
 188         else
 189                 rq->cmd[1] = MI_REPORT_TARGET_PGS;
 190         rq->cmd[6] = (h->bufflen >> 24) & 0xff;
 191         rq->cmd[7] = (h->bufflen >> 16) & 0xff;
 192         rq->cmd[8] = (h->bufflen >>  8) & 0xff;
 193         rq->cmd[9] = h->bufflen & 0xff;
 194         rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
 195
 196         rq->sense = h->sense;
 197         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 198         rq->sense_len = h->senselen = 0;
 199
 200         err = blk_execute_rq(rq->q, NULL, rq, 1);
 201         if (err == -EIO) {
 202                 sdev_printk(KERN_INFO, sdev,
 203                             "%s: rtpg failed with %x\n",
 204                             ALUA_DH_NAME, rq->errors);
 205                 h->senselen = rq->sense_len;
 206                 err = SCSI_DH_IO;
 207         }
 208         blk_put_request(rq);
 209 done:
 210         return err;
 211 }
 212
 213 /*
 214  * alua_stpg - Evaluate SET TARGET GROUP STATES
 215  * @sdev: the device to be evaluated
 216  * @state: the new target group state
 217  *
 218  * Send a SET TARGET GROUP STATES command to the device.
 219  * We only have to test here if we should resubmit the command;
 220  * any other error is assumed as a failure.
 221  */
 222 static void stpg_endio(struct request *req, int error)
 223 {
 224         struct alua_dh_data *h = req->end_io_data;
 225         struct scsi_sense_hdr sense_hdr;
 226         unsigned err = SCSI_DH_OK;
 227
 228         if (host_byte(req->errors) != DID_OK ||
 229             msg_byte(req->errors) != COMMAND_COMPLETE) {
 230                 err = SCSI_DH_IO;
 231                 goto done;
 232         }
 233
 234         if (req->sense_len > 0) {
 235                 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
 236                                            &sense_hdr);
 237                 if (!err) {
 238                         err = SCSI_DH_IO;
 239                         goto done;
 240                 }
 241                 err = alua_check_sense(h->sdev, &sense_hdr);
 242                 if (err == ADD_TO_MLQUEUE) {
 243                         err = SCSI_DH_RETRY;
 244                         goto done;
 245                 }
 246                 sdev_printk(KERN_INFO, h->sdev,
 247                             "%s: stpg sense code: %02x/%02x/%02x\n",
 248                             ALUA_DH_NAME, sense_hdr.sense_key,
 249                             sense_hdr.asc, sense_hdr.ascq);
 250                 err = SCSI_DH_IO;
 251         } else if (error)
 252                 err = SCSI_DH_IO;
 253
 254         if (err == SCSI_DH_OK) {
 255                 h->state = TPGS_STATE_OPTIMIZED;
 256                 sdev_printk(KERN_INFO, h->sdev,
 257                             "%s: port group %02x switched to state %c\n",
 258                             ALUA_DH_NAME, h->group_id,
 259                             print_alua_state(h->state));
 260         }
 261 done:
 262         req->end_io_data = NULL;
 263         __blk_put_request(req->q, req);
 264         if (h->callback_fn) {
 265                 h->callback_fn(h->callback_data, err);
 266                 h->callback_fn = h->callback_data = NULL;
 267         }
 268         return;
 269 }
 270
 271 /*
 272  * submit_stpg - Issue a SET TARGET GROUP STATES command
 273  *
 274  * Currently we're only setting the current target port group state
 275  * to 'active/optimized' and let the array firmware figure out
 276  * the states of the remaining groups.
 277  */
 278 static unsigned submit_stpg(struct alua_dh_data *h)
 279 {
 280         struct request *rq;
 281         int stpg_len = 8;
 282         struct scsi_device *sdev = h->sdev;
 283
 284         /* Prepare the data buffer */
 285         memset(h->buff, 0, stpg_len);
 286         h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
 287         h->buff[6] = (h->group_id >> 8) & 0xff;
 288         h->buff[7] = h->group_id & 0xff;
 289
 290         rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
 291         if (!rq)
 292                 return SCSI_DH_RES_TEMP_UNAVAIL;
 293
 294         /* Prepare the command. */
 295         rq->cmd[0] = MAINTENANCE_OUT;
 296         rq->cmd[1] = MO_SET_TARGET_PGS;
 297         rq->cmd[6] = (stpg_len >> 24) & 0xff;
 298         rq->cmd[7] = (stpg_len >> 16) & 0xff;
 299         rq->cmd[8] = (stpg_len >>  8) & 0xff;
 300         rq->cmd[9] = stpg_len & 0xff;
 301         rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
 302
 303         rq->sense = h->sense;
 304         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 305         rq->sense_len = h->senselen = 0;
 306         rq->end_io_data = h;
 307
 308         blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
 309         return SCSI_DH_OK;
 310 }
 311
 312 /*
 313  * alua_check_tpgs - Evaluate TPGS setting
 314  * @sdev: device to be checked
 315  *
 316  * Examine the TPGS setting of the sdev to find out if ALUA
 317  * is supported.
 318  */
 319 static int alua_check_tpgs(struct scsi_device *sdev, struct alua_dh_data *h)
 320 {
 321         int err = SCSI_DH_OK;
 322
 323         h->tpgs = scsi_device_tpgs(sdev);
 324         switch (h->tpgs) {
 325         case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
 326                 sdev_printk(KERN_INFO, sdev,
 327                             "%s: supports implicit and explicit TPGS\n",
 328                             ALUA_DH_NAME);
 329                 break;
 330         case TPGS_MODE_EXPLICIT:
 331                 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
 332                             ALUA_DH_NAME);
 333                 break;
 334         case TPGS_MODE_IMPLICIT:
 335                 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
 336                             ALUA_DH_NAME);
 337                 break;
 338         default:
 339                 h->tpgs = TPGS_MODE_NONE;
 340                 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
 341                             ALUA_DH_NAME);
 342                 err = SCSI_DH_DEV_UNSUPP;
 343                 break;
 344         }
 345
 346         return err;
 347 }
 348
 349 /*
 350  * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
 351  * @sdev: device to be checked
 352  *
 353  * Extract the relative target port and the target port group
 354  * descriptor from the list of identificators.
 355  */
 356 static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
 357 {
 358         int len;
 359         unsigned err;
 360         unsigned char *d;
 361
 362  retry:
 363         err = submit_vpd_inquiry(sdev, h);
 364
 365         if (err != SCSI_DH_OK)
 366                 return err;
 367
 368         /* Check if vpd page exceeds initial buffer */
 369         len = (h->buff[2] << 8) + h->buff[3] + 4;
 370         if (len > h->bufflen) {
 371                 /* Resubmit with the correct length */
 372                 if (realloc_buffer(h, len)) {
 373                         sdev_printk(KERN_WARNING, sdev,
 374                                     "%s: kmalloc buffer failed\n",
 375                                     ALUA_DH_NAME);
 376                         /* Temporary failure, bypass */
 377                         return SCSI_DH_DEV_TEMP_BUSY;
 378                 }
 379                 goto retry;
 380         }
 381
 382         /*
 383          * Now look for the correct descriptor.
 384          */
 385         d = h->buff + 4;
 386         while (d < h->buff + len) {
 387                 switch (d[1] & 0xf) {
 388                 case 0x4:
 389                         /* Relative target port */
 390                         h->rel_port = (d[6] << 8) + d[7];
 391                         break;
 392                 case 0x5:
 393                         /* Target port group */
 394                         h->group_id = (d[6] << 8) + d[7];
 395                         break;
 396                 default:
 397                         break;
 398                 }
 399                 d += d[3] + 4;
 400         }
 401
 402         if (h->group_id == -1) {
 403                 /*
 404                  * Internal error; TPGS supported but required
 405                  * VPD identification descriptors not present.
 406                  * Disable ALUA support
 407                  */
 408                 sdev_printk(KERN_INFO, sdev,
 409                             "%s: No target port descriptors found\n",
 410                             ALUA_DH_NAME);
 411                 h->state = TPGS_STATE_OPTIMIZED;
 412                 h->tpgs = TPGS_MODE_NONE;
 413                 err = SCSI_DH_DEV_UNSUPP;
 414         } else {
 415                 sdev_printk(KERN_INFO, sdev,
 416                             "%s: port group %02x rel port %02x\n",
 417                             ALUA_DH_NAME, h->group_id, h->rel_port);
 418         }
 419
 420         return err;
 421 }
 422
 423 static char print_alua_state(int state)
 424 {
 425         switch (state) {
 426         case TPGS_STATE_OPTIMIZED:
 427                 return 'A';
 428         case TPGS_STATE_NONOPTIMIZED:
 429                 return 'N';
 430         case TPGS_STATE_STANDBY:
 431                 return 'S';
 432         case TPGS_STATE_UNAVAILABLE:
 433                 return 'U';
 434         case TPGS_STATE_LBA_DEPENDENT:
 435                 return 'L';
 436         case TPGS_STATE_OFFLINE:
 437                 return 'O';
 438         case TPGS_STATE_TRANSITIONING:
 439                 return 'T';
 440         default:
 441                 return 'X';
 442         }
 443 }
 444
 445 static int alua_check_sense(struct scsi_device *sdev,
 446                             struct scsi_sense_hdr *sense_hdr)
 447 {
 448         switch (sense_hdr->sense_key) {
 449         case NOT_READY:
 450                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
 451                         /*
 452                          * LUN Not Accessible - ALUA state transition
 453                          */
 454                         return ADD_TO_MLQUEUE;
 455                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
 456                         /*
 457                          * LUN Not Accessible -- Target port in standby state
 458                          */
 459                         return SUCCESS;
 460                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
 461                         /*
 462                          * LUN Not Accessible -- Target port in unavailable state
 463                          */
 464                         return SUCCESS;
 465                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
 466                         /*
 467                          * LUN Not Ready -- Offline
 468                          */
 469                         return SUCCESS;
 470                 if (sdev->allow_restart &&
 471                     sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x02)
 472                         /*
 473                          * if the device is not started, we need to wake
 474                          * the error handler to start the motor
 475                          */
 476                         return FAILED;
 477                 break;
 478         case UNIT_ATTENTION:
 479                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
 480                         /*
 481                          * Power On, Reset, or Bus Device Reset, just retry.
 482                          */
 483                         return ADD_TO_MLQUEUE;
 484                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
 485                         /*
 486                          * Device internal reset
 487                          */
 488                         return ADD_TO_MLQUEUE;
 489                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
 490                         /*
 491                          * Mode Parameters Changed
 492                          */
 493                         return ADD_TO_MLQUEUE;
 494                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06)
 495                         /*
 496                          * ALUA state changed
 497                          */
 498                         return ADD_TO_MLQUEUE;
 499                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07)
 500                         /*
 501                          * Implicit ALUA state transition failed
 502                          */
 503                         return ADD_TO_MLQUEUE;
 504                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
 505                         /*
 506                          * Inquiry data has changed
 507                          */
 508                         return ADD_TO_MLQUEUE;
 509                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
 510                         /*
 511                          * REPORTED_LUNS_DATA_HAS_CHANGED is reported
 512                          * when switching controllers on targets like
 513                          * Intel Multi-Flex. We can just retry.
 514                          */
 515                         return ADD_TO_MLQUEUE;
 516                 break;
 517         }
 518
 519         return SCSI_RETURN_NOT_HANDLED;
 520 }
 521
 522 /*
 523  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
 524  * @sdev: the device to be evaluated.
 525  * @wait_for_transition: if nonzero, wait ALUA_FAILOVER_TIMEOUT seconds for device to exit transitioning state
 526  *
 527  * Evaluate the Target Port Group State.
 528  * Returns SCSI_DH_DEV_OFFLINED if the path is
 529  * found to be unusable.
 530  */
 531 static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h, int wait_for_transition)
 532 {
 533         struct scsi_sense_hdr sense_hdr;
 534         int len, k, off, valid_states = 0;
 535         unsigned char *ucp;
 536         unsigned err;
 537         bool rtpg_ext_hdr_req = 1;
 538         unsigned long expiry, interval = 0;
 539         unsigned int tpg_desc_tbl_off;
 540         unsigned char orig_transition_tmo;
 541
 542         if (!h->transition_tmo)
 543                 expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT * HZ);
 544         else
 545                 expiry = round_jiffies_up(jiffies + h->transition_tmo * HZ);
 546
 547  retry:
 548         err = submit_rtpg(sdev, h, rtpg_ext_hdr_req);
 549
 550         if (err == SCSI_DH_IO && h->senselen > 0) {
 551                 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
 552                                            &sense_hdr);
 553                 if (!err)
 554                         return SCSI_DH_IO;
 555
 556                 /*
 557                  * submit_rtpg() has failed on existing arrays
 558                  * when requesting extended header info, and
 559                  * the array doesn't support extended headers,
 560                  * even though it shouldn't according to T10.
 561                  * The retry without rtpg_ext_hdr_req set
 562                  * handles this.
 563                  */
 564                 if (rtpg_ext_hdr_req == 1 &&
 565                     sense_hdr.sense_key == ILLEGAL_REQUEST &&
 566                     sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
 567                         rtpg_ext_hdr_req = 0;
 568                         goto retry;
 569                 }
 570
 571                 err = alua_check_sense(sdev, &sense_hdr);
 572                 if (err == ADD_TO_MLQUEUE && time_before(jiffies, expiry))
 573                         goto retry;
 574                 sdev_printk(KERN_INFO, sdev,
 575                             "%s: rtpg sense code %02x/%02x/%02x\n",
 576                             ALUA_DH_NAME, sense_hdr.sense_key,
 577                             sense_hdr.asc, sense_hdr.ascq);
 578                 err = SCSI_DH_IO;
 579         }
 580         if (err != SCSI_DH_OK)
 581                 return err;
 582
 583         len = (h->buff[0] << 24) + (h->buff[1] << 16) +
 584                 (h->buff[2] << 8) + h->buff[3] + 4;
 585
 586         if (len > h->bufflen) {
 587                 /* Resubmit with the correct length */
 588                 if (realloc_buffer(h, len)) {
 589                         sdev_printk(KERN_WARNING, sdev,
 590                                     "%s: kmalloc buffer failed\n",__func__);
 591                         /* Temporary failure, bypass */
 592                         return SCSI_DH_DEV_TEMP_BUSY;
 593                 }
 594                 goto retry;
 595         }
 596
 597         orig_transition_tmo = h->transition_tmo;
 598         if ((h->buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && h->buff[5] != 0)
 599                 h->transition_tmo = h->buff[5];
 600         else
 601                 h->transition_tmo = ALUA_FAILOVER_TIMEOUT;
 602
 603         if (wait_for_transition && (orig_transition_tmo != h->transition_tmo)) {
 604                 sdev_printk(KERN_INFO, sdev,
 605                             "%s: transition timeout set to %d seconds\n",
 606                             ALUA_DH_NAME, h->transition_tmo);
 607                 expiry = jiffies + h->transition_tmo * HZ;
 608         }
 609
 610         if ((h->buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
 611                 tpg_desc_tbl_off = 8;
 612         else
 613                 tpg_desc_tbl_off = 4;
 614
 615         for (k = tpg_desc_tbl_off, ucp = h->buff + tpg_desc_tbl_off;
 616              k < len;
 617              k += off, ucp += off) {
 618
 619                 if (h->group_id == (ucp[2] << 8) + ucp[3]) {
 620                         h->state = ucp[0] & 0x0f;
 621                         h->pref = ucp[0] >> 7;
 622                         valid_states = ucp[1];
 623                 }
 624                 off = 8 + (ucp[7] * 4);
 625         }
 626
 627         sdev_printk(KERN_INFO, sdev,
 628                     "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
 629                     ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
 630                     h->pref ? "preferred" : "non-preferred",
 631                     valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
 632                     valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
 633                     valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
 634                     valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
 635                     valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
 636                     valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
 637                     valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
 638
 639         switch (h->state) {
 640         case TPGS_STATE_TRANSITIONING:
 641                 if (wait_for_transition) {
 642                         if (time_before(jiffies, expiry)) {
 643                                 /* State transition, retry */
 644                                 interval += 2000;
 645                                 msleep(interval);
 646                                 goto retry;
 647                         }
 648                         err = SCSI_DH_RETRY;
 649                 } else {
 650                         err = SCSI_DH_OK;
 651                 }
 652
 653                 /* Transitioning time exceeded, set port to standby */
 654                 h->state = TPGS_STATE_STANDBY;
 655                 break;
 656         case TPGS_STATE_OFFLINE:
 657                 /* Path unusable */
 658                 err = SCSI_DH_DEV_OFFLINED;
 659                 break;
 660         default:
 661                 /* Useable path if active */
 662                 err = SCSI_DH_OK;
 663                 break;
 664         }
 665         return err;
 666 }
 667
 668 /*
 669  * alua_initialize - Initialize ALUA state
 670  * @sdev: the device to be initialized
 671  *
 672  * For the prep_fn to work correctly we have
 673  * to initialize the ALUA state for the device.
 674  */
 675 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
 676 {
 677         int err;
 678
 679         err = alua_check_tpgs(sdev, h);
 680         if (err != SCSI_DH_OK)
 681                 goto out;
 682
 683         err = alua_vpd_inquiry(sdev, h);
 684         if (err != SCSI_DH_OK)
 685                 goto out;
 686
 687         err = alua_rtpg(sdev, h, 0);
 688         if (err != SCSI_DH_OK)
 689                 goto out;
 690
 691 out:
 692         return err;
 693 }
 694 /*
 695  * alua_set_params - set/unset the optimize flag
 696  * @sdev: device on the path to be activated
 697  * params - parameters in the following format
 698  *      "no_of_params\0param1\0param2\0param3\0...\0"
 699  * For example, to set the flag pass the following parameters
 700  * from multipath.conf
 701  *     hardware_handler        "2 alua 1"
 702  */
 703 static int alua_set_params(struct scsi_device *sdev, const char *params)
 704 {
 705         struct alua_dh_data *h = sdev->handler_data;
 706         unsigned int optimize = 0, argc;
 707         const char *p = params;
 708         int result = SCSI_DH_OK;
 709
 710         if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
 711                 return -EINVAL;
 712
 713         while (*p++)
 714                 ;
 715         if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
 716                 return -EINVAL;
 717
 718         if (optimize)
 719                 h->flags |= ALUA_OPTIMIZE_STPG;
 720         else
 721                 h->flags &= ~ALUA_OPTIMIZE_STPG;
 722
 723         return result;
 724 }
 725
 726 static uint optimize_stpg;
 727 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
 728 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.");
 729
 730 /*
 731  * alua_activate - activate a path
 732  * @sdev: device on the path to be activated
 733  *
 734  * We're currently switching the port group to be activated only and
 735  * let the array figure out the rest.
 736  * There may be other arrays which require us to switch all port groups
 737  * based on a certain policy. But until we actually encounter them it
 738  * should be okay.
 739  */
 740 static int alua_activate(struct scsi_device *sdev,
 741                         activate_complete fn, void *data)
 742 {
 743         struct alua_dh_data *h = sdev->handler_data;
 744         int err = SCSI_DH_OK;
 745         int stpg = 0;
 746
 747         err = alua_rtpg(sdev, h, 1);
 748         if (err != SCSI_DH_OK)
 749                 goto out;
 750
 751         if (optimize_stpg)
 752                 h->flags |= ALUA_OPTIMIZE_STPG;
 753
 754         if (h->tpgs & TPGS_MODE_EXPLICIT) {
 755                 switch (h->state) {
 756                 case TPGS_STATE_NONOPTIMIZED:
 757                         stpg = 1;
 758                         if ((h->flags & ALUA_OPTIMIZE_STPG) &&
 759                             (!h->pref) &&
 760                             (h->tpgs & TPGS_MODE_IMPLICIT))
 761                                 stpg = 0;
 762                         break;
 763                 case TPGS_STATE_STANDBY:
 764                 case TPGS_STATE_UNAVAILABLE:
 765                         stpg = 1;
 766                         break;
 767                 case TPGS_STATE_OFFLINE:
 768                         err = SCSI_DH_IO;
 769                         break;
 770                 case TPGS_STATE_TRANSITIONING:
 771                         err = SCSI_DH_RETRY;
 772                         break;
 773                 default:
 774                         break;
 775                 }
 776         }
 777
 778         if (stpg) {
 779                 h->callback_fn = fn;
 780                 h->callback_data = data;
 781                 err = submit_stpg(h);
 782                 if (err == SCSI_DH_OK)
 783                         return 0;
 784                 h->callback_fn = h->callback_data = NULL;
 785         }
 786
 787 out:
 788         if (fn)
 789                 fn(data, err);
 790         return 0;
 791 }
 792
 793 /*
 794  * alua_prep_fn - request callback
 795  *
 796  * Fail I/O to all paths not in state
 797  * active/optimized or active/non-optimized.
 798  */
 799 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
 800 {
 801         struct alua_dh_data *h = sdev->handler_data;
 802         int ret = BLKPREP_OK;
 803
 804         if (h->state == TPGS_STATE_TRANSITIONING)
 805                 ret = BLKPREP_DEFER;
 806         else if (h->state != TPGS_STATE_OPTIMIZED &&
 807                  h->state != TPGS_STATE_NONOPTIMIZED &&
 808                  h->state != TPGS_STATE_LBA_DEPENDENT) {
 809                 ret = BLKPREP_KILL;
 810                 req->cmd_flags |= REQ_QUIET;
 811         }
 812         return ret;
 813
 814 }
 815
 816 /*
 817  * alua_bus_attach - Attach device handler
 818  * @sdev: device to be attached to
 819  */
 820 static int alua_bus_attach(struct scsi_device *sdev)
 821 {
 822         struct alua_dh_data *h;
 823         int err;
 824
 825         h = kzalloc(sizeof(*h) , GFP_KERNEL);
 826         if (!h)
 827                 return -ENOMEM;
 828         h->tpgs = TPGS_MODE_UNINITIALIZED;
 829         h->state = TPGS_STATE_OPTIMIZED;
 830         h->group_id = -1;
 831         h->rel_port = -1;
 832         h->buff = h->inq;
 833         h->bufflen = ALUA_INQUIRY_SIZE;
 834         h->sdev = sdev;
 835
 836         err = alua_initialize(sdev, h);
 837         if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
 838                 goto failed;
 839
 840         sdev->handler_data = h;
 841         return 0;
 842 failed:
 843         kfree(h);
 844         return -EINVAL;
 845 }
 846
 847 /*
 848  * alua_bus_detach - Detach device handler
 849  * @sdev: device to be detached from
 850  */
 851 static void alua_bus_detach(struct scsi_device *sdev)
 852 {
 853         struct alua_dh_data *h = sdev->handler_data;
 854
 855         if (h->buff && h->inq != h->buff)
 856                 kfree(h->buff);
 857         sdev->handler_data = NULL;
 858         kfree(h);
 859 }
 860
 861 static struct scsi_device_handler alua_dh = {
 862         .name = ALUA_DH_NAME,
 863         .module = THIS_MODULE,
 864         .attach = alua_bus_attach,
 865         .detach = alua_bus_detach,
 866         .prep_fn = alua_prep_fn,
 867         .check_sense = alua_check_sense,
 868         .activate = alua_activate,
 869         .set_params = alua_set_params,
 870 };
 871
 872 static int __init alua_init(void)
 873 {
 874         int r;
 875
 876         r = scsi_register_device_handler(&alua_dh);
 877         if (r != 0)
 878                 printk(KERN_ERR "%s: Failed to register scsi device handler",
 879                         ALUA_DH_NAME);
 880         return r;
 881 }
 882
 883 static void __exit alua_exit(void)
 884 {
 885         scsi_unregister_device_handler(&alua_dh);
 886 }
 887
 888 module_init(alua_init);
 889 module_exit(alua_exit);
 890
 891 MODULE_DESCRIPTION("DM Multipath ALUA support");
 892 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
 893 MODULE_LICENSE("GPL");
 894 MODULE_VERSION(ALUA_DH_VER);