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 TPGS_MODE_UNINITIALIZED          -1
  50 #define TPGS_MODE_NONE                  0x0
  51 #define TPGS_MODE_IMPLICIT              0x1
  52 #define TPGS_MODE_EXPLICIT              0x2
  53
  54 #define ALUA_INQUIRY_SIZE               36
  55 #define ALUA_FAILOVER_TIMEOUT           (60 * HZ)
  56 #define ALUA_FAILOVER_RETRIES           5
  57
  58 /* flags passed from user level */
  59 #define ALUA_OPTIMIZE_STPG              1
  60
  61 struct alua_dh_data {
  62         int                     group_id;
  63         int                     rel_port;
  64         int                     tpgs;
  65         int                     state;
  66         int                     pref;
  67         unsigned                flags; /* used for optimizing STPG */
  68         unsigned char           inq[ALUA_INQUIRY_SIZE];
  69         unsigned char           *buff;
  70         int                     bufflen;
  71         unsigned char           sense[SCSI_SENSE_BUFFERSIZE];
  72         int                     senselen;
  73         struct scsi_device      *sdev;
  74         activate_complete       callback_fn;
  75         void                    *callback_data;
  76 };
  77
  78 #define ALUA_POLICY_SWITCH_CURRENT      0
  79 #define ALUA_POLICY_SWITCH_ALL          1
  80
  81 static char print_alua_state(int);
  82 static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
  83
  84 static inline struct alua_dh_data *get_alua_data(struct scsi_device *sdev)
  85 {
  86         struct scsi_dh_data *scsi_dh_data = sdev->scsi_dh_data;
  87         BUG_ON(scsi_dh_data == NULL);
  88         return ((struct alua_dh_data *) scsi_dh_data->buf);
  89 }
  90
  91 static int realloc_buffer(struct alua_dh_data *h, unsigned len)
  92 {
  93         if (h->buff && h->buff != h->inq)
  94                 kfree(h->buff);
  95
  96         h->buff = kmalloc(len, GFP_NOIO);
  97         if (!h->buff) {
  98                 h->buff = h->inq;
  99                 h->bufflen = ALUA_INQUIRY_SIZE;
 100                 return 1;
 101         }
 102         h->bufflen = len;
 103         return 0;
 104 }
 105
 106 static struct request *get_alua_req(struct scsi_device *sdev,
 107                                     void *buffer, unsigned buflen, int rw)
 108 {
 109         struct request *rq;
 110         struct request_queue *q = sdev->request_queue;
 111
 112         rq = blk_get_request(q, rw, GFP_NOIO);
 113
 114         if (!rq) {
 115                 sdev_printk(KERN_INFO, sdev,
 116                             "%s: blk_get_request failed\n", __func__);
 117                 return NULL;
 118         }
 119
 120         if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
 121                 blk_put_request(rq);
 122                 sdev_printk(KERN_INFO, sdev,
 123                             "%s: blk_rq_map_kern failed\n", __func__);
 124                 return NULL;
 125         }
 126
 127         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 128         rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
 129                          REQ_FAILFAST_DRIVER;
 130         rq->retries = ALUA_FAILOVER_RETRIES;
 131         rq->timeout = ALUA_FAILOVER_TIMEOUT;
 132
 133         return rq;
 134 }
 135
 136 /*
 137  * submit_vpd_inquiry - Issue an INQUIRY VPD page 0x83 command
 138  * @sdev: sdev the command should be sent to
 139  */
 140 static int submit_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
 141 {
 142         struct request *rq;
 143         int err = SCSI_DH_RES_TEMP_UNAVAIL;
 144
 145         rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
 146         if (!rq)
 147                 goto done;
 148
 149         /* Prepare the command. */
 150         rq->cmd[0] = INQUIRY;
 151         rq->cmd[1] = 1;
 152         rq->cmd[2] = 0x83;
 153         rq->cmd[4] = h->bufflen;
 154         rq->cmd_len = COMMAND_SIZE(INQUIRY);
 155
 156         rq->sense = h->sense;
 157         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 158         rq->sense_len = h->senselen = 0;
 159
 160         err = blk_execute_rq(rq->q, NULL, rq, 1);
 161         if (err == -EIO) {
 162                 sdev_printk(KERN_INFO, sdev,
 163                             "%s: evpd inquiry failed with %x\n",
 164                             ALUA_DH_NAME, rq->errors);
 165                 h->senselen = rq->sense_len;
 166                 err = SCSI_DH_IO;
 167         }
 168         blk_put_request(rq);
 169 done:
 170         return err;
 171 }
 172
 173 /*
 174  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
 175  * @sdev: sdev the command should be sent to
 176  */
 177 static unsigned submit_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
 178 {
 179         struct request *rq;
 180         int err = SCSI_DH_RES_TEMP_UNAVAIL;
 181
 182         rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
 183         if (!rq)
 184                 goto done;
 185
 186         /* Prepare the command. */
 187         rq->cmd[0] = MAINTENANCE_IN;
 188         rq->cmd[1] = MI_REPORT_TARGET_PGS;
 189         rq->cmd[6] = (h->bufflen >> 24) & 0xff;
 190         rq->cmd[7] = (h->bufflen >> 16) & 0xff;
 191         rq->cmd[8] = (h->bufflen >>  8) & 0xff;
 192         rq->cmd[9] = h->bufflen & 0xff;
 193         rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
 194
 195         rq->sense = h->sense;
 196         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 197         rq->sense_len = h->senselen = 0;
 198
 199         err = blk_execute_rq(rq->q, NULL, rq, 1);
 200         if (err == -EIO) {
 201                 sdev_printk(KERN_INFO, sdev,
 202                             "%s: rtpg failed with %x\n",
 203                             ALUA_DH_NAME, rq->errors);
 204                 h->senselen = rq->sense_len;
 205                 err = SCSI_DH_IO;
 206         }
 207         blk_put_request(rq);
 208 done:
 209         return err;
 210 }
 211
 212 /*
 213  * alua_stpg - Evaluate SET TARGET GROUP STATES
 214  * @sdev: the device to be evaluated
 215  * @state: the new target group state
 216  *
 217  * Send a SET TARGET GROUP STATES command to the device.
 218  * We only have to test here if we should resubmit the command;
 219  * any other error is assumed as a failure.
 220  */
 221 static void stpg_endio(struct request *req, int error)
 222 {
 223         struct alua_dh_data *h = req->end_io_data;
 224         struct scsi_sense_hdr sense_hdr;
 225         unsigned err = SCSI_DH_OK;
 226
 227         if (error || host_byte(req->errors) != DID_OK ||
 228                         msg_byte(req->errors) != COMMAND_COMPLETE) {
 229                 err = SCSI_DH_IO;
 230                 goto done;
 231         }
 232
 233         if (h->senselen > 0) {
 234                 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
 235                                            &sense_hdr);
 236                 if (!err) {
 237                         err = SCSI_DH_IO;
 238                         goto done;
 239                 }
 240                 err = alua_check_sense(h->sdev, &sense_hdr);
 241                 if (err == ADD_TO_MLQUEUE) {
 242                         err = SCSI_DH_RETRY;
 243                         goto done;
 244                 }
 245                 sdev_printk(KERN_INFO, h->sdev,
 246                             "%s: stpg sense code: %02x/%02x/%02x\n",
 247                             ALUA_DH_NAME, sense_hdr.sense_key,
 248                             sense_hdr.asc, sense_hdr.ascq);
 249                 err = SCSI_DH_IO;
 250         }
 251         if (err == SCSI_DH_OK) {
 252                 h->state = TPGS_STATE_OPTIMIZED;
 253                 sdev_printk(KERN_INFO, h->sdev,
 254                             "%s: port group %02x switched to state %c\n",
 255                             ALUA_DH_NAME, h->group_id,
 256                             print_alua_state(h->state));
 257         }
 258 done:
 259         req->end_io_data = NULL;
 260         __blk_put_request(req->q, req);
 261         if (h->callback_fn) {
 262                 h->callback_fn(h->callback_data, err);
 263                 h->callback_fn = h->callback_data = NULL;
 264         }
 265         return;
 266 }
 267
 268 /*
 269  * submit_stpg - Issue a SET TARGET GROUP STATES command
 270  *
 271  * Currently we're only setting the current target port group state
 272  * to 'active/optimized' and let the array firmware figure out
 273  * the states of the remaining groups.
 274  */
 275 static unsigned submit_stpg(struct alua_dh_data *h)
 276 {
 277         struct request *rq;
 278         int stpg_len = 8;
 279         struct scsi_device *sdev = h->sdev;
 280
 281         /* Prepare the data buffer */
 282         memset(h->buff, 0, stpg_len);
 283         h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
 284         h->buff[6] = (h->group_id >> 8) & 0xff;
 285         h->buff[7] = h->group_id & 0xff;
 286
 287         rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
 288         if (!rq)
 289                 return SCSI_DH_RES_TEMP_UNAVAIL;
 290
 291         /* Prepare the command. */
 292         rq->cmd[0] = MAINTENANCE_OUT;
 293         rq->cmd[1] = MO_SET_TARGET_PGS;
 294         rq->cmd[6] = (stpg_len >> 24) & 0xff;
 295         rq->cmd[7] = (stpg_len >> 16) & 0xff;
 296         rq->cmd[8] = (stpg_len >>  8) & 0xff;
 297         rq->cmd[9] = stpg_len & 0xff;
 298         rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
 299
 300         rq->sense = h->sense;
 301         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 302         rq->sense_len = h->senselen = 0;
 303         rq->end_io_data = h;
 304
 305         blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
 306         return SCSI_DH_OK;
 307 }
 308
 309 /*
 310  * alua_check_tpgs - Evaluate TPGS setting
 311  * @sdev: device to be checked
 312  *
 313  * Examine the TPGS setting of the sdev to find out if ALUA
 314  * is supported.
 315  */
 316 static int alua_check_tpgs(struct scsi_device *sdev, struct alua_dh_data *h)
 317 {
 318         int err = SCSI_DH_OK;
 319
 320         h->tpgs = scsi_device_tpgs(sdev);
 321         switch (h->tpgs) {
 322         case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
 323                 sdev_printk(KERN_INFO, sdev,
 324                             "%s: supports implicit and explicit TPGS\n",
 325                             ALUA_DH_NAME);
 326                 break;
 327         case TPGS_MODE_EXPLICIT:
 328                 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
 329                             ALUA_DH_NAME);
 330                 break;
 331         case TPGS_MODE_IMPLICIT:
 332                 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
 333                             ALUA_DH_NAME);
 334                 break;
 335         default:
 336                 h->tpgs = TPGS_MODE_NONE;
 337                 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
 338                             ALUA_DH_NAME);
 339                 err = SCSI_DH_DEV_UNSUPP;
 340                 break;
 341         }
 342
 343         return err;
 344 }
 345
 346 /*
 347  * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
 348  * @sdev: device to be checked
 349  *
 350  * Extract the relative target port and the target port group
 351  * descriptor from the list of identificators.
 352  */
 353 static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
 354 {
 355         int len;
 356         unsigned err;
 357         unsigned char *d;
 358
 359  retry:
 360         err = submit_vpd_inquiry(sdev, h);
 361
 362         if (err != SCSI_DH_OK)
 363                 return err;
 364
 365         /* Check if vpd page exceeds initial buffer */
 366         len = (h->buff[2] << 8) + h->buff[3] + 4;
 367         if (len > h->bufflen) {
 368                 /* Resubmit with the correct length */
 369                 if (realloc_buffer(h, len)) {
 370                         sdev_printk(KERN_WARNING, sdev,
 371                                     "%s: kmalloc buffer failed\n",
 372                                     ALUA_DH_NAME);
 373                         /* Temporary failure, bypass */
 374                         return SCSI_DH_DEV_TEMP_BUSY;
 375                 }
 376                 goto retry;
 377         }
 378
 379         /*
 380          * Now look for the correct descriptor.
 381          */
 382         d = h->buff + 4;
 383         while (d < h->buff + len) {
 384                 switch (d[1] & 0xf) {
 385                 case 0x4:
 386                         /* Relative target port */
 387                         h->rel_port = (d[6] << 8) + d[7];
 388                         break;
 389                 case 0x5:
 390                         /* Target port group */
 391                         h->group_id = (d[6] << 8) + d[7];
 392                         break;
 393                 default:
 394                         break;
 395                 }
 396                 d += d[3] + 4;
 397         }
 398
 399         if (h->group_id == -1) {
 400                 /*
 401                  * Internal error; TPGS supported but required
 402                  * VPD identification descriptors not present.
 403                  * Disable ALUA support
 404                  */
 405                 sdev_printk(KERN_INFO, sdev,
 406                             "%s: No target port descriptors found\n",
 407                             ALUA_DH_NAME);
 408                 h->state = TPGS_STATE_OPTIMIZED;
 409                 h->tpgs = TPGS_MODE_NONE;
 410                 err = SCSI_DH_DEV_UNSUPP;
 411         } else {
 412                 sdev_printk(KERN_INFO, sdev,
 413                             "%s: port group %02x rel port %02x\n",
 414                             ALUA_DH_NAME, h->group_id, h->rel_port);
 415         }
 416
 417         return err;
 418 }
 419
 420 static char print_alua_state(int state)
 421 {
 422         switch (state) {
 423         case TPGS_STATE_OPTIMIZED:
 424                 return 'A';
 425         case TPGS_STATE_NONOPTIMIZED:
 426                 return 'N';
 427         case TPGS_STATE_STANDBY:
 428                 return 'S';
 429         case TPGS_STATE_UNAVAILABLE:
 430                 return 'U';
 431         case TPGS_STATE_LBA_DEPENDENT:
 432                 return 'L';
 433         case TPGS_STATE_OFFLINE:
 434                 return 'O';
 435         case TPGS_STATE_TRANSITIONING:
 436                 return 'T';
 437         default:
 438                 return 'X';
 439         }
 440 }
 441
 442 static int alua_check_sense(struct scsi_device *sdev,
 443                             struct scsi_sense_hdr *sense_hdr)
 444 {
 445         switch (sense_hdr->sense_key) {
 446         case NOT_READY:
 447                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
 448                         /*
 449                          * LUN Not Accessible - ALUA state transition
 450                          */
 451                         return ADD_TO_MLQUEUE;
 452                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
 453                         /*
 454                          * LUN Not Accessible -- Target port in standby state
 455                          */
 456                         return SUCCESS;
 457                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
 458                         /*
 459                          * LUN Not Accessible -- Target port in unavailable state
 460                          */
 461                         return SUCCESS;
 462                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
 463                         /*
 464                          * LUN Not Ready -- Offline
 465                          */
 466                         return SUCCESS;
 467                 break;
 468         case UNIT_ATTENTION:
 469                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
 470                         /*
 471                          * Power On, Reset, or Bus Device Reset, just retry.
 472                          */
 473                         return ADD_TO_MLQUEUE;
 474                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
 475                         /*
 476                          * Mode Parameters Changed
 477                          */
 478                         return ADD_TO_MLQUEUE;
 479                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06)
 480                         /*
 481                          * ALUA state changed
 482                          */
 483                         return ADD_TO_MLQUEUE;
 484                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07)
 485                         /*
 486                          * Implicit ALUA state transition failed
 487                          */
 488                         return ADD_TO_MLQUEUE;
 489                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
 490                         /*
 491                          * Inquiry data has changed
 492                          */
 493                         return ADD_TO_MLQUEUE;
 494                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
 495                         /*
 496                          * REPORTED_LUNS_DATA_HAS_CHANGED is reported
 497                          * when switching controllers on targets like
 498                          * Intel Multi-Flex. We can just retry.
 499                          */
 500                         return ADD_TO_MLQUEUE;
 501                 break;
 502         }
 503
 504         return SCSI_RETURN_NOT_HANDLED;
 505 }
 506
 507 /*
 508  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
 509  * @sdev: the device to be evaluated.
 510  *
 511  * Evaluate the Target Port Group State.
 512  * Returns SCSI_DH_DEV_OFFLINED if the path is
 513  * found to be unusable.
 514  */
 515 static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
 516 {
 517         struct scsi_sense_hdr sense_hdr;
 518         int len, k, off, valid_states = 0;
 519         unsigned char *ucp;
 520         unsigned err;
 521         unsigned long expiry, interval = 1000;
 522
 523         expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT);
 524  retry:
 525         err = submit_rtpg(sdev, h);
 526
 527         if (err == SCSI_DH_IO && h->senselen > 0) {
 528                 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
 529                                            &sense_hdr);
 530                 if (!err)
 531                         return SCSI_DH_IO;
 532
 533                 err = alua_check_sense(sdev, &sense_hdr);
 534                 if (err == ADD_TO_MLQUEUE && time_before(jiffies, expiry))
 535                         goto retry;
 536                 sdev_printk(KERN_INFO, sdev,
 537                             "%s: rtpg sense code %02x/%02x/%02x\n",
 538                             ALUA_DH_NAME, sense_hdr.sense_key,
 539                             sense_hdr.asc, sense_hdr.ascq);
 540                 err = SCSI_DH_IO;
 541         }
 542         if (err != SCSI_DH_OK)
 543                 return err;
 544
 545         len = (h->buff[0] << 24) + (h->buff[1] << 16) +
 546                 (h->buff[2] << 8) + h->buff[3] + 4;
 547
 548         if (len > h->bufflen) {
 549                 /* Resubmit with the correct length */
 550                 if (realloc_buffer(h, len)) {
 551                         sdev_printk(KERN_WARNING, sdev,
 552                                     "%s: kmalloc buffer failed\n",__func__);
 553                         /* Temporary failure, bypass */
 554                         return SCSI_DH_DEV_TEMP_BUSY;
 555                 }
 556                 goto retry;
 557         }
 558
 559         for (k = 4, ucp = h->buff + 4; k < len; k += off, ucp += off) {
 560                 if (h->group_id == (ucp[2] << 8) + ucp[3]) {
 561                         h->state = ucp[0] & 0x0f;
 562                         h->pref = ucp[0] >> 7;
 563                         valid_states = ucp[1];
 564                 }
 565                 off = 8 + (ucp[7] * 4);
 566         }
 567
 568         sdev_printk(KERN_INFO, sdev,
 569                     "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
 570                     ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
 571                     h->pref ? "preferred" : "non-preferred",
 572                     valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
 573                     valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
 574                     valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
 575                     valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
 576                     valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
 577                     valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
 578                     valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
 579
 580         switch (h->state) {
 581         case TPGS_STATE_TRANSITIONING:
 582                 if (time_before(jiffies, expiry)) {
 583                         /* State transition, retry */
 584                         interval *= 2;
 585                         msleep(interval);
 586                         goto retry;
 587                 }
 588                 /* Transitioning time exceeded, set port to standby */
 589                 err = SCSI_DH_RETRY;
 590                 h->state = TPGS_STATE_STANDBY;
 591                 break;
 592         case TPGS_STATE_OFFLINE:
 593         case TPGS_STATE_UNAVAILABLE:
 594                 /* Path unusable for unavailable/offline */
 595                 err = SCSI_DH_DEV_OFFLINED;
 596                 break;
 597         default:
 598                 /* Useable path if active */
 599                 err = SCSI_DH_OK;
 600                 break;
 601         }
 602         return err;
 603 }
 604
 605 /*
 606  * alua_initialize - Initialize ALUA state
 607  * @sdev: the device to be initialized
 608  *
 609  * For the prep_fn to work correctly we have
 610  * to initialize the ALUA state for the device.
 611  */
 612 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
 613 {
 614         int err;
 615
 616         err = alua_check_tpgs(sdev, h);
 617         if (err != SCSI_DH_OK)
 618                 goto out;
 619
 620         err = alua_vpd_inquiry(sdev, h);
 621         if (err != SCSI_DH_OK)
 622                 goto out;
 623
 624         err = alua_rtpg(sdev, h);
 625         if (err != SCSI_DH_OK)
 626                 goto out;
 627
 628 out:
 629         return err;
 630 }
 631 /*
 632  * alua_set_params - set/unset the optimize flag
 633  * @sdev: device on the path to be activated
 634  * params - parameters in the following format
 635  *      "no_of_params\0param1\0param2\0param3\0...\0"
 636  * For example, to set the flag pass the following parameters
 637  * from multipath.conf
 638  *     hardware_handler        "2 alua 1"
 639  */
 640 static int alua_set_params(struct scsi_device *sdev, const char *params)
 641 {
 642         struct alua_dh_data *h = get_alua_data(sdev);
 643         unsigned int optimize = 0, argc;
 644         const char *p = params;
 645         int result = SCSI_DH_OK;
 646
 647         if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
 648                 return -EINVAL;
 649
 650         while (*p++)
 651                 ;
 652         if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
 653                 return -EINVAL;
 654
 655         if (optimize)
 656                 h->flags |= ALUA_OPTIMIZE_STPG;
 657         else
 658                 h->flags &= ~ALUA_OPTIMIZE_STPG;
 659
 660         return result;
 661 }
 662
 663 /*
 664  * alua_activate - activate a path
 665  * @sdev: device on the path to be activated
 666  *
 667  * We're currently switching the port group to be activated only and
 668  * let the array figure out the rest.
 669  * There may be other arrays which require us to switch all port groups
 670  * based on a certain policy. But until we actually encounter them it
 671  * should be okay.
 672  */
 673 static int alua_activate(struct scsi_device *sdev,
 674                         activate_complete fn, void *data)
 675 {
 676         struct alua_dh_data *h = get_alua_data(sdev);
 677         int err = SCSI_DH_OK;
 678         int stpg = 0;
 679
 680         err = alua_rtpg(sdev, h);
 681         if (err != SCSI_DH_OK)
 682                 goto out;
 683
 684         if (h->tpgs & TPGS_MODE_EXPLICIT) {
 685                 switch (h->state) {
 686                 case TPGS_STATE_NONOPTIMIZED:
 687                         stpg = 1;
 688                         if ((h->flags & ALUA_OPTIMIZE_STPG) &&
 689                             (!h->pref) &&
 690                             (h->tpgs & TPGS_MODE_IMPLICIT))
 691                                 stpg = 0;
 692                         break;
 693                 case TPGS_STATE_STANDBY:
 694                         stpg = 1;
 695                         break;
 696                 case TPGS_STATE_UNAVAILABLE:
 697                 case TPGS_STATE_OFFLINE:
 698                         err = SCSI_DH_IO;
 699                         break;
 700                 case TPGS_STATE_TRANSITIONING:
 701                         err = SCSI_DH_RETRY;
 702                         break;
 703                 default:
 704                         break;
 705                 }
 706         }
 707
 708         if (stpg) {
 709                 h->callback_fn = fn;
 710                 h->callback_data = data;
 711                 err = submit_stpg(h);
 712                 if (err == SCSI_DH_OK)
 713                         return 0;
 714                 h->callback_fn = h->callback_data = NULL;
 715         }
 716
 717 out:
 718         if (fn)
 719                 fn(data, err);
 720         return 0;
 721 }
 722
 723 /*
 724  * alua_prep_fn - request callback
 725  *
 726  * Fail I/O to all paths not in state
 727  * active/optimized or active/non-optimized.
 728  */
 729 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
 730 {
 731         struct alua_dh_data *h = get_alua_data(sdev);
 732         int ret = BLKPREP_OK;
 733
 734         if (h->state == TPGS_STATE_TRANSITIONING)
 735                 ret = BLKPREP_DEFER;
 736         else if (h->state != TPGS_STATE_OPTIMIZED &&
 737                  h->state != TPGS_STATE_NONOPTIMIZED &&
 738                  h->state != TPGS_STATE_LBA_DEPENDENT) {
 739                 ret = BLKPREP_KILL;
 740                 req->cmd_flags |= REQ_QUIET;
 741         }
 742         return ret;
 743
 744 }
 745
 746 static bool alua_match(struct scsi_device *sdev)
 747 {
 748         return (scsi_device_tpgs(sdev) != 0);
 749 }
 750
 751 static int alua_bus_attach(struct scsi_device *sdev);
 752 static void alua_bus_detach(struct scsi_device *sdev);
 753
 754 static struct scsi_device_handler alua_dh = {
 755         .name = ALUA_DH_NAME,
 756         .module = THIS_MODULE,
 757         .attach = alua_bus_attach,
 758         .detach = alua_bus_detach,
 759         .prep_fn = alua_prep_fn,
 760         .check_sense = alua_check_sense,
 761         .activate = alua_activate,
 762         .set_params = alua_set_params,
 763         .match = alua_match,
 764 };
 765
 766 /*
 767  * alua_bus_attach - Attach device handler
 768  * @sdev: device to be attached to
 769  */
 770 static int alua_bus_attach(struct scsi_device *sdev)
 771 {
 772         struct scsi_dh_data *scsi_dh_data;
 773         struct alua_dh_data *h;
 774         unsigned long flags;
 775         int err = SCSI_DH_OK;
 776
 777         scsi_dh_data = kzalloc(sizeof(*scsi_dh_data)
 778                                + sizeof(*h) , GFP_KERNEL);
 779         if (!scsi_dh_data) {
 780                 sdev_printk(KERN_ERR, sdev, "%s: Attach failed\n",
 781                             ALUA_DH_NAME);
 782                 return -ENOMEM;
 783         }
 784
 785         scsi_dh_data->scsi_dh = &alua_dh;
 786         h = (struct alua_dh_data *) scsi_dh_data->buf;
 787         h->tpgs = TPGS_MODE_UNINITIALIZED;
 788         h->state = TPGS_STATE_OPTIMIZED;
 789         h->group_id = -1;
 790         h->rel_port = -1;
 791         h->buff = h->inq;
 792         h->bufflen = ALUA_INQUIRY_SIZE;
 793         h->sdev = sdev;
 794
 795         err = alua_initialize(sdev, h);
 796         if ((err != SCSI_DH_OK) && (err != SCSI_DH_DEV_OFFLINED))
 797                 goto failed;
 798
 799         if (!try_module_get(THIS_MODULE))
 800                 goto failed;
 801
 802         spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
 803         sdev->scsi_dh_data = scsi_dh_data;
 804         spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
 805         sdev_printk(KERN_NOTICE, sdev, "%s: Attached\n", ALUA_DH_NAME);
 806
 807         return 0;
 808
 809 failed:
 810         kfree(scsi_dh_data);
 811         sdev_printk(KERN_ERR, sdev, "%s: not attached\n", ALUA_DH_NAME);
 812         return -EINVAL;
 813 }
 814
 815 /*
 816  * alua_bus_detach - Detach device handler
 817  * @sdev: device to be detached from
 818  */
 819 static void alua_bus_detach(struct scsi_device *sdev)
 820 {
 821         struct scsi_dh_data *scsi_dh_data;
 822         struct alua_dh_data *h;
 823         unsigned long flags;
 824
 825         spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
 826         scsi_dh_data = sdev->scsi_dh_data;
 827         sdev->scsi_dh_data = NULL;
 828         spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
 829
 830         h = (struct alua_dh_data *) scsi_dh_data->buf;
 831         if (h->buff && h->inq != h->buff)
 832                 kfree(h->buff);
 833         kfree(scsi_dh_data);
 834         module_put(THIS_MODULE);
 835         sdev_printk(KERN_NOTICE, sdev, "%s: Detached\n", ALUA_DH_NAME);
 836 }
 837
 838 static int __init alua_init(void)
 839 {
 840         int r;
 841
 842         r = scsi_register_device_handler(&alua_dh);
 843         if (r != 0)
 844                 printk(KERN_ERR "%s: Failed to register scsi device handler",
 845                         ALUA_DH_NAME);
 846         return r;
 847 }
 848
 849 static void __exit alua_exit(void)
 850 {
 851         scsi_unregister_device_handler(&alua_dh);
 852 }
 853
 854 module_init(alua_init);
 855 module_exit(alua_exit);
 856
 857 MODULE_DESCRIPTION("DM Multipath ALUA support");
 858 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
 859 MODULE_LICENSE("GPL");
 860 MODULE_VERSION(ALUA_DH_VER);