drivers/scsi/sd.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *      sd.c Copyright (C) 1992 Drew Eckhardt
   4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
   5  *
   6  *      Linux scsi disk driver
   7  *              Initial versions: Drew Eckhardt
   8  *              Subsequent revisions: Eric Youngdale
   9  *      Modification history:
  10  *       - Drew Eckhardt <[email protected]> original
  11  *       - Eric Youngdale <[email protected]> add scatter-gather, multiple
  12  *         outstanding request, and other enhancements.
  13  *         Support loadable low-level scsi drivers.
  14  *       - Jirka Hanika <[email protected]> support more scsi disks using
  15  *         eight major numbers.
  16  *       - Richard Gooch <[email protected]> support devfs.
  17  *       - Torben Mathiasen <[email protected]> Resource allocation fixes in
  18  *         sd_init and cleanups.
  19  *       - Alex Davis <[email protected]> Fix problem where partition info
  20  *         not being read in sd_open. Fix problem where removable media
  21  *         could be ejected after sd_open.
  22  *       - Douglas Gilbert <[email protected]> cleanup for lk 2.5.x
  23  *       - Badari Pulavarty <[email protected]>, Matthew Wilcox
  24  *         <[email protected]>, Kurt Garloff <[email protected]>:
  25  *         Support 32k/1M disks.
  26  *
  27  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
  28  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
  29  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
  30  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
  31  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
  32  *      Note: when the logging level is set by the user, it must be greater
  33  *      than the level indicated above to trigger output.
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/fs.h>
  38 #include <linux/kernel.h>
  39 #include <linux/mm.h>
  40 #include <linux/bio.h>
  41 #include <linux/genhd.h>
  42 #include <linux/hdreg.h>
  43 #include <linux/errno.h>
  44 #include <linux/idr.h>
  45 #include <linux/interrupt.h>
  46 #include <linux/init.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/blkpg.h>
  49 #include <linux/blk-pm.h>
  50 #include <linux/delay.h>
  51 #include <linux/mutex.h>
  52 #include <linux/string_helpers.h>
  53 #include <linux/async.h>
  54 #include <linux/slab.h>
  55 #include <linux/sed-opal.h>
  56 #include <linux/pm_runtime.h>
  57 #include <linux/pr.h>
  58 #include <linux/t10-pi.h>
  59 #include <linux/uaccess.h>
  60 #include <asm/unaligned.h>
  61
  62 #include <scsi/scsi.h>
  63 #include <scsi/scsi_cmnd.h>
  64 #include <scsi/scsi_dbg.h>
  65 #include <scsi/scsi_device.h>
  66 #include <scsi/scsi_driver.h>
  67 #include <scsi/scsi_eh.h>
  68 #include <scsi/scsi_host.h>
  69 #include <scsi/scsi_ioctl.h>
  70 #include <scsi/scsicam.h>
  71
  72 #include "sd.h"
  73 #include "scsi_priv.h"
  74 #include "scsi_logging.h"
  75
  76 MODULE_AUTHOR("Eric Youngdale");
  77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
  78 MODULE_LICENSE("GPL");
  79
  80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
  81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
  82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
  83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
  84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
  85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
  86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
  87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
  88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
  89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
  90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
  91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
  92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
  93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
  94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
  95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
  96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
  97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
  98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
  99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
 100
 101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
 102 #define SD_MINORS       16
 103 #else
 104 #define SD_MINORS       0
 105 #endif
 106
 107 static void sd_config_discard(struct scsi_disk *, unsigned int);
 108 static void sd_config_write_same(struct scsi_disk *);
 109 static int  sd_revalidate_disk(struct gendisk *);
 110 static void sd_unlock_native_capacity(struct gendisk *disk);
 111 static int  sd_probe(struct device *);
 112 static int  sd_remove(struct device *);
 113 static void sd_shutdown(struct device *);
 114 static int sd_suspend_system(struct device *);
 115 static int sd_suspend_runtime(struct device *);
 116 static int sd_resume(struct device *);
 117 static void sd_rescan(struct device *);
 118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
 119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
 120 static int sd_done(struct scsi_cmnd *);
 121 static void sd_eh_reset(struct scsi_cmnd *);
 122 static int sd_eh_action(struct scsi_cmnd *, int);
 123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 124 static void scsi_disk_release(struct device *cdev);
 125
 126 static DEFINE_IDA(sd_index_ida);
 127
 128 /* This semaphore is used to mediate the 0->1 reference get in the
 129  * face of object destruction (i.e. we can't allow a get on an
 130  * object after last put) */
 131 static DEFINE_MUTEX(sd_ref_mutex);
 132
 133 static struct kmem_cache *sd_cdb_cache;
 134 static mempool_t *sd_cdb_pool;
 135 static mempool_t *sd_page_pool;
 136
 137 static const char *sd_cache_types[] = {
 138         "write through", "none", "write back",
 139         "write back, no read (daft)"
 140 };
 141
 142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 143 {
 144         bool wc = false, fua = false;
 145
 146         if (sdkp->WCE) {
 147                 wc = true;
 148                 if (sdkp->DPOFUA)
 149                         fua = true;
 150         }
 151
 152         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 153 }
 154
 155 static ssize_t
 156 cache_type_store(struct device *dev, struct device_attribute *attr,
 157                  const char *buf, size_t count)
 158 {
 159         int ct, rcd, wce, sp;
 160         struct scsi_disk *sdkp = to_scsi_disk(dev);
 161         struct scsi_device *sdp = sdkp->device;
 162         char buffer[64];
 163         char *buffer_data;
 164         struct scsi_mode_data data;
 165         struct scsi_sense_hdr sshdr;
 166         static const char temp[] = "temporary ";
 167         int len;
 168
 169         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 170                 /* no cache control on RBC devices; theoretically they
 171                  * can do it, but there's probably so many exceptions
 172                  * it's not worth the risk */
 173                 return -EINVAL;
 174
 175         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 176                 buf += sizeof(temp) - 1;
 177                 sdkp->cache_override = 1;
 178         } else {
 179                 sdkp->cache_override = 0;
 180         }
 181
 182         ct = sysfs_match_string(sd_cache_types, buf);
 183         if (ct < 0)
 184                 return -EINVAL;
 185
 186         rcd = ct & 0x01 ? 1 : 0;
 187         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 188
 189         if (sdkp->cache_override) {
 190                 sdkp->WCE = wce;
 191                 sdkp->RCD = rcd;
 192                 sd_set_flush_flag(sdkp);
 193                 return count;
 194         }
 195
 196         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 197                             sdkp->max_retries, &data, NULL))
 198                 return -EINVAL;
 199         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 200                   data.block_descriptor_length);
 201         buffer_data = buffer + data.header_length +
 202                 data.block_descriptor_length;
 203         buffer_data[2] &= ~0x05;
 204         buffer_data[2] |= wce << 2 | rcd;
 205         sp = buffer_data[0] & 0x80 ? 1 : 0;
 206         buffer_data[0] &= ~0x80;
 207
 208         /*
 209          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 210          * received mode parameter buffer before doing MODE SELECT.
 211          */
 212         data.device_specific = 0;
 213
 214         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
 215                              sdkp->max_retries, &data, &sshdr)) {
 216                 if (scsi_sense_valid(&sshdr))
 217                         sd_print_sense_hdr(sdkp, &sshdr);
 218                 return -EINVAL;
 219         }
 220         sd_revalidate_disk(sdkp->disk);
 221         return count;
 222 }
 223
 224 static ssize_t
 225 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
 226                        char *buf)
 227 {
 228         struct scsi_disk *sdkp = to_scsi_disk(dev);
 229         struct scsi_device *sdp = sdkp->device;
 230
 231         return sprintf(buf, "%u\n", sdp->manage_start_stop);
 232 }
 233
 234 static ssize_t
 235 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
 236                         const char *buf, size_t count)
 237 {
 238         struct scsi_disk *sdkp = to_scsi_disk(dev);
 239         struct scsi_device *sdp = sdkp->device;
 240         bool v;
 241
 242         if (!capable(CAP_SYS_ADMIN))
 243                 return -EACCES;
 244
 245         if (kstrtobool(buf, &v))
 246                 return -EINVAL;
 247
 248         sdp->manage_start_stop = v;
 249
 250         return count;
 251 }
 252 static DEVICE_ATTR_RW(manage_start_stop);
 253
 254 static ssize_t
 255 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 256 {
 257         struct scsi_disk *sdkp = to_scsi_disk(dev);
 258
 259         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 260 }
 261
 262 static ssize_t
 263 allow_restart_store(struct device *dev, struct device_attribute *attr,
 264                     const char *buf, size_t count)
 265 {
 266         bool v;
 267         struct scsi_disk *sdkp = to_scsi_disk(dev);
 268         struct scsi_device *sdp = sdkp->device;
 269
 270         if (!capable(CAP_SYS_ADMIN))
 271                 return -EACCES;
 272
 273         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 274                 return -EINVAL;
 275
 276         if (kstrtobool(buf, &v))
 277                 return -EINVAL;
 278
 279         sdp->allow_restart = v;
 280
 281         return count;
 282 }
 283 static DEVICE_ATTR_RW(allow_restart);
 284
 285 static ssize_t
 286 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 287 {
 288         struct scsi_disk *sdkp = to_scsi_disk(dev);
 289         int ct = sdkp->RCD + 2*sdkp->WCE;
 290
 291         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 292 }
 293 static DEVICE_ATTR_RW(cache_type);
 294
 295 static ssize_t
 296 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 297 {
 298         struct scsi_disk *sdkp = to_scsi_disk(dev);
 299
 300         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 301 }
 302 static DEVICE_ATTR_RO(FUA);
 303
 304 static ssize_t
 305 protection_type_show(struct device *dev, struct device_attribute *attr,
 306                      char *buf)
 307 {
 308         struct scsi_disk *sdkp = to_scsi_disk(dev);
 309
 310         return sprintf(buf, "%u\n", sdkp->protection_type);
 311 }
 312
 313 static ssize_t
 314 protection_type_store(struct device *dev, struct device_attribute *attr,
 315                       const char *buf, size_t count)
 316 {
 317         struct scsi_disk *sdkp = to_scsi_disk(dev);
 318         unsigned int val;
 319         int err;
 320
 321         if (!capable(CAP_SYS_ADMIN))
 322                 return -EACCES;
 323
 324         err = kstrtouint(buf, 10, &val);
 325
 326         if (err)
 327                 return err;
 328
 329         if (val <= T10_PI_TYPE3_PROTECTION)
 330                 sdkp->protection_type = val;
 331
 332         return count;
 333 }
 334 static DEVICE_ATTR_RW(protection_type);
 335
 336 static ssize_t
 337 protection_mode_show(struct device *dev, struct device_attribute *attr,
 338                      char *buf)
 339 {
 340         struct scsi_disk *sdkp = to_scsi_disk(dev);
 341         struct scsi_device *sdp = sdkp->device;
 342         unsigned int dif, dix;
 343
 344         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 345         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 346
 347         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 348                 dif = 0;
 349                 dix = 1;
 350         }
 351
 352         if (!dif && !dix)
 353                 return sprintf(buf, "none\n");
 354
 355         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 356 }
 357 static DEVICE_ATTR_RO(protection_mode);
 358
 359 static ssize_t
 360 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 361 {
 362         struct scsi_disk *sdkp = to_scsi_disk(dev);
 363
 364         return sprintf(buf, "%u\n", sdkp->ATO);
 365 }
 366 static DEVICE_ATTR_RO(app_tag_own);
 367
 368 static ssize_t
 369 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 370                        char *buf)
 371 {
 372         struct scsi_disk *sdkp = to_scsi_disk(dev);
 373
 374         return sprintf(buf, "%u\n", sdkp->lbpme);
 375 }
 376 static DEVICE_ATTR_RO(thin_provisioning);
 377
 378 /* sysfs_match_string() requires dense arrays */
 379 static const char *lbp_mode[] = {
 380         [SD_LBP_FULL]           = "full",
 381         [SD_LBP_UNMAP]          = "unmap",
 382         [SD_LBP_WS16]           = "writesame_16",
 383         [SD_LBP_WS10]           = "writesame_10",
 384         [SD_LBP_ZERO]           = "writesame_zero",
 385         [SD_LBP_DISABLE]        = "disabled",
 386 };
 387
 388 static ssize_t
 389 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 390                        char *buf)
 391 {
 392         struct scsi_disk *sdkp = to_scsi_disk(dev);
 393
 394         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 395 }
 396
 397 static ssize_t
 398 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 399                         const char *buf, size_t count)
 400 {
 401         struct scsi_disk *sdkp = to_scsi_disk(dev);
 402         struct scsi_device *sdp = sdkp->device;
 403         int mode;
 404
 405         if (!capable(CAP_SYS_ADMIN))
 406                 return -EACCES;
 407
 408         if (sd_is_zoned(sdkp)) {
 409                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 410                 return count;
 411         }
 412
 413         if (sdp->type != TYPE_DISK)
 414                 return -EINVAL;
 415
 416         mode = sysfs_match_string(lbp_mode, buf);
 417         if (mode < 0)
 418                 return -EINVAL;
 419
 420         sd_config_discard(sdkp, mode);
 421
 422         return count;
 423 }
 424 static DEVICE_ATTR_RW(provisioning_mode);
 425
 426 /* sysfs_match_string() requires dense arrays */
 427 static const char *zeroing_mode[] = {
 428         [SD_ZERO_WRITE]         = "write",
 429         [SD_ZERO_WS]            = "writesame",
 430         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 431         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 432 };
 433
 434 static ssize_t
 435 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 436                   char *buf)
 437 {
 438         struct scsi_disk *sdkp = to_scsi_disk(dev);
 439
 440         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 441 }
 442
 443 static ssize_t
 444 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 445                    const char *buf, size_t count)
 446 {
 447         struct scsi_disk *sdkp = to_scsi_disk(dev);
 448         int mode;
 449
 450         if (!capable(CAP_SYS_ADMIN))
 451                 return -EACCES;
 452
 453         mode = sysfs_match_string(zeroing_mode, buf);
 454         if (mode < 0)
 455                 return -EINVAL;
 456
 457         sdkp->zeroing_mode = mode;
 458
 459         return count;
 460 }
 461 static DEVICE_ATTR_RW(zeroing_mode);
 462
 463 static ssize_t
 464 max_medium_access_timeouts_show(struct device *dev,
 465                                 struct device_attribute *attr, char *buf)
 466 {
 467         struct scsi_disk *sdkp = to_scsi_disk(dev);
 468
 469         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 470 }
 471
 472 static ssize_t
 473 max_medium_access_timeouts_store(struct device *dev,
 474                                  struct device_attribute *attr, const char *buf,
 475                                  size_t count)
 476 {
 477         struct scsi_disk *sdkp = to_scsi_disk(dev);
 478         int err;
 479
 480         if (!capable(CAP_SYS_ADMIN))
 481                 return -EACCES;
 482
 483         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 484
 485         return err ? err : count;
 486 }
 487 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 488
 489 static ssize_t
 490 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 491                            char *buf)
 492 {
 493         struct scsi_disk *sdkp = to_scsi_disk(dev);
 494
 495         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 496 }
 497
 498 static ssize_t
 499 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 500                             const char *buf, size_t count)
 501 {
 502         struct scsi_disk *sdkp = to_scsi_disk(dev);
 503         struct scsi_device *sdp = sdkp->device;
 504         unsigned long max;
 505         int err;
 506
 507         if (!capable(CAP_SYS_ADMIN))
 508                 return -EACCES;
 509
 510         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 511                 return -EINVAL;
 512
 513         err = kstrtoul(buf, 10, &max);
 514
 515         if (err)
 516                 return err;
 517
 518         if (max == 0)
 519                 sdp->no_write_same = 1;
 520         else if (max <= SD_MAX_WS16_BLOCKS) {
 521                 sdp->no_write_same = 0;
 522                 sdkp->max_ws_blocks = max;
 523         }
 524
 525         sd_config_write_same(sdkp);
 526
 527         return count;
 528 }
 529 static DEVICE_ATTR_RW(max_write_same_blocks);
 530
 531 static ssize_t
 532 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
 533 {
 534         struct scsi_disk *sdkp = to_scsi_disk(dev);
 535
 536         if (sdkp->device->type == TYPE_ZBC)
 537                 return sprintf(buf, "host-managed\n");
 538         if (sdkp->zoned == 1)
 539                 return sprintf(buf, "host-aware\n");
 540         if (sdkp->zoned == 2)
 541                 return sprintf(buf, "drive-managed\n");
 542         return sprintf(buf, "none\n");
 543 }
 544 static DEVICE_ATTR_RO(zoned_cap);
 545
 546 static ssize_t
 547 max_retries_store(struct device *dev, struct device_attribute *attr,
 548                   const char *buf, size_t count)
 549 {
 550         struct scsi_disk *sdkp = to_scsi_disk(dev);
 551         struct scsi_device *sdev = sdkp->device;
 552         int retries, err;
 553
 554         err = kstrtoint(buf, 10, &retries);
 555         if (err)
 556                 return err;
 557
 558         if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
 559                 sdkp->max_retries = retries;
 560                 return count;
 561         }
 562
 563         sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
 564                     SD_MAX_RETRIES);
 565         return -EINVAL;
 566 }
 567
 568 static ssize_t
 569 max_retries_show(struct device *dev, struct device_attribute *attr,
 570                  char *buf)
 571 {
 572         struct scsi_disk *sdkp = to_scsi_disk(dev);
 573
 574         return sprintf(buf, "%d\n", sdkp->max_retries);
 575 }
 576
 577 static DEVICE_ATTR_RW(max_retries);
 578
 579 static struct attribute *sd_disk_attrs[] = {
 580         &dev_attr_cache_type.attr,
 581         &dev_attr_FUA.attr,
 582         &dev_attr_allow_restart.attr,
 583         &dev_attr_manage_start_stop.attr,
 584         &dev_attr_protection_type.attr,
 585         &dev_attr_protection_mode.attr,
 586         &dev_attr_app_tag_own.attr,
 587         &dev_attr_thin_provisioning.attr,
 588         &dev_attr_provisioning_mode.attr,
 589         &dev_attr_zeroing_mode.attr,
 590         &dev_attr_max_write_same_blocks.attr,
 591         &dev_attr_max_medium_access_timeouts.attr,
 592         &dev_attr_zoned_cap.attr,
 593         &dev_attr_max_retries.attr,
 594         NULL,
 595 };
 596 ATTRIBUTE_GROUPS(sd_disk);
 597
 598 static struct class sd_disk_class = {
 599         .name           = "scsi_disk",
 600         .owner          = THIS_MODULE,
 601         .dev_release    = scsi_disk_release,
 602         .dev_groups     = sd_disk_groups,
 603 };
 604
 605 static const struct dev_pm_ops sd_pm_ops = {
 606         .suspend                = sd_suspend_system,
 607         .resume                 = sd_resume,
 608         .poweroff               = sd_suspend_system,
 609         .restore                = sd_resume,
 610         .runtime_suspend        = sd_suspend_runtime,
 611         .runtime_resume         = sd_resume,
 612 };
 613
 614 static struct scsi_driver sd_template = {
 615         .gendrv = {
 616                 .name           = "sd",
 617                 .owner          = THIS_MODULE,
 618                 .probe          = sd_probe,
 619                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
 620                 .remove         = sd_remove,
 621                 .shutdown       = sd_shutdown,
 622                 .pm             = &sd_pm_ops,
 623         },
 624         .rescan                 = sd_rescan,
 625         .init_command           = sd_init_command,
 626         .uninit_command         = sd_uninit_command,
 627         .done                   = sd_done,
 628         .eh_action              = sd_eh_action,
 629         .eh_reset               = sd_eh_reset,
 630 };
 631
 632 /*
 633  * Don't request a new module, as that could deadlock in multipath
 634  * environment.
 635  */
 636 static void sd_default_probe(dev_t devt)
 637 {
 638 }
 639
 640 /*
 641  * Device no to disk mapping:
 642  *
 643  *       major         disc2     disc  p1
 644  *   |............|.............|....|....| <- dev_t
 645  *    31        20 19          8 7  4 3  0
 646  *
 647  * Inside a major, we have 16k disks, however mapped non-
 648  * contiguously. The first 16 disks are for major0, the next
 649  * ones with major1, ... Disk 256 is for major0 again, disk 272
 650  * for major1, ...
 651  * As we stay compatible with our numbering scheme, we can reuse
 652  * the well-know SCSI majors 8, 65--71, 136--143.
 653  */
 654 static int sd_major(int major_idx)
 655 {
 656         switch (major_idx) {
 657         case 0:
 658                 return SCSI_DISK0_MAJOR;
 659         case 1 ... 7:
 660                 return SCSI_DISK1_MAJOR + major_idx - 1;
 661         case 8 ... 15:
 662                 return SCSI_DISK8_MAJOR + major_idx - 8;
 663         default:
 664                 BUG();
 665                 return 0;       /* shut up gcc */
 666         }
 667 }
 668
 669 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 670 {
 671         struct scsi_disk *sdkp = NULL;
 672
 673         mutex_lock(&sd_ref_mutex);
 674
 675         if (disk->private_data) {
 676                 sdkp = scsi_disk(disk);
 677                 if (scsi_device_get(sdkp->device) == 0)
 678                         get_device(&sdkp->dev);
 679                 else
 680                         sdkp = NULL;
 681         }
 682         mutex_unlock(&sd_ref_mutex);
 683         return sdkp;
 684 }
 685
 686 static void scsi_disk_put(struct scsi_disk *sdkp)
 687 {
 688         struct scsi_device *sdev = sdkp->device;
 689
 690         mutex_lock(&sd_ref_mutex);
 691         put_device(&sdkp->dev);
 692         scsi_device_put(sdev);
 693         mutex_unlock(&sd_ref_mutex);
 694 }
 695
 696 #ifdef CONFIG_BLK_SED_OPAL
 697 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 698                 size_t len, bool send)
 699 {
 700         struct scsi_disk *sdkp = data;
 701         struct scsi_device *sdev = sdkp->device;
 702         u8 cdb[12] = { 0, };
 703         int ret;
 704
 705         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 706         cdb[1] = secp;
 707         put_unaligned_be16(spsp, &cdb[2]);
 708         put_unaligned_be32(len, &cdb[6]);
 709
 710         ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 711                 buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
 712                 RQF_PM, NULL);
 713         return ret <= 0 ? ret : -EIO;
 714 }
 715 #endif /* CONFIG_BLK_SED_OPAL */
 716
 717 /*
 718  * Look up the DIX operation based on whether the command is read or
 719  * write and whether dix and dif are enabled.
 720  */
 721 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 722 {
 723         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 724         static const unsigned int ops[] = {     /* wrt dix dif */
 725                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 726                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 727                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 728                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 729                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 730                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 731                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 732                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 733         };
 734
 735         return ops[write << 2 | dix << 1 | dif];
 736 }
 737
 738 /*
 739  * Returns a mask of the protection flags that are valid for a given DIX
 740  * operation.
 741  */
 742 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 743 {
 744         static const unsigned int flag_mask[] = {
 745                 [SCSI_PROT_NORMAL]              = 0,
 746
 747                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 748                                                   SCSI_PROT_GUARD_CHECK |
 749                                                   SCSI_PROT_REF_CHECK |
 750                                                   SCSI_PROT_REF_INCREMENT,
 751
 752                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 753                                                   SCSI_PROT_IP_CHECKSUM,
 754
 755                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 756                                                   SCSI_PROT_GUARD_CHECK |
 757                                                   SCSI_PROT_REF_CHECK |
 758                                                   SCSI_PROT_REF_INCREMENT |
 759                                                   SCSI_PROT_IP_CHECKSUM,
 760
 761                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 762                                                   SCSI_PROT_REF_INCREMENT,
 763
 764                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 765                                                   SCSI_PROT_REF_CHECK |
 766                                                   SCSI_PROT_REF_INCREMENT |
 767                                                   SCSI_PROT_IP_CHECKSUM,
 768
 769                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 770                                                   SCSI_PROT_GUARD_CHECK |
 771                                                   SCSI_PROT_REF_CHECK |
 772                                                   SCSI_PROT_REF_INCREMENT |
 773                                                   SCSI_PROT_IP_CHECKSUM,
 774         };
 775
 776         return flag_mask[prot_op];
 777 }
 778
 779 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 780                                            unsigned int dix, unsigned int dif)
 781 {
 782         struct bio *bio = scmd->request->bio;
 783         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
 784         unsigned int protect = 0;
 785
 786         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 787                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 788                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 789
 790                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 791                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 792         }
 793
 794         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 795                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 796
 797                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 798                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 799         }
 800
 801         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 802                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 803
 804                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 805                         protect = 3 << 5;       /* Disable target PI checking */
 806                 else
 807                         protect = 1 << 5;       /* Enable target PI checking */
 808         }
 809
 810         scsi_set_prot_op(scmd, prot_op);
 811         scsi_set_prot_type(scmd, dif);
 812         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 813
 814         return protect;
 815 }
 816
 817 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 818 {
 819         struct request_queue *q = sdkp->disk->queue;
 820         unsigned int logical_block_size = sdkp->device->sector_size;
 821         unsigned int max_blocks = 0;
 822
 823         q->limits.discard_alignment =
 824                 sdkp->unmap_alignment * logical_block_size;
 825         q->limits.discard_granularity =
 826                 max(sdkp->physical_block_size,
 827                     sdkp->unmap_granularity * logical_block_size);
 828         sdkp->provisioning_mode = mode;
 829
 830         switch (mode) {
 831
 832         case SD_LBP_FULL:
 833         case SD_LBP_DISABLE:
 834                 blk_queue_max_discard_sectors(q, 0);
 835                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 836                 return;
 837
 838         case SD_LBP_UNMAP:
 839                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 840                                           (u32)SD_MAX_WS16_BLOCKS);
 841                 break;
 842
 843         case SD_LBP_WS16:
 844                 if (sdkp->device->unmap_limit_for_ws)
 845                         max_blocks = sdkp->max_unmap_blocks;
 846                 else
 847                         max_blocks = sdkp->max_ws_blocks;
 848
 849                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 850                 break;
 851
 852         case SD_LBP_WS10:
 853                 if (sdkp->device->unmap_limit_for_ws)
 854                         max_blocks = sdkp->max_unmap_blocks;
 855                 else
 856                         max_blocks = sdkp->max_ws_blocks;
 857
 858                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 859                 break;
 860
 861         case SD_LBP_ZERO:
 862                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 863                                           (u32)SD_MAX_WS10_BLOCKS);
 864                 break;
 865         }
 866
 867         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 868         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 869 }
 870
 871 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 872 {
 873         struct scsi_device *sdp = cmd->device;
 874         struct request *rq = cmd->request;
 875         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 876         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 877         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 878         unsigned int data_len = 24;
 879         char *buf;
 880
 881         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 882         if (!rq->special_vec.bv_page)
 883                 return BLK_STS_RESOURCE;
 884         clear_highpage(rq->special_vec.bv_page);
 885         rq->special_vec.bv_offset = 0;
 886         rq->special_vec.bv_len = data_len;
 887         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 888
 889         cmd->cmd_len = 10;
 890         cmd->cmnd[0] = UNMAP;
 891         cmd->cmnd[8] = 24;
 892
 893         buf = page_address(rq->special_vec.bv_page);
 894         put_unaligned_be16(6 + 16, &buf[0]);
 895         put_unaligned_be16(16, &buf[2]);
 896         put_unaligned_be64(lba, &buf[8]);
 897         put_unaligned_be32(nr_blocks, &buf[16]);
 898
 899         cmd->allowed = sdkp->max_retries;
 900         cmd->transfersize = data_len;
 901         rq->timeout = SD_TIMEOUT;
 902
 903         return scsi_alloc_sgtables(cmd);
 904 }
 905
 906 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 907                 bool unmap)
 908 {
 909         struct scsi_device *sdp = cmd->device;
 910         struct request *rq = cmd->request;
 911         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 912         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 913         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 914         u32 data_len = sdp->sector_size;
 915
 916         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 917         if (!rq->special_vec.bv_page)
 918                 return BLK_STS_RESOURCE;
 919         clear_highpage(rq->special_vec.bv_page);
 920         rq->special_vec.bv_offset = 0;
 921         rq->special_vec.bv_len = data_len;
 922         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 923
 924         cmd->cmd_len = 16;
 925         cmd->cmnd[0] = WRITE_SAME_16;
 926         if (unmap)
 927                 cmd->cmnd[1] = 0x8; /* UNMAP */
 928         put_unaligned_be64(lba, &cmd->cmnd[2]);
 929         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 930
 931         cmd->allowed = sdkp->max_retries;
 932         cmd->transfersize = data_len;
 933         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 934
 935         return scsi_alloc_sgtables(cmd);
 936 }
 937
 938 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 939                 bool unmap)
 940 {
 941         struct scsi_device *sdp = cmd->device;
 942         struct request *rq = cmd->request;
 943         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 944         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 945         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 946         u32 data_len = sdp->sector_size;
 947
 948         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 949         if (!rq->special_vec.bv_page)
 950                 return BLK_STS_RESOURCE;
 951         clear_highpage(rq->special_vec.bv_page);
 952         rq->special_vec.bv_offset = 0;
 953         rq->special_vec.bv_len = data_len;
 954         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 955
 956         cmd->cmd_len = 10;
 957         cmd->cmnd[0] = WRITE_SAME;
 958         if (unmap)
 959                 cmd->cmnd[1] = 0x8; /* UNMAP */
 960         put_unaligned_be32(lba, &cmd->cmnd[2]);
 961         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 962
 963         cmd->allowed = sdkp->max_retries;
 964         cmd->transfersize = data_len;
 965         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 966
 967         return scsi_alloc_sgtables(cmd);
 968 }
 969
 970 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 971 {
 972         struct request *rq = cmd->request;
 973         struct scsi_device *sdp = cmd->device;
 974         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 975         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 976         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 977
 978         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 979                 switch (sdkp->zeroing_mode) {
 980                 case SD_ZERO_WS16_UNMAP:
 981                         return sd_setup_write_same16_cmnd(cmd, true);
 982                 case SD_ZERO_WS10_UNMAP:
 983                         return sd_setup_write_same10_cmnd(cmd, true);
 984                 }
 985         }
 986
 987         if (sdp->no_write_same) {
 988                 rq->rq_flags |= RQF_QUIET;
 989                 return BLK_STS_TARGET;
 990         }
 991
 992         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 993                 return sd_setup_write_same16_cmnd(cmd, false);
 994
 995         return sd_setup_write_same10_cmnd(cmd, false);
 996 }
 997
 998 static void sd_config_write_same(struct scsi_disk *sdkp)
 999 {
1000         struct request_queue *q = sdkp->disk->queue;
1001         unsigned int logical_block_size = sdkp->device->sector_size;
1002
1003         if (sdkp->device->no_write_same) {
1004                 sdkp->max_ws_blocks = 0;
1005                 goto out;
1006         }
1007
1008         /* Some devices can not handle block counts above 0xffff despite
1009          * supporting WRITE SAME(16). Consequently we default to 64k
1010          * blocks per I/O unless the device explicitly advertises a
1011          * bigger limit.
1012          */
1013         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1014                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1015                                                    (u32)SD_MAX_WS16_BLOCKS);
1016         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1017                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1018                                                    (u32)SD_MAX_WS10_BLOCKS);
1019         else {
1020                 sdkp->device->no_write_same = 1;
1021                 sdkp->max_ws_blocks = 0;
1022         }
1023
1024         if (sdkp->lbprz && sdkp->lbpws)
1025                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1026         else if (sdkp->lbprz && sdkp->lbpws10)
1027                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1028         else if (sdkp->max_ws_blocks)
1029                 sdkp->zeroing_mode = SD_ZERO_WS;
1030         else
1031                 sdkp->zeroing_mode = SD_ZERO_WRITE;
1032
1033         if (sdkp->max_ws_blocks &&
1034             sdkp->physical_block_size > logical_block_size) {
1035                 /*
1036                  * Reporting a maximum number of blocks that is not aligned
1037                  * on the device physical size would cause a large write same
1038                  * request to be split into physically unaligned chunks by
1039                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1040                  * even if the caller of these functions took care to align the
1041                  * large request. So make sure the maximum reported is aligned
1042                  * to the device physical block size. This is only an optional
1043                  * optimization for regular disks, but this is mandatory to
1044                  * avoid failure of large write same requests directed at
1045                  * sequential write required zones of host-managed ZBC disks.
1046                  */
1047                 sdkp->max_ws_blocks =
1048                         round_down(sdkp->max_ws_blocks,
1049                                    bytes_to_logical(sdkp->device,
1050                                                     sdkp->physical_block_size));
1051         }
1052
1053 out:
1054         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1055                                          (logical_block_size >> 9));
1056         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1057                                          (logical_block_size >> 9));
1058 }
1059
1060 /**
1061  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1062  * @cmd: command to prepare
1063  *
1064  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1065  * the preference indicated by the target device.
1066  **/
1067 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1068 {
1069         struct request *rq = cmd->request;
1070         struct scsi_device *sdp = cmd->device;
1071         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1072         struct bio *bio = rq->bio;
1073         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1074         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1075         blk_status_t ret;
1076
1077         if (sdkp->device->no_write_same)
1078                 return BLK_STS_TARGET;
1079
1080         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1081
1082         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1083
1084         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1085                 cmd->cmd_len = 16;
1086                 cmd->cmnd[0] = WRITE_SAME_16;
1087                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1088                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1089         } else {
1090                 cmd->cmd_len = 10;
1091                 cmd->cmnd[0] = WRITE_SAME;
1092                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1093                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1094         }
1095
1096         cmd->transfersize = sdp->sector_size;
1097         cmd->allowed = sdkp->max_retries;
1098
1099         /*
1100          * For WRITE SAME the data transferred via the DATA OUT buffer is
1101          * different from the amount of data actually written to the target.
1102          *
1103          * We set up __data_len to the amount of data transferred via the
1104          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1105          * to transfer a single sector of data first, but then reset it to
1106          * the amount of data to be written right after so that the I/O path
1107          * knows how much to actually write.
1108          */
1109         rq->__data_len = sdp->sector_size;
1110         ret = scsi_alloc_sgtables(cmd);
1111         rq->__data_len = blk_rq_bytes(rq);
1112
1113         return ret;
1114 }
1115
1116 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1117 {
1118         struct request *rq = cmd->request;
1119         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1120
1121         /* flush requests don't perform I/O, zero the S/G table */
1122         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1123
1124         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1125         cmd->cmd_len = 10;
1126         cmd->transfersize = 0;
1127         cmd->allowed = sdkp->max_retries;
1128
1129         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1130         return BLK_STS_OK;
1131 }
1132
1133 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1134                                        sector_t lba, unsigned int nr_blocks,
1135                                        unsigned char flags)
1136 {
1137         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1138         if (unlikely(cmd->cmnd == NULL))
1139                 return BLK_STS_RESOURCE;
1140
1141         cmd->cmd_len = SD_EXT_CDB_SIZE;
1142         memset(cmd->cmnd, 0, cmd->cmd_len);
1143
1144         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1145         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1146         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1147         cmd->cmnd[10] = flags;
1148         put_unaligned_be64(lba, &cmd->cmnd[12]);
1149         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1150         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1151
1152         return BLK_STS_OK;
1153 }
1154
1155 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1156                                        sector_t lba, unsigned int nr_blocks,
1157                                        unsigned char flags)
1158 {
1159         cmd->cmd_len  = 16;
1160         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1161         cmd->cmnd[1]  = flags;
1162         cmd->cmnd[14] = 0;
1163         cmd->cmnd[15] = 0;
1164         put_unaligned_be64(lba, &cmd->cmnd[2]);
1165         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1166
1167         return BLK_STS_OK;
1168 }
1169
1170 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1171                                        sector_t lba, unsigned int nr_blocks,
1172                                        unsigned char flags)
1173 {
1174         cmd->cmd_len = 10;
1175         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1176         cmd->cmnd[1] = flags;
1177         cmd->cmnd[6] = 0;
1178         cmd->cmnd[9] = 0;
1179         put_unaligned_be32(lba, &cmd->cmnd[2]);
1180         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1181
1182         return BLK_STS_OK;
1183 }
1184
1185 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1186                                       sector_t lba, unsigned int nr_blocks,
1187                                       unsigned char flags)
1188 {
1189         /* Avoid that 0 blocks gets translated into 256 blocks. */
1190         if (WARN_ON_ONCE(nr_blocks == 0))
1191                 return BLK_STS_IOERR;
1192
1193         if (unlikely(flags & 0x8)) {
1194                 /*
1195                  * This happens only if this drive failed 10byte rw
1196                  * command with ILLEGAL_REQUEST during operation and
1197                  * thus turned off use_10_for_rw.
1198                  */
1199                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1200                 return BLK_STS_IOERR;
1201         }
1202
1203         cmd->cmd_len = 6;
1204         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1205         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1206         cmd->cmnd[2] = (lba >> 8) & 0xff;
1207         cmd->cmnd[3] = lba & 0xff;
1208         cmd->cmnd[4] = nr_blocks;
1209         cmd->cmnd[5] = 0;
1210
1211         return BLK_STS_OK;
1212 }
1213
1214 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1215 {
1216         struct request *rq = cmd->request;
1217         struct scsi_device *sdp = cmd->device;
1218         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1219         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1220         sector_t threshold;
1221         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1222         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1223         bool write = rq_data_dir(rq) == WRITE;
1224         unsigned char protect, fua;
1225         blk_status_t ret;
1226         unsigned int dif;
1227         bool dix;
1228
1229         ret = scsi_alloc_sgtables(cmd);
1230         if (ret != BLK_STS_OK)
1231                 return ret;
1232
1233         ret = BLK_STS_IOERR;
1234         if (!scsi_device_online(sdp) || sdp->changed) {
1235                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1236                 goto fail;
1237         }
1238
1239         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1240                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1241                 goto fail;
1242         }
1243
1244         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1245                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1246                 goto fail;
1247         }
1248
1249         /*
1250          * Some SD card readers can't handle accesses which touch the
1251          * last one or two logical blocks. Split accesses as needed.
1252          */
1253         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1254
1255         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1256                 if (lba < threshold) {
1257                         /* Access up to the threshold but not beyond */
1258                         nr_blocks = threshold - lba;
1259                 } else {
1260                         /* Access only a single logical block */
1261                         nr_blocks = 1;
1262                 }
1263         }
1264
1265         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1266                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1267                 if (ret)
1268                         goto fail;
1269         }
1270
1271         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1272         dix = scsi_prot_sg_count(cmd);
1273         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1274
1275         if (dif || dix)
1276                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1277         else
1278                 protect = 0;
1279
1280         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1281                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1282                                          protect | fua);
1283         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1284                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1285                                          protect | fua);
1286         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1287                    sdp->use_10_for_rw || protect) {
1288                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1289                                          protect | fua);
1290         } else {
1291                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1292                                         protect | fua);
1293         }
1294
1295         if (unlikely(ret != BLK_STS_OK))
1296                 goto fail;
1297
1298         /*
1299          * We shouldn't disconnect in the middle of a sector, so with a dumb
1300          * host adapter, it's safe to assume that we can at least transfer
1301          * this many bytes between each connect / disconnect.
1302          */
1303         cmd->transfersize = sdp->sector_size;
1304         cmd->underflow = nr_blocks << 9;
1305         cmd->allowed = sdkp->max_retries;
1306         cmd->sdb.length = nr_blocks * sdp->sector_size;
1307
1308         SCSI_LOG_HLQUEUE(1,
1309                          scmd_printk(KERN_INFO, cmd,
1310                                      "%s: block=%llu, count=%d\n", __func__,
1311                                      (unsigned long long)blk_rq_pos(rq),
1312                                      blk_rq_sectors(rq)));
1313         SCSI_LOG_HLQUEUE(2,
1314                          scmd_printk(KERN_INFO, cmd,
1315                                      "%s %d/%u 512 byte blocks.\n",
1316                                      write ? "writing" : "reading", nr_blocks,
1317                                      blk_rq_sectors(rq)));
1318
1319         /*
1320          * This indicates that the command is ready from our end to be queued.
1321          */
1322         return BLK_STS_OK;
1323 fail:
1324         scsi_free_sgtables(cmd);
1325         return ret;
1326 }
1327
1328 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1329 {
1330         struct request *rq = cmd->request;
1331
1332         switch (req_op(rq)) {
1333         case REQ_OP_DISCARD:
1334                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1335                 case SD_LBP_UNMAP:
1336                         return sd_setup_unmap_cmnd(cmd);
1337                 case SD_LBP_WS16:
1338                         return sd_setup_write_same16_cmnd(cmd, true);
1339                 case SD_LBP_WS10:
1340                         return sd_setup_write_same10_cmnd(cmd, true);
1341                 case SD_LBP_ZERO:
1342                         return sd_setup_write_same10_cmnd(cmd, false);
1343                 default:
1344                         return BLK_STS_TARGET;
1345                 }
1346         case REQ_OP_WRITE_ZEROES:
1347                 return sd_setup_write_zeroes_cmnd(cmd);
1348         case REQ_OP_WRITE_SAME:
1349                 return sd_setup_write_same_cmnd(cmd);
1350         case REQ_OP_FLUSH:
1351                 return sd_setup_flush_cmnd(cmd);
1352         case REQ_OP_READ:
1353         case REQ_OP_WRITE:
1354         case REQ_OP_ZONE_APPEND:
1355                 return sd_setup_read_write_cmnd(cmd);
1356         case REQ_OP_ZONE_RESET:
1357                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358                                                    false);
1359         case REQ_OP_ZONE_RESET_ALL:
1360                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1361                                                    true);
1362         case REQ_OP_ZONE_OPEN:
1363                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1364         case REQ_OP_ZONE_CLOSE:
1365                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1366         case REQ_OP_ZONE_FINISH:
1367                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1368         default:
1369                 WARN_ON_ONCE(1);
1370                 return BLK_STS_NOTSUPP;
1371         }
1372 }
1373
1374 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1375 {
1376         struct request *rq = SCpnt->request;
1377         u8 *cmnd;
1378
1379         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1380                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1381
1382         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1383                 cmnd = SCpnt->cmnd;
1384                 SCpnt->cmnd = NULL;
1385                 SCpnt->cmd_len = 0;
1386                 mempool_free(cmnd, sd_cdb_pool);
1387         }
1388 }
1389
1390 static bool sd_need_revalidate(struct block_device *bdev,
1391                 struct scsi_disk *sdkp)
1392 {
1393         if (sdkp->device->removable || sdkp->write_prot) {
1394                 if (bdev_check_media_change(bdev))
1395                         return true;
1396         }
1397
1398         /*
1399          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1400          * nothing to do with partitions, BLKRRPART is used to force a full
1401          * revalidate after things like a format for historical reasons.
1402          */
1403         return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1404 }
1405
1406 /**
1407  *      sd_open - open a scsi disk device
1408  *      @bdev: Block device of the scsi disk to open
1409  *      @mode: FMODE_* mask
1410  *
1411  *      Returns 0 if successful. Returns a negated errno value in case
1412  *      of error.
1413  *
1414  *      Note: This can be called from a user context (e.g. fsck(1) )
1415  *      or from within the kernel (e.g. as a result of a mount(1) ).
1416  *      In the latter case @inode and @filp carry an abridged amount
1417  *      of information as noted above.
1418  *
1419  *      Locking: called with bdev->bd_disk->open_mutex held.
1420  **/
1421 static int sd_open(struct block_device *bdev, fmode_t mode)
1422 {
1423         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1424         struct scsi_device *sdev;
1425         int retval;
1426
1427         if (!sdkp)
1428                 return -ENXIO;
1429
1430         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1431
1432         sdev = sdkp->device;
1433
1434         /*
1435          * If the device is in error recovery, wait until it is done.
1436          * If the device is offline, then disallow any access to it.
1437          */
1438         retval = -ENXIO;
1439         if (!scsi_block_when_processing_errors(sdev))
1440                 goto error_out;
1441
1442         if (sd_need_revalidate(bdev, sdkp))
1443                 sd_revalidate_disk(bdev->bd_disk);
1444
1445         /*
1446          * If the drive is empty, just let the open fail.
1447          */
1448         retval = -ENOMEDIUM;
1449         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1450                 goto error_out;
1451
1452         /*
1453          * If the device has the write protect tab set, have the open fail
1454          * if the user expects to be able to write to the thing.
1455          */
1456         retval = -EROFS;
1457         if (sdkp->write_prot && (mode & FMODE_WRITE))
1458                 goto error_out;
1459
1460         /*
1461          * It is possible that the disk changing stuff resulted in
1462          * the device being taken offline.  If this is the case,
1463          * report this to the user, and don't pretend that the
1464          * open actually succeeded.
1465          */
1466         retval = -ENXIO;
1467         if (!scsi_device_online(sdev))
1468                 goto error_out;
1469
1470         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1471                 if (scsi_block_when_processing_errors(sdev))
1472                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1473         }
1474
1475         return 0;
1476
1477 error_out:
1478         scsi_disk_put(sdkp);
1479         return retval;
1480 }
1481
1482 /**
1483  *      sd_release - invoked when the (last) close(2) is called on this
1484  *      scsi disk.
1485  *      @disk: disk to release
1486  *      @mode: FMODE_* mask
1487  *
1488  *      Returns 0.
1489  *
1490  *      Note: may block (uninterruptible) if error recovery is underway
1491  *      on this disk.
1492  *
1493  *      Locking: called with bdev->bd_disk->open_mutex held.
1494  **/
1495 static void sd_release(struct gendisk *disk, fmode_t mode)
1496 {
1497         struct scsi_disk *sdkp = scsi_disk(disk);
1498         struct scsi_device *sdev = sdkp->device;
1499
1500         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1501
1502         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1503                 if (scsi_block_when_processing_errors(sdev))
1504                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1505         }
1506
1507         scsi_disk_put(sdkp);
1508 }
1509
1510 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1511 {
1512         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1513         struct scsi_device *sdp = sdkp->device;
1514         struct Scsi_Host *host = sdp->host;
1515         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1516         int diskinfo[4];
1517
1518         /* default to most commonly used values */
1519         diskinfo[0] = 0x40;     /* 1 << 6 */
1520         diskinfo[1] = 0x20;     /* 1 << 5 */
1521         diskinfo[2] = capacity >> 11;
1522
1523         /* override with calculated, extended default, or driver values */
1524         if (host->hostt->bios_param)
1525                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1526         else
1527                 scsicam_bios_param(bdev, capacity, diskinfo);
1528
1529         geo->heads = diskinfo[0];
1530         geo->sectors = diskinfo[1];
1531         geo->cylinders = diskinfo[2];
1532         return 0;
1533 }
1534
1535 /**
1536  *      sd_ioctl_common - process an ioctl
1537  *      @bdev: target block device
1538  *      @mode: FMODE_* mask
1539  *      @cmd: ioctl command number
1540  *      @p: this is third argument given to ioctl(2) system call.
1541  *      Often contains a pointer.
1542  *
1543  *      Returns 0 if successful (some ioctls return positive numbers on
1544  *      success as well). Returns a negated errno value in case of error.
1545  *
1546  *      Note: most ioctls are forward onto the block subsystem or further
1547  *      down in the scsi subsystem.
1548  **/
1549 static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1550                            unsigned int cmd, void __user *p)
1551 {
1552         struct gendisk *disk = bdev->bd_disk;
1553         struct scsi_disk *sdkp = scsi_disk(disk);
1554         struct scsi_device *sdp = sdkp->device;
1555         int error;
1556
1557         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1558                                     "cmd=0x%x\n", disk->disk_name, cmd));
1559
1560         error = scsi_verify_blk_ioctl(bdev, cmd);
1561         if (error < 0)
1562                 return error;
1563
1564         /*
1565          * If we are in the middle of error recovery, don't let anyone
1566          * else try and use this device.  Also, if error recovery fails, it
1567          * may try and take the device offline, in which case all further
1568          * access to the device is prohibited.
1569          */
1570         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1571                         (mode & FMODE_NDELAY) != 0);
1572         if (error)
1573                 goto out;
1574
1575         if (is_sed_ioctl(cmd))
1576                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1577
1578         /*
1579          * Send SCSI addressing ioctls directly to mid level, send other
1580          * ioctls to block level and then onto mid level if they can't be
1581          * resolved.
1582          */
1583         switch (cmd) {
1584                 case SCSI_IOCTL_GET_IDLUN:
1585                 case SCSI_IOCTL_GET_BUS_NUMBER:
1586                         error = scsi_ioctl(sdp, cmd, p);
1587                         break;
1588                 default:
1589                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1590                         break;
1591         }
1592 out:
1593         return error;
1594 }
1595
1596 static void set_media_not_present(struct scsi_disk *sdkp)
1597 {
1598         if (sdkp->media_present)
1599                 sdkp->device->changed = 1;
1600
1601         if (sdkp->device->removable) {
1602                 sdkp->media_present = 0;
1603                 sdkp->capacity = 0;
1604         }
1605 }
1606
1607 static int media_not_present(struct scsi_disk *sdkp,
1608                              struct scsi_sense_hdr *sshdr)
1609 {
1610         if (!scsi_sense_valid(sshdr))
1611                 return 0;
1612
1613         /* not invoked for commands that could return deferred errors */
1614         switch (sshdr->sense_key) {
1615         case UNIT_ATTENTION:
1616         case NOT_READY:
1617                 /* medium not present */
1618                 if (sshdr->asc == 0x3A) {
1619                         set_media_not_present(sdkp);
1620                         return 1;
1621                 }
1622         }
1623         return 0;
1624 }
1625
1626 /**
1627  *      sd_check_events - check media events
1628  *      @disk: kernel device descriptor
1629  *      @clearing: disk events currently being cleared
1630  *
1631  *      Returns mask of DISK_EVENT_*.
1632  *
1633  *      Note: this function is invoked from the block subsystem.
1634  **/
1635 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1636 {
1637         struct scsi_disk *sdkp = scsi_disk_get(disk);
1638         struct scsi_device *sdp;
1639         int retval;
1640         bool disk_changed;
1641
1642         if (!sdkp)
1643                 return 0;
1644
1645         sdp = sdkp->device;
1646         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1647
1648         /*
1649          * If the device is offline, don't send any commands - just pretend as
1650          * if the command failed.  If the device ever comes back online, we
1651          * can deal with it then.  It is only because of unrecoverable errors
1652          * that we would ever take a device offline in the first place.
1653          */
1654         if (!scsi_device_online(sdp)) {
1655                 set_media_not_present(sdkp);
1656                 goto out;
1657         }
1658
1659         /*
1660          * Using TEST_UNIT_READY enables differentiation between drive with
1661          * no cartridge loaded - NOT READY, drive with changed cartridge -
1662          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1663          *
1664          * Drives that auto spin down. eg iomega jaz 1G, will be started
1665          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1666          * sd_revalidate() is called.
1667          */
1668         if (scsi_block_when_processing_errors(sdp)) {
1669                 struct scsi_sense_hdr sshdr = { 0, };
1670
1671                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1672                                               &sshdr);
1673
1674                 /* failed to execute TUR, assume media not present */
1675                 if (retval < 0 || host_byte(retval)) {
1676                         set_media_not_present(sdkp);
1677                         goto out;
1678                 }
1679
1680                 if (media_not_present(sdkp, &sshdr))
1681                         goto out;
1682         }
1683
1684         /*
1685          * For removable scsi disk we have to recognise the presence
1686          * of a disk in the drive.
1687          */
1688         if (!sdkp->media_present)
1689                 sdp->changed = 1;
1690         sdkp->media_present = 1;
1691 out:
1692         /*
1693          * sdp->changed is set under the following conditions:
1694          *
1695          *      Medium present state has changed in either direction.
1696          *      Device has indicated UNIT_ATTENTION.
1697          */
1698         disk_changed = sdp->changed;
1699         sdp->changed = 0;
1700         scsi_disk_put(sdkp);
1701         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1702 }
1703
1704 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1705 {
1706         int retries, res;
1707         struct scsi_device *sdp = sdkp->device;
1708         const int timeout = sdp->request_queue->rq_timeout
1709                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1710         struct scsi_sense_hdr my_sshdr;
1711
1712         if (!scsi_device_online(sdp))
1713                 return -ENODEV;
1714
1715         /* caller might not be interested in sense, but we need it */
1716         if (!sshdr)
1717                 sshdr = &my_sshdr;
1718
1719         for (retries = 3; retries > 0; --retries) {
1720                 unsigned char cmd[10] = { 0 };
1721
1722                 cmd[0] = SYNCHRONIZE_CACHE;
1723                 /*
1724                  * Leave the rest of the command zero to indicate
1725                  * flush everything.
1726                  */
1727                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1728                                 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1729                 if (res == 0)
1730                         break;
1731         }
1732
1733         if (res) {
1734                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1735
1736                 if (res < 0)
1737                         return res;
1738
1739                 if (scsi_status_is_check_condition(res) &&
1740                     scsi_sense_valid(sshdr)) {
1741                         sd_print_sense_hdr(sdkp, sshdr);
1742
1743                         /* we need to evaluate the error return  */
1744                         if (sshdr->asc == 0x3a ||       /* medium not present */
1745                             sshdr->asc == 0x20 ||       /* invalid command */
1746                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1747                                 /* this is no error here */
1748                                 return 0;
1749                 }
1750
1751                 switch (host_byte(res)) {
1752                 /* ignore errors due to racing a disconnection */
1753                 case DID_BAD_TARGET:
1754                 case DID_NO_CONNECT:
1755                         return 0;
1756                 /* signal the upper layer it might try again */
1757                 case DID_BUS_BUSY:
1758                 case DID_IMM_RETRY:
1759                 case DID_REQUEUE:
1760                 case DID_SOFT_ERROR:
1761                         return -EBUSY;
1762                 default:
1763                         return -EIO;
1764                 }
1765         }
1766         return 0;
1767 }
1768
1769 static void sd_rescan(struct device *dev)
1770 {
1771         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1772
1773         sd_revalidate_disk(sdkp->disk);
1774 }
1775
1776 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1777                     unsigned int cmd, unsigned long arg)
1778 {
1779         void __user *p = (void __user *)arg;
1780         int ret;
1781
1782         ret = sd_ioctl_common(bdev, mode, cmd, p);
1783         if (ret != -ENOTTY)
1784                 return ret;
1785
1786         return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1787 }
1788
1789 #ifdef CONFIG_COMPAT
1790 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1791                            unsigned int cmd, unsigned long arg)
1792 {
1793         void __user *p = compat_ptr(arg);
1794         int ret;
1795
1796         ret = sd_ioctl_common(bdev, mode, cmd, p);
1797         if (ret != -ENOTTY)
1798                 return ret;
1799
1800         return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1801 }
1802 #endif
1803
1804 static char sd_pr_type(enum pr_type type)
1805 {
1806         switch (type) {
1807         case PR_WRITE_EXCLUSIVE:
1808                 return 0x01;
1809         case PR_EXCLUSIVE_ACCESS:
1810                 return 0x03;
1811         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1812                 return 0x05;
1813         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1814                 return 0x06;
1815         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1816                 return 0x07;
1817         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1818                 return 0x08;
1819         default:
1820                 return 0;
1821         }
1822 };
1823
1824 static int sd_pr_command(struct block_device *bdev, u8 sa,
1825                 u64 key, u64 sa_key, u8 type, u8 flags)
1826 {
1827         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1828         struct scsi_device *sdev = sdkp->device;
1829         struct scsi_sense_hdr sshdr;
1830         int result;
1831         u8 cmd[16] = { 0, };
1832         u8 data[24] = { 0, };
1833
1834         cmd[0] = PERSISTENT_RESERVE_OUT;
1835         cmd[1] = sa;
1836         cmd[2] = type;
1837         put_unaligned_be32(sizeof(data), &cmd[5]);
1838
1839         put_unaligned_be64(key, &data[0]);
1840         put_unaligned_be64(sa_key, &data[8]);
1841         data[20] = flags;
1842
1843         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1844                         &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1845
1846         if (scsi_status_is_check_condition(result) &&
1847             scsi_sense_valid(&sshdr)) {
1848                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1849                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1850         }
1851
1852         return result;
1853 }
1854
1855 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1856                 u32 flags)
1857 {
1858         if (flags & ~PR_FL_IGNORE_KEY)
1859                 return -EOPNOTSUPP;
1860         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1861                         old_key, new_key, 0,
1862                         (1 << 0) /* APTPL */);
1863 }
1864
1865 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1866                 u32 flags)
1867 {
1868         if (flags)
1869                 return -EOPNOTSUPP;
1870         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1871 }
1872
1873 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1874 {
1875         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1876 }
1877
1878 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1879                 enum pr_type type, bool abort)
1880 {
1881         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1882                              sd_pr_type(type), 0);
1883 }
1884
1885 static int sd_pr_clear(struct block_device *bdev, u64 key)
1886 {
1887         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1888 }
1889
1890 static const struct pr_ops sd_pr_ops = {
1891         .pr_register    = sd_pr_register,
1892         .pr_reserve     = sd_pr_reserve,
1893         .pr_release     = sd_pr_release,
1894         .pr_preempt     = sd_pr_preempt,
1895         .pr_clear       = sd_pr_clear,
1896 };
1897
1898 static const struct block_device_operations sd_fops = {
1899         .owner                  = THIS_MODULE,
1900         .open                   = sd_open,
1901         .release                = sd_release,
1902         .ioctl                  = sd_ioctl,
1903         .getgeo                 = sd_getgeo,
1904 #ifdef CONFIG_COMPAT
1905         .compat_ioctl           = sd_compat_ioctl,
1906 #endif
1907         .check_events           = sd_check_events,
1908         .unlock_native_capacity = sd_unlock_native_capacity,
1909         .report_zones           = sd_zbc_report_zones,
1910         .pr_ops                 = &sd_pr_ops,
1911 };
1912
1913 /**
1914  *      sd_eh_reset - reset error handling callback
1915  *      @scmd:          sd-issued command that has failed
1916  *
1917  *      This function is called by the SCSI midlayer before starting
1918  *      SCSI EH. When counting medium access failures we have to be
1919  *      careful to register it only only once per device and SCSI EH run;
1920  *      there might be several timed out commands which will cause the
1921  *      'max_medium_access_timeouts' counter to trigger after the first
1922  *      SCSI EH run already and set the device to offline.
1923  *      So this function resets the internal counter before starting SCSI EH.
1924  **/
1925 static void sd_eh_reset(struct scsi_cmnd *scmd)
1926 {
1927         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1928
1929         /* New SCSI EH run, reset gate variable */
1930         sdkp->ignore_medium_access_errors = false;
1931 }
1932
1933 /**
1934  *      sd_eh_action - error handling callback
1935  *      @scmd:          sd-issued command that has failed
1936  *      @eh_disp:       The recovery disposition suggested by the midlayer
1937  *
1938  *      This function is called by the SCSI midlayer upon completion of an
1939  *      error test command (currently TEST UNIT READY). The result of sending
1940  *      the eh command is passed in eh_disp.  We're looking for devices that
1941  *      fail medium access commands but are OK with non access commands like
1942  *      test unit ready (so wrongly see the device as having a successful
1943  *      recovery)
1944  **/
1945 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1946 {
1947         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1948         struct scsi_device *sdev = scmd->device;
1949
1950         if (!scsi_device_online(sdev) ||
1951             !scsi_medium_access_command(scmd) ||
1952             host_byte(scmd->result) != DID_TIME_OUT ||
1953             eh_disp != SUCCESS)
1954                 return eh_disp;
1955
1956         /*
1957          * The device has timed out executing a medium access command.
1958          * However, the TEST UNIT READY command sent during error
1959          * handling completed successfully. Either the device is in the
1960          * process of recovering or has it suffered an internal failure
1961          * that prevents access to the storage medium.
1962          */
1963         if (!sdkp->ignore_medium_access_errors) {
1964                 sdkp->medium_access_timed_out++;
1965                 sdkp->ignore_medium_access_errors = true;
1966         }
1967
1968         /*
1969          * If the device keeps failing read/write commands but TEST UNIT
1970          * READY always completes successfully we assume that medium
1971          * access is no longer possible and take the device offline.
1972          */
1973         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1974                 scmd_printk(KERN_ERR, scmd,
1975                             "Medium access timeout failure. Offlining disk!\n");
1976                 mutex_lock(&sdev->state_mutex);
1977                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1978                 mutex_unlock(&sdev->state_mutex);
1979
1980                 return SUCCESS;
1981         }
1982
1983         return eh_disp;
1984 }
1985
1986 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1987 {
1988         struct request *req = scmd->request;
1989         struct scsi_device *sdev = scmd->device;
1990         unsigned int transferred, good_bytes;
1991         u64 start_lba, end_lba, bad_lba;
1992
1993         /*
1994          * Some commands have a payload smaller than the device logical
1995          * block size (e.g. INQUIRY on a 4K disk).
1996          */
1997         if (scsi_bufflen(scmd) <= sdev->sector_size)
1998                 return 0;
1999
2000         /* Check if we have a 'bad_lba' information */
2001         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2002                                      SCSI_SENSE_BUFFERSIZE,
2003                                      &bad_lba))
2004                 return 0;
2005
2006         /*
2007          * If the bad lba was reported incorrectly, we have no idea where
2008          * the error is.
2009          */
2010         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2011         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2012         if (bad_lba < start_lba || bad_lba >= end_lba)
2013                 return 0;
2014
2015         /*
2016          * resid is optional but mostly filled in.  When it's unused,
2017          * its value is zero, so we assume the whole buffer transferred
2018          */
2019         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2020
2021         /* This computation should always be done in terms of the
2022          * resolution of the device's medium.
2023          */
2024         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2025
2026         return min(good_bytes, transferred);
2027 }
2028
2029 /**
2030  *      sd_done - bottom half handler: called when the lower level
2031  *      driver has completed (successfully or otherwise) a scsi command.
2032  *      @SCpnt: mid-level's per command structure.
2033  *
2034  *      Note: potentially run from within an ISR. Must not block.
2035  **/
2036 static int sd_done(struct scsi_cmnd *SCpnt)
2037 {
2038         int result = SCpnt->result;
2039         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2040         unsigned int sector_size = SCpnt->device->sector_size;
2041         unsigned int resid;
2042         struct scsi_sense_hdr sshdr;
2043         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
2044         struct request *req = SCpnt->request;
2045         int sense_valid = 0;
2046         int sense_deferred = 0;
2047
2048         switch (req_op(req)) {
2049         case REQ_OP_DISCARD:
2050         case REQ_OP_WRITE_ZEROES:
2051         case REQ_OP_WRITE_SAME:
2052         case REQ_OP_ZONE_RESET:
2053         case REQ_OP_ZONE_RESET_ALL:
2054         case REQ_OP_ZONE_OPEN:
2055         case REQ_OP_ZONE_CLOSE:
2056         case REQ_OP_ZONE_FINISH:
2057                 if (!result) {
2058                         good_bytes = blk_rq_bytes(req);
2059                         scsi_set_resid(SCpnt, 0);
2060                 } else {
2061                         good_bytes = 0;
2062                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
2063                 }
2064                 break;
2065         default:
2066                 /*
2067                  * In case of bogus fw or device, we could end up having
2068                  * an unaligned partial completion. Check this here and force
2069                  * alignment.
2070                  */
2071                 resid = scsi_get_resid(SCpnt);
2072                 if (resid & (sector_size - 1)) {
2073                         sd_printk(KERN_INFO, sdkp,
2074                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2075                                 resid, sector_size);
2076                         scsi_print_command(SCpnt);
2077                         resid = min(scsi_bufflen(SCpnt),
2078                                     round_up(resid, sector_size));
2079                         scsi_set_resid(SCpnt, resid);
2080                 }
2081         }
2082
2083         if (result) {
2084                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2085                 if (sense_valid)
2086                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2087         }
2088         sdkp->medium_access_timed_out = 0;
2089
2090         if (!scsi_status_is_check_condition(result) &&
2091             (!sense_valid || sense_deferred))
2092                 goto out;
2093
2094         switch (sshdr.sense_key) {
2095         case HARDWARE_ERROR:
2096         case MEDIUM_ERROR:
2097                 good_bytes = sd_completed_bytes(SCpnt);
2098                 break;
2099         case RECOVERED_ERROR:
2100                 good_bytes = scsi_bufflen(SCpnt);
2101                 break;
2102         case NO_SENSE:
2103                 /* This indicates a false check condition, so ignore it.  An
2104                  * unknown amount of data was transferred so treat it as an
2105                  * error.
2106                  */
2107                 SCpnt->result = 0;
2108                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2109                 break;
2110         case ABORTED_COMMAND:
2111                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2112                         good_bytes = sd_completed_bytes(SCpnt);
2113                 break;
2114         case ILLEGAL_REQUEST:
2115                 switch (sshdr.asc) {
2116                 case 0x10:      /* DIX: Host detected corruption */
2117                         good_bytes = sd_completed_bytes(SCpnt);
2118                         break;
2119                 case 0x20:      /* INVALID COMMAND OPCODE */
2120                 case 0x24:      /* INVALID FIELD IN CDB */
2121                         switch (SCpnt->cmnd[0]) {
2122                         case UNMAP:
2123                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2124                                 break;
2125                         case WRITE_SAME_16:
2126                         case WRITE_SAME:
2127                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2128                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2129                                 } else {
2130                                         sdkp->device->no_write_same = 1;
2131                                         sd_config_write_same(sdkp);
2132                                         req->rq_flags |= RQF_QUIET;
2133                                 }
2134                                 break;
2135                         }
2136                 }
2137                 break;
2138         default:
2139                 break;
2140         }
2141
2142  out:
2143         if (sd_is_zoned(sdkp))
2144                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2145
2146         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2147                                            "sd_done: completed %d of %d bytes\n",
2148                                            good_bytes, scsi_bufflen(SCpnt)));
2149
2150         return good_bytes;
2151 }
2152
2153 /*
2154  * spinup disk - called only in sd_revalidate_disk()
2155  */
2156 static void
2157 sd_spinup_disk(struct scsi_disk *sdkp)
2158 {
2159         unsigned char cmd[10];
2160         unsigned long spintime_expire = 0;
2161         int retries, spintime;
2162         unsigned int the_result;
2163         struct scsi_sense_hdr sshdr;
2164         int sense_valid = 0;
2165
2166         spintime = 0;
2167
2168         /* Spin up drives, as required.  Only do this at boot time */
2169         /* Spinup needs to be done for module loads too. */
2170         do {
2171                 retries = 0;
2172
2173                 do {
2174                         cmd[0] = TEST_UNIT_READY;
2175                         memset((void *) &cmd[1], 0, 9);
2176
2177                         the_result = scsi_execute_req(sdkp->device, cmd,
2178                                                       DMA_NONE, NULL, 0,
2179                                                       &sshdr, SD_TIMEOUT,
2180                                                       sdkp->max_retries, NULL);
2181
2182                         /*
2183                          * If the drive has indicated to us that it
2184                          * doesn't have any media in it, don't bother
2185                          * with any more polling.
2186                          */
2187                         if (media_not_present(sdkp, &sshdr))
2188                                 return;
2189
2190                         if (the_result)
2191                                 sense_valid = scsi_sense_valid(&sshdr);
2192                         retries++;
2193                 } while (retries < 3 &&
2194                          (!scsi_status_is_good(the_result) ||
2195                           (scsi_status_is_check_condition(the_result) &&
2196                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2197
2198                 if (!scsi_status_is_check_condition(the_result)) {
2199                         /* no sense, TUR either succeeded or failed
2200                          * with a status error */
2201                         if(!spintime && !scsi_status_is_good(the_result)) {
2202                                 sd_print_result(sdkp, "Test Unit Ready failed",
2203                                                 the_result);
2204                         }
2205                         break;
2206                 }
2207
2208                 /*
2209                  * The device does not want the automatic start to be issued.
2210                  */
2211                 if (sdkp->device->no_start_on_add)
2212                         break;
2213
2214                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2215                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2216                                 break;  /* manual intervention required */
2217                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2218                                 break;  /* standby */
2219                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2220                                 break;  /* unavailable */
2221                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2222                                 break;  /* sanitize in progress */
2223                         /*
2224                          * Issue command to spin up drive when not ready
2225                          */
2226                         if (!spintime) {
2227                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2228                                 cmd[0] = START_STOP;
2229                                 cmd[1] = 1;     /* Return immediately */
2230                                 memset((void *) &cmd[2], 0, 8);
2231                                 cmd[4] = 1;     /* Start spin cycle */
2232                                 if (sdkp->device->start_stop_pwr_cond)
2233                                         cmd[4] |= 1 << 4;
2234                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2235                                                  NULL, 0, &sshdr,
2236                                                  SD_TIMEOUT, sdkp->max_retries,
2237                                                  NULL);
2238                                 spintime_expire = jiffies + 100 * HZ;
2239                                 spintime = 1;
2240                         }
2241                         /* Wait 1 second for next try */
2242                         msleep(1000);
2243                         printk(KERN_CONT ".");
2244
2245                 /*
2246                  * Wait for USB flash devices with slow firmware.
2247                  * Yes, this sense key/ASC combination shouldn't
2248                  * occur here.  It's characteristic of these devices.
2249                  */
2250                 } else if (sense_valid &&
2251                                 sshdr.sense_key == UNIT_ATTENTION &&
2252                                 sshdr.asc == 0x28) {
2253                         if (!spintime) {
2254                                 spintime_expire = jiffies + 5 * HZ;
2255                                 spintime = 1;
2256                         }
2257                         /* Wait 1 second for next try */
2258                         msleep(1000);
2259                 } else {
2260                         /* we don't understand the sense code, so it's
2261                          * probably pointless to loop */
2262                         if(!spintime) {
2263                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2264                                 sd_print_sense_hdr(sdkp, &sshdr);
2265                         }
2266                         break;
2267                 }
2268
2269         } while (spintime && time_before_eq(jiffies, spintime_expire));
2270
2271         if (spintime) {
2272                 if (scsi_status_is_good(the_result))
2273                         printk(KERN_CONT "ready\n");
2274                 else
2275                         printk(KERN_CONT "not responding...\n");
2276         }
2277 }
2278
2279 /*
2280  * Determine whether disk supports Data Integrity Field.
2281  */
2282 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2283 {
2284         struct scsi_device *sdp = sdkp->device;
2285         u8 type;
2286         int ret = 0;
2287
2288         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2289                 sdkp->protection_type = 0;
2290                 return ret;
2291         }
2292
2293         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2294
2295         if (type > T10_PI_TYPE3_PROTECTION)
2296                 ret = -ENODEV;
2297         else if (scsi_host_dif_capable(sdp->host, type))
2298                 ret = 1;
2299
2300         if (sdkp->first_scan || type != sdkp->protection_type)
2301                 switch (ret) {
2302                 case -ENODEV:
2303                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2304                                   " protection type %u. Disabling disk!\n",
2305                                   type);
2306                         break;
2307                 case 1:
2308                         sd_printk(KERN_NOTICE, sdkp,
2309                                   "Enabling DIF Type %u protection\n", type);
2310                         break;
2311                 case 0:
2312                         sd_printk(KERN_NOTICE, sdkp,
2313                                   "Disabling DIF Type %u protection\n", type);
2314                         break;
2315                 }
2316
2317         sdkp->protection_type = type;
2318
2319         return ret;
2320 }
2321
2322 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2323                         struct scsi_sense_hdr *sshdr, int sense_valid,
2324                         int the_result)
2325 {
2326         if (sense_valid)
2327                 sd_print_sense_hdr(sdkp, sshdr);
2328         else
2329                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2330
2331         /*
2332          * Set dirty bit for removable devices if not ready -
2333          * sometimes drives will not report this properly.
2334          */
2335         if (sdp->removable &&
2336             sense_valid && sshdr->sense_key == NOT_READY)
2337                 set_media_not_present(sdkp);
2338
2339         /*
2340          * We used to set media_present to 0 here to indicate no media
2341          * in the drive, but some drives fail read capacity even with
2342          * media present, so we can't do that.
2343          */
2344         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2345 }
2346
2347 #define RC16_LEN 32
2348 #if RC16_LEN > SD_BUF_SIZE
2349 #error RC16_LEN must not be more than SD_BUF_SIZE
2350 #endif
2351
2352 #define READ_CAPACITY_RETRIES_ON_RESET  10
2353
2354 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2355                                                 unsigned char *buffer)
2356 {
2357         unsigned char cmd[16];
2358         struct scsi_sense_hdr sshdr;
2359         int sense_valid = 0;
2360         int the_result;
2361         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2362         unsigned int alignment;
2363         unsigned long long lba;
2364         unsigned sector_size;
2365
2366         if (sdp->no_read_capacity_16)
2367                 return -EINVAL;
2368
2369         do {
2370                 memset(cmd, 0, 16);
2371                 cmd[0] = SERVICE_ACTION_IN_16;
2372                 cmd[1] = SAI_READ_CAPACITY_16;
2373                 cmd[13] = RC16_LEN;
2374                 memset(buffer, 0, RC16_LEN);
2375
2376                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2377                                         buffer, RC16_LEN, &sshdr,
2378                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2379
2380                 if (media_not_present(sdkp, &sshdr))
2381                         return -ENODEV;
2382
2383                 if (the_result > 0) {
2384                         sense_valid = scsi_sense_valid(&sshdr);
2385                         if (sense_valid &&
2386                             sshdr.sense_key == ILLEGAL_REQUEST &&
2387                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2388                             sshdr.ascq == 0x00)
2389                                 /* Invalid Command Operation Code or
2390                                  * Invalid Field in CDB, just retry
2391                                  * silently with RC10 */
2392                                 return -EINVAL;
2393                         if (sense_valid &&
2394                             sshdr.sense_key == UNIT_ATTENTION &&
2395                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2396                                 /* Device reset might occur several times,
2397                                  * give it one more chance */
2398                                 if (--reset_retries > 0)
2399                                         continue;
2400                 }
2401                 retries--;
2402
2403         } while (the_result && retries);
2404
2405         if (the_result) {
2406                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2407                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2408                 return -EINVAL;
2409         }
2410
2411         sector_size = get_unaligned_be32(&buffer[8]);
2412         lba = get_unaligned_be64(&buffer[0]);
2413
2414         if (sd_read_protection_type(sdkp, buffer) < 0) {
2415                 sdkp->capacity = 0;
2416                 return -ENODEV;
2417         }
2418
2419         /* Logical blocks per physical block exponent */
2420         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2421
2422         /* RC basis */
2423         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2424
2425         /* Lowest aligned logical block */
2426         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2427         blk_queue_alignment_offset(sdp->request_queue, alignment);
2428         if (alignment && sdkp->first_scan)
2429                 sd_printk(KERN_NOTICE, sdkp,
2430                           "physical block alignment offset: %u\n", alignment);
2431
2432         if (buffer[14] & 0x80) { /* LBPME */
2433                 sdkp->lbpme = 1;
2434
2435                 if (buffer[14] & 0x40) /* LBPRZ */
2436                         sdkp->lbprz = 1;
2437
2438                 sd_config_discard(sdkp, SD_LBP_WS16);
2439         }
2440
2441         sdkp->capacity = lba + 1;
2442         return sector_size;
2443 }
2444
2445 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2446                                                 unsigned char *buffer)
2447 {
2448         unsigned char cmd[16];
2449         struct scsi_sense_hdr sshdr;
2450         int sense_valid = 0;
2451         int the_result;
2452         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2453         sector_t lba;
2454         unsigned sector_size;
2455
2456         do {
2457                 cmd[0] = READ_CAPACITY;
2458                 memset(&cmd[1], 0, 9);
2459                 memset(buffer, 0, 8);
2460
2461                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2462                                         buffer, 8, &sshdr,
2463                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2464
2465                 if (media_not_present(sdkp, &sshdr))
2466                         return -ENODEV;
2467
2468                 if (the_result > 0) {
2469                         sense_valid = scsi_sense_valid(&sshdr);
2470                         if (sense_valid &&
2471                             sshdr.sense_key == UNIT_ATTENTION &&
2472                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2473                                 /* Device reset might occur several times,
2474                                  * give it one more chance */
2475                                 if (--reset_retries > 0)
2476                                         continue;
2477                 }
2478                 retries--;
2479
2480         } while (the_result && retries);
2481
2482         if (the_result) {
2483                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2484                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2485                 return -EINVAL;
2486         }
2487
2488         sector_size = get_unaligned_be32(&buffer[4]);
2489         lba = get_unaligned_be32(&buffer[0]);
2490
2491         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2492                 /* Some buggy (usb cardreader) devices return an lba of
2493                    0xffffffff when the want to report a size of 0 (with
2494                    which they really mean no media is present) */
2495                 sdkp->capacity = 0;
2496                 sdkp->physical_block_size = sector_size;
2497                 return sector_size;
2498         }
2499
2500         sdkp->capacity = lba + 1;
2501         sdkp->physical_block_size = sector_size;
2502         return sector_size;
2503 }
2504
2505 static int sd_try_rc16_first(struct scsi_device *sdp)
2506 {
2507         if (sdp->host->max_cmd_len < 16)
2508                 return 0;
2509         if (sdp->try_rc_10_first)
2510                 return 0;
2511         if (sdp->scsi_level > SCSI_SPC_2)
2512                 return 1;
2513         if (scsi_device_protection(sdp))
2514                 return 1;
2515         return 0;
2516 }
2517
2518 /*
2519  * read disk capacity
2520  */
2521 static void
2522 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2523 {
2524         int sector_size;
2525         struct scsi_device *sdp = sdkp->device;
2526
2527         if (sd_try_rc16_first(sdp)) {
2528                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2529                 if (sector_size == -EOVERFLOW)
2530                         goto got_data;
2531                 if (sector_size == -ENODEV)
2532                         return;
2533                 if (sector_size < 0)
2534                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2535                 if (sector_size < 0)
2536                         return;
2537         } else {
2538                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2539                 if (sector_size == -EOVERFLOW)
2540                         goto got_data;
2541                 if (sector_size < 0)
2542                         return;
2543                 if ((sizeof(sdkp->capacity) > 4) &&
2544                     (sdkp->capacity > 0xffffffffULL)) {
2545                         int old_sector_size = sector_size;
2546                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2547                                         "Trying to use READ CAPACITY(16).\n");
2548                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2549                         if (sector_size < 0) {
2550                                 sd_printk(KERN_NOTICE, sdkp,
2551                                         "Using 0xffffffff as device size\n");
2552                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2553                                 sector_size = old_sector_size;
2554                                 goto got_data;
2555                         }
2556                         /* Remember that READ CAPACITY(16) succeeded */
2557                         sdp->try_rc_10_first = 0;
2558                 }
2559         }
2560
2561         /* Some devices are known to return the total number of blocks,
2562          * not the highest block number.  Some devices have versions
2563          * which do this and others which do not.  Some devices we might
2564          * suspect of doing this but we don't know for certain.
2565          *
2566          * If we know the reported capacity is wrong, decrement it.  If
2567          * we can only guess, then assume the number of blocks is even
2568          * (usually true but not always) and err on the side of lowering
2569          * the capacity.
2570          */
2571         if (sdp->fix_capacity ||
2572             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2573                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2574                                 "from its reported value: %llu\n",
2575                                 (unsigned long long) sdkp->capacity);
2576                 --sdkp->capacity;
2577         }
2578
2579 got_data:
2580         if (sector_size == 0) {
2581                 sector_size = 512;
2582                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2583                           "assuming 512.\n");
2584         }
2585
2586         if (sector_size != 512 &&
2587             sector_size != 1024 &&
2588             sector_size != 2048 &&
2589             sector_size != 4096) {
2590                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2591                           sector_size);
2592                 /*
2593                  * The user might want to re-format the drive with
2594                  * a supported sectorsize.  Once this happens, it
2595                  * would be relatively trivial to set the thing up.
2596                  * For this reason, we leave the thing in the table.
2597                  */
2598                 sdkp->capacity = 0;
2599                 /*
2600                  * set a bogus sector size so the normal read/write
2601                  * logic in the block layer will eventually refuse any
2602                  * request on this device without tripping over power
2603                  * of two sector size assumptions
2604                  */
2605                 sector_size = 512;
2606         }
2607         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2608         blk_queue_physical_block_size(sdp->request_queue,
2609                                       sdkp->physical_block_size);
2610         sdkp->device->sector_size = sector_size;
2611
2612         if (sdkp->capacity > 0xffffffff)
2613                 sdp->use_16_for_rw = 1;
2614
2615 }
2616
2617 /*
2618  * Print disk capacity
2619  */
2620 static void
2621 sd_print_capacity(struct scsi_disk *sdkp,
2622                   sector_t old_capacity)
2623 {
2624         int sector_size = sdkp->device->sector_size;
2625         char cap_str_2[10], cap_str_10[10];
2626
2627         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2628                 return;
2629
2630         string_get_size(sdkp->capacity, sector_size,
2631                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2632         string_get_size(sdkp->capacity, sector_size,
2633                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2634
2635         sd_printk(KERN_NOTICE, sdkp,
2636                   "%llu %d-byte logical blocks: (%s/%s)\n",
2637                   (unsigned long long)sdkp->capacity,
2638                   sector_size, cap_str_10, cap_str_2);
2639
2640         if (sdkp->physical_block_size != sector_size)
2641                 sd_printk(KERN_NOTICE, sdkp,
2642                           "%u-byte physical blocks\n",
2643                           sdkp->physical_block_size);
2644 }
2645
2646 /* called with buffer of length 512 */
2647 static inline int
2648 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2649                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2650                  struct scsi_sense_hdr *sshdr)
2651 {
2652         return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2653                                SD_TIMEOUT, sdkp->max_retries, data,
2654                                sshdr);
2655 }
2656
2657 /*
2658  * read write protect setting, if possible - called only in sd_revalidate_disk()
2659  * called with buffer of length SD_BUF_SIZE
2660  */
2661 static void
2662 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2663 {
2664         int res;
2665         struct scsi_device *sdp = sdkp->device;
2666         struct scsi_mode_data data;
2667         int old_wp = sdkp->write_prot;
2668
2669         set_disk_ro(sdkp->disk, 0);
2670         if (sdp->skip_ms_page_3f) {
2671                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2672                 return;
2673         }
2674
2675         if (sdp->use_192_bytes_for_3f) {
2676                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2677         } else {
2678                 /*
2679                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2680                  * We have to start carefully: some devices hang if we ask
2681                  * for more than is available.
2682                  */
2683                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2684
2685                 /*
2686                  * Second attempt: ask for page 0 When only page 0 is
2687                  * implemented, a request for page 3F may return Sense Key
2688                  * 5: Illegal Request, Sense Code 24: Invalid field in
2689                  * CDB.
2690                  */
2691                 if (res < 0)
2692                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2693
2694                 /*
2695                  * Third attempt: ask 255 bytes, as we did earlier.
2696                  */
2697                 if (res < 0)
2698                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2699                                                &data, NULL);
2700         }
2701
2702         if (res < 0) {
2703                 sd_first_printk(KERN_WARNING, sdkp,
2704                           "Test WP failed, assume Write Enabled\n");
2705         } else {
2706                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2707                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2708                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2709                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2710                                   sdkp->write_prot ? "on" : "off");
2711                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2712                 }
2713         }
2714 }
2715
2716 /*
2717  * sd_read_cache_type - called only from sd_revalidate_disk()
2718  * called with buffer of length SD_BUF_SIZE
2719  */
2720 static void
2721 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2722 {
2723         int len = 0, res;
2724         struct scsi_device *sdp = sdkp->device;
2725
2726         int dbd;
2727         int modepage;
2728         int first_len;
2729         struct scsi_mode_data data;
2730         struct scsi_sense_hdr sshdr;
2731         int old_wce = sdkp->WCE;
2732         int old_rcd = sdkp->RCD;
2733         int old_dpofua = sdkp->DPOFUA;
2734
2735
2736         if (sdkp->cache_override)
2737                 return;
2738
2739         first_len = 4;
2740         if (sdp->skip_ms_page_8) {
2741                 if (sdp->type == TYPE_RBC)
2742                         goto defaults;
2743                 else {
2744                         if (sdp->skip_ms_page_3f)
2745                                 goto defaults;
2746                         modepage = 0x3F;
2747                         if (sdp->use_192_bytes_for_3f)
2748                                 first_len = 192;
2749                         dbd = 0;
2750                 }
2751         } else if (sdp->type == TYPE_RBC) {
2752                 modepage = 6;
2753                 dbd = 8;
2754         } else {
2755                 modepage = 8;
2756                 dbd = 0;
2757         }
2758
2759         /* cautiously ask */
2760         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2761                         &data, &sshdr);
2762
2763         if (res < 0)
2764                 goto bad_sense;
2765
2766         if (!data.header_length) {
2767                 modepage = 6;
2768                 first_len = 0;
2769                 sd_first_printk(KERN_ERR, sdkp,
2770                                 "Missing header in MODE_SENSE response\n");
2771         }
2772
2773         /* that went OK, now ask for the proper length */
2774         len = data.length;
2775
2776         /*
2777          * We're only interested in the first three bytes, actually.
2778          * But the data cache page is defined for the first 20.
2779          */
2780         if (len < 3)
2781                 goto bad_sense;
2782         else if (len > SD_BUF_SIZE) {
2783                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2784                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2785                 len = SD_BUF_SIZE;
2786         }
2787         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2788                 len = 192;
2789
2790         /* Get the data */
2791         if (len > first_len)
2792                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2793                                 &data, &sshdr);
2794
2795         if (!res) {
2796                 int offset = data.header_length + data.block_descriptor_length;
2797
2798                 while (offset < len) {
2799                         u8 page_code = buffer[offset] & 0x3F;
2800                         u8 spf       = buffer[offset] & 0x40;
2801
2802                         if (page_code == 8 || page_code == 6) {
2803                                 /* We're interested only in the first 3 bytes.
2804                                  */
2805                                 if (len - offset <= 2) {
2806                                         sd_first_printk(KERN_ERR, sdkp,
2807                                                 "Incomplete mode parameter "
2808                                                         "data\n");
2809                                         goto defaults;
2810                                 } else {
2811                                         modepage = page_code;
2812                                         goto Page_found;
2813                                 }
2814                         } else {
2815                                 /* Go to the next page */
2816                                 if (spf && len - offset > 3)
2817                                         offset += 4 + (buffer[offset+2] << 8) +
2818                                                 buffer[offset+3];
2819                                 else if (!spf && len - offset > 1)
2820                                         offset += 2 + buffer[offset+1];
2821                                 else {
2822                                         sd_first_printk(KERN_ERR, sdkp,
2823                                                         "Incomplete mode "
2824                                                         "parameter data\n");
2825                                         goto defaults;
2826                                 }
2827                         }
2828                 }
2829
2830                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2831                 goto defaults;
2832
2833         Page_found:
2834                 if (modepage == 8) {
2835                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2836                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2837                 } else {
2838                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2839                         sdkp->RCD = 0;
2840                 }
2841
2842                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2843                 if (sdp->broken_fua) {
2844                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2845                         sdkp->DPOFUA = 0;
2846                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2847                            !sdkp->device->use_16_for_rw) {
2848                         sd_first_printk(KERN_NOTICE, sdkp,
2849                                   "Uses READ/WRITE(6), disabling FUA\n");
2850                         sdkp->DPOFUA = 0;
2851                 }
2852
2853                 /* No cache flush allowed for write protected devices */
2854                 if (sdkp->WCE && sdkp->write_prot)
2855                         sdkp->WCE = 0;
2856
2857                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2858                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2859                         sd_printk(KERN_NOTICE, sdkp,
2860                                   "Write cache: %s, read cache: %s, %s\n",
2861                                   sdkp->WCE ? "enabled" : "disabled",
2862                                   sdkp->RCD ? "disabled" : "enabled",
2863                                   sdkp->DPOFUA ? "supports DPO and FUA"
2864                                   : "doesn't support DPO or FUA");
2865
2866                 return;
2867         }
2868
2869 bad_sense:
2870         if (scsi_sense_valid(&sshdr) &&
2871             sshdr.sense_key == ILLEGAL_REQUEST &&
2872             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2873                 /* Invalid field in CDB */
2874                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2875         else
2876                 sd_first_printk(KERN_ERR, sdkp,
2877                                 "Asking for cache data failed\n");
2878
2879 defaults:
2880         if (sdp->wce_default_on) {
2881                 sd_first_printk(KERN_NOTICE, sdkp,
2882                                 "Assuming drive cache: write back\n");
2883                 sdkp->WCE = 1;
2884         } else {
2885                 sd_first_printk(KERN_ERR, sdkp,
2886                                 "Assuming drive cache: write through\n");
2887                 sdkp->WCE = 0;
2888         }
2889         sdkp->RCD = 0;
2890         sdkp->DPOFUA = 0;
2891 }
2892
2893 /*
2894  * The ATO bit indicates whether the DIF application tag is available
2895  * for use by the operating system.
2896  */
2897 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2898 {
2899         int res, offset;
2900         struct scsi_device *sdp = sdkp->device;
2901         struct scsi_mode_data data;
2902         struct scsi_sense_hdr sshdr;
2903
2904         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2905                 return;
2906
2907         if (sdkp->protection_type == 0)
2908                 return;
2909
2910         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2911                               sdkp->max_retries, &data, &sshdr);
2912
2913         if (res < 0 || !data.header_length ||
2914             data.length < 6) {
2915                 sd_first_printk(KERN_WARNING, sdkp,
2916                           "getting Control mode page failed, assume no ATO\n");
2917
2918                 if (scsi_sense_valid(&sshdr))
2919                         sd_print_sense_hdr(sdkp, &sshdr);
2920
2921                 return;
2922         }
2923
2924         offset = data.header_length + data.block_descriptor_length;
2925
2926         if ((buffer[offset] & 0x3f) != 0x0a) {
2927                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2928                 return;
2929         }
2930
2931         if ((buffer[offset + 5] & 0x80) == 0)
2932                 return;
2933
2934         sdkp->ATO = 1;
2935
2936         return;
2937 }
2938
2939 /**
2940  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2941  * @sdkp: disk to query
2942  */
2943 static void sd_read_block_limits(struct scsi_disk *sdkp)
2944 {
2945         unsigned int sector_sz = sdkp->device->sector_size;
2946         const int vpd_len = 64;
2947         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2948
2949         if (!buffer ||
2950             /* Block Limits VPD */
2951             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2952                 goto out;
2953
2954         blk_queue_io_min(sdkp->disk->queue,
2955                          get_unaligned_be16(&buffer[6]) * sector_sz);
2956
2957         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2958         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2959
2960         if (buffer[3] == 0x3c) {
2961                 unsigned int lba_count, desc_count;
2962
2963                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2964
2965                 if (!sdkp->lbpme)
2966                         goto out;
2967
2968                 lba_count = get_unaligned_be32(&buffer[20]);
2969                 desc_count = get_unaligned_be32(&buffer[24]);
2970
2971                 if (lba_count && desc_count)
2972                         sdkp->max_unmap_blocks = lba_count;
2973
2974                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2975
2976                 if (buffer[32] & 0x80)
2977                         sdkp->unmap_alignment =
2978                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2979
2980                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2981
2982                         if (sdkp->max_unmap_blocks)
2983                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2984                         else
2985                                 sd_config_discard(sdkp, SD_LBP_WS16);
2986
2987                 } else {        /* LBP VPD page tells us what to use */
2988                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2989                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2990                         else if (sdkp->lbpws)
2991                                 sd_config_discard(sdkp, SD_LBP_WS16);
2992                         else if (sdkp->lbpws10)
2993                                 sd_config_discard(sdkp, SD_LBP_WS10);
2994                         else
2995                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2996                 }
2997         }
2998
2999  out:
3000         kfree(buffer);
3001 }
3002
3003 /**
3004  * sd_read_block_characteristics - Query block dev. characteristics
3005  * @sdkp: disk to query
3006  */
3007 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3008 {
3009         struct request_queue *q = sdkp->disk->queue;
3010         unsigned char *buffer;
3011         u16 rot;
3012         const int vpd_len = 64;
3013
3014         buffer = kmalloc(vpd_len, GFP_KERNEL);
3015
3016         if (!buffer ||
3017             /* Block Device Characteristics VPD */
3018             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
3019                 goto out;
3020
3021         rot = get_unaligned_be16(&buffer[4]);
3022
3023         if (rot == 1) {
3024                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3025                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3026         }
3027
3028         if (sdkp->device->type == TYPE_ZBC) {
3029                 /* Host-managed */
3030                 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3031         } else {
3032                 sdkp->zoned = (buffer[8] >> 4) & 3;
3033                 if (sdkp->zoned == 1) {
3034                         /* Host-aware */
3035                         blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3036                 } else {
3037                         /* Regular disk or drive managed disk */
3038                         blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3039                 }
3040         }
3041
3042         if (!sdkp->first_scan)
3043                 goto out;
3044
3045         if (blk_queue_is_zoned(q)) {
3046                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3047                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3048         } else {
3049                 if (sdkp->zoned == 1)
3050                         sd_printk(KERN_NOTICE, sdkp,
3051                                   "Host-aware SMR disk used as regular disk\n");
3052                 else if (sdkp->zoned == 2)
3053                         sd_printk(KERN_NOTICE, sdkp,
3054                                   "Drive-managed SMR disk\n");
3055         }
3056
3057  out:
3058         kfree(buffer);
3059 }
3060
3061 /**
3062  * sd_read_block_provisioning - Query provisioning VPD page
3063  * @sdkp: disk to query
3064  */
3065 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3066 {
3067         unsigned char *buffer;
3068         const int vpd_len = 8;
3069
3070         if (sdkp->lbpme == 0)
3071                 return;
3072
3073         buffer = kmalloc(vpd_len, GFP_KERNEL);
3074
3075         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3076                 goto out;
3077
3078         sdkp->lbpvpd    = 1;
3079         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3080         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3081         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3082
3083  out:
3084         kfree(buffer);
3085 }
3086
3087 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3088 {
3089         struct scsi_device *sdev = sdkp->device;
3090
3091         if (sdev->host->no_write_same) {
3092                 sdev->no_write_same = 1;
3093
3094                 return;
3095         }
3096
3097         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3098                 /* too large values might cause issues with arcmsr */
3099                 int vpd_buf_len = 64;
3100
3101                 sdev->no_report_opcodes = 1;
3102
3103                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3104                  * CODES is unsupported and the device has an ATA
3105                  * Information VPD page (SAT).
3106                  */
3107                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3108                         sdev->no_write_same = 1;
3109         }
3110
3111         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3112                 sdkp->ws16 = 1;
3113
3114         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3115                 sdkp->ws10 = 1;
3116 }
3117
3118 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3119 {
3120         struct scsi_device *sdev = sdkp->device;
3121
3122         if (!sdev->security_supported)
3123                 return;
3124
3125         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3126                         SECURITY_PROTOCOL_IN) == 1 &&
3127             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3128                         SECURITY_PROTOCOL_OUT) == 1)
3129                 sdkp->security = 1;
3130 }
3131
3132 /*
3133  * Determine the device's preferred I/O size for reads and writes
3134  * unless the reported value is unreasonably small, large, not a
3135  * multiple of the physical block size, or simply garbage.
3136  */
3137 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3138                                       unsigned int dev_max)
3139 {
3140         struct scsi_device *sdp = sdkp->device;
3141         unsigned int opt_xfer_bytes =
3142                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3143
3144         if (sdkp->opt_xfer_blocks == 0)
3145                 return false;
3146
3147         if (sdkp->opt_xfer_blocks > dev_max) {
3148                 sd_first_printk(KERN_WARNING, sdkp,
3149                                 "Optimal transfer size %u logical blocks " \
3150                                 "> dev_max (%u logical blocks)\n",
3151                                 sdkp->opt_xfer_blocks, dev_max);
3152                 return false;
3153         }
3154
3155         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3156                 sd_first_printk(KERN_WARNING, sdkp,
3157                                 "Optimal transfer size %u logical blocks " \
3158                                 "> sd driver limit (%u logical blocks)\n",
3159                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3160                 return false;
3161         }
3162
3163         if (opt_xfer_bytes < PAGE_SIZE) {
3164                 sd_first_printk(KERN_WARNING, sdkp,
3165                                 "Optimal transfer size %u bytes < " \
3166                                 "PAGE_SIZE (%u bytes)\n",
3167                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3168                 return false;
3169         }
3170
3171         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3172                 sd_first_printk(KERN_WARNING, sdkp,
3173                                 "Optimal transfer size %u bytes not a " \
3174                                 "multiple of physical block size (%u bytes)\n",
3175                                 opt_xfer_bytes, sdkp->physical_block_size);
3176                 return false;
3177         }
3178
3179         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3180                         opt_xfer_bytes);
3181         return true;
3182 }
3183
3184 /**
3185  *      sd_revalidate_disk - called the first time a new disk is seen,
3186  *      performs disk spin up, read_capacity, etc.
3187  *      @disk: struct gendisk we care about
3188  **/
3189 static int sd_revalidate_disk(struct gendisk *disk)
3190 {
3191         struct scsi_disk *sdkp = scsi_disk(disk);
3192         struct scsi_device *sdp = sdkp->device;
3193         struct request_queue *q = sdkp->disk->queue;
3194         sector_t old_capacity = sdkp->capacity;
3195         unsigned char *buffer;
3196         unsigned int dev_max, rw_max;
3197
3198         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3199                                       "sd_revalidate_disk\n"));
3200
3201         /*
3202          * If the device is offline, don't try and read capacity or any
3203          * of the other niceties.
3204          */
3205         if (!scsi_device_online(sdp))
3206                 goto out;
3207
3208         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3209         if (!buffer) {
3210                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3211                           "allocation failure.\n");
3212                 goto out;
3213         }
3214
3215         sd_spinup_disk(sdkp);
3216
3217         /*
3218          * Without media there is no reason to ask; moreover, some devices
3219          * react badly if we do.
3220          */
3221         if (sdkp->media_present) {
3222                 sd_read_capacity(sdkp, buffer);
3223
3224                 /*
3225                  * set the default to rotational.  All non-rotational devices
3226                  * support the block characteristics VPD page, which will
3227                  * cause this to be updated correctly and any device which
3228                  * doesn't support it should be treated as rotational.
3229                  */
3230                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3231                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3232
3233                 if (scsi_device_supports_vpd(sdp)) {
3234                         sd_read_block_provisioning(sdkp);
3235                         sd_read_block_limits(sdkp);
3236                         sd_read_block_characteristics(sdkp);
3237                         sd_zbc_read_zones(sdkp, buffer);
3238                 }
3239
3240                 sd_print_capacity(sdkp, old_capacity);
3241
3242                 sd_read_write_protect_flag(sdkp, buffer);
3243                 sd_read_cache_type(sdkp, buffer);
3244                 sd_read_app_tag_own(sdkp, buffer);
3245                 sd_read_write_same(sdkp, buffer);
3246                 sd_read_security(sdkp, buffer);
3247         }
3248
3249         /*
3250          * We now have all cache related info, determine how we deal
3251          * with flush requests.
3252          */
3253         sd_set_flush_flag(sdkp);
3254
3255         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3256         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3257
3258         /* Some devices report a maximum block count for READ/WRITE requests. */
3259         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3260         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3261
3262         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3263                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3264                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3265         } else {
3266                 q->limits.io_opt = 0;
3267                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3268                                       (sector_t)BLK_DEF_MAX_SECTORS);
3269         }
3270
3271         /* Do not exceed controller limit */
3272         rw_max = min(rw_max, queue_max_hw_sectors(q));
3273
3274         /*
3275          * Only update max_sectors if previously unset or if the current value
3276          * exceeds the capabilities of the hardware.
3277          */
3278         if (sdkp->first_scan ||
3279             q->limits.max_sectors > q->limits.max_dev_sectors ||
3280             q->limits.max_sectors > q->limits.max_hw_sectors)
3281                 q->limits.max_sectors = rw_max;
3282
3283         sdkp->first_scan = 0;
3284
3285         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3286         sd_config_write_same(sdkp);
3287         kfree(buffer);
3288
3289         /*
3290          * For a zoned drive, revalidating the zones can be done only once
3291          * the gendisk capacity is set. So if this fails, set back the gendisk
3292          * capacity to 0.
3293          */
3294         if (sd_zbc_revalidate_zones(sdkp))
3295                 set_capacity_and_notify(disk, 0);
3296
3297  out:
3298         return 0;
3299 }
3300
3301 /**
3302  *      sd_unlock_native_capacity - unlock native capacity
3303  *      @disk: struct gendisk to set capacity for
3304  *
3305  *      Block layer calls this function if it detects that partitions
3306  *      on @disk reach beyond the end of the device.  If the SCSI host
3307  *      implements ->unlock_native_capacity() method, it's invoked to
3308  *      give it a chance to adjust the device capacity.
3309  *
3310  *      CONTEXT:
3311  *      Defined by block layer.  Might sleep.
3312  */
3313 static void sd_unlock_native_capacity(struct gendisk *disk)
3314 {
3315         struct scsi_device *sdev = scsi_disk(disk)->device;
3316
3317         if (sdev->host->hostt->unlock_native_capacity)
3318                 sdev->host->hostt->unlock_native_capacity(sdev);
3319 }
3320
3321 /**
3322  *      sd_format_disk_name - format disk name
3323  *      @prefix: name prefix - ie. "sd" for SCSI disks
3324  *      @index: index of the disk to format name for
3325  *      @buf: output buffer
3326  *      @buflen: length of the output buffer
3327  *
3328  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3329  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3330  *      which is followed by sdaaa.
3331  *
3332  *      This is basically 26 base counting with one extra 'nil' entry
3333  *      at the beginning from the second digit on and can be
3334  *      determined using similar method as 26 base conversion with the
3335  *      index shifted -1 after each digit is computed.
3336  *
3337  *      CONTEXT:
3338  *      Don't care.
3339  *
3340  *      RETURNS:
3341  *      0 on success, -errno on failure.
3342  */
3343 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3344 {
3345         const int base = 'z' - 'a' + 1;
3346         char *begin = buf + strlen(prefix);
3347         char *end = buf + buflen;
3348         char *p;
3349         int unit;
3350
3351         p = end - 1;
3352         *p = '\0';
3353         unit = base;
3354         do {
3355                 if (p == begin)
3356                         return -EINVAL;
3357                 *--p = 'a' + (index % unit);
3358                 index = (index / unit) - 1;
3359         } while (index >= 0);
3360
3361         memmove(begin, p, end - p);
3362         memcpy(buf, prefix, strlen(prefix));
3363
3364         return 0;
3365 }
3366
3367 /**
3368  *      sd_probe - called during driver initialization and whenever a
3369  *      new scsi device is attached to the system. It is called once
3370  *      for each scsi device (not just disks) present.
3371  *      @dev: pointer to device object
3372  *
3373  *      Returns 0 if successful (or not interested in this scsi device
3374  *      (e.g. scanner)); 1 when there is an error.
3375  *
3376  *      Note: this function is invoked from the scsi mid-level.
3377  *      This function sets up the mapping between a given
3378  *      <host,channel,id,lun> (found in sdp) and new device name
3379  *      (e.g. /dev/sda). More precisely it is the block device major
3380  *      and minor number that is chosen here.
3381  *
3382  *      Assume sd_probe is not re-entrant (for time being)
3383  *      Also think about sd_probe() and sd_remove() running coincidentally.
3384  **/
3385 static int sd_probe(struct device *dev)
3386 {
3387         struct scsi_device *sdp = to_scsi_device(dev);
3388         struct scsi_disk *sdkp;
3389         struct gendisk *gd;
3390         int index;
3391         int error;
3392
3393         scsi_autopm_get_device(sdp);
3394         error = -ENODEV;
3395         if (sdp->type != TYPE_DISK &&
3396             sdp->type != TYPE_ZBC &&
3397             sdp->type != TYPE_MOD &&
3398             sdp->type != TYPE_RBC)
3399                 goto out;
3400
3401         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3402                 sdev_printk(KERN_WARNING, sdp,
3403                             "Unsupported ZBC host-managed device.\n");
3404                 goto out;
3405         }
3406
3407         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3408                                         "sd_probe\n"));
3409
3410         error = -ENOMEM;
3411         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3412         if (!sdkp)
3413                 goto out;
3414
3415         gd = alloc_disk(SD_MINORS);
3416         if (!gd)
3417                 goto out_free;
3418
3419         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3420         if (index < 0) {
3421                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3422                 goto out_put;
3423         }
3424
3425         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3426         if (error) {
3427                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3428                 goto out_free_index;
3429         }
3430
3431         sdkp->device = sdp;
3432         sdkp->driver = &sd_template;
3433         sdkp->disk = gd;
3434         sdkp->index = index;
3435         sdkp->max_retries = SD_MAX_RETRIES;
3436         atomic_set(&sdkp->openers, 0);
3437         atomic_set(&sdkp->device->ioerr_cnt, 0);
3438
3439         if (!sdp->request_queue->rq_timeout) {
3440                 if (sdp->type != TYPE_MOD)
3441                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3442                 else
3443                         blk_queue_rq_timeout(sdp->request_queue,
3444                                              SD_MOD_TIMEOUT);
3445         }
3446
3447         device_initialize(&sdkp->dev);
3448         sdkp->dev.parent = dev;
3449         sdkp->dev.class = &sd_disk_class;
3450         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3451
3452         error = device_add(&sdkp->dev);
3453         if (error)
3454                 goto out_free_index;
3455
3456         get_device(dev);
3457         dev_set_drvdata(dev, sdkp);
3458
3459         gd->major = sd_major((index & 0xf0) >> 4);
3460         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3461
3462         gd->fops = &sd_fops;
3463         gd->private_data = &sdkp->driver;
3464         gd->queue = sdkp->device->request_queue;
3465
3466         /* defaults, until the device tells us otherwise */
3467         sdp->sector_size = 512;
3468         sdkp->capacity = 0;
3469         sdkp->media_present = 1;
3470         sdkp->write_prot = 0;
3471         sdkp->cache_override = 0;
3472         sdkp->WCE = 0;
3473         sdkp->RCD = 0;
3474         sdkp->ATO = 0;
3475         sdkp->first_scan = 1;
3476         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3477
3478         sd_revalidate_disk(gd);
3479
3480         gd->flags = GENHD_FL_EXT_DEVT;
3481         if (sdp->removable) {
3482                 gd->flags |= GENHD_FL_REMOVABLE;
3483                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3484                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3485         }
3486
3487         blk_pm_runtime_init(sdp->request_queue, dev);
3488         if (sdp->rpm_autosuspend) {
3489                 pm_runtime_set_autosuspend_delay(dev,
3490                         sdp->host->hostt->rpm_autosuspend_delay);
3491         }
3492         device_add_disk(dev, gd, NULL);
3493         if (sdkp->capacity)
3494                 sd_dif_config_host(sdkp);
3495
3496         sd_revalidate_disk(gd);
3497
3498         if (sdkp->security) {
3499                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3500                 if (sdkp->opal_dev)
3501                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3502         }
3503
3504         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3505                   sdp->removable ? "removable " : "");
3506         scsi_autopm_put_device(sdp);
3507
3508         return 0;
3509
3510  out_free_index:
3511         ida_free(&sd_index_ida, index);
3512  out_put:
3513         put_disk(gd);
3514  out_free:
3515         sd_zbc_release_disk(sdkp);
3516         kfree(sdkp);
3517  out:
3518         scsi_autopm_put_device(sdp);
3519         return error;
3520 }
3521
3522 /**
3523  *      sd_remove - called whenever a scsi disk (previously recognized by
3524  *      sd_probe) is detached from the system. It is called (potentially
3525  *      multiple times) during sd module unload.
3526  *      @dev: pointer to device object
3527  *
3528  *      Note: this function is invoked from the scsi mid-level.
3529  *      This function potentially frees up a device name (e.g. /dev/sdc)
3530  *      that could be re-used by a subsequent sd_probe().
3531  *      This function is not called when the built-in sd driver is "exit-ed".
3532  **/
3533 static int sd_remove(struct device *dev)
3534 {
3535         struct scsi_disk *sdkp;
3536
3537         sdkp = dev_get_drvdata(dev);
3538         scsi_autopm_get_device(sdkp->device);
3539
3540         async_synchronize_full_domain(&scsi_sd_pm_domain);
3541         device_del(&sdkp->dev);
3542         del_gendisk(sdkp->disk);
3543         sd_shutdown(dev);
3544
3545         free_opal_dev(sdkp->opal_dev);
3546
3547         mutex_lock(&sd_ref_mutex);
3548         dev_set_drvdata(dev, NULL);
3549         put_device(&sdkp->dev);
3550         mutex_unlock(&sd_ref_mutex);
3551
3552         return 0;
3553 }
3554
3555 /**
3556  *      scsi_disk_release - Called to free the scsi_disk structure
3557  *      @dev: pointer to embedded class device
3558  *
3559  *      sd_ref_mutex must be held entering this routine.  Because it is
3560  *      called on last put, you should always use the scsi_disk_get()
3561  *      scsi_disk_put() helpers which manipulate the semaphore directly
3562  *      and never do a direct put_device.
3563  **/
3564 static void scsi_disk_release(struct device *dev)
3565 {
3566         struct scsi_disk *sdkp = to_scsi_disk(dev);
3567         struct gendisk *disk = sdkp->disk;
3568         struct request_queue *q = disk->queue;
3569
3570         ida_free(&sd_index_ida, sdkp->index);
3571
3572         /*
3573          * Wait until all requests that are in progress have completed.
3574          * This is necessary to avoid that e.g. scsi_end_request() crashes
3575          * due to clearing the disk->private_data pointer. Wait from inside
3576          * scsi_disk_release() instead of from sd_release() to avoid that
3577          * freezing and unfreezing the request queue affects user space I/O
3578          * in case multiple processes open a /dev/sd... node concurrently.
3579          */
3580         blk_mq_freeze_queue(q);
3581         blk_mq_unfreeze_queue(q);
3582
3583         disk->private_data = NULL;
3584         put_disk(disk);
3585         put_device(&sdkp->device->sdev_gendev);
3586
3587         sd_zbc_release_disk(sdkp);
3588
3589         kfree(sdkp);
3590 }
3591
3592 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3593 {
3594         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3595         struct scsi_sense_hdr sshdr;
3596         struct scsi_device *sdp = sdkp->device;
3597         int res;
3598
3599         if (start)
3600                 cmd[4] |= 1;    /* START */
3601
3602         if (sdp->start_stop_pwr_cond)
3603                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3604
3605         if (!scsi_device_online(sdp))
3606                 return -ENODEV;
3607
3608         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3609                         SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3610         if (res) {
3611                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3612                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3613                         sd_print_sense_hdr(sdkp, &sshdr);
3614                         /* 0x3a is medium not present */
3615                         if (sshdr.asc == 0x3a)
3616                                 res = 0;
3617                 }
3618         }
3619
3620         /* SCSI error codes must not go to the generic layer */
3621         if (res)
3622                 return -EIO;
3623
3624         return 0;
3625 }
3626
3627 /*
3628  * Send a SYNCHRONIZE CACHE instruction down to the device through
3629  * the normal SCSI command structure.  Wait for the command to
3630  * complete.
3631  */
3632 static void sd_shutdown(struct device *dev)
3633 {
3634         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3635
3636         if (!sdkp)
3637                 return;         /* this can happen */
3638
3639         if (pm_runtime_suspended(dev))
3640                 return;
3641
3642         if (sdkp->WCE && sdkp->media_present) {
3643                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3644                 sd_sync_cache(sdkp, NULL);
3645         }
3646
3647         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3648                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3649                 sd_start_stop_device(sdkp, 0);
3650         }
3651 }
3652
3653 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3654 {
3655         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3656         struct scsi_sense_hdr sshdr;
3657         int ret = 0;
3658
3659         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3660                 return 0;
3661
3662         if (sdkp->WCE && sdkp->media_present) {
3663                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3664                 ret = sd_sync_cache(sdkp, &sshdr);
3665
3666                 if (ret) {
3667                         /* ignore OFFLINE device */
3668                         if (ret == -ENODEV)
3669                                 return 0;
3670
3671                         if (!scsi_sense_valid(&sshdr) ||
3672                             sshdr.sense_key != ILLEGAL_REQUEST)
3673                                 return ret;
3674
3675                         /*
3676                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3677                          * doesn't support sync. There's not much to do and
3678                          * suspend shouldn't fail.
3679                          */
3680                         ret = 0;
3681                 }
3682         }
3683
3684         if (sdkp->device->manage_start_stop) {
3685                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3686                 /* an error is not worth aborting a system sleep */
3687                 ret = sd_start_stop_device(sdkp, 0);
3688                 if (ignore_stop_errors)
3689                         ret = 0;
3690         }
3691
3692         return ret;
3693 }
3694
3695 static int sd_suspend_system(struct device *dev)
3696 {
3697         return sd_suspend_common(dev, true);
3698 }
3699
3700 static int sd_suspend_runtime(struct device *dev)
3701 {
3702         return sd_suspend_common(dev, false);
3703 }
3704
3705 static int sd_resume(struct device *dev)
3706 {
3707         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3708         int ret;
3709
3710         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3711                 return 0;
3712
3713         if (!sdkp->device->manage_start_stop)
3714                 return 0;
3715
3716         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3717         ret = sd_start_stop_device(sdkp, 1);
3718         if (!ret)
3719                 opal_unlock_from_suspend(sdkp->opal_dev);
3720         return ret;
3721 }
3722
3723 /**
3724  *      init_sd - entry point for this driver (both when built in or when
3725  *      a module).
3726  *
3727  *      Note: this function registers this driver with the scsi mid-level.
3728  **/
3729 static int __init init_sd(void)
3730 {
3731         int majors = 0, i, err;
3732
3733         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3734
3735         for (i = 0; i < SD_MAJORS; i++) {
3736                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3737                         continue;
3738                 majors++;
3739         }
3740
3741         if (!majors)
3742                 return -ENODEV;
3743
3744         err = class_register(&sd_disk_class);
3745         if (err)
3746                 goto err_out;
3747
3748         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3749                                          0, 0, NULL);
3750         if (!sd_cdb_cache) {
3751                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3752                 err = -ENOMEM;
3753                 goto err_out_class;
3754         }
3755
3756         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3757         if (!sd_cdb_pool) {
3758                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3759                 err = -ENOMEM;
3760                 goto err_out_cache;
3761         }
3762
3763         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3764         if (!sd_page_pool) {
3765                 printk(KERN_ERR "sd: can't init discard page pool\n");
3766                 err = -ENOMEM;
3767                 goto err_out_ppool;
3768         }
3769
3770         err = scsi_register_driver(&sd_template.gendrv);
3771         if (err)
3772                 goto err_out_driver;
3773
3774         return 0;
3775
3776 err_out_driver:
3777         mempool_destroy(sd_page_pool);
3778
3779 err_out_ppool:
3780         mempool_destroy(sd_cdb_pool);
3781
3782 err_out_cache:
3783         kmem_cache_destroy(sd_cdb_cache);
3784
3785 err_out_class:
3786         class_unregister(&sd_disk_class);
3787 err_out:
3788         for (i = 0; i < SD_MAJORS; i++)
3789                 unregister_blkdev(sd_major(i), "sd");
3790         return err;
3791 }
3792
3793 /**
3794  *      exit_sd - exit point for this driver (when it is a module).
3795  *
3796  *      Note: this function unregisters this driver from the scsi mid-level.
3797  **/
3798 static void __exit exit_sd(void)
3799 {
3800         int i;
3801
3802         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3803
3804         scsi_unregister_driver(&sd_template.gendrv);
3805         mempool_destroy(sd_cdb_pool);
3806         mempool_destroy(sd_page_pool);
3807         kmem_cache_destroy(sd_cdb_cache);
3808
3809         class_unregister(&sd_disk_class);
3810
3811         for (i = 0; i < SD_MAJORS; i++)
3812                 unregister_blkdev(sd_major(i), "sd");
3813 }
3814
3815 module_init(init_sd);
3816 module_exit(exit_sd);
3817
3818 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3819 {
3820         scsi_print_sense_hdr(sdkp->device,
3821                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3822 }
3823
3824 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3825 {
3826         const char *hb_string = scsi_hostbyte_string(result);
3827
3828         if (hb_string)
3829                 sd_printk(KERN_INFO, sdkp,
3830                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3831                           hb_string ? hb_string : "invalid",
3832                           "DRIVER_OK");
3833         else
3834                 sd_printk(KERN_INFO, sdkp,
3835                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3836                           msg, host_byte(result), "DRIVER_OK");
3837 }