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