drivers/scsi/sd.c

   1 /*
   2  *      sd.c Copyright (C) 1992 Drew Eckhardt
   3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
   4  *
   5  *      Linux scsi disk driver
   6  *              Initial versions: Drew Eckhardt
   7  *              Subsequent revisions: Eric Youngdale
   8  *      Modification history:
   9  *       - Drew Eckhardt <[email protected]> original
  10  *       - Eric Youngdale <[email protected]> add scatter-gather, multiple
  11  *         outstanding request, and other enhancements.
  12  *         Support loadable low-level scsi drivers.
  13  *       - Jirka Hanika <[email protected]> support more scsi disks using
  14  *         eight major numbers.
  15  *       - Richard Gooch <[email protected]> support devfs.
  16  *       - Torben Mathiasen <[email protected]> Resource allocation fixes in
  17  *         sd_init and cleanups.
  18  *       - Alex Davis <[email protected]> Fix problem where partition info
  19  *         not being read in sd_open. Fix problem where removable media
  20  *         could be ejected after sd_open.
  21  *       - Douglas Gilbert <[email protected]> cleanup for lk 2.5.x
  22  *       - Badari Pulavarty <[email protected]>, Matthew Wilcox
  23  *         <[email protected]>, Kurt Garloff <[email protected]>:
  24  *         Support 32k/1M disks.
  25  *
  26  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
  27  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
  28  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
  29  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
  30  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
  31  *      Note: when the logging level is set by the user, it must be greater
  32  *      than the level indicated above to trigger output.
  33  */
  34
  35 #include <linux/module.h>
  36 #include <linux/fs.h>
  37 #include <linux/kernel.h>
  38 #include <linux/mm.h>
  39 #include <linux/bio.h>
  40 #include <linux/genhd.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/errno.h>
  43 #include <linux/idr.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/init.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/blkpg.h>
  48 #include <linux/blk-pm.h>
  49 #include <linux/delay.h>
  50 #include <linux/mutex.h>
  51 #include <linux/string_helpers.h>
  52 #include <linux/async.h>
  53 #include <linux/slab.h>
  54 #include <linux/sed-opal.h>
  55 #include <linux/pm_runtime.h>
  56 #include <linux/pr.h>
  57 #include <linux/t10-pi.h>
  58 #include <linux/uaccess.h>
  59 #include <asm/unaligned.h>
  60
  61 #include <scsi/scsi.h>
  62 #include <scsi/scsi_cmnd.h>
  63 #include <scsi/scsi_dbg.h>
  64 #include <scsi/scsi_device.h>
  65 #include <scsi/scsi_driver.h>
  66 #include <scsi/scsi_eh.h>
  67 #include <scsi/scsi_host.h>
  68 #include <scsi/scsi_ioctl.h>
  69 #include <scsi/scsicam.h>
  70
  71 #include "sd.h"
  72 #include "scsi_priv.h"
  73 #include "scsi_logging.h"
  74
  75 MODULE_AUTHOR("Eric Youngdale");
  76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
  77 MODULE_LICENSE("GPL");
  78
  79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
  80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
  81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
  82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
  83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
  84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
  85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
  86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
  87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
  88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
  89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
  90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
  91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
  92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
  93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
  94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
  95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
  96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
  97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
  98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
  99
 100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
 101 #define SD_MINORS       16
 102 #else
 103 #define SD_MINORS       0
 104 #endif
 105
 106 static void sd_config_discard(struct scsi_disk *, unsigned int);
 107 static void sd_config_write_same(struct scsi_disk *);
 108 static int  sd_revalidate_disk(struct gendisk *);
 109 static void sd_unlock_native_capacity(struct gendisk *disk);
 110 static int  sd_probe(struct device *);
 111 static int  sd_remove(struct device *);
 112 static void sd_shutdown(struct device *);
 113 static int sd_suspend_system(struct device *);
 114 static int sd_suspend_runtime(struct device *);
 115 static int sd_resume(struct device *);
 116 static void sd_rescan(struct device *);
 117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
 118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
 119 static int sd_done(struct scsi_cmnd *);
 120 static void sd_eh_reset(struct scsi_cmnd *);
 121 static int sd_eh_action(struct scsi_cmnd *, int);
 122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 123 static void scsi_disk_release(struct device *cdev);
 124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
 125 static void sd_print_result(const struct scsi_disk *, const char *, int);
 126
 127 static DEFINE_IDA(sd_index_ida);
 128
 129 /* This semaphore is used to mediate the 0->1 reference get in the
 130  * face of object destruction (i.e. we can't allow a get on an
 131  * object after last put) */
 132 static DEFINE_MUTEX(sd_ref_mutex);
 133
 134 static struct kmem_cache *sd_cdb_cache;
 135 static mempool_t *sd_cdb_pool;
 136 static mempool_t *sd_page_pool;
 137
 138 static const char *sd_cache_types[] = {
 139         "write through", "none", "write back",
 140         "write back, no read (daft)"
 141 };
 142
 143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 144 {
 145         bool wc = false, fua = false;
 146
 147         if (sdkp->WCE) {
 148                 wc = true;
 149                 if (sdkp->DPOFUA)
 150                         fua = true;
 151         }
 152
 153         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 154 }
 155
 156 static ssize_t
 157 cache_type_store(struct device *dev, struct device_attribute *attr,
 158                  const char *buf, size_t count)
 159 {
 160         int ct, rcd, wce, sp;
 161         struct scsi_disk *sdkp = to_scsi_disk(dev);
 162         struct scsi_device *sdp = sdkp->device;
 163         char buffer[64];
 164         char *buffer_data;
 165         struct scsi_mode_data data;
 166         struct scsi_sense_hdr sshdr;
 167         static const char temp[] = "temporary ";
 168         int len;
 169
 170         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 171                 /* no cache control on RBC devices; theoretically they
 172                  * can do it, but there's probably so many exceptions
 173                  * it's not worth the risk */
 174                 return -EINVAL;
 175
 176         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 177                 buf += sizeof(temp) - 1;
 178                 sdkp->cache_override = 1;
 179         } else {
 180                 sdkp->cache_override = 0;
 181         }
 182
 183         ct = sysfs_match_string(sd_cache_types, buf);
 184         if (ct < 0)
 185                 return -EINVAL;
 186
 187         rcd = ct & 0x01 ? 1 : 0;
 188         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 189
 190         if (sdkp->cache_override) {
 191                 sdkp->WCE = wce;
 192                 sdkp->RCD = rcd;
 193                 sd_set_flush_flag(sdkp);
 194                 return count;
 195         }
 196
 197         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 198                             SD_MAX_RETRIES, &data, NULL))
 199                 return -EINVAL;
 200         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 201                   data.block_descriptor_length);
 202         buffer_data = buffer + data.header_length +
 203                 data.block_descriptor_length;
 204         buffer_data[2] &= ~0x05;
 205         buffer_data[2] |= wce << 2 | rcd;
 206         sp = buffer_data[0] & 0x80 ? 1 : 0;
 207         buffer_data[0] &= ~0x80;
 208
 209         /*
 210          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 211          * received mode parameter buffer before doing MODE SELECT.
 212          */
 213         data.device_specific = 0;
 214
 215         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
 216                              SD_MAX_RETRIES, &data, &sshdr)) {
 217                 if (scsi_sense_valid(&sshdr))
 218                         sd_print_sense_hdr(sdkp, &sshdr);
 219                 return -EINVAL;
 220         }
 221         revalidate_disk(sdkp->disk);
 222         return count;
 223 }
 224
 225 static ssize_t
 226 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
 227                        char *buf)
 228 {
 229         struct scsi_disk *sdkp = to_scsi_disk(dev);
 230         struct scsi_device *sdp = sdkp->device;
 231
 232         return sprintf(buf, "%u\n", sdp->manage_start_stop);
 233 }
 234
 235 static ssize_t
 236 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
 237                         const char *buf, size_t count)
 238 {
 239         struct scsi_disk *sdkp = to_scsi_disk(dev);
 240         struct scsi_device *sdp = sdkp->device;
 241         bool v;
 242
 243         if (!capable(CAP_SYS_ADMIN))
 244                 return -EACCES;
 245
 246         if (kstrtobool(buf, &v))
 247                 return -EINVAL;
 248
 249         sdp->manage_start_stop = v;
 250
 251         return count;
 252 }
 253 static DEVICE_ATTR_RW(manage_start_stop);
 254
 255 static ssize_t
 256 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 257 {
 258         struct scsi_disk *sdkp = to_scsi_disk(dev);
 259
 260         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 261 }
 262
 263 static ssize_t
 264 allow_restart_store(struct device *dev, struct device_attribute *attr,
 265                     const char *buf, size_t count)
 266 {
 267         bool v;
 268         struct scsi_disk *sdkp = to_scsi_disk(dev);
 269         struct scsi_device *sdp = sdkp->device;
 270
 271         if (!capable(CAP_SYS_ADMIN))
 272                 return -EACCES;
 273
 274         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 275                 return -EINVAL;
 276
 277         if (kstrtobool(buf, &v))
 278                 return -EINVAL;
 279
 280         sdp->allow_restart = v;
 281
 282         return count;
 283 }
 284 static DEVICE_ATTR_RW(allow_restart);
 285
 286 static ssize_t
 287 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 288 {
 289         struct scsi_disk *sdkp = to_scsi_disk(dev);
 290         int ct = sdkp->RCD + 2*sdkp->WCE;
 291
 292         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 293 }
 294 static DEVICE_ATTR_RW(cache_type);
 295
 296 static ssize_t
 297 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 298 {
 299         struct scsi_disk *sdkp = to_scsi_disk(dev);
 300
 301         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 302 }
 303 static DEVICE_ATTR_RO(FUA);
 304
 305 static ssize_t
 306 protection_type_show(struct device *dev, struct device_attribute *attr,
 307                      char *buf)
 308 {
 309         struct scsi_disk *sdkp = to_scsi_disk(dev);
 310
 311         return sprintf(buf, "%u\n", sdkp->protection_type);
 312 }
 313
 314 static ssize_t
 315 protection_type_store(struct device *dev, struct device_attribute *attr,
 316                       const char *buf, size_t count)
 317 {
 318         struct scsi_disk *sdkp = to_scsi_disk(dev);
 319         unsigned int val;
 320         int err;
 321
 322         if (!capable(CAP_SYS_ADMIN))
 323                 return -EACCES;
 324
 325         err = kstrtouint(buf, 10, &val);
 326
 327         if (err)
 328                 return err;
 329
 330         if (val <= T10_PI_TYPE3_PROTECTION)
 331                 sdkp->protection_type = val;
 332
 333         return count;
 334 }
 335 static DEVICE_ATTR_RW(protection_type);
 336
 337 static ssize_t
 338 protection_mode_show(struct device *dev, struct device_attribute *attr,
 339                      char *buf)
 340 {
 341         struct scsi_disk *sdkp = to_scsi_disk(dev);
 342         struct scsi_device *sdp = sdkp->device;
 343         unsigned int dif, dix;
 344
 345         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 346         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 347
 348         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 349                 dif = 0;
 350                 dix = 1;
 351         }
 352
 353         if (!dif && !dix)
 354                 return sprintf(buf, "none\n");
 355
 356         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 357 }
 358 static DEVICE_ATTR_RO(protection_mode);
 359
 360 static ssize_t
 361 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 362 {
 363         struct scsi_disk *sdkp = to_scsi_disk(dev);
 364
 365         return sprintf(buf, "%u\n", sdkp->ATO);
 366 }
 367 static DEVICE_ATTR_RO(app_tag_own);
 368
 369 static ssize_t
 370 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 371                        char *buf)
 372 {
 373         struct scsi_disk *sdkp = to_scsi_disk(dev);
 374
 375         return sprintf(buf, "%u\n", sdkp->lbpme);
 376 }
 377 static DEVICE_ATTR_RO(thin_provisioning);
 378
 379 /* sysfs_match_string() requires dense arrays */
 380 static const char *lbp_mode[] = {
 381         [SD_LBP_FULL]           = "full",
 382         [SD_LBP_UNMAP]          = "unmap",
 383         [SD_LBP_WS16]           = "writesame_16",
 384         [SD_LBP_WS10]           = "writesame_10",
 385         [SD_LBP_ZERO]           = "writesame_zero",
 386         [SD_LBP_DISABLE]        = "disabled",
 387 };
 388
 389 static ssize_t
 390 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 391                        char *buf)
 392 {
 393         struct scsi_disk *sdkp = to_scsi_disk(dev);
 394
 395         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 396 }
 397
 398 static ssize_t
 399 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 400                         const char *buf, size_t count)
 401 {
 402         struct scsi_disk *sdkp = to_scsi_disk(dev);
 403         struct scsi_device *sdp = sdkp->device;
 404         int mode;
 405
 406         if (!capable(CAP_SYS_ADMIN))
 407                 return -EACCES;
 408
 409         if (sd_is_zoned(sdkp)) {
 410                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 411                 return count;
 412         }
 413
 414         if (sdp->type != TYPE_DISK)
 415                 return -EINVAL;
 416
 417         mode = sysfs_match_string(lbp_mode, buf);
 418         if (mode < 0)
 419                 return -EINVAL;
 420
 421         sd_config_discard(sdkp, mode);
 422
 423         return count;
 424 }
 425 static DEVICE_ATTR_RW(provisioning_mode);
 426
 427 /* sysfs_match_string() requires dense arrays */
 428 static const char *zeroing_mode[] = {
 429         [SD_ZERO_WRITE]         = "write",
 430         [SD_ZERO_WS]            = "writesame",
 431         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 432         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 433 };
 434
 435 static ssize_t
 436 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 437                   char *buf)
 438 {
 439         struct scsi_disk *sdkp = to_scsi_disk(dev);
 440
 441         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 442 }
 443
 444 static ssize_t
 445 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 446                    const char *buf, size_t count)
 447 {
 448         struct scsi_disk *sdkp = to_scsi_disk(dev);
 449         int mode;
 450
 451         if (!capable(CAP_SYS_ADMIN))
 452                 return -EACCES;
 453
 454         mode = sysfs_match_string(zeroing_mode, buf);
 455         if (mode < 0)
 456                 return -EINVAL;
 457
 458         sdkp->zeroing_mode = mode;
 459
 460         return count;
 461 }
 462 static DEVICE_ATTR_RW(zeroing_mode);
 463
 464 static ssize_t
 465 max_medium_access_timeouts_show(struct device *dev,
 466                                 struct device_attribute *attr, char *buf)
 467 {
 468         struct scsi_disk *sdkp = to_scsi_disk(dev);
 469
 470         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 471 }
 472
 473 static ssize_t
 474 max_medium_access_timeouts_store(struct device *dev,
 475                                  struct device_attribute *attr, const char *buf,
 476                                  size_t count)
 477 {
 478         struct scsi_disk *sdkp = to_scsi_disk(dev);
 479         int err;
 480
 481         if (!capable(CAP_SYS_ADMIN))
 482                 return -EACCES;
 483
 484         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 485
 486         return err ? err : count;
 487 }
 488 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 489
 490 static ssize_t
 491 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 492                            char *buf)
 493 {
 494         struct scsi_disk *sdkp = to_scsi_disk(dev);
 495
 496         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 497 }
 498
 499 static ssize_t
 500 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 501                             const char *buf, size_t count)
 502 {
 503         struct scsi_disk *sdkp = to_scsi_disk(dev);
 504         struct scsi_device *sdp = sdkp->device;
 505         unsigned long max;
 506         int err;
 507
 508         if (!capable(CAP_SYS_ADMIN))
 509                 return -EACCES;
 510
 511         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 512                 return -EINVAL;
 513
 514         err = kstrtoul(buf, 10, &max);
 515
 516         if (err)
 517                 return err;
 518
 519         if (max == 0)
 520                 sdp->no_write_same = 1;
 521         else if (max <= SD_MAX_WS16_BLOCKS) {
 522                 sdp->no_write_same = 0;
 523                 sdkp->max_ws_blocks = max;
 524         }
 525
 526         sd_config_write_same(sdkp);
 527
 528         return count;
 529 }
 530 static DEVICE_ATTR_RW(max_write_same_blocks);
 531
 532 static struct attribute *sd_disk_attrs[] = {
 533         &dev_attr_cache_type.attr,
 534         &dev_attr_FUA.attr,
 535         &dev_attr_allow_restart.attr,
 536         &dev_attr_manage_start_stop.attr,
 537         &dev_attr_protection_type.attr,
 538         &dev_attr_protection_mode.attr,
 539         &dev_attr_app_tag_own.attr,
 540         &dev_attr_thin_provisioning.attr,
 541         &dev_attr_provisioning_mode.attr,
 542         &dev_attr_zeroing_mode.attr,
 543         &dev_attr_max_write_same_blocks.attr,
 544         &dev_attr_max_medium_access_timeouts.attr,
 545         NULL,
 546 };
 547 ATTRIBUTE_GROUPS(sd_disk);
 548
 549 static struct class sd_disk_class = {
 550         .name           = "scsi_disk",
 551         .owner          = THIS_MODULE,
 552         .dev_release    = scsi_disk_release,
 553         .dev_groups     = sd_disk_groups,
 554 };
 555
 556 static const struct dev_pm_ops sd_pm_ops = {
 557         .suspend                = sd_suspend_system,
 558         .resume                 = sd_resume,
 559         .poweroff               = sd_suspend_system,
 560         .restore                = sd_resume,
 561         .runtime_suspend        = sd_suspend_runtime,
 562         .runtime_resume         = sd_resume,
 563 };
 564
 565 static struct scsi_driver sd_template = {
 566         .gendrv = {
 567                 .name           = "sd",
 568                 .owner          = THIS_MODULE,
 569                 .probe          = sd_probe,
 570                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
 571                 .remove         = sd_remove,
 572                 .shutdown       = sd_shutdown,
 573                 .pm             = &sd_pm_ops,
 574         },
 575         .rescan                 = sd_rescan,
 576         .init_command           = sd_init_command,
 577         .uninit_command         = sd_uninit_command,
 578         .done                   = sd_done,
 579         .eh_action              = sd_eh_action,
 580         .eh_reset               = sd_eh_reset,
 581 };
 582
 583 /*
 584  * Dummy kobj_map->probe function.
 585  * The default ->probe function will call modprobe, which is
 586  * pointless as this module is already loaded.
 587  */
 588 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
 589 {
 590         return NULL;
 591 }
 592
 593 /*
 594  * Device no to disk mapping:
 595  *
 596  *       major         disc2     disc  p1
 597  *   |............|.............|....|....| <- dev_t
 598  *    31        20 19          8 7  4 3  0
 599  *
 600  * Inside a major, we have 16k disks, however mapped non-
 601  * contiguously. The first 16 disks are for major0, the next
 602  * ones with major1, ... Disk 256 is for major0 again, disk 272
 603  * for major1, ...
 604  * As we stay compatible with our numbering scheme, we can reuse
 605  * the well-know SCSI majors 8, 65--71, 136--143.
 606  */
 607 static int sd_major(int major_idx)
 608 {
 609         switch (major_idx) {
 610         case 0:
 611                 return SCSI_DISK0_MAJOR;
 612         case 1 ... 7:
 613                 return SCSI_DISK1_MAJOR + major_idx - 1;
 614         case 8 ... 15:
 615                 return SCSI_DISK8_MAJOR + major_idx - 8;
 616         default:
 617                 BUG();
 618                 return 0;       /* shut up gcc */
 619         }
 620 }
 621
 622 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 623 {
 624         struct scsi_disk *sdkp = NULL;
 625
 626         mutex_lock(&sd_ref_mutex);
 627
 628         if (disk->private_data) {
 629                 sdkp = scsi_disk(disk);
 630                 if (scsi_device_get(sdkp->device) == 0)
 631                         get_device(&sdkp->dev);
 632                 else
 633                         sdkp = NULL;
 634         }
 635         mutex_unlock(&sd_ref_mutex);
 636         return sdkp;
 637 }
 638
 639 static void scsi_disk_put(struct scsi_disk *sdkp)
 640 {
 641         struct scsi_device *sdev = sdkp->device;
 642
 643         mutex_lock(&sd_ref_mutex);
 644         put_device(&sdkp->dev);
 645         scsi_device_put(sdev);
 646         mutex_unlock(&sd_ref_mutex);
 647 }
 648
 649 #ifdef CONFIG_BLK_SED_OPAL
 650 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 651                 size_t len, bool send)
 652 {
 653         struct scsi_device *sdev = data;
 654         u8 cdb[12] = { 0, };
 655         int ret;
 656
 657         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 658         cdb[1] = secp;
 659         put_unaligned_be16(spsp, &cdb[2]);
 660         put_unaligned_be32(len, &cdb[6]);
 661
 662         ret = scsi_execute_req(sdev, cdb,
 663                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 664                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
 665         return ret <= 0 ? ret : -EIO;
 666 }
 667 #endif /* CONFIG_BLK_SED_OPAL */
 668
 669 /*
 670  * Look up the DIX operation based on whether the command is read or
 671  * write and whether dix and dif are enabled.
 672  */
 673 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 674 {
 675         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 676         static const unsigned int ops[] = {     /* wrt dix dif */
 677                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 678                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 679                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 680                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 681                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 682                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 683                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 684                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 685         };
 686
 687         return ops[write << 2 | dix << 1 | dif];
 688 }
 689
 690 /*
 691  * Returns a mask of the protection flags that are valid for a given DIX
 692  * operation.
 693  */
 694 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 695 {
 696         static const unsigned int flag_mask[] = {
 697                 [SCSI_PROT_NORMAL]              = 0,
 698
 699                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 700                                                   SCSI_PROT_GUARD_CHECK |
 701                                                   SCSI_PROT_REF_CHECK |
 702                                                   SCSI_PROT_REF_INCREMENT,
 703
 704                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 705                                                   SCSI_PROT_IP_CHECKSUM,
 706
 707                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 708                                                   SCSI_PROT_GUARD_CHECK |
 709                                                   SCSI_PROT_REF_CHECK |
 710                                                   SCSI_PROT_REF_INCREMENT |
 711                                                   SCSI_PROT_IP_CHECKSUM,
 712
 713                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 714                                                   SCSI_PROT_REF_INCREMENT,
 715
 716                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 717                                                   SCSI_PROT_REF_CHECK |
 718                                                   SCSI_PROT_REF_INCREMENT |
 719                                                   SCSI_PROT_IP_CHECKSUM,
 720
 721                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 722                                                   SCSI_PROT_GUARD_CHECK |
 723                                                   SCSI_PROT_REF_CHECK |
 724                                                   SCSI_PROT_REF_INCREMENT |
 725                                                   SCSI_PROT_IP_CHECKSUM,
 726         };
 727
 728         return flag_mask[prot_op];
 729 }
 730
 731 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 732                                            unsigned int dix, unsigned int dif)
 733 {
 734         struct bio *bio = scmd->request->bio;
 735         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
 736         unsigned int protect = 0;
 737
 738         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 739                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 740                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 741
 742                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 743                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 744         }
 745
 746         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 747                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 748
 749                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 750                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 751         }
 752
 753         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 754                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 755
 756                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 757                         protect = 3 << 5;       /* Disable target PI checking */
 758                 else
 759                         protect = 1 << 5;       /* Enable target PI checking */
 760         }
 761
 762         scsi_set_prot_op(scmd, prot_op);
 763         scsi_set_prot_type(scmd, dif);
 764         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 765
 766         return protect;
 767 }
 768
 769 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 770 {
 771         struct request_queue *q = sdkp->disk->queue;
 772         unsigned int logical_block_size = sdkp->device->sector_size;
 773         unsigned int max_blocks = 0;
 774
 775         q->limits.discard_alignment =
 776                 sdkp->unmap_alignment * logical_block_size;
 777         q->limits.discard_granularity =
 778                 max(sdkp->physical_block_size,
 779                     sdkp->unmap_granularity * logical_block_size);
 780         sdkp->provisioning_mode = mode;
 781
 782         switch (mode) {
 783
 784         case SD_LBP_FULL:
 785         case SD_LBP_DISABLE:
 786                 blk_queue_max_discard_sectors(q, 0);
 787                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 788                 return;
 789
 790         case SD_LBP_UNMAP:
 791                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 792                                           (u32)SD_MAX_WS16_BLOCKS);
 793                 break;
 794
 795         case SD_LBP_WS16:
 796                 if (sdkp->device->unmap_limit_for_ws)
 797                         max_blocks = sdkp->max_unmap_blocks;
 798                 else
 799                         max_blocks = sdkp->max_ws_blocks;
 800
 801                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 802                 break;
 803
 804         case SD_LBP_WS10:
 805                 if (sdkp->device->unmap_limit_for_ws)
 806                         max_blocks = sdkp->max_unmap_blocks;
 807                 else
 808                         max_blocks = sdkp->max_ws_blocks;
 809
 810                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 811                 break;
 812
 813         case SD_LBP_ZERO:
 814                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 815                                           (u32)SD_MAX_WS10_BLOCKS);
 816                 break;
 817         }
 818
 819         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 820         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 821 }
 822
 823 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 824 {
 825         struct scsi_device *sdp = cmd->device;
 826         struct request *rq = cmd->request;
 827         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 828         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 829         unsigned int data_len = 24;
 830         char *buf;
 831
 832         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 833         if (!rq->special_vec.bv_page)
 834                 return BLK_STS_RESOURCE;
 835         clear_highpage(rq->special_vec.bv_page);
 836         rq->special_vec.bv_offset = 0;
 837         rq->special_vec.bv_len = data_len;
 838         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 839
 840         cmd->cmd_len = 10;
 841         cmd->cmnd[0] = UNMAP;
 842         cmd->cmnd[8] = 24;
 843
 844         buf = page_address(rq->special_vec.bv_page);
 845         put_unaligned_be16(6 + 16, &buf[0]);
 846         put_unaligned_be16(16, &buf[2]);
 847         put_unaligned_be64(lba, &buf[8]);
 848         put_unaligned_be32(nr_blocks, &buf[16]);
 849
 850         cmd->allowed = SD_MAX_RETRIES;
 851         cmd->transfersize = data_len;
 852         rq->timeout = SD_TIMEOUT;
 853
 854         return scsi_init_io(cmd);
 855 }
 856
 857 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 858                 bool unmap)
 859 {
 860         struct scsi_device *sdp = cmd->device;
 861         struct request *rq = cmd->request;
 862         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 863         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 864         u32 data_len = sdp->sector_size;
 865
 866         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 867         if (!rq->special_vec.bv_page)
 868                 return BLK_STS_RESOURCE;
 869         clear_highpage(rq->special_vec.bv_page);
 870         rq->special_vec.bv_offset = 0;
 871         rq->special_vec.bv_len = data_len;
 872         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 873
 874         cmd->cmd_len = 16;
 875         cmd->cmnd[0] = WRITE_SAME_16;
 876         if (unmap)
 877                 cmd->cmnd[1] = 0x8; /* UNMAP */
 878         put_unaligned_be64(lba, &cmd->cmnd[2]);
 879         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 880
 881         cmd->allowed = SD_MAX_RETRIES;
 882         cmd->transfersize = data_len;
 883         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 884
 885         return scsi_init_io(cmd);
 886 }
 887
 888 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 889                 bool unmap)
 890 {
 891         struct scsi_device *sdp = cmd->device;
 892         struct request *rq = cmd->request;
 893         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 894         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 895         u32 data_len = sdp->sector_size;
 896
 897         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 898         if (!rq->special_vec.bv_page)
 899                 return BLK_STS_RESOURCE;
 900         clear_highpage(rq->special_vec.bv_page);
 901         rq->special_vec.bv_offset = 0;
 902         rq->special_vec.bv_len = data_len;
 903         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 904
 905         cmd->cmd_len = 10;
 906         cmd->cmnd[0] = WRITE_SAME;
 907         if (unmap)
 908                 cmd->cmnd[1] = 0x8; /* UNMAP */
 909         put_unaligned_be32(lba, &cmd->cmnd[2]);
 910         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 911
 912         cmd->allowed = SD_MAX_RETRIES;
 913         cmd->transfersize = data_len;
 914         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 915
 916         return scsi_init_io(cmd);
 917 }
 918
 919 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 920 {
 921         struct request *rq = cmd->request;
 922         struct scsi_device *sdp = cmd->device;
 923         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 924         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 925         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 926
 927         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 928                 switch (sdkp->zeroing_mode) {
 929                 case SD_ZERO_WS16_UNMAP:
 930                         return sd_setup_write_same16_cmnd(cmd, true);
 931                 case SD_ZERO_WS10_UNMAP:
 932                         return sd_setup_write_same10_cmnd(cmd, true);
 933                 }
 934         }
 935
 936         if (sdp->no_write_same)
 937                 return BLK_STS_TARGET;
 938
 939         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 940                 return sd_setup_write_same16_cmnd(cmd, false);
 941
 942         return sd_setup_write_same10_cmnd(cmd, false);
 943 }
 944
 945 static void sd_config_write_same(struct scsi_disk *sdkp)
 946 {
 947         struct request_queue *q = sdkp->disk->queue;
 948         unsigned int logical_block_size = sdkp->device->sector_size;
 949
 950         if (sdkp->device->no_write_same) {
 951                 sdkp->max_ws_blocks = 0;
 952                 goto out;
 953         }
 954
 955         /* Some devices can not handle block counts above 0xffff despite
 956          * supporting WRITE SAME(16). Consequently we default to 64k
 957          * blocks per I/O unless the device explicitly advertises a
 958          * bigger limit.
 959          */
 960         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
 961                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 962                                                    (u32)SD_MAX_WS16_BLOCKS);
 963         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
 964                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 965                                                    (u32)SD_MAX_WS10_BLOCKS);
 966         else {
 967                 sdkp->device->no_write_same = 1;
 968                 sdkp->max_ws_blocks = 0;
 969         }
 970
 971         if (sdkp->lbprz && sdkp->lbpws)
 972                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
 973         else if (sdkp->lbprz && sdkp->lbpws10)
 974                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
 975         else if (sdkp->max_ws_blocks)
 976                 sdkp->zeroing_mode = SD_ZERO_WS;
 977         else
 978                 sdkp->zeroing_mode = SD_ZERO_WRITE;
 979
 980         if (sdkp->max_ws_blocks &&
 981             sdkp->physical_block_size > logical_block_size) {
 982                 /*
 983                  * Reporting a maximum number of blocks that is not aligned
 984                  * on the device physical size would cause a large write same
 985                  * request to be split into physically unaligned chunks by
 986                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
 987                  * even if the caller of these functions took care to align the
 988                  * large request. So make sure the maximum reported is aligned
 989                  * to the device physical block size. This is only an optional
 990                  * optimization for regular disks, but this is mandatory to
 991                  * avoid failure of large write same requests directed at
 992                  * sequential write required zones of host-managed ZBC disks.
 993                  */
 994                 sdkp->max_ws_blocks =
 995                         round_down(sdkp->max_ws_blocks,
 996                                    bytes_to_logical(sdkp->device,
 997                                                     sdkp->physical_block_size));
 998         }
 999
1000 out:
1001         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1002                                          (logical_block_size >> 9));
1003         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1004                                          (logical_block_size >> 9));
1005 }
1006
1007 /**
1008  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1009  * @cmd: command to prepare
1010  *
1011  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1012  * the preference indicated by the target device.
1013  **/
1014 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1015 {
1016         struct request *rq = cmd->request;
1017         struct scsi_device *sdp = cmd->device;
1018         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1019         struct bio *bio = rq->bio;
1020         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1021         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1022         blk_status_t ret;
1023
1024         if (sdkp->device->no_write_same)
1025                 return BLK_STS_TARGET;
1026
1027         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1028
1029         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1030
1031         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1032                 cmd->cmd_len = 16;
1033                 cmd->cmnd[0] = WRITE_SAME_16;
1034                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1035                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1036         } else {
1037                 cmd->cmd_len = 10;
1038                 cmd->cmnd[0] = WRITE_SAME;
1039                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1040                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1041         }
1042
1043         cmd->transfersize = sdp->sector_size;
1044         cmd->allowed = SD_MAX_RETRIES;
1045
1046         /*
1047          * For WRITE SAME the data transferred via the DATA OUT buffer is
1048          * different from the amount of data actually written to the target.
1049          *
1050          * We set up __data_len to the amount of data transferred via the
1051          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1052          * to transfer a single sector of data first, but then reset it to
1053          * the amount of data to be written right after so that the I/O path
1054          * knows how much to actually write.
1055          */
1056         rq->__data_len = sdp->sector_size;
1057         ret = scsi_init_io(cmd);
1058         rq->__data_len = blk_rq_bytes(rq);
1059
1060         return ret;
1061 }
1062
1063 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1064 {
1065         struct request *rq = cmd->request;
1066
1067         /* flush requests don't perform I/O, zero the S/G table */
1068         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1069
1070         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1071         cmd->cmd_len = 10;
1072         cmd->transfersize = 0;
1073         cmd->allowed = SD_MAX_RETRIES;
1074
1075         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1076         return BLK_STS_OK;
1077 }
1078
1079 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1080                                        sector_t lba, unsigned int nr_blocks,
1081                                        unsigned char flags)
1082 {
1083         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1084         if (unlikely(cmd->cmnd == NULL))
1085                 return BLK_STS_RESOURCE;
1086
1087         cmd->cmd_len = SD_EXT_CDB_SIZE;
1088         memset(cmd->cmnd, 0, cmd->cmd_len);
1089
1090         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1091         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1092         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1093         cmd->cmnd[10] = flags;
1094         put_unaligned_be64(lba, &cmd->cmnd[12]);
1095         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1096         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1097
1098         return BLK_STS_OK;
1099 }
1100
1101 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1102                                        sector_t lba, unsigned int nr_blocks,
1103                                        unsigned char flags)
1104 {
1105         cmd->cmd_len  = 16;
1106         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1107         cmd->cmnd[1]  = flags;
1108         cmd->cmnd[14] = 0;
1109         cmd->cmnd[15] = 0;
1110         put_unaligned_be64(lba, &cmd->cmnd[2]);
1111         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1112
1113         return BLK_STS_OK;
1114 }
1115
1116 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1117                                        sector_t lba, unsigned int nr_blocks,
1118                                        unsigned char flags)
1119 {
1120         cmd->cmd_len = 10;
1121         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1122         cmd->cmnd[1] = flags;
1123         cmd->cmnd[6] = 0;
1124         cmd->cmnd[9] = 0;
1125         put_unaligned_be32(lba, &cmd->cmnd[2]);
1126         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1127
1128         return BLK_STS_OK;
1129 }
1130
1131 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1132                                       sector_t lba, unsigned int nr_blocks,
1133                                       unsigned char flags)
1134 {
1135         /* Avoid that 0 blocks gets translated into 256 blocks. */
1136         if (WARN_ON_ONCE(nr_blocks == 0))
1137                 return BLK_STS_IOERR;
1138
1139         if (unlikely(flags & 0x8)) {
1140                 /*
1141                  * This happens only if this drive failed 10byte rw
1142                  * command with ILLEGAL_REQUEST during operation and
1143                  * thus turned off use_10_for_rw.
1144                  */
1145                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1146                 return BLK_STS_IOERR;
1147         }
1148
1149         cmd->cmd_len = 6;
1150         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1151         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1152         cmd->cmnd[2] = (lba >> 8) & 0xff;
1153         cmd->cmnd[3] = lba & 0xff;
1154         cmd->cmnd[4] = nr_blocks;
1155         cmd->cmnd[5] = 0;
1156
1157         return BLK_STS_OK;
1158 }
1159
1160 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1161 {
1162         struct request *rq = cmd->request;
1163         struct scsi_device *sdp = cmd->device;
1164         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1165         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1166         sector_t threshold;
1167         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1168         bool dif, dix;
1169         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170         bool write = rq_data_dir(rq) == WRITE;
1171         unsigned char protect, fua;
1172         blk_status_t ret;
1173
1174         ret = scsi_init_io(cmd);
1175         if (ret != BLK_STS_OK)
1176                 return ret;
1177
1178         if (!scsi_device_online(sdp) || sdp->changed) {
1179                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1180                 return BLK_STS_IOERR;
1181         }
1182
1183         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1184                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1185                 return BLK_STS_IOERR;
1186         }
1187
1188         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1189                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1190                 return BLK_STS_IOERR;
1191         }
1192
1193         /*
1194          * Some SD card readers can't handle accesses which touch the
1195          * last one or two logical blocks. Split accesses as needed.
1196          */
1197         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1198
1199         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1200                 if (lba < threshold) {
1201                         /* Access up to the threshold but not beyond */
1202                         nr_blocks = threshold - lba;
1203                 } else {
1204                         /* Access only a single logical block */
1205                         nr_blocks = 1;
1206                 }
1207         }
1208
1209         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1210         dix = scsi_prot_sg_count(cmd);
1211         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1212
1213         if (write && dix)
1214                 t10_pi_prepare(cmd->request, sdkp->protection_type);
1215
1216         if (dif || dix)
1217                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1218         else
1219                 protect = 0;
1220
1221         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1222                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1223                                          protect | fua);
1224         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1225                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1226                                          protect | fua);
1227         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1228                    sdp->use_10_for_rw || protect) {
1229                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1230                                          protect | fua);
1231         } else {
1232                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1233                                         protect | fua);
1234         }
1235
1236         if (unlikely(ret != BLK_STS_OK))
1237                 return ret;
1238
1239         /*
1240          * We shouldn't disconnect in the middle of a sector, so with a dumb
1241          * host adapter, it's safe to assume that we can at least transfer
1242          * this many bytes between each connect / disconnect.
1243          */
1244         cmd->transfersize = sdp->sector_size;
1245         cmd->underflow = nr_blocks << 9;
1246         cmd->allowed = SD_MAX_RETRIES;
1247         cmd->sdb.length = nr_blocks * sdp->sector_size;
1248
1249         SCSI_LOG_HLQUEUE(1,
1250                          scmd_printk(KERN_INFO, cmd,
1251                                      "%s: block=%llu, count=%d\n", __func__,
1252                                      (unsigned long long)blk_rq_pos(rq),
1253                                      blk_rq_sectors(rq)));
1254         SCSI_LOG_HLQUEUE(2,
1255                          scmd_printk(KERN_INFO, cmd,
1256                                      "%s %d/%u 512 byte blocks.\n",
1257                                      write ? "writing" : "reading", nr_blocks,
1258                                      blk_rq_sectors(rq)));
1259
1260         /*
1261          * This indicates that the command is ready from our end to be
1262          * queued.
1263          */
1264         return BLK_STS_OK;
1265 }
1266
1267 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1268 {
1269         struct request *rq = cmd->request;
1270
1271         switch (req_op(rq)) {
1272         case REQ_OP_DISCARD:
1273                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1274                 case SD_LBP_UNMAP:
1275                         return sd_setup_unmap_cmnd(cmd);
1276                 case SD_LBP_WS16:
1277                         return sd_setup_write_same16_cmnd(cmd, true);
1278                 case SD_LBP_WS10:
1279                         return sd_setup_write_same10_cmnd(cmd, true);
1280                 case SD_LBP_ZERO:
1281                         return sd_setup_write_same10_cmnd(cmd, false);
1282                 default:
1283                         return BLK_STS_TARGET;
1284                 }
1285         case REQ_OP_WRITE_ZEROES:
1286                 return sd_setup_write_zeroes_cmnd(cmd);
1287         case REQ_OP_WRITE_SAME:
1288                 return sd_setup_write_same_cmnd(cmd);
1289         case REQ_OP_FLUSH:
1290                 return sd_setup_flush_cmnd(cmd);
1291         case REQ_OP_READ:
1292         case REQ_OP_WRITE:
1293                 return sd_setup_read_write_cmnd(cmd);
1294         case REQ_OP_ZONE_RESET:
1295                 return sd_zbc_setup_reset_cmnd(cmd);
1296         default:
1297                 WARN_ON_ONCE(1);
1298                 return BLK_STS_NOTSUPP;
1299         }
1300 }
1301
1302 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1303 {
1304         struct request *rq = SCpnt->request;
1305         u8 *cmnd;
1306
1307         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1308                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1309
1310         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1311                 cmnd = SCpnt->cmnd;
1312                 SCpnt->cmnd = NULL;
1313                 SCpnt->cmd_len = 0;
1314                 mempool_free(cmnd, sd_cdb_pool);
1315         }
1316 }
1317
1318 /**
1319  *      sd_open - open a scsi disk device
1320  *      @bdev: Block device of the scsi disk to open
1321  *      @mode: FMODE_* mask
1322  *
1323  *      Returns 0 if successful. Returns a negated errno value in case
1324  *      of error.
1325  *
1326  *      Note: This can be called from a user context (e.g. fsck(1) )
1327  *      or from within the kernel (e.g. as a result of a mount(1) ).
1328  *      In the latter case @inode and @filp carry an abridged amount
1329  *      of information as noted above.
1330  *
1331  *      Locking: called with bdev->bd_mutex held.
1332  **/
1333 static int sd_open(struct block_device *bdev, fmode_t mode)
1334 {
1335         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1336         struct scsi_device *sdev;
1337         int retval;
1338
1339         if (!sdkp)
1340                 return -ENXIO;
1341
1342         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1343
1344         sdev = sdkp->device;
1345
1346         /*
1347          * If the device is in error recovery, wait until it is done.
1348          * If the device is offline, then disallow any access to it.
1349          */
1350         retval = -ENXIO;
1351         if (!scsi_block_when_processing_errors(sdev))
1352                 goto error_out;
1353
1354         if (sdev->removable || sdkp->write_prot)
1355                 check_disk_change(bdev);
1356
1357         /*
1358          * If the drive is empty, just let the open fail.
1359          */
1360         retval = -ENOMEDIUM;
1361         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1362                 goto error_out;
1363
1364         /*
1365          * If the device has the write protect tab set, have the open fail
1366          * if the user expects to be able to write to the thing.
1367          */
1368         retval = -EROFS;
1369         if (sdkp->write_prot && (mode & FMODE_WRITE))
1370                 goto error_out;
1371
1372         /*
1373          * It is possible that the disk changing stuff resulted in
1374          * the device being taken offline.  If this is the case,
1375          * report this to the user, and don't pretend that the
1376          * open actually succeeded.
1377          */
1378         retval = -ENXIO;
1379         if (!scsi_device_online(sdev))
1380                 goto error_out;
1381
1382         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1383                 if (scsi_block_when_processing_errors(sdev))
1384                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1385         }
1386
1387         return 0;
1388
1389 error_out:
1390         scsi_disk_put(sdkp);
1391         return retval;
1392 }
1393
1394 /**
1395  *      sd_release - invoked when the (last) close(2) is called on this
1396  *      scsi disk.
1397  *      @disk: disk to release
1398  *      @mode: FMODE_* mask
1399  *
1400  *      Returns 0.
1401  *
1402  *      Note: may block (uninterruptible) if error recovery is underway
1403  *      on this disk.
1404  *
1405  *      Locking: called with bdev->bd_mutex held.
1406  **/
1407 static void sd_release(struct gendisk *disk, fmode_t mode)
1408 {
1409         struct scsi_disk *sdkp = scsi_disk(disk);
1410         struct scsi_device *sdev = sdkp->device;
1411
1412         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1413
1414         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1415                 if (scsi_block_when_processing_errors(sdev))
1416                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1417         }
1418
1419         /*
1420          * XXX and what if there are packets in flight and this close()
1421          * XXX is followed by a "rmmod sd_mod"?
1422          */
1423
1424         scsi_disk_put(sdkp);
1425 }
1426
1427 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1428 {
1429         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1430         struct scsi_device *sdp = sdkp->device;
1431         struct Scsi_Host *host = sdp->host;
1432         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1433         int diskinfo[4];
1434
1435         /* default to most commonly used values */
1436         diskinfo[0] = 0x40;     /* 1 << 6 */
1437         diskinfo[1] = 0x20;     /* 1 << 5 */
1438         diskinfo[2] = capacity >> 11;
1439
1440         /* override with calculated, extended default, or driver values */
1441         if (host->hostt->bios_param)
1442                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1443         else
1444                 scsicam_bios_param(bdev, capacity, diskinfo);
1445
1446         geo->heads = diskinfo[0];
1447         geo->sectors = diskinfo[1];
1448         geo->cylinders = diskinfo[2];
1449         return 0;
1450 }
1451
1452 /**
1453  *      sd_ioctl - process an ioctl
1454  *      @bdev: target block device
1455  *      @mode: FMODE_* mask
1456  *      @cmd: ioctl command number
1457  *      @arg: this is third argument given to ioctl(2) system call.
1458  *      Often contains a pointer.
1459  *
1460  *      Returns 0 if successful (some ioctls return positive numbers on
1461  *      success as well). Returns a negated errno value in case of error.
1462  *
1463  *      Note: most ioctls are forward onto the block subsystem or further
1464  *      down in the scsi subsystem.
1465  **/
1466 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1467                     unsigned int cmd, unsigned long arg)
1468 {
1469         struct gendisk *disk = bdev->bd_disk;
1470         struct scsi_disk *sdkp = scsi_disk(disk);
1471         struct scsi_device *sdp = sdkp->device;
1472         void __user *p = (void __user *)arg;
1473         int error;
1474
1475         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1476                                     "cmd=0x%x\n", disk->disk_name, cmd));
1477
1478         error = scsi_verify_blk_ioctl(bdev, cmd);
1479         if (error < 0)
1480                 return error;
1481
1482         /*
1483          * If we are in the middle of error recovery, don't let anyone
1484          * else try and use this device.  Also, if error recovery fails, it
1485          * may try and take the device offline, in which case all further
1486          * access to the device is prohibited.
1487          */
1488         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1489                         (mode & FMODE_NDELAY) != 0);
1490         if (error)
1491                 goto out;
1492
1493         if (is_sed_ioctl(cmd))
1494                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1495
1496         /*
1497          * Send SCSI addressing ioctls directly to mid level, send other
1498          * ioctls to block level and then onto mid level if they can't be
1499          * resolved.
1500          */
1501         switch (cmd) {
1502                 case SCSI_IOCTL_GET_IDLUN:
1503                 case SCSI_IOCTL_GET_BUS_NUMBER:
1504                         error = scsi_ioctl(sdp, cmd, p);
1505                         break;
1506                 default:
1507                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1508                         if (error != -ENOTTY)
1509                                 break;
1510                         error = scsi_ioctl(sdp, cmd, p);
1511                         break;
1512         }
1513 out:
1514         return error;
1515 }
1516
1517 static void set_media_not_present(struct scsi_disk *sdkp)
1518 {
1519         if (sdkp->media_present)
1520                 sdkp->device->changed = 1;
1521
1522         if (sdkp->device->removable) {
1523                 sdkp->media_present = 0;
1524                 sdkp->capacity = 0;
1525         }
1526 }
1527
1528 static int media_not_present(struct scsi_disk *sdkp,
1529                              struct scsi_sense_hdr *sshdr)
1530 {
1531         if (!scsi_sense_valid(sshdr))
1532                 return 0;
1533
1534         /* not invoked for commands that could return deferred errors */
1535         switch (sshdr->sense_key) {
1536         case UNIT_ATTENTION:
1537         case NOT_READY:
1538                 /* medium not present */
1539                 if (sshdr->asc == 0x3A) {
1540                         set_media_not_present(sdkp);
1541                         return 1;
1542                 }
1543         }
1544         return 0;
1545 }
1546
1547 /**
1548  *      sd_check_events - check media events
1549  *      @disk: kernel device descriptor
1550  *      @clearing: disk events currently being cleared
1551  *
1552  *      Returns mask of DISK_EVENT_*.
1553  *
1554  *      Note: this function is invoked from the block subsystem.
1555  **/
1556 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1557 {
1558         struct scsi_disk *sdkp = scsi_disk_get(disk);
1559         struct scsi_device *sdp;
1560         int retval;
1561
1562         if (!sdkp)
1563                 return 0;
1564
1565         sdp = sdkp->device;
1566         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1567
1568         /*
1569          * If the device is offline, don't send any commands - just pretend as
1570          * if the command failed.  If the device ever comes back online, we
1571          * can deal with it then.  It is only because of unrecoverable errors
1572          * that we would ever take a device offline in the first place.
1573          */
1574         if (!scsi_device_online(sdp)) {
1575                 set_media_not_present(sdkp);
1576                 goto out;
1577         }
1578
1579         /*
1580          * Using TEST_UNIT_READY enables differentiation between drive with
1581          * no cartridge loaded - NOT READY, drive with changed cartridge -
1582          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1583          *
1584          * Drives that auto spin down. eg iomega jaz 1G, will be started
1585          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1586          * sd_revalidate() is called.
1587          */
1588         if (scsi_block_when_processing_errors(sdp)) {
1589                 struct scsi_sense_hdr sshdr = { 0, };
1590
1591                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1592                                               &sshdr);
1593
1594                 /* failed to execute TUR, assume media not present */
1595                 if (host_byte(retval)) {
1596                         set_media_not_present(sdkp);
1597                         goto out;
1598                 }
1599
1600                 if (media_not_present(sdkp, &sshdr))
1601                         goto out;
1602         }
1603
1604         /*
1605          * For removable scsi disk we have to recognise the presence
1606          * of a disk in the drive.
1607          */
1608         if (!sdkp->media_present)
1609                 sdp->changed = 1;
1610         sdkp->media_present = 1;
1611 out:
1612         /*
1613          * sdp->changed is set under the following conditions:
1614          *
1615          *      Medium present state has changed in either direction.
1616          *      Device has indicated UNIT_ATTENTION.
1617          */
1618         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1619         sdp->changed = 0;
1620         scsi_disk_put(sdkp);
1621         return retval;
1622 }
1623
1624 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1625 {
1626         int retries, res;
1627         struct scsi_device *sdp = sdkp->device;
1628         const int timeout = sdp->request_queue->rq_timeout
1629                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1630         struct scsi_sense_hdr my_sshdr;
1631
1632         if (!scsi_device_online(sdp))
1633                 return -ENODEV;
1634
1635         /* caller might not be interested in sense, but we need it */
1636         if (!sshdr)
1637                 sshdr = &my_sshdr;
1638
1639         for (retries = 3; retries > 0; --retries) {
1640                 unsigned char cmd[10] = { 0 };
1641
1642                 cmd[0] = SYNCHRONIZE_CACHE;
1643                 /*
1644                  * Leave the rest of the command zero to indicate
1645                  * flush everything.
1646                  */
1647                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1648                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1649                 if (res == 0)
1650                         break;
1651         }
1652
1653         if (res) {
1654                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1655
1656                 if (driver_byte(res) == DRIVER_SENSE)
1657                         sd_print_sense_hdr(sdkp, sshdr);
1658
1659                 /* we need to evaluate the error return  */
1660                 if (scsi_sense_valid(sshdr) &&
1661                         (sshdr->asc == 0x3a ||  /* medium not present */
1662                          sshdr->asc == 0x20))   /* invalid command */
1663                                 /* this is no error here */
1664                                 return 0;
1665
1666                 switch (host_byte(res)) {
1667                 /* ignore errors due to racing a disconnection */
1668                 case DID_BAD_TARGET:
1669                 case DID_NO_CONNECT:
1670                         return 0;
1671                 /* signal the upper layer it might try again */
1672                 case DID_BUS_BUSY:
1673                 case DID_IMM_RETRY:
1674                 case DID_REQUEUE:
1675                 case DID_SOFT_ERROR:
1676                         return -EBUSY;
1677                 default:
1678                         return -EIO;
1679                 }
1680         }
1681         return 0;
1682 }
1683
1684 static void sd_rescan(struct device *dev)
1685 {
1686         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1687
1688         revalidate_disk(sdkp->disk);
1689 }
1690
1691
1692 #ifdef CONFIG_COMPAT
1693 /*
1694  * This gets directly called from VFS. When the ioctl
1695  * is not recognized we go back to the other translation paths.
1696  */
1697 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1698                            unsigned int cmd, unsigned long arg)
1699 {
1700         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1701         int error;
1702
1703         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1704                         (mode & FMODE_NDELAY) != 0);
1705         if (error)
1706                 return error;
1707
1708         /*
1709          * Let the static ioctl translation table take care of it.
1710          */
1711         if (!sdev->host->hostt->compat_ioctl)
1712                 return -ENOIOCTLCMD;
1713         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1714 }
1715 #endif
1716
1717 static char sd_pr_type(enum pr_type type)
1718 {
1719         switch (type) {
1720         case PR_WRITE_EXCLUSIVE:
1721                 return 0x01;
1722         case PR_EXCLUSIVE_ACCESS:
1723                 return 0x03;
1724         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1725                 return 0x05;
1726         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1727                 return 0x06;
1728         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1729                 return 0x07;
1730         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1731                 return 0x08;
1732         default:
1733                 return 0;
1734         }
1735 };
1736
1737 static int sd_pr_command(struct block_device *bdev, u8 sa,
1738                 u64 key, u64 sa_key, u8 type, u8 flags)
1739 {
1740         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1741         struct scsi_sense_hdr sshdr;
1742         int result;
1743         u8 cmd[16] = { 0, };
1744         u8 data[24] = { 0, };
1745
1746         cmd[0] = PERSISTENT_RESERVE_OUT;
1747         cmd[1] = sa;
1748         cmd[2] = type;
1749         put_unaligned_be32(sizeof(data), &cmd[5]);
1750
1751         put_unaligned_be64(key, &data[0]);
1752         put_unaligned_be64(sa_key, &data[8]);
1753         data[20] = flags;
1754
1755         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1756                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1757
1758         if (driver_byte(result) == DRIVER_SENSE &&
1759             scsi_sense_valid(&sshdr)) {
1760                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1761                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1762         }
1763
1764         return result;
1765 }
1766
1767 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1768                 u32 flags)
1769 {
1770         if (flags & ~PR_FL_IGNORE_KEY)
1771                 return -EOPNOTSUPP;
1772         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1773                         old_key, new_key, 0,
1774                         (1 << 0) /* APTPL */);
1775 }
1776
1777 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1778                 u32 flags)
1779 {
1780         if (flags)
1781                 return -EOPNOTSUPP;
1782         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1783 }
1784
1785 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1786 {
1787         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1788 }
1789
1790 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1791                 enum pr_type type, bool abort)
1792 {
1793         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1794                              sd_pr_type(type), 0);
1795 }
1796
1797 static int sd_pr_clear(struct block_device *bdev, u64 key)
1798 {
1799         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1800 }
1801
1802 static const struct pr_ops sd_pr_ops = {
1803         .pr_register    = sd_pr_register,
1804         .pr_reserve     = sd_pr_reserve,
1805         .pr_release     = sd_pr_release,
1806         .pr_preempt     = sd_pr_preempt,
1807         .pr_clear       = sd_pr_clear,
1808 };
1809
1810 static const struct block_device_operations sd_fops = {
1811         .owner                  = THIS_MODULE,
1812         .open                   = sd_open,
1813         .release                = sd_release,
1814         .ioctl                  = sd_ioctl,
1815         .getgeo                 = sd_getgeo,
1816 #ifdef CONFIG_COMPAT
1817         .compat_ioctl           = sd_compat_ioctl,
1818 #endif
1819         .check_events           = sd_check_events,
1820         .revalidate_disk        = sd_revalidate_disk,
1821         .unlock_native_capacity = sd_unlock_native_capacity,
1822         .report_zones           = sd_zbc_report_zones,
1823         .pr_ops                 = &sd_pr_ops,
1824 };
1825
1826 /**
1827  *      sd_eh_reset - reset error handling callback
1828  *      @scmd:          sd-issued command that has failed
1829  *
1830  *      This function is called by the SCSI midlayer before starting
1831  *      SCSI EH. When counting medium access failures we have to be
1832  *      careful to register it only only once per device and SCSI EH run;
1833  *      there might be several timed out commands which will cause the
1834  *      'max_medium_access_timeouts' counter to trigger after the first
1835  *      SCSI EH run already and set the device to offline.
1836  *      So this function resets the internal counter before starting SCSI EH.
1837  **/
1838 static void sd_eh_reset(struct scsi_cmnd *scmd)
1839 {
1840         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1841
1842         /* New SCSI EH run, reset gate variable */
1843         sdkp->ignore_medium_access_errors = false;
1844 }
1845
1846 /**
1847  *      sd_eh_action - error handling callback
1848  *      @scmd:          sd-issued command that has failed
1849  *      @eh_disp:       The recovery disposition suggested by the midlayer
1850  *
1851  *      This function is called by the SCSI midlayer upon completion of an
1852  *      error test command (currently TEST UNIT READY). The result of sending
1853  *      the eh command is passed in eh_disp.  We're looking for devices that
1854  *      fail medium access commands but are OK with non access commands like
1855  *      test unit ready (so wrongly see the device as having a successful
1856  *      recovery)
1857  **/
1858 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1859 {
1860         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1861         struct scsi_device *sdev = scmd->device;
1862
1863         if (!scsi_device_online(sdev) ||
1864             !scsi_medium_access_command(scmd) ||
1865             host_byte(scmd->result) != DID_TIME_OUT ||
1866             eh_disp != SUCCESS)
1867                 return eh_disp;
1868
1869         /*
1870          * The device has timed out executing a medium access command.
1871          * However, the TEST UNIT READY command sent during error
1872          * handling completed successfully. Either the device is in the
1873          * process of recovering or has it suffered an internal failure
1874          * that prevents access to the storage medium.
1875          */
1876         if (!sdkp->ignore_medium_access_errors) {
1877                 sdkp->medium_access_timed_out++;
1878                 sdkp->ignore_medium_access_errors = true;
1879         }
1880
1881         /*
1882          * If the device keeps failing read/write commands but TEST UNIT
1883          * READY always completes successfully we assume that medium
1884          * access is no longer possible and take the device offline.
1885          */
1886         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1887                 scmd_printk(KERN_ERR, scmd,
1888                             "Medium access timeout failure. Offlining disk!\n");
1889                 mutex_lock(&sdev->state_mutex);
1890                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1891                 mutex_unlock(&sdev->state_mutex);
1892
1893                 return SUCCESS;
1894         }
1895
1896         return eh_disp;
1897 }
1898
1899 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1900 {
1901         struct request *req = scmd->request;
1902         struct scsi_device *sdev = scmd->device;
1903         unsigned int transferred, good_bytes;
1904         u64 start_lba, end_lba, bad_lba;
1905
1906         /*
1907          * Some commands have a payload smaller than the device logical
1908          * block size (e.g. INQUIRY on a 4K disk).
1909          */
1910         if (scsi_bufflen(scmd) <= sdev->sector_size)
1911                 return 0;
1912
1913         /* Check if we have a 'bad_lba' information */
1914         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1915                                      SCSI_SENSE_BUFFERSIZE,
1916                                      &bad_lba))
1917                 return 0;
1918
1919         /*
1920          * If the bad lba was reported incorrectly, we have no idea where
1921          * the error is.
1922          */
1923         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1924         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1925         if (bad_lba < start_lba || bad_lba >= end_lba)
1926                 return 0;
1927
1928         /*
1929          * resid is optional but mostly filled in.  When it's unused,
1930          * its value is zero, so we assume the whole buffer transferred
1931          */
1932         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1933
1934         /* This computation should always be done in terms of the
1935          * resolution of the device's medium.
1936          */
1937         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1938
1939         return min(good_bytes, transferred);
1940 }
1941
1942 /**
1943  *      sd_done - bottom half handler: called when the lower level
1944  *      driver has completed (successfully or otherwise) a scsi command.
1945  *      @SCpnt: mid-level's per command structure.
1946  *
1947  *      Note: potentially run from within an ISR. Must not block.
1948  **/
1949 static int sd_done(struct scsi_cmnd *SCpnt)
1950 {
1951         int result = SCpnt->result;
1952         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1953         unsigned int sector_size = SCpnt->device->sector_size;
1954         unsigned int resid;
1955         struct scsi_sense_hdr sshdr;
1956         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1957         struct request *req = SCpnt->request;
1958         int sense_valid = 0;
1959         int sense_deferred = 0;
1960
1961         switch (req_op(req)) {
1962         case REQ_OP_DISCARD:
1963         case REQ_OP_WRITE_ZEROES:
1964         case REQ_OP_WRITE_SAME:
1965         case REQ_OP_ZONE_RESET:
1966                 if (!result) {
1967                         good_bytes = blk_rq_bytes(req);
1968                         scsi_set_resid(SCpnt, 0);
1969                 } else {
1970                         good_bytes = 0;
1971                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1972                 }
1973                 break;
1974         default:
1975                 /*
1976                  * In case of bogus fw or device, we could end up having
1977                  * an unaligned partial completion. Check this here and force
1978                  * alignment.
1979                  */
1980                 resid = scsi_get_resid(SCpnt);
1981                 if (resid & (sector_size - 1)) {
1982                         sd_printk(KERN_INFO, sdkp,
1983                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1984                                 resid, sector_size);
1985                         resid = min(scsi_bufflen(SCpnt),
1986                                     round_up(resid, sector_size));
1987                         scsi_set_resid(SCpnt, resid);
1988                 }
1989         }
1990
1991         if (result) {
1992                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1993                 if (sense_valid)
1994                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1995         }
1996         sdkp->medium_access_timed_out = 0;
1997
1998         if (driver_byte(result) != DRIVER_SENSE &&
1999             (!sense_valid || sense_deferred))
2000                 goto out;
2001
2002         switch (sshdr.sense_key) {
2003         case HARDWARE_ERROR:
2004         case MEDIUM_ERROR:
2005                 good_bytes = sd_completed_bytes(SCpnt);
2006                 break;
2007         case RECOVERED_ERROR:
2008                 good_bytes = scsi_bufflen(SCpnt);
2009                 break;
2010         case NO_SENSE:
2011                 /* This indicates a false check condition, so ignore it.  An
2012                  * unknown amount of data was transferred so treat it as an
2013                  * error.
2014                  */
2015                 SCpnt->result = 0;
2016                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2017                 break;
2018         case ABORTED_COMMAND:
2019                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2020                         good_bytes = sd_completed_bytes(SCpnt);
2021                 break;
2022         case ILLEGAL_REQUEST:
2023                 switch (sshdr.asc) {
2024                 case 0x10:      /* DIX: Host detected corruption */
2025                         good_bytes = sd_completed_bytes(SCpnt);
2026                         break;
2027                 case 0x20:      /* INVALID COMMAND OPCODE */
2028                 case 0x24:      /* INVALID FIELD IN CDB */
2029                         switch (SCpnt->cmnd[0]) {
2030                         case UNMAP:
2031                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2032                                 break;
2033                         case WRITE_SAME_16:
2034                         case WRITE_SAME:
2035                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2036                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2037                                 } else {
2038                                         sdkp->device->no_write_same = 1;
2039                                         sd_config_write_same(sdkp);
2040                                         req->rq_flags |= RQF_QUIET;
2041                                 }
2042                                 break;
2043                         }
2044                 }
2045                 break;
2046         default:
2047                 break;
2048         }
2049
2050  out:
2051         if (sd_is_zoned(sdkp))
2052                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2053
2054         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2055                                            "sd_done: completed %d of %d bytes\n",
2056                                            good_bytes, scsi_bufflen(SCpnt)));
2057
2058         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2059             good_bytes)
2060                 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2061                                 good_bytes / scsi_prot_interval(SCpnt));
2062
2063         return good_bytes;
2064 }
2065
2066 /*
2067  * spinup disk - called only in sd_revalidate_disk()
2068  */
2069 static void
2070 sd_spinup_disk(struct scsi_disk *sdkp)
2071 {
2072         unsigned char cmd[10];
2073         unsigned long spintime_expire = 0;
2074         int retries, spintime;
2075         unsigned int the_result;
2076         struct scsi_sense_hdr sshdr;
2077         int sense_valid = 0;
2078
2079         spintime = 0;
2080
2081         /* Spin up drives, as required.  Only do this at boot time */
2082         /* Spinup needs to be done for module loads too. */
2083         do {
2084                 retries = 0;
2085
2086                 do {
2087                         cmd[0] = TEST_UNIT_READY;
2088                         memset((void *) &cmd[1], 0, 9);
2089
2090                         the_result = scsi_execute_req(sdkp->device, cmd,
2091                                                       DMA_NONE, NULL, 0,
2092                                                       &sshdr, SD_TIMEOUT,
2093                                                       SD_MAX_RETRIES, NULL);
2094
2095                         /*
2096                          * If the drive has indicated to us that it
2097                          * doesn't have any media in it, don't bother
2098                          * with any more polling.
2099                          */
2100                         if (media_not_present(sdkp, &sshdr))
2101                                 return;
2102
2103                         if (the_result)
2104                                 sense_valid = scsi_sense_valid(&sshdr);
2105                         retries++;
2106                 } while (retries < 3 &&
2107                          (!scsi_status_is_good(the_result) ||
2108                           ((driver_byte(the_result) == DRIVER_SENSE) &&
2109                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2110
2111                 if (driver_byte(the_result) != DRIVER_SENSE) {
2112                         /* no sense, TUR either succeeded or failed
2113                          * with a status error */
2114                         if(!spintime && !scsi_status_is_good(the_result)) {
2115                                 sd_print_result(sdkp, "Test Unit Ready failed",
2116                                                 the_result);
2117                         }
2118                         break;
2119                 }
2120
2121                 /*
2122                  * The device does not want the automatic start to be issued.
2123                  */
2124                 if (sdkp->device->no_start_on_add)
2125                         break;
2126
2127                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2128                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2129                                 break;  /* manual intervention required */
2130                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2131                                 break;  /* standby */
2132                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2133                                 break;  /* unavailable */
2134                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2135                                 break;  /* sanitize in progress */
2136                         /*
2137                          * Issue command to spin up drive when not ready
2138                          */
2139                         if (!spintime) {
2140                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2141                                 cmd[0] = START_STOP;
2142                                 cmd[1] = 1;     /* Return immediately */
2143                                 memset((void *) &cmd[2], 0, 8);
2144                                 cmd[4] = 1;     /* Start spin cycle */
2145                                 if (sdkp->device->start_stop_pwr_cond)
2146                                         cmd[4] |= 1 << 4;
2147                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2148                                                  NULL, 0, &sshdr,
2149                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2150                                                  NULL);
2151                                 spintime_expire = jiffies + 100 * HZ;
2152                                 spintime = 1;
2153                         }
2154                         /* Wait 1 second for next try */
2155                         msleep(1000);
2156                         printk(KERN_CONT ".");
2157
2158                 /*
2159                  * Wait for USB flash devices with slow firmware.
2160                  * Yes, this sense key/ASC combination shouldn't
2161                  * occur here.  It's characteristic of these devices.
2162                  */
2163                 } else if (sense_valid &&
2164                                 sshdr.sense_key == UNIT_ATTENTION &&
2165                                 sshdr.asc == 0x28) {
2166                         if (!spintime) {
2167                                 spintime_expire = jiffies + 5 * HZ;
2168                                 spintime = 1;
2169                         }
2170                         /* Wait 1 second for next try */
2171                         msleep(1000);
2172                 } else {
2173                         /* we don't understand the sense code, so it's
2174                          * probably pointless to loop */
2175                         if(!spintime) {
2176                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2177                                 sd_print_sense_hdr(sdkp, &sshdr);
2178                         }
2179                         break;
2180                 }
2181
2182         } while (spintime && time_before_eq(jiffies, spintime_expire));
2183
2184         if (spintime) {
2185                 if (scsi_status_is_good(the_result))
2186                         printk(KERN_CONT "ready\n");
2187                 else
2188                         printk(KERN_CONT "not responding...\n");
2189         }
2190 }
2191
2192 /*
2193  * Determine whether disk supports Data Integrity Field.
2194  */
2195 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2196 {
2197         struct scsi_device *sdp = sdkp->device;
2198         u8 type;
2199         int ret = 0;
2200
2201         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2202                 return ret;
2203
2204         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2205
2206         if (type > T10_PI_TYPE3_PROTECTION)
2207                 ret = -ENODEV;
2208         else if (scsi_host_dif_capable(sdp->host, type))
2209                 ret = 1;
2210
2211         if (sdkp->first_scan || type != sdkp->protection_type)
2212                 switch (ret) {
2213                 case -ENODEV:
2214                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2215                                   " protection type %u. Disabling disk!\n",
2216                                   type);
2217                         break;
2218                 case 1:
2219                         sd_printk(KERN_NOTICE, sdkp,
2220                                   "Enabling DIF Type %u protection\n", type);
2221                         break;
2222                 case 0:
2223                         sd_printk(KERN_NOTICE, sdkp,
2224                                   "Disabling DIF Type %u protection\n", type);
2225                         break;
2226                 }
2227
2228         sdkp->protection_type = type;
2229
2230         return ret;
2231 }
2232
2233 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2234                         struct scsi_sense_hdr *sshdr, int sense_valid,
2235                         int the_result)
2236 {
2237         if (driver_byte(the_result) == DRIVER_SENSE)
2238                 sd_print_sense_hdr(sdkp, sshdr);
2239         else
2240                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2241
2242         /*
2243          * Set dirty bit for removable devices if not ready -
2244          * sometimes drives will not report this properly.
2245          */
2246         if (sdp->removable &&
2247             sense_valid && sshdr->sense_key == NOT_READY)
2248                 set_media_not_present(sdkp);
2249
2250         /*
2251          * We used to set media_present to 0 here to indicate no media
2252          * in the drive, but some drives fail read capacity even with
2253          * media present, so we can't do that.
2254          */
2255         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2256 }
2257
2258 #define RC16_LEN 32
2259 #if RC16_LEN > SD_BUF_SIZE
2260 #error RC16_LEN must not be more than SD_BUF_SIZE
2261 #endif
2262
2263 #define READ_CAPACITY_RETRIES_ON_RESET  10
2264
2265 /*
2266  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2267  * and the reported logical block size is bigger than 512 bytes. Note
2268  * that last_sector is a u64 and therefore logical_to_sectors() is not
2269  * applicable.
2270  */
2271 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2272 {
2273         u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2274
2275         if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2276                 return false;
2277
2278         return true;
2279 }
2280
2281 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2282                                                 unsigned char *buffer)
2283 {
2284         unsigned char cmd[16];
2285         struct scsi_sense_hdr sshdr;
2286         int sense_valid = 0;
2287         int the_result;
2288         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2289         unsigned int alignment;
2290         unsigned long long lba;
2291         unsigned sector_size;
2292
2293         if (sdp->no_read_capacity_16)
2294                 return -EINVAL;
2295
2296         do {
2297                 memset(cmd, 0, 16);
2298                 cmd[0] = SERVICE_ACTION_IN_16;
2299                 cmd[1] = SAI_READ_CAPACITY_16;
2300                 cmd[13] = RC16_LEN;
2301                 memset(buffer, 0, RC16_LEN);
2302
2303                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2304                                         buffer, RC16_LEN, &sshdr,
2305                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2306
2307                 if (media_not_present(sdkp, &sshdr))
2308                         return -ENODEV;
2309
2310                 if (the_result) {
2311                         sense_valid = scsi_sense_valid(&sshdr);
2312                         if (sense_valid &&
2313                             sshdr.sense_key == ILLEGAL_REQUEST &&
2314                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2315                             sshdr.ascq == 0x00)
2316                                 /* Invalid Command Operation Code or
2317                                  * Invalid Field in CDB, just retry
2318                                  * silently with RC10 */
2319                                 return -EINVAL;
2320                         if (sense_valid &&
2321                             sshdr.sense_key == UNIT_ATTENTION &&
2322                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2323                                 /* Device reset might occur several times,
2324                                  * give it one more chance */
2325                                 if (--reset_retries > 0)
2326                                         continue;
2327                 }
2328                 retries--;
2329
2330         } while (the_result && retries);
2331
2332         if (the_result) {
2333                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2334                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2335                 return -EINVAL;
2336         }
2337
2338         sector_size = get_unaligned_be32(&buffer[8]);
2339         lba = get_unaligned_be64(&buffer[0]);
2340
2341         if (sd_read_protection_type(sdkp, buffer) < 0) {
2342                 sdkp->capacity = 0;
2343                 return -ENODEV;
2344         }
2345
2346         if (!sd_addressable_capacity(lba, sector_size)) {
2347                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2348                         "kernel compiled with support for large block "
2349                         "devices.\n");
2350                 sdkp->capacity = 0;
2351                 return -EOVERFLOW;
2352         }
2353
2354         /* Logical blocks per physical block exponent */
2355         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2356
2357         /* RC basis */
2358         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2359
2360         /* Lowest aligned logical block */
2361         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2362         blk_queue_alignment_offset(sdp->request_queue, alignment);
2363         if (alignment && sdkp->first_scan)
2364                 sd_printk(KERN_NOTICE, sdkp,
2365                           "physical block alignment offset: %u\n", alignment);
2366
2367         if (buffer[14] & 0x80) { /* LBPME */
2368                 sdkp->lbpme = 1;
2369
2370                 if (buffer[14] & 0x40) /* LBPRZ */
2371                         sdkp->lbprz = 1;
2372
2373                 sd_config_discard(sdkp, SD_LBP_WS16);
2374         }
2375
2376         sdkp->capacity = lba + 1;
2377         return sector_size;
2378 }
2379
2380 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2381                                                 unsigned char *buffer)
2382 {
2383         unsigned char cmd[16];
2384         struct scsi_sense_hdr sshdr;
2385         int sense_valid = 0;
2386         int the_result;
2387         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2388         sector_t lba;
2389         unsigned sector_size;
2390
2391         do {
2392                 cmd[0] = READ_CAPACITY;
2393                 memset(&cmd[1], 0, 9);
2394                 memset(buffer, 0, 8);
2395
2396                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2397                                         buffer, 8, &sshdr,
2398                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2399
2400                 if (media_not_present(sdkp, &sshdr))
2401                         return -ENODEV;
2402
2403                 if (the_result) {
2404                         sense_valid = scsi_sense_valid(&sshdr);
2405                         if (sense_valid &&
2406                             sshdr.sense_key == UNIT_ATTENTION &&
2407                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2408                                 /* Device reset might occur several times,
2409                                  * give it one more chance */
2410                                 if (--reset_retries > 0)
2411                                         continue;
2412                 }
2413                 retries--;
2414
2415         } while (the_result && retries);
2416
2417         if (the_result) {
2418                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2419                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2420                 return -EINVAL;
2421         }
2422
2423         sector_size = get_unaligned_be32(&buffer[4]);
2424         lba = get_unaligned_be32(&buffer[0]);
2425
2426         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2427                 /* Some buggy (usb cardreader) devices return an lba of
2428                    0xffffffff when the want to report a size of 0 (with
2429                    which they really mean no media is present) */
2430                 sdkp->capacity = 0;
2431                 sdkp->physical_block_size = sector_size;
2432                 return sector_size;
2433         }
2434
2435         if (!sd_addressable_capacity(lba, sector_size)) {
2436                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2437                         "kernel compiled with support for large block "
2438                         "devices.\n");
2439                 sdkp->capacity = 0;
2440                 return -EOVERFLOW;
2441         }
2442
2443         sdkp->capacity = lba + 1;
2444         sdkp->physical_block_size = sector_size;
2445         return sector_size;
2446 }
2447
2448 static int sd_try_rc16_first(struct scsi_device *sdp)
2449 {
2450         if (sdp->host->max_cmd_len < 16)
2451                 return 0;
2452         if (sdp->try_rc_10_first)
2453                 return 0;
2454         if (sdp->scsi_level > SCSI_SPC_2)
2455                 return 1;
2456         if (scsi_device_protection(sdp))
2457                 return 1;
2458         return 0;
2459 }
2460
2461 /*
2462  * read disk capacity
2463  */
2464 static void
2465 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2466 {
2467         int sector_size;
2468         struct scsi_device *sdp = sdkp->device;
2469
2470         if (sd_try_rc16_first(sdp)) {
2471                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2472                 if (sector_size == -EOVERFLOW)
2473                         goto got_data;
2474                 if (sector_size == -ENODEV)
2475                         return;
2476                 if (sector_size < 0)
2477                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2478                 if (sector_size < 0)
2479                         return;
2480         } else {
2481                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2482                 if (sector_size == -EOVERFLOW)
2483                         goto got_data;
2484                 if (sector_size < 0)
2485                         return;
2486                 if ((sizeof(sdkp->capacity) > 4) &&
2487                     (sdkp->capacity > 0xffffffffULL)) {
2488                         int old_sector_size = sector_size;
2489                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2490                                         "Trying to use READ CAPACITY(16).\n");
2491                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2492                         if (sector_size < 0) {
2493                                 sd_printk(KERN_NOTICE, sdkp,
2494                                         "Using 0xffffffff as device size\n");
2495                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2496                                 sector_size = old_sector_size;
2497                                 goto got_data;
2498                         }
2499                         /* Remember that READ CAPACITY(16) succeeded */
2500                         sdp->try_rc_10_first = 0;
2501                 }
2502         }
2503
2504         /* Some devices are known to return the total number of blocks,
2505          * not the highest block number.  Some devices have versions
2506          * which do this and others which do not.  Some devices we might
2507          * suspect of doing this but we don't know for certain.
2508          *
2509          * If we know the reported capacity is wrong, decrement it.  If
2510          * we can only guess, then assume the number of blocks is even
2511          * (usually true but not always) and err on the side of lowering
2512          * the capacity.
2513          */
2514         if (sdp->fix_capacity ||
2515             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2516                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2517                                 "from its reported value: %llu\n",
2518                                 (unsigned long long) sdkp->capacity);
2519                 --sdkp->capacity;
2520         }
2521
2522 got_data:
2523         if (sector_size == 0) {
2524                 sector_size = 512;
2525                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2526                           "assuming 512.\n");
2527         }
2528
2529         if (sector_size != 512 &&
2530             sector_size != 1024 &&
2531             sector_size != 2048 &&
2532             sector_size != 4096) {
2533                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2534                           sector_size);
2535                 /*
2536                  * The user might want to re-format the drive with
2537                  * a supported sectorsize.  Once this happens, it
2538                  * would be relatively trivial to set the thing up.
2539                  * For this reason, we leave the thing in the table.
2540                  */
2541                 sdkp->capacity = 0;
2542                 /*
2543                  * set a bogus sector size so the normal read/write
2544                  * logic in the block layer will eventually refuse any
2545                  * request on this device without tripping over power
2546                  * of two sector size assumptions
2547                  */
2548                 sector_size = 512;
2549         }
2550         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2551         blk_queue_physical_block_size(sdp->request_queue,
2552                                       sdkp->physical_block_size);
2553         sdkp->device->sector_size = sector_size;
2554
2555         if (sdkp->capacity > 0xffffffff)
2556                 sdp->use_16_for_rw = 1;
2557
2558 }
2559
2560 /*
2561  * Print disk capacity
2562  */
2563 static void
2564 sd_print_capacity(struct scsi_disk *sdkp,
2565                   sector_t old_capacity)
2566 {
2567         int sector_size = sdkp->device->sector_size;
2568         char cap_str_2[10], cap_str_10[10];
2569
2570         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2571                 return;
2572
2573         string_get_size(sdkp->capacity, sector_size,
2574                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2575         string_get_size(sdkp->capacity, sector_size,
2576                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2577
2578         sd_printk(KERN_NOTICE, sdkp,
2579                   "%llu %d-byte logical blocks: (%s/%s)\n",
2580                   (unsigned long long)sdkp->capacity,
2581                   sector_size, cap_str_10, cap_str_2);
2582
2583         if (sdkp->physical_block_size != sector_size)
2584                 sd_printk(KERN_NOTICE, sdkp,
2585                           "%u-byte physical blocks\n",
2586                           sdkp->physical_block_size);
2587
2588         sd_zbc_print_zones(sdkp);
2589 }
2590
2591 /* called with buffer of length 512 */
2592 static inline int
2593 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2594                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2595                  struct scsi_sense_hdr *sshdr)
2596 {
2597         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2598                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2599                                sshdr);
2600 }
2601
2602 /*
2603  * read write protect setting, if possible - called only in sd_revalidate_disk()
2604  * called with buffer of length SD_BUF_SIZE
2605  */
2606 static void
2607 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2608 {
2609         int res;
2610         struct scsi_device *sdp = sdkp->device;
2611         struct scsi_mode_data data;
2612         int disk_ro = get_disk_ro(sdkp->disk);
2613         int old_wp = sdkp->write_prot;
2614
2615         set_disk_ro(sdkp->disk, 0);
2616         if (sdp->skip_ms_page_3f) {
2617                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2618                 return;
2619         }
2620
2621         if (sdp->use_192_bytes_for_3f) {
2622                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2623         } else {
2624                 /*
2625                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2626                  * We have to start carefully: some devices hang if we ask
2627                  * for more than is available.
2628                  */
2629                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2630
2631                 /*
2632                  * Second attempt: ask for page 0 When only page 0 is
2633                  * implemented, a request for page 3F may return Sense Key
2634                  * 5: Illegal Request, Sense Code 24: Invalid field in
2635                  * CDB.
2636                  */
2637                 if (!scsi_status_is_good(res))
2638                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2639
2640                 /*
2641                  * Third attempt: ask 255 bytes, as we did earlier.
2642                  */
2643                 if (!scsi_status_is_good(res))
2644                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2645                                                &data, NULL);
2646         }
2647
2648         if (!scsi_status_is_good(res)) {
2649                 sd_first_printk(KERN_WARNING, sdkp,
2650                           "Test WP failed, assume Write Enabled\n");
2651         } else {
2652                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2653                 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2654                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2655                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2656                                   sdkp->write_prot ? "on" : "off");
2657                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2658                 }
2659         }
2660 }
2661
2662 /*
2663  * sd_read_cache_type - called only from sd_revalidate_disk()
2664  * called with buffer of length SD_BUF_SIZE
2665  */
2666 static void
2667 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2668 {
2669         int len = 0, res;
2670         struct scsi_device *sdp = sdkp->device;
2671
2672         int dbd;
2673         int modepage;
2674         int first_len;
2675         struct scsi_mode_data data;
2676         struct scsi_sense_hdr sshdr;
2677         int old_wce = sdkp->WCE;
2678         int old_rcd = sdkp->RCD;
2679         int old_dpofua = sdkp->DPOFUA;
2680
2681
2682         if (sdkp->cache_override)
2683                 return;
2684
2685         first_len = 4;
2686         if (sdp->skip_ms_page_8) {
2687                 if (sdp->type == TYPE_RBC)
2688                         goto defaults;
2689                 else {
2690                         if (sdp->skip_ms_page_3f)
2691                                 goto defaults;
2692                         modepage = 0x3F;
2693                         if (sdp->use_192_bytes_for_3f)
2694                                 first_len = 192;
2695                         dbd = 0;
2696                 }
2697         } else if (sdp->type == TYPE_RBC) {
2698                 modepage = 6;
2699                 dbd = 8;
2700         } else {
2701                 modepage = 8;
2702                 dbd = 0;
2703         }
2704
2705         /* cautiously ask */
2706         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2707                         &data, &sshdr);
2708
2709         if (!scsi_status_is_good(res))
2710                 goto bad_sense;
2711
2712         if (!data.header_length) {
2713                 modepage = 6;
2714                 first_len = 0;
2715                 sd_first_printk(KERN_ERR, sdkp,
2716                                 "Missing header in MODE_SENSE response\n");
2717         }
2718
2719         /* that went OK, now ask for the proper length */
2720         len = data.length;
2721
2722         /*
2723          * We're only interested in the first three bytes, actually.
2724          * But the data cache page is defined for the first 20.
2725          */
2726         if (len < 3)
2727                 goto bad_sense;
2728         else if (len > SD_BUF_SIZE) {
2729                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2730                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2731                 len = SD_BUF_SIZE;
2732         }
2733         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2734                 len = 192;
2735
2736         /* Get the data */
2737         if (len > first_len)
2738                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2739                                 &data, &sshdr);
2740
2741         if (scsi_status_is_good(res)) {
2742                 int offset = data.header_length + data.block_descriptor_length;
2743
2744                 while (offset < len) {
2745                         u8 page_code = buffer[offset] & 0x3F;
2746                         u8 spf       = buffer[offset] & 0x40;
2747
2748                         if (page_code == 8 || page_code == 6) {
2749                                 /* We're interested only in the first 3 bytes.
2750                                  */
2751                                 if (len - offset <= 2) {
2752                                         sd_first_printk(KERN_ERR, sdkp,
2753                                                 "Incomplete mode parameter "
2754                                                         "data\n");
2755                                         goto defaults;
2756                                 } else {
2757                                         modepage = page_code;
2758                                         goto Page_found;
2759                                 }
2760                         } else {
2761                                 /* Go to the next page */
2762                                 if (spf && len - offset > 3)
2763                                         offset += 4 + (buffer[offset+2] << 8) +
2764                                                 buffer[offset+3];
2765                                 else if (!spf && len - offset > 1)
2766                                         offset += 2 + buffer[offset+1];
2767                                 else {
2768                                         sd_first_printk(KERN_ERR, sdkp,
2769                                                         "Incomplete mode "
2770                                                         "parameter data\n");
2771                                         goto defaults;
2772                                 }
2773                         }
2774                 }
2775
2776                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2777                 goto defaults;
2778
2779         Page_found:
2780                 if (modepage == 8) {
2781                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2782                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2783                 } else {
2784                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2785                         sdkp->RCD = 0;
2786                 }
2787
2788                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2789                 if (sdp->broken_fua) {
2790                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2791                         sdkp->DPOFUA = 0;
2792                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2793                            !sdkp->device->use_16_for_rw) {
2794                         sd_first_printk(KERN_NOTICE, sdkp,
2795                                   "Uses READ/WRITE(6), disabling FUA\n");
2796                         sdkp->DPOFUA = 0;
2797                 }
2798
2799                 /* No cache flush allowed for write protected devices */
2800                 if (sdkp->WCE && sdkp->write_prot)
2801                         sdkp->WCE = 0;
2802
2803                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2804                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2805                         sd_printk(KERN_NOTICE, sdkp,
2806                                   "Write cache: %s, read cache: %s, %s\n",
2807                                   sdkp->WCE ? "enabled" : "disabled",
2808                                   sdkp->RCD ? "disabled" : "enabled",
2809                                   sdkp->DPOFUA ? "supports DPO and FUA"
2810                                   : "doesn't support DPO or FUA");
2811
2812                 return;
2813         }
2814
2815 bad_sense:
2816         if (scsi_sense_valid(&sshdr) &&
2817             sshdr.sense_key == ILLEGAL_REQUEST &&
2818             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2819                 /* Invalid field in CDB */
2820                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2821         else
2822                 sd_first_printk(KERN_ERR, sdkp,
2823                                 "Asking for cache data failed\n");
2824
2825 defaults:
2826         if (sdp->wce_default_on) {
2827                 sd_first_printk(KERN_NOTICE, sdkp,
2828                                 "Assuming drive cache: write back\n");
2829                 sdkp->WCE = 1;
2830         } else {
2831                 sd_first_printk(KERN_ERR, sdkp,
2832                                 "Assuming drive cache: write through\n");
2833                 sdkp->WCE = 0;
2834         }
2835         sdkp->RCD = 0;
2836         sdkp->DPOFUA = 0;
2837 }
2838
2839 /*
2840  * The ATO bit indicates whether the DIF application tag is available
2841  * for use by the operating system.
2842  */
2843 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2844 {
2845         int res, offset;
2846         struct scsi_device *sdp = sdkp->device;
2847         struct scsi_mode_data data;
2848         struct scsi_sense_hdr sshdr;
2849
2850         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2851                 return;
2852
2853         if (sdkp->protection_type == 0)
2854                 return;
2855
2856         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2857                               SD_MAX_RETRIES, &data, &sshdr);
2858
2859         if (!scsi_status_is_good(res) || !data.header_length ||
2860             data.length < 6) {
2861                 sd_first_printk(KERN_WARNING, sdkp,
2862                           "getting Control mode page failed, assume no ATO\n");
2863
2864                 if (scsi_sense_valid(&sshdr))
2865                         sd_print_sense_hdr(sdkp, &sshdr);
2866
2867                 return;
2868         }
2869
2870         offset = data.header_length + data.block_descriptor_length;
2871
2872         if ((buffer[offset] & 0x3f) != 0x0a) {
2873                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2874                 return;
2875         }
2876
2877         if ((buffer[offset + 5] & 0x80) == 0)
2878                 return;
2879
2880         sdkp->ATO = 1;
2881
2882         return;
2883 }
2884
2885 /**
2886  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2887  * @sdkp: disk to query
2888  */
2889 static void sd_read_block_limits(struct scsi_disk *sdkp)
2890 {
2891         unsigned int sector_sz = sdkp->device->sector_size;
2892         const int vpd_len = 64;
2893         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2894
2895         if (!buffer ||
2896             /* Block Limits VPD */
2897             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2898                 goto out;
2899
2900         blk_queue_io_min(sdkp->disk->queue,
2901                          get_unaligned_be16(&buffer[6]) * sector_sz);
2902
2903         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2904         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2905
2906         if (buffer[3] == 0x3c) {
2907                 unsigned int lba_count, desc_count;
2908
2909                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2910
2911                 if (!sdkp->lbpme)
2912                         goto out;
2913
2914                 lba_count = get_unaligned_be32(&buffer[20]);
2915                 desc_count = get_unaligned_be32(&buffer[24]);
2916
2917                 if (lba_count && desc_count)
2918                         sdkp->max_unmap_blocks = lba_count;
2919
2920                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2921
2922                 if (buffer[32] & 0x80)
2923                         sdkp->unmap_alignment =
2924                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2925
2926                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2927
2928                         if (sdkp->max_unmap_blocks)
2929                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2930                         else
2931                                 sd_config_discard(sdkp, SD_LBP_WS16);
2932
2933                 } else {        /* LBP VPD page tells us what to use */
2934                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2935                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2936                         else if (sdkp->lbpws)
2937                                 sd_config_discard(sdkp, SD_LBP_WS16);
2938                         else if (sdkp->lbpws10)
2939                                 sd_config_discard(sdkp, SD_LBP_WS10);
2940                         else
2941                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2942                 }
2943         }
2944
2945  out:
2946         kfree(buffer);
2947 }
2948
2949 /**
2950  * sd_read_block_characteristics - Query block dev. characteristics
2951  * @sdkp: disk to query
2952  */
2953 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2954 {
2955         struct request_queue *q = sdkp->disk->queue;
2956         unsigned char *buffer;
2957         u16 rot;
2958         const int vpd_len = 64;
2959
2960         buffer = kmalloc(vpd_len, GFP_KERNEL);
2961
2962         if (!buffer ||
2963             /* Block Device Characteristics VPD */
2964             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2965                 goto out;
2966
2967         rot = get_unaligned_be16(&buffer[4]);
2968
2969         if (rot == 1) {
2970                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2971                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2972         }
2973
2974         if (sdkp->device->type == TYPE_ZBC) {
2975                 /* Host-managed */
2976                 q->limits.zoned = BLK_ZONED_HM;
2977         } else {
2978                 sdkp->zoned = (buffer[8] >> 4) & 3;
2979                 if (sdkp->zoned == 1)
2980                         /* Host-aware */
2981                         q->limits.zoned = BLK_ZONED_HA;
2982                 else
2983                         /*
2984                          * Treat drive-managed devices as
2985                          * regular block devices.
2986                          */
2987                         q->limits.zoned = BLK_ZONED_NONE;
2988         }
2989         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2990                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2991                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2992
2993  out:
2994         kfree(buffer);
2995 }
2996
2997 /**
2998  * sd_read_block_provisioning - Query provisioning VPD page
2999  * @sdkp: disk to query
3000  */
3001 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3002 {
3003         unsigned char *buffer;
3004         const int vpd_len = 8;
3005
3006         if (sdkp->lbpme == 0)
3007                 return;
3008
3009         buffer = kmalloc(vpd_len, GFP_KERNEL);
3010
3011         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3012                 goto out;
3013
3014         sdkp->lbpvpd    = 1;
3015         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3016         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3017         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3018
3019  out:
3020         kfree(buffer);
3021 }
3022
3023 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3024 {
3025         struct scsi_device *sdev = sdkp->device;
3026
3027         if (sdev->host->no_write_same) {
3028                 sdev->no_write_same = 1;
3029
3030                 return;
3031         }
3032
3033         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3034                 /* too large values might cause issues with arcmsr */
3035                 int vpd_buf_len = 64;
3036
3037                 sdev->no_report_opcodes = 1;
3038
3039                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3040                  * CODES is unsupported and the device has an ATA
3041                  * Information VPD page (SAT).
3042                  */
3043                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3044                         sdev->no_write_same = 1;
3045         }
3046
3047         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3048                 sdkp->ws16 = 1;
3049
3050         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3051                 sdkp->ws10 = 1;
3052 }
3053
3054 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3055 {
3056         struct scsi_device *sdev = sdkp->device;
3057
3058         if (!sdev->security_supported)
3059                 return;
3060
3061         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3062                         SECURITY_PROTOCOL_IN) == 1 &&
3063             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3064                         SECURITY_PROTOCOL_OUT) == 1)
3065                 sdkp->security = 1;
3066 }
3067
3068 /*
3069  * Determine the device's preferred I/O size for reads and writes
3070  * unless the reported value is unreasonably small, large, not a
3071  * multiple of the physical block size, or simply garbage.
3072  */
3073 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3074                                       unsigned int dev_max)
3075 {
3076         struct scsi_device *sdp = sdkp->device;
3077         unsigned int opt_xfer_bytes =
3078                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
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                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3200                                       (sector_t)BLK_DEF_MAX_SECTORS);
3201
3202         /* Do not exceed controller limit */
3203         rw_max = min(rw_max, queue_max_hw_sectors(q));
3204
3205         /*
3206          * Only update max_sectors if previously unset or if the current value
3207          * exceeds the capabilities of the hardware.
3208          */
3209         if (sdkp->first_scan ||
3210             q->limits.max_sectors > q->limits.max_dev_sectors ||
3211             q->limits.max_sectors > q->limits.max_hw_sectors)
3212                 q->limits.max_sectors = rw_max;
3213
3214         sdkp->first_scan = 0;
3215
3216         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3217         sd_config_write_same(sdkp);
3218         kfree(buffer);
3219
3220  out:
3221         return 0;
3222 }
3223
3224 /**
3225  *      sd_unlock_native_capacity - unlock native capacity
3226  *      @disk: struct gendisk to set capacity for
3227  *
3228  *      Block layer calls this function if it detects that partitions
3229  *      on @disk reach beyond the end of the device.  If the SCSI host
3230  *      implements ->unlock_native_capacity() method, it's invoked to
3231  *      give it a chance to adjust the device capacity.
3232  *
3233  *      CONTEXT:
3234  *      Defined by block layer.  Might sleep.
3235  */
3236 static void sd_unlock_native_capacity(struct gendisk *disk)
3237 {
3238         struct scsi_device *sdev = scsi_disk(disk)->device;
3239
3240         if (sdev->host->hostt->unlock_native_capacity)
3241                 sdev->host->hostt->unlock_native_capacity(sdev);
3242 }
3243
3244 /**
3245  *      sd_format_disk_name - format disk name
3246  *      @prefix: name prefix - ie. "sd" for SCSI disks
3247  *      @index: index of the disk to format name for
3248  *      @buf: output buffer
3249  *      @buflen: length of the output buffer
3250  *
3251  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3252  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3253  *      which is followed by sdaaa.
3254  *
3255  *      This is basically 26 base counting with one extra 'nil' entry
3256  *      at the beginning from the second digit on and can be
3257  *      determined using similar method as 26 base conversion with the
3258  *      index shifted -1 after each digit is computed.
3259  *
3260  *      CONTEXT:
3261  *      Don't care.
3262  *
3263  *      RETURNS:
3264  *      0 on success, -errno on failure.
3265  */
3266 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3267 {
3268         const int base = 'z' - 'a' + 1;
3269         char *begin = buf + strlen(prefix);
3270         char *end = buf + buflen;
3271         char *p;
3272         int unit;
3273
3274         p = end - 1;
3275         *p = '\0';
3276         unit = base;
3277         do {
3278                 if (p == begin)
3279                         return -EINVAL;
3280                 *--p = 'a' + (index % unit);
3281                 index = (index / unit) - 1;
3282         } while (index >= 0);
3283
3284         memmove(begin, p, end - p);
3285         memcpy(buf, prefix, strlen(prefix));
3286
3287         return 0;
3288 }
3289
3290 /**
3291  *      sd_probe - called during driver initialization and whenever a
3292  *      new scsi device is attached to the system. It is called once
3293  *      for each scsi device (not just disks) present.
3294  *      @dev: pointer to device object
3295  *
3296  *      Returns 0 if successful (or not interested in this scsi device
3297  *      (e.g. scanner)); 1 when there is an error.
3298  *
3299  *      Note: this function is invoked from the scsi mid-level.
3300  *      This function sets up the mapping between a given
3301  *      <host,channel,id,lun> (found in sdp) and new device name
3302  *      (e.g. /dev/sda). More precisely it is the block device major
3303  *      and minor number that is chosen here.
3304  *
3305  *      Assume sd_probe is not re-entrant (for time being)
3306  *      Also think about sd_probe() and sd_remove() running coincidentally.
3307  **/
3308 static int sd_probe(struct device *dev)
3309 {
3310         struct scsi_device *sdp = to_scsi_device(dev);
3311         struct scsi_disk *sdkp;
3312         struct gendisk *gd;
3313         int index;
3314         int error;
3315
3316         scsi_autopm_get_device(sdp);
3317         error = -ENODEV;
3318         if (sdp->type != TYPE_DISK &&
3319             sdp->type != TYPE_ZBC &&
3320             sdp->type != TYPE_MOD &&
3321             sdp->type != TYPE_RBC)
3322                 goto out;
3323
3324 #ifndef CONFIG_BLK_DEV_ZONED
3325         if (sdp->type == TYPE_ZBC)
3326                 goto out;
3327 #endif
3328         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3329                                         "sd_probe\n"));
3330
3331         error = -ENOMEM;
3332         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3333         if (!sdkp)
3334                 goto out;
3335
3336         gd = alloc_disk(SD_MINORS);
3337         if (!gd)
3338                 goto out_free;
3339
3340         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3341         if (index < 0) {
3342                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3343                 goto out_put;
3344         }
3345
3346         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3347         if (error) {
3348                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3349                 goto out_free_index;
3350         }
3351
3352         sdkp->device = sdp;
3353         sdkp->driver = &sd_template;
3354         sdkp->disk = gd;
3355         sdkp->index = index;
3356         atomic_set(&sdkp->openers, 0);
3357         atomic_set(&sdkp->device->ioerr_cnt, 0);
3358
3359         if (!sdp->request_queue->rq_timeout) {
3360                 if (sdp->type != TYPE_MOD)
3361                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3362                 else
3363                         blk_queue_rq_timeout(sdp->request_queue,
3364                                              SD_MOD_TIMEOUT);
3365         }
3366
3367         device_initialize(&sdkp->dev);
3368         sdkp->dev.parent = dev;
3369         sdkp->dev.class = &sd_disk_class;
3370         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3371
3372         error = device_add(&sdkp->dev);
3373         if (error)
3374                 goto out_free_index;
3375
3376         get_device(dev);
3377         dev_set_drvdata(dev, sdkp);
3378
3379         gd->major = sd_major((index & 0xf0) >> 4);
3380         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3381
3382         gd->fops = &sd_fops;
3383         gd->private_data = &sdkp->driver;
3384         gd->queue = sdkp->device->request_queue;
3385
3386         /* defaults, until the device tells us otherwise */
3387         sdp->sector_size = 512;
3388         sdkp->capacity = 0;
3389         sdkp->media_present = 1;
3390         sdkp->write_prot = 0;
3391         sdkp->cache_override = 0;
3392         sdkp->WCE = 0;
3393         sdkp->RCD = 0;
3394         sdkp->ATO = 0;
3395         sdkp->first_scan = 1;
3396         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3397
3398         sd_revalidate_disk(gd);
3399
3400         gd->flags = GENHD_FL_EXT_DEVT;
3401         if (sdp->removable) {
3402                 gd->flags |= GENHD_FL_REMOVABLE;
3403                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3404         }
3405
3406         blk_pm_runtime_init(sdp->request_queue, dev);
3407         device_add_disk(dev, gd, NULL);
3408         if (sdkp->capacity)
3409                 sd_dif_config_host(sdkp);
3410
3411         sd_revalidate_disk(gd);
3412
3413         if (sdkp->security) {
3414                 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3415                 if (sdkp->opal_dev)
3416                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3417         }
3418
3419         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3420                   sdp->removable ? "removable " : "");
3421         scsi_autopm_put_device(sdp);
3422
3423         return 0;
3424
3425  out_free_index:
3426         ida_free(&sd_index_ida, index);
3427  out_put:
3428         put_disk(gd);
3429  out_free:
3430         kfree(sdkp);
3431  out:
3432         scsi_autopm_put_device(sdp);
3433         return error;
3434 }
3435
3436 /**
3437  *      sd_remove - called whenever a scsi disk (previously recognized by
3438  *      sd_probe) is detached from the system. It is called (potentially
3439  *      multiple times) during sd module unload.
3440  *      @dev: pointer to device object
3441  *
3442  *      Note: this function is invoked from the scsi mid-level.
3443  *      This function potentially frees up a device name (e.g. /dev/sdc)
3444  *      that could be re-used by a subsequent sd_probe().
3445  *      This function is not called when the built-in sd driver is "exit-ed".
3446  **/
3447 static int sd_remove(struct device *dev)
3448 {
3449         struct scsi_disk *sdkp;
3450         dev_t devt;
3451
3452         sdkp = dev_get_drvdata(dev);
3453         devt = disk_devt(sdkp->disk);
3454         scsi_autopm_get_device(sdkp->device);
3455
3456         device_del(&sdkp->dev);
3457         del_gendisk(sdkp->disk);
3458         sd_shutdown(dev);
3459
3460         free_opal_dev(sdkp->opal_dev);
3461
3462         blk_register_region(devt, SD_MINORS, NULL,
3463                             sd_default_probe, NULL, NULL);
3464
3465         mutex_lock(&sd_ref_mutex);
3466         dev_set_drvdata(dev, NULL);
3467         put_device(&sdkp->dev);
3468         mutex_unlock(&sd_ref_mutex);
3469
3470         return 0;
3471 }
3472
3473 /**
3474  *      scsi_disk_release - Called to free the scsi_disk structure
3475  *      @dev: pointer to embedded class device
3476  *
3477  *      sd_ref_mutex must be held entering this routine.  Because it is
3478  *      called on last put, you should always use the scsi_disk_get()
3479  *      scsi_disk_put() helpers which manipulate the semaphore directly
3480  *      and never do a direct put_device.
3481  **/
3482 static void scsi_disk_release(struct device *dev)
3483 {
3484         struct scsi_disk *sdkp = to_scsi_disk(dev);
3485         struct gendisk *disk = sdkp->disk;
3486
3487         ida_free(&sd_index_ida, sdkp->index);
3488
3489         disk->private_data = NULL;
3490         put_disk(disk);
3491         put_device(&sdkp->device->sdev_gendev);
3492
3493         kfree(sdkp);
3494 }
3495
3496 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3497 {
3498         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3499         struct scsi_sense_hdr sshdr;
3500         struct scsi_device *sdp = sdkp->device;
3501         int res;
3502
3503         if (start)
3504                 cmd[4] |= 1;    /* START */
3505
3506         if (sdp->start_stop_pwr_cond)
3507                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3508
3509         if (!scsi_device_online(sdp))
3510                 return -ENODEV;
3511
3512         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3513                         SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3514         if (res) {
3515                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3516                 if (driver_byte(res) == DRIVER_SENSE)
3517                         sd_print_sense_hdr(sdkp, &sshdr);
3518                 if (scsi_sense_valid(&sshdr) &&
3519                         /* 0x3a is medium not present */
3520                         sshdr.asc == 0x3a)
3521                         res = 0;
3522         }
3523
3524         /* SCSI error codes must not go to the generic layer */
3525         if (res)
3526                 return -EIO;
3527
3528         return 0;
3529 }
3530
3531 /*
3532  * Send a SYNCHRONIZE CACHE instruction down to the device through
3533  * the normal SCSI command structure.  Wait for the command to
3534  * complete.
3535  */
3536 static void sd_shutdown(struct device *dev)
3537 {
3538         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3539
3540         if (!sdkp)
3541                 return;         /* this can happen */
3542
3543         if (pm_runtime_suspended(dev))
3544                 return;
3545
3546         if (sdkp->WCE && sdkp->media_present) {
3547                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3548                 sd_sync_cache(sdkp, NULL);
3549         }
3550
3551         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3552                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3553                 sd_start_stop_device(sdkp, 0);
3554         }
3555 }
3556
3557 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3558 {
3559         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3560         struct scsi_sense_hdr sshdr;
3561         int ret = 0;
3562
3563         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3564                 return 0;
3565
3566         if (sdkp->WCE && sdkp->media_present) {
3567                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3568                 ret = sd_sync_cache(sdkp, &sshdr);
3569
3570                 if (ret) {
3571                         /* ignore OFFLINE device */
3572                         if (ret == -ENODEV)
3573                                 return 0;
3574
3575                         if (!scsi_sense_valid(&sshdr) ||
3576                             sshdr.sense_key != ILLEGAL_REQUEST)
3577                                 return ret;
3578
3579                         /*
3580                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3581                          * doesn't support sync. There's not much to do and
3582                          * suspend shouldn't fail.
3583                          */
3584                         ret = 0;
3585                 }
3586         }
3587
3588         if (sdkp->device->manage_start_stop) {
3589                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3590                 /* an error is not worth aborting a system sleep */
3591                 ret = sd_start_stop_device(sdkp, 0);
3592                 if (ignore_stop_errors)
3593                         ret = 0;
3594         }
3595
3596         return ret;
3597 }
3598
3599 static int sd_suspend_system(struct device *dev)
3600 {
3601         return sd_suspend_common(dev, true);
3602 }
3603
3604 static int sd_suspend_runtime(struct device *dev)
3605 {
3606         return sd_suspend_common(dev, false);
3607 }
3608
3609 static int sd_resume(struct device *dev)
3610 {
3611         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3612         int ret;
3613
3614         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3615                 return 0;
3616
3617         if (!sdkp->device->manage_start_stop)
3618                 return 0;
3619
3620         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3621         ret = sd_start_stop_device(sdkp, 1);
3622         if (!ret)
3623                 opal_unlock_from_suspend(sdkp->opal_dev);
3624         return ret;
3625 }
3626
3627 /**
3628  *      init_sd - entry point for this driver (both when built in or when
3629  *      a module).
3630  *
3631  *      Note: this function registers this driver with the scsi mid-level.
3632  **/
3633 static int __init init_sd(void)
3634 {
3635         int majors = 0, i, err;
3636
3637         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3638
3639         for (i = 0; i < SD_MAJORS; i++) {
3640                 if (register_blkdev(sd_major(i), "sd") != 0)
3641                         continue;
3642                 majors++;
3643                 blk_register_region(sd_major(i), SD_MINORS, NULL,
3644                                     sd_default_probe, NULL, NULL);
3645         }
3646
3647         if (!majors)
3648                 return -ENODEV;
3649
3650         err = class_register(&sd_disk_class);
3651         if (err)
3652                 goto err_out;
3653
3654         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3655                                          0, 0, NULL);
3656         if (!sd_cdb_cache) {
3657                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3658                 err = -ENOMEM;
3659                 goto err_out_class;
3660         }
3661
3662         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3663         if (!sd_cdb_pool) {
3664                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3665                 err = -ENOMEM;
3666                 goto err_out_cache;
3667         }
3668
3669         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3670         if (!sd_page_pool) {
3671                 printk(KERN_ERR "sd: can't init discard page pool\n");
3672                 err = -ENOMEM;
3673                 goto err_out_ppool;
3674         }
3675
3676         err = scsi_register_driver(&sd_template.gendrv);
3677         if (err)
3678                 goto err_out_driver;
3679
3680         return 0;
3681
3682 err_out_driver:
3683         mempool_destroy(sd_page_pool);
3684
3685 err_out_ppool:
3686         mempool_destroy(sd_cdb_pool);
3687
3688 err_out_cache:
3689         kmem_cache_destroy(sd_cdb_cache);
3690
3691 err_out_class:
3692         class_unregister(&sd_disk_class);
3693 err_out:
3694         for (i = 0; i < SD_MAJORS; i++)
3695                 unregister_blkdev(sd_major(i), "sd");
3696         return err;
3697 }
3698
3699 /**
3700  *      exit_sd - exit point for this driver (when it is a module).
3701  *
3702  *      Note: this function unregisters this driver from the scsi mid-level.
3703  **/
3704 static void __exit exit_sd(void)
3705 {
3706         int i;
3707
3708         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3709
3710         scsi_unregister_driver(&sd_template.gendrv);
3711         mempool_destroy(sd_cdb_pool);
3712         mempool_destroy(sd_page_pool);
3713         kmem_cache_destroy(sd_cdb_cache);
3714
3715         class_unregister(&sd_disk_class);
3716
3717         for (i = 0; i < SD_MAJORS; i++) {
3718                 blk_unregister_region(sd_major(i), SD_MINORS);
3719                 unregister_blkdev(sd_major(i), "sd");
3720         }
3721 }
3722
3723 module_init(init_sd);
3724 module_exit(exit_sd);
3725
3726 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3727                                struct scsi_sense_hdr *sshdr)
3728 {
3729         scsi_print_sense_hdr(sdkp->device,
3730                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3731 }
3732
3733 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3734                             int result)
3735 {
3736         const char *hb_string = scsi_hostbyte_string(result);
3737         const char *db_string = scsi_driverbyte_string(result);
3738
3739         if (hb_string || db_string)
3740                 sd_printk(KERN_INFO, sdkp,
3741                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3742                           hb_string ? hb_string : "invalid",
3743                           db_string ? db_string : "invalid");
3744         else
3745                 sd_printk(KERN_INFO, sdkp,
3746                           "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3747                           msg, host_byte(result), driver_byte(result));
3748 }
3749