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                 .remove         = sd_remove,
 571                 .shutdown       = sd_shutdown,
 572                 .pm             = &sd_pm_ops,
 573         },
 574         .rescan                 = sd_rescan,
 575         .init_command           = sd_init_command,
 576         .uninit_command         = sd_uninit_command,
 577         .done                   = sd_done,
 578         .eh_action              = sd_eh_action,
 579         .eh_reset               = sd_eh_reset,
 580 };
 581
 582 /*
 583  * Dummy kobj_map->probe function.
 584  * The default ->probe function will call modprobe, which is
 585  * pointless as this module is already loaded.
 586  */
 587 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
 588 {
 589         return NULL;
 590 }
 591
 592 /*
 593  * Device no to disk mapping:
 594  *
 595  *       major         disc2     disc  p1
 596  *   |............|.............|....|....| <- dev_t
 597  *    31        20 19          8 7  4 3  0
 598  *
 599  * Inside a major, we have 16k disks, however mapped non-
 600  * contiguously. The first 16 disks are for major0, the next
 601  * ones with major1, ... Disk 256 is for major0 again, disk 272
 602  * for major1, ...
 603  * As we stay compatible with our numbering scheme, we can reuse
 604  * the well-know SCSI majors 8, 65--71, 136--143.
 605  */
 606 static int sd_major(int major_idx)
 607 {
 608         switch (major_idx) {
 609         case 0:
 610                 return SCSI_DISK0_MAJOR;
 611         case 1 ... 7:
 612                 return SCSI_DISK1_MAJOR + major_idx - 1;
 613         case 8 ... 15:
 614                 return SCSI_DISK8_MAJOR + major_idx - 8;
 615         default:
 616                 BUG();
 617                 return 0;       /* shut up gcc */
 618         }
 619 }
 620
 621 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 622 {
 623         struct scsi_disk *sdkp = NULL;
 624
 625         mutex_lock(&sd_ref_mutex);
 626
 627         if (disk->private_data) {
 628                 sdkp = scsi_disk(disk);
 629                 if (scsi_device_get(sdkp->device) == 0)
 630                         get_device(&sdkp->dev);
 631                 else
 632                         sdkp = NULL;
 633         }
 634         mutex_unlock(&sd_ref_mutex);
 635         return sdkp;
 636 }
 637
 638 static void scsi_disk_put(struct scsi_disk *sdkp)
 639 {
 640         struct scsi_device *sdev = sdkp->device;
 641
 642         mutex_lock(&sd_ref_mutex);
 643         put_device(&sdkp->dev);
 644         scsi_device_put(sdev);
 645         mutex_unlock(&sd_ref_mutex);
 646 }
 647
 648 #ifdef CONFIG_BLK_SED_OPAL
 649 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 650                 size_t len, bool send)
 651 {
 652         struct scsi_device *sdev = data;
 653         u8 cdb[12] = { 0, };
 654         int ret;
 655
 656         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 657         cdb[1] = secp;
 658         put_unaligned_be16(spsp, &cdb[2]);
 659         put_unaligned_be32(len, &cdb[6]);
 660
 661         ret = scsi_execute_req(sdev, cdb,
 662                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 663                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
 664         return ret <= 0 ? ret : -EIO;
 665 }
 666 #endif /* CONFIG_BLK_SED_OPAL */
 667
 668 /*
 669  * Look up the DIX operation based on whether the command is read or
 670  * write and whether dix and dif are enabled.
 671  */
 672 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 673 {
 674         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 675         static const unsigned int ops[] = {     /* wrt dix dif */
 676                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 677                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 678                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 679                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 680                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 681                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 682                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 683                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 684         };
 685
 686         return ops[write << 2 | dix << 1 | dif];
 687 }
 688
 689 /*
 690  * Returns a mask of the protection flags that are valid for a given DIX
 691  * operation.
 692  */
 693 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 694 {
 695         static const unsigned int flag_mask[] = {
 696                 [SCSI_PROT_NORMAL]              = 0,
 697
 698                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 699                                                   SCSI_PROT_GUARD_CHECK |
 700                                                   SCSI_PROT_REF_CHECK |
 701                                                   SCSI_PROT_REF_INCREMENT,
 702
 703                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 704                                                   SCSI_PROT_IP_CHECKSUM,
 705
 706                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 707                                                   SCSI_PROT_GUARD_CHECK |
 708                                                   SCSI_PROT_REF_CHECK |
 709                                                   SCSI_PROT_REF_INCREMENT |
 710                                                   SCSI_PROT_IP_CHECKSUM,
 711
 712                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 713                                                   SCSI_PROT_REF_INCREMENT,
 714
 715                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 716                                                   SCSI_PROT_REF_CHECK |
 717                                                   SCSI_PROT_REF_INCREMENT |
 718                                                   SCSI_PROT_IP_CHECKSUM,
 719
 720                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 721                                                   SCSI_PROT_GUARD_CHECK |
 722                                                   SCSI_PROT_REF_CHECK |
 723                                                   SCSI_PROT_REF_INCREMENT |
 724                                                   SCSI_PROT_IP_CHECKSUM,
 725         };
 726
 727         return flag_mask[prot_op];
 728 }
 729
 730 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 731                                            unsigned int dix, unsigned int dif)
 732 {
 733         struct bio *bio = scmd->request->bio;
 734         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
 735         unsigned int protect = 0;
 736
 737         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 738                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 739                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 740
 741                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 742                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 743         }
 744
 745         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 746                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 747
 748                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 749                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 750         }
 751
 752         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 753                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 754
 755                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 756                         protect = 3 << 5;       /* Disable target PI checking */
 757                 else
 758                         protect = 1 << 5;       /* Enable target PI checking */
 759         }
 760
 761         scsi_set_prot_op(scmd, prot_op);
 762         scsi_set_prot_type(scmd, dif);
 763         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 764
 765         return protect;
 766 }
 767
 768 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 769 {
 770         struct request_queue *q = sdkp->disk->queue;
 771         unsigned int logical_block_size = sdkp->device->sector_size;
 772         unsigned int max_blocks = 0;
 773
 774         q->limits.discard_alignment =
 775                 sdkp->unmap_alignment * logical_block_size;
 776         q->limits.discard_granularity =
 777                 max(sdkp->physical_block_size,
 778                     sdkp->unmap_granularity * logical_block_size);
 779         sdkp->provisioning_mode = mode;
 780
 781         switch (mode) {
 782
 783         case SD_LBP_FULL:
 784         case SD_LBP_DISABLE:
 785                 blk_queue_max_discard_sectors(q, 0);
 786                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 787                 return;
 788
 789         case SD_LBP_UNMAP:
 790                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 791                                           (u32)SD_MAX_WS16_BLOCKS);
 792                 break;
 793
 794         case SD_LBP_WS16:
 795                 if (sdkp->device->unmap_limit_for_ws)
 796                         max_blocks = sdkp->max_unmap_blocks;
 797                 else
 798                         max_blocks = sdkp->max_ws_blocks;
 799
 800                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 801                 break;
 802
 803         case SD_LBP_WS10:
 804                 if (sdkp->device->unmap_limit_for_ws)
 805                         max_blocks = sdkp->max_unmap_blocks;
 806                 else
 807                         max_blocks = sdkp->max_ws_blocks;
 808
 809                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 810                 break;
 811
 812         case SD_LBP_ZERO:
 813                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 814                                           (u32)SD_MAX_WS10_BLOCKS);
 815                 break;
 816         }
 817
 818         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 819         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 820 }
 821
 822 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 823 {
 824         struct scsi_device *sdp = cmd->device;
 825         struct request *rq = cmd->request;
 826         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 827         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 828         unsigned int data_len = 24;
 829         char *buf;
 830
 831         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 832         if (!rq->special_vec.bv_page)
 833                 return BLK_STS_RESOURCE;
 834         clear_highpage(rq->special_vec.bv_page);
 835         rq->special_vec.bv_offset = 0;
 836         rq->special_vec.bv_len = data_len;
 837         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 838
 839         cmd->cmd_len = 10;
 840         cmd->cmnd[0] = UNMAP;
 841         cmd->cmnd[8] = 24;
 842
 843         buf = page_address(rq->special_vec.bv_page);
 844         put_unaligned_be16(6 + 16, &buf[0]);
 845         put_unaligned_be16(16, &buf[2]);
 846         put_unaligned_be64(lba, &buf[8]);
 847         put_unaligned_be32(nr_blocks, &buf[16]);
 848
 849         cmd->allowed = SD_MAX_RETRIES;
 850         cmd->transfersize = data_len;
 851         rq->timeout = SD_TIMEOUT;
 852
 853         return scsi_init_io(cmd);
 854 }
 855
 856 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 857                 bool unmap)
 858 {
 859         struct scsi_device *sdp = cmd->device;
 860         struct request *rq = cmd->request;
 861         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 862         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 863         u32 data_len = sdp->sector_size;
 864
 865         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 866         if (!rq->special_vec.bv_page)
 867                 return BLK_STS_RESOURCE;
 868         clear_highpage(rq->special_vec.bv_page);
 869         rq->special_vec.bv_offset = 0;
 870         rq->special_vec.bv_len = data_len;
 871         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 872
 873         cmd->cmd_len = 16;
 874         cmd->cmnd[0] = WRITE_SAME_16;
 875         if (unmap)
 876                 cmd->cmnd[1] = 0x8; /* UNMAP */
 877         put_unaligned_be64(lba, &cmd->cmnd[2]);
 878         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 879
 880         cmd->allowed = SD_MAX_RETRIES;
 881         cmd->transfersize = data_len;
 882         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 883
 884         return scsi_init_io(cmd);
 885 }
 886
 887 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 888                 bool unmap)
 889 {
 890         struct scsi_device *sdp = cmd->device;
 891         struct request *rq = cmd->request;
 892         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 893         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 894         u32 data_len = sdp->sector_size;
 895
 896         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 897         if (!rq->special_vec.bv_page)
 898                 return BLK_STS_RESOURCE;
 899         clear_highpage(rq->special_vec.bv_page);
 900         rq->special_vec.bv_offset = 0;
 901         rq->special_vec.bv_len = data_len;
 902         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 903
 904         cmd->cmd_len = 10;
 905         cmd->cmnd[0] = WRITE_SAME;
 906         if (unmap)
 907                 cmd->cmnd[1] = 0x8; /* UNMAP */
 908         put_unaligned_be32(lba, &cmd->cmnd[2]);
 909         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 910
 911         cmd->allowed = SD_MAX_RETRIES;
 912         cmd->transfersize = data_len;
 913         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 914
 915         return scsi_init_io(cmd);
 916 }
 917
 918 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 919 {
 920         struct request *rq = cmd->request;
 921         struct scsi_device *sdp = cmd->device;
 922         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 923         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 924         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 925
 926         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 927                 switch (sdkp->zeroing_mode) {
 928                 case SD_ZERO_WS16_UNMAP:
 929                         return sd_setup_write_same16_cmnd(cmd, true);
 930                 case SD_ZERO_WS10_UNMAP:
 931                         return sd_setup_write_same10_cmnd(cmd, true);
 932                 }
 933         }
 934
 935         if (sdp->no_write_same)
 936                 return BLK_STS_TARGET;
 937
 938         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 939                 return sd_setup_write_same16_cmnd(cmd, false);
 940
 941         return sd_setup_write_same10_cmnd(cmd, false);
 942 }
 943
 944 static void sd_config_write_same(struct scsi_disk *sdkp)
 945 {
 946         struct request_queue *q = sdkp->disk->queue;
 947         unsigned int logical_block_size = sdkp->device->sector_size;
 948
 949         if (sdkp->device->no_write_same) {
 950                 sdkp->max_ws_blocks = 0;
 951                 goto out;
 952         }
 953
 954         /* Some devices can not handle block counts above 0xffff despite
 955          * supporting WRITE SAME(16). Consequently we default to 64k
 956          * blocks per I/O unless the device explicitly advertises a
 957          * bigger limit.
 958          */
 959         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
 960                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 961                                                    (u32)SD_MAX_WS16_BLOCKS);
 962         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
 963                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 964                                                    (u32)SD_MAX_WS10_BLOCKS);
 965         else {
 966                 sdkp->device->no_write_same = 1;
 967                 sdkp->max_ws_blocks = 0;
 968         }
 969
 970         if (sdkp->lbprz && sdkp->lbpws)
 971                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
 972         else if (sdkp->lbprz && sdkp->lbpws10)
 973                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
 974         else if (sdkp->max_ws_blocks)
 975                 sdkp->zeroing_mode = SD_ZERO_WS;
 976         else
 977                 sdkp->zeroing_mode = SD_ZERO_WRITE;
 978
 979         if (sdkp->max_ws_blocks &&
 980             sdkp->physical_block_size > logical_block_size) {
 981                 /*
 982                  * Reporting a maximum number of blocks that is not aligned
 983                  * on the device physical size would cause a large write same
 984                  * request to be split into physically unaligned chunks by
 985                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
 986                  * even if the caller of these functions took care to align the
 987                  * large request. So make sure the maximum reported is aligned
 988                  * to the device physical block size. This is only an optional
 989                  * optimization for regular disks, but this is mandatory to
 990                  * avoid failure of large write same requests directed at
 991                  * sequential write required zones of host-managed ZBC disks.
 992                  */
 993                 sdkp->max_ws_blocks =
 994                         round_down(sdkp->max_ws_blocks,
 995                                    bytes_to_logical(sdkp->device,
 996                                                     sdkp->physical_block_size));
 997         }
 998
 999 out:
1000         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1001                                          (logical_block_size >> 9));
1002         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1003                                          (logical_block_size >> 9));
1004 }
1005
1006 /**
1007  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1008  * @cmd: command to prepare
1009  *
1010  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1011  * the preference indicated by the target device.
1012  **/
1013 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1014 {
1015         struct request *rq = cmd->request;
1016         struct scsi_device *sdp = cmd->device;
1017         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1018         struct bio *bio = rq->bio;
1019         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1020         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1021         blk_status_t ret;
1022
1023         if (sdkp->device->no_write_same)
1024                 return BLK_STS_TARGET;
1025
1026         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1027
1028         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1029
1030         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1031                 cmd->cmd_len = 16;
1032                 cmd->cmnd[0] = WRITE_SAME_16;
1033                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1034                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1035         } else {
1036                 cmd->cmd_len = 10;
1037                 cmd->cmnd[0] = WRITE_SAME;
1038                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1039                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1040         }
1041
1042         cmd->transfersize = sdp->sector_size;
1043         cmd->allowed = SD_MAX_RETRIES;
1044
1045         /*
1046          * For WRITE SAME the data transferred via the DATA OUT buffer is
1047          * different from the amount of data actually written to the target.
1048          *
1049          * We set up __data_len to the amount of data transferred via the
1050          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1051          * to transfer a single sector of data first, but then reset it to
1052          * the amount of data to be written right after so that the I/O path
1053          * knows how much to actually write.
1054          */
1055         rq->__data_len = sdp->sector_size;
1056         ret = scsi_init_io(cmd);
1057         rq->__data_len = blk_rq_bytes(rq);
1058
1059         return ret;
1060 }
1061
1062 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1063 {
1064         struct request *rq = cmd->request;
1065
1066         /* flush requests don't perform I/O, zero the S/G table */
1067         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1068
1069         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1070         cmd->cmd_len = 10;
1071         cmd->transfersize = 0;
1072         cmd->allowed = SD_MAX_RETRIES;
1073
1074         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1075         return BLK_STS_OK;
1076 }
1077
1078 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1079                                        sector_t lba, unsigned int nr_blocks,
1080                                        unsigned char flags)
1081 {
1082         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1083         if (unlikely(cmd->cmnd == NULL))
1084                 return BLK_STS_RESOURCE;
1085
1086         cmd->cmd_len = SD_EXT_CDB_SIZE;
1087         memset(cmd->cmnd, 0, cmd->cmd_len);
1088
1089         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1090         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1091         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1092         cmd->cmnd[10] = flags;
1093         put_unaligned_be64(lba, &cmd->cmnd[12]);
1094         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1095         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1096
1097         return BLK_STS_OK;
1098 }
1099
1100 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1101                                        sector_t lba, unsigned int nr_blocks,
1102                                        unsigned char flags)
1103 {
1104         cmd->cmd_len  = 16;
1105         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1106         cmd->cmnd[1]  = flags;
1107         cmd->cmnd[14] = 0;
1108         cmd->cmnd[15] = 0;
1109         put_unaligned_be64(lba, &cmd->cmnd[2]);
1110         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1111
1112         return BLK_STS_OK;
1113 }
1114
1115 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1116                                        sector_t lba, unsigned int nr_blocks,
1117                                        unsigned char flags)
1118 {
1119         cmd->cmd_len = 10;
1120         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1121         cmd->cmnd[1] = flags;
1122         cmd->cmnd[6] = 0;
1123         cmd->cmnd[9] = 0;
1124         put_unaligned_be32(lba, &cmd->cmnd[2]);
1125         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1126
1127         return BLK_STS_OK;
1128 }
1129
1130 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1131                                       sector_t lba, unsigned int nr_blocks,
1132                                       unsigned char flags)
1133 {
1134         /* Avoid that 0 blocks gets translated into 256 blocks. */
1135         if (WARN_ON_ONCE(nr_blocks == 0))
1136                 return BLK_STS_IOERR;
1137
1138         if (unlikely(flags & 0x8)) {
1139                 /*
1140                  * This happens only if this drive failed 10byte rw
1141                  * command with ILLEGAL_REQUEST during operation and
1142                  * thus turned off use_10_for_rw.
1143                  */
1144                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1145                 return BLK_STS_IOERR;
1146         }
1147
1148         cmd->cmd_len = 6;
1149         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1150         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1151         cmd->cmnd[2] = (lba >> 8) & 0xff;
1152         cmd->cmnd[3] = lba & 0xff;
1153         cmd->cmnd[4] = nr_blocks;
1154         cmd->cmnd[5] = 0;
1155
1156         return BLK_STS_OK;
1157 }
1158
1159 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1160 {
1161         struct request *rq = cmd->request;
1162         struct scsi_device *sdp = cmd->device;
1163         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1164         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1165         sector_t threshold;
1166         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1167         bool dif, dix;
1168         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1169         bool write = rq_data_dir(rq) == WRITE;
1170         unsigned char protect, fua;
1171         blk_status_t ret;
1172
1173         ret = scsi_init_io(cmd);
1174         if (ret != BLK_STS_OK)
1175                 return ret;
1176
1177         if (!scsi_device_online(sdp) || sdp->changed) {
1178                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1179                 return BLK_STS_IOERR;
1180         }
1181
1182         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1183                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1184                 return BLK_STS_IOERR;
1185         }
1186
1187         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1188                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1189                 return BLK_STS_IOERR;
1190         }
1191
1192         /*
1193          * Some SD card readers can't handle accesses which touch the
1194          * last one or two logical blocks. Split accesses as needed.
1195          */
1196         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1197
1198         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1199                 if (lba < threshold) {
1200                         /* Access up to the threshold but not beyond */
1201                         nr_blocks = threshold - lba;
1202                 } else {
1203                         /* Access only a single logical block */
1204                         nr_blocks = 1;
1205                 }
1206         }
1207
1208         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1209         dix = scsi_prot_sg_count(cmd);
1210         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1211
1212         if (write && dix)
1213                 t10_pi_prepare(cmd->request, sdkp->protection_type);
1214
1215         if (dif || dix)
1216                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1217         else
1218                 protect = 0;
1219
1220         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1221                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1222                                          protect | fua);
1223         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1224                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1225                                          protect | fua);
1226         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1227                    sdp->use_10_for_rw || protect) {
1228                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1229                                          protect | fua);
1230         } else {
1231                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1232                                         protect | fua);
1233         }
1234
1235         if (unlikely(ret != BLK_STS_OK))
1236                 return ret;
1237
1238         /*
1239          * We shouldn't disconnect in the middle of a sector, so with a dumb
1240          * host adapter, it's safe to assume that we can at least transfer
1241          * this many bytes between each connect / disconnect.
1242          */
1243         cmd->transfersize = sdp->sector_size;
1244         cmd->underflow = nr_blocks << 9;
1245         cmd->allowed = SD_MAX_RETRIES;
1246         cmd->sdb.length = nr_blocks * sdp->sector_size;
1247
1248         SCSI_LOG_HLQUEUE(1,
1249                          scmd_printk(KERN_INFO, cmd,
1250                                      "%s: block=%llu, count=%d\n", __func__,
1251                                      (unsigned long long)blk_rq_pos(rq),
1252                                      blk_rq_sectors(rq)));
1253         SCSI_LOG_HLQUEUE(2,
1254                          scmd_printk(KERN_INFO, cmd,
1255                                      "%s %d/%u 512 byte blocks.\n",
1256                                      write ? "writing" : "reading", nr_blocks,
1257                                      blk_rq_sectors(rq)));
1258
1259         /*
1260          * This indicates that the command is ready from our end to be
1261          * queued.
1262          */
1263         return BLK_STS_OK;
1264 }
1265
1266 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1267 {
1268         struct request *rq = cmd->request;
1269
1270         switch (req_op(rq)) {
1271         case REQ_OP_DISCARD:
1272                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1273                 case SD_LBP_UNMAP:
1274                         return sd_setup_unmap_cmnd(cmd);
1275                 case SD_LBP_WS16:
1276                         return sd_setup_write_same16_cmnd(cmd, true);
1277                 case SD_LBP_WS10:
1278                         return sd_setup_write_same10_cmnd(cmd, true);
1279                 case SD_LBP_ZERO:
1280                         return sd_setup_write_same10_cmnd(cmd, false);
1281                 default:
1282                         return BLK_STS_TARGET;
1283                 }
1284         case REQ_OP_WRITE_ZEROES:
1285                 return sd_setup_write_zeroes_cmnd(cmd);
1286         case REQ_OP_WRITE_SAME:
1287                 return sd_setup_write_same_cmnd(cmd);
1288         case REQ_OP_FLUSH:
1289                 return sd_setup_flush_cmnd(cmd);
1290         case REQ_OP_READ:
1291         case REQ_OP_WRITE:
1292                 return sd_setup_read_write_cmnd(cmd);
1293         case REQ_OP_ZONE_RESET:
1294                 return sd_zbc_setup_reset_cmnd(cmd);
1295         default:
1296                 WARN_ON_ONCE(1);
1297                 return BLK_STS_NOTSUPP;
1298         }
1299 }
1300
1301 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1302 {
1303         struct request *rq = SCpnt->request;
1304         u8 *cmnd;
1305
1306         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1307                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1308
1309         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1310                 cmnd = SCpnt->cmnd;
1311                 SCpnt->cmnd = NULL;
1312                 SCpnt->cmd_len = 0;
1313                 mempool_free(cmnd, sd_cdb_pool);
1314         }
1315 }
1316
1317 /**
1318  *      sd_open - open a scsi disk device
1319  *      @bdev: Block device of the scsi disk to open
1320  *      @mode: FMODE_* mask
1321  *
1322  *      Returns 0 if successful. Returns a negated errno value in case
1323  *      of error.
1324  *
1325  *      Note: This can be called from a user context (e.g. fsck(1) )
1326  *      or from within the kernel (e.g. as a result of a mount(1) ).
1327  *      In the latter case @inode and @filp carry an abridged amount
1328  *      of information as noted above.
1329  *
1330  *      Locking: called with bdev->bd_mutex held.
1331  **/
1332 static int sd_open(struct block_device *bdev, fmode_t mode)
1333 {
1334         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1335         struct scsi_device *sdev;
1336         int retval;
1337
1338         if (!sdkp)
1339                 return -ENXIO;
1340
1341         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1342
1343         sdev = sdkp->device;
1344
1345         /*
1346          * If the device is in error recovery, wait until it is done.
1347          * If the device is offline, then disallow any access to it.
1348          */
1349         retval = -ENXIO;
1350         if (!scsi_block_when_processing_errors(sdev))
1351                 goto error_out;
1352
1353         if (sdev->removable || sdkp->write_prot)
1354                 check_disk_change(bdev);
1355
1356         /*
1357          * If the drive is empty, just let the open fail.
1358          */
1359         retval = -ENOMEDIUM;
1360         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1361                 goto error_out;
1362
1363         /*
1364          * If the device has the write protect tab set, have the open fail
1365          * if the user expects to be able to write to the thing.
1366          */
1367         retval = -EROFS;
1368         if (sdkp->write_prot && (mode & FMODE_WRITE))
1369                 goto error_out;
1370
1371         /*
1372          * It is possible that the disk changing stuff resulted in
1373          * the device being taken offline.  If this is the case,
1374          * report this to the user, and don't pretend that the
1375          * open actually succeeded.
1376          */
1377         retval = -ENXIO;
1378         if (!scsi_device_online(sdev))
1379                 goto error_out;
1380
1381         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1382                 if (scsi_block_when_processing_errors(sdev))
1383                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1384         }
1385
1386         return 0;
1387
1388 error_out:
1389         scsi_disk_put(sdkp);
1390         return retval;
1391 }
1392
1393 /**
1394  *      sd_release - invoked when the (last) close(2) is called on this
1395  *      scsi disk.
1396  *      @disk: disk to release
1397  *      @mode: FMODE_* mask
1398  *
1399  *      Returns 0.
1400  *
1401  *      Note: may block (uninterruptible) if error recovery is underway
1402  *      on this disk.
1403  *
1404  *      Locking: called with bdev->bd_mutex held.
1405  **/
1406 static void sd_release(struct gendisk *disk, fmode_t mode)
1407 {
1408         struct scsi_disk *sdkp = scsi_disk(disk);
1409         struct scsi_device *sdev = sdkp->device;
1410
1411         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1412
1413         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1414                 if (scsi_block_when_processing_errors(sdev))
1415                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1416         }
1417
1418         scsi_disk_put(sdkp);
1419 }
1420
1421 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1422 {
1423         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1424         struct scsi_device *sdp = sdkp->device;
1425         struct Scsi_Host *host = sdp->host;
1426         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1427         int diskinfo[4];
1428
1429         /* default to most commonly used values */
1430         diskinfo[0] = 0x40;     /* 1 << 6 */
1431         diskinfo[1] = 0x20;     /* 1 << 5 */
1432         diskinfo[2] = capacity >> 11;
1433
1434         /* override with calculated, extended default, or driver values */
1435         if (host->hostt->bios_param)
1436                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1437         else
1438                 scsicam_bios_param(bdev, capacity, diskinfo);
1439
1440         geo->heads = diskinfo[0];
1441         geo->sectors = diskinfo[1];
1442         geo->cylinders = diskinfo[2];
1443         return 0;
1444 }
1445
1446 /**
1447  *      sd_ioctl - process an ioctl
1448  *      @bdev: target block device
1449  *      @mode: FMODE_* mask
1450  *      @cmd: ioctl command number
1451  *      @arg: this is third argument given to ioctl(2) system call.
1452  *      Often contains a pointer.
1453  *
1454  *      Returns 0 if successful (some ioctls return positive numbers on
1455  *      success as well). Returns a negated errno value in case of error.
1456  *
1457  *      Note: most ioctls are forward onto the block subsystem or further
1458  *      down in the scsi subsystem.
1459  **/
1460 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1461                     unsigned int cmd, unsigned long arg)
1462 {
1463         struct gendisk *disk = bdev->bd_disk;
1464         struct scsi_disk *sdkp = scsi_disk(disk);
1465         struct scsi_device *sdp = sdkp->device;
1466         void __user *p = (void __user *)arg;
1467         int error;
1468
1469         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1470                                     "cmd=0x%x\n", disk->disk_name, cmd));
1471
1472         error = scsi_verify_blk_ioctl(bdev, cmd);
1473         if (error < 0)
1474                 return error;
1475
1476         /*
1477          * If we are in the middle of error recovery, don't let anyone
1478          * else try and use this device.  Also, if error recovery fails, it
1479          * may try and take the device offline, in which case all further
1480          * access to the device is prohibited.
1481          */
1482         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1483                         (mode & FMODE_NDELAY) != 0);
1484         if (error)
1485                 goto out;
1486
1487         if (is_sed_ioctl(cmd))
1488                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1489
1490         /*
1491          * Send SCSI addressing ioctls directly to mid level, send other
1492          * ioctls to block level and then onto mid level if they can't be
1493          * resolved.
1494          */
1495         switch (cmd) {
1496                 case SCSI_IOCTL_GET_IDLUN:
1497                 case SCSI_IOCTL_GET_BUS_NUMBER:
1498                         error = scsi_ioctl(sdp, cmd, p);
1499                         break;
1500                 default:
1501                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1502                         if (error != -ENOTTY)
1503                                 break;
1504                         error = scsi_ioctl(sdp, cmd, p);
1505                         break;
1506         }
1507 out:
1508         return error;
1509 }
1510
1511 static void set_media_not_present(struct scsi_disk *sdkp)
1512 {
1513         if (sdkp->media_present)
1514                 sdkp->device->changed = 1;
1515
1516         if (sdkp->device->removable) {
1517                 sdkp->media_present = 0;
1518                 sdkp->capacity = 0;
1519         }
1520 }
1521
1522 static int media_not_present(struct scsi_disk *sdkp,
1523                              struct scsi_sense_hdr *sshdr)
1524 {
1525         if (!scsi_sense_valid(sshdr))
1526                 return 0;
1527
1528         /* not invoked for commands that could return deferred errors */
1529         switch (sshdr->sense_key) {
1530         case UNIT_ATTENTION:
1531         case NOT_READY:
1532                 /* medium not present */
1533                 if (sshdr->asc == 0x3A) {
1534                         set_media_not_present(sdkp);
1535                         return 1;
1536                 }
1537         }
1538         return 0;
1539 }
1540
1541 /**
1542  *      sd_check_events - check media events
1543  *      @disk: kernel device descriptor
1544  *      @clearing: disk events currently being cleared
1545  *
1546  *      Returns mask of DISK_EVENT_*.
1547  *
1548  *      Note: this function is invoked from the block subsystem.
1549  **/
1550 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1551 {
1552         struct scsi_disk *sdkp = scsi_disk_get(disk);
1553         struct scsi_device *sdp;
1554         int retval;
1555
1556         if (!sdkp)
1557                 return 0;
1558
1559         sdp = sdkp->device;
1560         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1561
1562         /*
1563          * If the device is offline, don't send any commands - just pretend as
1564          * if the command failed.  If the device ever comes back online, we
1565          * can deal with it then.  It is only because of unrecoverable errors
1566          * that we would ever take a device offline in the first place.
1567          */
1568         if (!scsi_device_online(sdp)) {
1569                 set_media_not_present(sdkp);
1570                 goto out;
1571         }
1572
1573         /*
1574          * Using TEST_UNIT_READY enables differentiation between drive with
1575          * no cartridge loaded - NOT READY, drive with changed cartridge -
1576          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1577          *
1578          * Drives that auto spin down. eg iomega jaz 1G, will be started
1579          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1580          * sd_revalidate() is called.
1581          */
1582         if (scsi_block_when_processing_errors(sdp)) {
1583                 struct scsi_sense_hdr sshdr = { 0, };
1584
1585                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1586                                               &sshdr);
1587
1588                 /* failed to execute TUR, assume media not present */
1589                 if (host_byte(retval)) {
1590                         set_media_not_present(sdkp);
1591                         goto out;
1592                 }
1593
1594                 if (media_not_present(sdkp, &sshdr))
1595                         goto out;
1596         }
1597
1598         /*
1599          * For removable scsi disk we have to recognise the presence
1600          * of a disk in the drive.
1601          */
1602         if (!sdkp->media_present)
1603                 sdp->changed = 1;
1604         sdkp->media_present = 1;
1605 out:
1606         /*
1607          * sdp->changed is set under the following conditions:
1608          *
1609          *      Medium present state has changed in either direction.
1610          *      Device has indicated UNIT_ATTENTION.
1611          */
1612         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1613         sdp->changed = 0;
1614         scsi_disk_put(sdkp);
1615         return retval;
1616 }
1617
1618 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1619 {
1620         int retries, res;
1621         struct scsi_device *sdp = sdkp->device;
1622         const int timeout = sdp->request_queue->rq_timeout
1623                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1624         struct scsi_sense_hdr my_sshdr;
1625
1626         if (!scsi_device_online(sdp))
1627                 return -ENODEV;
1628
1629         /* caller might not be interested in sense, but we need it */
1630         if (!sshdr)
1631                 sshdr = &my_sshdr;
1632
1633         for (retries = 3; retries > 0; --retries) {
1634                 unsigned char cmd[10] = { 0 };
1635
1636                 cmd[0] = SYNCHRONIZE_CACHE;
1637                 /*
1638                  * Leave the rest of the command zero to indicate
1639                  * flush everything.
1640                  */
1641                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1642                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1643                 if (res == 0)
1644                         break;
1645         }
1646
1647         if (res) {
1648                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1649
1650                 if (driver_byte(res) == DRIVER_SENSE)
1651                         sd_print_sense_hdr(sdkp, sshdr);
1652
1653                 /* we need to evaluate the error return  */
1654                 if (scsi_sense_valid(sshdr) &&
1655                         (sshdr->asc == 0x3a ||  /* medium not present */
1656                          sshdr->asc == 0x20))   /* invalid command */
1657                                 /* this is no error here */
1658                                 return 0;
1659
1660                 switch (host_byte(res)) {
1661                 /* ignore errors due to racing a disconnection */
1662                 case DID_BAD_TARGET:
1663                 case DID_NO_CONNECT:
1664                         return 0;
1665                 /* signal the upper layer it might try again */
1666                 case DID_BUS_BUSY:
1667                 case DID_IMM_RETRY:
1668                 case DID_REQUEUE:
1669                 case DID_SOFT_ERROR:
1670                         return -EBUSY;
1671                 default:
1672                         return -EIO;
1673                 }
1674         }
1675         return 0;
1676 }
1677
1678 static void sd_rescan(struct device *dev)
1679 {
1680         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1681
1682         revalidate_disk(sdkp->disk);
1683 }
1684
1685
1686 #ifdef CONFIG_COMPAT
1687 /*
1688  * This gets directly called from VFS. When the ioctl
1689  * is not recognized we go back to the other translation paths.
1690  */
1691 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1692                            unsigned int cmd, unsigned long arg)
1693 {
1694         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1695         int error;
1696
1697         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1698                         (mode & FMODE_NDELAY) != 0);
1699         if (error)
1700                 return error;
1701
1702         /*
1703          * Let the static ioctl translation table take care of it.
1704          */
1705         if (!sdev->host->hostt->compat_ioctl)
1706                 return -ENOIOCTLCMD;
1707         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1708 }
1709 #endif
1710
1711 static char sd_pr_type(enum pr_type type)
1712 {
1713         switch (type) {
1714         case PR_WRITE_EXCLUSIVE:
1715                 return 0x01;
1716         case PR_EXCLUSIVE_ACCESS:
1717                 return 0x03;
1718         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1719                 return 0x05;
1720         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1721                 return 0x06;
1722         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1723                 return 0x07;
1724         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1725                 return 0x08;
1726         default:
1727                 return 0;
1728         }
1729 };
1730
1731 static int sd_pr_command(struct block_device *bdev, u8 sa,
1732                 u64 key, u64 sa_key, u8 type, u8 flags)
1733 {
1734         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1735         struct scsi_sense_hdr sshdr;
1736         int result;
1737         u8 cmd[16] = { 0, };
1738         u8 data[24] = { 0, };
1739
1740         cmd[0] = PERSISTENT_RESERVE_OUT;
1741         cmd[1] = sa;
1742         cmd[2] = type;
1743         put_unaligned_be32(sizeof(data), &cmd[5]);
1744
1745         put_unaligned_be64(key, &data[0]);
1746         put_unaligned_be64(sa_key, &data[8]);
1747         data[20] = flags;
1748
1749         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1750                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1751
1752         if (driver_byte(result) == DRIVER_SENSE &&
1753             scsi_sense_valid(&sshdr)) {
1754                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1755                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1756         }
1757
1758         return result;
1759 }
1760
1761 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1762                 u32 flags)
1763 {
1764         if (flags & ~PR_FL_IGNORE_KEY)
1765                 return -EOPNOTSUPP;
1766         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1767                         old_key, new_key, 0,
1768                         (1 << 0) /* APTPL */);
1769 }
1770
1771 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1772                 u32 flags)
1773 {
1774         if (flags)
1775                 return -EOPNOTSUPP;
1776         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1777 }
1778
1779 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1780 {
1781         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1782 }
1783
1784 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1785                 enum pr_type type, bool abort)
1786 {
1787         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1788                              sd_pr_type(type), 0);
1789 }
1790
1791 static int sd_pr_clear(struct block_device *bdev, u64 key)
1792 {
1793         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1794 }
1795
1796 static const struct pr_ops sd_pr_ops = {
1797         .pr_register    = sd_pr_register,
1798         .pr_reserve     = sd_pr_reserve,
1799         .pr_release     = sd_pr_release,
1800         .pr_preempt     = sd_pr_preempt,
1801         .pr_clear       = sd_pr_clear,
1802 };
1803
1804 static const struct block_device_operations sd_fops = {
1805         .owner                  = THIS_MODULE,
1806         .open                   = sd_open,
1807         .release                = sd_release,
1808         .ioctl                  = sd_ioctl,
1809         .getgeo                 = sd_getgeo,
1810 #ifdef CONFIG_COMPAT
1811         .compat_ioctl           = sd_compat_ioctl,
1812 #endif
1813         .check_events           = sd_check_events,
1814         .revalidate_disk        = sd_revalidate_disk,
1815         .unlock_native_capacity = sd_unlock_native_capacity,
1816         .report_zones           = sd_zbc_report_zones,
1817         .pr_ops                 = &sd_pr_ops,
1818 };
1819
1820 /**
1821  *      sd_eh_reset - reset error handling callback
1822  *      @scmd:          sd-issued command that has failed
1823  *
1824  *      This function is called by the SCSI midlayer before starting
1825  *      SCSI EH. When counting medium access failures we have to be
1826  *      careful to register it only only once per device and SCSI EH run;
1827  *      there might be several timed out commands which will cause the
1828  *      'max_medium_access_timeouts' counter to trigger after the first
1829  *      SCSI EH run already and set the device to offline.
1830  *      So this function resets the internal counter before starting SCSI EH.
1831  **/
1832 static void sd_eh_reset(struct scsi_cmnd *scmd)
1833 {
1834         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1835
1836         /* New SCSI EH run, reset gate variable */
1837         sdkp->ignore_medium_access_errors = false;
1838 }
1839
1840 /**
1841  *      sd_eh_action - error handling callback
1842  *      @scmd:          sd-issued command that has failed
1843  *      @eh_disp:       The recovery disposition suggested by the midlayer
1844  *
1845  *      This function is called by the SCSI midlayer upon completion of an
1846  *      error test command (currently TEST UNIT READY). The result of sending
1847  *      the eh command is passed in eh_disp.  We're looking for devices that
1848  *      fail medium access commands but are OK with non access commands like
1849  *      test unit ready (so wrongly see the device as having a successful
1850  *      recovery)
1851  **/
1852 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1853 {
1854         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1855         struct scsi_device *sdev = scmd->device;
1856
1857         if (!scsi_device_online(sdev) ||
1858             !scsi_medium_access_command(scmd) ||
1859             host_byte(scmd->result) != DID_TIME_OUT ||
1860             eh_disp != SUCCESS)
1861                 return eh_disp;
1862
1863         /*
1864          * The device has timed out executing a medium access command.
1865          * However, the TEST UNIT READY command sent during error
1866          * handling completed successfully. Either the device is in the
1867          * process of recovering or has it suffered an internal failure
1868          * that prevents access to the storage medium.
1869          */
1870         if (!sdkp->ignore_medium_access_errors) {
1871                 sdkp->medium_access_timed_out++;
1872                 sdkp->ignore_medium_access_errors = true;
1873         }
1874
1875         /*
1876          * If the device keeps failing read/write commands but TEST UNIT
1877          * READY always completes successfully we assume that medium
1878          * access is no longer possible and take the device offline.
1879          */
1880         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1881                 scmd_printk(KERN_ERR, scmd,
1882                             "Medium access timeout failure. Offlining disk!\n");
1883                 mutex_lock(&sdev->state_mutex);
1884                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1885                 mutex_unlock(&sdev->state_mutex);
1886
1887                 return SUCCESS;
1888         }
1889
1890         return eh_disp;
1891 }
1892
1893 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1894 {
1895         struct request *req = scmd->request;
1896         struct scsi_device *sdev = scmd->device;
1897         unsigned int transferred, good_bytes;
1898         u64 start_lba, end_lba, bad_lba;
1899
1900         /*
1901          * Some commands have a payload smaller than the device logical
1902          * block size (e.g. INQUIRY on a 4K disk).
1903          */
1904         if (scsi_bufflen(scmd) <= sdev->sector_size)
1905                 return 0;
1906
1907         /* Check if we have a 'bad_lba' information */
1908         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1909                                      SCSI_SENSE_BUFFERSIZE,
1910                                      &bad_lba))
1911                 return 0;
1912
1913         /*
1914          * If the bad lba was reported incorrectly, we have no idea where
1915          * the error is.
1916          */
1917         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1918         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1919         if (bad_lba < start_lba || bad_lba >= end_lba)
1920                 return 0;
1921
1922         /*
1923          * resid is optional but mostly filled in.  When it's unused,
1924          * its value is zero, so we assume the whole buffer transferred
1925          */
1926         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1927
1928         /* This computation should always be done in terms of the
1929          * resolution of the device's medium.
1930          */
1931         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1932
1933         return min(good_bytes, transferred);
1934 }
1935
1936 /**
1937  *      sd_done - bottom half handler: called when the lower level
1938  *      driver has completed (successfully or otherwise) a scsi command.
1939  *      @SCpnt: mid-level's per command structure.
1940  *
1941  *      Note: potentially run from within an ISR. Must not block.
1942  **/
1943 static int sd_done(struct scsi_cmnd *SCpnt)
1944 {
1945         int result = SCpnt->result;
1946         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1947         unsigned int sector_size = SCpnt->device->sector_size;
1948         unsigned int resid;
1949         struct scsi_sense_hdr sshdr;
1950         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1951         struct request *req = SCpnt->request;
1952         int sense_valid = 0;
1953         int sense_deferred = 0;
1954
1955         switch (req_op(req)) {
1956         case REQ_OP_DISCARD:
1957         case REQ_OP_WRITE_ZEROES:
1958         case REQ_OP_WRITE_SAME:
1959         case REQ_OP_ZONE_RESET:
1960                 if (!result) {
1961                         good_bytes = blk_rq_bytes(req);
1962                         scsi_set_resid(SCpnt, 0);
1963                 } else {
1964                         good_bytes = 0;
1965                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1966                 }
1967                 break;
1968         default:
1969                 /*
1970                  * In case of bogus fw or device, we could end up having
1971                  * an unaligned partial completion. Check this here and force
1972                  * alignment.
1973                  */
1974                 resid = scsi_get_resid(SCpnt);
1975                 if (resid & (sector_size - 1)) {
1976                         sd_printk(KERN_INFO, sdkp,
1977                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1978                                 resid, sector_size);
1979                         resid = min(scsi_bufflen(SCpnt),
1980                                     round_up(resid, sector_size));
1981                         scsi_set_resid(SCpnt, resid);
1982                 }
1983         }
1984
1985         if (result) {
1986                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1987                 if (sense_valid)
1988                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1989         }
1990         sdkp->medium_access_timed_out = 0;
1991
1992         if (driver_byte(result) != DRIVER_SENSE &&
1993             (!sense_valid || sense_deferred))
1994                 goto out;
1995
1996         switch (sshdr.sense_key) {
1997         case HARDWARE_ERROR:
1998         case MEDIUM_ERROR:
1999                 good_bytes = sd_completed_bytes(SCpnt);
2000                 break;
2001         case RECOVERED_ERROR:
2002                 good_bytes = scsi_bufflen(SCpnt);
2003                 break;
2004         case NO_SENSE:
2005                 /* This indicates a false check condition, so ignore it.  An
2006                  * unknown amount of data was transferred so treat it as an
2007                  * error.
2008                  */
2009                 SCpnt->result = 0;
2010                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2011                 break;
2012         case ABORTED_COMMAND:
2013                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2014                         good_bytes = sd_completed_bytes(SCpnt);
2015                 break;
2016         case ILLEGAL_REQUEST:
2017                 switch (sshdr.asc) {
2018                 case 0x10:      /* DIX: Host detected corruption */
2019                         good_bytes = sd_completed_bytes(SCpnt);
2020                         break;
2021                 case 0x20:      /* INVALID COMMAND OPCODE */
2022                 case 0x24:      /* INVALID FIELD IN CDB */
2023                         switch (SCpnt->cmnd[0]) {
2024                         case UNMAP:
2025                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2026                                 break;
2027                         case WRITE_SAME_16:
2028                         case WRITE_SAME:
2029                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2030                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2031                                 } else {
2032                                         sdkp->device->no_write_same = 1;
2033                                         sd_config_write_same(sdkp);
2034                                         req->rq_flags |= RQF_QUIET;
2035                                 }
2036                                 break;
2037                         }
2038                 }
2039                 break;
2040         default:
2041                 break;
2042         }
2043
2044  out:
2045         if (sd_is_zoned(sdkp))
2046                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2047
2048         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2049                                            "sd_done: completed %d of %d bytes\n",
2050                                            good_bytes, scsi_bufflen(SCpnt)));
2051
2052         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2053             good_bytes)
2054                 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2055                                 good_bytes / scsi_prot_interval(SCpnt));
2056
2057         return good_bytes;
2058 }
2059
2060 /*
2061  * spinup disk - called only in sd_revalidate_disk()
2062  */
2063 static void
2064 sd_spinup_disk(struct scsi_disk *sdkp)
2065 {
2066         unsigned char cmd[10];
2067         unsigned long spintime_expire = 0;
2068         int retries, spintime;
2069         unsigned int the_result;
2070         struct scsi_sense_hdr sshdr;
2071         int sense_valid = 0;
2072
2073         spintime = 0;
2074
2075         /* Spin up drives, as required.  Only do this at boot time */
2076         /* Spinup needs to be done for module loads too. */
2077         do {
2078                 retries = 0;
2079
2080                 do {
2081                         cmd[0] = TEST_UNIT_READY;
2082                         memset((void *) &cmd[1], 0, 9);
2083
2084                         the_result = scsi_execute_req(sdkp->device, cmd,
2085                                                       DMA_NONE, NULL, 0,
2086                                                       &sshdr, SD_TIMEOUT,
2087                                                       SD_MAX_RETRIES, NULL);
2088
2089                         /*
2090                          * If the drive has indicated to us that it
2091                          * doesn't have any media in it, don't bother
2092                          * with any more polling.
2093                          */
2094                         if (media_not_present(sdkp, &sshdr))
2095                                 return;
2096
2097                         if (the_result)
2098                                 sense_valid = scsi_sense_valid(&sshdr);
2099                         retries++;
2100                 } while (retries < 3 &&
2101                          (!scsi_status_is_good(the_result) ||
2102                           ((driver_byte(the_result) == DRIVER_SENSE) &&
2103                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2104
2105                 if (driver_byte(the_result) != DRIVER_SENSE) {
2106                         /* no sense, TUR either succeeded or failed
2107                          * with a status error */
2108                         if(!spintime && !scsi_status_is_good(the_result)) {
2109                                 sd_print_result(sdkp, "Test Unit Ready failed",
2110                                                 the_result);
2111                         }
2112                         break;
2113                 }
2114
2115                 /*
2116                  * The device does not want the automatic start to be issued.
2117                  */
2118                 if (sdkp->device->no_start_on_add)
2119                         break;
2120
2121                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2122                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2123                                 break;  /* manual intervention required */
2124                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2125                                 break;  /* standby */
2126                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2127                                 break;  /* unavailable */
2128                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2129                                 break;  /* sanitize in progress */
2130                         /*
2131                          * Issue command to spin up drive when not ready
2132                          */
2133                         if (!spintime) {
2134                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2135                                 cmd[0] = START_STOP;
2136                                 cmd[1] = 1;     /* Return immediately */
2137                                 memset((void *) &cmd[2], 0, 8);
2138                                 cmd[4] = 1;     /* Start spin cycle */
2139                                 if (sdkp->device->start_stop_pwr_cond)
2140                                         cmd[4] |= 1 << 4;
2141                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2142                                                  NULL, 0, &sshdr,
2143                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2144                                                  NULL);
2145                                 spintime_expire = jiffies + 100 * HZ;
2146                                 spintime = 1;
2147                         }
2148                         /* Wait 1 second for next try */
2149                         msleep(1000);
2150                         printk(KERN_CONT ".");
2151
2152                 /*
2153                  * Wait for USB flash devices with slow firmware.
2154                  * Yes, this sense key/ASC combination shouldn't
2155                  * occur here.  It's characteristic of these devices.
2156                  */
2157                 } else if (sense_valid &&
2158                                 sshdr.sense_key == UNIT_ATTENTION &&
2159                                 sshdr.asc == 0x28) {
2160                         if (!spintime) {
2161                                 spintime_expire = jiffies + 5 * HZ;
2162                                 spintime = 1;
2163                         }
2164                         /* Wait 1 second for next try */
2165                         msleep(1000);
2166                 } else {
2167                         /* we don't understand the sense code, so it's
2168                          * probably pointless to loop */
2169                         if(!spintime) {
2170                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2171                                 sd_print_sense_hdr(sdkp, &sshdr);
2172                         }
2173                         break;
2174                 }
2175
2176         } while (spintime && time_before_eq(jiffies, spintime_expire));
2177
2178         if (spintime) {
2179                 if (scsi_status_is_good(the_result))
2180                         printk(KERN_CONT "ready\n");
2181                 else
2182                         printk(KERN_CONT "not responding...\n");
2183         }
2184 }
2185
2186 /*
2187  * Determine whether disk supports Data Integrity Field.
2188  */
2189 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2190 {
2191         struct scsi_device *sdp = sdkp->device;
2192         u8 type;
2193         int ret = 0;
2194
2195         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2196                 return ret;
2197
2198         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2199
2200         if (type > T10_PI_TYPE3_PROTECTION)
2201                 ret = -ENODEV;
2202         else if (scsi_host_dif_capable(sdp->host, type))
2203                 ret = 1;
2204
2205         if (sdkp->first_scan || type != sdkp->protection_type)
2206                 switch (ret) {
2207                 case -ENODEV:
2208                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2209                                   " protection type %u. Disabling disk!\n",
2210                                   type);
2211                         break;
2212                 case 1:
2213                         sd_printk(KERN_NOTICE, sdkp,
2214                                   "Enabling DIF Type %u protection\n", type);
2215                         break;
2216                 case 0:
2217                         sd_printk(KERN_NOTICE, sdkp,
2218                                   "Disabling DIF Type %u protection\n", type);
2219                         break;
2220                 }
2221
2222         sdkp->protection_type = type;
2223
2224         return ret;
2225 }
2226
2227 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2228                         struct scsi_sense_hdr *sshdr, int sense_valid,
2229                         int the_result)
2230 {
2231         if (driver_byte(the_result) == DRIVER_SENSE)
2232                 sd_print_sense_hdr(sdkp, sshdr);
2233         else
2234                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2235
2236         /*
2237          * Set dirty bit for removable devices if not ready -
2238          * sometimes drives will not report this properly.
2239          */
2240         if (sdp->removable &&
2241             sense_valid && sshdr->sense_key == NOT_READY)
2242                 set_media_not_present(sdkp);
2243
2244         /*
2245          * We used to set media_present to 0 here to indicate no media
2246          * in the drive, but some drives fail read capacity even with
2247          * media present, so we can't do that.
2248          */
2249         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2250 }
2251
2252 #define RC16_LEN 32
2253 #if RC16_LEN > SD_BUF_SIZE
2254 #error RC16_LEN must not be more than SD_BUF_SIZE
2255 #endif
2256
2257 #define READ_CAPACITY_RETRIES_ON_RESET  10
2258
2259 /*
2260  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2261  * and the reported logical block size is bigger than 512 bytes. Note
2262  * that last_sector is a u64 and therefore logical_to_sectors() is not
2263  * applicable.
2264  */
2265 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2266 {
2267         u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2268
2269         if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2270                 return false;
2271
2272         return true;
2273 }
2274
2275 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2276                                                 unsigned char *buffer)
2277 {
2278         unsigned char cmd[16];
2279         struct scsi_sense_hdr sshdr;
2280         int sense_valid = 0;
2281         int the_result;
2282         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2283         unsigned int alignment;
2284         unsigned long long lba;
2285         unsigned sector_size;
2286
2287         if (sdp->no_read_capacity_16)
2288                 return -EINVAL;
2289
2290         do {
2291                 memset(cmd, 0, 16);
2292                 cmd[0] = SERVICE_ACTION_IN_16;
2293                 cmd[1] = SAI_READ_CAPACITY_16;
2294                 cmd[13] = RC16_LEN;
2295                 memset(buffer, 0, RC16_LEN);
2296
2297                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2298                                         buffer, RC16_LEN, &sshdr,
2299                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2300
2301                 if (media_not_present(sdkp, &sshdr))
2302                         return -ENODEV;
2303
2304                 if (the_result) {
2305                         sense_valid = scsi_sense_valid(&sshdr);
2306                         if (sense_valid &&
2307                             sshdr.sense_key == ILLEGAL_REQUEST &&
2308                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2309                             sshdr.ascq == 0x00)
2310                                 /* Invalid Command Operation Code or
2311                                  * Invalid Field in CDB, just retry
2312                                  * silently with RC10 */
2313                                 return -EINVAL;
2314                         if (sense_valid &&
2315                             sshdr.sense_key == UNIT_ATTENTION &&
2316                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2317                                 /* Device reset might occur several times,
2318                                  * give it one more chance */
2319                                 if (--reset_retries > 0)
2320                                         continue;
2321                 }
2322                 retries--;
2323
2324         } while (the_result && retries);
2325
2326         if (the_result) {
2327                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2328                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2329                 return -EINVAL;
2330         }
2331
2332         sector_size = get_unaligned_be32(&buffer[8]);
2333         lba = get_unaligned_be64(&buffer[0]);
2334
2335         if (sd_read_protection_type(sdkp, buffer) < 0) {
2336                 sdkp->capacity = 0;
2337                 return -ENODEV;
2338         }
2339
2340         if (!sd_addressable_capacity(lba, sector_size)) {
2341                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2342                         "kernel compiled with support for large block "
2343                         "devices.\n");
2344                 sdkp->capacity = 0;
2345                 return -EOVERFLOW;
2346         }
2347
2348         /* Logical blocks per physical block exponent */
2349         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2350
2351         /* RC basis */
2352         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2353
2354         /* Lowest aligned logical block */
2355         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2356         blk_queue_alignment_offset(sdp->request_queue, alignment);
2357         if (alignment && sdkp->first_scan)
2358                 sd_printk(KERN_NOTICE, sdkp,
2359                           "physical block alignment offset: %u\n", alignment);
2360
2361         if (buffer[14] & 0x80) { /* LBPME */
2362                 sdkp->lbpme = 1;
2363
2364                 if (buffer[14] & 0x40) /* LBPRZ */
2365                         sdkp->lbprz = 1;
2366
2367                 sd_config_discard(sdkp, SD_LBP_WS16);
2368         }
2369
2370         sdkp->capacity = lba + 1;
2371         return sector_size;
2372 }
2373
2374 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2375                                                 unsigned char *buffer)
2376 {
2377         unsigned char cmd[16];
2378         struct scsi_sense_hdr sshdr;
2379         int sense_valid = 0;
2380         int the_result;
2381         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2382         sector_t lba;
2383         unsigned sector_size;
2384
2385         do {
2386                 cmd[0] = READ_CAPACITY;
2387                 memset(&cmd[1], 0, 9);
2388                 memset(buffer, 0, 8);
2389
2390                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2391                                         buffer, 8, &sshdr,
2392                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2393
2394                 if (media_not_present(sdkp, &sshdr))
2395                         return -ENODEV;
2396
2397                 if (the_result) {
2398                         sense_valid = scsi_sense_valid(&sshdr);
2399                         if (sense_valid &&
2400                             sshdr.sense_key == UNIT_ATTENTION &&
2401                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2402                                 /* Device reset might occur several times,
2403                                  * give it one more chance */
2404                                 if (--reset_retries > 0)
2405                                         continue;
2406                 }
2407                 retries--;
2408
2409         } while (the_result && retries);
2410
2411         if (the_result) {
2412                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2413                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2414                 return -EINVAL;
2415         }
2416
2417         sector_size = get_unaligned_be32(&buffer[4]);
2418         lba = get_unaligned_be32(&buffer[0]);
2419
2420         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2421                 /* Some buggy (usb cardreader) devices return an lba of
2422                    0xffffffff when the want to report a size of 0 (with
2423                    which they really mean no media is present) */
2424                 sdkp->capacity = 0;
2425                 sdkp->physical_block_size = sector_size;
2426                 return sector_size;
2427         }
2428
2429         if (!sd_addressable_capacity(lba, sector_size)) {
2430                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2431                         "kernel compiled with support for large block "
2432                         "devices.\n");
2433                 sdkp->capacity = 0;
2434                 return -EOVERFLOW;
2435         }
2436
2437         sdkp->capacity = lba + 1;
2438         sdkp->physical_block_size = sector_size;
2439         return sector_size;
2440 }
2441
2442 static int sd_try_rc16_first(struct scsi_device *sdp)
2443 {
2444         if (sdp->host->max_cmd_len < 16)
2445                 return 0;
2446         if (sdp->try_rc_10_first)
2447                 return 0;
2448         if (sdp->scsi_level > SCSI_SPC_2)
2449                 return 1;
2450         if (scsi_device_protection(sdp))
2451                 return 1;
2452         return 0;
2453 }
2454
2455 /*
2456  * read disk capacity
2457  */
2458 static void
2459 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2460 {
2461         int sector_size;
2462         struct scsi_device *sdp = sdkp->device;
2463
2464         if (sd_try_rc16_first(sdp)) {
2465                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2466                 if (sector_size == -EOVERFLOW)
2467                         goto got_data;
2468                 if (sector_size == -ENODEV)
2469                         return;
2470                 if (sector_size < 0)
2471                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2472                 if (sector_size < 0)
2473                         return;
2474         } else {
2475                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2476                 if (sector_size == -EOVERFLOW)
2477                         goto got_data;
2478                 if (sector_size < 0)
2479                         return;
2480                 if ((sizeof(sdkp->capacity) > 4) &&
2481                     (sdkp->capacity > 0xffffffffULL)) {
2482                         int old_sector_size = sector_size;
2483                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2484                                         "Trying to use READ CAPACITY(16).\n");
2485                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2486                         if (sector_size < 0) {
2487                                 sd_printk(KERN_NOTICE, sdkp,
2488                                         "Using 0xffffffff as device size\n");
2489                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2490                                 sector_size = old_sector_size;
2491                                 goto got_data;
2492                         }
2493                         /* Remember that READ CAPACITY(16) succeeded */
2494                         sdp->try_rc_10_first = 0;
2495                 }
2496         }
2497
2498         /* Some devices are known to return the total number of blocks,
2499          * not the highest block number.  Some devices have versions
2500          * which do this and others which do not.  Some devices we might
2501          * suspect of doing this but we don't know for certain.
2502          *
2503          * If we know the reported capacity is wrong, decrement it.  If
2504          * we can only guess, then assume the number of blocks is even
2505          * (usually true but not always) and err on the side of lowering
2506          * the capacity.
2507          */
2508         if (sdp->fix_capacity ||
2509             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2510                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2511                                 "from its reported value: %llu\n",
2512                                 (unsigned long long) sdkp->capacity);
2513                 --sdkp->capacity;
2514         }
2515
2516 got_data:
2517         if (sector_size == 0) {
2518                 sector_size = 512;
2519                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2520                           "assuming 512.\n");
2521         }
2522
2523         if (sector_size != 512 &&
2524             sector_size != 1024 &&
2525             sector_size != 2048 &&
2526             sector_size != 4096) {
2527                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2528                           sector_size);
2529                 /*
2530                  * The user might want to re-format the drive with
2531                  * a supported sectorsize.  Once this happens, it
2532                  * would be relatively trivial to set the thing up.
2533                  * For this reason, we leave the thing in the table.
2534                  */
2535                 sdkp->capacity = 0;
2536                 /*
2537                  * set a bogus sector size so the normal read/write
2538                  * logic in the block layer will eventually refuse any
2539                  * request on this device without tripping over power
2540                  * of two sector size assumptions
2541                  */
2542                 sector_size = 512;
2543         }
2544         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2545         blk_queue_physical_block_size(sdp->request_queue,
2546                                       sdkp->physical_block_size);
2547         sdkp->device->sector_size = sector_size;
2548
2549         if (sdkp->capacity > 0xffffffff)
2550                 sdp->use_16_for_rw = 1;
2551
2552 }
2553
2554 /*
2555  * Print disk capacity
2556  */
2557 static void
2558 sd_print_capacity(struct scsi_disk *sdkp,
2559                   sector_t old_capacity)
2560 {
2561         int sector_size = sdkp->device->sector_size;
2562         char cap_str_2[10], cap_str_10[10];
2563
2564         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2565                 return;
2566
2567         string_get_size(sdkp->capacity, sector_size,
2568                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2569         string_get_size(sdkp->capacity, sector_size,
2570                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2571
2572         sd_printk(KERN_NOTICE, sdkp,
2573                   "%llu %d-byte logical blocks: (%s/%s)\n",
2574                   (unsigned long long)sdkp->capacity,
2575                   sector_size, cap_str_10, cap_str_2);
2576
2577         if (sdkp->physical_block_size != sector_size)
2578                 sd_printk(KERN_NOTICE, sdkp,
2579                           "%u-byte physical blocks\n",
2580                           sdkp->physical_block_size);
2581
2582         sd_zbc_print_zones(sdkp);
2583 }
2584
2585 /* called with buffer of length 512 */
2586 static inline int
2587 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2588                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2589                  struct scsi_sense_hdr *sshdr)
2590 {
2591         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2592                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2593                                sshdr);
2594 }
2595
2596 /*
2597  * read write protect setting, if possible - called only in sd_revalidate_disk()
2598  * called with buffer of length SD_BUF_SIZE
2599  */
2600 static void
2601 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2602 {
2603         int res;
2604         struct scsi_device *sdp = sdkp->device;
2605         struct scsi_mode_data data;
2606         int disk_ro = get_disk_ro(sdkp->disk);
2607         int old_wp = sdkp->write_prot;
2608
2609         set_disk_ro(sdkp->disk, 0);
2610         if (sdp->skip_ms_page_3f) {
2611                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2612                 return;
2613         }
2614
2615         if (sdp->use_192_bytes_for_3f) {
2616                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2617         } else {
2618                 /*
2619                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2620                  * We have to start carefully: some devices hang if we ask
2621                  * for more than is available.
2622                  */
2623                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2624
2625                 /*
2626                  * Second attempt: ask for page 0 When only page 0 is
2627                  * implemented, a request for page 3F may return Sense Key
2628                  * 5: Illegal Request, Sense Code 24: Invalid field in
2629                  * CDB.
2630                  */
2631                 if (!scsi_status_is_good(res))
2632                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2633
2634                 /*
2635                  * Third attempt: ask 255 bytes, as we did earlier.
2636                  */
2637                 if (!scsi_status_is_good(res))
2638                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2639                                                &data, NULL);
2640         }
2641
2642         if (!scsi_status_is_good(res)) {
2643                 sd_first_printk(KERN_WARNING, sdkp,
2644                           "Test WP failed, assume Write Enabled\n");
2645         } else {
2646                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2647                 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2648                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2649                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2650                                   sdkp->write_prot ? "on" : "off");
2651                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2652                 }
2653         }
2654 }
2655
2656 /*
2657  * sd_read_cache_type - called only from sd_revalidate_disk()
2658  * called with buffer of length SD_BUF_SIZE
2659  */
2660 static void
2661 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2662 {
2663         int len = 0, res;
2664         struct scsi_device *sdp = sdkp->device;
2665
2666         int dbd;
2667         int modepage;
2668         int first_len;
2669         struct scsi_mode_data data;
2670         struct scsi_sense_hdr sshdr;
2671         int old_wce = sdkp->WCE;
2672         int old_rcd = sdkp->RCD;
2673         int old_dpofua = sdkp->DPOFUA;
2674
2675
2676         if (sdkp->cache_override)
2677                 return;
2678
2679         first_len = 4;
2680         if (sdp->skip_ms_page_8) {
2681                 if (sdp->type == TYPE_RBC)
2682                         goto defaults;
2683                 else {
2684                         if (sdp->skip_ms_page_3f)
2685                                 goto defaults;
2686                         modepage = 0x3F;
2687                         if (sdp->use_192_bytes_for_3f)
2688                                 first_len = 192;
2689                         dbd = 0;
2690                 }
2691         } else if (sdp->type == TYPE_RBC) {
2692                 modepage = 6;
2693                 dbd = 8;
2694         } else {
2695                 modepage = 8;
2696                 dbd = 0;
2697         }
2698
2699         /* cautiously ask */
2700         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2701                         &data, &sshdr);
2702
2703         if (!scsi_status_is_good(res))
2704                 goto bad_sense;
2705
2706         if (!data.header_length) {
2707                 modepage = 6;
2708                 first_len = 0;
2709                 sd_first_printk(KERN_ERR, sdkp,
2710                                 "Missing header in MODE_SENSE response\n");
2711         }
2712
2713         /* that went OK, now ask for the proper length */
2714         len = data.length;
2715
2716         /*
2717          * We're only interested in the first three bytes, actually.
2718          * But the data cache page is defined for the first 20.
2719          */
2720         if (len < 3)
2721                 goto bad_sense;
2722         else if (len > SD_BUF_SIZE) {
2723                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2724                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2725                 len = SD_BUF_SIZE;
2726         }
2727         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2728                 len = 192;
2729
2730         /* Get the data */
2731         if (len > first_len)
2732                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2733                                 &data, &sshdr);
2734
2735         if (scsi_status_is_good(res)) {
2736                 int offset = data.header_length + data.block_descriptor_length;
2737
2738                 while (offset < len) {
2739                         u8 page_code = buffer[offset] & 0x3F;
2740                         u8 spf       = buffer[offset] & 0x40;
2741
2742                         if (page_code == 8 || page_code == 6) {
2743                                 /* We're interested only in the first 3 bytes.
2744                                  */
2745                                 if (len - offset <= 2) {
2746                                         sd_first_printk(KERN_ERR, sdkp,
2747                                                 "Incomplete mode parameter "
2748                                                         "data\n");
2749                                         goto defaults;
2750                                 } else {
2751                                         modepage = page_code;
2752                                         goto Page_found;
2753                                 }
2754                         } else {
2755                                 /* Go to the next page */
2756                                 if (spf && len - offset > 3)
2757                                         offset += 4 + (buffer[offset+2] << 8) +
2758                                                 buffer[offset+3];
2759                                 else if (!spf && len - offset > 1)
2760                                         offset += 2 + buffer[offset+1];
2761                                 else {
2762                                         sd_first_printk(KERN_ERR, sdkp,
2763                                                         "Incomplete mode "
2764                                                         "parameter data\n");
2765                                         goto defaults;
2766                                 }
2767                         }
2768                 }
2769
2770                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2771                 goto defaults;
2772
2773         Page_found:
2774                 if (modepage == 8) {
2775                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2776                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2777                 } else {
2778                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2779                         sdkp->RCD = 0;
2780                 }
2781
2782                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2783                 if (sdp->broken_fua) {
2784                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2785                         sdkp->DPOFUA = 0;
2786                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2787                            !sdkp->device->use_16_for_rw) {
2788                         sd_first_printk(KERN_NOTICE, sdkp,
2789                                   "Uses READ/WRITE(6), disabling FUA\n");
2790                         sdkp->DPOFUA = 0;
2791                 }
2792
2793                 /* No cache flush allowed for write protected devices */
2794                 if (sdkp->WCE && sdkp->write_prot)
2795                         sdkp->WCE = 0;
2796
2797                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2798                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2799                         sd_printk(KERN_NOTICE, sdkp,
2800                                   "Write cache: %s, read cache: %s, %s\n",
2801                                   sdkp->WCE ? "enabled" : "disabled",
2802                                   sdkp->RCD ? "disabled" : "enabled",
2803                                   sdkp->DPOFUA ? "supports DPO and FUA"
2804                                   : "doesn't support DPO or FUA");
2805
2806                 return;
2807         }
2808
2809 bad_sense:
2810         if (scsi_sense_valid(&sshdr) &&
2811             sshdr.sense_key == ILLEGAL_REQUEST &&
2812             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2813                 /* Invalid field in CDB */
2814                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2815         else
2816                 sd_first_printk(KERN_ERR, sdkp,
2817                                 "Asking for cache data failed\n");
2818
2819 defaults:
2820         if (sdp->wce_default_on) {
2821                 sd_first_printk(KERN_NOTICE, sdkp,
2822                                 "Assuming drive cache: write back\n");
2823                 sdkp->WCE = 1;
2824         } else {
2825                 sd_first_printk(KERN_ERR, sdkp,
2826                                 "Assuming drive cache: write through\n");
2827                 sdkp->WCE = 0;
2828         }
2829         sdkp->RCD = 0;
2830         sdkp->DPOFUA = 0;
2831 }
2832
2833 /*
2834  * The ATO bit indicates whether the DIF application tag is available
2835  * for use by the operating system.
2836  */
2837 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2838 {
2839         int res, offset;
2840         struct scsi_device *sdp = sdkp->device;
2841         struct scsi_mode_data data;
2842         struct scsi_sense_hdr sshdr;
2843
2844         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2845                 return;
2846
2847         if (sdkp->protection_type == 0)
2848                 return;
2849
2850         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2851                               SD_MAX_RETRIES, &data, &sshdr);
2852
2853         if (!scsi_status_is_good(res) || !data.header_length ||
2854             data.length < 6) {
2855                 sd_first_printk(KERN_WARNING, sdkp,
2856                           "getting Control mode page failed, assume no ATO\n");
2857
2858                 if (scsi_sense_valid(&sshdr))
2859                         sd_print_sense_hdr(sdkp, &sshdr);
2860
2861                 return;
2862         }
2863
2864         offset = data.header_length + data.block_descriptor_length;
2865
2866         if ((buffer[offset] & 0x3f) != 0x0a) {
2867                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2868                 return;
2869         }
2870
2871         if ((buffer[offset + 5] & 0x80) == 0)
2872                 return;
2873
2874         sdkp->ATO = 1;
2875
2876         return;
2877 }
2878
2879 /**
2880  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2881  * @sdkp: disk to query
2882  */
2883 static void sd_read_block_limits(struct scsi_disk *sdkp)
2884 {
2885         unsigned int sector_sz = sdkp->device->sector_size;
2886         const int vpd_len = 64;
2887         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2888
2889         if (!buffer ||
2890             /* Block Limits VPD */
2891             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2892                 goto out;
2893
2894         blk_queue_io_min(sdkp->disk->queue,
2895                          get_unaligned_be16(&buffer[6]) * sector_sz);
2896
2897         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2898         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2899
2900         if (buffer[3] == 0x3c) {
2901                 unsigned int lba_count, desc_count;
2902
2903                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2904
2905                 if (!sdkp->lbpme)
2906                         goto out;
2907
2908                 lba_count = get_unaligned_be32(&buffer[20]);
2909                 desc_count = get_unaligned_be32(&buffer[24]);
2910
2911                 if (lba_count && desc_count)
2912                         sdkp->max_unmap_blocks = lba_count;
2913
2914                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2915
2916                 if (buffer[32] & 0x80)
2917                         sdkp->unmap_alignment =
2918                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2919
2920                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2921
2922                         if (sdkp->max_unmap_blocks)
2923                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2924                         else
2925                                 sd_config_discard(sdkp, SD_LBP_WS16);
2926
2927                 } else {        /* LBP VPD page tells us what to use */
2928                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2929                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2930                         else if (sdkp->lbpws)
2931                                 sd_config_discard(sdkp, SD_LBP_WS16);
2932                         else if (sdkp->lbpws10)
2933                                 sd_config_discard(sdkp, SD_LBP_WS10);
2934                         else
2935                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2936                 }
2937         }
2938
2939  out:
2940         kfree(buffer);
2941 }
2942
2943 /**
2944  * sd_read_block_characteristics - Query block dev. characteristics
2945  * @sdkp: disk to query
2946  */
2947 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2948 {
2949         struct request_queue *q = sdkp->disk->queue;
2950         unsigned char *buffer;
2951         u16 rot;
2952         const int vpd_len = 64;
2953
2954         buffer = kmalloc(vpd_len, GFP_KERNEL);
2955
2956         if (!buffer ||
2957             /* Block Device Characteristics VPD */
2958             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2959                 goto out;
2960
2961         rot = get_unaligned_be16(&buffer[4]);
2962
2963         if (rot == 1) {
2964                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2965                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2966         }
2967
2968         if (sdkp->device->type == TYPE_ZBC) {
2969                 /* Host-managed */
2970                 q->limits.zoned = BLK_ZONED_HM;
2971         } else {
2972                 sdkp->zoned = (buffer[8] >> 4) & 3;
2973                 if (sdkp->zoned == 1)
2974                         /* Host-aware */
2975                         q->limits.zoned = BLK_ZONED_HA;
2976                 else
2977                         /*
2978                          * Treat drive-managed devices as
2979                          * regular block devices.
2980                          */
2981                         q->limits.zoned = BLK_ZONED_NONE;
2982         }
2983         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2984                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2985                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2986
2987  out:
2988         kfree(buffer);
2989 }
2990
2991 /**
2992  * sd_read_block_provisioning - Query provisioning VPD page
2993  * @sdkp: disk to query
2994  */
2995 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2996 {
2997         unsigned char *buffer;
2998         const int vpd_len = 8;
2999
3000         if (sdkp->lbpme == 0)
3001                 return;
3002
3003         buffer = kmalloc(vpd_len, GFP_KERNEL);
3004
3005         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3006                 goto out;
3007
3008         sdkp->lbpvpd    = 1;
3009         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3010         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3011         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3012
3013  out:
3014         kfree(buffer);
3015 }
3016
3017 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3018 {
3019         struct scsi_device *sdev = sdkp->device;
3020
3021         if (sdev->host->no_write_same) {
3022                 sdev->no_write_same = 1;
3023
3024                 return;
3025         }
3026
3027         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3028                 /* too large values might cause issues with arcmsr */
3029                 int vpd_buf_len = 64;
3030
3031                 sdev->no_report_opcodes = 1;
3032
3033                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3034                  * CODES is unsupported and the device has an ATA
3035                  * Information VPD page (SAT).
3036                  */
3037                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3038                         sdev->no_write_same = 1;
3039         }
3040
3041         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3042                 sdkp->ws16 = 1;
3043
3044         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3045                 sdkp->ws10 = 1;
3046 }
3047
3048 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3049 {
3050         struct scsi_device *sdev = sdkp->device;
3051
3052         if (!sdev->security_supported)
3053                 return;
3054
3055         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3056                         SECURITY_PROTOCOL_IN) == 1 &&
3057             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3058                         SECURITY_PROTOCOL_OUT) == 1)
3059                 sdkp->security = 1;
3060 }
3061
3062 /*
3063  * Determine the device's preferred I/O size for reads and writes
3064  * unless the reported value is unreasonably small, large, not a
3065  * multiple of the physical block size, or simply garbage.
3066  */
3067 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3068                                       unsigned int dev_max)
3069 {
3070         struct scsi_device *sdp = sdkp->device;
3071         unsigned int opt_xfer_bytes =
3072                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3073
3074         if (sdkp->opt_xfer_blocks == 0)
3075                 return false;
3076
3077         if (sdkp->opt_xfer_blocks > dev_max) {
3078                 sd_first_printk(KERN_WARNING, sdkp,
3079                                 "Optimal transfer size %u logical blocks " \
3080                                 "> dev_max (%u logical blocks)\n",
3081                                 sdkp->opt_xfer_blocks, dev_max);
3082                 return false;
3083         }
3084
3085         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3086                 sd_first_printk(KERN_WARNING, sdkp,
3087                                 "Optimal transfer size %u logical blocks " \
3088                                 "> sd driver limit (%u logical blocks)\n",
3089                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3090                 return false;
3091         }
3092
3093         if (opt_xfer_bytes < PAGE_SIZE) {
3094                 sd_first_printk(KERN_WARNING, sdkp,
3095                                 "Optimal transfer size %u bytes < " \
3096                                 "PAGE_SIZE (%u bytes)\n",
3097                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3098                 return false;
3099         }
3100
3101         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3102                 sd_first_printk(KERN_WARNING, sdkp,
3103                                 "Optimal transfer size %u bytes not a " \
3104                                 "multiple of physical block size (%u bytes)\n",
3105                                 opt_xfer_bytes, sdkp->physical_block_size);
3106                 return false;
3107         }
3108
3109         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3110                         opt_xfer_bytes);
3111         return true;
3112 }
3113
3114 /**
3115  *      sd_revalidate_disk - called the first time a new disk is seen,
3116  *      performs disk spin up, read_capacity, etc.
3117  *      @disk: struct gendisk we care about
3118  **/
3119 static int sd_revalidate_disk(struct gendisk *disk)
3120 {
3121         struct scsi_disk *sdkp = scsi_disk(disk);
3122         struct scsi_device *sdp = sdkp->device;
3123         struct request_queue *q = sdkp->disk->queue;
3124         sector_t old_capacity = sdkp->capacity;
3125         unsigned char *buffer;
3126         unsigned int dev_max, rw_max;
3127
3128         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3129                                       "sd_revalidate_disk\n"));
3130
3131         /*
3132          * If the device is offline, don't try and read capacity or any
3133          * of the other niceties.
3134          */
3135         if (!scsi_device_online(sdp))
3136                 goto out;
3137
3138         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3139         if (!buffer) {
3140                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3141                           "allocation failure.\n");
3142                 goto out;
3143         }
3144
3145         sd_spinup_disk(sdkp);
3146
3147         /*
3148          * Without media there is no reason to ask; moreover, some devices
3149          * react badly if we do.
3150          */
3151         if (sdkp->media_present) {
3152                 sd_read_capacity(sdkp, buffer);
3153
3154                 /*
3155                  * set the default to rotational.  All non-rotational devices
3156                  * support the block characteristics VPD page, which will
3157                  * cause this to be updated correctly and any device which
3158                  * doesn't support it should be treated as rotational.
3159                  */
3160                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3161                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3162
3163                 if (scsi_device_supports_vpd(sdp)) {
3164                         sd_read_block_provisioning(sdkp);
3165                         sd_read_block_limits(sdkp);
3166                         sd_read_block_characteristics(sdkp);
3167                         sd_zbc_read_zones(sdkp, buffer);
3168                 }
3169
3170                 sd_print_capacity(sdkp, old_capacity);
3171
3172                 sd_read_write_protect_flag(sdkp, buffer);
3173                 sd_read_cache_type(sdkp, buffer);
3174                 sd_read_app_tag_own(sdkp, buffer);
3175                 sd_read_write_same(sdkp, buffer);
3176                 sd_read_security(sdkp, buffer);
3177         }
3178
3179         /*
3180          * We now have all cache related info, determine how we deal
3181          * with flush requests.
3182          */
3183         sd_set_flush_flag(sdkp);
3184
3185         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3186         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3187
3188         /* Some devices report a maximum block count for READ/WRITE requests. */
3189         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3190         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3191
3192         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3193                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3194                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3195         } else
3196                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3197                                       (sector_t)BLK_DEF_MAX_SECTORS);
3198
3199         /* Do not exceed controller limit */
3200         rw_max = min(rw_max, queue_max_hw_sectors(q));
3201
3202         /*
3203          * Only update max_sectors if previously unset or if the current value
3204          * exceeds the capabilities of the hardware.
3205          */
3206         if (sdkp->first_scan ||
3207             q->limits.max_sectors > q->limits.max_dev_sectors ||
3208             q->limits.max_sectors > q->limits.max_hw_sectors)
3209                 q->limits.max_sectors = rw_max;
3210
3211         sdkp->first_scan = 0;
3212
3213         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3214         sd_config_write_same(sdkp);
3215         kfree(buffer);
3216
3217  out:
3218         return 0;
3219 }
3220
3221 /**
3222  *      sd_unlock_native_capacity - unlock native capacity
3223  *      @disk: struct gendisk to set capacity for
3224  *
3225  *      Block layer calls this function if it detects that partitions
3226  *      on @disk reach beyond the end of the device.  If the SCSI host
3227  *      implements ->unlock_native_capacity() method, it's invoked to
3228  *      give it a chance to adjust the device capacity.
3229  *
3230  *      CONTEXT:
3231  *      Defined by block layer.  Might sleep.
3232  */
3233 static void sd_unlock_native_capacity(struct gendisk *disk)
3234 {
3235         struct scsi_device *sdev = scsi_disk(disk)->device;
3236
3237         if (sdev->host->hostt->unlock_native_capacity)
3238                 sdev->host->hostt->unlock_native_capacity(sdev);
3239 }
3240
3241 /**
3242  *      sd_format_disk_name - format disk name
3243  *      @prefix: name prefix - ie. "sd" for SCSI disks
3244  *      @index: index of the disk to format name for
3245  *      @buf: output buffer
3246  *      @buflen: length of the output buffer
3247  *
3248  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3249  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3250  *      which is followed by sdaaa.
3251  *
3252  *      This is basically 26 base counting with one extra 'nil' entry
3253  *      at the beginning from the second digit on and can be
3254  *      determined using similar method as 26 base conversion with the
3255  *      index shifted -1 after each digit is computed.
3256  *
3257  *      CONTEXT:
3258  *      Don't care.
3259  *
3260  *      RETURNS:
3261  *      0 on success, -errno on failure.
3262  */
3263 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3264 {
3265         const int base = 'z' - 'a' + 1;
3266         char *begin = buf + strlen(prefix);
3267         char *end = buf + buflen;
3268         char *p;
3269         int unit;
3270
3271         p = end - 1;
3272         *p = '\0';
3273         unit = base;
3274         do {
3275                 if (p == begin)
3276                         return -EINVAL;
3277                 *--p = 'a' + (index % unit);
3278                 index = (index / unit) - 1;
3279         } while (index >= 0);
3280
3281         memmove(begin, p, end - p);
3282         memcpy(buf, prefix, strlen(prefix));
3283
3284         return 0;
3285 }
3286
3287 /*
3288  * The asynchronous part of sd_probe
3289  */
3290 static void sd_probe_async(void *data, async_cookie_t cookie)
3291 {
3292         struct scsi_disk *sdkp = data;
3293         struct scsi_device *sdp;
3294         struct gendisk *gd;
3295         u32 index;
3296         struct device *dev;
3297
3298         sdp = sdkp->device;
3299         gd = sdkp->disk;
3300         index = sdkp->index;
3301         dev = &sdp->sdev_gendev;
3302
3303         gd->major = sd_major((index & 0xf0) >> 4);
3304         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3305
3306         gd->fops = &sd_fops;
3307         gd->private_data = &sdkp->driver;
3308         gd->queue = sdkp->device->request_queue;
3309
3310         /* defaults, until the device tells us otherwise */
3311         sdp->sector_size = 512;
3312         sdkp->capacity = 0;
3313         sdkp->media_present = 1;
3314         sdkp->write_prot = 0;
3315         sdkp->cache_override = 0;
3316         sdkp->WCE = 0;
3317         sdkp->RCD = 0;
3318         sdkp->ATO = 0;
3319         sdkp->first_scan = 1;
3320         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3321
3322         sd_revalidate_disk(gd);
3323
3324         gd->flags = GENHD_FL_EXT_DEVT;
3325         if (sdp->removable) {
3326                 gd->flags |= GENHD_FL_REMOVABLE;
3327                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3328         }
3329
3330         blk_pm_runtime_init(sdp->request_queue, dev);
3331         device_add_disk(dev, gd, NULL);
3332         if (sdkp->capacity)
3333                 sd_dif_config_host(sdkp);
3334
3335         sd_revalidate_disk(gd);
3336
3337         if (sdkp->security) {
3338                 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3339                 if (sdkp->opal_dev)
3340                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3341         }
3342
3343         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3344                   sdp->removable ? "removable " : "");
3345         scsi_autopm_put_device(sdp);
3346         put_device(&sdkp->dev);
3347 }
3348
3349 /**
3350  *      sd_probe - called during driver initialization and whenever a
3351  *      new scsi device is attached to the system. It is called once
3352  *      for each scsi device (not just disks) present.
3353  *      @dev: pointer to device object
3354  *
3355  *      Returns 0 if successful (or not interested in this scsi device
3356  *      (e.g. scanner)); 1 when there is an error.
3357  *
3358  *      Note: this function is invoked from the scsi mid-level.
3359  *      This function sets up the mapping between a given
3360  *      <host,channel,id,lun> (found in sdp) and new device name
3361  *      (e.g. /dev/sda). More precisely it is the block device major
3362  *      and minor number that is chosen here.
3363  *
3364  *      Assume sd_probe is not re-entrant (for time being)
3365  *      Also think about sd_probe() and sd_remove() running coincidentally.
3366  **/
3367 static int sd_probe(struct device *dev)
3368 {
3369         struct scsi_device *sdp = to_scsi_device(dev);
3370         struct scsi_disk *sdkp;
3371         struct gendisk *gd;
3372         int index;
3373         int error;
3374
3375         scsi_autopm_get_device(sdp);
3376         error = -ENODEV;
3377         if (sdp->type != TYPE_DISK &&
3378             sdp->type != TYPE_ZBC &&
3379             sdp->type != TYPE_MOD &&
3380             sdp->type != TYPE_RBC)
3381                 goto out;
3382
3383 #ifndef CONFIG_BLK_DEV_ZONED
3384         if (sdp->type == TYPE_ZBC)
3385                 goto out;
3386 #endif
3387         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3388                                         "sd_probe\n"));
3389
3390         error = -ENOMEM;
3391         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3392         if (!sdkp)
3393                 goto out;
3394
3395         gd = alloc_disk(SD_MINORS);
3396         if (!gd)
3397                 goto out_free;
3398
3399         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3400         if (index < 0) {
3401                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3402                 goto out_put;
3403         }
3404
3405         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3406         if (error) {
3407                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3408                 goto out_free_index;
3409         }
3410
3411         sdkp->device = sdp;
3412         sdkp->driver = &sd_template;
3413         sdkp->disk = gd;
3414         sdkp->index = index;
3415         atomic_set(&sdkp->openers, 0);
3416         atomic_set(&sdkp->device->ioerr_cnt, 0);
3417
3418         if (!sdp->request_queue->rq_timeout) {
3419                 if (sdp->type != TYPE_MOD)
3420                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3421                 else
3422                         blk_queue_rq_timeout(sdp->request_queue,
3423                                              SD_MOD_TIMEOUT);
3424         }
3425
3426         device_initialize(&sdkp->dev);
3427         sdkp->dev.parent = dev;
3428         sdkp->dev.class = &sd_disk_class;
3429         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3430
3431         error = device_add(&sdkp->dev);
3432         if (error)
3433                 goto out_free_index;
3434
3435         get_device(dev);
3436         dev_set_drvdata(dev, sdkp);
3437
3438         get_device(&sdkp->dev); /* prevent release before async_schedule */
3439         async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3440
3441         return 0;
3442
3443  out_free_index:
3444         ida_free(&sd_index_ida, index);
3445  out_put:
3446         put_disk(gd);
3447  out_free:
3448         kfree(sdkp);
3449  out:
3450         scsi_autopm_put_device(sdp);
3451         return error;
3452 }
3453
3454 /**
3455  *      sd_remove - called whenever a scsi disk (previously recognized by
3456  *      sd_probe) is detached from the system. It is called (potentially
3457  *      multiple times) during sd module unload.
3458  *      @dev: pointer to device object
3459  *
3460  *      Note: this function is invoked from the scsi mid-level.
3461  *      This function potentially frees up a device name (e.g. /dev/sdc)
3462  *      that could be re-used by a subsequent sd_probe().
3463  *      This function is not called when the built-in sd driver is "exit-ed".
3464  **/
3465 static int sd_remove(struct device *dev)
3466 {
3467         struct scsi_disk *sdkp;
3468         dev_t devt;
3469
3470         sdkp = dev_get_drvdata(dev);
3471         devt = disk_devt(sdkp->disk);
3472         scsi_autopm_get_device(sdkp->device);
3473
3474         async_synchronize_full_domain(&scsi_sd_pm_domain);
3475         async_synchronize_full_domain(&scsi_sd_probe_domain);
3476         device_del(&sdkp->dev);
3477         del_gendisk(sdkp->disk);
3478         sd_shutdown(dev);
3479
3480         free_opal_dev(sdkp->opal_dev);
3481
3482         blk_register_region(devt, SD_MINORS, NULL,
3483                             sd_default_probe, NULL, NULL);
3484
3485         mutex_lock(&sd_ref_mutex);
3486         dev_set_drvdata(dev, NULL);
3487         put_device(&sdkp->dev);
3488         mutex_unlock(&sd_ref_mutex);
3489
3490         return 0;
3491 }
3492
3493 /**
3494  *      scsi_disk_release - Called to free the scsi_disk structure
3495  *      @dev: pointer to embedded class device
3496  *
3497  *      sd_ref_mutex must be held entering this routine.  Because it is
3498  *      called on last put, you should always use the scsi_disk_get()
3499  *      scsi_disk_put() helpers which manipulate the semaphore directly
3500  *      and never do a direct put_device.
3501  **/
3502 static void scsi_disk_release(struct device *dev)
3503 {
3504         struct scsi_disk *sdkp = to_scsi_disk(dev);
3505         struct gendisk *disk = sdkp->disk;
3506         struct request_queue *q = disk->queue;
3507
3508         ida_free(&sd_index_ida, sdkp->index);
3509
3510         /*
3511          * Wait until all requests that are in progress have completed.
3512          * This is necessary to avoid that e.g. scsi_end_request() crashes
3513          * due to clearing the disk->private_data pointer. Wait from inside
3514          * scsi_disk_release() instead of from sd_release() to avoid that
3515          * freezing and unfreezing the request queue affects user space I/O
3516          * in case multiple processes open a /dev/sd... node concurrently.
3517          */
3518         blk_mq_freeze_queue(q);
3519         blk_mq_unfreeze_queue(q);
3520
3521         disk->private_data = NULL;
3522         put_disk(disk);
3523         put_device(&sdkp->device->sdev_gendev);
3524
3525         kfree(sdkp);
3526 }
3527
3528 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3529 {
3530         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3531         struct scsi_sense_hdr sshdr;
3532         struct scsi_device *sdp = sdkp->device;
3533         int res;
3534
3535         if (start)
3536                 cmd[4] |= 1;    /* START */
3537
3538         if (sdp->start_stop_pwr_cond)
3539                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3540
3541         if (!scsi_device_online(sdp))
3542                 return -ENODEV;
3543
3544         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3545                         SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3546         if (res) {
3547                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3548                 if (driver_byte(res) == DRIVER_SENSE)
3549                         sd_print_sense_hdr(sdkp, &sshdr);
3550                 if (scsi_sense_valid(&sshdr) &&
3551                         /* 0x3a is medium not present */
3552                         sshdr.asc == 0x3a)
3553                         res = 0;
3554         }
3555
3556         /* SCSI error codes must not go to the generic layer */
3557         if (res)
3558                 return -EIO;
3559
3560         return 0;
3561 }
3562
3563 /*
3564  * Send a SYNCHRONIZE CACHE instruction down to the device through
3565  * the normal SCSI command structure.  Wait for the command to
3566  * complete.
3567  */
3568 static void sd_shutdown(struct device *dev)
3569 {
3570         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3571
3572         if (!sdkp)
3573                 return;         /* this can happen */
3574
3575         if (pm_runtime_suspended(dev))
3576                 return;
3577
3578         if (sdkp->WCE && sdkp->media_present) {
3579                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3580                 sd_sync_cache(sdkp, NULL);
3581         }
3582
3583         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3584                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3585                 sd_start_stop_device(sdkp, 0);
3586         }
3587 }
3588
3589 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3590 {
3591         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3592         struct scsi_sense_hdr sshdr;
3593         int ret = 0;
3594
3595         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3596                 return 0;
3597
3598         if (sdkp->WCE && sdkp->media_present) {
3599                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3600                 ret = sd_sync_cache(sdkp, &sshdr);
3601
3602                 if (ret) {
3603                         /* ignore OFFLINE device */
3604                         if (ret == -ENODEV)
3605                                 return 0;
3606
3607                         if (!scsi_sense_valid(&sshdr) ||
3608                             sshdr.sense_key != ILLEGAL_REQUEST)
3609                                 return ret;
3610
3611                         /*
3612                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3613                          * doesn't support sync. There's not much to do and
3614                          * suspend shouldn't fail.
3615                          */
3616                         ret = 0;
3617                 }
3618         }
3619
3620         if (sdkp->device->manage_start_stop) {
3621                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3622                 /* an error is not worth aborting a system sleep */
3623                 ret = sd_start_stop_device(sdkp, 0);
3624                 if (ignore_stop_errors)
3625                         ret = 0;
3626         }
3627
3628         return ret;
3629 }
3630
3631 static int sd_suspend_system(struct device *dev)
3632 {
3633         return sd_suspend_common(dev, true);
3634 }
3635
3636 static int sd_suspend_runtime(struct device *dev)
3637 {
3638         return sd_suspend_common(dev, false);
3639 }
3640
3641 static int sd_resume(struct device *dev)
3642 {
3643         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3644         int ret;
3645
3646         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3647                 return 0;
3648
3649         if (!sdkp->device->manage_start_stop)
3650                 return 0;
3651
3652         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3653         ret = sd_start_stop_device(sdkp, 1);
3654         if (!ret)
3655                 opal_unlock_from_suspend(sdkp->opal_dev);
3656         return ret;
3657 }
3658
3659 /**
3660  *      init_sd - entry point for this driver (both when built in or when
3661  *      a module).
3662  *
3663  *      Note: this function registers this driver with the scsi mid-level.
3664  **/
3665 static int __init init_sd(void)
3666 {
3667         int majors = 0, i, err;
3668
3669         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3670
3671         for (i = 0; i < SD_MAJORS; i++) {
3672                 if (register_blkdev(sd_major(i), "sd") != 0)
3673                         continue;
3674                 majors++;
3675                 blk_register_region(sd_major(i), SD_MINORS, NULL,
3676                                     sd_default_probe, NULL, NULL);
3677         }
3678
3679         if (!majors)
3680                 return -ENODEV;
3681
3682         err = class_register(&sd_disk_class);
3683         if (err)
3684                 goto err_out;
3685
3686         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3687                                          0, 0, NULL);
3688         if (!sd_cdb_cache) {
3689                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3690                 err = -ENOMEM;
3691                 goto err_out_class;
3692         }
3693
3694         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3695         if (!sd_cdb_pool) {
3696                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3697                 err = -ENOMEM;
3698                 goto err_out_cache;
3699         }
3700
3701         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3702         if (!sd_page_pool) {
3703                 printk(KERN_ERR "sd: can't init discard page pool\n");
3704                 err = -ENOMEM;
3705                 goto err_out_ppool;
3706         }
3707
3708         err = scsi_register_driver(&sd_template.gendrv);
3709         if (err)
3710                 goto err_out_driver;
3711
3712         return 0;
3713
3714 err_out_driver:
3715         mempool_destroy(sd_page_pool);
3716
3717 err_out_ppool:
3718         mempool_destroy(sd_cdb_pool);
3719
3720 err_out_cache:
3721         kmem_cache_destroy(sd_cdb_cache);
3722
3723 err_out_class:
3724         class_unregister(&sd_disk_class);
3725 err_out:
3726         for (i = 0; i < SD_MAJORS; i++)
3727                 unregister_blkdev(sd_major(i), "sd");
3728         return err;
3729 }
3730
3731 /**
3732  *      exit_sd - exit point for this driver (when it is a module).
3733  *
3734  *      Note: this function unregisters this driver from the scsi mid-level.
3735  **/
3736 static void __exit exit_sd(void)
3737 {
3738         int i;
3739
3740         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3741
3742         scsi_unregister_driver(&sd_template.gendrv);
3743         mempool_destroy(sd_cdb_pool);
3744         mempool_destroy(sd_page_pool);
3745         kmem_cache_destroy(sd_cdb_cache);
3746
3747         class_unregister(&sd_disk_class);
3748
3749         for (i = 0; i < SD_MAJORS; i++) {
3750                 blk_unregister_region(sd_major(i), SD_MINORS);
3751                 unregister_blkdev(sd_major(i), "sd");
3752         }
3753 }
3754
3755 module_init(init_sd);
3756 module_exit(exit_sd);
3757
3758 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3759                                struct scsi_sense_hdr *sshdr)
3760 {
3761         scsi_print_sense_hdr(sdkp->device,
3762                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3763 }
3764
3765 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3766                             int result)
3767 {
3768         const char *hb_string = scsi_hostbyte_string(result);
3769         const char *db_string = scsi_driverbyte_string(result);
3770
3771         if (hb_string || db_string)
3772                 sd_printk(KERN_INFO, sdkp,
3773                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3774                           hb_string ? hb_string : "invalid",
3775                           db_string ? db_string : "invalid");
3776         else
3777                 sd_printk(KERN_INFO, sdkp,
3778                           "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3779                           msg, host_byte(result), driver_byte(result));
3780 }
3781