drivers/scsi/sd.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *      sd.c Copyright (C) 1992 Drew Eckhardt
   4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
   5  *
   6  *      Linux scsi disk driver
   7  *              Initial versions: Drew Eckhardt
   8  *              Subsequent revisions: Eric Youngdale
   9  *      Modification history:
  10  *       - Drew Eckhardt <drew@colorado.edu> original
  11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
  12  *         outstanding request, and other enhancements.
  13  *         Support loadable low-level scsi drivers.
  14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
  15  *         eight major numbers.
  16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
  17  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
  18  *         sd_init and cleanups.
  19  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
  20  *         not being read in sd_open. Fix problem where removable media
  21  *         could be ejected after sd_open.
  22  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
  23  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
  24  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
  25  *         Support 32k/1M disks.
  26  *
  27  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
  28  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
  29  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
  30  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
  31  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
  32  *      Note: when the logging level is set by the user, it must be greater
  33  *      than the level indicated above to trigger output.
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/fs.h>
  38 #include <linux/kernel.h>
  39 #include <linux/mm.h>
  40 #include <linux/bio.h>
  41 #include <linux/genhd.h>
  42 #include <linux/hdreg.h>
  43 #include <linux/errno.h>
  44 #include <linux/idr.h>
  45 #include <linux/interrupt.h>
  46 #include <linux/init.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/blkpg.h>
  49 #include <linux/blk-pm.h>
  50 #include <linux/delay.h>
  51 #include <linux/mutex.h>
  52 #include <linux/string_helpers.h>
  53 #include <linux/async.h>
  54 #include <linux/slab.h>
  55 #include <linux/sed-opal.h>
  56 #include <linux/pm_runtime.h>
  57 #include <linux/pr.h>
  58 #include <linux/t10-pi.h>
  59 #include <linux/uaccess.h>
  60 #include <asm/unaligned.h>
  61
  62 #include <scsi/scsi.h>
  63 #include <scsi/scsi_cmnd.h>
  64 #include <scsi/scsi_dbg.h>
  65 #include <scsi/scsi_device.h>
  66 #include <scsi/scsi_driver.h>
  67 #include <scsi/scsi_eh.h>
  68 #include <scsi/scsi_host.h>
  69 #include <scsi/scsi_ioctl.h>
  70 #include <scsi/scsicam.h>
  71
  72 #include "sd.h"
  73 #include "scsi_priv.h"
  74 #include "scsi_logging.h"
  75
  76 MODULE_AUTHOR("Eric Youngdale");
  77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
  78 MODULE_LICENSE("GPL");
  79
  80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
  81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
  82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
  83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
  84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
  85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
  86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
  87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
  88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
  89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
  90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
  91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
  92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
  93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
  94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
  95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
  96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
  97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
  98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
  99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
 100
 101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
 102 #define SD_MINORS       16
 103 #else
 104 #define SD_MINORS       0
 105 #endif
 106
 107 static void sd_config_discard(struct scsi_disk *, unsigned int);
 108 static void sd_config_write_same(struct scsi_disk *);
 109 static int  sd_revalidate_disk(struct gendisk *);
 110 static void sd_unlock_native_capacity(struct gendisk *disk);
 111 static int  sd_probe(struct device *);
 112 static int  sd_remove(struct device *);
 113 static void sd_shutdown(struct device *);
 114 static int sd_suspend_system(struct device *);
 115 static int sd_suspend_runtime(struct device *);
 116 static int sd_resume(struct device *);
 117 static void sd_rescan(struct device *);
 118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
 119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
 120 static int sd_done(struct scsi_cmnd *);
 121 static void sd_eh_reset(struct scsi_cmnd *);
 122 static int sd_eh_action(struct scsi_cmnd *, int);
 123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 124 static void scsi_disk_release(struct device *cdev);
 125 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
 126 static void sd_print_result(const struct scsi_disk *, const char *, int);
 127
 128 static DEFINE_IDA(sd_index_ida);
 129
 130 /* This semaphore is used to mediate the 0->1 reference get in the
 131  * face of object destruction (i.e. we can't allow a get on an
 132  * object after last put) */
 133 static DEFINE_MUTEX(sd_ref_mutex);
 134
 135 static struct kmem_cache *sd_cdb_cache;
 136 static mempool_t *sd_cdb_pool;
 137 static mempool_t *sd_page_pool;
 138
 139 static const char *sd_cache_types[] = {
 140         "write through", "none", "write back",
 141         "write back, no read (daft)"
 142 };
 143
 144 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 145 {
 146         bool wc = false, fua = false;
 147
 148         if (sdkp->WCE) {
 149                 wc = true;
 150                 if (sdkp->DPOFUA)
 151                         fua = true;
 152         }
 153
 154         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 155 }
 156
 157 static ssize_t
 158 cache_type_store(struct device *dev, struct device_attribute *attr,
 159                  const char *buf, size_t count)
 160 {
 161         int ct, rcd, wce, sp;
 162         struct scsi_disk *sdkp = to_scsi_disk(dev);
 163         struct scsi_device *sdp = sdkp->device;
 164         char buffer[64];
 165         char *buffer_data;
 166         struct scsi_mode_data data;
 167         struct scsi_sense_hdr sshdr;
 168         static const char temp[] = "temporary ";
 169         int len;
 170
 171         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 172                 /* no cache control on RBC devices; theoretically they
 173                  * can do it, but there's probably so many exceptions
 174                  * it's not worth the risk */
 175                 return -EINVAL;
 176
 177         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 178                 buf += sizeof(temp) - 1;
 179                 sdkp->cache_override = 1;
 180         } else {
 181                 sdkp->cache_override = 0;
 182         }
 183
 184         ct = sysfs_match_string(sd_cache_types, buf);
 185         if (ct < 0)
 186                 return -EINVAL;
 187
 188         rcd = ct & 0x01 ? 1 : 0;
 189         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 190
 191         if (sdkp->cache_override) {
 192                 sdkp->WCE = wce;
 193                 sdkp->RCD = rcd;
 194                 sd_set_flush_flag(sdkp);
 195                 return count;
 196         }
 197
 198         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 199                             SD_MAX_RETRIES, &data, NULL))
 200                 return -EINVAL;
 201         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 202                   data.block_descriptor_length);
 203         buffer_data = buffer + data.header_length +
 204                 data.block_descriptor_length;
 205         buffer_data[2] &= ~0x05;
 206         buffer_data[2] |= wce << 2 | rcd;
 207         sp = buffer_data[0] & 0x80 ? 1 : 0;
 208         buffer_data[0] &= ~0x80;
 209
 210         /*
 211          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 212          * received mode parameter buffer before doing MODE SELECT.
 213          */
 214         data.device_specific = 0;
 215
 216         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
 217                              SD_MAX_RETRIES, &data, &sshdr)) {
 218                 if (scsi_sense_valid(&sshdr))
 219                         sd_print_sense_hdr(sdkp, &sshdr);
 220                 return -EINVAL;
 221         }
 222         revalidate_disk(sdkp->disk);
 223         return count;
 224 }
 225
 226 static ssize_t
 227 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
 228                        char *buf)
 229 {
 230         struct scsi_disk *sdkp = to_scsi_disk(dev);
 231         struct scsi_device *sdp = sdkp->device;
 232
 233         return sprintf(buf, "%u\n", sdp->manage_start_stop);
 234 }
 235
 236 static ssize_t
 237 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
 238                         const char *buf, size_t count)
 239 {
 240         struct scsi_disk *sdkp = to_scsi_disk(dev);
 241         struct scsi_device *sdp = sdkp->device;
 242         bool v;
 243
 244         if (!capable(CAP_SYS_ADMIN))
 245                 return -EACCES;
 246
 247         if (kstrtobool(buf, &v))
 248                 return -EINVAL;
 249
 250         sdp->manage_start_stop = v;
 251
 252         return count;
 253 }
 254 static DEVICE_ATTR_RW(manage_start_stop);
 255
 256 static ssize_t
 257 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 258 {
 259         struct scsi_disk *sdkp = to_scsi_disk(dev);
 260
 261         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 262 }
 263
 264 static ssize_t
 265 allow_restart_store(struct device *dev, struct device_attribute *attr,
 266                     const char *buf, size_t count)
 267 {
 268         bool v;
 269         struct scsi_disk *sdkp = to_scsi_disk(dev);
 270         struct scsi_device *sdp = sdkp->device;
 271
 272         if (!capable(CAP_SYS_ADMIN))
 273                 return -EACCES;
 274
 275         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 276                 return -EINVAL;
 277
 278         if (kstrtobool(buf, &v))
 279                 return -EINVAL;
 280
 281         sdp->allow_restart = v;
 282
 283         return count;
 284 }
 285 static DEVICE_ATTR_RW(allow_restart);
 286
 287 static ssize_t
 288 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 289 {
 290         struct scsi_disk *sdkp = to_scsi_disk(dev);
 291         int ct = sdkp->RCD + 2*sdkp->WCE;
 292
 293         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 294 }
 295 static DEVICE_ATTR_RW(cache_type);
 296
 297 static ssize_t
 298 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 299 {
 300         struct scsi_disk *sdkp = to_scsi_disk(dev);
 301
 302         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 303 }
 304 static DEVICE_ATTR_RO(FUA);
 305
 306 static ssize_t
 307 protection_type_show(struct device *dev, struct device_attribute *attr,
 308                      char *buf)
 309 {
 310         struct scsi_disk *sdkp = to_scsi_disk(dev);
 311
 312         return sprintf(buf, "%u\n", sdkp->protection_type);
 313 }
 314
 315 static ssize_t
 316 protection_type_store(struct device *dev, struct device_attribute *attr,
 317                       const char *buf, size_t count)
 318 {
 319         struct scsi_disk *sdkp = to_scsi_disk(dev);
 320         unsigned int val;
 321         int err;
 322
 323         if (!capable(CAP_SYS_ADMIN))
 324                 return -EACCES;
 325
 326         err = kstrtouint(buf, 10, &val);
 327
 328         if (err)
 329                 return err;
 330
 331         if (val <= T10_PI_TYPE3_PROTECTION)
 332                 sdkp->protection_type = val;
 333
 334         return count;
 335 }
 336 static DEVICE_ATTR_RW(protection_type);
 337
 338 static ssize_t
 339 protection_mode_show(struct device *dev, struct device_attribute *attr,
 340                      char *buf)
 341 {
 342         struct scsi_disk *sdkp = to_scsi_disk(dev);
 343         struct scsi_device *sdp = sdkp->device;
 344         unsigned int dif, dix;
 345
 346         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 347         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 348
 349         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 350                 dif = 0;
 351                 dix = 1;
 352         }
 353
 354         if (!dif && !dix)
 355                 return sprintf(buf, "none\n");
 356
 357         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 358 }
 359 static DEVICE_ATTR_RO(protection_mode);
 360
 361 static ssize_t
 362 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 363 {
 364         struct scsi_disk *sdkp = to_scsi_disk(dev);
 365
 366         return sprintf(buf, "%u\n", sdkp->ATO);
 367 }
 368 static DEVICE_ATTR_RO(app_tag_own);
 369
 370 static ssize_t
 371 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 372                        char *buf)
 373 {
 374         struct scsi_disk *sdkp = to_scsi_disk(dev);
 375
 376         return sprintf(buf, "%u\n", sdkp->lbpme);
 377 }
 378 static DEVICE_ATTR_RO(thin_provisioning);
 379
 380 /* sysfs_match_string() requires dense arrays */
 381 static const char *lbp_mode[] = {
 382         [SD_LBP_FULL]           = "full",
 383         [SD_LBP_UNMAP]          = "unmap",
 384         [SD_LBP_WS16]           = "writesame_16",
 385         [SD_LBP_WS10]           = "writesame_10",
 386         [SD_LBP_ZERO]           = "writesame_zero",
 387         [SD_LBP_DISABLE]        = "disabled",
 388 };
 389
 390 static ssize_t
 391 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 392                        char *buf)
 393 {
 394         struct scsi_disk *sdkp = to_scsi_disk(dev);
 395
 396         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 397 }
 398
 399 static ssize_t
 400 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 401                         const char *buf, size_t count)
 402 {
 403         struct scsi_disk *sdkp = to_scsi_disk(dev);
 404         struct scsi_device *sdp = sdkp->device;
 405         int mode;
 406
 407         if (!capable(CAP_SYS_ADMIN))
 408                 return -EACCES;
 409
 410         if (sd_is_zoned(sdkp)) {
 411                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 412                 return count;
 413         }
 414
 415         if (sdp->type != TYPE_DISK)
 416                 return -EINVAL;
 417
 418         mode = sysfs_match_string(lbp_mode, buf);
 419         if (mode < 0)
 420                 return -EINVAL;
 421
 422         sd_config_discard(sdkp, mode);
 423
 424         return count;
 425 }
 426 static DEVICE_ATTR_RW(provisioning_mode);
 427
 428 /* sysfs_match_string() requires dense arrays */
 429 static const char *zeroing_mode[] = {
 430         [SD_ZERO_WRITE]         = "write",
 431         [SD_ZERO_WS]            = "writesame",
 432         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 433         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 434 };
 435
 436 static ssize_t
 437 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 438                   char *buf)
 439 {
 440         struct scsi_disk *sdkp = to_scsi_disk(dev);
 441
 442         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 443 }
 444
 445 static ssize_t
 446 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 447                    const char *buf, size_t count)
 448 {
 449         struct scsi_disk *sdkp = to_scsi_disk(dev);
 450         int mode;
 451
 452         if (!capable(CAP_SYS_ADMIN))
 453                 return -EACCES;
 454
 455         mode = sysfs_match_string(zeroing_mode, buf);
 456         if (mode < 0)
 457                 return -EINVAL;
 458
 459         sdkp->zeroing_mode = mode;
 460
 461         return count;
 462 }
 463 static DEVICE_ATTR_RW(zeroing_mode);
 464
 465 static ssize_t
 466 max_medium_access_timeouts_show(struct device *dev,
 467                                 struct device_attribute *attr, char *buf)
 468 {
 469         struct scsi_disk *sdkp = to_scsi_disk(dev);
 470
 471         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 472 }
 473
 474 static ssize_t
 475 max_medium_access_timeouts_store(struct device *dev,
 476                                  struct device_attribute *attr, const char *buf,
 477                                  size_t count)
 478 {
 479         struct scsi_disk *sdkp = to_scsi_disk(dev);
 480         int err;
 481
 482         if (!capable(CAP_SYS_ADMIN))
 483                 return -EACCES;
 484
 485         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 486
 487         return err ? err : count;
 488 }
 489 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 490
 491 static ssize_t
 492 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 493                            char *buf)
 494 {
 495         struct scsi_disk *sdkp = to_scsi_disk(dev);
 496
 497         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 498 }
 499
 500 static ssize_t
 501 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 502                             const char *buf, size_t count)
 503 {
 504         struct scsi_disk *sdkp = to_scsi_disk(dev);
 505         struct scsi_device *sdp = sdkp->device;
 506         unsigned long max;
 507         int err;
 508
 509         if (!capable(CAP_SYS_ADMIN))
 510                 return -EACCES;
 511
 512         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 513                 return -EINVAL;
 514
 515         err = kstrtoul(buf, 10, &max);
 516
 517         if (err)
 518                 return err;
 519
 520         if (max == 0)
 521                 sdp->no_write_same = 1;
 522         else if (max <= SD_MAX_WS16_BLOCKS) {
 523                 sdp->no_write_same = 0;
 524                 sdkp->max_ws_blocks = max;
 525         }
 526
 527         sd_config_write_same(sdkp);
 528
 529         return count;
 530 }
 531 static DEVICE_ATTR_RW(max_write_same_blocks);
 532
 533 static struct attribute *sd_disk_attrs[] = {
 534         &dev_attr_cache_type.attr,
 535         &dev_attr_FUA.attr,
 536         &dev_attr_allow_restart.attr,
 537         &dev_attr_manage_start_stop.attr,
 538         &dev_attr_protection_type.attr,
 539         &dev_attr_protection_mode.attr,
 540         &dev_attr_app_tag_own.attr,
 541         &dev_attr_thin_provisioning.attr,
 542         &dev_attr_provisioning_mode.attr,
 543         &dev_attr_zeroing_mode.attr,
 544         &dev_attr_max_write_same_blocks.attr,
 545         &dev_attr_max_medium_access_timeouts.attr,
 546         NULL,
 547 };
 548 ATTRIBUTE_GROUPS(sd_disk);
 549
 550 static struct class sd_disk_class = {
 551         .name           = "scsi_disk",
 552         .owner          = THIS_MODULE,
 553         .dev_release    = scsi_disk_release,
 554         .dev_groups     = sd_disk_groups,
 555 };
 556
 557 static const struct dev_pm_ops sd_pm_ops = {
 558         .suspend                = sd_suspend_system,
 559         .resume                 = sd_resume,
 560         .poweroff               = sd_suspend_system,
 561         .restore                = sd_resume,
 562         .runtime_suspend        = sd_suspend_runtime,
 563         .runtime_resume         = sd_resume,
 564 };
 565
 566 static struct scsi_driver sd_template = {
 567         .gendrv = {
 568                 .name           = "sd",
 569                 .owner          = THIS_MODULE,
 570                 .probe          = sd_probe,
 571                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
 572                 .remove         = sd_remove,
 573                 .shutdown       = sd_shutdown,
 574                 .pm             = &sd_pm_ops,
 575         },
 576         .rescan                 = sd_rescan,
 577         .init_command           = sd_init_command,
 578         .uninit_command         = sd_uninit_command,
 579         .done                   = sd_done,
 580         .eh_action              = sd_eh_action,
 581         .eh_reset               = sd_eh_reset,
 582 };
 583
 584 /*
 585  * Dummy kobj_map->probe function.
 586  * The default ->probe function will call modprobe, which is
 587  * pointless as this module is already loaded.
 588  */
 589 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
 590 {
 591         return NULL;
 592 }
 593
 594 /*
 595  * Device no to disk mapping:
 596  *
 597  *       major         disc2     disc  p1
 598  *   |............|.............|....|....| <- dev_t
 599  *    31        20 19          8 7  4 3  0
 600  *
 601  * Inside a major, we have 16k disks, however mapped non-
 602  * contiguously. The first 16 disks are for major0, the next
 603  * ones with major1, ... Disk 256 is for major0 again, disk 272
 604  * for major1, ...
 605  * As we stay compatible with our numbering scheme, we can reuse
 606  * the well-know SCSI majors 8, 65--71, 136--143.
 607  */
 608 static int sd_major(int major_idx)
 609 {
 610         switch (major_idx) {
 611         case 0:
 612                 return SCSI_DISK0_MAJOR;
 613         case 1 ... 7:
 614                 return SCSI_DISK1_MAJOR + major_idx - 1;
 615         case 8 ... 15:
 616                 return SCSI_DISK8_MAJOR + major_idx - 8;
 617         default:
 618                 BUG();
 619                 return 0;       /* shut up gcc */
 620         }
 621 }
 622
 623 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 624 {
 625         struct scsi_disk *sdkp = NULL;
 626
 627         mutex_lock(&sd_ref_mutex);
 628
 629         if (disk->private_data) {
 630                 sdkp = scsi_disk(disk);
 631                 if (scsi_device_get(sdkp->device) == 0)
 632                         get_device(&sdkp->dev);
 633                 else
 634                         sdkp = NULL;
 635         }
 636         mutex_unlock(&sd_ref_mutex);
 637         return sdkp;
 638 }
 639
 640 static void scsi_disk_put(struct scsi_disk *sdkp)
 641 {
 642         struct scsi_device *sdev = sdkp->device;
 643
 644         mutex_lock(&sd_ref_mutex);
 645         put_device(&sdkp->dev);
 646         scsi_device_put(sdev);
 647         mutex_unlock(&sd_ref_mutex);
 648 }
 649
 650 #ifdef CONFIG_BLK_SED_OPAL
 651 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 652                 size_t len, bool send)
 653 {
 654         struct scsi_device *sdev = data;
 655         u8 cdb[12] = { 0, };
 656         int ret;
 657
 658         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 659         cdb[1] = secp;
 660         put_unaligned_be16(spsp, &cdb[2]);
 661         put_unaligned_be32(len, &cdb[6]);
 662
 663         ret = scsi_execute_req(sdev, cdb,
 664                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 665                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
 666         return ret <= 0 ? ret : -EIO;
 667 }
 668 #endif /* CONFIG_BLK_SED_OPAL */
 669
 670 /*
 671  * Look up the DIX operation based on whether the command is read or
 672  * write and whether dix and dif are enabled.
 673  */
 674 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 675 {
 676         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 677         static const unsigned int ops[] = {     /* wrt dix dif */
 678                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 679                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 680                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 681                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 682                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 683                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 684                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 685                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 686         };
 687
 688         return ops[write << 2 | dix << 1 | dif];
 689 }
 690
 691 /*
 692  * Returns a mask of the protection flags that are valid for a given DIX
 693  * operation.
 694  */
 695 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 696 {
 697         static const unsigned int flag_mask[] = {
 698                 [SCSI_PROT_NORMAL]              = 0,
 699
 700                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 701                                                   SCSI_PROT_GUARD_CHECK |
 702                                                   SCSI_PROT_REF_CHECK |
 703                                                   SCSI_PROT_REF_INCREMENT,
 704
 705                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 706                                                   SCSI_PROT_IP_CHECKSUM,
 707
 708                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 709                                                   SCSI_PROT_GUARD_CHECK |
 710                                                   SCSI_PROT_REF_CHECK |
 711                                                   SCSI_PROT_REF_INCREMENT |
 712                                                   SCSI_PROT_IP_CHECKSUM,
 713
 714                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 715                                                   SCSI_PROT_REF_INCREMENT,
 716
 717                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 718                                                   SCSI_PROT_REF_CHECK |
 719                                                   SCSI_PROT_REF_INCREMENT |
 720                                                   SCSI_PROT_IP_CHECKSUM,
 721
 722                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 723                                                   SCSI_PROT_GUARD_CHECK |
 724                                                   SCSI_PROT_REF_CHECK |
 725                                                   SCSI_PROT_REF_INCREMENT |
 726                                                   SCSI_PROT_IP_CHECKSUM,
 727         };
 728
 729         return flag_mask[prot_op];
 730 }
 731
 732 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 733                                            unsigned int dix, unsigned int dif)
 734 {
 735         struct bio *bio = scmd->request->bio;
 736         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
 737         unsigned int protect = 0;
 738
 739         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 740                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 741                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 742
 743                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 744                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 745         }
 746
 747         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 748                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 749
 750                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 751                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 752         }
 753
 754         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 755                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 756
 757                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 758                         protect = 3 << 5;       /* Disable target PI checking */
 759                 else
 760                         protect = 1 << 5;       /* Enable target PI checking */
 761         }
 762
 763         scsi_set_prot_op(scmd, prot_op);
 764         scsi_set_prot_type(scmd, dif);
 765         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 766
 767         return protect;
 768 }
 769
 770 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 771 {
 772         struct request_queue *q = sdkp->disk->queue;
 773         unsigned int logical_block_size = sdkp->device->sector_size;
 774         unsigned int max_blocks = 0;
 775
 776         q->limits.discard_alignment =
 777                 sdkp->unmap_alignment * logical_block_size;
 778         q->limits.discard_granularity =
 779                 max(sdkp->physical_block_size,
 780                     sdkp->unmap_granularity * logical_block_size);
 781         sdkp->provisioning_mode = mode;
 782
 783         switch (mode) {
 784
 785         case SD_LBP_FULL:
 786         case SD_LBP_DISABLE:
 787                 blk_queue_max_discard_sectors(q, 0);
 788                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 789                 return;
 790
 791         case SD_LBP_UNMAP:
 792                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 793                                           (u32)SD_MAX_WS16_BLOCKS);
 794                 break;
 795
 796         case SD_LBP_WS16:
 797                 if (sdkp->device->unmap_limit_for_ws)
 798                         max_blocks = sdkp->max_unmap_blocks;
 799                 else
 800                         max_blocks = sdkp->max_ws_blocks;
 801
 802                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 803                 break;
 804
 805         case SD_LBP_WS10:
 806                 if (sdkp->device->unmap_limit_for_ws)
 807                         max_blocks = sdkp->max_unmap_blocks;
 808                 else
 809                         max_blocks = sdkp->max_ws_blocks;
 810
 811                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 812                 break;
 813
 814         case SD_LBP_ZERO:
 815                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 816                                           (u32)SD_MAX_WS10_BLOCKS);
 817                 break;
 818         }
 819
 820         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 821         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 822 }
 823
 824 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 825 {
 826         struct scsi_device *sdp = cmd->device;
 827         struct request *rq = cmd->request;
 828         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 829         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 830         unsigned int data_len = 24;
 831         char *buf;
 832
 833         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 834         if (!rq->special_vec.bv_page)
 835                 return BLK_STS_RESOURCE;
 836         clear_highpage(rq->special_vec.bv_page);
 837         rq->special_vec.bv_offset = 0;
 838         rq->special_vec.bv_len = data_len;
 839         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 840
 841         cmd->cmd_len = 10;
 842         cmd->cmnd[0] = UNMAP;
 843         cmd->cmnd[8] = 24;
 844
 845         buf = page_address(rq->special_vec.bv_page);
 846         put_unaligned_be16(6 + 16, &buf[0]);
 847         put_unaligned_be16(16, &buf[2]);
 848         put_unaligned_be64(lba, &buf[8]);
 849         put_unaligned_be32(nr_blocks, &buf[16]);
 850
 851         cmd->allowed = SD_MAX_RETRIES;
 852         cmd->transfersize = data_len;
 853         rq->timeout = SD_TIMEOUT;
 854
 855         return scsi_init_io(cmd);
 856 }
 857
 858 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 859                 bool unmap)
 860 {
 861         struct scsi_device *sdp = cmd->device;
 862         struct request *rq = cmd->request;
 863         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 864         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 865         u32 data_len = sdp->sector_size;
 866
 867         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 868         if (!rq->special_vec.bv_page)
 869                 return BLK_STS_RESOURCE;
 870         clear_highpage(rq->special_vec.bv_page);
 871         rq->special_vec.bv_offset = 0;
 872         rq->special_vec.bv_len = data_len;
 873         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 874
 875         cmd->cmd_len = 16;
 876         cmd->cmnd[0] = WRITE_SAME_16;
 877         if (unmap)
 878                 cmd->cmnd[1] = 0x8; /* UNMAP */
 879         put_unaligned_be64(lba, &cmd->cmnd[2]);
 880         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 881
 882         cmd->allowed = SD_MAX_RETRIES;
 883         cmd->transfersize = data_len;
 884         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 885
 886         return scsi_init_io(cmd);
 887 }
 888
 889 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 890                 bool unmap)
 891 {
 892         struct scsi_device *sdp = cmd->device;
 893         struct request *rq = cmd->request;
 894         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 895         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 896         u32 data_len = sdp->sector_size;
 897
 898         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 899         if (!rq->special_vec.bv_page)
 900                 return BLK_STS_RESOURCE;
 901         clear_highpage(rq->special_vec.bv_page);
 902         rq->special_vec.bv_offset = 0;
 903         rq->special_vec.bv_len = data_len;
 904         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 905
 906         cmd->cmd_len = 10;
 907         cmd->cmnd[0] = WRITE_SAME;
 908         if (unmap)
 909                 cmd->cmnd[1] = 0x8; /* UNMAP */
 910         put_unaligned_be32(lba, &cmd->cmnd[2]);
 911         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 912
 913         cmd->allowed = SD_MAX_RETRIES;
 914         cmd->transfersize = data_len;
 915         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 916
 917         return scsi_init_io(cmd);
 918 }
 919
 920 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 921 {
 922         struct request *rq = cmd->request;
 923         struct scsi_device *sdp = cmd->device;
 924         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 925         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 926         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 927
 928         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 929                 switch (sdkp->zeroing_mode) {
 930                 case SD_ZERO_WS16_UNMAP:
 931                         return sd_setup_write_same16_cmnd(cmd, true);
 932                 case SD_ZERO_WS10_UNMAP:
 933                         return sd_setup_write_same10_cmnd(cmd, true);
 934                 }
 935         }
 936
 937         if (sdp->no_write_same)
 938                 return BLK_STS_TARGET;
 939
 940         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 941                 return sd_setup_write_same16_cmnd(cmd, false);
 942
 943         return sd_setup_write_same10_cmnd(cmd, false);
 944 }
 945
 946 static void sd_config_write_same(struct scsi_disk *sdkp)
 947 {
 948         struct request_queue *q = sdkp->disk->queue;
 949         unsigned int logical_block_size = sdkp->device->sector_size;
 950
 951         if (sdkp->device->no_write_same) {
 952                 sdkp->max_ws_blocks = 0;
 953                 goto out;
 954         }
 955
 956         /* Some devices can not handle block counts above 0xffff despite
 957          * supporting WRITE SAME(16). Consequently we default to 64k
 958          * blocks per I/O unless the device explicitly advertises a
 959          * bigger limit.
 960          */
 961         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
 962                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 963                                                    (u32)SD_MAX_WS16_BLOCKS);
 964         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
 965                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 966                                                    (u32)SD_MAX_WS10_BLOCKS);
 967         else {
 968                 sdkp->device->no_write_same = 1;
 969                 sdkp->max_ws_blocks = 0;
 970         }
 971
 972         if (sdkp->lbprz && sdkp->lbpws)
 973                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
 974         else if (sdkp->lbprz && sdkp->lbpws10)
 975                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
 976         else if (sdkp->max_ws_blocks)
 977                 sdkp->zeroing_mode = SD_ZERO_WS;
 978         else
 979                 sdkp->zeroing_mode = SD_ZERO_WRITE;
 980
 981         if (sdkp->max_ws_blocks &&
 982             sdkp->physical_block_size > logical_block_size) {
 983                 /*
 984                  * Reporting a maximum number of blocks that is not aligned
 985                  * on the device physical size would cause a large write same
 986                  * request to be split into physically unaligned chunks by
 987                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
 988                  * even if the caller of these functions took care to align the
 989                  * large request. So make sure the maximum reported is aligned
 990                  * to the device physical block size. This is only an optional
 991                  * optimization for regular disks, but this is mandatory to
 992                  * avoid failure of large write same requests directed at
 993                  * sequential write required zones of host-managed ZBC disks.
 994                  */
 995                 sdkp->max_ws_blocks =
 996                         round_down(sdkp->max_ws_blocks,
 997                                    bytes_to_logical(sdkp->device,
 998                                                     sdkp->physical_block_size));
 999         }
1000
1001 out:
1002         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1003                                          (logical_block_size >> 9));
1004         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1005                                          (logical_block_size >> 9));
1006 }
1007
1008 /**
1009  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1010  * @cmd: command to prepare
1011  *
1012  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1013  * the preference indicated by the target device.
1014  **/
1015 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1016 {
1017         struct request *rq = cmd->request;
1018         struct scsi_device *sdp = cmd->device;
1019         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1020         struct bio *bio = rq->bio;
1021         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1022         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023         blk_status_t ret;
1024
1025         if (sdkp->device->no_write_same)
1026                 return BLK_STS_TARGET;
1027
1028         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1029
1030         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1031
1032         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1033                 cmd->cmd_len = 16;
1034                 cmd->cmnd[0] = WRITE_SAME_16;
1035                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1036                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037         } else {
1038                 cmd->cmd_len = 10;
1039                 cmd->cmnd[0] = WRITE_SAME;
1040                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1041                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042         }
1043
1044         cmd->transfersize = sdp->sector_size;
1045         cmd->allowed = SD_MAX_RETRIES;
1046
1047         /*
1048          * For WRITE SAME the data transferred via the DATA OUT buffer is
1049          * different from the amount of data actually written to the target.
1050          *
1051          * We set up __data_len to the amount of data transferred via the
1052          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1053          * to transfer a single sector of data first, but then reset it to
1054          * the amount of data to be written right after so that the I/O path
1055          * knows how much to actually write.
1056          */
1057         rq->__data_len = sdp->sector_size;
1058         ret = scsi_init_io(cmd);
1059         rq->__data_len = blk_rq_bytes(rq);
1060
1061         return ret;
1062 }
1063
1064 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1065 {
1066         struct request *rq = cmd->request;
1067
1068         /* flush requests don't perform I/O, zero the S/G table */
1069         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070
1071         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1072         cmd->cmd_len = 10;
1073         cmd->transfersize = 0;
1074         cmd->allowed = SD_MAX_RETRIES;
1075
1076         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1077         return BLK_STS_OK;
1078 }
1079
1080 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1081                                        sector_t lba, unsigned int nr_blocks,
1082                                        unsigned char flags)
1083 {
1084         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1085         if (unlikely(cmd->cmnd == NULL))
1086                 return BLK_STS_RESOURCE;
1087
1088         cmd->cmd_len = SD_EXT_CDB_SIZE;
1089         memset(cmd->cmnd, 0, cmd->cmd_len);
1090
1091         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1092         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1093         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1094         cmd->cmnd[10] = flags;
1095         put_unaligned_be64(lba, &cmd->cmnd[12]);
1096         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1097         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1098
1099         return BLK_STS_OK;
1100 }
1101
1102 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1103                                        sector_t lba, unsigned int nr_blocks,
1104                                        unsigned char flags)
1105 {
1106         cmd->cmd_len  = 16;
1107         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1108         cmd->cmnd[1]  = flags;
1109         cmd->cmnd[14] = 0;
1110         cmd->cmnd[15] = 0;
1111         put_unaligned_be64(lba, &cmd->cmnd[2]);
1112         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1113
1114         return BLK_STS_OK;
1115 }
1116
1117 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1118                                        sector_t lba, unsigned int nr_blocks,
1119                                        unsigned char flags)
1120 {
1121         cmd->cmd_len = 10;
1122         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1123         cmd->cmnd[1] = flags;
1124         cmd->cmnd[6] = 0;
1125         cmd->cmnd[9] = 0;
1126         put_unaligned_be32(lba, &cmd->cmnd[2]);
1127         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1128
1129         return BLK_STS_OK;
1130 }
1131
1132 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1133                                       sector_t lba, unsigned int nr_blocks,
1134                                       unsigned char flags)
1135 {
1136         /* Avoid that 0 blocks gets translated into 256 blocks. */
1137         if (WARN_ON_ONCE(nr_blocks == 0))
1138                 return BLK_STS_IOERR;
1139
1140         if (unlikely(flags & 0x8)) {
1141                 /*
1142                  * This happens only if this drive failed 10byte rw
1143                  * command with ILLEGAL_REQUEST during operation and
1144                  * thus turned off use_10_for_rw.
1145                  */
1146                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1147                 return BLK_STS_IOERR;
1148         }
1149
1150         cmd->cmd_len = 6;
1151         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1152         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1153         cmd->cmnd[2] = (lba >> 8) & 0xff;
1154         cmd->cmnd[3] = lba & 0xff;
1155         cmd->cmnd[4] = nr_blocks;
1156         cmd->cmnd[5] = 0;
1157
1158         return BLK_STS_OK;
1159 }
1160
1161 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1162 {
1163         struct request *rq = cmd->request;
1164         struct scsi_device *sdp = cmd->device;
1165         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1166         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1167         sector_t threshold;
1168         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1169         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170         bool write = rq_data_dir(rq) == WRITE;
1171         unsigned char protect, fua;
1172         blk_status_t ret;
1173         unsigned int dif;
1174         bool dix;
1175
1176         ret = scsi_init_io(cmd);
1177         if (ret != BLK_STS_OK)
1178                 return ret;
1179
1180         if (!scsi_device_online(sdp) || sdp->changed) {
1181                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1182                 return BLK_STS_IOERR;
1183         }
1184
1185         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1186                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1187                 return BLK_STS_IOERR;
1188         }
1189
1190         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1191                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1192                 return BLK_STS_IOERR;
1193         }
1194
1195         /*
1196          * Some SD card readers can't handle accesses which touch the
1197          * last one or two logical blocks. Split accesses as needed.
1198          */
1199         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1200
1201         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1202                 if (lba < threshold) {
1203                         /* Access up to the threshold but not beyond */
1204                         nr_blocks = threshold - lba;
1205                 } else {
1206                         /* Access only a single logical block */
1207                         nr_blocks = 1;
1208                 }
1209         }
1210
1211         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1212         dix = scsi_prot_sg_count(cmd);
1213         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1214
1215         if (write && dix)
1216                 t10_pi_prepare(cmd->request, sdkp->protection_type);
1217
1218         if (dif || dix)
1219                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1220         else
1221                 protect = 0;
1222
1223         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1224                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1225                                          protect | fua);
1226         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1227                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1228                                          protect | fua);
1229         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1230                    sdp->use_10_for_rw || protect) {
1231                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1232                                          protect | fua);
1233         } else {
1234                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1235                                         protect | fua);
1236         }
1237
1238         if (unlikely(ret != BLK_STS_OK))
1239                 return ret;
1240
1241         /*
1242          * We shouldn't disconnect in the middle of a sector, so with a dumb
1243          * host adapter, it's safe to assume that we can at least transfer
1244          * this many bytes between each connect / disconnect.
1245          */
1246         cmd->transfersize = sdp->sector_size;
1247         cmd->underflow = nr_blocks << 9;
1248         cmd->allowed = SD_MAX_RETRIES;
1249         cmd->sdb.length = nr_blocks * sdp->sector_size;
1250
1251         SCSI_LOG_HLQUEUE(1,
1252                          scmd_printk(KERN_INFO, cmd,
1253                                      "%s: block=%llu, count=%d\n", __func__,
1254                                      (unsigned long long)blk_rq_pos(rq),
1255                                      blk_rq_sectors(rq)));
1256         SCSI_LOG_HLQUEUE(2,
1257                          scmd_printk(KERN_INFO, cmd,
1258                                      "%s %d/%u 512 byte blocks.\n",
1259                                      write ? "writing" : "reading", nr_blocks,
1260                                      blk_rq_sectors(rq)));
1261
1262         /*
1263          * This indicates that the command is ready from our end to be
1264          * queued.
1265          */
1266         return BLK_STS_OK;
1267 }
1268
1269 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1270 {
1271         struct request *rq = cmd->request;
1272
1273         switch (req_op(rq)) {
1274         case REQ_OP_DISCARD:
1275                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1276                 case SD_LBP_UNMAP:
1277                         return sd_setup_unmap_cmnd(cmd);
1278                 case SD_LBP_WS16:
1279                         return sd_setup_write_same16_cmnd(cmd, true);
1280                 case SD_LBP_WS10:
1281                         return sd_setup_write_same10_cmnd(cmd, true);
1282                 case SD_LBP_ZERO:
1283                         return sd_setup_write_same10_cmnd(cmd, false);
1284                 default:
1285                         return BLK_STS_TARGET;
1286                 }
1287         case REQ_OP_WRITE_ZEROES:
1288                 return sd_setup_write_zeroes_cmnd(cmd);
1289         case REQ_OP_WRITE_SAME:
1290                 return sd_setup_write_same_cmnd(cmd);
1291         case REQ_OP_FLUSH:
1292                 return sd_setup_flush_cmnd(cmd);
1293         case REQ_OP_READ:
1294         case REQ_OP_WRITE:
1295                 return sd_setup_read_write_cmnd(cmd);
1296         case REQ_OP_ZONE_RESET:
1297                 return sd_zbc_setup_reset_cmnd(cmd);
1298         default:
1299                 WARN_ON_ONCE(1);
1300                 return BLK_STS_NOTSUPP;
1301         }
1302 }
1303
1304 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1305 {
1306         struct request *rq = SCpnt->request;
1307         u8 *cmnd;
1308
1309         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1310                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1311
1312         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1313                 cmnd = SCpnt->cmnd;
1314                 SCpnt->cmnd = NULL;
1315                 SCpnt->cmd_len = 0;
1316                 mempool_free(cmnd, sd_cdb_pool);
1317         }
1318 }
1319
1320 /**
1321  *      sd_open - open a scsi disk device
1322  *      @bdev: Block device of the scsi disk to open
1323  *      @mode: FMODE_* mask
1324  *
1325  *      Returns 0 if successful. Returns a negated errno value in case
1326  *      of error.
1327  *
1328  *      Note: This can be called from a user context (e.g. fsck(1) )
1329  *      or from within the kernel (e.g. as a result of a mount(1) ).
1330  *      In the latter case @inode and @filp carry an abridged amount
1331  *      of information as noted above.
1332  *
1333  *      Locking: called with bdev->bd_mutex held.
1334  **/
1335 static int sd_open(struct block_device *bdev, fmode_t mode)
1336 {
1337         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1338         struct scsi_device *sdev;
1339         int retval;
1340
1341         if (!sdkp)
1342                 return -ENXIO;
1343
1344         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1345
1346         sdev = sdkp->device;
1347
1348         /*
1349          * If the device is in error recovery, wait until it is done.
1350          * If the device is offline, then disallow any access to it.
1351          */
1352         retval = -ENXIO;
1353         if (!scsi_block_when_processing_errors(sdev))
1354                 goto error_out;
1355
1356         if (sdev->removable || sdkp->write_prot)
1357                 check_disk_change(bdev);
1358
1359         /*
1360          * If the drive is empty, just let the open fail.
1361          */
1362         retval = -ENOMEDIUM;
1363         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1364                 goto error_out;
1365
1366         /*
1367          * If the device has the write protect tab set, have the open fail
1368          * if the user expects to be able to write to the thing.
1369          */
1370         retval = -EROFS;
1371         if (sdkp->write_prot && (mode & FMODE_WRITE))
1372                 goto error_out;
1373
1374         /*
1375          * It is possible that the disk changing stuff resulted in
1376          * the device being taken offline.  If this is the case,
1377          * report this to the user, and don't pretend that the
1378          * open actually succeeded.
1379          */
1380         retval = -ENXIO;
1381         if (!scsi_device_online(sdev))
1382                 goto error_out;
1383
1384         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1385                 if (scsi_block_when_processing_errors(sdev))
1386                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1387         }
1388
1389         return 0;
1390
1391 error_out:
1392         scsi_disk_put(sdkp);
1393         return retval;
1394 }
1395
1396 /**
1397  *      sd_release - invoked when the (last) close(2) is called on this
1398  *      scsi disk.
1399  *      @disk: disk to release
1400  *      @mode: FMODE_* mask
1401  *
1402  *      Returns 0.
1403  *
1404  *      Note: may block (uninterruptible) if error recovery is underway
1405  *      on this disk.
1406  *
1407  *      Locking: called with bdev->bd_mutex held.
1408  **/
1409 static void sd_release(struct gendisk *disk, fmode_t mode)
1410 {
1411         struct scsi_disk *sdkp = scsi_disk(disk);
1412         struct scsi_device *sdev = sdkp->device;
1413
1414         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1415
1416         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1417                 if (scsi_block_when_processing_errors(sdev))
1418                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1419         }
1420
1421         scsi_disk_put(sdkp);
1422 }
1423
1424 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1425 {
1426         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1427         struct scsi_device *sdp = sdkp->device;
1428         struct Scsi_Host *host = sdp->host;
1429         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1430         int diskinfo[4];
1431
1432         /* default to most commonly used values */
1433         diskinfo[0] = 0x40;     /* 1 << 6 */
1434         diskinfo[1] = 0x20;     /* 1 << 5 */
1435         diskinfo[2] = capacity >> 11;
1436
1437         /* override with calculated, extended default, or driver values */
1438         if (host->hostt->bios_param)
1439                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1440         else
1441                 scsicam_bios_param(bdev, capacity, diskinfo);
1442
1443         geo->heads = diskinfo[0];
1444         geo->sectors = diskinfo[1];
1445         geo->cylinders = diskinfo[2];
1446         return 0;
1447 }
1448
1449 /**
1450  *      sd_ioctl - process an ioctl
1451  *      @bdev: target block device
1452  *      @mode: FMODE_* mask
1453  *      @cmd: ioctl command number
1454  *      @arg: this is third argument given to ioctl(2) system call.
1455  *      Often contains a pointer.
1456  *
1457  *      Returns 0 if successful (some ioctls return positive numbers on
1458  *      success as well). Returns a negated errno value in case of error.
1459  *
1460  *      Note: most ioctls are forward onto the block subsystem or further
1461  *      down in the scsi subsystem.
1462  **/
1463 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1464                     unsigned int cmd, unsigned long arg)
1465 {
1466         struct gendisk *disk = bdev->bd_disk;
1467         struct scsi_disk *sdkp = scsi_disk(disk);
1468         struct scsi_device *sdp = sdkp->device;
1469         void __user *p = (void __user *)arg;
1470         int error;
1471
1472         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1473                                     "cmd=0x%x\n", disk->disk_name, cmd));
1474
1475         error = scsi_verify_blk_ioctl(bdev, cmd);
1476         if (error < 0)
1477                 return error;
1478
1479         /*
1480          * If we are in the middle of error recovery, don't let anyone
1481          * else try and use this device.  Also, if error recovery fails, it
1482          * may try and take the device offline, in which case all further
1483          * access to the device is prohibited.
1484          */
1485         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1486                         (mode & FMODE_NDELAY) != 0);
1487         if (error)
1488                 goto out;
1489
1490         if (is_sed_ioctl(cmd))
1491                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1492
1493         /*
1494          * Send SCSI addressing ioctls directly to mid level, send other
1495          * ioctls to block level and then onto mid level if they can't be
1496          * resolved.
1497          */
1498         switch (cmd) {
1499                 case SCSI_IOCTL_GET_IDLUN:
1500                 case SCSI_IOCTL_GET_BUS_NUMBER:
1501                         error = scsi_ioctl(sdp, cmd, p);
1502                         break;
1503                 default:
1504                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1505                         if (error != -ENOTTY)
1506                                 break;
1507                         error = scsi_ioctl(sdp, cmd, p);
1508                         break;
1509         }
1510 out:
1511         return error;
1512 }
1513
1514 static void set_media_not_present(struct scsi_disk *sdkp)
1515 {
1516         if (sdkp->media_present)
1517                 sdkp->device->changed = 1;
1518
1519         if (sdkp->device->removable) {
1520                 sdkp->media_present = 0;
1521                 sdkp->capacity = 0;
1522         }
1523 }
1524
1525 static int media_not_present(struct scsi_disk *sdkp,
1526                              struct scsi_sense_hdr *sshdr)
1527 {
1528         if (!scsi_sense_valid(sshdr))
1529                 return 0;
1530
1531         /* not invoked for commands that could return deferred errors */
1532         switch (sshdr->sense_key) {
1533         case UNIT_ATTENTION:
1534         case NOT_READY:
1535                 /* medium not present */
1536                 if (sshdr->asc == 0x3A) {
1537                         set_media_not_present(sdkp);
1538                         return 1;
1539                 }
1540         }
1541         return 0;
1542 }
1543
1544 /**
1545  *      sd_check_events - check media events
1546  *      @disk: kernel device descriptor
1547  *      @clearing: disk events currently being cleared
1548  *
1549  *      Returns mask of DISK_EVENT_*.
1550  *
1551  *      Note: this function is invoked from the block subsystem.
1552  **/
1553 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1554 {
1555         struct scsi_disk *sdkp = scsi_disk_get(disk);
1556         struct scsi_device *sdp;
1557         int retval;
1558
1559         if (!sdkp)
1560                 return 0;
1561
1562         sdp = sdkp->device;
1563         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1564
1565         /*
1566          * If the device is offline, don't send any commands - just pretend as
1567          * if the command failed.  If the device ever comes back online, we
1568          * can deal with it then.  It is only because of unrecoverable errors
1569          * that we would ever take a device offline in the first place.
1570          */
1571         if (!scsi_device_online(sdp)) {
1572                 set_media_not_present(sdkp);
1573                 goto out;
1574         }
1575
1576         /*
1577          * Using TEST_UNIT_READY enables differentiation between drive with
1578          * no cartridge loaded - NOT READY, drive with changed cartridge -
1579          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1580          *
1581          * Drives that auto spin down. eg iomega jaz 1G, will be started
1582          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1583          * sd_revalidate() is called.
1584          */
1585         if (scsi_block_when_processing_errors(sdp)) {
1586                 struct scsi_sense_hdr sshdr = { 0, };
1587
1588                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1589                                               &sshdr);
1590
1591                 /* failed to execute TUR, assume media not present */
1592                 if (host_byte(retval)) {
1593                         set_media_not_present(sdkp);
1594                         goto out;
1595                 }
1596
1597                 if (media_not_present(sdkp, &sshdr))
1598                         goto out;
1599         }
1600
1601         /*
1602          * For removable scsi disk we have to recognise the presence
1603          * of a disk in the drive.
1604          */
1605         if (!sdkp->media_present)
1606                 sdp->changed = 1;
1607         sdkp->media_present = 1;
1608 out:
1609         /*
1610          * sdp->changed is set under the following conditions:
1611          *
1612          *      Medium present state has changed in either direction.
1613          *      Device has indicated UNIT_ATTENTION.
1614          */
1615         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1616         sdp->changed = 0;
1617         scsi_disk_put(sdkp);
1618         return retval;
1619 }
1620
1621 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1622 {
1623         int retries, res;
1624         struct scsi_device *sdp = sdkp->device;
1625         const int timeout = sdp->request_queue->rq_timeout
1626                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1627         struct scsi_sense_hdr my_sshdr;
1628
1629         if (!scsi_device_online(sdp))
1630                 return -ENODEV;
1631
1632         /* caller might not be interested in sense, but we need it */
1633         if (!sshdr)
1634                 sshdr = &my_sshdr;
1635
1636         for (retries = 3; retries > 0; --retries) {
1637                 unsigned char cmd[10] = { 0 };
1638
1639                 cmd[0] = SYNCHRONIZE_CACHE;
1640                 /*
1641                  * Leave the rest of the command zero to indicate
1642                  * flush everything.
1643                  */
1644                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1645                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1646                 if (res == 0)
1647                         break;
1648         }
1649
1650         if (res) {
1651                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1652
1653                 if (driver_byte(res) == DRIVER_SENSE)
1654                         sd_print_sense_hdr(sdkp, sshdr);
1655
1656                 /* we need to evaluate the error return  */
1657                 if (scsi_sense_valid(sshdr) &&
1658                         (sshdr->asc == 0x3a ||  /* medium not present */
1659                          sshdr->asc == 0x20 ||  /* invalid command */
1660                          (sshdr->asc == 0x74 && sshdr->ascq == 0x71)))  /* drive is password locked */
1661                                 /* this is no error here */
1662                                 return 0;
1663
1664                 switch (host_byte(res)) {
1665                 /* ignore errors due to racing a disconnection */
1666                 case DID_BAD_TARGET:
1667                 case DID_NO_CONNECT:
1668                         return 0;
1669                 /* signal the upper layer it might try again */
1670                 case DID_BUS_BUSY:
1671                 case DID_IMM_RETRY:
1672                 case DID_REQUEUE:
1673                 case DID_SOFT_ERROR:
1674                         return -EBUSY;
1675                 default:
1676                         return -EIO;
1677                 }
1678         }
1679         return 0;
1680 }
1681
1682 static void sd_rescan(struct device *dev)
1683 {
1684         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1685
1686         revalidate_disk(sdkp->disk);
1687 }
1688
1689
1690 #ifdef CONFIG_COMPAT
1691 /*
1692  * This gets directly called from VFS. When the ioctl
1693  * is not recognized we go back to the other translation paths.
1694  */
1695 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1696                            unsigned int cmd, unsigned long arg)
1697 {
1698         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1699         int error;
1700
1701         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1702                         (mode & FMODE_NDELAY) != 0);
1703         if (error)
1704                 return error;
1705
1706         /*
1707          * Let the static ioctl translation table take care of it.
1708          */
1709         if (!sdev->host->hostt->compat_ioctl)
1710                 return -ENOIOCTLCMD;
1711         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1712 }
1713 #endif
1714
1715 static char sd_pr_type(enum pr_type type)
1716 {
1717         switch (type) {
1718         case PR_WRITE_EXCLUSIVE:
1719                 return 0x01;
1720         case PR_EXCLUSIVE_ACCESS:
1721                 return 0x03;
1722         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1723                 return 0x05;
1724         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1725                 return 0x06;
1726         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1727                 return 0x07;
1728         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1729                 return 0x08;
1730         default:
1731                 return 0;
1732         }
1733 };
1734
1735 static int sd_pr_command(struct block_device *bdev, u8 sa,
1736                 u64 key, u64 sa_key, u8 type, u8 flags)
1737 {
1738         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1739         struct scsi_sense_hdr sshdr;
1740         int result;
1741         u8 cmd[16] = { 0, };
1742         u8 data[24] = { 0, };
1743
1744         cmd[0] = PERSISTENT_RESERVE_OUT;
1745         cmd[1] = sa;
1746         cmd[2] = type;
1747         put_unaligned_be32(sizeof(data), &cmd[5]);
1748
1749         put_unaligned_be64(key, &data[0]);
1750         put_unaligned_be64(sa_key, &data[8]);
1751         data[20] = flags;
1752
1753         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1754                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1755
1756         if (driver_byte(result) == DRIVER_SENSE &&
1757             scsi_sense_valid(&sshdr)) {
1758                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1759                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1760         }
1761
1762         return result;
1763 }
1764
1765 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1766                 u32 flags)
1767 {
1768         if (flags & ~PR_FL_IGNORE_KEY)
1769                 return -EOPNOTSUPP;
1770         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1771                         old_key, new_key, 0,
1772                         (1 << 0) /* APTPL */);
1773 }
1774
1775 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1776                 u32 flags)
1777 {
1778         if (flags)
1779                 return -EOPNOTSUPP;
1780         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1781 }
1782
1783 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1784 {
1785         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1786 }
1787
1788 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1789                 enum pr_type type, bool abort)
1790 {
1791         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1792                              sd_pr_type(type), 0);
1793 }
1794
1795 static int sd_pr_clear(struct block_device *bdev, u64 key)
1796 {
1797         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1798 }
1799
1800 static const struct pr_ops sd_pr_ops = {
1801         .pr_register    = sd_pr_register,
1802         .pr_reserve     = sd_pr_reserve,
1803         .pr_release     = sd_pr_release,
1804         .pr_preempt     = sd_pr_preempt,
1805         .pr_clear       = sd_pr_clear,
1806 };
1807
1808 static const struct block_device_operations sd_fops = {
1809         .owner                  = THIS_MODULE,
1810         .open                   = sd_open,
1811         .release                = sd_release,
1812         .ioctl                  = sd_ioctl,
1813         .getgeo                 = sd_getgeo,
1814 #ifdef CONFIG_COMPAT
1815         .compat_ioctl           = sd_compat_ioctl,
1816 #endif
1817         .check_events           = sd_check_events,
1818         .revalidate_disk        = sd_revalidate_disk,
1819         .unlock_native_capacity = sd_unlock_native_capacity,
1820         .report_zones           = sd_zbc_report_zones,
1821         .pr_ops                 = &sd_pr_ops,
1822 };
1823
1824 /**
1825  *      sd_eh_reset - reset error handling callback
1826  *      @scmd:          sd-issued command that has failed
1827  *
1828  *      This function is called by the SCSI midlayer before starting
1829  *      SCSI EH. When counting medium access failures we have to be
1830  *      careful to register it only only once per device and SCSI EH run;
1831  *      there might be several timed out commands which will cause the
1832  *      'max_medium_access_timeouts' counter to trigger after the first
1833  *      SCSI EH run already and set the device to offline.
1834  *      So this function resets the internal counter before starting SCSI EH.
1835  **/
1836 static void sd_eh_reset(struct scsi_cmnd *scmd)
1837 {
1838         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1839
1840         /* New SCSI EH run, reset gate variable */
1841         sdkp->ignore_medium_access_errors = false;
1842 }
1843
1844 /**
1845  *      sd_eh_action - error handling callback
1846  *      @scmd:          sd-issued command that has failed
1847  *      @eh_disp:       The recovery disposition suggested by the midlayer
1848  *
1849  *      This function is called by the SCSI midlayer upon completion of an
1850  *      error test command (currently TEST UNIT READY). The result of sending
1851  *      the eh command is passed in eh_disp.  We're looking for devices that
1852  *      fail medium access commands but are OK with non access commands like
1853  *      test unit ready (so wrongly see the device as having a successful
1854  *      recovery)
1855  **/
1856 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1857 {
1858         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1859         struct scsi_device *sdev = scmd->device;
1860
1861         if (!scsi_device_online(sdev) ||
1862             !scsi_medium_access_command(scmd) ||
1863             host_byte(scmd->result) != DID_TIME_OUT ||
1864             eh_disp != SUCCESS)
1865                 return eh_disp;
1866
1867         /*
1868          * The device has timed out executing a medium access command.
1869          * However, the TEST UNIT READY command sent during error
1870          * handling completed successfully. Either the device is in the
1871          * process of recovering or has it suffered an internal failure
1872          * that prevents access to the storage medium.
1873          */
1874         if (!sdkp->ignore_medium_access_errors) {
1875                 sdkp->medium_access_timed_out++;
1876                 sdkp->ignore_medium_access_errors = true;
1877         }
1878
1879         /*
1880          * If the device keeps failing read/write commands but TEST UNIT
1881          * READY always completes successfully we assume that medium
1882          * access is no longer possible and take the device offline.
1883          */
1884         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1885                 scmd_printk(KERN_ERR, scmd,
1886                             "Medium access timeout failure. Offlining disk!\n");
1887                 mutex_lock(&sdev->state_mutex);
1888                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1889                 mutex_unlock(&sdev->state_mutex);
1890
1891                 return SUCCESS;
1892         }
1893
1894         return eh_disp;
1895 }
1896
1897 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1898 {
1899         struct request *req = scmd->request;
1900         struct scsi_device *sdev = scmd->device;
1901         unsigned int transferred, good_bytes;
1902         u64 start_lba, end_lba, bad_lba;
1903
1904         /*
1905          * Some commands have a payload smaller than the device logical
1906          * block size (e.g. INQUIRY on a 4K disk).
1907          */
1908         if (scsi_bufflen(scmd) <= sdev->sector_size)
1909                 return 0;
1910
1911         /* Check if we have a 'bad_lba' information */
1912         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1913                                      SCSI_SENSE_BUFFERSIZE,
1914                                      &bad_lba))
1915                 return 0;
1916
1917         /*
1918          * If the bad lba was reported incorrectly, we have no idea where
1919          * the error is.
1920          */
1921         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1922         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1923         if (bad_lba < start_lba || bad_lba >= end_lba)
1924                 return 0;
1925
1926         /*
1927          * resid is optional but mostly filled in.  When it's unused,
1928          * its value is zero, so we assume the whole buffer transferred
1929          */
1930         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1931
1932         /* This computation should always be done in terms of the
1933          * resolution of the device's medium.
1934          */
1935         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1936
1937         return min(good_bytes, transferred);
1938 }
1939
1940 /**
1941  *      sd_done - bottom half handler: called when the lower level
1942  *      driver has completed (successfully or otherwise) a scsi command.
1943  *      @SCpnt: mid-level's per command structure.
1944  *
1945  *      Note: potentially run from within an ISR. Must not block.
1946  **/
1947 static int sd_done(struct scsi_cmnd *SCpnt)
1948 {
1949         int result = SCpnt->result;
1950         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1951         unsigned int sector_size = SCpnt->device->sector_size;
1952         unsigned int resid;
1953         struct scsi_sense_hdr sshdr;
1954         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1955         struct request *req = SCpnt->request;
1956         int sense_valid = 0;
1957         int sense_deferred = 0;
1958
1959         switch (req_op(req)) {
1960         case REQ_OP_DISCARD:
1961         case REQ_OP_WRITE_ZEROES:
1962         case REQ_OP_WRITE_SAME:
1963         case REQ_OP_ZONE_RESET:
1964                 if (!result) {
1965                         good_bytes = blk_rq_bytes(req);
1966                         scsi_set_resid(SCpnt, 0);
1967                 } else {
1968                         good_bytes = 0;
1969                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1970                 }
1971                 break;
1972         default:
1973                 /*
1974                  * In case of bogus fw or device, we could end up having
1975                  * an unaligned partial completion. Check this here and force
1976                  * alignment.
1977                  */
1978                 resid = scsi_get_resid(SCpnt);
1979                 if (resid & (sector_size - 1)) {
1980                         sd_printk(KERN_INFO, sdkp,
1981                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1982                                 resid, sector_size);
1983                         resid = min(scsi_bufflen(SCpnt),
1984                                     round_up(resid, sector_size));
1985                         scsi_set_resid(SCpnt, resid);
1986                 }
1987         }
1988
1989         if (result) {
1990                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1991                 if (sense_valid)
1992                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1993         }
1994         sdkp->medium_access_timed_out = 0;
1995
1996         if (driver_byte(result) != DRIVER_SENSE &&
1997             (!sense_valid || sense_deferred))
1998                 goto out;
1999
2000         switch (sshdr.sense_key) {
2001         case HARDWARE_ERROR:
2002         case MEDIUM_ERROR:
2003                 good_bytes = sd_completed_bytes(SCpnt);
2004                 break;
2005         case RECOVERED_ERROR:
2006                 good_bytes = scsi_bufflen(SCpnt);
2007                 break;
2008         case NO_SENSE:
2009                 /* This indicates a false check condition, so ignore it.  An
2010                  * unknown amount of data was transferred so treat it as an
2011                  * error.
2012                  */
2013                 SCpnt->result = 0;
2014                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2015                 break;
2016         case ABORTED_COMMAND:
2017                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2018                         good_bytes = sd_completed_bytes(SCpnt);
2019                 break;
2020         case ILLEGAL_REQUEST:
2021                 switch (sshdr.asc) {
2022                 case 0x10:      /* DIX: Host detected corruption */
2023                         good_bytes = sd_completed_bytes(SCpnt);
2024                         break;
2025                 case 0x20:      /* INVALID COMMAND OPCODE */
2026                 case 0x24:      /* INVALID FIELD IN CDB */
2027                         switch (SCpnt->cmnd[0]) {
2028                         case UNMAP:
2029                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2030                                 break;
2031                         case WRITE_SAME_16:
2032                         case WRITE_SAME:
2033                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2034                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2035                                 } else {
2036                                         sdkp->device->no_write_same = 1;
2037                                         sd_config_write_same(sdkp);
2038                                         req->rq_flags |= RQF_QUIET;
2039                                 }
2040                                 break;
2041                         }
2042                 }
2043                 break;
2044         default:
2045                 break;
2046         }
2047
2048  out:
2049         if (sd_is_zoned(sdkp))
2050                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2051
2052         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2053                                            "sd_done: completed %d of %d bytes\n",
2054                                            good_bytes, scsi_bufflen(SCpnt)));
2055
2056         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2057             good_bytes)
2058                 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2059                                 good_bytes / scsi_prot_interval(SCpnt));
2060
2061         return good_bytes;
2062 }
2063
2064 /*
2065  * spinup disk - called only in sd_revalidate_disk()
2066  */
2067 static void
2068 sd_spinup_disk(struct scsi_disk *sdkp)
2069 {
2070         unsigned char cmd[10];
2071         unsigned long spintime_expire = 0;
2072         int retries, spintime;
2073         unsigned int the_result;
2074         struct scsi_sense_hdr sshdr;
2075         int sense_valid = 0;
2076
2077         spintime = 0;
2078
2079         /* Spin up drives, as required.  Only do this at boot time */
2080         /* Spinup needs to be done for module loads too. */
2081         do {
2082                 retries = 0;
2083
2084                 do {
2085                         cmd[0] = TEST_UNIT_READY;
2086                         memset((void *) &cmd[1], 0, 9);
2087
2088                         the_result = scsi_execute_req(sdkp->device, cmd,
2089                                                       DMA_NONE, NULL, 0,
2090                                                       &sshdr, SD_TIMEOUT,
2091                                                       SD_MAX_RETRIES, NULL);
2092
2093                         /*
2094                          * If the drive has indicated to us that it
2095                          * doesn't have any media in it, don't bother
2096                          * with any more polling.
2097                          */
2098                         if (media_not_present(sdkp, &sshdr))
2099                                 return;
2100
2101                         if (the_result)
2102                                 sense_valid = scsi_sense_valid(&sshdr);
2103                         retries++;
2104                 } while (retries < 3 &&
2105                          (!scsi_status_is_good(the_result) ||
2106                           ((driver_byte(the_result) == DRIVER_SENSE) &&
2107                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2108
2109                 if (driver_byte(the_result) != DRIVER_SENSE) {
2110                         /* no sense, TUR either succeeded or failed
2111                          * with a status error */
2112                         if(!spintime && !scsi_status_is_good(the_result)) {
2113                                 sd_print_result(sdkp, "Test Unit Ready failed",
2114                                                 the_result);
2115                         }
2116                         break;
2117                 }
2118
2119                 /*
2120                  * The device does not want the automatic start to be issued.
2121                  */
2122                 if (sdkp->device->no_start_on_add)
2123                         break;
2124
2125                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2126                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2127                                 break;  /* manual intervention required */
2128                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2129                                 break;  /* standby */
2130                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2131                                 break;  /* unavailable */
2132                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2133                                 break;  /* sanitize in progress */
2134                         /*
2135                          * Issue command to spin up drive when not ready
2136                          */
2137                         if (!spintime) {
2138                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2139                                 cmd[0] = START_STOP;
2140                                 cmd[1] = 1;     /* Return immediately */
2141                                 memset((void *) &cmd[2], 0, 8);
2142                                 cmd[4] = 1;     /* Start spin cycle */
2143                                 if (sdkp->device->start_stop_pwr_cond)
2144                                         cmd[4] |= 1 << 4;
2145                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2146                                                  NULL, 0, &sshdr,
2147                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2148                                                  NULL);
2149                                 spintime_expire = jiffies + 100 * HZ;
2150                                 spintime = 1;
2151                         }
2152                         /* Wait 1 second for next try */
2153                         msleep(1000);
2154                         printk(KERN_CONT ".");
2155
2156                 /*
2157                  * Wait for USB flash devices with slow firmware.
2158                  * Yes, this sense key/ASC combination shouldn't
2159                  * occur here.  It's characteristic of these devices.
2160                  */
2161                 } else if (sense_valid &&
2162                                 sshdr.sense_key == UNIT_ATTENTION &&
2163                                 sshdr.asc == 0x28) {
2164                         if (!spintime) {
2165                                 spintime_expire = jiffies + 5 * HZ;
2166                                 spintime = 1;
2167                         }
2168                         /* Wait 1 second for next try */
2169                         msleep(1000);
2170                 } else {
2171                         /* we don't understand the sense code, so it's
2172                          * probably pointless to loop */
2173                         if(!spintime) {
2174                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2175                                 sd_print_sense_hdr(sdkp, &sshdr);
2176                         }
2177                         break;
2178                 }
2179
2180         } while (spintime && time_before_eq(jiffies, spintime_expire));
2181
2182         if (spintime) {
2183                 if (scsi_status_is_good(the_result))
2184                         printk(KERN_CONT "ready\n");
2185                 else
2186                         printk(KERN_CONT "not responding...\n");
2187         }
2188 }
2189
2190 /*
2191  * Determine whether disk supports Data Integrity Field.
2192  */
2193 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2194 {
2195         struct scsi_device *sdp = sdkp->device;
2196         u8 type;
2197         int ret = 0;
2198
2199         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2200                 return ret;
2201
2202         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2203
2204         if (type > T10_PI_TYPE3_PROTECTION)
2205                 ret = -ENODEV;
2206         else if (scsi_host_dif_capable(sdp->host, type))
2207                 ret = 1;
2208
2209         if (sdkp->first_scan || type != sdkp->protection_type)
2210                 switch (ret) {
2211                 case -ENODEV:
2212                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2213                                   " protection type %u. Disabling disk!\n",
2214                                   type);
2215                         break;
2216                 case 1:
2217                         sd_printk(KERN_NOTICE, sdkp,
2218                                   "Enabling DIF Type %u protection\n", type);
2219                         break;
2220                 case 0:
2221                         sd_printk(KERN_NOTICE, sdkp,
2222                                   "Disabling DIF Type %u protection\n", type);
2223                         break;
2224                 }
2225
2226         sdkp->protection_type = type;
2227
2228         return ret;
2229 }
2230
2231 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2232                         struct scsi_sense_hdr *sshdr, int sense_valid,
2233                         int the_result)
2234 {
2235         if (driver_byte(the_result) == DRIVER_SENSE)
2236                 sd_print_sense_hdr(sdkp, sshdr);
2237         else
2238                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2239
2240         /*
2241          * Set dirty bit for removable devices if not ready -
2242          * sometimes drives will not report this properly.
2243          */
2244         if (sdp->removable &&
2245             sense_valid && sshdr->sense_key == NOT_READY)
2246                 set_media_not_present(sdkp);
2247
2248         /*
2249          * We used to set media_present to 0 here to indicate no media
2250          * in the drive, but some drives fail read capacity even with
2251          * media present, so we can't do that.
2252          */
2253         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2254 }
2255
2256 #define RC16_LEN 32
2257 #if RC16_LEN > SD_BUF_SIZE
2258 #error RC16_LEN must not be more than SD_BUF_SIZE
2259 #endif
2260
2261 #define READ_CAPACITY_RETRIES_ON_RESET  10
2262
2263 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2264                                                 unsigned char *buffer)
2265 {
2266         unsigned char cmd[16];
2267         struct scsi_sense_hdr sshdr;
2268         int sense_valid = 0;
2269         int the_result;
2270         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2271         unsigned int alignment;
2272         unsigned long long lba;
2273         unsigned sector_size;
2274
2275         if (sdp->no_read_capacity_16)
2276                 return -EINVAL;
2277
2278         do {
2279                 memset(cmd, 0, 16);
2280                 cmd[0] = SERVICE_ACTION_IN_16;
2281                 cmd[1] = SAI_READ_CAPACITY_16;
2282                 cmd[13] = RC16_LEN;
2283                 memset(buffer, 0, RC16_LEN);
2284
2285                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2286                                         buffer, RC16_LEN, &sshdr,
2287                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2288
2289                 if (media_not_present(sdkp, &sshdr))
2290                         return -ENODEV;
2291
2292                 if (the_result) {
2293                         sense_valid = scsi_sense_valid(&sshdr);
2294                         if (sense_valid &&
2295                             sshdr.sense_key == ILLEGAL_REQUEST &&
2296                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2297                             sshdr.ascq == 0x00)
2298                                 /* Invalid Command Operation Code or
2299                                  * Invalid Field in CDB, just retry
2300                                  * silently with RC10 */
2301                                 return -EINVAL;
2302                         if (sense_valid &&
2303                             sshdr.sense_key == UNIT_ATTENTION &&
2304                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2305                                 /* Device reset might occur several times,
2306                                  * give it one more chance */
2307                                 if (--reset_retries > 0)
2308                                         continue;
2309                 }
2310                 retries--;
2311
2312         } while (the_result && retries);
2313
2314         if (the_result) {
2315                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2316                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2317                 return -EINVAL;
2318         }
2319
2320         sector_size = get_unaligned_be32(&buffer[8]);
2321         lba = get_unaligned_be64(&buffer[0]);
2322
2323         if (sd_read_protection_type(sdkp, buffer) < 0) {
2324                 sdkp->capacity = 0;
2325                 return -ENODEV;
2326         }
2327
2328         /* Logical blocks per physical block exponent */
2329         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2330
2331         /* RC basis */
2332         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2333
2334         /* Lowest aligned logical block */
2335         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2336         blk_queue_alignment_offset(sdp->request_queue, alignment);
2337         if (alignment && sdkp->first_scan)
2338                 sd_printk(KERN_NOTICE, sdkp,
2339                           "physical block alignment offset: %u\n", alignment);
2340
2341         if (buffer[14] & 0x80) { /* LBPME */
2342                 sdkp->lbpme = 1;
2343
2344                 if (buffer[14] & 0x40) /* LBPRZ */
2345                         sdkp->lbprz = 1;
2346
2347                 sd_config_discard(sdkp, SD_LBP_WS16);
2348         }
2349
2350         sdkp->capacity = lba + 1;
2351         return sector_size;
2352 }
2353
2354 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2355                                                 unsigned char *buffer)
2356 {
2357         unsigned char cmd[16];
2358         struct scsi_sense_hdr sshdr;
2359         int sense_valid = 0;
2360         int the_result;
2361         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2362         sector_t lba;
2363         unsigned sector_size;
2364
2365         do {
2366                 cmd[0] = READ_CAPACITY;
2367                 memset(&cmd[1], 0, 9);
2368                 memset(buffer, 0, 8);
2369
2370                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2371                                         buffer, 8, &sshdr,
2372                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2373
2374                 if (media_not_present(sdkp, &sshdr))
2375                         return -ENODEV;
2376
2377                 if (the_result) {
2378                         sense_valid = scsi_sense_valid(&sshdr);
2379                         if (sense_valid &&
2380                             sshdr.sense_key == UNIT_ATTENTION &&
2381                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2382                                 /* Device reset might occur several times,
2383                                  * give it one more chance */
2384                                 if (--reset_retries > 0)
2385                                         continue;
2386                 }
2387                 retries--;
2388
2389         } while (the_result && retries);
2390
2391         if (the_result) {
2392                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2393                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2394                 return -EINVAL;
2395         }
2396
2397         sector_size = get_unaligned_be32(&buffer[4]);
2398         lba = get_unaligned_be32(&buffer[0]);
2399
2400         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2401                 /* Some buggy (usb cardreader) devices return an lba of
2402                    0xffffffff when the want to report a size of 0 (with
2403                    which they really mean no media is present) */
2404                 sdkp->capacity = 0;
2405                 sdkp->physical_block_size = sector_size;
2406                 return sector_size;
2407         }
2408
2409         sdkp->capacity = lba + 1;
2410         sdkp->physical_block_size = sector_size;
2411         return sector_size;
2412 }
2413
2414 static int sd_try_rc16_first(struct scsi_device *sdp)
2415 {
2416         if (sdp->host->max_cmd_len < 16)
2417                 return 0;
2418         if (sdp->try_rc_10_first)
2419                 return 0;
2420         if (sdp->scsi_level > SCSI_SPC_2)
2421                 return 1;
2422         if (scsi_device_protection(sdp))
2423                 return 1;
2424         return 0;
2425 }
2426
2427 /*
2428  * read disk capacity
2429  */
2430 static void
2431 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2432 {
2433         int sector_size;
2434         struct scsi_device *sdp = sdkp->device;
2435
2436         if (sd_try_rc16_first(sdp)) {
2437                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2438                 if (sector_size == -EOVERFLOW)
2439                         goto got_data;
2440                 if (sector_size == -ENODEV)
2441                         return;
2442                 if (sector_size < 0)
2443                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2444                 if (sector_size < 0)
2445                         return;
2446         } else {
2447                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2448                 if (sector_size == -EOVERFLOW)
2449                         goto got_data;
2450                 if (sector_size < 0)
2451                         return;
2452                 if ((sizeof(sdkp->capacity) > 4) &&
2453                     (sdkp->capacity > 0xffffffffULL)) {
2454                         int old_sector_size = sector_size;
2455                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2456                                         "Trying to use READ CAPACITY(16).\n");
2457                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2458                         if (sector_size < 0) {
2459                                 sd_printk(KERN_NOTICE, sdkp,
2460                                         "Using 0xffffffff as device size\n");
2461                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2462                                 sector_size = old_sector_size;
2463                                 goto got_data;
2464                         }
2465                         /* Remember that READ CAPACITY(16) succeeded */
2466                         sdp->try_rc_10_first = 0;
2467                 }
2468         }
2469
2470         /* Some devices are known to return the total number of blocks,
2471          * not the highest block number.  Some devices have versions
2472          * which do this and others which do not.  Some devices we might
2473          * suspect of doing this but we don't know for certain.
2474          *
2475          * If we know the reported capacity is wrong, decrement it.  If
2476          * we can only guess, then assume the number of blocks is even
2477          * (usually true but not always) and err on the side of lowering
2478          * the capacity.
2479          */
2480         if (sdp->fix_capacity ||
2481             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2482                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2483                                 "from its reported value: %llu\n",
2484                                 (unsigned long long) sdkp->capacity);
2485                 --sdkp->capacity;
2486         }
2487
2488 got_data:
2489         if (sector_size == 0) {
2490                 sector_size = 512;
2491                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2492                           "assuming 512.\n");
2493         }
2494
2495         if (sector_size != 512 &&
2496             sector_size != 1024 &&
2497             sector_size != 2048 &&
2498             sector_size != 4096) {
2499                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2500                           sector_size);
2501                 /*
2502                  * The user might want to re-format the drive with
2503                  * a supported sectorsize.  Once this happens, it
2504                  * would be relatively trivial to set the thing up.
2505                  * For this reason, we leave the thing in the table.
2506                  */
2507                 sdkp->capacity = 0;
2508                 /*
2509                  * set a bogus sector size so the normal read/write
2510                  * logic in the block layer will eventually refuse any
2511                  * request on this device without tripping over power
2512                  * of two sector size assumptions
2513                  */
2514                 sector_size = 512;
2515         }
2516         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2517         blk_queue_physical_block_size(sdp->request_queue,
2518                                       sdkp->physical_block_size);
2519         sdkp->device->sector_size = sector_size;
2520
2521         if (sdkp->capacity > 0xffffffff)
2522                 sdp->use_16_for_rw = 1;
2523
2524 }
2525
2526 /*
2527  * Print disk capacity
2528  */
2529 static void
2530 sd_print_capacity(struct scsi_disk *sdkp,
2531                   sector_t old_capacity)
2532 {
2533         int sector_size = sdkp->device->sector_size;
2534         char cap_str_2[10], cap_str_10[10];
2535
2536         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2537                 return;
2538
2539         string_get_size(sdkp->capacity, sector_size,
2540                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2541         string_get_size(sdkp->capacity, sector_size,
2542                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2543
2544         sd_printk(KERN_NOTICE, sdkp,
2545                   "%llu %d-byte logical blocks: (%s/%s)\n",
2546                   (unsigned long long)sdkp->capacity,
2547                   sector_size, cap_str_10, cap_str_2);
2548
2549         if (sdkp->physical_block_size != sector_size)
2550                 sd_printk(KERN_NOTICE, sdkp,
2551                           "%u-byte physical blocks\n",
2552                           sdkp->physical_block_size);
2553
2554         sd_zbc_print_zones(sdkp);
2555 }
2556
2557 /* called with buffer of length 512 */
2558 static inline int
2559 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2560                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2561                  struct scsi_sense_hdr *sshdr)
2562 {
2563         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2564                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2565                                sshdr);
2566 }
2567
2568 /*
2569  * read write protect setting, if possible - called only in sd_revalidate_disk()
2570  * called with buffer of length SD_BUF_SIZE
2571  */
2572 static void
2573 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2574 {
2575         int res;
2576         struct scsi_device *sdp = sdkp->device;
2577         struct scsi_mode_data data;
2578         int old_wp = sdkp->write_prot;
2579
2580         set_disk_ro(sdkp->disk, 0);
2581         if (sdp->skip_ms_page_3f) {
2582                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2583                 return;
2584         }
2585
2586         if (sdp->use_192_bytes_for_3f) {
2587                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2588         } else {
2589                 /*
2590                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2591                  * We have to start carefully: some devices hang if we ask
2592                  * for more than is available.
2593                  */
2594                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2595
2596                 /*
2597                  * Second attempt: ask for page 0 When only page 0 is
2598                  * implemented, a request for page 3F may return Sense Key
2599                  * 5: Illegal Request, Sense Code 24: Invalid field in
2600                  * CDB.
2601                  */
2602                 if (!scsi_status_is_good(res))
2603                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2604
2605                 /*
2606                  * Third attempt: ask 255 bytes, as we did earlier.
2607                  */
2608                 if (!scsi_status_is_good(res))
2609                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2610                                                &data, NULL);
2611         }
2612
2613         if (!scsi_status_is_good(res)) {
2614                 sd_first_printk(KERN_WARNING, sdkp,
2615                           "Test WP failed, assume Write Enabled\n");
2616         } else {
2617                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2618                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2619                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2620                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2621                                   sdkp->write_prot ? "on" : "off");
2622                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2623                 }
2624         }
2625 }
2626
2627 /*
2628  * sd_read_cache_type - called only from sd_revalidate_disk()
2629  * called with buffer of length SD_BUF_SIZE
2630  */
2631 static void
2632 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2633 {
2634         int len = 0, res;
2635         struct scsi_device *sdp = sdkp->device;
2636
2637         int dbd;
2638         int modepage;
2639         int first_len;
2640         struct scsi_mode_data data;
2641         struct scsi_sense_hdr sshdr;
2642         int old_wce = sdkp->WCE;
2643         int old_rcd = sdkp->RCD;
2644         int old_dpofua = sdkp->DPOFUA;
2645
2646
2647         if (sdkp->cache_override)
2648                 return;
2649
2650         first_len = 4;
2651         if (sdp->skip_ms_page_8) {
2652                 if (sdp->type == TYPE_RBC)
2653                         goto defaults;
2654                 else {
2655                         if (sdp->skip_ms_page_3f)
2656                                 goto defaults;
2657                         modepage = 0x3F;
2658                         if (sdp->use_192_bytes_for_3f)
2659                                 first_len = 192;
2660                         dbd = 0;
2661                 }
2662         } else if (sdp->type == TYPE_RBC) {
2663                 modepage = 6;
2664                 dbd = 8;
2665         } else {
2666                 modepage = 8;
2667                 dbd = 0;
2668         }
2669
2670         /* cautiously ask */
2671         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2672                         &data, &sshdr);
2673
2674         if (!scsi_status_is_good(res))
2675                 goto bad_sense;
2676
2677         if (!data.header_length) {
2678                 modepage = 6;
2679                 first_len = 0;
2680                 sd_first_printk(KERN_ERR, sdkp,
2681                                 "Missing header in MODE_SENSE response\n");
2682         }
2683
2684         /* that went OK, now ask for the proper length */
2685         len = data.length;
2686
2687         /*
2688          * We're only interested in the first three bytes, actually.
2689          * But the data cache page is defined for the first 20.
2690          */
2691         if (len < 3)
2692                 goto bad_sense;
2693         else if (len > SD_BUF_SIZE) {
2694                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2695                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2696                 len = SD_BUF_SIZE;
2697         }
2698         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2699                 len = 192;
2700
2701         /* Get the data */
2702         if (len > first_len)
2703                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2704                                 &data, &sshdr);
2705
2706         if (scsi_status_is_good(res)) {
2707                 int offset = data.header_length + data.block_descriptor_length;
2708
2709                 while (offset < len) {
2710                         u8 page_code = buffer[offset] & 0x3F;
2711                         u8 spf       = buffer[offset] & 0x40;
2712
2713                         if (page_code == 8 || page_code == 6) {
2714                                 /* We're interested only in the first 3 bytes.
2715                                  */
2716                                 if (len - offset <= 2) {
2717                                         sd_first_printk(KERN_ERR, sdkp,
2718                                                 "Incomplete mode parameter "
2719                                                         "data\n");
2720                                         goto defaults;
2721                                 } else {
2722                                         modepage = page_code;
2723                                         goto Page_found;
2724                                 }
2725                         } else {
2726                                 /* Go to the next page */
2727                                 if (spf && len - offset > 3)
2728                                         offset += 4 + (buffer[offset+2] << 8) +
2729                                                 buffer[offset+3];
2730                                 else if (!spf && len - offset > 1)
2731                                         offset += 2 + buffer[offset+1];
2732                                 else {
2733                                         sd_first_printk(KERN_ERR, sdkp,
2734                                                         "Incomplete mode "
2735                                                         "parameter data\n");
2736                                         goto defaults;
2737                                 }
2738                         }
2739                 }
2740
2741                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2742                 goto defaults;
2743
2744         Page_found:
2745                 if (modepage == 8) {
2746                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2747                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2748                 } else {
2749                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2750                         sdkp->RCD = 0;
2751                 }
2752
2753                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2754                 if (sdp->broken_fua) {
2755                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2756                         sdkp->DPOFUA = 0;
2757                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2758                            !sdkp->device->use_16_for_rw) {
2759                         sd_first_printk(KERN_NOTICE, sdkp,
2760                                   "Uses READ/WRITE(6), disabling FUA\n");
2761                         sdkp->DPOFUA = 0;
2762                 }
2763
2764                 /* No cache flush allowed for write protected devices */
2765                 if (sdkp->WCE && sdkp->write_prot)
2766                         sdkp->WCE = 0;
2767
2768                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2769                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2770                         sd_printk(KERN_NOTICE, sdkp,
2771                                   "Write cache: %s, read cache: %s, %s\n",
2772                                   sdkp->WCE ? "enabled" : "disabled",
2773                                   sdkp->RCD ? "disabled" : "enabled",
2774                                   sdkp->DPOFUA ? "supports DPO and FUA"
2775                                   : "doesn't support DPO or FUA");
2776
2777                 return;
2778         }
2779
2780 bad_sense:
2781         if (scsi_sense_valid(&sshdr) &&
2782             sshdr.sense_key == ILLEGAL_REQUEST &&
2783             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2784                 /* Invalid field in CDB */
2785                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2786         else
2787                 sd_first_printk(KERN_ERR, sdkp,
2788                                 "Asking for cache data failed\n");
2789
2790 defaults:
2791         if (sdp->wce_default_on) {
2792                 sd_first_printk(KERN_NOTICE, sdkp,
2793                                 "Assuming drive cache: write back\n");
2794                 sdkp->WCE = 1;
2795         } else {
2796                 sd_first_printk(KERN_ERR, sdkp,
2797                                 "Assuming drive cache: write through\n");
2798                 sdkp->WCE = 0;
2799         }
2800         sdkp->RCD = 0;
2801         sdkp->DPOFUA = 0;
2802 }
2803
2804 /*
2805  * The ATO bit indicates whether the DIF application tag is available
2806  * for use by the operating system.
2807  */
2808 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2809 {
2810         int res, offset;
2811         struct scsi_device *sdp = sdkp->device;
2812         struct scsi_mode_data data;
2813         struct scsi_sense_hdr sshdr;
2814
2815         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2816                 return;
2817
2818         if (sdkp->protection_type == 0)
2819                 return;
2820
2821         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2822                               SD_MAX_RETRIES, &data, &sshdr);
2823
2824         if (!scsi_status_is_good(res) || !data.header_length ||
2825             data.length < 6) {
2826                 sd_first_printk(KERN_WARNING, sdkp,
2827                           "getting Control mode page failed, assume no ATO\n");
2828
2829                 if (scsi_sense_valid(&sshdr))
2830                         sd_print_sense_hdr(sdkp, &sshdr);
2831
2832                 return;
2833         }
2834
2835         offset = data.header_length + data.block_descriptor_length;
2836
2837         if ((buffer[offset] & 0x3f) != 0x0a) {
2838                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2839                 return;
2840         }
2841
2842         if ((buffer[offset + 5] & 0x80) == 0)
2843                 return;
2844
2845         sdkp->ATO = 1;
2846
2847         return;
2848 }
2849
2850 /**
2851  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2852  * @sdkp: disk to query
2853  */
2854 static void sd_read_block_limits(struct scsi_disk *sdkp)
2855 {
2856         unsigned int sector_sz = sdkp->device->sector_size;
2857         const int vpd_len = 64;
2858         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2859
2860         if (!buffer ||
2861             /* Block Limits VPD */
2862             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2863                 goto out;
2864
2865         blk_queue_io_min(sdkp->disk->queue,
2866                          get_unaligned_be16(&buffer[6]) * sector_sz);
2867
2868         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2869         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2870
2871         if (buffer[3] == 0x3c) {
2872                 unsigned int lba_count, desc_count;
2873
2874                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2875
2876                 if (!sdkp->lbpme)
2877                         goto out;
2878
2879                 lba_count = get_unaligned_be32(&buffer[20]);
2880                 desc_count = get_unaligned_be32(&buffer[24]);
2881
2882                 if (lba_count && desc_count)
2883                         sdkp->max_unmap_blocks = lba_count;
2884
2885                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2886
2887                 if (buffer[32] & 0x80)
2888                         sdkp->unmap_alignment =
2889                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2890
2891                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2892
2893                         if (sdkp->max_unmap_blocks)
2894                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2895                         else
2896                                 sd_config_discard(sdkp, SD_LBP_WS16);
2897
2898                 } else {        /* LBP VPD page tells us what to use */
2899                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2900                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2901                         else if (sdkp->lbpws)
2902                                 sd_config_discard(sdkp, SD_LBP_WS16);
2903                         else if (sdkp->lbpws10)
2904                                 sd_config_discard(sdkp, SD_LBP_WS10);
2905                         else
2906                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2907                 }
2908         }
2909
2910  out:
2911         kfree(buffer);
2912 }
2913
2914 /**
2915  * sd_read_block_characteristics - Query block dev. characteristics
2916  * @sdkp: disk to query
2917  */
2918 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2919 {
2920         struct request_queue *q = sdkp->disk->queue;
2921         unsigned char *buffer;
2922         u16 rot;
2923         const int vpd_len = 64;
2924
2925         buffer = kmalloc(vpd_len, GFP_KERNEL);
2926
2927         if (!buffer ||
2928             /* Block Device Characteristics VPD */
2929             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2930                 goto out;
2931
2932         rot = get_unaligned_be16(&buffer[4]);
2933
2934         if (rot == 1) {
2935                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2936                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2937         }
2938
2939         if (sdkp->device->type == TYPE_ZBC) {
2940                 /* Host-managed */
2941                 q->limits.zoned = BLK_ZONED_HM;
2942         } else {
2943                 sdkp->zoned = (buffer[8] >> 4) & 3;
2944                 if (sdkp->zoned == 1)
2945                         /* Host-aware */
2946                         q->limits.zoned = BLK_ZONED_HA;
2947                 else
2948                         /*
2949                          * Treat drive-managed devices as
2950                          * regular block devices.
2951                          */
2952                         q->limits.zoned = BLK_ZONED_NONE;
2953         }
2954         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2955                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2956                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2957
2958  out:
2959         kfree(buffer);
2960 }
2961
2962 /**
2963  * sd_read_block_provisioning - Query provisioning VPD page
2964  * @sdkp: disk to query
2965  */
2966 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2967 {
2968         unsigned char *buffer;
2969         const int vpd_len = 8;
2970
2971         if (sdkp->lbpme == 0)
2972                 return;
2973
2974         buffer = kmalloc(vpd_len, GFP_KERNEL);
2975
2976         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2977                 goto out;
2978
2979         sdkp->lbpvpd    = 1;
2980         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
2981         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2982         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2983
2984  out:
2985         kfree(buffer);
2986 }
2987
2988 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2989 {
2990         struct scsi_device *sdev = sdkp->device;
2991
2992         if (sdev->host->no_write_same) {
2993                 sdev->no_write_same = 1;
2994
2995                 return;
2996         }
2997
2998         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2999                 /* too large values might cause issues with arcmsr */
3000                 int vpd_buf_len = 64;
3001
3002                 sdev->no_report_opcodes = 1;
3003
3004                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3005                  * CODES is unsupported and the device has an ATA
3006                  * Information VPD page (SAT).
3007                  */
3008                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3009                         sdev->no_write_same = 1;
3010         }
3011
3012         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3013                 sdkp->ws16 = 1;
3014
3015         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3016                 sdkp->ws10 = 1;
3017 }
3018
3019 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3020 {
3021         struct scsi_device *sdev = sdkp->device;
3022
3023         if (!sdev->security_supported)
3024                 return;
3025
3026         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3027                         SECURITY_PROTOCOL_IN) == 1 &&
3028             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3029                         SECURITY_PROTOCOL_OUT) == 1)
3030                 sdkp->security = 1;
3031 }
3032
3033 /*
3034  * Determine the device's preferred I/O size for reads and writes
3035  * unless the reported value is unreasonably small, large, not a
3036  * multiple of the physical block size, or simply garbage.
3037  */
3038 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3039                                       unsigned int dev_max)
3040 {
3041         struct scsi_device *sdp = sdkp->device;
3042         unsigned int opt_xfer_bytes =
3043                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3044
3045         if (sdkp->opt_xfer_blocks == 0)
3046                 return false;
3047
3048         if (sdkp->opt_xfer_blocks > dev_max) {
3049                 sd_first_printk(KERN_WARNING, sdkp,
3050                                 "Optimal transfer size %u logical blocks " \
3051                                 "> dev_max (%u logical blocks)\n",
3052                                 sdkp->opt_xfer_blocks, dev_max);
3053                 return false;
3054         }
3055
3056         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3057                 sd_first_printk(KERN_WARNING, sdkp,
3058                                 "Optimal transfer size %u logical blocks " \
3059                                 "> sd driver limit (%u logical blocks)\n",
3060                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3061                 return false;
3062         }
3063
3064         if (opt_xfer_bytes < PAGE_SIZE) {
3065                 sd_first_printk(KERN_WARNING, sdkp,
3066                                 "Optimal transfer size %u bytes < " \
3067                                 "PAGE_SIZE (%u bytes)\n",
3068                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3069                 return false;
3070         }
3071
3072         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3073                 sd_first_printk(KERN_WARNING, sdkp,
3074                                 "Optimal transfer size %u bytes not a " \
3075                                 "multiple of physical block size (%u bytes)\n",
3076                                 opt_xfer_bytes, sdkp->physical_block_size);
3077                 return false;
3078         }
3079
3080         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3081                         opt_xfer_bytes);
3082         return true;
3083 }
3084
3085 /**
3086  *      sd_revalidate_disk - called the first time a new disk is seen,
3087  *      performs disk spin up, read_capacity, etc.
3088  *      @disk: struct gendisk we care about
3089  **/
3090 static int sd_revalidate_disk(struct gendisk *disk)
3091 {
3092         struct scsi_disk *sdkp = scsi_disk(disk);
3093         struct scsi_device *sdp = sdkp->device;
3094         struct request_queue *q = sdkp->disk->queue;
3095         sector_t old_capacity = sdkp->capacity;
3096         unsigned char *buffer;
3097         unsigned int dev_max, rw_max;
3098
3099         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3100                                       "sd_revalidate_disk\n"));
3101
3102         /*
3103          * If the device is offline, don't try and read capacity or any
3104          * of the other niceties.
3105          */
3106         if (!scsi_device_online(sdp))
3107                 goto out;
3108
3109         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3110         if (!buffer) {
3111                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3112                           "allocation failure.\n");
3113                 goto out;
3114         }
3115
3116         sd_spinup_disk(sdkp);
3117
3118         /*
3119          * Without media there is no reason to ask; moreover, some devices
3120          * react badly if we do.
3121          */
3122         if (sdkp->media_present) {
3123                 sd_read_capacity(sdkp, buffer);
3124
3125                 /*
3126                  * set the default to rotational.  All non-rotational devices
3127                  * support the block characteristics VPD page, which will
3128                  * cause this to be updated correctly and any device which
3129                  * doesn't support it should be treated as rotational.
3130                  */
3131                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3132                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3133
3134                 if (scsi_device_supports_vpd(sdp)) {
3135                         sd_read_block_provisioning(sdkp);
3136                         sd_read_block_limits(sdkp);
3137                         sd_read_block_characteristics(sdkp);
3138                         sd_zbc_read_zones(sdkp, buffer);
3139                 }
3140
3141                 sd_print_capacity(sdkp, old_capacity);
3142
3143                 sd_read_write_protect_flag(sdkp, buffer);
3144                 sd_read_cache_type(sdkp, buffer);
3145                 sd_read_app_tag_own(sdkp, buffer);
3146                 sd_read_write_same(sdkp, buffer);
3147                 sd_read_security(sdkp, buffer);
3148         }
3149
3150         /*
3151          * We now have all cache related info, determine how we deal
3152          * with flush requests.
3153          */
3154         sd_set_flush_flag(sdkp);
3155
3156         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3157         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3158
3159         /* Some devices report a maximum block count for READ/WRITE requests. */
3160         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3161         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3162
3163         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3164                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3165                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3166         } else
3167                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3168                                       (sector_t)BLK_DEF_MAX_SECTORS);
3169
3170         /* Do not exceed controller limit */
3171         rw_max = min(rw_max, queue_max_hw_sectors(q));
3172
3173         /*
3174          * Only update max_sectors if previously unset or if the current value
3175          * exceeds the capabilities of the hardware.
3176          */
3177         if (sdkp->first_scan ||
3178             q->limits.max_sectors > q->limits.max_dev_sectors ||
3179             q->limits.max_sectors > q->limits.max_hw_sectors)
3180                 q->limits.max_sectors = rw_max;
3181
3182         sdkp->first_scan = 0;
3183
3184         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3185         sd_config_write_same(sdkp);
3186         kfree(buffer);
3187
3188  out:
3189         return 0;
3190 }
3191
3192 /**
3193  *      sd_unlock_native_capacity - unlock native capacity
3194  *      @disk: struct gendisk to set capacity for
3195  *
3196  *      Block layer calls this function if it detects that partitions
3197  *      on @disk reach beyond the end of the device.  If the SCSI host
3198  *      implements ->unlock_native_capacity() method, it's invoked to
3199  *      give it a chance to adjust the device capacity.
3200  *
3201  *      CONTEXT:
3202  *      Defined by block layer.  Might sleep.
3203  */
3204 static void sd_unlock_native_capacity(struct gendisk *disk)
3205 {
3206         struct scsi_device *sdev = scsi_disk(disk)->device;
3207
3208         if (sdev->host->hostt->unlock_native_capacity)
3209                 sdev->host->hostt->unlock_native_capacity(sdev);
3210 }
3211
3212 /**
3213  *      sd_format_disk_name - format disk name
3214  *      @prefix: name prefix - ie. "sd" for SCSI disks
3215  *      @index: index of the disk to format name for
3216  *      @buf: output buffer
3217  *      @buflen: length of the output buffer
3218  *
3219  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3220  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3221  *      which is followed by sdaaa.
3222  *
3223  *      This is basically 26 base counting with one extra 'nil' entry
3224  *      at the beginning from the second digit on and can be
3225  *      determined using similar method as 26 base conversion with the
3226  *      index shifted -1 after each digit is computed.
3227  *
3228  *      CONTEXT:
3229  *      Don't care.
3230  *
3231  *      RETURNS:
3232  *      0 on success, -errno on failure.
3233  */
3234 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3235 {
3236         const int base = 'z' - 'a' + 1;
3237         char *begin = buf + strlen(prefix);
3238         char *end = buf + buflen;
3239         char *p;
3240         int unit;
3241
3242         p = end - 1;
3243         *p = '\0';
3244         unit = base;
3245         do {
3246                 if (p == begin)
3247                         return -EINVAL;
3248                 *--p = 'a' + (index % unit);
3249                 index = (index / unit) - 1;
3250         } while (index >= 0);
3251
3252         memmove(begin, p, end - p);
3253         memcpy(buf, prefix, strlen(prefix));
3254
3255         return 0;
3256 }
3257
3258 /**
3259  *      sd_probe - called during driver initialization and whenever a
3260  *      new scsi device is attached to the system. It is called once
3261  *      for each scsi device (not just disks) present.
3262  *      @dev: pointer to device object
3263  *
3264  *      Returns 0 if successful (or not interested in this scsi device
3265  *      (e.g. scanner)); 1 when there is an error.
3266  *
3267  *      Note: this function is invoked from the scsi mid-level.
3268  *      This function sets up the mapping between a given
3269  *      <host,channel,id,lun> (found in sdp) and new device name
3270  *      (e.g. /dev/sda). More precisely it is the block device major
3271  *      and minor number that is chosen here.
3272  *
3273  *      Assume sd_probe is not re-entrant (for time being)
3274  *      Also think about sd_probe() and sd_remove() running coincidentally.
3275  **/
3276 static int sd_probe(struct device *dev)
3277 {
3278         struct scsi_device *sdp = to_scsi_device(dev);
3279         struct scsi_disk *sdkp;
3280         struct gendisk *gd;
3281         int index;
3282         int error;
3283
3284         scsi_autopm_get_device(sdp);
3285         error = -ENODEV;
3286         if (sdp->type != TYPE_DISK &&
3287             sdp->type != TYPE_ZBC &&
3288             sdp->type != TYPE_MOD &&
3289             sdp->type != TYPE_RBC)
3290                 goto out;
3291
3292 #ifndef CONFIG_BLK_DEV_ZONED
3293         if (sdp->type == TYPE_ZBC)
3294                 goto out;
3295 #endif
3296         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3297                                         "sd_probe\n"));
3298
3299         error = -ENOMEM;
3300         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3301         if (!sdkp)
3302                 goto out;
3303
3304         gd = alloc_disk(SD_MINORS);
3305         if (!gd)
3306                 goto out_free;
3307
3308         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3309         if (index < 0) {
3310                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3311                 goto out_put;
3312         }
3313
3314         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3315         if (error) {
3316                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3317                 goto out_free_index;
3318         }
3319
3320         sdkp->device = sdp;
3321         sdkp->driver = &sd_template;
3322         sdkp->disk = gd;
3323         sdkp->index = index;
3324         atomic_set(&sdkp->openers, 0);
3325         atomic_set(&sdkp->device->ioerr_cnt, 0);
3326
3327         if (!sdp->request_queue->rq_timeout) {
3328                 if (sdp->type != TYPE_MOD)
3329                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3330                 else
3331                         blk_queue_rq_timeout(sdp->request_queue,
3332                                              SD_MOD_TIMEOUT);
3333         }
3334
3335         device_initialize(&sdkp->dev);
3336         sdkp->dev.parent = dev;
3337         sdkp->dev.class = &sd_disk_class;
3338         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3339
3340         error = device_add(&sdkp->dev);
3341         if (error)
3342                 goto out_free_index;
3343
3344         get_device(dev);
3345         dev_set_drvdata(dev, sdkp);
3346
3347         gd->major = sd_major((index & 0xf0) >> 4);
3348         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3349
3350         gd->fops = &sd_fops;
3351         gd->private_data = &sdkp->driver;
3352         gd->queue = sdkp->device->request_queue;
3353
3354         /* defaults, until the device tells us otherwise */
3355         sdp->sector_size = 512;
3356         sdkp->capacity = 0;
3357         sdkp->media_present = 1;
3358         sdkp->write_prot = 0;
3359         sdkp->cache_override = 0;
3360         sdkp->WCE = 0;
3361         sdkp->RCD = 0;
3362         sdkp->ATO = 0;
3363         sdkp->first_scan = 1;
3364         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3365
3366         sd_revalidate_disk(gd);
3367
3368         gd->flags = GENHD_FL_EXT_DEVT;
3369         if (sdp->removable) {
3370                 gd->flags |= GENHD_FL_REMOVABLE;
3371                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3372                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3373         }
3374
3375         blk_pm_runtime_init(sdp->request_queue, dev);
3376         device_add_disk(dev, gd, NULL);
3377         if (sdkp->capacity)
3378                 sd_dif_config_host(sdkp);
3379
3380         sd_revalidate_disk(gd);
3381
3382         if (sdkp->security) {
3383                 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3384                 if (sdkp->opal_dev)
3385                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3386         }
3387
3388         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3389                   sdp->removable ? "removable " : "");
3390         scsi_autopm_put_device(sdp);
3391
3392         return 0;
3393
3394  out_free_index:
3395         ida_free(&sd_index_ida, index);
3396  out_put:
3397         put_disk(gd);
3398  out_free:
3399         kfree(sdkp);
3400  out:
3401         scsi_autopm_put_device(sdp);
3402         return error;
3403 }
3404
3405 /**
3406  *      sd_remove - called whenever a scsi disk (previously recognized by
3407  *      sd_probe) is detached from the system. It is called (potentially
3408  *      multiple times) during sd module unload.
3409  *      @dev: pointer to device object
3410  *
3411  *      Note: this function is invoked from the scsi mid-level.
3412  *      This function potentially frees up a device name (e.g. /dev/sdc)
3413  *      that could be re-used by a subsequent sd_probe().
3414  *      This function is not called when the built-in sd driver is "exit-ed".
3415  **/
3416 static int sd_remove(struct device *dev)
3417 {
3418         struct scsi_disk *sdkp;
3419         dev_t devt;
3420
3421         sdkp = dev_get_drvdata(dev);
3422         devt = disk_devt(sdkp->disk);
3423         scsi_autopm_get_device(sdkp->device);
3424
3425         async_synchronize_full_domain(&scsi_sd_pm_domain);
3426         device_del(&sdkp->dev);
3427         del_gendisk(sdkp->disk);
3428         sd_shutdown(dev);
3429
3430         free_opal_dev(sdkp->opal_dev);
3431
3432         blk_register_region(devt, SD_MINORS, NULL,
3433                             sd_default_probe, NULL, NULL);
3434
3435         mutex_lock(&sd_ref_mutex);
3436         dev_set_drvdata(dev, NULL);
3437         put_device(&sdkp->dev);
3438         mutex_unlock(&sd_ref_mutex);
3439
3440         return 0;
3441 }
3442
3443 /**
3444  *      scsi_disk_release - Called to free the scsi_disk structure
3445  *      @dev: pointer to embedded class device
3446  *
3447  *      sd_ref_mutex must be held entering this routine.  Because it is
3448  *      called on last put, you should always use the scsi_disk_get()
3449  *      scsi_disk_put() helpers which manipulate the semaphore directly
3450  *      and never do a direct put_device.
3451  **/
3452 static void scsi_disk_release(struct device *dev)
3453 {
3454         struct scsi_disk *sdkp = to_scsi_disk(dev);
3455         struct gendisk *disk = sdkp->disk;
3456         struct request_queue *q = disk->queue;
3457
3458         ida_free(&sd_index_ida, sdkp->index);
3459
3460         /*
3461          * Wait until all requests that are in progress have completed.
3462          * This is necessary to avoid that e.g. scsi_end_request() crashes
3463          * due to clearing the disk->private_data pointer. Wait from inside
3464          * scsi_disk_release() instead of from sd_release() to avoid that
3465          * freezing and unfreezing the request queue affects user space I/O
3466          * in case multiple processes open a /dev/sd... node concurrently.
3467          */
3468         blk_mq_freeze_queue(q);
3469         blk_mq_unfreeze_queue(q);
3470
3471         disk->private_data = NULL;
3472         put_disk(disk);
3473         put_device(&sdkp->device->sdev_gendev);
3474
3475         kfree(sdkp);
3476 }
3477
3478 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3479 {
3480         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3481         struct scsi_sense_hdr sshdr;
3482         struct scsi_device *sdp = sdkp->device;
3483         int res;
3484
3485         if (start)
3486                 cmd[4] |= 1;    /* START */
3487
3488         if (sdp->start_stop_pwr_cond)
3489                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3490
3491         if (!scsi_device_online(sdp))
3492                 return -ENODEV;
3493
3494         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3495                         SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3496         if (res) {
3497                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3498                 if (driver_byte(res) == DRIVER_SENSE)
3499                         sd_print_sense_hdr(sdkp, &sshdr);
3500                 if (scsi_sense_valid(&sshdr) &&
3501                         /* 0x3a is medium not present */
3502                         sshdr.asc == 0x3a)
3503                         res = 0;
3504         }
3505
3506         /* SCSI error codes must not go to the generic layer */
3507         if (res)
3508                 return -EIO;
3509
3510         return 0;
3511 }
3512
3513 /*
3514  * Send a SYNCHRONIZE CACHE instruction down to the device through
3515  * the normal SCSI command structure.  Wait for the command to
3516  * complete.
3517  */
3518 static void sd_shutdown(struct device *dev)
3519 {
3520         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3521
3522         if (!sdkp)
3523                 return;         /* this can happen */
3524
3525         if (pm_runtime_suspended(dev))
3526                 return;
3527
3528         if (sdkp->WCE && sdkp->media_present) {
3529                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3530                 sd_sync_cache(sdkp, NULL);
3531         }
3532
3533         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3534                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3535                 sd_start_stop_device(sdkp, 0);
3536         }
3537 }
3538
3539 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3540 {
3541         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3542         struct scsi_sense_hdr sshdr;
3543         int ret = 0;
3544
3545         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3546                 return 0;
3547
3548         if (sdkp->WCE && sdkp->media_present) {
3549                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3550                 ret = sd_sync_cache(sdkp, &sshdr);
3551
3552                 if (ret) {
3553                         /* ignore OFFLINE device */
3554                         if (ret == -ENODEV)
3555                                 return 0;
3556
3557                         if (!scsi_sense_valid(&sshdr) ||
3558                             sshdr.sense_key != ILLEGAL_REQUEST)
3559                                 return ret;
3560
3561                         /*
3562                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3563                          * doesn't support sync. There's not much to do and
3564                          * suspend shouldn't fail.
3565                          */
3566                         ret = 0;
3567                 }
3568         }
3569
3570         if (sdkp->device->manage_start_stop) {
3571                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3572                 /* an error is not worth aborting a system sleep */
3573                 ret = sd_start_stop_device(sdkp, 0);
3574                 if (ignore_stop_errors)
3575                         ret = 0;
3576         }
3577
3578         return ret;
3579 }
3580
3581 static int sd_suspend_system(struct device *dev)
3582 {
3583         return sd_suspend_common(dev, true);
3584 }
3585
3586 static int sd_suspend_runtime(struct device *dev)
3587 {
3588         return sd_suspend_common(dev, false);
3589 }
3590
3591 static int sd_resume(struct device *dev)
3592 {
3593         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3594         int ret;
3595
3596         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3597                 return 0;
3598
3599         if (!sdkp->device->manage_start_stop)
3600                 return 0;
3601
3602         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3603         ret = sd_start_stop_device(sdkp, 1);
3604         if (!ret)
3605                 opal_unlock_from_suspend(sdkp->opal_dev);
3606         return ret;
3607 }
3608
3609 /**
3610  *      init_sd - entry point for this driver (both when built in or when
3611  *      a module).
3612  *
3613  *      Note: this function registers this driver with the scsi mid-level.
3614  **/
3615 static int __init init_sd(void)
3616 {
3617         int majors = 0, i, err;
3618
3619         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3620
3621         for (i = 0; i < SD_MAJORS; i++) {
3622                 if (register_blkdev(sd_major(i), "sd") != 0)
3623                         continue;
3624                 majors++;
3625                 blk_register_region(sd_major(i), SD_MINORS, NULL,
3626                                     sd_default_probe, NULL, NULL);
3627         }
3628
3629         if (!majors)
3630                 return -ENODEV;
3631
3632         err = class_register(&sd_disk_class);
3633         if (err)
3634                 goto err_out;
3635
3636         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3637                                          0, 0, NULL);
3638         if (!sd_cdb_cache) {
3639                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3640                 err = -ENOMEM;
3641                 goto err_out_class;
3642         }
3643
3644         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3645         if (!sd_cdb_pool) {
3646                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3647                 err = -ENOMEM;
3648                 goto err_out_cache;
3649         }
3650
3651         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3652         if (!sd_page_pool) {
3653                 printk(KERN_ERR "sd: can't init discard page pool\n");
3654                 err = -ENOMEM;
3655                 goto err_out_ppool;
3656         }
3657
3658         err = scsi_register_driver(&sd_template.gendrv);
3659         if (err)
3660                 goto err_out_driver;
3661
3662         return 0;
3663
3664 err_out_driver:
3665         mempool_destroy(sd_page_pool);
3666
3667 err_out_ppool:
3668         mempool_destroy(sd_cdb_pool);
3669
3670 err_out_cache:
3671         kmem_cache_destroy(sd_cdb_cache);
3672
3673 err_out_class:
3674         class_unregister(&sd_disk_class);
3675 err_out:
3676         for (i = 0; i < SD_MAJORS; i++)
3677                 unregister_blkdev(sd_major(i), "sd");
3678         return err;
3679 }
3680
3681 /**
3682  *      exit_sd - exit point for this driver (when it is a module).
3683  *
3684  *      Note: this function unregisters this driver from the scsi mid-level.
3685  **/
3686 static void __exit exit_sd(void)
3687 {
3688         int i;
3689
3690         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3691
3692         scsi_unregister_driver(&sd_template.gendrv);
3693         mempool_destroy(sd_cdb_pool);
3694         mempool_destroy(sd_page_pool);
3695         kmem_cache_destroy(sd_cdb_cache);
3696
3697         class_unregister(&sd_disk_class);
3698
3699         for (i = 0; i < SD_MAJORS; i++) {
3700                 blk_unregister_region(sd_major(i), SD_MINORS);
3701                 unregister_blkdev(sd_major(i), "sd");
3702         }
3703 }
3704
3705 module_init(init_sd);
3706 module_exit(exit_sd);
3707
3708 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3709                                struct scsi_sense_hdr *sshdr)
3710 {
3711         scsi_print_sense_hdr(sdkp->device,
3712                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3713 }
3714
3715 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3716                             int result)
3717 {
3718         const char *hb_string = scsi_hostbyte_string(result);
3719         const char *db_string = scsi_driverbyte_string(result);
3720
3721         if (hb_string || db_string)
3722                 sd_printk(KERN_INFO, sdkp,
3723                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3724                           hb_string ? hb_string : "invalid",
3725                           db_string ? db_string : "invalid");
3726         else
3727                 sd_printk(KERN_INFO, sdkp,
3728                           "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3729                           msg, host_byte(result), driver_byte(result));
3730 }
3731