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