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/hdreg.h>
  42 #include <linux/errno.h>
  43 #include <linux/idr.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/init.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/blkpg.h>
  48 #include <linux/blk-pm.h>
  49 #include <linux/delay.h>
  50 #include <linux/major.h>
  51 #include <linux/mutex.h>
  52 #include <linux/string_helpers.h>
  53 #include <linux/slab.h>
  54 #include <linux/sed-opal.h>
  55 #include <linux/pm_runtime.h>
  56 #include <linux/pr.h>
  57 #include <linux/t10-pi.h>
  58 #include <linux/uaccess.h>
  59 #include <asm/unaligned.h>
  60
  61 #include <scsi/scsi.h>
  62 #include <scsi/scsi_cmnd.h>
  63 #include <scsi/scsi_dbg.h>
  64 #include <scsi/scsi_device.h>
  65 #include <scsi/scsi_driver.h>
  66 #include <scsi/scsi_eh.h>
  67 #include <scsi/scsi_host.h>
  68 #include <scsi/scsi_ioctl.h>
  69 #include <scsi/scsicam.h>
  70 #include <scsi/scsi_common.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 #define SD_MINORS       16
 102
 103 static void sd_config_discard(struct scsi_disk *, unsigned int);
 104 static void sd_config_write_same(struct scsi_disk *);
 105 static int  sd_revalidate_disk(struct gendisk *);
 106 static void sd_unlock_native_capacity(struct gendisk *disk);
 107 static void sd_shutdown(struct device *);
 108 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 109 static void scsi_disk_release(struct device *cdev);
 110
 111 static DEFINE_IDA(sd_index_ida);
 112
 113 static mempool_t *sd_page_pool;
 114 static struct lock_class_key sd_bio_compl_lkclass;
 115
 116 static const char *sd_cache_types[] = {
 117         "write through", "none", "write back",
 118         "write back, no read (daft)"
 119 };
 120
 121 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 122 {
 123         bool wc = false, fua = false;
 124
 125         if (sdkp->WCE) {
 126                 wc = true;
 127                 if (sdkp->DPOFUA)
 128                         fua = true;
 129         }
 130
 131         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 132 }
 133
 134 static ssize_t
 135 cache_type_store(struct device *dev, struct device_attribute *attr,
 136                  const char *buf, size_t count)
 137 {
 138         int ct, rcd, wce, sp;
 139         struct scsi_disk *sdkp = to_scsi_disk(dev);
 140         struct scsi_device *sdp = sdkp->device;
 141         char buffer[64];
 142         char *buffer_data;
 143         struct scsi_mode_data data;
 144         struct scsi_sense_hdr sshdr;
 145         static const char temp[] = "temporary ";
 146         int len;
 147
 148         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 149                 /* no cache control on RBC devices; theoretically they
 150                  * can do it, but there's probably so many exceptions
 151                  * it's not worth the risk */
 152                 return -EINVAL;
 153
 154         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 155                 buf += sizeof(temp) - 1;
 156                 sdkp->cache_override = 1;
 157         } else {
 158                 sdkp->cache_override = 0;
 159         }
 160
 161         ct = sysfs_match_string(sd_cache_types, buf);
 162         if (ct < 0)
 163                 return -EINVAL;
 164
 165         rcd = ct & 0x01 ? 1 : 0;
 166         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 167
 168         if (sdkp->cache_override) {
 169                 sdkp->WCE = wce;
 170                 sdkp->RCD = rcd;
 171                 sd_set_flush_flag(sdkp);
 172                 return count;
 173         }
 174
 175         if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
 176                             sdkp->max_retries, &data, NULL))
 177                 return -EINVAL;
 178         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 179                   data.block_descriptor_length);
 180         buffer_data = buffer + data.header_length +
 181                 data.block_descriptor_length;
 182         buffer_data[2] &= ~0x05;
 183         buffer_data[2] |= wce << 2 | rcd;
 184         sp = buffer_data[0] & 0x80 ? 1 : 0;
 185         buffer_data[0] &= ~0x80;
 186
 187         /*
 188          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 189          * received mode parameter buffer before doing MODE SELECT.
 190          */
 191         data.device_specific = 0;
 192
 193         if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
 194                              sdkp->max_retries, &data, &sshdr)) {
 195                 if (scsi_sense_valid(&sshdr))
 196                         sd_print_sense_hdr(sdkp, &sshdr);
 197                 return -EINVAL;
 198         }
 199         sd_revalidate_disk(sdkp->disk);
 200         return count;
 201 }
 202
 203 static ssize_t
 204 manage_start_stop_show(struct device *dev,
 205                        struct device_attribute *attr, char *buf)
 206 {
 207         struct scsi_disk *sdkp = to_scsi_disk(dev);
 208         struct scsi_device *sdp = sdkp->device;
 209
 210         return sysfs_emit(buf, "%u\n",
 211                           sdp->manage_system_start_stop &&
 212                           sdp->manage_runtime_start_stop &&
 213                           sdp->manage_shutdown);
 214 }
 215 static DEVICE_ATTR_RO(manage_start_stop);
 216
 217 static ssize_t
 218 manage_system_start_stop_show(struct device *dev,
 219                               struct device_attribute *attr, char *buf)
 220 {
 221         struct scsi_disk *sdkp = to_scsi_disk(dev);
 222         struct scsi_device *sdp = sdkp->device;
 223
 224         return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
 225 }
 226
 227 static ssize_t
 228 manage_system_start_stop_store(struct device *dev,
 229                                struct device_attribute *attr,
 230                                const char *buf, size_t count)
 231 {
 232         struct scsi_disk *sdkp = to_scsi_disk(dev);
 233         struct scsi_device *sdp = sdkp->device;
 234         bool v;
 235
 236         if (!capable(CAP_SYS_ADMIN))
 237                 return -EACCES;
 238
 239         if (kstrtobool(buf, &v))
 240                 return -EINVAL;
 241
 242         sdp->manage_system_start_stop = v;
 243
 244         return count;
 245 }
 246 static DEVICE_ATTR_RW(manage_system_start_stop);
 247
 248 static ssize_t
 249 manage_runtime_start_stop_show(struct device *dev,
 250                                struct device_attribute *attr, char *buf)
 251 {
 252         struct scsi_disk *sdkp = to_scsi_disk(dev);
 253         struct scsi_device *sdp = sdkp->device;
 254
 255         return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
 256 }
 257
 258 static ssize_t
 259 manage_runtime_start_stop_store(struct device *dev,
 260                                 struct device_attribute *attr,
 261                                 const char *buf, size_t count)
 262 {
 263         struct scsi_disk *sdkp = to_scsi_disk(dev);
 264         struct scsi_device *sdp = sdkp->device;
 265         bool v;
 266
 267         if (!capable(CAP_SYS_ADMIN))
 268                 return -EACCES;
 269
 270         if (kstrtobool(buf, &v))
 271                 return -EINVAL;
 272
 273         sdp->manage_runtime_start_stop = v;
 274
 275         return count;
 276 }
 277 static DEVICE_ATTR_RW(manage_runtime_start_stop);
 278
 279 static ssize_t manage_shutdown_show(struct device *dev,
 280                                     struct device_attribute *attr, char *buf)
 281 {
 282         struct scsi_disk *sdkp = to_scsi_disk(dev);
 283         struct scsi_device *sdp = sdkp->device;
 284
 285         return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
 286 }
 287
 288 static ssize_t manage_shutdown_store(struct device *dev,
 289                                      struct device_attribute *attr,
 290                                      const char *buf, size_t count)
 291 {
 292         struct scsi_disk *sdkp = to_scsi_disk(dev);
 293         struct scsi_device *sdp = sdkp->device;
 294         bool v;
 295
 296         if (!capable(CAP_SYS_ADMIN))
 297                 return -EACCES;
 298
 299         if (kstrtobool(buf, &v))
 300                 return -EINVAL;
 301
 302         sdp->manage_shutdown = v;
 303
 304         return count;
 305 }
 306 static DEVICE_ATTR_RW(manage_shutdown);
 307
 308 static ssize_t
 309 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 310 {
 311         struct scsi_disk *sdkp = to_scsi_disk(dev);
 312
 313         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 314 }
 315
 316 static ssize_t
 317 allow_restart_store(struct device *dev, struct device_attribute *attr,
 318                     const char *buf, size_t count)
 319 {
 320         bool v;
 321         struct scsi_disk *sdkp = to_scsi_disk(dev);
 322         struct scsi_device *sdp = sdkp->device;
 323
 324         if (!capable(CAP_SYS_ADMIN))
 325                 return -EACCES;
 326
 327         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 328                 return -EINVAL;
 329
 330         if (kstrtobool(buf, &v))
 331                 return -EINVAL;
 332
 333         sdp->allow_restart = v;
 334
 335         return count;
 336 }
 337 static DEVICE_ATTR_RW(allow_restart);
 338
 339 static ssize_t
 340 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 341 {
 342         struct scsi_disk *sdkp = to_scsi_disk(dev);
 343         int ct = sdkp->RCD + 2*sdkp->WCE;
 344
 345         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 346 }
 347 static DEVICE_ATTR_RW(cache_type);
 348
 349 static ssize_t
 350 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 351 {
 352         struct scsi_disk *sdkp = to_scsi_disk(dev);
 353
 354         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 355 }
 356 static DEVICE_ATTR_RO(FUA);
 357
 358 static ssize_t
 359 protection_type_show(struct device *dev, struct device_attribute *attr,
 360                      char *buf)
 361 {
 362         struct scsi_disk *sdkp = to_scsi_disk(dev);
 363
 364         return sprintf(buf, "%u\n", sdkp->protection_type);
 365 }
 366
 367 static ssize_t
 368 protection_type_store(struct device *dev, struct device_attribute *attr,
 369                       const char *buf, size_t count)
 370 {
 371         struct scsi_disk *sdkp = to_scsi_disk(dev);
 372         unsigned int val;
 373         int err;
 374
 375         if (!capable(CAP_SYS_ADMIN))
 376                 return -EACCES;
 377
 378         err = kstrtouint(buf, 10, &val);
 379
 380         if (err)
 381                 return err;
 382
 383         if (val <= T10_PI_TYPE3_PROTECTION)
 384                 sdkp->protection_type = val;
 385
 386         return count;
 387 }
 388 static DEVICE_ATTR_RW(protection_type);
 389
 390 static ssize_t
 391 protection_mode_show(struct device *dev, struct device_attribute *attr,
 392                      char *buf)
 393 {
 394         struct scsi_disk *sdkp = to_scsi_disk(dev);
 395         struct scsi_device *sdp = sdkp->device;
 396         unsigned int dif, dix;
 397
 398         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 399         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 400
 401         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 402                 dif = 0;
 403                 dix = 1;
 404         }
 405
 406         if (!dif && !dix)
 407                 return sprintf(buf, "none\n");
 408
 409         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 410 }
 411 static DEVICE_ATTR_RO(protection_mode);
 412
 413 static ssize_t
 414 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 415 {
 416         struct scsi_disk *sdkp = to_scsi_disk(dev);
 417
 418         return sprintf(buf, "%u\n", sdkp->ATO);
 419 }
 420 static DEVICE_ATTR_RO(app_tag_own);
 421
 422 static ssize_t
 423 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 424                        char *buf)
 425 {
 426         struct scsi_disk *sdkp = to_scsi_disk(dev);
 427
 428         return sprintf(buf, "%u\n", sdkp->lbpme);
 429 }
 430 static DEVICE_ATTR_RO(thin_provisioning);
 431
 432 /* sysfs_match_string() requires dense arrays */
 433 static const char *lbp_mode[] = {
 434         [SD_LBP_FULL]           = "full",
 435         [SD_LBP_UNMAP]          = "unmap",
 436         [SD_LBP_WS16]           = "writesame_16",
 437         [SD_LBP_WS10]           = "writesame_10",
 438         [SD_LBP_ZERO]           = "writesame_zero",
 439         [SD_LBP_DISABLE]        = "disabled",
 440 };
 441
 442 static ssize_t
 443 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 444                        char *buf)
 445 {
 446         struct scsi_disk *sdkp = to_scsi_disk(dev);
 447
 448         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 449 }
 450
 451 static ssize_t
 452 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 453                         const char *buf, size_t count)
 454 {
 455         struct scsi_disk *sdkp = to_scsi_disk(dev);
 456         struct scsi_device *sdp = sdkp->device;
 457         int mode;
 458
 459         if (!capable(CAP_SYS_ADMIN))
 460                 return -EACCES;
 461
 462         if (sd_is_zoned(sdkp)) {
 463                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 464                 return count;
 465         }
 466
 467         if (sdp->type != TYPE_DISK)
 468                 return -EINVAL;
 469
 470         mode = sysfs_match_string(lbp_mode, buf);
 471         if (mode < 0)
 472                 return -EINVAL;
 473
 474         sd_config_discard(sdkp, mode);
 475
 476         return count;
 477 }
 478 static DEVICE_ATTR_RW(provisioning_mode);
 479
 480 /* sysfs_match_string() requires dense arrays */
 481 static const char *zeroing_mode[] = {
 482         [SD_ZERO_WRITE]         = "write",
 483         [SD_ZERO_WS]            = "writesame",
 484         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 485         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 486 };
 487
 488 static ssize_t
 489 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 490                   char *buf)
 491 {
 492         struct scsi_disk *sdkp = to_scsi_disk(dev);
 493
 494         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 495 }
 496
 497 static ssize_t
 498 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 499                    const char *buf, size_t count)
 500 {
 501         struct scsi_disk *sdkp = to_scsi_disk(dev);
 502         int mode;
 503
 504         if (!capable(CAP_SYS_ADMIN))
 505                 return -EACCES;
 506
 507         mode = sysfs_match_string(zeroing_mode, buf);
 508         if (mode < 0)
 509                 return -EINVAL;
 510
 511         sdkp->zeroing_mode = mode;
 512
 513         return count;
 514 }
 515 static DEVICE_ATTR_RW(zeroing_mode);
 516
 517 static ssize_t
 518 max_medium_access_timeouts_show(struct device *dev,
 519                                 struct device_attribute *attr, char *buf)
 520 {
 521         struct scsi_disk *sdkp = to_scsi_disk(dev);
 522
 523         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 524 }
 525
 526 static ssize_t
 527 max_medium_access_timeouts_store(struct device *dev,
 528                                  struct device_attribute *attr, const char *buf,
 529                                  size_t count)
 530 {
 531         struct scsi_disk *sdkp = to_scsi_disk(dev);
 532         int err;
 533
 534         if (!capable(CAP_SYS_ADMIN))
 535                 return -EACCES;
 536
 537         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 538
 539         return err ? err : count;
 540 }
 541 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 542
 543 static ssize_t
 544 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 545                            char *buf)
 546 {
 547         struct scsi_disk *sdkp = to_scsi_disk(dev);
 548
 549         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 550 }
 551
 552 static ssize_t
 553 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 554                             const char *buf, size_t count)
 555 {
 556         struct scsi_disk *sdkp = to_scsi_disk(dev);
 557         struct scsi_device *sdp = sdkp->device;
 558         unsigned long max;
 559         int err;
 560
 561         if (!capable(CAP_SYS_ADMIN))
 562                 return -EACCES;
 563
 564         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 565                 return -EINVAL;
 566
 567         err = kstrtoul(buf, 10, &max);
 568
 569         if (err)
 570                 return err;
 571
 572         if (max == 0)
 573                 sdp->no_write_same = 1;
 574         else if (max <= SD_MAX_WS16_BLOCKS) {
 575                 sdp->no_write_same = 0;
 576                 sdkp->max_ws_blocks = max;
 577         }
 578
 579         sd_config_write_same(sdkp);
 580
 581         return count;
 582 }
 583 static DEVICE_ATTR_RW(max_write_same_blocks);
 584
 585 static ssize_t
 586 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
 587 {
 588         struct scsi_disk *sdkp = to_scsi_disk(dev);
 589
 590         if (sdkp->device->type == TYPE_ZBC)
 591                 return sprintf(buf, "host-managed\n");
 592         if (sdkp->zoned == 1)
 593                 return sprintf(buf, "host-aware\n");
 594         if (sdkp->zoned == 2)
 595                 return sprintf(buf, "drive-managed\n");
 596         return sprintf(buf, "none\n");
 597 }
 598 static DEVICE_ATTR_RO(zoned_cap);
 599
 600 static ssize_t
 601 max_retries_store(struct device *dev, struct device_attribute *attr,
 602                   const char *buf, size_t count)
 603 {
 604         struct scsi_disk *sdkp = to_scsi_disk(dev);
 605         struct scsi_device *sdev = sdkp->device;
 606         int retries, err;
 607
 608         err = kstrtoint(buf, 10, &retries);
 609         if (err)
 610                 return err;
 611
 612         if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
 613                 sdkp->max_retries = retries;
 614                 return count;
 615         }
 616
 617         sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
 618                     SD_MAX_RETRIES);
 619         return -EINVAL;
 620 }
 621
 622 static ssize_t
 623 max_retries_show(struct device *dev, struct device_attribute *attr,
 624                  char *buf)
 625 {
 626         struct scsi_disk *sdkp = to_scsi_disk(dev);
 627
 628         return sprintf(buf, "%d\n", sdkp->max_retries);
 629 }
 630
 631 static DEVICE_ATTR_RW(max_retries);
 632
 633 static struct attribute *sd_disk_attrs[] = {
 634         &dev_attr_cache_type.attr,
 635         &dev_attr_FUA.attr,
 636         &dev_attr_allow_restart.attr,
 637         &dev_attr_manage_start_stop.attr,
 638         &dev_attr_manage_system_start_stop.attr,
 639         &dev_attr_manage_runtime_start_stop.attr,
 640         &dev_attr_manage_shutdown.attr,
 641         &dev_attr_protection_type.attr,
 642         &dev_attr_protection_mode.attr,
 643         &dev_attr_app_tag_own.attr,
 644         &dev_attr_thin_provisioning.attr,
 645         &dev_attr_provisioning_mode.attr,
 646         &dev_attr_zeroing_mode.attr,
 647         &dev_attr_max_write_same_blocks.attr,
 648         &dev_attr_max_medium_access_timeouts.attr,
 649         &dev_attr_zoned_cap.attr,
 650         &dev_attr_max_retries.attr,
 651         NULL,
 652 };
 653 ATTRIBUTE_GROUPS(sd_disk);
 654
 655 static struct class sd_disk_class = {
 656         .name           = "scsi_disk",
 657         .dev_release    = scsi_disk_release,
 658         .dev_groups     = sd_disk_groups,
 659 };
 660
 661 /*
 662  * Don't request a new module, as that could deadlock in multipath
 663  * environment.
 664  */
 665 static void sd_default_probe(dev_t devt)
 666 {
 667 }
 668
 669 /*
 670  * Device no to disk mapping:
 671  *
 672  *       major         disc2     disc  p1
 673  *   |............|.............|....|....| <- dev_t
 674  *    31        20 19          8 7  4 3  0
 675  *
 676  * Inside a major, we have 16k disks, however mapped non-
 677  * contiguously. The first 16 disks are for major0, the next
 678  * ones with major1, ... Disk 256 is for major0 again, disk 272
 679  * for major1, ...
 680  * As we stay compatible with our numbering scheme, we can reuse
 681  * the well-know SCSI majors 8, 65--71, 136--143.
 682  */
 683 static int sd_major(int major_idx)
 684 {
 685         switch (major_idx) {
 686         case 0:
 687                 return SCSI_DISK0_MAJOR;
 688         case 1 ... 7:
 689                 return SCSI_DISK1_MAJOR + major_idx - 1;
 690         case 8 ... 15:
 691                 return SCSI_DISK8_MAJOR + major_idx - 8;
 692         default:
 693                 BUG();
 694                 return 0;       /* shut up gcc */
 695         }
 696 }
 697
 698 #ifdef CONFIG_BLK_SED_OPAL
 699 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 700                 size_t len, bool send)
 701 {
 702         struct scsi_disk *sdkp = data;
 703         struct scsi_device *sdev = sdkp->device;
 704         u8 cdb[12] = { 0, };
 705         const struct scsi_exec_args exec_args = {
 706                 .req_flags = BLK_MQ_REQ_PM,
 707         };
 708         int ret;
 709
 710         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 711         cdb[1] = secp;
 712         put_unaligned_be16(spsp, &cdb[2]);
 713         put_unaligned_be32(len, &cdb[6]);
 714
 715         ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
 716                                buffer, len, SD_TIMEOUT, sdkp->max_retries,
 717                                &exec_args);
 718         return ret <= 0 ? ret : -EIO;
 719 }
 720 #endif /* CONFIG_BLK_SED_OPAL */
 721
 722 /*
 723  * Look up the DIX operation based on whether the command is read or
 724  * write and whether dix and dif are enabled.
 725  */
 726 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 727 {
 728         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 729         static const unsigned int ops[] = {     /* wrt dix dif */
 730                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 731                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 732                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 733                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 734                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 735                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 736                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 737                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 738         };
 739
 740         return ops[write << 2 | dix << 1 | dif];
 741 }
 742
 743 /*
 744  * Returns a mask of the protection flags that are valid for a given DIX
 745  * operation.
 746  */
 747 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 748 {
 749         static const unsigned int flag_mask[] = {
 750                 [SCSI_PROT_NORMAL]              = 0,
 751
 752                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 753                                                   SCSI_PROT_GUARD_CHECK |
 754                                                   SCSI_PROT_REF_CHECK |
 755                                                   SCSI_PROT_REF_INCREMENT,
 756
 757                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 758                                                   SCSI_PROT_IP_CHECKSUM,
 759
 760                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 761                                                   SCSI_PROT_GUARD_CHECK |
 762                                                   SCSI_PROT_REF_CHECK |
 763                                                   SCSI_PROT_REF_INCREMENT |
 764                                                   SCSI_PROT_IP_CHECKSUM,
 765
 766                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 767                                                   SCSI_PROT_REF_INCREMENT,
 768
 769                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 770                                                   SCSI_PROT_REF_CHECK |
 771                                                   SCSI_PROT_REF_INCREMENT |
 772                                                   SCSI_PROT_IP_CHECKSUM,
 773
 774                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 775                                                   SCSI_PROT_GUARD_CHECK |
 776                                                   SCSI_PROT_REF_CHECK |
 777                                                   SCSI_PROT_REF_INCREMENT |
 778                                                   SCSI_PROT_IP_CHECKSUM,
 779         };
 780
 781         return flag_mask[prot_op];
 782 }
 783
 784 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 785                                            unsigned int dix, unsigned int dif)
 786 {
 787         struct request *rq = scsi_cmd_to_rq(scmd);
 788         struct bio *bio = rq->bio;
 789         unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
 790         unsigned int protect = 0;
 791
 792         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 793                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 794                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 795
 796                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 797                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 798         }
 799
 800         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 801                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 802
 803                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 804                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 805         }
 806
 807         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 808                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 809
 810                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 811                         protect = 3 << 5;       /* Disable target PI checking */
 812                 else
 813                         protect = 1 << 5;       /* Enable target PI checking */
 814         }
 815
 816         scsi_set_prot_op(scmd, prot_op);
 817         scsi_set_prot_type(scmd, dif);
 818         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 819
 820         return protect;
 821 }
 822
 823 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 824 {
 825         struct request_queue *q = sdkp->disk->queue;
 826         unsigned int logical_block_size = sdkp->device->sector_size;
 827         unsigned int max_blocks = 0;
 828
 829         q->limits.discard_alignment =
 830                 sdkp->unmap_alignment * logical_block_size;
 831         q->limits.discard_granularity =
 832                 max(sdkp->physical_block_size,
 833                     sdkp->unmap_granularity * logical_block_size);
 834         sdkp->provisioning_mode = mode;
 835
 836         switch (mode) {
 837
 838         case SD_LBP_FULL:
 839         case SD_LBP_DISABLE:
 840                 blk_queue_max_discard_sectors(q, 0);
 841                 return;
 842
 843         case SD_LBP_UNMAP:
 844                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 845                                           (u32)SD_MAX_WS16_BLOCKS);
 846                 break;
 847
 848         case SD_LBP_WS16:
 849                 if (sdkp->device->unmap_limit_for_ws)
 850                         max_blocks = sdkp->max_unmap_blocks;
 851                 else
 852                         max_blocks = sdkp->max_ws_blocks;
 853
 854                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 855                 break;
 856
 857         case SD_LBP_WS10:
 858                 if (sdkp->device->unmap_limit_for_ws)
 859                         max_blocks = sdkp->max_unmap_blocks;
 860                 else
 861                         max_blocks = sdkp->max_ws_blocks;
 862
 863                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 864                 break;
 865
 866         case SD_LBP_ZERO:
 867                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 868                                           (u32)SD_MAX_WS10_BLOCKS);
 869                 break;
 870         }
 871
 872         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 873 }
 874
 875 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
 876 {
 877         struct page *page;
 878
 879         page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 880         if (!page)
 881                 return NULL;
 882         clear_highpage(page);
 883         bvec_set_page(&rq->special_vec, page, data_len, 0);
 884         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 885         return bvec_virt(&rq->special_vec);
 886 }
 887
 888 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 889 {
 890         struct scsi_device *sdp = cmd->device;
 891         struct request *rq = scsi_cmd_to_rq(cmd);
 892         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 893         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 894         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 895         unsigned int data_len = 24;
 896         char *buf;
 897
 898         buf = sd_set_special_bvec(rq, data_len);
 899         if (!buf)
 900                 return BLK_STS_RESOURCE;
 901
 902         cmd->cmd_len = 10;
 903         cmd->cmnd[0] = UNMAP;
 904         cmd->cmnd[8] = 24;
 905
 906         put_unaligned_be16(6 + 16, &buf[0]);
 907         put_unaligned_be16(16, &buf[2]);
 908         put_unaligned_be64(lba, &buf[8]);
 909         put_unaligned_be32(nr_blocks, &buf[16]);
 910
 911         cmd->allowed = sdkp->max_retries;
 912         cmd->transfersize = data_len;
 913         rq->timeout = SD_TIMEOUT;
 914
 915         return scsi_alloc_sgtables(cmd);
 916 }
 917
 918 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 919                 bool unmap)
 920 {
 921         struct scsi_device *sdp = cmd->device;
 922         struct request *rq = scsi_cmd_to_rq(cmd);
 923         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 924         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 925         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 926         u32 data_len = sdp->sector_size;
 927
 928         if (!sd_set_special_bvec(rq, data_len))
 929                 return BLK_STS_RESOURCE;
 930
 931         cmd->cmd_len = 16;
 932         cmd->cmnd[0] = WRITE_SAME_16;
 933         if (unmap)
 934                 cmd->cmnd[1] = 0x8; /* UNMAP */
 935         put_unaligned_be64(lba, &cmd->cmnd[2]);
 936         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 937
 938         cmd->allowed = sdkp->max_retries;
 939         cmd->transfersize = data_len;
 940         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 941
 942         return scsi_alloc_sgtables(cmd);
 943 }
 944
 945 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 946                 bool unmap)
 947 {
 948         struct scsi_device *sdp = cmd->device;
 949         struct request *rq = scsi_cmd_to_rq(cmd);
 950         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 951         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 952         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 953         u32 data_len = sdp->sector_size;
 954
 955         if (!sd_set_special_bvec(rq, data_len))
 956                 return BLK_STS_RESOURCE;
 957
 958         cmd->cmd_len = 10;
 959         cmd->cmnd[0] = WRITE_SAME;
 960         if (unmap)
 961                 cmd->cmnd[1] = 0x8; /* UNMAP */
 962         put_unaligned_be32(lba, &cmd->cmnd[2]);
 963         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 964
 965         cmd->allowed = sdkp->max_retries;
 966         cmd->transfersize = data_len;
 967         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 968
 969         return scsi_alloc_sgtables(cmd);
 970 }
 971
 972 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 973 {
 974         struct request *rq = scsi_cmd_to_rq(cmd);
 975         struct scsi_device *sdp = cmd->device;
 976         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 977         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 978         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 979
 980         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 981                 switch (sdkp->zeroing_mode) {
 982                 case SD_ZERO_WS16_UNMAP:
 983                         return sd_setup_write_same16_cmnd(cmd, true);
 984                 case SD_ZERO_WS10_UNMAP:
 985                         return sd_setup_write_same10_cmnd(cmd, true);
 986                 }
 987         }
 988
 989         if (sdp->no_write_same) {
 990                 rq->rq_flags |= RQF_QUIET;
 991                 return BLK_STS_TARGET;
 992         }
 993
 994         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 995                 return sd_setup_write_same16_cmnd(cmd, false);
 996
 997         return sd_setup_write_same10_cmnd(cmd, false);
 998 }
 999
1000 static void sd_config_write_same(struct scsi_disk *sdkp)
1001 {
1002         struct request_queue *q = sdkp->disk->queue;
1003         unsigned int logical_block_size = sdkp->device->sector_size;
1004
1005         if (sdkp->device->no_write_same) {
1006                 sdkp->max_ws_blocks = 0;
1007                 goto out;
1008         }
1009
1010         /* Some devices can not handle block counts above 0xffff despite
1011          * supporting WRITE SAME(16). Consequently we default to 64k
1012          * blocks per I/O unless the device explicitly advertises a
1013          * bigger limit.
1014          */
1015         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1016                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1017                                                    (u32)SD_MAX_WS16_BLOCKS);
1018         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1019                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1020                                                    (u32)SD_MAX_WS10_BLOCKS);
1021         else {
1022                 sdkp->device->no_write_same = 1;
1023                 sdkp->max_ws_blocks = 0;
1024         }
1025
1026         if (sdkp->lbprz && sdkp->lbpws)
1027                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1028         else if (sdkp->lbprz && sdkp->lbpws10)
1029                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1030         else if (sdkp->max_ws_blocks)
1031                 sdkp->zeroing_mode = SD_ZERO_WS;
1032         else
1033                 sdkp->zeroing_mode = SD_ZERO_WRITE;
1034
1035         if (sdkp->max_ws_blocks &&
1036             sdkp->physical_block_size > logical_block_size) {
1037                 /*
1038                  * Reporting a maximum number of blocks that is not aligned
1039                  * on the device physical size would cause a large write same
1040                  * request to be split into physically unaligned chunks by
1041                  * __blkdev_issue_write_zeroes() even if the caller of this
1042                  * functions took care to align the large request. So make sure
1043                  * the maximum reported is aligned to the device physical block
1044                  * size. This is only an optional optimization for regular
1045                  * disks, but this is mandatory to avoid failure of large write
1046                  * same requests directed at sequential write required zones of
1047                  * host-managed ZBC disks.
1048                  */
1049                 sdkp->max_ws_blocks =
1050                         round_down(sdkp->max_ws_blocks,
1051                                    bytes_to_logical(sdkp->device,
1052                                                     sdkp->physical_block_size));
1053         }
1054
1055 out:
1056         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1057                                          (logical_block_size >> 9));
1058 }
1059
1060 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1061 {
1062         struct request *rq = scsi_cmd_to_rq(cmd);
1063         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1064
1065         /* flush requests don't perform I/O, zero the S/G table */
1066         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1067
1068         if (cmd->device->use_16_for_sync) {
1069                 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1070                 cmd->cmd_len = 16;
1071         } else {
1072                 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1073                 cmd->cmd_len = 10;
1074         }
1075         cmd->transfersize = 0;
1076         cmd->allowed = sdkp->max_retries;
1077
1078         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1079         return BLK_STS_OK;
1080 }
1081
1082 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1083                                        sector_t lba, unsigned int nr_blocks,
1084                                        unsigned char flags, unsigned int dld)
1085 {
1086         cmd->cmd_len = SD_EXT_CDB_SIZE;
1087         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1088         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1089         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1090         cmd->cmnd[10] = flags;
1091         cmd->cmnd[11] = dld & 0x07;
1092         put_unaligned_be64(lba, &cmd->cmnd[12]);
1093         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1094         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1095
1096         return BLK_STS_OK;
1097 }
1098
1099 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1100                                        sector_t lba, unsigned int nr_blocks,
1101                                        unsigned char flags, unsigned int dld)
1102 {
1103         cmd->cmd_len  = 16;
1104         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1105         cmd->cmnd[1]  = flags | ((dld >> 2) & 0x01);
1106         cmd->cmnd[14] = (dld & 0x03) << 6;
1107         cmd->cmnd[15] = 0;
1108         put_unaligned_be64(lba, &cmd->cmnd[2]);
1109         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1110
1111         return BLK_STS_OK;
1112 }
1113
1114 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1115                                        sector_t lba, unsigned int nr_blocks,
1116                                        unsigned char flags)
1117 {
1118         cmd->cmd_len = 10;
1119         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1120         cmd->cmnd[1] = flags;
1121         cmd->cmnd[6] = 0;
1122         cmd->cmnd[9] = 0;
1123         put_unaligned_be32(lba, &cmd->cmnd[2]);
1124         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1125
1126         return BLK_STS_OK;
1127 }
1128
1129 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1130                                       sector_t lba, unsigned int nr_blocks,
1131                                       unsigned char flags)
1132 {
1133         /* Avoid that 0 blocks gets translated into 256 blocks. */
1134         if (WARN_ON_ONCE(nr_blocks == 0))
1135                 return BLK_STS_IOERR;
1136
1137         if (unlikely(flags & 0x8)) {
1138                 /*
1139                  * This happens only if this drive failed 10byte rw
1140                  * command with ILLEGAL_REQUEST during operation and
1141                  * thus turned off use_10_for_rw.
1142                  */
1143                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1144                 return BLK_STS_IOERR;
1145         }
1146
1147         cmd->cmd_len = 6;
1148         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1149         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1150         cmd->cmnd[2] = (lba >> 8) & 0xff;
1151         cmd->cmnd[3] = lba & 0xff;
1152         cmd->cmnd[4] = nr_blocks;
1153         cmd->cmnd[5] = 0;
1154
1155         return BLK_STS_OK;
1156 }
1157
1158 /*
1159  * Check if a command has a duration limit set. If it does, and the target
1160  * device supports CDL and the feature is enabled, return the limit
1161  * descriptor index to use. Return 0 (no limit) otherwise.
1162  */
1163 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1164 {
1165         struct scsi_device *sdp = sdkp->device;
1166         int hint;
1167
1168         if (!sdp->cdl_supported || !sdp->cdl_enable)
1169                 return 0;
1170
1171         /*
1172          * Use "no limit" if the request ioprio does not specify a duration
1173          * limit hint.
1174          */
1175         hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1176         if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1177             hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1178                 return 0;
1179
1180         return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1181 }
1182
1183 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1184 {
1185         struct request *rq = scsi_cmd_to_rq(cmd);
1186         struct scsi_device *sdp = cmd->device;
1187         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1188         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1189         sector_t threshold;
1190         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1191         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1192         bool write = rq_data_dir(rq) == WRITE;
1193         unsigned char protect, fua;
1194         unsigned int dld;
1195         blk_status_t ret;
1196         unsigned int dif;
1197         bool dix;
1198
1199         ret = scsi_alloc_sgtables(cmd);
1200         if (ret != BLK_STS_OK)
1201                 return ret;
1202
1203         ret = BLK_STS_IOERR;
1204         if (!scsi_device_online(sdp) || sdp->changed) {
1205                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1206                 goto fail;
1207         }
1208
1209         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1210                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1211                 goto fail;
1212         }
1213
1214         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1215                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1216                 goto fail;
1217         }
1218
1219         /*
1220          * Some SD card readers can't handle accesses which touch the
1221          * last one or two logical blocks. Split accesses as needed.
1222          */
1223         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1224
1225         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1226                 if (lba < threshold) {
1227                         /* Access up to the threshold but not beyond */
1228                         nr_blocks = threshold - lba;
1229                 } else {
1230                         /* Access only a single logical block */
1231                         nr_blocks = 1;
1232                 }
1233         }
1234
1235         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1236                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1237                 if (ret)
1238                         goto fail;
1239         }
1240
1241         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1242         dix = scsi_prot_sg_count(cmd);
1243         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1244         dld = sd_cdl_dld(sdkp, cmd);
1245
1246         if (dif || dix)
1247                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1248         else
1249                 protect = 0;
1250
1251         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1252                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1253                                          protect | fua, dld);
1254         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1255                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1256                                          protect | fua, dld);
1257         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1258                    sdp->use_10_for_rw || protect) {
1259                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1260                                          protect | fua);
1261         } else {
1262                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1263                                         protect | fua);
1264         }
1265
1266         if (unlikely(ret != BLK_STS_OK))
1267                 goto fail;
1268
1269         /*
1270          * We shouldn't disconnect in the middle of a sector, so with a dumb
1271          * host adapter, it's safe to assume that we can at least transfer
1272          * this many bytes between each connect / disconnect.
1273          */
1274         cmd->transfersize = sdp->sector_size;
1275         cmd->underflow = nr_blocks << 9;
1276         cmd->allowed = sdkp->max_retries;
1277         cmd->sdb.length = nr_blocks * sdp->sector_size;
1278
1279         SCSI_LOG_HLQUEUE(1,
1280                          scmd_printk(KERN_INFO, cmd,
1281                                      "%s: block=%llu, count=%d\n", __func__,
1282                                      (unsigned long long)blk_rq_pos(rq),
1283                                      blk_rq_sectors(rq)));
1284         SCSI_LOG_HLQUEUE(2,
1285                          scmd_printk(KERN_INFO, cmd,
1286                                      "%s %d/%u 512 byte blocks.\n",
1287                                      write ? "writing" : "reading", nr_blocks,
1288                                      blk_rq_sectors(rq)));
1289
1290         /*
1291          * This indicates that the command is ready from our end to be queued.
1292          */
1293         return BLK_STS_OK;
1294 fail:
1295         scsi_free_sgtables(cmd);
1296         return ret;
1297 }
1298
1299 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1300 {
1301         struct request *rq = scsi_cmd_to_rq(cmd);
1302
1303         switch (req_op(rq)) {
1304         case REQ_OP_DISCARD:
1305                 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1306                 case SD_LBP_UNMAP:
1307                         return sd_setup_unmap_cmnd(cmd);
1308                 case SD_LBP_WS16:
1309                         return sd_setup_write_same16_cmnd(cmd, true);
1310                 case SD_LBP_WS10:
1311                         return sd_setup_write_same10_cmnd(cmd, true);
1312                 case SD_LBP_ZERO:
1313                         return sd_setup_write_same10_cmnd(cmd, false);
1314                 default:
1315                         return BLK_STS_TARGET;
1316                 }
1317         case REQ_OP_WRITE_ZEROES:
1318                 return sd_setup_write_zeroes_cmnd(cmd);
1319         case REQ_OP_FLUSH:
1320                 return sd_setup_flush_cmnd(cmd);
1321         case REQ_OP_READ:
1322         case REQ_OP_WRITE:
1323         case REQ_OP_ZONE_APPEND:
1324                 return sd_setup_read_write_cmnd(cmd);
1325         case REQ_OP_ZONE_RESET:
1326                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1327                                                    false);
1328         case REQ_OP_ZONE_RESET_ALL:
1329                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1330                                                    true);
1331         case REQ_OP_ZONE_OPEN:
1332                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1333         case REQ_OP_ZONE_CLOSE:
1334                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1335         case REQ_OP_ZONE_FINISH:
1336                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1337         default:
1338                 WARN_ON_ONCE(1);
1339                 return BLK_STS_NOTSUPP;
1340         }
1341 }
1342
1343 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1344 {
1345         struct request *rq = scsi_cmd_to_rq(SCpnt);
1346
1347         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1348                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1349 }
1350
1351 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1352 {
1353         if (sdkp->device->removable || sdkp->write_prot) {
1354                 if (disk_check_media_change(disk))
1355                         return true;
1356         }
1357
1358         /*
1359          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1360          * nothing to do with partitions, BLKRRPART is used to force a full
1361          * revalidate after things like a format for historical reasons.
1362          */
1363         return test_bit(GD_NEED_PART_SCAN, &disk->state);
1364 }
1365
1366 /**
1367  *      sd_open - open a scsi disk device
1368  *      @disk: disk to open
1369  *      @mode: open mode
1370  *
1371  *      Returns 0 if successful. Returns a negated errno value in case
1372  *      of error.
1373  *
1374  *      Note: This can be called from a user context (e.g. fsck(1) )
1375  *      or from within the kernel (e.g. as a result of a mount(1) ).
1376  *      In the latter case @inode and @filp carry an abridged amount
1377  *      of information as noted above.
1378  *
1379  *      Locking: called with disk->open_mutex held.
1380  **/
1381 static int sd_open(struct gendisk *disk, blk_mode_t mode)
1382 {
1383         struct scsi_disk *sdkp = scsi_disk(disk);
1384         struct scsi_device *sdev = sdkp->device;
1385         int retval;
1386
1387         if (scsi_device_get(sdev))
1388                 return -ENXIO;
1389
1390         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1391
1392         /*
1393          * If the device is in error recovery, wait until it is done.
1394          * If the device is offline, then disallow any access to it.
1395          */
1396         retval = -ENXIO;
1397         if (!scsi_block_when_processing_errors(sdev))
1398                 goto error_out;
1399
1400         if (sd_need_revalidate(disk, sdkp))
1401                 sd_revalidate_disk(disk);
1402
1403         /*
1404          * If the drive is empty, just let the open fail.
1405          */
1406         retval = -ENOMEDIUM;
1407         if (sdev->removable && !sdkp->media_present &&
1408             !(mode & BLK_OPEN_NDELAY))
1409                 goto error_out;
1410
1411         /*
1412          * If the device has the write protect tab set, have the open fail
1413          * if the user expects to be able to write to the thing.
1414          */
1415         retval = -EROFS;
1416         if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1417                 goto error_out;
1418
1419         /*
1420          * It is possible that the disk changing stuff resulted in
1421          * the device being taken offline.  If this is the case,
1422          * report this to the user, and don't pretend that the
1423          * open actually succeeded.
1424          */
1425         retval = -ENXIO;
1426         if (!scsi_device_online(sdev))
1427                 goto error_out;
1428
1429         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1430                 if (scsi_block_when_processing_errors(sdev))
1431                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1432         }
1433
1434         return 0;
1435
1436 error_out:
1437         scsi_device_put(sdev);
1438         return retval;
1439 }
1440
1441 /**
1442  *      sd_release - invoked when the (last) close(2) is called on this
1443  *      scsi disk.
1444  *      @disk: disk to release
1445  *
1446  *      Returns 0.
1447  *
1448  *      Note: may block (uninterruptible) if error recovery is underway
1449  *      on this disk.
1450  *
1451  *      Locking: called with disk->open_mutex held.
1452  **/
1453 static void sd_release(struct gendisk *disk)
1454 {
1455         struct scsi_disk *sdkp = scsi_disk(disk);
1456         struct scsi_device *sdev = sdkp->device;
1457
1458         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1459
1460         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1461                 if (scsi_block_when_processing_errors(sdev))
1462                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1463         }
1464
1465         scsi_device_put(sdev);
1466 }
1467
1468 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1469 {
1470         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1471         struct scsi_device *sdp = sdkp->device;
1472         struct Scsi_Host *host = sdp->host;
1473         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1474         int diskinfo[4];
1475
1476         /* default to most commonly used values */
1477         diskinfo[0] = 0x40;     /* 1 << 6 */
1478         diskinfo[1] = 0x20;     /* 1 << 5 */
1479         diskinfo[2] = capacity >> 11;
1480
1481         /* override with calculated, extended default, or driver values */
1482         if (host->hostt->bios_param)
1483                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1484         else
1485                 scsicam_bios_param(bdev, capacity, diskinfo);
1486
1487         geo->heads = diskinfo[0];
1488         geo->sectors = diskinfo[1];
1489         geo->cylinders = diskinfo[2];
1490         return 0;
1491 }
1492
1493 /**
1494  *      sd_ioctl - process an ioctl
1495  *      @bdev: target block device
1496  *      @mode: open mode
1497  *      @cmd: ioctl command number
1498  *      @arg: this is third argument given to ioctl(2) system call.
1499  *      Often contains a pointer.
1500  *
1501  *      Returns 0 if successful (some ioctls return positive numbers on
1502  *      success as well). Returns a negated errno value in case of error.
1503  *
1504  *      Note: most ioctls are forward onto the block subsystem or further
1505  *      down in the scsi subsystem.
1506  **/
1507 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1508                     unsigned int cmd, unsigned long arg)
1509 {
1510         struct gendisk *disk = bdev->bd_disk;
1511         struct scsi_disk *sdkp = scsi_disk(disk);
1512         struct scsi_device *sdp = sdkp->device;
1513         void __user *p = (void __user *)arg;
1514         int error;
1515
1516         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1517                                     "cmd=0x%x\n", disk->disk_name, cmd));
1518
1519         if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1520                 return -ENOIOCTLCMD;
1521
1522         /*
1523          * If we are in the middle of error recovery, don't let anyone
1524          * else try and use this device.  Also, if error recovery fails, it
1525          * may try and take the device offline, in which case all further
1526          * access to the device is prohibited.
1527          */
1528         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1529                         (mode & BLK_OPEN_NDELAY));
1530         if (error)
1531                 return error;
1532
1533         if (is_sed_ioctl(cmd))
1534                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1535         return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1536 }
1537
1538 static void set_media_not_present(struct scsi_disk *sdkp)
1539 {
1540         if (sdkp->media_present)
1541                 sdkp->device->changed = 1;
1542
1543         if (sdkp->device->removable) {
1544                 sdkp->media_present = 0;
1545                 sdkp->capacity = 0;
1546         }
1547 }
1548
1549 static int media_not_present(struct scsi_disk *sdkp,
1550                              struct scsi_sense_hdr *sshdr)
1551 {
1552         if (!scsi_sense_valid(sshdr))
1553                 return 0;
1554
1555         /* not invoked for commands that could return deferred errors */
1556         switch (sshdr->sense_key) {
1557         case UNIT_ATTENTION:
1558         case NOT_READY:
1559                 /* medium not present */
1560                 if (sshdr->asc == 0x3A) {
1561                         set_media_not_present(sdkp);
1562                         return 1;
1563                 }
1564         }
1565         return 0;
1566 }
1567
1568 /**
1569  *      sd_check_events - check media events
1570  *      @disk: kernel device descriptor
1571  *      @clearing: disk events currently being cleared
1572  *
1573  *      Returns mask of DISK_EVENT_*.
1574  *
1575  *      Note: this function is invoked from the block subsystem.
1576  **/
1577 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1578 {
1579         struct scsi_disk *sdkp = disk->private_data;
1580         struct scsi_device *sdp;
1581         int retval;
1582         bool disk_changed;
1583
1584         if (!sdkp)
1585                 return 0;
1586
1587         sdp = sdkp->device;
1588         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1589
1590         /*
1591          * If the device is offline, don't send any commands - just pretend as
1592          * if the command failed.  If the device ever comes back online, we
1593          * can deal with it then.  It is only because of unrecoverable errors
1594          * that we would ever take a device offline in the first place.
1595          */
1596         if (!scsi_device_online(sdp)) {
1597                 set_media_not_present(sdkp);
1598                 goto out;
1599         }
1600
1601         /*
1602          * Using TEST_UNIT_READY enables differentiation between drive with
1603          * no cartridge loaded - NOT READY, drive with changed cartridge -
1604          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1605          *
1606          * Drives that auto spin down. eg iomega jaz 1G, will be started
1607          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1608          * sd_revalidate() is called.
1609          */
1610         if (scsi_block_when_processing_errors(sdp)) {
1611                 struct scsi_sense_hdr sshdr = { 0, };
1612
1613                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1614                                               &sshdr);
1615
1616                 /* failed to execute TUR, assume media not present */
1617                 if (retval < 0 || host_byte(retval)) {
1618                         set_media_not_present(sdkp);
1619                         goto out;
1620                 }
1621
1622                 if (media_not_present(sdkp, &sshdr))
1623                         goto out;
1624         }
1625
1626         /*
1627          * For removable scsi disk we have to recognise the presence
1628          * of a disk in the drive.
1629          */
1630         if (!sdkp->media_present)
1631                 sdp->changed = 1;
1632         sdkp->media_present = 1;
1633 out:
1634         /*
1635          * sdp->changed is set under the following conditions:
1636          *
1637          *      Medium present state has changed in either direction.
1638          *      Device has indicated UNIT_ATTENTION.
1639          */
1640         disk_changed = sdp->changed;
1641         sdp->changed = 0;
1642         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1643 }
1644
1645 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1646 {
1647         int retries, res;
1648         struct scsi_device *sdp = sdkp->device;
1649         const int timeout = sdp->request_queue->rq_timeout
1650                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1651         struct scsi_sense_hdr my_sshdr;
1652         const struct scsi_exec_args exec_args = {
1653                 .req_flags = BLK_MQ_REQ_PM,
1654                 /* caller might not be interested in sense, but we need it */
1655                 .sshdr = sshdr ? : &my_sshdr,
1656         };
1657
1658         if (!scsi_device_online(sdp))
1659                 return -ENODEV;
1660
1661         sshdr = exec_args.sshdr;
1662
1663         for (retries = 3; retries > 0; --retries) {
1664                 unsigned char cmd[16] = { 0 };
1665
1666                 if (sdp->use_16_for_sync)
1667                         cmd[0] = SYNCHRONIZE_CACHE_16;
1668                 else
1669                         cmd[0] = SYNCHRONIZE_CACHE;
1670                 /*
1671                  * Leave the rest of the command zero to indicate
1672                  * flush everything.
1673                  */
1674                 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1675                                        timeout, sdkp->max_retries, &exec_args);
1676                 if (res == 0)
1677                         break;
1678         }
1679
1680         if (res) {
1681                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1682
1683                 if (res < 0)
1684                         return res;
1685
1686                 if (scsi_status_is_check_condition(res) &&
1687                     scsi_sense_valid(sshdr)) {
1688                         sd_print_sense_hdr(sdkp, sshdr);
1689
1690                         /* we need to evaluate the error return  */
1691                         if (sshdr->asc == 0x3a ||       /* medium not present */
1692                             sshdr->asc == 0x20 ||       /* invalid command */
1693                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1694                                 /* this is no error here */
1695                                 return 0;
1696                 }
1697
1698                 switch (host_byte(res)) {
1699                 /* ignore errors due to racing a disconnection */
1700                 case DID_BAD_TARGET:
1701                 case DID_NO_CONNECT:
1702                         return 0;
1703                 /* signal the upper layer it might try again */
1704                 case DID_BUS_BUSY:
1705                 case DID_IMM_RETRY:
1706                 case DID_REQUEUE:
1707                 case DID_SOFT_ERROR:
1708                         return -EBUSY;
1709                 default:
1710                         return -EIO;
1711                 }
1712         }
1713         return 0;
1714 }
1715
1716 static void sd_rescan(struct device *dev)
1717 {
1718         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1719
1720         sd_revalidate_disk(sdkp->disk);
1721 }
1722
1723 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1724                 enum blk_unique_id type)
1725 {
1726         struct scsi_device *sdev = scsi_disk(disk)->device;
1727         const struct scsi_vpd *vpd;
1728         const unsigned char *d;
1729         int ret = -ENXIO, len;
1730
1731         rcu_read_lock();
1732         vpd = rcu_dereference(sdev->vpd_pg83);
1733         if (!vpd)
1734                 goto out_unlock;
1735
1736         ret = -EINVAL;
1737         for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1738                 /* we only care about designators with LU association */
1739                 if (((d[1] >> 4) & 0x3) != 0x00)
1740                         continue;
1741                 if ((d[1] & 0xf) != type)
1742                         continue;
1743
1744                 /*
1745                  * Only exit early if a 16-byte descriptor was found.  Otherwise
1746                  * keep looking as one with more entropy might still show up.
1747                  */
1748                 len = d[3];
1749                 if (len != 8 && len != 12 && len != 16)
1750                         continue;
1751                 ret = len;
1752                 memcpy(id, d + 4, len);
1753                 if (len == 16)
1754                         break;
1755         }
1756 out_unlock:
1757         rcu_read_unlock();
1758         return ret;
1759 }
1760
1761 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1762 {
1763         switch (host_byte(result)) {
1764         case DID_TRANSPORT_MARGINAL:
1765         case DID_TRANSPORT_DISRUPTED:
1766         case DID_BUS_BUSY:
1767                 return PR_STS_RETRY_PATH_FAILURE;
1768         case DID_NO_CONNECT:
1769                 return PR_STS_PATH_FAILED;
1770         case DID_TRANSPORT_FAILFAST:
1771                 return PR_STS_PATH_FAST_FAILED;
1772         }
1773
1774         switch (status_byte(result)) {
1775         case SAM_STAT_RESERVATION_CONFLICT:
1776                 return PR_STS_RESERVATION_CONFLICT;
1777         case SAM_STAT_CHECK_CONDITION:
1778                 if (!scsi_sense_valid(sshdr))
1779                         return PR_STS_IOERR;
1780
1781                 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1782                     (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1783                         return -EINVAL;
1784
1785                 fallthrough;
1786         default:
1787                 return PR_STS_IOERR;
1788         }
1789 }
1790
1791 static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1792                             unsigned char *data, int data_len)
1793 {
1794         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1795         struct scsi_device *sdev = sdkp->device;
1796         struct scsi_sense_hdr sshdr;
1797         u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1798         const struct scsi_exec_args exec_args = {
1799                 .sshdr = &sshdr,
1800         };
1801         int result;
1802
1803         put_unaligned_be16(data_len, &cmd[7]);
1804
1805         result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1806                                   SD_TIMEOUT, sdkp->max_retries, &exec_args);
1807         if (scsi_status_is_check_condition(result) &&
1808             scsi_sense_valid(&sshdr)) {
1809                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1810                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1811         }
1812
1813         if (result <= 0)
1814                 return result;
1815
1816         return sd_scsi_to_pr_err(&sshdr, result);
1817 }
1818
1819 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1820 {
1821         int result, i, data_offset, num_copy_keys;
1822         u32 num_keys = keys_info->num_keys;
1823         int data_len = num_keys * 8 + 8;
1824         u8 *data;
1825
1826         data = kzalloc(data_len, GFP_KERNEL);
1827         if (!data)
1828                 return -ENOMEM;
1829
1830         result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1831         if (result)
1832                 goto free_data;
1833
1834         keys_info->generation = get_unaligned_be32(&data[0]);
1835         keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1836
1837         data_offset = 8;
1838         num_copy_keys = min(num_keys, keys_info->num_keys);
1839
1840         for (i = 0; i < num_copy_keys; i++) {
1841                 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1842                 data_offset += 8;
1843         }
1844
1845 free_data:
1846         kfree(data);
1847         return result;
1848 }
1849
1850 static int sd_pr_read_reservation(struct block_device *bdev,
1851                                   struct pr_held_reservation *rsv)
1852 {
1853         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1854         struct scsi_device *sdev = sdkp->device;
1855         u8 data[24] = { };
1856         int result, len;
1857
1858         result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1859         if (result)
1860                 return result;
1861
1862         len = get_unaligned_be32(&data[4]);
1863         if (!len)
1864                 return 0;
1865
1866         /* Make sure we have at least the key and type */
1867         if (len < 14) {
1868                 sdev_printk(KERN_INFO, sdev,
1869                             "READ RESERVATION failed due to short return buffer of %d bytes\n",
1870                             len);
1871                 return -EINVAL;
1872         }
1873
1874         rsv->generation = get_unaligned_be32(&data[0]);
1875         rsv->key = get_unaligned_be64(&data[8]);
1876         rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1877         return 0;
1878 }
1879
1880 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1881                              u64 sa_key, enum scsi_pr_type type, u8 flags)
1882 {
1883         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1884         struct scsi_device *sdev = sdkp->device;
1885         struct scsi_sense_hdr sshdr;
1886         const struct scsi_exec_args exec_args = {
1887                 .sshdr = &sshdr,
1888         };
1889         int result;
1890         u8 cmd[16] = { 0, };
1891         u8 data[24] = { 0, };
1892
1893         cmd[0] = PERSISTENT_RESERVE_OUT;
1894         cmd[1] = sa;
1895         cmd[2] = type;
1896         put_unaligned_be32(sizeof(data), &cmd[5]);
1897
1898         put_unaligned_be64(key, &data[0]);
1899         put_unaligned_be64(sa_key, &data[8]);
1900         data[20] = flags;
1901
1902         result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1903                                   sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1904                                   &exec_args);
1905
1906         if (scsi_status_is_check_condition(result) &&
1907             scsi_sense_valid(&sshdr)) {
1908                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1909                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1910         }
1911
1912         if (result <= 0)
1913                 return result;
1914
1915         return sd_scsi_to_pr_err(&sshdr, result);
1916 }
1917
1918 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1919                 u32 flags)
1920 {
1921         if (flags & ~PR_FL_IGNORE_KEY)
1922                 return -EOPNOTSUPP;
1923         return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1924                         old_key, new_key, 0,
1925                         (1 << 0) /* APTPL */);
1926 }
1927
1928 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1929                 u32 flags)
1930 {
1931         if (flags)
1932                 return -EOPNOTSUPP;
1933         return sd_pr_out_command(bdev, 0x01, key, 0,
1934                                  block_pr_type_to_scsi(type), 0);
1935 }
1936
1937 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1938 {
1939         return sd_pr_out_command(bdev, 0x02, key, 0,
1940                                  block_pr_type_to_scsi(type), 0);
1941 }
1942
1943 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1944                 enum pr_type type, bool abort)
1945 {
1946         return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1947                                  block_pr_type_to_scsi(type), 0);
1948 }
1949
1950 static int sd_pr_clear(struct block_device *bdev, u64 key)
1951 {
1952         return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
1953 }
1954
1955 static const struct pr_ops sd_pr_ops = {
1956         .pr_register    = sd_pr_register,
1957         .pr_reserve     = sd_pr_reserve,
1958         .pr_release     = sd_pr_release,
1959         .pr_preempt     = sd_pr_preempt,
1960         .pr_clear       = sd_pr_clear,
1961         .pr_read_keys   = sd_pr_read_keys,
1962         .pr_read_reservation = sd_pr_read_reservation,
1963 };
1964
1965 static void scsi_disk_free_disk(struct gendisk *disk)
1966 {
1967         struct scsi_disk *sdkp = scsi_disk(disk);
1968
1969         put_device(&sdkp->disk_dev);
1970 }
1971
1972 static const struct block_device_operations sd_fops = {
1973         .owner                  = THIS_MODULE,
1974         .open                   = sd_open,
1975         .release                = sd_release,
1976         .ioctl                  = sd_ioctl,
1977         .getgeo                 = sd_getgeo,
1978         .compat_ioctl           = blkdev_compat_ptr_ioctl,
1979         .check_events           = sd_check_events,
1980         .unlock_native_capacity = sd_unlock_native_capacity,
1981         .report_zones           = sd_zbc_report_zones,
1982         .get_unique_id          = sd_get_unique_id,
1983         .free_disk              = scsi_disk_free_disk,
1984         .pr_ops                 = &sd_pr_ops,
1985 };
1986
1987 /**
1988  *      sd_eh_reset - reset error handling callback
1989  *      @scmd:          sd-issued command that has failed
1990  *
1991  *      This function is called by the SCSI midlayer before starting
1992  *      SCSI EH. When counting medium access failures we have to be
1993  *      careful to register it only only once per device and SCSI EH run;
1994  *      there might be several timed out commands which will cause the
1995  *      'max_medium_access_timeouts' counter to trigger after the first
1996  *      SCSI EH run already and set the device to offline.
1997  *      So this function resets the internal counter before starting SCSI EH.
1998  **/
1999 static void sd_eh_reset(struct scsi_cmnd *scmd)
2000 {
2001         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2002
2003         /* New SCSI EH run, reset gate variable */
2004         sdkp->ignore_medium_access_errors = false;
2005 }
2006
2007 /**
2008  *      sd_eh_action - error handling callback
2009  *      @scmd:          sd-issued command that has failed
2010  *      @eh_disp:       The recovery disposition suggested by the midlayer
2011  *
2012  *      This function is called by the SCSI midlayer upon completion of an
2013  *      error test command (currently TEST UNIT READY). The result of sending
2014  *      the eh command is passed in eh_disp.  We're looking for devices that
2015  *      fail medium access commands but are OK with non access commands like
2016  *      test unit ready (so wrongly see the device as having a successful
2017  *      recovery)
2018  **/
2019 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2020 {
2021         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2022         struct scsi_device *sdev = scmd->device;
2023
2024         if (!scsi_device_online(sdev) ||
2025             !scsi_medium_access_command(scmd) ||
2026             host_byte(scmd->result) != DID_TIME_OUT ||
2027             eh_disp != SUCCESS)
2028                 return eh_disp;
2029
2030         /*
2031          * The device has timed out executing a medium access command.
2032          * However, the TEST UNIT READY command sent during error
2033          * handling completed successfully. Either the device is in the
2034          * process of recovering or has it suffered an internal failure
2035          * that prevents access to the storage medium.
2036          */
2037         if (!sdkp->ignore_medium_access_errors) {
2038                 sdkp->medium_access_timed_out++;
2039                 sdkp->ignore_medium_access_errors = true;
2040         }
2041
2042         /*
2043          * If the device keeps failing read/write commands but TEST UNIT
2044          * READY always completes successfully we assume that medium
2045          * access is no longer possible and take the device offline.
2046          */
2047         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2048                 scmd_printk(KERN_ERR, scmd,
2049                             "Medium access timeout failure. Offlining disk!\n");
2050                 mutex_lock(&sdev->state_mutex);
2051                 scsi_device_set_state(sdev, SDEV_OFFLINE);
2052                 mutex_unlock(&sdev->state_mutex);
2053
2054                 return SUCCESS;
2055         }
2056
2057         return eh_disp;
2058 }
2059
2060 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2061 {
2062         struct request *req = scsi_cmd_to_rq(scmd);
2063         struct scsi_device *sdev = scmd->device;
2064         unsigned int transferred, good_bytes;
2065         u64 start_lba, end_lba, bad_lba;
2066
2067         /*
2068          * Some commands have a payload smaller than the device logical
2069          * block size (e.g. INQUIRY on a 4K disk).
2070          */
2071         if (scsi_bufflen(scmd) <= sdev->sector_size)
2072                 return 0;
2073
2074         /* Check if we have a 'bad_lba' information */
2075         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2076                                      SCSI_SENSE_BUFFERSIZE,
2077                                      &bad_lba))
2078                 return 0;
2079
2080         /*
2081          * If the bad lba was reported incorrectly, we have no idea where
2082          * the error is.
2083          */
2084         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2085         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2086         if (bad_lba < start_lba || bad_lba >= end_lba)
2087                 return 0;
2088
2089         /*
2090          * resid is optional but mostly filled in.  When it's unused,
2091          * its value is zero, so we assume the whole buffer transferred
2092          */
2093         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2094
2095         /* This computation should always be done in terms of the
2096          * resolution of the device's medium.
2097          */
2098         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2099
2100         return min(good_bytes, transferred);
2101 }
2102
2103 /**
2104  *      sd_done - bottom half handler: called when the lower level
2105  *      driver has completed (successfully or otherwise) a scsi command.
2106  *      @SCpnt: mid-level's per command structure.
2107  *
2108  *      Note: potentially run from within an ISR. Must not block.
2109  **/
2110 static int sd_done(struct scsi_cmnd *SCpnt)
2111 {
2112         int result = SCpnt->result;
2113         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2114         unsigned int sector_size = SCpnt->device->sector_size;
2115         unsigned int resid;
2116         struct scsi_sense_hdr sshdr;
2117         struct request *req = scsi_cmd_to_rq(SCpnt);
2118         struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2119         int sense_valid = 0;
2120         int sense_deferred = 0;
2121
2122         switch (req_op(req)) {
2123         case REQ_OP_DISCARD:
2124         case REQ_OP_WRITE_ZEROES:
2125         case REQ_OP_ZONE_RESET:
2126         case REQ_OP_ZONE_RESET_ALL:
2127         case REQ_OP_ZONE_OPEN:
2128         case REQ_OP_ZONE_CLOSE:
2129         case REQ_OP_ZONE_FINISH:
2130                 if (!result) {
2131                         good_bytes = blk_rq_bytes(req);
2132                         scsi_set_resid(SCpnt, 0);
2133                 } else {
2134                         good_bytes = 0;
2135                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
2136                 }
2137                 break;
2138         default:
2139                 /*
2140                  * In case of bogus fw or device, we could end up having
2141                  * an unaligned partial completion. Check this here and force
2142                  * alignment.
2143                  */
2144                 resid = scsi_get_resid(SCpnt);
2145                 if (resid & (sector_size - 1)) {
2146                         sd_printk(KERN_INFO, sdkp,
2147                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2148                                 resid, sector_size);
2149                         scsi_print_command(SCpnt);
2150                         resid = min(scsi_bufflen(SCpnt),
2151                                     round_up(resid, sector_size));
2152                         scsi_set_resid(SCpnt, resid);
2153                 }
2154         }
2155
2156         if (result) {
2157                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2158                 if (sense_valid)
2159                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2160         }
2161         sdkp->medium_access_timed_out = 0;
2162
2163         if (!scsi_status_is_check_condition(result) &&
2164             (!sense_valid || sense_deferred))
2165                 goto out;
2166
2167         switch (sshdr.sense_key) {
2168         case HARDWARE_ERROR:
2169         case MEDIUM_ERROR:
2170                 good_bytes = sd_completed_bytes(SCpnt);
2171                 break;
2172         case RECOVERED_ERROR:
2173                 good_bytes = scsi_bufflen(SCpnt);
2174                 break;
2175         case NO_SENSE:
2176                 /* This indicates a false check condition, so ignore it.  An
2177                  * unknown amount of data was transferred so treat it as an
2178                  * error.
2179                  */
2180                 SCpnt->result = 0;
2181                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2182                 break;
2183         case ABORTED_COMMAND:
2184                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2185                         good_bytes = sd_completed_bytes(SCpnt);
2186                 break;
2187         case ILLEGAL_REQUEST:
2188                 switch (sshdr.asc) {
2189                 case 0x10:      /* DIX: Host detected corruption */
2190                         good_bytes = sd_completed_bytes(SCpnt);
2191                         break;
2192                 case 0x20:      /* INVALID COMMAND OPCODE */
2193                 case 0x24:      /* INVALID FIELD IN CDB */
2194                         switch (SCpnt->cmnd[0]) {
2195                         case UNMAP:
2196                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2197                                 break;
2198                         case WRITE_SAME_16:
2199                         case WRITE_SAME:
2200                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2201                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2202                                 } else {
2203                                         sdkp->device->no_write_same = 1;
2204                                         sd_config_write_same(sdkp);
2205                                         req->rq_flags |= RQF_QUIET;
2206                                 }
2207                                 break;
2208                         }
2209                 }
2210                 break;
2211         default:
2212                 break;
2213         }
2214
2215  out:
2216         if (sd_is_zoned(sdkp))
2217                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2218
2219         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2220                                            "sd_done: completed %d of %d bytes\n",
2221                                            good_bytes, scsi_bufflen(SCpnt)));
2222
2223         return good_bytes;
2224 }
2225
2226 /*
2227  * spinup disk - called only in sd_revalidate_disk()
2228  */
2229 static void
2230 sd_spinup_disk(struct scsi_disk *sdkp)
2231 {
2232         unsigned char cmd[10];
2233         unsigned long spintime_expire = 0;
2234         int retries, spintime;
2235         unsigned int the_result;
2236         struct scsi_sense_hdr sshdr;
2237         const struct scsi_exec_args exec_args = {
2238                 .sshdr = &sshdr,
2239         };
2240         int sense_valid = 0;
2241
2242         spintime = 0;
2243
2244         /* Spin up drives, as required.  Only do this at boot time */
2245         /* Spinup needs to be done for module loads too. */
2246         do {
2247                 retries = 0;
2248
2249                 do {
2250                         bool media_was_present = sdkp->media_present;
2251
2252                         cmd[0] = TEST_UNIT_READY;
2253                         memset((void *) &cmd[1], 0, 9);
2254
2255                         the_result = scsi_execute_cmd(sdkp->device, cmd,
2256                                                       REQ_OP_DRV_IN, NULL, 0,
2257                                                       SD_TIMEOUT,
2258                                                       sdkp->max_retries,
2259                                                       &exec_args);
2260
2261                         /*
2262                          * If the drive has indicated to us that it
2263                          * doesn't have any media in it, don't bother
2264                          * with any more polling.
2265                          */
2266                         if (media_not_present(sdkp, &sshdr)) {
2267                                 if (media_was_present)
2268                                         sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2269                                 return;
2270                         }
2271
2272                         if (the_result)
2273                                 sense_valid = scsi_sense_valid(&sshdr);
2274                         retries++;
2275                 } while (retries < 3 &&
2276                          (!scsi_status_is_good(the_result) ||
2277                           (scsi_status_is_check_condition(the_result) &&
2278                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2279
2280                 if (!scsi_status_is_check_condition(the_result)) {
2281                         /* no sense, TUR either succeeded or failed
2282                          * with a status error */
2283                         if(!spintime && !scsi_status_is_good(the_result)) {
2284                                 sd_print_result(sdkp, "Test Unit Ready failed",
2285                                                 the_result);
2286                         }
2287                         break;
2288                 }
2289
2290                 /*
2291                  * The device does not want the automatic start to be issued.
2292                  */
2293                 if (sdkp->device->no_start_on_add)
2294                         break;
2295
2296                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2297                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2298                                 break;  /* manual intervention required */
2299                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2300                                 break;  /* standby */
2301                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2302                                 break;  /* unavailable */
2303                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2304                                 break;  /* sanitize in progress */
2305                         /*
2306                          * Issue command to spin up drive when not ready
2307                          */
2308                         if (!spintime) {
2309                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2310                                 cmd[0] = START_STOP;
2311                                 cmd[1] = 1;     /* Return immediately */
2312                                 memset((void *) &cmd[2], 0, 8);
2313                                 cmd[4] = 1;     /* Start spin cycle */
2314                                 if (sdkp->device->start_stop_pwr_cond)
2315                                         cmd[4] |= 1 << 4;
2316                                 scsi_execute_cmd(sdkp->device, cmd,
2317                                                  REQ_OP_DRV_IN, NULL, 0,
2318                                                  SD_TIMEOUT, sdkp->max_retries,
2319                                                  &exec_args);
2320                                 spintime_expire = jiffies + 100 * HZ;
2321                                 spintime = 1;
2322                         }
2323                         /* Wait 1 second for next try */
2324                         msleep(1000);
2325                         printk(KERN_CONT ".");
2326
2327                 /*
2328                  * Wait for USB flash devices with slow firmware.
2329                  * Yes, this sense key/ASC combination shouldn't
2330                  * occur here.  It's characteristic of these devices.
2331                  */
2332                 } else if (sense_valid &&
2333                                 sshdr.sense_key == UNIT_ATTENTION &&
2334                                 sshdr.asc == 0x28) {
2335                         if (!spintime) {
2336                                 spintime_expire = jiffies + 5 * HZ;
2337                                 spintime = 1;
2338                         }
2339                         /* Wait 1 second for next try */
2340                         msleep(1000);
2341                 } else {
2342                         /* we don't understand the sense code, so it's
2343                          * probably pointless to loop */
2344                         if(!spintime) {
2345                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2346                                 sd_print_sense_hdr(sdkp, &sshdr);
2347                         }
2348                         break;
2349                 }
2350
2351         } while (spintime && time_before_eq(jiffies, spintime_expire));
2352
2353         if (spintime) {
2354                 if (scsi_status_is_good(the_result))
2355                         printk(KERN_CONT "ready\n");
2356                 else
2357                         printk(KERN_CONT "not responding...\n");
2358         }
2359 }
2360
2361 /*
2362  * Determine whether disk supports Data Integrity Field.
2363  */
2364 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2365 {
2366         struct scsi_device *sdp = sdkp->device;
2367         u8 type;
2368
2369         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2370                 sdkp->protection_type = 0;
2371                 return 0;
2372         }
2373
2374         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2375
2376         if (type > T10_PI_TYPE3_PROTECTION) {
2377                 sd_printk(KERN_ERR, sdkp, "formatted with unsupported"  \
2378                           " protection type %u. Disabling disk!\n",
2379                           type);
2380                 sdkp->protection_type = 0;
2381                 return -ENODEV;
2382         }
2383
2384         sdkp->protection_type = type;
2385
2386         return 0;
2387 }
2388
2389 static void sd_config_protection(struct scsi_disk *sdkp)
2390 {
2391         struct scsi_device *sdp = sdkp->device;
2392
2393         sd_dif_config_host(sdkp);
2394
2395         if (!sdkp->protection_type)
2396                 return;
2397
2398         if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2399                 sd_first_printk(KERN_NOTICE, sdkp,
2400                                 "Disabling DIF Type %u protection\n",
2401                                 sdkp->protection_type);
2402                 sdkp->protection_type = 0;
2403         }
2404
2405         sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2406                         sdkp->protection_type);
2407 }
2408
2409 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2410                         struct scsi_sense_hdr *sshdr, int sense_valid,
2411                         int the_result)
2412 {
2413         if (sense_valid)
2414                 sd_print_sense_hdr(sdkp, sshdr);
2415         else
2416                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2417
2418         /*
2419          * Set dirty bit for removable devices if not ready -
2420          * sometimes drives will not report this properly.
2421          */
2422         if (sdp->removable &&
2423             sense_valid && sshdr->sense_key == NOT_READY)
2424                 set_media_not_present(sdkp);
2425
2426         /*
2427          * We used to set media_present to 0 here to indicate no media
2428          * in the drive, but some drives fail read capacity even with
2429          * media present, so we can't do that.
2430          */
2431         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2432 }
2433
2434 #define RC16_LEN 32
2435 #if RC16_LEN > SD_BUF_SIZE
2436 #error RC16_LEN must not be more than SD_BUF_SIZE
2437 #endif
2438
2439 #define READ_CAPACITY_RETRIES_ON_RESET  10
2440
2441 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2442                                                 unsigned char *buffer)
2443 {
2444         unsigned char cmd[16];
2445         struct scsi_sense_hdr sshdr;
2446         const struct scsi_exec_args exec_args = {
2447                 .sshdr = &sshdr,
2448         };
2449         int sense_valid = 0;
2450         int the_result;
2451         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2452         unsigned int alignment;
2453         unsigned long long lba;
2454         unsigned sector_size;
2455
2456         if (sdp->no_read_capacity_16)
2457                 return -EINVAL;
2458
2459         do {
2460                 memset(cmd, 0, 16);
2461                 cmd[0] = SERVICE_ACTION_IN_16;
2462                 cmd[1] = SAI_READ_CAPACITY_16;
2463                 cmd[13] = RC16_LEN;
2464                 memset(buffer, 0, RC16_LEN);
2465
2466                 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2467                                               buffer, RC16_LEN, SD_TIMEOUT,
2468                                               sdkp->max_retries, &exec_args);
2469
2470                 if (media_not_present(sdkp, &sshdr))
2471                         return -ENODEV;
2472
2473                 if (the_result > 0) {
2474                         sense_valid = scsi_sense_valid(&sshdr);
2475                         if (sense_valid &&
2476                             sshdr.sense_key == ILLEGAL_REQUEST &&
2477                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2478                             sshdr.ascq == 0x00)
2479                                 /* Invalid Command Operation Code or
2480                                  * Invalid Field in CDB, just retry
2481                                  * silently with RC10 */
2482                                 return -EINVAL;
2483                         if (sense_valid &&
2484                             sshdr.sense_key == UNIT_ATTENTION &&
2485                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2486                                 /* Device reset might occur several times,
2487                                  * give it one more chance */
2488                                 if (--reset_retries > 0)
2489                                         continue;
2490                 }
2491                 retries--;
2492
2493         } while (the_result && retries);
2494
2495         if (the_result) {
2496                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2497                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2498                 return -EINVAL;
2499         }
2500
2501         sector_size = get_unaligned_be32(&buffer[8]);
2502         lba = get_unaligned_be64(&buffer[0]);
2503
2504         if (sd_read_protection_type(sdkp, buffer) < 0) {
2505                 sdkp->capacity = 0;
2506                 return -ENODEV;
2507         }
2508
2509         /* Logical blocks per physical block exponent */
2510         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2511
2512         /* RC basis */
2513         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2514
2515         /* Lowest aligned logical block */
2516         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2517         blk_queue_alignment_offset(sdp->request_queue, alignment);
2518         if (alignment && sdkp->first_scan)
2519                 sd_printk(KERN_NOTICE, sdkp,
2520                           "physical block alignment offset: %u\n", alignment);
2521
2522         if (buffer[14] & 0x80) { /* LBPME */
2523                 sdkp->lbpme = 1;
2524
2525                 if (buffer[14] & 0x40) /* LBPRZ */
2526                         sdkp->lbprz = 1;
2527
2528                 sd_config_discard(sdkp, SD_LBP_WS16);
2529         }
2530
2531         sdkp->capacity = lba + 1;
2532         return sector_size;
2533 }
2534
2535 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2536                                                 unsigned char *buffer)
2537 {
2538         unsigned char cmd[16];
2539         struct scsi_sense_hdr sshdr;
2540         const struct scsi_exec_args exec_args = {
2541                 .sshdr = &sshdr,
2542         };
2543         int sense_valid = 0;
2544         int the_result;
2545         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2546         sector_t lba;
2547         unsigned sector_size;
2548
2549         do {
2550                 cmd[0] = READ_CAPACITY;
2551                 memset(&cmd[1], 0, 9);
2552                 memset(buffer, 0, 8);
2553
2554                 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2555                                               8, SD_TIMEOUT, sdkp->max_retries,
2556                                               &exec_args);
2557
2558                 if (media_not_present(sdkp, &sshdr))
2559                         return -ENODEV;
2560
2561                 if (the_result > 0) {
2562                         sense_valid = scsi_sense_valid(&sshdr);
2563                         if (sense_valid &&
2564                             sshdr.sense_key == UNIT_ATTENTION &&
2565                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2566                                 /* Device reset might occur several times,
2567                                  * give it one more chance */
2568                                 if (--reset_retries > 0)
2569                                         continue;
2570                 }
2571                 retries--;
2572
2573         } while (the_result && retries);
2574
2575         if (the_result) {
2576                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2577                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2578                 return -EINVAL;
2579         }
2580
2581         sector_size = get_unaligned_be32(&buffer[4]);
2582         lba = get_unaligned_be32(&buffer[0]);
2583
2584         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2585                 /* Some buggy (usb cardreader) devices return an lba of
2586                    0xffffffff when the want to report a size of 0 (with
2587                    which they really mean no media is present) */
2588                 sdkp->capacity = 0;
2589                 sdkp->physical_block_size = sector_size;
2590                 return sector_size;
2591         }
2592
2593         sdkp->capacity = lba + 1;
2594         sdkp->physical_block_size = sector_size;
2595         return sector_size;
2596 }
2597
2598 static int sd_try_rc16_first(struct scsi_device *sdp)
2599 {
2600         if (sdp->host->max_cmd_len < 16)
2601                 return 0;
2602         if (sdp->try_rc_10_first)
2603                 return 0;
2604         if (sdp->scsi_level > SCSI_SPC_2)
2605                 return 1;
2606         if (scsi_device_protection(sdp))
2607                 return 1;
2608         return 0;
2609 }
2610
2611 /*
2612  * read disk capacity
2613  */
2614 static void
2615 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2616 {
2617         int sector_size;
2618         struct scsi_device *sdp = sdkp->device;
2619
2620         if (sd_try_rc16_first(sdp)) {
2621                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2622                 if (sector_size == -EOVERFLOW)
2623                         goto got_data;
2624                 if (sector_size == -ENODEV)
2625                         return;
2626                 if (sector_size < 0)
2627                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2628                 if (sector_size < 0)
2629                         return;
2630         } else {
2631                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2632                 if (sector_size == -EOVERFLOW)
2633                         goto got_data;
2634                 if (sector_size < 0)
2635                         return;
2636                 if ((sizeof(sdkp->capacity) > 4) &&
2637                     (sdkp->capacity > 0xffffffffULL)) {
2638                         int old_sector_size = sector_size;
2639                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2640                                         "Trying to use READ CAPACITY(16).\n");
2641                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2642                         if (sector_size < 0) {
2643                                 sd_printk(KERN_NOTICE, sdkp,
2644                                         "Using 0xffffffff as device size\n");
2645                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2646                                 sector_size = old_sector_size;
2647                                 goto got_data;
2648                         }
2649                         /* Remember that READ CAPACITY(16) succeeded */
2650                         sdp->try_rc_10_first = 0;
2651                 }
2652         }
2653
2654         /* Some devices are known to return the total number of blocks,
2655          * not the highest block number.  Some devices have versions
2656          * which do this and others which do not.  Some devices we might
2657          * suspect of doing this but we don't know for certain.
2658          *
2659          * If we know the reported capacity is wrong, decrement it.  If
2660          * we can only guess, then assume the number of blocks is even
2661          * (usually true but not always) and err on the side of lowering
2662          * the capacity.
2663          */
2664         if (sdp->fix_capacity ||
2665             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2666                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2667                                 "from its reported value: %llu\n",
2668                                 (unsigned long long) sdkp->capacity);
2669                 --sdkp->capacity;
2670         }
2671
2672 got_data:
2673         if (sector_size == 0) {
2674                 sector_size = 512;
2675                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2676                           "assuming 512.\n");
2677         }
2678
2679         if (sector_size != 512 &&
2680             sector_size != 1024 &&
2681             sector_size != 2048 &&
2682             sector_size != 4096) {
2683                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2684                           sector_size);
2685                 /*
2686                  * The user might want to re-format the drive with
2687                  * a supported sectorsize.  Once this happens, it
2688                  * would be relatively trivial to set the thing up.
2689                  * For this reason, we leave the thing in the table.
2690                  */
2691                 sdkp->capacity = 0;
2692                 /*
2693                  * set a bogus sector size so the normal read/write
2694                  * logic in the block layer will eventually refuse any
2695                  * request on this device without tripping over power
2696                  * of two sector size assumptions
2697                  */
2698                 sector_size = 512;
2699         }
2700         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2701         blk_queue_physical_block_size(sdp->request_queue,
2702                                       sdkp->physical_block_size);
2703         sdkp->device->sector_size = sector_size;
2704
2705         if (sdkp->capacity > 0xffffffff)
2706                 sdp->use_16_for_rw = 1;
2707
2708 }
2709
2710 /*
2711  * Print disk capacity
2712  */
2713 static void
2714 sd_print_capacity(struct scsi_disk *sdkp,
2715                   sector_t old_capacity)
2716 {
2717         int sector_size = sdkp->device->sector_size;
2718         char cap_str_2[10], cap_str_10[10];
2719
2720         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2721                 return;
2722
2723         string_get_size(sdkp->capacity, sector_size,
2724                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2725         string_get_size(sdkp->capacity, sector_size,
2726                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2727
2728         sd_printk(KERN_NOTICE, sdkp,
2729                   "%llu %d-byte logical blocks: (%s/%s)\n",
2730                   (unsigned long long)sdkp->capacity,
2731                   sector_size, cap_str_10, cap_str_2);
2732
2733         if (sdkp->physical_block_size != sector_size)
2734                 sd_printk(KERN_NOTICE, sdkp,
2735                           "%u-byte physical blocks\n",
2736                           sdkp->physical_block_size);
2737 }
2738
2739 /* called with buffer of length 512 */
2740 static inline int
2741 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2742                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2743                  struct scsi_sense_hdr *sshdr)
2744 {
2745         /*
2746          * If we must use MODE SENSE(10), make sure that the buffer length
2747          * is at least 8 bytes so that the mode sense header fits.
2748          */
2749         if (sdkp->device->use_10_for_ms && len < 8)
2750                 len = 8;
2751
2752         return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2753                                SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2754 }
2755
2756 /*
2757  * read write protect setting, if possible - called only in sd_revalidate_disk()
2758  * called with buffer of length SD_BUF_SIZE
2759  */
2760 static void
2761 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2762 {
2763         int res;
2764         struct scsi_device *sdp = sdkp->device;
2765         struct scsi_mode_data data;
2766         int old_wp = sdkp->write_prot;
2767
2768         set_disk_ro(sdkp->disk, 0);
2769         if (sdp->skip_ms_page_3f) {
2770                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2771                 return;
2772         }
2773
2774         if (sdp->use_192_bytes_for_3f) {
2775                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2776         } else {
2777                 /*
2778                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2779                  * We have to start carefully: some devices hang if we ask
2780                  * for more than is available.
2781                  */
2782                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2783
2784                 /*
2785                  * Second attempt: ask for page 0 When only page 0 is
2786                  * implemented, a request for page 3F may return Sense Key
2787                  * 5: Illegal Request, Sense Code 24: Invalid field in
2788                  * CDB.
2789                  */
2790                 if (res < 0)
2791                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2792
2793                 /*
2794                  * Third attempt: ask 255 bytes, as we did earlier.
2795                  */
2796                 if (res < 0)
2797                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2798                                                &data, NULL);
2799         }
2800
2801         if (res < 0) {
2802                 sd_first_printk(KERN_WARNING, sdkp,
2803                           "Test WP failed, assume Write Enabled\n");
2804         } else {
2805                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2806                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2807                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2808                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2809                                   sdkp->write_prot ? "on" : "off");
2810                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2811                 }
2812         }
2813 }
2814
2815 /*
2816  * sd_read_cache_type - called only from sd_revalidate_disk()
2817  * called with buffer of length SD_BUF_SIZE
2818  */
2819 static void
2820 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2821 {
2822         int len = 0, res;
2823         struct scsi_device *sdp = sdkp->device;
2824
2825         int dbd;
2826         int modepage;
2827         int first_len;
2828         struct scsi_mode_data data;
2829         struct scsi_sense_hdr sshdr;
2830         int old_wce = sdkp->WCE;
2831         int old_rcd = sdkp->RCD;
2832         int old_dpofua = sdkp->DPOFUA;
2833
2834
2835         if (sdkp->cache_override)
2836                 return;
2837
2838         first_len = 4;
2839         if (sdp->skip_ms_page_8) {
2840                 if (sdp->type == TYPE_RBC)
2841                         goto defaults;
2842                 else {
2843                         if (sdp->skip_ms_page_3f)
2844                                 goto defaults;
2845                         modepage = 0x3F;
2846                         if (sdp->use_192_bytes_for_3f)
2847                                 first_len = 192;
2848                         dbd = 0;
2849                 }
2850         } else if (sdp->type == TYPE_RBC) {
2851                 modepage = 6;
2852                 dbd = 8;
2853         } else {
2854                 modepage = 8;
2855                 dbd = 0;
2856         }
2857
2858         /* cautiously ask */
2859         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2860                         &data, &sshdr);
2861
2862         if (res < 0)
2863                 goto bad_sense;
2864
2865         if (!data.header_length) {
2866                 modepage = 6;
2867                 first_len = 0;
2868                 sd_first_printk(KERN_ERR, sdkp,
2869                                 "Missing header in MODE_SENSE response\n");
2870         }
2871
2872         /* that went OK, now ask for the proper length */
2873         len = data.length;
2874
2875         /*
2876          * We're only interested in the first three bytes, actually.
2877          * But the data cache page is defined for the first 20.
2878          */
2879         if (len < 3)
2880                 goto bad_sense;
2881         else if (len > SD_BUF_SIZE) {
2882                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2883                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2884                 len = SD_BUF_SIZE;
2885         }
2886         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2887                 len = 192;
2888
2889         /* Get the data */
2890         if (len > first_len)
2891                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2892                                 &data, &sshdr);
2893
2894         if (!res) {
2895                 int offset = data.header_length + data.block_descriptor_length;
2896
2897                 while (offset < len) {
2898                         u8 page_code = buffer[offset] & 0x3F;
2899                         u8 spf       = buffer[offset] & 0x40;
2900
2901                         if (page_code == 8 || page_code == 6) {
2902                                 /* We're interested only in the first 3 bytes.
2903                                  */
2904                                 if (len - offset <= 2) {
2905                                         sd_first_printk(KERN_ERR, sdkp,
2906                                                 "Incomplete mode parameter "
2907                                                         "data\n");
2908                                         goto defaults;
2909                                 } else {
2910                                         modepage = page_code;
2911                                         goto Page_found;
2912                                 }
2913                         } else {
2914                                 /* Go to the next page */
2915                                 if (spf && len - offset > 3)
2916                                         offset += 4 + (buffer[offset+2] << 8) +
2917                                                 buffer[offset+3];
2918                                 else if (!spf && len - offset > 1)
2919                                         offset += 2 + buffer[offset+1];
2920                                 else {
2921                                         sd_first_printk(KERN_ERR, sdkp,
2922                                                         "Incomplete mode "
2923                                                         "parameter data\n");
2924                                         goto defaults;
2925                                 }
2926                         }
2927                 }
2928
2929                 sd_first_printk(KERN_WARNING, sdkp,
2930                                 "No Caching mode page found\n");
2931                 goto defaults;
2932
2933         Page_found:
2934                 if (modepage == 8) {
2935                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2936                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2937                 } else {
2938                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2939                         sdkp->RCD = 0;
2940                 }
2941
2942                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2943                 if (sdp->broken_fua) {
2944                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2945                         sdkp->DPOFUA = 0;
2946                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2947                            !sdkp->device->use_16_for_rw) {
2948                         sd_first_printk(KERN_NOTICE, sdkp,
2949                                   "Uses READ/WRITE(6), disabling FUA\n");
2950                         sdkp->DPOFUA = 0;
2951                 }
2952
2953                 /* No cache flush allowed for write protected devices */
2954                 if (sdkp->WCE && sdkp->write_prot)
2955                         sdkp->WCE = 0;
2956
2957                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2958                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2959                         sd_printk(KERN_NOTICE, sdkp,
2960                                   "Write cache: %s, read cache: %s, %s\n",
2961                                   sdkp->WCE ? "enabled" : "disabled",
2962                                   sdkp->RCD ? "disabled" : "enabled",
2963                                   sdkp->DPOFUA ? "supports DPO and FUA"
2964                                   : "doesn't support DPO or FUA");
2965
2966                 return;
2967         }
2968
2969 bad_sense:
2970         if (scsi_sense_valid(&sshdr) &&
2971             sshdr.sense_key == ILLEGAL_REQUEST &&
2972             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2973                 /* Invalid field in CDB */
2974                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2975         else
2976                 sd_first_printk(KERN_ERR, sdkp,
2977                                 "Asking for cache data failed\n");
2978
2979 defaults:
2980         if (sdp->wce_default_on) {
2981                 sd_first_printk(KERN_NOTICE, sdkp,
2982                                 "Assuming drive cache: write back\n");
2983                 sdkp->WCE = 1;
2984         } else {
2985                 sd_first_printk(KERN_WARNING, sdkp,
2986                                 "Assuming drive cache: write through\n");
2987                 sdkp->WCE = 0;
2988         }
2989         sdkp->RCD = 0;
2990         sdkp->DPOFUA = 0;
2991 }
2992
2993 /*
2994  * The ATO bit indicates whether the DIF application tag is available
2995  * for use by the operating system.
2996  */
2997 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2998 {
2999         int res, offset;
3000         struct scsi_device *sdp = sdkp->device;
3001         struct scsi_mode_data data;
3002         struct scsi_sense_hdr sshdr;
3003
3004         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3005                 return;
3006
3007         if (sdkp->protection_type == 0)
3008                 return;
3009
3010         res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3011                               sdkp->max_retries, &data, &sshdr);
3012
3013         if (res < 0 || !data.header_length ||
3014             data.length < 6) {
3015                 sd_first_printk(KERN_WARNING, sdkp,
3016                           "getting Control mode page failed, assume no ATO\n");
3017
3018                 if (scsi_sense_valid(&sshdr))
3019                         sd_print_sense_hdr(sdkp, &sshdr);
3020
3021                 return;
3022         }
3023
3024         offset = data.header_length + data.block_descriptor_length;
3025
3026         if ((buffer[offset] & 0x3f) != 0x0a) {
3027                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3028                 return;
3029         }
3030
3031         if ((buffer[offset + 5] & 0x80) == 0)
3032                 return;
3033
3034         sdkp->ATO = 1;
3035
3036         return;
3037 }
3038
3039 /**
3040  * sd_read_block_limits - Query disk device for preferred I/O sizes.
3041  * @sdkp: disk to query
3042  */
3043 static void sd_read_block_limits(struct scsi_disk *sdkp)
3044 {
3045         struct scsi_vpd *vpd;
3046
3047         rcu_read_lock();
3048
3049         vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3050         if (!vpd || vpd->len < 16)
3051                 goto out;
3052
3053         sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3054         sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3055         sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3056
3057         if (vpd->len >= 64) {
3058                 unsigned int lba_count, desc_count;
3059
3060                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3061
3062                 if (!sdkp->lbpme)
3063                         goto out;
3064
3065                 lba_count = get_unaligned_be32(&vpd->data[20]);
3066                 desc_count = get_unaligned_be32(&vpd->data[24]);
3067
3068                 if (lba_count && desc_count)
3069                         sdkp->max_unmap_blocks = lba_count;
3070
3071                 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3072
3073                 if (vpd->data[32] & 0x80)
3074                         sdkp->unmap_alignment =
3075                                 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3076
3077                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3078
3079                         if (sdkp->max_unmap_blocks)
3080                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
3081                         else
3082                                 sd_config_discard(sdkp, SD_LBP_WS16);
3083
3084                 } else {        /* LBP VPD page tells us what to use */
3085                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
3086                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
3087                         else if (sdkp->lbpws)
3088                                 sd_config_discard(sdkp, SD_LBP_WS16);
3089                         else if (sdkp->lbpws10)
3090                                 sd_config_discard(sdkp, SD_LBP_WS10);
3091                         else
3092                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
3093                 }
3094         }
3095
3096  out:
3097         rcu_read_unlock();
3098 }
3099
3100 /**
3101  * sd_read_block_characteristics - Query block dev. characteristics
3102  * @sdkp: disk to query
3103  */
3104 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3105 {
3106         struct request_queue *q = sdkp->disk->queue;
3107         struct scsi_vpd *vpd;
3108         u16 rot;
3109         u8 zoned;
3110
3111         rcu_read_lock();
3112         vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3113
3114         if (!vpd || vpd->len < 8) {
3115                 rcu_read_unlock();
3116                 return;
3117         }
3118
3119         rot = get_unaligned_be16(&vpd->data[4]);
3120         zoned = (vpd->data[8] >> 4) & 3;
3121         rcu_read_unlock();
3122
3123         if (rot == 1) {
3124                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3125                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3126         }
3127
3128         if (sdkp->device->type == TYPE_ZBC) {
3129                 /*
3130                  * Host-managed: Per ZBC and ZAC specifications, writes in
3131                  * sequential write required zones of host-managed devices must
3132                  * be aligned to the device physical block size.
3133                  */
3134                 disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
3135                 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3136         } else {
3137                 sdkp->zoned = zoned;
3138                 if (sdkp->zoned == 1) {
3139                         /* Host-aware */
3140                         disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
3141                 } else {
3142                         /* Regular disk or drive managed disk */
3143                         disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3144                 }
3145         }
3146
3147         if (!sdkp->first_scan)
3148                 return;
3149
3150         if (blk_queue_is_zoned(q)) {
3151                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3152                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3153         } else {
3154                 if (sdkp->zoned == 1)
3155                         sd_printk(KERN_NOTICE, sdkp,
3156                                   "Host-aware SMR disk used as regular disk\n");
3157                 else if (sdkp->zoned == 2)
3158                         sd_printk(KERN_NOTICE, sdkp,
3159                                   "Drive-managed SMR disk\n");
3160         }
3161 }
3162
3163 /**
3164  * sd_read_block_provisioning - Query provisioning VPD page
3165  * @sdkp: disk to query
3166  */
3167 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3168 {
3169         struct scsi_vpd *vpd;
3170
3171         if (sdkp->lbpme == 0)
3172                 return;
3173
3174         rcu_read_lock();
3175         vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3176
3177         if (!vpd || vpd->len < 8) {
3178                 rcu_read_unlock();
3179                 return;
3180         }
3181
3182         sdkp->lbpvpd    = 1;
3183         sdkp->lbpu      = (vpd->data[5] >> 7) & 1; /* UNMAP */
3184         sdkp->lbpws     = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3185         sdkp->lbpws10   = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3186         rcu_read_unlock();
3187 }
3188
3189 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3190 {
3191         struct scsi_device *sdev = sdkp->device;
3192
3193         if (sdev->host->no_write_same) {
3194                 sdev->no_write_same = 1;
3195
3196                 return;
3197         }
3198
3199         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3200                 struct scsi_vpd *vpd;
3201
3202                 sdev->no_report_opcodes = 1;
3203
3204                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3205                  * CODES is unsupported and the device has an ATA
3206                  * Information VPD page (SAT).
3207                  */
3208                 rcu_read_lock();
3209                 vpd = rcu_dereference(sdev->vpd_pg89);
3210                 if (vpd)
3211                         sdev->no_write_same = 1;
3212                 rcu_read_unlock();
3213         }
3214
3215         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3216                 sdkp->ws16 = 1;
3217
3218         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3219                 sdkp->ws10 = 1;
3220 }
3221
3222 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3223 {
3224         struct scsi_device *sdev = sdkp->device;
3225
3226         if (!sdev->security_supported)
3227                 return;
3228
3229         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3230                         SECURITY_PROTOCOL_IN, 0) == 1 &&
3231             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3232                         SECURITY_PROTOCOL_OUT, 0) == 1)
3233                 sdkp->security = 1;
3234 }
3235
3236 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3237 {
3238         return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3239 }
3240
3241 /**
3242  * sd_read_cpr - Query concurrent positioning ranges
3243  * @sdkp:       disk to query
3244  */
3245 static void sd_read_cpr(struct scsi_disk *sdkp)
3246 {
3247         struct blk_independent_access_ranges *iars = NULL;
3248         unsigned char *buffer = NULL;
3249         unsigned int nr_cpr = 0;
3250         int i, vpd_len, buf_len = SD_BUF_SIZE;
3251         u8 *desc;
3252
3253         /*
3254          * We need to have the capacity set first for the block layer to be
3255          * able to check the ranges.
3256          */
3257         if (sdkp->first_scan)
3258                 return;
3259
3260         if (!sdkp->capacity)
3261                 goto out;
3262
3263         /*
3264          * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3265          * leading to a maximum page size of 64 + 256*32 bytes.
3266          */
3267         buf_len = 64 + 256*32;
3268         buffer = kmalloc(buf_len, GFP_KERNEL);
3269         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3270                 goto out;
3271
3272         /* We must have at least a 64B header and one 32B range descriptor */
3273         vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3274         if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3275                 sd_printk(KERN_ERR, sdkp,
3276                           "Invalid Concurrent Positioning Ranges VPD page\n");
3277                 goto out;
3278         }
3279
3280         nr_cpr = (vpd_len - 64) / 32;
3281         if (nr_cpr == 1) {
3282                 nr_cpr = 0;
3283                 goto out;
3284         }
3285
3286         iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3287         if (!iars) {
3288                 nr_cpr = 0;
3289                 goto out;
3290         }
3291
3292         desc = &buffer[64];
3293         for (i = 0; i < nr_cpr; i++, desc += 32) {
3294                 if (desc[0] != i) {
3295                         sd_printk(KERN_ERR, sdkp,
3296                                 "Invalid Concurrent Positioning Range number\n");
3297                         nr_cpr = 0;
3298                         break;
3299                 }
3300
3301                 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3302                 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3303         }
3304
3305 out:
3306         disk_set_independent_access_ranges(sdkp->disk, iars);
3307         if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3308                 sd_printk(KERN_NOTICE, sdkp,
3309                           "%u concurrent positioning ranges\n", nr_cpr);
3310                 sdkp->nr_actuators = nr_cpr;
3311         }
3312
3313         kfree(buffer);
3314 }
3315
3316 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3317 {
3318         struct scsi_device *sdp = sdkp->device;
3319         unsigned int min_xfer_bytes =
3320                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3321
3322         if (sdkp->min_xfer_blocks == 0)
3323                 return false;
3324
3325         if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3326                 sd_first_printk(KERN_WARNING, sdkp,
3327                                 "Preferred minimum I/O size %u bytes not a " \
3328                                 "multiple of physical block size (%u bytes)\n",
3329                                 min_xfer_bytes, sdkp->physical_block_size);
3330                 sdkp->min_xfer_blocks = 0;
3331                 return false;
3332         }
3333
3334         sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3335                         min_xfer_bytes);
3336         return true;
3337 }
3338
3339 /*
3340  * Determine the device's preferred I/O size for reads and writes
3341  * unless the reported value is unreasonably small, large, not a
3342  * multiple of the physical block size, or simply garbage.
3343  */
3344 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3345                                       unsigned int dev_max)
3346 {
3347         struct scsi_device *sdp = sdkp->device;
3348         unsigned int opt_xfer_bytes =
3349                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3350         unsigned int min_xfer_bytes =
3351                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3352
3353         if (sdkp->opt_xfer_blocks == 0)
3354                 return false;
3355
3356         if (sdkp->opt_xfer_blocks > dev_max) {
3357                 sd_first_printk(KERN_WARNING, sdkp,
3358                                 "Optimal transfer size %u logical blocks " \
3359                                 "> dev_max (%u logical blocks)\n",
3360                                 sdkp->opt_xfer_blocks, dev_max);
3361                 return false;
3362         }
3363
3364         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3365                 sd_first_printk(KERN_WARNING, sdkp,
3366                                 "Optimal transfer size %u logical blocks " \
3367                                 "> sd driver limit (%u logical blocks)\n",
3368                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3369                 return false;
3370         }
3371
3372         if (opt_xfer_bytes < PAGE_SIZE) {
3373                 sd_first_printk(KERN_WARNING, sdkp,
3374                                 "Optimal transfer size %u bytes < " \
3375                                 "PAGE_SIZE (%u bytes)\n",
3376                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3377                 return false;
3378         }
3379
3380         if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3381                 sd_first_printk(KERN_WARNING, sdkp,
3382                                 "Optimal transfer size %u bytes not a " \
3383                                 "multiple of preferred minimum block " \
3384                                 "size (%u bytes)\n",
3385                                 opt_xfer_bytes, min_xfer_bytes);
3386                 return false;
3387         }
3388
3389         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3390                 sd_first_printk(KERN_WARNING, sdkp,
3391                                 "Optimal transfer size %u bytes not a " \
3392                                 "multiple of physical block size (%u bytes)\n",
3393                                 opt_xfer_bytes, sdkp->physical_block_size);
3394                 return false;
3395         }
3396
3397         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3398                         opt_xfer_bytes);
3399         return true;
3400 }
3401
3402 /**
3403  *      sd_revalidate_disk - called the first time a new disk is seen,
3404  *      performs disk spin up, read_capacity, etc.
3405  *      @disk: struct gendisk we care about
3406  **/
3407 static int sd_revalidate_disk(struct gendisk *disk)
3408 {
3409         struct scsi_disk *sdkp = scsi_disk(disk);
3410         struct scsi_device *sdp = sdkp->device;
3411         struct request_queue *q = sdkp->disk->queue;
3412         sector_t old_capacity = sdkp->capacity;
3413         unsigned char *buffer;
3414         unsigned int dev_max, rw_max;
3415
3416         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3417                                       "sd_revalidate_disk\n"));
3418
3419         /*
3420          * If the device is offline, don't try and read capacity or any
3421          * of the other niceties.
3422          */
3423         if (!scsi_device_online(sdp))
3424                 goto out;
3425
3426         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3427         if (!buffer) {
3428                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3429                           "allocation failure.\n");
3430                 goto out;
3431         }
3432
3433         sd_spinup_disk(sdkp);
3434
3435         /*
3436          * Without media there is no reason to ask; moreover, some devices
3437          * react badly if we do.
3438          */
3439         if (sdkp->media_present) {
3440                 sd_read_capacity(sdkp, buffer);
3441
3442                 /*
3443                  * set the default to rotational.  All non-rotational devices
3444                  * support the block characteristics VPD page, which will
3445                  * cause this to be updated correctly and any device which
3446                  * doesn't support it should be treated as rotational.
3447                  */
3448                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3449                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3450
3451                 if (scsi_device_supports_vpd(sdp)) {
3452                         sd_read_block_provisioning(sdkp);
3453                         sd_read_block_limits(sdkp);
3454                         sd_read_block_characteristics(sdkp);
3455                         sd_zbc_read_zones(sdkp, buffer);
3456                         sd_read_cpr(sdkp);
3457                 }
3458
3459                 sd_print_capacity(sdkp, old_capacity);
3460
3461                 sd_read_write_protect_flag(sdkp, buffer);
3462                 sd_read_cache_type(sdkp, buffer);
3463                 sd_read_app_tag_own(sdkp, buffer);
3464                 sd_read_write_same(sdkp, buffer);
3465                 sd_read_security(sdkp, buffer);
3466                 sd_config_protection(sdkp);
3467         }
3468
3469         /*
3470          * We now have all cache related info, determine how we deal
3471          * with flush requests.
3472          */
3473         sd_set_flush_flag(sdkp);
3474
3475         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3476         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3477
3478         /* Some devices report a maximum block count for READ/WRITE requests. */
3479         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3480         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3481
3482         if (sd_validate_min_xfer_size(sdkp))
3483                 blk_queue_io_min(sdkp->disk->queue,
3484                                  logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3485         else
3486                 blk_queue_io_min(sdkp->disk->queue, 0);
3487
3488         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3489                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3490                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3491         } else {
3492                 q->limits.io_opt = 0;
3493                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3494                                       (sector_t)BLK_DEF_MAX_SECTORS);
3495         }
3496
3497         /*
3498          * Limit default to SCSI host optimal sector limit if set. There may be
3499          * an impact on performance for when the size of a request exceeds this
3500          * host limit.
3501          */
3502         rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3503
3504         /* Do not exceed controller limit */
3505         rw_max = min(rw_max, queue_max_hw_sectors(q));
3506
3507         /*
3508          * Only update max_sectors if previously unset or if the current value
3509          * exceeds the capabilities of the hardware.
3510          */
3511         if (sdkp->first_scan ||
3512             q->limits.max_sectors > q->limits.max_dev_sectors ||
3513             q->limits.max_sectors > q->limits.max_hw_sectors)
3514                 q->limits.max_sectors = rw_max;
3515
3516         sdkp->first_scan = 0;
3517
3518         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3519         sd_config_write_same(sdkp);
3520         kfree(buffer);
3521
3522         /*
3523          * For a zoned drive, revalidating the zones can be done only once
3524          * the gendisk capacity is set. So if this fails, set back the gendisk
3525          * capacity to 0.
3526          */
3527         if (sd_zbc_revalidate_zones(sdkp))
3528                 set_capacity_and_notify(disk, 0);
3529
3530  out:
3531         return 0;
3532 }
3533
3534 /**
3535  *      sd_unlock_native_capacity - unlock native capacity
3536  *      @disk: struct gendisk to set capacity for
3537  *
3538  *      Block layer calls this function if it detects that partitions
3539  *      on @disk reach beyond the end of the device.  If the SCSI host
3540  *      implements ->unlock_native_capacity() method, it's invoked to
3541  *      give it a chance to adjust the device capacity.
3542  *
3543  *      CONTEXT:
3544  *      Defined by block layer.  Might sleep.
3545  */
3546 static void sd_unlock_native_capacity(struct gendisk *disk)
3547 {
3548         struct scsi_device *sdev = scsi_disk(disk)->device;
3549
3550         if (sdev->host->hostt->unlock_native_capacity)
3551                 sdev->host->hostt->unlock_native_capacity(sdev);
3552 }
3553
3554 /**
3555  *      sd_format_disk_name - format disk name
3556  *      @prefix: name prefix - ie. "sd" for SCSI disks
3557  *      @index: index of the disk to format name for
3558  *      @buf: output buffer
3559  *      @buflen: length of the output buffer
3560  *
3561  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3562  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3563  *      which is followed by sdaaa.
3564  *
3565  *      This is basically 26 base counting with one extra 'nil' entry
3566  *      at the beginning from the second digit on and can be
3567  *      determined using similar method as 26 base conversion with the
3568  *      index shifted -1 after each digit is computed.
3569  *
3570  *      CONTEXT:
3571  *      Don't care.
3572  *
3573  *      RETURNS:
3574  *      0 on success, -errno on failure.
3575  */
3576 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3577 {
3578         const int base = 'z' - 'a' + 1;
3579         char *begin = buf + strlen(prefix);
3580         char *end = buf + buflen;
3581         char *p;
3582         int unit;
3583
3584         p = end - 1;
3585         *p = '\0';
3586         unit = base;
3587         do {
3588                 if (p == begin)
3589                         return -EINVAL;
3590                 *--p = 'a' + (index % unit);
3591                 index = (index / unit) - 1;
3592         } while (index >= 0);
3593
3594         memmove(begin, p, end - p);
3595         memcpy(buf, prefix, strlen(prefix));
3596
3597         return 0;
3598 }
3599
3600 /**
3601  *      sd_probe - called during driver initialization and whenever a
3602  *      new scsi device is attached to the system. It is called once
3603  *      for each scsi device (not just disks) present.
3604  *      @dev: pointer to device object
3605  *
3606  *      Returns 0 if successful (or not interested in this scsi device
3607  *      (e.g. scanner)); 1 when there is an error.
3608  *
3609  *      Note: this function is invoked from the scsi mid-level.
3610  *      This function sets up the mapping between a given
3611  *      <host,channel,id,lun> (found in sdp) and new device name
3612  *      (e.g. /dev/sda). More precisely it is the block device major
3613  *      and minor number that is chosen here.
3614  *
3615  *      Assume sd_probe is not re-entrant (for time being)
3616  *      Also think about sd_probe() and sd_remove() running coincidentally.
3617  **/
3618 static int sd_probe(struct device *dev)
3619 {
3620         struct scsi_device *sdp = to_scsi_device(dev);
3621         struct scsi_disk *sdkp;
3622         struct gendisk *gd;
3623         int index;
3624         int error;
3625
3626         scsi_autopm_get_device(sdp);
3627         error = -ENODEV;
3628         if (sdp->type != TYPE_DISK &&
3629             sdp->type != TYPE_ZBC &&
3630             sdp->type != TYPE_MOD &&
3631             sdp->type != TYPE_RBC)
3632                 goto out;
3633
3634         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3635                 sdev_printk(KERN_WARNING, sdp,
3636                             "Unsupported ZBC host-managed device.\n");
3637                 goto out;
3638         }
3639
3640         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3641                                         "sd_probe\n"));
3642
3643         error = -ENOMEM;
3644         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3645         if (!sdkp)
3646                 goto out;
3647
3648         gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3649                                          &sd_bio_compl_lkclass);
3650         if (!gd)
3651                 goto out_free;
3652
3653         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3654         if (index < 0) {
3655                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3656                 goto out_put;
3657         }
3658
3659         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3660         if (error) {
3661                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3662                 goto out_free_index;
3663         }
3664
3665         sdkp->device = sdp;
3666         sdkp->disk = gd;
3667         sdkp->index = index;
3668         sdkp->max_retries = SD_MAX_RETRIES;
3669         atomic_set(&sdkp->openers, 0);
3670         atomic_set(&sdkp->device->ioerr_cnt, 0);
3671
3672         if (!sdp->request_queue->rq_timeout) {
3673                 if (sdp->type != TYPE_MOD)
3674                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3675                 else
3676                         blk_queue_rq_timeout(sdp->request_queue,
3677                                              SD_MOD_TIMEOUT);
3678         }
3679
3680         device_initialize(&sdkp->disk_dev);
3681         sdkp->disk_dev.parent = get_device(dev);
3682         sdkp->disk_dev.class = &sd_disk_class;
3683         dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3684
3685         error = device_add(&sdkp->disk_dev);
3686         if (error) {
3687                 put_device(&sdkp->disk_dev);
3688                 goto out;
3689         }
3690
3691         dev_set_drvdata(dev, sdkp);
3692
3693         gd->major = sd_major((index & 0xf0) >> 4);
3694         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3695         gd->minors = SD_MINORS;
3696
3697         gd->fops = &sd_fops;
3698         gd->private_data = sdkp;
3699
3700         /* defaults, until the device tells us otherwise */
3701         sdp->sector_size = 512;
3702         sdkp->capacity = 0;
3703         sdkp->media_present = 1;
3704         sdkp->write_prot = 0;
3705         sdkp->cache_override = 0;
3706         sdkp->WCE = 0;
3707         sdkp->RCD = 0;
3708         sdkp->ATO = 0;
3709         sdkp->first_scan = 1;
3710         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3711
3712         sd_revalidate_disk(gd);
3713
3714         if (sdp->removable) {
3715                 gd->flags |= GENHD_FL_REMOVABLE;
3716                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3717                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3718         }
3719
3720         blk_pm_runtime_init(sdp->request_queue, dev);
3721         if (sdp->rpm_autosuspend) {
3722                 pm_runtime_set_autosuspend_delay(dev,
3723                         sdp->host->hostt->rpm_autosuspend_delay);
3724         }
3725
3726         error = device_add_disk(dev, gd, NULL);
3727         if (error) {
3728                 put_device(&sdkp->disk_dev);
3729                 put_disk(gd);
3730                 goto out;
3731         }
3732
3733         if (sdkp->security) {
3734                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3735                 if (sdkp->opal_dev)
3736                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3737         }
3738
3739         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3740                   sdp->removable ? "removable " : "");
3741         scsi_autopm_put_device(sdp);
3742
3743         return 0;
3744
3745  out_free_index:
3746         ida_free(&sd_index_ida, index);
3747  out_put:
3748         put_disk(gd);
3749  out_free:
3750         kfree(sdkp);
3751  out:
3752         scsi_autopm_put_device(sdp);
3753         return error;
3754 }
3755
3756 /**
3757  *      sd_remove - called whenever a scsi disk (previously recognized by
3758  *      sd_probe) is detached from the system. It is called (potentially
3759  *      multiple times) during sd module unload.
3760  *      @dev: pointer to device object
3761  *
3762  *      Note: this function is invoked from the scsi mid-level.
3763  *      This function potentially frees up a device name (e.g. /dev/sdc)
3764  *      that could be re-used by a subsequent sd_probe().
3765  *      This function is not called when the built-in sd driver is "exit-ed".
3766  **/
3767 static int sd_remove(struct device *dev)
3768 {
3769         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3770
3771         scsi_autopm_get_device(sdkp->device);
3772
3773         device_del(&sdkp->disk_dev);
3774         del_gendisk(sdkp->disk);
3775         if (!sdkp->suspended)
3776                 sd_shutdown(dev);
3777
3778         put_disk(sdkp->disk);
3779         return 0;
3780 }
3781
3782 static void scsi_disk_release(struct device *dev)
3783 {
3784         struct scsi_disk *sdkp = to_scsi_disk(dev);
3785
3786         ida_free(&sd_index_ida, sdkp->index);
3787         sd_zbc_free_zone_info(sdkp);
3788         put_device(&sdkp->device->sdev_gendev);
3789         free_opal_dev(sdkp->opal_dev);
3790
3791         kfree(sdkp);
3792 }
3793
3794 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3795 {
3796         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3797         struct scsi_sense_hdr sshdr;
3798         const struct scsi_exec_args exec_args = {
3799                 .sshdr = &sshdr,
3800                 .req_flags = BLK_MQ_REQ_PM,
3801         };
3802         struct scsi_device *sdp = sdkp->device;
3803         int res;
3804
3805         if (start)
3806                 cmd[4] |= 1;    /* START */
3807
3808         if (sdp->start_stop_pwr_cond)
3809                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3810
3811         if (!scsi_device_online(sdp))
3812                 return -ENODEV;
3813
3814         res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
3815                                sdkp->max_retries, &exec_args);
3816         if (res) {
3817                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3818                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3819                         sd_print_sense_hdr(sdkp, &sshdr);
3820                         /* 0x3a is medium not present */
3821                         if (sshdr.asc == 0x3a)
3822                                 res = 0;
3823                 }
3824         }
3825
3826         /* SCSI error codes must not go to the generic layer */
3827         if (res)
3828                 return -EIO;
3829
3830         return 0;
3831 }
3832
3833 /*
3834  * Send a SYNCHRONIZE CACHE instruction down to the device through
3835  * the normal SCSI command structure.  Wait for the command to
3836  * complete.
3837  */
3838 static void sd_shutdown(struct device *dev)
3839 {
3840         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3841
3842         if (!sdkp)
3843                 return;         /* this can happen */
3844
3845         if (pm_runtime_suspended(dev))
3846                 return;
3847
3848         if (sdkp->WCE && sdkp->media_present) {
3849                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3850                 sd_sync_cache(sdkp, NULL);
3851         }
3852
3853         if ((system_state != SYSTEM_RESTART &&
3854              sdkp->device->manage_system_start_stop) ||
3855             (system_state == SYSTEM_POWER_OFF &&
3856              sdkp->device->manage_shutdown)) {
3857                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3858                 sd_start_stop_device(sdkp, 0);
3859         }
3860 }
3861
3862 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
3863 {
3864         return (sdev->manage_system_start_stop && !runtime) ||
3865                 (sdev->manage_runtime_start_stop && runtime);
3866 }
3867
3868 static int sd_suspend_common(struct device *dev, bool runtime)
3869 {
3870         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3871         struct scsi_sense_hdr sshdr;
3872         int ret = 0;
3873
3874         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3875                 return 0;
3876
3877         if (sdkp->WCE && sdkp->media_present) {
3878                 if (!sdkp->device->silence_suspend)
3879                         sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3880                 ret = sd_sync_cache(sdkp, &sshdr);
3881
3882                 if (ret) {
3883                         /* ignore OFFLINE device */
3884                         if (ret == -ENODEV)
3885                                 return 0;
3886
3887                         if (!scsi_sense_valid(&sshdr) ||
3888                             sshdr.sense_key != ILLEGAL_REQUEST)
3889                                 return ret;
3890
3891                         /*
3892                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3893                          * doesn't support sync. There's not much to do and
3894                          * suspend shouldn't fail.
3895                          */
3896                         ret = 0;
3897                 }
3898         }
3899
3900         if (sd_do_start_stop(sdkp->device, runtime)) {
3901                 if (!sdkp->device->silence_suspend)
3902                         sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3903                 /* an error is not worth aborting a system sleep */
3904                 ret = sd_start_stop_device(sdkp, 0);
3905                 if (!runtime)
3906                         ret = 0;
3907         }
3908
3909         if (!ret)
3910                 sdkp->suspended = true;
3911
3912         return ret;
3913 }
3914
3915 static int sd_suspend_system(struct device *dev)
3916 {
3917         if (pm_runtime_suspended(dev))
3918                 return 0;
3919
3920         return sd_suspend_common(dev, false);
3921 }
3922
3923 static int sd_suspend_runtime(struct device *dev)
3924 {
3925         return sd_suspend_common(dev, true);
3926 }
3927
3928 static int sd_resume(struct device *dev, bool runtime)
3929 {
3930         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3931         int ret = 0;
3932
3933         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3934                 return 0;
3935
3936         if (!sd_do_start_stop(sdkp->device, runtime)) {
3937                 sdkp->suspended = false;
3938                 return 0;
3939         }
3940
3941         if (!sdkp->device->no_start_on_resume) {
3942                 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3943                 ret = sd_start_stop_device(sdkp, 1);
3944         }
3945
3946         if (!ret) {
3947                 opal_unlock_from_suspend(sdkp->opal_dev);
3948                 sdkp->suspended = false;
3949         }
3950
3951         return ret;
3952 }
3953
3954 static int sd_resume_system(struct device *dev)
3955 {
3956         if (pm_runtime_suspended(dev))
3957                 return 0;
3958
3959         return sd_resume(dev, false);
3960 }
3961
3962 static int sd_resume_runtime(struct device *dev)
3963 {
3964         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3965         struct scsi_device *sdp;
3966
3967         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3968                 return 0;
3969
3970         sdp = sdkp->device;
3971
3972         if (sdp->ignore_media_change) {
3973                 /* clear the device's sense data */
3974                 static const u8 cmd[10] = { REQUEST_SENSE };
3975                 const struct scsi_exec_args exec_args = {
3976                         .req_flags = BLK_MQ_REQ_PM,
3977                 };
3978
3979                 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
3980                                      sdp->request_queue->rq_timeout, 1,
3981                                      &exec_args))
3982                         sd_printk(KERN_NOTICE, sdkp,
3983                                   "Failed to clear sense data\n");
3984         }
3985
3986         return sd_resume(dev, true);
3987 }
3988
3989 static const struct dev_pm_ops sd_pm_ops = {
3990         .suspend                = sd_suspend_system,
3991         .resume                 = sd_resume_system,
3992         .poweroff               = sd_suspend_system,
3993         .restore                = sd_resume_system,
3994         .runtime_suspend        = sd_suspend_runtime,
3995         .runtime_resume         = sd_resume_runtime,
3996 };
3997
3998 static struct scsi_driver sd_template = {
3999         .gendrv = {
4000                 .name           = "sd",
4001                 .owner          = THIS_MODULE,
4002                 .probe          = sd_probe,
4003                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
4004                 .remove         = sd_remove,
4005                 .shutdown       = sd_shutdown,
4006                 .pm             = &sd_pm_ops,
4007         },
4008         .rescan                 = sd_rescan,
4009         .init_command           = sd_init_command,
4010         .uninit_command         = sd_uninit_command,
4011         .done                   = sd_done,
4012         .eh_action              = sd_eh_action,
4013         .eh_reset               = sd_eh_reset,
4014 };
4015
4016 /**
4017  *      init_sd - entry point for this driver (both when built in or when
4018  *      a module).
4019  *
4020  *      Note: this function registers this driver with the scsi mid-level.
4021  **/
4022 static int __init init_sd(void)
4023 {
4024         int majors = 0, i, err;
4025
4026         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4027
4028         for (i = 0; i < SD_MAJORS; i++) {
4029                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4030                         continue;
4031                 majors++;
4032         }
4033
4034         if (!majors)
4035                 return -ENODEV;
4036
4037         err = class_register(&sd_disk_class);
4038         if (err)
4039                 goto err_out;
4040
4041         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4042         if (!sd_page_pool) {
4043                 printk(KERN_ERR "sd: can't init discard page pool\n");
4044                 err = -ENOMEM;
4045                 goto err_out_class;
4046         }
4047
4048         err = scsi_register_driver(&sd_template.gendrv);
4049         if (err)
4050                 goto err_out_driver;
4051
4052         return 0;
4053
4054 err_out_driver:
4055         mempool_destroy(sd_page_pool);
4056 err_out_class:
4057         class_unregister(&sd_disk_class);
4058 err_out:
4059         for (i = 0; i < SD_MAJORS; i++)
4060                 unregister_blkdev(sd_major(i), "sd");
4061         return err;
4062 }
4063
4064 /**
4065  *      exit_sd - exit point for this driver (when it is a module).
4066  *
4067  *      Note: this function unregisters this driver from the scsi mid-level.
4068  **/
4069 static void __exit exit_sd(void)
4070 {
4071         int i;
4072
4073         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4074
4075         scsi_unregister_driver(&sd_template.gendrv);
4076         mempool_destroy(sd_page_pool);
4077
4078         class_unregister(&sd_disk_class);
4079
4080         for (i = 0; i < SD_MAJORS; i++)
4081                 unregister_blkdev(sd_major(i), "sd");
4082 }
4083
4084 module_init(init_sd);
4085 module_exit(exit_sd);
4086
4087 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4088 {
4089         scsi_print_sense_hdr(sdkp->device,
4090                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4091 }
4092
4093 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4094 {
4095         const char *hb_string = scsi_hostbyte_string(result);
4096
4097         if (hb_string)
4098                 sd_printk(KERN_INFO, sdkp,
4099                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4100                           hb_string ? hb_string : "invalid",
4101                           "DRIVER_OK");
4102         else
4103                 sd_printk(KERN_INFO, sdkp,
4104                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4105                           msg, host_byte(result), "DRIVER_OK");
4106 }