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 <[email protected]> original
  11  *       - Eric Youngdale <[email protected]> add scatter-gather, multiple
  12  *         outstanding request, and other enhancements.
  13  *         Support loadable low-level scsi drivers.
  14  *       - Jirka Hanika <[email protected]> support more scsi disks using
  15  *         eight major numbers.
  16  *       - Richard Gooch <[email protected]> support devfs.
  17  *       - Torben Mathiasen <[email protected]> Resource allocation fixes in
  18  *         sd_init and cleanups.
  19  *       - Alex Davis <[email protected]> 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 <[email protected]> cleanup for lk 2.5.x
  23  *       - Badari Pulavarty <[email protected]>, Matthew Wilcox
  24  *         <[email protected]>, Kurt Garloff <[email protected]>:
  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, ret;
 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         ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
 194                                sdkp->max_retries, &data, &sshdr);
 195         if (ret) {
 196                 if (ret > 0 && scsi_sense_valid(&sshdr))
 197                         sd_print_sense_hdr(sdkp, &sshdr);
 198                 return -EINVAL;
 199         }
 200         sd_revalidate_disk(sdkp->disk);
 201         return count;
 202 }
 203
 204 static ssize_t
 205 manage_start_stop_show(struct device *dev,
 206                        struct device_attribute *attr, char *buf)
 207 {
 208         struct scsi_disk *sdkp = to_scsi_disk(dev);
 209         struct scsi_device *sdp = sdkp->device;
 210
 211         return sysfs_emit(buf, "%u\n",
 212                           sdp->manage_system_start_stop &&
 213                           sdp->manage_runtime_start_stop &&
 214                           sdp->manage_shutdown);
 215 }
 216 static DEVICE_ATTR_RO(manage_start_stop);
 217
 218 static ssize_t
 219 manage_system_start_stop_show(struct device *dev,
 220                               struct device_attribute *attr, char *buf)
 221 {
 222         struct scsi_disk *sdkp = to_scsi_disk(dev);
 223         struct scsi_device *sdp = sdkp->device;
 224
 225         return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
 226 }
 227
 228 static ssize_t
 229 manage_system_start_stop_store(struct device *dev,
 230                                struct device_attribute *attr,
 231                                const char *buf, size_t count)
 232 {
 233         struct scsi_disk *sdkp = to_scsi_disk(dev);
 234         struct scsi_device *sdp = sdkp->device;
 235         bool v;
 236
 237         if (!capable(CAP_SYS_ADMIN))
 238                 return -EACCES;
 239
 240         if (kstrtobool(buf, &v))
 241                 return -EINVAL;
 242
 243         sdp->manage_system_start_stop = v;
 244
 245         return count;
 246 }
 247 static DEVICE_ATTR_RW(manage_system_start_stop);
 248
 249 static ssize_t
 250 manage_runtime_start_stop_show(struct device *dev,
 251                                struct device_attribute *attr, char *buf)
 252 {
 253         struct scsi_disk *sdkp = to_scsi_disk(dev);
 254         struct scsi_device *sdp = sdkp->device;
 255
 256         return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
 257 }
 258
 259 static ssize_t
 260 manage_runtime_start_stop_store(struct device *dev,
 261                                 struct device_attribute *attr,
 262                                 const char *buf, size_t count)
 263 {
 264         struct scsi_disk *sdkp = to_scsi_disk(dev);
 265         struct scsi_device *sdp = sdkp->device;
 266         bool v;
 267
 268         if (!capable(CAP_SYS_ADMIN))
 269                 return -EACCES;
 270
 271         if (kstrtobool(buf, &v))
 272                 return -EINVAL;
 273
 274         sdp->manage_runtime_start_stop = v;
 275
 276         return count;
 277 }
 278 static DEVICE_ATTR_RW(manage_runtime_start_stop);
 279
 280 static ssize_t manage_shutdown_show(struct device *dev,
 281                                     struct device_attribute *attr, char *buf)
 282 {
 283         struct scsi_disk *sdkp = to_scsi_disk(dev);
 284         struct scsi_device *sdp = sdkp->device;
 285
 286         return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
 287 }
 288
 289 static ssize_t manage_shutdown_store(struct device *dev,
 290                                      struct device_attribute *attr,
 291                                      const char *buf, size_t count)
 292 {
 293         struct scsi_disk *sdkp = to_scsi_disk(dev);
 294         struct scsi_device *sdp = sdkp->device;
 295         bool v;
 296
 297         if (!capable(CAP_SYS_ADMIN))
 298                 return -EACCES;
 299
 300         if (kstrtobool(buf, &v))
 301                 return -EINVAL;
 302
 303         sdp->manage_shutdown = v;
 304
 305         return count;
 306 }
 307 static DEVICE_ATTR_RW(manage_shutdown);
 308
 309 static ssize_t
 310 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 311 {
 312         struct scsi_disk *sdkp = to_scsi_disk(dev);
 313
 314         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 315 }
 316
 317 static ssize_t
 318 allow_restart_store(struct device *dev, struct device_attribute *attr,
 319                     const char *buf, size_t count)
 320 {
 321         bool v;
 322         struct scsi_disk *sdkp = to_scsi_disk(dev);
 323         struct scsi_device *sdp = sdkp->device;
 324
 325         if (!capable(CAP_SYS_ADMIN))
 326                 return -EACCES;
 327
 328         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 329                 return -EINVAL;
 330
 331         if (kstrtobool(buf, &v))
 332                 return -EINVAL;
 333
 334         sdp->allow_restart = v;
 335
 336         return count;
 337 }
 338 static DEVICE_ATTR_RW(allow_restart);
 339
 340 static ssize_t
 341 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 342 {
 343         struct scsi_disk *sdkp = to_scsi_disk(dev);
 344         int ct = sdkp->RCD + 2*sdkp->WCE;
 345
 346         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 347 }
 348 static DEVICE_ATTR_RW(cache_type);
 349
 350 static ssize_t
 351 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 352 {
 353         struct scsi_disk *sdkp = to_scsi_disk(dev);
 354
 355         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 356 }
 357 static DEVICE_ATTR_RO(FUA);
 358
 359 static ssize_t
 360 protection_type_show(struct device *dev, struct device_attribute *attr,
 361                      char *buf)
 362 {
 363         struct scsi_disk *sdkp = to_scsi_disk(dev);
 364
 365         return sprintf(buf, "%u\n", sdkp->protection_type);
 366 }
 367
 368 static ssize_t
 369 protection_type_store(struct device *dev, struct device_attribute *attr,
 370                       const char *buf, size_t count)
 371 {
 372         struct scsi_disk *sdkp = to_scsi_disk(dev);
 373         unsigned int val;
 374         int err;
 375
 376         if (!capable(CAP_SYS_ADMIN))
 377                 return -EACCES;
 378
 379         err = kstrtouint(buf, 10, &val);
 380
 381         if (err)
 382                 return err;
 383
 384         if (val <= T10_PI_TYPE3_PROTECTION)
 385                 sdkp->protection_type = val;
 386
 387         return count;
 388 }
 389 static DEVICE_ATTR_RW(protection_type);
 390
 391 static ssize_t
 392 protection_mode_show(struct device *dev, struct device_attribute *attr,
 393                      char *buf)
 394 {
 395         struct scsi_disk *sdkp = to_scsi_disk(dev);
 396         struct scsi_device *sdp = sdkp->device;
 397         unsigned int dif, dix;
 398
 399         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 400         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 401
 402         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 403                 dif = 0;
 404                 dix = 1;
 405         }
 406
 407         if (!dif && !dix)
 408                 return sprintf(buf, "none\n");
 409
 410         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 411 }
 412 static DEVICE_ATTR_RO(protection_mode);
 413
 414 static ssize_t
 415 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 416 {
 417         struct scsi_disk *sdkp = to_scsi_disk(dev);
 418
 419         return sprintf(buf, "%u\n", sdkp->ATO);
 420 }
 421 static DEVICE_ATTR_RO(app_tag_own);
 422
 423 static ssize_t
 424 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 425                        char *buf)
 426 {
 427         struct scsi_disk *sdkp = to_scsi_disk(dev);
 428
 429         return sprintf(buf, "%u\n", sdkp->lbpme);
 430 }
 431 static DEVICE_ATTR_RO(thin_provisioning);
 432
 433 /* sysfs_match_string() requires dense arrays */
 434 static const char *lbp_mode[] = {
 435         [SD_LBP_FULL]           = "full",
 436         [SD_LBP_UNMAP]          = "unmap",
 437         [SD_LBP_WS16]           = "writesame_16",
 438         [SD_LBP_WS10]           = "writesame_10",
 439         [SD_LBP_ZERO]           = "writesame_zero",
 440         [SD_LBP_DISABLE]        = "disabled",
 441 };
 442
 443 static ssize_t
 444 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 445                        char *buf)
 446 {
 447         struct scsi_disk *sdkp = to_scsi_disk(dev);
 448
 449         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 450 }
 451
 452 static ssize_t
 453 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 454                         const char *buf, size_t count)
 455 {
 456         struct scsi_disk *sdkp = to_scsi_disk(dev);
 457         struct scsi_device *sdp = sdkp->device;
 458         int mode;
 459
 460         if (!capable(CAP_SYS_ADMIN))
 461                 return -EACCES;
 462
 463         if (sd_is_zoned(sdkp)) {
 464                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 465                 return count;
 466         }
 467
 468         if (sdp->type != TYPE_DISK)
 469                 return -EINVAL;
 470
 471         mode = sysfs_match_string(lbp_mode, buf);
 472         if (mode < 0)
 473                 return -EINVAL;
 474
 475         sd_config_discard(sdkp, mode);
 476
 477         return count;
 478 }
 479 static DEVICE_ATTR_RW(provisioning_mode);
 480
 481 /* sysfs_match_string() requires dense arrays */
 482 static const char *zeroing_mode[] = {
 483         [SD_ZERO_WRITE]         = "write",
 484         [SD_ZERO_WS]            = "writesame",
 485         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 486         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 487 };
 488
 489 static ssize_t
 490 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 491                   char *buf)
 492 {
 493         struct scsi_disk *sdkp = to_scsi_disk(dev);
 494
 495         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 496 }
 497
 498 static ssize_t
 499 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 500                    const char *buf, size_t count)
 501 {
 502         struct scsi_disk *sdkp = to_scsi_disk(dev);
 503         int mode;
 504
 505         if (!capable(CAP_SYS_ADMIN))
 506                 return -EACCES;
 507
 508         mode = sysfs_match_string(zeroing_mode, buf);
 509         if (mode < 0)
 510                 return -EINVAL;
 511
 512         sdkp->zeroing_mode = mode;
 513
 514         return count;
 515 }
 516 static DEVICE_ATTR_RW(zeroing_mode);
 517
 518 static ssize_t
 519 max_medium_access_timeouts_show(struct device *dev,
 520                                 struct device_attribute *attr, char *buf)
 521 {
 522         struct scsi_disk *sdkp = to_scsi_disk(dev);
 523
 524         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 525 }
 526
 527 static ssize_t
 528 max_medium_access_timeouts_store(struct device *dev,
 529                                  struct device_attribute *attr, const char *buf,
 530                                  size_t count)
 531 {
 532         struct scsi_disk *sdkp = to_scsi_disk(dev);
 533         int err;
 534
 535         if (!capable(CAP_SYS_ADMIN))
 536                 return -EACCES;
 537
 538         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 539
 540         return err ? err : count;
 541 }
 542 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 543
 544 static ssize_t
 545 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 546                            char *buf)
 547 {
 548         struct scsi_disk *sdkp = to_scsi_disk(dev);
 549
 550         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 551 }
 552
 553 static ssize_t
 554 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 555                             const char *buf, size_t count)
 556 {
 557         struct scsi_disk *sdkp = to_scsi_disk(dev);
 558         struct scsi_device *sdp = sdkp->device;
 559         unsigned long max;
 560         int err;
 561
 562         if (!capable(CAP_SYS_ADMIN))
 563                 return -EACCES;
 564
 565         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 566                 return -EINVAL;
 567
 568         err = kstrtoul(buf, 10, &max);
 569
 570         if (err)
 571                 return err;
 572
 573         if (max == 0)
 574                 sdp->no_write_same = 1;
 575         else if (max <= SD_MAX_WS16_BLOCKS) {
 576                 sdp->no_write_same = 0;
 577                 sdkp->max_ws_blocks = max;
 578         }
 579
 580         sd_config_write_same(sdkp);
 581
 582         return count;
 583 }
 584 static DEVICE_ATTR_RW(max_write_same_blocks);
 585
 586 static ssize_t
 587 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
 588 {
 589         struct scsi_disk *sdkp = to_scsi_disk(dev);
 590
 591         if (sdkp->device->type == TYPE_ZBC)
 592                 return sprintf(buf, "host-managed\n");
 593         if (sdkp->zoned == 1)
 594                 return sprintf(buf, "host-aware\n");
 595         if (sdkp->zoned == 2)
 596                 return sprintf(buf, "drive-managed\n");
 597         return sprintf(buf, "none\n");
 598 }
 599 static DEVICE_ATTR_RO(zoned_cap);
 600
 601 static ssize_t
 602 max_retries_store(struct device *dev, struct device_attribute *attr,
 603                   const char *buf, size_t count)
 604 {
 605         struct scsi_disk *sdkp = to_scsi_disk(dev);
 606         struct scsi_device *sdev = sdkp->device;
 607         int retries, err;
 608
 609         err = kstrtoint(buf, 10, &retries);
 610         if (err)
 611                 return err;
 612
 613         if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
 614                 sdkp->max_retries = retries;
 615                 return count;
 616         }
 617
 618         sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
 619                     SD_MAX_RETRIES);
 620         return -EINVAL;
 621 }
 622
 623 static ssize_t
 624 max_retries_show(struct device *dev, struct device_attribute *attr,
 625                  char *buf)
 626 {
 627         struct scsi_disk *sdkp = to_scsi_disk(dev);
 628
 629         return sprintf(buf, "%d\n", sdkp->max_retries);
 630 }
 631
 632 static DEVICE_ATTR_RW(max_retries);
 633
 634 static struct attribute *sd_disk_attrs[] = {
 635         &dev_attr_cache_type.attr,
 636         &dev_attr_FUA.attr,
 637         &dev_attr_allow_restart.attr,
 638         &dev_attr_manage_start_stop.attr,
 639         &dev_attr_manage_system_start_stop.attr,
 640         &dev_attr_manage_runtime_start_stop.attr,
 641         &dev_attr_manage_shutdown.attr,
 642         &dev_attr_protection_type.attr,
 643         &dev_attr_protection_mode.attr,
 644         &dev_attr_app_tag_own.attr,
 645         &dev_attr_thin_provisioning.attr,
 646         &dev_attr_provisioning_mode.attr,
 647         &dev_attr_zeroing_mode.attr,
 648         &dev_attr_max_write_same_blocks.attr,
 649         &dev_attr_max_medium_access_timeouts.attr,
 650         &dev_attr_zoned_cap.attr,
 651         &dev_attr_max_retries.attr,
 652         NULL,
 653 };
 654 ATTRIBUTE_GROUPS(sd_disk);
 655
 656 static struct class sd_disk_class = {
 657         .name           = "scsi_disk",
 658         .dev_release    = scsi_disk_release,
 659         .dev_groups     = sd_disk_groups,
 660 };
 661
 662 /*
 663  * Don't request a new module, as that could deadlock in multipath
 664  * environment.
 665  */
 666 static void sd_default_probe(dev_t devt)
 667 {
 668 }
 669
 670 /*
 671  * Device no to disk mapping:
 672  *
 673  *       major         disc2     disc  p1
 674  *   |............|.............|....|....| <- dev_t
 675  *    31        20 19          8 7  4 3  0
 676  *
 677  * Inside a major, we have 16k disks, however mapped non-
 678  * contiguously. The first 16 disks are for major0, the next
 679  * ones with major1, ... Disk 256 is for major0 again, disk 272
 680  * for major1, ...
 681  * As we stay compatible with our numbering scheme, we can reuse
 682  * the well-know SCSI majors 8, 65--71, 136--143.
 683  */
 684 static int sd_major(int major_idx)
 685 {
 686         switch (major_idx) {
 687         case 0:
 688                 return SCSI_DISK0_MAJOR;
 689         case 1 ... 7:
 690                 return SCSI_DISK1_MAJOR + major_idx - 1;
 691         case 8 ... 15:
 692                 return SCSI_DISK8_MAJOR + major_idx - 8;
 693         default:
 694                 BUG();
 695                 return 0;       /* shut up gcc */
 696         }
 697 }
 698
 699 #ifdef CONFIG_BLK_SED_OPAL
 700 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 701                 size_t len, bool send)
 702 {
 703         struct scsi_disk *sdkp = data;
 704         struct scsi_device *sdev = sdkp->device;
 705         u8 cdb[12] = { 0, };
 706         const struct scsi_exec_args exec_args = {
 707                 .req_flags = BLK_MQ_REQ_PM,
 708         };
 709         int ret;
 710
 711         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 712         cdb[1] = secp;
 713         put_unaligned_be16(spsp, &cdb[2]);
 714         put_unaligned_be32(len, &cdb[6]);
 715
 716         ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
 717                                buffer, len, SD_TIMEOUT, sdkp->max_retries,
 718                                &exec_args);
 719         return ret <= 0 ? ret : -EIO;
 720 }
 721 #endif /* CONFIG_BLK_SED_OPAL */
 722
 723 /*
 724  * Look up the DIX operation based on whether the command is read or
 725  * write and whether dix and dif are enabled.
 726  */
 727 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 728 {
 729         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 730         static const unsigned int ops[] = {     /* wrt dix dif */
 731                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 732                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 733                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 734                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 735                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 736                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 737                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 738                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 739         };
 740
 741         return ops[write << 2 | dix << 1 | dif];
 742 }
 743
 744 /*
 745  * Returns a mask of the protection flags that are valid for a given DIX
 746  * operation.
 747  */
 748 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 749 {
 750         static const unsigned int flag_mask[] = {
 751                 [SCSI_PROT_NORMAL]              = 0,
 752
 753                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 754                                                   SCSI_PROT_GUARD_CHECK |
 755                                                   SCSI_PROT_REF_CHECK |
 756                                                   SCSI_PROT_REF_INCREMENT,
 757
 758                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 759                                                   SCSI_PROT_IP_CHECKSUM,
 760
 761                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 762                                                   SCSI_PROT_GUARD_CHECK |
 763                                                   SCSI_PROT_REF_CHECK |
 764                                                   SCSI_PROT_REF_INCREMENT |
 765                                                   SCSI_PROT_IP_CHECKSUM,
 766
 767                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 768                                                   SCSI_PROT_REF_INCREMENT,
 769
 770                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 771                                                   SCSI_PROT_REF_CHECK |
 772                                                   SCSI_PROT_REF_INCREMENT |
 773                                                   SCSI_PROT_IP_CHECKSUM,
 774
 775                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 776                                                   SCSI_PROT_GUARD_CHECK |
 777                                                   SCSI_PROT_REF_CHECK |
 778                                                   SCSI_PROT_REF_INCREMENT |
 779                                                   SCSI_PROT_IP_CHECKSUM,
 780         };
 781
 782         return flag_mask[prot_op];
 783 }
 784
 785 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 786                                            unsigned int dix, unsigned int dif)
 787 {
 788         struct request *rq = scsi_cmd_to_rq(scmd);
 789         struct bio *bio = rq->bio;
 790         unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
 791         unsigned int protect = 0;
 792
 793         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 794                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 795                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 796
 797                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 798                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 799         }
 800
 801         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 802                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 803
 804                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 805                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 806         }
 807
 808         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 809                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 810
 811                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 812                         protect = 3 << 5;       /* Disable target PI checking */
 813                 else
 814                         protect = 1 << 5;       /* Enable target PI checking */
 815         }
 816
 817         scsi_set_prot_op(scmd, prot_op);
 818         scsi_set_prot_type(scmd, dif);
 819         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 820
 821         return protect;
 822 }
 823
 824 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 825 {
 826         struct request_queue *q = sdkp->disk->queue;
 827         unsigned int logical_block_size = sdkp->device->sector_size;
 828         unsigned int max_blocks = 0;
 829
 830         q->limits.discard_alignment =
 831                 sdkp->unmap_alignment * logical_block_size;
 832         q->limits.discard_granularity =
 833                 max(sdkp->physical_block_size,
 834                     sdkp->unmap_granularity * logical_block_size);
 835         sdkp->provisioning_mode = mode;
 836
 837         switch (mode) {
 838
 839         case SD_LBP_FULL:
 840         case SD_LBP_DISABLE:
 841                 blk_queue_max_discard_sectors(q, 0);
 842                 return;
 843
 844         case SD_LBP_UNMAP:
 845                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 846                                           (u32)SD_MAX_WS16_BLOCKS);
 847                 break;
 848
 849         case SD_LBP_WS16:
 850                 if (sdkp->device->unmap_limit_for_ws)
 851                         max_blocks = sdkp->max_unmap_blocks;
 852                 else
 853                         max_blocks = sdkp->max_ws_blocks;
 854
 855                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 856                 break;
 857
 858         case SD_LBP_WS10:
 859                 if (sdkp->device->unmap_limit_for_ws)
 860                         max_blocks = sdkp->max_unmap_blocks;
 861                 else
 862                         max_blocks = sdkp->max_ws_blocks;
 863
 864                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 865                 break;
 866
 867         case SD_LBP_ZERO:
 868                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 869                                           (u32)SD_MAX_WS10_BLOCKS);
 870                 break;
 871         }
 872
 873         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 874 }
 875
 876 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
 877 {
 878         struct page *page;
 879
 880         page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 881         if (!page)
 882                 return NULL;
 883         clear_highpage(page);
 884         bvec_set_page(&rq->special_vec, page, data_len, 0);
 885         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 886         return bvec_virt(&rq->special_vec);
 887 }
 888
 889 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 890 {
 891         struct scsi_device *sdp = cmd->device;
 892         struct request *rq = scsi_cmd_to_rq(cmd);
 893         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 894         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 895         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 896         unsigned int data_len = 24;
 897         char *buf;
 898
 899         buf = sd_set_special_bvec(rq, data_len);
 900         if (!buf)
 901                 return BLK_STS_RESOURCE;
 902
 903         cmd->cmd_len = 10;
 904         cmd->cmnd[0] = UNMAP;
 905         cmd->cmnd[8] = 24;
 906
 907         put_unaligned_be16(6 + 16, &buf[0]);
 908         put_unaligned_be16(16, &buf[2]);
 909         put_unaligned_be64(lba, &buf[8]);
 910         put_unaligned_be32(nr_blocks, &buf[16]);
 911
 912         cmd->allowed = sdkp->max_retries;
 913         cmd->transfersize = data_len;
 914         rq->timeout = SD_TIMEOUT;
 915
 916         return scsi_alloc_sgtables(cmd);
 917 }
 918
 919 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 920                 bool unmap)
 921 {
 922         struct scsi_device *sdp = cmd->device;
 923         struct request *rq = scsi_cmd_to_rq(cmd);
 924         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 925         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 926         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 927         u32 data_len = sdp->sector_size;
 928
 929         if (!sd_set_special_bvec(rq, data_len))
 930                 return BLK_STS_RESOURCE;
 931
 932         cmd->cmd_len = 16;
 933         cmd->cmnd[0] = WRITE_SAME_16;
 934         if (unmap)
 935                 cmd->cmnd[1] = 0x8; /* UNMAP */
 936         put_unaligned_be64(lba, &cmd->cmnd[2]);
 937         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 938
 939         cmd->allowed = sdkp->max_retries;
 940         cmd->transfersize = data_len;
 941         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 942
 943         return scsi_alloc_sgtables(cmd);
 944 }
 945
 946 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 947                 bool unmap)
 948 {
 949         struct scsi_device *sdp = cmd->device;
 950         struct request *rq = scsi_cmd_to_rq(cmd);
 951         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 952         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 953         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 954         u32 data_len = sdp->sector_size;
 955
 956         if (!sd_set_special_bvec(rq, data_len))
 957                 return BLK_STS_RESOURCE;
 958
 959         cmd->cmd_len = 10;
 960         cmd->cmnd[0] = WRITE_SAME;
 961         if (unmap)
 962                 cmd->cmnd[1] = 0x8; /* UNMAP */
 963         put_unaligned_be32(lba, &cmd->cmnd[2]);
 964         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 965
 966         cmd->allowed = sdkp->max_retries;
 967         cmd->transfersize = data_len;
 968         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 969
 970         return scsi_alloc_sgtables(cmd);
 971 }
 972
 973 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 974 {
 975         struct request *rq = scsi_cmd_to_rq(cmd);
 976         struct scsi_device *sdp = cmd->device;
 977         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 978         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 979         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 980
 981         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 982                 switch (sdkp->zeroing_mode) {
 983                 case SD_ZERO_WS16_UNMAP:
 984                         return sd_setup_write_same16_cmnd(cmd, true);
 985                 case SD_ZERO_WS10_UNMAP:
 986                         return sd_setup_write_same10_cmnd(cmd, true);
 987                 }
 988         }
 989
 990         if (sdp->no_write_same) {
 991                 rq->rq_flags |= RQF_QUIET;
 992                 return BLK_STS_TARGET;
 993         }
 994
 995         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 996                 return sd_setup_write_same16_cmnd(cmd, false);
 997
 998         return sd_setup_write_same10_cmnd(cmd, false);
 999 }
1000
1001 static void sd_config_write_same(struct scsi_disk *sdkp)
1002 {
1003         struct request_queue *q = sdkp->disk->queue;
1004         unsigned int logical_block_size = sdkp->device->sector_size;
1005
1006         if (sdkp->device->no_write_same) {
1007                 sdkp->max_ws_blocks = 0;
1008                 goto out;
1009         }
1010
1011         /* Some devices can not handle block counts above 0xffff despite
1012          * supporting WRITE SAME(16). Consequently we default to 64k
1013          * blocks per I/O unless the device explicitly advertises a
1014          * bigger limit.
1015          */
1016         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1017                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1018                                                    (u32)SD_MAX_WS16_BLOCKS);
1019         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1020                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1021                                                    (u32)SD_MAX_WS10_BLOCKS);
1022         else {
1023                 sdkp->device->no_write_same = 1;
1024                 sdkp->max_ws_blocks = 0;
1025         }
1026
1027         if (sdkp->lbprz && sdkp->lbpws)
1028                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1029         else if (sdkp->lbprz && sdkp->lbpws10)
1030                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1031         else if (sdkp->max_ws_blocks)
1032                 sdkp->zeroing_mode = SD_ZERO_WS;
1033         else
1034                 sdkp->zeroing_mode = SD_ZERO_WRITE;
1035
1036         if (sdkp->max_ws_blocks &&
1037             sdkp->physical_block_size > logical_block_size) {
1038                 /*
1039                  * Reporting a maximum number of blocks that is not aligned
1040                  * on the device physical size would cause a large write same
1041                  * request to be split into physically unaligned chunks by
1042                  * __blkdev_issue_write_zeroes() even if the caller of this
1043                  * functions took care to align the large request. So make sure
1044                  * the maximum reported is aligned to the device physical block
1045                  * size. This is only an optional optimization for regular
1046                  * disks, but this is mandatory to avoid failure of large write
1047                  * same requests directed at sequential write required zones of
1048                  * host-managed ZBC disks.
1049                  */
1050                 sdkp->max_ws_blocks =
1051                         round_down(sdkp->max_ws_blocks,
1052                                    bytes_to_logical(sdkp->device,
1053                                                     sdkp->physical_block_size));
1054         }
1055
1056 out:
1057         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1058                                          (logical_block_size >> 9));
1059 }
1060
1061 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1062 {
1063         struct request *rq = scsi_cmd_to_rq(cmd);
1064         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1065
1066         /* flush requests don't perform I/O, zero the S/G table */
1067         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1068
1069         if (cmd->device->use_16_for_sync) {
1070                 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1071                 cmd->cmd_len = 16;
1072         } else {
1073                 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1074                 cmd->cmd_len = 10;
1075         }
1076         cmd->transfersize = 0;
1077         cmd->allowed = sdkp->max_retries;
1078
1079         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1080         return BLK_STS_OK;
1081 }
1082
1083 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1084                                        sector_t lba, unsigned int nr_blocks,
1085                                        unsigned char flags, unsigned int dld)
1086 {
1087         cmd->cmd_len = SD_EXT_CDB_SIZE;
1088         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1089         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1090         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1091         cmd->cmnd[10] = flags;
1092         cmd->cmnd[11] = dld & 0x07;
1093         put_unaligned_be64(lba, &cmd->cmnd[12]);
1094         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1095         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1096
1097         return BLK_STS_OK;
1098 }
1099
1100 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1101                                        sector_t lba, unsigned int nr_blocks,
1102                                        unsigned char flags, unsigned int dld)
1103 {
1104         cmd->cmd_len  = 16;
1105         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1106         cmd->cmnd[1]  = flags | ((dld >> 2) & 0x01);
1107         cmd->cmnd[14] = (dld & 0x03) << 6;
1108         cmd->cmnd[15] = 0;
1109         put_unaligned_be64(lba, &cmd->cmnd[2]);
1110         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1111
1112         return BLK_STS_OK;
1113 }
1114
1115 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1116                                        sector_t lba, unsigned int nr_blocks,
1117                                        unsigned char flags)
1118 {
1119         cmd->cmd_len = 10;
1120         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1121         cmd->cmnd[1] = flags;
1122         cmd->cmnd[6] = 0;
1123         cmd->cmnd[9] = 0;
1124         put_unaligned_be32(lba, &cmd->cmnd[2]);
1125         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1126
1127         return BLK_STS_OK;
1128 }
1129
1130 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1131                                       sector_t lba, unsigned int nr_blocks,
1132                                       unsigned char flags)
1133 {
1134         /* Avoid that 0 blocks gets translated into 256 blocks. */
1135         if (WARN_ON_ONCE(nr_blocks == 0))
1136                 return BLK_STS_IOERR;
1137
1138         if (unlikely(flags & 0x8)) {
1139                 /*
1140                  * This happens only if this drive failed 10byte rw
1141                  * command with ILLEGAL_REQUEST during operation and
1142                  * thus turned off use_10_for_rw.
1143                  */
1144                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1145                 return BLK_STS_IOERR;
1146         }
1147
1148         cmd->cmd_len = 6;
1149         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1150         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1151         cmd->cmnd[2] = (lba >> 8) & 0xff;
1152         cmd->cmnd[3] = lba & 0xff;
1153         cmd->cmnd[4] = nr_blocks;
1154         cmd->cmnd[5] = 0;
1155
1156         return BLK_STS_OK;
1157 }
1158
1159 /*
1160  * Check if a command has a duration limit set. If it does, and the target
1161  * device supports CDL and the feature is enabled, return the limit
1162  * descriptor index to use. Return 0 (no limit) otherwise.
1163  */
1164 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1165 {
1166         struct scsi_device *sdp = sdkp->device;
1167         int hint;
1168
1169         if (!sdp->cdl_supported || !sdp->cdl_enable)
1170                 return 0;
1171
1172         /*
1173          * Use "no limit" if the request ioprio does not specify a duration
1174          * limit hint.
1175          */
1176         hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1177         if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1178             hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1179                 return 0;
1180
1181         return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1182 }
1183
1184 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1185 {
1186         struct request *rq = scsi_cmd_to_rq(cmd);
1187         struct scsi_device *sdp = cmd->device;
1188         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1189         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1190         sector_t threshold;
1191         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1192         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1193         bool write = rq_data_dir(rq) == WRITE;
1194         unsigned char protect, fua;
1195         unsigned int dld;
1196         blk_status_t ret;
1197         unsigned int dif;
1198         bool dix;
1199
1200         ret = scsi_alloc_sgtables(cmd);
1201         if (ret != BLK_STS_OK)
1202                 return ret;
1203
1204         ret = BLK_STS_IOERR;
1205         if (!scsi_device_online(sdp) || sdp->changed) {
1206                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1207                 goto fail;
1208         }
1209
1210         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1211                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1212                 goto fail;
1213         }
1214
1215         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1216                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1217                 goto fail;
1218         }
1219
1220         /*
1221          * Some SD card readers can't handle accesses which touch the
1222          * last one or two logical blocks. Split accesses as needed.
1223          */
1224         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1225
1226         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1227                 if (lba < threshold) {
1228                         /* Access up to the threshold but not beyond */
1229                         nr_blocks = threshold - lba;
1230                 } else {
1231                         /* Access only a single logical block */
1232                         nr_blocks = 1;
1233                 }
1234         }
1235
1236         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1237                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1238                 if (ret)
1239                         goto fail;
1240         }
1241
1242         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1243         dix = scsi_prot_sg_count(cmd);
1244         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1245         dld = sd_cdl_dld(sdkp, cmd);
1246
1247         if (dif || dix)
1248                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1249         else
1250                 protect = 0;
1251
1252         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1253                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1254                                          protect | fua, dld);
1255         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1256                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1257                                          protect | fua, dld);
1258         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1259                    sdp->use_10_for_rw || protect) {
1260                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1261                                          protect | fua);
1262         } else {
1263                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1264                                         protect | fua);
1265         }
1266
1267         if (unlikely(ret != BLK_STS_OK))
1268                 goto fail;
1269
1270         /*
1271          * We shouldn't disconnect in the middle of a sector, so with a dumb
1272          * host adapter, it's safe to assume that we can at least transfer
1273          * this many bytes between each connect / disconnect.
1274          */
1275         cmd->transfersize = sdp->sector_size;
1276         cmd->underflow = nr_blocks << 9;
1277         cmd->allowed = sdkp->max_retries;
1278         cmd->sdb.length = nr_blocks * sdp->sector_size;
1279
1280         SCSI_LOG_HLQUEUE(1,
1281                          scmd_printk(KERN_INFO, cmd,
1282                                      "%s: block=%llu, count=%d\n", __func__,
1283                                      (unsigned long long)blk_rq_pos(rq),
1284                                      blk_rq_sectors(rq)));
1285         SCSI_LOG_HLQUEUE(2,
1286                          scmd_printk(KERN_INFO, cmd,
1287                                      "%s %d/%u 512 byte blocks.\n",
1288                                      write ? "writing" : "reading", nr_blocks,
1289                                      blk_rq_sectors(rq)));
1290
1291         /*
1292          * This indicates that the command is ready from our end to be queued.
1293          */
1294         return BLK_STS_OK;
1295 fail:
1296         scsi_free_sgtables(cmd);
1297         return ret;
1298 }
1299
1300 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1301 {
1302         struct request *rq = scsi_cmd_to_rq(cmd);
1303
1304         switch (req_op(rq)) {
1305         case REQ_OP_DISCARD:
1306                 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1307                 case SD_LBP_UNMAP:
1308                         return sd_setup_unmap_cmnd(cmd);
1309                 case SD_LBP_WS16:
1310                         return sd_setup_write_same16_cmnd(cmd, true);
1311                 case SD_LBP_WS10:
1312                         return sd_setup_write_same10_cmnd(cmd, true);
1313                 case SD_LBP_ZERO:
1314                         return sd_setup_write_same10_cmnd(cmd, false);
1315                 default:
1316                         return BLK_STS_TARGET;
1317                 }
1318         case REQ_OP_WRITE_ZEROES:
1319                 return sd_setup_write_zeroes_cmnd(cmd);
1320         case REQ_OP_FLUSH:
1321                 return sd_setup_flush_cmnd(cmd);
1322         case REQ_OP_READ:
1323         case REQ_OP_WRITE:
1324         case REQ_OP_ZONE_APPEND:
1325                 return sd_setup_read_write_cmnd(cmd);
1326         case REQ_OP_ZONE_RESET:
1327                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1328                                                    false);
1329         case REQ_OP_ZONE_RESET_ALL:
1330                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1331                                                    true);
1332         case REQ_OP_ZONE_OPEN:
1333                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1334         case REQ_OP_ZONE_CLOSE:
1335                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1336         case REQ_OP_ZONE_FINISH:
1337                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1338         default:
1339                 WARN_ON_ONCE(1);
1340                 return BLK_STS_NOTSUPP;
1341         }
1342 }
1343
1344 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1345 {
1346         struct request *rq = scsi_cmd_to_rq(SCpnt);
1347
1348         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1349                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1350 }
1351
1352 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1353 {
1354         if (sdkp->device->removable || sdkp->write_prot) {
1355                 if (disk_check_media_change(disk))
1356                         return true;
1357         }
1358
1359         /*
1360          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1361          * nothing to do with partitions, BLKRRPART is used to force a full
1362          * revalidate after things like a format for historical reasons.
1363          */
1364         return test_bit(GD_NEED_PART_SCAN, &disk->state);
1365 }
1366
1367 /**
1368  *      sd_open - open a scsi disk device
1369  *      @disk: disk to open
1370  *      @mode: open mode
1371  *
1372  *      Returns 0 if successful. Returns a negated errno value in case
1373  *      of error.
1374  *
1375  *      Note: This can be called from a user context (e.g. fsck(1) )
1376  *      or from within the kernel (e.g. as a result of a mount(1) ).
1377  *      In the latter case @inode and @filp carry an abridged amount
1378  *      of information as noted above.
1379  *
1380  *      Locking: called with disk->open_mutex held.
1381  **/
1382 static int sd_open(struct gendisk *disk, blk_mode_t mode)
1383 {
1384         struct scsi_disk *sdkp = scsi_disk(disk);
1385         struct scsi_device *sdev = sdkp->device;
1386         int retval;
1387
1388         if (scsi_device_get(sdev))
1389                 return -ENXIO;
1390
1391         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1392
1393         /*
1394          * If the device is in error recovery, wait until it is done.
1395          * If the device is offline, then disallow any access to it.
1396          */
1397         retval = -ENXIO;
1398         if (!scsi_block_when_processing_errors(sdev))
1399                 goto error_out;
1400
1401         if (sd_need_revalidate(disk, sdkp))
1402                 sd_revalidate_disk(disk);
1403
1404         /*
1405          * If the drive is empty, just let the open fail.
1406          */
1407         retval = -ENOMEDIUM;
1408         if (sdev->removable && !sdkp->media_present &&
1409             !(mode & BLK_OPEN_NDELAY))
1410                 goto error_out;
1411
1412         /*
1413          * If the device has the write protect tab set, have the open fail
1414          * if the user expects to be able to write to the thing.
1415          */
1416         retval = -EROFS;
1417         if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1418                 goto error_out;
1419
1420         /*
1421          * It is possible that the disk changing stuff resulted in
1422          * the device being taken offline.  If this is the case,
1423          * report this to the user, and don't pretend that the
1424          * open actually succeeded.
1425          */
1426         retval = -ENXIO;
1427         if (!scsi_device_online(sdev))
1428                 goto error_out;
1429
1430         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1431                 if (scsi_block_when_processing_errors(sdev))
1432                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1433         }
1434
1435         return 0;
1436
1437 error_out:
1438         scsi_device_put(sdev);
1439         return retval;
1440 }
1441
1442 /**
1443  *      sd_release - invoked when the (last) close(2) is called on this
1444  *      scsi disk.
1445  *      @disk: disk to release
1446  *
1447  *      Returns 0.
1448  *
1449  *      Note: may block (uninterruptible) if error recovery is underway
1450  *      on this disk.
1451  *
1452  *      Locking: called with disk->open_mutex held.
1453  **/
1454 static void sd_release(struct gendisk *disk)
1455 {
1456         struct scsi_disk *sdkp = scsi_disk(disk);
1457         struct scsi_device *sdev = sdkp->device;
1458
1459         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1460
1461         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1462                 if (scsi_block_when_processing_errors(sdev))
1463                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1464         }
1465
1466         scsi_device_put(sdev);
1467 }
1468
1469 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1470 {
1471         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1472         struct scsi_device *sdp = sdkp->device;
1473         struct Scsi_Host *host = sdp->host;
1474         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1475         int diskinfo[4];
1476
1477         /* default to most commonly used values */
1478         diskinfo[0] = 0x40;     /* 1 << 6 */
1479         diskinfo[1] = 0x20;     /* 1 << 5 */
1480         diskinfo[2] = capacity >> 11;
1481
1482         /* override with calculated, extended default, or driver values */
1483         if (host->hostt->bios_param)
1484                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1485         else
1486                 scsicam_bios_param(bdev, capacity, diskinfo);
1487
1488         geo->heads = diskinfo[0];
1489         geo->sectors = diskinfo[1];
1490         geo->cylinders = diskinfo[2];
1491         return 0;
1492 }
1493
1494 /**
1495  *      sd_ioctl - process an ioctl
1496  *      @bdev: target block device
1497  *      @mode: open mode
1498  *      @cmd: ioctl command number
1499  *      @arg: this is third argument given to ioctl(2) system call.
1500  *      Often contains a pointer.
1501  *
1502  *      Returns 0 if successful (some ioctls return positive numbers on
1503  *      success as well). Returns a negated errno value in case of error.
1504  *
1505  *      Note: most ioctls are forward onto the block subsystem or further
1506  *      down in the scsi subsystem.
1507  **/
1508 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1509                     unsigned int cmd, unsigned long arg)
1510 {
1511         struct gendisk *disk = bdev->bd_disk;
1512         struct scsi_disk *sdkp = scsi_disk(disk);
1513         struct scsi_device *sdp = sdkp->device;
1514         void __user *p = (void __user *)arg;
1515         int error;
1516
1517         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1518                                     "cmd=0x%x\n", disk->disk_name, cmd));
1519
1520         if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1521                 return -ENOIOCTLCMD;
1522
1523         /*
1524          * If we are in the middle of error recovery, don't let anyone
1525          * else try and use this device.  Also, if error recovery fails, it
1526          * may try and take the device offline, in which case all further
1527          * access to the device is prohibited.
1528          */
1529         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1530                         (mode & BLK_OPEN_NDELAY));
1531         if (error)
1532                 return error;
1533
1534         if (is_sed_ioctl(cmd))
1535                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1536         return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1537 }
1538
1539 static void set_media_not_present(struct scsi_disk *sdkp)
1540 {
1541         if (sdkp->media_present)
1542                 sdkp->device->changed = 1;
1543
1544         if (sdkp->device->removable) {
1545                 sdkp->media_present = 0;
1546                 sdkp->capacity = 0;
1547         }
1548 }
1549
1550 static int media_not_present(struct scsi_disk *sdkp,
1551                              struct scsi_sense_hdr *sshdr)
1552 {
1553         if (!scsi_sense_valid(sshdr))
1554                 return 0;
1555
1556         /* not invoked for commands that could return deferred errors */
1557         switch (sshdr->sense_key) {
1558         case UNIT_ATTENTION:
1559         case NOT_READY:
1560                 /* medium not present */
1561                 if (sshdr->asc == 0x3A) {
1562                         set_media_not_present(sdkp);
1563                         return 1;
1564                 }
1565         }
1566         return 0;
1567 }
1568
1569 /**
1570  *      sd_check_events - check media events
1571  *      @disk: kernel device descriptor
1572  *      @clearing: disk events currently being cleared
1573  *
1574  *      Returns mask of DISK_EVENT_*.
1575  *
1576  *      Note: this function is invoked from the block subsystem.
1577  **/
1578 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1579 {
1580         struct scsi_disk *sdkp = disk->private_data;
1581         struct scsi_device *sdp;
1582         int retval;
1583         bool disk_changed;
1584
1585         if (!sdkp)
1586                 return 0;
1587
1588         sdp = sdkp->device;
1589         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1590
1591         /*
1592          * If the device is offline, don't send any commands - just pretend as
1593          * if the command failed.  If the device ever comes back online, we
1594          * can deal with it then.  It is only because of unrecoverable errors
1595          * that we would ever take a device offline in the first place.
1596          */
1597         if (!scsi_device_online(sdp)) {
1598                 set_media_not_present(sdkp);
1599                 goto out;
1600         }
1601
1602         /*
1603          * Using TEST_UNIT_READY enables differentiation between drive with
1604          * no cartridge loaded - NOT READY, drive with changed cartridge -
1605          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1606          *
1607          * Drives that auto spin down. eg iomega jaz 1G, will be started
1608          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1609          * sd_revalidate() is called.
1610          */
1611         if (scsi_block_when_processing_errors(sdp)) {
1612                 struct scsi_sense_hdr sshdr = { 0, };
1613
1614                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1615                                               &sshdr);
1616
1617                 /* failed to execute TUR, assume media not present */
1618                 if (retval < 0 || host_byte(retval)) {
1619                         set_media_not_present(sdkp);
1620                         goto out;
1621                 }
1622
1623                 if (media_not_present(sdkp, &sshdr))
1624                         goto out;
1625         }
1626
1627         /*
1628          * For removable scsi disk we have to recognise the presence
1629          * of a disk in the drive.
1630          */
1631         if (!sdkp->media_present)
1632                 sdp->changed = 1;
1633         sdkp->media_present = 1;
1634 out:
1635         /*
1636          * sdp->changed is set under the following conditions:
1637          *
1638          *      Medium present state has changed in either direction.
1639          *      Device has indicated UNIT_ATTENTION.
1640          */
1641         disk_changed = sdp->changed;
1642         sdp->changed = 0;
1643         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1644 }
1645
1646 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1647 {
1648         int retries, res;
1649         struct scsi_device *sdp = sdkp->device;
1650         const int timeout = sdp->request_queue->rq_timeout
1651                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1652         struct scsi_sense_hdr my_sshdr;
1653         const struct scsi_exec_args exec_args = {
1654                 .req_flags = BLK_MQ_REQ_PM,
1655                 /* caller might not be interested in sense, but we need it */
1656                 .sshdr = sshdr ? : &my_sshdr,
1657         };
1658
1659         if (!scsi_device_online(sdp))
1660                 return -ENODEV;
1661
1662         sshdr = exec_args.sshdr;
1663
1664         for (retries = 3; retries > 0; --retries) {
1665                 unsigned char cmd[16] = { 0 };
1666
1667                 if (sdp->use_16_for_sync)
1668                         cmd[0] = SYNCHRONIZE_CACHE_16;
1669                 else
1670                         cmd[0] = SYNCHRONIZE_CACHE;
1671                 /*
1672                  * Leave the rest of the command zero to indicate
1673                  * flush everything.
1674                  */
1675                 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1676                                        timeout, sdkp->max_retries, &exec_args);
1677                 if (res == 0)
1678                         break;
1679         }
1680
1681         if (res) {
1682                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1683
1684                 if (res < 0)
1685                         return res;
1686
1687                 if (scsi_status_is_check_condition(res) &&
1688                     scsi_sense_valid(sshdr)) {
1689                         sd_print_sense_hdr(sdkp, sshdr);
1690
1691                         /* we need to evaluate the error return  */
1692                         if (sshdr->asc == 0x3a ||       /* medium not present */
1693                             sshdr->asc == 0x20 ||       /* invalid command */
1694                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1695                                 /* this is no error here */
1696                                 return 0;
1697                 }
1698
1699                 switch (host_byte(res)) {
1700                 /* ignore errors due to racing a disconnection */
1701                 case DID_BAD_TARGET:
1702                 case DID_NO_CONNECT:
1703                         return 0;
1704                 /* signal the upper layer it might try again */
1705                 case DID_BUS_BUSY:
1706                 case DID_IMM_RETRY:
1707                 case DID_REQUEUE:
1708                 case DID_SOFT_ERROR:
1709                         return -EBUSY;
1710                 default:
1711                         return -EIO;
1712                 }
1713         }
1714         return 0;
1715 }
1716
1717 static void sd_rescan(struct device *dev)
1718 {
1719         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1720
1721         sd_revalidate_disk(sdkp->disk);
1722 }
1723
1724 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1725                 enum blk_unique_id type)
1726 {
1727         struct scsi_device *sdev = scsi_disk(disk)->device;
1728         const struct scsi_vpd *vpd;
1729         const unsigned char *d;
1730         int ret = -ENXIO, len;
1731
1732         rcu_read_lock();
1733         vpd = rcu_dereference(sdev->vpd_pg83);
1734         if (!vpd)
1735                 goto out_unlock;
1736
1737         ret = -EINVAL;
1738         for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1739                 /* we only care about designators with LU association */
1740                 if (((d[1] >> 4) & 0x3) != 0x00)
1741                         continue;
1742                 if ((d[1] & 0xf) != type)
1743                         continue;
1744
1745                 /*
1746                  * Only exit early if a 16-byte descriptor was found.  Otherwise
1747                  * keep looking as one with more entropy might still show up.
1748                  */
1749                 len = d[3];
1750                 if (len != 8 && len != 12 && len != 16)
1751                         continue;
1752                 ret = len;
1753                 memcpy(id, d + 4, len);
1754                 if (len == 16)
1755                         break;
1756         }
1757 out_unlock:
1758         rcu_read_unlock();
1759         return ret;
1760 }
1761
1762 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1763 {
1764         switch (host_byte(result)) {
1765         case DID_TRANSPORT_MARGINAL:
1766         case DID_TRANSPORT_DISRUPTED:
1767         case DID_BUS_BUSY:
1768                 return PR_STS_RETRY_PATH_FAILURE;
1769         case DID_NO_CONNECT:
1770                 return PR_STS_PATH_FAILED;
1771         case DID_TRANSPORT_FAILFAST:
1772                 return PR_STS_PATH_FAST_FAILED;
1773         }
1774
1775         switch (status_byte(result)) {
1776         case SAM_STAT_RESERVATION_CONFLICT:
1777                 return PR_STS_RESERVATION_CONFLICT;
1778         case SAM_STAT_CHECK_CONDITION:
1779                 if (!scsi_sense_valid(sshdr))
1780                         return PR_STS_IOERR;
1781
1782                 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1783                     (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1784                         return -EINVAL;
1785
1786                 fallthrough;
1787         default:
1788                 return PR_STS_IOERR;
1789         }
1790 }
1791
1792 static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1793                             unsigned char *data, int data_len)
1794 {
1795         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1796         struct scsi_device *sdev = sdkp->device;
1797         struct scsi_sense_hdr sshdr;
1798         u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1799         const struct scsi_exec_args exec_args = {
1800                 .sshdr = &sshdr,
1801         };
1802         int result;
1803
1804         put_unaligned_be16(data_len, &cmd[7]);
1805
1806         result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1807                                   SD_TIMEOUT, sdkp->max_retries, &exec_args);
1808         if (scsi_status_is_check_condition(result) &&
1809             scsi_sense_valid(&sshdr)) {
1810                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1811                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1812         }
1813
1814         if (result <= 0)
1815                 return result;
1816
1817         return sd_scsi_to_pr_err(&sshdr, result);
1818 }
1819
1820 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1821 {
1822         int result, i, data_offset, num_copy_keys;
1823         u32 num_keys = keys_info->num_keys;
1824         int data_len = num_keys * 8 + 8;
1825         u8 *data;
1826
1827         data = kzalloc(data_len, GFP_KERNEL);
1828         if (!data)
1829                 return -ENOMEM;
1830
1831         result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1832         if (result)
1833                 goto free_data;
1834
1835         keys_info->generation = get_unaligned_be32(&data[0]);
1836         keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1837
1838         data_offset = 8;
1839         num_copy_keys = min(num_keys, keys_info->num_keys);
1840
1841         for (i = 0; i < num_copy_keys; i++) {
1842                 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1843                 data_offset += 8;
1844         }
1845
1846 free_data:
1847         kfree(data);
1848         return result;
1849 }
1850
1851 static int sd_pr_read_reservation(struct block_device *bdev,
1852                                   struct pr_held_reservation *rsv)
1853 {
1854         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1855         struct scsi_device *sdev = sdkp->device;
1856         u8 data[24] = { };
1857         int result, len;
1858
1859         result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1860         if (result)
1861                 return result;
1862
1863         len = get_unaligned_be32(&data[4]);
1864         if (!len)
1865                 return 0;
1866
1867         /* Make sure we have at least the key and type */
1868         if (len < 14) {
1869                 sdev_printk(KERN_INFO, sdev,
1870                             "READ RESERVATION failed due to short return buffer of %d bytes\n",
1871                             len);
1872                 return -EINVAL;
1873         }
1874
1875         rsv->generation = get_unaligned_be32(&data[0]);
1876         rsv->key = get_unaligned_be64(&data[8]);
1877         rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1878         return 0;
1879 }
1880
1881 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1882                              u64 sa_key, enum scsi_pr_type type, u8 flags)
1883 {
1884         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1885         struct scsi_device *sdev = sdkp->device;
1886         struct scsi_sense_hdr sshdr;
1887         const struct scsi_exec_args exec_args = {
1888                 .sshdr = &sshdr,
1889         };
1890         int result;
1891         u8 cmd[16] = { 0, };
1892         u8 data[24] = { 0, };
1893
1894         cmd[0] = PERSISTENT_RESERVE_OUT;
1895         cmd[1] = sa;
1896         cmd[2] = type;
1897         put_unaligned_be32(sizeof(data), &cmd[5]);
1898
1899         put_unaligned_be64(key, &data[0]);
1900         put_unaligned_be64(sa_key, &data[8]);
1901         data[20] = flags;
1902
1903         result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1904                                   sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1905                                   &exec_args);
1906
1907         if (scsi_status_is_check_condition(result) &&
1908             scsi_sense_valid(&sshdr)) {
1909                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1910                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1911         }
1912
1913         if (result <= 0)
1914                 return result;
1915
1916         return sd_scsi_to_pr_err(&sshdr, result);
1917 }
1918
1919 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1920                 u32 flags)
1921 {
1922         if (flags & ~PR_FL_IGNORE_KEY)
1923                 return -EOPNOTSUPP;
1924         return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1925                         old_key, new_key, 0,
1926                         (1 << 0) /* APTPL */);
1927 }
1928
1929 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1930                 u32 flags)
1931 {
1932         if (flags)
1933                 return -EOPNOTSUPP;
1934         return sd_pr_out_command(bdev, 0x01, key, 0,
1935                                  block_pr_type_to_scsi(type), 0);
1936 }
1937
1938 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1939 {
1940         return sd_pr_out_command(bdev, 0x02, key, 0,
1941                                  block_pr_type_to_scsi(type), 0);
1942 }
1943
1944 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1945                 enum pr_type type, bool abort)
1946 {
1947         return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1948                                  block_pr_type_to_scsi(type), 0);
1949 }
1950
1951 static int sd_pr_clear(struct block_device *bdev, u64 key)
1952 {
1953         return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
1954 }
1955
1956 static const struct pr_ops sd_pr_ops = {
1957         .pr_register    = sd_pr_register,
1958         .pr_reserve     = sd_pr_reserve,
1959         .pr_release     = sd_pr_release,
1960         .pr_preempt     = sd_pr_preempt,
1961         .pr_clear       = sd_pr_clear,
1962         .pr_read_keys   = sd_pr_read_keys,
1963         .pr_read_reservation = sd_pr_read_reservation,
1964 };
1965
1966 static void scsi_disk_free_disk(struct gendisk *disk)
1967 {
1968         struct scsi_disk *sdkp = scsi_disk(disk);
1969
1970         put_device(&sdkp->disk_dev);
1971 }
1972
1973 static const struct block_device_operations sd_fops = {
1974         .owner                  = THIS_MODULE,
1975         .open                   = sd_open,
1976         .release                = sd_release,
1977         .ioctl                  = sd_ioctl,
1978         .getgeo                 = sd_getgeo,
1979         .compat_ioctl           = blkdev_compat_ptr_ioctl,
1980         .check_events           = sd_check_events,
1981         .unlock_native_capacity = sd_unlock_native_capacity,
1982         .report_zones           = sd_zbc_report_zones,
1983         .get_unique_id          = sd_get_unique_id,
1984         .free_disk              = scsi_disk_free_disk,
1985         .pr_ops                 = &sd_pr_ops,
1986 };
1987
1988 /**
1989  *      sd_eh_reset - reset error handling callback
1990  *      @scmd:          sd-issued command that has failed
1991  *
1992  *      This function is called by the SCSI midlayer before starting
1993  *      SCSI EH. When counting medium access failures we have to be
1994  *      careful to register it only only once per device and SCSI EH run;
1995  *      there might be several timed out commands which will cause the
1996  *      'max_medium_access_timeouts' counter to trigger after the first
1997  *      SCSI EH run already and set the device to offline.
1998  *      So this function resets the internal counter before starting SCSI EH.
1999  **/
2000 static void sd_eh_reset(struct scsi_cmnd *scmd)
2001 {
2002         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2003
2004         /* New SCSI EH run, reset gate variable */
2005         sdkp->ignore_medium_access_errors = false;
2006 }
2007
2008 /**
2009  *      sd_eh_action - error handling callback
2010  *      @scmd:          sd-issued command that has failed
2011  *      @eh_disp:       The recovery disposition suggested by the midlayer
2012  *
2013  *      This function is called by the SCSI midlayer upon completion of an
2014  *      error test command (currently TEST UNIT READY). The result of sending
2015  *      the eh command is passed in eh_disp.  We're looking for devices that
2016  *      fail medium access commands but are OK with non access commands like
2017  *      test unit ready (so wrongly see the device as having a successful
2018  *      recovery)
2019  **/
2020 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2021 {
2022         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2023         struct scsi_device *sdev = scmd->device;
2024
2025         if (!scsi_device_online(sdev) ||
2026             !scsi_medium_access_command(scmd) ||
2027             host_byte(scmd->result) != DID_TIME_OUT ||
2028             eh_disp != SUCCESS)
2029                 return eh_disp;
2030
2031         /*
2032          * The device has timed out executing a medium access command.
2033          * However, the TEST UNIT READY command sent during error
2034          * handling completed successfully. Either the device is in the
2035          * process of recovering or has it suffered an internal failure
2036          * that prevents access to the storage medium.
2037          */
2038         if (!sdkp->ignore_medium_access_errors) {
2039                 sdkp->medium_access_timed_out++;
2040                 sdkp->ignore_medium_access_errors = true;
2041         }
2042
2043         /*
2044          * If the device keeps failing read/write commands but TEST UNIT
2045          * READY always completes successfully we assume that medium
2046          * access is no longer possible and take the device offline.
2047          */
2048         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2049                 scmd_printk(KERN_ERR, scmd,
2050                             "Medium access timeout failure. Offlining disk!\n");
2051                 mutex_lock(&sdev->state_mutex);
2052                 scsi_device_set_state(sdev, SDEV_OFFLINE);
2053                 mutex_unlock(&sdev->state_mutex);
2054
2055                 return SUCCESS;
2056         }
2057
2058         return eh_disp;
2059 }
2060
2061 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2062 {
2063         struct request *req = scsi_cmd_to_rq(scmd);
2064         struct scsi_device *sdev = scmd->device;
2065         unsigned int transferred, good_bytes;
2066         u64 start_lba, end_lba, bad_lba;
2067
2068         /*
2069          * Some commands have a payload smaller than the device logical
2070          * block size (e.g. INQUIRY on a 4K disk).
2071          */
2072         if (scsi_bufflen(scmd) <= sdev->sector_size)
2073                 return 0;
2074
2075         /* Check if we have a 'bad_lba' information */
2076         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2077                                      SCSI_SENSE_BUFFERSIZE,
2078                                      &bad_lba))
2079                 return 0;
2080
2081         /*
2082          * If the bad lba was reported incorrectly, we have no idea where
2083          * the error is.
2084          */
2085         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2086         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2087         if (bad_lba < start_lba || bad_lba >= end_lba)
2088                 return 0;
2089
2090         /*
2091          * resid is optional but mostly filled in.  When it's unused,
2092          * its value is zero, so we assume the whole buffer transferred
2093          */
2094         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2095
2096         /* This computation should always be done in terms of the
2097          * resolution of the device's medium.
2098          */
2099         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2100
2101         return min(good_bytes, transferred);
2102 }
2103
2104 /**
2105  *      sd_done - bottom half handler: called when the lower level
2106  *      driver has completed (successfully or otherwise) a scsi command.
2107  *      @SCpnt: mid-level's per command structure.
2108  *
2109  *      Note: potentially run from within an ISR. Must not block.
2110  **/
2111 static int sd_done(struct scsi_cmnd *SCpnt)
2112 {
2113         int result = SCpnt->result;
2114         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2115         unsigned int sector_size = SCpnt->device->sector_size;
2116         unsigned int resid;
2117         struct scsi_sense_hdr sshdr;
2118         struct request *req = scsi_cmd_to_rq(SCpnt);
2119         struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2120         int sense_valid = 0;
2121         int sense_deferred = 0;
2122
2123         switch (req_op(req)) {
2124         case REQ_OP_DISCARD:
2125         case REQ_OP_WRITE_ZEROES:
2126         case REQ_OP_ZONE_RESET:
2127         case REQ_OP_ZONE_RESET_ALL:
2128         case REQ_OP_ZONE_OPEN:
2129         case REQ_OP_ZONE_CLOSE:
2130         case REQ_OP_ZONE_FINISH:
2131                 if (!result) {
2132                         good_bytes = blk_rq_bytes(req);
2133                         scsi_set_resid(SCpnt, 0);
2134                 } else {
2135                         good_bytes = 0;
2136                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
2137                 }
2138                 break;
2139         default:
2140                 /*
2141                  * In case of bogus fw or device, we could end up having
2142                  * an unaligned partial completion. Check this here and force
2143                  * alignment.
2144                  */
2145                 resid = scsi_get_resid(SCpnt);
2146                 if (resid & (sector_size - 1)) {
2147                         sd_printk(KERN_INFO, sdkp,
2148                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2149                                 resid, sector_size);
2150                         scsi_print_command(SCpnt);
2151                         resid = min(scsi_bufflen(SCpnt),
2152                                     round_up(resid, sector_size));
2153                         scsi_set_resid(SCpnt, resid);
2154                 }
2155         }
2156
2157         if (result) {
2158                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2159                 if (sense_valid)
2160                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2161         }
2162         sdkp->medium_access_timed_out = 0;
2163
2164         if (!scsi_status_is_check_condition(result) &&
2165             (!sense_valid || sense_deferred))
2166                 goto out;
2167
2168         switch (sshdr.sense_key) {
2169         case HARDWARE_ERROR:
2170         case MEDIUM_ERROR:
2171                 good_bytes = sd_completed_bytes(SCpnt);
2172                 break;
2173         case RECOVERED_ERROR:
2174                 good_bytes = scsi_bufflen(SCpnt);
2175                 break;
2176         case NO_SENSE:
2177                 /* This indicates a false check condition, so ignore it.  An
2178                  * unknown amount of data was transferred so treat it as an
2179                  * error.
2180                  */
2181                 SCpnt->result = 0;
2182                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2183                 break;
2184         case ABORTED_COMMAND:
2185                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2186                         good_bytes = sd_completed_bytes(SCpnt);
2187                 break;
2188         case ILLEGAL_REQUEST:
2189                 switch (sshdr.asc) {
2190                 case 0x10:      /* DIX: Host detected corruption */
2191                         good_bytes = sd_completed_bytes(SCpnt);
2192                         break;
2193                 case 0x20:      /* INVALID COMMAND OPCODE */
2194                 case 0x24:      /* INVALID FIELD IN CDB */
2195                         switch (SCpnt->cmnd[0]) {
2196                         case UNMAP:
2197                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2198                                 break;
2199                         case WRITE_SAME_16:
2200                         case WRITE_SAME:
2201                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2202                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2203                                 } else {
2204                                         sdkp->device->no_write_same = 1;
2205                                         sd_config_write_same(sdkp);
2206                                         req->rq_flags |= RQF_QUIET;
2207                                 }
2208                                 break;
2209                         }
2210                 }
2211                 break;
2212         default:
2213                 break;
2214         }
2215
2216  out:
2217         if (sd_is_zoned(sdkp))
2218                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2219
2220         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2221                                            "sd_done: completed %d of %d bytes\n",
2222                                            good_bytes, scsi_bufflen(SCpnt)));
2223
2224         return good_bytes;
2225 }
2226
2227 /*
2228  * spinup disk - called only in sd_revalidate_disk()
2229  */
2230 static void
2231 sd_spinup_disk(struct scsi_disk *sdkp)
2232 {
2233         unsigned char cmd[10];
2234         unsigned long spintime_expire = 0;
2235         int retries, spintime;
2236         unsigned int the_result;
2237         struct scsi_sense_hdr sshdr;
2238         const struct scsi_exec_args exec_args = {
2239                 .sshdr = &sshdr,
2240         };
2241         int sense_valid = 0;
2242
2243         spintime = 0;
2244
2245         /* Spin up drives, as required.  Only do this at boot time */
2246         /* Spinup needs to be done for module loads too. */
2247         do {
2248                 retries = 0;
2249
2250                 do {
2251                         bool media_was_present = sdkp->media_present;
2252
2253                         cmd[0] = TEST_UNIT_READY;
2254                         memset((void *) &cmd[1], 0, 9);
2255
2256                         the_result = scsi_execute_cmd(sdkp->device, cmd,
2257                                                       REQ_OP_DRV_IN, NULL, 0,
2258                                                       SD_TIMEOUT,
2259                                                       sdkp->max_retries,
2260                                                       &exec_args);
2261
2262                         if (the_result > 0) {
2263                                 /*
2264                                  * If the drive has indicated to us that it
2265                                  * doesn't have any media in it, don't bother
2266                                  * with any more polling.
2267                                  */
2268                                 if (media_not_present(sdkp, &sshdr)) {
2269                                         if (media_was_present)
2270                                                 sd_printk(KERN_NOTICE, sdkp,
2271                                                           "Media removed, stopped polling\n");
2272                                         return;
2273                                 }
2274
2275                                 sense_valid = scsi_sense_valid(&sshdr);
2276                         }
2277                         retries++;
2278                 } while (retries < 3 &&
2279                          (!scsi_status_is_good(the_result) ||
2280                           (scsi_status_is_check_condition(the_result) &&
2281                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2282
2283                 if (!scsi_status_is_check_condition(the_result)) {
2284                         /* no sense, TUR either succeeded or failed
2285                          * with a status error */
2286                         if(!spintime && !scsi_status_is_good(the_result)) {
2287                                 sd_print_result(sdkp, "Test Unit Ready failed",
2288                                                 the_result);
2289                         }
2290                         break;
2291                 }
2292
2293                 /*
2294                  * The device does not want the automatic start to be issued.
2295                  */
2296                 if (sdkp->device->no_start_on_add)
2297                         break;
2298
2299                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2300                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2301                                 break;  /* manual intervention required */
2302                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2303                                 break;  /* standby */
2304                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2305                                 break;  /* unavailable */
2306                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2307                                 break;  /* sanitize in progress */
2308                         if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2309                                 break;  /* depopulation in progress */
2310                         if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2311                                 break;  /* depopulation restoration in progress */
2312                         /*
2313                          * Issue command to spin up drive when not ready
2314                          */
2315                         if (!spintime) {
2316                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2317                                 cmd[0] = START_STOP;
2318                                 cmd[1] = 1;     /* Return immediately */
2319                                 memset((void *) &cmd[2], 0, 8);
2320                                 cmd[4] = 1;     /* Start spin cycle */
2321                                 if (sdkp->device->start_stop_pwr_cond)
2322                                         cmd[4] |= 1 << 4;
2323                                 scsi_execute_cmd(sdkp->device, cmd,
2324                                                  REQ_OP_DRV_IN, NULL, 0,
2325                                                  SD_TIMEOUT, sdkp->max_retries,
2326                                                  &exec_args);
2327                                 spintime_expire = jiffies + 100 * HZ;
2328                                 spintime = 1;
2329                         }
2330                         /* Wait 1 second for next try */
2331                         msleep(1000);
2332                         printk(KERN_CONT ".");
2333
2334                 /*
2335                  * Wait for USB flash devices with slow firmware.
2336                  * Yes, this sense key/ASC combination shouldn't
2337                  * occur here.  It's characteristic of these devices.
2338                  */
2339                 } else if (sense_valid &&
2340                                 sshdr.sense_key == UNIT_ATTENTION &&
2341                                 sshdr.asc == 0x28) {
2342                         if (!spintime) {
2343                                 spintime_expire = jiffies + 5 * HZ;
2344                                 spintime = 1;
2345                         }
2346                         /* Wait 1 second for next try */
2347                         msleep(1000);
2348                 } else {
2349                         /* we don't understand the sense code, so it's
2350                          * probably pointless to loop */
2351                         if(!spintime) {
2352                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2353                                 sd_print_sense_hdr(sdkp, &sshdr);
2354                         }
2355                         break;
2356                 }
2357
2358         } while (spintime && time_before_eq(jiffies, spintime_expire));
2359
2360         if (spintime) {
2361                 if (scsi_status_is_good(the_result))
2362                         printk(KERN_CONT "ready\n");
2363                 else
2364                         printk(KERN_CONT "not responding...\n");
2365         }
2366 }
2367
2368 /*
2369  * Determine whether disk supports Data Integrity Field.
2370  */
2371 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2372 {
2373         struct scsi_device *sdp = sdkp->device;
2374         u8 type;
2375
2376         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2377                 sdkp->protection_type = 0;
2378                 return 0;
2379         }
2380
2381         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2382
2383         if (type > T10_PI_TYPE3_PROTECTION) {
2384                 sd_printk(KERN_ERR, sdkp, "formatted with unsupported"  \
2385                           " protection type %u. Disabling disk!\n",
2386                           type);
2387                 sdkp->protection_type = 0;
2388                 return -ENODEV;
2389         }
2390
2391         sdkp->protection_type = type;
2392
2393         return 0;
2394 }
2395
2396 static void sd_config_protection(struct scsi_disk *sdkp)
2397 {
2398         struct scsi_device *sdp = sdkp->device;
2399
2400         sd_dif_config_host(sdkp);
2401
2402         if (!sdkp->protection_type)
2403                 return;
2404
2405         if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2406                 sd_first_printk(KERN_NOTICE, sdkp,
2407                                 "Disabling DIF Type %u protection\n",
2408                                 sdkp->protection_type);
2409                 sdkp->protection_type = 0;
2410         }
2411
2412         sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2413                         sdkp->protection_type);
2414 }
2415
2416 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2417                         struct scsi_sense_hdr *sshdr, int sense_valid,
2418                         int the_result)
2419 {
2420         if (sense_valid)
2421                 sd_print_sense_hdr(sdkp, sshdr);
2422         else
2423                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2424
2425         /*
2426          * Set dirty bit for removable devices if not ready -
2427          * sometimes drives will not report this properly.
2428          */
2429         if (sdp->removable &&
2430             sense_valid && sshdr->sense_key == NOT_READY)
2431                 set_media_not_present(sdkp);
2432
2433         /*
2434          * We used to set media_present to 0 here to indicate no media
2435          * in the drive, but some drives fail read capacity even with
2436          * media present, so we can't do that.
2437          */
2438         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2439 }
2440
2441 #define RC16_LEN 32
2442 #if RC16_LEN > SD_BUF_SIZE
2443 #error RC16_LEN must not be more than SD_BUF_SIZE
2444 #endif
2445
2446 #define READ_CAPACITY_RETRIES_ON_RESET  10
2447
2448 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2449                                                 unsigned char *buffer)
2450 {
2451         unsigned char cmd[16];
2452         struct scsi_sense_hdr sshdr;
2453         const struct scsi_exec_args exec_args = {
2454                 .sshdr = &sshdr,
2455         };
2456         int sense_valid = 0;
2457         int the_result;
2458         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2459         unsigned int alignment;
2460         unsigned long long lba;
2461         unsigned sector_size;
2462
2463         if (sdp->no_read_capacity_16)
2464                 return -EINVAL;
2465
2466         do {
2467                 memset(cmd, 0, 16);
2468                 cmd[0] = SERVICE_ACTION_IN_16;
2469                 cmd[1] = SAI_READ_CAPACITY_16;
2470                 cmd[13] = RC16_LEN;
2471                 memset(buffer, 0, RC16_LEN);
2472
2473                 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2474                                               buffer, RC16_LEN, SD_TIMEOUT,
2475                                               sdkp->max_retries, &exec_args);
2476                 if (the_result > 0) {
2477                         if (media_not_present(sdkp, &sshdr))
2478                                 return -ENODEV;
2479
2480                         sense_valid = scsi_sense_valid(&sshdr);
2481                         if (sense_valid &&
2482                             sshdr.sense_key == ILLEGAL_REQUEST &&
2483                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2484                             sshdr.ascq == 0x00)
2485                                 /* Invalid Command Operation Code or
2486                                  * Invalid Field in CDB, just retry
2487                                  * silently with RC10 */
2488                                 return -EINVAL;
2489                         if (sense_valid &&
2490                             sshdr.sense_key == UNIT_ATTENTION &&
2491                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2492                                 /* Device reset might occur several times,
2493                                  * give it one more chance */
2494                                 if (--reset_retries > 0)
2495                                         continue;
2496                 }
2497                 retries--;
2498
2499         } while (the_result && retries);
2500
2501         if (the_result) {
2502                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2503                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2504                 return -EINVAL;
2505         }
2506
2507         sector_size = get_unaligned_be32(&buffer[8]);
2508         lba = get_unaligned_be64(&buffer[0]);
2509
2510         if (sd_read_protection_type(sdkp, buffer) < 0) {
2511                 sdkp->capacity = 0;
2512                 return -ENODEV;
2513         }
2514
2515         /* Logical blocks per physical block exponent */
2516         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2517
2518         /* RC basis */
2519         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2520
2521         /* Lowest aligned logical block */
2522         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2523         blk_queue_alignment_offset(sdp->request_queue, alignment);
2524         if (alignment && sdkp->first_scan)
2525                 sd_printk(KERN_NOTICE, sdkp,
2526                           "physical block alignment offset: %u\n", alignment);
2527
2528         if (buffer[14] & 0x80) { /* LBPME */
2529                 sdkp->lbpme = 1;
2530
2531                 if (buffer[14] & 0x40) /* LBPRZ */
2532                         sdkp->lbprz = 1;
2533
2534                 sd_config_discard(sdkp, SD_LBP_WS16);
2535         }
2536
2537         sdkp->capacity = lba + 1;
2538         return sector_size;
2539 }
2540
2541 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2542                                                 unsigned char *buffer)
2543 {
2544         unsigned char cmd[16];
2545         struct scsi_sense_hdr sshdr;
2546         const struct scsi_exec_args exec_args = {
2547                 .sshdr = &sshdr,
2548         };
2549         int sense_valid = 0;
2550         int the_result;
2551         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2552         sector_t lba;
2553         unsigned sector_size;
2554
2555         do {
2556                 cmd[0] = READ_CAPACITY;
2557                 memset(&cmd[1], 0, 9);
2558                 memset(buffer, 0, 8);
2559
2560                 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2561                                               8, SD_TIMEOUT, sdkp->max_retries,
2562                                               &exec_args);
2563
2564                 if (media_not_present(sdkp, &sshdr))
2565                         return -ENODEV;
2566
2567                 if (the_result > 0) {
2568                         sense_valid = scsi_sense_valid(&sshdr);
2569                         if (sense_valid &&
2570                             sshdr.sense_key == UNIT_ATTENTION &&
2571                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2572                                 /* Device reset might occur several times,
2573                                  * give it one more chance */
2574                                 if (--reset_retries > 0)
2575                                         continue;
2576                 }
2577                 retries--;
2578
2579         } while (the_result && retries);
2580
2581         if (the_result) {
2582                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2583                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2584                 return -EINVAL;
2585         }
2586
2587         sector_size = get_unaligned_be32(&buffer[4]);
2588         lba = get_unaligned_be32(&buffer[0]);
2589
2590         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2591                 /* Some buggy (usb cardreader) devices return an lba of
2592                    0xffffffff when the want to report a size of 0 (with
2593                    which they really mean no media is present) */
2594                 sdkp->capacity = 0;
2595                 sdkp->physical_block_size = sector_size;
2596                 return sector_size;
2597         }
2598
2599         sdkp->capacity = lba + 1;
2600         sdkp->physical_block_size = sector_size;
2601         return sector_size;
2602 }
2603
2604 static int sd_try_rc16_first(struct scsi_device *sdp)
2605 {
2606         if (sdp->host->max_cmd_len < 16)
2607                 return 0;
2608         if (sdp->try_rc_10_first)
2609                 return 0;
2610         if (sdp->scsi_level > SCSI_SPC_2)
2611                 return 1;
2612         if (scsi_device_protection(sdp))
2613                 return 1;
2614         return 0;
2615 }
2616
2617 /*
2618  * read disk capacity
2619  */
2620 static void
2621 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2622 {
2623         int sector_size;
2624         struct scsi_device *sdp = sdkp->device;
2625
2626         if (sd_try_rc16_first(sdp)) {
2627                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2628                 if (sector_size == -EOVERFLOW)
2629                         goto got_data;
2630                 if (sector_size == -ENODEV)
2631                         return;
2632                 if (sector_size < 0)
2633                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2634                 if (sector_size < 0)
2635                         return;
2636         } else {
2637                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2638                 if (sector_size == -EOVERFLOW)
2639                         goto got_data;
2640                 if (sector_size < 0)
2641                         return;
2642                 if ((sizeof(sdkp->capacity) > 4) &&
2643                     (sdkp->capacity > 0xffffffffULL)) {
2644                         int old_sector_size = sector_size;
2645                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2646                                         "Trying to use READ CAPACITY(16).\n");
2647                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2648                         if (sector_size < 0) {
2649                                 sd_printk(KERN_NOTICE, sdkp,
2650                                         "Using 0xffffffff as device size\n");
2651                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2652                                 sector_size = old_sector_size;
2653                                 goto got_data;
2654                         }
2655                         /* Remember that READ CAPACITY(16) succeeded */
2656                         sdp->try_rc_10_first = 0;
2657                 }
2658         }
2659
2660         /* Some devices are known to return the total number of blocks,
2661          * not the highest block number.  Some devices have versions
2662          * which do this and others which do not.  Some devices we might
2663          * suspect of doing this but we don't know for certain.
2664          *
2665          * If we know the reported capacity is wrong, decrement it.  If
2666          * we can only guess, then assume the number of blocks is even
2667          * (usually true but not always) and err on the side of lowering
2668          * the capacity.
2669          */
2670         if (sdp->fix_capacity ||
2671             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2672                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2673                                 "from its reported value: %llu\n",
2674                                 (unsigned long long) sdkp->capacity);
2675                 --sdkp->capacity;
2676         }
2677
2678 got_data:
2679         if (sector_size == 0) {
2680                 sector_size = 512;
2681                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2682                           "assuming 512.\n");
2683         }
2684
2685         if (sector_size != 512 &&
2686             sector_size != 1024 &&
2687             sector_size != 2048 &&
2688             sector_size != 4096) {
2689                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2690                           sector_size);
2691                 /*
2692                  * The user might want to re-format the drive with
2693                  * a supported sectorsize.  Once this happens, it
2694                  * would be relatively trivial to set the thing up.
2695                  * For this reason, we leave the thing in the table.
2696                  */
2697                 sdkp->capacity = 0;
2698                 /*
2699                  * set a bogus sector size so the normal read/write
2700                  * logic in the block layer will eventually refuse any
2701                  * request on this device without tripping over power
2702                  * of two sector size assumptions
2703                  */
2704                 sector_size = 512;
2705         }
2706         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2707         blk_queue_physical_block_size(sdp->request_queue,
2708                                       sdkp->physical_block_size);
2709         sdkp->device->sector_size = sector_size;
2710
2711         if (sdkp->capacity > 0xffffffff)
2712                 sdp->use_16_for_rw = 1;
2713
2714 }
2715
2716 /*
2717  * Print disk capacity
2718  */
2719 static void
2720 sd_print_capacity(struct scsi_disk *sdkp,
2721                   sector_t old_capacity)
2722 {
2723         int sector_size = sdkp->device->sector_size;
2724         char cap_str_2[10], cap_str_10[10];
2725
2726         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2727                 return;
2728
2729         string_get_size(sdkp->capacity, sector_size,
2730                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2731         string_get_size(sdkp->capacity, sector_size,
2732                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2733
2734         sd_printk(KERN_NOTICE, sdkp,
2735                   "%llu %d-byte logical blocks: (%s/%s)\n",
2736                   (unsigned long long)sdkp->capacity,
2737                   sector_size, cap_str_10, cap_str_2);
2738
2739         if (sdkp->physical_block_size != sector_size)
2740                 sd_printk(KERN_NOTICE, sdkp,
2741                           "%u-byte physical blocks\n",
2742                           sdkp->physical_block_size);
2743 }
2744
2745 /* called with buffer of length 512 */
2746 static inline int
2747 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2748                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2749                  struct scsi_sense_hdr *sshdr)
2750 {
2751         /*
2752          * If we must use MODE SENSE(10), make sure that the buffer length
2753          * is at least 8 bytes so that the mode sense header fits.
2754          */
2755         if (sdkp->device->use_10_for_ms && len < 8)
2756                 len = 8;
2757
2758         return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2759                                SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2760 }
2761
2762 /*
2763  * read write protect setting, if possible - called only in sd_revalidate_disk()
2764  * called with buffer of length SD_BUF_SIZE
2765  */
2766 static void
2767 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2768 {
2769         int res;
2770         struct scsi_device *sdp = sdkp->device;
2771         struct scsi_mode_data data;
2772         int old_wp = sdkp->write_prot;
2773
2774         set_disk_ro(sdkp->disk, 0);
2775         if (sdp->skip_ms_page_3f) {
2776                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2777                 return;
2778         }
2779
2780         if (sdp->use_192_bytes_for_3f) {
2781                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2782         } else {
2783                 /*
2784                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2785                  * We have to start carefully: some devices hang if we ask
2786                  * for more than is available.
2787                  */
2788                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2789
2790                 /*
2791                  * Second attempt: ask for page 0 When only page 0 is
2792                  * implemented, a request for page 3F may return Sense Key
2793                  * 5: Illegal Request, Sense Code 24: Invalid field in
2794                  * CDB.
2795                  */
2796                 if (res < 0)
2797                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2798
2799                 /*
2800                  * Third attempt: ask 255 bytes, as we did earlier.
2801                  */
2802                 if (res < 0)
2803                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2804                                                &data, NULL);
2805         }
2806
2807         if (res < 0) {
2808                 sd_first_printk(KERN_WARNING, sdkp,
2809                           "Test WP failed, assume Write Enabled\n");
2810         } else {
2811                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2812                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2813                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2814                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2815                                   sdkp->write_prot ? "on" : "off");
2816                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2817                 }
2818         }
2819 }
2820
2821 /*
2822  * sd_read_cache_type - called only from sd_revalidate_disk()
2823  * called with buffer of length SD_BUF_SIZE
2824  */
2825 static void
2826 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2827 {
2828         int len = 0, res;
2829         struct scsi_device *sdp = sdkp->device;
2830
2831         int dbd;
2832         int modepage;
2833         int first_len;
2834         struct scsi_mode_data data;
2835         struct scsi_sense_hdr sshdr;
2836         int old_wce = sdkp->WCE;
2837         int old_rcd = sdkp->RCD;
2838         int old_dpofua = sdkp->DPOFUA;
2839
2840
2841         if (sdkp->cache_override)
2842                 return;
2843
2844         first_len = 4;
2845         if (sdp->skip_ms_page_8) {
2846                 if (sdp->type == TYPE_RBC)
2847                         goto defaults;
2848                 else {
2849                         if (sdp->skip_ms_page_3f)
2850                                 goto defaults;
2851                         modepage = 0x3F;
2852                         if (sdp->use_192_bytes_for_3f)
2853                                 first_len = 192;
2854                         dbd = 0;
2855                 }
2856         } else if (sdp->type == TYPE_RBC) {
2857                 modepage = 6;
2858                 dbd = 8;
2859         } else {
2860                 modepage = 8;
2861                 dbd = 0;
2862         }
2863
2864         /* cautiously ask */
2865         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2866                         &data, &sshdr);
2867
2868         if (res < 0)
2869                 goto bad_sense;
2870
2871         if (!data.header_length) {
2872                 modepage = 6;
2873                 first_len = 0;
2874                 sd_first_printk(KERN_ERR, sdkp,
2875                                 "Missing header in MODE_SENSE response\n");
2876         }
2877
2878         /* that went OK, now ask for the proper length */
2879         len = data.length;
2880
2881         /*
2882          * We're only interested in the first three bytes, actually.
2883          * But the data cache page is defined for the first 20.
2884          */
2885         if (len < 3)
2886                 goto bad_sense;
2887         else if (len > SD_BUF_SIZE) {
2888                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2889                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2890                 len = SD_BUF_SIZE;
2891         }
2892         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2893                 len = 192;
2894
2895         /* Get the data */
2896         if (len > first_len)
2897                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2898                                 &data, &sshdr);
2899
2900         if (!res) {
2901                 int offset = data.header_length + data.block_descriptor_length;
2902
2903                 while (offset < len) {
2904                         u8 page_code = buffer[offset] & 0x3F;
2905                         u8 spf       = buffer[offset] & 0x40;
2906
2907                         if (page_code == 8 || page_code == 6) {
2908                                 /* We're interested only in the first 3 bytes.
2909                                  */
2910                                 if (len - offset <= 2) {
2911                                         sd_first_printk(KERN_ERR, sdkp,
2912                                                 "Incomplete mode parameter "
2913                                                         "data\n");
2914                                         goto defaults;
2915                                 } else {
2916                                         modepage = page_code;
2917                                         goto Page_found;
2918                                 }
2919                         } else {
2920                                 /* Go to the next page */
2921                                 if (spf && len - offset > 3)
2922                                         offset += 4 + (buffer[offset+2] << 8) +
2923                                                 buffer[offset+3];
2924                                 else if (!spf && len - offset > 1)
2925                                         offset += 2 + buffer[offset+1];
2926                                 else {
2927                                         sd_first_printk(KERN_ERR, sdkp,
2928                                                         "Incomplete mode "
2929                                                         "parameter data\n");
2930                                         goto defaults;
2931                                 }
2932                         }
2933                 }
2934
2935                 sd_first_printk(KERN_WARNING, sdkp,
2936                                 "No Caching mode page found\n");
2937                 goto defaults;
2938
2939         Page_found:
2940                 if (modepage == 8) {
2941                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2942                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2943                 } else {
2944                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2945                         sdkp->RCD = 0;
2946                 }
2947
2948                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2949                 if (sdp->broken_fua) {
2950                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2951                         sdkp->DPOFUA = 0;
2952                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2953                            !sdkp->device->use_16_for_rw) {
2954                         sd_first_printk(KERN_NOTICE, sdkp,
2955                                   "Uses READ/WRITE(6), disabling FUA\n");
2956                         sdkp->DPOFUA = 0;
2957                 }
2958
2959                 /* No cache flush allowed for write protected devices */
2960                 if (sdkp->WCE && sdkp->write_prot)
2961                         sdkp->WCE = 0;
2962
2963                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2964                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2965                         sd_printk(KERN_NOTICE, sdkp,
2966                                   "Write cache: %s, read cache: %s, %s\n",
2967                                   sdkp->WCE ? "enabled" : "disabled",
2968                                   sdkp->RCD ? "disabled" : "enabled",
2969                                   sdkp->DPOFUA ? "supports DPO and FUA"
2970                                   : "doesn't support DPO or FUA");
2971
2972                 return;
2973         }
2974
2975 bad_sense:
2976         if (res == -EIO && scsi_sense_valid(&sshdr) &&
2977             sshdr.sense_key == ILLEGAL_REQUEST &&
2978             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2979                 /* Invalid field in CDB */
2980                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2981         else
2982                 sd_first_printk(KERN_ERR, sdkp,
2983                                 "Asking for cache data failed\n");
2984
2985 defaults:
2986         if (sdp->wce_default_on) {
2987                 sd_first_printk(KERN_NOTICE, sdkp,
2988                                 "Assuming drive cache: write back\n");
2989                 sdkp->WCE = 1;
2990         } else {
2991                 sd_first_printk(KERN_WARNING, sdkp,
2992                                 "Assuming drive cache: write through\n");
2993                 sdkp->WCE = 0;
2994         }
2995         sdkp->RCD = 0;
2996         sdkp->DPOFUA = 0;
2997 }
2998
2999 /*
3000  * The ATO bit indicates whether the DIF application tag is available
3001  * for use by the operating system.
3002  */
3003 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3004 {
3005         int res, offset;
3006         struct scsi_device *sdp = sdkp->device;
3007         struct scsi_mode_data data;
3008         struct scsi_sense_hdr sshdr;
3009
3010         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3011                 return;
3012
3013         if (sdkp->protection_type == 0)
3014                 return;
3015
3016         res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3017                               sdkp->max_retries, &data, &sshdr);
3018
3019         if (res < 0 || !data.header_length ||
3020             data.length < 6) {
3021                 sd_first_printk(KERN_WARNING, sdkp,
3022                           "getting Control mode page failed, assume no ATO\n");
3023
3024                 if (res == -EIO && scsi_sense_valid(&sshdr))
3025                         sd_print_sense_hdr(sdkp, &sshdr);
3026
3027                 return;
3028         }
3029
3030         offset = data.header_length + data.block_descriptor_length;
3031
3032         if ((buffer[offset] & 0x3f) != 0x0a) {
3033                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3034                 return;
3035         }
3036
3037         if ((buffer[offset + 5] & 0x80) == 0)
3038                 return;
3039
3040         sdkp->ATO = 1;
3041
3042         return;
3043 }
3044
3045 /**
3046  * sd_read_block_limits - Query disk device for preferred I/O sizes.
3047  * @sdkp: disk to query
3048  */
3049 static void sd_read_block_limits(struct scsi_disk *sdkp)
3050 {
3051         struct scsi_vpd *vpd;
3052
3053         rcu_read_lock();
3054
3055         vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3056         if (!vpd || vpd->len < 16)
3057                 goto out;
3058
3059         sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3060         sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3061         sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3062
3063         if (vpd->len >= 64) {
3064                 unsigned int lba_count, desc_count;
3065
3066                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3067
3068                 if (!sdkp->lbpme)
3069                         goto out;
3070
3071                 lba_count = get_unaligned_be32(&vpd->data[20]);
3072                 desc_count = get_unaligned_be32(&vpd->data[24]);
3073
3074                 if (lba_count && desc_count)
3075                         sdkp->max_unmap_blocks = lba_count;
3076
3077                 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3078
3079                 if (vpd->data[32] & 0x80)
3080                         sdkp->unmap_alignment =
3081                                 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3082
3083                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3084
3085                         if (sdkp->max_unmap_blocks)
3086                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
3087                         else
3088                                 sd_config_discard(sdkp, SD_LBP_WS16);
3089
3090                 } else {        /* LBP VPD page tells us what to use */
3091                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
3092                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
3093                         else if (sdkp->lbpws)
3094                                 sd_config_discard(sdkp, SD_LBP_WS16);
3095                         else if (sdkp->lbpws10)
3096                                 sd_config_discard(sdkp, SD_LBP_WS10);
3097                         else
3098                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
3099                 }
3100         }
3101
3102  out:
3103         rcu_read_unlock();
3104 }
3105
3106 /**
3107  * sd_read_block_characteristics - Query block dev. characteristics
3108  * @sdkp: disk to query
3109  */
3110 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3111 {
3112         struct request_queue *q = sdkp->disk->queue;
3113         struct scsi_vpd *vpd;
3114         u16 rot;
3115         u8 zoned;
3116
3117         rcu_read_lock();
3118         vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3119
3120         if (!vpd || vpd->len < 8) {
3121                 rcu_read_unlock();
3122                 return;
3123         }
3124
3125         rot = get_unaligned_be16(&vpd->data[4]);
3126         zoned = (vpd->data[8] >> 4) & 3;
3127         rcu_read_unlock();
3128
3129         if (rot == 1) {
3130                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3131                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3132         }
3133
3134         if (sdkp->device->type == TYPE_ZBC) {
3135                 /*
3136                  * Host-managed: Per ZBC and ZAC specifications, writes in
3137                  * sequential write required zones of host-managed devices must
3138                  * be aligned to the device physical block size.
3139                  */
3140                 disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
3141                 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3142         } else {
3143                 sdkp->zoned = zoned;
3144                 if (sdkp->zoned == 1) {
3145                         /* Host-aware */
3146                         disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
3147                 } else {
3148                         /* Regular disk or drive managed disk */
3149                         disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3150                 }
3151         }
3152
3153         if (!sdkp->first_scan)
3154                 return;
3155
3156         if (blk_queue_is_zoned(q)) {
3157                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3158                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3159         } else {
3160                 if (sdkp->zoned == 1)
3161                         sd_printk(KERN_NOTICE, sdkp,
3162                                   "Host-aware SMR disk used as regular disk\n");
3163                 else if (sdkp->zoned == 2)
3164                         sd_printk(KERN_NOTICE, sdkp,
3165                                   "Drive-managed SMR disk\n");
3166         }
3167 }
3168
3169 /**
3170  * sd_read_block_provisioning - Query provisioning VPD page
3171  * @sdkp: disk to query
3172  */
3173 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3174 {
3175         struct scsi_vpd *vpd;
3176
3177         if (sdkp->lbpme == 0)
3178                 return;
3179
3180         rcu_read_lock();
3181         vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3182
3183         if (!vpd || vpd->len < 8) {
3184                 rcu_read_unlock();
3185                 return;
3186         }
3187
3188         sdkp->lbpvpd    = 1;
3189         sdkp->lbpu      = (vpd->data[5] >> 7) & 1; /* UNMAP */
3190         sdkp->lbpws     = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3191         sdkp->lbpws10   = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3192         rcu_read_unlock();
3193 }
3194
3195 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3196 {
3197         struct scsi_device *sdev = sdkp->device;
3198
3199         if (sdev->host->no_write_same) {
3200                 sdev->no_write_same = 1;
3201
3202                 return;
3203         }
3204
3205         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3206                 struct scsi_vpd *vpd;
3207
3208                 sdev->no_report_opcodes = 1;
3209
3210                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3211                  * CODES is unsupported and the device has an ATA
3212                  * Information VPD page (SAT).
3213                  */
3214                 rcu_read_lock();
3215                 vpd = rcu_dereference(sdev->vpd_pg89);
3216                 if (vpd)
3217                         sdev->no_write_same = 1;
3218                 rcu_read_unlock();
3219         }
3220
3221         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3222                 sdkp->ws16 = 1;
3223
3224         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3225                 sdkp->ws10 = 1;
3226 }
3227
3228 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3229 {
3230         struct scsi_device *sdev = sdkp->device;
3231
3232         if (!sdev->security_supported)
3233                 return;
3234
3235         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3236                         SECURITY_PROTOCOL_IN, 0) == 1 &&
3237             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3238                         SECURITY_PROTOCOL_OUT, 0) == 1)
3239                 sdkp->security = 1;
3240 }
3241
3242 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3243 {
3244         return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3245 }
3246
3247 /**
3248  * sd_read_cpr - Query concurrent positioning ranges
3249  * @sdkp:       disk to query
3250  */
3251 static void sd_read_cpr(struct scsi_disk *sdkp)
3252 {
3253         struct blk_independent_access_ranges *iars = NULL;
3254         unsigned char *buffer = NULL;
3255         unsigned int nr_cpr = 0;
3256         int i, vpd_len, buf_len = SD_BUF_SIZE;
3257         u8 *desc;
3258
3259         /*
3260          * We need to have the capacity set first for the block layer to be
3261          * able to check the ranges.
3262          */
3263         if (sdkp->first_scan)
3264                 return;
3265
3266         if (!sdkp->capacity)
3267                 goto out;
3268
3269         /*
3270          * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3271          * leading to a maximum page size of 64 + 256*32 bytes.
3272          */
3273         buf_len = 64 + 256*32;
3274         buffer = kmalloc(buf_len, GFP_KERNEL);
3275         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3276                 goto out;
3277
3278         /* We must have at least a 64B header and one 32B range descriptor */
3279         vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3280         if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3281                 sd_printk(KERN_ERR, sdkp,
3282                           "Invalid Concurrent Positioning Ranges VPD page\n");
3283                 goto out;
3284         }
3285
3286         nr_cpr = (vpd_len - 64) / 32;
3287         if (nr_cpr == 1) {
3288                 nr_cpr = 0;
3289                 goto out;
3290         }
3291
3292         iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3293         if (!iars) {
3294                 nr_cpr = 0;
3295                 goto out;
3296         }
3297
3298         desc = &buffer[64];
3299         for (i = 0; i < nr_cpr; i++, desc += 32) {
3300                 if (desc[0] != i) {
3301                         sd_printk(KERN_ERR, sdkp,
3302                                 "Invalid Concurrent Positioning Range number\n");
3303                         nr_cpr = 0;
3304                         break;
3305                 }
3306
3307                 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3308                 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3309         }
3310
3311 out:
3312         disk_set_independent_access_ranges(sdkp->disk, iars);
3313         if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3314                 sd_printk(KERN_NOTICE, sdkp,
3315                           "%u concurrent positioning ranges\n", nr_cpr);
3316                 sdkp->nr_actuators = nr_cpr;
3317         }
3318
3319         kfree(buffer);
3320 }
3321
3322 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3323 {
3324         struct scsi_device *sdp = sdkp->device;
3325         unsigned int min_xfer_bytes =
3326                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3327
3328         if (sdkp->min_xfer_blocks == 0)
3329                 return false;
3330
3331         if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3332                 sd_first_printk(KERN_WARNING, sdkp,
3333                                 "Preferred minimum I/O size %u bytes not a " \
3334                                 "multiple of physical block size (%u bytes)\n",
3335                                 min_xfer_bytes, sdkp->physical_block_size);
3336                 sdkp->min_xfer_blocks = 0;
3337                 return false;
3338         }
3339
3340         sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3341                         min_xfer_bytes);
3342         return true;
3343 }
3344
3345 /*
3346  * Determine the device's preferred I/O size for reads and writes
3347  * unless the reported value is unreasonably small, large, not a
3348  * multiple of the physical block size, or simply garbage.
3349  */
3350 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3351                                       unsigned int dev_max)
3352 {
3353         struct scsi_device *sdp = sdkp->device;
3354         unsigned int opt_xfer_bytes =
3355                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3356         unsigned int min_xfer_bytes =
3357                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3358
3359         if (sdkp->opt_xfer_blocks == 0)
3360                 return false;
3361
3362         if (sdkp->opt_xfer_blocks > dev_max) {
3363                 sd_first_printk(KERN_WARNING, sdkp,
3364                                 "Optimal transfer size %u logical blocks " \
3365                                 "> dev_max (%u logical blocks)\n",
3366                                 sdkp->opt_xfer_blocks, dev_max);
3367                 return false;
3368         }
3369
3370         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3371                 sd_first_printk(KERN_WARNING, sdkp,
3372                                 "Optimal transfer size %u logical blocks " \
3373                                 "> sd driver limit (%u logical blocks)\n",
3374                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3375                 return false;
3376         }
3377
3378         if (opt_xfer_bytes < PAGE_SIZE) {
3379                 sd_first_printk(KERN_WARNING, sdkp,
3380                                 "Optimal transfer size %u bytes < " \
3381                                 "PAGE_SIZE (%u bytes)\n",
3382                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3383                 return false;
3384         }
3385
3386         if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3387                 sd_first_printk(KERN_WARNING, sdkp,
3388                                 "Optimal transfer size %u bytes not a " \
3389                                 "multiple of preferred minimum block " \
3390                                 "size (%u bytes)\n",
3391                                 opt_xfer_bytes, min_xfer_bytes);
3392                 return false;
3393         }
3394
3395         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3396                 sd_first_printk(KERN_WARNING, sdkp,
3397                                 "Optimal transfer size %u bytes not a " \
3398                                 "multiple of physical block size (%u bytes)\n",
3399                                 opt_xfer_bytes, sdkp->physical_block_size);
3400                 return false;
3401         }
3402
3403         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3404                         opt_xfer_bytes);
3405         return true;
3406 }
3407
3408 /**
3409  *      sd_revalidate_disk - called the first time a new disk is seen,
3410  *      performs disk spin up, read_capacity, etc.
3411  *      @disk: struct gendisk we care about
3412  **/
3413 static int sd_revalidate_disk(struct gendisk *disk)
3414 {
3415         struct scsi_disk *sdkp = scsi_disk(disk);
3416         struct scsi_device *sdp = sdkp->device;
3417         struct request_queue *q = sdkp->disk->queue;
3418         sector_t old_capacity = sdkp->capacity;
3419         unsigned char *buffer;
3420         unsigned int dev_max, rw_max;
3421
3422         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3423                                       "sd_revalidate_disk\n"));
3424
3425         /*
3426          * If the device is offline, don't try and read capacity or any
3427          * of the other niceties.
3428          */
3429         if (!scsi_device_online(sdp))
3430                 goto out;
3431
3432         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3433         if (!buffer) {
3434                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3435                           "allocation failure.\n");
3436                 goto out;
3437         }
3438
3439         sd_spinup_disk(sdkp);
3440
3441         /*
3442          * Without media there is no reason to ask; moreover, some devices
3443          * react badly if we do.
3444          */
3445         if (sdkp->media_present) {
3446                 sd_read_capacity(sdkp, buffer);
3447
3448                 /*
3449                  * set the default to rotational.  All non-rotational devices
3450                  * support the block characteristics VPD page, which will
3451                  * cause this to be updated correctly and any device which
3452                  * doesn't support it should be treated as rotational.
3453                  */
3454                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3455                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3456
3457                 if (scsi_device_supports_vpd(sdp)) {
3458                         sd_read_block_provisioning(sdkp);
3459                         sd_read_block_limits(sdkp);
3460                         sd_read_block_characteristics(sdkp);
3461                         sd_zbc_read_zones(sdkp, buffer);
3462                         sd_read_cpr(sdkp);
3463                 }
3464
3465                 sd_print_capacity(sdkp, old_capacity);
3466
3467                 sd_read_write_protect_flag(sdkp, buffer);
3468                 sd_read_cache_type(sdkp, buffer);
3469                 sd_read_app_tag_own(sdkp, buffer);
3470                 sd_read_write_same(sdkp, buffer);
3471                 sd_read_security(sdkp, buffer);
3472                 sd_config_protection(sdkp);
3473         }
3474
3475         /*
3476          * We now have all cache related info, determine how we deal
3477          * with flush requests.
3478          */
3479         sd_set_flush_flag(sdkp);
3480
3481         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3482         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3483
3484         /* Some devices report a maximum block count for READ/WRITE requests. */
3485         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3486         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3487
3488         if (sd_validate_min_xfer_size(sdkp))
3489                 blk_queue_io_min(sdkp->disk->queue,
3490                                  logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3491         else
3492                 blk_queue_io_min(sdkp->disk->queue, 0);
3493
3494         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3495                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3496                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3497         } else {
3498                 q->limits.io_opt = 0;
3499                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3500                                       (sector_t)BLK_DEF_MAX_SECTORS);
3501         }
3502
3503         /*
3504          * Limit default to SCSI host optimal sector limit if set. There may be
3505          * an impact on performance for when the size of a request exceeds this
3506          * host limit.
3507          */
3508         rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3509
3510         /* Do not exceed controller limit */
3511         rw_max = min(rw_max, queue_max_hw_sectors(q));
3512
3513         /*
3514          * Only update max_sectors if previously unset or if the current value
3515          * exceeds the capabilities of the hardware.
3516          */
3517         if (sdkp->first_scan ||
3518             q->limits.max_sectors > q->limits.max_dev_sectors ||
3519             q->limits.max_sectors > q->limits.max_hw_sectors)
3520                 q->limits.max_sectors = rw_max;
3521
3522         sdkp->first_scan = 0;
3523
3524         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3525         sd_config_write_same(sdkp);
3526         kfree(buffer);
3527
3528         /*
3529          * For a zoned drive, revalidating the zones can be done only once
3530          * the gendisk capacity is set. So if this fails, set back the gendisk
3531          * capacity to 0.
3532          */
3533         if (sd_zbc_revalidate_zones(sdkp))
3534                 set_capacity_and_notify(disk, 0);
3535
3536  out:
3537         return 0;
3538 }
3539
3540 /**
3541  *      sd_unlock_native_capacity - unlock native capacity
3542  *      @disk: struct gendisk to set capacity for
3543  *
3544  *      Block layer calls this function if it detects that partitions
3545  *      on @disk reach beyond the end of the device.  If the SCSI host
3546  *      implements ->unlock_native_capacity() method, it's invoked to
3547  *      give it a chance to adjust the device capacity.
3548  *
3549  *      CONTEXT:
3550  *      Defined by block layer.  Might sleep.
3551  */
3552 static void sd_unlock_native_capacity(struct gendisk *disk)
3553 {
3554         struct scsi_device *sdev = scsi_disk(disk)->device;
3555
3556         if (sdev->host->hostt->unlock_native_capacity)
3557                 sdev->host->hostt->unlock_native_capacity(sdev);
3558 }
3559
3560 /**
3561  *      sd_format_disk_name - format disk name
3562  *      @prefix: name prefix - ie. "sd" for SCSI disks
3563  *      @index: index of the disk to format name for
3564  *      @buf: output buffer
3565  *      @buflen: length of the output buffer
3566  *
3567  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3568  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3569  *      which is followed by sdaaa.
3570  *
3571  *      This is basically 26 base counting with one extra 'nil' entry
3572  *      at the beginning from the second digit on and can be
3573  *      determined using similar method as 26 base conversion with the
3574  *      index shifted -1 after each digit is computed.
3575  *
3576  *      CONTEXT:
3577  *      Don't care.
3578  *
3579  *      RETURNS:
3580  *      0 on success, -errno on failure.
3581  */
3582 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3583 {
3584         const int base = 'z' - 'a' + 1;
3585         char *begin = buf + strlen(prefix);
3586         char *end = buf + buflen;
3587         char *p;
3588         int unit;
3589
3590         p = end - 1;
3591         *p = '\0';
3592         unit = base;
3593         do {
3594                 if (p == begin)
3595                         return -EINVAL;
3596                 *--p = 'a' + (index % unit);
3597                 index = (index / unit) - 1;
3598         } while (index >= 0);
3599
3600         memmove(begin, p, end - p);
3601         memcpy(buf, prefix, strlen(prefix));
3602
3603         return 0;
3604 }
3605
3606 /**
3607  *      sd_probe - called during driver initialization and whenever a
3608  *      new scsi device is attached to the system. It is called once
3609  *      for each scsi device (not just disks) present.
3610  *      @dev: pointer to device object
3611  *
3612  *      Returns 0 if successful (or not interested in this scsi device
3613  *      (e.g. scanner)); 1 when there is an error.
3614  *
3615  *      Note: this function is invoked from the scsi mid-level.
3616  *      This function sets up the mapping between a given
3617  *      <host,channel,id,lun> (found in sdp) and new device name
3618  *      (e.g. /dev/sda). More precisely it is the block device major
3619  *      and minor number that is chosen here.
3620  *
3621  *      Assume sd_probe is not re-entrant (for time being)
3622  *      Also think about sd_probe() and sd_remove() running coincidentally.
3623  **/
3624 static int sd_probe(struct device *dev)
3625 {
3626         struct scsi_device *sdp = to_scsi_device(dev);
3627         struct scsi_disk *sdkp;
3628         struct gendisk *gd;
3629         int index;
3630         int error;
3631
3632         scsi_autopm_get_device(sdp);
3633         error = -ENODEV;
3634         if (sdp->type != TYPE_DISK &&
3635             sdp->type != TYPE_ZBC &&
3636             sdp->type != TYPE_MOD &&
3637             sdp->type != TYPE_RBC)
3638                 goto out;
3639
3640         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3641                 sdev_printk(KERN_WARNING, sdp,
3642                             "Unsupported ZBC host-managed device.\n");
3643                 goto out;
3644         }
3645
3646         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3647                                         "sd_probe\n"));
3648
3649         error = -ENOMEM;
3650         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3651         if (!sdkp)
3652                 goto out;
3653
3654         gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3655                                          &sd_bio_compl_lkclass);
3656         if (!gd)
3657                 goto out_free;
3658
3659         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3660         if (index < 0) {
3661                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3662                 goto out_put;
3663         }
3664
3665         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3666         if (error) {
3667                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3668                 goto out_free_index;
3669         }
3670
3671         sdkp->device = sdp;
3672         sdkp->disk = gd;
3673         sdkp->index = index;
3674         sdkp->max_retries = SD_MAX_RETRIES;
3675         atomic_set(&sdkp->openers, 0);
3676         atomic_set(&sdkp->device->ioerr_cnt, 0);
3677
3678         if (!sdp->request_queue->rq_timeout) {
3679                 if (sdp->type != TYPE_MOD)
3680                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3681                 else
3682                         blk_queue_rq_timeout(sdp->request_queue,
3683                                              SD_MOD_TIMEOUT);
3684         }
3685
3686         device_initialize(&sdkp->disk_dev);
3687         sdkp->disk_dev.parent = get_device(dev);
3688         sdkp->disk_dev.class = &sd_disk_class;
3689         dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3690
3691         error = device_add(&sdkp->disk_dev);
3692         if (error) {
3693                 put_device(&sdkp->disk_dev);
3694                 goto out;
3695         }
3696
3697         dev_set_drvdata(dev, sdkp);
3698
3699         gd->major = sd_major((index & 0xf0) >> 4);
3700         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3701         gd->minors = SD_MINORS;
3702
3703         gd->fops = &sd_fops;
3704         gd->private_data = sdkp;
3705
3706         /* defaults, until the device tells us otherwise */
3707         sdp->sector_size = 512;
3708         sdkp->capacity = 0;
3709         sdkp->media_present = 1;
3710         sdkp->write_prot = 0;
3711         sdkp->cache_override = 0;
3712         sdkp->WCE = 0;
3713         sdkp->RCD = 0;
3714         sdkp->ATO = 0;
3715         sdkp->first_scan = 1;
3716         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3717
3718         sd_revalidate_disk(gd);
3719
3720         if (sdp->removable) {
3721                 gd->flags |= GENHD_FL_REMOVABLE;
3722                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3723                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3724         }
3725
3726         blk_pm_runtime_init(sdp->request_queue, dev);
3727         if (sdp->rpm_autosuspend) {
3728                 pm_runtime_set_autosuspend_delay(dev,
3729                         sdp->host->hostt->rpm_autosuspend_delay);
3730         }
3731
3732         error = device_add_disk(dev, gd, NULL);
3733         if (error) {
3734                 put_device(&sdkp->disk_dev);
3735                 put_disk(gd);
3736                 goto out;
3737         }
3738
3739         if (sdkp->security) {
3740                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3741                 if (sdkp->opal_dev)
3742                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3743         }
3744
3745         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3746                   sdp->removable ? "removable " : "");
3747         scsi_autopm_put_device(sdp);
3748
3749         return 0;
3750
3751  out_free_index:
3752         ida_free(&sd_index_ida, index);
3753  out_put:
3754         put_disk(gd);
3755  out_free:
3756         kfree(sdkp);
3757  out:
3758         scsi_autopm_put_device(sdp);
3759         return error;
3760 }
3761
3762 /**
3763  *      sd_remove - called whenever a scsi disk (previously recognized by
3764  *      sd_probe) is detached from the system. It is called (potentially
3765  *      multiple times) during sd module unload.
3766  *      @dev: pointer to device object
3767  *
3768  *      Note: this function is invoked from the scsi mid-level.
3769  *      This function potentially frees up a device name (e.g. /dev/sdc)
3770  *      that could be re-used by a subsequent sd_probe().
3771  *      This function is not called when the built-in sd driver is "exit-ed".
3772  **/
3773 static int sd_remove(struct device *dev)
3774 {
3775         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3776
3777         scsi_autopm_get_device(sdkp->device);
3778
3779         device_del(&sdkp->disk_dev);
3780         del_gendisk(sdkp->disk);
3781         if (!sdkp->suspended)
3782                 sd_shutdown(dev);
3783
3784         put_disk(sdkp->disk);
3785         return 0;
3786 }
3787
3788 static void scsi_disk_release(struct device *dev)
3789 {
3790         struct scsi_disk *sdkp = to_scsi_disk(dev);
3791
3792         ida_free(&sd_index_ida, sdkp->index);
3793         sd_zbc_free_zone_info(sdkp);
3794         put_device(&sdkp->device->sdev_gendev);
3795         free_opal_dev(sdkp->opal_dev);
3796
3797         kfree(sdkp);
3798 }
3799
3800 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3801 {
3802         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3803         struct scsi_sense_hdr sshdr;
3804         const struct scsi_exec_args exec_args = {
3805                 .sshdr = &sshdr,
3806                 .req_flags = BLK_MQ_REQ_PM,
3807         };
3808         struct scsi_device *sdp = sdkp->device;
3809         int res;
3810
3811         if (start)
3812                 cmd[4] |= 1;    /* START */
3813
3814         if (sdp->start_stop_pwr_cond)
3815                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3816
3817         if (!scsi_device_online(sdp))
3818                 return -ENODEV;
3819
3820         res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
3821                                sdkp->max_retries, &exec_args);
3822         if (res) {
3823                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3824                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3825                         sd_print_sense_hdr(sdkp, &sshdr);
3826                         /* 0x3a is medium not present */
3827                         if (sshdr.asc == 0x3a)
3828                                 res = 0;
3829                 }
3830         }
3831
3832         /* SCSI error codes must not go to the generic layer */
3833         if (res)
3834                 return -EIO;
3835
3836         return 0;
3837 }
3838
3839 /*
3840  * Send a SYNCHRONIZE CACHE instruction down to the device through
3841  * the normal SCSI command structure.  Wait for the command to
3842  * complete.
3843  */
3844 static void sd_shutdown(struct device *dev)
3845 {
3846         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3847
3848         if (!sdkp)
3849                 return;         /* this can happen */
3850
3851         if (pm_runtime_suspended(dev))
3852                 return;
3853
3854         if (sdkp->WCE && sdkp->media_present) {
3855                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3856                 sd_sync_cache(sdkp, NULL);
3857         }
3858
3859         if ((system_state != SYSTEM_RESTART &&
3860              sdkp->device->manage_system_start_stop) ||
3861             (system_state == SYSTEM_POWER_OFF &&
3862              sdkp->device->manage_shutdown)) {
3863                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3864                 sd_start_stop_device(sdkp, 0);
3865         }
3866 }
3867
3868 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
3869 {
3870         return (sdev->manage_system_start_stop && !runtime) ||
3871                 (sdev->manage_runtime_start_stop && runtime);
3872 }
3873
3874 static int sd_suspend_common(struct device *dev, bool runtime)
3875 {
3876         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3877         struct scsi_sense_hdr sshdr;
3878         int ret = 0;
3879
3880         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3881                 return 0;
3882
3883         if (sdkp->WCE && sdkp->media_present) {
3884                 if (!sdkp->device->silence_suspend)
3885                         sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3886                 ret = sd_sync_cache(sdkp, &sshdr);
3887
3888                 if (ret) {
3889                         /* ignore OFFLINE device */
3890                         if (ret == -ENODEV)
3891                                 return 0;
3892
3893                         if (!scsi_sense_valid(&sshdr) ||
3894                             sshdr.sense_key != ILLEGAL_REQUEST)
3895                                 return ret;
3896
3897                         /*
3898                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3899                          * doesn't support sync. There's not much to do and
3900                          * suspend shouldn't fail.
3901                          */
3902                         ret = 0;
3903                 }
3904         }
3905
3906         if (sd_do_start_stop(sdkp->device, runtime)) {
3907                 if (!sdkp->device->silence_suspend)
3908                         sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3909                 /* an error is not worth aborting a system sleep */
3910                 ret = sd_start_stop_device(sdkp, 0);
3911                 if (!runtime)
3912                         ret = 0;
3913         }
3914
3915         if (!ret)
3916                 sdkp->suspended = true;
3917
3918         return ret;
3919 }
3920
3921 static int sd_suspend_system(struct device *dev)
3922 {
3923         if (pm_runtime_suspended(dev))
3924                 return 0;
3925
3926         return sd_suspend_common(dev, false);
3927 }
3928
3929 static int sd_suspend_runtime(struct device *dev)
3930 {
3931         return sd_suspend_common(dev, true);
3932 }
3933
3934 static int sd_resume(struct device *dev, bool runtime)
3935 {
3936         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3937         int ret;
3938
3939         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3940                 return 0;
3941
3942         if (!sd_do_start_stop(sdkp->device, runtime)) {
3943                 sdkp->suspended = false;
3944                 return 0;
3945         }
3946
3947         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3948         ret = sd_start_stop_device(sdkp, 1);
3949         if (!ret) {
3950                 opal_unlock_from_suspend(sdkp->opal_dev);
3951                 sdkp->suspended = false;
3952         }
3953
3954         return ret;
3955 }
3956
3957 static int sd_resume_system(struct device *dev)
3958 {
3959         if (pm_runtime_suspended(dev))
3960                 return 0;
3961
3962         return sd_resume(dev, false);
3963 }
3964
3965 static int sd_resume_runtime(struct device *dev)
3966 {
3967         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3968         struct scsi_device *sdp;
3969
3970         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3971                 return 0;
3972
3973         sdp = sdkp->device;
3974
3975         if (sdp->ignore_media_change) {
3976                 /* clear the device's sense data */
3977                 static const u8 cmd[10] = { REQUEST_SENSE };
3978                 const struct scsi_exec_args exec_args = {
3979                         .req_flags = BLK_MQ_REQ_PM,
3980                 };
3981
3982                 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
3983                                      sdp->request_queue->rq_timeout, 1,
3984                                      &exec_args))
3985                         sd_printk(KERN_NOTICE, sdkp,
3986                                   "Failed to clear sense data\n");
3987         }
3988
3989         return sd_resume(dev, true);
3990 }
3991
3992 static const struct dev_pm_ops sd_pm_ops = {
3993         .suspend                = sd_suspend_system,
3994         .resume                 = sd_resume_system,
3995         .poweroff               = sd_suspend_system,
3996         .restore                = sd_resume_system,
3997         .runtime_suspend        = sd_suspend_runtime,
3998         .runtime_resume         = sd_resume_runtime,
3999 };
4000
4001 static struct scsi_driver sd_template = {
4002         .gendrv = {
4003                 .name           = "sd",
4004                 .owner          = THIS_MODULE,
4005                 .probe          = sd_probe,
4006                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
4007                 .remove         = sd_remove,
4008                 .shutdown       = sd_shutdown,
4009                 .pm             = &sd_pm_ops,
4010         },
4011         .rescan                 = sd_rescan,
4012         .init_command           = sd_init_command,
4013         .uninit_command         = sd_uninit_command,
4014         .done                   = sd_done,
4015         .eh_action              = sd_eh_action,
4016         .eh_reset               = sd_eh_reset,
4017 };
4018
4019 /**
4020  *      init_sd - entry point for this driver (both when built in or when
4021  *      a module).
4022  *
4023  *      Note: this function registers this driver with the scsi mid-level.
4024  **/
4025 static int __init init_sd(void)
4026 {
4027         int majors = 0, i, err;
4028
4029         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4030
4031         for (i = 0; i < SD_MAJORS; i++) {
4032                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4033                         continue;
4034                 majors++;
4035         }
4036
4037         if (!majors)
4038                 return -ENODEV;
4039
4040         err = class_register(&sd_disk_class);
4041         if (err)
4042                 goto err_out;
4043
4044         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4045         if (!sd_page_pool) {
4046                 printk(KERN_ERR "sd: can't init discard page pool\n");
4047                 err = -ENOMEM;
4048                 goto err_out_class;
4049         }
4050
4051         err = scsi_register_driver(&sd_template.gendrv);
4052         if (err)
4053                 goto err_out_driver;
4054
4055         return 0;
4056
4057 err_out_driver:
4058         mempool_destroy(sd_page_pool);
4059 err_out_class:
4060         class_unregister(&sd_disk_class);
4061 err_out:
4062         for (i = 0; i < SD_MAJORS; i++)
4063                 unregister_blkdev(sd_major(i), "sd");
4064         return err;
4065 }
4066
4067 /**
4068  *      exit_sd - exit point for this driver (when it is a module).
4069  *
4070  *      Note: this function unregisters this driver from the scsi mid-level.
4071  **/
4072 static void __exit exit_sd(void)
4073 {
4074         int i;
4075
4076         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4077
4078         scsi_unregister_driver(&sd_template.gendrv);
4079         mempool_destroy(sd_page_pool);
4080
4081         class_unregister(&sd_disk_class);
4082
4083         for (i = 0; i < SD_MAJORS; i++)
4084                 unregister_blkdev(sd_major(i), "sd");
4085 }
4086
4087 module_init(init_sd);
4088 module_exit(exit_sd);
4089
4090 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4091 {
4092         scsi_print_sense_hdr(sdkp->device,
4093                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4094 }
4095
4096 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4097 {
4098         const char *hb_string = scsi_hostbyte_string(result);
4099
4100         if (hb_string)
4101                 sd_printk(KERN_INFO, sdkp,
4102                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4103                           hb_string ? hb_string : "invalid",
4104                           "DRIVER_OK");
4105         else
4106                 sd_printk(KERN_INFO, sdkp,
4107                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4108                           msg, host_byte(result), "DRIVER_OK");
4109 }