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
  71 #include "sd.h"
  72 #include "scsi_priv.h"
  73 #include "scsi_logging.h"
  74
  75 MODULE_AUTHOR("Eric Youngdale");
  76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
  77 MODULE_LICENSE("GPL");
  78
  79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
  80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
  81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
  82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
  83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
  84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
  85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
  86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
  87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
  88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
  89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
  90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
  91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
  92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
  93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
  94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
  95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
  96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
  97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
  98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
  99
 100 #define SD_MINORS       16
 101
 102 static void sd_config_discard(struct scsi_disk *, unsigned int);
 103 static void sd_config_write_same(struct scsi_disk *);
 104 static int  sd_revalidate_disk(struct gendisk *);
 105 static void sd_unlock_native_capacity(struct gendisk *disk);
 106 static int  sd_probe(struct device *);
 107 static int  sd_remove(struct device *);
 108 static void sd_shutdown(struct device *);
 109 static int sd_suspend_system(struct device *);
 110 static int sd_suspend_runtime(struct device *);
 111 static int sd_resume_system(struct device *);
 112 static int sd_resume_runtime(struct device *);
 113 static void sd_rescan(struct device *);
 114 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
 115 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
 116 static int sd_done(struct scsi_cmnd *);
 117 static void sd_eh_reset(struct scsi_cmnd *);
 118 static int sd_eh_action(struct scsi_cmnd *, int);
 119 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 120 static void scsi_disk_release(struct device *cdev);
 121
 122 static DEFINE_IDA(sd_index_ida);
 123
 124 static mempool_t *sd_page_pool;
 125 static struct lock_class_key sd_bio_compl_lkclass;
 126
 127 static const char *sd_cache_types[] = {
 128         "write through", "none", "write back",
 129         "write back, no read (daft)"
 130 };
 131
 132 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 133 {
 134         bool wc = false, fua = false;
 135
 136         if (sdkp->WCE) {
 137                 wc = true;
 138                 if (sdkp->DPOFUA)
 139                         fua = true;
 140         }
 141
 142         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 143 }
 144
 145 static ssize_t
 146 cache_type_store(struct device *dev, struct device_attribute *attr,
 147                  const char *buf, size_t count)
 148 {
 149         int ct, rcd, wce, sp;
 150         struct scsi_disk *sdkp = to_scsi_disk(dev);
 151         struct scsi_device *sdp = sdkp->device;
 152         char buffer[64];
 153         char *buffer_data;
 154         struct scsi_mode_data data;
 155         struct scsi_sense_hdr sshdr;
 156         static const char temp[] = "temporary ";
 157         int len;
 158
 159         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 160                 /* no cache control on RBC devices; theoretically they
 161                  * can do it, but there's probably so many exceptions
 162                  * it's not worth the risk */
 163                 return -EINVAL;
 164
 165         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 166                 buf += sizeof(temp) - 1;
 167                 sdkp->cache_override = 1;
 168         } else {
 169                 sdkp->cache_override = 0;
 170         }
 171
 172         ct = sysfs_match_string(sd_cache_types, buf);
 173         if (ct < 0)
 174                 return -EINVAL;
 175
 176         rcd = ct & 0x01 ? 1 : 0;
 177         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 178
 179         if (sdkp->cache_override) {
 180                 sdkp->WCE = wce;
 181                 sdkp->RCD = rcd;
 182                 sd_set_flush_flag(sdkp);
 183                 return count;
 184         }
 185
 186         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 187                             sdkp->max_retries, &data, NULL))
 188                 return -EINVAL;
 189         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 190                   data.block_descriptor_length);
 191         buffer_data = buffer + data.header_length +
 192                 data.block_descriptor_length;
 193         buffer_data[2] &= ~0x05;
 194         buffer_data[2] |= wce << 2 | rcd;
 195         sp = buffer_data[0] & 0x80 ? 1 : 0;
 196         buffer_data[0] &= ~0x80;
 197
 198         /*
 199          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 200          * received mode parameter buffer before doing MODE SELECT.
 201          */
 202         data.device_specific = 0;
 203
 204         if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
 205                              sdkp->max_retries, &data, &sshdr)) {
 206                 if (scsi_sense_valid(&sshdr))
 207                         sd_print_sense_hdr(sdkp, &sshdr);
 208                 return -EINVAL;
 209         }
 210         sd_revalidate_disk(sdkp->disk);
 211         return count;
 212 }
 213
 214 static ssize_t
 215 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
 216                        char *buf)
 217 {
 218         struct scsi_disk *sdkp = to_scsi_disk(dev);
 219         struct scsi_device *sdp = sdkp->device;
 220
 221         return sprintf(buf, "%u\n", sdp->manage_start_stop);
 222 }
 223
 224 static ssize_t
 225 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
 226                         const char *buf, size_t count)
 227 {
 228         struct scsi_disk *sdkp = to_scsi_disk(dev);
 229         struct scsi_device *sdp = sdkp->device;
 230         bool v;
 231
 232         if (!capable(CAP_SYS_ADMIN))
 233                 return -EACCES;
 234
 235         if (kstrtobool(buf, &v))
 236                 return -EINVAL;
 237
 238         sdp->manage_start_stop = v;
 239
 240         return count;
 241 }
 242 static DEVICE_ATTR_RW(manage_start_stop);
 243
 244 static ssize_t
 245 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 246 {
 247         struct scsi_disk *sdkp = to_scsi_disk(dev);
 248
 249         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 250 }
 251
 252 static ssize_t
 253 allow_restart_store(struct device *dev, struct device_attribute *attr,
 254                     const char *buf, size_t count)
 255 {
 256         bool v;
 257         struct scsi_disk *sdkp = to_scsi_disk(dev);
 258         struct scsi_device *sdp = sdkp->device;
 259
 260         if (!capable(CAP_SYS_ADMIN))
 261                 return -EACCES;
 262
 263         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 264                 return -EINVAL;
 265
 266         if (kstrtobool(buf, &v))
 267                 return -EINVAL;
 268
 269         sdp->allow_restart = v;
 270
 271         return count;
 272 }
 273 static DEVICE_ATTR_RW(allow_restart);
 274
 275 static ssize_t
 276 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 277 {
 278         struct scsi_disk *sdkp = to_scsi_disk(dev);
 279         int ct = sdkp->RCD + 2*sdkp->WCE;
 280
 281         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 282 }
 283 static DEVICE_ATTR_RW(cache_type);
 284
 285 static ssize_t
 286 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 287 {
 288         struct scsi_disk *sdkp = to_scsi_disk(dev);
 289
 290         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 291 }
 292 static DEVICE_ATTR_RO(FUA);
 293
 294 static ssize_t
 295 protection_type_show(struct device *dev, struct device_attribute *attr,
 296                      char *buf)
 297 {
 298         struct scsi_disk *sdkp = to_scsi_disk(dev);
 299
 300         return sprintf(buf, "%u\n", sdkp->protection_type);
 301 }
 302
 303 static ssize_t
 304 protection_type_store(struct device *dev, struct device_attribute *attr,
 305                       const char *buf, size_t count)
 306 {
 307         struct scsi_disk *sdkp = to_scsi_disk(dev);
 308         unsigned int val;
 309         int err;
 310
 311         if (!capable(CAP_SYS_ADMIN))
 312                 return -EACCES;
 313
 314         err = kstrtouint(buf, 10, &val);
 315
 316         if (err)
 317                 return err;
 318
 319         if (val <= T10_PI_TYPE3_PROTECTION)
 320                 sdkp->protection_type = val;
 321
 322         return count;
 323 }
 324 static DEVICE_ATTR_RW(protection_type);
 325
 326 static ssize_t
 327 protection_mode_show(struct device *dev, struct device_attribute *attr,
 328                      char *buf)
 329 {
 330         struct scsi_disk *sdkp = to_scsi_disk(dev);
 331         struct scsi_device *sdp = sdkp->device;
 332         unsigned int dif, dix;
 333
 334         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 335         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 336
 337         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 338                 dif = 0;
 339                 dix = 1;
 340         }
 341
 342         if (!dif && !dix)
 343                 return sprintf(buf, "none\n");
 344
 345         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 346 }
 347 static DEVICE_ATTR_RO(protection_mode);
 348
 349 static ssize_t
 350 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 351 {
 352         struct scsi_disk *sdkp = to_scsi_disk(dev);
 353
 354         return sprintf(buf, "%u\n", sdkp->ATO);
 355 }
 356 static DEVICE_ATTR_RO(app_tag_own);
 357
 358 static ssize_t
 359 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 360                        char *buf)
 361 {
 362         struct scsi_disk *sdkp = to_scsi_disk(dev);
 363
 364         return sprintf(buf, "%u\n", sdkp->lbpme);
 365 }
 366 static DEVICE_ATTR_RO(thin_provisioning);
 367
 368 /* sysfs_match_string() requires dense arrays */
 369 static const char *lbp_mode[] = {
 370         [SD_LBP_FULL]           = "full",
 371         [SD_LBP_UNMAP]          = "unmap",
 372         [SD_LBP_WS16]           = "writesame_16",
 373         [SD_LBP_WS10]           = "writesame_10",
 374         [SD_LBP_ZERO]           = "writesame_zero",
 375         [SD_LBP_DISABLE]        = "disabled",
 376 };
 377
 378 static ssize_t
 379 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 380                        char *buf)
 381 {
 382         struct scsi_disk *sdkp = to_scsi_disk(dev);
 383
 384         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 385 }
 386
 387 static ssize_t
 388 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 389                         const char *buf, size_t count)
 390 {
 391         struct scsi_disk *sdkp = to_scsi_disk(dev);
 392         struct scsi_device *sdp = sdkp->device;
 393         int mode;
 394
 395         if (!capable(CAP_SYS_ADMIN))
 396                 return -EACCES;
 397
 398         if (sd_is_zoned(sdkp)) {
 399                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 400                 return count;
 401         }
 402
 403         if (sdp->type != TYPE_DISK)
 404                 return -EINVAL;
 405
 406         mode = sysfs_match_string(lbp_mode, buf);
 407         if (mode < 0)
 408                 return -EINVAL;
 409
 410         sd_config_discard(sdkp, mode);
 411
 412         return count;
 413 }
 414 static DEVICE_ATTR_RW(provisioning_mode);
 415
 416 /* sysfs_match_string() requires dense arrays */
 417 static const char *zeroing_mode[] = {
 418         [SD_ZERO_WRITE]         = "write",
 419         [SD_ZERO_WS]            = "writesame",
 420         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 421         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 422 };
 423
 424 static ssize_t
 425 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 426                   char *buf)
 427 {
 428         struct scsi_disk *sdkp = to_scsi_disk(dev);
 429
 430         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 431 }
 432
 433 static ssize_t
 434 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 435                    const char *buf, size_t count)
 436 {
 437         struct scsi_disk *sdkp = to_scsi_disk(dev);
 438         int mode;
 439
 440         if (!capable(CAP_SYS_ADMIN))
 441                 return -EACCES;
 442
 443         mode = sysfs_match_string(zeroing_mode, buf);
 444         if (mode < 0)
 445                 return -EINVAL;
 446
 447         sdkp->zeroing_mode = mode;
 448
 449         return count;
 450 }
 451 static DEVICE_ATTR_RW(zeroing_mode);
 452
 453 static ssize_t
 454 max_medium_access_timeouts_show(struct device *dev,
 455                                 struct device_attribute *attr, char *buf)
 456 {
 457         struct scsi_disk *sdkp = to_scsi_disk(dev);
 458
 459         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 460 }
 461
 462 static ssize_t
 463 max_medium_access_timeouts_store(struct device *dev,
 464                                  struct device_attribute *attr, const char *buf,
 465                                  size_t count)
 466 {
 467         struct scsi_disk *sdkp = to_scsi_disk(dev);
 468         int err;
 469
 470         if (!capable(CAP_SYS_ADMIN))
 471                 return -EACCES;
 472
 473         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 474
 475         return err ? err : count;
 476 }
 477 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 478
 479 static ssize_t
 480 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 481                            char *buf)
 482 {
 483         struct scsi_disk *sdkp = to_scsi_disk(dev);
 484
 485         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 486 }
 487
 488 static ssize_t
 489 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 490                             const char *buf, size_t count)
 491 {
 492         struct scsi_disk *sdkp = to_scsi_disk(dev);
 493         struct scsi_device *sdp = sdkp->device;
 494         unsigned long max;
 495         int err;
 496
 497         if (!capable(CAP_SYS_ADMIN))
 498                 return -EACCES;
 499
 500         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 501                 return -EINVAL;
 502
 503         err = kstrtoul(buf, 10, &max);
 504
 505         if (err)
 506                 return err;
 507
 508         if (max == 0)
 509                 sdp->no_write_same = 1;
 510         else if (max <= SD_MAX_WS16_BLOCKS) {
 511                 sdp->no_write_same = 0;
 512                 sdkp->max_ws_blocks = max;
 513         }
 514
 515         sd_config_write_same(sdkp);
 516
 517         return count;
 518 }
 519 static DEVICE_ATTR_RW(max_write_same_blocks);
 520
 521 static ssize_t
 522 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
 523 {
 524         struct scsi_disk *sdkp = to_scsi_disk(dev);
 525
 526         if (sdkp->device->type == TYPE_ZBC)
 527                 return sprintf(buf, "host-managed\n");
 528         if (sdkp->zoned == 1)
 529                 return sprintf(buf, "host-aware\n");
 530         if (sdkp->zoned == 2)
 531                 return sprintf(buf, "drive-managed\n");
 532         return sprintf(buf, "none\n");
 533 }
 534 static DEVICE_ATTR_RO(zoned_cap);
 535
 536 static ssize_t
 537 max_retries_store(struct device *dev, struct device_attribute *attr,
 538                   const char *buf, size_t count)
 539 {
 540         struct scsi_disk *sdkp = to_scsi_disk(dev);
 541         struct scsi_device *sdev = sdkp->device;
 542         int retries, err;
 543
 544         err = kstrtoint(buf, 10, &retries);
 545         if (err)
 546                 return err;
 547
 548         if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
 549                 sdkp->max_retries = retries;
 550                 return count;
 551         }
 552
 553         sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
 554                     SD_MAX_RETRIES);
 555         return -EINVAL;
 556 }
 557
 558 static ssize_t
 559 max_retries_show(struct device *dev, struct device_attribute *attr,
 560                  char *buf)
 561 {
 562         struct scsi_disk *sdkp = to_scsi_disk(dev);
 563
 564         return sprintf(buf, "%d\n", sdkp->max_retries);
 565 }
 566
 567 static DEVICE_ATTR_RW(max_retries);
 568
 569 static struct attribute *sd_disk_attrs[] = {
 570         &dev_attr_cache_type.attr,
 571         &dev_attr_FUA.attr,
 572         &dev_attr_allow_restart.attr,
 573         &dev_attr_manage_start_stop.attr,
 574         &dev_attr_protection_type.attr,
 575         &dev_attr_protection_mode.attr,
 576         &dev_attr_app_tag_own.attr,
 577         &dev_attr_thin_provisioning.attr,
 578         &dev_attr_provisioning_mode.attr,
 579         &dev_attr_zeroing_mode.attr,
 580         &dev_attr_max_write_same_blocks.attr,
 581         &dev_attr_max_medium_access_timeouts.attr,
 582         &dev_attr_zoned_cap.attr,
 583         &dev_attr_max_retries.attr,
 584         NULL,
 585 };
 586 ATTRIBUTE_GROUPS(sd_disk);
 587
 588 static struct class sd_disk_class = {
 589         .name           = "scsi_disk",
 590         .owner          = THIS_MODULE,
 591         .dev_release    = scsi_disk_release,
 592         .dev_groups     = sd_disk_groups,
 593 };
 594
 595 static const struct dev_pm_ops sd_pm_ops = {
 596         .suspend                = sd_suspend_system,
 597         .resume                 = sd_resume_system,
 598         .poweroff               = sd_suspend_system,
 599         .restore                = sd_resume_system,
 600         .runtime_suspend        = sd_suspend_runtime,
 601         .runtime_resume         = sd_resume_runtime,
 602 };
 603
 604 static struct scsi_driver sd_template = {
 605         .gendrv = {
 606                 .name           = "sd",
 607                 .owner          = THIS_MODULE,
 608                 .probe          = sd_probe,
 609                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
 610                 .remove         = sd_remove,
 611                 .shutdown       = sd_shutdown,
 612                 .pm             = &sd_pm_ops,
 613         },
 614         .rescan                 = sd_rescan,
 615         .init_command           = sd_init_command,
 616         .uninit_command         = sd_uninit_command,
 617         .done                   = sd_done,
 618         .eh_action              = sd_eh_action,
 619         .eh_reset               = sd_eh_reset,
 620 };
 621
 622 /*
 623  * Don't request a new module, as that could deadlock in multipath
 624  * environment.
 625  */
 626 static void sd_default_probe(dev_t devt)
 627 {
 628 }
 629
 630 /*
 631  * Device no to disk mapping:
 632  *
 633  *       major         disc2     disc  p1
 634  *   |............|.............|....|....| <- dev_t
 635  *    31        20 19          8 7  4 3  0
 636  *
 637  * Inside a major, we have 16k disks, however mapped non-
 638  * contiguously. The first 16 disks are for major0, the next
 639  * ones with major1, ... Disk 256 is for major0 again, disk 272
 640  * for major1, ...
 641  * As we stay compatible with our numbering scheme, we can reuse
 642  * the well-know SCSI majors 8, 65--71, 136--143.
 643  */
 644 static int sd_major(int major_idx)
 645 {
 646         switch (major_idx) {
 647         case 0:
 648                 return SCSI_DISK0_MAJOR;
 649         case 1 ... 7:
 650                 return SCSI_DISK1_MAJOR + major_idx - 1;
 651         case 8 ... 15:
 652                 return SCSI_DISK8_MAJOR + major_idx - 8;
 653         default:
 654                 BUG();
 655                 return 0;       /* shut up gcc */
 656         }
 657 }
 658
 659 #ifdef CONFIG_BLK_SED_OPAL
 660 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 661                 size_t len, bool send)
 662 {
 663         struct scsi_disk *sdkp = data;
 664         struct scsi_device *sdev = sdkp->device;
 665         u8 cdb[12] = { 0, };
 666         const struct scsi_exec_args exec_args = {
 667                 .req_flags = BLK_MQ_REQ_PM,
 668         };
 669         int ret;
 670
 671         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 672         cdb[1] = secp;
 673         put_unaligned_be16(spsp, &cdb[2]);
 674         put_unaligned_be32(len, &cdb[6]);
 675
 676         ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
 677                                buffer, len, SD_TIMEOUT, sdkp->max_retries,
 678                                &exec_args);
 679         return ret <= 0 ? ret : -EIO;
 680 }
 681 #endif /* CONFIG_BLK_SED_OPAL */
 682
 683 /*
 684  * Look up the DIX operation based on whether the command is read or
 685  * write and whether dix and dif are enabled.
 686  */
 687 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 688 {
 689         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 690         static const unsigned int ops[] = {     /* wrt dix dif */
 691                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 692                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 693                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 694                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 695                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 696                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 697                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 698                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 699         };
 700
 701         return ops[write << 2 | dix << 1 | dif];
 702 }
 703
 704 /*
 705  * Returns a mask of the protection flags that are valid for a given DIX
 706  * operation.
 707  */
 708 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 709 {
 710         static const unsigned int flag_mask[] = {
 711                 [SCSI_PROT_NORMAL]              = 0,
 712
 713                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 714                                                   SCSI_PROT_GUARD_CHECK |
 715                                                   SCSI_PROT_REF_CHECK |
 716                                                   SCSI_PROT_REF_INCREMENT,
 717
 718                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 719                                                   SCSI_PROT_IP_CHECKSUM,
 720
 721                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 722                                                   SCSI_PROT_GUARD_CHECK |
 723                                                   SCSI_PROT_REF_CHECK |
 724                                                   SCSI_PROT_REF_INCREMENT |
 725                                                   SCSI_PROT_IP_CHECKSUM,
 726
 727                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 728                                                   SCSI_PROT_REF_INCREMENT,
 729
 730                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 731                                                   SCSI_PROT_REF_CHECK |
 732                                                   SCSI_PROT_REF_INCREMENT |
 733                                                   SCSI_PROT_IP_CHECKSUM,
 734
 735                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 736                                                   SCSI_PROT_GUARD_CHECK |
 737                                                   SCSI_PROT_REF_CHECK |
 738                                                   SCSI_PROT_REF_INCREMENT |
 739                                                   SCSI_PROT_IP_CHECKSUM,
 740         };
 741
 742         return flag_mask[prot_op];
 743 }
 744
 745 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 746                                            unsigned int dix, unsigned int dif)
 747 {
 748         struct request *rq = scsi_cmd_to_rq(scmd);
 749         struct bio *bio = rq->bio;
 750         unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
 751         unsigned int protect = 0;
 752
 753         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 754                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 755                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 756
 757                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 758                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 759         }
 760
 761         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 762                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 763
 764                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 765                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 766         }
 767
 768         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 769                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 770
 771                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 772                         protect = 3 << 5;       /* Disable target PI checking */
 773                 else
 774                         protect = 1 << 5;       /* Enable target PI checking */
 775         }
 776
 777         scsi_set_prot_op(scmd, prot_op);
 778         scsi_set_prot_type(scmd, dif);
 779         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 780
 781         return protect;
 782 }
 783
 784 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 785 {
 786         struct request_queue *q = sdkp->disk->queue;
 787         unsigned int logical_block_size = sdkp->device->sector_size;
 788         unsigned int max_blocks = 0;
 789
 790         q->limits.discard_alignment =
 791                 sdkp->unmap_alignment * logical_block_size;
 792         q->limits.discard_granularity =
 793                 max(sdkp->physical_block_size,
 794                     sdkp->unmap_granularity * logical_block_size);
 795         sdkp->provisioning_mode = mode;
 796
 797         switch (mode) {
 798
 799         case SD_LBP_FULL:
 800         case SD_LBP_DISABLE:
 801                 blk_queue_max_discard_sectors(q, 0);
 802                 return;
 803
 804         case SD_LBP_UNMAP:
 805                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 806                                           (u32)SD_MAX_WS16_BLOCKS);
 807                 break;
 808
 809         case SD_LBP_WS16:
 810                 if (sdkp->device->unmap_limit_for_ws)
 811                         max_blocks = sdkp->max_unmap_blocks;
 812                 else
 813                         max_blocks = sdkp->max_ws_blocks;
 814
 815                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 816                 break;
 817
 818         case SD_LBP_WS10:
 819                 if (sdkp->device->unmap_limit_for_ws)
 820                         max_blocks = sdkp->max_unmap_blocks;
 821                 else
 822                         max_blocks = sdkp->max_ws_blocks;
 823
 824                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 825                 break;
 826
 827         case SD_LBP_ZERO:
 828                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 829                                           (u32)SD_MAX_WS10_BLOCKS);
 830                 break;
 831         }
 832
 833         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 834 }
 835
 836 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 837 {
 838         struct scsi_device *sdp = cmd->device;
 839         struct request *rq = scsi_cmd_to_rq(cmd);
 840         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 841         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 842         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 843         unsigned int data_len = 24;
 844         char *buf;
 845
 846         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 847         if (!rq->special_vec.bv_page)
 848                 return BLK_STS_RESOURCE;
 849         clear_highpage(rq->special_vec.bv_page);
 850         rq->special_vec.bv_offset = 0;
 851         rq->special_vec.bv_len = data_len;
 852         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 853
 854         cmd->cmd_len = 10;
 855         cmd->cmnd[0] = UNMAP;
 856         cmd->cmnd[8] = 24;
 857
 858         buf = bvec_virt(&rq->special_vec);
 859         put_unaligned_be16(6 + 16, &buf[0]);
 860         put_unaligned_be16(16, &buf[2]);
 861         put_unaligned_be64(lba, &buf[8]);
 862         put_unaligned_be32(nr_blocks, &buf[16]);
 863
 864         cmd->allowed = sdkp->max_retries;
 865         cmd->transfersize = data_len;
 866         rq->timeout = SD_TIMEOUT;
 867
 868         return scsi_alloc_sgtables(cmd);
 869 }
 870
 871 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 872                 bool unmap)
 873 {
 874         struct scsi_device *sdp = cmd->device;
 875         struct request *rq = scsi_cmd_to_rq(cmd);
 876         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 877         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 878         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 879         u32 data_len = sdp->sector_size;
 880
 881         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 882         if (!rq->special_vec.bv_page)
 883                 return BLK_STS_RESOURCE;
 884         clear_highpage(rq->special_vec.bv_page);
 885         rq->special_vec.bv_offset = 0;
 886         rq->special_vec.bv_len = data_len;
 887         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 888
 889         cmd->cmd_len = 16;
 890         cmd->cmnd[0] = WRITE_SAME_16;
 891         if (unmap)
 892                 cmd->cmnd[1] = 0x8; /* UNMAP */
 893         put_unaligned_be64(lba, &cmd->cmnd[2]);
 894         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 895
 896         cmd->allowed = sdkp->max_retries;
 897         cmd->transfersize = data_len;
 898         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 899
 900         return scsi_alloc_sgtables(cmd);
 901 }
 902
 903 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 904                 bool unmap)
 905 {
 906         struct scsi_device *sdp = cmd->device;
 907         struct request *rq = scsi_cmd_to_rq(cmd);
 908         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 909         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 910         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 911         u32 data_len = sdp->sector_size;
 912
 913         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 914         if (!rq->special_vec.bv_page)
 915                 return BLK_STS_RESOURCE;
 916         clear_highpage(rq->special_vec.bv_page);
 917         rq->special_vec.bv_offset = 0;
 918         rq->special_vec.bv_len = data_len;
 919         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 920
 921         cmd->cmd_len = 10;
 922         cmd->cmnd[0] = WRITE_SAME;
 923         if (unmap)
 924                 cmd->cmnd[1] = 0x8; /* UNMAP */
 925         put_unaligned_be32(lba, &cmd->cmnd[2]);
 926         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 927
 928         cmd->allowed = sdkp->max_retries;
 929         cmd->transfersize = data_len;
 930         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 931
 932         return scsi_alloc_sgtables(cmd);
 933 }
 934
 935 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 936 {
 937         struct request *rq = scsi_cmd_to_rq(cmd);
 938         struct scsi_device *sdp = cmd->device;
 939         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 940         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 941         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 942
 943         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 944                 switch (sdkp->zeroing_mode) {
 945                 case SD_ZERO_WS16_UNMAP:
 946                         return sd_setup_write_same16_cmnd(cmd, true);
 947                 case SD_ZERO_WS10_UNMAP:
 948                         return sd_setup_write_same10_cmnd(cmd, true);
 949                 }
 950         }
 951
 952         if (sdp->no_write_same) {
 953                 rq->rq_flags |= RQF_QUIET;
 954                 return BLK_STS_TARGET;
 955         }
 956
 957         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 958                 return sd_setup_write_same16_cmnd(cmd, false);
 959
 960         return sd_setup_write_same10_cmnd(cmd, false);
 961 }
 962
 963 static void sd_config_write_same(struct scsi_disk *sdkp)
 964 {
 965         struct request_queue *q = sdkp->disk->queue;
 966         unsigned int logical_block_size = sdkp->device->sector_size;
 967
 968         if (sdkp->device->no_write_same) {
 969                 sdkp->max_ws_blocks = 0;
 970                 goto out;
 971         }
 972
 973         /* Some devices can not handle block counts above 0xffff despite
 974          * supporting WRITE SAME(16). Consequently we default to 64k
 975          * blocks per I/O unless the device explicitly advertises a
 976          * bigger limit.
 977          */
 978         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
 979                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 980                                                    (u32)SD_MAX_WS16_BLOCKS);
 981         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
 982                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 983                                                    (u32)SD_MAX_WS10_BLOCKS);
 984         else {
 985                 sdkp->device->no_write_same = 1;
 986                 sdkp->max_ws_blocks = 0;
 987         }
 988
 989         if (sdkp->lbprz && sdkp->lbpws)
 990                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
 991         else if (sdkp->lbprz && sdkp->lbpws10)
 992                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
 993         else if (sdkp->max_ws_blocks)
 994                 sdkp->zeroing_mode = SD_ZERO_WS;
 995         else
 996                 sdkp->zeroing_mode = SD_ZERO_WRITE;
 997
 998         if (sdkp->max_ws_blocks &&
 999             sdkp->physical_block_size > logical_block_size) {
1000                 /*
1001                  * Reporting a maximum number of blocks that is not aligned
1002                  * on the device physical size would cause a large write same
1003                  * request to be split into physically unaligned chunks by
1004                  * __blkdev_issue_write_zeroes() even if the caller of this
1005                  * functions took care to align the large request. So make sure
1006                  * the maximum reported is aligned to the device physical block
1007                  * size. This is only an optional optimization for regular
1008                  * disks, but this is mandatory to avoid failure of large write
1009                  * same requests directed at sequential write required zones of
1010                  * host-managed ZBC disks.
1011                  */
1012                 sdkp->max_ws_blocks =
1013                         round_down(sdkp->max_ws_blocks,
1014                                    bytes_to_logical(sdkp->device,
1015                                                     sdkp->physical_block_size));
1016         }
1017
1018 out:
1019         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1020                                          (logical_block_size >> 9));
1021 }
1022
1023 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1024 {
1025         struct request *rq = scsi_cmd_to_rq(cmd);
1026         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1027
1028         /* flush requests don't perform I/O, zero the S/G table */
1029         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1030
1031         if (cmd->device->use_16_for_sync) {
1032                 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1033                 cmd->cmd_len = 16;
1034         } else {
1035                 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1036                 cmd->cmd_len = 10;
1037         }
1038         cmd->transfersize = 0;
1039         cmd->allowed = sdkp->max_retries;
1040
1041         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1042         return BLK_STS_OK;
1043 }
1044
1045 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1046                                        sector_t lba, unsigned int nr_blocks,
1047                                        unsigned char flags)
1048 {
1049         cmd->cmd_len = SD_EXT_CDB_SIZE;
1050         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1051         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1052         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1053         cmd->cmnd[10] = flags;
1054         put_unaligned_be64(lba, &cmd->cmnd[12]);
1055         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1056         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1057
1058         return BLK_STS_OK;
1059 }
1060
1061 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1062                                        sector_t lba, unsigned int nr_blocks,
1063                                        unsigned char flags)
1064 {
1065         cmd->cmd_len  = 16;
1066         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1067         cmd->cmnd[1]  = flags;
1068         cmd->cmnd[14] = 0;
1069         cmd->cmnd[15] = 0;
1070         put_unaligned_be64(lba, &cmd->cmnd[2]);
1071         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1072
1073         return BLK_STS_OK;
1074 }
1075
1076 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1077                                        sector_t lba, unsigned int nr_blocks,
1078                                        unsigned char flags)
1079 {
1080         cmd->cmd_len = 10;
1081         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1082         cmd->cmnd[1] = flags;
1083         cmd->cmnd[6] = 0;
1084         cmd->cmnd[9] = 0;
1085         put_unaligned_be32(lba, &cmd->cmnd[2]);
1086         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1087
1088         return BLK_STS_OK;
1089 }
1090
1091 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1092                                       sector_t lba, unsigned int nr_blocks,
1093                                       unsigned char flags)
1094 {
1095         /* Avoid that 0 blocks gets translated into 256 blocks. */
1096         if (WARN_ON_ONCE(nr_blocks == 0))
1097                 return BLK_STS_IOERR;
1098
1099         if (unlikely(flags & 0x8)) {
1100                 /*
1101                  * This happens only if this drive failed 10byte rw
1102                  * command with ILLEGAL_REQUEST during operation and
1103                  * thus turned off use_10_for_rw.
1104                  */
1105                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1106                 return BLK_STS_IOERR;
1107         }
1108
1109         cmd->cmd_len = 6;
1110         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1111         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1112         cmd->cmnd[2] = (lba >> 8) & 0xff;
1113         cmd->cmnd[3] = lba & 0xff;
1114         cmd->cmnd[4] = nr_blocks;
1115         cmd->cmnd[5] = 0;
1116
1117         return BLK_STS_OK;
1118 }
1119
1120 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1121 {
1122         struct request *rq = scsi_cmd_to_rq(cmd);
1123         struct scsi_device *sdp = cmd->device;
1124         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1125         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1126         sector_t threshold;
1127         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1128         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1129         bool write = rq_data_dir(rq) == WRITE;
1130         unsigned char protect, fua;
1131         blk_status_t ret;
1132         unsigned int dif;
1133         bool dix;
1134
1135         ret = scsi_alloc_sgtables(cmd);
1136         if (ret != BLK_STS_OK)
1137                 return ret;
1138
1139         ret = BLK_STS_IOERR;
1140         if (!scsi_device_online(sdp) || sdp->changed) {
1141                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1142                 goto fail;
1143         }
1144
1145         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1146                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1147                 goto fail;
1148         }
1149
1150         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1151                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1152                 goto fail;
1153         }
1154
1155         /*
1156          * Some SD card readers can't handle accesses which touch the
1157          * last one or two logical blocks. Split accesses as needed.
1158          */
1159         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1160
1161         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1162                 if (lba < threshold) {
1163                         /* Access up to the threshold but not beyond */
1164                         nr_blocks = threshold - lba;
1165                 } else {
1166                         /* Access only a single logical block */
1167                         nr_blocks = 1;
1168                 }
1169         }
1170
1171         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1172                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1173                 if (ret)
1174                         goto fail;
1175         }
1176
1177         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1178         dix = scsi_prot_sg_count(cmd);
1179         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1180
1181         if (dif || dix)
1182                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1183         else
1184                 protect = 0;
1185
1186         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1187                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1188                                          protect | fua);
1189         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1190                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1191                                          protect | fua);
1192         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1193                    sdp->use_10_for_rw || protect) {
1194                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1195                                          protect | fua);
1196         } else {
1197                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1198                                         protect | fua);
1199         }
1200
1201         if (unlikely(ret != BLK_STS_OK))
1202                 goto fail;
1203
1204         /*
1205          * We shouldn't disconnect in the middle of a sector, so with a dumb
1206          * host adapter, it's safe to assume that we can at least transfer
1207          * this many bytes between each connect / disconnect.
1208          */
1209         cmd->transfersize = sdp->sector_size;
1210         cmd->underflow = nr_blocks << 9;
1211         cmd->allowed = sdkp->max_retries;
1212         cmd->sdb.length = nr_blocks * sdp->sector_size;
1213
1214         SCSI_LOG_HLQUEUE(1,
1215                          scmd_printk(KERN_INFO, cmd,
1216                                      "%s: block=%llu, count=%d\n", __func__,
1217                                      (unsigned long long)blk_rq_pos(rq),
1218                                      blk_rq_sectors(rq)));
1219         SCSI_LOG_HLQUEUE(2,
1220                          scmd_printk(KERN_INFO, cmd,
1221                                      "%s %d/%u 512 byte blocks.\n",
1222                                      write ? "writing" : "reading", nr_blocks,
1223                                      blk_rq_sectors(rq)));
1224
1225         /*
1226          * This indicates that the command is ready from our end to be queued.
1227          */
1228         return BLK_STS_OK;
1229 fail:
1230         scsi_free_sgtables(cmd);
1231         return ret;
1232 }
1233
1234 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1235 {
1236         struct request *rq = scsi_cmd_to_rq(cmd);
1237
1238         switch (req_op(rq)) {
1239         case REQ_OP_DISCARD:
1240                 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1241                 case SD_LBP_UNMAP:
1242                         return sd_setup_unmap_cmnd(cmd);
1243                 case SD_LBP_WS16:
1244                         return sd_setup_write_same16_cmnd(cmd, true);
1245                 case SD_LBP_WS10:
1246                         return sd_setup_write_same10_cmnd(cmd, true);
1247                 case SD_LBP_ZERO:
1248                         return sd_setup_write_same10_cmnd(cmd, false);
1249                 default:
1250                         return BLK_STS_TARGET;
1251                 }
1252         case REQ_OP_WRITE_ZEROES:
1253                 return sd_setup_write_zeroes_cmnd(cmd);
1254         case REQ_OP_FLUSH:
1255                 return sd_setup_flush_cmnd(cmd);
1256         case REQ_OP_READ:
1257         case REQ_OP_WRITE:
1258         case REQ_OP_ZONE_APPEND:
1259                 return sd_setup_read_write_cmnd(cmd);
1260         case REQ_OP_ZONE_RESET:
1261                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1262                                                    false);
1263         case REQ_OP_ZONE_RESET_ALL:
1264                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1265                                                    true);
1266         case REQ_OP_ZONE_OPEN:
1267                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1268         case REQ_OP_ZONE_CLOSE:
1269                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1270         case REQ_OP_ZONE_FINISH:
1271                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1272         default:
1273                 WARN_ON_ONCE(1);
1274                 return BLK_STS_NOTSUPP;
1275         }
1276 }
1277
1278 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1279 {
1280         struct request *rq = scsi_cmd_to_rq(SCpnt);
1281
1282         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1283                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1284 }
1285
1286 static bool sd_need_revalidate(struct block_device *bdev,
1287                 struct scsi_disk *sdkp)
1288 {
1289         if (sdkp->device->removable || sdkp->write_prot) {
1290                 if (bdev_check_media_change(bdev))
1291                         return true;
1292         }
1293
1294         /*
1295          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1296          * nothing to do with partitions, BLKRRPART is used to force a full
1297          * revalidate after things like a format for historical reasons.
1298          */
1299         return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1300 }
1301
1302 /**
1303  *      sd_open - open a scsi disk device
1304  *      @bdev: Block device of the scsi disk to open
1305  *      @mode: FMODE_* mask
1306  *
1307  *      Returns 0 if successful. Returns a negated errno value in case
1308  *      of error.
1309  *
1310  *      Note: This can be called from a user context (e.g. fsck(1) )
1311  *      or from within the kernel (e.g. as a result of a mount(1) ).
1312  *      In the latter case @inode and @filp carry an abridged amount
1313  *      of information as noted above.
1314  *
1315  *      Locking: called with bdev->bd_disk->open_mutex held.
1316  **/
1317 static int sd_open(struct block_device *bdev, fmode_t mode)
1318 {
1319         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1320         struct scsi_device *sdev = sdkp->device;
1321         int retval;
1322
1323         if (scsi_device_get(sdev))
1324                 return -ENXIO;
1325
1326         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1327
1328         /*
1329          * If the device is in error recovery, wait until it is done.
1330          * If the device is offline, then disallow any access to it.
1331          */
1332         retval = -ENXIO;
1333         if (!scsi_block_when_processing_errors(sdev))
1334                 goto error_out;
1335
1336         if (sd_need_revalidate(bdev, sdkp))
1337                 sd_revalidate_disk(bdev->bd_disk);
1338
1339         /*
1340          * If the drive is empty, just let the open fail.
1341          */
1342         retval = -ENOMEDIUM;
1343         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1344                 goto error_out;
1345
1346         /*
1347          * If the device has the write protect tab set, have the open fail
1348          * if the user expects to be able to write to the thing.
1349          */
1350         retval = -EROFS;
1351         if (sdkp->write_prot && (mode & FMODE_WRITE))
1352                 goto error_out;
1353
1354         /*
1355          * It is possible that the disk changing stuff resulted in
1356          * the device being taken offline.  If this is the case,
1357          * report this to the user, and don't pretend that the
1358          * open actually succeeded.
1359          */
1360         retval = -ENXIO;
1361         if (!scsi_device_online(sdev))
1362                 goto error_out;
1363
1364         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1365                 if (scsi_block_when_processing_errors(sdev))
1366                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1367         }
1368
1369         return 0;
1370
1371 error_out:
1372         scsi_device_put(sdev);
1373         return retval;
1374 }
1375
1376 /**
1377  *      sd_release - invoked when the (last) close(2) is called on this
1378  *      scsi disk.
1379  *      @disk: disk to release
1380  *      @mode: FMODE_* mask
1381  *
1382  *      Returns 0.
1383  *
1384  *      Note: may block (uninterruptible) if error recovery is underway
1385  *      on this disk.
1386  *
1387  *      Locking: called with bdev->bd_disk->open_mutex held.
1388  **/
1389 static void sd_release(struct gendisk *disk, fmode_t mode)
1390 {
1391         struct scsi_disk *sdkp = scsi_disk(disk);
1392         struct scsi_device *sdev = sdkp->device;
1393
1394         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1395
1396         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1397                 if (scsi_block_when_processing_errors(sdev))
1398                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1399         }
1400
1401         scsi_device_put(sdev);
1402 }
1403
1404 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1405 {
1406         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1407         struct scsi_device *sdp = sdkp->device;
1408         struct Scsi_Host *host = sdp->host;
1409         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1410         int diskinfo[4];
1411
1412         /* default to most commonly used values */
1413         diskinfo[0] = 0x40;     /* 1 << 6 */
1414         diskinfo[1] = 0x20;     /* 1 << 5 */
1415         diskinfo[2] = capacity >> 11;
1416
1417         /* override with calculated, extended default, or driver values */
1418         if (host->hostt->bios_param)
1419                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1420         else
1421                 scsicam_bios_param(bdev, capacity, diskinfo);
1422
1423         geo->heads = diskinfo[0];
1424         geo->sectors = diskinfo[1];
1425         geo->cylinders = diskinfo[2];
1426         return 0;
1427 }
1428
1429 /**
1430  *      sd_ioctl - process an ioctl
1431  *      @bdev: target block device
1432  *      @mode: FMODE_* mask
1433  *      @cmd: ioctl command number
1434  *      @arg: this is third argument given to ioctl(2) system call.
1435  *      Often contains a pointer.
1436  *
1437  *      Returns 0 if successful (some ioctls return positive numbers on
1438  *      success as well). Returns a negated errno value in case of error.
1439  *
1440  *      Note: most ioctls are forward onto the block subsystem or further
1441  *      down in the scsi subsystem.
1442  **/
1443 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1444                     unsigned int cmd, unsigned long arg)
1445 {
1446         struct gendisk *disk = bdev->bd_disk;
1447         struct scsi_disk *sdkp = scsi_disk(disk);
1448         struct scsi_device *sdp = sdkp->device;
1449         void __user *p = (void __user *)arg;
1450         int error;
1451
1452         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1453                                     "cmd=0x%x\n", disk->disk_name, cmd));
1454
1455         if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1456                 return -ENOIOCTLCMD;
1457
1458         /*
1459          * If we are in the middle of error recovery, don't let anyone
1460          * else try and use this device.  Also, if error recovery fails, it
1461          * may try and take the device offline, in which case all further
1462          * access to the device is prohibited.
1463          */
1464         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1465                         (mode & FMODE_NDELAY) != 0);
1466         if (error)
1467                 return error;
1468
1469         if (is_sed_ioctl(cmd))
1470                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1471         return scsi_ioctl(sdp, mode, cmd, p);
1472 }
1473
1474 static void set_media_not_present(struct scsi_disk *sdkp)
1475 {
1476         if (sdkp->media_present)
1477                 sdkp->device->changed = 1;
1478
1479         if (sdkp->device->removable) {
1480                 sdkp->media_present = 0;
1481                 sdkp->capacity = 0;
1482         }
1483 }
1484
1485 static int media_not_present(struct scsi_disk *sdkp,
1486                              struct scsi_sense_hdr *sshdr)
1487 {
1488         if (!scsi_sense_valid(sshdr))
1489                 return 0;
1490
1491         /* not invoked for commands that could return deferred errors */
1492         switch (sshdr->sense_key) {
1493         case UNIT_ATTENTION:
1494         case NOT_READY:
1495                 /* medium not present */
1496                 if (sshdr->asc == 0x3A) {
1497                         set_media_not_present(sdkp);
1498                         return 1;
1499                 }
1500         }
1501         return 0;
1502 }
1503
1504 /**
1505  *      sd_check_events - check media events
1506  *      @disk: kernel device descriptor
1507  *      @clearing: disk events currently being cleared
1508  *
1509  *      Returns mask of DISK_EVENT_*.
1510  *
1511  *      Note: this function is invoked from the block subsystem.
1512  **/
1513 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1514 {
1515         struct scsi_disk *sdkp = disk->private_data;
1516         struct scsi_device *sdp;
1517         int retval;
1518         bool disk_changed;
1519
1520         if (!sdkp)
1521                 return 0;
1522
1523         sdp = sdkp->device;
1524         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1525
1526         /*
1527          * If the device is offline, don't send any commands - just pretend as
1528          * if the command failed.  If the device ever comes back online, we
1529          * can deal with it then.  It is only because of unrecoverable errors
1530          * that we would ever take a device offline in the first place.
1531          */
1532         if (!scsi_device_online(sdp)) {
1533                 set_media_not_present(sdkp);
1534                 goto out;
1535         }
1536
1537         /*
1538          * Using TEST_UNIT_READY enables differentiation between drive with
1539          * no cartridge loaded - NOT READY, drive with changed cartridge -
1540          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1541          *
1542          * Drives that auto spin down. eg iomega jaz 1G, will be started
1543          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1544          * sd_revalidate() is called.
1545          */
1546         if (scsi_block_when_processing_errors(sdp)) {
1547                 struct scsi_sense_hdr sshdr = { 0, };
1548
1549                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1550                                               &sshdr);
1551
1552                 /* failed to execute TUR, assume media not present */
1553                 if (retval < 0 || host_byte(retval)) {
1554                         set_media_not_present(sdkp);
1555                         goto out;
1556                 }
1557
1558                 if (media_not_present(sdkp, &sshdr))
1559                         goto out;
1560         }
1561
1562         /*
1563          * For removable scsi disk we have to recognise the presence
1564          * of a disk in the drive.
1565          */
1566         if (!sdkp->media_present)
1567                 sdp->changed = 1;
1568         sdkp->media_present = 1;
1569 out:
1570         /*
1571          * sdp->changed is set under the following conditions:
1572          *
1573          *      Medium present state has changed in either direction.
1574          *      Device has indicated UNIT_ATTENTION.
1575          */
1576         disk_changed = sdp->changed;
1577         sdp->changed = 0;
1578         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1579 }
1580
1581 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1582 {
1583         int retries, res;
1584         struct scsi_device *sdp = sdkp->device;
1585         const int timeout = sdp->request_queue->rq_timeout
1586                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1587         struct scsi_sense_hdr my_sshdr;
1588         const struct scsi_exec_args exec_args = {
1589                 .req_flags = BLK_MQ_REQ_PM,
1590                 /* caller might not be interested in sense, but we need it */
1591                 .sshdr = sshdr ? : &my_sshdr,
1592         };
1593
1594         if (!scsi_device_online(sdp))
1595                 return -ENODEV;
1596
1597         sshdr = exec_args.sshdr;
1598
1599         for (retries = 3; retries > 0; --retries) {
1600                 unsigned char cmd[16] = { 0 };
1601
1602                 if (sdp->use_16_for_sync)
1603                         cmd[0] = SYNCHRONIZE_CACHE_16;
1604                 else
1605                         cmd[0] = SYNCHRONIZE_CACHE;
1606                 /*
1607                  * Leave the rest of the command zero to indicate
1608                  * flush everything.
1609                  */
1610                 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
1611                                        timeout, sdkp->max_retries, &exec_args);
1612                 if (res == 0)
1613                         break;
1614         }
1615
1616         if (res) {
1617                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1618
1619                 if (res < 0)
1620                         return res;
1621
1622                 if (scsi_status_is_check_condition(res) &&
1623                     scsi_sense_valid(sshdr)) {
1624                         sd_print_sense_hdr(sdkp, sshdr);
1625
1626                         /* we need to evaluate the error return  */
1627                         if (sshdr->asc == 0x3a ||       /* medium not present */
1628                             sshdr->asc == 0x20 ||       /* invalid command */
1629                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1630                                 /* this is no error here */
1631                                 return 0;
1632                 }
1633
1634                 switch (host_byte(res)) {
1635                 /* ignore errors due to racing a disconnection */
1636                 case DID_BAD_TARGET:
1637                 case DID_NO_CONNECT:
1638                         return 0;
1639                 /* signal the upper layer it might try again */
1640                 case DID_BUS_BUSY:
1641                 case DID_IMM_RETRY:
1642                 case DID_REQUEUE:
1643                 case DID_SOFT_ERROR:
1644                         return -EBUSY;
1645                 default:
1646                         return -EIO;
1647                 }
1648         }
1649         return 0;
1650 }
1651
1652 static void sd_rescan(struct device *dev)
1653 {
1654         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1655
1656         sd_revalidate_disk(sdkp->disk);
1657 }
1658
1659 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1660                 enum blk_unique_id type)
1661 {
1662         struct scsi_device *sdev = scsi_disk(disk)->device;
1663         const struct scsi_vpd *vpd;
1664         const unsigned char *d;
1665         int ret = -ENXIO, len;
1666
1667         rcu_read_lock();
1668         vpd = rcu_dereference(sdev->vpd_pg83);
1669         if (!vpd)
1670                 goto out_unlock;
1671
1672         ret = -EINVAL;
1673         for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1674                 /* we only care about designators with LU association */
1675                 if (((d[1] >> 4) & 0x3) != 0x00)
1676                         continue;
1677                 if ((d[1] & 0xf) != type)
1678                         continue;
1679
1680                 /*
1681                  * Only exit early if a 16-byte descriptor was found.  Otherwise
1682                  * keep looking as one with more entropy might still show up.
1683                  */
1684                 len = d[3];
1685                 if (len != 8 && len != 12 && len != 16)
1686                         continue;
1687                 ret = len;
1688                 memcpy(id, d + 4, len);
1689                 if (len == 16)
1690                         break;
1691         }
1692 out_unlock:
1693         rcu_read_unlock();
1694         return ret;
1695 }
1696
1697 static char sd_pr_type(enum pr_type type)
1698 {
1699         switch (type) {
1700         case PR_WRITE_EXCLUSIVE:
1701                 return 0x01;
1702         case PR_EXCLUSIVE_ACCESS:
1703                 return 0x03;
1704         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1705                 return 0x05;
1706         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1707                 return 0x06;
1708         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1709                 return 0x07;
1710         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1711                 return 0x08;
1712         default:
1713                 return 0;
1714         }
1715 };
1716
1717 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1718 {
1719         switch (host_byte(result)) {
1720         case DID_TRANSPORT_MARGINAL:
1721         case DID_TRANSPORT_DISRUPTED:
1722         case DID_BUS_BUSY:
1723                 return PR_STS_RETRY_PATH_FAILURE;
1724         case DID_NO_CONNECT:
1725                 return PR_STS_PATH_FAILED;
1726         case DID_TRANSPORT_FAILFAST:
1727                 return PR_STS_PATH_FAST_FAILED;
1728         }
1729
1730         switch (status_byte(result)) {
1731         case SAM_STAT_RESERVATION_CONFLICT:
1732                 return PR_STS_RESERVATION_CONFLICT;
1733         case SAM_STAT_CHECK_CONDITION:
1734                 if (!scsi_sense_valid(sshdr))
1735                         return PR_STS_IOERR;
1736
1737                 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1738                     (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1739                         return -EINVAL;
1740
1741                 fallthrough;
1742         default:
1743                 return PR_STS_IOERR;
1744         }
1745 }
1746
1747 static int sd_pr_command(struct block_device *bdev, u8 sa,
1748                 u64 key, u64 sa_key, u8 type, u8 flags)
1749 {
1750         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1751         struct scsi_device *sdev = sdkp->device;
1752         struct scsi_sense_hdr sshdr;
1753         const struct scsi_exec_args exec_args = {
1754                 .sshdr = &sshdr,
1755         };
1756         int result;
1757         u8 cmd[16] = { 0, };
1758         u8 data[24] = { 0, };
1759
1760         cmd[0] = PERSISTENT_RESERVE_OUT;
1761         cmd[1] = sa;
1762         cmd[2] = type;
1763         put_unaligned_be32(sizeof(data), &cmd[5]);
1764
1765         put_unaligned_be64(key, &data[0]);
1766         put_unaligned_be64(sa_key, &data[8]);
1767         data[20] = flags;
1768
1769         result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1770                                   sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1771                                   &exec_args);
1772
1773         if (scsi_status_is_check_condition(result) &&
1774             scsi_sense_valid(&sshdr)) {
1775                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1776                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1777         }
1778
1779         if (result <= 0)
1780                 return result;
1781
1782         return sd_scsi_to_pr_err(&sshdr, result);
1783 }
1784
1785 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1786                 u32 flags)
1787 {
1788         if (flags & ~PR_FL_IGNORE_KEY)
1789                 return -EOPNOTSUPP;
1790         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1791                         old_key, new_key, 0,
1792                         (1 << 0) /* APTPL */);
1793 }
1794
1795 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1796                 u32 flags)
1797 {
1798         if (flags)
1799                 return -EOPNOTSUPP;
1800         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1801 }
1802
1803 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1804 {
1805         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1806 }
1807
1808 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1809                 enum pr_type type, bool abort)
1810 {
1811         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1812                              sd_pr_type(type), 0);
1813 }
1814
1815 static int sd_pr_clear(struct block_device *bdev, u64 key)
1816 {
1817         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1818 }
1819
1820 static const struct pr_ops sd_pr_ops = {
1821         .pr_register    = sd_pr_register,
1822         .pr_reserve     = sd_pr_reserve,
1823         .pr_release     = sd_pr_release,
1824         .pr_preempt     = sd_pr_preempt,
1825         .pr_clear       = sd_pr_clear,
1826 };
1827
1828 static void scsi_disk_free_disk(struct gendisk *disk)
1829 {
1830         struct scsi_disk *sdkp = scsi_disk(disk);
1831
1832         put_device(&sdkp->disk_dev);
1833 }
1834
1835 static const struct block_device_operations sd_fops = {
1836         .owner                  = THIS_MODULE,
1837         .open                   = sd_open,
1838         .release                = sd_release,
1839         .ioctl                  = sd_ioctl,
1840         .getgeo                 = sd_getgeo,
1841         .compat_ioctl           = blkdev_compat_ptr_ioctl,
1842         .check_events           = sd_check_events,
1843         .unlock_native_capacity = sd_unlock_native_capacity,
1844         .report_zones           = sd_zbc_report_zones,
1845         .get_unique_id          = sd_get_unique_id,
1846         .free_disk              = scsi_disk_free_disk,
1847         .pr_ops                 = &sd_pr_ops,
1848 };
1849
1850 /**
1851  *      sd_eh_reset - reset error handling callback
1852  *      @scmd:          sd-issued command that has failed
1853  *
1854  *      This function is called by the SCSI midlayer before starting
1855  *      SCSI EH. When counting medium access failures we have to be
1856  *      careful to register it only only once per device and SCSI EH run;
1857  *      there might be several timed out commands which will cause the
1858  *      'max_medium_access_timeouts' counter to trigger after the first
1859  *      SCSI EH run already and set the device to offline.
1860  *      So this function resets the internal counter before starting SCSI EH.
1861  **/
1862 static void sd_eh_reset(struct scsi_cmnd *scmd)
1863 {
1864         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1865
1866         /* New SCSI EH run, reset gate variable */
1867         sdkp->ignore_medium_access_errors = false;
1868 }
1869
1870 /**
1871  *      sd_eh_action - error handling callback
1872  *      @scmd:          sd-issued command that has failed
1873  *      @eh_disp:       The recovery disposition suggested by the midlayer
1874  *
1875  *      This function is called by the SCSI midlayer upon completion of an
1876  *      error test command (currently TEST UNIT READY). The result of sending
1877  *      the eh command is passed in eh_disp.  We're looking for devices that
1878  *      fail medium access commands but are OK with non access commands like
1879  *      test unit ready (so wrongly see the device as having a successful
1880  *      recovery)
1881  **/
1882 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1883 {
1884         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1885         struct scsi_device *sdev = scmd->device;
1886
1887         if (!scsi_device_online(sdev) ||
1888             !scsi_medium_access_command(scmd) ||
1889             host_byte(scmd->result) != DID_TIME_OUT ||
1890             eh_disp != SUCCESS)
1891                 return eh_disp;
1892
1893         /*
1894          * The device has timed out executing a medium access command.
1895          * However, the TEST UNIT READY command sent during error
1896          * handling completed successfully. Either the device is in the
1897          * process of recovering or has it suffered an internal failure
1898          * that prevents access to the storage medium.
1899          */
1900         if (!sdkp->ignore_medium_access_errors) {
1901                 sdkp->medium_access_timed_out++;
1902                 sdkp->ignore_medium_access_errors = true;
1903         }
1904
1905         /*
1906          * If the device keeps failing read/write commands but TEST UNIT
1907          * READY always completes successfully we assume that medium
1908          * access is no longer possible and take the device offline.
1909          */
1910         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1911                 scmd_printk(KERN_ERR, scmd,
1912                             "Medium access timeout failure. Offlining disk!\n");
1913                 mutex_lock(&sdev->state_mutex);
1914                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1915                 mutex_unlock(&sdev->state_mutex);
1916
1917                 return SUCCESS;
1918         }
1919
1920         return eh_disp;
1921 }
1922
1923 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1924 {
1925         struct request *req = scsi_cmd_to_rq(scmd);
1926         struct scsi_device *sdev = scmd->device;
1927         unsigned int transferred, good_bytes;
1928         u64 start_lba, end_lba, bad_lba;
1929
1930         /*
1931          * Some commands have a payload smaller than the device logical
1932          * block size (e.g. INQUIRY on a 4K disk).
1933          */
1934         if (scsi_bufflen(scmd) <= sdev->sector_size)
1935                 return 0;
1936
1937         /* Check if we have a 'bad_lba' information */
1938         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1939                                      SCSI_SENSE_BUFFERSIZE,
1940                                      &bad_lba))
1941                 return 0;
1942
1943         /*
1944          * If the bad lba was reported incorrectly, we have no idea where
1945          * the error is.
1946          */
1947         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1948         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1949         if (bad_lba < start_lba || bad_lba >= end_lba)
1950                 return 0;
1951
1952         /*
1953          * resid is optional but mostly filled in.  When it's unused,
1954          * its value is zero, so we assume the whole buffer transferred
1955          */
1956         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1957
1958         /* This computation should always be done in terms of the
1959          * resolution of the device's medium.
1960          */
1961         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1962
1963         return min(good_bytes, transferred);
1964 }
1965
1966 /**
1967  *      sd_done - bottom half handler: called when the lower level
1968  *      driver has completed (successfully or otherwise) a scsi command.
1969  *      @SCpnt: mid-level's per command structure.
1970  *
1971  *      Note: potentially run from within an ISR. Must not block.
1972  **/
1973 static int sd_done(struct scsi_cmnd *SCpnt)
1974 {
1975         int result = SCpnt->result;
1976         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1977         unsigned int sector_size = SCpnt->device->sector_size;
1978         unsigned int resid;
1979         struct scsi_sense_hdr sshdr;
1980         struct request *req = scsi_cmd_to_rq(SCpnt);
1981         struct scsi_disk *sdkp = scsi_disk(req->q->disk);
1982         int sense_valid = 0;
1983         int sense_deferred = 0;
1984
1985         switch (req_op(req)) {
1986         case REQ_OP_DISCARD:
1987         case REQ_OP_WRITE_ZEROES:
1988         case REQ_OP_ZONE_RESET:
1989         case REQ_OP_ZONE_RESET_ALL:
1990         case REQ_OP_ZONE_OPEN:
1991         case REQ_OP_ZONE_CLOSE:
1992         case REQ_OP_ZONE_FINISH:
1993                 if (!result) {
1994                         good_bytes = blk_rq_bytes(req);
1995                         scsi_set_resid(SCpnt, 0);
1996                 } else {
1997                         good_bytes = 0;
1998                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1999                 }
2000                 break;
2001         default:
2002                 /*
2003                  * In case of bogus fw or device, we could end up having
2004                  * an unaligned partial completion. Check this here and force
2005                  * alignment.
2006                  */
2007                 resid = scsi_get_resid(SCpnt);
2008                 if (resid & (sector_size - 1)) {
2009                         sd_printk(KERN_INFO, sdkp,
2010                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2011                                 resid, sector_size);
2012                         scsi_print_command(SCpnt);
2013                         resid = min(scsi_bufflen(SCpnt),
2014                                     round_up(resid, sector_size));
2015                         scsi_set_resid(SCpnt, resid);
2016                 }
2017         }
2018
2019         if (result) {
2020                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2021                 if (sense_valid)
2022                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2023         }
2024         sdkp->medium_access_timed_out = 0;
2025
2026         if (!scsi_status_is_check_condition(result) &&
2027             (!sense_valid || sense_deferred))
2028                 goto out;
2029
2030         switch (sshdr.sense_key) {
2031         case HARDWARE_ERROR:
2032         case MEDIUM_ERROR:
2033                 good_bytes = sd_completed_bytes(SCpnt);
2034                 break;
2035         case RECOVERED_ERROR:
2036                 good_bytes = scsi_bufflen(SCpnt);
2037                 break;
2038         case NO_SENSE:
2039                 /* This indicates a false check condition, so ignore it.  An
2040                  * unknown amount of data was transferred so treat it as an
2041                  * error.
2042                  */
2043                 SCpnt->result = 0;
2044                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2045                 break;
2046         case ABORTED_COMMAND:
2047                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2048                         good_bytes = sd_completed_bytes(SCpnt);
2049                 break;
2050         case ILLEGAL_REQUEST:
2051                 switch (sshdr.asc) {
2052                 case 0x10:      /* DIX: Host detected corruption */
2053                         good_bytes = sd_completed_bytes(SCpnt);
2054                         break;
2055                 case 0x20:      /* INVALID COMMAND OPCODE */
2056                 case 0x24:      /* INVALID FIELD IN CDB */
2057                         switch (SCpnt->cmnd[0]) {
2058                         case UNMAP:
2059                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2060                                 break;
2061                         case WRITE_SAME_16:
2062                         case WRITE_SAME:
2063                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2064                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2065                                 } else {
2066                                         sdkp->device->no_write_same = 1;
2067                                         sd_config_write_same(sdkp);
2068                                         req->rq_flags |= RQF_QUIET;
2069                                 }
2070                                 break;
2071                         }
2072                 }
2073                 break;
2074         default:
2075                 break;
2076         }
2077
2078  out:
2079         if (sd_is_zoned(sdkp))
2080                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2081
2082         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2083                                            "sd_done: completed %d of %d bytes\n",
2084                                            good_bytes, scsi_bufflen(SCpnt)));
2085
2086         return good_bytes;
2087 }
2088
2089 /*
2090  * spinup disk - called only in sd_revalidate_disk()
2091  */
2092 static void
2093 sd_spinup_disk(struct scsi_disk *sdkp)
2094 {
2095         unsigned char cmd[10];
2096         unsigned long spintime_expire = 0;
2097         int retries, spintime;
2098         unsigned int the_result;
2099         struct scsi_sense_hdr sshdr;
2100         const struct scsi_exec_args exec_args = {
2101                 .sshdr = &sshdr,
2102         };
2103         int sense_valid = 0;
2104
2105         spintime = 0;
2106
2107         /* Spin up drives, as required.  Only do this at boot time */
2108         /* Spinup needs to be done for module loads too. */
2109         do {
2110                 retries = 0;
2111
2112                 do {
2113                         bool media_was_present = sdkp->media_present;
2114
2115                         cmd[0] = TEST_UNIT_READY;
2116                         memset((void *) &cmd[1], 0, 9);
2117
2118                         the_result = scsi_execute_cmd(sdkp->device, cmd,
2119                                                       REQ_OP_DRV_IN, NULL, 0,
2120                                                       SD_TIMEOUT,
2121                                                       sdkp->max_retries,
2122                                                       &exec_args);
2123
2124                         /*
2125                          * If the drive has indicated to us that it
2126                          * doesn't have any media in it, don't bother
2127                          * with any more polling.
2128                          */
2129                         if (media_not_present(sdkp, &sshdr)) {
2130                                 if (media_was_present)
2131                                         sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2132                                 return;
2133                         }
2134
2135                         if (the_result)
2136                                 sense_valid = scsi_sense_valid(&sshdr);
2137                         retries++;
2138                 } while (retries < 3 &&
2139                          (!scsi_status_is_good(the_result) ||
2140                           (scsi_status_is_check_condition(the_result) &&
2141                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2142
2143                 if (!scsi_status_is_check_condition(the_result)) {
2144                         /* no sense, TUR either succeeded or failed
2145                          * with a status error */
2146                         if(!spintime && !scsi_status_is_good(the_result)) {
2147                                 sd_print_result(sdkp, "Test Unit Ready failed",
2148                                                 the_result);
2149                         }
2150                         break;
2151                 }
2152
2153                 /*
2154                  * The device does not want the automatic start to be issued.
2155                  */
2156                 if (sdkp->device->no_start_on_add)
2157                         break;
2158
2159                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2160                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2161                                 break;  /* manual intervention required */
2162                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2163                                 break;  /* standby */
2164                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2165                                 break;  /* unavailable */
2166                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2167                                 break;  /* sanitize in progress */
2168                         /*
2169                          * Issue command to spin up drive when not ready
2170                          */
2171                         if (!spintime) {
2172                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2173                                 cmd[0] = START_STOP;
2174                                 cmd[1] = 1;     /* Return immediately */
2175                                 memset((void *) &cmd[2], 0, 8);
2176                                 cmd[4] = 1;     /* Start spin cycle */
2177                                 if (sdkp->device->start_stop_pwr_cond)
2178                                         cmd[4] |= 1 << 4;
2179                                 scsi_execute_cmd(sdkp->device, cmd,
2180                                                  REQ_OP_DRV_IN, NULL, 0,
2181                                                  SD_TIMEOUT, sdkp->max_retries,
2182                                                  &exec_args);
2183                                 spintime_expire = jiffies + 100 * HZ;
2184                                 spintime = 1;
2185                         }
2186                         /* Wait 1 second for next try */
2187                         msleep(1000);
2188                         printk(KERN_CONT ".");
2189
2190                 /*
2191                  * Wait for USB flash devices with slow firmware.
2192                  * Yes, this sense key/ASC combination shouldn't
2193                  * occur here.  It's characteristic of these devices.
2194                  */
2195                 } else if (sense_valid &&
2196                                 sshdr.sense_key == UNIT_ATTENTION &&
2197                                 sshdr.asc == 0x28) {
2198                         if (!spintime) {
2199                                 spintime_expire = jiffies + 5 * HZ;
2200                                 spintime = 1;
2201                         }
2202                         /* Wait 1 second for next try */
2203                         msleep(1000);
2204                 } else {
2205                         /* we don't understand the sense code, so it's
2206                          * probably pointless to loop */
2207                         if(!spintime) {
2208                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2209                                 sd_print_sense_hdr(sdkp, &sshdr);
2210                         }
2211                         break;
2212                 }
2213
2214         } while (spintime && time_before_eq(jiffies, spintime_expire));
2215
2216         if (spintime) {
2217                 if (scsi_status_is_good(the_result))
2218                         printk(KERN_CONT "ready\n");
2219                 else
2220                         printk(KERN_CONT "not responding...\n");
2221         }
2222 }
2223
2224 /*
2225  * Determine whether disk supports Data Integrity Field.
2226  */
2227 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2228 {
2229         struct scsi_device *sdp = sdkp->device;
2230         u8 type;
2231
2232         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2233                 sdkp->protection_type = 0;
2234                 return 0;
2235         }
2236
2237         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2238
2239         if (type > T10_PI_TYPE3_PROTECTION) {
2240                 sd_printk(KERN_ERR, sdkp, "formatted with unsupported"  \
2241                           " protection type %u. Disabling disk!\n",
2242                           type);
2243                 sdkp->protection_type = 0;
2244                 return -ENODEV;
2245         }
2246
2247         sdkp->protection_type = type;
2248
2249         return 0;
2250 }
2251
2252 static void sd_config_protection(struct scsi_disk *sdkp)
2253 {
2254         struct scsi_device *sdp = sdkp->device;
2255
2256         sd_dif_config_host(sdkp);
2257
2258         if (!sdkp->protection_type)
2259                 return;
2260
2261         if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2262                 sd_first_printk(KERN_NOTICE, sdkp,
2263                                 "Disabling DIF Type %u protection\n",
2264                                 sdkp->protection_type);
2265                 sdkp->protection_type = 0;
2266         }
2267
2268         sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2269                         sdkp->protection_type);
2270 }
2271
2272 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2273                         struct scsi_sense_hdr *sshdr, int sense_valid,
2274                         int the_result)
2275 {
2276         if (sense_valid)
2277                 sd_print_sense_hdr(sdkp, sshdr);
2278         else
2279                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2280
2281         /*
2282          * Set dirty bit for removable devices if not ready -
2283          * sometimes drives will not report this properly.
2284          */
2285         if (sdp->removable &&
2286             sense_valid && sshdr->sense_key == NOT_READY)
2287                 set_media_not_present(sdkp);
2288
2289         /*
2290          * We used to set media_present to 0 here to indicate no media
2291          * in the drive, but some drives fail read capacity even with
2292          * media present, so we can't do that.
2293          */
2294         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2295 }
2296
2297 #define RC16_LEN 32
2298 #if RC16_LEN > SD_BUF_SIZE
2299 #error RC16_LEN must not be more than SD_BUF_SIZE
2300 #endif
2301
2302 #define READ_CAPACITY_RETRIES_ON_RESET  10
2303
2304 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2305                                                 unsigned char *buffer)
2306 {
2307         unsigned char cmd[16];
2308         struct scsi_sense_hdr sshdr;
2309         const struct scsi_exec_args exec_args = {
2310                 .sshdr = &sshdr,
2311         };
2312         int sense_valid = 0;
2313         int the_result;
2314         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2315         unsigned int alignment;
2316         unsigned long long lba;
2317         unsigned sector_size;
2318
2319         if (sdp->no_read_capacity_16)
2320                 return -EINVAL;
2321
2322         do {
2323                 memset(cmd, 0, 16);
2324                 cmd[0] = SERVICE_ACTION_IN_16;
2325                 cmd[1] = SAI_READ_CAPACITY_16;
2326                 cmd[13] = RC16_LEN;
2327                 memset(buffer, 0, RC16_LEN);
2328
2329                 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2330                                               buffer, RC16_LEN, SD_TIMEOUT,
2331                                               sdkp->max_retries, &exec_args);
2332
2333                 if (media_not_present(sdkp, &sshdr))
2334                         return -ENODEV;
2335
2336                 if (the_result > 0) {
2337                         sense_valid = scsi_sense_valid(&sshdr);
2338                         if (sense_valid &&
2339                             sshdr.sense_key == ILLEGAL_REQUEST &&
2340                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2341                             sshdr.ascq == 0x00)
2342                                 /* Invalid Command Operation Code or
2343                                  * Invalid Field in CDB, just retry
2344                                  * silently with RC10 */
2345                                 return -EINVAL;
2346                         if (sense_valid &&
2347                             sshdr.sense_key == UNIT_ATTENTION &&
2348                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2349                                 /* Device reset might occur several times,
2350                                  * give it one more chance */
2351                                 if (--reset_retries > 0)
2352                                         continue;
2353                 }
2354                 retries--;
2355
2356         } while (the_result && retries);
2357
2358         if (the_result) {
2359                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2360                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2361                 return -EINVAL;
2362         }
2363
2364         sector_size = get_unaligned_be32(&buffer[8]);
2365         lba = get_unaligned_be64(&buffer[0]);
2366
2367         if (sd_read_protection_type(sdkp, buffer) < 0) {
2368                 sdkp->capacity = 0;
2369                 return -ENODEV;
2370         }
2371
2372         /* Logical blocks per physical block exponent */
2373         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2374
2375         /* RC basis */
2376         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2377
2378         /* Lowest aligned logical block */
2379         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2380         blk_queue_alignment_offset(sdp->request_queue, alignment);
2381         if (alignment && sdkp->first_scan)
2382                 sd_printk(KERN_NOTICE, sdkp,
2383                           "physical block alignment offset: %u\n", alignment);
2384
2385         if (buffer[14] & 0x80) { /* LBPME */
2386                 sdkp->lbpme = 1;
2387
2388                 if (buffer[14] & 0x40) /* LBPRZ */
2389                         sdkp->lbprz = 1;
2390
2391                 sd_config_discard(sdkp, SD_LBP_WS16);
2392         }
2393
2394         sdkp->capacity = lba + 1;
2395         return sector_size;
2396 }
2397
2398 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2399                                                 unsigned char *buffer)
2400 {
2401         unsigned char cmd[16];
2402         struct scsi_sense_hdr sshdr;
2403         const struct scsi_exec_args exec_args = {
2404                 .sshdr = &sshdr,
2405         };
2406         int sense_valid = 0;
2407         int the_result;
2408         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2409         sector_t lba;
2410         unsigned sector_size;
2411
2412         do {
2413                 cmd[0] = READ_CAPACITY;
2414                 memset(&cmd[1], 0, 9);
2415                 memset(buffer, 0, 8);
2416
2417                 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2418                                               8, SD_TIMEOUT, sdkp->max_retries,
2419                                               &exec_args);
2420
2421                 if (media_not_present(sdkp, &sshdr))
2422                         return -ENODEV;
2423
2424                 if (the_result > 0) {
2425                         sense_valid = scsi_sense_valid(&sshdr);
2426                         if (sense_valid &&
2427                             sshdr.sense_key == UNIT_ATTENTION &&
2428                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2429                                 /* Device reset might occur several times,
2430                                  * give it one more chance */
2431                                 if (--reset_retries > 0)
2432                                         continue;
2433                 }
2434                 retries--;
2435
2436         } while (the_result && retries);
2437
2438         if (the_result) {
2439                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2440                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2441                 return -EINVAL;
2442         }
2443
2444         sector_size = get_unaligned_be32(&buffer[4]);
2445         lba = get_unaligned_be32(&buffer[0]);
2446
2447         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2448                 /* Some buggy (usb cardreader) devices return an lba of
2449                    0xffffffff when the want to report a size of 0 (with
2450                    which they really mean no media is present) */
2451                 sdkp->capacity = 0;
2452                 sdkp->physical_block_size = sector_size;
2453                 return sector_size;
2454         }
2455
2456         sdkp->capacity = lba + 1;
2457         sdkp->physical_block_size = sector_size;
2458         return sector_size;
2459 }
2460
2461 static int sd_try_rc16_first(struct scsi_device *sdp)
2462 {
2463         if (sdp->host->max_cmd_len < 16)
2464                 return 0;
2465         if (sdp->try_rc_10_first)
2466                 return 0;
2467         if (sdp->scsi_level > SCSI_SPC_2)
2468                 return 1;
2469         if (scsi_device_protection(sdp))
2470                 return 1;
2471         return 0;
2472 }
2473
2474 /*
2475  * read disk capacity
2476  */
2477 static void
2478 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2479 {
2480         int sector_size;
2481         struct scsi_device *sdp = sdkp->device;
2482
2483         if (sd_try_rc16_first(sdp)) {
2484                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2485                 if (sector_size == -EOVERFLOW)
2486                         goto got_data;
2487                 if (sector_size == -ENODEV)
2488                         return;
2489                 if (sector_size < 0)
2490                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2491                 if (sector_size < 0)
2492                         return;
2493         } else {
2494                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2495                 if (sector_size == -EOVERFLOW)
2496                         goto got_data;
2497                 if (sector_size < 0)
2498                         return;
2499                 if ((sizeof(sdkp->capacity) > 4) &&
2500                     (sdkp->capacity > 0xffffffffULL)) {
2501                         int old_sector_size = sector_size;
2502                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2503                                         "Trying to use READ CAPACITY(16).\n");
2504                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2505                         if (sector_size < 0) {
2506                                 sd_printk(KERN_NOTICE, sdkp,
2507                                         "Using 0xffffffff as device size\n");
2508                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2509                                 sector_size = old_sector_size;
2510                                 goto got_data;
2511                         }
2512                         /* Remember that READ CAPACITY(16) succeeded */
2513                         sdp->try_rc_10_first = 0;
2514                 }
2515         }
2516
2517         /* Some devices are known to return the total number of blocks,
2518          * not the highest block number.  Some devices have versions
2519          * which do this and others which do not.  Some devices we might
2520          * suspect of doing this but we don't know for certain.
2521          *
2522          * If we know the reported capacity is wrong, decrement it.  If
2523          * we can only guess, then assume the number of blocks is even
2524          * (usually true but not always) and err on the side of lowering
2525          * the capacity.
2526          */
2527         if (sdp->fix_capacity ||
2528             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2529                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2530                                 "from its reported value: %llu\n",
2531                                 (unsigned long long) sdkp->capacity);
2532                 --sdkp->capacity;
2533         }
2534
2535 got_data:
2536         if (sector_size == 0) {
2537                 sector_size = 512;
2538                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2539                           "assuming 512.\n");
2540         }
2541
2542         if (sector_size != 512 &&
2543             sector_size != 1024 &&
2544             sector_size != 2048 &&
2545             sector_size != 4096) {
2546                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2547                           sector_size);
2548                 /*
2549                  * The user might want to re-format the drive with
2550                  * a supported sectorsize.  Once this happens, it
2551                  * would be relatively trivial to set the thing up.
2552                  * For this reason, we leave the thing in the table.
2553                  */
2554                 sdkp->capacity = 0;
2555                 /*
2556                  * set a bogus sector size so the normal read/write
2557                  * logic in the block layer will eventually refuse any
2558                  * request on this device without tripping over power
2559                  * of two sector size assumptions
2560                  */
2561                 sector_size = 512;
2562         }
2563         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2564         blk_queue_physical_block_size(sdp->request_queue,
2565                                       sdkp->physical_block_size);
2566         sdkp->device->sector_size = sector_size;
2567
2568         if (sdkp->capacity > 0xffffffff)
2569                 sdp->use_16_for_rw = 1;
2570
2571 }
2572
2573 /*
2574  * Print disk capacity
2575  */
2576 static void
2577 sd_print_capacity(struct scsi_disk *sdkp,
2578                   sector_t old_capacity)
2579 {
2580         int sector_size = sdkp->device->sector_size;
2581         char cap_str_2[10], cap_str_10[10];
2582
2583         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2584                 return;
2585
2586         string_get_size(sdkp->capacity, sector_size,
2587                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2588         string_get_size(sdkp->capacity, sector_size,
2589                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2590
2591         sd_printk(KERN_NOTICE, sdkp,
2592                   "%llu %d-byte logical blocks: (%s/%s)\n",
2593                   (unsigned long long)sdkp->capacity,
2594                   sector_size, cap_str_10, cap_str_2);
2595
2596         if (sdkp->physical_block_size != sector_size)
2597                 sd_printk(KERN_NOTICE, sdkp,
2598                           "%u-byte physical blocks\n",
2599                           sdkp->physical_block_size);
2600 }
2601
2602 /* called with buffer of length 512 */
2603 static inline int
2604 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2605                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2606                  struct scsi_sense_hdr *sshdr)
2607 {
2608         /*
2609          * If we must use MODE SENSE(10), make sure that the buffer length
2610          * is at least 8 bytes so that the mode sense header fits.
2611          */
2612         if (sdkp->device->use_10_for_ms && len < 8)
2613                 len = 8;
2614
2615         return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2616                                SD_TIMEOUT, sdkp->max_retries, data,
2617                                sshdr);
2618 }
2619
2620 /*
2621  * read write protect setting, if possible - called only in sd_revalidate_disk()
2622  * called with buffer of length SD_BUF_SIZE
2623  */
2624 static void
2625 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2626 {
2627         int res;
2628         struct scsi_device *sdp = sdkp->device;
2629         struct scsi_mode_data data;
2630         int old_wp = sdkp->write_prot;
2631
2632         set_disk_ro(sdkp->disk, 0);
2633         if (sdp->skip_ms_page_3f) {
2634                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2635                 return;
2636         }
2637
2638         if (sdp->use_192_bytes_for_3f) {
2639                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2640         } else {
2641                 /*
2642                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2643                  * We have to start carefully: some devices hang if we ask
2644                  * for more than is available.
2645                  */
2646                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2647
2648                 /*
2649                  * Second attempt: ask for page 0 When only page 0 is
2650                  * implemented, a request for page 3F may return Sense Key
2651                  * 5: Illegal Request, Sense Code 24: Invalid field in
2652                  * CDB.
2653                  */
2654                 if (res < 0)
2655                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2656
2657                 /*
2658                  * Third attempt: ask 255 bytes, as we did earlier.
2659                  */
2660                 if (res < 0)
2661                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2662                                                &data, NULL);
2663         }
2664
2665         if (res < 0) {
2666                 sd_first_printk(KERN_WARNING, sdkp,
2667                           "Test WP failed, assume Write Enabled\n");
2668         } else {
2669                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2670                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2671                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2672                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2673                                   sdkp->write_prot ? "on" : "off");
2674                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2675                 }
2676         }
2677 }
2678
2679 /*
2680  * sd_read_cache_type - called only from sd_revalidate_disk()
2681  * called with buffer of length SD_BUF_SIZE
2682  */
2683 static void
2684 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2685 {
2686         int len = 0, res;
2687         struct scsi_device *sdp = sdkp->device;
2688
2689         int dbd;
2690         int modepage;
2691         int first_len;
2692         struct scsi_mode_data data;
2693         struct scsi_sense_hdr sshdr;
2694         int old_wce = sdkp->WCE;
2695         int old_rcd = sdkp->RCD;
2696         int old_dpofua = sdkp->DPOFUA;
2697
2698
2699         if (sdkp->cache_override)
2700                 return;
2701
2702         first_len = 4;
2703         if (sdp->skip_ms_page_8) {
2704                 if (sdp->type == TYPE_RBC)
2705                         goto defaults;
2706                 else {
2707                         if (sdp->skip_ms_page_3f)
2708                                 goto defaults;
2709                         modepage = 0x3F;
2710                         if (sdp->use_192_bytes_for_3f)
2711                                 first_len = 192;
2712                         dbd = 0;
2713                 }
2714         } else if (sdp->type == TYPE_RBC) {
2715                 modepage = 6;
2716                 dbd = 8;
2717         } else {
2718                 modepage = 8;
2719                 dbd = 0;
2720         }
2721
2722         /* cautiously ask */
2723         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2724                         &data, &sshdr);
2725
2726         if (res < 0)
2727                 goto bad_sense;
2728
2729         if (!data.header_length) {
2730                 modepage = 6;
2731                 first_len = 0;
2732                 sd_first_printk(KERN_ERR, sdkp,
2733                                 "Missing header in MODE_SENSE response\n");
2734         }
2735
2736         /* that went OK, now ask for the proper length */
2737         len = data.length;
2738
2739         /*
2740          * We're only interested in the first three bytes, actually.
2741          * But the data cache page is defined for the first 20.
2742          */
2743         if (len < 3)
2744                 goto bad_sense;
2745         else if (len > SD_BUF_SIZE) {
2746                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2747                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2748                 len = SD_BUF_SIZE;
2749         }
2750         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2751                 len = 192;
2752
2753         /* Get the data */
2754         if (len > first_len)
2755                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2756                                 &data, &sshdr);
2757
2758         if (!res) {
2759                 int offset = data.header_length + data.block_descriptor_length;
2760
2761                 while (offset < len) {
2762                         u8 page_code = buffer[offset] & 0x3F;
2763                         u8 spf       = buffer[offset] & 0x40;
2764
2765                         if (page_code == 8 || page_code == 6) {
2766                                 /* We're interested only in the first 3 bytes.
2767                                  */
2768                                 if (len - offset <= 2) {
2769                                         sd_first_printk(KERN_ERR, sdkp,
2770                                                 "Incomplete mode parameter "
2771                                                         "data\n");
2772                                         goto defaults;
2773                                 } else {
2774                                         modepage = page_code;
2775                                         goto Page_found;
2776                                 }
2777                         } else {
2778                                 /* Go to the next page */
2779                                 if (spf && len - offset > 3)
2780                                         offset += 4 + (buffer[offset+2] << 8) +
2781                                                 buffer[offset+3];
2782                                 else if (!spf && len - offset > 1)
2783                                         offset += 2 + buffer[offset+1];
2784                                 else {
2785                                         sd_first_printk(KERN_ERR, sdkp,
2786                                                         "Incomplete mode "
2787                                                         "parameter data\n");
2788                                         goto defaults;
2789                                 }
2790                         }
2791                 }
2792
2793                 sd_first_printk(KERN_WARNING, sdkp,
2794                                 "No Caching mode page found\n");
2795                 goto defaults;
2796
2797         Page_found:
2798                 if (modepage == 8) {
2799                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2800                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2801                 } else {
2802                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2803                         sdkp->RCD = 0;
2804                 }
2805
2806                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2807                 if (sdp->broken_fua) {
2808                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2809                         sdkp->DPOFUA = 0;
2810                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2811                            !sdkp->device->use_16_for_rw) {
2812                         sd_first_printk(KERN_NOTICE, sdkp,
2813                                   "Uses READ/WRITE(6), disabling FUA\n");
2814                         sdkp->DPOFUA = 0;
2815                 }
2816
2817                 /* No cache flush allowed for write protected devices */
2818                 if (sdkp->WCE && sdkp->write_prot)
2819                         sdkp->WCE = 0;
2820
2821                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2822                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2823                         sd_printk(KERN_NOTICE, sdkp,
2824                                   "Write cache: %s, read cache: %s, %s\n",
2825                                   sdkp->WCE ? "enabled" : "disabled",
2826                                   sdkp->RCD ? "disabled" : "enabled",
2827                                   sdkp->DPOFUA ? "supports DPO and FUA"
2828                                   : "doesn't support DPO or FUA");
2829
2830                 return;
2831         }
2832
2833 bad_sense:
2834         if (scsi_sense_valid(&sshdr) &&
2835             sshdr.sense_key == ILLEGAL_REQUEST &&
2836             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2837                 /* Invalid field in CDB */
2838                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2839         else
2840                 sd_first_printk(KERN_ERR, sdkp,
2841                                 "Asking for cache data failed\n");
2842
2843 defaults:
2844         if (sdp->wce_default_on) {
2845                 sd_first_printk(KERN_NOTICE, sdkp,
2846                                 "Assuming drive cache: write back\n");
2847                 sdkp->WCE = 1;
2848         } else {
2849                 sd_first_printk(KERN_WARNING, sdkp,
2850                                 "Assuming drive cache: write through\n");
2851                 sdkp->WCE = 0;
2852         }
2853         sdkp->RCD = 0;
2854         sdkp->DPOFUA = 0;
2855 }
2856
2857 /*
2858  * The ATO bit indicates whether the DIF application tag is available
2859  * for use by the operating system.
2860  */
2861 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2862 {
2863         int res, offset;
2864         struct scsi_device *sdp = sdkp->device;
2865         struct scsi_mode_data data;
2866         struct scsi_sense_hdr sshdr;
2867
2868         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2869                 return;
2870
2871         if (sdkp->protection_type == 0)
2872                 return;
2873
2874         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2875                               sdkp->max_retries, &data, &sshdr);
2876
2877         if (res < 0 || !data.header_length ||
2878             data.length < 6) {
2879                 sd_first_printk(KERN_WARNING, sdkp,
2880                           "getting Control mode page failed, assume no ATO\n");
2881
2882                 if (scsi_sense_valid(&sshdr))
2883                         sd_print_sense_hdr(sdkp, &sshdr);
2884
2885                 return;
2886         }
2887
2888         offset = data.header_length + data.block_descriptor_length;
2889
2890         if ((buffer[offset] & 0x3f) != 0x0a) {
2891                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2892                 return;
2893         }
2894
2895         if ((buffer[offset + 5] & 0x80) == 0)
2896                 return;
2897
2898         sdkp->ATO = 1;
2899
2900         return;
2901 }
2902
2903 /**
2904  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2905  * @sdkp: disk to query
2906  */
2907 static void sd_read_block_limits(struct scsi_disk *sdkp)
2908 {
2909         struct scsi_vpd *vpd;
2910
2911         rcu_read_lock();
2912
2913         vpd = rcu_dereference(sdkp->device->vpd_pgb0);
2914         if (!vpd || vpd->len < 16)
2915                 goto out;
2916
2917         sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
2918         sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
2919         sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
2920
2921         if (vpd->len >= 64) {
2922                 unsigned int lba_count, desc_count;
2923
2924                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
2925
2926                 if (!sdkp->lbpme)
2927                         goto out;
2928
2929                 lba_count = get_unaligned_be32(&vpd->data[20]);
2930                 desc_count = get_unaligned_be32(&vpd->data[24]);
2931
2932                 if (lba_count && desc_count)
2933                         sdkp->max_unmap_blocks = lba_count;
2934
2935                 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
2936
2937                 if (vpd->data[32] & 0x80)
2938                         sdkp->unmap_alignment =
2939                                 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
2940
2941                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2942
2943                         if (sdkp->max_unmap_blocks)
2944                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2945                         else
2946                                 sd_config_discard(sdkp, SD_LBP_WS16);
2947
2948                 } else {        /* LBP VPD page tells us what to use */
2949                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2950                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2951                         else if (sdkp->lbpws)
2952                                 sd_config_discard(sdkp, SD_LBP_WS16);
2953                         else if (sdkp->lbpws10)
2954                                 sd_config_discard(sdkp, SD_LBP_WS10);
2955                         else
2956                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2957                 }
2958         }
2959
2960  out:
2961         rcu_read_unlock();
2962 }
2963
2964 /**
2965  * sd_read_block_characteristics - Query block dev. characteristics
2966  * @sdkp: disk to query
2967  */
2968 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2969 {
2970         struct request_queue *q = sdkp->disk->queue;
2971         struct scsi_vpd *vpd;
2972         u16 rot;
2973         u8 zoned;
2974
2975         rcu_read_lock();
2976         vpd = rcu_dereference(sdkp->device->vpd_pgb1);
2977
2978         if (!vpd || vpd->len < 8) {
2979                 rcu_read_unlock();
2980                 return;
2981         }
2982
2983         rot = get_unaligned_be16(&vpd->data[4]);
2984         zoned = (vpd->data[8] >> 4) & 3;
2985         rcu_read_unlock();
2986
2987         if (rot == 1) {
2988                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2989                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2990         }
2991
2992         if (sdkp->device->type == TYPE_ZBC) {
2993                 /* Host-managed */
2994                 disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
2995         } else {
2996                 sdkp->zoned = zoned;
2997                 if (sdkp->zoned == 1) {
2998                         /* Host-aware */
2999                         disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
3000                 } else {
3001                         /* Regular disk or drive managed disk */
3002                         disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3003                 }
3004         }
3005
3006         if (!sdkp->first_scan)
3007                 return;
3008
3009         if (blk_queue_is_zoned(q)) {
3010                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3011                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3012         } else {
3013                 if (sdkp->zoned == 1)
3014                         sd_printk(KERN_NOTICE, sdkp,
3015                                   "Host-aware SMR disk used as regular disk\n");
3016                 else if (sdkp->zoned == 2)
3017                         sd_printk(KERN_NOTICE, sdkp,
3018                                   "Drive-managed SMR disk\n");
3019         }
3020 }
3021
3022 /**
3023  * sd_read_block_provisioning - Query provisioning VPD page
3024  * @sdkp: disk to query
3025  */
3026 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3027 {
3028         struct scsi_vpd *vpd;
3029
3030         if (sdkp->lbpme == 0)
3031                 return;
3032
3033         rcu_read_lock();
3034         vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3035
3036         if (!vpd || vpd->len < 8) {
3037                 rcu_read_unlock();
3038                 return;
3039         }
3040
3041         sdkp->lbpvpd    = 1;
3042         sdkp->lbpu      = (vpd->data[5] >> 7) & 1; /* UNMAP */
3043         sdkp->lbpws     = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3044         sdkp->lbpws10   = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3045         rcu_read_unlock();
3046 }
3047
3048 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3049 {
3050         struct scsi_device *sdev = sdkp->device;
3051
3052         if (sdev->host->no_write_same) {
3053                 sdev->no_write_same = 1;
3054
3055                 return;
3056         }
3057
3058         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3059                 struct scsi_vpd *vpd;
3060
3061                 sdev->no_report_opcodes = 1;
3062
3063                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3064                  * CODES is unsupported and the device has an ATA
3065                  * Information VPD page (SAT).
3066                  */
3067                 rcu_read_lock();
3068                 vpd = rcu_dereference(sdev->vpd_pg89);
3069                 if (vpd)
3070                         sdev->no_write_same = 1;
3071                 rcu_read_unlock();
3072         }
3073
3074         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3075                 sdkp->ws16 = 1;
3076
3077         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3078                 sdkp->ws10 = 1;
3079 }
3080
3081 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3082 {
3083         struct scsi_device *sdev = sdkp->device;
3084
3085         if (!sdev->security_supported)
3086                 return;
3087
3088         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3089                         SECURITY_PROTOCOL_IN) == 1 &&
3090             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3091                         SECURITY_PROTOCOL_OUT) == 1)
3092                 sdkp->security = 1;
3093 }
3094
3095 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3096 {
3097         return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3098 }
3099
3100 /**
3101  * sd_read_cpr - Query concurrent positioning ranges
3102  * @sdkp:       disk to query
3103  */
3104 static void sd_read_cpr(struct scsi_disk *sdkp)
3105 {
3106         struct blk_independent_access_ranges *iars = NULL;
3107         unsigned char *buffer = NULL;
3108         unsigned int nr_cpr = 0;
3109         int i, vpd_len, buf_len = SD_BUF_SIZE;
3110         u8 *desc;
3111
3112         /*
3113          * We need to have the capacity set first for the block layer to be
3114          * able to check the ranges.
3115          */
3116         if (sdkp->first_scan)
3117                 return;
3118
3119         if (!sdkp->capacity)
3120                 goto out;
3121
3122         /*
3123          * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3124          * leading to a maximum page size of 64 + 256*32 bytes.
3125          */
3126         buf_len = 64 + 256*32;
3127         buffer = kmalloc(buf_len, GFP_KERNEL);
3128         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3129                 goto out;
3130
3131         /* We must have at least a 64B header and one 32B range descriptor */
3132         vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3133         if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3134                 sd_printk(KERN_ERR, sdkp,
3135                           "Invalid Concurrent Positioning Ranges VPD page\n");
3136                 goto out;
3137         }
3138
3139         nr_cpr = (vpd_len - 64) / 32;
3140         if (nr_cpr == 1) {
3141                 nr_cpr = 0;
3142                 goto out;
3143         }
3144
3145         iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3146         if (!iars) {
3147                 nr_cpr = 0;
3148                 goto out;
3149         }
3150
3151         desc = &buffer[64];
3152         for (i = 0; i < nr_cpr; i++, desc += 32) {
3153                 if (desc[0] != i) {
3154                         sd_printk(KERN_ERR, sdkp,
3155                                 "Invalid Concurrent Positioning Range number\n");
3156                         nr_cpr = 0;
3157                         break;
3158                 }
3159
3160                 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3161                 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3162         }
3163
3164 out:
3165         disk_set_independent_access_ranges(sdkp->disk, iars);
3166         if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3167                 sd_printk(KERN_NOTICE, sdkp,
3168                           "%u concurrent positioning ranges\n", nr_cpr);
3169                 sdkp->nr_actuators = nr_cpr;
3170         }
3171
3172         kfree(buffer);
3173 }
3174
3175 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3176 {
3177         struct scsi_device *sdp = sdkp->device;
3178         unsigned int min_xfer_bytes =
3179                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3180
3181         if (sdkp->min_xfer_blocks == 0)
3182                 return false;
3183
3184         if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3185                 sd_first_printk(KERN_WARNING, sdkp,
3186                                 "Preferred minimum I/O size %u bytes not a " \
3187                                 "multiple of physical block size (%u bytes)\n",
3188                                 min_xfer_bytes, sdkp->physical_block_size);
3189                 sdkp->min_xfer_blocks = 0;
3190                 return false;
3191         }
3192
3193         sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3194                         min_xfer_bytes);
3195         return true;
3196 }
3197
3198 /*
3199  * Determine the device's preferred I/O size for reads and writes
3200  * unless the reported value is unreasonably small, large, not a
3201  * multiple of the physical block size, or simply garbage.
3202  */
3203 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3204                                       unsigned int dev_max)
3205 {
3206         struct scsi_device *sdp = sdkp->device;
3207         unsigned int opt_xfer_bytes =
3208                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3209         unsigned int min_xfer_bytes =
3210                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3211
3212         if (sdkp->opt_xfer_blocks == 0)
3213                 return false;
3214
3215         if (sdkp->opt_xfer_blocks > dev_max) {
3216                 sd_first_printk(KERN_WARNING, sdkp,
3217                                 "Optimal transfer size %u logical blocks " \
3218                                 "> dev_max (%u logical blocks)\n",
3219                                 sdkp->opt_xfer_blocks, dev_max);
3220                 return false;
3221         }
3222
3223         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3224                 sd_first_printk(KERN_WARNING, sdkp,
3225                                 "Optimal transfer size %u logical blocks " \
3226                                 "> sd driver limit (%u logical blocks)\n",
3227                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3228                 return false;
3229         }
3230
3231         if (opt_xfer_bytes < PAGE_SIZE) {
3232                 sd_first_printk(KERN_WARNING, sdkp,
3233                                 "Optimal transfer size %u bytes < " \
3234                                 "PAGE_SIZE (%u bytes)\n",
3235                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3236                 return false;
3237         }
3238
3239         if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3240                 sd_first_printk(KERN_WARNING, sdkp,
3241                                 "Optimal transfer size %u bytes not a " \
3242                                 "multiple of preferred minimum block " \
3243                                 "size (%u bytes)\n",
3244                                 opt_xfer_bytes, min_xfer_bytes);
3245                 return false;
3246         }
3247
3248         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3249                 sd_first_printk(KERN_WARNING, sdkp,
3250                                 "Optimal transfer size %u bytes not a " \
3251                                 "multiple of physical block size (%u bytes)\n",
3252                                 opt_xfer_bytes, sdkp->physical_block_size);
3253                 return false;
3254         }
3255
3256         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3257                         opt_xfer_bytes);
3258         return true;
3259 }
3260
3261 /**
3262  *      sd_revalidate_disk - called the first time a new disk is seen,
3263  *      performs disk spin up, read_capacity, etc.
3264  *      @disk: struct gendisk we care about
3265  **/
3266 static int sd_revalidate_disk(struct gendisk *disk)
3267 {
3268         struct scsi_disk *sdkp = scsi_disk(disk);
3269         struct scsi_device *sdp = sdkp->device;
3270         struct request_queue *q = sdkp->disk->queue;
3271         sector_t old_capacity = sdkp->capacity;
3272         unsigned char *buffer;
3273         unsigned int dev_max, rw_max;
3274
3275         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3276                                       "sd_revalidate_disk\n"));
3277
3278         /*
3279          * If the device is offline, don't try and read capacity or any
3280          * of the other niceties.
3281          */
3282         if (!scsi_device_online(sdp))
3283                 goto out;
3284
3285         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3286         if (!buffer) {
3287                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3288                           "allocation failure.\n");
3289                 goto out;
3290         }
3291
3292         sd_spinup_disk(sdkp);
3293
3294         /*
3295          * Without media there is no reason to ask; moreover, some devices
3296          * react badly if we do.
3297          */
3298         if (sdkp->media_present) {
3299                 sd_read_capacity(sdkp, buffer);
3300
3301                 /*
3302                  * set the default to rotational.  All non-rotational devices
3303                  * support the block characteristics VPD page, which will
3304                  * cause this to be updated correctly and any device which
3305                  * doesn't support it should be treated as rotational.
3306                  */
3307                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3308                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3309
3310                 if (scsi_device_supports_vpd(sdp)) {
3311                         sd_read_block_provisioning(sdkp);
3312                         sd_read_block_limits(sdkp);
3313                         sd_read_block_characteristics(sdkp);
3314                         sd_zbc_read_zones(sdkp, buffer);
3315                         sd_read_cpr(sdkp);
3316                 }
3317
3318                 sd_print_capacity(sdkp, old_capacity);
3319
3320                 sd_read_write_protect_flag(sdkp, buffer);
3321                 sd_read_cache_type(sdkp, buffer);
3322                 sd_read_app_tag_own(sdkp, buffer);
3323                 sd_read_write_same(sdkp, buffer);
3324                 sd_read_security(sdkp, buffer);
3325                 sd_config_protection(sdkp);
3326         }
3327
3328         /*
3329          * We now have all cache related info, determine how we deal
3330          * with flush requests.
3331          */
3332         sd_set_flush_flag(sdkp);
3333
3334         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3335         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3336
3337         /* Some devices report a maximum block count for READ/WRITE requests. */
3338         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3339         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3340
3341         if (sd_validate_min_xfer_size(sdkp))
3342                 blk_queue_io_min(sdkp->disk->queue,
3343                                  logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3344         else
3345                 blk_queue_io_min(sdkp->disk->queue, 0);
3346
3347         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3348                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3349                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3350         } else {
3351                 q->limits.io_opt = 0;
3352                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3353                                       (sector_t)BLK_DEF_MAX_SECTORS);
3354         }
3355
3356         /*
3357          * Limit default to SCSI host optimal sector limit if set. There may be
3358          * an impact on performance for when the size of a request exceeds this
3359          * host limit.
3360          */
3361         rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3362
3363         /* Do not exceed controller limit */
3364         rw_max = min(rw_max, queue_max_hw_sectors(q));
3365
3366         /*
3367          * Only update max_sectors if previously unset or if the current value
3368          * exceeds the capabilities of the hardware.
3369          */
3370         if (sdkp->first_scan ||
3371             q->limits.max_sectors > q->limits.max_dev_sectors ||
3372             q->limits.max_sectors > q->limits.max_hw_sectors)
3373                 q->limits.max_sectors = rw_max;
3374
3375         sdkp->first_scan = 0;
3376
3377         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3378         sd_config_write_same(sdkp);
3379         kfree(buffer);
3380
3381         /*
3382          * For a zoned drive, revalidating the zones can be done only once
3383          * the gendisk capacity is set. So if this fails, set back the gendisk
3384          * capacity to 0.
3385          */
3386         if (sd_zbc_revalidate_zones(sdkp))
3387                 set_capacity_and_notify(disk, 0);
3388
3389  out:
3390         return 0;
3391 }
3392
3393 /**
3394  *      sd_unlock_native_capacity - unlock native capacity
3395  *      @disk: struct gendisk to set capacity for
3396  *
3397  *      Block layer calls this function if it detects that partitions
3398  *      on @disk reach beyond the end of the device.  If the SCSI host
3399  *      implements ->unlock_native_capacity() method, it's invoked to
3400  *      give it a chance to adjust the device capacity.
3401  *
3402  *      CONTEXT:
3403  *      Defined by block layer.  Might sleep.
3404  */
3405 static void sd_unlock_native_capacity(struct gendisk *disk)
3406 {
3407         struct scsi_device *sdev = scsi_disk(disk)->device;
3408
3409         if (sdev->host->hostt->unlock_native_capacity)
3410                 sdev->host->hostt->unlock_native_capacity(sdev);
3411 }
3412
3413 /**
3414  *      sd_format_disk_name - format disk name
3415  *      @prefix: name prefix - ie. "sd" for SCSI disks
3416  *      @index: index of the disk to format name for
3417  *      @buf: output buffer
3418  *      @buflen: length of the output buffer
3419  *
3420  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3421  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3422  *      which is followed by sdaaa.
3423  *
3424  *      This is basically 26 base counting with one extra 'nil' entry
3425  *      at the beginning from the second digit on and can be
3426  *      determined using similar method as 26 base conversion with the
3427  *      index shifted -1 after each digit is computed.
3428  *
3429  *      CONTEXT:
3430  *      Don't care.
3431  *
3432  *      RETURNS:
3433  *      0 on success, -errno on failure.
3434  */
3435 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3436 {
3437         const int base = 'z' - 'a' + 1;
3438         char *begin = buf + strlen(prefix);
3439         char *end = buf + buflen;
3440         char *p;
3441         int unit;
3442
3443         p = end - 1;
3444         *p = '\0';
3445         unit = base;
3446         do {
3447                 if (p == begin)
3448                         return -EINVAL;
3449                 *--p = 'a' + (index % unit);
3450                 index = (index / unit) - 1;
3451         } while (index >= 0);
3452
3453         memmove(begin, p, end - p);
3454         memcpy(buf, prefix, strlen(prefix));
3455
3456         return 0;
3457 }
3458
3459 /**
3460  *      sd_probe - called during driver initialization and whenever a
3461  *      new scsi device is attached to the system. It is called once
3462  *      for each scsi device (not just disks) present.
3463  *      @dev: pointer to device object
3464  *
3465  *      Returns 0 if successful (or not interested in this scsi device
3466  *      (e.g. scanner)); 1 when there is an error.
3467  *
3468  *      Note: this function is invoked from the scsi mid-level.
3469  *      This function sets up the mapping between a given
3470  *      <host,channel,id,lun> (found in sdp) and new device name
3471  *      (e.g. /dev/sda). More precisely it is the block device major
3472  *      and minor number that is chosen here.
3473  *
3474  *      Assume sd_probe is not re-entrant (for time being)
3475  *      Also think about sd_probe() and sd_remove() running coincidentally.
3476  **/
3477 static int sd_probe(struct device *dev)
3478 {
3479         struct scsi_device *sdp = to_scsi_device(dev);
3480         struct scsi_disk *sdkp;
3481         struct gendisk *gd;
3482         int index;
3483         int error;
3484
3485         scsi_autopm_get_device(sdp);
3486         error = -ENODEV;
3487         if (sdp->type != TYPE_DISK &&
3488             sdp->type != TYPE_ZBC &&
3489             sdp->type != TYPE_MOD &&
3490             sdp->type != TYPE_RBC)
3491                 goto out;
3492
3493         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3494                 sdev_printk(KERN_WARNING, sdp,
3495                             "Unsupported ZBC host-managed device.\n");
3496                 goto out;
3497         }
3498
3499         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3500                                         "sd_probe\n"));
3501
3502         error = -ENOMEM;
3503         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3504         if (!sdkp)
3505                 goto out;
3506
3507         gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3508                                          &sd_bio_compl_lkclass);
3509         if (!gd)
3510                 goto out_free;
3511
3512         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3513         if (index < 0) {
3514                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3515                 goto out_put;
3516         }
3517
3518         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3519         if (error) {
3520                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3521                 goto out_free_index;
3522         }
3523
3524         sdkp->device = sdp;
3525         sdkp->disk = gd;
3526         sdkp->index = index;
3527         sdkp->max_retries = SD_MAX_RETRIES;
3528         atomic_set(&sdkp->openers, 0);
3529         atomic_set(&sdkp->device->ioerr_cnt, 0);
3530
3531         if (!sdp->request_queue->rq_timeout) {
3532                 if (sdp->type != TYPE_MOD)
3533                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3534                 else
3535                         blk_queue_rq_timeout(sdp->request_queue,
3536                                              SD_MOD_TIMEOUT);
3537         }
3538
3539         device_initialize(&sdkp->disk_dev);
3540         sdkp->disk_dev.parent = get_device(dev);
3541         sdkp->disk_dev.class = &sd_disk_class;
3542         dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3543
3544         error = device_add(&sdkp->disk_dev);
3545         if (error) {
3546                 put_device(&sdkp->disk_dev);
3547                 goto out;
3548         }
3549
3550         dev_set_drvdata(dev, sdkp);
3551
3552         gd->major = sd_major((index & 0xf0) >> 4);
3553         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3554         gd->minors = SD_MINORS;
3555
3556         gd->fops = &sd_fops;
3557         gd->private_data = sdkp;
3558
3559         /* defaults, until the device tells us otherwise */
3560         sdp->sector_size = 512;
3561         sdkp->capacity = 0;
3562         sdkp->media_present = 1;
3563         sdkp->write_prot = 0;
3564         sdkp->cache_override = 0;
3565         sdkp->WCE = 0;
3566         sdkp->RCD = 0;
3567         sdkp->ATO = 0;
3568         sdkp->first_scan = 1;
3569         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3570
3571         sd_revalidate_disk(gd);
3572
3573         if (sdp->removable) {
3574                 gd->flags |= GENHD_FL_REMOVABLE;
3575                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3576                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3577         }
3578
3579         blk_pm_runtime_init(sdp->request_queue, dev);
3580         if (sdp->rpm_autosuspend) {
3581                 pm_runtime_set_autosuspend_delay(dev,
3582                         sdp->host->hostt->rpm_autosuspend_delay);
3583         }
3584
3585         error = device_add_disk(dev, gd, NULL);
3586         if (error) {
3587                 put_device(&sdkp->disk_dev);
3588                 put_disk(gd);
3589                 goto out;
3590         }
3591
3592         if (sdkp->security) {
3593                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3594                 if (sdkp->opal_dev)
3595                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3596         }
3597
3598         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3599                   sdp->removable ? "removable " : "");
3600         scsi_autopm_put_device(sdp);
3601
3602         return 0;
3603
3604  out_free_index:
3605         ida_free(&sd_index_ida, index);
3606  out_put:
3607         put_disk(gd);
3608  out_free:
3609         kfree(sdkp);
3610  out:
3611         scsi_autopm_put_device(sdp);
3612         return error;
3613 }
3614
3615 /**
3616  *      sd_remove - called whenever a scsi disk (previously recognized by
3617  *      sd_probe) is detached from the system. It is called (potentially
3618  *      multiple times) during sd module unload.
3619  *      @dev: pointer to device object
3620  *
3621  *      Note: this function is invoked from the scsi mid-level.
3622  *      This function potentially frees up a device name (e.g. /dev/sdc)
3623  *      that could be re-used by a subsequent sd_probe().
3624  *      This function is not called when the built-in sd driver is "exit-ed".
3625  **/
3626 static int sd_remove(struct device *dev)
3627 {
3628         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3629
3630         scsi_autopm_get_device(sdkp->device);
3631
3632         device_del(&sdkp->disk_dev);
3633         del_gendisk(sdkp->disk);
3634         sd_shutdown(dev);
3635
3636         put_disk(sdkp->disk);
3637         return 0;
3638 }
3639
3640 static void scsi_disk_release(struct device *dev)
3641 {
3642         struct scsi_disk *sdkp = to_scsi_disk(dev);
3643
3644         ida_free(&sd_index_ida, sdkp->index);
3645         sd_zbc_free_zone_info(sdkp);
3646         put_device(&sdkp->device->sdev_gendev);
3647         free_opal_dev(sdkp->opal_dev);
3648
3649         kfree(sdkp);
3650 }
3651
3652 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3653 {
3654         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3655         struct scsi_sense_hdr sshdr;
3656         const struct scsi_exec_args exec_args = {
3657                 .sshdr = &sshdr,
3658                 .req_flags = BLK_MQ_REQ_PM,
3659         };
3660         struct scsi_device *sdp = sdkp->device;
3661         int res;
3662
3663         if (start)
3664                 cmd[4] |= 1;    /* START */
3665
3666         if (sdp->start_stop_pwr_cond)
3667                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3668
3669         if (!scsi_device_online(sdp))
3670                 return -ENODEV;
3671
3672         res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
3673                                sdkp->max_retries, &exec_args);
3674         if (res) {
3675                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3676                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3677                         sd_print_sense_hdr(sdkp, &sshdr);
3678                         /* 0x3a is medium not present */
3679                         if (sshdr.asc == 0x3a)
3680                                 res = 0;
3681                 }
3682         }
3683
3684         /* SCSI error codes must not go to the generic layer */
3685         if (res)
3686                 return -EIO;
3687
3688         return 0;
3689 }
3690
3691 /*
3692  * Send a SYNCHRONIZE CACHE instruction down to the device through
3693  * the normal SCSI command structure.  Wait for the command to
3694  * complete.
3695  */
3696 static void sd_shutdown(struct device *dev)
3697 {
3698         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3699
3700         if (!sdkp)
3701                 return;         /* this can happen */
3702
3703         if (pm_runtime_suspended(dev))
3704                 return;
3705
3706         if (sdkp->WCE && sdkp->media_present) {
3707                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3708                 sd_sync_cache(sdkp, NULL);
3709         }
3710
3711         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3712                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3713                 sd_start_stop_device(sdkp, 0);
3714         }
3715 }
3716
3717 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3718 {
3719         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3720         struct scsi_sense_hdr sshdr;
3721         int ret = 0;
3722
3723         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3724                 return 0;
3725
3726         if (sdkp->WCE && sdkp->media_present) {
3727                 if (!sdkp->device->silence_suspend)
3728                         sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3729                 ret = sd_sync_cache(sdkp, &sshdr);
3730
3731                 if (ret) {
3732                         /* ignore OFFLINE device */
3733                         if (ret == -ENODEV)
3734                                 return 0;
3735
3736                         if (!scsi_sense_valid(&sshdr) ||
3737                             sshdr.sense_key != ILLEGAL_REQUEST)
3738                                 return ret;
3739
3740                         /*
3741                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3742                          * doesn't support sync. There's not much to do and
3743                          * suspend shouldn't fail.
3744                          */
3745                         ret = 0;
3746                 }
3747         }
3748
3749         if (sdkp->device->manage_start_stop) {
3750                 if (!sdkp->device->silence_suspend)
3751                         sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3752                 /* an error is not worth aborting a system sleep */
3753                 ret = sd_start_stop_device(sdkp, 0);
3754                 if (ignore_stop_errors)
3755                         ret = 0;
3756         }
3757
3758         return ret;
3759 }
3760
3761 static int sd_suspend_system(struct device *dev)
3762 {
3763         if (pm_runtime_suspended(dev))
3764                 return 0;
3765
3766         return sd_suspend_common(dev, true);
3767 }
3768
3769 static int sd_suspend_runtime(struct device *dev)
3770 {
3771         return sd_suspend_common(dev, false);
3772 }
3773
3774 static int sd_resume(struct device *dev)
3775 {
3776         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3777         int ret;
3778
3779         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3780                 return 0;
3781
3782         if (!sdkp->device->manage_start_stop)
3783                 return 0;
3784
3785         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3786         ret = sd_start_stop_device(sdkp, 1);
3787         if (!ret)
3788                 opal_unlock_from_suspend(sdkp->opal_dev);
3789         return ret;
3790 }
3791
3792 static int sd_resume_system(struct device *dev)
3793 {
3794         if (pm_runtime_suspended(dev))
3795                 return 0;
3796
3797         return sd_resume(dev);
3798 }
3799
3800 static int sd_resume_runtime(struct device *dev)
3801 {
3802         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3803         struct scsi_device *sdp;
3804
3805         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3806                 return 0;
3807
3808         sdp = sdkp->device;
3809
3810         if (sdp->ignore_media_change) {
3811                 /* clear the device's sense data */
3812                 static const u8 cmd[10] = { REQUEST_SENSE };
3813                 const struct scsi_exec_args exec_args = {
3814                         .req_flags = BLK_MQ_REQ_PM,
3815                 };
3816
3817                 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
3818                                      sdp->request_queue->rq_timeout, 1,
3819                                      &exec_args))
3820                         sd_printk(KERN_NOTICE, sdkp,
3821                                   "Failed to clear sense data\n");
3822         }
3823
3824         return sd_resume(dev);
3825 }
3826
3827 /**
3828  *      init_sd - entry point for this driver (both when built in or when
3829  *      a module).
3830  *
3831  *      Note: this function registers this driver with the scsi mid-level.
3832  **/
3833 static int __init init_sd(void)
3834 {
3835         int majors = 0, i, err;
3836
3837         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3838
3839         for (i = 0; i < SD_MAJORS; i++) {
3840                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3841                         continue;
3842                 majors++;
3843         }
3844
3845         if (!majors)
3846                 return -ENODEV;
3847
3848         err = class_register(&sd_disk_class);
3849         if (err)
3850                 goto err_out;
3851
3852         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3853         if (!sd_page_pool) {
3854                 printk(KERN_ERR "sd: can't init discard page pool\n");
3855                 err = -ENOMEM;
3856                 goto err_out_class;
3857         }
3858
3859         err = scsi_register_driver(&sd_template.gendrv);
3860         if (err)
3861                 goto err_out_driver;
3862
3863         return 0;
3864
3865 err_out_driver:
3866         mempool_destroy(sd_page_pool);
3867 err_out_class:
3868         class_unregister(&sd_disk_class);
3869 err_out:
3870         for (i = 0; i < SD_MAJORS; i++)
3871                 unregister_blkdev(sd_major(i), "sd");
3872         return err;
3873 }
3874
3875 /**
3876  *      exit_sd - exit point for this driver (when it is a module).
3877  *
3878  *      Note: this function unregisters this driver from the scsi mid-level.
3879  **/
3880 static void __exit exit_sd(void)
3881 {
3882         int i;
3883
3884         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3885
3886         scsi_unregister_driver(&sd_template.gendrv);
3887         mempool_destroy(sd_page_pool);
3888
3889         class_unregister(&sd_disk_class);
3890
3891         for (i = 0; i < SD_MAJORS; i++)
3892                 unregister_blkdev(sd_major(i), "sd");
3893 }
3894
3895 module_init(init_sd);
3896 module_exit(exit_sd);
3897
3898 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3899 {
3900         scsi_print_sense_hdr(sdkp->device,
3901                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3902 }
3903
3904 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3905 {
3906         const char *hb_string = scsi_hostbyte_string(result);
3907
3908         if (hb_string)
3909                 sd_printk(KERN_INFO, sdkp,
3910                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3911                           hb_string ? hb_string : "invalid",
3912                           "DRIVER_OK");
3913         else
3914                 sd_printk(KERN_INFO, sdkp,
3915                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3916                           msg, host_byte(result), "DRIVER_OK");
3917 }