drivers/scsi/sd.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *      sd.c Copyright (C) 1992 Drew Eckhardt
   4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
   5  *
   6  *      Linux scsi disk driver
   7  *              Initial versions: Drew Eckhardt
   8  *              Subsequent revisions: Eric Youngdale
   9  *      Modification history:
  10  *       - Drew Eckhardt <drew@colorado.edu> original
  11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
  12  *         outstanding request, and other enhancements.
  13  *         Support loadable low-level scsi drivers.
  14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
  15  *         eight major numbers.
  16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
  17  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
  18  *         sd_init and cleanups.
  19  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
  20  *         not being read in sd_open. Fix problem where removable media
  21  *         could be ejected after sd_open.
  22  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
  23  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
  24  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
  25  *         Support 32k/1M disks.
  26  *
  27  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
  28  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
  29  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
  30  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
  31  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
  32  *      Note: when the logging level is set by the user, it must be greater
  33  *      than the level indicated above to trigger output.
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/fs.h>
  38 #include <linux/kernel.h>
  39 #include <linux/mm.h>
  40 #include <linux/bio.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/errno.h>
  43 #include <linux/idr.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/init.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/blkpg.h>
  48 #include <linux/blk-pm.h>
  49 #include <linux/delay.h>
  50 #include <linux/major.h>
  51 #include <linux/mutex.h>
  52 #include <linux/string_helpers.h>
  53 #include <linux/slab.h>
  54 #include <linux/sed-opal.h>
  55 #include <linux/pm_runtime.h>
  56 #include <linux/pr.h>
  57 #include <linux/t10-pi.h>
  58 #include <linux/uaccess.h>
  59 #include <asm/unaligned.h>
  60
  61 #include <scsi/scsi.h>
  62 #include <scsi/scsi_cmnd.h>
  63 #include <scsi/scsi_dbg.h>
  64 #include <scsi/scsi_device.h>
  65 #include <scsi/scsi_driver.h>
  66 #include <scsi/scsi_eh.h>
  67 #include <scsi/scsi_host.h>
  68 #include <scsi/scsi_ioctl.h>
  69 #include <scsi/scsicam.h>
  70
  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 struct kmem_cache *sd_cdb_cache;
 125 static mempool_t *sd_page_pool;
 126 static struct lock_class_key sd_bio_compl_lkclass;
 127
 128 static const char *sd_cache_types[] = {
 129         "write through", "none", "write back",
 130         "write back, no read (daft)"
 131 };
 132
 133 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 134 {
 135         bool wc = false, fua = false;
 136
 137         if (sdkp->WCE) {
 138                 wc = true;
 139                 if (sdkp->DPOFUA)
 140                         fua = true;
 141         }
 142
 143         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 144 }
 145
 146 static ssize_t
 147 cache_type_store(struct device *dev, struct device_attribute *attr,
 148                  const char *buf, size_t count)
 149 {
 150         int ct, rcd, wce, sp;
 151         struct scsi_disk *sdkp = to_scsi_disk(dev);
 152         struct scsi_device *sdp = sdkp->device;
 153         char buffer[64];
 154         char *buffer_data;
 155         struct scsi_mode_data data;
 156         struct scsi_sense_hdr sshdr;
 157         static const char temp[] = "temporary ";
 158         int len;
 159
 160         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 161                 /* no cache control on RBC devices; theoretically they
 162                  * can do it, but there's probably so many exceptions
 163                  * it's not worth the risk */
 164                 return -EINVAL;
 165
 166         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 167                 buf += sizeof(temp) - 1;
 168                 sdkp->cache_override = 1;
 169         } else {
 170                 sdkp->cache_override = 0;
 171         }
 172
 173         ct = sysfs_match_string(sd_cache_types, buf);
 174         if (ct < 0)
 175                 return -EINVAL;
 176
 177         rcd = ct & 0x01 ? 1 : 0;
 178         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 179
 180         if (sdkp->cache_override) {
 181                 sdkp->WCE = wce;
 182                 sdkp->RCD = rcd;
 183                 sd_set_flush_flag(sdkp);
 184                 return count;
 185         }
 186
 187         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 188                             sdkp->max_retries, &data, NULL))
 189                 return -EINVAL;
 190         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 191                   data.block_descriptor_length);
 192         buffer_data = buffer + data.header_length +
 193                 data.block_descriptor_length;
 194         buffer_data[2] &= ~0x05;
 195         buffer_data[2] |= wce << 2 | rcd;
 196         sp = buffer_data[0] & 0x80 ? 1 : 0;
 197         buffer_data[0] &= ~0x80;
 198
 199         /*
 200          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 201          * received mode parameter buffer before doing MODE SELECT.
 202          */
 203         data.device_specific = 0;
 204
 205         if (scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
 206                              sdkp->max_retries, &data, &sshdr)) {
 207                 if (scsi_sense_valid(&sshdr))
 208                         sd_print_sense_hdr(sdkp, &sshdr);
 209                 return -EINVAL;
 210         }
 211         sd_revalidate_disk(sdkp->disk);
 212         return count;
 213 }
 214
 215 static ssize_t
 216 manage_start_stop_show(struct device *dev,
 217                        struct device_attribute *attr, char *buf)
 218 {
 219         struct scsi_disk *sdkp = to_scsi_disk(dev);
 220         struct scsi_device *sdp = sdkp->device;
 221
 222         return sysfs_emit(buf, "%u\n",
 223                           sdp->manage_system_start_stop &&
 224                           sdp->manage_runtime_start_stop);
 225 }
 226 static DEVICE_ATTR_RO(manage_start_stop);
 227
 228 static ssize_t
 229 manage_system_start_stop_show(struct device *dev,
 230                               struct device_attribute *attr, char *buf)
 231 {
 232         struct scsi_disk *sdkp = to_scsi_disk(dev);
 233         struct scsi_device *sdp = sdkp->device;
 234
 235         return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
 236 }
 237
 238 static ssize_t
 239 manage_system_start_stop_store(struct device *dev,
 240                                struct device_attribute *attr,
 241                                const char *buf, size_t count)
 242 {
 243         struct scsi_disk *sdkp = to_scsi_disk(dev);
 244         struct scsi_device *sdp = sdkp->device;
 245         bool v;
 246
 247         if (!capable(CAP_SYS_ADMIN))
 248                 return -EACCES;
 249
 250         if (kstrtobool(buf, &v))
 251                 return -EINVAL;
 252
 253         sdp->manage_system_start_stop = v;
 254
 255         return count;
 256 }
 257 static DEVICE_ATTR_RW(manage_system_start_stop);
 258
 259 static ssize_t
 260 manage_runtime_start_stop_show(struct device *dev,
 261                                struct device_attribute *attr, char *buf)
 262 {
 263         struct scsi_disk *sdkp = to_scsi_disk(dev);
 264         struct scsi_device *sdp = sdkp->device;
 265
 266         return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
 267 }
 268
 269 static ssize_t
 270 manage_runtime_start_stop_store(struct device *dev,
 271                                 struct device_attribute *attr,
 272                                 const char *buf, size_t count)
 273 {
 274         struct scsi_disk *sdkp = to_scsi_disk(dev);
 275         struct scsi_device *sdp = sdkp->device;
 276         bool v;
 277
 278         if (!capable(CAP_SYS_ADMIN))
 279                 return -EACCES;
 280
 281         if (kstrtobool(buf, &v))
 282                 return -EINVAL;
 283
 284         sdp->manage_runtime_start_stop = v;
 285
 286         return count;
 287 }
 288 static DEVICE_ATTR_RW(manage_runtime_start_stop);
 289
 290 static ssize_t
 291 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 292 {
 293         struct scsi_disk *sdkp = to_scsi_disk(dev);
 294
 295         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 296 }
 297
 298 static ssize_t
 299 allow_restart_store(struct device *dev, struct device_attribute *attr,
 300                     const char *buf, size_t count)
 301 {
 302         bool v;
 303         struct scsi_disk *sdkp = to_scsi_disk(dev);
 304         struct scsi_device *sdp = sdkp->device;
 305
 306         if (!capable(CAP_SYS_ADMIN))
 307                 return -EACCES;
 308
 309         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 310                 return -EINVAL;
 311
 312         if (kstrtobool(buf, &v))
 313                 return -EINVAL;
 314
 315         sdp->allow_restart = v;
 316
 317         return count;
 318 }
 319 static DEVICE_ATTR_RW(allow_restart);
 320
 321 static ssize_t
 322 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 323 {
 324         struct scsi_disk *sdkp = to_scsi_disk(dev);
 325         int ct = sdkp->RCD + 2*sdkp->WCE;
 326
 327         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 328 }
 329 static DEVICE_ATTR_RW(cache_type);
 330
 331 static ssize_t
 332 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 333 {
 334         struct scsi_disk *sdkp = to_scsi_disk(dev);
 335
 336         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 337 }
 338 static DEVICE_ATTR_RO(FUA);
 339
 340 static ssize_t
 341 protection_type_show(struct device *dev, struct device_attribute *attr,
 342                      char *buf)
 343 {
 344         struct scsi_disk *sdkp = to_scsi_disk(dev);
 345
 346         return sprintf(buf, "%u\n", sdkp->protection_type);
 347 }
 348
 349 static ssize_t
 350 protection_type_store(struct device *dev, struct device_attribute *attr,
 351                       const char *buf, size_t count)
 352 {
 353         struct scsi_disk *sdkp = to_scsi_disk(dev);
 354         unsigned int val;
 355         int err;
 356
 357         if (!capable(CAP_SYS_ADMIN))
 358                 return -EACCES;
 359
 360         err = kstrtouint(buf, 10, &val);
 361
 362         if (err)
 363                 return err;
 364
 365         if (val <= T10_PI_TYPE3_PROTECTION)
 366                 sdkp->protection_type = val;
 367
 368         return count;
 369 }
 370 static DEVICE_ATTR_RW(protection_type);
 371
 372 static ssize_t
 373 protection_mode_show(struct device *dev, struct device_attribute *attr,
 374                      char *buf)
 375 {
 376         struct scsi_disk *sdkp = to_scsi_disk(dev);
 377         struct scsi_device *sdp = sdkp->device;
 378         unsigned int dif, dix;
 379
 380         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 381         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 382
 383         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 384                 dif = 0;
 385                 dix = 1;
 386         }
 387
 388         if (!dif && !dix)
 389                 return sprintf(buf, "none\n");
 390
 391         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 392 }
 393 static DEVICE_ATTR_RO(protection_mode);
 394
 395 static ssize_t
 396 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 397 {
 398         struct scsi_disk *sdkp = to_scsi_disk(dev);
 399
 400         return sprintf(buf, "%u\n", sdkp->ATO);
 401 }
 402 static DEVICE_ATTR_RO(app_tag_own);
 403
 404 static ssize_t
 405 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 406                        char *buf)
 407 {
 408         struct scsi_disk *sdkp = to_scsi_disk(dev);
 409
 410         return sprintf(buf, "%u\n", sdkp->lbpme);
 411 }
 412 static DEVICE_ATTR_RO(thin_provisioning);
 413
 414 /* sysfs_match_string() requires dense arrays */
 415 static const char *lbp_mode[] = {
 416         [SD_LBP_FULL]           = "full",
 417         [SD_LBP_UNMAP]          = "unmap",
 418         [SD_LBP_WS16]           = "writesame_16",
 419         [SD_LBP_WS10]           = "writesame_10",
 420         [SD_LBP_ZERO]           = "writesame_zero",
 421         [SD_LBP_DISABLE]        = "disabled",
 422 };
 423
 424 static ssize_t
 425 provisioning_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", lbp_mode[sdkp->provisioning_mode]);
 431 }
 432
 433 static ssize_t
 434 provisioning_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         struct scsi_device *sdp = sdkp->device;
 439         int mode;
 440
 441         if (!capable(CAP_SYS_ADMIN))
 442                 return -EACCES;
 443
 444         if (sd_is_zoned(sdkp)) {
 445                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 446                 return count;
 447         }
 448
 449         if (sdp->type != TYPE_DISK)
 450                 return -EINVAL;
 451
 452         mode = sysfs_match_string(lbp_mode, buf);
 453         if (mode < 0)
 454                 return -EINVAL;
 455
 456         sd_config_discard(sdkp, mode);
 457
 458         return count;
 459 }
 460 static DEVICE_ATTR_RW(provisioning_mode);
 461
 462 /* sysfs_match_string() requires dense arrays */
 463 static const char *zeroing_mode[] = {
 464         [SD_ZERO_WRITE]         = "write",
 465         [SD_ZERO_WS]            = "writesame",
 466         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 467         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 468 };
 469
 470 static ssize_t
 471 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 472                   char *buf)
 473 {
 474         struct scsi_disk *sdkp = to_scsi_disk(dev);
 475
 476         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 477 }
 478
 479 static ssize_t
 480 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 481                    const char *buf, size_t count)
 482 {
 483         struct scsi_disk *sdkp = to_scsi_disk(dev);
 484         int mode;
 485
 486         if (!capable(CAP_SYS_ADMIN))
 487                 return -EACCES;
 488
 489         mode = sysfs_match_string(zeroing_mode, buf);
 490         if (mode < 0)
 491                 return -EINVAL;
 492
 493         sdkp->zeroing_mode = mode;
 494
 495         return count;
 496 }
 497 static DEVICE_ATTR_RW(zeroing_mode);
 498
 499 static ssize_t
 500 max_medium_access_timeouts_show(struct device *dev,
 501                                 struct device_attribute *attr, char *buf)
 502 {
 503         struct scsi_disk *sdkp = to_scsi_disk(dev);
 504
 505         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 506 }
 507
 508 static ssize_t
 509 max_medium_access_timeouts_store(struct device *dev,
 510                                  struct device_attribute *attr, const char *buf,
 511                                  size_t count)
 512 {
 513         struct scsi_disk *sdkp = to_scsi_disk(dev);
 514         int err;
 515
 516         if (!capable(CAP_SYS_ADMIN))
 517                 return -EACCES;
 518
 519         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 520
 521         return err ? err : count;
 522 }
 523 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 524
 525 static ssize_t
 526 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 527                            char *buf)
 528 {
 529         struct scsi_disk *sdkp = to_scsi_disk(dev);
 530
 531         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 532 }
 533
 534 static ssize_t
 535 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 536                             const char *buf, size_t count)
 537 {
 538         struct scsi_disk *sdkp = to_scsi_disk(dev);
 539         struct scsi_device *sdp = sdkp->device;
 540         unsigned long max;
 541         int err;
 542
 543         if (!capable(CAP_SYS_ADMIN))
 544                 return -EACCES;
 545
 546         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 547                 return -EINVAL;
 548
 549         err = kstrtoul(buf, 10, &max);
 550
 551         if (err)
 552                 return err;
 553
 554         if (max == 0)
 555                 sdp->no_write_same = 1;
 556         else if (max <= SD_MAX_WS16_BLOCKS) {
 557                 sdp->no_write_same = 0;
 558                 sdkp->max_ws_blocks = max;
 559         }
 560
 561         sd_config_write_same(sdkp);
 562
 563         return count;
 564 }
 565 static DEVICE_ATTR_RW(max_write_same_blocks);
 566
 567 static ssize_t
 568 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
 569 {
 570         struct scsi_disk *sdkp = to_scsi_disk(dev);
 571
 572         if (sdkp->device->type == TYPE_ZBC)
 573                 return sprintf(buf, "host-managed\n");
 574         if (sdkp->zoned == 1)
 575                 return sprintf(buf, "host-aware\n");
 576         if (sdkp->zoned == 2)
 577                 return sprintf(buf, "drive-managed\n");
 578         return sprintf(buf, "none\n");
 579 }
 580 static DEVICE_ATTR_RO(zoned_cap);
 581
 582 static ssize_t
 583 max_retries_store(struct device *dev, struct device_attribute *attr,
 584                   const char *buf, size_t count)
 585 {
 586         struct scsi_disk *sdkp = to_scsi_disk(dev);
 587         struct scsi_device *sdev = sdkp->device;
 588         int retries, err;
 589
 590         err = kstrtoint(buf, 10, &retries);
 591         if (err)
 592                 return err;
 593
 594         if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
 595                 sdkp->max_retries = retries;
 596                 return count;
 597         }
 598
 599         sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
 600                     SD_MAX_RETRIES);
 601         return -EINVAL;
 602 }
 603
 604 static ssize_t
 605 max_retries_show(struct device *dev, struct device_attribute *attr,
 606                  char *buf)
 607 {
 608         struct scsi_disk *sdkp = to_scsi_disk(dev);
 609
 610         return sprintf(buf, "%d\n", sdkp->max_retries);
 611 }
 612
 613 static DEVICE_ATTR_RW(max_retries);
 614
 615 static struct attribute *sd_disk_attrs[] = {
 616         &dev_attr_cache_type.attr,
 617         &dev_attr_FUA.attr,
 618         &dev_attr_allow_restart.attr,
 619         &dev_attr_manage_start_stop.attr,
 620         &dev_attr_manage_system_start_stop.attr,
 621         &dev_attr_manage_runtime_start_stop.attr,
 622         &dev_attr_protection_type.attr,
 623         &dev_attr_protection_mode.attr,
 624         &dev_attr_app_tag_own.attr,
 625         &dev_attr_thin_provisioning.attr,
 626         &dev_attr_provisioning_mode.attr,
 627         &dev_attr_zeroing_mode.attr,
 628         &dev_attr_max_write_same_blocks.attr,
 629         &dev_attr_max_medium_access_timeouts.attr,
 630         &dev_attr_zoned_cap.attr,
 631         &dev_attr_max_retries.attr,
 632         NULL,
 633 };
 634 ATTRIBUTE_GROUPS(sd_disk);
 635
 636 static struct class sd_disk_class = {
 637         .name           = "scsi_disk",
 638         .owner          = THIS_MODULE,
 639         .dev_release    = scsi_disk_release,
 640         .dev_groups     = sd_disk_groups,
 641 };
 642
 643 static const struct dev_pm_ops sd_pm_ops = {
 644         .suspend                = sd_suspend_system,
 645         .resume                 = sd_resume_system,
 646         .poweroff               = sd_suspend_system,
 647         .restore                = sd_resume_system,
 648         .runtime_suspend        = sd_suspend_runtime,
 649         .runtime_resume         = sd_resume_runtime,
 650 };
 651
 652 static struct scsi_driver sd_template = {
 653         .gendrv = {
 654                 .name           = "sd",
 655                 .owner          = THIS_MODULE,
 656                 .probe          = sd_probe,
 657                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
 658                 .remove         = sd_remove,
 659                 .shutdown       = sd_shutdown,
 660                 .pm             = &sd_pm_ops,
 661         },
 662         .rescan                 = sd_rescan,
 663         .init_command           = sd_init_command,
 664         .uninit_command         = sd_uninit_command,
 665         .done                   = sd_done,
 666         .eh_action              = sd_eh_action,
 667         .eh_reset               = sd_eh_reset,
 668 };
 669
 670 /*
 671  * Don't request a new module, as that could deadlock in multipath
 672  * environment.
 673  */
 674 static void sd_default_probe(dev_t devt)
 675 {
 676 }
 677
 678 /*
 679  * Device no to disk mapping:
 680  *
 681  *       major         disc2     disc  p1
 682  *   |............|.............|....|....| <- dev_t
 683  *    31        20 19          8 7  4 3  0
 684  *
 685  * Inside a major, we have 16k disks, however mapped non-
 686  * contiguously. The first 16 disks are for major0, the next
 687  * ones with major1, ... Disk 256 is for major0 again, disk 272
 688  * for major1, ...
 689  * As we stay compatible with our numbering scheme, we can reuse
 690  * the well-know SCSI majors 8, 65--71, 136--143.
 691  */
 692 static int sd_major(int major_idx)
 693 {
 694         switch (major_idx) {
 695         case 0:
 696                 return SCSI_DISK0_MAJOR;
 697         case 1 ... 7:
 698                 return SCSI_DISK1_MAJOR + major_idx - 1;
 699         case 8 ... 15:
 700                 return SCSI_DISK8_MAJOR + major_idx - 8;
 701         default:
 702                 BUG();
 703                 return 0;       /* shut up gcc */
 704         }
 705 }
 706
 707 #ifdef CONFIG_BLK_SED_OPAL
 708 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 709                 size_t len, bool send)
 710 {
 711         struct scsi_disk *sdkp = data;
 712         struct scsi_device *sdev = sdkp->device;
 713         u8 cdb[12] = { 0, };
 714         int ret;
 715
 716         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 717         cdb[1] = secp;
 718         put_unaligned_be16(spsp, &cdb[2]);
 719         put_unaligned_be32(len, &cdb[6]);
 720
 721         ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 722                 buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
 723                 RQF_PM, NULL);
 724         return ret <= 0 ? ret : -EIO;
 725 }
 726 #endif /* CONFIG_BLK_SED_OPAL */
 727
 728 /*
 729  * Look up the DIX operation based on whether the command is read or
 730  * write and whether dix and dif are enabled.
 731  */
 732 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 733 {
 734         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 735         static const unsigned int ops[] = {     /* wrt dix dif */
 736                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 737                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 738                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 739                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 740                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 741                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 742                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 743                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 744         };
 745
 746         return ops[write << 2 | dix << 1 | dif];
 747 }
 748
 749 /*
 750  * Returns a mask of the protection flags that are valid for a given DIX
 751  * operation.
 752  */
 753 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 754 {
 755         static const unsigned int flag_mask[] = {
 756                 [SCSI_PROT_NORMAL]              = 0,
 757
 758                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 759                                                   SCSI_PROT_GUARD_CHECK |
 760                                                   SCSI_PROT_REF_CHECK |
 761                                                   SCSI_PROT_REF_INCREMENT,
 762
 763                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 764                                                   SCSI_PROT_IP_CHECKSUM,
 765
 766                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 767                                                   SCSI_PROT_GUARD_CHECK |
 768                                                   SCSI_PROT_REF_CHECK |
 769                                                   SCSI_PROT_REF_INCREMENT |
 770                                                   SCSI_PROT_IP_CHECKSUM,
 771
 772                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 773                                                   SCSI_PROT_REF_INCREMENT,
 774
 775                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 776                                                   SCSI_PROT_REF_CHECK |
 777                                                   SCSI_PROT_REF_INCREMENT |
 778                                                   SCSI_PROT_IP_CHECKSUM,
 779
 780                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 781                                                   SCSI_PROT_GUARD_CHECK |
 782                                                   SCSI_PROT_REF_CHECK |
 783                                                   SCSI_PROT_REF_INCREMENT |
 784                                                   SCSI_PROT_IP_CHECKSUM,
 785         };
 786
 787         return flag_mask[prot_op];
 788 }
 789
 790 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 791                                            unsigned int dix, unsigned int dif)
 792 {
 793         struct request *rq = scsi_cmd_to_rq(scmd);
 794         struct bio *bio = rq->bio;
 795         unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
 796         unsigned int protect = 0;
 797
 798         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 799                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 800                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 801
 802                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 803                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 804         }
 805
 806         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 807                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 808
 809                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 810                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 811         }
 812
 813         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 814                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 815
 816                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 817                         protect = 3 << 5;       /* Disable target PI checking */
 818                 else
 819                         protect = 1 << 5;       /* Enable target PI checking */
 820         }
 821
 822         scsi_set_prot_op(scmd, prot_op);
 823         scsi_set_prot_type(scmd, dif);
 824         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 825
 826         return protect;
 827 }
 828
 829 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 830 {
 831         struct request_queue *q = sdkp->disk->queue;
 832         unsigned int logical_block_size = sdkp->device->sector_size;
 833         unsigned int max_blocks = 0;
 834
 835         q->limits.discard_alignment =
 836                 sdkp->unmap_alignment * logical_block_size;
 837         q->limits.discard_granularity =
 838                 max(sdkp->physical_block_size,
 839                     sdkp->unmap_granularity * logical_block_size);
 840         sdkp->provisioning_mode = mode;
 841
 842         switch (mode) {
 843
 844         case SD_LBP_FULL:
 845         case SD_LBP_DISABLE:
 846                 blk_queue_max_discard_sectors(q, 0);
 847                 return;
 848
 849         case SD_LBP_UNMAP:
 850                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 851                                           (u32)SD_MAX_WS16_BLOCKS);
 852                 break;
 853
 854         case SD_LBP_WS16:
 855                 if (sdkp->device->unmap_limit_for_ws)
 856                         max_blocks = sdkp->max_unmap_blocks;
 857                 else
 858                         max_blocks = sdkp->max_ws_blocks;
 859
 860                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 861                 break;
 862
 863         case SD_LBP_WS10:
 864                 if (sdkp->device->unmap_limit_for_ws)
 865                         max_blocks = sdkp->max_unmap_blocks;
 866                 else
 867                         max_blocks = sdkp->max_ws_blocks;
 868
 869                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 870                 break;
 871
 872         case SD_LBP_ZERO:
 873                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 874                                           (u32)SD_MAX_WS10_BLOCKS);
 875                 break;
 876         }
 877
 878         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 879 }
 880
 881 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 882 {
 883         struct scsi_device *sdp = cmd->device;
 884         struct request *rq = scsi_cmd_to_rq(cmd);
 885         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 886         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 887         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 888         unsigned int data_len = 24;
 889         char *buf;
 890
 891         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 892         if (!rq->special_vec.bv_page)
 893                 return BLK_STS_RESOURCE;
 894         clear_highpage(rq->special_vec.bv_page);
 895         rq->special_vec.bv_offset = 0;
 896         rq->special_vec.bv_len = data_len;
 897         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 898
 899         cmd->cmd_len = 10;
 900         cmd->cmnd[0] = UNMAP;
 901         cmd->cmnd[8] = 24;
 902
 903         buf = bvec_virt(&rq->special_vec);
 904         put_unaligned_be16(6 + 16, &buf[0]);
 905         put_unaligned_be16(16, &buf[2]);
 906         put_unaligned_be64(lba, &buf[8]);
 907         put_unaligned_be32(nr_blocks, &buf[16]);
 908
 909         cmd->allowed = sdkp->max_retries;
 910         cmd->transfersize = data_len;
 911         rq->timeout = SD_TIMEOUT;
 912
 913         return scsi_alloc_sgtables(cmd);
 914 }
 915
 916 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 917                 bool unmap)
 918 {
 919         struct scsi_device *sdp = cmd->device;
 920         struct request *rq = scsi_cmd_to_rq(cmd);
 921         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 922         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 923         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 924         u32 data_len = sdp->sector_size;
 925
 926         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 927         if (!rq->special_vec.bv_page)
 928                 return BLK_STS_RESOURCE;
 929         clear_highpage(rq->special_vec.bv_page);
 930         rq->special_vec.bv_offset = 0;
 931         rq->special_vec.bv_len = data_len;
 932         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 933
 934         cmd->cmd_len = 16;
 935         cmd->cmnd[0] = WRITE_SAME_16;
 936         if (unmap)
 937                 cmd->cmnd[1] = 0x8; /* UNMAP */
 938         put_unaligned_be64(lba, &cmd->cmnd[2]);
 939         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 940
 941         cmd->allowed = sdkp->max_retries;
 942         cmd->transfersize = data_len;
 943         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 944
 945         return scsi_alloc_sgtables(cmd);
 946 }
 947
 948 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 949                 bool unmap)
 950 {
 951         struct scsi_device *sdp = cmd->device;
 952         struct request *rq = scsi_cmd_to_rq(cmd);
 953         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 954         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 955         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 956         u32 data_len = sdp->sector_size;
 957
 958         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 959         if (!rq->special_vec.bv_page)
 960                 return BLK_STS_RESOURCE;
 961         clear_highpage(rq->special_vec.bv_page);
 962         rq->special_vec.bv_offset = 0;
 963         rq->special_vec.bv_len = data_len;
 964         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 965
 966         cmd->cmd_len = 10;
 967         cmd->cmnd[0] = WRITE_SAME;
 968         if (unmap)
 969                 cmd->cmnd[1] = 0x8; /* UNMAP */
 970         put_unaligned_be32(lba, &cmd->cmnd[2]);
 971         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 972
 973         cmd->allowed = sdkp->max_retries;
 974         cmd->transfersize = data_len;
 975         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 976
 977         return scsi_alloc_sgtables(cmd);
 978 }
 979
 980 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 981 {
 982         struct request *rq = scsi_cmd_to_rq(cmd);
 983         struct scsi_device *sdp = cmd->device;
 984         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
 985         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 986         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 987
 988         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 989                 switch (sdkp->zeroing_mode) {
 990                 case SD_ZERO_WS16_UNMAP:
 991                         return sd_setup_write_same16_cmnd(cmd, true);
 992                 case SD_ZERO_WS10_UNMAP:
 993                         return sd_setup_write_same10_cmnd(cmd, true);
 994                 }
 995         }
 996
 997         if (sdp->no_write_same) {
 998                 rq->rq_flags |= RQF_QUIET;
 999                 return BLK_STS_TARGET;
1000         }
1001
1002         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
1003                 return sd_setup_write_same16_cmnd(cmd, false);
1004
1005         return sd_setup_write_same10_cmnd(cmd, false);
1006 }
1007
1008 static void sd_config_write_same(struct scsi_disk *sdkp)
1009 {
1010         struct request_queue *q = sdkp->disk->queue;
1011         unsigned int logical_block_size = sdkp->device->sector_size;
1012
1013         if (sdkp->device->no_write_same) {
1014                 sdkp->max_ws_blocks = 0;
1015                 goto out;
1016         }
1017
1018         /* Some devices can not handle block counts above 0xffff despite
1019          * supporting WRITE SAME(16). Consequently we default to 64k
1020          * blocks per I/O unless the device explicitly advertises a
1021          * bigger limit.
1022          */
1023         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1024                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1025                                                    (u32)SD_MAX_WS16_BLOCKS);
1026         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1027                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1028                                                    (u32)SD_MAX_WS10_BLOCKS);
1029         else {
1030                 sdkp->device->no_write_same = 1;
1031                 sdkp->max_ws_blocks = 0;
1032         }
1033
1034         if (sdkp->lbprz && sdkp->lbpws)
1035                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1036         else if (sdkp->lbprz && sdkp->lbpws10)
1037                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1038         else if (sdkp->max_ws_blocks)
1039                 sdkp->zeroing_mode = SD_ZERO_WS;
1040         else
1041                 sdkp->zeroing_mode = SD_ZERO_WRITE;
1042
1043         if (sdkp->max_ws_blocks &&
1044             sdkp->physical_block_size > logical_block_size) {
1045                 /*
1046                  * Reporting a maximum number of blocks that is not aligned
1047                  * on the device physical size would cause a large write same
1048                  * request to be split into physically unaligned chunks by
1049                  * __blkdev_issue_write_zeroes() even if the caller of this
1050                  * functions took care to align the large request. So make sure
1051                  * the maximum reported is aligned to the device physical block
1052                  * size. This is only an optional optimization for regular
1053                  * disks, but this is mandatory to avoid failure of large write
1054                  * same requests directed at sequential write required zones of
1055                  * host-managed ZBC disks.
1056                  */
1057                 sdkp->max_ws_blocks =
1058                         round_down(sdkp->max_ws_blocks,
1059                                    bytes_to_logical(sdkp->device,
1060                                                     sdkp->physical_block_size));
1061         }
1062
1063 out:
1064         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1065                                          (logical_block_size >> 9));
1066 }
1067
1068 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1069 {
1070         struct request *rq = scsi_cmd_to_rq(cmd);
1071         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1072
1073         /* flush requests don't perform I/O, zero the S/G table */
1074         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1075
1076         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1077         cmd->cmd_len = 10;
1078         cmd->transfersize = 0;
1079         cmd->allowed = sdkp->max_retries;
1080
1081         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1082         return BLK_STS_OK;
1083 }
1084
1085 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1086                                        sector_t lba, unsigned int nr_blocks,
1087                                        unsigned char flags)
1088 {
1089         cmd->cmd_len = SD_EXT_CDB_SIZE;
1090         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1091         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1092         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1093         cmd->cmnd[10] = flags;
1094         put_unaligned_be64(lba, &cmd->cmnd[12]);
1095         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1096         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1097
1098         return BLK_STS_OK;
1099 }
1100
1101 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1102                                        sector_t lba, unsigned int nr_blocks,
1103                                        unsigned char flags)
1104 {
1105         cmd->cmd_len  = 16;
1106         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1107         cmd->cmnd[1]  = flags;
1108         cmd->cmnd[14] = 0;
1109         cmd->cmnd[15] = 0;
1110         put_unaligned_be64(lba, &cmd->cmnd[2]);
1111         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1112
1113         return BLK_STS_OK;
1114 }
1115
1116 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1117                                        sector_t lba, unsigned int nr_blocks,
1118                                        unsigned char flags)
1119 {
1120         cmd->cmd_len = 10;
1121         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1122         cmd->cmnd[1] = flags;
1123         cmd->cmnd[6] = 0;
1124         cmd->cmnd[9] = 0;
1125         put_unaligned_be32(lba, &cmd->cmnd[2]);
1126         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1127
1128         return BLK_STS_OK;
1129 }
1130
1131 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1132                                       sector_t lba, unsigned int nr_blocks,
1133                                       unsigned char flags)
1134 {
1135         /* Avoid that 0 blocks gets translated into 256 blocks. */
1136         if (WARN_ON_ONCE(nr_blocks == 0))
1137                 return BLK_STS_IOERR;
1138
1139         if (unlikely(flags & 0x8)) {
1140                 /*
1141                  * This happens only if this drive failed 10byte rw
1142                  * command with ILLEGAL_REQUEST during operation and
1143                  * thus turned off use_10_for_rw.
1144                  */
1145                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1146                 return BLK_STS_IOERR;
1147         }
1148
1149         cmd->cmd_len = 6;
1150         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1151         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1152         cmd->cmnd[2] = (lba >> 8) & 0xff;
1153         cmd->cmnd[3] = lba & 0xff;
1154         cmd->cmnd[4] = nr_blocks;
1155         cmd->cmnd[5] = 0;
1156
1157         return BLK_STS_OK;
1158 }
1159
1160 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1161 {
1162         struct request *rq = scsi_cmd_to_rq(cmd);
1163         struct scsi_device *sdp = cmd->device;
1164         struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1165         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1166         sector_t threshold;
1167         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1168         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1169         bool write = rq_data_dir(rq) == WRITE;
1170         unsigned char protect, fua;
1171         blk_status_t ret;
1172         unsigned int dif;
1173         bool dix;
1174
1175         ret = scsi_alloc_sgtables(cmd);
1176         if (ret != BLK_STS_OK)
1177                 return ret;
1178
1179         ret = BLK_STS_IOERR;
1180         if (!scsi_device_online(sdp) || sdp->changed) {
1181                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1182                 goto fail;
1183         }
1184
1185         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1186                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1187                 goto fail;
1188         }
1189
1190         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1191                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1192                 goto fail;
1193         }
1194
1195         /*
1196          * Some SD card readers can't handle accesses which touch the
1197          * last one or two logical blocks. Split accesses as needed.
1198          */
1199         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1200
1201         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1202                 if (lba < threshold) {
1203                         /* Access up to the threshold but not beyond */
1204                         nr_blocks = threshold - lba;
1205                 } else {
1206                         /* Access only a single logical block */
1207                         nr_blocks = 1;
1208                 }
1209         }
1210
1211         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1212                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1213                 if (ret)
1214                         goto fail;
1215         }
1216
1217         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1218         dix = scsi_prot_sg_count(cmd);
1219         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1220
1221         if (dif || dix)
1222                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1223         else
1224                 protect = 0;
1225
1226         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1227                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1228                                          protect | fua);
1229         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1230                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1231                                          protect | fua);
1232         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1233                    sdp->use_10_for_rw || protect) {
1234                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1235                                          protect | fua);
1236         } else {
1237                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1238                                         protect | fua);
1239         }
1240
1241         if (unlikely(ret != BLK_STS_OK))
1242                 goto fail;
1243
1244         /*
1245          * We shouldn't disconnect in the middle of a sector, so with a dumb
1246          * host adapter, it's safe to assume that we can at least transfer
1247          * this many bytes between each connect / disconnect.
1248          */
1249         cmd->transfersize = sdp->sector_size;
1250         cmd->underflow = nr_blocks << 9;
1251         cmd->allowed = sdkp->max_retries;
1252         cmd->sdb.length = nr_blocks * sdp->sector_size;
1253
1254         SCSI_LOG_HLQUEUE(1,
1255                          scmd_printk(KERN_INFO, cmd,
1256                                      "%s: block=%llu, count=%d\n", __func__,
1257                                      (unsigned long long)blk_rq_pos(rq),
1258                                      blk_rq_sectors(rq)));
1259         SCSI_LOG_HLQUEUE(2,
1260                          scmd_printk(KERN_INFO, cmd,
1261                                      "%s %d/%u 512 byte blocks.\n",
1262                                      write ? "writing" : "reading", nr_blocks,
1263                                      blk_rq_sectors(rq)));
1264
1265         /*
1266          * This indicates that the command is ready from our end to be queued.
1267          */
1268         return BLK_STS_OK;
1269 fail:
1270         scsi_free_sgtables(cmd);
1271         return ret;
1272 }
1273
1274 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1275 {
1276         struct request *rq = scsi_cmd_to_rq(cmd);
1277
1278         switch (req_op(rq)) {
1279         case REQ_OP_DISCARD:
1280                 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1281                 case SD_LBP_UNMAP:
1282                         return sd_setup_unmap_cmnd(cmd);
1283                 case SD_LBP_WS16:
1284                         return sd_setup_write_same16_cmnd(cmd, true);
1285                 case SD_LBP_WS10:
1286                         return sd_setup_write_same10_cmnd(cmd, true);
1287                 case SD_LBP_ZERO:
1288                         return sd_setup_write_same10_cmnd(cmd, false);
1289                 default:
1290                         return BLK_STS_TARGET;
1291                 }
1292         case REQ_OP_WRITE_ZEROES:
1293                 return sd_setup_write_zeroes_cmnd(cmd);
1294         case REQ_OP_FLUSH:
1295                 return sd_setup_flush_cmnd(cmd);
1296         case REQ_OP_READ:
1297         case REQ_OP_WRITE:
1298         case REQ_OP_ZONE_APPEND:
1299                 return sd_setup_read_write_cmnd(cmd);
1300         case REQ_OP_ZONE_RESET:
1301                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1302                                                    false);
1303         case REQ_OP_ZONE_RESET_ALL:
1304                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1305                                                    true);
1306         case REQ_OP_ZONE_OPEN:
1307                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1308         case REQ_OP_ZONE_CLOSE:
1309                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1310         case REQ_OP_ZONE_FINISH:
1311                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1312         default:
1313                 WARN_ON_ONCE(1);
1314                 return BLK_STS_NOTSUPP;
1315         }
1316 }
1317
1318 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1319 {
1320         struct request *rq = scsi_cmd_to_rq(SCpnt);
1321
1322         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1323                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1324 }
1325
1326 static bool sd_need_revalidate(struct block_device *bdev,
1327                 struct scsi_disk *sdkp)
1328 {
1329         if (sdkp->device->removable || sdkp->write_prot) {
1330                 if (bdev_check_media_change(bdev))
1331                         return true;
1332         }
1333
1334         /*
1335          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1336          * nothing to do with partitions, BLKRRPART is used to force a full
1337          * revalidate after things like a format for historical reasons.
1338          */
1339         return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1340 }
1341
1342 /**
1343  *      sd_open - open a scsi disk device
1344  *      @bdev: Block device of the scsi disk to open
1345  *      @mode: FMODE_* mask
1346  *
1347  *      Returns 0 if successful. Returns a negated errno value in case
1348  *      of error.
1349  *
1350  *      Note: This can be called from a user context (e.g. fsck(1) )
1351  *      or from within the kernel (e.g. as a result of a mount(1) ).
1352  *      In the latter case @inode and @filp carry an abridged amount
1353  *      of information as noted above.
1354  *
1355  *      Locking: called with bdev->bd_disk->open_mutex held.
1356  **/
1357 static int sd_open(struct block_device *bdev, fmode_t mode)
1358 {
1359         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1360         struct scsi_device *sdev = sdkp->device;
1361         int retval;
1362
1363         if (scsi_device_get(sdev))
1364                 return -ENXIO;
1365
1366         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1367
1368         /*
1369          * If the device is in error recovery, wait until it is done.
1370          * If the device is offline, then disallow any access to it.
1371          */
1372         retval = -ENXIO;
1373         if (!scsi_block_when_processing_errors(sdev))
1374                 goto error_out;
1375
1376         if (sd_need_revalidate(bdev, sdkp))
1377                 sd_revalidate_disk(bdev->bd_disk);
1378
1379         /*
1380          * If the drive is empty, just let the open fail.
1381          */
1382         retval = -ENOMEDIUM;
1383         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1384                 goto error_out;
1385
1386         /*
1387          * If the device has the write protect tab set, have the open fail
1388          * if the user expects to be able to write to the thing.
1389          */
1390         retval = -EROFS;
1391         if (sdkp->write_prot && (mode & FMODE_WRITE))
1392                 goto error_out;
1393
1394         /*
1395          * It is possible that the disk changing stuff resulted in
1396          * the device being taken offline.  If this is the case,
1397          * report this to the user, and don't pretend that the
1398          * open actually succeeded.
1399          */
1400         retval = -ENXIO;
1401         if (!scsi_device_online(sdev))
1402                 goto error_out;
1403
1404         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1405                 if (scsi_block_when_processing_errors(sdev))
1406                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1407         }
1408
1409         return 0;
1410
1411 error_out:
1412         scsi_device_put(sdev);
1413         return retval;
1414 }
1415
1416 /**
1417  *      sd_release - invoked when the (last) close(2) is called on this
1418  *      scsi disk.
1419  *      @disk: disk to release
1420  *      @mode: FMODE_* mask
1421  *
1422  *      Returns 0.
1423  *
1424  *      Note: may block (uninterruptible) if error recovery is underway
1425  *      on this disk.
1426  *
1427  *      Locking: called with bdev->bd_disk->open_mutex held.
1428  **/
1429 static void sd_release(struct gendisk *disk, fmode_t mode)
1430 {
1431         struct scsi_disk *sdkp = scsi_disk(disk);
1432         struct scsi_device *sdev = sdkp->device;
1433
1434         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1435
1436         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1437                 if (scsi_block_when_processing_errors(sdev))
1438                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1439         }
1440
1441         scsi_device_put(sdev);
1442 }
1443
1444 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1445 {
1446         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1447         struct scsi_device *sdp = sdkp->device;
1448         struct Scsi_Host *host = sdp->host;
1449         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1450         int diskinfo[4];
1451
1452         /* default to most commonly used values */
1453         diskinfo[0] = 0x40;     /* 1 << 6 */
1454         diskinfo[1] = 0x20;     /* 1 << 5 */
1455         diskinfo[2] = capacity >> 11;
1456
1457         /* override with calculated, extended default, or driver values */
1458         if (host->hostt->bios_param)
1459                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1460         else
1461                 scsicam_bios_param(bdev, capacity, diskinfo);
1462
1463         geo->heads = diskinfo[0];
1464         geo->sectors = diskinfo[1];
1465         geo->cylinders = diskinfo[2];
1466         return 0;
1467 }
1468
1469 /**
1470  *      sd_ioctl - process an ioctl
1471  *      @bdev: target block device
1472  *      @mode: FMODE_* mask
1473  *      @cmd: ioctl command number
1474  *      @arg: this is third argument given to ioctl(2) system call.
1475  *      Often contains a pointer.
1476  *
1477  *      Returns 0 if successful (some ioctls return positive numbers on
1478  *      success as well). Returns a negated errno value in case of error.
1479  *
1480  *      Note: most ioctls are forward onto the block subsystem or further
1481  *      down in the scsi subsystem.
1482  **/
1483 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1484                     unsigned int cmd, unsigned long arg)
1485 {
1486         struct gendisk *disk = bdev->bd_disk;
1487         struct scsi_disk *sdkp = scsi_disk(disk);
1488         struct scsi_device *sdp = sdkp->device;
1489         void __user *p = (void __user *)arg;
1490         int error;
1491
1492         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1493                                     "cmd=0x%x\n", disk->disk_name, cmd));
1494
1495         if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1496                 return -ENOIOCTLCMD;
1497
1498         /*
1499          * If we are in the middle of error recovery, don't let anyone
1500          * else try and use this device.  Also, if error recovery fails, it
1501          * may try and take the device offline, in which case all further
1502          * access to the device is prohibited.
1503          */
1504         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1505                         (mode & FMODE_NDELAY) != 0);
1506         if (error)
1507                 return error;
1508
1509         if (is_sed_ioctl(cmd))
1510                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1511         return scsi_ioctl(sdp, mode, cmd, p);
1512 }
1513
1514 static void set_media_not_present(struct scsi_disk *sdkp)
1515 {
1516         if (sdkp->media_present)
1517                 sdkp->device->changed = 1;
1518
1519         if (sdkp->device->removable) {
1520                 sdkp->media_present = 0;
1521                 sdkp->capacity = 0;
1522         }
1523 }
1524
1525 static int media_not_present(struct scsi_disk *sdkp,
1526                              struct scsi_sense_hdr *sshdr)
1527 {
1528         if (!scsi_sense_valid(sshdr))
1529                 return 0;
1530
1531         /* not invoked for commands that could return deferred errors */
1532         switch (sshdr->sense_key) {
1533         case UNIT_ATTENTION:
1534         case NOT_READY:
1535                 /* medium not present */
1536                 if (sshdr->asc == 0x3A) {
1537                         set_media_not_present(sdkp);
1538                         return 1;
1539                 }
1540         }
1541         return 0;
1542 }
1543
1544 /**
1545  *      sd_check_events - check media events
1546  *      @disk: kernel device descriptor
1547  *      @clearing: disk events currently being cleared
1548  *
1549  *      Returns mask of DISK_EVENT_*.
1550  *
1551  *      Note: this function is invoked from the block subsystem.
1552  **/
1553 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1554 {
1555         struct scsi_disk *sdkp = disk->private_data;
1556         struct scsi_device *sdp;
1557         int retval;
1558         bool disk_changed;
1559
1560         if (!sdkp)
1561                 return 0;
1562
1563         sdp = sdkp->device;
1564         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1565
1566         /*
1567          * If the device is offline, don't send any commands - just pretend as
1568          * if the command failed.  If the device ever comes back online, we
1569          * can deal with it then.  It is only because of unrecoverable errors
1570          * that we would ever take a device offline in the first place.
1571          */
1572         if (!scsi_device_online(sdp)) {
1573                 set_media_not_present(sdkp);
1574                 goto out;
1575         }
1576
1577         /*
1578          * Using TEST_UNIT_READY enables differentiation between drive with
1579          * no cartridge loaded - NOT READY, drive with changed cartridge -
1580          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1581          *
1582          * Drives that auto spin down. eg iomega jaz 1G, will be started
1583          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1584          * sd_revalidate() is called.
1585          */
1586         if (scsi_block_when_processing_errors(sdp)) {
1587                 struct scsi_sense_hdr sshdr = { 0, };
1588
1589                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1590                                               &sshdr);
1591
1592                 /* failed to execute TUR, assume media not present */
1593                 if (retval < 0 || host_byte(retval)) {
1594                         set_media_not_present(sdkp);
1595                         goto out;
1596                 }
1597
1598                 if (media_not_present(sdkp, &sshdr))
1599                         goto out;
1600         }
1601
1602         /*
1603          * For removable scsi disk we have to recognise the presence
1604          * of a disk in the drive.
1605          */
1606         if (!sdkp->media_present)
1607                 sdp->changed = 1;
1608         sdkp->media_present = 1;
1609 out:
1610         /*
1611          * sdp->changed is set under the following conditions:
1612          *
1613          *      Medium present state has changed in either direction.
1614          *      Device has indicated UNIT_ATTENTION.
1615          */
1616         disk_changed = sdp->changed;
1617         sdp->changed = 0;
1618         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1619 }
1620
1621 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1622 {
1623         int retries, res;
1624         struct scsi_device *sdp = sdkp->device;
1625         const int timeout = sdp->request_queue->rq_timeout
1626                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1627         struct scsi_sense_hdr my_sshdr;
1628
1629         if (!scsi_device_online(sdp))
1630                 return -ENODEV;
1631
1632         /* caller might not be interested in sense, but we need it */
1633         if (!sshdr)
1634                 sshdr = &my_sshdr;
1635
1636         for (retries = 3; retries > 0; --retries) {
1637                 unsigned char cmd[10] = { 0 };
1638
1639                 cmd[0] = SYNCHRONIZE_CACHE;
1640                 /*
1641                  * Leave the rest of the command zero to indicate
1642                  * flush everything.
1643                  */
1644                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1645                                 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1646                 if (res == 0)
1647                         break;
1648         }
1649
1650         if (res) {
1651                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1652
1653                 if (res < 0)
1654                         return res;
1655
1656                 if (scsi_status_is_check_condition(res) &&
1657                     scsi_sense_valid(sshdr)) {
1658                         sd_print_sense_hdr(sdkp, sshdr);
1659
1660                         /* we need to evaluate the error return  */
1661                         if (sshdr->asc == 0x3a ||       /* medium not present */
1662                             sshdr->asc == 0x20 ||       /* invalid command */
1663                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1664                                 /* this is no error here */
1665                                 return 0;
1666                 }
1667
1668                 switch (host_byte(res)) {
1669                 /* ignore errors due to racing a disconnection */
1670                 case DID_BAD_TARGET:
1671                 case DID_NO_CONNECT:
1672                         return 0;
1673                 /* signal the upper layer it might try again */
1674                 case DID_BUS_BUSY:
1675                 case DID_IMM_RETRY:
1676                 case DID_REQUEUE:
1677                 case DID_SOFT_ERROR:
1678                         return -EBUSY;
1679                 default:
1680                         return -EIO;
1681                 }
1682         }
1683         return 0;
1684 }
1685
1686 static void sd_rescan(struct device *dev)
1687 {
1688         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1689
1690         sd_revalidate_disk(sdkp->disk);
1691 }
1692
1693 static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1694                 enum blk_unique_id type)
1695 {
1696         struct scsi_device *sdev = scsi_disk(disk)->device;
1697         const struct scsi_vpd *vpd;
1698         const unsigned char *d;
1699         int ret = -ENXIO, len;
1700
1701         rcu_read_lock();
1702         vpd = rcu_dereference(sdev->vpd_pg83);
1703         if (!vpd)
1704                 goto out_unlock;
1705
1706         ret = -EINVAL;
1707         for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1708                 /* we only care about designators with LU association */
1709                 if (((d[1] >> 4) & 0x3) != 0x00)
1710                         continue;
1711                 if ((d[1] & 0xf) != type)
1712                         continue;
1713
1714                 /*
1715                  * Only exit early if a 16-byte descriptor was found.  Otherwise
1716                  * keep looking as one with more entropy might still show up.
1717                  */
1718                 len = d[3];
1719                 if (len != 8 && len != 12 && len != 16)
1720                         continue;
1721                 ret = len;
1722                 memcpy(id, d + 4, len);
1723                 if (len == 16)
1724                         break;
1725         }
1726 out_unlock:
1727         rcu_read_unlock();
1728         return ret;
1729 }
1730
1731 static char sd_pr_type(enum pr_type type)
1732 {
1733         switch (type) {
1734         case PR_WRITE_EXCLUSIVE:
1735                 return 0x01;
1736         case PR_EXCLUSIVE_ACCESS:
1737                 return 0x03;
1738         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1739                 return 0x05;
1740         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1741                 return 0x06;
1742         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1743                 return 0x07;
1744         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1745                 return 0x08;
1746         default:
1747                 return 0;
1748         }
1749 };
1750
1751 static int sd_pr_command(struct block_device *bdev, u8 sa,
1752                 u64 key, u64 sa_key, u8 type, u8 flags)
1753 {
1754         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1755         struct scsi_device *sdev = sdkp->device;
1756         struct scsi_sense_hdr sshdr;
1757         int result;
1758         u8 cmd[16] = { 0, };
1759         u8 data[24] = { 0, };
1760
1761         cmd[0] = PERSISTENT_RESERVE_OUT;
1762         cmd[1] = sa;
1763         cmd[2] = type;
1764         put_unaligned_be32(sizeof(data), &cmd[5]);
1765
1766         put_unaligned_be64(key, &data[0]);
1767         put_unaligned_be64(sa_key, &data[8]);
1768         data[20] = flags;
1769
1770         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1771                         &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
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         return result;
1780 }
1781
1782 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1783                 u32 flags)
1784 {
1785         if (flags & ~PR_FL_IGNORE_KEY)
1786                 return -EOPNOTSUPP;
1787         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1788                         old_key, new_key, 0,
1789                         (1 << 0) /* APTPL */);
1790 }
1791
1792 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1793                 u32 flags)
1794 {
1795         if (flags)
1796                 return -EOPNOTSUPP;
1797         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1798 }
1799
1800 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1801 {
1802         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1803 }
1804
1805 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1806                 enum pr_type type, bool abort)
1807 {
1808         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1809                              sd_pr_type(type), 0);
1810 }
1811
1812 static int sd_pr_clear(struct block_device *bdev, u64 key)
1813 {
1814         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1815 }
1816
1817 static const struct pr_ops sd_pr_ops = {
1818         .pr_register    = sd_pr_register,
1819         .pr_reserve     = sd_pr_reserve,
1820         .pr_release     = sd_pr_release,
1821         .pr_preempt     = sd_pr_preempt,
1822         .pr_clear       = sd_pr_clear,
1823 };
1824
1825 static void scsi_disk_free_disk(struct gendisk *disk)
1826 {
1827         struct scsi_disk *sdkp = scsi_disk(disk);
1828
1829         put_device(&sdkp->disk_dev);
1830 }
1831
1832 static const struct block_device_operations sd_fops = {
1833         .owner                  = THIS_MODULE,
1834         .open                   = sd_open,
1835         .release                = sd_release,
1836         .ioctl                  = sd_ioctl,
1837         .getgeo                 = sd_getgeo,
1838         .compat_ioctl           = blkdev_compat_ptr_ioctl,
1839         .check_events           = sd_check_events,
1840         .unlock_native_capacity = sd_unlock_native_capacity,
1841         .report_zones           = sd_zbc_report_zones,
1842         .get_unique_id          = sd_get_unique_id,
1843         .free_disk              = scsi_disk_free_disk,
1844         .pr_ops                 = &sd_pr_ops,
1845 };
1846
1847 /**
1848  *      sd_eh_reset - reset error handling callback
1849  *      @scmd:          sd-issued command that has failed
1850  *
1851  *      This function is called by the SCSI midlayer before starting
1852  *      SCSI EH. When counting medium access failures we have to be
1853  *      careful to register it only only once per device and SCSI EH run;
1854  *      there might be several timed out commands which will cause the
1855  *      'max_medium_access_timeouts' counter to trigger after the first
1856  *      SCSI EH run already and set the device to offline.
1857  *      So this function resets the internal counter before starting SCSI EH.
1858  **/
1859 static void sd_eh_reset(struct scsi_cmnd *scmd)
1860 {
1861         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1862
1863         /* New SCSI EH run, reset gate variable */
1864         sdkp->ignore_medium_access_errors = false;
1865 }
1866
1867 /**
1868  *      sd_eh_action - error handling callback
1869  *      @scmd:          sd-issued command that has failed
1870  *      @eh_disp:       The recovery disposition suggested by the midlayer
1871  *
1872  *      This function is called by the SCSI midlayer upon completion of an
1873  *      error test command (currently TEST UNIT READY). The result of sending
1874  *      the eh command is passed in eh_disp.  We're looking for devices that
1875  *      fail medium access commands but are OK with non access commands like
1876  *      test unit ready (so wrongly see the device as having a successful
1877  *      recovery)
1878  **/
1879 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1880 {
1881         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
1882         struct scsi_device *sdev = scmd->device;
1883
1884         if (!scsi_device_online(sdev) ||
1885             !scsi_medium_access_command(scmd) ||
1886             host_byte(scmd->result) != DID_TIME_OUT ||
1887             eh_disp != SUCCESS)
1888                 return eh_disp;
1889
1890         /*
1891          * The device has timed out executing a medium access command.
1892          * However, the TEST UNIT READY command sent during error
1893          * handling completed successfully. Either the device is in the
1894          * process of recovering or has it suffered an internal failure
1895          * that prevents access to the storage medium.
1896          */
1897         if (!sdkp->ignore_medium_access_errors) {
1898                 sdkp->medium_access_timed_out++;
1899                 sdkp->ignore_medium_access_errors = true;
1900         }
1901
1902         /*
1903          * If the device keeps failing read/write commands but TEST UNIT
1904          * READY always completes successfully we assume that medium
1905          * access is no longer possible and take the device offline.
1906          */
1907         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1908                 scmd_printk(KERN_ERR, scmd,
1909                             "Medium access timeout failure. Offlining disk!\n");
1910                 mutex_lock(&sdev->state_mutex);
1911                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1912                 mutex_unlock(&sdev->state_mutex);
1913
1914                 return SUCCESS;
1915         }
1916
1917         return eh_disp;
1918 }
1919
1920 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1921 {
1922         struct request *req = scsi_cmd_to_rq(scmd);
1923         struct scsi_device *sdev = scmd->device;
1924         unsigned int transferred, good_bytes;
1925         u64 start_lba, end_lba, bad_lba;
1926
1927         /*
1928          * Some commands have a payload smaller than the device logical
1929          * block size (e.g. INQUIRY on a 4K disk).
1930          */
1931         if (scsi_bufflen(scmd) <= sdev->sector_size)
1932                 return 0;
1933
1934         /* Check if we have a 'bad_lba' information */
1935         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1936                                      SCSI_SENSE_BUFFERSIZE,
1937                                      &bad_lba))
1938                 return 0;
1939
1940         /*
1941          * If the bad lba was reported incorrectly, we have no idea where
1942          * the error is.
1943          */
1944         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1945         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1946         if (bad_lba < start_lba || bad_lba >= end_lba)
1947                 return 0;
1948
1949         /*
1950          * resid is optional but mostly filled in.  When it's unused,
1951          * its value is zero, so we assume the whole buffer transferred
1952          */
1953         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1954
1955         /* This computation should always be done in terms of the
1956          * resolution of the device's medium.
1957          */
1958         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1959
1960         return min(good_bytes, transferred);
1961 }
1962
1963 /**
1964  *      sd_done - bottom half handler: called when the lower level
1965  *      driver has completed (successfully or otherwise) a scsi command.
1966  *      @SCpnt: mid-level's per command structure.
1967  *
1968  *      Note: potentially run from within an ISR. Must not block.
1969  **/
1970 static int sd_done(struct scsi_cmnd *SCpnt)
1971 {
1972         int result = SCpnt->result;
1973         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1974         unsigned int sector_size = SCpnt->device->sector_size;
1975         unsigned int resid;
1976         struct scsi_sense_hdr sshdr;
1977         struct request *req = scsi_cmd_to_rq(SCpnt);
1978         struct scsi_disk *sdkp = scsi_disk(req->q->disk);
1979         int sense_valid = 0;
1980         int sense_deferred = 0;
1981
1982         switch (req_op(req)) {
1983         case REQ_OP_DISCARD:
1984         case REQ_OP_WRITE_ZEROES:
1985         case REQ_OP_ZONE_RESET:
1986         case REQ_OP_ZONE_RESET_ALL:
1987         case REQ_OP_ZONE_OPEN:
1988         case REQ_OP_ZONE_CLOSE:
1989         case REQ_OP_ZONE_FINISH:
1990                 if (!result) {
1991                         good_bytes = blk_rq_bytes(req);
1992                         scsi_set_resid(SCpnt, 0);
1993                 } else {
1994                         good_bytes = 0;
1995                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1996                 }
1997                 break;
1998         default:
1999                 /*
2000                  * In case of bogus fw or device, we could end up having
2001                  * an unaligned partial completion. Check this here and force
2002                  * alignment.
2003                  */
2004                 resid = scsi_get_resid(SCpnt);
2005                 if (resid & (sector_size - 1)) {
2006                         sd_printk(KERN_INFO, sdkp,
2007                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2008                                 resid, sector_size);
2009                         scsi_print_command(SCpnt);
2010                         resid = min(scsi_bufflen(SCpnt),
2011                                     round_up(resid, sector_size));
2012                         scsi_set_resid(SCpnt, resid);
2013                 }
2014         }
2015
2016         if (result) {
2017                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2018                 if (sense_valid)
2019                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2020         }
2021         sdkp->medium_access_timed_out = 0;
2022
2023         if (!scsi_status_is_check_condition(result) &&
2024             (!sense_valid || sense_deferred))
2025                 goto out;
2026
2027         switch (sshdr.sense_key) {
2028         case HARDWARE_ERROR:
2029         case MEDIUM_ERROR:
2030                 good_bytes = sd_completed_bytes(SCpnt);
2031                 break;
2032         case RECOVERED_ERROR:
2033                 good_bytes = scsi_bufflen(SCpnt);
2034                 break;
2035         case NO_SENSE:
2036                 /* This indicates a false check condition, so ignore it.  An
2037                  * unknown amount of data was transferred so treat it as an
2038                  * error.
2039                  */
2040                 SCpnt->result = 0;
2041                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2042                 break;
2043         case ABORTED_COMMAND:
2044                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2045                         good_bytes = sd_completed_bytes(SCpnt);
2046                 break;
2047         case ILLEGAL_REQUEST:
2048                 switch (sshdr.asc) {
2049                 case 0x10:      /* DIX: Host detected corruption */
2050                         good_bytes = sd_completed_bytes(SCpnt);
2051                         break;
2052                 case 0x20:      /* INVALID COMMAND OPCODE */
2053                 case 0x24:      /* INVALID FIELD IN CDB */
2054                         switch (SCpnt->cmnd[0]) {
2055                         case UNMAP:
2056                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2057                                 break;
2058                         case WRITE_SAME_16:
2059                         case WRITE_SAME:
2060                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2061                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2062                                 } else {
2063                                         sdkp->device->no_write_same = 1;
2064                                         sd_config_write_same(sdkp);
2065                                         req->rq_flags |= RQF_QUIET;
2066                                 }
2067                                 break;
2068                         }
2069                 }
2070                 break;
2071         default:
2072                 break;
2073         }
2074
2075  out:
2076         if (sd_is_zoned(sdkp))
2077                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2078
2079         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2080                                            "sd_done: completed %d of %d bytes\n",
2081                                            good_bytes, scsi_bufflen(SCpnt)));
2082
2083         return good_bytes;
2084 }
2085
2086 /*
2087  * spinup disk - called only in sd_revalidate_disk()
2088  */
2089 static void
2090 sd_spinup_disk(struct scsi_disk *sdkp)
2091 {
2092         unsigned char cmd[10];
2093         unsigned long spintime_expire = 0;
2094         int retries, spintime;
2095         unsigned int the_result;
2096         struct scsi_sense_hdr sshdr;
2097         int sense_valid = 0;
2098
2099         spintime = 0;
2100
2101         /* Spin up drives, as required.  Only do this at boot time */
2102         /* Spinup needs to be done for module loads too. */
2103         do {
2104                 retries = 0;
2105
2106                 do {
2107                         bool media_was_present = sdkp->media_present;
2108
2109                         cmd[0] = TEST_UNIT_READY;
2110                         memset((void *) &cmd[1], 0, 9);
2111
2112                         the_result = scsi_execute_req(sdkp->device, cmd,
2113                                                       DMA_NONE, NULL, 0,
2114                                                       &sshdr, SD_TIMEOUT,
2115                                                       sdkp->max_retries, NULL);
2116
2117                         /*
2118                          * If the drive has indicated to us that it
2119                          * doesn't have any media in it, don't bother
2120                          * with any more polling.
2121                          */
2122                         if (media_not_present(sdkp, &sshdr)) {
2123                                 if (media_was_present)
2124                                         sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2125                                 return;
2126                         }
2127
2128                         if (the_result)
2129                                 sense_valid = scsi_sense_valid(&sshdr);
2130                         retries++;
2131                 } while (retries < 3 &&
2132                          (!scsi_status_is_good(the_result) ||
2133                           (scsi_status_is_check_condition(the_result) &&
2134                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2135
2136                 if (!scsi_status_is_check_condition(the_result)) {
2137                         /* no sense, TUR either succeeded or failed
2138                          * with a status error */
2139                         if(!spintime && !scsi_status_is_good(the_result)) {
2140                                 sd_print_result(sdkp, "Test Unit Ready failed",
2141                                                 the_result);
2142                         }
2143                         break;
2144                 }
2145
2146                 /*
2147                  * The device does not want the automatic start to be issued.
2148                  */
2149                 if (sdkp->device->no_start_on_add)
2150                         break;
2151
2152                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2153                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2154                                 break;  /* manual intervention required */
2155                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2156                                 break;  /* standby */
2157                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2158                                 break;  /* unavailable */
2159                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2160                                 break;  /* sanitize in progress */
2161                         /*
2162                          * Issue command to spin up drive when not ready
2163                          */
2164                         if (!spintime) {
2165                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2166                                 cmd[0] = START_STOP;
2167                                 cmd[1] = 1;     /* Return immediately */
2168                                 memset((void *) &cmd[2], 0, 8);
2169                                 cmd[4] = 1;     /* Start spin cycle */
2170                                 if (sdkp->device->start_stop_pwr_cond)
2171                                         cmd[4] |= 1 << 4;
2172                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2173                                                  NULL, 0, &sshdr,
2174                                                  SD_TIMEOUT, sdkp->max_retries,
2175                                                  NULL);
2176                                 spintime_expire = jiffies + 100 * HZ;
2177                                 spintime = 1;
2178                         }
2179                         /* Wait 1 second for next try */
2180                         msleep(1000);
2181                         printk(KERN_CONT ".");
2182
2183                 /*
2184                  * Wait for USB flash devices with slow firmware.
2185                  * Yes, this sense key/ASC combination shouldn't
2186                  * occur here.  It's characteristic of these devices.
2187                  */
2188                 } else if (sense_valid &&
2189                                 sshdr.sense_key == UNIT_ATTENTION &&
2190                                 sshdr.asc == 0x28) {
2191                         if (!spintime) {
2192                                 spintime_expire = jiffies + 5 * HZ;
2193                                 spintime = 1;
2194                         }
2195                         /* Wait 1 second for next try */
2196                         msleep(1000);
2197                 } else {
2198                         /* we don't understand the sense code, so it's
2199                          * probably pointless to loop */
2200                         if(!spintime) {
2201                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2202                                 sd_print_sense_hdr(sdkp, &sshdr);
2203                         }
2204                         break;
2205                 }
2206
2207         } while (spintime && time_before_eq(jiffies, spintime_expire));
2208
2209         if (spintime) {
2210                 if (scsi_status_is_good(the_result))
2211                         printk(KERN_CONT "ready\n");
2212                 else
2213                         printk(KERN_CONT "not responding...\n");
2214         }
2215 }
2216
2217 /*
2218  * Determine whether disk supports Data Integrity Field.
2219  */
2220 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2221 {
2222         struct scsi_device *sdp = sdkp->device;
2223         u8 type;
2224
2225         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2226                 sdkp->protection_type = 0;
2227                 return 0;
2228         }
2229
2230         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2231
2232         if (type > T10_PI_TYPE3_PROTECTION) {
2233                 sd_printk(KERN_ERR, sdkp, "formatted with unsupported"  \
2234                           " protection type %u. Disabling disk!\n",
2235                           type);
2236                 sdkp->protection_type = 0;
2237                 return -ENODEV;
2238         }
2239
2240         sdkp->protection_type = type;
2241
2242         return 0;
2243 }
2244
2245 static void sd_config_protection(struct scsi_disk *sdkp)
2246 {
2247         struct scsi_device *sdp = sdkp->device;
2248
2249         if (!sdkp->first_scan)
2250                 return;
2251
2252         sd_dif_config_host(sdkp);
2253
2254         if (!sdkp->protection_type)
2255                 return;
2256
2257         if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2258                 sd_printk(KERN_NOTICE, sdkp,
2259                           "Disabling DIF Type %u protection\n",
2260                           sdkp->protection_type);
2261                 sdkp->protection_type = 0;
2262         }
2263
2264         sd_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2265                   sdkp->protection_type);
2266 }
2267
2268 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2269                         struct scsi_sense_hdr *sshdr, int sense_valid,
2270                         int the_result)
2271 {
2272         if (sense_valid)
2273                 sd_print_sense_hdr(sdkp, sshdr);
2274         else
2275                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2276
2277         /*
2278          * Set dirty bit for removable devices if not ready -
2279          * sometimes drives will not report this properly.
2280          */
2281         if (sdp->removable &&
2282             sense_valid && sshdr->sense_key == NOT_READY)
2283                 set_media_not_present(sdkp);
2284
2285         /*
2286          * We used to set media_present to 0 here to indicate no media
2287          * in the drive, but some drives fail read capacity even with
2288          * media present, so we can't do that.
2289          */
2290         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2291 }
2292
2293 #define RC16_LEN 32
2294 #if RC16_LEN > SD_BUF_SIZE
2295 #error RC16_LEN must not be more than SD_BUF_SIZE
2296 #endif
2297
2298 #define READ_CAPACITY_RETRIES_ON_RESET  10
2299
2300 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2301                                                 unsigned char *buffer)
2302 {
2303         unsigned char cmd[16];
2304         struct scsi_sense_hdr sshdr;
2305         int sense_valid = 0;
2306         int the_result;
2307         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2308         unsigned int alignment;
2309         unsigned long long lba;
2310         unsigned sector_size;
2311
2312         if (sdp->no_read_capacity_16)
2313                 return -EINVAL;
2314
2315         do {
2316                 memset(cmd, 0, 16);
2317                 cmd[0] = SERVICE_ACTION_IN_16;
2318                 cmd[1] = SAI_READ_CAPACITY_16;
2319                 cmd[13] = RC16_LEN;
2320                 memset(buffer, 0, RC16_LEN);
2321
2322                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2323                                         buffer, RC16_LEN, &sshdr,
2324                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2325
2326                 if (media_not_present(sdkp, &sshdr))
2327                         return -ENODEV;
2328
2329                 if (the_result > 0) {
2330                         sense_valid = scsi_sense_valid(&sshdr);
2331                         if (sense_valid &&
2332                             sshdr.sense_key == ILLEGAL_REQUEST &&
2333                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2334                             sshdr.ascq == 0x00)
2335                                 /* Invalid Command Operation Code or
2336                                  * Invalid Field in CDB, just retry
2337                                  * silently with RC10 */
2338                                 return -EINVAL;
2339                         if (sense_valid &&
2340                             sshdr.sense_key == UNIT_ATTENTION &&
2341                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2342                                 /* Device reset might occur several times,
2343                                  * give it one more chance */
2344                                 if (--reset_retries > 0)
2345                                         continue;
2346                 }
2347                 retries--;
2348
2349         } while (the_result && retries);
2350
2351         if (the_result) {
2352                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2353                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2354                 return -EINVAL;
2355         }
2356
2357         sector_size = get_unaligned_be32(&buffer[8]);
2358         lba = get_unaligned_be64(&buffer[0]);
2359
2360         if (sd_read_protection_type(sdkp, buffer) < 0) {
2361                 sdkp->capacity = 0;
2362                 return -ENODEV;
2363         }
2364
2365         /* Logical blocks per physical block exponent */
2366         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2367
2368         /* RC basis */
2369         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2370
2371         /* Lowest aligned logical block */
2372         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2373         blk_queue_alignment_offset(sdp->request_queue, alignment);
2374         if (alignment && sdkp->first_scan)
2375                 sd_printk(KERN_NOTICE, sdkp,
2376                           "physical block alignment offset: %u\n", alignment);
2377
2378         if (buffer[14] & 0x80) { /* LBPME */
2379                 sdkp->lbpme = 1;
2380
2381                 if (buffer[14] & 0x40) /* LBPRZ */
2382                         sdkp->lbprz = 1;
2383
2384                 sd_config_discard(sdkp, SD_LBP_WS16);
2385         }
2386
2387         sdkp->capacity = lba + 1;
2388         return sector_size;
2389 }
2390
2391 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2392                                                 unsigned char *buffer)
2393 {
2394         unsigned char cmd[16];
2395         struct scsi_sense_hdr sshdr;
2396         int sense_valid = 0;
2397         int the_result;
2398         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2399         sector_t lba;
2400         unsigned sector_size;
2401
2402         do {
2403                 cmd[0] = READ_CAPACITY;
2404                 memset(&cmd[1], 0, 9);
2405                 memset(buffer, 0, 8);
2406
2407                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2408                                         buffer, 8, &sshdr,
2409                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2410
2411                 if (media_not_present(sdkp, &sshdr))
2412                         return -ENODEV;
2413
2414                 if (the_result > 0) {
2415                         sense_valid = scsi_sense_valid(&sshdr);
2416                         if (sense_valid &&
2417                             sshdr.sense_key == UNIT_ATTENTION &&
2418                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2419                                 /* Device reset might occur several times,
2420                                  * give it one more chance */
2421                                 if (--reset_retries > 0)
2422                                         continue;
2423                 }
2424                 retries--;
2425
2426         } while (the_result && retries);
2427
2428         if (the_result) {
2429                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2430                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2431                 return -EINVAL;
2432         }
2433
2434         sector_size = get_unaligned_be32(&buffer[4]);
2435         lba = get_unaligned_be32(&buffer[0]);
2436
2437         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2438                 /* Some buggy (usb cardreader) devices return an lba of
2439                    0xffffffff when the want to report a size of 0 (with
2440                    which they really mean no media is present) */
2441                 sdkp->capacity = 0;
2442                 sdkp->physical_block_size = sector_size;
2443                 return sector_size;
2444         }
2445
2446         sdkp->capacity = lba + 1;
2447         sdkp->physical_block_size = sector_size;
2448         return sector_size;
2449 }
2450
2451 static int sd_try_rc16_first(struct scsi_device *sdp)
2452 {
2453         if (sdp->host->max_cmd_len < 16)
2454                 return 0;
2455         if (sdp->try_rc_10_first)
2456                 return 0;
2457         if (sdp->scsi_level > SCSI_SPC_2)
2458                 return 1;
2459         if (scsi_device_protection(sdp))
2460                 return 1;
2461         return 0;
2462 }
2463
2464 /*
2465  * read disk capacity
2466  */
2467 static void
2468 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2469 {
2470         int sector_size;
2471         struct scsi_device *sdp = sdkp->device;
2472
2473         if (sd_try_rc16_first(sdp)) {
2474                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2475                 if (sector_size == -EOVERFLOW)
2476                         goto got_data;
2477                 if (sector_size == -ENODEV)
2478                         return;
2479                 if (sector_size < 0)
2480                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2481                 if (sector_size < 0)
2482                         return;
2483         } else {
2484                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2485                 if (sector_size == -EOVERFLOW)
2486                         goto got_data;
2487                 if (sector_size < 0)
2488                         return;
2489                 if ((sizeof(sdkp->capacity) > 4) &&
2490                     (sdkp->capacity > 0xffffffffULL)) {
2491                         int old_sector_size = sector_size;
2492                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2493                                         "Trying to use READ CAPACITY(16).\n");
2494                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2495                         if (sector_size < 0) {
2496                                 sd_printk(KERN_NOTICE, sdkp,
2497                                         "Using 0xffffffff as device size\n");
2498                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2499                                 sector_size = old_sector_size;
2500                                 goto got_data;
2501                         }
2502                         /* Remember that READ CAPACITY(16) succeeded */
2503                         sdp->try_rc_10_first = 0;
2504                 }
2505         }
2506
2507         /* Some devices are known to return the total number of blocks,
2508          * not the highest block number.  Some devices have versions
2509          * which do this and others which do not.  Some devices we might
2510          * suspect of doing this but we don't know for certain.
2511          *
2512          * If we know the reported capacity is wrong, decrement it.  If
2513          * we can only guess, then assume the number of blocks is even
2514          * (usually true but not always) and err on the side of lowering
2515          * the capacity.
2516          */
2517         if (sdp->fix_capacity ||
2518             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2519                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2520                                 "from its reported value: %llu\n",
2521                                 (unsigned long long) sdkp->capacity);
2522                 --sdkp->capacity;
2523         }
2524
2525 got_data:
2526         if (sector_size == 0) {
2527                 sector_size = 512;
2528                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2529                           "assuming 512.\n");
2530         }
2531
2532         if (sector_size != 512 &&
2533             sector_size != 1024 &&
2534             sector_size != 2048 &&
2535             sector_size != 4096) {
2536                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2537                           sector_size);
2538                 /*
2539                  * The user might want to re-format the drive with
2540                  * a supported sectorsize.  Once this happens, it
2541                  * would be relatively trivial to set the thing up.
2542                  * For this reason, we leave the thing in the table.
2543                  */
2544                 sdkp->capacity = 0;
2545                 /*
2546                  * set a bogus sector size so the normal read/write
2547                  * logic in the block layer will eventually refuse any
2548                  * request on this device without tripping over power
2549                  * of two sector size assumptions
2550                  */
2551                 sector_size = 512;
2552         }
2553         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2554         blk_queue_physical_block_size(sdp->request_queue,
2555                                       sdkp->physical_block_size);
2556         sdkp->device->sector_size = sector_size;
2557
2558         if (sdkp->capacity > 0xffffffff)
2559                 sdp->use_16_for_rw = 1;
2560
2561 }
2562
2563 /*
2564  * Print disk capacity
2565  */
2566 static void
2567 sd_print_capacity(struct scsi_disk *sdkp,
2568                   sector_t old_capacity)
2569 {
2570         int sector_size = sdkp->device->sector_size;
2571         char cap_str_2[10], cap_str_10[10];
2572
2573         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2574                 return;
2575
2576         string_get_size(sdkp->capacity, sector_size,
2577                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2578         string_get_size(sdkp->capacity, sector_size,
2579                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2580
2581         sd_printk(KERN_NOTICE, sdkp,
2582                   "%llu %d-byte logical blocks: (%s/%s)\n",
2583                   (unsigned long long)sdkp->capacity,
2584                   sector_size, cap_str_10, cap_str_2);
2585
2586         if (sdkp->physical_block_size != sector_size)
2587                 sd_printk(KERN_NOTICE, sdkp,
2588                           "%u-byte physical blocks\n",
2589                           sdkp->physical_block_size);
2590 }
2591
2592 /* called with buffer of length 512 */
2593 static inline int
2594 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2595                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2596                  struct scsi_sense_hdr *sshdr)
2597 {
2598         /*
2599          * If we must use MODE SENSE(10), make sure that the buffer length
2600          * is at least 8 bytes so that the mode sense header fits.
2601          */
2602         if (sdkp->device->use_10_for_ms && len < 8)
2603                 len = 8;
2604
2605         return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2606                                SD_TIMEOUT, sdkp->max_retries, data,
2607                                sshdr);
2608 }
2609
2610 /*
2611  * read write protect setting, if possible - called only in sd_revalidate_disk()
2612  * called with buffer of length SD_BUF_SIZE
2613  */
2614 static void
2615 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2616 {
2617         int res;
2618         struct scsi_device *sdp = sdkp->device;
2619         struct scsi_mode_data data;
2620         int old_wp = sdkp->write_prot;
2621
2622         set_disk_ro(sdkp->disk, 0);
2623         if (sdp->skip_ms_page_3f) {
2624                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2625                 return;
2626         }
2627
2628         if (sdp->use_192_bytes_for_3f) {
2629                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2630         } else {
2631                 /*
2632                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2633                  * We have to start carefully: some devices hang if we ask
2634                  * for more than is available.
2635                  */
2636                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2637
2638                 /*
2639                  * Second attempt: ask for page 0 When only page 0 is
2640                  * implemented, a request for page 3F may return Sense Key
2641                  * 5: Illegal Request, Sense Code 24: Invalid field in
2642                  * CDB.
2643                  */
2644                 if (res < 0)
2645                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2646
2647                 /*
2648                  * Third attempt: ask 255 bytes, as we did earlier.
2649                  */
2650                 if (res < 0)
2651                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2652                                                &data, NULL);
2653         }
2654
2655         if (res < 0) {
2656                 sd_first_printk(KERN_WARNING, sdkp,
2657                           "Test WP failed, assume Write Enabled\n");
2658         } else {
2659                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2660                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2661                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2662                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2663                                   sdkp->write_prot ? "on" : "off");
2664                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2665                 }
2666         }
2667 }
2668
2669 /*
2670  * sd_read_cache_type - called only from sd_revalidate_disk()
2671  * called with buffer of length SD_BUF_SIZE
2672  */
2673 static void
2674 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2675 {
2676         int len = 0, res;
2677         struct scsi_device *sdp = sdkp->device;
2678
2679         int dbd;
2680         int modepage;
2681         int first_len;
2682         struct scsi_mode_data data;
2683         struct scsi_sense_hdr sshdr;
2684         int old_wce = sdkp->WCE;
2685         int old_rcd = sdkp->RCD;
2686         int old_dpofua = sdkp->DPOFUA;
2687
2688
2689         if (sdkp->cache_override)
2690                 return;
2691
2692         first_len = 4;
2693         if (sdp->skip_ms_page_8) {
2694                 if (sdp->type == TYPE_RBC)
2695                         goto defaults;
2696                 else {
2697                         if (sdp->skip_ms_page_3f)
2698                                 goto defaults;
2699                         modepage = 0x3F;
2700                         if (sdp->use_192_bytes_for_3f)
2701                                 first_len = 192;
2702                         dbd = 0;
2703                 }
2704         } else if (sdp->type == TYPE_RBC) {
2705                 modepage = 6;
2706                 dbd = 8;
2707         } else {
2708                 modepage = 8;
2709                 dbd = 0;
2710         }
2711
2712         /* cautiously ask */
2713         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2714                         &data, &sshdr);
2715
2716         if (res < 0)
2717                 goto bad_sense;
2718
2719         if (!data.header_length) {
2720                 modepage = 6;
2721                 first_len = 0;
2722                 sd_first_printk(KERN_ERR, sdkp,
2723                                 "Missing header in MODE_SENSE response\n");
2724         }
2725
2726         /* that went OK, now ask for the proper length */
2727         len = data.length;
2728
2729         /*
2730          * We're only interested in the first three bytes, actually.
2731          * But the data cache page is defined for the first 20.
2732          */
2733         if (len < 3)
2734                 goto bad_sense;
2735         else if (len > SD_BUF_SIZE) {
2736                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2737                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2738                 len = SD_BUF_SIZE;
2739         }
2740         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2741                 len = 192;
2742
2743         /* Get the data */
2744         if (len > first_len)
2745                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2746                                 &data, &sshdr);
2747
2748         if (!res) {
2749                 int offset = data.header_length + data.block_descriptor_length;
2750
2751                 while (offset < len) {
2752                         u8 page_code = buffer[offset] & 0x3F;
2753                         u8 spf       = buffer[offset] & 0x40;
2754
2755                         if (page_code == 8 || page_code == 6) {
2756                                 /* We're interested only in the first 3 bytes.
2757                                  */
2758                                 if (len - offset <= 2) {
2759                                         sd_first_printk(KERN_ERR, sdkp,
2760                                                 "Incomplete mode parameter "
2761                                                         "data\n");
2762                                         goto defaults;
2763                                 } else {
2764                                         modepage = page_code;
2765                                         goto Page_found;
2766                                 }
2767                         } else {
2768                                 /* Go to the next page */
2769                                 if (spf && len - offset > 3)
2770                                         offset += 4 + (buffer[offset+2] << 8) +
2771                                                 buffer[offset+3];
2772                                 else if (!spf && len - offset > 1)
2773                                         offset += 2 + buffer[offset+1];
2774                                 else {
2775                                         sd_first_printk(KERN_ERR, sdkp,
2776                                                         "Incomplete mode "
2777                                                         "parameter data\n");
2778                                         goto defaults;
2779                                 }
2780                         }
2781                 }
2782
2783                 sd_first_printk(KERN_WARNING, sdkp,
2784                                 "No Caching mode page found\n");
2785                 goto defaults;
2786
2787         Page_found:
2788                 if (modepage == 8) {
2789                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2790                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2791                 } else {
2792                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2793                         sdkp->RCD = 0;
2794                 }
2795
2796                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2797                 if (sdp->broken_fua) {
2798                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2799                         sdkp->DPOFUA = 0;
2800                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2801                            !sdkp->device->use_16_for_rw) {
2802                         sd_first_printk(KERN_NOTICE, sdkp,
2803                                   "Uses READ/WRITE(6), disabling FUA\n");
2804                         sdkp->DPOFUA = 0;
2805                 }
2806
2807                 /* No cache flush allowed for write protected devices */
2808                 if (sdkp->WCE && sdkp->write_prot)
2809                         sdkp->WCE = 0;
2810
2811                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2812                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2813                         sd_printk(KERN_NOTICE, sdkp,
2814                                   "Write cache: %s, read cache: %s, %s\n",
2815                                   sdkp->WCE ? "enabled" : "disabled",
2816                                   sdkp->RCD ? "disabled" : "enabled",
2817                                   sdkp->DPOFUA ? "supports DPO and FUA"
2818                                   : "doesn't support DPO or FUA");
2819
2820                 return;
2821         }
2822
2823 bad_sense:
2824         if (scsi_sense_valid(&sshdr) &&
2825             sshdr.sense_key == ILLEGAL_REQUEST &&
2826             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2827                 /* Invalid field in CDB */
2828                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2829         else
2830                 sd_first_printk(KERN_ERR, sdkp,
2831                                 "Asking for cache data failed\n");
2832
2833 defaults:
2834         if (sdp->wce_default_on) {
2835                 sd_first_printk(KERN_NOTICE, sdkp,
2836                                 "Assuming drive cache: write back\n");
2837                 sdkp->WCE = 1;
2838         } else {
2839                 sd_first_printk(KERN_WARNING, sdkp,
2840                                 "Assuming drive cache: write through\n");
2841                 sdkp->WCE = 0;
2842         }
2843         sdkp->RCD = 0;
2844         sdkp->DPOFUA = 0;
2845 }
2846
2847 /*
2848  * The ATO bit indicates whether the DIF application tag is available
2849  * for use by the operating system.
2850  */
2851 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2852 {
2853         int res, offset;
2854         struct scsi_device *sdp = sdkp->device;
2855         struct scsi_mode_data data;
2856         struct scsi_sense_hdr sshdr;
2857
2858         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2859                 return;
2860
2861         if (sdkp->protection_type == 0)
2862                 return;
2863
2864         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2865                               sdkp->max_retries, &data, &sshdr);
2866
2867         if (res < 0 || !data.header_length ||
2868             data.length < 6) {
2869                 sd_first_printk(KERN_WARNING, sdkp,
2870                           "getting Control mode page failed, assume no ATO\n");
2871
2872                 if (scsi_sense_valid(&sshdr))
2873                         sd_print_sense_hdr(sdkp, &sshdr);
2874
2875                 return;
2876         }
2877
2878         offset = data.header_length + data.block_descriptor_length;
2879
2880         if ((buffer[offset] & 0x3f) != 0x0a) {
2881                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2882                 return;
2883         }
2884
2885         if ((buffer[offset + 5] & 0x80) == 0)
2886                 return;
2887
2888         sdkp->ATO = 1;
2889
2890         return;
2891 }
2892
2893 /**
2894  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2895  * @sdkp: disk to query
2896  */
2897 static void sd_read_block_limits(struct scsi_disk *sdkp)
2898 {
2899         struct scsi_vpd *vpd;
2900
2901         rcu_read_lock();
2902
2903         vpd = rcu_dereference(sdkp->device->vpd_pgb0);
2904         if (!vpd || vpd->len < 16)
2905                 goto out;
2906
2907         sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
2908         sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
2909         sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
2910
2911         if (vpd->len >= 64) {
2912                 unsigned int lba_count, desc_count;
2913
2914                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
2915
2916                 if (!sdkp->lbpme)
2917                         goto out;
2918
2919                 lba_count = get_unaligned_be32(&vpd->data[20]);
2920                 desc_count = get_unaligned_be32(&vpd->data[24]);
2921
2922                 if (lba_count && desc_count)
2923                         sdkp->max_unmap_blocks = lba_count;
2924
2925                 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
2926
2927                 if (vpd->data[32] & 0x80)
2928                         sdkp->unmap_alignment =
2929                                 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
2930
2931                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2932
2933                         if (sdkp->max_unmap_blocks)
2934                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2935                         else
2936                                 sd_config_discard(sdkp, SD_LBP_WS16);
2937
2938                 } else {        /* LBP VPD page tells us what to use */
2939                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2940                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2941                         else if (sdkp->lbpws)
2942                                 sd_config_discard(sdkp, SD_LBP_WS16);
2943                         else if (sdkp->lbpws10)
2944                                 sd_config_discard(sdkp, SD_LBP_WS10);
2945                         else
2946                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2947                 }
2948         }
2949
2950  out:
2951         rcu_read_unlock();
2952 }
2953
2954 /**
2955  * sd_read_block_characteristics - Query block dev. characteristics
2956  * @sdkp: disk to query
2957  */
2958 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2959 {
2960         struct request_queue *q = sdkp->disk->queue;
2961         struct scsi_vpd *vpd;
2962         u16 rot;
2963         u8 zoned;
2964
2965         rcu_read_lock();
2966         vpd = rcu_dereference(sdkp->device->vpd_pgb1);
2967
2968         if (!vpd || vpd->len < 8) {
2969                 rcu_read_unlock();
2970                 return;
2971         }
2972
2973         rot = get_unaligned_be16(&vpd->data[4]);
2974         zoned = (vpd->data[8] >> 4) & 3;
2975         rcu_read_unlock();
2976
2977         if (rot == 1) {
2978                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2979                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2980         }
2981
2982         if (sdkp->device->type == TYPE_ZBC) {
2983                 /*
2984                  * Host-managed: Per ZBC and ZAC specifications, writes in
2985                  * sequential write required zones of host-managed devices must
2986                  * be aligned to the device physical block size.
2987                  */
2988                 disk_set_zoned(sdkp->disk, BLK_ZONED_HM);
2989                 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
2990         } else {
2991                 sdkp->zoned = zoned;
2992                 if (sdkp->zoned == 1) {
2993                         /* Host-aware */
2994                         disk_set_zoned(sdkp->disk, BLK_ZONED_HA);
2995                 } else {
2996                         /* Regular disk or drive managed disk */
2997                         disk_set_zoned(sdkp->disk, BLK_ZONED_NONE);
2998                 }
2999         }
3000
3001         if (!sdkp->first_scan)
3002                 return;
3003
3004         if (blk_queue_is_zoned(q)) {
3005                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3006                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3007         } else {
3008                 if (sdkp->zoned == 1)
3009                         sd_printk(KERN_NOTICE, sdkp,
3010                                   "Host-aware SMR disk used as regular disk\n");
3011                 else if (sdkp->zoned == 2)
3012                         sd_printk(KERN_NOTICE, sdkp,
3013                                   "Drive-managed SMR disk\n");
3014         }
3015 }
3016
3017 /**
3018  * sd_read_block_provisioning - Query provisioning VPD page
3019  * @sdkp: disk to query
3020  */
3021 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3022 {
3023         struct scsi_vpd *vpd;
3024
3025         if (sdkp->lbpme == 0)
3026                 return;
3027
3028         rcu_read_lock();
3029         vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3030
3031         if (!vpd || vpd->len < 8) {
3032                 rcu_read_unlock();
3033                 return;
3034         }
3035
3036         sdkp->lbpvpd    = 1;
3037         sdkp->lbpu      = (vpd->data[5] >> 7) & 1; /* UNMAP */
3038         sdkp->lbpws     = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3039         sdkp->lbpws10   = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3040         rcu_read_unlock();
3041 }
3042
3043 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3044 {
3045         struct scsi_device *sdev = sdkp->device;
3046
3047         if (sdev->host->no_write_same) {
3048                 sdev->no_write_same = 1;
3049
3050                 return;
3051         }
3052
3053         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3054                 struct scsi_vpd *vpd;
3055
3056                 sdev->no_report_opcodes = 1;
3057
3058                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3059                  * CODES is unsupported and the device has an ATA
3060                  * Information VPD page (SAT).
3061                  */
3062                 rcu_read_lock();
3063                 vpd = rcu_dereference(sdev->vpd_pg89);
3064                 if (vpd)
3065                         sdev->no_write_same = 1;
3066                 rcu_read_unlock();
3067         }
3068
3069         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3070                 sdkp->ws16 = 1;
3071
3072         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3073                 sdkp->ws10 = 1;
3074 }
3075
3076 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3077 {
3078         struct scsi_device *sdev = sdkp->device;
3079
3080         if (!sdev->security_supported)
3081                 return;
3082
3083         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3084                         SECURITY_PROTOCOL_IN) == 1 &&
3085             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3086                         SECURITY_PROTOCOL_OUT) == 1)
3087                 sdkp->security = 1;
3088 }
3089
3090 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3091 {
3092         return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3093 }
3094
3095 /**
3096  * sd_read_cpr - Query concurrent positioning ranges
3097  * @sdkp:       disk to query
3098  */
3099 static void sd_read_cpr(struct scsi_disk *sdkp)
3100 {
3101         struct blk_independent_access_ranges *iars = NULL;
3102         unsigned char *buffer = NULL;
3103         unsigned int nr_cpr = 0;
3104         int i, vpd_len, buf_len = SD_BUF_SIZE;
3105         u8 *desc;
3106
3107         /*
3108          * We need to have the capacity set first for the block layer to be
3109          * able to check the ranges.
3110          */
3111         if (sdkp->first_scan)
3112                 return;
3113
3114         if (!sdkp->capacity)
3115                 goto out;
3116
3117         /*
3118          * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3119          * leading to a maximum page size of 64 + 256*32 bytes.
3120          */
3121         buf_len = 64 + 256*32;
3122         buffer = kmalloc(buf_len, GFP_KERNEL);
3123         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3124                 goto out;
3125
3126         /* We must have at least a 64B header and one 32B range descriptor */
3127         vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3128         if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3129                 sd_printk(KERN_ERR, sdkp,
3130                           "Invalid Concurrent Positioning Ranges VPD page\n");
3131                 goto out;
3132         }
3133
3134         nr_cpr = (vpd_len - 64) / 32;
3135         if (nr_cpr == 1) {
3136                 nr_cpr = 0;
3137                 goto out;
3138         }
3139
3140         iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3141         if (!iars) {
3142                 nr_cpr = 0;
3143                 goto out;
3144         }
3145
3146         desc = &buffer[64];
3147         for (i = 0; i < nr_cpr; i++, desc += 32) {
3148                 if (desc[0] != i) {
3149                         sd_printk(KERN_ERR, sdkp,
3150                                 "Invalid Concurrent Positioning Range number\n");
3151                         nr_cpr = 0;
3152                         break;
3153                 }
3154
3155                 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3156                 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3157         }
3158
3159 out:
3160         disk_set_independent_access_ranges(sdkp->disk, iars);
3161         if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3162                 sd_printk(KERN_NOTICE, sdkp,
3163                           "%u concurrent positioning ranges\n", nr_cpr);
3164                 sdkp->nr_actuators = nr_cpr;
3165         }
3166
3167         kfree(buffer);
3168 }
3169
3170 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3171 {
3172         struct scsi_device *sdp = sdkp->device;
3173         unsigned int min_xfer_bytes =
3174                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3175
3176         if (sdkp->min_xfer_blocks == 0)
3177                 return false;
3178
3179         if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3180                 sd_first_printk(KERN_WARNING, sdkp,
3181                                 "Preferred minimum I/O size %u bytes not a " \
3182                                 "multiple of physical block size (%u bytes)\n",
3183                                 min_xfer_bytes, sdkp->physical_block_size);
3184                 sdkp->min_xfer_blocks = 0;
3185                 return false;
3186         }
3187
3188         sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3189                         min_xfer_bytes);
3190         return true;
3191 }
3192
3193 /*
3194  * Determine the device's preferred I/O size for reads and writes
3195  * unless the reported value is unreasonably small, large, not a
3196  * multiple of the physical block size, or simply garbage.
3197  */
3198 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3199                                       unsigned int dev_max)
3200 {
3201         struct scsi_device *sdp = sdkp->device;
3202         unsigned int opt_xfer_bytes =
3203                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3204         unsigned int min_xfer_bytes =
3205                 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3206
3207         if (sdkp->opt_xfer_blocks == 0)
3208                 return false;
3209
3210         if (sdkp->opt_xfer_blocks > dev_max) {
3211                 sd_first_printk(KERN_WARNING, sdkp,
3212                                 "Optimal transfer size %u logical blocks " \
3213                                 "> dev_max (%u logical blocks)\n",
3214                                 sdkp->opt_xfer_blocks, dev_max);
3215                 return false;
3216         }
3217
3218         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3219                 sd_first_printk(KERN_WARNING, sdkp,
3220                                 "Optimal transfer size %u logical blocks " \
3221                                 "> sd driver limit (%u logical blocks)\n",
3222                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3223                 return false;
3224         }
3225
3226         if (opt_xfer_bytes < PAGE_SIZE) {
3227                 sd_first_printk(KERN_WARNING, sdkp,
3228                                 "Optimal transfer size %u bytes < " \
3229                                 "PAGE_SIZE (%u bytes)\n",
3230                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3231                 return false;
3232         }
3233
3234         if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3235                 sd_first_printk(KERN_WARNING, sdkp,
3236                                 "Optimal transfer size %u bytes not a " \
3237                                 "multiple of preferred minimum block " \
3238                                 "size (%u bytes)\n",
3239                                 opt_xfer_bytes, min_xfer_bytes);
3240                 return false;
3241         }
3242
3243         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3244                 sd_first_printk(KERN_WARNING, sdkp,
3245                                 "Optimal transfer size %u bytes not a " \
3246                                 "multiple of physical block size (%u bytes)\n",
3247                                 opt_xfer_bytes, sdkp->physical_block_size);
3248                 return false;
3249         }
3250
3251         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3252                         opt_xfer_bytes);
3253         return true;
3254 }
3255
3256 /**
3257  *      sd_revalidate_disk - called the first time a new disk is seen,
3258  *      performs disk spin up, read_capacity, etc.
3259  *      @disk: struct gendisk we care about
3260  **/
3261 static int sd_revalidate_disk(struct gendisk *disk)
3262 {
3263         struct scsi_disk *sdkp = scsi_disk(disk);
3264         struct scsi_device *sdp = sdkp->device;
3265         struct request_queue *q = sdkp->disk->queue;
3266         sector_t old_capacity = sdkp->capacity;
3267         unsigned char *buffer;
3268         unsigned int dev_max, rw_max;
3269
3270         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3271                                       "sd_revalidate_disk\n"));
3272
3273         /*
3274          * If the device is offline, don't try and read capacity or any
3275          * of the other niceties.
3276          */
3277         if (!scsi_device_online(sdp))
3278                 goto out;
3279
3280         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3281         if (!buffer) {
3282                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3283                           "allocation failure.\n");
3284                 goto out;
3285         }
3286
3287         sd_spinup_disk(sdkp);
3288
3289         /*
3290          * Without media there is no reason to ask; moreover, some devices
3291          * react badly if we do.
3292          */
3293         if (sdkp->media_present) {
3294                 sd_read_capacity(sdkp, buffer);
3295
3296                 /*
3297                  * set the default to rotational.  All non-rotational devices
3298                  * support the block characteristics VPD page, which will
3299                  * cause this to be updated correctly and any device which
3300                  * doesn't support it should be treated as rotational.
3301                  */
3302                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3303                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3304
3305                 if (scsi_device_supports_vpd(sdp)) {
3306                         sd_read_block_provisioning(sdkp);
3307                         sd_read_block_limits(sdkp);
3308                         sd_read_block_characteristics(sdkp);
3309                         sd_zbc_read_zones(sdkp, buffer);
3310                         sd_read_cpr(sdkp);
3311                 }
3312
3313                 sd_print_capacity(sdkp, old_capacity);
3314
3315                 sd_read_write_protect_flag(sdkp, buffer);
3316                 sd_read_cache_type(sdkp, buffer);
3317                 sd_read_app_tag_own(sdkp, buffer);
3318                 sd_read_write_same(sdkp, buffer);
3319                 sd_read_security(sdkp, buffer);
3320                 sd_config_protection(sdkp);
3321         }
3322
3323         /*
3324          * We now have all cache related info, determine how we deal
3325          * with flush requests.
3326          */
3327         sd_set_flush_flag(sdkp);
3328
3329         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3330         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3331
3332         /* Some devices report a maximum block count for READ/WRITE requests. */
3333         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3334         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3335
3336         if (sd_validate_min_xfer_size(sdkp))
3337                 blk_queue_io_min(sdkp->disk->queue,
3338                                  logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3339         else
3340                 blk_queue_io_min(sdkp->disk->queue, 0);
3341
3342         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3343                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3344                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3345         } else {
3346                 q->limits.io_opt = 0;
3347                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3348                                       (sector_t)BLK_DEF_MAX_SECTORS);
3349         }
3350
3351         /*
3352          * Limit default to SCSI host optimal sector limit if set. There may be
3353          * an impact on performance for when the size of a request exceeds this
3354          * host limit.
3355          */
3356         rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3357
3358         /* Do not exceed controller limit */
3359         rw_max = min(rw_max, queue_max_hw_sectors(q));
3360
3361         /*
3362          * Only update max_sectors if previously unset or if the current value
3363          * exceeds the capabilities of the hardware.
3364          */
3365         if (sdkp->first_scan ||
3366             q->limits.max_sectors > q->limits.max_dev_sectors ||
3367             q->limits.max_sectors > q->limits.max_hw_sectors)
3368                 q->limits.max_sectors = rw_max;
3369
3370         sdkp->first_scan = 0;
3371
3372         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3373         sd_config_write_same(sdkp);
3374         kfree(buffer);
3375
3376         /*
3377          * For a zoned drive, revalidating the zones can be done only once
3378          * the gendisk capacity is set. So if this fails, set back the gendisk
3379          * capacity to 0.
3380          */
3381         if (sd_zbc_revalidate_zones(sdkp))
3382                 set_capacity_and_notify(disk, 0);
3383
3384  out:
3385         return 0;
3386 }
3387
3388 /**
3389  *      sd_unlock_native_capacity - unlock native capacity
3390  *      @disk: struct gendisk to set capacity for
3391  *
3392  *      Block layer calls this function if it detects that partitions
3393  *      on @disk reach beyond the end of the device.  If the SCSI host
3394  *      implements ->unlock_native_capacity() method, it's invoked to
3395  *      give it a chance to adjust the device capacity.
3396  *
3397  *      CONTEXT:
3398  *      Defined by block layer.  Might sleep.
3399  */
3400 static void sd_unlock_native_capacity(struct gendisk *disk)
3401 {
3402         struct scsi_device *sdev = scsi_disk(disk)->device;
3403
3404         if (sdev->host->hostt->unlock_native_capacity)
3405                 sdev->host->hostt->unlock_native_capacity(sdev);
3406 }
3407
3408 /**
3409  *      sd_format_disk_name - format disk name
3410  *      @prefix: name prefix - ie. "sd" for SCSI disks
3411  *      @index: index of the disk to format name for
3412  *      @buf: output buffer
3413  *      @buflen: length of the output buffer
3414  *
3415  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3416  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3417  *      which is followed by sdaaa.
3418  *
3419  *      This is basically 26 base counting with one extra 'nil' entry
3420  *      at the beginning from the second digit on and can be
3421  *      determined using similar method as 26 base conversion with the
3422  *      index shifted -1 after each digit is computed.
3423  *
3424  *      CONTEXT:
3425  *      Don't care.
3426  *
3427  *      RETURNS:
3428  *      0 on success, -errno on failure.
3429  */
3430 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3431 {
3432         const int base = 'z' - 'a' + 1;
3433         char *begin = buf + strlen(prefix);
3434         char *end = buf + buflen;
3435         char *p;
3436         int unit;
3437
3438         p = end - 1;
3439         *p = '\0';
3440         unit = base;
3441         do {
3442                 if (p == begin)
3443                         return -EINVAL;
3444                 *--p = 'a' + (index % unit);
3445                 index = (index / unit) - 1;
3446         } while (index >= 0);
3447
3448         memmove(begin, p, end - p);
3449         memcpy(buf, prefix, strlen(prefix));
3450
3451         return 0;
3452 }
3453
3454 /**
3455  *      sd_probe - called during driver initialization and whenever a
3456  *      new scsi device is attached to the system. It is called once
3457  *      for each scsi device (not just disks) present.
3458  *      @dev: pointer to device object
3459  *
3460  *      Returns 0 if successful (or not interested in this scsi device
3461  *      (e.g. scanner)); 1 when there is an error.
3462  *
3463  *      Note: this function is invoked from the scsi mid-level.
3464  *      This function sets up the mapping between a given
3465  *      <host,channel,id,lun> (found in sdp) and new device name
3466  *      (e.g. /dev/sda). More precisely it is the block device major
3467  *      and minor number that is chosen here.
3468  *
3469  *      Assume sd_probe is not re-entrant (for time being)
3470  *      Also think about sd_probe() and sd_remove() running coincidentally.
3471  **/
3472 static int sd_probe(struct device *dev)
3473 {
3474         struct scsi_device *sdp = to_scsi_device(dev);
3475         struct scsi_disk *sdkp;
3476         struct gendisk *gd;
3477         int index;
3478         int error;
3479
3480         scsi_autopm_get_device(sdp);
3481         error = -ENODEV;
3482         if (sdp->type != TYPE_DISK &&
3483             sdp->type != TYPE_ZBC &&
3484             sdp->type != TYPE_MOD &&
3485             sdp->type != TYPE_RBC)
3486                 goto out;
3487
3488         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3489                 sdev_printk(KERN_WARNING, sdp,
3490                             "Unsupported ZBC host-managed device.\n");
3491                 goto out;
3492         }
3493
3494         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3495                                         "sd_probe\n"));
3496
3497         error = -ENOMEM;
3498         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3499         if (!sdkp)
3500                 goto out;
3501
3502         gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3503                                          &sd_bio_compl_lkclass);
3504         if (!gd)
3505                 goto out_free;
3506
3507         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3508         if (index < 0) {
3509                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3510                 goto out_put;
3511         }
3512
3513         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3514         if (error) {
3515                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3516                 goto out_free_index;
3517         }
3518
3519         sdkp->device = sdp;
3520         sdkp->disk = gd;
3521         sdkp->index = index;
3522         sdkp->max_retries = SD_MAX_RETRIES;
3523         atomic_set(&sdkp->openers, 0);
3524         atomic_set(&sdkp->device->ioerr_cnt, 0);
3525
3526         if (!sdp->request_queue->rq_timeout) {
3527                 if (sdp->type != TYPE_MOD)
3528                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3529                 else
3530                         blk_queue_rq_timeout(sdp->request_queue,
3531                                              SD_MOD_TIMEOUT);
3532         }
3533
3534         device_initialize(&sdkp->disk_dev);
3535         sdkp->disk_dev.parent = get_device(dev);
3536         sdkp->disk_dev.class = &sd_disk_class;
3537         dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3538
3539         error = device_add(&sdkp->disk_dev);
3540         if (error) {
3541                 put_device(&sdkp->disk_dev);
3542                 goto out;
3543         }
3544
3545         dev_set_drvdata(dev, sdkp);
3546
3547         gd->major = sd_major((index & 0xf0) >> 4);
3548         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3549         gd->minors = SD_MINORS;
3550
3551         gd->fops = &sd_fops;
3552         gd->private_data = sdkp;
3553
3554         /* defaults, until the device tells us otherwise */
3555         sdp->sector_size = 512;
3556         sdkp->capacity = 0;
3557         sdkp->media_present = 1;
3558         sdkp->write_prot = 0;
3559         sdkp->cache_override = 0;
3560         sdkp->WCE = 0;
3561         sdkp->RCD = 0;
3562         sdkp->ATO = 0;
3563         sdkp->first_scan = 1;
3564         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3565
3566         sd_revalidate_disk(gd);
3567
3568         if (sdp->removable) {
3569                 gd->flags |= GENHD_FL_REMOVABLE;
3570                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3571                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3572         }
3573
3574         blk_pm_runtime_init(sdp->request_queue, dev);
3575         if (sdp->rpm_autosuspend) {
3576                 pm_runtime_set_autosuspend_delay(dev,
3577                         sdp->host->hostt->rpm_autosuspend_delay);
3578         }
3579
3580         error = device_add_disk(dev, gd, NULL);
3581         if (error) {
3582                 put_device(&sdkp->disk_dev);
3583                 put_disk(gd);
3584                 goto out;
3585         }
3586
3587         if (sdkp->security) {
3588                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3589                 if (sdkp->opal_dev)
3590                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3591         }
3592
3593         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3594                   sdp->removable ? "removable " : "");
3595         scsi_autopm_put_device(sdp);
3596
3597         return 0;
3598
3599  out_free_index:
3600         ida_free(&sd_index_ida, index);
3601  out_put:
3602         put_disk(gd);
3603  out_free:
3604         kfree(sdkp);
3605  out:
3606         scsi_autopm_put_device(sdp);
3607         return error;
3608 }
3609
3610 /**
3611  *      sd_remove - called whenever a scsi disk (previously recognized by
3612  *      sd_probe) is detached from the system. It is called (potentially
3613  *      multiple times) during sd module unload.
3614  *      @dev: pointer to device object
3615  *
3616  *      Note: this function is invoked from the scsi mid-level.
3617  *      This function potentially frees up a device name (e.g. /dev/sdc)
3618  *      that could be re-used by a subsequent sd_probe().
3619  *      This function is not called when the built-in sd driver is "exit-ed".
3620  **/
3621 static int sd_remove(struct device *dev)
3622 {
3623         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3624
3625         scsi_autopm_get_device(sdkp->device);
3626
3627         device_del(&sdkp->disk_dev);
3628         del_gendisk(sdkp->disk);
3629         if (!sdkp->suspended)
3630                 sd_shutdown(dev);
3631
3632         put_disk(sdkp->disk);
3633         return 0;
3634 }
3635
3636 static void scsi_disk_release(struct device *dev)
3637 {
3638         struct scsi_disk *sdkp = to_scsi_disk(dev);
3639
3640         ida_free(&sd_index_ida, sdkp->index);
3641         sd_zbc_free_zone_info(sdkp);
3642         put_device(&sdkp->device->sdev_gendev);
3643         free_opal_dev(sdkp->opal_dev);
3644
3645         kfree(sdkp);
3646 }
3647
3648 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3649 {
3650         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3651         struct scsi_sense_hdr sshdr;
3652         struct scsi_device *sdp = sdkp->device;
3653         int res;
3654
3655         if (start)
3656                 cmd[4] |= 1;    /* START */
3657
3658         if (sdp->start_stop_pwr_cond)
3659                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3660
3661         if (!scsi_device_online(sdp))
3662                 return -ENODEV;
3663
3664         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3665                         SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3666         if (res) {
3667                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3668                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3669                         sd_print_sense_hdr(sdkp, &sshdr);
3670                         /* 0x3a is medium not present */
3671                         if (sshdr.asc == 0x3a)
3672                                 res = 0;
3673                 }
3674         }
3675
3676         /* SCSI error codes must not go to the generic layer */
3677         if (res)
3678                 return -EIO;
3679
3680         return 0;
3681 }
3682
3683 /*
3684  * Send a SYNCHRONIZE CACHE instruction down to the device through
3685  * the normal SCSI command structure.  Wait for the command to
3686  * complete.
3687  */
3688 static void sd_shutdown(struct device *dev)
3689 {
3690         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3691
3692         if (!sdkp)
3693                 return;         /* this can happen */
3694
3695         if (pm_runtime_suspended(dev))
3696                 return;
3697
3698         if (sdkp->WCE && sdkp->media_present) {
3699                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3700                 sd_sync_cache(sdkp, NULL);
3701         }
3702
3703         if (system_state != SYSTEM_RESTART &&
3704             sdkp->device->manage_system_start_stop) {
3705                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3706                 sd_start_stop_device(sdkp, 0);
3707         }
3708 }
3709
3710 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
3711 {
3712         return (sdev->manage_system_start_stop && !runtime) ||
3713                 (sdev->manage_runtime_start_stop && runtime);
3714 }
3715
3716 static int sd_suspend_common(struct device *dev, bool runtime)
3717 {
3718         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3719         struct scsi_sense_hdr sshdr;
3720         int ret = 0;
3721
3722         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3723                 return 0;
3724
3725         if (sdkp->WCE && sdkp->media_present) {
3726                 if (!sdkp->device->silence_suspend)
3727                         sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3728                 ret = sd_sync_cache(sdkp, &sshdr);
3729
3730                 if (ret) {
3731                         /* ignore OFFLINE device */
3732                         if (ret == -ENODEV)
3733                                 return 0;
3734
3735                         if (!scsi_sense_valid(&sshdr) ||
3736                             sshdr.sense_key != ILLEGAL_REQUEST)
3737                                 return ret;
3738
3739                         /*
3740                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3741                          * doesn't support sync. There's not much to do and
3742                          * suspend shouldn't fail.
3743                          */
3744                         ret = 0;
3745                 }
3746         }
3747
3748         if (sd_do_start_stop(sdkp->device, runtime)) {
3749                 if (!sdkp->device->silence_suspend)
3750                         sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3751                 /* an error is not worth aborting a system sleep */
3752                 ret = sd_start_stop_device(sdkp, 0);
3753                 if (!runtime)
3754                         ret = 0;
3755         }
3756
3757         if (!ret)
3758                 sdkp->suspended = true;
3759
3760         return ret;
3761 }
3762
3763 static int sd_suspend_system(struct device *dev)
3764 {
3765         if (pm_runtime_suspended(dev))
3766                 return 0;
3767
3768         return sd_suspend_common(dev, false);
3769 }
3770
3771 static int sd_suspend_runtime(struct device *dev)
3772 {
3773         return sd_suspend_common(dev, true);
3774 }
3775
3776 static int sd_resume(struct device *dev, bool runtime)
3777 {
3778         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3779         int ret = 0;
3780
3781         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3782                 return 0;
3783
3784         if (!sd_do_start_stop(sdkp->device, runtime)) {
3785                 sdkp->suspended = false;
3786                 return 0;
3787         }
3788
3789         if (!sdkp->device->no_start_on_resume) {
3790                 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3791                 ret = sd_start_stop_device(sdkp, 1);
3792         }
3793
3794         if (!ret) {
3795                 opal_unlock_from_suspend(sdkp->opal_dev);
3796                 sdkp->suspended = false;
3797         }
3798
3799         return ret;
3800 }
3801
3802 static int sd_resume_system(struct device *dev)
3803 {
3804         if (pm_runtime_suspended(dev))
3805                 return 0;
3806
3807         return sd_resume(dev, false);
3808 }
3809
3810 static int sd_resume_runtime(struct device *dev)
3811 {
3812         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3813         struct scsi_device *sdp;
3814
3815         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3816                 return 0;
3817
3818         sdp = sdkp->device;
3819
3820         if (sdp->ignore_media_change) {
3821                 /* clear the device's sense data */
3822                 static const u8 cmd[10] = { REQUEST_SENSE };
3823
3824                 if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3825                                  NULL, sdp->request_queue->rq_timeout, 1, 0,
3826                                  RQF_PM, NULL))
3827                         sd_printk(KERN_NOTICE, sdkp,
3828                                   "Failed to clear sense data\n");
3829         }
3830
3831         return sd_resume(dev, true);
3832 }
3833
3834 /**
3835  *      init_sd - entry point for this driver (both when built in or when
3836  *      a module).
3837  *
3838  *      Note: this function registers this driver with the scsi mid-level.
3839  **/
3840 static int __init init_sd(void)
3841 {
3842         int majors = 0, i, err;
3843
3844         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3845
3846         for (i = 0; i < SD_MAJORS; i++) {
3847                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3848                         continue;
3849                 majors++;
3850         }
3851
3852         if (!majors)
3853                 return -ENODEV;
3854
3855         err = class_register(&sd_disk_class);
3856         if (err)
3857                 goto err_out;
3858
3859         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3860                                          0, 0, NULL);
3861         if (!sd_cdb_cache) {
3862                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3863                 err = -ENOMEM;
3864                 goto err_out_class;
3865         }
3866
3867         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3868         if (!sd_page_pool) {
3869                 printk(KERN_ERR "sd: can't init discard page pool\n");
3870                 err = -ENOMEM;
3871                 goto err_out_cache;
3872         }
3873
3874         err = scsi_register_driver(&sd_template.gendrv);
3875         if (err)
3876                 goto err_out_driver;
3877
3878         return 0;
3879
3880 err_out_driver:
3881         mempool_destroy(sd_page_pool);
3882
3883 err_out_cache:
3884         kmem_cache_destroy(sd_cdb_cache);
3885
3886 err_out_class:
3887         class_unregister(&sd_disk_class);
3888 err_out:
3889         for (i = 0; i < SD_MAJORS; i++)
3890                 unregister_blkdev(sd_major(i), "sd");
3891         return err;
3892 }
3893
3894 /**
3895  *      exit_sd - exit point for this driver (when it is a module).
3896  *
3897  *      Note: this function unregisters this driver from the scsi mid-level.
3898  **/
3899 static void __exit exit_sd(void)
3900 {
3901         int i;
3902
3903         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3904
3905         scsi_unregister_driver(&sd_template.gendrv);
3906         mempool_destroy(sd_page_pool);
3907         kmem_cache_destroy(sd_cdb_cache);
3908
3909         class_unregister(&sd_disk_class);
3910
3911         for (i = 0; i < SD_MAJORS; i++)
3912                 unregister_blkdev(sd_major(i), "sd");
3913 }
3914
3915 module_init(init_sd);
3916 module_exit(exit_sd);
3917
3918 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3919 {
3920         scsi_print_sense_hdr(sdkp->device,
3921                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3922 }
3923
3924 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3925 {
3926         const char *hb_string = scsi_hostbyte_string(result);
3927
3928         if (hb_string)
3929                 sd_printk(KERN_INFO, sdkp,
3930                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3931                           hb_string ? hb_string : "invalid",
3932                           "DRIVER_OK");
3933         else
3934                 sd_printk(KERN_INFO, sdkp,
3935                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3936                           msg, host_byte(result), "DRIVER_OK");
3937 }