drivers/scsi/sd.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *      sd.c Copyright (C) 1992 Drew Eckhardt
   4  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
   5  *
   6  *      Linux scsi disk driver
   7  *              Initial versions: Drew Eckhardt
   8  *              Subsequent revisions: Eric Youngdale
   9  *      Modification history:
  10  *       - Drew Eckhardt <drew@colorado.edu> original
  11  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
  12  *         outstanding request, and other enhancements.
  13  *         Support loadable low-level scsi drivers.
  14  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
  15  *         eight major numbers.
  16  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
  17  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
  18  *         sd_init and cleanups.
  19  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
  20  *         not being read in sd_open. Fix problem where removable media
  21  *         could be ejected after sd_open.
  22  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
  23  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
  24  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
  25  *         Support 32k/1M disks.
  26  *
  27  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
  28  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
  29  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
  30  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
  31  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
  32  *      Note: when the logging level is set by the user, it must be greater
  33  *      than the level indicated above to trigger output.
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/fs.h>
  38 #include <linux/kernel.h>
  39 #include <linux/mm.h>
  40 #include <linux/bio.h>
  41 #include <linux/genhd.h>
  42 #include <linux/hdreg.h>
  43 #include <linux/errno.h>
  44 #include <linux/idr.h>
  45 #include <linux/interrupt.h>
  46 #include <linux/init.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/blkpg.h>
  49 #include <linux/blk-pm.h>
  50 #include <linux/delay.h>
  51 #include <linux/major.h>
  52 #include <linux/mutex.h>
  53 #include <linux/string_helpers.h>
  54 #include <linux/async.h>
  55 #include <linux/slab.h>
  56 #include <linux/sed-opal.h>
  57 #include <linux/pm_runtime.h>
  58 #include <linux/pr.h>
  59 #include <linux/t10-pi.h>
  60 #include <linux/uaccess.h>
  61 #include <asm/unaligned.h>
  62
  63 #include <scsi/scsi.h>
  64 #include <scsi/scsi_cmnd.h>
  65 #include <scsi/scsi_dbg.h>
  66 #include <scsi/scsi_device.h>
  67 #include <scsi/scsi_driver.h>
  68 #include <scsi/scsi_eh.h>
  69 #include <scsi/scsi_host.h>
  70 #include <scsi/scsi_ioctl.h>
  71 #include <scsi/scsicam.h>
  72
  73 #include "sd.h"
  74 #include "scsi_priv.h"
  75 #include "scsi_logging.h"
  76
  77 MODULE_AUTHOR("Eric Youngdale");
  78 MODULE_DESCRIPTION("SCSI disk (sd) driver");
  79 MODULE_LICENSE("GPL");
  80
  81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
  82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
  83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
  84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
  85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
  86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
  87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
  88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
  89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
  90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
  91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
  92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
  93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
  94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
  95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
  96 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
  97 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
  98 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
  99 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
 100 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
 101
 102 #define SD_MINORS       16
 103
 104 static void sd_config_discard(struct scsi_disk *, unsigned int);
 105 static void sd_config_write_same(struct scsi_disk *);
 106 static int  sd_revalidate_disk(struct gendisk *);
 107 static void sd_unlock_native_capacity(struct gendisk *disk);
 108 static int  sd_probe(struct device *);
 109 static int  sd_remove(struct device *);
 110 static void sd_shutdown(struct device *);
 111 static int sd_suspend_system(struct device *);
 112 static int sd_suspend_runtime(struct device *);
 113 static int sd_resume(struct device *);
 114 static int sd_resume_runtime(struct device *);
 115 static void sd_rescan(struct device *);
 116 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
 117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
 118 static int sd_done(struct scsi_cmnd *);
 119 static void sd_eh_reset(struct scsi_cmnd *);
 120 static int sd_eh_action(struct scsi_cmnd *, int);
 121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 122 static void scsi_disk_release(struct device *cdev);
 123
 124 static DEFINE_IDA(sd_index_ida);
 125
 126 /* This semaphore is used to mediate the 0->1 reference get in the
 127  * face of object destruction (i.e. we can't allow a get on an
 128  * object after last put) */
 129 static DEFINE_MUTEX(sd_ref_mutex);
 130
 131 static struct kmem_cache *sd_cdb_cache;
 132 static mempool_t *sd_cdb_pool;
 133 static mempool_t *sd_page_pool;
 134 static struct lock_class_key sd_bio_compl_lkclass;
 135
 136 static const char *sd_cache_types[] = {
 137         "write through", "none", "write back",
 138         "write back, no read (daft)"
 139 };
 140
 141 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 142 {
 143         bool wc = false, fua = false;
 144
 145         if (sdkp->WCE) {
 146                 wc = true;
 147                 if (sdkp->DPOFUA)
 148                         fua = true;
 149         }
 150
 151         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 152 }
 153
 154 static ssize_t
 155 cache_type_store(struct device *dev, struct device_attribute *attr,
 156                  const char *buf, size_t count)
 157 {
 158         int ct, rcd, wce, sp;
 159         struct scsi_disk *sdkp = to_scsi_disk(dev);
 160         struct scsi_device *sdp = sdkp->device;
 161         char buffer[64];
 162         char *buffer_data;
 163         struct scsi_mode_data data;
 164         struct scsi_sense_hdr sshdr;
 165         static const char temp[] = "temporary ";
 166         int len;
 167
 168         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 169                 /* no cache control on RBC devices; theoretically they
 170                  * can do it, but there's probably so many exceptions
 171                  * it's not worth the risk */
 172                 return -EINVAL;
 173
 174         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 175                 buf += sizeof(temp) - 1;
 176                 sdkp->cache_override = 1;
 177         } else {
 178                 sdkp->cache_override = 0;
 179         }
 180
 181         ct = sysfs_match_string(sd_cache_types, buf);
 182         if (ct < 0)
 183                 return -EINVAL;
 184
 185         rcd = ct & 0x01 ? 1 : 0;
 186         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 187
 188         if (sdkp->cache_override) {
 189                 sdkp->WCE = wce;
 190                 sdkp->RCD = rcd;
 191                 sd_set_flush_flag(sdkp);
 192                 return count;
 193         }
 194
 195         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 196                             sdkp->max_retries, &data, NULL))
 197                 return -EINVAL;
 198         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 199                   data.block_descriptor_length);
 200         buffer_data = buffer + data.header_length +
 201                 data.block_descriptor_length;
 202         buffer_data[2] &= ~0x05;
 203         buffer_data[2] |= wce << 2 | rcd;
 204         sp = buffer_data[0] & 0x80 ? 1 : 0;
 205         buffer_data[0] &= ~0x80;
 206
 207         /*
 208          * Ensure WP, DPOFUA, and RESERVED fields are cleared in
 209          * received mode parameter buffer before doing MODE SELECT.
 210          */
 211         data.device_specific = 0;
 212
 213         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
 214                              sdkp->max_retries, &data, &sshdr)) {
 215                 if (scsi_sense_valid(&sshdr))
 216                         sd_print_sense_hdr(sdkp, &sshdr);
 217                 return -EINVAL;
 218         }
 219         sd_revalidate_disk(sdkp->disk);
 220         return count;
 221 }
 222
 223 static ssize_t
 224 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
 225                        char *buf)
 226 {
 227         struct scsi_disk *sdkp = to_scsi_disk(dev);
 228         struct scsi_device *sdp = sdkp->device;
 229
 230         return sprintf(buf, "%u\n", sdp->manage_start_stop);
 231 }
 232
 233 static ssize_t
 234 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
 235                         const char *buf, size_t count)
 236 {
 237         struct scsi_disk *sdkp = to_scsi_disk(dev);
 238         struct scsi_device *sdp = sdkp->device;
 239         bool v;
 240
 241         if (!capable(CAP_SYS_ADMIN))
 242                 return -EACCES;
 243
 244         if (kstrtobool(buf, &v))
 245                 return -EINVAL;
 246
 247         sdp->manage_start_stop = v;
 248
 249         return count;
 250 }
 251 static DEVICE_ATTR_RW(manage_start_stop);
 252
 253 static ssize_t
 254 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 255 {
 256         struct scsi_disk *sdkp = to_scsi_disk(dev);
 257
 258         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 259 }
 260
 261 static ssize_t
 262 allow_restart_store(struct device *dev, struct device_attribute *attr,
 263                     const char *buf, size_t count)
 264 {
 265         bool v;
 266         struct scsi_disk *sdkp = to_scsi_disk(dev);
 267         struct scsi_device *sdp = sdkp->device;
 268
 269         if (!capable(CAP_SYS_ADMIN))
 270                 return -EACCES;
 271
 272         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 273                 return -EINVAL;
 274
 275         if (kstrtobool(buf, &v))
 276                 return -EINVAL;
 277
 278         sdp->allow_restart = v;
 279
 280         return count;
 281 }
 282 static DEVICE_ATTR_RW(allow_restart);
 283
 284 static ssize_t
 285 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 286 {
 287         struct scsi_disk *sdkp = to_scsi_disk(dev);
 288         int ct = sdkp->RCD + 2*sdkp->WCE;
 289
 290         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 291 }
 292 static DEVICE_ATTR_RW(cache_type);
 293
 294 static ssize_t
 295 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 296 {
 297         struct scsi_disk *sdkp = to_scsi_disk(dev);
 298
 299         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 300 }
 301 static DEVICE_ATTR_RO(FUA);
 302
 303 static ssize_t
 304 protection_type_show(struct device *dev, struct device_attribute *attr,
 305                      char *buf)
 306 {
 307         struct scsi_disk *sdkp = to_scsi_disk(dev);
 308
 309         return sprintf(buf, "%u\n", sdkp->protection_type);
 310 }
 311
 312 static ssize_t
 313 protection_type_store(struct device *dev, struct device_attribute *attr,
 314                       const char *buf, size_t count)
 315 {
 316         struct scsi_disk *sdkp = to_scsi_disk(dev);
 317         unsigned int val;
 318         int err;
 319
 320         if (!capable(CAP_SYS_ADMIN))
 321                 return -EACCES;
 322
 323         err = kstrtouint(buf, 10, &val);
 324
 325         if (err)
 326                 return err;
 327
 328         if (val <= T10_PI_TYPE3_PROTECTION)
 329                 sdkp->protection_type = val;
 330
 331         return count;
 332 }
 333 static DEVICE_ATTR_RW(protection_type);
 334
 335 static ssize_t
 336 protection_mode_show(struct device *dev, struct device_attribute *attr,
 337                      char *buf)
 338 {
 339         struct scsi_disk *sdkp = to_scsi_disk(dev);
 340         struct scsi_device *sdp = sdkp->device;
 341         unsigned int dif, dix;
 342
 343         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 344         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 345
 346         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 347                 dif = 0;
 348                 dix = 1;
 349         }
 350
 351         if (!dif && !dix)
 352                 return sprintf(buf, "none\n");
 353
 354         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 355 }
 356 static DEVICE_ATTR_RO(protection_mode);
 357
 358 static ssize_t
 359 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 360 {
 361         struct scsi_disk *sdkp = to_scsi_disk(dev);
 362
 363         return sprintf(buf, "%u\n", sdkp->ATO);
 364 }
 365 static DEVICE_ATTR_RO(app_tag_own);
 366
 367 static ssize_t
 368 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 369                        char *buf)
 370 {
 371         struct scsi_disk *sdkp = to_scsi_disk(dev);
 372
 373         return sprintf(buf, "%u\n", sdkp->lbpme);
 374 }
 375 static DEVICE_ATTR_RO(thin_provisioning);
 376
 377 /* sysfs_match_string() requires dense arrays */
 378 static const char *lbp_mode[] = {
 379         [SD_LBP_FULL]           = "full",
 380         [SD_LBP_UNMAP]          = "unmap",
 381         [SD_LBP_WS16]           = "writesame_16",
 382         [SD_LBP_WS10]           = "writesame_10",
 383         [SD_LBP_ZERO]           = "writesame_zero",
 384         [SD_LBP_DISABLE]        = "disabled",
 385 };
 386
 387 static ssize_t
 388 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 389                        char *buf)
 390 {
 391         struct scsi_disk *sdkp = to_scsi_disk(dev);
 392
 393         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 394 }
 395
 396 static ssize_t
 397 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 398                         const char *buf, size_t count)
 399 {
 400         struct scsi_disk *sdkp = to_scsi_disk(dev);
 401         struct scsi_device *sdp = sdkp->device;
 402         int mode;
 403
 404         if (!capable(CAP_SYS_ADMIN))
 405                 return -EACCES;
 406
 407         if (sd_is_zoned(sdkp)) {
 408                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 409                 return count;
 410         }
 411
 412         if (sdp->type != TYPE_DISK)
 413                 return -EINVAL;
 414
 415         mode = sysfs_match_string(lbp_mode, buf);
 416         if (mode < 0)
 417                 return -EINVAL;
 418
 419         sd_config_discard(sdkp, mode);
 420
 421         return count;
 422 }
 423 static DEVICE_ATTR_RW(provisioning_mode);
 424
 425 /* sysfs_match_string() requires dense arrays */
 426 static const char *zeroing_mode[] = {
 427         [SD_ZERO_WRITE]         = "write",
 428         [SD_ZERO_WS]            = "writesame",
 429         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 430         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 431 };
 432
 433 static ssize_t
 434 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 435                   char *buf)
 436 {
 437         struct scsi_disk *sdkp = to_scsi_disk(dev);
 438
 439         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 440 }
 441
 442 static ssize_t
 443 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 444                    const char *buf, size_t count)
 445 {
 446         struct scsi_disk *sdkp = to_scsi_disk(dev);
 447         int mode;
 448
 449         if (!capable(CAP_SYS_ADMIN))
 450                 return -EACCES;
 451
 452         mode = sysfs_match_string(zeroing_mode, buf);
 453         if (mode < 0)
 454                 return -EINVAL;
 455
 456         sdkp->zeroing_mode = mode;
 457
 458         return count;
 459 }
 460 static DEVICE_ATTR_RW(zeroing_mode);
 461
 462 static ssize_t
 463 max_medium_access_timeouts_show(struct device *dev,
 464                                 struct device_attribute *attr, char *buf)
 465 {
 466         struct scsi_disk *sdkp = to_scsi_disk(dev);
 467
 468         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 469 }
 470
 471 static ssize_t
 472 max_medium_access_timeouts_store(struct device *dev,
 473                                  struct device_attribute *attr, const char *buf,
 474                                  size_t count)
 475 {
 476         struct scsi_disk *sdkp = to_scsi_disk(dev);
 477         int err;
 478
 479         if (!capable(CAP_SYS_ADMIN))
 480                 return -EACCES;
 481
 482         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 483
 484         return err ? err : count;
 485 }
 486 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 487
 488 static ssize_t
 489 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 490                            char *buf)
 491 {
 492         struct scsi_disk *sdkp = to_scsi_disk(dev);
 493
 494         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 495 }
 496
 497 static ssize_t
 498 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 499                             const char *buf, size_t count)
 500 {
 501         struct scsi_disk *sdkp = to_scsi_disk(dev);
 502         struct scsi_device *sdp = sdkp->device;
 503         unsigned long max;
 504         int err;
 505
 506         if (!capable(CAP_SYS_ADMIN))
 507                 return -EACCES;
 508
 509         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 510                 return -EINVAL;
 511
 512         err = kstrtoul(buf, 10, &max);
 513
 514         if (err)
 515                 return err;
 516
 517         if (max == 0)
 518                 sdp->no_write_same = 1;
 519         else if (max <= SD_MAX_WS16_BLOCKS) {
 520                 sdp->no_write_same = 0;
 521                 sdkp->max_ws_blocks = max;
 522         }
 523
 524         sd_config_write_same(sdkp);
 525
 526         return count;
 527 }
 528 static DEVICE_ATTR_RW(max_write_same_blocks);
 529
 530 static ssize_t
 531 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
 532 {
 533         struct scsi_disk *sdkp = to_scsi_disk(dev);
 534
 535         if (sdkp->device->type == TYPE_ZBC)
 536                 return sprintf(buf, "host-managed\n");
 537         if (sdkp->zoned == 1)
 538                 return sprintf(buf, "host-aware\n");
 539         if (sdkp->zoned == 2)
 540                 return sprintf(buf, "drive-managed\n");
 541         return sprintf(buf, "none\n");
 542 }
 543 static DEVICE_ATTR_RO(zoned_cap);
 544
 545 static ssize_t
 546 max_retries_store(struct device *dev, struct device_attribute *attr,
 547                   const char *buf, size_t count)
 548 {
 549         struct scsi_disk *sdkp = to_scsi_disk(dev);
 550         struct scsi_device *sdev = sdkp->device;
 551         int retries, err;
 552
 553         err = kstrtoint(buf, 10, &retries);
 554         if (err)
 555                 return err;
 556
 557         if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
 558                 sdkp->max_retries = retries;
 559                 return count;
 560         }
 561
 562         sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
 563                     SD_MAX_RETRIES);
 564         return -EINVAL;
 565 }
 566
 567 static ssize_t
 568 max_retries_show(struct device *dev, struct device_attribute *attr,
 569                  char *buf)
 570 {
 571         struct scsi_disk *sdkp = to_scsi_disk(dev);
 572
 573         return sprintf(buf, "%d\n", sdkp->max_retries);
 574 }
 575
 576 static DEVICE_ATTR_RW(max_retries);
 577
 578 static struct attribute *sd_disk_attrs[] = {
 579         &dev_attr_cache_type.attr,
 580         &dev_attr_FUA.attr,
 581         &dev_attr_allow_restart.attr,
 582         &dev_attr_manage_start_stop.attr,
 583         &dev_attr_protection_type.attr,
 584         &dev_attr_protection_mode.attr,
 585         &dev_attr_app_tag_own.attr,
 586         &dev_attr_thin_provisioning.attr,
 587         &dev_attr_provisioning_mode.attr,
 588         &dev_attr_zeroing_mode.attr,
 589         &dev_attr_max_write_same_blocks.attr,
 590         &dev_attr_max_medium_access_timeouts.attr,
 591         &dev_attr_zoned_cap.attr,
 592         &dev_attr_max_retries.attr,
 593         NULL,
 594 };
 595 ATTRIBUTE_GROUPS(sd_disk);
 596
 597 static struct class sd_disk_class = {
 598         .name           = "scsi_disk",
 599         .owner          = THIS_MODULE,
 600         .dev_release    = scsi_disk_release,
 601         .dev_groups     = sd_disk_groups,
 602 };
 603
 604 static const struct dev_pm_ops sd_pm_ops = {
 605         .suspend                = sd_suspend_system,
 606         .resume                 = sd_resume,
 607         .poweroff               = sd_suspend_system,
 608         .restore                = sd_resume,
 609         .runtime_suspend        = sd_suspend_runtime,
 610         .runtime_resume         = sd_resume_runtime,
 611 };
 612
 613 static struct scsi_driver sd_template = {
 614         .gendrv = {
 615                 .name           = "sd",
 616                 .owner          = THIS_MODULE,
 617                 .probe          = sd_probe,
 618                 .probe_type     = PROBE_PREFER_ASYNCHRONOUS,
 619                 .remove         = sd_remove,
 620                 .shutdown       = sd_shutdown,
 621                 .pm             = &sd_pm_ops,
 622         },
 623         .rescan                 = sd_rescan,
 624         .init_command           = sd_init_command,
 625         .uninit_command         = sd_uninit_command,
 626         .done                   = sd_done,
 627         .eh_action              = sd_eh_action,
 628         .eh_reset               = sd_eh_reset,
 629 };
 630
 631 /*
 632  * Don't request a new module, as that could deadlock in multipath
 633  * environment.
 634  */
 635 static void sd_default_probe(dev_t devt)
 636 {
 637 }
 638
 639 /*
 640  * Device no to disk mapping:
 641  *
 642  *       major         disc2     disc  p1
 643  *   |............|.............|....|....| <- dev_t
 644  *    31        20 19          8 7  4 3  0
 645  *
 646  * Inside a major, we have 16k disks, however mapped non-
 647  * contiguously. The first 16 disks are for major0, the next
 648  * ones with major1, ... Disk 256 is for major0 again, disk 272
 649  * for major1, ...
 650  * As we stay compatible with our numbering scheme, we can reuse
 651  * the well-know SCSI majors 8, 65--71, 136--143.
 652  */
 653 static int sd_major(int major_idx)
 654 {
 655         switch (major_idx) {
 656         case 0:
 657                 return SCSI_DISK0_MAJOR;
 658         case 1 ... 7:
 659                 return SCSI_DISK1_MAJOR + major_idx - 1;
 660         case 8 ... 15:
 661                 return SCSI_DISK8_MAJOR + major_idx - 8;
 662         default:
 663                 BUG();
 664                 return 0;       /* shut up gcc */
 665         }
 666 }
 667
 668 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 669 {
 670         struct scsi_disk *sdkp = NULL;
 671
 672         mutex_lock(&sd_ref_mutex);
 673
 674         if (disk->private_data) {
 675                 sdkp = scsi_disk(disk);
 676                 if (scsi_device_get(sdkp->device) == 0)
 677                         get_device(&sdkp->dev);
 678                 else
 679                         sdkp = NULL;
 680         }
 681         mutex_unlock(&sd_ref_mutex);
 682         return sdkp;
 683 }
 684
 685 static void scsi_disk_put(struct scsi_disk *sdkp)
 686 {
 687         struct scsi_device *sdev = sdkp->device;
 688
 689         mutex_lock(&sd_ref_mutex);
 690         put_device(&sdkp->dev);
 691         scsi_device_put(sdev);
 692         mutex_unlock(&sd_ref_mutex);
 693 }
 694
 695 #ifdef CONFIG_BLK_SED_OPAL
 696 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 697                 size_t len, bool send)
 698 {
 699         struct scsi_disk *sdkp = data;
 700         struct scsi_device *sdev = sdkp->device;
 701         u8 cdb[12] = { 0, };
 702         int ret;
 703
 704         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 705         cdb[1] = secp;
 706         put_unaligned_be16(spsp, &cdb[2]);
 707         put_unaligned_be32(len, &cdb[6]);
 708
 709         ret = scsi_execute(sdev, cdb, send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 710                 buffer, len, NULL, NULL, SD_TIMEOUT, sdkp->max_retries, 0,
 711                 RQF_PM, NULL);
 712         return ret <= 0 ? ret : -EIO;
 713 }
 714 #endif /* CONFIG_BLK_SED_OPAL */
 715
 716 /*
 717  * Look up the DIX operation based on whether the command is read or
 718  * write and whether dix and dif are enabled.
 719  */
 720 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 721 {
 722         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 723         static const unsigned int ops[] = {     /* wrt dix dif */
 724                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 725                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 726                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 727                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 728                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 729                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 730                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 731                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 732         };
 733
 734         return ops[write << 2 | dix << 1 | dif];
 735 }
 736
 737 /*
 738  * Returns a mask of the protection flags that are valid for a given DIX
 739  * operation.
 740  */
 741 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 742 {
 743         static const unsigned int flag_mask[] = {
 744                 [SCSI_PROT_NORMAL]              = 0,
 745
 746                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 747                                                   SCSI_PROT_GUARD_CHECK |
 748                                                   SCSI_PROT_REF_CHECK |
 749                                                   SCSI_PROT_REF_INCREMENT,
 750
 751                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 752                                                   SCSI_PROT_IP_CHECKSUM,
 753
 754                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 755                                                   SCSI_PROT_GUARD_CHECK |
 756                                                   SCSI_PROT_REF_CHECK |
 757                                                   SCSI_PROT_REF_INCREMENT |
 758                                                   SCSI_PROT_IP_CHECKSUM,
 759
 760                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 761                                                   SCSI_PROT_REF_INCREMENT,
 762
 763                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 764                                                   SCSI_PROT_REF_CHECK |
 765                                                   SCSI_PROT_REF_INCREMENT |
 766                                                   SCSI_PROT_IP_CHECKSUM,
 767
 768                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 769                                                   SCSI_PROT_GUARD_CHECK |
 770                                                   SCSI_PROT_REF_CHECK |
 771                                                   SCSI_PROT_REF_INCREMENT |
 772                                                   SCSI_PROT_IP_CHECKSUM,
 773         };
 774
 775         return flag_mask[prot_op];
 776 }
 777
 778 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 779                                            unsigned int dix, unsigned int dif)
 780 {
 781         struct request *rq = scsi_cmd_to_rq(scmd);
 782         struct bio *bio = rq->bio;
 783         unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
 784         unsigned int protect = 0;
 785
 786         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 787                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 788                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 789
 790                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 791                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 792         }
 793
 794         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 795                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 796
 797                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 798                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 799         }
 800
 801         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 802                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 803
 804                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 805                         protect = 3 << 5;       /* Disable target PI checking */
 806                 else
 807                         protect = 1 << 5;       /* Enable target PI checking */
 808         }
 809
 810         scsi_set_prot_op(scmd, prot_op);
 811         scsi_set_prot_type(scmd, dif);
 812         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 813
 814         return protect;
 815 }
 816
 817 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 818 {
 819         struct request_queue *q = sdkp->disk->queue;
 820         unsigned int logical_block_size = sdkp->device->sector_size;
 821         unsigned int max_blocks = 0;
 822
 823         q->limits.discard_alignment =
 824                 sdkp->unmap_alignment * logical_block_size;
 825         q->limits.discard_granularity =
 826                 max(sdkp->physical_block_size,
 827                     sdkp->unmap_granularity * logical_block_size);
 828         sdkp->provisioning_mode = mode;
 829
 830         switch (mode) {
 831
 832         case SD_LBP_FULL:
 833         case SD_LBP_DISABLE:
 834                 blk_queue_max_discard_sectors(q, 0);
 835                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 836                 return;
 837
 838         case SD_LBP_UNMAP:
 839                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 840                                           (u32)SD_MAX_WS16_BLOCKS);
 841                 break;
 842
 843         case SD_LBP_WS16:
 844                 if (sdkp->device->unmap_limit_for_ws)
 845                         max_blocks = sdkp->max_unmap_blocks;
 846                 else
 847                         max_blocks = sdkp->max_ws_blocks;
 848
 849                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 850                 break;
 851
 852         case SD_LBP_WS10:
 853                 if (sdkp->device->unmap_limit_for_ws)
 854                         max_blocks = sdkp->max_unmap_blocks;
 855                 else
 856                         max_blocks = sdkp->max_ws_blocks;
 857
 858                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 859                 break;
 860
 861         case SD_LBP_ZERO:
 862                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 863                                           (u32)SD_MAX_WS10_BLOCKS);
 864                 break;
 865         }
 866
 867         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 868         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 869 }
 870
 871 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 872 {
 873         struct scsi_device *sdp = cmd->device;
 874         struct request *rq = scsi_cmd_to_rq(cmd);
 875         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 876         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 877         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 878         unsigned int data_len = 24;
 879         char *buf;
 880
 881         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 882         if (!rq->special_vec.bv_page)
 883                 return BLK_STS_RESOURCE;
 884         clear_highpage(rq->special_vec.bv_page);
 885         rq->special_vec.bv_offset = 0;
 886         rq->special_vec.bv_len = data_len;
 887         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 888
 889         cmd->cmd_len = 10;
 890         cmd->cmnd[0] = UNMAP;
 891         cmd->cmnd[8] = 24;
 892
 893         buf = bvec_virt(&rq->special_vec);
 894         put_unaligned_be16(6 + 16, &buf[0]);
 895         put_unaligned_be16(16, &buf[2]);
 896         put_unaligned_be64(lba, &buf[8]);
 897         put_unaligned_be32(nr_blocks, &buf[16]);
 898
 899         cmd->allowed = sdkp->max_retries;
 900         cmd->transfersize = data_len;
 901         rq->timeout = SD_TIMEOUT;
 902
 903         return scsi_alloc_sgtables(cmd);
 904 }
 905
 906 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 907                 bool unmap)
 908 {
 909         struct scsi_device *sdp = cmd->device;
 910         struct request *rq = scsi_cmd_to_rq(cmd);
 911         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 912         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 913         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 914         u32 data_len = sdp->sector_size;
 915
 916         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 917         if (!rq->special_vec.bv_page)
 918                 return BLK_STS_RESOURCE;
 919         clear_highpage(rq->special_vec.bv_page);
 920         rq->special_vec.bv_offset = 0;
 921         rq->special_vec.bv_len = data_len;
 922         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 923
 924         cmd->cmd_len = 16;
 925         cmd->cmnd[0] = WRITE_SAME_16;
 926         if (unmap)
 927                 cmd->cmnd[1] = 0x8; /* UNMAP */
 928         put_unaligned_be64(lba, &cmd->cmnd[2]);
 929         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 930
 931         cmd->allowed = sdkp->max_retries;
 932         cmd->transfersize = data_len;
 933         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 934
 935         return scsi_alloc_sgtables(cmd);
 936 }
 937
 938 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 939                 bool unmap)
 940 {
 941         struct scsi_device *sdp = cmd->device;
 942         struct request *rq = scsi_cmd_to_rq(cmd);
 943         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 944         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 945         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 946         u32 data_len = sdp->sector_size;
 947
 948         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 949         if (!rq->special_vec.bv_page)
 950                 return BLK_STS_RESOURCE;
 951         clear_highpage(rq->special_vec.bv_page);
 952         rq->special_vec.bv_offset = 0;
 953         rq->special_vec.bv_len = data_len;
 954         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 955
 956         cmd->cmd_len = 10;
 957         cmd->cmnd[0] = WRITE_SAME;
 958         if (unmap)
 959                 cmd->cmnd[1] = 0x8; /* UNMAP */
 960         put_unaligned_be32(lba, &cmd->cmnd[2]);
 961         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 962
 963         cmd->allowed = sdkp->max_retries;
 964         cmd->transfersize = data_len;
 965         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 966
 967         return scsi_alloc_sgtables(cmd);
 968 }
 969
 970 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 971 {
 972         struct request *rq = scsi_cmd_to_rq(cmd);
 973         struct scsi_device *sdp = cmd->device;
 974         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 975         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 976         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 977
 978         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 979                 switch (sdkp->zeroing_mode) {
 980                 case SD_ZERO_WS16_UNMAP:
 981                         return sd_setup_write_same16_cmnd(cmd, true);
 982                 case SD_ZERO_WS10_UNMAP:
 983                         return sd_setup_write_same10_cmnd(cmd, true);
 984                 }
 985         }
 986
 987         if (sdp->no_write_same) {
 988                 rq->rq_flags |= RQF_QUIET;
 989                 return BLK_STS_TARGET;
 990         }
 991
 992         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 993                 return sd_setup_write_same16_cmnd(cmd, false);
 994
 995         return sd_setup_write_same10_cmnd(cmd, false);
 996 }
 997
 998 static void sd_config_write_same(struct scsi_disk *sdkp)
 999 {
1000         struct request_queue *q = sdkp->disk->queue;
1001         unsigned int logical_block_size = sdkp->device->sector_size;
1002
1003         if (sdkp->device->no_write_same) {
1004                 sdkp->max_ws_blocks = 0;
1005                 goto out;
1006         }
1007
1008         /* Some devices can not handle block counts above 0xffff despite
1009          * supporting WRITE SAME(16). Consequently we default to 64k
1010          * blocks per I/O unless the device explicitly advertises a
1011          * bigger limit.
1012          */
1013         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1014                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1015                                                    (u32)SD_MAX_WS16_BLOCKS);
1016         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1017                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1018                                                    (u32)SD_MAX_WS10_BLOCKS);
1019         else {
1020                 sdkp->device->no_write_same = 1;
1021                 sdkp->max_ws_blocks = 0;
1022         }
1023
1024         if (sdkp->lbprz && sdkp->lbpws)
1025                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1026         else if (sdkp->lbprz && sdkp->lbpws10)
1027                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1028         else if (sdkp->max_ws_blocks)
1029                 sdkp->zeroing_mode = SD_ZERO_WS;
1030         else
1031                 sdkp->zeroing_mode = SD_ZERO_WRITE;
1032
1033         if (sdkp->max_ws_blocks &&
1034             sdkp->physical_block_size > logical_block_size) {
1035                 /*
1036                  * Reporting a maximum number of blocks that is not aligned
1037                  * on the device physical size would cause a large write same
1038                  * request to be split into physically unaligned chunks by
1039                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1040                  * even if the caller of these functions took care to align the
1041                  * large request. So make sure the maximum reported is aligned
1042                  * to the device physical block size. This is only an optional
1043                  * optimization for regular disks, but this is mandatory to
1044                  * avoid failure of large write same requests directed at
1045                  * sequential write required zones of host-managed ZBC disks.
1046                  */
1047                 sdkp->max_ws_blocks =
1048                         round_down(sdkp->max_ws_blocks,
1049                                    bytes_to_logical(sdkp->device,
1050                                                     sdkp->physical_block_size));
1051         }
1052
1053 out:
1054         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1055                                          (logical_block_size >> 9));
1056         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1057                                          (logical_block_size >> 9));
1058 }
1059
1060 /**
1061  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1062  * @cmd: command to prepare
1063  *
1064  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1065  * the preference indicated by the target device.
1066  **/
1067 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1068 {
1069         struct request *rq = scsi_cmd_to_rq(cmd);
1070         struct scsi_device *sdp = cmd->device;
1071         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1072         struct bio *bio = rq->bio;
1073         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1074         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1075         blk_status_t ret;
1076
1077         if (sdkp->device->no_write_same)
1078                 return BLK_STS_TARGET;
1079
1080         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1081
1082         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1083
1084         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1085                 cmd->cmd_len = 16;
1086                 cmd->cmnd[0] = WRITE_SAME_16;
1087                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1088                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1089         } else {
1090                 cmd->cmd_len = 10;
1091                 cmd->cmnd[0] = WRITE_SAME;
1092                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1093                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1094         }
1095
1096         cmd->transfersize = sdp->sector_size;
1097         cmd->allowed = sdkp->max_retries;
1098
1099         /*
1100          * For WRITE SAME the data transferred via the DATA OUT buffer is
1101          * different from the amount of data actually written to the target.
1102          *
1103          * We set up __data_len to the amount of data transferred via the
1104          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1105          * to transfer a single sector of data first, but then reset it to
1106          * the amount of data to be written right after so that the I/O path
1107          * knows how much to actually write.
1108          */
1109         rq->__data_len = sdp->sector_size;
1110         ret = scsi_alloc_sgtables(cmd);
1111         rq->__data_len = blk_rq_bytes(rq);
1112
1113         return ret;
1114 }
1115
1116 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1117 {
1118         struct request *rq = scsi_cmd_to_rq(cmd);
1119         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1120
1121         /* flush requests don't perform I/O, zero the S/G table */
1122         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1123
1124         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1125         cmd->cmd_len = 10;
1126         cmd->transfersize = 0;
1127         cmd->allowed = sdkp->max_retries;
1128
1129         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1130         return BLK_STS_OK;
1131 }
1132
1133 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1134                                        sector_t lba, unsigned int nr_blocks,
1135                                        unsigned char flags)
1136 {
1137         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1138         if (unlikely(cmd->cmnd == NULL))
1139                 return BLK_STS_RESOURCE;
1140
1141         cmd->cmd_len = SD_EXT_CDB_SIZE;
1142         memset(cmd->cmnd, 0, cmd->cmd_len);
1143
1144         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1145         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1146         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1147         cmd->cmnd[10] = flags;
1148         put_unaligned_be64(lba, &cmd->cmnd[12]);
1149         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1150         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1151
1152         return BLK_STS_OK;
1153 }
1154
1155 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1156                                        sector_t lba, unsigned int nr_blocks,
1157                                        unsigned char flags)
1158 {
1159         cmd->cmd_len  = 16;
1160         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1161         cmd->cmnd[1]  = flags;
1162         cmd->cmnd[14] = 0;
1163         cmd->cmnd[15] = 0;
1164         put_unaligned_be64(lba, &cmd->cmnd[2]);
1165         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1166
1167         return BLK_STS_OK;
1168 }
1169
1170 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1171                                        sector_t lba, unsigned int nr_blocks,
1172                                        unsigned char flags)
1173 {
1174         cmd->cmd_len = 10;
1175         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1176         cmd->cmnd[1] = flags;
1177         cmd->cmnd[6] = 0;
1178         cmd->cmnd[9] = 0;
1179         put_unaligned_be32(lba, &cmd->cmnd[2]);
1180         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1181
1182         return BLK_STS_OK;
1183 }
1184
1185 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1186                                       sector_t lba, unsigned int nr_blocks,
1187                                       unsigned char flags)
1188 {
1189         /* Avoid that 0 blocks gets translated into 256 blocks. */
1190         if (WARN_ON_ONCE(nr_blocks == 0))
1191                 return BLK_STS_IOERR;
1192
1193         if (unlikely(flags & 0x8)) {
1194                 /*
1195                  * This happens only if this drive failed 10byte rw
1196                  * command with ILLEGAL_REQUEST during operation and
1197                  * thus turned off use_10_for_rw.
1198                  */
1199                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1200                 return BLK_STS_IOERR;
1201         }
1202
1203         cmd->cmd_len = 6;
1204         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1205         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1206         cmd->cmnd[2] = (lba >> 8) & 0xff;
1207         cmd->cmnd[3] = lba & 0xff;
1208         cmd->cmnd[4] = nr_blocks;
1209         cmd->cmnd[5] = 0;
1210
1211         return BLK_STS_OK;
1212 }
1213
1214 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1215 {
1216         struct request *rq = scsi_cmd_to_rq(cmd);
1217         struct scsi_device *sdp = cmd->device;
1218         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1219         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1220         sector_t threshold;
1221         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1222         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1223         bool write = rq_data_dir(rq) == WRITE;
1224         unsigned char protect, fua;
1225         blk_status_t ret;
1226         unsigned int dif;
1227         bool dix;
1228
1229         ret = scsi_alloc_sgtables(cmd);
1230         if (ret != BLK_STS_OK)
1231                 return ret;
1232
1233         ret = BLK_STS_IOERR;
1234         if (!scsi_device_online(sdp) || sdp->changed) {
1235                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1236                 goto fail;
1237         }
1238
1239         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1240                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1241                 goto fail;
1242         }
1243
1244         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1245                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1246                 goto fail;
1247         }
1248
1249         /*
1250          * Some SD card readers can't handle accesses which touch the
1251          * last one or two logical blocks. Split accesses as needed.
1252          */
1253         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1254
1255         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1256                 if (lba < threshold) {
1257                         /* Access up to the threshold but not beyond */
1258                         nr_blocks = threshold - lba;
1259                 } else {
1260                         /* Access only a single logical block */
1261                         nr_blocks = 1;
1262                 }
1263         }
1264
1265         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1266                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1267                 if (ret)
1268                         goto fail;
1269         }
1270
1271         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1272         dix = scsi_prot_sg_count(cmd);
1273         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1274
1275         if (dif || dix)
1276                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1277         else
1278                 protect = 0;
1279
1280         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1281                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1282                                          protect | fua);
1283         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1284                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1285                                          protect | fua);
1286         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1287                    sdp->use_10_for_rw || protect) {
1288                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1289                                          protect | fua);
1290         } else {
1291                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1292                                         protect | fua);
1293         }
1294
1295         if (unlikely(ret != BLK_STS_OK))
1296                 goto fail;
1297
1298         /*
1299          * We shouldn't disconnect in the middle of a sector, so with a dumb
1300          * host adapter, it's safe to assume that we can at least transfer
1301          * this many bytes between each connect / disconnect.
1302          */
1303         cmd->transfersize = sdp->sector_size;
1304         cmd->underflow = nr_blocks << 9;
1305         cmd->allowed = sdkp->max_retries;
1306         cmd->sdb.length = nr_blocks * sdp->sector_size;
1307
1308         SCSI_LOG_HLQUEUE(1,
1309                          scmd_printk(KERN_INFO, cmd,
1310                                      "%s: block=%llu, count=%d\n", __func__,
1311                                      (unsigned long long)blk_rq_pos(rq),
1312                                      blk_rq_sectors(rq)));
1313         SCSI_LOG_HLQUEUE(2,
1314                          scmd_printk(KERN_INFO, cmd,
1315                                      "%s %d/%u 512 byte blocks.\n",
1316                                      write ? "writing" : "reading", nr_blocks,
1317                                      blk_rq_sectors(rq)));
1318
1319         /*
1320          * This indicates that the command is ready from our end to be queued.
1321          */
1322         return BLK_STS_OK;
1323 fail:
1324         scsi_free_sgtables(cmd);
1325         return ret;
1326 }
1327
1328 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1329 {
1330         struct request *rq = scsi_cmd_to_rq(cmd);
1331
1332         switch (req_op(rq)) {
1333         case REQ_OP_DISCARD:
1334                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1335                 case SD_LBP_UNMAP:
1336                         return sd_setup_unmap_cmnd(cmd);
1337                 case SD_LBP_WS16:
1338                         return sd_setup_write_same16_cmnd(cmd, true);
1339                 case SD_LBP_WS10:
1340                         return sd_setup_write_same10_cmnd(cmd, true);
1341                 case SD_LBP_ZERO:
1342                         return sd_setup_write_same10_cmnd(cmd, false);
1343                 default:
1344                         return BLK_STS_TARGET;
1345                 }
1346         case REQ_OP_WRITE_ZEROES:
1347                 return sd_setup_write_zeroes_cmnd(cmd);
1348         case REQ_OP_WRITE_SAME:
1349                 return sd_setup_write_same_cmnd(cmd);
1350         case REQ_OP_FLUSH:
1351                 return sd_setup_flush_cmnd(cmd);
1352         case REQ_OP_READ:
1353         case REQ_OP_WRITE:
1354         case REQ_OP_ZONE_APPEND:
1355                 return sd_setup_read_write_cmnd(cmd);
1356         case REQ_OP_ZONE_RESET:
1357                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358                                                    false);
1359         case REQ_OP_ZONE_RESET_ALL:
1360                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1361                                                    true);
1362         case REQ_OP_ZONE_OPEN:
1363                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1364         case REQ_OP_ZONE_CLOSE:
1365                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1366         case REQ_OP_ZONE_FINISH:
1367                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1368         default:
1369                 WARN_ON_ONCE(1);
1370                 return BLK_STS_NOTSUPP;
1371         }
1372 }
1373
1374 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1375 {
1376         struct request *rq = scsi_cmd_to_rq(SCpnt);
1377         u8 *cmnd;
1378
1379         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1380                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1381
1382         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1383                 cmnd = SCpnt->cmnd;
1384                 SCpnt->cmnd = NULL;
1385                 SCpnt->cmd_len = 0;
1386                 mempool_free(cmnd, sd_cdb_pool);
1387         }
1388 }
1389
1390 static bool sd_need_revalidate(struct block_device *bdev,
1391                 struct scsi_disk *sdkp)
1392 {
1393         if (sdkp->device->removable || sdkp->write_prot) {
1394                 if (bdev_check_media_change(bdev))
1395                         return true;
1396         }
1397
1398         /*
1399          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1400          * nothing to do with partitions, BLKRRPART is used to force a full
1401          * revalidate after things like a format for historical reasons.
1402          */
1403         return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1404 }
1405
1406 /**
1407  *      sd_open - open a scsi disk device
1408  *      @bdev: Block device of the scsi disk to open
1409  *      @mode: FMODE_* mask
1410  *
1411  *      Returns 0 if successful. Returns a negated errno value in case
1412  *      of error.
1413  *
1414  *      Note: This can be called from a user context (e.g. fsck(1) )
1415  *      or from within the kernel (e.g. as a result of a mount(1) ).
1416  *      In the latter case @inode and @filp carry an abridged amount
1417  *      of information as noted above.
1418  *
1419  *      Locking: called with bdev->bd_disk->open_mutex held.
1420  **/
1421 static int sd_open(struct block_device *bdev, fmode_t mode)
1422 {
1423         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1424         struct scsi_device *sdev;
1425         int retval;
1426
1427         if (!sdkp)
1428                 return -ENXIO;
1429
1430         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1431
1432         sdev = sdkp->device;
1433
1434         /*
1435          * If the device is in error recovery, wait until it is done.
1436          * If the device is offline, then disallow any access to it.
1437          */
1438         retval = -ENXIO;
1439         if (!scsi_block_when_processing_errors(sdev))
1440                 goto error_out;
1441
1442         if (sd_need_revalidate(bdev, sdkp))
1443                 sd_revalidate_disk(bdev->bd_disk);
1444
1445         /*
1446          * If the drive is empty, just let the open fail.
1447          */
1448         retval = -ENOMEDIUM;
1449         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1450                 goto error_out;
1451
1452         /*
1453          * If the device has the write protect tab set, have the open fail
1454          * if the user expects to be able to write to the thing.
1455          */
1456         retval = -EROFS;
1457         if (sdkp->write_prot && (mode & FMODE_WRITE))
1458                 goto error_out;
1459
1460         /*
1461          * It is possible that the disk changing stuff resulted in
1462          * the device being taken offline.  If this is the case,
1463          * report this to the user, and don't pretend that the
1464          * open actually succeeded.
1465          */
1466         retval = -ENXIO;
1467         if (!scsi_device_online(sdev))
1468                 goto error_out;
1469
1470         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1471                 if (scsi_block_when_processing_errors(sdev))
1472                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1473         }
1474
1475         return 0;
1476
1477 error_out:
1478         scsi_disk_put(sdkp);
1479         return retval;
1480 }
1481
1482 /**
1483  *      sd_release - invoked when the (last) close(2) is called on this
1484  *      scsi disk.
1485  *      @disk: disk to release
1486  *      @mode: FMODE_* mask
1487  *
1488  *      Returns 0.
1489  *
1490  *      Note: may block (uninterruptible) if error recovery is underway
1491  *      on this disk.
1492  *
1493  *      Locking: called with bdev->bd_disk->open_mutex held.
1494  **/
1495 static void sd_release(struct gendisk *disk, fmode_t mode)
1496 {
1497         struct scsi_disk *sdkp = scsi_disk(disk);
1498         struct scsi_device *sdev = sdkp->device;
1499
1500         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1501
1502         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1503                 if (scsi_block_when_processing_errors(sdev))
1504                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1505         }
1506
1507         scsi_disk_put(sdkp);
1508 }
1509
1510 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1511 {
1512         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1513         struct scsi_device *sdp = sdkp->device;
1514         struct Scsi_Host *host = sdp->host;
1515         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1516         int diskinfo[4];
1517
1518         /* default to most commonly used values */
1519         diskinfo[0] = 0x40;     /* 1 << 6 */
1520         diskinfo[1] = 0x20;     /* 1 << 5 */
1521         diskinfo[2] = capacity >> 11;
1522
1523         /* override with calculated, extended default, or driver values */
1524         if (host->hostt->bios_param)
1525                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1526         else
1527                 scsicam_bios_param(bdev, capacity, diskinfo);
1528
1529         geo->heads = diskinfo[0];
1530         geo->sectors = diskinfo[1];
1531         geo->cylinders = diskinfo[2];
1532         return 0;
1533 }
1534
1535 /**
1536  *      sd_ioctl - process an ioctl
1537  *      @bdev: target block device
1538  *      @mode: FMODE_* mask
1539  *      @cmd: ioctl command number
1540  *      @arg: this is third argument given to ioctl(2) system call.
1541  *      Often contains a pointer.
1542  *
1543  *      Returns 0 if successful (some ioctls return positive numbers on
1544  *      success as well). Returns a negated errno value in case of error.
1545  *
1546  *      Note: most ioctls are forward onto the block subsystem or further
1547  *      down in the scsi subsystem.
1548  **/
1549 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1550                     unsigned int cmd, unsigned long arg)
1551 {
1552         struct gendisk *disk = bdev->bd_disk;
1553         struct scsi_disk *sdkp = scsi_disk(disk);
1554         struct scsi_device *sdp = sdkp->device;
1555         void __user *p = (void __user *)arg;
1556         int error;
1557
1558         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1559                                     "cmd=0x%x\n", disk->disk_name, cmd));
1560
1561         if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1562                 return -ENOIOCTLCMD;
1563
1564         /*
1565          * If we are in the middle of error recovery, don't let anyone
1566          * else try and use this device.  Also, if error recovery fails, it
1567          * may try and take the device offline, in which case all further
1568          * access to the device is prohibited.
1569          */
1570         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1571                         (mode & FMODE_NDELAY) != 0);
1572         if (error)
1573                 return error;
1574
1575         if (is_sed_ioctl(cmd))
1576                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1577         return scsi_ioctl(sdp, disk, mode, cmd, p);
1578 }
1579
1580 static void set_media_not_present(struct scsi_disk *sdkp)
1581 {
1582         if (sdkp->media_present)
1583                 sdkp->device->changed = 1;
1584
1585         if (sdkp->device->removable) {
1586                 sdkp->media_present = 0;
1587                 sdkp->capacity = 0;
1588         }
1589 }
1590
1591 static int media_not_present(struct scsi_disk *sdkp,
1592                              struct scsi_sense_hdr *sshdr)
1593 {
1594         if (!scsi_sense_valid(sshdr))
1595                 return 0;
1596
1597         /* not invoked for commands that could return deferred errors */
1598         switch (sshdr->sense_key) {
1599         case UNIT_ATTENTION:
1600         case NOT_READY:
1601                 /* medium not present */
1602                 if (sshdr->asc == 0x3A) {
1603                         set_media_not_present(sdkp);
1604                         return 1;
1605                 }
1606         }
1607         return 0;
1608 }
1609
1610 /**
1611  *      sd_check_events - check media events
1612  *      @disk: kernel device descriptor
1613  *      @clearing: disk events currently being cleared
1614  *
1615  *      Returns mask of DISK_EVENT_*.
1616  *
1617  *      Note: this function is invoked from the block subsystem.
1618  **/
1619 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1620 {
1621         struct scsi_disk *sdkp = scsi_disk_get(disk);
1622         struct scsi_device *sdp;
1623         int retval;
1624         bool disk_changed;
1625
1626         if (!sdkp)
1627                 return 0;
1628
1629         sdp = sdkp->device;
1630         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1631
1632         /*
1633          * If the device is offline, don't send any commands - just pretend as
1634          * if the command failed.  If the device ever comes back online, we
1635          * can deal with it then.  It is only because of unrecoverable errors
1636          * that we would ever take a device offline in the first place.
1637          */
1638         if (!scsi_device_online(sdp)) {
1639                 set_media_not_present(sdkp);
1640                 goto out;
1641         }
1642
1643         /*
1644          * Using TEST_UNIT_READY enables differentiation between drive with
1645          * no cartridge loaded - NOT READY, drive with changed cartridge -
1646          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1647          *
1648          * Drives that auto spin down. eg iomega jaz 1G, will be started
1649          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1650          * sd_revalidate() is called.
1651          */
1652         if (scsi_block_when_processing_errors(sdp)) {
1653                 struct scsi_sense_hdr sshdr = { 0, };
1654
1655                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1656                                               &sshdr);
1657
1658                 /* failed to execute TUR, assume media not present */
1659                 if (retval < 0 || host_byte(retval)) {
1660                         set_media_not_present(sdkp);
1661                         goto out;
1662                 }
1663
1664                 if (media_not_present(sdkp, &sshdr))
1665                         goto out;
1666         }
1667
1668         /*
1669          * For removable scsi disk we have to recognise the presence
1670          * of a disk in the drive.
1671          */
1672         if (!sdkp->media_present)
1673                 sdp->changed = 1;
1674         sdkp->media_present = 1;
1675 out:
1676         /*
1677          * sdp->changed is set under the following conditions:
1678          *
1679          *      Medium present state has changed in either direction.
1680          *      Device has indicated UNIT_ATTENTION.
1681          */
1682         disk_changed = sdp->changed;
1683         sdp->changed = 0;
1684         scsi_disk_put(sdkp);
1685         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1686 }
1687
1688 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1689 {
1690         int retries, res;
1691         struct scsi_device *sdp = sdkp->device;
1692         const int timeout = sdp->request_queue->rq_timeout
1693                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1694         struct scsi_sense_hdr my_sshdr;
1695
1696         if (!scsi_device_online(sdp))
1697                 return -ENODEV;
1698
1699         /* caller might not be interested in sense, but we need it */
1700         if (!sshdr)
1701                 sshdr = &my_sshdr;
1702
1703         for (retries = 3; retries > 0; --retries) {
1704                 unsigned char cmd[10] = { 0 };
1705
1706                 cmd[0] = SYNCHRONIZE_CACHE;
1707                 /*
1708                  * Leave the rest of the command zero to indicate
1709                  * flush everything.
1710                  */
1711                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1712                                 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1713                 if (res == 0)
1714                         break;
1715         }
1716
1717         if (res) {
1718                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1719
1720                 if (res < 0)
1721                         return res;
1722
1723                 if (scsi_status_is_check_condition(res) &&
1724                     scsi_sense_valid(sshdr)) {
1725                         sd_print_sense_hdr(sdkp, sshdr);
1726
1727                         /* we need to evaluate the error return  */
1728                         if (sshdr->asc == 0x3a ||       /* medium not present */
1729                             sshdr->asc == 0x20 ||       /* invalid command */
1730                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1731                                 /* this is no error here */
1732                                 return 0;
1733                 }
1734
1735                 switch (host_byte(res)) {
1736                 /* ignore errors due to racing a disconnection */
1737                 case DID_BAD_TARGET:
1738                 case DID_NO_CONNECT:
1739                         return 0;
1740                 /* signal the upper layer it might try again */
1741                 case DID_BUS_BUSY:
1742                 case DID_IMM_RETRY:
1743                 case DID_REQUEUE:
1744                 case DID_SOFT_ERROR:
1745                         return -EBUSY;
1746                 default:
1747                         return -EIO;
1748                 }
1749         }
1750         return 0;
1751 }
1752
1753 static void sd_rescan(struct device *dev)
1754 {
1755         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1756
1757         sd_revalidate_disk(sdkp->disk);
1758 }
1759
1760 static char sd_pr_type(enum pr_type type)
1761 {
1762         switch (type) {
1763         case PR_WRITE_EXCLUSIVE:
1764                 return 0x01;
1765         case PR_EXCLUSIVE_ACCESS:
1766                 return 0x03;
1767         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1768                 return 0x05;
1769         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1770                 return 0x06;
1771         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1772                 return 0x07;
1773         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1774                 return 0x08;
1775         default:
1776                 return 0;
1777         }
1778 };
1779
1780 static int sd_pr_command(struct block_device *bdev, u8 sa,
1781                 u64 key, u64 sa_key, u8 type, u8 flags)
1782 {
1783         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1784         struct scsi_device *sdev = sdkp->device;
1785         struct scsi_sense_hdr sshdr;
1786         int result;
1787         u8 cmd[16] = { 0, };
1788         u8 data[24] = { 0, };
1789
1790         cmd[0] = PERSISTENT_RESERVE_OUT;
1791         cmd[1] = sa;
1792         cmd[2] = type;
1793         put_unaligned_be32(sizeof(data), &cmd[5]);
1794
1795         put_unaligned_be64(key, &data[0]);
1796         put_unaligned_be64(sa_key, &data[8]);
1797         data[20] = flags;
1798
1799         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1800                         &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1801
1802         if (scsi_status_is_check_condition(result) &&
1803             scsi_sense_valid(&sshdr)) {
1804                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1805                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1806         }
1807
1808         return result;
1809 }
1810
1811 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1812                 u32 flags)
1813 {
1814         if (flags & ~PR_FL_IGNORE_KEY)
1815                 return -EOPNOTSUPP;
1816         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1817                         old_key, new_key, 0,
1818                         (1 << 0) /* APTPL */);
1819 }
1820
1821 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1822                 u32 flags)
1823 {
1824         if (flags)
1825                 return -EOPNOTSUPP;
1826         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1827 }
1828
1829 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1830 {
1831         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1832 }
1833
1834 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1835                 enum pr_type type, bool abort)
1836 {
1837         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1838                              sd_pr_type(type), 0);
1839 }
1840
1841 static int sd_pr_clear(struct block_device *bdev, u64 key)
1842 {
1843         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1844 }
1845
1846 static const struct pr_ops sd_pr_ops = {
1847         .pr_register    = sd_pr_register,
1848         .pr_reserve     = sd_pr_reserve,
1849         .pr_release     = sd_pr_release,
1850         .pr_preempt     = sd_pr_preempt,
1851         .pr_clear       = sd_pr_clear,
1852 };
1853
1854 static const struct block_device_operations sd_fops = {
1855         .owner                  = THIS_MODULE,
1856         .open                   = sd_open,
1857         .release                = sd_release,
1858         .ioctl                  = sd_ioctl,
1859         .getgeo                 = sd_getgeo,
1860         .compat_ioctl           = blkdev_compat_ptr_ioctl,
1861         .check_events           = sd_check_events,
1862         .unlock_native_capacity = sd_unlock_native_capacity,
1863         .report_zones           = sd_zbc_report_zones,
1864         .pr_ops                 = &sd_pr_ops,
1865 };
1866
1867 /**
1868  *      sd_eh_reset - reset error handling callback
1869  *      @scmd:          sd-issued command that has failed
1870  *
1871  *      This function is called by the SCSI midlayer before starting
1872  *      SCSI EH. When counting medium access failures we have to be
1873  *      careful to register it only only once per device and SCSI EH run;
1874  *      there might be several timed out commands which will cause the
1875  *      'max_medium_access_timeouts' counter to trigger after the first
1876  *      SCSI EH run already and set the device to offline.
1877  *      So this function resets the internal counter before starting SCSI EH.
1878  **/
1879 static void sd_eh_reset(struct scsi_cmnd *scmd)
1880 {
1881         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1882
1883         /* New SCSI EH run, reset gate variable */
1884         sdkp->ignore_medium_access_errors = false;
1885 }
1886
1887 /**
1888  *      sd_eh_action - error handling callback
1889  *      @scmd:          sd-issued command that has failed
1890  *      @eh_disp:       The recovery disposition suggested by the midlayer
1891  *
1892  *      This function is called by the SCSI midlayer upon completion of an
1893  *      error test command (currently TEST UNIT READY). The result of sending
1894  *      the eh command is passed in eh_disp.  We're looking for devices that
1895  *      fail medium access commands but are OK with non access commands like
1896  *      test unit ready (so wrongly see the device as having a successful
1897  *      recovery)
1898  **/
1899 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1900 {
1901         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1902         struct scsi_device *sdev = scmd->device;
1903
1904         if (!scsi_device_online(sdev) ||
1905             !scsi_medium_access_command(scmd) ||
1906             host_byte(scmd->result) != DID_TIME_OUT ||
1907             eh_disp != SUCCESS)
1908                 return eh_disp;
1909
1910         /*
1911          * The device has timed out executing a medium access command.
1912          * However, the TEST UNIT READY command sent during error
1913          * handling completed successfully. Either the device is in the
1914          * process of recovering or has it suffered an internal failure
1915          * that prevents access to the storage medium.
1916          */
1917         if (!sdkp->ignore_medium_access_errors) {
1918                 sdkp->medium_access_timed_out++;
1919                 sdkp->ignore_medium_access_errors = true;
1920         }
1921
1922         /*
1923          * If the device keeps failing read/write commands but TEST UNIT
1924          * READY always completes successfully we assume that medium
1925          * access is no longer possible and take the device offline.
1926          */
1927         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1928                 scmd_printk(KERN_ERR, scmd,
1929                             "Medium access timeout failure. Offlining disk!\n");
1930                 mutex_lock(&sdev->state_mutex);
1931                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1932                 mutex_unlock(&sdev->state_mutex);
1933
1934                 return SUCCESS;
1935         }
1936
1937         return eh_disp;
1938 }
1939
1940 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1941 {
1942         struct request *req = scsi_cmd_to_rq(scmd);
1943         struct scsi_device *sdev = scmd->device;
1944         unsigned int transferred, good_bytes;
1945         u64 start_lba, end_lba, bad_lba;
1946
1947         /*
1948          * Some commands have a payload smaller than the device logical
1949          * block size (e.g. INQUIRY on a 4K disk).
1950          */
1951         if (scsi_bufflen(scmd) <= sdev->sector_size)
1952                 return 0;
1953
1954         /* Check if we have a 'bad_lba' information */
1955         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1956                                      SCSI_SENSE_BUFFERSIZE,
1957                                      &bad_lba))
1958                 return 0;
1959
1960         /*
1961          * If the bad lba was reported incorrectly, we have no idea where
1962          * the error is.
1963          */
1964         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1965         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1966         if (bad_lba < start_lba || bad_lba >= end_lba)
1967                 return 0;
1968
1969         /*
1970          * resid is optional but mostly filled in.  When it's unused,
1971          * its value is zero, so we assume the whole buffer transferred
1972          */
1973         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1974
1975         /* This computation should always be done in terms of the
1976          * resolution of the device's medium.
1977          */
1978         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1979
1980         return min(good_bytes, transferred);
1981 }
1982
1983 /**
1984  *      sd_done - bottom half handler: called when the lower level
1985  *      driver has completed (successfully or otherwise) a scsi command.
1986  *      @SCpnt: mid-level's per command structure.
1987  *
1988  *      Note: potentially run from within an ISR. Must not block.
1989  **/
1990 static int sd_done(struct scsi_cmnd *SCpnt)
1991 {
1992         int result = SCpnt->result;
1993         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1994         unsigned int sector_size = SCpnt->device->sector_size;
1995         unsigned int resid;
1996         struct scsi_sense_hdr sshdr;
1997         struct request *req = scsi_cmd_to_rq(SCpnt);
1998         struct scsi_disk *sdkp = scsi_disk(req->rq_disk);
1999         int sense_valid = 0;
2000         int sense_deferred = 0;
2001
2002         switch (req_op(req)) {
2003         case REQ_OP_DISCARD:
2004         case REQ_OP_WRITE_ZEROES:
2005         case REQ_OP_WRITE_SAME:
2006         case REQ_OP_ZONE_RESET:
2007         case REQ_OP_ZONE_RESET_ALL:
2008         case REQ_OP_ZONE_OPEN:
2009         case REQ_OP_ZONE_CLOSE:
2010         case REQ_OP_ZONE_FINISH:
2011                 if (!result) {
2012                         good_bytes = blk_rq_bytes(req);
2013                         scsi_set_resid(SCpnt, 0);
2014                 } else {
2015                         good_bytes = 0;
2016                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
2017                 }
2018                 break;
2019         default:
2020                 /*
2021                  * In case of bogus fw or device, we could end up having
2022                  * an unaligned partial completion. Check this here and force
2023                  * alignment.
2024                  */
2025                 resid = scsi_get_resid(SCpnt);
2026                 if (resid & (sector_size - 1)) {
2027                         sd_printk(KERN_INFO, sdkp,
2028                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2029                                 resid, sector_size);
2030                         scsi_print_command(SCpnt);
2031                         resid = min(scsi_bufflen(SCpnt),
2032                                     round_up(resid, sector_size));
2033                         scsi_set_resid(SCpnt, resid);
2034                 }
2035         }
2036
2037         if (result) {
2038                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2039                 if (sense_valid)
2040                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2041         }
2042         sdkp->medium_access_timed_out = 0;
2043
2044         if (!scsi_status_is_check_condition(result) &&
2045             (!sense_valid || sense_deferred))
2046                 goto out;
2047
2048         switch (sshdr.sense_key) {
2049         case HARDWARE_ERROR:
2050         case MEDIUM_ERROR:
2051                 good_bytes = sd_completed_bytes(SCpnt);
2052                 break;
2053         case RECOVERED_ERROR:
2054                 good_bytes = scsi_bufflen(SCpnt);
2055                 break;
2056         case NO_SENSE:
2057                 /* This indicates a false check condition, so ignore it.  An
2058                  * unknown amount of data was transferred so treat it as an
2059                  * error.
2060                  */
2061                 SCpnt->result = 0;
2062                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2063                 break;
2064         case ABORTED_COMMAND:
2065                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2066                         good_bytes = sd_completed_bytes(SCpnt);
2067                 break;
2068         case ILLEGAL_REQUEST:
2069                 switch (sshdr.asc) {
2070                 case 0x10:      /* DIX: Host detected corruption */
2071                         good_bytes = sd_completed_bytes(SCpnt);
2072                         break;
2073                 case 0x20:      /* INVALID COMMAND OPCODE */
2074                 case 0x24:      /* INVALID FIELD IN CDB */
2075                         switch (SCpnt->cmnd[0]) {
2076                         case UNMAP:
2077                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2078                                 break;
2079                         case WRITE_SAME_16:
2080                         case WRITE_SAME:
2081                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2082                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2083                                 } else {
2084                                         sdkp->device->no_write_same = 1;
2085                                         sd_config_write_same(sdkp);
2086                                         req->rq_flags |= RQF_QUIET;
2087                                 }
2088                                 break;
2089                         }
2090                 }
2091                 break;
2092         default:
2093                 break;
2094         }
2095
2096  out:
2097         if (sd_is_zoned(sdkp))
2098                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2099
2100         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2101                                            "sd_done: completed %d of %d bytes\n",
2102                                            good_bytes, scsi_bufflen(SCpnt)));
2103
2104         return good_bytes;
2105 }
2106
2107 /*
2108  * spinup disk - called only in sd_revalidate_disk()
2109  */
2110 static void
2111 sd_spinup_disk(struct scsi_disk *sdkp)
2112 {
2113         unsigned char cmd[10];
2114         unsigned long spintime_expire = 0;
2115         int retries, spintime;
2116         unsigned int the_result;
2117         struct scsi_sense_hdr sshdr;
2118         int sense_valid = 0;
2119
2120         spintime = 0;
2121
2122         /* Spin up drives, as required.  Only do this at boot time */
2123         /* Spinup needs to be done for module loads too. */
2124         do {
2125                 retries = 0;
2126
2127                 do {
2128                         bool media_was_present = sdkp->media_present;
2129
2130                         cmd[0] = TEST_UNIT_READY;
2131                         memset((void *) &cmd[1], 0, 9);
2132
2133                         the_result = scsi_execute_req(sdkp->device, cmd,
2134                                                       DMA_NONE, NULL, 0,
2135                                                       &sshdr, SD_TIMEOUT,
2136                                                       sdkp->max_retries, NULL);
2137
2138                         /*
2139                          * If the drive has indicated to us that it
2140                          * doesn't have any media in it, don't bother
2141                          * with any more polling.
2142                          */
2143                         if (media_not_present(sdkp, &sshdr)) {
2144                                 if (media_was_present)
2145                                         sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2146                                 return;
2147                         }
2148
2149                         if (the_result)
2150                                 sense_valid = scsi_sense_valid(&sshdr);
2151                         retries++;
2152                 } while (retries < 3 &&
2153                          (!scsi_status_is_good(the_result) ||
2154                           (scsi_status_is_check_condition(the_result) &&
2155                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2156
2157                 if (!scsi_status_is_check_condition(the_result)) {
2158                         /* no sense, TUR either succeeded or failed
2159                          * with a status error */
2160                         if(!spintime && !scsi_status_is_good(the_result)) {
2161                                 sd_print_result(sdkp, "Test Unit Ready failed",
2162                                                 the_result);
2163                         }
2164                         break;
2165                 }
2166
2167                 /*
2168                  * The device does not want the automatic start to be issued.
2169                  */
2170                 if (sdkp->device->no_start_on_add)
2171                         break;
2172
2173                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2174                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2175                                 break;  /* manual intervention required */
2176                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2177                                 break;  /* standby */
2178                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2179                                 break;  /* unavailable */
2180                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2181                                 break;  /* sanitize in progress */
2182                         /*
2183                          * Issue command to spin up drive when not ready
2184                          */
2185                         if (!spintime) {
2186                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2187                                 cmd[0] = START_STOP;
2188                                 cmd[1] = 1;     /* Return immediately */
2189                                 memset((void *) &cmd[2], 0, 8);
2190                                 cmd[4] = 1;     /* Start spin cycle */
2191                                 if (sdkp->device->start_stop_pwr_cond)
2192                                         cmd[4] |= 1 << 4;
2193                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2194                                                  NULL, 0, &sshdr,
2195                                                  SD_TIMEOUT, sdkp->max_retries,
2196                                                  NULL);
2197                                 spintime_expire = jiffies + 100 * HZ;
2198                                 spintime = 1;
2199                         }
2200                         /* Wait 1 second for next try */
2201                         msleep(1000);
2202                         printk(KERN_CONT ".");
2203
2204                 /*
2205                  * Wait for USB flash devices with slow firmware.
2206                  * Yes, this sense key/ASC combination shouldn't
2207                  * occur here.  It's characteristic of these devices.
2208                  */
2209                 } else if (sense_valid &&
2210                                 sshdr.sense_key == UNIT_ATTENTION &&
2211                                 sshdr.asc == 0x28) {
2212                         if (!spintime) {
2213                                 spintime_expire = jiffies + 5 * HZ;
2214                                 spintime = 1;
2215                         }
2216                         /* Wait 1 second for next try */
2217                         msleep(1000);
2218                 } else {
2219                         /* we don't understand the sense code, so it's
2220                          * probably pointless to loop */
2221                         if(!spintime) {
2222                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2223                                 sd_print_sense_hdr(sdkp, &sshdr);
2224                         }
2225                         break;
2226                 }
2227
2228         } while (spintime && time_before_eq(jiffies, spintime_expire));
2229
2230         if (spintime) {
2231                 if (scsi_status_is_good(the_result))
2232                         printk(KERN_CONT "ready\n");
2233                 else
2234                         printk(KERN_CONT "not responding...\n");
2235         }
2236 }
2237
2238 /*
2239  * Determine whether disk supports Data Integrity Field.
2240  */
2241 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2242 {
2243         struct scsi_device *sdp = sdkp->device;
2244         u8 type;
2245         int ret = 0;
2246
2247         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2248                 sdkp->protection_type = 0;
2249                 return ret;
2250         }
2251
2252         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2253
2254         if (type > T10_PI_TYPE3_PROTECTION)
2255                 ret = -ENODEV;
2256         else if (scsi_host_dif_capable(sdp->host, type))
2257                 ret = 1;
2258
2259         if (sdkp->first_scan || type != sdkp->protection_type)
2260                 switch (ret) {
2261                 case -ENODEV:
2262                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2263                                   " protection type %u. Disabling disk!\n",
2264                                   type);
2265                         break;
2266                 case 1:
2267                         sd_printk(KERN_NOTICE, sdkp,
2268                                   "Enabling DIF Type %u protection\n", type);
2269                         break;
2270                 case 0:
2271                         sd_printk(KERN_NOTICE, sdkp,
2272                                   "Disabling DIF Type %u protection\n", type);
2273                         break;
2274                 }
2275
2276         sdkp->protection_type = type;
2277
2278         return ret;
2279 }
2280
2281 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2282                         struct scsi_sense_hdr *sshdr, int sense_valid,
2283                         int the_result)
2284 {
2285         if (sense_valid)
2286                 sd_print_sense_hdr(sdkp, sshdr);
2287         else
2288                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2289
2290         /*
2291          * Set dirty bit for removable devices if not ready -
2292          * sometimes drives will not report this properly.
2293          */
2294         if (sdp->removable &&
2295             sense_valid && sshdr->sense_key == NOT_READY)
2296                 set_media_not_present(sdkp);
2297
2298         /*
2299          * We used to set media_present to 0 here to indicate no media
2300          * in the drive, but some drives fail read capacity even with
2301          * media present, so we can't do that.
2302          */
2303         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2304 }
2305
2306 #define RC16_LEN 32
2307 #if RC16_LEN > SD_BUF_SIZE
2308 #error RC16_LEN must not be more than SD_BUF_SIZE
2309 #endif
2310
2311 #define READ_CAPACITY_RETRIES_ON_RESET  10
2312
2313 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2314                                                 unsigned char *buffer)
2315 {
2316         unsigned char cmd[16];
2317         struct scsi_sense_hdr sshdr;
2318         int sense_valid = 0;
2319         int the_result;
2320         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2321         unsigned int alignment;
2322         unsigned long long lba;
2323         unsigned sector_size;
2324
2325         if (sdp->no_read_capacity_16)
2326                 return -EINVAL;
2327
2328         do {
2329                 memset(cmd, 0, 16);
2330                 cmd[0] = SERVICE_ACTION_IN_16;
2331                 cmd[1] = SAI_READ_CAPACITY_16;
2332                 cmd[13] = RC16_LEN;
2333                 memset(buffer, 0, RC16_LEN);
2334
2335                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2336                                         buffer, RC16_LEN, &sshdr,
2337                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2338
2339                 if (media_not_present(sdkp, &sshdr))
2340                         return -ENODEV;
2341
2342                 if (the_result > 0) {
2343                         sense_valid = scsi_sense_valid(&sshdr);
2344                         if (sense_valid &&
2345                             sshdr.sense_key == ILLEGAL_REQUEST &&
2346                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2347                             sshdr.ascq == 0x00)
2348                                 /* Invalid Command Operation Code or
2349                                  * Invalid Field in CDB, just retry
2350                                  * silently with RC10 */
2351                                 return -EINVAL;
2352                         if (sense_valid &&
2353                             sshdr.sense_key == UNIT_ATTENTION &&
2354                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2355                                 /* Device reset might occur several times,
2356                                  * give it one more chance */
2357                                 if (--reset_retries > 0)
2358                                         continue;
2359                 }
2360                 retries--;
2361
2362         } while (the_result && retries);
2363
2364         if (the_result) {
2365                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2366                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2367                 return -EINVAL;
2368         }
2369
2370         sector_size = get_unaligned_be32(&buffer[8]);
2371         lba = get_unaligned_be64(&buffer[0]);
2372
2373         if (sd_read_protection_type(sdkp, buffer) < 0) {
2374                 sdkp->capacity = 0;
2375                 return -ENODEV;
2376         }
2377
2378         /* Logical blocks per physical block exponent */
2379         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2380
2381         /* RC basis */
2382         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2383
2384         /* Lowest aligned logical block */
2385         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2386         blk_queue_alignment_offset(sdp->request_queue, alignment);
2387         if (alignment && sdkp->first_scan)
2388                 sd_printk(KERN_NOTICE, sdkp,
2389                           "physical block alignment offset: %u\n", alignment);
2390
2391         if (buffer[14] & 0x80) { /* LBPME */
2392                 sdkp->lbpme = 1;
2393
2394                 if (buffer[14] & 0x40) /* LBPRZ */
2395                         sdkp->lbprz = 1;
2396
2397                 sd_config_discard(sdkp, SD_LBP_WS16);
2398         }
2399
2400         sdkp->capacity = lba + 1;
2401         return sector_size;
2402 }
2403
2404 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2405                                                 unsigned char *buffer)
2406 {
2407         unsigned char cmd[16];
2408         struct scsi_sense_hdr sshdr;
2409         int sense_valid = 0;
2410         int the_result;
2411         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2412         sector_t lba;
2413         unsigned sector_size;
2414
2415         do {
2416                 cmd[0] = READ_CAPACITY;
2417                 memset(&cmd[1], 0, 9);
2418                 memset(buffer, 0, 8);
2419
2420                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2421                                         buffer, 8, &sshdr,
2422                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2423
2424                 if (media_not_present(sdkp, &sshdr))
2425                         return -ENODEV;
2426
2427                 if (the_result > 0) {
2428                         sense_valid = scsi_sense_valid(&sshdr);
2429                         if (sense_valid &&
2430                             sshdr.sense_key == UNIT_ATTENTION &&
2431                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2432                                 /* Device reset might occur several times,
2433                                  * give it one more chance */
2434                                 if (--reset_retries > 0)
2435                                         continue;
2436                 }
2437                 retries--;
2438
2439         } while (the_result && retries);
2440
2441         if (the_result) {
2442                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2443                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2444                 return -EINVAL;
2445         }
2446
2447         sector_size = get_unaligned_be32(&buffer[4]);
2448         lba = get_unaligned_be32(&buffer[0]);
2449
2450         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2451                 /* Some buggy (usb cardreader) devices return an lba of
2452                    0xffffffff when the want to report a size of 0 (with
2453                    which they really mean no media is present) */
2454                 sdkp->capacity = 0;
2455                 sdkp->physical_block_size = sector_size;
2456                 return sector_size;
2457         }
2458
2459         sdkp->capacity = lba + 1;
2460         sdkp->physical_block_size = sector_size;
2461         return sector_size;
2462 }
2463
2464 static int sd_try_rc16_first(struct scsi_device *sdp)
2465 {
2466         if (sdp->host->max_cmd_len < 16)
2467                 return 0;
2468         if (sdp->try_rc_10_first)
2469                 return 0;
2470         if (sdp->scsi_level > SCSI_SPC_2)
2471                 return 1;
2472         if (scsi_device_protection(sdp))
2473                 return 1;
2474         return 0;
2475 }
2476
2477 /*
2478  * read disk capacity
2479  */
2480 static void
2481 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2482 {
2483         int sector_size;
2484         struct scsi_device *sdp = sdkp->device;
2485
2486         if (sd_try_rc16_first(sdp)) {
2487                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2488                 if (sector_size == -EOVERFLOW)
2489                         goto got_data;
2490                 if (sector_size == -ENODEV)
2491                         return;
2492                 if (sector_size < 0)
2493                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2494                 if (sector_size < 0)
2495                         return;
2496         } else {
2497                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2498                 if (sector_size == -EOVERFLOW)
2499                         goto got_data;
2500                 if (sector_size < 0)
2501                         return;
2502                 if ((sizeof(sdkp->capacity) > 4) &&
2503                     (sdkp->capacity > 0xffffffffULL)) {
2504                         int old_sector_size = sector_size;
2505                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2506                                         "Trying to use READ CAPACITY(16).\n");
2507                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2508                         if (sector_size < 0) {
2509                                 sd_printk(KERN_NOTICE, sdkp,
2510                                         "Using 0xffffffff as device size\n");
2511                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2512                                 sector_size = old_sector_size;
2513                                 goto got_data;
2514                         }
2515                         /* Remember that READ CAPACITY(16) succeeded */
2516                         sdp->try_rc_10_first = 0;
2517                 }
2518         }
2519
2520         /* Some devices are known to return the total number of blocks,
2521          * not the highest block number.  Some devices have versions
2522          * which do this and others which do not.  Some devices we might
2523          * suspect of doing this but we don't know for certain.
2524          *
2525          * If we know the reported capacity is wrong, decrement it.  If
2526          * we can only guess, then assume the number of blocks is even
2527          * (usually true but not always) and err on the side of lowering
2528          * the capacity.
2529          */
2530         if (sdp->fix_capacity ||
2531             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2532                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2533                                 "from its reported value: %llu\n",
2534                                 (unsigned long long) sdkp->capacity);
2535                 --sdkp->capacity;
2536         }
2537
2538 got_data:
2539         if (sector_size == 0) {
2540                 sector_size = 512;
2541                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2542                           "assuming 512.\n");
2543         }
2544
2545         if (sector_size != 512 &&
2546             sector_size != 1024 &&
2547             sector_size != 2048 &&
2548             sector_size != 4096) {
2549                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2550                           sector_size);
2551                 /*
2552                  * The user might want to re-format the drive with
2553                  * a supported sectorsize.  Once this happens, it
2554                  * would be relatively trivial to set the thing up.
2555                  * For this reason, we leave the thing in the table.
2556                  */
2557                 sdkp->capacity = 0;
2558                 /*
2559                  * set a bogus sector size so the normal read/write
2560                  * logic in the block layer will eventually refuse any
2561                  * request on this device without tripping over power
2562                  * of two sector size assumptions
2563                  */
2564                 sector_size = 512;
2565         }
2566         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2567         blk_queue_physical_block_size(sdp->request_queue,
2568                                       sdkp->physical_block_size);
2569         sdkp->device->sector_size = sector_size;
2570
2571         if (sdkp->capacity > 0xffffffff)
2572                 sdp->use_16_for_rw = 1;
2573
2574 }
2575
2576 /*
2577  * Print disk capacity
2578  */
2579 static void
2580 sd_print_capacity(struct scsi_disk *sdkp,
2581                   sector_t old_capacity)
2582 {
2583         int sector_size = sdkp->device->sector_size;
2584         char cap_str_2[10], cap_str_10[10];
2585
2586         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2587                 return;
2588
2589         string_get_size(sdkp->capacity, sector_size,
2590                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2591         string_get_size(sdkp->capacity, sector_size,
2592                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2593
2594         sd_printk(KERN_NOTICE, sdkp,
2595                   "%llu %d-byte logical blocks: (%s/%s)\n",
2596                   (unsigned long long)sdkp->capacity,
2597                   sector_size, cap_str_10, cap_str_2);
2598
2599         if (sdkp->physical_block_size != sector_size)
2600                 sd_printk(KERN_NOTICE, sdkp,
2601                           "%u-byte physical blocks\n",
2602                           sdkp->physical_block_size);
2603 }
2604
2605 /* called with buffer of length 512 */
2606 static inline int
2607 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2608                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2609                  struct scsi_sense_hdr *sshdr)
2610 {
2611         /*
2612          * If we must use MODE SENSE(10), make sure that the buffer length
2613          * is at least 8 bytes so that the mode sense header fits.
2614          */
2615         if (sdkp->device->use_10_for_ms && len < 8)
2616                 len = 8;
2617
2618         return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2619                                SD_TIMEOUT, sdkp->max_retries, data,
2620                                sshdr);
2621 }
2622
2623 /*
2624  * read write protect setting, if possible - called only in sd_revalidate_disk()
2625  * called with buffer of length SD_BUF_SIZE
2626  */
2627 static void
2628 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2629 {
2630         int res;
2631         struct scsi_device *sdp = sdkp->device;
2632         struct scsi_mode_data data;
2633         int old_wp = sdkp->write_prot;
2634
2635         set_disk_ro(sdkp->disk, 0);
2636         if (sdp->skip_ms_page_3f) {
2637                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2638                 return;
2639         }
2640
2641         if (sdp->use_192_bytes_for_3f) {
2642                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2643         } else {
2644                 /*
2645                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2646                  * We have to start carefully: some devices hang if we ask
2647                  * for more than is available.
2648                  */
2649                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2650
2651                 /*
2652                  * Second attempt: ask for page 0 When only page 0 is
2653                  * implemented, a request for page 3F may return Sense Key
2654                  * 5: Illegal Request, Sense Code 24: Invalid field in
2655                  * CDB.
2656                  */
2657                 if (res < 0)
2658                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2659
2660                 /*
2661                  * Third attempt: ask 255 bytes, as we did earlier.
2662                  */
2663                 if (res < 0)
2664                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2665                                                &data, NULL);
2666         }
2667
2668         if (res < 0) {
2669                 sd_first_printk(KERN_WARNING, sdkp,
2670                           "Test WP failed, assume Write Enabled\n");
2671         } else {
2672                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2673                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2674                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2675                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2676                                   sdkp->write_prot ? "on" : "off");
2677                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2678                 }
2679         }
2680 }
2681
2682 /*
2683  * sd_read_cache_type - called only from sd_revalidate_disk()
2684  * called with buffer of length SD_BUF_SIZE
2685  */
2686 static void
2687 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2688 {
2689         int len = 0, res;
2690         struct scsi_device *sdp = sdkp->device;
2691
2692         int dbd;
2693         int modepage;
2694         int first_len;
2695         struct scsi_mode_data data;
2696         struct scsi_sense_hdr sshdr;
2697         int old_wce = sdkp->WCE;
2698         int old_rcd = sdkp->RCD;
2699         int old_dpofua = sdkp->DPOFUA;
2700
2701
2702         if (sdkp->cache_override)
2703                 return;
2704
2705         first_len = 4;
2706         if (sdp->skip_ms_page_8) {
2707                 if (sdp->type == TYPE_RBC)
2708                         goto defaults;
2709                 else {
2710                         if (sdp->skip_ms_page_3f)
2711                                 goto defaults;
2712                         modepage = 0x3F;
2713                         if (sdp->use_192_bytes_for_3f)
2714                                 first_len = 192;
2715                         dbd = 0;
2716                 }
2717         } else if (sdp->type == TYPE_RBC) {
2718                 modepage = 6;
2719                 dbd = 8;
2720         } else {
2721                 modepage = 8;
2722                 dbd = 0;
2723         }
2724
2725         /* cautiously ask */
2726         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2727                         &data, &sshdr);
2728
2729         if (res < 0)
2730                 goto bad_sense;
2731
2732         if (!data.header_length) {
2733                 modepage = 6;
2734                 first_len = 0;
2735                 sd_first_printk(KERN_ERR, sdkp,
2736                                 "Missing header in MODE_SENSE response\n");
2737         }
2738
2739         /* that went OK, now ask for the proper length */
2740         len = data.length;
2741
2742         /*
2743          * We're only interested in the first three bytes, actually.
2744          * But the data cache page is defined for the first 20.
2745          */
2746         if (len < 3)
2747                 goto bad_sense;
2748         else if (len > SD_BUF_SIZE) {
2749                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2750                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2751                 len = SD_BUF_SIZE;
2752         }
2753         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2754                 len = 192;
2755
2756         /* Get the data */
2757         if (len > first_len)
2758                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2759                                 &data, &sshdr);
2760
2761         if (!res) {
2762                 int offset = data.header_length + data.block_descriptor_length;
2763
2764                 while (offset < len) {
2765                         u8 page_code = buffer[offset] & 0x3F;
2766                         u8 spf       = buffer[offset] & 0x40;
2767
2768                         if (page_code == 8 || page_code == 6) {
2769                                 /* We're interested only in the first 3 bytes.
2770                                  */
2771                                 if (len - offset <= 2) {
2772                                         sd_first_printk(KERN_ERR, sdkp,
2773                                                 "Incomplete mode parameter "
2774                                                         "data\n");
2775                                         goto defaults;
2776                                 } else {
2777                                         modepage = page_code;
2778                                         goto Page_found;
2779                                 }
2780                         } else {
2781                                 /* Go to the next page */
2782                                 if (spf && len - offset > 3)
2783                                         offset += 4 + (buffer[offset+2] << 8) +
2784                                                 buffer[offset+3];
2785                                 else if (!spf && len - offset > 1)
2786                                         offset += 2 + buffer[offset+1];
2787                                 else {
2788                                         sd_first_printk(KERN_ERR, sdkp,
2789                                                         "Incomplete mode "
2790                                                         "parameter data\n");
2791                                         goto defaults;
2792                                 }
2793                         }
2794                 }
2795
2796                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2797                 goto defaults;
2798
2799         Page_found:
2800                 if (modepage == 8) {
2801                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2802                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2803                 } else {
2804                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2805                         sdkp->RCD = 0;
2806                 }
2807
2808                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2809                 if (sdp->broken_fua) {
2810                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2811                         sdkp->DPOFUA = 0;
2812                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2813                            !sdkp->device->use_16_for_rw) {
2814                         sd_first_printk(KERN_NOTICE, sdkp,
2815                                   "Uses READ/WRITE(6), disabling FUA\n");
2816                         sdkp->DPOFUA = 0;
2817                 }
2818
2819                 /* No cache flush allowed for write protected devices */
2820                 if (sdkp->WCE && sdkp->write_prot)
2821                         sdkp->WCE = 0;
2822
2823                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2824                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2825                         sd_printk(KERN_NOTICE, sdkp,
2826                                   "Write cache: %s, read cache: %s, %s\n",
2827                                   sdkp->WCE ? "enabled" : "disabled",
2828                                   sdkp->RCD ? "disabled" : "enabled",
2829                                   sdkp->DPOFUA ? "supports DPO and FUA"
2830                                   : "doesn't support DPO or FUA");
2831
2832                 return;
2833         }
2834
2835 bad_sense:
2836         if (scsi_sense_valid(&sshdr) &&
2837             sshdr.sense_key == ILLEGAL_REQUEST &&
2838             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2839                 /* Invalid field in CDB */
2840                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2841         else
2842                 sd_first_printk(KERN_ERR, sdkp,
2843                                 "Asking for cache data failed\n");
2844
2845 defaults:
2846         if (sdp->wce_default_on) {
2847                 sd_first_printk(KERN_NOTICE, sdkp,
2848                                 "Assuming drive cache: write back\n");
2849                 sdkp->WCE = 1;
2850         } else {
2851                 sd_first_printk(KERN_ERR, sdkp,
2852                                 "Assuming drive cache: write through\n");
2853                 sdkp->WCE = 0;
2854         }
2855         sdkp->RCD = 0;
2856         sdkp->DPOFUA = 0;
2857 }
2858
2859 /*
2860  * The ATO bit indicates whether the DIF application tag is available
2861  * for use by the operating system.
2862  */
2863 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2864 {
2865         int res, offset;
2866         struct scsi_device *sdp = sdkp->device;
2867         struct scsi_mode_data data;
2868         struct scsi_sense_hdr sshdr;
2869
2870         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2871                 return;
2872
2873         if (sdkp->protection_type == 0)
2874                 return;
2875
2876         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2877                               sdkp->max_retries, &data, &sshdr);
2878
2879         if (res < 0 || !data.header_length ||
2880             data.length < 6) {
2881                 sd_first_printk(KERN_WARNING, sdkp,
2882                           "getting Control mode page failed, assume no ATO\n");
2883
2884                 if (scsi_sense_valid(&sshdr))
2885                         sd_print_sense_hdr(sdkp, &sshdr);
2886
2887                 return;
2888         }
2889
2890         offset = data.header_length + data.block_descriptor_length;
2891
2892         if ((buffer[offset] & 0x3f) != 0x0a) {
2893                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2894                 return;
2895         }
2896
2897         if ((buffer[offset + 5] & 0x80) == 0)
2898                 return;
2899
2900         sdkp->ATO = 1;
2901
2902         return;
2903 }
2904
2905 /**
2906  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2907  * @sdkp: disk to query
2908  */
2909 static void sd_read_block_limits(struct scsi_disk *sdkp)
2910 {
2911         unsigned int sector_sz = sdkp->device->sector_size;
2912         const int vpd_len = 64;
2913         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2914
2915         if (!buffer ||
2916             /* Block Limits VPD */
2917             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2918                 goto out;
2919
2920         blk_queue_io_min(sdkp->disk->queue,
2921                          get_unaligned_be16(&buffer[6]) * sector_sz);
2922
2923         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2924         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2925
2926         if (buffer[3] == 0x3c) {
2927                 unsigned int lba_count, desc_count;
2928
2929                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2930
2931                 if (!sdkp->lbpme)
2932                         goto out;
2933
2934                 lba_count = get_unaligned_be32(&buffer[20]);
2935                 desc_count = get_unaligned_be32(&buffer[24]);
2936
2937                 if (lba_count && desc_count)
2938                         sdkp->max_unmap_blocks = lba_count;
2939
2940                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2941
2942                 if (buffer[32] & 0x80)
2943                         sdkp->unmap_alignment =
2944                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2945
2946                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2947
2948                         if (sdkp->max_unmap_blocks)
2949                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2950                         else
2951                                 sd_config_discard(sdkp, SD_LBP_WS16);
2952
2953                 } else {        /* LBP VPD page tells us what to use */
2954                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2955                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2956                         else if (sdkp->lbpws)
2957                                 sd_config_discard(sdkp, SD_LBP_WS16);
2958                         else if (sdkp->lbpws10)
2959                                 sd_config_discard(sdkp, SD_LBP_WS10);
2960                         else
2961                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2962                 }
2963         }
2964
2965  out:
2966         kfree(buffer);
2967 }
2968
2969 /**
2970  * sd_read_block_characteristics - Query block dev. characteristics
2971  * @sdkp: disk to query
2972  */
2973 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2974 {
2975         struct request_queue *q = sdkp->disk->queue;
2976         unsigned char *buffer;
2977         u16 rot;
2978         const int vpd_len = 64;
2979
2980         buffer = kmalloc(vpd_len, GFP_KERNEL);
2981
2982         if (!buffer ||
2983             /* Block Device Characteristics VPD */
2984             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2985                 goto out;
2986
2987         rot = get_unaligned_be16(&buffer[4]);
2988
2989         if (rot == 1) {
2990                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2991                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2992         }
2993
2994         if (sdkp->device->type == TYPE_ZBC) {
2995                 /* Host-managed */
2996                 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
2997         } else {
2998                 sdkp->zoned = (buffer[8] >> 4) & 3;
2999                 if (sdkp->zoned == 1) {
3000                         /* Host-aware */
3001                         blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3002                 } else {
3003                         /* Regular disk or drive managed disk */
3004                         blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3005                 }
3006         }
3007
3008         if (!sdkp->first_scan)
3009                 goto out;
3010
3011         if (blk_queue_is_zoned(q)) {
3012                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3013                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3014         } else {
3015                 if (sdkp->zoned == 1)
3016                         sd_printk(KERN_NOTICE, sdkp,
3017                                   "Host-aware SMR disk used as regular disk\n");
3018                 else if (sdkp->zoned == 2)
3019                         sd_printk(KERN_NOTICE, sdkp,
3020                                   "Drive-managed SMR disk\n");
3021         }
3022
3023  out:
3024         kfree(buffer);
3025 }
3026
3027 /**
3028  * sd_read_block_provisioning - Query provisioning VPD page
3029  * @sdkp: disk to query
3030  */
3031 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3032 {
3033         unsigned char *buffer;
3034         const int vpd_len = 8;
3035
3036         if (sdkp->lbpme == 0)
3037                 return;
3038
3039         buffer = kmalloc(vpd_len, GFP_KERNEL);
3040
3041         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3042                 goto out;
3043
3044         sdkp->lbpvpd    = 1;
3045         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3046         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3047         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3048
3049  out:
3050         kfree(buffer);
3051 }
3052
3053 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3054 {
3055         struct scsi_device *sdev = sdkp->device;
3056
3057         if (sdev->host->no_write_same) {
3058                 sdev->no_write_same = 1;
3059
3060                 return;
3061         }
3062
3063         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3064                 /* too large values might cause issues with arcmsr */
3065                 int vpd_buf_len = 64;
3066
3067                 sdev->no_report_opcodes = 1;
3068
3069                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3070                  * CODES is unsupported and the device has an ATA
3071                  * Information VPD page (SAT).
3072                  */
3073                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3074                         sdev->no_write_same = 1;
3075         }
3076
3077         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3078                 sdkp->ws16 = 1;
3079
3080         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3081                 sdkp->ws10 = 1;
3082 }
3083
3084 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3085 {
3086         struct scsi_device *sdev = sdkp->device;
3087
3088         if (!sdev->security_supported)
3089                 return;
3090
3091         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3092                         SECURITY_PROTOCOL_IN) == 1 &&
3093             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3094                         SECURITY_PROTOCOL_OUT) == 1)
3095                 sdkp->security = 1;
3096 }
3097
3098 /*
3099  * Determine the device's preferred I/O size for reads and writes
3100  * unless the reported value is unreasonably small, large, not a
3101  * multiple of the physical block size, or simply garbage.
3102  */
3103 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3104                                       unsigned int dev_max)
3105 {
3106         struct scsi_device *sdp = sdkp->device;
3107         unsigned int opt_xfer_bytes =
3108                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3109
3110         if (sdkp->opt_xfer_blocks == 0)
3111                 return false;
3112
3113         if (sdkp->opt_xfer_blocks > dev_max) {
3114                 sd_first_printk(KERN_WARNING, sdkp,
3115                                 "Optimal transfer size %u logical blocks " \
3116                                 "> dev_max (%u logical blocks)\n",
3117                                 sdkp->opt_xfer_blocks, dev_max);
3118                 return false;
3119         }
3120
3121         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3122                 sd_first_printk(KERN_WARNING, sdkp,
3123                                 "Optimal transfer size %u logical blocks " \
3124                                 "> sd driver limit (%u logical blocks)\n",
3125                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3126                 return false;
3127         }
3128
3129         if (opt_xfer_bytes < PAGE_SIZE) {
3130                 sd_first_printk(KERN_WARNING, sdkp,
3131                                 "Optimal transfer size %u bytes < " \
3132                                 "PAGE_SIZE (%u bytes)\n",
3133                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3134                 return false;
3135         }
3136
3137         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3138                 sd_first_printk(KERN_WARNING, sdkp,
3139                                 "Optimal transfer size %u bytes not a " \
3140                                 "multiple of physical block size (%u bytes)\n",
3141                                 opt_xfer_bytes, sdkp->physical_block_size);
3142                 return false;
3143         }
3144
3145         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3146                         opt_xfer_bytes);
3147         return true;
3148 }
3149
3150 /**
3151  *      sd_revalidate_disk - called the first time a new disk is seen,
3152  *      performs disk spin up, read_capacity, etc.
3153  *      @disk: struct gendisk we care about
3154  **/
3155 static int sd_revalidate_disk(struct gendisk *disk)
3156 {
3157         struct scsi_disk *sdkp = scsi_disk(disk);
3158         struct scsi_device *sdp = sdkp->device;
3159         struct request_queue *q = sdkp->disk->queue;
3160         sector_t old_capacity = sdkp->capacity;
3161         unsigned char *buffer;
3162         unsigned int dev_max, rw_max;
3163
3164         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3165                                       "sd_revalidate_disk\n"));
3166
3167         /*
3168          * If the device is offline, don't try and read capacity or any
3169          * of the other niceties.
3170          */
3171         if (!scsi_device_online(sdp))
3172                 goto out;
3173
3174         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3175         if (!buffer) {
3176                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3177                           "allocation failure.\n");
3178                 goto out;
3179         }
3180
3181         sd_spinup_disk(sdkp);
3182
3183         /*
3184          * Without media there is no reason to ask; moreover, some devices
3185          * react badly if we do.
3186          */
3187         if (sdkp->media_present) {
3188                 sd_read_capacity(sdkp, buffer);
3189
3190                 /*
3191                  * set the default to rotational.  All non-rotational devices
3192                  * support the block characteristics VPD page, which will
3193                  * cause this to be updated correctly and any device which
3194                  * doesn't support it should be treated as rotational.
3195                  */
3196                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3197                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3198
3199                 if (scsi_device_supports_vpd(sdp)) {
3200                         sd_read_block_provisioning(sdkp);
3201                         sd_read_block_limits(sdkp);
3202                         sd_read_block_characteristics(sdkp);
3203                         sd_zbc_read_zones(sdkp, buffer);
3204                 }
3205
3206                 sd_print_capacity(sdkp, old_capacity);
3207
3208                 sd_read_write_protect_flag(sdkp, buffer);
3209                 sd_read_cache_type(sdkp, buffer);
3210                 sd_read_app_tag_own(sdkp, buffer);
3211                 sd_read_write_same(sdkp, buffer);
3212                 sd_read_security(sdkp, buffer);
3213         }
3214
3215         /*
3216          * We now have all cache related info, determine how we deal
3217          * with flush requests.
3218          */
3219         sd_set_flush_flag(sdkp);
3220
3221         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3222         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3223
3224         /* Some devices report a maximum block count for READ/WRITE requests. */
3225         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3226         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3227
3228         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3229                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3230                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3231         } else {
3232                 q->limits.io_opt = 0;
3233                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3234                                       (sector_t)BLK_DEF_MAX_SECTORS);
3235         }
3236
3237         /* Do not exceed controller limit */
3238         rw_max = min(rw_max, queue_max_hw_sectors(q));
3239
3240         /*
3241          * Only update max_sectors if previously unset or if the current value
3242          * exceeds the capabilities of the hardware.
3243          */
3244         if (sdkp->first_scan ||
3245             q->limits.max_sectors > q->limits.max_dev_sectors ||
3246             q->limits.max_sectors > q->limits.max_hw_sectors)
3247                 q->limits.max_sectors = rw_max;
3248
3249         sdkp->first_scan = 0;
3250
3251         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3252         sd_config_write_same(sdkp);
3253         kfree(buffer);
3254
3255         /*
3256          * For a zoned drive, revalidating the zones can be done only once
3257          * the gendisk capacity is set. So if this fails, set back the gendisk
3258          * capacity to 0.
3259          */
3260         if (sd_zbc_revalidate_zones(sdkp))
3261                 set_capacity_and_notify(disk, 0);
3262
3263  out:
3264         return 0;
3265 }
3266
3267 /**
3268  *      sd_unlock_native_capacity - unlock native capacity
3269  *      @disk: struct gendisk to set capacity for
3270  *
3271  *      Block layer calls this function if it detects that partitions
3272  *      on @disk reach beyond the end of the device.  If the SCSI host
3273  *      implements ->unlock_native_capacity() method, it's invoked to
3274  *      give it a chance to adjust the device capacity.
3275  *
3276  *      CONTEXT:
3277  *      Defined by block layer.  Might sleep.
3278  */
3279 static void sd_unlock_native_capacity(struct gendisk *disk)
3280 {
3281         struct scsi_device *sdev = scsi_disk(disk)->device;
3282
3283         if (sdev->host->hostt->unlock_native_capacity)
3284                 sdev->host->hostt->unlock_native_capacity(sdev);
3285 }
3286
3287 /**
3288  *      sd_format_disk_name - format disk name
3289  *      @prefix: name prefix - ie. "sd" for SCSI disks
3290  *      @index: index of the disk to format name for
3291  *      @buf: output buffer
3292  *      @buflen: length of the output buffer
3293  *
3294  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3295  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3296  *      which is followed by sdaaa.
3297  *
3298  *      This is basically 26 base counting with one extra 'nil' entry
3299  *      at the beginning from the second digit on and can be
3300  *      determined using similar method as 26 base conversion with the
3301  *      index shifted -1 after each digit is computed.
3302  *
3303  *      CONTEXT:
3304  *      Don't care.
3305  *
3306  *      RETURNS:
3307  *      0 on success, -errno on failure.
3308  */
3309 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3310 {
3311         const int base = 'z' - 'a' + 1;
3312         char *begin = buf + strlen(prefix);
3313         char *end = buf + buflen;
3314         char *p;
3315         int unit;
3316
3317         p = end - 1;
3318         *p = '\0';
3319         unit = base;
3320         do {
3321                 if (p == begin)
3322                         return -EINVAL;
3323                 *--p = 'a' + (index % unit);
3324                 index = (index / unit) - 1;
3325         } while (index >= 0);
3326
3327         memmove(begin, p, end - p);
3328         memcpy(buf, prefix, strlen(prefix));
3329
3330         return 0;
3331 }
3332
3333 /**
3334  *      sd_probe - called during driver initialization and whenever a
3335  *      new scsi device is attached to the system. It is called once
3336  *      for each scsi device (not just disks) present.
3337  *      @dev: pointer to device object
3338  *
3339  *      Returns 0 if successful (or not interested in this scsi device
3340  *      (e.g. scanner)); 1 when there is an error.
3341  *
3342  *      Note: this function is invoked from the scsi mid-level.
3343  *      This function sets up the mapping between a given
3344  *      <host,channel,id,lun> (found in sdp) and new device name
3345  *      (e.g. /dev/sda). More precisely it is the block device major
3346  *      and minor number that is chosen here.
3347  *
3348  *      Assume sd_probe is not re-entrant (for time being)
3349  *      Also think about sd_probe() and sd_remove() running coincidentally.
3350  **/
3351 static int sd_probe(struct device *dev)
3352 {
3353         struct scsi_device *sdp = to_scsi_device(dev);
3354         struct scsi_disk *sdkp;
3355         struct gendisk *gd;
3356         int index;
3357         int error;
3358
3359         scsi_autopm_get_device(sdp);
3360         error = -ENODEV;
3361         if (sdp->type != TYPE_DISK &&
3362             sdp->type != TYPE_ZBC &&
3363             sdp->type != TYPE_MOD &&
3364             sdp->type != TYPE_RBC)
3365                 goto out;
3366
3367         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3368                 sdev_printk(KERN_WARNING, sdp,
3369                             "Unsupported ZBC host-managed device.\n");
3370                 goto out;
3371         }
3372
3373         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3374                                         "sd_probe\n"));
3375
3376         error = -ENOMEM;
3377         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3378         if (!sdkp)
3379                 goto out;
3380
3381         gd = __alloc_disk_node(sdp->request_queue, NUMA_NO_NODE,
3382                                &sd_bio_compl_lkclass);
3383         if (!gd)
3384                 goto out_free;
3385
3386         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3387         if (index < 0) {
3388                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3389                 goto out_put;
3390         }
3391
3392         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3393         if (error) {
3394                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3395                 goto out_free_index;
3396         }
3397
3398         sdkp->device = sdp;
3399         sdkp->driver = &sd_template;
3400         sdkp->disk = gd;
3401         sdkp->index = index;
3402         sdkp->max_retries = SD_MAX_RETRIES;
3403         atomic_set(&sdkp->openers, 0);
3404         atomic_set(&sdkp->device->ioerr_cnt, 0);
3405
3406         if (!sdp->request_queue->rq_timeout) {
3407                 if (sdp->type != TYPE_MOD)
3408                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3409                 else
3410                         blk_queue_rq_timeout(sdp->request_queue,
3411                                              SD_MOD_TIMEOUT);
3412         }
3413
3414         device_initialize(&sdkp->dev);
3415         sdkp->dev.parent = get_device(dev);
3416         sdkp->dev.class = &sd_disk_class;
3417         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3418
3419         error = device_add(&sdkp->dev);
3420         if (error) {
3421                 put_device(&sdkp->dev);
3422                 goto out;
3423         }
3424
3425         dev_set_drvdata(dev, sdkp);
3426
3427         gd->major = sd_major((index & 0xf0) >> 4);
3428         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3429         gd->minors = SD_MINORS;
3430
3431         gd->fops = &sd_fops;
3432         gd->private_data = &sdkp->driver;
3433
3434         /* defaults, until the device tells us otherwise */
3435         sdp->sector_size = 512;
3436         sdkp->capacity = 0;
3437         sdkp->media_present = 1;
3438         sdkp->write_prot = 0;
3439         sdkp->cache_override = 0;
3440         sdkp->WCE = 0;
3441         sdkp->RCD = 0;
3442         sdkp->ATO = 0;
3443         sdkp->first_scan = 1;
3444         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3445
3446         sd_revalidate_disk(gd);
3447
3448         gd->flags = GENHD_FL_EXT_DEVT;
3449         if (sdp->removable) {
3450                 gd->flags |= GENHD_FL_REMOVABLE;
3451                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3452                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3453         }
3454
3455         blk_pm_runtime_init(sdp->request_queue, dev);
3456         if (sdp->rpm_autosuspend) {
3457                 pm_runtime_set_autosuspend_delay(dev,
3458                         sdp->host->hostt->rpm_autosuspend_delay);
3459         }
3460         device_add_disk(dev, gd, NULL);
3461         if (sdkp->capacity)
3462                 sd_dif_config_host(sdkp);
3463
3464         sd_revalidate_disk(gd);
3465
3466         if (sdkp->security) {
3467                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3468                 if (sdkp->opal_dev)
3469                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3470         }
3471
3472         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3473                   sdp->removable ? "removable " : "");
3474         scsi_autopm_put_device(sdp);
3475
3476         return 0;
3477
3478  out_free_index:
3479         ida_free(&sd_index_ida, index);
3480  out_put:
3481         put_disk(gd);
3482  out_free:
3483         kfree(sdkp);
3484  out:
3485         scsi_autopm_put_device(sdp);
3486         return error;
3487 }
3488
3489 /**
3490  *      sd_remove - called whenever a scsi disk (previously recognized by
3491  *      sd_probe) is detached from the system. It is called (potentially
3492  *      multiple times) during sd module unload.
3493  *      @dev: pointer to device object
3494  *
3495  *      Note: this function is invoked from the scsi mid-level.
3496  *      This function potentially frees up a device name (e.g. /dev/sdc)
3497  *      that could be re-used by a subsequent sd_probe().
3498  *      This function is not called when the built-in sd driver is "exit-ed".
3499  **/
3500 static int sd_remove(struct device *dev)
3501 {
3502         struct scsi_disk *sdkp;
3503
3504         sdkp = dev_get_drvdata(dev);
3505         scsi_autopm_get_device(sdkp->device);
3506
3507         async_synchronize_full_domain(&scsi_sd_pm_domain);
3508         device_del(&sdkp->dev);
3509         del_gendisk(sdkp->disk);
3510         sd_shutdown(dev);
3511
3512         free_opal_dev(sdkp->opal_dev);
3513
3514         mutex_lock(&sd_ref_mutex);
3515         dev_set_drvdata(dev, NULL);
3516         put_device(&sdkp->dev);
3517         mutex_unlock(&sd_ref_mutex);
3518
3519         return 0;
3520 }
3521
3522 /**
3523  *      scsi_disk_release - Called to free the scsi_disk structure
3524  *      @dev: pointer to embedded class device
3525  *
3526  *      sd_ref_mutex must be held entering this routine.  Because it is
3527  *      called on last put, you should always use the scsi_disk_get()
3528  *      scsi_disk_put() helpers which manipulate the semaphore directly
3529  *      and never do a direct put_device.
3530  **/
3531 static void scsi_disk_release(struct device *dev)
3532 {
3533         struct scsi_disk *sdkp = to_scsi_disk(dev);
3534         struct gendisk *disk = sdkp->disk;
3535         struct request_queue *q = disk->queue;
3536
3537         ida_free(&sd_index_ida, sdkp->index);
3538
3539         /*
3540          * Wait until all requests that are in progress have completed.
3541          * This is necessary to avoid that e.g. scsi_end_request() crashes
3542          * due to clearing the disk->private_data pointer. Wait from inside
3543          * scsi_disk_release() instead of from sd_release() to avoid that
3544          * freezing and unfreezing the request queue affects user space I/O
3545          * in case multiple processes open a /dev/sd... node concurrently.
3546          */
3547         blk_mq_freeze_queue(q);
3548         blk_mq_unfreeze_queue(q);
3549
3550         disk->private_data = NULL;
3551         put_disk(disk);
3552         put_device(&sdkp->device->sdev_gendev);
3553
3554         sd_zbc_release_disk(sdkp);
3555
3556         kfree(sdkp);
3557 }
3558
3559 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3560 {
3561         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3562         struct scsi_sense_hdr sshdr;
3563         struct scsi_device *sdp = sdkp->device;
3564         int res;
3565
3566         if (start)
3567                 cmd[4] |= 1;    /* START */
3568
3569         if (sdp->start_stop_pwr_cond)
3570                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3571
3572         if (!scsi_device_online(sdp))
3573                 return -ENODEV;
3574
3575         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3576                         SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3577         if (res) {
3578                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3579                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3580                         sd_print_sense_hdr(sdkp, &sshdr);
3581                         /* 0x3a is medium not present */
3582                         if (sshdr.asc == 0x3a)
3583                                 res = 0;
3584                 }
3585         }
3586
3587         /* SCSI error codes must not go to the generic layer */
3588         if (res)
3589                 return -EIO;
3590
3591         return 0;
3592 }
3593
3594 /*
3595  * Send a SYNCHRONIZE CACHE instruction down to the device through
3596  * the normal SCSI command structure.  Wait for the command to
3597  * complete.
3598  */
3599 static void sd_shutdown(struct device *dev)
3600 {
3601         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3602
3603         if (!sdkp)
3604                 return;         /* this can happen */
3605
3606         if (pm_runtime_suspended(dev))
3607                 return;
3608
3609         if (sdkp->WCE && sdkp->media_present) {
3610                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3611                 sd_sync_cache(sdkp, NULL);
3612         }
3613
3614         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3615                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3616                 sd_start_stop_device(sdkp, 0);
3617         }
3618 }
3619
3620 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3621 {
3622         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3623         struct scsi_sense_hdr sshdr;
3624         int ret = 0;
3625
3626         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3627                 return 0;
3628
3629         if (sdkp->WCE && sdkp->media_present) {
3630                 if (!sdkp->device->silence_suspend)
3631                         sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3632                 ret = sd_sync_cache(sdkp, &sshdr);
3633
3634                 if (ret) {
3635                         /* ignore OFFLINE device */
3636                         if (ret == -ENODEV)
3637                                 return 0;
3638
3639                         if (!scsi_sense_valid(&sshdr) ||
3640                             sshdr.sense_key != ILLEGAL_REQUEST)
3641                                 return ret;
3642
3643                         /*
3644                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3645                          * doesn't support sync. There's not much to do and
3646                          * suspend shouldn't fail.
3647                          */
3648                         ret = 0;
3649                 }
3650         }
3651
3652         if (sdkp->device->manage_start_stop) {
3653                 if (!sdkp->device->silence_suspend)
3654                         sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3655                 /* an error is not worth aborting a system sleep */
3656                 ret = sd_start_stop_device(sdkp, 0);
3657                 if (ignore_stop_errors)
3658                         ret = 0;
3659         }
3660
3661         return ret;
3662 }
3663
3664 static int sd_suspend_system(struct device *dev)
3665 {
3666         return sd_suspend_common(dev, true);
3667 }
3668
3669 static int sd_suspend_runtime(struct device *dev)
3670 {
3671         return sd_suspend_common(dev, false);
3672 }
3673
3674 static int sd_resume(struct device *dev)
3675 {
3676         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3677         int ret;
3678
3679         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3680                 return 0;
3681
3682         if (!sdkp->device->manage_start_stop)
3683                 return 0;
3684
3685         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3686         ret = sd_start_stop_device(sdkp, 1);
3687         if (!ret)
3688                 opal_unlock_from_suspend(sdkp->opal_dev);
3689         return ret;
3690 }
3691
3692 static int sd_resume_runtime(struct device *dev)
3693 {
3694         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3695         struct scsi_device *sdp;
3696
3697         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3698                 return 0;
3699
3700         sdp = sdkp->device;
3701
3702         if (sdp->ignore_media_change) {
3703                 /* clear the device's sense data */
3704                 static const u8 cmd[10] = { REQUEST_SENSE };
3705
3706                 if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3707                                  NULL, sdp->request_queue->rq_timeout, 1, 0,
3708                                  RQF_PM, NULL))
3709                         sd_printk(KERN_NOTICE, sdkp,
3710                                   "Failed to clear sense data\n");
3711         }
3712
3713         return sd_resume(dev);
3714 }
3715
3716 /**
3717  *      init_sd - entry point for this driver (both when built in or when
3718  *      a module).
3719  *
3720  *      Note: this function registers this driver with the scsi mid-level.
3721  **/
3722 static int __init init_sd(void)
3723 {
3724         int majors = 0, i, err;
3725
3726         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3727
3728         for (i = 0; i < SD_MAJORS; i++) {
3729                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3730                         continue;
3731                 majors++;
3732         }
3733
3734         if (!majors)
3735                 return -ENODEV;
3736
3737         err = class_register(&sd_disk_class);
3738         if (err)
3739                 goto err_out;
3740
3741         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3742                                          0, 0, NULL);
3743         if (!sd_cdb_cache) {
3744                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3745                 err = -ENOMEM;
3746                 goto err_out_class;
3747         }
3748
3749         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3750         if (!sd_cdb_pool) {
3751                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3752                 err = -ENOMEM;
3753                 goto err_out_cache;
3754         }
3755
3756         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3757         if (!sd_page_pool) {
3758                 printk(KERN_ERR "sd: can't init discard page pool\n");
3759                 err = -ENOMEM;
3760                 goto err_out_ppool;
3761         }
3762
3763         err = scsi_register_driver(&sd_template.gendrv);
3764         if (err)
3765                 goto err_out_driver;
3766
3767         return 0;
3768
3769 err_out_driver:
3770         mempool_destroy(sd_page_pool);
3771
3772 err_out_ppool:
3773         mempool_destroy(sd_cdb_pool);
3774
3775 err_out_cache:
3776         kmem_cache_destroy(sd_cdb_cache);
3777
3778 err_out_class:
3779         class_unregister(&sd_disk_class);
3780 err_out:
3781         for (i = 0; i < SD_MAJORS; i++)
3782                 unregister_blkdev(sd_major(i), "sd");
3783         return err;
3784 }
3785
3786 /**
3787  *      exit_sd - exit point for this driver (when it is a module).
3788  *
3789  *      Note: this function unregisters this driver from the scsi mid-level.
3790  **/
3791 static void __exit exit_sd(void)
3792 {
3793         int i;
3794
3795         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3796
3797         scsi_unregister_driver(&sd_template.gendrv);
3798         mempool_destroy(sd_cdb_pool);
3799         mempool_destroy(sd_page_pool);
3800         kmem_cache_destroy(sd_cdb_cache);
3801
3802         class_unregister(&sd_disk_class);
3803
3804         for (i = 0; i < SD_MAJORS; i++)
3805                 unregister_blkdev(sd_major(i), "sd");
3806 }
3807
3808 module_init(init_sd);
3809 module_exit(exit_sd);
3810
3811 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3812 {
3813         scsi_print_sense_hdr(sdkp->device,
3814                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3815 }
3816
3817 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3818 {
3819         const char *hb_string = scsi_hostbyte_string(result);
3820
3821         if (hb_string)
3822                 sd_printk(KERN_INFO, sdkp,
3823                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3824                           hb_string ? hb_string : "invalid",
3825                           "DRIVER_OK");
3826         else
3827                 sd_printk(KERN_INFO, sdkp,
3828                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3829                           msg, host_byte(result), "DRIVER_OK");
3830 }