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         unsigned int nr_bytes = blk_rq_bytes(rq);
1076         blk_status_t ret;
1077
1078         if (sdkp->device->no_write_same)
1079                 return BLK_STS_TARGET;
1080
1081         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1082
1083         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1084
1085         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1086                 cmd->cmd_len = 16;
1087                 cmd->cmnd[0] = WRITE_SAME_16;
1088                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1089                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1090         } else {
1091                 cmd->cmd_len = 10;
1092                 cmd->cmnd[0] = WRITE_SAME;
1093                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1094                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1095         }
1096
1097         cmd->transfersize = sdp->sector_size;
1098         cmd->allowed = sdkp->max_retries;
1099
1100         /*
1101          * For WRITE SAME the data transferred via the DATA OUT buffer is
1102          * different from the amount of data actually written to the target.
1103          *
1104          * We set up __data_len to the amount of data transferred via the
1105          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1106          * to transfer a single sector of data first, but then reset it to
1107          * the amount of data to be written right after so that the I/O path
1108          * knows how much to actually write.
1109          */
1110         rq->__data_len = sdp->sector_size;
1111         ret = scsi_alloc_sgtables(cmd);
1112         rq->__data_len = nr_bytes;
1113
1114         return ret;
1115 }
1116
1117 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1118 {
1119         struct request *rq = scsi_cmd_to_rq(cmd);
1120         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1121
1122         /* flush requests don't perform I/O, zero the S/G table */
1123         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1124
1125         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1126         cmd->cmd_len = 10;
1127         cmd->transfersize = 0;
1128         cmd->allowed = sdkp->max_retries;
1129
1130         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1131         return BLK_STS_OK;
1132 }
1133
1134 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1135                                        sector_t lba, unsigned int nr_blocks,
1136                                        unsigned char flags)
1137 {
1138         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1139         if (unlikely(cmd->cmnd == NULL))
1140                 return BLK_STS_RESOURCE;
1141
1142         cmd->cmd_len = SD_EXT_CDB_SIZE;
1143         memset(cmd->cmnd, 0, cmd->cmd_len);
1144
1145         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1146         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1147         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1148         cmd->cmnd[10] = flags;
1149         put_unaligned_be64(lba, &cmd->cmnd[12]);
1150         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1151         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1152
1153         return BLK_STS_OK;
1154 }
1155
1156 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1157                                        sector_t lba, unsigned int nr_blocks,
1158                                        unsigned char flags)
1159 {
1160         cmd->cmd_len  = 16;
1161         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1162         cmd->cmnd[1]  = flags;
1163         cmd->cmnd[14] = 0;
1164         cmd->cmnd[15] = 0;
1165         put_unaligned_be64(lba, &cmd->cmnd[2]);
1166         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1167
1168         return BLK_STS_OK;
1169 }
1170
1171 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1172                                        sector_t lba, unsigned int nr_blocks,
1173                                        unsigned char flags)
1174 {
1175         cmd->cmd_len = 10;
1176         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1177         cmd->cmnd[1] = flags;
1178         cmd->cmnd[6] = 0;
1179         cmd->cmnd[9] = 0;
1180         put_unaligned_be32(lba, &cmd->cmnd[2]);
1181         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1182
1183         return BLK_STS_OK;
1184 }
1185
1186 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1187                                       sector_t lba, unsigned int nr_blocks,
1188                                       unsigned char flags)
1189 {
1190         /* Avoid that 0 blocks gets translated into 256 blocks. */
1191         if (WARN_ON_ONCE(nr_blocks == 0))
1192                 return BLK_STS_IOERR;
1193
1194         if (unlikely(flags & 0x8)) {
1195                 /*
1196                  * This happens only if this drive failed 10byte rw
1197                  * command with ILLEGAL_REQUEST during operation and
1198                  * thus turned off use_10_for_rw.
1199                  */
1200                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1201                 return BLK_STS_IOERR;
1202         }
1203
1204         cmd->cmd_len = 6;
1205         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1206         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1207         cmd->cmnd[2] = (lba >> 8) & 0xff;
1208         cmd->cmnd[3] = lba & 0xff;
1209         cmd->cmnd[4] = nr_blocks;
1210         cmd->cmnd[5] = 0;
1211
1212         return BLK_STS_OK;
1213 }
1214
1215 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1216 {
1217         struct request *rq = scsi_cmd_to_rq(cmd);
1218         struct scsi_device *sdp = cmd->device;
1219         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1220         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1221         sector_t threshold;
1222         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1223         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1224         bool write = rq_data_dir(rq) == WRITE;
1225         unsigned char protect, fua;
1226         blk_status_t ret;
1227         unsigned int dif;
1228         bool dix;
1229
1230         ret = scsi_alloc_sgtables(cmd);
1231         if (ret != BLK_STS_OK)
1232                 return ret;
1233
1234         ret = BLK_STS_IOERR;
1235         if (!scsi_device_online(sdp) || sdp->changed) {
1236                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1237                 goto fail;
1238         }
1239
1240         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1241                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1242                 goto fail;
1243         }
1244
1245         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1246                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1247                 goto fail;
1248         }
1249
1250         /*
1251          * Some SD card readers can't handle accesses which touch the
1252          * last one or two logical blocks. Split accesses as needed.
1253          */
1254         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1255
1256         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1257                 if (lba < threshold) {
1258                         /* Access up to the threshold but not beyond */
1259                         nr_blocks = threshold - lba;
1260                 } else {
1261                         /* Access only a single logical block */
1262                         nr_blocks = 1;
1263                 }
1264         }
1265
1266         if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1267                 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1268                 if (ret)
1269                         goto fail;
1270         }
1271
1272         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1273         dix = scsi_prot_sg_count(cmd);
1274         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1275
1276         if (dif || dix)
1277                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1278         else
1279                 protect = 0;
1280
1281         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1282                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1283                                          protect | fua);
1284         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1285                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1286                                          protect | fua);
1287         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1288                    sdp->use_10_for_rw || protect) {
1289                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1290                                          protect | fua);
1291         } else {
1292                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1293                                         protect | fua);
1294         }
1295
1296         if (unlikely(ret != BLK_STS_OK))
1297                 goto fail;
1298
1299         /*
1300          * We shouldn't disconnect in the middle of a sector, so with a dumb
1301          * host adapter, it's safe to assume that we can at least transfer
1302          * this many bytes between each connect / disconnect.
1303          */
1304         cmd->transfersize = sdp->sector_size;
1305         cmd->underflow = nr_blocks << 9;
1306         cmd->allowed = sdkp->max_retries;
1307         cmd->sdb.length = nr_blocks * sdp->sector_size;
1308
1309         SCSI_LOG_HLQUEUE(1,
1310                          scmd_printk(KERN_INFO, cmd,
1311                                      "%s: block=%llu, count=%d\n", __func__,
1312                                      (unsigned long long)blk_rq_pos(rq),
1313                                      blk_rq_sectors(rq)));
1314         SCSI_LOG_HLQUEUE(2,
1315                          scmd_printk(KERN_INFO, cmd,
1316                                      "%s %d/%u 512 byte blocks.\n",
1317                                      write ? "writing" : "reading", nr_blocks,
1318                                      blk_rq_sectors(rq)));
1319
1320         /*
1321          * This indicates that the command is ready from our end to be queued.
1322          */
1323         return BLK_STS_OK;
1324 fail:
1325         scsi_free_sgtables(cmd);
1326         return ret;
1327 }
1328
1329 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1330 {
1331         struct request *rq = scsi_cmd_to_rq(cmd);
1332
1333         switch (req_op(rq)) {
1334         case REQ_OP_DISCARD:
1335                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1336                 case SD_LBP_UNMAP:
1337                         return sd_setup_unmap_cmnd(cmd);
1338                 case SD_LBP_WS16:
1339                         return sd_setup_write_same16_cmnd(cmd, true);
1340                 case SD_LBP_WS10:
1341                         return sd_setup_write_same10_cmnd(cmd, true);
1342                 case SD_LBP_ZERO:
1343                         return sd_setup_write_same10_cmnd(cmd, false);
1344                 default:
1345                         return BLK_STS_TARGET;
1346                 }
1347         case REQ_OP_WRITE_ZEROES:
1348                 return sd_setup_write_zeroes_cmnd(cmd);
1349         case REQ_OP_WRITE_SAME:
1350                 return sd_setup_write_same_cmnd(cmd);
1351         case REQ_OP_FLUSH:
1352                 return sd_setup_flush_cmnd(cmd);
1353         case REQ_OP_READ:
1354         case REQ_OP_WRITE:
1355         case REQ_OP_ZONE_APPEND:
1356                 return sd_setup_read_write_cmnd(cmd);
1357         case REQ_OP_ZONE_RESET:
1358                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1359                                                    false);
1360         case REQ_OP_ZONE_RESET_ALL:
1361                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1362                                                    true);
1363         case REQ_OP_ZONE_OPEN:
1364                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1365         case REQ_OP_ZONE_CLOSE:
1366                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1367         case REQ_OP_ZONE_FINISH:
1368                 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1369         default:
1370                 WARN_ON_ONCE(1);
1371                 return BLK_STS_NOTSUPP;
1372         }
1373 }
1374
1375 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1376 {
1377         struct request *rq = scsi_cmd_to_rq(SCpnt);
1378         u8 *cmnd;
1379
1380         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1381                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1382
1383         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1384                 cmnd = SCpnt->cmnd;
1385                 SCpnt->cmnd = NULL;
1386                 SCpnt->cmd_len = 0;
1387                 mempool_free(cmnd, sd_cdb_pool);
1388         }
1389 }
1390
1391 static bool sd_need_revalidate(struct block_device *bdev,
1392                 struct scsi_disk *sdkp)
1393 {
1394         if (sdkp->device->removable || sdkp->write_prot) {
1395                 if (bdev_check_media_change(bdev))
1396                         return true;
1397         }
1398
1399         /*
1400          * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1401          * nothing to do with partitions, BLKRRPART is used to force a full
1402          * revalidate after things like a format for historical reasons.
1403          */
1404         return test_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1405 }
1406
1407 /**
1408  *      sd_open - open a scsi disk device
1409  *      @bdev: Block device of the scsi disk to open
1410  *      @mode: FMODE_* mask
1411  *
1412  *      Returns 0 if successful. Returns a negated errno value in case
1413  *      of error.
1414  *
1415  *      Note: This can be called from a user context (e.g. fsck(1) )
1416  *      or from within the kernel (e.g. as a result of a mount(1) ).
1417  *      In the latter case @inode and @filp carry an abridged amount
1418  *      of information as noted above.
1419  *
1420  *      Locking: called with bdev->bd_disk->open_mutex held.
1421  **/
1422 static int sd_open(struct block_device *bdev, fmode_t mode)
1423 {
1424         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1425         struct scsi_device *sdev;
1426         int retval;
1427
1428         if (!sdkp)
1429                 return -ENXIO;
1430
1431         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1432
1433         sdev = sdkp->device;
1434
1435         /*
1436          * If the device is in error recovery, wait until it is done.
1437          * If the device is offline, then disallow any access to it.
1438          */
1439         retval = -ENXIO;
1440         if (!scsi_block_when_processing_errors(sdev))
1441                 goto error_out;
1442
1443         if (sd_need_revalidate(bdev, sdkp))
1444                 sd_revalidate_disk(bdev->bd_disk);
1445
1446         /*
1447          * If the drive is empty, just let the open fail.
1448          */
1449         retval = -ENOMEDIUM;
1450         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1451                 goto error_out;
1452
1453         /*
1454          * If the device has the write protect tab set, have the open fail
1455          * if the user expects to be able to write to the thing.
1456          */
1457         retval = -EROFS;
1458         if (sdkp->write_prot && (mode & FMODE_WRITE))
1459                 goto error_out;
1460
1461         /*
1462          * It is possible that the disk changing stuff resulted in
1463          * the device being taken offline.  If this is the case,
1464          * report this to the user, and don't pretend that the
1465          * open actually succeeded.
1466          */
1467         retval = -ENXIO;
1468         if (!scsi_device_online(sdev))
1469                 goto error_out;
1470
1471         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1472                 if (scsi_block_when_processing_errors(sdev))
1473                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1474         }
1475
1476         return 0;
1477
1478 error_out:
1479         scsi_disk_put(sdkp);
1480         return retval;
1481 }
1482
1483 /**
1484  *      sd_release - invoked when the (last) close(2) is called on this
1485  *      scsi disk.
1486  *      @disk: disk to release
1487  *      @mode: FMODE_* mask
1488  *
1489  *      Returns 0.
1490  *
1491  *      Note: may block (uninterruptible) if error recovery is underway
1492  *      on this disk.
1493  *
1494  *      Locking: called with bdev->bd_disk->open_mutex held.
1495  **/
1496 static void sd_release(struct gendisk *disk, fmode_t mode)
1497 {
1498         struct scsi_disk *sdkp = scsi_disk(disk);
1499         struct scsi_device *sdev = sdkp->device;
1500
1501         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1502
1503         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1504                 if (scsi_block_when_processing_errors(sdev))
1505                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1506         }
1507
1508         scsi_disk_put(sdkp);
1509 }
1510
1511 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1512 {
1513         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1514         struct scsi_device *sdp = sdkp->device;
1515         struct Scsi_Host *host = sdp->host;
1516         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1517         int diskinfo[4];
1518
1519         /* default to most commonly used values */
1520         diskinfo[0] = 0x40;     /* 1 << 6 */
1521         diskinfo[1] = 0x20;     /* 1 << 5 */
1522         diskinfo[2] = capacity >> 11;
1523
1524         /* override with calculated, extended default, or driver values */
1525         if (host->hostt->bios_param)
1526                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1527         else
1528                 scsicam_bios_param(bdev, capacity, diskinfo);
1529
1530         geo->heads = diskinfo[0];
1531         geo->sectors = diskinfo[1];
1532         geo->cylinders = diskinfo[2];
1533         return 0;
1534 }
1535
1536 /**
1537  *      sd_ioctl - process an ioctl
1538  *      @bdev: target block device
1539  *      @mode: FMODE_* mask
1540  *      @cmd: ioctl command number
1541  *      @arg: this is third argument given to ioctl(2) system call.
1542  *      Often contains a pointer.
1543  *
1544  *      Returns 0 if successful (some ioctls return positive numbers on
1545  *      success as well). Returns a negated errno value in case of error.
1546  *
1547  *      Note: most ioctls are forward onto the block subsystem or further
1548  *      down in the scsi subsystem.
1549  **/
1550 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1551                     unsigned int cmd, unsigned long arg)
1552 {
1553         struct gendisk *disk = bdev->bd_disk;
1554         struct scsi_disk *sdkp = scsi_disk(disk);
1555         struct scsi_device *sdp = sdkp->device;
1556         void __user *p = (void __user *)arg;
1557         int error;
1558
1559         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1560                                     "cmd=0x%x\n", disk->disk_name, cmd));
1561
1562         if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1563                 return -ENOIOCTLCMD;
1564
1565         /*
1566          * If we are in the middle of error recovery, don't let anyone
1567          * else try and use this device.  Also, if error recovery fails, it
1568          * may try and take the device offline, in which case all further
1569          * access to the device is prohibited.
1570          */
1571         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1572                         (mode & FMODE_NDELAY) != 0);
1573         if (error)
1574                 return error;
1575
1576         if (is_sed_ioctl(cmd))
1577                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1578         return scsi_ioctl(sdp, disk, mode, cmd, p);
1579 }
1580
1581 static void set_media_not_present(struct scsi_disk *sdkp)
1582 {
1583         if (sdkp->media_present)
1584                 sdkp->device->changed = 1;
1585
1586         if (sdkp->device->removable) {
1587                 sdkp->media_present = 0;
1588                 sdkp->capacity = 0;
1589         }
1590 }
1591
1592 static int media_not_present(struct scsi_disk *sdkp,
1593                              struct scsi_sense_hdr *sshdr)
1594 {
1595         if (!scsi_sense_valid(sshdr))
1596                 return 0;
1597
1598         /* not invoked for commands that could return deferred errors */
1599         switch (sshdr->sense_key) {
1600         case UNIT_ATTENTION:
1601         case NOT_READY:
1602                 /* medium not present */
1603                 if (sshdr->asc == 0x3A) {
1604                         set_media_not_present(sdkp);
1605                         return 1;
1606                 }
1607         }
1608         return 0;
1609 }
1610
1611 /**
1612  *      sd_check_events - check media events
1613  *      @disk: kernel device descriptor
1614  *      @clearing: disk events currently being cleared
1615  *
1616  *      Returns mask of DISK_EVENT_*.
1617  *
1618  *      Note: this function is invoked from the block subsystem.
1619  **/
1620 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1621 {
1622         struct scsi_disk *sdkp = scsi_disk_get(disk);
1623         struct scsi_device *sdp;
1624         int retval;
1625         bool disk_changed;
1626
1627         if (!sdkp)
1628                 return 0;
1629
1630         sdp = sdkp->device;
1631         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1632
1633         /*
1634          * If the device is offline, don't send any commands - just pretend as
1635          * if the command failed.  If the device ever comes back online, we
1636          * can deal with it then.  It is only because of unrecoverable errors
1637          * that we would ever take a device offline in the first place.
1638          */
1639         if (!scsi_device_online(sdp)) {
1640                 set_media_not_present(sdkp);
1641                 goto out;
1642         }
1643
1644         /*
1645          * Using TEST_UNIT_READY enables differentiation between drive with
1646          * no cartridge loaded - NOT READY, drive with changed cartridge -
1647          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1648          *
1649          * Drives that auto spin down. eg iomega jaz 1G, will be started
1650          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1651          * sd_revalidate() is called.
1652          */
1653         if (scsi_block_when_processing_errors(sdp)) {
1654                 struct scsi_sense_hdr sshdr = { 0, };
1655
1656                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1657                                               &sshdr);
1658
1659                 /* failed to execute TUR, assume media not present */
1660                 if (retval < 0 || host_byte(retval)) {
1661                         set_media_not_present(sdkp);
1662                         goto out;
1663                 }
1664
1665                 if (media_not_present(sdkp, &sshdr))
1666                         goto out;
1667         }
1668
1669         /*
1670          * For removable scsi disk we have to recognise the presence
1671          * of a disk in the drive.
1672          */
1673         if (!sdkp->media_present)
1674                 sdp->changed = 1;
1675         sdkp->media_present = 1;
1676 out:
1677         /*
1678          * sdp->changed is set under the following conditions:
1679          *
1680          *      Medium present state has changed in either direction.
1681          *      Device has indicated UNIT_ATTENTION.
1682          */
1683         disk_changed = sdp->changed;
1684         sdp->changed = 0;
1685         scsi_disk_put(sdkp);
1686         return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1687 }
1688
1689 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1690 {
1691         int retries, res;
1692         struct scsi_device *sdp = sdkp->device;
1693         const int timeout = sdp->request_queue->rq_timeout
1694                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1695         struct scsi_sense_hdr my_sshdr;
1696
1697         if (!scsi_device_online(sdp))
1698                 return -ENODEV;
1699
1700         /* caller might not be interested in sense, but we need it */
1701         if (!sshdr)
1702                 sshdr = &my_sshdr;
1703
1704         for (retries = 3; retries > 0; --retries) {
1705                 unsigned char cmd[10] = { 0 };
1706
1707                 cmd[0] = SYNCHRONIZE_CACHE;
1708                 /*
1709                  * Leave the rest of the command zero to indicate
1710                  * flush everything.
1711                  */
1712                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1713                                 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1714                 if (res == 0)
1715                         break;
1716         }
1717
1718         if (res) {
1719                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1720
1721                 if (res < 0)
1722                         return res;
1723
1724                 if (scsi_status_is_check_condition(res) &&
1725                     scsi_sense_valid(sshdr)) {
1726                         sd_print_sense_hdr(sdkp, sshdr);
1727
1728                         /* we need to evaluate the error return  */
1729                         if (sshdr->asc == 0x3a ||       /* medium not present */
1730                             sshdr->asc == 0x20 ||       /* invalid command */
1731                             (sshdr->asc == 0x74 && sshdr->ascq == 0x71))        /* drive is password locked */
1732                                 /* this is no error here */
1733                                 return 0;
1734                 }
1735
1736                 switch (host_byte(res)) {
1737                 /* ignore errors due to racing a disconnection */
1738                 case DID_BAD_TARGET:
1739                 case DID_NO_CONNECT:
1740                         return 0;
1741                 /* signal the upper layer it might try again */
1742                 case DID_BUS_BUSY:
1743                 case DID_IMM_RETRY:
1744                 case DID_REQUEUE:
1745                 case DID_SOFT_ERROR:
1746                         return -EBUSY;
1747                 default:
1748                         return -EIO;
1749                 }
1750         }
1751         return 0;
1752 }
1753
1754 static void sd_rescan(struct device *dev)
1755 {
1756         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1757
1758         sd_revalidate_disk(sdkp->disk);
1759 }
1760
1761 static char sd_pr_type(enum pr_type type)
1762 {
1763         switch (type) {
1764         case PR_WRITE_EXCLUSIVE:
1765                 return 0x01;
1766         case PR_EXCLUSIVE_ACCESS:
1767                 return 0x03;
1768         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1769                 return 0x05;
1770         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1771                 return 0x06;
1772         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1773                 return 0x07;
1774         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1775                 return 0x08;
1776         default:
1777                 return 0;
1778         }
1779 };
1780
1781 static int sd_pr_command(struct block_device *bdev, u8 sa,
1782                 u64 key, u64 sa_key, u8 type, u8 flags)
1783 {
1784         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1785         struct scsi_device *sdev = sdkp->device;
1786         struct scsi_sense_hdr sshdr;
1787         int result;
1788         u8 cmd[16] = { 0, };
1789         u8 data[24] = { 0, };
1790
1791         cmd[0] = PERSISTENT_RESERVE_OUT;
1792         cmd[1] = sa;
1793         cmd[2] = type;
1794         put_unaligned_be32(sizeof(data), &cmd[5]);
1795
1796         put_unaligned_be64(key, &data[0]);
1797         put_unaligned_be64(sa_key, &data[8]);
1798         data[20] = flags;
1799
1800         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1801                         &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1802
1803         if (scsi_status_is_check_condition(result) &&
1804             scsi_sense_valid(&sshdr)) {
1805                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1806                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1807         }
1808
1809         return result;
1810 }
1811
1812 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1813                 u32 flags)
1814 {
1815         if (flags & ~PR_FL_IGNORE_KEY)
1816                 return -EOPNOTSUPP;
1817         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1818                         old_key, new_key, 0,
1819                         (1 << 0) /* APTPL */);
1820 }
1821
1822 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1823                 u32 flags)
1824 {
1825         if (flags)
1826                 return -EOPNOTSUPP;
1827         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1828 }
1829
1830 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1831 {
1832         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1833 }
1834
1835 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1836                 enum pr_type type, bool abort)
1837 {
1838         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1839                              sd_pr_type(type), 0);
1840 }
1841
1842 static int sd_pr_clear(struct block_device *bdev, u64 key)
1843 {
1844         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1845 }
1846
1847 static const struct pr_ops sd_pr_ops = {
1848         .pr_register    = sd_pr_register,
1849         .pr_reserve     = sd_pr_reserve,
1850         .pr_release     = sd_pr_release,
1851         .pr_preempt     = sd_pr_preempt,
1852         .pr_clear       = sd_pr_clear,
1853 };
1854
1855 static const struct block_device_operations sd_fops = {
1856         .owner                  = THIS_MODULE,
1857         .open                   = sd_open,
1858         .release                = sd_release,
1859         .ioctl                  = sd_ioctl,
1860         .getgeo                 = sd_getgeo,
1861         .compat_ioctl           = blkdev_compat_ptr_ioctl,
1862         .check_events           = sd_check_events,
1863         .unlock_native_capacity = sd_unlock_native_capacity,
1864         .report_zones           = sd_zbc_report_zones,
1865         .pr_ops                 = &sd_pr_ops,
1866 };
1867
1868 /**
1869  *      sd_eh_reset - reset error handling callback
1870  *      @scmd:          sd-issued command that has failed
1871  *
1872  *      This function is called by the SCSI midlayer before starting
1873  *      SCSI EH. When counting medium access failures we have to be
1874  *      careful to register it only only once per device and SCSI EH run;
1875  *      there might be several timed out commands which will cause the
1876  *      'max_medium_access_timeouts' counter to trigger after the first
1877  *      SCSI EH run already and set the device to offline.
1878  *      So this function resets the internal counter before starting SCSI EH.
1879  **/
1880 static void sd_eh_reset(struct scsi_cmnd *scmd)
1881 {
1882         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1883
1884         /* New SCSI EH run, reset gate variable */
1885         sdkp->ignore_medium_access_errors = false;
1886 }
1887
1888 /**
1889  *      sd_eh_action - error handling callback
1890  *      @scmd:          sd-issued command that has failed
1891  *      @eh_disp:       The recovery disposition suggested by the midlayer
1892  *
1893  *      This function is called by the SCSI midlayer upon completion of an
1894  *      error test command (currently TEST UNIT READY). The result of sending
1895  *      the eh command is passed in eh_disp.  We're looking for devices that
1896  *      fail medium access commands but are OK with non access commands like
1897  *      test unit ready (so wrongly see the device as having a successful
1898  *      recovery)
1899  **/
1900 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1901 {
1902         struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->rq_disk);
1903         struct scsi_device *sdev = scmd->device;
1904
1905         if (!scsi_device_online(sdev) ||
1906             !scsi_medium_access_command(scmd) ||
1907             host_byte(scmd->result) != DID_TIME_OUT ||
1908             eh_disp != SUCCESS)
1909                 return eh_disp;
1910
1911         /*
1912          * The device has timed out executing a medium access command.
1913          * However, the TEST UNIT READY command sent during error
1914          * handling completed successfully. Either the device is in the
1915          * process of recovering or has it suffered an internal failure
1916          * that prevents access to the storage medium.
1917          */
1918         if (!sdkp->ignore_medium_access_errors) {
1919                 sdkp->medium_access_timed_out++;
1920                 sdkp->ignore_medium_access_errors = true;
1921         }
1922
1923         /*
1924          * If the device keeps failing read/write commands but TEST UNIT
1925          * READY always completes successfully we assume that medium
1926          * access is no longer possible and take the device offline.
1927          */
1928         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1929                 scmd_printk(KERN_ERR, scmd,
1930                             "Medium access timeout failure. Offlining disk!\n");
1931                 mutex_lock(&sdev->state_mutex);
1932                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1933                 mutex_unlock(&sdev->state_mutex);
1934
1935                 return SUCCESS;
1936         }
1937
1938         return eh_disp;
1939 }
1940
1941 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1942 {
1943         struct request *req = scsi_cmd_to_rq(scmd);
1944         struct scsi_device *sdev = scmd->device;
1945         unsigned int transferred, good_bytes;
1946         u64 start_lba, end_lba, bad_lba;
1947
1948         /*
1949          * Some commands have a payload smaller than the device logical
1950          * block size (e.g. INQUIRY on a 4K disk).
1951          */
1952         if (scsi_bufflen(scmd) <= sdev->sector_size)
1953                 return 0;
1954
1955         /* Check if we have a 'bad_lba' information */
1956         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1957                                      SCSI_SENSE_BUFFERSIZE,
1958                                      &bad_lba))
1959                 return 0;
1960
1961         /*
1962          * If the bad lba was reported incorrectly, we have no idea where
1963          * the error is.
1964          */
1965         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1966         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1967         if (bad_lba < start_lba || bad_lba >= end_lba)
1968                 return 0;
1969
1970         /*
1971          * resid is optional but mostly filled in.  When it's unused,
1972          * its value is zero, so we assume the whole buffer transferred
1973          */
1974         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1975
1976         /* This computation should always be done in terms of the
1977          * resolution of the device's medium.
1978          */
1979         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1980
1981         return min(good_bytes, transferred);
1982 }
1983
1984 /**
1985  *      sd_done - bottom half handler: called when the lower level
1986  *      driver has completed (successfully or otherwise) a scsi command.
1987  *      @SCpnt: mid-level's per command structure.
1988  *
1989  *      Note: potentially run from within an ISR. Must not block.
1990  **/
1991 static int sd_done(struct scsi_cmnd *SCpnt)
1992 {
1993         int result = SCpnt->result;
1994         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1995         unsigned int sector_size = SCpnt->device->sector_size;
1996         unsigned int resid;
1997         struct scsi_sense_hdr sshdr;
1998         struct request *req = scsi_cmd_to_rq(SCpnt);
1999         struct scsi_disk *sdkp = scsi_disk(req->rq_disk);
2000         int sense_valid = 0;
2001         int sense_deferred = 0;
2002
2003         switch (req_op(req)) {
2004         case REQ_OP_DISCARD:
2005         case REQ_OP_WRITE_ZEROES:
2006         case REQ_OP_WRITE_SAME:
2007         case REQ_OP_ZONE_RESET:
2008         case REQ_OP_ZONE_RESET_ALL:
2009         case REQ_OP_ZONE_OPEN:
2010         case REQ_OP_ZONE_CLOSE:
2011         case REQ_OP_ZONE_FINISH:
2012                 if (!result) {
2013                         good_bytes = blk_rq_bytes(req);
2014                         scsi_set_resid(SCpnt, 0);
2015                 } else {
2016                         good_bytes = 0;
2017                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
2018                 }
2019                 break;
2020         default:
2021                 /*
2022                  * In case of bogus fw or device, we could end up having
2023                  * an unaligned partial completion. Check this here and force
2024                  * alignment.
2025                  */
2026                 resid = scsi_get_resid(SCpnt);
2027                 if (resid & (sector_size - 1)) {
2028                         sd_printk(KERN_INFO, sdkp,
2029                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2030                                 resid, sector_size);
2031                         scsi_print_command(SCpnt);
2032                         resid = min(scsi_bufflen(SCpnt),
2033                                     round_up(resid, sector_size));
2034                         scsi_set_resid(SCpnt, resid);
2035                 }
2036         }
2037
2038         if (result) {
2039                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2040                 if (sense_valid)
2041                         sense_deferred = scsi_sense_is_deferred(&sshdr);
2042         }
2043         sdkp->medium_access_timed_out = 0;
2044
2045         if (!scsi_status_is_check_condition(result) &&
2046             (!sense_valid || sense_deferred))
2047                 goto out;
2048
2049         switch (sshdr.sense_key) {
2050         case HARDWARE_ERROR:
2051         case MEDIUM_ERROR:
2052                 good_bytes = sd_completed_bytes(SCpnt);
2053                 break;
2054         case RECOVERED_ERROR:
2055                 good_bytes = scsi_bufflen(SCpnt);
2056                 break;
2057         case NO_SENSE:
2058                 /* This indicates a false check condition, so ignore it.  An
2059                  * unknown amount of data was transferred so treat it as an
2060                  * error.
2061                  */
2062                 SCpnt->result = 0;
2063                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2064                 break;
2065         case ABORTED_COMMAND:
2066                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2067                         good_bytes = sd_completed_bytes(SCpnt);
2068                 break;
2069         case ILLEGAL_REQUEST:
2070                 switch (sshdr.asc) {
2071                 case 0x10:      /* DIX: Host detected corruption */
2072                         good_bytes = sd_completed_bytes(SCpnt);
2073                         break;
2074                 case 0x20:      /* INVALID COMMAND OPCODE */
2075                 case 0x24:      /* INVALID FIELD IN CDB */
2076                         switch (SCpnt->cmnd[0]) {
2077                         case UNMAP:
2078                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2079                                 break;
2080                         case WRITE_SAME_16:
2081                         case WRITE_SAME:
2082                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2083                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2084                                 } else {
2085                                         sdkp->device->no_write_same = 1;
2086                                         sd_config_write_same(sdkp);
2087                                         req->rq_flags |= RQF_QUIET;
2088                                 }
2089                                 break;
2090                         }
2091                 }
2092                 break;
2093         default:
2094                 break;
2095         }
2096
2097  out:
2098         if (sd_is_zoned(sdkp))
2099                 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2100
2101         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2102                                            "sd_done: completed %d of %d bytes\n",
2103                                            good_bytes, scsi_bufflen(SCpnt)));
2104
2105         return good_bytes;
2106 }
2107
2108 /*
2109  * spinup disk - called only in sd_revalidate_disk()
2110  */
2111 static void
2112 sd_spinup_disk(struct scsi_disk *sdkp)
2113 {
2114         unsigned char cmd[10];
2115         unsigned long spintime_expire = 0;
2116         int retries, spintime;
2117         unsigned int the_result;
2118         struct scsi_sense_hdr sshdr;
2119         int sense_valid = 0;
2120
2121         spintime = 0;
2122
2123         /* Spin up drives, as required.  Only do this at boot time */
2124         /* Spinup needs to be done for module loads too. */
2125         do {
2126                 retries = 0;
2127
2128                 do {
2129                         bool media_was_present = sdkp->media_present;
2130
2131                         cmd[0] = TEST_UNIT_READY;
2132                         memset((void *) &cmd[1], 0, 9);
2133
2134                         the_result = scsi_execute_req(sdkp->device, cmd,
2135                                                       DMA_NONE, NULL, 0,
2136                                                       &sshdr, SD_TIMEOUT,
2137                                                       sdkp->max_retries, NULL);
2138
2139                         /*
2140                          * If the drive has indicated to us that it
2141                          * doesn't have any media in it, don't bother
2142                          * with any more polling.
2143                          */
2144                         if (media_not_present(sdkp, &sshdr)) {
2145                                 if (media_was_present)
2146                                         sd_printk(KERN_NOTICE, sdkp, "Media removed, stopped polling\n");
2147                                 return;
2148                         }
2149
2150                         if (the_result)
2151                                 sense_valid = scsi_sense_valid(&sshdr);
2152                         retries++;
2153                 } while (retries < 3 &&
2154                          (!scsi_status_is_good(the_result) ||
2155                           (scsi_status_is_check_condition(the_result) &&
2156                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2157
2158                 if (!scsi_status_is_check_condition(the_result)) {
2159                         /* no sense, TUR either succeeded or failed
2160                          * with a status error */
2161                         if(!spintime && !scsi_status_is_good(the_result)) {
2162                                 sd_print_result(sdkp, "Test Unit Ready failed",
2163                                                 the_result);
2164                         }
2165                         break;
2166                 }
2167
2168                 /*
2169                  * The device does not want the automatic start to be issued.
2170                  */
2171                 if (sdkp->device->no_start_on_add)
2172                         break;
2173
2174                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2175                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2176                                 break;  /* manual intervention required */
2177                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2178                                 break;  /* standby */
2179                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2180                                 break;  /* unavailable */
2181                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2182                                 break;  /* sanitize in progress */
2183                         /*
2184                          * Issue command to spin up drive when not ready
2185                          */
2186                         if (!spintime) {
2187                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2188                                 cmd[0] = START_STOP;
2189                                 cmd[1] = 1;     /* Return immediately */
2190                                 memset((void *) &cmd[2], 0, 8);
2191                                 cmd[4] = 1;     /* Start spin cycle */
2192                                 if (sdkp->device->start_stop_pwr_cond)
2193                                         cmd[4] |= 1 << 4;
2194                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2195                                                  NULL, 0, &sshdr,
2196                                                  SD_TIMEOUT, sdkp->max_retries,
2197                                                  NULL);
2198                                 spintime_expire = jiffies + 100 * HZ;
2199                                 spintime = 1;
2200                         }
2201                         /* Wait 1 second for next try */
2202                         msleep(1000);
2203                         printk(KERN_CONT ".");
2204
2205                 /*
2206                  * Wait for USB flash devices with slow firmware.
2207                  * Yes, this sense key/ASC combination shouldn't
2208                  * occur here.  It's characteristic of these devices.
2209                  */
2210                 } else if (sense_valid &&
2211                                 sshdr.sense_key == UNIT_ATTENTION &&
2212                                 sshdr.asc == 0x28) {
2213                         if (!spintime) {
2214                                 spintime_expire = jiffies + 5 * HZ;
2215                                 spintime = 1;
2216                         }
2217                         /* Wait 1 second for next try */
2218                         msleep(1000);
2219                 } else {
2220                         /* we don't understand the sense code, so it's
2221                          * probably pointless to loop */
2222                         if(!spintime) {
2223                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2224                                 sd_print_sense_hdr(sdkp, &sshdr);
2225                         }
2226                         break;
2227                 }
2228
2229         } while (spintime && time_before_eq(jiffies, spintime_expire));
2230
2231         if (spintime) {
2232                 if (scsi_status_is_good(the_result))
2233                         printk(KERN_CONT "ready\n");
2234                 else
2235                         printk(KERN_CONT "not responding...\n");
2236         }
2237 }
2238
2239 /*
2240  * Determine whether disk supports Data Integrity Field.
2241  */
2242 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2243 {
2244         struct scsi_device *sdp = sdkp->device;
2245         u8 type;
2246         int ret = 0;
2247
2248         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2249                 sdkp->protection_type = 0;
2250                 return ret;
2251         }
2252
2253         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2254
2255         if (type > T10_PI_TYPE3_PROTECTION)
2256                 ret = -ENODEV;
2257         else if (scsi_host_dif_capable(sdp->host, type))
2258                 ret = 1;
2259
2260         if (sdkp->first_scan || type != sdkp->protection_type)
2261                 switch (ret) {
2262                 case -ENODEV:
2263                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2264                                   " protection type %u. Disabling disk!\n",
2265                                   type);
2266                         break;
2267                 case 1:
2268                         sd_printk(KERN_NOTICE, sdkp,
2269                                   "Enabling DIF Type %u protection\n", type);
2270                         break;
2271                 case 0:
2272                         sd_printk(KERN_NOTICE, sdkp,
2273                                   "Disabling DIF Type %u protection\n", type);
2274                         break;
2275                 }
2276
2277         sdkp->protection_type = type;
2278
2279         return ret;
2280 }
2281
2282 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2283                         struct scsi_sense_hdr *sshdr, int sense_valid,
2284                         int the_result)
2285 {
2286         if (sense_valid)
2287                 sd_print_sense_hdr(sdkp, sshdr);
2288         else
2289                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2290
2291         /*
2292          * Set dirty bit for removable devices if not ready -
2293          * sometimes drives will not report this properly.
2294          */
2295         if (sdp->removable &&
2296             sense_valid && sshdr->sense_key == NOT_READY)
2297                 set_media_not_present(sdkp);
2298
2299         /*
2300          * We used to set media_present to 0 here to indicate no media
2301          * in the drive, but some drives fail read capacity even with
2302          * media present, so we can't do that.
2303          */
2304         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2305 }
2306
2307 #define RC16_LEN 32
2308 #if RC16_LEN > SD_BUF_SIZE
2309 #error RC16_LEN must not be more than SD_BUF_SIZE
2310 #endif
2311
2312 #define READ_CAPACITY_RETRIES_ON_RESET  10
2313
2314 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2315                                                 unsigned char *buffer)
2316 {
2317         unsigned char cmd[16];
2318         struct scsi_sense_hdr sshdr;
2319         int sense_valid = 0;
2320         int the_result;
2321         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2322         unsigned int alignment;
2323         unsigned long long lba;
2324         unsigned sector_size;
2325
2326         if (sdp->no_read_capacity_16)
2327                 return -EINVAL;
2328
2329         do {
2330                 memset(cmd, 0, 16);
2331                 cmd[0] = SERVICE_ACTION_IN_16;
2332                 cmd[1] = SAI_READ_CAPACITY_16;
2333                 cmd[13] = RC16_LEN;
2334                 memset(buffer, 0, RC16_LEN);
2335
2336                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2337                                         buffer, RC16_LEN, &sshdr,
2338                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2339
2340                 if (media_not_present(sdkp, &sshdr))
2341                         return -ENODEV;
2342
2343                 if (the_result > 0) {
2344                         sense_valid = scsi_sense_valid(&sshdr);
2345                         if (sense_valid &&
2346                             sshdr.sense_key == ILLEGAL_REQUEST &&
2347                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2348                             sshdr.ascq == 0x00)
2349                                 /* Invalid Command Operation Code or
2350                                  * Invalid Field in CDB, just retry
2351                                  * silently with RC10 */
2352                                 return -EINVAL;
2353                         if (sense_valid &&
2354                             sshdr.sense_key == UNIT_ATTENTION &&
2355                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2356                                 /* Device reset might occur several times,
2357                                  * give it one more chance */
2358                                 if (--reset_retries > 0)
2359                                         continue;
2360                 }
2361                 retries--;
2362
2363         } while (the_result && retries);
2364
2365         if (the_result) {
2366                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2367                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2368                 return -EINVAL;
2369         }
2370
2371         sector_size = get_unaligned_be32(&buffer[8]);
2372         lba = get_unaligned_be64(&buffer[0]);
2373
2374         if (sd_read_protection_type(sdkp, buffer) < 0) {
2375                 sdkp->capacity = 0;
2376                 return -ENODEV;
2377         }
2378
2379         /* Logical blocks per physical block exponent */
2380         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2381
2382         /* RC basis */
2383         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2384
2385         /* Lowest aligned logical block */
2386         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2387         blk_queue_alignment_offset(sdp->request_queue, alignment);
2388         if (alignment && sdkp->first_scan)
2389                 sd_printk(KERN_NOTICE, sdkp,
2390                           "physical block alignment offset: %u\n", alignment);
2391
2392         if (buffer[14] & 0x80) { /* LBPME */
2393                 sdkp->lbpme = 1;
2394
2395                 if (buffer[14] & 0x40) /* LBPRZ */
2396                         sdkp->lbprz = 1;
2397
2398                 sd_config_discard(sdkp, SD_LBP_WS16);
2399         }
2400
2401         sdkp->capacity = lba + 1;
2402         return sector_size;
2403 }
2404
2405 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2406                                                 unsigned char *buffer)
2407 {
2408         unsigned char cmd[16];
2409         struct scsi_sense_hdr sshdr;
2410         int sense_valid = 0;
2411         int the_result;
2412         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2413         sector_t lba;
2414         unsigned sector_size;
2415
2416         do {
2417                 cmd[0] = READ_CAPACITY;
2418                 memset(&cmd[1], 0, 9);
2419                 memset(buffer, 0, 8);
2420
2421                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2422                                         buffer, 8, &sshdr,
2423                                         SD_TIMEOUT, sdkp->max_retries, NULL);
2424
2425                 if (media_not_present(sdkp, &sshdr))
2426                         return -ENODEV;
2427
2428                 if (the_result > 0) {
2429                         sense_valid = scsi_sense_valid(&sshdr);
2430                         if (sense_valid &&
2431                             sshdr.sense_key == UNIT_ATTENTION &&
2432                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2433                                 /* Device reset might occur several times,
2434                                  * give it one more chance */
2435                                 if (--reset_retries > 0)
2436                                         continue;
2437                 }
2438                 retries--;
2439
2440         } while (the_result && retries);
2441
2442         if (the_result) {
2443                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2444                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2445                 return -EINVAL;
2446         }
2447
2448         sector_size = get_unaligned_be32(&buffer[4]);
2449         lba = get_unaligned_be32(&buffer[0]);
2450
2451         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2452                 /* Some buggy (usb cardreader) devices return an lba of
2453                    0xffffffff when the want to report a size of 0 (with
2454                    which they really mean no media is present) */
2455                 sdkp->capacity = 0;
2456                 sdkp->physical_block_size = sector_size;
2457                 return sector_size;
2458         }
2459
2460         sdkp->capacity = lba + 1;
2461         sdkp->physical_block_size = sector_size;
2462         return sector_size;
2463 }
2464
2465 static int sd_try_rc16_first(struct scsi_device *sdp)
2466 {
2467         if (sdp->host->max_cmd_len < 16)
2468                 return 0;
2469         if (sdp->try_rc_10_first)
2470                 return 0;
2471         if (sdp->scsi_level > SCSI_SPC_2)
2472                 return 1;
2473         if (scsi_device_protection(sdp))
2474                 return 1;
2475         return 0;
2476 }
2477
2478 /*
2479  * read disk capacity
2480  */
2481 static void
2482 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2483 {
2484         int sector_size;
2485         struct scsi_device *sdp = sdkp->device;
2486
2487         if (sd_try_rc16_first(sdp)) {
2488                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2489                 if (sector_size == -EOVERFLOW)
2490                         goto got_data;
2491                 if (sector_size == -ENODEV)
2492                         return;
2493                 if (sector_size < 0)
2494                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2495                 if (sector_size < 0)
2496                         return;
2497         } else {
2498                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2499                 if (sector_size == -EOVERFLOW)
2500                         goto got_data;
2501                 if (sector_size < 0)
2502                         return;
2503                 if ((sizeof(sdkp->capacity) > 4) &&
2504                     (sdkp->capacity > 0xffffffffULL)) {
2505                         int old_sector_size = sector_size;
2506                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2507                                         "Trying to use READ CAPACITY(16).\n");
2508                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2509                         if (sector_size < 0) {
2510                                 sd_printk(KERN_NOTICE, sdkp,
2511                                         "Using 0xffffffff as device size\n");
2512                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2513                                 sector_size = old_sector_size;
2514                                 goto got_data;
2515                         }
2516                         /* Remember that READ CAPACITY(16) succeeded */
2517                         sdp->try_rc_10_first = 0;
2518                 }
2519         }
2520
2521         /* Some devices are known to return the total number of blocks,
2522          * not the highest block number.  Some devices have versions
2523          * which do this and others which do not.  Some devices we might
2524          * suspect of doing this but we don't know for certain.
2525          *
2526          * If we know the reported capacity is wrong, decrement it.  If
2527          * we can only guess, then assume the number of blocks is even
2528          * (usually true but not always) and err on the side of lowering
2529          * the capacity.
2530          */
2531         if (sdp->fix_capacity ||
2532             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2533                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2534                                 "from its reported value: %llu\n",
2535                                 (unsigned long long) sdkp->capacity);
2536                 --sdkp->capacity;
2537         }
2538
2539 got_data:
2540         if (sector_size == 0) {
2541                 sector_size = 512;
2542                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2543                           "assuming 512.\n");
2544         }
2545
2546         if (sector_size != 512 &&
2547             sector_size != 1024 &&
2548             sector_size != 2048 &&
2549             sector_size != 4096) {
2550                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2551                           sector_size);
2552                 /*
2553                  * The user might want to re-format the drive with
2554                  * a supported sectorsize.  Once this happens, it
2555                  * would be relatively trivial to set the thing up.
2556                  * For this reason, we leave the thing in the table.
2557                  */
2558                 sdkp->capacity = 0;
2559                 /*
2560                  * set a bogus sector size so the normal read/write
2561                  * logic in the block layer will eventually refuse any
2562                  * request on this device without tripping over power
2563                  * of two sector size assumptions
2564                  */
2565                 sector_size = 512;
2566         }
2567         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2568         blk_queue_physical_block_size(sdp->request_queue,
2569                                       sdkp->physical_block_size);
2570         sdkp->device->sector_size = sector_size;
2571
2572         if (sdkp->capacity > 0xffffffff)
2573                 sdp->use_16_for_rw = 1;
2574
2575 }
2576
2577 /*
2578  * Print disk capacity
2579  */
2580 static void
2581 sd_print_capacity(struct scsi_disk *sdkp,
2582                   sector_t old_capacity)
2583 {
2584         int sector_size = sdkp->device->sector_size;
2585         char cap_str_2[10], cap_str_10[10];
2586
2587         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2588                 return;
2589
2590         string_get_size(sdkp->capacity, sector_size,
2591                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2592         string_get_size(sdkp->capacity, sector_size,
2593                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2594
2595         sd_printk(KERN_NOTICE, sdkp,
2596                   "%llu %d-byte logical blocks: (%s/%s)\n",
2597                   (unsigned long long)sdkp->capacity,
2598                   sector_size, cap_str_10, cap_str_2);
2599
2600         if (sdkp->physical_block_size != sector_size)
2601                 sd_printk(KERN_NOTICE, sdkp,
2602                           "%u-byte physical blocks\n",
2603                           sdkp->physical_block_size);
2604 }
2605
2606 /* called with buffer of length 512 */
2607 static inline int
2608 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2609                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2610                  struct scsi_sense_hdr *sshdr)
2611 {
2612         /*
2613          * If we must use MODE SENSE(10), make sure that the buffer length
2614          * is at least 8 bytes so that the mode sense header fits.
2615          */
2616         if (sdkp->device->use_10_for_ms && len < 8)
2617                 len = 8;
2618
2619         return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2620                                SD_TIMEOUT, sdkp->max_retries, data,
2621                                sshdr);
2622 }
2623
2624 /*
2625  * read write protect setting, if possible - called only in sd_revalidate_disk()
2626  * called with buffer of length SD_BUF_SIZE
2627  */
2628 static void
2629 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2630 {
2631         int res;
2632         struct scsi_device *sdp = sdkp->device;
2633         struct scsi_mode_data data;
2634         int old_wp = sdkp->write_prot;
2635
2636         set_disk_ro(sdkp->disk, 0);
2637         if (sdp->skip_ms_page_3f) {
2638                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2639                 return;
2640         }
2641
2642         if (sdp->use_192_bytes_for_3f) {
2643                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2644         } else {
2645                 /*
2646                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2647                  * We have to start carefully: some devices hang if we ask
2648                  * for more than is available.
2649                  */
2650                 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2651
2652                 /*
2653                  * Second attempt: ask for page 0 When only page 0 is
2654                  * implemented, a request for page 3F may return Sense Key
2655                  * 5: Illegal Request, Sense Code 24: Invalid field in
2656                  * CDB.
2657                  */
2658                 if (res < 0)
2659                         res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2660
2661                 /*
2662                  * Third attempt: ask 255 bytes, as we did earlier.
2663                  */
2664                 if (res < 0)
2665                         res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2666                                                &data, NULL);
2667         }
2668
2669         if (res < 0) {
2670                 sd_first_printk(KERN_WARNING, sdkp,
2671                           "Test WP failed, assume Write Enabled\n");
2672         } else {
2673                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2674                 set_disk_ro(sdkp->disk, sdkp->write_prot);
2675                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2676                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2677                                   sdkp->write_prot ? "on" : "off");
2678                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2679                 }
2680         }
2681 }
2682
2683 /*
2684  * sd_read_cache_type - called only from sd_revalidate_disk()
2685  * called with buffer of length SD_BUF_SIZE
2686  */
2687 static void
2688 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2689 {
2690         int len = 0, res;
2691         struct scsi_device *sdp = sdkp->device;
2692
2693         int dbd;
2694         int modepage;
2695         int first_len;
2696         struct scsi_mode_data data;
2697         struct scsi_sense_hdr sshdr;
2698         int old_wce = sdkp->WCE;
2699         int old_rcd = sdkp->RCD;
2700         int old_dpofua = sdkp->DPOFUA;
2701
2702
2703         if (sdkp->cache_override)
2704                 return;
2705
2706         first_len = 4;
2707         if (sdp->skip_ms_page_8) {
2708                 if (sdp->type == TYPE_RBC)
2709                         goto defaults;
2710                 else {
2711                         if (sdp->skip_ms_page_3f)
2712                                 goto defaults;
2713                         modepage = 0x3F;
2714                         if (sdp->use_192_bytes_for_3f)
2715                                 first_len = 192;
2716                         dbd = 0;
2717                 }
2718         } else if (sdp->type == TYPE_RBC) {
2719                 modepage = 6;
2720                 dbd = 8;
2721         } else {
2722                 modepage = 8;
2723                 dbd = 0;
2724         }
2725
2726         /* cautiously ask */
2727         res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2728                         &data, &sshdr);
2729
2730         if (res < 0)
2731                 goto bad_sense;
2732
2733         if (!data.header_length) {
2734                 modepage = 6;
2735                 first_len = 0;
2736                 sd_first_printk(KERN_ERR, sdkp,
2737                                 "Missing header in MODE_SENSE response\n");
2738         }
2739
2740         /* that went OK, now ask for the proper length */
2741         len = data.length;
2742
2743         /*
2744          * We're only interested in the first three bytes, actually.
2745          * But the data cache page is defined for the first 20.
2746          */
2747         if (len < 3)
2748                 goto bad_sense;
2749         else if (len > SD_BUF_SIZE) {
2750                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2751                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2752                 len = SD_BUF_SIZE;
2753         }
2754         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2755                 len = 192;
2756
2757         /* Get the data */
2758         if (len > first_len)
2759                 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2760                                 &data, &sshdr);
2761
2762         if (!res) {
2763                 int offset = data.header_length + data.block_descriptor_length;
2764
2765                 while (offset < len) {
2766                         u8 page_code = buffer[offset] & 0x3F;
2767                         u8 spf       = buffer[offset] & 0x40;
2768
2769                         if (page_code == 8 || page_code == 6) {
2770                                 /* We're interested only in the first 3 bytes.
2771                                  */
2772                                 if (len - offset <= 2) {
2773                                         sd_first_printk(KERN_ERR, sdkp,
2774                                                 "Incomplete mode parameter "
2775                                                         "data\n");
2776                                         goto defaults;
2777                                 } else {
2778                                         modepage = page_code;
2779                                         goto Page_found;
2780                                 }
2781                         } else {
2782                                 /* Go to the next page */
2783                                 if (spf && len - offset > 3)
2784                                         offset += 4 + (buffer[offset+2] << 8) +
2785                                                 buffer[offset+3];
2786                                 else if (!spf && len - offset > 1)
2787                                         offset += 2 + buffer[offset+1];
2788                                 else {
2789                                         sd_first_printk(KERN_ERR, sdkp,
2790                                                         "Incomplete mode "
2791                                                         "parameter data\n");
2792                                         goto defaults;
2793                                 }
2794                         }
2795                 }
2796
2797                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2798                 goto defaults;
2799
2800         Page_found:
2801                 if (modepage == 8) {
2802                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2803                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2804                 } else {
2805                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2806                         sdkp->RCD = 0;
2807                 }
2808
2809                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2810                 if (sdp->broken_fua) {
2811                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2812                         sdkp->DPOFUA = 0;
2813                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2814                            !sdkp->device->use_16_for_rw) {
2815                         sd_first_printk(KERN_NOTICE, sdkp,
2816                                   "Uses READ/WRITE(6), disabling FUA\n");
2817                         sdkp->DPOFUA = 0;
2818                 }
2819
2820                 /* No cache flush allowed for write protected devices */
2821                 if (sdkp->WCE && sdkp->write_prot)
2822                         sdkp->WCE = 0;
2823
2824                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2825                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2826                         sd_printk(KERN_NOTICE, sdkp,
2827                                   "Write cache: %s, read cache: %s, %s\n",
2828                                   sdkp->WCE ? "enabled" : "disabled",
2829                                   sdkp->RCD ? "disabled" : "enabled",
2830                                   sdkp->DPOFUA ? "supports DPO and FUA"
2831                                   : "doesn't support DPO or FUA");
2832
2833                 return;
2834         }
2835
2836 bad_sense:
2837         if (scsi_sense_valid(&sshdr) &&
2838             sshdr.sense_key == ILLEGAL_REQUEST &&
2839             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2840                 /* Invalid field in CDB */
2841                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2842         else
2843                 sd_first_printk(KERN_ERR, sdkp,
2844                                 "Asking for cache data failed\n");
2845
2846 defaults:
2847         if (sdp->wce_default_on) {
2848                 sd_first_printk(KERN_NOTICE, sdkp,
2849                                 "Assuming drive cache: write back\n");
2850                 sdkp->WCE = 1;
2851         } else {
2852                 sd_first_printk(KERN_ERR, sdkp,
2853                                 "Assuming drive cache: write through\n");
2854                 sdkp->WCE = 0;
2855         }
2856         sdkp->RCD = 0;
2857         sdkp->DPOFUA = 0;
2858 }
2859
2860 /*
2861  * The ATO bit indicates whether the DIF application tag is available
2862  * for use by the operating system.
2863  */
2864 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2865 {
2866         int res, offset;
2867         struct scsi_device *sdp = sdkp->device;
2868         struct scsi_mode_data data;
2869         struct scsi_sense_hdr sshdr;
2870
2871         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2872                 return;
2873
2874         if (sdkp->protection_type == 0)
2875                 return;
2876
2877         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2878                               sdkp->max_retries, &data, &sshdr);
2879
2880         if (res < 0 || !data.header_length ||
2881             data.length < 6) {
2882                 sd_first_printk(KERN_WARNING, sdkp,
2883                           "getting Control mode page failed, assume no ATO\n");
2884
2885                 if (scsi_sense_valid(&sshdr))
2886                         sd_print_sense_hdr(sdkp, &sshdr);
2887
2888                 return;
2889         }
2890
2891         offset = data.header_length + data.block_descriptor_length;
2892
2893         if ((buffer[offset] & 0x3f) != 0x0a) {
2894                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2895                 return;
2896         }
2897
2898         if ((buffer[offset + 5] & 0x80) == 0)
2899                 return;
2900
2901         sdkp->ATO = 1;
2902
2903         return;
2904 }
2905
2906 /**
2907  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2908  * @sdkp: disk to query
2909  */
2910 static void sd_read_block_limits(struct scsi_disk *sdkp)
2911 {
2912         unsigned int sector_sz = sdkp->device->sector_size;
2913         const int vpd_len = 64;
2914         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2915
2916         if (!buffer ||
2917             /* Block Limits VPD */
2918             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2919                 goto out;
2920
2921         blk_queue_io_min(sdkp->disk->queue,
2922                          get_unaligned_be16(&buffer[6]) * sector_sz);
2923
2924         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2925         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2926
2927         if (buffer[3] == 0x3c) {
2928                 unsigned int lba_count, desc_count;
2929
2930                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2931
2932                 if (!sdkp->lbpme)
2933                         goto out;
2934
2935                 lba_count = get_unaligned_be32(&buffer[20]);
2936                 desc_count = get_unaligned_be32(&buffer[24]);
2937
2938                 if (lba_count && desc_count)
2939                         sdkp->max_unmap_blocks = lba_count;
2940
2941                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2942
2943                 if (buffer[32] & 0x80)
2944                         sdkp->unmap_alignment =
2945                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2946
2947                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2948
2949                         if (sdkp->max_unmap_blocks)
2950                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2951                         else
2952                                 sd_config_discard(sdkp, SD_LBP_WS16);
2953
2954                 } else {        /* LBP VPD page tells us what to use */
2955                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2956                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2957                         else if (sdkp->lbpws)
2958                                 sd_config_discard(sdkp, SD_LBP_WS16);
2959                         else if (sdkp->lbpws10)
2960                                 sd_config_discard(sdkp, SD_LBP_WS10);
2961                         else
2962                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2963                 }
2964         }
2965
2966  out:
2967         kfree(buffer);
2968 }
2969
2970 /**
2971  * sd_read_block_characteristics - Query block dev. characteristics
2972  * @sdkp: disk to query
2973  */
2974 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2975 {
2976         struct request_queue *q = sdkp->disk->queue;
2977         unsigned char *buffer;
2978         u16 rot;
2979         const int vpd_len = 64;
2980
2981         buffer = kmalloc(vpd_len, GFP_KERNEL);
2982
2983         if (!buffer ||
2984             /* Block Device Characteristics VPD */
2985             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2986                 goto out;
2987
2988         rot = get_unaligned_be16(&buffer[4]);
2989
2990         if (rot == 1) {
2991                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2992                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2993         }
2994
2995         if (sdkp->device->type == TYPE_ZBC) {
2996                 /* Host-managed */
2997                 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
2998         } else {
2999                 sdkp->zoned = (buffer[8] >> 4) & 3;
3000                 if (sdkp->zoned == 1) {
3001                         /* Host-aware */
3002                         blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3003                 } else {
3004                         /* Regular disk or drive managed disk */
3005                         blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3006                 }
3007         }
3008
3009         if (!sdkp->first_scan)
3010                 goto out;
3011
3012         if (blk_queue_is_zoned(q)) {
3013                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3014                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3015         } else {
3016                 if (sdkp->zoned == 1)
3017                         sd_printk(KERN_NOTICE, sdkp,
3018                                   "Host-aware SMR disk used as regular disk\n");
3019                 else if (sdkp->zoned == 2)
3020                         sd_printk(KERN_NOTICE, sdkp,
3021                                   "Drive-managed SMR disk\n");
3022         }
3023
3024  out:
3025         kfree(buffer);
3026 }
3027
3028 /**
3029  * sd_read_block_provisioning - Query provisioning VPD page
3030  * @sdkp: disk to query
3031  */
3032 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3033 {
3034         unsigned char *buffer;
3035         const int vpd_len = 8;
3036
3037         if (sdkp->lbpme == 0)
3038                 return;
3039
3040         buffer = kmalloc(vpd_len, GFP_KERNEL);
3041
3042         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3043                 goto out;
3044
3045         sdkp->lbpvpd    = 1;
3046         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3047         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3048         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3049
3050  out:
3051         kfree(buffer);
3052 }
3053
3054 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3055 {
3056         struct scsi_device *sdev = sdkp->device;
3057
3058         if (sdev->host->no_write_same) {
3059                 sdev->no_write_same = 1;
3060
3061                 return;
3062         }
3063
3064         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3065                 /* too large values might cause issues with arcmsr */
3066                 int vpd_buf_len = 64;
3067
3068                 sdev->no_report_opcodes = 1;
3069
3070                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3071                  * CODES is unsupported and the device has an ATA
3072                  * Information VPD page (SAT).
3073                  */
3074                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3075                         sdev->no_write_same = 1;
3076         }
3077
3078         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3079                 sdkp->ws16 = 1;
3080
3081         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3082                 sdkp->ws10 = 1;
3083 }
3084
3085 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3086 {
3087         struct scsi_device *sdev = sdkp->device;
3088
3089         if (!sdev->security_supported)
3090                 return;
3091
3092         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3093                         SECURITY_PROTOCOL_IN) == 1 &&
3094             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3095                         SECURITY_PROTOCOL_OUT) == 1)
3096                 sdkp->security = 1;
3097 }
3098
3099 /*
3100  * Determine the device's preferred I/O size for reads and writes
3101  * unless the reported value is unreasonably small, large, not a
3102  * multiple of the physical block size, or simply garbage.
3103  */
3104 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3105                                       unsigned int dev_max)
3106 {
3107         struct scsi_device *sdp = sdkp->device;
3108         unsigned int opt_xfer_bytes =
3109                 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3110
3111         if (sdkp->opt_xfer_blocks == 0)
3112                 return false;
3113
3114         if (sdkp->opt_xfer_blocks > dev_max) {
3115                 sd_first_printk(KERN_WARNING, sdkp,
3116                                 "Optimal transfer size %u logical blocks " \
3117                                 "> dev_max (%u logical blocks)\n",
3118                                 sdkp->opt_xfer_blocks, dev_max);
3119                 return false;
3120         }
3121
3122         if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3123                 sd_first_printk(KERN_WARNING, sdkp,
3124                                 "Optimal transfer size %u logical blocks " \
3125                                 "> sd driver limit (%u logical blocks)\n",
3126                                 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3127                 return false;
3128         }
3129
3130         if (opt_xfer_bytes < PAGE_SIZE) {
3131                 sd_first_printk(KERN_WARNING, sdkp,
3132                                 "Optimal transfer size %u bytes < " \
3133                                 "PAGE_SIZE (%u bytes)\n",
3134                                 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3135                 return false;
3136         }
3137
3138         if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3139                 sd_first_printk(KERN_WARNING, sdkp,
3140                                 "Optimal transfer size %u bytes not a " \
3141                                 "multiple of physical block size (%u bytes)\n",
3142                                 opt_xfer_bytes, sdkp->physical_block_size);
3143                 return false;
3144         }
3145
3146         sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3147                         opt_xfer_bytes);
3148         return true;
3149 }
3150
3151 /**
3152  *      sd_revalidate_disk - called the first time a new disk is seen,
3153  *      performs disk spin up, read_capacity, etc.
3154  *      @disk: struct gendisk we care about
3155  **/
3156 static int sd_revalidate_disk(struct gendisk *disk)
3157 {
3158         struct scsi_disk *sdkp = scsi_disk(disk);
3159         struct scsi_device *sdp = sdkp->device;
3160         struct request_queue *q = sdkp->disk->queue;
3161         sector_t old_capacity = sdkp->capacity;
3162         unsigned char *buffer;
3163         unsigned int dev_max, rw_max;
3164
3165         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3166                                       "sd_revalidate_disk\n"));
3167
3168         /*
3169          * If the device is offline, don't try and read capacity or any
3170          * of the other niceties.
3171          */
3172         if (!scsi_device_online(sdp))
3173                 goto out;
3174
3175         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3176         if (!buffer) {
3177                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3178                           "allocation failure.\n");
3179                 goto out;
3180         }
3181
3182         sd_spinup_disk(sdkp);
3183
3184         /*
3185          * Without media there is no reason to ask; moreover, some devices
3186          * react badly if we do.
3187          */
3188         if (sdkp->media_present) {
3189                 sd_read_capacity(sdkp, buffer);
3190
3191                 /*
3192                  * set the default to rotational.  All non-rotational devices
3193                  * support the block characteristics VPD page, which will
3194                  * cause this to be updated correctly and any device which
3195                  * doesn't support it should be treated as rotational.
3196                  */
3197                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3198                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3199
3200                 if (scsi_device_supports_vpd(sdp)) {
3201                         sd_read_block_provisioning(sdkp);
3202                         sd_read_block_limits(sdkp);
3203                         sd_read_block_characteristics(sdkp);
3204                         sd_zbc_read_zones(sdkp, buffer);
3205                 }
3206
3207                 sd_print_capacity(sdkp, old_capacity);
3208
3209                 sd_read_write_protect_flag(sdkp, buffer);
3210                 sd_read_cache_type(sdkp, buffer);
3211                 sd_read_app_tag_own(sdkp, buffer);
3212                 sd_read_write_same(sdkp, buffer);
3213                 sd_read_security(sdkp, buffer);
3214         }
3215
3216         /*
3217          * We now have all cache related info, determine how we deal
3218          * with flush requests.
3219          */
3220         sd_set_flush_flag(sdkp);
3221
3222         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3223         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3224
3225         /* Some devices report a maximum block count for READ/WRITE requests. */
3226         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3227         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3228
3229         if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3230                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3231                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3232         } else {
3233                 q->limits.io_opt = 0;
3234                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3235                                       (sector_t)BLK_DEF_MAX_SECTORS);
3236         }
3237
3238         /* Do not exceed controller limit */
3239         rw_max = min(rw_max, queue_max_hw_sectors(q));
3240
3241         /*
3242          * Only update max_sectors if previously unset or if the current value
3243          * exceeds the capabilities of the hardware.
3244          */
3245         if (sdkp->first_scan ||
3246             q->limits.max_sectors > q->limits.max_dev_sectors ||
3247             q->limits.max_sectors > q->limits.max_hw_sectors)
3248                 q->limits.max_sectors = rw_max;
3249
3250         sdkp->first_scan = 0;
3251
3252         set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3253         sd_config_write_same(sdkp);
3254         kfree(buffer);
3255
3256         /*
3257          * For a zoned drive, revalidating the zones can be done only once
3258          * the gendisk capacity is set. So if this fails, set back the gendisk
3259          * capacity to 0.
3260          */
3261         if (sd_zbc_revalidate_zones(sdkp))
3262                 set_capacity_and_notify(disk, 0);
3263
3264  out:
3265         return 0;
3266 }
3267
3268 /**
3269  *      sd_unlock_native_capacity - unlock native capacity
3270  *      @disk: struct gendisk to set capacity for
3271  *
3272  *      Block layer calls this function if it detects that partitions
3273  *      on @disk reach beyond the end of the device.  If the SCSI host
3274  *      implements ->unlock_native_capacity() method, it's invoked to
3275  *      give it a chance to adjust the device capacity.
3276  *
3277  *      CONTEXT:
3278  *      Defined by block layer.  Might sleep.
3279  */
3280 static void sd_unlock_native_capacity(struct gendisk *disk)
3281 {
3282         struct scsi_device *sdev = scsi_disk(disk)->device;
3283
3284         if (sdev->host->hostt->unlock_native_capacity)
3285                 sdev->host->hostt->unlock_native_capacity(sdev);
3286 }
3287
3288 /**
3289  *      sd_format_disk_name - format disk name
3290  *      @prefix: name prefix - ie. "sd" for SCSI disks
3291  *      @index: index of the disk to format name for
3292  *      @buf: output buffer
3293  *      @buflen: length of the output buffer
3294  *
3295  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3296  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3297  *      which is followed by sdaaa.
3298  *
3299  *      This is basically 26 base counting with one extra 'nil' entry
3300  *      at the beginning from the second digit on and can be
3301  *      determined using similar method as 26 base conversion with the
3302  *      index shifted -1 after each digit is computed.
3303  *
3304  *      CONTEXT:
3305  *      Don't care.
3306  *
3307  *      RETURNS:
3308  *      0 on success, -errno on failure.
3309  */
3310 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3311 {
3312         const int base = 'z' - 'a' + 1;
3313         char *begin = buf + strlen(prefix);
3314         char *end = buf + buflen;
3315         char *p;
3316         int unit;
3317
3318         p = end - 1;
3319         *p = '\0';
3320         unit = base;
3321         do {
3322                 if (p == begin)
3323                         return -EINVAL;
3324                 *--p = 'a' + (index % unit);
3325                 index = (index / unit) - 1;
3326         } while (index >= 0);
3327
3328         memmove(begin, p, end - p);
3329         memcpy(buf, prefix, strlen(prefix));
3330
3331         return 0;
3332 }
3333
3334 /**
3335  *      sd_probe - called during driver initialization and whenever a
3336  *      new scsi device is attached to the system. It is called once
3337  *      for each scsi device (not just disks) present.
3338  *      @dev: pointer to device object
3339  *
3340  *      Returns 0 if successful (or not interested in this scsi device
3341  *      (e.g. scanner)); 1 when there is an error.
3342  *
3343  *      Note: this function is invoked from the scsi mid-level.
3344  *      This function sets up the mapping between a given
3345  *      <host,channel,id,lun> (found in sdp) and new device name
3346  *      (e.g. /dev/sda). More precisely it is the block device major
3347  *      and minor number that is chosen here.
3348  *
3349  *      Assume sd_probe is not re-entrant (for time being)
3350  *      Also think about sd_probe() and sd_remove() running coincidentally.
3351  **/
3352 static int sd_probe(struct device *dev)
3353 {
3354         struct scsi_device *sdp = to_scsi_device(dev);
3355         struct scsi_disk *sdkp;
3356         struct gendisk *gd;
3357         int index;
3358         int error;
3359
3360         scsi_autopm_get_device(sdp);
3361         error = -ENODEV;
3362         if (sdp->type != TYPE_DISK &&
3363             sdp->type != TYPE_ZBC &&
3364             sdp->type != TYPE_MOD &&
3365             sdp->type != TYPE_RBC)
3366                 goto out;
3367
3368         if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3369                 sdev_printk(KERN_WARNING, sdp,
3370                             "Unsupported ZBC host-managed device.\n");
3371                 goto out;
3372         }
3373
3374         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3375                                         "sd_probe\n"));
3376
3377         error = -ENOMEM;
3378         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3379         if (!sdkp)
3380                 goto out;
3381
3382         gd = __alloc_disk_node(sdp->request_queue, NUMA_NO_NODE,
3383                                &sd_bio_compl_lkclass);
3384         if (!gd)
3385                 goto out_free;
3386
3387         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3388         if (index < 0) {
3389                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3390                 goto out_put;
3391         }
3392
3393         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3394         if (error) {
3395                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3396                 goto out_free_index;
3397         }
3398
3399         sdkp->device = sdp;
3400         sdkp->driver = &sd_template;
3401         sdkp->disk = gd;
3402         sdkp->index = index;
3403         sdkp->max_retries = SD_MAX_RETRIES;
3404         atomic_set(&sdkp->openers, 0);
3405         atomic_set(&sdkp->device->ioerr_cnt, 0);
3406
3407         if (!sdp->request_queue->rq_timeout) {
3408                 if (sdp->type != TYPE_MOD)
3409                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3410                 else
3411                         blk_queue_rq_timeout(sdp->request_queue,
3412                                              SD_MOD_TIMEOUT);
3413         }
3414
3415         device_initialize(&sdkp->dev);
3416         sdkp->dev.parent = get_device(dev);
3417         sdkp->dev.class = &sd_disk_class;
3418         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3419
3420         error = device_add(&sdkp->dev);
3421         if (error) {
3422                 put_device(&sdkp->dev);
3423                 goto out;
3424         }
3425
3426         dev_set_drvdata(dev, sdkp);
3427
3428         gd->major = sd_major((index & 0xf0) >> 4);
3429         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3430         gd->minors = SD_MINORS;
3431
3432         gd->fops = &sd_fops;
3433         gd->private_data = &sdkp->driver;
3434
3435         /* defaults, until the device tells us otherwise */
3436         sdp->sector_size = 512;
3437         sdkp->capacity = 0;
3438         sdkp->media_present = 1;
3439         sdkp->write_prot = 0;
3440         sdkp->cache_override = 0;
3441         sdkp->WCE = 0;
3442         sdkp->RCD = 0;
3443         sdkp->ATO = 0;
3444         sdkp->first_scan = 1;
3445         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3446
3447         sd_revalidate_disk(gd);
3448
3449         gd->flags = GENHD_FL_EXT_DEVT;
3450         if (sdp->removable) {
3451                 gd->flags |= GENHD_FL_REMOVABLE;
3452                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3453                 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3454         }
3455
3456         blk_pm_runtime_init(sdp->request_queue, dev);
3457         if (sdp->rpm_autosuspend) {
3458                 pm_runtime_set_autosuspend_delay(dev,
3459                         sdp->host->hostt->rpm_autosuspend_delay);
3460         }
3461         device_add_disk(dev, gd, NULL);
3462         if (sdkp->capacity)
3463                 sd_dif_config_host(sdkp);
3464
3465         sd_revalidate_disk(gd);
3466
3467         if (sdkp->security) {
3468                 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3469                 if (sdkp->opal_dev)
3470                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3471         }
3472
3473         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3474                   sdp->removable ? "removable " : "");
3475         scsi_autopm_put_device(sdp);
3476
3477         return 0;
3478
3479  out_free_index:
3480         ida_free(&sd_index_ida, index);
3481  out_put:
3482         put_disk(gd);
3483  out_free:
3484         kfree(sdkp);
3485  out:
3486         scsi_autopm_put_device(sdp);
3487         return error;
3488 }
3489
3490 /**
3491  *      sd_remove - called whenever a scsi disk (previously recognized by
3492  *      sd_probe) is detached from the system. It is called (potentially
3493  *      multiple times) during sd module unload.
3494  *      @dev: pointer to device object
3495  *
3496  *      Note: this function is invoked from the scsi mid-level.
3497  *      This function potentially frees up a device name (e.g. /dev/sdc)
3498  *      that could be re-used by a subsequent sd_probe().
3499  *      This function is not called when the built-in sd driver is "exit-ed".
3500  **/
3501 static int sd_remove(struct device *dev)
3502 {
3503         struct scsi_disk *sdkp;
3504
3505         sdkp = dev_get_drvdata(dev);
3506         scsi_autopm_get_device(sdkp->device);
3507
3508         async_synchronize_full_domain(&scsi_sd_pm_domain);
3509         device_del(&sdkp->dev);
3510         del_gendisk(sdkp->disk);
3511         sd_shutdown(dev);
3512
3513         free_opal_dev(sdkp->opal_dev);
3514
3515         mutex_lock(&sd_ref_mutex);
3516         dev_set_drvdata(dev, NULL);
3517         put_device(&sdkp->dev);
3518         mutex_unlock(&sd_ref_mutex);
3519
3520         return 0;
3521 }
3522
3523 /**
3524  *      scsi_disk_release - Called to free the scsi_disk structure
3525  *      @dev: pointer to embedded class device
3526  *
3527  *      sd_ref_mutex must be held entering this routine.  Because it is
3528  *      called on last put, you should always use the scsi_disk_get()
3529  *      scsi_disk_put() helpers which manipulate the semaphore directly
3530  *      and never do a direct put_device.
3531  **/
3532 static void scsi_disk_release(struct device *dev)
3533 {
3534         struct scsi_disk *sdkp = to_scsi_disk(dev);
3535         struct gendisk *disk = sdkp->disk;
3536         struct request_queue *q = disk->queue;
3537
3538         ida_free(&sd_index_ida, sdkp->index);
3539
3540         /*
3541          * Wait until all requests that are in progress have completed.
3542          * This is necessary to avoid that e.g. scsi_end_request() crashes
3543          * due to clearing the disk->private_data pointer. Wait from inside
3544          * scsi_disk_release() instead of from sd_release() to avoid that
3545          * freezing and unfreezing the request queue affects user space I/O
3546          * in case multiple processes open a /dev/sd... node concurrently.
3547          */
3548         blk_mq_freeze_queue(q);
3549         blk_mq_unfreeze_queue(q);
3550
3551         disk->private_data = NULL;
3552         put_disk(disk);
3553         put_device(&sdkp->device->sdev_gendev);
3554
3555         sd_zbc_release_disk(sdkp);
3556
3557         kfree(sdkp);
3558 }
3559
3560 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3561 {
3562         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3563         struct scsi_sense_hdr sshdr;
3564         struct scsi_device *sdp = sdkp->device;
3565         int res;
3566
3567         if (start)
3568                 cmd[4] |= 1;    /* START */
3569
3570         if (sdp->start_stop_pwr_cond)
3571                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3572
3573         if (!scsi_device_online(sdp))
3574                 return -ENODEV;
3575
3576         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3577                         SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3578         if (res) {
3579                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3580                 if (res > 0 && scsi_sense_valid(&sshdr)) {
3581                         sd_print_sense_hdr(sdkp, &sshdr);
3582                         /* 0x3a is medium not present */
3583                         if (sshdr.asc == 0x3a)
3584                                 res = 0;
3585                 }
3586         }
3587
3588         /* SCSI error codes must not go to the generic layer */
3589         if (res)
3590                 return -EIO;
3591
3592         return 0;
3593 }
3594
3595 /*
3596  * Send a SYNCHRONIZE CACHE instruction down to the device through
3597  * the normal SCSI command structure.  Wait for the command to
3598  * complete.
3599  */
3600 static void sd_shutdown(struct device *dev)
3601 {
3602         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3603
3604         if (!sdkp)
3605                 return;         /* this can happen */
3606
3607         if (pm_runtime_suspended(dev))
3608                 return;
3609
3610         if (sdkp->WCE && sdkp->media_present) {
3611                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3612                 sd_sync_cache(sdkp, NULL);
3613         }
3614
3615         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3616                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3617                 sd_start_stop_device(sdkp, 0);
3618         }
3619 }
3620
3621 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3622 {
3623         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3624         struct scsi_sense_hdr sshdr;
3625         int ret = 0;
3626
3627         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3628                 return 0;
3629
3630         if (sdkp->WCE && sdkp->media_present) {
3631                 if (!sdkp->device->silence_suspend)
3632                         sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3633                 ret = sd_sync_cache(sdkp, &sshdr);
3634
3635                 if (ret) {
3636                         /* ignore OFFLINE device */
3637                         if (ret == -ENODEV)
3638                                 return 0;
3639
3640                         if (!scsi_sense_valid(&sshdr) ||
3641                             sshdr.sense_key != ILLEGAL_REQUEST)
3642                                 return ret;
3643
3644                         /*
3645                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3646                          * doesn't support sync. There's not much to do and
3647                          * suspend shouldn't fail.
3648                          */
3649                         ret = 0;
3650                 }
3651         }
3652
3653         if (sdkp->device->manage_start_stop) {
3654                 if (!sdkp->device->silence_suspend)
3655                         sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3656                 /* an error is not worth aborting a system sleep */
3657                 ret = sd_start_stop_device(sdkp, 0);
3658                 if (ignore_stop_errors)
3659                         ret = 0;
3660         }
3661
3662         return ret;
3663 }
3664
3665 static int sd_suspend_system(struct device *dev)
3666 {
3667         return sd_suspend_common(dev, true);
3668 }
3669
3670 static int sd_suspend_runtime(struct device *dev)
3671 {
3672         return sd_suspend_common(dev, false);
3673 }
3674
3675 static int sd_resume(struct device *dev)
3676 {
3677         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3678         int ret;
3679
3680         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3681                 return 0;
3682
3683         if (!sdkp->device->manage_start_stop)
3684                 return 0;
3685
3686         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3687         ret = sd_start_stop_device(sdkp, 1);
3688         if (!ret)
3689                 opal_unlock_from_suspend(sdkp->opal_dev);
3690         return ret;
3691 }
3692
3693 static int sd_resume_runtime(struct device *dev)
3694 {
3695         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3696         struct scsi_device *sdp;
3697
3698         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3699                 return 0;
3700
3701         sdp = sdkp->device;
3702
3703         if (sdp->ignore_media_change) {
3704                 /* clear the device's sense data */
3705                 static const u8 cmd[10] = { REQUEST_SENSE };
3706
3707                 if (scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL,
3708                                  NULL, sdp->request_queue->rq_timeout, 1, 0,
3709                                  RQF_PM, NULL))
3710                         sd_printk(KERN_NOTICE, sdkp,
3711                                   "Failed to clear sense data\n");
3712         }
3713
3714         return sd_resume(dev);
3715 }
3716
3717 /**
3718  *      init_sd - entry point for this driver (both when built in or when
3719  *      a module).
3720  *
3721  *      Note: this function registers this driver with the scsi mid-level.
3722  **/
3723 static int __init init_sd(void)
3724 {
3725         int majors = 0, i, err;
3726
3727         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3728
3729         for (i = 0; i < SD_MAJORS; i++) {
3730                 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3731                         continue;
3732                 majors++;
3733         }
3734
3735         if (!majors)
3736                 return -ENODEV;
3737
3738         err = class_register(&sd_disk_class);
3739         if (err)
3740                 goto err_out;
3741
3742         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3743                                          0, 0, NULL);
3744         if (!sd_cdb_cache) {
3745                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3746                 err = -ENOMEM;
3747                 goto err_out_class;
3748         }
3749
3750         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3751         if (!sd_cdb_pool) {
3752                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3753                 err = -ENOMEM;
3754                 goto err_out_cache;
3755         }
3756
3757         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3758         if (!sd_page_pool) {
3759                 printk(KERN_ERR "sd: can't init discard page pool\n");
3760                 err = -ENOMEM;
3761                 goto err_out_ppool;
3762         }
3763
3764         err = scsi_register_driver(&sd_template.gendrv);
3765         if (err)
3766                 goto err_out_driver;
3767
3768         return 0;
3769
3770 err_out_driver:
3771         mempool_destroy(sd_page_pool);
3772
3773 err_out_ppool:
3774         mempool_destroy(sd_cdb_pool);
3775
3776 err_out_cache:
3777         kmem_cache_destroy(sd_cdb_cache);
3778
3779 err_out_class:
3780         class_unregister(&sd_disk_class);
3781 err_out:
3782         for (i = 0; i < SD_MAJORS; i++)
3783                 unregister_blkdev(sd_major(i), "sd");
3784         return err;
3785 }
3786
3787 /**
3788  *      exit_sd - exit point for this driver (when it is a module).
3789  *
3790  *      Note: this function unregisters this driver from the scsi mid-level.
3791  **/
3792 static void __exit exit_sd(void)
3793 {
3794         int i;
3795
3796         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3797
3798         scsi_unregister_driver(&sd_template.gendrv);
3799         mempool_destroy(sd_cdb_pool);
3800         mempool_destroy(sd_page_pool);
3801         kmem_cache_destroy(sd_cdb_cache);
3802
3803         class_unregister(&sd_disk_class);
3804
3805         for (i = 0; i < SD_MAJORS; i++)
3806                 unregister_blkdev(sd_major(i), "sd");
3807 }
3808
3809 module_init(init_sd);
3810 module_exit(exit_sd);
3811
3812 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3813 {
3814         scsi_print_sense_hdr(sdkp->device,
3815                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3816 }
3817
3818 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3819 {
3820         const char *hb_string = scsi_hostbyte_string(result);
3821
3822         if (hb_string)
3823                 sd_printk(KERN_INFO, sdkp,
3824                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3825                           hb_string ? hb_string : "invalid",
3826                           "DRIVER_OK");
3827         else
3828                 sd_printk(KERN_INFO, sdkp,
3829                           "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
3830                           msg, host_byte(result), "DRIVER_OK");
3831 }