drivers/scsi/sd.c

   1 /*
   2  *      sd.c Copyright (C) 1992 Drew Eckhardt
   3  *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
   4  *
   5  *      Linux scsi disk driver
   6  *              Initial versions: Drew Eckhardt
   7  *              Subsequent revisions: Eric Youngdale
   8  *      Modification history:
   9  *       - Drew Eckhardt <drew@colorado.edu> original
  10  *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
  11  *         outstanding request, and other enhancements.
  12  *         Support loadable low-level scsi drivers.
  13  *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
  14  *         eight major numbers.
  15  *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
  16  *       - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
  17  *         sd_init and cleanups.
  18  *       - Alex Davis <letmein@erols.com> Fix problem where partition info
  19  *         not being read in sd_open. Fix problem where removable media
  20  *         could be ejected after sd_open.
  21  *       - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
  22  *       - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
  23  *         <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
  24  *         Support 32k/1M disks.
  25  *
  26  *      Logging policy (needs CONFIG_SCSI_LOGGING defined):
  27  *       - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
  28  *       - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
  29  *       - entering sd_ioctl: SCSI_LOG_IOCTL level 1
  30  *       - entering other commands: SCSI_LOG_HLQUEUE level 3
  31  *      Note: when the logging level is set by the user, it must be greater
  32  *      than the level indicated above to trigger output.
  33  */
  34
  35 #include <linux/module.h>
  36 #include <linux/fs.h>
  37 #include <linux/kernel.h>
  38 #include <linux/mm.h>
  39 #include <linux/bio.h>
  40 #include <linux/genhd.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/errno.h>
  43 #include <linux/idr.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/init.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/blkpg.h>
  48 #include <linux/blk-pm.h>
  49 #include <linux/delay.h>
  50 #include <linux/mutex.h>
  51 #include <linux/string_helpers.h>
  52 #include <linux/async.h>
  53 #include <linux/slab.h>
  54 #include <linux/sed-opal.h>
  55 #include <linux/pm_runtime.h>
  56 #include <linux/pr.h>
  57 #include <linux/t10-pi.h>
  58 #include <linux/uaccess.h>
  59 #include <asm/unaligned.h>
  60
  61 #include <scsi/scsi.h>
  62 #include <scsi/scsi_cmnd.h>
  63 #include <scsi/scsi_dbg.h>
  64 #include <scsi/scsi_device.h>
  65 #include <scsi/scsi_driver.h>
  66 #include <scsi/scsi_eh.h>
  67 #include <scsi/scsi_host.h>
  68 #include <scsi/scsi_ioctl.h>
  69 #include <scsi/scsicam.h>
  70
  71 #include "sd.h"
  72 #include "scsi_priv.h"
  73 #include "scsi_logging.h"
  74
  75 MODULE_AUTHOR("Eric Youngdale");
  76 MODULE_DESCRIPTION("SCSI disk (sd) driver");
  77 MODULE_LICENSE("GPL");
  78
  79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
  80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
  81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
  82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
  83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
  84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
  85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
  86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
  87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
  88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
  89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
  90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
  91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
  92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
  93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
  94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
  95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
  96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
  97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
  98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
  99
 100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
 101 #define SD_MINORS       16
 102 #else
 103 #define SD_MINORS       0
 104 #endif
 105
 106 static void sd_config_discard(struct scsi_disk *, unsigned int);
 107 static void sd_config_write_same(struct scsi_disk *);
 108 static int  sd_revalidate_disk(struct gendisk *);
 109 static void sd_unlock_native_capacity(struct gendisk *disk);
 110 static int  sd_probe(struct device *);
 111 static int  sd_remove(struct device *);
 112 static void sd_shutdown(struct device *);
 113 static int sd_suspend_system(struct device *);
 114 static int sd_suspend_runtime(struct device *);
 115 static int sd_resume(struct device *);
 116 static void sd_rescan(struct device *);
 117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
 118 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
 119 static int sd_done(struct scsi_cmnd *);
 120 static void sd_eh_reset(struct scsi_cmnd *);
 121 static int sd_eh_action(struct scsi_cmnd *, int);
 122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 123 static void scsi_disk_release(struct device *cdev);
 124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
 125 static void sd_print_result(const struct scsi_disk *, const char *, int);
 126
 127 static DEFINE_IDA(sd_index_ida);
 128
 129 /* This semaphore is used to mediate the 0->1 reference get in the
 130  * face of object destruction (i.e. we can't allow a get on an
 131  * object after last put) */
 132 static DEFINE_MUTEX(sd_ref_mutex);
 133
 134 static struct kmem_cache *sd_cdb_cache;
 135 static mempool_t *sd_cdb_pool;
 136 static mempool_t *sd_page_pool;
 137
 138 static const char *sd_cache_types[] = {
 139         "write through", "none", "write back",
 140         "write back, no read (daft)"
 141 };
 142
 143 static void sd_set_flush_flag(struct scsi_disk *sdkp)
 144 {
 145         bool wc = false, fua = false;
 146
 147         if (sdkp->WCE) {
 148                 wc = true;
 149                 if (sdkp->DPOFUA)
 150                         fua = true;
 151         }
 152
 153         blk_queue_write_cache(sdkp->disk->queue, wc, fua);
 154 }
 155
 156 static ssize_t
 157 cache_type_store(struct device *dev, struct device_attribute *attr,
 158                  const char *buf, size_t count)
 159 {
 160         int ct, rcd, wce, sp;
 161         struct scsi_disk *sdkp = to_scsi_disk(dev);
 162         struct scsi_device *sdp = sdkp->device;
 163         char buffer[64];
 164         char *buffer_data;
 165         struct scsi_mode_data data;
 166         struct scsi_sense_hdr sshdr;
 167         static const char temp[] = "temporary ";
 168         int len;
 169
 170         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 171                 /* no cache control on RBC devices; theoretically they
 172                  * can do it, but there's probably so many exceptions
 173                  * it's not worth the risk */
 174                 return -EINVAL;
 175
 176         if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
 177                 buf += sizeof(temp) - 1;
 178                 sdkp->cache_override = 1;
 179         } else {
 180                 sdkp->cache_override = 0;
 181         }
 182
 183         ct = sysfs_match_string(sd_cache_types, buf);
 184         if (ct < 0)
 185                 return -EINVAL;
 186
 187         rcd = ct & 0x01 ? 1 : 0;
 188         wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
 189
 190         if (sdkp->cache_override) {
 191                 sdkp->WCE = wce;
 192                 sdkp->RCD = rcd;
 193                 sd_set_flush_flag(sdkp);
 194                 return count;
 195         }
 196
 197         if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
 198                             SD_MAX_RETRIES, &data, NULL))
 199                 return -EINVAL;
 200         len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
 201                   data.block_descriptor_length);
 202         buffer_data = buffer + data.header_length +
 203                 data.block_descriptor_length;
 204         buffer_data[2] &= ~0x05;
 205         buffer_data[2] |= wce << 2 | rcd;
 206         sp = buffer_data[0] & 0x80 ? 1 : 0;
 207         buffer_data[0] &= ~0x80;
 208
 209         if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
 210                              SD_MAX_RETRIES, &data, &sshdr)) {
 211                 if (scsi_sense_valid(&sshdr))
 212                         sd_print_sense_hdr(sdkp, &sshdr);
 213                 return -EINVAL;
 214         }
 215         revalidate_disk(sdkp->disk);
 216         return count;
 217 }
 218
 219 static ssize_t
 220 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
 221                        char *buf)
 222 {
 223         struct scsi_disk *sdkp = to_scsi_disk(dev);
 224         struct scsi_device *sdp = sdkp->device;
 225
 226         return sprintf(buf, "%u\n", sdp->manage_start_stop);
 227 }
 228
 229 static ssize_t
 230 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
 231                         const char *buf, size_t count)
 232 {
 233         struct scsi_disk *sdkp = to_scsi_disk(dev);
 234         struct scsi_device *sdp = sdkp->device;
 235         bool v;
 236
 237         if (!capable(CAP_SYS_ADMIN))
 238                 return -EACCES;
 239
 240         if (kstrtobool(buf, &v))
 241                 return -EINVAL;
 242
 243         sdp->manage_start_stop = v;
 244
 245         return count;
 246 }
 247 static DEVICE_ATTR_RW(manage_start_stop);
 248
 249 static ssize_t
 250 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
 251 {
 252         struct scsi_disk *sdkp = to_scsi_disk(dev);
 253
 254         return sprintf(buf, "%u\n", sdkp->device->allow_restart);
 255 }
 256
 257 static ssize_t
 258 allow_restart_store(struct device *dev, struct device_attribute *attr,
 259                     const char *buf, size_t count)
 260 {
 261         bool v;
 262         struct scsi_disk *sdkp = to_scsi_disk(dev);
 263         struct scsi_device *sdp = sdkp->device;
 264
 265         if (!capable(CAP_SYS_ADMIN))
 266                 return -EACCES;
 267
 268         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 269                 return -EINVAL;
 270
 271         if (kstrtobool(buf, &v))
 272                 return -EINVAL;
 273
 274         sdp->allow_restart = v;
 275
 276         return count;
 277 }
 278 static DEVICE_ATTR_RW(allow_restart);
 279
 280 static ssize_t
 281 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 282 {
 283         struct scsi_disk *sdkp = to_scsi_disk(dev);
 284         int ct = sdkp->RCD + 2*sdkp->WCE;
 285
 286         return sprintf(buf, "%s\n", sd_cache_types[ct]);
 287 }
 288 static DEVICE_ATTR_RW(cache_type);
 289
 290 static ssize_t
 291 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
 292 {
 293         struct scsi_disk *sdkp = to_scsi_disk(dev);
 294
 295         return sprintf(buf, "%u\n", sdkp->DPOFUA);
 296 }
 297 static DEVICE_ATTR_RO(FUA);
 298
 299 static ssize_t
 300 protection_type_show(struct device *dev, struct device_attribute *attr,
 301                      char *buf)
 302 {
 303         struct scsi_disk *sdkp = to_scsi_disk(dev);
 304
 305         return sprintf(buf, "%u\n", sdkp->protection_type);
 306 }
 307
 308 static ssize_t
 309 protection_type_store(struct device *dev, struct device_attribute *attr,
 310                       const char *buf, size_t count)
 311 {
 312         struct scsi_disk *sdkp = to_scsi_disk(dev);
 313         unsigned int val;
 314         int err;
 315
 316         if (!capable(CAP_SYS_ADMIN))
 317                 return -EACCES;
 318
 319         err = kstrtouint(buf, 10, &val);
 320
 321         if (err)
 322                 return err;
 323
 324         if (val <= T10_PI_TYPE3_PROTECTION)
 325                 sdkp->protection_type = val;
 326
 327         return count;
 328 }
 329 static DEVICE_ATTR_RW(protection_type);
 330
 331 static ssize_t
 332 protection_mode_show(struct device *dev, struct device_attribute *attr,
 333                      char *buf)
 334 {
 335         struct scsi_disk *sdkp = to_scsi_disk(dev);
 336         struct scsi_device *sdp = sdkp->device;
 337         unsigned int dif, dix;
 338
 339         dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
 340         dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
 341
 342         if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
 343                 dif = 0;
 344                 dix = 1;
 345         }
 346
 347         if (!dif && !dix)
 348                 return sprintf(buf, "none\n");
 349
 350         return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
 351 }
 352 static DEVICE_ATTR_RO(protection_mode);
 353
 354 static ssize_t
 355 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
 356 {
 357         struct scsi_disk *sdkp = to_scsi_disk(dev);
 358
 359         return sprintf(buf, "%u\n", sdkp->ATO);
 360 }
 361 static DEVICE_ATTR_RO(app_tag_own);
 362
 363 static ssize_t
 364 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
 365                        char *buf)
 366 {
 367         struct scsi_disk *sdkp = to_scsi_disk(dev);
 368
 369         return sprintf(buf, "%u\n", sdkp->lbpme);
 370 }
 371 static DEVICE_ATTR_RO(thin_provisioning);
 372
 373 /* sysfs_match_string() requires dense arrays */
 374 static const char *lbp_mode[] = {
 375         [SD_LBP_FULL]           = "full",
 376         [SD_LBP_UNMAP]          = "unmap",
 377         [SD_LBP_WS16]           = "writesame_16",
 378         [SD_LBP_WS10]           = "writesame_10",
 379         [SD_LBP_ZERO]           = "writesame_zero",
 380         [SD_LBP_DISABLE]        = "disabled",
 381 };
 382
 383 static ssize_t
 384 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
 385                        char *buf)
 386 {
 387         struct scsi_disk *sdkp = to_scsi_disk(dev);
 388
 389         return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
 390 }
 391
 392 static ssize_t
 393 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
 394                         const char *buf, size_t count)
 395 {
 396         struct scsi_disk *sdkp = to_scsi_disk(dev);
 397         struct scsi_device *sdp = sdkp->device;
 398         int mode;
 399
 400         if (!capable(CAP_SYS_ADMIN))
 401                 return -EACCES;
 402
 403         if (sd_is_zoned(sdkp)) {
 404                 sd_config_discard(sdkp, SD_LBP_DISABLE);
 405                 return count;
 406         }
 407
 408         if (sdp->type != TYPE_DISK)
 409                 return -EINVAL;
 410
 411         mode = sysfs_match_string(lbp_mode, buf);
 412         if (mode < 0)
 413                 return -EINVAL;
 414
 415         sd_config_discard(sdkp, mode);
 416
 417         return count;
 418 }
 419 static DEVICE_ATTR_RW(provisioning_mode);
 420
 421 /* sysfs_match_string() requires dense arrays */
 422 static const char *zeroing_mode[] = {
 423         [SD_ZERO_WRITE]         = "write",
 424         [SD_ZERO_WS]            = "writesame",
 425         [SD_ZERO_WS16_UNMAP]    = "writesame_16_unmap",
 426         [SD_ZERO_WS10_UNMAP]    = "writesame_10_unmap",
 427 };
 428
 429 static ssize_t
 430 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
 431                   char *buf)
 432 {
 433         struct scsi_disk *sdkp = to_scsi_disk(dev);
 434
 435         return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
 436 }
 437
 438 static ssize_t
 439 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
 440                    const char *buf, size_t count)
 441 {
 442         struct scsi_disk *sdkp = to_scsi_disk(dev);
 443         int mode;
 444
 445         if (!capable(CAP_SYS_ADMIN))
 446                 return -EACCES;
 447
 448         mode = sysfs_match_string(zeroing_mode, buf);
 449         if (mode < 0)
 450                 return -EINVAL;
 451
 452         sdkp->zeroing_mode = mode;
 453
 454         return count;
 455 }
 456 static DEVICE_ATTR_RW(zeroing_mode);
 457
 458 static ssize_t
 459 max_medium_access_timeouts_show(struct device *dev,
 460                                 struct device_attribute *attr, char *buf)
 461 {
 462         struct scsi_disk *sdkp = to_scsi_disk(dev);
 463
 464         return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
 465 }
 466
 467 static ssize_t
 468 max_medium_access_timeouts_store(struct device *dev,
 469                                  struct device_attribute *attr, const char *buf,
 470                                  size_t count)
 471 {
 472         struct scsi_disk *sdkp = to_scsi_disk(dev);
 473         int err;
 474
 475         if (!capable(CAP_SYS_ADMIN))
 476                 return -EACCES;
 477
 478         err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
 479
 480         return err ? err : count;
 481 }
 482 static DEVICE_ATTR_RW(max_medium_access_timeouts);
 483
 484 static ssize_t
 485 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
 486                            char *buf)
 487 {
 488         struct scsi_disk *sdkp = to_scsi_disk(dev);
 489
 490         return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
 491 }
 492
 493 static ssize_t
 494 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
 495                             const char *buf, size_t count)
 496 {
 497         struct scsi_disk *sdkp = to_scsi_disk(dev);
 498         struct scsi_device *sdp = sdkp->device;
 499         unsigned long max;
 500         int err;
 501
 502         if (!capable(CAP_SYS_ADMIN))
 503                 return -EACCES;
 504
 505         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
 506                 return -EINVAL;
 507
 508         err = kstrtoul(buf, 10, &max);
 509
 510         if (err)
 511                 return err;
 512
 513         if (max == 0)
 514                 sdp->no_write_same = 1;
 515         else if (max <= SD_MAX_WS16_BLOCKS) {
 516                 sdp->no_write_same = 0;
 517                 sdkp->max_ws_blocks = max;
 518         }
 519
 520         sd_config_write_same(sdkp);
 521
 522         return count;
 523 }
 524 static DEVICE_ATTR_RW(max_write_same_blocks);
 525
 526 static struct attribute *sd_disk_attrs[] = {
 527         &dev_attr_cache_type.attr,
 528         &dev_attr_FUA.attr,
 529         &dev_attr_allow_restart.attr,
 530         &dev_attr_manage_start_stop.attr,
 531         &dev_attr_protection_type.attr,
 532         &dev_attr_protection_mode.attr,
 533         &dev_attr_app_tag_own.attr,
 534         &dev_attr_thin_provisioning.attr,
 535         &dev_attr_provisioning_mode.attr,
 536         &dev_attr_zeroing_mode.attr,
 537         &dev_attr_max_write_same_blocks.attr,
 538         &dev_attr_max_medium_access_timeouts.attr,
 539         NULL,
 540 };
 541 ATTRIBUTE_GROUPS(sd_disk);
 542
 543 static struct class sd_disk_class = {
 544         .name           = "scsi_disk",
 545         .owner          = THIS_MODULE,
 546         .dev_release    = scsi_disk_release,
 547         .dev_groups     = sd_disk_groups,
 548 };
 549
 550 static const struct dev_pm_ops sd_pm_ops = {
 551         .suspend                = sd_suspend_system,
 552         .resume                 = sd_resume,
 553         .poweroff               = sd_suspend_system,
 554         .restore                = sd_resume,
 555         .runtime_suspend        = sd_suspend_runtime,
 556         .runtime_resume         = sd_resume,
 557 };
 558
 559 static struct scsi_driver sd_template = {
 560         .gendrv = {
 561                 .name           = "sd",
 562                 .owner          = THIS_MODULE,
 563                 .probe          = sd_probe,
 564                 .remove         = sd_remove,
 565                 .shutdown       = sd_shutdown,
 566                 .pm             = &sd_pm_ops,
 567         },
 568         .rescan                 = sd_rescan,
 569         .init_command           = sd_init_command,
 570         .uninit_command         = sd_uninit_command,
 571         .done                   = sd_done,
 572         .eh_action              = sd_eh_action,
 573         .eh_reset               = sd_eh_reset,
 574 };
 575
 576 /*
 577  * Dummy kobj_map->probe function.
 578  * The default ->probe function will call modprobe, which is
 579  * pointless as this module is already loaded.
 580  */
 581 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
 582 {
 583         return NULL;
 584 }
 585
 586 /*
 587  * Device no to disk mapping:
 588  *
 589  *       major         disc2     disc  p1
 590  *   |............|.............|....|....| <- dev_t
 591  *    31        20 19          8 7  4 3  0
 592  *
 593  * Inside a major, we have 16k disks, however mapped non-
 594  * contiguously. The first 16 disks are for major0, the next
 595  * ones with major1, ... Disk 256 is for major0 again, disk 272
 596  * for major1, ...
 597  * As we stay compatible with our numbering scheme, we can reuse
 598  * the well-know SCSI majors 8, 65--71, 136--143.
 599  */
 600 static int sd_major(int major_idx)
 601 {
 602         switch (major_idx) {
 603         case 0:
 604                 return SCSI_DISK0_MAJOR;
 605         case 1 ... 7:
 606                 return SCSI_DISK1_MAJOR + major_idx - 1;
 607         case 8 ... 15:
 608                 return SCSI_DISK8_MAJOR + major_idx - 8;
 609         default:
 610                 BUG();
 611                 return 0;       /* shut up gcc */
 612         }
 613 }
 614
 615 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 616 {
 617         struct scsi_disk *sdkp = NULL;
 618
 619         mutex_lock(&sd_ref_mutex);
 620
 621         if (disk->private_data) {
 622                 sdkp = scsi_disk(disk);
 623                 if (scsi_device_get(sdkp->device) == 0)
 624                         get_device(&sdkp->dev);
 625                 else
 626                         sdkp = NULL;
 627         }
 628         mutex_unlock(&sd_ref_mutex);
 629         return sdkp;
 630 }
 631
 632 static void scsi_disk_put(struct scsi_disk *sdkp)
 633 {
 634         struct scsi_device *sdev = sdkp->device;
 635
 636         mutex_lock(&sd_ref_mutex);
 637         put_device(&sdkp->dev);
 638         scsi_device_put(sdev);
 639         mutex_unlock(&sd_ref_mutex);
 640 }
 641
 642 #ifdef CONFIG_BLK_SED_OPAL
 643 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
 644                 size_t len, bool send)
 645 {
 646         struct scsi_device *sdev = data;
 647         u8 cdb[12] = { 0, };
 648         int ret;
 649
 650         cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 651         cdb[1] = secp;
 652         put_unaligned_be16(spsp, &cdb[2]);
 653         put_unaligned_be32(len, &cdb[6]);
 654
 655         ret = scsi_execute_req(sdev, cdb,
 656                         send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
 657                         buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
 658         return ret <= 0 ? ret : -EIO;
 659 }
 660 #endif /* CONFIG_BLK_SED_OPAL */
 661
 662 /*
 663  * Look up the DIX operation based on whether the command is read or
 664  * write and whether dix and dif are enabled.
 665  */
 666 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
 667 {
 668         /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
 669         static const unsigned int ops[] = {     /* wrt dix dif */
 670                 SCSI_PROT_NORMAL,               /*  0   0   0  */
 671                 SCSI_PROT_READ_STRIP,           /*  0   0   1  */
 672                 SCSI_PROT_READ_INSERT,          /*  0   1   0  */
 673                 SCSI_PROT_READ_PASS,            /*  0   1   1  */
 674                 SCSI_PROT_NORMAL,               /*  1   0   0  */
 675                 SCSI_PROT_WRITE_INSERT,         /*  1   0   1  */
 676                 SCSI_PROT_WRITE_STRIP,          /*  1   1   0  */
 677                 SCSI_PROT_WRITE_PASS,           /*  1   1   1  */
 678         };
 679
 680         return ops[write << 2 | dix << 1 | dif];
 681 }
 682
 683 /*
 684  * Returns a mask of the protection flags that are valid for a given DIX
 685  * operation.
 686  */
 687 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
 688 {
 689         static const unsigned int flag_mask[] = {
 690                 [SCSI_PROT_NORMAL]              = 0,
 691
 692                 [SCSI_PROT_READ_STRIP]          = SCSI_PROT_TRANSFER_PI |
 693                                                   SCSI_PROT_GUARD_CHECK |
 694                                                   SCSI_PROT_REF_CHECK |
 695                                                   SCSI_PROT_REF_INCREMENT,
 696
 697                 [SCSI_PROT_READ_INSERT]         = SCSI_PROT_REF_INCREMENT |
 698                                                   SCSI_PROT_IP_CHECKSUM,
 699
 700                 [SCSI_PROT_READ_PASS]           = SCSI_PROT_TRANSFER_PI |
 701                                                   SCSI_PROT_GUARD_CHECK |
 702                                                   SCSI_PROT_REF_CHECK |
 703                                                   SCSI_PROT_REF_INCREMENT |
 704                                                   SCSI_PROT_IP_CHECKSUM,
 705
 706                 [SCSI_PROT_WRITE_INSERT]        = SCSI_PROT_TRANSFER_PI |
 707                                                   SCSI_PROT_REF_INCREMENT,
 708
 709                 [SCSI_PROT_WRITE_STRIP]         = SCSI_PROT_GUARD_CHECK |
 710                                                   SCSI_PROT_REF_CHECK |
 711                                                   SCSI_PROT_REF_INCREMENT |
 712                                                   SCSI_PROT_IP_CHECKSUM,
 713
 714                 [SCSI_PROT_WRITE_PASS]          = SCSI_PROT_TRANSFER_PI |
 715                                                   SCSI_PROT_GUARD_CHECK |
 716                                                   SCSI_PROT_REF_CHECK |
 717                                                   SCSI_PROT_REF_INCREMENT |
 718                                                   SCSI_PROT_IP_CHECKSUM,
 719         };
 720
 721         return flag_mask[prot_op];
 722 }
 723
 724 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
 725                                            unsigned int dix, unsigned int dif)
 726 {
 727         struct bio *bio = scmd->request->bio;
 728         unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
 729         unsigned int protect = 0;
 730
 731         if (dix) {                              /* DIX Type 0, 1, 2, 3 */
 732                 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
 733                         scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
 734
 735                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 736                         scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
 737         }
 738
 739         if (dif != T10_PI_TYPE3_PROTECTION) {   /* DIX/DIF Type 0, 1, 2 */
 740                 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
 741
 742                 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
 743                         scmd->prot_flags |= SCSI_PROT_REF_CHECK;
 744         }
 745
 746         if (dif) {                              /* DIX/DIF Type 1, 2, 3 */
 747                 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
 748
 749                 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
 750                         protect = 3 << 5;       /* Disable target PI checking */
 751                 else
 752                         protect = 1 << 5;       /* Enable target PI checking */
 753         }
 754
 755         scsi_set_prot_op(scmd, prot_op);
 756         scsi_set_prot_type(scmd, dif);
 757         scmd->prot_flags &= sd_prot_flag_mask(prot_op);
 758
 759         return protect;
 760 }
 761
 762 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
 763 {
 764         struct request_queue *q = sdkp->disk->queue;
 765         unsigned int logical_block_size = sdkp->device->sector_size;
 766         unsigned int max_blocks = 0;
 767
 768         q->limits.discard_alignment =
 769                 sdkp->unmap_alignment * logical_block_size;
 770         q->limits.discard_granularity =
 771                 max(sdkp->physical_block_size,
 772                     sdkp->unmap_granularity * logical_block_size);
 773         sdkp->provisioning_mode = mode;
 774
 775         switch (mode) {
 776
 777         case SD_LBP_FULL:
 778         case SD_LBP_DISABLE:
 779                 blk_queue_max_discard_sectors(q, 0);
 780                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 781                 return;
 782
 783         case SD_LBP_UNMAP:
 784                 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
 785                                           (u32)SD_MAX_WS16_BLOCKS);
 786                 break;
 787
 788         case SD_LBP_WS16:
 789                 if (sdkp->device->unmap_limit_for_ws)
 790                         max_blocks = sdkp->max_unmap_blocks;
 791                 else
 792                         max_blocks = sdkp->max_ws_blocks;
 793
 794                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
 795                 break;
 796
 797         case SD_LBP_WS10:
 798                 if (sdkp->device->unmap_limit_for_ws)
 799                         max_blocks = sdkp->max_unmap_blocks;
 800                 else
 801                         max_blocks = sdkp->max_ws_blocks;
 802
 803                 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
 804                 break;
 805
 806         case SD_LBP_ZERO:
 807                 max_blocks = min_not_zero(sdkp->max_ws_blocks,
 808                                           (u32)SD_MAX_WS10_BLOCKS);
 809                 break;
 810         }
 811
 812         blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
 813         blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 814 }
 815
 816 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 817 {
 818         struct scsi_device *sdp = cmd->device;
 819         struct request *rq = cmd->request;
 820         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 821         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 822         unsigned int data_len = 24;
 823         char *buf;
 824
 825         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 826         if (!rq->special_vec.bv_page)
 827                 return BLK_STS_RESOURCE;
 828         clear_highpage(rq->special_vec.bv_page);
 829         rq->special_vec.bv_offset = 0;
 830         rq->special_vec.bv_len = data_len;
 831         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 832
 833         cmd->cmd_len = 10;
 834         cmd->cmnd[0] = UNMAP;
 835         cmd->cmnd[8] = 24;
 836
 837         buf = page_address(rq->special_vec.bv_page);
 838         put_unaligned_be16(6 + 16, &buf[0]);
 839         put_unaligned_be16(16, &buf[2]);
 840         put_unaligned_be64(lba, &buf[8]);
 841         put_unaligned_be32(nr_blocks, &buf[16]);
 842
 843         cmd->allowed = SD_MAX_RETRIES;
 844         cmd->transfersize = data_len;
 845         rq->timeout = SD_TIMEOUT;
 846         scsi_req(rq)->resid_len = data_len;
 847
 848         return scsi_init_io(cmd);
 849 }
 850
 851 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
 852                 bool unmap)
 853 {
 854         struct scsi_device *sdp = cmd->device;
 855         struct request *rq = cmd->request;
 856         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 857         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 858         u32 data_len = sdp->sector_size;
 859
 860         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 861         if (!rq->special_vec.bv_page)
 862                 return BLK_STS_RESOURCE;
 863         clear_highpage(rq->special_vec.bv_page);
 864         rq->special_vec.bv_offset = 0;
 865         rq->special_vec.bv_len = data_len;
 866         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 867
 868         cmd->cmd_len = 16;
 869         cmd->cmnd[0] = WRITE_SAME_16;
 870         if (unmap)
 871                 cmd->cmnd[1] = 0x8; /* UNMAP */
 872         put_unaligned_be64(lba, &cmd->cmnd[2]);
 873         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
 874
 875         cmd->allowed = SD_MAX_RETRIES;
 876         cmd->transfersize = data_len;
 877         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 878         scsi_req(rq)->resid_len = data_len;
 879
 880         return scsi_init_io(cmd);
 881 }
 882
 883 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
 884                 bool unmap)
 885 {
 886         struct scsi_device *sdp = cmd->device;
 887         struct request *rq = cmd->request;
 888         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 889         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 890         u32 data_len = sdp->sector_size;
 891
 892         rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
 893         if (!rq->special_vec.bv_page)
 894                 return BLK_STS_RESOURCE;
 895         clear_highpage(rq->special_vec.bv_page);
 896         rq->special_vec.bv_offset = 0;
 897         rq->special_vec.bv_len = data_len;
 898         rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
 899
 900         cmd->cmd_len = 10;
 901         cmd->cmnd[0] = WRITE_SAME;
 902         if (unmap)
 903                 cmd->cmnd[1] = 0x8; /* UNMAP */
 904         put_unaligned_be32(lba, &cmd->cmnd[2]);
 905         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
 906
 907         cmd->allowed = SD_MAX_RETRIES;
 908         cmd->transfersize = data_len;
 909         rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
 910         scsi_req(rq)->resid_len = data_len;
 911
 912         return scsi_init_io(cmd);
 913 }
 914
 915 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 916 {
 917         struct request *rq = cmd->request;
 918         struct scsi_device *sdp = cmd->device;
 919         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 920         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
 921         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
 922
 923         if (!(rq->cmd_flags & REQ_NOUNMAP)) {
 924                 switch (sdkp->zeroing_mode) {
 925                 case SD_ZERO_WS16_UNMAP:
 926                         return sd_setup_write_same16_cmnd(cmd, true);
 927                 case SD_ZERO_WS10_UNMAP:
 928                         return sd_setup_write_same10_cmnd(cmd, true);
 929                 }
 930         }
 931
 932         if (sdp->no_write_same)
 933                 return BLK_STS_TARGET;
 934
 935         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
 936                 return sd_setup_write_same16_cmnd(cmd, false);
 937
 938         return sd_setup_write_same10_cmnd(cmd, false);
 939 }
 940
 941 static void sd_config_write_same(struct scsi_disk *sdkp)
 942 {
 943         struct request_queue *q = sdkp->disk->queue;
 944         unsigned int logical_block_size = sdkp->device->sector_size;
 945
 946         if (sdkp->device->no_write_same) {
 947                 sdkp->max_ws_blocks = 0;
 948                 goto out;
 949         }
 950
 951         /* Some devices can not handle block counts above 0xffff despite
 952          * supporting WRITE SAME(16). Consequently we default to 64k
 953          * blocks per I/O unless the device explicitly advertises a
 954          * bigger limit.
 955          */
 956         if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
 957                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 958                                                    (u32)SD_MAX_WS16_BLOCKS);
 959         else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
 960                 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
 961                                                    (u32)SD_MAX_WS10_BLOCKS);
 962         else {
 963                 sdkp->device->no_write_same = 1;
 964                 sdkp->max_ws_blocks = 0;
 965         }
 966
 967         if (sdkp->lbprz && sdkp->lbpws)
 968                 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
 969         else if (sdkp->lbprz && sdkp->lbpws10)
 970                 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
 971         else if (sdkp->max_ws_blocks)
 972                 sdkp->zeroing_mode = SD_ZERO_WS;
 973         else
 974                 sdkp->zeroing_mode = SD_ZERO_WRITE;
 975
 976         if (sdkp->max_ws_blocks &&
 977             sdkp->physical_block_size > logical_block_size) {
 978                 /*
 979                  * Reporting a maximum number of blocks that is not aligned
 980                  * on the device physical size would cause a large write same
 981                  * request to be split into physically unaligned chunks by
 982                  * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
 983                  * even if the caller of these functions took care to align the
 984                  * large request. So make sure the maximum reported is aligned
 985                  * to the device physical block size. This is only an optional
 986                  * optimization for regular disks, but this is mandatory to
 987                  * avoid failure of large write same requests directed at
 988                  * sequential write required zones of host-managed ZBC disks.
 989                  */
 990                 sdkp->max_ws_blocks =
 991                         round_down(sdkp->max_ws_blocks,
 992                                    bytes_to_logical(sdkp->device,
 993                                                     sdkp->physical_block_size));
 994         }
 995
 996 out:
 997         blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
 998                                          (logical_block_size >> 9));
 999         blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1000                                          (logical_block_size >> 9));
1001 }
1002
1003 /**
1004  * sd_setup_write_same_cmnd - write the same data to multiple blocks
1005  * @cmd: command to prepare
1006  *
1007  * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1008  * the preference indicated by the target device.
1009  **/
1010 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1011 {
1012         struct request *rq = cmd->request;
1013         struct scsi_device *sdp = cmd->device;
1014         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1015         struct bio *bio = rq->bio;
1016         u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1017         u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1018         blk_status_t ret;
1019
1020         if (sdkp->device->no_write_same)
1021                 return BLK_STS_TARGET;
1022
1023         BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1024
1025         rq->timeout = SD_WRITE_SAME_TIMEOUT;
1026
1027         if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1028                 cmd->cmd_len = 16;
1029                 cmd->cmnd[0] = WRITE_SAME_16;
1030                 put_unaligned_be64(lba, &cmd->cmnd[2]);
1031                 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1032         } else {
1033                 cmd->cmd_len = 10;
1034                 cmd->cmnd[0] = WRITE_SAME;
1035                 put_unaligned_be32(lba, &cmd->cmnd[2]);
1036                 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1037         }
1038
1039         cmd->transfersize = sdp->sector_size;
1040         cmd->allowed = SD_MAX_RETRIES;
1041
1042         /*
1043          * For WRITE SAME the data transferred via the DATA OUT buffer is
1044          * different from the amount of data actually written to the target.
1045          *
1046          * We set up __data_len to the amount of data transferred via the
1047          * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1048          * to transfer a single sector of data first, but then reset it to
1049          * the amount of data to be written right after so that the I/O path
1050          * knows how much to actually write.
1051          */
1052         rq->__data_len = sdp->sector_size;
1053         ret = scsi_init_io(cmd);
1054         rq->__data_len = blk_rq_bytes(rq);
1055
1056         return ret;
1057 }
1058
1059 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1060 {
1061         struct request *rq = cmd->request;
1062
1063         /* flush requests don't perform I/O, zero the S/G table */
1064         memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1065
1066         cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1067         cmd->cmd_len = 10;
1068         cmd->transfersize = 0;
1069         cmd->allowed = SD_MAX_RETRIES;
1070
1071         rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1072         return BLK_STS_OK;
1073 }
1074
1075 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1076                                        sector_t lba, unsigned int nr_blocks,
1077                                        unsigned char flags)
1078 {
1079         cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1080         if (unlikely(cmd->cmnd == NULL))
1081                 return BLK_STS_RESOURCE;
1082
1083         cmd->cmd_len = SD_EXT_CDB_SIZE;
1084         memset(cmd->cmnd, 0, cmd->cmd_len);
1085
1086         cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1087         cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1088         cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1089         cmd->cmnd[10] = flags;
1090         put_unaligned_be64(lba, &cmd->cmnd[12]);
1091         put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1092         put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1093
1094         return BLK_STS_OK;
1095 }
1096
1097 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1098                                        sector_t lba, unsigned int nr_blocks,
1099                                        unsigned char flags)
1100 {
1101         cmd->cmd_len  = 16;
1102         cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1103         cmd->cmnd[1]  = flags;
1104         cmd->cmnd[14] = 0;
1105         cmd->cmnd[15] = 0;
1106         put_unaligned_be64(lba, &cmd->cmnd[2]);
1107         put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1108
1109         return BLK_STS_OK;
1110 }
1111
1112 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1113                                        sector_t lba, unsigned int nr_blocks,
1114                                        unsigned char flags)
1115 {
1116         cmd->cmd_len = 10;
1117         cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1118         cmd->cmnd[1] = flags;
1119         cmd->cmnd[6] = 0;
1120         cmd->cmnd[9] = 0;
1121         put_unaligned_be32(lba, &cmd->cmnd[2]);
1122         put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1123
1124         return BLK_STS_OK;
1125 }
1126
1127 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1128                                       sector_t lba, unsigned int nr_blocks,
1129                                       unsigned char flags)
1130 {
1131         /* Avoid that 0 blocks gets translated into 256 blocks. */
1132         if (WARN_ON_ONCE(nr_blocks == 0))
1133                 return BLK_STS_IOERR;
1134
1135         if (unlikely(flags & 0x8)) {
1136                 /*
1137                  * This happens only if this drive failed 10byte rw
1138                  * command with ILLEGAL_REQUEST during operation and
1139                  * thus turned off use_10_for_rw.
1140                  */
1141                 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1142                 return BLK_STS_IOERR;
1143         }
1144
1145         cmd->cmd_len = 6;
1146         cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1147         cmd->cmnd[1] = (lba >> 16) & 0x1f;
1148         cmd->cmnd[2] = (lba >> 8) & 0xff;
1149         cmd->cmnd[3] = lba & 0xff;
1150         cmd->cmnd[4] = nr_blocks;
1151         cmd->cmnd[5] = 0;
1152
1153         return BLK_STS_OK;
1154 }
1155
1156 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1157 {
1158         struct request *rq = cmd->request;
1159         struct scsi_device *sdp = cmd->device;
1160         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1161         sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1162         sector_t threshold;
1163         unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1164         bool dif, dix;
1165         unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1166         bool write = rq_data_dir(rq) == WRITE;
1167         unsigned char protect, fua;
1168         blk_status_t ret;
1169
1170         ret = scsi_init_io(cmd);
1171         if (ret != BLK_STS_OK)
1172                 return ret;
1173
1174         if (!scsi_device_online(sdp) || sdp->changed) {
1175                 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1176                 return BLK_STS_IOERR;
1177         }
1178
1179         if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1180                 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1181                 return BLK_STS_IOERR;
1182         }
1183
1184         if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1185                 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1186                 return BLK_STS_IOERR;
1187         }
1188
1189         /*
1190          * Some SD card readers can't handle accesses which touch the
1191          * last one or two logical blocks. Split accesses as needed.
1192          */
1193         threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1194
1195         if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1196                 if (lba < threshold) {
1197                         /* Access up to the threshold but not beyond */
1198                         nr_blocks = threshold - lba;
1199                 } else {
1200                         /* Access only a single logical block */
1201                         nr_blocks = 1;
1202                 }
1203         }
1204
1205         fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1206         dix = scsi_prot_sg_count(cmd);
1207         dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1208
1209         if (write && dix)
1210                 t10_pi_prepare(cmd->request, sdkp->protection_type);
1211
1212         if (dif || dix)
1213                 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1214         else
1215                 protect = 0;
1216
1217         if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1218                 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1219                                          protect | fua);
1220         } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1221                 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1222                                          protect | fua);
1223         } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1224                    sdp->use_10_for_rw || protect) {
1225                 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1226                                          protect | fua);
1227         } else {
1228                 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1229                                         protect | fua);
1230         }
1231
1232         if (unlikely(ret != BLK_STS_OK))
1233                 return ret;
1234
1235         /*
1236          * We shouldn't disconnect in the middle of a sector, so with a dumb
1237          * host adapter, it's safe to assume that we can at least transfer
1238          * this many bytes between each connect / disconnect.
1239          */
1240         cmd->transfersize = sdp->sector_size;
1241         cmd->underflow = nr_blocks << 9;
1242         cmd->allowed = SD_MAX_RETRIES;
1243         cmd->sdb.length = nr_blocks * sdp->sector_size;
1244
1245         SCSI_LOG_HLQUEUE(1,
1246                          scmd_printk(KERN_INFO, cmd,
1247                                      "%s: block=%llu, count=%d\n", __func__,
1248                                      (unsigned long long)blk_rq_pos(rq),
1249                                      blk_rq_sectors(rq)));
1250         SCSI_LOG_HLQUEUE(2,
1251                          scmd_printk(KERN_INFO, cmd,
1252                                      "%s %d/%u 512 byte blocks.\n",
1253                                      write ? "writing" : "reading", nr_blocks,
1254                                      blk_rq_sectors(rq)));
1255
1256         /*
1257          * This indicates that the command is ready from our end to be
1258          * queued.
1259          */
1260         return BLK_STS_OK;
1261 }
1262
1263 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1264 {
1265         struct request *rq = cmd->request;
1266
1267         switch (req_op(rq)) {
1268         case REQ_OP_DISCARD:
1269                 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1270                 case SD_LBP_UNMAP:
1271                         return sd_setup_unmap_cmnd(cmd);
1272                 case SD_LBP_WS16:
1273                         return sd_setup_write_same16_cmnd(cmd, true);
1274                 case SD_LBP_WS10:
1275                         return sd_setup_write_same10_cmnd(cmd, true);
1276                 case SD_LBP_ZERO:
1277                         return sd_setup_write_same10_cmnd(cmd, false);
1278                 default:
1279                         return BLK_STS_TARGET;
1280                 }
1281         case REQ_OP_WRITE_ZEROES:
1282                 return sd_setup_write_zeroes_cmnd(cmd);
1283         case REQ_OP_WRITE_SAME:
1284                 return sd_setup_write_same_cmnd(cmd);
1285         case REQ_OP_FLUSH:
1286                 return sd_setup_flush_cmnd(cmd);
1287         case REQ_OP_READ:
1288         case REQ_OP_WRITE:
1289                 return sd_setup_read_write_cmnd(cmd);
1290         case REQ_OP_ZONE_RESET:
1291                 return sd_zbc_setup_reset_cmnd(cmd);
1292         default:
1293                 WARN_ON_ONCE(1);
1294                 return BLK_STS_NOTSUPP;
1295         }
1296 }
1297
1298 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1299 {
1300         struct request *rq = SCpnt->request;
1301         u8 *cmnd;
1302
1303         if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1304                 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1305
1306         if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1307                 cmnd = SCpnt->cmnd;
1308                 SCpnt->cmnd = NULL;
1309                 SCpnt->cmd_len = 0;
1310                 mempool_free(cmnd, sd_cdb_pool);
1311         }
1312 }
1313
1314 /**
1315  *      sd_open - open a scsi disk device
1316  *      @bdev: Block device of the scsi disk to open
1317  *      @mode: FMODE_* mask
1318  *
1319  *      Returns 0 if successful. Returns a negated errno value in case
1320  *      of error.
1321  *
1322  *      Note: This can be called from a user context (e.g. fsck(1) )
1323  *      or from within the kernel (e.g. as a result of a mount(1) ).
1324  *      In the latter case @inode and @filp carry an abridged amount
1325  *      of information as noted above.
1326  *
1327  *      Locking: called with bdev->bd_mutex held.
1328  **/
1329 static int sd_open(struct block_device *bdev, fmode_t mode)
1330 {
1331         struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1332         struct scsi_device *sdev;
1333         int retval;
1334
1335         if (!sdkp)
1336                 return -ENXIO;
1337
1338         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1339
1340         sdev = sdkp->device;
1341
1342         /*
1343          * If the device is in error recovery, wait until it is done.
1344          * If the device is offline, then disallow any access to it.
1345          */
1346         retval = -ENXIO;
1347         if (!scsi_block_when_processing_errors(sdev))
1348                 goto error_out;
1349
1350         if (sdev->removable || sdkp->write_prot)
1351                 check_disk_change(bdev);
1352
1353         /*
1354          * If the drive is empty, just let the open fail.
1355          */
1356         retval = -ENOMEDIUM;
1357         if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1358                 goto error_out;
1359
1360         /*
1361          * If the device has the write protect tab set, have the open fail
1362          * if the user expects to be able to write to the thing.
1363          */
1364         retval = -EROFS;
1365         if (sdkp->write_prot && (mode & FMODE_WRITE))
1366                 goto error_out;
1367
1368         /*
1369          * It is possible that the disk changing stuff resulted in
1370          * the device being taken offline.  If this is the case,
1371          * report this to the user, and don't pretend that the
1372          * open actually succeeded.
1373          */
1374         retval = -ENXIO;
1375         if (!scsi_device_online(sdev))
1376                 goto error_out;
1377
1378         if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1379                 if (scsi_block_when_processing_errors(sdev))
1380                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1381         }
1382
1383         return 0;
1384
1385 error_out:
1386         scsi_disk_put(sdkp);
1387         return retval;
1388 }
1389
1390 /**
1391  *      sd_release - invoked when the (last) close(2) is called on this
1392  *      scsi disk.
1393  *      @disk: disk to release
1394  *      @mode: FMODE_* mask
1395  *
1396  *      Returns 0.
1397  *
1398  *      Note: may block (uninterruptible) if error recovery is underway
1399  *      on this disk.
1400  *
1401  *      Locking: called with bdev->bd_mutex held.
1402  **/
1403 static void sd_release(struct gendisk *disk, fmode_t mode)
1404 {
1405         struct scsi_disk *sdkp = scsi_disk(disk);
1406         struct scsi_device *sdev = sdkp->device;
1407
1408         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1409
1410         if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1411                 if (scsi_block_when_processing_errors(sdev))
1412                         scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1413         }
1414
1415         /*
1416          * XXX and what if there are packets in flight and this close()
1417          * XXX is followed by a "rmmod sd_mod"?
1418          */
1419
1420         scsi_disk_put(sdkp);
1421 }
1422
1423 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1424 {
1425         struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1426         struct scsi_device *sdp = sdkp->device;
1427         struct Scsi_Host *host = sdp->host;
1428         sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1429         int diskinfo[4];
1430
1431         /* default to most commonly used values */
1432         diskinfo[0] = 0x40;     /* 1 << 6 */
1433         diskinfo[1] = 0x20;     /* 1 << 5 */
1434         diskinfo[2] = capacity >> 11;
1435
1436         /* override with calculated, extended default, or driver values */
1437         if (host->hostt->bios_param)
1438                 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1439         else
1440                 scsicam_bios_param(bdev, capacity, diskinfo);
1441
1442         geo->heads = diskinfo[0];
1443         geo->sectors = diskinfo[1];
1444         geo->cylinders = diskinfo[2];
1445         return 0;
1446 }
1447
1448 /**
1449  *      sd_ioctl - process an ioctl
1450  *      @bdev: target block device
1451  *      @mode: FMODE_* mask
1452  *      @cmd: ioctl command number
1453  *      @arg: this is third argument given to ioctl(2) system call.
1454  *      Often contains a pointer.
1455  *
1456  *      Returns 0 if successful (some ioctls return positive numbers on
1457  *      success as well). Returns a negated errno value in case of error.
1458  *
1459  *      Note: most ioctls are forward onto the block subsystem or further
1460  *      down in the scsi subsystem.
1461  **/
1462 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1463                     unsigned int cmd, unsigned long arg)
1464 {
1465         struct gendisk *disk = bdev->bd_disk;
1466         struct scsi_disk *sdkp = scsi_disk(disk);
1467         struct scsi_device *sdp = sdkp->device;
1468         void __user *p = (void __user *)arg;
1469         int error;
1470
1471         SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1472                                     "cmd=0x%x\n", disk->disk_name, cmd));
1473
1474         error = scsi_verify_blk_ioctl(bdev, cmd);
1475         if (error < 0)
1476                 return error;
1477
1478         /*
1479          * If we are in the middle of error recovery, don't let anyone
1480          * else try and use this device.  Also, if error recovery fails, it
1481          * may try and take the device offline, in which case all further
1482          * access to the device is prohibited.
1483          */
1484         error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1485                         (mode & FMODE_NDELAY) != 0);
1486         if (error)
1487                 goto out;
1488
1489         if (is_sed_ioctl(cmd))
1490                 return sed_ioctl(sdkp->opal_dev, cmd, p);
1491
1492         /*
1493          * Send SCSI addressing ioctls directly to mid level, send other
1494          * ioctls to block level and then onto mid level if they can't be
1495          * resolved.
1496          */
1497         switch (cmd) {
1498                 case SCSI_IOCTL_GET_IDLUN:
1499                 case SCSI_IOCTL_GET_BUS_NUMBER:
1500                         error = scsi_ioctl(sdp, cmd, p);
1501                         break;
1502                 default:
1503                         error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1504                         if (error != -ENOTTY)
1505                                 break;
1506                         error = scsi_ioctl(sdp, cmd, p);
1507                         break;
1508         }
1509 out:
1510         return error;
1511 }
1512
1513 static void set_media_not_present(struct scsi_disk *sdkp)
1514 {
1515         if (sdkp->media_present)
1516                 sdkp->device->changed = 1;
1517
1518         if (sdkp->device->removable) {
1519                 sdkp->media_present = 0;
1520                 sdkp->capacity = 0;
1521         }
1522 }
1523
1524 static int media_not_present(struct scsi_disk *sdkp,
1525                              struct scsi_sense_hdr *sshdr)
1526 {
1527         if (!scsi_sense_valid(sshdr))
1528                 return 0;
1529
1530         /* not invoked for commands that could return deferred errors */
1531         switch (sshdr->sense_key) {
1532         case UNIT_ATTENTION:
1533         case NOT_READY:
1534                 /* medium not present */
1535                 if (sshdr->asc == 0x3A) {
1536                         set_media_not_present(sdkp);
1537                         return 1;
1538                 }
1539         }
1540         return 0;
1541 }
1542
1543 /**
1544  *      sd_check_events - check media events
1545  *      @disk: kernel device descriptor
1546  *      @clearing: disk events currently being cleared
1547  *
1548  *      Returns mask of DISK_EVENT_*.
1549  *
1550  *      Note: this function is invoked from the block subsystem.
1551  **/
1552 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1553 {
1554         struct scsi_disk *sdkp = scsi_disk_get(disk);
1555         struct scsi_device *sdp;
1556         int retval;
1557
1558         if (!sdkp)
1559                 return 0;
1560
1561         sdp = sdkp->device;
1562         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1563
1564         /*
1565          * If the device is offline, don't send any commands - just pretend as
1566          * if the command failed.  If the device ever comes back online, we
1567          * can deal with it then.  It is only because of unrecoverable errors
1568          * that we would ever take a device offline in the first place.
1569          */
1570         if (!scsi_device_online(sdp)) {
1571                 set_media_not_present(sdkp);
1572                 goto out;
1573         }
1574
1575         /*
1576          * Using TEST_UNIT_READY enables differentiation between drive with
1577          * no cartridge loaded - NOT READY, drive with changed cartridge -
1578          * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1579          *
1580          * Drives that auto spin down. eg iomega jaz 1G, will be started
1581          * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1582          * sd_revalidate() is called.
1583          */
1584         if (scsi_block_when_processing_errors(sdp)) {
1585                 struct scsi_sense_hdr sshdr = { 0, };
1586
1587                 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1588                                               &sshdr);
1589
1590                 /* failed to execute TUR, assume media not present */
1591                 if (host_byte(retval)) {
1592                         set_media_not_present(sdkp);
1593                         goto out;
1594                 }
1595
1596                 if (media_not_present(sdkp, &sshdr))
1597                         goto out;
1598         }
1599
1600         /*
1601          * For removable scsi disk we have to recognise the presence
1602          * of a disk in the drive.
1603          */
1604         if (!sdkp->media_present)
1605                 sdp->changed = 1;
1606         sdkp->media_present = 1;
1607 out:
1608         /*
1609          * sdp->changed is set under the following conditions:
1610          *
1611          *      Medium present state has changed in either direction.
1612          *      Device has indicated UNIT_ATTENTION.
1613          */
1614         retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1615         sdp->changed = 0;
1616         scsi_disk_put(sdkp);
1617         return retval;
1618 }
1619
1620 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1621 {
1622         int retries, res;
1623         struct scsi_device *sdp = sdkp->device;
1624         const int timeout = sdp->request_queue->rq_timeout
1625                 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1626         struct scsi_sense_hdr my_sshdr;
1627
1628         if (!scsi_device_online(sdp))
1629                 return -ENODEV;
1630
1631         /* caller might not be interested in sense, but we need it */
1632         if (!sshdr)
1633                 sshdr = &my_sshdr;
1634
1635         for (retries = 3; retries > 0; --retries) {
1636                 unsigned char cmd[10] = { 0 };
1637
1638                 cmd[0] = SYNCHRONIZE_CACHE;
1639                 /*
1640                  * Leave the rest of the command zero to indicate
1641                  * flush everything.
1642                  */
1643                 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1644                                 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1645                 if (res == 0)
1646                         break;
1647         }
1648
1649         if (res) {
1650                 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1651
1652                 if (driver_byte(res) == DRIVER_SENSE)
1653                         sd_print_sense_hdr(sdkp, sshdr);
1654
1655                 /* we need to evaluate the error return  */
1656                 if (scsi_sense_valid(sshdr) &&
1657                         (sshdr->asc == 0x3a ||  /* medium not present */
1658                          sshdr->asc == 0x20))   /* invalid command */
1659                                 /* this is no error here */
1660                                 return 0;
1661
1662                 switch (host_byte(res)) {
1663                 /* ignore errors due to racing a disconnection */
1664                 case DID_BAD_TARGET:
1665                 case DID_NO_CONNECT:
1666                         return 0;
1667                 /* signal the upper layer it might try again */
1668                 case DID_BUS_BUSY:
1669                 case DID_IMM_RETRY:
1670                 case DID_REQUEUE:
1671                 case DID_SOFT_ERROR:
1672                         return -EBUSY;
1673                 default:
1674                         return -EIO;
1675                 }
1676         }
1677         return 0;
1678 }
1679
1680 static void sd_rescan(struct device *dev)
1681 {
1682         struct scsi_disk *sdkp = dev_get_drvdata(dev);
1683
1684         revalidate_disk(sdkp->disk);
1685 }
1686
1687
1688 #ifdef CONFIG_COMPAT
1689 /*
1690  * This gets directly called from VFS. When the ioctl
1691  * is not recognized we go back to the other translation paths.
1692  */
1693 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1694                            unsigned int cmd, unsigned long arg)
1695 {
1696         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1697         int error;
1698
1699         error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1700                         (mode & FMODE_NDELAY) != 0);
1701         if (error)
1702                 return error;
1703
1704         /*
1705          * Let the static ioctl translation table take care of it.
1706          */
1707         if (!sdev->host->hostt->compat_ioctl)
1708                 return -ENOIOCTLCMD;
1709         return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1710 }
1711 #endif
1712
1713 static char sd_pr_type(enum pr_type type)
1714 {
1715         switch (type) {
1716         case PR_WRITE_EXCLUSIVE:
1717                 return 0x01;
1718         case PR_EXCLUSIVE_ACCESS:
1719                 return 0x03;
1720         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1721                 return 0x05;
1722         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1723                 return 0x06;
1724         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1725                 return 0x07;
1726         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1727                 return 0x08;
1728         default:
1729                 return 0;
1730         }
1731 };
1732
1733 static int sd_pr_command(struct block_device *bdev, u8 sa,
1734                 u64 key, u64 sa_key, u8 type, u8 flags)
1735 {
1736         struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1737         struct scsi_sense_hdr sshdr;
1738         int result;
1739         u8 cmd[16] = { 0, };
1740         u8 data[24] = { 0, };
1741
1742         cmd[0] = PERSISTENT_RESERVE_OUT;
1743         cmd[1] = sa;
1744         cmd[2] = type;
1745         put_unaligned_be32(sizeof(data), &cmd[5]);
1746
1747         put_unaligned_be64(key, &data[0]);
1748         put_unaligned_be64(sa_key, &data[8]);
1749         data[20] = flags;
1750
1751         result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1752                         &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1753
1754         if (driver_byte(result) == DRIVER_SENSE &&
1755             scsi_sense_valid(&sshdr)) {
1756                 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1757                 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1758         }
1759
1760         return result;
1761 }
1762
1763 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1764                 u32 flags)
1765 {
1766         if (flags & ~PR_FL_IGNORE_KEY)
1767                 return -EOPNOTSUPP;
1768         return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1769                         old_key, new_key, 0,
1770                         (1 << 0) /* APTPL */);
1771 }
1772
1773 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1774                 u32 flags)
1775 {
1776         if (flags)
1777                 return -EOPNOTSUPP;
1778         return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1779 }
1780
1781 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1782 {
1783         return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1784 }
1785
1786 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1787                 enum pr_type type, bool abort)
1788 {
1789         return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1790                              sd_pr_type(type), 0);
1791 }
1792
1793 static int sd_pr_clear(struct block_device *bdev, u64 key)
1794 {
1795         return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1796 }
1797
1798 static const struct pr_ops sd_pr_ops = {
1799         .pr_register    = sd_pr_register,
1800         .pr_reserve     = sd_pr_reserve,
1801         .pr_release     = sd_pr_release,
1802         .pr_preempt     = sd_pr_preempt,
1803         .pr_clear       = sd_pr_clear,
1804 };
1805
1806 static const struct block_device_operations sd_fops = {
1807         .owner                  = THIS_MODULE,
1808         .open                   = sd_open,
1809         .release                = sd_release,
1810         .ioctl                  = sd_ioctl,
1811         .getgeo                 = sd_getgeo,
1812 #ifdef CONFIG_COMPAT
1813         .compat_ioctl           = sd_compat_ioctl,
1814 #endif
1815         .check_events           = sd_check_events,
1816         .revalidate_disk        = sd_revalidate_disk,
1817         .unlock_native_capacity = sd_unlock_native_capacity,
1818         .report_zones           = sd_zbc_report_zones,
1819         .pr_ops                 = &sd_pr_ops,
1820 };
1821
1822 /**
1823  *      sd_eh_reset - reset error handling callback
1824  *      @scmd:          sd-issued command that has failed
1825  *
1826  *      This function is called by the SCSI midlayer before starting
1827  *      SCSI EH. When counting medium access failures we have to be
1828  *      careful to register it only only once per device and SCSI EH run;
1829  *      there might be several timed out commands which will cause the
1830  *      'max_medium_access_timeouts' counter to trigger after the first
1831  *      SCSI EH run already and set the device to offline.
1832  *      So this function resets the internal counter before starting SCSI EH.
1833  **/
1834 static void sd_eh_reset(struct scsi_cmnd *scmd)
1835 {
1836         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1837
1838         /* New SCSI EH run, reset gate variable */
1839         sdkp->ignore_medium_access_errors = false;
1840 }
1841
1842 /**
1843  *      sd_eh_action - error handling callback
1844  *      @scmd:          sd-issued command that has failed
1845  *      @eh_disp:       The recovery disposition suggested by the midlayer
1846  *
1847  *      This function is called by the SCSI midlayer upon completion of an
1848  *      error test command (currently TEST UNIT READY). The result of sending
1849  *      the eh command is passed in eh_disp.  We're looking for devices that
1850  *      fail medium access commands but are OK with non access commands like
1851  *      test unit ready (so wrongly see the device as having a successful
1852  *      recovery)
1853  **/
1854 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1855 {
1856         struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1857         struct scsi_device *sdev = scmd->device;
1858
1859         if (!scsi_device_online(sdev) ||
1860             !scsi_medium_access_command(scmd) ||
1861             host_byte(scmd->result) != DID_TIME_OUT ||
1862             eh_disp != SUCCESS)
1863                 return eh_disp;
1864
1865         /*
1866          * The device has timed out executing a medium access command.
1867          * However, the TEST UNIT READY command sent during error
1868          * handling completed successfully. Either the device is in the
1869          * process of recovering or has it suffered an internal failure
1870          * that prevents access to the storage medium.
1871          */
1872         if (!sdkp->ignore_medium_access_errors) {
1873                 sdkp->medium_access_timed_out++;
1874                 sdkp->ignore_medium_access_errors = true;
1875         }
1876
1877         /*
1878          * If the device keeps failing read/write commands but TEST UNIT
1879          * READY always completes successfully we assume that medium
1880          * access is no longer possible and take the device offline.
1881          */
1882         if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1883                 scmd_printk(KERN_ERR, scmd,
1884                             "Medium access timeout failure. Offlining disk!\n");
1885                 mutex_lock(&sdev->state_mutex);
1886                 scsi_device_set_state(sdev, SDEV_OFFLINE);
1887                 mutex_unlock(&sdev->state_mutex);
1888
1889                 return SUCCESS;
1890         }
1891
1892         return eh_disp;
1893 }
1894
1895 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1896 {
1897         struct request *req = scmd->request;
1898         struct scsi_device *sdev = scmd->device;
1899         unsigned int transferred, good_bytes;
1900         u64 start_lba, end_lba, bad_lba;
1901
1902         /*
1903          * Some commands have a payload smaller than the device logical
1904          * block size (e.g. INQUIRY on a 4K disk).
1905          */
1906         if (scsi_bufflen(scmd) <= sdev->sector_size)
1907                 return 0;
1908
1909         /* Check if we have a 'bad_lba' information */
1910         if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1911                                      SCSI_SENSE_BUFFERSIZE,
1912                                      &bad_lba))
1913                 return 0;
1914
1915         /*
1916          * If the bad lba was reported incorrectly, we have no idea where
1917          * the error is.
1918          */
1919         start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1920         end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1921         if (bad_lba < start_lba || bad_lba >= end_lba)
1922                 return 0;
1923
1924         /*
1925          * resid is optional but mostly filled in.  When it's unused,
1926          * its value is zero, so we assume the whole buffer transferred
1927          */
1928         transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1929
1930         /* This computation should always be done in terms of the
1931          * resolution of the device's medium.
1932          */
1933         good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1934
1935         return min(good_bytes, transferred);
1936 }
1937
1938 /**
1939  *      sd_done - bottom half handler: called when the lower level
1940  *      driver has completed (successfully or otherwise) a scsi command.
1941  *      @SCpnt: mid-level's per command structure.
1942  *
1943  *      Note: potentially run from within an ISR. Must not block.
1944  **/
1945 static int sd_done(struct scsi_cmnd *SCpnt)
1946 {
1947         int result = SCpnt->result;
1948         unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1949         unsigned int sector_size = SCpnt->device->sector_size;
1950         unsigned int resid;
1951         struct scsi_sense_hdr sshdr;
1952         struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1953         struct request *req = SCpnt->request;
1954         int sense_valid = 0;
1955         int sense_deferred = 0;
1956
1957         switch (req_op(req)) {
1958         case REQ_OP_DISCARD:
1959         case REQ_OP_WRITE_ZEROES:
1960         case REQ_OP_WRITE_SAME:
1961         case REQ_OP_ZONE_RESET:
1962                 if (!result) {
1963                         good_bytes = blk_rq_bytes(req);
1964                         scsi_set_resid(SCpnt, 0);
1965                 } else {
1966                         good_bytes = 0;
1967                         scsi_set_resid(SCpnt, blk_rq_bytes(req));
1968                 }
1969                 break;
1970         default:
1971                 /*
1972                  * In case of bogus fw or device, we could end up having
1973                  * an unaligned partial completion. Check this here and force
1974                  * alignment.
1975                  */
1976                 resid = scsi_get_resid(SCpnt);
1977                 if (resid & (sector_size - 1)) {
1978                         sd_printk(KERN_INFO, sdkp,
1979                                 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1980                                 resid, sector_size);
1981                         resid = min(scsi_bufflen(SCpnt),
1982                                     round_up(resid, sector_size));
1983                         scsi_set_resid(SCpnt, resid);
1984                 }
1985         }
1986
1987         if (result) {
1988                 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1989                 if (sense_valid)
1990                         sense_deferred = scsi_sense_is_deferred(&sshdr);
1991         }
1992         sdkp->medium_access_timed_out = 0;
1993
1994         if (driver_byte(result) != DRIVER_SENSE &&
1995             (!sense_valid || sense_deferred))
1996                 goto out;
1997
1998         switch (sshdr.sense_key) {
1999         case HARDWARE_ERROR:
2000         case MEDIUM_ERROR:
2001                 good_bytes = sd_completed_bytes(SCpnt);
2002                 break;
2003         case RECOVERED_ERROR:
2004                 good_bytes = scsi_bufflen(SCpnt);
2005                 break;
2006         case NO_SENSE:
2007                 /* This indicates a false check condition, so ignore it.  An
2008                  * unknown amount of data was transferred so treat it as an
2009                  * error.
2010                  */
2011                 SCpnt->result = 0;
2012                 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2013                 break;
2014         case ABORTED_COMMAND:
2015                 if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2016                         good_bytes = sd_completed_bytes(SCpnt);
2017                 break;
2018         case ILLEGAL_REQUEST:
2019                 switch (sshdr.asc) {
2020                 case 0x10:      /* DIX: Host detected corruption */
2021                         good_bytes = sd_completed_bytes(SCpnt);
2022                         break;
2023                 case 0x20:      /* INVALID COMMAND OPCODE */
2024                 case 0x24:      /* INVALID FIELD IN CDB */
2025                         switch (SCpnt->cmnd[0]) {
2026                         case UNMAP:
2027                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2028                                 break;
2029                         case WRITE_SAME_16:
2030                         case WRITE_SAME:
2031                                 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2032                                         sd_config_discard(sdkp, SD_LBP_DISABLE);
2033                                 } else {
2034                                         sdkp->device->no_write_same = 1;
2035                                         sd_config_write_same(sdkp);
2036                                         req->rq_flags |= RQF_QUIET;
2037                                 }
2038                                 break;
2039                         }
2040                 }
2041                 break;
2042         default:
2043                 break;
2044         }
2045
2046  out:
2047         if (sd_is_zoned(sdkp))
2048                 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2049
2050         SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2051                                            "sd_done: completed %d of %d bytes\n",
2052                                            good_bytes, scsi_bufflen(SCpnt)));
2053
2054         if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2055             good_bytes)
2056                 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2057                                 good_bytes / scsi_prot_interval(SCpnt));
2058
2059         return good_bytes;
2060 }
2061
2062 /*
2063  * spinup disk - called only in sd_revalidate_disk()
2064  */
2065 static void
2066 sd_spinup_disk(struct scsi_disk *sdkp)
2067 {
2068         unsigned char cmd[10];
2069         unsigned long spintime_expire = 0;
2070         int retries, spintime;
2071         unsigned int the_result;
2072         struct scsi_sense_hdr sshdr;
2073         int sense_valid = 0;
2074
2075         spintime = 0;
2076
2077         /* Spin up drives, as required.  Only do this at boot time */
2078         /* Spinup needs to be done for module loads too. */
2079         do {
2080                 retries = 0;
2081
2082                 do {
2083                         cmd[0] = TEST_UNIT_READY;
2084                         memset((void *) &cmd[1], 0, 9);
2085
2086                         the_result = scsi_execute_req(sdkp->device, cmd,
2087                                                       DMA_NONE, NULL, 0,
2088                                                       &sshdr, SD_TIMEOUT,
2089                                                       SD_MAX_RETRIES, NULL);
2090
2091                         /*
2092                          * If the drive has indicated to us that it
2093                          * doesn't have any media in it, don't bother
2094                          * with any more polling.
2095                          */
2096                         if (media_not_present(sdkp, &sshdr))
2097                                 return;
2098
2099                         if (the_result)
2100                                 sense_valid = scsi_sense_valid(&sshdr);
2101                         retries++;
2102                 } while (retries < 3 &&
2103                          (!scsi_status_is_good(the_result) ||
2104                           ((driver_byte(the_result) == DRIVER_SENSE) &&
2105                           sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2106
2107                 if (driver_byte(the_result) != DRIVER_SENSE) {
2108                         /* no sense, TUR either succeeded or failed
2109                          * with a status error */
2110                         if(!spintime && !scsi_status_is_good(the_result)) {
2111                                 sd_print_result(sdkp, "Test Unit Ready failed",
2112                                                 the_result);
2113                         }
2114                         break;
2115                 }
2116
2117                 /*
2118                  * The device does not want the automatic start to be issued.
2119                  */
2120                 if (sdkp->device->no_start_on_add)
2121                         break;
2122
2123                 if (sense_valid && sshdr.sense_key == NOT_READY) {
2124                         if (sshdr.asc == 4 && sshdr.ascq == 3)
2125                                 break;  /* manual intervention required */
2126                         if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2127                                 break;  /* standby */
2128                         if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2129                                 break;  /* unavailable */
2130                         if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2131                                 break;  /* sanitize in progress */
2132                         /*
2133                          * Issue command to spin up drive when not ready
2134                          */
2135                         if (!spintime) {
2136                                 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2137                                 cmd[0] = START_STOP;
2138                                 cmd[1] = 1;     /* Return immediately */
2139                                 memset((void *) &cmd[2], 0, 8);
2140                                 cmd[4] = 1;     /* Start spin cycle */
2141                                 if (sdkp->device->start_stop_pwr_cond)
2142                                         cmd[4] |= 1 << 4;
2143                                 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2144                                                  NULL, 0, &sshdr,
2145                                                  SD_TIMEOUT, SD_MAX_RETRIES,
2146                                                  NULL);
2147                                 spintime_expire = jiffies + 100 * HZ;
2148                                 spintime = 1;
2149                         }
2150                         /* Wait 1 second for next try */
2151                         msleep(1000);
2152                         printk(KERN_CONT ".");
2153
2154                 /*
2155                  * Wait for USB flash devices with slow firmware.
2156                  * Yes, this sense key/ASC combination shouldn't
2157                  * occur here.  It's characteristic of these devices.
2158                  */
2159                 } else if (sense_valid &&
2160                                 sshdr.sense_key == UNIT_ATTENTION &&
2161                                 sshdr.asc == 0x28) {
2162                         if (!spintime) {
2163                                 spintime_expire = jiffies + 5 * HZ;
2164                                 spintime = 1;
2165                         }
2166                         /* Wait 1 second for next try */
2167                         msleep(1000);
2168                 } else {
2169                         /* we don't understand the sense code, so it's
2170                          * probably pointless to loop */
2171                         if(!spintime) {
2172                                 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2173                                 sd_print_sense_hdr(sdkp, &sshdr);
2174                         }
2175                         break;
2176                 }
2177
2178         } while (spintime && time_before_eq(jiffies, spintime_expire));
2179
2180         if (spintime) {
2181                 if (scsi_status_is_good(the_result))
2182                         printk(KERN_CONT "ready\n");
2183                 else
2184                         printk(KERN_CONT "not responding...\n");
2185         }
2186 }
2187
2188 /*
2189  * Determine whether disk supports Data Integrity Field.
2190  */
2191 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2192 {
2193         struct scsi_device *sdp = sdkp->device;
2194         u8 type;
2195         int ret = 0;
2196
2197         if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2198                 return ret;
2199
2200         type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2201
2202         if (type > T10_PI_TYPE3_PROTECTION)
2203                 ret = -ENODEV;
2204         else if (scsi_host_dif_capable(sdp->host, type))
2205                 ret = 1;
2206
2207         if (sdkp->first_scan || type != sdkp->protection_type)
2208                 switch (ret) {
2209                 case -ENODEV:
2210                         sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2211                                   " protection type %u. Disabling disk!\n",
2212                                   type);
2213                         break;
2214                 case 1:
2215                         sd_printk(KERN_NOTICE, sdkp,
2216                                   "Enabling DIF Type %u protection\n", type);
2217                         break;
2218                 case 0:
2219                         sd_printk(KERN_NOTICE, sdkp,
2220                                   "Disabling DIF Type %u protection\n", type);
2221                         break;
2222                 }
2223
2224         sdkp->protection_type = type;
2225
2226         return ret;
2227 }
2228
2229 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2230                         struct scsi_sense_hdr *sshdr, int sense_valid,
2231                         int the_result)
2232 {
2233         if (driver_byte(the_result) == DRIVER_SENSE)
2234                 sd_print_sense_hdr(sdkp, sshdr);
2235         else
2236                 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2237
2238         /*
2239          * Set dirty bit for removable devices if not ready -
2240          * sometimes drives will not report this properly.
2241          */
2242         if (sdp->removable &&
2243             sense_valid && sshdr->sense_key == NOT_READY)
2244                 set_media_not_present(sdkp);
2245
2246         /*
2247          * We used to set media_present to 0 here to indicate no media
2248          * in the drive, but some drives fail read capacity even with
2249          * media present, so we can't do that.
2250          */
2251         sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2252 }
2253
2254 #define RC16_LEN 32
2255 #if RC16_LEN > SD_BUF_SIZE
2256 #error RC16_LEN must not be more than SD_BUF_SIZE
2257 #endif
2258
2259 #define READ_CAPACITY_RETRIES_ON_RESET  10
2260
2261 /*
2262  * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2263  * and the reported logical block size is bigger than 512 bytes. Note
2264  * that last_sector is a u64 and therefore logical_to_sectors() is not
2265  * applicable.
2266  */
2267 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2268 {
2269         u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2270
2271         if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2272                 return false;
2273
2274         return true;
2275 }
2276
2277 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2278                                                 unsigned char *buffer)
2279 {
2280         unsigned char cmd[16];
2281         struct scsi_sense_hdr sshdr;
2282         int sense_valid = 0;
2283         int the_result;
2284         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2285         unsigned int alignment;
2286         unsigned long long lba;
2287         unsigned sector_size;
2288
2289         if (sdp->no_read_capacity_16)
2290                 return -EINVAL;
2291
2292         do {
2293                 memset(cmd, 0, 16);
2294                 cmd[0] = SERVICE_ACTION_IN_16;
2295                 cmd[1] = SAI_READ_CAPACITY_16;
2296                 cmd[13] = RC16_LEN;
2297                 memset(buffer, 0, RC16_LEN);
2298
2299                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2300                                         buffer, RC16_LEN, &sshdr,
2301                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2302
2303                 if (media_not_present(sdkp, &sshdr))
2304                         return -ENODEV;
2305
2306                 if (the_result) {
2307                         sense_valid = scsi_sense_valid(&sshdr);
2308                         if (sense_valid &&
2309                             sshdr.sense_key == ILLEGAL_REQUEST &&
2310                             (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2311                             sshdr.ascq == 0x00)
2312                                 /* Invalid Command Operation Code or
2313                                  * Invalid Field in CDB, just retry
2314                                  * silently with RC10 */
2315                                 return -EINVAL;
2316                         if (sense_valid &&
2317                             sshdr.sense_key == UNIT_ATTENTION &&
2318                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2319                                 /* Device reset might occur several times,
2320                                  * give it one more chance */
2321                                 if (--reset_retries > 0)
2322                                         continue;
2323                 }
2324                 retries--;
2325
2326         } while (the_result && retries);
2327
2328         if (the_result) {
2329                 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2330                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2331                 return -EINVAL;
2332         }
2333
2334         sector_size = get_unaligned_be32(&buffer[8]);
2335         lba = get_unaligned_be64(&buffer[0]);
2336
2337         if (sd_read_protection_type(sdkp, buffer) < 0) {
2338                 sdkp->capacity = 0;
2339                 return -ENODEV;
2340         }
2341
2342         if (!sd_addressable_capacity(lba, sector_size)) {
2343                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2344                         "kernel compiled with support for large block "
2345                         "devices.\n");
2346                 sdkp->capacity = 0;
2347                 return -EOVERFLOW;
2348         }
2349
2350         /* Logical blocks per physical block exponent */
2351         sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2352
2353         /* RC basis */
2354         sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2355
2356         /* Lowest aligned logical block */
2357         alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2358         blk_queue_alignment_offset(sdp->request_queue, alignment);
2359         if (alignment && sdkp->first_scan)
2360                 sd_printk(KERN_NOTICE, sdkp,
2361                           "physical block alignment offset: %u\n", alignment);
2362
2363         if (buffer[14] & 0x80) { /* LBPME */
2364                 sdkp->lbpme = 1;
2365
2366                 if (buffer[14] & 0x40) /* LBPRZ */
2367                         sdkp->lbprz = 1;
2368
2369                 sd_config_discard(sdkp, SD_LBP_WS16);
2370         }
2371
2372         sdkp->capacity = lba + 1;
2373         return sector_size;
2374 }
2375
2376 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2377                                                 unsigned char *buffer)
2378 {
2379         unsigned char cmd[16];
2380         struct scsi_sense_hdr sshdr;
2381         int sense_valid = 0;
2382         int the_result;
2383         int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2384         sector_t lba;
2385         unsigned sector_size;
2386
2387         do {
2388                 cmd[0] = READ_CAPACITY;
2389                 memset(&cmd[1], 0, 9);
2390                 memset(buffer, 0, 8);
2391
2392                 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2393                                         buffer, 8, &sshdr,
2394                                         SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2395
2396                 if (media_not_present(sdkp, &sshdr))
2397                         return -ENODEV;
2398
2399                 if (the_result) {
2400                         sense_valid = scsi_sense_valid(&sshdr);
2401                         if (sense_valid &&
2402                             sshdr.sense_key == UNIT_ATTENTION &&
2403                             sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2404                                 /* Device reset might occur several times,
2405                                  * give it one more chance */
2406                                 if (--reset_retries > 0)
2407                                         continue;
2408                 }
2409                 retries--;
2410
2411         } while (the_result && retries);
2412
2413         if (the_result) {
2414                 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2415                 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2416                 return -EINVAL;
2417         }
2418
2419         sector_size = get_unaligned_be32(&buffer[4]);
2420         lba = get_unaligned_be32(&buffer[0]);
2421
2422         if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2423                 /* Some buggy (usb cardreader) devices return an lba of
2424                    0xffffffff when the want to report a size of 0 (with
2425                    which they really mean no media is present) */
2426                 sdkp->capacity = 0;
2427                 sdkp->physical_block_size = sector_size;
2428                 return sector_size;
2429         }
2430
2431         if (!sd_addressable_capacity(lba, sector_size)) {
2432                 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2433                         "kernel compiled with support for large block "
2434                         "devices.\n");
2435                 sdkp->capacity = 0;
2436                 return -EOVERFLOW;
2437         }
2438
2439         sdkp->capacity = lba + 1;
2440         sdkp->physical_block_size = sector_size;
2441         return sector_size;
2442 }
2443
2444 static int sd_try_rc16_first(struct scsi_device *sdp)
2445 {
2446         if (sdp->host->max_cmd_len < 16)
2447                 return 0;
2448         if (sdp->try_rc_10_first)
2449                 return 0;
2450         if (sdp->scsi_level > SCSI_SPC_2)
2451                 return 1;
2452         if (scsi_device_protection(sdp))
2453                 return 1;
2454         return 0;
2455 }
2456
2457 /*
2458  * read disk capacity
2459  */
2460 static void
2461 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2462 {
2463         int sector_size;
2464         struct scsi_device *sdp = sdkp->device;
2465
2466         if (sd_try_rc16_first(sdp)) {
2467                 sector_size = read_capacity_16(sdkp, sdp, buffer);
2468                 if (sector_size == -EOVERFLOW)
2469                         goto got_data;
2470                 if (sector_size == -ENODEV)
2471                         return;
2472                 if (sector_size < 0)
2473                         sector_size = read_capacity_10(sdkp, sdp, buffer);
2474                 if (sector_size < 0)
2475                         return;
2476         } else {
2477                 sector_size = read_capacity_10(sdkp, sdp, buffer);
2478                 if (sector_size == -EOVERFLOW)
2479                         goto got_data;
2480                 if (sector_size < 0)
2481                         return;
2482                 if ((sizeof(sdkp->capacity) > 4) &&
2483                     (sdkp->capacity > 0xffffffffULL)) {
2484                         int old_sector_size = sector_size;
2485                         sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2486                                         "Trying to use READ CAPACITY(16).\n");
2487                         sector_size = read_capacity_16(sdkp, sdp, buffer);
2488                         if (sector_size < 0) {
2489                                 sd_printk(KERN_NOTICE, sdkp,
2490                                         "Using 0xffffffff as device size\n");
2491                                 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2492                                 sector_size = old_sector_size;
2493                                 goto got_data;
2494                         }
2495                         /* Remember that READ CAPACITY(16) succeeded */
2496                         sdp->try_rc_10_first = 0;
2497                 }
2498         }
2499
2500         /* Some devices are known to return the total number of blocks,
2501          * not the highest block number.  Some devices have versions
2502          * which do this and others which do not.  Some devices we might
2503          * suspect of doing this but we don't know for certain.
2504          *
2505          * If we know the reported capacity is wrong, decrement it.  If
2506          * we can only guess, then assume the number of blocks is even
2507          * (usually true but not always) and err on the side of lowering
2508          * the capacity.
2509          */
2510         if (sdp->fix_capacity ||
2511             (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2512                 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2513                                 "from its reported value: %llu\n",
2514                                 (unsigned long long) sdkp->capacity);
2515                 --sdkp->capacity;
2516         }
2517
2518 got_data:
2519         if (sector_size == 0) {
2520                 sector_size = 512;
2521                 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2522                           "assuming 512.\n");
2523         }
2524
2525         if (sector_size != 512 &&
2526             sector_size != 1024 &&
2527             sector_size != 2048 &&
2528             sector_size != 4096) {
2529                 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2530                           sector_size);
2531                 /*
2532                  * The user might want to re-format the drive with
2533                  * a supported sectorsize.  Once this happens, it
2534                  * would be relatively trivial to set the thing up.
2535                  * For this reason, we leave the thing in the table.
2536                  */
2537                 sdkp->capacity = 0;
2538                 /*
2539                  * set a bogus sector size so the normal read/write
2540                  * logic in the block layer will eventually refuse any
2541                  * request on this device without tripping over power
2542                  * of two sector size assumptions
2543                  */
2544                 sector_size = 512;
2545         }
2546         blk_queue_logical_block_size(sdp->request_queue, sector_size);
2547         blk_queue_physical_block_size(sdp->request_queue,
2548                                       sdkp->physical_block_size);
2549         sdkp->device->sector_size = sector_size;
2550
2551         if (sdkp->capacity > 0xffffffff)
2552                 sdp->use_16_for_rw = 1;
2553
2554 }
2555
2556 /*
2557  * Print disk capacity
2558  */
2559 static void
2560 sd_print_capacity(struct scsi_disk *sdkp,
2561                   sector_t old_capacity)
2562 {
2563         int sector_size = sdkp->device->sector_size;
2564         char cap_str_2[10], cap_str_10[10];
2565
2566         if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2567                 return;
2568
2569         string_get_size(sdkp->capacity, sector_size,
2570                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2571         string_get_size(sdkp->capacity, sector_size,
2572                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2573
2574         sd_printk(KERN_NOTICE, sdkp,
2575                   "%llu %d-byte logical blocks: (%s/%s)\n",
2576                   (unsigned long long)sdkp->capacity,
2577                   sector_size, cap_str_10, cap_str_2);
2578
2579         if (sdkp->physical_block_size != sector_size)
2580                 sd_printk(KERN_NOTICE, sdkp,
2581                           "%u-byte physical blocks\n",
2582                           sdkp->physical_block_size);
2583
2584         sd_zbc_print_zones(sdkp);
2585 }
2586
2587 /* called with buffer of length 512 */
2588 static inline int
2589 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2590                  unsigned char *buffer, int len, struct scsi_mode_data *data,
2591                  struct scsi_sense_hdr *sshdr)
2592 {
2593         return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2594                                SD_TIMEOUT, SD_MAX_RETRIES, data,
2595                                sshdr);
2596 }
2597
2598 /*
2599  * read write protect setting, if possible - called only in sd_revalidate_disk()
2600  * called with buffer of length SD_BUF_SIZE
2601  */
2602 static void
2603 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2604 {
2605         int res;
2606         struct scsi_device *sdp = sdkp->device;
2607         struct scsi_mode_data data;
2608         int disk_ro = get_disk_ro(sdkp->disk);
2609         int old_wp = sdkp->write_prot;
2610
2611         set_disk_ro(sdkp->disk, 0);
2612         if (sdp->skip_ms_page_3f) {
2613                 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2614                 return;
2615         }
2616
2617         if (sdp->use_192_bytes_for_3f) {
2618                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2619         } else {
2620                 /*
2621                  * First attempt: ask for all pages (0x3F), but only 4 bytes.
2622                  * We have to start carefully: some devices hang if we ask
2623                  * for more than is available.
2624                  */
2625                 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2626
2627                 /*
2628                  * Second attempt: ask for page 0 When only page 0 is
2629                  * implemented, a request for page 3F may return Sense Key
2630                  * 5: Illegal Request, Sense Code 24: Invalid field in
2631                  * CDB.
2632                  */
2633                 if (!scsi_status_is_good(res))
2634                         res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2635
2636                 /*
2637                  * Third attempt: ask 255 bytes, as we did earlier.
2638                  */
2639                 if (!scsi_status_is_good(res))
2640                         res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2641                                                &data, NULL);
2642         }
2643
2644         if (!scsi_status_is_good(res)) {
2645                 sd_first_printk(KERN_WARNING, sdkp,
2646                           "Test WP failed, assume Write Enabled\n");
2647         } else {
2648                 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2649                 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2650                 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2651                         sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2652                                   sdkp->write_prot ? "on" : "off");
2653                         sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2654                 }
2655         }
2656 }
2657
2658 /*
2659  * sd_read_cache_type - called only from sd_revalidate_disk()
2660  * called with buffer of length SD_BUF_SIZE
2661  */
2662 static void
2663 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2664 {
2665         int len = 0, res;
2666         struct scsi_device *sdp = sdkp->device;
2667
2668         int dbd;
2669         int modepage;
2670         int first_len;
2671         struct scsi_mode_data data;
2672         struct scsi_sense_hdr sshdr;
2673         int old_wce = sdkp->WCE;
2674         int old_rcd = sdkp->RCD;
2675         int old_dpofua = sdkp->DPOFUA;
2676
2677
2678         if (sdkp->cache_override)
2679                 return;
2680
2681         first_len = 4;
2682         if (sdp->skip_ms_page_8) {
2683                 if (sdp->type == TYPE_RBC)
2684                         goto defaults;
2685                 else {
2686                         if (sdp->skip_ms_page_3f)
2687                                 goto defaults;
2688                         modepage = 0x3F;
2689                         if (sdp->use_192_bytes_for_3f)
2690                                 first_len = 192;
2691                         dbd = 0;
2692                 }
2693         } else if (sdp->type == TYPE_RBC) {
2694                 modepage = 6;
2695                 dbd = 8;
2696         } else {
2697                 modepage = 8;
2698                 dbd = 0;
2699         }
2700
2701         /* cautiously ask */
2702         res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2703                         &data, &sshdr);
2704
2705         if (!scsi_status_is_good(res))
2706                 goto bad_sense;
2707
2708         if (!data.header_length) {
2709                 modepage = 6;
2710                 first_len = 0;
2711                 sd_first_printk(KERN_ERR, sdkp,
2712                                 "Missing header in MODE_SENSE response\n");
2713         }
2714
2715         /* that went OK, now ask for the proper length */
2716         len = data.length;
2717
2718         /*
2719          * We're only interested in the first three bytes, actually.
2720          * But the data cache page is defined for the first 20.
2721          */
2722         if (len < 3)
2723                 goto bad_sense;
2724         else if (len > SD_BUF_SIZE) {
2725                 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2726                           "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2727                 len = SD_BUF_SIZE;
2728         }
2729         if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2730                 len = 192;
2731
2732         /* Get the data */
2733         if (len > first_len)
2734                 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2735                                 &data, &sshdr);
2736
2737         if (scsi_status_is_good(res)) {
2738                 int offset = data.header_length + data.block_descriptor_length;
2739
2740                 while (offset < len) {
2741                         u8 page_code = buffer[offset] & 0x3F;
2742                         u8 spf       = buffer[offset] & 0x40;
2743
2744                         if (page_code == 8 || page_code == 6) {
2745                                 /* We're interested only in the first 3 bytes.
2746                                  */
2747                                 if (len - offset <= 2) {
2748                                         sd_first_printk(KERN_ERR, sdkp,
2749                                                 "Incomplete mode parameter "
2750                                                         "data\n");
2751                                         goto defaults;
2752                                 } else {
2753                                         modepage = page_code;
2754                                         goto Page_found;
2755                                 }
2756                         } else {
2757                                 /* Go to the next page */
2758                                 if (spf && len - offset > 3)
2759                                         offset += 4 + (buffer[offset+2] << 8) +
2760                                                 buffer[offset+3];
2761                                 else if (!spf && len - offset > 1)
2762                                         offset += 2 + buffer[offset+1];
2763                                 else {
2764                                         sd_first_printk(KERN_ERR, sdkp,
2765                                                         "Incomplete mode "
2766                                                         "parameter data\n");
2767                                         goto defaults;
2768                                 }
2769                         }
2770                 }
2771
2772                 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2773                 goto defaults;
2774
2775         Page_found:
2776                 if (modepage == 8) {
2777                         sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2778                         sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2779                 } else {
2780                         sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2781                         sdkp->RCD = 0;
2782                 }
2783
2784                 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2785                 if (sdp->broken_fua) {
2786                         sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2787                         sdkp->DPOFUA = 0;
2788                 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2789                            !sdkp->device->use_16_for_rw) {
2790                         sd_first_printk(KERN_NOTICE, sdkp,
2791                                   "Uses READ/WRITE(6), disabling FUA\n");
2792                         sdkp->DPOFUA = 0;
2793                 }
2794
2795                 /* No cache flush allowed for write protected devices */
2796                 if (sdkp->WCE && sdkp->write_prot)
2797                         sdkp->WCE = 0;
2798
2799                 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2800                     old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2801                         sd_printk(KERN_NOTICE, sdkp,
2802                                   "Write cache: %s, read cache: %s, %s\n",
2803                                   sdkp->WCE ? "enabled" : "disabled",
2804                                   sdkp->RCD ? "disabled" : "enabled",
2805                                   sdkp->DPOFUA ? "supports DPO and FUA"
2806                                   : "doesn't support DPO or FUA");
2807
2808                 return;
2809         }
2810
2811 bad_sense:
2812         if (scsi_sense_valid(&sshdr) &&
2813             sshdr.sense_key == ILLEGAL_REQUEST &&
2814             sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2815                 /* Invalid field in CDB */
2816                 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2817         else
2818                 sd_first_printk(KERN_ERR, sdkp,
2819                                 "Asking for cache data failed\n");
2820
2821 defaults:
2822         if (sdp->wce_default_on) {
2823                 sd_first_printk(KERN_NOTICE, sdkp,
2824                                 "Assuming drive cache: write back\n");
2825                 sdkp->WCE = 1;
2826         } else {
2827                 sd_first_printk(KERN_ERR, sdkp,
2828                                 "Assuming drive cache: write through\n");
2829                 sdkp->WCE = 0;
2830         }
2831         sdkp->RCD = 0;
2832         sdkp->DPOFUA = 0;
2833 }
2834
2835 /*
2836  * The ATO bit indicates whether the DIF application tag is available
2837  * for use by the operating system.
2838  */
2839 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2840 {
2841         int res, offset;
2842         struct scsi_device *sdp = sdkp->device;
2843         struct scsi_mode_data data;
2844         struct scsi_sense_hdr sshdr;
2845
2846         if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2847                 return;
2848
2849         if (sdkp->protection_type == 0)
2850                 return;
2851
2852         res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2853                               SD_MAX_RETRIES, &data, &sshdr);
2854
2855         if (!scsi_status_is_good(res) || !data.header_length ||
2856             data.length < 6) {
2857                 sd_first_printk(KERN_WARNING, sdkp,
2858                           "getting Control mode page failed, assume no ATO\n");
2859
2860                 if (scsi_sense_valid(&sshdr))
2861                         sd_print_sense_hdr(sdkp, &sshdr);
2862
2863                 return;
2864         }
2865
2866         offset = data.header_length + data.block_descriptor_length;
2867
2868         if ((buffer[offset] & 0x3f) != 0x0a) {
2869                 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2870                 return;
2871         }
2872
2873         if ((buffer[offset + 5] & 0x80) == 0)
2874                 return;
2875
2876         sdkp->ATO = 1;
2877
2878         return;
2879 }
2880
2881 /**
2882  * sd_read_block_limits - Query disk device for preferred I/O sizes.
2883  * @sdkp: disk to query
2884  */
2885 static void sd_read_block_limits(struct scsi_disk *sdkp)
2886 {
2887         unsigned int sector_sz = sdkp->device->sector_size;
2888         const int vpd_len = 64;
2889         unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2890
2891         if (!buffer ||
2892             /* Block Limits VPD */
2893             scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2894                 goto out;
2895
2896         blk_queue_io_min(sdkp->disk->queue,
2897                          get_unaligned_be16(&buffer[6]) * sector_sz);
2898
2899         sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2900         sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2901
2902         if (buffer[3] == 0x3c) {
2903                 unsigned int lba_count, desc_count;
2904
2905                 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2906
2907                 if (!sdkp->lbpme)
2908                         goto out;
2909
2910                 lba_count = get_unaligned_be32(&buffer[20]);
2911                 desc_count = get_unaligned_be32(&buffer[24]);
2912
2913                 if (lba_count && desc_count)
2914                         sdkp->max_unmap_blocks = lba_count;
2915
2916                 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2917
2918                 if (buffer[32] & 0x80)
2919                         sdkp->unmap_alignment =
2920                                 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2921
2922                 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2923
2924                         if (sdkp->max_unmap_blocks)
2925                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2926                         else
2927                                 sd_config_discard(sdkp, SD_LBP_WS16);
2928
2929                 } else {        /* LBP VPD page tells us what to use */
2930                         if (sdkp->lbpu && sdkp->max_unmap_blocks)
2931                                 sd_config_discard(sdkp, SD_LBP_UNMAP);
2932                         else if (sdkp->lbpws)
2933                                 sd_config_discard(sdkp, SD_LBP_WS16);
2934                         else if (sdkp->lbpws10)
2935                                 sd_config_discard(sdkp, SD_LBP_WS10);
2936                         else
2937                                 sd_config_discard(sdkp, SD_LBP_DISABLE);
2938                 }
2939         }
2940
2941  out:
2942         kfree(buffer);
2943 }
2944
2945 /**
2946  * sd_read_block_characteristics - Query block dev. characteristics
2947  * @sdkp: disk to query
2948  */
2949 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2950 {
2951         struct request_queue *q = sdkp->disk->queue;
2952         unsigned char *buffer;
2953         u16 rot;
2954         const int vpd_len = 64;
2955
2956         buffer = kmalloc(vpd_len, GFP_KERNEL);
2957
2958         if (!buffer ||
2959             /* Block Device Characteristics VPD */
2960             scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2961                 goto out;
2962
2963         rot = get_unaligned_be16(&buffer[4]);
2964
2965         if (rot == 1) {
2966                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2967                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2968         } else {
2969                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2970                 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
2971         }
2972
2973         if (sdkp->device->type == TYPE_ZBC) {
2974                 /* Host-managed */
2975                 q->limits.zoned = BLK_ZONED_HM;
2976         } else {
2977                 sdkp->zoned = (buffer[8] >> 4) & 3;
2978                 if (sdkp->zoned == 1)
2979                         /* Host-aware */
2980                         q->limits.zoned = BLK_ZONED_HA;
2981                 else
2982                         /*
2983                          * Treat drive-managed devices as
2984                          * regular block devices.
2985                          */
2986                         q->limits.zoned = BLK_ZONED_NONE;
2987         }
2988         if (blk_queue_is_zoned(q) && sdkp->first_scan)
2989                 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2990                       q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2991
2992  out:
2993         kfree(buffer);
2994 }
2995
2996 /**
2997  * sd_read_block_provisioning - Query provisioning VPD page
2998  * @sdkp: disk to query
2999  */
3000 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3001 {
3002         unsigned char *buffer;
3003         const int vpd_len = 8;
3004
3005         if (sdkp->lbpme == 0)
3006                 return;
3007
3008         buffer = kmalloc(vpd_len, GFP_KERNEL);
3009
3010         if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3011                 goto out;
3012
3013         sdkp->lbpvpd    = 1;
3014         sdkp->lbpu      = (buffer[5] >> 7) & 1; /* UNMAP */
3015         sdkp->lbpws     = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3016         sdkp->lbpws10   = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3017
3018  out:
3019         kfree(buffer);
3020 }
3021
3022 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3023 {
3024         struct scsi_device *sdev = sdkp->device;
3025
3026         if (sdev->host->no_write_same) {
3027                 sdev->no_write_same = 1;
3028
3029                 return;
3030         }
3031
3032         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3033                 /* too large values might cause issues with arcmsr */
3034                 int vpd_buf_len = 64;
3035
3036                 sdev->no_report_opcodes = 1;
3037
3038                 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3039                  * CODES is unsupported and the device has an ATA
3040                  * Information VPD page (SAT).
3041                  */
3042                 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3043                         sdev->no_write_same = 1;
3044         }
3045
3046         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3047                 sdkp->ws16 = 1;
3048
3049         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3050                 sdkp->ws10 = 1;
3051 }
3052
3053 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3054 {
3055         struct scsi_device *sdev = sdkp->device;
3056
3057         if (!sdev->security_supported)
3058                 return;
3059
3060         if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3061                         SECURITY_PROTOCOL_IN) == 1 &&
3062             scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3063                         SECURITY_PROTOCOL_OUT) == 1)
3064                 sdkp->security = 1;
3065 }
3066
3067 /**
3068  *      sd_revalidate_disk - called the first time a new disk is seen,
3069  *      performs disk spin up, read_capacity, etc.
3070  *      @disk: struct gendisk we care about
3071  **/
3072 static int sd_revalidate_disk(struct gendisk *disk)
3073 {
3074         struct scsi_disk *sdkp = scsi_disk(disk);
3075         struct scsi_device *sdp = sdkp->device;
3076         struct request_queue *q = sdkp->disk->queue;
3077         sector_t old_capacity = sdkp->capacity;
3078         unsigned char *buffer;
3079         unsigned int dev_max, rw_max;
3080
3081         SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3082                                       "sd_revalidate_disk\n"));
3083
3084         /*
3085          * If the device is offline, don't try and read capacity or any
3086          * of the other niceties.
3087          */
3088         if (!scsi_device_online(sdp))
3089                 goto out;
3090
3091         buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3092         if (!buffer) {
3093                 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3094                           "allocation failure.\n");
3095                 goto out;
3096         }
3097
3098         sd_spinup_disk(sdkp);
3099
3100         /*
3101          * Without media there is no reason to ask; moreover, some devices
3102          * react badly if we do.
3103          */
3104         if (sdkp->media_present) {
3105                 sd_read_capacity(sdkp, buffer);
3106
3107                 if (scsi_device_supports_vpd(sdp)) {
3108                         sd_read_block_provisioning(sdkp);
3109                         sd_read_block_limits(sdkp);
3110                         sd_read_block_characteristics(sdkp);
3111                         sd_zbc_read_zones(sdkp, buffer);
3112                 }
3113
3114                 sd_print_capacity(sdkp, old_capacity);
3115
3116                 sd_read_write_protect_flag(sdkp, buffer);
3117                 sd_read_cache_type(sdkp, buffer);
3118                 sd_read_app_tag_own(sdkp, buffer);
3119                 sd_read_write_same(sdkp, buffer);
3120                 sd_read_security(sdkp, buffer);
3121         }
3122
3123         /*
3124          * We now have all cache related info, determine how we deal
3125          * with flush requests.
3126          */
3127         sd_set_flush_flag(sdkp);
3128
3129         /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3130         dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3131
3132         /* Some devices report a maximum block count for READ/WRITE requests. */
3133         dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3134         q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3135
3136         /*
3137          * Determine the device's preferred I/O size for reads and writes
3138          * unless the reported value is unreasonably small, large, or
3139          * garbage.
3140          */
3141         if (sdkp->opt_xfer_blocks &&
3142             sdkp->opt_xfer_blocks <= dev_max &&
3143             sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3144             logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3145                 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3146                 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3147         } else
3148                 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3149                                       (sector_t)BLK_DEF_MAX_SECTORS);
3150
3151         /* Do not exceed controller limit */
3152         rw_max = min(rw_max, queue_max_hw_sectors(q));
3153
3154         /*
3155          * Only update max_sectors if previously unset or if the current value
3156          * exceeds the capabilities of the hardware.
3157          */
3158         if (sdkp->first_scan ||
3159             q->limits.max_sectors > q->limits.max_dev_sectors ||
3160             q->limits.max_sectors > q->limits.max_hw_sectors)
3161                 q->limits.max_sectors = rw_max;
3162
3163         sdkp->first_scan = 0;
3164
3165         set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3166         sd_config_write_same(sdkp);
3167         kfree(buffer);
3168
3169  out:
3170         return 0;
3171 }
3172
3173 /**
3174  *      sd_unlock_native_capacity - unlock native capacity
3175  *      @disk: struct gendisk to set capacity for
3176  *
3177  *      Block layer calls this function if it detects that partitions
3178  *      on @disk reach beyond the end of the device.  If the SCSI host
3179  *      implements ->unlock_native_capacity() method, it's invoked to
3180  *      give it a chance to adjust the device capacity.
3181  *
3182  *      CONTEXT:
3183  *      Defined by block layer.  Might sleep.
3184  */
3185 static void sd_unlock_native_capacity(struct gendisk *disk)
3186 {
3187         struct scsi_device *sdev = scsi_disk(disk)->device;
3188
3189         if (sdev->host->hostt->unlock_native_capacity)
3190                 sdev->host->hostt->unlock_native_capacity(sdev);
3191 }
3192
3193 /**
3194  *      sd_format_disk_name - format disk name
3195  *      @prefix: name prefix - ie. "sd" for SCSI disks
3196  *      @index: index of the disk to format name for
3197  *      @buf: output buffer
3198  *      @buflen: length of the output buffer
3199  *
3200  *      SCSI disk names starts at sda.  The 26th device is sdz and the
3201  *      27th is sdaa.  The last one for two lettered suffix is sdzz
3202  *      which is followed by sdaaa.
3203  *
3204  *      This is basically 26 base counting with one extra 'nil' entry
3205  *      at the beginning from the second digit on and can be
3206  *      determined using similar method as 26 base conversion with the
3207  *      index shifted -1 after each digit is computed.
3208  *
3209  *      CONTEXT:
3210  *      Don't care.
3211  *
3212  *      RETURNS:
3213  *      0 on success, -errno on failure.
3214  */
3215 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3216 {
3217         const int base = 'z' - 'a' + 1;
3218         char *begin = buf + strlen(prefix);
3219         char *end = buf + buflen;
3220         char *p;
3221         int unit;
3222
3223         p = end - 1;
3224         *p = '\0';
3225         unit = base;
3226         do {
3227                 if (p == begin)
3228                         return -EINVAL;
3229                 *--p = 'a' + (index % unit);
3230                 index = (index / unit) - 1;
3231         } while (index >= 0);
3232
3233         memmove(begin, p, end - p);
3234         memcpy(buf, prefix, strlen(prefix));
3235
3236         return 0;
3237 }
3238
3239 /*
3240  * The asynchronous part of sd_probe
3241  */
3242 static void sd_probe_async(void *data, async_cookie_t cookie)
3243 {
3244         struct scsi_disk *sdkp = data;
3245         struct scsi_device *sdp;
3246         struct gendisk *gd;
3247         u32 index;
3248         struct device *dev;
3249
3250         sdp = sdkp->device;
3251         gd = sdkp->disk;
3252         index = sdkp->index;
3253         dev = &sdp->sdev_gendev;
3254
3255         gd->major = sd_major((index & 0xf0) >> 4);
3256         gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3257
3258         gd->fops = &sd_fops;
3259         gd->private_data = &sdkp->driver;
3260         gd->queue = sdkp->device->request_queue;
3261
3262         /* defaults, until the device tells us otherwise */
3263         sdp->sector_size = 512;
3264         sdkp->capacity = 0;
3265         sdkp->media_present = 1;
3266         sdkp->write_prot = 0;
3267         sdkp->cache_override = 0;
3268         sdkp->WCE = 0;
3269         sdkp->RCD = 0;
3270         sdkp->ATO = 0;
3271         sdkp->first_scan = 1;
3272         sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3273
3274         sd_revalidate_disk(gd);
3275
3276         gd->flags = GENHD_FL_EXT_DEVT;
3277         if (sdp->removable) {
3278                 gd->flags |= GENHD_FL_REMOVABLE;
3279                 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3280         }
3281
3282         blk_pm_runtime_init(sdp->request_queue, dev);
3283         device_add_disk(dev, gd, NULL);
3284         if (sdkp->capacity)
3285                 sd_dif_config_host(sdkp);
3286
3287         sd_revalidate_disk(gd);
3288
3289         if (sdkp->security) {
3290                 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3291                 if (sdkp->opal_dev)
3292                         sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3293         }
3294
3295         sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3296                   sdp->removable ? "removable " : "");
3297         scsi_autopm_put_device(sdp);
3298         put_device(&sdkp->dev);
3299 }
3300
3301 /**
3302  *      sd_probe - called during driver initialization and whenever a
3303  *      new scsi device is attached to the system. It is called once
3304  *      for each scsi device (not just disks) present.
3305  *      @dev: pointer to device object
3306  *
3307  *      Returns 0 if successful (or not interested in this scsi device
3308  *      (e.g. scanner)); 1 when there is an error.
3309  *
3310  *      Note: this function is invoked from the scsi mid-level.
3311  *      This function sets up the mapping between a given
3312  *      <host,channel,id,lun> (found in sdp) and new device name
3313  *      (e.g. /dev/sda). More precisely it is the block device major
3314  *      and minor number that is chosen here.
3315  *
3316  *      Assume sd_probe is not re-entrant (for time being)
3317  *      Also think about sd_probe() and sd_remove() running coincidentally.
3318  **/
3319 static int sd_probe(struct device *dev)
3320 {
3321         struct scsi_device *sdp = to_scsi_device(dev);
3322         struct scsi_disk *sdkp;
3323         struct gendisk *gd;
3324         int index;
3325         int error;
3326
3327         scsi_autopm_get_device(sdp);
3328         error = -ENODEV;
3329         if (sdp->type != TYPE_DISK &&
3330             sdp->type != TYPE_ZBC &&
3331             sdp->type != TYPE_MOD &&
3332             sdp->type != TYPE_RBC)
3333                 goto out;
3334
3335 #ifndef CONFIG_BLK_DEV_ZONED
3336         if (sdp->type == TYPE_ZBC)
3337                 goto out;
3338 #endif
3339         SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3340                                         "sd_probe\n"));
3341
3342         error = -ENOMEM;
3343         sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3344         if (!sdkp)
3345                 goto out;
3346
3347         gd = alloc_disk(SD_MINORS);
3348         if (!gd)
3349                 goto out_free;
3350
3351         index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3352         if (index < 0) {
3353                 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3354                 goto out_put;
3355         }
3356
3357         error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3358         if (error) {
3359                 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3360                 goto out_free_index;
3361         }
3362
3363         sdkp->device = sdp;
3364         sdkp->driver = &sd_template;
3365         sdkp->disk = gd;
3366         sdkp->index = index;
3367         atomic_set(&sdkp->openers, 0);
3368         atomic_set(&sdkp->device->ioerr_cnt, 0);
3369
3370         if (!sdp->request_queue->rq_timeout) {
3371                 if (sdp->type != TYPE_MOD)
3372                         blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3373                 else
3374                         blk_queue_rq_timeout(sdp->request_queue,
3375                                              SD_MOD_TIMEOUT);
3376         }
3377
3378         device_initialize(&sdkp->dev);
3379         sdkp->dev.parent = dev;
3380         sdkp->dev.class = &sd_disk_class;
3381         dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3382
3383         error = device_add(&sdkp->dev);
3384         if (error)
3385                 goto out_free_index;
3386
3387         get_device(dev);
3388         dev_set_drvdata(dev, sdkp);
3389
3390         get_device(&sdkp->dev); /* prevent release before async_schedule */
3391         async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3392
3393         return 0;
3394
3395  out_free_index:
3396         ida_free(&sd_index_ida, index);
3397  out_put:
3398         put_disk(gd);
3399  out_free:
3400         kfree(sdkp);
3401  out:
3402         scsi_autopm_put_device(sdp);
3403         return error;
3404 }
3405
3406 /**
3407  *      sd_remove - called whenever a scsi disk (previously recognized by
3408  *      sd_probe) is detached from the system. It is called (potentially
3409  *      multiple times) during sd module unload.
3410  *      @dev: pointer to device object
3411  *
3412  *      Note: this function is invoked from the scsi mid-level.
3413  *      This function potentially frees up a device name (e.g. /dev/sdc)
3414  *      that could be re-used by a subsequent sd_probe().
3415  *      This function is not called when the built-in sd driver is "exit-ed".
3416  **/
3417 static int sd_remove(struct device *dev)
3418 {
3419         struct scsi_disk *sdkp;
3420         dev_t devt;
3421
3422         sdkp = dev_get_drvdata(dev);
3423         devt = disk_devt(sdkp->disk);
3424         scsi_autopm_get_device(sdkp->device);
3425
3426         async_synchronize_full_domain(&scsi_sd_pm_domain);
3427         async_synchronize_full_domain(&scsi_sd_probe_domain);
3428         device_del(&sdkp->dev);
3429         del_gendisk(sdkp->disk);
3430         sd_shutdown(dev);
3431
3432         free_opal_dev(sdkp->opal_dev);
3433
3434         blk_register_region(devt, SD_MINORS, NULL,
3435                             sd_default_probe, NULL, NULL);
3436
3437         mutex_lock(&sd_ref_mutex);
3438         dev_set_drvdata(dev, NULL);
3439         put_device(&sdkp->dev);
3440         mutex_unlock(&sd_ref_mutex);
3441
3442         return 0;
3443 }
3444
3445 /**
3446  *      scsi_disk_release - Called to free the scsi_disk structure
3447  *      @dev: pointer to embedded class device
3448  *
3449  *      sd_ref_mutex must be held entering this routine.  Because it is
3450  *      called on last put, you should always use the scsi_disk_get()
3451  *      scsi_disk_put() helpers which manipulate the semaphore directly
3452  *      and never do a direct put_device.
3453  **/
3454 static void scsi_disk_release(struct device *dev)
3455 {
3456         struct scsi_disk *sdkp = to_scsi_disk(dev);
3457         struct gendisk *disk = sdkp->disk;
3458
3459         ida_free(&sd_index_ida, sdkp->index);
3460
3461         disk->private_data = NULL;
3462         put_disk(disk);
3463         put_device(&sdkp->device->sdev_gendev);
3464
3465         kfree(sdkp);
3466 }
3467
3468 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3469 {
3470         unsigned char cmd[6] = { START_STOP };  /* START_VALID */
3471         struct scsi_sense_hdr sshdr;
3472         struct scsi_device *sdp = sdkp->device;
3473         int res;
3474
3475         if (start)
3476                 cmd[4] |= 1;    /* START */
3477
3478         if (sdp->start_stop_pwr_cond)
3479                 cmd[4] |= start ? 1 << 4 : 3 << 4;      /* Active or Standby */
3480
3481         if (!scsi_device_online(sdp))
3482                 return -ENODEV;
3483
3484         res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3485                         SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3486         if (res) {
3487                 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3488                 if (driver_byte(res) == DRIVER_SENSE)
3489                         sd_print_sense_hdr(sdkp, &sshdr);
3490                 if (scsi_sense_valid(&sshdr) &&
3491                         /* 0x3a is medium not present */
3492                         sshdr.asc == 0x3a)
3493                         res = 0;
3494         }
3495
3496         /* SCSI error codes must not go to the generic layer */
3497         if (res)
3498                 return -EIO;
3499
3500         return 0;
3501 }
3502
3503 /*
3504  * Send a SYNCHRONIZE CACHE instruction down to the device through
3505  * the normal SCSI command structure.  Wait for the command to
3506  * complete.
3507  */
3508 static void sd_shutdown(struct device *dev)
3509 {
3510         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3511
3512         if (!sdkp)
3513                 return;         /* this can happen */
3514
3515         if (pm_runtime_suspended(dev))
3516                 return;
3517
3518         if (sdkp->WCE && sdkp->media_present) {
3519                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3520                 sd_sync_cache(sdkp, NULL);
3521         }
3522
3523         if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3524                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3525                 sd_start_stop_device(sdkp, 0);
3526         }
3527 }
3528
3529 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3530 {
3531         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3532         struct scsi_sense_hdr sshdr;
3533         int ret = 0;
3534
3535         if (!sdkp)      /* E.g.: runtime suspend following sd_remove() */
3536                 return 0;
3537
3538         if (sdkp->WCE && sdkp->media_present) {
3539                 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3540                 ret = sd_sync_cache(sdkp, &sshdr);
3541
3542                 if (ret) {
3543                         /* ignore OFFLINE device */
3544                         if (ret == -ENODEV)
3545                                 return 0;
3546
3547                         if (!scsi_sense_valid(&sshdr) ||
3548                             sshdr.sense_key != ILLEGAL_REQUEST)
3549                                 return ret;
3550
3551                         /*
3552                          * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3553                          * doesn't support sync. There's not much to do and
3554                          * suspend shouldn't fail.
3555                          */
3556                         ret = 0;
3557                 }
3558         }
3559
3560         if (sdkp->device->manage_start_stop) {
3561                 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3562                 /* an error is not worth aborting a system sleep */
3563                 ret = sd_start_stop_device(sdkp, 0);
3564                 if (ignore_stop_errors)
3565                         ret = 0;
3566         }
3567
3568         return ret;
3569 }
3570
3571 static int sd_suspend_system(struct device *dev)
3572 {
3573         return sd_suspend_common(dev, true);
3574 }
3575
3576 static int sd_suspend_runtime(struct device *dev)
3577 {
3578         return sd_suspend_common(dev, false);
3579 }
3580
3581 static int sd_resume(struct device *dev)
3582 {
3583         struct scsi_disk *sdkp = dev_get_drvdata(dev);
3584         int ret;
3585
3586         if (!sdkp)      /* E.g.: runtime resume at the start of sd_probe() */
3587                 return 0;
3588
3589         if (!sdkp->device->manage_start_stop)
3590                 return 0;
3591
3592         sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3593         ret = sd_start_stop_device(sdkp, 1);
3594         if (!ret)
3595                 opal_unlock_from_suspend(sdkp->opal_dev);
3596         return ret;
3597 }
3598
3599 /**
3600  *      init_sd - entry point for this driver (both when built in or when
3601  *      a module).
3602  *
3603  *      Note: this function registers this driver with the scsi mid-level.
3604  **/
3605 static int __init init_sd(void)
3606 {
3607         int majors = 0, i, err;
3608
3609         SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3610
3611         for (i = 0; i < SD_MAJORS; i++) {
3612                 if (register_blkdev(sd_major(i), "sd") != 0)
3613                         continue;
3614                 majors++;
3615                 blk_register_region(sd_major(i), SD_MINORS, NULL,
3616                                     sd_default_probe, NULL, NULL);
3617         }
3618
3619         if (!majors)
3620                 return -ENODEV;
3621
3622         err = class_register(&sd_disk_class);
3623         if (err)
3624                 goto err_out;
3625
3626         sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3627                                          0, 0, NULL);
3628         if (!sd_cdb_cache) {
3629                 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3630                 err = -ENOMEM;
3631                 goto err_out_class;
3632         }
3633
3634         sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3635         if (!sd_cdb_pool) {
3636                 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3637                 err = -ENOMEM;
3638                 goto err_out_cache;
3639         }
3640
3641         sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3642         if (!sd_page_pool) {
3643                 printk(KERN_ERR "sd: can't init discard page pool\n");
3644                 err = -ENOMEM;
3645                 goto err_out_ppool;
3646         }
3647
3648         err = scsi_register_driver(&sd_template.gendrv);
3649         if (err)
3650                 goto err_out_driver;
3651
3652         return 0;
3653
3654 err_out_driver:
3655         mempool_destroy(sd_page_pool);
3656
3657 err_out_ppool:
3658         mempool_destroy(sd_cdb_pool);
3659
3660 err_out_cache:
3661         kmem_cache_destroy(sd_cdb_cache);
3662
3663 err_out_class:
3664         class_unregister(&sd_disk_class);
3665 err_out:
3666         for (i = 0; i < SD_MAJORS; i++)
3667                 unregister_blkdev(sd_major(i), "sd");
3668         return err;
3669 }
3670
3671 /**
3672  *      exit_sd - exit point for this driver (when it is a module).
3673  *
3674  *      Note: this function unregisters this driver from the scsi mid-level.
3675  **/
3676 static void __exit exit_sd(void)
3677 {
3678         int i;
3679
3680         SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3681
3682         scsi_unregister_driver(&sd_template.gendrv);
3683         mempool_destroy(sd_cdb_pool);
3684         mempool_destroy(sd_page_pool);
3685         kmem_cache_destroy(sd_cdb_cache);
3686
3687         class_unregister(&sd_disk_class);
3688
3689         for (i = 0; i < SD_MAJORS; i++) {
3690                 blk_unregister_region(sd_major(i), SD_MINORS);
3691                 unregister_blkdev(sd_major(i), "sd");
3692         }
3693 }
3694
3695 module_init(init_sd);
3696 module_exit(exit_sd);
3697
3698 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3699                                struct scsi_sense_hdr *sshdr)
3700 {
3701         scsi_print_sense_hdr(sdkp->device,
3702                              sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3703 }
3704
3705 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3706                             int result)
3707 {
3708         const char *hb_string = scsi_hostbyte_string(result);
3709         const char *db_string = scsi_driverbyte_string(result);
3710
3711         if (hb_string || db_string)
3712                 sd_printk(KERN_INFO, sdkp,
3713                           "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3714                           hb_string ? hb_string : "invalid",
3715                           db_string ? db_string : "invalid");
3716         else
3717                 sd_printk(KERN_INFO, sdkp,
3718                           "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3719                           msg, host_byte(result), driver_byte(result));
3720 }
3721