drivers/scsi/sd.c

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