1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * sd.c Copyright (C) 1992 Drew Eckhardt 4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale 5 * 6 * Linux scsi disk driver 7 * Initial versions: Drew Eckhardt 8 * Subsequent revisions: Eric Youngdale 9 * Modification history: 10 * - Drew Eckhardt <drew@colorado.edu> original 11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple 12 * outstanding request, and other enhancements. 13 * Support loadable low-level scsi drivers. 14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using 15 * eight major numbers. 16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs. 17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in 18 * sd_init and cleanups. 19 * - Alex Davis <letmein@erols.com> Fix problem where partition info 20 * not being read in sd_open. Fix problem where removable media 21 * could be ejected after sd_open. 22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x 23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox 24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>: 25 * Support 32k/1M disks. 26 * 27 * Logging policy (needs CONFIG_SCSI_LOGGING defined): 28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2 29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1 30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1 31 * - entering other commands: SCSI_LOG_HLQUEUE level 3 32 * Note: when the logging level is set by the user, it must be greater 33 * than the level indicated above to trigger output. 34 */ 35 36 #include <linux/module.h> 37 #include <linux/fs.h> 38 #include <linux/kernel.h> 39 #include <linux/mm.h> 40 #include <linux/bio.h> 41 #include <linux/hdreg.h> 42 #include <linux/errno.h> 43 #include <linux/idr.h> 44 #include <linux/interrupt.h> 45 #include <linux/init.h> 46 #include <linux/blkdev.h> 47 #include <linux/blkpg.h> 48 #include <linux/blk-pm.h> 49 #include <linux/delay.h> 50 #include <linux/rw_hint.h> 51 #include <linux/major.h> 52 #include <linux/mutex.h> 53 #include <linux/string_helpers.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_devinfo.h> 67 #include <scsi/scsi_driver.h> 68 #include <scsi/scsi_eh.h> 69 #include <scsi/scsi_host.h> 70 #include <scsi/scsi_ioctl.h> 71 #include <scsi/scsicam.h> 72 #include <scsi/scsi_common.h> 73 74 #include "sd.h" 75 #include "scsi_priv.h" 76 #include "scsi_logging.h" 77 78 MODULE_AUTHOR("Eric Youngdale"); 79 MODULE_DESCRIPTION("SCSI disk (sd) driver"); 80 MODULE_LICENSE("GPL"); 81 82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR); 83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR); 84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR); 85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR); 86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR); 87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR); 88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR); 89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR); 90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR); 91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR); 92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR); 93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR); 94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR); 95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR); 96 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR); 97 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR); 98 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK); 99 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD); 100 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC); 101 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC); 102 103 #define SD_MINORS 16 104 105 static void sd_config_discard(struct scsi_disk *, unsigned int); 106 static void sd_config_write_same(struct scsi_disk *); 107 static int sd_revalidate_disk(struct gendisk *); 108 static void sd_unlock_native_capacity(struct gendisk *disk); 109 static void sd_shutdown(struct device *); 110 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer); 111 static void scsi_disk_release(struct device *cdev); 112 113 static DEFINE_IDA(sd_index_ida); 114 115 static mempool_t *sd_page_pool; 116 static struct lock_class_key sd_bio_compl_lkclass; 117 118 static const char *sd_cache_types[] = { 119 "write through", "none", "write back", 120 "write back, no read (daft)" 121 }; 122 123 static void sd_set_flush_flag(struct scsi_disk *sdkp) 124 { 125 bool wc = false, fua = false; 126 127 if (sdkp->WCE) { 128 wc = true; 129 if (sdkp->DPOFUA) 130 fua = true; 131 } 132 133 blk_queue_write_cache(sdkp->disk->queue, wc, fua); 134 } 135 136 static ssize_t 137 cache_type_store(struct device *dev, struct device_attribute *attr, 138 const char *buf, size_t count) 139 { 140 int ct, rcd, wce, sp; 141 struct scsi_disk *sdkp = to_scsi_disk(dev); 142 struct scsi_device *sdp = sdkp->device; 143 char buffer[64]; 144 char *buffer_data; 145 struct scsi_mode_data data; 146 struct scsi_sense_hdr sshdr; 147 static const char temp[] = "temporary "; 148 int len, ret; 149 150 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 151 /* no cache control on RBC devices; theoretically they 152 * can do it, but there's probably so many exceptions 153 * it's not worth the risk */ 154 return -EINVAL; 155 156 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) { 157 buf += sizeof(temp) - 1; 158 sdkp->cache_override = 1; 159 } else { 160 sdkp->cache_override = 0; 161 } 162 163 ct = sysfs_match_string(sd_cache_types, buf); 164 if (ct < 0) 165 return -EINVAL; 166 167 rcd = ct & 0x01 ? 1 : 0; 168 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0; 169 170 if (sdkp->cache_override) { 171 sdkp->WCE = wce; 172 sdkp->RCD = rcd; 173 sd_set_flush_flag(sdkp); 174 return count; 175 } 176 177 if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT, 178 sdkp->max_retries, &data, NULL)) 179 return -EINVAL; 180 len = min_t(size_t, sizeof(buffer), data.length - data.header_length - 181 data.block_descriptor_length); 182 buffer_data = buffer + data.header_length + 183 data.block_descriptor_length; 184 buffer_data[2] &= ~0x05; 185 buffer_data[2] |= wce << 2 | rcd; 186 sp = buffer_data[0] & 0x80 ? 1 : 0; 187 buffer_data[0] &= ~0x80; 188 189 /* 190 * Ensure WP, DPOFUA, and RESERVED fields are cleared in 191 * received mode parameter buffer before doing MODE SELECT. 192 */ 193 data.device_specific = 0; 194 195 ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT, 196 sdkp->max_retries, &data, &sshdr); 197 if (ret) { 198 if (ret > 0 && scsi_sense_valid(&sshdr)) 199 sd_print_sense_hdr(sdkp, &sshdr); 200 return -EINVAL; 201 } 202 sd_revalidate_disk(sdkp->disk); 203 return count; 204 } 205 206 static ssize_t 207 manage_start_stop_show(struct device *dev, 208 struct device_attribute *attr, char *buf) 209 { 210 struct scsi_disk *sdkp = to_scsi_disk(dev); 211 struct scsi_device *sdp = sdkp->device; 212 213 return sysfs_emit(buf, "%u\n", 214 sdp->manage_system_start_stop && 215 sdp->manage_runtime_start_stop && 216 sdp->manage_shutdown); 217 } 218 static DEVICE_ATTR_RO(manage_start_stop); 219 220 static ssize_t 221 manage_system_start_stop_show(struct device *dev, 222 struct device_attribute *attr, char *buf) 223 { 224 struct scsi_disk *sdkp = to_scsi_disk(dev); 225 struct scsi_device *sdp = sdkp->device; 226 227 return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop); 228 } 229 230 static ssize_t 231 manage_system_start_stop_store(struct device *dev, 232 struct device_attribute *attr, 233 const char *buf, size_t count) 234 { 235 struct scsi_disk *sdkp = to_scsi_disk(dev); 236 struct scsi_device *sdp = sdkp->device; 237 bool v; 238 239 if (!capable(CAP_SYS_ADMIN)) 240 return -EACCES; 241 242 if (kstrtobool(buf, &v)) 243 return -EINVAL; 244 245 sdp->manage_system_start_stop = v; 246 247 return count; 248 } 249 static DEVICE_ATTR_RW(manage_system_start_stop); 250 251 static ssize_t 252 manage_runtime_start_stop_show(struct device *dev, 253 struct device_attribute *attr, char *buf) 254 { 255 struct scsi_disk *sdkp = to_scsi_disk(dev); 256 struct scsi_device *sdp = sdkp->device; 257 258 return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop); 259 } 260 261 static ssize_t 262 manage_runtime_start_stop_store(struct device *dev, 263 struct device_attribute *attr, 264 const char *buf, size_t count) 265 { 266 struct scsi_disk *sdkp = to_scsi_disk(dev); 267 struct scsi_device *sdp = sdkp->device; 268 bool v; 269 270 if (!capable(CAP_SYS_ADMIN)) 271 return -EACCES; 272 273 if (kstrtobool(buf, &v)) 274 return -EINVAL; 275 276 sdp->manage_runtime_start_stop = v; 277 278 return count; 279 } 280 static DEVICE_ATTR_RW(manage_runtime_start_stop); 281 282 static ssize_t manage_shutdown_show(struct device *dev, 283 struct device_attribute *attr, char *buf) 284 { 285 struct scsi_disk *sdkp = to_scsi_disk(dev); 286 struct scsi_device *sdp = sdkp->device; 287 288 return sysfs_emit(buf, "%u\n", sdp->manage_shutdown); 289 } 290 291 static ssize_t manage_shutdown_store(struct device *dev, 292 struct device_attribute *attr, 293 const char *buf, size_t count) 294 { 295 struct scsi_disk *sdkp = to_scsi_disk(dev); 296 struct scsi_device *sdp = sdkp->device; 297 bool v; 298 299 if (!capable(CAP_SYS_ADMIN)) 300 return -EACCES; 301 302 if (kstrtobool(buf, &v)) 303 return -EINVAL; 304 305 sdp->manage_shutdown = v; 306 307 return count; 308 } 309 static DEVICE_ATTR_RW(manage_shutdown); 310 311 static ssize_t 312 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf) 313 { 314 struct scsi_disk *sdkp = to_scsi_disk(dev); 315 316 return sprintf(buf, "%u\n", sdkp->device->allow_restart); 317 } 318 319 static ssize_t 320 allow_restart_store(struct device *dev, struct device_attribute *attr, 321 const char *buf, size_t count) 322 { 323 bool v; 324 struct scsi_disk *sdkp = to_scsi_disk(dev); 325 struct scsi_device *sdp = sdkp->device; 326 327 if (!capable(CAP_SYS_ADMIN)) 328 return -EACCES; 329 330 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 331 return -EINVAL; 332 333 if (kstrtobool(buf, &v)) 334 return -EINVAL; 335 336 sdp->allow_restart = v; 337 338 return count; 339 } 340 static DEVICE_ATTR_RW(allow_restart); 341 342 static ssize_t 343 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf) 344 { 345 struct scsi_disk *sdkp = to_scsi_disk(dev); 346 int ct = sdkp->RCD + 2*sdkp->WCE; 347 348 return sprintf(buf, "%s\n", sd_cache_types[ct]); 349 } 350 static DEVICE_ATTR_RW(cache_type); 351 352 static ssize_t 353 FUA_show(struct device *dev, struct device_attribute *attr, char *buf) 354 { 355 struct scsi_disk *sdkp = to_scsi_disk(dev); 356 357 return sprintf(buf, "%u\n", sdkp->DPOFUA); 358 } 359 static DEVICE_ATTR_RO(FUA); 360 361 static ssize_t 362 protection_type_show(struct device *dev, struct device_attribute *attr, 363 char *buf) 364 { 365 struct scsi_disk *sdkp = to_scsi_disk(dev); 366 367 return sprintf(buf, "%u\n", sdkp->protection_type); 368 } 369 370 static ssize_t 371 protection_type_store(struct device *dev, struct device_attribute *attr, 372 const char *buf, size_t count) 373 { 374 struct scsi_disk *sdkp = to_scsi_disk(dev); 375 unsigned int val; 376 int err; 377 378 if (!capable(CAP_SYS_ADMIN)) 379 return -EACCES; 380 381 err = kstrtouint(buf, 10, &val); 382 383 if (err) 384 return err; 385 386 if (val <= T10_PI_TYPE3_PROTECTION) 387 sdkp->protection_type = val; 388 389 return count; 390 } 391 static DEVICE_ATTR_RW(protection_type); 392 393 static ssize_t 394 protection_mode_show(struct device *dev, struct device_attribute *attr, 395 char *buf) 396 { 397 struct scsi_disk *sdkp = to_scsi_disk(dev); 398 struct scsi_device *sdp = sdkp->device; 399 unsigned int dif, dix; 400 401 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type); 402 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type); 403 404 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) { 405 dif = 0; 406 dix = 1; 407 } 408 409 if (!dif && !dix) 410 return sprintf(buf, "none\n"); 411 412 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif); 413 } 414 static DEVICE_ATTR_RO(protection_mode); 415 416 static ssize_t 417 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf) 418 { 419 struct scsi_disk *sdkp = to_scsi_disk(dev); 420 421 return sprintf(buf, "%u\n", sdkp->ATO); 422 } 423 static DEVICE_ATTR_RO(app_tag_own); 424 425 static ssize_t 426 thin_provisioning_show(struct device *dev, struct device_attribute *attr, 427 char *buf) 428 { 429 struct scsi_disk *sdkp = to_scsi_disk(dev); 430 431 return sprintf(buf, "%u\n", sdkp->lbpme); 432 } 433 static DEVICE_ATTR_RO(thin_provisioning); 434 435 /* sysfs_match_string() requires dense arrays */ 436 static const char *lbp_mode[] = { 437 [SD_LBP_FULL] = "full", 438 [SD_LBP_UNMAP] = "unmap", 439 [SD_LBP_WS16] = "writesame_16", 440 [SD_LBP_WS10] = "writesame_10", 441 [SD_LBP_ZERO] = "writesame_zero", 442 [SD_LBP_DISABLE] = "disabled", 443 }; 444 445 static ssize_t 446 provisioning_mode_show(struct device *dev, struct device_attribute *attr, 447 char *buf) 448 { 449 struct scsi_disk *sdkp = to_scsi_disk(dev); 450 451 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]); 452 } 453 454 static ssize_t 455 provisioning_mode_store(struct device *dev, struct device_attribute *attr, 456 const char *buf, size_t count) 457 { 458 struct scsi_disk *sdkp = to_scsi_disk(dev); 459 struct scsi_device *sdp = sdkp->device; 460 int mode; 461 462 if (!capable(CAP_SYS_ADMIN)) 463 return -EACCES; 464 465 if (sd_is_zoned(sdkp)) { 466 sd_config_discard(sdkp, SD_LBP_DISABLE); 467 return count; 468 } 469 470 if (sdp->type != TYPE_DISK) 471 return -EINVAL; 472 473 mode = sysfs_match_string(lbp_mode, buf); 474 if (mode < 0) 475 return -EINVAL; 476 477 sd_config_discard(sdkp, mode); 478 479 return count; 480 } 481 static DEVICE_ATTR_RW(provisioning_mode); 482 483 /* sysfs_match_string() requires dense arrays */ 484 static const char *zeroing_mode[] = { 485 [SD_ZERO_WRITE] = "write", 486 [SD_ZERO_WS] = "writesame", 487 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap", 488 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap", 489 }; 490 491 static ssize_t 492 zeroing_mode_show(struct device *dev, struct device_attribute *attr, 493 char *buf) 494 { 495 struct scsi_disk *sdkp = to_scsi_disk(dev); 496 497 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]); 498 } 499 500 static ssize_t 501 zeroing_mode_store(struct device *dev, struct device_attribute *attr, 502 const char *buf, size_t count) 503 { 504 struct scsi_disk *sdkp = to_scsi_disk(dev); 505 int mode; 506 507 if (!capable(CAP_SYS_ADMIN)) 508 return -EACCES; 509 510 mode = sysfs_match_string(zeroing_mode, buf); 511 if (mode < 0) 512 return -EINVAL; 513 514 sdkp->zeroing_mode = mode; 515 516 return count; 517 } 518 static DEVICE_ATTR_RW(zeroing_mode); 519 520 static ssize_t 521 max_medium_access_timeouts_show(struct device *dev, 522 struct device_attribute *attr, char *buf) 523 { 524 struct scsi_disk *sdkp = to_scsi_disk(dev); 525 526 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts); 527 } 528 529 static ssize_t 530 max_medium_access_timeouts_store(struct device *dev, 531 struct device_attribute *attr, const char *buf, 532 size_t count) 533 { 534 struct scsi_disk *sdkp = to_scsi_disk(dev); 535 int err; 536 537 if (!capable(CAP_SYS_ADMIN)) 538 return -EACCES; 539 540 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts); 541 542 return err ? err : count; 543 } 544 static DEVICE_ATTR_RW(max_medium_access_timeouts); 545 546 static ssize_t 547 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr, 548 char *buf) 549 { 550 struct scsi_disk *sdkp = to_scsi_disk(dev); 551 552 return sprintf(buf, "%u\n", sdkp->max_ws_blocks); 553 } 554 555 static ssize_t 556 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr, 557 const char *buf, size_t count) 558 { 559 struct scsi_disk *sdkp = to_scsi_disk(dev); 560 struct scsi_device *sdp = sdkp->device; 561 unsigned long max; 562 int err; 563 564 if (!capable(CAP_SYS_ADMIN)) 565 return -EACCES; 566 567 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 568 return -EINVAL; 569 570 err = kstrtoul(buf, 10, &max); 571 572 if (err) 573 return err; 574 575 if (max == 0) 576 sdp->no_write_same = 1; 577 else if (max <= SD_MAX_WS16_BLOCKS) { 578 sdp->no_write_same = 0; 579 sdkp->max_ws_blocks = max; 580 } 581 582 sd_config_write_same(sdkp); 583 584 return count; 585 } 586 static DEVICE_ATTR_RW(max_write_same_blocks); 587 588 static ssize_t 589 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf) 590 { 591 struct scsi_disk *sdkp = to_scsi_disk(dev); 592 593 if (sdkp->device->type == TYPE_ZBC) 594 return sprintf(buf, "host-managed\n"); 595 if (sdkp->zoned == 1) 596 return sprintf(buf, "host-aware\n"); 597 if (sdkp->zoned == 2) 598 return sprintf(buf, "drive-managed\n"); 599 return sprintf(buf, "none\n"); 600 } 601 static DEVICE_ATTR_RO(zoned_cap); 602 603 static ssize_t 604 max_retries_store(struct device *dev, struct device_attribute *attr, 605 const char *buf, size_t count) 606 { 607 struct scsi_disk *sdkp = to_scsi_disk(dev); 608 struct scsi_device *sdev = sdkp->device; 609 int retries, err; 610 611 err = kstrtoint(buf, 10, &retries); 612 if (err) 613 return err; 614 615 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) { 616 sdkp->max_retries = retries; 617 return count; 618 } 619 620 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n", 621 SD_MAX_RETRIES); 622 return -EINVAL; 623 } 624 625 static ssize_t 626 max_retries_show(struct device *dev, struct device_attribute *attr, 627 char *buf) 628 { 629 struct scsi_disk *sdkp = to_scsi_disk(dev); 630 631 return sprintf(buf, "%d\n", sdkp->max_retries); 632 } 633 634 static DEVICE_ATTR_RW(max_retries); 635 636 static struct attribute *sd_disk_attrs[] = { 637 &dev_attr_cache_type.attr, 638 &dev_attr_FUA.attr, 639 &dev_attr_allow_restart.attr, 640 &dev_attr_manage_start_stop.attr, 641 &dev_attr_manage_system_start_stop.attr, 642 &dev_attr_manage_runtime_start_stop.attr, 643 &dev_attr_manage_shutdown.attr, 644 &dev_attr_protection_type.attr, 645 &dev_attr_protection_mode.attr, 646 &dev_attr_app_tag_own.attr, 647 &dev_attr_thin_provisioning.attr, 648 &dev_attr_provisioning_mode.attr, 649 &dev_attr_zeroing_mode.attr, 650 &dev_attr_max_write_same_blocks.attr, 651 &dev_attr_max_medium_access_timeouts.attr, 652 &dev_attr_zoned_cap.attr, 653 &dev_attr_max_retries.attr, 654 NULL, 655 }; 656 ATTRIBUTE_GROUPS(sd_disk); 657 658 static struct class sd_disk_class = { 659 .name = "scsi_disk", 660 .dev_release = scsi_disk_release, 661 .dev_groups = sd_disk_groups, 662 }; 663 664 /* 665 * Don't request a new module, as that could deadlock in multipath 666 * environment. 667 */ 668 static void sd_default_probe(dev_t devt) 669 { 670 } 671 672 /* 673 * Device no to disk mapping: 674 * 675 * major disc2 disc p1 676 * |............|.............|....|....| <- dev_t 677 * 31 20 19 8 7 4 3 0 678 * 679 * Inside a major, we have 16k disks, however mapped non- 680 * contiguously. The first 16 disks are for major0, the next 681 * ones with major1, ... Disk 256 is for major0 again, disk 272 682 * for major1, ... 683 * As we stay compatible with our numbering scheme, we can reuse 684 * the well-know SCSI majors 8, 65--71, 136--143. 685 */ 686 static int sd_major(int major_idx) 687 { 688 switch (major_idx) { 689 case 0: 690 return SCSI_DISK0_MAJOR; 691 case 1 ... 7: 692 return SCSI_DISK1_MAJOR + major_idx - 1; 693 case 8 ... 15: 694 return SCSI_DISK8_MAJOR + major_idx - 8; 695 default: 696 BUG(); 697 return 0; /* shut up gcc */ 698 } 699 } 700 701 #ifdef CONFIG_BLK_SED_OPAL 702 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, 703 size_t len, bool send) 704 { 705 struct scsi_disk *sdkp = data; 706 struct scsi_device *sdev = sdkp->device; 707 u8 cdb[12] = { 0, }; 708 const struct scsi_exec_args exec_args = { 709 .req_flags = BLK_MQ_REQ_PM, 710 }; 711 int ret; 712 713 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN; 714 cdb[1] = secp; 715 put_unaligned_be16(spsp, &cdb[2]); 716 put_unaligned_be32(len, &cdb[6]); 717 718 ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 719 buffer, len, SD_TIMEOUT, sdkp->max_retries, 720 &exec_args); 721 return ret <= 0 ? ret : -EIO; 722 } 723 #endif /* CONFIG_BLK_SED_OPAL */ 724 725 /* 726 * Look up the DIX operation based on whether the command is read or 727 * write and whether dix and dif are enabled. 728 */ 729 static unsigned int sd_prot_op(bool write, bool dix, bool dif) 730 { 731 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */ 732 static const unsigned int ops[] = { /* wrt dix dif */ 733 SCSI_PROT_NORMAL, /* 0 0 0 */ 734 SCSI_PROT_READ_STRIP, /* 0 0 1 */ 735 SCSI_PROT_READ_INSERT, /* 0 1 0 */ 736 SCSI_PROT_READ_PASS, /* 0 1 1 */ 737 SCSI_PROT_NORMAL, /* 1 0 0 */ 738 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */ 739 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */ 740 SCSI_PROT_WRITE_PASS, /* 1 1 1 */ 741 }; 742 743 return ops[write << 2 | dix << 1 | dif]; 744 } 745 746 /* 747 * Returns a mask of the protection flags that are valid for a given DIX 748 * operation. 749 */ 750 static unsigned int sd_prot_flag_mask(unsigned int prot_op) 751 { 752 static const unsigned int flag_mask[] = { 753 [SCSI_PROT_NORMAL] = 0, 754 755 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI | 756 SCSI_PROT_GUARD_CHECK | 757 SCSI_PROT_REF_CHECK | 758 SCSI_PROT_REF_INCREMENT, 759 760 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT | 761 SCSI_PROT_IP_CHECKSUM, 762 763 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI | 764 SCSI_PROT_GUARD_CHECK | 765 SCSI_PROT_REF_CHECK | 766 SCSI_PROT_REF_INCREMENT | 767 SCSI_PROT_IP_CHECKSUM, 768 769 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI | 770 SCSI_PROT_REF_INCREMENT, 771 772 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK | 773 SCSI_PROT_REF_CHECK | 774 SCSI_PROT_REF_INCREMENT | 775 SCSI_PROT_IP_CHECKSUM, 776 777 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI | 778 SCSI_PROT_GUARD_CHECK | 779 SCSI_PROT_REF_CHECK | 780 SCSI_PROT_REF_INCREMENT | 781 SCSI_PROT_IP_CHECKSUM, 782 }; 783 784 return flag_mask[prot_op]; 785 } 786 787 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd, 788 unsigned int dix, unsigned int dif) 789 { 790 struct request *rq = scsi_cmd_to_rq(scmd); 791 struct bio *bio = rq->bio; 792 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif); 793 unsigned int protect = 0; 794 795 if (dix) { /* DIX Type 0, 1, 2, 3 */ 796 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM)) 797 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM; 798 799 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false) 800 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK; 801 } 802 803 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */ 804 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT; 805 806 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false) 807 scmd->prot_flags |= SCSI_PROT_REF_CHECK; 808 } 809 810 if (dif) { /* DIX/DIF Type 1, 2, 3 */ 811 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI; 812 813 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK)) 814 protect = 3 << 5; /* Disable target PI checking */ 815 else 816 protect = 1 << 5; /* Enable target PI checking */ 817 } 818 819 scsi_set_prot_op(scmd, prot_op); 820 scsi_set_prot_type(scmd, dif); 821 scmd->prot_flags &= sd_prot_flag_mask(prot_op); 822 823 return protect; 824 } 825 826 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode) 827 { 828 struct request_queue *q = sdkp->disk->queue; 829 unsigned int logical_block_size = sdkp->device->sector_size; 830 unsigned int max_blocks = 0; 831 832 q->limits.discard_alignment = 833 sdkp->unmap_alignment * logical_block_size; 834 q->limits.discard_granularity = 835 max(sdkp->physical_block_size, 836 sdkp->unmap_granularity * logical_block_size); 837 sdkp->provisioning_mode = mode; 838 839 switch (mode) { 840 841 case SD_LBP_FULL: 842 case SD_LBP_DISABLE: 843 blk_queue_max_discard_sectors(q, 0); 844 return; 845 846 case SD_LBP_UNMAP: 847 max_blocks = min_not_zero(sdkp->max_unmap_blocks, 848 (u32)SD_MAX_WS16_BLOCKS); 849 break; 850 851 case SD_LBP_WS16: 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_WS16_BLOCKS); 858 break; 859 860 case SD_LBP_WS10: 861 if (sdkp->device->unmap_limit_for_ws) 862 max_blocks = sdkp->max_unmap_blocks; 863 else 864 max_blocks = sdkp->max_ws_blocks; 865 866 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS); 867 break; 868 869 case SD_LBP_ZERO: 870 max_blocks = min_not_zero(sdkp->max_ws_blocks, 871 (u32)SD_MAX_WS10_BLOCKS); 872 break; 873 } 874 875 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9)); 876 } 877 878 static void *sd_set_special_bvec(struct request *rq, unsigned int data_len) 879 { 880 struct page *page; 881 882 page = mempool_alloc(sd_page_pool, GFP_ATOMIC); 883 if (!page) 884 return NULL; 885 clear_highpage(page); 886 bvec_set_page(&rq->special_vec, page, data_len, 0); 887 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 888 return bvec_virt(&rq->special_vec); 889 } 890 891 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd) 892 { 893 struct scsi_device *sdp = cmd->device; 894 struct request *rq = scsi_cmd_to_rq(cmd); 895 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 896 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 897 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 898 unsigned int data_len = 24; 899 char *buf; 900 901 buf = sd_set_special_bvec(rq, data_len); 902 if (!buf) 903 return BLK_STS_RESOURCE; 904 905 cmd->cmd_len = 10; 906 cmd->cmnd[0] = UNMAP; 907 cmd->cmnd[8] = 24; 908 909 put_unaligned_be16(6 + 16, &buf[0]); 910 put_unaligned_be16(16, &buf[2]); 911 put_unaligned_be64(lba, &buf[8]); 912 put_unaligned_be32(nr_blocks, &buf[16]); 913 914 cmd->allowed = sdkp->max_retries; 915 cmd->transfersize = data_len; 916 rq->timeout = SD_TIMEOUT; 917 918 return scsi_alloc_sgtables(cmd); 919 } 920 921 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, 922 bool unmap) 923 { 924 struct scsi_device *sdp = cmd->device; 925 struct request *rq = scsi_cmd_to_rq(cmd); 926 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 927 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 928 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 929 u32 data_len = sdp->sector_size; 930 931 if (!sd_set_special_bvec(rq, data_len)) 932 return BLK_STS_RESOURCE; 933 934 cmd->cmd_len = 16; 935 cmd->cmnd[0] = WRITE_SAME_16; 936 if (unmap) 937 cmd->cmnd[1] = 0x8; /* UNMAP */ 938 put_unaligned_be64(lba, &cmd->cmnd[2]); 939 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]); 940 941 cmd->allowed = sdkp->max_retries; 942 cmd->transfersize = data_len; 943 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT; 944 945 return scsi_alloc_sgtables(cmd); 946 } 947 948 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, 949 bool unmap) 950 { 951 struct scsi_device *sdp = cmd->device; 952 struct request *rq = scsi_cmd_to_rq(cmd); 953 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 954 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 955 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 956 u32 data_len = sdp->sector_size; 957 958 if (!sd_set_special_bvec(rq, data_len)) 959 return BLK_STS_RESOURCE; 960 961 cmd->cmd_len = 10; 962 cmd->cmnd[0] = WRITE_SAME; 963 if (unmap) 964 cmd->cmnd[1] = 0x8; /* UNMAP */ 965 put_unaligned_be32(lba, &cmd->cmnd[2]); 966 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]); 967 968 cmd->allowed = sdkp->max_retries; 969 cmd->transfersize = data_len; 970 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT; 971 972 return scsi_alloc_sgtables(cmd); 973 } 974 975 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd) 976 { 977 struct request *rq = scsi_cmd_to_rq(cmd); 978 struct scsi_device *sdp = cmd->device; 979 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 980 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 981 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 982 983 if (!(rq->cmd_flags & REQ_NOUNMAP)) { 984 switch (sdkp->zeroing_mode) { 985 case SD_ZERO_WS16_UNMAP: 986 return sd_setup_write_same16_cmnd(cmd, true); 987 case SD_ZERO_WS10_UNMAP: 988 return sd_setup_write_same10_cmnd(cmd, true); 989 } 990 } 991 992 if (sdp->no_write_same) { 993 rq->rq_flags |= RQF_QUIET; 994 return BLK_STS_TARGET; 995 } 996 997 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) 998 return sd_setup_write_same16_cmnd(cmd, false); 999 1000 return sd_setup_write_same10_cmnd(cmd, false); 1001 } 1002 1003 static void sd_config_write_same(struct scsi_disk *sdkp) 1004 { 1005 struct request_queue *q = sdkp->disk->queue; 1006 unsigned int logical_block_size = sdkp->device->sector_size; 1007 1008 if (sdkp->device->no_write_same) { 1009 sdkp->max_ws_blocks = 0; 1010 goto out; 1011 } 1012 1013 /* Some devices can not handle block counts above 0xffff despite 1014 * supporting WRITE SAME(16). Consequently we default to 64k 1015 * blocks per I/O unless the device explicitly advertises a 1016 * bigger limit. 1017 */ 1018 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS) 1019 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks, 1020 (u32)SD_MAX_WS16_BLOCKS); 1021 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes) 1022 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks, 1023 (u32)SD_MAX_WS10_BLOCKS); 1024 else { 1025 sdkp->device->no_write_same = 1; 1026 sdkp->max_ws_blocks = 0; 1027 } 1028 1029 if (sdkp->lbprz && sdkp->lbpws) 1030 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP; 1031 else if (sdkp->lbprz && sdkp->lbpws10) 1032 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP; 1033 else if (sdkp->max_ws_blocks) 1034 sdkp->zeroing_mode = SD_ZERO_WS; 1035 else 1036 sdkp->zeroing_mode = SD_ZERO_WRITE; 1037 1038 if (sdkp->max_ws_blocks && 1039 sdkp->physical_block_size > logical_block_size) { 1040 /* 1041 * Reporting a maximum number of blocks that is not aligned 1042 * on the device physical size would cause a large write same 1043 * request to be split into physically unaligned chunks by 1044 * __blkdev_issue_write_zeroes() even if the caller of this 1045 * functions took care to align the large request. So make sure 1046 * the maximum reported is aligned to the device physical block 1047 * size. This is only an optional optimization for regular 1048 * disks, but this is mandatory to avoid failure of large write 1049 * same requests directed at sequential write required zones of 1050 * host-managed ZBC disks. 1051 */ 1052 sdkp->max_ws_blocks = 1053 round_down(sdkp->max_ws_blocks, 1054 bytes_to_logical(sdkp->device, 1055 sdkp->physical_block_size)); 1056 } 1057 1058 out: 1059 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks * 1060 (logical_block_size >> 9)); 1061 } 1062 1063 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd) 1064 { 1065 struct request *rq = scsi_cmd_to_rq(cmd); 1066 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 1067 1068 /* flush requests don't perform I/O, zero the S/G table */ 1069 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1070 1071 if (cmd->device->use_16_for_sync) { 1072 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16; 1073 cmd->cmd_len = 16; 1074 } else { 1075 cmd->cmnd[0] = SYNCHRONIZE_CACHE; 1076 cmd->cmd_len = 10; 1077 } 1078 cmd->transfersize = 0; 1079 cmd->allowed = sdkp->max_retries; 1080 1081 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER; 1082 return BLK_STS_OK; 1083 } 1084 1085 /** 1086 * sd_group_number() - Compute the GROUP NUMBER field 1087 * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER 1088 * field. 1089 * 1090 * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf): 1091 * 0: no relative lifetime. 1092 * 1: shortest relative lifetime. 1093 * 2: second shortest relative lifetime. 1094 * 3 - 0x3d: intermediate relative lifetimes. 1095 * 0x3e: second longest relative lifetime. 1096 * 0x3f: longest relative lifetime. 1097 */ 1098 static u8 sd_group_number(struct scsi_cmnd *cmd) 1099 { 1100 const struct request *rq = scsi_cmd_to_rq(cmd); 1101 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 1102 1103 if (!sdkp->rscs) 1104 return 0; 1105 1106 return min3((u32)rq->write_hint, (u32)sdkp->permanent_stream_count, 1107 0x3fu); 1108 } 1109 1110 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write, 1111 sector_t lba, unsigned int nr_blocks, 1112 unsigned char flags, unsigned int dld) 1113 { 1114 cmd->cmd_len = SD_EXT_CDB_SIZE; 1115 cmd->cmnd[0] = VARIABLE_LENGTH_CMD; 1116 cmd->cmnd[6] = sd_group_number(cmd); 1117 cmd->cmnd[7] = 0x18; /* Additional CDB len */ 1118 cmd->cmnd[9] = write ? WRITE_32 : READ_32; 1119 cmd->cmnd[10] = flags; 1120 cmd->cmnd[11] = dld & 0x07; 1121 put_unaligned_be64(lba, &cmd->cmnd[12]); 1122 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */ 1123 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]); 1124 1125 return BLK_STS_OK; 1126 } 1127 1128 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write, 1129 sector_t lba, unsigned int nr_blocks, 1130 unsigned char flags, unsigned int dld) 1131 { 1132 cmd->cmd_len = 16; 1133 cmd->cmnd[0] = write ? WRITE_16 : READ_16; 1134 cmd->cmnd[1] = flags | ((dld >> 2) & 0x01); 1135 cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd); 1136 cmd->cmnd[15] = 0; 1137 put_unaligned_be64(lba, &cmd->cmnd[2]); 1138 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]); 1139 1140 return BLK_STS_OK; 1141 } 1142 1143 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write, 1144 sector_t lba, unsigned int nr_blocks, 1145 unsigned char flags) 1146 { 1147 cmd->cmd_len = 10; 1148 cmd->cmnd[0] = write ? WRITE_10 : READ_10; 1149 cmd->cmnd[1] = flags; 1150 cmd->cmnd[6] = sd_group_number(cmd); 1151 cmd->cmnd[9] = 0; 1152 put_unaligned_be32(lba, &cmd->cmnd[2]); 1153 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]); 1154 1155 return BLK_STS_OK; 1156 } 1157 1158 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write, 1159 sector_t lba, unsigned int nr_blocks, 1160 unsigned char flags) 1161 { 1162 /* Avoid that 0 blocks gets translated into 256 blocks. */ 1163 if (WARN_ON_ONCE(nr_blocks == 0)) 1164 return BLK_STS_IOERR; 1165 1166 if (unlikely(flags & 0x8)) { 1167 /* 1168 * This happens only if this drive failed 10byte rw 1169 * command with ILLEGAL_REQUEST during operation and 1170 * thus turned off use_10_for_rw. 1171 */ 1172 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n"); 1173 return BLK_STS_IOERR; 1174 } 1175 1176 cmd->cmd_len = 6; 1177 cmd->cmnd[0] = write ? WRITE_6 : READ_6; 1178 cmd->cmnd[1] = (lba >> 16) & 0x1f; 1179 cmd->cmnd[2] = (lba >> 8) & 0xff; 1180 cmd->cmnd[3] = lba & 0xff; 1181 cmd->cmnd[4] = nr_blocks; 1182 cmd->cmnd[5] = 0; 1183 1184 return BLK_STS_OK; 1185 } 1186 1187 /* 1188 * Check if a command has a duration limit set. If it does, and the target 1189 * device supports CDL and the feature is enabled, return the limit 1190 * descriptor index to use. Return 0 (no limit) otherwise. 1191 */ 1192 static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd) 1193 { 1194 struct scsi_device *sdp = sdkp->device; 1195 int hint; 1196 1197 if (!sdp->cdl_supported || !sdp->cdl_enable) 1198 return 0; 1199 1200 /* 1201 * Use "no limit" if the request ioprio does not specify a duration 1202 * limit hint. 1203 */ 1204 hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd))); 1205 if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 || 1206 hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7) 1207 return 0; 1208 1209 return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1; 1210 } 1211 1212 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd) 1213 { 1214 struct request *rq = scsi_cmd_to_rq(cmd); 1215 struct scsi_device *sdp = cmd->device; 1216 struct scsi_disk *sdkp = scsi_disk(rq->q->disk); 1217 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 1218 sector_t threshold; 1219 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 1220 unsigned int mask = logical_to_sectors(sdp, 1) - 1; 1221 bool write = rq_data_dir(rq) == WRITE; 1222 unsigned char protect, fua; 1223 unsigned int dld; 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->q->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 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0; 1265 dix = scsi_prot_sg_count(cmd); 1266 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type); 1267 dld = sd_cdl_dld(sdkp, cmd); 1268 1269 if (dif || dix) 1270 protect = sd_setup_protect_cmnd(cmd, dix, dif); 1271 else 1272 protect = 0; 1273 1274 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) { 1275 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks, 1276 protect | fua, dld); 1277 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) { 1278 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks, 1279 protect | fua, dld); 1280 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) || 1281 sdp->use_10_for_rw || protect || rq->write_hint) { 1282 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks, 1283 protect | fua); 1284 } else { 1285 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks, 1286 protect | fua); 1287 } 1288 1289 if (unlikely(ret != BLK_STS_OK)) 1290 goto fail; 1291 1292 /* 1293 * We shouldn't disconnect in the middle of a sector, so with a dumb 1294 * host adapter, it's safe to assume that we can at least transfer 1295 * this many bytes between each connect / disconnect. 1296 */ 1297 cmd->transfersize = sdp->sector_size; 1298 cmd->underflow = nr_blocks << 9; 1299 cmd->allowed = sdkp->max_retries; 1300 cmd->sdb.length = nr_blocks * sdp->sector_size; 1301 1302 SCSI_LOG_HLQUEUE(1, 1303 scmd_printk(KERN_INFO, cmd, 1304 "%s: block=%llu, count=%d\n", __func__, 1305 (unsigned long long)blk_rq_pos(rq), 1306 blk_rq_sectors(rq))); 1307 SCSI_LOG_HLQUEUE(2, 1308 scmd_printk(KERN_INFO, cmd, 1309 "%s %d/%u 512 byte blocks.\n", 1310 write ? "writing" : "reading", nr_blocks, 1311 blk_rq_sectors(rq))); 1312 1313 /* 1314 * This indicates that the command is ready from our end to be queued. 1315 */ 1316 return BLK_STS_OK; 1317 fail: 1318 scsi_free_sgtables(cmd); 1319 return ret; 1320 } 1321 1322 static blk_status_t sd_init_command(struct scsi_cmnd *cmd) 1323 { 1324 struct request *rq = scsi_cmd_to_rq(cmd); 1325 1326 switch (req_op(rq)) { 1327 case REQ_OP_DISCARD: 1328 switch (scsi_disk(rq->q->disk)->provisioning_mode) { 1329 case SD_LBP_UNMAP: 1330 return sd_setup_unmap_cmnd(cmd); 1331 case SD_LBP_WS16: 1332 return sd_setup_write_same16_cmnd(cmd, true); 1333 case SD_LBP_WS10: 1334 return sd_setup_write_same10_cmnd(cmd, true); 1335 case SD_LBP_ZERO: 1336 return sd_setup_write_same10_cmnd(cmd, false); 1337 default: 1338 return BLK_STS_TARGET; 1339 } 1340 case REQ_OP_WRITE_ZEROES: 1341 return sd_setup_write_zeroes_cmnd(cmd); 1342 case REQ_OP_FLUSH: 1343 return sd_setup_flush_cmnd(cmd); 1344 case REQ_OP_READ: 1345 case REQ_OP_WRITE: 1346 return sd_setup_read_write_cmnd(cmd); 1347 case REQ_OP_ZONE_RESET: 1348 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER, 1349 false); 1350 case REQ_OP_ZONE_RESET_ALL: 1351 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER, 1352 true); 1353 case REQ_OP_ZONE_OPEN: 1354 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false); 1355 case REQ_OP_ZONE_CLOSE: 1356 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false); 1357 case REQ_OP_ZONE_FINISH: 1358 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false); 1359 default: 1360 WARN_ON_ONCE(1); 1361 return BLK_STS_NOTSUPP; 1362 } 1363 } 1364 1365 static void sd_uninit_command(struct scsi_cmnd *SCpnt) 1366 { 1367 struct request *rq = scsi_cmd_to_rq(SCpnt); 1368 1369 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1370 mempool_free(rq->special_vec.bv_page, sd_page_pool); 1371 } 1372 1373 static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp) 1374 { 1375 if (sdkp->device->removable || sdkp->write_prot) { 1376 if (disk_check_media_change(disk)) 1377 return true; 1378 } 1379 1380 /* 1381 * Force a full rescan after ioctl(BLKRRPART). While the disk state has 1382 * nothing to do with partitions, BLKRRPART is used to force a full 1383 * revalidate after things like a format for historical reasons. 1384 */ 1385 return test_bit(GD_NEED_PART_SCAN, &disk->state); 1386 } 1387 1388 /** 1389 * sd_open - open a scsi disk device 1390 * @disk: disk to open 1391 * @mode: open mode 1392 * 1393 * Returns 0 if successful. Returns a negated errno value in case 1394 * of error. 1395 * 1396 * Note: This can be called from a user context (e.g. fsck(1) ) 1397 * or from within the kernel (e.g. as a result of a mount(1) ). 1398 * In the latter case @inode and @filp carry an abridged amount 1399 * of information as noted above. 1400 * 1401 * Locking: called with disk->open_mutex held. 1402 **/ 1403 static int sd_open(struct gendisk *disk, blk_mode_t mode) 1404 { 1405 struct scsi_disk *sdkp = scsi_disk(disk); 1406 struct scsi_device *sdev = sdkp->device; 1407 int retval; 1408 1409 if (scsi_device_get(sdev)) 1410 return -ENXIO; 1411 1412 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n")); 1413 1414 /* 1415 * If the device is in error recovery, wait until it is done. 1416 * If the device is offline, then disallow any access to it. 1417 */ 1418 retval = -ENXIO; 1419 if (!scsi_block_when_processing_errors(sdev)) 1420 goto error_out; 1421 1422 if (sd_need_revalidate(disk, sdkp)) 1423 sd_revalidate_disk(disk); 1424 1425 /* 1426 * If the drive is empty, just let the open fail. 1427 */ 1428 retval = -ENOMEDIUM; 1429 if (sdev->removable && !sdkp->media_present && 1430 !(mode & BLK_OPEN_NDELAY)) 1431 goto error_out; 1432 1433 /* 1434 * If the device has the write protect tab set, have the open fail 1435 * if the user expects to be able to write to the thing. 1436 */ 1437 retval = -EROFS; 1438 if (sdkp->write_prot && (mode & BLK_OPEN_WRITE)) 1439 goto error_out; 1440 1441 /* 1442 * It is possible that the disk changing stuff resulted in 1443 * the device being taken offline. If this is the case, 1444 * report this to the user, and don't pretend that the 1445 * open actually succeeded. 1446 */ 1447 retval = -ENXIO; 1448 if (!scsi_device_online(sdev)) 1449 goto error_out; 1450 1451 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) { 1452 if (scsi_block_when_processing_errors(sdev)) 1453 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT); 1454 } 1455 1456 return 0; 1457 1458 error_out: 1459 scsi_device_put(sdev); 1460 return retval; 1461 } 1462 1463 /** 1464 * sd_release - invoked when the (last) close(2) is called on this 1465 * scsi disk. 1466 * @disk: disk to release 1467 * 1468 * Returns 0. 1469 * 1470 * Note: may block (uninterruptible) if error recovery is underway 1471 * on this disk. 1472 * 1473 * Locking: called with disk->open_mutex held. 1474 **/ 1475 static void sd_release(struct gendisk *disk) 1476 { 1477 struct scsi_disk *sdkp = scsi_disk(disk); 1478 struct scsi_device *sdev = sdkp->device; 1479 1480 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n")); 1481 1482 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) { 1483 if (scsi_block_when_processing_errors(sdev)) 1484 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW); 1485 } 1486 1487 scsi_device_put(sdev); 1488 } 1489 1490 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo) 1491 { 1492 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk); 1493 struct scsi_device *sdp = sdkp->device; 1494 struct Scsi_Host *host = sdp->host; 1495 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity); 1496 int diskinfo[4]; 1497 1498 /* default to most commonly used values */ 1499 diskinfo[0] = 0x40; /* 1 << 6 */ 1500 diskinfo[1] = 0x20; /* 1 << 5 */ 1501 diskinfo[2] = capacity >> 11; 1502 1503 /* override with calculated, extended default, or driver values */ 1504 if (host->hostt->bios_param) 1505 host->hostt->bios_param(sdp, bdev, capacity, diskinfo); 1506 else 1507 scsicam_bios_param(bdev, capacity, diskinfo); 1508 1509 geo->heads = diskinfo[0]; 1510 geo->sectors = diskinfo[1]; 1511 geo->cylinders = diskinfo[2]; 1512 return 0; 1513 } 1514 1515 /** 1516 * sd_ioctl - process an ioctl 1517 * @bdev: target block device 1518 * @mode: open mode 1519 * @cmd: ioctl command number 1520 * @arg: this is third argument given to ioctl(2) system call. 1521 * Often contains a pointer. 1522 * 1523 * Returns 0 if successful (some ioctls return positive numbers on 1524 * success as well). Returns a negated errno value in case of error. 1525 * 1526 * Note: most ioctls are forward onto the block subsystem or further 1527 * down in the scsi subsystem. 1528 **/ 1529 static int sd_ioctl(struct block_device *bdev, blk_mode_t mode, 1530 unsigned int cmd, unsigned long arg) 1531 { 1532 struct gendisk *disk = bdev->bd_disk; 1533 struct scsi_disk *sdkp = scsi_disk(disk); 1534 struct scsi_device *sdp = sdkp->device; 1535 void __user *p = (void __user *)arg; 1536 int error; 1537 1538 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, " 1539 "cmd=0x%x\n", disk->disk_name, cmd)); 1540 1541 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO)) 1542 return -ENOIOCTLCMD; 1543 1544 /* 1545 * If we are in the middle of error recovery, don't let anyone 1546 * else try and use this device. Also, if error recovery fails, it 1547 * may try and take the device offline, in which case all further 1548 * access to the device is prohibited. 1549 */ 1550 error = scsi_ioctl_block_when_processing_errors(sdp, cmd, 1551 (mode & BLK_OPEN_NDELAY)); 1552 if (error) 1553 return error; 1554 1555 if (is_sed_ioctl(cmd)) 1556 return sed_ioctl(sdkp->opal_dev, cmd, p); 1557 return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p); 1558 } 1559 1560 static void set_media_not_present(struct scsi_disk *sdkp) 1561 { 1562 if (sdkp->media_present) 1563 sdkp->device->changed = 1; 1564 1565 if (sdkp->device->removable) { 1566 sdkp->media_present = 0; 1567 sdkp->capacity = 0; 1568 } 1569 } 1570 1571 static int media_not_present(struct scsi_disk *sdkp, 1572 struct scsi_sense_hdr *sshdr) 1573 { 1574 if (!scsi_sense_valid(sshdr)) 1575 return 0; 1576 1577 /* not invoked for commands that could return deferred errors */ 1578 switch (sshdr->sense_key) { 1579 case UNIT_ATTENTION: 1580 case NOT_READY: 1581 /* medium not present */ 1582 if (sshdr->asc == 0x3A) { 1583 set_media_not_present(sdkp); 1584 return 1; 1585 } 1586 } 1587 return 0; 1588 } 1589 1590 /** 1591 * sd_check_events - check media events 1592 * @disk: kernel device descriptor 1593 * @clearing: disk events currently being cleared 1594 * 1595 * Returns mask of DISK_EVENT_*. 1596 * 1597 * Note: this function is invoked from the block subsystem. 1598 **/ 1599 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing) 1600 { 1601 struct scsi_disk *sdkp = disk->private_data; 1602 struct scsi_device *sdp; 1603 int retval; 1604 bool disk_changed; 1605 1606 if (!sdkp) 1607 return 0; 1608 1609 sdp = sdkp->device; 1610 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n")); 1611 1612 /* 1613 * If the device is offline, don't send any commands - just pretend as 1614 * if the command failed. If the device ever comes back online, we 1615 * can deal with it then. It is only because of unrecoverable errors 1616 * that we would ever take a device offline in the first place. 1617 */ 1618 if (!scsi_device_online(sdp)) { 1619 set_media_not_present(sdkp); 1620 goto out; 1621 } 1622 1623 /* 1624 * Using TEST_UNIT_READY enables differentiation between drive with 1625 * no cartridge loaded - NOT READY, drive with changed cartridge - 1626 * UNIT ATTENTION, or with same cartridge - GOOD STATUS. 1627 * 1628 * Drives that auto spin down. eg iomega jaz 1G, will be started 1629 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever 1630 * sd_revalidate() is called. 1631 */ 1632 if (scsi_block_when_processing_errors(sdp)) { 1633 struct scsi_sense_hdr sshdr = { 0, }; 1634 1635 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries, 1636 &sshdr); 1637 1638 /* failed to execute TUR, assume media not present */ 1639 if (retval < 0 || host_byte(retval)) { 1640 set_media_not_present(sdkp); 1641 goto out; 1642 } 1643 1644 if (media_not_present(sdkp, &sshdr)) 1645 goto out; 1646 } 1647 1648 /* 1649 * For removable scsi disk we have to recognise the presence 1650 * of a disk in the drive. 1651 */ 1652 if (!sdkp->media_present) 1653 sdp->changed = 1; 1654 sdkp->media_present = 1; 1655 out: 1656 /* 1657 * sdp->changed is set under the following conditions: 1658 * 1659 * Medium present state has changed in either direction. 1660 * Device has indicated UNIT_ATTENTION. 1661 */ 1662 disk_changed = sdp->changed; 1663 sdp->changed = 0; 1664 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0; 1665 } 1666 1667 static int sd_sync_cache(struct scsi_disk *sdkp) 1668 { 1669 int res; 1670 struct scsi_device *sdp = sdkp->device; 1671 const int timeout = sdp->request_queue->rq_timeout 1672 * SD_FLUSH_TIMEOUT_MULTIPLIER; 1673 /* Leave the rest of the command zero to indicate flush everything. */ 1674 const unsigned char cmd[16] = { sdp->use_16_for_sync ? 1675 SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE }; 1676 struct scsi_sense_hdr sshdr; 1677 struct scsi_failure failure_defs[] = { 1678 { 1679 .allowed = 3, 1680 .result = SCMD_FAILURE_RESULT_ANY, 1681 }, 1682 {} 1683 }; 1684 struct scsi_failures failures = { 1685 .failure_definitions = failure_defs, 1686 }; 1687 const struct scsi_exec_args exec_args = { 1688 .req_flags = BLK_MQ_REQ_PM, 1689 .sshdr = &sshdr, 1690 .failures = &failures, 1691 }; 1692 1693 if (!scsi_device_online(sdp)) 1694 return -ENODEV; 1695 1696 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout, 1697 sdkp->max_retries, &exec_args); 1698 if (res) { 1699 sd_print_result(sdkp, "Synchronize Cache(10) failed", res); 1700 1701 if (res < 0) 1702 return res; 1703 1704 if (scsi_status_is_check_condition(res) && 1705 scsi_sense_valid(&sshdr)) { 1706 sd_print_sense_hdr(sdkp, &sshdr); 1707 1708 /* we need to evaluate the error return */ 1709 if (sshdr.asc == 0x3a || /* medium not present */ 1710 sshdr.asc == 0x20 || /* invalid command */ 1711 (sshdr.asc == 0x74 && sshdr.ascq == 0x71)) /* drive is password locked */ 1712 /* this is no error here */ 1713 return 0; 1714 /* 1715 * This drive doesn't support sync and there's not much 1716 * we can do because this is called during shutdown 1717 * or suspend so just return success so those operations 1718 * can proceed. 1719 */ 1720 if (sshdr.sense_key == ILLEGAL_REQUEST) 1721 return 0; 1722 } 1723 1724 switch (host_byte(res)) { 1725 /* ignore errors due to racing a disconnection */ 1726 case DID_BAD_TARGET: 1727 case DID_NO_CONNECT: 1728 return 0; 1729 /* signal the upper layer it might try again */ 1730 case DID_BUS_BUSY: 1731 case DID_IMM_RETRY: 1732 case DID_REQUEUE: 1733 case DID_SOFT_ERROR: 1734 return -EBUSY; 1735 default: 1736 return -EIO; 1737 } 1738 } 1739 return 0; 1740 } 1741 1742 static void sd_rescan(struct device *dev) 1743 { 1744 struct scsi_disk *sdkp = dev_get_drvdata(dev); 1745 1746 sd_revalidate_disk(sdkp->disk); 1747 } 1748 1749 static int sd_get_unique_id(struct gendisk *disk, u8 id[16], 1750 enum blk_unique_id type) 1751 { 1752 struct scsi_device *sdev = scsi_disk(disk)->device; 1753 const struct scsi_vpd *vpd; 1754 const unsigned char *d; 1755 int ret = -ENXIO, len; 1756 1757 rcu_read_lock(); 1758 vpd = rcu_dereference(sdev->vpd_pg83); 1759 if (!vpd) 1760 goto out_unlock; 1761 1762 ret = -EINVAL; 1763 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) { 1764 /* we only care about designators with LU association */ 1765 if (((d[1] >> 4) & 0x3) != 0x00) 1766 continue; 1767 if ((d[1] & 0xf) != type) 1768 continue; 1769 1770 /* 1771 * Only exit early if a 16-byte descriptor was found. Otherwise 1772 * keep looking as one with more entropy might still show up. 1773 */ 1774 len = d[3]; 1775 if (len != 8 && len != 12 && len != 16) 1776 continue; 1777 ret = len; 1778 memcpy(id, d + 4, len); 1779 if (len == 16) 1780 break; 1781 } 1782 out_unlock: 1783 rcu_read_unlock(); 1784 return ret; 1785 } 1786 1787 static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result) 1788 { 1789 switch (host_byte(result)) { 1790 case DID_TRANSPORT_MARGINAL: 1791 case DID_TRANSPORT_DISRUPTED: 1792 case DID_BUS_BUSY: 1793 return PR_STS_RETRY_PATH_FAILURE; 1794 case DID_NO_CONNECT: 1795 return PR_STS_PATH_FAILED; 1796 case DID_TRANSPORT_FAILFAST: 1797 return PR_STS_PATH_FAST_FAILED; 1798 } 1799 1800 switch (status_byte(result)) { 1801 case SAM_STAT_RESERVATION_CONFLICT: 1802 return PR_STS_RESERVATION_CONFLICT; 1803 case SAM_STAT_CHECK_CONDITION: 1804 if (!scsi_sense_valid(sshdr)) 1805 return PR_STS_IOERR; 1806 1807 if (sshdr->sense_key == ILLEGAL_REQUEST && 1808 (sshdr->asc == 0x26 || sshdr->asc == 0x24)) 1809 return -EINVAL; 1810 1811 fallthrough; 1812 default: 1813 return PR_STS_IOERR; 1814 } 1815 } 1816 1817 static int sd_pr_in_command(struct block_device *bdev, u8 sa, 1818 unsigned char *data, int data_len) 1819 { 1820 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk); 1821 struct scsi_device *sdev = sdkp->device; 1822 struct scsi_sense_hdr sshdr; 1823 u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa }; 1824 struct scsi_failure failure_defs[] = { 1825 { 1826 .sense = UNIT_ATTENTION, 1827 .asc = SCMD_FAILURE_ASC_ANY, 1828 .ascq = SCMD_FAILURE_ASCQ_ANY, 1829 .allowed = 5, 1830 .result = SAM_STAT_CHECK_CONDITION, 1831 }, 1832 {} 1833 }; 1834 struct scsi_failures failures = { 1835 .failure_definitions = failure_defs, 1836 }; 1837 const struct scsi_exec_args exec_args = { 1838 .sshdr = &sshdr, 1839 .failures = &failures, 1840 }; 1841 int result; 1842 1843 put_unaligned_be16(data_len, &cmd[7]); 1844 1845 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len, 1846 SD_TIMEOUT, sdkp->max_retries, &exec_args); 1847 if (scsi_status_is_check_condition(result) && 1848 scsi_sense_valid(&sshdr)) { 1849 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result); 1850 scsi_print_sense_hdr(sdev, NULL, &sshdr); 1851 } 1852 1853 if (result <= 0) 1854 return result; 1855 1856 return sd_scsi_to_pr_err(&sshdr, result); 1857 } 1858 1859 static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info) 1860 { 1861 int result, i, data_offset, num_copy_keys; 1862 u32 num_keys = keys_info->num_keys; 1863 int data_len = num_keys * 8 + 8; 1864 u8 *data; 1865 1866 data = kzalloc(data_len, GFP_KERNEL); 1867 if (!data) 1868 return -ENOMEM; 1869 1870 result = sd_pr_in_command(bdev, READ_KEYS, data, data_len); 1871 if (result) 1872 goto free_data; 1873 1874 keys_info->generation = get_unaligned_be32(&data[0]); 1875 keys_info->num_keys = get_unaligned_be32(&data[4]) / 8; 1876 1877 data_offset = 8; 1878 num_copy_keys = min(num_keys, keys_info->num_keys); 1879 1880 for (i = 0; i < num_copy_keys; i++) { 1881 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]); 1882 data_offset += 8; 1883 } 1884 1885 free_data: 1886 kfree(data); 1887 return result; 1888 } 1889 1890 static int sd_pr_read_reservation(struct block_device *bdev, 1891 struct pr_held_reservation *rsv) 1892 { 1893 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk); 1894 struct scsi_device *sdev = sdkp->device; 1895 u8 data[24] = { }; 1896 int result, len; 1897 1898 result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data)); 1899 if (result) 1900 return result; 1901 1902 len = get_unaligned_be32(&data[4]); 1903 if (!len) 1904 return 0; 1905 1906 /* Make sure we have at least the key and type */ 1907 if (len < 14) { 1908 sdev_printk(KERN_INFO, sdev, 1909 "READ RESERVATION failed due to short return buffer of %d bytes\n", 1910 len); 1911 return -EINVAL; 1912 } 1913 1914 rsv->generation = get_unaligned_be32(&data[0]); 1915 rsv->key = get_unaligned_be64(&data[8]); 1916 rsv->type = scsi_pr_type_to_block(data[21] & 0x0f); 1917 return 0; 1918 } 1919 1920 static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key, 1921 u64 sa_key, enum scsi_pr_type type, u8 flags) 1922 { 1923 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk); 1924 struct scsi_device *sdev = sdkp->device; 1925 struct scsi_sense_hdr sshdr; 1926 struct scsi_failure failure_defs[] = { 1927 { 1928 .sense = UNIT_ATTENTION, 1929 .asc = SCMD_FAILURE_ASC_ANY, 1930 .ascq = SCMD_FAILURE_ASCQ_ANY, 1931 .allowed = 5, 1932 .result = SAM_STAT_CHECK_CONDITION, 1933 }, 1934 {} 1935 }; 1936 struct scsi_failures failures = { 1937 .failure_definitions = failure_defs, 1938 }; 1939 const struct scsi_exec_args exec_args = { 1940 .sshdr = &sshdr, 1941 .failures = &failures, 1942 }; 1943 int result; 1944 u8 cmd[16] = { 0, }; 1945 u8 data[24] = { 0, }; 1946 1947 cmd[0] = PERSISTENT_RESERVE_OUT; 1948 cmd[1] = sa; 1949 cmd[2] = type; 1950 put_unaligned_be32(sizeof(data), &cmd[5]); 1951 1952 put_unaligned_be64(key, &data[0]); 1953 put_unaligned_be64(sa_key, &data[8]); 1954 data[20] = flags; 1955 1956 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data, 1957 sizeof(data), SD_TIMEOUT, sdkp->max_retries, 1958 &exec_args); 1959 1960 if (scsi_status_is_check_condition(result) && 1961 scsi_sense_valid(&sshdr)) { 1962 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result); 1963 scsi_print_sense_hdr(sdev, NULL, &sshdr); 1964 } 1965 1966 if (result <= 0) 1967 return result; 1968 1969 return sd_scsi_to_pr_err(&sshdr, result); 1970 } 1971 1972 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, 1973 u32 flags) 1974 { 1975 if (flags & ~PR_FL_IGNORE_KEY) 1976 return -EOPNOTSUPP; 1977 return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00, 1978 old_key, new_key, 0, 1979 (1 << 0) /* APTPL */); 1980 } 1981 1982 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, 1983 u32 flags) 1984 { 1985 if (flags) 1986 return -EOPNOTSUPP; 1987 return sd_pr_out_command(bdev, 0x01, key, 0, 1988 block_pr_type_to_scsi(type), 0); 1989 } 1990 1991 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type) 1992 { 1993 return sd_pr_out_command(bdev, 0x02, key, 0, 1994 block_pr_type_to_scsi(type), 0); 1995 } 1996 1997 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, 1998 enum pr_type type, bool abort) 1999 { 2000 return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key, 2001 block_pr_type_to_scsi(type), 0); 2002 } 2003 2004 static int sd_pr_clear(struct block_device *bdev, u64 key) 2005 { 2006 return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0); 2007 } 2008 2009 static const struct pr_ops sd_pr_ops = { 2010 .pr_register = sd_pr_register, 2011 .pr_reserve = sd_pr_reserve, 2012 .pr_release = sd_pr_release, 2013 .pr_preempt = sd_pr_preempt, 2014 .pr_clear = sd_pr_clear, 2015 .pr_read_keys = sd_pr_read_keys, 2016 .pr_read_reservation = sd_pr_read_reservation, 2017 }; 2018 2019 static void scsi_disk_free_disk(struct gendisk *disk) 2020 { 2021 struct scsi_disk *sdkp = scsi_disk(disk); 2022 2023 put_device(&sdkp->disk_dev); 2024 } 2025 2026 static const struct block_device_operations sd_fops = { 2027 .owner = THIS_MODULE, 2028 .open = sd_open, 2029 .release = sd_release, 2030 .ioctl = sd_ioctl, 2031 .getgeo = sd_getgeo, 2032 .compat_ioctl = blkdev_compat_ptr_ioctl, 2033 .check_events = sd_check_events, 2034 .unlock_native_capacity = sd_unlock_native_capacity, 2035 .report_zones = sd_zbc_report_zones, 2036 .get_unique_id = sd_get_unique_id, 2037 .free_disk = scsi_disk_free_disk, 2038 .pr_ops = &sd_pr_ops, 2039 }; 2040 2041 /** 2042 * sd_eh_reset - reset error handling callback 2043 * @scmd: sd-issued command that has failed 2044 * 2045 * This function is called by the SCSI midlayer before starting 2046 * SCSI EH. When counting medium access failures we have to be 2047 * careful to register it only only once per device and SCSI EH run; 2048 * there might be several timed out commands which will cause the 2049 * 'max_medium_access_timeouts' counter to trigger after the first 2050 * SCSI EH run already and set the device to offline. 2051 * So this function resets the internal counter before starting SCSI EH. 2052 **/ 2053 static void sd_eh_reset(struct scsi_cmnd *scmd) 2054 { 2055 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk); 2056 2057 /* New SCSI EH run, reset gate variable */ 2058 sdkp->ignore_medium_access_errors = false; 2059 } 2060 2061 /** 2062 * sd_eh_action - error handling callback 2063 * @scmd: sd-issued command that has failed 2064 * @eh_disp: The recovery disposition suggested by the midlayer 2065 * 2066 * This function is called by the SCSI midlayer upon completion of an 2067 * error test command (currently TEST UNIT READY). The result of sending 2068 * the eh command is passed in eh_disp. We're looking for devices that 2069 * fail medium access commands but are OK with non access commands like 2070 * test unit ready (so wrongly see the device as having a successful 2071 * recovery) 2072 **/ 2073 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp) 2074 { 2075 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk); 2076 struct scsi_device *sdev = scmd->device; 2077 2078 if (!scsi_device_online(sdev) || 2079 !scsi_medium_access_command(scmd) || 2080 host_byte(scmd->result) != DID_TIME_OUT || 2081 eh_disp != SUCCESS) 2082 return eh_disp; 2083 2084 /* 2085 * The device has timed out executing a medium access command. 2086 * However, the TEST UNIT READY command sent during error 2087 * handling completed successfully. Either the device is in the 2088 * process of recovering or has it suffered an internal failure 2089 * that prevents access to the storage medium. 2090 */ 2091 if (!sdkp->ignore_medium_access_errors) { 2092 sdkp->medium_access_timed_out++; 2093 sdkp->ignore_medium_access_errors = true; 2094 } 2095 2096 /* 2097 * If the device keeps failing read/write commands but TEST UNIT 2098 * READY always completes successfully we assume that medium 2099 * access is no longer possible and take the device offline. 2100 */ 2101 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) { 2102 scmd_printk(KERN_ERR, scmd, 2103 "Medium access timeout failure. Offlining disk!\n"); 2104 mutex_lock(&sdev->state_mutex); 2105 scsi_device_set_state(sdev, SDEV_OFFLINE); 2106 mutex_unlock(&sdev->state_mutex); 2107 2108 return SUCCESS; 2109 } 2110 2111 return eh_disp; 2112 } 2113 2114 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd) 2115 { 2116 struct request *req = scsi_cmd_to_rq(scmd); 2117 struct scsi_device *sdev = scmd->device; 2118 unsigned int transferred, good_bytes; 2119 u64 start_lba, end_lba, bad_lba; 2120 2121 /* 2122 * Some commands have a payload smaller than the device logical 2123 * block size (e.g. INQUIRY on a 4K disk). 2124 */ 2125 if (scsi_bufflen(scmd) <= sdev->sector_size) 2126 return 0; 2127 2128 /* Check if we have a 'bad_lba' information */ 2129 if (!scsi_get_sense_info_fld(scmd->sense_buffer, 2130 SCSI_SENSE_BUFFERSIZE, 2131 &bad_lba)) 2132 return 0; 2133 2134 /* 2135 * If the bad lba was reported incorrectly, we have no idea where 2136 * the error is. 2137 */ 2138 start_lba = sectors_to_logical(sdev, blk_rq_pos(req)); 2139 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd)); 2140 if (bad_lba < start_lba || bad_lba >= end_lba) 2141 return 0; 2142 2143 /* 2144 * resid is optional but mostly filled in. When it's unused, 2145 * its value is zero, so we assume the whole buffer transferred 2146 */ 2147 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd); 2148 2149 /* This computation should always be done in terms of the 2150 * resolution of the device's medium. 2151 */ 2152 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba); 2153 2154 return min(good_bytes, transferred); 2155 } 2156 2157 /** 2158 * sd_done - bottom half handler: called when the lower level 2159 * driver has completed (successfully or otherwise) a scsi command. 2160 * @SCpnt: mid-level's per command structure. 2161 * 2162 * Note: potentially run from within an ISR. Must not block. 2163 **/ 2164 static int sd_done(struct scsi_cmnd *SCpnt) 2165 { 2166 int result = SCpnt->result; 2167 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt); 2168 unsigned int sector_size = SCpnt->device->sector_size; 2169 unsigned int resid; 2170 struct scsi_sense_hdr sshdr; 2171 struct request *req = scsi_cmd_to_rq(SCpnt); 2172 struct scsi_disk *sdkp = scsi_disk(req->q->disk); 2173 int sense_valid = 0; 2174 int sense_deferred = 0; 2175 2176 switch (req_op(req)) { 2177 case REQ_OP_DISCARD: 2178 case REQ_OP_WRITE_ZEROES: 2179 case REQ_OP_ZONE_RESET: 2180 case REQ_OP_ZONE_RESET_ALL: 2181 case REQ_OP_ZONE_OPEN: 2182 case REQ_OP_ZONE_CLOSE: 2183 case REQ_OP_ZONE_FINISH: 2184 if (!result) { 2185 good_bytes = blk_rq_bytes(req); 2186 scsi_set_resid(SCpnt, 0); 2187 } else { 2188 good_bytes = 0; 2189 scsi_set_resid(SCpnt, blk_rq_bytes(req)); 2190 } 2191 break; 2192 default: 2193 /* 2194 * In case of bogus fw or device, we could end up having 2195 * an unaligned partial completion. Check this here and force 2196 * alignment. 2197 */ 2198 resid = scsi_get_resid(SCpnt); 2199 if (resid & (sector_size - 1)) { 2200 sd_printk(KERN_INFO, sdkp, 2201 "Unaligned partial completion (resid=%u, sector_sz=%u)\n", 2202 resid, sector_size); 2203 scsi_print_command(SCpnt); 2204 resid = min(scsi_bufflen(SCpnt), 2205 round_up(resid, sector_size)); 2206 scsi_set_resid(SCpnt, resid); 2207 } 2208 } 2209 2210 if (result) { 2211 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr); 2212 if (sense_valid) 2213 sense_deferred = scsi_sense_is_deferred(&sshdr); 2214 } 2215 sdkp->medium_access_timed_out = 0; 2216 2217 if (!scsi_status_is_check_condition(result) && 2218 (!sense_valid || sense_deferred)) 2219 goto out; 2220 2221 switch (sshdr.sense_key) { 2222 case HARDWARE_ERROR: 2223 case MEDIUM_ERROR: 2224 good_bytes = sd_completed_bytes(SCpnt); 2225 break; 2226 case RECOVERED_ERROR: 2227 good_bytes = scsi_bufflen(SCpnt); 2228 break; 2229 case NO_SENSE: 2230 /* This indicates a false check condition, so ignore it. An 2231 * unknown amount of data was transferred so treat it as an 2232 * error. 2233 */ 2234 SCpnt->result = 0; 2235 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 2236 break; 2237 case ABORTED_COMMAND: 2238 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */ 2239 good_bytes = sd_completed_bytes(SCpnt); 2240 break; 2241 case ILLEGAL_REQUEST: 2242 switch (sshdr.asc) { 2243 case 0x10: /* DIX: Host detected corruption */ 2244 good_bytes = sd_completed_bytes(SCpnt); 2245 break; 2246 case 0x20: /* INVALID COMMAND OPCODE */ 2247 case 0x24: /* INVALID FIELD IN CDB */ 2248 switch (SCpnt->cmnd[0]) { 2249 case UNMAP: 2250 sd_config_discard(sdkp, SD_LBP_DISABLE); 2251 break; 2252 case WRITE_SAME_16: 2253 case WRITE_SAME: 2254 if (SCpnt->cmnd[1] & 8) { /* UNMAP */ 2255 sd_config_discard(sdkp, SD_LBP_DISABLE); 2256 } else { 2257 sdkp->device->no_write_same = 1; 2258 sd_config_write_same(sdkp); 2259 req->rq_flags |= RQF_QUIET; 2260 } 2261 break; 2262 } 2263 } 2264 break; 2265 default: 2266 break; 2267 } 2268 2269 out: 2270 if (sd_is_zoned(sdkp)) 2271 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr); 2272 2273 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt, 2274 "sd_done: completed %d of %d bytes\n", 2275 good_bytes, scsi_bufflen(SCpnt))); 2276 2277 return good_bytes; 2278 } 2279 2280 /* 2281 * spinup disk - called only in sd_revalidate_disk() 2282 */ 2283 static void 2284 sd_spinup_disk(struct scsi_disk *sdkp) 2285 { 2286 static const u8 cmd[10] = { TEST_UNIT_READY }; 2287 unsigned long spintime_expire = 0; 2288 int spintime, sense_valid = 0; 2289 unsigned int the_result; 2290 struct scsi_sense_hdr sshdr; 2291 struct scsi_failure failure_defs[] = { 2292 /* Do not retry Medium Not Present */ 2293 { 2294 .sense = UNIT_ATTENTION, 2295 .asc = 0x3A, 2296 .ascq = SCMD_FAILURE_ASCQ_ANY, 2297 .result = SAM_STAT_CHECK_CONDITION, 2298 }, 2299 { 2300 .sense = NOT_READY, 2301 .asc = 0x3A, 2302 .ascq = SCMD_FAILURE_ASCQ_ANY, 2303 .result = SAM_STAT_CHECK_CONDITION, 2304 }, 2305 /* Retry when scsi_status_is_good would return false 3 times */ 2306 { 2307 .result = SCMD_FAILURE_STAT_ANY, 2308 .allowed = 3, 2309 }, 2310 {} 2311 }; 2312 struct scsi_failures failures = { 2313 .failure_definitions = failure_defs, 2314 }; 2315 const struct scsi_exec_args exec_args = { 2316 .sshdr = &sshdr, 2317 .failures = &failures, 2318 }; 2319 2320 spintime = 0; 2321 2322 /* Spin up drives, as required. Only do this at boot time */ 2323 /* Spinup needs to be done for module loads too. */ 2324 do { 2325 bool media_was_present = sdkp->media_present; 2326 2327 scsi_failures_reset_retries(&failures); 2328 2329 the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, 2330 NULL, 0, SD_TIMEOUT, 2331 sdkp->max_retries, &exec_args); 2332 2333 2334 if (the_result > 0) { 2335 /* 2336 * If the drive has indicated to us that it doesn't 2337 * have any media in it, don't bother with any more 2338 * polling. 2339 */ 2340 if (media_not_present(sdkp, &sshdr)) { 2341 if (media_was_present) 2342 sd_printk(KERN_NOTICE, sdkp, 2343 "Media removed, stopped polling\n"); 2344 return; 2345 } 2346 sense_valid = scsi_sense_valid(&sshdr); 2347 } 2348 2349 if (!scsi_status_is_check_condition(the_result)) { 2350 /* no sense, TUR either succeeded or failed 2351 * with a status error */ 2352 if(!spintime && !scsi_status_is_good(the_result)) { 2353 sd_print_result(sdkp, "Test Unit Ready failed", 2354 the_result); 2355 } 2356 break; 2357 } 2358 2359 /* 2360 * The device does not want the automatic start to be issued. 2361 */ 2362 if (sdkp->device->no_start_on_add) 2363 break; 2364 2365 if (sense_valid && sshdr.sense_key == NOT_READY) { 2366 if (sshdr.asc == 4 && sshdr.ascq == 3) 2367 break; /* manual intervention required */ 2368 if (sshdr.asc == 4 && sshdr.ascq == 0xb) 2369 break; /* standby */ 2370 if (sshdr.asc == 4 && sshdr.ascq == 0xc) 2371 break; /* unavailable */ 2372 if (sshdr.asc == 4 && sshdr.ascq == 0x1b) 2373 break; /* sanitize in progress */ 2374 if (sshdr.asc == 4 && sshdr.ascq == 0x24) 2375 break; /* depopulation in progress */ 2376 if (sshdr.asc == 4 && sshdr.ascq == 0x25) 2377 break; /* depopulation restoration in progress */ 2378 /* 2379 * Issue command to spin up drive when not ready 2380 */ 2381 if (!spintime) { 2382 /* Return immediately and start spin cycle */ 2383 const u8 start_cmd[10] = { 2384 [0] = START_STOP, 2385 [1] = 1, 2386 [4] = sdkp->device->start_stop_pwr_cond ? 2387 0x11 : 1, 2388 }; 2389 2390 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk..."); 2391 scsi_execute_cmd(sdkp->device, start_cmd, 2392 REQ_OP_DRV_IN, NULL, 0, 2393 SD_TIMEOUT, sdkp->max_retries, 2394 &exec_args); 2395 spintime_expire = jiffies + 100 * HZ; 2396 spintime = 1; 2397 } 2398 /* Wait 1 second for next try */ 2399 msleep(1000); 2400 printk(KERN_CONT "."); 2401 2402 /* 2403 * Wait for USB flash devices with slow firmware. 2404 * Yes, this sense key/ASC combination shouldn't 2405 * occur here. It's characteristic of these devices. 2406 */ 2407 } else if (sense_valid && 2408 sshdr.sense_key == UNIT_ATTENTION && 2409 sshdr.asc == 0x28) { 2410 if (!spintime) { 2411 spintime_expire = jiffies + 5 * HZ; 2412 spintime = 1; 2413 } 2414 /* Wait 1 second for next try */ 2415 msleep(1000); 2416 } else { 2417 /* we don't understand the sense code, so it's 2418 * probably pointless to loop */ 2419 if(!spintime) { 2420 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n"); 2421 sd_print_sense_hdr(sdkp, &sshdr); 2422 } 2423 break; 2424 } 2425 2426 } while (spintime && time_before_eq(jiffies, spintime_expire)); 2427 2428 if (spintime) { 2429 if (scsi_status_is_good(the_result)) 2430 printk(KERN_CONT "ready\n"); 2431 else 2432 printk(KERN_CONT "not responding...\n"); 2433 } 2434 } 2435 2436 /* 2437 * Determine whether disk supports Data Integrity Field. 2438 */ 2439 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer) 2440 { 2441 struct scsi_device *sdp = sdkp->device; 2442 u8 type; 2443 2444 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) { 2445 sdkp->protection_type = 0; 2446 return 0; 2447 } 2448 2449 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */ 2450 2451 if (type > T10_PI_TYPE3_PROTECTION) { 2452 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \ 2453 " protection type %u. Disabling disk!\n", 2454 type); 2455 sdkp->protection_type = 0; 2456 return -ENODEV; 2457 } 2458 2459 sdkp->protection_type = type; 2460 2461 return 0; 2462 } 2463 2464 static void sd_config_protection(struct scsi_disk *sdkp) 2465 { 2466 struct scsi_device *sdp = sdkp->device; 2467 2468 sd_dif_config_host(sdkp); 2469 2470 if (!sdkp->protection_type) 2471 return; 2472 2473 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) { 2474 sd_first_printk(KERN_NOTICE, sdkp, 2475 "Disabling DIF Type %u protection\n", 2476 sdkp->protection_type); 2477 sdkp->protection_type = 0; 2478 } 2479 2480 sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n", 2481 sdkp->protection_type); 2482 } 2483 2484 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp, 2485 struct scsi_sense_hdr *sshdr, int sense_valid, 2486 int the_result) 2487 { 2488 if (sense_valid) 2489 sd_print_sense_hdr(sdkp, sshdr); 2490 else 2491 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n"); 2492 2493 /* 2494 * Set dirty bit for removable devices if not ready - 2495 * sometimes drives will not report this properly. 2496 */ 2497 if (sdp->removable && 2498 sense_valid && sshdr->sense_key == NOT_READY) 2499 set_media_not_present(sdkp); 2500 2501 /* 2502 * We used to set media_present to 0 here to indicate no media 2503 * in the drive, but some drives fail read capacity even with 2504 * media present, so we can't do that. 2505 */ 2506 sdkp->capacity = 0; /* unknown mapped to zero - as usual */ 2507 } 2508 2509 #define RC16_LEN 32 2510 #if RC16_LEN > SD_BUF_SIZE 2511 #error RC16_LEN must not be more than SD_BUF_SIZE 2512 #endif 2513 2514 #define READ_CAPACITY_RETRIES_ON_RESET 10 2515 2516 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp, 2517 unsigned char *buffer) 2518 { 2519 unsigned char cmd[16]; 2520 struct scsi_sense_hdr sshdr; 2521 const struct scsi_exec_args exec_args = { 2522 .sshdr = &sshdr, 2523 }; 2524 int sense_valid = 0; 2525 int the_result; 2526 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET; 2527 unsigned int alignment; 2528 unsigned long long lba; 2529 unsigned sector_size; 2530 2531 if (sdp->no_read_capacity_16) 2532 return -EINVAL; 2533 2534 do { 2535 memset(cmd, 0, 16); 2536 cmd[0] = SERVICE_ACTION_IN_16; 2537 cmd[1] = SAI_READ_CAPACITY_16; 2538 cmd[13] = RC16_LEN; 2539 memset(buffer, 0, RC16_LEN); 2540 2541 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, 2542 buffer, RC16_LEN, SD_TIMEOUT, 2543 sdkp->max_retries, &exec_args); 2544 if (the_result > 0) { 2545 if (media_not_present(sdkp, &sshdr)) 2546 return -ENODEV; 2547 2548 sense_valid = scsi_sense_valid(&sshdr); 2549 if (sense_valid && 2550 sshdr.sense_key == ILLEGAL_REQUEST && 2551 (sshdr.asc == 0x20 || sshdr.asc == 0x24) && 2552 sshdr.ascq == 0x00) 2553 /* Invalid Command Operation Code or 2554 * Invalid Field in CDB, just retry 2555 * silently with RC10 */ 2556 return -EINVAL; 2557 if (sense_valid && 2558 sshdr.sense_key == UNIT_ATTENTION && 2559 sshdr.asc == 0x29 && sshdr.ascq == 0x00) 2560 /* Device reset might occur several times, 2561 * give it one more chance */ 2562 if (--reset_retries > 0) 2563 continue; 2564 } 2565 retries--; 2566 2567 } while (the_result && retries); 2568 2569 if (the_result) { 2570 sd_print_result(sdkp, "Read Capacity(16) failed", the_result); 2571 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result); 2572 return -EINVAL; 2573 } 2574 2575 sector_size = get_unaligned_be32(&buffer[8]); 2576 lba = get_unaligned_be64(&buffer[0]); 2577 2578 if (sd_read_protection_type(sdkp, buffer) < 0) { 2579 sdkp->capacity = 0; 2580 return -ENODEV; 2581 } 2582 2583 /* Logical blocks per physical block exponent */ 2584 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size; 2585 2586 /* RC basis */ 2587 sdkp->rc_basis = (buffer[12] >> 4) & 0x3; 2588 2589 /* Lowest aligned logical block */ 2590 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size; 2591 blk_queue_alignment_offset(sdp->request_queue, alignment); 2592 if (alignment && sdkp->first_scan) 2593 sd_printk(KERN_NOTICE, sdkp, 2594 "physical block alignment offset: %u\n", alignment); 2595 2596 if (buffer[14] & 0x80) { /* LBPME */ 2597 sdkp->lbpme = 1; 2598 2599 if (buffer[14] & 0x40) /* LBPRZ */ 2600 sdkp->lbprz = 1; 2601 2602 sd_config_discard(sdkp, SD_LBP_WS16); 2603 } 2604 2605 sdkp->capacity = lba + 1; 2606 return sector_size; 2607 } 2608 2609 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp, 2610 unsigned char *buffer) 2611 { 2612 static const u8 cmd[10] = { READ_CAPACITY }; 2613 struct scsi_sense_hdr sshdr; 2614 struct scsi_failure failure_defs[] = { 2615 /* Do not retry Medium Not Present */ 2616 { 2617 .sense = UNIT_ATTENTION, 2618 .asc = 0x3A, 2619 .result = SAM_STAT_CHECK_CONDITION, 2620 }, 2621 { 2622 .sense = NOT_READY, 2623 .asc = 0x3A, 2624 .result = SAM_STAT_CHECK_CONDITION, 2625 }, 2626 /* Device reset might occur several times so retry a lot */ 2627 { 2628 .sense = UNIT_ATTENTION, 2629 .asc = 0x29, 2630 .allowed = READ_CAPACITY_RETRIES_ON_RESET, 2631 .result = SAM_STAT_CHECK_CONDITION, 2632 }, 2633 /* Any other error not listed above retry 3 times */ 2634 { 2635 .result = SCMD_FAILURE_RESULT_ANY, 2636 .allowed = 3, 2637 }, 2638 {} 2639 }; 2640 struct scsi_failures failures = { 2641 .failure_definitions = failure_defs, 2642 }; 2643 const struct scsi_exec_args exec_args = { 2644 .sshdr = &sshdr, 2645 .failures = &failures, 2646 }; 2647 int sense_valid = 0; 2648 int the_result; 2649 sector_t lba; 2650 unsigned sector_size; 2651 2652 memset(buffer, 0, 8); 2653 2654 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer, 2655 8, SD_TIMEOUT, sdkp->max_retries, 2656 &exec_args); 2657 2658 if (the_result > 0) { 2659 sense_valid = scsi_sense_valid(&sshdr); 2660 2661 if (media_not_present(sdkp, &sshdr)) 2662 return -ENODEV; 2663 } 2664 2665 if (the_result) { 2666 sd_print_result(sdkp, "Read Capacity(10) failed", the_result); 2667 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result); 2668 return -EINVAL; 2669 } 2670 2671 sector_size = get_unaligned_be32(&buffer[4]); 2672 lba = get_unaligned_be32(&buffer[0]); 2673 2674 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) { 2675 /* Some buggy (usb cardreader) devices return an lba of 2676 0xffffffff when the want to report a size of 0 (with 2677 which they really mean no media is present) */ 2678 sdkp->capacity = 0; 2679 sdkp->physical_block_size = sector_size; 2680 return sector_size; 2681 } 2682 2683 sdkp->capacity = lba + 1; 2684 sdkp->physical_block_size = sector_size; 2685 return sector_size; 2686 } 2687 2688 static int sd_try_rc16_first(struct scsi_device *sdp) 2689 { 2690 if (sdp->host->max_cmd_len < 16) 2691 return 0; 2692 if (sdp->try_rc_10_first) 2693 return 0; 2694 if (sdp->scsi_level > SCSI_SPC_2) 2695 return 1; 2696 if (scsi_device_protection(sdp)) 2697 return 1; 2698 return 0; 2699 } 2700 2701 /* 2702 * read disk capacity 2703 */ 2704 static void 2705 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer) 2706 { 2707 int sector_size; 2708 struct scsi_device *sdp = sdkp->device; 2709 2710 if (sd_try_rc16_first(sdp)) { 2711 sector_size = read_capacity_16(sdkp, sdp, buffer); 2712 if (sector_size == -EOVERFLOW) 2713 goto got_data; 2714 if (sector_size == -ENODEV) 2715 return; 2716 if (sector_size < 0) 2717 sector_size = read_capacity_10(sdkp, sdp, buffer); 2718 if (sector_size < 0) 2719 return; 2720 } else { 2721 sector_size = read_capacity_10(sdkp, sdp, buffer); 2722 if (sector_size == -EOVERFLOW) 2723 goto got_data; 2724 if (sector_size < 0) 2725 return; 2726 if ((sizeof(sdkp->capacity) > 4) && 2727 (sdkp->capacity > 0xffffffffULL)) { 2728 int old_sector_size = sector_size; 2729 sd_printk(KERN_NOTICE, sdkp, "Very big device. " 2730 "Trying to use READ CAPACITY(16).\n"); 2731 sector_size = read_capacity_16(sdkp, sdp, buffer); 2732 if (sector_size < 0) { 2733 sd_printk(KERN_NOTICE, sdkp, 2734 "Using 0xffffffff as device size\n"); 2735 sdkp->capacity = 1 + (sector_t) 0xffffffff; 2736 sector_size = old_sector_size; 2737 goto got_data; 2738 } 2739 /* Remember that READ CAPACITY(16) succeeded */ 2740 sdp->try_rc_10_first = 0; 2741 } 2742 } 2743 2744 /* Some devices are known to return the total number of blocks, 2745 * not the highest block number. Some devices have versions 2746 * which do this and others which do not. Some devices we might 2747 * suspect of doing this but we don't know for certain. 2748 * 2749 * If we know the reported capacity is wrong, decrement it. If 2750 * we can only guess, then assume the number of blocks is even 2751 * (usually true but not always) and err on the side of lowering 2752 * the capacity. 2753 */ 2754 if (sdp->fix_capacity || 2755 (sdp->guess_capacity && (sdkp->capacity & 0x01))) { 2756 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count " 2757 "from its reported value: %llu\n", 2758 (unsigned long long) sdkp->capacity); 2759 --sdkp->capacity; 2760 } 2761 2762 got_data: 2763 if (sector_size == 0) { 2764 sector_size = 512; 2765 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, " 2766 "assuming 512.\n"); 2767 } 2768 2769 if (sector_size != 512 && 2770 sector_size != 1024 && 2771 sector_size != 2048 && 2772 sector_size != 4096) { 2773 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n", 2774 sector_size); 2775 /* 2776 * The user might want to re-format the drive with 2777 * a supported sectorsize. Once this happens, it 2778 * would be relatively trivial to set the thing up. 2779 * For this reason, we leave the thing in the table. 2780 */ 2781 sdkp->capacity = 0; 2782 /* 2783 * set a bogus sector size so the normal read/write 2784 * logic in the block layer will eventually refuse any 2785 * request on this device without tripping over power 2786 * of two sector size assumptions 2787 */ 2788 sector_size = 512; 2789 } 2790 blk_queue_logical_block_size(sdp->request_queue, sector_size); 2791 blk_queue_physical_block_size(sdp->request_queue, 2792 sdkp->physical_block_size); 2793 sdkp->device->sector_size = sector_size; 2794 2795 if (sdkp->capacity > 0xffffffff) 2796 sdp->use_16_for_rw = 1; 2797 2798 } 2799 2800 /* 2801 * Print disk capacity 2802 */ 2803 static void 2804 sd_print_capacity(struct scsi_disk *sdkp, 2805 sector_t old_capacity) 2806 { 2807 int sector_size = sdkp->device->sector_size; 2808 char cap_str_2[10], cap_str_10[10]; 2809 2810 if (!sdkp->first_scan && old_capacity == sdkp->capacity) 2811 return; 2812 2813 string_get_size(sdkp->capacity, sector_size, 2814 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); 2815 string_get_size(sdkp->capacity, sector_size, 2816 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); 2817 2818 sd_printk(KERN_NOTICE, sdkp, 2819 "%llu %d-byte logical blocks: (%s/%s)\n", 2820 (unsigned long long)sdkp->capacity, 2821 sector_size, cap_str_10, cap_str_2); 2822 2823 if (sdkp->physical_block_size != sector_size) 2824 sd_printk(KERN_NOTICE, sdkp, 2825 "%u-byte physical blocks\n", 2826 sdkp->physical_block_size); 2827 } 2828 2829 /* called with buffer of length 512 */ 2830 static inline int 2831 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage, 2832 unsigned char *buffer, int len, struct scsi_mode_data *data, 2833 struct scsi_sense_hdr *sshdr) 2834 { 2835 /* 2836 * If we must use MODE SENSE(10), make sure that the buffer length 2837 * is at least 8 bytes so that the mode sense header fits. 2838 */ 2839 if (sdkp->device->use_10_for_ms && len < 8) 2840 len = 8; 2841 2842 return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len, 2843 SD_TIMEOUT, sdkp->max_retries, data, sshdr); 2844 } 2845 2846 /* 2847 * read write protect setting, if possible - called only in sd_revalidate_disk() 2848 * called with buffer of length SD_BUF_SIZE 2849 */ 2850 static void 2851 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer) 2852 { 2853 int res; 2854 struct scsi_device *sdp = sdkp->device; 2855 struct scsi_mode_data data; 2856 int old_wp = sdkp->write_prot; 2857 2858 set_disk_ro(sdkp->disk, 0); 2859 if (sdp->skip_ms_page_3f) { 2860 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n"); 2861 return; 2862 } 2863 2864 if (sdp->use_192_bytes_for_3f) { 2865 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL); 2866 } else { 2867 /* 2868 * First attempt: ask for all pages (0x3F), but only 4 bytes. 2869 * We have to start carefully: some devices hang if we ask 2870 * for more than is available. 2871 */ 2872 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL); 2873 2874 /* 2875 * Second attempt: ask for page 0 When only page 0 is 2876 * implemented, a request for page 3F may return Sense Key 2877 * 5: Illegal Request, Sense Code 24: Invalid field in 2878 * CDB. 2879 */ 2880 if (res < 0) 2881 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL); 2882 2883 /* 2884 * Third attempt: ask 255 bytes, as we did earlier. 2885 */ 2886 if (res < 0) 2887 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255, 2888 &data, NULL); 2889 } 2890 2891 if (res < 0) { 2892 sd_first_printk(KERN_WARNING, sdkp, 2893 "Test WP failed, assume Write Enabled\n"); 2894 } else { 2895 sdkp->write_prot = ((data.device_specific & 0x80) != 0); 2896 set_disk_ro(sdkp->disk, sdkp->write_prot); 2897 if (sdkp->first_scan || old_wp != sdkp->write_prot) { 2898 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n", 2899 sdkp->write_prot ? "on" : "off"); 2900 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer); 2901 } 2902 } 2903 } 2904 2905 /* 2906 * sd_read_cache_type - called only from sd_revalidate_disk() 2907 * called with buffer of length SD_BUF_SIZE 2908 */ 2909 static void 2910 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer) 2911 { 2912 int len = 0, res; 2913 struct scsi_device *sdp = sdkp->device; 2914 2915 int dbd; 2916 int modepage; 2917 int first_len; 2918 struct scsi_mode_data data; 2919 struct scsi_sense_hdr sshdr; 2920 int old_wce = sdkp->WCE; 2921 int old_rcd = sdkp->RCD; 2922 int old_dpofua = sdkp->DPOFUA; 2923 2924 2925 if (sdkp->cache_override) 2926 return; 2927 2928 first_len = 4; 2929 if (sdp->skip_ms_page_8) { 2930 if (sdp->type == TYPE_RBC) 2931 goto defaults; 2932 else { 2933 if (sdp->skip_ms_page_3f) 2934 goto defaults; 2935 modepage = 0x3F; 2936 if (sdp->use_192_bytes_for_3f) 2937 first_len = 192; 2938 dbd = 0; 2939 } 2940 } else if (sdp->type == TYPE_RBC) { 2941 modepage = 6; 2942 dbd = 8; 2943 } else { 2944 modepage = 8; 2945 dbd = 0; 2946 } 2947 2948 /* cautiously ask */ 2949 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len, 2950 &data, &sshdr); 2951 2952 if (res < 0) 2953 goto bad_sense; 2954 2955 if (!data.header_length) { 2956 modepage = 6; 2957 first_len = 0; 2958 sd_first_printk(KERN_ERR, sdkp, 2959 "Missing header in MODE_SENSE response\n"); 2960 } 2961 2962 /* that went OK, now ask for the proper length */ 2963 len = data.length; 2964 2965 /* 2966 * We're only interested in the first three bytes, actually. 2967 * But the data cache page is defined for the first 20. 2968 */ 2969 if (len < 3) 2970 goto bad_sense; 2971 else if (len > SD_BUF_SIZE) { 2972 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter " 2973 "data from %d to %d bytes\n", len, SD_BUF_SIZE); 2974 len = SD_BUF_SIZE; 2975 } 2976 if (modepage == 0x3F && sdp->use_192_bytes_for_3f) 2977 len = 192; 2978 2979 /* Get the data */ 2980 if (len > first_len) 2981 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len, 2982 &data, &sshdr); 2983 2984 if (!res) { 2985 int offset = data.header_length + data.block_descriptor_length; 2986 2987 while (offset < len) { 2988 u8 page_code = buffer[offset] & 0x3F; 2989 u8 spf = buffer[offset] & 0x40; 2990 2991 if (page_code == 8 || page_code == 6) { 2992 /* We're interested only in the first 3 bytes. 2993 */ 2994 if (len - offset <= 2) { 2995 sd_first_printk(KERN_ERR, sdkp, 2996 "Incomplete mode parameter " 2997 "data\n"); 2998 goto defaults; 2999 } else { 3000 modepage = page_code; 3001 goto Page_found; 3002 } 3003 } else { 3004 /* Go to the next page */ 3005 if (spf && len - offset > 3) 3006 offset += 4 + (buffer[offset+2] << 8) + 3007 buffer[offset+3]; 3008 else if (!spf && len - offset > 1) 3009 offset += 2 + buffer[offset+1]; 3010 else { 3011 sd_first_printk(KERN_ERR, sdkp, 3012 "Incomplete mode " 3013 "parameter data\n"); 3014 goto defaults; 3015 } 3016 } 3017 } 3018 3019 sd_first_printk(KERN_WARNING, sdkp, 3020 "No Caching mode page found\n"); 3021 goto defaults; 3022 3023 Page_found: 3024 if (modepage == 8) { 3025 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0); 3026 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0); 3027 } else { 3028 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0); 3029 sdkp->RCD = 0; 3030 } 3031 3032 sdkp->DPOFUA = (data.device_specific & 0x10) != 0; 3033 if (sdp->broken_fua) { 3034 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n"); 3035 sdkp->DPOFUA = 0; 3036 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw && 3037 !sdkp->device->use_16_for_rw) { 3038 sd_first_printk(KERN_NOTICE, sdkp, 3039 "Uses READ/WRITE(6), disabling FUA\n"); 3040 sdkp->DPOFUA = 0; 3041 } 3042 3043 /* No cache flush allowed for write protected devices */ 3044 if (sdkp->WCE && sdkp->write_prot) 3045 sdkp->WCE = 0; 3046 3047 if (sdkp->first_scan || old_wce != sdkp->WCE || 3048 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA) 3049 sd_printk(KERN_NOTICE, sdkp, 3050 "Write cache: %s, read cache: %s, %s\n", 3051 sdkp->WCE ? "enabled" : "disabled", 3052 sdkp->RCD ? "disabled" : "enabled", 3053 sdkp->DPOFUA ? "supports DPO and FUA" 3054 : "doesn't support DPO or FUA"); 3055 3056 return; 3057 } 3058 3059 bad_sense: 3060 if (res == -EIO && scsi_sense_valid(&sshdr) && 3061 sshdr.sense_key == ILLEGAL_REQUEST && 3062 sshdr.asc == 0x24 && sshdr.ascq == 0x0) 3063 /* Invalid field in CDB */ 3064 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n"); 3065 else 3066 sd_first_printk(KERN_ERR, sdkp, 3067 "Asking for cache data failed\n"); 3068 3069 defaults: 3070 if (sdp->wce_default_on) { 3071 sd_first_printk(KERN_NOTICE, sdkp, 3072 "Assuming drive cache: write back\n"); 3073 sdkp->WCE = 1; 3074 } else { 3075 sd_first_printk(KERN_WARNING, sdkp, 3076 "Assuming drive cache: write through\n"); 3077 sdkp->WCE = 0; 3078 } 3079 sdkp->RCD = 0; 3080 sdkp->DPOFUA = 0; 3081 } 3082 3083 static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id) 3084 { 3085 u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS }; 3086 struct { 3087 struct scsi_stream_status_header h; 3088 struct scsi_stream_status s; 3089 } buf; 3090 struct scsi_device *sdev = sdkp->device; 3091 struct scsi_sense_hdr sshdr; 3092 const struct scsi_exec_args exec_args = { 3093 .sshdr = &sshdr, 3094 }; 3095 int res; 3096 3097 put_unaligned_be16(stream_id, &cdb[4]); 3098 put_unaligned_be32(sizeof(buf), &cdb[10]); 3099 3100 res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf), 3101 SD_TIMEOUT, sdkp->max_retries, &exec_args); 3102 if (res < 0) 3103 return false; 3104 if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr)) 3105 sd_print_sense_hdr(sdkp, &sshdr); 3106 if (res) 3107 return false; 3108 if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status)) 3109 return false; 3110 return buf.h.stream_status[0].perm; 3111 } 3112 3113 static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer) 3114 { 3115 struct scsi_device *sdp = sdkp->device; 3116 const struct scsi_io_group_descriptor *desc, *start, *end; 3117 u16 permanent_stream_count_old; 3118 struct scsi_sense_hdr sshdr; 3119 struct scsi_mode_data data; 3120 int res; 3121 3122 if (sdp->sdev_bflags & BLIST_SKIP_IO_HINTS) 3123 return; 3124 3125 res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a, 3126 /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT, 3127 sdkp->max_retries, &data, &sshdr); 3128 if (res < 0) 3129 return; 3130 start = (void *)buffer + data.header_length + 16; 3131 end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length, 3132 sizeof(*end)); 3133 /* 3134 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs 3135 * should assign the lowest numbered stream identifiers to permanent 3136 * streams. 3137 */ 3138 for (desc = start; desc < end; desc++) 3139 if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start)) 3140 break; 3141 permanent_stream_count_old = sdkp->permanent_stream_count; 3142 sdkp->permanent_stream_count = desc - start; 3143 if (sdkp->rscs && sdkp->permanent_stream_count < 2) 3144 sd_printk(KERN_INFO, sdkp, 3145 "Unexpected: RSCS has been set and the permanent stream count is %u\n", 3146 sdkp->permanent_stream_count); 3147 else if (sdkp->permanent_stream_count != permanent_stream_count_old) 3148 sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n", 3149 sdkp->permanent_stream_count); 3150 } 3151 3152 /* 3153 * The ATO bit indicates whether the DIF application tag is available 3154 * for use by the operating system. 3155 */ 3156 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer) 3157 { 3158 int res, offset; 3159 struct scsi_device *sdp = sdkp->device; 3160 struct scsi_mode_data data; 3161 struct scsi_sense_hdr sshdr; 3162 3163 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 3164 return; 3165 3166 if (sdkp->protection_type == 0) 3167 return; 3168 3169 res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT, 3170 sdkp->max_retries, &data, &sshdr); 3171 3172 if (res < 0 || !data.header_length || 3173 data.length < 6) { 3174 sd_first_printk(KERN_WARNING, sdkp, 3175 "getting Control mode page failed, assume no ATO\n"); 3176 3177 if (res == -EIO && scsi_sense_valid(&sshdr)) 3178 sd_print_sense_hdr(sdkp, &sshdr); 3179 3180 return; 3181 } 3182 3183 offset = data.header_length + data.block_descriptor_length; 3184 3185 if ((buffer[offset] & 0x3f) != 0x0a) { 3186 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n"); 3187 return; 3188 } 3189 3190 if ((buffer[offset + 5] & 0x80) == 0) 3191 return; 3192 3193 sdkp->ATO = 1; 3194 3195 return; 3196 } 3197 3198 /** 3199 * sd_read_block_limits - Query disk device for preferred I/O sizes. 3200 * @sdkp: disk to query 3201 */ 3202 static void sd_read_block_limits(struct scsi_disk *sdkp) 3203 { 3204 struct scsi_vpd *vpd; 3205 3206 rcu_read_lock(); 3207 3208 vpd = rcu_dereference(sdkp->device->vpd_pgb0); 3209 if (!vpd || vpd->len < 16) 3210 goto out; 3211 3212 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]); 3213 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]); 3214 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]); 3215 3216 if (vpd->len >= 64) { 3217 unsigned int lba_count, desc_count; 3218 3219 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]); 3220 3221 if (!sdkp->lbpme) 3222 goto out; 3223 3224 lba_count = get_unaligned_be32(&vpd->data[20]); 3225 desc_count = get_unaligned_be32(&vpd->data[24]); 3226 3227 if (lba_count && desc_count) 3228 sdkp->max_unmap_blocks = lba_count; 3229 3230 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]); 3231 3232 if (vpd->data[32] & 0x80) 3233 sdkp->unmap_alignment = 3234 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31); 3235 3236 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */ 3237 3238 if (sdkp->max_unmap_blocks) 3239 sd_config_discard(sdkp, SD_LBP_UNMAP); 3240 else 3241 sd_config_discard(sdkp, SD_LBP_WS16); 3242 3243 } else { /* LBP VPD page tells us what to use */ 3244 if (sdkp->lbpu && sdkp->max_unmap_blocks) 3245 sd_config_discard(sdkp, SD_LBP_UNMAP); 3246 else if (sdkp->lbpws) 3247 sd_config_discard(sdkp, SD_LBP_WS16); 3248 else if (sdkp->lbpws10) 3249 sd_config_discard(sdkp, SD_LBP_WS10); 3250 else 3251 sd_config_discard(sdkp, SD_LBP_DISABLE); 3252 } 3253 } 3254 3255 out: 3256 rcu_read_unlock(); 3257 } 3258 3259 /* Parse the Block Limits Extension VPD page (0xb7) */ 3260 static void sd_read_block_limits_ext(struct scsi_disk *sdkp) 3261 { 3262 struct scsi_vpd *vpd; 3263 3264 rcu_read_lock(); 3265 vpd = rcu_dereference(sdkp->device->vpd_pgb7); 3266 if (vpd && vpd->len >= 2) 3267 sdkp->rscs = vpd->data[5] & 1; 3268 rcu_read_unlock(); 3269 } 3270 3271 /** 3272 * sd_read_block_characteristics - Query block dev. characteristics 3273 * @sdkp: disk to query 3274 */ 3275 static void sd_read_block_characteristics(struct scsi_disk *sdkp) 3276 { 3277 struct request_queue *q = sdkp->disk->queue; 3278 struct scsi_vpd *vpd; 3279 u16 rot; 3280 3281 rcu_read_lock(); 3282 vpd = rcu_dereference(sdkp->device->vpd_pgb1); 3283 3284 if (!vpd || vpd->len < 8) { 3285 rcu_read_unlock(); 3286 return; 3287 } 3288 3289 rot = get_unaligned_be16(&vpd->data[4]); 3290 sdkp->zoned = (vpd->data[8] >> 4) & 3; 3291 rcu_read_unlock(); 3292 3293 if (rot == 1) { 3294 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 3295 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q); 3296 } 3297 3298 3299 #ifdef CONFIG_BLK_DEV_ZONED /* sd_probe rejects ZBD devices early otherwise */ 3300 if (sdkp->device->type == TYPE_ZBC) { 3301 /* 3302 * Host-managed. 3303 */ 3304 disk_set_zoned(sdkp->disk); 3305 3306 /* 3307 * Per ZBC and ZAC specifications, writes in sequential write 3308 * required zones of host-managed devices must be aligned to 3309 * the device physical block size. 3310 */ 3311 blk_queue_zone_write_granularity(q, sdkp->physical_block_size); 3312 } else { 3313 /* 3314 * Host-aware devices are treated as conventional. 3315 */ 3316 WARN_ON_ONCE(blk_queue_is_zoned(q)); 3317 } 3318 #endif /* CONFIG_BLK_DEV_ZONED */ 3319 3320 if (!sdkp->first_scan) 3321 return; 3322 3323 if (blk_queue_is_zoned(q)) 3324 sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n"); 3325 else if (sdkp->zoned == 1) 3326 sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n"); 3327 else if (sdkp->zoned == 2) 3328 sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n"); 3329 } 3330 3331 /** 3332 * sd_read_block_provisioning - Query provisioning VPD page 3333 * @sdkp: disk to query 3334 */ 3335 static void sd_read_block_provisioning(struct scsi_disk *sdkp) 3336 { 3337 struct scsi_vpd *vpd; 3338 3339 if (sdkp->lbpme == 0) 3340 return; 3341 3342 rcu_read_lock(); 3343 vpd = rcu_dereference(sdkp->device->vpd_pgb2); 3344 3345 if (!vpd || vpd->len < 8) { 3346 rcu_read_unlock(); 3347 return; 3348 } 3349 3350 sdkp->lbpvpd = 1; 3351 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */ 3352 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */ 3353 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */ 3354 rcu_read_unlock(); 3355 } 3356 3357 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer) 3358 { 3359 struct scsi_device *sdev = sdkp->device; 3360 3361 if (sdev->host->no_write_same) { 3362 sdev->no_write_same = 1; 3363 3364 return; 3365 } 3366 3367 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) { 3368 struct scsi_vpd *vpd; 3369 3370 sdev->no_report_opcodes = 1; 3371 3372 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION 3373 * CODES is unsupported and the device has an ATA 3374 * Information VPD page (SAT). 3375 */ 3376 rcu_read_lock(); 3377 vpd = rcu_dereference(sdev->vpd_pg89); 3378 if (vpd) 3379 sdev->no_write_same = 1; 3380 rcu_read_unlock(); 3381 } 3382 3383 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1) 3384 sdkp->ws16 = 1; 3385 3386 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1) 3387 sdkp->ws10 = 1; 3388 } 3389 3390 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer) 3391 { 3392 struct scsi_device *sdev = sdkp->device; 3393 3394 if (!sdev->security_supported) 3395 return; 3396 3397 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, 3398 SECURITY_PROTOCOL_IN, 0) == 1 && 3399 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, 3400 SECURITY_PROTOCOL_OUT, 0) == 1) 3401 sdkp->security = 1; 3402 } 3403 3404 static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf) 3405 { 3406 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf)); 3407 } 3408 3409 /** 3410 * sd_read_cpr - Query concurrent positioning ranges 3411 * @sdkp: disk to query 3412 */ 3413 static void sd_read_cpr(struct scsi_disk *sdkp) 3414 { 3415 struct blk_independent_access_ranges *iars = NULL; 3416 unsigned char *buffer = NULL; 3417 unsigned int nr_cpr = 0; 3418 int i, vpd_len, buf_len = SD_BUF_SIZE; 3419 u8 *desc; 3420 3421 /* 3422 * We need to have the capacity set first for the block layer to be 3423 * able to check the ranges. 3424 */ 3425 if (sdkp->first_scan) 3426 return; 3427 3428 if (!sdkp->capacity) 3429 goto out; 3430 3431 /* 3432 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges, 3433 * leading to a maximum page size of 64 + 256*32 bytes. 3434 */ 3435 buf_len = 64 + 256*32; 3436 buffer = kmalloc(buf_len, GFP_KERNEL); 3437 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len)) 3438 goto out; 3439 3440 /* We must have at least a 64B header and one 32B range descriptor */ 3441 vpd_len = get_unaligned_be16(&buffer[2]) + 4; 3442 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) { 3443 sd_printk(KERN_ERR, sdkp, 3444 "Invalid Concurrent Positioning Ranges VPD page\n"); 3445 goto out; 3446 } 3447 3448 nr_cpr = (vpd_len - 64) / 32; 3449 if (nr_cpr == 1) { 3450 nr_cpr = 0; 3451 goto out; 3452 } 3453 3454 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr); 3455 if (!iars) { 3456 nr_cpr = 0; 3457 goto out; 3458 } 3459 3460 desc = &buffer[64]; 3461 for (i = 0; i < nr_cpr; i++, desc += 32) { 3462 if (desc[0] != i) { 3463 sd_printk(KERN_ERR, sdkp, 3464 "Invalid Concurrent Positioning Range number\n"); 3465 nr_cpr = 0; 3466 break; 3467 } 3468 3469 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8); 3470 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16); 3471 } 3472 3473 out: 3474 disk_set_independent_access_ranges(sdkp->disk, iars); 3475 if (nr_cpr && sdkp->nr_actuators != nr_cpr) { 3476 sd_printk(KERN_NOTICE, sdkp, 3477 "%u concurrent positioning ranges\n", nr_cpr); 3478 sdkp->nr_actuators = nr_cpr; 3479 } 3480 3481 kfree(buffer); 3482 } 3483 3484 static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp) 3485 { 3486 struct scsi_device *sdp = sdkp->device; 3487 unsigned int min_xfer_bytes = 3488 logical_to_bytes(sdp, sdkp->min_xfer_blocks); 3489 3490 if (sdkp->min_xfer_blocks == 0) 3491 return false; 3492 3493 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) { 3494 sd_first_printk(KERN_WARNING, sdkp, 3495 "Preferred minimum I/O size %u bytes not a " \ 3496 "multiple of physical block size (%u bytes)\n", 3497 min_xfer_bytes, sdkp->physical_block_size); 3498 sdkp->min_xfer_blocks = 0; 3499 return false; 3500 } 3501 3502 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n", 3503 min_xfer_bytes); 3504 return true; 3505 } 3506 3507 /* 3508 * Determine the device's preferred I/O size for reads and writes 3509 * unless the reported value is unreasonably small, large, not a 3510 * multiple of the physical block size, or simply garbage. 3511 */ 3512 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp, 3513 unsigned int dev_max) 3514 { 3515 struct scsi_device *sdp = sdkp->device; 3516 unsigned int opt_xfer_bytes = 3517 logical_to_bytes(sdp, sdkp->opt_xfer_blocks); 3518 unsigned int min_xfer_bytes = 3519 logical_to_bytes(sdp, sdkp->min_xfer_blocks); 3520 3521 if (sdkp->opt_xfer_blocks == 0) 3522 return false; 3523 3524 if (sdkp->opt_xfer_blocks > dev_max) { 3525 sd_first_printk(KERN_WARNING, sdkp, 3526 "Optimal transfer size %u logical blocks " \ 3527 "> dev_max (%u logical blocks)\n", 3528 sdkp->opt_xfer_blocks, dev_max); 3529 return false; 3530 } 3531 3532 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) { 3533 sd_first_printk(KERN_WARNING, sdkp, 3534 "Optimal transfer size %u logical blocks " \ 3535 "> sd driver limit (%u logical blocks)\n", 3536 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS); 3537 return false; 3538 } 3539 3540 if (opt_xfer_bytes < PAGE_SIZE) { 3541 sd_first_printk(KERN_WARNING, sdkp, 3542 "Optimal transfer size %u bytes < " \ 3543 "PAGE_SIZE (%u bytes)\n", 3544 opt_xfer_bytes, (unsigned int)PAGE_SIZE); 3545 return false; 3546 } 3547 3548 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) { 3549 sd_first_printk(KERN_WARNING, sdkp, 3550 "Optimal transfer size %u bytes not a " \ 3551 "multiple of preferred minimum block " \ 3552 "size (%u bytes)\n", 3553 opt_xfer_bytes, min_xfer_bytes); 3554 return false; 3555 } 3556 3557 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) { 3558 sd_first_printk(KERN_WARNING, sdkp, 3559 "Optimal transfer size %u bytes not a " \ 3560 "multiple of physical block size (%u bytes)\n", 3561 opt_xfer_bytes, sdkp->physical_block_size); 3562 return false; 3563 } 3564 3565 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n", 3566 opt_xfer_bytes); 3567 return true; 3568 } 3569 3570 static void sd_read_block_zero(struct scsi_disk *sdkp) 3571 { 3572 struct scsi_device *sdev = sdkp->device; 3573 unsigned int buf_len = sdev->sector_size; 3574 u8 *buffer, cmd[16] = { }; 3575 3576 buffer = kmalloc(buf_len, GFP_KERNEL); 3577 if (!buffer) 3578 return; 3579 3580 if (sdev->use_16_for_rw) { 3581 cmd[0] = READ_16; 3582 put_unaligned_be64(0, &cmd[2]); /* Logical block address 0 */ 3583 put_unaligned_be32(1, &cmd[10]);/* Transfer 1 logical block */ 3584 } else { 3585 cmd[0] = READ_10; 3586 put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */ 3587 put_unaligned_be16(1, &cmd[7]); /* Transfer 1 logical block */ 3588 } 3589 3590 scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len, 3591 SD_TIMEOUT, sdkp->max_retries, NULL); 3592 kfree(buffer); 3593 } 3594 3595 /** 3596 * sd_revalidate_disk - called the first time a new disk is seen, 3597 * performs disk spin up, read_capacity, etc. 3598 * @disk: struct gendisk we care about 3599 **/ 3600 static int sd_revalidate_disk(struct gendisk *disk) 3601 { 3602 struct scsi_disk *sdkp = scsi_disk(disk); 3603 struct scsi_device *sdp = sdkp->device; 3604 struct request_queue *q = sdkp->disk->queue; 3605 sector_t old_capacity = sdkp->capacity; 3606 unsigned char *buffer; 3607 unsigned int dev_max, rw_max; 3608 3609 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, 3610 "sd_revalidate_disk\n")); 3611 3612 /* 3613 * If the device is offline, don't try and read capacity or any 3614 * of the other niceties. 3615 */ 3616 if (!scsi_device_online(sdp)) 3617 goto out; 3618 3619 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL); 3620 if (!buffer) { 3621 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory " 3622 "allocation failure.\n"); 3623 goto out; 3624 } 3625 3626 sd_spinup_disk(sdkp); 3627 3628 /* 3629 * Without media there is no reason to ask; moreover, some devices 3630 * react badly if we do. 3631 */ 3632 if (sdkp->media_present) { 3633 sd_read_capacity(sdkp, buffer); 3634 /* 3635 * Some USB/UAS devices return generic values for mode pages 3636 * until the media has been accessed. Trigger a READ operation 3637 * to force the device to populate mode pages. 3638 */ 3639 if (sdp->read_before_ms) 3640 sd_read_block_zero(sdkp); 3641 /* 3642 * set the default to rotational. All non-rotational devices 3643 * support the block characteristics VPD page, which will 3644 * cause this to be updated correctly and any device which 3645 * doesn't support it should be treated as rotational. 3646 */ 3647 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 3648 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q); 3649 3650 if (scsi_device_supports_vpd(sdp)) { 3651 sd_read_block_provisioning(sdkp); 3652 sd_read_block_limits(sdkp); 3653 sd_read_block_limits_ext(sdkp); 3654 sd_read_block_characteristics(sdkp); 3655 sd_zbc_read_zones(sdkp, buffer); 3656 sd_read_cpr(sdkp); 3657 } 3658 3659 sd_print_capacity(sdkp, old_capacity); 3660 3661 sd_read_write_protect_flag(sdkp, buffer); 3662 sd_read_cache_type(sdkp, buffer); 3663 sd_read_io_hints(sdkp, buffer); 3664 sd_read_app_tag_own(sdkp, buffer); 3665 sd_read_write_same(sdkp, buffer); 3666 sd_read_security(sdkp, buffer); 3667 sd_config_protection(sdkp); 3668 } 3669 3670 /* 3671 * We now have all cache related info, determine how we deal 3672 * with flush requests. 3673 */ 3674 sd_set_flush_flag(sdkp); 3675 3676 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */ 3677 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS; 3678 3679 /* Some devices report a maximum block count for READ/WRITE requests. */ 3680 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks); 3681 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max); 3682 3683 if (sd_validate_min_xfer_size(sdkp)) 3684 blk_queue_io_min(sdkp->disk->queue, 3685 logical_to_bytes(sdp, sdkp->min_xfer_blocks)); 3686 else 3687 blk_queue_io_min(sdkp->disk->queue, 0); 3688 3689 if (sd_validate_opt_xfer_size(sdkp, dev_max)) { 3690 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks); 3691 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks); 3692 } else { 3693 q->limits.io_opt = 0; 3694 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max), 3695 (sector_t)BLK_DEF_MAX_SECTORS_CAP); 3696 } 3697 3698 /* 3699 * Limit default to SCSI host optimal sector limit if set. There may be 3700 * an impact on performance for when the size of a request exceeds this 3701 * host limit. 3702 */ 3703 rw_max = min_not_zero(rw_max, sdp->host->opt_sectors); 3704 3705 /* Do not exceed controller limit */ 3706 rw_max = min(rw_max, queue_max_hw_sectors(q)); 3707 3708 /* 3709 * Only update max_sectors if previously unset or if the current value 3710 * exceeds the capabilities of the hardware. 3711 */ 3712 if (sdkp->first_scan || 3713 q->limits.max_sectors > q->limits.max_dev_sectors || 3714 q->limits.max_sectors > q->limits.max_hw_sectors) { 3715 q->limits.max_sectors = rw_max; 3716 q->limits.max_user_sectors = rw_max; 3717 } 3718 3719 sdkp->first_scan = 0; 3720 3721 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity)); 3722 sd_config_write_same(sdkp); 3723 kfree(buffer); 3724 3725 /* 3726 * For a zoned drive, revalidating the zones can be done only once 3727 * the gendisk capacity is set. So if this fails, set back the gendisk 3728 * capacity to 0. 3729 */ 3730 if (sd_zbc_revalidate_zones(sdkp)) 3731 set_capacity_and_notify(disk, 0); 3732 3733 out: 3734 return 0; 3735 } 3736 3737 /** 3738 * sd_unlock_native_capacity - unlock native capacity 3739 * @disk: struct gendisk to set capacity for 3740 * 3741 * Block layer calls this function if it detects that partitions 3742 * on @disk reach beyond the end of the device. If the SCSI host 3743 * implements ->unlock_native_capacity() method, it's invoked to 3744 * give it a chance to adjust the device capacity. 3745 * 3746 * CONTEXT: 3747 * Defined by block layer. Might sleep. 3748 */ 3749 static void sd_unlock_native_capacity(struct gendisk *disk) 3750 { 3751 struct scsi_device *sdev = scsi_disk(disk)->device; 3752 3753 if (sdev->host->hostt->unlock_native_capacity) 3754 sdev->host->hostt->unlock_native_capacity(sdev); 3755 } 3756 3757 /** 3758 * sd_format_disk_name - format disk name 3759 * @prefix: name prefix - ie. "sd" for SCSI disks 3760 * @index: index of the disk to format name for 3761 * @buf: output buffer 3762 * @buflen: length of the output buffer 3763 * 3764 * SCSI disk names starts at sda. The 26th device is sdz and the 3765 * 27th is sdaa. The last one for two lettered suffix is sdzz 3766 * which is followed by sdaaa. 3767 * 3768 * This is basically 26 base counting with one extra 'nil' entry 3769 * at the beginning from the second digit on and can be 3770 * determined using similar method as 26 base conversion with the 3771 * index shifted -1 after each digit is computed. 3772 * 3773 * CONTEXT: 3774 * Don't care. 3775 * 3776 * RETURNS: 3777 * 0 on success, -errno on failure. 3778 */ 3779 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen) 3780 { 3781 const int base = 'z' - 'a' + 1; 3782 char *begin = buf + strlen(prefix); 3783 char *end = buf + buflen; 3784 char *p; 3785 int unit; 3786 3787 p = end - 1; 3788 *p = '\0'; 3789 unit = base; 3790 do { 3791 if (p == begin) 3792 return -EINVAL; 3793 *--p = 'a' + (index % unit); 3794 index = (index / unit) - 1; 3795 } while (index >= 0); 3796 3797 memmove(begin, p, end - p); 3798 memcpy(buf, prefix, strlen(prefix)); 3799 3800 return 0; 3801 } 3802 3803 /** 3804 * sd_probe - called during driver initialization and whenever a 3805 * new scsi device is attached to the system. It is called once 3806 * for each scsi device (not just disks) present. 3807 * @dev: pointer to device object 3808 * 3809 * Returns 0 if successful (or not interested in this scsi device 3810 * (e.g. scanner)); 1 when there is an error. 3811 * 3812 * Note: this function is invoked from the scsi mid-level. 3813 * This function sets up the mapping between a given 3814 * <host,channel,id,lun> (found in sdp) and new device name 3815 * (e.g. /dev/sda). More precisely it is the block device major 3816 * and minor number that is chosen here. 3817 * 3818 * Assume sd_probe is not re-entrant (for time being) 3819 * Also think about sd_probe() and sd_remove() running coincidentally. 3820 **/ 3821 static int sd_probe(struct device *dev) 3822 { 3823 struct scsi_device *sdp = to_scsi_device(dev); 3824 struct scsi_disk *sdkp; 3825 struct gendisk *gd; 3826 int index; 3827 int error; 3828 3829 scsi_autopm_get_device(sdp); 3830 error = -ENODEV; 3831 if (sdp->type != TYPE_DISK && 3832 sdp->type != TYPE_ZBC && 3833 sdp->type != TYPE_MOD && 3834 sdp->type != TYPE_RBC) 3835 goto out; 3836 3837 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) { 3838 sdev_printk(KERN_WARNING, sdp, 3839 "Unsupported ZBC host-managed device.\n"); 3840 goto out; 3841 } 3842 3843 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp, 3844 "sd_probe\n")); 3845 3846 error = -ENOMEM; 3847 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL); 3848 if (!sdkp) 3849 goto out; 3850 3851 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue, 3852 &sd_bio_compl_lkclass); 3853 if (!gd) 3854 goto out_free; 3855 3856 index = ida_alloc(&sd_index_ida, GFP_KERNEL); 3857 if (index < 0) { 3858 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n"); 3859 goto out_put; 3860 } 3861 3862 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN); 3863 if (error) { 3864 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n"); 3865 goto out_free_index; 3866 } 3867 3868 sdkp->device = sdp; 3869 sdkp->disk = gd; 3870 sdkp->index = index; 3871 sdkp->max_retries = SD_MAX_RETRIES; 3872 atomic_set(&sdkp->openers, 0); 3873 atomic_set(&sdkp->device->ioerr_cnt, 0); 3874 3875 if (!sdp->request_queue->rq_timeout) { 3876 if (sdp->type != TYPE_MOD) 3877 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT); 3878 else 3879 blk_queue_rq_timeout(sdp->request_queue, 3880 SD_MOD_TIMEOUT); 3881 } 3882 3883 device_initialize(&sdkp->disk_dev); 3884 sdkp->disk_dev.parent = get_device(dev); 3885 sdkp->disk_dev.class = &sd_disk_class; 3886 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev)); 3887 3888 error = device_add(&sdkp->disk_dev); 3889 if (error) { 3890 put_device(&sdkp->disk_dev); 3891 goto out; 3892 } 3893 3894 dev_set_drvdata(dev, sdkp); 3895 3896 gd->major = sd_major((index & 0xf0) >> 4); 3897 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00); 3898 gd->minors = SD_MINORS; 3899 3900 gd->fops = &sd_fops; 3901 gd->private_data = sdkp; 3902 3903 /* defaults, until the device tells us otherwise */ 3904 sdp->sector_size = 512; 3905 sdkp->capacity = 0; 3906 sdkp->media_present = 1; 3907 sdkp->write_prot = 0; 3908 sdkp->cache_override = 0; 3909 sdkp->WCE = 0; 3910 sdkp->RCD = 0; 3911 sdkp->ATO = 0; 3912 sdkp->first_scan = 1; 3913 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS; 3914 3915 sd_revalidate_disk(gd); 3916 3917 if (sdp->removable) { 3918 gd->flags |= GENHD_FL_REMOVABLE; 3919 gd->events |= DISK_EVENT_MEDIA_CHANGE; 3920 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT; 3921 } 3922 3923 blk_pm_runtime_init(sdp->request_queue, dev); 3924 if (sdp->rpm_autosuspend) { 3925 pm_runtime_set_autosuspend_delay(dev, 3926 sdp->host->rpm_autosuspend_delay); 3927 } 3928 3929 error = device_add_disk(dev, gd, NULL); 3930 if (error) { 3931 device_unregister(&sdkp->disk_dev); 3932 put_disk(gd); 3933 goto out; 3934 } 3935 3936 if (sdkp->security) { 3937 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit); 3938 if (sdkp->opal_dev) 3939 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n"); 3940 } 3941 3942 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n", 3943 sdp->removable ? "removable " : ""); 3944 scsi_autopm_put_device(sdp); 3945 3946 return 0; 3947 3948 out_free_index: 3949 ida_free(&sd_index_ida, index); 3950 out_put: 3951 put_disk(gd); 3952 out_free: 3953 kfree(sdkp); 3954 out: 3955 scsi_autopm_put_device(sdp); 3956 return error; 3957 } 3958 3959 /** 3960 * sd_remove - called whenever a scsi disk (previously recognized by 3961 * sd_probe) is detached from the system. It is called (potentially 3962 * multiple times) during sd module unload. 3963 * @dev: pointer to device object 3964 * 3965 * Note: this function is invoked from the scsi mid-level. 3966 * This function potentially frees up a device name (e.g. /dev/sdc) 3967 * that could be re-used by a subsequent sd_probe(). 3968 * This function is not called when the built-in sd driver is "exit-ed". 3969 **/ 3970 static int sd_remove(struct device *dev) 3971 { 3972 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3973 3974 scsi_autopm_get_device(sdkp->device); 3975 3976 device_del(&sdkp->disk_dev); 3977 del_gendisk(sdkp->disk); 3978 if (!sdkp->suspended) 3979 sd_shutdown(dev); 3980 3981 put_disk(sdkp->disk); 3982 return 0; 3983 } 3984 3985 static void scsi_disk_release(struct device *dev) 3986 { 3987 struct scsi_disk *sdkp = to_scsi_disk(dev); 3988 3989 ida_free(&sd_index_ida, sdkp->index); 3990 put_device(&sdkp->device->sdev_gendev); 3991 free_opal_dev(sdkp->opal_dev); 3992 3993 kfree(sdkp); 3994 } 3995 3996 static int sd_start_stop_device(struct scsi_disk *sdkp, int start) 3997 { 3998 unsigned char cmd[6] = { START_STOP }; /* START_VALID */ 3999 struct scsi_sense_hdr sshdr; 4000 const struct scsi_exec_args exec_args = { 4001 .sshdr = &sshdr, 4002 .req_flags = BLK_MQ_REQ_PM, 4003 }; 4004 struct scsi_device *sdp = sdkp->device; 4005 int res; 4006 4007 if (start) 4008 cmd[4] |= 1; /* START */ 4009 4010 if (sdp->start_stop_pwr_cond) 4011 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */ 4012 4013 if (!scsi_device_online(sdp)) 4014 return -ENODEV; 4015 4016 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT, 4017 sdkp->max_retries, &exec_args); 4018 if (res) { 4019 sd_print_result(sdkp, "Start/Stop Unit failed", res); 4020 if (res > 0 && scsi_sense_valid(&sshdr)) { 4021 sd_print_sense_hdr(sdkp, &sshdr); 4022 /* 0x3a is medium not present */ 4023 if (sshdr.asc == 0x3a) 4024 res = 0; 4025 } 4026 } 4027 4028 /* SCSI error codes must not go to the generic layer */ 4029 if (res) 4030 return -EIO; 4031 4032 return 0; 4033 } 4034 4035 /* 4036 * Send a SYNCHRONIZE CACHE instruction down to the device through 4037 * the normal SCSI command structure. Wait for the command to 4038 * complete. 4039 */ 4040 static void sd_shutdown(struct device *dev) 4041 { 4042 struct scsi_disk *sdkp = dev_get_drvdata(dev); 4043 4044 if (!sdkp) 4045 return; /* this can happen */ 4046 4047 if (pm_runtime_suspended(dev)) 4048 return; 4049 4050 if (sdkp->WCE && sdkp->media_present) { 4051 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n"); 4052 sd_sync_cache(sdkp); 4053 } 4054 4055 if ((system_state != SYSTEM_RESTART && 4056 sdkp->device->manage_system_start_stop) || 4057 (system_state == SYSTEM_POWER_OFF && 4058 sdkp->device->manage_shutdown)) { 4059 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n"); 4060 sd_start_stop_device(sdkp, 0); 4061 } 4062 } 4063 4064 static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime) 4065 { 4066 return (sdev->manage_system_start_stop && !runtime) || 4067 (sdev->manage_runtime_start_stop && runtime); 4068 } 4069 4070 static int sd_suspend_common(struct device *dev, bool runtime) 4071 { 4072 struct scsi_disk *sdkp = dev_get_drvdata(dev); 4073 int ret = 0; 4074 4075 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */ 4076 return 0; 4077 4078 if (sdkp->WCE && sdkp->media_present) { 4079 if (!sdkp->device->silence_suspend) 4080 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n"); 4081 ret = sd_sync_cache(sdkp); 4082 /* ignore OFFLINE device */ 4083 if (ret == -ENODEV) 4084 return 0; 4085 4086 if (ret) 4087 return ret; 4088 } 4089 4090 if (sd_do_start_stop(sdkp->device, runtime)) { 4091 if (!sdkp->device->silence_suspend) 4092 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n"); 4093 /* an error is not worth aborting a system sleep */ 4094 ret = sd_start_stop_device(sdkp, 0); 4095 if (!runtime) 4096 ret = 0; 4097 } 4098 4099 if (!ret) 4100 sdkp->suspended = true; 4101 4102 return ret; 4103 } 4104 4105 static int sd_suspend_system(struct device *dev) 4106 { 4107 if (pm_runtime_suspended(dev)) 4108 return 0; 4109 4110 return sd_suspend_common(dev, false); 4111 } 4112 4113 static int sd_suspend_runtime(struct device *dev) 4114 { 4115 return sd_suspend_common(dev, true); 4116 } 4117 4118 static int sd_resume(struct device *dev) 4119 { 4120 struct scsi_disk *sdkp = dev_get_drvdata(dev); 4121 4122 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n"); 4123 4124 if (opal_unlock_from_suspend(sdkp->opal_dev)) { 4125 sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n"); 4126 return -EIO; 4127 } 4128 4129 return 0; 4130 } 4131 4132 static int sd_resume_common(struct device *dev, bool runtime) 4133 { 4134 struct scsi_disk *sdkp = dev_get_drvdata(dev); 4135 int ret; 4136 4137 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */ 4138 return 0; 4139 4140 if (!sd_do_start_stop(sdkp->device, runtime)) { 4141 sdkp->suspended = false; 4142 return 0; 4143 } 4144 4145 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n"); 4146 ret = sd_start_stop_device(sdkp, 1); 4147 if (!ret) { 4148 sd_resume(dev); 4149 sdkp->suspended = false; 4150 } 4151 4152 return ret; 4153 } 4154 4155 static int sd_resume_system(struct device *dev) 4156 { 4157 if (pm_runtime_suspended(dev)) { 4158 struct scsi_disk *sdkp = dev_get_drvdata(dev); 4159 struct scsi_device *sdp = sdkp ? sdkp->device : NULL; 4160 4161 if (sdp && sdp->force_runtime_start_on_system_start) 4162 pm_request_resume(dev); 4163 4164 return 0; 4165 } 4166 4167 return sd_resume_common(dev, false); 4168 } 4169 4170 static int sd_resume_runtime(struct device *dev) 4171 { 4172 struct scsi_disk *sdkp = dev_get_drvdata(dev); 4173 struct scsi_device *sdp; 4174 4175 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */ 4176 return 0; 4177 4178 sdp = sdkp->device; 4179 4180 if (sdp->ignore_media_change) { 4181 /* clear the device's sense data */ 4182 static const u8 cmd[10] = { REQUEST_SENSE }; 4183 const struct scsi_exec_args exec_args = { 4184 .req_flags = BLK_MQ_REQ_PM, 4185 }; 4186 4187 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, 4188 sdp->request_queue->rq_timeout, 1, 4189 &exec_args)) 4190 sd_printk(KERN_NOTICE, sdkp, 4191 "Failed to clear sense data\n"); 4192 } 4193 4194 return sd_resume_common(dev, true); 4195 } 4196 4197 static const struct dev_pm_ops sd_pm_ops = { 4198 .suspend = sd_suspend_system, 4199 .resume = sd_resume_system, 4200 .poweroff = sd_suspend_system, 4201 .restore = sd_resume_system, 4202 .runtime_suspend = sd_suspend_runtime, 4203 .runtime_resume = sd_resume_runtime, 4204 }; 4205 4206 static struct scsi_driver sd_template = { 4207 .gendrv = { 4208 .name = "sd", 4209 .probe = sd_probe, 4210 .probe_type = PROBE_PREFER_ASYNCHRONOUS, 4211 .remove = sd_remove, 4212 .shutdown = sd_shutdown, 4213 .pm = &sd_pm_ops, 4214 }, 4215 .rescan = sd_rescan, 4216 .resume = sd_resume, 4217 .init_command = sd_init_command, 4218 .uninit_command = sd_uninit_command, 4219 .done = sd_done, 4220 .eh_action = sd_eh_action, 4221 .eh_reset = sd_eh_reset, 4222 }; 4223 4224 /** 4225 * init_sd - entry point for this driver (both when built in or when 4226 * a module). 4227 * 4228 * Note: this function registers this driver with the scsi mid-level. 4229 **/ 4230 static int __init init_sd(void) 4231 { 4232 int majors = 0, i, err; 4233 4234 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n")); 4235 4236 for (i = 0; i < SD_MAJORS; i++) { 4237 if (__register_blkdev(sd_major(i), "sd", sd_default_probe)) 4238 continue; 4239 majors++; 4240 } 4241 4242 if (!majors) 4243 return -ENODEV; 4244 4245 err = class_register(&sd_disk_class); 4246 if (err) 4247 goto err_out; 4248 4249 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0); 4250 if (!sd_page_pool) { 4251 printk(KERN_ERR "sd: can't init discard page pool\n"); 4252 err = -ENOMEM; 4253 goto err_out_class; 4254 } 4255 4256 err = scsi_register_driver(&sd_template.gendrv); 4257 if (err) 4258 goto err_out_driver; 4259 4260 return 0; 4261 4262 err_out_driver: 4263 mempool_destroy(sd_page_pool); 4264 err_out_class: 4265 class_unregister(&sd_disk_class); 4266 err_out: 4267 for (i = 0; i < SD_MAJORS; i++) 4268 unregister_blkdev(sd_major(i), "sd"); 4269 return err; 4270 } 4271 4272 /** 4273 * exit_sd - exit point for this driver (when it is a module). 4274 * 4275 * Note: this function unregisters this driver from the scsi mid-level. 4276 **/ 4277 static void __exit exit_sd(void) 4278 { 4279 int i; 4280 4281 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n")); 4282 4283 scsi_unregister_driver(&sd_template.gendrv); 4284 mempool_destroy(sd_page_pool); 4285 4286 class_unregister(&sd_disk_class); 4287 4288 for (i = 0; i < SD_MAJORS; i++) 4289 unregister_blkdev(sd_major(i), "sd"); 4290 } 4291 4292 module_init(init_sd); 4293 module_exit(exit_sd); 4294 4295 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr) 4296 { 4297 scsi_print_sense_hdr(sdkp->device, 4298 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr); 4299 } 4300 4301 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result) 4302 { 4303 const char *hb_string = scsi_hostbyte_string(result); 4304 4305 if (hb_string) 4306 sd_printk(KERN_INFO, sdkp, 4307 "%s: Result: hostbyte=%s driverbyte=%s\n", msg, 4308 hb_string ? hb_string : "invalid", 4309 "DRIVER_OK"); 4310 else 4311 sd_printk(KERN_INFO, sdkp, 4312 "%s: Result: hostbyte=0x%02x driverbyte=%s\n", 4313 msg, host_byte(result), "DRIVER_OK"); 4314 } 4315