1 /* 2 * sd.c Copyright (C) 1992 Drew Eckhardt 3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale 4 * 5 * Linux scsi disk driver 6 * Initial versions: Drew Eckhardt 7 * Subsequent revisions: Eric Youngdale 8 * Modification history: 9 * - Drew Eckhardt <drew@colorado.edu> original 10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple 11 * outstanding request, and other enhancements. 12 * Support loadable low-level scsi drivers. 13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using 14 * eight major numbers. 15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs. 16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in 17 * sd_init and cleanups. 18 * - Alex Davis <letmein@erols.com> Fix problem where partition info 19 * not being read in sd_open. Fix problem where removable media 20 * could be ejected after sd_open. 21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x 22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox 23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>: 24 * Support 32k/1M disks. 25 * 26 * Logging policy (needs CONFIG_SCSI_LOGGING defined): 27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2 28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1 29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1 30 * - entering other commands: SCSI_LOG_HLQUEUE level 3 31 * Note: when the logging level is set by the user, it must be greater 32 * than the level indicated above to trigger output. 33 */ 34 35 #include <linux/module.h> 36 #include <linux/fs.h> 37 #include <linux/kernel.h> 38 #include <linux/mm.h> 39 #include <linux/bio.h> 40 #include <linux/genhd.h> 41 #include <linux/hdreg.h> 42 #include <linux/errno.h> 43 #include <linux/idr.h> 44 #include <linux/interrupt.h> 45 #include <linux/init.h> 46 #include <linux/blkdev.h> 47 #include <linux/blkpg.h> 48 #include <linux/blk-pm.h> 49 #include <linux/delay.h> 50 #include <linux/mutex.h> 51 #include <linux/string_helpers.h> 52 #include <linux/async.h> 53 #include <linux/slab.h> 54 #include <linux/sed-opal.h> 55 #include <linux/pm_runtime.h> 56 #include <linux/pr.h> 57 #include <linux/t10-pi.h> 58 #include <linux/uaccess.h> 59 #include <asm/unaligned.h> 60 61 #include <scsi/scsi.h> 62 #include <scsi/scsi_cmnd.h> 63 #include <scsi/scsi_dbg.h> 64 #include <scsi/scsi_device.h> 65 #include <scsi/scsi_driver.h> 66 #include <scsi/scsi_eh.h> 67 #include <scsi/scsi_host.h> 68 #include <scsi/scsi_ioctl.h> 69 #include <scsi/scsicam.h> 70 71 #include "sd.h" 72 #include "scsi_priv.h" 73 #include "scsi_logging.h" 74 75 MODULE_AUTHOR("Eric Youngdale"); 76 MODULE_DESCRIPTION("SCSI disk (sd) driver"); 77 MODULE_LICENSE("GPL"); 78 79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR); 80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR); 81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR); 82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR); 83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR); 84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR); 85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR); 86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR); 87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR); 88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR); 89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR); 90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR); 91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR); 92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR); 93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR); 94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR); 95 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK); 96 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD); 97 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC); 98 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC); 99 100 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT) 101 #define SD_MINORS 16 102 #else 103 #define SD_MINORS 0 104 #endif 105 106 static void sd_config_discard(struct scsi_disk *, unsigned int); 107 static void sd_config_write_same(struct scsi_disk *); 108 static int sd_revalidate_disk(struct gendisk *); 109 static void sd_unlock_native_capacity(struct gendisk *disk); 110 static int sd_probe(struct device *); 111 static int sd_remove(struct device *); 112 static void sd_shutdown(struct device *); 113 static int sd_suspend_system(struct device *); 114 static int sd_suspend_runtime(struct device *); 115 static int sd_resume(struct device *); 116 static void sd_rescan(struct device *); 117 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt); 118 static void sd_uninit_command(struct scsi_cmnd *SCpnt); 119 static int sd_done(struct scsi_cmnd *); 120 static void sd_eh_reset(struct scsi_cmnd *); 121 static int sd_eh_action(struct scsi_cmnd *, int); 122 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer); 123 static void scsi_disk_release(struct device *cdev); 124 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *); 125 static void sd_print_result(const struct scsi_disk *, const char *, int); 126 127 static DEFINE_IDA(sd_index_ida); 128 129 /* This semaphore is used to mediate the 0->1 reference get in the 130 * face of object destruction (i.e. we can't allow a get on an 131 * object after last put) */ 132 static DEFINE_MUTEX(sd_ref_mutex); 133 134 static struct kmem_cache *sd_cdb_cache; 135 static mempool_t *sd_cdb_pool; 136 static mempool_t *sd_page_pool; 137 138 static const char *sd_cache_types[] = { 139 "write through", "none", "write back", 140 "write back, no read (daft)" 141 }; 142 143 static void sd_set_flush_flag(struct scsi_disk *sdkp) 144 { 145 bool wc = false, fua = false; 146 147 if (sdkp->WCE) { 148 wc = true; 149 if (sdkp->DPOFUA) 150 fua = true; 151 } 152 153 blk_queue_write_cache(sdkp->disk->queue, wc, fua); 154 } 155 156 static ssize_t 157 cache_type_store(struct device *dev, struct device_attribute *attr, 158 const char *buf, size_t count) 159 { 160 int ct, rcd, wce, sp; 161 struct scsi_disk *sdkp = to_scsi_disk(dev); 162 struct scsi_device *sdp = sdkp->device; 163 char buffer[64]; 164 char *buffer_data; 165 struct scsi_mode_data data; 166 struct scsi_sense_hdr sshdr; 167 static const char temp[] = "temporary "; 168 int len; 169 170 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 171 /* no cache control on RBC devices; theoretically they 172 * can do it, but there's probably so many exceptions 173 * it's not worth the risk */ 174 return -EINVAL; 175 176 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) { 177 buf += sizeof(temp) - 1; 178 sdkp->cache_override = 1; 179 } else { 180 sdkp->cache_override = 0; 181 } 182 183 ct = sysfs_match_string(sd_cache_types, buf); 184 if (ct < 0) 185 return -EINVAL; 186 187 rcd = ct & 0x01 ? 1 : 0; 188 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0; 189 190 if (sdkp->cache_override) { 191 sdkp->WCE = wce; 192 sdkp->RCD = rcd; 193 sd_set_flush_flag(sdkp); 194 return count; 195 } 196 197 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT, 198 SD_MAX_RETRIES, &data, NULL)) 199 return -EINVAL; 200 len = min_t(size_t, sizeof(buffer), data.length - data.header_length - 201 data.block_descriptor_length); 202 buffer_data = buffer + data.header_length + 203 data.block_descriptor_length; 204 buffer_data[2] &= ~0x05; 205 buffer_data[2] |= wce << 2 | rcd; 206 sp = buffer_data[0] & 0x80 ? 1 : 0; 207 buffer_data[0] &= ~0x80; 208 209 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT, 210 SD_MAX_RETRIES, &data, &sshdr)) { 211 if (scsi_sense_valid(&sshdr)) 212 sd_print_sense_hdr(sdkp, &sshdr); 213 return -EINVAL; 214 } 215 revalidate_disk(sdkp->disk); 216 return count; 217 } 218 219 static ssize_t 220 manage_start_stop_show(struct device *dev, struct device_attribute *attr, 221 char *buf) 222 { 223 struct scsi_disk *sdkp = to_scsi_disk(dev); 224 struct scsi_device *sdp = sdkp->device; 225 226 return sprintf(buf, "%u\n", sdp->manage_start_stop); 227 } 228 229 static ssize_t 230 manage_start_stop_store(struct device *dev, struct device_attribute *attr, 231 const char *buf, size_t count) 232 { 233 struct scsi_disk *sdkp = to_scsi_disk(dev); 234 struct scsi_device *sdp = sdkp->device; 235 bool v; 236 237 if (!capable(CAP_SYS_ADMIN)) 238 return -EACCES; 239 240 if (kstrtobool(buf, &v)) 241 return -EINVAL; 242 243 sdp->manage_start_stop = v; 244 245 return count; 246 } 247 static DEVICE_ATTR_RW(manage_start_stop); 248 249 static ssize_t 250 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf) 251 { 252 struct scsi_disk *sdkp = to_scsi_disk(dev); 253 254 return sprintf(buf, "%u\n", sdkp->device->allow_restart); 255 } 256 257 static ssize_t 258 allow_restart_store(struct device *dev, struct device_attribute *attr, 259 const char *buf, size_t count) 260 { 261 bool v; 262 struct scsi_disk *sdkp = to_scsi_disk(dev); 263 struct scsi_device *sdp = sdkp->device; 264 265 if (!capable(CAP_SYS_ADMIN)) 266 return -EACCES; 267 268 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 269 return -EINVAL; 270 271 if (kstrtobool(buf, &v)) 272 return -EINVAL; 273 274 sdp->allow_restart = v; 275 276 return count; 277 } 278 static DEVICE_ATTR_RW(allow_restart); 279 280 static ssize_t 281 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf) 282 { 283 struct scsi_disk *sdkp = to_scsi_disk(dev); 284 int ct = sdkp->RCD + 2*sdkp->WCE; 285 286 return sprintf(buf, "%s\n", sd_cache_types[ct]); 287 } 288 static DEVICE_ATTR_RW(cache_type); 289 290 static ssize_t 291 FUA_show(struct device *dev, struct device_attribute *attr, char *buf) 292 { 293 struct scsi_disk *sdkp = to_scsi_disk(dev); 294 295 return sprintf(buf, "%u\n", sdkp->DPOFUA); 296 } 297 static DEVICE_ATTR_RO(FUA); 298 299 static ssize_t 300 protection_type_show(struct device *dev, struct device_attribute *attr, 301 char *buf) 302 { 303 struct scsi_disk *sdkp = to_scsi_disk(dev); 304 305 return sprintf(buf, "%u\n", sdkp->protection_type); 306 } 307 308 static ssize_t 309 protection_type_store(struct device *dev, struct device_attribute *attr, 310 const char *buf, size_t count) 311 { 312 struct scsi_disk *sdkp = to_scsi_disk(dev); 313 unsigned int val; 314 int err; 315 316 if (!capable(CAP_SYS_ADMIN)) 317 return -EACCES; 318 319 err = kstrtouint(buf, 10, &val); 320 321 if (err) 322 return err; 323 324 if (val <= T10_PI_TYPE3_PROTECTION) 325 sdkp->protection_type = val; 326 327 return count; 328 } 329 static DEVICE_ATTR_RW(protection_type); 330 331 static ssize_t 332 protection_mode_show(struct device *dev, struct device_attribute *attr, 333 char *buf) 334 { 335 struct scsi_disk *sdkp = to_scsi_disk(dev); 336 struct scsi_device *sdp = sdkp->device; 337 unsigned int dif, dix; 338 339 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type); 340 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type); 341 342 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) { 343 dif = 0; 344 dix = 1; 345 } 346 347 if (!dif && !dix) 348 return sprintf(buf, "none\n"); 349 350 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif); 351 } 352 static DEVICE_ATTR_RO(protection_mode); 353 354 static ssize_t 355 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf) 356 { 357 struct scsi_disk *sdkp = to_scsi_disk(dev); 358 359 return sprintf(buf, "%u\n", sdkp->ATO); 360 } 361 static DEVICE_ATTR_RO(app_tag_own); 362 363 static ssize_t 364 thin_provisioning_show(struct device *dev, struct device_attribute *attr, 365 char *buf) 366 { 367 struct scsi_disk *sdkp = to_scsi_disk(dev); 368 369 return sprintf(buf, "%u\n", sdkp->lbpme); 370 } 371 static DEVICE_ATTR_RO(thin_provisioning); 372 373 /* sysfs_match_string() requires dense arrays */ 374 static const char *lbp_mode[] = { 375 [SD_LBP_FULL] = "full", 376 [SD_LBP_UNMAP] = "unmap", 377 [SD_LBP_WS16] = "writesame_16", 378 [SD_LBP_WS10] = "writesame_10", 379 [SD_LBP_ZERO] = "writesame_zero", 380 [SD_LBP_DISABLE] = "disabled", 381 }; 382 383 static ssize_t 384 provisioning_mode_show(struct device *dev, struct device_attribute *attr, 385 char *buf) 386 { 387 struct scsi_disk *sdkp = to_scsi_disk(dev); 388 389 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]); 390 } 391 392 static ssize_t 393 provisioning_mode_store(struct device *dev, struct device_attribute *attr, 394 const char *buf, size_t count) 395 { 396 struct scsi_disk *sdkp = to_scsi_disk(dev); 397 struct scsi_device *sdp = sdkp->device; 398 int mode; 399 400 if (!capable(CAP_SYS_ADMIN)) 401 return -EACCES; 402 403 if (sd_is_zoned(sdkp)) { 404 sd_config_discard(sdkp, SD_LBP_DISABLE); 405 return count; 406 } 407 408 if (sdp->type != TYPE_DISK) 409 return -EINVAL; 410 411 mode = sysfs_match_string(lbp_mode, buf); 412 if (mode < 0) 413 return -EINVAL; 414 415 sd_config_discard(sdkp, mode); 416 417 return count; 418 } 419 static DEVICE_ATTR_RW(provisioning_mode); 420 421 /* sysfs_match_string() requires dense arrays */ 422 static const char *zeroing_mode[] = { 423 [SD_ZERO_WRITE] = "write", 424 [SD_ZERO_WS] = "writesame", 425 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap", 426 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap", 427 }; 428 429 static ssize_t 430 zeroing_mode_show(struct device *dev, struct device_attribute *attr, 431 char *buf) 432 { 433 struct scsi_disk *sdkp = to_scsi_disk(dev); 434 435 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]); 436 } 437 438 static ssize_t 439 zeroing_mode_store(struct device *dev, struct device_attribute *attr, 440 const char *buf, size_t count) 441 { 442 struct scsi_disk *sdkp = to_scsi_disk(dev); 443 int mode; 444 445 if (!capable(CAP_SYS_ADMIN)) 446 return -EACCES; 447 448 mode = sysfs_match_string(zeroing_mode, buf); 449 if (mode < 0) 450 return -EINVAL; 451 452 sdkp->zeroing_mode = mode; 453 454 return count; 455 } 456 static DEVICE_ATTR_RW(zeroing_mode); 457 458 static ssize_t 459 max_medium_access_timeouts_show(struct device *dev, 460 struct device_attribute *attr, char *buf) 461 { 462 struct scsi_disk *sdkp = to_scsi_disk(dev); 463 464 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts); 465 } 466 467 static ssize_t 468 max_medium_access_timeouts_store(struct device *dev, 469 struct device_attribute *attr, const char *buf, 470 size_t count) 471 { 472 struct scsi_disk *sdkp = to_scsi_disk(dev); 473 int err; 474 475 if (!capable(CAP_SYS_ADMIN)) 476 return -EACCES; 477 478 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts); 479 480 return err ? err : count; 481 } 482 static DEVICE_ATTR_RW(max_medium_access_timeouts); 483 484 static ssize_t 485 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr, 486 char *buf) 487 { 488 struct scsi_disk *sdkp = to_scsi_disk(dev); 489 490 return sprintf(buf, "%u\n", sdkp->max_ws_blocks); 491 } 492 493 static ssize_t 494 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr, 495 const char *buf, size_t count) 496 { 497 struct scsi_disk *sdkp = to_scsi_disk(dev); 498 struct scsi_device *sdp = sdkp->device; 499 unsigned long max; 500 int err; 501 502 if (!capable(CAP_SYS_ADMIN)) 503 return -EACCES; 504 505 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 506 return -EINVAL; 507 508 err = kstrtoul(buf, 10, &max); 509 510 if (err) 511 return err; 512 513 if (max == 0) 514 sdp->no_write_same = 1; 515 else if (max <= SD_MAX_WS16_BLOCKS) { 516 sdp->no_write_same = 0; 517 sdkp->max_ws_blocks = max; 518 } 519 520 sd_config_write_same(sdkp); 521 522 return count; 523 } 524 static DEVICE_ATTR_RW(max_write_same_blocks); 525 526 static struct attribute *sd_disk_attrs[] = { 527 &dev_attr_cache_type.attr, 528 &dev_attr_FUA.attr, 529 &dev_attr_allow_restart.attr, 530 &dev_attr_manage_start_stop.attr, 531 &dev_attr_protection_type.attr, 532 &dev_attr_protection_mode.attr, 533 &dev_attr_app_tag_own.attr, 534 &dev_attr_thin_provisioning.attr, 535 &dev_attr_provisioning_mode.attr, 536 &dev_attr_zeroing_mode.attr, 537 &dev_attr_max_write_same_blocks.attr, 538 &dev_attr_max_medium_access_timeouts.attr, 539 NULL, 540 }; 541 ATTRIBUTE_GROUPS(sd_disk); 542 543 static struct class sd_disk_class = { 544 .name = "scsi_disk", 545 .owner = THIS_MODULE, 546 .dev_release = scsi_disk_release, 547 .dev_groups = sd_disk_groups, 548 }; 549 550 static const struct dev_pm_ops sd_pm_ops = { 551 .suspend = sd_suspend_system, 552 .resume = sd_resume, 553 .poweroff = sd_suspend_system, 554 .restore = sd_resume, 555 .runtime_suspend = sd_suspend_runtime, 556 .runtime_resume = sd_resume, 557 }; 558 559 static struct scsi_driver sd_template = { 560 .gendrv = { 561 .name = "sd", 562 .owner = THIS_MODULE, 563 .probe = sd_probe, 564 .remove = sd_remove, 565 .shutdown = sd_shutdown, 566 .pm = &sd_pm_ops, 567 }, 568 .rescan = sd_rescan, 569 .init_command = sd_init_command, 570 .uninit_command = sd_uninit_command, 571 .done = sd_done, 572 .eh_action = sd_eh_action, 573 .eh_reset = sd_eh_reset, 574 }; 575 576 /* 577 * Dummy kobj_map->probe function. 578 * The default ->probe function will call modprobe, which is 579 * pointless as this module is already loaded. 580 */ 581 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data) 582 { 583 return NULL; 584 } 585 586 /* 587 * Device no to disk mapping: 588 * 589 * major disc2 disc p1 590 * |............|.............|....|....| <- dev_t 591 * 31 20 19 8 7 4 3 0 592 * 593 * Inside a major, we have 16k disks, however mapped non- 594 * contiguously. The first 16 disks are for major0, the next 595 * ones with major1, ... Disk 256 is for major0 again, disk 272 596 * for major1, ... 597 * As we stay compatible with our numbering scheme, we can reuse 598 * the well-know SCSI majors 8, 65--71, 136--143. 599 */ 600 static int sd_major(int major_idx) 601 { 602 switch (major_idx) { 603 case 0: 604 return SCSI_DISK0_MAJOR; 605 case 1 ... 7: 606 return SCSI_DISK1_MAJOR + major_idx - 1; 607 case 8 ... 15: 608 return SCSI_DISK8_MAJOR + major_idx - 8; 609 default: 610 BUG(); 611 return 0; /* shut up gcc */ 612 } 613 } 614 615 static struct scsi_disk *scsi_disk_get(struct gendisk *disk) 616 { 617 struct scsi_disk *sdkp = NULL; 618 619 mutex_lock(&sd_ref_mutex); 620 621 if (disk->private_data) { 622 sdkp = scsi_disk(disk); 623 if (scsi_device_get(sdkp->device) == 0) 624 get_device(&sdkp->dev); 625 else 626 sdkp = NULL; 627 } 628 mutex_unlock(&sd_ref_mutex); 629 return sdkp; 630 } 631 632 static void scsi_disk_put(struct scsi_disk *sdkp) 633 { 634 struct scsi_device *sdev = sdkp->device; 635 636 mutex_lock(&sd_ref_mutex); 637 put_device(&sdkp->dev); 638 scsi_device_put(sdev); 639 mutex_unlock(&sd_ref_mutex); 640 } 641 642 #ifdef CONFIG_BLK_SED_OPAL 643 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, 644 size_t len, bool send) 645 { 646 struct scsi_device *sdev = data; 647 u8 cdb[12] = { 0, }; 648 int ret; 649 650 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN; 651 cdb[1] = secp; 652 put_unaligned_be16(spsp, &cdb[2]); 653 put_unaligned_be32(len, &cdb[6]); 654 655 ret = scsi_execute_req(sdev, cdb, 656 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE, 657 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL); 658 return ret <= 0 ? ret : -EIO; 659 } 660 #endif /* CONFIG_BLK_SED_OPAL */ 661 662 /* 663 * Look up the DIX operation based on whether the command is read or 664 * write and whether dix and dif are enabled. 665 */ 666 static unsigned int sd_prot_op(bool write, bool dix, bool dif) 667 { 668 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */ 669 static const unsigned int ops[] = { /* wrt dix dif */ 670 SCSI_PROT_NORMAL, /* 0 0 0 */ 671 SCSI_PROT_READ_STRIP, /* 0 0 1 */ 672 SCSI_PROT_READ_INSERT, /* 0 1 0 */ 673 SCSI_PROT_READ_PASS, /* 0 1 1 */ 674 SCSI_PROT_NORMAL, /* 1 0 0 */ 675 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */ 676 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */ 677 SCSI_PROT_WRITE_PASS, /* 1 1 1 */ 678 }; 679 680 return ops[write << 2 | dix << 1 | dif]; 681 } 682 683 /* 684 * Returns a mask of the protection flags that are valid for a given DIX 685 * operation. 686 */ 687 static unsigned int sd_prot_flag_mask(unsigned int prot_op) 688 { 689 static const unsigned int flag_mask[] = { 690 [SCSI_PROT_NORMAL] = 0, 691 692 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI | 693 SCSI_PROT_GUARD_CHECK | 694 SCSI_PROT_REF_CHECK | 695 SCSI_PROT_REF_INCREMENT, 696 697 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT | 698 SCSI_PROT_IP_CHECKSUM, 699 700 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI | 701 SCSI_PROT_GUARD_CHECK | 702 SCSI_PROT_REF_CHECK | 703 SCSI_PROT_REF_INCREMENT | 704 SCSI_PROT_IP_CHECKSUM, 705 706 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI | 707 SCSI_PROT_REF_INCREMENT, 708 709 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK | 710 SCSI_PROT_REF_CHECK | 711 SCSI_PROT_REF_INCREMENT | 712 SCSI_PROT_IP_CHECKSUM, 713 714 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI | 715 SCSI_PROT_GUARD_CHECK | 716 SCSI_PROT_REF_CHECK | 717 SCSI_PROT_REF_INCREMENT | 718 SCSI_PROT_IP_CHECKSUM, 719 }; 720 721 return flag_mask[prot_op]; 722 } 723 724 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd, 725 unsigned int dix, unsigned int dif) 726 { 727 struct bio *bio = scmd->request->bio; 728 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif); 729 unsigned int protect = 0; 730 731 if (dix) { /* DIX Type 0, 1, 2, 3 */ 732 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM)) 733 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM; 734 735 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false) 736 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK; 737 } 738 739 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */ 740 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT; 741 742 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false) 743 scmd->prot_flags |= SCSI_PROT_REF_CHECK; 744 } 745 746 if (dif) { /* DIX/DIF Type 1, 2, 3 */ 747 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI; 748 749 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK)) 750 protect = 3 << 5; /* Disable target PI checking */ 751 else 752 protect = 1 << 5; /* Enable target PI checking */ 753 } 754 755 scsi_set_prot_op(scmd, prot_op); 756 scsi_set_prot_type(scmd, dif); 757 scmd->prot_flags &= sd_prot_flag_mask(prot_op); 758 759 return protect; 760 } 761 762 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode) 763 { 764 struct request_queue *q = sdkp->disk->queue; 765 unsigned int logical_block_size = sdkp->device->sector_size; 766 unsigned int max_blocks = 0; 767 768 q->limits.discard_alignment = 769 sdkp->unmap_alignment * logical_block_size; 770 q->limits.discard_granularity = 771 max(sdkp->physical_block_size, 772 sdkp->unmap_granularity * logical_block_size); 773 sdkp->provisioning_mode = mode; 774 775 switch (mode) { 776 777 case SD_LBP_FULL: 778 case SD_LBP_DISABLE: 779 blk_queue_max_discard_sectors(q, 0); 780 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 781 return; 782 783 case SD_LBP_UNMAP: 784 max_blocks = min_not_zero(sdkp->max_unmap_blocks, 785 (u32)SD_MAX_WS16_BLOCKS); 786 break; 787 788 case SD_LBP_WS16: 789 if (sdkp->device->unmap_limit_for_ws) 790 max_blocks = sdkp->max_unmap_blocks; 791 else 792 max_blocks = sdkp->max_ws_blocks; 793 794 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS); 795 break; 796 797 case SD_LBP_WS10: 798 if (sdkp->device->unmap_limit_for_ws) 799 max_blocks = sdkp->max_unmap_blocks; 800 else 801 max_blocks = sdkp->max_ws_blocks; 802 803 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS); 804 break; 805 806 case SD_LBP_ZERO: 807 max_blocks = min_not_zero(sdkp->max_ws_blocks, 808 (u32)SD_MAX_WS10_BLOCKS); 809 break; 810 } 811 812 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9)); 813 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 814 } 815 816 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd) 817 { 818 struct scsi_device *sdp = cmd->device; 819 struct request *rq = cmd->request; 820 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 821 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 822 unsigned int data_len = 24; 823 char *buf; 824 825 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC); 826 if (!rq->special_vec.bv_page) 827 return BLK_STS_RESOURCE; 828 clear_highpage(rq->special_vec.bv_page); 829 rq->special_vec.bv_offset = 0; 830 rq->special_vec.bv_len = data_len; 831 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 832 833 cmd->cmd_len = 10; 834 cmd->cmnd[0] = UNMAP; 835 cmd->cmnd[8] = 24; 836 837 buf = page_address(rq->special_vec.bv_page); 838 put_unaligned_be16(6 + 16, &buf[0]); 839 put_unaligned_be16(16, &buf[2]); 840 put_unaligned_be64(lba, &buf[8]); 841 put_unaligned_be32(nr_blocks, &buf[16]); 842 843 cmd->allowed = SD_MAX_RETRIES; 844 cmd->transfersize = data_len; 845 rq->timeout = SD_TIMEOUT; 846 scsi_req(rq)->resid_len = data_len; 847 848 return scsi_init_io(cmd); 849 } 850 851 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, 852 bool unmap) 853 { 854 struct scsi_device *sdp = cmd->device; 855 struct request *rq = cmd->request; 856 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 857 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 858 u32 data_len = sdp->sector_size; 859 860 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC); 861 if (!rq->special_vec.bv_page) 862 return BLK_STS_RESOURCE; 863 clear_highpage(rq->special_vec.bv_page); 864 rq->special_vec.bv_offset = 0; 865 rq->special_vec.bv_len = data_len; 866 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 867 868 cmd->cmd_len = 16; 869 cmd->cmnd[0] = WRITE_SAME_16; 870 if (unmap) 871 cmd->cmnd[1] = 0x8; /* UNMAP */ 872 put_unaligned_be64(lba, &cmd->cmnd[2]); 873 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]); 874 875 cmd->allowed = SD_MAX_RETRIES; 876 cmd->transfersize = data_len; 877 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT; 878 scsi_req(rq)->resid_len = data_len; 879 880 return scsi_init_io(cmd); 881 } 882 883 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, 884 bool unmap) 885 { 886 struct scsi_device *sdp = cmd->device; 887 struct request *rq = cmd->request; 888 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 889 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 890 u32 data_len = sdp->sector_size; 891 892 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC); 893 if (!rq->special_vec.bv_page) 894 return BLK_STS_RESOURCE; 895 clear_highpage(rq->special_vec.bv_page); 896 rq->special_vec.bv_offset = 0; 897 rq->special_vec.bv_len = data_len; 898 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 899 900 cmd->cmd_len = 10; 901 cmd->cmnd[0] = WRITE_SAME; 902 if (unmap) 903 cmd->cmnd[1] = 0x8; /* UNMAP */ 904 put_unaligned_be32(lba, &cmd->cmnd[2]); 905 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]); 906 907 cmd->allowed = SD_MAX_RETRIES; 908 cmd->transfersize = data_len; 909 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT; 910 scsi_req(rq)->resid_len = data_len; 911 912 return scsi_init_io(cmd); 913 } 914 915 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd) 916 { 917 struct request *rq = cmd->request; 918 struct scsi_device *sdp = cmd->device; 919 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk); 920 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 921 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 922 923 if (!(rq->cmd_flags & REQ_NOUNMAP)) { 924 switch (sdkp->zeroing_mode) { 925 case SD_ZERO_WS16_UNMAP: 926 return sd_setup_write_same16_cmnd(cmd, true); 927 case SD_ZERO_WS10_UNMAP: 928 return sd_setup_write_same10_cmnd(cmd, true); 929 } 930 } 931 932 if (sdp->no_write_same) 933 return BLK_STS_TARGET; 934 935 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) 936 return sd_setup_write_same16_cmnd(cmd, false); 937 938 return sd_setup_write_same10_cmnd(cmd, false); 939 } 940 941 static void sd_config_write_same(struct scsi_disk *sdkp) 942 { 943 struct request_queue *q = sdkp->disk->queue; 944 unsigned int logical_block_size = sdkp->device->sector_size; 945 946 if (sdkp->device->no_write_same) { 947 sdkp->max_ws_blocks = 0; 948 goto out; 949 } 950 951 /* Some devices can not handle block counts above 0xffff despite 952 * supporting WRITE SAME(16). Consequently we default to 64k 953 * blocks per I/O unless the device explicitly advertises a 954 * bigger limit. 955 */ 956 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS) 957 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks, 958 (u32)SD_MAX_WS16_BLOCKS); 959 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes) 960 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks, 961 (u32)SD_MAX_WS10_BLOCKS); 962 else { 963 sdkp->device->no_write_same = 1; 964 sdkp->max_ws_blocks = 0; 965 } 966 967 if (sdkp->lbprz && sdkp->lbpws) 968 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP; 969 else if (sdkp->lbprz && sdkp->lbpws10) 970 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP; 971 else if (sdkp->max_ws_blocks) 972 sdkp->zeroing_mode = SD_ZERO_WS; 973 else 974 sdkp->zeroing_mode = SD_ZERO_WRITE; 975 976 if (sdkp->max_ws_blocks && 977 sdkp->physical_block_size > logical_block_size) { 978 /* 979 * Reporting a maximum number of blocks that is not aligned 980 * on the device physical size would cause a large write same 981 * request to be split into physically unaligned chunks by 982 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same() 983 * even if the caller of these functions took care to align the 984 * large request. So make sure the maximum reported is aligned 985 * to the device physical block size. This is only an optional 986 * optimization for regular disks, but this is mandatory to 987 * avoid failure of large write same requests directed at 988 * sequential write required zones of host-managed ZBC disks. 989 */ 990 sdkp->max_ws_blocks = 991 round_down(sdkp->max_ws_blocks, 992 bytes_to_logical(sdkp->device, 993 sdkp->physical_block_size)); 994 } 995 996 out: 997 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks * 998 (logical_block_size >> 9)); 999 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks * 1000 (logical_block_size >> 9)); 1001 } 1002 1003 /** 1004 * sd_setup_write_same_cmnd - write the same data to multiple blocks 1005 * @cmd: command to prepare 1006 * 1007 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on 1008 * the preference indicated by the target device. 1009 **/ 1010 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd) 1011 { 1012 struct request *rq = cmd->request; 1013 struct scsi_device *sdp = cmd->device; 1014 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk); 1015 struct bio *bio = rq->bio; 1016 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 1017 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 1018 blk_status_t ret; 1019 1020 if (sdkp->device->no_write_same) 1021 return BLK_STS_TARGET; 1022 1023 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size); 1024 1025 rq->timeout = SD_WRITE_SAME_TIMEOUT; 1026 1027 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) { 1028 cmd->cmd_len = 16; 1029 cmd->cmnd[0] = WRITE_SAME_16; 1030 put_unaligned_be64(lba, &cmd->cmnd[2]); 1031 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]); 1032 } else { 1033 cmd->cmd_len = 10; 1034 cmd->cmnd[0] = WRITE_SAME; 1035 put_unaligned_be32(lba, &cmd->cmnd[2]); 1036 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]); 1037 } 1038 1039 cmd->transfersize = sdp->sector_size; 1040 cmd->allowed = SD_MAX_RETRIES; 1041 1042 /* 1043 * For WRITE SAME the data transferred via the DATA OUT buffer is 1044 * different from the amount of data actually written to the target. 1045 * 1046 * We set up __data_len to the amount of data transferred via the 1047 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list 1048 * to transfer a single sector of data first, but then reset it to 1049 * the amount of data to be written right after so that the I/O path 1050 * knows how much to actually write. 1051 */ 1052 rq->__data_len = sdp->sector_size; 1053 ret = scsi_init_io(cmd); 1054 rq->__data_len = blk_rq_bytes(rq); 1055 1056 return ret; 1057 } 1058 1059 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd) 1060 { 1061 struct request *rq = cmd->request; 1062 1063 /* flush requests don't perform I/O, zero the S/G table */ 1064 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1065 1066 cmd->cmnd[0] = SYNCHRONIZE_CACHE; 1067 cmd->cmd_len = 10; 1068 cmd->transfersize = 0; 1069 cmd->allowed = SD_MAX_RETRIES; 1070 1071 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER; 1072 return BLK_STS_OK; 1073 } 1074 1075 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write, 1076 sector_t lba, unsigned int nr_blocks, 1077 unsigned char flags) 1078 { 1079 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC); 1080 if (unlikely(cmd->cmnd == NULL)) 1081 return BLK_STS_RESOURCE; 1082 1083 cmd->cmd_len = SD_EXT_CDB_SIZE; 1084 memset(cmd->cmnd, 0, cmd->cmd_len); 1085 1086 cmd->cmnd[0] = VARIABLE_LENGTH_CMD; 1087 cmd->cmnd[7] = 0x18; /* Additional CDB len */ 1088 cmd->cmnd[9] = write ? WRITE_32 : READ_32; 1089 cmd->cmnd[10] = flags; 1090 put_unaligned_be64(lba, &cmd->cmnd[12]); 1091 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */ 1092 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]); 1093 1094 return BLK_STS_OK; 1095 } 1096 1097 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write, 1098 sector_t lba, unsigned int nr_blocks, 1099 unsigned char flags) 1100 { 1101 cmd->cmd_len = 16; 1102 cmd->cmnd[0] = write ? WRITE_16 : READ_16; 1103 cmd->cmnd[1] = flags; 1104 cmd->cmnd[14] = 0; 1105 cmd->cmnd[15] = 0; 1106 put_unaligned_be64(lba, &cmd->cmnd[2]); 1107 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]); 1108 1109 return BLK_STS_OK; 1110 } 1111 1112 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write, 1113 sector_t lba, unsigned int nr_blocks, 1114 unsigned char flags) 1115 { 1116 cmd->cmd_len = 10; 1117 cmd->cmnd[0] = write ? WRITE_10 : READ_10; 1118 cmd->cmnd[1] = flags; 1119 cmd->cmnd[6] = 0; 1120 cmd->cmnd[9] = 0; 1121 put_unaligned_be32(lba, &cmd->cmnd[2]); 1122 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]); 1123 1124 return BLK_STS_OK; 1125 } 1126 1127 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write, 1128 sector_t lba, unsigned int nr_blocks, 1129 unsigned char flags) 1130 { 1131 /* Avoid that 0 blocks gets translated into 256 blocks. */ 1132 if (WARN_ON_ONCE(nr_blocks == 0)) 1133 return BLK_STS_IOERR; 1134 1135 if (unlikely(flags & 0x8)) { 1136 /* 1137 * This happens only if this drive failed 10byte rw 1138 * command with ILLEGAL_REQUEST during operation and 1139 * thus turned off use_10_for_rw. 1140 */ 1141 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n"); 1142 return BLK_STS_IOERR; 1143 } 1144 1145 cmd->cmd_len = 6; 1146 cmd->cmnd[0] = write ? WRITE_6 : READ_6; 1147 cmd->cmnd[1] = (lba >> 16) & 0x1f; 1148 cmd->cmnd[2] = (lba >> 8) & 0xff; 1149 cmd->cmnd[3] = lba & 0xff; 1150 cmd->cmnd[4] = nr_blocks; 1151 cmd->cmnd[5] = 0; 1152 1153 return BLK_STS_OK; 1154 } 1155 1156 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd) 1157 { 1158 struct request *rq = cmd->request; 1159 struct scsi_device *sdp = cmd->device; 1160 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk); 1161 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq)); 1162 sector_t threshold; 1163 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq)); 1164 bool dif, dix; 1165 unsigned int mask = logical_to_sectors(sdp, 1) - 1; 1166 bool write = rq_data_dir(rq) == WRITE; 1167 unsigned char protect, fua; 1168 blk_status_t ret; 1169 1170 ret = scsi_init_io(cmd); 1171 if (ret != BLK_STS_OK) 1172 return ret; 1173 1174 WARN_ON_ONCE(cmd != rq->special); 1175 1176 if (!scsi_device_online(sdp) || sdp->changed) { 1177 scmd_printk(KERN_ERR, cmd, "device offline or changed\n"); 1178 return BLK_STS_IOERR; 1179 } 1180 1181 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) { 1182 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n"); 1183 return BLK_STS_IOERR; 1184 } 1185 1186 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) { 1187 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n"); 1188 return BLK_STS_IOERR; 1189 } 1190 1191 /* 1192 * Some SD card readers can't handle accesses which touch the 1193 * last one or two logical blocks. Split accesses as needed. 1194 */ 1195 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS; 1196 1197 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) { 1198 if (lba < threshold) { 1199 /* Access up to the threshold but not beyond */ 1200 nr_blocks = threshold - lba; 1201 } else { 1202 /* Access only a single logical block */ 1203 nr_blocks = 1; 1204 } 1205 } 1206 1207 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0; 1208 dix = scsi_prot_sg_count(cmd); 1209 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type); 1210 1211 if (write && dix) 1212 t10_pi_prepare(cmd->request, sdkp->protection_type); 1213 1214 if (dif || dix) 1215 protect = sd_setup_protect_cmnd(cmd, dix, dif); 1216 else 1217 protect = 0; 1218 1219 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) { 1220 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks, 1221 protect | fua); 1222 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) { 1223 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks, 1224 protect | fua); 1225 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) || 1226 sdp->use_10_for_rw || protect) { 1227 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks, 1228 protect | fua); 1229 } else { 1230 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks, 1231 protect | fua); 1232 } 1233 1234 if (unlikely(ret != BLK_STS_OK)) 1235 return ret; 1236 1237 /* 1238 * We shouldn't disconnect in the middle of a sector, so with a dumb 1239 * host adapter, it's safe to assume that we can at least transfer 1240 * this many bytes between each connect / disconnect. 1241 */ 1242 cmd->transfersize = sdp->sector_size; 1243 cmd->underflow = nr_blocks << 9; 1244 cmd->allowed = SD_MAX_RETRIES; 1245 cmd->sdb.length = nr_blocks * sdp->sector_size; 1246 1247 SCSI_LOG_HLQUEUE(1, 1248 scmd_printk(KERN_INFO, cmd, 1249 "%s: block=%llu, count=%d\n", __func__, 1250 (unsigned long long)blk_rq_pos(rq), 1251 blk_rq_sectors(rq))); 1252 SCSI_LOG_HLQUEUE(2, 1253 scmd_printk(KERN_INFO, cmd, 1254 "%s %d/%u 512 byte blocks.\n", 1255 write ? "writing" : "reading", nr_blocks, 1256 blk_rq_sectors(rq))); 1257 1258 /* 1259 * This indicates that the command is ready from our end to be 1260 * queued. 1261 */ 1262 return BLK_STS_OK; 1263 } 1264 1265 static blk_status_t sd_init_command(struct scsi_cmnd *cmd) 1266 { 1267 struct request *rq = cmd->request; 1268 1269 switch (req_op(rq)) { 1270 case REQ_OP_DISCARD: 1271 switch (scsi_disk(rq->rq_disk)->provisioning_mode) { 1272 case SD_LBP_UNMAP: 1273 return sd_setup_unmap_cmnd(cmd); 1274 case SD_LBP_WS16: 1275 return sd_setup_write_same16_cmnd(cmd, true); 1276 case SD_LBP_WS10: 1277 return sd_setup_write_same10_cmnd(cmd, true); 1278 case SD_LBP_ZERO: 1279 return sd_setup_write_same10_cmnd(cmd, false); 1280 default: 1281 return BLK_STS_TARGET; 1282 } 1283 case REQ_OP_WRITE_ZEROES: 1284 return sd_setup_write_zeroes_cmnd(cmd); 1285 case REQ_OP_WRITE_SAME: 1286 return sd_setup_write_same_cmnd(cmd); 1287 case REQ_OP_FLUSH: 1288 return sd_setup_flush_cmnd(cmd); 1289 case REQ_OP_READ: 1290 case REQ_OP_WRITE: 1291 return sd_setup_read_write_cmnd(cmd); 1292 case REQ_OP_ZONE_RESET: 1293 return sd_zbc_setup_reset_cmnd(cmd); 1294 default: 1295 WARN_ON_ONCE(1); 1296 return BLK_STS_NOTSUPP; 1297 } 1298 } 1299 1300 static void sd_uninit_command(struct scsi_cmnd *SCpnt) 1301 { 1302 struct request *rq = SCpnt->request; 1303 u8 *cmnd; 1304 1305 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1306 mempool_free(rq->special_vec.bv_page, sd_page_pool); 1307 1308 if (SCpnt->cmnd != scsi_req(rq)->cmd) { 1309 cmnd = SCpnt->cmnd; 1310 SCpnt->cmnd = NULL; 1311 SCpnt->cmd_len = 0; 1312 mempool_free(cmnd, sd_cdb_pool); 1313 } 1314 } 1315 1316 /** 1317 * sd_open - open a scsi disk device 1318 * @bdev: Block device of the scsi disk to open 1319 * @mode: FMODE_* mask 1320 * 1321 * Returns 0 if successful. Returns a negated errno value in case 1322 * of error. 1323 * 1324 * Note: This can be called from a user context (e.g. fsck(1) ) 1325 * or from within the kernel (e.g. as a result of a mount(1) ). 1326 * In the latter case @inode and @filp carry an abridged amount 1327 * of information as noted above. 1328 * 1329 * Locking: called with bdev->bd_mutex held. 1330 **/ 1331 static int sd_open(struct block_device *bdev, fmode_t mode) 1332 { 1333 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk); 1334 struct scsi_device *sdev; 1335 int retval; 1336 1337 if (!sdkp) 1338 return -ENXIO; 1339 1340 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n")); 1341 1342 sdev = sdkp->device; 1343 1344 /* 1345 * If the device is in error recovery, wait until it is done. 1346 * If the device is offline, then disallow any access to it. 1347 */ 1348 retval = -ENXIO; 1349 if (!scsi_block_when_processing_errors(sdev)) 1350 goto error_out; 1351 1352 if (sdev->removable || sdkp->write_prot) 1353 check_disk_change(bdev); 1354 1355 /* 1356 * If the drive is empty, just let the open fail. 1357 */ 1358 retval = -ENOMEDIUM; 1359 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY)) 1360 goto error_out; 1361 1362 /* 1363 * If the device has the write protect tab set, have the open fail 1364 * if the user expects to be able to write to the thing. 1365 */ 1366 retval = -EROFS; 1367 if (sdkp->write_prot && (mode & FMODE_WRITE)) 1368 goto error_out; 1369 1370 /* 1371 * It is possible that the disk changing stuff resulted in 1372 * the device being taken offline. If this is the case, 1373 * report this to the user, and don't pretend that the 1374 * open actually succeeded. 1375 */ 1376 retval = -ENXIO; 1377 if (!scsi_device_online(sdev)) 1378 goto error_out; 1379 1380 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) { 1381 if (scsi_block_when_processing_errors(sdev)) 1382 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT); 1383 } 1384 1385 return 0; 1386 1387 error_out: 1388 scsi_disk_put(sdkp); 1389 return retval; 1390 } 1391 1392 /** 1393 * sd_release - invoked when the (last) close(2) is called on this 1394 * scsi disk. 1395 * @disk: disk to release 1396 * @mode: FMODE_* mask 1397 * 1398 * Returns 0. 1399 * 1400 * Note: may block (uninterruptible) if error recovery is underway 1401 * on this disk. 1402 * 1403 * Locking: called with bdev->bd_mutex held. 1404 **/ 1405 static void sd_release(struct gendisk *disk, fmode_t mode) 1406 { 1407 struct scsi_disk *sdkp = scsi_disk(disk); 1408 struct scsi_device *sdev = sdkp->device; 1409 1410 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n")); 1411 1412 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) { 1413 if (scsi_block_when_processing_errors(sdev)) 1414 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW); 1415 } 1416 1417 /* 1418 * XXX and what if there are packets in flight and this close() 1419 * XXX is followed by a "rmmod sd_mod"? 1420 */ 1421 1422 scsi_disk_put(sdkp); 1423 } 1424 1425 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo) 1426 { 1427 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk); 1428 struct scsi_device *sdp = sdkp->device; 1429 struct Scsi_Host *host = sdp->host; 1430 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity); 1431 int diskinfo[4]; 1432 1433 /* default to most commonly used values */ 1434 diskinfo[0] = 0x40; /* 1 << 6 */ 1435 diskinfo[1] = 0x20; /* 1 << 5 */ 1436 diskinfo[2] = capacity >> 11; 1437 1438 /* override with calculated, extended default, or driver values */ 1439 if (host->hostt->bios_param) 1440 host->hostt->bios_param(sdp, bdev, capacity, diskinfo); 1441 else 1442 scsicam_bios_param(bdev, capacity, diskinfo); 1443 1444 geo->heads = diskinfo[0]; 1445 geo->sectors = diskinfo[1]; 1446 geo->cylinders = diskinfo[2]; 1447 return 0; 1448 } 1449 1450 /** 1451 * sd_ioctl - process an ioctl 1452 * @bdev: target block device 1453 * @mode: FMODE_* mask 1454 * @cmd: ioctl command number 1455 * @arg: this is third argument given to ioctl(2) system call. 1456 * Often contains a pointer. 1457 * 1458 * Returns 0 if successful (some ioctls return positive numbers on 1459 * success as well). Returns a negated errno value in case of error. 1460 * 1461 * Note: most ioctls are forward onto the block subsystem or further 1462 * down in the scsi subsystem. 1463 **/ 1464 static int sd_ioctl(struct block_device *bdev, fmode_t mode, 1465 unsigned int cmd, unsigned long arg) 1466 { 1467 struct gendisk *disk = bdev->bd_disk; 1468 struct scsi_disk *sdkp = scsi_disk(disk); 1469 struct scsi_device *sdp = sdkp->device; 1470 void __user *p = (void __user *)arg; 1471 int error; 1472 1473 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, " 1474 "cmd=0x%x\n", disk->disk_name, cmd)); 1475 1476 error = scsi_verify_blk_ioctl(bdev, cmd); 1477 if (error < 0) 1478 return error; 1479 1480 /* 1481 * If we are in the middle of error recovery, don't let anyone 1482 * else try and use this device. Also, if error recovery fails, it 1483 * may try and take the device offline, in which case all further 1484 * access to the device is prohibited. 1485 */ 1486 error = scsi_ioctl_block_when_processing_errors(sdp, cmd, 1487 (mode & FMODE_NDELAY) != 0); 1488 if (error) 1489 goto out; 1490 1491 if (is_sed_ioctl(cmd)) 1492 return sed_ioctl(sdkp->opal_dev, cmd, p); 1493 1494 /* 1495 * Send SCSI addressing ioctls directly to mid level, send other 1496 * ioctls to block level and then onto mid level if they can't be 1497 * resolved. 1498 */ 1499 switch (cmd) { 1500 case SCSI_IOCTL_GET_IDLUN: 1501 case SCSI_IOCTL_GET_BUS_NUMBER: 1502 error = scsi_ioctl(sdp, cmd, p); 1503 break; 1504 default: 1505 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p); 1506 if (error != -ENOTTY) 1507 break; 1508 error = scsi_ioctl(sdp, cmd, p); 1509 break; 1510 } 1511 out: 1512 return error; 1513 } 1514 1515 static void set_media_not_present(struct scsi_disk *sdkp) 1516 { 1517 if (sdkp->media_present) 1518 sdkp->device->changed = 1; 1519 1520 if (sdkp->device->removable) { 1521 sdkp->media_present = 0; 1522 sdkp->capacity = 0; 1523 } 1524 } 1525 1526 static int media_not_present(struct scsi_disk *sdkp, 1527 struct scsi_sense_hdr *sshdr) 1528 { 1529 if (!scsi_sense_valid(sshdr)) 1530 return 0; 1531 1532 /* not invoked for commands that could return deferred errors */ 1533 switch (sshdr->sense_key) { 1534 case UNIT_ATTENTION: 1535 case NOT_READY: 1536 /* medium not present */ 1537 if (sshdr->asc == 0x3A) { 1538 set_media_not_present(sdkp); 1539 return 1; 1540 } 1541 } 1542 return 0; 1543 } 1544 1545 /** 1546 * sd_check_events - check media events 1547 * @disk: kernel device descriptor 1548 * @clearing: disk events currently being cleared 1549 * 1550 * Returns mask of DISK_EVENT_*. 1551 * 1552 * Note: this function is invoked from the block subsystem. 1553 **/ 1554 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing) 1555 { 1556 struct scsi_disk *sdkp = scsi_disk_get(disk); 1557 struct scsi_device *sdp; 1558 int retval; 1559 1560 if (!sdkp) 1561 return 0; 1562 1563 sdp = sdkp->device; 1564 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n")); 1565 1566 /* 1567 * If the device is offline, don't send any commands - just pretend as 1568 * if the command failed. If the device ever comes back online, we 1569 * can deal with it then. It is only because of unrecoverable errors 1570 * that we would ever take a device offline in the first place. 1571 */ 1572 if (!scsi_device_online(sdp)) { 1573 set_media_not_present(sdkp); 1574 goto out; 1575 } 1576 1577 /* 1578 * Using TEST_UNIT_READY enables differentiation between drive with 1579 * no cartridge loaded - NOT READY, drive with changed cartridge - 1580 * UNIT ATTENTION, or with same cartridge - GOOD STATUS. 1581 * 1582 * Drives that auto spin down. eg iomega jaz 1G, will be started 1583 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever 1584 * sd_revalidate() is called. 1585 */ 1586 if (scsi_block_when_processing_errors(sdp)) { 1587 struct scsi_sense_hdr sshdr = { 0, }; 1588 1589 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES, 1590 &sshdr); 1591 1592 /* failed to execute TUR, assume media not present */ 1593 if (host_byte(retval)) { 1594 set_media_not_present(sdkp); 1595 goto out; 1596 } 1597 1598 if (media_not_present(sdkp, &sshdr)) 1599 goto out; 1600 } 1601 1602 /* 1603 * For removable scsi disk we have to recognise the presence 1604 * of a disk in the drive. 1605 */ 1606 if (!sdkp->media_present) 1607 sdp->changed = 1; 1608 sdkp->media_present = 1; 1609 out: 1610 /* 1611 * sdp->changed is set under the following conditions: 1612 * 1613 * Medium present state has changed in either direction. 1614 * Device has indicated UNIT_ATTENTION. 1615 */ 1616 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0; 1617 sdp->changed = 0; 1618 scsi_disk_put(sdkp); 1619 return retval; 1620 } 1621 1622 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr) 1623 { 1624 int retries, res; 1625 struct scsi_device *sdp = sdkp->device; 1626 const int timeout = sdp->request_queue->rq_timeout 1627 * SD_FLUSH_TIMEOUT_MULTIPLIER; 1628 struct scsi_sense_hdr my_sshdr; 1629 1630 if (!scsi_device_online(sdp)) 1631 return -ENODEV; 1632 1633 /* caller might not be interested in sense, but we need it */ 1634 if (!sshdr) 1635 sshdr = &my_sshdr; 1636 1637 for (retries = 3; retries > 0; --retries) { 1638 unsigned char cmd[10] = { 0 }; 1639 1640 cmd[0] = SYNCHRONIZE_CACHE; 1641 /* 1642 * Leave the rest of the command zero to indicate 1643 * flush everything. 1644 */ 1645 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr, 1646 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL); 1647 if (res == 0) 1648 break; 1649 } 1650 1651 if (res) { 1652 sd_print_result(sdkp, "Synchronize Cache(10) failed", res); 1653 1654 if (driver_byte(res) == DRIVER_SENSE) 1655 sd_print_sense_hdr(sdkp, sshdr); 1656 1657 /* we need to evaluate the error return */ 1658 if (scsi_sense_valid(sshdr) && 1659 (sshdr->asc == 0x3a || /* medium not present */ 1660 sshdr->asc == 0x20)) /* invalid command */ 1661 /* this is no error here */ 1662 return 0; 1663 1664 switch (host_byte(res)) { 1665 /* ignore errors due to racing a disconnection */ 1666 case DID_BAD_TARGET: 1667 case DID_NO_CONNECT: 1668 return 0; 1669 /* signal the upper layer it might try again */ 1670 case DID_BUS_BUSY: 1671 case DID_IMM_RETRY: 1672 case DID_REQUEUE: 1673 case DID_SOFT_ERROR: 1674 return -EBUSY; 1675 default: 1676 return -EIO; 1677 } 1678 } 1679 return 0; 1680 } 1681 1682 static void sd_rescan(struct device *dev) 1683 { 1684 struct scsi_disk *sdkp = dev_get_drvdata(dev); 1685 1686 revalidate_disk(sdkp->disk); 1687 } 1688 1689 1690 #ifdef CONFIG_COMPAT 1691 /* 1692 * This gets directly called from VFS. When the ioctl 1693 * is not recognized we go back to the other translation paths. 1694 */ 1695 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode, 1696 unsigned int cmd, unsigned long arg) 1697 { 1698 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device; 1699 int error; 1700 1701 error = scsi_ioctl_block_when_processing_errors(sdev, cmd, 1702 (mode & FMODE_NDELAY) != 0); 1703 if (error) 1704 return error; 1705 1706 /* 1707 * Let the static ioctl translation table take care of it. 1708 */ 1709 if (!sdev->host->hostt->compat_ioctl) 1710 return -ENOIOCTLCMD; 1711 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg); 1712 } 1713 #endif 1714 1715 static char sd_pr_type(enum pr_type type) 1716 { 1717 switch (type) { 1718 case PR_WRITE_EXCLUSIVE: 1719 return 0x01; 1720 case PR_EXCLUSIVE_ACCESS: 1721 return 0x03; 1722 case PR_WRITE_EXCLUSIVE_REG_ONLY: 1723 return 0x05; 1724 case PR_EXCLUSIVE_ACCESS_REG_ONLY: 1725 return 0x06; 1726 case PR_WRITE_EXCLUSIVE_ALL_REGS: 1727 return 0x07; 1728 case PR_EXCLUSIVE_ACCESS_ALL_REGS: 1729 return 0x08; 1730 default: 1731 return 0; 1732 } 1733 }; 1734 1735 static int sd_pr_command(struct block_device *bdev, u8 sa, 1736 u64 key, u64 sa_key, u8 type, u8 flags) 1737 { 1738 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device; 1739 struct scsi_sense_hdr sshdr; 1740 int result; 1741 u8 cmd[16] = { 0, }; 1742 u8 data[24] = { 0, }; 1743 1744 cmd[0] = PERSISTENT_RESERVE_OUT; 1745 cmd[1] = sa; 1746 cmd[2] = type; 1747 put_unaligned_be32(sizeof(data), &cmd[5]); 1748 1749 put_unaligned_be64(key, &data[0]); 1750 put_unaligned_be64(sa_key, &data[8]); 1751 data[20] = flags; 1752 1753 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data), 1754 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL); 1755 1756 if (driver_byte(result) == DRIVER_SENSE && 1757 scsi_sense_valid(&sshdr)) { 1758 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result); 1759 scsi_print_sense_hdr(sdev, NULL, &sshdr); 1760 } 1761 1762 return result; 1763 } 1764 1765 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, 1766 u32 flags) 1767 { 1768 if (flags & ~PR_FL_IGNORE_KEY) 1769 return -EOPNOTSUPP; 1770 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00, 1771 old_key, new_key, 0, 1772 (1 << 0) /* APTPL */); 1773 } 1774 1775 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, 1776 u32 flags) 1777 { 1778 if (flags) 1779 return -EOPNOTSUPP; 1780 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0); 1781 } 1782 1783 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type) 1784 { 1785 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0); 1786 } 1787 1788 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, 1789 enum pr_type type, bool abort) 1790 { 1791 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key, 1792 sd_pr_type(type), 0); 1793 } 1794 1795 static int sd_pr_clear(struct block_device *bdev, u64 key) 1796 { 1797 return sd_pr_command(bdev, 0x03, key, 0, 0, 0); 1798 } 1799 1800 static const struct pr_ops sd_pr_ops = { 1801 .pr_register = sd_pr_register, 1802 .pr_reserve = sd_pr_reserve, 1803 .pr_release = sd_pr_release, 1804 .pr_preempt = sd_pr_preempt, 1805 .pr_clear = sd_pr_clear, 1806 }; 1807 1808 static const struct block_device_operations sd_fops = { 1809 .owner = THIS_MODULE, 1810 .open = sd_open, 1811 .release = sd_release, 1812 .ioctl = sd_ioctl, 1813 .getgeo = sd_getgeo, 1814 #ifdef CONFIG_COMPAT 1815 .compat_ioctl = sd_compat_ioctl, 1816 #endif 1817 .check_events = sd_check_events, 1818 .revalidate_disk = sd_revalidate_disk, 1819 .unlock_native_capacity = sd_unlock_native_capacity, 1820 .report_zones = sd_zbc_report_zones, 1821 .pr_ops = &sd_pr_ops, 1822 }; 1823 1824 /** 1825 * sd_eh_reset - reset error handling callback 1826 * @scmd: sd-issued command that has failed 1827 * 1828 * This function is called by the SCSI midlayer before starting 1829 * SCSI EH. When counting medium access failures we have to be 1830 * careful to register it only only once per device and SCSI EH run; 1831 * there might be several timed out commands which will cause the 1832 * 'max_medium_access_timeouts' counter to trigger after the first 1833 * SCSI EH run already and set the device to offline. 1834 * So this function resets the internal counter before starting SCSI EH. 1835 **/ 1836 static void sd_eh_reset(struct scsi_cmnd *scmd) 1837 { 1838 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk); 1839 1840 /* New SCSI EH run, reset gate variable */ 1841 sdkp->ignore_medium_access_errors = false; 1842 } 1843 1844 /** 1845 * sd_eh_action - error handling callback 1846 * @scmd: sd-issued command that has failed 1847 * @eh_disp: The recovery disposition suggested by the midlayer 1848 * 1849 * This function is called by the SCSI midlayer upon completion of an 1850 * error test command (currently TEST UNIT READY). The result of sending 1851 * the eh command is passed in eh_disp. We're looking for devices that 1852 * fail medium access commands but are OK with non access commands like 1853 * test unit ready (so wrongly see the device as having a successful 1854 * recovery) 1855 **/ 1856 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp) 1857 { 1858 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk); 1859 struct scsi_device *sdev = scmd->device; 1860 1861 if (!scsi_device_online(sdev) || 1862 !scsi_medium_access_command(scmd) || 1863 host_byte(scmd->result) != DID_TIME_OUT || 1864 eh_disp != SUCCESS) 1865 return eh_disp; 1866 1867 /* 1868 * The device has timed out executing a medium access command. 1869 * However, the TEST UNIT READY command sent during error 1870 * handling completed successfully. Either the device is in the 1871 * process of recovering or has it suffered an internal failure 1872 * that prevents access to the storage medium. 1873 */ 1874 if (!sdkp->ignore_medium_access_errors) { 1875 sdkp->medium_access_timed_out++; 1876 sdkp->ignore_medium_access_errors = true; 1877 } 1878 1879 /* 1880 * If the device keeps failing read/write commands but TEST UNIT 1881 * READY always completes successfully we assume that medium 1882 * access is no longer possible and take the device offline. 1883 */ 1884 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) { 1885 scmd_printk(KERN_ERR, scmd, 1886 "Medium access timeout failure. Offlining disk!\n"); 1887 mutex_lock(&sdev->state_mutex); 1888 scsi_device_set_state(sdev, SDEV_OFFLINE); 1889 mutex_unlock(&sdev->state_mutex); 1890 1891 return SUCCESS; 1892 } 1893 1894 return eh_disp; 1895 } 1896 1897 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd) 1898 { 1899 struct request *req = scmd->request; 1900 struct scsi_device *sdev = scmd->device; 1901 unsigned int transferred, good_bytes; 1902 u64 start_lba, end_lba, bad_lba; 1903 1904 /* 1905 * Some commands have a payload smaller than the device logical 1906 * block size (e.g. INQUIRY on a 4K disk). 1907 */ 1908 if (scsi_bufflen(scmd) <= sdev->sector_size) 1909 return 0; 1910 1911 /* Check if we have a 'bad_lba' information */ 1912 if (!scsi_get_sense_info_fld(scmd->sense_buffer, 1913 SCSI_SENSE_BUFFERSIZE, 1914 &bad_lba)) 1915 return 0; 1916 1917 /* 1918 * If the bad lba was reported incorrectly, we have no idea where 1919 * the error is. 1920 */ 1921 start_lba = sectors_to_logical(sdev, blk_rq_pos(req)); 1922 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd)); 1923 if (bad_lba < start_lba || bad_lba >= end_lba) 1924 return 0; 1925 1926 /* 1927 * resid is optional but mostly filled in. When it's unused, 1928 * its value is zero, so we assume the whole buffer transferred 1929 */ 1930 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd); 1931 1932 /* This computation should always be done in terms of the 1933 * resolution of the device's medium. 1934 */ 1935 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba); 1936 1937 return min(good_bytes, transferred); 1938 } 1939 1940 /** 1941 * sd_done - bottom half handler: called when the lower level 1942 * driver has completed (successfully or otherwise) a scsi command. 1943 * @SCpnt: mid-level's per command structure. 1944 * 1945 * Note: potentially run from within an ISR. Must not block. 1946 **/ 1947 static int sd_done(struct scsi_cmnd *SCpnt) 1948 { 1949 int result = SCpnt->result; 1950 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt); 1951 unsigned int sector_size = SCpnt->device->sector_size; 1952 unsigned int resid; 1953 struct scsi_sense_hdr sshdr; 1954 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk); 1955 struct request *req = SCpnt->request; 1956 int sense_valid = 0; 1957 int sense_deferred = 0; 1958 1959 switch (req_op(req)) { 1960 case REQ_OP_DISCARD: 1961 case REQ_OP_WRITE_ZEROES: 1962 case REQ_OP_WRITE_SAME: 1963 case REQ_OP_ZONE_RESET: 1964 if (!result) { 1965 good_bytes = blk_rq_bytes(req); 1966 scsi_set_resid(SCpnt, 0); 1967 } else { 1968 good_bytes = 0; 1969 scsi_set_resid(SCpnt, blk_rq_bytes(req)); 1970 } 1971 break; 1972 default: 1973 /* 1974 * In case of bogus fw or device, we could end up having 1975 * an unaligned partial completion. Check this here and force 1976 * alignment. 1977 */ 1978 resid = scsi_get_resid(SCpnt); 1979 if (resid & (sector_size - 1)) { 1980 sd_printk(KERN_INFO, sdkp, 1981 "Unaligned partial completion (resid=%u, sector_sz=%u)\n", 1982 resid, sector_size); 1983 resid = min(scsi_bufflen(SCpnt), 1984 round_up(resid, sector_size)); 1985 scsi_set_resid(SCpnt, resid); 1986 } 1987 } 1988 1989 if (result) { 1990 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr); 1991 if (sense_valid) 1992 sense_deferred = scsi_sense_is_deferred(&sshdr); 1993 } 1994 sdkp->medium_access_timed_out = 0; 1995 1996 if (driver_byte(result) != DRIVER_SENSE && 1997 (!sense_valid || sense_deferred)) 1998 goto out; 1999 2000 switch (sshdr.sense_key) { 2001 case HARDWARE_ERROR: 2002 case MEDIUM_ERROR: 2003 good_bytes = sd_completed_bytes(SCpnt); 2004 break; 2005 case RECOVERED_ERROR: 2006 good_bytes = scsi_bufflen(SCpnt); 2007 break; 2008 case NO_SENSE: 2009 /* This indicates a false check condition, so ignore it. An 2010 * unknown amount of data was transferred so treat it as an 2011 * error. 2012 */ 2013 SCpnt->result = 0; 2014 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 2015 break; 2016 case ABORTED_COMMAND: 2017 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */ 2018 good_bytes = sd_completed_bytes(SCpnt); 2019 break; 2020 case ILLEGAL_REQUEST: 2021 switch (sshdr.asc) { 2022 case 0x10: /* DIX: Host detected corruption */ 2023 good_bytes = sd_completed_bytes(SCpnt); 2024 break; 2025 case 0x20: /* INVALID COMMAND OPCODE */ 2026 case 0x24: /* INVALID FIELD IN CDB */ 2027 switch (SCpnt->cmnd[0]) { 2028 case UNMAP: 2029 sd_config_discard(sdkp, SD_LBP_DISABLE); 2030 break; 2031 case WRITE_SAME_16: 2032 case WRITE_SAME: 2033 if (SCpnt->cmnd[1] & 8) { /* UNMAP */ 2034 sd_config_discard(sdkp, SD_LBP_DISABLE); 2035 } else { 2036 sdkp->device->no_write_same = 1; 2037 sd_config_write_same(sdkp); 2038 req->rq_flags |= RQF_QUIET; 2039 } 2040 break; 2041 } 2042 } 2043 break; 2044 default: 2045 break; 2046 } 2047 2048 out: 2049 if (sd_is_zoned(sdkp)) 2050 sd_zbc_complete(SCpnt, good_bytes, &sshdr); 2051 2052 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt, 2053 "sd_done: completed %d of %d bytes\n", 2054 good_bytes, scsi_bufflen(SCpnt))); 2055 2056 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) && 2057 good_bytes) 2058 t10_pi_complete(SCpnt->request, sdkp->protection_type, 2059 good_bytes / scsi_prot_interval(SCpnt)); 2060 2061 return good_bytes; 2062 } 2063 2064 /* 2065 * spinup disk - called only in sd_revalidate_disk() 2066 */ 2067 static void 2068 sd_spinup_disk(struct scsi_disk *sdkp) 2069 { 2070 unsigned char cmd[10]; 2071 unsigned long spintime_expire = 0; 2072 int retries, spintime; 2073 unsigned int the_result; 2074 struct scsi_sense_hdr sshdr; 2075 int sense_valid = 0; 2076 2077 spintime = 0; 2078 2079 /* Spin up drives, as required. Only do this at boot time */ 2080 /* Spinup needs to be done for module loads too. */ 2081 do { 2082 retries = 0; 2083 2084 do { 2085 cmd[0] = TEST_UNIT_READY; 2086 memset((void *) &cmd[1], 0, 9); 2087 2088 the_result = scsi_execute_req(sdkp->device, cmd, 2089 DMA_NONE, NULL, 0, 2090 &sshdr, SD_TIMEOUT, 2091 SD_MAX_RETRIES, NULL); 2092 2093 /* 2094 * If the drive has indicated to us that it 2095 * doesn't have any media in it, don't bother 2096 * with any more polling. 2097 */ 2098 if (media_not_present(sdkp, &sshdr)) 2099 return; 2100 2101 if (the_result) 2102 sense_valid = scsi_sense_valid(&sshdr); 2103 retries++; 2104 } while (retries < 3 && 2105 (!scsi_status_is_good(the_result) || 2106 ((driver_byte(the_result) == DRIVER_SENSE) && 2107 sense_valid && sshdr.sense_key == UNIT_ATTENTION))); 2108 2109 if (driver_byte(the_result) != DRIVER_SENSE) { 2110 /* no sense, TUR either succeeded or failed 2111 * with a status error */ 2112 if(!spintime && !scsi_status_is_good(the_result)) { 2113 sd_print_result(sdkp, "Test Unit Ready failed", 2114 the_result); 2115 } 2116 break; 2117 } 2118 2119 /* 2120 * The device does not want the automatic start to be issued. 2121 */ 2122 if (sdkp->device->no_start_on_add) 2123 break; 2124 2125 if (sense_valid && sshdr.sense_key == NOT_READY) { 2126 if (sshdr.asc == 4 && sshdr.ascq == 3) 2127 break; /* manual intervention required */ 2128 if (sshdr.asc == 4 && sshdr.ascq == 0xb) 2129 break; /* standby */ 2130 if (sshdr.asc == 4 && sshdr.ascq == 0xc) 2131 break; /* unavailable */ 2132 if (sshdr.asc == 4 && sshdr.ascq == 0x1b) 2133 break; /* sanitize in progress */ 2134 /* 2135 * Issue command to spin up drive when not ready 2136 */ 2137 if (!spintime) { 2138 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk..."); 2139 cmd[0] = START_STOP; 2140 cmd[1] = 1; /* Return immediately */ 2141 memset((void *) &cmd[2], 0, 8); 2142 cmd[4] = 1; /* Start spin cycle */ 2143 if (sdkp->device->start_stop_pwr_cond) 2144 cmd[4] |= 1 << 4; 2145 scsi_execute_req(sdkp->device, cmd, DMA_NONE, 2146 NULL, 0, &sshdr, 2147 SD_TIMEOUT, SD_MAX_RETRIES, 2148 NULL); 2149 spintime_expire = jiffies + 100 * HZ; 2150 spintime = 1; 2151 } 2152 /* Wait 1 second for next try */ 2153 msleep(1000); 2154 printk(KERN_CONT "."); 2155 2156 /* 2157 * Wait for USB flash devices with slow firmware. 2158 * Yes, this sense key/ASC combination shouldn't 2159 * occur here. It's characteristic of these devices. 2160 */ 2161 } else if (sense_valid && 2162 sshdr.sense_key == UNIT_ATTENTION && 2163 sshdr.asc == 0x28) { 2164 if (!spintime) { 2165 spintime_expire = jiffies + 5 * HZ; 2166 spintime = 1; 2167 } 2168 /* Wait 1 second for next try */ 2169 msleep(1000); 2170 } else { 2171 /* we don't understand the sense code, so it's 2172 * probably pointless to loop */ 2173 if(!spintime) { 2174 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n"); 2175 sd_print_sense_hdr(sdkp, &sshdr); 2176 } 2177 break; 2178 } 2179 2180 } while (spintime && time_before_eq(jiffies, spintime_expire)); 2181 2182 if (spintime) { 2183 if (scsi_status_is_good(the_result)) 2184 printk(KERN_CONT "ready\n"); 2185 else 2186 printk(KERN_CONT "not responding...\n"); 2187 } 2188 } 2189 2190 /* 2191 * Determine whether disk supports Data Integrity Field. 2192 */ 2193 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer) 2194 { 2195 struct scsi_device *sdp = sdkp->device; 2196 u8 type; 2197 int ret = 0; 2198 2199 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) 2200 return ret; 2201 2202 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */ 2203 2204 if (type > T10_PI_TYPE3_PROTECTION) 2205 ret = -ENODEV; 2206 else if (scsi_host_dif_capable(sdp->host, type)) 2207 ret = 1; 2208 2209 if (sdkp->first_scan || type != sdkp->protection_type) 2210 switch (ret) { 2211 case -ENODEV: 2212 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \ 2213 " protection type %u. Disabling disk!\n", 2214 type); 2215 break; 2216 case 1: 2217 sd_printk(KERN_NOTICE, sdkp, 2218 "Enabling DIF Type %u protection\n", type); 2219 break; 2220 case 0: 2221 sd_printk(KERN_NOTICE, sdkp, 2222 "Disabling DIF Type %u protection\n", type); 2223 break; 2224 } 2225 2226 sdkp->protection_type = type; 2227 2228 return ret; 2229 } 2230 2231 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp, 2232 struct scsi_sense_hdr *sshdr, int sense_valid, 2233 int the_result) 2234 { 2235 if (driver_byte(the_result) == DRIVER_SENSE) 2236 sd_print_sense_hdr(sdkp, sshdr); 2237 else 2238 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n"); 2239 2240 /* 2241 * Set dirty bit for removable devices if not ready - 2242 * sometimes drives will not report this properly. 2243 */ 2244 if (sdp->removable && 2245 sense_valid && sshdr->sense_key == NOT_READY) 2246 set_media_not_present(sdkp); 2247 2248 /* 2249 * We used to set media_present to 0 here to indicate no media 2250 * in the drive, but some drives fail read capacity even with 2251 * media present, so we can't do that. 2252 */ 2253 sdkp->capacity = 0; /* unknown mapped to zero - as usual */ 2254 } 2255 2256 #define RC16_LEN 32 2257 #if RC16_LEN > SD_BUF_SIZE 2258 #error RC16_LEN must not be more than SD_BUF_SIZE 2259 #endif 2260 2261 #define READ_CAPACITY_RETRIES_ON_RESET 10 2262 2263 /* 2264 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set 2265 * and the reported logical block size is bigger than 512 bytes. Note 2266 * that last_sector is a u64 and therefore logical_to_sectors() is not 2267 * applicable. 2268 */ 2269 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size) 2270 { 2271 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9); 2272 2273 if (sizeof(sector_t) == 4 && last_sector > U32_MAX) 2274 return false; 2275 2276 return true; 2277 } 2278 2279 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp, 2280 unsigned char *buffer) 2281 { 2282 unsigned char cmd[16]; 2283 struct scsi_sense_hdr sshdr; 2284 int sense_valid = 0; 2285 int the_result; 2286 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET; 2287 unsigned int alignment; 2288 unsigned long long lba; 2289 unsigned sector_size; 2290 2291 if (sdp->no_read_capacity_16) 2292 return -EINVAL; 2293 2294 do { 2295 memset(cmd, 0, 16); 2296 cmd[0] = SERVICE_ACTION_IN_16; 2297 cmd[1] = SAI_READ_CAPACITY_16; 2298 cmd[13] = RC16_LEN; 2299 memset(buffer, 0, RC16_LEN); 2300 2301 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE, 2302 buffer, RC16_LEN, &sshdr, 2303 SD_TIMEOUT, SD_MAX_RETRIES, NULL); 2304 2305 if (media_not_present(sdkp, &sshdr)) 2306 return -ENODEV; 2307 2308 if (the_result) { 2309 sense_valid = scsi_sense_valid(&sshdr); 2310 if (sense_valid && 2311 sshdr.sense_key == ILLEGAL_REQUEST && 2312 (sshdr.asc == 0x20 || sshdr.asc == 0x24) && 2313 sshdr.ascq == 0x00) 2314 /* Invalid Command Operation Code or 2315 * Invalid Field in CDB, just retry 2316 * silently with RC10 */ 2317 return -EINVAL; 2318 if (sense_valid && 2319 sshdr.sense_key == UNIT_ATTENTION && 2320 sshdr.asc == 0x29 && sshdr.ascq == 0x00) 2321 /* Device reset might occur several times, 2322 * give it one more chance */ 2323 if (--reset_retries > 0) 2324 continue; 2325 } 2326 retries--; 2327 2328 } while (the_result && retries); 2329 2330 if (the_result) { 2331 sd_print_result(sdkp, "Read Capacity(16) failed", the_result); 2332 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result); 2333 return -EINVAL; 2334 } 2335 2336 sector_size = get_unaligned_be32(&buffer[8]); 2337 lba = get_unaligned_be64(&buffer[0]); 2338 2339 if (sd_read_protection_type(sdkp, buffer) < 0) { 2340 sdkp->capacity = 0; 2341 return -ENODEV; 2342 } 2343 2344 if (!sd_addressable_capacity(lba, sector_size)) { 2345 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a " 2346 "kernel compiled with support for large block " 2347 "devices.\n"); 2348 sdkp->capacity = 0; 2349 return -EOVERFLOW; 2350 } 2351 2352 /* Logical blocks per physical block exponent */ 2353 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size; 2354 2355 /* RC basis */ 2356 sdkp->rc_basis = (buffer[12] >> 4) & 0x3; 2357 2358 /* Lowest aligned logical block */ 2359 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size; 2360 blk_queue_alignment_offset(sdp->request_queue, alignment); 2361 if (alignment && sdkp->first_scan) 2362 sd_printk(KERN_NOTICE, sdkp, 2363 "physical block alignment offset: %u\n", alignment); 2364 2365 if (buffer[14] & 0x80) { /* LBPME */ 2366 sdkp->lbpme = 1; 2367 2368 if (buffer[14] & 0x40) /* LBPRZ */ 2369 sdkp->lbprz = 1; 2370 2371 sd_config_discard(sdkp, SD_LBP_WS16); 2372 } 2373 2374 sdkp->capacity = lba + 1; 2375 return sector_size; 2376 } 2377 2378 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp, 2379 unsigned char *buffer) 2380 { 2381 unsigned char cmd[16]; 2382 struct scsi_sense_hdr sshdr; 2383 int sense_valid = 0; 2384 int the_result; 2385 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET; 2386 sector_t lba; 2387 unsigned sector_size; 2388 2389 do { 2390 cmd[0] = READ_CAPACITY; 2391 memset(&cmd[1], 0, 9); 2392 memset(buffer, 0, 8); 2393 2394 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE, 2395 buffer, 8, &sshdr, 2396 SD_TIMEOUT, SD_MAX_RETRIES, NULL); 2397 2398 if (media_not_present(sdkp, &sshdr)) 2399 return -ENODEV; 2400 2401 if (the_result) { 2402 sense_valid = scsi_sense_valid(&sshdr); 2403 if (sense_valid && 2404 sshdr.sense_key == UNIT_ATTENTION && 2405 sshdr.asc == 0x29 && sshdr.ascq == 0x00) 2406 /* Device reset might occur several times, 2407 * give it one more chance */ 2408 if (--reset_retries > 0) 2409 continue; 2410 } 2411 retries--; 2412 2413 } while (the_result && retries); 2414 2415 if (the_result) { 2416 sd_print_result(sdkp, "Read Capacity(10) failed", the_result); 2417 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result); 2418 return -EINVAL; 2419 } 2420 2421 sector_size = get_unaligned_be32(&buffer[4]); 2422 lba = get_unaligned_be32(&buffer[0]); 2423 2424 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) { 2425 /* Some buggy (usb cardreader) devices return an lba of 2426 0xffffffff when the want to report a size of 0 (with 2427 which they really mean no media is present) */ 2428 sdkp->capacity = 0; 2429 sdkp->physical_block_size = sector_size; 2430 return sector_size; 2431 } 2432 2433 if (!sd_addressable_capacity(lba, sector_size)) { 2434 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a " 2435 "kernel compiled with support for large block " 2436 "devices.\n"); 2437 sdkp->capacity = 0; 2438 return -EOVERFLOW; 2439 } 2440 2441 sdkp->capacity = lba + 1; 2442 sdkp->physical_block_size = sector_size; 2443 return sector_size; 2444 } 2445 2446 static int sd_try_rc16_first(struct scsi_device *sdp) 2447 { 2448 if (sdp->host->max_cmd_len < 16) 2449 return 0; 2450 if (sdp->try_rc_10_first) 2451 return 0; 2452 if (sdp->scsi_level > SCSI_SPC_2) 2453 return 1; 2454 if (scsi_device_protection(sdp)) 2455 return 1; 2456 return 0; 2457 } 2458 2459 /* 2460 * read disk capacity 2461 */ 2462 static void 2463 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer) 2464 { 2465 int sector_size; 2466 struct scsi_device *sdp = sdkp->device; 2467 2468 if (sd_try_rc16_first(sdp)) { 2469 sector_size = read_capacity_16(sdkp, sdp, buffer); 2470 if (sector_size == -EOVERFLOW) 2471 goto got_data; 2472 if (sector_size == -ENODEV) 2473 return; 2474 if (sector_size < 0) 2475 sector_size = read_capacity_10(sdkp, sdp, buffer); 2476 if (sector_size < 0) 2477 return; 2478 } else { 2479 sector_size = read_capacity_10(sdkp, sdp, buffer); 2480 if (sector_size == -EOVERFLOW) 2481 goto got_data; 2482 if (sector_size < 0) 2483 return; 2484 if ((sizeof(sdkp->capacity) > 4) && 2485 (sdkp->capacity > 0xffffffffULL)) { 2486 int old_sector_size = sector_size; 2487 sd_printk(KERN_NOTICE, sdkp, "Very big device. " 2488 "Trying to use READ CAPACITY(16).\n"); 2489 sector_size = read_capacity_16(sdkp, sdp, buffer); 2490 if (sector_size < 0) { 2491 sd_printk(KERN_NOTICE, sdkp, 2492 "Using 0xffffffff as device size\n"); 2493 sdkp->capacity = 1 + (sector_t) 0xffffffff; 2494 sector_size = old_sector_size; 2495 goto got_data; 2496 } 2497 /* Remember that READ CAPACITY(16) succeeded */ 2498 sdp->try_rc_10_first = 0; 2499 } 2500 } 2501 2502 /* Some devices are known to return the total number of blocks, 2503 * not the highest block number. Some devices have versions 2504 * which do this and others which do not. Some devices we might 2505 * suspect of doing this but we don't know for certain. 2506 * 2507 * If we know the reported capacity is wrong, decrement it. If 2508 * we can only guess, then assume the number of blocks is even 2509 * (usually true but not always) and err on the side of lowering 2510 * the capacity. 2511 */ 2512 if (sdp->fix_capacity || 2513 (sdp->guess_capacity && (sdkp->capacity & 0x01))) { 2514 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count " 2515 "from its reported value: %llu\n", 2516 (unsigned long long) sdkp->capacity); 2517 --sdkp->capacity; 2518 } 2519 2520 got_data: 2521 if (sector_size == 0) { 2522 sector_size = 512; 2523 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, " 2524 "assuming 512.\n"); 2525 } 2526 2527 if (sector_size != 512 && 2528 sector_size != 1024 && 2529 sector_size != 2048 && 2530 sector_size != 4096) { 2531 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n", 2532 sector_size); 2533 /* 2534 * The user might want to re-format the drive with 2535 * a supported sectorsize. Once this happens, it 2536 * would be relatively trivial to set the thing up. 2537 * For this reason, we leave the thing in the table. 2538 */ 2539 sdkp->capacity = 0; 2540 /* 2541 * set a bogus sector size so the normal read/write 2542 * logic in the block layer will eventually refuse any 2543 * request on this device without tripping over power 2544 * of two sector size assumptions 2545 */ 2546 sector_size = 512; 2547 } 2548 blk_queue_logical_block_size(sdp->request_queue, sector_size); 2549 blk_queue_physical_block_size(sdp->request_queue, 2550 sdkp->physical_block_size); 2551 sdkp->device->sector_size = sector_size; 2552 2553 if (sdkp->capacity > 0xffffffff) 2554 sdp->use_16_for_rw = 1; 2555 2556 } 2557 2558 /* 2559 * Print disk capacity 2560 */ 2561 static void 2562 sd_print_capacity(struct scsi_disk *sdkp, 2563 sector_t old_capacity) 2564 { 2565 int sector_size = sdkp->device->sector_size; 2566 char cap_str_2[10], cap_str_10[10]; 2567 2568 string_get_size(sdkp->capacity, sector_size, 2569 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); 2570 string_get_size(sdkp->capacity, sector_size, 2571 STRING_UNITS_10, cap_str_10, 2572 sizeof(cap_str_10)); 2573 2574 if (sdkp->first_scan || old_capacity != sdkp->capacity) { 2575 sd_printk(KERN_NOTICE, sdkp, 2576 "%llu %d-byte logical blocks: (%s/%s)\n", 2577 (unsigned long long)sdkp->capacity, 2578 sector_size, cap_str_10, cap_str_2); 2579 2580 if (sdkp->physical_block_size != sector_size) 2581 sd_printk(KERN_NOTICE, sdkp, 2582 "%u-byte physical blocks\n", 2583 sdkp->physical_block_size); 2584 2585 sd_zbc_print_zones(sdkp); 2586 } 2587 } 2588 2589 /* called with buffer of length 512 */ 2590 static inline int 2591 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage, 2592 unsigned char *buffer, int len, struct scsi_mode_data *data, 2593 struct scsi_sense_hdr *sshdr) 2594 { 2595 return scsi_mode_sense(sdp, dbd, modepage, buffer, len, 2596 SD_TIMEOUT, SD_MAX_RETRIES, data, 2597 sshdr); 2598 } 2599 2600 /* 2601 * read write protect setting, if possible - called only in sd_revalidate_disk() 2602 * called with buffer of length SD_BUF_SIZE 2603 */ 2604 static void 2605 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer) 2606 { 2607 int res; 2608 struct scsi_device *sdp = sdkp->device; 2609 struct scsi_mode_data data; 2610 int disk_ro = get_disk_ro(sdkp->disk); 2611 int old_wp = sdkp->write_prot; 2612 2613 set_disk_ro(sdkp->disk, 0); 2614 if (sdp->skip_ms_page_3f) { 2615 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n"); 2616 return; 2617 } 2618 2619 if (sdp->use_192_bytes_for_3f) { 2620 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL); 2621 } else { 2622 /* 2623 * First attempt: ask for all pages (0x3F), but only 4 bytes. 2624 * We have to start carefully: some devices hang if we ask 2625 * for more than is available. 2626 */ 2627 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL); 2628 2629 /* 2630 * Second attempt: ask for page 0 When only page 0 is 2631 * implemented, a request for page 3F may return Sense Key 2632 * 5: Illegal Request, Sense Code 24: Invalid field in 2633 * CDB. 2634 */ 2635 if (!scsi_status_is_good(res)) 2636 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL); 2637 2638 /* 2639 * Third attempt: ask 255 bytes, as we did earlier. 2640 */ 2641 if (!scsi_status_is_good(res)) 2642 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255, 2643 &data, NULL); 2644 } 2645 2646 if (!scsi_status_is_good(res)) { 2647 sd_first_printk(KERN_WARNING, sdkp, 2648 "Test WP failed, assume Write Enabled\n"); 2649 } else { 2650 sdkp->write_prot = ((data.device_specific & 0x80) != 0); 2651 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro); 2652 if (sdkp->first_scan || old_wp != sdkp->write_prot) { 2653 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n", 2654 sdkp->write_prot ? "on" : "off"); 2655 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer); 2656 } 2657 } 2658 } 2659 2660 /* 2661 * sd_read_cache_type - called only from sd_revalidate_disk() 2662 * called with buffer of length SD_BUF_SIZE 2663 */ 2664 static void 2665 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer) 2666 { 2667 int len = 0, res; 2668 struct scsi_device *sdp = sdkp->device; 2669 2670 int dbd; 2671 int modepage; 2672 int first_len; 2673 struct scsi_mode_data data; 2674 struct scsi_sense_hdr sshdr; 2675 int old_wce = sdkp->WCE; 2676 int old_rcd = sdkp->RCD; 2677 int old_dpofua = sdkp->DPOFUA; 2678 2679 2680 if (sdkp->cache_override) 2681 return; 2682 2683 first_len = 4; 2684 if (sdp->skip_ms_page_8) { 2685 if (sdp->type == TYPE_RBC) 2686 goto defaults; 2687 else { 2688 if (sdp->skip_ms_page_3f) 2689 goto defaults; 2690 modepage = 0x3F; 2691 if (sdp->use_192_bytes_for_3f) 2692 first_len = 192; 2693 dbd = 0; 2694 } 2695 } else if (sdp->type == TYPE_RBC) { 2696 modepage = 6; 2697 dbd = 8; 2698 } else { 2699 modepage = 8; 2700 dbd = 0; 2701 } 2702 2703 /* cautiously ask */ 2704 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len, 2705 &data, &sshdr); 2706 2707 if (!scsi_status_is_good(res)) 2708 goto bad_sense; 2709 2710 if (!data.header_length) { 2711 modepage = 6; 2712 first_len = 0; 2713 sd_first_printk(KERN_ERR, sdkp, 2714 "Missing header in MODE_SENSE response\n"); 2715 } 2716 2717 /* that went OK, now ask for the proper length */ 2718 len = data.length; 2719 2720 /* 2721 * We're only interested in the first three bytes, actually. 2722 * But the data cache page is defined for the first 20. 2723 */ 2724 if (len < 3) 2725 goto bad_sense; 2726 else if (len > SD_BUF_SIZE) { 2727 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter " 2728 "data from %d to %d bytes\n", len, SD_BUF_SIZE); 2729 len = SD_BUF_SIZE; 2730 } 2731 if (modepage == 0x3F && sdp->use_192_bytes_for_3f) 2732 len = 192; 2733 2734 /* Get the data */ 2735 if (len > first_len) 2736 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, 2737 &data, &sshdr); 2738 2739 if (scsi_status_is_good(res)) { 2740 int offset = data.header_length + data.block_descriptor_length; 2741 2742 while (offset < len) { 2743 u8 page_code = buffer[offset] & 0x3F; 2744 u8 spf = buffer[offset] & 0x40; 2745 2746 if (page_code == 8 || page_code == 6) { 2747 /* We're interested only in the first 3 bytes. 2748 */ 2749 if (len - offset <= 2) { 2750 sd_first_printk(KERN_ERR, sdkp, 2751 "Incomplete mode parameter " 2752 "data\n"); 2753 goto defaults; 2754 } else { 2755 modepage = page_code; 2756 goto Page_found; 2757 } 2758 } else { 2759 /* Go to the next page */ 2760 if (spf && len - offset > 3) 2761 offset += 4 + (buffer[offset+2] << 8) + 2762 buffer[offset+3]; 2763 else if (!spf && len - offset > 1) 2764 offset += 2 + buffer[offset+1]; 2765 else { 2766 sd_first_printk(KERN_ERR, sdkp, 2767 "Incomplete mode " 2768 "parameter data\n"); 2769 goto defaults; 2770 } 2771 } 2772 } 2773 2774 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n"); 2775 goto defaults; 2776 2777 Page_found: 2778 if (modepage == 8) { 2779 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0); 2780 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0); 2781 } else { 2782 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0); 2783 sdkp->RCD = 0; 2784 } 2785 2786 sdkp->DPOFUA = (data.device_specific & 0x10) != 0; 2787 if (sdp->broken_fua) { 2788 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n"); 2789 sdkp->DPOFUA = 0; 2790 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw && 2791 !sdkp->device->use_16_for_rw) { 2792 sd_first_printk(KERN_NOTICE, sdkp, 2793 "Uses READ/WRITE(6), disabling FUA\n"); 2794 sdkp->DPOFUA = 0; 2795 } 2796 2797 /* No cache flush allowed for write protected devices */ 2798 if (sdkp->WCE && sdkp->write_prot) 2799 sdkp->WCE = 0; 2800 2801 if (sdkp->first_scan || old_wce != sdkp->WCE || 2802 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA) 2803 sd_printk(KERN_NOTICE, sdkp, 2804 "Write cache: %s, read cache: %s, %s\n", 2805 sdkp->WCE ? "enabled" : "disabled", 2806 sdkp->RCD ? "disabled" : "enabled", 2807 sdkp->DPOFUA ? "supports DPO and FUA" 2808 : "doesn't support DPO or FUA"); 2809 2810 return; 2811 } 2812 2813 bad_sense: 2814 if (scsi_sense_valid(&sshdr) && 2815 sshdr.sense_key == ILLEGAL_REQUEST && 2816 sshdr.asc == 0x24 && sshdr.ascq == 0x0) 2817 /* Invalid field in CDB */ 2818 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n"); 2819 else 2820 sd_first_printk(KERN_ERR, sdkp, 2821 "Asking for cache data failed\n"); 2822 2823 defaults: 2824 if (sdp->wce_default_on) { 2825 sd_first_printk(KERN_NOTICE, sdkp, 2826 "Assuming drive cache: write back\n"); 2827 sdkp->WCE = 1; 2828 } else { 2829 sd_first_printk(KERN_ERR, sdkp, 2830 "Assuming drive cache: write through\n"); 2831 sdkp->WCE = 0; 2832 } 2833 sdkp->RCD = 0; 2834 sdkp->DPOFUA = 0; 2835 } 2836 2837 /* 2838 * The ATO bit indicates whether the DIF application tag is available 2839 * for use by the operating system. 2840 */ 2841 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer) 2842 { 2843 int res, offset; 2844 struct scsi_device *sdp = sdkp->device; 2845 struct scsi_mode_data data; 2846 struct scsi_sense_hdr sshdr; 2847 2848 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 2849 return; 2850 2851 if (sdkp->protection_type == 0) 2852 return; 2853 2854 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT, 2855 SD_MAX_RETRIES, &data, &sshdr); 2856 2857 if (!scsi_status_is_good(res) || !data.header_length || 2858 data.length < 6) { 2859 sd_first_printk(KERN_WARNING, sdkp, 2860 "getting Control mode page failed, assume no ATO\n"); 2861 2862 if (scsi_sense_valid(&sshdr)) 2863 sd_print_sense_hdr(sdkp, &sshdr); 2864 2865 return; 2866 } 2867 2868 offset = data.header_length + data.block_descriptor_length; 2869 2870 if ((buffer[offset] & 0x3f) != 0x0a) { 2871 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n"); 2872 return; 2873 } 2874 2875 if ((buffer[offset + 5] & 0x80) == 0) 2876 return; 2877 2878 sdkp->ATO = 1; 2879 2880 return; 2881 } 2882 2883 /** 2884 * sd_read_block_limits - Query disk device for preferred I/O sizes. 2885 * @sdkp: disk to query 2886 */ 2887 static void sd_read_block_limits(struct scsi_disk *sdkp) 2888 { 2889 unsigned int sector_sz = sdkp->device->sector_size; 2890 const int vpd_len = 64; 2891 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL); 2892 2893 if (!buffer || 2894 /* Block Limits VPD */ 2895 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len)) 2896 goto out; 2897 2898 blk_queue_io_min(sdkp->disk->queue, 2899 get_unaligned_be16(&buffer[6]) * sector_sz); 2900 2901 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]); 2902 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]); 2903 2904 if (buffer[3] == 0x3c) { 2905 unsigned int lba_count, desc_count; 2906 2907 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]); 2908 2909 if (!sdkp->lbpme) 2910 goto out; 2911 2912 lba_count = get_unaligned_be32(&buffer[20]); 2913 desc_count = get_unaligned_be32(&buffer[24]); 2914 2915 if (lba_count && desc_count) 2916 sdkp->max_unmap_blocks = lba_count; 2917 2918 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]); 2919 2920 if (buffer[32] & 0x80) 2921 sdkp->unmap_alignment = 2922 get_unaligned_be32(&buffer[32]) & ~(1 << 31); 2923 2924 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */ 2925 2926 if (sdkp->max_unmap_blocks) 2927 sd_config_discard(sdkp, SD_LBP_UNMAP); 2928 else 2929 sd_config_discard(sdkp, SD_LBP_WS16); 2930 2931 } else { /* LBP VPD page tells us what to use */ 2932 if (sdkp->lbpu && sdkp->max_unmap_blocks) 2933 sd_config_discard(sdkp, SD_LBP_UNMAP); 2934 else if (sdkp->lbpws) 2935 sd_config_discard(sdkp, SD_LBP_WS16); 2936 else if (sdkp->lbpws10) 2937 sd_config_discard(sdkp, SD_LBP_WS10); 2938 else 2939 sd_config_discard(sdkp, SD_LBP_DISABLE); 2940 } 2941 } 2942 2943 out: 2944 kfree(buffer); 2945 } 2946 2947 /** 2948 * sd_read_block_characteristics - Query block dev. characteristics 2949 * @sdkp: disk to query 2950 */ 2951 static void sd_read_block_characteristics(struct scsi_disk *sdkp) 2952 { 2953 struct request_queue *q = sdkp->disk->queue; 2954 unsigned char *buffer; 2955 u16 rot; 2956 const int vpd_len = 64; 2957 2958 buffer = kmalloc(vpd_len, GFP_KERNEL); 2959 2960 if (!buffer || 2961 /* Block Device Characteristics VPD */ 2962 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len)) 2963 goto out; 2964 2965 rot = get_unaligned_be16(&buffer[4]); 2966 2967 if (rot == 1) { 2968 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 2969 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q); 2970 } else { 2971 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 2972 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q); 2973 } 2974 2975 if (sdkp->device->type == TYPE_ZBC) { 2976 /* Host-managed */ 2977 q->limits.zoned = BLK_ZONED_HM; 2978 } else { 2979 sdkp->zoned = (buffer[8] >> 4) & 3; 2980 if (sdkp->zoned == 1) 2981 /* Host-aware */ 2982 q->limits.zoned = BLK_ZONED_HA; 2983 else 2984 /* 2985 * Treat drive-managed devices as 2986 * regular block devices. 2987 */ 2988 q->limits.zoned = BLK_ZONED_NONE; 2989 } 2990 if (blk_queue_is_zoned(q) && sdkp->first_scan) 2991 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n", 2992 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware"); 2993 2994 out: 2995 kfree(buffer); 2996 } 2997 2998 /** 2999 * sd_read_block_provisioning - Query provisioning VPD page 3000 * @sdkp: disk to query 3001 */ 3002 static void sd_read_block_provisioning(struct scsi_disk *sdkp) 3003 { 3004 unsigned char *buffer; 3005 const int vpd_len = 8; 3006 3007 if (sdkp->lbpme == 0) 3008 return; 3009 3010 buffer = kmalloc(vpd_len, GFP_KERNEL); 3011 3012 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len)) 3013 goto out; 3014 3015 sdkp->lbpvpd = 1; 3016 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */ 3017 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */ 3018 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */ 3019 3020 out: 3021 kfree(buffer); 3022 } 3023 3024 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer) 3025 { 3026 struct scsi_device *sdev = sdkp->device; 3027 3028 if (sdev->host->no_write_same) { 3029 sdev->no_write_same = 1; 3030 3031 return; 3032 } 3033 3034 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) { 3035 /* too large values might cause issues with arcmsr */ 3036 int vpd_buf_len = 64; 3037 3038 sdev->no_report_opcodes = 1; 3039 3040 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION 3041 * CODES is unsupported and the device has an ATA 3042 * Information VPD page (SAT). 3043 */ 3044 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len)) 3045 sdev->no_write_same = 1; 3046 } 3047 3048 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1) 3049 sdkp->ws16 = 1; 3050 3051 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1) 3052 sdkp->ws10 = 1; 3053 } 3054 3055 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer) 3056 { 3057 struct scsi_device *sdev = sdkp->device; 3058 3059 if (!sdev->security_supported) 3060 return; 3061 3062 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, 3063 SECURITY_PROTOCOL_IN) == 1 && 3064 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, 3065 SECURITY_PROTOCOL_OUT) == 1) 3066 sdkp->security = 1; 3067 } 3068 3069 /** 3070 * sd_revalidate_disk - called the first time a new disk is seen, 3071 * performs disk spin up, read_capacity, etc. 3072 * @disk: struct gendisk we care about 3073 **/ 3074 static int sd_revalidate_disk(struct gendisk *disk) 3075 { 3076 struct scsi_disk *sdkp = scsi_disk(disk); 3077 struct scsi_device *sdp = sdkp->device; 3078 struct request_queue *q = sdkp->disk->queue; 3079 sector_t old_capacity = sdkp->capacity; 3080 unsigned char *buffer; 3081 unsigned int dev_max, rw_max; 3082 3083 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, 3084 "sd_revalidate_disk\n")); 3085 3086 /* 3087 * If the device is offline, don't try and read capacity or any 3088 * of the other niceties. 3089 */ 3090 if (!scsi_device_online(sdp)) 3091 goto out; 3092 3093 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL); 3094 if (!buffer) { 3095 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory " 3096 "allocation failure.\n"); 3097 goto out; 3098 } 3099 3100 sd_spinup_disk(sdkp); 3101 3102 /* 3103 * Without media there is no reason to ask; moreover, some devices 3104 * react badly if we do. 3105 */ 3106 if (sdkp->media_present) { 3107 sd_read_capacity(sdkp, buffer); 3108 3109 if (scsi_device_supports_vpd(sdp)) { 3110 sd_read_block_provisioning(sdkp); 3111 sd_read_block_limits(sdkp); 3112 sd_read_block_characteristics(sdkp); 3113 sd_zbc_read_zones(sdkp, buffer); 3114 } 3115 3116 sd_print_capacity(sdkp, old_capacity); 3117 3118 sd_read_write_protect_flag(sdkp, buffer); 3119 sd_read_cache_type(sdkp, buffer); 3120 sd_read_app_tag_own(sdkp, buffer); 3121 sd_read_write_same(sdkp, buffer); 3122 sd_read_security(sdkp, buffer); 3123 } 3124 3125 /* 3126 * We now have all cache related info, determine how we deal 3127 * with flush requests. 3128 */ 3129 sd_set_flush_flag(sdkp); 3130 3131 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */ 3132 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS; 3133 3134 /* Some devices report a maximum block count for READ/WRITE requests. */ 3135 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks); 3136 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max); 3137 3138 /* 3139 * Determine the device's preferred I/O size for reads and writes 3140 * unless the reported value is unreasonably small, large, or 3141 * garbage. 3142 */ 3143 if (sdkp->opt_xfer_blocks && 3144 sdkp->opt_xfer_blocks <= dev_max && 3145 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS && 3146 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) { 3147 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks); 3148 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks); 3149 } else 3150 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max), 3151 (sector_t)BLK_DEF_MAX_SECTORS); 3152 3153 /* Do not exceed controller limit */ 3154 rw_max = min(rw_max, queue_max_hw_sectors(q)); 3155 3156 /* 3157 * Only update max_sectors if previously unset or if the current value 3158 * exceeds the capabilities of the hardware. 3159 */ 3160 if (sdkp->first_scan || 3161 q->limits.max_sectors > q->limits.max_dev_sectors || 3162 q->limits.max_sectors > q->limits.max_hw_sectors) 3163 q->limits.max_sectors = rw_max; 3164 3165 sdkp->first_scan = 0; 3166 3167 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity)); 3168 sd_config_write_same(sdkp); 3169 kfree(buffer); 3170 3171 out: 3172 return 0; 3173 } 3174 3175 /** 3176 * sd_unlock_native_capacity - unlock native capacity 3177 * @disk: struct gendisk to set capacity for 3178 * 3179 * Block layer calls this function if it detects that partitions 3180 * on @disk reach beyond the end of the device. If the SCSI host 3181 * implements ->unlock_native_capacity() method, it's invoked to 3182 * give it a chance to adjust the device capacity. 3183 * 3184 * CONTEXT: 3185 * Defined by block layer. Might sleep. 3186 */ 3187 static void sd_unlock_native_capacity(struct gendisk *disk) 3188 { 3189 struct scsi_device *sdev = scsi_disk(disk)->device; 3190 3191 if (sdev->host->hostt->unlock_native_capacity) 3192 sdev->host->hostt->unlock_native_capacity(sdev); 3193 } 3194 3195 /** 3196 * sd_format_disk_name - format disk name 3197 * @prefix: name prefix - ie. "sd" for SCSI disks 3198 * @index: index of the disk to format name for 3199 * @buf: output buffer 3200 * @buflen: length of the output buffer 3201 * 3202 * SCSI disk names starts at sda. The 26th device is sdz and the 3203 * 27th is sdaa. The last one for two lettered suffix is sdzz 3204 * which is followed by sdaaa. 3205 * 3206 * This is basically 26 base counting with one extra 'nil' entry 3207 * at the beginning from the second digit on and can be 3208 * determined using similar method as 26 base conversion with the 3209 * index shifted -1 after each digit is computed. 3210 * 3211 * CONTEXT: 3212 * Don't care. 3213 * 3214 * RETURNS: 3215 * 0 on success, -errno on failure. 3216 */ 3217 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen) 3218 { 3219 const int base = 'z' - 'a' + 1; 3220 char *begin = buf + strlen(prefix); 3221 char *end = buf + buflen; 3222 char *p; 3223 int unit; 3224 3225 p = end - 1; 3226 *p = '\0'; 3227 unit = base; 3228 do { 3229 if (p == begin) 3230 return -EINVAL; 3231 *--p = 'a' + (index % unit); 3232 index = (index / unit) - 1; 3233 } while (index >= 0); 3234 3235 memmove(begin, p, end - p); 3236 memcpy(buf, prefix, strlen(prefix)); 3237 3238 return 0; 3239 } 3240 3241 /* 3242 * The asynchronous part of sd_probe 3243 */ 3244 static void sd_probe_async(void *data, async_cookie_t cookie) 3245 { 3246 struct scsi_disk *sdkp = data; 3247 struct scsi_device *sdp; 3248 struct gendisk *gd; 3249 u32 index; 3250 struct device *dev; 3251 3252 sdp = sdkp->device; 3253 gd = sdkp->disk; 3254 index = sdkp->index; 3255 dev = &sdp->sdev_gendev; 3256 3257 gd->major = sd_major((index & 0xf0) >> 4); 3258 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00); 3259 3260 gd->fops = &sd_fops; 3261 gd->private_data = &sdkp->driver; 3262 gd->queue = sdkp->device->request_queue; 3263 3264 /* defaults, until the device tells us otherwise */ 3265 sdp->sector_size = 512; 3266 sdkp->capacity = 0; 3267 sdkp->media_present = 1; 3268 sdkp->write_prot = 0; 3269 sdkp->cache_override = 0; 3270 sdkp->WCE = 0; 3271 sdkp->RCD = 0; 3272 sdkp->ATO = 0; 3273 sdkp->first_scan = 1; 3274 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS; 3275 3276 sd_revalidate_disk(gd); 3277 3278 gd->flags = GENHD_FL_EXT_DEVT; 3279 if (sdp->removable) { 3280 gd->flags |= GENHD_FL_REMOVABLE; 3281 gd->events |= DISK_EVENT_MEDIA_CHANGE; 3282 } 3283 3284 blk_pm_runtime_init(sdp->request_queue, dev); 3285 device_add_disk(dev, gd, NULL); 3286 if (sdkp->capacity) 3287 sd_dif_config_host(sdkp); 3288 3289 sd_revalidate_disk(gd); 3290 3291 if (sdkp->security) { 3292 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit); 3293 if (sdkp->opal_dev) 3294 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n"); 3295 } 3296 3297 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n", 3298 sdp->removable ? "removable " : ""); 3299 scsi_autopm_put_device(sdp); 3300 put_device(&sdkp->dev); 3301 } 3302 3303 /** 3304 * sd_probe - called during driver initialization and whenever a 3305 * new scsi device is attached to the system. It is called once 3306 * for each scsi device (not just disks) present. 3307 * @dev: pointer to device object 3308 * 3309 * Returns 0 if successful (or not interested in this scsi device 3310 * (e.g. scanner)); 1 when there is an error. 3311 * 3312 * Note: this function is invoked from the scsi mid-level. 3313 * This function sets up the mapping between a given 3314 * <host,channel,id,lun> (found in sdp) and new device name 3315 * (e.g. /dev/sda). More precisely it is the block device major 3316 * and minor number that is chosen here. 3317 * 3318 * Assume sd_probe is not re-entrant (for time being) 3319 * Also think about sd_probe() and sd_remove() running coincidentally. 3320 **/ 3321 static int sd_probe(struct device *dev) 3322 { 3323 struct scsi_device *sdp = to_scsi_device(dev); 3324 struct scsi_disk *sdkp; 3325 struct gendisk *gd; 3326 int index; 3327 int error; 3328 3329 scsi_autopm_get_device(sdp); 3330 error = -ENODEV; 3331 if (sdp->type != TYPE_DISK && 3332 sdp->type != TYPE_ZBC && 3333 sdp->type != TYPE_MOD && 3334 sdp->type != TYPE_RBC) 3335 goto out; 3336 3337 #ifndef CONFIG_BLK_DEV_ZONED 3338 if (sdp->type == TYPE_ZBC) 3339 goto out; 3340 #endif 3341 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp, 3342 "sd_probe\n")); 3343 3344 error = -ENOMEM; 3345 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL); 3346 if (!sdkp) 3347 goto out; 3348 3349 gd = alloc_disk(SD_MINORS); 3350 if (!gd) 3351 goto out_free; 3352 3353 index = ida_alloc(&sd_index_ida, GFP_KERNEL); 3354 if (index < 0) { 3355 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n"); 3356 goto out_put; 3357 } 3358 3359 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN); 3360 if (error) { 3361 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n"); 3362 goto out_free_index; 3363 } 3364 3365 sdkp->device = sdp; 3366 sdkp->driver = &sd_template; 3367 sdkp->disk = gd; 3368 sdkp->index = index; 3369 atomic_set(&sdkp->openers, 0); 3370 atomic_set(&sdkp->device->ioerr_cnt, 0); 3371 3372 if (!sdp->request_queue->rq_timeout) { 3373 if (sdp->type != TYPE_MOD) 3374 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT); 3375 else 3376 blk_queue_rq_timeout(sdp->request_queue, 3377 SD_MOD_TIMEOUT); 3378 } 3379 3380 device_initialize(&sdkp->dev); 3381 sdkp->dev.parent = dev; 3382 sdkp->dev.class = &sd_disk_class; 3383 dev_set_name(&sdkp->dev, "%s", dev_name(dev)); 3384 3385 error = device_add(&sdkp->dev); 3386 if (error) 3387 goto out_free_index; 3388 3389 get_device(dev); 3390 dev_set_drvdata(dev, sdkp); 3391 3392 get_device(&sdkp->dev); /* prevent release before async_schedule */ 3393 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain); 3394 3395 return 0; 3396 3397 out_free_index: 3398 ida_free(&sd_index_ida, index); 3399 out_put: 3400 put_disk(gd); 3401 out_free: 3402 kfree(sdkp); 3403 out: 3404 scsi_autopm_put_device(sdp); 3405 return error; 3406 } 3407 3408 /** 3409 * sd_remove - called whenever a scsi disk (previously recognized by 3410 * sd_probe) is detached from the system. It is called (potentially 3411 * multiple times) during sd module unload. 3412 * @dev: pointer to device object 3413 * 3414 * Note: this function is invoked from the scsi mid-level. 3415 * This function potentially frees up a device name (e.g. /dev/sdc) 3416 * that could be re-used by a subsequent sd_probe(). 3417 * This function is not called when the built-in sd driver is "exit-ed". 3418 **/ 3419 static int sd_remove(struct device *dev) 3420 { 3421 struct scsi_disk *sdkp; 3422 dev_t devt; 3423 3424 sdkp = dev_get_drvdata(dev); 3425 devt = disk_devt(sdkp->disk); 3426 scsi_autopm_get_device(sdkp->device); 3427 3428 async_synchronize_full_domain(&scsi_sd_pm_domain); 3429 async_synchronize_full_domain(&scsi_sd_probe_domain); 3430 device_del(&sdkp->dev); 3431 del_gendisk(sdkp->disk); 3432 sd_shutdown(dev); 3433 3434 free_opal_dev(sdkp->opal_dev); 3435 3436 blk_register_region(devt, SD_MINORS, NULL, 3437 sd_default_probe, NULL, NULL); 3438 3439 mutex_lock(&sd_ref_mutex); 3440 dev_set_drvdata(dev, NULL); 3441 put_device(&sdkp->dev); 3442 mutex_unlock(&sd_ref_mutex); 3443 3444 return 0; 3445 } 3446 3447 /** 3448 * scsi_disk_release - Called to free the scsi_disk structure 3449 * @dev: pointer to embedded class device 3450 * 3451 * sd_ref_mutex must be held entering this routine. Because it is 3452 * called on last put, you should always use the scsi_disk_get() 3453 * scsi_disk_put() helpers which manipulate the semaphore directly 3454 * and never do a direct put_device. 3455 **/ 3456 static void scsi_disk_release(struct device *dev) 3457 { 3458 struct scsi_disk *sdkp = to_scsi_disk(dev); 3459 struct gendisk *disk = sdkp->disk; 3460 3461 ida_free(&sd_index_ida, sdkp->index); 3462 3463 disk->private_data = NULL; 3464 put_disk(disk); 3465 put_device(&sdkp->device->sdev_gendev); 3466 3467 kfree(sdkp); 3468 } 3469 3470 static int sd_start_stop_device(struct scsi_disk *sdkp, int start) 3471 { 3472 unsigned char cmd[6] = { START_STOP }; /* START_VALID */ 3473 struct scsi_sense_hdr sshdr; 3474 struct scsi_device *sdp = sdkp->device; 3475 int res; 3476 3477 if (start) 3478 cmd[4] |= 1; /* START */ 3479 3480 if (sdp->start_stop_pwr_cond) 3481 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */ 3482 3483 if (!scsi_device_online(sdp)) 3484 return -ENODEV; 3485 3486 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr, 3487 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL); 3488 if (res) { 3489 sd_print_result(sdkp, "Start/Stop Unit failed", res); 3490 if (driver_byte(res) == DRIVER_SENSE) 3491 sd_print_sense_hdr(sdkp, &sshdr); 3492 if (scsi_sense_valid(&sshdr) && 3493 /* 0x3a is medium not present */ 3494 sshdr.asc == 0x3a) 3495 res = 0; 3496 } 3497 3498 /* SCSI error codes must not go to the generic layer */ 3499 if (res) 3500 return -EIO; 3501 3502 return 0; 3503 } 3504 3505 /* 3506 * Send a SYNCHRONIZE CACHE instruction down to the device through 3507 * the normal SCSI command structure. Wait for the command to 3508 * complete. 3509 */ 3510 static void sd_shutdown(struct device *dev) 3511 { 3512 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3513 3514 if (!sdkp) 3515 return; /* this can happen */ 3516 3517 if (pm_runtime_suspended(dev)) 3518 return; 3519 3520 if (sdkp->WCE && sdkp->media_present) { 3521 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n"); 3522 sd_sync_cache(sdkp, NULL); 3523 } 3524 3525 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) { 3526 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n"); 3527 sd_start_stop_device(sdkp, 0); 3528 } 3529 } 3530 3531 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors) 3532 { 3533 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3534 struct scsi_sense_hdr sshdr; 3535 int ret = 0; 3536 3537 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */ 3538 return 0; 3539 3540 if (sdkp->WCE && sdkp->media_present) { 3541 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n"); 3542 ret = sd_sync_cache(sdkp, &sshdr); 3543 3544 if (ret) { 3545 /* ignore OFFLINE device */ 3546 if (ret == -ENODEV) 3547 return 0; 3548 3549 if (!scsi_sense_valid(&sshdr) || 3550 sshdr.sense_key != ILLEGAL_REQUEST) 3551 return ret; 3552 3553 /* 3554 * sshdr.sense_key == ILLEGAL_REQUEST means this drive 3555 * doesn't support sync. There's not much to do and 3556 * suspend shouldn't fail. 3557 */ 3558 ret = 0; 3559 } 3560 } 3561 3562 if (sdkp->device->manage_start_stop) { 3563 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n"); 3564 /* an error is not worth aborting a system sleep */ 3565 ret = sd_start_stop_device(sdkp, 0); 3566 if (ignore_stop_errors) 3567 ret = 0; 3568 } 3569 3570 return ret; 3571 } 3572 3573 static int sd_suspend_system(struct device *dev) 3574 { 3575 return sd_suspend_common(dev, true); 3576 } 3577 3578 static int sd_suspend_runtime(struct device *dev) 3579 { 3580 return sd_suspend_common(dev, false); 3581 } 3582 3583 static int sd_resume(struct device *dev) 3584 { 3585 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3586 int ret; 3587 3588 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */ 3589 return 0; 3590 3591 if (!sdkp->device->manage_start_stop) 3592 return 0; 3593 3594 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n"); 3595 ret = sd_start_stop_device(sdkp, 1); 3596 if (!ret) 3597 opal_unlock_from_suspend(sdkp->opal_dev); 3598 return ret; 3599 } 3600 3601 /** 3602 * init_sd - entry point for this driver (both when built in or when 3603 * a module). 3604 * 3605 * Note: this function registers this driver with the scsi mid-level. 3606 **/ 3607 static int __init init_sd(void) 3608 { 3609 int majors = 0, i, err; 3610 3611 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n")); 3612 3613 for (i = 0; i < SD_MAJORS; i++) { 3614 if (register_blkdev(sd_major(i), "sd") != 0) 3615 continue; 3616 majors++; 3617 blk_register_region(sd_major(i), SD_MINORS, NULL, 3618 sd_default_probe, NULL, NULL); 3619 } 3620 3621 if (!majors) 3622 return -ENODEV; 3623 3624 err = class_register(&sd_disk_class); 3625 if (err) 3626 goto err_out; 3627 3628 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE, 3629 0, 0, NULL); 3630 if (!sd_cdb_cache) { 3631 printk(KERN_ERR "sd: can't init extended cdb cache\n"); 3632 err = -ENOMEM; 3633 goto err_out_class; 3634 } 3635 3636 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache); 3637 if (!sd_cdb_pool) { 3638 printk(KERN_ERR "sd: can't init extended cdb pool\n"); 3639 err = -ENOMEM; 3640 goto err_out_cache; 3641 } 3642 3643 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0); 3644 if (!sd_page_pool) { 3645 printk(KERN_ERR "sd: can't init discard page pool\n"); 3646 err = -ENOMEM; 3647 goto err_out_ppool; 3648 } 3649 3650 err = scsi_register_driver(&sd_template.gendrv); 3651 if (err) 3652 goto err_out_driver; 3653 3654 return 0; 3655 3656 err_out_driver: 3657 mempool_destroy(sd_page_pool); 3658 3659 err_out_ppool: 3660 mempool_destroy(sd_cdb_pool); 3661 3662 err_out_cache: 3663 kmem_cache_destroy(sd_cdb_cache); 3664 3665 err_out_class: 3666 class_unregister(&sd_disk_class); 3667 err_out: 3668 for (i = 0; i < SD_MAJORS; i++) 3669 unregister_blkdev(sd_major(i), "sd"); 3670 return err; 3671 } 3672 3673 /** 3674 * exit_sd - exit point for this driver (when it is a module). 3675 * 3676 * Note: this function unregisters this driver from the scsi mid-level. 3677 **/ 3678 static void __exit exit_sd(void) 3679 { 3680 int i; 3681 3682 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n")); 3683 3684 scsi_unregister_driver(&sd_template.gendrv); 3685 mempool_destroy(sd_cdb_pool); 3686 mempool_destroy(sd_page_pool); 3687 kmem_cache_destroy(sd_cdb_cache); 3688 3689 class_unregister(&sd_disk_class); 3690 3691 for (i = 0; i < SD_MAJORS; i++) { 3692 blk_unregister_region(sd_major(i), SD_MINORS); 3693 unregister_blkdev(sd_major(i), "sd"); 3694 } 3695 } 3696 3697 module_init(init_sd); 3698 module_exit(exit_sd); 3699 3700 static void sd_print_sense_hdr(struct scsi_disk *sdkp, 3701 struct scsi_sense_hdr *sshdr) 3702 { 3703 scsi_print_sense_hdr(sdkp->device, 3704 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr); 3705 } 3706 3707 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg, 3708 int result) 3709 { 3710 const char *hb_string = scsi_hostbyte_string(result); 3711 const char *db_string = scsi_driverbyte_string(result); 3712 3713 if (hb_string || db_string) 3714 sd_printk(KERN_INFO, sdkp, 3715 "%s: Result: hostbyte=%s driverbyte=%s\n", msg, 3716 hb_string ? hb_string : "invalid", 3717 db_string ? db_string : "invalid"); 3718 else 3719 sd_printk(KERN_INFO, sdkp, 3720 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n", 3721 msg, host_byte(result), driver_byte(result)); 3722 } 3723 3724