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/delay.h> 49 #include <linux/mutex.h> 50 #include <linux/string_helpers.h> 51 #include <linux/async.h> 52 #include <linux/slab.h> 53 #include <linux/sed-opal.h> 54 #include <linux/pm_runtime.h> 55 #include <linux/pr.h> 56 #include <linux/t10-pi.h> 57 #include <linux/uaccess.h> 58 #include <asm/unaligned.h> 59 60 #include <scsi/scsi.h> 61 #include <scsi/scsi_cmnd.h> 62 #include <scsi/scsi_dbg.h> 63 #include <scsi/scsi_device.h> 64 #include <scsi/scsi_driver.h> 65 #include <scsi/scsi_eh.h> 66 #include <scsi/scsi_host.h> 67 #include <scsi/scsi_ioctl.h> 68 #include <scsi/scsicam.h> 69 70 #include "sd.h" 71 #include "scsi_priv.h" 72 #include "scsi_logging.h" 73 74 MODULE_AUTHOR("Eric Youngdale"); 75 MODULE_DESCRIPTION("SCSI disk (sd) driver"); 76 MODULE_LICENSE("GPL"); 77 78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR); 79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR); 80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR); 81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR); 82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR); 83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR); 84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR); 85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR); 86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR); 87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR); 88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR); 89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR); 90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR); 91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR); 92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR); 93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR); 94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK); 95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD); 96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC); 97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC); 98 99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT) 100 #define SD_MINORS 16 101 #else 102 #define SD_MINORS 0 103 #endif 104 105 static void sd_config_discard(struct scsi_disk *, unsigned int); 106 static void sd_config_write_same(struct scsi_disk *); 107 static int sd_revalidate_disk(struct gendisk *); 108 static void sd_unlock_native_capacity(struct gendisk *disk); 109 static int sd_probe(struct device *); 110 static int sd_remove(struct device *); 111 static void sd_shutdown(struct device *); 112 static int sd_suspend_system(struct device *); 113 static int sd_suspend_runtime(struct device *); 114 static int sd_resume(struct device *); 115 static void sd_rescan(struct device *); 116 static int sd_init_command(struct scsi_cmnd *SCpnt); 117 static void sd_uninit_command(struct scsi_cmnd *SCpnt); 118 static int sd_done(struct scsi_cmnd *); 119 static void sd_eh_reset(struct scsi_cmnd *); 120 static int sd_eh_action(struct scsi_cmnd *, int); 121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer); 122 static void scsi_disk_release(struct device *cdev); 123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *); 124 static void sd_print_result(const struct scsi_disk *, const char *, int); 125 126 static DEFINE_SPINLOCK(sd_index_lock); 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; 136 137 static const char *sd_cache_types[] = { 138 "write through", "none", "write back", 139 "write back, no read (daft)" 140 }; 141 142 static void sd_set_flush_flag(struct scsi_disk *sdkp) 143 { 144 bool wc = false, fua = false; 145 146 if (sdkp->WCE) { 147 wc = true; 148 if (sdkp->DPOFUA) 149 fua = true; 150 } 151 152 blk_queue_write_cache(sdkp->disk->queue, wc, fua); 153 } 154 155 static ssize_t 156 cache_type_store(struct device *dev, struct device_attribute *attr, 157 const char *buf, size_t count) 158 { 159 int ct, rcd, wce, sp; 160 struct scsi_disk *sdkp = to_scsi_disk(dev); 161 struct scsi_device *sdp = sdkp->device; 162 char buffer[64]; 163 char *buffer_data; 164 struct scsi_mode_data data; 165 struct scsi_sense_hdr sshdr; 166 static const char temp[] = "temporary "; 167 int len; 168 169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 170 /* no cache control on RBC devices; theoretically they 171 * can do it, but there's probably so many exceptions 172 * it's not worth the risk */ 173 return -EINVAL; 174 175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) { 176 buf += sizeof(temp) - 1; 177 sdkp->cache_override = 1; 178 } else { 179 sdkp->cache_override = 0; 180 } 181 182 ct = sysfs_match_string(sd_cache_types, buf); 183 if (ct < 0) 184 return -EINVAL; 185 186 rcd = ct & 0x01 ? 1 : 0; 187 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0; 188 189 if (sdkp->cache_override) { 190 sdkp->WCE = wce; 191 sdkp->RCD = rcd; 192 sd_set_flush_flag(sdkp); 193 return count; 194 } 195 196 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT, 197 SD_MAX_RETRIES, &data, NULL)) 198 return -EINVAL; 199 len = min_t(size_t, sizeof(buffer), data.length - data.header_length - 200 data.block_descriptor_length); 201 buffer_data = buffer + data.header_length + 202 data.block_descriptor_length; 203 buffer_data[2] &= ~0x05; 204 buffer_data[2] |= wce << 2 | rcd; 205 sp = buffer_data[0] & 0x80 ? 1 : 0; 206 buffer_data[0] &= ~0x80; 207 208 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT, 209 SD_MAX_RETRIES, &data, &sshdr)) { 210 if (scsi_sense_valid(&sshdr)) 211 sd_print_sense_hdr(sdkp, &sshdr); 212 return -EINVAL; 213 } 214 revalidate_disk(sdkp->disk); 215 return count; 216 } 217 218 static ssize_t 219 manage_start_stop_show(struct device *dev, struct device_attribute *attr, 220 char *buf) 221 { 222 struct scsi_disk *sdkp = to_scsi_disk(dev); 223 struct scsi_device *sdp = sdkp->device; 224 225 return sprintf(buf, "%u\n", sdp->manage_start_stop); 226 } 227 228 static ssize_t 229 manage_start_stop_store(struct device *dev, struct device_attribute *attr, 230 const char *buf, size_t count) 231 { 232 struct scsi_disk *sdkp = to_scsi_disk(dev); 233 struct scsi_device *sdp = sdkp->device; 234 bool v; 235 236 if (!capable(CAP_SYS_ADMIN)) 237 return -EACCES; 238 239 if (kstrtobool(buf, &v)) 240 return -EINVAL; 241 242 sdp->manage_start_stop = v; 243 244 return count; 245 } 246 static DEVICE_ATTR_RW(manage_start_stop); 247 248 static ssize_t 249 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf) 250 { 251 struct scsi_disk *sdkp = to_scsi_disk(dev); 252 253 return sprintf(buf, "%u\n", sdkp->device->allow_restart); 254 } 255 256 static ssize_t 257 allow_restart_store(struct device *dev, struct device_attribute *attr, 258 const char *buf, size_t count) 259 { 260 bool v; 261 struct scsi_disk *sdkp = to_scsi_disk(dev); 262 struct scsi_device *sdp = sdkp->device; 263 264 if (!capable(CAP_SYS_ADMIN)) 265 return -EACCES; 266 267 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 268 return -EINVAL; 269 270 if (kstrtobool(buf, &v)) 271 return -EINVAL; 272 273 sdp->allow_restart = v; 274 275 return count; 276 } 277 static DEVICE_ATTR_RW(allow_restart); 278 279 static ssize_t 280 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf) 281 { 282 struct scsi_disk *sdkp = to_scsi_disk(dev); 283 int ct = sdkp->RCD + 2*sdkp->WCE; 284 285 return sprintf(buf, "%s\n", sd_cache_types[ct]); 286 } 287 static DEVICE_ATTR_RW(cache_type); 288 289 static ssize_t 290 FUA_show(struct device *dev, struct device_attribute *attr, char *buf) 291 { 292 struct scsi_disk *sdkp = to_scsi_disk(dev); 293 294 return sprintf(buf, "%u\n", sdkp->DPOFUA); 295 } 296 static DEVICE_ATTR_RO(FUA); 297 298 static ssize_t 299 protection_type_show(struct device *dev, struct device_attribute *attr, 300 char *buf) 301 { 302 struct scsi_disk *sdkp = to_scsi_disk(dev); 303 304 return sprintf(buf, "%u\n", sdkp->protection_type); 305 } 306 307 static ssize_t 308 protection_type_store(struct device *dev, struct device_attribute *attr, 309 const char *buf, size_t count) 310 { 311 struct scsi_disk *sdkp = to_scsi_disk(dev); 312 unsigned int val; 313 int err; 314 315 if (!capable(CAP_SYS_ADMIN)) 316 return -EACCES; 317 318 err = kstrtouint(buf, 10, &val); 319 320 if (err) 321 return err; 322 323 if (val <= T10_PI_TYPE3_PROTECTION) 324 sdkp->protection_type = val; 325 326 return count; 327 } 328 static DEVICE_ATTR_RW(protection_type); 329 330 static ssize_t 331 protection_mode_show(struct device *dev, struct device_attribute *attr, 332 char *buf) 333 { 334 struct scsi_disk *sdkp = to_scsi_disk(dev); 335 struct scsi_device *sdp = sdkp->device; 336 unsigned int dif, dix; 337 338 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type); 339 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type); 340 341 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) { 342 dif = 0; 343 dix = 1; 344 } 345 346 if (!dif && !dix) 347 return sprintf(buf, "none\n"); 348 349 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif); 350 } 351 static DEVICE_ATTR_RO(protection_mode); 352 353 static ssize_t 354 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf) 355 { 356 struct scsi_disk *sdkp = to_scsi_disk(dev); 357 358 return sprintf(buf, "%u\n", sdkp->ATO); 359 } 360 static DEVICE_ATTR_RO(app_tag_own); 361 362 static ssize_t 363 thin_provisioning_show(struct device *dev, struct device_attribute *attr, 364 char *buf) 365 { 366 struct scsi_disk *sdkp = to_scsi_disk(dev); 367 368 return sprintf(buf, "%u\n", sdkp->lbpme); 369 } 370 static DEVICE_ATTR_RO(thin_provisioning); 371 372 /* sysfs_match_string() requires dense arrays */ 373 static const char *lbp_mode[] = { 374 [SD_LBP_FULL] = "full", 375 [SD_LBP_UNMAP] = "unmap", 376 [SD_LBP_WS16] = "writesame_16", 377 [SD_LBP_WS10] = "writesame_10", 378 [SD_LBP_ZERO] = "writesame_zero", 379 [SD_LBP_DISABLE] = "disabled", 380 }; 381 382 static ssize_t 383 provisioning_mode_show(struct device *dev, struct device_attribute *attr, 384 char *buf) 385 { 386 struct scsi_disk *sdkp = to_scsi_disk(dev); 387 388 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]); 389 } 390 391 static ssize_t 392 provisioning_mode_store(struct device *dev, struct device_attribute *attr, 393 const char *buf, size_t count) 394 { 395 struct scsi_disk *sdkp = to_scsi_disk(dev); 396 struct scsi_device *sdp = sdkp->device; 397 int mode; 398 399 if (!capable(CAP_SYS_ADMIN)) 400 return -EACCES; 401 402 if (sd_is_zoned(sdkp)) { 403 sd_config_discard(sdkp, SD_LBP_DISABLE); 404 return count; 405 } 406 407 if (sdp->type != TYPE_DISK) 408 return -EINVAL; 409 410 mode = sysfs_match_string(lbp_mode, buf); 411 if (mode < 0) 412 return -EINVAL; 413 414 sd_config_discard(sdkp, mode); 415 416 return count; 417 } 418 static DEVICE_ATTR_RW(provisioning_mode); 419 420 /* sysfs_match_string() requires dense arrays */ 421 static const char *zeroing_mode[] = { 422 [SD_ZERO_WRITE] = "write", 423 [SD_ZERO_WS] = "writesame", 424 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap", 425 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap", 426 }; 427 428 static ssize_t 429 zeroing_mode_show(struct device *dev, struct device_attribute *attr, 430 char *buf) 431 { 432 struct scsi_disk *sdkp = to_scsi_disk(dev); 433 434 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]); 435 } 436 437 static ssize_t 438 zeroing_mode_store(struct device *dev, struct device_attribute *attr, 439 const char *buf, size_t count) 440 { 441 struct scsi_disk *sdkp = to_scsi_disk(dev); 442 int mode; 443 444 if (!capable(CAP_SYS_ADMIN)) 445 return -EACCES; 446 447 mode = sysfs_match_string(zeroing_mode, buf); 448 if (mode < 0) 449 return -EINVAL; 450 451 sdkp->zeroing_mode = mode; 452 453 return count; 454 } 455 static DEVICE_ATTR_RW(zeroing_mode); 456 457 static ssize_t 458 max_medium_access_timeouts_show(struct device *dev, 459 struct device_attribute *attr, char *buf) 460 { 461 struct scsi_disk *sdkp = to_scsi_disk(dev); 462 463 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts); 464 } 465 466 static ssize_t 467 max_medium_access_timeouts_store(struct device *dev, 468 struct device_attribute *attr, const char *buf, 469 size_t count) 470 { 471 struct scsi_disk *sdkp = to_scsi_disk(dev); 472 int err; 473 474 if (!capable(CAP_SYS_ADMIN)) 475 return -EACCES; 476 477 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts); 478 479 return err ? err : count; 480 } 481 static DEVICE_ATTR_RW(max_medium_access_timeouts); 482 483 static ssize_t 484 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr, 485 char *buf) 486 { 487 struct scsi_disk *sdkp = to_scsi_disk(dev); 488 489 return sprintf(buf, "%u\n", sdkp->max_ws_blocks); 490 } 491 492 static ssize_t 493 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr, 494 const char *buf, size_t count) 495 { 496 struct scsi_disk *sdkp = to_scsi_disk(dev); 497 struct scsi_device *sdp = sdkp->device; 498 unsigned long max; 499 int err; 500 501 if (!capable(CAP_SYS_ADMIN)) 502 return -EACCES; 503 504 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 505 return -EINVAL; 506 507 err = kstrtoul(buf, 10, &max); 508 509 if (err) 510 return err; 511 512 if (max == 0) 513 sdp->no_write_same = 1; 514 else if (max <= SD_MAX_WS16_BLOCKS) { 515 sdp->no_write_same = 0; 516 sdkp->max_ws_blocks = max; 517 } 518 519 sd_config_write_same(sdkp); 520 521 return count; 522 } 523 static DEVICE_ATTR_RW(max_write_same_blocks); 524 525 static struct attribute *sd_disk_attrs[] = { 526 &dev_attr_cache_type.attr, 527 &dev_attr_FUA.attr, 528 &dev_attr_allow_restart.attr, 529 &dev_attr_manage_start_stop.attr, 530 &dev_attr_protection_type.attr, 531 &dev_attr_protection_mode.attr, 532 &dev_attr_app_tag_own.attr, 533 &dev_attr_thin_provisioning.attr, 534 &dev_attr_provisioning_mode.attr, 535 &dev_attr_zeroing_mode.attr, 536 &dev_attr_max_write_same_blocks.attr, 537 &dev_attr_max_medium_access_timeouts.attr, 538 NULL, 539 }; 540 ATTRIBUTE_GROUPS(sd_disk); 541 542 static struct class sd_disk_class = { 543 .name = "scsi_disk", 544 .owner = THIS_MODULE, 545 .dev_release = scsi_disk_release, 546 .dev_groups = sd_disk_groups, 547 }; 548 549 static const struct dev_pm_ops sd_pm_ops = { 550 .suspend = sd_suspend_system, 551 .resume = sd_resume, 552 .poweroff = sd_suspend_system, 553 .restore = sd_resume, 554 .runtime_suspend = sd_suspend_runtime, 555 .runtime_resume = sd_resume, 556 }; 557 558 static struct scsi_driver sd_template = { 559 .gendrv = { 560 .name = "sd", 561 .owner = THIS_MODULE, 562 .probe = sd_probe, 563 .remove = sd_remove, 564 .shutdown = sd_shutdown, 565 .pm = &sd_pm_ops, 566 }, 567 .rescan = sd_rescan, 568 .init_command = sd_init_command, 569 .uninit_command = sd_uninit_command, 570 .done = sd_done, 571 .eh_action = sd_eh_action, 572 .eh_reset = sd_eh_reset, 573 }; 574 575 /* 576 * Dummy kobj_map->probe function. 577 * The default ->probe function will call modprobe, which is 578 * pointless as this module is already loaded. 579 */ 580 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data) 581 { 582 return NULL; 583 } 584 585 /* 586 * Device no to disk mapping: 587 * 588 * major disc2 disc p1 589 * |............|.............|....|....| <- dev_t 590 * 31 20 19 8 7 4 3 0 591 * 592 * Inside a major, we have 16k disks, however mapped non- 593 * contiguously. The first 16 disks are for major0, the next 594 * ones with major1, ... Disk 256 is for major0 again, disk 272 595 * for major1, ... 596 * As we stay compatible with our numbering scheme, we can reuse 597 * the well-know SCSI majors 8, 65--71, 136--143. 598 */ 599 static int sd_major(int major_idx) 600 { 601 switch (major_idx) { 602 case 0: 603 return SCSI_DISK0_MAJOR; 604 case 1 ... 7: 605 return SCSI_DISK1_MAJOR + major_idx - 1; 606 case 8 ... 15: 607 return SCSI_DISK8_MAJOR + major_idx - 8; 608 default: 609 BUG(); 610 return 0; /* shut up gcc */ 611 } 612 } 613 614 static struct scsi_disk *scsi_disk_get(struct gendisk *disk) 615 { 616 struct scsi_disk *sdkp = NULL; 617 618 mutex_lock(&sd_ref_mutex); 619 620 if (disk->private_data) { 621 sdkp = scsi_disk(disk); 622 if (scsi_device_get(sdkp->device) == 0) 623 get_device(&sdkp->dev); 624 else 625 sdkp = NULL; 626 } 627 mutex_unlock(&sd_ref_mutex); 628 return sdkp; 629 } 630 631 static void scsi_disk_put(struct scsi_disk *sdkp) 632 { 633 struct scsi_device *sdev = sdkp->device; 634 635 mutex_lock(&sd_ref_mutex); 636 put_device(&sdkp->dev); 637 scsi_device_put(sdev); 638 mutex_unlock(&sd_ref_mutex); 639 } 640 641 #ifdef CONFIG_BLK_SED_OPAL 642 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, 643 size_t len, bool send) 644 { 645 struct scsi_device *sdev = data; 646 u8 cdb[12] = { 0, }; 647 int ret; 648 649 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN; 650 cdb[1] = secp; 651 put_unaligned_be16(spsp, &cdb[2]); 652 put_unaligned_be32(len, &cdb[6]); 653 654 ret = scsi_execute_req(sdev, cdb, 655 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE, 656 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL); 657 return ret <= 0 ? ret : -EIO; 658 } 659 #endif /* CONFIG_BLK_SED_OPAL */ 660 661 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd, 662 unsigned int dix, unsigned int dif) 663 { 664 struct bio *bio = scmd->request->bio; 665 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif); 666 unsigned int protect = 0; 667 668 if (dix) { /* DIX Type 0, 1, 2, 3 */ 669 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM)) 670 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM; 671 672 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false) 673 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK; 674 } 675 676 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */ 677 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT; 678 679 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false) 680 scmd->prot_flags |= SCSI_PROT_REF_CHECK; 681 } 682 683 if (dif) { /* DIX/DIF Type 1, 2, 3 */ 684 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI; 685 686 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK)) 687 protect = 3 << 5; /* Disable target PI checking */ 688 else 689 protect = 1 << 5; /* Enable target PI checking */ 690 } 691 692 scsi_set_prot_op(scmd, prot_op); 693 scsi_set_prot_type(scmd, dif); 694 scmd->prot_flags &= sd_prot_flag_mask(prot_op); 695 696 return protect; 697 } 698 699 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode) 700 { 701 struct request_queue *q = sdkp->disk->queue; 702 unsigned int logical_block_size = sdkp->device->sector_size; 703 unsigned int max_blocks = 0; 704 705 q->limits.discard_alignment = 706 sdkp->unmap_alignment * logical_block_size; 707 q->limits.discard_granularity = 708 max(sdkp->physical_block_size, 709 sdkp->unmap_granularity * logical_block_size); 710 sdkp->provisioning_mode = mode; 711 712 switch (mode) { 713 714 case SD_LBP_FULL: 715 case SD_LBP_DISABLE: 716 blk_queue_max_discard_sectors(q, 0); 717 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 718 return; 719 720 case SD_LBP_UNMAP: 721 max_blocks = min_not_zero(sdkp->max_unmap_blocks, 722 (u32)SD_MAX_WS16_BLOCKS); 723 break; 724 725 case SD_LBP_WS16: 726 if (sdkp->device->unmap_limit_for_ws) 727 max_blocks = sdkp->max_unmap_blocks; 728 else 729 max_blocks = sdkp->max_ws_blocks; 730 731 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS); 732 break; 733 734 case SD_LBP_WS10: 735 if (sdkp->device->unmap_limit_for_ws) 736 max_blocks = sdkp->max_unmap_blocks; 737 else 738 max_blocks = sdkp->max_ws_blocks; 739 740 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS); 741 break; 742 743 case SD_LBP_ZERO: 744 max_blocks = min_not_zero(sdkp->max_ws_blocks, 745 (u32)SD_MAX_WS10_BLOCKS); 746 break; 747 } 748 749 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9)); 750 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 751 } 752 753 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd) 754 { 755 struct scsi_device *sdp = cmd->device; 756 struct request *rq = cmd->request; 757 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9); 758 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9); 759 unsigned int data_len = 24; 760 char *buf; 761 762 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO); 763 if (!rq->special_vec.bv_page) 764 return BLKPREP_DEFER; 765 rq->special_vec.bv_offset = 0; 766 rq->special_vec.bv_len = data_len; 767 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 768 769 cmd->cmd_len = 10; 770 cmd->cmnd[0] = UNMAP; 771 cmd->cmnd[8] = 24; 772 773 buf = page_address(rq->special_vec.bv_page); 774 put_unaligned_be16(6 + 16, &buf[0]); 775 put_unaligned_be16(16, &buf[2]); 776 put_unaligned_be64(sector, &buf[8]); 777 put_unaligned_be32(nr_sectors, &buf[16]); 778 779 cmd->allowed = SD_MAX_RETRIES; 780 cmd->transfersize = data_len; 781 rq->timeout = SD_TIMEOUT; 782 scsi_req(rq)->resid_len = data_len; 783 784 return scsi_init_io(cmd); 785 } 786 787 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap) 788 { 789 struct scsi_device *sdp = cmd->device; 790 struct request *rq = cmd->request; 791 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9); 792 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9); 793 u32 data_len = sdp->sector_size; 794 795 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO); 796 if (!rq->special_vec.bv_page) 797 return BLKPREP_DEFER; 798 rq->special_vec.bv_offset = 0; 799 rq->special_vec.bv_len = data_len; 800 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 801 802 cmd->cmd_len = 16; 803 cmd->cmnd[0] = WRITE_SAME_16; 804 if (unmap) 805 cmd->cmnd[1] = 0x8; /* UNMAP */ 806 put_unaligned_be64(sector, &cmd->cmnd[2]); 807 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]); 808 809 cmd->allowed = SD_MAX_RETRIES; 810 cmd->transfersize = data_len; 811 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT; 812 scsi_req(rq)->resid_len = data_len; 813 814 return scsi_init_io(cmd); 815 } 816 817 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap) 818 { 819 struct scsi_device *sdp = cmd->device; 820 struct request *rq = cmd->request; 821 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9); 822 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9); 823 u32 data_len = sdp->sector_size; 824 825 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO); 826 if (!rq->special_vec.bv_page) 827 return BLKPREP_DEFER; 828 rq->special_vec.bv_offset = 0; 829 rq->special_vec.bv_len = data_len; 830 rq->rq_flags |= RQF_SPECIAL_PAYLOAD; 831 832 cmd->cmd_len = 10; 833 cmd->cmnd[0] = WRITE_SAME; 834 if (unmap) 835 cmd->cmnd[1] = 0x8; /* UNMAP */ 836 put_unaligned_be32(sector, &cmd->cmnd[2]); 837 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]); 838 839 cmd->allowed = SD_MAX_RETRIES; 840 cmd->transfersize = data_len; 841 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT; 842 scsi_req(rq)->resid_len = data_len; 843 844 return scsi_init_io(cmd); 845 } 846 847 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd) 848 { 849 struct request *rq = cmd->request; 850 struct scsi_device *sdp = cmd->device; 851 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk); 852 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9); 853 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9); 854 int ret; 855 856 if (!(rq->cmd_flags & REQ_NOUNMAP)) { 857 switch (sdkp->zeroing_mode) { 858 case SD_ZERO_WS16_UNMAP: 859 ret = sd_setup_write_same16_cmnd(cmd, true); 860 goto out; 861 case SD_ZERO_WS10_UNMAP: 862 ret = sd_setup_write_same10_cmnd(cmd, true); 863 goto out; 864 } 865 } 866 867 if (sdp->no_write_same) 868 return BLKPREP_INVALID; 869 870 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) 871 ret = sd_setup_write_same16_cmnd(cmd, false); 872 else 873 ret = sd_setup_write_same10_cmnd(cmd, false); 874 875 out: 876 if (sd_is_zoned(sdkp) && ret == BLKPREP_OK) 877 return sd_zbc_write_lock_zone(cmd); 878 879 return ret; 880 } 881 882 static void sd_config_write_same(struct scsi_disk *sdkp) 883 { 884 struct request_queue *q = sdkp->disk->queue; 885 unsigned int logical_block_size = sdkp->device->sector_size; 886 887 if (sdkp->device->no_write_same) { 888 sdkp->max_ws_blocks = 0; 889 goto out; 890 } 891 892 /* Some devices can not handle block counts above 0xffff despite 893 * supporting WRITE SAME(16). Consequently we default to 64k 894 * blocks per I/O unless the device explicitly advertises a 895 * bigger limit. 896 */ 897 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS) 898 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks, 899 (u32)SD_MAX_WS16_BLOCKS); 900 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes) 901 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks, 902 (u32)SD_MAX_WS10_BLOCKS); 903 else { 904 sdkp->device->no_write_same = 1; 905 sdkp->max_ws_blocks = 0; 906 } 907 908 if (sdkp->lbprz && sdkp->lbpws) 909 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP; 910 else if (sdkp->lbprz && sdkp->lbpws10) 911 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP; 912 else if (sdkp->max_ws_blocks) 913 sdkp->zeroing_mode = SD_ZERO_WS; 914 else 915 sdkp->zeroing_mode = SD_ZERO_WRITE; 916 917 if (sdkp->max_ws_blocks && 918 sdkp->physical_block_size > logical_block_size) { 919 /* 920 * Reporting a maximum number of blocks that is not aligned 921 * on the device physical size would cause a large write same 922 * request to be split into physically unaligned chunks by 923 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same() 924 * even if the caller of these functions took care to align the 925 * large request. So make sure the maximum reported is aligned 926 * to the device physical block size. This is only an optional 927 * optimization for regular disks, but this is mandatory to 928 * avoid failure of large write same requests directed at 929 * sequential write required zones of host-managed ZBC disks. 930 */ 931 sdkp->max_ws_blocks = 932 round_down(sdkp->max_ws_blocks, 933 bytes_to_logical(sdkp->device, 934 sdkp->physical_block_size)); 935 } 936 937 out: 938 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks * 939 (logical_block_size >> 9)); 940 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks * 941 (logical_block_size >> 9)); 942 } 943 944 /** 945 * sd_setup_write_same_cmnd - write the same data to multiple blocks 946 * @cmd: command to prepare 947 * 948 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on 949 * the preference indicated by the target device. 950 **/ 951 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd) 952 { 953 struct request *rq = cmd->request; 954 struct scsi_device *sdp = cmd->device; 955 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk); 956 struct bio *bio = rq->bio; 957 sector_t sector = blk_rq_pos(rq); 958 unsigned int nr_sectors = blk_rq_sectors(rq); 959 unsigned int nr_bytes = blk_rq_bytes(rq); 960 int ret; 961 962 if (sdkp->device->no_write_same) 963 return BLKPREP_INVALID; 964 965 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size); 966 967 if (sd_is_zoned(sdkp)) { 968 ret = sd_zbc_write_lock_zone(cmd); 969 if (ret != BLKPREP_OK) 970 return ret; 971 } 972 973 sector >>= ilog2(sdp->sector_size) - 9; 974 nr_sectors >>= ilog2(sdp->sector_size) - 9; 975 976 rq->timeout = SD_WRITE_SAME_TIMEOUT; 977 978 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) { 979 cmd->cmd_len = 16; 980 cmd->cmnd[0] = WRITE_SAME_16; 981 put_unaligned_be64(sector, &cmd->cmnd[2]); 982 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]); 983 } else { 984 cmd->cmd_len = 10; 985 cmd->cmnd[0] = WRITE_SAME; 986 put_unaligned_be32(sector, &cmd->cmnd[2]); 987 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]); 988 } 989 990 cmd->transfersize = sdp->sector_size; 991 cmd->allowed = SD_MAX_RETRIES; 992 993 /* 994 * For WRITE SAME the data transferred via the DATA OUT buffer is 995 * different from the amount of data actually written to the target. 996 * 997 * We set up __data_len to the amount of data transferred via the 998 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list 999 * to transfer a single sector of data first, but then reset it to 1000 * the amount of data to be written right after so that the I/O path 1001 * knows how much to actually write. 1002 */ 1003 rq->__data_len = sdp->sector_size; 1004 ret = scsi_init_io(cmd); 1005 rq->__data_len = nr_bytes; 1006 1007 if (sd_is_zoned(sdkp) && ret != BLKPREP_OK) 1008 sd_zbc_write_unlock_zone(cmd); 1009 1010 return ret; 1011 } 1012 1013 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd) 1014 { 1015 struct request *rq = cmd->request; 1016 1017 /* flush requests don't perform I/O, zero the S/G table */ 1018 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1019 1020 cmd->cmnd[0] = SYNCHRONIZE_CACHE; 1021 cmd->cmd_len = 10; 1022 cmd->transfersize = 0; 1023 cmd->allowed = SD_MAX_RETRIES; 1024 1025 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER; 1026 return BLKPREP_OK; 1027 } 1028 1029 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt) 1030 { 1031 struct request *rq = SCpnt->request; 1032 struct scsi_device *sdp = SCpnt->device; 1033 struct gendisk *disk = rq->rq_disk; 1034 struct scsi_disk *sdkp = scsi_disk(disk); 1035 sector_t block = blk_rq_pos(rq); 1036 sector_t threshold; 1037 unsigned int this_count = blk_rq_sectors(rq); 1038 unsigned int dif, dix; 1039 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE; 1040 int ret; 1041 unsigned char protect; 1042 1043 if (zoned_write) { 1044 ret = sd_zbc_write_lock_zone(SCpnt); 1045 if (ret != BLKPREP_OK) 1046 return ret; 1047 } 1048 1049 ret = scsi_init_io(SCpnt); 1050 if (ret != BLKPREP_OK) 1051 goto out; 1052 WARN_ON_ONCE(SCpnt != rq->special); 1053 1054 /* from here on until we're complete, any goto out 1055 * is used for a killable error condition */ 1056 ret = BLKPREP_KILL; 1057 1058 SCSI_LOG_HLQUEUE(1, 1059 scmd_printk(KERN_INFO, SCpnt, 1060 "%s: block=%llu, count=%d\n", 1061 __func__, (unsigned long long)block, this_count)); 1062 1063 if (!sdp || !scsi_device_online(sdp) || 1064 block + blk_rq_sectors(rq) > get_capacity(disk)) { 1065 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, 1066 "Finishing %u sectors\n", 1067 blk_rq_sectors(rq))); 1068 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, 1069 "Retry with 0x%p\n", SCpnt)); 1070 goto out; 1071 } 1072 1073 if (sdp->changed) { 1074 /* 1075 * quietly refuse to do anything to a changed disc until 1076 * the changed bit has been reset 1077 */ 1078 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */ 1079 goto out; 1080 } 1081 1082 /* 1083 * Some SD card readers can't handle multi-sector accesses which touch 1084 * the last one or two hardware sectors. Split accesses as needed. 1085 */ 1086 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS * 1087 (sdp->sector_size / 512); 1088 1089 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) { 1090 if (block < threshold) { 1091 /* Access up to the threshold but not beyond */ 1092 this_count = threshold - block; 1093 } else { 1094 /* Access only a single hardware sector */ 1095 this_count = sdp->sector_size / 512; 1096 } 1097 } 1098 1099 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n", 1100 (unsigned long long)block)); 1101 1102 /* 1103 * If we have a 1K hardware sectorsize, prevent access to single 1104 * 512 byte sectors. In theory we could handle this - in fact 1105 * the scsi cdrom driver must be able to handle this because 1106 * we typically use 1K blocksizes, and cdroms typically have 1107 * 2K hardware sectorsizes. Of course, things are simpler 1108 * with the cdrom, since it is read-only. For performance 1109 * reasons, the filesystems should be able to handle this 1110 * and not force the scsi disk driver to use bounce buffers 1111 * for this. 1112 */ 1113 if (sdp->sector_size == 1024) { 1114 if ((block & 1) || (blk_rq_sectors(rq) & 1)) { 1115 scmd_printk(KERN_ERR, SCpnt, 1116 "Bad block number requested\n"); 1117 goto out; 1118 } else { 1119 block = block >> 1; 1120 this_count = this_count >> 1; 1121 } 1122 } 1123 if (sdp->sector_size == 2048) { 1124 if ((block & 3) || (blk_rq_sectors(rq) & 3)) { 1125 scmd_printk(KERN_ERR, SCpnt, 1126 "Bad block number requested\n"); 1127 goto out; 1128 } else { 1129 block = block >> 2; 1130 this_count = this_count >> 2; 1131 } 1132 } 1133 if (sdp->sector_size == 4096) { 1134 if ((block & 7) || (blk_rq_sectors(rq) & 7)) { 1135 scmd_printk(KERN_ERR, SCpnt, 1136 "Bad block number requested\n"); 1137 goto out; 1138 } else { 1139 block = block >> 3; 1140 this_count = this_count >> 3; 1141 } 1142 } 1143 if (rq_data_dir(rq) == WRITE) { 1144 SCpnt->cmnd[0] = WRITE_6; 1145 1146 if (blk_integrity_rq(rq)) 1147 sd_dif_prepare(SCpnt); 1148 1149 } else if (rq_data_dir(rq) == READ) { 1150 SCpnt->cmnd[0] = READ_6; 1151 } else { 1152 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq)); 1153 goto out; 1154 } 1155 1156 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, 1157 "%s %d/%u 512 byte blocks.\n", 1158 (rq_data_dir(rq) == WRITE) ? 1159 "writing" : "reading", this_count, 1160 blk_rq_sectors(rq))); 1161 1162 dix = scsi_prot_sg_count(SCpnt); 1163 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type); 1164 1165 if (dif || dix) 1166 protect = sd_setup_protect_cmnd(SCpnt, dix, dif); 1167 else 1168 protect = 0; 1169 1170 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) { 1171 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC); 1172 1173 if (unlikely(SCpnt->cmnd == NULL)) { 1174 ret = BLKPREP_DEFER; 1175 goto out; 1176 } 1177 1178 SCpnt->cmd_len = SD_EXT_CDB_SIZE; 1179 memset(SCpnt->cmnd, 0, SCpnt->cmd_len); 1180 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD; 1181 SCpnt->cmnd[7] = 0x18; 1182 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32; 1183 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0); 1184 1185 /* LBA */ 1186 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0; 1187 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0; 1188 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0; 1189 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0; 1190 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff; 1191 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff; 1192 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff; 1193 SCpnt->cmnd[19] = (unsigned char) block & 0xff; 1194 1195 /* Expected Indirect LBA */ 1196 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff; 1197 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff; 1198 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff; 1199 SCpnt->cmnd[23] = (unsigned char) block & 0xff; 1200 1201 /* Transfer length */ 1202 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff; 1203 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff; 1204 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff; 1205 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff; 1206 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) { 1207 SCpnt->cmnd[0] += READ_16 - READ_6; 1208 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0); 1209 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0; 1210 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0; 1211 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0; 1212 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0; 1213 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff; 1214 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff; 1215 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff; 1216 SCpnt->cmnd[9] = (unsigned char) block & 0xff; 1217 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff; 1218 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff; 1219 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff; 1220 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff; 1221 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0; 1222 } else if ((this_count > 0xff) || (block > 0x1fffff) || 1223 scsi_device_protection(SCpnt->device) || 1224 SCpnt->device->use_10_for_rw) { 1225 SCpnt->cmnd[0] += READ_10 - READ_6; 1226 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0); 1227 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff; 1228 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff; 1229 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff; 1230 SCpnt->cmnd[5] = (unsigned char) block & 0xff; 1231 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0; 1232 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff; 1233 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff; 1234 } else { 1235 if (unlikely(rq->cmd_flags & REQ_FUA)) { 1236 /* 1237 * This happens only if this drive failed 1238 * 10byte rw command with ILLEGAL_REQUEST 1239 * during operation and thus turned off 1240 * use_10_for_rw. 1241 */ 1242 scmd_printk(KERN_ERR, SCpnt, 1243 "FUA write on READ/WRITE(6) drive\n"); 1244 goto out; 1245 } 1246 1247 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f); 1248 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff); 1249 SCpnt->cmnd[3] = (unsigned char) block & 0xff; 1250 SCpnt->cmnd[4] = (unsigned char) this_count; 1251 SCpnt->cmnd[5] = 0; 1252 } 1253 SCpnt->sdb.length = this_count * sdp->sector_size; 1254 1255 /* 1256 * We shouldn't disconnect in the middle of a sector, so with a dumb 1257 * host adapter, it's safe to assume that we can at least transfer 1258 * this many bytes between each connect / disconnect. 1259 */ 1260 SCpnt->transfersize = sdp->sector_size; 1261 SCpnt->underflow = this_count << 9; 1262 SCpnt->allowed = SD_MAX_RETRIES; 1263 1264 /* 1265 * This indicates that the command is ready from our end to be 1266 * queued. 1267 */ 1268 ret = BLKPREP_OK; 1269 out: 1270 if (zoned_write && ret != BLKPREP_OK) 1271 sd_zbc_write_unlock_zone(SCpnt); 1272 1273 return ret; 1274 } 1275 1276 static int sd_init_command(struct scsi_cmnd *cmd) 1277 { 1278 struct request *rq = cmd->request; 1279 1280 switch (req_op(rq)) { 1281 case REQ_OP_DISCARD: 1282 switch (scsi_disk(rq->rq_disk)->provisioning_mode) { 1283 case SD_LBP_UNMAP: 1284 return sd_setup_unmap_cmnd(cmd); 1285 case SD_LBP_WS16: 1286 return sd_setup_write_same16_cmnd(cmd, true); 1287 case SD_LBP_WS10: 1288 return sd_setup_write_same10_cmnd(cmd, true); 1289 case SD_LBP_ZERO: 1290 return sd_setup_write_same10_cmnd(cmd, false); 1291 default: 1292 return BLKPREP_INVALID; 1293 } 1294 case REQ_OP_WRITE_ZEROES: 1295 return sd_setup_write_zeroes_cmnd(cmd); 1296 case REQ_OP_WRITE_SAME: 1297 return sd_setup_write_same_cmnd(cmd); 1298 case REQ_OP_FLUSH: 1299 return sd_setup_flush_cmnd(cmd); 1300 case REQ_OP_READ: 1301 case REQ_OP_WRITE: 1302 return sd_setup_read_write_cmnd(cmd); 1303 case REQ_OP_ZONE_REPORT: 1304 return sd_zbc_setup_report_cmnd(cmd); 1305 case REQ_OP_ZONE_RESET: 1306 return sd_zbc_setup_reset_cmnd(cmd); 1307 default: 1308 BUG(); 1309 } 1310 } 1311 1312 static void sd_uninit_command(struct scsi_cmnd *SCpnt) 1313 { 1314 struct request *rq = SCpnt->request; 1315 u8 *cmnd; 1316 1317 if (SCpnt->flags & SCMD_ZONE_WRITE_LOCK) 1318 sd_zbc_write_unlock_zone(SCpnt); 1319 1320 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1321 __free_page(rq->special_vec.bv_page); 1322 1323 if (SCpnt->cmnd != scsi_req(rq)->cmd) { 1324 cmnd = SCpnt->cmnd; 1325 SCpnt->cmnd = NULL; 1326 SCpnt->cmd_len = 0; 1327 mempool_free(cmnd, sd_cdb_pool); 1328 } 1329 } 1330 1331 /** 1332 * sd_open - open a scsi disk device 1333 * @bdev: Block device of the scsi disk to open 1334 * @mode: FMODE_* mask 1335 * 1336 * Returns 0 if successful. Returns a negated errno value in case 1337 * of error. 1338 * 1339 * Note: This can be called from a user context (e.g. fsck(1) ) 1340 * or from within the kernel (e.g. as a result of a mount(1) ). 1341 * In the latter case @inode and @filp carry an abridged amount 1342 * of information as noted above. 1343 * 1344 * Locking: called with bdev->bd_mutex held. 1345 **/ 1346 static int sd_open(struct block_device *bdev, fmode_t mode) 1347 { 1348 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk); 1349 struct scsi_device *sdev; 1350 int retval; 1351 1352 if (!sdkp) 1353 return -ENXIO; 1354 1355 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n")); 1356 1357 sdev = sdkp->device; 1358 1359 /* 1360 * If the device is in error recovery, wait until it is done. 1361 * If the device is offline, then disallow any access to it. 1362 */ 1363 retval = -ENXIO; 1364 if (!scsi_block_when_processing_errors(sdev)) 1365 goto error_out; 1366 1367 if (sdev->removable || sdkp->write_prot) 1368 check_disk_change(bdev); 1369 1370 /* 1371 * If the drive is empty, just let the open fail. 1372 */ 1373 retval = -ENOMEDIUM; 1374 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY)) 1375 goto error_out; 1376 1377 /* 1378 * If the device has the write protect tab set, have the open fail 1379 * if the user expects to be able to write to the thing. 1380 */ 1381 retval = -EROFS; 1382 if (sdkp->write_prot && (mode & FMODE_WRITE)) 1383 goto error_out; 1384 1385 /* 1386 * It is possible that the disk changing stuff resulted in 1387 * the device being taken offline. If this is the case, 1388 * report this to the user, and don't pretend that the 1389 * open actually succeeded. 1390 */ 1391 retval = -ENXIO; 1392 if (!scsi_device_online(sdev)) 1393 goto error_out; 1394 1395 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) { 1396 if (scsi_block_when_processing_errors(sdev)) 1397 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT); 1398 } 1399 1400 return 0; 1401 1402 error_out: 1403 scsi_disk_put(sdkp); 1404 return retval; 1405 } 1406 1407 /** 1408 * sd_release - invoked when the (last) close(2) is called on this 1409 * scsi disk. 1410 * @disk: disk to release 1411 * @mode: FMODE_* mask 1412 * 1413 * Returns 0. 1414 * 1415 * Note: may block (uninterruptible) if error recovery is underway 1416 * on this disk. 1417 * 1418 * Locking: called with bdev->bd_mutex held. 1419 **/ 1420 static void sd_release(struct gendisk *disk, fmode_t mode) 1421 { 1422 struct scsi_disk *sdkp = scsi_disk(disk); 1423 struct scsi_device *sdev = sdkp->device; 1424 1425 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n")); 1426 1427 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) { 1428 if (scsi_block_when_processing_errors(sdev)) 1429 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW); 1430 } 1431 1432 /* 1433 * XXX and what if there are packets in flight and this close() 1434 * XXX is followed by a "rmmod sd_mod"? 1435 */ 1436 1437 scsi_disk_put(sdkp); 1438 } 1439 1440 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo) 1441 { 1442 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk); 1443 struct scsi_device *sdp = sdkp->device; 1444 struct Scsi_Host *host = sdp->host; 1445 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity); 1446 int diskinfo[4]; 1447 1448 /* default to most commonly used values */ 1449 diskinfo[0] = 0x40; /* 1 << 6 */ 1450 diskinfo[1] = 0x20; /* 1 << 5 */ 1451 diskinfo[2] = capacity >> 11; 1452 1453 /* override with calculated, extended default, or driver values */ 1454 if (host->hostt->bios_param) 1455 host->hostt->bios_param(sdp, bdev, capacity, diskinfo); 1456 else 1457 scsicam_bios_param(bdev, capacity, diskinfo); 1458 1459 geo->heads = diskinfo[0]; 1460 geo->sectors = diskinfo[1]; 1461 geo->cylinders = diskinfo[2]; 1462 return 0; 1463 } 1464 1465 /** 1466 * sd_ioctl - process an ioctl 1467 * @bdev: target block device 1468 * @mode: FMODE_* mask 1469 * @cmd: ioctl command number 1470 * @arg: this is third argument given to ioctl(2) system call. 1471 * Often contains a pointer. 1472 * 1473 * Returns 0 if successful (some ioctls return positive numbers on 1474 * success as well). Returns a negated errno value in case of error. 1475 * 1476 * Note: most ioctls are forward onto the block subsystem or further 1477 * down in the scsi subsystem. 1478 **/ 1479 static int sd_ioctl(struct block_device *bdev, fmode_t mode, 1480 unsigned int cmd, unsigned long arg) 1481 { 1482 struct gendisk *disk = bdev->bd_disk; 1483 struct scsi_disk *sdkp = scsi_disk(disk); 1484 struct scsi_device *sdp = sdkp->device; 1485 void __user *p = (void __user *)arg; 1486 int error; 1487 1488 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, " 1489 "cmd=0x%x\n", disk->disk_name, cmd)); 1490 1491 error = scsi_verify_blk_ioctl(bdev, cmd); 1492 if (error < 0) 1493 return error; 1494 1495 /* 1496 * If we are in the middle of error recovery, don't let anyone 1497 * else try and use this device. Also, if error recovery fails, it 1498 * may try and take the device offline, in which case all further 1499 * access to the device is prohibited. 1500 */ 1501 error = scsi_ioctl_block_when_processing_errors(sdp, cmd, 1502 (mode & FMODE_NDELAY) != 0); 1503 if (error) 1504 goto out; 1505 1506 if (is_sed_ioctl(cmd)) 1507 return sed_ioctl(sdkp->opal_dev, cmd, p); 1508 1509 /* 1510 * Send SCSI addressing ioctls directly to mid level, send other 1511 * ioctls to block level and then onto mid level if they can't be 1512 * resolved. 1513 */ 1514 switch (cmd) { 1515 case SCSI_IOCTL_GET_IDLUN: 1516 case SCSI_IOCTL_GET_BUS_NUMBER: 1517 error = scsi_ioctl(sdp, cmd, p); 1518 break; 1519 default: 1520 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p); 1521 if (error != -ENOTTY) 1522 break; 1523 error = scsi_ioctl(sdp, cmd, p); 1524 break; 1525 } 1526 out: 1527 return error; 1528 } 1529 1530 static void set_media_not_present(struct scsi_disk *sdkp) 1531 { 1532 if (sdkp->media_present) 1533 sdkp->device->changed = 1; 1534 1535 if (sdkp->device->removable) { 1536 sdkp->media_present = 0; 1537 sdkp->capacity = 0; 1538 } 1539 } 1540 1541 static int media_not_present(struct scsi_disk *sdkp, 1542 struct scsi_sense_hdr *sshdr) 1543 { 1544 if (!scsi_sense_valid(sshdr)) 1545 return 0; 1546 1547 /* not invoked for commands that could return deferred errors */ 1548 switch (sshdr->sense_key) { 1549 case UNIT_ATTENTION: 1550 case NOT_READY: 1551 /* medium not present */ 1552 if (sshdr->asc == 0x3A) { 1553 set_media_not_present(sdkp); 1554 return 1; 1555 } 1556 } 1557 return 0; 1558 } 1559 1560 /** 1561 * sd_check_events - check media events 1562 * @disk: kernel device descriptor 1563 * @clearing: disk events currently being cleared 1564 * 1565 * Returns mask of DISK_EVENT_*. 1566 * 1567 * Note: this function is invoked from the block subsystem. 1568 **/ 1569 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing) 1570 { 1571 struct scsi_disk *sdkp = scsi_disk_get(disk); 1572 struct scsi_device *sdp; 1573 int retval; 1574 1575 if (!sdkp) 1576 return 0; 1577 1578 sdp = sdkp->device; 1579 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n")); 1580 1581 /* 1582 * If the device is offline, don't send any commands - just pretend as 1583 * if the command failed. If the device ever comes back online, we 1584 * can deal with it then. It is only because of unrecoverable errors 1585 * that we would ever take a device offline in the first place. 1586 */ 1587 if (!scsi_device_online(sdp)) { 1588 set_media_not_present(sdkp); 1589 goto out; 1590 } 1591 1592 /* 1593 * Using TEST_UNIT_READY enables differentiation between drive with 1594 * no cartridge loaded - NOT READY, drive with changed cartridge - 1595 * UNIT ATTENTION, or with same cartridge - GOOD STATUS. 1596 * 1597 * Drives that auto spin down. eg iomega jaz 1G, will be started 1598 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever 1599 * sd_revalidate() is called. 1600 */ 1601 if (scsi_block_when_processing_errors(sdp)) { 1602 struct scsi_sense_hdr sshdr = { 0, }; 1603 1604 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES, 1605 &sshdr); 1606 1607 /* failed to execute TUR, assume media not present */ 1608 if (host_byte(retval)) { 1609 set_media_not_present(sdkp); 1610 goto out; 1611 } 1612 1613 if (media_not_present(sdkp, &sshdr)) 1614 goto out; 1615 } 1616 1617 /* 1618 * For removable scsi disk we have to recognise the presence 1619 * of a disk in the drive. 1620 */ 1621 if (!sdkp->media_present) 1622 sdp->changed = 1; 1623 sdkp->media_present = 1; 1624 out: 1625 /* 1626 * sdp->changed is set under the following conditions: 1627 * 1628 * Medium present state has changed in either direction. 1629 * Device has indicated UNIT_ATTENTION. 1630 */ 1631 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0; 1632 sdp->changed = 0; 1633 scsi_disk_put(sdkp); 1634 return retval; 1635 } 1636 1637 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr) 1638 { 1639 int retries, res; 1640 struct scsi_device *sdp = sdkp->device; 1641 const int timeout = sdp->request_queue->rq_timeout 1642 * SD_FLUSH_TIMEOUT_MULTIPLIER; 1643 struct scsi_sense_hdr my_sshdr; 1644 1645 if (!scsi_device_online(sdp)) 1646 return -ENODEV; 1647 1648 /* caller might not be interested in sense, but we need it */ 1649 if (!sshdr) 1650 sshdr = &my_sshdr; 1651 1652 for (retries = 3; retries > 0; --retries) { 1653 unsigned char cmd[10] = { 0 }; 1654 1655 cmd[0] = SYNCHRONIZE_CACHE; 1656 /* 1657 * Leave the rest of the command zero to indicate 1658 * flush everything. 1659 */ 1660 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr, 1661 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL); 1662 if (res == 0) 1663 break; 1664 } 1665 1666 if (res) { 1667 sd_print_result(sdkp, "Synchronize Cache(10) failed", res); 1668 1669 if (driver_byte(res) & DRIVER_SENSE) 1670 sd_print_sense_hdr(sdkp, sshdr); 1671 1672 /* we need to evaluate the error return */ 1673 if (scsi_sense_valid(sshdr) && 1674 (sshdr->asc == 0x3a || /* medium not present */ 1675 sshdr->asc == 0x20)) /* invalid command */ 1676 /* this is no error here */ 1677 return 0; 1678 1679 switch (host_byte(res)) { 1680 /* ignore errors due to racing a disconnection */ 1681 case DID_BAD_TARGET: 1682 case DID_NO_CONNECT: 1683 return 0; 1684 /* signal the upper layer it might try again */ 1685 case DID_BUS_BUSY: 1686 case DID_IMM_RETRY: 1687 case DID_REQUEUE: 1688 case DID_SOFT_ERROR: 1689 return -EBUSY; 1690 default: 1691 return -EIO; 1692 } 1693 } 1694 return 0; 1695 } 1696 1697 static void sd_rescan(struct device *dev) 1698 { 1699 struct scsi_disk *sdkp = dev_get_drvdata(dev); 1700 1701 revalidate_disk(sdkp->disk); 1702 } 1703 1704 1705 #ifdef CONFIG_COMPAT 1706 /* 1707 * This gets directly called from VFS. When the ioctl 1708 * is not recognized we go back to the other translation paths. 1709 */ 1710 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode, 1711 unsigned int cmd, unsigned long arg) 1712 { 1713 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device; 1714 int error; 1715 1716 error = scsi_ioctl_block_when_processing_errors(sdev, cmd, 1717 (mode & FMODE_NDELAY) != 0); 1718 if (error) 1719 return error; 1720 1721 /* 1722 * Let the static ioctl translation table take care of it. 1723 */ 1724 if (!sdev->host->hostt->compat_ioctl) 1725 return -ENOIOCTLCMD; 1726 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg); 1727 } 1728 #endif 1729 1730 static char sd_pr_type(enum pr_type type) 1731 { 1732 switch (type) { 1733 case PR_WRITE_EXCLUSIVE: 1734 return 0x01; 1735 case PR_EXCLUSIVE_ACCESS: 1736 return 0x03; 1737 case PR_WRITE_EXCLUSIVE_REG_ONLY: 1738 return 0x05; 1739 case PR_EXCLUSIVE_ACCESS_REG_ONLY: 1740 return 0x06; 1741 case PR_WRITE_EXCLUSIVE_ALL_REGS: 1742 return 0x07; 1743 case PR_EXCLUSIVE_ACCESS_ALL_REGS: 1744 return 0x08; 1745 default: 1746 return 0; 1747 } 1748 }; 1749 1750 static int sd_pr_command(struct block_device *bdev, u8 sa, 1751 u64 key, u64 sa_key, u8 type, u8 flags) 1752 { 1753 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device; 1754 struct scsi_sense_hdr sshdr; 1755 int result; 1756 u8 cmd[16] = { 0, }; 1757 u8 data[24] = { 0, }; 1758 1759 cmd[0] = PERSISTENT_RESERVE_OUT; 1760 cmd[1] = sa; 1761 cmd[2] = type; 1762 put_unaligned_be32(sizeof(data), &cmd[5]); 1763 1764 put_unaligned_be64(key, &data[0]); 1765 put_unaligned_be64(sa_key, &data[8]); 1766 data[20] = flags; 1767 1768 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data), 1769 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL); 1770 1771 if ((driver_byte(result) & DRIVER_SENSE) && 1772 (scsi_sense_valid(&sshdr))) { 1773 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result); 1774 scsi_print_sense_hdr(sdev, NULL, &sshdr); 1775 } 1776 1777 return result; 1778 } 1779 1780 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, 1781 u32 flags) 1782 { 1783 if (flags & ~PR_FL_IGNORE_KEY) 1784 return -EOPNOTSUPP; 1785 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00, 1786 old_key, new_key, 0, 1787 (1 << 0) /* APTPL */); 1788 } 1789 1790 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, 1791 u32 flags) 1792 { 1793 if (flags) 1794 return -EOPNOTSUPP; 1795 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0); 1796 } 1797 1798 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type) 1799 { 1800 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0); 1801 } 1802 1803 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, 1804 enum pr_type type, bool abort) 1805 { 1806 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key, 1807 sd_pr_type(type), 0); 1808 } 1809 1810 static int sd_pr_clear(struct block_device *bdev, u64 key) 1811 { 1812 return sd_pr_command(bdev, 0x03, key, 0, 0, 0); 1813 } 1814 1815 static const struct pr_ops sd_pr_ops = { 1816 .pr_register = sd_pr_register, 1817 .pr_reserve = sd_pr_reserve, 1818 .pr_release = sd_pr_release, 1819 .pr_preempt = sd_pr_preempt, 1820 .pr_clear = sd_pr_clear, 1821 }; 1822 1823 static const struct block_device_operations sd_fops = { 1824 .owner = THIS_MODULE, 1825 .open = sd_open, 1826 .release = sd_release, 1827 .ioctl = sd_ioctl, 1828 .getgeo = sd_getgeo, 1829 #ifdef CONFIG_COMPAT 1830 .compat_ioctl = sd_compat_ioctl, 1831 #endif 1832 .check_events = sd_check_events, 1833 .revalidate_disk = sd_revalidate_disk, 1834 .unlock_native_capacity = sd_unlock_native_capacity, 1835 .pr_ops = &sd_pr_ops, 1836 }; 1837 1838 /** 1839 * sd_eh_reset - reset error handling callback 1840 * @scmd: sd-issued command that has failed 1841 * 1842 * This function is called by the SCSI midlayer before starting 1843 * SCSI EH. When counting medium access failures we have to be 1844 * careful to register it only only once per device and SCSI EH run; 1845 * there might be several timed out commands which will cause the 1846 * 'max_medium_access_timeouts' counter to trigger after the first 1847 * SCSI EH run already and set the device to offline. 1848 * So this function resets the internal counter before starting SCSI EH. 1849 **/ 1850 static void sd_eh_reset(struct scsi_cmnd *scmd) 1851 { 1852 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk); 1853 1854 /* New SCSI EH run, reset gate variable */ 1855 sdkp->ignore_medium_access_errors = false; 1856 } 1857 1858 /** 1859 * sd_eh_action - error handling callback 1860 * @scmd: sd-issued command that has failed 1861 * @eh_disp: The recovery disposition suggested by the midlayer 1862 * 1863 * This function is called by the SCSI midlayer upon completion of an 1864 * error test command (currently TEST UNIT READY). The result of sending 1865 * the eh command is passed in eh_disp. We're looking for devices that 1866 * fail medium access commands but are OK with non access commands like 1867 * test unit ready (so wrongly see the device as having a successful 1868 * recovery) 1869 **/ 1870 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp) 1871 { 1872 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk); 1873 struct scsi_device *sdev = scmd->device; 1874 1875 if (!scsi_device_online(sdev) || 1876 !scsi_medium_access_command(scmd) || 1877 host_byte(scmd->result) != DID_TIME_OUT || 1878 eh_disp != SUCCESS) 1879 return eh_disp; 1880 1881 /* 1882 * The device has timed out executing a medium access command. 1883 * However, the TEST UNIT READY command sent during error 1884 * handling completed successfully. Either the device is in the 1885 * process of recovering or has it suffered an internal failure 1886 * that prevents access to the storage medium. 1887 */ 1888 if (!sdkp->ignore_medium_access_errors) { 1889 sdkp->medium_access_timed_out++; 1890 sdkp->ignore_medium_access_errors = true; 1891 } 1892 1893 /* 1894 * If the device keeps failing read/write commands but TEST UNIT 1895 * READY always completes successfully we assume that medium 1896 * access is no longer possible and take the device offline. 1897 */ 1898 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) { 1899 scmd_printk(KERN_ERR, scmd, 1900 "Medium access timeout failure. Offlining disk!\n"); 1901 mutex_lock(&sdev->state_mutex); 1902 scsi_device_set_state(sdev, SDEV_OFFLINE); 1903 mutex_unlock(&sdev->state_mutex); 1904 1905 return SUCCESS; 1906 } 1907 1908 return eh_disp; 1909 } 1910 1911 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd) 1912 { 1913 struct request *req = scmd->request; 1914 struct scsi_device *sdev = scmd->device; 1915 unsigned int transferred, good_bytes; 1916 u64 start_lba, end_lba, bad_lba; 1917 1918 /* 1919 * Some commands have a payload smaller than the device logical 1920 * block size (e.g. INQUIRY on a 4K disk). 1921 */ 1922 if (scsi_bufflen(scmd) <= sdev->sector_size) 1923 return 0; 1924 1925 /* Check if we have a 'bad_lba' information */ 1926 if (!scsi_get_sense_info_fld(scmd->sense_buffer, 1927 SCSI_SENSE_BUFFERSIZE, 1928 &bad_lba)) 1929 return 0; 1930 1931 /* 1932 * If the bad lba was reported incorrectly, we have no idea where 1933 * the error is. 1934 */ 1935 start_lba = sectors_to_logical(sdev, blk_rq_pos(req)); 1936 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd)); 1937 if (bad_lba < start_lba || bad_lba >= end_lba) 1938 return 0; 1939 1940 /* 1941 * resid is optional but mostly filled in. When it's unused, 1942 * its value is zero, so we assume the whole buffer transferred 1943 */ 1944 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd); 1945 1946 /* This computation should always be done in terms of the 1947 * resolution of the device's medium. 1948 */ 1949 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba); 1950 1951 return min(good_bytes, transferred); 1952 } 1953 1954 /** 1955 * sd_done - bottom half handler: called when the lower level 1956 * driver has completed (successfully or otherwise) a scsi command. 1957 * @SCpnt: mid-level's per command structure. 1958 * 1959 * Note: potentially run from within an ISR. Must not block. 1960 **/ 1961 static int sd_done(struct scsi_cmnd *SCpnt) 1962 { 1963 int result = SCpnt->result; 1964 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt); 1965 unsigned int sector_size = SCpnt->device->sector_size; 1966 unsigned int resid; 1967 struct scsi_sense_hdr sshdr; 1968 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk); 1969 struct request *req = SCpnt->request; 1970 int sense_valid = 0; 1971 int sense_deferred = 0; 1972 1973 switch (req_op(req)) { 1974 case REQ_OP_DISCARD: 1975 case REQ_OP_WRITE_ZEROES: 1976 case REQ_OP_WRITE_SAME: 1977 case REQ_OP_ZONE_RESET: 1978 if (!result) { 1979 good_bytes = blk_rq_bytes(req); 1980 scsi_set_resid(SCpnt, 0); 1981 } else { 1982 good_bytes = 0; 1983 scsi_set_resid(SCpnt, blk_rq_bytes(req)); 1984 } 1985 break; 1986 case REQ_OP_ZONE_REPORT: 1987 if (!result) { 1988 good_bytes = scsi_bufflen(SCpnt) 1989 - scsi_get_resid(SCpnt); 1990 scsi_set_resid(SCpnt, 0); 1991 } else { 1992 good_bytes = 0; 1993 scsi_set_resid(SCpnt, blk_rq_bytes(req)); 1994 } 1995 break; 1996 default: 1997 /* 1998 * In case of bogus fw or device, we could end up having 1999 * an unaligned partial completion. Check this here and force 2000 * alignment. 2001 */ 2002 resid = scsi_get_resid(SCpnt); 2003 if (resid & (sector_size - 1)) { 2004 sd_printk(KERN_INFO, sdkp, 2005 "Unaligned partial completion (resid=%u, sector_sz=%u)\n", 2006 resid, sector_size); 2007 resid = min(scsi_bufflen(SCpnt), 2008 round_up(resid, sector_size)); 2009 scsi_set_resid(SCpnt, resid); 2010 } 2011 } 2012 2013 if (result) { 2014 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr); 2015 if (sense_valid) 2016 sense_deferred = scsi_sense_is_deferred(&sshdr); 2017 } 2018 sdkp->medium_access_timed_out = 0; 2019 2020 if (driver_byte(result) != DRIVER_SENSE && 2021 (!sense_valid || sense_deferred)) 2022 goto out; 2023 2024 switch (sshdr.sense_key) { 2025 case HARDWARE_ERROR: 2026 case MEDIUM_ERROR: 2027 good_bytes = sd_completed_bytes(SCpnt); 2028 break; 2029 case RECOVERED_ERROR: 2030 good_bytes = scsi_bufflen(SCpnt); 2031 break; 2032 case NO_SENSE: 2033 /* This indicates a false check condition, so ignore it. An 2034 * unknown amount of data was transferred so treat it as an 2035 * error. 2036 */ 2037 SCpnt->result = 0; 2038 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 2039 break; 2040 case ABORTED_COMMAND: 2041 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */ 2042 good_bytes = sd_completed_bytes(SCpnt); 2043 break; 2044 case ILLEGAL_REQUEST: 2045 switch (sshdr.asc) { 2046 case 0x10: /* DIX: Host detected corruption */ 2047 good_bytes = sd_completed_bytes(SCpnt); 2048 break; 2049 case 0x20: /* INVALID COMMAND OPCODE */ 2050 case 0x24: /* INVALID FIELD IN CDB */ 2051 switch (SCpnt->cmnd[0]) { 2052 case UNMAP: 2053 sd_config_discard(sdkp, SD_LBP_DISABLE); 2054 break; 2055 case WRITE_SAME_16: 2056 case WRITE_SAME: 2057 if (SCpnt->cmnd[1] & 8) { /* UNMAP */ 2058 sd_config_discard(sdkp, SD_LBP_DISABLE); 2059 } else { 2060 sdkp->device->no_write_same = 1; 2061 sd_config_write_same(sdkp); 2062 req->__data_len = blk_rq_bytes(req); 2063 req->rq_flags |= RQF_QUIET; 2064 } 2065 break; 2066 } 2067 } 2068 break; 2069 default: 2070 break; 2071 } 2072 2073 out: 2074 if (sd_is_zoned(sdkp)) 2075 sd_zbc_complete(SCpnt, good_bytes, &sshdr); 2076 2077 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt, 2078 "sd_done: completed %d of %d bytes\n", 2079 good_bytes, scsi_bufflen(SCpnt))); 2080 2081 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt)) 2082 sd_dif_complete(SCpnt, good_bytes); 2083 2084 return good_bytes; 2085 } 2086 2087 /* 2088 * spinup disk - called only in sd_revalidate_disk() 2089 */ 2090 static void 2091 sd_spinup_disk(struct scsi_disk *sdkp) 2092 { 2093 unsigned char cmd[10]; 2094 unsigned long spintime_expire = 0; 2095 int retries, spintime; 2096 unsigned int the_result; 2097 struct scsi_sense_hdr sshdr; 2098 int sense_valid = 0; 2099 2100 spintime = 0; 2101 2102 /* Spin up drives, as required. Only do this at boot time */ 2103 /* Spinup needs to be done for module loads too. */ 2104 do { 2105 retries = 0; 2106 2107 do { 2108 cmd[0] = TEST_UNIT_READY; 2109 memset((void *) &cmd[1], 0, 9); 2110 2111 the_result = scsi_execute_req(sdkp->device, cmd, 2112 DMA_NONE, NULL, 0, 2113 &sshdr, SD_TIMEOUT, 2114 SD_MAX_RETRIES, NULL); 2115 2116 /* 2117 * If the drive has indicated to us that it 2118 * doesn't have any media in it, don't bother 2119 * with any more polling. 2120 */ 2121 if (media_not_present(sdkp, &sshdr)) 2122 return; 2123 2124 if (the_result) 2125 sense_valid = scsi_sense_valid(&sshdr); 2126 retries++; 2127 } while (retries < 3 && 2128 (!scsi_status_is_good(the_result) || 2129 ((driver_byte(the_result) & DRIVER_SENSE) && 2130 sense_valid && sshdr.sense_key == UNIT_ATTENTION))); 2131 2132 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) { 2133 /* no sense, TUR either succeeded or failed 2134 * with a status error */ 2135 if(!spintime && !scsi_status_is_good(the_result)) { 2136 sd_print_result(sdkp, "Test Unit Ready failed", 2137 the_result); 2138 } 2139 break; 2140 } 2141 2142 /* 2143 * The device does not want the automatic start to be issued. 2144 */ 2145 if (sdkp->device->no_start_on_add) 2146 break; 2147 2148 if (sense_valid && sshdr.sense_key == NOT_READY) { 2149 if (sshdr.asc == 4 && sshdr.ascq == 3) 2150 break; /* manual intervention required */ 2151 if (sshdr.asc == 4 && sshdr.ascq == 0xb) 2152 break; /* standby */ 2153 if (sshdr.asc == 4 && sshdr.ascq == 0xc) 2154 break; /* unavailable */ 2155 /* 2156 * Issue command to spin up drive when not ready 2157 */ 2158 if (!spintime) { 2159 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk..."); 2160 cmd[0] = START_STOP; 2161 cmd[1] = 1; /* Return immediately */ 2162 memset((void *) &cmd[2], 0, 8); 2163 cmd[4] = 1; /* Start spin cycle */ 2164 if (sdkp->device->start_stop_pwr_cond) 2165 cmd[4] |= 1 << 4; 2166 scsi_execute_req(sdkp->device, cmd, DMA_NONE, 2167 NULL, 0, &sshdr, 2168 SD_TIMEOUT, SD_MAX_RETRIES, 2169 NULL); 2170 spintime_expire = jiffies + 100 * HZ; 2171 spintime = 1; 2172 } 2173 /* Wait 1 second for next try */ 2174 msleep(1000); 2175 printk("."); 2176 2177 /* 2178 * Wait for USB flash devices with slow firmware. 2179 * Yes, this sense key/ASC combination shouldn't 2180 * occur here. It's characteristic of these devices. 2181 */ 2182 } else if (sense_valid && 2183 sshdr.sense_key == UNIT_ATTENTION && 2184 sshdr.asc == 0x28) { 2185 if (!spintime) { 2186 spintime_expire = jiffies + 5 * HZ; 2187 spintime = 1; 2188 } 2189 /* Wait 1 second for next try */ 2190 msleep(1000); 2191 } else { 2192 /* we don't understand the sense code, so it's 2193 * probably pointless to loop */ 2194 if(!spintime) { 2195 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n"); 2196 sd_print_sense_hdr(sdkp, &sshdr); 2197 } 2198 break; 2199 } 2200 2201 } while (spintime && time_before_eq(jiffies, spintime_expire)); 2202 2203 if (spintime) { 2204 if (scsi_status_is_good(the_result)) 2205 printk("ready\n"); 2206 else 2207 printk("not responding...\n"); 2208 } 2209 } 2210 2211 /* 2212 * Determine whether disk supports Data Integrity Field. 2213 */ 2214 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer) 2215 { 2216 struct scsi_device *sdp = sdkp->device; 2217 u8 type; 2218 int ret = 0; 2219 2220 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) 2221 return ret; 2222 2223 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */ 2224 2225 if (type > T10_PI_TYPE3_PROTECTION) 2226 ret = -ENODEV; 2227 else if (scsi_host_dif_capable(sdp->host, type)) 2228 ret = 1; 2229 2230 if (sdkp->first_scan || type != sdkp->protection_type) 2231 switch (ret) { 2232 case -ENODEV: 2233 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \ 2234 " protection type %u. Disabling disk!\n", 2235 type); 2236 break; 2237 case 1: 2238 sd_printk(KERN_NOTICE, sdkp, 2239 "Enabling DIF Type %u protection\n", type); 2240 break; 2241 case 0: 2242 sd_printk(KERN_NOTICE, sdkp, 2243 "Disabling DIF Type %u protection\n", type); 2244 break; 2245 } 2246 2247 sdkp->protection_type = type; 2248 2249 return ret; 2250 } 2251 2252 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp, 2253 struct scsi_sense_hdr *sshdr, int sense_valid, 2254 int the_result) 2255 { 2256 if (driver_byte(the_result) & DRIVER_SENSE) 2257 sd_print_sense_hdr(sdkp, sshdr); 2258 else 2259 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n"); 2260 2261 /* 2262 * Set dirty bit for removable devices if not ready - 2263 * sometimes drives will not report this properly. 2264 */ 2265 if (sdp->removable && 2266 sense_valid && sshdr->sense_key == NOT_READY) 2267 set_media_not_present(sdkp); 2268 2269 /* 2270 * We used to set media_present to 0 here to indicate no media 2271 * in the drive, but some drives fail read capacity even with 2272 * media present, so we can't do that. 2273 */ 2274 sdkp->capacity = 0; /* unknown mapped to zero - as usual */ 2275 } 2276 2277 #define RC16_LEN 32 2278 #if RC16_LEN > SD_BUF_SIZE 2279 #error RC16_LEN must not be more than SD_BUF_SIZE 2280 #endif 2281 2282 #define READ_CAPACITY_RETRIES_ON_RESET 10 2283 2284 /* 2285 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set 2286 * and the reported logical block size is bigger than 512 bytes. Note 2287 * that last_sector is a u64 and therefore logical_to_sectors() is not 2288 * applicable. 2289 */ 2290 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size) 2291 { 2292 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9); 2293 2294 if (sizeof(sector_t) == 4 && last_sector > U32_MAX) 2295 return false; 2296 2297 return true; 2298 } 2299 2300 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp, 2301 unsigned char *buffer) 2302 { 2303 unsigned char cmd[16]; 2304 struct scsi_sense_hdr sshdr; 2305 int sense_valid = 0; 2306 int the_result; 2307 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET; 2308 unsigned int alignment; 2309 unsigned long long lba; 2310 unsigned sector_size; 2311 2312 if (sdp->no_read_capacity_16) 2313 return -EINVAL; 2314 2315 do { 2316 memset(cmd, 0, 16); 2317 cmd[0] = SERVICE_ACTION_IN_16; 2318 cmd[1] = SAI_READ_CAPACITY_16; 2319 cmd[13] = RC16_LEN; 2320 memset(buffer, 0, RC16_LEN); 2321 2322 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE, 2323 buffer, RC16_LEN, &sshdr, 2324 SD_TIMEOUT, SD_MAX_RETRIES, NULL); 2325 2326 if (media_not_present(sdkp, &sshdr)) 2327 return -ENODEV; 2328 2329 if (the_result) { 2330 sense_valid = scsi_sense_valid(&sshdr); 2331 if (sense_valid && 2332 sshdr.sense_key == ILLEGAL_REQUEST && 2333 (sshdr.asc == 0x20 || sshdr.asc == 0x24) && 2334 sshdr.ascq == 0x00) 2335 /* Invalid Command Operation Code or 2336 * Invalid Field in CDB, just retry 2337 * silently with RC10 */ 2338 return -EINVAL; 2339 if (sense_valid && 2340 sshdr.sense_key == UNIT_ATTENTION && 2341 sshdr.asc == 0x29 && sshdr.ascq == 0x00) 2342 /* Device reset might occur several times, 2343 * give it one more chance */ 2344 if (--reset_retries > 0) 2345 continue; 2346 } 2347 retries--; 2348 2349 } while (the_result && retries); 2350 2351 if (the_result) { 2352 sd_print_result(sdkp, "Read Capacity(16) failed", the_result); 2353 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result); 2354 return -EINVAL; 2355 } 2356 2357 sector_size = get_unaligned_be32(&buffer[8]); 2358 lba = get_unaligned_be64(&buffer[0]); 2359 2360 if (sd_read_protection_type(sdkp, buffer) < 0) { 2361 sdkp->capacity = 0; 2362 return -ENODEV; 2363 } 2364 2365 if (!sd_addressable_capacity(lba, sector_size)) { 2366 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a " 2367 "kernel compiled with support for large block " 2368 "devices.\n"); 2369 sdkp->capacity = 0; 2370 return -EOVERFLOW; 2371 } 2372 2373 /* Logical blocks per physical block exponent */ 2374 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size; 2375 2376 /* RC basis */ 2377 sdkp->rc_basis = (buffer[12] >> 4) & 0x3; 2378 2379 /* Lowest aligned logical block */ 2380 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size; 2381 blk_queue_alignment_offset(sdp->request_queue, alignment); 2382 if (alignment && sdkp->first_scan) 2383 sd_printk(KERN_NOTICE, sdkp, 2384 "physical block alignment offset: %u\n", alignment); 2385 2386 if (buffer[14] & 0x80) { /* LBPME */ 2387 sdkp->lbpme = 1; 2388 2389 if (buffer[14] & 0x40) /* LBPRZ */ 2390 sdkp->lbprz = 1; 2391 2392 sd_config_discard(sdkp, SD_LBP_WS16); 2393 } 2394 2395 sdkp->capacity = lba + 1; 2396 return sector_size; 2397 } 2398 2399 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp, 2400 unsigned char *buffer) 2401 { 2402 unsigned char cmd[16]; 2403 struct scsi_sense_hdr sshdr; 2404 int sense_valid = 0; 2405 int the_result; 2406 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET; 2407 sector_t lba; 2408 unsigned sector_size; 2409 2410 do { 2411 cmd[0] = READ_CAPACITY; 2412 memset(&cmd[1], 0, 9); 2413 memset(buffer, 0, 8); 2414 2415 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE, 2416 buffer, 8, &sshdr, 2417 SD_TIMEOUT, SD_MAX_RETRIES, NULL); 2418 2419 if (media_not_present(sdkp, &sshdr)) 2420 return -ENODEV; 2421 2422 if (the_result) { 2423 sense_valid = scsi_sense_valid(&sshdr); 2424 if (sense_valid && 2425 sshdr.sense_key == UNIT_ATTENTION && 2426 sshdr.asc == 0x29 && sshdr.ascq == 0x00) 2427 /* Device reset might occur several times, 2428 * give it one more chance */ 2429 if (--reset_retries > 0) 2430 continue; 2431 } 2432 retries--; 2433 2434 } while (the_result && retries); 2435 2436 if (the_result) { 2437 sd_print_result(sdkp, "Read Capacity(10) failed", the_result); 2438 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result); 2439 return -EINVAL; 2440 } 2441 2442 sector_size = get_unaligned_be32(&buffer[4]); 2443 lba = get_unaligned_be32(&buffer[0]); 2444 2445 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) { 2446 /* Some buggy (usb cardreader) devices return an lba of 2447 0xffffffff when the want to report a size of 0 (with 2448 which they really mean no media is present) */ 2449 sdkp->capacity = 0; 2450 sdkp->physical_block_size = sector_size; 2451 return sector_size; 2452 } 2453 2454 if (!sd_addressable_capacity(lba, sector_size)) { 2455 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a " 2456 "kernel compiled with support for large block " 2457 "devices.\n"); 2458 sdkp->capacity = 0; 2459 return -EOVERFLOW; 2460 } 2461 2462 sdkp->capacity = lba + 1; 2463 sdkp->physical_block_size = sector_size; 2464 return sector_size; 2465 } 2466 2467 static int sd_try_rc16_first(struct scsi_device *sdp) 2468 { 2469 if (sdp->host->max_cmd_len < 16) 2470 return 0; 2471 if (sdp->try_rc_10_first) 2472 return 0; 2473 if (sdp->scsi_level > SCSI_SPC_2) 2474 return 1; 2475 if (scsi_device_protection(sdp)) 2476 return 1; 2477 return 0; 2478 } 2479 2480 /* 2481 * read disk capacity 2482 */ 2483 static void 2484 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer) 2485 { 2486 int sector_size; 2487 struct scsi_device *sdp = sdkp->device; 2488 2489 if (sd_try_rc16_first(sdp)) { 2490 sector_size = read_capacity_16(sdkp, sdp, buffer); 2491 if (sector_size == -EOVERFLOW) 2492 goto got_data; 2493 if (sector_size == -ENODEV) 2494 return; 2495 if (sector_size < 0) 2496 sector_size = read_capacity_10(sdkp, sdp, buffer); 2497 if (sector_size < 0) 2498 return; 2499 } else { 2500 sector_size = read_capacity_10(sdkp, sdp, buffer); 2501 if (sector_size == -EOVERFLOW) 2502 goto got_data; 2503 if (sector_size < 0) 2504 return; 2505 if ((sizeof(sdkp->capacity) > 4) && 2506 (sdkp->capacity > 0xffffffffULL)) { 2507 int old_sector_size = sector_size; 2508 sd_printk(KERN_NOTICE, sdkp, "Very big device. " 2509 "Trying to use READ CAPACITY(16).\n"); 2510 sector_size = read_capacity_16(sdkp, sdp, buffer); 2511 if (sector_size < 0) { 2512 sd_printk(KERN_NOTICE, sdkp, 2513 "Using 0xffffffff as device size\n"); 2514 sdkp->capacity = 1 + (sector_t) 0xffffffff; 2515 sector_size = old_sector_size; 2516 goto got_data; 2517 } 2518 } 2519 } 2520 2521 /* Some devices are known to return the total number of blocks, 2522 * not the highest block number. Some devices have versions 2523 * which do this and others which do not. Some devices we might 2524 * suspect of doing this but we don't know for certain. 2525 * 2526 * If we know the reported capacity is wrong, decrement it. If 2527 * we can only guess, then assume the number of blocks is even 2528 * (usually true but not always) and err on the side of lowering 2529 * the capacity. 2530 */ 2531 if (sdp->fix_capacity || 2532 (sdp->guess_capacity && (sdkp->capacity & 0x01))) { 2533 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count " 2534 "from its reported value: %llu\n", 2535 (unsigned long long) sdkp->capacity); 2536 --sdkp->capacity; 2537 } 2538 2539 got_data: 2540 if (sector_size == 0) { 2541 sector_size = 512; 2542 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, " 2543 "assuming 512.\n"); 2544 } 2545 2546 if (sector_size != 512 && 2547 sector_size != 1024 && 2548 sector_size != 2048 && 2549 sector_size != 4096) { 2550 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n", 2551 sector_size); 2552 /* 2553 * The user might want to re-format the drive with 2554 * a supported sectorsize. Once this happens, it 2555 * would be relatively trivial to set the thing up. 2556 * For this reason, we leave the thing in the table. 2557 */ 2558 sdkp->capacity = 0; 2559 /* 2560 * set a bogus sector size so the normal read/write 2561 * logic in the block layer will eventually refuse any 2562 * request on this device without tripping over power 2563 * of two sector size assumptions 2564 */ 2565 sector_size = 512; 2566 } 2567 blk_queue_logical_block_size(sdp->request_queue, sector_size); 2568 blk_queue_physical_block_size(sdp->request_queue, 2569 sdkp->physical_block_size); 2570 sdkp->device->sector_size = sector_size; 2571 2572 if (sdkp->capacity > 0xffffffff) 2573 sdp->use_16_for_rw = 1; 2574 2575 } 2576 2577 /* 2578 * Print disk capacity 2579 */ 2580 static void 2581 sd_print_capacity(struct scsi_disk *sdkp, 2582 sector_t old_capacity) 2583 { 2584 int sector_size = sdkp->device->sector_size; 2585 char cap_str_2[10], cap_str_10[10]; 2586 2587 string_get_size(sdkp->capacity, sector_size, 2588 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); 2589 string_get_size(sdkp->capacity, sector_size, 2590 STRING_UNITS_10, cap_str_10, 2591 sizeof(cap_str_10)); 2592 2593 if (sdkp->first_scan || old_capacity != sdkp->capacity) { 2594 sd_printk(KERN_NOTICE, sdkp, 2595 "%llu %d-byte logical blocks: (%s/%s)\n", 2596 (unsigned long long)sdkp->capacity, 2597 sector_size, cap_str_10, cap_str_2); 2598 2599 if (sdkp->physical_block_size != sector_size) 2600 sd_printk(KERN_NOTICE, sdkp, 2601 "%u-byte physical blocks\n", 2602 sdkp->physical_block_size); 2603 2604 sd_zbc_print_zones(sdkp); 2605 } 2606 } 2607 2608 /* called with buffer of length 512 */ 2609 static inline int 2610 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage, 2611 unsigned char *buffer, int len, struct scsi_mode_data *data, 2612 struct scsi_sense_hdr *sshdr) 2613 { 2614 return scsi_mode_sense(sdp, dbd, modepage, buffer, len, 2615 SD_TIMEOUT, SD_MAX_RETRIES, data, 2616 sshdr); 2617 } 2618 2619 /* 2620 * read write protect setting, if possible - called only in sd_revalidate_disk() 2621 * called with buffer of length SD_BUF_SIZE 2622 */ 2623 static void 2624 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer) 2625 { 2626 int res; 2627 struct scsi_device *sdp = sdkp->device; 2628 struct scsi_mode_data data; 2629 int old_wp = sdkp->write_prot; 2630 2631 set_disk_ro(sdkp->disk, 0); 2632 if (sdp->skip_ms_page_3f) { 2633 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n"); 2634 return; 2635 } 2636 2637 if (sdp->use_192_bytes_for_3f) { 2638 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL); 2639 } else { 2640 /* 2641 * First attempt: ask for all pages (0x3F), but only 4 bytes. 2642 * We have to start carefully: some devices hang if we ask 2643 * for more than is available. 2644 */ 2645 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL); 2646 2647 /* 2648 * Second attempt: ask for page 0 When only page 0 is 2649 * implemented, a request for page 3F may return Sense Key 2650 * 5: Illegal Request, Sense Code 24: Invalid field in 2651 * CDB. 2652 */ 2653 if (!scsi_status_is_good(res)) 2654 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL); 2655 2656 /* 2657 * Third attempt: ask 255 bytes, as we did earlier. 2658 */ 2659 if (!scsi_status_is_good(res)) 2660 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255, 2661 &data, NULL); 2662 } 2663 2664 if (!scsi_status_is_good(res)) { 2665 sd_first_printk(KERN_WARNING, sdkp, 2666 "Test WP failed, assume Write Enabled\n"); 2667 } else { 2668 sdkp->write_prot = ((data.device_specific & 0x80) != 0); 2669 set_disk_ro(sdkp->disk, sdkp->write_prot); 2670 if (sdkp->first_scan || old_wp != sdkp->write_prot) { 2671 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n", 2672 sdkp->write_prot ? "on" : "off"); 2673 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer); 2674 } 2675 } 2676 } 2677 2678 /* 2679 * sd_read_cache_type - called only from sd_revalidate_disk() 2680 * called with buffer of length SD_BUF_SIZE 2681 */ 2682 static void 2683 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer) 2684 { 2685 int len = 0, res; 2686 struct scsi_device *sdp = sdkp->device; 2687 2688 int dbd; 2689 int modepage; 2690 int first_len; 2691 struct scsi_mode_data data; 2692 struct scsi_sense_hdr sshdr; 2693 int old_wce = sdkp->WCE; 2694 int old_rcd = sdkp->RCD; 2695 int old_dpofua = sdkp->DPOFUA; 2696 2697 2698 if (sdkp->cache_override) 2699 return; 2700 2701 first_len = 4; 2702 if (sdp->skip_ms_page_8) { 2703 if (sdp->type == TYPE_RBC) 2704 goto defaults; 2705 else { 2706 if (sdp->skip_ms_page_3f) 2707 goto defaults; 2708 modepage = 0x3F; 2709 if (sdp->use_192_bytes_for_3f) 2710 first_len = 192; 2711 dbd = 0; 2712 } 2713 } else if (sdp->type == TYPE_RBC) { 2714 modepage = 6; 2715 dbd = 8; 2716 } else { 2717 modepage = 8; 2718 dbd = 0; 2719 } 2720 2721 /* cautiously ask */ 2722 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len, 2723 &data, &sshdr); 2724 2725 if (!scsi_status_is_good(res)) 2726 goto bad_sense; 2727 2728 if (!data.header_length) { 2729 modepage = 6; 2730 first_len = 0; 2731 sd_first_printk(KERN_ERR, sdkp, 2732 "Missing header in MODE_SENSE response\n"); 2733 } 2734 2735 /* that went OK, now ask for the proper length */ 2736 len = data.length; 2737 2738 /* 2739 * We're only interested in the first three bytes, actually. 2740 * But the data cache page is defined for the first 20. 2741 */ 2742 if (len < 3) 2743 goto bad_sense; 2744 else if (len > SD_BUF_SIZE) { 2745 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter " 2746 "data from %d to %d bytes\n", len, SD_BUF_SIZE); 2747 len = SD_BUF_SIZE; 2748 } 2749 if (modepage == 0x3F && sdp->use_192_bytes_for_3f) 2750 len = 192; 2751 2752 /* Get the data */ 2753 if (len > first_len) 2754 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, 2755 &data, &sshdr); 2756 2757 if (scsi_status_is_good(res)) { 2758 int offset = data.header_length + data.block_descriptor_length; 2759 2760 while (offset < len) { 2761 u8 page_code = buffer[offset] & 0x3F; 2762 u8 spf = buffer[offset] & 0x40; 2763 2764 if (page_code == 8 || page_code == 6) { 2765 /* We're interested only in the first 3 bytes. 2766 */ 2767 if (len - offset <= 2) { 2768 sd_first_printk(KERN_ERR, sdkp, 2769 "Incomplete mode parameter " 2770 "data\n"); 2771 goto defaults; 2772 } else { 2773 modepage = page_code; 2774 goto Page_found; 2775 } 2776 } else { 2777 /* Go to the next page */ 2778 if (spf && len - offset > 3) 2779 offset += 4 + (buffer[offset+2] << 8) + 2780 buffer[offset+3]; 2781 else if (!spf && len - offset > 1) 2782 offset += 2 + buffer[offset+1]; 2783 else { 2784 sd_first_printk(KERN_ERR, sdkp, 2785 "Incomplete mode " 2786 "parameter data\n"); 2787 goto defaults; 2788 } 2789 } 2790 } 2791 2792 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n"); 2793 goto defaults; 2794 2795 Page_found: 2796 if (modepage == 8) { 2797 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0); 2798 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0); 2799 } else { 2800 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0); 2801 sdkp->RCD = 0; 2802 } 2803 2804 sdkp->DPOFUA = (data.device_specific & 0x10) != 0; 2805 if (sdp->broken_fua) { 2806 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n"); 2807 sdkp->DPOFUA = 0; 2808 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw && 2809 !sdkp->device->use_16_for_rw) { 2810 sd_first_printk(KERN_NOTICE, sdkp, 2811 "Uses READ/WRITE(6), disabling FUA\n"); 2812 sdkp->DPOFUA = 0; 2813 } 2814 2815 /* No cache flush allowed for write protected devices */ 2816 if (sdkp->WCE && sdkp->write_prot) 2817 sdkp->WCE = 0; 2818 2819 if (sdkp->first_scan || old_wce != sdkp->WCE || 2820 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA) 2821 sd_printk(KERN_NOTICE, sdkp, 2822 "Write cache: %s, read cache: %s, %s\n", 2823 sdkp->WCE ? "enabled" : "disabled", 2824 sdkp->RCD ? "disabled" : "enabled", 2825 sdkp->DPOFUA ? "supports DPO and FUA" 2826 : "doesn't support DPO or FUA"); 2827 2828 return; 2829 } 2830 2831 bad_sense: 2832 if (scsi_sense_valid(&sshdr) && 2833 sshdr.sense_key == ILLEGAL_REQUEST && 2834 sshdr.asc == 0x24 && sshdr.ascq == 0x0) 2835 /* Invalid field in CDB */ 2836 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n"); 2837 else 2838 sd_first_printk(KERN_ERR, sdkp, 2839 "Asking for cache data failed\n"); 2840 2841 defaults: 2842 if (sdp->wce_default_on) { 2843 sd_first_printk(KERN_NOTICE, sdkp, 2844 "Assuming drive cache: write back\n"); 2845 sdkp->WCE = 1; 2846 } else { 2847 sd_first_printk(KERN_ERR, sdkp, 2848 "Assuming drive cache: write through\n"); 2849 sdkp->WCE = 0; 2850 } 2851 sdkp->RCD = 0; 2852 sdkp->DPOFUA = 0; 2853 } 2854 2855 /* 2856 * The ATO bit indicates whether the DIF application tag is available 2857 * for use by the operating system. 2858 */ 2859 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer) 2860 { 2861 int res, offset; 2862 struct scsi_device *sdp = sdkp->device; 2863 struct scsi_mode_data data; 2864 struct scsi_sense_hdr sshdr; 2865 2866 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC) 2867 return; 2868 2869 if (sdkp->protection_type == 0) 2870 return; 2871 2872 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT, 2873 SD_MAX_RETRIES, &data, &sshdr); 2874 2875 if (!scsi_status_is_good(res) || !data.header_length || 2876 data.length < 6) { 2877 sd_first_printk(KERN_WARNING, sdkp, 2878 "getting Control mode page failed, assume no ATO\n"); 2879 2880 if (scsi_sense_valid(&sshdr)) 2881 sd_print_sense_hdr(sdkp, &sshdr); 2882 2883 return; 2884 } 2885 2886 offset = data.header_length + data.block_descriptor_length; 2887 2888 if ((buffer[offset] & 0x3f) != 0x0a) { 2889 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n"); 2890 return; 2891 } 2892 2893 if ((buffer[offset + 5] & 0x80) == 0) 2894 return; 2895 2896 sdkp->ATO = 1; 2897 2898 return; 2899 } 2900 2901 /** 2902 * sd_read_block_limits - Query disk device for preferred I/O sizes. 2903 * @sdkp: disk to query 2904 */ 2905 static void sd_read_block_limits(struct scsi_disk *sdkp) 2906 { 2907 unsigned int sector_sz = sdkp->device->sector_size; 2908 const int vpd_len = 64; 2909 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL); 2910 2911 if (!buffer || 2912 /* Block Limits VPD */ 2913 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len)) 2914 goto out; 2915 2916 blk_queue_io_min(sdkp->disk->queue, 2917 get_unaligned_be16(&buffer[6]) * sector_sz); 2918 2919 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]); 2920 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]); 2921 2922 if (buffer[3] == 0x3c) { 2923 unsigned int lba_count, desc_count; 2924 2925 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]); 2926 2927 if (!sdkp->lbpme) 2928 goto out; 2929 2930 lba_count = get_unaligned_be32(&buffer[20]); 2931 desc_count = get_unaligned_be32(&buffer[24]); 2932 2933 if (lba_count && desc_count) 2934 sdkp->max_unmap_blocks = lba_count; 2935 2936 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]); 2937 2938 if (buffer[32] & 0x80) 2939 sdkp->unmap_alignment = 2940 get_unaligned_be32(&buffer[32]) & ~(1 << 31); 2941 2942 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */ 2943 2944 if (sdkp->max_unmap_blocks) 2945 sd_config_discard(sdkp, SD_LBP_UNMAP); 2946 else 2947 sd_config_discard(sdkp, SD_LBP_WS16); 2948 2949 } else { /* LBP VPD page tells us what to use */ 2950 if (sdkp->lbpu && sdkp->max_unmap_blocks) 2951 sd_config_discard(sdkp, SD_LBP_UNMAP); 2952 else if (sdkp->lbpws) 2953 sd_config_discard(sdkp, SD_LBP_WS16); 2954 else if (sdkp->lbpws10) 2955 sd_config_discard(sdkp, SD_LBP_WS10); 2956 else 2957 sd_config_discard(sdkp, SD_LBP_DISABLE); 2958 } 2959 } 2960 2961 out: 2962 kfree(buffer); 2963 } 2964 2965 /** 2966 * sd_read_block_characteristics - Query block dev. characteristics 2967 * @sdkp: disk to query 2968 */ 2969 static void sd_read_block_characteristics(struct scsi_disk *sdkp) 2970 { 2971 struct request_queue *q = sdkp->disk->queue; 2972 unsigned char *buffer; 2973 u16 rot; 2974 const int vpd_len = 64; 2975 2976 buffer = kmalloc(vpd_len, GFP_KERNEL); 2977 2978 if (!buffer || 2979 /* Block Device Characteristics VPD */ 2980 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len)) 2981 goto out; 2982 2983 rot = get_unaligned_be16(&buffer[4]); 2984 2985 if (rot == 1) { 2986 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); 2987 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q); 2988 } 2989 2990 if (sdkp->device->type == TYPE_ZBC) { 2991 /* Host-managed */ 2992 q->limits.zoned = BLK_ZONED_HM; 2993 } else { 2994 sdkp->zoned = (buffer[8] >> 4) & 3; 2995 if (sdkp->zoned == 1) 2996 /* Host-aware */ 2997 q->limits.zoned = BLK_ZONED_HA; 2998 else 2999 /* 3000 * Treat drive-managed devices as 3001 * regular block devices. 3002 */ 3003 q->limits.zoned = BLK_ZONED_NONE; 3004 } 3005 if (blk_queue_is_zoned(q) && sdkp->first_scan) 3006 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n", 3007 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware"); 3008 3009 out: 3010 kfree(buffer); 3011 } 3012 3013 /** 3014 * sd_read_block_provisioning - Query provisioning VPD page 3015 * @sdkp: disk to query 3016 */ 3017 static void sd_read_block_provisioning(struct scsi_disk *sdkp) 3018 { 3019 unsigned char *buffer; 3020 const int vpd_len = 8; 3021 3022 if (sdkp->lbpme == 0) 3023 return; 3024 3025 buffer = kmalloc(vpd_len, GFP_KERNEL); 3026 3027 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len)) 3028 goto out; 3029 3030 sdkp->lbpvpd = 1; 3031 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */ 3032 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */ 3033 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */ 3034 3035 out: 3036 kfree(buffer); 3037 } 3038 3039 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer) 3040 { 3041 struct scsi_device *sdev = sdkp->device; 3042 3043 if (sdev->host->no_write_same) { 3044 sdev->no_write_same = 1; 3045 3046 return; 3047 } 3048 3049 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) { 3050 /* too large values might cause issues with arcmsr */ 3051 int vpd_buf_len = 64; 3052 3053 sdev->no_report_opcodes = 1; 3054 3055 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION 3056 * CODES is unsupported and the device has an ATA 3057 * Information VPD page (SAT). 3058 */ 3059 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len)) 3060 sdev->no_write_same = 1; 3061 } 3062 3063 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1) 3064 sdkp->ws16 = 1; 3065 3066 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1) 3067 sdkp->ws10 = 1; 3068 } 3069 3070 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer) 3071 { 3072 struct scsi_device *sdev = sdkp->device; 3073 3074 if (!sdev->security_supported) 3075 return; 3076 3077 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, 3078 SECURITY_PROTOCOL_IN) == 1 && 3079 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, 3080 SECURITY_PROTOCOL_OUT) == 1) 3081 sdkp->security = 1; 3082 } 3083 3084 /** 3085 * sd_revalidate_disk - called the first time a new disk is seen, 3086 * performs disk spin up, read_capacity, etc. 3087 * @disk: struct gendisk we care about 3088 **/ 3089 static int sd_revalidate_disk(struct gendisk *disk) 3090 { 3091 struct scsi_disk *sdkp = scsi_disk(disk); 3092 struct scsi_device *sdp = sdkp->device; 3093 struct request_queue *q = sdkp->disk->queue; 3094 sector_t old_capacity = sdkp->capacity; 3095 unsigned char *buffer; 3096 unsigned int dev_max, rw_max; 3097 3098 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, 3099 "sd_revalidate_disk\n")); 3100 3101 /* 3102 * If the device is offline, don't try and read capacity or any 3103 * of the other niceties. 3104 */ 3105 if (!scsi_device_online(sdp)) 3106 goto out; 3107 3108 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL); 3109 if (!buffer) { 3110 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory " 3111 "allocation failure.\n"); 3112 goto out; 3113 } 3114 3115 sd_spinup_disk(sdkp); 3116 3117 /* 3118 * Without media there is no reason to ask; moreover, some devices 3119 * react badly if we do. 3120 */ 3121 if (sdkp->media_present) { 3122 sd_read_capacity(sdkp, buffer); 3123 3124 if (scsi_device_supports_vpd(sdp)) { 3125 sd_read_block_provisioning(sdkp); 3126 sd_read_block_limits(sdkp); 3127 sd_read_block_characteristics(sdkp); 3128 sd_zbc_read_zones(sdkp, buffer); 3129 } 3130 3131 sd_print_capacity(sdkp, old_capacity); 3132 3133 sd_read_write_protect_flag(sdkp, buffer); 3134 sd_read_cache_type(sdkp, buffer); 3135 sd_read_app_tag_own(sdkp, buffer); 3136 sd_read_write_same(sdkp, buffer); 3137 sd_read_security(sdkp, buffer); 3138 } 3139 3140 /* 3141 * We now have all cache related info, determine how we deal 3142 * with flush requests. 3143 */ 3144 sd_set_flush_flag(sdkp); 3145 3146 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */ 3147 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS; 3148 3149 /* Some devices report a maximum block count for READ/WRITE requests. */ 3150 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks); 3151 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max); 3152 3153 /* 3154 * Determine the device's preferred I/O size for reads and writes 3155 * unless the reported value is unreasonably small, large, or 3156 * garbage. 3157 */ 3158 if (sdkp->opt_xfer_blocks && 3159 sdkp->opt_xfer_blocks <= dev_max && 3160 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS && 3161 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) { 3162 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks); 3163 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks); 3164 } else 3165 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max), 3166 (sector_t)BLK_DEF_MAX_SECTORS); 3167 3168 /* Do not exceed controller limit */ 3169 rw_max = min(rw_max, queue_max_hw_sectors(q)); 3170 3171 /* 3172 * Only update max_sectors if previously unset or if the current value 3173 * exceeds the capabilities of the hardware. 3174 */ 3175 if (sdkp->first_scan || 3176 q->limits.max_sectors > q->limits.max_dev_sectors || 3177 q->limits.max_sectors > q->limits.max_hw_sectors) 3178 q->limits.max_sectors = rw_max; 3179 3180 sdkp->first_scan = 0; 3181 3182 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity)); 3183 sd_config_write_same(sdkp); 3184 kfree(buffer); 3185 3186 out: 3187 return 0; 3188 } 3189 3190 /** 3191 * sd_unlock_native_capacity - unlock native capacity 3192 * @disk: struct gendisk to set capacity for 3193 * 3194 * Block layer calls this function if it detects that partitions 3195 * on @disk reach beyond the end of the device. If the SCSI host 3196 * implements ->unlock_native_capacity() method, it's invoked to 3197 * give it a chance to adjust the device capacity. 3198 * 3199 * CONTEXT: 3200 * Defined by block layer. Might sleep. 3201 */ 3202 static void sd_unlock_native_capacity(struct gendisk *disk) 3203 { 3204 struct scsi_device *sdev = scsi_disk(disk)->device; 3205 3206 if (sdev->host->hostt->unlock_native_capacity) 3207 sdev->host->hostt->unlock_native_capacity(sdev); 3208 } 3209 3210 /** 3211 * sd_format_disk_name - format disk name 3212 * @prefix: name prefix - ie. "sd" for SCSI disks 3213 * @index: index of the disk to format name for 3214 * @buf: output buffer 3215 * @buflen: length of the output buffer 3216 * 3217 * SCSI disk names starts at sda. The 26th device is sdz and the 3218 * 27th is sdaa. The last one for two lettered suffix is sdzz 3219 * which is followed by sdaaa. 3220 * 3221 * This is basically 26 base counting with one extra 'nil' entry 3222 * at the beginning from the second digit on and can be 3223 * determined using similar method as 26 base conversion with the 3224 * index shifted -1 after each digit is computed. 3225 * 3226 * CONTEXT: 3227 * Don't care. 3228 * 3229 * RETURNS: 3230 * 0 on success, -errno on failure. 3231 */ 3232 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen) 3233 { 3234 const int base = 'z' - 'a' + 1; 3235 char *begin = buf + strlen(prefix); 3236 char *end = buf + buflen; 3237 char *p; 3238 int unit; 3239 3240 p = end - 1; 3241 *p = '\0'; 3242 unit = base; 3243 do { 3244 if (p == begin) 3245 return -EINVAL; 3246 *--p = 'a' + (index % unit); 3247 index = (index / unit) - 1; 3248 } while (index >= 0); 3249 3250 memmove(begin, p, end - p); 3251 memcpy(buf, prefix, strlen(prefix)); 3252 3253 return 0; 3254 } 3255 3256 /* 3257 * The asynchronous part of sd_probe 3258 */ 3259 static void sd_probe_async(void *data, async_cookie_t cookie) 3260 { 3261 struct scsi_disk *sdkp = data; 3262 struct scsi_device *sdp; 3263 struct gendisk *gd; 3264 u32 index; 3265 struct device *dev; 3266 3267 sdp = sdkp->device; 3268 gd = sdkp->disk; 3269 index = sdkp->index; 3270 dev = &sdp->sdev_gendev; 3271 3272 gd->major = sd_major((index & 0xf0) >> 4); 3273 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00); 3274 3275 gd->fops = &sd_fops; 3276 gd->private_data = &sdkp->driver; 3277 gd->queue = sdkp->device->request_queue; 3278 3279 /* defaults, until the device tells us otherwise */ 3280 sdp->sector_size = 512; 3281 sdkp->capacity = 0; 3282 sdkp->media_present = 1; 3283 sdkp->write_prot = 0; 3284 sdkp->cache_override = 0; 3285 sdkp->WCE = 0; 3286 sdkp->RCD = 0; 3287 sdkp->ATO = 0; 3288 sdkp->first_scan = 1; 3289 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS; 3290 3291 sd_revalidate_disk(gd); 3292 3293 gd->flags = GENHD_FL_EXT_DEVT; 3294 if (sdp->removable) { 3295 gd->flags |= GENHD_FL_REMOVABLE; 3296 gd->events |= DISK_EVENT_MEDIA_CHANGE; 3297 } 3298 3299 blk_pm_runtime_init(sdp->request_queue, dev); 3300 device_add_disk(dev, gd); 3301 if (sdkp->capacity) 3302 sd_dif_config_host(sdkp); 3303 3304 sd_revalidate_disk(gd); 3305 3306 if (sdkp->security) { 3307 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit); 3308 if (sdkp->opal_dev) 3309 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n"); 3310 } 3311 3312 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n", 3313 sdp->removable ? "removable " : ""); 3314 scsi_autopm_put_device(sdp); 3315 put_device(&sdkp->dev); 3316 } 3317 3318 /** 3319 * sd_probe - called during driver initialization and whenever a 3320 * new scsi device is attached to the system. It is called once 3321 * for each scsi device (not just disks) present. 3322 * @dev: pointer to device object 3323 * 3324 * Returns 0 if successful (or not interested in this scsi device 3325 * (e.g. scanner)); 1 when there is an error. 3326 * 3327 * Note: this function is invoked from the scsi mid-level. 3328 * This function sets up the mapping between a given 3329 * <host,channel,id,lun> (found in sdp) and new device name 3330 * (e.g. /dev/sda). More precisely it is the block device major 3331 * and minor number that is chosen here. 3332 * 3333 * Assume sd_probe is not re-entrant (for time being) 3334 * Also think about sd_probe() and sd_remove() running coincidentally. 3335 **/ 3336 static int sd_probe(struct device *dev) 3337 { 3338 struct scsi_device *sdp = to_scsi_device(dev); 3339 struct scsi_disk *sdkp; 3340 struct gendisk *gd; 3341 int index; 3342 int error; 3343 3344 scsi_autopm_get_device(sdp); 3345 error = -ENODEV; 3346 if (sdp->type != TYPE_DISK && 3347 sdp->type != TYPE_ZBC && 3348 sdp->type != TYPE_MOD && 3349 sdp->type != TYPE_RBC) 3350 goto out; 3351 3352 #ifndef CONFIG_BLK_DEV_ZONED 3353 if (sdp->type == TYPE_ZBC) 3354 goto out; 3355 #endif 3356 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp, 3357 "sd_probe\n")); 3358 3359 error = -ENOMEM; 3360 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL); 3361 if (!sdkp) 3362 goto out; 3363 3364 gd = alloc_disk(SD_MINORS); 3365 if (!gd) 3366 goto out_free; 3367 3368 do { 3369 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL)) 3370 goto out_put; 3371 3372 spin_lock(&sd_index_lock); 3373 error = ida_get_new(&sd_index_ida, &index); 3374 spin_unlock(&sd_index_lock); 3375 } while (error == -EAGAIN); 3376 3377 if (error) { 3378 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n"); 3379 goto out_put; 3380 } 3381 3382 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN); 3383 if (error) { 3384 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n"); 3385 goto out_free_index; 3386 } 3387 3388 sdkp->device = sdp; 3389 sdkp->driver = &sd_template; 3390 sdkp->disk = gd; 3391 sdkp->index = index; 3392 atomic_set(&sdkp->openers, 0); 3393 atomic_set(&sdkp->device->ioerr_cnt, 0); 3394 3395 if (!sdp->request_queue->rq_timeout) { 3396 if (sdp->type != TYPE_MOD) 3397 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT); 3398 else 3399 blk_queue_rq_timeout(sdp->request_queue, 3400 SD_MOD_TIMEOUT); 3401 } 3402 3403 device_initialize(&sdkp->dev); 3404 sdkp->dev.parent = dev; 3405 sdkp->dev.class = &sd_disk_class; 3406 dev_set_name(&sdkp->dev, "%s", dev_name(dev)); 3407 3408 error = device_add(&sdkp->dev); 3409 if (error) 3410 goto out_free_index; 3411 3412 get_device(dev); 3413 dev_set_drvdata(dev, sdkp); 3414 3415 get_device(&sdkp->dev); /* prevent release before async_schedule */ 3416 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain); 3417 3418 return 0; 3419 3420 out_free_index: 3421 spin_lock(&sd_index_lock); 3422 ida_remove(&sd_index_ida, index); 3423 spin_unlock(&sd_index_lock); 3424 out_put: 3425 put_disk(gd); 3426 out_free: 3427 kfree(sdkp); 3428 out: 3429 scsi_autopm_put_device(sdp); 3430 return error; 3431 } 3432 3433 /** 3434 * sd_remove - called whenever a scsi disk (previously recognized by 3435 * sd_probe) is detached from the system. It is called (potentially 3436 * multiple times) during sd module unload. 3437 * @dev: pointer to device object 3438 * 3439 * Note: this function is invoked from the scsi mid-level. 3440 * This function potentially frees up a device name (e.g. /dev/sdc) 3441 * that could be re-used by a subsequent sd_probe(). 3442 * This function is not called when the built-in sd driver is "exit-ed". 3443 **/ 3444 static int sd_remove(struct device *dev) 3445 { 3446 struct scsi_disk *sdkp; 3447 dev_t devt; 3448 3449 sdkp = dev_get_drvdata(dev); 3450 devt = disk_devt(sdkp->disk); 3451 scsi_autopm_get_device(sdkp->device); 3452 3453 async_synchronize_full_domain(&scsi_sd_pm_domain); 3454 async_synchronize_full_domain(&scsi_sd_probe_domain); 3455 device_del(&sdkp->dev); 3456 del_gendisk(sdkp->disk); 3457 sd_shutdown(dev); 3458 3459 sd_zbc_remove(sdkp); 3460 3461 free_opal_dev(sdkp->opal_dev); 3462 3463 blk_register_region(devt, SD_MINORS, NULL, 3464 sd_default_probe, NULL, NULL); 3465 3466 mutex_lock(&sd_ref_mutex); 3467 dev_set_drvdata(dev, NULL); 3468 put_device(&sdkp->dev); 3469 mutex_unlock(&sd_ref_mutex); 3470 3471 return 0; 3472 } 3473 3474 /** 3475 * scsi_disk_release - Called to free the scsi_disk structure 3476 * @dev: pointer to embedded class device 3477 * 3478 * sd_ref_mutex must be held entering this routine. Because it is 3479 * called on last put, you should always use the scsi_disk_get() 3480 * scsi_disk_put() helpers which manipulate the semaphore directly 3481 * and never do a direct put_device. 3482 **/ 3483 static void scsi_disk_release(struct device *dev) 3484 { 3485 struct scsi_disk *sdkp = to_scsi_disk(dev); 3486 struct gendisk *disk = sdkp->disk; 3487 3488 spin_lock(&sd_index_lock); 3489 ida_remove(&sd_index_ida, sdkp->index); 3490 spin_unlock(&sd_index_lock); 3491 3492 disk->private_data = NULL; 3493 put_disk(disk); 3494 put_device(&sdkp->device->sdev_gendev); 3495 3496 kfree(sdkp); 3497 } 3498 3499 static int sd_start_stop_device(struct scsi_disk *sdkp, int start) 3500 { 3501 unsigned char cmd[6] = { START_STOP }; /* START_VALID */ 3502 struct scsi_sense_hdr sshdr; 3503 struct scsi_device *sdp = sdkp->device; 3504 int res; 3505 3506 if (start) 3507 cmd[4] |= 1; /* START */ 3508 3509 if (sdp->start_stop_pwr_cond) 3510 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */ 3511 3512 if (!scsi_device_online(sdp)) 3513 return -ENODEV; 3514 3515 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr, 3516 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL); 3517 if (res) { 3518 sd_print_result(sdkp, "Start/Stop Unit failed", res); 3519 if (driver_byte(res) & DRIVER_SENSE) 3520 sd_print_sense_hdr(sdkp, &sshdr); 3521 if (scsi_sense_valid(&sshdr) && 3522 /* 0x3a is medium not present */ 3523 sshdr.asc == 0x3a) 3524 res = 0; 3525 } 3526 3527 /* SCSI error codes must not go to the generic layer */ 3528 if (res) 3529 return -EIO; 3530 3531 return 0; 3532 } 3533 3534 /* 3535 * Send a SYNCHRONIZE CACHE instruction down to the device through 3536 * the normal SCSI command structure. Wait for the command to 3537 * complete. 3538 */ 3539 static void sd_shutdown(struct device *dev) 3540 { 3541 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3542 3543 if (!sdkp) 3544 return; /* this can happen */ 3545 3546 if (pm_runtime_suspended(dev)) 3547 return; 3548 3549 if (sdkp->WCE && sdkp->media_present) { 3550 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n"); 3551 sd_sync_cache(sdkp, NULL); 3552 } 3553 3554 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) { 3555 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n"); 3556 sd_start_stop_device(sdkp, 0); 3557 } 3558 } 3559 3560 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors) 3561 { 3562 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3563 struct scsi_sense_hdr sshdr; 3564 int ret = 0; 3565 3566 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */ 3567 return 0; 3568 3569 if (sdkp->WCE && sdkp->media_present) { 3570 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n"); 3571 ret = sd_sync_cache(sdkp, &sshdr); 3572 3573 if (ret) { 3574 /* ignore OFFLINE device */ 3575 if (ret == -ENODEV) 3576 return 0; 3577 3578 if (!scsi_sense_valid(&sshdr) || 3579 sshdr.sense_key != ILLEGAL_REQUEST) 3580 return ret; 3581 3582 /* 3583 * sshdr.sense_key == ILLEGAL_REQUEST means this drive 3584 * doesn't support sync. There's not much to do and 3585 * suspend shouldn't fail. 3586 */ 3587 ret = 0; 3588 } 3589 } 3590 3591 if (sdkp->device->manage_start_stop) { 3592 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n"); 3593 /* an error is not worth aborting a system sleep */ 3594 ret = sd_start_stop_device(sdkp, 0); 3595 if (ignore_stop_errors) 3596 ret = 0; 3597 } 3598 3599 return ret; 3600 } 3601 3602 static int sd_suspend_system(struct device *dev) 3603 { 3604 return sd_suspend_common(dev, true); 3605 } 3606 3607 static int sd_suspend_runtime(struct device *dev) 3608 { 3609 return sd_suspend_common(dev, false); 3610 } 3611 3612 static int sd_resume(struct device *dev) 3613 { 3614 struct scsi_disk *sdkp = dev_get_drvdata(dev); 3615 int ret; 3616 3617 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */ 3618 return 0; 3619 3620 if (!sdkp->device->manage_start_stop) 3621 return 0; 3622 3623 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n"); 3624 ret = sd_start_stop_device(sdkp, 1); 3625 if (!ret) 3626 opal_unlock_from_suspend(sdkp->opal_dev); 3627 return ret; 3628 } 3629 3630 /** 3631 * init_sd - entry point for this driver (both when built in or when 3632 * a module). 3633 * 3634 * Note: this function registers this driver with the scsi mid-level. 3635 **/ 3636 static int __init init_sd(void) 3637 { 3638 int majors = 0, i, err; 3639 3640 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n")); 3641 3642 for (i = 0; i < SD_MAJORS; i++) { 3643 if (register_blkdev(sd_major(i), "sd") != 0) 3644 continue; 3645 majors++; 3646 blk_register_region(sd_major(i), SD_MINORS, NULL, 3647 sd_default_probe, NULL, NULL); 3648 } 3649 3650 if (!majors) 3651 return -ENODEV; 3652 3653 err = class_register(&sd_disk_class); 3654 if (err) 3655 goto err_out; 3656 3657 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE, 3658 0, 0, NULL); 3659 if (!sd_cdb_cache) { 3660 printk(KERN_ERR "sd: can't init extended cdb cache\n"); 3661 err = -ENOMEM; 3662 goto err_out_class; 3663 } 3664 3665 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache); 3666 if (!sd_cdb_pool) { 3667 printk(KERN_ERR "sd: can't init extended cdb pool\n"); 3668 err = -ENOMEM; 3669 goto err_out_cache; 3670 } 3671 3672 err = scsi_register_driver(&sd_template.gendrv); 3673 if (err) 3674 goto err_out_driver; 3675 3676 return 0; 3677 3678 err_out_driver: 3679 mempool_destroy(sd_cdb_pool); 3680 3681 err_out_cache: 3682 kmem_cache_destroy(sd_cdb_cache); 3683 3684 err_out_class: 3685 class_unregister(&sd_disk_class); 3686 err_out: 3687 for (i = 0; i < SD_MAJORS; i++) 3688 unregister_blkdev(sd_major(i), "sd"); 3689 return err; 3690 } 3691 3692 /** 3693 * exit_sd - exit point for this driver (when it is a module). 3694 * 3695 * Note: this function unregisters this driver from the scsi mid-level. 3696 **/ 3697 static void __exit exit_sd(void) 3698 { 3699 int i; 3700 3701 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n")); 3702 3703 scsi_unregister_driver(&sd_template.gendrv); 3704 mempool_destroy(sd_cdb_pool); 3705 kmem_cache_destroy(sd_cdb_cache); 3706 3707 class_unregister(&sd_disk_class); 3708 3709 for (i = 0; i < SD_MAJORS; i++) { 3710 blk_unregister_region(sd_major(i), SD_MINORS); 3711 unregister_blkdev(sd_major(i), "sd"); 3712 } 3713 } 3714 3715 module_init(init_sd); 3716 module_exit(exit_sd); 3717 3718 static void sd_print_sense_hdr(struct scsi_disk *sdkp, 3719 struct scsi_sense_hdr *sshdr) 3720 { 3721 scsi_print_sense_hdr(sdkp->device, 3722 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr); 3723 } 3724 3725 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg, 3726 int result) 3727 { 3728 const char *hb_string = scsi_hostbyte_string(result); 3729 const char *db_string = scsi_driverbyte_string(result); 3730 3731 if (hb_string || db_string) 3732 sd_printk(KERN_INFO, sdkp, 3733 "%s: Result: hostbyte=%s driverbyte=%s\n", msg, 3734 hb_string ? hb_string : "invalid", 3735 db_string ? db_string : "invalid"); 3736 else 3737 sd_printk(KERN_INFO, sdkp, 3738 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n", 3739 msg, host_byte(result), driver_byte(result)); 3740 } 3741 3742