1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Functions related to setting various queue properties from drivers 4 */ 5 #include <linux/kernel.h> 6 #include <linux/module.h> 7 #include <linux/init.h> 8 #include <linux/bio.h> 9 #include <linux/blk-integrity.h> 10 #include <linux/pagemap.h> 11 #include <linux/backing-dev-defs.h> 12 #include <linux/gcd.h> 13 #include <linux/lcm.h> 14 #include <linux/jiffies.h> 15 #include <linux/gfp.h> 16 #include <linux/dma-mapping.h> 17 #include <linux/t10-pi.h> 18 #include <linux/crc64.h> 19 20 #include "blk.h" 21 #include "blk-rq-qos.h" 22 #include "blk-wbt.h" 23 24 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) 25 { 26 WRITE_ONCE(q->rq_timeout, timeout); 27 } 28 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); 29 30 /** 31 * blk_set_stacking_limits - set default limits for stacking devices 32 * @lim: the queue_limits structure to reset 33 * 34 * Prepare queue limits for applying limits from underlying devices using 35 * blk_stack_limits(). 36 */ 37 void blk_set_stacking_limits(struct queue_limits *lim) 38 { 39 memset(lim, 0, sizeof(*lim)); 40 lim->logical_block_size = SECTOR_SIZE; 41 lim->physical_block_size = SECTOR_SIZE; 42 lim->io_min = SECTOR_SIZE; 43 lim->discard_granularity = SECTOR_SIZE; 44 lim->dma_alignment = SECTOR_SIZE - 1; 45 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; 46 47 /* Inherit limits from component devices */ 48 lim->max_segments = USHRT_MAX; 49 lim->max_discard_segments = USHRT_MAX; 50 lim->max_hw_sectors = UINT_MAX; 51 lim->max_segment_size = UINT_MAX; 52 lim->max_sectors = UINT_MAX; 53 lim->max_dev_sectors = UINT_MAX; 54 lim->max_write_zeroes_sectors = UINT_MAX; 55 lim->max_hw_wzeroes_unmap_sectors = UINT_MAX; 56 lim->max_user_wzeroes_unmap_sectors = UINT_MAX; 57 lim->max_hw_zone_append_sectors = UINT_MAX; 58 lim->max_user_discard_sectors = UINT_MAX; 59 lim->atomic_write_hw_max = UINT_MAX; 60 } 61 EXPORT_SYMBOL(blk_set_stacking_limits); 62 63 void blk_apply_bdi_limits(struct backing_dev_info *bdi, 64 struct queue_limits *lim) 65 { 66 u64 io_opt = lim->io_opt; 67 68 /* 69 * For read-ahead of large files to be effective, we need to read ahead 70 * at least twice the optimal I/O size. For rotational devices that do 71 * not report an optimal I/O size (e.g. ATA HDDs), use the maximum I/O 72 * size to avoid falling back to the (rather inefficient) small default 73 * read-ahead size. 74 * 75 * There is no hardware limitation for the read-ahead size and the user 76 * might have increased the read-ahead size through sysfs, so don't ever 77 * decrease it. 78 */ 79 if (!io_opt && (lim->features & BLK_FEAT_ROTATIONAL)) 80 io_opt = (u64)lim->max_sectors << SECTOR_SHIFT; 81 82 bdi->ra_pages = max3(bdi->ra_pages, 83 io_opt * 2 >> PAGE_SHIFT, 84 VM_READAHEAD_PAGES); 85 bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT; 86 } 87 88 static int blk_validate_zoned_limits(struct queue_limits *lim) 89 { 90 if (!(lim->features & BLK_FEAT_ZONED)) { 91 if (WARN_ON_ONCE(lim->max_open_zones) || 92 WARN_ON_ONCE(lim->max_active_zones) || 93 WARN_ON_ONCE(lim->zone_write_granularity) || 94 WARN_ON_ONCE(lim->max_zone_append_sectors)) 95 return -EINVAL; 96 return 0; 97 } 98 99 if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED))) 100 return -EINVAL; 101 102 /* 103 * Given that active zones include open zones, the maximum number of 104 * open zones cannot be larger than the maximum number of active zones. 105 */ 106 if (lim->max_active_zones && 107 lim->max_open_zones > lim->max_active_zones) 108 return -EINVAL; 109 110 if (lim->zone_write_granularity < lim->logical_block_size) 111 lim->zone_write_granularity = lim->logical_block_size; 112 113 /* 114 * The Zone Append size is limited by the maximum I/O size and the zone 115 * size given that it can't span zones. 116 * 117 * If no max_hw_zone_append_sectors limit is provided, the block layer 118 * will emulated it, else we're also bound by the hardware limit. 119 */ 120 lim->max_zone_append_sectors = 121 min_not_zero(lim->max_hw_zone_append_sectors, 122 min(lim->chunk_sectors, lim->max_hw_sectors)); 123 return 0; 124 } 125 126 /* 127 * Maximum size of I/O that needs a block layer integrity buffer. Limited 128 * by the number of intervals for which we can fit the integrity buffer into 129 * the buffer size. Because the buffer is a single segment it is also limited 130 * by the maximum segment size. 131 */ 132 static inline unsigned int max_integrity_io_size(struct queue_limits *lim) 133 { 134 return min_t(unsigned int, lim->max_segment_size, 135 (BLK_INTEGRITY_MAX_SIZE / lim->integrity.metadata_size) << 136 lim->integrity.interval_exp); 137 } 138 139 static int blk_validate_integrity_limits(struct queue_limits *lim) 140 { 141 struct blk_integrity *bi = &lim->integrity; 142 143 if (!bi->metadata_size) { 144 if (bi->csum_type != BLK_INTEGRITY_CSUM_NONE || 145 bi->tag_size || ((bi->flags & BLK_INTEGRITY_REF_TAG))) { 146 pr_warn("invalid PI settings.\n"); 147 return -EINVAL; 148 } 149 bi->flags |= BLK_INTEGRITY_NOGENERATE | BLK_INTEGRITY_NOVERIFY; 150 return 0; 151 } 152 153 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) { 154 pr_warn("integrity support disabled.\n"); 155 return -EINVAL; 156 } 157 158 if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE && 159 (bi->flags & BLK_INTEGRITY_REF_TAG)) { 160 pr_warn("ref tag not support without checksum.\n"); 161 return -EINVAL; 162 } 163 164 if (bi->pi_tuple_size > bi->metadata_size) { 165 pr_warn("pi_tuple_size (%u) exceeds metadata_size (%u)\n", 166 bi->pi_tuple_size, 167 bi->metadata_size); 168 return -EINVAL; 169 } 170 171 switch (bi->csum_type) { 172 case BLK_INTEGRITY_CSUM_NONE: 173 if (bi->pi_tuple_size) { 174 pr_warn("pi_tuple_size must be 0 when checksum type is none\n"); 175 return -EINVAL; 176 } 177 break; 178 case BLK_INTEGRITY_CSUM_CRC: 179 case BLK_INTEGRITY_CSUM_IP: 180 if (bi->pi_tuple_size != sizeof(struct t10_pi_tuple)) { 181 pr_warn("pi_tuple_size mismatch for T10 PI: expected %zu, got %u\n", 182 sizeof(struct t10_pi_tuple), 183 bi->pi_tuple_size); 184 return -EINVAL; 185 } 186 break; 187 case BLK_INTEGRITY_CSUM_CRC64: 188 if (bi->pi_tuple_size != sizeof(struct crc64_pi_tuple)) { 189 pr_warn("pi_tuple_size mismatch for CRC64 PI: expected %zu, got %u\n", 190 sizeof(struct crc64_pi_tuple), 191 bi->pi_tuple_size); 192 return -EINVAL; 193 } 194 break; 195 } 196 197 if (!bi->interval_exp) 198 bi->interval_exp = ilog2(lim->logical_block_size); 199 200 /* 201 * The block layer automatically adds integrity data for bios that don't 202 * already have it. Limit the I/O size so that a single maximum size 203 * metadata segment can cover the integrity data for the entire I/O. 204 */ 205 lim->max_sectors = min(lim->max_sectors, 206 max_integrity_io_size(lim) >> SECTOR_SHIFT); 207 208 return 0; 209 } 210 211 /* 212 * Returns max guaranteed bytes which we can fit in a bio. 213 * 214 * We request that an atomic_write is ITER_UBUF iov_iter (so a single vector), 215 * so we assume that we can fit in at least PAGE_SIZE in a segment, apart from 216 * the first and last segments. 217 */ 218 static unsigned int blk_queue_max_guaranteed_bio(struct queue_limits *lim) 219 { 220 unsigned int max_segments = min(BIO_MAX_VECS, lim->max_segments); 221 unsigned int length; 222 223 length = min(max_segments, 2) * lim->logical_block_size; 224 if (max_segments > 2) 225 length += (max_segments - 2) * PAGE_SIZE; 226 227 return length; 228 } 229 230 static void blk_atomic_writes_update_limits(struct queue_limits *lim) 231 { 232 unsigned int unit_limit = min(lim->max_hw_sectors << SECTOR_SHIFT, 233 blk_queue_max_guaranteed_bio(lim)); 234 235 unit_limit = rounddown_pow_of_two(unit_limit); 236 237 lim->atomic_write_max_sectors = 238 min(lim->atomic_write_hw_max >> SECTOR_SHIFT, 239 lim->max_hw_sectors); 240 lim->atomic_write_unit_min = 241 min(lim->atomic_write_hw_unit_min, unit_limit); 242 lim->atomic_write_unit_max = 243 min(lim->atomic_write_hw_unit_max, unit_limit); 244 lim->atomic_write_boundary_sectors = 245 lim->atomic_write_hw_boundary >> SECTOR_SHIFT; 246 } 247 248 /* 249 * Test whether any boundary is aligned with any chunk size. Stacked 250 * devices store any stripe size in t->chunk_sectors. 251 */ 252 static bool blk_valid_atomic_writes_boundary(unsigned int chunk_sectors, 253 unsigned int boundary_sectors) 254 { 255 if (!chunk_sectors || !boundary_sectors) 256 return true; 257 258 if (boundary_sectors > chunk_sectors && 259 boundary_sectors % chunk_sectors) 260 return false; 261 262 if (chunk_sectors > boundary_sectors && 263 chunk_sectors % boundary_sectors) 264 return false; 265 266 return true; 267 } 268 269 static void blk_validate_atomic_write_limits(struct queue_limits *lim) 270 { 271 unsigned int boundary_sectors; 272 unsigned int atomic_write_hw_max_sectors = 273 lim->atomic_write_hw_max >> SECTOR_SHIFT; 274 275 if (!(lim->features & BLK_FEAT_ATOMIC_WRITES)) 276 goto unsupported; 277 278 /* UINT_MAX indicates stacked limits in initial state */ 279 if (lim->atomic_write_hw_max == UINT_MAX) 280 goto unsupported; 281 282 if (!lim->atomic_write_hw_max) 283 goto unsupported; 284 285 if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_min))) 286 goto unsupported; 287 288 if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_max))) 289 goto unsupported; 290 291 if (WARN_ON_ONCE(lim->atomic_write_hw_unit_min > 292 lim->atomic_write_hw_unit_max)) 293 goto unsupported; 294 295 if (WARN_ON_ONCE(lim->atomic_write_hw_unit_max > 296 lim->atomic_write_hw_max)) 297 goto unsupported; 298 299 if (WARN_ON_ONCE(lim->chunk_sectors && 300 atomic_write_hw_max_sectors > lim->chunk_sectors)) 301 goto unsupported; 302 303 boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT; 304 305 if (boundary_sectors) { 306 if (WARN_ON_ONCE(lim->atomic_write_hw_max > 307 lim->atomic_write_hw_boundary)) 308 goto unsupported; 309 310 if (WARN_ON_ONCE(!blk_valid_atomic_writes_boundary( 311 lim->chunk_sectors, boundary_sectors))) 312 goto unsupported; 313 314 /* 315 * The boundary size just needs to be a multiple of unit_max 316 * (and not necessarily a power-of-2), so this following check 317 * could be relaxed in future. 318 * Furthermore, if needed, unit_max could even be reduced so 319 * that it is compliant with a !power-of-2 boundary. 320 */ 321 if (!is_power_of_2(boundary_sectors)) 322 goto unsupported; 323 } 324 325 blk_atomic_writes_update_limits(lim); 326 return; 327 328 unsupported: 329 lim->atomic_write_max_sectors = 0; 330 lim->atomic_write_boundary_sectors = 0; 331 lim->atomic_write_unit_min = 0; 332 lim->atomic_write_unit_max = 0; 333 } 334 335 /* 336 * Check that the limits in lim are valid, initialize defaults for unset 337 * values, and cap values based on others where needed. 338 */ 339 int blk_validate_limits(struct queue_limits *lim) 340 { 341 unsigned int max_hw_sectors; 342 unsigned int logical_block_sectors; 343 unsigned long seg_size; 344 int err; 345 346 /* 347 * Unless otherwise specified, default to 512 byte logical blocks and a 348 * physical block size equal to the logical block size. 349 */ 350 if (!lim->logical_block_size) 351 lim->logical_block_size = SECTOR_SIZE; 352 else if (blk_validate_block_size(lim->logical_block_size)) { 353 pr_warn("Invalid logical block size (%d)\n", lim->logical_block_size); 354 return -EINVAL; 355 } 356 if (lim->physical_block_size < lim->logical_block_size) { 357 lim->physical_block_size = lim->logical_block_size; 358 } else if (!is_power_of_2(lim->physical_block_size)) { 359 pr_warn("Invalid physical block size (%d)\n", lim->physical_block_size); 360 return -EINVAL; 361 } 362 363 /* 364 * The minimum I/O size defaults to the physical block size unless 365 * explicitly overridden. 366 */ 367 if (lim->io_min < lim->physical_block_size) 368 lim->io_min = lim->physical_block_size; 369 370 /* 371 * The optimal I/O size may not be aligned to physical block size 372 * (because it may be limited by dma engines which have no clue about 373 * block size of the disks attached to them), so we round it down here. 374 */ 375 lim->io_opt = round_down(lim->io_opt, lim->physical_block_size); 376 377 /* 378 * max_hw_sectors has a somewhat weird default for historical reason, 379 * but driver really should set their own instead of relying on this 380 * value. 381 * 382 * The block layer relies on the fact that every driver can 383 * handle at lest a page worth of data per I/O, and needs the value 384 * aligned to the logical block size. 385 */ 386 if (!lim->max_hw_sectors) 387 lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; 388 if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS)) 389 return -EINVAL; 390 logical_block_sectors = lim->logical_block_size >> SECTOR_SHIFT; 391 if (WARN_ON_ONCE(logical_block_sectors > lim->max_hw_sectors)) 392 return -EINVAL; 393 lim->max_hw_sectors = round_down(lim->max_hw_sectors, 394 logical_block_sectors); 395 396 /* 397 * The actual max_sectors value is a complex beast and also takes the 398 * max_dev_sectors value (set by SCSI ULPs) and a user configurable 399 * value into account. The ->max_sectors value is always calculated 400 * from these, so directly setting it won't have any effect. 401 */ 402 max_hw_sectors = min_not_zero(lim->max_hw_sectors, 403 lim->max_dev_sectors); 404 if (lim->max_user_sectors) { 405 if (lim->max_user_sectors < BLK_MIN_SEGMENT_SIZE / SECTOR_SIZE) 406 return -EINVAL; 407 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors); 408 } else if (lim->io_opt > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) { 409 lim->max_sectors = 410 min(max_hw_sectors, lim->io_opt >> SECTOR_SHIFT); 411 } else if (lim->io_min > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) { 412 lim->max_sectors = 413 min(max_hw_sectors, lim->io_min >> SECTOR_SHIFT); 414 } else { 415 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP); 416 } 417 lim->max_sectors = round_down(lim->max_sectors, 418 logical_block_sectors); 419 420 /* 421 * Random default for the maximum number of segments. Driver should not 422 * rely on this and set their own. 423 */ 424 if (!lim->max_segments) 425 lim->max_segments = BLK_MAX_SEGMENTS; 426 427 if (lim->max_hw_wzeroes_unmap_sectors && 428 lim->max_hw_wzeroes_unmap_sectors != lim->max_write_zeroes_sectors) 429 return -EINVAL; 430 lim->max_wzeroes_unmap_sectors = min(lim->max_hw_wzeroes_unmap_sectors, 431 lim->max_user_wzeroes_unmap_sectors); 432 433 lim->max_discard_sectors = 434 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors); 435 436 /* 437 * When discard is not supported, discard_granularity should be reported 438 * as 0 to userspace. 439 */ 440 if (lim->max_discard_sectors) 441 lim->discard_granularity = 442 max(lim->discard_granularity, lim->physical_block_size); 443 else 444 lim->discard_granularity = 0; 445 446 if (!lim->max_discard_segments) 447 lim->max_discard_segments = 1; 448 449 /* 450 * By default there is no limit on the segment boundary alignment, 451 * but if there is one it can't be smaller than the page size as 452 * that would break all the normal I/O patterns. 453 */ 454 if (!lim->seg_boundary_mask) 455 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; 456 if (WARN_ON_ONCE(lim->seg_boundary_mask < BLK_MIN_SEGMENT_SIZE - 1)) 457 return -EINVAL; 458 459 /* 460 * Stacking device may have both virtual boundary and max segment 461 * size limit, so allow this setting now, and long-term the two 462 * might need to move out of stacking limits since we have immutable 463 * bvec and lower layer bio splitting is supposed to handle the two 464 * correctly. 465 */ 466 if (lim->virt_boundary_mask) { 467 if (!lim->max_segment_size) 468 lim->max_segment_size = UINT_MAX; 469 } else { 470 /* 471 * The maximum segment size has an odd historic 64k default that 472 * drivers probably should override. Just like the I/O size we 473 * require drivers to at least handle a full page per segment. 474 */ 475 if (!lim->max_segment_size) 476 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; 477 if (WARN_ON_ONCE(lim->max_segment_size < BLK_MIN_SEGMENT_SIZE)) 478 return -EINVAL; 479 } 480 481 /* setup max segment size for building new segment in fast path */ 482 if (lim->seg_boundary_mask > lim->max_segment_size - 1) 483 seg_size = lim->max_segment_size; 484 else 485 seg_size = lim->seg_boundary_mask + 1; 486 lim->max_fast_segment_size = min_t(unsigned int, seg_size, PAGE_SIZE); 487 488 /* 489 * We require drivers to at least do logical block aligned I/O, but 490 * historically could not check for that due to the separate calls 491 * to set the limits. Once the transition is finished the check 492 * below should be narrowed down to check the logical block size. 493 */ 494 if (!lim->dma_alignment) 495 lim->dma_alignment = SECTOR_SIZE - 1; 496 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE)) 497 return -EINVAL; 498 499 if (lim->alignment_offset) { 500 lim->alignment_offset &= (lim->physical_block_size - 1); 501 lim->flags &= ~BLK_FLAG_MISALIGNED; 502 } 503 504 if (!(lim->features & BLK_FEAT_WRITE_CACHE)) 505 lim->features &= ~BLK_FEAT_FUA; 506 507 blk_validate_atomic_write_limits(lim); 508 509 err = blk_validate_integrity_limits(lim); 510 if (err) 511 return err; 512 return blk_validate_zoned_limits(lim); 513 } 514 EXPORT_SYMBOL_GPL(blk_validate_limits); 515 516 /* 517 * Set the default limits for a newly allocated queue. @lim contains the 518 * initial limits set by the driver, which could be no limit in which case 519 * all fields are cleared to zero. 520 */ 521 int blk_set_default_limits(struct queue_limits *lim) 522 { 523 /* 524 * Most defaults are set by capping the bounds in blk_validate_limits, 525 * but these limits are special and need an explicit initialization to 526 * the max value here. 527 */ 528 lim->max_user_discard_sectors = UINT_MAX; 529 lim->max_user_wzeroes_unmap_sectors = UINT_MAX; 530 return blk_validate_limits(lim); 531 } 532 533 /** 534 * queue_limits_commit_update - commit an atomic update of queue limits 535 * @q: queue to update 536 * @lim: limits to apply 537 * 538 * Apply the limits in @lim that were obtained from queue_limits_start_update() 539 * and updated by the caller to @q. The caller must have frozen the queue or 540 * ensure that there are no outstanding I/Os by other means. 541 * 542 * Returns 0 if successful, else a negative error code. 543 */ 544 int queue_limits_commit_update(struct request_queue *q, 545 struct queue_limits *lim) 546 { 547 int error; 548 549 lockdep_assert_held(&q->limits_lock); 550 551 error = blk_validate_limits(lim); 552 if (error) 553 goto out_unlock; 554 555 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 556 if (q->crypto_profile && lim->integrity.tag_size) { 557 pr_warn("blk-integrity: Integrity and hardware inline encryption are not supported together.\n"); 558 error = -EINVAL; 559 goto out_unlock; 560 } 561 #endif 562 563 q->limits = *lim; 564 if (q->disk) 565 blk_apply_bdi_limits(q->disk->bdi, lim); 566 out_unlock: 567 mutex_unlock(&q->limits_lock); 568 return error; 569 } 570 EXPORT_SYMBOL_GPL(queue_limits_commit_update); 571 572 /** 573 * queue_limits_commit_update_frozen - commit an atomic update of queue limits 574 * @q: queue to update 575 * @lim: limits to apply 576 * 577 * Apply the limits in @lim that were obtained from queue_limits_start_update() 578 * and updated with the new values by the caller to @q. Freezes the queue 579 * before the update and unfreezes it after. 580 * 581 * Returns 0 if successful, else a negative error code. 582 */ 583 int queue_limits_commit_update_frozen(struct request_queue *q, 584 struct queue_limits *lim) 585 { 586 unsigned int memflags; 587 int ret; 588 589 memflags = blk_mq_freeze_queue(q); 590 ret = queue_limits_commit_update(q, lim); 591 blk_mq_unfreeze_queue(q, memflags); 592 593 return ret; 594 } 595 EXPORT_SYMBOL_GPL(queue_limits_commit_update_frozen); 596 597 /** 598 * queue_limits_set - apply queue limits to queue 599 * @q: queue to update 600 * @lim: limits to apply 601 * 602 * Apply the limits in @lim that were freshly initialized to @q. 603 * To update existing limits use queue_limits_start_update() and 604 * queue_limits_commit_update() instead. 605 * 606 * Returns 0 if successful, else a negative error code. 607 */ 608 int queue_limits_set(struct request_queue *q, struct queue_limits *lim) 609 { 610 mutex_lock(&q->limits_lock); 611 return queue_limits_commit_update(q, lim); 612 } 613 EXPORT_SYMBOL_GPL(queue_limits_set); 614 615 static int queue_limit_alignment_offset(const struct queue_limits *lim, 616 sector_t sector) 617 { 618 unsigned int granularity = max(lim->physical_block_size, lim->io_min); 619 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT) 620 << SECTOR_SHIFT; 621 622 return (granularity + lim->alignment_offset - alignment) % granularity; 623 } 624 625 static unsigned int queue_limit_discard_alignment( 626 const struct queue_limits *lim, sector_t sector) 627 { 628 unsigned int alignment, granularity, offset; 629 630 if (!lim->max_discard_sectors) 631 return 0; 632 633 /* Why are these in bytes, not sectors? */ 634 alignment = lim->discard_alignment >> SECTOR_SHIFT; 635 granularity = lim->discard_granularity >> SECTOR_SHIFT; 636 637 /* Offset of the partition start in 'granularity' sectors */ 638 offset = sector_div(sector, granularity); 639 640 /* And why do we do this modulus *again* in blkdev_issue_discard()? */ 641 offset = (granularity + alignment - offset) % granularity; 642 643 /* Turn it back into bytes, gaah */ 644 return offset << SECTOR_SHIFT; 645 } 646 647 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs) 648 { 649 sectors = round_down(sectors, lbs >> SECTOR_SHIFT); 650 if (sectors < PAGE_SIZE >> SECTOR_SHIFT) 651 sectors = PAGE_SIZE >> SECTOR_SHIFT; 652 return sectors; 653 } 654 655 /* Check if second and later bottom devices are compliant */ 656 static bool blk_stack_atomic_writes_tail(struct queue_limits *t, 657 struct queue_limits *b) 658 { 659 /* We're not going to support different boundary sizes.. yet */ 660 if (t->atomic_write_hw_boundary != b->atomic_write_hw_boundary) 661 return false; 662 663 /* Can't support this */ 664 if (t->atomic_write_hw_unit_min > b->atomic_write_hw_unit_max) 665 return false; 666 667 /* Or this */ 668 if (t->atomic_write_hw_unit_max < b->atomic_write_hw_unit_min) 669 return false; 670 671 t->atomic_write_hw_max = min(t->atomic_write_hw_max, 672 b->atomic_write_hw_max); 673 t->atomic_write_hw_unit_min = max(t->atomic_write_hw_unit_min, 674 b->atomic_write_hw_unit_min); 675 t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max, 676 b->atomic_write_hw_unit_max); 677 return true; 678 } 679 680 static void blk_stack_atomic_writes_chunk_sectors(struct queue_limits *t) 681 { 682 unsigned int chunk_bytes; 683 684 if (!t->chunk_sectors) 685 return; 686 687 /* 688 * If chunk sectors is so large that its value in bytes overflows 689 * UINT_MAX, then just shift it down so it definitely will fit. 690 * We don't support atomic writes of such a large size anyway. 691 */ 692 if (check_shl_overflow(t->chunk_sectors, SECTOR_SHIFT, &chunk_bytes)) 693 chunk_bytes = t->chunk_sectors; 694 695 /* 696 * Find values for limits which work for chunk size. 697 * b->atomic_write_hw_unit_{min, max} may not be aligned with chunk 698 * size, as the chunk size is not restricted to a power-of-2. 699 * So we need to find highest power-of-2 which works for the chunk 700 * size. 701 * As an example scenario, we could have t->unit_max = 16K and 702 * t->chunk_sectors = 24KB. For this case, reduce t->unit_max to a 703 * value aligned with both limits, i.e. 8K in this example. 704 */ 705 t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max, 706 max_pow_of_two_factor(chunk_bytes)); 707 708 t->atomic_write_hw_unit_min = min(t->atomic_write_hw_unit_min, 709 t->atomic_write_hw_unit_max); 710 t->atomic_write_hw_max = min(t->atomic_write_hw_max, chunk_bytes); 711 } 712 713 /* Check stacking of first bottom device */ 714 static bool blk_stack_atomic_writes_head(struct queue_limits *t, 715 struct queue_limits *b) 716 { 717 if (!blk_valid_atomic_writes_boundary(t->chunk_sectors, 718 b->atomic_write_hw_boundary >> SECTOR_SHIFT)) 719 return false; 720 721 t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max; 722 t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min; 723 t->atomic_write_hw_max = b->atomic_write_hw_max; 724 t->atomic_write_hw_boundary = b->atomic_write_hw_boundary; 725 return true; 726 } 727 728 static void blk_stack_atomic_writes_limits(struct queue_limits *t, 729 struct queue_limits *b, sector_t start) 730 { 731 if (!(b->features & BLK_FEAT_ATOMIC_WRITES)) 732 goto unsupported; 733 734 if (!b->atomic_write_hw_unit_min) 735 goto unsupported; 736 737 if (!blk_atomic_write_start_sect_aligned(start, b)) 738 goto unsupported; 739 740 /* UINT_MAX indicates no stacking of bottom devices yet */ 741 if (t->atomic_write_hw_max == UINT_MAX) { 742 if (!blk_stack_atomic_writes_head(t, b)) 743 goto unsupported; 744 } else { 745 if (!blk_stack_atomic_writes_tail(t, b)) 746 goto unsupported; 747 } 748 blk_stack_atomic_writes_chunk_sectors(t); 749 return; 750 751 unsupported: 752 t->atomic_write_hw_max = 0; 753 t->atomic_write_hw_unit_max = 0; 754 t->atomic_write_hw_unit_min = 0; 755 t->atomic_write_hw_boundary = 0; 756 } 757 758 /** 759 * blk_stack_limits - adjust queue_limits for stacked devices 760 * @t: the stacking driver limits (top device) 761 * @b: the underlying queue limits (bottom, component device) 762 * @start: first data sector within component device 763 * 764 * Description: 765 * This function is used by stacking drivers like MD and DM to ensure 766 * that all component devices have compatible block sizes and 767 * alignments. The stacking driver must provide a queue_limits 768 * struct (top) and then iteratively call the stacking function for 769 * all component (bottom) devices. The stacking function will 770 * attempt to combine the values and ensure proper alignment. 771 * 772 * Returns 0 if the top and bottom queue_limits are compatible. The 773 * top device's block sizes and alignment offsets may be adjusted to 774 * ensure alignment with the bottom device. If no compatible sizes 775 * and alignments exist, -1 is returned and the resulting top 776 * queue_limits will have the misaligned flag set to indicate that 777 * the alignment_offset is undefined. 778 */ 779 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, 780 sector_t start) 781 { 782 unsigned int top, bottom, alignment; 783 int ret = 0; 784 785 t->features |= (b->features & BLK_FEAT_INHERIT_MASK); 786 787 /* 788 * Some feaures need to be supported both by the stacking driver and all 789 * underlying devices. The stacking driver sets these flags before 790 * stacking the limits, and this will clear the flags if any of the 791 * underlying devices does not support it. 792 */ 793 if (!(b->features & BLK_FEAT_NOWAIT)) 794 t->features &= ~BLK_FEAT_NOWAIT; 795 if (!(b->features & BLK_FEAT_POLL)) 796 t->features &= ~BLK_FEAT_POLL; 797 798 t->flags |= (b->flags & BLK_FLAG_MISALIGNED); 799 800 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); 801 t->max_user_sectors = min_not_zero(t->max_user_sectors, 802 b->max_user_sectors); 803 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); 804 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); 805 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, 806 b->max_write_zeroes_sectors); 807 t->max_user_wzeroes_unmap_sectors = 808 min(t->max_user_wzeroes_unmap_sectors, 809 b->max_user_wzeroes_unmap_sectors); 810 t->max_hw_wzeroes_unmap_sectors = 811 min(t->max_hw_wzeroes_unmap_sectors, 812 b->max_hw_wzeroes_unmap_sectors); 813 814 t->max_hw_zone_append_sectors = min(t->max_hw_zone_append_sectors, 815 b->max_hw_zone_append_sectors); 816 817 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, 818 b->seg_boundary_mask); 819 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, 820 b->virt_boundary_mask); 821 822 t->max_segments = min_not_zero(t->max_segments, b->max_segments); 823 t->max_discard_segments = min_not_zero(t->max_discard_segments, 824 b->max_discard_segments); 825 t->max_integrity_segments = min_not_zero(t->max_integrity_segments, 826 b->max_integrity_segments); 827 828 t->max_segment_size = min_not_zero(t->max_segment_size, 829 b->max_segment_size); 830 831 alignment = queue_limit_alignment_offset(b, start); 832 833 /* Bottom device has different alignment. Check that it is 834 * compatible with the current top alignment. 835 */ 836 if (t->alignment_offset != alignment) { 837 838 top = max(t->physical_block_size, t->io_min) 839 + t->alignment_offset; 840 bottom = max(b->physical_block_size, b->io_min) + alignment; 841 842 /* Verify that top and bottom intervals line up */ 843 if (max(top, bottom) % min(top, bottom)) { 844 t->flags |= BLK_FLAG_MISALIGNED; 845 ret = -1; 846 } 847 } 848 849 t->logical_block_size = max(t->logical_block_size, 850 b->logical_block_size); 851 852 t->physical_block_size = max(t->physical_block_size, 853 b->physical_block_size); 854 855 t->io_min = max(t->io_min, b->io_min); 856 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt); 857 t->dma_alignment = max(t->dma_alignment, b->dma_alignment); 858 859 /* Set non-power-of-2 compatible chunk_sectors boundary */ 860 if (b->chunk_sectors) 861 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors); 862 863 /* Physical block size a multiple of the logical block size? */ 864 if (t->physical_block_size & (t->logical_block_size - 1)) { 865 t->physical_block_size = t->logical_block_size; 866 t->flags |= BLK_FLAG_MISALIGNED; 867 ret = -1; 868 } 869 870 /* Minimum I/O a multiple of the physical block size? */ 871 if (t->io_min & (t->physical_block_size - 1)) { 872 t->io_min = t->physical_block_size; 873 t->flags |= BLK_FLAG_MISALIGNED; 874 ret = -1; 875 } 876 877 /* Optimal I/O a multiple of the physical block size? */ 878 if (t->io_opt & (t->physical_block_size - 1)) { 879 t->io_opt = 0; 880 t->flags |= BLK_FLAG_MISALIGNED; 881 ret = -1; 882 } 883 884 /* chunk_sectors a multiple of the physical block size? */ 885 if (t->chunk_sectors % (t->physical_block_size >> SECTOR_SHIFT)) { 886 t->chunk_sectors = 0; 887 t->flags |= BLK_FLAG_MISALIGNED; 888 ret = -1; 889 } 890 891 /* Find lowest common alignment_offset */ 892 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment) 893 % max(t->physical_block_size, t->io_min); 894 895 /* Verify that new alignment_offset is on a logical block boundary */ 896 if (t->alignment_offset & (t->logical_block_size - 1)) { 897 t->flags |= BLK_FLAG_MISALIGNED; 898 ret = -1; 899 } 900 901 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size); 902 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size); 903 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size); 904 905 /* Discard alignment and granularity */ 906 if (b->discard_granularity) { 907 alignment = queue_limit_discard_alignment(b, start); 908 909 t->max_discard_sectors = min_not_zero(t->max_discard_sectors, 910 b->max_discard_sectors); 911 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, 912 b->max_hw_discard_sectors); 913 t->discard_granularity = max(t->discard_granularity, 914 b->discard_granularity); 915 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) % 916 t->discard_granularity; 917 } 918 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors, 919 b->max_secure_erase_sectors); 920 t->zone_write_granularity = max(t->zone_write_granularity, 921 b->zone_write_granularity); 922 if (!(t->features & BLK_FEAT_ZONED)) { 923 t->zone_write_granularity = 0; 924 t->max_zone_append_sectors = 0; 925 } 926 blk_stack_atomic_writes_limits(t, b, start); 927 928 return ret; 929 } 930 EXPORT_SYMBOL(blk_stack_limits); 931 932 /** 933 * queue_limits_stack_bdev - adjust queue_limits for stacked devices 934 * @t: the stacking driver limits (top device) 935 * @bdev: the underlying block device (bottom) 936 * @offset: offset to beginning of data within component device 937 * @pfx: prefix to use for warnings logged 938 * 939 * Description: 940 * This function is used by stacking drivers like MD and DM to ensure 941 * that all component devices have compatible block sizes and 942 * alignments. The stacking driver must provide a queue_limits 943 * struct (top) and then iteratively call the stacking function for 944 * all component (bottom) devices. The stacking function will 945 * attempt to combine the values and ensure proper alignment. 946 */ 947 void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev, 948 sector_t offset, const char *pfx) 949 { 950 if (blk_stack_limits(t, bdev_limits(bdev), 951 get_start_sect(bdev) + offset)) 952 pr_notice("%s: Warning: Device %pg is misaligned\n", 953 pfx, bdev); 954 } 955 EXPORT_SYMBOL_GPL(queue_limits_stack_bdev); 956 957 /** 958 * queue_limits_stack_integrity - stack integrity profile 959 * @t: target queue limits 960 * @b: base queue limits 961 * 962 * Check if the integrity profile in the @b can be stacked into the 963 * target @t. Stacking is possible if either: 964 * 965 * a) does not have any integrity information stacked into it yet 966 * b) the integrity profile in @b is identical to the one in @t 967 * 968 * If @b can be stacked into @t, return %true. Else return %false and clear the 969 * integrity information in @t. 970 */ 971 bool queue_limits_stack_integrity(struct queue_limits *t, 972 struct queue_limits *b) 973 { 974 struct blk_integrity *ti = &t->integrity; 975 struct blk_integrity *bi = &b->integrity; 976 977 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) 978 return true; 979 980 if (ti->flags & BLK_INTEGRITY_STACKED) { 981 if (ti->metadata_size != bi->metadata_size) 982 goto incompatible; 983 if (ti->interval_exp != bi->interval_exp) 984 goto incompatible; 985 if (ti->tag_size != bi->tag_size) 986 goto incompatible; 987 if (ti->csum_type != bi->csum_type) 988 goto incompatible; 989 if (ti->pi_tuple_size != bi->pi_tuple_size) 990 goto incompatible; 991 if ((ti->flags & BLK_INTEGRITY_REF_TAG) != 992 (bi->flags & BLK_INTEGRITY_REF_TAG)) 993 goto incompatible; 994 } else { 995 ti->flags = BLK_INTEGRITY_STACKED; 996 ti->flags |= (bi->flags & BLK_INTEGRITY_DEVICE_CAPABLE) | 997 (bi->flags & BLK_INTEGRITY_REF_TAG); 998 ti->csum_type = bi->csum_type; 999 ti->pi_tuple_size = bi->pi_tuple_size; 1000 ti->metadata_size = bi->metadata_size; 1001 ti->pi_offset = bi->pi_offset; 1002 ti->interval_exp = bi->interval_exp; 1003 ti->tag_size = bi->tag_size; 1004 } 1005 return true; 1006 1007 incompatible: 1008 memset(ti, 0, sizeof(*ti)); 1009 return false; 1010 } 1011 EXPORT_SYMBOL_GPL(queue_limits_stack_integrity); 1012 1013 /** 1014 * blk_set_queue_depth - tell the block layer about the device queue depth 1015 * @q: the request queue for the device 1016 * @depth: queue depth 1017 * 1018 */ 1019 void blk_set_queue_depth(struct request_queue *q, unsigned int depth) 1020 { 1021 q->queue_depth = depth; 1022 rq_qos_queue_depth_changed(q); 1023 } 1024 EXPORT_SYMBOL(blk_set_queue_depth); 1025 1026 int bdev_alignment_offset(struct block_device *bdev) 1027 { 1028 struct request_queue *q = bdev_get_queue(bdev); 1029 1030 if (q->limits.flags & BLK_FLAG_MISALIGNED) 1031 return -1; 1032 if (bdev_is_partition(bdev)) 1033 return queue_limit_alignment_offset(&q->limits, 1034 bdev->bd_start_sect); 1035 return q->limits.alignment_offset; 1036 } 1037 EXPORT_SYMBOL_GPL(bdev_alignment_offset); 1038 1039 unsigned int bdev_discard_alignment(struct block_device *bdev) 1040 { 1041 struct request_queue *q = bdev_get_queue(bdev); 1042 1043 if (bdev_is_partition(bdev)) 1044 return queue_limit_discard_alignment(&q->limits, 1045 bdev->bd_start_sect); 1046 return q->limits.discard_alignment; 1047 } 1048 EXPORT_SYMBOL_GPL(bdev_discard_alignment); 1049