blk-settings.c

   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
  24void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
  25{
  26	WRITE_ONCE(q->rq_timeout, timeout);
  27}
  28EXPORT_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 */
  37void 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}
  61EXPORT_SYMBOL(blk_set_stacking_limits);
  62
  63void 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
  88static 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 */
 132static 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
 139static 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_offset + bi->pi_tuple_size > bi->metadata_size) {
 165		pr_warn("pi_offset (%u) + pi_tuple_size (%u) exceeds metadata_size (%u)\n",
 166			bi->pi_offset, bi->pi_tuple_size, bi->metadata_size);
 167		return -EINVAL;
 168	}
 169
 170	switch (bi->csum_type) {
 171	case BLK_INTEGRITY_CSUM_NONE:
 172		if (bi->pi_tuple_size) {
 173			pr_warn("pi_tuple_size must be 0 when checksum type is none\n");
 174			return -EINVAL;
 175		}
 176		break;
 177	case BLK_INTEGRITY_CSUM_CRC:
 178	case BLK_INTEGRITY_CSUM_IP:
 179		if (bi->pi_tuple_size != sizeof(struct t10_pi_tuple)) {
 180			pr_warn("pi_tuple_size mismatch for T10 PI: expected %zu, got %u\n",
 181				 sizeof(struct t10_pi_tuple),
 182				 bi->pi_tuple_size);
 183			return -EINVAL;
 184		}
 185		break;
 186	case BLK_INTEGRITY_CSUM_CRC64:
 187		if (bi->pi_tuple_size != sizeof(struct crc64_pi_tuple)) {
 188			pr_warn("pi_tuple_size mismatch for CRC64 PI: expected %zu, got %u\n",
 189				 sizeof(struct crc64_pi_tuple),
 190				 bi->pi_tuple_size);
 191			return -EINVAL;
 192		}
 193		break;
 194	}
 195
 196	if (!bi->interval_exp) {
 197		bi->interval_exp = ilog2(lim->logical_block_size);
 198	} else if (bi->interval_exp < SECTOR_SHIFT ||
 199		   bi->interval_exp > ilog2(lim->logical_block_size)) {
 200		pr_warn("invalid interval_exp %u\n", bi->interval_exp);
 201		return -EINVAL;
 202	}
 203
 204	/*
 205	 * The PI generation / validation helpers do not expect intervals to
 206	 * straddle multiple bio_vecs.  Enforce alignment so that those are
 207	 * never generated, and that each buffer is aligned as expected.
 208	 */
 209	if (bi->csum_type) {
 210		lim->dma_alignment = max(lim->dma_alignment,
 211					(1U << bi->interval_exp) - 1);
 212	}
 213
 214	/*
 215	 * The block layer automatically adds integrity data for bios that don't
 216	 * already have it.  Limit the I/O size so that a single maximum size
 217	 * metadata segment can cover the integrity data for the entire I/O.
 218	 */
 219	lim->max_sectors = min(lim->max_sectors,
 220		max_integrity_io_size(lim) >> SECTOR_SHIFT);
 221
 222	return 0;
 223}
 224
 225/*
 226 * Returns max guaranteed bytes which we can fit in a bio.
 227 *
 228 * We request that an atomic_write is ITER_UBUF iov_iter (so a single vector),
 229 * so we assume that we can fit in at least PAGE_SIZE in a segment, apart from
 230 * the first and last segments.
 231 */
 232static unsigned int blk_queue_max_guaranteed_bio(struct queue_limits *lim)
 233{
 234	unsigned int max_segments = min(BIO_MAX_VECS, lim->max_segments);
 235	unsigned int length;
 236
 237	length = min(max_segments, 2) * lim->logical_block_size;
 238	if (max_segments > 2)
 239		length += (max_segments - 2) * PAGE_SIZE;
 240
 241	return length;
 242}
 243
 244static void blk_atomic_writes_update_limits(struct queue_limits *lim)
 245{
 246	unsigned int unit_limit = min(lim->max_hw_sectors << SECTOR_SHIFT,
 247					blk_queue_max_guaranteed_bio(lim));
 248
 249	unit_limit = rounddown_pow_of_two(unit_limit);
 250
 251	lim->atomic_write_max_sectors =
 252		min(lim->atomic_write_hw_max >> SECTOR_SHIFT,
 253			lim->max_hw_sectors);
 254	lim->atomic_write_unit_min =
 255		min(lim->atomic_write_hw_unit_min, unit_limit);
 256	lim->atomic_write_unit_max =
 257		min(lim->atomic_write_hw_unit_max, unit_limit);
 258	lim->atomic_write_boundary_sectors =
 259		lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
 260}
 261
 262/*
 263 * Test whether any boundary is aligned with any chunk size. Stacked
 264 * devices store any stripe size in t->chunk_sectors.
 265 */
 266static bool blk_valid_atomic_writes_boundary(unsigned int chunk_sectors,
 267					unsigned int boundary_sectors)
 268{
 269	if (!chunk_sectors || !boundary_sectors)
 270		return true;
 271
 272	if (boundary_sectors > chunk_sectors &&
 273	    boundary_sectors % chunk_sectors)
 274		return false;
 275
 276	if (chunk_sectors > boundary_sectors &&
 277	    chunk_sectors % boundary_sectors)
 278		return false;
 279
 280	return true;
 281}
 282
 283static void blk_validate_atomic_write_limits(struct queue_limits *lim)
 284{
 285	unsigned int boundary_sectors;
 286	unsigned int atomic_write_hw_max_sectors =
 287			lim->atomic_write_hw_max >> SECTOR_SHIFT;
 288
 289	if (!(lim->features & BLK_FEAT_ATOMIC_WRITES))
 290		goto unsupported;
 291
 292	/* UINT_MAX indicates stacked limits in initial state */
 293	if (lim->atomic_write_hw_max == UINT_MAX)
 294		goto unsupported;
 295
 296	if (!lim->atomic_write_hw_max)
 297		goto unsupported;
 298
 299	if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_min)))
 300		goto unsupported;
 301
 302	if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_max)))
 303		goto unsupported;
 304
 305	if (WARN_ON_ONCE(lim->atomic_write_hw_unit_min >
 306			 lim->atomic_write_hw_unit_max))
 307		goto unsupported;
 308
 309	if (WARN_ON_ONCE(lim->atomic_write_hw_unit_max >
 310			 lim->atomic_write_hw_max))
 311		goto unsupported;
 312
 313	if (WARN_ON_ONCE(lim->chunk_sectors &&
 314			atomic_write_hw_max_sectors > lim->chunk_sectors))
 315		goto unsupported;
 316
 317	boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
 318
 319	if (boundary_sectors) {
 320		if (WARN_ON_ONCE(lim->atomic_write_hw_max >
 321				 lim->atomic_write_hw_boundary))
 322			goto unsupported;
 323
 324		if (WARN_ON_ONCE(!blk_valid_atomic_writes_boundary(
 325			lim->chunk_sectors, boundary_sectors)))
 326			goto unsupported;
 327
 328		/*
 329		 * The boundary size just needs to be a multiple of unit_max
 330		 * (and not necessarily a power-of-2), so this following check
 331		 * could be relaxed in future.
 332		 * Furthermore, if needed, unit_max could even be reduced so
 333		 * that it is compliant with a !power-of-2 boundary.
 334		 */
 335		if (!is_power_of_2(boundary_sectors))
 336			goto unsupported;
 337	}
 338
 339	blk_atomic_writes_update_limits(lim);
 340	return;
 341
 342unsupported:
 343	lim->atomic_write_max_sectors = 0;
 344	lim->atomic_write_boundary_sectors = 0;
 345	lim->atomic_write_unit_min = 0;
 346	lim->atomic_write_unit_max = 0;
 347}
 348
 349/*
 350 * Check that the limits in lim are valid, initialize defaults for unset
 351 * values, and cap values based on others where needed.
 352 */
 353int blk_validate_limits(struct queue_limits *lim)
 354{
 355	unsigned int max_hw_sectors;
 356	unsigned int logical_block_sectors;
 357	unsigned long seg_size;
 358	int err;
 359
 360	/*
 361	 * Unless otherwise specified, default to 512 byte logical blocks and a
 362	 * physical block size equal to the logical block size.
 363	 */
 364	if (!lim->logical_block_size)
 365		lim->logical_block_size = SECTOR_SIZE;
 366	else if (blk_validate_block_size(lim->logical_block_size)) {
 367		pr_warn("Invalid logical block size (%d)\n", lim->logical_block_size);
 368		return -EINVAL;
 369	}
 370	if (lim->physical_block_size < lim->logical_block_size) {
 371		lim->physical_block_size = lim->logical_block_size;
 372	} else if (!is_power_of_2(lim->physical_block_size)) {
 373		pr_warn("Invalid physical block size (%d)\n", lim->physical_block_size);
 374		return -EINVAL;
 375	}
 376
 377	/*
 378	 * The minimum I/O size defaults to the physical block size unless
 379	 * explicitly overridden.
 380	 */
 381	if (lim->io_min < lim->physical_block_size)
 382		lim->io_min = lim->physical_block_size;
 383
 384	/*
 385	 * The optimal I/O size may not be aligned to physical block size
 386	 * (because it may be limited by dma engines which have no clue about
 387	 * block size of the disks attached to them), so we round it down here.
 388	 */
 389	lim->io_opt = round_down(lim->io_opt, lim->physical_block_size);
 390
 391	/*
 392	 * max_hw_sectors has a somewhat weird default for historical reason,
 393	 * but driver really should set their own instead of relying on this
 394	 * value.
 395	 *
 396	 * The block layer relies on the fact that every driver can
 397	 * handle at lest a page worth of data per I/O, and needs the value
 398	 * aligned to the logical block size.
 399	 */
 400	if (!lim->max_hw_sectors)
 401		lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
 402	if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
 403		return -EINVAL;
 404	logical_block_sectors = lim->logical_block_size >> SECTOR_SHIFT;
 405	if (WARN_ON_ONCE(logical_block_sectors > lim->max_hw_sectors))
 406		return -EINVAL;
 407	lim->max_hw_sectors = round_down(lim->max_hw_sectors,
 408			logical_block_sectors);
 409
 410	/*
 411	 * The actual max_sectors value is a complex beast and also takes the
 412	 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
 413	 * value into account.  The ->max_sectors value is always calculated
 414	 * from these, so directly setting it won't have any effect.
 415	 */
 416	max_hw_sectors = min_not_zero(lim->max_hw_sectors,
 417				lim->max_dev_sectors);
 418	if (lim->max_user_sectors) {
 419		if (lim->max_user_sectors < BLK_MIN_SEGMENT_SIZE / SECTOR_SIZE)
 420			return -EINVAL;
 421		lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
 422	} else if (lim->io_opt > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
 423		lim->max_sectors =
 424			min(max_hw_sectors, lim->io_opt >> SECTOR_SHIFT);
 425	} else if (lim->io_min > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
 426		lim->max_sectors =
 427			min(max_hw_sectors, lim->io_min >> SECTOR_SHIFT);
 428	} else {
 429		lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
 430	}
 431	lim->max_sectors = round_down(lim->max_sectors,
 432			logical_block_sectors);
 433
 434	/*
 435	 * Random default for the maximum number of segments.  Driver should not
 436	 * rely on this and set their own.
 437	 */
 438	if (!lim->max_segments)
 439		lim->max_segments = BLK_MAX_SEGMENTS;
 440
 441	if (lim->max_hw_wzeroes_unmap_sectors &&
 442	    lim->max_hw_wzeroes_unmap_sectors != lim->max_write_zeroes_sectors)
 443		return -EINVAL;
 444	lim->max_wzeroes_unmap_sectors = min(lim->max_hw_wzeroes_unmap_sectors,
 445			lim->max_user_wzeroes_unmap_sectors);
 446
 447	lim->max_discard_sectors =
 448		min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
 449
 450	/*
 451	 * When discard is not supported, discard_granularity should be reported
 452	 * as 0 to userspace.
 453	 */
 454	if (lim->max_discard_sectors)
 455		lim->discard_granularity =
 456			max(lim->discard_granularity, lim->physical_block_size);
 457	else
 458		lim->discard_granularity = 0;
 459
 460	if (!lim->max_discard_segments)
 461		lim->max_discard_segments = 1;
 462
 463	/*
 464	 * By default there is no limit on the segment boundary alignment,
 465	 * but if there is one it can't be smaller than the page size as
 466	 * that would break all the normal I/O patterns.
 467	 */
 468	if (!lim->seg_boundary_mask)
 469		lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
 470	if (WARN_ON_ONCE(lim->seg_boundary_mask < BLK_MIN_SEGMENT_SIZE - 1))
 471		return -EINVAL;
 472
 473	/*
 474	 * Stacking device may have both virtual boundary and max segment
 475	 * size limit, so allow this setting now, and long-term the two
 476	 * might need to move out of stacking limits since we have immutable
 477	 * bvec and lower layer bio splitting is supposed to handle the two
 478	 * correctly.
 479	 */
 480	if (lim->virt_boundary_mask) {
 481		if (!lim->max_segment_size)
 482			lim->max_segment_size = UINT_MAX;
 483	} else {
 484		/*
 485		 * The maximum segment size has an odd historic 64k default that
 486		 * drivers probably should override.  Just like the I/O size we
 487		 * require drivers to at least handle a full page per segment.
 488		 */
 489		if (!lim->max_segment_size)
 490			lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
 491		if (WARN_ON_ONCE(lim->max_segment_size < BLK_MIN_SEGMENT_SIZE))
 492			return -EINVAL;
 493	}
 494
 495	/* setup max segment size for building new segment in fast path */
 496	if (lim->seg_boundary_mask > lim->max_segment_size - 1)
 497		seg_size = lim->max_segment_size;
 498	else
 499		seg_size = lim->seg_boundary_mask + 1;
 500	lim->max_fast_segment_size = min_t(unsigned int, seg_size, PAGE_SIZE);
 501
 502	/*
 503	 * We require drivers to at least do logical block aligned I/O, but
 504	 * historically could not check for that due to the separate calls
 505	 * to set the limits.  Once the transition is finished the check
 506	 * below should be narrowed down to check the logical block size.
 507	 */
 508	if (!lim->dma_alignment)
 509		lim->dma_alignment = SECTOR_SIZE - 1;
 510	if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
 511		return -EINVAL;
 512
 513	if (lim->alignment_offset) {
 514		lim->alignment_offset &= (lim->physical_block_size - 1);
 515		lim->flags &= ~BLK_FLAG_MISALIGNED;
 516	}
 517
 518	if (!(lim->features & BLK_FEAT_WRITE_CACHE))
 519		lim->features &= ~BLK_FEAT_FUA;
 520
 521	blk_validate_atomic_write_limits(lim);
 522
 523	err = blk_validate_integrity_limits(lim);
 524	if (err)
 525		return err;
 526	return blk_validate_zoned_limits(lim);
 527}
 528EXPORT_SYMBOL_GPL(blk_validate_limits);
 529
 530/*
 531 * Set the default limits for a newly allocated queue.  @lim contains the
 532 * initial limits set by the driver, which could be no limit in which case
 533 * all fields are cleared to zero.
 534 */
 535int blk_set_default_limits(struct queue_limits *lim)
 536{
 537	/*
 538	 * Most defaults are set by capping the bounds in blk_validate_limits,
 539	 * but these limits are special and need an explicit initialization to
 540	 * the max value here.
 541	 */
 542	lim->max_user_discard_sectors = UINT_MAX;
 543	lim->max_user_wzeroes_unmap_sectors = UINT_MAX;
 544	return blk_validate_limits(lim);
 545}
 546
 547/**
 548 * queue_limits_commit_update - commit an atomic update of queue limits
 549 * @q:		queue to update
 550 * @lim:	limits to apply
 551 *
 552 * Apply the limits in @lim that were obtained from queue_limits_start_update()
 553 * and updated by the caller to @q.  The caller must have frozen the queue or
 554 * ensure that there are no outstanding I/Os by other means.
 555 *
 556 * Returns 0 if successful, else a negative error code.
 557 */
 558int queue_limits_commit_update(struct request_queue *q,
 559		struct queue_limits *lim)
 560{
 561	int error;
 562
 563	lockdep_assert_held(&q->limits_lock);
 564
 565	error = blk_validate_limits(lim);
 566	if (error)
 567		goto out_unlock;
 568
 569#ifdef CONFIG_BLK_INLINE_ENCRYPTION
 570	if (q->crypto_profile && lim->integrity.tag_size) {
 571		pr_warn("blk-integrity: Integrity and hardware inline encryption are not supported together.\n");
 572		error = -EINVAL;
 573		goto out_unlock;
 574	}
 575#endif
 576
 577	q->limits = *lim;
 578	if (q->disk)
 579		blk_apply_bdi_limits(q->disk->bdi, lim);
 580out_unlock:
 581	mutex_unlock(&q->limits_lock);
 582	return error;
 583}
 584EXPORT_SYMBOL_GPL(queue_limits_commit_update);
 585
 586/**
 587 * queue_limits_commit_update_frozen - commit an atomic update of queue limits
 588 * @q:		queue to update
 589 * @lim:	limits to apply
 590 *
 591 * Apply the limits in @lim that were obtained from queue_limits_start_update()
 592 * and updated with the new values by the caller to @q.  Freezes the queue
 593 * before the update and unfreezes it after.
 594 *
 595 * Returns 0 if successful, else a negative error code.
 596 */
 597int queue_limits_commit_update_frozen(struct request_queue *q,
 598		struct queue_limits *lim)
 599{
 600	unsigned int memflags;
 601	int ret;
 602
 603	memflags = blk_mq_freeze_queue(q);
 604	ret = queue_limits_commit_update(q, lim);
 605	blk_mq_unfreeze_queue(q, memflags);
 606
 607	return ret;
 608}
 609EXPORT_SYMBOL_GPL(queue_limits_commit_update_frozen);
 610
 611/**
 612 * queue_limits_set - apply queue limits to queue
 613 * @q:		queue to update
 614 * @lim:	limits to apply
 615 *
 616 * Apply the limits in @lim that were freshly initialized to @q.
 617 * To update existing limits use queue_limits_start_update() and
 618 * queue_limits_commit_update() instead.
 619 *
 620 * Returns 0 if successful, else a negative error code.
 621 */
 622int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
 623{
 624	mutex_lock(&q->limits_lock);
 625	return queue_limits_commit_update(q, lim);
 626}
 627EXPORT_SYMBOL_GPL(queue_limits_set);
 628
 629static int queue_limit_alignment_offset(const struct queue_limits *lim,
 630		sector_t sector)
 631{
 632	unsigned int granularity = max(lim->physical_block_size, lim->io_min);
 633	unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
 634		<< SECTOR_SHIFT;
 635
 636	return (granularity + lim->alignment_offset - alignment) % granularity;
 637}
 638
 639static unsigned int queue_limit_discard_alignment(
 640		const struct queue_limits *lim, sector_t sector)
 641{
 642	unsigned int alignment, granularity, offset;
 643
 644	if (!lim->max_discard_sectors)
 645		return 0;
 646
 647	/* Why are these in bytes, not sectors? */
 648	alignment = lim->discard_alignment >> SECTOR_SHIFT;
 649	granularity = lim->discard_granularity >> SECTOR_SHIFT;
 650
 651	/* Offset of the partition start in 'granularity' sectors */
 652	offset = sector_div(sector, granularity);
 653
 654	/* And why do we do this modulus *again* in blkdev_issue_discard()? */
 655	offset = (granularity + alignment - offset) % granularity;
 656
 657	/* Turn it back into bytes, gaah */
 658	return offset << SECTOR_SHIFT;
 659}
 660
 661static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
 662{
 663	sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
 664	if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
 665		sectors = PAGE_SIZE >> SECTOR_SHIFT;
 666	return sectors;
 667}
 668
 669/* Check if second and later bottom devices are compliant */
 670static bool blk_stack_atomic_writes_tail(struct queue_limits *t,
 671				struct queue_limits *b)
 672{
 673	/* We're not going to support different boundary sizes.. yet */
 674	if (t->atomic_write_hw_boundary != b->atomic_write_hw_boundary)
 675		return false;
 676
 677	/* Can't support this */
 678	if (t->atomic_write_hw_unit_min > b->atomic_write_hw_unit_max)
 679		return false;
 680
 681	/* Or this */
 682	if (t->atomic_write_hw_unit_max < b->atomic_write_hw_unit_min)
 683		return false;
 684
 685	t->atomic_write_hw_max = min(t->atomic_write_hw_max,
 686				b->atomic_write_hw_max);
 687	t->atomic_write_hw_unit_min = max(t->atomic_write_hw_unit_min,
 688				b->atomic_write_hw_unit_min);
 689	t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max,
 690				b->atomic_write_hw_unit_max);
 691	return true;
 692}
 693
 694static void blk_stack_atomic_writes_chunk_sectors(struct queue_limits *t)
 695{
 696	unsigned int chunk_bytes;
 697
 698	if (!t->chunk_sectors)
 699		return;
 700
 701	/*
 702	 * If chunk sectors is so large that its value in bytes overflows
 703	 * UINT_MAX, then just shift it down so it definitely will fit.
 704	 * We don't support atomic writes of such a large size anyway.
 705	 */
 706	if (check_shl_overflow(t->chunk_sectors, SECTOR_SHIFT, &chunk_bytes))
 707		chunk_bytes = t->chunk_sectors;
 708
 709	/*
 710	 * Find values for limits which work for chunk size.
 711	 * b->atomic_write_hw_unit_{min, max} may not be aligned with chunk
 712	 * size, as the chunk size is not restricted to a power-of-2.
 713	 * So we need to find highest power-of-2 which works for the chunk
 714	 * size.
 715	 * As an example scenario, we could have t->unit_max = 16K and
 716	 * t->chunk_sectors = 24KB. For this case, reduce t->unit_max to a
 717	 * value aligned with both limits, i.e. 8K in this example.
 718	 */
 719	t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max,
 720					max_pow_of_two_factor(chunk_bytes));
 721
 722	t->atomic_write_hw_unit_min = min(t->atomic_write_hw_unit_min,
 723					  t->atomic_write_hw_unit_max);
 724	t->atomic_write_hw_max = min(t->atomic_write_hw_max, chunk_bytes);
 725}
 726
 727/* Check stacking of first bottom device */
 728static bool blk_stack_atomic_writes_head(struct queue_limits *t,
 729				struct queue_limits *b)
 730{
 731	if (!blk_valid_atomic_writes_boundary(t->chunk_sectors,
 732			b->atomic_write_hw_boundary >> SECTOR_SHIFT))
 733		return false;
 734
 735	t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
 736	t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
 737	t->atomic_write_hw_max = b->atomic_write_hw_max;
 738	t->atomic_write_hw_boundary = b->atomic_write_hw_boundary;
 739	return true;
 740}
 741
 742static void blk_stack_atomic_writes_limits(struct queue_limits *t,
 743				struct queue_limits *b, sector_t start)
 744{
 745	if (!(b->features & BLK_FEAT_ATOMIC_WRITES))
 746		goto unsupported;
 747
 748	if (!b->atomic_write_hw_unit_min)
 749		goto unsupported;
 750
 751	if (!blk_atomic_write_start_sect_aligned(start, b))
 752		goto unsupported;
 753
 754	/* UINT_MAX indicates no stacking of bottom devices yet */
 755	if (t->atomic_write_hw_max == UINT_MAX) {
 756		if (!blk_stack_atomic_writes_head(t, b))
 757			goto unsupported;
 758	} else {
 759		if (!blk_stack_atomic_writes_tail(t, b))
 760			goto unsupported;
 761	}
 762	blk_stack_atomic_writes_chunk_sectors(t);
 763	return;
 764
 765unsupported:
 766	t->atomic_write_hw_max = 0;
 767	t->atomic_write_hw_unit_max = 0;
 768	t->atomic_write_hw_unit_min = 0;
 769	t->atomic_write_hw_boundary = 0;
 770}
 771
 772/**
 773 * blk_stack_limits - adjust queue_limits for stacked devices
 774 * @t:	the stacking driver limits (top device)
 775 * @b:  the underlying queue limits (bottom, component device)
 776 * @start:  first data sector within component device
 777 *
 778 * Description:
 779 *    This function is used by stacking drivers like MD and DM to ensure
 780 *    that all component devices have compatible block sizes and
 781 *    alignments.  The stacking driver must provide a queue_limits
 782 *    struct (top) and then iteratively call the stacking function for
 783 *    all component (bottom) devices.  The stacking function will
 784 *    attempt to combine the values and ensure proper alignment.
 785 *
 786 *    Returns 0 if the top and bottom queue_limits are compatible.  The
 787 *    top device's block sizes and alignment offsets may be adjusted to
 788 *    ensure alignment with the bottom device. If no compatible sizes
 789 *    and alignments exist, -1 is returned and the resulting top
 790 *    queue_limits will have the misaligned flag set to indicate that
 791 *    the alignment_offset is undefined.
 792 */
 793int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
 794		     sector_t start)
 795{
 796	unsigned int top, bottom, alignment;
 797	int ret = 0;
 798
 799	t->features |= (b->features & BLK_FEAT_INHERIT_MASK);
 800
 801	/*
 802	 * Some feaures need to be supported both by the stacking driver and all
 803	 * underlying devices.  The stacking driver sets these flags before
 804	 * stacking the limits, and this will clear the flags if any of the
 805	 * underlying devices does not support it.
 806	 */
 807	if (!(b->features & BLK_FEAT_NOWAIT))
 808		t->features &= ~BLK_FEAT_NOWAIT;
 809	if (!(b->features & BLK_FEAT_POLL))
 810		t->features &= ~BLK_FEAT_POLL;
 811
 812	t->flags |= (b->flags & BLK_FLAG_MISALIGNED);
 813
 814	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
 815	t->max_user_sectors = min_not_zero(t->max_user_sectors,
 816			b->max_user_sectors);
 817	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
 818	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
 819	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
 820					b->max_write_zeroes_sectors);
 821	t->max_user_wzeroes_unmap_sectors =
 822			min(t->max_user_wzeroes_unmap_sectors,
 823			    b->max_user_wzeroes_unmap_sectors);
 824	t->max_hw_wzeroes_unmap_sectors =
 825			min(t->max_hw_wzeroes_unmap_sectors,
 826			    b->max_hw_wzeroes_unmap_sectors);
 827
 828	t->max_hw_zone_append_sectors = min(t->max_hw_zone_append_sectors,
 829					b->max_hw_zone_append_sectors);
 830
 831	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
 832					    b->seg_boundary_mask);
 833	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
 834					    b->virt_boundary_mask);
 835
 836	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
 837	t->max_discard_segments = min_not_zero(t->max_discard_segments,
 838					       b->max_discard_segments);
 839	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
 840						 b->max_integrity_segments);
 841
 842	t->max_segment_size = min_not_zero(t->max_segment_size,
 843					   b->max_segment_size);
 844
 845	alignment = queue_limit_alignment_offset(b, start);
 846
 847	/* Bottom device has different alignment.  Check that it is
 848	 * compatible with the current top alignment.
 849	 */
 850	if (t->alignment_offset != alignment) {
 851
 852		top = max(t->physical_block_size, t->io_min)
 853			+ t->alignment_offset;
 854		bottom = max(b->physical_block_size, b->io_min) + alignment;
 855
 856		/* Verify that top and bottom intervals line up */
 857		if (max(top, bottom) % min(top, bottom)) {
 858			t->flags |= BLK_FLAG_MISALIGNED;
 859			ret = -1;
 860		}
 861	}
 862
 863	t->logical_block_size = max(t->logical_block_size,
 864				    b->logical_block_size);
 865
 866	t->physical_block_size = max(t->physical_block_size,
 867				     b->physical_block_size);
 868
 869	t->io_min = max(t->io_min, b->io_min);
 870	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
 871	t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
 872
 873	/* Set non-power-of-2 compatible chunk_sectors boundary */
 874	if (b->chunk_sectors)
 875		t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
 876
 877	/* Physical block size a multiple of the logical block size? */
 878	if (t->physical_block_size & (t->logical_block_size - 1)) {
 879		t->physical_block_size = t->logical_block_size;
 880		t->flags |= BLK_FLAG_MISALIGNED;
 881		ret = -1;
 882	}
 883
 884	/* Minimum I/O a multiple of the physical block size? */
 885	if (t->io_min & (t->physical_block_size - 1)) {
 886		t->io_min = t->physical_block_size;
 887		t->flags |= BLK_FLAG_MISALIGNED;
 888		ret = -1;
 889	}
 890
 891	/* Optimal I/O a multiple of the physical block size? */
 892	if (t->io_opt & (t->physical_block_size - 1)) {
 893		t->io_opt = 0;
 894		t->flags |= BLK_FLAG_MISALIGNED;
 895		ret = -1;
 896	}
 897
 898	/* chunk_sectors a multiple of the physical block size? */
 899	if (t->chunk_sectors % (t->physical_block_size >> SECTOR_SHIFT)) {
 900		t->chunk_sectors = 0;
 901		t->flags |= BLK_FLAG_MISALIGNED;
 902		ret = -1;
 903	}
 904
 905	/* Find lowest common alignment_offset */
 906	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
 907		% max(t->physical_block_size, t->io_min);
 908
 909	/* Verify that new alignment_offset is on a logical block boundary */
 910	if (t->alignment_offset & (t->logical_block_size - 1)) {
 911		t->flags |= BLK_FLAG_MISALIGNED;
 912		ret = -1;
 913	}
 914
 915	t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
 916	t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
 917	t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
 918
 919	/* Discard alignment and granularity */
 920	if (b->discard_granularity) {
 921		alignment = queue_limit_discard_alignment(b, start);
 922
 923		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
 924						      b->max_discard_sectors);
 925		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
 926							 b->max_hw_discard_sectors);
 927		t->discard_granularity = max(t->discard_granularity,
 928					     b->discard_granularity);
 929		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
 930			t->discard_granularity;
 931	}
 932	t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
 933						   b->max_secure_erase_sectors);
 934	t->zone_write_granularity = max(t->zone_write_granularity,
 935					b->zone_write_granularity);
 936	if (!(t->features & BLK_FEAT_ZONED)) {
 937		t->zone_write_granularity = 0;
 938		t->max_zone_append_sectors = 0;
 939	}
 940	blk_stack_atomic_writes_limits(t, b, start);
 941
 942	return ret;
 943}
 944EXPORT_SYMBOL(blk_stack_limits);
 945
 946/**
 947 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
 948 * @t:	the stacking driver limits (top device)
 949 * @bdev:  the underlying block device (bottom)
 950 * @offset:  offset to beginning of data within component device
 951 * @pfx: prefix to use for warnings logged
 952 *
 953 * Description:
 954 *    This function is used by stacking drivers like MD and DM to ensure
 955 *    that all component devices have compatible block sizes and
 956 *    alignments.  The stacking driver must provide a queue_limits
 957 *    struct (top) and then iteratively call the stacking function for
 958 *    all component (bottom) devices.  The stacking function will
 959 *    attempt to combine the values and ensure proper alignment.
 960 */
 961void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
 962		sector_t offset, const char *pfx)
 963{
 964	if (blk_stack_limits(t, bdev_limits(bdev),
 965			get_start_sect(bdev) + offset))
 966		pr_notice("%s: Warning: Device %pg is misaligned\n",
 967			pfx, bdev);
 968}
 969EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
 970
 971/**
 972 * queue_limits_stack_integrity - stack integrity profile
 973 * @t: target queue limits
 974 * @b: base queue limits
 975 *
 976 * Check if the integrity profile in the @b can be stacked into the
 977 * target @t.  Stacking is possible if either:
 978 *
 979 *   a) does not have any integrity information stacked into it yet
 980 *   b) the integrity profile in @b is identical to the one in @t
 981 *
 982 * If @b can be stacked into @t, return %true.  Else return %false and clear the
 983 * integrity information in @t.
 984 */
 985bool queue_limits_stack_integrity(struct queue_limits *t,
 986		struct queue_limits *b)
 987{
 988	struct blk_integrity *ti = &t->integrity;
 989	struct blk_integrity *bi = &b->integrity;
 990
 991	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
 992		return true;
 993
 994	if (ti->flags & BLK_INTEGRITY_STACKED) {
 995		if (ti->metadata_size != bi->metadata_size)
 996			goto incompatible;
 997		if (ti->interval_exp != bi->interval_exp)
 998			goto incompatible;
 999		if (ti->tag_size != bi->tag_size)
1000			goto incompatible;
1001		if (ti->csum_type != bi->csum_type)
1002			goto incompatible;
1003		if (ti->pi_tuple_size != bi->pi_tuple_size)
1004			goto incompatible;
1005		if ((ti->flags & BLK_INTEGRITY_REF_TAG) !=
1006		    (bi->flags & BLK_INTEGRITY_REF_TAG))
1007			goto incompatible;
1008	} else {
1009		ti->flags = BLK_INTEGRITY_STACKED;
1010		ti->flags |= (bi->flags & BLK_INTEGRITY_DEVICE_CAPABLE) |
1011			     (bi->flags & BLK_INTEGRITY_REF_TAG);
1012		ti->csum_type = bi->csum_type;
1013		ti->pi_tuple_size = bi->pi_tuple_size;
1014		ti->metadata_size = bi->metadata_size;
1015		ti->pi_offset = bi->pi_offset;
1016		ti->interval_exp = bi->interval_exp;
1017		ti->tag_size = bi->tag_size;
1018	}
1019	return true;
1020
1021incompatible:
1022	memset(ti, 0, sizeof(*ti));
1023	return false;
1024}
1025EXPORT_SYMBOL_GPL(queue_limits_stack_integrity);
1026
1027/**
1028 * blk_set_queue_depth - tell the block layer about the device queue depth
1029 * @q:		the request queue for the device
1030 * @depth:		queue depth
1031 *
1032 */
1033void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
1034{
1035	q->queue_depth = depth;
1036	rq_qos_queue_depth_changed(q);
1037}
1038EXPORT_SYMBOL(blk_set_queue_depth);
1039
1040int bdev_alignment_offset(struct block_device *bdev)
1041{
1042	struct request_queue *q = bdev_get_queue(bdev);
1043
1044	if (q->limits.flags & BLK_FLAG_MISALIGNED)
1045		return -1;
1046	if (bdev_is_partition(bdev))
1047		return queue_limit_alignment_offset(&q->limits,
1048				bdev->bd_start_sect);
1049	return q->limits.alignment_offset;
1050}
1051EXPORT_SYMBOL_GPL(bdev_alignment_offset);
1052
1053unsigned int bdev_discard_alignment(struct block_device *bdev)
1054{
1055	struct request_queue *q = bdev_get_queue(bdev);
1056
1057	if (bdev_is_partition(bdev))
1058		return queue_limit_discard_alignment(&q->limits,
1059				bdev->bd_start_sect);
1060	return q->limits.discard_alignment;
1061}
1062EXPORT_SYMBOL_GPL(bdev_discard_alignment);