Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * Copyright (C) 2016 Oracle.  All Rights Reserved.
   4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_mount.h"
  13#include "xfs_defer.h"
  14#include "xfs_btree.h"
  15#include "xfs_bmap.h"
  16#include "xfs_refcount_btree.h"
  17#include "xfs_alloc.h"
  18#include "xfs_errortag.h"
  19#include "xfs_error.h"
  20#include "xfs_trace.h"
  21#include "xfs_trans.h"
  22#include "xfs_bit.h"
  23#include "xfs_refcount.h"
  24#include "xfs_rmap.h"
  25#include "xfs_ag.h"
  26
  27struct kmem_cache	*xfs_refcount_intent_cache;
  28
  29/* Allowable refcount adjustment amounts. */
  30enum xfs_refc_adjust_op {
  31	XFS_REFCOUNT_ADJUST_INCREASE	= 1,
  32	XFS_REFCOUNT_ADJUST_DECREASE	= -1,
  33	XFS_REFCOUNT_ADJUST_COW_ALLOC	= 0,
  34	XFS_REFCOUNT_ADJUST_COW_FREE	= -1,
  35};
  36
  37STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
  38		xfs_agblock_t agbno, xfs_extlen_t aglen);
  39STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
  40		xfs_agblock_t agbno, xfs_extlen_t aglen);
  41
  42/*
  43 * Look up the first record less than or equal to [bno, len] in the btree
  44 * given by cur.
  45 */
  46int
  47xfs_refcount_lookup_le(
  48	struct xfs_btree_cur	*cur,
  49	enum xfs_refc_domain	domain,
  50	xfs_agblock_t		bno,
  51	int			*stat)
  52{
  53	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.pag->pag_agno,
  54			xfs_refcount_encode_startblock(bno, domain),
  55			XFS_LOOKUP_LE);
  56	cur->bc_rec.rc.rc_startblock = bno;
  57	cur->bc_rec.rc.rc_blockcount = 0;
  58	cur->bc_rec.rc.rc_domain = domain;
  59	return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
  60}
  61
  62/*
  63 * Look up the first record greater than or equal to [bno, len] in the btree
  64 * given by cur.
  65 */
  66int
  67xfs_refcount_lookup_ge(
  68	struct xfs_btree_cur	*cur,
  69	enum xfs_refc_domain	domain,
  70	xfs_agblock_t		bno,
  71	int			*stat)
  72{
  73	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.pag->pag_agno,
  74			xfs_refcount_encode_startblock(bno, domain),
  75			XFS_LOOKUP_GE);
  76	cur->bc_rec.rc.rc_startblock = bno;
  77	cur->bc_rec.rc.rc_blockcount = 0;
  78	cur->bc_rec.rc.rc_domain = domain;
  79	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
  80}
  81
  82/*
  83 * Look up the first record equal to [bno, len] in the btree
  84 * given by cur.
  85 */
  86int
  87xfs_refcount_lookup_eq(
  88	struct xfs_btree_cur	*cur,
  89	enum xfs_refc_domain	domain,
  90	xfs_agblock_t		bno,
  91	int			*stat)
  92{
  93	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.pag->pag_agno,
  94			xfs_refcount_encode_startblock(bno, domain),
  95			XFS_LOOKUP_LE);
  96	cur->bc_rec.rc.rc_startblock = bno;
  97	cur->bc_rec.rc.rc_blockcount = 0;
  98	cur->bc_rec.rc.rc_domain = domain;
  99	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 100}
 101
 102/* Convert on-disk record to in-core format. */
 103void
 104xfs_refcount_btrec_to_irec(
 105	const union xfs_btree_rec	*rec,
 106	struct xfs_refcount_irec	*irec)
 107{
 108	uint32_t			start;
 109
 110	start = be32_to_cpu(rec->refc.rc_startblock);
 111	if (start & XFS_REFC_COWFLAG) {
 112		start &= ~XFS_REFC_COWFLAG;
 113		irec->rc_domain = XFS_REFC_DOMAIN_COW;
 114	} else {
 115		irec->rc_domain = XFS_REFC_DOMAIN_SHARED;
 116	}
 117
 118	irec->rc_startblock = start;
 119	irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
 120	irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
 121}
 122
 123/* Simple checks for refcount records. */
 124xfs_failaddr_t
 125xfs_refcount_check_irec(
 126	struct xfs_perag		*pag,
 127	const struct xfs_refcount_irec	*irec)
 128{
 129	if (irec->rc_blockcount == 0 || irec->rc_blockcount > MAXREFCEXTLEN)
 130		return __this_address;
 131
 132	if (!xfs_refcount_check_domain(irec))
 133		return __this_address;
 134
 135	/* check for valid extent range, including overflow */
 136	if (!xfs_verify_agbext(pag, irec->rc_startblock, irec->rc_blockcount))
 137		return __this_address;
 138
 139	if (irec->rc_refcount == 0 || irec->rc_refcount > MAXREFCOUNT)
 140		return __this_address;
 141
 142	return NULL;
 143}
 144
 145static inline int
 146xfs_refcount_complain_bad_rec(
 147	struct xfs_btree_cur		*cur,
 148	xfs_failaddr_t			fa,
 149	const struct xfs_refcount_irec	*irec)
 150{
 151	struct xfs_mount		*mp = cur->bc_mp;
 152
 153	xfs_warn(mp,
 154 "Refcount BTree record corruption in AG %d detected at %pS!",
 155				cur->bc_ag.pag->pag_agno, fa);
 156	xfs_warn(mp,
 157		"Start block 0x%x, block count 0x%x, references 0x%x",
 158		irec->rc_startblock, irec->rc_blockcount, irec->rc_refcount);
 159	return -EFSCORRUPTED;
 160}
 161
 162/*
 163 * Get the data from the pointed-to record.
 164 */
 165int
 166xfs_refcount_get_rec(
 167	struct xfs_btree_cur		*cur,
 168	struct xfs_refcount_irec	*irec,
 169	int				*stat)
 170{
 171	union xfs_btree_rec		*rec;
 172	xfs_failaddr_t			fa;
 173	int				error;
 174
 175	error = xfs_btree_get_rec(cur, &rec, stat);
 176	if (error || !*stat)
 177		return error;
 178
 179	xfs_refcount_btrec_to_irec(rec, irec);
 180	fa = xfs_refcount_check_irec(cur->bc_ag.pag, irec);
 181	if (fa)
 182		return xfs_refcount_complain_bad_rec(cur, fa, irec);
 183
 184	trace_xfs_refcount_get(cur->bc_mp, cur->bc_ag.pag->pag_agno, irec);
 185	return 0;
 186}
 187
 188/*
 189 * Update the record referred to by cur to the value given
 190 * by [bno, len, refcount].
 191 * This either works (return 0) or gets an EFSCORRUPTED error.
 192 */
 193STATIC int
 194xfs_refcount_update(
 195	struct xfs_btree_cur		*cur,
 196	struct xfs_refcount_irec	*irec)
 197{
 198	union xfs_btree_rec	rec;
 199	uint32_t		start;
 200	int			error;
 201
 202	trace_xfs_refcount_update(cur->bc_mp, cur->bc_ag.pag->pag_agno, irec);
 203
 204	start = xfs_refcount_encode_startblock(irec->rc_startblock,
 205			irec->rc_domain);
 206	rec.refc.rc_startblock = cpu_to_be32(start);
 207	rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
 208	rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
 209
 210	error = xfs_btree_update(cur, &rec);
 211	if (error)
 212		trace_xfs_refcount_update_error(cur->bc_mp,
 213				cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 214	return error;
 215}
 216
 217/*
 218 * Insert the record referred to by cur to the value given
 219 * by [bno, len, refcount].
 220 * This either works (return 0) or gets an EFSCORRUPTED error.
 221 */
 222int
 223xfs_refcount_insert(
 224	struct xfs_btree_cur		*cur,
 225	struct xfs_refcount_irec	*irec,
 226	int				*i)
 227{
 228	int				error;
 229
 230	trace_xfs_refcount_insert(cur->bc_mp, cur->bc_ag.pag->pag_agno, irec);
 231
 232	cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
 233	cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
 234	cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
 235	cur->bc_rec.rc.rc_domain = irec->rc_domain;
 236
 237	error = xfs_btree_insert(cur, i);
 238	if (error)
 239		goto out_error;
 240	if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
 241		error = -EFSCORRUPTED;
 242		goto out_error;
 243	}
 244
 245out_error:
 246	if (error)
 247		trace_xfs_refcount_insert_error(cur->bc_mp,
 248				cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 249	return error;
 250}
 251
 252/*
 253 * Remove the record referred to by cur, then set the pointer to the spot
 254 * where the record could be re-inserted, in case we want to increment or
 255 * decrement the cursor.
 256 * This either works (return 0) or gets an EFSCORRUPTED error.
 257 */
 258STATIC int
 259xfs_refcount_delete(
 260	struct xfs_btree_cur	*cur,
 261	int			*i)
 262{
 263	struct xfs_refcount_irec	irec;
 264	int			found_rec;
 265	int			error;
 266
 267	error = xfs_refcount_get_rec(cur, &irec, &found_rec);
 268	if (error)
 269		goto out_error;
 270	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 271		error = -EFSCORRUPTED;
 272		goto out_error;
 273	}
 274	trace_xfs_refcount_delete(cur->bc_mp, cur->bc_ag.pag->pag_agno, &irec);
 275	error = xfs_btree_delete(cur, i);
 276	if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
 277		error = -EFSCORRUPTED;
 278		goto out_error;
 279	}
 280	if (error)
 281		goto out_error;
 282	error = xfs_refcount_lookup_ge(cur, irec.rc_domain, irec.rc_startblock,
 283			&found_rec);
 284out_error:
 285	if (error)
 286		trace_xfs_refcount_delete_error(cur->bc_mp,
 287				cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 288	return error;
 289}
 290
 291/*
 292 * Adjusting the Reference Count
 293 *
 294 * As stated elsewhere, the reference count btree (refcbt) stores
 295 * >1 reference counts for extents of physical blocks.  In this
 296 * operation, we're either raising or lowering the reference count of
 297 * some subrange stored in the tree:
 298 *
 299 *      <------ adjustment range ------>
 300 * ----+   +---+-----+ +--+--------+---------
 301 *  2  |   | 3 |  4  | |17|   55   |   10
 302 * ----+   +---+-----+ +--+--------+---------
 303 * X axis is physical blocks number;
 304 * reference counts are the numbers inside the rectangles
 305 *
 306 * The first thing we need to do is to ensure that there are no
 307 * refcount extents crossing either boundary of the range to be
 308 * adjusted.  For any extent that does cross a boundary, split it into
 309 * two extents so that we can increment the refcount of one of the
 310 * pieces later:
 311 *
 312 *      <------ adjustment range ------>
 313 * ----+   +---+-----+ +--+--------+----+----
 314 *  2  |   | 3 |  2  | |17|   55   | 10 | 10
 315 * ----+   +---+-----+ +--+--------+----+----
 316 *
 317 * For this next step, let's assume that all the physical blocks in
 318 * the adjustment range are mapped to a file and are therefore in use
 319 * at least once.  Therefore, we can infer that any gap in the
 320 * refcount tree within the adjustment range represents a physical
 321 * extent with refcount == 1:
 322 *
 323 *      <------ adjustment range ------>
 324 * ----+---+---+-----+-+--+--------+----+----
 325 *  2  |"1"| 3 |  2  |1|17|   55   | 10 | 10
 326 * ----+---+---+-----+-+--+--------+----+----
 327 *      ^
 328 *
 329 * For each extent that falls within the interval range, figure out
 330 * which extent is to the left or the right of that extent.  Now we
 331 * have a left, current, and right extent.  If the new reference count
 332 * of the center extent enables us to merge left, center, and right
 333 * into one record covering all three, do so.  If the center extent is
 334 * at the left end of the range, abuts the left extent, and its new
 335 * reference count matches the left extent's record, then merge them.
 336 * If the center extent is at the right end of the range, abuts the
 337 * right extent, and the reference counts match, merge those.  In the
 338 * example, we can left merge (assuming an increment operation):
 339 *
 340 *      <------ adjustment range ------>
 341 * --------+---+-----+-+--+--------+----+----
 342 *    2    | 3 |  2  |1|17|   55   | 10 | 10
 343 * --------+---+-----+-+--+--------+----+----
 344 *          ^
 345 *
 346 * For all other extents within the range, adjust the reference count
 347 * or delete it if the refcount falls below 2.  If we were
 348 * incrementing, the end result looks like this:
 349 *
 350 *      <------ adjustment range ------>
 351 * --------+---+-----+-+--+--------+----+----
 352 *    2    | 4 |  3  |2|18|   56   | 11 | 10
 353 * --------+---+-----+-+--+--------+----+----
 354 *
 355 * The result of a decrement operation looks as such:
 356 *
 357 *      <------ adjustment range ------>
 358 * ----+   +---+       +--+--------+----+----
 359 *  2  |   | 2 |       |16|   54   |  9 | 10
 360 * ----+   +---+       +--+--------+----+----
 361 *      DDDD    111111DD
 362 *
 363 * The blocks marked "D" are freed; the blocks marked "1" are only
 364 * referenced once and therefore the record is removed from the
 365 * refcount btree.
 366 */
 367
 368/* Next block after this extent. */
 369static inline xfs_agblock_t
 370xfs_refc_next(
 371	struct xfs_refcount_irec	*rc)
 372{
 373	return rc->rc_startblock + rc->rc_blockcount;
 374}
 375
 376/*
 377 * Split a refcount extent that crosses agbno.
 378 */
 379STATIC int
 380xfs_refcount_split_extent(
 381	struct xfs_btree_cur		*cur,
 382	enum xfs_refc_domain		domain,
 383	xfs_agblock_t			agbno,
 384	bool				*shape_changed)
 385{
 386	struct xfs_refcount_irec	rcext, tmp;
 387	int				found_rec;
 388	int				error;
 389
 390	*shape_changed = false;
 391	error = xfs_refcount_lookup_le(cur, domain, agbno, &found_rec);
 392	if (error)
 393		goto out_error;
 394	if (!found_rec)
 395		return 0;
 396
 397	error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
 398	if (error)
 399		goto out_error;
 400	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 401		error = -EFSCORRUPTED;
 402		goto out_error;
 403	}
 404	if (rcext.rc_domain != domain)
 405		return 0;
 406	if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
 407		return 0;
 408
 409	*shape_changed = true;
 410	trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_ag.pag->pag_agno,
 411			&rcext, agbno);
 412
 413	/* Establish the right extent. */
 414	tmp = rcext;
 415	tmp.rc_startblock = agbno;
 416	tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
 417	error = xfs_refcount_update(cur, &tmp);
 418	if (error)
 419		goto out_error;
 420
 421	/* Insert the left extent. */
 422	tmp = rcext;
 423	tmp.rc_blockcount = agbno - rcext.rc_startblock;
 424	error = xfs_refcount_insert(cur, &tmp, &found_rec);
 425	if (error)
 426		goto out_error;
 427	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 428		error = -EFSCORRUPTED;
 429		goto out_error;
 430	}
 431	return error;
 432
 433out_error:
 434	trace_xfs_refcount_split_extent_error(cur->bc_mp,
 435			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 436	return error;
 437}
 438
 439/*
 440 * Merge the left, center, and right extents.
 441 */
 442STATIC int
 443xfs_refcount_merge_center_extents(
 444	struct xfs_btree_cur		*cur,
 445	struct xfs_refcount_irec	*left,
 446	struct xfs_refcount_irec	*center,
 447	struct xfs_refcount_irec	*right,
 448	unsigned long long		extlen,
 449	xfs_extlen_t			*aglen)
 450{
 451	int				error;
 452	int				found_rec;
 453
 454	trace_xfs_refcount_merge_center_extents(cur->bc_mp,
 455			cur->bc_ag.pag->pag_agno, left, center, right);
 456
 457	ASSERT(left->rc_domain == center->rc_domain);
 458	ASSERT(right->rc_domain == center->rc_domain);
 459
 460	/*
 461	 * Make sure the center and right extents are not in the btree.
 462	 * If the center extent was synthesized, the first delete call
 463	 * removes the right extent and we skip the second deletion.
 464	 * If center and right were in the btree, then the first delete
 465	 * call removes the center and the second one removes the right
 466	 * extent.
 467	 */
 468	error = xfs_refcount_lookup_ge(cur, center->rc_domain,
 469			center->rc_startblock, &found_rec);
 470	if (error)
 471		goto out_error;
 472	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 473		error = -EFSCORRUPTED;
 474		goto out_error;
 475	}
 476
 477	error = xfs_refcount_delete(cur, &found_rec);
 478	if (error)
 479		goto out_error;
 480	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 481		error = -EFSCORRUPTED;
 482		goto out_error;
 483	}
 484
 485	if (center->rc_refcount > 1) {
 486		error = xfs_refcount_delete(cur, &found_rec);
 487		if (error)
 488			goto out_error;
 489		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 490			error = -EFSCORRUPTED;
 491			goto out_error;
 492		}
 493	}
 494
 495	/* Enlarge the left extent. */
 496	error = xfs_refcount_lookup_le(cur, left->rc_domain,
 497			left->rc_startblock, &found_rec);
 498	if (error)
 499		goto out_error;
 500	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 501		error = -EFSCORRUPTED;
 502		goto out_error;
 503	}
 504
 505	left->rc_blockcount = extlen;
 506	error = xfs_refcount_update(cur, left);
 507	if (error)
 508		goto out_error;
 509
 510	*aglen = 0;
 511	return error;
 512
 513out_error:
 514	trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
 515			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 516	return error;
 517}
 518
 519/*
 520 * Merge with the left extent.
 521 */
 522STATIC int
 523xfs_refcount_merge_left_extent(
 524	struct xfs_btree_cur		*cur,
 525	struct xfs_refcount_irec	*left,
 526	struct xfs_refcount_irec	*cleft,
 527	xfs_agblock_t			*agbno,
 528	xfs_extlen_t			*aglen)
 529{
 530	int				error;
 531	int				found_rec;
 532
 533	trace_xfs_refcount_merge_left_extent(cur->bc_mp,
 534			cur->bc_ag.pag->pag_agno, left, cleft);
 535
 536	ASSERT(left->rc_domain == cleft->rc_domain);
 537
 538	/* If the extent at agbno (cleft) wasn't synthesized, remove it. */
 539	if (cleft->rc_refcount > 1) {
 540		error = xfs_refcount_lookup_le(cur, cleft->rc_domain,
 541				cleft->rc_startblock, &found_rec);
 542		if (error)
 543			goto out_error;
 544		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 545			error = -EFSCORRUPTED;
 546			goto out_error;
 547		}
 548
 549		error = xfs_refcount_delete(cur, &found_rec);
 550		if (error)
 551			goto out_error;
 552		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 553			error = -EFSCORRUPTED;
 554			goto out_error;
 555		}
 556	}
 557
 558	/* Enlarge the left extent. */
 559	error = xfs_refcount_lookup_le(cur, left->rc_domain,
 560			left->rc_startblock, &found_rec);
 561	if (error)
 562		goto out_error;
 563	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 564		error = -EFSCORRUPTED;
 565		goto out_error;
 566	}
 567
 568	left->rc_blockcount += cleft->rc_blockcount;
 569	error = xfs_refcount_update(cur, left);
 570	if (error)
 571		goto out_error;
 572
 573	*agbno += cleft->rc_blockcount;
 574	*aglen -= cleft->rc_blockcount;
 575	return error;
 576
 577out_error:
 578	trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
 579			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 580	return error;
 581}
 582
 583/*
 584 * Merge with the right extent.
 585 */
 586STATIC int
 587xfs_refcount_merge_right_extent(
 588	struct xfs_btree_cur		*cur,
 589	struct xfs_refcount_irec	*right,
 590	struct xfs_refcount_irec	*cright,
 591	xfs_extlen_t			*aglen)
 592{
 593	int				error;
 594	int				found_rec;
 595
 596	trace_xfs_refcount_merge_right_extent(cur->bc_mp,
 597			cur->bc_ag.pag->pag_agno, cright, right);
 598
 599	ASSERT(right->rc_domain == cright->rc_domain);
 600
 601	/*
 602	 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
 603	 * remove it.
 604	 */
 605	if (cright->rc_refcount > 1) {
 606		error = xfs_refcount_lookup_le(cur, cright->rc_domain,
 607				cright->rc_startblock, &found_rec);
 608		if (error)
 609			goto out_error;
 610		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 611			error = -EFSCORRUPTED;
 612			goto out_error;
 613		}
 614
 615		error = xfs_refcount_delete(cur, &found_rec);
 616		if (error)
 617			goto out_error;
 618		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 619			error = -EFSCORRUPTED;
 620			goto out_error;
 621		}
 622	}
 623
 624	/* Enlarge the right extent. */
 625	error = xfs_refcount_lookup_le(cur, right->rc_domain,
 626			right->rc_startblock, &found_rec);
 627	if (error)
 628		goto out_error;
 629	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 630		error = -EFSCORRUPTED;
 631		goto out_error;
 632	}
 633
 634	right->rc_startblock -= cright->rc_blockcount;
 635	right->rc_blockcount += cright->rc_blockcount;
 636	error = xfs_refcount_update(cur, right);
 637	if (error)
 638		goto out_error;
 639
 640	*aglen -= cright->rc_blockcount;
 641	return error;
 642
 643out_error:
 644	trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
 645			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 646	return error;
 647}
 648
 649/*
 650 * Find the left extent and the one after it (cleft).  This function assumes
 651 * that we've already split any extent crossing agbno.
 652 */
 653STATIC int
 654xfs_refcount_find_left_extents(
 655	struct xfs_btree_cur		*cur,
 656	struct xfs_refcount_irec	*left,
 657	struct xfs_refcount_irec	*cleft,
 658	enum xfs_refc_domain		domain,
 659	xfs_agblock_t			agbno,
 660	xfs_extlen_t			aglen)
 661{
 662	struct xfs_refcount_irec	tmp;
 663	int				error;
 664	int				found_rec;
 665
 666	left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
 667	error = xfs_refcount_lookup_le(cur, domain, agbno - 1, &found_rec);
 668	if (error)
 669		goto out_error;
 670	if (!found_rec)
 671		return 0;
 672
 673	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 674	if (error)
 675		goto out_error;
 676	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 677		error = -EFSCORRUPTED;
 678		goto out_error;
 679	}
 680
 681	if (tmp.rc_domain != domain)
 682		return 0;
 683	if (xfs_refc_next(&tmp) != agbno)
 684		return 0;
 685	/* We have a left extent; retrieve (or invent) the next right one */
 686	*left = tmp;
 687
 688	error = xfs_btree_increment(cur, 0, &found_rec);
 689	if (error)
 690		goto out_error;
 691	if (found_rec) {
 692		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 693		if (error)
 694			goto out_error;
 695		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 696			error = -EFSCORRUPTED;
 697			goto out_error;
 698		}
 699
 700		if (tmp.rc_domain != domain)
 701			goto not_found;
 702
 703		/* if tmp starts at the end of our range, just use that */
 704		if (tmp.rc_startblock == agbno)
 705			*cleft = tmp;
 706		else {
 707			/*
 708			 * There's a gap in the refcntbt at the start of the
 709			 * range we're interested in (refcount == 1) so
 710			 * synthesize the implied extent and pass it back.
 711			 * We assume here that the agbno/aglen range was
 712			 * passed in from a data fork extent mapping and
 713			 * therefore is allocated to exactly one owner.
 714			 */
 715			cleft->rc_startblock = agbno;
 716			cleft->rc_blockcount = min(aglen,
 717					tmp.rc_startblock - agbno);
 718			cleft->rc_refcount = 1;
 719			cleft->rc_domain = domain;
 720		}
 721	} else {
 722not_found:
 723		/*
 724		 * No extents, so pretend that there's one covering the whole
 725		 * range.
 726		 */
 727		cleft->rc_startblock = agbno;
 728		cleft->rc_blockcount = aglen;
 729		cleft->rc_refcount = 1;
 730		cleft->rc_domain = domain;
 731	}
 732	trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_ag.pag->pag_agno,
 733			left, cleft, agbno);
 734	return error;
 735
 736out_error:
 737	trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
 738			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 739	return error;
 740}
 741
 742/*
 743 * Find the right extent and the one before it (cright).  This function
 744 * assumes that we've already split any extents crossing agbno + aglen.
 745 */
 746STATIC int
 747xfs_refcount_find_right_extents(
 748	struct xfs_btree_cur		*cur,
 749	struct xfs_refcount_irec	*right,
 750	struct xfs_refcount_irec	*cright,
 751	enum xfs_refc_domain		domain,
 752	xfs_agblock_t			agbno,
 753	xfs_extlen_t			aglen)
 754{
 755	struct xfs_refcount_irec	tmp;
 756	int				error;
 757	int				found_rec;
 758
 759	right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
 760	error = xfs_refcount_lookup_ge(cur, domain, agbno + aglen, &found_rec);
 761	if (error)
 762		goto out_error;
 763	if (!found_rec)
 764		return 0;
 765
 766	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 767	if (error)
 768		goto out_error;
 769	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 770		error = -EFSCORRUPTED;
 771		goto out_error;
 772	}
 773
 774	if (tmp.rc_domain != domain)
 775		return 0;
 776	if (tmp.rc_startblock != agbno + aglen)
 777		return 0;
 778	/* We have a right extent; retrieve (or invent) the next left one */
 779	*right = tmp;
 780
 781	error = xfs_btree_decrement(cur, 0, &found_rec);
 782	if (error)
 783		goto out_error;
 784	if (found_rec) {
 785		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 786		if (error)
 787			goto out_error;
 788		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
 789			error = -EFSCORRUPTED;
 790			goto out_error;
 791		}
 792
 793		if (tmp.rc_domain != domain)
 794			goto not_found;
 795
 796		/* if tmp ends at the end of our range, just use that */
 797		if (xfs_refc_next(&tmp) == agbno + aglen)
 798			*cright = tmp;
 799		else {
 800			/*
 801			 * There's a gap in the refcntbt at the end of the
 802			 * range we're interested in (refcount == 1) so
 803			 * create the implied extent and pass it back.
 804			 * We assume here that the agbno/aglen range was
 805			 * passed in from a data fork extent mapping and
 806			 * therefore is allocated to exactly one owner.
 807			 */
 808			cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
 809			cright->rc_blockcount = right->rc_startblock -
 810					cright->rc_startblock;
 811			cright->rc_refcount = 1;
 812			cright->rc_domain = domain;
 813		}
 814	} else {
 815not_found:
 816		/*
 817		 * No extents, so pretend that there's one covering the whole
 818		 * range.
 819		 */
 820		cright->rc_startblock = agbno;
 821		cright->rc_blockcount = aglen;
 822		cright->rc_refcount = 1;
 823		cright->rc_domain = domain;
 824	}
 825	trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_ag.pag->pag_agno,
 826			cright, right, agbno + aglen);
 827	return error;
 828
 829out_error:
 830	trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
 831			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
 832	return error;
 833}
 834
 835/* Is this extent valid? */
 836static inline bool
 837xfs_refc_valid(
 838	const struct xfs_refcount_irec	*rc)
 839{
 840	return rc->rc_startblock != NULLAGBLOCK;
 841}
 842
 843static inline xfs_nlink_t
 844xfs_refc_merge_refcount(
 845	const struct xfs_refcount_irec	*irec,
 846	enum xfs_refc_adjust_op		adjust)
 847{
 848	/* Once a record hits MAXREFCOUNT, it is pinned there forever */
 849	if (irec->rc_refcount == MAXREFCOUNT)
 850		return MAXREFCOUNT;
 851	return irec->rc_refcount + adjust;
 852}
 853
 854static inline bool
 855xfs_refc_want_merge_center(
 856	const struct xfs_refcount_irec	*left,
 857	const struct xfs_refcount_irec	*cleft,
 858	const struct xfs_refcount_irec	*cright,
 859	const struct xfs_refcount_irec	*right,
 860	bool				cleft_is_cright,
 861	enum xfs_refc_adjust_op		adjust,
 862	unsigned long long		*ulenp)
 863{
 864	unsigned long long		ulen = left->rc_blockcount;
 865	xfs_nlink_t			new_refcount;
 866
 867	/*
 868	 * To merge with a center record, both shoulder records must be
 869	 * adjacent to the record we want to adjust.  This is only true if
 870	 * find_left and find_right made all four records valid.
 871	 */
 872	if (!xfs_refc_valid(left)  || !xfs_refc_valid(right) ||
 873	    !xfs_refc_valid(cleft) || !xfs_refc_valid(cright))
 874		return false;
 875
 876	/* There must only be one record for the entire range. */
 877	if (!cleft_is_cright)
 878		return false;
 879
 880	/* The shoulder record refcounts must match the new refcount. */
 881	new_refcount = xfs_refc_merge_refcount(cleft, adjust);
 882	if (left->rc_refcount != new_refcount)
 883		return false;
 884	if (right->rc_refcount != new_refcount)
 885		return false;
 886
 887	/*
 888	 * The new record cannot exceed the max length.  ulen is a ULL as the
 889	 * individual record block counts can be up to (u32 - 1) in length
 890	 * hence we need to catch u32 addition overflows here.
 891	 */
 892	ulen += cleft->rc_blockcount + right->rc_blockcount;
 893	if (ulen >= MAXREFCEXTLEN)
 894		return false;
 895
 896	*ulenp = ulen;
 897	return true;
 898}
 899
 900static inline bool
 901xfs_refc_want_merge_left(
 902	const struct xfs_refcount_irec	*left,
 903	const struct xfs_refcount_irec	*cleft,
 904	enum xfs_refc_adjust_op		adjust)
 905{
 906	unsigned long long		ulen = left->rc_blockcount;
 907	xfs_nlink_t			new_refcount;
 908
 909	/*
 910	 * For a left merge, the left shoulder record must be adjacent to the
 911	 * start of the range.  If this is true, find_left made left and cleft
 912	 * contain valid contents.
 913	 */
 914	if (!xfs_refc_valid(left) || !xfs_refc_valid(cleft))
 915		return false;
 916
 917	/* Left shoulder record refcount must match the new refcount. */
 918	new_refcount = xfs_refc_merge_refcount(cleft, adjust);
 919	if (left->rc_refcount != new_refcount)
 920		return false;
 921
 922	/*
 923	 * The new record cannot exceed the max length.  ulen is a ULL as the
 924	 * individual record block counts can be up to (u32 - 1) in length
 925	 * hence we need to catch u32 addition overflows here.
 926	 */
 927	ulen += cleft->rc_blockcount;
 928	if (ulen >= MAXREFCEXTLEN)
 929		return false;
 930
 931	return true;
 932}
 933
 934static inline bool
 935xfs_refc_want_merge_right(
 936	const struct xfs_refcount_irec	*cright,
 937	const struct xfs_refcount_irec	*right,
 938	enum xfs_refc_adjust_op		adjust)
 939{
 940	unsigned long long		ulen = right->rc_blockcount;
 941	xfs_nlink_t			new_refcount;
 942
 943	/*
 944	 * For a right merge, the right shoulder record must be adjacent to the
 945	 * end of the range.  If this is true, find_right made cright and right
 946	 * contain valid contents.
 947	 */
 948	if (!xfs_refc_valid(right) || !xfs_refc_valid(cright))
 949		return false;
 950
 951	/* Right shoulder record refcount must match the new refcount. */
 952	new_refcount = xfs_refc_merge_refcount(cright, adjust);
 953	if (right->rc_refcount != new_refcount)
 954		return false;
 955
 956	/*
 957	 * The new record cannot exceed the max length.  ulen is a ULL as the
 958	 * individual record block counts can be up to (u32 - 1) in length
 959	 * hence we need to catch u32 addition overflows here.
 960	 */
 961	ulen += cright->rc_blockcount;
 962	if (ulen >= MAXREFCEXTLEN)
 963		return false;
 964
 965	return true;
 966}
 967
 968/*
 969 * Try to merge with any extents on the boundaries of the adjustment range.
 970 */
 971STATIC int
 972xfs_refcount_merge_extents(
 973	struct xfs_btree_cur	*cur,
 974	enum xfs_refc_domain	domain,
 975	xfs_agblock_t		*agbno,
 976	xfs_extlen_t		*aglen,
 977	enum xfs_refc_adjust_op adjust,
 978	bool			*shape_changed)
 979{
 980	struct xfs_refcount_irec	left = {0}, cleft = {0};
 981	struct xfs_refcount_irec	cright = {0}, right = {0};
 982	int				error;
 983	unsigned long long		ulen;
 984	bool				cequal;
 985
 986	*shape_changed = false;
 987	/*
 988	 * Find the extent just below agbno [left], just above agbno [cleft],
 989	 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
 990	 * [right].
 991	 */
 992	error = xfs_refcount_find_left_extents(cur, &left, &cleft, domain,
 993			*agbno, *aglen);
 994	if (error)
 995		return error;
 996	error = xfs_refcount_find_right_extents(cur, &right, &cright, domain,
 997			*agbno, *aglen);
 998	if (error)
 999		return error;
1000
1001	/* No left or right extent to merge; exit. */
1002	if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
1003		return 0;
1004
1005	cequal = (cleft.rc_startblock == cright.rc_startblock) &&
1006		 (cleft.rc_blockcount == cright.rc_blockcount);
1007
1008	/* Try to merge left, cleft, and right.  cleft must == cright. */
1009	if (xfs_refc_want_merge_center(&left, &cleft, &cright, &right, cequal,
1010				adjust, &ulen)) {
1011		*shape_changed = true;
1012		return xfs_refcount_merge_center_extents(cur, &left, &cleft,
1013				&right, ulen, aglen);
1014	}
1015
1016	/* Try to merge left and cleft. */
1017	if (xfs_refc_want_merge_left(&left, &cleft, adjust)) {
1018		*shape_changed = true;
1019		error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
1020				agbno, aglen);
1021		if (error)
1022			return error;
1023
1024		/*
1025		 * If we just merged left + cleft and cleft == cright,
1026		 * we no longer have a cright to merge with right.  We're done.
1027		 */
1028		if (cequal)
1029			return 0;
1030	}
1031
1032	/* Try to merge cright and right. */
1033	if (xfs_refc_want_merge_right(&cright, &right, adjust)) {
1034		*shape_changed = true;
1035		return xfs_refcount_merge_right_extent(cur, &right, &cright,
1036				aglen);
1037	}
1038
1039	return 0;
1040}
1041
1042/*
1043 * XXX: This is a pretty hand-wavy estimate.  The penalty for guessing
1044 * true incorrectly is a shutdown FS; the penalty for guessing false
1045 * incorrectly is more transaction rolls than might be necessary.
1046 * Be conservative here.
1047 */
1048static bool
1049xfs_refcount_still_have_space(
1050	struct xfs_btree_cur		*cur)
1051{
1052	unsigned long			overhead;
1053
1054	/*
1055	 * Worst case estimate: full splits of the free space and rmap btrees
1056	 * to handle each of the shape changes to the refcount btree.
1057	 */
1058	overhead = xfs_allocfree_block_count(cur->bc_mp,
1059				cur->bc_ag.refc.shape_changes);
1060	overhead += cur->bc_mp->m_refc_maxlevels;
1061	overhead *= cur->bc_mp->m_sb.sb_blocksize;
1062
1063	/*
1064	 * Only allow 2 refcount extent updates per transaction if the
1065	 * refcount continue update "error" has been injected.
1066	 */
1067	if (cur->bc_ag.refc.nr_ops > 2 &&
1068	    XFS_TEST_ERROR(false, cur->bc_mp,
1069			XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
1070		return false;
1071
1072	if (cur->bc_ag.refc.nr_ops == 0)
1073		return true;
1074	else if (overhead > cur->bc_tp->t_log_res)
1075		return false;
1076	return  cur->bc_tp->t_log_res - overhead >
1077		cur->bc_ag.refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
1078}
1079
1080/*
1081 * Adjust the refcounts of middle extents.  At this point we should have
1082 * split extents that crossed the adjustment range; merged with adjacent
1083 * extents; and updated agbno/aglen to reflect the merges.  Therefore,
1084 * all we have to do is update the extents inside [agbno, agbno + aglen].
1085 */
1086STATIC int
1087xfs_refcount_adjust_extents(
1088	struct xfs_btree_cur	*cur,
1089	xfs_agblock_t		*agbno,
1090	xfs_extlen_t		*aglen,
1091	enum xfs_refc_adjust_op	adj)
1092{
1093	struct xfs_refcount_irec	ext, tmp;
1094	int				error;
1095	int				found_rec, found_tmp;
1096	xfs_fsblock_t			fsbno;
1097
1098	/* Merging did all the work already. */
1099	if (*aglen == 0)
1100		return 0;
1101
1102	error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_SHARED, *agbno,
1103			&found_rec);
1104	if (error)
1105		goto out_error;
1106
1107	while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
1108		error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1109		if (error)
1110			goto out_error;
1111		if (!found_rec || ext.rc_domain != XFS_REFC_DOMAIN_SHARED) {
1112			ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
1113			ext.rc_blockcount = 0;
1114			ext.rc_refcount = 0;
1115			ext.rc_domain = XFS_REFC_DOMAIN_SHARED;
1116		}
1117
1118		/*
1119		 * Deal with a hole in the refcount tree; if a file maps to
1120		 * these blocks and there's no refcountbt record, pretend that
1121		 * there is one with refcount == 1.
1122		 */
1123		if (ext.rc_startblock != *agbno) {
1124			tmp.rc_startblock = *agbno;
1125			tmp.rc_blockcount = min(*aglen,
1126					ext.rc_startblock - *agbno);
1127			tmp.rc_refcount = 1 + adj;
1128			tmp.rc_domain = XFS_REFC_DOMAIN_SHARED;
1129
1130			trace_xfs_refcount_modify_extent(cur->bc_mp,
1131					cur->bc_ag.pag->pag_agno, &tmp);
1132
1133			/*
1134			 * Either cover the hole (increment) or
1135			 * delete the range (decrement).
1136			 */
1137			cur->bc_ag.refc.nr_ops++;
1138			if (tmp.rc_refcount) {
1139				error = xfs_refcount_insert(cur, &tmp,
1140						&found_tmp);
1141				if (error)
1142					goto out_error;
1143				if (XFS_IS_CORRUPT(cur->bc_mp,
1144						   found_tmp != 1)) {
1145					error = -EFSCORRUPTED;
1146					goto out_error;
1147				}
1148			} else {
1149				fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1150						cur->bc_ag.pag->pag_agno,
1151						tmp.rc_startblock);
1152				error = xfs_free_extent_later(cur->bc_tp, fsbno,
1153						  tmp.rc_blockcount, NULL,
1154						  XFS_AG_RESV_NONE, false);
1155				if (error)
1156					goto out_error;
1157			}
1158
1159			(*agbno) += tmp.rc_blockcount;
1160			(*aglen) -= tmp.rc_blockcount;
1161
1162			/* Stop if there's nothing left to modify */
1163			if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
1164				break;
1165
1166			/* Move the cursor to the start of ext. */
1167			error = xfs_refcount_lookup_ge(cur,
1168					XFS_REFC_DOMAIN_SHARED, *agbno,
1169					&found_rec);
1170			if (error)
1171				goto out_error;
1172		}
1173
1174		/*
1175		 * A previous step trimmed agbno/aglen such that the end of the
1176		 * range would not be in the middle of the record.  If this is
1177		 * no longer the case, something is seriously wrong with the
1178		 * btree.  Make sure we never feed the synthesized record into
1179		 * the processing loop below.
1180		 */
1181		if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount == 0) ||
1182		    XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount > *aglen)) {
1183			error = -EFSCORRUPTED;
1184			goto out_error;
1185		}
1186
1187		/*
1188		 * Adjust the reference count and either update the tree
1189		 * (incr) or free the blocks (decr).
1190		 */
1191		if (ext.rc_refcount == MAXREFCOUNT)
1192			goto skip;
1193		ext.rc_refcount += adj;
1194		trace_xfs_refcount_modify_extent(cur->bc_mp,
1195				cur->bc_ag.pag->pag_agno, &ext);
1196		cur->bc_ag.refc.nr_ops++;
1197		if (ext.rc_refcount > 1) {
1198			error = xfs_refcount_update(cur, &ext);
1199			if (error)
1200				goto out_error;
1201		} else if (ext.rc_refcount == 1) {
1202			error = xfs_refcount_delete(cur, &found_rec);
1203			if (error)
1204				goto out_error;
1205			if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1206				error = -EFSCORRUPTED;
1207				goto out_error;
1208			}
1209			goto advloop;
1210		} else {
1211			fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1212					cur->bc_ag.pag->pag_agno,
1213					ext.rc_startblock);
1214			error = xfs_free_extent_later(cur->bc_tp, fsbno,
1215					ext.rc_blockcount, NULL,
1216					XFS_AG_RESV_NONE, false);
1217			if (error)
1218				goto out_error;
1219		}
1220
1221skip:
1222		error = xfs_btree_increment(cur, 0, &found_rec);
1223		if (error)
1224			goto out_error;
1225
1226advloop:
1227		(*agbno) += ext.rc_blockcount;
1228		(*aglen) -= ext.rc_blockcount;
1229	}
1230
1231	return error;
1232out_error:
1233	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1234			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
1235	return error;
1236}
1237
1238/* Adjust the reference count of a range of AG blocks. */
1239STATIC int
1240xfs_refcount_adjust(
1241	struct xfs_btree_cur	*cur,
1242	xfs_agblock_t		*agbno,
1243	xfs_extlen_t		*aglen,
1244	enum xfs_refc_adjust_op	adj)
1245{
1246	bool			shape_changed;
1247	int			shape_changes = 0;
1248	int			error;
1249
1250	if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
1251		trace_xfs_refcount_increase(cur->bc_mp,
1252				cur->bc_ag.pag->pag_agno, *agbno, *aglen);
1253	else
1254		trace_xfs_refcount_decrease(cur->bc_mp,
1255				cur->bc_ag.pag->pag_agno, *agbno, *aglen);
1256
1257	/*
1258	 * Ensure that no rcextents cross the boundary of the adjustment range.
1259	 */
1260	error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1261			*agbno, &shape_changed);
1262	if (error)
1263		goto out_error;
1264	if (shape_changed)
1265		shape_changes++;
1266
1267	error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1268			*agbno + *aglen, &shape_changed);
1269	if (error)
1270		goto out_error;
1271	if (shape_changed)
1272		shape_changes++;
1273
1274	/*
1275	 * Try to merge with the left or right extents of the range.
1276	 */
1277	error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_SHARED,
1278			agbno, aglen, adj, &shape_changed);
1279	if (error)
1280		goto out_error;
1281	if (shape_changed)
1282		shape_changes++;
1283	if (shape_changes)
1284		cur->bc_ag.refc.shape_changes++;
1285
1286	/* Now that we've taken care of the ends, adjust the middle extents */
1287	error = xfs_refcount_adjust_extents(cur, agbno, aglen, adj);
1288	if (error)
1289		goto out_error;
1290
1291	return 0;
1292
1293out_error:
1294	trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1295			error, _RET_IP_);
1296	return error;
1297}
1298
1299/* Clean up after calling xfs_refcount_finish_one. */
1300void
1301xfs_refcount_finish_one_cleanup(
1302	struct xfs_trans	*tp,
1303	struct xfs_btree_cur	*rcur,
1304	int			error)
1305{
1306	struct xfs_buf		*agbp;
1307
1308	if (rcur == NULL)
1309		return;
1310	agbp = rcur->bc_ag.agbp;
1311	xfs_btree_del_cursor(rcur, error);
1312	if (error)
1313		xfs_trans_brelse(tp, agbp);
1314}
1315
1316/*
1317 * Set up a continuation a deferred refcount operation by updating the intent.
1318 * Checks to make sure we're not going to run off the end of the AG.
1319 */
1320static inline int
1321xfs_refcount_continue_op(
1322	struct xfs_btree_cur		*cur,
1323	struct xfs_refcount_intent	*ri,
1324	xfs_agblock_t			new_agbno)
1325{
1326	struct xfs_mount		*mp = cur->bc_mp;
1327	struct xfs_perag		*pag = cur->bc_ag.pag;
1328
1329	if (XFS_IS_CORRUPT(mp, !xfs_verify_agbext(pag, new_agbno,
1330					ri->ri_blockcount)))
1331		return -EFSCORRUPTED;
1332
1333	ri->ri_startblock = XFS_AGB_TO_FSB(mp, pag->pag_agno, new_agbno);
1334
1335	ASSERT(xfs_verify_fsbext(mp, ri->ri_startblock, ri->ri_blockcount));
1336	ASSERT(pag->pag_agno == XFS_FSB_TO_AGNO(mp, ri->ri_startblock));
1337
1338	return 0;
1339}
1340
1341/*
1342 * Process one of the deferred refcount operations.  We pass back the
1343 * btree cursor to maintain our lock on the btree between calls.
1344 * This saves time and eliminates a buffer deadlock between the
1345 * superblock and the AGF because we'll always grab them in the same
1346 * order.
1347 */
1348int
1349xfs_refcount_finish_one(
1350	struct xfs_trans		*tp,
1351	struct xfs_refcount_intent	*ri,
1352	struct xfs_btree_cur		**pcur)
1353{
1354	struct xfs_mount		*mp = tp->t_mountp;
1355	struct xfs_btree_cur		*rcur;
1356	struct xfs_buf			*agbp = NULL;
1357	int				error = 0;
1358	xfs_agblock_t			bno;
1359	unsigned long			nr_ops = 0;
1360	int				shape_changes = 0;
1361
1362	bno = XFS_FSB_TO_AGBNO(mp, ri->ri_startblock);
1363
1364	trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, ri->ri_startblock),
1365			ri->ri_type, XFS_FSB_TO_AGBNO(mp, ri->ri_startblock),
1366			ri->ri_blockcount);
1367
1368	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1369		return -EIO;
1370
1371	/*
1372	 * If we haven't gotten a cursor or the cursor AG doesn't match
1373	 * the startblock, get one now.
1374	 */
1375	rcur = *pcur;
1376	if (rcur != NULL && rcur->bc_ag.pag != ri->ri_pag) {
1377		nr_ops = rcur->bc_ag.refc.nr_ops;
1378		shape_changes = rcur->bc_ag.refc.shape_changes;
1379		xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1380		rcur = NULL;
1381		*pcur = NULL;
1382	}
1383	if (rcur == NULL) {
1384		error = xfs_alloc_read_agf(ri->ri_pag, tp,
1385				XFS_ALLOC_FLAG_FREEING, &agbp);
1386		if (error)
1387			return error;
1388
1389		rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, ri->ri_pag);
1390		rcur->bc_ag.refc.nr_ops = nr_ops;
1391		rcur->bc_ag.refc.shape_changes = shape_changes;
1392	}
1393	*pcur = rcur;
1394
1395	switch (ri->ri_type) {
1396	case XFS_REFCOUNT_INCREASE:
1397		error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1398				XFS_REFCOUNT_ADJUST_INCREASE);
1399		if (error)
1400			return error;
1401		if (ri->ri_blockcount > 0)
1402			error = xfs_refcount_continue_op(rcur, ri, bno);
1403		break;
1404	case XFS_REFCOUNT_DECREASE:
1405		error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1406				XFS_REFCOUNT_ADJUST_DECREASE);
1407		if (error)
1408			return error;
1409		if (ri->ri_blockcount > 0)
1410			error = xfs_refcount_continue_op(rcur, ri, bno);
1411		break;
1412	case XFS_REFCOUNT_ALLOC_COW:
1413		error = __xfs_refcount_cow_alloc(rcur, bno, ri->ri_blockcount);
1414		if (error)
1415			return error;
1416		ri->ri_blockcount = 0;
1417		break;
1418	case XFS_REFCOUNT_FREE_COW:
1419		error = __xfs_refcount_cow_free(rcur, bno, ri->ri_blockcount);
1420		if (error)
1421			return error;
1422		ri->ri_blockcount = 0;
1423		break;
1424	default:
1425		ASSERT(0);
1426		return -EFSCORRUPTED;
1427	}
1428	if (!error && ri->ri_blockcount > 0)
1429		trace_xfs_refcount_finish_one_leftover(mp, ri->ri_pag->pag_agno,
1430				ri->ri_type, bno, ri->ri_blockcount);
1431	return error;
1432}
1433
1434/*
1435 * Record a refcount intent for later processing.
1436 */
1437static void
1438__xfs_refcount_add(
1439	struct xfs_trans		*tp,
1440	enum xfs_refcount_intent_type	type,
1441	xfs_fsblock_t			startblock,
1442	xfs_extlen_t			blockcount)
1443{
1444	struct xfs_refcount_intent	*ri;
1445
1446	trace_xfs_refcount_defer(tp->t_mountp,
1447			XFS_FSB_TO_AGNO(tp->t_mountp, startblock),
1448			type, XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
1449			blockcount);
1450
1451	ri = kmem_cache_alloc(xfs_refcount_intent_cache,
1452			GFP_NOFS | __GFP_NOFAIL);
1453	INIT_LIST_HEAD(&ri->ri_list);
1454	ri->ri_type = type;
1455	ri->ri_startblock = startblock;
1456	ri->ri_blockcount = blockcount;
1457
1458	xfs_refcount_update_get_group(tp->t_mountp, ri);
1459	xfs_defer_add(tp, &ri->ri_list, &xfs_refcount_update_defer_type);
1460}
1461
1462/*
1463 * Increase the reference count of the blocks backing a file's extent.
1464 */
1465void
1466xfs_refcount_increase_extent(
1467	struct xfs_trans		*tp,
1468	struct xfs_bmbt_irec		*PREV)
1469{
1470	if (!xfs_has_reflink(tp->t_mountp))
1471		return;
1472
1473	__xfs_refcount_add(tp, XFS_REFCOUNT_INCREASE, PREV->br_startblock,
1474			PREV->br_blockcount);
1475}
1476
1477/*
1478 * Decrease the reference count of the blocks backing a file's extent.
1479 */
1480void
1481xfs_refcount_decrease_extent(
1482	struct xfs_trans		*tp,
1483	struct xfs_bmbt_irec		*PREV)
1484{
1485	if (!xfs_has_reflink(tp->t_mountp))
1486		return;
1487
1488	__xfs_refcount_add(tp, XFS_REFCOUNT_DECREASE, PREV->br_startblock,
1489			PREV->br_blockcount);
1490}
1491
1492/*
1493 * Given an AG extent, find the lowest-numbered run of shared blocks
1494 * within that range and return the range in fbno/flen.  If
1495 * find_end_of_shared is set, return the longest contiguous extent of
1496 * shared blocks; if not, just return the first extent we find.  If no
1497 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1498 * and 0, respectively.
1499 */
1500int
1501xfs_refcount_find_shared(
1502	struct xfs_btree_cur		*cur,
1503	xfs_agblock_t			agbno,
1504	xfs_extlen_t			aglen,
1505	xfs_agblock_t			*fbno,
1506	xfs_extlen_t			*flen,
1507	bool				find_end_of_shared)
1508{
1509	struct xfs_refcount_irec	tmp;
1510	int				i;
1511	int				have;
1512	int				error;
1513
1514	trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1515			agbno, aglen);
1516
1517	/* By default, skip the whole range */
1518	*fbno = NULLAGBLOCK;
1519	*flen = 0;
1520
1521	/* Try to find a refcount extent that crosses the start */
1522	error = xfs_refcount_lookup_le(cur, XFS_REFC_DOMAIN_SHARED, agbno,
1523			&have);
1524	if (error)
1525		goto out_error;
1526	if (!have) {
1527		/* No left extent, look at the next one */
1528		error = xfs_btree_increment(cur, 0, &have);
1529		if (error)
1530			goto out_error;
1531		if (!have)
1532			goto done;
1533	}
1534	error = xfs_refcount_get_rec(cur, &tmp, &i);
1535	if (error)
1536		goto out_error;
1537	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1538		error = -EFSCORRUPTED;
1539		goto out_error;
1540	}
1541	if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1542		goto done;
1543
1544	/* If the extent ends before the start, look at the next one */
1545	if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1546		error = xfs_btree_increment(cur, 0, &have);
1547		if (error)
1548			goto out_error;
1549		if (!have)
1550			goto done;
1551		error = xfs_refcount_get_rec(cur, &tmp, &i);
1552		if (error)
1553			goto out_error;
1554		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1555			error = -EFSCORRUPTED;
1556			goto out_error;
1557		}
1558		if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1559			goto done;
1560	}
1561
1562	/* If the extent starts after the range we want, bail out */
1563	if (tmp.rc_startblock >= agbno + aglen)
1564		goto done;
1565
1566	/* We found the start of a shared extent! */
1567	if (tmp.rc_startblock < agbno) {
1568		tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1569		tmp.rc_startblock = agbno;
1570	}
1571
1572	*fbno = tmp.rc_startblock;
1573	*flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1574	if (!find_end_of_shared)
1575		goto done;
1576
1577	/* Otherwise, find the end of this shared extent */
1578	while (*fbno + *flen < agbno + aglen) {
1579		error = xfs_btree_increment(cur, 0, &have);
1580		if (error)
1581			goto out_error;
1582		if (!have)
1583			break;
1584		error = xfs_refcount_get_rec(cur, &tmp, &i);
1585		if (error)
1586			goto out_error;
1587		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1588			error = -EFSCORRUPTED;
1589			goto out_error;
1590		}
1591		if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED ||
1592		    tmp.rc_startblock >= agbno + aglen ||
1593		    tmp.rc_startblock != *fbno + *flen)
1594			break;
1595		*flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1596	}
1597
1598done:
1599	trace_xfs_refcount_find_shared_result(cur->bc_mp,
1600			cur->bc_ag.pag->pag_agno, *fbno, *flen);
1601
1602out_error:
1603	if (error)
1604		trace_xfs_refcount_find_shared_error(cur->bc_mp,
1605				cur->bc_ag.pag->pag_agno, error, _RET_IP_);
1606	return error;
1607}
1608
1609/*
1610 * Recovering CoW Blocks After a Crash
1611 *
1612 * Due to the way that the copy on write mechanism works, there's a window of
1613 * opportunity in which we can lose track of allocated blocks during a crash.
1614 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1615 * causes blocks to be allocated and stored in the CoW fork.  The blocks are
1616 * no longer in the free space btree but are not otherwise recorded anywhere
1617 * until the write completes and the blocks are mapped into the file.  A crash
1618 * in between allocation and remapping results in the replacement blocks being
1619 * lost.  This situation is exacerbated by the CoW extent size hint because
1620 * allocations can hang around for long time.
1621 *
1622 * However, there is a place where we can record these allocations before they
1623 * become mappings -- the reference count btree.  The btree does not record
1624 * extents with refcount == 1, so we can record allocations with a refcount of
1625 * 1.  Blocks being used for CoW writeout cannot be shared, so there should be
1626 * no conflict with shared block records.  These mappings should be created
1627 * when we allocate blocks to the CoW fork and deleted when they're removed
1628 * from the CoW fork.
1629 *
1630 * Minor nit: records for in-progress CoW allocations and records for shared
1631 * extents must never be merged, to preserve the property that (except for CoW
1632 * allocations) there are no refcount btree entries with refcount == 1.  The
1633 * only time this could potentially happen is when unsharing a block that's
1634 * adjacent to CoW allocations, so we must be careful to avoid this.
1635 *
1636 * At mount time we recover lost CoW allocations by searching the refcount
1637 * btree for these refcount == 1 mappings.  These represent CoW allocations
1638 * that were in progress at the time the filesystem went down, so we can free
1639 * them to get the space back.
1640 *
1641 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1642 * several reasons -- first, EFIs pin the tail of the log and would have to be
1643 * periodically relogged to avoid filling up the log.  Second, CoW completions
1644 * will have to file an EFD and create new EFIs for whatever remains in the
1645 * CoW fork; this partially takes care of (1) but extent-size reservations
1646 * will have to periodically relog even if there's no writeout in progress.
1647 * This can happen if the CoW extent size hint is set, which you really want.
1648 * Third, EFIs cannot currently be automatically relogged into newer
1649 * transactions to advance the log tail.  Fourth, stuffing the log full of
1650 * EFIs places an upper bound on the number of CoW allocations that can be
1651 * held filesystem-wide at any given time.  Recording them in the refcount
1652 * btree doesn't require us to maintain any state in memory and doesn't pin
1653 * the log.
1654 */
1655/*
1656 * Adjust the refcounts of CoW allocations.  These allocations are "magic"
1657 * in that they're not referenced anywhere else in the filesystem, so we
1658 * stash them in the refcount btree with a refcount of 1 until either file
1659 * remapping (or CoW cancellation) happens.
1660 */
1661STATIC int
1662xfs_refcount_adjust_cow_extents(
1663	struct xfs_btree_cur	*cur,
1664	xfs_agblock_t		agbno,
1665	xfs_extlen_t		aglen,
1666	enum xfs_refc_adjust_op	adj)
1667{
1668	struct xfs_refcount_irec	ext, tmp;
1669	int				error;
1670	int				found_rec, found_tmp;
1671
1672	if (aglen == 0)
1673		return 0;
1674
1675	/* Find any overlapping refcount records */
1676	error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_COW, agbno,
1677			&found_rec);
1678	if (error)
1679		goto out_error;
1680	error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1681	if (error)
1682		goto out_error;
1683	if (XFS_IS_CORRUPT(cur->bc_mp, found_rec &&
1684				ext.rc_domain != XFS_REFC_DOMAIN_COW)) {
1685		error = -EFSCORRUPTED;
1686		goto out_error;
1687	}
1688	if (!found_rec) {
1689		ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
1690		ext.rc_blockcount = 0;
1691		ext.rc_refcount = 0;
1692		ext.rc_domain = XFS_REFC_DOMAIN_COW;
1693	}
1694
1695	switch (adj) {
1696	case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1697		/* Adding a CoW reservation, there should be nothing here. */
1698		if (XFS_IS_CORRUPT(cur->bc_mp,
1699				   agbno + aglen > ext.rc_startblock)) {
1700			error = -EFSCORRUPTED;
1701			goto out_error;
1702		}
1703
1704		tmp.rc_startblock = agbno;
1705		tmp.rc_blockcount = aglen;
1706		tmp.rc_refcount = 1;
1707		tmp.rc_domain = XFS_REFC_DOMAIN_COW;
1708
1709		trace_xfs_refcount_modify_extent(cur->bc_mp,
1710				cur->bc_ag.pag->pag_agno, &tmp);
1711
1712		error = xfs_refcount_insert(cur, &tmp,
1713				&found_tmp);
1714		if (error)
1715			goto out_error;
1716		if (XFS_IS_CORRUPT(cur->bc_mp, found_tmp != 1)) {
1717			error = -EFSCORRUPTED;
1718			goto out_error;
1719		}
1720		break;
1721	case XFS_REFCOUNT_ADJUST_COW_FREE:
1722		/* Removing a CoW reservation, there should be one extent. */
1723		if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_startblock != agbno)) {
1724			error = -EFSCORRUPTED;
1725			goto out_error;
1726		}
1727		if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount != aglen)) {
1728			error = -EFSCORRUPTED;
1729			goto out_error;
1730		}
1731		if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_refcount != 1)) {
1732			error = -EFSCORRUPTED;
1733			goto out_error;
1734		}
1735
1736		ext.rc_refcount = 0;
1737		trace_xfs_refcount_modify_extent(cur->bc_mp,
1738				cur->bc_ag.pag->pag_agno, &ext);
1739		error = xfs_refcount_delete(cur, &found_rec);
1740		if (error)
1741			goto out_error;
1742		if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1743			error = -EFSCORRUPTED;
1744			goto out_error;
1745		}
1746		break;
1747	default:
1748		ASSERT(0);
1749	}
1750
1751	return error;
1752out_error:
1753	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1754			cur->bc_ag.pag->pag_agno, error, _RET_IP_);
1755	return error;
1756}
1757
1758/*
1759 * Add or remove refcount btree entries for CoW reservations.
1760 */
1761STATIC int
1762xfs_refcount_adjust_cow(
1763	struct xfs_btree_cur	*cur,
1764	xfs_agblock_t		agbno,
1765	xfs_extlen_t		aglen,
1766	enum xfs_refc_adjust_op	adj)
1767{
1768	bool			shape_changed;
1769	int			error;
1770
1771	/*
1772	 * Ensure that no rcextents cross the boundary of the adjustment range.
1773	 */
1774	error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1775			agbno, &shape_changed);
1776	if (error)
1777		goto out_error;
1778
1779	error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1780			agbno + aglen, &shape_changed);
1781	if (error)
1782		goto out_error;
1783
1784	/*
1785	 * Try to merge with the left or right extents of the range.
1786	 */
1787	error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_COW, &agbno,
1788			&aglen, adj, &shape_changed);
1789	if (error)
1790		goto out_error;
1791
1792	/* Now that we've taken care of the ends, adjust the middle extents */
1793	error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1794	if (error)
1795		goto out_error;
1796
1797	return 0;
1798
1799out_error:
1800	trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1801			error, _RET_IP_);
1802	return error;
1803}
1804
1805/*
1806 * Record a CoW allocation in the refcount btree.
1807 */
1808STATIC int
1809__xfs_refcount_cow_alloc(
1810	struct xfs_btree_cur	*rcur,
1811	xfs_agblock_t		agbno,
1812	xfs_extlen_t		aglen)
1813{
1814	trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_ag.pag->pag_agno,
1815			agbno, aglen);
1816
1817	/* Add refcount btree reservation */
1818	return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1819			XFS_REFCOUNT_ADJUST_COW_ALLOC);
1820}
1821
1822/*
1823 * Remove a CoW allocation from the refcount btree.
1824 */
1825STATIC int
1826__xfs_refcount_cow_free(
1827	struct xfs_btree_cur	*rcur,
1828	xfs_agblock_t		agbno,
1829	xfs_extlen_t		aglen)
1830{
1831	trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_ag.pag->pag_agno,
1832			agbno, aglen);
1833
1834	/* Remove refcount btree reservation */
1835	return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1836			XFS_REFCOUNT_ADJUST_COW_FREE);
1837}
1838
1839/* Record a CoW staging extent in the refcount btree. */
1840void
1841xfs_refcount_alloc_cow_extent(
1842	struct xfs_trans		*tp,
1843	xfs_fsblock_t			fsb,
1844	xfs_extlen_t			len)
1845{
1846	struct xfs_mount		*mp = tp->t_mountp;
1847
1848	if (!xfs_has_reflink(mp))
1849		return;
1850
1851	__xfs_refcount_add(tp, XFS_REFCOUNT_ALLOC_COW, fsb, len);
1852
1853	/* Add rmap entry */
1854	xfs_rmap_alloc_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1855			XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1856}
1857
1858/* Forget a CoW staging event in the refcount btree. */
1859void
1860xfs_refcount_free_cow_extent(
1861	struct xfs_trans		*tp,
1862	xfs_fsblock_t			fsb,
1863	xfs_extlen_t			len)
1864{
1865	struct xfs_mount		*mp = tp->t_mountp;
1866
1867	if (!xfs_has_reflink(mp))
1868		return;
1869
1870	/* Remove rmap entry */
1871	xfs_rmap_free_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1872			XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1873	__xfs_refcount_add(tp, XFS_REFCOUNT_FREE_COW, fsb, len);
1874}
1875
1876struct xfs_refcount_recovery {
1877	struct list_head		rr_list;
1878	struct xfs_refcount_irec	rr_rrec;
1879};
1880
1881/* Stuff an extent on the recovery list. */
1882STATIC int
1883xfs_refcount_recover_extent(
1884	struct xfs_btree_cur		*cur,
1885	const union xfs_btree_rec	*rec,
1886	void				*priv)
1887{
1888	struct list_head		*debris = priv;
1889	struct xfs_refcount_recovery	*rr;
1890
1891	if (XFS_IS_CORRUPT(cur->bc_mp,
1892			   be32_to_cpu(rec->refc.rc_refcount) != 1))
1893		return -EFSCORRUPTED;
1894
1895	rr = kmalloc(sizeof(struct xfs_refcount_recovery),
1896			GFP_KERNEL | __GFP_NOFAIL);
1897	INIT_LIST_HEAD(&rr->rr_list);
1898	xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1899
1900	if (xfs_refcount_check_irec(cur->bc_ag.pag, &rr->rr_rrec) != NULL ||
1901	    XFS_IS_CORRUPT(cur->bc_mp,
1902			   rr->rr_rrec.rc_domain != XFS_REFC_DOMAIN_COW)) {
1903		kfree(rr);
1904		return -EFSCORRUPTED;
1905	}
1906
1907	list_add_tail(&rr->rr_list, debris);
1908	return 0;
1909}
1910
1911/* Find and remove leftover CoW reservations. */
1912int
1913xfs_refcount_recover_cow_leftovers(
1914	struct xfs_mount		*mp,
1915	struct xfs_perag		*pag)
1916{
1917	struct xfs_trans		*tp;
1918	struct xfs_btree_cur		*cur;
1919	struct xfs_buf			*agbp;
1920	struct xfs_refcount_recovery	*rr, *n;
1921	struct list_head		debris;
1922	union xfs_btree_irec		low = {
1923		.rc.rc_domain		= XFS_REFC_DOMAIN_COW,
1924	};
1925	union xfs_btree_irec		high = {
1926		.rc.rc_domain		= XFS_REFC_DOMAIN_COW,
1927		.rc.rc_startblock	= -1U,
1928	};
1929	xfs_fsblock_t			fsb;
1930	int				error;
1931
1932	/* reflink filesystems mustn't have AGs larger than 2^31-1 blocks */
1933	BUILD_BUG_ON(XFS_MAX_CRC_AG_BLOCKS >= XFS_REFC_COWFLAG);
1934	if (mp->m_sb.sb_agblocks > XFS_MAX_CRC_AG_BLOCKS)
1935		return -EOPNOTSUPP;
1936
1937	INIT_LIST_HEAD(&debris);
1938
1939	/*
1940	 * In this first part, we use an empty transaction to gather up
1941	 * all the leftover CoW extents so that we can subsequently
1942	 * delete them.  The empty transaction is used to avoid
1943	 * a buffer lock deadlock if there happens to be a loop in the
1944	 * refcountbt because we're allowed to re-grab a buffer that is
1945	 * already attached to our transaction.  When we're done
1946	 * recording the CoW debris we cancel the (empty) transaction
1947	 * and everything goes away cleanly.
1948	 */
1949	error = xfs_trans_alloc_empty(mp, &tp);
1950	if (error)
1951		return error;
1952
1953	error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
1954	if (error)
1955		goto out_trans;
1956	cur = xfs_refcountbt_init_cursor(mp, tp, agbp, pag);
1957
1958	/* Find all the leftover CoW staging extents. */
1959	error = xfs_btree_query_range(cur, &low, &high,
1960			xfs_refcount_recover_extent, &debris);
1961	xfs_btree_del_cursor(cur, error);
1962	xfs_trans_brelse(tp, agbp);
1963	xfs_trans_cancel(tp);
1964	if (error)
1965		goto out_free;
1966
1967	/* Now iterate the list to free the leftovers */
1968	list_for_each_entry_safe(rr, n, &debris, rr_list) {
1969		/* Set up transaction. */
1970		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1971		if (error)
1972			goto out_free;
1973
1974		trace_xfs_refcount_recover_extent(mp, pag->pag_agno,
1975				&rr->rr_rrec);
1976
1977		/* Free the orphan record */
1978		fsb = XFS_AGB_TO_FSB(mp, pag->pag_agno,
1979				rr->rr_rrec.rc_startblock);
1980		xfs_refcount_free_cow_extent(tp, fsb,
1981				rr->rr_rrec.rc_blockcount);
1982
1983		/* Free the block. */
1984		error = xfs_free_extent_later(tp, fsb,
1985				rr->rr_rrec.rc_blockcount, NULL,
1986				XFS_AG_RESV_NONE, false);
1987		if (error)
1988			goto out_trans;
1989
1990		error = xfs_trans_commit(tp);
1991		if (error)
1992			goto out_free;
1993
1994		list_del(&rr->rr_list);
1995		kfree(rr);
1996	}
1997
1998	return error;
1999out_trans:
2000	xfs_trans_cancel(tp);
2001out_free:
2002	/* Free the leftover list */
2003	list_for_each_entry_safe(rr, n, &debris, rr_list) {
2004		list_del(&rr->rr_list);
2005		kfree(rr);
2006	}
2007	return error;
2008}
2009
2010/*
2011 * Scan part of the keyspace of the refcount records and tell us if the area
2012 * has no records, is fully mapped by records, or is partially filled.
2013 */
2014int
2015xfs_refcount_has_records(
2016	struct xfs_btree_cur	*cur,
2017	enum xfs_refc_domain	domain,
2018	xfs_agblock_t		bno,
2019	xfs_extlen_t		len,
2020	enum xbtree_recpacking	*outcome)
2021{
2022	union xfs_btree_irec	low;
2023	union xfs_btree_irec	high;
2024
2025	memset(&low, 0, sizeof(low));
2026	low.rc.rc_startblock = bno;
2027	memset(&high, 0xFF, sizeof(high));
2028	high.rc.rc_startblock = bno + len - 1;
2029	low.rc.rc_domain = high.rc.rc_domain = domain;
2030
2031	return xfs_btree_has_records(cur, &low, &high, NULL, outcome);
2032}
2033
2034struct xfs_refcount_query_range_info {
2035	xfs_refcount_query_range_fn	fn;
2036	void				*priv;
2037};
2038
2039/* Format btree record and pass to our callback. */
2040STATIC int
2041xfs_refcount_query_range_helper(
2042	struct xfs_btree_cur		*cur,
2043	const union xfs_btree_rec	*rec,
2044	void				*priv)
2045{
2046	struct xfs_refcount_query_range_info	*query = priv;
2047	struct xfs_refcount_irec	irec;
2048	xfs_failaddr_t			fa;
2049
2050	xfs_refcount_btrec_to_irec(rec, &irec);
2051	fa = xfs_refcount_check_irec(cur->bc_ag.pag, &irec);
2052	if (fa)
2053		return xfs_refcount_complain_bad_rec(cur, fa, &irec);
2054
2055	return query->fn(cur, &irec, query->priv);
2056}
2057
2058/* Find all refcount records between two keys. */
2059int
2060xfs_refcount_query_range(
2061	struct xfs_btree_cur		*cur,
2062	const struct xfs_refcount_irec	*low_rec,
2063	const struct xfs_refcount_irec	*high_rec,
2064	xfs_refcount_query_range_fn	fn,
2065	void				*priv)
2066{
2067	union xfs_btree_irec		low_brec = { .rc = *low_rec };
2068	union xfs_btree_irec		high_brec = { .rc = *high_rec };
2069	struct xfs_refcount_query_range_info query = { .priv = priv, .fn = fn };
2070
2071	return xfs_btree_query_range(cur, &low_brec, &high_brec,
2072			xfs_refcount_query_range_helper, &query);
2073}
2074
2075int __init
2076xfs_refcount_intent_init_cache(void)
2077{
2078	xfs_refcount_intent_cache = kmem_cache_create("xfs_refc_intent",
2079			sizeof(struct xfs_refcount_intent),
2080			0, 0, NULL);
2081
2082	return xfs_refcount_intent_cache != NULL ? 0 : -ENOMEM;
2083}
2084
2085void
2086xfs_refcount_intent_destroy_cache(void)
2087{
2088	kmem_cache_destroy(xfs_refcount_intent_cache);
2089	xfs_refcount_intent_cache = NULL;
2090}