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