Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.15.
   1/*
   2 * Copyright (C) 2016 Oracle.  All Rights Reserved.
   3 *
   4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License
   8 * as published by the Free Software Foundation; either version 2
   9 * of the License, or (at your option) any later version.
  10 *
  11 * This program is distributed in the hope that it would be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with this program; if not, write the Free Software Foundation,
  18 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
  19 */
  20#include "xfs.h"
  21#include "xfs_fs.h"
  22#include "xfs_shared.h"
  23#include "xfs_format.h"
  24#include "xfs_log_format.h"
  25#include "xfs_trans_resv.h"
  26#include "xfs_sb.h"
  27#include "xfs_mount.h"
  28#include "xfs_defer.h"
  29#include "xfs_btree.h"
  30#include "xfs_bmap.h"
  31#include "xfs_refcount_btree.h"
  32#include "xfs_alloc.h"
  33#include "xfs_errortag.h"
  34#include "xfs_error.h"
  35#include "xfs_trace.h"
  36#include "xfs_cksum.h"
  37#include "xfs_trans.h"
  38#include "xfs_bit.h"
  39#include "xfs_refcount.h"
  40#include "xfs_rmap.h"
  41
  42/* Allowable refcount adjustment amounts. */
  43enum xfs_refc_adjust_op {
  44	XFS_REFCOUNT_ADJUST_INCREASE	= 1,
  45	XFS_REFCOUNT_ADJUST_DECREASE	= -1,
  46	XFS_REFCOUNT_ADJUST_COW_ALLOC	= 0,
  47	XFS_REFCOUNT_ADJUST_COW_FREE	= -1,
  48};
  49
  50STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
  51		xfs_agblock_t agbno, xfs_extlen_t aglen,
  52		struct xfs_defer_ops *dfops);
  53STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
  54		xfs_agblock_t agbno, xfs_extlen_t aglen,
  55		struct xfs_defer_ops *dfops);
  56
  57/*
  58 * Look up the first record less than or equal to [bno, len] in the btree
  59 * given by cur.
  60 */
  61int
  62xfs_refcount_lookup_le(
  63	struct xfs_btree_cur	*cur,
  64	xfs_agblock_t		bno,
  65	int			*stat)
  66{
  67	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
  68			XFS_LOOKUP_LE);
  69	cur->bc_rec.rc.rc_startblock = bno;
  70	cur->bc_rec.rc.rc_blockcount = 0;
  71	return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
  72}
  73
  74/*
  75 * Look up the first record greater than or equal to [bno, len] in the btree
  76 * given by cur.
  77 */
  78int
  79xfs_refcount_lookup_ge(
  80	struct xfs_btree_cur	*cur,
  81	xfs_agblock_t		bno,
  82	int			*stat)
  83{
  84	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
  85			XFS_LOOKUP_GE);
  86	cur->bc_rec.rc.rc_startblock = bno;
  87	cur->bc_rec.rc.rc_blockcount = 0;
  88	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
  89}
  90
  91/* Convert on-disk record to in-core format. */
  92static inline void
  93xfs_refcount_btrec_to_irec(
  94	union xfs_btree_rec		*rec,
  95	struct xfs_refcount_irec	*irec)
  96{
  97	irec->rc_startblock = be32_to_cpu(rec->refc.rc_startblock);
  98	irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
  99	irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
 100}
 101
 102/*
 103 * Get the data from the pointed-to record.
 104 */
 105int
 106xfs_refcount_get_rec(
 107	struct xfs_btree_cur		*cur,
 108	struct xfs_refcount_irec	*irec,
 109	int				*stat)
 110{
 111	union xfs_btree_rec		*rec;
 112	int				error;
 113
 114	error = xfs_btree_get_rec(cur, &rec, stat);
 115	if (!error && *stat == 1) {
 116		xfs_refcount_btrec_to_irec(rec, irec);
 117		trace_xfs_refcount_get(cur->bc_mp, cur->bc_private.a.agno,
 118				irec);
 119	}
 120	return error;
 121}
 122
 123/*
 124 * Update the record referred to by cur to the value given
 125 * by [bno, len, refcount].
 126 * This either works (return 0) or gets an EFSCORRUPTED error.
 127 */
 128STATIC int
 129xfs_refcount_update(
 130	struct xfs_btree_cur		*cur,
 131	struct xfs_refcount_irec	*irec)
 132{
 133	union xfs_btree_rec	rec;
 134	int			error;
 135
 136	trace_xfs_refcount_update(cur->bc_mp, cur->bc_private.a.agno, irec);
 137	rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
 138	rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
 139	rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
 140	error = xfs_btree_update(cur, &rec);
 141	if (error)
 142		trace_xfs_refcount_update_error(cur->bc_mp,
 143				cur->bc_private.a.agno, error, _RET_IP_);
 144	return error;
 145}
 146
 147/*
 148 * Insert the record referred to by cur to the value given
 149 * by [bno, len, refcount].
 150 * This either works (return 0) or gets an EFSCORRUPTED error.
 151 */
 152STATIC int
 153xfs_refcount_insert(
 154	struct xfs_btree_cur		*cur,
 155	struct xfs_refcount_irec	*irec,
 156	int				*i)
 157{
 158	int				error;
 159
 160	trace_xfs_refcount_insert(cur->bc_mp, cur->bc_private.a.agno, irec);
 161	cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
 162	cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
 163	cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
 164	error = xfs_btree_insert(cur, i);
 165	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
 166out_error:
 167	if (error)
 168		trace_xfs_refcount_insert_error(cur->bc_mp,
 169				cur->bc_private.a.agno, error, _RET_IP_);
 170	return error;
 171}
 172
 173/*
 174 * Remove the record referred to by cur, then set the pointer to the spot
 175 * where the record could be re-inserted, in case we want to increment or
 176 * decrement the cursor.
 177 * This either works (return 0) or gets an EFSCORRUPTED error.
 178 */
 179STATIC int
 180xfs_refcount_delete(
 181	struct xfs_btree_cur	*cur,
 182	int			*i)
 183{
 184	struct xfs_refcount_irec	irec;
 185	int			found_rec;
 186	int			error;
 187
 188	error = xfs_refcount_get_rec(cur, &irec, &found_rec);
 189	if (error)
 190		goto out_error;
 191	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 192	trace_xfs_refcount_delete(cur->bc_mp, cur->bc_private.a.agno, &irec);
 193	error = xfs_btree_delete(cur, i);
 194	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
 195	if (error)
 196		goto out_error;
 197	error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
 198out_error:
 199	if (error)
 200		trace_xfs_refcount_delete_error(cur->bc_mp,
 201				cur->bc_private.a.agno, error, _RET_IP_);
 202	return error;
 203}
 204
 205/*
 206 * Adjusting the Reference Count
 207 *
 208 * As stated elsewhere, the reference count btree (refcbt) stores
 209 * >1 reference counts for extents of physical blocks.  In this
 210 * operation, we're either raising or lowering the reference count of
 211 * some subrange stored in the tree:
 212 *
 213 *      <------ adjustment range ------>
 214 * ----+   +---+-----+ +--+--------+---------
 215 *  2  |   | 3 |  4  | |17|   55   |   10
 216 * ----+   +---+-----+ +--+--------+---------
 217 * X axis is physical blocks number;
 218 * reference counts are the numbers inside the rectangles
 219 *
 220 * The first thing we need to do is to ensure that there are no
 221 * refcount extents crossing either boundary of the range to be
 222 * adjusted.  For any extent that does cross a boundary, split it into
 223 * two extents so that we can increment the refcount of one of the
 224 * pieces later:
 225 *
 226 *      <------ adjustment range ------>
 227 * ----+   +---+-----+ +--+--------+----+----
 228 *  2  |   | 3 |  2  | |17|   55   | 10 | 10
 229 * ----+   +---+-----+ +--+--------+----+----
 230 *
 231 * For this next step, let's assume that all the physical blocks in
 232 * the adjustment range are mapped to a file and are therefore in use
 233 * at least once.  Therefore, we can infer that any gap in the
 234 * refcount tree within the adjustment range represents a physical
 235 * extent with refcount == 1:
 236 *
 237 *      <------ adjustment range ------>
 238 * ----+---+---+-----+-+--+--------+----+----
 239 *  2  |"1"| 3 |  2  |1|17|   55   | 10 | 10
 240 * ----+---+---+-----+-+--+--------+----+----
 241 *      ^
 242 *
 243 * For each extent that falls within the interval range, figure out
 244 * which extent is to the left or the right of that extent.  Now we
 245 * have a left, current, and right extent.  If the new reference count
 246 * of the center extent enables us to merge left, center, and right
 247 * into one record covering all three, do so.  If the center extent is
 248 * at the left end of the range, abuts the left extent, and its new
 249 * reference count matches the left extent's record, then merge them.
 250 * If the center extent is at the right end of the range, abuts the
 251 * right extent, and the reference counts match, merge those.  In the
 252 * example, we can left merge (assuming an increment operation):
 253 *
 254 *      <------ adjustment range ------>
 255 * --------+---+-----+-+--+--------+----+----
 256 *    2    | 3 |  2  |1|17|   55   | 10 | 10
 257 * --------+---+-----+-+--+--------+----+----
 258 *          ^
 259 *
 260 * For all other extents within the range, adjust the reference count
 261 * or delete it if the refcount falls below 2.  If we were
 262 * incrementing, the end result looks like this:
 263 *
 264 *      <------ adjustment range ------>
 265 * --------+---+-----+-+--+--------+----+----
 266 *    2    | 4 |  3  |2|18|   56   | 11 | 10
 267 * --------+---+-----+-+--+--------+----+----
 268 *
 269 * The result of a decrement operation looks as such:
 270 *
 271 *      <------ adjustment range ------>
 272 * ----+   +---+       +--+--------+----+----
 273 *  2  |   | 2 |       |16|   54   |  9 | 10
 274 * ----+   +---+       +--+--------+----+----
 275 *      DDDD    111111DD
 276 *
 277 * The blocks marked "D" are freed; the blocks marked "1" are only
 278 * referenced once and therefore the record is removed from the
 279 * refcount btree.
 280 */
 281
 282/* Next block after this extent. */
 283static inline xfs_agblock_t
 284xfs_refc_next(
 285	struct xfs_refcount_irec	*rc)
 286{
 287	return rc->rc_startblock + rc->rc_blockcount;
 288}
 289
 290/*
 291 * Split a refcount extent that crosses agbno.
 292 */
 293STATIC int
 294xfs_refcount_split_extent(
 295	struct xfs_btree_cur		*cur,
 296	xfs_agblock_t			agbno,
 297	bool				*shape_changed)
 298{
 299	struct xfs_refcount_irec	rcext, tmp;
 300	int				found_rec;
 301	int				error;
 302
 303	*shape_changed = false;
 304	error = xfs_refcount_lookup_le(cur, agbno, &found_rec);
 305	if (error)
 306		goto out_error;
 307	if (!found_rec)
 308		return 0;
 309
 310	error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
 311	if (error)
 312		goto out_error;
 313	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 314	if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
 315		return 0;
 316
 317	*shape_changed = true;
 318	trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_private.a.agno,
 319			&rcext, agbno);
 320
 321	/* Establish the right extent. */
 322	tmp = rcext;
 323	tmp.rc_startblock = agbno;
 324	tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
 325	error = xfs_refcount_update(cur, &tmp);
 326	if (error)
 327		goto out_error;
 328
 329	/* Insert the left extent. */
 330	tmp = rcext;
 331	tmp.rc_blockcount = agbno - rcext.rc_startblock;
 332	error = xfs_refcount_insert(cur, &tmp, &found_rec);
 333	if (error)
 334		goto out_error;
 335	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 336	return error;
 337
 338out_error:
 339	trace_xfs_refcount_split_extent_error(cur->bc_mp,
 340			cur->bc_private.a.agno, error, _RET_IP_);
 341	return error;
 342}
 343
 344/*
 345 * Merge the left, center, and right extents.
 346 */
 347STATIC int
 348xfs_refcount_merge_center_extents(
 349	struct xfs_btree_cur		*cur,
 350	struct xfs_refcount_irec	*left,
 351	struct xfs_refcount_irec	*center,
 352	struct xfs_refcount_irec	*right,
 353	unsigned long long		extlen,
 354	xfs_extlen_t			*aglen)
 355{
 356	int				error;
 357	int				found_rec;
 358
 359	trace_xfs_refcount_merge_center_extents(cur->bc_mp,
 360			cur->bc_private.a.agno, left, center, right);
 361
 362	/*
 363	 * Make sure the center and right extents are not in the btree.
 364	 * If the center extent was synthesized, the first delete call
 365	 * removes the right extent and we skip the second deletion.
 366	 * If center and right were in the btree, then the first delete
 367	 * call removes the center and the second one removes the right
 368	 * extent.
 369	 */
 370	error = xfs_refcount_lookup_ge(cur, center->rc_startblock,
 371			&found_rec);
 372	if (error)
 373		goto out_error;
 374	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 375
 376	error = xfs_refcount_delete(cur, &found_rec);
 377	if (error)
 378		goto out_error;
 379	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 380
 381	if (center->rc_refcount > 1) {
 382		error = xfs_refcount_delete(cur, &found_rec);
 383		if (error)
 384			goto out_error;
 385		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 386				out_error);
 387	}
 388
 389	/* Enlarge the left extent. */
 390	error = xfs_refcount_lookup_le(cur, left->rc_startblock,
 391			&found_rec);
 392	if (error)
 393		goto out_error;
 394	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 395
 396	left->rc_blockcount = extlen;
 397	error = xfs_refcount_update(cur, left);
 398	if (error)
 399		goto out_error;
 400
 401	*aglen = 0;
 402	return error;
 403
 404out_error:
 405	trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
 406			cur->bc_private.a.agno, error, _RET_IP_);
 407	return error;
 408}
 409
 410/*
 411 * Merge with the left extent.
 412 */
 413STATIC int
 414xfs_refcount_merge_left_extent(
 415	struct xfs_btree_cur		*cur,
 416	struct xfs_refcount_irec	*left,
 417	struct xfs_refcount_irec	*cleft,
 418	xfs_agblock_t			*agbno,
 419	xfs_extlen_t			*aglen)
 420{
 421	int				error;
 422	int				found_rec;
 423
 424	trace_xfs_refcount_merge_left_extent(cur->bc_mp,
 425			cur->bc_private.a.agno, left, cleft);
 426
 427	/* If the extent at agbno (cleft) wasn't synthesized, remove it. */
 428	if (cleft->rc_refcount > 1) {
 429		error = xfs_refcount_lookup_le(cur, cleft->rc_startblock,
 430				&found_rec);
 431		if (error)
 432			goto out_error;
 433		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 434				out_error);
 435
 436		error = xfs_refcount_delete(cur, &found_rec);
 437		if (error)
 438			goto out_error;
 439		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 440				out_error);
 441	}
 442
 443	/* Enlarge the left extent. */
 444	error = xfs_refcount_lookup_le(cur, left->rc_startblock,
 445			&found_rec);
 446	if (error)
 447		goto out_error;
 448	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 449
 450	left->rc_blockcount += cleft->rc_blockcount;
 451	error = xfs_refcount_update(cur, left);
 452	if (error)
 453		goto out_error;
 454
 455	*agbno += cleft->rc_blockcount;
 456	*aglen -= cleft->rc_blockcount;
 457	return error;
 458
 459out_error:
 460	trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
 461			cur->bc_private.a.agno, error, _RET_IP_);
 462	return error;
 463}
 464
 465/*
 466 * Merge with the right extent.
 467 */
 468STATIC int
 469xfs_refcount_merge_right_extent(
 470	struct xfs_btree_cur		*cur,
 471	struct xfs_refcount_irec	*right,
 472	struct xfs_refcount_irec	*cright,
 473	xfs_extlen_t			*aglen)
 474{
 475	int				error;
 476	int				found_rec;
 477
 478	trace_xfs_refcount_merge_right_extent(cur->bc_mp,
 479			cur->bc_private.a.agno, cright, right);
 480
 481	/*
 482	 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
 483	 * remove it.
 484	 */
 485	if (cright->rc_refcount > 1) {
 486		error = xfs_refcount_lookup_le(cur, cright->rc_startblock,
 487			&found_rec);
 488		if (error)
 489			goto out_error;
 490		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 491				out_error);
 492
 493		error = xfs_refcount_delete(cur, &found_rec);
 494		if (error)
 495			goto out_error;
 496		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 497				out_error);
 498	}
 499
 500	/* Enlarge the right extent. */
 501	error = xfs_refcount_lookup_le(cur, right->rc_startblock,
 502			&found_rec);
 503	if (error)
 504		goto out_error;
 505	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 506
 507	right->rc_startblock -= cright->rc_blockcount;
 508	right->rc_blockcount += cright->rc_blockcount;
 509	error = xfs_refcount_update(cur, right);
 510	if (error)
 511		goto out_error;
 512
 513	*aglen -= cright->rc_blockcount;
 514	return error;
 515
 516out_error:
 517	trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
 518			cur->bc_private.a.agno, error, _RET_IP_);
 519	return error;
 520}
 521
 522#define XFS_FIND_RCEXT_SHARED	1
 523#define XFS_FIND_RCEXT_COW	2
 524/*
 525 * Find the left extent and the one after it (cleft).  This function assumes
 526 * that we've already split any extent crossing agbno.
 527 */
 528STATIC int
 529xfs_refcount_find_left_extents(
 530	struct xfs_btree_cur		*cur,
 531	struct xfs_refcount_irec	*left,
 532	struct xfs_refcount_irec	*cleft,
 533	xfs_agblock_t			agbno,
 534	xfs_extlen_t			aglen,
 535	int				flags)
 536{
 537	struct xfs_refcount_irec	tmp;
 538	int				error;
 539	int				found_rec;
 540
 541	left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
 542	error = xfs_refcount_lookup_le(cur, agbno - 1, &found_rec);
 543	if (error)
 544		goto out_error;
 545	if (!found_rec)
 546		return 0;
 547
 548	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 549	if (error)
 550		goto out_error;
 551	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 552
 553	if (xfs_refc_next(&tmp) != agbno)
 554		return 0;
 555	if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
 556		return 0;
 557	if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
 558		return 0;
 559	/* We have a left extent; retrieve (or invent) the next right one */
 560	*left = tmp;
 561
 562	error = xfs_btree_increment(cur, 0, &found_rec);
 563	if (error)
 564		goto out_error;
 565	if (found_rec) {
 566		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 567		if (error)
 568			goto out_error;
 569		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 570				out_error);
 571
 572		/* if tmp starts at the end of our range, just use that */
 573		if (tmp.rc_startblock == agbno)
 574			*cleft = tmp;
 575		else {
 576			/*
 577			 * There's a gap in the refcntbt at the start of the
 578			 * range we're interested in (refcount == 1) so
 579			 * synthesize the implied extent and pass it back.
 580			 * We assume here that the agbno/aglen range was
 581			 * passed in from a data fork extent mapping and
 582			 * therefore is allocated to exactly one owner.
 583			 */
 584			cleft->rc_startblock = agbno;
 585			cleft->rc_blockcount = min(aglen,
 586					tmp.rc_startblock - agbno);
 587			cleft->rc_refcount = 1;
 588		}
 589	} else {
 590		/*
 591		 * No extents, so pretend that there's one covering the whole
 592		 * range.
 593		 */
 594		cleft->rc_startblock = agbno;
 595		cleft->rc_blockcount = aglen;
 596		cleft->rc_refcount = 1;
 597	}
 598	trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_private.a.agno,
 599			left, cleft, agbno);
 600	return error;
 601
 602out_error:
 603	trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
 604			cur->bc_private.a.agno, error, _RET_IP_);
 605	return error;
 606}
 607
 608/*
 609 * Find the right extent and the one before it (cright).  This function
 610 * assumes that we've already split any extents crossing agbno + aglen.
 611 */
 612STATIC int
 613xfs_refcount_find_right_extents(
 614	struct xfs_btree_cur		*cur,
 615	struct xfs_refcount_irec	*right,
 616	struct xfs_refcount_irec	*cright,
 617	xfs_agblock_t			agbno,
 618	xfs_extlen_t			aglen,
 619	int				flags)
 620{
 621	struct xfs_refcount_irec	tmp;
 622	int				error;
 623	int				found_rec;
 624
 625	right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
 626	error = xfs_refcount_lookup_ge(cur, agbno + aglen, &found_rec);
 627	if (error)
 628		goto out_error;
 629	if (!found_rec)
 630		return 0;
 631
 632	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 633	if (error)
 634		goto out_error;
 635	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
 636
 637	if (tmp.rc_startblock != agbno + aglen)
 638		return 0;
 639	if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
 640		return 0;
 641	if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
 642		return 0;
 643	/* We have a right extent; retrieve (or invent) the next left one */
 644	*right = tmp;
 645
 646	error = xfs_btree_decrement(cur, 0, &found_rec);
 647	if (error)
 648		goto out_error;
 649	if (found_rec) {
 650		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
 651		if (error)
 652			goto out_error;
 653		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
 654				out_error);
 655
 656		/* if tmp ends at the end of our range, just use that */
 657		if (xfs_refc_next(&tmp) == agbno + aglen)
 658			*cright = tmp;
 659		else {
 660			/*
 661			 * There's a gap in the refcntbt at the end of the
 662			 * range we're interested in (refcount == 1) so
 663			 * create the implied extent and pass it back.
 664			 * We assume here that the agbno/aglen range was
 665			 * passed in from a data fork extent mapping and
 666			 * therefore is allocated to exactly one owner.
 667			 */
 668			cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
 669			cright->rc_blockcount = right->rc_startblock -
 670					cright->rc_startblock;
 671			cright->rc_refcount = 1;
 672		}
 673	} else {
 674		/*
 675		 * No extents, so pretend that there's one covering the whole
 676		 * range.
 677		 */
 678		cright->rc_startblock = agbno;
 679		cright->rc_blockcount = aglen;
 680		cright->rc_refcount = 1;
 681	}
 682	trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_private.a.agno,
 683			cright, right, agbno + aglen);
 684	return error;
 685
 686out_error:
 687	trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
 688			cur->bc_private.a.agno, error, _RET_IP_);
 689	return error;
 690}
 691
 692/* Is this extent valid? */
 693static inline bool
 694xfs_refc_valid(
 695	struct xfs_refcount_irec	*rc)
 696{
 697	return rc->rc_startblock != NULLAGBLOCK;
 698}
 699
 700/*
 701 * Try to merge with any extents on the boundaries of the adjustment range.
 702 */
 703STATIC int
 704xfs_refcount_merge_extents(
 705	struct xfs_btree_cur	*cur,
 706	xfs_agblock_t		*agbno,
 707	xfs_extlen_t		*aglen,
 708	enum xfs_refc_adjust_op adjust,
 709	int			flags,
 710	bool			*shape_changed)
 711{
 712	struct xfs_refcount_irec	left = {0}, cleft = {0};
 713	struct xfs_refcount_irec	cright = {0}, right = {0};
 714	int				error;
 715	unsigned long long		ulen;
 716	bool				cequal;
 717
 718	*shape_changed = false;
 719	/*
 720	 * Find the extent just below agbno [left], just above agbno [cleft],
 721	 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
 722	 * [right].
 723	 */
 724	error = xfs_refcount_find_left_extents(cur, &left, &cleft, *agbno,
 725			*aglen, flags);
 726	if (error)
 727		return error;
 728	error = xfs_refcount_find_right_extents(cur, &right, &cright, *agbno,
 729			*aglen, flags);
 730	if (error)
 731		return error;
 732
 733	/* No left or right extent to merge; exit. */
 734	if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
 735		return 0;
 736
 737	cequal = (cleft.rc_startblock == cright.rc_startblock) &&
 738		 (cleft.rc_blockcount == cright.rc_blockcount);
 739
 740	/* Try to merge left, cleft, and right.  cleft must == cright. */
 741	ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount +
 742			right.rc_blockcount;
 743	if (xfs_refc_valid(&left) && xfs_refc_valid(&right) &&
 744	    xfs_refc_valid(&cleft) && xfs_refc_valid(&cright) && cequal &&
 745	    left.rc_refcount == cleft.rc_refcount + adjust &&
 746	    right.rc_refcount == cleft.rc_refcount + adjust &&
 747	    ulen < MAXREFCEXTLEN) {
 748		*shape_changed = true;
 749		return xfs_refcount_merge_center_extents(cur, &left, &cleft,
 750				&right, ulen, aglen);
 751	}
 752
 753	/* Try to merge left and cleft. */
 754	ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount;
 755	if (xfs_refc_valid(&left) && xfs_refc_valid(&cleft) &&
 756	    left.rc_refcount == cleft.rc_refcount + adjust &&
 757	    ulen < MAXREFCEXTLEN) {
 758		*shape_changed = true;
 759		error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
 760				agbno, aglen);
 761		if (error)
 762			return error;
 763
 764		/*
 765		 * If we just merged left + cleft and cleft == cright,
 766		 * we no longer have a cright to merge with right.  We're done.
 767		 */
 768		if (cequal)
 769			return 0;
 770	}
 771
 772	/* Try to merge cright and right. */
 773	ulen = (unsigned long long)right.rc_blockcount + cright.rc_blockcount;
 774	if (xfs_refc_valid(&right) && xfs_refc_valid(&cright) &&
 775	    right.rc_refcount == cright.rc_refcount + adjust &&
 776	    ulen < MAXREFCEXTLEN) {
 777		*shape_changed = true;
 778		return xfs_refcount_merge_right_extent(cur, &right, &cright,
 779				aglen);
 780	}
 781
 782	return error;
 783}
 784
 785/*
 786 * XXX: This is a pretty hand-wavy estimate.  The penalty for guessing
 787 * true incorrectly is a shutdown FS; the penalty for guessing false
 788 * incorrectly is more transaction rolls than might be necessary.
 789 * Be conservative here.
 790 */
 791static bool
 792xfs_refcount_still_have_space(
 793	struct xfs_btree_cur		*cur)
 794{
 795	unsigned long			overhead;
 796
 797	overhead = cur->bc_private.a.priv.refc.shape_changes *
 798			xfs_allocfree_log_count(cur->bc_mp, 1);
 799	overhead *= cur->bc_mp->m_sb.sb_blocksize;
 800
 801	/*
 802	 * Only allow 2 refcount extent updates per transaction if the
 803	 * refcount continue update "error" has been injected.
 804	 */
 805	if (cur->bc_private.a.priv.refc.nr_ops > 2 &&
 806	    XFS_TEST_ERROR(false, cur->bc_mp,
 807			XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
 808		return false;
 809
 810	if (cur->bc_private.a.priv.refc.nr_ops == 0)
 811		return true;
 812	else if (overhead > cur->bc_tp->t_log_res)
 813		return false;
 814	return  cur->bc_tp->t_log_res - overhead >
 815		cur->bc_private.a.priv.refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
 816}
 817
 818/*
 819 * Adjust the refcounts of middle extents.  At this point we should have
 820 * split extents that crossed the adjustment range; merged with adjacent
 821 * extents; and updated agbno/aglen to reflect the merges.  Therefore,
 822 * all we have to do is update the extents inside [agbno, agbno + aglen].
 823 */
 824STATIC int
 825xfs_refcount_adjust_extents(
 826	struct xfs_btree_cur	*cur,
 827	xfs_agblock_t		*agbno,
 828	xfs_extlen_t		*aglen,
 829	enum xfs_refc_adjust_op	adj,
 830	struct xfs_defer_ops	*dfops,
 831	struct xfs_owner_info	*oinfo)
 832{
 833	struct xfs_refcount_irec	ext, tmp;
 834	int				error;
 835	int				found_rec, found_tmp;
 836	xfs_fsblock_t			fsbno;
 837
 838	/* Merging did all the work already. */
 839	if (*aglen == 0)
 840		return 0;
 841
 842	error = xfs_refcount_lookup_ge(cur, *agbno, &found_rec);
 843	if (error)
 844		goto out_error;
 845
 846	while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
 847		error = xfs_refcount_get_rec(cur, &ext, &found_rec);
 848		if (error)
 849			goto out_error;
 850		if (!found_rec) {
 851			ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
 852			ext.rc_blockcount = 0;
 853			ext.rc_refcount = 0;
 854		}
 855
 856		/*
 857		 * Deal with a hole in the refcount tree; if a file maps to
 858		 * these blocks and there's no refcountbt record, pretend that
 859		 * there is one with refcount == 1.
 860		 */
 861		if (ext.rc_startblock != *agbno) {
 862			tmp.rc_startblock = *agbno;
 863			tmp.rc_blockcount = min(*aglen,
 864					ext.rc_startblock - *agbno);
 865			tmp.rc_refcount = 1 + adj;
 866			trace_xfs_refcount_modify_extent(cur->bc_mp,
 867					cur->bc_private.a.agno, &tmp);
 868
 869			/*
 870			 * Either cover the hole (increment) or
 871			 * delete the range (decrement).
 872			 */
 873			if (tmp.rc_refcount) {
 874				error = xfs_refcount_insert(cur, &tmp,
 875						&found_tmp);
 876				if (error)
 877					goto out_error;
 878				XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
 879						found_tmp == 1, out_error);
 880				cur->bc_private.a.priv.refc.nr_ops++;
 881			} else {
 882				fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
 883						cur->bc_private.a.agno,
 884						tmp.rc_startblock);
 885				xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
 886						tmp.rc_blockcount, oinfo);
 887			}
 888
 889			(*agbno) += tmp.rc_blockcount;
 890			(*aglen) -= tmp.rc_blockcount;
 891
 892			error = xfs_refcount_lookup_ge(cur, *agbno,
 893					&found_rec);
 894			if (error)
 895				goto out_error;
 896		}
 897
 898		/* Stop if there's nothing left to modify */
 899		if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
 900			break;
 901
 902		/*
 903		 * Adjust the reference count and either update the tree
 904		 * (incr) or free the blocks (decr).
 905		 */
 906		if (ext.rc_refcount == MAXREFCOUNT)
 907			goto skip;
 908		ext.rc_refcount += adj;
 909		trace_xfs_refcount_modify_extent(cur->bc_mp,
 910				cur->bc_private.a.agno, &ext);
 911		if (ext.rc_refcount > 1) {
 912			error = xfs_refcount_update(cur, &ext);
 913			if (error)
 914				goto out_error;
 915			cur->bc_private.a.priv.refc.nr_ops++;
 916		} else if (ext.rc_refcount == 1) {
 917			error = xfs_refcount_delete(cur, &found_rec);
 918			if (error)
 919				goto out_error;
 920			XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
 921					found_rec == 1, out_error);
 922			cur->bc_private.a.priv.refc.nr_ops++;
 923			goto advloop;
 924		} else {
 925			fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
 926					cur->bc_private.a.agno,
 927					ext.rc_startblock);
 928			xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
 929					ext.rc_blockcount, oinfo);
 930		}
 931
 932skip:
 933		error = xfs_btree_increment(cur, 0, &found_rec);
 934		if (error)
 935			goto out_error;
 936
 937advloop:
 938		(*agbno) += ext.rc_blockcount;
 939		(*aglen) -= ext.rc_blockcount;
 940	}
 941
 942	return error;
 943out_error:
 944	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
 945			cur->bc_private.a.agno, error, _RET_IP_);
 946	return error;
 947}
 948
 949/* Adjust the reference count of a range of AG blocks. */
 950STATIC int
 951xfs_refcount_adjust(
 952	struct xfs_btree_cur	*cur,
 953	xfs_agblock_t		agbno,
 954	xfs_extlen_t		aglen,
 955	xfs_agblock_t		*new_agbno,
 956	xfs_extlen_t		*new_aglen,
 957	enum xfs_refc_adjust_op	adj,
 958	struct xfs_defer_ops	*dfops,
 959	struct xfs_owner_info	*oinfo)
 960{
 961	bool			shape_changed;
 962	int			shape_changes = 0;
 963	int			error;
 964
 965	*new_agbno = agbno;
 966	*new_aglen = aglen;
 967	if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
 968		trace_xfs_refcount_increase(cur->bc_mp, cur->bc_private.a.agno,
 969				agbno, aglen);
 970	else
 971		trace_xfs_refcount_decrease(cur->bc_mp, cur->bc_private.a.agno,
 972				agbno, aglen);
 973
 974	/*
 975	 * Ensure that no rcextents cross the boundary of the adjustment range.
 976	 */
 977	error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
 978	if (error)
 979		goto out_error;
 980	if (shape_changed)
 981		shape_changes++;
 982
 983	error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
 984	if (error)
 985		goto out_error;
 986	if (shape_changed)
 987		shape_changes++;
 988
 989	/*
 990	 * Try to merge with the left or right extents of the range.
 991	 */
 992	error = xfs_refcount_merge_extents(cur, new_agbno, new_aglen, adj,
 993			XFS_FIND_RCEXT_SHARED, &shape_changed);
 994	if (error)
 995		goto out_error;
 996	if (shape_changed)
 997		shape_changes++;
 998	if (shape_changes)
 999		cur->bc_private.a.priv.refc.shape_changes++;
1000
1001	/* Now that we've taken care of the ends, adjust the middle extents */
1002	error = xfs_refcount_adjust_extents(cur, new_agbno, new_aglen,
1003			adj, dfops, oinfo);
1004	if (error)
1005		goto out_error;
1006
1007	return 0;
1008
1009out_error:
1010	trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_private.a.agno,
1011			error, _RET_IP_);
1012	return error;
1013}
1014
1015/* Clean up after calling xfs_refcount_finish_one. */
1016void
1017xfs_refcount_finish_one_cleanup(
1018	struct xfs_trans	*tp,
1019	struct xfs_btree_cur	*rcur,
1020	int			error)
1021{
1022	struct xfs_buf		*agbp;
1023
1024	if (rcur == NULL)
1025		return;
1026	agbp = rcur->bc_private.a.agbp;
1027	xfs_btree_del_cursor(rcur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1028	if (error)
1029		xfs_trans_brelse(tp, agbp);
1030}
1031
1032/*
1033 * Process one of the deferred refcount operations.  We pass back the
1034 * btree cursor to maintain our lock on the btree between calls.
1035 * This saves time and eliminates a buffer deadlock between the
1036 * superblock and the AGF because we'll always grab them in the same
1037 * order.
1038 */
1039int
1040xfs_refcount_finish_one(
1041	struct xfs_trans		*tp,
1042	struct xfs_defer_ops		*dfops,
1043	enum xfs_refcount_intent_type	type,
1044	xfs_fsblock_t			startblock,
1045	xfs_extlen_t			blockcount,
1046	xfs_fsblock_t			*new_fsb,
1047	xfs_extlen_t			*new_len,
1048	struct xfs_btree_cur		**pcur)
1049{
1050	struct xfs_mount		*mp = tp->t_mountp;
1051	struct xfs_btree_cur		*rcur;
1052	struct xfs_buf			*agbp = NULL;
1053	int				error = 0;
1054	xfs_agnumber_t			agno;
1055	xfs_agblock_t			bno;
1056	xfs_agblock_t			new_agbno;
1057	unsigned long			nr_ops = 0;
1058	int				shape_changes = 0;
1059
1060	agno = XFS_FSB_TO_AGNO(mp, startblock);
1061	ASSERT(agno != NULLAGNUMBER);
1062	bno = XFS_FSB_TO_AGBNO(mp, startblock);
1063
1064	trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, startblock),
1065			type, XFS_FSB_TO_AGBNO(mp, startblock),
1066			blockcount);
1067
1068	if (XFS_TEST_ERROR(false, mp,
1069			XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1070		return -EIO;
1071
1072	/*
1073	 * If we haven't gotten a cursor or the cursor AG doesn't match
1074	 * the startblock, get one now.
1075	 */
1076	rcur = *pcur;
1077	if (rcur != NULL && rcur->bc_private.a.agno != agno) {
1078		nr_ops = rcur->bc_private.a.priv.refc.nr_ops;
1079		shape_changes = rcur->bc_private.a.priv.refc.shape_changes;
1080		xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1081		rcur = NULL;
1082		*pcur = NULL;
1083	}
1084	if (rcur == NULL) {
1085		error = xfs_alloc_read_agf(tp->t_mountp, tp, agno,
1086				XFS_ALLOC_FLAG_FREEING, &agbp);
1087		if (error)
1088			return error;
1089		if (!agbp)
1090			return -EFSCORRUPTED;
1091
1092		rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, dfops);
1093		if (!rcur) {
1094			error = -ENOMEM;
1095			goto out_cur;
1096		}
1097		rcur->bc_private.a.priv.refc.nr_ops = nr_ops;
1098		rcur->bc_private.a.priv.refc.shape_changes = shape_changes;
1099	}
1100	*pcur = rcur;
1101
1102	switch (type) {
1103	case XFS_REFCOUNT_INCREASE:
1104		error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1105			new_len, XFS_REFCOUNT_ADJUST_INCREASE, dfops, NULL);
1106		*new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1107		break;
1108	case XFS_REFCOUNT_DECREASE:
1109		error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1110			new_len, XFS_REFCOUNT_ADJUST_DECREASE, dfops, NULL);
1111		*new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1112		break;
1113	case XFS_REFCOUNT_ALLOC_COW:
1114		*new_fsb = startblock + blockcount;
1115		*new_len = 0;
1116		error = __xfs_refcount_cow_alloc(rcur, bno, blockcount, dfops);
1117		break;
1118	case XFS_REFCOUNT_FREE_COW:
1119		*new_fsb = startblock + blockcount;
1120		*new_len = 0;
1121		error = __xfs_refcount_cow_free(rcur, bno, blockcount, dfops);
1122		break;
1123	default:
1124		ASSERT(0);
1125		error = -EFSCORRUPTED;
1126	}
1127	if (!error && *new_len > 0)
1128		trace_xfs_refcount_finish_one_leftover(mp, agno, type,
1129				bno, blockcount, new_agbno, *new_len);
1130	return error;
1131
1132out_cur:
1133	xfs_trans_brelse(tp, agbp);
1134
1135	return error;
1136}
1137
1138/*
1139 * Record a refcount intent for later processing.
1140 */
1141static int
1142__xfs_refcount_add(
1143	struct xfs_mount		*mp,
1144	struct xfs_defer_ops		*dfops,
1145	enum xfs_refcount_intent_type	type,
1146	xfs_fsblock_t			startblock,
1147	xfs_extlen_t			blockcount)
1148{
1149	struct xfs_refcount_intent	*ri;
1150
1151	trace_xfs_refcount_defer(mp, XFS_FSB_TO_AGNO(mp, startblock),
1152			type, XFS_FSB_TO_AGBNO(mp, startblock),
1153			blockcount);
1154
1155	ri = kmem_alloc(sizeof(struct xfs_refcount_intent),
1156			KM_SLEEP | KM_NOFS);
1157	INIT_LIST_HEAD(&ri->ri_list);
1158	ri->ri_type = type;
1159	ri->ri_startblock = startblock;
1160	ri->ri_blockcount = blockcount;
1161
1162	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
1163	return 0;
1164}
1165
1166/*
1167 * Increase the reference count of the blocks backing a file's extent.
1168 */
1169int
1170xfs_refcount_increase_extent(
1171	struct xfs_mount		*mp,
1172	struct xfs_defer_ops		*dfops,
1173	struct xfs_bmbt_irec		*PREV)
1174{
1175	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1176		return 0;
1177
1178	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_INCREASE,
1179			PREV->br_startblock, PREV->br_blockcount);
1180}
1181
1182/*
1183 * Decrease the reference count of the blocks backing a file's extent.
1184 */
1185int
1186xfs_refcount_decrease_extent(
1187	struct xfs_mount		*mp,
1188	struct xfs_defer_ops		*dfops,
1189	struct xfs_bmbt_irec		*PREV)
1190{
1191	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1192		return 0;
1193
1194	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_DECREASE,
1195			PREV->br_startblock, PREV->br_blockcount);
1196}
1197
1198/*
1199 * Given an AG extent, find the lowest-numbered run of shared blocks
1200 * within that range and return the range in fbno/flen.  If
1201 * find_end_of_shared is set, return the longest contiguous extent of
1202 * shared blocks; if not, just return the first extent we find.  If no
1203 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1204 * and 0, respectively.
1205 */
1206int
1207xfs_refcount_find_shared(
1208	struct xfs_btree_cur		*cur,
1209	xfs_agblock_t			agbno,
1210	xfs_extlen_t			aglen,
1211	xfs_agblock_t			*fbno,
1212	xfs_extlen_t			*flen,
1213	bool				find_end_of_shared)
1214{
1215	struct xfs_refcount_irec	tmp;
1216	int				i;
1217	int				have;
1218	int				error;
1219
1220	trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_private.a.agno,
1221			agbno, aglen);
1222
1223	/* By default, skip the whole range */
1224	*fbno = NULLAGBLOCK;
1225	*flen = 0;
1226
1227	/* Try to find a refcount extent that crosses the start */
1228	error = xfs_refcount_lookup_le(cur, agbno, &have);
1229	if (error)
1230		goto out_error;
1231	if (!have) {
1232		/* No left extent, look at the next one */
1233		error = xfs_btree_increment(cur, 0, &have);
1234		if (error)
1235			goto out_error;
1236		if (!have)
1237			goto done;
1238	}
1239	error = xfs_refcount_get_rec(cur, &tmp, &i);
1240	if (error)
1241		goto out_error;
1242	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1243
1244	/* If the extent ends before the start, look at the next one */
1245	if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1246		error = xfs_btree_increment(cur, 0, &have);
1247		if (error)
1248			goto out_error;
1249		if (!have)
1250			goto done;
1251		error = xfs_refcount_get_rec(cur, &tmp, &i);
1252		if (error)
1253			goto out_error;
1254		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1255	}
1256
1257	/* If the extent starts after the range we want, bail out */
1258	if (tmp.rc_startblock >= agbno + aglen)
1259		goto done;
1260
1261	/* We found the start of a shared extent! */
1262	if (tmp.rc_startblock < agbno) {
1263		tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1264		tmp.rc_startblock = agbno;
1265	}
1266
1267	*fbno = tmp.rc_startblock;
1268	*flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1269	if (!find_end_of_shared)
1270		goto done;
1271
1272	/* Otherwise, find the end of this shared extent */
1273	while (*fbno + *flen < agbno + aglen) {
1274		error = xfs_btree_increment(cur, 0, &have);
1275		if (error)
1276			goto out_error;
1277		if (!have)
1278			break;
1279		error = xfs_refcount_get_rec(cur, &tmp, &i);
1280		if (error)
1281			goto out_error;
1282		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1283		if (tmp.rc_startblock >= agbno + aglen ||
1284		    tmp.rc_startblock != *fbno + *flen)
1285			break;
1286		*flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1287	}
1288
1289done:
1290	trace_xfs_refcount_find_shared_result(cur->bc_mp,
1291			cur->bc_private.a.agno, *fbno, *flen);
1292
1293out_error:
1294	if (error)
1295		trace_xfs_refcount_find_shared_error(cur->bc_mp,
1296				cur->bc_private.a.agno, error, _RET_IP_);
1297	return error;
1298}
1299
1300/*
1301 * Recovering CoW Blocks After a Crash
1302 *
1303 * Due to the way that the copy on write mechanism works, there's a window of
1304 * opportunity in which we can lose track of allocated blocks during a crash.
1305 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1306 * causes blocks to be allocated and stored in the CoW fork.  The blocks are
1307 * no longer in the free space btree but are not otherwise recorded anywhere
1308 * until the write completes and the blocks are mapped into the file.  A crash
1309 * in between allocation and remapping results in the replacement blocks being
1310 * lost.  This situation is exacerbated by the CoW extent size hint because
1311 * allocations can hang around for long time.
1312 *
1313 * However, there is a place where we can record these allocations before they
1314 * become mappings -- the reference count btree.  The btree does not record
1315 * extents with refcount == 1, so we can record allocations with a refcount of
1316 * 1.  Blocks being used for CoW writeout cannot be shared, so there should be
1317 * no conflict with shared block records.  These mappings should be created
1318 * when we allocate blocks to the CoW fork and deleted when they're removed
1319 * from the CoW fork.
1320 *
1321 * Minor nit: records for in-progress CoW allocations and records for shared
1322 * extents must never be merged, to preserve the property that (except for CoW
1323 * allocations) there are no refcount btree entries with refcount == 1.  The
1324 * only time this could potentially happen is when unsharing a block that's
1325 * adjacent to CoW allocations, so we must be careful to avoid this.
1326 *
1327 * At mount time we recover lost CoW allocations by searching the refcount
1328 * btree for these refcount == 1 mappings.  These represent CoW allocations
1329 * that were in progress at the time the filesystem went down, so we can free
1330 * them to get the space back.
1331 *
1332 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1333 * several reasons -- first, EFIs pin the tail of the log and would have to be
1334 * periodically relogged to avoid filling up the log.  Second, CoW completions
1335 * will have to file an EFD and create new EFIs for whatever remains in the
1336 * CoW fork; this partially takes care of (1) but extent-size reservations
1337 * will have to periodically relog even if there's no writeout in progress.
1338 * This can happen if the CoW extent size hint is set, which you really want.
1339 * Third, EFIs cannot currently be automatically relogged into newer
1340 * transactions to advance the log tail.  Fourth, stuffing the log full of
1341 * EFIs places an upper bound on the number of CoW allocations that can be
1342 * held filesystem-wide at any given time.  Recording them in the refcount
1343 * btree doesn't require us to maintain any state in memory and doesn't pin
1344 * the log.
1345 */
1346/*
1347 * Adjust the refcounts of CoW allocations.  These allocations are "magic"
1348 * in that they're not referenced anywhere else in the filesystem, so we
1349 * stash them in the refcount btree with a refcount of 1 until either file
1350 * remapping (or CoW cancellation) happens.
1351 */
1352STATIC int
1353xfs_refcount_adjust_cow_extents(
1354	struct xfs_btree_cur	*cur,
1355	xfs_agblock_t		agbno,
1356	xfs_extlen_t		aglen,
1357	enum xfs_refc_adjust_op	adj)
1358{
1359	struct xfs_refcount_irec	ext, tmp;
1360	int				error;
1361	int				found_rec, found_tmp;
1362
1363	if (aglen == 0)
1364		return 0;
1365
1366	/* Find any overlapping refcount records */
1367	error = xfs_refcount_lookup_ge(cur, agbno, &found_rec);
1368	if (error)
1369		goto out_error;
1370	error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1371	if (error)
1372		goto out_error;
1373	if (!found_rec) {
1374		ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks +
1375				XFS_REFC_COW_START;
1376		ext.rc_blockcount = 0;
1377		ext.rc_refcount = 0;
1378	}
1379
1380	switch (adj) {
1381	case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1382		/* Adding a CoW reservation, there should be nothing here. */
1383		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1384				ext.rc_startblock >= agbno + aglen, out_error);
1385
1386		tmp.rc_startblock = agbno;
1387		tmp.rc_blockcount = aglen;
1388		tmp.rc_refcount = 1;
1389		trace_xfs_refcount_modify_extent(cur->bc_mp,
1390				cur->bc_private.a.agno, &tmp);
1391
1392		error = xfs_refcount_insert(cur, &tmp,
1393				&found_tmp);
1394		if (error)
1395			goto out_error;
1396		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1397				found_tmp == 1, out_error);
1398		break;
1399	case XFS_REFCOUNT_ADJUST_COW_FREE:
1400		/* Removing a CoW reservation, there should be one extent. */
1401		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1402			ext.rc_startblock == agbno, out_error);
1403		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1404			ext.rc_blockcount == aglen, out_error);
1405		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1406			ext.rc_refcount == 1, out_error);
1407
1408		ext.rc_refcount = 0;
1409		trace_xfs_refcount_modify_extent(cur->bc_mp,
1410				cur->bc_private.a.agno, &ext);
1411		error = xfs_refcount_delete(cur, &found_rec);
1412		if (error)
1413			goto out_error;
1414		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1415				found_rec == 1, out_error);
1416		break;
1417	default:
1418		ASSERT(0);
1419	}
1420
1421	return error;
1422out_error:
1423	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1424			cur->bc_private.a.agno, error, _RET_IP_);
1425	return error;
1426}
1427
1428/*
1429 * Add or remove refcount btree entries for CoW reservations.
1430 */
1431STATIC int
1432xfs_refcount_adjust_cow(
1433	struct xfs_btree_cur	*cur,
1434	xfs_agblock_t		agbno,
1435	xfs_extlen_t		aglen,
1436	enum xfs_refc_adjust_op	adj)
1437{
1438	bool			shape_changed;
1439	int			error;
1440
1441	agbno += XFS_REFC_COW_START;
1442
1443	/*
1444	 * Ensure that no rcextents cross the boundary of the adjustment range.
1445	 */
1446	error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1447	if (error)
1448		goto out_error;
1449
1450	error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1451	if (error)
1452		goto out_error;
1453
1454	/*
1455	 * Try to merge with the left or right extents of the range.
1456	 */
1457	error = xfs_refcount_merge_extents(cur, &agbno, &aglen, adj,
1458			XFS_FIND_RCEXT_COW, &shape_changed);
1459	if (error)
1460		goto out_error;
1461
1462	/* Now that we've taken care of the ends, adjust the middle extents */
1463	error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1464	if (error)
1465		goto out_error;
1466
1467	return 0;
1468
1469out_error:
1470	trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_private.a.agno,
1471			error, _RET_IP_);
1472	return error;
1473}
1474
1475/*
1476 * Record a CoW allocation in the refcount btree.
1477 */
1478STATIC int
1479__xfs_refcount_cow_alloc(
1480	struct xfs_btree_cur	*rcur,
1481	xfs_agblock_t		agbno,
1482	xfs_extlen_t		aglen,
1483	struct xfs_defer_ops	*dfops)
1484{
1485	trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_private.a.agno,
1486			agbno, aglen);
1487
1488	/* Add refcount btree reservation */
1489	return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1490			XFS_REFCOUNT_ADJUST_COW_ALLOC);
1491}
1492
1493/*
1494 * Remove a CoW allocation from the refcount btree.
1495 */
1496STATIC int
1497__xfs_refcount_cow_free(
1498	struct xfs_btree_cur	*rcur,
1499	xfs_agblock_t		agbno,
1500	xfs_extlen_t		aglen,
1501	struct xfs_defer_ops	*dfops)
1502{
1503	trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_private.a.agno,
1504			agbno, aglen);
1505
1506	/* Remove refcount btree reservation */
1507	return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1508			XFS_REFCOUNT_ADJUST_COW_FREE);
1509}
1510
1511/* Record a CoW staging extent in the refcount btree. */
1512int
1513xfs_refcount_alloc_cow_extent(
1514	struct xfs_mount		*mp,
1515	struct xfs_defer_ops		*dfops,
1516	xfs_fsblock_t			fsb,
1517	xfs_extlen_t			len)
1518{
1519	int				error;
1520
1521	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1522		return 0;
1523
1524	error = __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_ALLOC_COW,
1525			fsb, len);
1526	if (error)
1527		return error;
1528
1529	/* Add rmap entry */
1530	return xfs_rmap_alloc_extent(mp, dfops, XFS_FSB_TO_AGNO(mp, fsb),
1531			XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1532}
1533
1534/* Forget a CoW staging event in the refcount btree. */
1535int
1536xfs_refcount_free_cow_extent(
1537	struct xfs_mount		*mp,
1538	struct xfs_defer_ops		*dfops,
1539	xfs_fsblock_t			fsb,
1540	xfs_extlen_t			len)
1541{
1542	int				error;
1543
1544	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1545		return 0;
1546
1547	/* Remove rmap entry */
1548	error = xfs_rmap_free_extent(mp, dfops, XFS_FSB_TO_AGNO(mp, fsb),
1549			XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1550	if (error)
1551		return error;
1552
1553	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_FREE_COW,
1554			fsb, len);
1555}
1556
1557struct xfs_refcount_recovery {
1558	struct list_head		rr_list;
1559	struct xfs_refcount_irec	rr_rrec;
1560};
1561
1562/* Stuff an extent on the recovery list. */
1563STATIC int
1564xfs_refcount_recover_extent(
1565	struct xfs_btree_cur 		*cur,
1566	union xfs_btree_rec		*rec,
1567	void				*priv)
1568{
1569	struct list_head		*debris = priv;
1570	struct xfs_refcount_recovery	*rr;
1571
1572	if (be32_to_cpu(rec->refc.rc_refcount) != 1)
1573		return -EFSCORRUPTED;
1574
1575	rr = kmem_alloc(sizeof(struct xfs_refcount_recovery), KM_SLEEP);
1576	xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1577	list_add_tail(&rr->rr_list, debris);
1578
1579	return 0;
1580}
1581
1582/* Find and remove leftover CoW reservations. */
1583int
1584xfs_refcount_recover_cow_leftovers(
1585	struct xfs_mount		*mp,
1586	xfs_agnumber_t			agno)
1587{
1588	struct xfs_trans		*tp;
1589	struct xfs_btree_cur		*cur;
1590	struct xfs_buf			*agbp;
1591	struct xfs_refcount_recovery	*rr, *n;
1592	struct list_head		debris;
1593	union xfs_btree_irec		low;
1594	union xfs_btree_irec		high;
1595	struct xfs_defer_ops		dfops;
1596	xfs_fsblock_t			fsb;
1597	xfs_agblock_t			agbno;
1598	int				error;
1599
1600	if (mp->m_sb.sb_agblocks >= XFS_REFC_COW_START)
1601		return -EOPNOTSUPP;
1602
1603	INIT_LIST_HEAD(&debris);
1604
1605	/*
1606	 * In this first part, we use an empty transaction to gather up
1607	 * all the leftover CoW extents so that we can subsequently
1608	 * delete them.  The empty transaction is used to avoid
1609	 * a buffer lock deadlock if there happens to be a loop in the
1610	 * refcountbt because we're allowed to re-grab a buffer that is
1611	 * already attached to our transaction.  When we're done
1612	 * recording the CoW debris we cancel the (empty) transaction
1613	 * and everything goes away cleanly.
1614	 */
1615	error = xfs_trans_alloc_empty(mp, &tp);
1616	if (error)
1617		return error;
1618
1619	error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
1620	if (error)
1621		goto out_trans;
1622	if (!agbp) {
1623		error = -ENOMEM;
1624		goto out_trans;
1625	}
1626	cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, NULL);
1627
1628	/* Find all the leftover CoW staging extents. */
1629	memset(&low, 0, sizeof(low));
1630	memset(&high, 0, sizeof(high));
1631	low.rc.rc_startblock = XFS_REFC_COW_START;
1632	high.rc.rc_startblock = -1U;
1633	error = xfs_btree_query_range(cur, &low, &high,
1634			xfs_refcount_recover_extent, &debris);
1635	if (error)
1636		goto out_cursor;
1637	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1638	xfs_trans_brelse(tp, agbp);
1639	xfs_trans_cancel(tp);
1640
1641	/* Now iterate the list to free the leftovers */
1642	list_for_each_entry_safe(rr, n, &debris, rr_list) {
1643		/* Set up transaction. */
1644		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1645		if (error)
1646			goto out_free;
1647
1648		trace_xfs_refcount_recover_extent(mp, agno, &rr->rr_rrec);
1649
1650		/* Free the orphan record */
1651		xfs_defer_init(&dfops, &fsb);
1652		agbno = rr->rr_rrec.rc_startblock - XFS_REFC_COW_START;
1653		fsb = XFS_AGB_TO_FSB(mp, agno, agbno);
1654		error = xfs_refcount_free_cow_extent(mp, &dfops, fsb,
1655				rr->rr_rrec.rc_blockcount);
1656		if (error)
1657			goto out_defer;
1658
1659		/* Free the block. */
1660		xfs_bmap_add_free(mp, &dfops, fsb,
1661				rr->rr_rrec.rc_blockcount, NULL);
1662
1663		error = xfs_defer_finish(&tp, &dfops);
1664		if (error)
1665			goto out_defer;
1666
1667		error = xfs_trans_commit(tp);
1668		if (error)
1669			goto out_free;
1670
1671		list_del(&rr->rr_list);
1672		kmem_free(rr);
1673	}
1674
1675	return error;
1676out_defer:
1677	xfs_defer_cancel(&dfops);
1678out_trans:
1679	xfs_trans_cancel(tp);
1680out_free:
1681	/* Free the leftover list */
1682	list_for_each_entry_safe(rr, n, &debris, rr_list) {
1683		list_del(&rr->rr_list);
1684		kmem_free(rr);
1685	}
1686	return error;
1687
1688out_cursor:
1689	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1690	xfs_trans_brelse(tp, agbp);
1691	goto out_trans;
1692}
1693
1694/* Is there a record covering a given extent? */
1695int
1696xfs_refcount_has_record(
1697	struct xfs_btree_cur	*cur,
1698	xfs_agblock_t		bno,
1699	xfs_extlen_t		len,
1700	bool			*exists)
1701{
1702	union xfs_btree_irec	low;
1703	union xfs_btree_irec	high;
1704
1705	memset(&low, 0, sizeof(low));
1706	low.rc.rc_startblock = bno;
1707	memset(&high, 0xFF, sizeof(high));
1708	high.rc.rc_startblock = bno + len - 1;
1709
1710	return xfs_btree_has_record(cur, &low, &high, exists);
1711}