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