xref: /openbmc/linux/fs/xfs/libxfs/xfs_refcount.c (revision 8730046c)
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_error.h"
34 #include "xfs_trace.h"
35 #include "xfs_cksum.h"
36 #include "xfs_trans.h"
37 #include "xfs_bit.h"
38 #include "xfs_refcount.h"
39 #include "xfs_rmap.h"
40 
41 /* Allowable refcount adjustment amounts. */
42 enum xfs_refc_adjust_op {
43 	XFS_REFCOUNT_ADJUST_INCREASE	= 1,
44 	XFS_REFCOUNT_ADJUST_DECREASE	= -1,
45 	XFS_REFCOUNT_ADJUST_COW_ALLOC	= 0,
46 	XFS_REFCOUNT_ADJUST_COW_FREE	= -1,
47 };
48 
49 STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
50 		xfs_agblock_t agbno, xfs_extlen_t aglen,
51 		struct xfs_defer_ops *dfops);
52 STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
53 		xfs_agblock_t agbno, xfs_extlen_t aglen,
54 		struct xfs_defer_ops *dfops);
55 
56 /*
57  * Look up the first record less than or equal to [bno, len] in the btree
58  * given by cur.
59  */
60 int
61 xfs_refcount_lookup_le(
62 	struct xfs_btree_cur	*cur,
63 	xfs_agblock_t		bno,
64 	int			*stat)
65 {
66 	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
67 			XFS_LOOKUP_LE);
68 	cur->bc_rec.rc.rc_startblock = bno;
69 	cur->bc_rec.rc.rc_blockcount = 0;
70 	return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
71 }
72 
73 /*
74  * Look up the first record greater than or equal to [bno, len] in the btree
75  * given by cur.
76  */
77 int
78 xfs_refcount_lookup_ge(
79 	struct xfs_btree_cur	*cur,
80 	xfs_agblock_t		bno,
81 	int			*stat)
82 {
83 	trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
84 			XFS_LOOKUP_GE);
85 	cur->bc_rec.rc.rc_startblock = bno;
86 	cur->bc_rec.rc.rc_blockcount = 0;
87 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
88 }
89 
90 /* Convert on-disk record to in-core format. */
91 static inline void
92 xfs_refcount_btrec_to_irec(
93 	union xfs_btree_rec		*rec,
94 	struct xfs_refcount_irec	*irec)
95 {
96 	irec->rc_startblock = be32_to_cpu(rec->refc.rc_startblock);
97 	irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
98 	irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
99 }
100 
101 /*
102  * Get the data from the pointed-to record.
103  */
104 int
105 xfs_refcount_get_rec(
106 	struct xfs_btree_cur		*cur,
107 	struct xfs_refcount_irec	*irec,
108 	int				*stat)
109 {
110 	union xfs_btree_rec		*rec;
111 	int				error;
112 
113 	error = xfs_btree_get_rec(cur, &rec, stat);
114 	if (!error && *stat == 1) {
115 		xfs_refcount_btrec_to_irec(rec, irec);
116 		trace_xfs_refcount_get(cur->bc_mp, cur->bc_private.a.agno,
117 				irec);
118 	}
119 	return error;
120 }
121 
122 /*
123  * Update the record referred to by cur to the value given
124  * by [bno, len, refcount].
125  * This either works (return 0) or gets an EFSCORRUPTED error.
126  */
127 STATIC int
128 xfs_refcount_update(
129 	struct xfs_btree_cur		*cur,
130 	struct xfs_refcount_irec	*irec)
131 {
132 	union xfs_btree_rec	rec;
133 	int			error;
134 
135 	trace_xfs_refcount_update(cur->bc_mp, cur->bc_private.a.agno, irec);
136 	rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
137 	rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
138 	rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
139 	error = xfs_btree_update(cur, &rec);
140 	if (error)
141 		trace_xfs_refcount_update_error(cur->bc_mp,
142 				cur->bc_private.a.agno, error, _RET_IP_);
143 	return error;
144 }
145 
146 /*
147  * Insert the record referred to by cur to the value given
148  * by [bno, len, refcount].
149  * This either works (return 0) or gets an EFSCORRUPTED error.
150  */
151 STATIC int
152 xfs_refcount_insert(
153 	struct xfs_btree_cur		*cur,
154 	struct xfs_refcount_irec	*irec,
155 	int				*i)
156 {
157 	int				error;
158 
159 	trace_xfs_refcount_insert(cur->bc_mp, cur->bc_private.a.agno, irec);
160 	cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
161 	cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
162 	cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
163 	error = xfs_btree_insert(cur, i);
164 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
165 out_error:
166 	if (error)
167 		trace_xfs_refcount_insert_error(cur->bc_mp,
168 				cur->bc_private.a.agno, error, _RET_IP_);
169 	return error;
170 }
171 
172 /*
173  * Remove the record referred to by cur, then set the pointer to the spot
174  * where the record could be re-inserted, in case we want to increment or
175  * decrement the cursor.
176  * This either works (return 0) or gets an EFSCORRUPTED error.
177  */
178 STATIC int
179 xfs_refcount_delete(
180 	struct xfs_btree_cur	*cur,
181 	int			*i)
182 {
183 	struct xfs_refcount_irec	irec;
184 	int			found_rec;
185 	int			error;
186 
187 	error = xfs_refcount_get_rec(cur, &irec, &found_rec);
188 	if (error)
189 		goto out_error;
190 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
191 	trace_xfs_refcount_delete(cur->bc_mp, cur->bc_private.a.agno, &irec);
192 	error = xfs_btree_delete(cur, i);
193 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
194 	if (error)
195 		goto out_error;
196 	error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
197 out_error:
198 	if (error)
199 		trace_xfs_refcount_delete_error(cur->bc_mp,
200 				cur->bc_private.a.agno, error, _RET_IP_);
201 	return error;
202 }
203 
204 /*
205  * Adjusting the Reference Count
206  *
207  * As stated elsewhere, the reference count btree (refcbt) stores
208  * >1 reference counts for extents of physical blocks.  In this
209  * operation, we're either raising or lowering the reference count of
210  * some subrange stored in the tree:
211  *
212  *      <------ adjustment range ------>
213  * ----+   +---+-----+ +--+--------+---------
214  *  2  |   | 3 |  4  | |17|   55   |   10
215  * ----+   +---+-----+ +--+--------+---------
216  * X axis is physical blocks number;
217  * reference counts are the numbers inside the rectangles
218  *
219  * The first thing we need to do is to ensure that there are no
220  * refcount extents crossing either boundary of the range to be
221  * adjusted.  For any extent that does cross a boundary, split it into
222  * two extents so that we can increment the refcount of one of the
223  * pieces later:
224  *
225  *      <------ adjustment range ------>
226  * ----+   +---+-----+ +--+--------+----+----
227  *  2  |   | 3 |  2  | |17|   55   | 10 | 10
228  * ----+   +---+-----+ +--+--------+----+----
229  *
230  * For this next step, let's assume that all the physical blocks in
231  * the adjustment range are mapped to a file and are therefore in use
232  * at least once.  Therefore, we can infer that any gap in the
233  * refcount tree within the adjustment range represents a physical
234  * extent with refcount == 1:
235  *
236  *      <------ adjustment range ------>
237  * ----+---+---+-----+-+--+--------+----+----
238  *  2  |"1"| 3 |  2  |1|17|   55   | 10 | 10
239  * ----+---+---+-----+-+--+--------+----+----
240  *      ^
241  *
242  * For each extent that falls within the interval range, figure out
243  * which extent is to the left or the right of that extent.  Now we
244  * have a left, current, and right extent.  If the new reference count
245  * of the center extent enables us to merge left, center, and right
246  * into one record covering all three, do so.  If the center extent is
247  * at the left end of the range, abuts the left extent, and its new
248  * reference count matches the left extent's record, then merge them.
249  * If the center extent is at the right end of the range, abuts the
250  * right extent, and the reference counts match, merge those.  In the
251  * example, we can left merge (assuming an increment operation):
252  *
253  *      <------ adjustment range ------>
254  * --------+---+-----+-+--+--------+----+----
255  *    2    | 3 |  2  |1|17|   55   | 10 | 10
256  * --------+---+-----+-+--+--------+----+----
257  *          ^
258  *
259  * For all other extents within the range, adjust the reference count
260  * or delete it if the refcount falls below 2.  If we were
261  * incrementing, the end result looks like this:
262  *
263  *      <------ adjustment range ------>
264  * --------+---+-----+-+--+--------+----+----
265  *    2    | 4 |  3  |2|18|   56   | 11 | 10
266  * --------+---+-----+-+--+--------+----+----
267  *
268  * The result of a decrement operation looks as such:
269  *
270  *      <------ adjustment range ------>
271  * ----+   +---+       +--+--------+----+----
272  *  2  |   | 2 |       |16|   54   |  9 | 10
273  * ----+   +---+       +--+--------+----+----
274  *      DDDD    111111DD
275  *
276  * The blocks marked "D" are freed; the blocks marked "1" are only
277  * referenced once and therefore the record is removed from the
278  * refcount btree.
279  */
280 
281 /* Next block after this extent. */
282 static inline xfs_agblock_t
283 xfs_refc_next(
284 	struct xfs_refcount_irec	*rc)
285 {
286 	return rc->rc_startblock + rc->rc_blockcount;
287 }
288 
289 /*
290  * Split a refcount extent that crosses agbno.
291  */
292 STATIC int
293 xfs_refcount_split_extent(
294 	struct xfs_btree_cur		*cur,
295 	xfs_agblock_t			agbno,
296 	bool				*shape_changed)
297 {
298 	struct xfs_refcount_irec	rcext, tmp;
299 	int				found_rec;
300 	int				error;
301 
302 	*shape_changed = false;
303 	error = xfs_refcount_lookup_le(cur, agbno, &found_rec);
304 	if (error)
305 		goto out_error;
306 	if (!found_rec)
307 		return 0;
308 
309 	error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
310 	if (error)
311 		goto out_error;
312 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
313 	if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
314 		return 0;
315 
316 	*shape_changed = true;
317 	trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_private.a.agno,
318 			&rcext, agbno);
319 
320 	/* Establish the right extent. */
321 	tmp = rcext;
322 	tmp.rc_startblock = agbno;
323 	tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
324 	error = xfs_refcount_update(cur, &tmp);
325 	if (error)
326 		goto out_error;
327 
328 	/* Insert the left extent. */
329 	tmp = rcext;
330 	tmp.rc_blockcount = agbno - rcext.rc_startblock;
331 	error = xfs_refcount_insert(cur, &tmp, &found_rec);
332 	if (error)
333 		goto out_error;
334 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
335 	return error;
336 
337 out_error:
338 	trace_xfs_refcount_split_extent_error(cur->bc_mp,
339 			cur->bc_private.a.agno, error, _RET_IP_);
340 	return error;
341 }
342 
343 /*
344  * Merge the left, center, and right extents.
345  */
346 STATIC int
347 xfs_refcount_merge_center_extents(
348 	struct xfs_btree_cur		*cur,
349 	struct xfs_refcount_irec	*left,
350 	struct xfs_refcount_irec	*center,
351 	struct xfs_refcount_irec	*right,
352 	unsigned long long		extlen,
353 	xfs_agblock_t			*agbno,
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 
404 out_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  */
413 STATIC int
414 xfs_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 
459 out_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  */
468 STATIC int
469 xfs_refcount_merge_right_extent(
470 	struct xfs_btree_cur		*cur,
471 	struct xfs_refcount_irec	*right,
472 	struct xfs_refcount_irec	*cright,
473 	xfs_agblock_t			*agbno,
474 	xfs_extlen_t			*aglen)
475 {
476 	int				error;
477 	int				found_rec;
478 
479 	trace_xfs_refcount_merge_right_extent(cur->bc_mp,
480 			cur->bc_private.a.agno, cright, right);
481 
482 	/*
483 	 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
484 	 * remove it.
485 	 */
486 	if (cright->rc_refcount > 1) {
487 		error = xfs_refcount_lookup_le(cur, cright->rc_startblock,
488 			&found_rec);
489 		if (error)
490 			goto out_error;
491 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
492 				out_error);
493 
494 		error = xfs_refcount_delete(cur, &found_rec);
495 		if (error)
496 			goto out_error;
497 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
498 				out_error);
499 	}
500 
501 	/* Enlarge the right extent. */
502 	error = xfs_refcount_lookup_le(cur, right->rc_startblock,
503 			&found_rec);
504 	if (error)
505 		goto out_error;
506 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
507 
508 	right->rc_startblock -= cright->rc_blockcount;
509 	right->rc_blockcount += cright->rc_blockcount;
510 	error = xfs_refcount_update(cur, right);
511 	if (error)
512 		goto out_error;
513 
514 	*aglen -= cright->rc_blockcount;
515 	return error;
516 
517 out_error:
518 	trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
519 			cur->bc_private.a.agno, error, _RET_IP_);
520 	return error;
521 }
522 
523 #define XFS_FIND_RCEXT_SHARED	1
524 #define XFS_FIND_RCEXT_COW	2
525 /*
526  * Find the left extent and the one after it (cleft).  This function assumes
527  * that we've already split any extent crossing agbno.
528  */
529 STATIC int
530 xfs_refcount_find_left_extents(
531 	struct xfs_btree_cur		*cur,
532 	struct xfs_refcount_irec	*left,
533 	struct xfs_refcount_irec	*cleft,
534 	xfs_agblock_t			agbno,
535 	xfs_extlen_t			aglen,
536 	int				flags)
537 {
538 	struct xfs_refcount_irec	tmp;
539 	int				error;
540 	int				found_rec;
541 
542 	left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
543 	error = xfs_refcount_lookup_le(cur, agbno - 1, &found_rec);
544 	if (error)
545 		goto out_error;
546 	if (!found_rec)
547 		return 0;
548 
549 	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
550 	if (error)
551 		goto out_error;
552 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
553 
554 	if (xfs_refc_next(&tmp) != agbno)
555 		return 0;
556 	if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
557 		return 0;
558 	if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
559 		return 0;
560 	/* We have a left extent; retrieve (or invent) the next right one */
561 	*left = tmp;
562 
563 	error = xfs_btree_increment(cur, 0, &found_rec);
564 	if (error)
565 		goto out_error;
566 	if (found_rec) {
567 		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
568 		if (error)
569 			goto out_error;
570 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
571 				out_error);
572 
573 		/* if tmp starts at the end of our range, just use that */
574 		if (tmp.rc_startblock == agbno)
575 			*cleft = tmp;
576 		else {
577 			/*
578 			 * There's a gap in the refcntbt at the start of the
579 			 * range we're interested in (refcount == 1) so
580 			 * synthesize the implied extent and pass it back.
581 			 * We assume here that the agbno/aglen range was
582 			 * passed in from a data fork extent mapping and
583 			 * therefore is allocated to exactly one owner.
584 			 */
585 			cleft->rc_startblock = agbno;
586 			cleft->rc_blockcount = min(aglen,
587 					tmp.rc_startblock - agbno);
588 			cleft->rc_refcount = 1;
589 		}
590 	} else {
591 		/*
592 		 * No extents, so pretend that there's one covering the whole
593 		 * range.
594 		 */
595 		cleft->rc_startblock = agbno;
596 		cleft->rc_blockcount = aglen;
597 		cleft->rc_refcount = 1;
598 	}
599 	trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_private.a.agno,
600 			left, cleft, agbno);
601 	return error;
602 
603 out_error:
604 	trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
605 			cur->bc_private.a.agno, error, _RET_IP_);
606 	return error;
607 }
608 
609 /*
610  * Find the right extent and the one before it (cright).  This function
611  * assumes that we've already split any extents crossing agbno + aglen.
612  */
613 STATIC int
614 xfs_refcount_find_right_extents(
615 	struct xfs_btree_cur		*cur,
616 	struct xfs_refcount_irec	*right,
617 	struct xfs_refcount_irec	*cright,
618 	xfs_agblock_t			agbno,
619 	xfs_extlen_t			aglen,
620 	int				flags)
621 {
622 	struct xfs_refcount_irec	tmp;
623 	int				error;
624 	int				found_rec;
625 
626 	right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
627 	error = xfs_refcount_lookup_ge(cur, agbno + aglen, &found_rec);
628 	if (error)
629 		goto out_error;
630 	if (!found_rec)
631 		return 0;
632 
633 	error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
634 	if (error)
635 		goto out_error;
636 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
637 
638 	if (tmp.rc_startblock != agbno + aglen)
639 		return 0;
640 	if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
641 		return 0;
642 	if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
643 		return 0;
644 	/* We have a right extent; retrieve (or invent) the next left one */
645 	*right = tmp;
646 
647 	error = xfs_btree_decrement(cur, 0, &found_rec);
648 	if (error)
649 		goto out_error;
650 	if (found_rec) {
651 		error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
652 		if (error)
653 			goto out_error;
654 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
655 				out_error);
656 
657 		/* if tmp ends at the end of our range, just use that */
658 		if (xfs_refc_next(&tmp) == agbno + aglen)
659 			*cright = tmp;
660 		else {
661 			/*
662 			 * There's a gap in the refcntbt at the end of the
663 			 * range we're interested in (refcount == 1) so
664 			 * create the implied extent and pass it back.
665 			 * We assume here that the agbno/aglen range was
666 			 * passed in from a data fork extent mapping and
667 			 * therefore is allocated to exactly one owner.
668 			 */
669 			cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
670 			cright->rc_blockcount = right->rc_startblock -
671 					cright->rc_startblock;
672 			cright->rc_refcount = 1;
673 		}
674 	} else {
675 		/*
676 		 * No extents, so pretend that there's one covering the whole
677 		 * range.
678 		 */
679 		cright->rc_startblock = agbno;
680 		cright->rc_blockcount = aglen;
681 		cright->rc_refcount = 1;
682 	}
683 	trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_private.a.agno,
684 			cright, right, agbno + aglen);
685 	return error;
686 
687 out_error:
688 	trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
689 			cur->bc_private.a.agno, error, _RET_IP_);
690 	return error;
691 }
692 
693 /* Is this extent valid? */
694 static inline bool
695 xfs_refc_valid(
696 	struct xfs_refcount_irec	*rc)
697 {
698 	return rc->rc_startblock != NULLAGBLOCK;
699 }
700 
701 /*
702  * Try to merge with any extents on the boundaries of the adjustment range.
703  */
704 STATIC int
705 xfs_refcount_merge_extents(
706 	struct xfs_btree_cur	*cur,
707 	xfs_agblock_t		*agbno,
708 	xfs_extlen_t		*aglen,
709 	enum xfs_refc_adjust_op adjust,
710 	int			flags,
711 	bool			*shape_changed)
712 {
713 	struct xfs_refcount_irec	left = {0}, cleft = {0};
714 	struct xfs_refcount_irec	cright = {0}, right = {0};
715 	int				error;
716 	unsigned long long		ulen;
717 	bool				cequal;
718 
719 	*shape_changed = false;
720 	/*
721 	 * Find the extent just below agbno [left], just above agbno [cleft],
722 	 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
723 	 * [right].
724 	 */
725 	error = xfs_refcount_find_left_extents(cur, &left, &cleft, *agbno,
726 			*aglen, flags);
727 	if (error)
728 		return error;
729 	error = xfs_refcount_find_right_extents(cur, &right, &cright, *agbno,
730 			*aglen, flags);
731 	if (error)
732 		return error;
733 
734 	/* No left or right extent to merge; exit. */
735 	if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
736 		return 0;
737 
738 	cequal = (cleft.rc_startblock == cright.rc_startblock) &&
739 		 (cleft.rc_blockcount == cright.rc_blockcount);
740 
741 	/* Try to merge left, cleft, and right.  cleft must == cright. */
742 	ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount +
743 			right.rc_blockcount;
744 	if (xfs_refc_valid(&left) && xfs_refc_valid(&right) &&
745 	    xfs_refc_valid(&cleft) && xfs_refc_valid(&cright) && cequal &&
746 	    left.rc_refcount == cleft.rc_refcount + adjust &&
747 	    right.rc_refcount == cleft.rc_refcount + adjust &&
748 	    ulen < MAXREFCEXTLEN) {
749 		*shape_changed = true;
750 		return xfs_refcount_merge_center_extents(cur, &left, &cleft,
751 				&right, ulen, agbno, aglen);
752 	}
753 
754 	/* Try to merge left and cleft. */
755 	ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount;
756 	if (xfs_refc_valid(&left) && xfs_refc_valid(&cleft) &&
757 	    left.rc_refcount == cleft.rc_refcount + adjust &&
758 	    ulen < MAXREFCEXTLEN) {
759 		*shape_changed = true;
760 		error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
761 				agbno, aglen);
762 		if (error)
763 			return error;
764 
765 		/*
766 		 * If we just merged left + cleft and cleft == cright,
767 		 * we no longer have a cright to merge with right.  We're done.
768 		 */
769 		if (cequal)
770 			return 0;
771 	}
772 
773 	/* Try to merge cright and right. */
774 	ulen = (unsigned long long)right.rc_blockcount + cright.rc_blockcount;
775 	if (xfs_refc_valid(&right) && xfs_refc_valid(&cright) &&
776 	    right.rc_refcount == cright.rc_refcount + adjust &&
777 	    ulen < MAXREFCEXTLEN) {
778 		*shape_changed = true;
779 		return xfs_refcount_merge_right_extent(cur, &right, &cright,
780 				agbno, aglen);
781 	}
782 
783 	return error;
784 }
785 
786 /*
787  * While we're adjusting the refcounts records of an extent, we have
788  * to keep an eye on the number of extents we're dirtying -- run too
789  * many in a single transaction and we'll exceed the transaction's
790  * reservation and crash the fs.  Each record adds 12 bytes to the
791  * log (plus any key updates) so we'll conservatively assume 24 bytes
792  * per record.  We must also leave space for btree splits on both ends
793  * of the range and space for the CUD and a new CUI.
794  *
795  * XXX: This is a pretty hand-wavy estimate.  The penalty for guessing
796  * true incorrectly is a shutdown FS; the penalty for guessing false
797  * incorrectly is more transaction rolls than might be necessary.
798  * Be conservative here.
799  */
800 static bool
801 xfs_refcount_still_have_space(
802 	struct xfs_btree_cur		*cur)
803 {
804 	unsigned long			overhead;
805 
806 	overhead = cur->bc_private.a.priv.refc.shape_changes *
807 			xfs_allocfree_log_count(cur->bc_mp, 1);
808 	overhead *= cur->bc_mp->m_sb.sb_blocksize;
809 
810 	/*
811 	 * Only allow 2 refcount extent updates per transaction if the
812 	 * refcount continue update "error" has been injected.
813 	 */
814 	if (cur->bc_private.a.priv.refc.nr_ops > 2 &&
815 	    XFS_TEST_ERROR(false, cur->bc_mp,
816 			XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE,
817 			XFS_RANDOM_REFCOUNT_CONTINUE_UPDATE))
818 		return false;
819 
820 	if (cur->bc_private.a.priv.refc.nr_ops == 0)
821 		return true;
822 	else if (overhead > cur->bc_tp->t_log_res)
823 		return false;
824 	return  cur->bc_tp->t_log_res - overhead >
825 		cur->bc_private.a.priv.refc.nr_ops * 32;
826 }
827 
828 /*
829  * Adjust the refcounts of middle extents.  At this point we should have
830  * split extents that crossed the adjustment range; merged with adjacent
831  * extents; and updated agbno/aglen to reflect the merges.  Therefore,
832  * all we have to do is update the extents inside [agbno, agbno + aglen].
833  */
834 STATIC int
835 xfs_refcount_adjust_extents(
836 	struct xfs_btree_cur	*cur,
837 	xfs_agblock_t		*agbno,
838 	xfs_extlen_t		*aglen,
839 	enum xfs_refc_adjust_op	adj,
840 	struct xfs_defer_ops	*dfops,
841 	struct xfs_owner_info	*oinfo)
842 {
843 	struct xfs_refcount_irec	ext, tmp;
844 	int				error;
845 	int				found_rec, found_tmp;
846 	xfs_fsblock_t			fsbno;
847 
848 	/* Merging did all the work already. */
849 	if (*aglen == 0)
850 		return 0;
851 
852 	error = xfs_refcount_lookup_ge(cur, *agbno, &found_rec);
853 	if (error)
854 		goto out_error;
855 
856 	while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
857 		error = xfs_refcount_get_rec(cur, &ext, &found_rec);
858 		if (error)
859 			goto out_error;
860 		if (!found_rec) {
861 			ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
862 			ext.rc_blockcount = 0;
863 			ext.rc_refcount = 0;
864 		}
865 
866 		/*
867 		 * Deal with a hole in the refcount tree; if a file maps to
868 		 * these blocks and there's no refcountbt record, pretend that
869 		 * there is one with refcount == 1.
870 		 */
871 		if (ext.rc_startblock != *agbno) {
872 			tmp.rc_startblock = *agbno;
873 			tmp.rc_blockcount = min(*aglen,
874 					ext.rc_startblock - *agbno);
875 			tmp.rc_refcount = 1 + adj;
876 			trace_xfs_refcount_modify_extent(cur->bc_mp,
877 					cur->bc_private.a.agno, &tmp);
878 
879 			/*
880 			 * Either cover the hole (increment) or
881 			 * delete the range (decrement).
882 			 */
883 			if (tmp.rc_refcount) {
884 				error = xfs_refcount_insert(cur, &tmp,
885 						&found_tmp);
886 				if (error)
887 					goto out_error;
888 				XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
889 						found_tmp == 1, out_error);
890 				cur->bc_private.a.priv.refc.nr_ops++;
891 			} else {
892 				fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
893 						cur->bc_private.a.agno,
894 						tmp.rc_startblock);
895 				xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
896 						tmp.rc_blockcount, oinfo);
897 			}
898 
899 			(*agbno) += tmp.rc_blockcount;
900 			(*aglen) -= tmp.rc_blockcount;
901 
902 			error = xfs_refcount_lookup_ge(cur, *agbno,
903 					&found_rec);
904 			if (error)
905 				goto out_error;
906 		}
907 
908 		/* Stop if there's nothing left to modify */
909 		if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
910 			break;
911 
912 		/*
913 		 * Adjust the reference count and either update the tree
914 		 * (incr) or free the blocks (decr).
915 		 */
916 		if (ext.rc_refcount == MAXREFCOUNT)
917 			goto skip;
918 		ext.rc_refcount += adj;
919 		trace_xfs_refcount_modify_extent(cur->bc_mp,
920 				cur->bc_private.a.agno, &ext);
921 		if (ext.rc_refcount > 1) {
922 			error = xfs_refcount_update(cur, &ext);
923 			if (error)
924 				goto out_error;
925 			cur->bc_private.a.priv.refc.nr_ops++;
926 		} else if (ext.rc_refcount == 1) {
927 			error = xfs_refcount_delete(cur, &found_rec);
928 			if (error)
929 				goto out_error;
930 			XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
931 					found_rec == 1, out_error);
932 			cur->bc_private.a.priv.refc.nr_ops++;
933 			goto advloop;
934 		} else {
935 			fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
936 					cur->bc_private.a.agno,
937 					ext.rc_startblock);
938 			xfs_bmap_add_free(cur->bc_mp, dfops, fsbno,
939 					ext.rc_blockcount, oinfo);
940 		}
941 
942 skip:
943 		error = xfs_btree_increment(cur, 0, &found_rec);
944 		if (error)
945 			goto out_error;
946 
947 advloop:
948 		(*agbno) += ext.rc_blockcount;
949 		(*aglen) -= ext.rc_blockcount;
950 	}
951 
952 	return error;
953 out_error:
954 	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
955 			cur->bc_private.a.agno, error, _RET_IP_);
956 	return error;
957 }
958 
959 /* Adjust the reference count of a range of AG blocks. */
960 STATIC int
961 xfs_refcount_adjust(
962 	struct xfs_btree_cur	*cur,
963 	xfs_agblock_t		agbno,
964 	xfs_extlen_t		aglen,
965 	xfs_agblock_t		*new_agbno,
966 	xfs_extlen_t		*new_aglen,
967 	enum xfs_refc_adjust_op	adj,
968 	struct xfs_defer_ops	*dfops,
969 	struct xfs_owner_info	*oinfo)
970 {
971 	bool			shape_changed;
972 	int			shape_changes = 0;
973 	int			error;
974 
975 	*new_agbno = agbno;
976 	*new_aglen = aglen;
977 	if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
978 		trace_xfs_refcount_increase(cur->bc_mp, cur->bc_private.a.agno,
979 				agbno, aglen);
980 	else
981 		trace_xfs_refcount_decrease(cur->bc_mp, cur->bc_private.a.agno,
982 				agbno, aglen);
983 
984 	/*
985 	 * Ensure that no rcextents cross the boundary of the adjustment range.
986 	 */
987 	error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
988 	if (error)
989 		goto out_error;
990 	if (shape_changed)
991 		shape_changes++;
992 
993 	error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
994 	if (error)
995 		goto out_error;
996 	if (shape_changed)
997 		shape_changes++;
998 
999 	/*
1000 	 * Try to merge with the left or right extents of the range.
1001 	 */
1002 	error = xfs_refcount_merge_extents(cur, new_agbno, new_aglen, adj,
1003 			XFS_FIND_RCEXT_SHARED, &shape_changed);
1004 	if (error)
1005 		goto out_error;
1006 	if (shape_changed)
1007 		shape_changes++;
1008 	if (shape_changes)
1009 		cur->bc_private.a.priv.refc.shape_changes++;
1010 
1011 	/* Now that we've taken care of the ends, adjust the middle extents */
1012 	error = xfs_refcount_adjust_extents(cur, new_agbno, new_aglen,
1013 			adj, dfops, oinfo);
1014 	if (error)
1015 		goto out_error;
1016 
1017 	return 0;
1018 
1019 out_error:
1020 	trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_private.a.agno,
1021 			error, _RET_IP_);
1022 	return error;
1023 }
1024 
1025 /* Clean up after calling xfs_refcount_finish_one. */
1026 void
1027 xfs_refcount_finish_one_cleanup(
1028 	struct xfs_trans	*tp,
1029 	struct xfs_btree_cur	*rcur,
1030 	int			error)
1031 {
1032 	struct xfs_buf		*agbp;
1033 
1034 	if (rcur == NULL)
1035 		return;
1036 	agbp = rcur->bc_private.a.agbp;
1037 	xfs_btree_del_cursor(rcur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1038 	if (error)
1039 		xfs_trans_brelse(tp, agbp);
1040 }
1041 
1042 /*
1043  * Process one of the deferred refcount operations.  We pass back the
1044  * btree cursor to maintain our lock on the btree between calls.
1045  * This saves time and eliminates a buffer deadlock between the
1046  * superblock and the AGF because we'll always grab them in the same
1047  * order.
1048  */
1049 int
1050 xfs_refcount_finish_one(
1051 	struct xfs_trans		*tp,
1052 	struct xfs_defer_ops		*dfops,
1053 	enum xfs_refcount_intent_type	type,
1054 	xfs_fsblock_t			startblock,
1055 	xfs_extlen_t			blockcount,
1056 	xfs_fsblock_t			*new_fsb,
1057 	xfs_extlen_t			*new_len,
1058 	struct xfs_btree_cur		**pcur)
1059 {
1060 	struct xfs_mount		*mp = tp->t_mountp;
1061 	struct xfs_btree_cur		*rcur;
1062 	struct xfs_buf			*agbp = NULL;
1063 	int				error = 0;
1064 	xfs_agnumber_t			agno;
1065 	xfs_agblock_t			bno;
1066 	xfs_agblock_t			new_agbno;
1067 	unsigned long			nr_ops = 0;
1068 	int				shape_changes = 0;
1069 
1070 	agno = XFS_FSB_TO_AGNO(mp, startblock);
1071 	ASSERT(agno != NULLAGNUMBER);
1072 	bno = XFS_FSB_TO_AGBNO(mp, startblock);
1073 
1074 	trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, startblock),
1075 			type, XFS_FSB_TO_AGBNO(mp, startblock),
1076 			blockcount);
1077 
1078 	if (XFS_TEST_ERROR(false, mp,
1079 			XFS_ERRTAG_REFCOUNT_FINISH_ONE,
1080 			XFS_RANDOM_REFCOUNT_FINISH_ONE))
1081 		return -EIO;
1082 
1083 	/*
1084 	 * If we haven't gotten a cursor or the cursor AG doesn't match
1085 	 * the startblock, get one now.
1086 	 */
1087 	rcur = *pcur;
1088 	if (rcur != NULL && rcur->bc_private.a.agno != agno) {
1089 		nr_ops = rcur->bc_private.a.priv.refc.nr_ops;
1090 		shape_changes = rcur->bc_private.a.priv.refc.shape_changes;
1091 		xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1092 		rcur = NULL;
1093 		*pcur = NULL;
1094 	}
1095 	if (rcur == NULL) {
1096 		error = xfs_alloc_read_agf(tp->t_mountp, tp, agno,
1097 				XFS_ALLOC_FLAG_FREEING, &agbp);
1098 		if (error)
1099 			return error;
1100 		if (!agbp)
1101 			return -EFSCORRUPTED;
1102 
1103 		rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno, dfops);
1104 		if (!rcur) {
1105 			error = -ENOMEM;
1106 			goto out_cur;
1107 		}
1108 		rcur->bc_private.a.priv.refc.nr_ops = nr_ops;
1109 		rcur->bc_private.a.priv.refc.shape_changes = shape_changes;
1110 	}
1111 	*pcur = rcur;
1112 
1113 	switch (type) {
1114 	case XFS_REFCOUNT_INCREASE:
1115 		error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1116 			new_len, XFS_REFCOUNT_ADJUST_INCREASE, dfops, NULL);
1117 		*new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1118 		break;
1119 	case XFS_REFCOUNT_DECREASE:
1120 		error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1121 			new_len, XFS_REFCOUNT_ADJUST_DECREASE, dfops, NULL);
1122 		*new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1123 		break;
1124 	case XFS_REFCOUNT_ALLOC_COW:
1125 		*new_fsb = startblock + blockcount;
1126 		*new_len = 0;
1127 		error = __xfs_refcount_cow_alloc(rcur, bno, blockcount, dfops);
1128 		break;
1129 	case XFS_REFCOUNT_FREE_COW:
1130 		*new_fsb = startblock + blockcount;
1131 		*new_len = 0;
1132 		error = __xfs_refcount_cow_free(rcur, bno, blockcount, dfops);
1133 		break;
1134 	default:
1135 		ASSERT(0);
1136 		error = -EFSCORRUPTED;
1137 	}
1138 	if (!error && *new_len > 0)
1139 		trace_xfs_refcount_finish_one_leftover(mp, agno, type,
1140 				bno, blockcount, new_agbno, *new_len);
1141 	return error;
1142 
1143 out_cur:
1144 	xfs_trans_brelse(tp, agbp);
1145 
1146 	return error;
1147 }
1148 
1149 /*
1150  * Record a refcount intent for later processing.
1151  */
1152 static int
1153 __xfs_refcount_add(
1154 	struct xfs_mount		*mp,
1155 	struct xfs_defer_ops		*dfops,
1156 	enum xfs_refcount_intent_type	type,
1157 	xfs_fsblock_t			startblock,
1158 	xfs_extlen_t			blockcount)
1159 {
1160 	struct xfs_refcount_intent	*ri;
1161 
1162 	trace_xfs_refcount_defer(mp, XFS_FSB_TO_AGNO(mp, startblock),
1163 			type, XFS_FSB_TO_AGBNO(mp, startblock),
1164 			blockcount);
1165 
1166 	ri = kmem_alloc(sizeof(struct xfs_refcount_intent),
1167 			KM_SLEEP | KM_NOFS);
1168 	INIT_LIST_HEAD(&ri->ri_list);
1169 	ri->ri_type = type;
1170 	ri->ri_startblock = startblock;
1171 	ri->ri_blockcount = blockcount;
1172 
1173 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
1174 	return 0;
1175 }
1176 
1177 /*
1178  * Increase the reference count of the blocks backing a file's extent.
1179  */
1180 int
1181 xfs_refcount_increase_extent(
1182 	struct xfs_mount		*mp,
1183 	struct xfs_defer_ops		*dfops,
1184 	struct xfs_bmbt_irec		*PREV)
1185 {
1186 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1187 		return 0;
1188 
1189 	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_INCREASE,
1190 			PREV->br_startblock, PREV->br_blockcount);
1191 }
1192 
1193 /*
1194  * Decrease the reference count of the blocks backing a file's extent.
1195  */
1196 int
1197 xfs_refcount_decrease_extent(
1198 	struct xfs_mount		*mp,
1199 	struct xfs_defer_ops		*dfops,
1200 	struct xfs_bmbt_irec		*PREV)
1201 {
1202 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1203 		return 0;
1204 
1205 	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_DECREASE,
1206 			PREV->br_startblock, PREV->br_blockcount);
1207 }
1208 
1209 /*
1210  * Given an AG extent, find the lowest-numbered run of shared blocks
1211  * within that range and return the range in fbno/flen.  If
1212  * find_end_of_shared is set, return the longest contiguous extent of
1213  * shared blocks; if not, just return the first extent we find.  If no
1214  * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1215  * and 0, respectively.
1216  */
1217 int
1218 xfs_refcount_find_shared(
1219 	struct xfs_btree_cur		*cur,
1220 	xfs_agblock_t			agbno,
1221 	xfs_extlen_t			aglen,
1222 	xfs_agblock_t			*fbno,
1223 	xfs_extlen_t			*flen,
1224 	bool				find_end_of_shared)
1225 {
1226 	struct xfs_refcount_irec	tmp;
1227 	int				i;
1228 	int				have;
1229 	int				error;
1230 
1231 	trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_private.a.agno,
1232 			agbno, aglen);
1233 
1234 	/* By default, skip the whole range */
1235 	*fbno = NULLAGBLOCK;
1236 	*flen = 0;
1237 
1238 	/* Try to find a refcount extent that crosses the start */
1239 	error = xfs_refcount_lookup_le(cur, agbno, &have);
1240 	if (error)
1241 		goto out_error;
1242 	if (!have) {
1243 		/* No left extent, look at the next one */
1244 		error = xfs_btree_increment(cur, 0, &have);
1245 		if (error)
1246 			goto out_error;
1247 		if (!have)
1248 			goto done;
1249 	}
1250 	error = xfs_refcount_get_rec(cur, &tmp, &i);
1251 	if (error)
1252 		goto out_error;
1253 	XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1254 
1255 	/* If the extent ends before the start, look at the next one */
1256 	if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1257 		error = xfs_btree_increment(cur, 0, &have);
1258 		if (error)
1259 			goto out_error;
1260 		if (!have)
1261 			goto done;
1262 		error = xfs_refcount_get_rec(cur, &tmp, &i);
1263 		if (error)
1264 			goto out_error;
1265 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1266 	}
1267 
1268 	/* If the extent starts after the range we want, bail out */
1269 	if (tmp.rc_startblock >= agbno + aglen)
1270 		goto done;
1271 
1272 	/* We found the start of a shared extent! */
1273 	if (tmp.rc_startblock < agbno) {
1274 		tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1275 		tmp.rc_startblock = agbno;
1276 	}
1277 
1278 	*fbno = tmp.rc_startblock;
1279 	*flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1280 	if (!find_end_of_shared)
1281 		goto done;
1282 
1283 	/* Otherwise, find the end of this shared extent */
1284 	while (*fbno + *flen < agbno + aglen) {
1285 		error = xfs_btree_increment(cur, 0, &have);
1286 		if (error)
1287 			goto out_error;
1288 		if (!have)
1289 			break;
1290 		error = xfs_refcount_get_rec(cur, &tmp, &i);
1291 		if (error)
1292 			goto out_error;
1293 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1294 		if (tmp.rc_startblock >= agbno + aglen ||
1295 		    tmp.rc_startblock != *fbno + *flen)
1296 			break;
1297 		*flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1298 	}
1299 
1300 done:
1301 	trace_xfs_refcount_find_shared_result(cur->bc_mp,
1302 			cur->bc_private.a.agno, *fbno, *flen);
1303 
1304 out_error:
1305 	if (error)
1306 		trace_xfs_refcount_find_shared_error(cur->bc_mp,
1307 				cur->bc_private.a.agno, error, _RET_IP_);
1308 	return error;
1309 }
1310 
1311 /*
1312  * Recovering CoW Blocks After a Crash
1313  *
1314  * Due to the way that the copy on write mechanism works, there's a window of
1315  * opportunity in which we can lose track of allocated blocks during a crash.
1316  * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1317  * causes blocks to be allocated and stored in the CoW fork.  The blocks are
1318  * no longer in the free space btree but are not otherwise recorded anywhere
1319  * until the write completes and the blocks are mapped into the file.  A crash
1320  * in between allocation and remapping results in the replacement blocks being
1321  * lost.  This situation is exacerbated by the CoW extent size hint because
1322  * allocations can hang around for long time.
1323  *
1324  * However, there is a place where we can record these allocations before they
1325  * become mappings -- the reference count btree.  The btree does not record
1326  * extents with refcount == 1, so we can record allocations with a refcount of
1327  * 1.  Blocks being used for CoW writeout cannot be shared, so there should be
1328  * no conflict with shared block records.  These mappings should be created
1329  * when we allocate blocks to the CoW fork and deleted when they're removed
1330  * from the CoW fork.
1331  *
1332  * Minor nit: records for in-progress CoW allocations and records for shared
1333  * extents must never be merged, to preserve the property that (except for CoW
1334  * allocations) there are no refcount btree entries with refcount == 1.  The
1335  * only time this could potentially happen is when unsharing a block that's
1336  * adjacent to CoW allocations, so we must be careful to avoid this.
1337  *
1338  * At mount time we recover lost CoW allocations by searching the refcount
1339  * btree for these refcount == 1 mappings.  These represent CoW allocations
1340  * that were in progress at the time the filesystem went down, so we can free
1341  * them to get the space back.
1342  *
1343  * This mechanism is superior to creating EFIs for unmapped CoW extents for
1344  * several reasons -- first, EFIs pin the tail of the log and would have to be
1345  * periodically relogged to avoid filling up the log.  Second, CoW completions
1346  * will have to file an EFD and create new EFIs for whatever remains in the
1347  * CoW fork; this partially takes care of (1) but extent-size reservations
1348  * will have to periodically relog even if there's no writeout in progress.
1349  * This can happen if the CoW extent size hint is set, which you really want.
1350  * Third, EFIs cannot currently be automatically relogged into newer
1351  * transactions to advance the log tail.  Fourth, stuffing the log full of
1352  * EFIs places an upper bound on the number of CoW allocations that can be
1353  * held filesystem-wide at any given time.  Recording them in the refcount
1354  * btree doesn't require us to maintain any state in memory and doesn't pin
1355  * the log.
1356  */
1357 /*
1358  * Adjust the refcounts of CoW allocations.  These allocations are "magic"
1359  * in that they're not referenced anywhere else in the filesystem, so we
1360  * stash them in the refcount btree with a refcount of 1 until either file
1361  * remapping (or CoW cancellation) happens.
1362  */
1363 STATIC int
1364 xfs_refcount_adjust_cow_extents(
1365 	struct xfs_btree_cur	*cur,
1366 	xfs_agblock_t		agbno,
1367 	xfs_extlen_t		aglen,
1368 	enum xfs_refc_adjust_op	adj,
1369 	struct xfs_defer_ops	*dfops,
1370 	struct xfs_owner_info	*oinfo)
1371 {
1372 	struct xfs_refcount_irec	ext, tmp;
1373 	int				error;
1374 	int				found_rec, found_tmp;
1375 
1376 	if (aglen == 0)
1377 		return 0;
1378 
1379 	/* Find any overlapping refcount records */
1380 	error = xfs_refcount_lookup_ge(cur, agbno, &found_rec);
1381 	if (error)
1382 		goto out_error;
1383 	error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1384 	if (error)
1385 		goto out_error;
1386 	if (!found_rec) {
1387 		ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks +
1388 				XFS_REFC_COW_START;
1389 		ext.rc_blockcount = 0;
1390 		ext.rc_refcount = 0;
1391 	}
1392 
1393 	switch (adj) {
1394 	case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1395 		/* Adding a CoW reservation, there should be nothing here. */
1396 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1397 				ext.rc_startblock >= agbno + aglen, out_error);
1398 
1399 		tmp.rc_startblock = agbno;
1400 		tmp.rc_blockcount = aglen;
1401 		tmp.rc_refcount = 1;
1402 		trace_xfs_refcount_modify_extent(cur->bc_mp,
1403 				cur->bc_private.a.agno, &tmp);
1404 
1405 		error = xfs_refcount_insert(cur, &tmp,
1406 				&found_tmp);
1407 		if (error)
1408 			goto out_error;
1409 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1410 				found_tmp == 1, out_error);
1411 		break;
1412 	case XFS_REFCOUNT_ADJUST_COW_FREE:
1413 		/* Removing a CoW reservation, there should be one extent. */
1414 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1415 			ext.rc_startblock == agbno, out_error);
1416 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1417 			ext.rc_blockcount == aglen, out_error);
1418 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1419 			ext.rc_refcount == 1, out_error);
1420 
1421 		ext.rc_refcount = 0;
1422 		trace_xfs_refcount_modify_extent(cur->bc_mp,
1423 				cur->bc_private.a.agno, &ext);
1424 		error = xfs_refcount_delete(cur, &found_rec);
1425 		if (error)
1426 			goto out_error;
1427 		XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1428 				found_rec == 1, out_error);
1429 		break;
1430 	default:
1431 		ASSERT(0);
1432 	}
1433 
1434 	return error;
1435 out_error:
1436 	trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1437 			cur->bc_private.a.agno, error, _RET_IP_);
1438 	return error;
1439 }
1440 
1441 /*
1442  * Add or remove refcount btree entries for CoW reservations.
1443  */
1444 STATIC int
1445 xfs_refcount_adjust_cow(
1446 	struct xfs_btree_cur	*cur,
1447 	xfs_agblock_t		agbno,
1448 	xfs_extlen_t		aglen,
1449 	enum xfs_refc_adjust_op	adj,
1450 	struct xfs_defer_ops	*dfops)
1451 {
1452 	bool			shape_changed;
1453 	int			error;
1454 
1455 	agbno += XFS_REFC_COW_START;
1456 
1457 	/*
1458 	 * Ensure that no rcextents cross the boundary of the adjustment range.
1459 	 */
1460 	error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1461 	if (error)
1462 		goto out_error;
1463 
1464 	error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1465 	if (error)
1466 		goto out_error;
1467 
1468 	/*
1469 	 * Try to merge with the left or right extents of the range.
1470 	 */
1471 	error = xfs_refcount_merge_extents(cur, &agbno, &aglen, adj,
1472 			XFS_FIND_RCEXT_COW, &shape_changed);
1473 	if (error)
1474 		goto out_error;
1475 
1476 	/* Now that we've taken care of the ends, adjust the middle extents */
1477 	error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj,
1478 			dfops, NULL);
1479 	if (error)
1480 		goto out_error;
1481 
1482 	return 0;
1483 
1484 out_error:
1485 	trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_private.a.agno,
1486 			error, _RET_IP_);
1487 	return error;
1488 }
1489 
1490 /*
1491  * Record a CoW allocation in the refcount btree.
1492  */
1493 STATIC int
1494 __xfs_refcount_cow_alloc(
1495 	struct xfs_btree_cur	*rcur,
1496 	xfs_agblock_t		agbno,
1497 	xfs_extlen_t		aglen,
1498 	struct xfs_defer_ops	*dfops)
1499 {
1500 	int			error;
1501 
1502 	trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_private.a.agno,
1503 			agbno, aglen);
1504 
1505 	/* Add refcount btree reservation */
1506 	error = xfs_refcount_adjust_cow(rcur, agbno, aglen,
1507 			XFS_REFCOUNT_ADJUST_COW_ALLOC, dfops);
1508 	if (error)
1509 		return error;
1510 
1511 	/* Add rmap entry */
1512 	if (xfs_sb_version_hasrmapbt(&rcur->bc_mp->m_sb)) {
1513 		error = xfs_rmap_alloc_extent(rcur->bc_mp, dfops,
1514 				rcur->bc_private.a.agno,
1515 				agbno, aglen, XFS_RMAP_OWN_COW);
1516 		if (error)
1517 			return error;
1518 	}
1519 
1520 	return error;
1521 }
1522 
1523 /*
1524  * Remove a CoW allocation from the refcount btree.
1525  */
1526 STATIC int
1527 __xfs_refcount_cow_free(
1528 	struct xfs_btree_cur	*rcur,
1529 	xfs_agblock_t		agbno,
1530 	xfs_extlen_t		aglen,
1531 	struct xfs_defer_ops	*dfops)
1532 {
1533 	int			error;
1534 
1535 	trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_private.a.agno,
1536 			agbno, aglen);
1537 
1538 	/* Remove refcount btree reservation */
1539 	error = xfs_refcount_adjust_cow(rcur, agbno, aglen,
1540 			XFS_REFCOUNT_ADJUST_COW_FREE, dfops);
1541 	if (error)
1542 		return error;
1543 
1544 	/* Remove rmap entry */
1545 	if (xfs_sb_version_hasrmapbt(&rcur->bc_mp->m_sb)) {
1546 		error = xfs_rmap_free_extent(rcur->bc_mp, dfops,
1547 				rcur->bc_private.a.agno,
1548 				agbno, aglen, XFS_RMAP_OWN_COW);
1549 		if (error)
1550 			return error;
1551 	}
1552 
1553 	return error;
1554 }
1555 
1556 /* Record a CoW staging extent in the refcount btree. */
1557 int
1558 xfs_refcount_alloc_cow_extent(
1559 	struct xfs_mount		*mp,
1560 	struct xfs_defer_ops		*dfops,
1561 	xfs_fsblock_t			fsb,
1562 	xfs_extlen_t			len)
1563 {
1564 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1565 		return 0;
1566 
1567 	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_ALLOC_COW,
1568 			fsb, len);
1569 }
1570 
1571 /* Forget a CoW staging event in the refcount btree. */
1572 int
1573 xfs_refcount_free_cow_extent(
1574 	struct xfs_mount		*mp,
1575 	struct xfs_defer_ops		*dfops,
1576 	xfs_fsblock_t			fsb,
1577 	xfs_extlen_t			len)
1578 {
1579 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
1580 		return 0;
1581 
1582 	return __xfs_refcount_add(mp, dfops, XFS_REFCOUNT_FREE_COW,
1583 			fsb, len);
1584 }
1585 
1586 struct xfs_refcount_recovery {
1587 	struct list_head		rr_list;
1588 	struct xfs_refcount_irec	rr_rrec;
1589 };
1590 
1591 /* Stuff an extent on the recovery list. */
1592 STATIC int
1593 xfs_refcount_recover_extent(
1594 	struct xfs_btree_cur		*cur,
1595 	union xfs_btree_rec		*rec,
1596 	void				*priv)
1597 {
1598 	struct list_head		*debris = priv;
1599 	struct xfs_refcount_recovery	*rr;
1600 
1601 	if (be32_to_cpu(rec->refc.rc_refcount) != 1)
1602 		return -EFSCORRUPTED;
1603 
1604 	rr = kmem_alloc(sizeof(struct xfs_refcount_recovery), KM_SLEEP);
1605 	xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1606 	list_add_tail(&rr->rr_list, debris);
1607 
1608 	return 0;
1609 }
1610 
1611 /* Find and remove leftover CoW reservations. */
1612 int
1613 xfs_refcount_recover_cow_leftovers(
1614 	struct xfs_mount		*mp,
1615 	xfs_agnumber_t			agno)
1616 {
1617 	struct xfs_trans		*tp;
1618 	struct xfs_btree_cur		*cur;
1619 	struct xfs_buf			*agbp;
1620 	struct xfs_refcount_recovery	*rr, *n;
1621 	struct list_head		debris;
1622 	union xfs_btree_irec		low;
1623 	union xfs_btree_irec		high;
1624 	struct xfs_defer_ops		dfops;
1625 	xfs_fsblock_t			fsb;
1626 	xfs_agblock_t			agbno;
1627 	int				error;
1628 
1629 	if (mp->m_sb.sb_agblocks >= XFS_REFC_COW_START)
1630 		return -EOPNOTSUPP;
1631 
1632 	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
1633 	if (error)
1634 		return error;
1635 	cur = xfs_refcountbt_init_cursor(mp, NULL, agbp, agno, NULL);
1636 
1637 	/* Find all the leftover CoW staging extents. */
1638 	INIT_LIST_HEAD(&debris);
1639 	memset(&low, 0, sizeof(low));
1640 	memset(&high, 0, sizeof(high));
1641 	low.rc.rc_startblock = XFS_REFC_COW_START;
1642 	high.rc.rc_startblock = -1U;
1643 	error = xfs_btree_query_range(cur, &low, &high,
1644 			xfs_refcount_recover_extent, &debris);
1645 	if (error)
1646 		goto out_cursor;
1647 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1648 	xfs_buf_relse(agbp);
1649 
1650 	/* Now iterate the list to free the leftovers */
1651 	list_for_each_entry(rr, &debris, rr_list) {
1652 		/* Set up transaction. */
1653 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1654 		if (error)
1655 			goto out_free;
1656 
1657 		trace_xfs_refcount_recover_extent(mp, agno, &rr->rr_rrec);
1658 
1659 		/* Free the orphan record */
1660 		xfs_defer_init(&dfops, &fsb);
1661 		agbno = rr->rr_rrec.rc_startblock - XFS_REFC_COW_START;
1662 		fsb = XFS_AGB_TO_FSB(mp, agno, agbno);
1663 		error = xfs_refcount_free_cow_extent(mp, &dfops, fsb,
1664 				rr->rr_rrec.rc_blockcount);
1665 		if (error)
1666 			goto out_defer;
1667 
1668 		/* Free the block. */
1669 		xfs_bmap_add_free(mp, &dfops, fsb,
1670 				rr->rr_rrec.rc_blockcount, NULL);
1671 
1672 		error = xfs_defer_finish(&tp, &dfops, NULL);
1673 		if (error)
1674 			goto out_defer;
1675 
1676 		error = xfs_trans_commit(tp);
1677 		if (error)
1678 			goto out_free;
1679 	}
1680 
1681 out_free:
1682 	/* Free the leftover list */
1683 	list_for_each_entry_safe(rr, n, &debris, rr_list) {
1684 		list_del(&rr->rr_list);
1685 		kmem_free(rr);
1686 	}
1687 	return error;
1688 
1689 out_cursor:
1690 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1691 	xfs_buf_relse(agbp);
1692 	goto out_free;
1693 
1694 out_defer:
1695 	xfs_defer_cancel(&dfops);
1696 	xfs_trans_cancel(tp);
1697 	goto out_free;
1698 }
1699