1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_defer.h"
14 #include "xfs_btree.h"
16 #include "xfs_refcount_btree.h"
17 #include "xfs_alloc.h"
18 #include "xfs_errortag.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
23 #include "xfs_refcount.h"
26 /* Allowable refcount adjustment amounts. */
27 enum xfs_refc_adjust_op {
28 XFS_REFCOUNT_ADJUST_INCREASE = 1,
29 XFS_REFCOUNT_ADJUST_DECREASE = -1,
30 XFS_REFCOUNT_ADJUST_COW_ALLOC = 0,
31 XFS_REFCOUNT_ADJUST_COW_FREE = -1,
34 STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
35 xfs_agblock_t agbno, xfs_extlen_t aglen);
36 STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
37 xfs_agblock_t agbno, xfs_extlen_t aglen);
40 * Look up the first record less than or equal to [bno, len] in the btree
44 xfs_refcount_lookup_le(
45 struct xfs_btree_cur *cur,
49 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.agno, bno,
51 cur->bc_rec.rc.rc_startblock = bno;
52 cur->bc_rec.rc.rc_blockcount = 0;
53 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
57 * Look up the first record greater than or equal to [bno, len] in the btree
61 xfs_refcount_lookup_ge(
62 struct xfs_btree_cur *cur,
66 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.agno, bno,
68 cur->bc_rec.rc.rc_startblock = bno;
69 cur->bc_rec.rc.rc_blockcount = 0;
70 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
74 * Look up the first record equal to [bno, len] in the btree
78 xfs_refcount_lookup_eq(
79 struct xfs_btree_cur *cur,
83 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.agno, bno,
85 cur->bc_rec.rc.rc_startblock = bno;
86 cur->bc_rec.rc.rc_blockcount = 0;
87 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
90 /* Convert on-disk record to in-core format. */
92 xfs_refcount_btrec_to_irec(
93 union xfs_btree_rec *rec,
94 struct xfs_refcount_irec *irec)
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);
102 * Get the data from the pointed-to record.
105 xfs_refcount_get_rec(
106 struct xfs_btree_cur *cur,
107 struct xfs_refcount_irec *irec,
110 struct xfs_mount *mp = cur->bc_mp;
111 xfs_agnumber_t agno = cur->bc_ag.agno;
112 union xfs_btree_rec *rec;
114 xfs_agblock_t realstart;
116 error = xfs_btree_get_rec(cur, &rec, stat);
120 xfs_refcount_btrec_to_irec(rec, irec);
122 agno = cur->bc_ag.agno;
123 if (irec->rc_blockcount == 0 || irec->rc_blockcount > MAXREFCEXTLEN)
126 /* handle special COW-staging state */
127 realstart = irec->rc_startblock;
128 if (realstart & XFS_REFC_COW_START) {
129 if (irec->rc_refcount != 1)
131 realstart &= ~XFS_REFC_COW_START;
132 } else if (irec->rc_refcount < 2) {
136 /* check for valid extent range, including overflow */
137 if (!xfs_verify_agbno(mp, agno, realstart))
139 if (realstart > realstart + irec->rc_blockcount)
141 if (!xfs_verify_agbno(mp, agno, realstart + irec->rc_blockcount - 1))
144 if (irec->rc_refcount == 0 || irec->rc_refcount > MAXREFCOUNT)
147 trace_xfs_refcount_get(cur->bc_mp, cur->bc_ag.agno, irec);
152 "Refcount BTree record corruption in AG %d detected!", agno);
154 "Start block 0x%x, block count 0x%x, references 0x%x",
155 irec->rc_startblock, irec->rc_blockcount, irec->rc_refcount);
156 return -EFSCORRUPTED;
160 * Update the record referred to by cur to the value given
161 * by [bno, len, refcount].
162 * This either works (return 0) or gets an EFSCORRUPTED error.
166 struct xfs_btree_cur *cur,
167 struct xfs_refcount_irec *irec)
169 union xfs_btree_rec rec;
172 trace_xfs_refcount_update(cur->bc_mp, cur->bc_ag.agno, irec);
173 rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
174 rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
175 rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
176 error = xfs_btree_update(cur, &rec);
178 trace_xfs_refcount_update_error(cur->bc_mp,
179 cur->bc_ag.agno, error, _RET_IP_);
184 * Insert the record referred to by cur to the value given
185 * by [bno, len, refcount].
186 * This either works (return 0) or gets an EFSCORRUPTED error.
190 struct xfs_btree_cur *cur,
191 struct xfs_refcount_irec *irec,
196 trace_xfs_refcount_insert(cur->bc_mp, cur->bc_ag.agno, irec);
197 cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
198 cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
199 cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
200 error = xfs_btree_insert(cur, i);
203 if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
204 error = -EFSCORRUPTED;
210 trace_xfs_refcount_insert_error(cur->bc_mp,
211 cur->bc_ag.agno, error, _RET_IP_);
216 * Remove the record referred to by cur, then set the pointer to the spot
217 * where the record could be re-inserted, in case we want to increment or
218 * decrement the cursor.
219 * This either works (return 0) or gets an EFSCORRUPTED error.
223 struct xfs_btree_cur *cur,
226 struct xfs_refcount_irec irec;
230 error = xfs_refcount_get_rec(cur, &irec, &found_rec);
233 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
234 error = -EFSCORRUPTED;
237 trace_xfs_refcount_delete(cur->bc_mp, cur->bc_ag.agno, &irec);
238 error = xfs_btree_delete(cur, i);
239 if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
240 error = -EFSCORRUPTED;
245 error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
248 trace_xfs_refcount_delete_error(cur->bc_mp,
249 cur->bc_ag.agno, error, _RET_IP_);
254 * Adjusting the Reference Count
256 * As stated elsewhere, the reference count btree (refcbt) stores
257 * >1 reference counts for extents of physical blocks. In this
258 * operation, we're either raising or lowering the reference count of
259 * some subrange stored in the tree:
261 * <------ adjustment range ------>
262 * ----+ +---+-----+ +--+--------+---------
263 * 2 | | 3 | 4 | |17| 55 | 10
264 * ----+ +---+-----+ +--+--------+---------
265 * X axis is physical blocks number;
266 * reference counts are the numbers inside the rectangles
268 * The first thing we need to do is to ensure that there are no
269 * refcount extents crossing either boundary of the range to be
270 * adjusted. For any extent that does cross a boundary, split it into
271 * two extents so that we can increment the refcount of one of the
274 * <------ adjustment range ------>
275 * ----+ +---+-----+ +--+--------+----+----
276 * 2 | | 3 | 2 | |17| 55 | 10 | 10
277 * ----+ +---+-----+ +--+--------+----+----
279 * For this next step, let's assume that all the physical blocks in
280 * the adjustment range are mapped to a file and are therefore in use
281 * at least once. Therefore, we can infer that any gap in the
282 * refcount tree within the adjustment range represents a physical
283 * extent with refcount == 1:
285 * <------ adjustment range ------>
286 * ----+---+---+-----+-+--+--------+----+----
287 * 2 |"1"| 3 | 2 |1|17| 55 | 10 | 10
288 * ----+---+---+-----+-+--+--------+----+----
291 * For each extent that falls within the interval range, figure out
292 * which extent is to the left or the right of that extent. Now we
293 * have a left, current, and right extent. If the new reference count
294 * of the center extent enables us to merge left, center, and right
295 * into one record covering all three, do so. If the center extent is
296 * at the left end of the range, abuts the left extent, and its new
297 * reference count matches the left extent's record, then merge them.
298 * If the center extent is at the right end of the range, abuts the
299 * right extent, and the reference counts match, merge those. In the
300 * example, we can left merge (assuming an increment operation):
302 * <------ adjustment range ------>
303 * --------+---+-----+-+--+--------+----+----
304 * 2 | 3 | 2 |1|17| 55 | 10 | 10
305 * --------+---+-----+-+--+--------+----+----
308 * For all other extents within the range, adjust the reference count
309 * or delete it if the refcount falls below 2. If we were
310 * incrementing, the end result looks like this:
312 * <------ adjustment range ------>
313 * --------+---+-----+-+--+--------+----+----
314 * 2 | 4 | 3 |2|18| 56 | 11 | 10
315 * --------+---+-----+-+--+--------+----+----
317 * The result of a decrement operation looks as such:
319 * <------ adjustment range ------>
320 * ----+ +---+ +--+--------+----+----
321 * 2 | | 2 | |16| 54 | 9 | 10
322 * ----+ +---+ +--+--------+----+----
325 * The blocks marked "D" are freed; the blocks marked "1" are only
326 * referenced once and therefore the record is removed from the
330 /* Next block after this extent. */
331 static inline xfs_agblock_t
333 struct xfs_refcount_irec *rc)
335 return rc->rc_startblock + rc->rc_blockcount;
339 * Split a refcount extent that crosses agbno.
342 xfs_refcount_split_extent(
343 struct xfs_btree_cur *cur,
347 struct xfs_refcount_irec rcext, tmp;
351 *shape_changed = false;
352 error = xfs_refcount_lookup_le(cur, agbno, &found_rec);
358 error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
361 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
362 error = -EFSCORRUPTED;
365 if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
368 *shape_changed = true;
369 trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_ag.agno,
372 /* Establish the right extent. */
374 tmp.rc_startblock = agbno;
375 tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
376 error = xfs_refcount_update(cur, &tmp);
380 /* Insert the left extent. */
382 tmp.rc_blockcount = agbno - rcext.rc_startblock;
383 error = xfs_refcount_insert(cur, &tmp, &found_rec);
386 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
387 error = -EFSCORRUPTED;
393 trace_xfs_refcount_split_extent_error(cur->bc_mp,
394 cur->bc_ag.agno, error, _RET_IP_);
399 * Merge the left, center, and right extents.
402 xfs_refcount_merge_center_extents(
403 struct xfs_btree_cur *cur,
404 struct xfs_refcount_irec *left,
405 struct xfs_refcount_irec *center,
406 struct xfs_refcount_irec *right,
407 unsigned long long extlen,
413 trace_xfs_refcount_merge_center_extents(cur->bc_mp,
414 cur->bc_ag.agno, left, center, right);
417 * Make sure the center and right extents are not in the btree.
418 * If the center extent was synthesized, the first delete call
419 * removes the right extent and we skip the second deletion.
420 * If center and right were in the btree, then the first delete
421 * call removes the center and the second one removes the right
424 error = xfs_refcount_lookup_ge(cur, center->rc_startblock,
428 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
429 error = -EFSCORRUPTED;
433 error = xfs_refcount_delete(cur, &found_rec);
436 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
437 error = -EFSCORRUPTED;
441 if (center->rc_refcount > 1) {
442 error = xfs_refcount_delete(cur, &found_rec);
445 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
446 error = -EFSCORRUPTED;
451 /* Enlarge the left extent. */
452 error = xfs_refcount_lookup_le(cur, left->rc_startblock,
456 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
457 error = -EFSCORRUPTED;
461 left->rc_blockcount = extlen;
462 error = xfs_refcount_update(cur, left);
470 trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
471 cur->bc_ag.agno, error, _RET_IP_);
476 * Merge with the left extent.
479 xfs_refcount_merge_left_extent(
480 struct xfs_btree_cur *cur,
481 struct xfs_refcount_irec *left,
482 struct xfs_refcount_irec *cleft,
483 xfs_agblock_t *agbno,
489 trace_xfs_refcount_merge_left_extent(cur->bc_mp,
490 cur->bc_ag.agno, left, cleft);
492 /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
493 if (cleft->rc_refcount > 1) {
494 error = xfs_refcount_lookup_le(cur, cleft->rc_startblock,
498 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
499 error = -EFSCORRUPTED;
503 error = xfs_refcount_delete(cur, &found_rec);
506 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
507 error = -EFSCORRUPTED;
512 /* Enlarge the left extent. */
513 error = xfs_refcount_lookup_le(cur, left->rc_startblock,
517 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
518 error = -EFSCORRUPTED;
522 left->rc_blockcount += cleft->rc_blockcount;
523 error = xfs_refcount_update(cur, left);
527 *agbno += cleft->rc_blockcount;
528 *aglen -= cleft->rc_blockcount;
532 trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
533 cur->bc_ag.agno, error, _RET_IP_);
538 * Merge with the right extent.
541 xfs_refcount_merge_right_extent(
542 struct xfs_btree_cur *cur,
543 struct xfs_refcount_irec *right,
544 struct xfs_refcount_irec *cright,
550 trace_xfs_refcount_merge_right_extent(cur->bc_mp,
551 cur->bc_ag.agno, cright, right);
554 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
557 if (cright->rc_refcount > 1) {
558 error = xfs_refcount_lookup_le(cur, cright->rc_startblock,
562 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
563 error = -EFSCORRUPTED;
567 error = xfs_refcount_delete(cur, &found_rec);
570 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
571 error = -EFSCORRUPTED;
576 /* Enlarge the right extent. */
577 error = xfs_refcount_lookup_le(cur, right->rc_startblock,
581 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
582 error = -EFSCORRUPTED;
586 right->rc_startblock -= cright->rc_blockcount;
587 right->rc_blockcount += cright->rc_blockcount;
588 error = xfs_refcount_update(cur, right);
592 *aglen -= cright->rc_blockcount;
596 trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
597 cur->bc_ag.agno, error, _RET_IP_);
601 #define XFS_FIND_RCEXT_SHARED 1
602 #define XFS_FIND_RCEXT_COW 2
604 * Find the left extent and the one after it (cleft). This function assumes
605 * that we've already split any extent crossing agbno.
608 xfs_refcount_find_left_extents(
609 struct xfs_btree_cur *cur,
610 struct xfs_refcount_irec *left,
611 struct xfs_refcount_irec *cleft,
616 struct xfs_refcount_irec tmp;
620 left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
621 error = xfs_refcount_lookup_le(cur, agbno - 1, &found_rec);
627 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
630 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
631 error = -EFSCORRUPTED;
635 if (xfs_refc_next(&tmp) != agbno)
637 if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
639 if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
641 /* We have a left extent; retrieve (or invent) the next right one */
644 error = xfs_btree_increment(cur, 0, &found_rec);
648 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
651 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
652 error = -EFSCORRUPTED;
656 /* if tmp starts at the end of our range, just use that */
657 if (tmp.rc_startblock == agbno)
661 * There's a gap in the refcntbt at the start of the
662 * range we're interested in (refcount == 1) so
663 * synthesize the implied extent and pass it back.
664 * We assume here that the agbno/aglen range was
665 * passed in from a data fork extent mapping and
666 * therefore is allocated to exactly one owner.
668 cleft->rc_startblock = agbno;
669 cleft->rc_blockcount = min(aglen,
670 tmp.rc_startblock - agbno);
671 cleft->rc_refcount = 1;
675 * No extents, so pretend that there's one covering the whole
678 cleft->rc_startblock = agbno;
679 cleft->rc_blockcount = aglen;
680 cleft->rc_refcount = 1;
682 trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_ag.agno,
687 trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
688 cur->bc_ag.agno, error, _RET_IP_);
693 * Find the right extent and the one before it (cright). This function
694 * assumes that we've already split any extents crossing agbno + aglen.
697 xfs_refcount_find_right_extents(
698 struct xfs_btree_cur *cur,
699 struct xfs_refcount_irec *right,
700 struct xfs_refcount_irec *cright,
705 struct xfs_refcount_irec tmp;
709 right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
710 error = xfs_refcount_lookup_ge(cur, agbno + aglen, &found_rec);
716 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
719 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
720 error = -EFSCORRUPTED;
724 if (tmp.rc_startblock != agbno + aglen)
726 if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
728 if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
730 /* We have a right extent; retrieve (or invent) the next left one */
733 error = xfs_btree_decrement(cur, 0, &found_rec);
737 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
740 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
741 error = -EFSCORRUPTED;
745 /* if tmp ends at the end of our range, just use that */
746 if (xfs_refc_next(&tmp) == agbno + aglen)
750 * There's a gap in the refcntbt at the end of the
751 * range we're interested in (refcount == 1) so
752 * create the implied extent and pass it back.
753 * We assume here that the agbno/aglen range was
754 * passed in from a data fork extent mapping and
755 * therefore is allocated to exactly one owner.
757 cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
758 cright->rc_blockcount = right->rc_startblock -
759 cright->rc_startblock;
760 cright->rc_refcount = 1;
764 * No extents, so pretend that there's one covering the whole
767 cright->rc_startblock = agbno;
768 cright->rc_blockcount = aglen;
769 cright->rc_refcount = 1;
771 trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_ag.agno,
772 cright, right, agbno + aglen);
776 trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
777 cur->bc_ag.agno, error, _RET_IP_);
781 /* Is this extent valid? */
784 struct xfs_refcount_irec *rc)
786 return rc->rc_startblock != NULLAGBLOCK;
790 * Try to merge with any extents on the boundaries of the adjustment range.
793 xfs_refcount_merge_extents(
794 struct xfs_btree_cur *cur,
795 xfs_agblock_t *agbno,
797 enum xfs_refc_adjust_op adjust,
801 struct xfs_refcount_irec left = {0}, cleft = {0};
802 struct xfs_refcount_irec cright = {0}, right = {0};
804 unsigned long long ulen;
807 *shape_changed = false;
809 * Find the extent just below agbno [left], just above agbno [cleft],
810 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
813 error = xfs_refcount_find_left_extents(cur, &left, &cleft, *agbno,
817 error = xfs_refcount_find_right_extents(cur, &right, &cright, *agbno,
822 /* No left or right extent to merge; exit. */
823 if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
826 cequal = (cleft.rc_startblock == cright.rc_startblock) &&
827 (cleft.rc_blockcount == cright.rc_blockcount);
829 /* Try to merge left, cleft, and right. cleft must == cright. */
830 ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount +
832 if (xfs_refc_valid(&left) && xfs_refc_valid(&right) &&
833 xfs_refc_valid(&cleft) && xfs_refc_valid(&cright) && cequal &&
834 left.rc_refcount == cleft.rc_refcount + adjust &&
835 right.rc_refcount == cleft.rc_refcount + adjust &&
836 ulen < MAXREFCEXTLEN) {
837 *shape_changed = true;
838 return xfs_refcount_merge_center_extents(cur, &left, &cleft,
839 &right, ulen, aglen);
842 /* Try to merge left and cleft. */
843 ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount;
844 if (xfs_refc_valid(&left) && xfs_refc_valid(&cleft) &&
845 left.rc_refcount == cleft.rc_refcount + adjust &&
846 ulen < MAXREFCEXTLEN) {
847 *shape_changed = true;
848 error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
854 * If we just merged left + cleft and cleft == cright,
855 * we no longer have a cright to merge with right. We're done.
861 /* Try to merge cright and right. */
862 ulen = (unsigned long long)right.rc_blockcount + cright.rc_blockcount;
863 if (xfs_refc_valid(&right) && xfs_refc_valid(&cright) &&
864 right.rc_refcount == cright.rc_refcount + adjust &&
865 ulen < MAXREFCEXTLEN) {
866 *shape_changed = true;
867 return xfs_refcount_merge_right_extent(cur, &right, &cright,
875 * XXX: This is a pretty hand-wavy estimate. The penalty for guessing
876 * true incorrectly is a shutdown FS; the penalty for guessing false
877 * incorrectly is more transaction rolls than might be necessary.
878 * Be conservative here.
881 xfs_refcount_still_have_space(
882 struct xfs_btree_cur *cur)
884 unsigned long overhead;
886 overhead = cur->bc_ag.refc.shape_changes *
887 xfs_allocfree_log_count(cur->bc_mp, 1);
888 overhead *= cur->bc_mp->m_sb.sb_blocksize;
891 * Only allow 2 refcount extent updates per transaction if the
892 * refcount continue update "error" has been injected.
894 if (cur->bc_ag.refc.nr_ops > 2 &&
895 XFS_TEST_ERROR(false, cur->bc_mp,
896 XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
899 if (cur->bc_ag.refc.nr_ops == 0)
901 else if (overhead > cur->bc_tp->t_log_res)
903 return cur->bc_tp->t_log_res - overhead >
904 cur->bc_ag.refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
908 * Adjust the refcounts of middle extents. At this point we should have
909 * split extents that crossed the adjustment range; merged with adjacent
910 * extents; and updated agbno/aglen to reflect the merges. Therefore,
911 * all we have to do is update the extents inside [agbno, agbno + aglen].
914 xfs_refcount_adjust_extents(
915 struct xfs_btree_cur *cur,
916 xfs_agblock_t *agbno,
918 enum xfs_refc_adjust_op adj,
919 struct xfs_owner_info *oinfo)
921 struct xfs_refcount_irec ext, tmp;
923 int found_rec, found_tmp;
926 /* Merging did all the work already. */
930 error = xfs_refcount_lookup_ge(cur, *agbno, &found_rec);
934 while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
935 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
939 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
940 ext.rc_blockcount = 0;
945 * Deal with a hole in the refcount tree; if a file maps to
946 * these blocks and there's no refcountbt record, pretend that
947 * there is one with refcount == 1.
949 if (ext.rc_startblock != *agbno) {
950 tmp.rc_startblock = *agbno;
951 tmp.rc_blockcount = min(*aglen,
952 ext.rc_startblock - *agbno);
953 tmp.rc_refcount = 1 + adj;
954 trace_xfs_refcount_modify_extent(cur->bc_mp,
955 cur->bc_ag.agno, &tmp);
958 * Either cover the hole (increment) or
959 * delete the range (decrement).
961 if (tmp.rc_refcount) {
962 error = xfs_refcount_insert(cur, &tmp,
966 if (XFS_IS_CORRUPT(cur->bc_mp,
968 error = -EFSCORRUPTED;
971 cur->bc_ag.refc.nr_ops++;
973 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
976 xfs_bmap_add_free(cur->bc_tp, fsbno,
977 tmp.rc_blockcount, oinfo);
980 (*agbno) += tmp.rc_blockcount;
981 (*aglen) -= tmp.rc_blockcount;
983 error = xfs_refcount_lookup_ge(cur, *agbno,
989 /* Stop if there's nothing left to modify */
990 if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
994 * Adjust the reference count and either update the tree
995 * (incr) or free the blocks (decr).
997 if (ext.rc_refcount == MAXREFCOUNT)
999 ext.rc_refcount += adj;
1000 trace_xfs_refcount_modify_extent(cur->bc_mp,
1001 cur->bc_ag.agno, &ext);
1002 if (ext.rc_refcount > 1) {
1003 error = xfs_refcount_update(cur, &ext);
1006 cur->bc_ag.refc.nr_ops++;
1007 } else if (ext.rc_refcount == 1) {
1008 error = xfs_refcount_delete(cur, &found_rec);
1011 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1012 error = -EFSCORRUPTED;
1015 cur->bc_ag.refc.nr_ops++;
1018 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1021 xfs_bmap_add_free(cur->bc_tp, fsbno, ext.rc_blockcount,
1026 error = xfs_btree_increment(cur, 0, &found_rec);
1031 (*agbno) += ext.rc_blockcount;
1032 (*aglen) -= ext.rc_blockcount;
1037 trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1038 cur->bc_ag.agno, error, _RET_IP_);
1042 /* Adjust the reference count of a range of AG blocks. */
1044 xfs_refcount_adjust(
1045 struct xfs_btree_cur *cur,
1046 xfs_agblock_t agbno,
1048 xfs_agblock_t *new_agbno,
1049 xfs_extlen_t *new_aglen,
1050 enum xfs_refc_adjust_op adj,
1051 struct xfs_owner_info *oinfo)
1054 int shape_changes = 0;
1059 if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
1060 trace_xfs_refcount_increase(cur->bc_mp, cur->bc_ag.agno,
1063 trace_xfs_refcount_decrease(cur->bc_mp, cur->bc_ag.agno,
1067 * Ensure that no rcextents cross the boundary of the adjustment range.
1069 error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1075 error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1082 * Try to merge with the left or right extents of the range.
1084 error = xfs_refcount_merge_extents(cur, new_agbno, new_aglen, adj,
1085 XFS_FIND_RCEXT_SHARED, &shape_changed);
1091 cur->bc_ag.refc.shape_changes++;
1093 /* Now that we've taken care of the ends, adjust the middle extents */
1094 error = xfs_refcount_adjust_extents(cur, new_agbno, new_aglen,
1102 trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_ag.agno,
1107 /* Clean up after calling xfs_refcount_finish_one. */
1109 xfs_refcount_finish_one_cleanup(
1110 struct xfs_trans *tp,
1111 struct xfs_btree_cur *rcur,
1114 struct xfs_buf *agbp;
1118 agbp = rcur->bc_ag.agbp;
1119 xfs_btree_del_cursor(rcur, error);
1121 xfs_trans_brelse(tp, agbp);
1125 * Process one of the deferred refcount operations. We pass back the
1126 * btree cursor to maintain our lock on the btree between calls.
1127 * This saves time and eliminates a buffer deadlock between the
1128 * superblock and the AGF because we'll always grab them in the same
1132 xfs_refcount_finish_one(
1133 struct xfs_trans *tp,
1134 enum xfs_refcount_intent_type type,
1135 xfs_fsblock_t startblock,
1136 xfs_extlen_t blockcount,
1137 xfs_fsblock_t *new_fsb,
1138 xfs_extlen_t *new_len,
1139 struct xfs_btree_cur **pcur)
1141 struct xfs_mount *mp = tp->t_mountp;
1142 struct xfs_btree_cur *rcur;
1143 struct xfs_buf *agbp = NULL;
1145 xfs_agnumber_t agno;
1147 xfs_agblock_t new_agbno;
1148 unsigned long nr_ops = 0;
1149 int shape_changes = 0;
1151 agno = XFS_FSB_TO_AGNO(mp, startblock);
1152 ASSERT(agno != NULLAGNUMBER);
1153 bno = XFS_FSB_TO_AGBNO(mp, startblock);
1155 trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, startblock),
1156 type, XFS_FSB_TO_AGBNO(mp, startblock),
1159 if (XFS_TEST_ERROR(false, mp,
1160 XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1164 * If we haven't gotten a cursor or the cursor AG doesn't match
1165 * the startblock, get one now.
1168 if (rcur != NULL && rcur->bc_ag.agno != agno) {
1169 nr_ops = rcur->bc_ag.refc.nr_ops;
1170 shape_changes = rcur->bc_ag.refc.shape_changes;
1171 xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1176 error = xfs_alloc_read_agf(tp->t_mountp, tp, agno,
1177 XFS_ALLOC_FLAG_FREEING, &agbp);
1181 rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno);
1186 rcur->bc_ag.refc.nr_ops = nr_ops;
1187 rcur->bc_ag.refc.shape_changes = shape_changes;
1192 case XFS_REFCOUNT_INCREASE:
1193 error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1194 new_len, XFS_REFCOUNT_ADJUST_INCREASE, NULL);
1195 *new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1197 case XFS_REFCOUNT_DECREASE:
1198 error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1199 new_len, XFS_REFCOUNT_ADJUST_DECREASE, NULL);
1200 *new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1202 case XFS_REFCOUNT_ALLOC_COW:
1203 *new_fsb = startblock + blockcount;
1205 error = __xfs_refcount_cow_alloc(rcur, bno, blockcount);
1207 case XFS_REFCOUNT_FREE_COW:
1208 *new_fsb = startblock + blockcount;
1210 error = __xfs_refcount_cow_free(rcur, bno, blockcount);
1214 error = -EFSCORRUPTED;
1216 if (!error && *new_len > 0)
1217 trace_xfs_refcount_finish_one_leftover(mp, agno, type,
1218 bno, blockcount, new_agbno, *new_len);
1222 xfs_trans_brelse(tp, agbp);
1228 * Record a refcount intent for later processing.
1232 struct xfs_trans *tp,
1233 enum xfs_refcount_intent_type type,
1234 xfs_fsblock_t startblock,
1235 xfs_extlen_t blockcount)
1237 struct xfs_refcount_intent *ri;
1239 trace_xfs_refcount_defer(tp->t_mountp,
1240 XFS_FSB_TO_AGNO(tp->t_mountp, startblock),
1241 type, XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
1244 ri = kmem_alloc(sizeof(struct xfs_refcount_intent),
1246 INIT_LIST_HEAD(&ri->ri_list);
1248 ri->ri_startblock = startblock;
1249 ri->ri_blockcount = blockcount;
1251 xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
1255 * Increase the reference count of the blocks backing a file's extent.
1258 xfs_refcount_increase_extent(
1259 struct xfs_trans *tp,
1260 struct xfs_bmbt_irec *PREV)
1262 if (!xfs_sb_version_hasreflink(&tp->t_mountp->m_sb))
1265 __xfs_refcount_add(tp, XFS_REFCOUNT_INCREASE, PREV->br_startblock,
1266 PREV->br_blockcount);
1270 * Decrease the reference count of the blocks backing a file's extent.
1273 xfs_refcount_decrease_extent(
1274 struct xfs_trans *tp,
1275 struct xfs_bmbt_irec *PREV)
1277 if (!xfs_sb_version_hasreflink(&tp->t_mountp->m_sb))
1280 __xfs_refcount_add(tp, XFS_REFCOUNT_DECREASE, PREV->br_startblock,
1281 PREV->br_blockcount);
1285 * Given an AG extent, find the lowest-numbered run of shared blocks
1286 * within that range and return the range in fbno/flen. If
1287 * find_end_of_shared is set, return the longest contiguous extent of
1288 * shared blocks; if not, just return the first extent we find. If no
1289 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1290 * and 0, respectively.
1293 xfs_refcount_find_shared(
1294 struct xfs_btree_cur *cur,
1295 xfs_agblock_t agbno,
1297 xfs_agblock_t *fbno,
1299 bool find_end_of_shared)
1301 struct xfs_refcount_irec tmp;
1306 trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_ag.agno,
1309 /* By default, skip the whole range */
1310 *fbno = NULLAGBLOCK;
1313 /* Try to find a refcount extent that crosses the start */
1314 error = xfs_refcount_lookup_le(cur, agbno, &have);
1318 /* No left extent, look at the next one */
1319 error = xfs_btree_increment(cur, 0, &have);
1325 error = xfs_refcount_get_rec(cur, &tmp, &i);
1328 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1329 error = -EFSCORRUPTED;
1333 /* If the extent ends before the start, look at the next one */
1334 if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1335 error = xfs_btree_increment(cur, 0, &have);
1340 error = xfs_refcount_get_rec(cur, &tmp, &i);
1343 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1344 error = -EFSCORRUPTED;
1349 /* If the extent starts after the range we want, bail out */
1350 if (tmp.rc_startblock >= agbno + aglen)
1353 /* We found the start of a shared extent! */
1354 if (tmp.rc_startblock < agbno) {
1355 tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1356 tmp.rc_startblock = agbno;
1359 *fbno = tmp.rc_startblock;
1360 *flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1361 if (!find_end_of_shared)
1364 /* Otherwise, find the end of this shared extent */
1365 while (*fbno + *flen < agbno + aglen) {
1366 error = xfs_btree_increment(cur, 0, &have);
1371 error = xfs_refcount_get_rec(cur, &tmp, &i);
1374 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1375 error = -EFSCORRUPTED;
1378 if (tmp.rc_startblock >= agbno + aglen ||
1379 tmp.rc_startblock != *fbno + *flen)
1381 *flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1385 trace_xfs_refcount_find_shared_result(cur->bc_mp,
1386 cur->bc_ag.agno, *fbno, *flen);
1390 trace_xfs_refcount_find_shared_error(cur->bc_mp,
1391 cur->bc_ag.agno, error, _RET_IP_);
1396 * Recovering CoW Blocks After a Crash
1398 * Due to the way that the copy on write mechanism works, there's a window of
1399 * opportunity in which we can lose track of allocated blocks during a crash.
1400 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1401 * causes blocks to be allocated and stored in the CoW fork. The blocks are
1402 * no longer in the free space btree but are not otherwise recorded anywhere
1403 * until the write completes and the blocks are mapped into the file. A crash
1404 * in between allocation and remapping results in the replacement blocks being
1405 * lost. This situation is exacerbated by the CoW extent size hint because
1406 * allocations can hang around for long time.
1408 * However, there is a place where we can record these allocations before they
1409 * become mappings -- the reference count btree. The btree does not record
1410 * extents with refcount == 1, so we can record allocations with a refcount of
1411 * 1. Blocks being used for CoW writeout cannot be shared, so there should be
1412 * no conflict with shared block records. These mappings should be created
1413 * when we allocate blocks to the CoW fork and deleted when they're removed
1414 * from the CoW fork.
1416 * Minor nit: records for in-progress CoW allocations and records for shared
1417 * extents must never be merged, to preserve the property that (except for CoW
1418 * allocations) there are no refcount btree entries with refcount == 1. The
1419 * only time this could potentially happen is when unsharing a block that's
1420 * adjacent to CoW allocations, so we must be careful to avoid this.
1422 * At mount time we recover lost CoW allocations by searching the refcount
1423 * btree for these refcount == 1 mappings. These represent CoW allocations
1424 * that were in progress at the time the filesystem went down, so we can free
1425 * them to get the space back.
1427 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1428 * several reasons -- first, EFIs pin the tail of the log and would have to be
1429 * periodically relogged to avoid filling up the log. Second, CoW completions
1430 * will have to file an EFD and create new EFIs for whatever remains in the
1431 * CoW fork; this partially takes care of (1) but extent-size reservations
1432 * will have to periodically relog even if there's no writeout in progress.
1433 * This can happen if the CoW extent size hint is set, which you really want.
1434 * Third, EFIs cannot currently be automatically relogged into newer
1435 * transactions to advance the log tail. Fourth, stuffing the log full of
1436 * EFIs places an upper bound on the number of CoW allocations that can be
1437 * held filesystem-wide at any given time. Recording them in the refcount
1438 * btree doesn't require us to maintain any state in memory and doesn't pin
1442 * Adjust the refcounts of CoW allocations. These allocations are "magic"
1443 * in that they're not referenced anywhere else in the filesystem, so we
1444 * stash them in the refcount btree with a refcount of 1 until either file
1445 * remapping (or CoW cancellation) happens.
1448 xfs_refcount_adjust_cow_extents(
1449 struct xfs_btree_cur *cur,
1450 xfs_agblock_t agbno,
1452 enum xfs_refc_adjust_op adj)
1454 struct xfs_refcount_irec ext, tmp;
1456 int found_rec, found_tmp;
1461 /* Find any overlapping refcount records */
1462 error = xfs_refcount_lookup_ge(cur, agbno, &found_rec);
1465 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1469 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks +
1471 ext.rc_blockcount = 0;
1472 ext.rc_refcount = 0;
1476 case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1477 /* Adding a CoW reservation, there should be nothing here. */
1478 if (XFS_IS_CORRUPT(cur->bc_mp,
1479 agbno + aglen > ext.rc_startblock)) {
1480 error = -EFSCORRUPTED;
1484 tmp.rc_startblock = agbno;
1485 tmp.rc_blockcount = aglen;
1486 tmp.rc_refcount = 1;
1487 trace_xfs_refcount_modify_extent(cur->bc_mp,
1488 cur->bc_ag.agno, &tmp);
1490 error = xfs_refcount_insert(cur, &tmp,
1494 if (XFS_IS_CORRUPT(cur->bc_mp, found_tmp != 1)) {
1495 error = -EFSCORRUPTED;
1499 case XFS_REFCOUNT_ADJUST_COW_FREE:
1500 /* Removing a CoW reservation, there should be one extent. */
1501 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_startblock != agbno)) {
1502 error = -EFSCORRUPTED;
1505 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount != aglen)) {
1506 error = -EFSCORRUPTED;
1509 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_refcount != 1)) {
1510 error = -EFSCORRUPTED;
1514 ext.rc_refcount = 0;
1515 trace_xfs_refcount_modify_extent(cur->bc_mp,
1516 cur->bc_ag.agno, &ext);
1517 error = xfs_refcount_delete(cur, &found_rec);
1520 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1521 error = -EFSCORRUPTED;
1531 trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1532 cur->bc_ag.agno, error, _RET_IP_);
1537 * Add or remove refcount btree entries for CoW reservations.
1540 xfs_refcount_adjust_cow(
1541 struct xfs_btree_cur *cur,
1542 xfs_agblock_t agbno,
1544 enum xfs_refc_adjust_op adj)
1549 agbno += XFS_REFC_COW_START;
1552 * Ensure that no rcextents cross the boundary of the adjustment range.
1554 error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1558 error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1563 * Try to merge with the left or right extents of the range.
1565 error = xfs_refcount_merge_extents(cur, &agbno, &aglen, adj,
1566 XFS_FIND_RCEXT_COW, &shape_changed);
1570 /* Now that we've taken care of the ends, adjust the middle extents */
1571 error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1578 trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_ag.agno,
1584 * Record a CoW allocation in the refcount btree.
1587 __xfs_refcount_cow_alloc(
1588 struct xfs_btree_cur *rcur,
1589 xfs_agblock_t agbno,
1592 trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_ag.agno,
1595 /* Add refcount btree reservation */
1596 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1597 XFS_REFCOUNT_ADJUST_COW_ALLOC);
1601 * Remove a CoW allocation from the refcount btree.
1604 __xfs_refcount_cow_free(
1605 struct xfs_btree_cur *rcur,
1606 xfs_agblock_t agbno,
1609 trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_ag.agno,
1612 /* Remove refcount btree reservation */
1613 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1614 XFS_REFCOUNT_ADJUST_COW_FREE);
1617 /* Record a CoW staging extent in the refcount btree. */
1619 xfs_refcount_alloc_cow_extent(
1620 struct xfs_trans *tp,
1624 struct xfs_mount *mp = tp->t_mountp;
1626 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1629 __xfs_refcount_add(tp, XFS_REFCOUNT_ALLOC_COW, fsb, len);
1631 /* Add rmap entry */
1632 xfs_rmap_alloc_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1633 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1636 /* Forget a CoW staging event in the refcount btree. */
1638 xfs_refcount_free_cow_extent(
1639 struct xfs_trans *tp,
1643 struct xfs_mount *mp = tp->t_mountp;
1645 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1648 /* Remove rmap entry */
1649 xfs_rmap_free_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1650 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1651 __xfs_refcount_add(tp, XFS_REFCOUNT_FREE_COW, fsb, len);
1654 struct xfs_refcount_recovery {
1655 struct list_head rr_list;
1656 struct xfs_refcount_irec rr_rrec;
1659 /* Stuff an extent on the recovery list. */
1661 xfs_refcount_recover_extent(
1662 struct xfs_btree_cur *cur,
1663 union xfs_btree_rec *rec,
1666 struct list_head *debris = priv;
1667 struct xfs_refcount_recovery *rr;
1669 if (XFS_IS_CORRUPT(cur->bc_mp,
1670 be32_to_cpu(rec->refc.rc_refcount) != 1))
1671 return -EFSCORRUPTED;
1673 rr = kmem_alloc(sizeof(struct xfs_refcount_recovery), 0);
1674 xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1675 list_add_tail(&rr->rr_list, debris);
1680 /* Find and remove leftover CoW reservations. */
1682 xfs_refcount_recover_cow_leftovers(
1683 struct xfs_mount *mp,
1684 xfs_agnumber_t agno)
1686 struct xfs_trans *tp;
1687 struct xfs_btree_cur *cur;
1688 struct xfs_buf *agbp;
1689 struct xfs_refcount_recovery *rr, *n;
1690 struct list_head debris;
1691 union xfs_btree_irec low;
1692 union xfs_btree_irec high;
1694 xfs_agblock_t agbno;
1697 if (mp->m_sb.sb_agblocks >= XFS_REFC_COW_START)
1700 INIT_LIST_HEAD(&debris);
1703 * In this first part, we use an empty transaction to gather up
1704 * all the leftover CoW extents so that we can subsequently
1705 * delete them. The empty transaction is used to avoid
1706 * a buffer lock deadlock if there happens to be a loop in the
1707 * refcountbt because we're allowed to re-grab a buffer that is
1708 * already attached to our transaction. When we're done
1709 * recording the CoW debris we cancel the (empty) transaction
1710 * and everything goes away cleanly.
1712 error = xfs_trans_alloc_empty(mp, &tp);
1716 error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
1719 cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno);
1721 /* Find all the leftover CoW staging extents. */
1722 memset(&low, 0, sizeof(low));
1723 memset(&high, 0, sizeof(high));
1724 low.rc.rc_startblock = XFS_REFC_COW_START;
1725 high.rc.rc_startblock = -1U;
1726 error = xfs_btree_query_range(cur, &low, &high,
1727 xfs_refcount_recover_extent, &debris);
1728 xfs_btree_del_cursor(cur, error);
1729 xfs_trans_brelse(tp, agbp);
1730 xfs_trans_cancel(tp);
1734 /* Now iterate the list to free the leftovers */
1735 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1736 /* Set up transaction. */
1737 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1741 trace_xfs_refcount_recover_extent(mp, agno, &rr->rr_rrec);
1743 /* Free the orphan record */
1744 agbno = rr->rr_rrec.rc_startblock - XFS_REFC_COW_START;
1745 fsb = XFS_AGB_TO_FSB(mp, agno, agbno);
1746 xfs_refcount_free_cow_extent(tp, fsb,
1747 rr->rr_rrec.rc_blockcount);
1749 /* Free the block. */
1750 xfs_bmap_add_free(tp, fsb, rr->rr_rrec.rc_blockcount, NULL);
1752 error = xfs_trans_commit(tp);
1756 list_del(&rr->rr_list);
1762 xfs_trans_cancel(tp);
1764 /* Free the leftover list */
1765 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1766 list_del(&rr->rr_list);
1772 /* Is there a record covering a given extent? */
1774 xfs_refcount_has_record(
1775 struct xfs_btree_cur *cur,
1780 union xfs_btree_irec low;
1781 union xfs_btree_irec high;
1783 memset(&low, 0, sizeof(low));
1784 low.rc.rc_startblock = bno;
1785 memset(&high, 0xFF, sizeof(high));
1786 high.rc.rc_startblock = bno + len - 1;
1788 return xfs_btree_has_record(cur, &low, &high, exists);