2 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
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.
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.
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.
20 #include "libxfs_priv.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_btree.h"
31 #include "xfs_refcount_btree.h"
32 #include "xfs_alloc.h"
33 #include "xfs_errortag.h"
34 #include "xfs_trace.h"
35 #include "xfs_cksum.h"
36 #include "xfs_trans.h"
38 #include "xfs_refcount.h"
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,
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
);
57 * Look up the first record less than or equal to [bno, len] in the btree
61 xfs_refcount_lookup_le(
62 struct xfs_btree_cur
*cur
,
66 trace_xfs_refcount_lookup(cur
->bc_mp
, cur
->bc_private
.a
.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_LE
, stat
);
74 * Look up the first record greater than or equal to [bno, len] in the btree
78 xfs_refcount_lookup_ge(
79 struct xfs_btree_cur
*cur
,
83 trace_xfs_refcount_lookup(cur
->bc_mp
, cur
->bc_private
.a
.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_GE
, 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 union xfs_btree_rec
*rec
;
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
,
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.
129 struct xfs_btree_cur
*cur
,
130 struct xfs_refcount_irec
*irec
)
132 union xfs_btree_rec rec
;
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
);
141 trace_xfs_refcount_update_error(cur
->bc_mp
,
142 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
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.
153 struct xfs_btree_cur
*cur
,
154 struct xfs_refcount_irec
*irec
,
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
);
167 trace_xfs_refcount_insert_error(cur
->bc_mp
,
168 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
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.
180 struct xfs_btree_cur
*cur
,
183 struct xfs_refcount_irec irec
;
187 error
= xfs_refcount_get_rec(cur
, &irec
, &found_rec
);
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
);
196 error
= xfs_refcount_lookup_ge(cur
, irec
.rc_startblock
, &found_rec
);
199 trace_xfs_refcount_delete_error(cur
->bc_mp
,
200 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
205 * Adjusting the Reference Count
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:
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
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
225 * <------ adjustment range ------>
226 * ----+ +---+-----+ +--+--------+----+----
227 * 2 | | 3 | 2 | |17| 55 | 10 | 10
228 * ----+ +---+-----+ +--+--------+----+----
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:
236 * <------ adjustment range ------>
237 * ----+---+---+-----+-+--+--------+----+----
238 * 2 |"1"| 3 | 2 |1|17| 55 | 10 | 10
239 * ----+---+---+-----+-+--+--------+----+----
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):
253 * <------ adjustment range ------>
254 * --------+---+-----+-+--+--------+----+----
255 * 2 | 3 | 2 |1|17| 55 | 10 | 10
256 * --------+---+-----+-+--+--------+----+----
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:
263 * <------ adjustment range ------>
264 * --------+---+-----+-+--+--------+----+----
265 * 2 | 4 | 3 |2|18| 56 | 11 | 10
266 * --------+---+-----+-+--+--------+----+----
268 * The result of a decrement operation looks as such:
270 * <------ adjustment range ------>
271 * ----+ +---+ +--+--------+----+----
272 * 2 | | 2 | |16| 54 | 9 | 10
273 * ----+ +---+ +--+--------+----+----
276 * The blocks marked "D" are freed; the blocks marked "1" are only
277 * referenced once and therefore the record is removed from the
281 /* Next block after this extent. */
282 static inline xfs_agblock_t
284 struct xfs_refcount_irec
*rc
)
286 return rc
->rc_startblock
+ rc
->rc_blockcount
;
290 * Split a refcount extent that crosses agbno.
293 xfs_refcount_split_extent(
294 struct xfs_btree_cur
*cur
,
298 struct xfs_refcount_irec rcext
, tmp
;
302 *shape_changed
= false;
303 error
= xfs_refcount_lookup_le(cur
, agbno
, &found_rec
);
309 error
= xfs_refcount_get_rec(cur
, &rcext
, &found_rec
);
312 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
313 if (rcext
.rc_startblock
== agbno
|| xfs_refc_next(&rcext
) <= agbno
)
316 *shape_changed
= true;
317 trace_xfs_refcount_split_extent(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
320 /* Establish the right extent. */
322 tmp
.rc_startblock
= agbno
;
323 tmp
.rc_blockcount
-= (agbno
- rcext
.rc_startblock
);
324 error
= xfs_refcount_update(cur
, &tmp
);
328 /* Insert the left extent. */
330 tmp
.rc_blockcount
= agbno
- rcext
.rc_startblock
;
331 error
= xfs_refcount_insert(cur
, &tmp
, &found_rec
);
334 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
338 trace_xfs_refcount_split_extent_error(cur
->bc_mp
,
339 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
344 * Merge the left, center, and right extents.
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
,
358 trace_xfs_refcount_merge_center_extents(cur
->bc_mp
,
359 cur
->bc_private
.a
.agno
, left
, center
, right
);
362 * Make sure the center and right extents are not in the btree.
363 * If the center extent was synthesized, the first delete call
364 * removes the right extent and we skip the second deletion.
365 * If center and right were in the btree, then the first delete
366 * call removes the center and the second one removes the right
369 error
= xfs_refcount_lookup_ge(cur
, center
->rc_startblock
,
373 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
375 error
= xfs_refcount_delete(cur
, &found_rec
);
378 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
380 if (center
->rc_refcount
> 1) {
381 error
= xfs_refcount_delete(cur
, &found_rec
);
384 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
388 /* Enlarge the left extent. */
389 error
= xfs_refcount_lookup_le(cur
, left
->rc_startblock
,
393 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
395 left
->rc_blockcount
= extlen
;
396 error
= xfs_refcount_update(cur
, left
);
404 trace_xfs_refcount_merge_center_extents_error(cur
->bc_mp
,
405 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
410 * Merge with the left extent.
413 xfs_refcount_merge_left_extent(
414 struct xfs_btree_cur
*cur
,
415 struct xfs_refcount_irec
*left
,
416 struct xfs_refcount_irec
*cleft
,
417 xfs_agblock_t
*agbno
,
423 trace_xfs_refcount_merge_left_extent(cur
->bc_mp
,
424 cur
->bc_private
.a
.agno
, left
, cleft
);
426 /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
427 if (cleft
->rc_refcount
> 1) {
428 error
= xfs_refcount_lookup_le(cur
, cleft
->rc_startblock
,
432 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
435 error
= xfs_refcount_delete(cur
, &found_rec
);
438 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
442 /* Enlarge the left extent. */
443 error
= xfs_refcount_lookup_le(cur
, left
->rc_startblock
,
447 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
449 left
->rc_blockcount
+= cleft
->rc_blockcount
;
450 error
= xfs_refcount_update(cur
, left
);
454 *agbno
+= cleft
->rc_blockcount
;
455 *aglen
-= cleft
->rc_blockcount
;
459 trace_xfs_refcount_merge_left_extent_error(cur
->bc_mp
,
460 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
465 * Merge with the right extent.
468 xfs_refcount_merge_right_extent(
469 struct xfs_btree_cur
*cur
,
470 struct xfs_refcount_irec
*right
,
471 struct xfs_refcount_irec
*cright
,
477 trace_xfs_refcount_merge_right_extent(cur
->bc_mp
,
478 cur
->bc_private
.a
.agno
, cright
, right
);
481 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
484 if (cright
->rc_refcount
> 1) {
485 error
= xfs_refcount_lookup_le(cur
, cright
->rc_startblock
,
489 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
492 error
= xfs_refcount_delete(cur
, &found_rec
);
495 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
499 /* Enlarge the right extent. */
500 error
= xfs_refcount_lookup_le(cur
, right
->rc_startblock
,
504 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
506 right
->rc_startblock
-= cright
->rc_blockcount
;
507 right
->rc_blockcount
+= cright
->rc_blockcount
;
508 error
= xfs_refcount_update(cur
, right
);
512 *aglen
-= cright
->rc_blockcount
;
516 trace_xfs_refcount_merge_right_extent_error(cur
->bc_mp
,
517 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
521 #define XFS_FIND_RCEXT_SHARED 1
522 #define XFS_FIND_RCEXT_COW 2
524 * Find the left extent and the one after it (cleft). This function assumes
525 * that we've already split any extent crossing agbno.
528 xfs_refcount_find_left_extents(
529 struct xfs_btree_cur
*cur
,
530 struct xfs_refcount_irec
*left
,
531 struct xfs_refcount_irec
*cleft
,
536 struct xfs_refcount_irec tmp
;
540 left
->rc_startblock
= cleft
->rc_startblock
= NULLAGBLOCK
;
541 error
= xfs_refcount_lookup_le(cur
, agbno
- 1, &found_rec
);
547 error
= xfs_refcount_get_rec(cur
, &tmp
, &found_rec
);
550 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
552 if (xfs_refc_next(&tmp
) != agbno
)
554 if ((flags
& XFS_FIND_RCEXT_SHARED
) && tmp
.rc_refcount
< 2)
556 if ((flags
& XFS_FIND_RCEXT_COW
) && tmp
.rc_refcount
> 1)
558 /* We have a left extent; retrieve (or invent) the next right one */
561 error
= xfs_btree_increment(cur
, 0, &found_rec
);
565 error
= xfs_refcount_get_rec(cur
, &tmp
, &found_rec
);
568 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
571 /* if tmp starts at the end of our range, just use that */
572 if (tmp
.rc_startblock
== agbno
)
576 * There's a gap in the refcntbt at the start of the
577 * range we're interested in (refcount == 1) so
578 * synthesize the implied extent and pass it back.
579 * We assume here that the agbno/aglen range was
580 * passed in from a data fork extent mapping and
581 * therefore is allocated to exactly one owner.
583 cleft
->rc_startblock
= agbno
;
584 cleft
->rc_blockcount
= min(aglen
,
585 tmp
.rc_startblock
- agbno
);
586 cleft
->rc_refcount
= 1;
590 * No extents, so pretend that there's one covering the whole
593 cleft
->rc_startblock
= agbno
;
594 cleft
->rc_blockcount
= aglen
;
595 cleft
->rc_refcount
= 1;
597 trace_xfs_refcount_find_left_extent(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
602 trace_xfs_refcount_find_left_extent_error(cur
->bc_mp
,
603 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
608 * Find the right extent and the one before it (cright). This function
609 * assumes that we've already split any extents crossing agbno + aglen.
612 xfs_refcount_find_right_extents(
613 struct xfs_btree_cur
*cur
,
614 struct xfs_refcount_irec
*right
,
615 struct xfs_refcount_irec
*cright
,
620 struct xfs_refcount_irec tmp
;
624 right
->rc_startblock
= cright
->rc_startblock
= NULLAGBLOCK
;
625 error
= xfs_refcount_lookup_ge(cur
, agbno
+ aglen
, &found_rec
);
631 error
= xfs_refcount_get_rec(cur
, &tmp
, &found_rec
);
634 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1, out_error
);
636 if (tmp
.rc_startblock
!= agbno
+ aglen
)
638 if ((flags
& XFS_FIND_RCEXT_SHARED
) && tmp
.rc_refcount
< 2)
640 if ((flags
& XFS_FIND_RCEXT_COW
) && tmp
.rc_refcount
> 1)
642 /* We have a right extent; retrieve (or invent) the next left one */
645 error
= xfs_btree_decrement(cur
, 0, &found_rec
);
649 error
= xfs_refcount_get_rec(cur
, &tmp
, &found_rec
);
652 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, found_rec
== 1,
655 /* if tmp ends at the end of our range, just use that */
656 if (xfs_refc_next(&tmp
) == agbno
+ aglen
)
660 * There's a gap in the refcntbt at the end of the
661 * range we're interested in (refcount == 1) so
662 * create the implied extent and pass it back.
663 * We assume here that the agbno/aglen range was
664 * passed in from a data fork extent mapping and
665 * therefore is allocated to exactly one owner.
667 cright
->rc_startblock
= max(agbno
, xfs_refc_next(&tmp
));
668 cright
->rc_blockcount
= right
->rc_startblock
-
669 cright
->rc_startblock
;
670 cright
->rc_refcount
= 1;
674 * No extents, so pretend that there's one covering the whole
677 cright
->rc_startblock
= agbno
;
678 cright
->rc_blockcount
= aglen
;
679 cright
->rc_refcount
= 1;
681 trace_xfs_refcount_find_right_extent(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
682 cright
, right
, agbno
+ aglen
);
686 trace_xfs_refcount_find_right_extent_error(cur
->bc_mp
,
687 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
691 /* Is this extent valid? */
694 struct xfs_refcount_irec
*rc
)
696 return rc
->rc_startblock
!= NULLAGBLOCK
;
700 * Try to merge with any extents on the boundaries of the adjustment range.
703 xfs_refcount_merge_extents(
704 struct xfs_btree_cur
*cur
,
705 xfs_agblock_t
*agbno
,
707 enum xfs_refc_adjust_op adjust
,
711 struct xfs_refcount_irec left
= {0}, cleft
= {0};
712 struct xfs_refcount_irec cright
= {0}, right
= {0};
714 unsigned long long ulen
;
717 *shape_changed
= false;
719 * Find the extent just below agbno [left], just above agbno [cleft],
720 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
723 error
= xfs_refcount_find_left_extents(cur
, &left
, &cleft
, *agbno
,
727 error
= xfs_refcount_find_right_extents(cur
, &right
, &cright
, *agbno
,
732 /* No left or right extent to merge; exit. */
733 if (!xfs_refc_valid(&left
) && !xfs_refc_valid(&right
))
736 cequal
= (cleft
.rc_startblock
== cright
.rc_startblock
) &&
737 (cleft
.rc_blockcount
== cright
.rc_blockcount
);
739 /* Try to merge left, cleft, and right. cleft must == cright. */
740 ulen
= (unsigned long long)left
.rc_blockcount
+ cleft
.rc_blockcount
+
742 if (xfs_refc_valid(&left
) && xfs_refc_valid(&right
) &&
743 xfs_refc_valid(&cleft
) && xfs_refc_valid(&cright
) && cequal
&&
744 left
.rc_refcount
== cleft
.rc_refcount
+ adjust
&&
745 right
.rc_refcount
== cleft
.rc_refcount
+ adjust
&&
746 ulen
< MAXREFCEXTLEN
) {
747 *shape_changed
= true;
748 return xfs_refcount_merge_center_extents(cur
, &left
, &cleft
,
749 &right
, ulen
, aglen
);
752 /* Try to merge left and cleft. */
753 ulen
= (unsigned long long)left
.rc_blockcount
+ cleft
.rc_blockcount
;
754 if (xfs_refc_valid(&left
) && xfs_refc_valid(&cleft
) &&
755 left
.rc_refcount
== cleft
.rc_refcount
+ adjust
&&
756 ulen
< MAXREFCEXTLEN
) {
757 *shape_changed
= true;
758 error
= xfs_refcount_merge_left_extent(cur
, &left
, &cleft
,
764 * If we just merged left + cleft and cleft == cright,
765 * we no longer have a cright to merge with right. We're done.
771 /* Try to merge cright and right. */
772 ulen
= (unsigned long long)right
.rc_blockcount
+ cright
.rc_blockcount
;
773 if (xfs_refc_valid(&right
) && xfs_refc_valid(&cright
) &&
774 right
.rc_refcount
== cright
.rc_refcount
+ adjust
&&
775 ulen
< MAXREFCEXTLEN
) {
776 *shape_changed
= true;
777 return xfs_refcount_merge_right_extent(cur
, &right
, &cright
,
785 * XXX: This is a pretty hand-wavy estimate. The penalty for guessing
786 * true incorrectly is a shutdown FS; the penalty for guessing false
787 * incorrectly is more transaction rolls than might be necessary.
788 * Be conservative here.
791 xfs_refcount_still_have_space(
792 struct xfs_btree_cur
*cur
)
794 unsigned long overhead
;
796 overhead
= cur
->bc_private
.a
.priv
.refc
.shape_changes
*
797 xfs_allocfree_log_count(cur
->bc_mp
, 1);
798 overhead
*= cur
->bc_mp
->m_sb
.sb_blocksize
;
801 * Only allow 2 refcount extent updates per transaction if the
802 * refcount continue update "error" has been injected.
804 if (cur
->bc_private
.a
.priv
.refc
.nr_ops
> 2 &&
805 XFS_TEST_ERROR(false, cur
->bc_mp
,
806 XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE
))
809 if (cur
->bc_private
.a
.priv
.refc
.nr_ops
== 0)
811 else if (overhead
> cur
->bc_tp
->t_log_res
)
813 return cur
->bc_tp
->t_log_res
- overhead
>
814 cur
->bc_private
.a
.priv
.refc
.nr_ops
* XFS_REFCOUNT_ITEM_OVERHEAD
;
818 * Adjust the refcounts of middle extents. At this point we should have
819 * split extents that crossed the adjustment range; merged with adjacent
820 * extents; and updated agbno/aglen to reflect the merges. Therefore,
821 * all we have to do is update the extents inside [agbno, agbno + aglen].
824 xfs_refcount_adjust_extents(
825 struct xfs_btree_cur
*cur
,
826 xfs_agblock_t
*agbno
,
828 enum xfs_refc_adjust_op adj
,
829 struct xfs_defer_ops
*dfops
,
830 struct xfs_owner_info
*oinfo
)
832 struct xfs_refcount_irec ext
, tmp
;
834 int found_rec
, found_tmp
;
837 /* Merging did all the work already. */
841 error
= xfs_refcount_lookup_ge(cur
, *agbno
, &found_rec
);
845 while (*aglen
> 0 && xfs_refcount_still_have_space(cur
)) {
846 error
= xfs_refcount_get_rec(cur
, &ext
, &found_rec
);
850 ext
.rc_startblock
= cur
->bc_mp
->m_sb
.sb_agblocks
;
851 ext
.rc_blockcount
= 0;
856 * Deal with a hole in the refcount tree; if a file maps to
857 * these blocks and there's no refcountbt record, pretend that
858 * there is one with refcount == 1.
860 if (ext
.rc_startblock
!= *agbno
) {
861 tmp
.rc_startblock
= *agbno
;
862 tmp
.rc_blockcount
= min(*aglen
,
863 ext
.rc_startblock
- *agbno
);
864 tmp
.rc_refcount
= 1 + adj
;
865 trace_xfs_refcount_modify_extent(cur
->bc_mp
,
866 cur
->bc_private
.a
.agno
, &tmp
);
869 * Either cover the hole (increment) or
870 * delete the range (decrement).
872 if (tmp
.rc_refcount
) {
873 error
= xfs_refcount_insert(cur
, &tmp
,
877 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
878 found_tmp
== 1, out_error
);
879 cur
->bc_private
.a
.priv
.refc
.nr_ops
++;
881 fsbno
= XFS_AGB_TO_FSB(cur
->bc_mp
,
882 cur
->bc_private
.a
.agno
,
884 xfs_bmap_add_free(cur
->bc_mp
, dfops
, fsbno
,
885 tmp
.rc_blockcount
, oinfo
);
888 (*agbno
) += tmp
.rc_blockcount
;
889 (*aglen
) -= tmp
.rc_blockcount
;
891 error
= xfs_refcount_lookup_ge(cur
, *agbno
,
897 /* Stop if there's nothing left to modify */
898 if (*aglen
== 0 || !xfs_refcount_still_have_space(cur
))
902 * Adjust the reference count and either update the tree
903 * (incr) or free the blocks (decr).
905 if (ext
.rc_refcount
== MAXREFCOUNT
)
907 ext
.rc_refcount
+= adj
;
908 trace_xfs_refcount_modify_extent(cur
->bc_mp
,
909 cur
->bc_private
.a
.agno
, &ext
);
910 if (ext
.rc_refcount
> 1) {
911 error
= xfs_refcount_update(cur
, &ext
);
914 cur
->bc_private
.a
.priv
.refc
.nr_ops
++;
915 } else if (ext
.rc_refcount
== 1) {
916 error
= xfs_refcount_delete(cur
, &found_rec
);
919 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
920 found_rec
== 1, out_error
);
921 cur
->bc_private
.a
.priv
.refc
.nr_ops
++;
924 fsbno
= XFS_AGB_TO_FSB(cur
->bc_mp
,
925 cur
->bc_private
.a
.agno
,
927 xfs_bmap_add_free(cur
->bc_mp
, dfops
, fsbno
,
928 ext
.rc_blockcount
, oinfo
);
932 error
= xfs_btree_increment(cur
, 0, &found_rec
);
937 (*agbno
) += ext
.rc_blockcount
;
938 (*aglen
) -= ext
.rc_blockcount
;
943 trace_xfs_refcount_modify_extent_error(cur
->bc_mp
,
944 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
948 /* Adjust the reference count of a range of AG blocks. */
951 struct xfs_btree_cur
*cur
,
954 xfs_agblock_t
*new_agbno
,
955 xfs_extlen_t
*new_aglen
,
956 enum xfs_refc_adjust_op adj
,
957 struct xfs_defer_ops
*dfops
,
958 struct xfs_owner_info
*oinfo
)
961 int shape_changes
= 0;
966 if (adj
== XFS_REFCOUNT_ADJUST_INCREASE
)
967 trace_xfs_refcount_increase(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
970 trace_xfs_refcount_decrease(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
974 * Ensure that no rcextents cross the boundary of the adjustment range.
976 error
= xfs_refcount_split_extent(cur
, agbno
, &shape_changed
);
982 error
= xfs_refcount_split_extent(cur
, agbno
+ aglen
, &shape_changed
);
989 * Try to merge with the left or right extents of the range.
991 error
= xfs_refcount_merge_extents(cur
, new_agbno
, new_aglen
, adj
,
992 XFS_FIND_RCEXT_SHARED
, &shape_changed
);
998 cur
->bc_private
.a
.priv
.refc
.shape_changes
++;
1000 /* Now that we've taken care of the ends, adjust the middle extents */
1001 error
= xfs_refcount_adjust_extents(cur
, new_agbno
, new_aglen
,
1009 trace_xfs_refcount_adjust_error(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
1014 /* Clean up after calling xfs_refcount_finish_one. */
1016 xfs_refcount_finish_one_cleanup(
1017 struct xfs_trans
*tp
,
1018 struct xfs_btree_cur
*rcur
,
1021 struct xfs_buf
*agbp
;
1025 agbp
= rcur
->bc_private
.a
.agbp
;
1026 xfs_btree_del_cursor(rcur
, error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
1028 xfs_trans_brelse(tp
, agbp
);
1032 * Process one of the deferred refcount operations. We pass back the
1033 * btree cursor to maintain our lock on the btree between calls.
1034 * This saves time and eliminates a buffer deadlock between the
1035 * superblock and the AGF because we'll always grab them in the same
1039 xfs_refcount_finish_one(
1040 struct xfs_trans
*tp
,
1041 struct xfs_defer_ops
*dfops
,
1042 enum xfs_refcount_intent_type type
,
1043 xfs_fsblock_t startblock
,
1044 xfs_extlen_t blockcount
,
1045 xfs_fsblock_t
*new_fsb
,
1046 xfs_extlen_t
*new_len
,
1047 struct xfs_btree_cur
**pcur
)
1049 struct xfs_mount
*mp
= tp
->t_mountp
;
1050 struct xfs_btree_cur
*rcur
;
1051 struct xfs_buf
*agbp
= NULL
;
1053 xfs_agnumber_t agno
;
1055 xfs_agblock_t new_agbno
;
1056 unsigned long nr_ops
= 0;
1057 int shape_changes
= 0;
1059 agno
= XFS_FSB_TO_AGNO(mp
, startblock
);
1060 ASSERT(agno
!= NULLAGNUMBER
);
1061 bno
= XFS_FSB_TO_AGBNO(mp
, startblock
);
1063 trace_xfs_refcount_deferred(mp
, XFS_FSB_TO_AGNO(mp
, startblock
),
1064 type
, XFS_FSB_TO_AGBNO(mp
, startblock
),
1067 if (XFS_TEST_ERROR(false, mp
,
1068 XFS_ERRTAG_REFCOUNT_FINISH_ONE
))
1072 * If we haven't gotten a cursor or the cursor AG doesn't match
1073 * the startblock, get one now.
1076 if (rcur
!= NULL
&& rcur
->bc_private
.a
.agno
!= agno
) {
1077 nr_ops
= rcur
->bc_private
.a
.priv
.refc
.nr_ops
;
1078 shape_changes
= rcur
->bc_private
.a
.priv
.refc
.shape_changes
;
1079 xfs_refcount_finish_one_cleanup(tp
, rcur
, 0);
1084 error
= xfs_alloc_read_agf(tp
->t_mountp
, tp
, agno
,
1085 XFS_ALLOC_FLAG_FREEING
, &agbp
);
1089 return -EFSCORRUPTED
;
1091 rcur
= xfs_refcountbt_init_cursor(mp
, tp
, agbp
, agno
, dfops
);
1096 rcur
->bc_private
.a
.priv
.refc
.nr_ops
= nr_ops
;
1097 rcur
->bc_private
.a
.priv
.refc
.shape_changes
= shape_changes
;
1102 case XFS_REFCOUNT_INCREASE
:
1103 error
= xfs_refcount_adjust(rcur
, bno
, blockcount
, &new_agbno
,
1104 new_len
, XFS_REFCOUNT_ADJUST_INCREASE
, dfops
, NULL
);
1105 *new_fsb
= XFS_AGB_TO_FSB(mp
, agno
, new_agbno
);
1107 case XFS_REFCOUNT_DECREASE
:
1108 error
= xfs_refcount_adjust(rcur
, bno
, blockcount
, &new_agbno
,
1109 new_len
, XFS_REFCOUNT_ADJUST_DECREASE
, dfops
, NULL
);
1110 *new_fsb
= XFS_AGB_TO_FSB(mp
, agno
, new_agbno
);
1112 case XFS_REFCOUNT_ALLOC_COW
:
1113 *new_fsb
= startblock
+ blockcount
;
1115 error
= __xfs_refcount_cow_alloc(rcur
, bno
, blockcount
, dfops
);
1117 case XFS_REFCOUNT_FREE_COW
:
1118 *new_fsb
= startblock
+ blockcount
;
1120 error
= __xfs_refcount_cow_free(rcur
, bno
, blockcount
, dfops
);
1124 error
= -EFSCORRUPTED
;
1126 if (!error
&& *new_len
> 0)
1127 trace_xfs_refcount_finish_one_leftover(mp
, agno
, type
,
1128 bno
, blockcount
, new_agbno
, *new_len
);
1132 xfs_trans_brelse(tp
, agbp
);
1138 * Record a refcount intent for later processing.
1142 struct xfs_mount
*mp
,
1143 struct xfs_defer_ops
*dfops
,
1144 enum xfs_refcount_intent_type type
,
1145 xfs_fsblock_t startblock
,
1146 xfs_extlen_t blockcount
)
1148 struct xfs_refcount_intent
*ri
;
1150 trace_xfs_refcount_defer(mp
, XFS_FSB_TO_AGNO(mp
, startblock
),
1151 type
, XFS_FSB_TO_AGBNO(mp
, startblock
),
1154 ri
= kmem_alloc(sizeof(struct xfs_refcount_intent
),
1155 KM_SLEEP
| KM_NOFS
);
1156 INIT_LIST_HEAD(&ri
->ri_list
);
1158 ri
->ri_startblock
= startblock
;
1159 ri
->ri_blockcount
= blockcount
;
1161 xfs_defer_add(dfops
, XFS_DEFER_OPS_TYPE_REFCOUNT
, &ri
->ri_list
);
1166 * Increase the reference count of the blocks backing a file's extent.
1169 xfs_refcount_increase_extent(
1170 struct xfs_mount
*mp
,
1171 struct xfs_defer_ops
*dfops
,
1172 struct xfs_bmbt_irec
*PREV
)
1174 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
1177 return __xfs_refcount_add(mp
, dfops
, XFS_REFCOUNT_INCREASE
,
1178 PREV
->br_startblock
, PREV
->br_blockcount
);
1182 * Decrease the reference count of the blocks backing a file's extent.
1185 xfs_refcount_decrease_extent(
1186 struct xfs_mount
*mp
,
1187 struct xfs_defer_ops
*dfops
,
1188 struct xfs_bmbt_irec
*PREV
)
1190 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
1193 return __xfs_refcount_add(mp
, dfops
, XFS_REFCOUNT_DECREASE
,
1194 PREV
->br_startblock
, PREV
->br_blockcount
);
1198 * Given an AG extent, find the lowest-numbered run of shared blocks
1199 * within that range and return the range in fbno/flen. If
1200 * find_end_of_shared is set, return the longest contiguous extent of
1201 * shared blocks; if not, just return the first extent we find. If no
1202 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1203 * and 0, respectively.
1206 xfs_refcount_find_shared(
1207 struct xfs_btree_cur
*cur
,
1208 xfs_agblock_t agbno
,
1210 xfs_agblock_t
*fbno
,
1212 bool find_end_of_shared
)
1214 struct xfs_refcount_irec tmp
;
1219 trace_xfs_refcount_find_shared(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
1222 /* By default, skip the whole range */
1223 *fbno
= NULLAGBLOCK
;
1226 /* Try to find a refcount extent that crosses the start */
1227 error
= xfs_refcount_lookup_le(cur
, agbno
, &have
);
1231 /* No left extent, look at the next one */
1232 error
= xfs_btree_increment(cur
, 0, &have
);
1238 error
= xfs_refcount_get_rec(cur
, &tmp
, &i
);
1241 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, out_error
);
1243 /* If the extent ends before the start, look at the next one */
1244 if (tmp
.rc_startblock
+ tmp
.rc_blockcount
<= agbno
) {
1245 error
= xfs_btree_increment(cur
, 0, &have
);
1250 error
= xfs_refcount_get_rec(cur
, &tmp
, &i
);
1253 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, out_error
);
1256 /* If the extent starts after the range we want, bail out */
1257 if (tmp
.rc_startblock
>= agbno
+ aglen
)
1260 /* We found the start of a shared extent! */
1261 if (tmp
.rc_startblock
< agbno
) {
1262 tmp
.rc_blockcount
-= (agbno
- tmp
.rc_startblock
);
1263 tmp
.rc_startblock
= agbno
;
1266 *fbno
= tmp
.rc_startblock
;
1267 *flen
= min(tmp
.rc_blockcount
, agbno
+ aglen
- *fbno
);
1268 if (!find_end_of_shared
)
1271 /* Otherwise, find the end of this shared extent */
1272 while (*fbno
+ *flen
< agbno
+ aglen
) {
1273 error
= xfs_btree_increment(cur
, 0, &have
);
1278 error
= xfs_refcount_get_rec(cur
, &tmp
, &i
);
1281 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, out_error
);
1282 if (tmp
.rc_startblock
>= agbno
+ aglen
||
1283 tmp
.rc_startblock
!= *fbno
+ *flen
)
1285 *flen
= min(*flen
+ tmp
.rc_blockcount
, agbno
+ aglen
- *fbno
);
1289 trace_xfs_refcount_find_shared_result(cur
->bc_mp
,
1290 cur
->bc_private
.a
.agno
, *fbno
, *flen
);
1294 trace_xfs_refcount_find_shared_error(cur
->bc_mp
,
1295 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
1300 * Recovering CoW Blocks After a Crash
1302 * Due to the way that the copy on write mechanism works, there's a window of
1303 * opportunity in which we can lose track of allocated blocks during a crash.
1304 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1305 * causes blocks to be allocated and stored in the CoW fork. The blocks are
1306 * no longer in the free space btree but are not otherwise recorded anywhere
1307 * until the write completes and the blocks are mapped into the file. A crash
1308 * in between allocation and remapping results in the replacement blocks being
1309 * lost. This situation is exacerbated by the CoW extent size hint because
1310 * allocations can hang around for long time.
1312 * However, there is a place where we can record these allocations before they
1313 * become mappings -- the reference count btree. The btree does not record
1314 * extents with refcount == 1, so we can record allocations with a refcount of
1315 * 1. Blocks being used for CoW writeout cannot be shared, so there should be
1316 * no conflict with shared block records. These mappings should be created
1317 * when we allocate blocks to the CoW fork and deleted when they're removed
1318 * from the CoW fork.
1320 * Minor nit: records for in-progress CoW allocations and records for shared
1321 * extents must never be merged, to preserve the property that (except for CoW
1322 * allocations) there are no refcount btree entries with refcount == 1. The
1323 * only time this could potentially happen is when unsharing a block that's
1324 * adjacent to CoW allocations, so we must be careful to avoid this.
1326 * At mount time we recover lost CoW allocations by searching the refcount
1327 * btree for these refcount == 1 mappings. These represent CoW allocations
1328 * that were in progress at the time the filesystem went down, so we can free
1329 * them to get the space back.
1331 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1332 * several reasons -- first, EFIs pin the tail of the log and would have to be
1333 * periodically relogged to avoid filling up the log. Second, CoW completions
1334 * will have to file an EFD and create new EFIs for whatever remains in the
1335 * CoW fork; this partially takes care of (1) but extent-size reservations
1336 * will have to periodically relog even if there's no writeout in progress.
1337 * This can happen if the CoW extent size hint is set, which you really want.
1338 * Third, EFIs cannot currently be automatically relogged into newer
1339 * transactions to advance the log tail. Fourth, stuffing the log full of
1340 * EFIs places an upper bound on the number of CoW allocations that can be
1341 * held filesystem-wide at any given time. Recording them in the refcount
1342 * btree doesn't require us to maintain any state in memory and doesn't pin
1346 * Adjust the refcounts of CoW allocations. These allocations are "magic"
1347 * in that they're not referenced anywhere else in the filesystem, so we
1348 * stash them in the refcount btree with a refcount of 1 until either file
1349 * remapping (or CoW cancellation) happens.
1352 xfs_refcount_adjust_cow_extents(
1353 struct xfs_btree_cur
*cur
,
1354 xfs_agblock_t agbno
,
1356 enum xfs_refc_adjust_op adj
)
1358 struct xfs_refcount_irec ext
, tmp
;
1360 int found_rec
, found_tmp
;
1365 /* Find any overlapping refcount records */
1366 error
= xfs_refcount_lookup_ge(cur
, agbno
, &found_rec
);
1369 error
= xfs_refcount_get_rec(cur
, &ext
, &found_rec
);
1373 ext
.rc_startblock
= cur
->bc_mp
->m_sb
.sb_agblocks
+
1375 ext
.rc_blockcount
= 0;
1376 ext
.rc_refcount
= 0;
1380 case XFS_REFCOUNT_ADJUST_COW_ALLOC
:
1381 /* Adding a CoW reservation, there should be nothing here. */
1382 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
1383 ext
.rc_startblock
>= agbno
+ aglen
, out_error
);
1385 tmp
.rc_startblock
= agbno
;
1386 tmp
.rc_blockcount
= aglen
;
1387 tmp
.rc_refcount
= 1;
1388 trace_xfs_refcount_modify_extent(cur
->bc_mp
,
1389 cur
->bc_private
.a
.agno
, &tmp
);
1391 error
= xfs_refcount_insert(cur
, &tmp
,
1395 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
1396 found_tmp
== 1, out_error
);
1398 case XFS_REFCOUNT_ADJUST_COW_FREE
:
1399 /* Removing a CoW reservation, there should be one extent. */
1400 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
1401 ext
.rc_startblock
== agbno
, out_error
);
1402 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
1403 ext
.rc_blockcount
== aglen
, out_error
);
1404 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
1405 ext
.rc_refcount
== 1, out_error
);
1407 ext
.rc_refcount
= 0;
1408 trace_xfs_refcount_modify_extent(cur
->bc_mp
,
1409 cur
->bc_private
.a
.agno
, &ext
);
1410 error
= xfs_refcount_delete(cur
, &found_rec
);
1413 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
,
1414 found_rec
== 1, out_error
);
1422 trace_xfs_refcount_modify_extent_error(cur
->bc_mp
,
1423 cur
->bc_private
.a
.agno
, error
, _RET_IP_
);
1428 * Add or remove refcount btree entries for CoW reservations.
1431 xfs_refcount_adjust_cow(
1432 struct xfs_btree_cur
*cur
,
1433 xfs_agblock_t agbno
,
1435 enum xfs_refc_adjust_op adj
)
1440 agbno
+= XFS_REFC_COW_START
;
1443 * Ensure that no rcextents cross the boundary of the adjustment range.
1445 error
= xfs_refcount_split_extent(cur
, agbno
, &shape_changed
);
1449 error
= xfs_refcount_split_extent(cur
, agbno
+ aglen
, &shape_changed
);
1454 * Try to merge with the left or right extents of the range.
1456 error
= xfs_refcount_merge_extents(cur
, &agbno
, &aglen
, adj
,
1457 XFS_FIND_RCEXT_COW
, &shape_changed
);
1461 /* Now that we've taken care of the ends, adjust the middle extents */
1462 error
= xfs_refcount_adjust_cow_extents(cur
, agbno
, aglen
, adj
);
1469 trace_xfs_refcount_adjust_cow_error(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
1475 * Record a CoW allocation in the refcount btree.
1478 __xfs_refcount_cow_alloc(
1479 struct xfs_btree_cur
*rcur
,
1480 xfs_agblock_t agbno
,
1482 struct xfs_defer_ops
*dfops
)
1484 trace_xfs_refcount_cow_increase(rcur
->bc_mp
, rcur
->bc_private
.a
.agno
,
1487 /* Add refcount btree reservation */
1488 return xfs_refcount_adjust_cow(rcur
, agbno
, aglen
,
1489 XFS_REFCOUNT_ADJUST_COW_ALLOC
);
1493 * Remove a CoW allocation from the refcount btree.
1496 __xfs_refcount_cow_free(
1497 struct xfs_btree_cur
*rcur
,
1498 xfs_agblock_t agbno
,
1500 struct xfs_defer_ops
*dfops
)
1502 trace_xfs_refcount_cow_decrease(rcur
->bc_mp
, rcur
->bc_private
.a
.agno
,
1505 /* Remove refcount btree reservation */
1506 return xfs_refcount_adjust_cow(rcur
, agbno
, aglen
,
1507 XFS_REFCOUNT_ADJUST_COW_FREE
);
1510 /* Record a CoW staging extent in the refcount btree. */
1512 xfs_refcount_alloc_cow_extent(
1513 struct xfs_mount
*mp
,
1514 struct xfs_defer_ops
*dfops
,
1520 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
1523 error
= __xfs_refcount_add(mp
, dfops
, XFS_REFCOUNT_ALLOC_COW
,
1528 /* Add rmap entry */
1529 return xfs_rmap_alloc_extent(mp
, dfops
, XFS_FSB_TO_AGNO(mp
, fsb
),
1530 XFS_FSB_TO_AGBNO(mp
, fsb
), len
, XFS_RMAP_OWN_COW
);
1533 /* Forget a CoW staging event in the refcount btree. */
1535 xfs_refcount_free_cow_extent(
1536 struct xfs_mount
*mp
,
1537 struct xfs_defer_ops
*dfops
,
1543 if (!xfs_sb_version_hasreflink(&mp
->m_sb
))
1546 /* Remove rmap entry */
1547 error
= xfs_rmap_free_extent(mp
, dfops
, XFS_FSB_TO_AGNO(mp
, fsb
),
1548 XFS_FSB_TO_AGBNO(mp
, fsb
), len
, XFS_RMAP_OWN_COW
);
1552 return __xfs_refcount_add(mp
, dfops
, XFS_REFCOUNT_FREE_COW
,
1556 struct xfs_refcount_recovery
{
1557 struct list_head rr_list
;
1558 struct xfs_refcount_irec rr_rrec
;
1561 /* Stuff an extent on the recovery list. */
1563 xfs_refcount_recover_extent(
1564 struct xfs_btree_cur
*cur
,
1565 union xfs_btree_rec
*rec
,
1568 struct list_head
*debris
= priv
;
1569 struct xfs_refcount_recovery
*rr
;
1571 if (be32_to_cpu(rec
->refc
.rc_refcount
) != 1)
1572 return -EFSCORRUPTED
;
1574 rr
= kmem_alloc(sizeof(struct xfs_refcount_recovery
), KM_SLEEP
);
1575 xfs_refcount_btrec_to_irec(rec
, &rr
->rr_rrec
);
1576 list_add_tail(&rr
->rr_list
, debris
);
1581 /* Find and remove leftover CoW reservations. */
1583 xfs_refcount_recover_cow_leftovers(
1584 struct xfs_mount
*mp
,
1585 xfs_agnumber_t agno
)
1587 struct xfs_trans
*tp
;
1588 struct xfs_btree_cur
*cur
;
1589 struct xfs_buf
*agbp
;
1590 struct xfs_refcount_recovery
*rr
, *n
;
1591 struct list_head debris
;
1592 union xfs_btree_irec low
;
1593 union xfs_btree_irec high
;
1594 struct xfs_defer_ops dfops
;
1596 xfs_agblock_t agbno
;
1599 if (mp
->m_sb
.sb_agblocks
>= XFS_REFC_COW_START
)
1602 INIT_LIST_HEAD(&debris
);
1605 * In this first part, we use an empty transaction to gather up
1606 * all the leftover CoW extents so that we can subsequently
1607 * delete them. The empty transaction is used to avoid
1608 * a buffer lock deadlock if there happens to be a loop in the
1609 * refcountbt because we're allowed to re-grab a buffer that is
1610 * already attached to our transaction. When we're done
1611 * recording the CoW debris we cancel the (empty) transaction
1612 * and everything goes away cleanly.
1614 error
= xfs_trans_alloc_empty(mp
, &tp
);
1618 error
= xfs_alloc_read_agf(mp
, tp
, agno
, 0, &agbp
);
1625 cur
= xfs_refcountbt_init_cursor(mp
, tp
, agbp
, agno
, NULL
);
1627 /* Find all the leftover CoW staging extents. */
1628 memset(&low
, 0, sizeof(low
));
1629 memset(&high
, 0, sizeof(high
));
1630 low
.rc
.rc_startblock
= XFS_REFC_COW_START
;
1631 high
.rc
.rc_startblock
= -1U;
1632 error
= xfs_btree_query_range(cur
, &low
, &high
,
1633 xfs_refcount_recover_extent
, &debris
);
1636 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1637 xfs_trans_brelse(tp
, agbp
);
1638 xfs_trans_cancel(tp
);
1640 /* Now iterate the list to free the leftovers */
1641 list_for_each_entry_safe(rr
, n
, &debris
, rr_list
) {
1642 /* Set up transaction. */
1643 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, 0, 0, 0, &tp
);
1647 trace_xfs_refcount_recover_extent(mp
, agno
, &rr
->rr_rrec
);
1649 /* Free the orphan record */
1650 xfs_defer_init(&dfops
, &fsb
);
1651 agbno
= rr
->rr_rrec
.rc_startblock
- XFS_REFC_COW_START
;
1652 fsb
= XFS_AGB_TO_FSB(mp
, agno
, agbno
);
1653 error
= xfs_refcount_free_cow_extent(mp
, &dfops
, fsb
,
1654 rr
->rr_rrec
.rc_blockcount
);
1658 /* Free the block. */
1659 xfs_bmap_add_free(mp
, &dfops
, fsb
,
1660 rr
->rr_rrec
.rc_blockcount
, NULL
);
1662 error
= xfs_defer_finish(&tp
, &dfops
);
1666 error
= xfs_trans_commit(tp
);
1670 list_del(&rr
->rr_list
);
1676 xfs_defer_cancel(&dfops
);
1678 xfs_trans_cancel(tp
);
1680 /* Free the leftover list */
1681 list_for_each_entry_safe(rr
, n
, &debris
, rr_list
) {
1682 list_del(&rr
->rr_list
);
1688 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1689 xfs_trans_brelse(tp
, agbp
);
1693 /* Is there a record covering a given extent? */
1695 xfs_refcount_has_record(
1696 struct xfs_btree_cur
*cur
,
1701 union xfs_btree_irec low
;
1702 union xfs_btree_irec high
;
1704 memset(&low
, 0, sizeof(low
));
1705 low
.rc
.rc_startblock
= bno
;
1706 memset(&high
, 0xFF, sizeof(high
));
1707 high
.rc
.rc_startblock
= bno
+ len
- 1;
1709 return xfs_btree_has_record(cur
, &low
, &high
, exists
);