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"
27 struct kmem_cache *xfs_refcount_intent_cache;
29 /* Allowable refcount adjustment amounts. */
30 enum xfs_refc_adjust_op {
31 XFS_REFCOUNT_ADJUST_INCREASE = 1,
32 XFS_REFCOUNT_ADJUST_DECREASE = -1,
33 XFS_REFCOUNT_ADJUST_COW_ALLOC = 0,
34 XFS_REFCOUNT_ADJUST_COW_FREE = -1,
37 STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
38 xfs_agblock_t agbno, xfs_extlen_t aglen);
39 STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
40 xfs_agblock_t agbno, xfs_extlen_t aglen);
43 * Look up the first record less than or equal to [bno, len] in the btree
47 xfs_refcount_lookup_le(
48 struct xfs_btree_cur *cur,
49 enum xfs_refc_domain domain,
53 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.pag->pag_agno,
54 xfs_refcount_encode_startblock(bno, domain),
56 cur->bc_rec.rc.rc_startblock = bno;
57 cur->bc_rec.rc.rc_blockcount = 0;
58 cur->bc_rec.rc.rc_domain = domain;
59 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
63 * Look up the first record greater than or equal to [bno, len] in the btree
67 xfs_refcount_lookup_ge(
68 struct xfs_btree_cur *cur,
69 enum xfs_refc_domain domain,
73 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.pag->pag_agno,
74 xfs_refcount_encode_startblock(bno, domain),
76 cur->bc_rec.rc.rc_startblock = bno;
77 cur->bc_rec.rc.rc_blockcount = 0;
78 cur->bc_rec.rc.rc_domain = domain;
79 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
83 * Look up the first record equal to [bno, len] in the btree
87 xfs_refcount_lookup_eq(
88 struct xfs_btree_cur *cur,
89 enum xfs_refc_domain domain,
93 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_ag.pag->pag_agno,
94 xfs_refcount_encode_startblock(bno, domain),
96 cur->bc_rec.rc.rc_startblock = bno;
97 cur->bc_rec.rc.rc_blockcount = 0;
98 cur->bc_rec.rc.rc_domain = domain;
99 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
102 /* Convert on-disk record to in-core format. */
104 xfs_refcount_btrec_to_irec(
105 const union xfs_btree_rec *rec,
106 struct xfs_refcount_irec *irec)
110 start = be32_to_cpu(rec->refc.rc_startblock);
111 if (start & XFS_REFC_COWFLAG) {
112 start &= ~XFS_REFC_COWFLAG;
113 irec->rc_domain = XFS_REFC_DOMAIN_COW;
115 irec->rc_domain = XFS_REFC_DOMAIN_SHARED;
118 irec->rc_startblock = start;
119 irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
120 irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
123 /* Simple checks for refcount records. */
125 xfs_refcount_check_irec(
126 struct xfs_btree_cur *cur,
127 const struct xfs_refcount_irec *irec)
129 struct xfs_perag *pag = cur->bc_ag.pag;
131 if (irec->rc_blockcount == 0 || irec->rc_blockcount > MAXREFCEXTLEN)
132 return __this_address;
134 if (!xfs_refcount_check_domain(irec))
135 return __this_address;
137 /* check for valid extent range, including overflow */
138 if (!xfs_verify_agbext(pag, irec->rc_startblock, irec->rc_blockcount))
139 return __this_address;
141 if (irec->rc_refcount == 0 || irec->rc_refcount > MAXREFCOUNT)
142 return __this_address;
148 xfs_refcount_complain_bad_rec(
149 struct xfs_btree_cur *cur,
151 const struct xfs_refcount_irec *irec)
153 struct xfs_mount *mp = cur->bc_mp;
156 "Refcount BTree record corruption in AG %d detected at %pS!",
157 cur->bc_ag.pag->pag_agno, fa);
159 "Start block 0x%x, block count 0x%x, references 0x%x",
160 irec->rc_startblock, irec->rc_blockcount, irec->rc_refcount);
161 return -EFSCORRUPTED;
165 * Get the data from the pointed-to record.
168 xfs_refcount_get_rec(
169 struct xfs_btree_cur *cur,
170 struct xfs_refcount_irec *irec,
173 union xfs_btree_rec *rec;
177 error = xfs_btree_get_rec(cur, &rec, stat);
181 xfs_refcount_btrec_to_irec(rec, irec);
182 fa = xfs_refcount_check_irec(cur, irec);
184 return xfs_refcount_complain_bad_rec(cur, fa, irec);
186 trace_xfs_refcount_get(cur->bc_mp, cur->bc_ag.pag->pag_agno, irec);
191 * Update the record referred to by cur to the value given
192 * by [bno, len, refcount].
193 * This either works (return 0) or gets an EFSCORRUPTED error.
197 struct xfs_btree_cur *cur,
198 struct xfs_refcount_irec *irec)
200 union xfs_btree_rec rec;
204 trace_xfs_refcount_update(cur->bc_mp, cur->bc_ag.pag->pag_agno, irec);
206 start = xfs_refcount_encode_startblock(irec->rc_startblock,
208 rec.refc.rc_startblock = cpu_to_be32(start);
209 rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
210 rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
212 error = xfs_btree_update(cur, &rec);
214 trace_xfs_refcount_update_error(cur->bc_mp,
215 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
220 * Insert the record referred to by cur to the value given
221 * by [bno, len, refcount].
222 * This either works (return 0) or gets an EFSCORRUPTED error.
226 struct xfs_btree_cur *cur,
227 struct xfs_refcount_irec *irec,
232 trace_xfs_refcount_insert(cur->bc_mp, cur->bc_ag.pag->pag_agno, irec);
234 cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
235 cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
236 cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
237 cur->bc_rec.rc.rc_domain = irec->rc_domain;
239 error = xfs_btree_insert(cur, i);
242 if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
243 error = -EFSCORRUPTED;
249 trace_xfs_refcount_insert_error(cur->bc_mp,
250 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
255 * Remove the record referred to by cur, then set the pointer to the spot
256 * where the record could be re-inserted, in case we want to increment or
257 * decrement the cursor.
258 * This either works (return 0) or gets an EFSCORRUPTED error.
262 struct xfs_btree_cur *cur,
265 struct xfs_refcount_irec irec;
269 error = xfs_refcount_get_rec(cur, &irec, &found_rec);
272 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
273 error = -EFSCORRUPTED;
276 trace_xfs_refcount_delete(cur->bc_mp, cur->bc_ag.pag->pag_agno, &irec);
277 error = xfs_btree_delete(cur, i);
278 if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
279 error = -EFSCORRUPTED;
284 error = xfs_refcount_lookup_ge(cur, irec.rc_domain, irec.rc_startblock,
288 trace_xfs_refcount_delete_error(cur->bc_mp,
289 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
294 * Adjusting the Reference Count
296 * As stated elsewhere, the reference count btree (refcbt) stores
297 * >1 reference counts for extents of physical blocks. In this
298 * operation, we're either raising or lowering the reference count of
299 * some subrange stored in the tree:
301 * <------ adjustment range ------>
302 * ----+ +---+-----+ +--+--------+---------
303 * 2 | | 3 | 4 | |17| 55 | 10
304 * ----+ +---+-----+ +--+--------+---------
305 * X axis is physical blocks number;
306 * reference counts are the numbers inside the rectangles
308 * The first thing we need to do is to ensure that there are no
309 * refcount extents crossing either boundary of the range to be
310 * adjusted. For any extent that does cross a boundary, split it into
311 * two extents so that we can increment the refcount of one of the
314 * <------ adjustment range ------>
315 * ----+ +---+-----+ +--+--------+----+----
316 * 2 | | 3 | 2 | |17| 55 | 10 | 10
317 * ----+ +---+-----+ +--+--------+----+----
319 * For this next step, let's assume that all the physical blocks in
320 * the adjustment range are mapped to a file and are therefore in use
321 * at least once. Therefore, we can infer that any gap in the
322 * refcount tree within the adjustment range represents a physical
323 * extent with refcount == 1:
325 * <------ adjustment range ------>
326 * ----+---+---+-----+-+--+--------+----+----
327 * 2 |"1"| 3 | 2 |1|17| 55 | 10 | 10
328 * ----+---+---+-----+-+--+--------+----+----
331 * For each extent that falls within the interval range, figure out
332 * which extent is to the left or the right of that extent. Now we
333 * have a left, current, and right extent. If the new reference count
334 * of the center extent enables us to merge left, center, and right
335 * into one record covering all three, do so. If the center extent is
336 * at the left end of the range, abuts the left extent, and its new
337 * reference count matches the left extent's record, then merge them.
338 * If the center extent is at the right end of the range, abuts the
339 * right extent, and the reference counts match, merge those. In the
340 * example, we can left merge (assuming an increment operation):
342 * <------ adjustment range ------>
343 * --------+---+-----+-+--+--------+----+----
344 * 2 | 3 | 2 |1|17| 55 | 10 | 10
345 * --------+---+-----+-+--+--------+----+----
348 * For all other extents within the range, adjust the reference count
349 * or delete it if the refcount falls below 2. If we were
350 * incrementing, the end result looks like this:
352 * <------ adjustment range ------>
353 * --------+---+-----+-+--+--------+----+----
354 * 2 | 4 | 3 |2|18| 56 | 11 | 10
355 * --------+---+-----+-+--+--------+----+----
357 * The result of a decrement operation looks as such:
359 * <------ adjustment range ------>
360 * ----+ +---+ +--+--------+----+----
361 * 2 | | 2 | |16| 54 | 9 | 10
362 * ----+ +---+ +--+--------+----+----
365 * The blocks marked "D" are freed; the blocks marked "1" are only
366 * referenced once and therefore the record is removed from the
370 /* Next block after this extent. */
371 static inline xfs_agblock_t
373 struct xfs_refcount_irec *rc)
375 return rc->rc_startblock + rc->rc_blockcount;
379 * Split a refcount extent that crosses agbno.
382 xfs_refcount_split_extent(
383 struct xfs_btree_cur *cur,
384 enum xfs_refc_domain domain,
388 struct xfs_refcount_irec rcext, tmp;
392 *shape_changed = false;
393 error = xfs_refcount_lookup_le(cur, domain, agbno, &found_rec);
399 error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
402 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
403 error = -EFSCORRUPTED;
406 if (rcext.rc_domain != domain)
408 if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
411 *shape_changed = true;
412 trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_ag.pag->pag_agno,
415 /* Establish the right extent. */
417 tmp.rc_startblock = agbno;
418 tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
419 error = xfs_refcount_update(cur, &tmp);
423 /* Insert the left extent. */
425 tmp.rc_blockcount = agbno - rcext.rc_startblock;
426 error = xfs_refcount_insert(cur, &tmp, &found_rec);
429 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
430 error = -EFSCORRUPTED;
436 trace_xfs_refcount_split_extent_error(cur->bc_mp,
437 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
442 * Merge the left, center, and right extents.
445 xfs_refcount_merge_center_extents(
446 struct xfs_btree_cur *cur,
447 struct xfs_refcount_irec *left,
448 struct xfs_refcount_irec *center,
449 struct xfs_refcount_irec *right,
450 unsigned long long extlen,
456 trace_xfs_refcount_merge_center_extents(cur->bc_mp,
457 cur->bc_ag.pag->pag_agno, left, center, right);
459 ASSERT(left->rc_domain == center->rc_domain);
460 ASSERT(right->rc_domain == center->rc_domain);
463 * Make sure the center and right extents are not in the btree.
464 * If the center extent was synthesized, the first delete call
465 * removes the right extent and we skip the second deletion.
466 * If center and right were in the btree, then the first delete
467 * call removes the center and the second one removes the right
470 error = xfs_refcount_lookup_ge(cur, center->rc_domain,
471 center->rc_startblock, &found_rec);
474 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
475 error = -EFSCORRUPTED;
479 error = xfs_refcount_delete(cur, &found_rec);
482 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
483 error = -EFSCORRUPTED;
487 if (center->rc_refcount > 1) {
488 error = xfs_refcount_delete(cur, &found_rec);
491 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
492 error = -EFSCORRUPTED;
497 /* Enlarge the left extent. */
498 error = xfs_refcount_lookup_le(cur, left->rc_domain,
499 left->rc_startblock, &found_rec);
502 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
503 error = -EFSCORRUPTED;
507 left->rc_blockcount = extlen;
508 error = xfs_refcount_update(cur, left);
516 trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
517 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
522 * Merge with the left extent.
525 xfs_refcount_merge_left_extent(
526 struct xfs_btree_cur *cur,
527 struct xfs_refcount_irec *left,
528 struct xfs_refcount_irec *cleft,
529 xfs_agblock_t *agbno,
535 trace_xfs_refcount_merge_left_extent(cur->bc_mp,
536 cur->bc_ag.pag->pag_agno, left, cleft);
538 ASSERT(left->rc_domain == cleft->rc_domain);
540 /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
541 if (cleft->rc_refcount > 1) {
542 error = xfs_refcount_lookup_le(cur, cleft->rc_domain,
543 cleft->rc_startblock, &found_rec);
546 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
547 error = -EFSCORRUPTED;
551 error = xfs_refcount_delete(cur, &found_rec);
554 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
555 error = -EFSCORRUPTED;
560 /* Enlarge the left extent. */
561 error = xfs_refcount_lookup_le(cur, left->rc_domain,
562 left->rc_startblock, &found_rec);
565 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
566 error = -EFSCORRUPTED;
570 left->rc_blockcount += cleft->rc_blockcount;
571 error = xfs_refcount_update(cur, left);
575 *agbno += cleft->rc_blockcount;
576 *aglen -= cleft->rc_blockcount;
580 trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
581 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
586 * Merge with the right extent.
589 xfs_refcount_merge_right_extent(
590 struct xfs_btree_cur *cur,
591 struct xfs_refcount_irec *right,
592 struct xfs_refcount_irec *cright,
598 trace_xfs_refcount_merge_right_extent(cur->bc_mp,
599 cur->bc_ag.pag->pag_agno, cright, right);
601 ASSERT(right->rc_domain == cright->rc_domain);
604 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
607 if (cright->rc_refcount > 1) {
608 error = xfs_refcount_lookup_le(cur, cright->rc_domain,
609 cright->rc_startblock, &found_rec);
612 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
613 error = -EFSCORRUPTED;
617 error = xfs_refcount_delete(cur, &found_rec);
620 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
621 error = -EFSCORRUPTED;
626 /* Enlarge the right extent. */
627 error = xfs_refcount_lookup_le(cur, right->rc_domain,
628 right->rc_startblock, &found_rec);
631 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
632 error = -EFSCORRUPTED;
636 right->rc_startblock -= cright->rc_blockcount;
637 right->rc_blockcount += cright->rc_blockcount;
638 error = xfs_refcount_update(cur, right);
642 *aglen -= cright->rc_blockcount;
646 trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
647 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
652 * Find the left extent and the one after it (cleft). This function assumes
653 * that we've already split any extent crossing agbno.
656 xfs_refcount_find_left_extents(
657 struct xfs_btree_cur *cur,
658 struct xfs_refcount_irec *left,
659 struct xfs_refcount_irec *cleft,
660 enum xfs_refc_domain domain,
664 struct xfs_refcount_irec tmp;
668 left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
669 error = xfs_refcount_lookup_le(cur, domain, agbno - 1, &found_rec);
675 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
678 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
679 error = -EFSCORRUPTED;
683 if (tmp.rc_domain != domain)
685 if (xfs_refc_next(&tmp) != agbno)
687 /* We have a left extent; retrieve (or invent) the next right one */
690 error = xfs_btree_increment(cur, 0, &found_rec);
694 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
697 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
698 error = -EFSCORRUPTED;
702 if (tmp.rc_domain != domain)
705 /* if tmp starts at the end of our range, just use that */
706 if (tmp.rc_startblock == agbno)
710 * There's a gap in the refcntbt at the start of the
711 * range we're interested in (refcount == 1) so
712 * synthesize the implied extent and pass it back.
713 * We assume here that the agbno/aglen range was
714 * passed in from a data fork extent mapping and
715 * therefore is allocated to exactly one owner.
717 cleft->rc_startblock = agbno;
718 cleft->rc_blockcount = min(aglen,
719 tmp.rc_startblock - agbno);
720 cleft->rc_refcount = 1;
721 cleft->rc_domain = domain;
726 * No extents, so pretend that there's one covering the whole
729 cleft->rc_startblock = agbno;
730 cleft->rc_blockcount = aglen;
731 cleft->rc_refcount = 1;
732 cleft->rc_domain = domain;
734 trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_ag.pag->pag_agno,
739 trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
740 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
745 * Find the right extent and the one before it (cright). This function
746 * assumes that we've already split any extents crossing agbno + aglen.
749 xfs_refcount_find_right_extents(
750 struct xfs_btree_cur *cur,
751 struct xfs_refcount_irec *right,
752 struct xfs_refcount_irec *cright,
753 enum xfs_refc_domain domain,
757 struct xfs_refcount_irec tmp;
761 right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
762 error = xfs_refcount_lookup_ge(cur, domain, agbno + aglen, &found_rec);
768 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
771 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
772 error = -EFSCORRUPTED;
776 if (tmp.rc_domain != domain)
778 if (tmp.rc_startblock != agbno + aglen)
780 /* We have a right extent; retrieve (or invent) the next left one */
783 error = xfs_btree_decrement(cur, 0, &found_rec);
787 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
790 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
791 error = -EFSCORRUPTED;
795 if (tmp.rc_domain != domain)
798 /* if tmp ends at the end of our range, just use that */
799 if (xfs_refc_next(&tmp) == agbno + aglen)
803 * There's a gap in the refcntbt at the end of the
804 * range we're interested in (refcount == 1) so
805 * create the implied extent and pass it back.
806 * We assume here that the agbno/aglen range was
807 * passed in from a data fork extent mapping and
808 * therefore is allocated to exactly one owner.
810 cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
811 cright->rc_blockcount = right->rc_startblock -
812 cright->rc_startblock;
813 cright->rc_refcount = 1;
814 cright->rc_domain = domain;
819 * No extents, so pretend that there's one covering the whole
822 cright->rc_startblock = agbno;
823 cright->rc_blockcount = aglen;
824 cright->rc_refcount = 1;
825 cright->rc_domain = domain;
827 trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_ag.pag->pag_agno,
828 cright, right, agbno + aglen);
832 trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
833 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
837 /* Is this extent valid? */
840 const struct xfs_refcount_irec *rc)
842 return rc->rc_startblock != NULLAGBLOCK;
845 static inline xfs_nlink_t
846 xfs_refc_merge_refcount(
847 const struct xfs_refcount_irec *irec,
848 enum xfs_refc_adjust_op adjust)
850 /* Once a record hits MAXREFCOUNT, it is pinned there forever */
851 if (irec->rc_refcount == MAXREFCOUNT)
853 return irec->rc_refcount + adjust;
857 xfs_refc_want_merge_center(
858 const struct xfs_refcount_irec *left,
859 const struct xfs_refcount_irec *cleft,
860 const struct xfs_refcount_irec *cright,
861 const struct xfs_refcount_irec *right,
862 bool cleft_is_cright,
863 enum xfs_refc_adjust_op adjust,
864 unsigned long long *ulenp)
866 unsigned long long ulen = left->rc_blockcount;
867 xfs_nlink_t new_refcount;
870 * To merge with a center record, both shoulder records must be
871 * adjacent to the record we want to adjust. This is only true if
872 * find_left and find_right made all four records valid.
874 if (!xfs_refc_valid(left) || !xfs_refc_valid(right) ||
875 !xfs_refc_valid(cleft) || !xfs_refc_valid(cright))
878 /* There must only be one record for the entire range. */
879 if (!cleft_is_cright)
882 /* The shoulder record refcounts must match the new refcount. */
883 new_refcount = xfs_refc_merge_refcount(cleft, adjust);
884 if (left->rc_refcount != new_refcount)
886 if (right->rc_refcount != new_refcount)
890 * The new record cannot exceed the max length. ulen is a ULL as the
891 * individual record block counts can be up to (u32 - 1) in length
892 * hence we need to catch u32 addition overflows here.
894 ulen += cleft->rc_blockcount + right->rc_blockcount;
895 if (ulen >= MAXREFCEXTLEN)
903 xfs_refc_want_merge_left(
904 const struct xfs_refcount_irec *left,
905 const struct xfs_refcount_irec *cleft,
906 enum xfs_refc_adjust_op adjust)
908 unsigned long long ulen = left->rc_blockcount;
909 xfs_nlink_t new_refcount;
912 * For a left merge, the left shoulder record must be adjacent to the
913 * start of the range. If this is true, find_left made left and cleft
914 * contain valid contents.
916 if (!xfs_refc_valid(left) || !xfs_refc_valid(cleft))
919 /* Left shoulder record refcount must match the new refcount. */
920 new_refcount = xfs_refc_merge_refcount(cleft, adjust);
921 if (left->rc_refcount != new_refcount)
925 * The new record cannot exceed the max length. ulen is a ULL as the
926 * individual record block counts can be up to (u32 - 1) in length
927 * hence we need to catch u32 addition overflows here.
929 ulen += cleft->rc_blockcount;
930 if (ulen >= MAXREFCEXTLEN)
937 xfs_refc_want_merge_right(
938 const struct xfs_refcount_irec *cright,
939 const struct xfs_refcount_irec *right,
940 enum xfs_refc_adjust_op adjust)
942 unsigned long long ulen = right->rc_blockcount;
943 xfs_nlink_t new_refcount;
946 * For a right merge, the right shoulder record must be adjacent to the
947 * end of the range. If this is true, find_right made cright and right
948 * contain valid contents.
950 if (!xfs_refc_valid(right) || !xfs_refc_valid(cright))
953 /* Right shoulder record refcount must match the new refcount. */
954 new_refcount = xfs_refc_merge_refcount(cright, adjust);
955 if (right->rc_refcount != new_refcount)
959 * The new record cannot exceed the max length. ulen is a ULL as the
960 * individual record block counts can be up to (u32 - 1) in length
961 * hence we need to catch u32 addition overflows here.
963 ulen += cright->rc_blockcount;
964 if (ulen >= MAXREFCEXTLEN)
971 * Try to merge with any extents on the boundaries of the adjustment range.
974 xfs_refcount_merge_extents(
975 struct xfs_btree_cur *cur,
976 enum xfs_refc_domain domain,
977 xfs_agblock_t *agbno,
979 enum xfs_refc_adjust_op adjust,
982 struct xfs_refcount_irec left = {0}, cleft = {0};
983 struct xfs_refcount_irec cright = {0}, right = {0};
985 unsigned long long ulen;
988 *shape_changed = false;
990 * Find the extent just below agbno [left], just above agbno [cleft],
991 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
994 error = xfs_refcount_find_left_extents(cur, &left, &cleft, domain,
998 error = xfs_refcount_find_right_extents(cur, &right, &cright, domain,
1003 /* No left or right extent to merge; exit. */
1004 if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
1007 cequal = (cleft.rc_startblock == cright.rc_startblock) &&
1008 (cleft.rc_blockcount == cright.rc_blockcount);
1010 /* Try to merge left, cleft, and right. cleft must == cright. */
1011 if (xfs_refc_want_merge_center(&left, &cleft, &cright, &right, cequal,
1013 *shape_changed = true;
1014 return xfs_refcount_merge_center_extents(cur, &left, &cleft,
1015 &right, ulen, aglen);
1018 /* Try to merge left and cleft. */
1019 if (xfs_refc_want_merge_left(&left, &cleft, adjust)) {
1020 *shape_changed = true;
1021 error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
1027 * If we just merged left + cleft and cleft == cright,
1028 * we no longer have a cright to merge with right. We're done.
1034 /* Try to merge cright and right. */
1035 if (xfs_refc_want_merge_right(&cright, &right, adjust)) {
1036 *shape_changed = true;
1037 return xfs_refcount_merge_right_extent(cur, &right, &cright,
1045 * XXX: This is a pretty hand-wavy estimate. The penalty for guessing
1046 * true incorrectly is a shutdown FS; the penalty for guessing false
1047 * incorrectly is more transaction rolls than might be necessary.
1048 * Be conservative here.
1051 xfs_refcount_still_have_space(
1052 struct xfs_btree_cur *cur)
1054 unsigned long overhead;
1057 * Worst case estimate: full splits of the free space and rmap btrees
1058 * to handle each of the shape changes to the refcount btree.
1060 overhead = xfs_allocfree_block_count(cur->bc_mp,
1061 cur->bc_ag.refc.shape_changes);
1062 overhead += cur->bc_mp->m_refc_maxlevels;
1063 overhead *= cur->bc_mp->m_sb.sb_blocksize;
1066 * Only allow 2 refcount extent updates per transaction if the
1067 * refcount continue update "error" has been injected.
1069 if (cur->bc_ag.refc.nr_ops > 2 &&
1070 XFS_TEST_ERROR(false, cur->bc_mp,
1071 XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
1074 if (cur->bc_ag.refc.nr_ops == 0)
1076 else if (overhead > cur->bc_tp->t_log_res)
1078 return cur->bc_tp->t_log_res - overhead >
1079 cur->bc_ag.refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
1083 * Adjust the refcounts of middle extents. At this point we should have
1084 * split extents that crossed the adjustment range; merged with adjacent
1085 * extents; and updated agbno/aglen to reflect the merges. Therefore,
1086 * all we have to do is update the extents inside [agbno, agbno + aglen].
1089 xfs_refcount_adjust_extents(
1090 struct xfs_btree_cur *cur,
1091 xfs_agblock_t *agbno,
1092 xfs_extlen_t *aglen,
1093 enum xfs_refc_adjust_op adj)
1095 struct xfs_refcount_irec ext, tmp;
1097 int found_rec, found_tmp;
1098 xfs_fsblock_t fsbno;
1100 /* Merging did all the work already. */
1104 error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_SHARED, *agbno,
1109 while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
1110 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1113 if (!found_rec || ext.rc_domain != XFS_REFC_DOMAIN_SHARED) {
1114 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
1115 ext.rc_blockcount = 0;
1116 ext.rc_refcount = 0;
1117 ext.rc_domain = XFS_REFC_DOMAIN_SHARED;
1121 * Deal with a hole in the refcount tree; if a file maps to
1122 * these blocks and there's no refcountbt record, pretend that
1123 * there is one with refcount == 1.
1125 if (ext.rc_startblock != *agbno) {
1126 tmp.rc_startblock = *agbno;
1127 tmp.rc_blockcount = min(*aglen,
1128 ext.rc_startblock - *agbno);
1129 tmp.rc_refcount = 1 + adj;
1130 tmp.rc_domain = XFS_REFC_DOMAIN_SHARED;
1132 trace_xfs_refcount_modify_extent(cur->bc_mp,
1133 cur->bc_ag.pag->pag_agno, &tmp);
1136 * Either cover the hole (increment) or
1137 * delete the range (decrement).
1139 cur->bc_ag.refc.nr_ops++;
1140 if (tmp.rc_refcount) {
1141 error = xfs_refcount_insert(cur, &tmp,
1145 if (XFS_IS_CORRUPT(cur->bc_mp,
1147 error = -EFSCORRUPTED;
1151 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1152 cur->bc_ag.pag->pag_agno,
1154 error = xfs_free_extent_later(cur->bc_tp, fsbno,
1155 tmp.rc_blockcount, NULL,
1161 (*agbno) += tmp.rc_blockcount;
1162 (*aglen) -= tmp.rc_blockcount;
1164 /* Stop if there's nothing left to modify */
1165 if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
1168 /* Move the cursor to the start of ext. */
1169 error = xfs_refcount_lookup_ge(cur,
1170 XFS_REFC_DOMAIN_SHARED, *agbno,
1177 * A previous step trimmed agbno/aglen such that the end of the
1178 * range would not be in the middle of the record. If this is
1179 * no longer the case, something is seriously wrong with the
1180 * btree. Make sure we never feed the synthesized record into
1181 * the processing loop below.
1183 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount == 0) ||
1184 XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount > *aglen)) {
1185 error = -EFSCORRUPTED;
1190 * Adjust the reference count and either update the tree
1191 * (incr) or free the blocks (decr).
1193 if (ext.rc_refcount == MAXREFCOUNT)
1195 ext.rc_refcount += adj;
1196 trace_xfs_refcount_modify_extent(cur->bc_mp,
1197 cur->bc_ag.pag->pag_agno, &ext);
1198 cur->bc_ag.refc.nr_ops++;
1199 if (ext.rc_refcount > 1) {
1200 error = xfs_refcount_update(cur, &ext);
1203 } else if (ext.rc_refcount == 1) {
1204 error = xfs_refcount_delete(cur, &found_rec);
1207 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1208 error = -EFSCORRUPTED;
1213 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1214 cur->bc_ag.pag->pag_agno,
1216 error = xfs_free_extent_later(cur->bc_tp, fsbno,
1217 ext.rc_blockcount, NULL,
1224 error = xfs_btree_increment(cur, 0, &found_rec);
1229 (*agbno) += ext.rc_blockcount;
1230 (*aglen) -= ext.rc_blockcount;
1235 trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1236 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
1240 /* Adjust the reference count of a range of AG blocks. */
1242 xfs_refcount_adjust(
1243 struct xfs_btree_cur *cur,
1244 xfs_agblock_t *agbno,
1245 xfs_extlen_t *aglen,
1246 enum xfs_refc_adjust_op adj)
1249 int shape_changes = 0;
1252 if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
1253 trace_xfs_refcount_increase(cur->bc_mp,
1254 cur->bc_ag.pag->pag_agno, *agbno, *aglen);
1256 trace_xfs_refcount_decrease(cur->bc_mp,
1257 cur->bc_ag.pag->pag_agno, *agbno, *aglen);
1260 * Ensure that no rcextents cross the boundary of the adjustment range.
1262 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1263 *agbno, &shape_changed);
1269 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1270 *agbno + *aglen, &shape_changed);
1277 * Try to merge with the left or right extents of the range.
1279 error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_SHARED,
1280 agbno, aglen, adj, &shape_changed);
1286 cur->bc_ag.refc.shape_changes++;
1288 /* Now that we've taken care of the ends, adjust the middle extents */
1289 error = xfs_refcount_adjust_extents(cur, agbno, aglen, adj);
1296 trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1301 /* Clean up after calling xfs_refcount_finish_one. */
1303 xfs_refcount_finish_one_cleanup(
1304 struct xfs_trans *tp,
1305 struct xfs_btree_cur *rcur,
1308 struct xfs_buf *agbp;
1312 agbp = rcur->bc_ag.agbp;
1313 xfs_btree_del_cursor(rcur, error);
1315 xfs_trans_brelse(tp, agbp);
1319 * Set up a continuation a deferred refcount operation by updating the intent.
1320 * Checks to make sure we're not going to run off the end of the AG.
1323 xfs_refcount_continue_op(
1324 struct xfs_btree_cur *cur,
1325 struct xfs_refcount_intent *ri,
1326 xfs_agblock_t new_agbno)
1328 struct xfs_mount *mp = cur->bc_mp;
1329 struct xfs_perag *pag = cur->bc_ag.pag;
1331 if (XFS_IS_CORRUPT(mp, !xfs_verify_agbext(pag, new_agbno,
1332 ri->ri_blockcount)))
1333 return -EFSCORRUPTED;
1335 ri->ri_startblock = XFS_AGB_TO_FSB(mp, pag->pag_agno, new_agbno);
1337 ASSERT(xfs_verify_fsbext(mp, ri->ri_startblock, ri->ri_blockcount));
1338 ASSERT(pag->pag_agno == XFS_FSB_TO_AGNO(mp, ri->ri_startblock));
1344 * Process one of the deferred refcount operations. We pass back the
1345 * btree cursor to maintain our lock on the btree between calls.
1346 * This saves time and eliminates a buffer deadlock between the
1347 * superblock and the AGF because we'll always grab them in the same
1351 xfs_refcount_finish_one(
1352 struct xfs_trans *tp,
1353 struct xfs_refcount_intent *ri,
1354 struct xfs_btree_cur **pcur)
1356 struct xfs_mount *mp = tp->t_mountp;
1357 struct xfs_btree_cur *rcur;
1358 struct xfs_buf *agbp = NULL;
1361 unsigned long nr_ops = 0;
1362 int shape_changes = 0;
1364 bno = XFS_FSB_TO_AGBNO(mp, ri->ri_startblock);
1366 trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, ri->ri_startblock),
1367 ri->ri_type, XFS_FSB_TO_AGBNO(mp, ri->ri_startblock),
1370 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1374 * If we haven't gotten a cursor or the cursor AG doesn't match
1375 * the startblock, get one now.
1378 if (rcur != NULL && rcur->bc_ag.pag != ri->ri_pag) {
1379 nr_ops = rcur->bc_ag.refc.nr_ops;
1380 shape_changes = rcur->bc_ag.refc.shape_changes;
1381 xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1386 error = xfs_alloc_read_agf(ri->ri_pag, tp,
1387 XFS_ALLOC_FLAG_FREEING, &agbp);
1391 rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, ri->ri_pag);
1392 rcur->bc_ag.refc.nr_ops = nr_ops;
1393 rcur->bc_ag.refc.shape_changes = shape_changes;
1397 switch (ri->ri_type) {
1398 case XFS_REFCOUNT_INCREASE:
1399 error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1400 XFS_REFCOUNT_ADJUST_INCREASE);
1403 if (ri->ri_blockcount > 0)
1404 error = xfs_refcount_continue_op(rcur, ri, bno);
1406 case XFS_REFCOUNT_DECREASE:
1407 error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1408 XFS_REFCOUNT_ADJUST_DECREASE);
1411 if (ri->ri_blockcount > 0)
1412 error = xfs_refcount_continue_op(rcur, ri, bno);
1414 case XFS_REFCOUNT_ALLOC_COW:
1415 error = __xfs_refcount_cow_alloc(rcur, bno, ri->ri_blockcount);
1418 ri->ri_blockcount = 0;
1420 case XFS_REFCOUNT_FREE_COW:
1421 error = __xfs_refcount_cow_free(rcur, bno, ri->ri_blockcount);
1424 ri->ri_blockcount = 0;
1428 return -EFSCORRUPTED;
1430 if (!error && ri->ri_blockcount > 0)
1431 trace_xfs_refcount_finish_one_leftover(mp, ri->ri_pag->pag_agno,
1432 ri->ri_type, bno, ri->ri_blockcount);
1437 * Record a refcount intent for later processing.
1441 struct xfs_trans *tp,
1442 enum xfs_refcount_intent_type type,
1443 xfs_fsblock_t startblock,
1444 xfs_extlen_t blockcount)
1446 struct xfs_refcount_intent *ri;
1448 trace_xfs_refcount_defer(tp->t_mountp,
1449 XFS_FSB_TO_AGNO(tp->t_mountp, startblock),
1450 type, XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
1453 ri = kmem_cache_alloc(xfs_refcount_intent_cache,
1454 GFP_NOFS | __GFP_NOFAIL);
1455 INIT_LIST_HEAD(&ri->ri_list);
1457 ri->ri_startblock = startblock;
1458 ri->ri_blockcount = blockcount;
1460 xfs_refcount_update_get_group(tp->t_mountp, ri);
1461 xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
1465 * Increase the reference count of the blocks backing a file's extent.
1468 xfs_refcount_increase_extent(
1469 struct xfs_trans *tp,
1470 struct xfs_bmbt_irec *PREV)
1472 if (!xfs_has_reflink(tp->t_mountp))
1475 __xfs_refcount_add(tp, XFS_REFCOUNT_INCREASE, PREV->br_startblock,
1476 PREV->br_blockcount);
1480 * Decrease the reference count of the blocks backing a file's extent.
1483 xfs_refcount_decrease_extent(
1484 struct xfs_trans *tp,
1485 struct xfs_bmbt_irec *PREV)
1487 if (!xfs_has_reflink(tp->t_mountp))
1490 __xfs_refcount_add(tp, XFS_REFCOUNT_DECREASE, PREV->br_startblock,
1491 PREV->br_blockcount);
1495 * Given an AG extent, find the lowest-numbered run of shared blocks
1496 * within that range and return the range in fbno/flen. If
1497 * find_end_of_shared is set, return the longest contiguous extent of
1498 * shared blocks; if not, just return the first extent we find. If no
1499 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1500 * and 0, respectively.
1503 xfs_refcount_find_shared(
1504 struct xfs_btree_cur *cur,
1505 xfs_agblock_t agbno,
1507 xfs_agblock_t *fbno,
1509 bool find_end_of_shared)
1511 struct xfs_refcount_irec tmp;
1516 trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1519 /* By default, skip the whole range */
1520 *fbno = NULLAGBLOCK;
1523 /* Try to find a refcount extent that crosses the start */
1524 error = xfs_refcount_lookup_le(cur, XFS_REFC_DOMAIN_SHARED, agbno,
1529 /* No left extent, look at the next one */
1530 error = xfs_btree_increment(cur, 0, &have);
1536 error = xfs_refcount_get_rec(cur, &tmp, &i);
1539 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1540 error = -EFSCORRUPTED;
1543 if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1546 /* If the extent ends before the start, look at the next one */
1547 if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1548 error = xfs_btree_increment(cur, 0, &have);
1553 error = xfs_refcount_get_rec(cur, &tmp, &i);
1556 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1557 error = -EFSCORRUPTED;
1560 if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1564 /* If the extent starts after the range we want, bail out */
1565 if (tmp.rc_startblock >= agbno + aglen)
1568 /* We found the start of a shared extent! */
1569 if (tmp.rc_startblock < agbno) {
1570 tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1571 tmp.rc_startblock = agbno;
1574 *fbno = tmp.rc_startblock;
1575 *flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1576 if (!find_end_of_shared)
1579 /* Otherwise, find the end of this shared extent */
1580 while (*fbno + *flen < agbno + aglen) {
1581 error = xfs_btree_increment(cur, 0, &have);
1586 error = xfs_refcount_get_rec(cur, &tmp, &i);
1589 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1590 error = -EFSCORRUPTED;
1593 if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED ||
1594 tmp.rc_startblock >= agbno + aglen ||
1595 tmp.rc_startblock != *fbno + *flen)
1597 *flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1601 trace_xfs_refcount_find_shared_result(cur->bc_mp,
1602 cur->bc_ag.pag->pag_agno, *fbno, *flen);
1606 trace_xfs_refcount_find_shared_error(cur->bc_mp,
1607 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
1612 * Recovering CoW Blocks After a Crash
1614 * Due to the way that the copy on write mechanism works, there's a window of
1615 * opportunity in which we can lose track of allocated blocks during a crash.
1616 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1617 * causes blocks to be allocated and stored in the CoW fork. The blocks are
1618 * no longer in the free space btree but are not otherwise recorded anywhere
1619 * until the write completes and the blocks are mapped into the file. A crash
1620 * in between allocation and remapping results in the replacement blocks being
1621 * lost. This situation is exacerbated by the CoW extent size hint because
1622 * allocations can hang around for long time.
1624 * However, there is a place where we can record these allocations before they
1625 * become mappings -- the reference count btree. The btree does not record
1626 * extents with refcount == 1, so we can record allocations with a refcount of
1627 * 1. Blocks being used for CoW writeout cannot be shared, so there should be
1628 * no conflict with shared block records. These mappings should be created
1629 * when we allocate blocks to the CoW fork and deleted when they're removed
1630 * from the CoW fork.
1632 * Minor nit: records for in-progress CoW allocations and records for shared
1633 * extents must never be merged, to preserve the property that (except for CoW
1634 * allocations) there are no refcount btree entries with refcount == 1. The
1635 * only time this could potentially happen is when unsharing a block that's
1636 * adjacent to CoW allocations, so we must be careful to avoid this.
1638 * At mount time we recover lost CoW allocations by searching the refcount
1639 * btree for these refcount == 1 mappings. These represent CoW allocations
1640 * that were in progress at the time the filesystem went down, so we can free
1641 * them to get the space back.
1643 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1644 * several reasons -- first, EFIs pin the tail of the log and would have to be
1645 * periodically relogged to avoid filling up the log. Second, CoW completions
1646 * will have to file an EFD and create new EFIs for whatever remains in the
1647 * CoW fork; this partially takes care of (1) but extent-size reservations
1648 * will have to periodically relog even if there's no writeout in progress.
1649 * This can happen if the CoW extent size hint is set, which you really want.
1650 * Third, EFIs cannot currently be automatically relogged into newer
1651 * transactions to advance the log tail. Fourth, stuffing the log full of
1652 * EFIs places an upper bound on the number of CoW allocations that can be
1653 * held filesystem-wide at any given time. Recording them in the refcount
1654 * btree doesn't require us to maintain any state in memory and doesn't pin
1658 * Adjust the refcounts of CoW allocations. These allocations are "magic"
1659 * in that they're not referenced anywhere else in the filesystem, so we
1660 * stash them in the refcount btree with a refcount of 1 until either file
1661 * remapping (or CoW cancellation) happens.
1664 xfs_refcount_adjust_cow_extents(
1665 struct xfs_btree_cur *cur,
1666 xfs_agblock_t agbno,
1668 enum xfs_refc_adjust_op adj)
1670 struct xfs_refcount_irec ext, tmp;
1672 int found_rec, found_tmp;
1677 /* Find any overlapping refcount records */
1678 error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_COW, agbno,
1682 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1685 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec &&
1686 ext.rc_domain != XFS_REFC_DOMAIN_COW)) {
1687 error = -EFSCORRUPTED;
1691 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
1692 ext.rc_blockcount = 0;
1693 ext.rc_refcount = 0;
1694 ext.rc_domain = XFS_REFC_DOMAIN_COW;
1698 case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1699 /* Adding a CoW reservation, there should be nothing here. */
1700 if (XFS_IS_CORRUPT(cur->bc_mp,
1701 agbno + aglen > ext.rc_startblock)) {
1702 error = -EFSCORRUPTED;
1706 tmp.rc_startblock = agbno;
1707 tmp.rc_blockcount = aglen;
1708 tmp.rc_refcount = 1;
1709 tmp.rc_domain = XFS_REFC_DOMAIN_COW;
1711 trace_xfs_refcount_modify_extent(cur->bc_mp,
1712 cur->bc_ag.pag->pag_agno, &tmp);
1714 error = xfs_refcount_insert(cur, &tmp,
1718 if (XFS_IS_CORRUPT(cur->bc_mp, found_tmp != 1)) {
1719 error = -EFSCORRUPTED;
1723 case XFS_REFCOUNT_ADJUST_COW_FREE:
1724 /* Removing a CoW reservation, there should be one extent. */
1725 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_startblock != agbno)) {
1726 error = -EFSCORRUPTED;
1729 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount != aglen)) {
1730 error = -EFSCORRUPTED;
1733 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_refcount != 1)) {
1734 error = -EFSCORRUPTED;
1738 ext.rc_refcount = 0;
1739 trace_xfs_refcount_modify_extent(cur->bc_mp,
1740 cur->bc_ag.pag->pag_agno, &ext);
1741 error = xfs_refcount_delete(cur, &found_rec);
1744 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1745 error = -EFSCORRUPTED;
1755 trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1756 cur->bc_ag.pag->pag_agno, error, _RET_IP_);
1761 * Add or remove refcount btree entries for CoW reservations.
1764 xfs_refcount_adjust_cow(
1765 struct xfs_btree_cur *cur,
1766 xfs_agblock_t agbno,
1768 enum xfs_refc_adjust_op adj)
1774 * Ensure that no rcextents cross the boundary of the adjustment range.
1776 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1777 agbno, &shape_changed);
1781 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1782 agbno + aglen, &shape_changed);
1787 * Try to merge with the left or right extents of the range.
1789 error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_COW, &agbno,
1790 &aglen, adj, &shape_changed);
1794 /* Now that we've taken care of the ends, adjust the middle extents */
1795 error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1802 trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_ag.pag->pag_agno,
1808 * Record a CoW allocation in the refcount btree.
1811 __xfs_refcount_cow_alloc(
1812 struct xfs_btree_cur *rcur,
1813 xfs_agblock_t agbno,
1816 trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_ag.pag->pag_agno,
1819 /* Add refcount btree reservation */
1820 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1821 XFS_REFCOUNT_ADJUST_COW_ALLOC);
1825 * Remove a CoW allocation from the refcount btree.
1828 __xfs_refcount_cow_free(
1829 struct xfs_btree_cur *rcur,
1830 xfs_agblock_t agbno,
1833 trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_ag.pag->pag_agno,
1836 /* Remove refcount btree reservation */
1837 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1838 XFS_REFCOUNT_ADJUST_COW_FREE);
1841 /* Record a CoW staging extent in the refcount btree. */
1843 xfs_refcount_alloc_cow_extent(
1844 struct xfs_trans *tp,
1848 struct xfs_mount *mp = tp->t_mountp;
1850 if (!xfs_has_reflink(mp))
1853 __xfs_refcount_add(tp, XFS_REFCOUNT_ALLOC_COW, fsb, len);
1855 /* Add rmap entry */
1856 xfs_rmap_alloc_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1857 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1860 /* Forget a CoW staging event in the refcount btree. */
1862 xfs_refcount_free_cow_extent(
1863 struct xfs_trans *tp,
1867 struct xfs_mount *mp = tp->t_mountp;
1869 if (!xfs_has_reflink(mp))
1872 /* Remove rmap entry */
1873 xfs_rmap_free_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1874 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1875 __xfs_refcount_add(tp, XFS_REFCOUNT_FREE_COW, fsb, len);
1878 struct xfs_refcount_recovery {
1879 struct list_head rr_list;
1880 struct xfs_refcount_irec rr_rrec;
1883 /* Stuff an extent on the recovery list. */
1885 xfs_refcount_recover_extent(
1886 struct xfs_btree_cur *cur,
1887 const union xfs_btree_rec *rec,
1890 struct list_head *debris = priv;
1891 struct xfs_refcount_recovery *rr;
1893 if (XFS_IS_CORRUPT(cur->bc_mp,
1894 be32_to_cpu(rec->refc.rc_refcount) != 1))
1895 return -EFSCORRUPTED;
1897 rr = kmalloc(sizeof(struct xfs_refcount_recovery),
1898 GFP_KERNEL | __GFP_NOFAIL);
1899 INIT_LIST_HEAD(&rr->rr_list);
1900 xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1902 if (xfs_refcount_check_irec(cur, &rr->rr_rrec) != NULL ||
1903 XFS_IS_CORRUPT(cur->bc_mp,
1904 rr->rr_rrec.rc_domain != XFS_REFC_DOMAIN_COW)) {
1906 return -EFSCORRUPTED;
1909 list_add_tail(&rr->rr_list, debris);
1913 /* Find and remove leftover CoW reservations. */
1915 xfs_refcount_recover_cow_leftovers(
1916 struct xfs_mount *mp,
1917 struct xfs_perag *pag)
1919 struct xfs_trans *tp;
1920 struct xfs_btree_cur *cur;
1921 struct xfs_buf *agbp;
1922 struct xfs_refcount_recovery *rr, *n;
1923 struct list_head debris;
1924 union xfs_btree_irec low = {
1925 .rc.rc_domain = XFS_REFC_DOMAIN_COW,
1927 union xfs_btree_irec high = {
1928 .rc.rc_domain = XFS_REFC_DOMAIN_COW,
1929 .rc.rc_startblock = -1U,
1934 /* reflink filesystems mustn't have AGs larger than 2^31-1 blocks */
1935 BUILD_BUG_ON(XFS_MAX_CRC_AG_BLOCKS >= XFS_REFC_COWFLAG);
1936 if (mp->m_sb.sb_agblocks > XFS_MAX_CRC_AG_BLOCKS)
1939 INIT_LIST_HEAD(&debris);
1942 * In this first part, we use an empty transaction to gather up
1943 * all the leftover CoW extents so that we can subsequently
1944 * delete them. The empty transaction is used to avoid
1945 * a buffer lock deadlock if there happens to be a loop in the
1946 * refcountbt because we're allowed to re-grab a buffer that is
1947 * already attached to our transaction. When we're done
1948 * recording the CoW debris we cancel the (empty) transaction
1949 * and everything goes away cleanly.
1951 error = xfs_trans_alloc_empty(mp, &tp);
1955 error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
1958 cur = xfs_refcountbt_init_cursor(mp, tp, agbp, pag);
1960 /* Find all the leftover CoW staging extents. */
1961 error = xfs_btree_query_range(cur, &low, &high,
1962 xfs_refcount_recover_extent, &debris);
1963 xfs_btree_del_cursor(cur, error);
1964 xfs_trans_brelse(tp, agbp);
1965 xfs_trans_cancel(tp);
1969 /* Now iterate the list to free the leftovers */
1970 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1971 /* Set up transaction. */
1972 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1976 trace_xfs_refcount_recover_extent(mp, pag->pag_agno,
1979 /* Free the orphan record */
1980 fsb = XFS_AGB_TO_FSB(mp, pag->pag_agno,
1981 rr->rr_rrec.rc_startblock);
1982 xfs_refcount_free_cow_extent(tp, fsb,
1983 rr->rr_rrec.rc_blockcount);
1985 /* Free the block. */
1986 error = xfs_free_extent_later(tp, fsb,
1987 rr->rr_rrec.rc_blockcount, NULL,
1992 error = xfs_trans_commit(tp);
1996 list_del(&rr->rr_list);
2002 xfs_trans_cancel(tp);
2004 /* Free the leftover list */
2005 list_for_each_entry_safe(rr, n, &debris, rr_list) {
2006 list_del(&rr->rr_list);
2013 * Scan part of the keyspace of the refcount records and tell us if the area
2014 * has no records, is fully mapped by records, or is partially filled.
2017 xfs_refcount_has_records(
2018 struct xfs_btree_cur *cur,
2019 enum xfs_refc_domain domain,
2022 enum xbtree_recpacking *outcome)
2024 union xfs_btree_irec low;
2025 union xfs_btree_irec high;
2027 memset(&low, 0, sizeof(low));
2028 low.rc.rc_startblock = bno;
2029 memset(&high, 0xFF, sizeof(high));
2030 high.rc.rc_startblock = bno + len - 1;
2031 low.rc.rc_domain = high.rc.rc_domain = domain;
2033 return xfs_btree_has_records(cur, &low, &high, NULL, outcome);
2037 xfs_refcount_intent_init_cache(void)
2039 xfs_refcount_intent_cache = kmem_cache_create("xfs_refc_intent",
2040 sizeof(struct xfs_refcount_intent),
2043 return xfs_refcount_intent_cache != NULL ? 0 : -ENOMEM;
2047 xfs_refcount_intent_destroy_cache(void)
2049 kmem_cache_destroy(xfs_refcount_intent_cache);
2050 xfs_refcount_intent_cache = NULL;