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 #include "xfs_health.h"
27 #include "xfs_refcount_item.h"
29 struct kmem_cache *xfs_refcount_intent_cache;
31 /* Allowable refcount adjustment amounts. */
32 enum xfs_refc_adjust_op {
33 XFS_REFCOUNT_ADJUST_INCREASE = 1,
34 XFS_REFCOUNT_ADJUST_DECREASE = -1,
35 XFS_REFCOUNT_ADJUST_COW_ALLOC = 0,
36 XFS_REFCOUNT_ADJUST_COW_FREE = -1,
39 STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
40 xfs_agblock_t agbno, xfs_extlen_t aglen);
41 STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
42 xfs_agblock_t agbno, xfs_extlen_t aglen);
45 * Look up the first record less than or equal to [bno, len] in the btree
49 xfs_refcount_lookup_le(
50 struct xfs_btree_cur *cur,
51 enum xfs_refc_domain domain,
55 trace_xfs_refcount_lookup(cur,
56 xfs_refcount_encode_startblock(bno, domain),
58 cur->bc_rec.rc.rc_startblock = bno;
59 cur->bc_rec.rc.rc_blockcount = 0;
60 cur->bc_rec.rc.rc_domain = domain;
61 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
65 * Look up the first record greater than or equal to [bno, len] in the btree
69 xfs_refcount_lookup_ge(
70 struct xfs_btree_cur *cur,
71 enum xfs_refc_domain domain,
75 trace_xfs_refcount_lookup(cur,
76 xfs_refcount_encode_startblock(bno, domain),
78 cur->bc_rec.rc.rc_startblock = bno;
79 cur->bc_rec.rc.rc_blockcount = 0;
80 cur->bc_rec.rc.rc_domain = domain;
81 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
85 * Look up the first record equal to [bno, len] in the btree
89 xfs_refcount_lookup_eq(
90 struct xfs_btree_cur *cur,
91 enum xfs_refc_domain domain,
95 trace_xfs_refcount_lookup(cur,
96 xfs_refcount_encode_startblock(bno, domain),
98 cur->bc_rec.rc.rc_startblock = bno;
99 cur->bc_rec.rc.rc_blockcount = 0;
100 cur->bc_rec.rc.rc_domain = domain;
101 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
104 /* Convert on-disk record to in-core format. */
106 xfs_refcount_btrec_to_irec(
107 const union xfs_btree_rec *rec,
108 struct xfs_refcount_irec *irec)
112 start = be32_to_cpu(rec->refc.rc_startblock);
113 if (start & XFS_REFC_COWFLAG) {
114 start &= ~XFS_REFC_COWFLAG;
115 irec->rc_domain = XFS_REFC_DOMAIN_COW;
117 irec->rc_domain = XFS_REFC_DOMAIN_SHARED;
120 irec->rc_startblock = start;
121 irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
122 irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
125 /* Simple checks for refcount records. */
127 xfs_refcount_check_irec(
128 struct xfs_perag *pag,
129 const struct xfs_refcount_irec *irec)
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 xfs_btree_mark_sick(cur);
162 return -EFSCORRUPTED;
166 * Get the data from the pointed-to record.
169 xfs_refcount_get_rec(
170 struct xfs_btree_cur *cur,
171 struct xfs_refcount_irec *irec,
174 union xfs_btree_rec *rec;
178 error = xfs_btree_get_rec(cur, &rec, stat);
182 xfs_refcount_btrec_to_irec(rec, irec);
183 fa = xfs_refcount_check_irec(cur->bc_ag.pag, irec);
185 return xfs_refcount_complain_bad_rec(cur, fa, irec);
187 trace_xfs_refcount_get(cur, irec);
192 * Update the record referred to by cur to the value given
193 * by [bno, len, refcount].
194 * This either works (return 0) or gets an EFSCORRUPTED error.
198 struct xfs_btree_cur *cur,
199 struct xfs_refcount_irec *irec)
201 union xfs_btree_rec rec;
205 trace_xfs_refcount_update(cur, irec);
207 start = xfs_refcount_encode_startblock(irec->rc_startblock,
209 rec.refc.rc_startblock = cpu_to_be32(start);
210 rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
211 rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
213 error = xfs_btree_update(cur, &rec);
215 trace_xfs_refcount_update_error(cur, 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, 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 xfs_btree_mark_sick(cur);
244 error = -EFSCORRUPTED;
250 trace_xfs_refcount_insert_error(cur, 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 xfs_btree_mark_sick(cur);
274 error = -EFSCORRUPTED;
277 trace_xfs_refcount_delete(cur, &irec);
278 error = xfs_btree_delete(cur, i);
279 if (XFS_IS_CORRUPT(cur->bc_mp, *i != 1)) {
280 xfs_btree_mark_sick(cur);
281 error = -EFSCORRUPTED;
286 error = xfs_refcount_lookup_ge(cur, irec.rc_domain, irec.rc_startblock,
290 trace_xfs_refcount_delete_error(cur, error, _RET_IP_);
295 * Adjusting the Reference Count
297 * As stated elsewhere, the reference count btree (refcbt) stores
298 * >1 reference counts for extents of physical blocks. In this
299 * operation, we're either raising or lowering the reference count of
300 * some subrange stored in the tree:
302 * <------ adjustment range ------>
303 * ----+ +---+-----+ +--+--------+---------
304 * 2 | | 3 | 4 | |17| 55 | 10
305 * ----+ +---+-----+ +--+--------+---------
306 * X axis is physical blocks number;
307 * reference counts are the numbers inside the rectangles
309 * The first thing we need to do is to ensure that there are no
310 * refcount extents crossing either boundary of the range to be
311 * adjusted. For any extent that does cross a boundary, split it into
312 * two extents so that we can increment the refcount of one of the
315 * <------ adjustment range ------>
316 * ----+ +---+-----+ +--+--------+----+----
317 * 2 | | 3 | 2 | |17| 55 | 10 | 10
318 * ----+ +---+-----+ +--+--------+----+----
320 * For this next step, let's assume that all the physical blocks in
321 * the adjustment range are mapped to a file and are therefore in use
322 * at least once. Therefore, we can infer that any gap in the
323 * refcount tree within the adjustment range represents a physical
324 * extent with refcount == 1:
326 * <------ adjustment range ------>
327 * ----+---+---+-----+-+--+--------+----+----
328 * 2 |"1"| 3 | 2 |1|17| 55 | 10 | 10
329 * ----+---+---+-----+-+--+--------+----+----
332 * For each extent that falls within the interval range, figure out
333 * which extent is to the left or the right of that extent. Now we
334 * have a left, current, and right extent. If the new reference count
335 * of the center extent enables us to merge left, center, and right
336 * into one record covering all three, do so. If the center extent is
337 * at the left end of the range, abuts the left extent, and its new
338 * reference count matches the left extent's record, then merge them.
339 * If the center extent is at the right end of the range, abuts the
340 * right extent, and the reference counts match, merge those. In the
341 * example, we can left merge (assuming an increment operation):
343 * <------ adjustment range ------>
344 * --------+---+-----+-+--+--------+----+----
345 * 2 | 3 | 2 |1|17| 55 | 10 | 10
346 * --------+---+-----+-+--+--------+----+----
349 * For all other extents within the range, adjust the reference count
350 * or delete it if the refcount falls below 2. If we were
351 * incrementing, the end result looks like this:
353 * <------ adjustment range ------>
354 * --------+---+-----+-+--+--------+----+----
355 * 2 | 4 | 3 |2|18| 56 | 11 | 10
356 * --------+---+-----+-+--+--------+----+----
358 * The result of a decrement operation looks as such:
360 * <------ adjustment range ------>
361 * ----+ +---+ +--+--------+----+----
362 * 2 | | 2 | |16| 54 | 9 | 10
363 * ----+ +---+ +--+--------+----+----
366 * The blocks marked "D" are freed; the blocks marked "1" are only
367 * referenced once and therefore the record is removed from the
371 /* Next block after this extent. */
372 static inline xfs_agblock_t
374 struct xfs_refcount_irec *rc)
376 return rc->rc_startblock + rc->rc_blockcount;
380 * Split a refcount extent that crosses agbno.
383 xfs_refcount_split_extent(
384 struct xfs_btree_cur *cur,
385 enum xfs_refc_domain domain,
389 struct xfs_refcount_irec rcext, tmp;
393 *shape_changed = false;
394 error = xfs_refcount_lookup_le(cur, domain, agbno, &found_rec);
400 error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
403 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
404 xfs_btree_mark_sick(cur);
405 error = -EFSCORRUPTED;
408 if (rcext.rc_domain != domain)
410 if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
413 *shape_changed = true;
414 trace_xfs_refcount_split_extent(cur, &rcext, agbno);
416 /* Establish the right extent. */
418 tmp.rc_startblock = agbno;
419 tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
420 error = xfs_refcount_update(cur, &tmp);
424 /* Insert the left extent. */
426 tmp.rc_blockcount = agbno - rcext.rc_startblock;
427 error = xfs_refcount_insert(cur, &tmp, &found_rec);
430 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
431 xfs_btree_mark_sick(cur);
432 error = -EFSCORRUPTED;
438 trace_xfs_refcount_split_extent_error(cur, error, _RET_IP_);
443 * Merge the left, center, and right extents.
446 xfs_refcount_merge_center_extents(
447 struct xfs_btree_cur *cur,
448 struct xfs_refcount_irec *left,
449 struct xfs_refcount_irec *center,
450 struct xfs_refcount_irec *right,
451 unsigned long long extlen,
457 trace_xfs_refcount_merge_center_extents(cur, 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 xfs_btree_mark_sick(cur);
476 error = -EFSCORRUPTED;
480 error = xfs_refcount_delete(cur, &found_rec);
483 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
484 xfs_btree_mark_sick(cur);
485 error = -EFSCORRUPTED;
489 if (center->rc_refcount > 1) {
490 error = xfs_refcount_delete(cur, &found_rec);
493 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
494 xfs_btree_mark_sick(cur);
495 error = -EFSCORRUPTED;
500 /* Enlarge the left extent. */
501 error = xfs_refcount_lookup_le(cur, left->rc_domain,
502 left->rc_startblock, &found_rec);
505 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
506 xfs_btree_mark_sick(cur);
507 error = -EFSCORRUPTED;
511 left->rc_blockcount = extlen;
512 error = xfs_refcount_update(cur, left);
520 trace_xfs_refcount_merge_center_extents_error(cur, error, _RET_IP_);
525 * Merge with the left extent.
528 xfs_refcount_merge_left_extent(
529 struct xfs_btree_cur *cur,
530 struct xfs_refcount_irec *left,
531 struct xfs_refcount_irec *cleft,
532 xfs_agblock_t *agbno,
538 trace_xfs_refcount_merge_left_extent(cur, left, cleft);
540 ASSERT(left->rc_domain == cleft->rc_domain);
542 /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
543 if (cleft->rc_refcount > 1) {
544 error = xfs_refcount_lookup_le(cur, cleft->rc_domain,
545 cleft->rc_startblock, &found_rec);
548 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
549 xfs_btree_mark_sick(cur);
550 error = -EFSCORRUPTED;
554 error = xfs_refcount_delete(cur, &found_rec);
557 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
558 xfs_btree_mark_sick(cur);
559 error = -EFSCORRUPTED;
564 /* Enlarge the left extent. */
565 error = xfs_refcount_lookup_le(cur, left->rc_domain,
566 left->rc_startblock, &found_rec);
569 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
570 xfs_btree_mark_sick(cur);
571 error = -EFSCORRUPTED;
575 left->rc_blockcount += cleft->rc_blockcount;
576 error = xfs_refcount_update(cur, left);
580 *agbno += cleft->rc_blockcount;
581 *aglen -= cleft->rc_blockcount;
585 trace_xfs_refcount_merge_left_extent_error(cur, error, _RET_IP_);
590 * Merge with the right extent.
593 xfs_refcount_merge_right_extent(
594 struct xfs_btree_cur *cur,
595 struct xfs_refcount_irec *right,
596 struct xfs_refcount_irec *cright,
602 trace_xfs_refcount_merge_right_extent(cur, cright, right);
604 ASSERT(right->rc_domain == cright->rc_domain);
607 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
610 if (cright->rc_refcount > 1) {
611 error = xfs_refcount_lookup_le(cur, cright->rc_domain,
612 cright->rc_startblock, &found_rec);
615 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
616 xfs_btree_mark_sick(cur);
617 error = -EFSCORRUPTED;
621 error = xfs_refcount_delete(cur, &found_rec);
624 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
625 xfs_btree_mark_sick(cur);
626 error = -EFSCORRUPTED;
631 /* Enlarge the right extent. */
632 error = xfs_refcount_lookup_le(cur, right->rc_domain,
633 right->rc_startblock, &found_rec);
636 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
637 xfs_btree_mark_sick(cur);
638 error = -EFSCORRUPTED;
642 right->rc_startblock -= cright->rc_blockcount;
643 right->rc_blockcount += cright->rc_blockcount;
644 error = xfs_refcount_update(cur, right);
648 *aglen -= cright->rc_blockcount;
652 trace_xfs_refcount_merge_right_extent_error(cur, error, _RET_IP_);
657 * Find the left extent and the one after it (cleft). This function assumes
658 * that we've already split any extent crossing agbno.
661 xfs_refcount_find_left_extents(
662 struct xfs_btree_cur *cur,
663 struct xfs_refcount_irec *left,
664 struct xfs_refcount_irec *cleft,
665 enum xfs_refc_domain domain,
669 struct xfs_refcount_irec tmp;
673 left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
674 error = xfs_refcount_lookup_le(cur, domain, agbno - 1, &found_rec);
680 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
683 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
684 xfs_btree_mark_sick(cur);
685 error = -EFSCORRUPTED;
689 if (tmp.rc_domain != domain)
691 if (xfs_refc_next(&tmp) != agbno)
693 /* We have a left extent; retrieve (or invent) the next right one */
696 error = xfs_btree_increment(cur, 0, &found_rec);
700 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
703 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
704 xfs_btree_mark_sick(cur);
705 error = -EFSCORRUPTED;
709 if (tmp.rc_domain != domain)
712 /* if tmp starts at the end of our range, just use that */
713 if (tmp.rc_startblock == agbno)
717 * There's a gap in the refcntbt at the start of the
718 * range we're interested in (refcount == 1) so
719 * synthesize the implied extent and pass it back.
720 * We assume here that the agbno/aglen range was
721 * passed in from a data fork extent mapping and
722 * therefore is allocated to exactly one owner.
724 cleft->rc_startblock = agbno;
725 cleft->rc_blockcount = min(aglen,
726 tmp.rc_startblock - agbno);
727 cleft->rc_refcount = 1;
728 cleft->rc_domain = domain;
733 * No extents, so pretend that there's one covering the whole
736 cleft->rc_startblock = agbno;
737 cleft->rc_blockcount = aglen;
738 cleft->rc_refcount = 1;
739 cleft->rc_domain = domain;
741 trace_xfs_refcount_find_left_extent(cur, left, cleft, agbno);
745 trace_xfs_refcount_find_left_extent_error(cur, error, _RET_IP_);
750 * Find the right extent and the one before it (cright). This function
751 * assumes that we've already split any extents crossing agbno + aglen.
754 xfs_refcount_find_right_extents(
755 struct xfs_btree_cur *cur,
756 struct xfs_refcount_irec *right,
757 struct xfs_refcount_irec *cright,
758 enum xfs_refc_domain domain,
762 struct xfs_refcount_irec tmp;
766 right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
767 error = xfs_refcount_lookup_ge(cur, domain, agbno + aglen, &found_rec);
773 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
776 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
777 xfs_btree_mark_sick(cur);
778 error = -EFSCORRUPTED;
782 if (tmp.rc_domain != domain)
784 if (tmp.rc_startblock != agbno + aglen)
786 /* We have a right extent; retrieve (or invent) the next left one */
789 error = xfs_btree_decrement(cur, 0, &found_rec);
793 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
796 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
797 xfs_btree_mark_sick(cur);
798 error = -EFSCORRUPTED;
802 if (tmp.rc_domain != domain)
805 /* if tmp ends at the end of our range, just use that */
806 if (xfs_refc_next(&tmp) == agbno + aglen)
810 * There's a gap in the refcntbt at the end of the
811 * range we're interested in (refcount == 1) so
812 * create the implied extent and pass it back.
813 * We assume here that the agbno/aglen range was
814 * passed in from a data fork extent mapping and
815 * therefore is allocated to exactly one owner.
817 cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
818 cright->rc_blockcount = right->rc_startblock -
819 cright->rc_startblock;
820 cright->rc_refcount = 1;
821 cright->rc_domain = domain;
826 * No extents, so pretend that there's one covering the whole
829 cright->rc_startblock = agbno;
830 cright->rc_blockcount = aglen;
831 cright->rc_refcount = 1;
832 cright->rc_domain = domain;
834 trace_xfs_refcount_find_right_extent(cur, cright, right,
839 trace_xfs_refcount_find_right_extent_error(cur, error, _RET_IP_);
843 /* Is this extent valid? */
846 const struct xfs_refcount_irec *rc)
848 return rc->rc_startblock != NULLAGBLOCK;
851 static inline xfs_nlink_t
852 xfs_refc_merge_refcount(
853 const struct xfs_refcount_irec *irec,
854 enum xfs_refc_adjust_op adjust)
856 /* Once a record hits MAXREFCOUNT, it is pinned there forever */
857 if (irec->rc_refcount == MAXREFCOUNT)
859 return irec->rc_refcount + adjust;
863 xfs_refc_want_merge_center(
864 const struct xfs_refcount_irec *left,
865 const struct xfs_refcount_irec *cleft,
866 const struct xfs_refcount_irec *cright,
867 const struct xfs_refcount_irec *right,
868 bool cleft_is_cright,
869 enum xfs_refc_adjust_op adjust,
870 unsigned long long *ulenp)
872 unsigned long long ulen = left->rc_blockcount;
873 xfs_nlink_t new_refcount;
876 * To merge with a center record, both shoulder records must be
877 * adjacent to the record we want to adjust. This is only true if
878 * find_left and find_right made all four records valid.
880 if (!xfs_refc_valid(left) || !xfs_refc_valid(right) ||
881 !xfs_refc_valid(cleft) || !xfs_refc_valid(cright))
884 /* There must only be one record for the entire range. */
885 if (!cleft_is_cright)
888 /* The shoulder record refcounts must match the new refcount. */
889 new_refcount = xfs_refc_merge_refcount(cleft, adjust);
890 if (left->rc_refcount != new_refcount)
892 if (right->rc_refcount != new_refcount)
896 * The new record cannot exceed the max length. ulen is a ULL as the
897 * individual record block counts can be up to (u32 - 1) in length
898 * hence we need to catch u32 addition overflows here.
900 ulen += cleft->rc_blockcount + right->rc_blockcount;
901 if (ulen >= MAXREFCEXTLEN)
909 xfs_refc_want_merge_left(
910 const struct xfs_refcount_irec *left,
911 const struct xfs_refcount_irec *cleft,
912 enum xfs_refc_adjust_op adjust)
914 unsigned long long ulen = left->rc_blockcount;
915 xfs_nlink_t new_refcount;
918 * For a left merge, the left shoulder record must be adjacent to the
919 * start of the range. If this is true, find_left made left and cleft
920 * contain valid contents.
922 if (!xfs_refc_valid(left) || !xfs_refc_valid(cleft))
925 /* Left shoulder record refcount must match the new refcount. */
926 new_refcount = xfs_refc_merge_refcount(cleft, adjust);
927 if (left->rc_refcount != new_refcount)
931 * The new record cannot exceed the max length. ulen is a ULL as the
932 * individual record block counts can be up to (u32 - 1) in length
933 * hence we need to catch u32 addition overflows here.
935 ulen += cleft->rc_blockcount;
936 if (ulen >= MAXREFCEXTLEN)
943 xfs_refc_want_merge_right(
944 const struct xfs_refcount_irec *cright,
945 const struct xfs_refcount_irec *right,
946 enum xfs_refc_adjust_op adjust)
948 unsigned long long ulen = right->rc_blockcount;
949 xfs_nlink_t new_refcount;
952 * For a right merge, the right shoulder record must be adjacent to the
953 * end of the range. If this is true, find_right made cright and right
954 * contain valid contents.
956 if (!xfs_refc_valid(right) || !xfs_refc_valid(cright))
959 /* Right shoulder record refcount must match the new refcount. */
960 new_refcount = xfs_refc_merge_refcount(cright, adjust);
961 if (right->rc_refcount != new_refcount)
965 * The new record cannot exceed the max length. ulen is a ULL as the
966 * individual record block counts can be up to (u32 - 1) in length
967 * hence we need to catch u32 addition overflows here.
969 ulen += cright->rc_blockcount;
970 if (ulen >= MAXREFCEXTLEN)
977 * Try to merge with any extents on the boundaries of the adjustment range.
980 xfs_refcount_merge_extents(
981 struct xfs_btree_cur *cur,
982 enum xfs_refc_domain domain,
983 xfs_agblock_t *agbno,
985 enum xfs_refc_adjust_op adjust,
988 struct xfs_refcount_irec left = {0}, cleft = {0};
989 struct xfs_refcount_irec cright = {0}, right = {0};
991 unsigned long long ulen;
994 *shape_changed = false;
996 * Find the extent just below agbno [left], just above agbno [cleft],
997 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
1000 error = xfs_refcount_find_left_extents(cur, &left, &cleft, domain,
1004 error = xfs_refcount_find_right_extents(cur, &right, &cright, domain,
1009 /* No left or right extent to merge; exit. */
1010 if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
1013 cequal = (cleft.rc_startblock == cright.rc_startblock) &&
1014 (cleft.rc_blockcount == cright.rc_blockcount);
1016 /* Try to merge left, cleft, and right. cleft must == cright. */
1017 if (xfs_refc_want_merge_center(&left, &cleft, &cright, &right, cequal,
1019 *shape_changed = true;
1020 return xfs_refcount_merge_center_extents(cur, &left, &cleft,
1021 &right, ulen, aglen);
1024 /* Try to merge left and cleft. */
1025 if (xfs_refc_want_merge_left(&left, &cleft, adjust)) {
1026 *shape_changed = true;
1027 error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
1033 * If we just merged left + cleft and cleft == cright,
1034 * we no longer have a cright to merge with right. We're done.
1040 /* Try to merge cright and right. */
1041 if (xfs_refc_want_merge_right(&cright, &right, adjust)) {
1042 *shape_changed = true;
1043 return xfs_refcount_merge_right_extent(cur, &right, &cright,
1051 * XXX: This is a pretty hand-wavy estimate. The penalty for guessing
1052 * true incorrectly is a shutdown FS; the penalty for guessing false
1053 * incorrectly is more transaction rolls than might be necessary.
1054 * Be conservative here.
1057 xfs_refcount_still_have_space(
1058 struct xfs_btree_cur *cur)
1060 unsigned long overhead;
1063 * Worst case estimate: full splits of the free space and rmap btrees
1064 * to handle each of the shape changes to the refcount btree.
1066 overhead = xfs_allocfree_block_count(cur->bc_mp,
1067 cur->bc_refc.shape_changes);
1068 overhead += cur->bc_mp->m_refc_maxlevels;
1069 overhead *= cur->bc_mp->m_sb.sb_blocksize;
1072 * Only allow 2 refcount extent updates per transaction if the
1073 * refcount continue update "error" has been injected.
1075 if (cur->bc_refc.nr_ops > 2 &&
1076 XFS_TEST_ERROR(false, cur->bc_mp,
1077 XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
1080 if (cur->bc_refc.nr_ops == 0)
1082 else if (overhead > cur->bc_tp->t_log_res)
1084 return cur->bc_tp->t_log_res - overhead >
1085 cur->bc_refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
1089 * Adjust the refcounts of middle extents. At this point we should have
1090 * split extents that crossed the adjustment range; merged with adjacent
1091 * extents; and updated agbno/aglen to reflect the merges. Therefore,
1092 * all we have to do is update the extents inside [agbno, agbno + aglen].
1095 xfs_refcount_adjust_extents(
1096 struct xfs_btree_cur *cur,
1097 xfs_agblock_t *agbno,
1098 xfs_extlen_t *aglen,
1099 enum xfs_refc_adjust_op adj)
1101 struct xfs_refcount_irec ext, tmp;
1103 int found_rec, found_tmp;
1104 xfs_fsblock_t fsbno;
1106 /* Merging did all the work already. */
1110 error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_SHARED, *agbno,
1115 while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
1116 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1119 if (!found_rec || ext.rc_domain != XFS_REFC_DOMAIN_SHARED) {
1120 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
1121 ext.rc_blockcount = 0;
1122 ext.rc_refcount = 0;
1123 ext.rc_domain = XFS_REFC_DOMAIN_SHARED;
1127 * Deal with a hole in the refcount tree; if a file maps to
1128 * these blocks and there's no refcountbt record, pretend that
1129 * there is one with refcount == 1.
1131 if (ext.rc_startblock != *agbno) {
1132 tmp.rc_startblock = *agbno;
1133 tmp.rc_blockcount = min(*aglen,
1134 ext.rc_startblock - *agbno);
1135 tmp.rc_refcount = 1 + adj;
1136 tmp.rc_domain = XFS_REFC_DOMAIN_SHARED;
1138 trace_xfs_refcount_modify_extent(cur, &tmp);
1141 * Either cover the hole (increment) or
1142 * delete the range (decrement).
1144 cur->bc_refc.nr_ops++;
1145 if (tmp.rc_refcount) {
1146 error = xfs_refcount_insert(cur, &tmp,
1150 if (XFS_IS_CORRUPT(cur->bc_mp,
1152 xfs_btree_mark_sick(cur);
1153 error = -EFSCORRUPTED;
1157 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1158 cur->bc_ag.pag->pag_agno,
1160 error = xfs_free_extent_later(cur->bc_tp, fsbno,
1161 tmp.rc_blockcount, NULL,
1162 XFS_AG_RESV_NONE, 0);
1167 (*agbno) += tmp.rc_blockcount;
1168 (*aglen) -= tmp.rc_blockcount;
1170 /* Stop if there's nothing left to modify */
1171 if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
1174 /* Move the cursor to the start of ext. */
1175 error = xfs_refcount_lookup_ge(cur,
1176 XFS_REFC_DOMAIN_SHARED, *agbno,
1183 * A previous step trimmed agbno/aglen such that the end of the
1184 * range would not be in the middle of the record. If this is
1185 * no longer the case, something is seriously wrong with the
1186 * btree. Make sure we never feed the synthesized record into
1187 * the processing loop below.
1189 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount == 0) ||
1190 XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount > *aglen)) {
1191 xfs_btree_mark_sick(cur);
1192 error = -EFSCORRUPTED;
1197 * Adjust the reference count and either update the tree
1198 * (incr) or free the blocks (decr).
1200 if (ext.rc_refcount == MAXREFCOUNT)
1202 ext.rc_refcount += adj;
1203 trace_xfs_refcount_modify_extent(cur, &ext);
1204 cur->bc_refc.nr_ops++;
1205 if (ext.rc_refcount > 1) {
1206 error = xfs_refcount_update(cur, &ext);
1209 } else if (ext.rc_refcount == 1) {
1210 error = xfs_refcount_delete(cur, &found_rec);
1213 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1214 xfs_btree_mark_sick(cur);
1215 error = -EFSCORRUPTED;
1220 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
1221 cur->bc_ag.pag->pag_agno,
1223 error = xfs_free_extent_later(cur->bc_tp, fsbno,
1224 ext.rc_blockcount, NULL,
1225 XFS_AG_RESV_NONE, 0);
1231 error = xfs_btree_increment(cur, 0, &found_rec);
1236 (*agbno) += ext.rc_blockcount;
1237 (*aglen) -= ext.rc_blockcount;
1242 trace_xfs_refcount_modify_extent_error(cur, error, _RET_IP_);
1246 /* Adjust the reference count of a range of AG blocks. */
1248 xfs_refcount_adjust(
1249 struct xfs_btree_cur *cur,
1250 xfs_agblock_t *agbno,
1251 xfs_extlen_t *aglen,
1252 enum xfs_refc_adjust_op adj)
1255 int shape_changes = 0;
1258 if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
1259 trace_xfs_refcount_increase(cur, *agbno, *aglen);
1261 trace_xfs_refcount_decrease(cur, *agbno, *aglen);
1264 * Ensure that no rcextents cross the boundary of the adjustment range.
1266 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1267 *agbno, &shape_changed);
1273 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_SHARED,
1274 *agbno + *aglen, &shape_changed);
1281 * Try to merge with the left or right extents of the range.
1283 error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_SHARED,
1284 agbno, aglen, adj, &shape_changed);
1290 cur->bc_refc.shape_changes++;
1292 /* Now that we've taken care of the ends, adjust the middle extents */
1293 error = xfs_refcount_adjust_extents(cur, agbno, aglen, adj);
1300 trace_xfs_refcount_adjust_error(cur, error, _RET_IP_);
1305 * Set up a continuation a deferred refcount operation by updating the intent.
1306 * Checks to make sure we're not going to run off the end of the AG.
1309 xfs_refcount_continue_op(
1310 struct xfs_btree_cur *cur,
1311 struct xfs_refcount_intent *ri,
1312 xfs_agblock_t new_agbno)
1314 struct xfs_mount *mp = cur->bc_mp;
1315 struct xfs_perag *pag = cur->bc_ag.pag;
1317 if (XFS_IS_CORRUPT(mp, !xfs_verify_agbext(pag, new_agbno,
1318 ri->ri_blockcount))) {
1319 xfs_btree_mark_sick(cur);
1320 return -EFSCORRUPTED;
1323 ri->ri_startblock = XFS_AGB_TO_FSB(mp, pag->pag_agno, new_agbno);
1325 ASSERT(xfs_verify_fsbext(mp, ri->ri_startblock, ri->ri_blockcount));
1326 ASSERT(pag->pag_agno == XFS_FSB_TO_AGNO(mp, ri->ri_startblock));
1332 * Process one of the deferred refcount operations. We pass back the
1333 * btree cursor to maintain our lock on the btree between calls.
1334 * This saves time and eliminates a buffer deadlock between the
1335 * superblock and the AGF because we'll always grab them in the same
1339 xfs_refcount_finish_one(
1340 struct xfs_trans *tp,
1341 struct xfs_refcount_intent *ri,
1342 struct xfs_btree_cur **pcur)
1344 struct xfs_mount *mp = tp->t_mountp;
1345 struct xfs_btree_cur *rcur = *pcur;
1346 struct xfs_buf *agbp = NULL;
1349 unsigned long nr_ops = 0;
1350 int shape_changes = 0;
1352 bno = XFS_FSB_TO_AGBNO(mp, ri->ri_startblock);
1354 trace_xfs_refcount_deferred(mp, ri);
1356 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1360 * If we haven't gotten a cursor or the cursor AG doesn't match
1361 * the startblock, get one now.
1363 if (rcur != NULL && rcur->bc_ag.pag != ri->ri_pag) {
1364 nr_ops = rcur->bc_refc.nr_ops;
1365 shape_changes = rcur->bc_refc.shape_changes;
1366 xfs_btree_del_cursor(rcur, 0);
1371 error = xfs_alloc_read_agf(ri->ri_pag, tp,
1372 XFS_ALLOC_FLAG_FREEING, &agbp);
1376 *pcur = rcur = xfs_refcountbt_init_cursor(mp, tp, agbp,
1378 rcur->bc_refc.nr_ops = nr_ops;
1379 rcur->bc_refc.shape_changes = shape_changes;
1382 switch (ri->ri_type) {
1383 case XFS_REFCOUNT_INCREASE:
1384 error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1385 XFS_REFCOUNT_ADJUST_INCREASE);
1388 if (ri->ri_blockcount > 0)
1389 error = xfs_refcount_continue_op(rcur, ri, bno);
1391 case XFS_REFCOUNT_DECREASE:
1392 error = xfs_refcount_adjust(rcur, &bno, &ri->ri_blockcount,
1393 XFS_REFCOUNT_ADJUST_DECREASE);
1396 if (ri->ri_blockcount > 0)
1397 error = xfs_refcount_continue_op(rcur, ri, bno);
1399 case XFS_REFCOUNT_ALLOC_COW:
1400 error = __xfs_refcount_cow_alloc(rcur, bno, ri->ri_blockcount);
1403 ri->ri_blockcount = 0;
1405 case XFS_REFCOUNT_FREE_COW:
1406 error = __xfs_refcount_cow_free(rcur, bno, ri->ri_blockcount);
1409 ri->ri_blockcount = 0;
1413 return -EFSCORRUPTED;
1415 if (!error && ri->ri_blockcount > 0)
1416 trace_xfs_refcount_finish_one_leftover(mp, ri);
1421 * Record a refcount intent for later processing.
1425 struct xfs_trans *tp,
1426 enum xfs_refcount_intent_type type,
1427 xfs_fsblock_t startblock,
1428 xfs_extlen_t blockcount)
1430 struct xfs_refcount_intent *ri;
1432 ri = kmem_cache_alloc(xfs_refcount_intent_cache,
1433 GFP_KERNEL | __GFP_NOFAIL);
1434 INIT_LIST_HEAD(&ri->ri_list);
1436 ri->ri_startblock = startblock;
1437 ri->ri_blockcount = blockcount;
1439 xfs_refcount_defer_add(tp, ri);
1443 * Increase the reference count of the blocks backing a file's extent.
1446 xfs_refcount_increase_extent(
1447 struct xfs_trans *tp,
1448 struct xfs_bmbt_irec *PREV)
1450 if (!xfs_has_reflink(tp->t_mountp))
1453 __xfs_refcount_add(tp, XFS_REFCOUNT_INCREASE, PREV->br_startblock,
1454 PREV->br_blockcount);
1458 * Decrease the reference count of the blocks backing a file's extent.
1461 xfs_refcount_decrease_extent(
1462 struct xfs_trans *tp,
1463 struct xfs_bmbt_irec *PREV)
1465 if (!xfs_has_reflink(tp->t_mountp))
1468 __xfs_refcount_add(tp, XFS_REFCOUNT_DECREASE, PREV->br_startblock,
1469 PREV->br_blockcount);
1473 * Given an AG extent, find the lowest-numbered run of shared blocks
1474 * within that range and return the range in fbno/flen. If
1475 * find_end_of_shared is set, return the longest contiguous extent of
1476 * shared blocks; if not, just return the first extent we find. If no
1477 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1478 * and 0, respectively.
1481 xfs_refcount_find_shared(
1482 struct xfs_btree_cur *cur,
1483 xfs_agblock_t agbno,
1485 xfs_agblock_t *fbno,
1487 bool find_end_of_shared)
1489 struct xfs_refcount_irec tmp;
1494 trace_xfs_refcount_find_shared(cur, agbno, aglen);
1496 /* By default, skip the whole range */
1497 *fbno = NULLAGBLOCK;
1500 /* Try to find a refcount extent that crosses the start */
1501 error = xfs_refcount_lookup_le(cur, XFS_REFC_DOMAIN_SHARED, agbno,
1506 /* No left extent, look at the next one */
1507 error = xfs_btree_increment(cur, 0, &have);
1513 error = xfs_refcount_get_rec(cur, &tmp, &i);
1516 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1517 xfs_btree_mark_sick(cur);
1518 error = -EFSCORRUPTED;
1521 if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1524 /* If the extent ends before the start, look at the next one */
1525 if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1526 error = xfs_btree_increment(cur, 0, &have);
1531 error = xfs_refcount_get_rec(cur, &tmp, &i);
1534 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1535 xfs_btree_mark_sick(cur);
1536 error = -EFSCORRUPTED;
1539 if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED)
1543 /* If the extent starts after the range we want, bail out */
1544 if (tmp.rc_startblock >= agbno + aglen)
1547 /* We found the start of a shared extent! */
1548 if (tmp.rc_startblock < agbno) {
1549 tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1550 tmp.rc_startblock = agbno;
1553 *fbno = tmp.rc_startblock;
1554 *flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1555 if (!find_end_of_shared)
1558 /* Otherwise, find the end of this shared extent */
1559 while (*fbno + *flen < agbno + aglen) {
1560 error = xfs_btree_increment(cur, 0, &have);
1565 error = xfs_refcount_get_rec(cur, &tmp, &i);
1568 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1569 xfs_btree_mark_sick(cur);
1570 error = -EFSCORRUPTED;
1573 if (tmp.rc_domain != XFS_REFC_DOMAIN_SHARED ||
1574 tmp.rc_startblock >= agbno + aglen ||
1575 tmp.rc_startblock != *fbno + *flen)
1577 *flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1581 trace_xfs_refcount_find_shared_result(cur, *fbno, *flen);
1585 trace_xfs_refcount_find_shared_error(cur, error, _RET_IP_);
1590 * Recovering CoW Blocks After a Crash
1592 * Due to the way that the copy on write mechanism works, there's a window of
1593 * opportunity in which we can lose track of allocated blocks during a crash.
1594 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1595 * causes blocks to be allocated and stored in the CoW fork. The blocks are
1596 * no longer in the free space btree but are not otherwise recorded anywhere
1597 * until the write completes and the blocks are mapped into the file. A crash
1598 * in between allocation and remapping results in the replacement blocks being
1599 * lost. This situation is exacerbated by the CoW extent size hint because
1600 * allocations can hang around for long time.
1602 * However, there is a place where we can record these allocations before they
1603 * become mappings -- the reference count btree. The btree does not record
1604 * extents with refcount == 1, so we can record allocations with a refcount of
1605 * 1. Blocks being used for CoW writeout cannot be shared, so there should be
1606 * no conflict with shared block records. These mappings should be created
1607 * when we allocate blocks to the CoW fork and deleted when they're removed
1608 * from the CoW fork.
1610 * Minor nit: records for in-progress CoW allocations and records for shared
1611 * extents must never be merged, to preserve the property that (except for CoW
1612 * allocations) there are no refcount btree entries with refcount == 1. The
1613 * only time this could potentially happen is when unsharing a block that's
1614 * adjacent to CoW allocations, so we must be careful to avoid this.
1616 * At mount time we recover lost CoW allocations by searching the refcount
1617 * btree for these refcount == 1 mappings. These represent CoW allocations
1618 * that were in progress at the time the filesystem went down, so we can free
1619 * them to get the space back.
1621 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1622 * several reasons -- first, EFIs pin the tail of the log and would have to be
1623 * periodically relogged to avoid filling up the log. Second, CoW completions
1624 * will have to file an EFD and create new EFIs for whatever remains in the
1625 * CoW fork; this partially takes care of (1) but extent-size reservations
1626 * will have to periodically relog even if there's no writeout in progress.
1627 * This can happen if the CoW extent size hint is set, which you really want.
1628 * Third, EFIs cannot currently be automatically relogged into newer
1629 * transactions to advance the log tail. Fourth, stuffing the log full of
1630 * EFIs places an upper bound on the number of CoW allocations that can be
1631 * held filesystem-wide at any given time. Recording them in the refcount
1632 * btree doesn't require us to maintain any state in memory and doesn't pin
1636 * Adjust the refcounts of CoW allocations. These allocations are "magic"
1637 * in that they're not referenced anywhere else in the filesystem, so we
1638 * stash them in the refcount btree with a refcount of 1 until either file
1639 * remapping (or CoW cancellation) happens.
1642 xfs_refcount_adjust_cow_extents(
1643 struct xfs_btree_cur *cur,
1644 xfs_agblock_t agbno,
1646 enum xfs_refc_adjust_op adj)
1648 struct xfs_refcount_irec ext, tmp;
1650 int found_rec, found_tmp;
1655 /* Find any overlapping refcount records */
1656 error = xfs_refcount_lookup_ge(cur, XFS_REFC_DOMAIN_COW, agbno,
1660 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1663 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec &&
1664 ext.rc_domain != XFS_REFC_DOMAIN_COW)) {
1665 xfs_btree_mark_sick(cur);
1666 error = -EFSCORRUPTED;
1670 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
1671 ext.rc_blockcount = 0;
1672 ext.rc_refcount = 0;
1673 ext.rc_domain = XFS_REFC_DOMAIN_COW;
1677 case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1678 /* Adding a CoW reservation, there should be nothing here. */
1679 if (XFS_IS_CORRUPT(cur->bc_mp,
1680 agbno + aglen > ext.rc_startblock)) {
1681 xfs_btree_mark_sick(cur);
1682 error = -EFSCORRUPTED;
1686 tmp.rc_startblock = agbno;
1687 tmp.rc_blockcount = aglen;
1688 tmp.rc_refcount = 1;
1689 tmp.rc_domain = XFS_REFC_DOMAIN_COW;
1691 trace_xfs_refcount_modify_extent(cur, &tmp);
1693 error = xfs_refcount_insert(cur, &tmp,
1697 if (XFS_IS_CORRUPT(cur->bc_mp, found_tmp != 1)) {
1698 xfs_btree_mark_sick(cur);
1699 error = -EFSCORRUPTED;
1703 case XFS_REFCOUNT_ADJUST_COW_FREE:
1704 /* Removing a CoW reservation, there should be one extent. */
1705 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_startblock != agbno)) {
1706 xfs_btree_mark_sick(cur);
1707 error = -EFSCORRUPTED;
1710 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_blockcount != aglen)) {
1711 xfs_btree_mark_sick(cur);
1712 error = -EFSCORRUPTED;
1715 if (XFS_IS_CORRUPT(cur->bc_mp, ext.rc_refcount != 1)) {
1716 xfs_btree_mark_sick(cur);
1717 error = -EFSCORRUPTED;
1721 ext.rc_refcount = 0;
1722 trace_xfs_refcount_modify_extent(cur, &ext);
1723 error = xfs_refcount_delete(cur, &found_rec);
1726 if (XFS_IS_CORRUPT(cur->bc_mp, found_rec != 1)) {
1727 xfs_btree_mark_sick(cur);
1728 error = -EFSCORRUPTED;
1738 trace_xfs_refcount_modify_extent_error(cur, error, _RET_IP_);
1743 * Add or remove refcount btree entries for CoW reservations.
1746 xfs_refcount_adjust_cow(
1747 struct xfs_btree_cur *cur,
1748 xfs_agblock_t agbno,
1750 enum xfs_refc_adjust_op adj)
1756 * Ensure that no rcextents cross the boundary of the adjustment range.
1758 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1759 agbno, &shape_changed);
1763 error = xfs_refcount_split_extent(cur, XFS_REFC_DOMAIN_COW,
1764 agbno + aglen, &shape_changed);
1769 * Try to merge with the left or right extents of the range.
1771 error = xfs_refcount_merge_extents(cur, XFS_REFC_DOMAIN_COW, &agbno,
1772 &aglen, adj, &shape_changed);
1776 /* Now that we've taken care of the ends, adjust the middle extents */
1777 error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1784 trace_xfs_refcount_adjust_cow_error(cur, error, _RET_IP_);
1789 * Record a CoW allocation in the refcount btree.
1792 __xfs_refcount_cow_alloc(
1793 struct xfs_btree_cur *rcur,
1794 xfs_agblock_t agbno,
1797 trace_xfs_refcount_cow_increase(rcur, agbno, aglen);
1799 /* Add refcount btree reservation */
1800 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1801 XFS_REFCOUNT_ADJUST_COW_ALLOC);
1805 * Remove a CoW allocation from the refcount btree.
1808 __xfs_refcount_cow_free(
1809 struct xfs_btree_cur *rcur,
1810 xfs_agblock_t agbno,
1813 trace_xfs_refcount_cow_decrease(rcur, agbno, aglen);
1815 /* Remove refcount btree reservation */
1816 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1817 XFS_REFCOUNT_ADJUST_COW_FREE);
1820 /* Record a CoW staging extent in the refcount btree. */
1822 xfs_refcount_alloc_cow_extent(
1823 struct xfs_trans *tp,
1827 struct xfs_mount *mp = tp->t_mountp;
1829 if (!xfs_has_reflink(mp))
1832 __xfs_refcount_add(tp, XFS_REFCOUNT_ALLOC_COW, fsb, len);
1834 /* Add rmap entry */
1835 xfs_rmap_alloc_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1836 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1839 /* Forget a CoW staging event in the refcount btree. */
1841 xfs_refcount_free_cow_extent(
1842 struct xfs_trans *tp,
1846 struct xfs_mount *mp = tp->t_mountp;
1848 if (!xfs_has_reflink(mp))
1851 /* Remove rmap entry */
1852 xfs_rmap_free_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1853 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1854 __xfs_refcount_add(tp, XFS_REFCOUNT_FREE_COW, fsb, len);
1857 struct xfs_refcount_recovery {
1858 struct list_head rr_list;
1859 struct xfs_refcount_irec rr_rrec;
1862 /* Stuff an extent on the recovery list. */
1864 xfs_refcount_recover_extent(
1865 struct xfs_btree_cur *cur,
1866 const union xfs_btree_rec *rec,
1869 struct list_head *debris = priv;
1870 struct xfs_refcount_recovery *rr;
1872 if (XFS_IS_CORRUPT(cur->bc_mp,
1873 be32_to_cpu(rec->refc.rc_refcount) != 1)) {
1874 xfs_btree_mark_sick(cur);
1875 return -EFSCORRUPTED;
1878 rr = kmalloc(sizeof(struct xfs_refcount_recovery),
1879 GFP_KERNEL | __GFP_NOFAIL);
1880 INIT_LIST_HEAD(&rr->rr_list);
1881 xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1883 if (xfs_refcount_check_irec(cur->bc_ag.pag, &rr->rr_rrec) != NULL ||
1884 XFS_IS_CORRUPT(cur->bc_mp,
1885 rr->rr_rrec.rc_domain != XFS_REFC_DOMAIN_COW)) {
1886 xfs_btree_mark_sick(cur);
1888 return -EFSCORRUPTED;
1891 list_add_tail(&rr->rr_list, debris);
1895 /* Find and remove leftover CoW reservations. */
1897 xfs_refcount_recover_cow_leftovers(
1898 struct xfs_mount *mp,
1899 struct xfs_perag *pag)
1901 struct xfs_trans *tp;
1902 struct xfs_btree_cur *cur;
1903 struct xfs_buf *agbp;
1904 struct xfs_refcount_recovery *rr, *n;
1905 struct list_head debris;
1906 union xfs_btree_irec low = {
1907 .rc.rc_domain = XFS_REFC_DOMAIN_COW,
1909 union xfs_btree_irec high = {
1910 .rc.rc_domain = XFS_REFC_DOMAIN_COW,
1911 .rc.rc_startblock = -1U,
1916 /* reflink filesystems mustn't have AGs larger than 2^31-1 blocks */
1917 BUILD_BUG_ON(XFS_MAX_CRC_AG_BLOCKS >= XFS_REFC_COWFLAG);
1918 if (mp->m_sb.sb_agblocks > XFS_MAX_CRC_AG_BLOCKS)
1921 INIT_LIST_HEAD(&debris);
1924 * In this first part, we use an empty transaction to gather up
1925 * all the leftover CoW extents so that we can subsequently
1926 * delete them. The empty transaction is used to avoid
1927 * a buffer lock deadlock if there happens to be a loop in the
1928 * refcountbt because we're allowed to re-grab a buffer that is
1929 * already attached to our transaction. When we're done
1930 * recording the CoW debris we cancel the (empty) transaction
1931 * and everything goes away cleanly.
1933 error = xfs_trans_alloc_empty(mp, &tp);
1937 error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
1940 cur = xfs_refcountbt_init_cursor(mp, tp, agbp, pag);
1942 /* Find all the leftover CoW staging extents. */
1943 error = xfs_btree_query_range(cur, &low, &high,
1944 xfs_refcount_recover_extent, &debris);
1945 xfs_btree_del_cursor(cur, error);
1946 xfs_trans_brelse(tp, agbp);
1947 xfs_trans_cancel(tp);
1951 /* Now iterate the list to free the leftovers */
1952 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1953 /* Set up transaction. */
1954 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1958 /* Free the orphan record */
1959 fsb = XFS_AGB_TO_FSB(mp, pag->pag_agno,
1960 rr->rr_rrec.rc_startblock);
1961 xfs_refcount_free_cow_extent(tp, fsb,
1962 rr->rr_rrec.rc_blockcount);
1964 /* Free the block. */
1965 error = xfs_free_extent_later(tp, fsb,
1966 rr->rr_rrec.rc_blockcount, NULL,
1967 XFS_AG_RESV_NONE, 0);
1971 error = xfs_trans_commit(tp);
1975 list_del(&rr->rr_list);
1981 xfs_trans_cancel(tp);
1983 /* Free the leftover list */
1984 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1985 list_del(&rr->rr_list);
1992 * Scan part of the keyspace of the refcount records and tell us if the area
1993 * has no records, is fully mapped by records, or is partially filled.
1996 xfs_refcount_has_records(
1997 struct xfs_btree_cur *cur,
1998 enum xfs_refc_domain domain,
2001 enum xbtree_recpacking *outcome)
2003 union xfs_btree_irec low;
2004 union xfs_btree_irec high;
2006 memset(&low, 0, sizeof(low));
2007 low.rc.rc_startblock = bno;
2008 memset(&high, 0xFF, sizeof(high));
2009 high.rc.rc_startblock = bno + len - 1;
2010 low.rc.rc_domain = high.rc.rc_domain = domain;
2012 return xfs_btree_has_records(cur, &low, &high, NULL, outcome);
2015 struct xfs_refcount_query_range_info {
2016 xfs_refcount_query_range_fn fn;
2020 /* Format btree record and pass to our callback. */
2022 xfs_refcount_query_range_helper(
2023 struct xfs_btree_cur *cur,
2024 const union xfs_btree_rec *rec,
2027 struct xfs_refcount_query_range_info *query = priv;
2028 struct xfs_refcount_irec irec;
2031 xfs_refcount_btrec_to_irec(rec, &irec);
2032 fa = xfs_refcount_check_irec(cur->bc_ag.pag, &irec);
2034 return xfs_refcount_complain_bad_rec(cur, fa, &irec);
2036 return query->fn(cur, &irec, query->priv);
2039 /* Find all refcount records between two keys. */
2041 xfs_refcount_query_range(
2042 struct xfs_btree_cur *cur,
2043 const struct xfs_refcount_irec *low_rec,
2044 const struct xfs_refcount_irec *high_rec,
2045 xfs_refcount_query_range_fn fn,
2048 union xfs_btree_irec low_brec = { .rc = *low_rec };
2049 union xfs_btree_irec high_brec = { .rc = *high_rec };
2050 struct xfs_refcount_query_range_info query = { .priv = priv, .fn = fn };
2052 return xfs_btree_query_range(cur, &low_brec, &high_brec,
2053 xfs_refcount_query_range_helper, &query);
2057 xfs_refcount_intent_init_cache(void)
2059 xfs_refcount_intent_cache = kmem_cache_create("xfs_refc_intent",
2060 sizeof(struct xfs_refcount_intent),
2063 return xfs_refcount_intent_cache != NULL ? 0 : -ENOMEM;
2067 xfs_refcount_intent_destroy_cache(void)
2069 kmem_cache_destroy(xfs_refcount_intent_cache);
2070 xfs_refcount_intent_cache = NULL;