]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blob - libxfs/xfs_refcount.c
libxfs: refactor manage_zones()
[thirdparty/xfsprogs-dev.git] / libxfs / xfs_refcount.c
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
5 */
6 #include "libxfs_priv.h"
7 #include "xfs_fs.h"
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_sb.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap.h"
17 #include "xfs_refcount_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_errortag.h"
20 #include "xfs_trace.h"
21 #include "xfs_cksum.h"
22 #include "xfs_trans.h"
23 #include "xfs_bit.h"
24 #include "xfs_refcount.h"
25 #include "xfs_rmap.h"
26
27 /* Allowable refcount adjustment amounts. */
28 enum xfs_refc_adjust_op {
29 XFS_REFCOUNT_ADJUST_INCREASE = 1,
30 XFS_REFCOUNT_ADJUST_DECREASE = -1,
31 XFS_REFCOUNT_ADJUST_COW_ALLOC = 0,
32 XFS_REFCOUNT_ADJUST_COW_FREE = -1,
33 };
34
35 STATIC int __xfs_refcount_cow_alloc(struct xfs_btree_cur *rcur,
36 xfs_agblock_t agbno, xfs_extlen_t aglen);
37 STATIC int __xfs_refcount_cow_free(struct xfs_btree_cur *rcur,
38 xfs_agblock_t agbno, xfs_extlen_t aglen);
39
40 /*
41 * Look up the first record less than or equal to [bno, len] in the btree
42 * given by cur.
43 */
44 int
45 xfs_refcount_lookup_le(
46 struct xfs_btree_cur *cur,
47 xfs_agblock_t bno,
48 int *stat)
49 {
50 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
51 XFS_LOOKUP_LE);
52 cur->bc_rec.rc.rc_startblock = bno;
53 cur->bc_rec.rc.rc_blockcount = 0;
54 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
55 }
56
57 /*
58 * Look up the first record greater than or equal to [bno, len] in the btree
59 * given by cur.
60 */
61 int
62 xfs_refcount_lookup_ge(
63 struct xfs_btree_cur *cur,
64 xfs_agblock_t bno,
65 int *stat)
66 {
67 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
68 XFS_LOOKUP_GE);
69 cur->bc_rec.rc.rc_startblock = bno;
70 cur->bc_rec.rc.rc_blockcount = 0;
71 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
72 }
73
74 /*
75 * Look up the first record equal to [bno, len] in the btree
76 * given by cur.
77 */
78 int
79 xfs_refcount_lookup_eq(
80 struct xfs_btree_cur *cur,
81 xfs_agblock_t bno,
82 int *stat)
83 {
84 trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
85 XFS_LOOKUP_LE);
86 cur->bc_rec.rc.rc_startblock = bno;
87 cur->bc_rec.rc.rc_blockcount = 0;
88 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
89 }
90
91 /* Convert on-disk record to in-core format. */
92 void
93 xfs_refcount_btrec_to_irec(
94 union xfs_btree_rec *rec,
95 struct xfs_refcount_irec *irec)
96 {
97 irec->rc_startblock = be32_to_cpu(rec->refc.rc_startblock);
98 irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
99 irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
100 }
101
102 /*
103 * Get the data from the pointed-to record.
104 */
105 int
106 xfs_refcount_get_rec(
107 struct xfs_btree_cur *cur,
108 struct xfs_refcount_irec *irec,
109 int *stat)
110 {
111 struct xfs_mount *mp = cur->bc_mp;
112 xfs_agnumber_t agno = cur->bc_private.a.agno;
113 union xfs_btree_rec *rec;
114 int error;
115 xfs_agblock_t realstart;
116
117 error = xfs_btree_get_rec(cur, &rec, stat);
118 if (error || !*stat)
119 return error;
120
121 xfs_refcount_btrec_to_irec(rec, irec);
122
123 agno = cur->bc_private.a.agno;
124 if (irec->rc_blockcount == 0 || irec->rc_blockcount > MAXREFCEXTLEN)
125 goto out_bad_rec;
126
127 /* handle special COW-staging state */
128 realstart = irec->rc_startblock;
129 if (realstart & XFS_REFC_COW_START) {
130 if (irec->rc_refcount != 1)
131 goto out_bad_rec;
132 realstart &= ~XFS_REFC_COW_START;
133 } else if (irec->rc_refcount < 2) {
134 goto out_bad_rec;
135 }
136
137 /* check for valid extent range, including overflow */
138 if (!xfs_verify_agbno(mp, agno, realstart))
139 goto out_bad_rec;
140 if (realstart > realstart + irec->rc_blockcount)
141 goto out_bad_rec;
142 if (!xfs_verify_agbno(mp, agno, realstart + irec->rc_blockcount - 1))
143 goto out_bad_rec;
144
145 if (irec->rc_refcount == 0 || irec->rc_refcount > MAXREFCOUNT)
146 goto out_bad_rec;
147
148 trace_xfs_refcount_get(cur->bc_mp, cur->bc_private.a.agno, irec);
149 return 0;
150
151 out_bad_rec:
152 xfs_warn(mp,
153 "Refcount BTree record corruption in AG %d detected!", agno);
154 xfs_warn(mp,
155 "Start block 0x%x, block count 0x%x, references 0x%x",
156 irec->rc_startblock, irec->rc_blockcount, irec->rc_refcount);
157 return -EFSCORRUPTED;
158 }
159
160 /*
161 * Update the record referred to by cur to the value given
162 * by [bno, len, refcount].
163 * This either works (return 0) or gets an EFSCORRUPTED error.
164 */
165 STATIC int
166 xfs_refcount_update(
167 struct xfs_btree_cur *cur,
168 struct xfs_refcount_irec *irec)
169 {
170 union xfs_btree_rec rec;
171 int error;
172
173 trace_xfs_refcount_update(cur->bc_mp, cur->bc_private.a.agno, irec);
174 rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
175 rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
176 rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
177 error = xfs_btree_update(cur, &rec);
178 if (error)
179 trace_xfs_refcount_update_error(cur->bc_mp,
180 cur->bc_private.a.agno, error, _RET_IP_);
181 return error;
182 }
183
184 /*
185 * Insert the record referred to by cur to the value given
186 * by [bno, len, refcount].
187 * This either works (return 0) or gets an EFSCORRUPTED error.
188 */
189 int
190 xfs_refcount_insert(
191 struct xfs_btree_cur *cur,
192 struct xfs_refcount_irec *irec,
193 int *i)
194 {
195 int error;
196
197 trace_xfs_refcount_insert(cur->bc_mp, cur->bc_private.a.agno, irec);
198 cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
199 cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
200 cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
201 error = xfs_btree_insert(cur, i);
202 if (error)
203 goto out_error;
204 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
205
206 out_error:
207 if (error)
208 trace_xfs_refcount_insert_error(cur->bc_mp,
209 cur->bc_private.a.agno, error, _RET_IP_);
210 return error;
211 }
212
213 /*
214 * Remove the record referred to by cur, then set the pointer to the spot
215 * where the record could be re-inserted, in case we want to increment or
216 * decrement the cursor.
217 * This either works (return 0) or gets an EFSCORRUPTED error.
218 */
219 STATIC int
220 xfs_refcount_delete(
221 struct xfs_btree_cur *cur,
222 int *i)
223 {
224 struct xfs_refcount_irec irec;
225 int found_rec;
226 int error;
227
228 error = xfs_refcount_get_rec(cur, &irec, &found_rec);
229 if (error)
230 goto out_error;
231 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
232 trace_xfs_refcount_delete(cur->bc_mp, cur->bc_private.a.agno, &irec);
233 error = xfs_btree_delete(cur, i);
234 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
235 if (error)
236 goto out_error;
237 error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
238 out_error:
239 if (error)
240 trace_xfs_refcount_delete_error(cur->bc_mp,
241 cur->bc_private.a.agno, error, _RET_IP_);
242 return error;
243 }
244
245 /*
246 * Adjusting the Reference Count
247 *
248 * As stated elsewhere, the reference count btree (refcbt) stores
249 * >1 reference counts for extents of physical blocks. In this
250 * operation, we're either raising or lowering the reference count of
251 * some subrange stored in the tree:
252 *
253 * <------ adjustment range ------>
254 * ----+ +---+-----+ +--+--------+---------
255 * 2 | | 3 | 4 | |17| 55 | 10
256 * ----+ +---+-----+ +--+--------+---------
257 * X axis is physical blocks number;
258 * reference counts are the numbers inside the rectangles
259 *
260 * The first thing we need to do is to ensure that there are no
261 * refcount extents crossing either boundary of the range to be
262 * adjusted. For any extent that does cross a boundary, split it into
263 * two extents so that we can increment the refcount of one of the
264 * pieces later:
265 *
266 * <------ adjustment range ------>
267 * ----+ +---+-----+ +--+--------+----+----
268 * 2 | | 3 | 2 | |17| 55 | 10 | 10
269 * ----+ +---+-----+ +--+--------+----+----
270 *
271 * For this next step, let's assume that all the physical blocks in
272 * the adjustment range are mapped to a file and are therefore in use
273 * at least once. Therefore, we can infer that any gap in the
274 * refcount tree within the adjustment range represents a physical
275 * extent with refcount == 1:
276 *
277 * <------ adjustment range ------>
278 * ----+---+---+-----+-+--+--------+----+----
279 * 2 |"1"| 3 | 2 |1|17| 55 | 10 | 10
280 * ----+---+---+-----+-+--+--------+----+----
281 * ^
282 *
283 * For each extent that falls within the interval range, figure out
284 * which extent is to the left or the right of that extent. Now we
285 * have a left, current, and right extent. If the new reference count
286 * of the center extent enables us to merge left, center, and right
287 * into one record covering all three, do so. If the center extent is
288 * at the left end of the range, abuts the left extent, and its new
289 * reference count matches the left extent's record, then merge them.
290 * If the center extent is at the right end of the range, abuts the
291 * right extent, and the reference counts match, merge those. In the
292 * example, we can left merge (assuming an increment operation):
293 *
294 * <------ adjustment range ------>
295 * --------+---+-----+-+--+--------+----+----
296 * 2 | 3 | 2 |1|17| 55 | 10 | 10
297 * --------+---+-----+-+--+--------+----+----
298 * ^
299 *
300 * For all other extents within the range, adjust the reference count
301 * or delete it if the refcount falls below 2. If we were
302 * incrementing, the end result looks like this:
303 *
304 * <------ adjustment range ------>
305 * --------+---+-----+-+--+--------+----+----
306 * 2 | 4 | 3 |2|18| 56 | 11 | 10
307 * --------+---+-----+-+--+--------+----+----
308 *
309 * The result of a decrement operation looks as such:
310 *
311 * <------ adjustment range ------>
312 * ----+ +---+ +--+--------+----+----
313 * 2 | | 2 | |16| 54 | 9 | 10
314 * ----+ +---+ +--+--------+----+----
315 * DDDD 111111DD
316 *
317 * The blocks marked "D" are freed; the blocks marked "1" are only
318 * referenced once and therefore the record is removed from the
319 * refcount btree.
320 */
321
322 /* Next block after this extent. */
323 static inline xfs_agblock_t
324 xfs_refc_next(
325 struct xfs_refcount_irec *rc)
326 {
327 return rc->rc_startblock + rc->rc_blockcount;
328 }
329
330 /*
331 * Split a refcount extent that crosses agbno.
332 */
333 STATIC int
334 xfs_refcount_split_extent(
335 struct xfs_btree_cur *cur,
336 xfs_agblock_t agbno,
337 bool *shape_changed)
338 {
339 struct xfs_refcount_irec rcext, tmp;
340 int found_rec;
341 int error;
342
343 *shape_changed = false;
344 error = xfs_refcount_lookup_le(cur, agbno, &found_rec);
345 if (error)
346 goto out_error;
347 if (!found_rec)
348 return 0;
349
350 error = xfs_refcount_get_rec(cur, &rcext, &found_rec);
351 if (error)
352 goto out_error;
353 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
354 if (rcext.rc_startblock == agbno || xfs_refc_next(&rcext) <= agbno)
355 return 0;
356
357 *shape_changed = true;
358 trace_xfs_refcount_split_extent(cur->bc_mp, cur->bc_private.a.agno,
359 &rcext, agbno);
360
361 /* Establish the right extent. */
362 tmp = rcext;
363 tmp.rc_startblock = agbno;
364 tmp.rc_blockcount -= (agbno - rcext.rc_startblock);
365 error = xfs_refcount_update(cur, &tmp);
366 if (error)
367 goto out_error;
368
369 /* Insert the left extent. */
370 tmp = rcext;
371 tmp.rc_blockcount = agbno - rcext.rc_startblock;
372 error = xfs_refcount_insert(cur, &tmp, &found_rec);
373 if (error)
374 goto out_error;
375 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
376 return error;
377
378 out_error:
379 trace_xfs_refcount_split_extent_error(cur->bc_mp,
380 cur->bc_private.a.agno, error, _RET_IP_);
381 return error;
382 }
383
384 /*
385 * Merge the left, center, and right extents.
386 */
387 STATIC int
388 xfs_refcount_merge_center_extents(
389 struct xfs_btree_cur *cur,
390 struct xfs_refcount_irec *left,
391 struct xfs_refcount_irec *center,
392 struct xfs_refcount_irec *right,
393 unsigned long long extlen,
394 xfs_extlen_t *aglen)
395 {
396 int error;
397 int found_rec;
398
399 trace_xfs_refcount_merge_center_extents(cur->bc_mp,
400 cur->bc_private.a.agno, left, center, right);
401
402 /*
403 * Make sure the center and right extents are not in the btree.
404 * If the center extent was synthesized, the first delete call
405 * removes the right extent and we skip the second deletion.
406 * If center and right were in the btree, then the first delete
407 * call removes the center and the second one removes the right
408 * extent.
409 */
410 error = xfs_refcount_lookup_ge(cur, center->rc_startblock,
411 &found_rec);
412 if (error)
413 goto out_error;
414 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
415
416 error = xfs_refcount_delete(cur, &found_rec);
417 if (error)
418 goto out_error;
419 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
420
421 if (center->rc_refcount > 1) {
422 error = xfs_refcount_delete(cur, &found_rec);
423 if (error)
424 goto out_error;
425 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
426 out_error);
427 }
428
429 /* Enlarge the left extent. */
430 error = xfs_refcount_lookup_le(cur, left->rc_startblock,
431 &found_rec);
432 if (error)
433 goto out_error;
434 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
435
436 left->rc_blockcount = extlen;
437 error = xfs_refcount_update(cur, left);
438 if (error)
439 goto out_error;
440
441 *aglen = 0;
442 return error;
443
444 out_error:
445 trace_xfs_refcount_merge_center_extents_error(cur->bc_mp,
446 cur->bc_private.a.agno, error, _RET_IP_);
447 return error;
448 }
449
450 /*
451 * Merge with the left extent.
452 */
453 STATIC int
454 xfs_refcount_merge_left_extent(
455 struct xfs_btree_cur *cur,
456 struct xfs_refcount_irec *left,
457 struct xfs_refcount_irec *cleft,
458 xfs_agblock_t *agbno,
459 xfs_extlen_t *aglen)
460 {
461 int error;
462 int found_rec;
463
464 trace_xfs_refcount_merge_left_extent(cur->bc_mp,
465 cur->bc_private.a.agno, left, cleft);
466
467 /* If the extent at agbno (cleft) wasn't synthesized, remove it. */
468 if (cleft->rc_refcount > 1) {
469 error = xfs_refcount_lookup_le(cur, cleft->rc_startblock,
470 &found_rec);
471 if (error)
472 goto out_error;
473 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
474 out_error);
475
476 error = xfs_refcount_delete(cur, &found_rec);
477 if (error)
478 goto out_error;
479 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
480 out_error);
481 }
482
483 /* Enlarge the left extent. */
484 error = xfs_refcount_lookup_le(cur, left->rc_startblock,
485 &found_rec);
486 if (error)
487 goto out_error;
488 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
489
490 left->rc_blockcount += cleft->rc_blockcount;
491 error = xfs_refcount_update(cur, left);
492 if (error)
493 goto out_error;
494
495 *agbno += cleft->rc_blockcount;
496 *aglen -= cleft->rc_blockcount;
497 return error;
498
499 out_error:
500 trace_xfs_refcount_merge_left_extent_error(cur->bc_mp,
501 cur->bc_private.a.agno, error, _RET_IP_);
502 return error;
503 }
504
505 /*
506 * Merge with the right extent.
507 */
508 STATIC int
509 xfs_refcount_merge_right_extent(
510 struct xfs_btree_cur *cur,
511 struct xfs_refcount_irec *right,
512 struct xfs_refcount_irec *cright,
513 xfs_extlen_t *aglen)
514 {
515 int error;
516 int found_rec;
517
518 trace_xfs_refcount_merge_right_extent(cur->bc_mp,
519 cur->bc_private.a.agno, cright, right);
520
521 /*
522 * If the extent ending at agbno+aglen (cright) wasn't synthesized,
523 * remove it.
524 */
525 if (cright->rc_refcount > 1) {
526 error = xfs_refcount_lookup_le(cur, cright->rc_startblock,
527 &found_rec);
528 if (error)
529 goto out_error;
530 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
531 out_error);
532
533 error = xfs_refcount_delete(cur, &found_rec);
534 if (error)
535 goto out_error;
536 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
537 out_error);
538 }
539
540 /* Enlarge the right extent. */
541 error = xfs_refcount_lookup_le(cur, right->rc_startblock,
542 &found_rec);
543 if (error)
544 goto out_error;
545 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
546
547 right->rc_startblock -= cright->rc_blockcount;
548 right->rc_blockcount += cright->rc_blockcount;
549 error = xfs_refcount_update(cur, right);
550 if (error)
551 goto out_error;
552
553 *aglen -= cright->rc_blockcount;
554 return error;
555
556 out_error:
557 trace_xfs_refcount_merge_right_extent_error(cur->bc_mp,
558 cur->bc_private.a.agno, error, _RET_IP_);
559 return error;
560 }
561
562 #define XFS_FIND_RCEXT_SHARED 1
563 #define XFS_FIND_RCEXT_COW 2
564 /*
565 * Find the left extent and the one after it (cleft). This function assumes
566 * that we've already split any extent crossing agbno.
567 */
568 STATIC int
569 xfs_refcount_find_left_extents(
570 struct xfs_btree_cur *cur,
571 struct xfs_refcount_irec *left,
572 struct xfs_refcount_irec *cleft,
573 xfs_agblock_t agbno,
574 xfs_extlen_t aglen,
575 int flags)
576 {
577 struct xfs_refcount_irec tmp;
578 int error;
579 int found_rec;
580
581 left->rc_startblock = cleft->rc_startblock = NULLAGBLOCK;
582 error = xfs_refcount_lookup_le(cur, agbno - 1, &found_rec);
583 if (error)
584 goto out_error;
585 if (!found_rec)
586 return 0;
587
588 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
589 if (error)
590 goto out_error;
591 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
592
593 if (xfs_refc_next(&tmp) != agbno)
594 return 0;
595 if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
596 return 0;
597 if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
598 return 0;
599 /* We have a left extent; retrieve (or invent) the next right one */
600 *left = tmp;
601
602 error = xfs_btree_increment(cur, 0, &found_rec);
603 if (error)
604 goto out_error;
605 if (found_rec) {
606 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
607 if (error)
608 goto out_error;
609 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
610 out_error);
611
612 /* if tmp starts at the end of our range, just use that */
613 if (tmp.rc_startblock == agbno)
614 *cleft = tmp;
615 else {
616 /*
617 * There's a gap in the refcntbt at the start of the
618 * range we're interested in (refcount == 1) so
619 * synthesize the implied extent and pass it back.
620 * We assume here that the agbno/aglen range was
621 * passed in from a data fork extent mapping and
622 * therefore is allocated to exactly one owner.
623 */
624 cleft->rc_startblock = agbno;
625 cleft->rc_blockcount = min(aglen,
626 tmp.rc_startblock - agbno);
627 cleft->rc_refcount = 1;
628 }
629 } else {
630 /*
631 * No extents, so pretend that there's one covering the whole
632 * range.
633 */
634 cleft->rc_startblock = agbno;
635 cleft->rc_blockcount = aglen;
636 cleft->rc_refcount = 1;
637 }
638 trace_xfs_refcount_find_left_extent(cur->bc_mp, cur->bc_private.a.agno,
639 left, cleft, agbno);
640 return error;
641
642 out_error:
643 trace_xfs_refcount_find_left_extent_error(cur->bc_mp,
644 cur->bc_private.a.agno, error, _RET_IP_);
645 return error;
646 }
647
648 /*
649 * Find the right extent and the one before it (cright). This function
650 * assumes that we've already split any extents crossing agbno + aglen.
651 */
652 STATIC int
653 xfs_refcount_find_right_extents(
654 struct xfs_btree_cur *cur,
655 struct xfs_refcount_irec *right,
656 struct xfs_refcount_irec *cright,
657 xfs_agblock_t agbno,
658 xfs_extlen_t aglen,
659 int flags)
660 {
661 struct xfs_refcount_irec tmp;
662 int error;
663 int found_rec;
664
665 right->rc_startblock = cright->rc_startblock = NULLAGBLOCK;
666 error = xfs_refcount_lookup_ge(cur, agbno + aglen, &found_rec);
667 if (error)
668 goto out_error;
669 if (!found_rec)
670 return 0;
671
672 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
673 if (error)
674 goto out_error;
675 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
676
677 if (tmp.rc_startblock != agbno + aglen)
678 return 0;
679 if ((flags & XFS_FIND_RCEXT_SHARED) && tmp.rc_refcount < 2)
680 return 0;
681 if ((flags & XFS_FIND_RCEXT_COW) && tmp.rc_refcount > 1)
682 return 0;
683 /* We have a right extent; retrieve (or invent) the next left one */
684 *right = tmp;
685
686 error = xfs_btree_decrement(cur, 0, &found_rec);
687 if (error)
688 goto out_error;
689 if (found_rec) {
690 error = xfs_refcount_get_rec(cur, &tmp, &found_rec);
691 if (error)
692 goto out_error;
693 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1,
694 out_error);
695
696 /* if tmp ends at the end of our range, just use that */
697 if (xfs_refc_next(&tmp) == agbno + aglen)
698 *cright = tmp;
699 else {
700 /*
701 * There's a gap in the refcntbt at the end of the
702 * range we're interested in (refcount == 1) so
703 * create the implied extent and pass it back.
704 * We assume here that the agbno/aglen range was
705 * passed in from a data fork extent mapping and
706 * therefore is allocated to exactly one owner.
707 */
708 cright->rc_startblock = max(agbno, xfs_refc_next(&tmp));
709 cright->rc_blockcount = right->rc_startblock -
710 cright->rc_startblock;
711 cright->rc_refcount = 1;
712 }
713 } else {
714 /*
715 * No extents, so pretend that there's one covering the whole
716 * range.
717 */
718 cright->rc_startblock = agbno;
719 cright->rc_blockcount = aglen;
720 cright->rc_refcount = 1;
721 }
722 trace_xfs_refcount_find_right_extent(cur->bc_mp, cur->bc_private.a.agno,
723 cright, right, agbno + aglen);
724 return error;
725
726 out_error:
727 trace_xfs_refcount_find_right_extent_error(cur->bc_mp,
728 cur->bc_private.a.agno, error, _RET_IP_);
729 return error;
730 }
731
732 /* Is this extent valid? */
733 static inline bool
734 xfs_refc_valid(
735 struct xfs_refcount_irec *rc)
736 {
737 return rc->rc_startblock != NULLAGBLOCK;
738 }
739
740 /*
741 * Try to merge with any extents on the boundaries of the adjustment range.
742 */
743 STATIC int
744 xfs_refcount_merge_extents(
745 struct xfs_btree_cur *cur,
746 xfs_agblock_t *agbno,
747 xfs_extlen_t *aglen,
748 enum xfs_refc_adjust_op adjust,
749 int flags,
750 bool *shape_changed)
751 {
752 struct xfs_refcount_irec left = {0}, cleft = {0};
753 struct xfs_refcount_irec cright = {0}, right = {0};
754 int error;
755 unsigned long long ulen;
756 bool cequal;
757
758 *shape_changed = false;
759 /*
760 * Find the extent just below agbno [left], just above agbno [cleft],
761 * just below (agbno + aglen) [cright], and just above (agbno + aglen)
762 * [right].
763 */
764 error = xfs_refcount_find_left_extents(cur, &left, &cleft, *agbno,
765 *aglen, flags);
766 if (error)
767 return error;
768 error = xfs_refcount_find_right_extents(cur, &right, &cright, *agbno,
769 *aglen, flags);
770 if (error)
771 return error;
772
773 /* No left or right extent to merge; exit. */
774 if (!xfs_refc_valid(&left) && !xfs_refc_valid(&right))
775 return 0;
776
777 cequal = (cleft.rc_startblock == cright.rc_startblock) &&
778 (cleft.rc_blockcount == cright.rc_blockcount);
779
780 /* Try to merge left, cleft, and right. cleft must == cright. */
781 ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount +
782 right.rc_blockcount;
783 if (xfs_refc_valid(&left) && xfs_refc_valid(&right) &&
784 xfs_refc_valid(&cleft) && xfs_refc_valid(&cright) && cequal &&
785 left.rc_refcount == cleft.rc_refcount + adjust &&
786 right.rc_refcount == cleft.rc_refcount + adjust &&
787 ulen < MAXREFCEXTLEN) {
788 *shape_changed = true;
789 return xfs_refcount_merge_center_extents(cur, &left, &cleft,
790 &right, ulen, aglen);
791 }
792
793 /* Try to merge left and cleft. */
794 ulen = (unsigned long long)left.rc_blockcount + cleft.rc_blockcount;
795 if (xfs_refc_valid(&left) && xfs_refc_valid(&cleft) &&
796 left.rc_refcount == cleft.rc_refcount + adjust &&
797 ulen < MAXREFCEXTLEN) {
798 *shape_changed = true;
799 error = xfs_refcount_merge_left_extent(cur, &left, &cleft,
800 agbno, aglen);
801 if (error)
802 return error;
803
804 /*
805 * If we just merged left + cleft and cleft == cright,
806 * we no longer have a cright to merge with right. We're done.
807 */
808 if (cequal)
809 return 0;
810 }
811
812 /* Try to merge cright and right. */
813 ulen = (unsigned long long)right.rc_blockcount + cright.rc_blockcount;
814 if (xfs_refc_valid(&right) && xfs_refc_valid(&cright) &&
815 right.rc_refcount == cright.rc_refcount + adjust &&
816 ulen < MAXREFCEXTLEN) {
817 *shape_changed = true;
818 return xfs_refcount_merge_right_extent(cur, &right, &cright,
819 aglen);
820 }
821
822 return error;
823 }
824
825 /*
826 * XXX: This is a pretty hand-wavy estimate. The penalty for guessing
827 * true incorrectly is a shutdown FS; the penalty for guessing false
828 * incorrectly is more transaction rolls than might be necessary.
829 * Be conservative here.
830 */
831 static bool
832 xfs_refcount_still_have_space(
833 struct xfs_btree_cur *cur)
834 {
835 unsigned long overhead;
836
837 overhead = cur->bc_private.a.priv.refc.shape_changes *
838 xfs_allocfree_log_count(cur->bc_mp, 1);
839 overhead *= cur->bc_mp->m_sb.sb_blocksize;
840
841 /*
842 * Only allow 2 refcount extent updates per transaction if the
843 * refcount continue update "error" has been injected.
844 */
845 if (cur->bc_private.a.priv.refc.nr_ops > 2 &&
846 XFS_TEST_ERROR(false, cur->bc_mp,
847 XFS_ERRTAG_REFCOUNT_CONTINUE_UPDATE))
848 return false;
849
850 if (cur->bc_private.a.priv.refc.nr_ops == 0)
851 return true;
852 else if (overhead > cur->bc_tp->t_log_res)
853 return false;
854 return cur->bc_tp->t_log_res - overhead >
855 cur->bc_private.a.priv.refc.nr_ops * XFS_REFCOUNT_ITEM_OVERHEAD;
856 }
857
858 /*
859 * Adjust the refcounts of middle extents. At this point we should have
860 * split extents that crossed the adjustment range; merged with adjacent
861 * extents; and updated agbno/aglen to reflect the merges. Therefore,
862 * all we have to do is update the extents inside [agbno, agbno + aglen].
863 */
864 STATIC int
865 xfs_refcount_adjust_extents(
866 struct xfs_btree_cur *cur,
867 xfs_agblock_t *agbno,
868 xfs_extlen_t *aglen,
869 enum xfs_refc_adjust_op adj,
870 struct xfs_owner_info *oinfo)
871 {
872 struct xfs_refcount_irec ext, tmp;
873 int error;
874 int found_rec, found_tmp;
875 xfs_fsblock_t fsbno;
876
877 /* Merging did all the work already. */
878 if (*aglen == 0)
879 return 0;
880
881 error = xfs_refcount_lookup_ge(cur, *agbno, &found_rec);
882 if (error)
883 goto out_error;
884
885 while (*aglen > 0 && xfs_refcount_still_have_space(cur)) {
886 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
887 if (error)
888 goto out_error;
889 if (!found_rec) {
890 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks;
891 ext.rc_blockcount = 0;
892 ext.rc_refcount = 0;
893 }
894
895 /*
896 * Deal with a hole in the refcount tree; if a file maps to
897 * these blocks and there's no refcountbt record, pretend that
898 * there is one with refcount == 1.
899 */
900 if (ext.rc_startblock != *agbno) {
901 tmp.rc_startblock = *agbno;
902 tmp.rc_blockcount = min(*aglen,
903 ext.rc_startblock - *agbno);
904 tmp.rc_refcount = 1 + adj;
905 trace_xfs_refcount_modify_extent(cur->bc_mp,
906 cur->bc_private.a.agno, &tmp);
907
908 /*
909 * Either cover the hole (increment) or
910 * delete the range (decrement).
911 */
912 if (tmp.rc_refcount) {
913 error = xfs_refcount_insert(cur, &tmp,
914 &found_tmp);
915 if (error)
916 goto out_error;
917 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
918 found_tmp == 1, out_error);
919 cur->bc_private.a.priv.refc.nr_ops++;
920 } else {
921 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
922 cur->bc_private.a.agno,
923 tmp.rc_startblock);
924 xfs_bmap_add_free(cur->bc_tp, fsbno,
925 tmp.rc_blockcount, oinfo);
926 }
927
928 (*agbno) += tmp.rc_blockcount;
929 (*aglen) -= tmp.rc_blockcount;
930
931 error = xfs_refcount_lookup_ge(cur, *agbno,
932 &found_rec);
933 if (error)
934 goto out_error;
935 }
936
937 /* Stop if there's nothing left to modify */
938 if (*aglen == 0 || !xfs_refcount_still_have_space(cur))
939 break;
940
941 /*
942 * Adjust the reference count and either update the tree
943 * (incr) or free the blocks (decr).
944 */
945 if (ext.rc_refcount == MAXREFCOUNT)
946 goto skip;
947 ext.rc_refcount += adj;
948 trace_xfs_refcount_modify_extent(cur->bc_mp,
949 cur->bc_private.a.agno, &ext);
950 if (ext.rc_refcount > 1) {
951 error = xfs_refcount_update(cur, &ext);
952 if (error)
953 goto out_error;
954 cur->bc_private.a.priv.refc.nr_ops++;
955 } else if (ext.rc_refcount == 1) {
956 error = xfs_refcount_delete(cur, &found_rec);
957 if (error)
958 goto out_error;
959 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
960 found_rec == 1, out_error);
961 cur->bc_private.a.priv.refc.nr_ops++;
962 goto advloop;
963 } else {
964 fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
965 cur->bc_private.a.agno,
966 ext.rc_startblock);
967 xfs_bmap_add_free(cur->bc_tp, fsbno, ext.rc_blockcount,
968 oinfo);
969 }
970
971 skip:
972 error = xfs_btree_increment(cur, 0, &found_rec);
973 if (error)
974 goto out_error;
975
976 advloop:
977 (*agbno) += ext.rc_blockcount;
978 (*aglen) -= ext.rc_blockcount;
979 }
980
981 return error;
982 out_error:
983 trace_xfs_refcount_modify_extent_error(cur->bc_mp,
984 cur->bc_private.a.agno, error, _RET_IP_);
985 return error;
986 }
987
988 /* Adjust the reference count of a range of AG blocks. */
989 STATIC int
990 xfs_refcount_adjust(
991 struct xfs_btree_cur *cur,
992 xfs_agblock_t agbno,
993 xfs_extlen_t aglen,
994 xfs_agblock_t *new_agbno,
995 xfs_extlen_t *new_aglen,
996 enum xfs_refc_adjust_op adj,
997 struct xfs_owner_info *oinfo)
998 {
999 bool shape_changed;
1000 int shape_changes = 0;
1001 int error;
1002
1003 *new_agbno = agbno;
1004 *new_aglen = aglen;
1005 if (adj == XFS_REFCOUNT_ADJUST_INCREASE)
1006 trace_xfs_refcount_increase(cur->bc_mp, cur->bc_private.a.agno,
1007 agbno, aglen);
1008 else
1009 trace_xfs_refcount_decrease(cur->bc_mp, cur->bc_private.a.agno,
1010 agbno, aglen);
1011
1012 /*
1013 * Ensure that no rcextents cross the boundary of the adjustment range.
1014 */
1015 error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1016 if (error)
1017 goto out_error;
1018 if (shape_changed)
1019 shape_changes++;
1020
1021 error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1022 if (error)
1023 goto out_error;
1024 if (shape_changed)
1025 shape_changes++;
1026
1027 /*
1028 * Try to merge with the left or right extents of the range.
1029 */
1030 error = xfs_refcount_merge_extents(cur, new_agbno, new_aglen, adj,
1031 XFS_FIND_RCEXT_SHARED, &shape_changed);
1032 if (error)
1033 goto out_error;
1034 if (shape_changed)
1035 shape_changes++;
1036 if (shape_changes)
1037 cur->bc_private.a.priv.refc.shape_changes++;
1038
1039 /* Now that we've taken care of the ends, adjust the middle extents */
1040 error = xfs_refcount_adjust_extents(cur, new_agbno, new_aglen,
1041 adj, oinfo);
1042 if (error)
1043 goto out_error;
1044
1045 return 0;
1046
1047 out_error:
1048 trace_xfs_refcount_adjust_error(cur->bc_mp, cur->bc_private.a.agno,
1049 error, _RET_IP_);
1050 return error;
1051 }
1052
1053 /* Clean up after calling xfs_refcount_finish_one. */
1054 void
1055 xfs_refcount_finish_one_cleanup(
1056 struct xfs_trans *tp,
1057 struct xfs_btree_cur *rcur,
1058 int error)
1059 {
1060 struct xfs_buf *agbp;
1061
1062 if (rcur == NULL)
1063 return;
1064 agbp = rcur->bc_private.a.agbp;
1065 xfs_btree_del_cursor(rcur, error);
1066 if (error)
1067 xfs_trans_brelse(tp, agbp);
1068 }
1069
1070 /*
1071 * Process one of the deferred refcount operations. We pass back the
1072 * btree cursor to maintain our lock on the btree between calls.
1073 * This saves time and eliminates a buffer deadlock between the
1074 * superblock and the AGF because we'll always grab them in the same
1075 * order.
1076 */
1077 int
1078 xfs_refcount_finish_one(
1079 struct xfs_trans *tp,
1080 enum xfs_refcount_intent_type type,
1081 xfs_fsblock_t startblock,
1082 xfs_extlen_t blockcount,
1083 xfs_fsblock_t *new_fsb,
1084 xfs_extlen_t *new_len,
1085 struct xfs_btree_cur **pcur)
1086 {
1087 struct xfs_mount *mp = tp->t_mountp;
1088 struct xfs_btree_cur *rcur;
1089 struct xfs_buf *agbp = NULL;
1090 int error = 0;
1091 xfs_agnumber_t agno;
1092 xfs_agblock_t bno;
1093 xfs_agblock_t new_agbno;
1094 unsigned long nr_ops = 0;
1095 int shape_changes = 0;
1096
1097 agno = XFS_FSB_TO_AGNO(mp, startblock);
1098 ASSERT(agno != NULLAGNUMBER);
1099 bno = XFS_FSB_TO_AGBNO(mp, startblock);
1100
1101 trace_xfs_refcount_deferred(mp, XFS_FSB_TO_AGNO(mp, startblock),
1102 type, XFS_FSB_TO_AGBNO(mp, startblock),
1103 blockcount);
1104
1105 if (XFS_TEST_ERROR(false, mp,
1106 XFS_ERRTAG_REFCOUNT_FINISH_ONE))
1107 return -EIO;
1108
1109 /*
1110 * If we haven't gotten a cursor or the cursor AG doesn't match
1111 * the startblock, get one now.
1112 */
1113 rcur = *pcur;
1114 if (rcur != NULL && rcur->bc_private.a.agno != agno) {
1115 nr_ops = rcur->bc_private.a.priv.refc.nr_ops;
1116 shape_changes = rcur->bc_private.a.priv.refc.shape_changes;
1117 xfs_refcount_finish_one_cleanup(tp, rcur, 0);
1118 rcur = NULL;
1119 *pcur = NULL;
1120 }
1121 if (rcur == NULL) {
1122 error = xfs_alloc_read_agf(tp->t_mountp, tp, agno,
1123 XFS_ALLOC_FLAG_FREEING, &agbp);
1124 if (error)
1125 return error;
1126 if (!agbp)
1127 return -EFSCORRUPTED;
1128
1129 rcur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno);
1130 if (!rcur) {
1131 error = -ENOMEM;
1132 goto out_cur;
1133 }
1134 rcur->bc_private.a.priv.refc.nr_ops = nr_ops;
1135 rcur->bc_private.a.priv.refc.shape_changes = shape_changes;
1136 }
1137 *pcur = rcur;
1138
1139 switch (type) {
1140 case XFS_REFCOUNT_INCREASE:
1141 error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1142 new_len, XFS_REFCOUNT_ADJUST_INCREASE, NULL);
1143 *new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1144 break;
1145 case XFS_REFCOUNT_DECREASE:
1146 error = xfs_refcount_adjust(rcur, bno, blockcount, &new_agbno,
1147 new_len, XFS_REFCOUNT_ADJUST_DECREASE, NULL);
1148 *new_fsb = XFS_AGB_TO_FSB(mp, agno, new_agbno);
1149 break;
1150 case XFS_REFCOUNT_ALLOC_COW:
1151 *new_fsb = startblock + blockcount;
1152 *new_len = 0;
1153 error = __xfs_refcount_cow_alloc(rcur, bno, blockcount);
1154 break;
1155 case XFS_REFCOUNT_FREE_COW:
1156 *new_fsb = startblock + blockcount;
1157 *new_len = 0;
1158 error = __xfs_refcount_cow_free(rcur, bno, blockcount);
1159 break;
1160 default:
1161 ASSERT(0);
1162 error = -EFSCORRUPTED;
1163 }
1164 if (!error && *new_len > 0)
1165 trace_xfs_refcount_finish_one_leftover(mp, agno, type,
1166 bno, blockcount, new_agbno, *new_len);
1167 return error;
1168
1169 out_cur:
1170 xfs_trans_brelse(tp, agbp);
1171
1172 return error;
1173 }
1174
1175 /*
1176 * Record a refcount intent for later processing.
1177 */
1178 static int
1179 __xfs_refcount_add(
1180 struct xfs_trans *tp,
1181 enum xfs_refcount_intent_type type,
1182 xfs_fsblock_t startblock,
1183 xfs_extlen_t blockcount)
1184 {
1185 struct xfs_refcount_intent *ri;
1186
1187 trace_xfs_refcount_defer(tp->t_mountp,
1188 XFS_FSB_TO_AGNO(tp->t_mountp, startblock),
1189 type, XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
1190 blockcount);
1191
1192 ri = kmem_alloc(sizeof(struct xfs_refcount_intent),
1193 KM_SLEEP | KM_NOFS);
1194 INIT_LIST_HEAD(&ri->ri_list);
1195 ri->ri_type = type;
1196 ri->ri_startblock = startblock;
1197 ri->ri_blockcount = blockcount;
1198
1199 xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_REFCOUNT, &ri->ri_list);
1200 return 0;
1201 }
1202
1203 /*
1204 * Increase the reference count of the blocks backing a file's extent.
1205 */
1206 int
1207 xfs_refcount_increase_extent(
1208 struct xfs_trans *tp,
1209 struct xfs_bmbt_irec *PREV)
1210 {
1211 if (!xfs_sb_version_hasreflink(&tp->t_mountp->m_sb))
1212 return 0;
1213
1214 return __xfs_refcount_add(tp, XFS_REFCOUNT_INCREASE,
1215 PREV->br_startblock, PREV->br_blockcount);
1216 }
1217
1218 /*
1219 * Decrease the reference count of the blocks backing a file's extent.
1220 */
1221 int
1222 xfs_refcount_decrease_extent(
1223 struct xfs_trans *tp,
1224 struct xfs_bmbt_irec *PREV)
1225 {
1226 if (!xfs_sb_version_hasreflink(&tp->t_mountp->m_sb))
1227 return 0;
1228
1229 return __xfs_refcount_add(tp, XFS_REFCOUNT_DECREASE,
1230 PREV->br_startblock, PREV->br_blockcount);
1231 }
1232
1233 /*
1234 * Given an AG extent, find the lowest-numbered run of shared blocks
1235 * within that range and return the range in fbno/flen. If
1236 * find_end_of_shared is set, return the longest contiguous extent of
1237 * shared blocks; if not, just return the first extent we find. If no
1238 * shared blocks are found, fbno and flen will be set to NULLAGBLOCK
1239 * and 0, respectively.
1240 */
1241 int
1242 xfs_refcount_find_shared(
1243 struct xfs_btree_cur *cur,
1244 xfs_agblock_t agbno,
1245 xfs_extlen_t aglen,
1246 xfs_agblock_t *fbno,
1247 xfs_extlen_t *flen,
1248 bool find_end_of_shared)
1249 {
1250 struct xfs_refcount_irec tmp;
1251 int i;
1252 int have;
1253 int error;
1254
1255 trace_xfs_refcount_find_shared(cur->bc_mp, cur->bc_private.a.agno,
1256 agbno, aglen);
1257
1258 /* By default, skip the whole range */
1259 *fbno = NULLAGBLOCK;
1260 *flen = 0;
1261
1262 /* Try to find a refcount extent that crosses the start */
1263 error = xfs_refcount_lookup_le(cur, agbno, &have);
1264 if (error)
1265 goto out_error;
1266 if (!have) {
1267 /* No left extent, look at the next one */
1268 error = xfs_btree_increment(cur, 0, &have);
1269 if (error)
1270 goto out_error;
1271 if (!have)
1272 goto done;
1273 }
1274 error = xfs_refcount_get_rec(cur, &tmp, &i);
1275 if (error)
1276 goto out_error;
1277 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1278
1279 /* If the extent ends before the start, look at the next one */
1280 if (tmp.rc_startblock + tmp.rc_blockcount <= agbno) {
1281 error = xfs_btree_increment(cur, 0, &have);
1282 if (error)
1283 goto out_error;
1284 if (!have)
1285 goto done;
1286 error = xfs_refcount_get_rec(cur, &tmp, &i);
1287 if (error)
1288 goto out_error;
1289 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1290 }
1291
1292 /* If the extent starts after the range we want, bail out */
1293 if (tmp.rc_startblock >= agbno + aglen)
1294 goto done;
1295
1296 /* We found the start of a shared extent! */
1297 if (tmp.rc_startblock < agbno) {
1298 tmp.rc_blockcount -= (agbno - tmp.rc_startblock);
1299 tmp.rc_startblock = agbno;
1300 }
1301
1302 *fbno = tmp.rc_startblock;
1303 *flen = min(tmp.rc_blockcount, agbno + aglen - *fbno);
1304 if (!find_end_of_shared)
1305 goto done;
1306
1307 /* Otherwise, find the end of this shared extent */
1308 while (*fbno + *flen < agbno + aglen) {
1309 error = xfs_btree_increment(cur, 0, &have);
1310 if (error)
1311 goto out_error;
1312 if (!have)
1313 break;
1314 error = xfs_refcount_get_rec(cur, &tmp, &i);
1315 if (error)
1316 goto out_error;
1317 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, out_error);
1318 if (tmp.rc_startblock >= agbno + aglen ||
1319 tmp.rc_startblock != *fbno + *flen)
1320 break;
1321 *flen = min(*flen + tmp.rc_blockcount, agbno + aglen - *fbno);
1322 }
1323
1324 done:
1325 trace_xfs_refcount_find_shared_result(cur->bc_mp,
1326 cur->bc_private.a.agno, *fbno, *flen);
1327
1328 out_error:
1329 if (error)
1330 trace_xfs_refcount_find_shared_error(cur->bc_mp,
1331 cur->bc_private.a.agno, error, _RET_IP_);
1332 return error;
1333 }
1334
1335 /*
1336 * Recovering CoW Blocks After a Crash
1337 *
1338 * Due to the way that the copy on write mechanism works, there's a window of
1339 * opportunity in which we can lose track of allocated blocks during a crash.
1340 * Because CoW uses delayed allocation in the in-core CoW fork, writeback
1341 * causes blocks to be allocated and stored in the CoW fork. The blocks are
1342 * no longer in the free space btree but are not otherwise recorded anywhere
1343 * until the write completes and the blocks are mapped into the file. A crash
1344 * in between allocation and remapping results in the replacement blocks being
1345 * lost. This situation is exacerbated by the CoW extent size hint because
1346 * allocations can hang around for long time.
1347 *
1348 * However, there is a place where we can record these allocations before they
1349 * become mappings -- the reference count btree. The btree does not record
1350 * extents with refcount == 1, so we can record allocations with a refcount of
1351 * 1. Blocks being used for CoW writeout cannot be shared, so there should be
1352 * no conflict with shared block records. These mappings should be created
1353 * when we allocate blocks to the CoW fork and deleted when they're removed
1354 * from the CoW fork.
1355 *
1356 * Minor nit: records for in-progress CoW allocations and records for shared
1357 * extents must never be merged, to preserve the property that (except for CoW
1358 * allocations) there are no refcount btree entries with refcount == 1. The
1359 * only time this could potentially happen is when unsharing a block that's
1360 * adjacent to CoW allocations, so we must be careful to avoid this.
1361 *
1362 * At mount time we recover lost CoW allocations by searching the refcount
1363 * btree for these refcount == 1 mappings. These represent CoW allocations
1364 * that were in progress at the time the filesystem went down, so we can free
1365 * them to get the space back.
1366 *
1367 * This mechanism is superior to creating EFIs for unmapped CoW extents for
1368 * several reasons -- first, EFIs pin the tail of the log and would have to be
1369 * periodically relogged to avoid filling up the log. Second, CoW completions
1370 * will have to file an EFD and create new EFIs for whatever remains in the
1371 * CoW fork; this partially takes care of (1) but extent-size reservations
1372 * will have to periodically relog even if there's no writeout in progress.
1373 * This can happen if the CoW extent size hint is set, which you really want.
1374 * Third, EFIs cannot currently be automatically relogged into newer
1375 * transactions to advance the log tail. Fourth, stuffing the log full of
1376 * EFIs places an upper bound on the number of CoW allocations that can be
1377 * held filesystem-wide at any given time. Recording them in the refcount
1378 * btree doesn't require us to maintain any state in memory and doesn't pin
1379 * the log.
1380 */
1381 /*
1382 * Adjust the refcounts of CoW allocations. These allocations are "magic"
1383 * in that they're not referenced anywhere else in the filesystem, so we
1384 * stash them in the refcount btree with a refcount of 1 until either file
1385 * remapping (or CoW cancellation) happens.
1386 */
1387 STATIC int
1388 xfs_refcount_adjust_cow_extents(
1389 struct xfs_btree_cur *cur,
1390 xfs_agblock_t agbno,
1391 xfs_extlen_t aglen,
1392 enum xfs_refc_adjust_op adj)
1393 {
1394 struct xfs_refcount_irec ext, tmp;
1395 int error;
1396 int found_rec, found_tmp;
1397
1398 if (aglen == 0)
1399 return 0;
1400
1401 /* Find any overlapping refcount records */
1402 error = xfs_refcount_lookup_ge(cur, agbno, &found_rec);
1403 if (error)
1404 goto out_error;
1405 error = xfs_refcount_get_rec(cur, &ext, &found_rec);
1406 if (error)
1407 goto out_error;
1408 if (!found_rec) {
1409 ext.rc_startblock = cur->bc_mp->m_sb.sb_agblocks +
1410 XFS_REFC_COW_START;
1411 ext.rc_blockcount = 0;
1412 ext.rc_refcount = 0;
1413 }
1414
1415 switch (adj) {
1416 case XFS_REFCOUNT_ADJUST_COW_ALLOC:
1417 /* Adding a CoW reservation, there should be nothing here. */
1418 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1419 ext.rc_startblock >= agbno + aglen, out_error);
1420
1421 tmp.rc_startblock = agbno;
1422 tmp.rc_blockcount = aglen;
1423 tmp.rc_refcount = 1;
1424 trace_xfs_refcount_modify_extent(cur->bc_mp,
1425 cur->bc_private.a.agno, &tmp);
1426
1427 error = xfs_refcount_insert(cur, &tmp,
1428 &found_tmp);
1429 if (error)
1430 goto out_error;
1431 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1432 found_tmp == 1, out_error);
1433 break;
1434 case XFS_REFCOUNT_ADJUST_COW_FREE:
1435 /* Removing a CoW reservation, there should be one extent. */
1436 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1437 ext.rc_startblock == agbno, out_error);
1438 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1439 ext.rc_blockcount == aglen, out_error);
1440 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1441 ext.rc_refcount == 1, out_error);
1442
1443 ext.rc_refcount = 0;
1444 trace_xfs_refcount_modify_extent(cur->bc_mp,
1445 cur->bc_private.a.agno, &ext);
1446 error = xfs_refcount_delete(cur, &found_rec);
1447 if (error)
1448 goto out_error;
1449 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp,
1450 found_rec == 1, out_error);
1451 break;
1452 default:
1453 ASSERT(0);
1454 }
1455
1456 return error;
1457 out_error:
1458 trace_xfs_refcount_modify_extent_error(cur->bc_mp,
1459 cur->bc_private.a.agno, error, _RET_IP_);
1460 return error;
1461 }
1462
1463 /*
1464 * Add or remove refcount btree entries for CoW reservations.
1465 */
1466 STATIC int
1467 xfs_refcount_adjust_cow(
1468 struct xfs_btree_cur *cur,
1469 xfs_agblock_t agbno,
1470 xfs_extlen_t aglen,
1471 enum xfs_refc_adjust_op adj)
1472 {
1473 bool shape_changed;
1474 int error;
1475
1476 agbno += XFS_REFC_COW_START;
1477
1478 /*
1479 * Ensure that no rcextents cross the boundary of the adjustment range.
1480 */
1481 error = xfs_refcount_split_extent(cur, agbno, &shape_changed);
1482 if (error)
1483 goto out_error;
1484
1485 error = xfs_refcount_split_extent(cur, agbno + aglen, &shape_changed);
1486 if (error)
1487 goto out_error;
1488
1489 /*
1490 * Try to merge with the left or right extents of the range.
1491 */
1492 error = xfs_refcount_merge_extents(cur, &agbno, &aglen, adj,
1493 XFS_FIND_RCEXT_COW, &shape_changed);
1494 if (error)
1495 goto out_error;
1496
1497 /* Now that we've taken care of the ends, adjust the middle extents */
1498 error = xfs_refcount_adjust_cow_extents(cur, agbno, aglen, adj);
1499 if (error)
1500 goto out_error;
1501
1502 return 0;
1503
1504 out_error:
1505 trace_xfs_refcount_adjust_cow_error(cur->bc_mp, cur->bc_private.a.agno,
1506 error, _RET_IP_);
1507 return error;
1508 }
1509
1510 /*
1511 * Record a CoW allocation in the refcount btree.
1512 */
1513 STATIC int
1514 __xfs_refcount_cow_alloc(
1515 struct xfs_btree_cur *rcur,
1516 xfs_agblock_t agbno,
1517 xfs_extlen_t aglen)
1518 {
1519 trace_xfs_refcount_cow_increase(rcur->bc_mp, rcur->bc_private.a.agno,
1520 agbno, aglen);
1521
1522 /* Add refcount btree reservation */
1523 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1524 XFS_REFCOUNT_ADJUST_COW_ALLOC);
1525 }
1526
1527 /*
1528 * Remove a CoW allocation from the refcount btree.
1529 */
1530 STATIC int
1531 __xfs_refcount_cow_free(
1532 struct xfs_btree_cur *rcur,
1533 xfs_agblock_t agbno,
1534 xfs_extlen_t aglen)
1535 {
1536 trace_xfs_refcount_cow_decrease(rcur->bc_mp, rcur->bc_private.a.agno,
1537 agbno, aglen);
1538
1539 /* Remove refcount btree reservation */
1540 return xfs_refcount_adjust_cow(rcur, agbno, aglen,
1541 XFS_REFCOUNT_ADJUST_COW_FREE);
1542 }
1543
1544 /* Record a CoW staging extent in the refcount btree. */
1545 int
1546 xfs_refcount_alloc_cow_extent(
1547 struct xfs_trans *tp,
1548 xfs_fsblock_t fsb,
1549 xfs_extlen_t len)
1550 {
1551 struct xfs_mount *mp = tp->t_mountp;
1552 int error;
1553
1554 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1555 return 0;
1556
1557 error = __xfs_refcount_add(tp, XFS_REFCOUNT_ALLOC_COW, fsb, len);
1558 if (error)
1559 return error;
1560
1561 /* Add rmap entry */
1562 return xfs_rmap_alloc_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1563 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1564 }
1565
1566 /* Forget a CoW staging event in the refcount btree. */
1567 int
1568 xfs_refcount_free_cow_extent(
1569 struct xfs_trans *tp,
1570 xfs_fsblock_t fsb,
1571 xfs_extlen_t len)
1572 {
1573 struct xfs_mount *mp = tp->t_mountp;
1574 int error;
1575
1576 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1577 return 0;
1578
1579 /* Remove rmap entry */
1580 error = xfs_rmap_free_extent(tp, XFS_FSB_TO_AGNO(mp, fsb),
1581 XFS_FSB_TO_AGBNO(mp, fsb), len, XFS_RMAP_OWN_COW);
1582 if (error)
1583 return error;
1584
1585 return __xfs_refcount_add(tp, XFS_REFCOUNT_FREE_COW, fsb, len);
1586 }
1587
1588 struct xfs_refcount_recovery {
1589 struct list_head rr_list;
1590 struct xfs_refcount_irec rr_rrec;
1591 };
1592
1593 /* Stuff an extent on the recovery list. */
1594 STATIC int
1595 xfs_refcount_recover_extent(
1596 struct xfs_btree_cur *cur,
1597 union xfs_btree_rec *rec,
1598 void *priv)
1599 {
1600 struct list_head *debris = priv;
1601 struct xfs_refcount_recovery *rr;
1602
1603 if (be32_to_cpu(rec->refc.rc_refcount) != 1)
1604 return -EFSCORRUPTED;
1605
1606 rr = kmem_alloc(sizeof(struct xfs_refcount_recovery), KM_SLEEP);
1607 xfs_refcount_btrec_to_irec(rec, &rr->rr_rrec);
1608 list_add_tail(&rr->rr_list, debris);
1609
1610 return 0;
1611 }
1612
1613 /* Find and remove leftover CoW reservations. */
1614 int
1615 xfs_refcount_recover_cow_leftovers(
1616 struct xfs_mount *mp,
1617 xfs_agnumber_t agno)
1618 {
1619 struct xfs_trans *tp;
1620 struct xfs_btree_cur *cur;
1621 struct xfs_buf *agbp;
1622 struct xfs_refcount_recovery *rr, *n;
1623 struct list_head debris;
1624 union xfs_btree_irec low;
1625 union xfs_btree_irec high;
1626 xfs_fsblock_t fsb;
1627 xfs_agblock_t agbno;
1628 int error;
1629
1630 if (mp->m_sb.sb_agblocks >= XFS_REFC_COW_START)
1631 return -EOPNOTSUPP;
1632
1633 INIT_LIST_HEAD(&debris);
1634
1635 /*
1636 * In this first part, we use an empty transaction to gather up
1637 * all the leftover CoW extents so that we can subsequently
1638 * delete them. The empty transaction is used to avoid
1639 * a buffer lock deadlock if there happens to be a loop in the
1640 * refcountbt because we're allowed to re-grab a buffer that is
1641 * already attached to our transaction. When we're done
1642 * recording the CoW debris we cancel the (empty) transaction
1643 * and everything goes away cleanly.
1644 */
1645 error = xfs_trans_alloc_empty(mp, &tp);
1646 if (error)
1647 return error;
1648
1649 error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
1650 if (error)
1651 goto out_trans;
1652 if (!agbp) {
1653 error = -ENOMEM;
1654 goto out_trans;
1655 }
1656 cur = xfs_refcountbt_init_cursor(mp, tp, agbp, agno);
1657
1658 /* Find all the leftover CoW staging extents. */
1659 memset(&low, 0, sizeof(low));
1660 memset(&high, 0, sizeof(high));
1661 low.rc.rc_startblock = XFS_REFC_COW_START;
1662 high.rc.rc_startblock = -1U;
1663 error = xfs_btree_query_range(cur, &low, &high,
1664 xfs_refcount_recover_extent, &debris);
1665 xfs_btree_del_cursor(cur, error);
1666 xfs_trans_brelse(tp, agbp);
1667 xfs_trans_cancel(tp);
1668 if (error)
1669 goto out_free;
1670
1671 /* Now iterate the list to free the leftovers */
1672 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1673 /* Set up transaction. */
1674 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1675 if (error)
1676 goto out_free;
1677
1678 trace_xfs_refcount_recover_extent(mp, agno, &rr->rr_rrec);
1679
1680 /* Free the orphan record */
1681 agbno = rr->rr_rrec.rc_startblock - XFS_REFC_COW_START;
1682 fsb = XFS_AGB_TO_FSB(mp, agno, agbno);
1683 error = xfs_refcount_free_cow_extent(tp, fsb,
1684 rr->rr_rrec.rc_blockcount);
1685 if (error)
1686 goto out_trans;
1687
1688 /* Free the block. */
1689 xfs_bmap_add_free(tp, fsb, rr->rr_rrec.rc_blockcount, NULL);
1690
1691 error = xfs_trans_commit(tp);
1692 if (error)
1693 goto out_free;
1694
1695 list_del(&rr->rr_list);
1696 kmem_free(rr);
1697 }
1698
1699 return error;
1700 out_trans:
1701 xfs_trans_cancel(tp);
1702 out_free:
1703 /* Free the leftover list */
1704 list_for_each_entry_safe(rr, n, &debris, rr_list) {
1705 list_del(&rr->rr_list);
1706 kmem_free(rr);
1707 }
1708 return error;
1709 }
1710
1711 /* Is there a record covering a given extent? */
1712 int
1713 xfs_refcount_has_record(
1714 struct xfs_btree_cur *cur,
1715 xfs_agblock_t bno,
1716 xfs_extlen_t len,
1717 bool *exists)
1718 {
1719 union xfs_btree_irec low;
1720 union xfs_btree_irec high;
1721
1722 memset(&low, 0, sizeof(low));
1723 low.rc.rc_startblock = bno;
1724 memset(&high, 0xFF, sizeof(high));
1725 high.rc.rc_startblock = bno + len - 1;
1726
1727 return xfs_btree_has_record(cur, &low, &high, exists);
1728 }