xref: /openbmc/linux/fs/xfs/libxfs/xfs_rmap_btree.c (revision caa80275)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2014 Red Hat, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.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_mount.h"
13 #include "xfs_trans.h"
14 #include "xfs_alloc.h"
15 #include "xfs_btree.h"
16 #include "xfs_btree_staging.h"
17 #include "xfs_rmap.h"
18 #include "xfs_rmap_btree.h"
19 #include "xfs_trace.h"
20 #include "xfs_error.h"
21 #include "xfs_extent_busy.h"
22 #include "xfs_ag.h"
23 #include "xfs_ag_resv.h"
24 
25 /*
26  * Reverse map btree.
27  *
28  * This is a per-ag tree used to track the owner(s) of a given extent. With
29  * reflink it is possible for there to be multiple owners, which is a departure
30  * from classic XFS. Owner records for data extents are inserted when the
31  * extent is mapped and removed when an extent is unmapped.  Owner records for
32  * all other block types (i.e. metadata) are inserted when an extent is
33  * allocated and removed when an extent is freed. There can only be one owner
34  * of a metadata extent, usually an inode or some other metadata structure like
35  * an AG btree.
36  *
37  * The rmap btree is part of the free space management, so blocks for the tree
38  * are sourced from the agfl. Hence we need transaction reservation support for
39  * this tree so that the freelist is always large enough. This also impacts on
40  * the minimum space we need to leave free in the AG.
41  *
42  * The tree is ordered by [ag block, owner, offset]. This is a large key size,
43  * but it is the only way to enforce unique keys when a block can be owned by
44  * multiple files at any offset. There's no need to order/search by extent
45  * size for online updating/management of the tree. It is intended that most
46  * reverse lookups will be to find the owner(s) of a particular block, or to
47  * try to recover tree and file data from corrupt primary metadata.
48  */
49 
50 static struct xfs_btree_cur *
51 xfs_rmapbt_dup_cursor(
52 	struct xfs_btree_cur	*cur)
53 {
54 	return xfs_rmapbt_init_cursor(cur->bc_mp, cur->bc_tp,
55 				cur->bc_ag.agbp, cur->bc_ag.pag);
56 }
57 
58 STATIC void
59 xfs_rmapbt_set_root(
60 	struct xfs_btree_cur		*cur,
61 	const union xfs_btree_ptr	*ptr,
62 	int				inc)
63 {
64 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
65 	struct xfs_agf		*agf = agbp->b_addr;
66 	int			btnum = cur->bc_btnum;
67 
68 	ASSERT(ptr->s != 0);
69 
70 	agf->agf_roots[btnum] = ptr->s;
71 	be32_add_cpu(&agf->agf_levels[btnum], inc);
72 	cur->bc_ag.pag->pagf_levels[btnum] += inc;
73 
74 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
75 }
76 
77 STATIC int
78 xfs_rmapbt_alloc_block(
79 	struct xfs_btree_cur		*cur,
80 	const union xfs_btree_ptr	*start,
81 	union xfs_btree_ptr		*new,
82 	int				*stat)
83 {
84 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
85 	struct xfs_agf		*agf = agbp->b_addr;
86 	struct xfs_perag	*pag = cur->bc_ag.pag;
87 	int			error;
88 	xfs_agblock_t		bno;
89 
90 	/* Allocate the new block from the freelist. If we can't, give up.  */
91 	error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_ag.agbp,
92 				       &bno, 1);
93 	if (error)
94 		return error;
95 
96 	trace_xfs_rmapbt_alloc_block(cur->bc_mp, pag->pag_agno, bno, 1);
97 	if (bno == NULLAGBLOCK) {
98 		*stat = 0;
99 		return 0;
100 	}
101 
102 	xfs_extent_busy_reuse(cur->bc_mp, pag, bno, 1, false);
103 
104 	new->s = cpu_to_be32(bno);
105 	be32_add_cpu(&agf->agf_rmap_blocks, 1);
106 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_RMAP_BLOCKS);
107 
108 	xfs_ag_resv_rmapbt_alloc(cur->bc_mp, pag->pag_agno);
109 
110 	*stat = 1;
111 	return 0;
112 }
113 
114 STATIC int
115 xfs_rmapbt_free_block(
116 	struct xfs_btree_cur	*cur,
117 	struct xfs_buf		*bp)
118 {
119 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
120 	struct xfs_agf		*agf = agbp->b_addr;
121 	struct xfs_perag	*pag = cur->bc_ag.pag;
122 	xfs_agblock_t		bno;
123 	int			error;
124 
125 	bno = xfs_daddr_to_agbno(cur->bc_mp, xfs_buf_daddr(bp));
126 	trace_xfs_rmapbt_free_block(cur->bc_mp, pag->pag_agno,
127 			bno, 1);
128 	be32_add_cpu(&agf->agf_rmap_blocks, -1);
129 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_RMAP_BLOCKS);
130 	error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
131 	if (error)
132 		return error;
133 
134 	xfs_extent_busy_insert(cur->bc_tp, pag, bno, 1,
135 			      XFS_EXTENT_BUSY_SKIP_DISCARD);
136 
137 	xfs_ag_resv_free_extent(pag, XFS_AG_RESV_RMAPBT, NULL, 1);
138 	return 0;
139 }
140 
141 STATIC int
142 xfs_rmapbt_get_minrecs(
143 	struct xfs_btree_cur	*cur,
144 	int			level)
145 {
146 	return cur->bc_mp->m_rmap_mnr[level != 0];
147 }
148 
149 STATIC int
150 xfs_rmapbt_get_maxrecs(
151 	struct xfs_btree_cur	*cur,
152 	int			level)
153 {
154 	return cur->bc_mp->m_rmap_mxr[level != 0];
155 }
156 
157 STATIC void
158 xfs_rmapbt_init_key_from_rec(
159 	union xfs_btree_key		*key,
160 	const union xfs_btree_rec	*rec)
161 {
162 	key->rmap.rm_startblock = rec->rmap.rm_startblock;
163 	key->rmap.rm_owner = rec->rmap.rm_owner;
164 	key->rmap.rm_offset = rec->rmap.rm_offset;
165 }
166 
167 /*
168  * The high key for a reverse mapping record can be computed by shifting
169  * the startblock and offset to the highest value that would still map
170  * to that record.  In practice this means that we add blockcount-1 to
171  * the startblock for all records, and if the record is for a data/attr
172  * fork mapping, we add blockcount-1 to the offset too.
173  */
174 STATIC void
175 xfs_rmapbt_init_high_key_from_rec(
176 	union xfs_btree_key		*key,
177 	const union xfs_btree_rec	*rec)
178 {
179 	uint64_t			off;
180 	int				adj;
181 
182 	adj = be32_to_cpu(rec->rmap.rm_blockcount) - 1;
183 
184 	key->rmap.rm_startblock = rec->rmap.rm_startblock;
185 	be32_add_cpu(&key->rmap.rm_startblock, adj);
186 	key->rmap.rm_owner = rec->rmap.rm_owner;
187 	key->rmap.rm_offset = rec->rmap.rm_offset;
188 	if (XFS_RMAP_NON_INODE_OWNER(be64_to_cpu(rec->rmap.rm_owner)) ||
189 	    XFS_RMAP_IS_BMBT_BLOCK(be64_to_cpu(rec->rmap.rm_offset)))
190 		return;
191 	off = be64_to_cpu(key->rmap.rm_offset);
192 	off = (XFS_RMAP_OFF(off) + adj) | (off & ~XFS_RMAP_OFF_MASK);
193 	key->rmap.rm_offset = cpu_to_be64(off);
194 }
195 
196 STATIC void
197 xfs_rmapbt_init_rec_from_cur(
198 	struct xfs_btree_cur	*cur,
199 	union xfs_btree_rec	*rec)
200 {
201 	rec->rmap.rm_startblock = cpu_to_be32(cur->bc_rec.r.rm_startblock);
202 	rec->rmap.rm_blockcount = cpu_to_be32(cur->bc_rec.r.rm_blockcount);
203 	rec->rmap.rm_owner = cpu_to_be64(cur->bc_rec.r.rm_owner);
204 	rec->rmap.rm_offset = cpu_to_be64(
205 			xfs_rmap_irec_offset_pack(&cur->bc_rec.r));
206 }
207 
208 STATIC void
209 xfs_rmapbt_init_ptr_from_cur(
210 	struct xfs_btree_cur	*cur,
211 	union xfs_btree_ptr	*ptr)
212 {
213 	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
214 
215 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
216 
217 	ptr->s = agf->agf_roots[cur->bc_btnum];
218 }
219 
220 STATIC int64_t
221 xfs_rmapbt_key_diff(
222 	struct xfs_btree_cur		*cur,
223 	const union xfs_btree_key	*key)
224 {
225 	struct xfs_rmap_irec		*rec = &cur->bc_rec.r;
226 	const struct xfs_rmap_key	*kp = &key->rmap;
227 	__u64				x, y;
228 	int64_t				d;
229 
230 	d = (int64_t)be32_to_cpu(kp->rm_startblock) - rec->rm_startblock;
231 	if (d)
232 		return d;
233 
234 	x = be64_to_cpu(kp->rm_owner);
235 	y = rec->rm_owner;
236 	if (x > y)
237 		return 1;
238 	else if (y > x)
239 		return -1;
240 
241 	x = XFS_RMAP_OFF(be64_to_cpu(kp->rm_offset));
242 	y = rec->rm_offset;
243 	if (x > y)
244 		return 1;
245 	else if (y > x)
246 		return -1;
247 	return 0;
248 }
249 
250 STATIC int64_t
251 xfs_rmapbt_diff_two_keys(
252 	struct xfs_btree_cur		*cur,
253 	const union xfs_btree_key	*k1,
254 	const union xfs_btree_key	*k2)
255 {
256 	const struct xfs_rmap_key	*kp1 = &k1->rmap;
257 	const struct xfs_rmap_key	*kp2 = &k2->rmap;
258 	int64_t				d;
259 	__u64				x, y;
260 
261 	d = (int64_t)be32_to_cpu(kp1->rm_startblock) -
262 		       be32_to_cpu(kp2->rm_startblock);
263 	if (d)
264 		return d;
265 
266 	x = be64_to_cpu(kp1->rm_owner);
267 	y = be64_to_cpu(kp2->rm_owner);
268 	if (x > y)
269 		return 1;
270 	else if (y > x)
271 		return -1;
272 
273 	x = XFS_RMAP_OFF(be64_to_cpu(kp1->rm_offset));
274 	y = XFS_RMAP_OFF(be64_to_cpu(kp2->rm_offset));
275 	if (x > y)
276 		return 1;
277 	else if (y > x)
278 		return -1;
279 	return 0;
280 }
281 
282 static xfs_failaddr_t
283 xfs_rmapbt_verify(
284 	struct xfs_buf		*bp)
285 {
286 	struct xfs_mount	*mp = bp->b_mount;
287 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
288 	struct xfs_perag	*pag = bp->b_pag;
289 	xfs_failaddr_t		fa;
290 	unsigned int		level;
291 
292 	/*
293 	 * magic number and level verification
294 	 *
295 	 * During growfs operations, we can't verify the exact level or owner as
296 	 * the perag is not fully initialised and hence not attached to the
297 	 * buffer.  In this case, check against the maximum tree depth.
298 	 *
299 	 * Similarly, during log recovery we will have a perag structure
300 	 * attached, but the agf information will not yet have been initialised
301 	 * from the on disk AGF. Again, we can only check against maximum limits
302 	 * in this case.
303 	 */
304 	if (!xfs_verify_magic(bp, block->bb_magic))
305 		return __this_address;
306 
307 	if (!xfs_has_rmapbt(mp))
308 		return __this_address;
309 	fa = xfs_btree_sblock_v5hdr_verify(bp);
310 	if (fa)
311 		return fa;
312 
313 	level = be16_to_cpu(block->bb_level);
314 	if (pag && pag->pagf_init) {
315 		if (level >= pag->pagf_levels[XFS_BTNUM_RMAPi])
316 			return __this_address;
317 	} else if (level >= mp->m_rmap_maxlevels)
318 		return __this_address;
319 
320 	return xfs_btree_sblock_verify(bp, mp->m_rmap_mxr[level != 0]);
321 }
322 
323 static void
324 xfs_rmapbt_read_verify(
325 	struct xfs_buf	*bp)
326 {
327 	xfs_failaddr_t	fa;
328 
329 	if (!xfs_btree_sblock_verify_crc(bp))
330 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
331 	else {
332 		fa = xfs_rmapbt_verify(bp);
333 		if (fa)
334 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
335 	}
336 
337 	if (bp->b_error)
338 		trace_xfs_btree_corrupt(bp, _RET_IP_);
339 }
340 
341 static void
342 xfs_rmapbt_write_verify(
343 	struct xfs_buf	*bp)
344 {
345 	xfs_failaddr_t	fa;
346 
347 	fa = xfs_rmapbt_verify(bp);
348 	if (fa) {
349 		trace_xfs_btree_corrupt(bp, _RET_IP_);
350 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
351 		return;
352 	}
353 	xfs_btree_sblock_calc_crc(bp);
354 
355 }
356 
357 const struct xfs_buf_ops xfs_rmapbt_buf_ops = {
358 	.name			= "xfs_rmapbt",
359 	.magic			= { 0, cpu_to_be32(XFS_RMAP_CRC_MAGIC) },
360 	.verify_read		= xfs_rmapbt_read_verify,
361 	.verify_write		= xfs_rmapbt_write_verify,
362 	.verify_struct		= xfs_rmapbt_verify,
363 };
364 
365 STATIC int
366 xfs_rmapbt_keys_inorder(
367 	struct xfs_btree_cur		*cur,
368 	const union xfs_btree_key	*k1,
369 	const union xfs_btree_key	*k2)
370 {
371 	uint32_t		x;
372 	uint32_t		y;
373 	uint64_t		a;
374 	uint64_t		b;
375 
376 	x = be32_to_cpu(k1->rmap.rm_startblock);
377 	y = be32_to_cpu(k2->rmap.rm_startblock);
378 	if (x < y)
379 		return 1;
380 	else if (x > y)
381 		return 0;
382 	a = be64_to_cpu(k1->rmap.rm_owner);
383 	b = be64_to_cpu(k2->rmap.rm_owner);
384 	if (a < b)
385 		return 1;
386 	else if (a > b)
387 		return 0;
388 	a = XFS_RMAP_OFF(be64_to_cpu(k1->rmap.rm_offset));
389 	b = XFS_RMAP_OFF(be64_to_cpu(k2->rmap.rm_offset));
390 	if (a <= b)
391 		return 1;
392 	return 0;
393 }
394 
395 STATIC int
396 xfs_rmapbt_recs_inorder(
397 	struct xfs_btree_cur		*cur,
398 	const union xfs_btree_rec	*r1,
399 	const union xfs_btree_rec	*r2)
400 {
401 	uint32_t		x;
402 	uint32_t		y;
403 	uint64_t		a;
404 	uint64_t		b;
405 
406 	x = be32_to_cpu(r1->rmap.rm_startblock);
407 	y = be32_to_cpu(r2->rmap.rm_startblock);
408 	if (x < y)
409 		return 1;
410 	else if (x > y)
411 		return 0;
412 	a = be64_to_cpu(r1->rmap.rm_owner);
413 	b = be64_to_cpu(r2->rmap.rm_owner);
414 	if (a < b)
415 		return 1;
416 	else if (a > b)
417 		return 0;
418 	a = XFS_RMAP_OFF(be64_to_cpu(r1->rmap.rm_offset));
419 	b = XFS_RMAP_OFF(be64_to_cpu(r2->rmap.rm_offset));
420 	if (a <= b)
421 		return 1;
422 	return 0;
423 }
424 
425 static const struct xfs_btree_ops xfs_rmapbt_ops = {
426 	.rec_len		= sizeof(struct xfs_rmap_rec),
427 	.key_len		= 2 * sizeof(struct xfs_rmap_key),
428 
429 	.dup_cursor		= xfs_rmapbt_dup_cursor,
430 	.set_root		= xfs_rmapbt_set_root,
431 	.alloc_block		= xfs_rmapbt_alloc_block,
432 	.free_block		= xfs_rmapbt_free_block,
433 	.get_minrecs		= xfs_rmapbt_get_minrecs,
434 	.get_maxrecs		= xfs_rmapbt_get_maxrecs,
435 	.init_key_from_rec	= xfs_rmapbt_init_key_from_rec,
436 	.init_high_key_from_rec	= xfs_rmapbt_init_high_key_from_rec,
437 	.init_rec_from_cur	= xfs_rmapbt_init_rec_from_cur,
438 	.init_ptr_from_cur	= xfs_rmapbt_init_ptr_from_cur,
439 	.key_diff		= xfs_rmapbt_key_diff,
440 	.buf_ops		= &xfs_rmapbt_buf_ops,
441 	.diff_two_keys		= xfs_rmapbt_diff_two_keys,
442 	.keys_inorder		= xfs_rmapbt_keys_inorder,
443 	.recs_inorder		= xfs_rmapbt_recs_inorder,
444 };
445 
446 static struct xfs_btree_cur *
447 xfs_rmapbt_init_common(
448 	struct xfs_mount	*mp,
449 	struct xfs_trans	*tp,
450 	struct xfs_perag	*pag)
451 {
452 	struct xfs_btree_cur	*cur;
453 
454 	cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
455 	cur->bc_tp = tp;
456 	cur->bc_mp = mp;
457 	/* Overlapping btree; 2 keys per pointer. */
458 	cur->bc_btnum = XFS_BTNUM_RMAP;
459 	cur->bc_flags = XFS_BTREE_CRC_BLOCKS | XFS_BTREE_OVERLAPPING;
460 	cur->bc_blocklog = mp->m_sb.sb_blocklog;
461 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_rmap_2);
462 	cur->bc_ops = &xfs_rmapbt_ops;
463 
464 	/* take a reference for the cursor */
465 	atomic_inc(&pag->pag_ref);
466 	cur->bc_ag.pag = pag;
467 
468 	return cur;
469 }
470 
471 /* Create a new reverse mapping btree cursor. */
472 struct xfs_btree_cur *
473 xfs_rmapbt_init_cursor(
474 	struct xfs_mount	*mp,
475 	struct xfs_trans	*tp,
476 	struct xfs_buf		*agbp,
477 	struct xfs_perag	*pag)
478 {
479 	struct xfs_agf		*agf = agbp->b_addr;
480 	struct xfs_btree_cur	*cur;
481 
482 	cur = xfs_rmapbt_init_common(mp, tp, pag);
483 	cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]);
484 	cur->bc_ag.agbp = agbp;
485 	return cur;
486 }
487 
488 /* Create a new reverse mapping btree cursor with a fake root for staging. */
489 struct xfs_btree_cur *
490 xfs_rmapbt_stage_cursor(
491 	struct xfs_mount	*mp,
492 	struct xbtree_afakeroot	*afake,
493 	struct xfs_perag	*pag)
494 {
495 	struct xfs_btree_cur	*cur;
496 
497 	cur = xfs_rmapbt_init_common(mp, NULL, pag);
498 	xfs_btree_stage_afakeroot(cur, afake);
499 	return cur;
500 }
501 
502 /*
503  * Install a new reverse mapping btree root.  Caller is responsible for
504  * invalidating and freeing the old btree blocks.
505  */
506 void
507 xfs_rmapbt_commit_staged_btree(
508 	struct xfs_btree_cur	*cur,
509 	struct xfs_trans	*tp,
510 	struct xfs_buf		*agbp)
511 {
512 	struct xfs_agf		*agf = agbp->b_addr;
513 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
514 
515 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
516 
517 	agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root);
518 	agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels);
519 	agf->agf_rmap_blocks = cpu_to_be32(afake->af_blocks);
520 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS |
521 				    XFS_AGF_RMAP_BLOCKS);
522 	xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_rmapbt_ops);
523 }
524 
525 /*
526  * Calculate number of records in an rmap btree block.
527  */
528 int
529 xfs_rmapbt_maxrecs(
530 	int			blocklen,
531 	int			leaf)
532 {
533 	blocklen -= XFS_RMAP_BLOCK_LEN;
534 
535 	if (leaf)
536 		return blocklen / sizeof(struct xfs_rmap_rec);
537 	return blocklen /
538 		(2 * sizeof(struct xfs_rmap_key) + sizeof(xfs_rmap_ptr_t));
539 }
540 
541 /* Compute the maximum height of an rmap btree. */
542 void
543 xfs_rmapbt_compute_maxlevels(
544 	struct xfs_mount		*mp)
545 {
546 	/*
547 	 * On a non-reflink filesystem, the maximum number of rmap
548 	 * records is the number of blocks in the AG, hence the max
549 	 * rmapbt height is log_$maxrecs($agblocks).  However, with
550 	 * reflink each AG block can have up to 2^32 (per the refcount
551 	 * record format) owners, which means that theoretically we
552 	 * could face up to 2^64 rmap records.
553 	 *
554 	 * That effectively means that the max rmapbt height must be
555 	 * XFS_BTREE_MAXLEVELS.  "Fortunately" we'll run out of AG
556 	 * blocks to feed the rmapbt long before the rmapbt reaches
557 	 * maximum height.  The reflink code uses ag_resv_critical to
558 	 * disallow reflinking when less than 10% of the per-AG metadata
559 	 * block reservation since the fallback is a regular file copy.
560 	 */
561 	if (xfs_has_reflink(mp))
562 		mp->m_rmap_maxlevels = XFS_BTREE_MAXLEVELS;
563 	else
564 		mp->m_rmap_maxlevels = xfs_btree_compute_maxlevels(
565 				mp->m_rmap_mnr, mp->m_sb.sb_agblocks);
566 }
567 
568 /* Calculate the refcount btree size for some records. */
569 xfs_extlen_t
570 xfs_rmapbt_calc_size(
571 	struct xfs_mount	*mp,
572 	unsigned long long	len)
573 {
574 	return xfs_btree_calc_size(mp->m_rmap_mnr, len);
575 }
576 
577 /*
578  * Calculate the maximum refcount btree size.
579  */
580 xfs_extlen_t
581 xfs_rmapbt_max_size(
582 	struct xfs_mount	*mp,
583 	xfs_agblock_t		agblocks)
584 {
585 	/* Bail out if we're uninitialized, which can happen in mkfs. */
586 	if (mp->m_rmap_mxr[0] == 0)
587 		return 0;
588 
589 	return xfs_rmapbt_calc_size(mp, agblocks);
590 }
591 
592 /*
593  * Figure out how many blocks to reserve and how many are used by this btree.
594  */
595 int
596 xfs_rmapbt_calc_reserves(
597 	struct xfs_mount	*mp,
598 	struct xfs_trans	*tp,
599 	struct xfs_perag	*pag,
600 	xfs_extlen_t		*ask,
601 	xfs_extlen_t		*used)
602 {
603 	struct xfs_buf		*agbp;
604 	struct xfs_agf		*agf;
605 	xfs_agblock_t		agblocks;
606 	xfs_extlen_t		tree_len;
607 	int			error;
608 
609 	if (!xfs_has_rmapbt(mp))
610 		return 0;
611 
612 	error = xfs_alloc_read_agf(mp, tp, pag->pag_agno, 0, &agbp);
613 	if (error)
614 		return error;
615 
616 	agf = agbp->b_addr;
617 	agblocks = be32_to_cpu(agf->agf_length);
618 	tree_len = be32_to_cpu(agf->agf_rmap_blocks);
619 	xfs_trans_brelse(tp, agbp);
620 
621 	/*
622 	 * The log is permanently allocated, so the space it occupies will
623 	 * never be available for the kinds of things that would require btree
624 	 * expansion.  We therefore can pretend the space isn't there.
625 	 */
626 	if (mp->m_sb.sb_logstart &&
627 	    XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart) == pag->pag_agno)
628 		agblocks -= mp->m_sb.sb_logblocks;
629 
630 	/* Reserve 1% of the AG or enough for 1 block per record. */
631 	*ask += max(agblocks / 100, xfs_rmapbt_max_size(mp, agblocks));
632 	*used += tree_len;
633 
634 	return error;
635 }
636