xref: /openbmc/linux/fs/xfs/libxfs/xfs_alloc_btree.c (revision efac162a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2001,2005 Silicon Graphics, 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_btree.h"
14 #include "xfs_btree_staging.h"
15 #include "xfs_alloc_btree.h"
16 #include "xfs_alloc.h"
17 #include "xfs_extent_busy.h"
18 #include "xfs_error.h"
19 #include "xfs_trace.h"
20 #include "xfs_trans.h"
21 #include "xfs_ag.h"
22 
23 static struct kmem_cache	*xfs_allocbt_cur_cache;
24 
25 STATIC struct xfs_btree_cur *
26 xfs_allocbt_dup_cursor(
27 	struct xfs_btree_cur	*cur)
28 {
29 	return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
30 			cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
31 }
32 
33 STATIC void
34 xfs_allocbt_set_root(
35 	struct xfs_btree_cur		*cur,
36 	const union xfs_btree_ptr	*ptr,
37 	int				inc)
38 {
39 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
40 	struct xfs_agf		*agf = agbp->b_addr;
41 	int			btnum = cur->bc_btnum;
42 
43 	ASSERT(ptr->s != 0);
44 
45 	agf->agf_roots[btnum] = ptr->s;
46 	be32_add_cpu(&agf->agf_levels[btnum], inc);
47 	cur->bc_ag.pag->pagf_levels[btnum] += inc;
48 
49 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
50 }
51 
52 STATIC int
53 xfs_allocbt_alloc_block(
54 	struct xfs_btree_cur		*cur,
55 	const union xfs_btree_ptr	*start,
56 	union xfs_btree_ptr		*new,
57 	int				*stat)
58 {
59 	int			error;
60 	xfs_agblock_t		bno;
61 
62 	/* Allocate the new block from the freelist. If we can't, give up.  */
63 	error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_ag.agbp,
64 				       &bno, 1);
65 	if (error)
66 		return error;
67 
68 	if (bno == NULLAGBLOCK) {
69 		*stat = 0;
70 		return 0;
71 	}
72 
73 	atomic64_inc(&cur->bc_mp->m_allocbt_blks);
74 	xfs_extent_busy_reuse(cur->bc_mp, cur->bc_ag.agbp->b_pag, bno, 1, false);
75 
76 	new->s = cpu_to_be32(bno);
77 
78 	*stat = 1;
79 	return 0;
80 }
81 
82 STATIC int
83 xfs_allocbt_free_block(
84 	struct xfs_btree_cur	*cur,
85 	struct xfs_buf		*bp)
86 {
87 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
88 	xfs_agblock_t		bno;
89 	int			error;
90 
91 	bno = xfs_daddr_to_agbno(cur->bc_mp, xfs_buf_daddr(bp));
92 	error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
93 	if (error)
94 		return error;
95 
96 	atomic64_dec(&cur->bc_mp->m_allocbt_blks);
97 	xfs_extent_busy_insert(cur->bc_tp, agbp->b_pag, bno, 1,
98 			      XFS_EXTENT_BUSY_SKIP_DISCARD);
99 	return 0;
100 }
101 
102 /*
103  * Update the longest extent in the AGF
104  */
105 STATIC void
106 xfs_allocbt_update_lastrec(
107 	struct xfs_btree_cur		*cur,
108 	const struct xfs_btree_block	*block,
109 	const union xfs_btree_rec	*rec,
110 	int				ptr,
111 	int				reason)
112 {
113 	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
114 	struct xfs_perag	*pag;
115 	__be32			len;
116 	int			numrecs;
117 
118 	ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
119 
120 	switch (reason) {
121 	case LASTREC_UPDATE:
122 		/*
123 		 * If this is the last leaf block and it's the last record,
124 		 * then update the size of the longest extent in the AG.
125 		 */
126 		if (ptr != xfs_btree_get_numrecs(block))
127 			return;
128 		len = rec->alloc.ar_blockcount;
129 		break;
130 	case LASTREC_INSREC:
131 		if (be32_to_cpu(rec->alloc.ar_blockcount) <=
132 		    be32_to_cpu(agf->agf_longest))
133 			return;
134 		len = rec->alloc.ar_blockcount;
135 		break;
136 	case LASTREC_DELREC:
137 		numrecs = xfs_btree_get_numrecs(block);
138 		if (ptr <= numrecs)
139 			return;
140 		ASSERT(ptr == numrecs + 1);
141 
142 		if (numrecs) {
143 			xfs_alloc_rec_t *rrp;
144 
145 			rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
146 			len = rrp->ar_blockcount;
147 		} else {
148 			len = 0;
149 		}
150 
151 		break;
152 	default:
153 		ASSERT(0);
154 		return;
155 	}
156 
157 	agf->agf_longest = len;
158 	pag = cur->bc_ag.agbp->b_pag;
159 	pag->pagf_longest = be32_to_cpu(len);
160 	xfs_alloc_log_agf(cur->bc_tp, cur->bc_ag.agbp, XFS_AGF_LONGEST);
161 }
162 
163 STATIC int
164 xfs_allocbt_get_minrecs(
165 	struct xfs_btree_cur	*cur,
166 	int			level)
167 {
168 	return cur->bc_mp->m_alloc_mnr[level != 0];
169 }
170 
171 STATIC int
172 xfs_allocbt_get_maxrecs(
173 	struct xfs_btree_cur	*cur,
174 	int			level)
175 {
176 	return cur->bc_mp->m_alloc_mxr[level != 0];
177 }
178 
179 STATIC void
180 xfs_allocbt_init_key_from_rec(
181 	union xfs_btree_key		*key,
182 	const union xfs_btree_rec	*rec)
183 {
184 	key->alloc.ar_startblock = rec->alloc.ar_startblock;
185 	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
186 }
187 
188 STATIC void
189 xfs_bnobt_init_high_key_from_rec(
190 	union xfs_btree_key		*key,
191 	const union xfs_btree_rec	*rec)
192 {
193 	__u32				x;
194 
195 	x = be32_to_cpu(rec->alloc.ar_startblock);
196 	x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
197 	key->alloc.ar_startblock = cpu_to_be32(x);
198 	key->alloc.ar_blockcount = 0;
199 }
200 
201 STATIC void
202 xfs_cntbt_init_high_key_from_rec(
203 	union xfs_btree_key		*key,
204 	const union xfs_btree_rec	*rec)
205 {
206 	key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
207 	key->alloc.ar_startblock = 0;
208 }
209 
210 STATIC void
211 xfs_allocbt_init_rec_from_cur(
212 	struct xfs_btree_cur	*cur,
213 	union xfs_btree_rec	*rec)
214 {
215 	rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
216 	rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
217 }
218 
219 STATIC void
220 xfs_allocbt_init_ptr_from_cur(
221 	struct xfs_btree_cur	*cur,
222 	union xfs_btree_ptr	*ptr)
223 {
224 	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
225 
226 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
227 
228 	ptr->s = agf->agf_roots[cur->bc_btnum];
229 }
230 
231 STATIC int64_t
232 xfs_bnobt_key_diff(
233 	struct xfs_btree_cur		*cur,
234 	const union xfs_btree_key	*key)
235 {
236 	struct xfs_alloc_rec_incore	*rec = &cur->bc_rec.a;
237 	const struct xfs_alloc_rec	*kp = &key->alloc;
238 
239 	return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
240 }
241 
242 STATIC int64_t
243 xfs_cntbt_key_diff(
244 	struct xfs_btree_cur		*cur,
245 	const union xfs_btree_key	*key)
246 {
247 	struct xfs_alloc_rec_incore	*rec = &cur->bc_rec.a;
248 	const struct xfs_alloc_rec	*kp = &key->alloc;
249 	int64_t				diff;
250 
251 	diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
252 	if (diff)
253 		return diff;
254 
255 	return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
256 }
257 
258 STATIC int64_t
259 xfs_bnobt_diff_two_keys(
260 	struct xfs_btree_cur		*cur,
261 	const union xfs_btree_key	*k1,
262 	const union xfs_btree_key	*k2)
263 {
264 	return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
265 			  be32_to_cpu(k2->alloc.ar_startblock);
266 }
267 
268 STATIC int64_t
269 xfs_cntbt_diff_two_keys(
270 	struct xfs_btree_cur		*cur,
271 	const union xfs_btree_key	*k1,
272 	const union xfs_btree_key	*k2)
273 {
274 	int64_t				diff;
275 
276 	diff =  be32_to_cpu(k1->alloc.ar_blockcount) -
277 		be32_to_cpu(k2->alloc.ar_blockcount);
278 	if (diff)
279 		return diff;
280 
281 	return  be32_to_cpu(k1->alloc.ar_startblock) -
282 		be32_to_cpu(k2->alloc.ar_startblock);
283 }
284 
285 static xfs_failaddr_t
286 xfs_allocbt_verify(
287 	struct xfs_buf		*bp)
288 {
289 	struct xfs_mount	*mp = bp->b_mount;
290 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
291 	struct xfs_perag	*pag = bp->b_pag;
292 	xfs_failaddr_t		fa;
293 	unsigned int		level;
294 	xfs_btnum_t		btnum = XFS_BTNUM_BNOi;
295 
296 	if (!xfs_verify_magic(bp, block->bb_magic))
297 		return __this_address;
298 
299 	if (xfs_has_crc(mp)) {
300 		fa = xfs_btree_sblock_v5hdr_verify(bp);
301 		if (fa)
302 			return fa;
303 	}
304 
305 	/*
306 	 * The perag may not be attached during grow operations or fully
307 	 * initialized from the AGF during log recovery. Therefore we can only
308 	 * check against maximum tree depth from those contexts.
309 	 *
310 	 * Otherwise check against the per-tree limit. Peek at one of the
311 	 * verifier magic values to determine the type of tree we're verifying
312 	 * against.
313 	 */
314 	level = be16_to_cpu(block->bb_level);
315 	if (bp->b_ops->magic[0] == cpu_to_be32(XFS_ABTC_MAGIC))
316 		btnum = XFS_BTNUM_CNTi;
317 	if (pag && pag->pagf_init) {
318 		if (level >= pag->pagf_levels[btnum])
319 			return __this_address;
320 	} else if (level >= mp->m_alloc_maxlevels)
321 		return __this_address;
322 
323 	return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
324 }
325 
326 static void
327 xfs_allocbt_read_verify(
328 	struct xfs_buf	*bp)
329 {
330 	xfs_failaddr_t	fa;
331 
332 	if (!xfs_btree_sblock_verify_crc(bp))
333 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
334 	else {
335 		fa = xfs_allocbt_verify(bp);
336 		if (fa)
337 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
338 	}
339 
340 	if (bp->b_error)
341 		trace_xfs_btree_corrupt(bp, _RET_IP_);
342 }
343 
344 static void
345 xfs_allocbt_write_verify(
346 	struct xfs_buf	*bp)
347 {
348 	xfs_failaddr_t	fa;
349 
350 	fa = xfs_allocbt_verify(bp);
351 	if (fa) {
352 		trace_xfs_btree_corrupt(bp, _RET_IP_);
353 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
354 		return;
355 	}
356 	xfs_btree_sblock_calc_crc(bp);
357 
358 }
359 
360 const struct xfs_buf_ops xfs_bnobt_buf_ops = {
361 	.name = "xfs_bnobt",
362 	.magic = { cpu_to_be32(XFS_ABTB_MAGIC),
363 		   cpu_to_be32(XFS_ABTB_CRC_MAGIC) },
364 	.verify_read = xfs_allocbt_read_verify,
365 	.verify_write = xfs_allocbt_write_verify,
366 	.verify_struct = xfs_allocbt_verify,
367 };
368 
369 const struct xfs_buf_ops xfs_cntbt_buf_ops = {
370 	.name = "xfs_cntbt",
371 	.magic = { cpu_to_be32(XFS_ABTC_MAGIC),
372 		   cpu_to_be32(XFS_ABTC_CRC_MAGIC) },
373 	.verify_read = xfs_allocbt_read_verify,
374 	.verify_write = xfs_allocbt_write_verify,
375 	.verify_struct = xfs_allocbt_verify,
376 };
377 
378 STATIC int
379 xfs_bnobt_keys_inorder(
380 	struct xfs_btree_cur		*cur,
381 	const union xfs_btree_key	*k1,
382 	const union xfs_btree_key	*k2)
383 {
384 	return be32_to_cpu(k1->alloc.ar_startblock) <
385 	       be32_to_cpu(k2->alloc.ar_startblock);
386 }
387 
388 STATIC int
389 xfs_bnobt_recs_inorder(
390 	struct xfs_btree_cur		*cur,
391 	const union xfs_btree_rec	*r1,
392 	const union xfs_btree_rec	*r2)
393 {
394 	return be32_to_cpu(r1->alloc.ar_startblock) +
395 		be32_to_cpu(r1->alloc.ar_blockcount) <=
396 		be32_to_cpu(r2->alloc.ar_startblock);
397 }
398 
399 STATIC int
400 xfs_cntbt_keys_inorder(
401 	struct xfs_btree_cur		*cur,
402 	const union xfs_btree_key	*k1,
403 	const union xfs_btree_key	*k2)
404 {
405 	return be32_to_cpu(k1->alloc.ar_blockcount) <
406 		be32_to_cpu(k2->alloc.ar_blockcount) ||
407 		(k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
408 		 be32_to_cpu(k1->alloc.ar_startblock) <
409 		 be32_to_cpu(k2->alloc.ar_startblock));
410 }
411 
412 STATIC int
413 xfs_cntbt_recs_inorder(
414 	struct xfs_btree_cur		*cur,
415 	const union xfs_btree_rec	*r1,
416 	const union xfs_btree_rec	*r2)
417 {
418 	return be32_to_cpu(r1->alloc.ar_blockcount) <
419 		be32_to_cpu(r2->alloc.ar_blockcount) ||
420 		(r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
421 		 be32_to_cpu(r1->alloc.ar_startblock) <
422 		 be32_to_cpu(r2->alloc.ar_startblock));
423 }
424 
425 static const struct xfs_btree_ops xfs_bnobt_ops = {
426 	.rec_len		= sizeof(xfs_alloc_rec_t),
427 	.key_len		= sizeof(xfs_alloc_key_t),
428 
429 	.dup_cursor		= xfs_allocbt_dup_cursor,
430 	.set_root		= xfs_allocbt_set_root,
431 	.alloc_block		= xfs_allocbt_alloc_block,
432 	.free_block		= xfs_allocbt_free_block,
433 	.update_lastrec		= xfs_allocbt_update_lastrec,
434 	.get_minrecs		= xfs_allocbt_get_minrecs,
435 	.get_maxrecs		= xfs_allocbt_get_maxrecs,
436 	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
437 	.init_high_key_from_rec	= xfs_bnobt_init_high_key_from_rec,
438 	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
439 	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
440 	.key_diff		= xfs_bnobt_key_diff,
441 	.buf_ops		= &xfs_bnobt_buf_ops,
442 	.diff_two_keys		= xfs_bnobt_diff_two_keys,
443 	.keys_inorder		= xfs_bnobt_keys_inorder,
444 	.recs_inorder		= xfs_bnobt_recs_inorder,
445 };
446 
447 static const struct xfs_btree_ops xfs_cntbt_ops = {
448 	.rec_len		= sizeof(xfs_alloc_rec_t),
449 	.key_len		= sizeof(xfs_alloc_key_t),
450 
451 	.dup_cursor		= xfs_allocbt_dup_cursor,
452 	.set_root		= xfs_allocbt_set_root,
453 	.alloc_block		= xfs_allocbt_alloc_block,
454 	.free_block		= xfs_allocbt_free_block,
455 	.update_lastrec		= xfs_allocbt_update_lastrec,
456 	.get_minrecs		= xfs_allocbt_get_minrecs,
457 	.get_maxrecs		= xfs_allocbt_get_maxrecs,
458 	.init_key_from_rec	= xfs_allocbt_init_key_from_rec,
459 	.init_high_key_from_rec	= xfs_cntbt_init_high_key_from_rec,
460 	.init_rec_from_cur	= xfs_allocbt_init_rec_from_cur,
461 	.init_ptr_from_cur	= xfs_allocbt_init_ptr_from_cur,
462 	.key_diff		= xfs_cntbt_key_diff,
463 	.buf_ops		= &xfs_cntbt_buf_ops,
464 	.diff_two_keys		= xfs_cntbt_diff_two_keys,
465 	.keys_inorder		= xfs_cntbt_keys_inorder,
466 	.recs_inorder		= xfs_cntbt_recs_inorder,
467 };
468 
469 /* Allocate most of a new allocation btree cursor. */
470 STATIC struct xfs_btree_cur *
471 xfs_allocbt_init_common(
472 	struct xfs_mount	*mp,
473 	struct xfs_trans	*tp,
474 	struct xfs_perag	*pag,
475 	xfs_btnum_t		btnum)
476 {
477 	struct xfs_btree_cur	*cur;
478 
479 	ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
480 
481 	cur = xfs_btree_alloc_cursor(mp, tp, btnum, mp->m_alloc_maxlevels,
482 			xfs_allocbt_cur_cache);
483 	cur->bc_ag.abt.active = false;
484 
485 	if (btnum == XFS_BTNUM_CNT) {
486 		cur->bc_ops = &xfs_cntbt_ops;
487 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
488 		cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
489 	} else {
490 		cur->bc_ops = &xfs_bnobt_ops;
491 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
492 	}
493 
494 	/* take a reference for the cursor */
495 	atomic_inc(&pag->pag_ref);
496 	cur->bc_ag.pag = pag;
497 
498 	if (xfs_has_crc(mp))
499 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
500 
501 	return cur;
502 }
503 
504 /*
505  * Allocate a new allocation btree cursor.
506  */
507 struct xfs_btree_cur *			/* new alloc btree cursor */
508 xfs_allocbt_init_cursor(
509 	struct xfs_mount	*mp,		/* file system mount point */
510 	struct xfs_trans	*tp,		/* transaction pointer */
511 	struct xfs_buf		*agbp,		/* buffer for agf structure */
512 	struct xfs_perag	*pag,
513 	xfs_btnum_t		btnum)		/* btree identifier */
514 {
515 	struct xfs_agf		*agf = agbp->b_addr;
516 	struct xfs_btree_cur	*cur;
517 
518 	cur = xfs_allocbt_init_common(mp, tp, pag, btnum);
519 	if (btnum == XFS_BTNUM_CNT)
520 		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
521 	else
522 		cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
523 
524 	cur->bc_ag.agbp = agbp;
525 
526 	return cur;
527 }
528 
529 /* Create a free space btree cursor with a fake root for staging. */
530 struct xfs_btree_cur *
531 xfs_allocbt_stage_cursor(
532 	struct xfs_mount	*mp,
533 	struct xbtree_afakeroot	*afake,
534 	struct xfs_perag	*pag,
535 	xfs_btnum_t		btnum)
536 {
537 	struct xfs_btree_cur	*cur;
538 
539 	cur = xfs_allocbt_init_common(mp, NULL, pag, btnum);
540 	xfs_btree_stage_afakeroot(cur, afake);
541 	return cur;
542 }
543 
544 /*
545  * Install a new free space btree root.  Caller is responsible for invalidating
546  * and freeing the old btree blocks.
547  */
548 void
549 xfs_allocbt_commit_staged_btree(
550 	struct xfs_btree_cur	*cur,
551 	struct xfs_trans	*tp,
552 	struct xfs_buf		*agbp)
553 {
554 	struct xfs_agf		*agf = agbp->b_addr;
555 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
556 
557 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
558 
559 	agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root);
560 	agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels);
561 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
562 
563 	if (cur->bc_btnum == XFS_BTNUM_BNO) {
564 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_bnobt_ops);
565 	} else {
566 		cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE;
567 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_cntbt_ops);
568 	}
569 }
570 
571 /* Calculate number of records in an alloc btree block. */
572 static inline unsigned int
573 xfs_allocbt_block_maxrecs(
574 	unsigned int		blocklen,
575 	bool			leaf)
576 {
577 	if (leaf)
578 		return blocklen / sizeof(xfs_alloc_rec_t);
579 	return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
580 }
581 
582 /*
583  * Calculate number of records in an alloc btree block.
584  */
585 int
586 xfs_allocbt_maxrecs(
587 	struct xfs_mount	*mp,
588 	int			blocklen,
589 	int			leaf)
590 {
591 	blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
592 	return xfs_allocbt_block_maxrecs(blocklen, leaf);
593 }
594 
595 /* Free space btrees are at their largest when every other block is free. */
596 #define XFS_MAX_FREESP_RECORDS	((XFS_MAX_AG_BLOCKS + 1) / 2)
597 
598 /* Compute the max possible height for free space btrees. */
599 unsigned int
600 xfs_allocbt_maxlevels_ondisk(void)
601 {
602 	unsigned int		minrecs[2];
603 	unsigned int		blocklen;
604 
605 	blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
606 		       XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
607 
608 	minrecs[0] = xfs_allocbt_block_maxrecs(blocklen, true) / 2;
609 	minrecs[1] = xfs_allocbt_block_maxrecs(blocklen, false) / 2;
610 
611 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_FREESP_RECORDS);
612 }
613 
614 /* Calculate the freespace btree size for some records. */
615 xfs_extlen_t
616 xfs_allocbt_calc_size(
617 	struct xfs_mount	*mp,
618 	unsigned long long	len)
619 {
620 	return xfs_btree_calc_size(mp->m_alloc_mnr, len);
621 }
622 
623 int __init
624 xfs_allocbt_init_cur_cache(void)
625 {
626 	xfs_allocbt_cur_cache = kmem_cache_create("xfs_bnobt_cur",
627 			xfs_btree_cur_sizeof(xfs_allocbt_maxlevels_ondisk()),
628 			0, 0, NULL);
629 
630 	if (!xfs_allocbt_cur_cache)
631 		return -ENOMEM;
632 	return 0;
633 }
634 
635 void
636 xfs_allocbt_destroy_cur_cache(void)
637 {
638 	kmem_cache_destroy(xfs_allocbt_cur_cache);
639 	xfs_allocbt_cur_cache = NULL;
640 }
641