xref: /openbmc/linux/fs/xfs/libxfs/xfs_ialloc_btree.c (revision 55b6f763)
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_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_btree.h"
15 #include "xfs_btree_staging.h"
16 #include "xfs_ialloc.h"
17 #include "xfs_ialloc_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
22 #include "xfs_rmap.h"
23 
24 STATIC int
25 xfs_inobt_get_minrecs(
26 	struct xfs_btree_cur	*cur,
27 	int			level)
28 {
29 	return M_IGEO(cur->bc_mp)->inobt_mnr[level != 0];
30 }
31 
32 STATIC struct xfs_btree_cur *
33 xfs_inobt_dup_cursor(
34 	struct xfs_btree_cur	*cur)
35 {
36 	return xfs_inobt_init_cursor(cur->bc_mp, cur->bc_tp,
37 			cur->bc_ag.agbp, cur->bc_ag.agno,
38 			cur->bc_btnum);
39 }
40 
41 STATIC void
42 xfs_inobt_set_root(
43 	struct xfs_btree_cur	*cur,
44 	union xfs_btree_ptr	*nptr,
45 	int			inc)	/* level change */
46 {
47 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
48 	struct xfs_agi		*agi = agbp->b_addr;
49 
50 	agi->agi_root = nptr->s;
51 	be32_add_cpu(&agi->agi_level, inc);
52 	xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_ROOT | XFS_AGI_LEVEL);
53 }
54 
55 STATIC void
56 xfs_finobt_set_root(
57 	struct xfs_btree_cur	*cur,
58 	union xfs_btree_ptr	*nptr,
59 	int			inc)	/* level change */
60 {
61 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
62 	struct xfs_agi		*agi = agbp->b_addr;
63 
64 	agi->agi_free_root = nptr->s;
65 	be32_add_cpu(&agi->agi_free_level, inc);
66 	xfs_ialloc_log_agi(cur->bc_tp, agbp,
67 			   XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL);
68 }
69 
70 /* Update the inode btree block counter for this btree. */
71 static inline void
72 xfs_inobt_mod_blockcount(
73 	struct xfs_btree_cur	*cur,
74 	int			howmuch)
75 {
76 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
77 	struct xfs_agi		*agi = agbp->b_addr;
78 
79 	if (!xfs_sb_version_hasinobtcounts(&cur->bc_mp->m_sb))
80 		return;
81 
82 	if (cur->bc_btnum == XFS_BTNUM_FINO)
83 		be32_add_cpu(&agi->agi_fblocks, howmuch);
84 	else if (cur->bc_btnum == XFS_BTNUM_INO)
85 		be32_add_cpu(&agi->agi_iblocks, howmuch);
86 	xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_IBLOCKS);
87 }
88 
89 STATIC int
90 __xfs_inobt_alloc_block(
91 	struct xfs_btree_cur	*cur,
92 	union xfs_btree_ptr	*start,
93 	union xfs_btree_ptr	*new,
94 	int			*stat,
95 	enum xfs_ag_resv_type	resv)
96 {
97 	xfs_alloc_arg_t		args;		/* block allocation args */
98 	int			error;		/* error return value */
99 	xfs_agblock_t		sbno = be32_to_cpu(start->s);
100 
101 	memset(&args, 0, sizeof(args));
102 	args.tp = cur->bc_tp;
103 	args.mp = cur->bc_mp;
104 	args.oinfo = XFS_RMAP_OINFO_INOBT;
105 	args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_ag.agno, sbno);
106 	args.minlen = 1;
107 	args.maxlen = 1;
108 	args.prod = 1;
109 	args.type = XFS_ALLOCTYPE_NEAR_BNO;
110 	args.resv = resv;
111 
112 	error = xfs_alloc_vextent(&args);
113 	if (error)
114 		return error;
115 
116 	if (args.fsbno == NULLFSBLOCK) {
117 		*stat = 0;
118 		return 0;
119 	}
120 	ASSERT(args.len == 1);
121 
122 	new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
123 	*stat = 1;
124 	xfs_inobt_mod_blockcount(cur, 1);
125 	return 0;
126 }
127 
128 STATIC int
129 xfs_inobt_alloc_block(
130 	struct xfs_btree_cur	*cur,
131 	union xfs_btree_ptr	*start,
132 	union xfs_btree_ptr	*new,
133 	int			*stat)
134 {
135 	return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
136 }
137 
138 STATIC int
139 xfs_finobt_alloc_block(
140 	struct xfs_btree_cur	*cur,
141 	union xfs_btree_ptr	*start,
142 	union xfs_btree_ptr	*new,
143 	int			*stat)
144 {
145 	if (cur->bc_mp->m_finobt_nores)
146 		return xfs_inobt_alloc_block(cur, start, new, stat);
147 	return __xfs_inobt_alloc_block(cur, start, new, stat,
148 			XFS_AG_RESV_METADATA);
149 }
150 
151 STATIC int
152 __xfs_inobt_free_block(
153 	struct xfs_btree_cur	*cur,
154 	struct xfs_buf		*bp,
155 	enum xfs_ag_resv_type	resv)
156 {
157 	xfs_inobt_mod_blockcount(cur, -1);
158 	return xfs_free_extent(cur->bc_tp,
159 			XFS_DADDR_TO_FSB(cur->bc_mp, XFS_BUF_ADDR(bp)), 1,
160 			&XFS_RMAP_OINFO_INOBT, resv);
161 }
162 
163 STATIC int
164 xfs_inobt_free_block(
165 	struct xfs_btree_cur	*cur,
166 	struct xfs_buf		*bp)
167 {
168 	return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
169 }
170 
171 STATIC int
172 xfs_finobt_free_block(
173 	struct xfs_btree_cur	*cur,
174 	struct xfs_buf		*bp)
175 {
176 	if (cur->bc_mp->m_finobt_nores)
177 		return xfs_inobt_free_block(cur, bp);
178 	return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_METADATA);
179 }
180 
181 STATIC int
182 xfs_inobt_get_maxrecs(
183 	struct xfs_btree_cur	*cur,
184 	int			level)
185 {
186 	return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
187 }
188 
189 STATIC void
190 xfs_inobt_init_key_from_rec(
191 	union xfs_btree_key	*key,
192 	union xfs_btree_rec	*rec)
193 {
194 	key->inobt.ir_startino = rec->inobt.ir_startino;
195 }
196 
197 STATIC void
198 xfs_inobt_init_high_key_from_rec(
199 	union xfs_btree_key	*key,
200 	union xfs_btree_rec	*rec)
201 {
202 	__u32			x;
203 
204 	x = be32_to_cpu(rec->inobt.ir_startino);
205 	x += XFS_INODES_PER_CHUNK - 1;
206 	key->inobt.ir_startino = cpu_to_be32(x);
207 }
208 
209 STATIC void
210 xfs_inobt_init_rec_from_cur(
211 	struct xfs_btree_cur	*cur,
212 	union xfs_btree_rec	*rec)
213 {
214 	rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
215 	if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
216 		rec->inobt.ir_u.sp.ir_holemask =
217 					cpu_to_be16(cur->bc_rec.i.ir_holemask);
218 		rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
219 		rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
220 	} else {
221 		/* ir_holemask/ir_count not supported on-disk */
222 		rec->inobt.ir_u.f.ir_freecount =
223 					cpu_to_be32(cur->bc_rec.i.ir_freecount);
224 	}
225 	rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
226 }
227 
228 /*
229  * initial value of ptr for lookup
230  */
231 STATIC void
232 xfs_inobt_init_ptr_from_cur(
233 	struct xfs_btree_cur	*cur,
234 	union xfs_btree_ptr	*ptr)
235 {
236 	struct xfs_agi		*agi = cur->bc_ag.agbp->b_addr;
237 
238 	ASSERT(cur->bc_ag.agno == be32_to_cpu(agi->agi_seqno));
239 
240 	ptr->s = agi->agi_root;
241 }
242 
243 STATIC void
244 xfs_finobt_init_ptr_from_cur(
245 	struct xfs_btree_cur	*cur,
246 	union xfs_btree_ptr	*ptr)
247 {
248 	struct xfs_agi		*agi = cur->bc_ag.agbp->b_addr;
249 
250 	ASSERT(cur->bc_ag.agno == be32_to_cpu(agi->agi_seqno));
251 	ptr->s = agi->agi_free_root;
252 }
253 
254 STATIC int64_t
255 xfs_inobt_key_diff(
256 	struct xfs_btree_cur	*cur,
257 	union xfs_btree_key	*key)
258 {
259 	return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
260 			  cur->bc_rec.i.ir_startino;
261 }
262 
263 STATIC int64_t
264 xfs_inobt_diff_two_keys(
265 	struct xfs_btree_cur	*cur,
266 	union xfs_btree_key	*k1,
267 	union xfs_btree_key	*k2)
268 {
269 	return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
270 			  be32_to_cpu(k2->inobt.ir_startino);
271 }
272 
273 static xfs_failaddr_t
274 xfs_inobt_verify(
275 	struct xfs_buf		*bp)
276 {
277 	struct xfs_mount	*mp = bp->b_mount;
278 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
279 	xfs_failaddr_t		fa;
280 	unsigned int		level;
281 
282 	if (!xfs_verify_magic(bp, block->bb_magic))
283 		return __this_address;
284 
285 	/*
286 	 * During growfs operations, we can't verify the exact owner as the
287 	 * perag is not fully initialised and hence not attached to the buffer.
288 	 *
289 	 * Similarly, during log recovery we will have a perag structure
290 	 * attached, but the agi information will not yet have been initialised
291 	 * from the on disk AGI. We don't currently use any of this information,
292 	 * but beware of the landmine (i.e. need to check pag->pagi_init) if we
293 	 * ever do.
294 	 */
295 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
296 		fa = xfs_btree_sblock_v5hdr_verify(bp);
297 		if (fa)
298 			return fa;
299 	}
300 
301 	/* level verification */
302 	level = be16_to_cpu(block->bb_level);
303 	if (level >= M_IGEO(mp)->inobt_maxlevels)
304 		return __this_address;
305 
306 	return xfs_btree_sblock_verify(bp,
307 			M_IGEO(mp)->inobt_mxr[level != 0]);
308 }
309 
310 static void
311 xfs_inobt_read_verify(
312 	struct xfs_buf	*bp)
313 {
314 	xfs_failaddr_t	fa;
315 
316 	if (!xfs_btree_sblock_verify_crc(bp))
317 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
318 	else {
319 		fa = xfs_inobt_verify(bp);
320 		if (fa)
321 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
322 	}
323 
324 	if (bp->b_error)
325 		trace_xfs_btree_corrupt(bp, _RET_IP_);
326 }
327 
328 static void
329 xfs_inobt_write_verify(
330 	struct xfs_buf	*bp)
331 {
332 	xfs_failaddr_t	fa;
333 
334 	fa = xfs_inobt_verify(bp);
335 	if (fa) {
336 		trace_xfs_btree_corrupt(bp, _RET_IP_);
337 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
338 		return;
339 	}
340 	xfs_btree_sblock_calc_crc(bp);
341 
342 }
343 
344 const struct xfs_buf_ops xfs_inobt_buf_ops = {
345 	.name = "xfs_inobt",
346 	.magic = { cpu_to_be32(XFS_IBT_MAGIC), cpu_to_be32(XFS_IBT_CRC_MAGIC) },
347 	.verify_read = xfs_inobt_read_verify,
348 	.verify_write = xfs_inobt_write_verify,
349 	.verify_struct = xfs_inobt_verify,
350 };
351 
352 const struct xfs_buf_ops xfs_finobt_buf_ops = {
353 	.name = "xfs_finobt",
354 	.magic = { cpu_to_be32(XFS_FIBT_MAGIC),
355 		   cpu_to_be32(XFS_FIBT_CRC_MAGIC) },
356 	.verify_read = xfs_inobt_read_verify,
357 	.verify_write = xfs_inobt_write_verify,
358 	.verify_struct = xfs_inobt_verify,
359 };
360 
361 STATIC int
362 xfs_inobt_keys_inorder(
363 	struct xfs_btree_cur	*cur,
364 	union xfs_btree_key	*k1,
365 	union xfs_btree_key	*k2)
366 {
367 	return be32_to_cpu(k1->inobt.ir_startino) <
368 		be32_to_cpu(k2->inobt.ir_startino);
369 }
370 
371 STATIC int
372 xfs_inobt_recs_inorder(
373 	struct xfs_btree_cur	*cur,
374 	union xfs_btree_rec	*r1,
375 	union xfs_btree_rec	*r2)
376 {
377 	return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
378 		be32_to_cpu(r2->inobt.ir_startino);
379 }
380 
381 static const struct xfs_btree_ops xfs_inobt_ops = {
382 	.rec_len		= sizeof(xfs_inobt_rec_t),
383 	.key_len		= sizeof(xfs_inobt_key_t),
384 
385 	.dup_cursor		= xfs_inobt_dup_cursor,
386 	.set_root		= xfs_inobt_set_root,
387 	.alloc_block		= xfs_inobt_alloc_block,
388 	.free_block		= xfs_inobt_free_block,
389 	.get_minrecs		= xfs_inobt_get_minrecs,
390 	.get_maxrecs		= xfs_inobt_get_maxrecs,
391 	.init_key_from_rec	= xfs_inobt_init_key_from_rec,
392 	.init_high_key_from_rec	= xfs_inobt_init_high_key_from_rec,
393 	.init_rec_from_cur	= xfs_inobt_init_rec_from_cur,
394 	.init_ptr_from_cur	= xfs_inobt_init_ptr_from_cur,
395 	.key_diff		= xfs_inobt_key_diff,
396 	.buf_ops		= &xfs_inobt_buf_ops,
397 	.diff_two_keys		= xfs_inobt_diff_two_keys,
398 	.keys_inorder		= xfs_inobt_keys_inorder,
399 	.recs_inorder		= xfs_inobt_recs_inorder,
400 };
401 
402 static const struct xfs_btree_ops xfs_finobt_ops = {
403 	.rec_len		= sizeof(xfs_inobt_rec_t),
404 	.key_len		= sizeof(xfs_inobt_key_t),
405 
406 	.dup_cursor		= xfs_inobt_dup_cursor,
407 	.set_root		= xfs_finobt_set_root,
408 	.alloc_block		= xfs_finobt_alloc_block,
409 	.free_block		= xfs_finobt_free_block,
410 	.get_minrecs		= xfs_inobt_get_minrecs,
411 	.get_maxrecs		= xfs_inobt_get_maxrecs,
412 	.init_key_from_rec	= xfs_inobt_init_key_from_rec,
413 	.init_high_key_from_rec	= xfs_inobt_init_high_key_from_rec,
414 	.init_rec_from_cur	= xfs_inobt_init_rec_from_cur,
415 	.init_ptr_from_cur	= xfs_finobt_init_ptr_from_cur,
416 	.key_diff		= xfs_inobt_key_diff,
417 	.buf_ops		= &xfs_finobt_buf_ops,
418 	.diff_two_keys		= xfs_inobt_diff_two_keys,
419 	.keys_inorder		= xfs_inobt_keys_inorder,
420 	.recs_inorder		= xfs_inobt_recs_inorder,
421 };
422 
423 /*
424  * Initialize a new inode btree cursor.
425  */
426 static struct xfs_btree_cur *
427 xfs_inobt_init_common(
428 	struct xfs_mount	*mp,		/* file system mount point */
429 	struct xfs_trans	*tp,		/* transaction pointer */
430 	xfs_agnumber_t		agno,		/* allocation group number */
431 	xfs_btnum_t		btnum)		/* ialloc or free ino btree */
432 {
433 	struct xfs_btree_cur	*cur;
434 
435 	cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
436 	cur->bc_tp = tp;
437 	cur->bc_mp = mp;
438 	cur->bc_btnum = btnum;
439 	if (btnum == XFS_BTNUM_INO) {
440 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_ibt_2);
441 		cur->bc_ops = &xfs_inobt_ops;
442 	} else {
443 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_fibt_2);
444 		cur->bc_ops = &xfs_finobt_ops;
445 	}
446 
447 	cur->bc_blocklog = mp->m_sb.sb_blocklog;
448 
449 	if (xfs_sb_version_hascrc(&mp->m_sb))
450 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
451 
452 	cur->bc_ag.agno = agno;
453 	return cur;
454 }
455 
456 /* Create an inode btree cursor. */
457 struct xfs_btree_cur *
458 xfs_inobt_init_cursor(
459 	struct xfs_mount	*mp,
460 	struct xfs_trans	*tp,
461 	struct xfs_buf		*agbp,
462 	xfs_agnumber_t		agno,
463 	xfs_btnum_t		btnum)
464 {
465 	struct xfs_btree_cur	*cur;
466 	struct xfs_agi		*agi = agbp->b_addr;
467 
468 	cur = xfs_inobt_init_common(mp, tp, agno, btnum);
469 	if (btnum == XFS_BTNUM_INO)
470 		cur->bc_nlevels = be32_to_cpu(agi->agi_level);
471 	else
472 		cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
473 	cur->bc_ag.agbp = agbp;
474 	return cur;
475 }
476 
477 /* Create an inode btree cursor with a fake root for staging. */
478 struct xfs_btree_cur *
479 xfs_inobt_stage_cursor(
480 	struct xfs_mount	*mp,
481 	struct xbtree_afakeroot	*afake,
482 	xfs_agnumber_t		agno,
483 	xfs_btnum_t		btnum)
484 {
485 	struct xfs_btree_cur	*cur;
486 
487 	cur = xfs_inobt_init_common(mp, NULL, agno, btnum);
488 	xfs_btree_stage_afakeroot(cur, afake);
489 	return cur;
490 }
491 
492 /*
493  * Install a new inobt btree root.  Caller is responsible for invalidating
494  * and freeing the old btree blocks.
495  */
496 void
497 xfs_inobt_commit_staged_btree(
498 	struct xfs_btree_cur	*cur,
499 	struct xfs_trans	*tp,
500 	struct xfs_buf		*agbp)
501 {
502 	struct xfs_agi		*agi = agbp->b_addr;
503 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
504 	int			fields;
505 
506 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
507 
508 	if (cur->bc_btnum == XFS_BTNUM_INO) {
509 		fields = XFS_AGI_ROOT | XFS_AGI_LEVEL;
510 		agi->agi_root = cpu_to_be32(afake->af_root);
511 		agi->agi_level = cpu_to_be32(afake->af_levels);
512 		if (xfs_sb_version_hasinobtcounts(&cur->bc_mp->m_sb)) {
513 			agi->agi_iblocks = cpu_to_be32(afake->af_blocks);
514 			fields |= XFS_AGI_IBLOCKS;
515 		}
516 		xfs_ialloc_log_agi(tp, agbp, fields);
517 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_inobt_ops);
518 	} else {
519 		fields = XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL;
520 		agi->agi_free_root = cpu_to_be32(afake->af_root);
521 		agi->agi_free_level = cpu_to_be32(afake->af_levels);
522 		if (xfs_sb_version_hasinobtcounts(&cur->bc_mp->m_sb)) {
523 			agi->agi_fblocks = cpu_to_be32(afake->af_blocks);
524 			fields |= XFS_AGI_IBLOCKS;
525 		}
526 		xfs_ialloc_log_agi(tp, agbp, fields);
527 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_finobt_ops);
528 	}
529 }
530 
531 /*
532  * Calculate number of records in an inobt btree block.
533  */
534 int
535 xfs_inobt_maxrecs(
536 	struct xfs_mount	*mp,
537 	int			blocklen,
538 	int			leaf)
539 {
540 	blocklen -= XFS_INOBT_BLOCK_LEN(mp);
541 
542 	if (leaf)
543 		return blocklen / sizeof(xfs_inobt_rec_t);
544 	return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
545 }
546 
547 /*
548  * Convert the inode record holemask to an inode allocation bitmap. The inode
549  * allocation bitmap is inode granularity and specifies whether an inode is
550  * physically allocated on disk (not whether the inode is considered allocated
551  * or free by the fs).
552  *
553  * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
554  */
555 uint64_t
556 xfs_inobt_irec_to_allocmask(
557 	struct xfs_inobt_rec_incore	*rec)
558 {
559 	uint64_t			bitmap = 0;
560 	uint64_t			inodespbit;
561 	int				nextbit;
562 	uint				allocbitmap;
563 
564 	/*
565 	 * The holemask has 16-bits for a 64 inode record. Therefore each
566 	 * holemask bit represents multiple inodes. Create a mask of bits to set
567 	 * in the allocmask for each holemask bit.
568 	 */
569 	inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
570 
571 	/*
572 	 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
573 	 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
574 	 * anything beyond the 16 holemask bits since this casts to a larger
575 	 * type.
576 	 */
577 	allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
578 
579 	/*
580 	 * allocbitmap is the inverted holemask so every set bit represents
581 	 * allocated inodes. To expand from 16-bit holemask granularity to
582 	 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
583 	 * bitmap for every holemask bit.
584 	 */
585 	nextbit = xfs_next_bit(&allocbitmap, 1, 0);
586 	while (nextbit != -1) {
587 		ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
588 
589 		bitmap |= (inodespbit <<
590 			   (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
591 
592 		nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
593 	}
594 
595 	return bitmap;
596 }
597 
598 #if defined(DEBUG) || defined(XFS_WARN)
599 /*
600  * Verify that an in-core inode record has a valid inode count.
601  */
602 int
603 xfs_inobt_rec_check_count(
604 	struct xfs_mount		*mp,
605 	struct xfs_inobt_rec_incore	*rec)
606 {
607 	int				inocount = 0;
608 	int				nextbit = 0;
609 	uint64_t			allocbmap;
610 	int				wordsz;
611 
612 	wordsz = sizeof(allocbmap) / sizeof(unsigned int);
613 	allocbmap = xfs_inobt_irec_to_allocmask(rec);
614 
615 	nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
616 	while (nextbit != -1) {
617 		inocount++;
618 		nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
619 				       nextbit + 1);
620 	}
621 
622 	if (inocount != rec->ir_count)
623 		return -EFSCORRUPTED;
624 
625 	return 0;
626 }
627 #endif	/* DEBUG */
628 
629 static xfs_extlen_t
630 xfs_inobt_max_size(
631 	struct xfs_mount	*mp,
632 	xfs_agnumber_t		agno)
633 {
634 	xfs_agblock_t		agblocks = xfs_ag_block_count(mp, agno);
635 
636 	/* Bail out if we're uninitialized, which can happen in mkfs. */
637 	if (M_IGEO(mp)->inobt_mxr[0] == 0)
638 		return 0;
639 
640 	/*
641 	 * The log is permanently allocated, so the space it occupies will
642 	 * never be available for the kinds of things that would require btree
643 	 * expansion.  We therefore can pretend the space isn't there.
644 	 */
645 	if (mp->m_sb.sb_logstart &&
646 	    XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart) == agno)
647 		agblocks -= mp->m_sb.sb_logblocks;
648 
649 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr,
650 				(uint64_t)agblocks * mp->m_sb.sb_inopblock /
651 					XFS_INODES_PER_CHUNK);
652 }
653 
654 /* Read AGI and create inobt cursor. */
655 int
656 xfs_inobt_cur(
657 	struct xfs_mount	*mp,
658 	struct xfs_trans	*tp,
659 	xfs_agnumber_t		agno,
660 	xfs_btnum_t		which,
661 	struct xfs_btree_cur	**curpp,
662 	struct xfs_buf		**agi_bpp)
663 {
664 	struct xfs_btree_cur	*cur;
665 	int			error;
666 
667 	ASSERT(*agi_bpp == NULL);
668 	ASSERT(*curpp == NULL);
669 
670 	error = xfs_ialloc_read_agi(mp, tp, agno, agi_bpp);
671 	if (error)
672 		return error;
673 
674 	cur = xfs_inobt_init_cursor(mp, tp, *agi_bpp, agno, which);
675 	*curpp = cur;
676 	return 0;
677 }
678 
679 static int
680 xfs_inobt_count_blocks(
681 	struct xfs_mount	*mp,
682 	struct xfs_trans	*tp,
683 	xfs_agnumber_t		agno,
684 	xfs_btnum_t		btnum,
685 	xfs_extlen_t		*tree_blocks)
686 {
687 	struct xfs_buf		*agbp = NULL;
688 	struct xfs_btree_cur	*cur = NULL;
689 	int			error;
690 
691 	error = xfs_inobt_cur(mp, tp, agno, btnum, &cur, &agbp);
692 	if (error)
693 		return error;
694 
695 	error = xfs_btree_count_blocks(cur, tree_blocks);
696 	xfs_btree_del_cursor(cur, error);
697 	xfs_trans_brelse(tp, agbp);
698 
699 	return error;
700 }
701 
702 /* Read finobt block count from AGI header. */
703 static int
704 xfs_finobt_read_blocks(
705 	struct xfs_mount	*mp,
706 	struct xfs_trans	*tp,
707 	xfs_agnumber_t		agno,
708 	xfs_extlen_t		*tree_blocks)
709 {
710 	struct xfs_buf		*agbp;
711 	struct xfs_agi		*agi;
712 	int			error;
713 
714 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
715 	if (error)
716 		return error;
717 
718 	agi = agbp->b_addr;
719 	*tree_blocks = be32_to_cpu(agi->agi_fblocks);
720 	xfs_trans_brelse(tp, agbp);
721 	return 0;
722 }
723 
724 /*
725  * Figure out how many blocks to reserve and how many are used by this btree.
726  */
727 int
728 xfs_finobt_calc_reserves(
729 	struct xfs_mount	*mp,
730 	struct xfs_trans	*tp,
731 	xfs_agnumber_t		agno,
732 	xfs_extlen_t		*ask,
733 	xfs_extlen_t		*used)
734 {
735 	xfs_extlen_t		tree_len = 0;
736 	int			error;
737 
738 	if (!xfs_sb_version_hasfinobt(&mp->m_sb))
739 		return 0;
740 
741 	if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
742 		error = xfs_finobt_read_blocks(mp, tp, agno, &tree_len);
743 	else
744 		error = xfs_inobt_count_blocks(mp, tp, agno, XFS_BTNUM_FINO,
745 				&tree_len);
746 	if (error)
747 		return error;
748 
749 	*ask += xfs_inobt_max_size(mp, agno);
750 	*used += tree_len;
751 	return 0;
752 }
753 
754 /* Calculate the inobt btree size for some records. */
755 xfs_extlen_t
756 xfs_iallocbt_calc_size(
757 	struct xfs_mount	*mp,
758 	unsigned long long	len)
759 {
760 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);
761 }
762