xref: /openbmc/linux/fs/xfs/libxfs/xfs_ialloc_btree.c (revision c4c3c32d)
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 #include "xfs_ag.h"
24 
25 static struct kmem_cache	*xfs_inobt_cur_cache;
26 
27 STATIC int
28 xfs_inobt_get_minrecs(
29 	struct xfs_btree_cur	*cur,
30 	int			level)
31 {
32 	return M_IGEO(cur->bc_mp)->inobt_mnr[level != 0];
33 }
34 
35 STATIC struct xfs_btree_cur *
36 xfs_inobt_dup_cursor(
37 	struct xfs_btree_cur	*cur)
38 {
39 	return xfs_inobt_init_cursor(cur->bc_ag.pag, cur->bc_tp,
40 			cur->bc_ag.agbp, cur->bc_btnum);
41 }
42 
43 STATIC void
44 xfs_inobt_set_root(
45 	struct xfs_btree_cur		*cur,
46 	const union xfs_btree_ptr	*nptr,
47 	int				inc)	/* level change */
48 {
49 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
50 	struct xfs_agi		*agi = agbp->b_addr;
51 
52 	agi->agi_root = nptr->s;
53 	be32_add_cpu(&agi->agi_level, inc);
54 	xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_ROOT | XFS_AGI_LEVEL);
55 }
56 
57 STATIC void
58 xfs_finobt_set_root(
59 	struct xfs_btree_cur		*cur,
60 	const union xfs_btree_ptr	*nptr,
61 	int				inc)	/* level change */
62 {
63 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
64 	struct xfs_agi		*agi = agbp->b_addr;
65 
66 	agi->agi_free_root = nptr->s;
67 	be32_add_cpu(&agi->agi_free_level, inc);
68 	xfs_ialloc_log_agi(cur->bc_tp, agbp,
69 			   XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL);
70 }
71 
72 /* Update the inode btree block counter for this btree. */
73 static inline void
74 xfs_inobt_mod_blockcount(
75 	struct xfs_btree_cur	*cur,
76 	int			howmuch)
77 {
78 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
79 	struct xfs_agi		*agi = agbp->b_addr;
80 
81 	if (!xfs_has_inobtcounts(cur->bc_mp))
82 		return;
83 
84 	if (cur->bc_btnum == XFS_BTNUM_FINO)
85 		be32_add_cpu(&agi->agi_fblocks, howmuch);
86 	else if (cur->bc_btnum == XFS_BTNUM_INO)
87 		be32_add_cpu(&agi->agi_iblocks, howmuch);
88 	xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_IBLOCKS);
89 }
90 
91 STATIC int
92 __xfs_inobt_alloc_block(
93 	struct xfs_btree_cur		*cur,
94 	const union xfs_btree_ptr	*start,
95 	union xfs_btree_ptr		*new,
96 	int				*stat,
97 	enum xfs_ag_resv_type		resv)
98 {
99 	xfs_alloc_arg_t		args;		/* block allocation args */
100 	int			error;		/* error return value */
101 	xfs_agblock_t		sbno = be32_to_cpu(start->s);
102 
103 	memset(&args, 0, sizeof(args));
104 	args.tp = cur->bc_tp;
105 	args.mp = cur->bc_mp;
106 	args.pag = cur->bc_ag.pag;
107 	args.oinfo = XFS_RMAP_OINFO_INOBT;
108 	args.minlen = 1;
109 	args.maxlen = 1;
110 	args.prod = 1;
111 	args.resv = resv;
112 
113 	error = xfs_alloc_vextent_near_bno(&args,
114 			XFS_AGB_TO_FSB(args.mp, args.pag->pag_agno, sbno));
115 	if (error)
116 		return error;
117 
118 	if (args.fsbno == NULLFSBLOCK) {
119 		*stat = 0;
120 		return 0;
121 	}
122 	ASSERT(args.len == 1);
123 
124 	new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
125 	*stat = 1;
126 	xfs_inobt_mod_blockcount(cur, 1);
127 	return 0;
128 }
129 
130 STATIC int
131 xfs_inobt_alloc_block(
132 	struct xfs_btree_cur		*cur,
133 	const union xfs_btree_ptr	*start,
134 	union xfs_btree_ptr		*new,
135 	int				*stat)
136 {
137 	return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
138 }
139 
140 STATIC int
141 xfs_finobt_alloc_block(
142 	struct xfs_btree_cur		*cur,
143 	const union xfs_btree_ptr	*start,
144 	union xfs_btree_ptr		*new,
145 	int				*stat)
146 {
147 	if (cur->bc_mp->m_finobt_nores)
148 		return xfs_inobt_alloc_block(cur, start, new, stat);
149 	return __xfs_inobt_alloc_block(cur, start, new, stat,
150 			XFS_AG_RESV_METADATA);
151 }
152 
153 STATIC int
154 __xfs_inobt_free_block(
155 	struct xfs_btree_cur	*cur,
156 	struct xfs_buf		*bp,
157 	enum xfs_ag_resv_type	resv)
158 {
159 	xfs_fsblock_t		fsbno;
160 
161 	xfs_inobt_mod_blockcount(cur, -1);
162 	fsbno = XFS_DADDR_TO_FSB(cur->bc_mp, xfs_buf_daddr(bp));
163 	return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
164 			&XFS_RMAP_OINFO_INOBT, resv);
165 }
166 
167 STATIC int
168 xfs_inobt_free_block(
169 	struct xfs_btree_cur	*cur,
170 	struct xfs_buf		*bp)
171 {
172 	return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
173 }
174 
175 STATIC int
176 xfs_finobt_free_block(
177 	struct xfs_btree_cur	*cur,
178 	struct xfs_buf		*bp)
179 {
180 	if (cur->bc_mp->m_finobt_nores)
181 		return xfs_inobt_free_block(cur, bp);
182 	return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_METADATA);
183 }
184 
185 STATIC int
186 xfs_inobt_get_maxrecs(
187 	struct xfs_btree_cur	*cur,
188 	int			level)
189 {
190 	return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
191 }
192 
193 STATIC void
194 xfs_inobt_init_key_from_rec(
195 	union xfs_btree_key		*key,
196 	const union xfs_btree_rec	*rec)
197 {
198 	key->inobt.ir_startino = rec->inobt.ir_startino;
199 }
200 
201 STATIC void
202 xfs_inobt_init_high_key_from_rec(
203 	union xfs_btree_key		*key,
204 	const union xfs_btree_rec	*rec)
205 {
206 	__u32				x;
207 
208 	x = be32_to_cpu(rec->inobt.ir_startino);
209 	x += XFS_INODES_PER_CHUNK - 1;
210 	key->inobt.ir_startino = cpu_to_be32(x);
211 }
212 
213 STATIC void
214 xfs_inobt_init_rec_from_cur(
215 	struct xfs_btree_cur	*cur,
216 	union xfs_btree_rec	*rec)
217 {
218 	rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
219 	if (xfs_has_sparseinodes(cur->bc_mp)) {
220 		rec->inobt.ir_u.sp.ir_holemask =
221 					cpu_to_be16(cur->bc_rec.i.ir_holemask);
222 		rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
223 		rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
224 	} else {
225 		/* ir_holemask/ir_count not supported on-disk */
226 		rec->inobt.ir_u.f.ir_freecount =
227 					cpu_to_be32(cur->bc_rec.i.ir_freecount);
228 	}
229 	rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
230 }
231 
232 /*
233  * initial value of ptr for lookup
234  */
235 STATIC void
236 xfs_inobt_init_ptr_from_cur(
237 	struct xfs_btree_cur	*cur,
238 	union xfs_btree_ptr	*ptr)
239 {
240 	struct xfs_agi		*agi = cur->bc_ag.agbp->b_addr;
241 
242 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
243 
244 	ptr->s = agi->agi_root;
245 }
246 
247 STATIC void
248 xfs_finobt_init_ptr_from_cur(
249 	struct xfs_btree_cur	*cur,
250 	union xfs_btree_ptr	*ptr)
251 {
252 	struct xfs_agi		*agi = cur->bc_ag.agbp->b_addr;
253 
254 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
255 	ptr->s = agi->agi_free_root;
256 }
257 
258 STATIC int64_t
259 xfs_inobt_key_diff(
260 	struct xfs_btree_cur		*cur,
261 	const union xfs_btree_key	*key)
262 {
263 	return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
264 			  cur->bc_rec.i.ir_startino;
265 }
266 
267 STATIC int64_t
268 xfs_inobt_diff_two_keys(
269 	struct xfs_btree_cur		*cur,
270 	const union xfs_btree_key	*k1,
271 	const union xfs_btree_key	*k2,
272 	const union xfs_btree_key	*mask)
273 {
274 	ASSERT(!mask || mask->inobt.ir_startino);
275 
276 	return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
277 			be32_to_cpu(k2->inobt.ir_startino);
278 }
279 
280 static xfs_failaddr_t
281 xfs_inobt_verify(
282 	struct xfs_buf		*bp)
283 {
284 	struct xfs_mount	*mp = bp->b_mount;
285 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
286 	xfs_failaddr_t		fa;
287 	unsigned int		level;
288 
289 	if (!xfs_verify_magic(bp, block->bb_magic))
290 		return __this_address;
291 
292 	/*
293 	 * During growfs operations, we can't verify the exact owner as the
294 	 * perag is not fully initialised and hence not attached to the buffer.
295 	 *
296 	 * Similarly, during log recovery we will have a perag structure
297 	 * attached, but the agi information will not yet have been initialised
298 	 * from the on disk AGI. We don't currently use any of this information,
299 	 * but beware of the landmine (i.e. need to check
300 	 * xfs_perag_initialised_agi(pag)) if we ever do.
301 	 */
302 	if (xfs_has_crc(mp)) {
303 		fa = xfs_btree_sblock_v5hdr_verify(bp);
304 		if (fa)
305 			return fa;
306 	}
307 
308 	/* level verification */
309 	level = be16_to_cpu(block->bb_level);
310 	if (level >= M_IGEO(mp)->inobt_maxlevels)
311 		return __this_address;
312 
313 	return xfs_btree_sblock_verify(bp,
314 			M_IGEO(mp)->inobt_mxr[level != 0]);
315 }
316 
317 static void
318 xfs_inobt_read_verify(
319 	struct xfs_buf	*bp)
320 {
321 	xfs_failaddr_t	fa;
322 
323 	if (!xfs_btree_sblock_verify_crc(bp))
324 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
325 	else {
326 		fa = xfs_inobt_verify(bp);
327 		if (fa)
328 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
329 	}
330 
331 	if (bp->b_error)
332 		trace_xfs_btree_corrupt(bp, _RET_IP_);
333 }
334 
335 static void
336 xfs_inobt_write_verify(
337 	struct xfs_buf	*bp)
338 {
339 	xfs_failaddr_t	fa;
340 
341 	fa = xfs_inobt_verify(bp);
342 	if (fa) {
343 		trace_xfs_btree_corrupt(bp, _RET_IP_);
344 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
345 		return;
346 	}
347 	xfs_btree_sblock_calc_crc(bp);
348 
349 }
350 
351 const struct xfs_buf_ops xfs_inobt_buf_ops = {
352 	.name = "xfs_inobt",
353 	.magic = { cpu_to_be32(XFS_IBT_MAGIC), cpu_to_be32(XFS_IBT_CRC_MAGIC) },
354 	.verify_read = xfs_inobt_read_verify,
355 	.verify_write = xfs_inobt_write_verify,
356 	.verify_struct = xfs_inobt_verify,
357 };
358 
359 const struct xfs_buf_ops xfs_finobt_buf_ops = {
360 	.name = "xfs_finobt",
361 	.magic = { cpu_to_be32(XFS_FIBT_MAGIC),
362 		   cpu_to_be32(XFS_FIBT_CRC_MAGIC) },
363 	.verify_read = xfs_inobt_read_verify,
364 	.verify_write = xfs_inobt_write_verify,
365 	.verify_struct = xfs_inobt_verify,
366 };
367 
368 STATIC int
369 xfs_inobt_keys_inorder(
370 	struct xfs_btree_cur		*cur,
371 	const union xfs_btree_key	*k1,
372 	const union xfs_btree_key	*k2)
373 {
374 	return be32_to_cpu(k1->inobt.ir_startino) <
375 		be32_to_cpu(k2->inobt.ir_startino);
376 }
377 
378 STATIC int
379 xfs_inobt_recs_inorder(
380 	struct xfs_btree_cur		*cur,
381 	const union xfs_btree_rec	*r1,
382 	const union xfs_btree_rec	*r2)
383 {
384 	return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
385 		be32_to_cpu(r2->inobt.ir_startino);
386 }
387 
388 STATIC enum xbtree_key_contig
389 xfs_inobt_keys_contiguous(
390 	struct xfs_btree_cur		*cur,
391 	const union xfs_btree_key	*key1,
392 	const union xfs_btree_key	*key2,
393 	const union xfs_btree_key	*mask)
394 {
395 	ASSERT(!mask || mask->inobt.ir_startino);
396 
397 	return xbtree_key_contig(be32_to_cpu(key1->inobt.ir_startino),
398 				 be32_to_cpu(key2->inobt.ir_startino));
399 }
400 
401 static const struct xfs_btree_ops xfs_inobt_ops = {
402 	.rec_len		= sizeof(xfs_inobt_rec_t),
403 	.key_len		= sizeof(xfs_inobt_key_t),
404 
405 	.dup_cursor		= xfs_inobt_dup_cursor,
406 	.set_root		= xfs_inobt_set_root,
407 	.alloc_block		= xfs_inobt_alloc_block,
408 	.free_block		= xfs_inobt_free_block,
409 	.get_minrecs		= xfs_inobt_get_minrecs,
410 	.get_maxrecs		= xfs_inobt_get_maxrecs,
411 	.init_key_from_rec	= xfs_inobt_init_key_from_rec,
412 	.init_high_key_from_rec	= xfs_inobt_init_high_key_from_rec,
413 	.init_rec_from_cur	= xfs_inobt_init_rec_from_cur,
414 	.init_ptr_from_cur	= xfs_inobt_init_ptr_from_cur,
415 	.key_diff		= xfs_inobt_key_diff,
416 	.buf_ops		= &xfs_inobt_buf_ops,
417 	.diff_two_keys		= xfs_inobt_diff_two_keys,
418 	.keys_inorder		= xfs_inobt_keys_inorder,
419 	.recs_inorder		= xfs_inobt_recs_inorder,
420 	.keys_contiguous	= xfs_inobt_keys_contiguous,
421 };
422 
423 static const struct xfs_btree_ops xfs_finobt_ops = {
424 	.rec_len		= sizeof(xfs_inobt_rec_t),
425 	.key_len		= sizeof(xfs_inobt_key_t),
426 
427 	.dup_cursor		= xfs_inobt_dup_cursor,
428 	.set_root		= xfs_finobt_set_root,
429 	.alloc_block		= xfs_finobt_alloc_block,
430 	.free_block		= xfs_finobt_free_block,
431 	.get_minrecs		= xfs_inobt_get_minrecs,
432 	.get_maxrecs		= xfs_inobt_get_maxrecs,
433 	.init_key_from_rec	= xfs_inobt_init_key_from_rec,
434 	.init_high_key_from_rec	= xfs_inobt_init_high_key_from_rec,
435 	.init_rec_from_cur	= xfs_inobt_init_rec_from_cur,
436 	.init_ptr_from_cur	= xfs_finobt_init_ptr_from_cur,
437 	.key_diff		= xfs_inobt_key_diff,
438 	.buf_ops		= &xfs_finobt_buf_ops,
439 	.diff_two_keys		= xfs_inobt_diff_two_keys,
440 	.keys_inorder		= xfs_inobt_keys_inorder,
441 	.recs_inorder		= xfs_inobt_recs_inorder,
442 	.keys_contiguous	= xfs_inobt_keys_contiguous,
443 };
444 
445 /*
446  * Initialize a new inode btree cursor.
447  */
448 static struct xfs_btree_cur *
449 xfs_inobt_init_common(
450 	struct xfs_perag	*pag,
451 	struct xfs_trans	*tp,		/* transaction pointer */
452 	xfs_btnum_t		btnum)		/* ialloc or free ino btree */
453 {
454 	struct xfs_mount	*mp = pag->pag_mount;
455 	struct xfs_btree_cur	*cur;
456 
457 	cur = xfs_btree_alloc_cursor(mp, tp, btnum,
458 			M_IGEO(mp)->inobt_maxlevels, xfs_inobt_cur_cache);
459 	if (btnum == XFS_BTNUM_INO) {
460 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_ibt_2);
461 		cur->bc_ops = &xfs_inobt_ops;
462 	} else {
463 		cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_fibt_2);
464 		cur->bc_ops = &xfs_finobt_ops;
465 	}
466 
467 	if (xfs_has_crc(mp))
468 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
469 
470 	cur->bc_ag.pag = xfs_perag_hold(pag);
471 	return cur;
472 }
473 
474 /* Create an inode btree cursor. */
475 struct xfs_btree_cur *
476 xfs_inobt_init_cursor(
477 	struct xfs_perag	*pag,
478 	struct xfs_trans	*tp,
479 	struct xfs_buf		*agbp,
480 	xfs_btnum_t		btnum)
481 {
482 	struct xfs_btree_cur	*cur;
483 	struct xfs_agi		*agi = agbp->b_addr;
484 
485 	cur = xfs_inobt_init_common(pag, tp, btnum);
486 	if (btnum == XFS_BTNUM_INO)
487 		cur->bc_nlevels = be32_to_cpu(agi->agi_level);
488 	else
489 		cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
490 	cur->bc_ag.agbp = agbp;
491 	return cur;
492 }
493 
494 /* Create an inode btree cursor with a fake root for staging. */
495 struct xfs_btree_cur *
496 xfs_inobt_stage_cursor(
497 	struct xfs_perag	*pag,
498 	struct xbtree_afakeroot	*afake,
499 	xfs_btnum_t		btnum)
500 {
501 	struct xfs_btree_cur	*cur;
502 
503 	cur = xfs_inobt_init_common(pag, NULL, btnum);
504 	xfs_btree_stage_afakeroot(cur, afake);
505 	return cur;
506 }
507 
508 /*
509  * Install a new inobt btree root.  Caller is responsible for invalidating
510  * and freeing the old btree blocks.
511  */
512 void
513 xfs_inobt_commit_staged_btree(
514 	struct xfs_btree_cur	*cur,
515 	struct xfs_trans	*tp,
516 	struct xfs_buf		*agbp)
517 {
518 	struct xfs_agi		*agi = agbp->b_addr;
519 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
520 	int			fields;
521 
522 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
523 
524 	if (cur->bc_btnum == XFS_BTNUM_INO) {
525 		fields = XFS_AGI_ROOT | XFS_AGI_LEVEL;
526 		agi->agi_root = cpu_to_be32(afake->af_root);
527 		agi->agi_level = cpu_to_be32(afake->af_levels);
528 		if (xfs_has_inobtcounts(cur->bc_mp)) {
529 			agi->agi_iblocks = cpu_to_be32(afake->af_blocks);
530 			fields |= XFS_AGI_IBLOCKS;
531 		}
532 		xfs_ialloc_log_agi(tp, agbp, fields);
533 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_inobt_ops);
534 	} else {
535 		fields = XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL;
536 		agi->agi_free_root = cpu_to_be32(afake->af_root);
537 		agi->agi_free_level = cpu_to_be32(afake->af_levels);
538 		if (xfs_has_inobtcounts(cur->bc_mp)) {
539 			agi->agi_fblocks = cpu_to_be32(afake->af_blocks);
540 			fields |= XFS_AGI_IBLOCKS;
541 		}
542 		xfs_ialloc_log_agi(tp, agbp, fields);
543 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_finobt_ops);
544 	}
545 }
546 
547 /* Calculate number of records in an inode btree block. */
548 static inline unsigned int
549 xfs_inobt_block_maxrecs(
550 	unsigned int		blocklen,
551 	bool			leaf)
552 {
553 	if (leaf)
554 		return blocklen / sizeof(xfs_inobt_rec_t);
555 	return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
556 }
557 
558 /*
559  * Calculate number of records in an inobt btree block.
560  */
561 int
562 xfs_inobt_maxrecs(
563 	struct xfs_mount	*mp,
564 	int			blocklen,
565 	int			leaf)
566 {
567 	blocklen -= XFS_INOBT_BLOCK_LEN(mp);
568 	return xfs_inobt_block_maxrecs(blocklen, leaf);
569 }
570 
571 /*
572  * Maximum number of inode btree records per AG.  Pretend that we can fill an
573  * entire AG completely full of inodes except for the AG headers.
574  */
575 #define XFS_MAX_INODE_RECORDS \
576 	((XFS_MAX_AG_BYTES - (4 * BBSIZE)) / XFS_DINODE_MIN_SIZE) / \
577 			XFS_INODES_PER_CHUNK
578 
579 /* Compute the max possible height for the inode btree. */
580 static inline unsigned int
581 xfs_inobt_maxlevels_ondisk(void)
582 {
583 	unsigned int		minrecs[2];
584 	unsigned int		blocklen;
585 
586 	blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
587 		       XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
588 
589 	minrecs[0] = xfs_inobt_block_maxrecs(blocklen, true) / 2;
590 	minrecs[1] = xfs_inobt_block_maxrecs(blocklen, false) / 2;
591 
592 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_INODE_RECORDS);
593 }
594 
595 /* Compute the max possible height for the free inode btree. */
596 static inline unsigned int
597 xfs_finobt_maxlevels_ondisk(void)
598 {
599 	unsigned int		minrecs[2];
600 	unsigned int		blocklen;
601 
602 	blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
603 
604 	minrecs[0] = xfs_inobt_block_maxrecs(blocklen, true) / 2;
605 	minrecs[1] = xfs_inobt_block_maxrecs(blocklen, false) / 2;
606 
607 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_INODE_RECORDS);
608 }
609 
610 /* Compute the max possible height for either inode btree. */
611 unsigned int
612 xfs_iallocbt_maxlevels_ondisk(void)
613 {
614 	return max(xfs_inobt_maxlevels_ondisk(),
615 		   xfs_finobt_maxlevels_ondisk());
616 }
617 
618 /*
619  * Convert the inode record holemask to an inode allocation bitmap. The inode
620  * allocation bitmap is inode granularity and specifies whether an inode is
621  * physically allocated on disk (not whether the inode is considered allocated
622  * or free by the fs).
623  *
624  * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
625  */
626 uint64_t
627 xfs_inobt_irec_to_allocmask(
628 	const struct xfs_inobt_rec_incore	*rec)
629 {
630 	uint64_t			bitmap = 0;
631 	uint64_t			inodespbit;
632 	int				nextbit;
633 	uint				allocbitmap;
634 
635 	/*
636 	 * The holemask has 16-bits for a 64 inode record. Therefore each
637 	 * holemask bit represents multiple inodes. Create a mask of bits to set
638 	 * in the allocmask for each holemask bit.
639 	 */
640 	inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
641 
642 	/*
643 	 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
644 	 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
645 	 * anything beyond the 16 holemask bits since this casts to a larger
646 	 * type.
647 	 */
648 	allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
649 
650 	/*
651 	 * allocbitmap is the inverted holemask so every set bit represents
652 	 * allocated inodes. To expand from 16-bit holemask granularity to
653 	 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
654 	 * bitmap for every holemask bit.
655 	 */
656 	nextbit = xfs_next_bit(&allocbitmap, 1, 0);
657 	while (nextbit != -1) {
658 		ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
659 
660 		bitmap |= (inodespbit <<
661 			   (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
662 
663 		nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
664 	}
665 
666 	return bitmap;
667 }
668 
669 #if defined(DEBUG) || defined(XFS_WARN)
670 /*
671  * Verify that an in-core inode record has a valid inode count.
672  */
673 int
674 xfs_inobt_rec_check_count(
675 	struct xfs_mount		*mp,
676 	struct xfs_inobt_rec_incore	*rec)
677 {
678 	int				inocount = 0;
679 	int				nextbit = 0;
680 	uint64_t			allocbmap;
681 	int				wordsz;
682 
683 	wordsz = sizeof(allocbmap) / sizeof(unsigned int);
684 	allocbmap = xfs_inobt_irec_to_allocmask(rec);
685 
686 	nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
687 	while (nextbit != -1) {
688 		inocount++;
689 		nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
690 				       nextbit + 1);
691 	}
692 
693 	if (inocount != rec->ir_count)
694 		return -EFSCORRUPTED;
695 
696 	return 0;
697 }
698 #endif	/* DEBUG */
699 
700 static xfs_extlen_t
701 xfs_inobt_max_size(
702 	struct xfs_perag	*pag)
703 {
704 	struct xfs_mount	*mp = pag->pag_mount;
705 	xfs_agblock_t		agblocks = pag->block_count;
706 
707 	/* Bail out if we're uninitialized, which can happen in mkfs. */
708 	if (M_IGEO(mp)->inobt_mxr[0] == 0)
709 		return 0;
710 
711 	/*
712 	 * The log is permanently allocated, so the space it occupies will
713 	 * never be available for the kinds of things that would require btree
714 	 * expansion.  We therefore can pretend the space isn't there.
715 	 */
716 	if (xfs_ag_contains_log(mp, pag->pag_agno))
717 		agblocks -= mp->m_sb.sb_logblocks;
718 
719 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr,
720 				(uint64_t)agblocks * mp->m_sb.sb_inopblock /
721 					XFS_INODES_PER_CHUNK);
722 }
723 
724 /* Read AGI and create inobt cursor. */
725 int
726 xfs_inobt_cur(
727 	struct xfs_perag	*pag,
728 	struct xfs_trans	*tp,
729 	xfs_btnum_t		which,
730 	struct xfs_btree_cur	**curpp,
731 	struct xfs_buf		**agi_bpp)
732 {
733 	struct xfs_btree_cur	*cur;
734 	int			error;
735 
736 	ASSERT(*agi_bpp == NULL);
737 	ASSERT(*curpp == NULL);
738 
739 	error = xfs_ialloc_read_agi(pag, tp, agi_bpp);
740 	if (error)
741 		return error;
742 
743 	cur = xfs_inobt_init_cursor(pag, tp, *agi_bpp, which);
744 	*curpp = cur;
745 	return 0;
746 }
747 
748 static int
749 xfs_inobt_count_blocks(
750 	struct xfs_perag	*pag,
751 	struct xfs_trans	*tp,
752 	xfs_btnum_t		btnum,
753 	xfs_extlen_t		*tree_blocks)
754 {
755 	struct xfs_buf		*agbp = NULL;
756 	struct xfs_btree_cur	*cur = NULL;
757 	int			error;
758 
759 	error = xfs_inobt_cur(pag, tp, btnum, &cur, &agbp);
760 	if (error)
761 		return error;
762 
763 	error = xfs_btree_count_blocks(cur, tree_blocks);
764 	xfs_btree_del_cursor(cur, error);
765 	xfs_trans_brelse(tp, agbp);
766 
767 	return error;
768 }
769 
770 /* Read finobt block count from AGI header. */
771 static int
772 xfs_finobt_read_blocks(
773 	struct xfs_perag	*pag,
774 	struct xfs_trans	*tp,
775 	xfs_extlen_t		*tree_blocks)
776 {
777 	struct xfs_buf		*agbp;
778 	struct xfs_agi		*agi;
779 	int			error;
780 
781 	error = xfs_ialloc_read_agi(pag, tp, &agbp);
782 	if (error)
783 		return error;
784 
785 	agi = agbp->b_addr;
786 	*tree_blocks = be32_to_cpu(agi->agi_fblocks);
787 	xfs_trans_brelse(tp, agbp);
788 	return 0;
789 }
790 
791 /*
792  * Figure out how many blocks to reserve and how many are used by this btree.
793  */
794 int
795 xfs_finobt_calc_reserves(
796 	struct xfs_perag	*pag,
797 	struct xfs_trans	*tp,
798 	xfs_extlen_t		*ask,
799 	xfs_extlen_t		*used)
800 {
801 	xfs_extlen_t		tree_len = 0;
802 	int			error;
803 
804 	if (!xfs_has_finobt(pag->pag_mount))
805 		return 0;
806 
807 	if (xfs_has_inobtcounts(pag->pag_mount))
808 		error = xfs_finobt_read_blocks(pag, tp, &tree_len);
809 	else
810 		error = xfs_inobt_count_blocks(pag, tp, XFS_BTNUM_FINO,
811 				&tree_len);
812 	if (error)
813 		return error;
814 
815 	*ask += xfs_inobt_max_size(pag);
816 	*used += tree_len;
817 	return 0;
818 }
819 
820 /* Calculate the inobt btree size for some records. */
821 xfs_extlen_t
822 xfs_iallocbt_calc_size(
823 	struct xfs_mount	*mp,
824 	unsigned long long	len)
825 {
826 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);
827 }
828 
829 int __init
830 xfs_inobt_init_cur_cache(void)
831 {
832 	xfs_inobt_cur_cache = kmem_cache_create("xfs_inobt_cur",
833 			xfs_btree_cur_sizeof(xfs_inobt_maxlevels_ondisk()),
834 			0, 0, NULL);
835 
836 	if (!xfs_inobt_cur_cache)
837 		return -ENOMEM;
838 	return 0;
839 }
840 
841 void
842 xfs_inobt_destroy_cur_cache(void)
843 {
844 	kmem_cache_destroy(xfs_inobt_cur_cache);
845 	xfs_inobt_cur_cache = NULL;
846 }
847