xref: /openbmc/linux/fs/xfs/libxfs/xfs_ialloc_btree.c (revision 71de0a05)
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_inobt_mod_blockcount(cur, -1);
160 	return xfs_free_extent(cur->bc_tp,
161 			XFS_DADDR_TO_FSB(cur->bc_mp, xfs_buf_daddr(bp)), 1,
162 			&XFS_RMAP_OINFO_INOBT, resv);
163 }
164 
165 STATIC int
166 xfs_inobt_free_block(
167 	struct xfs_btree_cur	*cur,
168 	struct xfs_buf		*bp)
169 {
170 	return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
171 }
172 
173 STATIC int
174 xfs_finobt_free_block(
175 	struct xfs_btree_cur	*cur,
176 	struct xfs_buf		*bp)
177 {
178 	if (cur->bc_mp->m_finobt_nores)
179 		return xfs_inobt_free_block(cur, bp);
180 	return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_METADATA);
181 }
182 
183 STATIC int
184 xfs_inobt_get_maxrecs(
185 	struct xfs_btree_cur	*cur,
186 	int			level)
187 {
188 	return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
189 }
190 
191 STATIC void
192 xfs_inobt_init_key_from_rec(
193 	union xfs_btree_key		*key,
194 	const union xfs_btree_rec	*rec)
195 {
196 	key->inobt.ir_startino = rec->inobt.ir_startino;
197 }
198 
199 STATIC void
200 xfs_inobt_init_high_key_from_rec(
201 	union xfs_btree_key		*key,
202 	const union xfs_btree_rec	*rec)
203 {
204 	__u32				x;
205 
206 	x = be32_to_cpu(rec->inobt.ir_startino);
207 	x += XFS_INODES_PER_CHUNK - 1;
208 	key->inobt.ir_startino = cpu_to_be32(x);
209 }
210 
211 STATIC void
212 xfs_inobt_init_rec_from_cur(
213 	struct xfs_btree_cur	*cur,
214 	union xfs_btree_rec	*rec)
215 {
216 	rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
217 	if (xfs_has_sparseinodes(cur->bc_mp)) {
218 		rec->inobt.ir_u.sp.ir_holemask =
219 					cpu_to_be16(cur->bc_rec.i.ir_holemask);
220 		rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
221 		rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
222 	} else {
223 		/* ir_holemask/ir_count not supported on-disk */
224 		rec->inobt.ir_u.f.ir_freecount =
225 					cpu_to_be32(cur->bc_rec.i.ir_freecount);
226 	}
227 	rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
228 }
229 
230 /*
231  * initial value of ptr for lookup
232  */
233 STATIC void
234 xfs_inobt_init_ptr_from_cur(
235 	struct xfs_btree_cur	*cur,
236 	union xfs_btree_ptr	*ptr)
237 {
238 	struct xfs_agi		*agi = cur->bc_ag.agbp->b_addr;
239 
240 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
241 
242 	ptr->s = agi->agi_root;
243 }
244 
245 STATIC void
246 xfs_finobt_init_ptr_from_cur(
247 	struct xfs_btree_cur	*cur,
248 	union xfs_btree_ptr	*ptr)
249 {
250 	struct xfs_agi		*agi = cur->bc_ag.agbp->b_addr;
251 
252 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
253 	ptr->s = agi->agi_free_root;
254 }
255 
256 STATIC int64_t
257 xfs_inobt_key_diff(
258 	struct xfs_btree_cur		*cur,
259 	const union xfs_btree_key	*key)
260 {
261 	return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
262 			  cur->bc_rec.i.ir_startino;
263 }
264 
265 STATIC int64_t
266 xfs_inobt_diff_two_keys(
267 	struct xfs_btree_cur		*cur,
268 	const union xfs_btree_key	*k1,
269 	const union xfs_btree_key	*k2)
270 {
271 	return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
272 			  be32_to_cpu(k2->inobt.ir_startino);
273 }
274 
275 static xfs_failaddr_t
276 xfs_inobt_verify(
277 	struct xfs_buf		*bp)
278 {
279 	struct xfs_mount	*mp = bp->b_mount;
280 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
281 	xfs_failaddr_t		fa;
282 	unsigned int		level;
283 
284 	if (!xfs_verify_magic(bp, block->bb_magic))
285 		return __this_address;
286 
287 	/*
288 	 * During growfs operations, we can't verify the exact owner as the
289 	 * perag is not fully initialised and hence not attached to the buffer.
290 	 *
291 	 * Similarly, during log recovery we will have a perag structure
292 	 * attached, but the agi information will not yet have been initialised
293 	 * from the on disk AGI. We don't currently use any of this information,
294 	 * but beware of the landmine (i.e. need to check
295 	 * xfs_perag_initialised_agi(pag)) if we ever do.
296 	 */
297 	if (xfs_has_crc(mp)) {
298 		fa = xfs_btree_sblock_v5hdr_verify(bp);
299 		if (fa)
300 			return fa;
301 	}
302 
303 	/* level verification */
304 	level = be16_to_cpu(block->bb_level);
305 	if (level >= M_IGEO(mp)->inobt_maxlevels)
306 		return __this_address;
307 
308 	return xfs_btree_sblock_verify(bp,
309 			M_IGEO(mp)->inobt_mxr[level != 0]);
310 }
311 
312 static void
313 xfs_inobt_read_verify(
314 	struct xfs_buf	*bp)
315 {
316 	xfs_failaddr_t	fa;
317 
318 	if (!xfs_btree_sblock_verify_crc(bp))
319 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
320 	else {
321 		fa = xfs_inobt_verify(bp);
322 		if (fa)
323 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
324 	}
325 
326 	if (bp->b_error)
327 		trace_xfs_btree_corrupt(bp, _RET_IP_);
328 }
329 
330 static void
331 xfs_inobt_write_verify(
332 	struct xfs_buf	*bp)
333 {
334 	xfs_failaddr_t	fa;
335 
336 	fa = xfs_inobt_verify(bp);
337 	if (fa) {
338 		trace_xfs_btree_corrupt(bp, _RET_IP_);
339 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
340 		return;
341 	}
342 	xfs_btree_sblock_calc_crc(bp);
343 
344 }
345 
346 const struct xfs_buf_ops xfs_inobt_buf_ops = {
347 	.name = "xfs_inobt",
348 	.magic = { cpu_to_be32(XFS_IBT_MAGIC), cpu_to_be32(XFS_IBT_CRC_MAGIC) },
349 	.verify_read = xfs_inobt_read_verify,
350 	.verify_write = xfs_inobt_write_verify,
351 	.verify_struct = xfs_inobt_verify,
352 };
353 
354 const struct xfs_buf_ops xfs_finobt_buf_ops = {
355 	.name = "xfs_finobt",
356 	.magic = { cpu_to_be32(XFS_FIBT_MAGIC),
357 		   cpu_to_be32(XFS_FIBT_CRC_MAGIC) },
358 	.verify_read = xfs_inobt_read_verify,
359 	.verify_write = xfs_inobt_write_verify,
360 	.verify_struct = xfs_inobt_verify,
361 };
362 
363 STATIC int
364 xfs_inobt_keys_inorder(
365 	struct xfs_btree_cur		*cur,
366 	const union xfs_btree_key	*k1,
367 	const union xfs_btree_key	*k2)
368 {
369 	return be32_to_cpu(k1->inobt.ir_startino) <
370 		be32_to_cpu(k2->inobt.ir_startino);
371 }
372 
373 STATIC int
374 xfs_inobt_recs_inorder(
375 	struct xfs_btree_cur		*cur,
376 	const union xfs_btree_rec	*r1,
377 	const union xfs_btree_rec	*r2)
378 {
379 	return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
380 		be32_to_cpu(r2->inobt.ir_startino);
381 }
382 
383 static const struct xfs_btree_ops xfs_inobt_ops = {
384 	.rec_len		= sizeof(xfs_inobt_rec_t),
385 	.key_len		= sizeof(xfs_inobt_key_t),
386 
387 	.dup_cursor		= xfs_inobt_dup_cursor,
388 	.set_root		= xfs_inobt_set_root,
389 	.alloc_block		= xfs_inobt_alloc_block,
390 	.free_block		= xfs_inobt_free_block,
391 	.get_minrecs		= xfs_inobt_get_minrecs,
392 	.get_maxrecs		= xfs_inobt_get_maxrecs,
393 	.init_key_from_rec	= xfs_inobt_init_key_from_rec,
394 	.init_high_key_from_rec	= xfs_inobt_init_high_key_from_rec,
395 	.init_rec_from_cur	= xfs_inobt_init_rec_from_cur,
396 	.init_ptr_from_cur	= xfs_inobt_init_ptr_from_cur,
397 	.key_diff		= xfs_inobt_key_diff,
398 	.buf_ops		= &xfs_inobt_buf_ops,
399 	.diff_two_keys		= xfs_inobt_diff_two_keys,
400 	.keys_inorder		= xfs_inobt_keys_inorder,
401 	.recs_inorder		= xfs_inobt_recs_inorder,
402 };
403 
404 static const struct xfs_btree_ops xfs_finobt_ops = {
405 	.rec_len		= sizeof(xfs_inobt_rec_t),
406 	.key_len		= sizeof(xfs_inobt_key_t),
407 
408 	.dup_cursor		= xfs_inobt_dup_cursor,
409 	.set_root		= xfs_finobt_set_root,
410 	.alloc_block		= xfs_finobt_alloc_block,
411 	.free_block		= xfs_finobt_free_block,
412 	.get_minrecs		= xfs_inobt_get_minrecs,
413 	.get_maxrecs		= xfs_inobt_get_maxrecs,
414 	.init_key_from_rec	= xfs_inobt_init_key_from_rec,
415 	.init_high_key_from_rec	= xfs_inobt_init_high_key_from_rec,
416 	.init_rec_from_cur	= xfs_inobt_init_rec_from_cur,
417 	.init_ptr_from_cur	= xfs_finobt_init_ptr_from_cur,
418 	.key_diff		= xfs_inobt_key_diff,
419 	.buf_ops		= &xfs_finobt_buf_ops,
420 	.diff_two_keys		= xfs_inobt_diff_two_keys,
421 	.keys_inorder		= xfs_inobt_keys_inorder,
422 	.recs_inorder		= xfs_inobt_recs_inorder,
423 };
424 
425 /*
426  * Initialize a new inode btree cursor.
427  */
428 static struct xfs_btree_cur *
429 xfs_inobt_init_common(
430 	struct xfs_perag	*pag,
431 	struct xfs_trans	*tp,		/* transaction pointer */
432 	xfs_btnum_t		btnum)		/* ialloc or free ino btree */
433 {
434 	struct xfs_mount	*mp = pag->pag_mount;
435 	struct xfs_btree_cur	*cur;
436 
437 	cur = xfs_btree_alloc_cursor(mp, tp, btnum,
438 			M_IGEO(mp)->inobt_maxlevels, xfs_inobt_cur_cache);
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 	if (xfs_has_crc(mp))
448 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
449 
450 	/* take a reference for the cursor */
451 	atomic_inc(&pag->pag_ref);
452 	cur->bc_ag.pag = pag;
453 	return cur;
454 }
455 
456 /* Create an inode btree cursor. */
457 struct xfs_btree_cur *
458 xfs_inobt_init_cursor(
459 	struct xfs_perag	*pag,
460 	struct xfs_trans	*tp,
461 	struct xfs_buf		*agbp,
462 	xfs_btnum_t		btnum)
463 {
464 	struct xfs_btree_cur	*cur;
465 	struct xfs_agi		*agi = agbp->b_addr;
466 
467 	cur = xfs_inobt_init_common(pag, tp, btnum);
468 	if (btnum == XFS_BTNUM_INO)
469 		cur->bc_nlevels = be32_to_cpu(agi->agi_level);
470 	else
471 		cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
472 	cur->bc_ag.agbp = agbp;
473 	return cur;
474 }
475 
476 /* Create an inode btree cursor with a fake root for staging. */
477 struct xfs_btree_cur *
478 xfs_inobt_stage_cursor(
479 	struct xfs_perag	*pag,
480 	struct xbtree_afakeroot	*afake,
481 	xfs_btnum_t		btnum)
482 {
483 	struct xfs_btree_cur	*cur;
484 
485 	cur = xfs_inobt_init_common(pag, NULL, btnum);
486 	xfs_btree_stage_afakeroot(cur, afake);
487 	return cur;
488 }
489 
490 /*
491  * Install a new inobt btree root.  Caller is responsible for invalidating
492  * and freeing the old btree blocks.
493  */
494 void
495 xfs_inobt_commit_staged_btree(
496 	struct xfs_btree_cur	*cur,
497 	struct xfs_trans	*tp,
498 	struct xfs_buf		*agbp)
499 {
500 	struct xfs_agi		*agi = agbp->b_addr;
501 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
502 	int			fields;
503 
504 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
505 
506 	if (cur->bc_btnum == XFS_BTNUM_INO) {
507 		fields = XFS_AGI_ROOT | XFS_AGI_LEVEL;
508 		agi->agi_root = cpu_to_be32(afake->af_root);
509 		agi->agi_level = cpu_to_be32(afake->af_levels);
510 		if (xfs_has_inobtcounts(cur->bc_mp)) {
511 			agi->agi_iblocks = cpu_to_be32(afake->af_blocks);
512 			fields |= XFS_AGI_IBLOCKS;
513 		}
514 		xfs_ialloc_log_agi(tp, agbp, fields);
515 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_inobt_ops);
516 	} else {
517 		fields = XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL;
518 		agi->agi_free_root = cpu_to_be32(afake->af_root);
519 		agi->agi_free_level = cpu_to_be32(afake->af_levels);
520 		if (xfs_has_inobtcounts(cur->bc_mp)) {
521 			agi->agi_fblocks = cpu_to_be32(afake->af_blocks);
522 			fields |= XFS_AGI_IBLOCKS;
523 		}
524 		xfs_ialloc_log_agi(tp, agbp, fields);
525 		xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_finobt_ops);
526 	}
527 }
528 
529 /* Calculate number of records in an inode btree block. */
530 static inline unsigned int
531 xfs_inobt_block_maxrecs(
532 	unsigned int		blocklen,
533 	bool			leaf)
534 {
535 	if (leaf)
536 		return blocklen / sizeof(xfs_inobt_rec_t);
537 	return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
538 }
539 
540 /*
541  * Calculate number of records in an inobt btree block.
542  */
543 int
544 xfs_inobt_maxrecs(
545 	struct xfs_mount	*mp,
546 	int			blocklen,
547 	int			leaf)
548 {
549 	blocklen -= XFS_INOBT_BLOCK_LEN(mp);
550 	return xfs_inobt_block_maxrecs(blocklen, leaf);
551 }
552 
553 /*
554  * Maximum number of inode btree records per AG.  Pretend that we can fill an
555  * entire AG completely full of inodes except for the AG headers.
556  */
557 #define XFS_MAX_INODE_RECORDS \
558 	((XFS_MAX_AG_BYTES - (4 * BBSIZE)) / XFS_DINODE_MIN_SIZE) / \
559 			XFS_INODES_PER_CHUNK
560 
561 /* Compute the max possible height for the inode btree. */
562 static inline unsigned int
563 xfs_inobt_maxlevels_ondisk(void)
564 {
565 	unsigned int		minrecs[2];
566 	unsigned int		blocklen;
567 
568 	blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
569 		       XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
570 
571 	minrecs[0] = xfs_inobt_block_maxrecs(blocklen, true) / 2;
572 	minrecs[1] = xfs_inobt_block_maxrecs(blocklen, false) / 2;
573 
574 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_INODE_RECORDS);
575 }
576 
577 /* Compute the max possible height for the free inode btree. */
578 static inline unsigned int
579 xfs_finobt_maxlevels_ondisk(void)
580 {
581 	unsigned int		minrecs[2];
582 	unsigned int		blocklen;
583 
584 	blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
585 
586 	minrecs[0] = xfs_inobt_block_maxrecs(blocklen, true) / 2;
587 	minrecs[1] = xfs_inobt_block_maxrecs(blocklen, false) / 2;
588 
589 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_INODE_RECORDS);
590 }
591 
592 /* Compute the max possible height for either inode btree. */
593 unsigned int
594 xfs_iallocbt_maxlevels_ondisk(void)
595 {
596 	return max(xfs_inobt_maxlevels_ondisk(),
597 		   xfs_finobt_maxlevels_ondisk());
598 }
599 
600 /*
601  * Convert the inode record holemask to an inode allocation bitmap. The inode
602  * allocation bitmap is inode granularity and specifies whether an inode is
603  * physically allocated on disk (not whether the inode is considered allocated
604  * or free by the fs).
605  *
606  * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
607  */
608 uint64_t
609 xfs_inobt_irec_to_allocmask(
610 	struct xfs_inobt_rec_incore	*rec)
611 {
612 	uint64_t			bitmap = 0;
613 	uint64_t			inodespbit;
614 	int				nextbit;
615 	uint				allocbitmap;
616 
617 	/*
618 	 * The holemask has 16-bits for a 64 inode record. Therefore each
619 	 * holemask bit represents multiple inodes. Create a mask of bits to set
620 	 * in the allocmask for each holemask bit.
621 	 */
622 	inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
623 
624 	/*
625 	 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
626 	 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
627 	 * anything beyond the 16 holemask bits since this casts to a larger
628 	 * type.
629 	 */
630 	allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
631 
632 	/*
633 	 * allocbitmap is the inverted holemask so every set bit represents
634 	 * allocated inodes. To expand from 16-bit holemask granularity to
635 	 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
636 	 * bitmap for every holemask bit.
637 	 */
638 	nextbit = xfs_next_bit(&allocbitmap, 1, 0);
639 	while (nextbit != -1) {
640 		ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
641 
642 		bitmap |= (inodespbit <<
643 			   (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
644 
645 		nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
646 	}
647 
648 	return bitmap;
649 }
650 
651 #if defined(DEBUG) || defined(XFS_WARN)
652 /*
653  * Verify that an in-core inode record has a valid inode count.
654  */
655 int
656 xfs_inobt_rec_check_count(
657 	struct xfs_mount		*mp,
658 	struct xfs_inobt_rec_incore	*rec)
659 {
660 	int				inocount = 0;
661 	int				nextbit = 0;
662 	uint64_t			allocbmap;
663 	int				wordsz;
664 
665 	wordsz = sizeof(allocbmap) / sizeof(unsigned int);
666 	allocbmap = xfs_inobt_irec_to_allocmask(rec);
667 
668 	nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
669 	while (nextbit != -1) {
670 		inocount++;
671 		nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
672 				       nextbit + 1);
673 	}
674 
675 	if (inocount != rec->ir_count)
676 		return -EFSCORRUPTED;
677 
678 	return 0;
679 }
680 #endif	/* DEBUG */
681 
682 static xfs_extlen_t
683 xfs_inobt_max_size(
684 	struct xfs_perag	*pag)
685 {
686 	struct xfs_mount	*mp = pag->pag_mount;
687 	xfs_agblock_t		agblocks = pag->block_count;
688 
689 	/* Bail out if we're uninitialized, which can happen in mkfs. */
690 	if (M_IGEO(mp)->inobt_mxr[0] == 0)
691 		return 0;
692 
693 	/*
694 	 * The log is permanently allocated, so the space it occupies will
695 	 * never be available for the kinds of things that would require btree
696 	 * expansion.  We therefore can pretend the space isn't there.
697 	 */
698 	if (xfs_ag_contains_log(mp, pag->pag_agno))
699 		agblocks -= mp->m_sb.sb_logblocks;
700 
701 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr,
702 				(uint64_t)agblocks * mp->m_sb.sb_inopblock /
703 					XFS_INODES_PER_CHUNK);
704 }
705 
706 /* Read AGI and create inobt cursor. */
707 int
708 xfs_inobt_cur(
709 	struct xfs_perag	*pag,
710 	struct xfs_trans	*tp,
711 	xfs_btnum_t		which,
712 	struct xfs_btree_cur	**curpp,
713 	struct xfs_buf		**agi_bpp)
714 {
715 	struct xfs_btree_cur	*cur;
716 	int			error;
717 
718 	ASSERT(*agi_bpp == NULL);
719 	ASSERT(*curpp == NULL);
720 
721 	error = xfs_ialloc_read_agi(pag, tp, agi_bpp);
722 	if (error)
723 		return error;
724 
725 	cur = xfs_inobt_init_cursor(pag, tp, *agi_bpp, which);
726 	*curpp = cur;
727 	return 0;
728 }
729 
730 static int
731 xfs_inobt_count_blocks(
732 	struct xfs_perag	*pag,
733 	struct xfs_trans	*tp,
734 	xfs_btnum_t		btnum,
735 	xfs_extlen_t		*tree_blocks)
736 {
737 	struct xfs_buf		*agbp = NULL;
738 	struct xfs_btree_cur	*cur = NULL;
739 	int			error;
740 
741 	error = xfs_inobt_cur(pag, tp, btnum, &cur, &agbp);
742 	if (error)
743 		return error;
744 
745 	error = xfs_btree_count_blocks(cur, tree_blocks);
746 	xfs_btree_del_cursor(cur, error);
747 	xfs_trans_brelse(tp, agbp);
748 
749 	return error;
750 }
751 
752 /* Read finobt block count from AGI header. */
753 static int
754 xfs_finobt_read_blocks(
755 	struct xfs_perag	*pag,
756 	struct xfs_trans	*tp,
757 	xfs_extlen_t		*tree_blocks)
758 {
759 	struct xfs_buf		*agbp;
760 	struct xfs_agi		*agi;
761 	int			error;
762 
763 	error = xfs_ialloc_read_agi(pag, tp, &agbp);
764 	if (error)
765 		return error;
766 
767 	agi = agbp->b_addr;
768 	*tree_blocks = be32_to_cpu(agi->agi_fblocks);
769 	xfs_trans_brelse(tp, agbp);
770 	return 0;
771 }
772 
773 /*
774  * Figure out how many blocks to reserve and how many are used by this btree.
775  */
776 int
777 xfs_finobt_calc_reserves(
778 	struct xfs_perag	*pag,
779 	struct xfs_trans	*tp,
780 	xfs_extlen_t		*ask,
781 	xfs_extlen_t		*used)
782 {
783 	xfs_extlen_t		tree_len = 0;
784 	int			error;
785 
786 	if (!xfs_has_finobt(pag->pag_mount))
787 		return 0;
788 
789 	if (xfs_has_inobtcounts(pag->pag_mount))
790 		error = xfs_finobt_read_blocks(pag, tp, &tree_len);
791 	else
792 		error = xfs_inobt_count_blocks(pag, tp, XFS_BTNUM_FINO,
793 				&tree_len);
794 	if (error)
795 		return error;
796 
797 	*ask += xfs_inobt_max_size(pag);
798 	*used += tree_len;
799 	return 0;
800 }
801 
802 /* Calculate the inobt btree size for some records. */
803 xfs_extlen_t
804 xfs_iallocbt_calc_size(
805 	struct xfs_mount	*mp,
806 	unsigned long long	len)
807 {
808 	return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);
809 }
810 
811 int __init
812 xfs_inobt_init_cur_cache(void)
813 {
814 	xfs_inobt_cur_cache = kmem_cache_create("xfs_inobt_cur",
815 			xfs_btree_cur_sizeof(xfs_inobt_maxlevels_ondisk()),
816 			0, 0, NULL);
817 
818 	if (!xfs_inobt_cur_cache)
819 		return -ENOMEM;
820 	return 0;
821 }
822 
823 void
824 xfs_inobt_destroy_cur_cache(void)
825 {
826 	kmem_cache_destroy(xfs_inobt_cur_cache);
827 	xfs_inobt_cur_cache = NULL;
828 }
829