xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap_btree.c (revision 692b6cdd)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2003,2005 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_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_alloc.h"
17 #include "xfs_btree.h"
18 #include "xfs_bmap_btree.h"
19 #include "xfs_bmap.h"
20 #include "xfs_error.h"
21 #include "xfs_quota.h"
22 #include "xfs_trace.h"
23 #include "xfs_rmap.h"
24 
25 static struct kmem_cache	*xfs_bmbt_cur_cache;
26 
27 /*
28  * Convert on-disk form of btree root to in-memory form.
29  */
30 void
31 xfs_bmdr_to_bmbt(
32 	struct xfs_inode	*ip,
33 	xfs_bmdr_block_t	*dblock,
34 	int			dblocklen,
35 	struct xfs_btree_block	*rblock,
36 	int			rblocklen)
37 {
38 	struct xfs_mount	*mp = ip->i_mount;
39 	int			dmxr;
40 	xfs_bmbt_key_t		*fkp;
41 	__be64			*fpp;
42 	xfs_bmbt_key_t		*tkp;
43 	__be64			*tpp;
44 
45 	xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
46 				 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
47 				 XFS_BTREE_LONG_PTRS);
48 	rblock->bb_level = dblock->bb_level;
49 	ASSERT(be16_to_cpu(rblock->bb_level) > 0);
50 	rblock->bb_numrecs = dblock->bb_numrecs;
51 	dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
52 	fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
53 	tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
54 	fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
55 	tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
56 	dmxr = be16_to_cpu(dblock->bb_numrecs);
57 	memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
58 	memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
59 }
60 
61 void
62 xfs_bmbt_disk_get_all(
63 	const struct xfs_bmbt_rec *rec,
64 	struct xfs_bmbt_irec	*irec)
65 {
66 	uint64_t		l0 = get_unaligned_be64(&rec->l0);
67 	uint64_t		l1 = get_unaligned_be64(&rec->l1);
68 
69 	irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
70 	irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
71 	irec->br_blockcount = l1 & xfs_mask64lo(21);
72 	if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
73 		irec->br_state = XFS_EXT_UNWRITTEN;
74 	else
75 		irec->br_state = XFS_EXT_NORM;
76 }
77 
78 /*
79  * Extract the blockcount field from an on disk bmap extent record.
80  */
81 xfs_filblks_t
82 xfs_bmbt_disk_get_blockcount(
83 	const struct xfs_bmbt_rec	*r)
84 {
85 	return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
86 }
87 
88 /*
89  * Extract the startoff field from a disk format bmap extent record.
90  */
91 xfs_fileoff_t
92 xfs_bmbt_disk_get_startoff(
93 	const struct xfs_bmbt_rec	*r)
94 {
95 	return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
96 		 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
97 }
98 
99 /*
100  * Set all the fields in a bmap extent record from the uncompressed form.
101  */
102 void
103 xfs_bmbt_disk_set_all(
104 	struct xfs_bmbt_rec	*r,
105 	struct xfs_bmbt_irec	*s)
106 {
107 	int			extent_flag = (s->br_state != XFS_EXT_NORM);
108 
109 	ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
110 	ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
111 	ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
112 	ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
113 
114 	put_unaligned_be64(
115 		((xfs_bmbt_rec_base_t)extent_flag << 63) |
116 		 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
117 		 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
118 	put_unaligned_be64(
119 		((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
120 		 ((xfs_bmbt_rec_base_t)s->br_blockcount &
121 		  (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
122 }
123 
124 /*
125  * Convert in-memory form of btree root to on-disk form.
126  */
127 void
128 xfs_bmbt_to_bmdr(
129 	struct xfs_mount	*mp,
130 	struct xfs_btree_block	*rblock,
131 	int			rblocklen,
132 	xfs_bmdr_block_t	*dblock,
133 	int			dblocklen)
134 {
135 	int			dmxr;
136 	xfs_bmbt_key_t		*fkp;
137 	__be64			*fpp;
138 	xfs_bmbt_key_t		*tkp;
139 	__be64			*tpp;
140 
141 	if (xfs_has_crc(mp)) {
142 		ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
143 		ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
144 		       &mp->m_sb.sb_meta_uuid));
145 		ASSERT(rblock->bb_u.l.bb_blkno ==
146 		       cpu_to_be64(XFS_BUF_DADDR_NULL));
147 	} else
148 		ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
149 	ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
150 	ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
151 	ASSERT(rblock->bb_level != 0);
152 	dblock->bb_level = rblock->bb_level;
153 	dblock->bb_numrecs = rblock->bb_numrecs;
154 	dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
155 	fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
156 	tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
157 	fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
158 	tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
159 	dmxr = be16_to_cpu(dblock->bb_numrecs);
160 	memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
161 	memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
162 }
163 
164 STATIC struct xfs_btree_cur *
165 xfs_bmbt_dup_cursor(
166 	struct xfs_btree_cur	*cur)
167 {
168 	struct xfs_btree_cur	*new;
169 
170 	new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
171 			cur->bc_ino.ip, cur->bc_ino.whichfork);
172 
173 	/*
174 	 * Copy the firstblock, dfops, and flags values,
175 	 * since init cursor doesn't get them.
176 	 */
177 	new->bc_ino.flags = cur->bc_ino.flags;
178 
179 	return new;
180 }
181 
182 STATIC void
183 xfs_bmbt_update_cursor(
184 	struct xfs_btree_cur	*src,
185 	struct xfs_btree_cur	*dst)
186 {
187 	ASSERT((dst->bc_tp->t_highest_agno != NULLAGNUMBER) ||
188 	       (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
189 
190 	dst->bc_ino.allocated += src->bc_ino.allocated;
191 	dst->bc_tp->t_highest_agno = src->bc_tp->t_highest_agno;
192 
193 	src->bc_ino.allocated = 0;
194 }
195 
196 STATIC int
197 xfs_bmbt_alloc_block(
198 	struct xfs_btree_cur		*cur,
199 	const union xfs_btree_ptr	*start,
200 	union xfs_btree_ptr		*new,
201 	int				*stat)
202 {
203 	xfs_alloc_arg_t		args;		/* block allocation args */
204 	int			error;		/* error return value */
205 
206 	memset(&args, 0, sizeof(args));
207 	args.tp = cur->bc_tp;
208 	args.mp = cur->bc_mp;
209 	xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
210 			cur->bc_ino.whichfork);
211 
212 	args.fsbno = be64_to_cpu(start->l);
213 	args.type = XFS_ALLOCTYPE_START_BNO;
214 
215 	/*
216 	 * If we are coming here from something like unwritten extent
217 	 * conversion, there has been no data extent allocation already done, so
218 	 * we have to ensure that we attempt to locate the entire set of bmbt
219 	 * allocations in the same AG, as xfs_bmapi_write() would have reserved.
220 	 */
221 	if (cur->bc_tp->t_highest_agno == NULLAGNUMBER)
222 		args.minleft = xfs_bmapi_minleft(cur->bc_tp, cur->bc_ino.ip,
223 					cur->bc_ino.whichfork);
224 
225 	args.minlen = args.maxlen = args.prod = 1;
226 	args.wasdel = cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL;
227 	if (!args.wasdel && args.tp->t_blk_res == 0) {
228 		error = -ENOSPC;
229 		goto error0;
230 	}
231 	error = xfs_alloc_vextent(&args);
232 	if (error)
233 		goto error0;
234 
235 	if (args.fsbno == NULLFSBLOCK && args.minleft) {
236 		/*
237 		 * Could not find an AG with enough free space to satisfy
238 		 * a full btree split.  Try again and if
239 		 * successful activate the lowspace algorithm.
240 		 */
241 		args.fsbno = 0;
242 		args.minleft = 0;
243 		args.type = XFS_ALLOCTYPE_START_BNO;
244 		error = xfs_alloc_vextent(&args);
245 		if (error)
246 			goto error0;
247 		cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
248 	}
249 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
250 		*stat = 0;
251 		return 0;
252 	}
253 
254 	ASSERT(args.len == 1);
255 	cur->bc_ino.allocated++;
256 	cur->bc_ino.ip->i_nblocks++;
257 	xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
258 	xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
259 			XFS_TRANS_DQ_BCOUNT, 1L);
260 
261 	new->l = cpu_to_be64(args.fsbno);
262 
263 	*stat = 1;
264 	return 0;
265 
266  error0:
267 	return error;
268 }
269 
270 STATIC int
271 xfs_bmbt_free_block(
272 	struct xfs_btree_cur	*cur,
273 	struct xfs_buf		*bp)
274 {
275 	struct xfs_mount	*mp = cur->bc_mp;
276 	struct xfs_inode	*ip = cur->bc_ino.ip;
277 	struct xfs_trans	*tp = cur->bc_tp;
278 	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
279 	struct xfs_owner_info	oinfo;
280 
281 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
282 	xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
283 	ip->i_nblocks--;
284 
285 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
286 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
287 	return 0;
288 }
289 
290 STATIC int
291 xfs_bmbt_get_minrecs(
292 	struct xfs_btree_cur	*cur,
293 	int			level)
294 {
295 	if (level == cur->bc_nlevels - 1) {
296 		struct xfs_ifork	*ifp;
297 
298 		ifp = xfs_ifork_ptr(cur->bc_ino.ip,
299 				    cur->bc_ino.whichfork);
300 
301 		return xfs_bmbt_maxrecs(cur->bc_mp,
302 					ifp->if_broot_bytes, level == 0) / 2;
303 	}
304 
305 	return cur->bc_mp->m_bmap_dmnr[level != 0];
306 }
307 
308 int
309 xfs_bmbt_get_maxrecs(
310 	struct xfs_btree_cur	*cur,
311 	int			level)
312 {
313 	if (level == cur->bc_nlevels - 1) {
314 		struct xfs_ifork	*ifp;
315 
316 		ifp = xfs_ifork_ptr(cur->bc_ino.ip,
317 				    cur->bc_ino.whichfork);
318 
319 		return xfs_bmbt_maxrecs(cur->bc_mp,
320 					ifp->if_broot_bytes, level == 0);
321 	}
322 
323 	return cur->bc_mp->m_bmap_dmxr[level != 0];
324 
325 }
326 
327 /*
328  * Get the maximum records we could store in the on-disk format.
329  *
330  * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
331  * for the root node this checks the available space in the dinode fork
332  * so that we can resize the in-memory buffer to match it.  After a
333  * resize to the maximum size this function returns the same value
334  * as xfs_bmbt_get_maxrecs for the root node, too.
335  */
336 STATIC int
337 xfs_bmbt_get_dmaxrecs(
338 	struct xfs_btree_cur	*cur,
339 	int			level)
340 {
341 	if (level != cur->bc_nlevels - 1)
342 		return cur->bc_mp->m_bmap_dmxr[level != 0];
343 	return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
344 }
345 
346 STATIC void
347 xfs_bmbt_init_key_from_rec(
348 	union xfs_btree_key		*key,
349 	const union xfs_btree_rec	*rec)
350 {
351 	key->bmbt.br_startoff =
352 		cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
353 }
354 
355 STATIC void
356 xfs_bmbt_init_high_key_from_rec(
357 	union xfs_btree_key		*key,
358 	const union xfs_btree_rec	*rec)
359 {
360 	key->bmbt.br_startoff = cpu_to_be64(
361 			xfs_bmbt_disk_get_startoff(&rec->bmbt) +
362 			xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
363 }
364 
365 STATIC void
366 xfs_bmbt_init_rec_from_cur(
367 	struct xfs_btree_cur	*cur,
368 	union xfs_btree_rec	*rec)
369 {
370 	xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
371 }
372 
373 STATIC void
374 xfs_bmbt_init_ptr_from_cur(
375 	struct xfs_btree_cur	*cur,
376 	union xfs_btree_ptr	*ptr)
377 {
378 	ptr->l = 0;
379 }
380 
381 STATIC int64_t
382 xfs_bmbt_key_diff(
383 	struct xfs_btree_cur		*cur,
384 	const union xfs_btree_key	*key)
385 {
386 	return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
387 				      cur->bc_rec.b.br_startoff;
388 }
389 
390 STATIC int64_t
391 xfs_bmbt_diff_two_keys(
392 	struct xfs_btree_cur		*cur,
393 	const union xfs_btree_key	*k1,
394 	const union xfs_btree_key	*k2)
395 {
396 	uint64_t			a = be64_to_cpu(k1->bmbt.br_startoff);
397 	uint64_t			b = be64_to_cpu(k2->bmbt.br_startoff);
398 
399 	/*
400 	 * Note: This routine previously casted a and b to int64 and subtracted
401 	 * them to generate a result.  This lead to problems if b was the
402 	 * "maximum" key value (all ones) being signed incorrectly, hence this
403 	 * somewhat less efficient version.
404 	 */
405 	if (a > b)
406 		return 1;
407 	if (b > a)
408 		return -1;
409 	return 0;
410 }
411 
412 static xfs_failaddr_t
413 xfs_bmbt_verify(
414 	struct xfs_buf		*bp)
415 {
416 	struct xfs_mount	*mp = bp->b_mount;
417 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
418 	xfs_failaddr_t		fa;
419 	unsigned int		level;
420 
421 	if (!xfs_verify_magic(bp, block->bb_magic))
422 		return __this_address;
423 
424 	if (xfs_has_crc(mp)) {
425 		/*
426 		 * XXX: need a better way of verifying the owner here. Right now
427 		 * just make sure there has been one set.
428 		 */
429 		fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
430 		if (fa)
431 			return fa;
432 	}
433 
434 	/*
435 	 * numrecs and level verification.
436 	 *
437 	 * We don't know what fork we belong to, so just verify that the level
438 	 * is less than the maximum of the two. Later checks will be more
439 	 * precise.
440 	 */
441 	level = be16_to_cpu(block->bb_level);
442 	if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
443 		return __this_address;
444 
445 	return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
446 }
447 
448 static void
449 xfs_bmbt_read_verify(
450 	struct xfs_buf	*bp)
451 {
452 	xfs_failaddr_t	fa;
453 
454 	if (!xfs_btree_lblock_verify_crc(bp))
455 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
456 	else {
457 		fa = xfs_bmbt_verify(bp);
458 		if (fa)
459 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
460 	}
461 
462 	if (bp->b_error)
463 		trace_xfs_btree_corrupt(bp, _RET_IP_);
464 }
465 
466 static void
467 xfs_bmbt_write_verify(
468 	struct xfs_buf	*bp)
469 {
470 	xfs_failaddr_t	fa;
471 
472 	fa = xfs_bmbt_verify(bp);
473 	if (fa) {
474 		trace_xfs_btree_corrupt(bp, _RET_IP_);
475 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
476 		return;
477 	}
478 	xfs_btree_lblock_calc_crc(bp);
479 }
480 
481 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
482 	.name = "xfs_bmbt",
483 	.magic = { cpu_to_be32(XFS_BMAP_MAGIC),
484 		   cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
485 	.verify_read = xfs_bmbt_read_verify,
486 	.verify_write = xfs_bmbt_write_verify,
487 	.verify_struct = xfs_bmbt_verify,
488 };
489 
490 
491 STATIC int
492 xfs_bmbt_keys_inorder(
493 	struct xfs_btree_cur		*cur,
494 	const union xfs_btree_key	*k1,
495 	const union xfs_btree_key	*k2)
496 {
497 	return be64_to_cpu(k1->bmbt.br_startoff) <
498 		be64_to_cpu(k2->bmbt.br_startoff);
499 }
500 
501 STATIC int
502 xfs_bmbt_recs_inorder(
503 	struct xfs_btree_cur		*cur,
504 	const union xfs_btree_rec	*r1,
505 	const union xfs_btree_rec	*r2)
506 {
507 	return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
508 		xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
509 		xfs_bmbt_disk_get_startoff(&r2->bmbt);
510 }
511 
512 static const struct xfs_btree_ops xfs_bmbt_ops = {
513 	.rec_len		= sizeof(xfs_bmbt_rec_t),
514 	.key_len		= sizeof(xfs_bmbt_key_t),
515 
516 	.dup_cursor		= xfs_bmbt_dup_cursor,
517 	.update_cursor		= xfs_bmbt_update_cursor,
518 	.alloc_block		= xfs_bmbt_alloc_block,
519 	.free_block		= xfs_bmbt_free_block,
520 	.get_maxrecs		= xfs_bmbt_get_maxrecs,
521 	.get_minrecs		= xfs_bmbt_get_minrecs,
522 	.get_dmaxrecs		= xfs_bmbt_get_dmaxrecs,
523 	.init_key_from_rec	= xfs_bmbt_init_key_from_rec,
524 	.init_high_key_from_rec	= xfs_bmbt_init_high_key_from_rec,
525 	.init_rec_from_cur	= xfs_bmbt_init_rec_from_cur,
526 	.init_ptr_from_cur	= xfs_bmbt_init_ptr_from_cur,
527 	.key_diff		= xfs_bmbt_key_diff,
528 	.diff_two_keys		= xfs_bmbt_diff_two_keys,
529 	.buf_ops		= &xfs_bmbt_buf_ops,
530 	.keys_inorder		= xfs_bmbt_keys_inorder,
531 	.recs_inorder		= xfs_bmbt_recs_inorder,
532 };
533 
534 /*
535  * Allocate a new bmap btree cursor.
536  */
537 struct xfs_btree_cur *				/* new bmap btree cursor */
538 xfs_bmbt_init_cursor(
539 	struct xfs_mount	*mp,		/* file system mount point */
540 	struct xfs_trans	*tp,		/* transaction pointer */
541 	struct xfs_inode	*ip,		/* inode owning the btree */
542 	int			whichfork)	/* data or attr fork */
543 {
544 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
545 	struct xfs_btree_cur	*cur;
546 	ASSERT(whichfork != XFS_COW_FORK);
547 
548 	cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
549 			mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
550 	cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
551 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
552 
553 	cur->bc_ops = &xfs_bmbt_ops;
554 	cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
555 	if (xfs_has_crc(mp))
556 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
557 
558 	cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
559 	cur->bc_ino.ip = ip;
560 	cur->bc_ino.allocated = 0;
561 	cur->bc_ino.flags = 0;
562 	cur->bc_ino.whichfork = whichfork;
563 
564 	return cur;
565 }
566 
567 /* Calculate number of records in a block mapping btree block. */
568 static inline unsigned int
569 xfs_bmbt_block_maxrecs(
570 	unsigned int		blocklen,
571 	bool			leaf)
572 {
573 	if (leaf)
574 		return blocklen / sizeof(xfs_bmbt_rec_t);
575 	return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
576 }
577 
578 /*
579  * Calculate number of records in a bmap btree block.
580  */
581 int
582 xfs_bmbt_maxrecs(
583 	struct xfs_mount	*mp,
584 	int			blocklen,
585 	int			leaf)
586 {
587 	blocklen -= XFS_BMBT_BLOCK_LEN(mp);
588 	return xfs_bmbt_block_maxrecs(blocklen, leaf);
589 }
590 
591 /*
592  * Calculate the maximum possible height of the btree that the on-disk format
593  * supports. This is used for sizing structures large enough to support every
594  * possible configuration of a filesystem that might get mounted.
595  */
596 unsigned int
597 xfs_bmbt_maxlevels_ondisk(void)
598 {
599 	unsigned int		minrecs[2];
600 	unsigned int		blocklen;
601 
602 	blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
603 		       XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
604 
605 	minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
606 	minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
607 
608 	/* One extra level for the inode root. */
609 	return xfs_btree_compute_maxlevels(minrecs,
610 			XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
611 }
612 
613 /*
614  * Calculate number of records in a bmap btree inode root.
615  */
616 int
617 xfs_bmdr_maxrecs(
618 	int			blocklen,
619 	int			leaf)
620 {
621 	blocklen -= sizeof(xfs_bmdr_block_t);
622 
623 	if (leaf)
624 		return blocklen / sizeof(xfs_bmdr_rec_t);
625 	return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
626 }
627 
628 /*
629  * Change the owner of a btree format fork fo the inode passed in. Change it to
630  * the owner of that is passed in so that we can change owners before or after
631  * we switch forks between inodes. The operation that the caller is doing will
632  * determine whether is needs to change owner before or after the switch.
633  *
634  * For demand paged transactional modification, the fork switch should be done
635  * after reading in all the blocks, modifying them and pinning them in the
636  * transaction. For modification when the buffers are already pinned in memory,
637  * the fork switch can be done before changing the owner as we won't need to
638  * validate the owner until the btree buffers are unpinned and writes can occur
639  * again.
640  *
641  * For recovery based ownership change, there is no transactional context and
642  * so a buffer list must be supplied so that we can record the buffers that we
643  * modified for the caller to issue IO on.
644  */
645 int
646 xfs_bmbt_change_owner(
647 	struct xfs_trans	*tp,
648 	struct xfs_inode	*ip,
649 	int			whichfork,
650 	xfs_ino_t		new_owner,
651 	struct list_head	*buffer_list)
652 {
653 	struct xfs_btree_cur	*cur;
654 	int			error;
655 
656 	ASSERT(tp || buffer_list);
657 	ASSERT(!(tp && buffer_list));
658 	ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
659 
660 	cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
661 	cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
662 
663 	error = xfs_btree_change_owner(cur, new_owner, buffer_list);
664 	xfs_btree_del_cursor(cur, error);
665 	return error;
666 }
667 
668 /* Calculate the bmap btree size for some records. */
669 unsigned long long
670 xfs_bmbt_calc_size(
671 	struct xfs_mount	*mp,
672 	unsigned long long	len)
673 {
674 	return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
675 }
676 
677 int __init
678 xfs_bmbt_init_cur_cache(void)
679 {
680 	xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
681 			xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
682 			0, 0, NULL);
683 
684 	if (!xfs_bmbt_cur_cache)
685 		return -ENOMEM;
686 	return 0;
687 }
688 
689 void
690 xfs_bmbt_destroy_cur_cache(void)
691 {
692 	kmem_cache_destroy(xfs_bmbt_cur_cache);
693 	xfs_bmbt_cur_cache = NULL;
694 }
695