xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap_btree.c (revision 74c36a86)
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 #include "xfs_ag.h"
25 
26 static struct kmem_cache	*xfs_bmbt_cur_cache;
27 
28 /*
29  * Convert on-disk form of btree root to in-memory form.
30  */
31 void
32 xfs_bmdr_to_bmbt(
33 	struct xfs_inode	*ip,
34 	xfs_bmdr_block_t	*dblock,
35 	int			dblocklen,
36 	struct xfs_btree_block	*rblock,
37 	int			rblocklen)
38 {
39 	struct xfs_mount	*mp = ip->i_mount;
40 	int			dmxr;
41 	xfs_bmbt_key_t		*fkp;
42 	__be64			*fpp;
43 	xfs_bmbt_key_t		*tkp;
44 	__be64			*tpp;
45 
46 	xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
47 				 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
48 				 XFS_BTREE_LONG_PTRS);
49 	rblock->bb_level = dblock->bb_level;
50 	ASSERT(be16_to_cpu(rblock->bb_level) > 0);
51 	rblock->bb_numrecs = dblock->bb_numrecs;
52 	dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
53 	fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
54 	tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
55 	fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
56 	tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
57 	dmxr = be16_to_cpu(dblock->bb_numrecs);
58 	memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
59 	memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
60 }
61 
62 void
63 xfs_bmbt_disk_get_all(
64 	const struct xfs_bmbt_rec *rec,
65 	struct xfs_bmbt_irec	*irec)
66 {
67 	uint64_t		l0 = get_unaligned_be64(&rec->l0);
68 	uint64_t		l1 = get_unaligned_be64(&rec->l1);
69 
70 	irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
71 	irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
72 	irec->br_blockcount = l1 & xfs_mask64lo(21);
73 	if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
74 		irec->br_state = XFS_EXT_UNWRITTEN;
75 	else
76 		irec->br_state = XFS_EXT_NORM;
77 }
78 
79 /*
80  * Extract the blockcount field from an on disk bmap extent record.
81  */
82 xfs_filblks_t
83 xfs_bmbt_disk_get_blockcount(
84 	const struct xfs_bmbt_rec	*r)
85 {
86 	return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
87 }
88 
89 /*
90  * Extract the startoff field from a disk format bmap extent record.
91  */
92 xfs_fileoff_t
93 xfs_bmbt_disk_get_startoff(
94 	const struct xfs_bmbt_rec	*r)
95 {
96 	return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
97 		 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
98 }
99 
100 /*
101  * Set all the fields in a bmap extent record from the uncompressed form.
102  */
103 void
104 xfs_bmbt_disk_set_all(
105 	struct xfs_bmbt_rec	*r,
106 	struct xfs_bmbt_irec	*s)
107 {
108 	int			extent_flag = (s->br_state != XFS_EXT_NORM);
109 
110 	ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
111 	ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
112 	ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
113 	ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
114 
115 	put_unaligned_be64(
116 		((xfs_bmbt_rec_base_t)extent_flag << 63) |
117 		 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
118 		 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
119 	put_unaligned_be64(
120 		((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
121 		 ((xfs_bmbt_rec_base_t)s->br_blockcount &
122 		  (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
123 }
124 
125 /*
126  * Convert in-memory form of btree root to on-disk form.
127  */
128 void
129 xfs_bmbt_to_bmdr(
130 	struct xfs_mount	*mp,
131 	struct xfs_btree_block	*rblock,
132 	int			rblocklen,
133 	xfs_bmdr_block_t	*dblock,
134 	int			dblocklen)
135 {
136 	int			dmxr;
137 	xfs_bmbt_key_t		*fkp;
138 	__be64			*fpp;
139 	xfs_bmbt_key_t		*tkp;
140 	__be64			*tpp;
141 
142 	if (xfs_has_crc(mp)) {
143 		ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
144 		ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
145 		       &mp->m_sb.sb_meta_uuid));
146 		ASSERT(rblock->bb_u.l.bb_blkno ==
147 		       cpu_to_be64(XFS_BUF_DADDR_NULL));
148 	} else
149 		ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
150 	ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
151 	ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
152 	ASSERT(rblock->bb_level != 0);
153 	dblock->bb_level = rblock->bb_level;
154 	dblock->bb_numrecs = rblock->bb_numrecs;
155 	dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
156 	fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
157 	tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
158 	fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
159 	tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
160 	dmxr = be16_to_cpu(dblock->bb_numrecs);
161 	memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
162 	memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
163 }
164 
165 STATIC struct xfs_btree_cur *
166 xfs_bmbt_dup_cursor(
167 	struct xfs_btree_cur	*cur)
168 {
169 	struct xfs_btree_cur	*new;
170 
171 	new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
172 			cur->bc_ino.ip, cur->bc_ino.whichfork);
173 
174 	/*
175 	 * Copy the firstblock, dfops, and flags values,
176 	 * since init cursor doesn't get them.
177 	 */
178 	new->bc_ino.flags = cur->bc_ino.flags;
179 
180 	return new;
181 }
182 
183 STATIC void
184 xfs_bmbt_update_cursor(
185 	struct xfs_btree_cur	*src,
186 	struct xfs_btree_cur	*dst)
187 {
188 	ASSERT((dst->bc_tp->t_highest_agno != NULLAGNUMBER) ||
189 	       (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
190 
191 	dst->bc_ino.allocated += src->bc_ino.allocated;
192 	dst->bc_tp->t_highest_agno = src->bc_tp->t_highest_agno;
193 
194 	src->bc_ino.allocated = 0;
195 }
196 
197 STATIC int
198 xfs_bmbt_alloc_block(
199 	struct xfs_btree_cur		*cur,
200 	const union xfs_btree_ptr	*start,
201 	union xfs_btree_ptr		*new,
202 	int				*stat)
203 {
204 	struct xfs_alloc_arg	args;
205 	int			error;
206 
207 	memset(&args, 0, sizeof(args));
208 	args.tp = cur->bc_tp;
209 	args.mp = cur->bc_mp;
210 	xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
211 			cur->bc_ino.whichfork);
212 	args.minlen = args.maxlen = args.prod = 1;
213 	args.wasdel = cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL;
214 	if (!args.wasdel && args.tp->t_blk_res == 0)
215 		return -ENOSPC;
216 
217 	args.fsbno = be64_to_cpu(start->l);
218 	args.type = XFS_ALLOCTYPE_START_BNO;
219 
220 	/*
221 	 * If we are coming here from something like unwritten extent
222 	 * conversion, there has been no data extent allocation already done, so
223 	 * we have to ensure that we attempt to locate the entire set of bmbt
224 	 * allocations in the same AG, as xfs_bmapi_write() would have reserved.
225 	 */
226 	if (cur->bc_tp->t_highest_agno == NULLAGNUMBER)
227 		args.minleft = xfs_bmapi_minleft(cur->bc_tp, cur->bc_ino.ip,
228 					cur->bc_ino.whichfork);
229 
230 	error = xfs_alloc_vextent(&args);
231 	if (error)
232 		return error;
233 
234 	if (args.fsbno == NULLFSBLOCK && args.minleft) {
235 		/*
236 		 * Could not find an AG with enough free space to satisfy
237 		 * a full btree split.  Try again and if
238 		 * successful activate the lowspace algorithm.
239 		 */
240 		args.fsbno = 0;
241 		args.minleft = 0;
242 		args.type = XFS_ALLOCTYPE_START_BNO;
243 		error = xfs_alloc_vextent(&args);
244 		if (error)
245 			return error;
246 		cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
247 	}
248 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
249 		*stat = 0;
250 		return 0;
251 	}
252 
253 	ASSERT(args.len == 1);
254 	cur->bc_ino.allocated++;
255 	cur->bc_ino.ip->i_nblocks++;
256 	xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
257 	xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
258 			XFS_TRANS_DQ_BCOUNT, 1L);
259 
260 	new->l = cpu_to_be64(args.fsbno);
261 
262 	*stat = 1;
263 	return 0;
264 }
265 
266 STATIC int
267 xfs_bmbt_free_block(
268 	struct xfs_btree_cur	*cur,
269 	struct xfs_buf		*bp)
270 {
271 	struct xfs_mount	*mp = cur->bc_mp;
272 	struct xfs_inode	*ip = cur->bc_ino.ip;
273 	struct xfs_trans	*tp = cur->bc_tp;
274 	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
275 	struct xfs_owner_info	oinfo;
276 
277 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
278 	xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
279 	ip->i_nblocks--;
280 
281 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
282 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
283 	return 0;
284 }
285 
286 STATIC int
287 xfs_bmbt_get_minrecs(
288 	struct xfs_btree_cur	*cur,
289 	int			level)
290 {
291 	if (level == cur->bc_nlevels - 1) {
292 		struct xfs_ifork	*ifp;
293 
294 		ifp = xfs_ifork_ptr(cur->bc_ino.ip,
295 				    cur->bc_ino.whichfork);
296 
297 		return xfs_bmbt_maxrecs(cur->bc_mp,
298 					ifp->if_broot_bytes, level == 0) / 2;
299 	}
300 
301 	return cur->bc_mp->m_bmap_dmnr[level != 0];
302 }
303 
304 int
305 xfs_bmbt_get_maxrecs(
306 	struct xfs_btree_cur	*cur,
307 	int			level)
308 {
309 	if (level == cur->bc_nlevels - 1) {
310 		struct xfs_ifork	*ifp;
311 
312 		ifp = xfs_ifork_ptr(cur->bc_ino.ip,
313 				    cur->bc_ino.whichfork);
314 
315 		return xfs_bmbt_maxrecs(cur->bc_mp,
316 					ifp->if_broot_bytes, level == 0);
317 	}
318 
319 	return cur->bc_mp->m_bmap_dmxr[level != 0];
320 
321 }
322 
323 /*
324  * Get the maximum records we could store in the on-disk format.
325  *
326  * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
327  * for the root node this checks the available space in the dinode fork
328  * so that we can resize the in-memory buffer to match it.  After a
329  * resize to the maximum size this function returns the same value
330  * as xfs_bmbt_get_maxrecs for the root node, too.
331  */
332 STATIC int
333 xfs_bmbt_get_dmaxrecs(
334 	struct xfs_btree_cur	*cur,
335 	int			level)
336 {
337 	if (level != cur->bc_nlevels - 1)
338 		return cur->bc_mp->m_bmap_dmxr[level != 0];
339 	return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
340 }
341 
342 STATIC void
343 xfs_bmbt_init_key_from_rec(
344 	union xfs_btree_key		*key,
345 	const union xfs_btree_rec	*rec)
346 {
347 	key->bmbt.br_startoff =
348 		cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
349 }
350 
351 STATIC void
352 xfs_bmbt_init_high_key_from_rec(
353 	union xfs_btree_key		*key,
354 	const union xfs_btree_rec	*rec)
355 {
356 	key->bmbt.br_startoff = cpu_to_be64(
357 			xfs_bmbt_disk_get_startoff(&rec->bmbt) +
358 			xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
359 }
360 
361 STATIC void
362 xfs_bmbt_init_rec_from_cur(
363 	struct xfs_btree_cur	*cur,
364 	union xfs_btree_rec	*rec)
365 {
366 	xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
367 }
368 
369 STATIC void
370 xfs_bmbt_init_ptr_from_cur(
371 	struct xfs_btree_cur	*cur,
372 	union xfs_btree_ptr	*ptr)
373 {
374 	ptr->l = 0;
375 }
376 
377 STATIC int64_t
378 xfs_bmbt_key_diff(
379 	struct xfs_btree_cur		*cur,
380 	const union xfs_btree_key	*key)
381 {
382 	return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
383 				      cur->bc_rec.b.br_startoff;
384 }
385 
386 STATIC int64_t
387 xfs_bmbt_diff_two_keys(
388 	struct xfs_btree_cur		*cur,
389 	const union xfs_btree_key	*k1,
390 	const union xfs_btree_key	*k2)
391 {
392 	uint64_t			a = be64_to_cpu(k1->bmbt.br_startoff);
393 	uint64_t			b = be64_to_cpu(k2->bmbt.br_startoff);
394 
395 	/*
396 	 * Note: This routine previously casted a and b to int64 and subtracted
397 	 * them to generate a result.  This lead to problems if b was the
398 	 * "maximum" key value (all ones) being signed incorrectly, hence this
399 	 * somewhat less efficient version.
400 	 */
401 	if (a > b)
402 		return 1;
403 	if (b > a)
404 		return -1;
405 	return 0;
406 }
407 
408 static xfs_failaddr_t
409 xfs_bmbt_verify(
410 	struct xfs_buf		*bp)
411 {
412 	struct xfs_mount	*mp = bp->b_mount;
413 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
414 	xfs_failaddr_t		fa;
415 	unsigned int		level;
416 
417 	if (!xfs_verify_magic(bp, block->bb_magic))
418 		return __this_address;
419 
420 	if (xfs_has_crc(mp)) {
421 		/*
422 		 * XXX: need a better way of verifying the owner here. Right now
423 		 * just make sure there has been one set.
424 		 */
425 		fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
426 		if (fa)
427 			return fa;
428 	}
429 
430 	/*
431 	 * numrecs and level verification.
432 	 *
433 	 * We don't know what fork we belong to, so just verify that the level
434 	 * is less than the maximum of the two. Later checks will be more
435 	 * precise.
436 	 */
437 	level = be16_to_cpu(block->bb_level);
438 	if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
439 		return __this_address;
440 
441 	return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
442 }
443 
444 static void
445 xfs_bmbt_read_verify(
446 	struct xfs_buf	*bp)
447 {
448 	xfs_failaddr_t	fa;
449 
450 	if (!xfs_btree_lblock_verify_crc(bp))
451 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
452 	else {
453 		fa = xfs_bmbt_verify(bp);
454 		if (fa)
455 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
456 	}
457 
458 	if (bp->b_error)
459 		trace_xfs_btree_corrupt(bp, _RET_IP_);
460 }
461 
462 static void
463 xfs_bmbt_write_verify(
464 	struct xfs_buf	*bp)
465 {
466 	xfs_failaddr_t	fa;
467 
468 	fa = xfs_bmbt_verify(bp);
469 	if (fa) {
470 		trace_xfs_btree_corrupt(bp, _RET_IP_);
471 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
472 		return;
473 	}
474 	xfs_btree_lblock_calc_crc(bp);
475 }
476 
477 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
478 	.name = "xfs_bmbt",
479 	.magic = { cpu_to_be32(XFS_BMAP_MAGIC),
480 		   cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
481 	.verify_read = xfs_bmbt_read_verify,
482 	.verify_write = xfs_bmbt_write_verify,
483 	.verify_struct = xfs_bmbt_verify,
484 };
485 
486 
487 STATIC int
488 xfs_bmbt_keys_inorder(
489 	struct xfs_btree_cur		*cur,
490 	const union xfs_btree_key	*k1,
491 	const union xfs_btree_key	*k2)
492 {
493 	return be64_to_cpu(k1->bmbt.br_startoff) <
494 		be64_to_cpu(k2->bmbt.br_startoff);
495 }
496 
497 STATIC int
498 xfs_bmbt_recs_inorder(
499 	struct xfs_btree_cur		*cur,
500 	const union xfs_btree_rec	*r1,
501 	const union xfs_btree_rec	*r2)
502 {
503 	return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
504 		xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
505 		xfs_bmbt_disk_get_startoff(&r2->bmbt);
506 }
507 
508 static const struct xfs_btree_ops xfs_bmbt_ops = {
509 	.rec_len		= sizeof(xfs_bmbt_rec_t),
510 	.key_len		= sizeof(xfs_bmbt_key_t),
511 
512 	.dup_cursor		= xfs_bmbt_dup_cursor,
513 	.update_cursor		= xfs_bmbt_update_cursor,
514 	.alloc_block		= xfs_bmbt_alloc_block,
515 	.free_block		= xfs_bmbt_free_block,
516 	.get_maxrecs		= xfs_bmbt_get_maxrecs,
517 	.get_minrecs		= xfs_bmbt_get_minrecs,
518 	.get_dmaxrecs		= xfs_bmbt_get_dmaxrecs,
519 	.init_key_from_rec	= xfs_bmbt_init_key_from_rec,
520 	.init_high_key_from_rec	= xfs_bmbt_init_high_key_from_rec,
521 	.init_rec_from_cur	= xfs_bmbt_init_rec_from_cur,
522 	.init_ptr_from_cur	= xfs_bmbt_init_ptr_from_cur,
523 	.key_diff		= xfs_bmbt_key_diff,
524 	.diff_two_keys		= xfs_bmbt_diff_two_keys,
525 	.buf_ops		= &xfs_bmbt_buf_ops,
526 	.keys_inorder		= xfs_bmbt_keys_inorder,
527 	.recs_inorder		= xfs_bmbt_recs_inorder,
528 };
529 
530 /*
531  * Allocate a new bmap btree cursor.
532  */
533 struct xfs_btree_cur *				/* new bmap btree cursor */
534 xfs_bmbt_init_cursor(
535 	struct xfs_mount	*mp,		/* file system mount point */
536 	struct xfs_trans	*tp,		/* transaction pointer */
537 	struct xfs_inode	*ip,		/* inode owning the btree */
538 	int			whichfork)	/* data or attr fork */
539 {
540 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
541 	struct xfs_btree_cur	*cur;
542 	ASSERT(whichfork != XFS_COW_FORK);
543 
544 	cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
545 			mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
546 	cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
547 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
548 
549 	cur->bc_ops = &xfs_bmbt_ops;
550 	cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
551 	if (xfs_has_crc(mp))
552 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
553 
554 	cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
555 	cur->bc_ino.ip = ip;
556 	cur->bc_ino.allocated = 0;
557 	cur->bc_ino.flags = 0;
558 	cur->bc_ino.whichfork = whichfork;
559 
560 	return cur;
561 }
562 
563 /* Calculate number of records in a block mapping btree block. */
564 static inline unsigned int
565 xfs_bmbt_block_maxrecs(
566 	unsigned int		blocklen,
567 	bool			leaf)
568 {
569 	if (leaf)
570 		return blocklen / sizeof(xfs_bmbt_rec_t);
571 	return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
572 }
573 
574 /*
575  * Calculate number of records in a bmap btree block.
576  */
577 int
578 xfs_bmbt_maxrecs(
579 	struct xfs_mount	*mp,
580 	int			blocklen,
581 	int			leaf)
582 {
583 	blocklen -= XFS_BMBT_BLOCK_LEN(mp);
584 	return xfs_bmbt_block_maxrecs(blocklen, leaf);
585 }
586 
587 /*
588  * Calculate the maximum possible height of the btree that the on-disk format
589  * supports. This is used for sizing structures large enough to support every
590  * possible configuration of a filesystem that might get mounted.
591  */
592 unsigned int
593 xfs_bmbt_maxlevels_ondisk(void)
594 {
595 	unsigned int		minrecs[2];
596 	unsigned int		blocklen;
597 
598 	blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
599 		       XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
600 
601 	minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
602 	minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
603 
604 	/* One extra level for the inode root. */
605 	return xfs_btree_compute_maxlevels(minrecs,
606 			XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
607 }
608 
609 /*
610  * Calculate number of records in a bmap btree inode root.
611  */
612 int
613 xfs_bmdr_maxrecs(
614 	int			blocklen,
615 	int			leaf)
616 {
617 	blocklen -= sizeof(xfs_bmdr_block_t);
618 
619 	if (leaf)
620 		return blocklen / sizeof(xfs_bmdr_rec_t);
621 	return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
622 }
623 
624 /*
625  * Change the owner of a btree format fork fo the inode passed in. Change it to
626  * the owner of that is passed in so that we can change owners before or after
627  * we switch forks between inodes. The operation that the caller is doing will
628  * determine whether is needs to change owner before or after the switch.
629  *
630  * For demand paged transactional modification, the fork switch should be done
631  * after reading in all the blocks, modifying them and pinning them in the
632  * transaction. For modification when the buffers are already pinned in memory,
633  * the fork switch can be done before changing the owner as we won't need to
634  * validate the owner until the btree buffers are unpinned and writes can occur
635  * again.
636  *
637  * For recovery based ownership change, there is no transactional context and
638  * so a buffer list must be supplied so that we can record the buffers that we
639  * modified for the caller to issue IO on.
640  */
641 int
642 xfs_bmbt_change_owner(
643 	struct xfs_trans	*tp,
644 	struct xfs_inode	*ip,
645 	int			whichfork,
646 	xfs_ino_t		new_owner,
647 	struct list_head	*buffer_list)
648 {
649 	struct xfs_btree_cur	*cur;
650 	int			error;
651 
652 	ASSERT(tp || buffer_list);
653 	ASSERT(!(tp && buffer_list));
654 	ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
655 
656 	cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
657 	cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
658 
659 	error = xfs_btree_change_owner(cur, new_owner, buffer_list);
660 	xfs_btree_del_cursor(cur, error);
661 	return error;
662 }
663 
664 /* Calculate the bmap btree size for some records. */
665 unsigned long long
666 xfs_bmbt_calc_size(
667 	struct xfs_mount	*mp,
668 	unsigned long long	len)
669 {
670 	return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
671 }
672 
673 int __init
674 xfs_bmbt_init_cur_cache(void)
675 {
676 	xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
677 			xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
678 			0, 0, NULL);
679 
680 	if (!xfs_bmbt_cur_cache)
681 		return -ENOMEM;
682 	return 0;
683 }
684 
685 void
686 xfs_bmbt_destroy_cur_cache(void)
687 {
688 	kmem_cache_destroy(xfs_bmbt_cur_cache);
689 	xfs_bmbt_cur_cache = NULL;
690 }
691