xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap_btree.c (revision 844f5ed5)
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 	/*
218 	 * If we are coming here from something like unwritten extent
219 	 * conversion, there has been no data extent allocation already done, so
220 	 * we have to ensure that we attempt to locate the entire set of bmbt
221 	 * allocations in the same AG, as xfs_bmapi_write() would have reserved.
222 	 */
223 	if (cur->bc_tp->t_highest_agno == NULLAGNUMBER)
224 		args.minleft = xfs_bmapi_minleft(cur->bc_tp, cur->bc_ino.ip,
225 					cur->bc_ino.whichfork);
226 
227 	error = xfs_alloc_vextent_start_ag(&args, be64_to_cpu(start->l));
228 	if (error)
229 		return error;
230 
231 	if (args.fsbno == NULLFSBLOCK && args.minleft) {
232 		/*
233 		 * Could not find an AG with enough free space to satisfy
234 		 * a full btree split.  Try again and if
235 		 * successful activate the lowspace algorithm.
236 		 */
237 		args.minleft = 0;
238 		error = xfs_alloc_vextent_start_ag(&args, 0);
239 		if (error)
240 			return error;
241 		cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
242 	}
243 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
244 		*stat = 0;
245 		return 0;
246 	}
247 
248 	ASSERT(args.len == 1);
249 	cur->bc_ino.allocated++;
250 	cur->bc_ino.ip->i_nblocks++;
251 	xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
252 	xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
253 			XFS_TRANS_DQ_BCOUNT, 1L);
254 
255 	new->l = cpu_to_be64(args.fsbno);
256 
257 	*stat = 1;
258 	return 0;
259 }
260 
261 STATIC int
262 xfs_bmbt_free_block(
263 	struct xfs_btree_cur	*cur,
264 	struct xfs_buf		*bp)
265 {
266 	struct xfs_mount	*mp = cur->bc_mp;
267 	struct xfs_inode	*ip = cur->bc_ino.ip;
268 	struct xfs_trans	*tp = cur->bc_tp;
269 	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
270 	struct xfs_owner_info	oinfo;
271 
272 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
273 	xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
274 	ip->i_nblocks--;
275 
276 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
277 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
278 	return 0;
279 }
280 
281 STATIC int
282 xfs_bmbt_get_minrecs(
283 	struct xfs_btree_cur	*cur,
284 	int			level)
285 {
286 	if (level == cur->bc_nlevels - 1) {
287 		struct xfs_ifork	*ifp;
288 
289 		ifp = xfs_ifork_ptr(cur->bc_ino.ip,
290 				    cur->bc_ino.whichfork);
291 
292 		return xfs_bmbt_maxrecs(cur->bc_mp,
293 					ifp->if_broot_bytes, level == 0) / 2;
294 	}
295 
296 	return cur->bc_mp->m_bmap_dmnr[level != 0];
297 }
298 
299 int
300 xfs_bmbt_get_maxrecs(
301 	struct xfs_btree_cur	*cur,
302 	int			level)
303 {
304 	if (level == cur->bc_nlevels - 1) {
305 		struct xfs_ifork	*ifp;
306 
307 		ifp = xfs_ifork_ptr(cur->bc_ino.ip,
308 				    cur->bc_ino.whichfork);
309 
310 		return xfs_bmbt_maxrecs(cur->bc_mp,
311 					ifp->if_broot_bytes, level == 0);
312 	}
313 
314 	return cur->bc_mp->m_bmap_dmxr[level != 0];
315 
316 }
317 
318 /*
319  * Get the maximum records we could store in the on-disk format.
320  *
321  * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
322  * for the root node this checks the available space in the dinode fork
323  * so that we can resize the in-memory buffer to match it.  After a
324  * resize to the maximum size this function returns the same value
325  * as xfs_bmbt_get_maxrecs for the root node, too.
326  */
327 STATIC int
328 xfs_bmbt_get_dmaxrecs(
329 	struct xfs_btree_cur	*cur,
330 	int			level)
331 {
332 	if (level != cur->bc_nlevels - 1)
333 		return cur->bc_mp->m_bmap_dmxr[level != 0];
334 	return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
335 }
336 
337 STATIC void
338 xfs_bmbt_init_key_from_rec(
339 	union xfs_btree_key		*key,
340 	const union xfs_btree_rec	*rec)
341 {
342 	key->bmbt.br_startoff =
343 		cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
344 }
345 
346 STATIC void
347 xfs_bmbt_init_high_key_from_rec(
348 	union xfs_btree_key		*key,
349 	const union xfs_btree_rec	*rec)
350 {
351 	key->bmbt.br_startoff = cpu_to_be64(
352 			xfs_bmbt_disk_get_startoff(&rec->bmbt) +
353 			xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
354 }
355 
356 STATIC void
357 xfs_bmbt_init_rec_from_cur(
358 	struct xfs_btree_cur	*cur,
359 	union xfs_btree_rec	*rec)
360 {
361 	xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
362 }
363 
364 STATIC void
365 xfs_bmbt_init_ptr_from_cur(
366 	struct xfs_btree_cur	*cur,
367 	union xfs_btree_ptr	*ptr)
368 {
369 	ptr->l = 0;
370 }
371 
372 STATIC int64_t
373 xfs_bmbt_key_diff(
374 	struct xfs_btree_cur		*cur,
375 	const union xfs_btree_key	*key)
376 {
377 	return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
378 				      cur->bc_rec.b.br_startoff;
379 }
380 
381 STATIC int64_t
382 xfs_bmbt_diff_two_keys(
383 	struct xfs_btree_cur		*cur,
384 	const union xfs_btree_key	*k1,
385 	const union xfs_btree_key	*k2)
386 {
387 	uint64_t			a = be64_to_cpu(k1->bmbt.br_startoff);
388 	uint64_t			b = be64_to_cpu(k2->bmbt.br_startoff);
389 
390 	/*
391 	 * Note: This routine previously casted a and b to int64 and subtracted
392 	 * them to generate a result.  This lead to problems if b was the
393 	 * "maximum" key value (all ones) being signed incorrectly, hence this
394 	 * somewhat less efficient version.
395 	 */
396 	if (a > b)
397 		return 1;
398 	if (b > a)
399 		return -1;
400 	return 0;
401 }
402 
403 static xfs_failaddr_t
404 xfs_bmbt_verify(
405 	struct xfs_buf		*bp)
406 {
407 	struct xfs_mount	*mp = bp->b_mount;
408 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
409 	xfs_failaddr_t		fa;
410 	unsigned int		level;
411 
412 	if (!xfs_verify_magic(bp, block->bb_magic))
413 		return __this_address;
414 
415 	if (xfs_has_crc(mp)) {
416 		/*
417 		 * XXX: need a better way of verifying the owner here. Right now
418 		 * just make sure there has been one set.
419 		 */
420 		fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
421 		if (fa)
422 			return fa;
423 	}
424 
425 	/*
426 	 * numrecs and level verification.
427 	 *
428 	 * We don't know what fork we belong to, so just verify that the level
429 	 * is less than the maximum of the two. Later checks will be more
430 	 * precise.
431 	 */
432 	level = be16_to_cpu(block->bb_level);
433 	if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
434 		return __this_address;
435 
436 	return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
437 }
438 
439 static void
440 xfs_bmbt_read_verify(
441 	struct xfs_buf	*bp)
442 {
443 	xfs_failaddr_t	fa;
444 
445 	if (!xfs_btree_lblock_verify_crc(bp))
446 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
447 	else {
448 		fa = xfs_bmbt_verify(bp);
449 		if (fa)
450 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
451 	}
452 
453 	if (bp->b_error)
454 		trace_xfs_btree_corrupt(bp, _RET_IP_);
455 }
456 
457 static void
458 xfs_bmbt_write_verify(
459 	struct xfs_buf	*bp)
460 {
461 	xfs_failaddr_t	fa;
462 
463 	fa = xfs_bmbt_verify(bp);
464 	if (fa) {
465 		trace_xfs_btree_corrupt(bp, _RET_IP_);
466 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
467 		return;
468 	}
469 	xfs_btree_lblock_calc_crc(bp);
470 }
471 
472 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
473 	.name = "xfs_bmbt",
474 	.magic = { cpu_to_be32(XFS_BMAP_MAGIC),
475 		   cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
476 	.verify_read = xfs_bmbt_read_verify,
477 	.verify_write = xfs_bmbt_write_verify,
478 	.verify_struct = xfs_bmbt_verify,
479 };
480 
481 
482 STATIC int
483 xfs_bmbt_keys_inorder(
484 	struct xfs_btree_cur		*cur,
485 	const union xfs_btree_key	*k1,
486 	const union xfs_btree_key	*k2)
487 {
488 	return be64_to_cpu(k1->bmbt.br_startoff) <
489 		be64_to_cpu(k2->bmbt.br_startoff);
490 }
491 
492 STATIC int
493 xfs_bmbt_recs_inorder(
494 	struct xfs_btree_cur		*cur,
495 	const union xfs_btree_rec	*r1,
496 	const union xfs_btree_rec	*r2)
497 {
498 	return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
499 		xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
500 		xfs_bmbt_disk_get_startoff(&r2->bmbt);
501 }
502 
503 static const struct xfs_btree_ops xfs_bmbt_ops = {
504 	.rec_len		= sizeof(xfs_bmbt_rec_t),
505 	.key_len		= sizeof(xfs_bmbt_key_t),
506 
507 	.dup_cursor		= xfs_bmbt_dup_cursor,
508 	.update_cursor		= xfs_bmbt_update_cursor,
509 	.alloc_block		= xfs_bmbt_alloc_block,
510 	.free_block		= xfs_bmbt_free_block,
511 	.get_maxrecs		= xfs_bmbt_get_maxrecs,
512 	.get_minrecs		= xfs_bmbt_get_minrecs,
513 	.get_dmaxrecs		= xfs_bmbt_get_dmaxrecs,
514 	.init_key_from_rec	= xfs_bmbt_init_key_from_rec,
515 	.init_high_key_from_rec	= xfs_bmbt_init_high_key_from_rec,
516 	.init_rec_from_cur	= xfs_bmbt_init_rec_from_cur,
517 	.init_ptr_from_cur	= xfs_bmbt_init_ptr_from_cur,
518 	.key_diff		= xfs_bmbt_key_diff,
519 	.diff_two_keys		= xfs_bmbt_diff_two_keys,
520 	.buf_ops		= &xfs_bmbt_buf_ops,
521 	.keys_inorder		= xfs_bmbt_keys_inorder,
522 	.recs_inorder		= xfs_bmbt_recs_inorder,
523 };
524 
525 /*
526  * Allocate a new bmap btree cursor.
527  */
528 struct xfs_btree_cur *				/* new bmap btree cursor */
529 xfs_bmbt_init_cursor(
530 	struct xfs_mount	*mp,		/* file system mount point */
531 	struct xfs_trans	*tp,		/* transaction pointer */
532 	struct xfs_inode	*ip,		/* inode owning the btree */
533 	int			whichfork)	/* data or attr fork */
534 {
535 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
536 	struct xfs_btree_cur	*cur;
537 	ASSERT(whichfork != XFS_COW_FORK);
538 
539 	cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
540 			mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
541 	cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
542 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
543 
544 	cur->bc_ops = &xfs_bmbt_ops;
545 	cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
546 	if (xfs_has_crc(mp))
547 		cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
548 
549 	cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
550 	cur->bc_ino.ip = ip;
551 	cur->bc_ino.allocated = 0;
552 	cur->bc_ino.flags = 0;
553 	cur->bc_ino.whichfork = whichfork;
554 
555 	return cur;
556 }
557 
558 /* Calculate number of records in a block mapping btree block. */
559 static inline unsigned int
560 xfs_bmbt_block_maxrecs(
561 	unsigned int		blocklen,
562 	bool			leaf)
563 {
564 	if (leaf)
565 		return blocklen / sizeof(xfs_bmbt_rec_t);
566 	return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
567 }
568 
569 /*
570  * Calculate number of records in a bmap btree block.
571  */
572 int
573 xfs_bmbt_maxrecs(
574 	struct xfs_mount	*mp,
575 	int			blocklen,
576 	int			leaf)
577 {
578 	blocklen -= XFS_BMBT_BLOCK_LEN(mp);
579 	return xfs_bmbt_block_maxrecs(blocklen, leaf);
580 }
581 
582 /*
583  * Calculate the maximum possible height of the btree that the on-disk format
584  * supports. This is used for sizing structures large enough to support every
585  * possible configuration of a filesystem that might get mounted.
586  */
587 unsigned int
588 xfs_bmbt_maxlevels_ondisk(void)
589 {
590 	unsigned int		minrecs[2];
591 	unsigned int		blocklen;
592 
593 	blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
594 		       XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
595 
596 	minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
597 	minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
598 
599 	/* One extra level for the inode root. */
600 	return xfs_btree_compute_maxlevels(minrecs,
601 			XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
602 }
603 
604 /*
605  * Calculate number of records in a bmap btree inode root.
606  */
607 int
608 xfs_bmdr_maxrecs(
609 	int			blocklen,
610 	int			leaf)
611 {
612 	blocklen -= sizeof(xfs_bmdr_block_t);
613 
614 	if (leaf)
615 		return blocklen / sizeof(xfs_bmdr_rec_t);
616 	return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
617 }
618 
619 /*
620  * Change the owner of a btree format fork fo the inode passed in. Change it to
621  * the owner of that is passed in so that we can change owners before or after
622  * we switch forks between inodes. The operation that the caller is doing will
623  * determine whether is needs to change owner before or after the switch.
624  *
625  * For demand paged transactional modification, the fork switch should be done
626  * after reading in all the blocks, modifying them and pinning them in the
627  * transaction. For modification when the buffers are already pinned in memory,
628  * the fork switch can be done before changing the owner as we won't need to
629  * validate the owner until the btree buffers are unpinned and writes can occur
630  * again.
631  *
632  * For recovery based ownership change, there is no transactional context and
633  * so a buffer list must be supplied so that we can record the buffers that we
634  * modified for the caller to issue IO on.
635  */
636 int
637 xfs_bmbt_change_owner(
638 	struct xfs_trans	*tp,
639 	struct xfs_inode	*ip,
640 	int			whichfork,
641 	xfs_ino_t		new_owner,
642 	struct list_head	*buffer_list)
643 {
644 	struct xfs_btree_cur	*cur;
645 	int			error;
646 
647 	ASSERT(tp || buffer_list);
648 	ASSERT(!(tp && buffer_list));
649 	ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
650 
651 	cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
652 	cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
653 
654 	error = xfs_btree_change_owner(cur, new_owner, buffer_list);
655 	xfs_btree_del_cursor(cur, error);
656 	return error;
657 }
658 
659 /* Calculate the bmap btree size for some records. */
660 unsigned long long
661 xfs_bmbt_calc_size(
662 	struct xfs_mount	*mp,
663 	unsigned long long	len)
664 {
665 	return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
666 }
667 
668 int __init
669 xfs_bmbt_init_cur_cache(void)
670 {
671 	xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
672 			xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
673 			0, 0, NULL);
674 
675 	if (!xfs_bmbt_cur_cache)
676 		return -ENOMEM;
677 	return 0;
678 }
679 
680 void
681 xfs_bmbt_destroy_cur_cache(void)
682 {
683 	kmem_cache_destroy(xfs_bmbt_cur_cache);
684 	xfs_bmbt_cur_cache = NULL;
685 }
686