xref: /linux/fs/xfs/libxfs/xfs_rtrefcount_btree.c (revision aa301322f72f82f26e4ba0826018d41388ab9896)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2021-2024 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
5  */
6 #include "xfs_platform.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_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_trans.h"
18 #include "xfs_alloc.h"
19 #include "xfs_btree.h"
20 #include "xfs_btree_staging.h"
21 #include "xfs_rtrefcount_btree.h"
22 #include "xfs_refcount.h"
23 #include "xfs_trace.h"
24 #include "xfs_cksum.h"
25 #include "xfs_error.h"
26 #include "xfs_extent_busy.h"
27 #include "xfs_rtgroup.h"
28 #include "xfs_rtbitmap.h"
29 #include "xfs_metafile.h"
30 #include "xfs_health.h"
31 
32 static struct kmem_cache	*xfs_rtrefcountbt_cur_cache;
33 
34 /*
35  * Realtime Reference Count btree.
36  *
37  * This is a btree used to track the owner(s) of a given extent in the realtime
38  * device.  See the comments in xfs_refcount_btree.c for more information.
39  *
40  * This tree is basically the same as the regular refcount btree except that
41  * it's rooted in an inode.
42  */
43 
44 static struct xfs_btree_cur *
45 xfs_rtrefcountbt_dup_cursor(
46 	struct xfs_btree_cur	*cur)
47 {
48 	return xfs_rtrefcountbt_init_cursor(cur->bc_tp, to_rtg(cur->bc_group));
49 }
50 
51 STATIC int
52 xfs_rtrefcountbt_get_minrecs(
53 	struct xfs_btree_cur	*cur,
54 	int			level)
55 {
56 	if (level == cur->bc_nlevels - 1) {
57 		struct xfs_ifork	*ifp = xfs_btree_ifork_ptr(cur);
58 
59 		return xfs_rtrefcountbt_maxrecs(cur->bc_mp, ifp->if_broot_bytes,
60 				level == 0) / 2;
61 	}
62 
63 	return cur->bc_mp->m_rtrefc_mnr[level != 0];
64 }
65 
66 STATIC int
67 xfs_rtrefcountbt_get_maxrecs(
68 	struct xfs_btree_cur	*cur,
69 	int			level)
70 {
71 	if (level == cur->bc_nlevels - 1) {
72 		struct xfs_ifork	*ifp = xfs_btree_ifork_ptr(cur);
73 
74 		return xfs_rtrefcountbt_maxrecs(cur->bc_mp, ifp->if_broot_bytes,
75 				level == 0);
76 	}
77 
78 	return cur->bc_mp->m_rtrefc_mxr[level != 0];
79 }
80 
81 /*
82  * Calculate number of records in a realtime refcount btree inode root.
83  */
84 unsigned int
85 xfs_rtrefcountbt_droot_maxrecs(
86 	unsigned int		blocklen,
87 	bool			leaf)
88 {
89 	blocklen -= sizeof(struct xfs_rtrefcount_root);
90 
91 	if (leaf)
92 		return blocklen / sizeof(struct xfs_refcount_rec);
93 	return blocklen / (2 * sizeof(struct xfs_refcount_key) +
94 			sizeof(xfs_rtrefcount_ptr_t));
95 }
96 
97 /*
98  * Get the maximum records we could store in the on-disk format.
99  *
100  * For non-root nodes this is equivalent to xfs_rtrefcountbt_get_maxrecs, but
101  * for the root node this checks the available space in the dinode fork so that
102  * we can resize the in-memory buffer to match it.  After a resize to the
103  * maximum size this function returns the same value as
104  * xfs_rtrefcountbt_get_maxrecs for the root node, too.
105  */
106 STATIC int
107 xfs_rtrefcountbt_get_dmaxrecs(
108 	struct xfs_btree_cur	*cur,
109 	int			level)
110 {
111 	if (level != cur->bc_nlevels - 1)
112 		return cur->bc_mp->m_rtrefc_mxr[level != 0];
113 	return xfs_rtrefcountbt_droot_maxrecs(cur->bc_ino.forksize, level == 0);
114 }
115 
116 STATIC void
117 xfs_rtrefcountbt_init_key_from_rec(
118 	union xfs_btree_key		*key,
119 	const union xfs_btree_rec	*rec)
120 {
121 	key->refc.rc_startblock = rec->refc.rc_startblock;
122 }
123 
124 STATIC void
125 xfs_rtrefcountbt_init_high_key_from_rec(
126 	union xfs_btree_key		*key,
127 	const union xfs_btree_rec	*rec)
128 {
129 	__u32				x;
130 
131 	x = be32_to_cpu(rec->refc.rc_startblock);
132 	x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
133 	key->refc.rc_startblock = cpu_to_be32(x);
134 }
135 
136 STATIC void
137 xfs_rtrefcountbt_init_rec_from_cur(
138 	struct xfs_btree_cur	*cur,
139 	union xfs_btree_rec	*rec)
140 {
141 	const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
142 	uint32_t		start;
143 
144 	start = xfs_refcount_encode_startblock(irec->rc_startblock,
145 			irec->rc_domain);
146 	rec->refc.rc_startblock = cpu_to_be32(start);
147 	rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
148 	rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
149 }
150 
151 STATIC void
152 xfs_rtrefcountbt_init_ptr_from_cur(
153 	struct xfs_btree_cur	*cur,
154 	union xfs_btree_ptr	*ptr)
155 {
156 	ptr->l = 0;
157 }
158 
159 STATIC int
160 xfs_rtrefcountbt_cmp_key_with_cur(
161 	struct xfs_btree_cur		*cur,
162 	const union xfs_btree_key	*key)
163 {
164 	const struct xfs_refcount_key	*kp = &key->refc;
165 	const struct xfs_refcount_irec	*irec = &cur->bc_rec.rc;
166 	uint32_t			start;
167 
168 	start = xfs_refcount_encode_startblock(irec->rc_startblock,
169 			irec->rc_domain);
170 	return cmp_int(be32_to_cpu(kp->rc_startblock), start);
171 }
172 
173 STATIC int
174 xfs_rtrefcountbt_cmp_two_keys(
175 	struct xfs_btree_cur		*cur,
176 	const union xfs_btree_key	*k1,
177 	const union xfs_btree_key	*k2,
178 	const union xfs_btree_key	*mask)
179 {
180 	ASSERT(!mask || mask->refc.rc_startblock);
181 
182 	return cmp_int(be32_to_cpu(k1->refc.rc_startblock),
183 		       be32_to_cpu(k2->refc.rc_startblock));
184 }
185 
186 static xfs_failaddr_t
187 xfs_rtrefcountbt_verify(
188 	struct xfs_buf		*bp)
189 {
190 	struct xfs_mount	*mp = bp->b_target->bt_mount;
191 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
192 	xfs_failaddr_t		fa;
193 	int			level;
194 
195 	if (!xfs_verify_magic(bp, block->bb_magic))
196 		return __this_address;
197 
198 	if (!xfs_has_reflink(mp))
199 		return __this_address;
200 	fa = xfs_btree_fsblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
201 	if (fa)
202 		return fa;
203 	level = be16_to_cpu(block->bb_level);
204 	if (level >= mp->m_rtrefc_maxlevels)
205 		return __this_address;
206 
207 	return xfs_btree_fsblock_verify(bp, mp->m_rtrefc_mxr[level != 0]);
208 }
209 
210 static void
211 xfs_rtrefcountbt_read_verify(
212 	struct xfs_buf	*bp)
213 {
214 	xfs_failaddr_t	fa;
215 
216 	if (!xfs_btree_fsblock_verify_crc(bp))
217 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
218 	else {
219 		fa = xfs_rtrefcountbt_verify(bp);
220 		if (fa)
221 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
222 	}
223 
224 	if (bp->b_error)
225 		trace_xfs_btree_corrupt(bp, _RET_IP_);
226 }
227 
228 static void
229 xfs_rtrefcountbt_write_verify(
230 	struct xfs_buf	*bp)
231 {
232 	xfs_failaddr_t	fa;
233 
234 	fa = xfs_rtrefcountbt_verify(bp);
235 	if (fa) {
236 		trace_xfs_btree_corrupt(bp, _RET_IP_);
237 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
238 		return;
239 	}
240 	xfs_btree_fsblock_calc_crc(bp);
241 
242 }
243 
244 const struct xfs_buf_ops xfs_rtrefcountbt_buf_ops = {
245 	.name			= "xfs_rtrefcountbt",
246 	.magic			= { 0, cpu_to_be32(XFS_RTREFC_CRC_MAGIC) },
247 	.verify_read		= xfs_rtrefcountbt_read_verify,
248 	.verify_write		= xfs_rtrefcountbt_write_verify,
249 	.verify_struct		= xfs_rtrefcountbt_verify,
250 };
251 
252 STATIC int
253 xfs_rtrefcountbt_keys_inorder(
254 	struct xfs_btree_cur		*cur,
255 	const union xfs_btree_key	*k1,
256 	const union xfs_btree_key	*k2)
257 {
258 	return be32_to_cpu(k1->refc.rc_startblock) <
259 	       be32_to_cpu(k2->refc.rc_startblock);
260 }
261 
262 STATIC int
263 xfs_rtrefcountbt_recs_inorder(
264 	struct xfs_btree_cur		*cur,
265 	const union xfs_btree_rec	*r1,
266 	const union xfs_btree_rec	*r2)
267 {
268 	return  be32_to_cpu(r1->refc.rc_startblock) +
269 		be32_to_cpu(r1->refc.rc_blockcount) <=
270 		be32_to_cpu(r2->refc.rc_startblock);
271 }
272 
273 STATIC enum xbtree_key_contig
274 xfs_rtrefcountbt_keys_contiguous(
275 	struct xfs_btree_cur		*cur,
276 	const union xfs_btree_key	*key1,
277 	const union xfs_btree_key	*key2,
278 	const union xfs_btree_key	*mask)
279 {
280 	ASSERT(!mask || mask->refc.rc_startblock);
281 
282 	return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
283 				 be32_to_cpu(key2->refc.rc_startblock));
284 }
285 
286 static inline void
287 xfs_rtrefcountbt_move_ptrs(
288 	struct xfs_mount	*mp,
289 	struct xfs_btree_block	*broot,
290 	short			old_size,
291 	size_t			new_size,
292 	unsigned int		numrecs)
293 {
294 	void			*dptr;
295 	void			*sptr;
296 
297 	sptr = xfs_rtrefcount_broot_ptr_addr(mp, broot, 1, old_size);
298 	dptr = xfs_rtrefcount_broot_ptr_addr(mp, broot, 1, new_size);
299 	memmove(dptr, sptr, numrecs * sizeof(xfs_rtrefcount_ptr_t));
300 }
301 
302 static struct xfs_btree_block *
303 xfs_rtrefcountbt_broot_realloc(
304 	struct xfs_btree_cur	*cur,
305 	unsigned int		new_numrecs)
306 {
307 	struct xfs_mount	*mp = cur->bc_mp;
308 	struct xfs_ifork	*ifp = xfs_btree_ifork_ptr(cur);
309 	struct xfs_btree_block	*broot;
310 	unsigned int		new_size;
311 	unsigned int		old_size = ifp->if_broot_bytes;
312 	const unsigned int	level = cur->bc_nlevels - 1;
313 
314 	new_size = xfs_rtrefcount_broot_space_calc(mp, level, new_numrecs);
315 
316 	/* Handle the nop case quietly. */
317 	if (new_size == old_size)
318 		return ifp->if_broot;
319 
320 	if (new_size > old_size) {
321 		unsigned int	old_numrecs;
322 
323 		/*
324 		 * If there wasn't any memory allocated before, just allocate
325 		 * it now and get out.
326 		 */
327 		if (old_size == 0)
328 			return xfs_broot_realloc(ifp, new_size);
329 
330 		/*
331 		 * If there is already an existing if_broot, then we need to
332 		 * realloc it and possibly move the node block pointers because
333 		 * those are not butted up against the btree block header.
334 		 */
335 		old_numrecs = xfs_rtrefcountbt_maxrecs(mp, old_size, level);
336 		broot = xfs_broot_realloc(ifp, new_size);
337 		if (level > 0)
338 			xfs_rtrefcountbt_move_ptrs(mp, broot, old_size,
339 					new_size, old_numrecs);
340 		goto out_broot;
341 	}
342 
343 	/*
344 	 * We're reducing numrecs.  If we're going all the way to zero, just
345 	 * free the block.
346 	 */
347 	ASSERT(ifp->if_broot != NULL && old_size > 0);
348 	if (new_size == 0)
349 		return xfs_broot_realloc(ifp, 0);
350 
351 	/*
352 	 * Shrink the btree root by possibly moving the rtrmapbt pointers,
353 	 * since they are not butted up against the btree block header.  Then
354 	 * reallocate broot.
355 	 */
356 	if (level > 0)
357 		xfs_rtrefcountbt_move_ptrs(mp, ifp->if_broot, old_size,
358 				new_size, new_numrecs);
359 	broot = xfs_broot_realloc(ifp, new_size);
360 
361 out_broot:
362 	ASSERT(xfs_rtrefcount_droot_space(broot) <=
363 	       xfs_inode_fork_size(cur->bc_ino.ip, cur->bc_ino.whichfork));
364 	return broot;
365 }
366 
367 const struct xfs_btree_ops xfs_rtrefcountbt_ops = {
368 	.name			= "rtrefcount",
369 	.type			= XFS_BTREE_TYPE_INODE,
370 	.geom_flags		= XFS_BTGEO_IROOT_RECORDS,
371 
372 	.rec_len		= sizeof(struct xfs_refcount_rec),
373 	.key_len		= sizeof(struct xfs_refcount_key),
374 	.ptr_len		= XFS_BTREE_LONG_PTR_LEN,
375 
376 	.lru_refs		= XFS_REFC_BTREE_REF,
377 	.statoff		= XFS_STATS_CALC_INDEX(xs_rtrefcbt_2),
378 	.sick_mask		= XFS_SICK_RG_REFCNTBT,
379 
380 	.dup_cursor		= xfs_rtrefcountbt_dup_cursor,
381 	.alloc_block		= xfs_btree_alloc_metafile_block,
382 	.free_block		= xfs_btree_free_metafile_block,
383 	.get_minrecs		= xfs_rtrefcountbt_get_minrecs,
384 	.get_maxrecs		= xfs_rtrefcountbt_get_maxrecs,
385 	.get_dmaxrecs		= xfs_rtrefcountbt_get_dmaxrecs,
386 	.init_key_from_rec	= xfs_rtrefcountbt_init_key_from_rec,
387 	.init_high_key_from_rec	= xfs_rtrefcountbt_init_high_key_from_rec,
388 	.init_rec_from_cur	= xfs_rtrefcountbt_init_rec_from_cur,
389 	.init_ptr_from_cur	= xfs_rtrefcountbt_init_ptr_from_cur,
390 	.cmp_key_with_cur	= xfs_rtrefcountbt_cmp_key_with_cur,
391 	.buf_ops		= &xfs_rtrefcountbt_buf_ops,
392 	.cmp_two_keys		= xfs_rtrefcountbt_cmp_two_keys,
393 	.keys_inorder		= xfs_rtrefcountbt_keys_inorder,
394 	.recs_inorder		= xfs_rtrefcountbt_recs_inorder,
395 	.keys_contiguous	= xfs_rtrefcountbt_keys_contiguous,
396 	.broot_realloc		= xfs_rtrefcountbt_broot_realloc,
397 };
398 
399 /* Allocate a new rt refcount btree cursor. */
400 struct xfs_btree_cur *
401 xfs_rtrefcountbt_init_cursor(
402 	struct xfs_trans	*tp,
403 	struct xfs_rtgroup	*rtg)
404 {
405 	struct xfs_inode	*ip = rtg_refcount(rtg);
406 	struct xfs_mount	*mp = rtg_mount(rtg);
407 	struct xfs_btree_cur	*cur;
408 
409 	xfs_assert_ilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL);
410 
411 	cur = xfs_btree_alloc_cursor(mp, tp, &xfs_rtrefcountbt_ops,
412 			mp->m_rtrefc_maxlevels, xfs_rtrefcountbt_cur_cache);
413 
414 	cur->bc_ino.ip = ip;
415 	cur->bc_refc.nr_ops = 0;
416 	cur->bc_refc.shape_changes = 0;
417 	cur->bc_group = xfs_group_hold(rtg_group(rtg));
418 	cur->bc_nlevels = be16_to_cpu(ip->i_df.if_broot->bb_level) + 1;
419 	cur->bc_ino.forksize = xfs_inode_fork_size(ip, XFS_DATA_FORK);
420 	cur->bc_ino.whichfork = XFS_DATA_FORK;
421 	return cur;
422 }
423 
424 /*
425  * Install a new rt reverse mapping btree root.  Caller is responsible for
426  * invalidating and freeing the old btree blocks.
427  */
428 void
429 xfs_rtrefcountbt_commit_staged_btree(
430 	struct xfs_btree_cur	*cur,
431 	struct xfs_trans	*tp)
432 {
433 	struct xbtree_ifakeroot	*ifake = cur->bc_ino.ifake;
434 	struct xfs_ifork	*ifp;
435 	int			flags = XFS_ILOG_CORE | XFS_ILOG_DBROOT;
436 
437 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
438 	ASSERT(ifake->if_fork->if_format == XFS_DINODE_FMT_META_BTREE);
439 
440 	/*
441 	 * Free any resources hanging off the real fork, then shallow-copy the
442 	 * staging fork's contents into the real fork to transfer everything
443 	 * we just built.
444 	 */
445 	ifp = xfs_ifork_ptr(cur->bc_ino.ip, XFS_DATA_FORK);
446 	xfs_idestroy_fork(ifp);
447 	memcpy(ifp, ifake->if_fork, sizeof(struct xfs_ifork));
448 
449 	cur->bc_ino.ip->i_projid = cur->bc_group->xg_gno;
450 	xfs_trans_log_inode(tp, cur->bc_ino.ip, flags);
451 	xfs_btree_commit_ifakeroot(cur, tp, XFS_DATA_FORK);
452 }
453 
454 /* Calculate number of records in a realtime refcount btree block. */
455 static inline unsigned int
456 xfs_rtrefcountbt_block_maxrecs(
457 	unsigned int		blocklen,
458 	bool			leaf)
459 {
460 
461 	if (leaf)
462 		return blocklen / sizeof(struct xfs_refcount_rec);
463 	return blocklen / (sizeof(struct xfs_refcount_key) +
464 			   sizeof(xfs_rtrefcount_ptr_t));
465 }
466 
467 /*
468  * Calculate number of records in an refcount btree block.
469  */
470 unsigned int
471 xfs_rtrefcountbt_maxrecs(
472 	struct xfs_mount	*mp,
473 	unsigned int		blocklen,
474 	bool			leaf)
475 {
476 	blocklen -= XFS_RTREFCOUNT_BLOCK_LEN;
477 	return xfs_rtrefcountbt_block_maxrecs(blocklen, leaf);
478 }
479 
480 /* Compute the max possible height for realtime refcount btrees. */
481 unsigned int
482 xfs_rtrefcountbt_maxlevels_ondisk(void)
483 {
484 	unsigned int		minrecs[2];
485 	unsigned int		blocklen;
486 
487 	blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_LBLOCK_CRC_LEN;
488 
489 	minrecs[0] = xfs_rtrefcountbt_block_maxrecs(blocklen, true) / 2;
490 	minrecs[1] = xfs_rtrefcountbt_block_maxrecs(blocklen, false) / 2;
491 
492 	/*
493 	 * We need at most one record for every block in an rt group, and
494 	 * one extra level for the inode root.
495 	 */
496 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_RGBLOCKS) + 1;
497 }
498 
499 int __init
500 xfs_rtrefcountbt_init_cur_cache(void)
501 {
502 	xfs_rtrefcountbt_cur_cache = kmem_cache_create("xfs_rtrefcountbt_cur",
503 			xfs_btree_cur_sizeof(
504 					xfs_rtrefcountbt_maxlevels_ondisk()),
505 			0, 0, NULL);
506 
507 	if (!xfs_rtrefcountbt_cur_cache)
508 		return -ENOMEM;
509 	return 0;
510 }
511 
512 void
513 xfs_rtrefcountbt_destroy_cur_cache(void)
514 {
515 	kmem_cache_destroy(xfs_rtrefcountbt_cur_cache);
516 	xfs_rtrefcountbt_cur_cache = NULL;
517 }
518 
519 /* Compute the maximum height of a realtime refcount btree. */
520 void
521 xfs_rtrefcountbt_compute_maxlevels(
522 	struct xfs_mount	*mp)
523 {
524 	unsigned int		d_maxlevels, r_maxlevels;
525 
526 	if (!xfs_has_rtreflink(mp)) {
527 		mp->m_rtrefc_maxlevels = 0;
528 		return;
529 	}
530 
531 	/*
532 	 * The realtime refcountbt lives on the data device, which means that
533 	 * its maximum height is constrained by the size of the data device and
534 	 * the height required to store one refcount record for each rtextent
535 	 * in an rt group.
536 	 */
537 	d_maxlevels = xfs_btree_space_to_height(mp->m_rtrefc_mnr,
538 				mp->m_sb.sb_dblocks);
539 	r_maxlevels = xfs_btree_compute_maxlevels(mp->m_rtrefc_mnr,
540 				mp->m_sb.sb_rgextents);
541 
542 	/* Add one level to handle the inode root level. */
543 	mp->m_rtrefc_maxlevels = min(d_maxlevels, r_maxlevels) + 1;
544 }
545 
546 /* Calculate the rtrefcount btree size for some records. */
547 unsigned long long
548 xfs_rtrefcountbt_calc_size(
549 	struct xfs_mount	*mp,
550 	unsigned long long	len)
551 {
552 	return xfs_btree_calc_size(mp->m_rtrefc_mnr, len);
553 }
554 
555 /*
556  * Calculate the maximum refcount btree size.
557  */
558 static unsigned long long
559 xfs_rtrefcountbt_max_size(
560 	struct xfs_mount	*mp,
561 	xfs_rtblock_t		rtblocks)
562 {
563 	/* Bail out if we're uninitialized, which can happen in mkfs. */
564 	if (mp->m_rtrefc_mxr[0] == 0)
565 		return 0;
566 
567 	return xfs_rtrefcountbt_calc_size(mp, rtblocks);
568 }
569 
570 /*
571  * Figure out how many blocks to reserve and how many are used by this btree.
572  * We need enough space to hold one record for every rt extent in the rtgroup.
573  */
574 xfs_filblks_t
575 xfs_rtrefcountbt_calc_reserves(
576 	struct xfs_mount	*mp)
577 {
578 	if (!xfs_has_rtreflink(mp))
579 		return 0;
580 
581 	return xfs_rtrefcountbt_max_size(mp, mp->m_sb.sb_rgextents);
582 }
583 
584 /*
585  * Convert on-disk form of btree root to in-memory form.
586  */
587 STATIC void
588 xfs_rtrefcountbt_from_disk(
589 	struct xfs_inode		*ip,
590 	struct xfs_rtrefcount_root	*dblock,
591 	int				dblocklen,
592 	struct xfs_btree_block		*rblock)
593 {
594 	struct xfs_mount		*mp = ip->i_mount;
595 	struct xfs_refcount_key	*fkp;
596 	__be64				*fpp;
597 	struct xfs_refcount_key	*tkp;
598 	__be64				*tpp;
599 	struct xfs_refcount_rec	*frp;
600 	struct xfs_refcount_rec	*trp;
601 	unsigned int			numrecs;
602 	unsigned int			maxrecs;
603 	unsigned int			rblocklen;
604 
605 	rblocklen = xfs_rtrefcount_broot_space(mp, dblock);
606 
607 	xfs_btree_init_block(mp, rblock, &xfs_rtrefcountbt_ops, 0, 0,
608 			I_INO(ip));
609 
610 	rblock->bb_level = dblock->bb_level;
611 	rblock->bb_numrecs = dblock->bb_numrecs;
612 
613 	if (be16_to_cpu(rblock->bb_level) > 0) {
614 		maxrecs = xfs_rtrefcountbt_droot_maxrecs(dblocklen, false);
615 		fkp = xfs_rtrefcount_droot_key_addr(dblock, 1);
616 		tkp = xfs_rtrefcount_key_addr(rblock, 1);
617 		fpp = xfs_rtrefcount_droot_ptr_addr(dblock, 1, maxrecs);
618 		tpp = xfs_rtrefcount_broot_ptr_addr(mp, rblock, 1, rblocklen);
619 		numrecs = be16_to_cpu(dblock->bb_numrecs);
620 		memcpy(tkp, fkp, 2 * sizeof(*fkp) * numrecs);
621 		memcpy(tpp, fpp, sizeof(*fpp) * numrecs);
622 	} else {
623 		frp = xfs_rtrefcount_droot_rec_addr(dblock, 1);
624 		trp = xfs_rtrefcount_rec_addr(rblock, 1);
625 		numrecs = be16_to_cpu(dblock->bb_numrecs);
626 		memcpy(trp, frp, sizeof(*frp) * numrecs);
627 	}
628 }
629 
630 /* Load a realtime reference count btree root in from disk. */
631 int
632 xfs_iformat_rtrefcount(
633 	struct xfs_inode	*ip,
634 	struct xfs_dinode	*dip)
635 {
636 	struct xfs_mount	*mp = ip->i_mount;
637 	struct xfs_rtrefcount_root *dfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);
638 	struct xfs_btree_block	*broot;
639 	unsigned int		numrecs;
640 	unsigned int		level;
641 	int			dsize;
642 
643 	/*
644 	 * growfs must create the rtrefcount inodes before adding a realtime
645 	 * volume to the filesystem, so we cannot use the rtrefcount predicate
646 	 * here.
647 	 */
648 	if (!xfs_has_reflink(ip->i_mount)) {
649 		xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
650 		return -EFSCORRUPTED;
651 	}
652 
653 	dsize = XFS_DFORK_SIZE(dip, mp, XFS_DATA_FORK);
654 	numrecs = be16_to_cpu(dfp->bb_numrecs);
655 	level = be16_to_cpu(dfp->bb_level);
656 
657 	if (level >= mp->m_rtrefc_maxlevels ||
658 	    xfs_rtrefcount_droot_space_calc(level, numrecs) > dsize) {
659 		xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
660 		return -EFSCORRUPTED;
661 	}
662 
663 	broot = xfs_broot_alloc(xfs_ifork_ptr(ip, XFS_DATA_FORK),
664 			xfs_rtrefcount_broot_space_calc(mp, level, numrecs));
665 	if (broot)
666 		xfs_rtrefcountbt_from_disk(ip, dfp, dsize, broot);
667 	return 0;
668 }
669 
670 /*
671  * Convert in-memory form of btree root to on-disk form.
672  */
673 void
674 xfs_rtrefcountbt_to_disk(
675 	struct xfs_mount		*mp,
676 	struct xfs_btree_block		*rblock,
677 	int				rblocklen,
678 	struct xfs_rtrefcount_root	*dblock,
679 	int				dblocklen)
680 {
681 	struct xfs_refcount_key	*fkp;
682 	__be64				*fpp;
683 	struct xfs_refcount_key	*tkp;
684 	__be64				*tpp;
685 	struct xfs_refcount_rec	*frp;
686 	struct xfs_refcount_rec	*trp;
687 	unsigned int			maxrecs;
688 	unsigned int			numrecs;
689 
690 	ASSERT(rblock->bb_magic == cpu_to_be32(XFS_RTREFC_CRC_MAGIC));
691 	ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid));
692 	ASSERT(rblock->bb_u.l.bb_blkno == cpu_to_be64(XFS_BUF_DADDR_NULL));
693 	ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
694 	ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
695 
696 	dblock->bb_level = rblock->bb_level;
697 	dblock->bb_numrecs = rblock->bb_numrecs;
698 
699 	if (be16_to_cpu(rblock->bb_level) > 0) {
700 		maxrecs = xfs_rtrefcountbt_droot_maxrecs(dblocklen, false);
701 		fkp = xfs_rtrefcount_key_addr(rblock, 1);
702 		tkp = xfs_rtrefcount_droot_key_addr(dblock, 1);
703 		fpp = xfs_rtrefcount_broot_ptr_addr(mp, rblock, 1, rblocklen);
704 		tpp = xfs_rtrefcount_droot_ptr_addr(dblock, 1, maxrecs);
705 		numrecs = be16_to_cpu(rblock->bb_numrecs);
706 		memcpy(tkp, fkp, 2 * sizeof(*fkp) * numrecs);
707 		memcpy(tpp, fpp, sizeof(*fpp) * numrecs);
708 	} else {
709 		frp = xfs_rtrefcount_rec_addr(rblock, 1);
710 		trp = xfs_rtrefcount_droot_rec_addr(dblock, 1);
711 		numrecs = be16_to_cpu(rblock->bb_numrecs);
712 		memcpy(trp, frp, sizeof(*frp) * numrecs);
713 	}
714 }
715 
716 /* Flush a realtime reference count btree root out to disk. */
717 void
718 xfs_iflush_rtrefcount(
719 	struct xfs_inode	*ip,
720 	struct xfs_dinode	*dip)
721 {
722 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
723 	struct xfs_rtrefcount_root *dfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);
724 
725 	ASSERT(ifp->if_broot != NULL);
726 	ASSERT(ifp->if_broot_bytes > 0);
727 	ASSERT(xfs_rtrefcount_droot_space(ifp->if_broot) <=
728 			xfs_inode_fork_size(ip, XFS_DATA_FORK));
729 	xfs_rtrefcountbt_to_disk(ip->i_mount, ifp->if_broot,
730 			ifp->if_broot_bytes, dfp,
731 			XFS_DFORK_SIZE(dip, ip->i_mount, XFS_DATA_FORK));
732 }
733 
734 /*
735  * Create a realtime refcount btree inode.
736  */
737 int
738 xfs_rtrefcountbt_create(
739 	struct xfs_rtgroup	*rtg,
740 	struct xfs_inode	*ip,
741 	struct xfs_trans	*tp,
742 	bool			init)
743 {
744 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
745 	struct xfs_mount	*mp = ip->i_mount;
746 	struct xfs_btree_block	*broot;
747 
748 	ifp->if_format = XFS_DINODE_FMT_META_BTREE;
749 	ASSERT(ifp->if_broot_bytes == 0);
750 	ASSERT(ifp->if_bytes == 0);
751 
752 	/* Initialize the empty incore btree root. */
753 	broot = xfs_broot_realloc(ifp,
754 			xfs_rtrefcount_broot_space_calc(mp, 0, 0));
755 	if (broot)
756 		xfs_btree_init_block(mp, broot, &xfs_rtrefcountbt_ops, 0, 0,
757 				I_INO(ip));
758 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE | XFS_ILOG_DBROOT);
759 	return 0;
760 }
761