xref: /linux/fs/xfs/libxfs/xfs_rmap_btree.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2014 Red Hat, 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_mount.h"
13 #include "xfs_trans.h"
14 #include "xfs_alloc.h"
15 #include "xfs_btree.h"
16 #include "xfs_btree_staging.h"
17 #include "xfs_rmap.h"
18 #include "xfs_rmap_btree.h"
19 #include "xfs_health.h"
20 #include "xfs_trace.h"
21 #include "xfs_error.h"
22 #include "xfs_extent_busy.h"
23 #include "xfs_ag.h"
24 #include "xfs_ag_resv.h"
25 #include "xfs_buf_mem.h"
26 #include "xfs_btree_mem.h"
27 
28 static struct kmem_cache	*xfs_rmapbt_cur_cache;
29 
30 /*
31  * Reverse map btree.
32  *
33  * This is a per-ag tree used to track the owner(s) of a given extent. With
34  * reflink it is possible for there to be multiple owners, which is a departure
35  * from classic XFS. Owner records for data extents are inserted when the
36  * extent is mapped and removed when an extent is unmapped.  Owner records for
37  * all other block types (i.e. metadata) are inserted when an extent is
38  * allocated and removed when an extent is freed. There can only be one owner
39  * of a metadata extent, usually an inode or some other metadata structure like
40  * an AG btree.
41  *
42  * The rmap btree is part of the free space management, so blocks for the tree
43  * are sourced from the agfl. Hence we need transaction reservation support for
44  * this tree so that the freelist is always large enough. This also impacts on
45  * the minimum space we need to leave free in the AG.
46  *
47  * The tree is ordered by [ag block, owner, offset]. This is a large key size,
48  * but it is the only way to enforce unique keys when a block can be owned by
49  * multiple files at any offset. There's no need to order/search by extent
50  * size for online updating/management of the tree. It is intended that most
51  * reverse lookups will be to find the owner(s) of a particular block, or to
52  * try to recover tree and file data from corrupt primary metadata.
53  */
54 
55 static struct xfs_btree_cur *
56 xfs_rmapbt_dup_cursor(
57 	struct xfs_btree_cur	*cur)
58 {
59 	return xfs_rmapbt_init_cursor(cur->bc_mp, cur->bc_tp,
60 				cur->bc_ag.agbp, cur->bc_ag.pag);
61 }
62 
63 STATIC void
64 xfs_rmapbt_set_root(
65 	struct xfs_btree_cur		*cur,
66 	const union xfs_btree_ptr	*ptr,
67 	int				inc)
68 {
69 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
70 	struct xfs_agf		*agf = agbp->b_addr;
71 
72 	ASSERT(ptr->s != 0);
73 
74 	agf->agf_rmap_root = ptr->s;
75 	be32_add_cpu(&agf->agf_rmap_level, inc);
76 	cur->bc_ag.pag->pagf_rmap_level += inc;
77 
78 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
79 }
80 
81 STATIC int
82 xfs_rmapbt_alloc_block(
83 	struct xfs_btree_cur		*cur,
84 	const union xfs_btree_ptr	*start,
85 	union xfs_btree_ptr		*new,
86 	int				*stat)
87 {
88 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
89 	struct xfs_agf		*agf = agbp->b_addr;
90 	struct xfs_perag	*pag = cur->bc_ag.pag;
91 	struct xfs_alloc_arg    args = { .len = 1 };
92 	int			error;
93 	xfs_agblock_t		bno;
94 
95 	/* Allocate the new block from the freelist. If we can't, give up.  */
96 	error = xfs_alloc_get_freelist(pag, cur->bc_tp, cur->bc_ag.agbp,
97 				       &bno, 1);
98 	if (error)
99 		return error;
100 	if (bno == NULLAGBLOCK) {
101 		*stat = 0;
102 		return 0;
103 	}
104 
105 	xfs_extent_busy_reuse(cur->bc_mp, pag, bno, 1, false);
106 
107 	new->s = cpu_to_be32(bno);
108 	be32_add_cpu(&agf->agf_rmap_blocks, 1);
109 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_RMAP_BLOCKS);
110 
111 	/*
112 	 * Since rmapbt blocks are sourced from the AGFL, they are allocated one
113 	 * at a time and the reservation updates don't require a transaction.
114 	 */
115 	xfs_ag_resv_alloc_extent(pag, XFS_AG_RESV_RMAPBT, &args);
116 
117 	*stat = 1;
118 	return 0;
119 }
120 
121 STATIC int
122 xfs_rmapbt_free_block(
123 	struct xfs_btree_cur	*cur,
124 	struct xfs_buf		*bp)
125 {
126 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
127 	struct xfs_agf		*agf = agbp->b_addr;
128 	struct xfs_perag	*pag = cur->bc_ag.pag;
129 	xfs_agblock_t		bno;
130 	int			error;
131 
132 	bno = xfs_daddr_to_agbno(cur->bc_mp, xfs_buf_daddr(bp));
133 	be32_add_cpu(&agf->agf_rmap_blocks, -1);
134 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_RMAP_BLOCKS);
135 	error = xfs_alloc_put_freelist(pag, cur->bc_tp, agbp, NULL, bno, 1);
136 	if (error)
137 		return error;
138 
139 	xfs_extent_busy_insert(cur->bc_tp, pag, bno, 1,
140 			      XFS_EXTENT_BUSY_SKIP_DISCARD);
141 
142 	xfs_ag_resv_free_extent(pag, XFS_AG_RESV_RMAPBT, NULL, 1);
143 	return 0;
144 }
145 
146 STATIC int
147 xfs_rmapbt_get_minrecs(
148 	struct xfs_btree_cur	*cur,
149 	int			level)
150 {
151 	return cur->bc_mp->m_rmap_mnr[level != 0];
152 }
153 
154 STATIC int
155 xfs_rmapbt_get_maxrecs(
156 	struct xfs_btree_cur	*cur,
157 	int			level)
158 {
159 	return cur->bc_mp->m_rmap_mxr[level != 0];
160 }
161 
162 /*
163  * Convert the ondisk record's offset field into the ondisk key's offset field.
164  * Fork and bmbt are significant parts of the rmap record key, but written
165  * status is merely a record attribute.
166  */
167 static inline __be64 ondisk_rec_offset_to_key(const union xfs_btree_rec *rec)
168 {
169 	return rec->rmap.rm_offset & ~cpu_to_be64(XFS_RMAP_OFF_UNWRITTEN);
170 }
171 
172 STATIC void
173 xfs_rmapbt_init_key_from_rec(
174 	union xfs_btree_key		*key,
175 	const union xfs_btree_rec	*rec)
176 {
177 	key->rmap.rm_startblock = rec->rmap.rm_startblock;
178 	key->rmap.rm_owner = rec->rmap.rm_owner;
179 	key->rmap.rm_offset = ondisk_rec_offset_to_key(rec);
180 }
181 
182 /*
183  * The high key for a reverse mapping record can be computed by shifting
184  * the startblock and offset to the highest value that would still map
185  * to that record.  In practice this means that we add blockcount-1 to
186  * the startblock for all records, and if the record is for a data/attr
187  * fork mapping, we add blockcount-1 to the offset too.
188  */
189 STATIC void
190 xfs_rmapbt_init_high_key_from_rec(
191 	union xfs_btree_key		*key,
192 	const union xfs_btree_rec	*rec)
193 {
194 	uint64_t			off;
195 	int				adj;
196 
197 	adj = be32_to_cpu(rec->rmap.rm_blockcount) - 1;
198 
199 	key->rmap.rm_startblock = rec->rmap.rm_startblock;
200 	be32_add_cpu(&key->rmap.rm_startblock, adj);
201 	key->rmap.rm_owner = rec->rmap.rm_owner;
202 	key->rmap.rm_offset = ondisk_rec_offset_to_key(rec);
203 	if (XFS_RMAP_NON_INODE_OWNER(be64_to_cpu(rec->rmap.rm_owner)) ||
204 	    XFS_RMAP_IS_BMBT_BLOCK(be64_to_cpu(rec->rmap.rm_offset)))
205 		return;
206 	off = be64_to_cpu(key->rmap.rm_offset);
207 	off = (XFS_RMAP_OFF(off) + adj) | (off & ~XFS_RMAP_OFF_MASK);
208 	key->rmap.rm_offset = cpu_to_be64(off);
209 }
210 
211 STATIC void
212 xfs_rmapbt_init_rec_from_cur(
213 	struct xfs_btree_cur	*cur,
214 	union xfs_btree_rec	*rec)
215 {
216 	rec->rmap.rm_startblock = cpu_to_be32(cur->bc_rec.r.rm_startblock);
217 	rec->rmap.rm_blockcount = cpu_to_be32(cur->bc_rec.r.rm_blockcount);
218 	rec->rmap.rm_owner = cpu_to_be64(cur->bc_rec.r.rm_owner);
219 	rec->rmap.rm_offset = cpu_to_be64(
220 			xfs_rmap_irec_offset_pack(&cur->bc_rec.r));
221 }
222 
223 STATIC void
224 xfs_rmapbt_init_ptr_from_cur(
225 	struct xfs_btree_cur	*cur,
226 	union xfs_btree_ptr	*ptr)
227 {
228 	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
229 
230 	ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
231 
232 	ptr->s = agf->agf_rmap_root;
233 }
234 
235 /*
236  * Mask the appropriate parts of the ondisk key field for a key comparison.
237  * Fork and bmbt are significant parts of the rmap record key, but written
238  * status is merely a record attribute.
239  */
240 static inline uint64_t offset_keymask(uint64_t offset)
241 {
242 	return offset & ~XFS_RMAP_OFF_UNWRITTEN;
243 }
244 
245 STATIC int64_t
246 xfs_rmapbt_key_diff(
247 	struct xfs_btree_cur		*cur,
248 	const union xfs_btree_key	*key)
249 {
250 	struct xfs_rmap_irec		*rec = &cur->bc_rec.r;
251 	const struct xfs_rmap_key	*kp = &key->rmap;
252 	__u64				x, y;
253 	int64_t				d;
254 
255 	d = (int64_t)be32_to_cpu(kp->rm_startblock) - rec->rm_startblock;
256 	if (d)
257 		return d;
258 
259 	x = be64_to_cpu(kp->rm_owner);
260 	y = rec->rm_owner;
261 	if (x > y)
262 		return 1;
263 	else if (y > x)
264 		return -1;
265 
266 	x = offset_keymask(be64_to_cpu(kp->rm_offset));
267 	y = offset_keymask(xfs_rmap_irec_offset_pack(rec));
268 	if (x > y)
269 		return 1;
270 	else if (y > x)
271 		return -1;
272 	return 0;
273 }
274 
275 STATIC int64_t
276 xfs_rmapbt_diff_two_keys(
277 	struct xfs_btree_cur		*cur,
278 	const union xfs_btree_key	*k1,
279 	const union xfs_btree_key	*k2,
280 	const union xfs_btree_key	*mask)
281 {
282 	const struct xfs_rmap_key	*kp1 = &k1->rmap;
283 	const struct xfs_rmap_key	*kp2 = &k2->rmap;
284 	int64_t				d;
285 	__u64				x, y;
286 
287 	/* Doesn't make sense to mask off the physical space part */
288 	ASSERT(!mask || mask->rmap.rm_startblock);
289 
290 	d = (int64_t)be32_to_cpu(kp1->rm_startblock) -
291 		     be32_to_cpu(kp2->rm_startblock);
292 	if (d)
293 		return d;
294 
295 	if (!mask || mask->rmap.rm_owner) {
296 		x = be64_to_cpu(kp1->rm_owner);
297 		y = be64_to_cpu(kp2->rm_owner);
298 		if (x > y)
299 			return 1;
300 		else if (y > x)
301 			return -1;
302 	}
303 
304 	if (!mask || mask->rmap.rm_offset) {
305 		/* Doesn't make sense to allow offset but not owner */
306 		ASSERT(!mask || mask->rmap.rm_owner);
307 
308 		x = offset_keymask(be64_to_cpu(kp1->rm_offset));
309 		y = offset_keymask(be64_to_cpu(kp2->rm_offset));
310 		if (x > y)
311 			return 1;
312 		else if (y > x)
313 			return -1;
314 	}
315 
316 	return 0;
317 }
318 
319 static xfs_failaddr_t
320 xfs_rmapbt_verify(
321 	struct xfs_buf		*bp)
322 {
323 	struct xfs_mount	*mp = bp->b_mount;
324 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
325 	struct xfs_perag	*pag = bp->b_pag;
326 	xfs_failaddr_t		fa;
327 	unsigned int		level;
328 
329 	/*
330 	 * magic number and level verification
331 	 *
332 	 * During growfs operations, we can't verify the exact level or owner as
333 	 * the perag is not fully initialised and hence not attached to the
334 	 * buffer.  In this case, check against the maximum tree depth.
335 	 *
336 	 * Similarly, during log recovery we will have a perag structure
337 	 * attached, but the agf information will not yet have been initialised
338 	 * from the on disk AGF. Again, we can only check against maximum limits
339 	 * in this case.
340 	 */
341 	if (!xfs_verify_magic(bp, block->bb_magic))
342 		return __this_address;
343 
344 	if (!xfs_has_rmapbt(mp))
345 		return __this_address;
346 	fa = xfs_btree_agblock_v5hdr_verify(bp);
347 	if (fa)
348 		return fa;
349 
350 	level = be16_to_cpu(block->bb_level);
351 	if (pag && xfs_perag_initialised_agf(pag)) {
352 		unsigned int	maxlevel = pag->pagf_rmap_level;
353 
354 #ifdef CONFIG_XFS_ONLINE_REPAIR
355 		/*
356 		 * Online repair could be rewriting the free space btrees, so
357 		 * we'll validate against the larger of either tree while this
358 		 * is going on.
359 		 */
360 		maxlevel = max_t(unsigned int, maxlevel,
361 				pag->pagf_repair_rmap_level);
362 #endif
363 		if (level >= maxlevel)
364 			return __this_address;
365 	} else if (level >= mp->m_rmap_maxlevels)
366 		return __this_address;
367 
368 	return xfs_btree_agblock_verify(bp, mp->m_rmap_mxr[level != 0]);
369 }
370 
371 static void
372 xfs_rmapbt_read_verify(
373 	struct xfs_buf	*bp)
374 {
375 	xfs_failaddr_t	fa;
376 
377 	if (!xfs_btree_agblock_verify_crc(bp))
378 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
379 	else {
380 		fa = xfs_rmapbt_verify(bp);
381 		if (fa)
382 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
383 	}
384 
385 	if (bp->b_error)
386 		trace_xfs_btree_corrupt(bp, _RET_IP_);
387 }
388 
389 static void
390 xfs_rmapbt_write_verify(
391 	struct xfs_buf	*bp)
392 {
393 	xfs_failaddr_t	fa;
394 
395 	fa = xfs_rmapbt_verify(bp);
396 	if (fa) {
397 		trace_xfs_btree_corrupt(bp, _RET_IP_);
398 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
399 		return;
400 	}
401 	xfs_btree_agblock_calc_crc(bp);
402 
403 }
404 
405 const struct xfs_buf_ops xfs_rmapbt_buf_ops = {
406 	.name			= "xfs_rmapbt",
407 	.magic			= { 0, cpu_to_be32(XFS_RMAP_CRC_MAGIC) },
408 	.verify_read		= xfs_rmapbt_read_verify,
409 	.verify_write		= xfs_rmapbt_write_verify,
410 	.verify_struct		= xfs_rmapbt_verify,
411 };
412 
413 STATIC int
414 xfs_rmapbt_keys_inorder(
415 	struct xfs_btree_cur		*cur,
416 	const union xfs_btree_key	*k1,
417 	const union xfs_btree_key	*k2)
418 {
419 	uint32_t		x;
420 	uint32_t		y;
421 	uint64_t		a;
422 	uint64_t		b;
423 
424 	x = be32_to_cpu(k1->rmap.rm_startblock);
425 	y = be32_to_cpu(k2->rmap.rm_startblock);
426 	if (x < y)
427 		return 1;
428 	else if (x > y)
429 		return 0;
430 	a = be64_to_cpu(k1->rmap.rm_owner);
431 	b = be64_to_cpu(k2->rmap.rm_owner);
432 	if (a < b)
433 		return 1;
434 	else if (a > b)
435 		return 0;
436 	a = offset_keymask(be64_to_cpu(k1->rmap.rm_offset));
437 	b = offset_keymask(be64_to_cpu(k2->rmap.rm_offset));
438 	if (a <= b)
439 		return 1;
440 	return 0;
441 }
442 
443 STATIC int
444 xfs_rmapbt_recs_inorder(
445 	struct xfs_btree_cur		*cur,
446 	const union xfs_btree_rec	*r1,
447 	const union xfs_btree_rec	*r2)
448 {
449 	uint32_t		x;
450 	uint32_t		y;
451 	uint64_t		a;
452 	uint64_t		b;
453 
454 	x = be32_to_cpu(r1->rmap.rm_startblock);
455 	y = be32_to_cpu(r2->rmap.rm_startblock);
456 	if (x < y)
457 		return 1;
458 	else if (x > y)
459 		return 0;
460 	a = be64_to_cpu(r1->rmap.rm_owner);
461 	b = be64_to_cpu(r2->rmap.rm_owner);
462 	if (a < b)
463 		return 1;
464 	else if (a > b)
465 		return 0;
466 	a = offset_keymask(be64_to_cpu(r1->rmap.rm_offset));
467 	b = offset_keymask(be64_to_cpu(r2->rmap.rm_offset));
468 	if (a <= b)
469 		return 1;
470 	return 0;
471 }
472 
473 STATIC enum xbtree_key_contig
474 xfs_rmapbt_keys_contiguous(
475 	struct xfs_btree_cur		*cur,
476 	const union xfs_btree_key	*key1,
477 	const union xfs_btree_key	*key2,
478 	const union xfs_btree_key	*mask)
479 {
480 	ASSERT(!mask || mask->rmap.rm_startblock);
481 
482 	/*
483 	 * We only support checking contiguity of the physical space component.
484 	 * If any callers ever need more specificity than that, they'll have to
485 	 * implement it here.
486 	 */
487 	ASSERT(!mask || (!mask->rmap.rm_owner && !mask->rmap.rm_offset));
488 
489 	return xbtree_key_contig(be32_to_cpu(key1->rmap.rm_startblock),
490 				 be32_to_cpu(key2->rmap.rm_startblock));
491 }
492 
493 const struct xfs_btree_ops xfs_rmapbt_ops = {
494 	.name			= "rmap",
495 	.type			= XFS_BTREE_TYPE_AG,
496 	.geom_flags		= XFS_BTGEO_OVERLAPPING,
497 
498 	.rec_len		= sizeof(struct xfs_rmap_rec),
499 	/* Overlapping btree; 2 keys per pointer. */
500 	.key_len		= 2 * sizeof(struct xfs_rmap_key),
501 	.ptr_len		= XFS_BTREE_SHORT_PTR_LEN,
502 
503 	.lru_refs		= XFS_RMAP_BTREE_REF,
504 	.statoff		= XFS_STATS_CALC_INDEX(xs_rmap_2),
505 	.sick_mask		= XFS_SICK_AG_RMAPBT,
506 
507 	.dup_cursor		= xfs_rmapbt_dup_cursor,
508 	.set_root		= xfs_rmapbt_set_root,
509 	.alloc_block		= xfs_rmapbt_alloc_block,
510 	.free_block		= xfs_rmapbt_free_block,
511 	.get_minrecs		= xfs_rmapbt_get_minrecs,
512 	.get_maxrecs		= xfs_rmapbt_get_maxrecs,
513 	.init_key_from_rec	= xfs_rmapbt_init_key_from_rec,
514 	.init_high_key_from_rec	= xfs_rmapbt_init_high_key_from_rec,
515 	.init_rec_from_cur	= xfs_rmapbt_init_rec_from_cur,
516 	.init_ptr_from_cur	= xfs_rmapbt_init_ptr_from_cur,
517 	.key_diff		= xfs_rmapbt_key_diff,
518 	.buf_ops		= &xfs_rmapbt_buf_ops,
519 	.diff_two_keys		= xfs_rmapbt_diff_two_keys,
520 	.keys_inorder		= xfs_rmapbt_keys_inorder,
521 	.recs_inorder		= xfs_rmapbt_recs_inorder,
522 	.keys_contiguous	= xfs_rmapbt_keys_contiguous,
523 };
524 
525 /*
526  * Create a new reverse mapping btree cursor.
527  *
528  * For staging cursors tp and agbp are NULL.
529  */
530 struct xfs_btree_cur *
531 xfs_rmapbt_init_cursor(
532 	struct xfs_mount	*mp,
533 	struct xfs_trans	*tp,
534 	struct xfs_buf		*agbp,
535 	struct xfs_perag	*pag)
536 {
537 	struct xfs_btree_cur	*cur;
538 
539 	cur = xfs_btree_alloc_cursor(mp, tp, &xfs_rmapbt_ops,
540 			mp->m_rmap_maxlevels, xfs_rmapbt_cur_cache);
541 	cur->bc_ag.pag = xfs_perag_hold(pag);
542 	cur->bc_ag.agbp = agbp;
543 	if (agbp) {
544 		struct xfs_agf		*agf = agbp->b_addr;
545 
546 		cur->bc_nlevels = be32_to_cpu(agf->agf_rmap_level);
547 	}
548 	return cur;
549 }
550 
551 #ifdef CONFIG_XFS_BTREE_IN_MEM
552 static inline unsigned int
553 xfs_rmapbt_mem_block_maxrecs(
554 	unsigned int		blocklen,
555 	bool			leaf)
556 {
557 	if (leaf)
558 		return blocklen / sizeof(struct xfs_rmap_rec);
559 	return blocklen /
560 		(2 * sizeof(struct xfs_rmap_key) + sizeof(__be64));
561 }
562 
563 /*
564  * Validate an in-memory rmap btree block.  Callers are allowed to generate an
565  * in-memory btree even if the ondisk feature is not enabled.
566  */
567 static xfs_failaddr_t
568 xfs_rmapbt_mem_verify(
569 	struct xfs_buf		*bp)
570 {
571 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
572 	xfs_failaddr_t		fa;
573 	unsigned int		level;
574 	unsigned int		maxrecs;
575 
576 	if (!xfs_verify_magic(bp, block->bb_magic))
577 		return __this_address;
578 
579 	fa = xfs_btree_fsblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
580 	if (fa)
581 		return fa;
582 
583 	level = be16_to_cpu(block->bb_level);
584 	if (level >= xfs_rmapbt_maxlevels_ondisk())
585 		return __this_address;
586 
587 	maxrecs = xfs_rmapbt_mem_block_maxrecs(
588 			XFBNO_BLOCKSIZE - XFS_BTREE_LBLOCK_CRC_LEN, level == 0);
589 	return xfs_btree_memblock_verify(bp, maxrecs);
590 }
591 
592 static void
593 xfs_rmapbt_mem_rw_verify(
594 	struct xfs_buf	*bp)
595 {
596 	xfs_failaddr_t	fa = xfs_rmapbt_mem_verify(bp);
597 
598 	if (fa)
599 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
600 }
601 
602 /* skip crc checks on in-memory btrees to save time */
603 static const struct xfs_buf_ops xfs_rmapbt_mem_buf_ops = {
604 	.name			= "xfs_rmapbt_mem",
605 	.magic			= { 0, cpu_to_be32(XFS_RMAP_CRC_MAGIC) },
606 	.verify_read		= xfs_rmapbt_mem_rw_verify,
607 	.verify_write		= xfs_rmapbt_mem_rw_verify,
608 	.verify_struct		= xfs_rmapbt_mem_verify,
609 };
610 
611 const struct xfs_btree_ops xfs_rmapbt_mem_ops = {
612 	.name			= "mem_rmap",
613 	.type			= XFS_BTREE_TYPE_MEM,
614 	.geom_flags		= XFS_BTGEO_OVERLAPPING,
615 
616 	.rec_len		= sizeof(struct xfs_rmap_rec),
617 	/* Overlapping btree; 2 keys per pointer. */
618 	.key_len		= 2 * sizeof(struct xfs_rmap_key),
619 	.ptr_len		= XFS_BTREE_LONG_PTR_LEN,
620 
621 	.lru_refs		= XFS_RMAP_BTREE_REF,
622 	.statoff		= XFS_STATS_CALC_INDEX(xs_rmap_mem_2),
623 
624 	.dup_cursor		= xfbtree_dup_cursor,
625 	.set_root		= xfbtree_set_root,
626 	.alloc_block		= xfbtree_alloc_block,
627 	.free_block		= xfbtree_free_block,
628 	.get_minrecs		= xfbtree_get_minrecs,
629 	.get_maxrecs		= xfbtree_get_maxrecs,
630 	.init_key_from_rec	= xfs_rmapbt_init_key_from_rec,
631 	.init_high_key_from_rec	= xfs_rmapbt_init_high_key_from_rec,
632 	.init_rec_from_cur	= xfs_rmapbt_init_rec_from_cur,
633 	.init_ptr_from_cur	= xfbtree_init_ptr_from_cur,
634 	.key_diff		= xfs_rmapbt_key_diff,
635 	.buf_ops		= &xfs_rmapbt_mem_buf_ops,
636 	.diff_two_keys		= xfs_rmapbt_diff_two_keys,
637 	.keys_inorder		= xfs_rmapbt_keys_inorder,
638 	.recs_inorder		= xfs_rmapbt_recs_inorder,
639 	.keys_contiguous	= xfs_rmapbt_keys_contiguous,
640 };
641 
642 /* Create a cursor for an in-memory btree. */
643 struct xfs_btree_cur *
644 xfs_rmapbt_mem_cursor(
645 	struct xfs_perag	*pag,
646 	struct xfs_trans	*tp,
647 	struct xfbtree		*xfbt)
648 {
649 	struct xfs_btree_cur	*cur;
650 	struct xfs_mount	*mp = pag->pag_mount;
651 
652 	cur = xfs_btree_alloc_cursor(mp, tp, &xfs_rmapbt_mem_ops,
653 			xfs_rmapbt_maxlevels_ondisk(), xfs_rmapbt_cur_cache);
654 	cur->bc_mem.xfbtree = xfbt;
655 	cur->bc_nlevels = xfbt->nlevels;
656 
657 	cur->bc_mem.pag = xfs_perag_hold(pag);
658 	return cur;
659 }
660 
661 /* Create an in-memory rmap btree. */
662 int
663 xfs_rmapbt_mem_init(
664 	struct xfs_mount	*mp,
665 	struct xfbtree		*xfbt,
666 	struct xfs_buftarg	*btp,
667 	xfs_agnumber_t		agno)
668 {
669 	xfbt->owner = agno;
670 	return xfbtree_init(mp, xfbt, btp, &xfs_rmapbt_mem_ops);
671 }
672 
673 /* Compute the max possible height for reverse mapping btrees in memory. */
674 static unsigned int
675 xfs_rmapbt_mem_maxlevels(void)
676 {
677 	unsigned int		minrecs[2];
678 	unsigned int		blocklen;
679 
680 	blocklen = XFBNO_BLOCKSIZE - XFS_BTREE_LBLOCK_CRC_LEN;
681 
682 	minrecs[0] = xfs_rmapbt_mem_block_maxrecs(blocklen, true) / 2;
683 	minrecs[1] = xfs_rmapbt_mem_block_maxrecs(blocklen, false) / 2;
684 
685 	/*
686 	 * How tall can an in-memory rmap btree become if we filled the entire
687 	 * AG with rmap records?
688 	 */
689 	return xfs_btree_compute_maxlevels(minrecs,
690 			XFS_MAX_AG_BYTES / sizeof(struct xfs_rmap_rec));
691 }
692 #else
693 # define xfs_rmapbt_mem_maxlevels()	(0)
694 #endif /* CONFIG_XFS_BTREE_IN_MEM */
695 
696 /*
697  * Install a new reverse mapping btree root.  Caller is responsible for
698  * invalidating and freeing the old btree blocks.
699  */
700 void
701 xfs_rmapbt_commit_staged_btree(
702 	struct xfs_btree_cur	*cur,
703 	struct xfs_trans	*tp,
704 	struct xfs_buf		*agbp)
705 {
706 	struct xfs_agf		*agf = agbp->b_addr;
707 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
708 
709 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
710 
711 	agf->agf_rmap_root = cpu_to_be32(afake->af_root);
712 	agf->agf_rmap_level = cpu_to_be32(afake->af_levels);
713 	agf->agf_rmap_blocks = cpu_to_be32(afake->af_blocks);
714 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS |
715 				    XFS_AGF_RMAP_BLOCKS);
716 	xfs_btree_commit_afakeroot(cur, tp, agbp);
717 }
718 
719 /* Calculate number of records in a reverse mapping btree block. */
720 static inline unsigned int
721 xfs_rmapbt_block_maxrecs(
722 	unsigned int		blocklen,
723 	bool			leaf)
724 {
725 	if (leaf)
726 		return blocklen / sizeof(struct xfs_rmap_rec);
727 	return blocklen /
728 		(2 * sizeof(struct xfs_rmap_key) + sizeof(xfs_rmap_ptr_t));
729 }
730 
731 /*
732  * Calculate number of records in an rmap btree block.
733  */
734 unsigned int
735 xfs_rmapbt_maxrecs(
736 	struct xfs_mount	*mp,
737 	unsigned int		blocklen,
738 	bool			leaf)
739 {
740 	blocklen -= XFS_RMAP_BLOCK_LEN;
741 	return xfs_rmapbt_block_maxrecs(blocklen, leaf);
742 }
743 
744 /* Compute the max possible height for reverse mapping btrees. */
745 unsigned int
746 xfs_rmapbt_maxlevels_ondisk(void)
747 {
748 	unsigned int		minrecs[2];
749 	unsigned int		blocklen;
750 
751 	blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
752 
753 	minrecs[0] = xfs_rmapbt_block_maxrecs(blocklen, true) / 2;
754 	minrecs[1] = xfs_rmapbt_block_maxrecs(blocklen, false) / 2;
755 
756 	/*
757 	 * Compute the asymptotic maxlevels for an rmapbt on any reflink fs.
758 	 *
759 	 * On a reflink filesystem, each AG block can have up to 2^32 (per the
760 	 * refcount record format) owners, which means that theoretically we
761 	 * could face up to 2^64 rmap records.  However, we're likely to run
762 	 * out of blocks in the AG long before that happens, which means that
763 	 * we must compute the max height based on what the btree will look
764 	 * like if it consumes almost all the blocks in the AG due to maximal
765 	 * sharing factor.
766 	 */
767 	return max(xfs_btree_space_to_height(minrecs, XFS_MAX_CRC_AG_BLOCKS),
768 		   xfs_rmapbt_mem_maxlevels());
769 }
770 
771 /* Compute the maximum height of an rmap btree. */
772 void
773 xfs_rmapbt_compute_maxlevels(
774 	struct xfs_mount		*mp)
775 {
776 	if (!xfs_has_rmapbt(mp)) {
777 		mp->m_rmap_maxlevels = 0;
778 		return;
779 	}
780 
781 	if (xfs_has_reflink(mp)) {
782 		/*
783 		 * Compute the asymptotic maxlevels for an rmap btree on a
784 		 * filesystem that supports reflink.
785 		 *
786 		 * On a reflink filesystem, each AG block can have up to 2^32
787 		 * (per the refcount record format) owners, which means that
788 		 * theoretically we could face up to 2^64 rmap records.
789 		 * However, we're likely to run out of blocks in the AG long
790 		 * before that happens, which means that we must compute the
791 		 * max height based on what the btree will look like if it
792 		 * consumes almost all the blocks in the AG due to maximal
793 		 * sharing factor.
794 		 */
795 		mp->m_rmap_maxlevels = xfs_btree_space_to_height(mp->m_rmap_mnr,
796 				mp->m_sb.sb_agblocks);
797 	} else {
798 		/*
799 		 * If there's no block sharing, compute the maximum rmapbt
800 		 * height assuming one rmap record per AG block.
801 		 */
802 		mp->m_rmap_maxlevels = xfs_btree_compute_maxlevels(
803 				mp->m_rmap_mnr, mp->m_sb.sb_agblocks);
804 	}
805 	ASSERT(mp->m_rmap_maxlevels <= xfs_rmapbt_maxlevels_ondisk());
806 }
807 
808 /* Calculate the refcount btree size for some records. */
809 xfs_extlen_t
810 xfs_rmapbt_calc_size(
811 	struct xfs_mount	*mp,
812 	unsigned long long	len)
813 {
814 	return xfs_btree_calc_size(mp->m_rmap_mnr, len);
815 }
816 
817 /*
818  * Calculate the maximum refcount btree size.
819  */
820 xfs_extlen_t
821 xfs_rmapbt_max_size(
822 	struct xfs_mount	*mp,
823 	xfs_agblock_t		agblocks)
824 {
825 	/* Bail out if we're uninitialized, which can happen in mkfs. */
826 	if (mp->m_rmap_mxr[0] == 0)
827 		return 0;
828 
829 	return xfs_rmapbt_calc_size(mp, agblocks);
830 }
831 
832 /*
833  * Figure out how many blocks to reserve and how many are used by this btree.
834  */
835 int
836 xfs_rmapbt_calc_reserves(
837 	struct xfs_mount	*mp,
838 	struct xfs_trans	*tp,
839 	struct xfs_perag	*pag,
840 	xfs_extlen_t		*ask,
841 	xfs_extlen_t		*used)
842 {
843 	struct xfs_buf		*agbp;
844 	struct xfs_agf		*agf;
845 	xfs_agblock_t		agblocks;
846 	xfs_extlen_t		tree_len;
847 	int			error;
848 
849 	if (!xfs_has_rmapbt(mp))
850 		return 0;
851 
852 	error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
853 	if (error)
854 		return error;
855 
856 	agf = agbp->b_addr;
857 	agblocks = be32_to_cpu(agf->agf_length);
858 	tree_len = be32_to_cpu(agf->agf_rmap_blocks);
859 	xfs_trans_brelse(tp, agbp);
860 
861 	/*
862 	 * The log is permanently allocated, so the space it occupies will
863 	 * never be available for the kinds of things that would require btree
864 	 * expansion.  We therefore can pretend the space isn't there.
865 	 */
866 	if (xfs_ag_contains_log(mp, pag->pag_agno))
867 		agblocks -= mp->m_sb.sb_logblocks;
868 
869 	/* Reserve 1% of the AG or enough for 1 block per record. */
870 	*ask += max(agblocks / 100, xfs_rmapbt_max_size(mp, agblocks));
871 	*used += tree_len;
872 
873 	return error;
874 }
875 
876 int __init
877 xfs_rmapbt_init_cur_cache(void)
878 {
879 	xfs_rmapbt_cur_cache = kmem_cache_create("xfs_rmapbt_cur",
880 			xfs_btree_cur_sizeof(xfs_rmapbt_maxlevels_ondisk()),
881 			0, 0, NULL);
882 
883 	if (!xfs_rmapbt_cur_cache)
884 		return -ENOMEM;
885 	return 0;
886 }
887 
888 void
889 xfs_rmapbt_destroy_cur_cache(void)
890 {
891 	kmem_cache_destroy(xfs_rmapbt_cur_cache);
892 	xfs_rmapbt_cur_cache = NULL;
893 }
894