xref: /linux/fs/xfs/libxfs/xfs_refcount_btree.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2016 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
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_btree.h"
14 #include "xfs_btree_staging.h"
15 #include "xfs_refcount_btree.h"
16 #include "xfs_refcount.h"
17 #include "xfs_alloc.h"
18 #include "xfs_error.h"
19 #include "xfs_health.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
22 #include "xfs_bit.h"
23 #include "xfs_rmap.h"
24 #include "xfs_ag.h"
25 
26 static struct kmem_cache	*xfs_refcountbt_cur_cache;
27 
28 static struct xfs_btree_cur *
29 xfs_refcountbt_dup_cursor(
30 	struct xfs_btree_cur	*cur)
31 {
32 	return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
33 			cur->bc_ag.agbp, to_perag(cur->bc_group));
34 }
35 
36 STATIC void
37 xfs_refcountbt_set_root(
38 	struct xfs_btree_cur		*cur,
39 	const union xfs_btree_ptr	*ptr,
40 	int				inc)
41 {
42 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
43 	struct xfs_agf		*agf = agbp->b_addr;
44 	struct xfs_perag	*pag = agbp->b_pag;
45 
46 	ASSERT(ptr->s != 0);
47 
48 	agf->agf_refcount_root = ptr->s;
49 	be32_add_cpu(&agf->agf_refcount_level, inc);
50 	pag->pagf_refcount_level += inc;
51 
52 	xfs_alloc_log_agf(cur->bc_tp, agbp,
53 			XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
54 }
55 
56 STATIC int
57 xfs_refcountbt_alloc_block(
58 	struct xfs_btree_cur		*cur,
59 	const union xfs_btree_ptr	*start,
60 	union xfs_btree_ptr		*new,
61 	int				*stat)
62 {
63 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
64 	struct xfs_agf		*agf = agbp->b_addr;
65 	struct xfs_alloc_arg	args;		/* block allocation args */
66 	int			error;		/* error return value */
67 
68 	memset(&args, 0, sizeof(args));
69 	args.tp = cur->bc_tp;
70 	args.mp = cur->bc_mp;
71 	args.pag = to_perag(cur->bc_group);
72 	args.oinfo = XFS_RMAP_OINFO_REFC;
73 	args.minlen = args.maxlen = args.prod = 1;
74 	args.resv = XFS_AG_RESV_METADATA;
75 
76 	error = xfs_alloc_vextent_near_bno(&args,
77 			xfs_agbno_to_fsb(args.pag, xfs_refc_block(args.mp)));
78 	if (error)
79 		goto out_error;
80 	if (args.fsbno == NULLFSBLOCK) {
81 		*stat = 0;
82 		return 0;
83 	}
84 	ASSERT(args.agno == cur->bc_group->xg_gno);
85 	ASSERT(args.len == 1);
86 
87 	new->s = cpu_to_be32(args.agbno);
88 	be32_add_cpu(&agf->agf_refcount_blocks, 1);
89 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
90 
91 	*stat = 1;
92 	return 0;
93 
94 out_error:
95 	return error;
96 }
97 
98 STATIC int
99 xfs_refcountbt_free_block(
100 	struct xfs_btree_cur	*cur,
101 	struct xfs_buf		*bp)
102 {
103 	struct xfs_mount	*mp = cur->bc_mp;
104 	struct xfs_buf		*agbp = cur->bc_ag.agbp;
105 	struct xfs_agf		*agf = agbp->b_addr;
106 	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
107 
108 	be32_add_cpu(&agf->agf_refcount_blocks, -1);
109 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
110 	return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
111 			&XFS_RMAP_OINFO_REFC, XFS_AG_RESV_METADATA, 0);
112 }
113 
114 STATIC int
115 xfs_refcountbt_get_minrecs(
116 	struct xfs_btree_cur	*cur,
117 	int			level)
118 {
119 	return cur->bc_mp->m_refc_mnr[level != 0];
120 }
121 
122 STATIC int
123 xfs_refcountbt_get_maxrecs(
124 	struct xfs_btree_cur	*cur,
125 	int			level)
126 {
127 	return cur->bc_mp->m_refc_mxr[level != 0];
128 }
129 
130 STATIC void
131 xfs_refcountbt_init_key_from_rec(
132 	union xfs_btree_key		*key,
133 	const union xfs_btree_rec	*rec)
134 {
135 	key->refc.rc_startblock = rec->refc.rc_startblock;
136 }
137 
138 STATIC void
139 xfs_refcountbt_init_high_key_from_rec(
140 	union xfs_btree_key		*key,
141 	const union xfs_btree_rec	*rec)
142 {
143 	__u32				x;
144 
145 	x = be32_to_cpu(rec->refc.rc_startblock);
146 	x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
147 	key->refc.rc_startblock = cpu_to_be32(x);
148 }
149 
150 STATIC void
151 xfs_refcountbt_init_rec_from_cur(
152 	struct xfs_btree_cur	*cur,
153 	union xfs_btree_rec	*rec)
154 {
155 	const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
156 	uint32_t		start;
157 
158 	start = xfs_refcount_encode_startblock(irec->rc_startblock,
159 			irec->rc_domain);
160 	rec->refc.rc_startblock = cpu_to_be32(start);
161 	rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
162 	rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
163 }
164 
165 STATIC void
166 xfs_refcountbt_init_ptr_from_cur(
167 	struct xfs_btree_cur	*cur,
168 	union xfs_btree_ptr	*ptr)
169 {
170 	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
171 
172 	ASSERT(cur->bc_group->xg_gno == be32_to_cpu(agf->agf_seqno));
173 
174 	ptr->s = agf->agf_refcount_root;
175 }
176 
177 STATIC int64_t
178 xfs_refcountbt_key_diff(
179 	struct xfs_btree_cur		*cur,
180 	const union xfs_btree_key	*key)
181 {
182 	const struct xfs_refcount_key	*kp = &key->refc;
183 	const struct xfs_refcount_irec	*irec = &cur->bc_rec.rc;
184 	uint32_t			start;
185 
186 	start = xfs_refcount_encode_startblock(irec->rc_startblock,
187 			irec->rc_domain);
188 	return (int64_t)be32_to_cpu(kp->rc_startblock) - start;
189 }
190 
191 STATIC int64_t
192 xfs_refcountbt_diff_two_keys(
193 	struct xfs_btree_cur		*cur,
194 	const union xfs_btree_key	*k1,
195 	const union xfs_btree_key	*k2,
196 	const union xfs_btree_key	*mask)
197 {
198 	ASSERT(!mask || mask->refc.rc_startblock);
199 
200 	return (int64_t)be32_to_cpu(k1->refc.rc_startblock) -
201 			be32_to_cpu(k2->refc.rc_startblock);
202 }
203 
204 STATIC xfs_failaddr_t
205 xfs_refcountbt_verify(
206 	struct xfs_buf		*bp)
207 {
208 	struct xfs_mount	*mp = bp->b_mount;
209 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
210 	struct xfs_perag	*pag = bp->b_pag;
211 	xfs_failaddr_t		fa;
212 	unsigned int		level;
213 
214 	if (!xfs_verify_magic(bp, block->bb_magic))
215 		return __this_address;
216 
217 	if (!xfs_has_reflink(mp))
218 		return __this_address;
219 	fa = xfs_btree_agblock_v5hdr_verify(bp);
220 	if (fa)
221 		return fa;
222 
223 	level = be16_to_cpu(block->bb_level);
224 	if (pag && xfs_perag_initialised_agf(pag)) {
225 		unsigned int	maxlevel = pag->pagf_refcount_level;
226 
227 #ifdef CONFIG_XFS_ONLINE_REPAIR
228 		/*
229 		 * Online repair could be rewriting the refcount btree, so
230 		 * we'll validate against the larger of either tree while this
231 		 * is going on.
232 		 */
233 		maxlevel = max_t(unsigned int, maxlevel,
234 				pag->pagf_repair_refcount_level);
235 #endif
236 		if (level >= maxlevel)
237 			return __this_address;
238 	} else if (level >= mp->m_refc_maxlevels)
239 		return __this_address;
240 
241 	return xfs_btree_agblock_verify(bp, mp->m_refc_mxr[level != 0]);
242 }
243 
244 STATIC void
245 xfs_refcountbt_read_verify(
246 	struct xfs_buf	*bp)
247 {
248 	xfs_failaddr_t	fa;
249 
250 	if (!xfs_btree_agblock_verify_crc(bp))
251 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
252 	else {
253 		fa = xfs_refcountbt_verify(bp);
254 		if (fa)
255 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
256 	}
257 
258 	if (bp->b_error)
259 		trace_xfs_btree_corrupt(bp, _RET_IP_);
260 }
261 
262 STATIC void
263 xfs_refcountbt_write_verify(
264 	struct xfs_buf	*bp)
265 {
266 	xfs_failaddr_t	fa;
267 
268 	fa = xfs_refcountbt_verify(bp);
269 	if (fa) {
270 		trace_xfs_btree_corrupt(bp, _RET_IP_);
271 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
272 		return;
273 	}
274 	xfs_btree_agblock_calc_crc(bp);
275 
276 }
277 
278 const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
279 	.name			= "xfs_refcountbt",
280 	.magic			= { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) },
281 	.verify_read		= xfs_refcountbt_read_verify,
282 	.verify_write		= xfs_refcountbt_write_verify,
283 	.verify_struct		= xfs_refcountbt_verify,
284 };
285 
286 STATIC int
287 xfs_refcountbt_keys_inorder(
288 	struct xfs_btree_cur		*cur,
289 	const union xfs_btree_key	*k1,
290 	const union xfs_btree_key	*k2)
291 {
292 	return be32_to_cpu(k1->refc.rc_startblock) <
293 	       be32_to_cpu(k2->refc.rc_startblock);
294 }
295 
296 STATIC int
297 xfs_refcountbt_recs_inorder(
298 	struct xfs_btree_cur		*cur,
299 	const union xfs_btree_rec	*r1,
300 	const union xfs_btree_rec	*r2)
301 {
302 	return  be32_to_cpu(r1->refc.rc_startblock) +
303 		be32_to_cpu(r1->refc.rc_blockcount) <=
304 		be32_to_cpu(r2->refc.rc_startblock);
305 }
306 
307 STATIC enum xbtree_key_contig
308 xfs_refcountbt_keys_contiguous(
309 	struct xfs_btree_cur		*cur,
310 	const union xfs_btree_key	*key1,
311 	const union xfs_btree_key	*key2,
312 	const union xfs_btree_key	*mask)
313 {
314 	ASSERT(!mask || mask->refc.rc_startblock);
315 
316 	return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
317 				 be32_to_cpu(key2->refc.rc_startblock));
318 }
319 
320 const struct xfs_btree_ops xfs_refcountbt_ops = {
321 	.name			= "refcount",
322 	.type			= XFS_BTREE_TYPE_AG,
323 
324 	.rec_len		= sizeof(struct xfs_refcount_rec),
325 	.key_len		= sizeof(struct xfs_refcount_key),
326 	.ptr_len		= XFS_BTREE_SHORT_PTR_LEN,
327 
328 	.lru_refs		= XFS_REFC_BTREE_REF,
329 	.statoff		= XFS_STATS_CALC_INDEX(xs_refcbt_2),
330 	.sick_mask		= XFS_SICK_AG_REFCNTBT,
331 
332 	.dup_cursor		= xfs_refcountbt_dup_cursor,
333 	.set_root		= xfs_refcountbt_set_root,
334 	.alloc_block		= xfs_refcountbt_alloc_block,
335 	.free_block		= xfs_refcountbt_free_block,
336 	.get_minrecs		= xfs_refcountbt_get_minrecs,
337 	.get_maxrecs		= xfs_refcountbt_get_maxrecs,
338 	.init_key_from_rec	= xfs_refcountbt_init_key_from_rec,
339 	.init_high_key_from_rec	= xfs_refcountbt_init_high_key_from_rec,
340 	.init_rec_from_cur	= xfs_refcountbt_init_rec_from_cur,
341 	.init_ptr_from_cur	= xfs_refcountbt_init_ptr_from_cur,
342 	.key_diff		= xfs_refcountbt_key_diff,
343 	.buf_ops		= &xfs_refcountbt_buf_ops,
344 	.diff_two_keys		= xfs_refcountbt_diff_two_keys,
345 	.keys_inorder		= xfs_refcountbt_keys_inorder,
346 	.recs_inorder		= xfs_refcountbt_recs_inorder,
347 	.keys_contiguous	= xfs_refcountbt_keys_contiguous,
348 };
349 
350 /*
351  * Create a new refcount btree cursor.
352  *
353  * For staging cursors tp and agbp are NULL.
354  */
355 struct xfs_btree_cur *
356 xfs_refcountbt_init_cursor(
357 	struct xfs_mount	*mp,
358 	struct xfs_trans	*tp,
359 	struct xfs_buf		*agbp,
360 	struct xfs_perag	*pag)
361 {
362 	struct xfs_btree_cur	*cur;
363 
364 	ASSERT(pag_agno(pag) < mp->m_sb.sb_agcount);
365 
366 	cur = xfs_btree_alloc_cursor(mp, tp, &xfs_refcountbt_ops,
367 			mp->m_refc_maxlevels, xfs_refcountbt_cur_cache);
368 	cur->bc_group = xfs_group_hold(pag_group(pag));
369 	cur->bc_refc.nr_ops = 0;
370 	cur->bc_refc.shape_changes = 0;
371 	cur->bc_ag.agbp = agbp;
372 	if (agbp) {
373 		struct xfs_agf		*agf = agbp->b_addr;
374 
375 		cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
376 	}
377 	return cur;
378 }
379 
380 /*
381  * Swap in the new btree root.  Once we pass this point the newly rebuilt btree
382  * is in place and we have to kill off all the old btree blocks.
383  */
384 void
385 xfs_refcountbt_commit_staged_btree(
386 	struct xfs_btree_cur	*cur,
387 	struct xfs_trans	*tp,
388 	struct xfs_buf		*agbp)
389 {
390 	struct xfs_agf		*agf = agbp->b_addr;
391 	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
392 
393 	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
394 
395 	agf->agf_refcount_root = cpu_to_be32(afake->af_root);
396 	agf->agf_refcount_level = cpu_to_be32(afake->af_levels);
397 	agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks);
398 	xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS |
399 				    XFS_AGF_REFCOUNT_ROOT |
400 				    XFS_AGF_REFCOUNT_LEVEL);
401 	xfs_btree_commit_afakeroot(cur, tp, agbp);
402 }
403 
404 /* Calculate number of records in a refcount btree block. */
405 static inline unsigned int
406 xfs_refcountbt_block_maxrecs(
407 	unsigned int		blocklen,
408 	bool			leaf)
409 {
410 	if (leaf)
411 		return blocklen / sizeof(struct xfs_refcount_rec);
412 	return blocklen / (sizeof(struct xfs_refcount_key) +
413 			   sizeof(xfs_refcount_ptr_t));
414 }
415 
416 /*
417  * Calculate the number of records in a refcount btree block.
418  */
419 unsigned int
420 xfs_refcountbt_maxrecs(
421 	struct xfs_mount	*mp,
422 	unsigned int		blocklen,
423 	bool			leaf)
424 {
425 	blocklen -= XFS_REFCOUNT_BLOCK_LEN;
426 	return xfs_refcountbt_block_maxrecs(blocklen, leaf);
427 }
428 
429 /* Compute the max possible height of the maximally sized refcount btree. */
430 unsigned int
431 xfs_refcountbt_maxlevels_ondisk(void)
432 {
433 	unsigned int		minrecs[2];
434 	unsigned int		blocklen;
435 
436 	blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
437 
438 	minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2;
439 	minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2;
440 
441 	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS);
442 }
443 
444 /* Compute the maximum height of a refcount btree. */
445 void
446 xfs_refcountbt_compute_maxlevels(
447 	struct xfs_mount		*mp)
448 {
449 	if (!xfs_has_reflink(mp)) {
450 		mp->m_refc_maxlevels = 0;
451 		return;
452 	}
453 
454 	mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(
455 			mp->m_refc_mnr, mp->m_sb.sb_agblocks);
456 	ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk());
457 }
458 
459 /* Calculate the refcount btree size for some records. */
460 xfs_extlen_t
461 xfs_refcountbt_calc_size(
462 	struct xfs_mount	*mp,
463 	unsigned long long	len)
464 {
465 	return xfs_btree_calc_size(mp->m_refc_mnr, len);
466 }
467 
468 /*
469  * Calculate the maximum refcount btree size.
470  */
471 xfs_extlen_t
472 xfs_refcountbt_max_size(
473 	struct xfs_mount	*mp,
474 	xfs_agblock_t		agblocks)
475 {
476 	/* Bail out if we're uninitialized, which can happen in mkfs. */
477 	if (mp->m_refc_mxr[0] == 0)
478 		return 0;
479 
480 	return xfs_refcountbt_calc_size(mp, agblocks);
481 }
482 
483 /*
484  * Figure out how many blocks to reserve and how many are used by this btree.
485  */
486 int
487 xfs_refcountbt_calc_reserves(
488 	struct xfs_mount	*mp,
489 	struct xfs_trans	*tp,
490 	struct xfs_perag	*pag,
491 	xfs_extlen_t		*ask,
492 	xfs_extlen_t		*used)
493 {
494 	struct xfs_buf		*agbp;
495 	struct xfs_agf		*agf;
496 	xfs_agblock_t		agblocks;
497 	xfs_extlen_t		tree_len;
498 	int			error;
499 
500 	if (!xfs_has_reflink(mp))
501 		return 0;
502 
503 	error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
504 	if (error)
505 		return error;
506 
507 	agf = agbp->b_addr;
508 	agblocks = be32_to_cpu(agf->agf_length);
509 	tree_len = be32_to_cpu(agf->agf_refcount_blocks);
510 	xfs_trans_brelse(tp, agbp);
511 
512 	/*
513 	 * The log is permanently allocated, so the space it occupies will
514 	 * never be available for the kinds of things that would require btree
515 	 * expansion.  We therefore can pretend the space isn't there.
516 	 */
517 	if (xfs_ag_contains_log(mp, pag_agno(pag)))
518 		agblocks -= mp->m_sb.sb_logblocks;
519 
520 	*ask += xfs_refcountbt_max_size(mp, agblocks);
521 	*used += tree_len;
522 
523 	return error;
524 }
525 
526 int __init
527 xfs_refcountbt_init_cur_cache(void)
528 {
529 	xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur",
530 			xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()),
531 			0, 0, NULL);
532 
533 	if (!xfs_refcountbt_cur_cache)
534 		return -ENOMEM;
535 	return 0;
536 }
537 
538 void
539 xfs_refcountbt_destroy_cur_cache(void)
540 {
541 	kmem_cache_destroy(xfs_refcountbt_cur_cache);
542 	xfs_refcountbt_cur_cache = NULL;
543 }
544