1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
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_trace.h"
20 #include "xfs_trans.h"
25 static struct kmem_cache *xfs_refcountbt_cur_cache;
27 static struct xfs_btree_cur *
28 xfs_refcountbt_dup_cursor(
29 struct xfs_btree_cur *cur)
31 return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
32 cur->bc_ag.agbp, cur->bc_ag.pag);
36 xfs_refcountbt_set_root(
37 struct xfs_btree_cur *cur,
38 const union xfs_btree_ptr *ptr,
41 struct xfs_buf *agbp = cur->bc_ag.agbp;
42 struct xfs_agf *agf = agbp->b_addr;
43 struct xfs_perag *pag = agbp->b_pag;
47 agf->agf_refcount_root = ptr->s;
48 be32_add_cpu(&agf->agf_refcount_level, inc);
49 pag->pagf_refcount_level += inc;
51 xfs_alloc_log_agf(cur->bc_tp, agbp,
52 XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
56 xfs_refcountbt_alloc_block(
57 struct xfs_btree_cur *cur,
58 const union xfs_btree_ptr *start,
59 union xfs_btree_ptr *new,
62 struct xfs_buf *agbp = cur->bc_ag.agbp;
63 struct xfs_agf *agf = agbp->b_addr;
64 struct xfs_alloc_arg args; /* block allocation args */
65 int error; /* error return value */
67 memset(&args, 0, sizeof(args));
70 args.pag = cur->bc_ag.pag;
71 args.oinfo = XFS_RMAP_OINFO_REFC;
72 args.minlen = args.maxlen = args.prod = 1;
73 args.resv = XFS_AG_RESV_METADATA;
75 error = xfs_alloc_vextent_near_bno(&args,
76 XFS_AGB_TO_FSB(args.mp, args.pag->pag_agno,
77 xfs_refc_block(args.mp)));
80 trace_xfs_refcountbt_alloc_block(cur->bc_mp, cur->bc_ag.pag->pag_agno,
82 if (args.fsbno == NULLFSBLOCK) {
86 ASSERT(args.agno == cur->bc_ag.pag->pag_agno);
87 ASSERT(args.len == 1);
89 new->s = cpu_to_be32(args.agbno);
90 be32_add_cpu(&agf->agf_refcount_blocks, 1);
91 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
101 xfs_refcountbt_free_block(
102 struct xfs_btree_cur *cur,
105 struct xfs_mount *mp = cur->bc_mp;
106 struct xfs_buf *agbp = cur->bc_ag.agbp;
107 struct xfs_agf *agf = agbp->b_addr;
108 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
110 trace_xfs_refcountbt_free_block(cur->bc_mp, cur->bc_ag.pag->pag_agno,
111 XFS_FSB_TO_AGBNO(cur->bc_mp, fsbno), 1);
112 be32_add_cpu(&agf->agf_refcount_blocks, -1);
113 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
114 return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
115 &XFS_RMAP_OINFO_REFC, XFS_AG_RESV_METADATA, false);
119 xfs_refcountbt_get_minrecs(
120 struct xfs_btree_cur *cur,
123 return cur->bc_mp->m_refc_mnr[level != 0];
127 xfs_refcountbt_get_maxrecs(
128 struct xfs_btree_cur *cur,
131 return cur->bc_mp->m_refc_mxr[level != 0];
135 xfs_refcountbt_init_key_from_rec(
136 union xfs_btree_key *key,
137 const union xfs_btree_rec *rec)
139 key->refc.rc_startblock = rec->refc.rc_startblock;
143 xfs_refcountbt_init_high_key_from_rec(
144 union xfs_btree_key *key,
145 const union xfs_btree_rec *rec)
149 x = be32_to_cpu(rec->refc.rc_startblock);
150 x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
151 key->refc.rc_startblock = cpu_to_be32(x);
155 xfs_refcountbt_init_rec_from_cur(
156 struct xfs_btree_cur *cur,
157 union xfs_btree_rec *rec)
159 const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
162 start = xfs_refcount_encode_startblock(irec->rc_startblock,
164 rec->refc.rc_startblock = cpu_to_be32(start);
165 rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
166 rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
170 xfs_refcountbt_init_ptr_from_cur(
171 struct xfs_btree_cur *cur,
172 union xfs_btree_ptr *ptr)
174 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
176 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
178 ptr->s = agf->agf_refcount_root;
182 xfs_refcountbt_key_diff(
183 struct xfs_btree_cur *cur,
184 const union xfs_btree_key *key)
186 const struct xfs_refcount_key *kp = &key->refc;
187 const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
190 start = xfs_refcount_encode_startblock(irec->rc_startblock,
192 return (int64_t)be32_to_cpu(kp->rc_startblock) - start;
196 xfs_refcountbt_diff_two_keys(
197 struct xfs_btree_cur *cur,
198 const union xfs_btree_key *k1,
199 const union xfs_btree_key *k2,
200 const union xfs_btree_key *mask)
202 ASSERT(!mask || mask->refc.rc_startblock);
204 return (int64_t)be32_to_cpu(k1->refc.rc_startblock) -
205 be32_to_cpu(k2->refc.rc_startblock);
208 STATIC xfs_failaddr_t
209 xfs_refcountbt_verify(
212 struct xfs_mount *mp = bp->b_mount;
213 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
214 struct xfs_perag *pag = bp->b_pag;
218 if (!xfs_verify_magic(bp, block->bb_magic))
219 return __this_address;
221 if (!xfs_has_reflink(mp))
222 return __this_address;
223 fa = xfs_btree_sblock_v5hdr_verify(bp);
227 level = be16_to_cpu(block->bb_level);
228 if (pag && xfs_perag_initialised_agf(pag)) {
229 unsigned int maxlevel = pag->pagf_refcount_level;
231 #ifdef CONFIG_XFS_ONLINE_REPAIR
233 * Online repair could be rewriting the refcount btree, so
234 * we'll validate against the larger of either tree while this
237 maxlevel = max_t(unsigned int, maxlevel,
238 pag->pagf_repair_refcount_level);
240 if (level >= maxlevel)
241 return __this_address;
242 } else if (level >= mp->m_refc_maxlevels)
243 return __this_address;
245 return xfs_btree_sblock_verify(bp, mp->m_refc_mxr[level != 0]);
249 xfs_refcountbt_read_verify(
254 if (!xfs_btree_sblock_verify_crc(bp))
255 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
257 fa = xfs_refcountbt_verify(bp);
259 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
263 trace_xfs_btree_corrupt(bp, _RET_IP_);
267 xfs_refcountbt_write_verify(
272 fa = xfs_refcountbt_verify(bp);
274 trace_xfs_btree_corrupt(bp, _RET_IP_);
275 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
278 xfs_btree_sblock_calc_crc(bp);
282 const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
283 .name = "xfs_refcountbt",
284 .magic = { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) },
285 .verify_read = xfs_refcountbt_read_verify,
286 .verify_write = xfs_refcountbt_write_verify,
287 .verify_struct = xfs_refcountbt_verify,
291 xfs_refcountbt_keys_inorder(
292 struct xfs_btree_cur *cur,
293 const union xfs_btree_key *k1,
294 const union xfs_btree_key *k2)
296 return be32_to_cpu(k1->refc.rc_startblock) <
297 be32_to_cpu(k2->refc.rc_startblock);
301 xfs_refcountbt_recs_inorder(
302 struct xfs_btree_cur *cur,
303 const union xfs_btree_rec *r1,
304 const union xfs_btree_rec *r2)
306 return be32_to_cpu(r1->refc.rc_startblock) +
307 be32_to_cpu(r1->refc.rc_blockcount) <=
308 be32_to_cpu(r2->refc.rc_startblock);
311 STATIC enum xbtree_key_contig
312 xfs_refcountbt_keys_contiguous(
313 struct xfs_btree_cur *cur,
314 const union xfs_btree_key *key1,
315 const union xfs_btree_key *key2,
316 const union xfs_btree_key *mask)
318 ASSERT(!mask || mask->refc.rc_startblock);
320 return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
321 be32_to_cpu(key2->refc.rc_startblock));
324 static const struct xfs_btree_ops xfs_refcountbt_ops = {
325 .rec_len = sizeof(struct xfs_refcount_rec),
326 .key_len = sizeof(struct xfs_refcount_key),
328 .dup_cursor = xfs_refcountbt_dup_cursor,
329 .set_root = xfs_refcountbt_set_root,
330 .alloc_block = xfs_refcountbt_alloc_block,
331 .free_block = xfs_refcountbt_free_block,
332 .get_minrecs = xfs_refcountbt_get_minrecs,
333 .get_maxrecs = xfs_refcountbt_get_maxrecs,
334 .init_key_from_rec = xfs_refcountbt_init_key_from_rec,
335 .init_high_key_from_rec = xfs_refcountbt_init_high_key_from_rec,
336 .init_rec_from_cur = xfs_refcountbt_init_rec_from_cur,
337 .init_ptr_from_cur = xfs_refcountbt_init_ptr_from_cur,
338 .key_diff = xfs_refcountbt_key_diff,
339 .buf_ops = &xfs_refcountbt_buf_ops,
340 .diff_two_keys = xfs_refcountbt_diff_two_keys,
341 .keys_inorder = xfs_refcountbt_keys_inorder,
342 .recs_inorder = xfs_refcountbt_recs_inorder,
343 .keys_contiguous = xfs_refcountbt_keys_contiguous,
347 * Initialize a new refcount btree cursor.
349 static struct xfs_btree_cur *
350 xfs_refcountbt_init_common(
351 struct xfs_mount *mp,
352 struct xfs_trans *tp,
353 struct xfs_perag *pag)
355 struct xfs_btree_cur *cur;
357 ASSERT(pag->pag_agno < mp->m_sb.sb_agcount);
359 cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_REFC,
360 mp->m_refc_maxlevels, xfs_refcountbt_cur_cache);
361 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2);
363 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
365 cur->bc_ag.pag = xfs_perag_hold(pag);
366 cur->bc_ag.refc.nr_ops = 0;
367 cur->bc_ag.refc.shape_changes = 0;
368 cur->bc_ops = &xfs_refcountbt_ops;
372 /* Create a btree cursor. */
373 struct xfs_btree_cur *
374 xfs_refcountbt_init_cursor(
375 struct xfs_mount *mp,
376 struct xfs_trans *tp,
377 struct xfs_buf *agbp,
378 struct xfs_perag *pag)
380 struct xfs_agf *agf = agbp->b_addr;
381 struct xfs_btree_cur *cur;
383 cur = xfs_refcountbt_init_common(mp, tp, pag);
384 cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
385 cur->bc_ag.agbp = agbp;
389 /* Create a btree cursor with a fake root for staging. */
390 struct xfs_btree_cur *
391 xfs_refcountbt_stage_cursor(
392 struct xfs_mount *mp,
393 struct xbtree_afakeroot *afake,
394 struct xfs_perag *pag)
396 struct xfs_btree_cur *cur;
398 cur = xfs_refcountbt_init_common(mp, NULL, pag);
399 xfs_btree_stage_afakeroot(cur, afake);
404 * Swap in the new btree root. Once we pass this point the newly rebuilt btree
405 * is in place and we have to kill off all the old btree blocks.
408 xfs_refcountbt_commit_staged_btree(
409 struct xfs_btree_cur *cur,
410 struct xfs_trans *tp,
411 struct xfs_buf *agbp)
413 struct xfs_agf *agf = agbp->b_addr;
414 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
416 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
418 agf->agf_refcount_root = cpu_to_be32(afake->af_root);
419 agf->agf_refcount_level = cpu_to_be32(afake->af_levels);
420 agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks);
421 xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS |
422 XFS_AGF_REFCOUNT_ROOT |
423 XFS_AGF_REFCOUNT_LEVEL);
424 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_refcountbt_ops);
427 /* Calculate number of records in a refcount btree block. */
428 static inline unsigned int
429 xfs_refcountbt_block_maxrecs(
430 unsigned int blocklen,
434 return blocklen / sizeof(struct xfs_refcount_rec);
435 return blocklen / (sizeof(struct xfs_refcount_key) +
436 sizeof(xfs_refcount_ptr_t));
440 * Calculate the number of records in a refcount btree block.
443 xfs_refcountbt_maxrecs(
447 blocklen -= XFS_REFCOUNT_BLOCK_LEN;
448 return xfs_refcountbt_block_maxrecs(blocklen, leaf);
451 /* Compute the max possible height of the maximally sized refcount btree. */
453 xfs_refcountbt_maxlevels_ondisk(void)
455 unsigned int minrecs[2];
456 unsigned int blocklen;
458 blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
460 minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2;
461 minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2;
463 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS);
466 /* Compute the maximum height of a refcount btree. */
468 xfs_refcountbt_compute_maxlevels(
469 struct xfs_mount *mp)
471 if (!xfs_has_reflink(mp)) {
472 mp->m_refc_maxlevels = 0;
476 mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(
477 mp->m_refc_mnr, mp->m_sb.sb_agblocks);
478 ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk());
481 /* Calculate the refcount btree size for some records. */
483 xfs_refcountbt_calc_size(
484 struct xfs_mount *mp,
485 unsigned long long len)
487 return xfs_btree_calc_size(mp->m_refc_mnr, len);
491 * Calculate the maximum refcount btree size.
494 xfs_refcountbt_max_size(
495 struct xfs_mount *mp,
496 xfs_agblock_t agblocks)
498 /* Bail out if we're uninitialized, which can happen in mkfs. */
499 if (mp->m_refc_mxr[0] == 0)
502 return xfs_refcountbt_calc_size(mp, agblocks);
506 * Figure out how many blocks to reserve and how many are used by this btree.
509 xfs_refcountbt_calc_reserves(
510 struct xfs_mount *mp,
511 struct xfs_trans *tp,
512 struct xfs_perag *pag,
516 struct xfs_buf *agbp;
518 xfs_agblock_t agblocks;
519 xfs_extlen_t tree_len;
522 if (!xfs_has_reflink(mp))
525 error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
530 agblocks = be32_to_cpu(agf->agf_length);
531 tree_len = be32_to_cpu(agf->agf_refcount_blocks);
532 xfs_trans_brelse(tp, agbp);
535 * The log is permanently allocated, so the space it occupies will
536 * never be available for the kinds of things that would require btree
537 * expansion. We therefore can pretend the space isn't there.
539 if (xfs_ag_contains_log(mp, pag->pag_agno))
540 agblocks -= mp->m_sb.sb_logblocks;
542 *ask += xfs_refcountbt_max_size(mp, agblocks);
549 xfs_refcountbt_init_cur_cache(void)
551 xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur",
552 xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()),
555 if (!xfs_refcountbt_cur_cache)
561 xfs_refcountbt_destroy_cur_cache(void)
563 kmem_cache_destroy(xfs_refcountbt_cur_cache);
564 xfs_refcountbt_cur_cache = NULL;