1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
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_alloc_btree.h"
16 #include "xfs_alloc.h"
17 #include "xfs_extent_busy.h"
18 #include "xfs_error.h"
19 #include "xfs_trace.h"
20 #include "xfs_trans.h"
23 static struct kmem_cache *xfs_allocbt_cur_cache;
25 STATIC struct xfs_btree_cur *
26 xfs_allocbt_dup_cursor(
27 struct xfs_btree_cur *cur)
29 return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
30 cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
35 struct xfs_btree_cur *cur,
36 const union xfs_btree_ptr *ptr,
39 struct xfs_buf *agbp = cur->bc_ag.agbp;
40 struct xfs_agf *agf = agbp->b_addr;
41 int btnum = cur->bc_btnum;
45 agf->agf_roots[btnum] = ptr->s;
46 be32_add_cpu(&agf->agf_levels[btnum], inc);
47 cur->bc_ag.pag->pagf_levels[btnum] += inc;
49 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
53 xfs_allocbt_alloc_block(
54 struct xfs_btree_cur *cur,
55 const union xfs_btree_ptr *start,
56 union xfs_btree_ptr *new,
62 /* Allocate the new block from the freelist. If we can't, give up. */
63 error = xfs_alloc_get_freelist(cur->bc_ag.pag, cur->bc_tp,
64 cur->bc_ag.agbp, &bno, 1);
68 if (bno == NULLAGBLOCK) {
73 atomic64_inc(&cur->bc_mp->m_allocbt_blks);
74 xfs_extent_busy_reuse(cur->bc_mp, cur->bc_ag.pag, bno, 1, false);
76 new->s = cpu_to_be32(bno);
83 xfs_allocbt_free_block(
84 struct xfs_btree_cur *cur,
87 struct xfs_buf *agbp = cur->bc_ag.agbp;
91 bno = xfs_daddr_to_agbno(cur->bc_mp, xfs_buf_daddr(bp));
92 error = xfs_alloc_put_freelist(cur->bc_ag.pag, cur->bc_tp, agbp, NULL,
97 atomic64_dec(&cur->bc_mp->m_allocbt_blks);
98 xfs_extent_busy_insert(cur->bc_tp, agbp->b_pag, bno, 1,
99 XFS_EXTENT_BUSY_SKIP_DISCARD);
104 * Update the longest extent in the AGF
107 xfs_allocbt_update_lastrec(
108 struct xfs_btree_cur *cur,
109 const struct xfs_btree_block *block,
110 const union xfs_btree_rec *rec,
114 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
115 struct xfs_perag *pag;
119 ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
124 * If this is the last leaf block and it's the last record,
125 * then update the size of the longest extent in the AG.
127 if (ptr != xfs_btree_get_numrecs(block))
129 len = rec->alloc.ar_blockcount;
132 if (be32_to_cpu(rec->alloc.ar_blockcount) <=
133 be32_to_cpu(agf->agf_longest))
135 len = rec->alloc.ar_blockcount;
138 numrecs = xfs_btree_get_numrecs(block);
141 ASSERT(ptr == numrecs + 1);
144 xfs_alloc_rec_t *rrp;
146 rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
147 len = rrp->ar_blockcount;
158 agf->agf_longest = len;
159 pag = cur->bc_ag.agbp->b_pag;
160 pag->pagf_longest = be32_to_cpu(len);
161 xfs_alloc_log_agf(cur->bc_tp, cur->bc_ag.agbp, XFS_AGF_LONGEST);
165 xfs_allocbt_get_minrecs(
166 struct xfs_btree_cur *cur,
169 return cur->bc_mp->m_alloc_mnr[level != 0];
173 xfs_allocbt_get_maxrecs(
174 struct xfs_btree_cur *cur,
177 return cur->bc_mp->m_alloc_mxr[level != 0];
181 xfs_allocbt_init_key_from_rec(
182 union xfs_btree_key *key,
183 const union xfs_btree_rec *rec)
185 key->alloc.ar_startblock = rec->alloc.ar_startblock;
186 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
190 xfs_bnobt_init_high_key_from_rec(
191 union xfs_btree_key *key,
192 const union xfs_btree_rec *rec)
196 x = be32_to_cpu(rec->alloc.ar_startblock);
197 x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
198 key->alloc.ar_startblock = cpu_to_be32(x);
199 key->alloc.ar_blockcount = 0;
203 xfs_cntbt_init_high_key_from_rec(
204 union xfs_btree_key *key,
205 const union xfs_btree_rec *rec)
207 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
208 key->alloc.ar_startblock = 0;
212 xfs_allocbt_init_rec_from_cur(
213 struct xfs_btree_cur *cur,
214 union xfs_btree_rec *rec)
216 rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
217 rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
221 xfs_allocbt_init_ptr_from_cur(
222 struct xfs_btree_cur *cur,
223 union xfs_btree_ptr *ptr)
225 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
227 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
229 ptr->s = agf->agf_roots[cur->bc_btnum];
234 struct xfs_btree_cur *cur,
235 const union xfs_btree_key *key)
237 struct xfs_alloc_rec_incore *rec = &cur->bc_rec.a;
238 const struct xfs_alloc_rec *kp = &key->alloc;
240 return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
245 struct xfs_btree_cur *cur,
246 const union xfs_btree_key *key)
248 struct xfs_alloc_rec_incore *rec = &cur->bc_rec.a;
249 const struct xfs_alloc_rec *kp = &key->alloc;
252 diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
256 return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
260 xfs_bnobt_diff_two_keys(
261 struct xfs_btree_cur *cur,
262 const union xfs_btree_key *k1,
263 const union xfs_btree_key *k2,
264 const union xfs_btree_key *mask)
266 ASSERT(!mask || mask->alloc.ar_startblock);
268 return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
269 be32_to_cpu(k2->alloc.ar_startblock);
273 xfs_cntbt_diff_two_keys(
274 struct xfs_btree_cur *cur,
275 const union xfs_btree_key *k1,
276 const union xfs_btree_key *k2,
277 const union xfs_btree_key *mask)
281 ASSERT(!mask || (mask->alloc.ar_blockcount &&
282 mask->alloc.ar_startblock));
284 diff = be32_to_cpu(k1->alloc.ar_blockcount) -
285 be32_to_cpu(k2->alloc.ar_blockcount);
289 return be32_to_cpu(k1->alloc.ar_startblock) -
290 be32_to_cpu(k2->alloc.ar_startblock);
293 static xfs_failaddr_t
297 struct xfs_mount *mp = bp->b_mount;
298 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
299 struct xfs_perag *pag = bp->b_pag;
302 xfs_btnum_t btnum = XFS_BTNUM_BNOi;
304 if (!xfs_verify_magic(bp, block->bb_magic))
305 return __this_address;
307 if (xfs_has_crc(mp)) {
308 fa = xfs_btree_sblock_v5hdr_verify(bp);
314 * The perag may not be attached during grow operations or fully
315 * initialized from the AGF during log recovery. Therefore we can only
316 * check against maximum tree depth from those contexts.
318 * Otherwise check against the per-tree limit. Peek at one of the
319 * verifier magic values to determine the type of tree we're verifying
322 level = be16_to_cpu(block->bb_level);
323 if (bp->b_ops->magic[0] == cpu_to_be32(XFS_ABTC_MAGIC))
324 btnum = XFS_BTNUM_CNTi;
325 if (pag && xfs_perag_initialised_agf(pag)) {
326 if (level >= pag->pagf_levels[btnum])
327 return __this_address;
328 } else if (level >= mp->m_alloc_maxlevels)
329 return __this_address;
331 return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
335 xfs_allocbt_read_verify(
340 if (!xfs_btree_sblock_verify_crc(bp))
341 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
343 fa = xfs_allocbt_verify(bp);
345 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
349 trace_xfs_btree_corrupt(bp, _RET_IP_);
353 xfs_allocbt_write_verify(
358 fa = xfs_allocbt_verify(bp);
360 trace_xfs_btree_corrupt(bp, _RET_IP_);
361 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
364 xfs_btree_sblock_calc_crc(bp);
368 const struct xfs_buf_ops xfs_bnobt_buf_ops = {
370 .magic = { cpu_to_be32(XFS_ABTB_MAGIC),
371 cpu_to_be32(XFS_ABTB_CRC_MAGIC) },
372 .verify_read = xfs_allocbt_read_verify,
373 .verify_write = xfs_allocbt_write_verify,
374 .verify_struct = xfs_allocbt_verify,
377 const struct xfs_buf_ops xfs_cntbt_buf_ops = {
379 .magic = { cpu_to_be32(XFS_ABTC_MAGIC),
380 cpu_to_be32(XFS_ABTC_CRC_MAGIC) },
381 .verify_read = xfs_allocbt_read_verify,
382 .verify_write = xfs_allocbt_write_verify,
383 .verify_struct = xfs_allocbt_verify,
387 xfs_bnobt_keys_inorder(
388 struct xfs_btree_cur *cur,
389 const union xfs_btree_key *k1,
390 const union xfs_btree_key *k2)
392 return be32_to_cpu(k1->alloc.ar_startblock) <
393 be32_to_cpu(k2->alloc.ar_startblock);
397 xfs_bnobt_recs_inorder(
398 struct xfs_btree_cur *cur,
399 const union xfs_btree_rec *r1,
400 const union xfs_btree_rec *r2)
402 return be32_to_cpu(r1->alloc.ar_startblock) +
403 be32_to_cpu(r1->alloc.ar_blockcount) <=
404 be32_to_cpu(r2->alloc.ar_startblock);
408 xfs_cntbt_keys_inorder(
409 struct xfs_btree_cur *cur,
410 const union xfs_btree_key *k1,
411 const union xfs_btree_key *k2)
413 return be32_to_cpu(k1->alloc.ar_blockcount) <
414 be32_to_cpu(k2->alloc.ar_blockcount) ||
415 (k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
416 be32_to_cpu(k1->alloc.ar_startblock) <
417 be32_to_cpu(k2->alloc.ar_startblock));
421 xfs_cntbt_recs_inorder(
422 struct xfs_btree_cur *cur,
423 const union xfs_btree_rec *r1,
424 const union xfs_btree_rec *r2)
426 return be32_to_cpu(r1->alloc.ar_blockcount) <
427 be32_to_cpu(r2->alloc.ar_blockcount) ||
428 (r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
429 be32_to_cpu(r1->alloc.ar_startblock) <
430 be32_to_cpu(r2->alloc.ar_startblock));
433 STATIC enum xbtree_key_contig
434 xfs_allocbt_keys_contiguous(
435 struct xfs_btree_cur *cur,
436 const union xfs_btree_key *key1,
437 const union xfs_btree_key *key2,
438 const union xfs_btree_key *mask)
440 ASSERT(!mask || mask->alloc.ar_startblock);
442 return xbtree_key_contig(be32_to_cpu(key1->alloc.ar_startblock),
443 be32_to_cpu(key2->alloc.ar_startblock));
446 static const struct xfs_btree_ops xfs_bnobt_ops = {
447 .rec_len = sizeof(xfs_alloc_rec_t),
448 .key_len = sizeof(xfs_alloc_key_t),
450 .dup_cursor = xfs_allocbt_dup_cursor,
451 .set_root = xfs_allocbt_set_root,
452 .alloc_block = xfs_allocbt_alloc_block,
453 .free_block = xfs_allocbt_free_block,
454 .update_lastrec = xfs_allocbt_update_lastrec,
455 .get_minrecs = xfs_allocbt_get_minrecs,
456 .get_maxrecs = xfs_allocbt_get_maxrecs,
457 .init_key_from_rec = xfs_allocbt_init_key_from_rec,
458 .init_high_key_from_rec = xfs_bnobt_init_high_key_from_rec,
459 .init_rec_from_cur = xfs_allocbt_init_rec_from_cur,
460 .init_ptr_from_cur = xfs_allocbt_init_ptr_from_cur,
461 .key_diff = xfs_bnobt_key_diff,
462 .buf_ops = &xfs_bnobt_buf_ops,
463 .diff_two_keys = xfs_bnobt_diff_two_keys,
464 .keys_inorder = xfs_bnobt_keys_inorder,
465 .recs_inorder = xfs_bnobt_recs_inorder,
466 .keys_contiguous = xfs_allocbt_keys_contiguous,
469 static const struct xfs_btree_ops xfs_cntbt_ops = {
470 .rec_len = sizeof(xfs_alloc_rec_t),
471 .key_len = sizeof(xfs_alloc_key_t),
473 .dup_cursor = xfs_allocbt_dup_cursor,
474 .set_root = xfs_allocbt_set_root,
475 .alloc_block = xfs_allocbt_alloc_block,
476 .free_block = xfs_allocbt_free_block,
477 .update_lastrec = xfs_allocbt_update_lastrec,
478 .get_minrecs = xfs_allocbt_get_minrecs,
479 .get_maxrecs = xfs_allocbt_get_maxrecs,
480 .init_key_from_rec = xfs_allocbt_init_key_from_rec,
481 .init_high_key_from_rec = xfs_cntbt_init_high_key_from_rec,
482 .init_rec_from_cur = xfs_allocbt_init_rec_from_cur,
483 .init_ptr_from_cur = xfs_allocbt_init_ptr_from_cur,
484 .key_diff = xfs_cntbt_key_diff,
485 .buf_ops = &xfs_cntbt_buf_ops,
486 .diff_two_keys = xfs_cntbt_diff_two_keys,
487 .keys_inorder = xfs_cntbt_keys_inorder,
488 .recs_inorder = xfs_cntbt_recs_inorder,
489 .keys_contiguous = NULL, /* not needed right now */
492 /* Allocate most of a new allocation btree cursor. */
493 STATIC struct xfs_btree_cur *
494 xfs_allocbt_init_common(
495 struct xfs_mount *mp,
496 struct xfs_trans *tp,
497 struct xfs_perag *pag,
500 struct xfs_btree_cur *cur;
502 ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
504 cur = xfs_btree_alloc_cursor(mp, tp, btnum, mp->m_alloc_maxlevels,
505 xfs_allocbt_cur_cache);
506 cur->bc_ag.abt.active = false;
508 if (btnum == XFS_BTNUM_CNT) {
509 cur->bc_ops = &xfs_cntbt_ops;
510 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
511 cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
513 cur->bc_ops = &xfs_bnobt_ops;
514 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
517 cur->bc_ag.pag = xfs_perag_hold(pag);
520 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
526 * Allocate a new allocation btree cursor.
528 struct xfs_btree_cur * /* new alloc btree cursor */
529 xfs_allocbt_init_cursor(
530 struct xfs_mount *mp, /* file system mount point */
531 struct xfs_trans *tp, /* transaction pointer */
532 struct xfs_buf *agbp, /* buffer for agf structure */
533 struct xfs_perag *pag,
534 xfs_btnum_t btnum) /* btree identifier */
536 struct xfs_agf *agf = agbp->b_addr;
537 struct xfs_btree_cur *cur;
539 cur = xfs_allocbt_init_common(mp, tp, pag, btnum);
540 if (btnum == XFS_BTNUM_CNT)
541 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
543 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
545 cur->bc_ag.agbp = agbp;
550 /* Create a free space btree cursor with a fake root for staging. */
551 struct xfs_btree_cur *
552 xfs_allocbt_stage_cursor(
553 struct xfs_mount *mp,
554 struct xbtree_afakeroot *afake,
555 struct xfs_perag *pag,
558 struct xfs_btree_cur *cur;
560 cur = xfs_allocbt_init_common(mp, NULL, pag, btnum);
561 xfs_btree_stage_afakeroot(cur, afake);
566 * Install a new free space btree root. Caller is responsible for invalidating
567 * and freeing the old btree blocks.
570 xfs_allocbt_commit_staged_btree(
571 struct xfs_btree_cur *cur,
572 struct xfs_trans *tp,
573 struct xfs_buf *agbp)
575 struct xfs_agf *agf = agbp->b_addr;
576 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
578 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
580 agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root);
581 agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels);
582 xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
584 if (cur->bc_btnum == XFS_BTNUM_BNO) {
585 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_bnobt_ops);
587 cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE;
588 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_cntbt_ops);
592 /* Calculate number of records in an alloc btree block. */
593 static inline unsigned int
594 xfs_allocbt_block_maxrecs(
595 unsigned int blocklen,
599 return blocklen / sizeof(xfs_alloc_rec_t);
600 return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
604 * Calculate number of records in an alloc btree block.
608 struct xfs_mount *mp,
612 blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
613 return xfs_allocbt_block_maxrecs(blocklen, leaf);
616 /* Free space btrees are at their largest when every other block is free. */
617 #define XFS_MAX_FREESP_RECORDS ((XFS_MAX_AG_BLOCKS + 1) / 2)
619 /* Compute the max possible height for free space btrees. */
621 xfs_allocbt_maxlevels_ondisk(void)
623 unsigned int minrecs[2];
624 unsigned int blocklen;
626 blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
627 XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
629 minrecs[0] = xfs_allocbt_block_maxrecs(blocklen, true) / 2;
630 minrecs[1] = xfs_allocbt_block_maxrecs(blocklen, false) / 2;
632 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_FREESP_RECORDS);
635 /* Calculate the freespace btree size for some records. */
637 xfs_allocbt_calc_size(
638 struct xfs_mount *mp,
639 unsigned long long len)
641 return xfs_btree_calc_size(mp->m_alloc_mnr, len);
645 xfs_allocbt_init_cur_cache(void)
647 xfs_allocbt_cur_cache = kmem_cache_create("xfs_bnobt_cur",
648 xfs_btree_cur_sizeof(xfs_allocbt_maxlevels_ondisk()),
651 if (!xfs_allocbt_cur_cache)
657 xfs_allocbt_destroy_cur_cache(void)
659 kmem_cache_destroy(xfs_allocbt_cur_cache);
660 xfs_allocbt_cur_cache = NULL;