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"
24 STATIC struct xfs_btree_cur *
25 xfs_allocbt_dup_cursor(
26 struct xfs_btree_cur *cur)
28 return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
29 cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
34 struct xfs_btree_cur *cur,
35 union xfs_btree_ptr *ptr,
38 struct xfs_buf *agbp = cur->bc_ag.agbp;
39 struct xfs_agf *agf = agbp->b_addr;
40 int btnum = cur->bc_btnum;
44 agf->agf_roots[btnum] = ptr->s;
45 be32_add_cpu(&agf->agf_levels[btnum], inc);
46 cur->bc_ag.pag->pagf_levels[btnum] += inc;
48 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
52 xfs_allocbt_alloc_block(
53 struct xfs_btree_cur *cur,
54 union xfs_btree_ptr *start,
55 union xfs_btree_ptr *new,
61 /* Allocate the new block from the freelist. If we can't, give up. */
62 error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_ag.agbp,
67 if (bno == NULLAGBLOCK) {
72 atomic64_inc(&cur->bc_mp->m_allocbt_blks);
73 xfs_extent_busy_reuse(cur->bc_mp, cur->bc_ag.agbp->b_pag, bno, 1, false);
75 new->s = cpu_to_be32(bno);
82 xfs_allocbt_free_block(
83 struct xfs_btree_cur *cur,
86 struct xfs_buf *agbp = cur->bc_ag.agbp;
90 bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
91 error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
95 atomic64_dec(&cur->bc_mp->m_allocbt_blks);
96 xfs_extent_busy_insert(cur->bc_tp, agbp->b_pag, bno, 1,
97 XFS_EXTENT_BUSY_SKIP_DISCARD);
102 * Update the longest extent in the AGF
105 xfs_allocbt_update_lastrec(
106 struct xfs_btree_cur *cur,
107 struct xfs_btree_block *block,
108 union xfs_btree_rec *rec,
112 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
113 struct xfs_perag *pag;
117 ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
122 * If this is the last leaf block and it's the last record,
123 * then update the size of the longest extent in the AG.
125 if (ptr != xfs_btree_get_numrecs(block))
127 len = rec->alloc.ar_blockcount;
130 if (be32_to_cpu(rec->alloc.ar_blockcount) <=
131 be32_to_cpu(agf->agf_longest))
133 len = rec->alloc.ar_blockcount;
136 numrecs = xfs_btree_get_numrecs(block);
139 ASSERT(ptr == numrecs + 1);
142 xfs_alloc_rec_t *rrp;
144 rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
145 len = rrp->ar_blockcount;
156 agf->agf_longest = len;
157 pag = cur->bc_ag.agbp->b_pag;
158 pag->pagf_longest = be32_to_cpu(len);
159 xfs_alloc_log_agf(cur->bc_tp, cur->bc_ag.agbp, XFS_AGF_LONGEST);
163 xfs_allocbt_get_minrecs(
164 struct xfs_btree_cur *cur,
167 return cur->bc_mp->m_alloc_mnr[level != 0];
171 xfs_allocbt_get_maxrecs(
172 struct xfs_btree_cur *cur,
175 return cur->bc_mp->m_alloc_mxr[level != 0];
179 xfs_allocbt_init_key_from_rec(
180 union xfs_btree_key *key,
181 union xfs_btree_rec *rec)
183 key->alloc.ar_startblock = rec->alloc.ar_startblock;
184 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
188 xfs_bnobt_init_high_key_from_rec(
189 union xfs_btree_key *key,
190 union xfs_btree_rec *rec)
194 x = be32_to_cpu(rec->alloc.ar_startblock);
195 x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
196 key->alloc.ar_startblock = cpu_to_be32(x);
197 key->alloc.ar_blockcount = 0;
201 xfs_cntbt_init_high_key_from_rec(
202 union xfs_btree_key *key,
203 union xfs_btree_rec *rec)
205 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
206 key->alloc.ar_startblock = 0;
210 xfs_allocbt_init_rec_from_cur(
211 struct xfs_btree_cur *cur,
212 union xfs_btree_rec *rec)
214 rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
215 rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
219 xfs_allocbt_init_ptr_from_cur(
220 struct xfs_btree_cur *cur,
221 union xfs_btree_ptr *ptr)
223 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
225 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
227 ptr->s = agf->agf_roots[cur->bc_btnum];
232 struct xfs_btree_cur *cur,
233 union xfs_btree_key *key)
235 xfs_alloc_rec_incore_t *rec = &cur->bc_rec.a;
236 xfs_alloc_key_t *kp = &key->alloc;
238 return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
243 struct xfs_btree_cur *cur,
244 union xfs_btree_key *key)
246 xfs_alloc_rec_incore_t *rec = &cur->bc_rec.a;
247 xfs_alloc_key_t *kp = &key->alloc;
250 diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
254 return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
258 xfs_bnobt_diff_two_keys(
259 struct xfs_btree_cur *cur,
260 union xfs_btree_key *k1,
261 union xfs_btree_key *k2)
263 return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
264 be32_to_cpu(k2->alloc.ar_startblock);
268 xfs_cntbt_diff_two_keys(
269 struct xfs_btree_cur *cur,
270 union xfs_btree_key *k1,
271 union xfs_btree_key *k2)
275 diff = be32_to_cpu(k1->alloc.ar_blockcount) -
276 be32_to_cpu(k2->alloc.ar_blockcount);
280 return be32_to_cpu(k1->alloc.ar_startblock) -
281 be32_to_cpu(k2->alloc.ar_startblock);
284 static xfs_failaddr_t
288 struct xfs_mount *mp = bp->b_mount;
289 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
290 struct xfs_perag *pag = bp->b_pag;
293 xfs_btnum_t btnum = XFS_BTNUM_BNOi;
295 if (!xfs_verify_magic(bp, block->bb_magic))
296 return __this_address;
298 if (xfs_sb_version_hascrc(&mp->m_sb)) {
299 fa = xfs_btree_sblock_v5hdr_verify(bp);
305 * The perag may not be attached during grow operations or fully
306 * initialized from the AGF during log recovery. Therefore we can only
307 * check against maximum tree depth from those contexts.
309 * Otherwise check against the per-tree limit. Peek at one of the
310 * verifier magic values to determine the type of tree we're verifying
313 level = be16_to_cpu(block->bb_level);
314 if (bp->b_ops->magic[0] == cpu_to_be32(XFS_ABTC_MAGIC))
315 btnum = XFS_BTNUM_CNTi;
316 if (pag && pag->pagf_init) {
317 if (level >= pag->pagf_levels[btnum])
318 return __this_address;
319 } else if (level >= mp->m_ag_maxlevels)
320 return __this_address;
322 return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
326 xfs_allocbt_read_verify(
331 if (!xfs_btree_sblock_verify_crc(bp))
332 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
334 fa = xfs_allocbt_verify(bp);
336 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
340 trace_xfs_btree_corrupt(bp, _RET_IP_);
344 xfs_allocbt_write_verify(
349 fa = xfs_allocbt_verify(bp);
351 trace_xfs_btree_corrupt(bp, _RET_IP_);
352 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
355 xfs_btree_sblock_calc_crc(bp);
359 const struct xfs_buf_ops xfs_bnobt_buf_ops = {
361 .magic = { cpu_to_be32(XFS_ABTB_MAGIC),
362 cpu_to_be32(XFS_ABTB_CRC_MAGIC) },
363 .verify_read = xfs_allocbt_read_verify,
364 .verify_write = xfs_allocbt_write_verify,
365 .verify_struct = xfs_allocbt_verify,
368 const struct xfs_buf_ops xfs_cntbt_buf_ops = {
370 .magic = { cpu_to_be32(XFS_ABTC_MAGIC),
371 cpu_to_be32(XFS_ABTC_CRC_MAGIC) },
372 .verify_read = xfs_allocbt_read_verify,
373 .verify_write = xfs_allocbt_write_verify,
374 .verify_struct = xfs_allocbt_verify,
378 xfs_bnobt_keys_inorder(
379 struct xfs_btree_cur *cur,
380 union xfs_btree_key *k1,
381 union xfs_btree_key *k2)
383 return be32_to_cpu(k1->alloc.ar_startblock) <
384 be32_to_cpu(k2->alloc.ar_startblock);
388 xfs_bnobt_recs_inorder(
389 struct xfs_btree_cur *cur,
390 union xfs_btree_rec *r1,
391 union xfs_btree_rec *r2)
393 return be32_to_cpu(r1->alloc.ar_startblock) +
394 be32_to_cpu(r1->alloc.ar_blockcount) <=
395 be32_to_cpu(r2->alloc.ar_startblock);
399 xfs_cntbt_keys_inorder(
400 struct xfs_btree_cur *cur,
401 union xfs_btree_key *k1,
402 union xfs_btree_key *k2)
404 return be32_to_cpu(k1->alloc.ar_blockcount) <
405 be32_to_cpu(k2->alloc.ar_blockcount) ||
406 (k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
407 be32_to_cpu(k1->alloc.ar_startblock) <
408 be32_to_cpu(k2->alloc.ar_startblock));
412 xfs_cntbt_recs_inorder(
413 struct xfs_btree_cur *cur,
414 union xfs_btree_rec *r1,
415 union xfs_btree_rec *r2)
417 return be32_to_cpu(r1->alloc.ar_blockcount) <
418 be32_to_cpu(r2->alloc.ar_blockcount) ||
419 (r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
420 be32_to_cpu(r1->alloc.ar_startblock) <
421 be32_to_cpu(r2->alloc.ar_startblock));
424 static const struct xfs_btree_ops xfs_bnobt_ops = {
425 .rec_len = sizeof(xfs_alloc_rec_t),
426 .key_len = sizeof(xfs_alloc_key_t),
428 .dup_cursor = xfs_allocbt_dup_cursor,
429 .set_root = xfs_allocbt_set_root,
430 .alloc_block = xfs_allocbt_alloc_block,
431 .free_block = xfs_allocbt_free_block,
432 .update_lastrec = xfs_allocbt_update_lastrec,
433 .get_minrecs = xfs_allocbt_get_minrecs,
434 .get_maxrecs = xfs_allocbt_get_maxrecs,
435 .init_key_from_rec = xfs_allocbt_init_key_from_rec,
436 .init_high_key_from_rec = xfs_bnobt_init_high_key_from_rec,
437 .init_rec_from_cur = xfs_allocbt_init_rec_from_cur,
438 .init_ptr_from_cur = xfs_allocbt_init_ptr_from_cur,
439 .key_diff = xfs_bnobt_key_diff,
440 .buf_ops = &xfs_bnobt_buf_ops,
441 .diff_two_keys = xfs_bnobt_diff_two_keys,
442 .keys_inorder = xfs_bnobt_keys_inorder,
443 .recs_inorder = xfs_bnobt_recs_inorder,
446 static const struct xfs_btree_ops xfs_cntbt_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_cntbt_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_cntbt_key_diff,
462 .buf_ops = &xfs_cntbt_buf_ops,
463 .diff_two_keys = xfs_cntbt_diff_two_keys,
464 .keys_inorder = xfs_cntbt_keys_inorder,
465 .recs_inorder = xfs_cntbt_recs_inorder,
468 /* Allocate most of a new allocation btree cursor. */
469 STATIC struct xfs_btree_cur *
470 xfs_allocbt_init_common(
471 struct xfs_mount *mp,
472 struct xfs_trans *tp,
473 struct xfs_perag *pag,
476 struct xfs_btree_cur *cur;
478 ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
480 cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
484 cur->bc_btnum = btnum;
485 cur->bc_blocklog = mp->m_sb.sb_blocklog;
486 cur->bc_ag.abt.active = false;
488 if (btnum == XFS_BTNUM_CNT) {
489 cur->bc_ops = &xfs_cntbt_ops;
490 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
491 cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
493 cur->bc_ops = &xfs_bnobt_ops;
494 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
497 /* take a reference for the cursor */
498 atomic_inc(&pag->pag_ref);
499 cur->bc_ag.pag = pag;
501 if (xfs_sb_version_hascrc(&mp->m_sb))
502 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
508 * Allocate a new allocation btree cursor.
510 struct xfs_btree_cur * /* new alloc btree cursor */
511 xfs_allocbt_init_cursor(
512 struct xfs_mount *mp, /* file system mount point */
513 struct xfs_trans *tp, /* transaction pointer */
514 struct xfs_buf *agbp, /* buffer for agf structure */
515 struct xfs_perag *pag,
516 xfs_btnum_t btnum) /* btree identifier */
518 struct xfs_agf *agf = agbp->b_addr;
519 struct xfs_btree_cur *cur;
521 cur = xfs_allocbt_init_common(mp, tp, pag, btnum);
522 if (btnum == XFS_BTNUM_CNT)
523 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
525 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
527 cur->bc_ag.agbp = agbp;
532 /* Create a free space btree cursor with a fake root for staging. */
533 struct xfs_btree_cur *
534 xfs_allocbt_stage_cursor(
535 struct xfs_mount *mp,
536 struct xbtree_afakeroot *afake,
537 struct xfs_perag *pag,
540 struct xfs_btree_cur *cur;
542 cur = xfs_allocbt_init_common(mp, NULL, pag, btnum);
543 xfs_btree_stage_afakeroot(cur, afake);
548 * Install a new free space btree root. Caller is responsible for invalidating
549 * and freeing the old btree blocks.
552 xfs_allocbt_commit_staged_btree(
553 struct xfs_btree_cur *cur,
554 struct xfs_trans *tp,
555 struct xfs_buf *agbp)
557 struct xfs_agf *agf = agbp->b_addr;
558 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
560 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
562 agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root);
563 agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels);
564 xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
566 if (cur->bc_btnum == XFS_BTNUM_BNO) {
567 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_bnobt_ops);
569 cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE;
570 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_cntbt_ops);
575 * Calculate number of records in an alloc btree block.
579 struct xfs_mount *mp,
583 blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
586 return blocklen / sizeof(xfs_alloc_rec_t);
587 return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
590 /* Calculate the freespace btree size for some records. */
592 xfs_allocbt_calc_size(
593 struct xfs_mount *mp,
594 unsigned long long len)
596 return xfs_btree_calc_size(mp->m_alloc_mnr, len);