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"
13 #include "xfs_mount.h"
14 #include "xfs_btree.h"
15 #include "xfs_btree_staging.h"
16 #include "xfs_ialloc.h"
17 #include "xfs_ialloc_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
25 xfs_inobt_get_minrecs(
26 struct xfs_btree_cur *cur,
29 return M_IGEO(cur->bc_mp)->inobt_mnr[level != 0];
32 STATIC struct xfs_btree_cur *
34 struct xfs_btree_cur *cur)
36 return xfs_inobt_init_cursor(cur->bc_mp, cur->bc_tp,
37 cur->bc_ag.agbp, cur->bc_ag.agno,
43 struct xfs_btree_cur *cur,
44 union xfs_btree_ptr *nptr,
45 int inc) /* level change */
47 struct xfs_buf *agbp = cur->bc_ag.agbp;
48 struct xfs_agi *agi = agbp->b_addr;
50 agi->agi_root = nptr->s;
51 be32_add_cpu(&agi->agi_level, inc);
52 xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_ROOT | XFS_AGI_LEVEL);
57 struct xfs_btree_cur *cur,
58 union xfs_btree_ptr *nptr,
59 int inc) /* level change */
61 struct xfs_buf *agbp = cur->bc_ag.agbp;
62 struct xfs_agi *agi = agbp->b_addr;
64 agi->agi_free_root = nptr->s;
65 be32_add_cpu(&agi->agi_free_level, inc);
66 xfs_ialloc_log_agi(cur->bc_tp, agbp,
67 XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL);
71 __xfs_inobt_alloc_block(
72 struct xfs_btree_cur *cur,
73 union xfs_btree_ptr *start,
74 union xfs_btree_ptr *new,
76 enum xfs_ag_resv_type resv)
78 xfs_alloc_arg_t args; /* block allocation args */
79 int error; /* error return value */
80 xfs_agblock_t sbno = be32_to_cpu(start->s);
82 memset(&args, 0, sizeof(args));
85 args.oinfo = XFS_RMAP_OINFO_INOBT;
86 args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_ag.agno, sbno);
90 args.type = XFS_ALLOCTYPE_NEAR_BNO;
93 error = xfs_alloc_vextent(&args);
97 if (args.fsbno == NULLFSBLOCK) {
101 ASSERT(args.len == 1);
103 new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
109 xfs_inobt_alloc_block(
110 struct xfs_btree_cur *cur,
111 union xfs_btree_ptr *start,
112 union xfs_btree_ptr *new,
115 return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
119 xfs_finobt_alloc_block(
120 struct xfs_btree_cur *cur,
121 union xfs_btree_ptr *start,
122 union xfs_btree_ptr *new,
125 if (cur->bc_mp->m_finobt_nores)
126 return xfs_inobt_alloc_block(cur, start, new, stat);
127 return __xfs_inobt_alloc_block(cur, start, new, stat,
128 XFS_AG_RESV_METADATA);
132 __xfs_inobt_free_block(
133 struct xfs_btree_cur *cur,
135 enum xfs_ag_resv_type resv)
137 return xfs_free_extent(cur->bc_tp,
138 XFS_DADDR_TO_FSB(cur->bc_mp, XFS_BUF_ADDR(bp)), 1,
139 &XFS_RMAP_OINFO_INOBT, resv);
143 xfs_inobt_free_block(
144 struct xfs_btree_cur *cur,
147 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
151 xfs_finobt_free_block(
152 struct xfs_btree_cur *cur,
155 if (cur->bc_mp->m_finobt_nores)
156 return xfs_inobt_free_block(cur, bp);
157 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_METADATA);
161 xfs_inobt_get_maxrecs(
162 struct xfs_btree_cur *cur,
165 return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
169 xfs_inobt_init_key_from_rec(
170 union xfs_btree_key *key,
171 union xfs_btree_rec *rec)
173 key->inobt.ir_startino = rec->inobt.ir_startino;
177 xfs_inobt_init_high_key_from_rec(
178 union xfs_btree_key *key,
179 union xfs_btree_rec *rec)
183 x = be32_to_cpu(rec->inobt.ir_startino);
184 x += XFS_INODES_PER_CHUNK - 1;
185 key->inobt.ir_startino = cpu_to_be32(x);
189 xfs_inobt_init_rec_from_cur(
190 struct xfs_btree_cur *cur,
191 union xfs_btree_rec *rec)
193 rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
194 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
195 rec->inobt.ir_u.sp.ir_holemask =
196 cpu_to_be16(cur->bc_rec.i.ir_holemask);
197 rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
198 rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
200 /* ir_holemask/ir_count not supported on-disk */
201 rec->inobt.ir_u.f.ir_freecount =
202 cpu_to_be32(cur->bc_rec.i.ir_freecount);
204 rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
208 * initial value of ptr for lookup
211 xfs_inobt_init_ptr_from_cur(
212 struct xfs_btree_cur *cur,
213 union xfs_btree_ptr *ptr)
215 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
217 ASSERT(cur->bc_ag.agno == be32_to_cpu(agi->agi_seqno));
219 ptr->s = agi->agi_root;
223 xfs_finobt_init_ptr_from_cur(
224 struct xfs_btree_cur *cur,
225 union xfs_btree_ptr *ptr)
227 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
229 ASSERT(cur->bc_ag.agno == be32_to_cpu(agi->agi_seqno));
230 ptr->s = agi->agi_free_root;
235 struct xfs_btree_cur *cur,
236 union xfs_btree_key *key)
238 return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
239 cur->bc_rec.i.ir_startino;
243 xfs_inobt_diff_two_keys(
244 struct xfs_btree_cur *cur,
245 union xfs_btree_key *k1,
246 union xfs_btree_key *k2)
248 return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
249 be32_to_cpu(k2->inobt.ir_startino);
252 static xfs_failaddr_t
256 struct xfs_mount *mp = bp->b_mount;
257 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
261 if (!xfs_verify_magic(bp, block->bb_magic))
262 return __this_address;
265 * During growfs operations, we can't verify the exact owner as the
266 * perag is not fully initialised and hence not attached to the buffer.
268 * Similarly, during log recovery we will have a perag structure
269 * attached, but the agi information will not yet have been initialised
270 * from the on disk AGI. We don't currently use any of this information,
271 * but beware of the landmine (i.e. need to check pag->pagi_init) if we
274 if (xfs_sb_version_hascrc(&mp->m_sb)) {
275 fa = xfs_btree_sblock_v5hdr_verify(bp);
280 /* level verification */
281 level = be16_to_cpu(block->bb_level);
282 if (level >= M_IGEO(mp)->inobt_maxlevels)
283 return __this_address;
285 return xfs_btree_sblock_verify(bp,
286 M_IGEO(mp)->inobt_mxr[level != 0]);
290 xfs_inobt_read_verify(
295 if (!xfs_btree_sblock_verify_crc(bp))
296 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
298 fa = xfs_inobt_verify(bp);
300 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
304 trace_xfs_btree_corrupt(bp, _RET_IP_);
308 xfs_inobt_write_verify(
313 fa = xfs_inobt_verify(bp);
315 trace_xfs_btree_corrupt(bp, _RET_IP_);
316 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
319 xfs_btree_sblock_calc_crc(bp);
323 const struct xfs_buf_ops xfs_inobt_buf_ops = {
325 .magic = { cpu_to_be32(XFS_IBT_MAGIC), cpu_to_be32(XFS_IBT_CRC_MAGIC) },
326 .verify_read = xfs_inobt_read_verify,
327 .verify_write = xfs_inobt_write_verify,
328 .verify_struct = xfs_inobt_verify,
331 const struct xfs_buf_ops xfs_finobt_buf_ops = {
332 .name = "xfs_finobt",
333 .magic = { cpu_to_be32(XFS_FIBT_MAGIC),
334 cpu_to_be32(XFS_FIBT_CRC_MAGIC) },
335 .verify_read = xfs_inobt_read_verify,
336 .verify_write = xfs_inobt_write_verify,
337 .verify_struct = xfs_inobt_verify,
341 xfs_inobt_keys_inorder(
342 struct xfs_btree_cur *cur,
343 union xfs_btree_key *k1,
344 union xfs_btree_key *k2)
346 return be32_to_cpu(k1->inobt.ir_startino) <
347 be32_to_cpu(k2->inobt.ir_startino);
351 xfs_inobt_recs_inorder(
352 struct xfs_btree_cur *cur,
353 union xfs_btree_rec *r1,
354 union xfs_btree_rec *r2)
356 return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
357 be32_to_cpu(r2->inobt.ir_startino);
360 static const struct xfs_btree_ops xfs_inobt_ops = {
361 .rec_len = sizeof(xfs_inobt_rec_t),
362 .key_len = sizeof(xfs_inobt_key_t),
364 .dup_cursor = xfs_inobt_dup_cursor,
365 .set_root = xfs_inobt_set_root,
366 .alloc_block = xfs_inobt_alloc_block,
367 .free_block = xfs_inobt_free_block,
368 .get_minrecs = xfs_inobt_get_minrecs,
369 .get_maxrecs = xfs_inobt_get_maxrecs,
370 .init_key_from_rec = xfs_inobt_init_key_from_rec,
371 .init_high_key_from_rec = xfs_inobt_init_high_key_from_rec,
372 .init_rec_from_cur = xfs_inobt_init_rec_from_cur,
373 .init_ptr_from_cur = xfs_inobt_init_ptr_from_cur,
374 .key_diff = xfs_inobt_key_diff,
375 .buf_ops = &xfs_inobt_buf_ops,
376 .diff_two_keys = xfs_inobt_diff_two_keys,
377 .keys_inorder = xfs_inobt_keys_inorder,
378 .recs_inorder = xfs_inobt_recs_inorder,
381 static const struct xfs_btree_ops xfs_finobt_ops = {
382 .rec_len = sizeof(xfs_inobt_rec_t),
383 .key_len = sizeof(xfs_inobt_key_t),
385 .dup_cursor = xfs_inobt_dup_cursor,
386 .set_root = xfs_finobt_set_root,
387 .alloc_block = xfs_finobt_alloc_block,
388 .free_block = xfs_finobt_free_block,
389 .get_minrecs = xfs_inobt_get_minrecs,
390 .get_maxrecs = xfs_inobt_get_maxrecs,
391 .init_key_from_rec = xfs_inobt_init_key_from_rec,
392 .init_high_key_from_rec = xfs_inobt_init_high_key_from_rec,
393 .init_rec_from_cur = xfs_inobt_init_rec_from_cur,
394 .init_ptr_from_cur = xfs_finobt_init_ptr_from_cur,
395 .key_diff = xfs_inobt_key_diff,
396 .buf_ops = &xfs_finobt_buf_ops,
397 .diff_two_keys = xfs_inobt_diff_two_keys,
398 .keys_inorder = xfs_inobt_keys_inorder,
399 .recs_inorder = xfs_inobt_recs_inorder,
403 * Initialize a new inode btree cursor.
405 static struct xfs_btree_cur *
406 xfs_inobt_init_common(
407 struct xfs_mount *mp, /* file system mount point */
408 struct xfs_trans *tp, /* transaction pointer */
409 xfs_agnumber_t agno, /* allocation group number */
410 xfs_btnum_t btnum) /* ialloc or free ino btree */
412 struct xfs_btree_cur *cur;
414 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
417 cur->bc_btnum = btnum;
418 if (btnum == XFS_BTNUM_INO) {
419 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_ibt_2);
420 cur->bc_ops = &xfs_inobt_ops;
422 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_fibt_2);
423 cur->bc_ops = &xfs_finobt_ops;
426 cur->bc_blocklog = mp->m_sb.sb_blocklog;
428 if (xfs_sb_version_hascrc(&mp->m_sb))
429 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
431 cur->bc_ag.agno = agno;
435 /* Create an inode btree cursor. */
436 struct xfs_btree_cur *
437 xfs_inobt_init_cursor(
438 struct xfs_mount *mp,
439 struct xfs_trans *tp,
440 struct xfs_buf *agbp,
444 struct xfs_btree_cur *cur;
445 struct xfs_agi *agi = agbp->b_addr;
447 cur = xfs_inobt_init_common(mp, tp, agno, btnum);
448 if (btnum == XFS_BTNUM_INO)
449 cur->bc_nlevels = be32_to_cpu(agi->agi_level);
451 cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
452 cur->bc_ag.agbp = agbp;
456 /* Create an inode btree cursor with a fake root for staging. */
457 struct xfs_btree_cur *
458 xfs_inobt_stage_cursor(
459 struct xfs_mount *mp,
460 struct xbtree_afakeroot *afake,
464 struct xfs_btree_cur *cur;
466 cur = xfs_inobt_init_common(mp, NULL, agno, btnum);
467 xfs_btree_stage_afakeroot(cur, afake);
472 * Install a new inobt btree root. Caller is responsible for invalidating
473 * and freeing the old btree blocks.
476 xfs_inobt_commit_staged_btree(
477 struct xfs_btree_cur *cur,
478 struct xfs_trans *tp,
479 struct xfs_buf *agbp)
481 struct xfs_agi *agi = agbp->b_addr;
482 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
484 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
486 if (cur->bc_btnum == XFS_BTNUM_INO) {
487 agi->agi_root = cpu_to_be32(afake->af_root);
488 agi->agi_level = cpu_to_be32(afake->af_levels);
489 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_ROOT | XFS_AGI_LEVEL);
490 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_inobt_ops);
492 agi->agi_free_root = cpu_to_be32(afake->af_root);
493 agi->agi_free_level = cpu_to_be32(afake->af_levels);
494 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREE_ROOT |
496 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_finobt_ops);
501 * Calculate number of records in an inobt btree block.
505 struct xfs_mount *mp,
509 blocklen -= XFS_INOBT_BLOCK_LEN(mp);
512 return blocklen / sizeof(xfs_inobt_rec_t);
513 return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
517 * Convert the inode record holemask to an inode allocation bitmap. The inode
518 * allocation bitmap is inode granularity and specifies whether an inode is
519 * physically allocated on disk (not whether the inode is considered allocated
520 * or free by the fs).
522 * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
525 xfs_inobt_irec_to_allocmask(
526 struct xfs_inobt_rec_incore *rec)
534 * The holemask has 16-bits for a 64 inode record. Therefore each
535 * holemask bit represents multiple inodes. Create a mask of bits to set
536 * in the allocmask for each holemask bit.
538 inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
541 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
542 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
543 * anything beyond the 16 holemask bits since this casts to a larger
546 allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
549 * allocbitmap is the inverted holemask so every set bit represents
550 * allocated inodes. To expand from 16-bit holemask granularity to
551 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
552 * bitmap for every holemask bit.
554 nextbit = xfs_next_bit(&allocbitmap, 1, 0);
555 while (nextbit != -1) {
556 ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
558 bitmap |= (inodespbit <<
559 (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
561 nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
567 #if defined(DEBUG) || defined(XFS_WARN)
569 * Verify that an in-core inode record has a valid inode count.
572 xfs_inobt_rec_check_count(
573 struct xfs_mount *mp,
574 struct xfs_inobt_rec_incore *rec)
581 wordsz = sizeof(allocbmap) / sizeof(unsigned int);
582 allocbmap = xfs_inobt_irec_to_allocmask(rec);
584 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
585 while (nextbit != -1) {
587 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
591 if (inocount != rec->ir_count)
592 return -EFSCORRUPTED;
600 struct xfs_mount *mp,
603 xfs_agblock_t agblocks = xfs_ag_block_count(mp, agno);
605 /* Bail out if we're uninitialized, which can happen in mkfs. */
606 if (M_IGEO(mp)->inobt_mxr[0] == 0)
610 * The log is permanently allocated, so the space it occupies will
611 * never be available for the kinds of things that would require btree
612 * expansion. We therefore can pretend the space isn't there.
614 if (mp->m_sb.sb_logstart &&
615 XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart) == agno)
616 agblocks -= mp->m_sb.sb_logblocks;
618 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr,
619 (uint64_t)agblocks * mp->m_sb.sb_inopblock /
620 XFS_INODES_PER_CHUNK);
623 /* Read AGI and create inobt cursor. */
626 struct xfs_mount *mp,
627 struct xfs_trans *tp,
630 struct xfs_btree_cur **curpp,
631 struct xfs_buf **agi_bpp)
633 struct xfs_btree_cur *cur;
636 ASSERT(*agi_bpp == NULL);
637 ASSERT(*curpp == NULL);
639 error = xfs_ialloc_read_agi(mp, tp, agno, agi_bpp);
643 cur = xfs_inobt_init_cursor(mp, tp, *agi_bpp, agno, which);
645 xfs_trans_brelse(tp, *agi_bpp);
654 xfs_inobt_count_blocks(
655 struct xfs_mount *mp,
656 struct xfs_trans *tp,
659 xfs_extlen_t *tree_blocks)
661 struct xfs_buf *agbp = NULL;
662 struct xfs_btree_cur *cur = NULL;
665 error = xfs_inobt_cur(mp, tp, agno, btnum, &cur, &agbp);
669 error = xfs_btree_count_blocks(cur, tree_blocks);
670 xfs_btree_del_cursor(cur, error);
671 xfs_trans_brelse(tp, agbp);
677 * Figure out how many blocks to reserve and how many are used by this btree.
680 xfs_finobt_calc_reserves(
681 struct xfs_mount *mp,
682 struct xfs_trans *tp,
687 xfs_extlen_t tree_len = 0;
690 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
693 error = xfs_inobt_count_blocks(mp, tp, agno, XFS_BTNUM_FINO, &tree_len);
697 *ask += xfs_inobt_max_size(mp, agno);
702 /* Calculate the inobt btree size for some records. */
704 xfs_iallocbt_calc_size(
705 struct xfs_mount *mp,
706 unsigned long long len)
708 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);