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"
26 xfs_inobt_get_minrecs(
27 struct xfs_btree_cur *cur,
30 return M_IGEO(cur->bc_mp)->inobt_mnr[level != 0];
33 STATIC struct xfs_btree_cur *
35 struct xfs_btree_cur *cur)
37 return xfs_inobt_init_cursor(cur->bc_mp, cur->bc_tp,
38 cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
43 struct xfs_btree_cur *cur,
44 const 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 const 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);
70 /* Update the inode btree block counter for this btree. */
72 xfs_inobt_mod_blockcount(
73 struct xfs_btree_cur *cur,
76 struct xfs_buf *agbp = cur->bc_ag.agbp;
77 struct xfs_agi *agi = agbp->b_addr;
79 if (!xfs_has_inobtcounts(cur->bc_mp))
82 if (cur->bc_btnum == XFS_BTNUM_FINO)
83 be32_add_cpu(&agi->agi_fblocks, howmuch);
84 else if (cur->bc_btnum == XFS_BTNUM_INO)
85 be32_add_cpu(&agi->agi_iblocks, howmuch);
86 xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_IBLOCKS);
90 __xfs_inobt_alloc_block(
91 struct xfs_btree_cur *cur,
92 const union xfs_btree_ptr *start,
93 union xfs_btree_ptr *new,
95 enum xfs_ag_resv_type resv)
97 xfs_alloc_arg_t args; /* block allocation args */
98 int error; /* error return value */
99 xfs_agblock_t sbno = be32_to_cpu(start->s);
101 memset(&args, 0, sizeof(args));
102 args.tp = cur->bc_tp;
103 args.mp = cur->bc_mp;
104 args.oinfo = XFS_RMAP_OINFO_INOBT;
105 args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_ag.pag->pag_agno, sbno);
109 args.type = XFS_ALLOCTYPE_NEAR_BNO;
112 error = xfs_alloc_vextent(&args);
116 if (args.fsbno == NULLFSBLOCK) {
120 ASSERT(args.len == 1);
122 new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
124 xfs_inobt_mod_blockcount(cur, 1);
129 xfs_inobt_alloc_block(
130 struct xfs_btree_cur *cur,
131 const union xfs_btree_ptr *start,
132 union xfs_btree_ptr *new,
135 return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
139 xfs_finobt_alloc_block(
140 struct xfs_btree_cur *cur,
141 const union xfs_btree_ptr *start,
142 union xfs_btree_ptr *new,
145 if (cur->bc_mp->m_finobt_nores)
146 return xfs_inobt_alloc_block(cur, start, new, stat);
147 return __xfs_inobt_alloc_block(cur, start, new, stat,
148 XFS_AG_RESV_METADATA);
152 __xfs_inobt_free_block(
153 struct xfs_btree_cur *cur,
155 enum xfs_ag_resv_type resv)
157 xfs_inobt_mod_blockcount(cur, -1);
158 return xfs_free_extent(cur->bc_tp,
159 XFS_DADDR_TO_FSB(cur->bc_mp, xfs_buf_daddr(bp)), 1,
160 &XFS_RMAP_OINFO_INOBT, resv);
164 xfs_inobt_free_block(
165 struct xfs_btree_cur *cur,
168 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
172 xfs_finobt_free_block(
173 struct xfs_btree_cur *cur,
176 if (cur->bc_mp->m_finobt_nores)
177 return xfs_inobt_free_block(cur, bp);
178 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_METADATA);
182 xfs_inobt_get_maxrecs(
183 struct xfs_btree_cur *cur,
186 return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
190 xfs_inobt_init_key_from_rec(
191 union xfs_btree_key *key,
192 const union xfs_btree_rec *rec)
194 key->inobt.ir_startino = rec->inobt.ir_startino;
198 xfs_inobt_init_high_key_from_rec(
199 union xfs_btree_key *key,
200 const union xfs_btree_rec *rec)
204 x = be32_to_cpu(rec->inobt.ir_startino);
205 x += XFS_INODES_PER_CHUNK - 1;
206 key->inobt.ir_startino = cpu_to_be32(x);
210 xfs_inobt_init_rec_from_cur(
211 struct xfs_btree_cur *cur,
212 union xfs_btree_rec *rec)
214 rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
215 if (xfs_has_sparseinodes(cur->bc_mp)) {
216 rec->inobt.ir_u.sp.ir_holemask =
217 cpu_to_be16(cur->bc_rec.i.ir_holemask);
218 rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
219 rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
221 /* ir_holemask/ir_count not supported on-disk */
222 rec->inobt.ir_u.f.ir_freecount =
223 cpu_to_be32(cur->bc_rec.i.ir_freecount);
225 rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
229 * initial value of ptr for lookup
232 xfs_inobt_init_ptr_from_cur(
233 struct xfs_btree_cur *cur,
234 union xfs_btree_ptr *ptr)
236 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
238 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
240 ptr->s = agi->agi_root;
244 xfs_finobt_init_ptr_from_cur(
245 struct xfs_btree_cur *cur,
246 union xfs_btree_ptr *ptr)
248 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
250 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
251 ptr->s = agi->agi_free_root;
256 struct xfs_btree_cur *cur,
257 const union xfs_btree_key *key)
259 return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
260 cur->bc_rec.i.ir_startino;
264 xfs_inobt_diff_two_keys(
265 struct xfs_btree_cur *cur,
266 const union xfs_btree_key *k1,
267 const union xfs_btree_key *k2)
269 return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
270 be32_to_cpu(k2->inobt.ir_startino);
273 static xfs_failaddr_t
277 struct xfs_mount *mp = bp->b_mount;
278 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
282 if (!xfs_verify_magic(bp, block->bb_magic))
283 return __this_address;
286 * During growfs operations, we can't verify the exact owner as the
287 * perag is not fully initialised and hence not attached to the buffer.
289 * Similarly, during log recovery we will have a perag structure
290 * attached, but the agi information will not yet have been initialised
291 * from the on disk AGI. We don't currently use any of this information,
292 * but beware of the landmine (i.e. need to check pag->pagi_init) if we
295 if (xfs_has_crc(mp)) {
296 fa = xfs_btree_sblock_v5hdr_verify(bp);
301 /* level verification */
302 level = be16_to_cpu(block->bb_level);
303 if (level >= M_IGEO(mp)->inobt_maxlevels)
304 return __this_address;
306 return xfs_btree_sblock_verify(bp,
307 M_IGEO(mp)->inobt_mxr[level != 0]);
311 xfs_inobt_read_verify(
316 if (!xfs_btree_sblock_verify_crc(bp))
317 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
319 fa = xfs_inobt_verify(bp);
321 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
325 trace_xfs_btree_corrupt(bp, _RET_IP_);
329 xfs_inobt_write_verify(
334 fa = xfs_inobt_verify(bp);
336 trace_xfs_btree_corrupt(bp, _RET_IP_);
337 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
340 xfs_btree_sblock_calc_crc(bp);
344 const struct xfs_buf_ops xfs_inobt_buf_ops = {
346 .magic = { cpu_to_be32(XFS_IBT_MAGIC), cpu_to_be32(XFS_IBT_CRC_MAGIC) },
347 .verify_read = xfs_inobt_read_verify,
348 .verify_write = xfs_inobt_write_verify,
349 .verify_struct = xfs_inobt_verify,
352 const struct xfs_buf_ops xfs_finobt_buf_ops = {
353 .name = "xfs_finobt",
354 .magic = { cpu_to_be32(XFS_FIBT_MAGIC),
355 cpu_to_be32(XFS_FIBT_CRC_MAGIC) },
356 .verify_read = xfs_inobt_read_verify,
357 .verify_write = xfs_inobt_write_verify,
358 .verify_struct = xfs_inobt_verify,
362 xfs_inobt_keys_inorder(
363 struct xfs_btree_cur *cur,
364 const union xfs_btree_key *k1,
365 const union xfs_btree_key *k2)
367 return be32_to_cpu(k1->inobt.ir_startino) <
368 be32_to_cpu(k2->inobt.ir_startino);
372 xfs_inobt_recs_inorder(
373 struct xfs_btree_cur *cur,
374 const union xfs_btree_rec *r1,
375 const union xfs_btree_rec *r2)
377 return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
378 be32_to_cpu(r2->inobt.ir_startino);
381 static const struct xfs_btree_ops xfs_inobt_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_inobt_set_root,
387 .alloc_block = xfs_inobt_alloc_block,
388 .free_block = xfs_inobt_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_inobt_init_ptr_from_cur,
395 .key_diff = xfs_inobt_key_diff,
396 .buf_ops = &xfs_inobt_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,
402 static const struct xfs_btree_ops xfs_finobt_ops = {
403 .rec_len = sizeof(xfs_inobt_rec_t),
404 .key_len = sizeof(xfs_inobt_key_t),
406 .dup_cursor = xfs_inobt_dup_cursor,
407 .set_root = xfs_finobt_set_root,
408 .alloc_block = xfs_finobt_alloc_block,
409 .free_block = xfs_finobt_free_block,
410 .get_minrecs = xfs_inobt_get_minrecs,
411 .get_maxrecs = xfs_inobt_get_maxrecs,
412 .init_key_from_rec = xfs_inobt_init_key_from_rec,
413 .init_high_key_from_rec = xfs_inobt_init_high_key_from_rec,
414 .init_rec_from_cur = xfs_inobt_init_rec_from_cur,
415 .init_ptr_from_cur = xfs_finobt_init_ptr_from_cur,
416 .key_diff = xfs_inobt_key_diff,
417 .buf_ops = &xfs_finobt_buf_ops,
418 .diff_two_keys = xfs_inobt_diff_two_keys,
419 .keys_inorder = xfs_inobt_keys_inorder,
420 .recs_inorder = xfs_inobt_recs_inorder,
424 * Initialize a new inode btree cursor.
426 static struct xfs_btree_cur *
427 xfs_inobt_init_common(
428 struct xfs_mount *mp, /* file system mount point */
429 struct xfs_trans *tp, /* transaction pointer */
430 struct xfs_perag *pag,
431 xfs_btnum_t btnum) /* ialloc or free ino btree */
433 struct xfs_btree_cur *cur;
435 cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
438 cur->bc_btnum = btnum;
439 if (btnum == XFS_BTNUM_INO) {
440 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_ibt_2);
441 cur->bc_ops = &xfs_inobt_ops;
443 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_fibt_2);
444 cur->bc_ops = &xfs_finobt_ops;
447 cur->bc_blocklog = mp->m_sb.sb_blocklog;
450 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
452 /* take a reference for the cursor */
453 atomic_inc(&pag->pag_ref);
454 cur->bc_ag.pag = pag;
458 /* Create an inode btree cursor. */
459 struct xfs_btree_cur *
460 xfs_inobt_init_cursor(
461 struct xfs_mount *mp,
462 struct xfs_trans *tp,
463 struct xfs_buf *agbp,
464 struct xfs_perag *pag,
467 struct xfs_btree_cur *cur;
468 struct xfs_agi *agi = agbp->b_addr;
470 cur = xfs_inobt_init_common(mp, tp, pag, btnum);
471 if (btnum == XFS_BTNUM_INO)
472 cur->bc_nlevels = be32_to_cpu(agi->agi_level);
474 cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
475 cur->bc_ag.agbp = agbp;
479 /* Create an inode btree cursor with a fake root for staging. */
480 struct xfs_btree_cur *
481 xfs_inobt_stage_cursor(
482 struct xfs_mount *mp,
483 struct xbtree_afakeroot *afake,
484 struct xfs_perag *pag,
487 struct xfs_btree_cur *cur;
489 cur = xfs_inobt_init_common(mp, NULL, pag, btnum);
490 xfs_btree_stage_afakeroot(cur, afake);
495 * Install a new inobt btree root. Caller is responsible for invalidating
496 * and freeing the old btree blocks.
499 xfs_inobt_commit_staged_btree(
500 struct xfs_btree_cur *cur,
501 struct xfs_trans *tp,
502 struct xfs_buf *agbp)
504 struct xfs_agi *agi = agbp->b_addr;
505 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
508 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
510 if (cur->bc_btnum == XFS_BTNUM_INO) {
511 fields = XFS_AGI_ROOT | XFS_AGI_LEVEL;
512 agi->agi_root = cpu_to_be32(afake->af_root);
513 agi->agi_level = cpu_to_be32(afake->af_levels);
514 if (xfs_has_inobtcounts(cur->bc_mp)) {
515 agi->agi_iblocks = cpu_to_be32(afake->af_blocks);
516 fields |= XFS_AGI_IBLOCKS;
518 xfs_ialloc_log_agi(tp, agbp, fields);
519 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_inobt_ops);
521 fields = XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL;
522 agi->agi_free_root = cpu_to_be32(afake->af_root);
523 agi->agi_free_level = cpu_to_be32(afake->af_levels);
524 if (xfs_has_inobtcounts(cur->bc_mp)) {
525 agi->agi_fblocks = cpu_to_be32(afake->af_blocks);
526 fields |= XFS_AGI_IBLOCKS;
528 xfs_ialloc_log_agi(tp, agbp, fields);
529 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_finobt_ops);
534 * Calculate number of records in an inobt btree block.
538 struct xfs_mount *mp,
542 blocklen -= XFS_INOBT_BLOCK_LEN(mp);
545 return blocklen / sizeof(xfs_inobt_rec_t);
546 return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
550 * Convert the inode record holemask to an inode allocation bitmap. The inode
551 * allocation bitmap is inode granularity and specifies whether an inode is
552 * physically allocated on disk (not whether the inode is considered allocated
553 * or free by the fs).
555 * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
558 xfs_inobt_irec_to_allocmask(
559 struct xfs_inobt_rec_incore *rec)
567 * The holemask has 16-bits for a 64 inode record. Therefore each
568 * holemask bit represents multiple inodes. Create a mask of bits to set
569 * in the allocmask for each holemask bit.
571 inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
574 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
575 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
576 * anything beyond the 16 holemask bits since this casts to a larger
579 allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
582 * allocbitmap is the inverted holemask so every set bit represents
583 * allocated inodes. To expand from 16-bit holemask granularity to
584 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
585 * bitmap for every holemask bit.
587 nextbit = xfs_next_bit(&allocbitmap, 1, 0);
588 while (nextbit != -1) {
589 ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
591 bitmap |= (inodespbit <<
592 (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
594 nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
600 #if defined(DEBUG) || defined(XFS_WARN)
602 * Verify that an in-core inode record has a valid inode count.
605 xfs_inobt_rec_check_count(
606 struct xfs_mount *mp,
607 struct xfs_inobt_rec_incore *rec)
614 wordsz = sizeof(allocbmap) / sizeof(unsigned int);
615 allocbmap = xfs_inobt_irec_to_allocmask(rec);
617 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
618 while (nextbit != -1) {
620 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
624 if (inocount != rec->ir_count)
625 return -EFSCORRUPTED;
633 struct xfs_mount *mp,
636 xfs_agblock_t agblocks = xfs_ag_block_count(mp, agno);
638 /* Bail out if we're uninitialized, which can happen in mkfs. */
639 if (M_IGEO(mp)->inobt_mxr[0] == 0)
643 * The log is permanently allocated, so the space it occupies will
644 * never be available for the kinds of things that would require btree
645 * expansion. We therefore can pretend the space isn't there.
647 if (mp->m_sb.sb_logstart &&
648 XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart) == agno)
649 agblocks -= mp->m_sb.sb_logblocks;
651 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr,
652 (uint64_t)agblocks * mp->m_sb.sb_inopblock /
653 XFS_INODES_PER_CHUNK);
656 /* Read AGI and create inobt cursor. */
659 struct xfs_mount *mp,
660 struct xfs_trans *tp,
661 struct xfs_perag *pag,
663 struct xfs_btree_cur **curpp,
664 struct xfs_buf **agi_bpp)
666 struct xfs_btree_cur *cur;
669 ASSERT(*agi_bpp == NULL);
670 ASSERT(*curpp == NULL);
672 error = xfs_ialloc_read_agi(mp, tp, pag->pag_agno, agi_bpp);
676 cur = xfs_inobt_init_cursor(mp, tp, *agi_bpp, pag, which);
682 xfs_inobt_count_blocks(
683 struct xfs_mount *mp,
684 struct xfs_trans *tp,
685 struct xfs_perag *pag,
687 xfs_extlen_t *tree_blocks)
689 struct xfs_buf *agbp = NULL;
690 struct xfs_btree_cur *cur = NULL;
693 error = xfs_inobt_cur(mp, tp, pag, btnum, &cur, &agbp);
697 error = xfs_btree_count_blocks(cur, tree_blocks);
698 xfs_btree_del_cursor(cur, error);
699 xfs_trans_brelse(tp, agbp);
704 /* Read finobt block count from AGI header. */
706 xfs_finobt_read_blocks(
707 struct xfs_mount *mp,
708 struct xfs_trans *tp,
709 struct xfs_perag *pag,
710 xfs_extlen_t *tree_blocks)
712 struct xfs_buf *agbp;
716 error = xfs_ialloc_read_agi(mp, tp, pag->pag_agno, &agbp);
721 *tree_blocks = be32_to_cpu(agi->agi_fblocks);
722 xfs_trans_brelse(tp, agbp);
727 * Figure out how many blocks to reserve and how many are used by this btree.
730 xfs_finobt_calc_reserves(
731 struct xfs_mount *mp,
732 struct xfs_trans *tp,
733 struct xfs_perag *pag,
737 xfs_extlen_t tree_len = 0;
740 if (!xfs_has_finobt(mp))
743 if (xfs_has_inobtcounts(mp))
744 error = xfs_finobt_read_blocks(mp, tp, pag, &tree_len);
746 error = xfs_inobt_count_blocks(mp, tp, pag, XFS_BTNUM_FINO,
751 *ask += xfs_inobt_max_size(mp, pag->pag_agno);
756 /* Calculate the inobt btree size for some records. */
758 xfs_iallocbt_calc_size(
759 struct xfs_mount *mp,
760 unsigned long long len)
762 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);