Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
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_bit.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"
22#include "xfs_rmap.h"
23#include "xfs_ag.h"
24
25STATIC int
26xfs_inobt_get_minrecs(
27 struct xfs_btree_cur *cur,
28 int level)
29{
30 return M_IGEO(cur->bc_mp)->inobt_mnr[level != 0];
31}
32
33STATIC struct xfs_btree_cur *
34xfs_inobt_dup_cursor(
35 struct xfs_btree_cur *cur)
36{
37 return xfs_inobt_init_cursor(cur->bc_mp, cur->bc_tp,
38 cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
39}
40
41STATIC void
42xfs_inobt_set_root(
43 struct xfs_btree_cur *cur,
44 union xfs_btree_ptr *nptr,
45 int inc) /* level change */
46{
47 struct xfs_buf *agbp = cur->bc_ag.agbp;
48 struct xfs_agi *agi = agbp->b_addr;
49
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);
53}
54
55STATIC void
56xfs_finobt_set_root(
57 struct xfs_btree_cur *cur,
58 union xfs_btree_ptr *nptr,
59 int inc) /* level change */
60{
61 struct xfs_buf *agbp = cur->bc_ag.agbp;
62 struct xfs_agi *agi = agbp->b_addr;
63
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);
68}
69
70/* Update the inode btree block counter for this btree. */
71static inline void
72xfs_inobt_mod_blockcount(
73 struct xfs_btree_cur *cur,
74 int howmuch)
75{
76 struct xfs_buf *agbp = cur->bc_ag.agbp;
77 struct xfs_agi *agi = agbp->b_addr;
78
79 if (!xfs_sb_version_hasinobtcounts(&cur->bc_mp->m_sb))
80 return;
81
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);
87}
88
89STATIC int
90__xfs_inobt_alloc_block(
91 struct xfs_btree_cur *cur,
92 union xfs_btree_ptr *start,
93 union xfs_btree_ptr *new,
94 int *stat,
95 enum xfs_ag_resv_type resv)
96{
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);
100
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);
106 args.minlen = 1;
107 args.maxlen = 1;
108 args.prod = 1;
109 args.type = XFS_ALLOCTYPE_NEAR_BNO;
110 args.resv = resv;
111
112 error = xfs_alloc_vextent(&args);
113 if (error)
114 return error;
115
116 if (args.fsbno == NULLFSBLOCK) {
117 *stat = 0;
118 return 0;
119 }
120 ASSERT(args.len == 1);
121
122 new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
123 *stat = 1;
124 xfs_inobt_mod_blockcount(cur, 1);
125 return 0;
126}
127
128STATIC int
129xfs_inobt_alloc_block(
130 struct xfs_btree_cur *cur,
131 union xfs_btree_ptr *start,
132 union xfs_btree_ptr *new,
133 int *stat)
134{
135 return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
136}
137
138STATIC int
139xfs_finobt_alloc_block(
140 struct xfs_btree_cur *cur,
141 union xfs_btree_ptr *start,
142 union xfs_btree_ptr *new,
143 int *stat)
144{
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);
149}
150
151STATIC int
152__xfs_inobt_free_block(
153 struct xfs_btree_cur *cur,
154 struct xfs_buf *bp,
155 enum xfs_ag_resv_type resv)
156{
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_ADDR(bp)), 1,
160 &XFS_RMAP_OINFO_INOBT, resv);
161}
162
163STATIC int
164xfs_inobt_free_block(
165 struct xfs_btree_cur *cur,
166 struct xfs_buf *bp)
167{
168 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
169}
170
171STATIC int
172xfs_finobt_free_block(
173 struct xfs_btree_cur *cur,
174 struct xfs_buf *bp)
175{
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);
179}
180
181STATIC int
182xfs_inobt_get_maxrecs(
183 struct xfs_btree_cur *cur,
184 int level)
185{
186 return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
187}
188
189STATIC void
190xfs_inobt_init_key_from_rec(
191 union xfs_btree_key *key,
192 union xfs_btree_rec *rec)
193{
194 key->inobt.ir_startino = rec->inobt.ir_startino;
195}
196
197STATIC void
198xfs_inobt_init_high_key_from_rec(
199 union xfs_btree_key *key,
200 union xfs_btree_rec *rec)
201{
202 __u32 x;
203
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);
207}
208
209STATIC void
210xfs_inobt_init_rec_from_cur(
211 struct xfs_btree_cur *cur,
212 union xfs_btree_rec *rec)
213{
214 rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
215 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
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;
220 } else {
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);
224 }
225 rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
226}
227
228/*
229 * initial value of ptr for lookup
230 */
231STATIC void
232xfs_inobt_init_ptr_from_cur(
233 struct xfs_btree_cur *cur,
234 union xfs_btree_ptr *ptr)
235{
236 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
237
238 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
239
240 ptr->s = agi->agi_root;
241}
242
243STATIC void
244xfs_finobt_init_ptr_from_cur(
245 struct xfs_btree_cur *cur,
246 union xfs_btree_ptr *ptr)
247{
248 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
249
250 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agi->agi_seqno));
251 ptr->s = agi->agi_free_root;
252}
253
254STATIC int64_t
255xfs_inobt_key_diff(
256 struct xfs_btree_cur *cur,
257 union xfs_btree_key *key)
258{
259 return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
260 cur->bc_rec.i.ir_startino;
261}
262
263STATIC int64_t
264xfs_inobt_diff_two_keys(
265 struct xfs_btree_cur *cur,
266 union xfs_btree_key *k1,
267 union xfs_btree_key *k2)
268{
269 return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
270 be32_to_cpu(k2->inobt.ir_startino);
271}
272
273static xfs_failaddr_t
274xfs_inobt_verify(
275 struct xfs_buf *bp)
276{
277 struct xfs_mount *mp = bp->b_mount;
278 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
279 xfs_failaddr_t fa;
280 unsigned int level;
281
282 if (!xfs_verify_magic(bp, block->bb_magic))
283 return __this_address;
284
285 /*
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.
288 *
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
293 * ever do.
294 */
295 if (xfs_sb_version_hascrc(&mp->m_sb)) {
296 fa = xfs_btree_sblock_v5hdr_verify(bp);
297 if (fa)
298 return fa;
299 }
300
301 /* level verification */
302 level = be16_to_cpu(block->bb_level);
303 if (level >= M_IGEO(mp)->inobt_maxlevels)
304 return __this_address;
305
306 return xfs_btree_sblock_verify(bp,
307 M_IGEO(mp)->inobt_mxr[level != 0]);
308}
309
310static void
311xfs_inobt_read_verify(
312 struct xfs_buf *bp)
313{
314 xfs_failaddr_t fa;
315
316 if (!xfs_btree_sblock_verify_crc(bp))
317 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
318 else {
319 fa = xfs_inobt_verify(bp);
320 if (fa)
321 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
322 }
323
324 if (bp->b_error)
325 trace_xfs_btree_corrupt(bp, _RET_IP_);
326}
327
328static void
329xfs_inobt_write_verify(
330 struct xfs_buf *bp)
331{
332 xfs_failaddr_t fa;
333
334 fa = xfs_inobt_verify(bp);
335 if (fa) {
336 trace_xfs_btree_corrupt(bp, _RET_IP_);
337 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
338 return;
339 }
340 xfs_btree_sblock_calc_crc(bp);
341
342}
343
344const struct xfs_buf_ops xfs_inobt_buf_ops = {
345 .name = "xfs_inobt",
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,
350};
351
352const 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,
359};
360
361STATIC int
362xfs_inobt_keys_inorder(
363 struct xfs_btree_cur *cur,
364 union xfs_btree_key *k1,
365 union xfs_btree_key *k2)
366{
367 return be32_to_cpu(k1->inobt.ir_startino) <
368 be32_to_cpu(k2->inobt.ir_startino);
369}
370
371STATIC int
372xfs_inobt_recs_inorder(
373 struct xfs_btree_cur *cur,
374 union xfs_btree_rec *r1,
375 union xfs_btree_rec *r2)
376{
377 return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
378 be32_to_cpu(r2->inobt.ir_startino);
379}
380
381static const struct xfs_btree_ops xfs_inobt_ops = {
382 .rec_len = sizeof(xfs_inobt_rec_t),
383 .key_len = sizeof(xfs_inobt_key_t),
384
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,
400};
401
402static const struct xfs_btree_ops xfs_finobt_ops = {
403 .rec_len = sizeof(xfs_inobt_rec_t),
404 .key_len = sizeof(xfs_inobt_key_t),
405
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,
421};
422
423/*
424 * Initialize a new inode btree cursor.
425 */
426static struct xfs_btree_cur *
427xfs_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 */
432{
433 struct xfs_btree_cur *cur;
434
435 cur = kmem_cache_zalloc(xfs_btree_cur_zone, GFP_NOFS | __GFP_NOFAIL);
436 cur->bc_tp = tp;
437 cur->bc_mp = mp;
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;
442 } else {
443 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_fibt_2);
444 cur->bc_ops = &xfs_finobt_ops;
445 }
446
447 cur->bc_blocklog = mp->m_sb.sb_blocklog;
448
449 if (xfs_sb_version_hascrc(&mp->m_sb))
450 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
451
452 /* take a reference for the cursor */
453 atomic_inc(&pag->pag_ref);
454 cur->bc_ag.pag = pag;
455 return cur;
456}
457
458/* Create an inode btree cursor. */
459struct xfs_btree_cur *
460xfs_inobt_init_cursor(
461 struct xfs_mount *mp,
462 struct xfs_trans *tp,
463 struct xfs_buf *agbp,
464 struct xfs_perag *pag,
465 xfs_btnum_t btnum)
466{
467 struct xfs_btree_cur *cur;
468 struct xfs_agi *agi = agbp->b_addr;
469
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);
473 else
474 cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
475 cur->bc_ag.agbp = agbp;
476 return cur;
477}
478
479/* Create an inode btree cursor with a fake root for staging. */
480struct xfs_btree_cur *
481xfs_inobt_stage_cursor(
482 struct xfs_mount *mp,
483 struct xbtree_afakeroot *afake,
484 struct xfs_perag *pag,
485 xfs_btnum_t btnum)
486{
487 struct xfs_btree_cur *cur;
488
489 cur = xfs_inobt_init_common(mp, NULL, pag, btnum);
490 xfs_btree_stage_afakeroot(cur, afake);
491 return cur;
492}
493
494/*
495 * Install a new inobt btree root. Caller is responsible for invalidating
496 * and freeing the old btree blocks.
497 */
498void
499xfs_inobt_commit_staged_btree(
500 struct xfs_btree_cur *cur,
501 struct xfs_trans *tp,
502 struct xfs_buf *agbp)
503{
504 struct xfs_agi *agi = agbp->b_addr;
505 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
506 int fields;
507
508 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
509
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_sb_version_hasinobtcounts(&cur->bc_mp->m_sb)) {
515 agi->agi_iblocks = cpu_to_be32(afake->af_blocks);
516 fields |= XFS_AGI_IBLOCKS;
517 }
518 xfs_ialloc_log_agi(tp, agbp, fields);
519 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_inobt_ops);
520 } else {
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_sb_version_hasinobtcounts(&cur->bc_mp->m_sb)) {
525 agi->agi_fblocks = cpu_to_be32(afake->af_blocks);
526 fields |= XFS_AGI_IBLOCKS;
527 }
528 xfs_ialloc_log_agi(tp, agbp, fields);
529 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_finobt_ops);
530 }
531}
532
533/*
534 * Calculate number of records in an inobt btree block.
535 */
536int
537xfs_inobt_maxrecs(
538 struct xfs_mount *mp,
539 int blocklen,
540 int leaf)
541{
542 blocklen -= XFS_INOBT_BLOCK_LEN(mp);
543
544 if (leaf)
545 return blocklen / sizeof(xfs_inobt_rec_t);
546 return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
547}
548
549/*
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).
554 *
555 * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
556 */
557uint64_t
558xfs_inobt_irec_to_allocmask(
559 struct xfs_inobt_rec_incore *rec)
560{
561 uint64_t bitmap = 0;
562 uint64_t inodespbit;
563 int nextbit;
564 uint allocbitmap;
565
566 /*
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.
570 */
571 inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
572
573 /*
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
577 * type.
578 */
579 allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
580
581 /*
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.
586 */
587 nextbit = xfs_next_bit(&allocbitmap, 1, 0);
588 while (nextbit != -1) {
589 ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
590
591 bitmap |= (inodespbit <<
592 (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
593
594 nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
595 }
596
597 return bitmap;
598}
599
600#if defined(DEBUG) || defined(XFS_WARN)
601/*
602 * Verify that an in-core inode record has a valid inode count.
603 */
604int
605xfs_inobt_rec_check_count(
606 struct xfs_mount *mp,
607 struct xfs_inobt_rec_incore *rec)
608{
609 int inocount = 0;
610 int nextbit = 0;
611 uint64_t allocbmap;
612 int wordsz;
613
614 wordsz = sizeof(allocbmap) / sizeof(unsigned int);
615 allocbmap = xfs_inobt_irec_to_allocmask(rec);
616
617 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
618 while (nextbit != -1) {
619 inocount++;
620 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
621 nextbit + 1);
622 }
623
624 if (inocount != rec->ir_count)
625 return -EFSCORRUPTED;
626
627 return 0;
628}
629#endif /* DEBUG */
630
631static xfs_extlen_t
632xfs_inobt_max_size(
633 struct xfs_mount *mp,
634 xfs_agnumber_t agno)
635{
636 xfs_agblock_t agblocks = xfs_ag_block_count(mp, agno);
637
638 /* Bail out if we're uninitialized, which can happen in mkfs. */
639 if (M_IGEO(mp)->inobt_mxr[0] == 0)
640 return 0;
641
642 /*
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.
646 */
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;
650
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);
654}
655
656/* Read AGI and create inobt cursor. */
657int
658xfs_inobt_cur(
659 struct xfs_mount *mp,
660 struct xfs_trans *tp,
661 struct xfs_perag *pag,
662 xfs_btnum_t which,
663 struct xfs_btree_cur **curpp,
664 struct xfs_buf **agi_bpp)
665{
666 struct xfs_btree_cur *cur;
667 int error;
668
669 ASSERT(*agi_bpp == NULL);
670 ASSERT(*curpp == NULL);
671
672 error = xfs_ialloc_read_agi(mp, tp, pag->pag_agno, agi_bpp);
673 if (error)
674 return error;
675
676 cur = xfs_inobt_init_cursor(mp, tp, *agi_bpp, pag, which);
677 *curpp = cur;
678 return 0;
679}
680
681static int
682xfs_inobt_count_blocks(
683 struct xfs_mount *mp,
684 struct xfs_trans *tp,
685 struct xfs_perag *pag,
686 xfs_btnum_t btnum,
687 xfs_extlen_t *tree_blocks)
688{
689 struct xfs_buf *agbp = NULL;
690 struct xfs_btree_cur *cur = NULL;
691 int error;
692
693 error = xfs_inobt_cur(mp, tp, pag, btnum, &cur, &agbp);
694 if (error)
695 return error;
696
697 error = xfs_btree_count_blocks(cur, tree_blocks);
698 xfs_btree_del_cursor(cur, error);
699 xfs_trans_brelse(tp, agbp);
700
701 return error;
702}
703
704/* Read finobt block count from AGI header. */
705static int
706xfs_finobt_read_blocks(
707 struct xfs_mount *mp,
708 struct xfs_trans *tp,
709 struct xfs_perag *pag,
710 xfs_extlen_t *tree_blocks)
711{
712 struct xfs_buf *agbp;
713 struct xfs_agi *agi;
714 int error;
715
716 error = xfs_ialloc_read_agi(mp, tp, pag->pag_agno, &agbp);
717 if (error)
718 return error;
719
720 agi = agbp->b_addr;
721 *tree_blocks = be32_to_cpu(agi->agi_fblocks);
722 xfs_trans_brelse(tp, agbp);
723 return 0;
724}
725
726/*
727 * Figure out how many blocks to reserve and how many are used by this btree.
728 */
729int
730xfs_finobt_calc_reserves(
731 struct xfs_mount *mp,
732 struct xfs_trans *tp,
733 struct xfs_perag *pag,
734 xfs_extlen_t *ask,
735 xfs_extlen_t *used)
736{
737 xfs_extlen_t tree_len = 0;
738 int error;
739
740 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
741 return 0;
742
743 if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
744 error = xfs_finobt_read_blocks(mp, tp, pag, &tree_len);
745 else
746 error = xfs_inobt_count_blocks(mp, tp, pag, XFS_BTNUM_FINO,
747 &tree_len);
748 if (error)
749 return error;
750
751 *ask += xfs_inobt_max_size(mp, pag->pag_agno);
752 *used += tree_len;
753 return 0;
754}
755
756/* Calculate the inobt btree size for some records. */
757xfs_extlen_t
758xfs_iallocbt_calc_size(
759 struct xfs_mount *mp,
760 unsigned long long len)
761{
762 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);
763}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
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_bit.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_health.h"
21#include "xfs_trace.h"
22#include "xfs_trans.h"
23#include "xfs_rmap.h"
24#include "xfs_ag.h"
25
26static struct kmem_cache *xfs_inobt_cur_cache;
27
28STATIC int
29xfs_inobt_get_minrecs(
30 struct xfs_btree_cur *cur,
31 int level)
32{
33 return M_IGEO(cur->bc_mp)->inobt_mnr[level != 0];
34}
35
36STATIC struct xfs_btree_cur *
37xfs_inobt_dup_cursor(
38 struct xfs_btree_cur *cur)
39{
40 return xfs_inobt_init_cursor(to_perag(cur->bc_group), cur->bc_tp,
41 cur->bc_ag.agbp);
42}
43
44STATIC struct xfs_btree_cur *
45xfs_finobt_dup_cursor(
46 struct xfs_btree_cur *cur)
47{
48 return xfs_finobt_init_cursor(to_perag(cur->bc_group), cur->bc_tp,
49 cur->bc_ag.agbp);
50}
51
52STATIC void
53xfs_inobt_set_root(
54 struct xfs_btree_cur *cur,
55 const union xfs_btree_ptr *nptr,
56 int inc) /* level change */
57{
58 struct xfs_buf *agbp = cur->bc_ag.agbp;
59 struct xfs_agi *agi = agbp->b_addr;
60
61 agi->agi_root = nptr->s;
62 be32_add_cpu(&agi->agi_level, inc);
63 xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_ROOT | XFS_AGI_LEVEL);
64}
65
66STATIC void
67xfs_finobt_set_root(
68 struct xfs_btree_cur *cur,
69 const union xfs_btree_ptr *nptr,
70 int inc) /* level change */
71{
72 struct xfs_buf *agbp = cur->bc_ag.agbp;
73 struct xfs_agi *agi = agbp->b_addr;
74
75 agi->agi_free_root = nptr->s;
76 be32_add_cpu(&agi->agi_free_level, inc);
77 xfs_ialloc_log_agi(cur->bc_tp, agbp,
78 XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL);
79}
80
81/* Update the inode btree block counter for this btree. */
82static inline void
83xfs_inobt_mod_blockcount(
84 struct xfs_btree_cur *cur,
85 int howmuch)
86{
87 struct xfs_buf *agbp = cur->bc_ag.agbp;
88 struct xfs_agi *agi = agbp->b_addr;
89
90 if (!xfs_has_inobtcounts(cur->bc_mp))
91 return;
92
93 if (xfs_btree_is_fino(cur->bc_ops))
94 be32_add_cpu(&agi->agi_fblocks, howmuch);
95 else
96 be32_add_cpu(&agi->agi_iblocks, howmuch);
97 xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_IBLOCKS);
98}
99
100STATIC int
101__xfs_inobt_alloc_block(
102 struct xfs_btree_cur *cur,
103 const union xfs_btree_ptr *start,
104 union xfs_btree_ptr *new,
105 int *stat,
106 enum xfs_ag_resv_type resv)
107{
108 xfs_alloc_arg_t args; /* block allocation args */
109 int error; /* error return value */
110 xfs_agblock_t sbno = be32_to_cpu(start->s);
111
112 memset(&args, 0, sizeof(args));
113 args.tp = cur->bc_tp;
114 args.mp = cur->bc_mp;
115 args.pag = to_perag(cur->bc_group);
116 args.oinfo = XFS_RMAP_OINFO_INOBT;
117 args.minlen = 1;
118 args.maxlen = 1;
119 args.prod = 1;
120 args.resv = resv;
121
122 error = xfs_alloc_vextent_near_bno(&args,
123 xfs_agbno_to_fsb(args.pag, sbno));
124 if (error)
125 return error;
126
127 if (args.fsbno == NULLFSBLOCK) {
128 *stat = 0;
129 return 0;
130 }
131 ASSERT(args.len == 1);
132
133 new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
134 *stat = 1;
135 xfs_inobt_mod_blockcount(cur, 1);
136 return 0;
137}
138
139STATIC int
140xfs_inobt_alloc_block(
141 struct xfs_btree_cur *cur,
142 const union xfs_btree_ptr *start,
143 union xfs_btree_ptr *new,
144 int *stat)
145{
146 return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
147}
148
149STATIC int
150xfs_finobt_alloc_block(
151 struct xfs_btree_cur *cur,
152 const union xfs_btree_ptr *start,
153 union xfs_btree_ptr *new,
154 int *stat)
155{
156 if (cur->bc_mp->m_finobt_nores)
157 return xfs_inobt_alloc_block(cur, start, new, stat);
158 return __xfs_inobt_alloc_block(cur, start, new, stat,
159 XFS_AG_RESV_METADATA);
160}
161
162STATIC int
163__xfs_inobt_free_block(
164 struct xfs_btree_cur *cur,
165 struct xfs_buf *bp,
166 enum xfs_ag_resv_type resv)
167{
168 xfs_fsblock_t fsbno;
169
170 xfs_inobt_mod_blockcount(cur, -1);
171 fsbno = XFS_DADDR_TO_FSB(cur->bc_mp, xfs_buf_daddr(bp));
172 return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
173 &XFS_RMAP_OINFO_INOBT, resv, 0);
174}
175
176STATIC int
177xfs_inobt_free_block(
178 struct xfs_btree_cur *cur,
179 struct xfs_buf *bp)
180{
181 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_NONE);
182}
183
184STATIC int
185xfs_finobt_free_block(
186 struct xfs_btree_cur *cur,
187 struct xfs_buf *bp)
188{
189 if (cur->bc_mp->m_finobt_nores)
190 return xfs_inobt_free_block(cur, bp);
191 return __xfs_inobt_free_block(cur, bp, XFS_AG_RESV_METADATA);
192}
193
194STATIC int
195xfs_inobt_get_maxrecs(
196 struct xfs_btree_cur *cur,
197 int level)
198{
199 return M_IGEO(cur->bc_mp)->inobt_mxr[level != 0];
200}
201
202STATIC void
203xfs_inobt_init_key_from_rec(
204 union xfs_btree_key *key,
205 const union xfs_btree_rec *rec)
206{
207 key->inobt.ir_startino = rec->inobt.ir_startino;
208}
209
210STATIC void
211xfs_inobt_init_high_key_from_rec(
212 union xfs_btree_key *key,
213 const union xfs_btree_rec *rec)
214{
215 __u32 x;
216
217 x = be32_to_cpu(rec->inobt.ir_startino);
218 x += XFS_INODES_PER_CHUNK - 1;
219 key->inobt.ir_startino = cpu_to_be32(x);
220}
221
222STATIC void
223xfs_inobt_init_rec_from_cur(
224 struct xfs_btree_cur *cur,
225 union xfs_btree_rec *rec)
226{
227 rec->inobt.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
228 if (xfs_has_sparseinodes(cur->bc_mp)) {
229 rec->inobt.ir_u.sp.ir_holemask =
230 cpu_to_be16(cur->bc_rec.i.ir_holemask);
231 rec->inobt.ir_u.sp.ir_count = cur->bc_rec.i.ir_count;
232 rec->inobt.ir_u.sp.ir_freecount = cur->bc_rec.i.ir_freecount;
233 } else {
234 /* ir_holemask/ir_count not supported on-disk */
235 rec->inobt.ir_u.f.ir_freecount =
236 cpu_to_be32(cur->bc_rec.i.ir_freecount);
237 }
238 rec->inobt.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
239}
240
241/*
242 * initial value of ptr for lookup
243 */
244STATIC void
245xfs_inobt_init_ptr_from_cur(
246 struct xfs_btree_cur *cur,
247 union xfs_btree_ptr *ptr)
248{
249 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
250
251 ASSERT(cur->bc_group->xg_gno == be32_to_cpu(agi->agi_seqno));
252
253 ptr->s = agi->agi_root;
254}
255
256STATIC void
257xfs_finobt_init_ptr_from_cur(
258 struct xfs_btree_cur *cur,
259 union xfs_btree_ptr *ptr)
260{
261 struct xfs_agi *agi = cur->bc_ag.agbp->b_addr;
262
263 ASSERT(cur->bc_group->xg_gno == be32_to_cpu(agi->agi_seqno));
264
265 ptr->s = agi->agi_free_root;
266}
267
268STATIC int64_t
269xfs_inobt_key_diff(
270 struct xfs_btree_cur *cur,
271 const union xfs_btree_key *key)
272{
273 return (int64_t)be32_to_cpu(key->inobt.ir_startino) -
274 cur->bc_rec.i.ir_startino;
275}
276
277STATIC int64_t
278xfs_inobt_diff_two_keys(
279 struct xfs_btree_cur *cur,
280 const union xfs_btree_key *k1,
281 const union xfs_btree_key *k2,
282 const union xfs_btree_key *mask)
283{
284 ASSERT(!mask || mask->inobt.ir_startino);
285
286 return (int64_t)be32_to_cpu(k1->inobt.ir_startino) -
287 be32_to_cpu(k2->inobt.ir_startino);
288}
289
290static xfs_failaddr_t
291xfs_inobt_verify(
292 struct xfs_buf *bp)
293{
294 struct xfs_mount *mp = bp->b_mount;
295 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
296 xfs_failaddr_t fa;
297 unsigned int level;
298
299 if (!xfs_verify_magic(bp, block->bb_magic))
300 return __this_address;
301
302 /*
303 * During growfs operations, we can't verify the exact owner as the
304 * perag is not fully initialised and hence not attached to the buffer.
305 *
306 * Similarly, during log recovery we will have a perag structure
307 * attached, but the agi information will not yet have been initialised
308 * from the on disk AGI. We don't currently use any of this information,
309 * but beware of the landmine (i.e. need to check
310 * xfs_perag_initialised_agi(pag)) if we ever do.
311 */
312 if (xfs_has_crc(mp)) {
313 fa = xfs_btree_agblock_v5hdr_verify(bp);
314 if (fa)
315 return fa;
316 }
317
318 /* level verification */
319 level = be16_to_cpu(block->bb_level);
320 if (level >= M_IGEO(mp)->inobt_maxlevels)
321 return __this_address;
322
323 return xfs_btree_agblock_verify(bp,
324 M_IGEO(mp)->inobt_mxr[level != 0]);
325}
326
327static void
328xfs_inobt_read_verify(
329 struct xfs_buf *bp)
330{
331 xfs_failaddr_t fa;
332
333 if (!xfs_btree_agblock_verify_crc(bp))
334 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
335 else {
336 fa = xfs_inobt_verify(bp);
337 if (fa)
338 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
339 }
340
341 if (bp->b_error)
342 trace_xfs_btree_corrupt(bp, _RET_IP_);
343}
344
345static void
346xfs_inobt_write_verify(
347 struct xfs_buf *bp)
348{
349 xfs_failaddr_t fa;
350
351 fa = xfs_inobt_verify(bp);
352 if (fa) {
353 trace_xfs_btree_corrupt(bp, _RET_IP_);
354 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
355 return;
356 }
357 xfs_btree_agblock_calc_crc(bp);
358
359}
360
361const struct xfs_buf_ops xfs_inobt_buf_ops = {
362 .name = "xfs_inobt",
363 .magic = { cpu_to_be32(XFS_IBT_MAGIC), cpu_to_be32(XFS_IBT_CRC_MAGIC) },
364 .verify_read = xfs_inobt_read_verify,
365 .verify_write = xfs_inobt_write_verify,
366 .verify_struct = xfs_inobt_verify,
367};
368
369const struct xfs_buf_ops xfs_finobt_buf_ops = {
370 .name = "xfs_finobt",
371 .magic = { cpu_to_be32(XFS_FIBT_MAGIC),
372 cpu_to_be32(XFS_FIBT_CRC_MAGIC) },
373 .verify_read = xfs_inobt_read_verify,
374 .verify_write = xfs_inobt_write_verify,
375 .verify_struct = xfs_inobt_verify,
376};
377
378STATIC int
379xfs_inobt_keys_inorder(
380 struct xfs_btree_cur *cur,
381 const union xfs_btree_key *k1,
382 const union xfs_btree_key *k2)
383{
384 return be32_to_cpu(k1->inobt.ir_startino) <
385 be32_to_cpu(k2->inobt.ir_startino);
386}
387
388STATIC int
389xfs_inobt_recs_inorder(
390 struct xfs_btree_cur *cur,
391 const union xfs_btree_rec *r1,
392 const union xfs_btree_rec *r2)
393{
394 return be32_to_cpu(r1->inobt.ir_startino) + XFS_INODES_PER_CHUNK <=
395 be32_to_cpu(r2->inobt.ir_startino);
396}
397
398STATIC enum xbtree_key_contig
399xfs_inobt_keys_contiguous(
400 struct xfs_btree_cur *cur,
401 const union xfs_btree_key *key1,
402 const union xfs_btree_key *key2,
403 const union xfs_btree_key *mask)
404{
405 ASSERT(!mask || mask->inobt.ir_startino);
406
407 return xbtree_key_contig(be32_to_cpu(key1->inobt.ir_startino),
408 be32_to_cpu(key2->inobt.ir_startino));
409}
410
411const struct xfs_btree_ops xfs_inobt_ops = {
412 .name = "ino",
413 .type = XFS_BTREE_TYPE_AG,
414
415 .rec_len = sizeof(xfs_inobt_rec_t),
416 .key_len = sizeof(xfs_inobt_key_t),
417 .ptr_len = XFS_BTREE_SHORT_PTR_LEN,
418
419 .lru_refs = XFS_INO_BTREE_REF,
420 .statoff = XFS_STATS_CALC_INDEX(xs_ibt_2),
421 .sick_mask = XFS_SICK_AG_INOBT,
422
423 .dup_cursor = xfs_inobt_dup_cursor,
424 .set_root = xfs_inobt_set_root,
425 .alloc_block = xfs_inobt_alloc_block,
426 .free_block = xfs_inobt_free_block,
427 .get_minrecs = xfs_inobt_get_minrecs,
428 .get_maxrecs = xfs_inobt_get_maxrecs,
429 .init_key_from_rec = xfs_inobt_init_key_from_rec,
430 .init_high_key_from_rec = xfs_inobt_init_high_key_from_rec,
431 .init_rec_from_cur = xfs_inobt_init_rec_from_cur,
432 .init_ptr_from_cur = xfs_inobt_init_ptr_from_cur,
433 .key_diff = xfs_inobt_key_diff,
434 .buf_ops = &xfs_inobt_buf_ops,
435 .diff_two_keys = xfs_inobt_diff_two_keys,
436 .keys_inorder = xfs_inobt_keys_inorder,
437 .recs_inorder = xfs_inobt_recs_inorder,
438 .keys_contiguous = xfs_inobt_keys_contiguous,
439};
440
441const struct xfs_btree_ops xfs_finobt_ops = {
442 .name = "fino",
443 .type = XFS_BTREE_TYPE_AG,
444
445 .rec_len = sizeof(xfs_inobt_rec_t),
446 .key_len = sizeof(xfs_inobt_key_t),
447 .ptr_len = XFS_BTREE_SHORT_PTR_LEN,
448
449 .lru_refs = XFS_INO_BTREE_REF,
450 .statoff = XFS_STATS_CALC_INDEX(xs_fibt_2),
451 .sick_mask = XFS_SICK_AG_FINOBT,
452
453 .dup_cursor = xfs_finobt_dup_cursor,
454 .set_root = xfs_finobt_set_root,
455 .alloc_block = xfs_finobt_alloc_block,
456 .free_block = xfs_finobt_free_block,
457 .get_minrecs = xfs_inobt_get_minrecs,
458 .get_maxrecs = xfs_inobt_get_maxrecs,
459 .init_key_from_rec = xfs_inobt_init_key_from_rec,
460 .init_high_key_from_rec = xfs_inobt_init_high_key_from_rec,
461 .init_rec_from_cur = xfs_inobt_init_rec_from_cur,
462 .init_ptr_from_cur = xfs_finobt_init_ptr_from_cur,
463 .key_diff = xfs_inobt_key_diff,
464 .buf_ops = &xfs_finobt_buf_ops,
465 .diff_two_keys = xfs_inobt_diff_two_keys,
466 .keys_inorder = xfs_inobt_keys_inorder,
467 .recs_inorder = xfs_inobt_recs_inorder,
468 .keys_contiguous = xfs_inobt_keys_contiguous,
469};
470
471/*
472 * Create an inode btree cursor.
473 *
474 * For staging cursors tp and agbp are NULL.
475 */
476struct xfs_btree_cur *
477xfs_inobt_init_cursor(
478 struct xfs_perag *pag,
479 struct xfs_trans *tp,
480 struct xfs_buf *agbp)
481{
482 struct xfs_mount *mp = pag_mount(pag);
483 struct xfs_btree_cur *cur;
484
485 cur = xfs_btree_alloc_cursor(mp, tp, &xfs_inobt_ops,
486 M_IGEO(mp)->inobt_maxlevels, xfs_inobt_cur_cache);
487 cur->bc_group = xfs_group_hold(pag_group(pag));
488 cur->bc_ag.agbp = agbp;
489 if (agbp) {
490 struct xfs_agi *agi = agbp->b_addr;
491
492 cur->bc_nlevels = be32_to_cpu(agi->agi_level);
493 }
494 return cur;
495}
496
497/*
498 * Create a free inode btree cursor.
499 *
500 * For staging cursors tp and agbp are NULL.
501 */
502struct xfs_btree_cur *
503xfs_finobt_init_cursor(
504 struct xfs_perag *pag,
505 struct xfs_trans *tp,
506 struct xfs_buf *agbp)
507{
508 struct xfs_mount *mp = pag_mount(pag);
509 struct xfs_btree_cur *cur;
510
511 cur = xfs_btree_alloc_cursor(mp, tp, &xfs_finobt_ops,
512 M_IGEO(mp)->inobt_maxlevels, xfs_inobt_cur_cache);
513 cur->bc_group = xfs_group_hold(pag_group(pag));
514 cur->bc_ag.agbp = agbp;
515 if (agbp) {
516 struct xfs_agi *agi = agbp->b_addr;
517
518 cur->bc_nlevels = be32_to_cpu(agi->agi_free_level);
519 }
520 return cur;
521}
522
523/*
524 * Install a new inobt btree root. Caller is responsible for invalidating
525 * and freeing the old btree blocks.
526 */
527void
528xfs_inobt_commit_staged_btree(
529 struct xfs_btree_cur *cur,
530 struct xfs_trans *tp,
531 struct xfs_buf *agbp)
532{
533 struct xfs_agi *agi = agbp->b_addr;
534 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
535 int fields;
536
537 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
538
539 if (xfs_btree_is_ino(cur->bc_ops)) {
540 fields = XFS_AGI_ROOT | XFS_AGI_LEVEL;
541 agi->agi_root = cpu_to_be32(afake->af_root);
542 agi->agi_level = cpu_to_be32(afake->af_levels);
543 if (xfs_has_inobtcounts(cur->bc_mp)) {
544 agi->agi_iblocks = cpu_to_be32(afake->af_blocks);
545 fields |= XFS_AGI_IBLOCKS;
546 }
547 xfs_ialloc_log_agi(tp, agbp, fields);
548 xfs_btree_commit_afakeroot(cur, tp, agbp);
549 } else {
550 fields = XFS_AGI_FREE_ROOT | XFS_AGI_FREE_LEVEL;
551 agi->agi_free_root = cpu_to_be32(afake->af_root);
552 agi->agi_free_level = cpu_to_be32(afake->af_levels);
553 if (xfs_has_inobtcounts(cur->bc_mp)) {
554 agi->agi_fblocks = cpu_to_be32(afake->af_blocks);
555 fields |= XFS_AGI_IBLOCKS;
556 }
557 xfs_ialloc_log_agi(tp, agbp, fields);
558 xfs_btree_commit_afakeroot(cur, tp, agbp);
559 }
560}
561
562/* Calculate number of records in an inode btree block. */
563static inline unsigned int
564xfs_inobt_block_maxrecs(
565 unsigned int blocklen,
566 bool leaf)
567{
568 if (leaf)
569 return blocklen / sizeof(xfs_inobt_rec_t);
570 return blocklen / (sizeof(xfs_inobt_key_t) + sizeof(xfs_inobt_ptr_t));
571}
572
573/*
574 * Calculate number of records in an inobt btree block.
575 */
576unsigned int
577xfs_inobt_maxrecs(
578 struct xfs_mount *mp,
579 unsigned int blocklen,
580 bool leaf)
581{
582 blocklen -= XFS_INOBT_BLOCK_LEN(mp);
583 return xfs_inobt_block_maxrecs(blocklen, leaf);
584}
585
586/*
587 * Maximum number of inode btree records per AG. Pretend that we can fill an
588 * entire AG completely full of inodes except for the AG headers.
589 */
590#define XFS_MAX_INODE_RECORDS \
591 ((XFS_MAX_AG_BYTES - (4 * BBSIZE)) / XFS_DINODE_MIN_SIZE) / \
592 XFS_INODES_PER_CHUNK
593
594/* Compute the max possible height for the inode btree. */
595static inline unsigned int
596xfs_inobt_maxlevels_ondisk(void)
597{
598 unsigned int minrecs[2];
599 unsigned int blocklen;
600
601 blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
602 XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
603
604 minrecs[0] = xfs_inobt_block_maxrecs(blocklen, true) / 2;
605 minrecs[1] = xfs_inobt_block_maxrecs(blocklen, false) / 2;
606
607 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_INODE_RECORDS);
608}
609
610/* Compute the max possible height for the free inode btree. */
611static inline unsigned int
612xfs_finobt_maxlevels_ondisk(void)
613{
614 unsigned int minrecs[2];
615 unsigned int blocklen;
616
617 blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
618
619 minrecs[0] = xfs_inobt_block_maxrecs(blocklen, true) / 2;
620 minrecs[1] = xfs_inobt_block_maxrecs(blocklen, false) / 2;
621
622 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_INODE_RECORDS);
623}
624
625/* Compute the max possible height for either inode btree. */
626unsigned int
627xfs_iallocbt_maxlevels_ondisk(void)
628{
629 return max(xfs_inobt_maxlevels_ondisk(),
630 xfs_finobt_maxlevels_ondisk());
631}
632
633/*
634 * Convert the inode record holemask to an inode allocation bitmap. The inode
635 * allocation bitmap is inode granularity and specifies whether an inode is
636 * physically allocated on disk (not whether the inode is considered allocated
637 * or free by the fs).
638 *
639 * A bit value of 1 means the inode is allocated, a value of 0 means it is free.
640 */
641uint64_t
642xfs_inobt_irec_to_allocmask(
643 const struct xfs_inobt_rec_incore *rec)
644{
645 uint64_t bitmap = 0;
646 uint64_t inodespbit;
647 int nextbit;
648 uint allocbitmap;
649
650 /*
651 * The holemask has 16-bits for a 64 inode record. Therefore each
652 * holemask bit represents multiple inodes. Create a mask of bits to set
653 * in the allocmask for each holemask bit.
654 */
655 inodespbit = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
656
657 /*
658 * Allocated inodes are represented by 0 bits in holemask. Invert the 0
659 * bits to 1 and convert to a uint so we can use xfs_next_bit(). Mask
660 * anything beyond the 16 holemask bits since this casts to a larger
661 * type.
662 */
663 allocbitmap = ~rec->ir_holemask & ((1 << XFS_INOBT_HOLEMASK_BITS) - 1);
664
665 /*
666 * allocbitmap is the inverted holemask so every set bit represents
667 * allocated inodes. To expand from 16-bit holemask granularity to
668 * 64-bit (e.g., bit-per-inode), set inodespbit bits in the target
669 * bitmap for every holemask bit.
670 */
671 nextbit = xfs_next_bit(&allocbitmap, 1, 0);
672 while (nextbit != -1) {
673 ASSERT(nextbit < (sizeof(rec->ir_holemask) * NBBY));
674
675 bitmap |= (inodespbit <<
676 (nextbit * XFS_INODES_PER_HOLEMASK_BIT));
677
678 nextbit = xfs_next_bit(&allocbitmap, 1, nextbit + 1);
679 }
680
681 return bitmap;
682}
683
684#if defined(DEBUG) || defined(XFS_WARN)
685/*
686 * Verify that an in-core inode record has a valid inode count.
687 */
688int
689xfs_inobt_rec_check_count(
690 struct xfs_mount *mp,
691 struct xfs_inobt_rec_incore *rec)
692{
693 int inocount = 0;
694 int nextbit = 0;
695 uint64_t allocbmap;
696 int wordsz;
697
698 wordsz = sizeof(allocbmap) / sizeof(unsigned int);
699 allocbmap = xfs_inobt_irec_to_allocmask(rec);
700
701 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz, nextbit);
702 while (nextbit != -1) {
703 inocount++;
704 nextbit = xfs_next_bit((uint *) &allocbmap, wordsz,
705 nextbit + 1);
706 }
707
708 if (inocount != rec->ir_count)
709 return -EFSCORRUPTED;
710
711 return 0;
712}
713#endif /* DEBUG */
714
715static xfs_extlen_t
716xfs_inobt_max_size(
717 struct xfs_perag *pag)
718{
719 struct xfs_mount *mp = pag_mount(pag);
720 xfs_agblock_t agblocks = pag_group(pag)->xg_block_count;
721
722 /* Bail out if we're uninitialized, which can happen in mkfs. */
723 if (M_IGEO(mp)->inobt_mxr[0] == 0)
724 return 0;
725
726 /*
727 * The log is permanently allocated, so the space it occupies will
728 * never be available for the kinds of things that would require btree
729 * expansion. We therefore can pretend the space isn't there.
730 */
731 if (xfs_ag_contains_log(mp, pag_agno(pag)))
732 agblocks -= mp->m_sb.sb_logblocks;
733
734 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr,
735 (uint64_t)agblocks * mp->m_sb.sb_inopblock /
736 XFS_INODES_PER_CHUNK);
737}
738
739static int
740xfs_finobt_count_blocks(
741 struct xfs_perag *pag,
742 struct xfs_trans *tp,
743 xfs_extlen_t *tree_blocks)
744{
745 struct xfs_buf *agbp = NULL;
746 struct xfs_btree_cur *cur;
747 xfs_filblks_t blocks;
748 int error;
749
750 error = xfs_ialloc_read_agi(pag, tp, 0, &agbp);
751 if (error)
752 return error;
753
754 cur = xfs_finobt_init_cursor(pag, tp, agbp);
755 error = xfs_btree_count_blocks(cur, &blocks);
756 xfs_btree_del_cursor(cur, error);
757 xfs_trans_brelse(tp, agbp);
758 *tree_blocks = blocks;
759
760 return error;
761}
762
763/* Read finobt block count from AGI header. */
764static int
765xfs_finobt_read_blocks(
766 struct xfs_perag *pag,
767 struct xfs_trans *tp,
768 xfs_extlen_t *tree_blocks)
769{
770 struct xfs_buf *agbp;
771 struct xfs_agi *agi;
772 int error;
773
774 error = xfs_ialloc_read_agi(pag, tp, 0, &agbp);
775 if (error)
776 return error;
777
778 agi = agbp->b_addr;
779 *tree_blocks = be32_to_cpu(agi->agi_fblocks);
780 xfs_trans_brelse(tp, agbp);
781 return 0;
782}
783
784/*
785 * Figure out how many blocks to reserve and how many are used by this btree.
786 */
787int
788xfs_finobt_calc_reserves(
789 struct xfs_perag *pag,
790 struct xfs_trans *tp,
791 xfs_extlen_t *ask,
792 xfs_extlen_t *used)
793{
794 xfs_extlen_t tree_len = 0;
795 int error;
796
797 if (!xfs_has_finobt(pag_mount(pag)))
798 return 0;
799
800 if (xfs_has_inobtcounts(pag_mount(pag)))
801 error = xfs_finobt_read_blocks(pag, tp, &tree_len);
802 else
803 error = xfs_finobt_count_blocks(pag, tp, &tree_len);
804 if (error)
805 return error;
806
807 *ask += xfs_inobt_max_size(pag);
808 *used += tree_len;
809 return 0;
810}
811
812/* Calculate the inobt btree size for some records. */
813xfs_extlen_t
814xfs_iallocbt_calc_size(
815 struct xfs_mount *mp,
816 unsigned long long len)
817{
818 return xfs_btree_calc_size(M_IGEO(mp)->inobt_mnr, len);
819}
820
821int __init
822xfs_inobt_init_cur_cache(void)
823{
824 xfs_inobt_cur_cache = kmem_cache_create("xfs_inobt_cur",
825 xfs_btree_cur_sizeof(xfs_inobt_maxlevels_ondisk()),
826 0, 0, NULL);
827
828 if (!xfs_inobt_cur_cache)
829 return -ENOMEM;
830 return 0;
831}
832
833void
834xfs_inobt_destroy_cur_cache(void)
835{
836 kmem_cache_destroy(xfs_inobt_cur_cache);
837 xfs_inobt_cur_cache = NULL;
838}