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