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_mount.h"
13#include "xfs_btree.h"
14#include "xfs_btree_staging.h"
15#include "xfs_alloc_btree.h"
16#include "xfs_alloc.h"
17#include "xfs_extent_busy.h"
18#include "xfs_error.h"
19#include "xfs_trace.h"
20#include "xfs_trans.h"
21#include "xfs_ag.h"
22
23static struct kmem_cache *xfs_allocbt_cur_cache;
24
25STATIC struct xfs_btree_cur *
26xfs_allocbt_dup_cursor(
27 struct xfs_btree_cur *cur)
28{
29 return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
30 cur->bc_ag.agbp, cur->bc_ag.pag, cur->bc_btnum);
31}
32
33STATIC void
34xfs_allocbt_set_root(
35 struct xfs_btree_cur *cur,
36 const union xfs_btree_ptr *ptr,
37 int inc)
38{
39 struct xfs_buf *agbp = cur->bc_ag.agbp;
40 struct xfs_agf *agf = agbp->b_addr;
41 int btnum = cur->bc_btnum;
42
43 ASSERT(ptr->s != 0);
44
45 agf->agf_roots[btnum] = ptr->s;
46 be32_add_cpu(&agf->agf_levels[btnum], inc);
47 cur->bc_ag.pag->pagf_levels[btnum] += inc;
48
49 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
50}
51
52STATIC int
53xfs_allocbt_alloc_block(
54 struct xfs_btree_cur *cur,
55 const union xfs_btree_ptr *start,
56 union xfs_btree_ptr *new,
57 int *stat)
58{
59 int error;
60 xfs_agblock_t bno;
61
62 /* Allocate the new block from the freelist. If we can't, give up. */
63 error = xfs_alloc_get_freelist(cur->bc_ag.pag, cur->bc_tp,
64 cur->bc_ag.agbp, &bno, 1);
65 if (error)
66 return error;
67
68 if (bno == NULLAGBLOCK) {
69 *stat = 0;
70 return 0;
71 }
72
73 atomic64_inc(&cur->bc_mp->m_allocbt_blks);
74 xfs_extent_busy_reuse(cur->bc_mp, cur->bc_ag.pag, bno, 1, false);
75
76 new->s = cpu_to_be32(bno);
77
78 *stat = 1;
79 return 0;
80}
81
82STATIC int
83xfs_allocbt_free_block(
84 struct xfs_btree_cur *cur,
85 struct xfs_buf *bp)
86{
87 struct xfs_buf *agbp = cur->bc_ag.agbp;
88 xfs_agblock_t bno;
89 int error;
90
91 bno = xfs_daddr_to_agbno(cur->bc_mp, xfs_buf_daddr(bp));
92 error = xfs_alloc_put_freelist(cur->bc_ag.pag, cur->bc_tp, agbp, NULL,
93 bno, 1);
94 if (error)
95 return error;
96
97 atomic64_dec(&cur->bc_mp->m_allocbt_blks);
98 xfs_extent_busy_insert(cur->bc_tp, agbp->b_pag, bno, 1,
99 XFS_EXTENT_BUSY_SKIP_DISCARD);
100 return 0;
101}
102
103/*
104 * Update the longest extent in the AGF
105 */
106STATIC void
107xfs_allocbt_update_lastrec(
108 struct xfs_btree_cur *cur,
109 const struct xfs_btree_block *block,
110 const union xfs_btree_rec *rec,
111 int ptr,
112 int reason)
113{
114 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
115 struct xfs_perag *pag;
116 __be32 len;
117 int numrecs;
118
119 ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
120
121 switch (reason) {
122 case LASTREC_UPDATE:
123 /*
124 * If this is the last leaf block and it's the last record,
125 * then update the size of the longest extent in the AG.
126 */
127 if (ptr != xfs_btree_get_numrecs(block))
128 return;
129 len = rec->alloc.ar_blockcount;
130 break;
131 case LASTREC_INSREC:
132 if (be32_to_cpu(rec->alloc.ar_blockcount) <=
133 be32_to_cpu(agf->agf_longest))
134 return;
135 len = rec->alloc.ar_blockcount;
136 break;
137 case LASTREC_DELREC:
138 numrecs = xfs_btree_get_numrecs(block);
139 if (ptr <= numrecs)
140 return;
141 ASSERT(ptr == numrecs + 1);
142
143 if (numrecs) {
144 xfs_alloc_rec_t *rrp;
145
146 rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
147 len = rrp->ar_blockcount;
148 } else {
149 len = 0;
150 }
151
152 break;
153 default:
154 ASSERT(0);
155 return;
156 }
157
158 agf->agf_longest = len;
159 pag = cur->bc_ag.agbp->b_pag;
160 pag->pagf_longest = be32_to_cpu(len);
161 xfs_alloc_log_agf(cur->bc_tp, cur->bc_ag.agbp, XFS_AGF_LONGEST);
162}
163
164STATIC int
165xfs_allocbt_get_minrecs(
166 struct xfs_btree_cur *cur,
167 int level)
168{
169 return cur->bc_mp->m_alloc_mnr[level != 0];
170}
171
172STATIC int
173xfs_allocbt_get_maxrecs(
174 struct xfs_btree_cur *cur,
175 int level)
176{
177 return cur->bc_mp->m_alloc_mxr[level != 0];
178}
179
180STATIC void
181xfs_allocbt_init_key_from_rec(
182 union xfs_btree_key *key,
183 const union xfs_btree_rec *rec)
184{
185 key->alloc.ar_startblock = rec->alloc.ar_startblock;
186 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
187}
188
189STATIC void
190xfs_bnobt_init_high_key_from_rec(
191 union xfs_btree_key *key,
192 const union xfs_btree_rec *rec)
193{
194 __u32 x;
195
196 x = be32_to_cpu(rec->alloc.ar_startblock);
197 x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
198 key->alloc.ar_startblock = cpu_to_be32(x);
199 key->alloc.ar_blockcount = 0;
200}
201
202STATIC void
203xfs_cntbt_init_high_key_from_rec(
204 union xfs_btree_key *key,
205 const union xfs_btree_rec *rec)
206{
207 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
208 key->alloc.ar_startblock = 0;
209}
210
211STATIC void
212xfs_allocbt_init_rec_from_cur(
213 struct xfs_btree_cur *cur,
214 union xfs_btree_rec *rec)
215{
216 rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
217 rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
218}
219
220STATIC void
221xfs_allocbt_init_ptr_from_cur(
222 struct xfs_btree_cur *cur,
223 union xfs_btree_ptr *ptr)
224{
225 struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
226
227 ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
228
229 ptr->s = agf->agf_roots[cur->bc_btnum];
230}
231
232STATIC int64_t
233xfs_bnobt_key_diff(
234 struct xfs_btree_cur *cur,
235 const union xfs_btree_key *key)
236{
237 struct xfs_alloc_rec_incore *rec = &cur->bc_rec.a;
238 const struct xfs_alloc_rec *kp = &key->alloc;
239
240 return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
241}
242
243STATIC int64_t
244xfs_cntbt_key_diff(
245 struct xfs_btree_cur *cur,
246 const union xfs_btree_key *key)
247{
248 struct xfs_alloc_rec_incore *rec = &cur->bc_rec.a;
249 const struct xfs_alloc_rec *kp = &key->alloc;
250 int64_t diff;
251
252 diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
253 if (diff)
254 return diff;
255
256 return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
257}
258
259STATIC int64_t
260xfs_bnobt_diff_two_keys(
261 struct xfs_btree_cur *cur,
262 const union xfs_btree_key *k1,
263 const union xfs_btree_key *k2,
264 const union xfs_btree_key *mask)
265{
266 ASSERT(!mask || mask->alloc.ar_startblock);
267
268 return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
269 be32_to_cpu(k2->alloc.ar_startblock);
270}
271
272STATIC int64_t
273xfs_cntbt_diff_two_keys(
274 struct xfs_btree_cur *cur,
275 const union xfs_btree_key *k1,
276 const union xfs_btree_key *k2,
277 const union xfs_btree_key *mask)
278{
279 int64_t diff;
280
281 ASSERT(!mask || (mask->alloc.ar_blockcount &&
282 mask->alloc.ar_startblock));
283
284 diff = be32_to_cpu(k1->alloc.ar_blockcount) -
285 be32_to_cpu(k2->alloc.ar_blockcount);
286 if (diff)
287 return diff;
288
289 return be32_to_cpu(k1->alloc.ar_startblock) -
290 be32_to_cpu(k2->alloc.ar_startblock);
291}
292
293static xfs_failaddr_t
294xfs_allocbt_verify(
295 struct xfs_buf *bp)
296{
297 struct xfs_mount *mp = bp->b_mount;
298 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
299 struct xfs_perag *pag = bp->b_pag;
300 xfs_failaddr_t fa;
301 unsigned int level;
302 xfs_btnum_t btnum = XFS_BTNUM_BNOi;
303
304 if (!xfs_verify_magic(bp, block->bb_magic))
305 return __this_address;
306
307 if (xfs_has_crc(mp)) {
308 fa = xfs_btree_sblock_v5hdr_verify(bp);
309 if (fa)
310 return fa;
311 }
312
313 /*
314 * The perag may not be attached during grow operations or fully
315 * initialized from the AGF during log recovery. Therefore we can only
316 * check against maximum tree depth from those contexts.
317 *
318 * Otherwise check against the per-tree limit. Peek at one of the
319 * verifier magic values to determine the type of tree we're verifying
320 * against.
321 */
322 level = be16_to_cpu(block->bb_level);
323 if (bp->b_ops->magic[0] == cpu_to_be32(XFS_ABTC_MAGIC))
324 btnum = XFS_BTNUM_CNTi;
325 if (pag && xfs_perag_initialised_agf(pag)) {
326 unsigned int maxlevel = pag->pagf_levels[btnum];
327
328#ifdef CONFIG_XFS_ONLINE_REPAIR
329 /*
330 * Online repair could be rewriting the free space btrees, so
331 * we'll validate against the larger of either tree while this
332 * is going on.
333 */
334 maxlevel = max_t(unsigned int, maxlevel,
335 pag->pagf_repair_levels[btnum]);
336#endif
337 if (level >= maxlevel)
338 return __this_address;
339 } else if (level >= mp->m_alloc_maxlevels)
340 return __this_address;
341
342 return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
343}
344
345static void
346xfs_allocbt_read_verify(
347 struct xfs_buf *bp)
348{
349 xfs_failaddr_t fa;
350
351 if (!xfs_btree_sblock_verify_crc(bp))
352 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
353 else {
354 fa = xfs_allocbt_verify(bp);
355 if (fa)
356 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
357 }
358
359 if (bp->b_error)
360 trace_xfs_btree_corrupt(bp, _RET_IP_);
361}
362
363static void
364xfs_allocbt_write_verify(
365 struct xfs_buf *bp)
366{
367 xfs_failaddr_t fa;
368
369 fa = xfs_allocbt_verify(bp);
370 if (fa) {
371 trace_xfs_btree_corrupt(bp, _RET_IP_);
372 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
373 return;
374 }
375 xfs_btree_sblock_calc_crc(bp);
376
377}
378
379const struct xfs_buf_ops xfs_bnobt_buf_ops = {
380 .name = "xfs_bnobt",
381 .magic = { cpu_to_be32(XFS_ABTB_MAGIC),
382 cpu_to_be32(XFS_ABTB_CRC_MAGIC) },
383 .verify_read = xfs_allocbt_read_verify,
384 .verify_write = xfs_allocbt_write_verify,
385 .verify_struct = xfs_allocbt_verify,
386};
387
388const struct xfs_buf_ops xfs_cntbt_buf_ops = {
389 .name = "xfs_cntbt",
390 .magic = { cpu_to_be32(XFS_ABTC_MAGIC),
391 cpu_to_be32(XFS_ABTC_CRC_MAGIC) },
392 .verify_read = xfs_allocbt_read_verify,
393 .verify_write = xfs_allocbt_write_verify,
394 .verify_struct = xfs_allocbt_verify,
395};
396
397STATIC int
398xfs_bnobt_keys_inorder(
399 struct xfs_btree_cur *cur,
400 const union xfs_btree_key *k1,
401 const union xfs_btree_key *k2)
402{
403 return be32_to_cpu(k1->alloc.ar_startblock) <
404 be32_to_cpu(k2->alloc.ar_startblock);
405}
406
407STATIC int
408xfs_bnobt_recs_inorder(
409 struct xfs_btree_cur *cur,
410 const union xfs_btree_rec *r1,
411 const union xfs_btree_rec *r2)
412{
413 return be32_to_cpu(r1->alloc.ar_startblock) +
414 be32_to_cpu(r1->alloc.ar_blockcount) <=
415 be32_to_cpu(r2->alloc.ar_startblock);
416}
417
418STATIC int
419xfs_cntbt_keys_inorder(
420 struct xfs_btree_cur *cur,
421 const union xfs_btree_key *k1,
422 const union xfs_btree_key *k2)
423{
424 return be32_to_cpu(k1->alloc.ar_blockcount) <
425 be32_to_cpu(k2->alloc.ar_blockcount) ||
426 (k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
427 be32_to_cpu(k1->alloc.ar_startblock) <
428 be32_to_cpu(k2->alloc.ar_startblock));
429}
430
431STATIC int
432xfs_cntbt_recs_inorder(
433 struct xfs_btree_cur *cur,
434 const union xfs_btree_rec *r1,
435 const union xfs_btree_rec *r2)
436{
437 return be32_to_cpu(r1->alloc.ar_blockcount) <
438 be32_to_cpu(r2->alloc.ar_blockcount) ||
439 (r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
440 be32_to_cpu(r1->alloc.ar_startblock) <
441 be32_to_cpu(r2->alloc.ar_startblock));
442}
443
444STATIC enum xbtree_key_contig
445xfs_allocbt_keys_contiguous(
446 struct xfs_btree_cur *cur,
447 const union xfs_btree_key *key1,
448 const union xfs_btree_key *key2,
449 const union xfs_btree_key *mask)
450{
451 ASSERT(!mask || mask->alloc.ar_startblock);
452
453 return xbtree_key_contig(be32_to_cpu(key1->alloc.ar_startblock),
454 be32_to_cpu(key2->alloc.ar_startblock));
455}
456
457static const struct xfs_btree_ops xfs_bnobt_ops = {
458 .rec_len = sizeof(xfs_alloc_rec_t),
459 .key_len = sizeof(xfs_alloc_key_t),
460
461 .dup_cursor = xfs_allocbt_dup_cursor,
462 .set_root = xfs_allocbt_set_root,
463 .alloc_block = xfs_allocbt_alloc_block,
464 .free_block = xfs_allocbt_free_block,
465 .update_lastrec = xfs_allocbt_update_lastrec,
466 .get_minrecs = xfs_allocbt_get_minrecs,
467 .get_maxrecs = xfs_allocbt_get_maxrecs,
468 .init_key_from_rec = xfs_allocbt_init_key_from_rec,
469 .init_high_key_from_rec = xfs_bnobt_init_high_key_from_rec,
470 .init_rec_from_cur = xfs_allocbt_init_rec_from_cur,
471 .init_ptr_from_cur = xfs_allocbt_init_ptr_from_cur,
472 .key_diff = xfs_bnobt_key_diff,
473 .buf_ops = &xfs_bnobt_buf_ops,
474 .diff_two_keys = xfs_bnobt_diff_two_keys,
475 .keys_inorder = xfs_bnobt_keys_inorder,
476 .recs_inorder = xfs_bnobt_recs_inorder,
477 .keys_contiguous = xfs_allocbt_keys_contiguous,
478};
479
480static const struct xfs_btree_ops xfs_cntbt_ops = {
481 .rec_len = sizeof(xfs_alloc_rec_t),
482 .key_len = sizeof(xfs_alloc_key_t),
483
484 .dup_cursor = xfs_allocbt_dup_cursor,
485 .set_root = xfs_allocbt_set_root,
486 .alloc_block = xfs_allocbt_alloc_block,
487 .free_block = xfs_allocbt_free_block,
488 .update_lastrec = xfs_allocbt_update_lastrec,
489 .get_minrecs = xfs_allocbt_get_minrecs,
490 .get_maxrecs = xfs_allocbt_get_maxrecs,
491 .init_key_from_rec = xfs_allocbt_init_key_from_rec,
492 .init_high_key_from_rec = xfs_cntbt_init_high_key_from_rec,
493 .init_rec_from_cur = xfs_allocbt_init_rec_from_cur,
494 .init_ptr_from_cur = xfs_allocbt_init_ptr_from_cur,
495 .key_diff = xfs_cntbt_key_diff,
496 .buf_ops = &xfs_cntbt_buf_ops,
497 .diff_two_keys = xfs_cntbt_diff_two_keys,
498 .keys_inorder = xfs_cntbt_keys_inorder,
499 .recs_inorder = xfs_cntbt_recs_inorder,
500 .keys_contiguous = NULL, /* not needed right now */
501};
502
503/* Allocate most of a new allocation btree cursor. */
504STATIC struct xfs_btree_cur *
505xfs_allocbt_init_common(
506 struct xfs_mount *mp,
507 struct xfs_trans *tp,
508 struct xfs_perag *pag,
509 xfs_btnum_t btnum)
510{
511 struct xfs_btree_cur *cur;
512
513 ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
514
515 cur = xfs_btree_alloc_cursor(mp, tp, btnum, mp->m_alloc_maxlevels,
516 xfs_allocbt_cur_cache);
517 cur->bc_ag.abt.active = false;
518
519 if (btnum == XFS_BTNUM_CNT) {
520 cur->bc_ops = &xfs_cntbt_ops;
521 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
522 cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
523 } else {
524 cur->bc_ops = &xfs_bnobt_ops;
525 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
526 }
527
528 cur->bc_ag.pag = xfs_perag_hold(pag);
529
530 if (xfs_has_crc(mp))
531 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
532
533 return cur;
534}
535
536/*
537 * Allocate a new allocation btree cursor.
538 */
539struct xfs_btree_cur * /* new alloc btree cursor */
540xfs_allocbt_init_cursor(
541 struct xfs_mount *mp, /* file system mount point */
542 struct xfs_trans *tp, /* transaction pointer */
543 struct xfs_buf *agbp, /* buffer for agf structure */
544 struct xfs_perag *pag,
545 xfs_btnum_t btnum) /* btree identifier */
546{
547 struct xfs_agf *agf = agbp->b_addr;
548 struct xfs_btree_cur *cur;
549
550 cur = xfs_allocbt_init_common(mp, tp, pag, btnum);
551 if (btnum == XFS_BTNUM_CNT)
552 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
553 else
554 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
555
556 cur->bc_ag.agbp = agbp;
557
558 return cur;
559}
560
561/* Create a free space btree cursor with a fake root for staging. */
562struct xfs_btree_cur *
563xfs_allocbt_stage_cursor(
564 struct xfs_mount *mp,
565 struct xbtree_afakeroot *afake,
566 struct xfs_perag *pag,
567 xfs_btnum_t btnum)
568{
569 struct xfs_btree_cur *cur;
570
571 cur = xfs_allocbt_init_common(mp, NULL, pag, btnum);
572 xfs_btree_stage_afakeroot(cur, afake);
573 return cur;
574}
575
576/*
577 * Install a new free space btree root. Caller is responsible for invalidating
578 * and freeing the old btree blocks.
579 */
580void
581xfs_allocbt_commit_staged_btree(
582 struct xfs_btree_cur *cur,
583 struct xfs_trans *tp,
584 struct xfs_buf *agbp)
585{
586 struct xfs_agf *agf = agbp->b_addr;
587 struct xbtree_afakeroot *afake = cur->bc_ag.afake;
588
589 ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
590
591 agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root);
592 agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels);
593 xfs_alloc_log_agf(tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
594
595 if (cur->bc_btnum == XFS_BTNUM_BNO) {
596 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_bnobt_ops);
597 } else {
598 cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE;
599 xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_cntbt_ops);
600 }
601}
602
603/* Calculate number of records in an alloc btree block. */
604static inline unsigned int
605xfs_allocbt_block_maxrecs(
606 unsigned int blocklen,
607 bool leaf)
608{
609 if (leaf)
610 return blocklen / sizeof(xfs_alloc_rec_t);
611 return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
612}
613
614/*
615 * Calculate number of records in an alloc btree block.
616 */
617int
618xfs_allocbt_maxrecs(
619 struct xfs_mount *mp,
620 int blocklen,
621 int leaf)
622{
623 blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
624 return xfs_allocbt_block_maxrecs(blocklen, leaf);
625}
626
627/* Free space btrees are at their largest when every other block is free. */
628#define XFS_MAX_FREESP_RECORDS ((XFS_MAX_AG_BLOCKS + 1) / 2)
629
630/* Compute the max possible height for free space btrees. */
631unsigned int
632xfs_allocbt_maxlevels_ondisk(void)
633{
634 unsigned int minrecs[2];
635 unsigned int blocklen;
636
637 blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
638 XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
639
640 minrecs[0] = xfs_allocbt_block_maxrecs(blocklen, true) / 2;
641 minrecs[1] = xfs_allocbt_block_maxrecs(blocklen, false) / 2;
642
643 return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_FREESP_RECORDS);
644}
645
646/* Calculate the freespace btree size for some records. */
647xfs_extlen_t
648xfs_allocbt_calc_size(
649 struct xfs_mount *mp,
650 unsigned long long len)
651{
652 return xfs_btree_calc_size(mp->m_alloc_mnr, len);
653}
654
655int __init
656xfs_allocbt_init_cur_cache(void)
657{
658 xfs_allocbt_cur_cache = kmem_cache_create("xfs_bnobt_cur",
659 xfs_btree_cur_sizeof(xfs_allocbt_maxlevels_ondisk()),
660 0, 0, NULL);
661
662 if (!xfs_allocbt_cur_cache)
663 return -ENOMEM;
664 return 0;
665}
666
667void
668xfs_allocbt_destroy_cur_cache(void)
669{
670 kmem_cache_destroy(xfs_allocbt_cur_cache);
671 xfs_allocbt_cur_cache = NULL;
672}
1/*
2 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_sb.h"
25#include "xfs_mount.h"
26#include "xfs_btree.h"
27#include "xfs_alloc_btree.h"
28#include "xfs_alloc.h"
29#include "xfs_extent_busy.h"
30#include "xfs_error.h"
31#include "xfs_trace.h"
32#include "xfs_cksum.h"
33#include "xfs_trans.h"
34
35
36STATIC struct xfs_btree_cur *
37xfs_allocbt_dup_cursor(
38 struct xfs_btree_cur *cur)
39{
40 return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
41 cur->bc_private.a.agbp, cur->bc_private.a.agno,
42 cur->bc_btnum);
43}
44
45STATIC void
46xfs_allocbt_set_root(
47 struct xfs_btree_cur *cur,
48 union xfs_btree_ptr *ptr,
49 int inc)
50{
51 struct xfs_buf *agbp = cur->bc_private.a.agbp;
52 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
53 xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno);
54 int btnum = cur->bc_btnum;
55 struct xfs_perag *pag = xfs_perag_get(cur->bc_mp, seqno);
56
57 ASSERT(ptr->s != 0);
58
59 agf->agf_roots[btnum] = ptr->s;
60 be32_add_cpu(&agf->agf_levels[btnum], inc);
61 pag->pagf_levels[btnum] += inc;
62 xfs_perag_put(pag);
63
64 xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
65}
66
67STATIC int
68xfs_allocbt_alloc_block(
69 struct xfs_btree_cur *cur,
70 union xfs_btree_ptr *start,
71 union xfs_btree_ptr *new,
72 int *stat)
73{
74 int error;
75 xfs_agblock_t bno;
76
77 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
78
79 /* Allocate the new block from the freelist. If we can't, give up. */
80 error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp,
81 &bno, 1);
82 if (error) {
83 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
84 return error;
85 }
86
87 if (bno == NULLAGBLOCK) {
88 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
89 *stat = 0;
90 return 0;
91 }
92
93 xfs_extent_busy_reuse(cur->bc_mp, cur->bc_private.a.agno, bno, 1, false);
94
95 xfs_trans_agbtree_delta(cur->bc_tp, 1);
96 new->s = cpu_to_be32(bno);
97
98 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
99 *stat = 1;
100 return 0;
101}
102
103STATIC int
104xfs_allocbt_free_block(
105 struct xfs_btree_cur *cur,
106 struct xfs_buf *bp)
107{
108 struct xfs_buf *agbp = cur->bc_private.a.agbp;
109 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
110 xfs_agblock_t bno;
111 int error;
112
113 bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
114 error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
115 if (error)
116 return error;
117
118 xfs_extent_busy_insert(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1,
119 XFS_EXTENT_BUSY_SKIP_DISCARD);
120 xfs_trans_agbtree_delta(cur->bc_tp, -1);
121 return 0;
122}
123
124/*
125 * Update the longest extent in the AGF
126 */
127STATIC void
128xfs_allocbt_update_lastrec(
129 struct xfs_btree_cur *cur,
130 struct xfs_btree_block *block,
131 union xfs_btree_rec *rec,
132 int ptr,
133 int reason)
134{
135 struct xfs_agf *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
136 xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno);
137 struct xfs_perag *pag;
138 __be32 len;
139 int numrecs;
140
141 ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
142
143 switch (reason) {
144 case LASTREC_UPDATE:
145 /*
146 * If this is the last leaf block and it's the last record,
147 * then update the size of the longest extent in the AG.
148 */
149 if (ptr != xfs_btree_get_numrecs(block))
150 return;
151 len = rec->alloc.ar_blockcount;
152 break;
153 case LASTREC_INSREC:
154 if (be32_to_cpu(rec->alloc.ar_blockcount) <=
155 be32_to_cpu(agf->agf_longest))
156 return;
157 len = rec->alloc.ar_blockcount;
158 break;
159 case LASTREC_DELREC:
160 numrecs = xfs_btree_get_numrecs(block);
161 if (ptr <= numrecs)
162 return;
163 ASSERT(ptr == numrecs + 1);
164
165 if (numrecs) {
166 xfs_alloc_rec_t *rrp;
167
168 rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
169 len = rrp->ar_blockcount;
170 } else {
171 len = 0;
172 }
173
174 break;
175 default:
176 ASSERT(0);
177 return;
178 }
179
180 agf->agf_longest = len;
181 pag = xfs_perag_get(cur->bc_mp, seqno);
182 pag->pagf_longest = be32_to_cpu(len);
183 xfs_perag_put(pag);
184 xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, XFS_AGF_LONGEST);
185}
186
187STATIC int
188xfs_allocbt_get_minrecs(
189 struct xfs_btree_cur *cur,
190 int level)
191{
192 return cur->bc_mp->m_alloc_mnr[level != 0];
193}
194
195STATIC int
196xfs_allocbt_get_maxrecs(
197 struct xfs_btree_cur *cur,
198 int level)
199{
200 return cur->bc_mp->m_alloc_mxr[level != 0];
201}
202
203STATIC void
204xfs_allocbt_init_key_from_rec(
205 union xfs_btree_key *key,
206 union xfs_btree_rec *rec)
207{
208 ASSERT(rec->alloc.ar_startblock != 0);
209
210 key->alloc.ar_startblock = rec->alloc.ar_startblock;
211 key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
212}
213
214STATIC void
215xfs_allocbt_init_rec_from_cur(
216 struct xfs_btree_cur *cur,
217 union xfs_btree_rec *rec)
218{
219 ASSERT(cur->bc_rec.a.ar_startblock != 0);
220
221 rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
222 rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
223}
224
225STATIC void
226xfs_allocbt_init_ptr_from_cur(
227 struct xfs_btree_cur *cur,
228 union xfs_btree_ptr *ptr)
229{
230 struct xfs_agf *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
231
232 ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
233 ASSERT(agf->agf_roots[cur->bc_btnum] != 0);
234
235 ptr->s = agf->agf_roots[cur->bc_btnum];
236}
237
238STATIC __int64_t
239xfs_allocbt_key_diff(
240 struct xfs_btree_cur *cur,
241 union xfs_btree_key *key)
242{
243 xfs_alloc_rec_incore_t *rec = &cur->bc_rec.a;
244 xfs_alloc_key_t *kp = &key->alloc;
245 __int64_t diff;
246
247 if (cur->bc_btnum == XFS_BTNUM_BNO) {
248 return (__int64_t)be32_to_cpu(kp->ar_startblock) -
249 rec->ar_startblock;
250 }
251
252 diff = (__int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
253 if (diff)
254 return diff;
255
256 return (__int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
257}
258
259static bool
260xfs_allocbt_verify(
261 struct xfs_buf *bp)
262{
263 struct xfs_mount *mp = bp->b_target->bt_mount;
264 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
265 struct xfs_perag *pag = bp->b_pag;
266 unsigned int level;
267
268 /*
269 * magic number and level verification
270 *
271 * During growfs operations, we can't verify the exact level or owner as
272 * the perag is not fully initialised and hence not attached to the
273 * buffer. In this case, check against the maximum tree depth.
274 *
275 * Similarly, during log recovery we will have a perag structure
276 * attached, but the agf information will not yet have been initialised
277 * from the on disk AGF. Again, we can only check against maximum limits
278 * in this case.
279 */
280 level = be16_to_cpu(block->bb_level);
281 switch (block->bb_magic) {
282 case cpu_to_be32(XFS_ABTB_CRC_MAGIC):
283 if (!xfs_btree_sblock_v5hdr_verify(bp))
284 return false;
285 /* fall through */
286 case cpu_to_be32(XFS_ABTB_MAGIC):
287 if (pag && pag->pagf_init) {
288 if (level >= pag->pagf_levels[XFS_BTNUM_BNOi])
289 return false;
290 } else if (level >= mp->m_ag_maxlevels)
291 return false;
292 break;
293 case cpu_to_be32(XFS_ABTC_CRC_MAGIC):
294 if (!xfs_btree_sblock_v5hdr_verify(bp))
295 return false;
296 /* fall through */
297 case cpu_to_be32(XFS_ABTC_MAGIC):
298 if (pag && pag->pagf_init) {
299 if (level >= pag->pagf_levels[XFS_BTNUM_CNTi])
300 return false;
301 } else if (level >= mp->m_ag_maxlevels)
302 return false;
303 break;
304 default:
305 return false;
306 }
307
308 return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
309}
310
311static void
312xfs_allocbt_read_verify(
313 struct xfs_buf *bp)
314{
315 if (!xfs_btree_sblock_verify_crc(bp))
316 xfs_buf_ioerror(bp, -EFSBADCRC);
317 else if (!xfs_allocbt_verify(bp))
318 xfs_buf_ioerror(bp, -EFSCORRUPTED);
319
320 if (bp->b_error) {
321 trace_xfs_btree_corrupt(bp, _RET_IP_);
322 xfs_verifier_error(bp);
323 }
324}
325
326static void
327xfs_allocbt_write_verify(
328 struct xfs_buf *bp)
329{
330 if (!xfs_allocbt_verify(bp)) {
331 trace_xfs_btree_corrupt(bp, _RET_IP_);
332 xfs_buf_ioerror(bp, -EFSCORRUPTED);
333 xfs_verifier_error(bp);
334 return;
335 }
336 xfs_btree_sblock_calc_crc(bp);
337
338}
339
340const struct xfs_buf_ops xfs_allocbt_buf_ops = {
341 .name = "xfs_allocbt",
342 .verify_read = xfs_allocbt_read_verify,
343 .verify_write = xfs_allocbt_write_verify,
344};
345
346
347#if defined(DEBUG) || defined(XFS_WARN)
348STATIC int
349xfs_allocbt_keys_inorder(
350 struct xfs_btree_cur *cur,
351 union xfs_btree_key *k1,
352 union xfs_btree_key *k2)
353{
354 if (cur->bc_btnum == XFS_BTNUM_BNO) {
355 return be32_to_cpu(k1->alloc.ar_startblock) <
356 be32_to_cpu(k2->alloc.ar_startblock);
357 } else {
358 return be32_to_cpu(k1->alloc.ar_blockcount) <
359 be32_to_cpu(k2->alloc.ar_blockcount) ||
360 (k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
361 be32_to_cpu(k1->alloc.ar_startblock) <
362 be32_to_cpu(k2->alloc.ar_startblock));
363 }
364}
365
366STATIC int
367xfs_allocbt_recs_inorder(
368 struct xfs_btree_cur *cur,
369 union xfs_btree_rec *r1,
370 union xfs_btree_rec *r2)
371{
372 if (cur->bc_btnum == XFS_BTNUM_BNO) {
373 return be32_to_cpu(r1->alloc.ar_startblock) +
374 be32_to_cpu(r1->alloc.ar_blockcount) <=
375 be32_to_cpu(r2->alloc.ar_startblock);
376 } else {
377 return be32_to_cpu(r1->alloc.ar_blockcount) <
378 be32_to_cpu(r2->alloc.ar_blockcount) ||
379 (r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
380 be32_to_cpu(r1->alloc.ar_startblock) <
381 be32_to_cpu(r2->alloc.ar_startblock));
382 }
383}
384#endif /* DEBUG */
385
386static const struct xfs_btree_ops xfs_allocbt_ops = {
387 .rec_len = sizeof(xfs_alloc_rec_t),
388 .key_len = sizeof(xfs_alloc_key_t),
389
390 .dup_cursor = xfs_allocbt_dup_cursor,
391 .set_root = xfs_allocbt_set_root,
392 .alloc_block = xfs_allocbt_alloc_block,
393 .free_block = xfs_allocbt_free_block,
394 .update_lastrec = xfs_allocbt_update_lastrec,
395 .get_minrecs = xfs_allocbt_get_minrecs,
396 .get_maxrecs = xfs_allocbt_get_maxrecs,
397 .init_key_from_rec = xfs_allocbt_init_key_from_rec,
398 .init_rec_from_cur = xfs_allocbt_init_rec_from_cur,
399 .init_ptr_from_cur = xfs_allocbt_init_ptr_from_cur,
400 .key_diff = xfs_allocbt_key_diff,
401 .buf_ops = &xfs_allocbt_buf_ops,
402#if defined(DEBUG) || defined(XFS_WARN)
403 .keys_inorder = xfs_allocbt_keys_inorder,
404 .recs_inorder = xfs_allocbt_recs_inorder,
405#endif
406};
407
408/*
409 * Allocate a new allocation btree cursor.
410 */
411struct xfs_btree_cur * /* new alloc btree cursor */
412xfs_allocbt_init_cursor(
413 struct xfs_mount *mp, /* file system mount point */
414 struct xfs_trans *tp, /* transaction pointer */
415 struct xfs_buf *agbp, /* buffer for agf structure */
416 xfs_agnumber_t agno, /* allocation group number */
417 xfs_btnum_t btnum) /* btree identifier */
418{
419 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
420 struct xfs_btree_cur *cur;
421
422 ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
423
424 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
425
426 cur->bc_tp = tp;
427 cur->bc_mp = mp;
428 cur->bc_btnum = btnum;
429 cur->bc_blocklog = mp->m_sb.sb_blocklog;
430 cur->bc_ops = &xfs_allocbt_ops;
431 if (btnum == XFS_BTNUM_BNO)
432 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
433 else
434 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
435
436 if (btnum == XFS_BTNUM_CNT) {
437 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
438 cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
439 } else {
440 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
441 }
442
443 cur->bc_private.a.agbp = agbp;
444 cur->bc_private.a.agno = agno;
445
446 if (xfs_sb_version_hascrc(&mp->m_sb))
447 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
448
449 return cur;
450}
451
452/*
453 * Calculate number of records in an alloc btree block.
454 */
455int
456xfs_allocbt_maxrecs(
457 struct xfs_mount *mp,
458 int blocklen,
459 int leaf)
460{
461 blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
462
463 if (leaf)
464 return blocklen / sizeof(xfs_alloc_rec_t);
465 return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
466}