Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0+
2 : /*
3 : * Copyright (C) 2016 Oracle. All Rights Reserved.
4 : * Author: Darrick J. Wong <darrick.wong@oracle.com>
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_refcount_btree.h"
16 : #include "xfs_refcount.h"
17 : #include "xfs_alloc.h"
18 : #include "xfs_error.h"
19 : #include "xfs_trace.h"
20 : #include "xfs_trans.h"
21 : #include "xfs_bit.h"
22 : #include "xfs_rmap.h"
23 : #include "xfs_ag.h"
24 :
25 : static struct kmem_cache *xfs_refcountbt_cur_cache;
26 :
27 : static struct xfs_btree_cur *
28 19370923 : xfs_refcountbt_dup_cursor(
29 : struct xfs_btree_cur *cur)
30 : {
31 19370923 : return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
32 : cur->bc_ag.agbp, cur->bc_ag.pag);
33 : }
34 :
35 : STATIC void
36 3355 : xfs_refcountbt_set_root(
37 : struct xfs_btree_cur *cur,
38 : const union xfs_btree_ptr *ptr,
39 : int inc)
40 : {
41 3355 : struct xfs_buf *agbp = cur->bc_ag.agbp;
42 3355 : struct xfs_agf *agf = agbp->b_addr;
43 3355 : struct xfs_perag *pag = agbp->b_pag;
44 :
45 3355 : ASSERT(ptr->s != 0);
46 :
47 3355 : agf->agf_refcount_root = ptr->s;
48 3355 : be32_add_cpu(&agf->agf_refcount_level, inc);
49 3355 : pag->pagf_refcount_level += inc;
50 :
51 3355 : xfs_alloc_log_agf(cur->bc_tp, agbp,
52 : XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
53 3355 : }
54 :
55 : STATIC int
56 121781 : xfs_refcountbt_alloc_block(
57 : struct xfs_btree_cur *cur,
58 : const union xfs_btree_ptr *start,
59 : union xfs_btree_ptr *new,
60 : int *stat)
61 : {
62 121781 : struct xfs_buf *agbp = cur->bc_ag.agbp;
63 121781 : struct xfs_agf *agf = agbp->b_addr;
64 121781 : struct xfs_alloc_arg args; /* block allocation args */
65 121781 : int error; /* error return value */
66 :
67 121781 : memset(&args, 0, sizeof(args));
68 121781 : args.tp = cur->bc_tp;
69 121781 : args.mp = cur->bc_mp;
70 121781 : args.pag = cur->bc_ag.pag;
71 121781 : args.oinfo = XFS_RMAP_OINFO_REFC;
72 121781 : args.minlen = args.maxlen = args.prod = 1;
73 121781 : args.resv = XFS_AG_RESV_METADATA;
74 :
75 243562 : error = xfs_alloc_vextent_near_bno(&args,
76 121781 : XFS_AGB_TO_FSB(args.mp, args.pag->pag_agno,
77 : xfs_refc_block(args.mp)));
78 121781 : if (error)
79 0 : goto out_error;
80 121781 : if (args.fsbno == NULLFSBLOCK) {
81 0 : *stat = 0;
82 0 : return 0;
83 : }
84 121781 : ASSERT(args.agno == cur->bc_ag.pag->pag_agno);
85 121781 : ASSERT(args.len == 1);
86 :
87 121781 : new->s = cpu_to_be32(args.agbno);
88 121781 : be32_add_cpu(&agf->agf_refcount_blocks, 1);
89 121781 : xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
90 :
91 121781 : *stat = 1;
92 121781 : return 0;
93 :
94 : out_error:
95 0 : return error;
96 : }
97 :
98 : STATIC int
99 99642 : xfs_refcountbt_free_block(
100 : struct xfs_btree_cur *cur,
101 : struct xfs_buf *bp)
102 : {
103 99642 : struct xfs_mount *mp = cur->bc_mp;
104 99642 : struct xfs_buf *agbp = cur->bc_ag.agbp;
105 99642 : struct xfs_agf *agf = agbp->b_addr;
106 99642 : xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
107 :
108 99642 : be32_add_cpu(&agf->agf_refcount_blocks, -1);
109 99642 : xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
110 99642 : return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
111 : &XFS_RMAP_OINFO_REFC, XFS_AG_RESV_METADATA, 0);
112 : }
113 :
114 : STATIC int
115 77921996 : xfs_refcountbt_get_minrecs(
116 : struct xfs_btree_cur *cur,
117 : int level)
118 : {
119 77921996 : return cur->bc_mp->m_refc_mnr[level != 0];
120 : }
121 :
122 : STATIC int
123 14318707897 : xfs_refcountbt_get_maxrecs(
124 : struct xfs_btree_cur *cur,
125 : int level)
126 : {
127 14318707897 : return cur->bc_mp->m_refc_mxr[level != 0];
128 : }
129 :
130 : STATIC void
131 38913679721 : xfs_refcountbt_init_key_from_rec(
132 : union xfs_btree_key *key,
133 : const union xfs_btree_rec *rec)
134 : {
135 38913679721 : key->refc.rc_startblock = rec->refc.rc_startblock;
136 38913679721 : }
137 :
138 : STATIC void
139 1537435129 : xfs_refcountbt_init_high_key_from_rec(
140 : union xfs_btree_key *key,
141 : const union xfs_btree_rec *rec)
142 : {
143 1537435129 : __u32 x;
144 :
145 1537435129 : x = be32_to_cpu(rec->refc.rc_startblock);
146 1537435129 : x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
147 1537435129 : key->refc.rc_startblock = cpu_to_be32(x);
148 1537435129 : }
149 :
150 : STATIC void
151 11094154405 : xfs_refcountbt_init_rec_from_cur(
152 : struct xfs_btree_cur *cur,
153 : union xfs_btree_rec *rec)
154 : {
155 11094154405 : const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
156 11094154405 : uint32_t start;
157 :
158 11094154405 : start = xfs_refcount_encode_startblock(irec->rc_startblock,
159 11094154405 : irec->rc_domain);
160 11094154405 : rec->refc.rc_startblock = cpu_to_be32(start);
161 11094154405 : rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
162 11094154405 : rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
163 11094154405 : }
164 :
165 : STATIC void
166 5759379454 : xfs_refcountbt_init_ptr_from_cur(
167 : struct xfs_btree_cur *cur,
168 : union xfs_btree_ptr *ptr)
169 : {
170 5759379454 : struct xfs_agf *agf = cur->bc_ag.agbp->b_addr;
171 :
172 5759379454 : ASSERT(cur->bc_ag.pag->pag_agno == be32_to_cpu(agf->agf_seqno));
173 :
174 5759379454 : ptr->s = agf->agf_refcount_root;
175 5759379454 : }
176 :
177 : STATIC int64_t
178 36286780873 : xfs_refcountbt_key_diff(
179 : struct xfs_btree_cur *cur,
180 : const union xfs_btree_key *key)
181 : {
182 36286780873 : const struct xfs_refcount_key *kp = &key->refc;
183 36286780873 : const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
184 36286780873 : uint32_t start;
185 :
186 36286780873 : start = xfs_refcount_encode_startblock(irec->rc_startblock,
187 36286780873 : irec->rc_domain);
188 36286780873 : return (int64_t)be32_to_cpu(kp->rc_startblock) - start;
189 : }
190 :
191 : STATIC int64_t
192 5712379484 : xfs_refcountbt_diff_two_keys(
193 : struct xfs_btree_cur *cur,
194 : const union xfs_btree_key *k1,
195 : const union xfs_btree_key *k2,
196 : const union xfs_btree_key *mask)
197 : {
198 5712379484 : ASSERT(!mask || mask->refc.rc_startblock);
199 :
200 5712379484 : return (int64_t)be32_to_cpu(k1->refc.rc_startblock) -
201 5712379484 : be32_to_cpu(k2->refc.rc_startblock);
202 : }
203 :
204 : STATIC xfs_failaddr_t
205 2321834 : xfs_refcountbt_verify(
206 : struct xfs_buf *bp)
207 : {
208 2321834 : struct xfs_mount *mp = bp->b_mount;
209 2321834 : struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
210 2321834 : struct xfs_perag *pag = bp->b_pag;
211 2321834 : xfs_failaddr_t fa;
212 2321834 : unsigned int level;
213 :
214 2321834 : if (!xfs_verify_magic(bp, block->bb_magic))
215 0 : return __this_address;
216 :
217 2321648 : if (!xfs_has_reflink(mp))
218 0 : return __this_address;
219 2321648 : fa = xfs_btree_sblock_v5hdr_verify(bp);
220 2321723 : if (fa)
221 : return fa;
222 :
223 2321704 : level = be16_to_cpu(block->bb_level);
224 4615921 : if (pag && xfs_perag_initialised_agf(pag)) {
225 1696620 : unsigned int maxlevel = pag->pagf_refcount_level;
226 :
227 : #ifdef CONFIG_XFS_ONLINE_REPAIR
228 : /*
229 : * Online repair could be rewriting the refcount btree, so
230 : * we'll validate against the larger of either tree while this
231 : * is going on.
232 : */
233 1696620 : maxlevel = max_t(unsigned int, maxlevel,
234 : pag->pagf_alt_refcount_level);
235 : #endif
236 1696620 : if (level >= maxlevel)
237 0 : return __this_address;
238 625084 : } else if (level >= mp->m_refc_maxlevels)
239 0 : return __this_address;
240 :
241 2321704 : return xfs_btree_sblock_verify(bp, mp->m_refc_mxr[level != 0]);
242 : }
243 :
244 : STATIC void
245 290611 : xfs_refcountbt_read_verify(
246 : struct xfs_buf *bp)
247 : {
248 290611 : xfs_failaddr_t fa;
249 :
250 290611 : if (!xfs_btree_sblock_verify_crc(bp))
251 20 : xfs_verifier_error(bp, -EFSBADCRC, __this_address);
252 : else {
253 290591 : fa = xfs_refcountbt_verify(bp);
254 290591 : if (fa)
255 0 : xfs_verifier_error(bp, -EFSCORRUPTED, fa);
256 : }
257 :
258 290611 : if (bp->b_error)
259 20 : trace_xfs_btree_corrupt(bp, _RET_IP_);
260 290611 : }
261 :
262 : STATIC void
263 1056520 : xfs_refcountbt_write_verify(
264 : struct xfs_buf *bp)
265 : {
266 1056520 : xfs_failaddr_t fa;
267 :
268 1056520 : fa = xfs_refcountbt_verify(bp);
269 1056558 : if (fa) {
270 0 : trace_xfs_btree_corrupt(bp, _RET_IP_);
271 0 : xfs_verifier_error(bp, -EFSCORRUPTED, fa);
272 0 : return;
273 : }
274 1056558 : xfs_btree_sblock_calc_crc(bp);
275 :
276 : }
277 :
278 : const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
279 : .name = "xfs_refcountbt",
280 : .magic = { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) },
281 : .verify_read = xfs_refcountbt_read_verify,
282 : .verify_write = xfs_refcountbt_write_verify,
283 : .verify_struct = xfs_refcountbt_verify,
284 : };
285 :
286 : STATIC int
287 300716 : xfs_refcountbt_keys_inorder(
288 : struct xfs_btree_cur *cur,
289 : const union xfs_btree_key *k1,
290 : const union xfs_btree_key *k2)
291 : {
292 300716 : return be32_to_cpu(k1->refc.rc_startblock) <
293 300716 : be32_to_cpu(k2->refc.rc_startblock);
294 : }
295 :
296 : STATIC int
297 208279529 : xfs_refcountbt_recs_inorder(
298 : struct xfs_btree_cur *cur,
299 : const union xfs_btree_rec *r1,
300 : const union xfs_btree_rec *r2)
301 : {
302 208279529 : return be32_to_cpu(r1->refc.rc_startblock) +
303 208279529 : be32_to_cpu(r1->refc.rc_blockcount) <=
304 208279529 : be32_to_cpu(r2->refc.rc_startblock);
305 : }
306 :
307 : STATIC enum xbtree_key_contig
308 0 : xfs_refcountbt_keys_contiguous(
309 : struct xfs_btree_cur *cur,
310 : const union xfs_btree_key *key1,
311 : const union xfs_btree_key *key2,
312 : const union xfs_btree_key *mask)
313 : {
314 0 : ASSERT(!mask || mask->refc.rc_startblock);
315 :
316 0 : return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
317 0 : be32_to_cpu(key2->refc.rc_startblock));
318 : }
319 :
320 : const struct xfs_btree_ops xfs_refcountbt_ops = {
321 : .rec_len = sizeof(struct xfs_refcount_rec),
322 : .key_len = sizeof(struct xfs_refcount_key),
323 : .lru_refs = XFS_REFC_BTREE_REF,
324 :
325 : .dup_cursor = xfs_refcountbt_dup_cursor,
326 : .set_root = xfs_refcountbt_set_root,
327 : .alloc_block = xfs_refcountbt_alloc_block,
328 : .free_block = xfs_refcountbt_free_block,
329 : .get_minrecs = xfs_refcountbt_get_minrecs,
330 : .get_maxrecs = xfs_refcountbt_get_maxrecs,
331 : .init_key_from_rec = xfs_refcountbt_init_key_from_rec,
332 : .init_high_key_from_rec = xfs_refcountbt_init_high_key_from_rec,
333 : .init_rec_from_cur = xfs_refcountbt_init_rec_from_cur,
334 : .init_ptr_from_cur = xfs_refcountbt_init_ptr_from_cur,
335 : .key_diff = xfs_refcountbt_key_diff,
336 : .buf_ops = &xfs_refcountbt_buf_ops,
337 : .diff_two_keys = xfs_refcountbt_diff_two_keys,
338 : .keys_inorder = xfs_refcountbt_keys_inorder,
339 : .recs_inorder = xfs_refcountbt_recs_inorder,
340 : .keys_contiguous = xfs_refcountbt_keys_contiguous,
341 : };
342 :
343 : /*
344 : * Initialize a new refcount btree cursor.
345 : */
346 : static struct xfs_btree_cur *
347 1885641692 : xfs_refcountbt_init_common(
348 : struct xfs_mount *mp,
349 : struct xfs_trans *tp,
350 : struct xfs_perag *pag)
351 : {
352 1885641692 : struct xfs_btree_cur *cur;
353 :
354 1885641692 : ASSERT(pag->pag_agno < mp->m_sb.sb_agcount);
355 :
356 1885641692 : cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_REFC,
357 1885641692 : &xfs_refcountbt_ops, mp->m_refc_maxlevels,
358 : xfs_refcountbt_cur_cache);
359 1887155218 : cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2);
360 :
361 1887155218 : cur->bc_ag.pag = xfs_perag_hold(pag);
362 1890043546 : cur->bc_ag.refc.nr_ops = 0;
363 1890043546 : cur->bc_ag.refc.shape_changes = 0;
364 1890043546 : return cur;
365 : }
366 :
367 : /* Create a btree cursor. */
368 : struct xfs_btree_cur *
369 1888141190 : xfs_refcountbt_init_cursor(
370 : struct xfs_mount *mp,
371 : struct xfs_trans *tp,
372 : struct xfs_buf *agbp,
373 : struct xfs_perag *pag)
374 : {
375 1888141190 : struct xfs_agf *agf = agbp->b_addr;
376 1888141190 : struct xfs_btree_cur *cur;
377 :
378 1888141190 : cur = xfs_refcountbt_init_common(mp, tp, pag);
379 1890050788 : cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
380 1890050788 : cur->bc_ag.agbp = agbp;
381 1890050788 : return cur;
382 : }
383 :
384 : /* Create a btree cursor with a fake root for staging. */
385 : struct xfs_btree_cur *
386 91902 : xfs_refcountbt_stage_cursor(
387 : struct xfs_mount *mp,
388 : struct xbtree_afakeroot *afake,
389 : struct xfs_perag *pag)
390 : {
391 91902 : struct xfs_btree_cur *cur;
392 :
393 91902 : cur = xfs_refcountbt_init_common(mp, NULL, pag);
394 92016 : xfs_btree_stage_afakeroot(cur, afake);
395 91903 : return cur;
396 : }
397 :
398 : /*
399 : * Swap in the new btree root. Once we pass this point the newly rebuilt btree
400 : * is in place and we have to kill off all the old btree blocks.
401 : */
402 : void
403 92047 : xfs_refcountbt_commit_staged_btree(
404 : struct xfs_btree_cur *cur,
405 : struct xfs_trans *tp,
406 : struct xfs_buf *agbp)
407 : {
408 92047 : struct xfs_agf *agf = agbp->b_addr;
409 92047 : struct xbtree_afakeroot *afake = cur->bc_ag.afake;
410 :
411 92047 : ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
412 :
413 92047 : agf->agf_refcount_root = cpu_to_be32(afake->af_root);
414 92047 : agf->agf_refcount_level = cpu_to_be32(afake->af_levels);
415 92047 : agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks);
416 92047 : xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS |
417 : XFS_AGF_REFCOUNT_ROOT |
418 : XFS_AGF_REFCOUNT_LEVEL);
419 92034 : xfs_btree_commit_afakeroot(cur, tp, agbp, &xfs_refcountbt_ops);
420 92028 : }
421 :
422 : /* Calculate number of records in a refcount btree block. */
423 : static inline unsigned int
424 : xfs_refcountbt_block_maxrecs(
425 : unsigned int blocklen,
426 : bool leaf)
427 : {
428 133964 : if (leaf)
429 66982 : return blocklen / sizeof(struct xfs_refcount_rec);
430 66982 : return blocklen / (sizeof(struct xfs_refcount_key) +
431 : sizeof(xfs_refcount_ptr_t));
432 : }
433 :
434 : /*
435 : * Calculate the number of records in a refcount btree block.
436 : */
437 : unsigned int
438 133964 : xfs_refcountbt_maxrecs(
439 : struct xfs_mount *mp,
440 : unsigned int blocklen,
441 : bool leaf)
442 : {
443 133964 : blocklen -= XFS_REFCOUNT_BLOCK_LEN;
444 133964 : return xfs_refcountbt_block_maxrecs(blocklen, leaf);
445 : }
446 :
447 : /* Compute the max possible height of the maximally sized refcount btree. */
448 : unsigned int
449 62366 : xfs_refcountbt_maxlevels_ondisk(void)
450 : {
451 62366 : unsigned int minrecs[2];
452 62366 : unsigned int blocklen;
453 :
454 62366 : blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
455 :
456 62366 : minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2;
457 62366 : minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2;
458 :
459 62366 : return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS);
460 : }
461 :
462 : /* Compute the maximum height of a refcount btree. */
463 : void
464 66972 : xfs_refcountbt_compute_maxlevels(
465 : struct xfs_mount *mp)
466 : {
467 66972 : if (!xfs_has_reflink(mp)) {
468 4665 : mp->m_refc_maxlevels = 0;
469 4665 : return;
470 : }
471 :
472 124614 : mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(
473 62307 : mp->m_refc_mnr, mp->m_sb.sb_agblocks);
474 62307 : ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk());
475 : }
476 :
477 : /* Calculate the refcount btree size for some records. */
478 : xfs_extlen_t
479 2994531 : xfs_refcountbt_calc_size(
480 : struct xfs_mount *mp,
481 : unsigned long long len)
482 : {
483 4214735 : return xfs_btree_calc_size(mp->m_refc_mnr, len);
484 : }
485 :
486 : /*
487 : * Calculate the maximum refcount btree size.
488 : */
489 : xfs_extlen_t
490 0 : xfs_refcountbt_max_size(
491 : struct xfs_mount *mp,
492 : xfs_agblock_t agblocks)
493 : {
494 : /* Bail out if we're uninitialized, which can happen in mkfs. */
495 0 : if (mp->m_refc_mxr[0] == 0)
496 : return 0;
497 :
498 1220189 : return xfs_refcountbt_calc_size(mp, agblocks);
499 : }
500 :
501 : /*
502 : * Figure out how many blocks to reserve and how many are used by this btree.
503 : */
504 : int
505 1277122 : xfs_refcountbt_calc_reserves(
506 : struct xfs_mount *mp,
507 : struct xfs_trans *tp,
508 : struct xfs_perag *pag,
509 : xfs_extlen_t *ask,
510 : xfs_extlen_t *used)
511 : {
512 1277122 : struct xfs_buf *agbp;
513 1277122 : struct xfs_agf *agf;
514 1277122 : xfs_agblock_t agblocks;
515 1277122 : xfs_extlen_t tree_len;
516 1277122 : int error;
517 :
518 1277122 : if (!xfs_has_reflink(mp))
519 : return 0;
520 :
521 1220867 : error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
522 1221349 : if (error)
523 : return error;
524 :
525 1220921 : agf = agbp->b_addr;
526 1220921 : agblocks = be32_to_cpu(agf->agf_length);
527 1220921 : tree_len = be32_to_cpu(agf->agf_refcount_blocks);
528 1220921 : xfs_trans_brelse(tp, agbp);
529 :
530 : /*
531 : * The log is permanently allocated, so the space it occupies will
532 : * never be available for the kinds of things that would require btree
533 : * expansion. We therefore can pretend the space isn't there.
534 : */
535 1220229 : if (xfs_ag_contains_log(mp, pag->pag_agno))
536 114867 : agblocks -= mp->m_sb.sb_logblocks;
537 :
538 1219981 : *ask += xfs_refcountbt_max_size(mp, agblocks);
539 1219996 : *used += tree_len;
540 :
541 1219996 : return error;
542 : }
543 :
544 : int __init
545 59 : xfs_refcountbt_init_cur_cache(void)
546 : {
547 59 : xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur",
548 59 : xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()),
549 : 0, 0, NULL);
550 :
551 59 : if (!xfs_refcountbt_cur_cache)
552 0 : return -ENOMEM;
553 : return 0;
554 : }
555 :
556 : void
557 58 : xfs_refcountbt_destroy_cur_cache(void)
558 : {
559 58 : kmem_cache_destroy(xfs_refcountbt_cur_cache);
560 58 : xfs_refcountbt_cur_cache = NULL;
561 58 : }
|
