]>
Commit | Line | Data |
---|---|---|
035e00ac DW |
1 | /* |
2 | * Copyright (c) 2014 Red Hat, 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_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
27 | #include "xfs_defer.h" | |
28 | #include "xfs_inode.h" | |
29 | #include "xfs_trans.h" | |
30 | #include "xfs_alloc.h" | |
31 | #include "xfs_btree.h" | |
4b8ed677 | 32 | #include "xfs_rmap.h" |
035e00ac DW |
33 | #include "xfs_rmap_btree.h" |
34 | #include "xfs_trace.h" | |
35 | #include "xfs_cksum.h" | |
36 | #include "xfs_error.h" | |
37 | #include "xfs_extent_busy.h" | |
38 | ||
4b8ed677 DW |
39 | /* |
40 | * Reverse map btree. | |
41 | * | |
42 | * This is a per-ag tree used to track the owner(s) of a given extent. With | |
43 | * reflink it is possible for there to be multiple owners, which is a departure | |
44 | * from classic XFS. Owner records for data extents are inserted when the | |
45 | * extent is mapped and removed when an extent is unmapped. Owner records for | |
46 | * all other block types (i.e. metadata) are inserted when an extent is | |
47 | * allocated and removed when an extent is freed. There can only be one owner | |
48 | * of a metadata extent, usually an inode or some other metadata structure like | |
49 | * an AG btree. | |
50 | * | |
51 | * The rmap btree is part of the free space management, so blocks for the tree | |
52 | * are sourced from the agfl. Hence we need transaction reservation support for | |
53 | * this tree so that the freelist is always large enough. This also impacts on | |
54 | * the minimum space we need to leave free in the AG. | |
55 | * | |
56 | * The tree is ordered by [ag block, owner, offset]. This is a large key size, | |
57 | * but it is the only way to enforce unique keys when a block can be owned by | |
58 | * multiple files at any offset. There's no need to order/search by extent | |
59 | * size for online updating/management of the tree. It is intended that most | |
60 | * reverse lookups will be to find the owner(s) of a particular block, or to | |
61 | * try to recover tree and file data from corrupt primary metadata. | |
62 | */ | |
63 | ||
035e00ac DW |
64 | static struct xfs_btree_cur * |
65 | xfs_rmapbt_dup_cursor( | |
66 | struct xfs_btree_cur *cur) | |
67 | { | |
68 | return xfs_rmapbt_init_cursor(cur->bc_mp, cur->bc_tp, | |
69 | cur->bc_private.a.agbp, cur->bc_private.a.agno); | |
70 | } | |
71 | ||
4b8ed677 DW |
72 | STATIC void |
73 | xfs_rmapbt_set_root( | |
74 | struct xfs_btree_cur *cur, | |
75 | union xfs_btree_ptr *ptr, | |
76 | int inc) | |
77 | { | |
78 | struct xfs_buf *agbp = cur->bc_private.a.agbp; | |
79 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
80 | xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno); | |
81 | int btnum = cur->bc_btnum; | |
82 | struct xfs_perag *pag = xfs_perag_get(cur->bc_mp, seqno); | |
83 | ||
84 | ASSERT(ptr->s != 0); | |
85 | ||
86 | agf->agf_roots[btnum] = ptr->s; | |
87 | be32_add_cpu(&agf->agf_levels[btnum], inc); | |
88 | pag->pagf_levels[btnum] += inc; | |
89 | xfs_perag_put(pag); | |
90 | ||
91 | xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS); | |
92 | } | |
93 | ||
94 | STATIC int | |
95 | xfs_rmapbt_alloc_block( | |
96 | struct xfs_btree_cur *cur, | |
97 | union xfs_btree_ptr *start, | |
98 | union xfs_btree_ptr *new, | |
99 | int *stat) | |
100 | { | |
f32866fd DW |
101 | struct xfs_buf *agbp = cur->bc_private.a.agbp; |
102 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
4b8ed677 DW |
103 | int error; |
104 | xfs_agblock_t bno; | |
105 | ||
106 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
107 | ||
108 | /* Allocate the new block from the freelist. If we can't, give up. */ | |
109 | error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp, | |
110 | &bno, 1); | |
111 | if (error) { | |
112 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
113 | return error; | |
114 | } | |
115 | ||
116 | trace_xfs_rmapbt_alloc_block(cur->bc_mp, cur->bc_private.a.agno, | |
117 | bno, 1); | |
118 | if (bno == NULLAGBLOCK) { | |
119 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
120 | *stat = 0; | |
121 | return 0; | |
122 | } | |
123 | ||
124 | xfs_extent_busy_reuse(cur->bc_mp, cur->bc_private.a.agno, bno, 1, | |
125 | false); | |
126 | ||
127 | xfs_trans_agbtree_delta(cur->bc_tp, 1); | |
128 | new->s = cpu_to_be32(bno); | |
f32866fd DW |
129 | be32_add_cpu(&agf->agf_rmap_blocks, 1); |
130 | xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_RMAP_BLOCKS); | |
4b8ed677 DW |
131 | |
132 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
133 | *stat = 1; | |
134 | return 0; | |
135 | } | |
136 | ||
137 | STATIC int | |
138 | xfs_rmapbt_free_block( | |
139 | struct xfs_btree_cur *cur, | |
140 | struct xfs_buf *bp) | |
141 | { | |
142 | struct xfs_buf *agbp = cur->bc_private.a.agbp; | |
143 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
144 | xfs_agblock_t bno; | |
145 | int error; | |
146 | ||
147 | bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp)); | |
148 | trace_xfs_rmapbt_free_block(cur->bc_mp, cur->bc_private.a.agno, | |
149 | bno, 1); | |
f32866fd DW |
150 | be32_add_cpu(&agf->agf_rmap_blocks, -1); |
151 | xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_RMAP_BLOCKS); | |
4b8ed677 DW |
152 | error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1); |
153 | if (error) | |
154 | return error; | |
155 | ||
156 | xfs_extent_busy_insert(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1, | |
157 | XFS_EXTENT_BUSY_SKIP_DISCARD); | |
158 | xfs_trans_agbtree_delta(cur->bc_tp, -1); | |
159 | ||
160 | return 0; | |
161 | } | |
162 | ||
163 | STATIC int | |
164 | xfs_rmapbt_get_minrecs( | |
165 | struct xfs_btree_cur *cur, | |
166 | int level) | |
167 | { | |
168 | return cur->bc_mp->m_rmap_mnr[level != 0]; | |
169 | } | |
170 | ||
171 | STATIC int | |
172 | xfs_rmapbt_get_maxrecs( | |
173 | struct xfs_btree_cur *cur, | |
174 | int level) | |
175 | { | |
176 | return cur->bc_mp->m_rmap_mxr[level != 0]; | |
177 | } | |
178 | ||
179 | STATIC void | |
180 | xfs_rmapbt_init_key_from_rec( | |
181 | union xfs_btree_key *key, | |
182 | union xfs_btree_rec *rec) | |
183 | { | |
184 | key->rmap.rm_startblock = rec->rmap.rm_startblock; | |
185 | key->rmap.rm_owner = rec->rmap.rm_owner; | |
186 | key->rmap.rm_offset = rec->rmap.rm_offset; | |
187 | } | |
188 | ||
cfed56ae DW |
189 | /* |
190 | * The high key for a reverse mapping record can be computed by shifting | |
191 | * the startblock and offset to the highest value that would still map | |
192 | * to that record. In practice this means that we add blockcount-1 to | |
193 | * the startblock for all records, and if the record is for a data/attr | |
194 | * fork mapping, we add blockcount-1 to the offset too. | |
195 | */ | |
196 | STATIC void | |
197 | xfs_rmapbt_init_high_key_from_rec( | |
198 | union xfs_btree_key *key, | |
199 | union xfs_btree_rec *rec) | |
200 | { | |
201 | __uint64_t off; | |
202 | int adj; | |
203 | ||
204 | adj = be32_to_cpu(rec->rmap.rm_blockcount) - 1; | |
205 | ||
206 | key->rmap.rm_startblock = rec->rmap.rm_startblock; | |
207 | be32_add_cpu(&key->rmap.rm_startblock, adj); | |
208 | key->rmap.rm_owner = rec->rmap.rm_owner; | |
209 | key->rmap.rm_offset = rec->rmap.rm_offset; | |
210 | if (XFS_RMAP_NON_INODE_OWNER(be64_to_cpu(rec->rmap.rm_owner)) || | |
211 | XFS_RMAP_IS_BMBT_BLOCK(be64_to_cpu(rec->rmap.rm_offset))) | |
212 | return; | |
213 | off = be64_to_cpu(key->rmap.rm_offset); | |
214 | off = (XFS_RMAP_OFF(off) + adj) | (off & ~XFS_RMAP_OFF_MASK); | |
215 | key->rmap.rm_offset = cpu_to_be64(off); | |
216 | } | |
217 | ||
4b8ed677 DW |
218 | STATIC void |
219 | xfs_rmapbt_init_rec_from_cur( | |
220 | struct xfs_btree_cur *cur, | |
221 | union xfs_btree_rec *rec) | |
222 | { | |
223 | rec->rmap.rm_startblock = cpu_to_be32(cur->bc_rec.r.rm_startblock); | |
224 | rec->rmap.rm_blockcount = cpu_to_be32(cur->bc_rec.r.rm_blockcount); | |
225 | rec->rmap.rm_owner = cpu_to_be64(cur->bc_rec.r.rm_owner); | |
226 | rec->rmap.rm_offset = cpu_to_be64( | |
227 | xfs_rmap_irec_offset_pack(&cur->bc_rec.r)); | |
228 | } | |
229 | ||
230 | STATIC void | |
231 | xfs_rmapbt_init_ptr_from_cur( | |
232 | struct xfs_btree_cur *cur, | |
233 | union xfs_btree_ptr *ptr) | |
234 | { | |
235 | struct xfs_agf *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
236 | ||
237 | ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno)); | |
238 | ASSERT(agf->agf_roots[cur->bc_btnum] != 0); | |
239 | ||
240 | ptr->s = agf->agf_roots[cur->bc_btnum]; | |
241 | } | |
242 | ||
243 | STATIC __int64_t | |
244 | xfs_rmapbt_key_diff( | |
245 | struct xfs_btree_cur *cur, | |
246 | union xfs_btree_key *key) | |
247 | { | |
248 | struct xfs_rmap_irec *rec = &cur->bc_rec.r; | |
249 | struct xfs_rmap_key *kp = &key->rmap; | |
250 | __u64 x, y; | |
251 | __int64_t d; | |
252 | ||
253 | d = (__int64_t)be32_to_cpu(kp->rm_startblock) - rec->rm_startblock; | |
254 | if (d) | |
255 | return d; | |
256 | ||
257 | x = be64_to_cpu(kp->rm_owner); | |
258 | y = rec->rm_owner; | |
259 | if (x > y) | |
260 | return 1; | |
261 | else if (y > x) | |
262 | return -1; | |
263 | ||
264 | x = XFS_RMAP_OFF(be64_to_cpu(kp->rm_offset)); | |
265 | y = rec->rm_offset; | |
266 | if (x > y) | |
267 | return 1; | |
268 | else if (y > x) | |
269 | return -1; | |
270 | return 0; | |
271 | } | |
272 | ||
cfed56ae DW |
273 | STATIC __int64_t |
274 | xfs_rmapbt_diff_two_keys( | |
275 | struct xfs_btree_cur *cur, | |
276 | union xfs_btree_key *k1, | |
277 | union xfs_btree_key *k2) | |
278 | { | |
279 | struct xfs_rmap_key *kp1 = &k1->rmap; | |
280 | struct xfs_rmap_key *kp2 = &k2->rmap; | |
281 | __int64_t d; | |
282 | __u64 x, y; | |
283 | ||
284 | d = (__int64_t)be32_to_cpu(kp1->rm_startblock) - | |
285 | be32_to_cpu(kp2->rm_startblock); | |
286 | if (d) | |
287 | return d; | |
288 | ||
289 | x = be64_to_cpu(kp1->rm_owner); | |
290 | y = be64_to_cpu(kp2->rm_owner); | |
291 | if (x > y) | |
292 | return 1; | |
293 | else if (y > x) | |
294 | return -1; | |
295 | ||
296 | x = XFS_RMAP_OFF(be64_to_cpu(kp1->rm_offset)); | |
297 | y = XFS_RMAP_OFF(be64_to_cpu(kp2->rm_offset)); | |
298 | if (x > y) | |
299 | return 1; | |
300 | else if (y > x) | |
301 | return -1; | |
302 | return 0; | |
303 | } | |
304 | ||
035e00ac DW |
305 | static bool |
306 | xfs_rmapbt_verify( | |
307 | struct xfs_buf *bp) | |
308 | { | |
309 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
310 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
311 | struct xfs_perag *pag = bp->b_pag; | |
312 | unsigned int level; | |
313 | ||
314 | /* | |
315 | * magic number and level verification | |
316 | * | |
317 | * During growfs operations, we can't verify the exact level or owner as | |
318 | * the perag is not fully initialised and hence not attached to the | |
319 | * buffer. In this case, check against the maximum tree depth. | |
320 | * | |
321 | * Similarly, during log recovery we will have a perag structure | |
322 | * attached, but the agf information will not yet have been initialised | |
323 | * from the on disk AGF. Again, we can only check against maximum limits | |
324 | * in this case. | |
325 | */ | |
326 | if (block->bb_magic != cpu_to_be32(XFS_RMAP_CRC_MAGIC)) | |
327 | return false; | |
328 | ||
329 | if (!xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
330 | return false; | |
331 | if (!xfs_btree_sblock_v5hdr_verify(bp)) | |
332 | return false; | |
333 | ||
334 | level = be16_to_cpu(block->bb_level); | |
335 | if (pag && pag->pagf_init) { | |
336 | if (level >= pag->pagf_levels[XFS_BTNUM_RMAPi]) | |
337 | return false; | |
338 | } else if (level >= mp->m_rmap_maxlevels) | |
339 | return false; | |
340 | ||
341 | return xfs_btree_sblock_verify(bp, mp->m_rmap_mxr[level != 0]); | |
342 | } | |
343 | ||
344 | static void | |
345 | xfs_rmapbt_read_verify( | |
346 | struct xfs_buf *bp) | |
347 | { | |
348 | if (!xfs_btree_sblock_verify_crc(bp)) | |
349 | xfs_buf_ioerror(bp, -EFSBADCRC); | |
350 | else if (!xfs_rmapbt_verify(bp)) | |
351 | xfs_buf_ioerror(bp, -EFSCORRUPTED); | |
352 | ||
353 | if (bp->b_error) { | |
354 | trace_xfs_btree_corrupt(bp, _RET_IP_); | |
355 | xfs_verifier_error(bp); | |
356 | } | |
357 | } | |
358 | ||
359 | static void | |
360 | xfs_rmapbt_write_verify( | |
361 | struct xfs_buf *bp) | |
362 | { | |
363 | if (!xfs_rmapbt_verify(bp)) { | |
364 | trace_xfs_btree_corrupt(bp, _RET_IP_); | |
365 | xfs_buf_ioerror(bp, -EFSCORRUPTED); | |
366 | xfs_verifier_error(bp); | |
367 | return; | |
368 | } | |
369 | xfs_btree_sblock_calc_crc(bp); | |
370 | ||
371 | } | |
372 | ||
373 | const struct xfs_buf_ops xfs_rmapbt_buf_ops = { | |
374 | .name = "xfs_rmapbt", | |
375 | .verify_read = xfs_rmapbt_read_verify, | |
376 | .verify_write = xfs_rmapbt_write_verify, | |
377 | }; | |
378 | ||
4b8ed677 DW |
379 | #if defined(DEBUG) || defined(XFS_WARN) |
380 | STATIC int | |
381 | xfs_rmapbt_keys_inorder( | |
382 | struct xfs_btree_cur *cur, | |
383 | union xfs_btree_key *k1, | |
384 | union xfs_btree_key *k2) | |
385 | { | |
386 | __uint32_t x; | |
387 | __uint32_t y; | |
388 | __uint64_t a; | |
389 | __uint64_t b; | |
390 | ||
391 | x = be32_to_cpu(k1->rmap.rm_startblock); | |
392 | y = be32_to_cpu(k2->rmap.rm_startblock); | |
393 | if (x < y) | |
394 | return 1; | |
395 | else if (x > y) | |
396 | return 0; | |
397 | a = be64_to_cpu(k1->rmap.rm_owner); | |
398 | b = be64_to_cpu(k2->rmap.rm_owner); | |
399 | if (a < b) | |
400 | return 1; | |
401 | else if (a > b) | |
402 | return 0; | |
403 | a = XFS_RMAP_OFF(be64_to_cpu(k1->rmap.rm_offset)); | |
404 | b = XFS_RMAP_OFF(be64_to_cpu(k2->rmap.rm_offset)); | |
405 | if (a <= b) | |
406 | return 1; | |
407 | return 0; | |
408 | } | |
409 | ||
410 | STATIC int | |
411 | xfs_rmapbt_recs_inorder( | |
412 | struct xfs_btree_cur *cur, | |
413 | union xfs_btree_rec *r1, | |
414 | union xfs_btree_rec *r2) | |
415 | { | |
416 | __uint32_t x; | |
417 | __uint32_t y; | |
418 | __uint64_t a; | |
419 | __uint64_t b; | |
420 | ||
421 | x = be32_to_cpu(r1->rmap.rm_startblock); | |
422 | y = be32_to_cpu(r2->rmap.rm_startblock); | |
423 | if (x < y) | |
424 | return 1; | |
425 | else if (x > y) | |
426 | return 0; | |
427 | a = be64_to_cpu(r1->rmap.rm_owner); | |
428 | b = be64_to_cpu(r2->rmap.rm_owner); | |
429 | if (a < b) | |
430 | return 1; | |
431 | else if (a > b) | |
432 | return 0; | |
433 | a = XFS_RMAP_OFF(be64_to_cpu(r1->rmap.rm_offset)); | |
434 | b = XFS_RMAP_OFF(be64_to_cpu(r2->rmap.rm_offset)); | |
435 | if (a <= b) | |
436 | return 1; | |
437 | return 0; | |
438 | } | |
439 | #endif /* DEBUG */ | |
440 | ||
035e00ac DW |
441 | static const struct xfs_btree_ops xfs_rmapbt_ops = { |
442 | .rec_len = sizeof(struct xfs_rmap_rec), | |
443 | .key_len = 2 * sizeof(struct xfs_rmap_key), | |
444 | ||
445 | .dup_cursor = xfs_rmapbt_dup_cursor, | |
4b8ed677 DW |
446 | .set_root = xfs_rmapbt_set_root, |
447 | .alloc_block = xfs_rmapbt_alloc_block, | |
448 | .free_block = xfs_rmapbt_free_block, | |
449 | .get_minrecs = xfs_rmapbt_get_minrecs, | |
450 | .get_maxrecs = xfs_rmapbt_get_maxrecs, | |
451 | .init_key_from_rec = xfs_rmapbt_init_key_from_rec, | |
cfed56ae | 452 | .init_high_key_from_rec = xfs_rmapbt_init_high_key_from_rec, |
4b8ed677 DW |
453 | .init_rec_from_cur = xfs_rmapbt_init_rec_from_cur, |
454 | .init_ptr_from_cur = xfs_rmapbt_init_ptr_from_cur, | |
455 | .key_diff = xfs_rmapbt_key_diff, | |
035e00ac | 456 | .buf_ops = &xfs_rmapbt_buf_ops, |
cfed56ae | 457 | .diff_two_keys = xfs_rmapbt_diff_two_keys, |
4b8ed677 DW |
458 | #if defined(DEBUG) || defined(XFS_WARN) |
459 | .keys_inorder = xfs_rmapbt_keys_inorder, | |
460 | .recs_inorder = xfs_rmapbt_recs_inorder, | |
461 | #endif | |
035e00ac DW |
462 | }; |
463 | ||
464 | /* | |
465 | * Allocate a new allocation btree cursor. | |
466 | */ | |
467 | struct xfs_btree_cur * | |
468 | xfs_rmapbt_init_cursor( | |
469 | struct xfs_mount *mp, | |
470 | struct xfs_trans *tp, | |
471 | struct xfs_buf *agbp, | |
472 | xfs_agnumber_t agno) | |
473 | { | |
474 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
475 | struct xfs_btree_cur *cur; | |
476 | ||
477 | cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); | |
478 | cur->bc_tp = tp; | |
479 | cur->bc_mp = mp; | |
cfed56ae | 480 | /* Overlapping btree; 2 keys per pointer. */ |
035e00ac | 481 | cur->bc_btnum = XFS_BTNUM_RMAP; |
cfed56ae | 482 | cur->bc_flags = XFS_BTREE_CRC_BLOCKS | XFS_BTREE_OVERLAPPING; |
035e00ac DW |
483 | cur->bc_blocklog = mp->m_sb.sb_blocklog; |
484 | cur->bc_ops = &xfs_rmapbt_ops; | |
485 | cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]); | |
486 | ||
487 | cur->bc_private.a.agbp = agbp; | |
488 | cur->bc_private.a.agno = agno; | |
489 | ||
490 | return cur; | |
491 | } | |
492 | ||
493 | /* | |
494 | * Calculate number of records in an rmap btree block. | |
495 | */ | |
496 | int | |
497 | xfs_rmapbt_maxrecs( | |
498 | struct xfs_mount *mp, | |
499 | int blocklen, | |
500 | int leaf) | |
501 | { | |
502 | blocklen -= XFS_RMAP_BLOCK_LEN; | |
503 | ||
504 | if (leaf) | |
505 | return blocklen / sizeof(struct xfs_rmap_rec); | |
506 | return blocklen / | |
cfed56ae | 507 | (2 * sizeof(struct xfs_rmap_key) + sizeof(xfs_rmap_ptr_t)); |
035e00ac DW |
508 | } |
509 | ||
510 | /* Compute the maximum height of an rmap btree. */ | |
511 | void | |
512 | xfs_rmapbt_compute_maxlevels( | |
513 | struct xfs_mount *mp) | |
514 | { | |
46eeb521 DW |
515 | /* |
516 | * On a non-reflink filesystem, the maximum number of rmap | |
517 | * records is the number of blocks in the AG, hence the max | |
518 | * rmapbt height is log_$maxrecs($agblocks). However, with | |
519 | * reflink each AG block can have up to 2^32 (per the refcount | |
520 | * record format) owners, which means that theoretically we | |
521 | * could face up to 2^64 rmap records. | |
522 | * | |
523 | * That effectively means that the max rmapbt height must be | |
524 | * XFS_BTREE_MAXLEVELS. "Fortunately" we'll run out of AG | |
525 | * blocks to feed the rmapbt long before the rmapbt reaches | |
526 | * maximum height. The reflink code uses ag_resv_critical to | |
527 | * disallow reflinking when less than 10% of the per-AG metadata | |
528 | * block reservation since the fallback is a regular file copy. | |
529 | */ | |
530 | if (xfs_sb_version_hasreflink(&mp->m_sb)) | |
531 | mp->m_rmap_maxlevels = XFS_BTREE_MAXLEVELS; | |
532 | else | |
533 | mp->m_rmap_maxlevels = xfs_btree_compute_maxlevels(mp, | |
534 | mp->m_rmap_mnr, mp->m_sb.sb_agblocks); | |
035e00ac | 535 | } |