2 * Copyright (c) 2014 Red Hat, Inc.
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.
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.
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
18 #include "libxfs_priv.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.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"
33 #include "xfs_rmap_btree.h"
34 #include "xfs_trace.h"
35 #include "xfs_cksum.h"
36 #include "xfs_ag_resv.h"
41 * This is a per-ag tree used to track the owner(s) of a given extent. With
42 * reflink it is possible for there to be multiple owners, which is a departure
43 * from classic XFS. Owner records for data extents are inserted when the
44 * extent is mapped and removed when an extent is unmapped. Owner records for
45 * all other block types (i.e. metadata) are inserted when an extent is
46 * allocated and removed when an extent is freed. There can only be one owner
47 * of a metadata extent, usually an inode or some other metadata structure like
50 * The rmap btree is part of the free space management, so blocks for the tree
51 * are sourced from the agfl. Hence we need transaction reservation support for
52 * this tree so that the freelist is always large enough. This also impacts on
53 * the minimum space we need to leave free in the AG.
55 * The tree is ordered by [ag block, owner, offset]. This is a large key size,
56 * but it is the only way to enforce unique keys when a block can be owned by
57 * multiple files at any offset. There's no need to order/search by extent
58 * size for online updating/management of the tree. It is intended that most
59 * reverse lookups will be to find the owner(s) of a particular block, or to
60 * try to recover tree and file data from corrupt primary metadata.
63 static struct xfs_btree_cur
*
64 xfs_rmapbt_dup_cursor(
65 struct xfs_btree_cur
*cur
)
67 return xfs_rmapbt_init_cursor(cur
->bc_mp
, cur
->bc_tp
,
68 cur
->bc_private
.a
.agbp
, cur
->bc_private
.a
.agno
);
73 struct xfs_btree_cur
*cur
,
74 union xfs_btree_ptr
*ptr
,
77 struct xfs_buf
*agbp
= cur
->bc_private
.a
.agbp
;
78 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
79 xfs_agnumber_t seqno
= be32_to_cpu(agf
->agf_seqno
);
80 int btnum
= cur
->bc_btnum
;
81 struct xfs_perag
*pag
= xfs_perag_get(cur
->bc_mp
, seqno
);
85 agf
->agf_roots
[btnum
] = ptr
->s
;
86 be32_add_cpu(&agf
->agf_levels
[btnum
], inc
);
87 pag
->pagf_levels
[btnum
] += inc
;
90 xfs_alloc_log_agf(cur
->bc_tp
, agbp
, XFS_AGF_ROOTS
| XFS_AGF_LEVELS
);
94 xfs_rmapbt_alloc_block(
95 struct xfs_btree_cur
*cur
,
96 union xfs_btree_ptr
*start
,
97 union xfs_btree_ptr
*new,
100 struct xfs_buf
*agbp
= cur
->bc_private
.a
.agbp
;
101 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
105 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
107 /* Allocate the new block from the freelist. If we can't, give up. */
108 error
= xfs_alloc_get_freelist(cur
->bc_tp
, cur
->bc_private
.a
.agbp
,
111 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
115 trace_xfs_rmapbt_alloc_block(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
117 if (bno
== NULLAGBLOCK
) {
118 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
123 xfs_extent_busy_reuse(cur
->bc_mp
, cur
->bc_private
.a
.agno
, bno
, 1,
126 xfs_trans_agbtree_delta(cur
->bc_tp
, 1);
127 new->s
= cpu_to_be32(bno
);
128 be32_add_cpu(&agf
->agf_rmap_blocks
, 1);
129 xfs_alloc_log_agf(cur
->bc_tp
, agbp
, XFS_AGF_RMAP_BLOCKS
);
131 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
137 xfs_rmapbt_free_block(
138 struct xfs_btree_cur
*cur
,
141 struct xfs_buf
*agbp
= cur
->bc_private
.a
.agbp
;
142 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
146 bno
= xfs_daddr_to_agbno(cur
->bc_mp
, XFS_BUF_ADDR(bp
));
147 trace_xfs_rmapbt_free_block(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
149 be32_add_cpu(&agf
->agf_rmap_blocks
, -1);
150 xfs_alloc_log_agf(cur
->bc_tp
, agbp
, XFS_AGF_RMAP_BLOCKS
);
151 error
= xfs_alloc_put_freelist(cur
->bc_tp
, agbp
, NULL
, bno
, 1);
155 xfs_extent_busy_insert(cur
->bc_tp
, be32_to_cpu(agf
->agf_seqno
), bno
, 1,
156 XFS_EXTENT_BUSY_SKIP_DISCARD
);
157 xfs_trans_agbtree_delta(cur
->bc_tp
, -1);
163 xfs_rmapbt_get_minrecs(
164 struct xfs_btree_cur
*cur
,
167 return cur
->bc_mp
->m_rmap_mnr
[level
!= 0];
171 xfs_rmapbt_get_maxrecs(
172 struct xfs_btree_cur
*cur
,
175 return cur
->bc_mp
->m_rmap_mxr
[level
!= 0];
179 xfs_rmapbt_init_key_from_rec(
180 union xfs_btree_key
*key
,
181 union xfs_btree_rec
*rec
)
183 key
->rmap
.rm_startblock
= rec
->rmap
.rm_startblock
;
184 key
->rmap
.rm_owner
= rec
->rmap
.rm_owner
;
185 key
->rmap
.rm_offset
= rec
->rmap
.rm_offset
;
189 * The high key for a reverse mapping record can be computed by shifting
190 * the startblock and offset to the highest value that would still map
191 * to that record. In practice this means that we add blockcount-1 to
192 * the startblock for all records, and if the record is for a data/attr
193 * fork mapping, we add blockcount-1 to the offset too.
196 xfs_rmapbt_init_high_key_from_rec(
197 union xfs_btree_key
*key
,
198 union xfs_btree_rec
*rec
)
203 adj
= be32_to_cpu(rec
->rmap
.rm_blockcount
) - 1;
205 key
->rmap
.rm_startblock
= rec
->rmap
.rm_startblock
;
206 be32_add_cpu(&key
->rmap
.rm_startblock
, adj
);
207 key
->rmap
.rm_owner
= rec
->rmap
.rm_owner
;
208 key
->rmap
.rm_offset
= rec
->rmap
.rm_offset
;
209 if (XFS_RMAP_NON_INODE_OWNER(be64_to_cpu(rec
->rmap
.rm_owner
)) ||
210 XFS_RMAP_IS_BMBT_BLOCK(be64_to_cpu(rec
->rmap
.rm_offset
)))
212 off
= be64_to_cpu(key
->rmap
.rm_offset
);
213 off
= (XFS_RMAP_OFF(off
) + adj
) | (off
& ~XFS_RMAP_OFF_MASK
);
214 key
->rmap
.rm_offset
= cpu_to_be64(off
);
218 xfs_rmapbt_init_rec_from_cur(
219 struct xfs_btree_cur
*cur
,
220 union xfs_btree_rec
*rec
)
222 rec
->rmap
.rm_startblock
= cpu_to_be32(cur
->bc_rec
.r
.rm_startblock
);
223 rec
->rmap
.rm_blockcount
= cpu_to_be32(cur
->bc_rec
.r
.rm_blockcount
);
224 rec
->rmap
.rm_owner
= cpu_to_be64(cur
->bc_rec
.r
.rm_owner
);
225 rec
->rmap
.rm_offset
= cpu_to_be64(
226 xfs_rmap_irec_offset_pack(&cur
->bc_rec
.r
));
230 xfs_rmapbt_init_ptr_from_cur(
231 struct xfs_btree_cur
*cur
,
232 union xfs_btree_ptr
*ptr
)
234 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(cur
->bc_private
.a
.agbp
);
236 ASSERT(cur
->bc_private
.a
.agno
== be32_to_cpu(agf
->agf_seqno
));
237 ASSERT(agf
->agf_roots
[cur
->bc_btnum
] != 0);
239 ptr
->s
= agf
->agf_roots
[cur
->bc_btnum
];
244 struct xfs_btree_cur
*cur
,
245 union xfs_btree_key
*key
)
247 struct xfs_rmap_irec
*rec
= &cur
->bc_rec
.r
;
248 struct xfs_rmap_key
*kp
= &key
->rmap
;
252 d
= (int64_t)be32_to_cpu(kp
->rm_startblock
) - rec
->rm_startblock
;
256 x
= be64_to_cpu(kp
->rm_owner
);
263 x
= XFS_RMAP_OFF(be64_to_cpu(kp
->rm_offset
));
273 xfs_rmapbt_diff_two_keys(
274 struct xfs_btree_cur
*cur
,
275 union xfs_btree_key
*k1
,
276 union xfs_btree_key
*k2
)
278 struct xfs_rmap_key
*kp1
= &k1
->rmap
;
279 struct xfs_rmap_key
*kp2
= &k2
->rmap
;
283 d
= (int64_t)be32_to_cpu(kp1
->rm_startblock
) -
284 be32_to_cpu(kp2
->rm_startblock
);
288 x
= be64_to_cpu(kp1
->rm_owner
);
289 y
= be64_to_cpu(kp2
->rm_owner
);
295 x
= XFS_RMAP_OFF(be64_to_cpu(kp1
->rm_offset
));
296 y
= XFS_RMAP_OFF(be64_to_cpu(kp2
->rm_offset
));
304 static xfs_failaddr_t
308 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
309 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
310 struct xfs_perag
*pag
= bp
->b_pag
;
315 * magic number and level verification
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.
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
326 if (block
->bb_magic
!= cpu_to_be32(XFS_RMAP_CRC_MAGIC
))
327 return __this_address
;
329 if (!xfs_sb_version_hasrmapbt(&mp
->m_sb
))
330 return __this_address
;
331 fa
= xfs_btree_sblock_v5hdr_verify(bp
);
335 level
= be16_to_cpu(block
->bb_level
);
336 if (pag
&& pag
->pagf_init
) {
337 if (level
>= pag
->pagf_levels
[XFS_BTNUM_RMAPi
])
338 return __this_address
;
339 } else if (level
>= mp
->m_rmap_maxlevels
)
340 return __this_address
;
342 return xfs_btree_sblock_verify(bp
, mp
->m_rmap_mxr
[level
!= 0]);
346 xfs_rmapbt_read_verify(
351 if (!xfs_btree_sblock_verify_crc(bp
))
352 xfs_verifier_error(bp
, -EFSBADCRC
, __this_address
);
354 fa
= xfs_rmapbt_verify(bp
);
356 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
360 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
364 xfs_rmapbt_write_verify(
369 fa
= xfs_rmapbt_verify(bp
);
371 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
372 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
375 xfs_btree_sblock_calc_crc(bp
);
379 const struct xfs_buf_ops xfs_rmapbt_buf_ops
= {
380 .name
= "xfs_rmapbt",
381 .verify_read
= xfs_rmapbt_read_verify
,
382 .verify_write
= xfs_rmapbt_write_verify
,
383 .verify_struct
= xfs_rmapbt_verify
,
387 xfs_rmapbt_keys_inorder(
388 struct xfs_btree_cur
*cur
,
389 union xfs_btree_key
*k1
,
390 union xfs_btree_key
*k2
)
397 x
= be32_to_cpu(k1
->rmap
.rm_startblock
);
398 y
= be32_to_cpu(k2
->rmap
.rm_startblock
);
403 a
= be64_to_cpu(k1
->rmap
.rm_owner
);
404 b
= be64_to_cpu(k2
->rmap
.rm_owner
);
409 a
= XFS_RMAP_OFF(be64_to_cpu(k1
->rmap
.rm_offset
));
410 b
= XFS_RMAP_OFF(be64_to_cpu(k2
->rmap
.rm_offset
));
417 xfs_rmapbt_recs_inorder(
418 struct xfs_btree_cur
*cur
,
419 union xfs_btree_rec
*r1
,
420 union xfs_btree_rec
*r2
)
427 x
= be32_to_cpu(r1
->rmap
.rm_startblock
);
428 y
= be32_to_cpu(r2
->rmap
.rm_startblock
);
433 a
= be64_to_cpu(r1
->rmap
.rm_owner
);
434 b
= be64_to_cpu(r2
->rmap
.rm_owner
);
439 a
= XFS_RMAP_OFF(be64_to_cpu(r1
->rmap
.rm_offset
));
440 b
= XFS_RMAP_OFF(be64_to_cpu(r2
->rmap
.rm_offset
));
446 static const struct xfs_btree_ops xfs_rmapbt_ops
= {
447 .rec_len
= sizeof(struct xfs_rmap_rec
),
448 .key_len
= 2 * sizeof(struct xfs_rmap_key
),
450 .dup_cursor
= xfs_rmapbt_dup_cursor
,
451 .set_root
= xfs_rmapbt_set_root
,
452 .alloc_block
= xfs_rmapbt_alloc_block
,
453 .free_block
= xfs_rmapbt_free_block
,
454 .get_minrecs
= xfs_rmapbt_get_minrecs
,
455 .get_maxrecs
= xfs_rmapbt_get_maxrecs
,
456 .init_key_from_rec
= xfs_rmapbt_init_key_from_rec
,
457 .init_high_key_from_rec
= xfs_rmapbt_init_high_key_from_rec
,
458 .init_rec_from_cur
= xfs_rmapbt_init_rec_from_cur
,
459 .init_ptr_from_cur
= xfs_rmapbt_init_ptr_from_cur
,
460 .key_diff
= xfs_rmapbt_key_diff
,
461 .buf_ops
= &xfs_rmapbt_buf_ops
,
462 .diff_two_keys
= xfs_rmapbt_diff_two_keys
,
463 .keys_inorder
= xfs_rmapbt_keys_inorder
,
464 .recs_inorder
= xfs_rmapbt_recs_inorder
,
468 * Allocate a new allocation btree cursor.
470 struct xfs_btree_cur
*
471 xfs_rmapbt_init_cursor(
472 struct xfs_mount
*mp
,
473 struct xfs_trans
*tp
,
474 struct xfs_buf
*agbp
,
477 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
478 struct xfs_btree_cur
*cur
;
480 cur
= kmem_zone_zalloc(xfs_btree_cur_zone
, KM_NOFS
);
483 /* Overlapping btree; 2 keys per pointer. */
484 cur
->bc_btnum
= XFS_BTNUM_RMAP
;
485 cur
->bc_flags
= XFS_BTREE_CRC_BLOCKS
| XFS_BTREE_OVERLAPPING
;
486 cur
->bc_blocklog
= mp
->m_sb
.sb_blocklog
;
487 cur
->bc_ops
= &xfs_rmapbt_ops
;
488 cur
->bc_nlevels
= be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_RMAP
]);
489 cur
->bc_statoff
= XFS_STATS_CALC_INDEX(xs_rmap_2
);
491 cur
->bc_private
.a
.agbp
= agbp
;
492 cur
->bc_private
.a
.agno
= agno
;
498 * Calculate number of records in an rmap btree block.
502 struct xfs_mount
*mp
,
506 blocklen
-= XFS_RMAP_BLOCK_LEN
;
509 return blocklen
/ sizeof(struct xfs_rmap_rec
);
511 (2 * sizeof(struct xfs_rmap_key
) + sizeof(xfs_rmap_ptr_t
));
514 /* Compute the maximum height of an rmap btree. */
516 xfs_rmapbt_compute_maxlevels(
517 struct xfs_mount
*mp
)
520 * On a non-reflink filesystem, the maximum number of rmap
521 * records is the number of blocks in the AG, hence the max
522 * rmapbt height is log_$maxrecs($agblocks). However, with
523 * reflink each AG block can have up to 2^32 (per the refcount
524 * record format) owners, which means that theoretically we
525 * could face up to 2^64 rmap records.
527 * That effectively means that the max rmapbt height must be
528 * XFS_BTREE_MAXLEVELS. "Fortunately" we'll run out of AG
529 * blocks to feed the rmapbt long before the rmapbt reaches
530 * maximum height. The reflink code uses ag_resv_critical to
531 * disallow reflinking when less than 10% of the per-AG metadata
532 * block reservation since the fallback is a regular file copy.
534 if (xfs_sb_version_hasreflink(&mp
->m_sb
))
535 mp
->m_rmap_maxlevels
= XFS_BTREE_MAXLEVELS
;
537 mp
->m_rmap_maxlevels
= xfs_btree_compute_maxlevels(mp
,
538 mp
->m_rmap_mnr
, mp
->m_sb
.sb_agblocks
);
541 /* Calculate the refcount btree size for some records. */
543 xfs_rmapbt_calc_size(
544 struct xfs_mount
*mp
,
545 unsigned long long len
)
547 return xfs_btree_calc_size(mp
, mp
->m_rmap_mnr
, len
);
551 * Calculate the maximum refcount btree size.
555 struct xfs_mount
*mp
,
556 xfs_agblock_t agblocks
)
558 /* Bail out if we're uninitialized, which can happen in mkfs. */
559 if (mp
->m_rmap_mxr
[0] == 0)
562 return xfs_rmapbt_calc_size(mp
, agblocks
);
566 * Figure out how many blocks to reserve and how many are used by this btree.
569 xfs_rmapbt_calc_reserves(
570 struct xfs_mount
*mp
,
575 struct xfs_buf
*agbp
;
577 xfs_agblock_t agblocks
;
578 xfs_extlen_t tree_len
;
581 if (!xfs_sb_version_hasrmapbt(&mp
->m_sb
))
584 error
= xfs_alloc_read_agf(mp
, NULL
, agno
, 0, &agbp
);
588 agf
= XFS_BUF_TO_AGF(agbp
);
589 agblocks
= be32_to_cpu(agf
->agf_length
);
590 tree_len
= be32_to_cpu(agf
->agf_rmap_blocks
);
593 /* Reserve 1% of the AG or enough for 1 block per record. */
594 *ask
+= max(agblocks
/ 100, xfs_rmapbt_max_size(mp
, agblocks
));