1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_alloc.h"
17 #include "xfs_btree.h"
18 #include "xfs_bmap_btree.h"
20 #include "xfs_error.h"
21 #include "xfs_quota.h"
22 #include "xfs_trace.h"
26 * Convert on-disk form of btree root to in-memory form.
31 xfs_bmdr_block_t
*dblock
,
33 struct xfs_btree_block
*rblock
,
36 struct xfs_mount
*mp
= ip
->i_mount
;
43 xfs_btree_init_block_int(mp
, rblock
, XFS_BUF_DADDR_NULL
,
44 XFS_BTNUM_BMAP
, 0, 0, ip
->i_ino
,
46 rblock
->bb_level
= dblock
->bb_level
;
47 ASSERT(be16_to_cpu(rblock
->bb_level
) > 0);
48 rblock
->bb_numrecs
= dblock
->bb_numrecs
;
49 dmxr
= xfs_bmdr_maxrecs(dblocklen
, 0);
50 fkp
= XFS_BMDR_KEY_ADDR(dblock
, 1);
51 tkp
= XFS_BMBT_KEY_ADDR(mp
, rblock
, 1);
52 fpp
= XFS_BMDR_PTR_ADDR(dblock
, 1, dmxr
);
53 tpp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, rblocklen
);
54 dmxr
= be16_to_cpu(dblock
->bb_numrecs
);
55 memcpy(tkp
, fkp
, sizeof(*fkp
) * dmxr
);
56 memcpy(tpp
, fpp
, sizeof(*fpp
) * dmxr
);
60 xfs_bmbt_disk_get_all(
61 struct xfs_bmbt_rec
*rec
,
62 struct xfs_bmbt_irec
*irec
)
64 uint64_t l0
= get_unaligned_be64(&rec
->l0
);
65 uint64_t l1
= get_unaligned_be64(&rec
->l1
);
67 irec
->br_startoff
= (l0
& xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN
)) >> 9;
68 irec
->br_startblock
= ((l0
& xfs_mask64lo(9)) << 43) | (l1
>> 21);
69 irec
->br_blockcount
= l1
& xfs_mask64lo(21);
70 if (l0
>> (64 - BMBT_EXNTFLAG_BITLEN
))
71 irec
->br_state
= XFS_EXT_UNWRITTEN
;
73 irec
->br_state
= XFS_EXT_NORM
;
77 * Extract the blockcount field from an on disk bmap extent record.
80 xfs_bmbt_disk_get_blockcount(
83 return (xfs_filblks_t
)(be64_to_cpu(r
->l1
) & xfs_mask64lo(21));
87 * Extract the startoff field from a disk format bmap extent record.
90 xfs_bmbt_disk_get_startoff(
93 return ((xfs_fileoff_t
)be64_to_cpu(r
->l0
) &
94 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN
)) >> 9;
98 * Set all the fields in a bmap extent record from the uncompressed form.
101 xfs_bmbt_disk_set_all(
102 struct xfs_bmbt_rec
*r
,
103 struct xfs_bmbt_irec
*s
)
105 int extent_flag
= (s
->br_state
!= XFS_EXT_NORM
);
107 ASSERT(s
->br_state
== XFS_EXT_NORM
|| s
->br_state
== XFS_EXT_UNWRITTEN
);
108 ASSERT(!(s
->br_startoff
& xfs_mask64hi(64-BMBT_STARTOFF_BITLEN
)));
109 ASSERT(!(s
->br_blockcount
& xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN
)));
110 ASSERT(!(s
->br_startblock
& xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN
)));
113 ((xfs_bmbt_rec_base_t
)extent_flag
<< 63) |
114 ((xfs_bmbt_rec_base_t
)s
->br_startoff
<< 9) |
115 ((xfs_bmbt_rec_base_t
)s
->br_startblock
>> 43), &r
->l0
);
117 ((xfs_bmbt_rec_base_t
)s
->br_startblock
<< 21) |
118 ((xfs_bmbt_rec_base_t
)s
->br_blockcount
&
119 (xfs_bmbt_rec_base_t
)xfs_mask64lo(21)), &r
->l1
);
123 * Convert in-memory form of btree root to on-disk form.
127 struct xfs_mount
*mp
,
128 struct xfs_btree_block
*rblock
,
130 xfs_bmdr_block_t
*dblock
,
139 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
140 ASSERT(rblock
->bb_magic
== cpu_to_be32(XFS_BMAP_CRC_MAGIC
));
141 ASSERT(uuid_equal(&rblock
->bb_u
.l
.bb_uuid
,
142 &mp
->m_sb
.sb_meta_uuid
));
143 ASSERT(rblock
->bb_u
.l
.bb_blkno
==
144 cpu_to_be64(XFS_BUF_DADDR_NULL
));
146 ASSERT(rblock
->bb_magic
== cpu_to_be32(XFS_BMAP_MAGIC
));
147 ASSERT(rblock
->bb_u
.l
.bb_leftsib
== cpu_to_be64(NULLFSBLOCK
));
148 ASSERT(rblock
->bb_u
.l
.bb_rightsib
== cpu_to_be64(NULLFSBLOCK
));
149 ASSERT(rblock
->bb_level
!= 0);
150 dblock
->bb_level
= rblock
->bb_level
;
151 dblock
->bb_numrecs
= rblock
->bb_numrecs
;
152 dmxr
= xfs_bmdr_maxrecs(dblocklen
, 0);
153 fkp
= XFS_BMBT_KEY_ADDR(mp
, rblock
, 1);
154 tkp
= XFS_BMDR_KEY_ADDR(dblock
, 1);
155 fpp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, rblocklen
);
156 tpp
= XFS_BMDR_PTR_ADDR(dblock
, 1, dmxr
);
157 dmxr
= be16_to_cpu(dblock
->bb_numrecs
);
158 memcpy(tkp
, fkp
, sizeof(*fkp
) * dmxr
);
159 memcpy(tpp
, fpp
, sizeof(*fpp
) * dmxr
);
162 STATIC
struct xfs_btree_cur
*
164 struct xfs_btree_cur
*cur
)
166 struct xfs_btree_cur
*new;
168 new = xfs_bmbt_init_cursor(cur
->bc_mp
, cur
->bc_tp
,
169 cur
->bc_ino
.ip
, cur
->bc_ino
.whichfork
);
172 * Copy the firstblock, dfops, and flags values,
173 * since init cursor doesn't get them.
175 new->bc_ino
.flags
= cur
->bc_ino
.flags
;
181 xfs_bmbt_update_cursor(
182 struct xfs_btree_cur
*src
,
183 struct xfs_btree_cur
*dst
)
185 ASSERT((dst
->bc_tp
->t_firstblock
!= NULLFSBLOCK
) ||
186 (dst
->bc_ino
.ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
));
188 dst
->bc_ino
.allocated
+= src
->bc_ino
.allocated
;
189 dst
->bc_tp
->t_firstblock
= src
->bc_tp
->t_firstblock
;
191 src
->bc_ino
.allocated
= 0;
195 xfs_bmbt_alloc_block(
196 struct xfs_btree_cur
*cur
,
197 union xfs_btree_ptr
*start
,
198 union xfs_btree_ptr
*new,
201 xfs_alloc_arg_t args
; /* block allocation args */
202 int error
; /* error return value */
204 memset(&args
, 0, sizeof(args
));
205 args
.tp
= cur
->bc_tp
;
206 args
.mp
= cur
->bc_mp
;
207 args
.fsbno
= cur
->bc_tp
->t_firstblock
;
208 xfs_rmap_ino_bmbt_owner(&args
.oinfo
, cur
->bc_ino
.ip
->i_ino
,
209 cur
->bc_ino
.whichfork
);
211 if (args
.fsbno
== NULLFSBLOCK
) {
212 args
.fsbno
= be64_to_cpu(start
->l
);
213 args
.type
= XFS_ALLOCTYPE_START_BNO
;
215 * Make sure there is sufficient room left in the AG to
216 * complete a full tree split for an extent insert. If
217 * we are converting the middle part of an extent then
218 * we may need space for two tree splits.
220 * We are relying on the caller to make the correct block
221 * reservation for this operation to succeed. If the
222 * reservation amount is insufficient then we may fail a
223 * block allocation here and corrupt the filesystem.
225 args
.minleft
= args
.tp
->t_blk_res
;
226 } else if (cur
->bc_tp
->t_flags
& XFS_TRANS_LOWMODE
) {
227 args
.type
= XFS_ALLOCTYPE_START_BNO
;
229 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
232 args
.minlen
= args
.maxlen
= args
.prod
= 1;
233 args
.wasdel
= cur
->bc_ino
.flags
& XFS_BTCUR_BMBT_WASDEL
;
234 if (!args
.wasdel
&& args
.tp
->t_blk_res
== 0) {
238 error
= xfs_alloc_vextent(&args
);
242 if (args
.fsbno
== NULLFSBLOCK
&& args
.minleft
) {
244 * Could not find an AG with enough free space to satisfy
245 * a full btree split. Try again and if
246 * successful activate the lowspace algorithm.
249 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
250 error
= xfs_alloc_vextent(&args
);
253 cur
->bc_tp
->t_flags
|= XFS_TRANS_LOWMODE
;
255 if (WARN_ON_ONCE(args
.fsbno
== NULLFSBLOCK
)) {
260 ASSERT(args
.len
== 1);
261 cur
->bc_tp
->t_firstblock
= args
.fsbno
;
262 cur
->bc_ino
.allocated
++;
263 cur
->bc_ino
.ip
->i_d
.di_nblocks
++;
264 xfs_trans_log_inode(args
.tp
, cur
->bc_ino
.ip
, XFS_ILOG_CORE
);
265 xfs_trans_mod_dquot_byino(args
.tp
, cur
->bc_ino
.ip
,
266 XFS_TRANS_DQ_BCOUNT
, 1L);
268 new->l
= cpu_to_be64(args
.fsbno
);
279 struct xfs_btree_cur
*cur
,
282 struct xfs_mount
*mp
= cur
->bc_mp
;
283 struct xfs_inode
*ip
= cur
->bc_ino
.ip
;
284 struct xfs_trans
*tp
= cur
->bc_tp
;
285 xfs_fsblock_t fsbno
= XFS_DADDR_TO_FSB(mp
, XFS_BUF_ADDR(bp
));
286 struct xfs_owner_info oinfo
;
288 xfs_rmap_ino_bmbt_owner(&oinfo
, ip
->i_ino
, cur
->bc_ino
.whichfork
);
289 xfs_bmap_add_free(cur
->bc_tp
, fsbno
, 1, &oinfo
);
290 ip
->i_d
.di_nblocks
--;
292 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
293 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
298 xfs_bmbt_get_minrecs(
299 struct xfs_btree_cur
*cur
,
302 if (level
== cur
->bc_nlevels
- 1) {
303 struct xfs_ifork
*ifp
;
305 ifp
= XFS_IFORK_PTR(cur
->bc_ino
.ip
,
306 cur
->bc_ino
.whichfork
);
308 return xfs_bmbt_maxrecs(cur
->bc_mp
,
309 ifp
->if_broot_bytes
, level
== 0) / 2;
312 return cur
->bc_mp
->m_bmap_dmnr
[level
!= 0];
316 xfs_bmbt_get_maxrecs(
317 struct xfs_btree_cur
*cur
,
320 if (level
== cur
->bc_nlevels
- 1) {
321 struct xfs_ifork
*ifp
;
323 ifp
= XFS_IFORK_PTR(cur
->bc_ino
.ip
,
324 cur
->bc_ino
.whichfork
);
326 return xfs_bmbt_maxrecs(cur
->bc_mp
,
327 ifp
->if_broot_bytes
, level
== 0);
330 return cur
->bc_mp
->m_bmap_dmxr
[level
!= 0];
335 * Get the maximum records we could store in the on-disk format.
337 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
338 * for the root node this checks the available space in the dinode fork
339 * so that we can resize the in-memory buffer to match it. After a
340 * resize to the maximum size this function returns the same value
341 * as xfs_bmbt_get_maxrecs for the root node, too.
344 xfs_bmbt_get_dmaxrecs(
345 struct xfs_btree_cur
*cur
,
348 if (level
!= cur
->bc_nlevels
- 1)
349 return cur
->bc_mp
->m_bmap_dmxr
[level
!= 0];
350 return xfs_bmdr_maxrecs(cur
->bc_ino
.forksize
, level
== 0);
354 xfs_bmbt_init_key_from_rec(
355 union xfs_btree_key
*key
,
356 union xfs_btree_rec
*rec
)
358 key
->bmbt
.br_startoff
=
359 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec
->bmbt
));
363 xfs_bmbt_init_high_key_from_rec(
364 union xfs_btree_key
*key
,
365 union xfs_btree_rec
*rec
)
367 key
->bmbt
.br_startoff
= cpu_to_be64(
368 xfs_bmbt_disk_get_startoff(&rec
->bmbt
) +
369 xfs_bmbt_disk_get_blockcount(&rec
->bmbt
) - 1);
373 xfs_bmbt_init_rec_from_cur(
374 struct xfs_btree_cur
*cur
,
375 union xfs_btree_rec
*rec
)
377 xfs_bmbt_disk_set_all(&rec
->bmbt
, &cur
->bc_rec
.b
);
381 xfs_bmbt_init_ptr_from_cur(
382 struct xfs_btree_cur
*cur
,
383 union xfs_btree_ptr
*ptr
)
390 struct xfs_btree_cur
*cur
,
391 union xfs_btree_key
*key
)
393 return (int64_t)be64_to_cpu(key
->bmbt
.br_startoff
) -
394 cur
->bc_rec
.b
.br_startoff
;
398 xfs_bmbt_diff_two_keys(
399 struct xfs_btree_cur
*cur
,
400 union xfs_btree_key
*k1
,
401 union xfs_btree_key
*k2
)
403 uint64_t a
= be64_to_cpu(k1
->bmbt
.br_startoff
);
404 uint64_t b
= be64_to_cpu(k2
->bmbt
.br_startoff
);
407 * Note: This routine previously casted a and b to int64 and subtracted
408 * them to generate a result. This lead to problems if b was the
409 * "maximum" key value (all ones) being signed incorrectly, hence this
410 * somewhat less efficient version.
419 static xfs_failaddr_t
423 struct xfs_mount
*mp
= bp
->b_mount
;
424 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
428 if (!xfs_verify_magic(bp
, block
->bb_magic
))
429 return __this_address
;
431 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
433 * XXX: need a better way of verifying the owner here. Right now
434 * just make sure there has been one set.
436 fa
= xfs_btree_lblock_v5hdr_verify(bp
, XFS_RMAP_OWN_UNKNOWN
);
442 * numrecs and level verification.
444 * We don't know what fork we belong to, so just verify that the level
445 * is less than the maximum of the two. Later checks will be more
448 level
= be16_to_cpu(block
->bb_level
);
449 if (level
> max(mp
->m_bm_maxlevels
[0], mp
->m_bm_maxlevels
[1]))
450 return __this_address
;
452 return xfs_btree_lblock_verify(bp
, mp
->m_bmap_dmxr
[level
!= 0]);
456 xfs_bmbt_read_verify(
461 if (!xfs_btree_lblock_verify_crc(bp
))
462 xfs_verifier_error(bp
, -EFSBADCRC
, __this_address
);
464 fa
= xfs_bmbt_verify(bp
);
466 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
470 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
474 xfs_bmbt_write_verify(
479 fa
= xfs_bmbt_verify(bp
);
481 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
482 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
485 xfs_btree_lblock_calc_crc(bp
);
488 const struct xfs_buf_ops xfs_bmbt_buf_ops
= {
490 .magic
= { cpu_to_be32(XFS_BMAP_MAGIC
),
491 cpu_to_be32(XFS_BMAP_CRC_MAGIC
) },
492 .verify_read
= xfs_bmbt_read_verify
,
493 .verify_write
= xfs_bmbt_write_verify
,
494 .verify_struct
= xfs_bmbt_verify
,
499 xfs_bmbt_keys_inorder(
500 struct xfs_btree_cur
*cur
,
501 union xfs_btree_key
*k1
,
502 union xfs_btree_key
*k2
)
504 return be64_to_cpu(k1
->bmbt
.br_startoff
) <
505 be64_to_cpu(k2
->bmbt
.br_startoff
);
509 xfs_bmbt_recs_inorder(
510 struct xfs_btree_cur
*cur
,
511 union xfs_btree_rec
*r1
,
512 union xfs_btree_rec
*r2
)
514 return xfs_bmbt_disk_get_startoff(&r1
->bmbt
) +
515 xfs_bmbt_disk_get_blockcount(&r1
->bmbt
) <=
516 xfs_bmbt_disk_get_startoff(&r2
->bmbt
);
519 static const struct xfs_btree_ops xfs_bmbt_ops
= {
520 .rec_len
= sizeof(xfs_bmbt_rec_t
),
521 .key_len
= sizeof(xfs_bmbt_key_t
),
523 .dup_cursor
= xfs_bmbt_dup_cursor
,
524 .update_cursor
= xfs_bmbt_update_cursor
,
525 .alloc_block
= xfs_bmbt_alloc_block
,
526 .free_block
= xfs_bmbt_free_block
,
527 .get_maxrecs
= xfs_bmbt_get_maxrecs
,
528 .get_minrecs
= xfs_bmbt_get_minrecs
,
529 .get_dmaxrecs
= xfs_bmbt_get_dmaxrecs
,
530 .init_key_from_rec
= xfs_bmbt_init_key_from_rec
,
531 .init_high_key_from_rec
= xfs_bmbt_init_high_key_from_rec
,
532 .init_rec_from_cur
= xfs_bmbt_init_rec_from_cur
,
533 .init_ptr_from_cur
= xfs_bmbt_init_ptr_from_cur
,
534 .key_diff
= xfs_bmbt_key_diff
,
535 .diff_two_keys
= xfs_bmbt_diff_two_keys
,
536 .buf_ops
= &xfs_bmbt_buf_ops
,
537 .keys_inorder
= xfs_bmbt_keys_inorder
,
538 .recs_inorder
= xfs_bmbt_recs_inorder
,
542 * Allocate a new bmap btree cursor.
544 struct xfs_btree_cur
* /* new bmap btree cursor */
545 xfs_bmbt_init_cursor(
546 struct xfs_mount
*mp
, /* file system mount point */
547 struct xfs_trans
*tp
, /* transaction pointer */
548 struct xfs_inode
*ip
, /* inode owning the btree */
549 int whichfork
) /* data or attr fork */
551 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
552 struct xfs_btree_cur
*cur
;
553 ASSERT(whichfork
!= XFS_COW_FORK
);
555 cur
= kmem_cache_zalloc(xfs_btree_cur_zone
, GFP_NOFS
| __GFP_NOFAIL
);
559 cur
->bc_nlevels
= be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
560 cur
->bc_btnum
= XFS_BTNUM_BMAP
;
561 cur
->bc_blocklog
= mp
->m_sb
.sb_blocklog
;
562 cur
->bc_statoff
= XFS_STATS_CALC_INDEX(xs_bmbt_2
);
564 cur
->bc_ops
= &xfs_bmbt_ops
;
565 cur
->bc_flags
= XFS_BTREE_LONG_PTRS
| XFS_BTREE_ROOT_IN_INODE
;
566 if (xfs_sb_version_hascrc(&mp
->m_sb
))
567 cur
->bc_flags
|= XFS_BTREE_CRC_BLOCKS
;
569 cur
->bc_ino
.forksize
= XFS_IFORK_SIZE(ip
, whichfork
);
571 cur
->bc_ino
.allocated
= 0;
572 cur
->bc_ino
.flags
= 0;
573 cur
->bc_ino
.whichfork
= whichfork
;
579 * Calculate number of records in a bmap btree block.
583 struct xfs_mount
*mp
,
587 blocklen
-= XFS_BMBT_BLOCK_LEN(mp
);
590 return blocklen
/ sizeof(xfs_bmbt_rec_t
);
591 return blocklen
/ (sizeof(xfs_bmbt_key_t
) + sizeof(xfs_bmbt_ptr_t
));
595 * Calculate number of records in a bmap btree inode root.
602 blocklen
-= sizeof(xfs_bmdr_block_t
);
605 return blocklen
/ sizeof(xfs_bmdr_rec_t
);
606 return blocklen
/ (sizeof(xfs_bmdr_key_t
) + sizeof(xfs_bmdr_ptr_t
));
610 * Change the owner of a btree format fork fo the inode passed in. Change it to
611 * the owner of that is passed in so that we can change owners before or after
612 * we switch forks between inodes. The operation that the caller is doing will
613 * determine whether is needs to change owner before or after the switch.
615 * For demand paged transactional modification, the fork switch should be done
616 * after reading in all the blocks, modifying them and pinning them in the
617 * transaction. For modification when the buffers are already pinned in memory,
618 * the fork switch can be done before changing the owner as we won't need to
619 * validate the owner until the btree buffers are unpinned and writes can occur
622 * For recovery based ownership change, there is no transactional context and
623 * so a buffer list must be supplied so that we can record the buffers that we
624 * modified for the caller to issue IO on.
627 xfs_bmbt_change_owner(
628 struct xfs_trans
*tp
,
629 struct xfs_inode
*ip
,
632 struct list_head
*buffer_list
)
634 struct xfs_btree_cur
*cur
;
637 ASSERT(tp
|| buffer_list
);
638 ASSERT(!(tp
&& buffer_list
));
639 ASSERT(XFS_IFORK_PTR(ip
, whichfork
)->if_format
== XFS_DINODE_FMT_BTREE
);
641 cur
= xfs_bmbt_init_cursor(ip
->i_mount
, tp
, ip
, whichfork
);
642 cur
->bc_ino
.flags
|= XFS_BTCUR_BMBT_INVALID_OWNER
;
644 error
= xfs_btree_change_owner(cur
, new_owner
, buffer_list
);
645 xfs_btree_del_cursor(cur
, error
);
649 /* Calculate the bmap btree size for some records. */
652 struct xfs_mount
*mp
,
653 unsigned long long len
)
655 return xfs_btree_calc_size(mp
->m_bmap_dmnr
, len
);