2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_format.h"
22 #include "xfs_shared.h"
25 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_alloc.h"
40 #include "xfs_attr_leaf.h"
41 #include "xfs_error.h"
42 #include "xfs_trace.h"
43 #include "xfs_cksum.h"
44 #include "xfs_buf_item.h"
49 * Routines to implement directories as Btrees of hashed names.
52 /*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
57 * Routines used for growing the Btree.
59 STATIC
int xfs_da3_root_split(xfs_da_state_t
*state
,
60 xfs_da_state_blk_t
*existing_root
,
61 xfs_da_state_blk_t
*new_child
);
62 STATIC
int xfs_da3_node_split(xfs_da_state_t
*state
,
63 xfs_da_state_blk_t
*existing_blk
,
64 xfs_da_state_blk_t
*split_blk
,
65 xfs_da_state_blk_t
*blk_to_add
,
68 STATIC
void xfs_da3_node_rebalance(xfs_da_state_t
*state
,
69 xfs_da_state_blk_t
*node_blk_1
,
70 xfs_da_state_blk_t
*node_blk_2
);
71 STATIC
void xfs_da3_node_add(xfs_da_state_t
*state
,
72 xfs_da_state_blk_t
*old_node_blk
,
73 xfs_da_state_blk_t
*new_node_blk
);
76 * Routines used for shrinking the Btree.
78 STATIC
int xfs_da3_root_join(xfs_da_state_t
*state
,
79 xfs_da_state_blk_t
*root_blk
);
80 STATIC
int xfs_da3_node_toosmall(xfs_da_state_t
*state
, int *retval
);
81 STATIC
void xfs_da3_node_remove(xfs_da_state_t
*state
,
82 xfs_da_state_blk_t
*drop_blk
);
83 STATIC
void xfs_da3_node_unbalance(xfs_da_state_t
*state
,
84 xfs_da_state_blk_t
*src_node_blk
,
85 xfs_da_state_blk_t
*dst_node_blk
);
90 STATIC
int xfs_da3_blk_unlink(xfs_da_state_t
*state
,
91 xfs_da_state_blk_t
*drop_blk
,
92 xfs_da_state_blk_t
*save_blk
);
95 kmem_zone_t
*xfs_da_state_zone
; /* anchor for state struct zone */
98 * Allocate a dir-state structure.
99 * We don't put them on the stack since they're large.
102 xfs_da_state_alloc(void)
104 return kmem_zone_zalloc(xfs_da_state_zone
, KM_NOFS
);
108 * Kill the altpath contents of a da-state structure.
111 xfs_da_state_kill_altpath(xfs_da_state_t
*state
)
115 for (i
= 0; i
< state
->altpath
.active
; i
++)
116 state
->altpath
.blk
[i
].bp
= NULL
;
117 state
->altpath
.active
= 0;
121 * Free a da-state structure.
124 xfs_da_state_free(xfs_da_state_t
*state
)
126 xfs_da_state_kill_altpath(state
);
128 memset((char *)state
, 0, sizeof(*state
));
130 kmem_zone_free(xfs_da_state_zone
, state
);
134 xfs_da3_node_hdr_from_disk(
135 struct xfs_da3_icnode_hdr
*to
,
136 struct xfs_da_intnode
*from
)
138 ASSERT(from
->hdr
.info
.magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
139 from
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
));
141 if (from
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
)) {
142 struct xfs_da3_node_hdr
*hdr3
= (struct xfs_da3_node_hdr
*)from
;
144 to
->forw
= be32_to_cpu(hdr3
->info
.hdr
.forw
);
145 to
->back
= be32_to_cpu(hdr3
->info
.hdr
.back
);
146 to
->magic
= be16_to_cpu(hdr3
->info
.hdr
.magic
);
147 to
->count
= be16_to_cpu(hdr3
->__count
);
148 to
->level
= be16_to_cpu(hdr3
->__level
);
151 to
->forw
= be32_to_cpu(from
->hdr
.info
.forw
);
152 to
->back
= be32_to_cpu(from
->hdr
.info
.back
);
153 to
->magic
= be16_to_cpu(from
->hdr
.info
.magic
);
154 to
->count
= be16_to_cpu(from
->hdr
.__count
);
155 to
->level
= be16_to_cpu(from
->hdr
.__level
);
159 xfs_da3_node_hdr_to_disk(
160 struct xfs_da_intnode
*to
,
161 struct xfs_da3_icnode_hdr
*from
)
163 ASSERT(from
->magic
== XFS_DA_NODE_MAGIC
||
164 from
->magic
== XFS_DA3_NODE_MAGIC
);
166 if (from
->magic
== XFS_DA3_NODE_MAGIC
) {
167 struct xfs_da3_node_hdr
*hdr3
= (struct xfs_da3_node_hdr
*)to
;
169 hdr3
->info
.hdr
.forw
= cpu_to_be32(from
->forw
);
170 hdr3
->info
.hdr
.back
= cpu_to_be32(from
->back
);
171 hdr3
->info
.hdr
.magic
= cpu_to_be16(from
->magic
);
172 hdr3
->__count
= cpu_to_be16(from
->count
);
173 hdr3
->__level
= cpu_to_be16(from
->level
);
176 to
->hdr
.info
.forw
= cpu_to_be32(from
->forw
);
177 to
->hdr
.info
.back
= cpu_to_be32(from
->back
);
178 to
->hdr
.info
.magic
= cpu_to_be16(from
->magic
);
179 to
->hdr
.__count
= cpu_to_be16(from
->count
);
180 to
->hdr
.__level
= cpu_to_be16(from
->level
);
187 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
188 struct xfs_da_intnode
*hdr
= bp
->b_addr
;
189 struct xfs_da3_icnode_hdr ichdr
;
191 xfs_da3_node_hdr_from_disk(&ichdr
, hdr
);
193 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
194 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
196 if (ichdr
.magic
!= XFS_DA3_NODE_MAGIC
)
199 if (!uuid_equal(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_uuid
))
201 if (be64_to_cpu(hdr3
->info
.blkno
) != bp
->b_bn
)
204 if (ichdr
.magic
!= XFS_DA_NODE_MAGIC
)
207 if (ichdr
.level
== 0)
209 if (ichdr
.level
> XFS_DA_NODE_MAXDEPTH
)
211 if (ichdr
.count
== 0)
215 * we don't know if the node is for and attribute or directory tree,
216 * so only fail if the count is outside both bounds
218 if (ichdr
.count
> mp
->m_dir_node_ents
&&
219 ichdr
.count
> mp
->m_attr_node_ents
)
222 /* XXX: hash order check? */
228 xfs_da3_node_write_verify(
231 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
232 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
233 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
235 if (!xfs_da3_node_verify(bp
)) {
236 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
237 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
241 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
245 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
247 xfs_update_cksum(bp
->b_addr
, BBTOB(bp
->b_length
), XFS_DA3_NODE_CRC_OFF
);
251 * leaf/node format detection on trees is sketchy, so a node read can be done on
252 * leaf level blocks when detection identifies the tree as a node format tree
253 * incorrectly. In this case, we need to swap the verifier to match the correct
254 * format of the block being read.
257 xfs_da3_node_read_verify(
260 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
261 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
263 switch (be16_to_cpu(info
->magic
)) {
264 case XFS_DA3_NODE_MAGIC
:
265 if (!xfs_verify_cksum(bp
->b_addr
, BBTOB(bp
->b_length
),
266 XFS_DA3_NODE_CRC_OFF
))
269 case XFS_DA_NODE_MAGIC
:
270 if (!xfs_da3_node_verify(bp
))
273 case XFS_ATTR_LEAF_MAGIC
:
274 case XFS_ATTR3_LEAF_MAGIC
:
275 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
276 bp
->b_ops
->verify_read(bp
);
278 case XFS_DIR2_LEAFN_MAGIC
:
279 case XFS_DIR3_LEAFN_MAGIC
:
280 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
281 bp
->b_ops
->verify_read(bp
);
288 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
289 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
292 const struct xfs_buf_ops xfs_da3_node_buf_ops
= {
293 .verify_read
= xfs_da3_node_read_verify
,
294 .verify_write
= xfs_da3_node_write_verify
,
299 struct xfs_trans
*tp
,
300 struct xfs_inode
*dp
,
302 xfs_daddr_t mappedbno
,
303 struct xfs_buf
**bpp
,
308 err
= xfs_da_read_buf(tp
, dp
, bno
, mappedbno
, bpp
,
309 which_fork
, &xfs_da3_node_buf_ops
);
311 struct xfs_da_blkinfo
*info
= (*bpp
)->b_addr
;
314 switch (be16_to_cpu(info
->magic
)) {
315 case XFS_DA_NODE_MAGIC
:
316 case XFS_DA3_NODE_MAGIC
:
317 type
= XFS_BLFT_DA_NODE_BUF
;
319 case XFS_ATTR_LEAF_MAGIC
:
320 case XFS_ATTR3_LEAF_MAGIC
:
321 type
= XFS_BLFT_ATTR_LEAF_BUF
;
323 case XFS_DIR2_LEAFN_MAGIC
:
324 case XFS_DIR3_LEAFN_MAGIC
:
325 type
= XFS_BLFT_DIR_LEAFN_BUF
;
332 xfs_trans_buf_set_type(tp
, *bpp
, type
);
337 /*========================================================================
338 * Routines used for growing the Btree.
339 *========================================================================*/
342 * Create the initial contents of an intermediate node.
346 struct xfs_da_args
*args
,
349 struct xfs_buf
**bpp
,
352 struct xfs_da_intnode
*node
;
353 struct xfs_trans
*tp
= args
->trans
;
354 struct xfs_mount
*mp
= tp
->t_mountp
;
355 struct xfs_da3_icnode_hdr ichdr
= {0};
359 trace_xfs_da_node_create(args
);
360 ASSERT(level
<= XFS_DA_NODE_MAXDEPTH
);
362 error
= xfs_da_get_buf(tp
, args
->dp
, blkno
, -1, &bp
, whichfork
);
365 bp
->b_ops
= &xfs_da3_node_buf_ops
;
366 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
369 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
370 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
372 ichdr
.magic
= XFS_DA3_NODE_MAGIC
;
373 hdr3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
374 hdr3
->info
.owner
= cpu_to_be64(args
->dp
->i_ino
);
375 uuid_copy(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_uuid
);
377 ichdr
.magic
= XFS_DA_NODE_MAGIC
;
381 xfs_da3_node_hdr_to_disk(node
, &ichdr
);
382 xfs_trans_log_buf(tp
, bp
,
383 XFS_DA_LOGRANGE(node
, &node
->hdr
, xfs_da3_node_hdr_size(node
)));
390 * Split a leaf node, rebalance, then possibly split
391 * intermediate nodes, rebalance, etc.
395 struct xfs_da_state
*state
)
397 struct xfs_da_state_blk
*oldblk
;
398 struct xfs_da_state_blk
*newblk
;
399 struct xfs_da_state_blk
*addblk
;
400 struct xfs_da_intnode
*node
;
407 trace_xfs_da_split(state
->args
);
410 * Walk back up the tree splitting/inserting/adjusting as necessary.
411 * If we need to insert and there isn't room, split the node, then
412 * decide which fragment to insert the new block from below into.
413 * Note that we may split the root this way, but we need more fixup.
415 max
= state
->path
.active
- 1;
416 ASSERT((max
>= 0) && (max
< XFS_DA_NODE_MAXDEPTH
));
417 ASSERT(state
->path
.blk
[max
].magic
== XFS_ATTR_LEAF_MAGIC
||
418 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
420 addblk
= &state
->path
.blk
[max
]; /* initial dummy value */
421 for (i
= max
; (i
>= 0) && addblk
; state
->path
.active
--, i
--) {
422 oldblk
= &state
->path
.blk
[i
];
423 newblk
= &state
->altpath
.blk
[i
];
426 * If a leaf node then
427 * Allocate a new leaf node, then rebalance across them.
428 * else if an intermediate node then
429 * We split on the last layer, must we split the node?
431 switch (oldblk
->magic
) {
432 case XFS_ATTR_LEAF_MAGIC
:
433 error
= xfs_attr3_leaf_split(state
, oldblk
, newblk
);
434 if ((error
!= 0) && (error
!= ENOSPC
)) {
435 return(error
); /* GROT: attr is inconsistent */
442 * Entry wouldn't fit, split the leaf again.
444 state
->extravalid
= 1;
446 state
->extraafter
= 0; /* before newblk */
447 trace_xfs_attr_leaf_split_before(state
->args
);
448 error
= xfs_attr3_leaf_split(state
, oldblk
,
451 state
->extraafter
= 1; /* after newblk */
452 trace_xfs_attr_leaf_split_after(state
->args
);
453 error
= xfs_attr3_leaf_split(state
, newblk
,
457 return(error
); /* GROT: attr inconsistent */
460 case XFS_DIR2_LEAFN_MAGIC
:
461 error
= xfs_dir2_leafn_split(state
, oldblk
, newblk
);
466 case XFS_DA_NODE_MAGIC
:
467 error
= xfs_da3_node_split(state
, oldblk
, newblk
, addblk
,
471 return(error
); /* GROT: dir is inconsistent */
473 * Record the newly split block for the next time thru?
483 * Update the btree to show the new hashval for this child.
485 xfs_da3_fixhashpath(state
, &state
->path
);
491 * Split the root node.
493 ASSERT(state
->path
.active
== 0);
494 oldblk
= &state
->path
.blk
[0];
495 error
= xfs_da3_root_split(state
, oldblk
, addblk
);
498 return(error
); /* GROT: dir is inconsistent */
502 * Update pointers to the node which used to be block 0 and
503 * just got bumped because of the addition of a new root node.
504 * There might be three blocks involved if a double split occurred,
505 * and the original block 0 could be at any position in the list.
507 * Note: the magic numbers and sibling pointers are in the same
508 * physical place for both v2 and v3 headers (by design). Hence it
509 * doesn't matter which version of the xfs_da_intnode structure we use
510 * here as the result will be the same using either structure.
512 node
= oldblk
->bp
->b_addr
;
513 if (node
->hdr
.info
.forw
) {
514 if (be32_to_cpu(node
->hdr
.info
.forw
) == addblk
->blkno
) {
517 ASSERT(state
->extravalid
);
518 bp
= state
->extrablk
.bp
;
521 node
->hdr
.info
.back
= cpu_to_be32(oldblk
->blkno
);
522 xfs_trans_log_buf(state
->args
->trans
, bp
,
523 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
524 sizeof(node
->hdr
.info
)));
526 node
= oldblk
->bp
->b_addr
;
527 if (node
->hdr
.info
.back
) {
528 if (be32_to_cpu(node
->hdr
.info
.back
) == addblk
->blkno
) {
531 ASSERT(state
->extravalid
);
532 bp
= state
->extrablk
.bp
;
535 node
->hdr
.info
.forw
= cpu_to_be32(oldblk
->blkno
);
536 xfs_trans_log_buf(state
->args
->trans
, bp
,
537 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
538 sizeof(node
->hdr
.info
)));
545 * Split the root. We have to create a new root and point to the two
546 * parts (the split old root) that we just created. Copy block zero to
547 * the EOF, extending the inode in process.
549 STATIC
int /* error */
551 struct xfs_da_state
*state
,
552 struct xfs_da_state_blk
*blk1
,
553 struct xfs_da_state_blk
*blk2
)
555 struct xfs_da_intnode
*node
;
556 struct xfs_da_intnode
*oldroot
;
557 struct xfs_da_node_entry
*btree
;
558 struct xfs_da3_icnode_hdr nodehdr
;
559 struct xfs_da_args
*args
;
561 struct xfs_inode
*dp
;
562 struct xfs_trans
*tp
;
563 struct xfs_mount
*mp
;
564 struct xfs_dir2_leaf
*leaf
;
570 trace_xfs_da_root_split(state
->args
);
573 * Copy the existing (incorrect) block from the root node position
574 * to a free space somewhere.
577 error
= xfs_da_grow_inode(args
, &blkno
);
584 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, args
->whichfork
);
588 oldroot
= blk1
->bp
->b_addr
;
589 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
590 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
)) {
591 struct xfs_da3_icnode_hdr nodehdr
;
593 xfs_da3_node_hdr_from_disk(&nodehdr
, oldroot
);
594 btree
= xfs_da3_node_tree_p(oldroot
);
595 size
= (int)((char *)&btree
[nodehdr
.count
] - (char *)oldroot
);
596 level
= nodehdr
.level
;
599 * we are about to copy oldroot to bp, so set up the type
600 * of bp while we know exactly what it will be.
602 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
604 struct xfs_dir3_icleaf_hdr leafhdr
;
605 struct xfs_dir2_leaf_entry
*ents
;
607 leaf
= (xfs_dir2_leaf_t
*)oldroot
;
608 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
609 ents
= xfs_dir3_leaf_ents_p(leaf
);
611 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
612 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
613 size
= (int)((char *)&ents
[leafhdr
.count
] - (char *)leaf
);
617 * we are about to copy oldroot to bp, so set up the type
618 * of bp while we know exactly what it will be.
620 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAFN_BUF
);
624 * we can copy most of the information in the node from one block to
625 * another, but for CRC enabled headers we have to make sure that the
626 * block specific identifiers are kept intact. We update the buffer
629 memcpy(node
, oldroot
, size
);
630 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
631 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
632 struct xfs_da3_intnode
*node3
= (struct xfs_da3_intnode
*)node
;
634 node3
->hdr
.info
.blkno
= cpu_to_be64(bp
->b_bn
);
636 xfs_trans_log_buf(tp
, bp
, 0, size
- 1);
638 bp
->b_ops
= blk1
->bp
->b_ops
;
639 xfs_trans_buf_copy_type(bp
, blk1
->bp
);
644 * Set up the new root node.
646 error
= xfs_da3_node_create(args
,
647 (args
->whichfork
== XFS_DATA_FORK
) ? mp
->m_dirleafblk
: 0,
648 level
+ 1, &bp
, args
->whichfork
);
653 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
654 btree
= xfs_da3_node_tree_p(node
);
655 btree
[0].hashval
= cpu_to_be32(blk1
->hashval
);
656 btree
[0].before
= cpu_to_be32(blk1
->blkno
);
657 btree
[1].hashval
= cpu_to_be32(blk2
->hashval
);
658 btree
[1].before
= cpu_to_be32(blk2
->blkno
);
660 xfs_da3_node_hdr_to_disk(node
, &nodehdr
);
663 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
664 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
665 ASSERT(blk1
->blkno
>= mp
->m_dirleafblk
&&
666 blk1
->blkno
< mp
->m_dirfreeblk
);
667 ASSERT(blk2
->blkno
>= mp
->m_dirleafblk
&&
668 blk2
->blkno
< mp
->m_dirfreeblk
);
672 /* Header is already logged by xfs_da_node_create */
673 xfs_trans_log_buf(tp
, bp
,
674 XFS_DA_LOGRANGE(node
, btree
, sizeof(xfs_da_node_entry_t
) * 2));
680 * Split the node, rebalance, then add the new entry.
682 STATIC
int /* error */
684 struct xfs_da_state
*state
,
685 struct xfs_da_state_blk
*oldblk
,
686 struct xfs_da_state_blk
*newblk
,
687 struct xfs_da_state_blk
*addblk
,
691 struct xfs_da_intnode
*node
;
692 struct xfs_da3_icnode_hdr nodehdr
;
698 trace_xfs_da_node_split(state
->args
);
700 node
= oldblk
->bp
->b_addr
;
701 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
704 * With V2 dirs the extra block is data or freespace.
706 useextra
= state
->extravalid
&& state
->args
->whichfork
== XFS_ATTR_FORK
;
707 newcount
= 1 + useextra
;
709 * Do we have to split the node?
711 if (nodehdr
.count
+ newcount
> state
->node_ents
) {
713 * Allocate a new node, add to the doubly linked chain of
714 * nodes, then move some of our excess entries into it.
716 error
= xfs_da_grow_inode(state
->args
, &blkno
);
718 return(error
); /* GROT: dir is inconsistent */
720 error
= xfs_da3_node_create(state
->args
, blkno
, treelevel
,
721 &newblk
->bp
, state
->args
->whichfork
);
723 return(error
); /* GROT: dir is inconsistent */
724 newblk
->blkno
= blkno
;
725 newblk
->magic
= XFS_DA_NODE_MAGIC
;
726 xfs_da3_node_rebalance(state
, oldblk
, newblk
);
727 error
= xfs_da3_blk_link(state
, oldblk
, newblk
);
736 * Insert the new entry(s) into the correct block
737 * (updating last hashval in the process).
739 * xfs_da3_node_add() inserts BEFORE the given index,
740 * and as a result of using node_lookup_int() we always
741 * point to a valid entry (not after one), but a split
742 * operation always results in a new block whose hashvals
743 * FOLLOW the current block.
745 * If we had double-split op below us, then add the extra block too.
747 node
= oldblk
->bp
->b_addr
;
748 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
749 if (oldblk
->index
<= nodehdr
.count
) {
751 xfs_da3_node_add(state
, oldblk
, addblk
);
753 if (state
->extraafter
)
755 xfs_da3_node_add(state
, oldblk
, &state
->extrablk
);
756 state
->extravalid
= 0;
760 xfs_da3_node_add(state
, newblk
, addblk
);
762 if (state
->extraafter
)
764 xfs_da3_node_add(state
, newblk
, &state
->extrablk
);
765 state
->extravalid
= 0;
773 * Balance the btree elements between two intermediate nodes,
774 * usually one full and one empty.
776 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
779 xfs_da3_node_rebalance(
780 struct xfs_da_state
*state
,
781 struct xfs_da_state_blk
*blk1
,
782 struct xfs_da_state_blk
*blk2
)
784 struct xfs_da_intnode
*node1
;
785 struct xfs_da_intnode
*node2
;
786 struct xfs_da_intnode
*tmpnode
;
787 struct xfs_da_node_entry
*btree1
;
788 struct xfs_da_node_entry
*btree2
;
789 struct xfs_da_node_entry
*btree_s
;
790 struct xfs_da_node_entry
*btree_d
;
791 struct xfs_da3_icnode_hdr nodehdr1
;
792 struct xfs_da3_icnode_hdr nodehdr2
;
793 struct xfs_trans
*tp
;
798 trace_xfs_da_node_rebalance(state
->args
);
800 node1
= blk1
->bp
->b_addr
;
801 node2
= blk2
->bp
->b_addr
;
802 xfs_da3_node_hdr_from_disk(&nodehdr1
, node1
);
803 xfs_da3_node_hdr_from_disk(&nodehdr2
, node2
);
804 btree1
= xfs_da3_node_tree_p(node1
);
805 btree2
= xfs_da3_node_tree_p(node2
);
808 * Figure out how many entries need to move, and in which direction.
809 * Swap the nodes around if that makes it simpler.
811 if (nodehdr1
.count
> 0 && nodehdr2
.count
> 0 &&
812 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
813 (be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
) <
814 be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
)))) {
818 xfs_da3_node_hdr_from_disk(&nodehdr1
, node1
);
819 xfs_da3_node_hdr_from_disk(&nodehdr2
, node2
);
820 btree1
= xfs_da3_node_tree_p(node1
);
821 btree2
= xfs_da3_node_tree_p(node2
);
825 count
= (nodehdr1
.count
- nodehdr2
.count
) / 2;
828 tp
= state
->args
->trans
;
830 * Two cases: high-to-low and low-to-high.
834 * Move elements in node2 up to make a hole.
836 tmp
= nodehdr2
.count
;
838 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
839 btree_s
= &btree2
[0];
840 btree_d
= &btree2
[count
];
841 memmove(btree_d
, btree_s
, tmp
);
845 * Move the req'd B-tree elements from high in node1 to
848 nodehdr2
.count
+= count
;
849 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
850 btree_s
= &btree1
[nodehdr1
.count
- count
];
851 btree_d
= &btree2
[0];
852 memcpy(btree_d
, btree_s
, tmp
);
853 nodehdr1
.count
-= count
;
856 * Move the req'd B-tree elements from low in node2 to
860 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
861 btree_s
= &btree2
[0];
862 btree_d
= &btree1
[nodehdr1
.count
];
863 memcpy(btree_d
, btree_s
, tmp
);
864 nodehdr1
.count
+= count
;
866 xfs_trans_log_buf(tp
, blk1
->bp
,
867 XFS_DA_LOGRANGE(node1
, btree_d
, tmp
));
870 * Move elements in node2 down to fill the hole.
872 tmp
= nodehdr2
.count
- count
;
873 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
874 btree_s
= &btree2
[count
];
875 btree_d
= &btree2
[0];
876 memmove(btree_d
, btree_s
, tmp
);
877 nodehdr2
.count
-= count
;
881 * Log header of node 1 and all current bits of node 2.
883 xfs_da3_node_hdr_to_disk(node1
, &nodehdr1
);
884 xfs_trans_log_buf(tp
, blk1
->bp
,
885 XFS_DA_LOGRANGE(node1
, &node1
->hdr
,
886 xfs_da3_node_hdr_size(node1
)));
888 xfs_da3_node_hdr_to_disk(node2
, &nodehdr2
);
889 xfs_trans_log_buf(tp
, blk2
->bp
,
890 XFS_DA_LOGRANGE(node2
, &node2
->hdr
,
891 xfs_da3_node_hdr_size(node2
) +
892 (sizeof(btree2
[0]) * nodehdr2
.count
)));
895 * Record the last hashval from each block for upward propagation.
896 * (note: don't use the swapped node pointers)
899 node1
= blk1
->bp
->b_addr
;
900 node2
= blk2
->bp
->b_addr
;
901 xfs_da3_node_hdr_from_disk(&nodehdr1
, node1
);
902 xfs_da3_node_hdr_from_disk(&nodehdr2
, node2
);
903 btree1
= xfs_da3_node_tree_p(node1
);
904 btree2
= xfs_da3_node_tree_p(node2
);
906 blk1
->hashval
= be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
);
907 blk2
->hashval
= be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
);
910 * Adjust the expected index for insertion.
912 if (blk1
->index
>= nodehdr1
.count
) {
913 blk2
->index
= blk1
->index
- nodehdr1
.count
;
914 blk1
->index
= nodehdr1
.count
+ 1; /* make it invalid */
919 * Add a new entry to an intermediate node.
923 struct xfs_da_state
*state
,
924 struct xfs_da_state_blk
*oldblk
,
925 struct xfs_da_state_blk
*newblk
)
927 struct xfs_da_intnode
*node
;
928 struct xfs_da3_icnode_hdr nodehdr
;
929 struct xfs_da_node_entry
*btree
;
932 trace_xfs_da_node_add(state
->args
);
934 node
= oldblk
->bp
->b_addr
;
935 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
936 btree
= xfs_da3_node_tree_p(node
);
938 ASSERT(oldblk
->index
>= 0 && oldblk
->index
<= nodehdr
.count
);
939 ASSERT(newblk
->blkno
!= 0);
940 if (state
->args
->whichfork
== XFS_DATA_FORK
)
941 ASSERT(newblk
->blkno
>= state
->mp
->m_dirleafblk
&&
942 newblk
->blkno
< state
->mp
->m_dirfreeblk
);
945 * We may need to make some room before we insert the new node.
948 if (oldblk
->index
< nodehdr
.count
) {
949 tmp
= (nodehdr
.count
- oldblk
->index
) * (uint
)sizeof(*btree
);
950 memmove(&btree
[oldblk
->index
+ 1], &btree
[oldblk
->index
], tmp
);
952 btree
[oldblk
->index
].hashval
= cpu_to_be32(newblk
->hashval
);
953 btree
[oldblk
->index
].before
= cpu_to_be32(newblk
->blkno
);
954 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
955 XFS_DA_LOGRANGE(node
, &btree
[oldblk
->index
],
956 tmp
+ sizeof(*btree
)));
959 xfs_da3_node_hdr_to_disk(node
, &nodehdr
);
960 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
961 XFS_DA_LOGRANGE(node
, &node
->hdr
, xfs_da3_node_hdr_size(node
)));
964 * Copy the last hash value from the oldblk to propagate upwards.
966 oldblk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
969 /*========================================================================
970 * Routines used for shrinking the Btree.
971 *========================================================================*/
974 * Deallocate an empty leaf node, remove it from its parent,
975 * possibly deallocating that block, etc...
979 struct xfs_da_state
*state
)
981 struct xfs_da_state_blk
*drop_blk
;
982 struct xfs_da_state_blk
*save_blk
;
986 trace_xfs_da_join(state
->args
);
988 drop_blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
989 save_blk
= &state
->altpath
.blk
[ state
->path
.active
-1 ];
990 ASSERT(state
->path
.blk
[0].magic
== XFS_DA_NODE_MAGIC
);
991 ASSERT(drop_blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
992 drop_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
995 * Walk back up the tree joining/deallocating as necessary.
996 * When we stop dropping blocks, break out.
998 for ( ; state
->path
.active
>= 2; drop_blk
--, save_blk
--,
999 state
->path
.active
--) {
1001 * See if we can combine the block with a neighbor.
1002 * (action == 0) => no options, just leave
1003 * (action == 1) => coalesce, then unlink
1004 * (action == 2) => block empty, unlink it
1006 switch (drop_blk
->magic
) {
1007 case XFS_ATTR_LEAF_MAGIC
:
1008 error
= xfs_attr3_leaf_toosmall(state
, &action
);
1013 xfs_attr3_leaf_unbalance(state
, drop_blk
, save_blk
);
1015 case XFS_DIR2_LEAFN_MAGIC
:
1016 error
= xfs_dir2_leafn_toosmall(state
, &action
);
1021 xfs_dir2_leafn_unbalance(state
, drop_blk
, save_blk
);
1023 case XFS_DA_NODE_MAGIC
:
1025 * Remove the offending node, fixup hashvals,
1026 * check for a toosmall neighbor.
1028 xfs_da3_node_remove(state
, drop_blk
);
1029 xfs_da3_fixhashpath(state
, &state
->path
);
1030 error
= xfs_da3_node_toosmall(state
, &action
);
1035 xfs_da3_node_unbalance(state
, drop_blk
, save_blk
);
1038 xfs_da3_fixhashpath(state
, &state
->altpath
);
1039 error
= xfs_da3_blk_unlink(state
, drop_blk
, save_blk
);
1040 xfs_da_state_kill_altpath(state
);
1043 error
= xfs_da_shrink_inode(state
->args
, drop_blk
->blkno
,
1045 drop_blk
->bp
= NULL
;
1050 * We joined all the way to the top. If it turns out that
1051 * we only have one entry in the root, make the child block
1054 xfs_da3_node_remove(state
, drop_blk
);
1055 xfs_da3_fixhashpath(state
, &state
->path
);
1056 error
= xfs_da3_root_join(state
, &state
->path
.blk
[0]);
1062 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo
*blkinfo
, __u16 level
)
1064 __be16 magic
= blkinfo
->magic
;
1067 ASSERT(magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1068 magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1069 magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1070 magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1072 ASSERT(magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1073 magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
));
1075 ASSERT(!blkinfo
->forw
);
1076 ASSERT(!blkinfo
->back
);
1079 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1083 * We have only one entry in the root. Copy the only remaining child of
1084 * the old root to block 0 as the new root node.
1088 struct xfs_da_state
*state
,
1089 struct xfs_da_state_blk
*root_blk
)
1091 struct xfs_da_intnode
*oldroot
;
1092 struct xfs_da_args
*args
;
1095 struct xfs_da3_icnode_hdr oldroothdr
;
1096 struct xfs_da_node_entry
*btree
;
1099 trace_xfs_da_root_join(state
->args
);
1101 ASSERT(root_blk
->magic
== XFS_DA_NODE_MAGIC
);
1104 oldroot
= root_blk
->bp
->b_addr
;
1105 xfs_da3_node_hdr_from_disk(&oldroothdr
, oldroot
);
1106 ASSERT(oldroothdr
.forw
== 0);
1107 ASSERT(oldroothdr
.back
== 0);
1110 * If the root has more than one child, then don't do anything.
1112 if (oldroothdr
.count
> 1)
1116 * Read in the (only) child block, then copy those bytes into
1117 * the root block's buffer and free the original child block.
1119 btree
= xfs_da3_node_tree_p(oldroot
);
1120 child
= be32_to_cpu(btree
[0].before
);
1122 error
= xfs_da3_node_read(args
->trans
, args
->dp
, child
, -1, &bp
,
1126 xfs_da_blkinfo_onlychild_validate(bp
->b_addr
, oldroothdr
.level
);
1129 * This could be copying a leaf back into the root block in the case of
1130 * there only being a single leaf block left in the tree. Hence we have
1131 * to update the b_ops pointer as well to match the buffer type change
1132 * that could occur. For dir3 blocks we also need to update the block
1133 * number in the buffer header.
1135 memcpy(root_blk
->bp
->b_addr
, bp
->b_addr
, state
->blocksize
);
1136 root_blk
->bp
->b_ops
= bp
->b_ops
;
1137 xfs_trans_buf_copy_type(root_blk
->bp
, bp
);
1138 if (oldroothdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1139 struct xfs_da3_blkinfo
*da3
= root_blk
->bp
->b_addr
;
1140 da3
->blkno
= cpu_to_be64(root_blk
->bp
->b_bn
);
1142 xfs_trans_log_buf(args
->trans
, root_blk
->bp
, 0, state
->blocksize
- 1);
1143 error
= xfs_da_shrink_inode(args
, child
, bp
);
1148 * Check a node block and its neighbors to see if the block should be
1149 * collapsed into one or the other neighbor. Always keep the block
1150 * with the smaller block number.
1151 * If the current block is over 50% full, don't try to join it, return 0.
1152 * If the block is empty, fill in the state structure and return 2.
1153 * If it can be collapsed, fill in the state structure and return 1.
1154 * If nothing can be done, return 0.
1157 xfs_da3_node_toosmall(
1158 struct xfs_da_state
*state
,
1161 struct xfs_da_intnode
*node
;
1162 struct xfs_da_state_blk
*blk
;
1163 struct xfs_da_blkinfo
*info
;
1166 struct xfs_da3_icnode_hdr nodehdr
;
1173 trace_xfs_da_node_toosmall(state
->args
);
1176 * Check for the degenerate case of the block being over 50% full.
1177 * If so, it's not worth even looking to see if we might be able
1178 * to coalesce with a sibling.
1180 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1181 info
= blk
->bp
->b_addr
;
1182 node
= (xfs_da_intnode_t
*)info
;
1183 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1184 if (nodehdr
.count
> (state
->node_ents
>> 1)) {
1185 *action
= 0; /* blk over 50%, don't try to join */
1186 return(0); /* blk over 50%, don't try to join */
1190 * Check for the degenerate case of the block being empty.
1191 * If the block is empty, we'll simply delete it, no need to
1192 * coalesce it with a sibling block. We choose (arbitrarily)
1193 * to merge with the forward block unless it is NULL.
1195 if (nodehdr
.count
== 0) {
1197 * Make altpath point to the block we want to keep and
1198 * path point to the block we want to drop (this one).
1200 forward
= (info
->forw
!= 0);
1201 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1202 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1215 * Examine each sibling block to see if we can coalesce with
1216 * at least 25% free space to spare. We need to figure out
1217 * whether to merge with the forward or the backward block.
1218 * We prefer coalescing with the lower numbered sibling so as
1219 * to shrink a directory over time.
1221 count
= state
->node_ents
;
1222 count
-= state
->node_ents
>> 2;
1223 count
-= nodehdr
.count
;
1225 /* start with smaller blk num */
1226 forward
= nodehdr
.forw
< nodehdr
.back
;
1227 for (i
= 0; i
< 2; forward
= !forward
, i
++) {
1228 struct xfs_da3_icnode_hdr thdr
;
1230 blkno
= nodehdr
.forw
;
1232 blkno
= nodehdr
.back
;
1235 error
= xfs_da3_node_read(state
->args
->trans
, state
->args
->dp
,
1236 blkno
, -1, &bp
, state
->args
->whichfork
);
1241 xfs_da3_node_hdr_from_disk(&thdr
, node
);
1242 xfs_trans_brelse(state
->args
->trans
, bp
);
1244 if (count
- thdr
.count
>= 0)
1245 break; /* fits with at least 25% to spare */
1253 * Make altpath point to the block we want to keep (the lower
1254 * numbered block) and path point to the block we want to drop.
1256 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1257 if (blkno
< blk
->blkno
) {
1258 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1261 error
= xfs_da3_path_shift(state
, &state
->path
, forward
,
1275 * Pick up the last hashvalue from an intermediate node.
1278 xfs_da3_node_lasthash(
1282 struct xfs_da_intnode
*node
;
1283 struct xfs_da_node_entry
*btree
;
1284 struct xfs_da3_icnode_hdr nodehdr
;
1287 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1289 *count
= nodehdr
.count
;
1292 btree
= xfs_da3_node_tree_p(node
);
1293 return be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1297 * Walk back up the tree adjusting hash values as necessary,
1298 * when we stop making changes, return.
1301 xfs_da3_fixhashpath(
1302 struct xfs_da_state
*state
,
1303 struct xfs_da_state_path
*path
)
1305 struct xfs_da_state_blk
*blk
;
1306 struct xfs_da_intnode
*node
;
1307 struct xfs_da_node_entry
*btree
;
1308 xfs_dahash_t lasthash
=0;
1312 trace_xfs_da_fixhashpath(state
->args
);
1314 level
= path
->active
-1;
1315 blk
= &path
->blk
[ level
];
1316 switch (blk
->magic
) {
1317 case XFS_ATTR_LEAF_MAGIC
:
1318 lasthash
= xfs_attr_leaf_lasthash(blk
->bp
, &count
);
1322 case XFS_DIR2_LEAFN_MAGIC
:
1323 lasthash
= xfs_dir2_leafn_lasthash(blk
->bp
, &count
);
1327 case XFS_DA_NODE_MAGIC
:
1328 lasthash
= xfs_da3_node_lasthash(blk
->bp
, &count
);
1333 for (blk
--, level
--; level
>= 0; blk
--, level
--) {
1334 struct xfs_da3_icnode_hdr nodehdr
;
1336 node
= blk
->bp
->b_addr
;
1337 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1338 btree
= xfs_da3_node_tree_p(node
);
1339 if (be32_to_cpu(btree
->hashval
) == lasthash
)
1341 blk
->hashval
= lasthash
;
1342 btree
[blk
->index
].hashval
= cpu_to_be32(lasthash
);
1343 xfs_trans_log_buf(state
->args
->trans
, blk
->bp
,
1344 XFS_DA_LOGRANGE(node
, &btree
[blk
->index
],
1347 lasthash
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1352 * Remove an entry from an intermediate node.
1355 xfs_da3_node_remove(
1356 struct xfs_da_state
*state
,
1357 struct xfs_da_state_blk
*drop_blk
)
1359 struct xfs_da_intnode
*node
;
1360 struct xfs_da3_icnode_hdr nodehdr
;
1361 struct xfs_da_node_entry
*btree
;
1365 trace_xfs_da_node_remove(state
->args
);
1367 node
= drop_blk
->bp
->b_addr
;
1368 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1369 ASSERT(drop_blk
->index
< nodehdr
.count
);
1370 ASSERT(drop_blk
->index
>= 0);
1373 * Copy over the offending entry, or just zero it out.
1375 index
= drop_blk
->index
;
1376 btree
= xfs_da3_node_tree_p(node
);
1377 if (index
< nodehdr
.count
- 1) {
1378 tmp
= nodehdr
.count
- index
- 1;
1379 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
1380 memmove(&btree
[index
], &btree
[index
+ 1], tmp
);
1381 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1382 XFS_DA_LOGRANGE(node
, &btree
[index
], tmp
));
1383 index
= nodehdr
.count
- 1;
1385 memset(&btree
[index
], 0, sizeof(xfs_da_node_entry_t
));
1386 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1387 XFS_DA_LOGRANGE(node
, &btree
[index
], sizeof(btree
[index
])));
1389 xfs_da3_node_hdr_to_disk(node
, &nodehdr
);
1390 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1391 XFS_DA_LOGRANGE(node
, &node
->hdr
, xfs_da3_node_hdr_size(node
)));
1394 * Copy the last hash value from the block to propagate upwards.
1396 drop_blk
->hashval
= be32_to_cpu(btree
[index
- 1].hashval
);
1400 * Unbalance the elements between two intermediate nodes,
1401 * move all Btree elements from one node into another.
1404 xfs_da3_node_unbalance(
1405 struct xfs_da_state
*state
,
1406 struct xfs_da_state_blk
*drop_blk
,
1407 struct xfs_da_state_blk
*save_blk
)
1409 struct xfs_da_intnode
*drop_node
;
1410 struct xfs_da_intnode
*save_node
;
1411 struct xfs_da_node_entry
*drop_btree
;
1412 struct xfs_da_node_entry
*save_btree
;
1413 struct xfs_da3_icnode_hdr drop_hdr
;
1414 struct xfs_da3_icnode_hdr save_hdr
;
1415 struct xfs_trans
*tp
;
1419 trace_xfs_da_node_unbalance(state
->args
);
1421 drop_node
= drop_blk
->bp
->b_addr
;
1422 save_node
= save_blk
->bp
->b_addr
;
1423 xfs_da3_node_hdr_from_disk(&drop_hdr
, drop_node
);
1424 xfs_da3_node_hdr_from_disk(&save_hdr
, save_node
);
1425 drop_btree
= xfs_da3_node_tree_p(drop_node
);
1426 save_btree
= xfs_da3_node_tree_p(save_node
);
1427 tp
= state
->args
->trans
;
1430 * If the dying block has lower hashvals, then move all the
1431 * elements in the remaining block up to make a hole.
1433 if ((be32_to_cpu(drop_btree
[0].hashval
) <
1434 be32_to_cpu(save_btree
[0].hashval
)) ||
1435 (be32_to_cpu(drop_btree
[drop_hdr
.count
- 1].hashval
) <
1436 be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
))) {
1437 /* XXX: check this - is memmove dst correct? */
1438 tmp
= save_hdr
.count
* sizeof(xfs_da_node_entry_t
);
1439 memmove(&save_btree
[drop_hdr
.count
], &save_btree
[0], tmp
);
1442 xfs_trans_log_buf(tp
, save_blk
->bp
,
1443 XFS_DA_LOGRANGE(save_node
, &save_btree
[0],
1444 (save_hdr
.count
+ drop_hdr
.count
) *
1445 sizeof(xfs_da_node_entry_t
)));
1447 sindex
= save_hdr
.count
;
1448 xfs_trans_log_buf(tp
, save_blk
->bp
,
1449 XFS_DA_LOGRANGE(save_node
, &save_btree
[sindex
],
1450 drop_hdr
.count
* sizeof(xfs_da_node_entry_t
)));
1454 * Move all the B-tree elements from drop_blk to save_blk.
1456 tmp
= drop_hdr
.count
* (uint
)sizeof(xfs_da_node_entry_t
);
1457 memcpy(&save_btree
[sindex
], &drop_btree
[0], tmp
);
1458 save_hdr
.count
+= drop_hdr
.count
;
1460 xfs_da3_node_hdr_to_disk(save_node
, &save_hdr
);
1461 xfs_trans_log_buf(tp
, save_blk
->bp
,
1462 XFS_DA_LOGRANGE(save_node
, &save_node
->hdr
,
1463 xfs_da3_node_hdr_size(save_node
)));
1466 * Save the last hashval in the remaining block for upward propagation.
1468 save_blk
->hashval
= be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
);
1471 /*========================================================================
1472 * Routines used for finding things in the Btree.
1473 *========================================================================*/
1476 * Walk down the Btree looking for a particular filename, filling
1477 * in the state structure as we go.
1479 * We will set the state structure to point to each of the elements
1480 * in each of the nodes where either the hashval is or should be.
1482 * We support duplicate hashval's so for each entry in the current
1483 * node that could contain the desired hashval, descend. This is a
1484 * pruned depth-first tree search.
1487 xfs_da3_node_lookup_int(
1488 struct xfs_da_state
*state
,
1491 struct xfs_da_state_blk
*blk
;
1492 struct xfs_da_blkinfo
*curr
;
1493 struct xfs_da_intnode
*node
;
1494 struct xfs_da_node_entry
*btree
;
1495 struct xfs_da3_icnode_hdr nodehdr
;
1496 struct xfs_da_args
*args
;
1498 xfs_dahash_t hashval
;
1499 xfs_dahash_t btreehashval
;
1509 * Descend thru the B-tree searching each level for the right
1510 * node to use, until the right hashval is found.
1512 blkno
= (args
->whichfork
== XFS_DATA_FORK
)? state
->mp
->m_dirleafblk
: 0;
1513 for (blk
= &state
->path
.blk
[0], state
->path
.active
= 1;
1514 state
->path
.active
<= XFS_DA_NODE_MAXDEPTH
;
1515 blk
++, state
->path
.active
++) {
1517 * Read the next node down in the tree.
1520 error
= xfs_da3_node_read(args
->trans
, args
->dp
, blkno
,
1521 -1, &blk
->bp
, args
->whichfork
);
1524 state
->path
.active
--;
1527 curr
= blk
->bp
->b_addr
;
1528 blk
->magic
= be16_to_cpu(curr
->magic
);
1530 if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
1531 blk
->magic
== XFS_ATTR3_LEAF_MAGIC
) {
1532 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1533 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1537 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1538 blk
->magic
== XFS_DIR3_LEAFN_MAGIC
) {
1539 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1540 blk
->hashval
= xfs_dir2_leafn_lasthash(blk
->bp
, NULL
);
1544 blk
->magic
= XFS_DA_NODE_MAGIC
;
1548 * Search an intermediate node for a match.
1550 node
= blk
->bp
->b_addr
;
1551 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1552 btree
= xfs_da3_node_tree_p(node
);
1554 max
= nodehdr
.count
;
1555 blk
->hashval
= be32_to_cpu(btree
[max
- 1].hashval
);
1558 * Binary search. (note: small blocks will skip loop)
1560 probe
= span
= max
/ 2;
1561 hashval
= args
->hashval
;
1564 btreehashval
= be32_to_cpu(btree
[probe
].hashval
);
1565 if (btreehashval
< hashval
)
1567 else if (btreehashval
> hashval
)
1572 ASSERT((probe
>= 0) && (probe
< max
));
1573 ASSERT((span
<= 4) ||
1574 (be32_to_cpu(btree
[probe
].hashval
) == hashval
));
1577 * Since we may have duplicate hashval's, find the first
1578 * matching hashval in the node.
1581 be32_to_cpu(btree
[probe
].hashval
) >= hashval
) {
1584 while (probe
< max
&&
1585 be32_to_cpu(btree
[probe
].hashval
) < hashval
) {
1590 * Pick the right block to descend on.
1593 blk
->index
= max
- 1;
1594 blkno
= be32_to_cpu(btree
[max
- 1].before
);
1597 blkno
= be32_to_cpu(btree
[probe
].before
);
1602 * A leaf block that ends in the hashval that we are interested in
1603 * (final hashval == search hashval) means that the next block may
1604 * contain more entries with the same hashval, shift upward to the
1605 * next leaf and keep searching.
1608 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
) {
1609 retval
= xfs_dir2_leafn_lookup_int(blk
->bp
, args
,
1610 &blk
->index
, state
);
1611 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1612 retval
= xfs_attr3_leaf_lookup_int(blk
->bp
, args
);
1613 blk
->index
= args
->index
;
1614 args
->blkno
= blk
->blkno
;
1617 return XFS_ERROR(EFSCORRUPTED
);
1619 if (((retval
== ENOENT
) || (retval
== ENOATTR
)) &&
1620 (blk
->hashval
== args
->hashval
)) {
1621 error
= xfs_da3_path_shift(state
, &state
->path
, 1, 1,
1627 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1628 /* path_shift() gives ENOENT */
1629 retval
= XFS_ERROR(ENOATTR
);
1638 /*========================================================================
1640 *========================================================================*/
1643 * Compare two intermediate nodes for "order".
1647 struct xfs_buf
*node1_bp
,
1648 struct xfs_buf
*node2_bp
)
1650 struct xfs_da_intnode
*node1
;
1651 struct xfs_da_intnode
*node2
;
1652 struct xfs_da_node_entry
*btree1
;
1653 struct xfs_da_node_entry
*btree2
;
1654 struct xfs_da3_icnode_hdr node1hdr
;
1655 struct xfs_da3_icnode_hdr node2hdr
;
1657 node1
= node1_bp
->b_addr
;
1658 node2
= node2_bp
->b_addr
;
1659 xfs_da3_node_hdr_from_disk(&node1hdr
, node1
);
1660 xfs_da3_node_hdr_from_disk(&node2hdr
, node2
);
1661 btree1
= xfs_da3_node_tree_p(node1
);
1662 btree2
= xfs_da3_node_tree_p(node2
);
1664 if (node1hdr
.count
> 0 && node2hdr
.count
> 0 &&
1665 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
1666 (be32_to_cpu(btree2
[node2hdr
.count
- 1].hashval
) <
1667 be32_to_cpu(btree1
[node1hdr
.count
- 1].hashval
)))) {
1674 * Link a new block into a doubly linked list of blocks (of whatever type).
1678 struct xfs_da_state
*state
,
1679 struct xfs_da_state_blk
*old_blk
,
1680 struct xfs_da_state_blk
*new_blk
)
1682 struct xfs_da_blkinfo
*old_info
;
1683 struct xfs_da_blkinfo
*new_info
;
1684 struct xfs_da_blkinfo
*tmp_info
;
1685 struct xfs_da_args
*args
;
1691 * Set up environment.
1694 ASSERT(args
!= NULL
);
1695 old_info
= old_blk
->bp
->b_addr
;
1696 new_info
= new_blk
->bp
->b_addr
;
1697 ASSERT(old_blk
->magic
== XFS_DA_NODE_MAGIC
||
1698 old_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1699 old_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1701 switch (old_blk
->magic
) {
1702 case XFS_ATTR_LEAF_MAGIC
:
1703 before
= xfs_attr_leaf_order(old_blk
->bp
, new_blk
->bp
);
1705 case XFS_DIR2_LEAFN_MAGIC
:
1706 before
= xfs_dir2_leafn_order(old_blk
->bp
, new_blk
->bp
);
1708 case XFS_DA_NODE_MAGIC
:
1709 before
= xfs_da3_node_order(old_blk
->bp
, new_blk
->bp
);
1714 * Link blocks in appropriate order.
1718 * Link new block in before existing block.
1720 trace_xfs_da_link_before(args
);
1721 new_info
->forw
= cpu_to_be32(old_blk
->blkno
);
1722 new_info
->back
= old_info
->back
;
1723 if (old_info
->back
) {
1724 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1725 be32_to_cpu(old_info
->back
),
1726 -1, &bp
, args
->whichfork
);
1730 tmp_info
= bp
->b_addr
;
1731 ASSERT(tmp_info
->magic
== old_info
->magic
);
1732 ASSERT(be32_to_cpu(tmp_info
->forw
) == old_blk
->blkno
);
1733 tmp_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1734 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1736 old_info
->back
= cpu_to_be32(new_blk
->blkno
);
1739 * Link new block in after existing block.
1741 trace_xfs_da_link_after(args
);
1742 new_info
->forw
= old_info
->forw
;
1743 new_info
->back
= cpu_to_be32(old_blk
->blkno
);
1744 if (old_info
->forw
) {
1745 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1746 be32_to_cpu(old_info
->forw
),
1747 -1, &bp
, args
->whichfork
);
1751 tmp_info
= bp
->b_addr
;
1752 ASSERT(tmp_info
->magic
== old_info
->magic
);
1753 ASSERT(be32_to_cpu(tmp_info
->back
) == old_blk
->blkno
);
1754 tmp_info
->back
= cpu_to_be32(new_blk
->blkno
);
1755 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1757 old_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1760 xfs_trans_log_buf(args
->trans
, old_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1761 xfs_trans_log_buf(args
->trans
, new_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1766 * Unlink a block from a doubly linked list of blocks.
1768 STATIC
int /* error */
1770 struct xfs_da_state
*state
,
1771 struct xfs_da_state_blk
*drop_blk
,
1772 struct xfs_da_state_blk
*save_blk
)
1774 struct xfs_da_blkinfo
*drop_info
;
1775 struct xfs_da_blkinfo
*save_info
;
1776 struct xfs_da_blkinfo
*tmp_info
;
1777 struct xfs_da_args
*args
;
1782 * Set up environment.
1785 ASSERT(args
!= NULL
);
1786 save_info
= save_blk
->bp
->b_addr
;
1787 drop_info
= drop_blk
->bp
->b_addr
;
1788 ASSERT(save_blk
->magic
== XFS_DA_NODE_MAGIC
||
1789 save_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1790 save_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1791 ASSERT(save_blk
->magic
== drop_blk
->magic
);
1792 ASSERT((be32_to_cpu(save_info
->forw
) == drop_blk
->blkno
) ||
1793 (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
));
1794 ASSERT((be32_to_cpu(drop_info
->forw
) == save_blk
->blkno
) ||
1795 (be32_to_cpu(drop_info
->back
) == save_blk
->blkno
));
1798 * Unlink the leaf block from the doubly linked chain of leaves.
1800 if (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
) {
1801 trace_xfs_da_unlink_back(args
);
1802 save_info
->back
= drop_info
->back
;
1803 if (drop_info
->back
) {
1804 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1805 be32_to_cpu(drop_info
->back
),
1806 -1, &bp
, args
->whichfork
);
1810 tmp_info
= bp
->b_addr
;
1811 ASSERT(tmp_info
->magic
== save_info
->magic
);
1812 ASSERT(be32_to_cpu(tmp_info
->forw
) == drop_blk
->blkno
);
1813 tmp_info
->forw
= cpu_to_be32(save_blk
->blkno
);
1814 xfs_trans_log_buf(args
->trans
, bp
, 0,
1815 sizeof(*tmp_info
) - 1);
1818 trace_xfs_da_unlink_forward(args
);
1819 save_info
->forw
= drop_info
->forw
;
1820 if (drop_info
->forw
) {
1821 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1822 be32_to_cpu(drop_info
->forw
),
1823 -1, &bp
, args
->whichfork
);
1827 tmp_info
= bp
->b_addr
;
1828 ASSERT(tmp_info
->magic
== save_info
->magic
);
1829 ASSERT(be32_to_cpu(tmp_info
->back
) == drop_blk
->blkno
);
1830 tmp_info
->back
= cpu_to_be32(save_blk
->blkno
);
1831 xfs_trans_log_buf(args
->trans
, bp
, 0,
1832 sizeof(*tmp_info
) - 1);
1836 xfs_trans_log_buf(args
->trans
, save_blk
->bp
, 0, sizeof(*save_info
) - 1);
1841 * Move a path "forward" or "!forward" one block at the current level.
1843 * This routine will adjust a "path" to point to the next block
1844 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1845 * Btree, including updating pointers to the intermediate nodes between
1846 * the new bottom and the root.
1850 struct xfs_da_state
*state
,
1851 struct xfs_da_state_path
*path
,
1856 struct xfs_da_state_blk
*blk
;
1857 struct xfs_da_blkinfo
*info
;
1858 struct xfs_da_intnode
*node
;
1859 struct xfs_da_args
*args
;
1860 struct xfs_da_node_entry
*btree
;
1861 struct xfs_da3_icnode_hdr nodehdr
;
1862 xfs_dablk_t blkno
= 0;
1866 trace_xfs_da_path_shift(state
->args
);
1869 * Roll up the Btree looking for the first block where our
1870 * current index is not at the edge of the block. Note that
1871 * we skip the bottom layer because we want the sibling block.
1874 ASSERT(args
!= NULL
);
1875 ASSERT(path
!= NULL
);
1876 ASSERT((path
->active
> 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1877 level
= (path
->active
-1) - 1; /* skip bottom layer in path */
1878 for (blk
= &path
->blk
[level
]; level
>= 0; blk
--, level
--) {
1879 node
= blk
->bp
->b_addr
;
1880 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1881 btree
= xfs_da3_node_tree_p(node
);
1883 if (forward
&& (blk
->index
< nodehdr
.count
- 1)) {
1885 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1887 } else if (!forward
&& (blk
->index
> 0)) {
1889 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1894 *result
= XFS_ERROR(ENOENT
); /* we're out of our tree */
1895 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1900 * Roll down the edge of the subtree until we reach the
1901 * same depth we were at originally.
1903 for (blk
++, level
++; level
< path
->active
; blk
++, level
++) {
1905 * Release the old block.
1906 * (if it's dirty, trans won't actually let go)
1909 xfs_trans_brelse(args
->trans
, blk
->bp
);
1912 * Read the next child block.
1915 error
= xfs_da3_node_read(args
->trans
, args
->dp
, blkno
, -1,
1916 &blk
->bp
, args
->whichfork
);
1919 info
= blk
->bp
->b_addr
;
1920 ASSERT(info
->magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1921 info
->magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
1922 info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1923 info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1924 info
->magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1925 info
->magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1929 * Note: we flatten the magic number to a single type so we
1930 * don't have to compare against crc/non-crc types elsewhere.
1932 switch (be16_to_cpu(info
->magic
)) {
1933 case XFS_DA_NODE_MAGIC
:
1934 case XFS_DA3_NODE_MAGIC
:
1935 blk
->magic
= XFS_DA_NODE_MAGIC
;
1936 node
= (xfs_da_intnode_t
*)info
;
1937 xfs_da3_node_hdr_from_disk(&nodehdr
, node
);
1938 btree
= xfs_da3_node_tree_p(node
);
1939 blk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1943 blk
->index
= nodehdr
.count
- 1;
1944 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1946 case XFS_ATTR_LEAF_MAGIC
:
1947 case XFS_ATTR3_LEAF_MAGIC
:
1948 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1949 ASSERT(level
== path
->active
-1);
1951 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
,
1954 case XFS_DIR2_LEAFN_MAGIC
:
1955 case XFS_DIR3_LEAFN_MAGIC
:
1956 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1957 ASSERT(level
== path
->active
-1);
1959 blk
->hashval
= xfs_dir2_leafn_lasthash(blk
->bp
,
1972 /*========================================================================
1974 *========================================================================*/
1977 * Implement a simple hash on a character string.
1978 * Rotate the hash value by 7 bits, then XOR each character in.
1979 * This is implemented with some source-level loop unrolling.
1982 xfs_da_hashname(const __uint8_t
*name
, int namelen
)
1987 * Do four characters at a time as long as we can.
1989 for (hash
= 0; namelen
>= 4; namelen
-= 4, name
+= 4)
1990 hash
= (name
[0] << 21) ^ (name
[1] << 14) ^ (name
[2] << 7) ^
1991 (name
[3] << 0) ^ rol32(hash
, 7 * 4);
1994 * Now do the rest of the characters.
1998 return (name
[0] << 14) ^ (name
[1] << 7) ^ (name
[2] << 0) ^
2001 return (name
[0] << 7) ^ (name
[1] << 0) ^ rol32(hash
, 7 * 2);
2003 return (name
[0] << 0) ^ rol32(hash
, 7 * 1);
2004 default: /* case 0: */
2011 struct xfs_da_args
*args
,
2012 const unsigned char *name
,
2015 return (args
->namelen
== len
&& memcmp(args
->name
, name
, len
) == 0) ?
2016 XFS_CMP_EXACT
: XFS_CMP_DIFFERENT
;
2020 xfs_default_hashname(
2021 struct xfs_name
*name
)
2023 return xfs_da_hashname(name
->name
, name
->len
);
2026 const struct xfs_nameops xfs_default_nameops
= {
2027 .hashname
= xfs_default_hashname
,
2028 .compname
= xfs_da_compname
2032 xfs_da_grow_inode_int(
2033 struct xfs_da_args
*args
,
2037 struct xfs_trans
*tp
= args
->trans
;
2038 struct xfs_inode
*dp
= args
->dp
;
2039 int w
= args
->whichfork
;
2040 xfs_drfsbno_t nblks
= dp
->i_d
.di_nblocks
;
2041 struct xfs_bmbt_irec map
, *mapp
;
2042 int nmap
, error
, got
, i
, mapi
;
2045 * Find a spot in the file space to put the new block.
2047 error
= xfs_bmap_first_unused(tp
, dp
, count
, bno
, w
);
2052 * Try mapping it in one filesystem block.
2055 ASSERT(args
->firstblock
!= NULL
);
2056 error
= xfs_bmapi_write(tp
, dp
, *bno
, count
,
2057 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
|XFS_BMAPI_CONTIG
,
2058 args
->firstblock
, args
->total
, &map
, &nmap
,
2067 } else if (nmap
== 0 && count
> 1) {
2072 * If we didn't get it and the block might work if fragmented,
2073 * try without the CONTIG flag. Loop until we get it all.
2075 mapp
= kmem_alloc(sizeof(*mapp
) * count
, KM_SLEEP
);
2076 for (b
= *bno
, mapi
= 0; b
< *bno
+ count
; ) {
2077 nmap
= MIN(XFS_BMAP_MAX_NMAP
, count
);
2078 c
= (int)(*bno
+ count
- b
);
2079 error
= xfs_bmapi_write(tp
, dp
, b
, c
,
2080 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
,
2081 args
->firstblock
, args
->total
,
2082 &mapp
[mapi
], &nmap
, args
->flist
);
2088 b
= mapp
[mapi
- 1].br_startoff
+
2089 mapp
[mapi
- 1].br_blockcount
;
2097 * Count the blocks we got, make sure it matches the total.
2099 for (i
= 0, got
= 0; i
< mapi
; i
++)
2100 got
+= mapp
[i
].br_blockcount
;
2101 if (got
!= count
|| mapp
[0].br_startoff
!= *bno
||
2102 mapp
[mapi
- 1].br_startoff
+ mapp
[mapi
- 1].br_blockcount
!=
2104 error
= XFS_ERROR(ENOSPC
);
2108 /* account for newly allocated blocks in reserved blocks total */
2109 args
->total
-= dp
->i_d
.di_nblocks
- nblks
;
2118 * Add a block to the btree ahead of the file.
2119 * Return the new block number to the caller.
2123 struct xfs_da_args
*args
,
2124 xfs_dablk_t
*new_blkno
)
2130 trace_xfs_da_grow_inode(args
);
2132 if (args
->whichfork
== XFS_DATA_FORK
) {
2133 bno
= args
->dp
->i_mount
->m_dirleafblk
;
2134 count
= args
->dp
->i_mount
->m_dirblkfsbs
;
2140 error
= xfs_da_grow_inode_int(args
, &bno
, count
);
2142 *new_blkno
= (xfs_dablk_t
)bno
;
2147 * Ick. We need to always be able to remove a btree block, even
2148 * if there's no space reservation because the filesystem is full.
2149 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2150 * It swaps the target block with the last block in the file. The
2151 * last block in the file can always be removed since it can't cause
2152 * a bmap btree split to do that.
2155 xfs_da3_swap_lastblock(
2156 struct xfs_da_args
*args
,
2157 xfs_dablk_t
*dead_blknop
,
2158 struct xfs_buf
**dead_bufp
)
2160 struct xfs_da_blkinfo
*dead_info
;
2161 struct xfs_da_blkinfo
*sib_info
;
2162 struct xfs_da_intnode
*par_node
;
2163 struct xfs_da_intnode
*dead_node
;
2164 struct xfs_dir2_leaf
*dead_leaf2
;
2165 struct xfs_da_node_entry
*btree
;
2166 struct xfs_da3_icnode_hdr par_hdr
;
2167 struct xfs_inode
*ip
;
2168 struct xfs_trans
*tp
;
2169 struct xfs_mount
*mp
;
2170 struct xfs_buf
*dead_buf
;
2171 struct xfs_buf
*last_buf
;
2172 struct xfs_buf
*sib_buf
;
2173 struct xfs_buf
*par_buf
;
2174 xfs_dahash_t dead_hash
;
2175 xfs_fileoff_t lastoff
;
2176 xfs_dablk_t dead_blkno
;
2177 xfs_dablk_t last_blkno
;
2178 xfs_dablk_t sib_blkno
;
2179 xfs_dablk_t par_blkno
;
2186 trace_xfs_da_swap_lastblock(args
);
2188 dead_buf
= *dead_bufp
;
2189 dead_blkno
= *dead_blknop
;
2192 w
= args
->whichfork
;
2193 ASSERT(w
== XFS_DATA_FORK
);
2195 lastoff
= mp
->m_dirfreeblk
;
2196 error
= xfs_bmap_last_before(tp
, ip
, &lastoff
, w
);
2199 if (unlikely(lastoff
== 0)) {
2200 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW
,
2202 return XFS_ERROR(EFSCORRUPTED
);
2205 * Read the last block in the btree space.
2207 last_blkno
= (xfs_dablk_t
)lastoff
- mp
->m_dirblkfsbs
;
2208 error
= xfs_da3_node_read(tp
, ip
, last_blkno
, -1, &last_buf
, w
);
2212 * Copy the last block into the dead buffer and log it.
2214 memcpy(dead_buf
->b_addr
, last_buf
->b_addr
, mp
->m_dirblksize
);
2215 xfs_trans_log_buf(tp
, dead_buf
, 0, mp
->m_dirblksize
- 1);
2216 dead_info
= dead_buf
->b_addr
;
2218 * Get values from the moved block.
2220 if (dead_info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
2221 dead_info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
2222 struct xfs_dir3_icleaf_hdr leafhdr
;
2223 struct xfs_dir2_leaf_entry
*ents
;
2225 dead_leaf2
= (xfs_dir2_leaf_t
*)dead_info
;
2226 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, dead_leaf2
);
2227 ents
= xfs_dir3_leaf_ents_p(dead_leaf2
);
2229 dead_hash
= be32_to_cpu(ents
[leafhdr
.count
- 1].hashval
);
2231 struct xfs_da3_icnode_hdr deadhdr
;
2233 dead_node
= (xfs_da_intnode_t
*)dead_info
;
2234 xfs_da3_node_hdr_from_disk(&deadhdr
, dead_node
);
2235 btree
= xfs_da3_node_tree_p(dead_node
);
2236 dead_level
= deadhdr
.level
;
2237 dead_hash
= be32_to_cpu(btree
[deadhdr
.count
- 1].hashval
);
2239 sib_buf
= par_buf
= NULL
;
2241 * If the moved block has a left sibling, fix up the pointers.
2243 if ((sib_blkno
= be32_to_cpu(dead_info
->back
))) {
2244 error
= xfs_da3_node_read(tp
, ip
, sib_blkno
, -1, &sib_buf
, w
);
2247 sib_info
= sib_buf
->b_addr
;
2249 be32_to_cpu(sib_info
->forw
) != last_blkno
||
2250 sib_info
->magic
!= dead_info
->magic
)) {
2251 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2252 XFS_ERRLEVEL_LOW
, mp
);
2253 error
= XFS_ERROR(EFSCORRUPTED
);
2256 sib_info
->forw
= cpu_to_be32(dead_blkno
);
2257 xfs_trans_log_buf(tp
, sib_buf
,
2258 XFS_DA_LOGRANGE(sib_info
, &sib_info
->forw
,
2259 sizeof(sib_info
->forw
)));
2263 * If the moved block has a right sibling, fix up the pointers.
2265 if ((sib_blkno
= be32_to_cpu(dead_info
->forw
))) {
2266 error
= xfs_da3_node_read(tp
, ip
, sib_blkno
, -1, &sib_buf
, w
);
2269 sib_info
= sib_buf
->b_addr
;
2271 be32_to_cpu(sib_info
->back
) != last_blkno
||
2272 sib_info
->magic
!= dead_info
->magic
)) {
2273 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2274 XFS_ERRLEVEL_LOW
, mp
);
2275 error
= XFS_ERROR(EFSCORRUPTED
);
2278 sib_info
->back
= cpu_to_be32(dead_blkno
);
2279 xfs_trans_log_buf(tp
, sib_buf
,
2280 XFS_DA_LOGRANGE(sib_info
, &sib_info
->back
,
2281 sizeof(sib_info
->back
)));
2284 par_blkno
= mp
->m_dirleafblk
;
2287 * Walk down the tree looking for the parent of the moved block.
2290 error
= xfs_da3_node_read(tp
, ip
, par_blkno
, -1, &par_buf
, w
);
2293 par_node
= par_buf
->b_addr
;
2294 xfs_da3_node_hdr_from_disk(&par_hdr
, par_node
);
2295 if (level
>= 0 && level
!= par_hdr
.level
+ 1) {
2296 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2297 XFS_ERRLEVEL_LOW
, mp
);
2298 error
= XFS_ERROR(EFSCORRUPTED
);
2301 level
= par_hdr
.level
;
2302 btree
= xfs_da3_node_tree_p(par_node
);
2304 entno
< par_hdr
.count
&&
2305 be32_to_cpu(btree
[entno
].hashval
) < dead_hash
;
2308 if (entno
== par_hdr
.count
) {
2309 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2310 XFS_ERRLEVEL_LOW
, mp
);
2311 error
= XFS_ERROR(EFSCORRUPTED
);
2314 par_blkno
= be32_to_cpu(btree
[entno
].before
);
2315 if (level
== dead_level
+ 1)
2317 xfs_trans_brelse(tp
, par_buf
);
2321 * We're in the right parent block.
2322 * Look for the right entry.
2326 entno
< par_hdr
.count
&&
2327 be32_to_cpu(btree
[entno
].before
) != last_blkno
;
2330 if (entno
< par_hdr
.count
)
2332 par_blkno
= par_hdr
.forw
;
2333 xfs_trans_brelse(tp
, par_buf
);
2335 if (unlikely(par_blkno
== 0)) {
2336 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2337 XFS_ERRLEVEL_LOW
, mp
);
2338 error
= XFS_ERROR(EFSCORRUPTED
);
2341 error
= xfs_da3_node_read(tp
, ip
, par_blkno
, -1, &par_buf
, w
);
2344 par_node
= par_buf
->b_addr
;
2345 xfs_da3_node_hdr_from_disk(&par_hdr
, par_node
);
2346 if (par_hdr
.level
!= level
) {
2347 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2348 XFS_ERRLEVEL_LOW
, mp
);
2349 error
= XFS_ERROR(EFSCORRUPTED
);
2352 btree
= xfs_da3_node_tree_p(par_node
);
2356 * Update the parent entry pointing to the moved block.
2358 btree
[entno
].before
= cpu_to_be32(dead_blkno
);
2359 xfs_trans_log_buf(tp
, par_buf
,
2360 XFS_DA_LOGRANGE(par_node
, &btree
[entno
].before
,
2361 sizeof(btree
[entno
].before
)));
2362 *dead_blknop
= last_blkno
;
2363 *dead_bufp
= last_buf
;
2367 xfs_trans_brelse(tp
, par_buf
);
2369 xfs_trans_brelse(tp
, sib_buf
);
2370 xfs_trans_brelse(tp
, last_buf
);
2375 * Remove a btree block from a directory or attribute.
2378 xfs_da_shrink_inode(
2379 xfs_da_args_t
*args
,
2380 xfs_dablk_t dead_blkno
,
2381 struct xfs_buf
*dead_buf
)
2384 int done
, error
, w
, count
;
2388 trace_xfs_da_shrink_inode(args
);
2391 w
= args
->whichfork
;
2394 if (w
== XFS_DATA_FORK
)
2395 count
= mp
->m_dirblkfsbs
;
2400 * Remove extents. If we get ENOSPC for a dir we have to move
2401 * the last block to the place we want to kill.
2403 error
= xfs_bunmapi(tp
, dp
, dead_blkno
, count
,
2404 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
,
2405 0, args
->firstblock
, args
->flist
, &done
);
2406 if (error
== ENOSPC
) {
2407 if (w
!= XFS_DATA_FORK
)
2409 error
= xfs_da3_swap_lastblock(args
, &dead_blkno
,
2417 xfs_trans_binval(tp
, dead_buf
);
2422 * See if the mapping(s) for this btree block are valid, i.e.
2423 * don't contain holes, are logically contiguous, and cover the whole range.
2426 xfs_da_map_covers_blocks(
2428 xfs_bmbt_irec_t
*mapp
,
2435 for (i
= 0, off
= bno
; i
< nmap
; i
++) {
2436 if (mapp
[i
].br_startblock
== HOLESTARTBLOCK
||
2437 mapp
[i
].br_startblock
== DELAYSTARTBLOCK
) {
2440 if (off
!= mapp
[i
].br_startoff
) {
2443 off
+= mapp
[i
].br_blockcount
;
2445 return off
== bno
+ count
;
2449 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2451 * For the single map case, it is assumed that the caller has provided a pointer
2452 * to a valid xfs_buf_map. For the multiple map case, this function will
2453 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2454 * map pointer with the allocated map.
2457 xfs_buf_map_from_irec(
2458 struct xfs_mount
*mp
,
2459 struct xfs_buf_map
**mapp
,
2461 struct xfs_bmbt_irec
*irecs
,
2464 struct xfs_buf_map
*map
;
2467 ASSERT(*nmaps
== 1);
2468 ASSERT(nirecs
>= 1);
2471 map
= kmem_zalloc(nirecs
* sizeof(struct xfs_buf_map
),
2472 KM_SLEEP
| KM_NOFS
);
2480 for (i
= 0; i
< *nmaps
; i
++) {
2481 ASSERT(irecs
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
2482 irecs
[i
].br_startblock
!= HOLESTARTBLOCK
);
2483 map
[i
].bm_bn
= XFS_FSB_TO_DADDR(mp
, irecs
[i
].br_startblock
);
2484 map
[i
].bm_len
= XFS_FSB_TO_BB(mp
, irecs
[i
].br_blockcount
);
2490 * Map the block we are given ready for reading. There are three possible return
2492 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2493 * caller knows not to execute a subsequent read.
2494 * 0 - if we mapped the block successfully
2495 * >0 - positive error number if there was an error.
2499 struct xfs_trans
*trans
,
2500 struct xfs_inode
*dp
,
2502 xfs_daddr_t mappedbno
,
2504 struct xfs_buf_map
**map
,
2507 struct xfs_mount
*mp
= dp
->i_mount
;
2510 struct xfs_bmbt_irec irec
;
2511 struct xfs_bmbt_irec
*irecs
= &irec
;
2514 ASSERT(map
&& *map
);
2515 ASSERT(*nmaps
== 1);
2517 nfsb
= (whichfork
== XFS_DATA_FORK
) ? mp
->m_dirblkfsbs
: 1;
2520 * Caller doesn't have a mapping. -2 means don't complain
2521 * if we land in a hole.
2523 if (mappedbno
== -1 || mappedbno
== -2) {
2525 * Optimize the one-block case.
2528 irecs
= kmem_zalloc(sizeof(irec
) * nfsb
,
2529 KM_SLEEP
| KM_NOFS
);
2532 error
= xfs_bmapi_read(dp
, (xfs_fileoff_t
)bno
, nfsb
, irecs
,
2533 &nirecs
, xfs_bmapi_aflag(whichfork
));
2537 irecs
->br_startblock
= XFS_DADDR_TO_FSB(mp
, mappedbno
);
2538 irecs
->br_startoff
= (xfs_fileoff_t
)bno
;
2539 irecs
->br_blockcount
= nfsb
;
2540 irecs
->br_state
= 0;
2544 if (!xfs_da_map_covers_blocks(nirecs
, irecs
, bno
, nfsb
)) {
2545 error
= mappedbno
== -2 ? -1 : XFS_ERROR(EFSCORRUPTED
);
2546 if (unlikely(error
== EFSCORRUPTED
)) {
2547 if (xfs_error_level
>= XFS_ERRLEVEL_LOW
) {
2549 xfs_alert(mp
, "%s: bno %lld dir: inode %lld",
2550 __func__
, (long long)bno
,
2551 (long long)dp
->i_ino
);
2552 for (i
= 0; i
< *nmaps
; i
++) {
2554 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2556 (long long)irecs
[i
].br_startoff
,
2557 (long long)irecs
[i
].br_startblock
,
2558 (long long)irecs
[i
].br_blockcount
,
2562 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2563 XFS_ERRLEVEL_LOW
, mp
);
2567 error
= xfs_buf_map_from_irec(mp
, map
, nmaps
, irecs
, nirecs
);
2575 * Get a buffer for the dir/attr block.
2579 struct xfs_trans
*trans
,
2580 struct xfs_inode
*dp
,
2582 xfs_daddr_t mappedbno
,
2583 struct xfs_buf
**bpp
,
2587 struct xfs_buf_map map
;
2588 struct xfs_buf_map
*mapp
;
2595 error
= xfs_dabuf_map(trans
, dp
, bno
, mappedbno
, whichfork
,
2598 /* mapping a hole is not an error, but we don't continue */
2604 bp
= xfs_trans_get_buf_map(trans
, dp
->i_mount
->m_ddev_targp
,
2606 error
= bp
? bp
->b_error
: XFS_ERROR(EIO
);
2608 xfs_trans_brelse(trans
, bp
);
2622 * Get a buffer for the dir/attr block, fill in the contents.
2626 struct xfs_trans
*trans
,
2627 struct xfs_inode
*dp
,
2629 xfs_daddr_t mappedbno
,
2630 struct xfs_buf
**bpp
,
2632 const struct xfs_buf_ops
*ops
)
2635 struct xfs_buf_map map
;
2636 struct xfs_buf_map
*mapp
;
2643 error
= xfs_dabuf_map(trans
, dp
, bno
, mappedbno
, whichfork
,
2646 /* mapping a hole is not an error, but we don't continue */
2652 error
= xfs_trans_read_buf_map(dp
->i_mount
, trans
,
2653 dp
->i_mount
->m_ddev_targp
,
2654 mapp
, nmap
, 0, &bp
, ops
);
2658 if (whichfork
== XFS_ATTR_FORK
)
2659 xfs_buf_set_ref(bp
, XFS_ATTR_BTREE_REF
);
2661 xfs_buf_set_ref(bp
, XFS_DIR_BTREE_REF
);
2664 * This verification code will be moved to a CRC verification callback
2665 * function so just leave it here unchanged until then.
2668 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
2669 xfs_dir2_free_t
*free
= bp
->b_addr
;
2670 xfs_da_blkinfo_t
*info
= bp
->b_addr
;
2672 struct xfs_mount
*mp
= dp
->i_mount
;
2674 magic
= be16_to_cpu(info
->magic
);
2675 magic1
= be32_to_cpu(hdr
->magic
);
2677 XFS_TEST_ERROR((magic
!= XFS_DA_NODE_MAGIC
) &&
2678 (magic
!= XFS_DA3_NODE_MAGIC
) &&
2679 (magic
!= XFS_ATTR_LEAF_MAGIC
) &&
2680 (magic
!= XFS_ATTR3_LEAF_MAGIC
) &&
2681 (magic
!= XFS_DIR2_LEAF1_MAGIC
) &&
2682 (magic
!= XFS_DIR3_LEAF1_MAGIC
) &&
2683 (magic
!= XFS_DIR2_LEAFN_MAGIC
) &&
2684 (magic
!= XFS_DIR3_LEAFN_MAGIC
) &&
2685 (magic1
!= XFS_DIR2_BLOCK_MAGIC
) &&
2686 (magic1
!= XFS_DIR3_BLOCK_MAGIC
) &&
2687 (magic1
!= XFS_DIR2_DATA_MAGIC
) &&
2688 (magic1
!= XFS_DIR3_DATA_MAGIC
) &&
2690 cpu_to_be32(XFS_DIR2_FREE_MAGIC
)) &&
2692 cpu_to_be32(XFS_DIR3_FREE_MAGIC
)),
2693 mp
, XFS_ERRTAG_DA_READ_BUF
,
2694 XFS_RANDOM_DA_READ_BUF
))) {
2695 trace_xfs_da_btree_corrupt(bp
, _RET_IP_
);
2696 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2697 XFS_ERRLEVEL_LOW
, mp
, info
);
2698 error
= XFS_ERROR(EFSCORRUPTED
);
2699 xfs_trans_brelse(trans
, bp
);
2712 * Readahead the dir/attr block.
2716 struct xfs_trans
*trans
,
2717 struct xfs_inode
*dp
,
2719 xfs_daddr_t mappedbno
,
2721 const struct xfs_buf_ops
*ops
)
2723 struct xfs_buf_map map
;
2724 struct xfs_buf_map
*mapp
;
2730 error
= xfs_dabuf_map(trans
, dp
, bno
, mappedbno
, whichfork
,
2733 /* mapping a hole is not an error, but we don't continue */
2739 mappedbno
= mapp
[0].bm_bn
;
2740 xfs_buf_readahead_map(dp
->i_mount
->m_ddev_targp
, mapp
, nmap
, ops
);