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
19 #include "libxfs_priv.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_inode.h"
32 #include "xfs_trans.h"
33 #include "xfs_alloc.h"
35 #include "xfs_attr_leaf.h"
36 #include "xfs_trace.h"
37 #include "xfs_cksum.h"
42 * Routines to implement directories as Btrees of hashed names.
45 /*========================================================================
46 * Function prototypes for the kernel.
47 *========================================================================*/
50 * Routines used for growing the Btree.
52 STATIC
int xfs_da3_root_split(xfs_da_state_t
*state
,
53 xfs_da_state_blk_t
*existing_root
,
54 xfs_da_state_blk_t
*new_child
);
55 STATIC
int xfs_da3_node_split(xfs_da_state_t
*state
,
56 xfs_da_state_blk_t
*existing_blk
,
57 xfs_da_state_blk_t
*split_blk
,
58 xfs_da_state_blk_t
*blk_to_add
,
61 STATIC
void xfs_da3_node_rebalance(xfs_da_state_t
*state
,
62 xfs_da_state_blk_t
*node_blk_1
,
63 xfs_da_state_blk_t
*node_blk_2
);
64 STATIC
void xfs_da3_node_add(xfs_da_state_t
*state
,
65 xfs_da_state_blk_t
*old_node_blk
,
66 xfs_da_state_blk_t
*new_node_blk
);
69 * Routines used for shrinking the Btree.
71 STATIC
int xfs_da3_root_join(xfs_da_state_t
*state
,
72 xfs_da_state_blk_t
*root_blk
);
73 STATIC
int xfs_da3_node_toosmall(xfs_da_state_t
*state
, int *retval
);
74 STATIC
void xfs_da3_node_remove(xfs_da_state_t
*state
,
75 xfs_da_state_blk_t
*drop_blk
);
76 STATIC
void xfs_da3_node_unbalance(xfs_da_state_t
*state
,
77 xfs_da_state_blk_t
*src_node_blk
,
78 xfs_da_state_blk_t
*dst_node_blk
);
83 STATIC
int xfs_da3_blk_unlink(xfs_da_state_t
*state
,
84 xfs_da_state_blk_t
*drop_blk
,
85 xfs_da_state_blk_t
*save_blk
);
88 kmem_zone_t
*xfs_da_state_zone
; /* anchor for state struct zone */
91 * Allocate a dir-state structure.
92 * We don't put them on the stack since they're large.
95 xfs_da_state_alloc(void)
97 return kmem_zone_zalloc(xfs_da_state_zone
, KM_NOFS
);
101 * Kill the altpath contents of a da-state structure.
104 xfs_da_state_kill_altpath(xfs_da_state_t
*state
)
108 for (i
= 0; i
< state
->altpath
.active
; i
++)
109 state
->altpath
.blk
[i
].bp
= NULL
;
110 state
->altpath
.active
= 0;
114 * Free a da-state structure.
117 xfs_da_state_free(xfs_da_state_t
*state
)
119 xfs_da_state_kill_altpath(state
);
121 memset((char *)state
, 0, sizeof(*state
));
123 kmem_zone_free(xfs_da_state_zone
, state
);
126 static xfs_failaddr_t
130 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
131 struct xfs_da_intnode
*hdr
= bp
->b_addr
;
132 struct xfs_da3_icnode_hdr ichdr
;
133 const struct xfs_dir_ops
*ops
;
135 ops
= xfs_dir_get_ops(mp
, NULL
);
137 ops
->node_hdr_from_disk(&ichdr
, hdr
);
139 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
140 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
142 if (ichdr
.magic
!= XFS_DA3_NODE_MAGIC
)
143 return __this_address
;
145 if (!uuid_equal(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_meta_uuid
))
146 return __this_address
;
147 if (be64_to_cpu(hdr3
->info
.blkno
) != bp
->b_bn
)
148 return __this_address
;
149 if (!xfs_log_check_lsn(mp
, be64_to_cpu(hdr3
->info
.lsn
)))
150 return __this_address
;
152 if (ichdr
.magic
!= XFS_DA_NODE_MAGIC
)
153 return __this_address
;
155 if (ichdr
.level
== 0)
156 return __this_address
;
157 if (ichdr
.level
> XFS_DA_NODE_MAXDEPTH
)
158 return __this_address
;
159 if (ichdr
.count
== 0)
160 return __this_address
;
163 * we don't know if the node is for and attribute or directory tree,
164 * so only fail if the count is outside both bounds
166 if (ichdr
.count
> mp
->m_dir_geo
->node_ents
&&
167 ichdr
.count
> mp
->m_attr_geo
->node_ents
)
168 return __this_address
;
170 /* XXX: hash order check? */
176 xfs_da3_node_write_verify(
179 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
180 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
181 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
184 fa
= xfs_da3_node_verify(bp
);
186 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
190 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
194 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
196 xfs_buf_update_cksum(bp
, XFS_DA3_NODE_CRC_OFF
);
200 * leaf/node format detection on trees is sketchy, so a node read can be done on
201 * leaf level blocks when detection identifies the tree as a node format tree
202 * incorrectly. In this case, we need to swap the verifier to match the correct
203 * format of the block being read.
206 xfs_da3_node_read_verify(
209 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
212 switch (be16_to_cpu(info
->magic
)) {
213 case XFS_DA3_NODE_MAGIC
:
214 if (!xfs_buf_verify_cksum(bp
, XFS_DA3_NODE_CRC_OFF
)) {
215 xfs_verifier_error(bp
, -EFSBADCRC
,
220 case XFS_DA_NODE_MAGIC
:
221 fa
= xfs_da3_node_verify(bp
);
223 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
225 case XFS_ATTR_LEAF_MAGIC
:
226 case XFS_ATTR3_LEAF_MAGIC
:
227 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
228 bp
->b_ops
->verify_read(bp
);
230 case XFS_DIR2_LEAFN_MAGIC
:
231 case XFS_DIR3_LEAFN_MAGIC
:
232 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
233 bp
->b_ops
->verify_read(bp
);
236 xfs_verifier_error(bp
, -EFSCORRUPTED
, __this_address
);
241 /* Verify the structure of a da3 block. */
242 static xfs_failaddr_t
243 xfs_da3_node_verify_struct(
246 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
248 switch (be16_to_cpu(info
->magic
)) {
249 case XFS_DA3_NODE_MAGIC
:
250 case XFS_DA_NODE_MAGIC
:
251 return xfs_da3_node_verify(bp
);
252 case XFS_ATTR_LEAF_MAGIC
:
253 case XFS_ATTR3_LEAF_MAGIC
:
254 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
255 return bp
->b_ops
->verify_struct(bp
);
256 case XFS_DIR2_LEAFN_MAGIC
:
257 case XFS_DIR3_LEAFN_MAGIC
:
258 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
259 return bp
->b_ops
->verify_struct(bp
);
261 return __this_address
;
265 const struct xfs_buf_ops xfs_da3_node_buf_ops
= {
266 .name
= "xfs_da3_node",
267 .verify_read
= xfs_da3_node_read_verify
,
268 .verify_write
= xfs_da3_node_write_verify
,
269 .verify_struct
= xfs_da3_node_verify_struct
,
274 struct xfs_trans
*tp
,
275 struct xfs_inode
*dp
,
277 xfs_daddr_t mappedbno
,
278 struct xfs_buf
**bpp
,
283 err
= xfs_da_read_buf(tp
, dp
, bno
, mappedbno
, bpp
,
284 which_fork
, &xfs_da3_node_buf_ops
);
285 if (!err
&& tp
&& *bpp
) {
286 struct xfs_da_blkinfo
*info
= (*bpp
)->b_addr
;
289 switch (be16_to_cpu(info
->magic
)) {
290 case XFS_DA_NODE_MAGIC
:
291 case XFS_DA3_NODE_MAGIC
:
292 type
= XFS_BLFT_DA_NODE_BUF
;
294 case XFS_ATTR_LEAF_MAGIC
:
295 case XFS_ATTR3_LEAF_MAGIC
:
296 type
= XFS_BLFT_ATTR_LEAF_BUF
;
298 case XFS_DIR2_LEAFN_MAGIC
:
299 case XFS_DIR3_LEAFN_MAGIC
:
300 type
= XFS_BLFT_DIR_LEAFN_BUF
;
303 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
,
304 tp
->t_mountp
, info
, sizeof(*info
));
305 xfs_trans_brelse(tp
, *bpp
);
307 return -EFSCORRUPTED
;
309 xfs_trans_buf_set_type(tp
, *bpp
, type
);
314 /*========================================================================
315 * Routines used for growing the Btree.
316 *========================================================================*/
319 * Create the initial contents of an intermediate node.
323 struct xfs_da_args
*args
,
326 struct xfs_buf
**bpp
,
329 struct xfs_da_intnode
*node
;
330 struct xfs_trans
*tp
= args
->trans
;
331 struct xfs_mount
*mp
= tp
->t_mountp
;
332 struct xfs_da3_icnode_hdr ichdr
= {0};
335 struct xfs_inode
*dp
= args
->dp
;
337 trace_xfs_da_node_create(args
);
338 ASSERT(level
<= XFS_DA_NODE_MAXDEPTH
);
340 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, whichfork
);
343 bp
->b_ops
= &xfs_da3_node_buf_ops
;
344 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
347 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
348 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
350 memset(hdr3
, 0, sizeof(struct xfs_da3_node_hdr
));
351 ichdr
.magic
= XFS_DA3_NODE_MAGIC
;
352 hdr3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
353 hdr3
->info
.owner
= cpu_to_be64(args
->dp
->i_ino
);
354 uuid_copy(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_meta_uuid
);
356 ichdr
.magic
= XFS_DA_NODE_MAGIC
;
360 dp
->d_ops
->node_hdr_to_disk(node
, &ichdr
);
361 xfs_trans_log_buf(tp
, bp
,
362 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
369 * Split a leaf node, rebalance, then possibly split
370 * intermediate nodes, rebalance, etc.
374 struct xfs_da_state
*state
)
376 struct xfs_da_state_blk
*oldblk
;
377 struct xfs_da_state_blk
*newblk
;
378 struct xfs_da_state_blk
*addblk
;
379 struct xfs_da_intnode
*node
;
385 trace_xfs_da_split(state
->args
);
388 * Walk back up the tree splitting/inserting/adjusting as necessary.
389 * If we need to insert and there isn't room, split the node, then
390 * decide which fragment to insert the new block from below into.
391 * Note that we may split the root this way, but we need more fixup.
393 max
= state
->path
.active
- 1;
394 ASSERT((max
>= 0) && (max
< XFS_DA_NODE_MAXDEPTH
));
395 ASSERT(state
->path
.blk
[max
].magic
== XFS_ATTR_LEAF_MAGIC
||
396 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
398 addblk
= &state
->path
.blk
[max
]; /* initial dummy value */
399 for (i
= max
; (i
>= 0) && addblk
; state
->path
.active
--, i
--) {
400 oldblk
= &state
->path
.blk
[i
];
401 newblk
= &state
->altpath
.blk
[i
];
404 * If a leaf node then
405 * Allocate a new leaf node, then rebalance across them.
406 * else if an intermediate node then
407 * We split on the last layer, must we split the node?
409 switch (oldblk
->magic
) {
410 case XFS_ATTR_LEAF_MAGIC
:
411 error
= xfs_attr3_leaf_split(state
, oldblk
, newblk
);
412 if ((error
!= 0) && (error
!= -ENOSPC
)) {
413 return error
; /* GROT: attr is inconsistent */
420 * Entry wouldn't fit, split the leaf again. The new
421 * extrablk will be consumed by xfs_da3_node_split if
424 state
->extravalid
= 1;
426 state
->extraafter
= 0; /* before newblk */
427 trace_xfs_attr_leaf_split_before(state
->args
);
428 error
= xfs_attr3_leaf_split(state
, oldblk
,
431 state
->extraafter
= 1; /* after newblk */
432 trace_xfs_attr_leaf_split_after(state
->args
);
433 error
= xfs_attr3_leaf_split(state
, newblk
,
437 return error
; /* GROT: attr inconsistent */
440 case XFS_DIR2_LEAFN_MAGIC
:
441 error
= xfs_dir2_leafn_split(state
, oldblk
, newblk
);
446 case XFS_DA_NODE_MAGIC
:
447 error
= xfs_da3_node_split(state
, oldblk
, newblk
, addblk
,
451 return error
; /* GROT: dir is inconsistent */
453 * Record the newly split block for the next time thru?
463 * Update the btree to show the new hashval for this child.
465 xfs_da3_fixhashpath(state
, &state
->path
);
471 * xfs_da3_node_split() should have consumed any extra blocks we added
472 * during a double leaf split in the attr fork. This is guaranteed as
473 * we can't be here if the attr fork only has a single leaf block.
475 ASSERT(state
->extravalid
== 0 ||
476 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
479 * Split the root node.
481 ASSERT(state
->path
.active
== 0);
482 oldblk
= &state
->path
.blk
[0];
483 error
= xfs_da3_root_split(state
, oldblk
, addblk
);
486 return error
; /* GROT: dir is inconsistent */
490 * Update pointers to the node which used to be block 0 and just got
491 * bumped because of the addition of a new root node. Note that the
492 * original block 0 could be at any position in the list of blocks in
495 * Note: the magic numbers and sibling pointers are in the same physical
496 * place for both v2 and v3 headers (by design). Hence it doesn't matter
497 * which version of the xfs_da_intnode structure we use here as the
498 * result will be the same using either structure.
500 node
= oldblk
->bp
->b_addr
;
501 if (node
->hdr
.info
.forw
) {
502 ASSERT(be32_to_cpu(node
->hdr
.info
.forw
) == addblk
->blkno
);
503 node
= addblk
->bp
->b_addr
;
504 node
->hdr
.info
.back
= cpu_to_be32(oldblk
->blkno
);
505 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
506 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
507 sizeof(node
->hdr
.info
)));
509 node
= oldblk
->bp
->b_addr
;
510 if (node
->hdr
.info
.back
) {
511 ASSERT(be32_to_cpu(node
->hdr
.info
.back
) == addblk
->blkno
);
512 node
= addblk
->bp
->b_addr
;
513 node
->hdr
.info
.forw
= cpu_to_be32(oldblk
->blkno
);
514 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
515 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
516 sizeof(node
->hdr
.info
)));
523 * Split the root. We have to create a new root and point to the two
524 * parts (the split old root) that we just created. Copy block zero to
525 * the EOF, extending the inode in process.
527 STATIC
int /* error */
529 struct xfs_da_state
*state
,
530 struct xfs_da_state_blk
*blk1
,
531 struct xfs_da_state_blk
*blk2
)
533 struct xfs_da_intnode
*node
;
534 struct xfs_da_intnode
*oldroot
;
535 struct xfs_da_node_entry
*btree
;
536 struct xfs_da3_icnode_hdr nodehdr
;
537 struct xfs_da_args
*args
;
539 struct xfs_inode
*dp
;
540 struct xfs_trans
*tp
;
541 struct xfs_dir2_leaf
*leaf
;
547 trace_xfs_da_root_split(state
->args
);
550 * Copy the existing (incorrect) block from the root node position
551 * to a free space somewhere.
554 error
= xfs_da_grow_inode(args
, &blkno
);
560 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, args
->whichfork
);
564 oldroot
= blk1
->bp
->b_addr
;
565 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
566 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
)) {
567 struct xfs_da3_icnode_hdr icnodehdr
;
569 dp
->d_ops
->node_hdr_from_disk(&icnodehdr
, oldroot
);
570 btree
= dp
->d_ops
->node_tree_p(oldroot
);
571 size
= (int)((char *)&btree
[icnodehdr
.count
] - (char *)oldroot
);
572 level
= icnodehdr
.level
;
575 * we are about to copy oldroot to bp, so set up the type
576 * of bp while we know exactly what it will be.
578 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
580 struct xfs_dir3_icleaf_hdr leafhdr
;
581 struct xfs_dir2_leaf_entry
*ents
;
583 leaf
= (xfs_dir2_leaf_t
*)oldroot
;
584 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
585 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
587 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
588 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
589 size
= (int)((char *)&ents
[leafhdr
.count
] - (char *)leaf
);
593 * we are about to copy oldroot to bp, so set up the type
594 * of bp while we know exactly what it will be.
596 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAFN_BUF
);
600 * we can copy most of the information in the node from one block to
601 * another, but for CRC enabled headers we have to make sure that the
602 * block specific identifiers are kept intact. We update the buffer
605 memcpy(node
, oldroot
, size
);
606 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
607 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
608 struct xfs_da3_intnode
*node3
= (struct xfs_da3_intnode
*)node
;
610 node3
->hdr
.info
.blkno
= cpu_to_be64(bp
->b_bn
);
612 xfs_trans_log_buf(tp
, bp
, 0, size
- 1);
614 bp
->b_ops
= blk1
->bp
->b_ops
;
615 xfs_trans_buf_copy_type(bp
, blk1
->bp
);
620 * Set up the new root node.
622 error
= xfs_da3_node_create(args
,
623 (args
->whichfork
== XFS_DATA_FORK
) ? args
->geo
->leafblk
: 0,
624 level
+ 1, &bp
, args
->whichfork
);
629 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
630 btree
= dp
->d_ops
->node_tree_p(node
);
631 btree
[0].hashval
= cpu_to_be32(blk1
->hashval
);
632 btree
[0].before
= cpu_to_be32(blk1
->blkno
);
633 btree
[1].hashval
= cpu_to_be32(blk2
->hashval
);
634 btree
[1].before
= cpu_to_be32(blk2
->blkno
);
636 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
639 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
640 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
641 ASSERT(blk1
->blkno
>= args
->geo
->leafblk
&&
642 blk1
->blkno
< args
->geo
->freeblk
);
643 ASSERT(blk2
->blkno
>= args
->geo
->leafblk
&&
644 blk2
->blkno
< args
->geo
->freeblk
);
648 /* Header is already logged by xfs_da_node_create */
649 xfs_trans_log_buf(tp
, bp
,
650 XFS_DA_LOGRANGE(node
, btree
, sizeof(xfs_da_node_entry_t
) * 2));
656 * Split the node, rebalance, then add the new entry.
658 STATIC
int /* error */
660 struct xfs_da_state
*state
,
661 struct xfs_da_state_blk
*oldblk
,
662 struct xfs_da_state_blk
*newblk
,
663 struct xfs_da_state_blk
*addblk
,
667 struct xfs_da_intnode
*node
;
668 struct xfs_da3_icnode_hdr nodehdr
;
673 struct xfs_inode
*dp
= state
->args
->dp
;
675 trace_xfs_da_node_split(state
->args
);
677 node
= oldblk
->bp
->b_addr
;
678 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
681 * With V2 dirs the extra block is data or freespace.
683 useextra
= state
->extravalid
&& state
->args
->whichfork
== XFS_ATTR_FORK
;
684 newcount
= 1 + useextra
;
686 * Do we have to split the node?
688 if (nodehdr
.count
+ newcount
> state
->args
->geo
->node_ents
) {
690 * Allocate a new node, add to the doubly linked chain of
691 * nodes, then move some of our excess entries into it.
693 error
= xfs_da_grow_inode(state
->args
, &blkno
);
695 return error
; /* GROT: dir is inconsistent */
697 error
= xfs_da3_node_create(state
->args
, blkno
, treelevel
,
698 &newblk
->bp
, state
->args
->whichfork
);
700 return error
; /* GROT: dir is inconsistent */
701 newblk
->blkno
= blkno
;
702 newblk
->magic
= XFS_DA_NODE_MAGIC
;
703 xfs_da3_node_rebalance(state
, oldblk
, newblk
);
704 error
= xfs_da3_blk_link(state
, oldblk
, newblk
);
713 * Insert the new entry(s) into the correct block
714 * (updating last hashval in the process).
716 * xfs_da3_node_add() inserts BEFORE the given index,
717 * and as a result of using node_lookup_int() we always
718 * point to a valid entry (not after one), but a split
719 * operation always results in a new block whose hashvals
720 * FOLLOW the current block.
722 * If we had double-split op below us, then add the extra block too.
724 node
= oldblk
->bp
->b_addr
;
725 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
726 if (oldblk
->index
<= nodehdr
.count
) {
728 xfs_da3_node_add(state
, oldblk
, addblk
);
730 if (state
->extraafter
)
732 xfs_da3_node_add(state
, oldblk
, &state
->extrablk
);
733 state
->extravalid
= 0;
737 xfs_da3_node_add(state
, newblk
, addblk
);
739 if (state
->extraafter
)
741 xfs_da3_node_add(state
, newblk
, &state
->extrablk
);
742 state
->extravalid
= 0;
750 * Balance the btree elements between two intermediate nodes,
751 * usually one full and one empty.
753 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
756 xfs_da3_node_rebalance(
757 struct xfs_da_state
*state
,
758 struct xfs_da_state_blk
*blk1
,
759 struct xfs_da_state_blk
*blk2
)
761 struct xfs_da_intnode
*node1
;
762 struct xfs_da_intnode
*node2
;
763 struct xfs_da_intnode
*tmpnode
;
764 struct xfs_da_node_entry
*btree1
;
765 struct xfs_da_node_entry
*btree2
;
766 struct xfs_da_node_entry
*btree_s
;
767 struct xfs_da_node_entry
*btree_d
;
768 struct xfs_da3_icnode_hdr nodehdr1
;
769 struct xfs_da3_icnode_hdr nodehdr2
;
770 struct xfs_trans
*tp
;
774 struct xfs_inode
*dp
= state
->args
->dp
;
776 trace_xfs_da_node_rebalance(state
->args
);
778 node1
= blk1
->bp
->b_addr
;
779 node2
= blk2
->bp
->b_addr
;
780 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
781 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
782 btree1
= dp
->d_ops
->node_tree_p(node1
);
783 btree2
= dp
->d_ops
->node_tree_p(node2
);
786 * Figure out how many entries need to move, and in which direction.
787 * Swap the nodes around if that makes it simpler.
789 if (nodehdr1
.count
> 0 && nodehdr2
.count
> 0 &&
790 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
791 (be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
) <
792 be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
)))) {
796 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
797 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
798 btree1
= dp
->d_ops
->node_tree_p(node1
);
799 btree2
= dp
->d_ops
->node_tree_p(node2
);
803 count
= (nodehdr1
.count
- nodehdr2
.count
) / 2;
806 tp
= state
->args
->trans
;
808 * Two cases: high-to-low and low-to-high.
812 * Move elements in node2 up to make a hole.
814 tmp
= nodehdr2
.count
;
816 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
817 btree_s
= &btree2
[0];
818 btree_d
= &btree2
[count
];
819 memmove(btree_d
, btree_s
, tmp
);
823 * Move the req'd B-tree elements from high in node1 to
826 nodehdr2
.count
+= count
;
827 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
828 btree_s
= &btree1
[nodehdr1
.count
- count
];
829 btree_d
= &btree2
[0];
830 memcpy(btree_d
, btree_s
, tmp
);
831 nodehdr1
.count
-= count
;
834 * Move the req'd B-tree elements from low in node2 to
838 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
839 btree_s
= &btree2
[0];
840 btree_d
= &btree1
[nodehdr1
.count
];
841 memcpy(btree_d
, btree_s
, tmp
);
842 nodehdr1
.count
+= count
;
844 xfs_trans_log_buf(tp
, blk1
->bp
,
845 XFS_DA_LOGRANGE(node1
, btree_d
, tmp
));
848 * Move elements in node2 down to fill the hole.
850 tmp
= nodehdr2
.count
- count
;
851 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
852 btree_s
= &btree2
[count
];
853 btree_d
= &btree2
[0];
854 memmove(btree_d
, btree_s
, tmp
);
855 nodehdr2
.count
-= count
;
859 * Log header of node 1 and all current bits of node 2.
861 dp
->d_ops
->node_hdr_to_disk(node1
, &nodehdr1
);
862 xfs_trans_log_buf(tp
, blk1
->bp
,
863 XFS_DA_LOGRANGE(node1
, &node1
->hdr
, dp
->d_ops
->node_hdr_size
));
865 dp
->d_ops
->node_hdr_to_disk(node2
, &nodehdr2
);
866 xfs_trans_log_buf(tp
, blk2
->bp
,
867 XFS_DA_LOGRANGE(node2
, &node2
->hdr
,
868 dp
->d_ops
->node_hdr_size
+
869 (sizeof(btree2
[0]) * nodehdr2
.count
)));
872 * Record the last hashval from each block for upward propagation.
873 * (note: don't use the swapped node pointers)
876 node1
= blk1
->bp
->b_addr
;
877 node2
= blk2
->bp
->b_addr
;
878 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
879 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
880 btree1
= dp
->d_ops
->node_tree_p(node1
);
881 btree2
= dp
->d_ops
->node_tree_p(node2
);
883 blk1
->hashval
= be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
);
884 blk2
->hashval
= be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
);
887 * Adjust the expected index for insertion.
889 if (blk1
->index
>= nodehdr1
.count
) {
890 blk2
->index
= blk1
->index
- nodehdr1
.count
;
891 blk1
->index
= nodehdr1
.count
+ 1; /* make it invalid */
896 * Add a new entry to an intermediate node.
900 struct xfs_da_state
*state
,
901 struct xfs_da_state_blk
*oldblk
,
902 struct xfs_da_state_blk
*newblk
)
904 struct xfs_da_intnode
*node
;
905 struct xfs_da3_icnode_hdr nodehdr
;
906 struct xfs_da_node_entry
*btree
;
908 struct xfs_inode
*dp
= state
->args
->dp
;
910 trace_xfs_da_node_add(state
->args
);
912 node
= oldblk
->bp
->b_addr
;
913 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
914 btree
= dp
->d_ops
->node_tree_p(node
);
916 ASSERT(oldblk
->index
>= 0 && oldblk
->index
<= nodehdr
.count
);
917 ASSERT(newblk
->blkno
!= 0);
918 if (state
->args
->whichfork
== XFS_DATA_FORK
)
919 ASSERT(newblk
->blkno
>= state
->args
->geo
->leafblk
&&
920 newblk
->blkno
< state
->args
->geo
->freeblk
);
923 * We may need to make some room before we insert the new node.
926 if (oldblk
->index
< nodehdr
.count
) {
927 tmp
= (nodehdr
.count
- oldblk
->index
) * (uint
)sizeof(*btree
);
928 memmove(&btree
[oldblk
->index
+ 1], &btree
[oldblk
->index
], tmp
);
930 btree
[oldblk
->index
].hashval
= cpu_to_be32(newblk
->hashval
);
931 btree
[oldblk
->index
].before
= cpu_to_be32(newblk
->blkno
);
932 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
933 XFS_DA_LOGRANGE(node
, &btree
[oldblk
->index
],
934 tmp
+ sizeof(*btree
)));
937 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
938 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
939 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
942 * Copy the last hash value from the oldblk to propagate upwards.
944 oldblk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
947 /*========================================================================
948 * Routines used for shrinking the Btree.
949 *========================================================================*/
952 * Deallocate an empty leaf node, remove it from its parent,
953 * possibly deallocating that block, etc...
957 struct xfs_da_state
*state
)
959 struct xfs_da_state_blk
*drop_blk
;
960 struct xfs_da_state_blk
*save_blk
;
964 trace_xfs_da_join(state
->args
);
966 drop_blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
967 save_blk
= &state
->altpath
.blk
[ state
->path
.active
-1 ];
968 ASSERT(state
->path
.blk
[0].magic
== XFS_DA_NODE_MAGIC
);
969 ASSERT(drop_blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
970 drop_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
973 * Walk back up the tree joining/deallocating as necessary.
974 * When we stop dropping blocks, break out.
976 for ( ; state
->path
.active
>= 2; drop_blk
--, save_blk
--,
977 state
->path
.active
--) {
979 * See if we can combine the block with a neighbor.
980 * (action == 0) => no options, just leave
981 * (action == 1) => coalesce, then unlink
982 * (action == 2) => block empty, unlink it
984 switch (drop_blk
->magic
) {
985 case XFS_ATTR_LEAF_MAGIC
:
986 error
= xfs_attr3_leaf_toosmall(state
, &action
);
991 xfs_attr3_leaf_unbalance(state
, drop_blk
, save_blk
);
993 case XFS_DIR2_LEAFN_MAGIC
:
994 error
= xfs_dir2_leafn_toosmall(state
, &action
);
999 xfs_dir2_leafn_unbalance(state
, drop_blk
, save_blk
);
1001 case XFS_DA_NODE_MAGIC
:
1003 * Remove the offending node, fixup hashvals,
1004 * check for a toosmall neighbor.
1006 xfs_da3_node_remove(state
, drop_blk
);
1007 xfs_da3_fixhashpath(state
, &state
->path
);
1008 error
= xfs_da3_node_toosmall(state
, &action
);
1013 xfs_da3_node_unbalance(state
, drop_blk
, save_blk
);
1016 xfs_da3_fixhashpath(state
, &state
->altpath
);
1017 error
= xfs_da3_blk_unlink(state
, drop_blk
, save_blk
);
1018 xfs_da_state_kill_altpath(state
);
1021 error
= xfs_da_shrink_inode(state
->args
, drop_blk
->blkno
,
1023 drop_blk
->bp
= NULL
;
1028 * We joined all the way to the top. If it turns out that
1029 * we only have one entry in the root, make the child block
1032 xfs_da3_node_remove(state
, drop_blk
);
1033 xfs_da3_fixhashpath(state
, &state
->path
);
1034 error
= xfs_da3_root_join(state
, &state
->path
.blk
[0]);
1040 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo
*blkinfo
, __u16 level
)
1042 __be16 magic
= blkinfo
->magic
;
1045 ASSERT(magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1046 magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1047 magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1048 magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1050 ASSERT(magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1051 magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
));
1053 ASSERT(!blkinfo
->forw
);
1054 ASSERT(!blkinfo
->back
);
1057 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1061 * We have only one entry in the root. Copy the only remaining child of
1062 * the old root to block 0 as the new root node.
1066 struct xfs_da_state
*state
,
1067 struct xfs_da_state_blk
*root_blk
)
1069 struct xfs_da_intnode
*oldroot
;
1070 struct xfs_da_args
*args
;
1073 struct xfs_da3_icnode_hdr oldroothdr
;
1074 struct xfs_da_node_entry
*btree
;
1076 struct xfs_inode
*dp
= state
->args
->dp
;
1078 trace_xfs_da_root_join(state
->args
);
1080 ASSERT(root_blk
->magic
== XFS_DA_NODE_MAGIC
);
1083 oldroot
= root_blk
->bp
->b_addr
;
1084 dp
->d_ops
->node_hdr_from_disk(&oldroothdr
, oldroot
);
1085 ASSERT(oldroothdr
.forw
== 0);
1086 ASSERT(oldroothdr
.back
== 0);
1089 * If the root has more than one child, then don't do anything.
1091 if (oldroothdr
.count
> 1)
1095 * Read in the (only) child block, then copy those bytes into
1096 * the root block's buffer and free the original child block.
1098 btree
= dp
->d_ops
->node_tree_p(oldroot
);
1099 child
= be32_to_cpu(btree
[0].before
);
1101 error
= xfs_da3_node_read(args
->trans
, dp
, child
, -1, &bp
,
1105 xfs_da_blkinfo_onlychild_validate(bp
->b_addr
, oldroothdr
.level
);
1108 * This could be copying a leaf back into the root block in the case of
1109 * there only being a single leaf block left in the tree. Hence we have
1110 * to update the b_ops pointer as well to match the buffer type change
1111 * that could occur. For dir3 blocks we also need to update the block
1112 * number in the buffer header.
1114 memcpy(root_blk
->bp
->b_addr
, bp
->b_addr
, args
->geo
->blksize
);
1115 root_blk
->bp
->b_ops
= bp
->b_ops
;
1116 xfs_trans_buf_copy_type(root_blk
->bp
, bp
);
1117 if (oldroothdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1118 struct xfs_da3_blkinfo
*da3
= root_blk
->bp
->b_addr
;
1119 da3
->blkno
= cpu_to_be64(root_blk
->bp
->b_bn
);
1121 xfs_trans_log_buf(args
->trans
, root_blk
->bp
, 0,
1122 args
->geo
->blksize
- 1);
1123 error
= xfs_da_shrink_inode(args
, child
, bp
);
1128 * Check a node block and its neighbors to see if the block should be
1129 * collapsed into one or the other neighbor. Always keep the block
1130 * with the smaller block number.
1131 * If the current block is over 50% full, don't try to join it, return 0.
1132 * If the block is empty, fill in the state structure and return 2.
1133 * If it can be collapsed, fill in the state structure and return 1.
1134 * If nothing can be done, return 0.
1137 xfs_da3_node_toosmall(
1138 struct xfs_da_state
*state
,
1141 struct xfs_da_intnode
*node
;
1142 struct xfs_da_state_blk
*blk
;
1143 struct xfs_da_blkinfo
*info
;
1146 struct xfs_da3_icnode_hdr nodehdr
;
1152 struct xfs_inode
*dp
= state
->args
->dp
;
1154 trace_xfs_da_node_toosmall(state
->args
);
1157 * Check for the degenerate case of the block being over 50% full.
1158 * If so, it's not worth even looking to see if we might be able
1159 * to coalesce with a sibling.
1161 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1162 info
= blk
->bp
->b_addr
;
1163 node
= (xfs_da_intnode_t
*)info
;
1164 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1165 if (nodehdr
.count
> (state
->args
->geo
->node_ents
>> 1)) {
1166 *action
= 0; /* blk over 50%, don't try to join */
1167 return 0; /* blk over 50%, don't try to join */
1171 * Check for the degenerate case of the block being empty.
1172 * If the block is empty, we'll simply delete it, no need to
1173 * coalesce it with a sibling block. We choose (arbitrarily)
1174 * to merge with the forward block unless it is NULL.
1176 if (nodehdr
.count
== 0) {
1178 * Make altpath point to the block we want to keep and
1179 * path point to the block we want to drop (this one).
1181 forward
= (info
->forw
!= 0);
1182 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1183 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1196 * Examine each sibling block to see if we can coalesce with
1197 * at least 25% free space to spare. We need to figure out
1198 * whether to merge with the forward or the backward block.
1199 * We prefer coalescing with the lower numbered sibling so as
1200 * to shrink a directory over time.
1202 count
= state
->args
->geo
->node_ents
;
1203 count
-= state
->args
->geo
->node_ents
>> 2;
1204 count
-= nodehdr
.count
;
1206 /* start with smaller blk num */
1207 forward
= nodehdr
.forw
< nodehdr
.back
;
1208 for (i
= 0; i
< 2; forward
= !forward
, i
++) {
1209 struct xfs_da3_icnode_hdr thdr
;
1211 blkno
= nodehdr
.forw
;
1213 blkno
= nodehdr
.back
;
1216 error
= xfs_da3_node_read(state
->args
->trans
, dp
,
1217 blkno
, -1, &bp
, state
->args
->whichfork
);
1222 dp
->d_ops
->node_hdr_from_disk(&thdr
, node
);
1223 xfs_trans_brelse(state
->args
->trans
, bp
);
1225 if (count
- thdr
.count
>= 0)
1226 break; /* fits with at least 25% to spare */
1234 * Make altpath point to the block we want to keep (the lower
1235 * numbered block) and path point to the block we want to drop.
1237 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1238 if (blkno
< blk
->blkno
) {
1239 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1242 error
= xfs_da3_path_shift(state
, &state
->path
, forward
,
1256 * Pick up the last hashvalue from an intermediate node.
1259 xfs_da3_node_lasthash(
1260 struct xfs_inode
*dp
,
1264 struct xfs_da_intnode
*node
;
1265 struct xfs_da_node_entry
*btree
;
1266 struct xfs_da3_icnode_hdr nodehdr
;
1269 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1271 *count
= nodehdr
.count
;
1274 btree
= dp
->d_ops
->node_tree_p(node
);
1275 return be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1279 * Walk back up the tree adjusting hash values as necessary,
1280 * when we stop making changes, return.
1283 xfs_da3_fixhashpath(
1284 struct xfs_da_state
*state
,
1285 struct xfs_da_state_path
*path
)
1287 struct xfs_da_state_blk
*blk
;
1288 struct xfs_da_intnode
*node
;
1289 struct xfs_da_node_entry
*btree
;
1290 xfs_dahash_t lasthash
=0;
1293 struct xfs_inode
*dp
= state
->args
->dp
;
1295 trace_xfs_da_fixhashpath(state
->args
);
1297 level
= path
->active
-1;
1298 blk
= &path
->blk
[ level
];
1299 switch (blk
->magic
) {
1300 case XFS_ATTR_LEAF_MAGIC
:
1301 lasthash
= xfs_attr_leaf_lasthash(blk
->bp
, &count
);
1305 case XFS_DIR2_LEAFN_MAGIC
:
1306 lasthash
= xfs_dir2_leaf_lasthash(dp
, blk
->bp
, &count
);
1310 case XFS_DA_NODE_MAGIC
:
1311 lasthash
= xfs_da3_node_lasthash(dp
, blk
->bp
, &count
);
1316 for (blk
--, level
--; level
>= 0; blk
--, level
--) {
1317 struct xfs_da3_icnode_hdr nodehdr
;
1319 node
= blk
->bp
->b_addr
;
1320 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1321 btree
= dp
->d_ops
->node_tree_p(node
);
1322 if (be32_to_cpu(btree
[blk
->index
].hashval
) == lasthash
)
1324 blk
->hashval
= lasthash
;
1325 btree
[blk
->index
].hashval
= cpu_to_be32(lasthash
);
1326 xfs_trans_log_buf(state
->args
->trans
, blk
->bp
,
1327 XFS_DA_LOGRANGE(node
, &btree
[blk
->index
],
1330 lasthash
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1335 * Remove an entry from an intermediate node.
1338 xfs_da3_node_remove(
1339 struct xfs_da_state
*state
,
1340 struct xfs_da_state_blk
*drop_blk
)
1342 struct xfs_da_intnode
*node
;
1343 struct xfs_da3_icnode_hdr nodehdr
;
1344 struct xfs_da_node_entry
*btree
;
1347 struct xfs_inode
*dp
= state
->args
->dp
;
1349 trace_xfs_da_node_remove(state
->args
);
1351 node
= drop_blk
->bp
->b_addr
;
1352 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1353 ASSERT(drop_blk
->index
< nodehdr
.count
);
1354 ASSERT(drop_blk
->index
>= 0);
1357 * Copy over the offending entry, or just zero it out.
1359 index
= drop_blk
->index
;
1360 btree
= dp
->d_ops
->node_tree_p(node
);
1361 if (index
< nodehdr
.count
- 1) {
1362 tmp
= nodehdr
.count
- index
- 1;
1363 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
1364 memmove(&btree
[index
], &btree
[index
+ 1], tmp
);
1365 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1366 XFS_DA_LOGRANGE(node
, &btree
[index
], tmp
));
1367 index
= nodehdr
.count
- 1;
1369 memset(&btree
[index
], 0, sizeof(xfs_da_node_entry_t
));
1370 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1371 XFS_DA_LOGRANGE(node
, &btree
[index
], sizeof(btree
[index
])));
1373 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
1374 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1375 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
1378 * Copy the last hash value from the block to propagate upwards.
1380 drop_blk
->hashval
= be32_to_cpu(btree
[index
- 1].hashval
);
1384 * Unbalance the elements between two intermediate nodes,
1385 * move all Btree elements from one node into another.
1388 xfs_da3_node_unbalance(
1389 struct xfs_da_state
*state
,
1390 struct xfs_da_state_blk
*drop_blk
,
1391 struct xfs_da_state_blk
*save_blk
)
1393 struct xfs_da_intnode
*drop_node
;
1394 struct xfs_da_intnode
*save_node
;
1395 struct xfs_da_node_entry
*drop_btree
;
1396 struct xfs_da_node_entry
*save_btree
;
1397 struct xfs_da3_icnode_hdr drop_hdr
;
1398 struct xfs_da3_icnode_hdr save_hdr
;
1399 struct xfs_trans
*tp
;
1402 struct xfs_inode
*dp
= state
->args
->dp
;
1404 trace_xfs_da_node_unbalance(state
->args
);
1406 drop_node
= drop_blk
->bp
->b_addr
;
1407 save_node
= save_blk
->bp
->b_addr
;
1408 dp
->d_ops
->node_hdr_from_disk(&drop_hdr
, drop_node
);
1409 dp
->d_ops
->node_hdr_from_disk(&save_hdr
, save_node
);
1410 drop_btree
= dp
->d_ops
->node_tree_p(drop_node
);
1411 save_btree
= dp
->d_ops
->node_tree_p(save_node
);
1412 tp
= state
->args
->trans
;
1415 * If the dying block has lower hashvals, then move all the
1416 * elements in the remaining block up to make a hole.
1418 if ((be32_to_cpu(drop_btree
[0].hashval
) <
1419 be32_to_cpu(save_btree
[0].hashval
)) ||
1420 (be32_to_cpu(drop_btree
[drop_hdr
.count
- 1].hashval
) <
1421 be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
))) {
1422 /* XXX: check this - is memmove dst correct? */
1423 tmp
= save_hdr
.count
* sizeof(xfs_da_node_entry_t
);
1424 memmove(&save_btree
[drop_hdr
.count
], &save_btree
[0], tmp
);
1427 xfs_trans_log_buf(tp
, save_blk
->bp
,
1428 XFS_DA_LOGRANGE(save_node
, &save_btree
[0],
1429 (save_hdr
.count
+ drop_hdr
.count
) *
1430 sizeof(xfs_da_node_entry_t
)));
1432 sindex
= save_hdr
.count
;
1433 xfs_trans_log_buf(tp
, save_blk
->bp
,
1434 XFS_DA_LOGRANGE(save_node
, &save_btree
[sindex
],
1435 drop_hdr
.count
* sizeof(xfs_da_node_entry_t
)));
1439 * Move all the B-tree elements from drop_blk to save_blk.
1441 tmp
= drop_hdr
.count
* (uint
)sizeof(xfs_da_node_entry_t
);
1442 memcpy(&save_btree
[sindex
], &drop_btree
[0], tmp
);
1443 save_hdr
.count
+= drop_hdr
.count
;
1445 dp
->d_ops
->node_hdr_to_disk(save_node
, &save_hdr
);
1446 xfs_trans_log_buf(tp
, save_blk
->bp
,
1447 XFS_DA_LOGRANGE(save_node
, &save_node
->hdr
,
1448 dp
->d_ops
->node_hdr_size
));
1451 * Save the last hashval in the remaining block for upward propagation.
1453 save_blk
->hashval
= be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
);
1456 /*========================================================================
1457 * Routines used for finding things in the Btree.
1458 *========================================================================*/
1461 * Walk down the Btree looking for a particular filename, filling
1462 * in the state structure as we go.
1464 * We will set the state structure to point to each of the elements
1465 * in each of the nodes where either the hashval is or should be.
1467 * We support duplicate hashval's so for each entry in the current
1468 * node that could contain the desired hashval, descend. This is a
1469 * pruned depth-first tree search.
1472 xfs_da3_node_lookup_int(
1473 struct xfs_da_state
*state
,
1476 struct xfs_da_state_blk
*blk
;
1477 struct xfs_da_blkinfo
*curr
;
1478 struct xfs_da_intnode
*node
;
1479 struct xfs_da_node_entry
*btree
;
1480 struct xfs_da3_icnode_hdr nodehdr
;
1481 struct xfs_da_args
*args
;
1483 xfs_dahash_t hashval
;
1484 xfs_dahash_t btreehashval
;
1490 unsigned int expected_level
= 0;
1491 struct xfs_inode
*dp
= state
->args
->dp
;
1496 * Descend thru the B-tree searching each level for the right
1497 * node to use, until the right hashval is found.
1499 blkno
= args
->geo
->leafblk
;
1500 for (blk
= &state
->path
.blk
[0], state
->path
.active
= 1;
1501 state
->path
.active
<= XFS_DA_NODE_MAXDEPTH
;
1502 blk
++, state
->path
.active
++) {
1504 * Read the next node down in the tree.
1507 error
= xfs_da3_node_read(args
->trans
, args
->dp
, blkno
,
1508 -1, &blk
->bp
, args
->whichfork
);
1511 state
->path
.active
--;
1514 curr
= blk
->bp
->b_addr
;
1515 blk
->magic
= be16_to_cpu(curr
->magic
);
1517 if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
1518 blk
->magic
== XFS_ATTR3_LEAF_MAGIC
) {
1519 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1520 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1524 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1525 blk
->magic
== XFS_DIR3_LEAFN_MAGIC
) {
1526 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1527 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1532 blk
->magic
= XFS_DA_NODE_MAGIC
;
1536 * Search an intermediate node for a match.
1538 node
= blk
->bp
->b_addr
;
1539 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1540 btree
= dp
->d_ops
->node_tree_p(node
);
1542 /* Tree taller than we can handle; bail out! */
1543 if (nodehdr
.level
>= XFS_DA_NODE_MAXDEPTH
)
1544 return -EFSCORRUPTED
;
1546 /* Check the level from the root. */
1547 if (blkno
== args
->geo
->leafblk
)
1548 expected_level
= nodehdr
.level
- 1;
1549 else if (expected_level
!= nodehdr
.level
)
1550 return -EFSCORRUPTED
;
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
);
1600 /* We can't point back to the root. */
1601 if (blkno
== args
->geo
->leafblk
)
1602 return -EFSCORRUPTED
;
1605 if (expected_level
!= 0)
1606 return -EFSCORRUPTED
;
1609 * A leaf block that ends in the hashval that we are interested in
1610 * (final hashval == search hashval) means that the next block may
1611 * contain more entries with the same hashval, shift upward to the
1612 * next leaf and keep searching.
1615 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
) {
1616 retval
= xfs_dir2_leafn_lookup_int(blk
->bp
, args
,
1617 &blk
->index
, state
);
1618 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1619 retval
= xfs_attr3_leaf_lookup_int(blk
->bp
, args
);
1620 blk
->index
= args
->index
;
1621 args
->blkno
= blk
->blkno
;
1624 return -EFSCORRUPTED
;
1626 if (((retval
== -ENOENT
) || (retval
== -ENOATTR
)) &&
1627 (blk
->hashval
== args
->hashval
)) {
1628 error
= xfs_da3_path_shift(state
, &state
->path
, 1, 1,
1634 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1635 /* path_shift() gives ENOENT */
1645 /*========================================================================
1647 *========================================================================*/
1650 * Compare two intermediate nodes for "order".
1654 struct xfs_inode
*dp
,
1655 struct xfs_buf
*node1_bp
,
1656 struct xfs_buf
*node2_bp
)
1658 struct xfs_da_intnode
*node1
;
1659 struct xfs_da_intnode
*node2
;
1660 struct xfs_da_node_entry
*btree1
;
1661 struct xfs_da_node_entry
*btree2
;
1662 struct xfs_da3_icnode_hdr node1hdr
;
1663 struct xfs_da3_icnode_hdr node2hdr
;
1665 node1
= node1_bp
->b_addr
;
1666 node2
= node2_bp
->b_addr
;
1667 dp
->d_ops
->node_hdr_from_disk(&node1hdr
, node1
);
1668 dp
->d_ops
->node_hdr_from_disk(&node2hdr
, node2
);
1669 btree1
= dp
->d_ops
->node_tree_p(node1
);
1670 btree2
= dp
->d_ops
->node_tree_p(node2
);
1672 if (node1hdr
.count
> 0 && node2hdr
.count
> 0 &&
1673 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
1674 (be32_to_cpu(btree2
[node2hdr
.count
- 1].hashval
) <
1675 be32_to_cpu(btree1
[node1hdr
.count
- 1].hashval
)))) {
1682 * Link a new block into a doubly linked list of blocks (of whatever type).
1686 struct xfs_da_state
*state
,
1687 struct xfs_da_state_blk
*old_blk
,
1688 struct xfs_da_state_blk
*new_blk
)
1690 struct xfs_da_blkinfo
*old_info
;
1691 struct xfs_da_blkinfo
*new_info
;
1692 struct xfs_da_blkinfo
*tmp_info
;
1693 struct xfs_da_args
*args
;
1697 struct xfs_inode
*dp
= state
->args
->dp
;
1700 * Set up environment.
1703 ASSERT(args
!= NULL
);
1704 old_info
= old_blk
->bp
->b_addr
;
1705 new_info
= new_blk
->bp
->b_addr
;
1706 ASSERT(old_blk
->magic
== XFS_DA_NODE_MAGIC
||
1707 old_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1708 old_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1710 switch (old_blk
->magic
) {
1711 case XFS_ATTR_LEAF_MAGIC
:
1712 before
= xfs_attr_leaf_order(old_blk
->bp
, new_blk
->bp
);
1714 case XFS_DIR2_LEAFN_MAGIC
:
1715 before
= xfs_dir2_leafn_order(dp
, old_blk
->bp
, new_blk
->bp
);
1717 case XFS_DA_NODE_MAGIC
:
1718 before
= xfs_da3_node_order(dp
, old_blk
->bp
, new_blk
->bp
);
1723 * Link blocks in appropriate order.
1727 * Link new block in before existing block.
1729 trace_xfs_da_link_before(args
);
1730 new_info
->forw
= cpu_to_be32(old_blk
->blkno
);
1731 new_info
->back
= old_info
->back
;
1732 if (old_info
->back
) {
1733 error
= xfs_da3_node_read(args
->trans
, dp
,
1734 be32_to_cpu(old_info
->back
),
1735 -1, &bp
, args
->whichfork
);
1739 tmp_info
= bp
->b_addr
;
1740 ASSERT(tmp_info
->magic
== old_info
->magic
);
1741 ASSERT(be32_to_cpu(tmp_info
->forw
) == old_blk
->blkno
);
1742 tmp_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1743 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1745 old_info
->back
= cpu_to_be32(new_blk
->blkno
);
1748 * Link new block in after existing block.
1750 trace_xfs_da_link_after(args
);
1751 new_info
->forw
= old_info
->forw
;
1752 new_info
->back
= cpu_to_be32(old_blk
->blkno
);
1753 if (old_info
->forw
) {
1754 error
= xfs_da3_node_read(args
->trans
, dp
,
1755 be32_to_cpu(old_info
->forw
),
1756 -1, &bp
, args
->whichfork
);
1760 tmp_info
= bp
->b_addr
;
1761 ASSERT(tmp_info
->magic
== old_info
->magic
);
1762 ASSERT(be32_to_cpu(tmp_info
->back
) == old_blk
->blkno
);
1763 tmp_info
->back
= cpu_to_be32(new_blk
->blkno
);
1764 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1766 old_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1769 xfs_trans_log_buf(args
->trans
, old_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1770 xfs_trans_log_buf(args
->trans
, new_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1775 * Unlink a block from a doubly linked list of blocks.
1777 STATIC
int /* error */
1779 struct xfs_da_state
*state
,
1780 struct xfs_da_state_blk
*drop_blk
,
1781 struct xfs_da_state_blk
*save_blk
)
1783 struct xfs_da_blkinfo
*drop_info
;
1784 struct xfs_da_blkinfo
*save_info
;
1785 struct xfs_da_blkinfo
*tmp_info
;
1786 struct xfs_da_args
*args
;
1791 * Set up environment.
1794 ASSERT(args
!= NULL
);
1795 save_info
= save_blk
->bp
->b_addr
;
1796 drop_info
= drop_blk
->bp
->b_addr
;
1797 ASSERT(save_blk
->magic
== XFS_DA_NODE_MAGIC
||
1798 save_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1799 save_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1800 ASSERT(save_blk
->magic
== drop_blk
->magic
);
1801 ASSERT((be32_to_cpu(save_info
->forw
) == drop_blk
->blkno
) ||
1802 (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
));
1803 ASSERT((be32_to_cpu(drop_info
->forw
) == save_blk
->blkno
) ||
1804 (be32_to_cpu(drop_info
->back
) == save_blk
->blkno
));
1807 * Unlink the leaf block from the doubly linked chain of leaves.
1809 if (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
) {
1810 trace_xfs_da_unlink_back(args
);
1811 save_info
->back
= drop_info
->back
;
1812 if (drop_info
->back
) {
1813 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1814 be32_to_cpu(drop_info
->back
),
1815 -1, &bp
, args
->whichfork
);
1819 tmp_info
= bp
->b_addr
;
1820 ASSERT(tmp_info
->magic
== save_info
->magic
);
1821 ASSERT(be32_to_cpu(tmp_info
->forw
) == drop_blk
->blkno
);
1822 tmp_info
->forw
= cpu_to_be32(save_blk
->blkno
);
1823 xfs_trans_log_buf(args
->trans
, bp
, 0,
1824 sizeof(*tmp_info
) - 1);
1827 trace_xfs_da_unlink_forward(args
);
1828 save_info
->forw
= drop_info
->forw
;
1829 if (drop_info
->forw
) {
1830 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1831 be32_to_cpu(drop_info
->forw
),
1832 -1, &bp
, args
->whichfork
);
1836 tmp_info
= bp
->b_addr
;
1837 ASSERT(tmp_info
->magic
== save_info
->magic
);
1838 ASSERT(be32_to_cpu(tmp_info
->back
) == drop_blk
->blkno
);
1839 tmp_info
->back
= cpu_to_be32(save_blk
->blkno
);
1840 xfs_trans_log_buf(args
->trans
, bp
, 0,
1841 sizeof(*tmp_info
) - 1);
1845 xfs_trans_log_buf(args
->trans
, save_blk
->bp
, 0, sizeof(*save_info
) - 1);
1850 * Move a path "forward" or "!forward" one block at the current level.
1852 * This routine will adjust a "path" to point to the next block
1853 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1854 * Btree, including updating pointers to the intermediate nodes between
1855 * the new bottom and the root.
1859 struct xfs_da_state
*state
,
1860 struct xfs_da_state_path
*path
,
1865 struct xfs_da_state_blk
*blk
;
1866 struct xfs_da_blkinfo
*info
;
1867 struct xfs_da_intnode
*node
;
1868 struct xfs_da_args
*args
;
1869 struct xfs_da_node_entry
*btree
;
1870 struct xfs_da3_icnode_hdr nodehdr
;
1872 xfs_dablk_t blkno
= 0;
1875 struct xfs_inode
*dp
= state
->args
->dp
;
1877 trace_xfs_da_path_shift(state
->args
);
1880 * Roll up the Btree looking for the first block where our
1881 * current index is not at the edge of the block. Note that
1882 * we skip the bottom layer because we want the sibling block.
1885 ASSERT(args
!= NULL
);
1886 ASSERT(path
!= NULL
);
1887 ASSERT((path
->active
> 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1888 level
= (path
->active
-1) - 1; /* skip bottom layer in path */
1889 for (blk
= &path
->blk
[level
]; level
>= 0; blk
--, level
--) {
1890 node
= blk
->bp
->b_addr
;
1891 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1892 btree
= dp
->d_ops
->node_tree_p(node
);
1894 if (forward
&& (blk
->index
< nodehdr
.count
- 1)) {
1896 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1898 } else if (!forward
&& (blk
->index
> 0)) {
1900 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1905 *result
= -ENOENT
; /* we're out of our tree */
1906 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1911 * Roll down the edge of the subtree until we reach the
1912 * same depth we were at originally.
1914 for (blk
++, level
++; level
< path
->active
; blk
++, level
++) {
1916 * Read the next child block into a local buffer.
1918 error
= xfs_da3_node_read(args
->trans
, dp
, blkno
, -1, &bp
,
1924 * Release the old block (if it's dirty, the trans doesn't
1925 * actually let go) and swap the local buffer into the path
1926 * structure. This ensures failure of the above read doesn't set
1927 * a NULL buffer in an active slot in the path.
1930 xfs_trans_brelse(args
->trans
, blk
->bp
);
1934 info
= blk
->bp
->b_addr
;
1935 ASSERT(info
->magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1936 info
->magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
1937 info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1938 info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1939 info
->magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1940 info
->magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1944 * Note: we flatten the magic number to a single type so we
1945 * don't have to compare against crc/non-crc types elsewhere.
1947 switch (be16_to_cpu(info
->magic
)) {
1948 case XFS_DA_NODE_MAGIC
:
1949 case XFS_DA3_NODE_MAGIC
:
1950 blk
->magic
= XFS_DA_NODE_MAGIC
;
1951 node
= (xfs_da_intnode_t
*)info
;
1952 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1953 btree
= dp
->d_ops
->node_tree_p(node
);
1954 blk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1958 blk
->index
= nodehdr
.count
- 1;
1959 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1961 case XFS_ATTR_LEAF_MAGIC
:
1962 case XFS_ATTR3_LEAF_MAGIC
:
1963 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1964 ASSERT(level
== path
->active
-1);
1966 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1968 case XFS_DIR2_LEAFN_MAGIC
:
1969 case XFS_DIR3_LEAFN_MAGIC
:
1970 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1971 ASSERT(level
== path
->active
-1);
1973 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1986 /*========================================================================
1988 *========================================================================*/
1991 * Implement a simple hash on a character string.
1992 * Rotate the hash value by 7 bits, then XOR each character in.
1993 * This is implemented with some source-level loop unrolling.
1996 xfs_da_hashname(const uint8_t *name
, int namelen
)
2001 * Do four characters at a time as long as we can.
2003 for (hash
= 0; namelen
>= 4; namelen
-= 4, name
+= 4)
2004 hash
= (name
[0] << 21) ^ (name
[1] << 14) ^ (name
[2] << 7) ^
2005 (name
[3] << 0) ^ rol32(hash
, 7 * 4);
2008 * Now do the rest of the characters.
2012 return (name
[0] << 14) ^ (name
[1] << 7) ^ (name
[2] << 0) ^
2015 return (name
[0] << 7) ^ (name
[1] << 0) ^ rol32(hash
, 7 * 2);
2017 return (name
[0] << 0) ^ rol32(hash
, 7 * 1);
2018 default: /* case 0: */
2025 struct xfs_da_args
*args
,
2026 const unsigned char *name
,
2029 return (args
->namelen
== len
&& memcmp(args
->name
, name
, len
) == 0) ?
2030 XFS_CMP_EXACT
: XFS_CMP_DIFFERENT
;
2034 xfs_default_hashname(
2035 struct xfs_name
*name
)
2037 return xfs_da_hashname(name
->name
, name
->len
);
2040 const struct xfs_nameops xfs_default_nameops
= {
2041 .hashname
= xfs_default_hashname
,
2042 .compname
= xfs_da_compname
2046 xfs_da_grow_inode_int(
2047 struct xfs_da_args
*args
,
2051 struct xfs_trans
*tp
= args
->trans
;
2052 struct xfs_inode
*dp
= args
->dp
;
2053 int w
= args
->whichfork
;
2054 xfs_rfsblock_t nblks
= dp
->i_d
.di_nblocks
;
2055 struct xfs_bmbt_irec map
, *mapp
;
2056 int nmap
, error
, got
, i
, mapi
;
2059 * Find a spot in the file space to put the new block.
2061 error
= xfs_bmap_first_unused(tp
, dp
, count
, bno
, w
);
2066 * Try mapping it in one filesystem block.
2069 ASSERT(args
->firstblock
!= NULL
);
2070 error
= xfs_bmapi_write(tp
, dp
, *bno
, count
,
2071 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
|XFS_BMAPI_CONTIG
,
2072 args
->firstblock
, args
->total
, &map
, &nmap
,
2081 } else if (nmap
== 0 && count
> 1) {
2086 * If we didn't get it and the block might work if fragmented,
2087 * try without the CONTIG flag. Loop until we get it all.
2089 mapp
= kmem_alloc(sizeof(*mapp
) * count
, KM_SLEEP
);
2090 for (b
= *bno
, mapi
= 0; b
< *bno
+ count
; ) {
2091 nmap
= MIN(XFS_BMAP_MAX_NMAP
, count
);
2092 c
= (int)(*bno
+ count
- b
);
2093 error
= xfs_bmapi_write(tp
, dp
, b
, c
,
2094 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
,
2095 args
->firstblock
, args
->total
,
2096 &mapp
[mapi
], &nmap
, args
->dfops
);
2102 b
= mapp
[mapi
- 1].br_startoff
+
2103 mapp
[mapi
- 1].br_blockcount
;
2111 * Count the blocks we got, make sure it matches the total.
2113 for (i
= 0, got
= 0; i
< mapi
; i
++)
2114 got
+= mapp
[i
].br_blockcount
;
2115 if (got
!= count
|| mapp
[0].br_startoff
!= *bno
||
2116 mapp
[mapi
- 1].br_startoff
+ mapp
[mapi
- 1].br_blockcount
!=
2122 /* account for newly allocated blocks in reserved blocks total */
2123 args
->total
-= dp
->i_d
.di_nblocks
- nblks
;
2132 * Add a block to the btree ahead of the file.
2133 * Return the new block number to the caller.
2137 struct xfs_da_args
*args
,
2138 xfs_dablk_t
*new_blkno
)
2143 trace_xfs_da_grow_inode(args
);
2145 bno
= args
->geo
->leafblk
;
2146 error
= xfs_da_grow_inode_int(args
, &bno
, args
->geo
->fsbcount
);
2148 *new_blkno
= (xfs_dablk_t
)bno
;
2153 * Ick. We need to always be able to remove a btree block, even
2154 * if there's no space reservation because the filesystem is full.
2155 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2156 * It swaps the target block with the last block in the file. The
2157 * last block in the file can always be removed since it can't cause
2158 * a bmap btree split to do that.
2161 xfs_da3_swap_lastblock(
2162 struct xfs_da_args
*args
,
2163 xfs_dablk_t
*dead_blknop
,
2164 struct xfs_buf
**dead_bufp
)
2166 struct xfs_da_blkinfo
*dead_info
;
2167 struct xfs_da_blkinfo
*sib_info
;
2168 struct xfs_da_intnode
*par_node
;
2169 struct xfs_da_intnode
*dead_node
;
2170 struct xfs_dir2_leaf
*dead_leaf2
;
2171 struct xfs_da_node_entry
*btree
;
2172 struct xfs_da3_icnode_hdr par_hdr
;
2173 struct xfs_inode
*dp
;
2174 struct xfs_trans
*tp
;
2175 struct xfs_mount
*mp
;
2176 struct xfs_buf
*dead_buf
;
2177 struct xfs_buf
*last_buf
;
2178 struct xfs_buf
*sib_buf
;
2179 struct xfs_buf
*par_buf
;
2180 xfs_dahash_t dead_hash
;
2181 xfs_fileoff_t lastoff
;
2182 xfs_dablk_t dead_blkno
;
2183 xfs_dablk_t last_blkno
;
2184 xfs_dablk_t sib_blkno
;
2185 xfs_dablk_t par_blkno
;
2192 trace_xfs_da_swap_lastblock(args
);
2194 dead_buf
= *dead_bufp
;
2195 dead_blkno
= *dead_blknop
;
2198 w
= args
->whichfork
;
2199 ASSERT(w
== XFS_DATA_FORK
);
2201 lastoff
= args
->geo
->freeblk
;
2202 error
= xfs_bmap_last_before(tp
, dp
, &lastoff
, w
);
2205 if (unlikely(lastoff
== 0)) {
2206 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW
,
2208 return -EFSCORRUPTED
;
2211 * Read the last block in the btree space.
2213 last_blkno
= (xfs_dablk_t
)lastoff
- args
->geo
->fsbcount
;
2214 error
= xfs_da3_node_read(tp
, dp
, last_blkno
, -1, &last_buf
, w
);
2218 * Copy the last block into the dead buffer and log it.
2220 memcpy(dead_buf
->b_addr
, last_buf
->b_addr
, args
->geo
->blksize
);
2221 xfs_trans_log_buf(tp
, dead_buf
, 0, args
->geo
->blksize
- 1);
2222 dead_info
= dead_buf
->b_addr
;
2224 * Get values from the moved block.
2226 if (dead_info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
2227 dead_info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
2228 struct xfs_dir3_icleaf_hdr leafhdr
;
2229 struct xfs_dir2_leaf_entry
*ents
;
2231 dead_leaf2
= (xfs_dir2_leaf_t
*)dead_info
;
2232 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, dead_leaf2
);
2233 ents
= dp
->d_ops
->leaf_ents_p(dead_leaf2
);
2235 dead_hash
= be32_to_cpu(ents
[leafhdr
.count
- 1].hashval
);
2237 struct xfs_da3_icnode_hdr deadhdr
;
2239 dead_node
= (xfs_da_intnode_t
*)dead_info
;
2240 dp
->d_ops
->node_hdr_from_disk(&deadhdr
, dead_node
);
2241 btree
= dp
->d_ops
->node_tree_p(dead_node
);
2242 dead_level
= deadhdr
.level
;
2243 dead_hash
= be32_to_cpu(btree
[deadhdr
.count
- 1].hashval
);
2245 sib_buf
= par_buf
= NULL
;
2247 * If the moved block has a left sibling, fix up the pointers.
2249 if ((sib_blkno
= be32_to_cpu(dead_info
->back
))) {
2250 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2253 sib_info
= sib_buf
->b_addr
;
2255 be32_to_cpu(sib_info
->forw
) != last_blkno
||
2256 sib_info
->magic
!= dead_info
->magic
)) {
2257 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2258 XFS_ERRLEVEL_LOW
, mp
);
2259 error
= -EFSCORRUPTED
;
2262 sib_info
->forw
= cpu_to_be32(dead_blkno
);
2263 xfs_trans_log_buf(tp
, sib_buf
,
2264 XFS_DA_LOGRANGE(sib_info
, &sib_info
->forw
,
2265 sizeof(sib_info
->forw
)));
2269 * If the moved block has a right sibling, fix up the pointers.
2271 if ((sib_blkno
= be32_to_cpu(dead_info
->forw
))) {
2272 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2275 sib_info
= sib_buf
->b_addr
;
2277 be32_to_cpu(sib_info
->back
) != last_blkno
||
2278 sib_info
->magic
!= dead_info
->magic
)) {
2279 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2280 XFS_ERRLEVEL_LOW
, mp
);
2281 error
= -EFSCORRUPTED
;
2284 sib_info
->back
= cpu_to_be32(dead_blkno
);
2285 xfs_trans_log_buf(tp
, sib_buf
,
2286 XFS_DA_LOGRANGE(sib_info
, &sib_info
->back
,
2287 sizeof(sib_info
->back
)));
2290 par_blkno
= args
->geo
->leafblk
;
2293 * Walk down the tree looking for the parent of the moved block.
2296 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2299 par_node
= par_buf
->b_addr
;
2300 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2301 if (level
>= 0 && level
!= par_hdr
.level
+ 1) {
2302 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2303 XFS_ERRLEVEL_LOW
, mp
);
2304 error
= -EFSCORRUPTED
;
2307 level
= par_hdr
.level
;
2308 btree
= dp
->d_ops
->node_tree_p(par_node
);
2310 entno
< par_hdr
.count
&&
2311 be32_to_cpu(btree
[entno
].hashval
) < dead_hash
;
2314 if (entno
== par_hdr
.count
) {
2315 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2316 XFS_ERRLEVEL_LOW
, mp
);
2317 error
= -EFSCORRUPTED
;
2320 par_blkno
= be32_to_cpu(btree
[entno
].before
);
2321 if (level
== dead_level
+ 1)
2323 xfs_trans_brelse(tp
, par_buf
);
2327 * We're in the right parent block.
2328 * Look for the right entry.
2332 entno
< par_hdr
.count
&&
2333 be32_to_cpu(btree
[entno
].before
) != last_blkno
;
2336 if (entno
< par_hdr
.count
)
2338 par_blkno
= par_hdr
.forw
;
2339 xfs_trans_brelse(tp
, par_buf
);
2341 if (unlikely(par_blkno
== 0)) {
2342 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2343 XFS_ERRLEVEL_LOW
, mp
);
2344 error
= -EFSCORRUPTED
;
2347 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2350 par_node
= par_buf
->b_addr
;
2351 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2352 if (par_hdr
.level
!= level
) {
2353 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2354 XFS_ERRLEVEL_LOW
, mp
);
2355 error
= -EFSCORRUPTED
;
2358 btree
= dp
->d_ops
->node_tree_p(par_node
);
2362 * Update the parent entry pointing to the moved block.
2364 btree
[entno
].before
= cpu_to_be32(dead_blkno
);
2365 xfs_trans_log_buf(tp
, par_buf
,
2366 XFS_DA_LOGRANGE(par_node
, &btree
[entno
].before
,
2367 sizeof(btree
[entno
].before
)));
2368 *dead_blknop
= last_blkno
;
2369 *dead_bufp
= last_buf
;
2373 xfs_trans_brelse(tp
, par_buf
);
2375 xfs_trans_brelse(tp
, sib_buf
);
2376 xfs_trans_brelse(tp
, last_buf
);
2381 * Remove a btree block from a directory or attribute.
2384 xfs_da_shrink_inode(
2385 xfs_da_args_t
*args
,
2386 xfs_dablk_t dead_blkno
,
2387 struct xfs_buf
*dead_buf
)
2390 int done
, error
, w
, count
;
2393 trace_xfs_da_shrink_inode(args
);
2396 w
= args
->whichfork
;
2398 count
= args
->geo
->fsbcount
;
2401 * Remove extents. If we get ENOSPC for a dir we have to move
2402 * the last block to the place we want to kill.
2404 error
= xfs_bunmapi(tp
, dp
, dead_blkno
, count
,
2405 xfs_bmapi_aflag(w
), 0, args
->firstblock
,
2406 args
->dfops
, &done
);
2407 if (error
== -ENOSPC
) {
2408 if (w
!= XFS_DATA_FORK
)
2410 error
= xfs_da3_swap_lastblock(args
, &dead_blkno
,
2418 xfs_trans_binval(tp
, dead_buf
);
2423 * See if the mapping(s) for this btree block are valid, i.e.
2424 * don't contain holes, are logically contiguous, and cover the whole range.
2427 xfs_da_map_covers_blocks(
2429 xfs_bmbt_irec_t
*mapp
,
2436 for (i
= 0, off
= bno
; i
< nmap
; i
++) {
2437 if (mapp
[i
].br_startblock
== HOLESTARTBLOCK
||
2438 mapp
[i
].br_startblock
== DELAYSTARTBLOCK
) {
2441 if (off
!= mapp
[i
].br_startoff
) {
2444 off
+= mapp
[i
].br_blockcount
;
2446 return off
== bno
+ count
;
2450 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2452 * For the single map case, it is assumed that the caller has provided a pointer
2453 * to a valid xfs_buf_map. For the multiple map case, this function will
2454 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2455 * map pointer with the allocated map.
2458 xfs_buf_map_from_irec(
2459 struct xfs_mount
*mp
,
2460 struct xfs_buf_map
**mapp
,
2462 struct xfs_bmbt_irec
*irecs
,
2465 struct xfs_buf_map
*map
;
2468 ASSERT(*nmaps
== 1);
2469 ASSERT(nirecs
>= 1);
2472 map
= kmem_zalloc(nirecs
* sizeof(struct xfs_buf_map
),
2473 KM_SLEEP
| KM_NOFS
);
2481 for (i
= 0; i
< *nmaps
; i
++) {
2482 ASSERT(irecs
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
2483 irecs
[i
].br_startblock
!= HOLESTARTBLOCK
);
2484 map
[i
].bm_bn
= XFS_FSB_TO_DADDR(mp
, irecs
[i
].br_startblock
);
2485 map
[i
].bm_len
= XFS_FSB_TO_BB(mp
, irecs
[i
].br_blockcount
);
2491 * Map the block we are given ready for reading. There are three possible return
2493 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2494 * caller knows not to execute a subsequent read.
2495 * 0 - if we mapped the block successfully
2496 * >0 - positive error number if there was an error.
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 if (whichfork
== XFS_DATA_FORK
)
2518 nfsb
= mp
->m_dir_geo
->fsbcount
;
2520 nfsb
= mp
->m_attr_geo
->fsbcount
;
2523 * Caller doesn't have a mapping. -2 means don't complain
2524 * if we land in a hole.
2526 if (mappedbno
== -1 || mappedbno
== -2) {
2528 * Optimize the one-block case.
2531 irecs
= kmem_zalloc(sizeof(irec
) * nfsb
,
2532 KM_SLEEP
| KM_NOFS
);
2535 error
= xfs_bmapi_read(dp
, (xfs_fileoff_t
)bno
, nfsb
, irecs
,
2536 &nirecs
, xfs_bmapi_aflag(whichfork
));
2540 irecs
->br_startblock
= XFS_DADDR_TO_FSB(mp
, mappedbno
);
2541 irecs
->br_startoff
= (xfs_fileoff_t
)bno
;
2542 irecs
->br_blockcount
= nfsb
;
2543 irecs
->br_state
= 0;
2547 if (!xfs_da_map_covers_blocks(nirecs
, irecs
, bno
, nfsb
)) {
2548 error
= mappedbno
== -2 ? -1 : -EFSCORRUPTED
;
2549 if (unlikely(error
== -EFSCORRUPTED
)) {
2550 if (xfs_error_level
>= XFS_ERRLEVEL_LOW
) {
2552 xfs_alert(mp
, "%s: bno %lld dir: inode %lld",
2553 __func__
, (long long)bno
,
2554 (long long)dp
->i_ino
);
2555 for (i
= 0; i
< *nmaps
; i
++) {
2557 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2559 (long long)irecs
[i
].br_startoff
,
2560 (long long)irecs
[i
].br_startblock
,
2561 (long long)irecs
[i
].br_blockcount
,
2565 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2566 XFS_ERRLEVEL_LOW
, mp
);
2570 error
= xfs_buf_map_from_irec(mp
, map
, nmaps
, irecs
, nirecs
);
2578 * Get a buffer for the dir/attr block.
2582 struct xfs_trans
*trans
,
2583 struct xfs_inode
*dp
,
2585 xfs_daddr_t mappedbno
,
2586 struct xfs_buf
**bpp
,
2590 struct xfs_buf_map map
;
2591 struct xfs_buf_map
*mapp
;
2598 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2601 /* mapping a hole is not an error, but we don't continue */
2607 bp
= xfs_trans_get_buf_map(trans
, dp
->i_mount
->m_ddev_targp
,
2609 error
= bp
? bp
->b_error
: -EIO
;
2612 xfs_trans_brelse(trans
, bp
);
2626 * Get a buffer for the dir/attr block, fill in the contents.
2630 struct xfs_trans
*trans
,
2631 struct xfs_inode
*dp
,
2633 xfs_daddr_t mappedbno
,
2634 struct xfs_buf
**bpp
,
2636 const struct xfs_buf_ops
*ops
)
2639 struct xfs_buf_map map
;
2640 struct xfs_buf_map
*mapp
;
2647 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2650 /* mapping a hole is not an error, but we don't continue */
2656 error
= xfs_trans_read_buf_map(dp
->i_mount
, trans
,
2657 dp
->i_mount
->m_ddev_targp
,
2658 mapp
, nmap
, 0, &bp
, ops
);
2662 if (whichfork
== XFS_ATTR_FORK
)
2663 xfs_buf_set_ref(bp
, XFS_ATTR_BTREE_REF
);
2665 xfs_buf_set_ref(bp
, XFS_DIR_BTREE_REF
);
2675 * Readahead the dir/attr block.
2679 struct xfs_inode
*dp
,
2681 xfs_daddr_t mappedbno
,
2683 const struct xfs_buf_ops
*ops
)
2685 struct xfs_buf_map map
;
2686 struct xfs_buf_map
*mapp
;
2692 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2695 /* mapping a hole is not an error, but we don't continue */
2701 mappedbno
= mapp
[0].bm_bn
;
2702 xfs_buf_readahead_map(dp
->i_mount
->m_ddev_targp
, mapp
, nmap
, ops
);