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_fspriv
;
181 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
183 if (xfs_da3_node_verify(bp
)) {
184 xfs_verifier_error(bp
, -EFSCORRUPTED
);
188 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
192 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
194 xfs_buf_update_cksum(bp
, XFS_DA3_NODE_CRC_OFF
);
198 * leaf/node format detection on trees is sketchy, so a node read can be done on
199 * leaf level blocks when detection identifies the tree as a node format tree
200 * incorrectly. In this case, we need to swap the verifier to match the correct
201 * format of the block being read.
204 xfs_da3_node_read_verify(
207 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
209 switch (be16_to_cpu(info
->magic
)) {
210 case XFS_DA3_NODE_MAGIC
:
211 if (!xfs_buf_verify_cksum(bp
, XFS_DA3_NODE_CRC_OFF
)) {
212 xfs_verifier_error(bp
, -EFSBADCRC
);
216 case XFS_DA_NODE_MAGIC
:
217 if (xfs_da3_node_verify(bp
))
218 xfs_verifier_error(bp
, -EFSCORRUPTED
);
220 case XFS_ATTR_LEAF_MAGIC
:
221 case XFS_ATTR3_LEAF_MAGIC
:
222 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
223 bp
->b_ops
->verify_read(bp
);
225 case XFS_DIR2_LEAFN_MAGIC
:
226 case XFS_DIR3_LEAFN_MAGIC
:
227 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
228 bp
->b_ops
->verify_read(bp
);
231 xfs_verifier_error(bp
, -EFSCORRUPTED
);
236 const struct xfs_buf_ops xfs_da3_node_buf_ops
= {
237 .name
= "xfs_da3_node",
238 .verify_read
= xfs_da3_node_read_verify
,
239 .verify_write
= xfs_da3_node_write_verify
,
244 struct xfs_trans
*tp
,
245 struct xfs_inode
*dp
,
247 xfs_daddr_t mappedbno
,
248 struct xfs_buf
**bpp
,
253 err
= xfs_da_read_buf(tp
, dp
, bno
, mappedbno
, bpp
,
254 which_fork
, &xfs_da3_node_buf_ops
);
255 if (!err
&& tp
&& *bpp
) {
256 struct xfs_da_blkinfo
*info
= (*bpp
)->b_addr
;
259 switch (be16_to_cpu(info
->magic
)) {
260 case XFS_DA_NODE_MAGIC
:
261 case XFS_DA3_NODE_MAGIC
:
262 type
= XFS_BLFT_DA_NODE_BUF
;
264 case XFS_ATTR_LEAF_MAGIC
:
265 case XFS_ATTR3_LEAF_MAGIC
:
266 type
= XFS_BLFT_ATTR_LEAF_BUF
;
268 case XFS_DIR2_LEAFN_MAGIC
:
269 case XFS_DIR3_LEAFN_MAGIC
:
270 type
= XFS_BLFT_DIR_LEAFN_BUF
;
277 xfs_trans_buf_set_type(tp
, *bpp
, type
);
282 /*========================================================================
283 * Routines used for growing the Btree.
284 *========================================================================*/
287 * Create the initial contents of an intermediate node.
291 struct xfs_da_args
*args
,
294 struct xfs_buf
**bpp
,
297 struct xfs_da_intnode
*node
;
298 struct xfs_trans
*tp
= args
->trans
;
299 struct xfs_mount
*mp
= tp
->t_mountp
;
300 struct xfs_da3_icnode_hdr ichdr
= {0};
303 struct xfs_inode
*dp
= args
->dp
;
305 trace_xfs_da_node_create(args
);
306 ASSERT(level
<= XFS_DA_NODE_MAXDEPTH
);
308 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, whichfork
);
311 bp
->b_ops
= &xfs_da3_node_buf_ops
;
312 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
315 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
316 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
318 memset(hdr3
, 0, sizeof(struct xfs_da3_node_hdr
));
319 ichdr
.magic
= XFS_DA3_NODE_MAGIC
;
320 hdr3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
321 hdr3
->info
.owner
= cpu_to_be64(args
->dp
->i_ino
);
322 uuid_copy(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_meta_uuid
);
324 ichdr
.magic
= XFS_DA_NODE_MAGIC
;
328 dp
->d_ops
->node_hdr_to_disk(node
, &ichdr
);
329 xfs_trans_log_buf(tp
, bp
,
330 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
337 * Split a leaf node, rebalance, then possibly split
338 * intermediate nodes, rebalance, etc.
342 struct xfs_da_state
*state
)
344 struct xfs_da_state_blk
*oldblk
;
345 struct xfs_da_state_blk
*newblk
;
346 struct xfs_da_state_blk
*addblk
;
347 struct xfs_da_intnode
*node
;
353 trace_xfs_da_split(state
->args
);
356 * Walk back up the tree splitting/inserting/adjusting as necessary.
357 * If we need to insert and there isn't room, split the node, then
358 * decide which fragment to insert the new block from below into.
359 * Note that we may split the root this way, but we need more fixup.
361 max
= state
->path
.active
- 1;
362 ASSERT((max
>= 0) && (max
< XFS_DA_NODE_MAXDEPTH
));
363 ASSERT(state
->path
.blk
[max
].magic
== XFS_ATTR_LEAF_MAGIC
||
364 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
366 addblk
= &state
->path
.blk
[max
]; /* initial dummy value */
367 for (i
= max
; (i
>= 0) && addblk
; state
->path
.active
--, i
--) {
368 oldblk
= &state
->path
.blk
[i
];
369 newblk
= &state
->altpath
.blk
[i
];
372 * If a leaf node then
373 * Allocate a new leaf node, then rebalance across them.
374 * else if an intermediate node then
375 * We split on the last layer, must we split the node?
377 switch (oldblk
->magic
) {
378 case XFS_ATTR_LEAF_MAGIC
:
379 error
= xfs_attr3_leaf_split(state
, oldblk
, newblk
);
380 if ((error
!= 0) && (error
!= -ENOSPC
)) {
381 return error
; /* GROT: attr is inconsistent */
388 * Entry wouldn't fit, split the leaf again. The new
389 * extrablk will be consumed by xfs_da3_node_split if
392 state
->extravalid
= 1;
394 state
->extraafter
= 0; /* before newblk */
395 trace_xfs_attr_leaf_split_before(state
->args
);
396 error
= xfs_attr3_leaf_split(state
, oldblk
,
399 state
->extraafter
= 1; /* after newblk */
400 trace_xfs_attr_leaf_split_after(state
->args
);
401 error
= xfs_attr3_leaf_split(state
, newblk
,
405 return error
; /* GROT: attr inconsistent */
408 case XFS_DIR2_LEAFN_MAGIC
:
409 error
= xfs_dir2_leafn_split(state
, oldblk
, newblk
);
414 case XFS_DA_NODE_MAGIC
:
415 error
= xfs_da3_node_split(state
, oldblk
, newblk
, addblk
,
419 return error
; /* GROT: dir is inconsistent */
421 * Record the newly split block for the next time thru?
431 * Update the btree to show the new hashval for this child.
433 xfs_da3_fixhashpath(state
, &state
->path
);
439 * xfs_da3_node_split() should have consumed any extra blocks we added
440 * during a double leaf split in the attr fork. This is guaranteed as
441 * we can't be here if the attr fork only has a single leaf block.
443 ASSERT(state
->extravalid
== 0 ||
444 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
447 * Split the root node.
449 ASSERT(state
->path
.active
== 0);
450 oldblk
= &state
->path
.blk
[0];
451 error
= xfs_da3_root_split(state
, oldblk
, addblk
);
454 return error
; /* GROT: dir is inconsistent */
458 * Update pointers to the node which used to be block 0 and just got
459 * bumped because of the addition of a new root node. Note that the
460 * original block 0 could be at any position in the list of blocks in
463 * Note: the magic numbers and sibling pointers are in the same physical
464 * place for both v2 and v3 headers (by design). Hence it doesn't matter
465 * which version of the xfs_da_intnode structure we use here as the
466 * result will be the same using either structure.
468 node
= oldblk
->bp
->b_addr
;
469 if (node
->hdr
.info
.forw
) {
470 ASSERT(be32_to_cpu(node
->hdr
.info
.forw
) == addblk
->blkno
);
471 node
= addblk
->bp
->b_addr
;
472 node
->hdr
.info
.back
= cpu_to_be32(oldblk
->blkno
);
473 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
474 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
475 sizeof(node
->hdr
.info
)));
477 node
= oldblk
->bp
->b_addr
;
478 if (node
->hdr
.info
.back
) {
479 ASSERT(be32_to_cpu(node
->hdr
.info
.back
) == addblk
->blkno
);
480 node
= addblk
->bp
->b_addr
;
481 node
->hdr
.info
.forw
= cpu_to_be32(oldblk
->blkno
);
482 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
483 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
484 sizeof(node
->hdr
.info
)));
491 * Split the root. We have to create a new root and point to the two
492 * parts (the split old root) that we just created. Copy block zero to
493 * the EOF, extending the inode in process.
495 STATIC
int /* error */
497 struct xfs_da_state
*state
,
498 struct xfs_da_state_blk
*blk1
,
499 struct xfs_da_state_blk
*blk2
)
501 struct xfs_da_intnode
*node
;
502 struct xfs_da_intnode
*oldroot
;
503 struct xfs_da_node_entry
*btree
;
504 struct xfs_da3_icnode_hdr nodehdr
;
505 struct xfs_da_args
*args
;
507 struct xfs_inode
*dp
;
508 struct xfs_trans
*tp
;
509 struct xfs_dir2_leaf
*leaf
;
515 trace_xfs_da_root_split(state
->args
);
518 * Copy the existing (incorrect) block from the root node position
519 * to a free space somewhere.
522 error
= xfs_da_grow_inode(args
, &blkno
);
528 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, args
->whichfork
);
532 oldroot
= blk1
->bp
->b_addr
;
533 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
534 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
)) {
535 struct xfs_da3_icnode_hdr icnodehdr
;
537 dp
->d_ops
->node_hdr_from_disk(&icnodehdr
, oldroot
);
538 btree
= dp
->d_ops
->node_tree_p(oldroot
);
539 size
= (int)((char *)&btree
[icnodehdr
.count
] - (char *)oldroot
);
540 level
= icnodehdr
.level
;
543 * we are about to copy oldroot to bp, so set up the type
544 * of bp while we know exactly what it will be.
546 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
548 struct xfs_dir3_icleaf_hdr leafhdr
;
549 struct xfs_dir2_leaf_entry
*ents
;
551 leaf
= (xfs_dir2_leaf_t
*)oldroot
;
552 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
553 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
555 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
556 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
557 size
= (int)((char *)&ents
[leafhdr
.count
] - (char *)leaf
);
561 * we are about to copy oldroot to bp, so set up the type
562 * of bp while we know exactly what it will be.
564 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAFN_BUF
);
568 * we can copy most of the information in the node from one block to
569 * another, but for CRC enabled headers we have to make sure that the
570 * block specific identifiers are kept intact. We update the buffer
573 memcpy(node
, oldroot
, size
);
574 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
575 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
576 struct xfs_da3_intnode
*node3
= (struct xfs_da3_intnode
*)node
;
578 node3
->hdr
.info
.blkno
= cpu_to_be64(bp
->b_bn
);
580 xfs_trans_log_buf(tp
, bp
, 0, size
- 1);
582 bp
->b_ops
= blk1
->bp
->b_ops
;
583 xfs_trans_buf_copy_type(bp
, blk1
->bp
);
588 * Set up the new root node.
590 error
= xfs_da3_node_create(args
,
591 (args
->whichfork
== XFS_DATA_FORK
) ? args
->geo
->leafblk
: 0,
592 level
+ 1, &bp
, args
->whichfork
);
597 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
598 btree
= dp
->d_ops
->node_tree_p(node
);
599 btree
[0].hashval
= cpu_to_be32(blk1
->hashval
);
600 btree
[0].before
= cpu_to_be32(blk1
->blkno
);
601 btree
[1].hashval
= cpu_to_be32(blk2
->hashval
);
602 btree
[1].before
= cpu_to_be32(blk2
->blkno
);
604 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
607 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
608 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
609 ASSERT(blk1
->blkno
>= args
->geo
->leafblk
&&
610 blk1
->blkno
< args
->geo
->freeblk
);
611 ASSERT(blk2
->blkno
>= args
->geo
->leafblk
&&
612 blk2
->blkno
< args
->geo
->freeblk
);
616 /* Header is already logged by xfs_da_node_create */
617 xfs_trans_log_buf(tp
, bp
,
618 XFS_DA_LOGRANGE(node
, btree
, sizeof(xfs_da_node_entry_t
) * 2));
624 * Split the node, rebalance, then add the new entry.
626 STATIC
int /* error */
628 struct xfs_da_state
*state
,
629 struct xfs_da_state_blk
*oldblk
,
630 struct xfs_da_state_blk
*newblk
,
631 struct xfs_da_state_blk
*addblk
,
635 struct xfs_da_intnode
*node
;
636 struct xfs_da3_icnode_hdr nodehdr
;
641 struct xfs_inode
*dp
= state
->args
->dp
;
643 trace_xfs_da_node_split(state
->args
);
645 node
= oldblk
->bp
->b_addr
;
646 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
649 * With V2 dirs the extra block is data or freespace.
651 useextra
= state
->extravalid
&& state
->args
->whichfork
== XFS_ATTR_FORK
;
652 newcount
= 1 + useextra
;
654 * Do we have to split the node?
656 if (nodehdr
.count
+ newcount
> state
->args
->geo
->node_ents
) {
658 * Allocate a new node, add to the doubly linked chain of
659 * nodes, then move some of our excess entries into it.
661 error
= xfs_da_grow_inode(state
->args
, &blkno
);
663 return error
; /* GROT: dir is inconsistent */
665 error
= xfs_da3_node_create(state
->args
, blkno
, treelevel
,
666 &newblk
->bp
, state
->args
->whichfork
);
668 return error
; /* GROT: dir is inconsistent */
669 newblk
->blkno
= blkno
;
670 newblk
->magic
= XFS_DA_NODE_MAGIC
;
671 xfs_da3_node_rebalance(state
, oldblk
, newblk
);
672 error
= xfs_da3_blk_link(state
, oldblk
, newblk
);
681 * Insert the new entry(s) into the correct block
682 * (updating last hashval in the process).
684 * xfs_da3_node_add() inserts BEFORE the given index,
685 * and as a result of using node_lookup_int() we always
686 * point to a valid entry (not after one), but a split
687 * operation always results in a new block whose hashvals
688 * FOLLOW the current block.
690 * If we had double-split op below us, then add the extra block too.
692 node
= oldblk
->bp
->b_addr
;
693 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
694 if (oldblk
->index
<= nodehdr
.count
) {
696 xfs_da3_node_add(state
, oldblk
, addblk
);
698 if (state
->extraafter
)
700 xfs_da3_node_add(state
, oldblk
, &state
->extrablk
);
701 state
->extravalid
= 0;
705 xfs_da3_node_add(state
, newblk
, addblk
);
707 if (state
->extraafter
)
709 xfs_da3_node_add(state
, newblk
, &state
->extrablk
);
710 state
->extravalid
= 0;
718 * Balance the btree elements between two intermediate nodes,
719 * usually one full and one empty.
721 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
724 xfs_da3_node_rebalance(
725 struct xfs_da_state
*state
,
726 struct xfs_da_state_blk
*blk1
,
727 struct xfs_da_state_blk
*blk2
)
729 struct xfs_da_intnode
*node1
;
730 struct xfs_da_intnode
*node2
;
731 struct xfs_da_intnode
*tmpnode
;
732 struct xfs_da_node_entry
*btree1
;
733 struct xfs_da_node_entry
*btree2
;
734 struct xfs_da_node_entry
*btree_s
;
735 struct xfs_da_node_entry
*btree_d
;
736 struct xfs_da3_icnode_hdr nodehdr1
;
737 struct xfs_da3_icnode_hdr nodehdr2
;
738 struct xfs_trans
*tp
;
742 struct xfs_inode
*dp
= state
->args
->dp
;
744 trace_xfs_da_node_rebalance(state
->args
);
746 node1
= blk1
->bp
->b_addr
;
747 node2
= blk2
->bp
->b_addr
;
748 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
749 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
750 btree1
= dp
->d_ops
->node_tree_p(node1
);
751 btree2
= dp
->d_ops
->node_tree_p(node2
);
754 * Figure out how many entries need to move, and in which direction.
755 * Swap the nodes around if that makes it simpler.
757 if (nodehdr1
.count
> 0 && nodehdr2
.count
> 0 &&
758 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
759 (be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
) <
760 be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
)))) {
764 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
765 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
766 btree1
= dp
->d_ops
->node_tree_p(node1
);
767 btree2
= dp
->d_ops
->node_tree_p(node2
);
771 count
= (nodehdr1
.count
- nodehdr2
.count
) / 2;
774 tp
= state
->args
->trans
;
776 * Two cases: high-to-low and low-to-high.
780 * Move elements in node2 up to make a hole.
782 tmp
= nodehdr2
.count
;
784 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
785 btree_s
= &btree2
[0];
786 btree_d
= &btree2
[count
];
787 memmove(btree_d
, btree_s
, tmp
);
791 * Move the req'd B-tree elements from high in node1 to
794 nodehdr2
.count
+= count
;
795 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
796 btree_s
= &btree1
[nodehdr1
.count
- count
];
797 btree_d
= &btree2
[0];
798 memcpy(btree_d
, btree_s
, tmp
);
799 nodehdr1
.count
-= count
;
802 * Move the req'd B-tree elements from low in node2 to
806 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
807 btree_s
= &btree2
[0];
808 btree_d
= &btree1
[nodehdr1
.count
];
809 memcpy(btree_d
, btree_s
, tmp
);
810 nodehdr1
.count
+= count
;
812 xfs_trans_log_buf(tp
, blk1
->bp
,
813 XFS_DA_LOGRANGE(node1
, btree_d
, tmp
));
816 * Move elements in node2 down to fill the hole.
818 tmp
= nodehdr2
.count
- count
;
819 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
820 btree_s
= &btree2
[count
];
821 btree_d
= &btree2
[0];
822 memmove(btree_d
, btree_s
, tmp
);
823 nodehdr2
.count
-= count
;
827 * Log header of node 1 and all current bits of node 2.
829 dp
->d_ops
->node_hdr_to_disk(node1
, &nodehdr1
);
830 xfs_trans_log_buf(tp
, blk1
->bp
,
831 XFS_DA_LOGRANGE(node1
, &node1
->hdr
, dp
->d_ops
->node_hdr_size
));
833 dp
->d_ops
->node_hdr_to_disk(node2
, &nodehdr2
);
834 xfs_trans_log_buf(tp
, blk2
->bp
,
835 XFS_DA_LOGRANGE(node2
, &node2
->hdr
,
836 dp
->d_ops
->node_hdr_size
+
837 (sizeof(btree2
[0]) * nodehdr2
.count
)));
840 * Record the last hashval from each block for upward propagation.
841 * (note: don't use the swapped node pointers)
844 node1
= blk1
->bp
->b_addr
;
845 node2
= blk2
->bp
->b_addr
;
846 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
847 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
848 btree1
= dp
->d_ops
->node_tree_p(node1
);
849 btree2
= dp
->d_ops
->node_tree_p(node2
);
851 blk1
->hashval
= be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
);
852 blk2
->hashval
= be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
);
855 * Adjust the expected index for insertion.
857 if (blk1
->index
>= nodehdr1
.count
) {
858 blk2
->index
= blk1
->index
- nodehdr1
.count
;
859 blk1
->index
= nodehdr1
.count
+ 1; /* make it invalid */
864 * Add a new entry to an intermediate node.
868 struct xfs_da_state
*state
,
869 struct xfs_da_state_blk
*oldblk
,
870 struct xfs_da_state_blk
*newblk
)
872 struct xfs_da_intnode
*node
;
873 struct xfs_da3_icnode_hdr nodehdr
;
874 struct xfs_da_node_entry
*btree
;
876 struct xfs_inode
*dp
= state
->args
->dp
;
878 trace_xfs_da_node_add(state
->args
);
880 node
= oldblk
->bp
->b_addr
;
881 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
882 btree
= dp
->d_ops
->node_tree_p(node
);
884 ASSERT(oldblk
->index
>= 0 && oldblk
->index
<= nodehdr
.count
);
885 ASSERT(newblk
->blkno
!= 0);
886 if (state
->args
->whichfork
== XFS_DATA_FORK
)
887 ASSERT(newblk
->blkno
>= state
->args
->geo
->leafblk
&&
888 newblk
->blkno
< state
->args
->geo
->freeblk
);
891 * We may need to make some room before we insert the new node.
894 if (oldblk
->index
< nodehdr
.count
) {
895 tmp
= (nodehdr
.count
- oldblk
->index
) * (uint
)sizeof(*btree
);
896 memmove(&btree
[oldblk
->index
+ 1], &btree
[oldblk
->index
], tmp
);
898 btree
[oldblk
->index
].hashval
= cpu_to_be32(newblk
->hashval
);
899 btree
[oldblk
->index
].before
= cpu_to_be32(newblk
->blkno
);
900 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
901 XFS_DA_LOGRANGE(node
, &btree
[oldblk
->index
],
902 tmp
+ sizeof(*btree
)));
905 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
906 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
907 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
910 * Copy the last hash value from the oldblk to propagate upwards.
912 oldblk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
915 /*========================================================================
916 * Routines used for shrinking the Btree.
917 *========================================================================*/
920 * Deallocate an empty leaf node, remove it from its parent,
921 * possibly deallocating that block, etc...
925 struct xfs_da_state
*state
)
927 struct xfs_da_state_blk
*drop_blk
;
928 struct xfs_da_state_blk
*save_blk
;
932 trace_xfs_da_join(state
->args
);
934 drop_blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
935 save_blk
= &state
->altpath
.blk
[ state
->path
.active
-1 ];
936 ASSERT(state
->path
.blk
[0].magic
== XFS_DA_NODE_MAGIC
);
937 ASSERT(drop_blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
938 drop_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
941 * Walk back up the tree joining/deallocating as necessary.
942 * When we stop dropping blocks, break out.
944 for ( ; state
->path
.active
>= 2; drop_blk
--, save_blk
--,
945 state
->path
.active
--) {
947 * See if we can combine the block with a neighbor.
948 * (action == 0) => no options, just leave
949 * (action == 1) => coalesce, then unlink
950 * (action == 2) => block empty, unlink it
952 switch (drop_blk
->magic
) {
953 case XFS_ATTR_LEAF_MAGIC
:
954 error
= xfs_attr3_leaf_toosmall(state
, &action
);
959 xfs_attr3_leaf_unbalance(state
, drop_blk
, save_blk
);
961 case XFS_DIR2_LEAFN_MAGIC
:
962 error
= xfs_dir2_leafn_toosmall(state
, &action
);
967 xfs_dir2_leafn_unbalance(state
, drop_blk
, save_blk
);
969 case XFS_DA_NODE_MAGIC
:
971 * Remove the offending node, fixup hashvals,
972 * check for a toosmall neighbor.
974 xfs_da3_node_remove(state
, drop_blk
);
975 xfs_da3_fixhashpath(state
, &state
->path
);
976 error
= xfs_da3_node_toosmall(state
, &action
);
981 xfs_da3_node_unbalance(state
, drop_blk
, save_blk
);
984 xfs_da3_fixhashpath(state
, &state
->altpath
);
985 error
= xfs_da3_blk_unlink(state
, drop_blk
, save_blk
);
986 xfs_da_state_kill_altpath(state
);
989 error
= xfs_da_shrink_inode(state
->args
, drop_blk
->blkno
,
996 * We joined all the way to the top. If it turns out that
997 * we only have one entry in the root, make the child block
1000 xfs_da3_node_remove(state
, drop_blk
);
1001 xfs_da3_fixhashpath(state
, &state
->path
);
1002 error
= xfs_da3_root_join(state
, &state
->path
.blk
[0]);
1008 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo
*blkinfo
, __u16 level
)
1010 __be16 magic
= blkinfo
->magic
;
1013 ASSERT(magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1014 magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1015 magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1016 magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1018 ASSERT(magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1019 magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
));
1021 ASSERT(!blkinfo
->forw
);
1022 ASSERT(!blkinfo
->back
);
1025 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1029 * We have only one entry in the root. Copy the only remaining child of
1030 * the old root to block 0 as the new root node.
1034 struct xfs_da_state
*state
,
1035 struct xfs_da_state_blk
*root_blk
)
1037 struct xfs_da_intnode
*oldroot
;
1038 struct xfs_da_args
*args
;
1041 struct xfs_da3_icnode_hdr oldroothdr
;
1042 struct xfs_da_node_entry
*btree
;
1044 struct xfs_inode
*dp
= state
->args
->dp
;
1046 trace_xfs_da_root_join(state
->args
);
1048 ASSERT(root_blk
->magic
== XFS_DA_NODE_MAGIC
);
1051 oldroot
= root_blk
->bp
->b_addr
;
1052 dp
->d_ops
->node_hdr_from_disk(&oldroothdr
, oldroot
);
1053 ASSERT(oldroothdr
.forw
== 0);
1054 ASSERT(oldroothdr
.back
== 0);
1057 * If the root has more than one child, then don't do anything.
1059 if (oldroothdr
.count
> 1)
1063 * Read in the (only) child block, then copy those bytes into
1064 * the root block's buffer and free the original child block.
1066 btree
= dp
->d_ops
->node_tree_p(oldroot
);
1067 child
= be32_to_cpu(btree
[0].before
);
1069 error
= xfs_da3_node_read(args
->trans
, dp
, child
, -1, &bp
,
1073 xfs_da_blkinfo_onlychild_validate(bp
->b_addr
, oldroothdr
.level
);
1076 * This could be copying a leaf back into the root block in the case of
1077 * there only being a single leaf block left in the tree. Hence we have
1078 * to update the b_ops pointer as well to match the buffer type change
1079 * that could occur. For dir3 blocks we also need to update the block
1080 * number in the buffer header.
1082 memcpy(root_blk
->bp
->b_addr
, bp
->b_addr
, args
->geo
->blksize
);
1083 root_blk
->bp
->b_ops
= bp
->b_ops
;
1084 xfs_trans_buf_copy_type(root_blk
->bp
, bp
);
1085 if (oldroothdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1086 struct xfs_da3_blkinfo
*da3
= root_blk
->bp
->b_addr
;
1087 da3
->blkno
= cpu_to_be64(root_blk
->bp
->b_bn
);
1089 xfs_trans_log_buf(args
->trans
, root_blk
->bp
, 0,
1090 args
->geo
->blksize
- 1);
1091 error
= xfs_da_shrink_inode(args
, child
, bp
);
1096 * Check a node block and its neighbors to see if the block should be
1097 * collapsed into one or the other neighbor. Always keep the block
1098 * with the smaller block number.
1099 * If the current block is over 50% full, don't try to join it, return 0.
1100 * If the block is empty, fill in the state structure and return 2.
1101 * If it can be collapsed, fill in the state structure and return 1.
1102 * If nothing can be done, return 0.
1105 xfs_da3_node_toosmall(
1106 struct xfs_da_state
*state
,
1109 struct xfs_da_intnode
*node
;
1110 struct xfs_da_state_blk
*blk
;
1111 struct xfs_da_blkinfo
*info
;
1114 struct xfs_da3_icnode_hdr nodehdr
;
1120 struct xfs_inode
*dp
= state
->args
->dp
;
1122 trace_xfs_da_node_toosmall(state
->args
);
1125 * Check for the degenerate case of the block being over 50% full.
1126 * If so, it's not worth even looking to see if we might be able
1127 * to coalesce with a sibling.
1129 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1130 info
= blk
->bp
->b_addr
;
1131 node
= (xfs_da_intnode_t
*)info
;
1132 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1133 if (nodehdr
.count
> (state
->args
->geo
->node_ents
>> 1)) {
1134 *action
= 0; /* blk over 50%, don't try to join */
1135 return 0; /* blk over 50%, don't try to join */
1139 * Check for the degenerate case of the block being empty.
1140 * If the block is empty, we'll simply delete it, no need to
1141 * coalesce it with a sibling block. We choose (arbitrarily)
1142 * to merge with the forward block unless it is NULL.
1144 if (nodehdr
.count
== 0) {
1146 * Make altpath point to the block we want to keep and
1147 * path point to the block we want to drop (this one).
1149 forward
= (info
->forw
!= 0);
1150 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1151 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1164 * Examine each sibling block to see if we can coalesce with
1165 * at least 25% free space to spare. We need to figure out
1166 * whether to merge with the forward or the backward block.
1167 * We prefer coalescing with the lower numbered sibling so as
1168 * to shrink a directory over time.
1170 count
= state
->args
->geo
->node_ents
;
1171 count
-= state
->args
->geo
->node_ents
>> 2;
1172 count
-= nodehdr
.count
;
1174 /* start with smaller blk num */
1175 forward
= nodehdr
.forw
< nodehdr
.back
;
1176 for (i
= 0; i
< 2; forward
= !forward
, i
++) {
1177 struct xfs_da3_icnode_hdr thdr
;
1179 blkno
= nodehdr
.forw
;
1181 blkno
= nodehdr
.back
;
1184 error
= xfs_da3_node_read(state
->args
->trans
, dp
,
1185 blkno
, -1, &bp
, state
->args
->whichfork
);
1190 dp
->d_ops
->node_hdr_from_disk(&thdr
, node
);
1191 xfs_trans_brelse(state
->args
->trans
, bp
);
1193 if (count
- thdr
.count
>= 0)
1194 break; /* fits with at least 25% to spare */
1202 * Make altpath point to the block we want to keep (the lower
1203 * numbered block) and path point to the block we want to drop.
1205 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1206 if (blkno
< blk
->blkno
) {
1207 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1210 error
= xfs_da3_path_shift(state
, &state
->path
, forward
,
1224 * Pick up the last hashvalue from an intermediate node.
1227 xfs_da3_node_lasthash(
1228 struct xfs_inode
*dp
,
1232 struct xfs_da_intnode
*node
;
1233 struct xfs_da_node_entry
*btree
;
1234 struct xfs_da3_icnode_hdr nodehdr
;
1237 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1239 *count
= nodehdr
.count
;
1242 btree
= dp
->d_ops
->node_tree_p(node
);
1243 return be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1247 * Walk back up the tree adjusting hash values as necessary,
1248 * when we stop making changes, return.
1251 xfs_da3_fixhashpath(
1252 struct xfs_da_state
*state
,
1253 struct xfs_da_state_path
*path
)
1255 struct xfs_da_state_blk
*blk
;
1256 struct xfs_da_intnode
*node
;
1257 struct xfs_da_node_entry
*btree
;
1258 xfs_dahash_t lasthash
=0;
1261 struct xfs_inode
*dp
= state
->args
->dp
;
1263 trace_xfs_da_fixhashpath(state
->args
);
1265 level
= path
->active
-1;
1266 blk
= &path
->blk
[ level
];
1267 switch (blk
->magic
) {
1268 case XFS_ATTR_LEAF_MAGIC
:
1269 lasthash
= xfs_attr_leaf_lasthash(blk
->bp
, &count
);
1273 case XFS_DIR2_LEAFN_MAGIC
:
1274 lasthash
= xfs_dir2_leaf_lasthash(dp
, blk
->bp
, &count
);
1278 case XFS_DA_NODE_MAGIC
:
1279 lasthash
= xfs_da3_node_lasthash(dp
, blk
->bp
, &count
);
1284 for (blk
--, level
--; level
>= 0; blk
--, level
--) {
1285 struct xfs_da3_icnode_hdr nodehdr
;
1287 node
= blk
->bp
->b_addr
;
1288 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1289 btree
= dp
->d_ops
->node_tree_p(node
);
1290 if (be32_to_cpu(btree
[blk
->index
].hashval
) == lasthash
)
1292 blk
->hashval
= lasthash
;
1293 btree
[blk
->index
].hashval
= cpu_to_be32(lasthash
);
1294 xfs_trans_log_buf(state
->args
->trans
, blk
->bp
,
1295 XFS_DA_LOGRANGE(node
, &btree
[blk
->index
],
1298 lasthash
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1303 * Remove an entry from an intermediate node.
1306 xfs_da3_node_remove(
1307 struct xfs_da_state
*state
,
1308 struct xfs_da_state_blk
*drop_blk
)
1310 struct xfs_da_intnode
*node
;
1311 struct xfs_da3_icnode_hdr nodehdr
;
1312 struct xfs_da_node_entry
*btree
;
1315 struct xfs_inode
*dp
= state
->args
->dp
;
1317 trace_xfs_da_node_remove(state
->args
);
1319 node
= drop_blk
->bp
->b_addr
;
1320 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1321 ASSERT(drop_blk
->index
< nodehdr
.count
);
1322 ASSERT(drop_blk
->index
>= 0);
1325 * Copy over the offending entry, or just zero it out.
1327 index
= drop_blk
->index
;
1328 btree
= dp
->d_ops
->node_tree_p(node
);
1329 if (index
< nodehdr
.count
- 1) {
1330 tmp
= nodehdr
.count
- index
- 1;
1331 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
1332 memmove(&btree
[index
], &btree
[index
+ 1], tmp
);
1333 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1334 XFS_DA_LOGRANGE(node
, &btree
[index
], tmp
));
1335 index
= nodehdr
.count
- 1;
1337 memset(&btree
[index
], 0, sizeof(xfs_da_node_entry_t
));
1338 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1339 XFS_DA_LOGRANGE(node
, &btree
[index
], sizeof(btree
[index
])));
1341 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
1342 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1343 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
1346 * Copy the last hash value from the block to propagate upwards.
1348 drop_blk
->hashval
= be32_to_cpu(btree
[index
- 1].hashval
);
1352 * Unbalance the elements between two intermediate nodes,
1353 * move all Btree elements from one node into another.
1356 xfs_da3_node_unbalance(
1357 struct xfs_da_state
*state
,
1358 struct xfs_da_state_blk
*drop_blk
,
1359 struct xfs_da_state_blk
*save_blk
)
1361 struct xfs_da_intnode
*drop_node
;
1362 struct xfs_da_intnode
*save_node
;
1363 struct xfs_da_node_entry
*drop_btree
;
1364 struct xfs_da_node_entry
*save_btree
;
1365 struct xfs_da3_icnode_hdr drop_hdr
;
1366 struct xfs_da3_icnode_hdr save_hdr
;
1367 struct xfs_trans
*tp
;
1370 struct xfs_inode
*dp
= state
->args
->dp
;
1372 trace_xfs_da_node_unbalance(state
->args
);
1374 drop_node
= drop_blk
->bp
->b_addr
;
1375 save_node
= save_blk
->bp
->b_addr
;
1376 dp
->d_ops
->node_hdr_from_disk(&drop_hdr
, drop_node
);
1377 dp
->d_ops
->node_hdr_from_disk(&save_hdr
, save_node
);
1378 drop_btree
= dp
->d_ops
->node_tree_p(drop_node
);
1379 save_btree
= dp
->d_ops
->node_tree_p(save_node
);
1380 tp
= state
->args
->trans
;
1383 * If the dying block has lower hashvals, then move all the
1384 * elements in the remaining block up to make a hole.
1386 if ((be32_to_cpu(drop_btree
[0].hashval
) <
1387 be32_to_cpu(save_btree
[0].hashval
)) ||
1388 (be32_to_cpu(drop_btree
[drop_hdr
.count
- 1].hashval
) <
1389 be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
))) {
1390 /* XXX: check this - is memmove dst correct? */
1391 tmp
= save_hdr
.count
* sizeof(xfs_da_node_entry_t
);
1392 memmove(&save_btree
[drop_hdr
.count
], &save_btree
[0], tmp
);
1395 xfs_trans_log_buf(tp
, save_blk
->bp
,
1396 XFS_DA_LOGRANGE(save_node
, &save_btree
[0],
1397 (save_hdr
.count
+ drop_hdr
.count
) *
1398 sizeof(xfs_da_node_entry_t
)));
1400 sindex
= save_hdr
.count
;
1401 xfs_trans_log_buf(tp
, save_blk
->bp
,
1402 XFS_DA_LOGRANGE(save_node
, &save_btree
[sindex
],
1403 drop_hdr
.count
* sizeof(xfs_da_node_entry_t
)));
1407 * Move all the B-tree elements from drop_blk to save_blk.
1409 tmp
= drop_hdr
.count
* (uint
)sizeof(xfs_da_node_entry_t
);
1410 memcpy(&save_btree
[sindex
], &drop_btree
[0], tmp
);
1411 save_hdr
.count
+= drop_hdr
.count
;
1413 dp
->d_ops
->node_hdr_to_disk(save_node
, &save_hdr
);
1414 xfs_trans_log_buf(tp
, save_blk
->bp
,
1415 XFS_DA_LOGRANGE(save_node
, &save_node
->hdr
,
1416 dp
->d_ops
->node_hdr_size
));
1419 * Save the last hashval in the remaining block for upward propagation.
1421 save_blk
->hashval
= be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
);
1424 /*========================================================================
1425 * Routines used for finding things in the Btree.
1426 *========================================================================*/
1429 * Walk down the Btree looking for a particular filename, filling
1430 * in the state structure as we go.
1432 * We will set the state structure to point to each of the elements
1433 * in each of the nodes where either the hashval is or should be.
1435 * We support duplicate hashval's so for each entry in the current
1436 * node that could contain the desired hashval, descend. This is a
1437 * pruned depth-first tree search.
1440 xfs_da3_node_lookup_int(
1441 struct xfs_da_state
*state
,
1444 struct xfs_da_state_blk
*blk
;
1445 struct xfs_da_blkinfo
*curr
;
1446 struct xfs_da_intnode
*node
;
1447 struct xfs_da_node_entry
*btree
;
1448 struct xfs_da3_icnode_hdr nodehdr
;
1449 struct xfs_da_args
*args
;
1451 xfs_dahash_t hashval
;
1452 xfs_dahash_t btreehashval
;
1458 unsigned int expected_level
= 0;
1459 struct xfs_inode
*dp
= state
->args
->dp
;
1464 * Descend thru the B-tree searching each level for the right
1465 * node to use, until the right hashval is found.
1467 blkno
= args
->geo
->leafblk
;
1468 for (blk
= &state
->path
.blk
[0], state
->path
.active
= 1;
1469 state
->path
.active
<= XFS_DA_NODE_MAXDEPTH
;
1470 blk
++, state
->path
.active
++) {
1472 * Read the next node down in the tree.
1475 error
= xfs_da3_node_read(args
->trans
, args
->dp
, blkno
,
1476 -1, &blk
->bp
, args
->whichfork
);
1479 state
->path
.active
--;
1482 curr
= blk
->bp
->b_addr
;
1483 blk
->magic
= be16_to_cpu(curr
->magic
);
1485 if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
1486 blk
->magic
== XFS_ATTR3_LEAF_MAGIC
) {
1487 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1488 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1492 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1493 blk
->magic
== XFS_DIR3_LEAFN_MAGIC
) {
1494 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1495 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1500 blk
->magic
= XFS_DA_NODE_MAGIC
;
1504 * Search an intermediate node for a match.
1506 node
= blk
->bp
->b_addr
;
1507 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1508 btree
= dp
->d_ops
->node_tree_p(node
);
1510 /* Tree taller than we can handle; bail out! */
1511 if (nodehdr
.level
>= XFS_DA_NODE_MAXDEPTH
)
1512 return -EFSCORRUPTED
;
1514 /* Check the level from the root. */
1515 if (blkno
== args
->geo
->leafblk
)
1516 expected_level
= nodehdr
.level
- 1;
1517 else if (expected_level
!= nodehdr
.level
)
1518 return -EFSCORRUPTED
;
1522 max
= nodehdr
.count
;
1523 blk
->hashval
= be32_to_cpu(btree
[max
- 1].hashval
);
1526 * Binary search. (note: small blocks will skip loop)
1528 probe
= span
= max
/ 2;
1529 hashval
= args
->hashval
;
1532 btreehashval
= be32_to_cpu(btree
[probe
].hashval
);
1533 if (btreehashval
< hashval
)
1535 else if (btreehashval
> hashval
)
1540 ASSERT((probe
>= 0) && (probe
< max
));
1541 ASSERT((span
<= 4) ||
1542 (be32_to_cpu(btree
[probe
].hashval
) == hashval
));
1545 * Since we may have duplicate hashval's, find the first
1546 * matching hashval in the node.
1549 be32_to_cpu(btree
[probe
].hashval
) >= hashval
) {
1552 while (probe
< max
&&
1553 be32_to_cpu(btree
[probe
].hashval
) < hashval
) {
1558 * Pick the right block to descend on.
1561 blk
->index
= max
- 1;
1562 blkno
= be32_to_cpu(btree
[max
- 1].before
);
1565 blkno
= be32_to_cpu(btree
[probe
].before
);
1568 /* We can't point back to the root. */
1569 if (blkno
== args
->geo
->leafblk
)
1570 return -EFSCORRUPTED
;
1573 if (expected_level
!= 0)
1574 return -EFSCORRUPTED
;
1577 * A leaf block that ends in the hashval that we are interested in
1578 * (final hashval == search hashval) means that the next block may
1579 * contain more entries with the same hashval, shift upward to the
1580 * next leaf and keep searching.
1583 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
) {
1584 retval
= xfs_dir2_leafn_lookup_int(blk
->bp
, args
,
1585 &blk
->index
, state
);
1586 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1587 retval
= xfs_attr3_leaf_lookup_int(blk
->bp
, args
);
1588 blk
->index
= args
->index
;
1589 args
->blkno
= blk
->blkno
;
1592 return -EFSCORRUPTED
;
1594 if (((retval
== -ENOENT
) || (retval
== -ENOATTR
)) &&
1595 (blk
->hashval
== args
->hashval
)) {
1596 error
= xfs_da3_path_shift(state
, &state
->path
, 1, 1,
1602 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1603 /* path_shift() gives ENOENT */
1613 /*========================================================================
1615 *========================================================================*/
1618 * Compare two intermediate nodes for "order".
1622 struct xfs_inode
*dp
,
1623 struct xfs_buf
*node1_bp
,
1624 struct xfs_buf
*node2_bp
)
1626 struct xfs_da_intnode
*node1
;
1627 struct xfs_da_intnode
*node2
;
1628 struct xfs_da_node_entry
*btree1
;
1629 struct xfs_da_node_entry
*btree2
;
1630 struct xfs_da3_icnode_hdr node1hdr
;
1631 struct xfs_da3_icnode_hdr node2hdr
;
1633 node1
= node1_bp
->b_addr
;
1634 node2
= node2_bp
->b_addr
;
1635 dp
->d_ops
->node_hdr_from_disk(&node1hdr
, node1
);
1636 dp
->d_ops
->node_hdr_from_disk(&node2hdr
, node2
);
1637 btree1
= dp
->d_ops
->node_tree_p(node1
);
1638 btree2
= dp
->d_ops
->node_tree_p(node2
);
1640 if (node1hdr
.count
> 0 && node2hdr
.count
> 0 &&
1641 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
1642 (be32_to_cpu(btree2
[node2hdr
.count
- 1].hashval
) <
1643 be32_to_cpu(btree1
[node1hdr
.count
- 1].hashval
)))) {
1650 * Link a new block into a doubly linked list of blocks (of whatever type).
1654 struct xfs_da_state
*state
,
1655 struct xfs_da_state_blk
*old_blk
,
1656 struct xfs_da_state_blk
*new_blk
)
1658 struct xfs_da_blkinfo
*old_info
;
1659 struct xfs_da_blkinfo
*new_info
;
1660 struct xfs_da_blkinfo
*tmp_info
;
1661 struct xfs_da_args
*args
;
1665 struct xfs_inode
*dp
= state
->args
->dp
;
1668 * Set up environment.
1671 ASSERT(args
!= NULL
);
1672 old_info
= old_blk
->bp
->b_addr
;
1673 new_info
= new_blk
->bp
->b_addr
;
1674 ASSERT(old_blk
->magic
== XFS_DA_NODE_MAGIC
||
1675 old_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1676 old_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1678 switch (old_blk
->magic
) {
1679 case XFS_ATTR_LEAF_MAGIC
:
1680 before
= xfs_attr_leaf_order(old_blk
->bp
, new_blk
->bp
);
1682 case XFS_DIR2_LEAFN_MAGIC
:
1683 before
= xfs_dir2_leafn_order(dp
, old_blk
->bp
, new_blk
->bp
);
1685 case XFS_DA_NODE_MAGIC
:
1686 before
= xfs_da3_node_order(dp
, old_blk
->bp
, new_blk
->bp
);
1691 * Link blocks in appropriate order.
1695 * Link new block in before existing block.
1697 trace_xfs_da_link_before(args
);
1698 new_info
->forw
= cpu_to_be32(old_blk
->blkno
);
1699 new_info
->back
= old_info
->back
;
1700 if (old_info
->back
) {
1701 error
= xfs_da3_node_read(args
->trans
, dp
,
1702 be32_to_cpu(old_info
->back
),
1703 -1, &bp
, args
->whichfork
);
1707 tmp_info
= bp
->b_addr
;
1708 ASSERT(tmp_info
->magic
== old_info
->magic
);
1709 ASSERT(be32_to_cpu(tmp_info
->forw
) == old_blk
->blkno
);
1710 tmp_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1711 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1713 old_info
->back
= cpu_to_be32(new_blk
->blkno
);
1716 * Link new block in after existing block.
1718 trace_xfs_da_link_after(args
);
1719 new_info
->forw
= old_info
->forw
;
1720 new_info
->back
= cpu_to_be32(old_blk
->blkno
);
1721 if (old_info
->forw
) {
1722 error
= xfs_da3_node_read(args
->trans
, dp
,
1723 be32_to_cpu(old_info
->forw
),
1724 -1, &bp
, args
->whichfork
);
1728 tmp_info
= bp
->b_addr
;
1729 ASSERT(tmp_info
->magic
== old_info
->magic
);
1730 ASSERT(be32_to_cpu(tmp_info
->back
) == old_blk
->blkno
);
1731 tmp_info
->back
= cpu_to_be32(new_blk
->blkno
);
1732 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1734 old_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1737 xfs_trans_log_buf(args
->trans
, old_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1738 xfs_trans_log_buf(args
->trans
, new_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1743 * Unlink a block from a doubly linked list of blocks.
1745 STATIC
int /* error */
1747 struct xfs_da_state
*state
,
1748 struct xfs_da_state_blk
*drop_blk
,
1749 struct xfs_da_state_blk
*save_blk
)
1751 struct xfs_da_blkinfo
*drop_info
;
1752 struct xfs_da_blkinfo
*save_info
;
1753 struct xfs_da_blkinfo
*tmp_info
;
1754 struct xfs_da_args
*args
;
1759 * Set up environment.
1762 ASSERT(args
!= NULL
);
1763 save_info
= save_blk
->bp
->b_addr
;
1764 drop_info
= drop_blk
->bp
->b_addr
;
1765 ASSERT(save_blk
->magic
== XFS_DA_NODE_MAGIC
||
1766 save_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1767 save_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1768 ASSERT(save_blk
->magic
== drop_blk
->magic
);
1769 ASSERT((be32_to_cpu(save_info
->forw
) == drop_blk
->blkno
) ||
1770 (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
));
1771 ASSERT((be32_to_cpu(drop_info
->forw
) == save_blk
->blkno
) ||
1772 (be32_to_cpu(drop_info
->back
) == save_blk
->blkno
));
1775 * Unlink the leaf block from the doubly linked chain of leaves.
1777 if (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
) {
1778 trace_xfs_da_unlink_back(args
);
1779 save_info
->back
= drop_info
->back
;
1780 if (drop_info
->back
) {
1781 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1782 be32_to_cpu(drop_info
->back
),
1783 -1, &bp
, args
->whichfork
);
1787 tmp_info
= bp
->b_addr
;
1788 ASSERT(tmp_info
->magic
== save_info
->magic
);
1789 ASSERT(be32_to_cpu(tmp_info
->forw
) == drop_blk
->blkno
);
1790 tmp_info
->forw
= cpu_to_be32(save_blk
->blkno
);
1791 xfs_trans_log_buf(args
->trans
, bp
, 0,
1792 sizeof(*tmp_info
) - 1);
1795 trace_xfs_da_unlink_forward(args
);
1796 save_info
->forw
= drop_info
->forw
;
1797 if (drop_info
->forw
) {
1798 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1799 be32_to_cpu(drop_info
->forw
),
1800 -1, &bp
, args
->whichfork
);
1804 tmp_info
= bp
->b_addr
;
1805 ASSERT(tmp_info
->magic
== save_info
->magic
);
1806 ASSERT(be32_to_cpu(tmp_info
->back
) == drop_blk
->blkno
);
1807 tmp_info
->back
= cpu_to_be32(save_blk
->blkno
);
1808 xfs_trans_log_buf(args
->trans
, bp
, 0,
1809 sizeof(*tmp_info
) - 1);
1813 xfs_trans_log_buf(args
->trans
, save_blk
->bp
, 0, sizeof(*save_info
) - 1);
1818 * Move a path "forward" or "!forward" one block at the current level.
1820 * This routine will adjust a "path" to point to the next block
1821 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1822 * Btree, including updating pointers to the intermediate nodes between
1823 * the new bottom and the root.
1827 struct xfs_da_state
*state
,
1828 struct xfs_da_state_path
*path
,
1833 struct xfs_da_state_blk
*blk
;
1834 struct xfs_da_blkinfo
*info
;
1835 struct xfs_da_intnode
*node
;
1836 struct xfs_da_args
*args
;
1837 struct xfs_da_node_entry
*btree
;
1838 struct xfs_da3_icnode_hdr nodehdr
;
1840 xfs_dablk_t blkno
= 0;
1843 struct xfs_inode
*dp
= state
->args
->dp
;
1845 trace_xfs_da_path_shift(state
->args
);
1848 * Roll up the Btree looking for the first block where our
1849 * current index is not at the edge of the block. Note that
1850 * we skip the bottom layer because we want the sibling block.
1853 ASSERT(args
!= NULL
);
1854 ASSERT(path
!= NULL
);
1855 ASSERT((path
->active
> 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1856 level
= (path
->active
-1) - 1; /* skip bottom layer in path */
1857 for (blk
= &path
->blk
[level
]; level
>= 0; blk
--, level
--) {
1858 node
= blk
->bp
->b_addr
;
1859 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1860 btree
= dp
->d_ops
->node_tree_p(node
);
1862 if (forward
&& (blk
->index
< nodehdr
.count
- 1)) {
1864 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1866 } else if (!forward
&& (blk
->index
> 0)) {
1868 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1873 *result
= -ENOENT
; /* we're out of our tree */
1874 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1879 * Roll down the edge of the subtree until we reach the
1880 * same depth we were at originally.
1882 for (blk
++, level
++; level
< path
->active
; blk
++, level
++) {
1884 * Read the next child block into a local buffer.
1886 error
= xfs_da3_node_read(args
->trans
, dp
, blkno
, -1, &bp
,
1892 * Release the old block (if it's dirty, the trans doesn't
1893 * actually let go) and swap the local buffer into the path
1894 * structure. This ensures failure of the above read doesn't set
1895 * a NULL buffer in an active slot in the path.
1898 xfs_trans_brelse(args
->trans
, blk
->bp
);
1902 info
= blk
->bp
->b_addr
;
1903 ASSERT(info
->magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1904 info
->magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
1905 info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1906 info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1907 info
->magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1908 info
->magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1912 * Note: we flatten the magic number to a single type so we
1913 * don't have to compare against crc/non-crc types elsewhere.
1915 switch (be16_to_cpu(info
->magic
)) {
1916 case XFS_DA_NODE_MAGIC
:
1917 case XFS_DA3_NODE_MAGIC
:
1918 blk
->magic
= XFS_DA_NODE_MAGIC
;
1919 node
= (xfs_da_intnode_t
*)info
;
1920 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1921 btree
= dp
->d_ops
->node_tree_p(node
);
1922 blk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1926 blk
->index
= nodehdr
.count
- 1;
1927 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1929 case XFS_ATTR_LEAF_MAGIC
:
1930 case XFS_ATTR3_LEAF_MAGIC
:
1931 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1932 ASSERT(level
== path
->active
-1);
1934 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1936 case XFS_DIR2_LEAFN_MAGIC
:
1937 case XFS_DIR3_LEAFN_MAGIC
:
1938 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1939 ASSERT(level
== path
->active
-1);
1941 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1954 /*========================================================================
1956 *========================================================================*/
1959 * Implement a simple hash on a character string.
1960 * Rotate the hash value by 7 bits, then XOR each character in.
1961 * This is implemented with some source-level loop unrolling.
1964 xfs_da_hashname(const uint8_t *name
, int namelen
)
1969 * Do four characters at a time as long as we can.
1971 for (hash
= 0; namelen
>= 4; namelen
-= 4, name
+= 4)
1972 hash
= (name
[0] << 21) ^ (name
[1] << 14) ^ (name
[2] << 7) ^
1973 (name
[3] << 0) ^ rol32(hash
, 7 * 4);
1976 * Now do the rest of the characters.
1980 return (name
[0] << 14) ^ (name
[1] << 7) ^ (name
[2] << 0) ^
1983 return (name
[0] << 7) ^ (name
[1] << 0) ^ rol32(hash
, 7 * 2);
1985 return (name
[0] << 0) ^ rol32(hash
, 7 * 1);
1986 default: /* case 0: */
1993 struct xfs_da_args
*args
,
1994 const unsigned char *name
,
1997 return (args
->namelen
== len
&& memcmp(args
->name
, name
, len
) == 0) ?
1998 XFS_CMP_EXACT
: XFS_CMP_DIFFERENT
;
2002 xfs_default_hashname(
2003 struct xfs_name
*name
)
2005 return xfs_da_hashname(name
->name
, name
->len
);
2008 const struct xfs_nameops xfs_default_nameops
= {
2009 .hashname
= xfs_default_hashname
,
2010 .compname
= xfs_da_compname
2014 xfs_da_grow_inode_int(
2015 struct xfs_da_args
*args
,
2019 struct xfs_trans
*tp
= args
->trans
;
2020 struct xfs_inode
*dp
= args
->dp
;
2021 int w
= args
->whichfork
;
2022 xfs_rfsblock_t nblks
= dp
->i_d
.di_nblocks
;
2023 struct xfs_bmbt_irec map
, *mapp
;
2024 int nmap
, error
, got
, i
, mapi
;
2027 * Find a spot in the file space to put the new block.
2029 error
= xfs_bmap_first_unused(tp
, dp
, count
, bno
, w
);
2034 * Try mapping it in one filesystem block.
2037 ASSERT(args
->firstblock
!= NULL
);
2038 error
= xfs_bmapi_write(tp
, dp
, *bno
, count
,
2039 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
|XFS_BMAPI_CONTIG
,
2040 args
->firstblock
, args
->total
, &map
, &nmap
,
2049 } else if (nmap
== 0 && count
> 1) {
2054 * If we didn't get it and the block might work if fragmented,
2055 * try without the CONTIG flag. Loop until we get it all.
2057 mapp
= kmem_alloc(sizeof(*mapp
) * count
, KM_SLEEP
);
2058 for (b
= *bno
, mapi
= 0; b
< *bno
+ count
; ) {
2059 nmap
= MIN(XFS_BMAP_MAX_NMAP
, count
);
2060 c
= (int)(*bno
+ count
- b
);
2061 error
= xfs_bmapi_write(tp
, dp
, b
, c
,
2062 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
,
2063 args
->firstblock
, args
->total
,
2064 &mapp
[mapi
], &nmap
, args
->dfops
);
2070 b
= mapp
[mapi
- 1].br_startoff
+
2071 mapp
[mapi
- 1].br_blockcount
;
2079 * Count the blocks we got, make sure it matches the total.
2081 for (i
= 0, got
= 0; i
< mapi
; i
++)
2082 got
+= mapp
[i
].br_blockcount
;
2083 if (got
!= count
|| mapp
[0].br_startoff
!= *bno
||
2084 mapp
[mapi
- 1].br_startoff
+ mapp
[mapi
- 1].br_blockcount
!=
2090 /* account for newly allocated blocks in reserved blocks total */
2091 args
->total
-= dp
->i_d
.di_nblocks
- nblks
;
2100 * Add a block to the btree ahead of the file.
2101 * Return the new block number to the caller.
2105 struct xfs_da_args
*args
,
2106 xfs_dablk_t
*new_blkno
)
2111 trace_xfs_da_grow_inode(args
);
2113 bno
= args
->geo
->leafblk
;
2114 error
= xfs_da_grow_inode_int(args
, &bno
, args
->geo
->fsbcount
);
2116 *new_blkno
= (xfs_dablk_t
)bno
;
2121 * Ick. We need to always be able to remove a btree block, even
2122 * if there's no space reservation because the filesystem is full.
2123 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2124 * It swaps the target block with the last block in the file. The
2125 * last block in the file can always be removed since it can't cause
2126 * a bmap btree split to do that.
2129 xfs_da3_swap_lastblock(
2130 struct xfs_da_args
*args
,
2131 xfs_dablk_t
*dead_blknop
,
2132 struct xfs_buf
**dead_bufp
)
2134 struct xfs_da_blkinfo
*dead_info
;
2135 struct xfs_da_blkinfo
*sib_info
;
2136 struct xfs_da_intnode
*par_node
;
2137 struct xfs_da_intnode
*dead_node
;
2138 struct xfs_dir2_leaf
*dead_leaf2
;
2139 struct xfs_da_node_entry
*btree
;
2140 struct xfs_da3_icnode_hdr par_hdr
;
2141 struct xfs_inode
*dp
;
2142 struct xfs_trans
*tp
;
2143 struct xfs_mount
*mp
;
2144 struct xfs_buf
*dead_buf
;
2145 struct xfs_buf
*last_buf
;
2146 struct xfs_buf
*sib_buf
;
2147 struct xfs_buf
*par_buf
;
2148 xfs_dahash_t dead_hash
;
2149 xfs_fileoff_t lastoff
;
2150 xfs_dablk_t dead_blkno
;
2151 xfs_dablk_t last_blkno
;
2152 xfs_dablk_t sib_blkno
;
2153 xfs_dablk_t par_blkno
;
2160 trace_xfs_da_swap_lastblock(args
);
2162 dead_buf
= *dead_bufp
;
2163 dead_blkno
= *dead_blknop
;
2166 w
= args
->whichfork
;
2167 ASSERT(w
== XFS_DATA_FORK
);
2169 lastoff
= args
->geo
->freeblk
;
2170 error
= xfs_bmap_last_before(tp
, dp
, &lastoff
, w
);
2173 if (unlikely(lastoff
== 0)) {
2174 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW
,
2176 return -EFSCORRUPTED
;
2179 * Read the last block in the btree space.
2181 last_blkno
= (xfs_dablk_t
)lastoff
- args
->geo
->fsbcount
;
2182 error
= xfs_da3_node_read(tp
, dp
, last_blkno
, -1, &last_buf
, w
);
2186 * Copy the last block into the dead buffer and log it.
2188 memcpy(dead_buf
->b_addr
, last_buf
->b_addr
, args
->geo
->blksize
);
2189 xfs_trans_log_buf(tp
, dead_buf
, 0, args
->geo
->blksize
- 1);
2190 dead_info
= dead_buf
->b_addr
;
2192 * Get values from the moved block.
2194 if (dead_info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
2195 dead_info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
2196 struct xfs_dir3_icleaf_hdr leafhdr
;
2197 struct xfs_dir2_leaf_entry
*ents
;
2199 dead_leaf2
= (xfs_dir2_leaf_t
*)dead_info
;
2200 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, dead_leaf2
);
2201 ents
= dp
->d_ops
->leaf_ents_p(dead_leaf2
);
2203 dead_hash
= be32_to_cpu(ents
[leafhdr
.count
- 1].hashval
);
2205 struct xfs_da3_icnode_hdr deadhdr
;
2207 dead_node
= (xfs_da_intnode_t
*)dead_info
;
2208 dp
->d_ops
->node_hdr_from_disk(&deadhdr
, dead_node
);
2209 btree
= dp
->d_ops
->node_tree_p(dead_node
);
2210 dead_level
= deadhdr
.level
;
2211 dead_hash
= be32_to_cpu(btree
[deadhdr
.count
- 1].hashval
);
2213 sib_buf
= par_buf
= NULL
;
2215 * If the moved block has a left sibling, fix up the pointers.
2217 if ((sib_blkno
= be32_to_cpu(dead_info
->back
))) {
2218 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2221 sib_info
= sib_buf
->b_addr
;
2223 be32_to_cpu(sib_info
->forw
) != last_blkno
||
2224 sib_info
->magic
!= dead_info
->magic
)) {
2225 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2226 XFS_ERRLEVEL_LOW
, mp
);
2227 error
= -EFSCORRUPTED
;
2230 sib_info
->forw
= cpu_to_be32(dead_blkno
);
2231 xfs_trans_log_buf(tp
, sib_buf
,
2232 XFS_DA_LOGRANGE(sib_info
, &sib_info
->forw
,
2233 sizeof(sib_info
->forw
)));
2237 * If the moved block has a right sibling, fix up the pointers.
2239 if ((sib_blkno
= be32_to_cpu(dead_info
->forw
))) {
2240 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2243 sib_info
= sib_buf
->b_addr
;
2245 be32_to_cpu(sib_info
->back
) != last_blkno
||
2246 sib_info
->magic
!= dead_info
->magic
)) {
2247 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2248 XFS_ERRLEVEL_LOW
, mp
);
2249 error
= -EFSCORRUPTED
;
2252 sib_info
->back
= cpu_to_be32(dead_blkno
);
2253 xfs_trans_log_buf(tp
, sib_buf
,
2254 XFS_DA_LOGRANGE(sib_info
, &sib_info
->back
,
2255 sizeof(sib_info
->back
)));
2258 par_blkno
= args
->geo
->leafblk
;
2261 * Walk down the tree looking for the parent of the moved block.
2264 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2267 par_node
= par_buf
->b_addr
;
2268 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2269 if (level
>= 0 && level
!= par_hdr
.level
+ 1) {
2270 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2271 XFS_ERRLEVEL_LOW
, mp
);
2272 error
= -EFSCORRUPTED
;
2275 level
= par_hdr
.level
;
2276 btree
= dp
->d_ops
->node_tree_p(par_node
);
2278 entno
< par_hdr
.count
&&
2279 be32_to_cpu(btree
[entno
].hashval
) < dead_hash
;
2282 if (entno
== par_hdr
.count
) {
2283 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2284 XFS_ERRLEVEL_LOW
, mp
);
2285 error
= -EFSCORRUPTED
;
2288 par_blkno
= be32_to_cpu(btree
[entno
].before
);
2289 if (level
== dead_level
+ 1)
2291 xfs_trans_brelse(tp
, par_buf
);
2295 * We're in the right parent block.
2296 * Look for the right entry.
2300 entno
< par_hdr
.count
&&
2301 be32_to_cpu(btree
[entno
].before
) != last_blkno
;
2304 if (entno
< par_hdr
.count
)
2306 par_blkno
= par_hdr
.forw
;
2307 xfs_trans_brelse(tp
, par_buf
);
2309 if (unlikely(par_blkno
== 0)) {
2310 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2311 XFS_ERRLEVEL_LOW
, mp
);
2312 error
= -EFSCORRUPTED
;
2315 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2318 par_node
= par_buf
->b_addr
;
2319 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2320 if (par_hdr
.level
!= level
) {
2321 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2322 XFS_ERRLEVEL_LOW
, mp
);
2323 error
= -EFSCORRUPTED
;
2326 btree
= dp
->d_ops
->node_tree_p(par_node
);
2330 * Update the parent entry pointing to the moved block.
2332 btree
[entno
].before
= cpu_to_be32(dead_blkno
);
2333 xfs_trans_log_buf(tp
, par_buf
,
2334 XFS_DA_LOGRANGE(par_node
, &btree
[entno
].before
,
2335 sizeof(btree
[entno
].before
)));
2336 *dead_blknop
= last_blkno
;
2337 *dead_bufp
= last_buf
;
2341 xfs_trans_brelse(tp
, par_buf
);
2343 xfs_trans_brelse(tp
, sib_buf
);
2344 xfs_trans_brelse(tp
, last_buf
);
2349 * Remove a btree block from a directory or attribute.
2352 xfs_da_shrink_inode(
2353 xfs_da_args_t
*args
,
2354 xfs_dablk_t dead_blkno
,
2355 struct xfs_buf
*dead_buf
)
2358 int done
, error
, w
, count
;
2361 trace_xfs_da_shrink_inode(args
);
2364 w
= args
->whichfork
;
2366 count
= args
->geo
->fsbcount
;
2369 * Remove extents. If we get ENOSPC for a dir we have to move
2370 * the last block to the place we want to kill.
2372 error
= xfs_bunmapi(tp
, dp
, dead_blkno
, count
,
2373 xfs_bmapi_aflag(w
), 0, args
->firstblock
,
2374 args
->dfops
, &done
);
2375 if (error
== -ENOSPC
) {
2376 if (w
!= XFS_DATA_FORK
)
2378 error
= xfs_da3_swap_lastblock(args
, &dead_blkno
,
2386 xfs_trans_binval(tp
, dead_buf
);
2391 * See if the mapping(s) for this btree block are valid, i.e.
2392 * don't contain holes, are logically contiguous, and cover the whole range.
2395 xfs_da_map_covers_blocks(
2397 xfs_bmbt_irec_t
*mapp
,
2404 for (i
= 0, off
= bno
; i
< nmap
; i
++) {
2405 if (mapp
[i
].br_startblock
== HOLESTARTBLOCK
||
2406 mapp
[i
].br_startblock
== DELAYSTARTBLOCK
) {
2409 if (off
!= mapp
[i
].br_startoff
) {
2412 off
+= mapp
[i
].br_blockcount
;
2414 return off
== bno
+ count
;
2418 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2420 * For the single map case, it is assumed that the caller has provided a pointer
2421 * to a valid xfs_buf_map. For the multiple map case, this function will
2422 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2423 * map pointer with the allocated map.
2426 xfs_buf_map_from_irec(
2427 struct xfs_mount
*mp
,
2428 struct xfs_buf_map
**mapp
,
2430 struct xfs_bmbt_irec
*irecs
,
2433 struct xfs_buf_map
*map
;
2436 ASSERT(*nmaps
== 1);
2437 ASSERT(nirecs
>= 1);
2440 map
= kmem_zalloc(nirecs
* sizeof(struct xfs_buf_map
),
2441 KM_SLEEP
| KM_NOFS
);
2449 for (i
= 0; i
< *nmaps
; i
++) {
2450 ASSERT(irecs
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
2451 irecs
[i
].br_startblock
!= HOLESTARTBLOCK
);
2452 map
[i
].bm_bn
= XFS_FSB_TO_DADDR(mp
, irecs
[i
].br_startblock
);
2453 map
[i
].bm_len
= XFS_FSB_TO_BB(mp
, irecs
[i
].br_blockcount
);
2459 * Map the block we are given ready for reading. There are three possible return
2461 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2462 * caller knows not to execute a subsequent read.
2463 * 0 - if we mapped the block successfully
2464 * >0 - positive error number if there was an error.
2468 struct xfs_inode
*dp
,
2470 xfs_daddr_t mappedbno
,
2472 struct xfs_buf_map
**map
,
2475 struct xfs_mount
*mp
= dp
->i_mount
;
2478 struct xfs_bmbt_irec irec
;
2479 struct xfs_bmbt_irec
*irecs
= &irec
;
2482 ASSERT(map
&& *map
);
2483 ASSERT(*nmaps
== 1);
2485 if (whichfork
== XFS_DATA_FORK
)
2486 nfsb
= mp
->m_dir_geo
->fsbcount
;
2488 nfsb
= mp
->m_attr_geo
->fsbcount
;
2491 * Caller doesn't have a mapping. -2 means don't complain
2492 * if we land in a hole.
2494 if (mappedbno
== -1 || mappedbno
== -2) {
2496 * Optimize the one-block case.
2499 irecs
= kmem_zalloc(sizeof(irec
) * nfsb
,
2500 KM_SLEEP
| KM_NOFS
);
2503 error
= xfs_bmapi_read(dp
, (xfs_fileoff_t
)bno
, nfsb
, irecs
,
2504 &nirecs
, xfs_bmapi_aflag(whichfork
));
2508 irecs
->br_startblock
= XFS_DADDR_TO_FSB(mp
, mappedbno
);
2509 irecs
->br_startoff
= (xfs_fileoff_t
)bno
;
2510 irecs
->br_blockcount
= nfsb
;
2511 irecs
->br_state
= 0;
2515 if (!xfs_da_map_covers_blocks(nirecs
, irecs
, bno
, nfsb
)) {
2516 error
= mappedbno
== -2 ? -1 : -EFSCORRUPTED
;
2517 if (unlikely(error
== -EFSCORRUPTED
)) {
2518 if (xfs_error_level
>= XFS_ERRLEVEL_LOW
) {
2520 xfs_alert(mp
, "%s: bno %lld dir: inode %lld",
2521 __func__
, (long long)bno
,
2522 (long long)dp
->i_ino
);
2523 for (i
= 0; i
< *nmaps
; i
++) {
2525 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2527 (long long)irecs
[i
].br_startoff
,
2528 (long long)irecs
[i
].br_startblock
,
2529 (long long)irecs
[i
].br_blockcount
,
2533 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2534 XFS_ERRLEVEL_LOW
, mp
);
2538 error
= xfs_buf_map_from_irec(mp
, map
, nmaps
, irecs
, nirecs
);
2546 * Get a buffer for the dir/attr block.
2550 struct xfs_trans
*trans
,
2551 struct xfs_inode
*dp
,
2553 xfs_daddr_t mappedbno
,
2554 struct xfs_buf
**bpp
,
2558 struct xfs_buf_map map
;
2559 struct xfs_buf_map
*mapp
;
2566 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2569 /* mapping a hole is not an error, but we don't continue */
2575 bp
= xfs_trans_get_buf_map(trans
, dp
->i_mount
->m_ddev_targp
,
2577 error
= bp
? bp
->b_error
: -EIO
;
2580 xfs_trans_brelse(trans
, bp
);
2594 * Get a buffer for the dir/attr block, fill in the contents.
2598 struct xfs_trans
*trans
,
2599 struct xfs_inode
*dp
,
2601 xfs_daddr_t mappedbno
,
2602 struct xfs_buf
**bpp
,
2604 const struct xfs_buf_ops
*ops
)
2607 struct xfs_buf_map map
;
2608 struct xfs_buf_map
*mapp
;
2615 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2618 /* mapping a hole is not an error, but we don't continue */
2624 error
= xfs_trans_read_buf_map(dp
->i_mount
, trans
,
2625 dp
->i_mount
->m_ddev_targp
,
2626 mapp
, nmap
, 0, &bp
, ops
);
2630 if (whichfork
== XFS_ATTR_FORK
)
2631 xfs_buf_set_ref(bp
, XFS_ATTR_BTREE_REF
);
2633 xfs_buf_set_ref(bp
, XFS_DIR_BTREE_REF
);
2643 * Readahead the dir/attr block.
2647 struct xfs_inode
*dp
,
2649 xfs_daddr_t mappedbno
,
2651 const struct xfs_buf_ops
*ops
)
2653 struct xfs_buf_map map
;
2654 struct xfs_buf_map
*mapp
;
2660 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2663 /* mapping a hole is not an error, but we don't continue */
2669 mappedbno
= mapp
[0].bm_bn
;
2670 xfs_buf_readahead_map(dp
->i_mount
->m_ddev_targp
, mapp
, nmap
, ops
);