1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
16 #include "xfs_dir2_priv.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
20 #include "xfs_attr_leaf.h"
21 #include "xfs_error.h"
22 #include "xfs_trace.h"
23 #include "xfs_buf_item.h"
29 * Routines to implement directories as Btrees of hashed names.
32 /*========================================================================
33 * Function prototypes for the kernel.
34 *========================================================================*/
37 * Routines used for growing the Btree.
39 STATIC
int xfs_da3_root_split(xfs_da_state_t
*state
,
40 xfs_da_state_blk_t
*existing_root
,
41 xfs_da_state_blk_t
*new_child
);
42 STATIC
int xfs_da3_node_split(xfs_da_state_t
*state
,
43 xfs_da_state_blk_t
*existing_blk
,
44 xfs_da_state_blk_t
*split_blk
,
45 xfs_da_state_blk_t
*blk_to_add
,
48 STATIC
void xfs_da3_node_rebalance(xfs_da_state_t
*state
,
49 xfs_da_state_blk_t
*node_blk_1
,
50 xfs_da_state_blk_t
*node_blk_2
);
51 STATIC
void xfs_da3_node_add(xfs_da_state_t
*state
,
52 xfs_da_state_blk_t
*old_node_blk
,
53 xfs_da_state_blk_t
*new_node_blk
);
56 * Routines used for shrinking the Btree.
58 STATIC
int xfs_da3_root_join(xfs_da_state_t
*state
,
59 xfs_da_state_blk_t
*root_blk
);
60 STATIC
int xfs_da3_node_toosmall(xfs_da_state_t
*state
, int *retval
);
61 STATIC
void xfs_da3_node_remove(xfs_da_state_t
*state
,
62 xfs_da_state_blk_t
*drop_blk
);
63 STATIC
void xfs_da3_node_unbalance(xfs_da_state_t
*state
,
64 xfs_da_state_blk_t
*src_node_blk
,
65 xfs_da_state_blk_t
*dst_node_blk
);
70 STATIC
int xfs_da3_blk_unlink(xfs_da_state_t
*state
,
71 xfs_da_state_blk_t
*drop_blk
,
72 xfs_da_state_blk_t
*save_blk
);
75 kmem_zone_t
*xfs_da_state_zone
; /* anchor for state struct zone */
78 * Allocate a dir-state structure.
79 * We don't put them on the stack since they're large.
82 xfs_da_state_alloc(void)
84 return kmem_zone_zalloc(xfs_da_state_zone
, KM_NOFS
);
88 * Kill the altpath contents of a da-state structure.
91 xfs_da_state_kill_altpath(xfs_da_state_t
*state
)
95 for (i
= 0; i
< state
->altpath
.active
; i
++)
96 state
->altpath
.blk
[i
].bp
= NULL
;
97 state
->altpath
.active
= 0;
101 * Free a da-state structure.
104 xfs_da_state_free(xfs_da_state_t
*state
)
106 xfs_da_state_kill_altpath(state
);
108 memset((char *)state
, 0, sizeof(*state
));
110 kmem_zone_free(xfs_da_state_zone
, state
);
114 * Verify an xfs_da3_blkinfo structure. Note that the da3 fields are only
115 * accessible on v5 filesystems. This header format is common across da node,
116 * attr leaf and dir leaf blocks.
119 xfs_da3_blkinfo_verify(
121 struct xfs_da3_blkinfo
*hdr3
)
123 struct xfs_mount
*mp
= bp
->b_mount
;
124 struct xfs_da_blkinfo
*hdr
= &hdr3
->hdr
;
126 if (!xfs_verify_magic16(bp
, hdr
->magic
))
127 return __this_address
;
129 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
130 if (!uuid_equal(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
))
131 return __this_address
;
132 if (be64_to_cpu(hdr3
->blkno
) != bp
->b_bn
)
133 return __this_address
;
134 if (!xfs_log_check_lsn(mp
, be64_to_cpu(hdr3
->lsn
)))
135 return __this_address
;
141 static xfs_failaddr_t
145 struct xfs_mount
*mp
= bp
->b_mount
;
146 struct xfs_da_intnode
*hdr
= bp
->b_addr
;
147 struct xfs_da3_icnode_hdr ichdr
;
148 const struct xfs_dir_ops
*ops
;
151 ops
= xfs_dir_get_ops(mp
, NULL
);
153 ops
->node_hdr_from_disk(&ichdr
, hdr
);
155 fa
= xfs_da3_blkinfo_verify(bp
, bp
->b_addr
);
159 if (ichdr
.level
== 0)
160 return __this_address
;
161 if (ichdr
.level
> XFS_DA_NODE_MAXDEPTH
)
162 return __this_address
;
163 if (ichdr
.count
== 0)
164 return __this_address
;
167 * we don't know if the node is for and attribute or directory tree,
168 * so only fail if the count is outside both bounds
170 if (ichdr
.count
> mp
->m_dir_geo
->node_ents
&&
171 ichdr
.count
> mp
->m_attr_geo
->node_ents
)
172 return __this_address
;
174 /* XXX: hash order check? */
180 xfs_da3_node_write_verify(
183 struct xfs_mount
*mp
= bp
->b_mount
;
184 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
185 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
188 fa
= xfs_da3_node_verify(bp
);
190 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
194 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
198 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
200 xfs_buf_update_cksum(bp
, XFS_DA3_NODE_CRC_OFF
);
204 * leaf/node format detection on trees is sketchy, so a node read can be done on
205 * leaf level blocks when detection identifies the tree as a node format tree
206 * incorrectly. In this case, we need to swap the verifier to match the correct
207 * format of the block being read.
210 xfs_da3_node_read_verify(
213 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
216 switch (be16_to_cpu(info
->magic
)) {
217 case XFS_DA3_NODE_MAGIC
:
218 if (!xfs_buf_verify_cksum(bp
, XFS_DA3_NODE_CRC_OFF
)) {
219 xfs_verifier_error(bp
, -EFSBADCRC
,
224 case XFS_DA_NODE_MAGIC
:
225 fa
= xfs_da3_node_verify(bp
);
227 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
229 case XFS_ATTR_LEAF_MAGIC
:
230 case XFS_ATTR3_LEAF_MAGIC
:
231 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
232 bp
->b_ops
->verify_read(bp
);
234 case XFS_DIR2_LEAFN_MAGIC
:
235 case XFS_DIR3_LEAFN_MAGIC
:
236 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
237 bp
->b_ops
->verify_read(bp
);
240 xfs_verifier_error(bp
, -EFSCORRUPTED
, __this_address
);
245 /* Verify the structure of a da3 block. */
246 static xfs_failaddr_t
247 xfs_da3_node_verify_struct(
250 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
252 switch (be16_to_cpu(info
->magic
)) {
253 case XFS_DA3_NODE_MAGIC
:
254 case XFS_DA_NODE_MAGIC
:
255 return xfs_da3_node_verify(bp
);
256 case XFS_ATTR_LEAF_MAGIC
:
257 case XFS_ATTR3_LEAF_MAGIC
:
258 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
259 return bp
->b_ops
->verify_struct(bp
);
260 case XFS_DIR2_LEAFN_MAGIC
:
261 case XFS_DIR3_LEAFN_MAGIC
:
262 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
263 return bp
->b_ops
->verify_struct(bp
);
265 return __this_address
;
269 const struct xfs_buf_ops xfs_da3_node_buf_ops
= {
270 .name
= "xfs_da3_node",
271 .magic16
= { cpu_to_be16(XFS_DA_NODE_MAGIC
),
272 cpu_to_be16(XFS_DA3_NODE_MAGIC
) },
273 .verify_read
= xfs_da3_node_read_verify
,
274 .verify_write
= xfs_da3_node_write_verify
,
275 .verify_struct
= xfs_da3_node_verify_struct
,
280 struct xfs_trans
*tp
,
281 struct xfs_inode
*dp
,
283 xfs_daddr_t mappedbno
,
284 struct xfs_buf
**bpp
,
289 err
= xfs_da_read_buf(tp
, dp
, bno
, mappedbno
, bpp
,
290 which_fork
, &xfs_da3_node_buf_ops
);
291 if (!err
&& tp
&& *bpp
) {
292 struct xfs_da_blkinfo
*info
= (*bpp
)->b_addr
;
295 switch (be16_to_cpu(info
->magic
)) {
296 case XFS_DA_NODE_MAGIC
:
297 case XFS_DA3_NODE_MAGIC
:
298 type
= XFS_BLFT_DA_NODE_BUF
;
300 case XFS_ATTR_LEAF_MAGIC
:
301 case XFS_ATTR3_LEAF_MAGIC
:
302 type
= XFS_BLFT_ATTR_LEAF_BUF
;
304 case XFS_DIR2_LEAFN_MAGIC
:
305 case XFS_DIR3_LEAFN_MAGIC
:
306 type
= XFS_BLFT_DIR_LEAFN_BUF
;
309 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
,
310 tp
->t_mountp
, info
, sizeof(*info
));
311 xfs_trans_brelse(tp
, *bpp
);
313 return -EFSCORRUPTED
;
315 xfs_trans_buf_set_type(tp
, *bpp
, type
);
320 /*========================================================================
321 * Routines used for growing the Btree.
322 *========================================================================*/
325 * Create the initial contents of an intermediate node.
329 struct xfs_da_args
*args
,
332 struct xfs_buf
**bpp
,
335 struct xfs_da_intnode
*node
;
336 struct xfs_trans
*tp
= args
->trans
;
337 struct xfs_mount
*mp
= tp
->t_mountp
;
338 struct xfs_da3_icnode_hdr ichdr
= {0};
341 struct xfs_inode
*dp
= args
->dp
;
343 trace_xfs_da_node_create(args
);
344 ASSERT(level
<= XFS_DA_NODE_MAXDEPTH
);
346 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, whichfork
);
349 bp
->b_ops
= &xfs_da3_node_buf_ops
;
350 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
353 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
354 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
356 memset(hdr3
, 0, sizeof(struct xfs_da3_node_hdr
));
357 ichdr
.magic
= XFS_DA3_NODE_MAGIC
;
358 hdr3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
359 hdr3
->info
.owner
= cpu_to_be64(args
->dp
->i_ino
);
360 uuid_copy(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_meta_uuid
);
362 ichdr
.magic
= XFS_DA_NODE_MAGIC
;
366 dp
->d_ops
->node_hdr_to_disk(node
, &ichdr
);
367 xfs_trans_log_buf(tp
, bp
,
368 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
375 * Split a leaf node, rebalance, then possibly split
376 * intermediate nodes, rebalance, etc.
380 struct xfs_da_state
*state
)
382 struct xfs_da_state_blk
*oldblk
;
383 struct xfs_da_state_blk
*newblk
;
384 struct xfs_da_state_blk
*addblk
;
385 struct xfs_da_intnode
*node
;
391 trace_xfs_da_split(state
->args
);
394 * Walk back up the tree splitting/inserting/adjusting as necessary.
395 * If we need to insert and there isn't room, split the node, then
396 * decide which fragment to insert the new block from below into.
397 * Note that we may split the root this way, but we need more fixup.
399 max
= state
->path
.active
- 1;
400 ASSERT((max
>= 0) && (max
< XFS_DA_NODE_MAXDEPTH
));
401 ASSERT(state
->path
.blk
[max
].magic
== XFS_ATTR_LEAF_MAGIC
||
402 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
404 addblk
= &state
->path
.blk
[max
]; /* initial dummy value */
405 for (i
= max
; (i
>= 0) && addblk
; state
->path
.active
--, i
--) {
406 oldblk
= &state
->path
.blk
[i
];
407 newblk
= &state
->altpath
.blk
[i
];
410 * If a leaf node then
411 * Allocate a new leaf node, then rebalance across them.
412 * else if an intermediate node then
413 * We split on the last layer, must we split the node?
415 switch (oldblk
->magic
) {
416 case XFS_ATTR_LEAF_MAGIC
:
417 error
= xfs_attr3_leaf_split(state
, oldblk
, newblk
);
418 if ((error
!= 0) && (error
!= -ENOSPC
)) {
419 return error
; /* GROT: attr is inconsistent */
426 * Entry wouldn't fit, split the leaf again. The new
427 * extrablk will be consumed by xfs_da3_node_split if
430 state
->extravalid
= 1;
432 state
->extraafter
= 0; /* before newblk */
433 trace_xfs_attr_leaf_split_before(state
->args
);
434 error
= xfs_attr3_leaf_split(state
, oldblk
,
437 state
->extraafter
= 1; /* after newblk */
438 trace_xfs_attr_leaf_split_after(state
->args
);
439 error
= xfs_attr3_leaf_split(state
, newblk
,
443 return error
; /* GROT: attr inconsistent */
446 case XFS_DIR2_LEAFN_MAGIC
:
447 error
= xfs_dir2_leafn_split(state
, oldblk
, newblk
);
452 case XFS_DA_NODE_MAGIC
:
453 error
= xfs_da3_node_split(state
, oldblk
, newblk
, addblk
,
457 return error
; /* GROT: dir is inconsistent */
459 * Record the newly split block for the next time thru?
469 * Update the btree to show the new hashval for this child.
471 xfs_da3_fixhashpath(state
, &state
->path
);
477 * xfs_da3_node_split() should have consumed any extra blocks we added
478 * during a double leaf split in the attr fork. This is guaranteed as
479 * we can't be here if the attr fork only has a single leaf block.
481 ASSERT(state
->extravalid
== 0 ||
482 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
485 * Split the root node.
487 ASSERT(state
->path
.active
== 0);
488 oldblk
= &state
->path
.blk
[0];
489 error
= xfs_da3_root_split(state
, oldblk
, addblk
);
492 return error
; /* GROT: dir is inconsistent */
496 * Update pointers to the node which used to be block 0 and just got
497 * bumped because of the addition of a new root node. Note that the
498 * original block 0 could be at any position in the list of blocks in
501 * Note: the magic numbers and sibling pointers are in the same physical
502 * place for both v2 and v3 headers (by design). Hence it doesn't matter
503 * which version of the xfs_da_intnode structure we use here as the
504 * result will be the same using either structure.
506 node
= oldblk
->bp
->b_addr
;
507 if (node
->hdr
.info
.forw
) {
508 ASSERT(be32_to_cpu(node
->hdr
.info
.forw
) == addblk
->blkno
);
509 node
= addblk
->bp
->b_addr
;
510 node
->hdr
.info
.back
= cpu_to_be32(oldblk
->blkno
);
511 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
512 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
513 sizeof(node
->hdr
.info
)));
515 node
= oldblk
->bp
->b_addr
;
516 if (node
->hdr
.info
.back
) {
517 ASSERT(be32_to_cpu(node
->hdr
.info
.back
) == addblk
->blkno
);
518 node
= addblk
->bp
->b_addr
;
519 node
->hdr
.info
.forw
= cpu_to_be32(oldblk
->blkno
);
520 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
521 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
522 sizeof(node
->hdr
.info
)));
529 * Split the root. We have to create a new root and point to the two
530 * parts (the split old root) that we just created. Copy block zero to
531 * the EOF, extending the inode in process.
533 STATIC
int /* error */
535 struct xfs_da_state
*state
,
536 struct xfs_da_state_blk
*blk1
,
537 struct xfs_da_state_blk
*blk2
)
539 struct xfs_da_intnode
*node
;
540 struct xfs_da_intnode
*oldroot
;
541 struct xfs_da_node_entry
*btree
;
542 struct xfs_da3_icnode_hdr nodehdr
;
543 struct xfs_da_args
*args
;
545 struct xfs_inode
*dp
;
546 struct xfs_trans
*tp
;
547 struct xfs_dir2_leaf
*leaf
;
553 trace_xfs_da_root_split(state
->args
);
556 * Copy the existing (incorrect) block from the root node position
557 * to a free space somewhere.
560 error
= xfs_da_grow_inode(args
, &blkno
);
566 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, args
->whichfork
);
570 oldroot
= blk1
->bp
->b_addr
;
571 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
572 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
)) {
573 struct xfs_da3_icnode_hdr icnodehdr
;
575 dp
->d_ops
->node_hdr_from_disk(&icnodehdr
, oldroot
);
576 btree
= dp
->d_ops
->node_tree_p(oldroot
);
577 size
= (int)((char *)&btree
[icnodehdr
.count
] - (char *)oldroot
);
578 level
= icnodehdr
.level
;
581 * we are about to copy oldroot to bp, so set up the type
582 * of bp while we know exactly what it will be.
584 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
586 struct xfs_dir3_icleaf_hdr leafhdr
;
587 struct xfs_dir2_leaf_entry
*ents
;
589 leaf
= (xfs_dir2_leaf_t
*)oldroot
;
590 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
591 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
593 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
594 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
595 size
= (int)((char *)&ents
[leafhdr
.count
] - (char *)leaf
);
599 * we are about to copy oldroot to bp, so set up the type
600 * of bp while we know exactly what it will be.
602 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAFN_BUF
);
606 * we can copy most of the information in the node from one block to
607 * another, but for CRC enabled headers we have to make sure that the
608 * block specific identifiers are kept intact. We update the buffer
611 memcpy(node
, oldroot
, size
);
612 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
613 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
614 struct xfs_da3_intnode
*node3
= (struct xfs_da3_intnode
*)node
;
616 node3
->hdr
.info
.blkno
= cpu_to_be64(bp
->b_bn
);
618 xfs_trans_log_buf(tp
, bp
, 0, size
- 1);
620 bp
->b_ops
= blk1
->bp
->b_ops
;
621 xfs_trans_buf_copy_type(bp
, blk1
->bp
);
626 * Set up the new root node.
628 error
= xfs_da3_node_create(args
,
629 (args
->whichfork
== XFS_DATA_FORK
) ? args
->geo
->leafblk
: 0,
630 level
+ 1, &bp
, args
->whichfork
);
635 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
636 btree
= dp
->d_ops
->node_tree_p(node
);
637 btree
[0].hashval
= cpu_to_be32(blk1
->hashval
);
638 btree
[0].before
= cpu_to_be32(blk1
->blkno
);
639 btree
[1].hashval
= cpu_to_be32(blk2
->hashval
);
640 btree
[1].before
= cpu_to_be32(blk2
->blkno
);
642 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
645 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
646 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
647 ASSERT(blk1
->blkno
>= args
->geo
->leafblk
&&
648 blk1
->blkno
< args
->geo
->freeblk
);
649 ASSERT(blk2
->blkno
>= args
->geo
->leafblk
&&
650 blk2
->blkno
< args
->geo
->freeblk
);
654 /* Header is already logged by xfs_da_node_create */
655 xfs_trans_log_buf(tp
, bp
,
656 XFS_DA_LOGRANGE(node
, btree
, sizeof(xfs_da_node_entry_t
) * 2));
662 * Split the node, rebalance, then add the new entry.
664 STATIC
int /* error */
666 struct xfs_da_state
*state
,
667 struct xfs_da_state_blk
*oldblk
,
668 struct xfs_da_state_blk
*newblk
,
669 struct xfs_da_state_blk
*addblk
,
673 struct xfs_da_intnode
*node
;
674 struct xfs_da3_icnode_hdr nodehdr
;
679 struct xfs_inode
*dp
= state
->args
->dp
;
681 trace_xfs_da_node_split(state
->args
);
683 node
= oldblk
->bp
->b_addr
;
684 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
687 * With V2 dirs the extra block is data or freespace.
689 useextra
= state
->extravalid
&& state
->args
->whichfork
== XFS_ATTR_FORK
;
690 newcount
= 1 + useextra
;
692 * Do we have to split the node?
694 if (nodehdr
.count
+ newcount
> state
->args
->geo
->node_ents
) {
696 * Allocate a new node, add to the doubly linked chain of
697 * nodes, then move some of our excess entries into it.
699 error
= xfs_da_grow_inode(state
->args
, &blkno
);
701 return error
; /* GROT: dir is inconsistent */
703 error
= xfs_da3_node_create(state
->args
, blkno
, treelevel
,
704 &newblk
->bp
, state
->args
->whichfork
);
706 return error
; /* GROT: dir is inconsistent */
707 newblk
->blkno
= blkno
;
708 newblk
->magic
= XFS_DA_NODE_MAGIC
;
709 xfs_da3_node_rebalance(state
, oldblk
, newblk
);
710 error
= xfs_da3_blk_link(state
, oldblk
, newblk
);
719 * Insert the new entry(s) into the correct block
720 * (updating last hashval in the process).
722 * xfs_da3_node_add() inserts BEFORE the given index,
723 * and as a result of using node_lookup_int() we always
724 * point to a valid entry (not after one), but a split
725 * operation always results in a new block whose hashvals
726 * FOLLOW the current block.
728 * If we had double-split op below us, then add the extra block too.
730 node
= oldblk
->bp
->b_addr
;
731 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
732 if (oldblk
->index
<= nodehdr
.count
) {
734 xfs_da3_node_add(state
, oldblk
, addblk
);
736 if (state
->extraafter
)
738 xfs_da3_node_add(state
, oldblk
, &state
->extrablk
);
739 state
->extravalid
= 0;
743 xfs_da3_node_add(state
, newblk
, addblk
);
745 if (state
->extraafter
)
747 xfs_da3_node_add(state
, newblk
, &state
->extrablk
);
748 state
->extravalid
= 0;
756 * Balance the btree elements between two intermediate nodes,
757 * usually one full and one empty.
759 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
762 xfs_da3_node_rebalance(
763 struct xfs_da_state
*state
,
764 struct xfs_da_state_blk
*blk1
,
765 struct xfs_da_state_blk
*blk2
)
767 struct xfs_da_intnode
*node1
;
768 struct xfs_da_intnode
*node2
;
769 struct xfs_da_intnode
*tmpnode
;
770 struct xfs_da_node_entry
*btree1
;
771 struct xfs_da_node_entry
*btree2
;
772 struct xfs_da_node_entry
*btree_s
;
773 struct xfs_da_node_entry
*btree_d
;
774 struct xfs_da3_icnode_hdr nodehdr1
;
775 struct xfs_da3_icnode_hdr nodehdr2
;
776 struct xfs_trans
*tp
;
780 struct xfs_inode
*dp
= state
->args
->dp
;
782 trace_xfs_da_node_rebalance(state
->args
);
784 node1
= blk1
->bp
->b_addr
;
785 node2
= blk2
->bp
->b_addr
;
786 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
787 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
788 btree1
= dp
->d_ops
->node_tree_p(node1
);
789 btree2
= dp
->d_ops
->node_tree_p(node2
);
792 * Figure out how many entries need to move, and in which direction.
793 * Swap the nodes around if that makes it simpler.
795 if (nodehdr1
.count
> 0 && nodehdr2
.count
> 0 &&
796 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
797 (be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
) <
798 be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
)))) {
802 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
803 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
804 btree1
= dp
->d_ops
->node_tree_p(node1
);
805 btree2
= dp
->d_ops
->node_tree_p(node2
);
809 count
= (nodehdr1
.count
- nodehdr2
.count
) / 2;
812 tp
= state
->args
->trans
;
814 * Two cases: high-to-low and low-to-high.
818 * Move elements in node2 up to make a hole.
820 tmp
= nodehdr2
.count
;
822 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
823 btree_s
= &btree2
[0];
824 btree_d
= &btree2
[count
];
825 memmove(btree_d
, btree_s
, tmp
);
829 * Move the req'd B-tree elements from high in node1 to
832 nodehdr2
.count
+= count
;
833 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
834 btree_s
= &btree1
[nodehdr1
.count
- count
];
835 btree_d
= &btree2
[0];
836 memcpy(btree_d
, btree_s
, tmp
);
837 nodehdr1
.count
-= count
;
840 * Move the req'd B-tree elements from low in node2 to
844 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
845 btree_s
= &btree2
[0];
846 btree_d
= &btree1
[nodehdr1
.count
];
847 memcpy(btree_d
, btree_s
, tmp
);
848 nodehdr1
.count
+= count
;
850 xfs_trans_log_buf(tp
, blk1
->bp
,
851 XFS_DA_LOGRANGE(node1
, btree_d
, tmp
));
854 * Move elements in node2 down to fill the hole.
856 tmp
= nodehdr2
.count
- count
;
857 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
858 btree_s
= &btree2
[count
];
859 btree_d
= &btree2
[0];
860 memmove(btree_d
, btree_s
, tmp
);
861 nodehdr2
.count
-= count
;
865 * Log header of node 1 and all current bits of node 2.
867 dp
->d_ops
->node_hdr_to_disk(node1
, &nodehdr1
);
868 xfs_trans_log_buf(tp
, blk1
->bp
,
869 XFS_DA_LOGRANGE(node1
, &node1
->hdr
, dp
->d_ops
->node_hdr_size
));
871 dp
->d_ops
->node_hdr_to_disk(node2
, &nodehdr2
);
872 xfs_trans_log_buf(tp
, blk2
->bp
,
873 XFS_DA_LOGRANGE(node2
, &node2
->hdr
,
874 dp
->d_ops
->node_hdr_size
+
875 (sizeof(btree2
[0]) * nodehdr2
.count
)));
878 * Record the last hashval from each block for upward propagation.
879 * (note: don't use the swapped node pointers)
882 node1
= blk1
->bp
->b_addr
;
883 node2
= blk2
->bp
->b_addr
;
884 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
885 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
886 btree1
= dp
->d_ops
->node_tree_p(node1
);
887 btree2
= dp
->d_ops
->node_tree_p(node2
);
889 blk1
->hashval
= be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
);
890 blk2
->hashval
= be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
);
893 * Adjust the expected index for insertion.
895 if (blk1
->index
>= nodehdr1
.count
) {
896 blk2
->index
= blk1
->index
- nodehdr1
.count
;
897 blk1
->index
= nodehdr1
.count
+ 1; /* make it invalid */
902 * Add a new entry to an intermediate node.
906 struct xfs_da_state
*state
,
907 struct xfs_da_state_blk
*oldblk
,
908 struct xfs_da_state_blk
*newblk
)
910 struct xfs_da_intnode
*node
;
911 struct xfs_da3_icnode_hdr nodehdr
;
912 struct xfs_da_node_entry
*btree
;
914 struct xfs_inode
*dp
= state
->args
->dp
;
916 trace_xfs_da_node_add(state
->args
);
918 node
= oldblk
->bp
->b_addr
;
919 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
920 btree
= dp
->d_ops
->node_tree_p(node
);
922 ASSERT(oldblk
->index
>= 0 && oldblk
->index
<= nodehdr
.count
);
923 ASSERT(newblk
->blkno
!= 0);
924 if (state
->args
->whichfork
== XFS_DATA_FORK
)
925 ASSERT(newblk
->blkno
>= state
->args
->geo
->leafblk
&&
926 newblk
->blkno
< state
->args
->geo
->freeblk
);
929 * We may need to make some room before we insert the new node.
932 if (oldblk
->index
< nodehdr
.count
) {
933 tmp
= (nodehdr
.count
- oldblk
->index
) * (uint
)sizeof(*btree
);
934 memmove(&btree
[oldblk
->index
+ 1], &btree
[oldblk
->index
], tmp
);
936 btree
[oldblk
->index
].hashval
= cpu_to_be32(newblk
->hashval
);
937 btree
[oldblk
->index
].before
= cpu_to_be32(newblk
->blkno
);
938 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
939 XFS_DA_LOGRANGE(node
, &btree
[oldblk
->index
],
940 tmp
+ sizeof(*btree
)));
943 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
944 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
945 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
948 * Copy the last hash value from the oldblk to propagate upwards.
950 oldblk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
953 /*========================================================================
954 * Routines used for shrinking the Btree.
955 *========================================================================*/
958 * Deallocate an empty leaf node, remove it from its parent,
959 * possibly deallocating that block, etc...
963 struct xfs_da_state
*state
)
965 struct xfs_da_state_blk
*drop_blk
;
966 struct xfs_da_state_blk
*save_blk
;
970 trace_xfs_da_join(state
->args
);
972 drop_blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
973 save_blk
= &state
->altpath
.blk
[ state
->path
.active
-1 ];
974 ASSERT(state
->path
.blk
[0].magic
== XFS_DA_NODE_MAGIC
);
975 ASSERT(drop_blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
976 drop_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
979 * Walk back up the tree joining/deallocating as necessary.
980 * When we stop dropping blocks, break out.
982 for ( ; state
->path
.active
>= 2; drop_blk
--, save_blk
--,
983 state
->path
.active
--) {
985 * See if we can combine the block with a neighbor.
986 * (action == 0) => no options, just leave
987 * (action == 1) => coalesce, then unlink
988 * (action == 2) => block empty, unlink it
990 switch (drop_blk
->magic
) {
991 case XFS_ATTR_LEAF_MAGIC
:
992 error
= xfs_attr3_leaf_toosmall(state
, &action
);
997 xfs_attr3_leaf_unbalance(state
, drop_blk
, save_blk
);
999 case XFS_DIR2_LEAFN_MAGIC
:
1000 error
= xfs_dir2_leafn_toosmall(state
, &action
);
1005 xfs_dir2_leafn_unbalance(state
, drop_blk
, save_blk
);
1007 case XFS_DA_NODE_MAGIC
:
1009 * Remove the offending node, fixup hashvals,
1010 * check for a toosmall neighbor.
1012 xfs_da3_node_remove(state
, drop_blk
);
1013 xfs_da3_fixhashpath(state
, &state
->path
);
1014 error
= xfs_da3_node_toosmall(state
, &action
);
1019 xfs_da3_node_unbalance(state
, drop_blk
, save_blk
);
1022 xfs_da3_fixhashpath(state
, &state
->altpath
);
1023 error
= xfs_da3_blk_unlink(state
, drop_blk
, save_blk
);
1024 xfs_da_state_kill_altpath(state
);
1027 error
= xfs_da_shrink_inode(state
->args
, drop_blk
->blkno
,
1029 drop_blk
->bp
= NULL
;
1034 * We joined all the way to the top. If it turns out that
1035 * we only have one entry in the root, make the child block
1038 xfs_da3_node_remove(state
, drop_blk
);
1039 xfs_da3_fixhashpath(state
, &state
->path
);
1040 error
= xfs_da3_root_join(state
, &state
->path
.blk
[0]);
1046 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo
*blkinfo
, __u16 level
)
1048 __be16 magic
= blkinfo
->magic
;
1051 ASSERT(magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1052 magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1053 magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1054 magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1056 ASSERT(magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1057 magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
));
1059 ASSERT(!blkinfo
->forw
);
1060 ASSERT(!blkinfo
->back
);
1063 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1067 * We have only one entry in the root. Copy the only remaining child of
1068 * the old root to block 0 as the new root node.
1072 struct xfs_da_state
*state
,
1073 struct xfs_da_state_blk
*root_blk
)
1075 struct xfs_da_intnode
*oldroot
;
1076 struct xfs_da_args
*args
;
1079 struct xfs_da3_icnode_hdr oldroothdr
;
1080 struct xfs_da_node_entry
*btree
;
1082 struct xfs_inode
*dp
= state
->args
->dp
;
1084 trace_xfs_da_root_join(state
->args
);
1086 ASSERT(root_blk
->magic
== XFS_DA_NODE_MAGIC
);
1089 oldroot
= root_blk
->bp
->b_addr
;
1090 dp
->d_ops
->node_hdr_from_disk(&oldroothdr
, oldroot
);
1091 ASSERT(oldroothdr
.forw
== 0);
1092 ASSERT(oldroothdr
.back
== 0);
1095 * If the root has more than one child, then don't do anything.
1097 if (oldroothdr
.count
> 1)
1101 * Read in the (only) child block, then copy those bytes into
1102 * the root block's buffer and free the original child block.
1104 btree
= dp
->d_ops
->node_tree_p(oldroot
);
1105 child
= be32_to_cpu(btree
[0].before
);
1107 error
= xfs_da3_node_read(args
->trans
, dp
, child
, -1, &bp
,
1111 xfs_da_blkinfo_onlychild_validate(bp
->b_addr
, oldroothdr
.level
);
1114 * This could be copying a leaf back into the root block in the case of
1115 * there only being a single leaf block left in the tree. Hence we have
1116 * to update the b_ops pointer as well to match the buffer type change
1117 * that could occur. For dir3 blocks we also need to update the block
1118 * number in the buffer header.
1120 memcpy(root_blk
->bp
->b_addr
, bp
->b_addr
, args
->geo
->blksize
);
1121 root_blk
->bp
->b_ops
= bp
->b_ops
;
1122 xfs_trans_buf_copy_type(root_blk
->bp
, bp
);
1123 if (oldroothdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1124 struct xfs_da3_blkinfo
*da3
= root_blk
->bp
->b_addr
;
1125 da3
->blkno
= cpu_to_be64(root_blk
->bp
->b_bn
);
1127 xfs_trans_log_buf(args
->trans
, root_blk
->bp
, 0,
1128 args
->geo
->blksize
- 1);
1129 error
= xfs_da_shrink_inode(args
, child
, bp
);
1134 * Check a node block and its neighbors to see if the block should be
1135 * collapsed into one or the other neighbor. Always keep the block
1136 * with the smaller block number.
1137 * If the current block is over 50% full, don't try to join it, return 0.
1138 * If the block is empty, fill in the state structure and return 2.
1139 * If it can be collapsed, fill in the state structure and return 1.
1140 * If nothing can be done, return 0.
1143 xfs_da3_node_toosmall(
1144 struct xfs_da_state
*state
,
1147 struct xfs_da_intnode
*node
;
1148 struct xfs_da_state_blk
*blk
;
1149 struct xfs_da_blkinfo
*info
;
1152 struct xfs_da3_icnode_hdr nodehdr
;
1158 struct xfs_inode
*dp
= state
->args
->dp
;
1160 trace_xfs_da_node_toosmall(state
->args
);
1163 * Check for the degenerate case of the block being over 50% full.
1164 * If so, it's not worth even looking to see if we might be able
1165 * to coalesce with a sibling.
1167 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1168 info
= blk
->bp
->b_addr
;
1169 node
= (xfs_da_intnode_t
*)info
;
1170 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1171 if (nodehdr
.count
> (state
->args
->geo
->node_ents
>> 1)) {
1172 *action
= 0; /* blk over 50%, don't try to join */
1173 return 0; /* blk over 50%, don't try to join */
1177 * Check for the degenerate case of the block being empty.
1178 * If the block is empty, we'll simply delete it, no need to
1179 * coalesce it with a sibling block. We choose (arbitrarily)
1180 * to merge with the forward block unless it is NULL.
1182 if (nodehdr
.count
== 0) {
1184 * Make altpath point to the block we want to keep and
1185 * path point to the block we want to drop (this one).
1187 forward
= (info
->forw
!= 0);
1188 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1189 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1202 * Examine each sibling block to see if we can coalesce with
1203 * at least 25% free space to spare. We need to figure out
1204 * whether to merge with the forward or the backward block.
1205 * We prefer coalescing with the lower numbered sibling so as
1206 * to shrink a directory over time.
1208 count
= state
->args
->geo
->node_ents
;
1209 count
-= state
->args
->geo
->node_ents
>> 2;
1210 count
-= nodehdr
.count
;
1212 /* start with smaller blk num */
1213 forward
= nodehdr
.forw
< nodehdr
.back
;
1214 for (i
= 0; i
< 2; forward
= !forward
, i
++) {
1215 struct xfs_da3_icnode_hdr thdr
;
1217 blkno
= nodehdr
.forw
;
1219 blkno
= nodehdr
.back
;
1222 error
= xfs_da3_node_read(state
->args
->trans
, dp
,
1223 blkno
, -1, &bp
, state
->args
->whichfork
);
1228 dp
->d_ops
->node_hdr_from_disk(&thdr
, node
);
1229 xfs_trans_brelse(state
->args
->trans
, bp
);
1231 if (count
- thdr
.count
>= 0)
1232 break; /* fits with at least 25% to spare */
1240 * Make altpath point to the block we want to keep (the lower
1241 * numbered block) and path point to the block we want to drop.
1243 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1244 if (blkno
< blk
->blkno
) {
1245 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1248 error
= xfs_da3_path_shift(state
, &state
->path
, forward
,
1262 * Pick up the last hashvalue from an intermediate node.
1265 xfs_da3_node_lasthash(
1266 struct xfs_inode
*dp
,
1270 struct xfs_da_intnode
*node
;
1271 struct xfs_da_node_entry
*btree
;
1272 struct xfs_da3_icnode_hdr nodehdr
;
1275 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1277 *count
= nodehdr
.count
;
1280 btree
= dp
->d_ops
->node_tree_p(node
);
1281 return be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1285 * Walk back up the tree adjusting hash values as necessary,
1286 * when we stop making changes, return.
1289 xfs_da3_fixhashpath(
1290 struct xfs_da_state
*state
,
1291 struct xfs_da_state_path
*path
)
1293 struct xfs_da_state_blk
*blk
;
1294 struct xfs_da_intnode
*node
;
1295 struct xfs_da_node_entry
*btree
;
1296 xfs_dahash_t lasthash
=0;
1299 struct xfs_inode
*dp
= state
->args
->dp
;
1301 trace_xfs_da_fixhashpath(state
->args
);
1303 level
= path
->active
-1;
1304 blk
= &path
->blk
[ level
];
1305 switch (blk
->magic
) {
1306 case XFS_ATTR_LEAF_MAGIC
:
1307 lasthash
= xfs_attr_leaf_lasthash(blk
->bp
, &count
);
1311 case XFS_DIR2_LEAFN_MAGIC
:
1312 lasthash
= xfs_dir2_leaf_lasthash(dp
, blk
->bp
, &count
);
1316 case XFS_DA_NODE_MAGIC
:
1317 lasthash
= xfs_da3_node_lasthash(dp
, blk
->bp
, &count
);
1322 for (blk
--, level
--; level
>= 0; blk
--, level
--) {
1323 struct xfs_da3_icnode_hdr nodehdr
;
1325 node
= blk
->bp
->b_addr
;
1326 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1327 btree
= dp
->d_ops
->node_tree_p(node
);
1328 if (be32_to_cpu(btree
[blk
->index
].hashval
) == lasthash
)
1330 blk
->hashval
= lasthash
;
1331 btree
[blk
->index
].hashval
= cpu_to_be32(lasthash
);
1332 xfs_trans_log_buf(state
->args
->trans
, blk
->bp
,
1333 XFS_DA_LOGRANGE(node
, &btree
[blk
->index
],
1336 lasthash
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1341 * Remove an entry from an intermediate node.
1344 xfs_da3_node_remove(
1345 struct xfs_da_state
*state
,
1346 struct xfs_da_state_blk
*drop_blk
)
1348 struct xfs_da_intnode
*node
;
1349 struct xfs_da3_icnode_hdr nodehdr
;
1350 struct xfs_da_node_entry
*btree
;
1353 struct xfs_inode
*dp
= state
->args
->dp
;
1355 trace_xfs_da_node_remove(state
->args
);
1357 node
= drop_blk
->bp
->b_addr
;
1358 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1359 ASSERT(drop_blk
->index
< nodehdr
.count
);
1360 ASSERT(drop_blk
->index
>= 0);
1363 * Copy over the offending entry, or just zero it out.
1365 index
= drop_blk
->index
;
1366 btree
= dp
->d_ops
->node_tree_p(node
);
1367 if (index
< nodehdr
.count
- 1) {
1368 tmp
= nodehdr
.count
- index
- 1;
1369 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
1370 memmove(&btree
[index
], &btree
[index
+ 1], tmp
);
1371 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1372 XFS_DA_LOGRANGE(node
, &btree
[index
], tmp
));
1373 index
= nodehdr
.count
- 1;
1375 memset(&btree
[index
], 0, sizeof(xfs_da_node_entry_t
));
1376 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1377 XFS_DA_LOGRANGE(node
, &btree
[index
], sizeof(btree
[index
])));
1379 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
1380 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1381 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
1384 * Copy the last hash value from the block to propagate upwards.
1386 drop_blk
->hashval
= be32_to_cpu(btree
[index
- 1].hashval
);
1390 * Unbalance the elements between two intermediate nodes,
1391 * move all Btree elements from one node into another.
1394 xfs_da3_node_unbalance(
1395 struct xfs_da_state
*state
,
1396 struct xfs_da_state_blk
*drop_blk
,
1397 struct xfs_da_state_blk
*save_blk
)
1399 struct xfs_da_intnode
*drop_node
;
1400 struct xfs_da_intnode
*save_node
;
1401 struct xfs_da_node_entry
*drop_btree
;
1402 struct xfs_da_node_entry
*save_btree
;
1403 struct xfs_da3_icnode_hdr drop_hdr
;
1404 struct xfs_da3_icnode_hdr save_hdr
;
1405 struct xfs_trans
*tp
;
1408 struct xfs_inode
*dp
= state
->args
->dp
;
1410 trace_xfs_da_node_unbalance(state
->args
);
1412 drop_node
= drop_blk
->bp
->b_addr
;
1413 save_node
= save_blk
->bp
->b_addr
;
1414 dp
->d_ops
->node_hdr_from_disk(&drop_hdr
, drop_node
);
1415 dp
->d_ops
->node_hdr_from_disk(&save_hdr
, save_node
);
1416 drop_btree
= dp
->d_ops
->node_tree_p(drop_node
);
1417 save_btree
= dp
->d_ops
->node_tree_p(save_node
);
1418 tp
= state
->args
->trans
;
1421 * If the dying block has lower hashvals, then move all the
1422 * elements in the remaining block up to make a hole.
1424 if ((be32_to_cpu(drop_btree
[0].hashval
) <
1425 be32_to_cpu(save_btree
[0].hashval
)) ||
1426 (be32_to_cpu(drop_btree
[drop_hdr
.count
- 1].hashval
) <
1427 be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
))) {
1428 /* XXX: check this - is memmove dst correct? */
1429 tmp
= save_hdr
.count
* sizeof(xfs_da_node_entry_t
);
1430 memmove(&save_btree
[drop_hdr
.count
], &save_btree
[0], tmp
);
1433 xfs_trans_log_buf(tp
, save_blk
->bp
,
1434 XFS_DA_LOGRANGE(save_node
, &save_btree
[0],
1435 (save_hdr
.count
+ drop_hdr
.count
) *
1436 sizeof(xfs_da_node_entry_t
)));
1438 sindex
= save_hdr
.count
;
1439 xfs_trans_log_buf(tp
, save_blk
->bp
,
1440 XFS_DA_LOGRANGE(save_node
, &save_btree
[sindex
],
1441 drop_hdr
.count
* sizeof(xfs_da_node_entry_t
)));
1445 * Move all the B-tree elements from drop_blk to save_blk.
1447 tmp
= drop_hdr
.count
* (uint
)sizeof(xfs_da_node_entry_t
);
1448 memcpy(&save_btree
[sindex
], &drop_btree
[0], tmp
);
1449 save_hdr
.count
+= drop_hdr
.count
;
1451 dp
->d_ops
->node_hdr_to_disk(save_node
, &save_hdr
);
1452 xfs_trans_log_buf(tp
, save_blk
->bp
,
1453 XFS_DA_LOGRANGE(save_node
, &save_node
->hdr
,
1454 dp
->d_ops
->node_hdr_size
));
1457 * Save the last hashval in the remaining block for upward propagation.
1459 save_blk
->hashval
= be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
);
1462 /*========================================================================
1463 * Routines used for finding things in the Btree.
1464 *========================================================================*/
1467 * Walk down the Btree looking for a particular filename, filling
1468 * in the state structure as we go.
1470 * We will set the state structure to point to each of the elements
1471 * in each of the nodes where either the hashval is or should be.
1473 * We support duplicate hashval's so for each entry in the current
1474 * node that could contain the desired hashval, descend. This is a
1475 * pruned depth-first tree search.
1478 xfs_da3_node_lookup_int(
1479 struct xfs_da_state
*state
,
1482 struct xfs_da_state_blk
*blk
;
1483 struct xfs_da_blkinfo
*curr
;
1484 struct xfs_da_intnode
*node
;
1485 struct xfs_da_node_entry
*btree
;
1486 struct xfs_da3_icnode_hdr nodehdr
;
1487 struct xfs_da_args
*args
;
1489 xfs_dahash_t hashval
;
1490 xfs_dahash_t btreehashval
;
1496 unsigned int expected_level
= 0;
1498 struct xfs_inode
*dp
= state
->args
->dp
;
1503 * Descend thru the B-tree searching each level for the right
1504 * node to use, until the right hashval is found.
1506 blkno
= args
->geo
->leafblk
;
1507 for (blk
= &state
->path
.blk
[0], state
->path
.active
= 1;
1508 state
->path
.active
<= XFS_DA_NODE_MAXDEPTH
;
1509 blk
++, state
->path
.active
++) {
1511 * Read the next node down in the tree.
1514 error
= xfs_da3_node_read(args
->trans
, args
->dp
, blkno
,
1515 -1, &blk
->bp
, args
->whichfork
);
1518 state
->path
.active
--;
1521 curr
= blk
->bp
->b_addr
;
1522 magic
= be16_to_cpu(curr
->magic
);
1524 if (magic
== XFS_ATTR_LEAF_MAGIC
||
1525 magic
== XFS_ATTR3_LEAF_MAGIC
) {
1526 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1527 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1531 if (magic
== XFS_DIR2_LEAFN_MAGIC
||
1532 magic
== XFS_DIR3_LEAFN_MAGIC
) {
1533 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1534 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1539 if (magic
!= XFS_DA_NODE_MAGIC
&& magic
!= XFS_DA3_NODE_MAGIC
)
1540 return -EFSCORRUPTED
;
1542 blk
->magic
= XFS_DA_NODE_MAGIC
;
1545 * Search an intermediate node for a match.
1547 node
= blk
->bp
->b_addr
;
1548 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1549 btree
= dp
->d_ops
->node_tree_p(node
);
1551 /* Tree taller than we can handle; bail out! */
1552 if (nodehdr
.level
>= XFS_DA_NODE_MAXDEPTH
)
1553 return -EFSCORRUPTED
;
1555 /* Check the level from the root. */
1556 if (blkno
== args
->geo
->leafblk
)
1557 expected_level
= nodehdr
.level
- 1;
1558 else if (expected_level
!= nodehdr
.level
)
1559 return -EFSCORRUPTED
;
1563 max
= nodehdr
.count
;
1564 blk
->hashval
= be32_to_cpu(btree
[max
- 1].hashval
);
1567 * Binary search. (note: small blocks will skip loop)
1569 probe
= span
= max
/ 2;
1570 hashval
= args
->hashval
;
1573 btreehashval
= be32_to_cpu(btree
[probe
].hashval
);
1574 if (btreehashval
< hashval
)
1576 else if (btreehashval
> hashval
)
1581 ASSERT((probe
>= 0) && (probe
< max
));
1582 ASSERT((span
<= 4) ||
1583 (be32_to_cpu(btree
[probe
].hashval
) == hashval
));
1586 * Since we may have duplicate hashval's, find the first
1587 * matching hashval in the node.
1590 be32_to_cpu(btree
[probe
].hashval
) >= hashval
) {
1593 while (probe
< max
&&
1594 be32_to_cpu(btree
[probe
].hashval
) < hashval
) {
1599 * Pick the right block to descend on.
1602 blk
->index
= max
- 1;
1603 blkno
= be32_to_cpu(btree
[max
- 1].before
);
1606 blkno
= be32_to_cpu(btree
[probe
].before
);
1609 /* We can't point back to the root. */
1610 if (blkno
== args
->geo
->leafblk
)
1611 return -EFSCORRUPTED
;
1614 if (expected_level
!= 0)
1615 return -EFSCORRUPTED
;
1618 * A leaf block that ends in the hashval that we are interested in
1619 * (final hashval == search hashval) means that the next block may
1620 * contain more entries with the same hashval, shift upward to the
1621 * next leaf and keep searching.
1624 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
) {
1625 retval
= xfs_dir2_leafn_lookup_int(blk
->bp
, args
,
1626 &blk
->index
, state
);
1627 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1628 retval
= xfs_attr3_leaf_lookup_int(blk
->bp
, args
);
1629 blk
->index
= args
->index
;
1630 args
->blkno
= blk
->blkno
;
1633 return -EFSCORRUPTED
;
1635 if (((retval
== -ENOENT
) || (retval
== -ENOATTR
)) &&
1636 (blk
->hashval
== args
->hashval
)) {
1637 error
= xfs_da3_path_shift(state
, &state
->path
, 1, 1,
1643 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1644 /* path_shift() gives ENOENT */
1654 /*========================================================================
1656 *========================================================================*/
1659 * Compare two intermediate nodes for "order".
1663 struct xfs_inode
*dp
,
1664 struct xfs_buf
*node1_bp
,
1665 struct xfs_buf
*node2_bp
)
1667 struct xfs_da_intnode
*node1
;
1668 struct xfs_da_intnode
*node2
;
1669 struct xfs_da_node_entry
*btree1
;
1670 struct xfs_da_node_entry
*btree2
;
1671 struct xfs_da3_icnode_hdr node1hdr
;
1672 struct xfs_da3_icnode_hdr node2hdr
;
1674 node1
= node1_bp
->b_addr
;
1675 node2
= node2_bp
->b_addr
;
1676 dp
->d_ops
->node_hdr_from_disk(&node1hdr
, node1
);
1677 dp
->d_ops
->node_hdr_from_disk(&node2hdr
, node2
);
1678 btree1
= dp
->d_ops
->node_tree_p(node1
);
1679 btree2
= dp
->d_ops
->node_tree_p(node2
);
1681 if (node1hdr
.count
> 0 && node2hdr
.count
> 0 &&
1682 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
1683 (be32_to_cpu(btree2
[node2hdr
.count
- 1].hashval
) <
1684 be32_to_cpu(btree1
[node1hdr
.count
- 1].hashval
)))) {
1691 * Link a new block into a doubly linked list of blocks (of whatever type).
1695 struct xfs_da_state
*state
,
1696 struct xfs_da_state_blk
*old_blk
,
1697 struct xfs_da_state_blk
*new_blk
)
1699 struct xfs_da_blkinfo
*old_info
;
1700 struct xfs_da_blkinfo
*new_info
;
1701 struct xfs_da_blkinfo
*tmp_info
;
1702 struct xfs_da_args
*args
;
1706 struct xfs_inode
*dp
= state
->args
->dp
;
1709 * Set up environment.
1712 ASSERT(args
!= NULL
);
1713 old_info
= old_blk
->bp
->b_addr
;
1714 new_info
= new_blk
->bp
->b_addr
;
1715 ASSERT(old_blk
->magic
== XFS_DA_NODE_MAGIC
||
1716 old_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1717 old_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1719 switch (old_blk
->magic
) {
1720 case XFS_ATTR_LEAF_MAGIC
:
1721 before
= xfs_attr_leaf_order(old_blk
->bp
, new_blk
->bp
);
1723 case XFS_DIR2_LEAFN_MAGIC
:
1724 before
= xfs_dir2_leafn_order(dp
, old_blk
->bp
, new_blk
->bp
);
1726 case XFS_DA_NODE_MAGIC
:
1727 before
= xfs_da3_node_order(dp
, old_blk
->bp
, new_blk
->bp
);
1732 * Link blocks in appropriate order.
1736 * Link new block in before existing block.
1738 trace_xfs_da_link_before(args
);
1739 new_info
->forw
= cpu_to_be32(old_blk
->blkno
);
1740 new_info
->back
= old_info
->back
;
1741 if (old_info
->back
) {
1742 error
= xfs_da3_node_read(args
->trans
, dp
,
1743 be32_to_cpu(old_info
->back
),
1744 -1, &bp
, args
->whichfork
);
1748 tmp_info
= bp
->b_addr
;
1749 ASSERT(tmp_info
->magic
== old_info
->magic
);
1750 ASSERT(be32_to_cpu(tmp_info
->forw
) == old_blk
->blkno
);
1751 tmp_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1752 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1754 old_info
->back
= cpu_to_be32(new_blk
->blkno
);
1757 * Link new block in after existing block.
1759 trace_xfs_da_link_after(args
);
1760 new_info
->forw
= old_info
->forw
;
1761 new_info
->back
= cpu_to_be32(old_blk
->blkno
);
1762 if (old_info
->forw
) {
1763 error
= xfs_da3_node_read(args
->trans
, dp
,
1764 be32_to_cpu(old_info
->forw
),
1765 -1, &bp
, args
->whichfork
);
1769 tmp_info
= bp
->b_addr
;
1770 ASSERT(tmp_info
->magic
== old_info
->magic
);
1771 ASSERT(be32_to_cpu(tmp_info
->back
) == old_blk
->blkno
);
1772 tmp_info
->back
= cpu_to_be32(new_blk
->blkno
);
1773 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1775 old_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1778 xfs_trans_log_buf(args
->trans
, old_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1779 xfs_trans_log_buf(args
->trans
, new_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1784 * Unlink a block from a doubly linked list of blocks.
1786 STATIC
int /* error */
1788 struct xfs_da_state
*state
,
1789 struct xfs_da_state_blk
*drop_blk
,
1790 struct xfs_da_state_blk
*save_blk
)
1792 struct xfs_da_blkinfo
*drop_info
;
1793 struct xfs_da_blkinfo
*save_info
;
1794 struct xfs_da_blkinfo
*tmp_info
;
1795 struct xfs_da_args
*args
;
1800 * Set up environment.
1803 ASSERT(args
!= NULL
);
1804 save_info
= save_blk
->bp
->b_addr
;
1805 drop_info
= drop_blk
->bp
->b_addr
;
1806 ASSERT(save_blk
->magic
== XFS_DA_NODE_MAGIC
||
1807 save_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1808 save_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1809 ASSERT(save_blk
->magic
== drop_blk
->magic
);
1810 ASSERT((be32_to_cpu(save_info
->forw
) == drop_blk
->blkno
) ||
1811 (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
));
1812 ASSERT((be32_to_cpu(drop_info
->forw
) == save_blk
->blkno
) ||
1813 (be32_to_cpu(drop_info
->back
) == save_blk
->blkno
));
1816 * Unlink the leaf block from the doubly linked chain of leaves.
1818 if (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
) {
1819 trace_xfs_da_unlink_back(args
);
1820 save_info
->back
= drop_info
->back
;
1821 if (drop_info
->back
) {
1822 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1823 be32_to_cpu(drop_info
->back
),
1824 -1, &bp
, args
->whichfork
);
1828 tmp_info
= bp
->b_addr
;
1829 ASSERT(tmp_info
->magic
== save_info
->magic
);
1830 ASSERT(be32_to_cpu(tmp_info
->forw
) == drop_blk
->blkno
);
1831 tmp_info
->forw
= cpu_to_be32(save_blk
->blkno
);
1832 xfs_trans_log_buf(args
->trans
, bp
, 0,
1833 sizeof(*tmp_info
) - 1);
1836 trace_xfs_da_unlink_forward(args
);
1837 save_info
->forw
= drop_info
->forw
;
1838 if (drop_info
->forw
) {
1839 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1840 be32_to_cpu(drop_info
->forw
),
1841 -1, &bp
, args
->whichfork
);
1845 tmp_info
= bp
->b_addr
;
1846 ASSERT(tmp_info
->magic
== save_info
->magic
);
1847 ASSERT(be32_to_cpu(tmp_info
->back
) == drop_blk
->blkno
);
1848 tmp_info
->back
= cpu_to_be32(save_blk
->blkno
);
1849 xfs_trans_log_buf(args
->trans
, bp
, 0,
1850 sizeof(*tmp_info
) - 1);
1854 xfs_trans_log_buf(args
->trans
, save_blk
->bp
, 0, sizeof(*save_info
) - 1);
1859 * Move a path "forward" or "!forward" one block at the current level.
1861 * This routine will adjust a "path" to point to the next block
1862 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1863 * Btree, including updating pointers to the intermediate nodes between
1864 * the new bottom and the root.
1868 struct xfs_da_state
*state
,
1869 struct xfs_da_state_path
*path
,
1874 struct xfs_da_state_blk
*blk
;
1875 struct xfs_da_blkinfo
*info
;
1876 struct xfs_da_intnode
*node
;
1877 struct xfs_da_args
*args
;
1878 struct xfs_da_node_entry
*btree
;
1879 struct xfs_da3_icnode_hdr nodehdr
;
1881 xfs_dablk_t blkno
= 0;
1884 struct xfs_inode
*dp
= state
->args
->dp
;
1886 trace_xfs_da_path_shift(state
->args
);
1889 * Roll up the Btree looking for the first block where our
1890 * current index is not at the edge of the block. Note that
1891 * we skip the bottom layer because we want the sibling block.
1894 ASSERT(args
!= NULL
);
1895 ASSERT(path
!= NULL
);
1896 ASSERT((path
->active
> 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1897 level
= (path
->active
-1) - 1; /* skip bottom layer in path */
1898 for (blk
= &path
->blk
[level
]; level
>= 0; blk
--, level
--) {
1899 node
= blk
->bp
->b_addr
;
1900 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1901 btree
= dp
->d_ops
->node_tree_p(node
);
1903 if (forward
&& (blk
->index
< nodehdr
.count
- 1)) {
1905 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1907 } else if (!forward
&& (blk
->index
> 0)) {
1909 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1914 *result
= -ENOENT
; /* we're out of our tree */
1915 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1920 * Roll down the edge of the subtree until we reach the
1921 * same depth we were at originally.
1923 for (blk
++, level
++; level
< path
->active
; blk
++, level
++) {
1925 * Read the next child block into a local buffer.
1927 error
= xfs_da3_node_read(args
->trans
, dp
, blkno
, -1, &bp
,
1933 * Release the old block (if it's dirty, the trans doesn't
1934 * actually let go) and swap the local buffer into the path
1935 * structure. This ensures failure of the above read doesn't set
1936 * a NULL buffer in an active slot in the path.
1939 xfs_trans_brelse(args
->trans
, blk
->bp
);
1943 info
= blk
->bp
->b_addr
;
1944 ASSERT(info
->magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1945 info
->magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
1946 info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1947 info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1948 info
->magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1949 info
->magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1953 * Note: we flatten the magic number to a single type so we
1954 * don't have to compare against crc/non-crc types elsewhere.
1956 switch (be16_to_cpu(info
->magic
)) {
1957 case XFS_DA_NODE_MAGIC
:
1958 case XFS_DA3_NODE_MAGIC
:
1959 blk
->magic
= XFS_DA_NODE_MAGIC
;
1960 node
= (xfs_da_intnode_t
*)info
;
1961 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1962 btree
= dp
->d_ops
->node_tree_p(node
);
1963 blk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1967 blk
->index
= nodehdr
.count
- 1;
1968 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1970 case XFS_ATTR_LEAF_MAGIC
:
1971 case XFS_ATTR3_LEAF_MAGIC
:
1972 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1973 ASSERT(level
== path
->active
-1);
1975 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1977 case XFS_DIR2_LEAFN_MAGIC
:
1978 case XFS_DIR3_LEAFN_MAGIC
:
1979 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1980 ASSERT(level
== path
->active
-1);
1982 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1995 /*========================================================================
1997 *========================================================================*/
2000 * Implement a simple hash on a character string.
2001 * Rotate the hash value by 7 bits, then XOR each character in.
2002 * This is implemented with some source-level loop unrolling.
2005 xfs_da_hashname(const uint8_t *name
, int namelen
)
2010 * Do four characters at a time as long as we can.
2012 for (hash
= 0; namelen
>= 4; namelen
-= 4, name
+= 4)
2013 hash
= (name
[0] << 21) ^ (name
[1] << 14) ^ (name
[2] << 7) ^
2014 (name
[3] << 0) ^ rol32(hash
, 7 * 4);
2017 * Now do the rest of the characters.
2021 return (name
[0] << 14) ^ (name
[1] << 7) ^ (name
[2] << 0) ^
2024 return (name
[0] << 7) ^ (name
[1] << 0) ^ rol32(hash
, 7 * 2);
2026 return (name
[0] << 0) ^ rol32(hash
, 7 * 1);
2027 default: /* case 0: */
2034 struct xfs_da_args
*args
,
2035 const unsigned char *name
,
2038 return (args
->namelen
== len
&& memcmp(args
->name
, name
, len
) == 0) ?
2039 XFS_CMP_EXACT
: XFS_CMP_DIFFERENT
;
2043 xfs_default_hashname(
2044 struct xfs_name
*name
)
2046 return xfs_da_hashname(name
->name
, name
->len
);
2049 const struct xfs_nameops xfs_default_nameops
= {
2050 .hashname
= xfs_default_hashname
,
2051 .compname
= xfs_da_compname
2055 xfs_da_grow_inode_int(
2056 struct xfs_da_args
*args
,
2060 struct xfs_trans
*tp
= args
->trans
;
2061 struct xfs_inode
*dp
= args
->dp
;
2062 int w
= args
->whichfork
;
2063 xfs_rfsblock_t nblks
= dp
->i_d
.di_nblocks
;
2064 struct xfs_bmbt_irec map
, *mapp
;
2065 int nmap
, error
, got
, i
, mapi
;
2068 * Find a spot in the file space to put the new block.
2070 error
= xfs_bmap_first_unused(tp
, dp
, count
, bno
, w
);
2075 * Try mapping it in one filesystem block.
2078 error
= xfs_bmapi_write(tp
, dp
, *bno
, count
,
2079 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
|XFS_BMAPI_CONTIG
,
2080 args
->total
, &map
, &nmap
);
2088 } else if (nmap
== 0 && count
> 1) {
2093 * If we didn't get it and the block might work if fragmented,
2094 * try without the CONTIG flag. Loop until we get it all.
2096 mapp
= kmem_alloc(sizeof(*mapp
) * count
, KM_SLEEP
);
2097 for (b
= *bno
, mapi
= 0; b
< *bno
+ count
; ) {
2098 nmap
= min(XFS_BMAP_MAX_NMAP
, count
);
2099 c
= (int)(*bno
+ count
- b
);
2100 error
= xfs_bmapi_write(tp
, dp
, b
, c
,
2101 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
,
2102 args
->total
, &mapp
[mapi
], &nmap
);
2108 b
= mapp
[mapi
- 1].br_startoff
+
2109 mapp
[mapi
- 1].br_blockcount
;
2117 * Count the blocks we got, make sure it matches the total.
2119 for (i
= 0, got
= 0; i
< mapi
; i
++)
2120 got
+= mapp
[i
].br_blockcount
;
2121 if (got
!= count
|| mapp
[0].br_startoff
!= *bno
||
2122 mapp
[mapi
- 1].br_startoff
+ mapp
[mapi
- 1].br_blockcount
!=
2128 /* account for newly allocated blocks in reserved blocks total */
2129 args
->total
-= dp
->i_d
.di_nblocks
- nblks
;
2138 * Add a block to the btree ahead of the file.
2139 * Return the new block number to the caller.
2143 struct xfs_da_args
*args
,
2144 xfs_dablk_t
*new_blkno
)
2149 trace_xfs_da_grow_inode(args
);
2151 bno
= args
->geo
->leafblk
;
2152 error
= xfs_da_grow_inode_int(args
, &bno
, args
->geo
->fsbcount
);
2154 *new_blkno
= (xfs_dablk_t
)bno
;
2159 * Ick. We need to always be able to remove a btree block, even
2160 * if there's no space reservation because the filesystem is full.
2161 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2162 * It swaps the target block with the last block in the file. The
2163 * last block in the file can always be removed since it can't cause
2164 * a bmap btree split to do that.
2167 xfs_da3_swap_lastblock(
2168 struct xfs_da_args
*args
,
2169 xfs_dablk_t
*dead_blknop
,
2170 struct xfs_buf
**dead_bufp
)
2172 struct xfs_da_blkinfo
*dead_info
;
2173 struct xfs_da_blkinfo
*sib_info
;
2174 struct xfs_da_intnode
*par_node
;
2175 struct xfs_da_intnode
*dead_node
;
2176 struct xfs_dir2_leaf
*dead_leaf2
;
2177 struct xfs_da_node_entry
*btree
;
2178 struct xfs_da3_icnode_hdr par_hdr
;
2179 struct xfs_inode
*dp
;
2180 struct xfs_trans
*tp
;
2181 struct xfs_mount
*mp
;
2182 struct xfs_buf
*dead_buf
;
2183 struct xfs_buf
*last_buf
;
2184 struct xfs_buf
*sib_buf
;
2185 struct xfs_buf
*par_buf
;
2186 xfs_dahash_t dead_hash
;
2187 xfs_fileoff_t lastoff
;
2188 xfs_dablk_t dead_blkno
;
2189 xfs_dablk_t last_blkno
;
2190 xfs_dablk_t sib_blkno
;
2191 xfs_dablk_t par_blkno
;
2198 trace_xfs_da_swap_lastblock(args
);
2200 dead_buf
= *dead_bufp
;
2201 dead_blkno
= *dead_blknop
;
2204 w
= args
->whichfork
;
2205 ASSERT(w
== XFS_DATA_FORK
);
2207 lastoff
= args
->geo
->freeblk
;
2208 error
= xfs_bmap_last_before(tp
, dp
, &lastoff
, w
);
2211 if (unlikely(lastoff
== 0)) {
2212 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW
,
2214 return -EFSCORRUPTED
;
2217 * Read the last block in the btree space.
2219 last_blkno
= (xfs_dablk_t
)lastoff
- args
->geo
->fsbcount
;
2220 error
= xfs_da3_node_read(tp
, dp
, last_blkno
, -1, &last_buf
, w
);
2224 * Copy the last block into the dead buffer and log it.
2226 memcpy(dead_buf
->b_addr
, last_buf
->b_addr
, args
->geo
->blksize
);
2227 xfs_trans_log_buf(tp
, dead_buf
, 0, args
->geo
->blksize
- 1);
2228 dead_info
= dead_buf
->b_addr
;
2230 * Get values from the moved block.
2232 if (dead_info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
2233 dead_info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
2234 struct xfs_dir3_icleaf_hdr leafhdr
;
2235 struct xfs_dir2_leaf_entry
*ents
;
2237 dead_leaf2
= (xfs_dir2_leaf_t
*)dead_info
;
2238 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, dead_leaf2
);
2239 ents
= dp
->d_ops
->leaf_ents_p(dead_leaf2
);
2241 dead_hash
= be32_to_cpu(ents
[leafhdr
.count
- 1].hashval
);
2243 struct xfs_da3_icnode_hdr deadhdr
;
2245 dead_node
= (xfs_da_intnode_t
*)dead_info
;
2246 dp
->d_ops
->node_hdr_from_disk(&deadhdr
, dead_node
);
2247 btree
= dp
->d_ops
->node_tree_p(dead_node
);
2248 dead_level
= deadhdr
.level
;
2249 dead_hash
= be32_to_cpu(btree
[deadhdr
.count
- 1].hashval
);
2251 sib_buf
= par_buf
= NULL
;
2253 * If the moved block has a left sibling, fix up the pointers.
2255 if ((sib_blkno
= be32_to_cpu(dead_info
->back
))) {
2256 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2259 sib_info
= sib_buf
->b_addr
;
2261 be32_to_cpu(sib_info
->forw
) != last_blkno
||
2262 sib_info
->magic
!= dead_info
->magic
)) {
2263 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2264 XFS_ERRLEVEL_LOW
, mp
);
2265 error
= -EFSCORRUPTED
;
2268 sib_info
->forw
= cpu_to_be32(dead_blkno
);
2269 xfs_trans_log_buf(tp
, sib_buf
,
2270 XFS_DA_LOGRANGE(sib_info
, &sib_info
->forw
,
2271 sizeof(sib_info
->forw
)));
2275 * If the moved block has a right sibling, fix up the pointers.
2277 if ((sib_blkno
= be32_to_cpu(dead_info
->forw
))) {
2278 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2281 sib_info
= sib_buf
->b_addr
;
2283 be32_to_cpu(sib_info
->back
) != last_blkno
||
2284 sib_info
->magic
!= dead_info
->magic
)) {
2285 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2286 XFS_ERRLEVEL_LOW
, mp
);
2287 error
= -EFSCORRUPTED
;
2290 sib_info
->back
= cpu_to_be32(dead_blkno
);
2291 xfs_trans_log_buf(tp
, sib_buf
,
2292 XFS_DA_LOGRANGE(sib_info
, &sib_info
->back
,
2293 sizeof(sib_info
->back
)));
2296 par_blkno
= args
->geo
->leafblk
;
2299 * Walk down the tree looking for the parent of the moved block.
2302 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2305 par_node
= par_buf
->b_addr
;
2306 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2307 if (level
>= 0 && level
!= par_hdr
.level
+ 1) {
2308 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2309 XFS_ERRLEVEL_LOW
, mp
);
2310 error
= -EFSCORRUPTED
;
2313 level
= par_hdr
.level
;
2314 btree
= dp
->d_ops
->node_tree_p(par_node
);
2316 entno
< par_hdr
.count
&&
2317 be32_to_cpu(btree
[entno
].hashval
) < dead_hash
;
2320 if (entno
== par_hdr
.count
) {
2321 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2322 XFS_ERRLEVEL_LOW
, mp
);
2323 error
= -EFSCORRUPTED
;
2326 par_blkno
= be32_to_cpu(btree
[entno
].before
);
2327 if (level
== dead_level
+ 1)
2329 xfs_trans_brelse(tp
, par_buf
);
2333 * We're in the right parent block.
2334 * Look for the right entry.
2338 entno
< par_hdr
.count
&&
2339 be32_to_cpu(btree
[entno
].before
) != last_blkno
;
2342 if (entno
< par_hdr
.count
)
2344 par_blkno
= par_hdr
.forw
;
2345 xfs_trans_brelse(tp
, par_buf
);
2347 if (unlikely(par_blkno
== 0)) {
2348 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2349 XFS_ERRLEVEL_LOW
, mp
);
2350 error
= -EFSCORRUPTED
;
2353 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2356 par_node
= par_buf
->b_addr
;
2357 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2358 if (par_hdr
.level
!= level
) {
2359 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2360 XFS_ERRLEVEL_LOW
, mp
);
2361 error
= -EFSCORRUPTED
;
2364 btree
= dp
->d_ops
->node_tree_p(par_node
);
2368 * Update the parent entry pointing to the moved block.
2370 btree
[entno
].before
= cpu_to_be32(dead_blkno
);
2371 xfs_trans_log_buf(tp
, par_buf
,
2372 XFS_DA_LOGRANGE(par_node
, &btree
[entno
].before
,
2373 sizeof(btree
[entno
].before
)));
2374 *dead_blknop
= last_blkno
;
2375 *dead_bufp
= last_buf
;
2379 xfs_trans_brelse(tp
, par_buf
);
2381 xfs_trans_brelse(tp
, sib_buf
);
2382 xfs_trans_brelse(tp
, last_buf
);
2387 * Remove a btree block from a directory or attribute.
2390 xfs_da_shrink_inode(
2391 struct xfs_da_args
*args
,
2392 xfs_dablk_t dead_blkno
,
2393 struct xfs_buf
*dead_buf
)
2395 struct xfs_inode
*dp
;
2396 int done
, error
, w
, count
;
2397 struct xfs_trans
*tp
;
2399 trace_xfs_da_shrink_inode(args
);
2402 w
= args
->whichfork
;
2404 count
= args
->geo
->fsbcount
;
2407 * Remove extents. If we get ENOSPC for a dir we have to move
2408 * the last block to the place we want to kill.
2410 error
= xfs_bunmapi(tp
, dp
, dead_blkno
, count
,
2411 xfs_bmapi_aflag(w
), 0, &done
);
2412 if (error
== -ENOSPC
) {
2413 if (w
!= XFS_DATA_FORK
)
2415 error
= xfs_da3_swap_lastblock(args
, &dead_blkno
,
2423 xfs_trans_binval(tp
, dead_buf
);
2428 * See if the mapping(s) for this btree block are valid, i.e.
2429 * don't contain holes, are logically contiguous, and cover the whole range.
2432 xfs_da_map_covers_blocks(
2434 xfs_bmbt_irec_t
*mapp
,
2441 for (i
= 0, off
= bno
; i
< nmap
; i
++) {
2442 if (mapp
[i
].br_startblock
== HOLESTARTBLOCK
||
2443 mapp
[i
].br_startblock
== DELAYSTARTBLOCK
) {
2446 if (off
!= mapp
[i
].br_startoff
) {
2449 off
+= mapp
[i
].br_blockcount
;
2451 return off
== bno
+ count
;
2455 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2457 * For the single map case, it is assumed that the caller has provided a pointer
2458 * to a valid xfs_buf_map. For the multiple map case, this function will
2459 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2460 * map pointer with the allocated map.
2463 xfs_buf_map_from_irec(
2464 struct xfs_mount
*mp
,
2465 struct xfs_buf_map
**mapp
,
2467 struct xfs_bmbt_irec
*irecs
,
2470 struct xfs_buf_map
*map
;
2473 ASSERT(*nmaps
== 1);
2474 ASSERT(nirecs
>= 1);
2477 map
= kmem_zalloc(nirecs
* sizeof(struct xfs_buf_map
),
2478 KM_SLEEP
| KM_NOFS
);
2486 for (i
= 0; i
< *nmaps
; i
++) {
2487 ASSERT(irecs
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
2488 irecs
[i
].br_startblock
!= HOLESTARTBLOCK
);
2489 map
[i
].bm_bn
= XFS_FSB_TO_DADDR(mp
, irecs
[i
].br_startblock
);
2490 map
[i
].bm_len
= XFS_FSB_TO_BB(mp
, irecs
[i
].br_blockcount
);
2496 * Map the block we are given ready for reading. There are three possible return
2498 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2499 * caller knows not to execute a subsequent read.
2500 * 0 - if we mapped the block successfully
2501 * >0 - positive error number if there was an error.
2505 struct xfs_inode
*dp
,
2507 xfs_daddr_t mappedbno
,
2509 struct xfs_buf_map
**map
,
2512 struct xfs_mount
*mp
= dp
->i_mount
;
2515 struct xfs_bmbt_irec irec
;
2516 struct xfs_bmbt_irec
*irecs
= &irec
;
2519 ASSERT(map
&& *map
);
2520 ASSERT(*nmaps
== 1);
2522 if (whichfork
== XFS_DATA_FORK
)
2523 nfsb
= mp
->m_dir_geo
->fsbcount
;
2525 nfsb
= mp
->m_attr_geo
->fsbcount
;
2528 * Caller doesn't have a mapping. -2 means don't complain
2529 * if we land in a hole.
2531 if (mappedbno
== -1 || mappedbno
== -2) {
2533 * Optimize the one-block case.
2536 irecs
= kmem_zalloc(sizeof(irec
) * nfsb
,
2537 KM_SLEEP
| KM_NOFS
);
2540 error
= xfs_bmapi_read(dp
, (xfs_fileoff_t
)bno
, nfsb
, irecs
,
2541 &nirecs
, xfs_bmapi_aflag(whichfork
));
2545 irecs
->br_startblock
= XFS_DADDR_TO_FSB(mp
, mappedbno
);
2546 irecs
->br_startoff
= (xfs_fileoff_t
)bno
;
2547 irecs
->br_blockcount
= nfsb
;
2548 irecs
->br_state
= 0;
2552 if (!xfs_da_map_covers_blocks(nirecs
, irecs
, bno
, nfsb
)) {
2553 error
= mappedbno
== -2 ? -1 : -EFSCORRUPTED
;
2554 if (unlikely(error
== -EFSCORRUPTED
)) {
2555 if (xfs_error_level
>= XFS_ERRLEVEL_LOW
) {
2557 xfs_alert(mp
, "%s: bno %lld dir: inode %lld",
2558 __func__
, (long long)bno
,
2559 (long long)dp
->i_ino
);
2560 for (i
= 0; i
< *nmaps
; i
++) {
2562 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2564 (long long)irecs
[i
].br_startoff
,
2565 (long long)irecs
[i
].br_startblock
,
2566 (long long)irecs
[i
].br_blockcount
,
2570 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2571 XFS_ERRLEVEL_LOW
, mp
);
2575 error
= xfs_buf_map_from_irec(mp
, map
, nmaps
, irecs
, nirecs
);
2583 * Get a buffer for the dir/attr block.
2587 struct xfs_trans
*trans
,
2588 struct xfs_inode
*dp
,
2590 xfs_daddr_t mappedbno
,
2591 struct xfs_buf
**bpp
,
2595 struct xfs_buf_map map
;
2596 struct xfs_buf_map
*mapp
;
2603 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2606 /* mapping a hole is not an error, but we don't continue */
2612 bp
= xfs_trans_get_buf_map(trans
, dp
->i_mount
->m_ddev_targp
,
2614 error
= bp
? bp
->b_error
: -EIO
;
2617 xfs_trans_brelse(trans
, bp
);
2631 * Get a buffer for the dir/attr block, fill in the contents.
2635 struct xfs_trans
*trans
,
2636 struct xfs_inode
*dp
,
2638 xfs_daddr_t mappedbno
,
2639 struct xfs_buf
**bpp
,
2641 const struct xfs_buf_ops
*ops
)
2644 struct xfs_buf_map map
;
2645 struct xfs_buf_map
*mapp
;
2652 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2655 /* mapping a hole is not an error, but we don't continue */
2661 error
= xfs_trans_read_buf_map(dp
->i_mount
, trans
,
2662 dp
->i_mount
->m_ddev_targp
,
2663 mapp
, nmap
, 0, &bp
, ops
);
2667 if (whichfork
== XFS_ATTR_FORK
)
2668 xfs_buf_set_ref(bp
, XFS_ATTR_BTREE_REF
);
2670 xfs_buf_set_ref(bp
, XFS_DIR_BTREE_REF
);
2680 * Readahead the dir/attr block.
2684 struct xfs_inode
*dp
,
2686 xfs_daddr_t mappedbno
,
2688 const struct xfs_buf_ops
*ops
)
2690 struct xfs_buf_map map
;
2691 struct xfs_buf_map
*mapp
;
2697 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2700 /* mapping a hole is not an error, but we don't continue */
2706 mappedbno
= mapp
[0].bm_bn
;
2707 xfs_buf_readahead_map(dp
->i_mount
->m_ddev_targp
, mapp
, nmap
, ops
);