1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
7 #include "libxfs_priv.h"
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"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
18 #include "xfs_dir2_priv.h"
19 #include "xfs_inode.h"
20 #include "xfs_trans.h"
21 #include "xfs_alloc.h"
23 #include "xfs_attr_leaf.h"
24 #include "xfs_trace.h"
25 #include "xfs_cksum.h"
30 * Routines to implement directories as Btrees of hashed names.
33 /*========================================================================
34 * Function prototypes for the kernel.
35 *========================================================================*/
38 * Routines used for growing the Btree.
40 STATIC
int xfs_da3_root_split(xfs_da_state_t
*state
,
41 xfs_da_state_blk_t
*existing_root
,
42 xfs_da_state_blk_t
*new_child
);
43 STATIC
int xfs_da3_node_split(xfs_da_state_t
*state
,
44 xfs_da_state_blk_t
*existing_blk
,
45 xfs_da_state_blk_t
*split_blk
,
46 xfs_da_state_blk_t
*blk_to_add
,
49 STATIC
void xfs_da3_node_rebalance(xfs_da_state_t
*state
,
50 xfs_da_state_blk_t
*node_blk_1
,
51 xfs_da_state_blk_t
*node_blk_2
);
52 STATIC
void xfs_da3_node_add(xfs_da_state_t
*state
,
53 xfs_da_state_blk_t
*old_node_blk
,
54 xfs_da_state_blk_t
*new_node_blk
);
57 * Routines used for shrinking the Btree.
59 STATIC
int xfs_da3_root_join(xfs_da_state_t
*state
,
60 xfs_da_state_blk_t
*root_blk
);
61 STATIC
int xfs_da3_node_toosmall(xfs_da_state_t
*state
, int *retval
);
62 STATIC
void xfs_da3_node_remove(xfs_da_state_t
*state
,
63 xfs_da_state_blk_t
*drop_blk
);
64 STATIC
void xfs_da3_node_unbalance(xfs_da_state_t
*state
,
65 xfs_da_state_blk_t
*src_node_blk
,
66 xfs_da_state_blk_t
*dst_node_blk
);
71 STATIC
int xfs_da3_blk_unlink(xfs_da_state_t
*state
,
72 xfs_da_state_blk_t
*drop_blk
,
73 xfs_da_state_blk_t
*save_blk
);
76 kmem_zone_t
*xfs_da_state_zone
; /* anchor for state struct zone */
79 * Allocate a dir-state structure.
80 * We don't put them on the stack since they're large.
83 xfs_da_state_alloc(void)
85 return kmem_zone_zalloc(xfs_da_state_zone
, KM_NOFS
);
89 * Kill the altpath contents of a da-state structure.
92 xfs_da_state_kill_altpath(xfs_da_state_t
*state
)
96 for (i
= 0; i
< state
->altpath
.active
; i
++)
97 state
->altpath
.blk
[i
].bp
= NULL
;
98 state
->altpath
.active
= 0;
102 * Free a da-state structure.
105 xfs_da_state_free(xfs_da_state_t
*state
)
107 xfs_da_state_kill_altpath(state
);
109 memset((char *)state
, 0, sizeof(*state
));
111 kmem_zone_free(xfs_da_state_zone
, state
);
114 static xfs_failaddr_t
118 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
119 struct xfs_da_intnode
*hdr
= bp
->b_addr
;
120 struct xfs_da3_icnode_hdr ichdr
;
121 const struct xfs_dir_ops
*ops
;
123 ops
= xfs_dir_get_ops(mp
, NULL
);
125 ops
->node_hdr_from_disk(&ichdr
, hdr
);
127 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
128 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
130 if (ichdr
.magic
!= XFS_DA3_NODE_MAGIC
)
131 return __this_address
;
133 if (!uuid_equal(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_meta_uuid
))
134 return __this_address
;
135 if (be64_to_cpu(hdr3
->info
.blkno
) != bp
->b_bn
)
136 return __this_address
;
137 if (!xfs_log_check_lsn(mp
, be64_to_cpu(hdr3
->info
.lsn
)))
138 return __this_address
;
140 if (ichdr
.magic
!= XFS_DA_NODE_MAGIC
)
141 return __this_address
;
143 if (ichdr
.level
== 0)
144 return __this_address
;
145 if (ichdr
.level
> XFS_DA_NODE_MAXDEPTH
)
146 return __this_address
;
147 if (ichdr
.count
== 0)
148 return __this_address
;
151 * we don't know if the node is for and attribute or directory tree,
152 * so only fail if the count is outside both bounds
154 if (ichdr
.count
> mp
->m_dir_geo
->node_ents
&&
155 ichdr
.count
> mp
->m_attr_geo
->node_ents
)
156 return __this_address
;
158 /* XXX: hash order check? */
164 xfs_da3_node_write_verify(
167 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
168 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
169 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
172 fa
= xfs_da3_node_verify(bp
);
174 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
178 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
182 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
184 xfs_buf_update_cksum(bp
, XFS_DA3_NODE_CRC_OFF
);
188 * leaf/node format detection on trees is sketchy, so a node read can be done on
189 * leaf level blocks when detection identifies the tree as a node format tree
190 * incorrectly. In this case, we need to swap the verifier to match the correct
191 * format of the block being read.
194 xfs_da3_node_read_verify(
197 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
200 switch (be16_to_cpu(info
->magic
)) {
201 case XFS_DA3_NODE_MAGIC
:
202 if (!xfs_buf_verify_cksum(bp
, XFS_DA3_NODE_CRC_OFF
)) {
203 xfs_verifier_error(bp
, -EFSBADCRC
,
208 case XFS_DA_NODE_MAGIC
:
209 fa
= xfs_da3_node_verify(bp
);
211 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
213 case XFS_ATTR_LEAF_MAGIC
:
214 case XFS_ATTR3_LEAF_MAGIC
:
215 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
216 bp
->b_ops
->verify_read(bp
);
218 case XFS_DIR2_LEAFN_MAGIC
:
219 case XFS_DIR3_LEAFN_MAGIC
:
220 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
221 bp
->b_ops
->verify_read(bp
);
224 xfs_verifier_error(bp
, -EFSCORRUPTED
, __this_address
);
229 /* Verify the structure of a da3 block. */
230 static xfs_failaddr_t
231 xfs_da3_node_verify_struct(
234 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
236 switch (be16_to_cpu(info
->magic
)) {
237 case XFS_DA3_NODE_MAGIC
:
238 case XFS_DA_NODE_MAGIC
:
239 return xfs_da3_node_verify(bp
);
240 case XFS_ATTR_LEAF_MAGIC
:
241 case XFS_ATTR3_LEAF_MAGIC
:
242 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
243 return bp
->b_ops
->verify_struct(bp
);
244 case XFS_DIR2_LEAFN_MAGIC
:
245 case XFS_DIR3_LEAFN_MAGIC
:
246 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
247 return bp
->b_ops
->verify_struct(bp
);
249 return __this_address
;
253 const struct xfs_buf_ops xfs_da3_node_buf_ops
= {
254 .name
= "xfs_da3_node",
255 .verify_read
= xfs_da3_node_read_verify
,
256 .verify_write
= xfs_da3_node_write_verify
,
257 .verify_struct
= xfs_da3_node_verify_struct
,
262 struct xfs_trans
*tp
,
263 struct xfs_inode
*dp
,
265 xfs_daddr_t mappedbno
,
266 struct xfs_buf
**bpp
,
271 err
= xfs_da_read_buf(tp
, dp
, bno
, mappedbno
, bpp
,
272 which_fork
, &xfs_da3_node_buf_ops
);
273 if (!err
&& tp
&& *bpp
) {
274 struct xfs_da_blkinfo
*info
= (*bpp
)->b_addr
;
277 switch (be16_to_cpu(info
->magic
)) {
278 case XFS_DA_NODE_MAGIC
:
279 case XFS_DA3_NODE_MAGIC
:
280 type
= XFS_BLFT_DA_NODE_BUF
;
282 case XFS_ATTR_LEAF_MAGIC
:
283 case XFS_ATTR3_LEAF_MAGIC
:
284 type
= XFS_BLFT_ATTR_LEAF_BUF
;
286 case XFS_DIR2_LEAFN_MAGIC
:
287 case XFS_DIR3_LEAFN_MAGIC
:
288 type
= XFS_BLFT_DIR_LEAFN_BUF
;
291 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
,
292 tp
->t_mountp
, info
, sizeof(*info
));
293 xfs_trans_brelse(tp
, *bpp
);
295 return -EFSCORRUPTED
;
297 xfs_trans_buf_set_type(tp
, *bpp
, type
);
302 /*========================================================================
303 * Routines used for growing the Btree.
304 *========================================================================*/
307 * Create the initial contents of an intermediate node.
311 struct xfs_da_args
*args
,
314 struct xfs_buf
**bpp
,
317 struct xfs_da_intnode
*node
;
318 struct xfs_trans
*tp
= args
->trans
;
319 struct xfs_mount
*mp
= tp
->t_mountp
;
320 struct xfs_da3_icnode_hdr ichdr
= {0};
323 struct xfs_inode
*dp
= args
->dp
;
325 trace_xfs_da_node_create(args
);
326 ASSERT(level
<= XFS_DA_NODE_MAXDEPTH
);
328 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, whichfork
);
331 bp
->b_ops
= &xfs_da3_node_buf_ops
;
332 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
335 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
336 struct xfs_da3_node_hdr
*hdr3
= bp
->b_addr
;
338 memset(hdr3
, 0, sizeof(struct xfs_da3_node_hdr
));
339 ichdr
.magic
= XFS_DA3_NODE_MAGIC
;
340 hdr3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
341 hdr3
->info
.owner
= cpu_to_be64(args
->dp
->i_ino
);
342 uuid_copy(&hdr3
->info
.uuid
, &mp
->m_sb
.sb_meta_uuid
);
344 ichdr
.magic
= XFS_DA_NODE_MAGIC
;
348 dp
->d_ops
->node_hdr_to_disk(node
, &ichdr
);
349 xfs_trans_log_buf(tp
, bp
,
350 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
357 * Split a leaf node, rebalance, then possibly split
358 * intermediate nodes, rebalance, etc.
362 struct xfs_da_state
*state
)
364 struct xfs_da_state_blk
*oldblk
;
365 struct xfs_da_state_blk
*newblk
;
366 struct xfs_da_state_blk
*addblk
;
367 struct xfs_da_intnode
*node
;
373 trace_xfs_da_split(state
->args
);
376 * Walk back up the tree splitting/inserting/adjusting as necessary.
377 * If we need to insert and there isn't room, split the node, then
378 * decide which fragment to insert the new block from below into.
379 * Note that we may split the root this way, but we need more fixup.
381 max
= state
->path
.active
- 1;
382 ASSERT((max
>= 0) && (max
< XFS_DA_NODE_MAXDEPTH
));
383 ASSERT(state
->path
.blk
[max
].magic
== XFS_ATTR_LEAF_MAGIC
||
384 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
386 addblk
= &state
->path
.blk
[max
]; /* initial dummy value */
387 for (i
= max
; (i
>= 0) && addblk
; state
->path
.active
--, i
--) {
388 oldblk
= &state
->path
.blk
[i
];
389 newblk
= &state
->altpath
.blk
[i
];
392 * If a leaf node then
393 * Allocate a new leaf node, then rebalance across them.
394 * else if an intermediate node then
395 * We split on the last layer, must we split the node?
397 switch (oldblk
->magic
) {
398 case XFS_ATTR_LEAF_MAGIC
:
399 error
= xfs_attr3_leaf_split(state
, oldblk
, newblk
);
400 if ((error
!= 0) && (error
!= -ENOSPC
)) {
401 return error
; /* GROT: attr is inconsistent */
408 * Entry wouldn't fit, split the leaf again. The new
409 * extrablk will be consumed by xfs_da3_node_split if
412 state
->extravalid
= 1;
414 state
->extraafter
= 0; /* before newblk */
415 trace_xfs_attr_leaf_split_before(state
->args
);
416 error
= xfs_attr3_leaf_split(state
, oldblk
,
419 state
->extraafter
= 1; /* after newblk */
420 trace_xfs_attr_leaf_split_after(state
->args
);
421 error
= xfs_attr3_leaf_split(state
, newblk
,
425 return error
; /* GROT: attr inconsistent */
428 case XFS_DIR2_LEAFN_MAGIC
:
429 error
= xfs_dir2_leafn_split(state
, oldblk
, newblk
);
434 case XFS_DA_NODE_MAGIC
:
435 error
= xfs_da3_node_split(state
, oldblk
, newblk
, addblk
,
439 return error
; /* GROT: dir is inconsistent */
441 * Record the newly split block for the next time thru?
451 * Update the btree to show the new hashval for this child.
453 xfs_da3_fixhashpath(state
, &state
->path
);
459 * xfs_da3_node_split() should have consumed any extra blocks we added
460 * during a double leaf split in the attr fork. This is guaranteed as
461 * we can't be here if the attr fork only has a single leaf block.
463 ASSERT(state
->extravalid
== 0 ||
464 state
->path
.blk
[max
].magic
== XFS_DIR2_LEAFN_MAGIC
);
467 * Split the root node.
469 ASSERT(state
->path
.active
== 0);
470 oldblk
= &state
->path
.blk
[0];
471 error
= xfs_da3_root_split(state
, oldblk
, addblk
);
474 return error
; /* GROT: dir is inconsistent */
478 * Update pointers to the node which used to be block 0 and just got
479 * bumped because of the addition of a new root node. Note that the
480 * original block 0 could be at any position in the list of blocks in
483 * Note: the magic numbers and sibling pointers are in the same physical
484 * place for both v2 and v3 headers (by design). Hence it doesn't matter
485 * which version of the xfs_da_intnode structure we use here as the
486 * result will be the same using either structure.
488 node
= oldblk
->bp
->b_addr
;
489 if (node
->hdr
.info
.forw
) {
490 ASSERT(be32_to_cpu(node
->hdr
.info
.forw
) == addblk
->blkno
);
491 node
= addblk
->bp
->b_addr
;
492 node
->hdr
.info
.back
= cpu_to_be32(oldblk
->blkno
);
493 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
494 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
495 sizeof(node
->hdr
.info
)));
497 node
= oldblk
->bp
->b_addr
;
498 if (node
->hdr
.info
.back
) {
499 ASSERT(be32_to_cpu(node
->hdr
.info
.back
) == addblk
->blkno
);
500 node
= addblk
->bp
->b_addr
;
501 node
->hdr
.info
.forw
= cpu_to_be32(oldblk
->blkno
);
502 xfs_trans_log_buf(state
->args
->trans
, addblk
->bp
,
503 XFS_DA_LOGRANGE(node
, &node
->hdr
.info
,
504 sizeof(node
->hdr
.info
)));
511 * Split the root. We have to create a new root and point to the two
512 * parts (the split old root) that we just created. Copy block zero to
513 * the EOF, extending the inode in process.
515 STATIC
int /* error */
517 struct xfs_da_state
*state
,
518 struct xfs_da_state_blk
*blk1
,
519 struct xfs_da_state_blk
*blk2
)
521 struct xfs_da_intnode
*node
;
522 struct xfs_da_intnode
*oldroot
;
523 struct xfs_da_node_entry
*btree
;
524 struct xfs_da3_icnode_hdr nodehdr
;
525 struct xfs_da_args
*args
;
527 struct xfs_inode
*dp
;
528 struct xfs_trans
*tp
;
529 struct xfs_dir2_leaf
*leaf
;
535 trace_xfs_da_root_split(state
->args
);
538 * Copy the existing (incorrect) block from the root node position
539 * to a free space somewhere.
542 error
= xfs_da_grow_inode(args
, &blkno
);
548 error
= xfs_da_get_buf(tp
, dp
, blkno
, -1, &bp
, args
->whichfork
);
552 oldroot
= blk1
->bp
->b_addr
;
553 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
554 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
)) {
555 struct xfs_da3_icnode_hdr icnodehdr
;
557 dp
->d_ops
->node_hdr_from_disk(&icnodehdr
, oldroot
);
558 btree
= dp
->d_ops
->node_tree_p(oldroot
);
559 size
= (int)((char *)&btree
[icnodehdr
.count
] - (char *)oldroot
);
560 level
= icnodehdr
.level
;
563 * we are about to copy oldroot to bp, so set up the type
564 * of bp while we know exactly what it will be.
566 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DA_NODE_BUF
);
568 struct xfs_dir3_icleaf_hdr leafhdr
;
569 struct xfs_dir2_leaf_entry
*ents
;
571 leaf
= (xfs_dir2_leaf_t
*)oldroot
;
572 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
573 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
575 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
576 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
577 size
= (int)((char *)&ents
[leafhdr
.count
] - (char *)leaf
);
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_DIR_LEAFN_BUF
);
588 * we can copy most of the information in the node from one block to
589 * another, but for CRC enabled headers we have to make sure that the
590 * block specific identifiers are kept intact. We update the buffer
593 memcpy(node
, oldroot
, size
);
594 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
595 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
596 struct xfs_da3_intnode
*node3
= (struct xfs_da3_intnode
*)node
;
598 node3
->hdr
.info
.blkno
= cpu_to_be64(bp
->b_bn
);
600 xfs_trans_log_buf(tp
, bp
, 0, size
- 1);
602 bp
->b_ops
= blk1
->bp
->b_ops
;
603 xfs_trans_buf_copy_type(bp
, blk1
->bp
);
608 * Set up the new root node.
610 error
= xfs_da3_node_create(args
,
611 (args
->whichfork
== XFS_DATA_FORK
) ? args
->geo
->leafblk
: 0,
612 level
+ 1, &bp
, args
->whichfork
);
617 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
618 btree
= dp
->d_ops
->node_tree_p(node
);
619 btree
[0].hashval
= cpu_to_be32(blk1
->hashval
);
620 btree
[0].before
= cpu_to_be32(blk1
->blkno
);
621 btree
[1].hashval
= cpu_to_be32(blk2
->hashval
);
622 btree
[1].before
= cpu_to_be32(blk2
->blkno
);
624 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
627 if (oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
628 oldroot
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
629 ASSERT(blk1
->blkno
>= args
->geo
->leafblk
&&
630 blk1
->blkno
< args
->geo
->freeblk
);
631 ASSERT(blk2
->blkno
>= args
->geo
->leafblk
&&
632 blk2
->blkno
< args
->geo
->freeblk
);
636 /* Header is already logged by xfs_da_node_create */
637 xfs_trans_log_buf(tp
, bp
,
638 XFS_DA_LOGRANGE(node
, btree
, sizeof(xfs_da_node_entry_t
) * 2));
644 * Split the node, rebalance, then add the new entry.
646 STATIC
int /* error */
648 struct xfs_da_state
*state
,
649 struct xfs_da_state_blk
*oldblk
,
650 struct xfs_da_state_blk
*newblk
,
651 struct xfs_da_state_blk
*addblk
,
655 struct xfs_da_intnode
*node
;
656 struct xfs_da3_icnode_hdr nodehdr
;
661 struct xfs_inode
*dp
= state
->args
->dp
;
663 trace_xfs_da_node_split(state
->args
);
665 node
= oldblk
->bp
->b_addr
;
666 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
669 * With V2 dirs the extra block is data or freespace.
671 useextra
= state
->extravalid
&& state
->args
->whichfork
== XFS_ATTR_FORK
;
672 newcount
= 1 + useextra
;
674 * Do we have to split the node?
676 if (nodehdr
.count
+ newcount
> state
->args
->geo
->node_ents
) {
678 * Allocate a new node, add to the doubly linked chain of
679 * nodes, then move some of our excess entries into it.
681 error
= xfs_da_grow_inode(state
->args
, &blkno
);
683 return error
; /* GROT: dir is inconsistent */
685 error
= xfs_da3_node_create(state
->args
, blkno
, treelevel
,
686 &newblk
->bp
, state
->args
->whichfork
);
688 return error
; /* GROT: dir is inconsistent */
689 newblk
->blkno
= blkno
;
690 newblk
->magic
= XFS_DA_NODE_MAGIC
;
691 xfs_da3_node_rebalance(state
, oldblk
, newblk
);
692 error
= xfs_da3_blk_link(state
, oldblk
, newblk
);
701 * Insert the new entry(s) into the correct block
702 * (updating last hashval in the process).
704 * xfs_da3_node_add() inserts BEFORE the given index,
705 * and as a result of using node_lookup_int() we always
706 * point to a valid entry (not after one), but a split
707 * operation always results in a new block whose hashvals
708 * FOLLOW the current block.
710 * If we had double-split op below us, then add the extra block too.
712 node
= oldblk
->bp
->b_addr
;
713 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
714 if (oldblk
->index
<= nodehdr
.count
) {
716 xfs_da3_node_add(state
, oldblk
, addblk
);
718 if (state
->extraafter
)
720 xfs_da3_node_add(state
, oldblk
, &state
->extrablk
);
721 state
->extravalid
= 0;
725 xfs_da3_node_add(state
, newblk
, addblk
);
727 if (state
->extraafter
)
729 xfs_da3_node_add(state
, newblk
, &state
->extrablk
);
730 state
->extravalid
= 0;
738 * Balance the btree elements between two intermediate nodes,
739 * usually one full and one empty.
741 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
744 xfs_da3_node_rebalance(
745 struct xfs_da_state
*state
,
746 struct xfs_da_state_blk
*blk1
,
747 struct xfs_da_state_blk
*blk2
)
749 struct xfs_da_intnode
*node1
;
750 struct xfs_da_intnode
*node2
;
751 struct xfs_da_intnode
*tmpnode
;
752 struct xfs_da_node_entry
*btree1
;
753 struct xfs_da_node_entry
*btree2
;
754 struct xfs_da_node_entry
*btree_s
;
755 struct xfs_da_node_entry
*btree_d
;
756 struct xfs_da3_icnode_hdr nodehdr1
;
757 struct xfs_da3_icnode_hdr nodehdr2
;
758 struct xfs_trans
*tp
;
762 struct xfs_inode
*dp
= state
->args
->dp
;
764 trace_xfs_da_node_rebalance(state
->args
);
766 node1
= blk1
->bp
->b_addr
;
767 node2
= blk2
->bp
->b_addr
;
768 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
769 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
770 btree1
= dp
->d_ops
->node_tree_p(node1
);
771 btree2
= dp
->d_ops
->node_tree_p(node2
);
774 * Figure out how many entries need to move, and in which direction.
775 * Swap the nodes around if that makes it simpler.
777 if (nodehdr1
.count
> 0 && nodehdr2
.count
> 0 &&
778 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
779 (be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
) <
780 be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
)))) {
784 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
785 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
786 btree1
= dp
->d_ops
->node_tree_p(node1
);
787 btree2
= dp
->d_ops
->node_tree_p(node2
);
791 count
= (nodehdr1
.count
- nodehdr2
.count
) / 2;
794 tp
= state
->args
->trans
;
796 * Two cases: high-to-low and low-to-high.
800 * Move elements in node2 up to make a hole.
802 tmp
= nodehdr2
.count
;
804 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
805 btree_s
= &btree2
[0];
806 btree_d
= &btree2
[count
];
807 memmove(btree_d
, btree_s
, tmp
);
811 * Move the req'd B-tree elements from high in node1 to
814 nodehdr2
.count
+= count
;
815 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
816 btree_s
= &btree1
[nodehdr1
.count
- count
];
817 btree_d
= &btree2
[0];
818 memcpy(btree_d
, btree_s
, tmp
);
819 nodehdr1
.count
-= count
;
822 * Move the req'd B-tree elements from low in node2 to
826 tmp
= count
* (uint
)sizeof(xfs_da_node_entry_t
);
827 btree_s
= &btree2
[0];
828 btree_d
= &btree1
[nodehdr1
.count
];
829 memcpy(btree_d
, btree_s
, tmp
);
830 nodehdr1
.count
+= count
;
832 xfs_trans_log_buf(tp
, blk1
->bp
,
833 XFS_DA_LOGRANGE(node1
, btree_d
, tmp
));
836 * Move elements in node2 down to fill the hole.
838 tmp
= nodehdr2
.count
- count
;
839 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
840 btree_s
= &btree2
[count
];
841 btree_d
= &btree2
[0];
842 memmove(btree_d
, btree_s
, tmp
);
843 nodehdr2
.count
-= count
;
847 * Log header of node 1 and all current bits of node 2.
849 dp
->d_ops
->node_hdr_to_disk(node1
, &nodehdr1
);
850 xfs_trans_log_buf(tp
, blk1
->bp
,
851 XFS_DA_LOGRANGE(node1
, &node1
->hdr
, dp
->d_ops
->node_hdr_size
));
853 dp
->d_ops
->node_hdr_to_disk(node2
, &nodehdr2
);
854 xfs_trans_log_buf(tp
, blk2
->bp
,
855 XFS_DA_LOGRANGE(node2
, &node2
->hdr
,
856 dp
->d_ops
->node_hdr_size
+
857 (sizeof(btree2
[0]) * nodehdr2
.count
)));
860 * Record the last hashval from each block for upward propagation.
861 * (note: don't use the swapped node pointers)
864 node1
= blk1
->bp
->b_addr
;
865 node2
= blk2
->bp
->b_addr
;
866 dp
->d_ops
->node_hdr_from_disk(&nodehdr1
, node1
);
867 dp
->d_ops
->node_hdr_from_disk(&nodehdr2
, node2
);
868 btree1
= dp
->d_ops
->node_tree_p(node1
);
869 btree2
= dp
->d_ops
->node_tree_p(node2
);
871 blk1
->hashval
= be32_to_cpu(btree1
[nodehdr1
.count
- 1].hashval
);
872 blk2
->hashval
= be32_to_cpu(btree2
[nodehdr2
.count
- 1].hashval
);
875 * Adjust the expected index for insertion.
877 if (blk1
->index
>= nodehdr1
.count
) {
878 blk2
->index
= blk1
->index
- nodehdr1
.count
;
879 blk1
->index
= nodehdr1
.count
+ 1; /* make it invalid */
884 * Add a new entry to an intermediate node.
888 struct xfs_da_state
*state
,
889 struct xfs_da_state_blk
*oldblk
,
890 struct xfs_da_state_blk
*newblk
)
892 struct xfs_da_intnode
*node
;
893 struct xfs_da3_icnode_hdr nodehdr
;
894 struct xfs_da_node_entry
*btree
;
896 struct xfs_inode
*dp
= state
->args
->dp
;
898 trace_xfs_da_node_add(state
->args
);
900 node
= oldblk
->bp
->b_addr
;
901 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
902 btree
= dp
->d_ops
->node_tree_p(node
);
904 ASSERT(oldblk
->index
>= 0 && oldblk
->index
<= nodehdr
.count
);
905 ASSERT(newblk
->blkno
!= 0);
906 if (state
->args
->whichfork
== XFS_DATA_FORK
)
907 ASSERT(newblk
->blkno
>= state
->args
->geo
->leafblk
&&
908 newblk
->blkno
< state
->args
->geo
->freeblk
);
911 * We may need to make some room before we insert the new node.
914 if (oldblk
->index
< nodehdr
.count
) {
915 tmp
= (nodehdr
.count
- oldblk
->index
) * (uint
)sizeof(*btree
);
916 memmove(&btree
[oldblk
->index
+ 1], &btree
[oldblk
->index
], tmp
);
918 btree
[oldblk
->index
].hashval
= cpu_to_be32(newblk
->hashval
);
919 btree
[oldblk
->index
].before
= cpu_to_be32(newblk
->blkno
);
920 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
921 XFS_DA_LOGRANGE(node
, &btree
[oldblk
->index
],
922 tmp
+ sizeof(*btree
)));
925 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
926 xfs_trans_log_buf(state
->args
->trans
, oldblk
->bp
,
927 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
930 * Copy the last hash value from the oldblk to propagate upwards.
932 oldblk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
935 /*========================================================================
936 * Routines used for shrinking the Btree.
937 *========================================================================*/
940 * Deallocate an empty leaf node, remove it from its parent,
941 * possibly deallocating that block, etc...
945 struct xfs_da_state
*state
)
947 struct xfs_da_state_blk
*drop_blk
;
948 struct xfs_da_state_blk
*save_blk
;
952 trace_xfs_da_join(state
->args
);
954 drop_blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
955 save_blk
= &state
->altpath
.blk
[ state
->path
.active
-1 ];
956 ASSERT(state
->path
.blk
[0].magic
== XFS_DA_NODE_MAGIC
);
957 ASSERT(drop_blk
->magic
== XFS_ATTR_LEAF_MAGIC
||
958 drop_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
961 * Walk back up the tree joining/deallocating as necessary.
962 * When we stop dropping blocks, break out.
964 for ( ; state
->path
.active
>= 2; drop_blk
--, save_blk
--,
965 state
->path
.active
--) {
967 * See if we can combine the block with a neighbor.
968 * (action == 0) => no options, just leave
969 * (action == 1) => coalesce, then unlink
970 * (action == 2) => block empty, unlink it
972 switch (drop_blk
->magic
) {
973 case XFS_ATTR_LEAF_MAGIC
:
974 error
= xfs_attr3_leaf_toosmall(state
, &action
);
979 xfs_attr3_leaf_unbalance(state
, drop_blk
, save_blk
);
981 case XFS_DIR2_LEAFN_MAGIC
:
982 error
= xfs_dir2_leafn_toosmall(state
, &action
);
987 xfs_dir2_leafn_unbalance(state
, drop_blk
, save_blk
);
989 case XFS_DA_NODE_MAGIC
:
991 * Remove the offending node, fixup hashvals,
992 * check for a toosmall neighbor.
994 xfs_da3_node_remove(state
, drop_blk
);
995 xfs_da3_fixhashpath(state
, &state
->path
);
996 error
= xfs_da3_node_toosmall(state
, &action
);
1001 xfs_da3_node_unbalance(state
, drop_blk
, save_blk
);
1004 xfs_da3_fixhashpath(state
, &state
->altpath
);
1005 error
= xfs_da3_blk_unlink(state
, drop_blk
, save_blk
);
1006 xfs_da_state_kill_altpath(state
);
1009 error
= xfs_da_shrink_inode(state
->args
, drop_blk
->blkno
,
1011 drop_blk
->bp
= NULL
;
1016 * We joined all the way to the top. If it turns out that
1017 * we only have one entry in the root, make the child block
1020 xfs_da3_node_remove(state
, drop_blk
);
1021 xfs_da3_fixhashpath(state
, &state
->path
);
1022 error
= xfs_da3_root_join(state
, &state
->path
.blk
[0]);
1028 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo
*blkinfo
, __u16 level
)
1030 __be16 magic
= blkinfo
->magic
;
1033 ASSERT(magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1034 magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1035 magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1036 magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1038 ASSERT(magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1039 magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
));
1041 ASSERT(!blkinfo
->forw
);
1042 ASSERT(!blkinfo
->back
);
1045 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1049 * We have only one entry in the root. Copy the only remaining child of
1050 * the old root to block 0 as the new root node.
1054 struct xfs_da_state
*state
,
1055 struct xfs_da_state_blk
*root_blk
)
1057 struct xfs_da_intnode
*oldroot
;
1058 struct xfs_da_args
*args
;
1061 struct xfs_da3_icnode_hdr oldroothdr
;
1062 struct xfs_da_node_entry
*btree
;
1064 struct xfs_inode
*dp
= state
->args
->dp
;
1066 trace_xfs_da_root_join(state
->args
);
1068 ASSERT(root_blk
->magic
== XFS_DA_NODE_MAGIC
);
1071 oldroot
= root_blk
->bp
->b_addr
;
1072 dp
->d_ops
->node_hdr_from_disk(&oldroothdr
, oldroot
);
1073 ASSERT(oldroothdr
.forw
== 0);
1074 ASSERT(oldroothdr
.back
== 0);
1077 * If the root has more than one child, then don't do anything.
1079 if (oldroothdr
.count
> 1)
1083 * Read in the (only) child block, then copy those bytes into
1084 * the root block's buffer and free the original child block.
1086 btree
= dp
->d_ops
->node_tree_p(oldroot
);
1087 child
= be32_to_cpu(btree
[0].before
);
1089 error
= xfs_da3_node_read(args
->trans
, dp
, child
, -1, &bp
,
1093 xfs_da_blkinfo_onlychild_validate(bp
->b_addr
, oldroothdr
.level
);
1096 * This could be copying a leaf back into the root block in the case of
1097 * there only being a single leaf block left in the tree. Hence we have
1098 * to update the b_ops pointer as well to match the buffer type change
1099 * that could occur. For dir3 blocks we also need to update the block
1100 * number in the buffer header.
1102 memcpy(root_blk
->bp
->b_addr
, bp
->b_addr
, args
->geo
->blksize
);
1103 root_blk
->bp
->b_ops
= bp
->b_ops
;
1104 xfs_trans_buf_copy_type(root_blk
->bp
, bp
);
1105 if (oldroothdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1106 struct xfs_da3_blkinfo
*da3
= root_blk
->bp
->b_addr
;
1107 da3
->blkno
= cpu_to_be64(root_blk
->bp
->b_bn
);
1109 xfs_trans_log_buf(args
->trans
, root_blk
->bp
, 0,
1110 args
->geo
->blksize
- 1);
1111 error
= xfs_da_shrink_inode(args
, child
, bp
);
1116 * Check a node block and its neighbors to see if the block should be
1117 * collapsed into one or the other neighbor. Always keep the block
1118 * with the smaller block number.
1119 * If the current block is over 50% full, don't try to join it, return 0.
1120 * If the block is empty, fill in the state structure and return 2.
1121 * If it can be collapsed, fill in the state structure and return 1.
1122 * If nothing can be done, return 0.
1125 xfs_da3_node_toosmall(
1126 struct xfs_da_state
*state
,
1129 struct xfs_da_intnode
*node
;
1130 struct xfs_da_state_blk
*blk
;
1131 struct xfs_da_blkinfo
*info
;
1134 struct xfs_da3_icnode_hdr nodehdr
;
1140 struct xfs_inode
*dp
= state
->args
->dp
;
1142 trace_xfs_da_node_toosmall(state
->args
);
1145 * Check for the degenerate case of the block being over 50% full.
1146 * If so, it's not worth even looking to see if we might be able
1147 * to coalesce with a sibling.
1149 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1150 info
= blk
->bp
->b_addr
;
1151 node
= (xfs_da_intnode_t
*)info
;
1152 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1153 if (nodehdr
.count
> (state
->args
->geo
->node_ents
>> 1)) {
1154 *action
= 0; /* blk over 50%, don't try to join */
1155 return 0; /* blk over 50%, don't try to join */
1159 * Check for the degenerate case of the block being empty.
1160 * If the block is empty, we'll simply delete it, no need to
1161 * coalesce it with a sibling block. We choose (arbitrarily)
1162 * to merge with the forward block unless it is NULL.
1164 if (nodehdr
.count
== 0) {
1166 * Make altpath point to the block we want to keep and
1167 * path point to the block we want to drop (this one).
1169 forward
= (info
->forw
!= 0);
1170 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1171 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1184 * Examine each sibling block to see if we can coalesce with
1185 * at least 25% free space to spare. We need to figure out
1186 * whether to merge with the forward or the backward block.
1187 * We prefer coalescing with the lower numbered sibling so as
1188 * to shrink a directory over time.
1190 count
= state
->args
->geo
->node_ents
;
1191 count
-= state
->args
->geo
->node_ents
>> 2;
1192 count
-= nodehdr
.count
;
1194 /* start with smaller blk num */
1195 forward
= nodehdr
.forw
< nodehdr
.back
;
1196 for (i
= 0; i
< 2; forward
= !forward
, i
++) {
1197 struct xfs_da3_icnode_hdr thdr
;
1199 blkno
= nodehdr
.forw
;
1201 blkno
= nodehdr
.back
;
1204 error
= xfs_da3_node_read(state
->args
->trans
, dp
,
1205 blkno
, -1, &bp
, state
->args
->whichfork
);
1210 dp
->d_ops
->node_hdr_from_disk(&thdr
, node
);
1211 xfs_trans_brelse(state
->args
->trans
, bp
);
1213 if (count
- thdr
.count
>= 0)
1214 break; /* fits with at least 25% to spare */
1222 * Make altpath point to the block we want to keep (the lower
1223 * numbered block) and path point to the block we want to drop.
1225 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1226 if (blkno
< blk
->blkno
) {
1227 error
= xfs_da3_path_shift(state
, &state
->altpath
, forward
,
1230 error
= xfs_da3_path_shift(state
, &state
->path
, forward
,
1244 * Pick up the last hashvalue from an intermediate node.
1247 xfs_da3_node_lasthash(
1248 struct xfs_inode
*dp
,
1252 struct xfs_da_intnode
*node
;
1253 struct xfs_da_node_entry
*btree
;
1254 struct xfs_da3_icnode_hdr nodehdr
;
1257 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1259 *count
= nodehdr
.count
;
1262 btree
= dp
->d_ops
->node_tree_p(node
);
1263 return be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1267 * Walk back up the tree adjusting hash values as necessary,
1268 * when we stop making changes, return.
1271 xfs_da3_fixhashpath(
1272 struct xfs_da_state
*state
,
1273 struct xfs_da_state_path
*path
)
1275 struct xfs_da_state_blk
*blk
;
1276 struct xfs_da_intnode
*node
;
1277 struct xfs_da_node_entry
*btree
;
1278 xfs_dahash_t lasthash
=0;
1281 struct xfs_inode
*dp
= state
->args
->dp
;
1283 trace_xfs_da_fixhashpath(state
->args
);
1285 level
= path
->active
-1;
1286 blk
= &path
->blk
[ level
];
1287 switch (blk
->magic
) {
1288 case XFS_ATTR_LEAF_MAGIC
:
1289 lasthash
= xfs_attr_leaf_lasthash(blk
->bp
, &count
);
1293 case XFS_DIR2_LEAFN_MAGIC
:
1294 lasthash
= xfs_dir2_leaf_lasthash(dp
, blk
->bp
, &count
);
1298 case XFS_DA_NODE_MAGIC
:
1299 lasthash
= xfs_da3_node_lasthash(dp
, blk
->bp
, &count
);
1304 for (blk
--, level
--; level
>= 0; blk
--, level
--) {
1305 struct xfs_da3_icnode_hdr nodehdr
;
1307 node
= blk
->bp
->b_addr
;
1308 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1309 btree
= dp
->d_ops
->node_tree_p(node
);
1310 if (be32_to_cpu(btree
[blk
->index
].hashval
) == lasthash
)
1312 blk
->hashval
= lasthash
;
1313 btree
[blk
->index
].hashval
= cpu_to_be32(lasthash
);
1314 xfs_trans_log_buf(state
->args
->trans
, blk
->bp
,
1315 XFS_DA_LOGRANGE(node
, &btree
[blk
->index
],
1318 lasthash
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1323 * Remove an entry from an intermediate node.
1326 xfs_da3_node_remove(
1327 struct xfs_da_state
*state
,
1328 struct xfs_da_state_blk
*drop_blk
)
1330 struct xfs_da_intnode
*node
;
1331 struct xfs_da3_icnode_hdr nodehdr
;
1332 struct xfs_da_node_entry
*btree
;
1335 struct xfs_inode
*dp
= state
->args
->dp
;
1337 trace_xfs_da_node_remove(state
->args
);
1339 node
= drop_blk
->bp
->b_addr
;
1340 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1341 ASSERT(drop_blk
->index
< nodehdr
.count
);
1342 ASSERT(drop_blk
->index
>= 0);
1345 * Copy over the offending entry, or just zero it out.
1347 index
= drop_blk
->index
;
1348 btree
= dp
->d_ops
->node_tree_p(node
);
1349 if (index
< nodehdr
.count
- 1) {
1350 tmp
= nodehdr
.count
- index
- 1;
1351 tmp
*= (uint
)sizeof(xfs_da_node_entry_t
);
1352 memmove(&btree
[index
], &btree
[index
+ 1], tmp
);
1353 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1354 XFS_DA_LOGRANGE(node
, &btree
[index
], tmp
));
1355 index
= nodehdr
.count
- 1;
1357 memset(&btree
[index
], 0, sizeof(xfs_da_node_entry_t
));
1358 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1359 XFS_DA_LOGRANGE(node
, &btree
[index
], sizeof(btree
[index
])));
1361 dp
->d_ops
->node_hdr_to_disk(node
, &nodehdr
);
1362 xfs_trans_log_buf(state
->args
->trans
, drop_blk
->bp
,
1363 XFS_DA_LOGRANGE(node
, &node
->hdr
, dp
->d_ops
->node_hdr_size
));
1366 * Copy the last hash value from the block to propagate upwards.
1368 drop_blk
->hashval
= be32_to_cpu(btree
[index
- 1].hashval
);
1372 * Unbalance the elements between two intermediate nodes,
1373 * move all Btree elements from one node into another.
1376 xfs_da3_node_unbalance(
1377 struct xfs_da_state
*state
,
1378 struct xfs_da_state_blk
*drop_blk
,
1379 struct xfs_da_state_blk
*save_blk
)
1381 struct xfs_da_intnode
*drop_node
;
1382 struct xfs_da_intnode
*save_node
;
1383 struct xfs_da_node_entry
*drop_btree
;
1384 struct xfs_da_node_entry
*save_btree
;
1385 struct xfs_da3_icnode_hdr drop_hdr
;
1386 struct xfs_da3_icnode_hdr save_hdr
;
1387 struct xfs_trans
*tp
;
1390 struct xfs_inode
*dp
= state
->args
->dp
;
1392 trace_xfs_da_node_unbalance(state
->args
);
1394 drop_node
= drop_blk
->bp
->b_addr
;
1395 save_node
= save_blk
->bp
->b_addr
;
1396 dp
->d_ops
->node_hdr_from_disk(&drop_hdr
, drop_node
);
1397 dp
->d_ops
->node_hdr_from_disk(&save_hdr
, save_node
);
1398 drop_btree
= dp
->d_ops
->node_tree_p(drop_node
);
1399 save_btree
= dp
->d_ops
->node_tree_p(save_node
);
1400 tp
= state
->args
->trans
;
1403 * If the dying block has lower hashvals, then move all the
1404 * elements in the remaining block up to make a hole.
1406 if ((be32_to_cpu(drop_btree
[0].hashval
) <
1407 be32_to_cpu(save_btree
[0].hashval
)) ||
1408 (be32_to_cpu(drop_btree
[drop_hdr
.count
- 1].hashval
) <
1409 be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
))) {
1410 /* XXX: check this - is memmove dst correct? */
1411 tmp
= save_hdr
.count
* sizeof(xfs_da_node_entry_t
);
1412 memmove(&save_btree
[drop_hdr
.count
], &save_btree
[0], tmp
);
1415 xfs_trans_log_buf(tp
, save_blk
->bp
,
1416 XFS_DA_LOGRANGE(save_node
, &save_btree
[0],
1417 (save_hdr
.count
+ drop_hdr
.count
) *
1418 sizeof(xfs_da_node_entry_t
)));
1420 sindex
= save_hdr
.count
;
1421 xfs_trans_log_buf(tp
, save_blk
->bp
,
1422 XFS_DA_LOGRANGE(save_node
, &save_btree
[sindex
],
1423 drop_hdr
.count
* sizeof(xfs_da_node_entry_t
)));
1427 * Move all the B-tree elements from drop_blk to save_blk.
1429 tmp
= drop_hdr
.count
* (uint
)sizeof(xfs_da_node_entry_t
);
1430 memcpy(&save_btree
[sindex
], &drop_btree
[0], tmp
);
1431 save_hdr
.count
+= drop_hdr
.count
;
1433 dp
->d_ops
->node_hdr_to_disk(save_node
, &save_hdr
);
1434 xfs_trans_log_buf(tp
, save_blk
->bp
,
1435 XFS_DA_LOGRANGE(save_node
, &save_node
->hdr
,
1436 dp
->d_ops
->node_hdr_size
));
1439 * Save the last hashval in the remaining block for upward propagation.
1441 save_blk
->hashval
= be32_to_cpu(save_btree
[save_hdr
.count
- 1].hashval
);
1444 /*========================================================================
1445 * Routines used for finding things in the Btree.
1446 *========================================================================*/
1449 * Walk down the Btree looking for a particular filename, filling
1450 * in the state structure as we go.
1452 * We will set the state structure to point to each of the elements
1453 * in each of the nodes where either the hashval is or should be.
1455 * We support duplicate hashval's so for each entry in the current
1456 * node that could contain the desired hashval, descend. This is a
1457 * pruned depth-first tree search.
1460 xfs_da3_node_lookup_int(
1461 struct xfs_da_state
*state
,
1464 struct xfs_da_state_blk
*blk
;
1465 struct xfs_da_blkinfo
*curr
;
1466 struct xfs_da_intnode
*node
;
1467 struct xfs_da_node_entry
*btree
;
1468 struct xfs_da3_icnode_hdr nodehdr
;
1469 struct xfs_da_args
*args
;
1471 xfs_dahash_t hashval
;
1472 xfs_dahash_t btreehashval
;
1478 unsigned int expected_level
= 0;
1480 struct xfs_inode
*dp
= state
->args
->dp
;
1485 * Descend thru the B-tree searching each level for the right
1486 * node to use, until the right hashval is found.
1488 blkno
= args
->geo
->leafblk
;
1489 for (blk
= &state
->path
.blk
[0], state
->path
.active
= 1;
1490 state
->path
.active
<= XFS_DA_NODE_MAXDEPTH
;
1491 blk
++, state
->path
.active
++) {
1493 * Read the next node down in the tree.
1496 error
= xfs_da3_node_read(args
->trans
, args
->dp
, blkno
,
1497 -1, &blk
->bp
, args
->whichfork
);
1500 state
->path
.active
--;
1503 curr
= blk
->bp
->b_addr
;
1504 magic
= be16_to_cpu(curr
->magic
);
1506 if (magic
== XFS_ATTR_LEAF_MAGIC
||
1507 magic
== XFS_ATTR3_LEAF_MAGIC
) {
1508 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1509 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1513 if (magic
== XFS_DIR2_LEAFN_MAGIC
||
1514 magic
== XFS_DIR3_LEAFN_MAGIC
) {
1515 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1516 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1521 if (magic
!= XFS_DA_NODE_MAGIC
&& magic
!= XFS_DA3_NODE_MAGIC
)
1522 return -EFSCORRUPTED
;
1524 blk
->magic
= XFS_DA_NODE_MAGIC
;
1527 * Search an intermediate node for a match.
1529 node
= blk
->bp
->b_addr
;
1530 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1531 btree
= dp
->d_ops
->node_tree_p(node
);
1533 /* Tree taller than we can handle; bail out! */
1534 if (nodehdr
.level
>= XFS_DA_NODE_MAXDEPTH
)
1535 return -EFSCORRUPTED
;
1537 /* Check the level from the root. */
1538 if (blkno
== args
->geo
->leafblk
)
1539 expected_level
= nodehdr
.level
- 1;
1540 else if (expected_level
!= nodehdr
.level
)
1541 return -EFSCORRUPTED
;
1545 max
= nodehdr
.count
;
1546 blk
->hashval
= be32_to_cpu(btree
[max
- 1].hashval
);
1549 * Binary search. (note: small blocks will skip loop)
1551 probe
= span
= max
/ 2;
1552 hashval
= args
->hashval
;
1555 btreehashval
= be32_to_cpu(btree
[probe
].hashval
);
1556 if (btreehashval
< hashval
)
1558 else if (btreehashval
> hashval
)
1563 ASSERT((probe
>= 0) && (probe
< max
));
1564 ASSERT((span
<= 4) ||
1565 (be32_to_cpu(btree
[probe
].hashval
) == hashval
));
1568 * Since we may have duplicate hashval's, find the first
1569 * matching hashval in the node.
1572 be32_to_cpu(btree
[probe
].hashval
) >= hashval
) {
1575 while (probe
< max
&&
1576 be32_to_cpu(btree
[probe
].hashval
) < hashval
) {
1581 * Pick the right block to descend on.
1584 blk
->index
= max
- 1;
1585 blkno
= be32_to_cpu(btree
[max
- 1].before
);
1588 blkno
= be32_to_cpu(btree
[probe
].before
);
1591 /* We can't point back to the root. */
1592 if (blkno
== args
->geo
->leafblk
)
1593 return -EFSCORRUPTED
;
1596 if (expected_level
!= 0)
1597 return -EFSCORRUPTED
;
1600 * A leaf block that ends in the hashval that we are interested in
1601 * (final hashval == search hashval) means that the next block may
1602 * contain more entries with the same hashval, shift upward to the
1603 * next leaf and keep searching.
1606 if (blk
->magic
== XFS_DIR2_LEAFN_MAGIC
) {
1607 retval
= xfs_dir2_leafn_lookup_int(blk
->bp
, args
,
1608 &blk
->index
, state
);
1609 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1610 retval
= xfs_attr3_leaf_lookup_int(blk
->bp
, args
);
1611 blk
->index
= args
->index
;
1612 args
->blkno
= blk
->blkno
;
1615 return -EFSCORRUPTED
;
1617 if (((retval
== -ENOENT
) || (retval
== -ENOATTR
)) &&
1618 (blk
->hashval
== args
->hashval
)) {
1619 error
= xfs_da3_path_shift(state
, &state
->path
, 1, 1,
1625 } else if (blk
->magic
== XFS_ATTR_LEAF_MAGIC
) {
1626 /* path_shift() gives ENOENT */
1636 /*========================================================================
1638 *========================================================================*/
1641 * Compare two intermediate nodes for "order".
1645 struct xfs_inode
*dp
,
1646 struct xfs_buf
*node1_bp
,
1647 struct xfs_buf
*node2_bp
)
1649 struct xfs_da_intnode
*node1
;
1650 struct xfs_da_intnode
*node2
;
1651 struct xfs_da_node_entry
*btree1
;
1652 struct xfs_da_node_entry
*btree2
;
1653 struct xfs_da3_icnode_hdr node1hdr
;
1654 struct xfs_da3_icnode_hdr node2hdr
;
1656 node1
= node1_bp
->b_addr
;
1657 node2
= node2_bp
->b_addr
;
1658 dp
->d_ops
->node_hdr_from_disk(&node1hdr
, node1
);
1659 dp
->d_ops
->node_hdr_from_disk(&node2hdr
, node2
);
1660 btree1
= dp
->d_ops
->node_tree_p(node1
);
1661 btree2
= dp
->d_ops
->node_tree_p(node2
);
1663 if (node1hdr
.count
> 0 && node2hdr
.count
> 0 &&
1664 ((be32_to_cpu(btree2
[0].hashval
) < be32_to_cpu(btree1
[0].hashval
)) ||
1665 (be32_to_cpu(btree2
[node2hdr
.count
- 1].hashval
) <
1666 be32_to_cpu(btree1
[node1hdr
.count
- 1].hashval
)))) {
1673 * Link a new block into a doubly linked list of blocks (of whatever type).
1677 struct xfs_da_state
*state
,
1678 struct xfs_da_state_blk
*old_blk
,
1679 struct xfs_da_state_blk
*new_blk
)
1681 struct xfs_da_blkinfo
*old_info
;
1682 struct xfs_da_blkinfo
*new_info
;
1683 struct xfs_da_blkinfo
*tmp_info
;
1684 struct xfs_da_args
*args
;
1688 struct xfs_inode
*dp
= state
->args
->dp
;
1691 * Set up environment.
1694 ASSERT(args
!= NULL
);
1695 old_info
= old_blk
->bp
->b_addr
;
1696 new_info
= new_blk
->bp
->b_addr
;
1697 ASSERT(old_blk
->magic
== XFS_DA_NODE_MAGIC
||
1698 old_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1699 old_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1701 switch (old_blk
->magic
) {
1702 case XFS_ATTR_LEAF_MAGIC
:
1703 before
= xfs_attr_leaf_order(old_blk
->bp
, new_blk
->bp
);
1705 case XFS_DIR2_LEAFN_MAGIC
:
1706 before
= xfs_dir2_leafn_order(dp
, old_blk
->bp
, new_blk
->bp
);
1708 case XFS_DA_NODE_MAGIC
:
1709 before
= xfs_da3_node_order(dp
, old_blk
->bp
, new_blk
->bp
);
1714 * Link blocks in appropriate order.
1718 * Link new block in before existing block.
1720 trace_xfs_da_link_before(args
);
1721 new_info
->forw
= cpu_to_be32(old_blk
->blkno
);
1722 new_info
->back
= old_info
->back
;
1723 if (old_info
->back
) {
1724 error
= xfs_da3_node_read(args
->trans
, dp
,
1725 be32_to_cpu(old_info
->back
),
1726 -1, &bp
, args
->whichfork
);
1730 tmp_info
= bp
->b_addr
;
1731 ASSERT(tmp_info
->magic
== old_info
->magic
);
1732 ASSERT(be32_to_cpu(tmp_info
->forw
) == old_blk
->blkno
);
1733 tmp_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1734 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1736 old_info
->back
= cpu_to_be32(new_blk
->blkno
);
1739 * Link new block in after existing block.
1741 trace_xfs_da_link_after(args
);
1742 new_info
->forw
= old_info
->forw
;
1743 new_info
->back
= cpu_to_be32(old_blk
->blkno
);
1744 if (old_info
->forw
) {
1745 error
= xfs_da3_node_read(args
->trans
, dp
,
1746 be32_to_cpu(old_info
->forw
),
1747 -1, &bp
, args
->whichfork
);
1751 tmp_info
= bp
->b_addr
;
1752 ASSERT(tmp_info
->magic
== old_info
->magic
);
1753 ASSERT(be32_to_cpu(tmp_info
->back
) == old_blk
->blkno
);
1754 tmp_info
->back
= cpu_to_be32(new_blk
->blkno
);
1755 xfs_trans_log_buf(args
->trans
, bp
, 0, sizeof(*tmp_info
)-1);
1757 old_info
->forw
= cpu_to_be32(new_blk
->blkno
);
1760 xfs_trans_log_buf(args
->trans
, old_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1761 xfs_trans_log_buf(args
->trans
, new_blk
->bp
, 0, sizeof(*tmp_info
) - 1);
1766 * Unlink a block from a doubly linked list of blocks.
1768 STATIC
int /* error */
1770 struct xfs_da_state
*state
,
1771 struct xfs_da_state_blk
*drop_blk
,
1772 struct xfs_da_state_blk
*save_blk
)
1774 struct xfs_da_blkinfo
*drop_info
;
1775 struct xfs_da_blkinfo
*save_info
;
1776 struct xfs_da_blkinfo
*tmp_info
;
1777 struct xfs_da_args
*args
;
1782 * Set up environment.
1785 ASSERT(args
!= NULL
);
1786 save_info
= save_blk
->bp
->b_addr
;
1787 drop_info
= drop_blk
->bp
->b_addr
;
1788 ASSERT(save_blk
->magic
== XFS_DA_NODE_MAGIC
||
1789 save_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
||
1790 save_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1791 ASSERT(save_blk
->magic
== drop_blk
->magic
);
1792 ASSERT((be32_to_cpu(save_info
->forw
) == drop_blk
->blkno
) ||
1793 (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
));
1794 ASSERT((be32_to_cpu(drop_info
->forw
) == save_blk
->blkno
) ||
1795 (be32_to_cpu(drop_info
->back
) == save_blk
->blkno
));
1798 * Unlink the leaf block from the doubly linked chain of leaves.
1800 if (be32_to_cpu(save_info
->back
) == drop_blk
->blkno
) {
1801 trace_xfs_da_unlink_back(args
);
1802 save_info
->back
= drop_info
->back
;
1803 if (drop_info
->back
) {
1804 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1805 be32_to_cpu(drop_info
->back
),
1806 -1, &bp
, args
->whichfork
);
1810 tmp_info
= bp
->b_addr
;
1811 ASSERT(tmp_info
->magic
== save_info
->magic
);
1812 ASSERT(be32_to_cpu(tmp_info
->forw
) == drop_blk
->blkno
);
1813 tmp_info
->forw
= cpu_to_be32(save_blk
->blkno
);
1814 xfs_trans_log_buf(args
->trans
, bp
, 0,
1815 sizeof(*tmp_info
) - 1);
1818 trace_xfs_da_unlink_forward(args
);
1819 save_info
->forw
= drop_info
->forw
;
1820 if (drop_info
->forw
) {
1821 error
= xfs_da3_node_read(args
->trans
, args
->dp
,
1822 be32_to_cpu(drop_info
->forw
),
1823 -1, &bp
, args
->whichfork
);
1827 tmp_info
= bp
->b_addr
;
1828 ASSERT(tmp_info
->magic
== save_info
->magic
);
1829 ASSERT(be32_to_cpu(tmp_info
->back
) == drop_blk
->blkno
);
1830 tmp_info
->back
= cpu_to_be32(save_blk
->blkno
);
1831 xfs_trans_log_buf(args
->trans
, bp
, 0,
1832 sizeof(*tmp_info
) - 1);
1836 xfs_trans_log_buf(args
->trans
, save_blk
->bp
, 0, sizeof(*save_info
) - 1);
1841 * Move a path "forward" or "!forward" one block at the current level.
1843 * This routine will adjust a "path" to point to the next block
1844 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1845 * Btree, including updating pointers to the intermediate nodes between
1846 * the new bottom and the root.
1850 struct xfs_da_state
*state
,
1851 struct xfs_da_state_path
*path
,
1856 struct xfs_da_state_blk
*blk
;
1857 struct xfs_da_blkinfo
*info
;
1858 struct xfs_da_intnode
*node
;
1859 struct xfs_da_args
*args
;
1860 struct xfs_da_node_entry
*btree
;
1861 struct xfs_da3_icnode_hdr nodehdr
;
1863 xfs_dablk_t blkno
= 0;
1866 struct xfs_inode
*dp
= state
->args
->dp
;
1868 trace_xfs_da_path_shift(state
->args
);
1871 * Roll up the Btree looking for the first block where our
1872 * current index is not at the edge of the block. Note that
1873 * we skip the bottom layer because we want the sibling block.
1876 ASSERT(args
!= NULL
);
1877 ASSERT(path
!= NULL
);
1878 ASSERT((path
->active
> 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1879 level
= (path
->active
-1) - 1; /* skip bottom layer in path */
1880 for (blk
= &path
->blk
[level
]; level
>= 0; blk
--, level
--) {
1881 node
= blk
->bp
->b_addr
;
1882 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1883 btree
= dp
->d_ops
->node_tree_p(node
);
1885 if (forward
&& (blk
->index
< nodehdr
.count
- 1)) {
1887 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1889 } else if (!forward
&& (blk
->index
> 0)) {
1891 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1896 *result
= -ENOENT
; /* we're out of our tree */
1897 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1902 * Roll down the edge of the subtree until we reach the
1903 * same depth we were at originally.
1905 for (blk
++, level
++; level
< path
->active
; blk
++, level
++) {
1907 * Read the next child block into a local buffer.
1909 error
= xfs_da3_node_read(args
->trans
, dp
, blkno
, -1, &bp
,
1915 * Release the old block (if it's dirty, the trans doesn't
1916 * actually let go) and swap the local buffer into the path
1917 * structure. This ensures failure of the above read doesn't set
1918 * a NULL buffer in an active slot in the path.
1921 xfs_trans_brelse(args
->trans
, blk
->bp
);
1925 info
= blk
->bp
->b_addr
;
1926 ASSERT(info
->magic
== cpu_to_be16(XFS_DA_NODE_MAGIC
) ||
1927 info
->magic
== cpu_to_be16(XFS_DA3_NODE_MAGIC
) ||
1928 info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1929 info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
) ||
1930 info
->magic
== cpu_to_be16(XFS_ATTR_LEAF_MAGIC
) ||
1931 info
->magic
== cpu_to_be16(XFS_ATTR3_LEAF_MAGIC
));
1935 * Note: we flatten the magic number to a single type so we
1936 * don't have to compare against crc/non-crc types elsewhere.
1938 switch (be16_to_cpu(info
->magic
)) {
1939 case XFS_DA_NODE_MAGIC
:
1940 case XFS_DA3_NODE_MAGIC
:
1941 blk
->magic
= XFS_DA_NODE_MAGIC
;
1942 node
= (xfs_da_intnode_t
*)info
;
1943 dp
->d_ops
->node_hdr_from_disk(&nodehdr
, node
);
1944 btree
= dp
->d_ops
->node_tree_p(node
);
1945 blk
->hashval
= be32_to_cpu(btree
[nodehdr
.count
- 1].hashval
);
1949 blk
->index
= nodehdr
.count
- 1;
1950 blkno
= be32_to_cpu(btree
[blk
->index
].before
);
1952 case XFS_ATTR_LEAF_MAGIC
:
1953 case XFS_ATTR3_LEAF_MAGIC
:
1954 blk
->magic
= XFS_ATTR_LEAF_MAGIC
;
1955 ASSERT(level
== path
->active
-1);
1957 blk
->hashval
= xfs_attr_leaf_lasthash(blk
->bp
, NULL
);
1959 case XFS_DIR2_LEAFN_MAGIC
:
1960 case XFS_DIR3_LEAFN_MAGIC
:
1961 blk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1962 ASSERT(level
== path
->active
-1);
1964 blk
->hashval
= xfs_dir2_leaf_lasthash(args
->dp
,
1977 /*========================================================================
1979 *========================================================================*/
1982 * Implement a simple hash on a character string.
1983 * Rotate the hash value by 7 bits, then XOR each character in.
1984 * This is implemented with some source-level loop unrolling.
1987 xfs_da_hashname(const uint8_t *name
, int namelen
)
1992 * Do four characters at a time as long as we can.
1994 for (hash
= 0; namelen
>= 4; namelen
-= 4, name
+= 4)
1995 hash
= (name
[0] << 21) ^ (name
[1] << 14) ^ (name
[2] << 7) ^
1996 (name
[3] << 0) ^ rol32(hash
, 7 * 4);
1999 * Now do the rest of the characters.
2003 return (name
[0] << 14) ^ (name
[1] << 7) ^ (name
[2] << 0) ^
2006 return (name
[0] << 7) ^ (name
[1] << 0) ^ rol32(hash
, 7 * 2);
2008 return (name
[0] << 0) ^ rol32(hash
, 7 * 1);
2009 default: /* case 0: */
2016 struct xfs_da_args
*args
,
2017 const unsigned char *name
,
2020 return (args
->namelen
== len
&& memcmp(args
->name
, name
, len
) == 0) ?
2021 XFS_CMP_EXACT
: XFS_CMP_DIFFERENT
;
2025 xfs_default_hashname(
2026 struct xfs_name
*name
)
2028 return xfs_da_hashname(name
->name
, name
->len
);
2031 const struct xfs_nameops xfs_default_nameops
= {
2032 .hashname
= xfs_default_hashname
,
2033 .compname
= xfs_da_compname
2037 xfs_da_grow_inode_int(
2038 struct xfs_da_args
*args
,
2042 struct xfs_trans
*tp
= args
->trans
;
2043 struct xfs_inode
*dp
= args
->dp
;
2044 int w
= args
->whichfork
;
2045 xfs_rfsblock_t nblks
= dp
->i_d
.di_nblocks
;
2046 struct xfs_bmbt_irec map
, *mapp
;
2047 int nmap
, error
, got
, i
, mapi
;
2050 * Find a spot in the file space to put the new block.
2052 error
= xfs_bmap_first_unused(tp
, dp
, count
, bno
, w
);
2057 * Try mapping it in one filesystem block.
2060 error
= xfs_bmapi_write(tp
, dp
, *bno
, count
,
2061 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
|XFS_BMAPI_CONTIG
,
2062 args
->total
, &map
, &nmap
);
2070 } else if (nmap
== 0 && count
> 1) {
2075 * If we didn't get it and the block might work if fragmented,
2076 * try without the CONTIG flag. Loop until we get it all.
2078 mapp
= kmem_alloc(sizeof(*mapp
) * count
, KM_SLEEP
);
2079 for (b
= *bno
, mapi
= 0; b
< *bno
+ count
; ) {
2080 nmap
= min(XFS_BMAP_MAX_NMAP
, count
);
2081 c
= (int)(*bno
+ count
- b
);
2082 error
= xfs_bmapi_write(tp
, dp
, b
, c
,
2083 xfs_bmapi_aflag(w
)|XFS_BMAPI_METADATA
,
2084 args
->total
, &mapp
[mapi
], &nmap
);
2090 b
= mapp
[mapi
- 1].br_startoff
+
2091 mapp
[mapi
- 1].br_blockcount
;
2099 * Count the blocks we got, make sure it matches the total.
2101 for (i
= 0, got
= 0; i
< mapi
; i
++)
2102 got
+= mapp
[i
].br_blockcount
;
2103 if (got
!= count
|| mapp
[0].br_startoff
!= *bno
||
2104 mapp
[mapi
- 1].br_startoff
+ mapp
[mapi
- 1].br_blockcount
!=
2110 /* account for newly allocated blocks in reserved blocks total */
2111 args
->total
-= dp
->i_d
.di_nblocks
- nblks
;
2120 * Add a block to the btree ahead of the file.
2121 * Return the new block number to the caller.
2125 struct xfs_da_args
*args
,
2126 xfs_dablk_t
*new_blkno
)
2131 trace_xfs_da_grow_inode(args
);
2133 bno
= args
->geo
->leafblk
;
2134 error
= xfs_da_grow_inode_int(args
, &bno
, args
->geo
->fsbcount
);
2136 *new_blkno
= (xfs_dablk_t
)bno
;
2141 * Ick. We need to always be able to remove a btree block, even
2142 * if there's no space reservation because the filesystem is full.
2143 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2144 * It swaps the target block with the last block in the file. The
2145 * last block in the file can always be removed since it can't cause
2146 * a bmap btree split to do that.
2149 xfs_da3_swap_lastblock(
2150 struct xfs_da_args
*args
,
2151 xfs_dablk_t
*dead_blknop
,
2152 struct xfs_buf
**dead_bufp
)
2154 struct xfs_da_blkinfo
*dead_info
;
2155 struct xfs_da_blkinfo
*sib_info
;
2156 struct xfs_da_intnode
*par_node
;
2157 struct xfs_da_intnode
*dead_node
;
2158 struct xfs_dir2_leaf
*dead_leaf2
;
2159 struct xfs_da_node_entry
*btree
;
2160 struct xfs_da3_icnode_hdr par_hdr
;
2161 struct xfs_inode
*dp
;
2162 struct xfs_trans
*tp
;
2163 struct xfs_mount
*mp
;
2164 struct xfs_buf
*dead_buf
;
2165 struct xfs_buf
*last_buf
;
2166 struct xfs_buf
*sib_buf
;
2167 struct xfs_buf
*par_buf
;
2168 xfs_dahash_t dead_hash
;
2169 xfs_fileoff_t lastoff
;
2170 xfs_dablk_t dead_blkno
;
2171 xfs_dablk_t last_blkno
;
2172 xfs_dablk_t sib_blkno
;
2173 xfs_dablk_t par_blkno
;
2180 trace_xfs_da_swap_lastblock(args
);
2182 dead_buf
= *dead_bufp
;
2183 dead_blkno
= *dead_blknop
;
2186 w
= args
->whichfork
;
2187 ASSERT(w
== XFS_DATA_FORK
);
2189 lastoff
= args
->geo
->freeblk
;
2190 error
= xfs_bmap_last_before(tp
, dp
, &lastoff
, w
);
2193 if (unlikely(lastoff
== 0)) {
2194 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW
,
2196 return -EFSCORRUPTED
;
2199 * Read the last block in the btree space.
2201 last_blkno
= (xfs_dablk_t
)lastoff
- args
->geo
->fsbcount
;
2202 error
= xfs_da3_node_read(tp
, dp
, last_blkno
, -1, &last_buf
, w
);
2206 * Copy the last block into the dead buffer and log it.
2208 memcpy(dead_buf
->b_addr
, last_buf
->b_addr
, args
->geo
->blksize
);
2209 xfs_trans_log_buf(tp
, dead_buf
, 0, args
->geo
->blksize
- 1);
2210 dead_info
= dead_buf
->b_addr
;
2212 * Get values from the moved block.
2214 if (dead_info
->magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
2215 dead_info
->magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
)) {
2216 struct xfs_dir3_icleaf_hdr leafhdr
;
2217 struct xfs_dir2_leaf_entry
*ents
;
2219 dead_leaf2
= (xfs_dir2_leaf_t
*)dead_info
;
2220 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, dead_leaf2
);
2221 ents
= dp
->d_ops
->leaf_ents_p(dead_leaf2
);
2223 dead_hash
= be32_to_cpu(ents
[leafhdr
.count
- 1].hashval
);
2225 struct xfs_da3_icnode_hdr deadhdr
;
2227 dead_node
= (xfs_da_intnode_t
*)dead_info
;
2228 dp
->d_ops
->node_hdr_from_disk(&deadhdr
, dead_node
);
2229 btree
= dp
->d_ops
->node_tree_p(dead_node
);
2230 dead_level
= deadhdr
.level
;
2231 dead_hash
= be32_to_cpu(btree
[deadhdr
.count
- 1].hashval
);
2233 sib_buf
= par_buf
= NULL
;
2235 * If the moved block has a left sibling, fix up the pointers.
2237 if ((sib_blkno
= be32_to_cpu(dead_info
->back
))) {
2238 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2241 sib_info
= sib_buf
->b_addr
;
2243 be32_to_cpu(sib_info
->forw
) != last_blkno
||
2244 sib_info
->magic
!= dead_info
->magic
)) {
2245 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2246 XFS_ERRLEVEL_LOW
, mp
);
2247 error
= -EFSCORRUPTED
;
2250 sib_info
->forw
= cpu_to_be32(dead_blkno
);
2251 xfs_trans_log_buf(tp
, sib_buf
,
2252 XFS_DA_LOGRANGE(sib_info
, &sib_info
->forw
,
2253 sizeof(sib_info
->forw
)));
2257 * If the moved block has a right sibling, fix up the pointers.
2259 if ((sib_blkno
= be32_to_cpu(dead_info
->forw
))) {
2260 error
= xfs_da3_node_read(tp
, dp
, sib_blkno
, -1, &sib_buf
, w
);
2263 sib_info
= sib_buf
->b_addr
;
2265 be32_to_cpu(sib_info
->back
) != last_blkno
||
2266 sib_info
->magic
!= dead_info
->magic
)) {
2267 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2268 XFS_ERRLEVEL_LOW
, mp
);
2269 error
= -EFSCORRUPTED
;
2272 sib_info
->back
= cpu_to_be32(dead_blkno
);
2273 xfs_trans_log_buf(tp
, sib_buf
,
2274 XFS_DA_LOGRANGE(sib_info
, &sib_info
->back
,
2275 sizeof(sib_info
->back
)));
2278 par_blkno
= args
->geo
->leafblk
;
2281 * Walk down the tree looking for the parent of the moved block.
2284 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2287 par_node
= par_buf
->b_addr
;
2288 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2289 if (level
>= 0 && level
!= par_hdr
.level
+ 1) {
2290 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2291 XFS_ERRLEVEL_LOW
, mp
);
2292 error
= -EFSCORRUPTED
;
2295 level
= par_hdr
.level
;
2296 btree
= dp
->d_ops
->node_tree_p(par_node
);
2298 entno
< par_hdr
.count
&&
2299 be32_to_cpu(btree
[entno
].hashval
) < dead_hash
;
2302 if (entno
== par_hdr
.count
) {
2303 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2304 XFS_ERRLEVEL_LOW
, mp
);
2305 error
= -EFSCORRUPTED
;
2308 par_blkno
= be32_to_cpu(btree
[entno
].before
);
2309 if (level
== dead_level
+ 1)
2311 xfs_trans_brelse(tp
, par_buf
);
2315 * We're in the right parent block.
2316 * Look for the right entry.
2320 entno
< par_hdr
.count
&&
2321 be32_to_cpu(btree
[entno
].before
) != last_blkno
;
2324 if (entno
< par_hdr
.count
)
2326 par_blkno
= par_hdr
.forw
;
2327 xfs_trans_brelse(tp
, par_buf
);
2329 if (unlikely(par_blkno
== 0)) {
2330 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2331 XFS_ERRLEVEL_LOW
, mp
);
2332 error
= -EFSCORRUPTED
;
2335 error
= xfs_da3_node_read(tp
, dp
, par_blkno
, -1, &par_buf
, w
);
2338 par_node
= par_buf
->b_addr
;
2339 dp
->d_ops
->node_hdr_from_disk(&par_hdr
, par_node
);
2340 if (par_hdr
.level
!= level
) {
2341 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2342 XFS_ERRLEVEL_LOW
, mp
);
2343 error
= -EFSCORRUPTED
;
2346 btree
= dp
->d_ops
->node_tree_p(par_node
);
2350 * Update the parent entry pointing to the moved block.
2352 btree
[entno
].before
= cpu_to_be32(dead_blkno
);
2353 xfs_trans_log_buf(tp
, par_buf
,
2354 XFS_DA_LOGRANGE(par_node
, &btree
[entno
].before
,
2355 sizeof(btree
[entno
].before
)));
2356 *dead_blknop
= last_blkno
;
2357 *dead_bufp
= last_buf
;
2361 xfs_trans_brelse(tp
, par_buf
);
2363 xfs_trans_brelse(tp
, sib_buf
);
2364 xfs_trans_brelse(tp
, last_buf
);
2369 * Remove a btree block from a directory or attribute.
2372 xfs_da_shrink_inode(
2373 struct xfs_da_args
*args
,
2374 xfs_dablk_t dead_blkno
,
2375 struct xfs_buf
*dead_buf
)
2377 struct xfs_inode
*dp
;
2378 int done
, error
, w
, count
;
2379 struct xfs_trans
*tp
;
2381 trace_xfs_da_shrink_inode(args
);
2384 w
= args
->whichfork
;
2386 count
= args
->geo
->fsbcount
;
2389 * Remove extents. If we get ENOSPC for a dir we have to move
2390 * the last block to the place we want to kill.
2392 error
= xfs_bunmapi(tp
, dp
, dead_blkno
, count
,
2393 xfs_bmapi_aflag(w
), 0, &done
);
2394 if (error
== -ENOSPC
) {
2395 if (w
!= XFS_DATA_FORK
)
2397 error
= xfs_da3_swap_lastblock(args
, &dead_blkno
,
2405 xfs_trans_binval(tp
, dead_buf
);
2410 * See if the mapping(s) for this btree block are valid, i.e.
2411 * don't contain holes, are logically contiguous, and cover the whole range.
2414 xfs_da_map_covers_blocks(
2416 xfs_bmbt_irec_t
*mapp
,
2423 for (i
= 0, off
= bno
; i
< nmap
; i
++) {
2424 if (mapp
[i
].br_startblock
== HOLESTARTBLOCK
||
2425 mapp
[i
].br_startblock
== DELAYSTARTBLOCK
) {
2428 if (off
!= mapp
[i
].br_startoff
) {
2431 off
+= mapp
[i
].br_blockcount
;
2433 return off
== bno
+ count
;
2437 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2439 * For the single map case, it is assumed that the caller has provided a pointer
2440 * to a valid xfs_buf_map. For the multiple map case, this function will
2441 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2442 * map pointer with the allocated map.
2445 xfs_buf_map_from_irec(
2446 struct xfs_mount
*mp
,
2447 struct xfs_buf_map
**mapp
,
2449 struct xfs_bmbt_irec
*irecs
,
2452 struct xfs_buf_map
*map
;
2455 ASSERT(*nmaps
== 1);
2456 ASSERT(nirecs
>= 1);
2459 map
= kmem_zalloc(nirecs
* sizeof(struct xfs_buf_map
),
2460 KM_SLEEP
| KM_NOFS
);
2468 for (i
= 0; i
< *nmaps
; i
++) {
2469 ASSERT(irecs
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
2470 irecs
[i
].br_startblock
!= HOLESTARTBLOCK
);
2471 map
[i
].bm_bn
= XFS_FSB_TO_DADDR(mp
, irecs
[i
].br_startblock
);
2472 map
[i
].bm_len
= XFS_FSB_TO_BB(mp
, irecs
[i
].br_blockcount
);
2478 * Map the block we are given ready for reading. There are three possible return
2480 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2481 * caller knows not to execute a subsequent read.
2482 * 0 - if we mapped the block successfully
2483 * >0 - positive error number if there was an error.
2487 struct xfs_inode
*dp
,
2489 xfs_daddr_t mappedbno
,
2491 struct xfs_buf_map
**map
,
2494 struct xfs_mount
*mp
= dp
->i_mount
;
2497 struct xfs_bmbt_irec irec
;
2498 struct xfs_bmbt_irec
*irecs
= &irec
;
2501 ASSERT(map
&& *map
);
2502 ASSERT(*nmaps
== 1);
2504 if (whichfork
== XFS_DATA_FORK
)
2505 nfsb
= mp
->m_dir_geo
->fsbcount
;
2507 nfsb
= mp
->m_attr_geo
->fsbcount
;
2510 * Caller doesn't have a mapping. -2 means don't complain
2511 * if we land in a hole.
2513 if (mappedbno
== -1 || mappedbno
== -2) {
2515 * Optimize the one-block case.
2518 irecs
= kmem_zalloc(sizeof(irec
) * nfsb
,
2519 KM_SLEEP
| KM_NOFS
);
2522 error
= xfs_bmapi_read(dp
, (xfs_fileoff_t
)bno
, nfsb
, irecs
,
2523 &nirecs
, xfs_bmapi_aflag(whichfork
));
2527 irecs
->br_startblock
= XFS_DADDR_TO_FSB(mp
, mappedbno
);
2528 irecs
->br_startoff
= (xfs_fileoff_t
)bno
;
2529 irecs
->br_blockcount
= nfsb
;
2530 irecs
->br_state
= 0;
2534 if (!xfs_da_map_covers_blocks(nirecs
, irecs
, bno
, nfsb
)) {
2535 error
= mappedbno
== -2 ? -1 : -EFSCORRUPTED
;
2536 if (unlikely(error
== -EFSCORRUPTED
)) {
2537 if (xfs_error_level
>= XFS_ERRLEVEL_LOW
) {
2539 xfs_alert(mp
, "%s: bno %lld dir: inode %lld",
2540 __func__
, (long long)bno
,
2541 (long long)dp
->i_ino
);
2542 for (i
= 0; i
< *nmaps
; i
++) {
2544 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2546 (long long)irecs
[i
].br_startoff
,
2547 (long long)irecs
[i
].br_startblock
,
2548 (long long)irecs
[i
].br_blockcount
,
2552 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2553 XFS_ERRLEVEL_LOW
, mp
);
2557 error
= xfs_buf_map_from_irec(mp
, map
, nmaps
, irecs
, nirecs
);
2565 * Get a buffer for the dir/attr block.
2569 struct xfs_trans
*trans
,
2570 struct xfs_inode
*dp
,
2572 xfs_daddr_t mappedbno
,
2573 struct xfs_buf
**bpp
,
2577 struct xfs_buf_map map
;
2578 struct xfs_buf_map
*mapp
;
2585 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2588 /* mapping a hole is not an error, but we don't continue */
2594 bp
= xfs_trans_get_buf_map(trans
, dp
->i_mount
->m_ddev_targp
,
2596 error
= bp
? bp
->b_error
: -EIO
;
2599 xfs_trans_brelse(trans
, bp
);
2613 * Get a buffer for the dir/attr block, fill in the contents.
2617 struct xfs_trans
*trans
,
2618 struct xfs_inode
*dp
,
2620 xfs_daddr_t mappedbno
,
2621 struct xfs_buf
**bpp
,
2623 const struct xfs_buf_ops
*ops
)
2626 struct xfs_buf_map map
;
2627 struct xfs_buf_map
*mapp
;
2634 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2637 /* mapping a hole is not an error, but we don't continue */
2643 error
= xfs_trans_read_buf_map(dp
->i_mount
, trans
,
2644 dp
->i_mount
->m_ddev_targp
,
2645 mapp
, nmap
, 0, &bp
, ops
);
2649 if (whichfork
== XFS_ATTR_FORK
)
2650 xfs_buf_set_ref(bp
, XFS_ATTR_BTREE_REF
);
2652 xfs_buf_set_ref(bp
, XFS_DIR_BTREE_REF
);
2662 * Readahead the dir/attr block.
2666 struct xfs_inode
*dp
,
2668 xfs_daddr_t mappedbno
,
2670 const struct xfs_buf_ops
*ops
)
2672 struct xfs_buf_map map
;
2673 struct xfs_buf_map
*mapp
;
2679 error
= xfs_dabuf_map(dp
, bno
, mappedbno
, whichfork
,
2682 /* mapping a hole is not an error, but we don't continue */
2688 mappedbno
= mapp
[0].bm_bn
;
2689 xfs_buf_readahead_map(dp
->i_mount
->m_ddev_targp
, mapp
, nmap
, ops
);