2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_alloc.h"
35 #include "xfs_btree.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_error.h"
49 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52 /*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
57 * Routines used for growing the Btree.
59 STATIC
int xfs_attr_leaf_create(xfs_da_args_t
*args
, xfs_dablk_t which_block
,
61 STATIC
int xfs_attr_leaf_add_work(xfs_dabuf_t
*leaf_buffer
, xfs_da_args_t
*args
,
63 STATIC
void xfs_attr_leaf_compact(xfs_trans_t
*trans
, xfs_dabuf_t
*leaf_buffer
);
64 STATIC
void xfs_attr_leaf_rebalance(xfs_da_state_t
*state
,
65 xfs_da_state_blk_t
*blk1
,
66 xfs_da_state_blk_t
*blk2
);
67 STATIC
int xfs_attr_leaf_figure_balance(xfs_da_state_t
*state
,
68 xfs_da_state_blk_t
*leaf_blk_1
,
69 xfs_da_state_blk_t
*leaf_blk_2
,
70 int *number_entries_in_blk1
,
71 int *number_usedbytes_in_blk1
);
74 * Routines used for shrinking the Btree.
76 STATIC
int xfs_attr_node_inactive(xfs_trans_t
**trans
, xfs_inode_t
*dp
,
77 xfs_dabuf_t
*bp
, int level
);
78 STATIC
int xfs_attr_leaf_inactive(xfs_trans_t
**trans
, xfs_inode_t
*dp
,
80 STATIC
int xfs_attr_leaf_freextent(xfs_trans_t
**trans
, xfs_inode_t
*dp
,
81 xfs_dablk_t blkno
, int blkcnt
);
86 STATIC
void xfs_attr_leaf_moveents(xfs_attr_leafblock_t
*src_leaf
,
88 xfs_attr_leafblock_t
*dst_leaf
,
89 int dst_start
, int move_count
,
91 STATIC
int xfs_attr_leaf_entsize(xfs_attr_leafblock_t
*leaf
, int index
);
92 STATIC
int xfs_attr_put_listent(xfs_attr_list_context_t
*context
,
93 attrnames_t
*, char *name
, int namelen
,
97 /*========================================================================
98 * External routines when attribute fork size < XFS_LITINO(mp).
99 *========================================================================*/
102 * Query whether the requested number of additional bytes of extended
103 * attribute space will be able to fit inline.
104 * Returns zero if not, else the di_forkoff fork offset to be used in the
105 * literal area for attribute data once the new bytes have been added.
107 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
108 * special case for dev/uuid inodes, they have fixed size data forks.
111 xfs_attr_shortform_bytesfit(xfs_inode_t
*dp
, int bytes
)
114 int minforkoff
; /* lower limit on valid forkoff locations */
115 int maxforkoff
; /* upper limit on valid forkoff locations */
116 xfs_mount_t
*mp
= dp
->i_mount
;
118 offset
= (XFS_LITINO(mp
) - bytes
) >> 3; /* rounded down */
120 switch (dp
->i_d
.di_format
) {
121 case XFS_DINODE_FMT_DEV
:
122 minforkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
123 return (offset
>= minforkoff
) ? minforkoff
: 0;
124 case XFS_DINODE_FMT_UUID
:
125 minforkoff
= roundup(sizeof(uuid_t
), 8) >> 3;
126 return (offset
>= minforkoff
) ? minforkoff
: 0;
129 if (!(mp
->m_flags
& XFS_MOUNT_ATTR2
)) {
130 if (bytes
<= XFS_IFORK_ASIZE(dp
))
131 return mp
->m_attroffset
>> 3;
135 /* data fork btree root can have at least this many key/ptr pairs */
136 minforkoff
= MAX(dp
->i_df
.if_bytes
, XFS_BMDR_SPACE_CALC(MINDBTPTRS
));
137 minforkoff
= roundup(minforkoff
, 8) >> 3;
139 /* attr fork btree root can have at least this many key/ptr pairs */
140 maxforkoff
= XFS_LITINO(mp
) - XFS_BMDR_SPACE_CALC(MINABTPTRS
);
141 maxforkoff
= maxforkoff
>> 3; /* rounded down */
143 if (offset
>= minforkoff
&& offset
< maxforkoff
)
145 if (offset
>= maxforkoff
)
151 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
154 xfs_sbversion_add_attr2(xfs_mount_t
*mp
, xfs_trans_t
*tp
)
158 if ((mp
->m_flags
& XFS_MOUNT_ATTR2
) &&
159 !(XFS_SB_VERSION_HASATTR2(&mp
->m_sb
))) {
161 if (!XFS_SB_VERSION_HASATTR2(&mp
->m_sb
)) {
162 XFS_SB_VERSION_ADDATTR2(&mp
->m_sb
);
163 XFS_SB_UNLOCK(mp
, s
);
164 xfs_mod_sb(tp
, XFS_SB_VERSIONNUM
| XFS_SB_FEATURES2
);
166 XFS_SB_UNLOCK(mp
, s
);
171 * Create the initial contents of a shortform attribute list.
174 xfs_attr_shortform_create(xfs_da_args_t
*args
)
176 xfs_attr_sf_hdr_t
*hdr
;
184 ASSERT(ifp
->if_bytes
== 0);
185 if (dp
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
) {
186 ifp
->if_flags
&= ~XFS_IFEXTENTS
; /* just in case */
187 dp
->i_d
.di_aformat
= XFS_DINODE_FMT_LOCAL
;
188 ifp
->if_flags
|= XFS_IFINLINE
;
190 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
192 xfs_idata_realloc(dp
, sizeof(*hdr
), XFS_ATTR_FORK
);
193 hdr
= (xfs_attr_sf_hdr_t
*)ifp
->if_u1
.if_data
;
195 hdr
->totsize
= cpu_to_be16(sizeof(*hdr
));
196 xfs_trans_log_inode(args
->trans
, dp
, XFS_ILOG_CORE
| XFS_ILOG_ADATA
);
200 * Add a name/value pair to the shortform attribute list.
201 * Overflow from the inode has already been checked for.
204 xfs_attr_shortform_add(xfs_da_args_t
*args
, int forkoff
)
206 xfs_attr_shortform_t
*sf
;
207 xfs_attr_sf_entry_t
*sfe
;
215 dp
->i_d
.di_forkoff
= forkoff
;
216 dp
->i_df
.if_ext_max
=
217 XFS_IFORK_DSIZE(dp
) / (uint
)sizeof(xfs_bmbt_rec_t
);
218 dp
->i_afp
->if_ext_max
=
219 XFS_IFORK_ASIZE(dp
) / (uint
)sizeof(xfs_bmbt_rec_t
);
222 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
223 sf
= (xfs_attr_shortform_t
*)ifp
->if_u1
.if_data
;
225 for (i
= 0; i
< sf
->hdr
.count
; sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
), i
++) {
227 if (sfe
->namelen
!= args
->namelen
)
229 if (memcmp(args
->name
, sfe
->nameval
, args
->namelen
) != 0)
231 if (((args
->flags
& ATTR_SECURE
) != 0) !=
232 ((sfe
->flags
& XFS_ATTR_SECURE
) != 0))
234 if (((args
->flags
& ATTR_ROOT
) != 0) !=
235 ((sfe
->flags
& XFS_ATTR_ROOT
) != 0))
241 offset
= (char *)sfe
- (char *)sf
;
242 size
= XFS_ATTR_SF_ENTSIZE_BYNAME(args
->namelen
, args
->valuelen
);
243 xfs_idata_realloc(dp
, size
, XFS_ATTR_FORK
);
244 sf
= (xfs_attr_shortform_t
*)ifp
->if_u1
.if_data
;
245 sfe
= (xfs_attr_sf_entry_t
*)((char *)sf
+ offset
);
247 sfe
->namelen
= args
->namelen
;
248 sfe
->valuelen
= args
->valuelen
;
249 sfe
->flags
= (args
->flags
& ATTR_SECURE
) ? XFS_ATTR_SECURE
:
250 ((args
->flags
& ATTR_ROOT
) ? XFS_ATTR_ROOT
: 0);
251 memcpy(sfe
->nameval
, args
->name
, args
->namelen
);
252 memcpy(&sfe
->nameval
[args
->namelen
], args
->value
, args
->valuelen
);
254 be16_add(&sf
->hdr
.totsize
, size
);
255 xfs_trans_log_inode(args
->trans
, dp
, XFS_ILOG_CORE
| XFS_ILOG_ADATA
);
257 xfs_sbversion_add_attr2(mp
, args
->trans
);
261 * Remove an attribute from the shortform attribute list structure.
264 xfs_attr_shortform_remove(xfs_da_args_t
*args
)
266 xfs_attr_shortform_t
*sf
;
267 xfs_attr_sf_entry_t
*sfe
;
268 int base
, size
=0, end
, totsize
, i
;
274 base
= sizeof(xfs_attr_sf_hdr_t
);
275 sf
= (xfs_attr_shortform_t
*)dp
->i_afp
->if_u1
.if_data
;
278 for (i
= 0; i
< end
; sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
),
280 size
= XFS_ATTR_SF_ENTSIZE(sfe
);
281 if (sfe
->namelen
!= args
->namelen
)
283 if (memcmp(sfe
->nameval
, args
->name
, args
->namelen
) != 0)
285 if (((args
->flags
& ATTR_SECURE
) != 0) !=
286 ((sfe
->flags
& XFS_ATTR_SECURE
) != 0))
288 if (((args
->flags
& ATTR_ROOT
) != 0) !=
289 ((sfe
->flags
& XFS_ATTR_ROOT
) != 0))
294 return(XFS_ERROR(ENOATTR
));
297 * Fix up the attribute fork data, covering the hole
300 totsize
= be16_to_cpu(sf
->hdr
.totsize
);
302 memmove(&((char *)sf
)[base
], &((char *)sf
)[end
], totsize
- end
);
304 be16_add(&sf
->hdr
.totsize
, -size
);
307 * Fix up the start offset of the attribute fork
310 if (totsize
== sizeof(xfs_attr_sf_hdr_t
) && !args
->addname
&&
311 (mp
->m_flags
& XFS_MOUNT_ATTR2
)) {
313 * Last attribute now removed, revert to original
314 * inode format making all literal area available
315 * to the data fork once more.
317 xfs_idestroy_fork(dp
, XFS_ATTR_FORK
);
318 dp
->i_d
.di_forkoff
= 0;
319 dp
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
320 ASSERT(dp
->i_d
.di_anextents
== 0);
321 ASSERT(dp
->i_afp
== NULL
);
322 dp
->i_df
.if_ext_max
=
323 XFS_IFORK_DSIZE(dp
) / (uint
)sizeof(xfs_bmbt_rec_t
);
324 xfs_trans_log_inode(args
->trans
, dp
, XFS_ILOG_CORE
);
326 xfs_idata_realloc(dp
, -size
, XFS_ATTR_FORK
);
327 dp
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(dp
, totsize
);
328 ASSERT(dp
->i_d
.di_forkoff
);
329 ASSERT(totsize
> sizeof(xfs_attr_sf_hdr_t
) || args
->addname
||
330 !(mp
->m_flags
& XFS_MOUNT_ATTR2
));
331 dp
->i_afp
->if_ext_max
=
332 XFS_IFORK_ASIZE(dp
) / (uint
)sizeof(xfs_bmbt_rec_t
);
333 dp
->i_df
.if_ext_max
=
334 XFS_IFORK_DSIZE(dp
) / (uint
)sizeof(xfs_bmbt_rec_t
);
335 xfs_trans_log_inode(args
->trans
, dp
,
336 XFS_ILOG_CORE
| XFS_ILOG_ADATA
);
339 xfs_sbversion_add_attr2(mp
, args
->trans
);
345 * Look up a name in a shortform attribute list structure.
349 xfs_attr_shortform_lookup(xfs_da_args_t
*args
)
351 xfs_attr_shortform_t
*sf
;
352 xfs_attr_sf_entry_t
*sfe
;
356 ifp
= args
->dp
->i_afp
;
357 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
358 sf
= (xfs_attr_shortform_t
*)ifp
->if_u1
.if_data
;
360 for (i
= 0; i
< sf
->hdr
.count
;
361 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
), i
++) {
362 if (sfe
->namelen
!= args
->namelen
)
364 if (memcmp(args
->name
, sfe
->nameval
, args
->namelen
) != 0)
366 if (((args
->flags
& ATTR_SECURE
) != 0) !=
367 ((sfe
->flags
& XFS_ATTR_SECURE
) != 0))
369 if (((args
->flags
& ATTR_ROOT
) != 0) !=
370 ((sfe
->flags
& XFS_ATTR_ROOT
) != 0))
372 return(XFS_ERROR(EEXIST
));
374 return(XFS_ERROR(ENOATTR
));
378 * Look up a name in a shortform attribute list structure.
382 xfs_attr_shortform_getvalue(xfs_da_args_t
*args
)
384 xfs_attr_shortform_t
*sf
;
385 xfs_attr_sf_entry_t
*sfe
;
388 ASSERT(args
->dp
->i_d
.di_aformat
== XFS_IFINLINE
);
389 sf
= (xfs_attr_shortform_t
*)args
->dp
->i_afp
->if_u1
.if_data
;
391 for (i
= 0; i
< sf
->hdr
.count
;
392 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
), i
++) {
393 if (sfe
->namelen
!= args
->namelen
)
395 if (memcmp(args
->name
, sfe
->nameval
, args
->namelen
) != 0)
397 if (((args
->flags
& ATTR_SECURE
) != 0) !=
398 ((sfe
->flags
& XFS_ATTR_SECURE
) != 0))
400 if (((args
->flags
& ATTR_ROOT
) != 0) !=
401 ((sfe
->flags
& XFS_ATTR_ROOT
) != 0))
403 if (args
->flags
& ATTR_KERNOVAL
) {
404 args
->valuelen
= sfe
->valuelen
;
405 return(XFS_ERROR(EEXIST
));
407 if (args
->valuelen
< sfe
->valuelen
) {
408 args
->valuelen
= sfe
->valuelen
;
409 return(XFS_ERROR(ERANGE
));
411 args
->valuelen
= sfe
->valuelen
;
412 memcpy(args
->value
, &sfe
->nameval
[args
->namelen
],
414 return(XFS_ERROR(EEXIST
));
416 return(XFS_ERROR(ENOATTR
));
420 * Convert from using the shortform to the leaf.
423 xfs_attr_shortform_to_leaf(xfs_da_args_t
*args
)
426 xfs_attr_shortform_t
*sf
;
427 xfs_attr_sf_entry_t
*sfe
;
437 sf
= (xfs_attr_shortform_t
*)ifp
->if_u1
.if_data
;
438 size
= be16_to_cpu(sf
->hdr
.totsize
);
439 tmpbuffer
= kmem_alloc(size
, KM_SLEEP
);
440 ASSERT(tmpbuffer
!= NULL
);
441 memcpy(tmpbuffer
, ifp
->if_u1
.if_data
, size
);
442 sf
= (xfs_attr_shortform_t
*)tmpbuffer
;
444 xfs_idata_realloc(dp
, -size
, XFS_ATTR_FORK
);
446 error
= xfs_da_grow_inode(args
, &blkno
);
449 * If we hit an IO error middle of the transaction inside
450 * grow_inode(), we may have inconsistent data. Bail out.
454 xfs_idata_realloc(dp
, size
, XFS_ATTR_FORK
); /* try to put */
455 memcpy(ifp
->if_u1
.if_data
, tmpbuffer
, size
); /* it back */
460 error
= xfs_attr_leaf_create(args
, blkno
, &bp
);
462 error
= xfs_da_shrink_inode(args
, 0, bp
);
466 xfs_idata_realloc(dp
, size
, XFS_ATTR_FORK
); /* try to put */
467 memcpy(ifp
->if_u1
.if_data
, tmpbuffer
, size
); /* it back */
471 memset((char *)&nargs
, 0, sizeof(nargs
));
473 nargs
.firstblock
= args
->firstblock
;
474 nargs
.flist
= args
->flist
;
475 nargs
.total
= args
->total
;
476 nargs
.whichfork
= XFS_ATTR_FORK
;
477 nargs
.trans
= args
->trans
;
481 for (i
= 0; i
< sf
->hdr
.count
; i
++) {
482 nargs
.name
= (char *)sfe
->nameval
;
483 nargs
.namelen
= sfe
->namelen
;
484 nargs
.value
= (char *)&sfe
->nameval
[nargs
.namelen
];
485 nargs
.valuelen
= sfe
->valuelen
;
486 nargs
.hashval
= xfs_da_hashname((char *)sfe
->nameval
,
488 nargs
.flags
= (sfe
->flags
& XFS_ATTR_SECURE
) ? ATTR_SECURE
:
489 ((sfe
->flags
& XFS_ATTR_ROOT
) ? ATTR_ROOT
: 0);
490 error
= xfs_attr_leaf_lookup_int(bp
, &nargs
); /* set a->index */
491 ASSERT(error
== ENOATTR
);
492 error
= xfs_attr_leaf_add(bp
, &nargs
);
493 ASSERT(error
!= ENOSPC
);
496 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
503 kmem_free(tmpbuffer
, size
);
508 xfs_attr_shortform_compare(const void *a
, const void *b
)
510 xfs_attr_sf_sort_t
*sa
, *sb
;
512 sa
= (xfs_attr_sf_sort_t
*)a
;
513 sb
= (xfs_attr_sf_sort_t
*)b
;
514 if (sa
->hash
< sb
->hash
) {
516 } else if (sa
->hash
> sb
->hash
) {
519 return(sa
->entno
- sb
->entno
);
524 * Copy out entries of shortform attribute lists for attr_list().
525 * Shortform attribute lists are not stored in hashval sorted order.
526 * If the output buffer is not large enough to hold them all, then we
527 * we have to calculate each entries' hashvalue and sort them before
528 * we can begin returning them to the user.
532 xfs_attr_shortform_list(xfs_attr_list_context_t
*context
)
534 attrlist_cursor_kern_t
*cursor
;
535 xfs_attr_sf_sort_t
*sbuf
, *sbp
;
536 xfs_attr_shortform_t
*sf
;
537 xfs_attr_sf_entry_t
*sfe
;
539 int sbsize
, nsbuf
, count
, i
;
541 ASSERT(context
!= NULL
);
544 ASSERT(dp
->i_afp
!= NULL
);
545 sf
= (xfs_attr_shortform_t
*)dp
->i_afp
->if_u1
.if_data
;
549 cursor
= context
->cursor
;
550 ASSERT(cursor
!= NULL
);
552 xfs_attr_trace_l_c("sf start", context
);
555 * If the buffer is large enough, do not bother with sorting.
556 * Note the generous fudge factor of 16 overhead bytes per entry.
558 if ((dp
->i_afp
->if_bytes
+ sf
->hdr
.count
* 16) < context
->bufsize
) {
559 for (i
= 0, sfe
= &sf
->list
[0]; i
< sf
->hdr
.count
; i
++) {
562 if (((context
->flags
& ATTR_SECURE
) != 0) !=
563 ((sfe
->flags
& XFS_ATTR_SECURE
) != 0) &&
564 !(context
->flags
& ATTR_KERNORMALS
)) {
565 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
568 if (((context
->flags
& ATTR_ROOT
) != 0) !=
569 ((sfe
->flags
& XFS_ATTR_ROOT
) != 0) &&
570 !(context
->flags
& ATTR_KERNROOTLS
)) {
571 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
574 namesp
= (sfe
->flags
& XFS_ATTR_SECURE
) ? &attr_secure
:
575 ((sfe
->flags
& XFS_ATTR_ROOT
) ? &attr_trusted
:
577 if (context
->flags
& ATTR_KERNOVAL
) {
578 ASSERT(context
->flags
& ATTR_KERNAMELS
);
579 context
->count
+= namesp
->attr_namelen
+
583 if (xfs_attr_put_listent(context
, namesp
,
584 (char *)sfe
->nameval
,
589 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
591 xfs_attr_trace_l_c("sf big-gulp", context
);
596 * It didn't all fit, so we have to sort everything on hashval.
598 sbsize
= sf
->hdr
.count
* sizeof(*sbuf
);
599 sbp
= sbuf
= kmem_alloc(sbsize
, KM_SLEEP
);
602 * Scan the attribute list for the rest of the entries, storing
603 * the relevant info from only those that match into a buffer.
606 for (i
= 0, sfe
= &sf
->list
[0]; i
< sf
->hdr
.count
; i
++) {
608 ((char *)sfe
< (char *)sf
) ||
609 ((char *)sfe
>= ((char *)sf
+ dp
->i_afp
->if_bytes
)))) {
610 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
612 context
->dp
->i_mount
, sfe
);
613 xfs_attr_trace_l_c("sf corrupted", context
);
614 kmem_free(sbuf
, sbsize
);
615 return XFS_ERROR(EFSCORRUPTED
);
617 if (((context
->flags
& ATTR_SECURE
) != 0) !=
618 ((sfe
->flags
& XFS_ATTR_SECURE
) != 0) &&
619 !(context
->flags
& ATTR_KERNORMALS
)) {
620 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
623 if (((context
->flags
& ATTR_ROOT
) != 0) !=
624 ((sfe
->flags
& XFS_ATTR_ROOT
) != 0) &&
625 !(context
->flags
& ATTR_KERNROOTLS
)) {
626 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
630 sbp
->hash
= xfs_da_hashname((char *)sfe
->nameval
, sfe
->namelen
);
631 sbp
->name
= (char *)sfe
->nameval
;
632 sbp
->namelen
= sfe
->namelen
;
633 /* These are bytes, and both on-disk, don't endian-flip */
634 sbp
->valuelen
= sfe
->valuelen
;
635 sbp
->flags
= sfe
->flags
;
636 sfe
= XFS_ATTR_SF_NEXTENTRY(sfe
);
642 * Sort the entries on hash then entno.
644 xfs_sort(sbuf
, nsbuf
, sizeof(*sbuf
), xfs_attr_shortform_compare
);
647 * Re-find our place IN THE SORTED LIST.
652 for (sbp
= sbuf
, i
= 0; i
< nsbuf
; i
++, sbp
++) {
653 if (sbp
->hash
== cursor
->hashval
) {
654 if (cursor
->offset
== count
) {
658 } else if (sbp
->hash
> cursor
->hashval
) {
663 kmem_free(sbuf
, sbsize
);
664 xfs_attr_trace_l_c("blk end", context
);
669 * Loop putting entries into the user buffer.
671 for ( ; i
< nsbuf
; i
++, sbp
++) {
674 namesp
= (sbp
->flags
& XFS_ATTR_SECURE
) ? &attr_secure
:
675 ((sbp
->flags
& XFS_ATTR_ROOT
) ? &attr_trusted
:
678 if (cursor
->hashval
!= sbp
->hash
) {
679 cursor
->hashval
= sbp
->hash
;
682 if (context
->flags
& ATTR_KERNOVAL
) {
683 ASSERT(context
->flags
& ATTR_KERNAMELS
);
684 context
->count
+= namesp
->attr_namelen
+
687 if (xfs_attr_put_listent(context
, namesp
,
688 sbp
->name
, sbp
->namelen
,
695 kmem_free(sbuf
, sbsize
);
696 xfs_attr_trace_l_c("sf E-O-F", context
);
701 * Check a leaf attribute block to see if all the entries would fit into
702 * a shortform attribute list.
705 xfs_attr_shortform_allfit(xfs_dabuf_t
*bp
, xfs_inode_t
*dp
)
707 xfs_attr_leafblock_t
*leaf
;
708 xfs_attr_leaf_entry_t
*entry
;
709 xfs_attr_leaf_name_local_t
*name_loc
;
713 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
715 entry
= &leaf
->entries
[0];
716 bytes
= sizeof(struct xfs_attr_sf_hdr
);
717 for (i
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); entry
++, i
++) {
718 if (entry
->flags
& XFS_ATTR_INCOMPLETE
)
719 continue; /* don't copy partial entries */
720 if (!(entry
->flags
& XFS_ATTR_LOCAL
))
722 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, i
);
723 if (name_loc
->namelen
>= XFS_ATTR_SF_ENTSIZE_MAX
)
725 if (be16_to_cpu(name_loc
->valuelen
) >= XFS_ATTR_SF_ENTSIZE_MAX
)
727 bytes
+= sizeof(struct xfs_attr_sf_entry
)-1
729 + be16_to_cpu(name_loc
->valuelen
);
731 if ((dp
->i_mount
->m_flags
& XFS_MOUNT_ATTR2
) &&
732 (bytes
== sizeof(struct xfs_attr_sf_hdr
)))
734 return(xfs_attr_shortform_bytesfit(dp
, bytes
));
738 * Convert a leaf attribute list to shortform attribute list
741 xfs_attr_leaf_to_shortform(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
, int forkoff
)
743 xfs_attr_leafblock_t
*leaf
;
744 xfs_attr_leaf_entry_t
*entry
;
745 xfs_attr_leaf_name_local_t
*name_loc
;
752 tmpbuffer
= kmem_alloc(XFS_LBSIZE(dp
->i_mount
), KM_SLEEP
);
753 ASSERT(tmpbuffer
!= NULL
);
756 memcpy(tmpbuffer
, bp
->data
, XFS_LBSIZE(dp
->i_mount
));
757 leaf
= (xfs_attr_leafblock_t
*)tmpbuffer
;
758 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
759 memset(bp
->data
, 0, XFS_LBSIZE(dp
->i_mount
));
762 * Clean out the prior contents of the attribute list.
764 error
= xfs_da_shrink_inode(args
, 0, bp
);
769 ASSERT(dp
->i_mount
->m_flags
& XFS_MOUNT_ATTR2
);
772 * Last attribute was removed, revert to original
773 * inode format making all literal area available
774 * to the data fork once more.
776 xfs_idestroy_fork(dp
, XFS_ATTR_FORK
);
777 dp
->i_d
.di_forkoff
= 0;
778 dp
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
779 ASSERT(dp
->i_d
.di_anextents
== 0);
780 ASSERT(dp
->i_afp
== NULL
);
781 dp
->i_df
.if_ext_max
=
782 XFS_IFORK_DSIZE(dp
) / (uint
)sizeof(xfs_bmbt_rec_t
);
783 xfs_trans_log_inode(args
->trans
, dp
, XFS_ILOG_CORE
);
787 xfs_attr_shortform_create(args
);
790 * Copy the attributes
792 memset((char *)&nargs
, 0, sizeof(nargs
));
794 nargs
.firstblock
= args
->firstblock
;
795 nargs
.flist
= args
->flist
;
796 nargs
.total
= args
->total
;
797 nargs
.whichfork
= XFS_ATTR_FORK
;
798 nargs
.trans
= args
->trans
;
800 entry
= &leaf
->entries
[0];
801 for (i
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); entry
++, i
++) {
802 if (entry
->flags
& XFS_ATTR_INCOMPLETE
)
803 continue; /* don't copy partial entries */
806 ASSERT(entry
->flags
& XFS_ATTR_LOCAL
);
807 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, i
);
808 nargs
.name
= (char *)name_loc
->nameval
;
809 nargs
.namelen
= name_loc
->namelen
;
810 nargs
.value
= (char *)&name_loc
->nameval
[nargs
.namelen
];
811 nargs
.valuelen
= be16_to_cpu(name_loc
->valuelen
);
812 nargs
.hashval
= be32_to_cpu(entry
->hashval
);
813 nargs
.flags
= (entry
->flags
& XFS_ATTR_SECURE
) ? ATTR_SECURE
:
814 ((entry
->flags
& XFS_ATTR_ROOT
) ? ATTR_ROOT
: 0);
815 xfs_attr_shortform_add(&nargs
, forkoff
);
820 kmem_free(tmpbuffer
, XFS_LBSIZE(dp
->i_mount
));
825 * Convert from using a single leaf to a root node and a leaf.
828 xfs_attr_leaf_to_node(xfs_da_args_t
*args
)
830 xfs_attr_leafblock_t
*leaf
;
831 xfs_da_intnode_t
*node
;
833 xfs_dabuf_t
*bp1
, *bp2
;
839 error
= xfs_da_grow_inode(args
, &blkno
);
842 error
= xfs_da_read_buf(args
->trans
, args
->dp
, 0, -1, &bp1
,
848 error
= xfs_da_get_buf(args
->trans
, args
->dp
, blkno
, -1, &bp2
,
853 memcpy(bp2
->data
, bp1
->data
, XFS_LBSIZE(dp
->i_mount
));
854 xfs_da_buf_done(bp1
);
856 xfs_da_log_buf(args
->trans
, bp2
, 0, XFS_LBSIZE(dp
->i_mount
) - 1);
859 * Set up the new root node.
861 error
= xfs_da_node_create(args
, 0, 1, &bp1
, XFS_ATTR_FORK
);
866 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
867 /* both on-disk, don't endian-flip twice */
868 node
->btree
[0].hashval
=
869 leaf
->entries
[be16_to_cpu(leaf
->hdr
.count
)-1 ].hashval
;
870 node
->btree
[0].before
= cpu_to_be32(blkno
);
871 node
->hdr
.count
= cpu_to_be16(1);
872 xfs_da_log_buf(args
->trans
, bp1
, 0, XFS_LBSIZE(dp
->i_mount
) - 1);
876 xfs_da_buf_done(bp1
);
878 xfs_da_buf_done(bp2
);
883 /*========================================================================
884 * Routines used for growing the Btree.
885 *========================================================================*/
888 * Create the initial contents of a leaf attribute list
889 * or a leaf in a node attribute list.
892 xfs_attr_leaf_create(xfs_da_args_t
*args
, xfs_dablk_t blkno
, xfs_dabuf_t
**bpp
)
894 xfs_attr_leafblock_t
*leaf
;
895 xfs_attr_leaf_hdr_t
*hdr
;
902 error
= xfs_da_get_buf(args
->trans
, args
->dp
, blkno
, -1, &bp
,
908 memset((char *)leaf
, 0, XFS_LBSIZE(dp
->i_mount
));
910 hdr
->info
.magic
= cpu_to_be16(XFS_ATTR_LEAF_MAGIC
);
911 hdr
->firstused
= cpu_to_be16(XFS_LBSIZE(dp
->i_mount
));
912 if (!hdr
->firstused
) {
913 hdr
->firstused
= cpu_to_be16(
914 XFS_LBSIZE(dp
->i_mount
) - XFS_ATTR_LEAF_NAME_ALIGN
);
917 hdr
->freemap
[0].base
= cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t
));
918 hdr
->freemap
[0].size
= cpu_to_be16(be16_to_cpu(hdr
->firstused
) -
919 sizeof(xfs_attr_leaf_hdr_t
));
921 xfs_da_log_buf(args
->trans
, bp
, 0, XFS_LBSIZE(dp
->i_mount
) - 1);
928 * Split the leaf node, rebalance, then add the new entry.
931 xfs_attr_leaf_split(xfs_da_state_t
*state
, xfs_da_state_blk_t
*oldblk
,
932 xfs_da_state_blk_t
*newblk
)
938 * Allocate space for a new leaf node.
940 ASSERT(oldblk
->magic
== XFS_ATTR_LEAF_MAGIC
);
941 error
= xfs_da_grow_inode(state
->args
, &blkno
);
944 error
= xfs_attr_leaf_create(state
->args
, blkno
, &newblk
->bp
);
947 newblk
->blkno
= blkno
;
948 newblk
->magic
= XFS_ATTR_LEAF_MAGIC
;
951 * Rebalance the entries across the two leaves.
952 * NOTE: rebalance() currently depends on the 2nd block being empty.
954 xfs_attr_leaf_rebalance(state
, oldblk
, newblk
);
955 error
= xfs_da_blk_link(state
, oldblk
, newblk
);
960 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
961 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
962 * "new" attrs info. Will need the "old" info to remove it later.
964 * Insert the "new" entry in the correct block.
967 error
= xfs_attr_leaf_add(oldblk
->bp
, state
->args
);
969 error
= xfs_attr_leaf_add(newblk
->bp
, state
->args
);
972 * Update last hashval in each block since we added the name.
974 oldblk
->hashval
= xfs_attr_leaf_lasthash(oldblk
->bp
, NULL
);
975 newblk
->hashval
= xfs_attr_leaf_lasthash(newblk
->bp
, NULL
);
980 * Add a name to the leaf attribute list structure.
983 xfs_attr_leaf_add(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
)
985 xfs_attr_leafblock_t
*leaf
;
986 xfs_attr_leaf_hdr_t
*hdr
;
987 xfs_attr_leaf_map_t
*map
;
988 int tablesize
, entsize
, sum
, tmp
, i
;
991 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
992 ASSERT((args
->index
>= 0)
993 && (args
->index
<= be16_to_cpu(leaf
->hdr
.count
)));
995 entsize
= xfs_attr_leaf_newentsize(args
->namelen
, args
->valuelen
,
996 args
->trans
->t_mountp
->m_sb
.sb_blocksize
, NULL
);
999 * Search through freemap for first-fit on new name length.
1000 * (may need to figure in size of entry struct too)
1002 tablesize
= (be16_to_cpu(hdr
->count
) + 1)
1003 * sizeof(xfs_attr_leaf_entry_t
)
1004 + sizeof(xfs_attr_leaf_hdr_t
);
1005 map
= &hdr
->freemap
[XFS_ATTR_LEAF_MAPSIZE
-1];
1006 for (sum
= 0, i
= XFS_ATTR_LEAF_MAPSIZE
-1; i
>= 0; map
--, i
--) {
1007 if (tablesize
> be16_to_cpu(hdr
->firstused
)) {
1008 sum
+= be16_to_cpu(map
->size
);
1012 continue; /* no space in this map */
1014 if (be16_to_cpu(map
->base
) < be16_to_cpu(hdr
->firstused
))
1015 tmp
+= sizeof(xfs_attr_leaf_entry_t
);
1016 if (be16_to_cpu(map
->size
) >= tmp
) {
1017 tmp
= xfs_attr_leaf_add_work(bp
, args
, i
);
1020 sum
+= be16_to_cpu(map
->size
);
1024 * If there are no holes in the address space of the block,
1025 * and we don't have enough freespace, then compaction will do us
1026 * no good and we should just give up.
1028 if (!hdr
->holes
&& (sum
< entsize
))
1029 return(XFS_ERROR(ENOSPC
));
1032 * Compact the entries to coalesce free space.
1033 * This may change the hdr->count via dropping INCOMPLETE entries.
1035 xfs_attr_leaf_compact(args
->trans
, bp
);
1038 * After compaction, the block is guaranteed to have only one
1039 * free region, in freemap[0]. If it is not big enough, give up.
1041 if (be16_to_cpu(hdr
->freemap
[0].size
)
1042 < (entsize
+ sizeof(xfs_attr_leaf_entry_t
)))
1043 return(XFS_ERROR(ENOSPC
));
1045 return(xfs_attr_leaf_add_work(bp
, args
, 0));
1049 * Add a name to a leaf attribute list structure.
1052 xfs_attr_leaf_add_work(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
, int mapindex
)
1054 xfs_attr_leafblock_t
*leaf
;
1055 xfs_attr_leaf_hdr_t
*hdr
;
1056 xfs_attr_leaf_entry_t
*entry
;
1057 xfs_attr_leaf_name_local_t
*name_loc
;
1058 xfs_attr_leaf_name_remote_t
*name_rmt
;
1059 xfs_attr_leaf_map_t
*map
;
1064 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1066 ASSERT((mapindex
>= 0) && (mapindex
< XFS_ATTR_LEAF_MAPSIZE
));
1067 ASSERT((args
->index
>= 0) && (args
->index
<= be16_to_cpu(hdr
->count
)));
1070 * Force open some space in the entry array and fill it in.
1072 entry
= &leaf
->entries
[args
->index
];
1073 if (args
->index
< be16_to_cpu(hdr
->count
)) {
1074 tmp
= be16_to_cpu(hdr
->count
) - args
->index
;
1075 tmp
*= sizeof(xfs_attr_leaf_entry_t
);
1076 memmove((char *)(entry
+1), (char *)entry
, tmp
);
1077 xfs_da_log_buf(args
->trans
, bp
,
1078 XFS_DA_LOGRANGE(leaf
, entry
, tmp
+ sizeof(*entry
)));
1080 be16_add(&hdr
->count
, 1);
1083 * Allocate space for the new string (at the end of the run).
1085 map
= &hdr
->freemap
[mapindex
];
1086 mp
= args
->trans
->t_mountp
;
1087 ASSERT(be16_to_cpu(map
->base
) < XFS_LBSIZE(mp
));
1088 ASSERT((be16_to_cpu(map
->base
) & 0x3) == 0);
1089 ASSERT(be16_to_cpu(map
->size
) >=
1090 xfs_attr_leaf_newentsize(args
->namelen
, args
->valuelen
,
1091 mp
->m_sb
.sb_blocksize
, NULL
));
1092 ASSERT(be16_to_cpu(map
->size
) < XFS_LBSIZE(mp
));
1093 ASSERT((be16_to_cpu(map
->size
) & 0x3) == 0);
1094 be16_add(&map
->size
,
1095 -xfs_attr_leaf_newentsize(args
->namelen
, args
->valuelen
,
1096 mp
->m_sb
.sb_blocksize
, &tmp
));
1097 entry
->nameidx
= cpu_to_be16(be16_to_cpu(map
->base
) +
1098 be16_to_cpu(map
->size
));
1099 entry
->hashval
= cpu_to_be32(args
->hashval
);
1100 entry
->flags
= tmp
? XFS_ATTR_LOCAL
: 0;
1101 entry
->flags
|= (args
->flags
& ATTR_SECURE
) ? XFS_ATTR_SECURE
:
1102 ((args
->flags
& ATTR_ROOT
) ? XFS_ATTR_ROOT
: 0);
1104 entry
->flags
|= XFS_ATTR_INCOMPLETE
;
1105 if ((args
->blkno2
== args
->blkno
) &&
1106 (args
->index2
<= args
->index
)) {
1110 xfs_da_log_buf(args
->trans
, bp
,
1111 XFS_DA_LOGRANGE(leaf
, entry
, sizeof(*entry
)));
1112 ASSERT((args
->index
== 0) ||
1113 (be32_to_cpu(entry
->hashval
) >= be32_to_cpu((entry
-1)->hashval
)));
1114 ASSERT((args
->index
== be16_to_cpu(hdr
->count
)-1) ||
1115 (be32_to_cpu(entry
->hashval
) <= be32_to_cpu((entry
+1)->hashval
)));
1118 * Copy the attribute name and value into the new space.
1120 * For "remote" attribute values, simply note that we need to
1121 * allocate space for the "remote" value. We can't actually
1122 * allocate the extents in this transaction, and we can't decide
1123 * which blocks they should be as we might allocate more blocks
1124 * as part of this transaction (a split operation for example).
1126 if (entry
->flags
& XFS_ATTR_LOCAL
) {
1127 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, args
->index
);
1128 name_loc
->namelen
= args
->namelen
;
1129 name_loc
->valuelen
= cpu_to_be16(args
->valuelen
);
1130 memcpy((char *)name_loc
->nameval
, args
->name
, args
->namelen
);
1131 memcpy((char *)&name_loc
->nameval
[args
->namelen
], args
->value
,
1132 be16_to_cpu(name_loc
->valuelen
));
1134 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, args
->index
);
1135 name_rmt
->namelen
= args
->namelen
;
1136 memcpy((char *)name_rmt
->name
, args
->name
, args
->namelen
);
1137 entry
->flags
|= XFS_ATTR_INCOMPLETE
;
1139 name_rmt
->valuelen
= 0;
1140 name_rmt
->valueblk
= 0;
1142 args
->rmtblkcnt
= XFS_B_TO_FSB(mp
, args
->valuelen
);
1144 xfs_da_log_buf(args
->trans
, bp
,
1145 XFS_DA_LOGRANGE(leaf
, XFS_ATTR_LEAF_NAME(leaf
, args
->index
),
1146 xfs_attr_leaf_entsize(leaf
, args
->index
)));
1149 * Update the control info for this leaf node
1151 if (be16_to_cpu(entry
->nameidx
) < be16_to_cpu(hdr
->firstused
)) {
1152 /* both on-disk, don't endian-flip twice */
1153 hdr
->firstused
= entry
->nameidx
;
1155 ASSERT(be16_to_cpu(hdr
->firstused
) >=
1156 ((be16_to_cpu(hdr
->count
) * sizeof(*entry
)) + sizeof(*hdr
)));
1157 tmp
= (be16_to_cpu(hdr
->count
)-1) * sizeof(xfs_attr_leaf_entry_t
)
1158 + sizeof(xfs_attr_leaf_hdr_t
);
1159 map
= &hdr
->freemap
[0];
1160 for (i
= 0; i
< XFS_ATTR_LEAF_MAPSIZE
; map
++, i
++) {
1161 if (be16_to_cpu(map
->base
) == tmp
) {
1162 be16_add(&map
->base
, sizeof(xfs_attr_leaf_entry_t
));
1163 be16_add(&map
->size
,
1164 -((int)sizeof(xfs_attr_leaf_entry_t
)));
1167 be16_add(&hdr
->usedbytes
, xfs_attr_leaf_entsize(leaf
, args
->index
));
1168 xfs_da_log_buf(args
->trans
, bp
,
1169 XFS_DA_LOGRANGE(leaf
, hdr
, sizeof(*hdr
)));
1174 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1177 xfs_attr_leaf_compact(xfs_trans_t
*trans
, xfs_dabuf_t
*bp
)
1179 xfs_attr_leafblock_t
*leaf_s
, *leaf_d
;
1180 xfs_attr_leaf_hdr_t
*hdr_s
, *hdr_d
;
1184 mp
= trans
->t_mountp
;
1185 tmpbuffer
= kmem_alloc(XFS_LBSIZE(mp
), KM_SLEEP
);
1186 ASSERT(tmpbuffer
!= NULL
);
1187 memcpy(tmpbuffer
, bp
->data
, XFS_LBSIZE(mp
));
1188 memset(bp
->data
, 0, XFS_LBSIZE(mp
));
1191 * Copy basic information
1193 leaf_s
= (xfs_attr_leafblock_t
*)tmpbuffer
;
1195 hdr_s
= &leaf_s
->hdr
;
1196 hdr_d
= &leaf_d
->hdr
;
1197 hdr_d
->info
= hdr_s
->info
; /* struct copy */
1198 hdr_d
->firstused
= cpu_to_be16(XFS_LBSIZE(mp
));
1199 /* handle truncation gracefully */
1200 if (!hdr_d
->firstused
) {
1201 hdr_d
->firstused
= cpu_to_be16(
1202 XFS_LBSIZE(mp
) - XFS_ATTR_LEAF_NAME_ALIGN
);
1204 hdr_d
->usedbytes
= 0;
1207 hdr_d
->freemap
[0].base
= cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t
));
1208 hdr_d
->freemap
[0].size
= cpu_to_be16(be16_to_cpu(hdr_d
->firstused
) -
1209 sizeof(xfs_attr_leaf_hdr_t
));
1212 * Copy all entry's in the same (sorted) order,
1213 * but allocate name/value pairs packed and in sequence.
1215 xfs_attr_leaf_moveents(leaf_s
, 0, leaf_d
, 0,
1216 be16_to_cpu(hdr_s
->count
), mp
);
1217 xfs_da_log_buf(trans
, bp
, 0, XFS_LBSIZE(mp
) - 1);
1219 kmem_free(tmpbuffer
, XFS_LBSIZE(mp
));
1223 * Redistribute the attribute list entries between two leaf nodes,
1224 * taking into account the size of the new entry.
1226 * NOTE: if new block is empty, then it will get the upper half of the
1227 * old block. At present, all (one) callers pass in an empty second block.
1229 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1230 * to match what it is doing in splitting the attribute leaf block. Those
1231 * values are used in "atomic rename" operations on attributes. Note that
1232 * the "new" and "old" values can end up in different blocks.
1235 xfs_attr_leaf_rebalance(xfs_da_state_t
*state
, xfs_da_state_blk_t
*blk1
,
1236 xfs_da_state_blk_t
*blk2
)
1238 xfs_da_args_t
*args
;
1239 xfs_da_state_blk_t
*tmp_blk
;
1240 xfs_attr_leafblock_t
*leaf1
, *leaf2
;
1241 xfs_attr_leaf_hdr_t
*hdr1
, *hdr2
;
1242 int count
, totallen
, max
, space
, swap
;
1245 * Set up environment.
1247 ASSERT(blk1
->magic
== XFS_ATTR_LEAF_MAGIC
);
1248 ASSERT(blk2
->magic
== XFS_ATTR_LEAF_MAGIC
);
1249 leaf1
= blk1
->bp
->data
;
1250 leaf2
= blk2
->bp
->data
;
1251 ASSERT(be16_to_cpu(leaf1
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1252 ASSERT(be16_to_cpu(leaf2
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1256 * Check ordering of blocks, reverse if it makes things simpler.
1258 * NOTE: Given that all (current) callers pass in an empty
1259 * second block, this code should never set "swap".
1262 if (xfs_attr_leaf_order(blk1
->bp
, blk2
->bp
)) {
1266 leaf1
= blk1
->bp
->data
;
1267 leaf2
= blk2
->bp
->data
;
1274 * Examine entries until we reduce the absolute difference in
1275 * byte usage between the two blocks to a minimum. Then get
1276 * the direction to copy and the number of elements to move.
1278 * "inleaf" is true if the new entry should be inserted into blk1.
1279 * If "swap" is also true, then reverse the sense of "inleaf".
1281 state
->inleaf
= xfs_attr_leaf_figure_balance(state
, blk1
, blk2
,
1284 state
->inleaf
= !state
->inleaf
;
1287 * Move any entries required from leaf to leaf:
1289 if (count
< be16_to_cpu(hdr1
->count
)) {
1291 * Figure the total bytes to be added to the destination leaf.
1293 /* number entries being moved */
1294 count
= be16_to_cpu(hdr1
->count
) - count
;
1295 space
= be16_to_cpu(hdr1
->usedbytes
) - totallen
;
1296 space
+= count
* sizeof(xfs_attr_leaf_entry_t
);
1299 * leaf2 is the destination, compact it if it looks tight.
1301 max
= be16_to_cpu(hdr2
->firstused
)
1302 - sizeof(xfs_attr_leaf_hdr_t
);
1303 max
-= be16_to_cpu(hdr2
->count
) * sizeof(xfs_attr_leaf_entry_t
);
1305 xfs_attr_leaf_compact(args
->trans
, blk2
->bp
);
1309 * Move high entries from leaf1 to low end of leaf2.
1311 xfs_attr_leaf_moveents(leaf1
, be16_to_cpu(hdr1
->count
) - count
,
1312 leaf2
, 0, count
, state
->mp
);
1314 xfs_da_log_buf(args
->trans
, blk1
->bp
, 0, state
->blocksize
-1);
1315 xfs_da_log_buf(args
->trans
, blk2
->bp
, 0, state
->blocksize
-1);
1316 } else if (count
> be16_to_cpu(hdr1
->count
)) {
1318 * I assert that since all callers pass in an empty
1319 * second buffer, this code should never execute.
1323 * Figure the total bytes to be added to the destination leaf.
1325 /* number entries being moved */
1326 count
-= be16_to_cpu(hdr1
->count
);
1327 space
= totallen
- be16_to_cpu(hdr1
->usedbytes
);
1328 space
+= count
* sizeof(xfs_attr_leaf_entry_t
);
1331 * leaf1 is the destination, compact it if it looks tight.
1333 max
= be16_to_cpu(hdr1
->firstused
)
1334 - sizeof(xfs_attr_leaf_hdr_t
);
1335 max
-= be16_to_cpu(hdr1
->count
) * sizeof(xfs_attr_leaf_entry_t
);
1337 xfs_attr_leaf_compact(args
->trans
, blk1
->bp
);
1341 * Move low entries from leaf2 to high end of leaf1.
1343 xfs_attr_leaf_moveents(leaf2
, 0, leaf1
,
1344 be16_to_cpu(hdr1
->count
), count
, state
->mp
);
1346 xfs_da_log_buf(args
->trans
, blk1
->bp
, 0, state
->blocksize
-1);
1347 xfs_da_log_buf(args
->trans
, blk2
->bp
, 0, state
->blocksize
-1);
1351 * Copy out last hashval in each block for B-tree code.
1353 blk1
->hashval
= be32_to_cpu(
1354 leaf1
->entries
[be16_to_cpu(leaf1
->hdr
.count
)-1].hashval
);
1355 blk2
->hashval
= be32_to_cpu(
1356 leaf2
->entries
[be16_to_cpu(leaf2
->hdr
.count
)-1].hashval
);
1359 * Adjust the expected index for insertion.
1360 * NOTE: this code depends on the (current) situation that the
1361 * second block was originally empty.
1363 * If the insertion point moved to the 2nd block, we must adjust
1364 * the index. We must also track the entry just following the
1365 * new entry for use in an "atomic rename" operation, that entry
1366 * is always the "old" entry and the "new" entry is what we are
1367 * inserting. The index/blkno fields refer to the "old" entry,
1368 * while the index2/blkno2 fields refer to the "new" entry.
1370 if (blk1
->index
> be16_to_cpu(leaf1
->hdr
.count
)) {
1371 ASSERT(state
->inleaf
== 0);
1372 blk2
->index
= blk1
->index
- be16_to_cpu(leaf1
->hdr
.count
);
1373 args
->index
= args
->index2
= blk2
->index
;
1374 args
->blkno
= args
->blkno2
= blk2
->blkno
;
1375 } else if (blk1
->index
== be16_to_cpu(leaf1
->hdr
.count
)) {
1376 if (state
->inleaf
) {
1377 args
->index
= blk1
->index
;
1378 args
->blkno
= blk1
->blkno
;
1380 args
->blkno2
= blk2
->blkno
;
1382 blk2
->index
= blk1
->index
1383 - be16_to_cpu(leaf1
->hdr
.count
);
1384 args
->index
= args
->index2
= blk2
->index
;
1385 args
->blkno
= args
->blkno2
= blk2
->blkno
;
1388 ASSERT(state
->inleaf
== 1);
1389 args
->index
= args
->index2
= blk1
->index
;
1390 args
->blkno
= args
->blkno2
= blk1
->blkno
;
1395 * Examine entries until we reduce the absolute difference in
1396 * byte usage between the two blocks to a minimum.
1397 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1398 * GROT: there will always be enough room in either block for a new entry.
1399 * GROT: Do a double-split for this case?
1402 xfs_attr_leaf_figure_balance(xfs_da_state_t
*state
,
1403 xfs_da_state_blk_t
*blk1
,
1404 xfs_da_state_blk_t
*blk2
,
1405 int *countarg
, int *usedbytesarg
)
1407 xfs_attr_leafblock_t
*leaf1
, *leaf2
;
1408 xfs_attr_leaf_hdr_t
*hdr1
, *hdr2
;
1409 xfs_attr_leaf_entry_t
*entry
;
1410 int count
, max
, index
, totallen
, half
;
1411 int lastdelta
, foundit
, tmp
;
1414 * Set up environment.
1416 leaf1
= blk1
->bp
->data
;
1417 leaf2
= blk2
->bp
->data
;
1424 * Examine entries until we reduce the absolute difference in
1425 * byte usage between the two blocks to a minimum.
1427 max
= be16_to_cpu(hdr1
->count
) + be16_to_cpu(hdr2
->count
);
1428 half
= (max
+1) * sizeof(*entry
);
1429 half
+= be16_to_cpu(hdr1
->usedbytes
) +
1430 be16_to_cpu(hdr2
->usedbytes
) +
1431 xfs_attr_leaf_newentsize(
1432 state
->args
->namelen
,
1433 state
->args
->valuelen
,
1434 state
->blocksize
, NULL
);
1436 lastdelta
= state
->blocksize
;
1437 entry
= &leaf1
->entries
[0];
1438 for (count
= index
= 0; count
< max
; entry
++, index
++, count
++) {
1440 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1442 * The new entry is in the first block, account for it.
1444 if (count
== blk1
->index
) {
1445 tmp
= totallen
+ sizeof(*entry
) +
1446 xfs_attr_leaf_newentsize(
1447 state
->args
->namelen
,
1448 state
->args
->valuelen
,
1449 state
->blocksize
, NULL
);
1450 if (XFS_ATTR_ABS(half
- tmp
) > lastdelta
)
1452 lastdelta
= XFS_ATTR_ABS(half
- tmp
);
1458 * Wrap around into the second block if necessary.
1460 if (count
== be16_to_cpu(hdr1
->count
)) {
1462 entry
= &leaf1
->entries
[0];
1467 * Figure out if next leaf entry would be too much.
1469 tmp
= totallen
+ sizeof(*entry
) + xfs_attr_leaf_entsize(leaf1
,
1471 if (XFS_ATTR_ABS(half
- tmp
) > lastdelta
)
1473 lastdelta
= XFS_ATTR_ABS(half
- tmp
);
1479 * Calculate the number of usedbytes that will end up in lower block.
1480 * If new entry not in lower block, fix up the count.
1482 totallen
-= count
* sizeof(*entry
);
1484 totallen
-= sizeof(*entry
) +
1485 xfs_attr_leaf_newentsize(
1486 state
->args
->namelen
,
1487 state
->args
->valuelen
,
1488 state
->blocksize
, NULL
);
1492 *usedbytesarg
= totallen
;
1496 /*========================================================================
1497 * Routines used for shrinking the Btree.
1498 *========================================================================*/
1501 * Check a leaf block and its neighbors to see if the block should be
1502 * collapsed into one or the other neighbor. Always keep the block
1503 * with the smaller block number.
1504 * If the current block is over 50% full, don't try to join it, return 0.
1505 * If the block is empty, fill in the state structure and return 2.
1506 * If it can be collapsed, fill in the state structure and return 1.
1507 * If nothing can be done, return 0.
1509 * GROT: allow for INCOMPLETE entries in calculation.
1512 xfs_attr_leaf_toosmall(xfs_da_state_t
*state
, int *action
)
1514 xfs_attr_leafblock_t
*leaf
;
1515 xfs_da_state_blk_t
*blk
;
1516 xfs_da_blkinfo_t
*info
;
1517 int count
, bytes
, forward
, error
, retval
, i
;
1522 * Check for the degenerate case of the block being over 50% full.
1523 * If so, it's not worth even looking to see if we might be able
1524 * to coalesce with a sibling.
1526 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1527 info
= blk
->bp
->data
;
1528 ASSERT(be16_to_cpu(info
->magic
) == XFS_ATTR_LEAF_MAGIC
);
1529 leaf
= (xfs_attr_leafblock_t
*)info
;
1530 count
= be16_to_cpu(leaf
->hdr
.count
);
1531 bytes
= sizeof(xfs_attr_leaf_hdr_t
) +
1532 count
* sizeof(xfs_attr_leaf_entry_t
) +
1533 be16_to_cpu(leaf
->hdr
.usedbytes
);
1534 if (bytes
> (state
->blocksize
>> 1)) {
1535 *action
= 0; /* blk over 50%, don't try to join */
1540 * Check for the degenerate case of the block being empty.
1541 * If the block is empty, we'll simply delete it, no need to
1542 * coalesce it with a sibling block. We choose (arbitrarily)
1543 * to merge with the forward block unless it is NULL.
1547 * Make altpath point to the block we want to keep and
1548 * path point to the block we want to drop (this one).
1550 forward
= (info
->forw
!= 0);
1551 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1552 error
= xfs_da_path_shift(state
, &state
->altpath
, forward
,
1565 * Examine each sibling block to see if we can coalesce with
1566 * at least 25% free space to spare. We need to figure out
1567 * whether to merge with the forward or the backward block.
1568 * We prefer coalescing with the lower numbered sibling so as
1569 * to shrink an attribute list over time.
1571 /* start with smaller blk num */
1572 forward
= (be32_to_cpu(info
->forw
) < be32_to_cpu(info
->back
));
1573 for (i
= 0; i
< 2; forward
= !forward
, i
++) {
1575 blkno
= be32_to_cpu(info
->forw
);
1577 blkno
= be32_to_cpu(info
->back
);
1580 error
= xfs_da_read_buf(state
->args
->trans
, state
->args
->dp
,
1581 blkno
, -1, &bp
, XFS_ATTR_FORK
);
1586 leaf
= (xfs_attr_leafblock_t
*)info
;
1587 count
= be16_to_cpu(leaf
->hdr
.count
);
1588 bytes
= state
->blocksize
- (state
->blocksize
>>2);
1589 bytes
-= be16_to_cpu(leaf
->hdr
.usedbytes
);
1591 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1592 count
+= be16_to_cpu(leaf
->hdr
.count
);
1593 bytes
-= be16_to_cpu(leaf
->hdr
.usedbytes
);
1594 bytes
-= count
* sizeof(xfs_attr_leaf_entry_t
);
1595 bytes
-= sizeof(xfs_attr_leaf_hdr_t
);
1596 xfs_da_brelse(state
->args
->trans
, bp
);
1598 break; /* fits with at least 25% to spare */
1606 * Make altpath point to the block we want to keep (the lower
1607 * numbered block) and path point to the block we want to drop.
1609 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1610 if (blkno
< blk
->blkno
) {
1611 error
= xfs_da_path_shift(state
, &state
->altpath
, forward
,
1614 error
= xfs_da_path_shift(state
, &state
->path
, forward
,
1628 * Remove a name from the leaf attribute list structure.
1630 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1631 * If two leaves are 37% full, when combined they will leave 25% free.
1634 xfs_attr_leaf_remove(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
)
1636 xfs_attr_leafblock_t
*leaf
;
1637 xfs_attr_leaf_hdr_t
*hdr
;
1638 xfs_attr_leaf_map_t
*map
;
1639 xfs_attr_leaf_entry_t
*entry
;
1640 int before
, after
, smallest
, entsize
;
1641 int tablesize
, tmp
, i
;
1645 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1647 mp
= args
->trans
->t_mountp
;
1648 ASSERT((be16_to_cpu(hdr
->count
) > 0)
1649 && (be16_to_cpu(hdr
->count
) < (XFS_LBSIZE(mp
)/8)));
1650 ASSERT((args
->index
>= 0)
1651 && (args
->index
< be16_to_cpu(hdr
->count
)));
1652 ASSERT(be16_to_cpu(hdr
->firstused
) >=
1653 ((be16_to_cpu(hdr
->count
) * sizeof(*entry
)) + sizeof(*hdr
)));
1654 entry
= &leaf
->entries
[args
->index
];
1655 ASSERT(be16_to_cpu(entry
->nameidx
) >= be16_to_cpu(hdr
->firstused
));
1656 ASSERT(be16_to_cpu(entry
->nameidx
) < XFS_LBSIZE(mp
));
1659 * Scan through free region table:
1660 * check for adjacency of free'd entry with an existing one,
1661 * find smallest free region in case we need to replace it,
1662 * adjust any map that borders the entry table,
1664 tablesize
= be16_to_cpu(hdr
->count
) * sizeof(xfs_attr_leaf_entry_t
)
1665 + sizeof(xfs_attr_leaf_hdr_t
);
1666 map
= &hdr
->freemap
[0];
1667 tmp
= be16_to_cpu(map
->size
);
1668 before
= after
= -1;
1669 smallest
= XFS_ATTR_LEAF_MAPSIZE
- 1;
1670 entsize
= xfs_attr_leaf_entsize(leaf
, args
->index
);
1671 for (i
= 0; i
< XFS_ATTR_LEAF_MAPSIZE
; map
++, i
++) {
1672 ASSERT(be16_to_cpu(map
->base
) < XFS_LBSIZE(mp
));
1673 ASSERT(be16_to_cpu(map
->size
) < XFS_LBSIZE(mp
));
1674 if (be16_to_cpu(map
->base
) == tablesize
) {
1675 be16_add(&map
->base
,
1676 -((int)sizeof(xfs_attr_leaf_entry_t
)));
1677 be16_add(&map
->size
, sizeof(xfs_attr_leaf_entry_t
));
1680 if ((be16_to_cpu(map
->base
) + be16_to_cpu(map
->size
))
1681 == be16_to_cpu(entry
->nameidx
)) {
1683 } else if (be16_to_cpu(map
->base
)
1684 == (be16_to_cpu(entry
->nameidx
) + entsize
)) {
1686 } else if (be16_to_cpu(map
->size
) < tmp
) {
1687 tmp
= be16_to_cpu(map
->size
);
1693 * Coalesce adjacent freemap regions,
1694 * or replace the smallest region.
1696 if ((before
>= 0) || (after
>= 0)) {
1697 if ((before
>= 0) && (after
>= 0)) {
1698 map
= &hdr
->freemap
[before
];
1699 be16_add(&map
->size
, entsize
);
1700 be16_add(&map
->size
,
1701 be16_to_cpu(hdr
->freemap
[after
].size
));
1702 hdr
->freemap
[after
].base
= 0;
1703 hdr
->freemap
[after
].size
= 0;
1704 } else if (before
>= 0) {
1705 map
= &hdr
->freemap
[before
];
1706 be16_add(&map
->size
, entsize
);
1708 map
= &hdr
->freemap
[after
];
1709 /* both on-disk, don't endian flip twice */
1710 map
->base
= entry
->nameidx
;
1711 be16_add(&map
->size
, entsize
);
1715 * Replace smallest region (if it is smaller than free'd entry)
1717 map
= &hdr
->freemap
[smallest
];
1718 if (be16_to_cpu(map
->size
) < entsize
) {
1719 map
->base
= cpu_to_be16(be16_to_cpu(entry
->nameidx
));
1720 map
->size
= cpu_to_be16(entsize
);
1725 * Did we remove the first entry?
1727 if (be16_to_cpu(entry
->nameidx
) == be16_to_cpu(hdr
->firstused
))
1733 * Compress the remaining entries and zero out the removed stuff.
1735 memset(XFS_ATTR_LEAF_NAME(leaf
, args
->index
), 0, entsize
);
1736 be16_add(&hdr
->usedbytes
, -entsize
);
1737 xfs_da_log_buf(args
->trans
, bp
,
1738 XFS_DA_LOGRANGE(leaf
, XFS_ATTR_LEAF_NAME(leaf
, args
->index
),
1741 tmp
= (be16_to_cpu(hdr
->count
) - args
->index
)
1742 * sizeof(xfs_attr_leaf_entry_t
);
1743 memmove((char *)entry
, (char *)(entry
+1), tmp
);
1744 be16_add(&hdr
->count
, -1);
1745 xfs_da_log_buf(args
->trans
, bp
,
1746 XFS_DA_LOGRANGE(leaf
, entry
, tmp
+ sizeof(*entry
)));
1747 entry
= &leaf
->entries
[be16_to_cpu(hdr
->count
)];
1748 memset((char *)entry
, 0, sizeof(xfs_attr_leaf_entry_t
));
1751 * If we removed the first entry, re-find the first used byte
1752 * in the name area. Note that if the entry was the "firstused",
1753 * then we don't have a "hole" in our block resulting from
1754 * removing the name.
1757 tmp
= XFS_LBSIZE(mp
);
1758 entry
= &leaf
->entries
[0];
1759 for (i
= be16_to_cpu(hdr
->count
)-1; i
>= 0; entry
++, i
--) {
1760 ASSERT(be16_to_cpu(entry
->nameidx
) >=
1761 be16_to_cpu(hdr
->firstused
));
1762 ASSERT(be16_to_cpu(entry
->nameidx
) < XFS_LBSIZE(mp
));
1764 if (be16_to_cpu(entry
->nameidx
) < tmp
)
1765 tmp
= be16_to_cpu(entry
->nameidx
);
1767 hdr
->firstused
= cpu_to_be16(tmp
);
1768 if (!hdr
->firstused
) {
1769 hdr
->firstused
= cpu_to_be16(
1770 tmp
- XFS_ATTR_LEAF_NAME_ALIGN
);
1773 hdr
->holes
= 1; /* mark as needing compaction */
1775 xfs_da_log_buf(args
->trans
, bp
,
1776 XFS_DA_LOGRANGE(leaf
, hdr
, sizeof(*hdr
)));
1779 * Check if leaf is less than 50% full, caller may want to
1780 * "join" the leaf with a sibling if so.
1782 tmp
= sizeof(xfs_attr_leaf_hdr_t
);
1783 tmp
+= be16_to_cpu(leaf
->hdr
.count
) * sizeof(xfs_attr_leaf_entry_t
);
1784 tmp
+= be16_to_cpu(leaf
->hdr
.usedbytes
);
1785 return(tmp
< mp
->m_attr_magicpct
); /* leaf is < 37% full */
1789 * Move all the attribute list entries from drop_leaf into save_leaf.
1792 xfs_attr_leaf_unbalance(xfs_da_state_t
*state
, xfs_da_state_blk_t
*drop_blk
,
1793 xfs_da_state_blk_t
*save_blk
)
1795 xfs_attr_leafblock_t
*drop_leaf
, *save_leaf
, *tmp_leaf
;
1796 xfs_attr_leaf_hdr_t
*drop_hdr
, *save_hdr
, *tmp_hdr
;
1801 * Set up environment.
1804 ASSERT(drop_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1805 ASSERT(save_blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1806 drop_leaf
= drop_blk
->bp
->data
;
1807 save_leaf
= save_blk
->bp
->data
;
1808 ASSERT(be16_to_cpu(drop_leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1809 ASSERT(be16_to_cpu(save_leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1810 drop_hdr
= &drop_leaf
->hdr
;
1811 save_hdr
= &save_leaf
->hdr
;
1814 * Save last hashval from dying block for later Btree fixup.
1816 drop_blk
->hashval
= be32_to_cpu(
1817 drop_leaf
->entries
[be16_to_cpu(drop_leaf
->hdr
.count
)-1].hashval
);
1820 * Check if we need a temp buffer, or can we do it in place.
1821 * Note that we don't check "leaf" for holes because we will
1822 * always be dropping it, toosmall() decided that for us already.
1824 if (save_hdr
->holes
== 0) {
1826 * dest leaf has no holes, so we add there. May need
1827 * to make some room in the entry array.
1829 if (xfs_attr_leaf_order(save_blk
->bp
, drop_blk
->bp
)) {
1830 xfs_attr_leaf_moveents(drop_leaf
, 0, save_leaf
, 0,
1831 be16_to_cpu(drop_hdr
->count
), mp
);
1833 xfs_attr_leaf_moveents(drop_leaf
, 0, save_leaf
,
1834 be16_to_cpu(save_hdr
->count
),
1835 be16_to_cpu(drop_hdr
->count
), mp
);
1839 * Destination has holes, so we make a temporary copy
1840 * of the leaf and add them both to that.
1842 tmpbuffer
= kmem_alloc(state
->blocksize
, KM_SLEEP
);
1843 ASSERT(tmpbuffer
!= NULL
);
1844 memset(tmpbuffer
, 0, state
->blocksize
);
1845 tmp_leaf
= (xfs_attr_leafblock_t
*)tmpbuffer
;
1846 tmp_hdr
= &tmp_leaf
->hdr
;
1847 tmp_hdr
->info
= save_hdr
->info
; /* struct copy */
1849 tmp_hdr
->firstused
= cpu_to_be16(state
->blocksize
);
1850 if (!tmp_hdr
->firstused
) {
1851 tmp_hdr
->firstused
= cpu_to_be16(
1852 state
->blocksize
- XFS_ATTR_LEAF_NAME_ALIGN
);
1854 tmp_hdr
->usedbytes
= 0;
1855 if (xfs_attr_leaf_order(save_blk
->bp
, drop_blk
->bp
)) {
1856 xfs_attr_leaf_moveents(drop_leaf
, 0, tmp_leaf
, 0,
1857 be16_to_cpu(drop_hdr
->count
), mp
);
1858 xfs_attr_leaf_moveents(save_leaf
, 0, tmp_leaf
,
1859 be16_to_cpu(tmp_leaf
->hdr
.count
),
1860 be16_to_cpu(save_hdr
->count
), mp
);
1862 xfs_attr_leaf_moveents(save_leaf
, 0, tmp_leaf
, 0,
1863 be16_to_cpu(save_hdr
->count
), mp
);
1864 xfs_attr_leaf_moveents(drop_leaf
, 0, tmp_leaf
,
1865 be16_to_cpu(tmp_leaf
->hdr
.count
),
1866 be16_to_cpu(drop_hdr
->count
), mp
);
1868 memcpy((char *)save_leaf
, (char *)tmp_leaf
, state
->blocksize
);
1869 kmem_free(tmpbuffer
, state
->blocksize
);
1872 xfs_da_log_buf(state
->args
->trans
, save_blk
->bp
, 0,
1873 state
->blocksize
- 1);
1876 * Copy out last hashval in each block for B-tree code.
1878 save_blk
->hashval
= be32_to_cpu(
1879 save_leaf
->entries
[be16_to_cpu(save_leaf
->hdr
.count
)-1].hashval
);
1882 /*========================================================================
1883 * Routines used for finding things in the Btree.
1884 *========================================================================*/
1887 * Look up a name in a leaf attribute list structure.
1888 * This is the internal routine, it uses the caller's buffer.
1890 * Note that duplicate keys are allowed, but only check within the
1891 * current leaf node. The Btree code must check in adjacent leaf nodes.
1893 * Return in args->index the index into the entry[] array of either
1894 * the found entry, or where the entry should have been (insert before
1897 * Don't change the args->value unless we find the attribute.
1900 xfs_attr_leaf_lookup_int(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
)
1902 xfs_attr_leafblock_t
*leaf
;
1903 xfs_attr_leaf_entry_t
*entry
;
1904 xfs_attr_leaf_name_local_t
*name_loc
;
1905 xfs_attr_leaf_name_remote_t
*name_rmt
;
1907 xfs_dahash_t hashval
;
1910 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
1911 ASSERT(be16_to_cpu(leaf
->hdr
.count
)
1912 < (XFS_LBSIZE(args
->dp
->i_mount
)/8));
1915 * Binary search. (note: small blocks will skip this loop)
1917 hashval
= args
->hashval
;
1918 probe
= span
= be16_to_cpu(leaf
->hdr
.count
) / 2;
1919 for (entry
= &leaf
->entries
[probe
]; span
> 4;
1920 entry
= &leaf
->entries
[probe
]) {
1922 if (be32_to_cpu(entry
->hashval
) < hashval
)
1924 else if (be32_to_cpu(entry
->hashval
) > hashval
)
1929 ASSERT((probe
>= 0) &&
1931 || (probe
< be16_to_cpu(leaf
->hdr
.count
))));
1932 ASSERT((span
<= 4) || (be32_to_cpu(entry
->hashval
) == hashval
));
1935 * Since we may have duplicate hashval's, find the first matching
1936 * hashval in the leaf.
1938 while ((probe
> 0) && (be32_to_cpu(entry
->hashval
) >= hashval
)) {
1942 while ((probe
< be16_to_cpu(leaf
->hdr
.count
)) &&
1943 (be32_to_cpu(entry
->hashval
) < hashval
)) {
1947 if ((probe
== be16_to_cpu(leaf
->hdr
.count
)) ||
1948 (be32_to_cpu(entry
->hashval
) != hashval
)) {
1949 args
->index
= probe
;
1950 return(XFS_ERROR(ENOATTR
));
1954 * Duplicate keys may be present, so search all of them for a match.
1956 for ( ; (probe
< be16_to_cpu(leaf
->hdr
.count
)) &&
1957 (be32_to_cpu(entry
->hashval
) == hashval
);
1960 * GROT: Add code to remove incomplete entries.
1963 * If we are looking for INCOMPLETE entries, show only those.
1964 * If we are looking for complete entries, show only those.
1966 if ((args
->flags
& XFS_ATTR_INCOMPLETE
) !=
1967 (entry
->flags
& XFS_ATTR_INCOMPLETE
)) {
1970 if (entry
->flags
& XFS_ATTR_LOCAL
) {
1971 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, probe
);
1972 if (name_loc
->namelen
!= args
->namelen
)
1974 if (memcmp(args
->name
, (char *)name_loc
->nameval
,
1975 args
->namelen
) != 0)
1977 if (((args
->flags
& ATTR_SECURE
) != 0) !=
1978 ((entry
->flags
& XFS_ATTR_SECURE
) != 0))
1980 if (((args
->flags
& ATTR_ROOT
) != 0) !=
1981 ((entry
->flags
& XFS_ATTR_ROOT
) != 0))
1983 args
->index
= probe
;
1984 return(XFS_ERROR(EEXIST
));
1986 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, probe
);
1987 if (name_rmt
->namelen
!= args
->namelen
)
1989 if (memcmp(args
->name
, (char *)name_rmt
->name
,
1990 args
->namelen
) != 0)
1992 if (((args
->flags
& ATTR_SECURE
) != 0) !=
1993 ((entry
->flags
& XFS_ATTR_SECURE
) != 0))
1995 if (((args
->flags
& ATTR_ROOT
) != 0) !=
1996 ((entry
->flags
& XFS_ATTR_ROOT
) != 0))
1998 args
->index
= probe
;
1999 args
->rmtblkno
= be32_to_cpu(name_rmt
->valueblk
);
2000 args
->rmtblkcnt
= XFS_B_TO_FSB(args
->dp
->i_mount
,
2001 be32_to_cpu(name_rmt
->valuelen
));
2002 return(XFS_ERROR(EEXIST
));
2005 args
->index
= probe
;
2006 return(XFS_ERROR(ENOATTR
));
2010 * Get the value associated with an attribute name from a leaf attribute
2014 xfs_attr_leaf_getvalue(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
)
2017 xfs_attr_leafblock_t
*leaf
;
2018 xfs_attr_leaf_entry_t
*entry
;
2019 xfs_attr_leaf_name_local_t
*name_loc
;
2020 xfs_attr_leaf_name_remote_t
*name_rmt
;
2023 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2024 ASSERT(be16_to_cpu(leaf
->hdr
.count
)
2025 < (XFS_LBSIZE(args
->dp
->i_mount
)/8));
2026 ASSERT(args
->index
< be16_to_cpu(leaf
->hdr
.count
));
2028 entry
= &leaf
->entries
[args
->index
];
2029 if (entry
->flags
& XFS_ATTR_LOCAL
) {
2030 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, args
->index
);
2031 ASSERT(name_loc
->namelen
== args
->namelen
);
2032 ASSERT(memcmp(args
->name
, name_loc
->nameval
, args
->namelen
) == 0);
2033 valuelen
= be16_to_cpu(name_loc
->valuelen
);
2034 if (args
->flags
& ATTR_KERNOVAL
) {
2035 args
->valuelen
= valuelen
;
2038 if (args
->valuelen
< valuelen
) {
2039 args
->valuelen
= valuelen
;
2040 return(XFS_ERROR(ERANGE
));
2042 args
->valuelen
= valuelen
;
2043 memcpy(args
->value
, &name_loc
->nameval
[args
->namelen
], valuelen
);
2045 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, args
->index
);
2046 ASSERT(name_rmt
->namelen
== args
->namelen
);
2047 ASSERT(memcmp(args
->name
, name_rmt
->name
, args
->namelen
) == 0);
2048 valuelen
= be32_to_cpu(name_rmt
->valuelen
);
2049 args
->rmtblkno
= be32_to_cpu(name_rmt
->valueblk
);
2050 args
->rmtblkcnt
= XFS_B_TO_FSB(args
->dp
->i_mount
, valuelen
);
2051 if (args
->flags
& ATTR_KERNOVAL
) {
2052 args
->valuelen
= valuelen
;
2055 if (args
->valuelen
< valuelen
) {
2056 args
->valuelen
= valuelen
;
2057 return(XFS_ERROR(ERANGE
));
2059 args
->valuelen
= valuelen
;
2064 /*========================================================================
2066 *========================================================================*/
2069 * Move the indicated entries from one leaf to another.
2070 * NOTE: this routine modifies both source and destination leaves.
2074 xfs_attr_leaf_moveents(xfs_attr_leafblock_t
*leaf_s
, int start_s
,
2075 xfs_attr_leafblock_t
*leaf_d
, int start_d
,
2076 int count
, xfs_mount_t
*mp
)
2078 xfs_attr_leaf_hdr_t
*hdr_s
, *hdr_d
;
2079 xfs_attr_leaf_entry_t
*entry_s
, *entry_d
;
2083 * Check for nothing to do.
2089 * Set up environment.
2091 ASSERT(be16_to_cpu(leaf_s
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2092 ASSERT(be16_to_cpu(leaf_d
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2093 hdr_s
= &leaf_s
->hdr
;
2094 hdr_d
= &leaf_d
->hdr
;
2095 ASSERT((be16_to_cpu(hdr_s
->count
) > 0) &&
2096 (be16_to_cpu(hdr_s
->count
) < (XFS_LBSIZE(mp
)/8)));
2097 ASSERT(be16_to_cpu(hdr_s
->firstused
) >=
2098 ((be16_to_cpu(hdr_s
->count
)
2099 * sizeof(*entry_s
))+sizeof(*hdr_s
)));
2100 ASSERT(be16_to_cpu(hdr_d
->count
) < (XFS_LBSIZE(mp
)/8));
2101 ASSERT(be16_to_cpu(hdr_d
->firstused
) >=
2102 ((be16_to_cpu(hdr_d
->count
)
2103 * sizeof(*entry_d
))+sizeof(*hdr_d
)));
2105 ASSERT(start_s
< be16_to_cpu(hdr_s
->count
));
2106 ASSERT(start_d
<= be16_to_cpu(hdr_d
->count
));
2107 ASSERT(count
<= be16_to_cpu(hdr_s
->count
));
2110 * Move the entries in the destination leaf up to make a hole?
2112 if (start_d
< be16_to_cpu(hdr_d
->count
)) {
2113 tmp
= be16_to_cpu(hdr_d
->count
) - start_d
;
2114 tmp
*= sizeof(xfs_attr_leaf_entry_t
);
2115 entry_s
= &leaf_d
->entries
[start_d
];
2116 entry_d
= &leaf_d
->entries
[start_d
+ count
];
2117 memmove((char *)entry_d
, (char *)entry_s
, tmp
);
2121 * Copy all entry's in the same (sorted) order,
2122 * but allocate attribute info packed and in sequence.
2124 entry_s
= &leaf_s
->entries
[start_s
];
2125 entry_d
= &leaf_d
->entries
[start_d
];
2127 for (i
= 0; i
< count
; entry_s
++, entry_d
++, desti
++, i
++) {
2128 ASSERT(be16_to_cpu(entry_s
->nameidx
)
2129 >= be16_to_cpu(hdr_s
->firstused
));
2130 tmp
= xfs_attr_leaf_entsize(leaf_s
, start_s
+ i
);
2133 * Code to drop INCOMPLETE entries. Difficult to use as we
2134 * may also need to change the insertion index. Code turned
2135 * off for 6.2, should be revisited later.
2137 if (entry_s
->flags
& XFS_ATTR_INCOMPLETE
) { /* skip partials? */
2138 memset(XFS_ATTR_LEAF_NAME(leaf_s
, start_s
+ i
), 0, tmp
);
2139 be16_add(&hdr_s
->usedbytes
, -tmp
);
2140 be16_add(&hdr_s
->count
, -1);
2141 entry_d
--; /* to compensate for ++ in loop hdr */
2143 if ((start_s
+ i
) < offset
)
2144 result
++; /* insertion index adjustment */
2147 be16_add(&hdr_d
->firstused
, -tmp
);
2148 /* both on-disk, don't endian flip twice */
2149 entry_d
->hashval
= entry_s
->hashval
;
2150 /* both on-disk, don't endian flip twice */
2151 entry_d
->nameidx
= hdr_d
->firstused
;
2152 entry_d
->flags
= entry_s
->flags
;
2153 ASSERT(be16_to_cpu(entry_d
->nameidx
) + tmp
2155 memmove(XFS_ATTR_LEAF_NAME(leaf_d
, desti
),
2156 XFS_ATTR_LEAF_NAME(leaf_s
, start_s
+ i
), tmp
);
2157 ASSERT(be16_to_cpu(entry_s
->nameidx
) + tmp
2159 memset(XFS_ATTR_LEAF_NAME(leaf_s
, start_s
+ i
), 0, tmp
);
2160 be16_add(&hdr_s
->usedbytes
, -tmp
);
2161 be16_add(&hdr_d
->usedbytes
, tmp
);
2162 be16_add(&hdr_s
->count
, -1);
2163 be16_add(&hdr_d
->count
, 1);
2164 tmp
= be16_to_cpu(hdr_d
->count
)
2165 * sizeof(xfs_attr_leaf_entry_t
)
2166 + sizeof(xfs_attr_leaf_hdr_t
);
2167 ASSERT(be16_to_cpu(hdr_d
->firstused
) >= tmp
);
2174 * Zero out the entries we just copied.
2176 if (start_s
== be16_to_cpu(hdr_s
->count
)) {
2177 tmp
= count
* sizeof(xfs_attr_leaf_entry_t
);
2178 entry_s
= &leaf_s
->entries
[start_s
];
2179 ASSERT(((char *)entry_s
+ tmp
) <=
2180 ((char *)leaf_s
+ XFS_LBSIZE(mp
)));
2181 memset((char *)entry_s
, 0, tmp
);
2184 * Move the remaining entries down to fill the hole,
2185 * then zero the entries at the top.
2187 tmp
= be16_to_cpu(hdr_s
->count
) - count
;
2188 tmp
*= sizeof(xfs_attr_leaf_entry_t
);
2189 entry_s
= &leaf_s
->entries
[start_s
+ count
];
2190 entry_d
= &leaf_s
->entries
[start_s
];
2191 memmove((char *)entry_d
, (char *)entry_s
, tmp
);
2193 tmp
= count
* sizeof(xfs_attr_leaf_entry_t
);
2194 entry_s
= &leaf_s
->entries
[be16_to_cpu(hdr_s
->count
)];
2195 ASSERT(((char *)entry_s
+ tmp
) <=
2196 ((char *)leaf_s
+ XFS_LBSIZE(mp
)));
2197 memset((char *)entry_s
, 0, tmp
);
2201 * Fill in the freemap information
2203 hdr_d
->freemap
[0].base
= cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t
));
2204 be16_add(&hdr_d
->freemap
[0].base
, be16_to_cpu(hdr_d
->count
) *
2205 sizeof(xfs_attr_leaf_entry_t
));
2206 hdr_d
->freemap
[0].size
= cpu_to_be16(be16_to_cpu(hdr_d
->firstused
)
2207 - be16_to_cpu(hdr_d
->freemap
[0].base
));
2208 hdr_d
->freemap
[1].base
= 0;
2209 hdr_d
->freemap
[2].base
= 0;
2210 hdr_d
->freemap
[1].size
= 0;
2211 hdr_d
->freemap
[2].size
= 0;
2212 hdr_s
->holes
= 1; /* leaf may not be compact */
2216 * Compare two leaf blocks "order".
2217 * Return 0 unless leaf2 should go before leaf1.
2220 xfs_attr_leaf_order(xfs_dabuf_t
*leaf1_bp
, xfs_dabuf_t
*leaf2_bp
)
2222 xfs_attr_leafblock_t
*leaf1
, *leaf2
;
2224 leaf1
= leaf1_bp
->data
;
2225 leaf2
= leaf2_bp
->data
;
2226 ASSERT((be16_to_cpu(leaf1
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
) &&
2227 (be16_to_cpu(leaf2
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
));
2228 if ((be16_to_cpu(leaf1
->hdr
.count
) > 0) &&
2229 (be16_to_cpu(leaf2
->hdr
.count
) > 0) &&
2230 ((be32_to_cpu(leaf2
->entries
[0].hashval
) <
2231 be32_to_cpu(leaf1
->entries
[0].hashval
)) ||
2232 (be32_to_cpu(leaf2
->entries
[
2233 be16_to_cpu(leaf2
->hdr
.count
)-1].hashval
) <
2234 be32_to_cpu(leaf1
->entries
[
2235 be16_to_cpu(leaf1
->hdr
.count
)-1].hashval
)))) {
2242 * Pick up the last hashvalue from a leaf block.
2245 xfs_attr_leaf_lasthash(xfs_dabuf_t
*bp
, int *count
)
2247 xfs_attr_leafblock_t
*leaf
;
2250 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2252 *count
= be16_to_cpu(leaf
->hdr
.count
);
2253 if (!leaf
->hdr
.count
)
2255 return be32_to_cpu(leaf
->entries
[be16_to_cpu(leaf
->hdr
.count
)-1].hashval
);
2259 * Calculate the number of bytes used to store the indicated attribute
2260 * (whether local or remote only calculate bytes in this block).
2263 xfs_attr_leaf_entsize(xfs_attr_leafblock_t
*leaf
, int index
)
2265 xfs_attr_leaf_name_local_t
*name_loc
;
2266 xfs_attr_leaf_name_remote_t
*name_rmt
;
2269 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2270 if (leaf
->entries
[index
].flags
& XFS_ATTR_LOCAL
) {
2271 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, index
);
2272 size
= XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc
->namelen
,
2273 be16_to_cpu(name_loc
->valuelen
));
2275 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, index
);
2276 size
= XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt
->namelen
);
2282 * Calculate the number of bytes that would be required to store the new
2283 * attribute (whether local or remote only calculate bytes in this block).
2284 * This routine decides as a side effect whether the attribute will be
2285 * a "local" or a "remote" attribute.
2288 xfs_attr_leaf_newentsize(int namelen
, int valuelen
, int blocksize
, int *local
)
2292 size
= XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen
, valuelen
);
2293 if (size
< XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize
)) {
2298 size
= XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen
);
2307 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2310 xfs_attr_leaf_list_int(xfs_dabuf_t
*bp
, xfs_attr_list_context_t
*context
)
2312 attrlist_cursor_kern_t
*cursor
;
2313 xfs_attr_leafblock_t
*leaf
;
2314 xfs_attr_leaf_entry_t
*entry
;
2315 xfs_attr_leaf_name_local_t
*name_loc
;
2316 xfs_attr_leaf_name_remote_t
*name_rmt
;
2321 cursor
= context
->cursor
;
2322 cursor
->initted
= 1;
2324 xfs_attr_trace_l_cl("blk start", context
, leaf
);
2327 * Re-find our place in the leaf block if this is a new syscall.
2329 if (context
->resynch
) {
2330 entry
= &leaf
->entries
[0];
2331 for (i
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); entry
++, i
++) {
2332 if (be32_to_cpu(entry
->hashval
) == cursor
->hashval
) {
2333 if (cursor
->offset
== context
->dupcnt
) {
2334 context
->dupcnt
= 0;
2338 } else if (be32_to_cpu(entry
->hashval
) >
2340 context
->dupcnt
= 0;
2344 if (i
== be16_to_cpu(leaf
->hdr
.count
)) {
2345 xfs_attr_trace_l_c("not found", context
);
2349 entry
= &leaf
->entries
[0];
2352 context
->resynch
= 0;
2355 * We have found our place, start copying out the new attributes.
2358 for ( ; (i
< be16_to_cpu(leaf
->hdr
.count
))
2359 && (retval
== 0); entry
++, i
++) {
2360 attrnames_t
*namesp
;
2362 if (be32_to_cpu(entry
->hashval
) != cursor
->hashval
) {
2363 cursor
->hashval
= be32_to_cpu(entry
->hashval
);
2367 if (entry
->flags
& XFS_ATTR_INCOMPLETE
)
2368 continue; /* skip incomplete entries */
2369 if (((context
->flags
& ATTR_SECURE
) != 0) !=
2370 ((entry
->flags
& XFS_ATTR_SECURE
) != 0) &&
2371 !(context
->flags
& ATTR_KERNORMALS
))
2372 continue; /* skip non-matching entries */
2373 if (((context
->flags
& ATTR_ROOT
) != 0) !=
2374 ((entry
->flags
& XFS_ATTR_ROOT
) != 0) &&
2375 !(context
->flags
& ATTR_KERNROOTLS
))
2376 continue; /* skip non-matching entries */
2378 namesp
= (entry
->flags
& XFS_ATTR_SECURE
) ? &attr_secure
:
2379 ((entry
->flags
& XFS_ATTR_ROOT
) ? &attr_trusted
:
2382 if (entry
->flags
& XFS_ATTR_LOCAL
) {
2383 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, i
);
2384 if (context
->flags
& ATTR_KERNOVAL
) {
2385 ASSERT(context
->flags
& ATTR_KERNAMELS
);
2386 context
->count
+= namesp
->attr_namelen
+
2387 (int)name_loc
->namelen
+ 1;
2389 retval
= xfs_attr_put_listent(context
, namesp
,
2390 (char *)name_loc
->nameval
,
2391 (int)name_loc
->namelen
,
2392 be16_to_cpu(name_loc
->valuelen
));
2395 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, i
);
2396 if (context
->flags
& ATTR_KERNOVAL
) {
2397 ASSERT(context
->flags
& ATTR_KERNAMELS
);
2398 context
->count
+= namesp
->attr_namelen
+
2399 (int)name_rmt
->namelen
+ 1;
2401 retval
= xfs_attr_put_listent(context
, namesp
,
2402 (char *)name_rmt
->name
,
2403 (int)name_rmt
->namelen
,
2404 be32_to_cpu(name_rmt
->valuelen
));
2411 xfs_attr_trace_l_cl("blk end", context
, leaf
);
2415 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
2416 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
2417 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
2418 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
2419 & ~(sizeof(u_int32_t)-1))
2422 * Format an attribute and copy it out to the user's buffer.
2423 * Take care to check values and protect against them changing later,
2424 * we may be reading them directly out of a user buffer.
2428 xfs_attr_put_listent(xfs_attr_list_context_t
*context
,
2429 attrnames_t
*namesp
, char *name
, int namelen
, int valuelen
)
2431 attrlist_ent_t
*aep
;
2434 ASSERT(!(context
->flags
& ATTR_KERNOVAL
));
2435 if (context
->flags
& ATTR_KERNAMELS
) {
2438 ASSERT(context
->count
>= 0);
2440 arraytop
= context
->count
+ namesp
->attr_namelen
+ namelen
+ 1;
2441 if (arraytop
> context
->firstu
) {
2442 context
->count
= -1; /* insufficient space */
2445 offset
= (char *)context
->alist
+ context
->count
;
2446 strncpy(offset
, namesp
->attr_name
, namesp
->attr_namelen
);
2447 offset
+= namesp
->attr_namelen
;
2448 strncpy(offset
, name
, namelen
); /* real name */
2451 context
->count
+= namesp
->attr_namelen
+ namelen
+ 1;
2455 ASSERT(context
->count
>= 0);
2456 ASSERT(context
->count
< (ATTR_MAX_VALUELEN
/8));
2457 ASSERT(context
->firstu
>= sizeof(*context
->alist
));
2458 ASSERT(context
->firstu
<= context
->bufsize
);
2460 arraytop
= sizeof(*context
->alist
) +
2461 context
->count
* sizeof(context
->alist
->al_offset
[0]);
2462 context
->firstu
-= ATTR_ENTSIZE(namelen
);
2463 if (context
->firstu
< arraytop
) {
2464 xfs_attr_trace_l_c("buffer full", context
);
2465 context
->alist
->al_more
= 1;
2469 aep
= (attrlist_ent_t
*)&(((char *)context
->alist
)[ context
->firstu
]);
2470 aep
->a_valuelen
= valuelen
;
2471 memcpy(aep
->a_name
, name
, namelen
);
2472 aep
->a_name
[ namelen
] = 0;
2473 context
->alist
->al_offset
[ context
->count
++ ] = context
->firstu
;
2474 context
->alist
->al_count
= context
->count
;
2475 xfs_attr_trace_l_c("add", context
);
2479 /*========================================================================
2480 * Manage the INCOMPLETE flag in a leaf entry
2481 *========================================================================*/
2484 * Clear the INCOMPLETE flag on an entry in a leaf block.
2487 xfs_attr_leaf_clearflag(xfs_da_args_t
*args
)
2489 xfs_attr_leafblock_t
*leaf
;
2490 xfs_attr_leaf_entry_t
*entry
;
2491 xfs_attr_leaf_name_remote_t
*name_rmt
;
2495 xfs_attr_leaf_name_local_t
*name_loc
;
2501 * Set up the operation.
2503 error
= xfs_da_read_buf(args
->trans
, args
->dp
, args
->blkno
, -1, &bp
,
2511 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2512 ASSERT(args
->index
< be16_to_cpu(leaf
->hdr
.count
));
2513 ASSERT(args
->index
>= 0);
2514 entry
= &leaf
->entries
[ args
->index
];
2515 ASSERT(entry
->flags
& XFS_ATTR_INCOMPLETE
);
2518 if (entry
->flags
& XFS_ATTR_LOCAL
) {
2519 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf
, args
->index
);
2520 namelen
= name_loc
->namelen
;
2521 name
= (char *)name_loc
->nameval
;
2523 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, args
->index
);
2524 namelen
= name_rmt
->namelen
;
2525 name
= (char *)name_rmt
->name
;
2527 ASSERT(be32_to_cpu(entry
->hashval
) == args
->hashval
);
2528 ASSERT(namelen
== args
->namelen
);
2529 ASSERT(memcmp(name
, args
->name
, namelen
) == 0);
2532 entry
->flags
&= ~XFS_ATTR_INCOMPLETE
;
2533 xfs_da_log_buf(args
->trans
, bp
,
2534 XFS_DA_LOGRANGE(leaf
, entry
, sizeof(*entry
)));
2536 if (args
->rmtblkno
) {
2537 ASSERT((entry
->flags
& XFS_ATTR_LOCAL
) == 0);
2538 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, args
->index
);
2539 name_rmt
->valueblk
= cpu_to_be32(args
->rmtblkno
);
2540 name_rmt
->valuelen
= cpu_to_be32(args
->valuelen
);
2541 xfs_da_log_buf(args
->trans
, bp
,
2542 XFS_DA_LOGRANGE(leaf
, name_rmt
, sizeof(*name_rmt
)));
2544 xfs_da_buf_done(bp
);
2547 * Commit the flag value change and start the next trans in series.
2549 error
= xfs_attr_rolltrans(&args
->trans
, args
->dp
);
2555 * Set the INCOMPLETE flag on an entry in a leaf block.
2558 xfs_attr_leaf_setflag(xfs_da_args_t
*args
)
2560 xfs_attr_leafblock_t
*leaf
;
2561 xfs_attr_leaf_entry_t
*entry
;
2562 xfs_attr_leaf_name_remote_t
*name_rmt
;
2567 * Set up the operation.
2569 error
= xfs_da_read_buf(args
->trans
, args
->dp
, args
->blkno
, -1, &bp
,
2577 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2578 ASSERT(args
->index
< be16_to_cpu(leaf
->hdr
.count
));
2579 ASSERT(args
->index
>= 0);
2580 entry
= &leaf
->entries
[ args
->index
];
2582 ASSERT((entry
->flags
& XFS_ATTR_INCOMPLETE
) == 0);
2583 entry
->flags
|= XFS_ATTR_INCOMPLETE
;
2584 xfs_da_log_buf(args
->trans
, bp
,
2585 XFS_DA_LOGRANGE(leaf
, entry
, sizeof(*entry
)));
2586 if ((entry
->flags
& XFS_ATTR_LOCAL
) == 0) {
2587 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, args
->index
);
2588 name_rmt
->valueblk
= 0;
2589 name_rmt
->valuelen
= 0;
2590 xfs_da_log_buf(args
->trans
, bp
,
2591 XFS_DA_LOGRANGE(leaf
, name_rmt
, sizeof(*name_rmt
)));
2593 xfs_da_buf_done(bp
);
2596 * Commit the flag value change and start the next trans in series.
2598 error
= xfs_attr_rolltrans(&args
->trans
, args
->dp
);
2604 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2605 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2606 * entry given by args->blkno2/index2.
2608 * Note that they could be in different blocks, or in the same block.
2611 xfs_attr_leaf_flipflags(xfs_da_args_t
*args
)
2613 xfs_attr_leafblock_t
*leaf1
, *leaf2
;
2614 xfs_attr_leaf_entry_t
*entry1
, *entry2
;
2615 xfs_attr_leaf_name_remote_t
*name_rmt
;
2616 xfs_dabuf_t
*bp1
, *bp2
;
2619 xfs_attr_leaf_name_local_t
*name_loc
;
2620 int namelen1
, namelen2
;
2621 char *name1
, *name2
;
2625 * Read the block containing the "old" attr
2627 error
= xfs_da_read_buf(args
->trans
, args
->dp
, args
->blkno
, -1, &bp1
,
2632 ASSERT(bp1
!= NULL
);
2635 * Read the block containing the "new" attr, if it is different
2637 if (args
->blkno2
!= args
->blkno
) {
2638 error
= xfs_da_read_buf(args
->trans
, args
->dp
, args
->blkno2
,
2639 -1, &bp2
, XFS_ATTR_FORK
);
2643 ASSERT(bp2
!= NULL
);
2649 ASSERT(be16_to_cpu(leaf1
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2650 ASSERT(args
->index
< be16_to_cpu(leaf1
->hdr
.count
));
2651 ASSERT(args
->index
>= 0);
2652 entry1
= &leaf1
->entries
[ args
->index
];
2655 ASSERT(be16_to_cpu(leaf2
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2656 ASSERT(args
->index2
< be16_to_cpu(leaf2
->hdr
.count
));
2657 ASSERT(args
->index2
>= 0);
2658 entry2
= &leaf2
->entries
[ args
->index2
];
2661 if (entry1
->flags
& XFS_ATTR_LOCAL
) {
2662 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf1
, args
->index
);
2663 namelen1
= name_loc
->namelen
;
2664 name1
= (char *)name_loc
->nameval
;
2666 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf1
, args
->index
);
2667 namelen1
= name_rmt
->namelen
;
2668 name1
= (char *)name_rmt
->name
;
2670 if (entry2
->flags
& XFS_ATTR_LOCAL
) {
2671 name_loc
= XFS_ATTR_LEAF_NAME_LOCAL(leaf2
, args
->index2
);
2672 namelen2
= name_loc
->namelen
;
2673 name2
= (char *)name_loc
->nameval
;
2675 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf2
, args
->index2
);
2676 namelen2
= name_rmt
->namelen
;
2677 name2
= (char *)name_rmt
->name
;
2679 ASSERT(be32_to_cpu(entry1
->hashval
) == be32_to_cpu(entry2
->hashval
));
2680 ASSERT(namelen1
== namelen2
);
2681 ASSERT(memcmp(name1
, name2
, namelen1
) == 0);
2684 ASSERT(entry1
->flags
& XFS_ATTR_INCOMPLETE
);
2685 ASSERT((entry2
->flags
& XFS_ATTR_INCOMPLETE
) == 0);
2687 entry1
->flags
&= ~XFS_ATTR_INCOMPLETE
;
2688 xfs_da_log_buf(args
->trans
, bp1
,
2689 XFS_DA_LOGRANGE(leaf1
, entry1
, sizeof(*entry1
)));
2690 if (args
->rmtblkno
) {
2691 ASSERT((entry1
->flags
& XFS_ATTR_LOCAL
) == 0);
2692 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf1
, args
->index
);
2693 name_rmt
->valueblk
= cpu_to_be32(args
->rmtblkno
);
2694 name_rmt
->valuelen
= cpu_to_be32(args
->valuelen
);
2695 xfs_da_log_buf(args
->trans
, bp1
,
2696 XFS_DA_LOGRANGE(leaf1
, name_rmt
, sizeof(*name_rmt
)));
2699 entry2
->flags
|= XFS_ATTR_INCOMPLETE
;
2700 xfs_da_log_buf(args
->trans
, bp2
,
2701 XFS_DA_LOGRANGE(leaf2
, entry2
, sizeof(*entry2
)));
2702 if ((entry2
->flags
& XFS_ATTR_LOCAL
) == 0) {
2703 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf2
, args
->index2
);
2704 name_rmt
->valueblk
= 0;
2705 name_rmt
->valuelen
= 0;
2706 xfs_da_log_buf(args
->trans
, bp2
,
2707 XFS_DA_LOGRANGE(leaf2
, name_rmt
, sizeof(*name_rmt
)));
2709 xfs_da_buf_done(bp1
);
2711 xfs_da_buf_done(bp2
);
2714 * Commit the flag value change and start the next trans in series.
2716 error
= xfs_attr_rolltrans(&args
->trans
, args
->dp
);
2721 /*========================================================================
2722 * Indiscriminately delete the entire attribute fork
2723 *========================================================================*/
2726 * Recurse (gasp!) through the attribute nodes until we find leaves.
2727 * We're doing a depth-first traversal in order to invalidate everything.
2730 xfs_attr_root_inactive(xfs_trans_t
**trans
, xfs_inode_t
*dp
)
2732 xfs_da_blkinfo_t
*info
;
2738 * Read block 0 to see what we have to work with.
2739 * We only get here if we have extents, since we remove
2740 * the extents in reverse order the extent containing
2741 * block 0 must still be there.
2743 error
= xfs_da_read_buf(*trans
, dp
, 0, -1, &bp
, XFS_ATTR_FORK
);
2746 blkno
= xfs_da_blkno(bp
);
2749 * Invalidate the tree, even if the "tree" is only a single leaf block.
2750 * This is a depth-first traversal!
2753 if (be16_to_cpu(info
->magic
) == XFS_DA_NODE_MAGIC
) {
2754 error
= xfs_attr_node_inactive(trans
, dp
, bp
, 1);
2755 } else if (be16_to_cpu(info
->magic
) == XFS_ATTR_LEAF_MAGIC
) {
2756 error
= xfs_attr_leaf_inactive(trans
, dp
, bp
);
2758 error
= XFS_ERROR(EIO
);
2759 xfs_da_brelse(*trans
, bp
);
2765 * Invalidate the incore copy of the root block.
2767 error
= xfs_da_get_buf(*trans
, dp
, 0, blkno
, &bp
, XFS_ATTR_FORK
);
2770 xfs_da_binval(*trans
, bp
); /* remove from cache */
2772 * Commit the invalidate and start the next transaction.
2774 error
= xfs_attr_rolltrans(trans
, dp
);
2780 * Recurse (gasp!) through the attribute nodes until we find leaves.
2781 * We're doing a depth-first traversal in order to invalidate everything.
2784 xfs_attr_node_inactive(xfs_trans_t
**trans
, xfs_inode_t
*dp
, xfs_dabuf_t
*bp
,
2787 xfs_da_blkinfo_t
*info
;
2788 xfs_da_intnode_t
*node
;
2789 xfs_dablk_t child_fsb
;
2790 xfs_daddr_t parent_blkno
, child_blkno
;
2791 int error
, count
, i
;
2792 xfs_dabuf_t
*child_bp
;
2795 * Since this code is recursive (gasp!) we must protect ourselves.
2797 if (level
> XFS_DA_NODE_MAXDEPTH
) {
2798 xfs_da_brelse(*trans
, bp
); /* no locks for later trans */
2799 return(XFS_ERROR(EIO
));
2803 ASSERT(be16_to_cpu(node
->hdr
.info
.magic
) == XFS_DA_NODE_MAGIC
);
2804 parent_blkno
= xfs_da_blkno(bp
); /* save for re-read later */
2805 count
= be16_to_cpu(node
->hdr
.count
);
2807 xfs_da_brelse(*trans
, bp
);
2810 child_fsb
= be32_to_cpu(node
->btree
[0].before
);
2811 xfs_da_brelse(*trans
, bp
); /* no locks for later trans */
2814 * If this is the node level just above the leaves, simply loop
2815 * over the leaves removing all of them. If this is higher up
2816 * in the tree, recurse downward.
2818 for (i
= 0; i
< count
; i
++) {
2820 * Read the subsidiary block to see what we have to work with.
2821 * Don't do this in a transaction. This is a depth-first
2822 * traversal of the tree so we may deal with many blocks
2823 * before we come back to this one.
2825 error
= xfs_da_read_buf(*trans
, dp
, child_fsb
, -2, &child_bp
,
2830 /* save for re-read later */
2831 child_blkno
= xfs_da_blkno(child_bp
);
2834 * Invalidate the subtree, however we have to.
2836 info
= child_bp
->data
;
2837 if (be16_to_cpu(info
->magic
) == XFS_DA_NODE_MAGIC
) {
2838 error
= xfs_attr_node_inactive(trans
, dp
,
2840 } else if (be16_to_cpu(info
->magic
) == XFS_ATTR_LEAF_MAGIC
) {
2841 error
= xfs_attr_leaf_inactive(trans
, dp
,
2844 error
= XFS_ERROR(EIO
);
2845 xfs_da_brelse(*trans
, child_bp
);
2851 * Remove the subsidiary block from the cache
2854 error
= xfs_da_get_buf(*trans
, dp
, 0, child_blkno
,
2855 &child_bp
, XFS_ATTR_FORK
);
2858 xfs_da_binval(*trans
, child_bp
);
2862 * If we're not done, re-read the parent to get the next
2863 * child block number.
2865 if ((i
+1) < count
) {
2866 error
= xfs_da_read_buf(*trans
, dp
, 0, parent_blkno
,
2867 &bp
, XFS_ATTR_FORK
);
2870 child_fsb
= be32_to_cpu(node
->btree
[i
+1].before
);
2871 xfs_da_brelse(*trans
, bp
);
2874 * Atomically commit the whole invalidate stuff.
2876 if ((error
= xfs_attr_rolltrans(trans
, dp
)))
2884 * Invalidate all of the "remote" value regions pointed to by a particular
2886 * Note that we must release the lock on the buffer so that we are not
2887 * caught holding something that the logging code wants to flush to disk.
2890 xfs_attr_leaf_inactive(xfs_trans_t
**trans
, xfs_inode_t
*dp
, xfs_dabuf_t
*bp
)
2892 xfs_attr_leafblock_t
*leaf
;
2893 xfs_attr_leaf_entry_t
*entry
;
2894 xfs_attr_leaf_name_remote_t
*name_rmt
;
2895 xfs_attr_inactive_list_t
*list
, *lp
;
2896 int error
, count
, size
, tmp
, i
;
2899 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_ATTR_LEAF_MAGIC
);
2902 * Count the number of "remote" value extents.
2905 entry
= &leaf
->entries
[0];
2906 for (i
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); entry
++, i
++) {
2907 if (be16_to_cpu(entry
->nameidx
) &&
2908 ((entry
->flags
& XFS_ATTR_LOCAL
) == 0)) {
2909 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, i
);
2910 if (name_rmt
->valueblk
)
2916 * If there are no "remote" values, we're done.
2919 xfs_da_brelse(*trans
, bp
);
2924 * Allocate storage for a list of all the "remote" value extents.
2926 size
= count
* sizeof(xfs_attr_inactive_list_t
);
2927 list
= (xfs_attr_inactive_list_t
*)kmem_alloc(size
, KM_SLEEP
);
2930 * Identify each of the "remote" value extents.
2933 entry
= &leaf
->entries
[0];
2934 for (i
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); entry
++, i
++) {
2935 if (be16_to_cpu(entry
->nameidx
) &&
2936 ((entry
->flags
& XFS_ATTR_LOCAL
) == 0)) {
2937 name_rmt
= XFS_ATTR_LEAF_NAME_REMOTE(leaf
, i
);
2938 if (name_rmt
->valueblk
) {
2939 lp
->valueblk
= be32_to_cpu(name_rmt
->valueblk
);
2940 lp
->valuelen
= XFS_B_TO_FSB(dp
->i_mount
,
2941 be32_to_cpu(name_rmt
->valuelen
));
2946 xfs_da_brelse(*trans
, bp
); /* unlock for trans. in freextent() */
2949 * Invalidate each of the "remote" value extents.
2952 for (lp
= list
, i
= 0; i
< count
; i
++, lp
++) {
2953 tmp
= xfs_attr_leaf_freextent(trans
, dp
,
2954 lp
->valueblk
, lp
->valuelen
);
2957 error
= tmp
; /* save only the 1st errno */
2960 kmem_free((xfs_caddr_t
)list
, size
);
2965 * Look at all the extents for this logical region,
2966 * invalidate any buffers that are incore/in transactions.
2969 xfs_attr_leaf_freextent(xfs_trans_t
**trans
, xfs_inode_t
*dp
,
2970 xfs_dablk_t blkno
, int blkcnt
)
2972 xfs_bmbt_irec_t map
;
2974 int tblkcnt
, dblkcnt
, nmap
, error
;
2979 * Roll through the "value", invalidating the attribute value's
2984 while (tblkcnt
> 0) {
2986 * Try to remember where we decided to put the value.
2989 error
= xfs_bmapi(*trans
, dp
, (xfs_fileoff_t
)tblkno
, tblkcnt
,
2990 XFS_BMAPI_ATTRFORK
| XFS_BMAPI_METADATA
,
2991 NULL
, 0, &map
, &nmap
, NULL
, NULL
);
2996 ASSERT(map
.br_startblock
!= DELAYSTARTBLOCK
);
2999 * If it's a hole, these are already unmapped
3000 * so there's nothing to invalidate.
3002 if (map
.br_startblock
!= HOLESTARTBLOCK
) {
3004 dblkno
= XFS_FSB_TO_DADDR(dp
->i_mount
,
3006 dblkcnt
= XFS_FSB_TO_BB(dp
->i_mount
,
3008 bp
= xfs_trans_get_buf(*trans
,
3009 dp
->i_mount
->m_ddev_targp
,
3010 dblkno
, dblkcnt
, XFS_BUF_LOCK
);
3011 xfs_trans_binval(*trans
, bp
);
3013 * Roll to next transaction.
3015 if ((error
= xfs_attr_rolltrans(trans
, dp
)))
3019 tblkno
+= map
.br_blockcount
;
3020 tblkcnt
-= map
.br_blockcount
;
3028 * Roll from one trans in the sequence of PERMANENT transactions to the next.
3031 xfs_attr_rolltrans(xfs_trans_t
**transp
, xfs_inode_t
*dp
)
3034 unsigned int logres
, count
;
3038 * Ensure that the inode is always logged.
3041 xfs_trans_log_inode(trans
, dp
, XFS_ILOG_CORE
);
3044 * Copy the critical parameters from one trans to the next.
3046 logres
= trans
->t_log_res
;
3047 count
= trans
->t_log_count
;
3048 *transp
= xfs_trans_dup(trans
);
3051 * Commit the current transaction.
3052 * If this commit failed, then it'd just unlock those items that
3053 * are not marked ihold. That also means that a filesystem shutdown
3054 * is in progress. The caller takes the responsibility to cancel
3055 * the duplicate transaction that gets returned.
3057 if ((error
= xfs_trans_commit(trans
, 0, NULL
)))
3063 * Reserve space in the log for th next transaction.
3064 * This also pushes items in the "AIL", the list of logged items,
3065 * out to disk if they are taking up space at the tail of the log
3066 * that we want to use. This requires that either nothing be locked
3067 * across this call, or that anything that is locked be logged in
3068 * the prior and the next transactions.
3070 error
= xfs_trans_reserve(trans
, 0, logres
, 0,
3071 XFS_TRANS_PERM_LOG_RES
, count
);
3073 * Ensure that the inode is in the new transaction and locked.
3076 xfs_trans_ijoin(trans
, dp
, XFS_ILOCK_EXCL
);
3077 xfs_trans_ihold(trans
, dp
);