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_format.h"
23 #include "xfs_trans.h"
24 #include "xfs_trans_priv.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_attr_sf.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_alloc.h"
34 #include "xfs_inode_item.h"
36 #include "xfs_bmap_util.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_attr_remote.h"
40 #include "xfs_error.h"
41 #include "xfs_quota.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_trace.h"
48 * Provide the external interfaces to manage attribute lists.
51 /*========================================================================
52 * Function prototypes for the kernel.
53 *========================================================================*/
56 * Internal routines when attribute list fits inside the inode.
58 STATIC
int xfs_attr_shortform_addname(xfs_da_args_t
*args
);
61 * Internal routines when attribute list is one block.
63 STATIC
int xfs_attr_leaf_get(xfs_da_args_t
*args
);
64 STATIC
int xfs_attr_leaf_addname(xfs_da_args_t
*args
);
65 STATIC
int xfs_attr_leaf_removename(xfs_da_args_t
*args
);
68 * Internal routines when attribute list is more than one block.
70 STATIC
int xfs_attr_node_get(xfs_da_args_t
*args
);
71 STATIC
int xfs_attr_node_addname(xfs_da_args_t
*args
);
72 STATIC
int xfs_attr_node_removename(xfs_da_args_t
*args
);
73 STATIC
int xfs_attr_fillstate(xfs_da_state_t
*state
);
74 STATIC
int xfs_attr_refillstate(xfs_da_state_t
*state
);
78 xfs_attr_name_to_xname(
79 struct xfs_name
*xname
,
80 const unsigned char *aname
)
85 xname
->len
= strlen((char *)aname
);
86 if (xname
->len
>= MAXNAMELEN
)
87 return EFAULT
; /* match IRIX behaviour */
96 if (!XFS_IFORK_Q(ip
) ||
97 (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
&&
98 ip
->i_d
.di_anextents
== 0))
103 /*========================================================================
104 * Overall external interface routines.
105 *========================================================================*/
109 struct xfs_inode
*ip
,
110 struct xfs_name
*name
,
111 unsigned char *value
,
118 if (!xfs_inode_hasattr(ip
))
122 * Fill in the arg structure for this request.
124 memset((char *)&args
, 0, sizeof(args
));
125 args
.name
= name
->name
;
126 args
.namelen
= name
->len
;
128 args
.valuelen
= *valuelenp
;
130 args
.hashval
= xfs_da_hashname(args
.name
, args
.namelen
);
132 args
.whichfork
= XFS_ATTR_FORK
;
135 * Decide on what work routines to call based on the inode size.
137 if (ip
->i_d
.di_aformat
== XFS_DINODE_FMT_LOCAL
) {
138 error
= xfs_attr_shortform_getvalue(&args
);
139 } else if (xfs_bmap_one_block(ip
, XFS_ATTR_FORK
)) {
140 error
= xfs_attr_leaf_get(&args
);
142 error
= xfs_attr_node_get(&args
);
146 * Return the number of bytes in the value to the caller.
148 *valuelenp
= args
.valuelen
;
158 const unsigned char *name
,
159 unsigned char *value
,
164 struct xfs_name xname
;
166 XFS_STATS_INC(xs_attr_get
);
168 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
171 error
= xfs_attr_name_to_xname(&xname
, name
);
175 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
176 error
= xfs_attr_get_int(ip
, &xname
, value
, valuelenp
, flags
);
177 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
182 * Calculate how many blocks we need for the new attribute,
186 struct xfs_inode
*ip
,
191 struct xfs_mount
*mp
= ip
->i_mount
;
196 * Determine space new attribute will use, and if it would be
197 * "local" or "remote" (note: local != inline).
199 size
= xfs_attr_leaf_newentsize(namelen
, valuelen
,
200 mp
->m_sb
.sb_blocksize
, local
);
202 nblks
= XFS_DAENTER_SPACE_RES(mp
, XFS_ATTR_FORK
);
204 if (size
> (mp
->m_sb
.sb_blocksize
>> 1)) {
205 /* Double split possible */
210 * Out of line attribute, cannot double split, but
211 * make room for the attribute value itself.
213 uint dblocks
= XFS_B_TO_FSB(mp
, valuelen
);
215 nblks
+= XFS_NEXTENTADD_SPACE_RES(mp
, dblocks
, XFS_ATTR_FORK
);
223 struct xfs_inode
*dp
,
224 struct xfs_name
*name
,
225 unsigned char *value
,
230 xfs_fsblock_t firstblock
;
231 xfs_bmap_free_t flist
;
232 int error
, err2
, committed
;
233 struct xfs_mount
*mp
= dp
->i_mount
;
234 struct xfs_trans_res tres
;
235 int rsvd
= (flags
& ATTR_ROOT
) != 0;
239 * Attach the dquots to the inode.
241 error
= xfs_qm_dqattach(dp
, 0);
246 * If the inode doesn't have an attribute fork, add one.
247 * (inode must not be locked when we call this routine)
249 if (XFS_IFORK_Q(dp
) == 0) {
250 int sf_size
= sizeof(xfs_attr_sf_hdr_t
) +
251 XFS_ATTR_SF_ENTSIZE_BYNAME(name
->len
, valuelen
);
253 if ((error
= xfs_bmap_add_attrfork(dp
, sf_size
, rsvd
)))
258 * Fill in the arg structure for this request.
260 memset((char *)&args
, 0, sizeof(args
));
261 args
.name
= name
->name
;
262 args
.namelen
= name
->len
;
264 args
.valuelen
= valuelen
;
266 args
.hashval
= xfs_da_hashname(args
.name
, args
.namelen
);
268 args
.firstblock
= &firstblock
;
270 args
.whichfork
= XFS_ATTR_FORK
;
271 args
.op_flags
= XFS_DA_OP_ADDNAME
| XFS_DA_OP_OKNOENT
;
273 /* Size is now blocks for attribute data */
274 args
.total
= xfs_attr_calc_size(dp
, name
->len
, valuelen
, &local
);
277 * Start our first transaction of the day.
279 * All future transactions during this code must be "chained" off
280 * this one via the trans_dup() call. All transactions will contain
281 * the inode, and the inode will always be marked with trans_ihold().
282 * Since the inode will be locked in all transactions, we must log
283 * the inode in every transaction to let it float upward through
286 args
.trans
= xfs_trans_alloc(mp
, XFS_TRANS_ATTR_SET
);
289 * Root fork attributes can use reserved data blocks for this
290 * operation if necessary
294 args
.trans
->t_flags
|= XFS_TRANS_RESERVE
;
296 tres
.tr_logres
= M_RES(mp
)->tr_attrsetm
.tr_logres
+
297 M_RES(mp
)->tr_attrsetrt
.tr_logres
* args
.total
;
298 tres
.tr_logcount
= XFS_ATTRSET_LOG_COUNT
;
299 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
300 error
= xfs_trans_reserve(args
.trans
, &tres
, args
.total
, 0);
302 xfs_trans_cancel(args
.trans
, 0);
305 xfs_ilock(dp
, XFS_ILOCK_EXCL
);
307 error
= xfs_trans_reserve_quota_nblks(args
.trans
, dp
, args
.total
, 0,
308 rsvd
? XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
309 XFS_QMOPT_RES_REGBLKS
);
311 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
312 xfs_trans_cancel(args
.trans
, XFS_TRANS_RELEASE_LOG_RES
);
316 xfs_trans_ijoin(args
.trans
, dp
, 0);
319 * If the attribute list is non-existent or a shortform list,
320 * upgrade it to a single-leaf-block attribute list.
322 if ((dp
->i_d
.di_aformat
== XFS_DINODE_FMT_LOCAL
) ||
323 ((dp
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
) &&
324 (dp
->i_d
.di_anextents
== 0))) {
327 * Build initial attribute list (if required).
329 if (dp
->i_d
.di_aformat
== XFS_DINODE_FMT_EXTENTS
)
330 xfs_attr_shortform_create(&args
);
333 * Try to add the attr to the attribute list in
336 error
= xfs_attr_shortform_addname(&args
);
337 if (error
!= ENOSPC
) {
339 * Commit the shortform mods, and we're done.
340 * NOTE: this is also the error path (EEXIST, etc).
342 ASSERT(args
.trans
!= NULL
);
345 * If this is a synchronous mount, make sure that
346 * the transaction goes to disk before returning
349 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
350 xfs_trans_set_sync(args
.trans
);
353 if (!error
&& (flags
& ATTR_KERNOTIME
) == 0) {
354 xfs_trans_ichgtime(args
.trans
, dp
,
357 err2
= xfs_trans_commit(args
.trans
,
358 XFS_TRANS_RELEASE_LOG_RES
);
359 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
361 return(error
== 0 ? err2
: error
);
365 * It won't fit in the shortform, transform to a leaf block.
366 * GROT: another possible req'mt for a double-split btree op.
368 xfs_bmap_init(args
.flist
, args
.firstblock
);
369 error
= xfs_attr_shortform_to_leaf(&args
);
371 error
= xfs_bmap_finish(&args
.trans
, args
.flist
,
377 xfs_bmap_cancel(&flist
);
382 * bmap_finish() may have committed the last trans and started
383 * a new one. We need the inode to be in all transactions.
386 xfs_trans_ijoin(args
.trans
, dp
, 0);
389 * Commit the leaf transformation. We'll need another (linked)
390 * transaction to add the new attribute to the leaf.
393 error
= xfs_trans_roll(&args
.trans
, dp
);
399 if (xfs_bmap_one_block(dp
, XFS_ATTR_FORK
)) {
400 error
= xfs_attr_leaf_addname(&args
);
402 error
= xfs_attr_node_addname(&args
);
409 * If this is a synchronous mount, make sure that the
410 * transaction goes to disk before returning to the user.
412 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
413 xfs_trans_set_sync(args
.trans
);
416 if ((flags
& ATTR_KERNOTIME
) == 0)
417 xfs_trans_ichgtime(args
.trans
, dp
, XFS_ICHGTIME_CHG
);
420 * Commit the last in the sequence of transactions.
422 xfs_trans_log_inode(args
.trans
, dp
, XFS_ILOG_CORE
);
423 error
= xfs_trans_commit(args
.trans
, XFS_TRANS_RELEASE_LOG_RES
);
424 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
430 xfs_trans_cancel(args
.trans
,
431 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
432 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
439 const unsigned char *name
,
440 unsigned char *value
,
445 struct xfs_name xname
;
447 XFS_STATS_INC(xs_attr_set
);
449 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
452 error
= xfs_attr_name_to_xname(&xname
, name
);
456 return xfs_attr_set_int(dp
, &xname
, value
, valuelen
, flags
);
460 * Generic handler routine to remove a name from an attribute list.
461 * Transitions attribute list from Btree to shortform as necessary.
464 xfs_attr_remove_int(xfs_inode_t
*dp
, struct xfs_name
*name
, int flags
)
467 xfs_fsblock_t firstblock
;
468 xfs_bmap_free_t flist
;
470 xfs_mount_t
*mp
= dp
->i_mount
;
473 * Fill in the arg structure for this request.
475 memset((char *)&args
, 0, sizeof(args
));
476 args
.name
= name
->name
;
477 args
.namelen
= name
->len
;
479 args
.hashval
= xfs_da_hashname(args
.name
, args
.namelen
);
481 args
.firstblock
= &firstblock
;
484 args
.whichfork
= XFS_ATTR_FORK
;
487 * we have no control over the attribute names that userspace passes us
488 * to remove, so we have to allow the name lookup prior to attribute
491 args
.op_flags
= XFS_DA_OP_OKNOENT
;
494 * Attach the dquots to the inode.
496 error
= xfs_qm_dqattach(dp
, 0);
501 * Start our first transaction of the day.
503 * All future transactions during this code must be "chained" off
504 * this one via the trans_dup() call. All transactions will contain
505 * the inode, and the inode will always be marked with trans_ihold().
506 * Since the inode will be locked in all transactions, we must log
507 * the inode in every transaction to let it float upward through
510 args
.trans
= xfs_trans_alloc(mp
, XFS_TRANS_ATTR_RM
);
513 * Root fork attributes can use reserved data blocks for this
514 * operation if necessary
517 if (flags
& ATTR_ROOT
)
518 args
.trans
->t_flags
|= XFS_TRANS_RESERVE
;
520 error
= xfs_trans_reserve(args
.trans
, &M_RES(mp
)->tr_attrrm
,
521 XFS_ATTRRM_SPACE_RES(mp
), 0);
523 xfs_trans_cancel(args
.trans
, 0);
527 xfs_ilock(dp
, XFS_ILOCK_EXCL
);
529 * No need to make quota reservations here. We expect to release some
530 * blocks not allocate in the common case.
532 xfs_trans_ijoin(args
.trans
, dp
, 0);
535 * Decide on what work routines to call based on the inode size.
537 if (!xfs_inode_hasattr(dp
)) {
538 error
= XFS_ERROR(ENOATTR
);
541 if (dp
->i_d
.di_aformat
== XFS_DINODE_FMT_LOCAL
) {
542 ASSERT(dp
->i_afp
->if_flags
& XFS_IFINLINE
);
543 error
= xfs_attr_shortform_remove(&args
);
547 } else if (xfs_bmap_one_block(dp
, XFS_ATTR_FORK
)) {
548 error
= xfs_attr_leaf_removename(&args
);
550 error
= xfs_attr_node_removename(&args
);
557 * If this is a synchronous mount, make sure that the
558 * transaction goes to disk before returning to the user.
560 if (mp
->m_flags
& XFS_MOUNT_WSYNC
) {
561 xfs_trans_set_sync(args
.trans
);
564 if ((flags
& ATTR_KERNOTIME
) == 0)
565 xfs_trans_ichgtime(args
.trans
, dp
, XFS_ICHGTIME_CHG
);
568 * Commit the last in the sequence of transactions.
570 xfs_trans_log_inode(args
.trans
, dp
, XFS_ILOG_CORE
);
571 error
= xfs_trans_commit(args
.trans
, XFS_TRANS_RELEASE_LOG_RES
);
572 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
578 xfs_trans_cancel(args
.trans
,
579 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
580 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
587 const unsigned char *name
,
591 struct xfs_name xname
;
593 XFS_STATS_INC(xs_attr_remove
);
595 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
598 error
= xfs_attr_name_to_xname(&xname
, name
);
602 xfs_ilock(dp
, XFS_ILOCK_SHARED
);
603 if (!xfs_inode_hasattr(dp
)) {
604 xfs_iunlock(dp
, XFS_ILOCK_SHARED
);
605 return XFS_ERROR(ENOATTR
);
607 xfs_iunlock(dp
, XFS_ILOCK_SHARED
);
609 return xfs_attr_remove_int(dp
, &xname
, flags
);
613 /*========================================================================
614 * External routines when attribute list is inside the inode
615 *========================================================================*/
618 * Add a name to the shortform attribute list structure
619 * This is the external routine.
622 xfs_attr_shortform_addname(xfs_da_args_t
*args
)
624 int newsize
, forkoff
, retval
;
626 trace_xfs_attr_sf_addname(args
);
628 retval
= xfs_attr_shortform_lookup(args
);
629 if ((args
->flags
& ATTR_REPLACE
) && (retval
== ENOATTR
)) {
631 } else if (retval
== EEXIST
) {
632 if (args
->flags
& ATTR_CREATE
)
634 retval
= xfs_attr_shortform_remove(args
);
638 if (args
->namelen
>= XFS_ATTR_SF_ENTSIZE_MAX
||
639 args
->valuelen
>= XFS_ATTR_SF_ENTSIZE_MAX
)
640 return(XFS_ERROR(ENOSPC
));
642 newsize
= XFS_ATTR_SF_TOTSIZE(args
->dp
);
643 newsize
+= XFS_ATTR_SF_ENTSIZE_BYNAME(args
->namelen
, args
->valuelen
);
645 forkoff
= xfs_attr_shortform_bytesfit(args
->dp
, newsize
);
647 return(XFS_ERROR(ENOSPC
));
649 xfs_attr_shortform_add(args
, forkoff
);
654 /*========================================================================
655 * External routines when attribute list is one block
656 *========================================================================*/
659 * Add a name to the leaf attribute list structure
661 * This leaf block cannot have a "remote" value, we only call this routine
662 * if bmap_one_block() says there is only one block (ie: no remote blks).
665 xfs_attr_leaf_addname(xfs_da_args_t
*args
)
669 int retval
, error
, committed
, forkoff
;
671 trace_xfs_attr_leaf_addname(args
);
674 * Read the (only) block in the attribute list in.
678 error
= xfs_attr3_leaf_read(args
->trans
, args
->dp
, args
->blkno
, -1, &bp
);
683 * Look up the given attribute in the leaf block. Figure out if
684 * the given flags produce an error or call for an atomic rename.
686 retval
= xfs_attr3_leaf_lookup_int(bp
, args
);
687 if ((args
->flags
& ATTR_REPLACE
) && (retval
== ENOATTR
)) {
688 xfs_trans_brelse(args
->trans
, bp
);
690 } else if (retval
== EEXIST
) {
691 if (args
->flags
& ATTR_CREATE
) { /* pure create op */
692 xfs_trans_brelse(args
->trans
, bp
);
696 trace_xfs_attr_leaf_replace(args
);
698 args
->op_flags
|= XFS_DA_OP_RENAME
; /* an atomic rename */
699 args
->blkno2
= args
->blkno
; /* set 2nd entry info*/
700 args
->index2
= args
->index
;
701 args
->rmtblkno2
= args
->rmtblkno
;
702 args
->rmtblkcnt2
= args
->rmtblkcnt
;
706 * Add the attribute to the leaf block, transitioning to a Btree
709 retval
= xfs_attr3_leaf_add(bp
, args
);
710 if (retval
== ENOSPC
) {
712 * Promote the attribute list to the Btree format, then
713 * Commit that transaction so that the node_addname() call
714 * can manage its own transactions.
716 xfs_bmap_init(args
->flist
, args
->firstblock
);
717 error
= xfs_attr3_leaf_to_node(args
);
719 error
= xfs_bmap_finish(&args
->trans
, args
->flist
,
725 xfs_bmap_cancel(args
->flist
);
730 * bmap_finish() may have committed the last trans and started
731 * a new one. We need the inode to be in all transactions.
734 xfs_trans_ijoin(args
->trans
, dp
, 0);
737 * Commit the current trans (including the inode) and start
740 error
= xfs_trans_roll(&args
->trans
, dp
);
745 * Fob the whole rest of the problem off on the Btree code.
747 error
= xfs_attr_node_addname(args
);
752 * Commit the transaction that added the attr name so that
753 * later routines can manage their own transactions.
755 error
= xfs_trans_roll(&args
->trans
, dp
);
760 * If there was an out-of-line value, allocate the blocks we
761 * identified for its storage and copy the value. This is done
762 * after we create the attribute so that we don't overflow the
763 * maximum size of a transaction and/or hit a deadlock.
765 if (args
->rmtblkno
> 0) {
766 error
= xfs_attr_rmtval_set(args
);
772 * If this is an atomic rename operation, we must "flip" the
773 * incomplete flags on the "new" and "old" attribute/value pairs
774 * so that one disappears and one appears atomically. Then we
775 * must remove the "old" attribute/value pair.
777 if (args
->op_flags
& XFS_DA_OP_RENAME
) {
779 * In a separate transaction, set the incomplete flag on the
780 * "old" attr and clear the incomplete flag on the "new" attr.
782 error
= xfs_attr3_leaf_flipflags(args
);
787 * Dismantle the "old" attribute/value pair by removing
788 * a "remote" value (if it exists).
790 args
->index
= args
->index2
;
791 args
->blkno
= args
->blkno2
;
792 args
->rmtblkno
= args
->rmtblkno2
;
793 args
->rmtblkcnt
= args
->rmtblkcnt2
;
794 if (args
->rmtblkno
) {
795 error
= xfs_attr_rmtval_remove(args
);
801 * Read in the block containing the "old" attr, then
802 * remove the "old" attr from that block (neat, huh!)
804 error
= xfs_attr3_leaf_read(args
->trans
, args
->dp
, args
->blkno
,
809 xfs_attr3_leaf_remove(bp
, args
);
812 * If the result is small enough, shrink it all into the inode.
814 if ((forkoff
= xfs_attr_shortform_allfit(bp
, dp
))) {
815 xfs_bmap_init(args
->flist
, args
->firstblock
);
816 error
= xfs_attr3_leaf_to_shortform(bp
, args
, forkoff
);
817 /* bp is gone due to xfs_da_shrink_inode */
819 error
= xfs_bmap_finish(&args
->trans
,
826 xfs_bmap_cancel(args
->flist
);
831 * bmap_finish() may have committed the last trans
832 * and started a new one. We need the inode to be
833 * in all transactions.
836 xfs_trans_ijoin(args
->trans
, dp
, 0);
840 * Commit the remove and start the next trans in series.
842 error
= xfs_trans_roll(&args
->trans
, dp
);
844 } else if (args
->rmtblkno
> 0) {
846 * Added a "remote" value, just clear the incomplete flag.
848 error
= xfs_attr3_leaf_clearflag(args
);
854 * Remove a name from the leaf attribute list structure
856 * This leaf block cannot have a "remote" value, we only call this routine
857 * if bmap_one_block() says there is only one block (ie: no remote blks).
860 xfs_attr_leaf_removename(xfs_da_args_t
*args
)
864 int error
, committed
, forkoff
;
866 trace_xfs_attr_leaf_removename(args
);
869 * Remove the attribute.
873 error
= xfs_attr3_leaf_read(args
->trans
, args
->dp
, args
->blkno
, -1, &bp
);
877 error
= xfs_attr3_leaf_lookup_int(bp
, args
);
878 if (error
== ENOATTR
) {
879 xfs_trans_brelse(args
->trans
, bp
);
883 xfs_attr3_leaf_remove(bp
, args
);
886 * If the result is small enough, shrink it all into the inode.
888 if ((forkoff
= xfs_attr_shortform_allfit(bp
, dp
))) {
889 xfs_bmap_init(args
->flist
, args
->firstblock
);
890 error
= xfs_attr3_leaf_to_shortform(bp
, args
, forkoff
);
891 /* bp is gone due to xfs_da_shrink_inode */
893 error
= xfs_bmap_finish(&args
->trans
, args
->flist
,
899 xfs_bmap_cancel(args
->flist
);
904 * bmap_finish() may have committed the last trans and started
905 * a new one. We need the inode to be in all transactions.
908 xfs_trans_ijoin(args
->trans
, dp
, 0);
914 * Look up a name in a leaf attribute list structure.
916 * This leaf block cannot have a "remote" value, we only call this routine
917 * if bmap_one_block() says there is only one block (ie: no remote blks).
920 xfs_attr_leaf_get(xfs_da_args_t
*args
)
925 trace_xfs_attr_leaf_get(args
);
928 error
= xfs_attr3_leaf_read(args
->trans
, args
->dp
, args
->blkno
, -1, &bp
);
932 error
= xfs_attr3_leaf_lookup_int(bp
, args
);
933 if (error
!= EEXIST
) {
934 xfs_trans_brelse(args
->trans
, bp
);
937 error
= xfs_attr3_leaf_getvalue(bp
, args
);
938 xfs_trans_brelse(args
->trans
, bp
);
939 if (!error
&& (args
->rmtblkno
> 0) && !(args
->flags
& ATTR_KERNOVAL
)) {
940 error
= xfs_attr_rmtval_get(args
);
945 /*========================================================================
946 * External routines when attribute list size > XFS_LBSIZE(mp).
947 *========================================================================*/
950 * Add a name to a Btree-format attribute list.
952 * This will involve walking down the Btree, and may involve splitting
953 * leaf nodes and even splitting intermediate nodes up to and including
954 * the root node (a special case of an intermediate node).
956 * "Remote" attribute values confuse the issue and atomic rename operations
957 * add a whole extra layer of confusion on top of that.
960 xfs_attr_node_addname(xfs_da_args_t
*args
)
962 xfs_da_state_t
*state
;
963 xfs_da_state_blk_t
*blk
;
966 int committed
, retval
, error
;
968 trace_xfs_attr_node_addname(args
);
971 * Fill in bucket of arguments/results/context to carry around.
976 state
= xfs_da_state_alloc();
979 state
->blocksize
= state
->mp
->m_sb
.sb_blocksize
;
980 state
->node_ents
= state
->mp
->m_attr_node_ents
;
983 * Search to see if name already exists, and get back a pointer
984 * to where it should go.
986 error
= xfs_da3_node_lookup_int(state
, &retval
);
989 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
990 ASSERT(blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
991 if ((args
->flags
& ATTR_REPLACE
) && (retval
== ENOATTR
)) {
993 } else if (retval
== EEXIST
) {
994 if (args
->flags
& ATTR_CREATE
)
997 trace_xfs_attr_node_replace(args
);
999 args
->op_flags
|= XFS_DA_OP_RENAME
; /* atomic rename op */
1000 args
->blkno2
= args
->blkno
; /* set 2nd entry info*/
1001 args
->index2
= args
->index
;
1002 args
->rmtblkno2
= args
->rmtblkno
;
1003 args
->rmtblkcnt2
= args
->rmtblkcnt
;
1005 args
->rmtblkcnt
= 0;
1008 retval
= xfs_attr3_leaf_add(blk
->bp
, state
->args
);
1009 if (retval
== ENOSPC
) {
1010 if (state
->path
.active
== 1) {
1012 * Its really a single leaf node, but it had
1013 * out-of-line values so it looked like it *might*
1014 * have been a b-tree.
1016 xfs_da_state_free(state
);
1018 xfs_bmap_init(args
->flist
, args
->firstblock
);
1019 error
= xfs_attr3_leaf_to_node(args
);
1021 error
= xfs_bmap_finish(&args
->trans
,
1028 xfs_bmap_cancel(args
->flist
);
1033 * bmap_finish() may have committed the last trans
1034 * and started a new one. We need the inode to be
1035 * in all transactions.
1038 xfs_trans_ijoin(args
->trans
, dp
, 0);
1041 * Commit the node conversion and start the next
1042 * trans in the chain.
1044 error
= xfs_trans_roll(&args
->trans
, dp
);
1052 * Split as many Btree elements as required.
1053 * This code tracks the new and old attr's location
1054 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1055 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1057 xfs_bmap_init(args
->flist
, args
->firstblock
);
1058 error
= xfs_da3_split(state
);
1060 error
= xfs_bmap_finish(&args
->trans
, args
->flist
,
1066 xfs_bmap_cancel(args
->flist
);
1071 * bmap_finish() may have committed the last trans and started
1072 * a new one. We need the inode to be in all transactions.
1075 xfs_trans_ijoin(args
->trans
, dp
, 0);
1078 * Addition succeeded, update Btree hashvals.
1080 xfs_da3_fixhashpath(state
, &state
->path
);
1084 * Kill the state structure, we're done with it and need to
1085 * allow the buffers to come back later.
1087 xfs_da_state_free(state
);
1091 * Commit the leaf addition or btree split and start the next
1092 * trans in the chain.
1094 error
= xfs_trans_roll(&args
->trans
, dp
);
1099 * If there was an out-of-line value, allocate the blocks we
1100 * identified for its storage and copy the value. This is done
1101 * after we create the attribute so that we don't overflow the
1102 * maximum size of a transaction and/or hit a deadlock.
1104 if (args
->rmtblkno
> 0) {
1105 error
= xfs_attr_rmtval_set(args
);
1111 * If this is an atomic rename operation, we must "flip" the
1112 * incomplete flags on the "new" and "old" attribute/value pairs
1113 * so that one disappears and one appears atomically. Then we
1114 * must remove the "old" attribute/value pair.
1116 if (args
->op_flags
& XFS_DA_OP_RENAME
) {
1118 * In a separate transaction, set the incomplete flag on the
1119 * "old" attr and clear the incomplete flag on the "new" attr.
1121 error
= xfs_attr3_leaf_flipflags(args
);
1126 * Dismantle the "old" attribute/value pair by removing
1127 * a "remote" value (if it exists).
1129 args
->index
= args
->index2
;
1130 args
->blkno
= args
->blkno2
;
1131 args
->rmtblkno
= args
->rmtblkno2
;
1132 args
->rmtblkcnt
= args
->rmtblkcnt2
;
1133 if (args
->rmtblkno
) {
1134 error
= xfs_attr_rmtval_remove(args
);
1140 * Re-find the "old" attribute entry after any split ops.
1141 * The INCOMPLETE flag means that we will find the "old"
1142 * attr, not the "new" one.
1144 args
->flags
|= XFS_ATTR_INCOMPLETE
;
1145 state
= xfs_da_state_alloc();
1148 state
->blocksize
= state
->mp
->m_sb
.sb_blocksize
;
1149 state
->node_ents
= state
->mp
->m_attr_node_ents
;
1151 error
= xfs_da3_node_lookup_int(state
, &retval
);
1156 * Remove the name and update the hashvals in the tree.
1158 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1159 ASSERT(blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1160 error
= xfs_attr3_leaf_remove(blk
->bp
, args
);
1161 xfs_da3_fixhashpath(state
, &state
->path
);
1164 * Check to see if the tree needs to be collapsed.
1166 if (retval
&& (state
->path
.active
> 1)) {
1167 xfs_bmap_init(args
->flist
, args
->firstblock
);
1168 error
= xfs_da3_join(state
);
1170 error
= xfs_bmap_finish(&args
->trans
,
1177 xfs_bmap_cancel(args
->flist
);
1182 * bmap_finish() may have committed the last trans
1183 * and started a new one. We need the inode to be
1184 * in all transactions.
1187 xfs_trans_ijoin(args
->trans
, dp
, 0);
1191 * Commit and start the next trans in the chain.
1193 error
= xfs_trans_roll(&args
->trans
, dp
);
1197 } else if (args
->rmtblkno
> 0) {
1199 * Added a "remote" value, just clear the incomplete flag.
1201 error
= xfs_attr3_leaf_clearflag(args
);
1209 xfs_da_state_free(state
);
1216 * Remove a name from a B-tree attribute list.
1218 * This will involve walking down the Btree, and may involve joining
1219 * leaf nodes and even joining intermediate nodes up to and including
1220 * the root node (a special case of an intermediate node).
1223 xfs_attr_node_removename(xfs_da_args_t
*args
)
1225 xfs_da_state_t
*state
;
1226 xfs_da_state_blk_t
*blk
;
1229 int retval
, error
, committed
, forkoff
;
1231 trace_xfs_attr_node_removename(args
);
1234 * Tie a string around our finger to remind us where we are.
1237 state
= xfs_da_state_alloc();
1239 state
->mp
= dp
->i_mount
;
1240 state
->blocksize
= state
->mp
->m_sb
.sb_blocksize
;
1241 state
->node_ents
= state
->mp
->m_attr_node_ents
;
1244 * Search to see if name exists, and get back a pointer to it.
1246 error
= xfs_da3_node_lookup_int(state
, &retval
);
1247 if (error
|| (retval
!= EEXIST
)) {
1254 * If there is an out-of-line value, de-allocate the blocks.
1255 * This is done before we remove the attribute so that we don't
1256 * overflow the maximum size of a transaction and/or hit a deadlock.
1258 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1259 ASSERT(blk
->bp
!= NULL
);
1260 ASSERT(blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1261 if (args
->rmtblkno
> 0) {
1263 * Fill in disk block numbers in the state structure
1264 * so that we can get the buffers back after we commit
1265 * several transactions in the following calls.
1267 error
= xfs_attr_fillstate(state
);
1272 * Mark the attribute as INCOMPLETE, then bunmapi() the
1275 error
= xfs_attr3_leaf_setflag(args
);
1278 error
= xfs_attr_rmtval_remove(args
);
1283 * Refill the state structure with buffers, the prior calls
1284 * released our buffers.
1286 error
= xfs_attr_refillstate(state
);
1292 * Remove the name and update the hashvals in the tree.
1294 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1295 ASSERT(blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1296 retval
= xfs_attr3_leaf_remove(blk
->bp
, args
);
1297 xfs_da3_fixhashpath(state
, &state
->path
);
1300 * Check to see if the tree needs to be collapsed.
1302 if (retval
&& (state
->path
.active
> 1)) {
1303 xfs_bmap_init(args
->flist
, args
->firstblock
);
1304 error
= xfs_da3_join(state
);
1306 error
= xfs_bmap_finish(&args
->trans
, args
->flist
,
1312 xfs_bmap_cancel(args
->flist
);
1317 * bmap_finish() may have committed the last trans and started
1318 * a new one. We need the inode to be in all transactions.
1321 xfs_trans_ijoin(args
->trans
, dp
, 0);
1324 * Commit the Btree join operation and start a new trans.
1326 error
= xfs_trans_roll(&args
->trans
, dp
);
1332 * If the result is small enough, push it all into the inode.
1334 if (xfs_bmap_one_block(dp
, XFS_ATTR_FORK
)) {
1336 * Have to get rid of the copy of this dabuf in the state.
1338 ASSERT(state
->path
.active
== 1);
1339 ASSERT(state
->path
.blk
[0].bp
);
1340 state
->path
.blk
[0].bp
= NULL
;
1342 error
= xfs_attr3_leaf_read(args
->trans
, args
->dp
, 0, -1, &bp
);
1346 if ((forkoff
= xfs_attr_shortform_allfit(bp
, dp
))) {
1347 xfs_bmap_init(args
->flist
, args
->firstblock
);
1348 error
= xfs_attr3_leaf_to_shortform(bp
, args
, forkoff
);
1349 /* bp is gone due to xfs_da_shrink_inode */
1351 error
= xfs_bmap_finish(&args
->trans
,
1358 xfs_bmap_cancel(args
->flist
);
1363 * bmap_finish() may have committed the last trans
1364 * and started a new one. We need the inode to be
1365 * in all transactions.
1368 xfs_trans_ijoin(args
->trans
, dp
, 0);
1370 xfs_trans_brelse(args
->trans
, bp
);
1375 xfs_da_state_free(state
);
1380 * Fill in the disk block numbers in the state structure for the buffers
1381 * that are attached to the state structure.
1382 * This is done so that we can quickly reattach ourselves to those buffers
1383 * after some set of transaction commits have released these buffers.
1386 xfs_attr_fillstate(xfs_da_state_t
*state
)
1388 xfs_da_state_path_t
*path
;
1389 xfs_da_state_blk_t
*blk
;
1392 trace_xfs_attr_fillstate(state
->args
);
1395 * Roll down the "path" in the state structure, storing the on-disk
1396 * block number for those buffers in the "path".
1398 path
= &state
->path
;
1399 ASSERT((path
->active
>= 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1400 for (blk
= path
->blk
, level
= 0; level
< path
->active
; blk
++, level
++) {
1402 blk
->disk_blkno
= XFS_BUF_ADDR(blk
->bp
);
1405 blk
->disk_blkno
= 0;
1410 * Roll down the "altpath" in the state structure, storing the on-disk
1411 * block number for those buffers in the "altpath".
1413 path
= &state
->altpath
;
1414 ASSERT((path
->active
>= 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1415 for (blk
= path
->blk
, level
= 0; level
< path
->active
; blk
++, level
++) {
1417 blk
->disk_blkno
= XFS_BUF_ADDR(blk
->bp
);
1420 blk
->disk_blkno
= 0;
1428 * Reattach the buffers to the state structure based on the disk block
1429 * numbers stored in the state structure.
1430 * This is done after some set of transaction commits have released those
1431 * buffers from our grip.
1434 xfs_attr_refillstate(xfs_da_state_t
*state
)
1436 xfs_da_state_path_t
*path
;
1437 xfs_da_state_blk_t
*blk
;
1440 trace_xfs_attr_refillstate(state
->args
);
1443 * Roll down the "path" in the state structure, storing the on-disk
1444 * block number for those buffers in the "path".
1446 path
= &state
->path
;
1447 ASSERT((path
->active
>= 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1448 for (blk
= path
->blk
, level
= 0; level
< path
->active
; blk
++, level
++) {
1449 if (blk
->disk_blkno
) {
1450 error
= xfs_da3_node_read(state
->args
->trans
,
1452 blk
->blkno
, blk
->disk_blkno
,
1453 &blk
->bp
, XFS_ATTR_FORK
);
1462 * Roll down the "altpath" in the state structure, storing the on-disk
1463 * block number for those buffers in the "altpath".
1465 path
= &state
->altpath
;
1466 ASSERT((path
->active
>= 0) && (path
->active
< XFS_DA_NODE_MAXDEPTH
));
1467 for (blk
= path
->blk
, level
= 0; level
< path
->active
; blk
++, level
++) {
1468 if (blk
->disk_blkno
) {
1469 error
= xfs_da3_node_read(state
->args
->trans
,
1471 blk
->blkno
, blk
->disk_blkno
,
1472 &blk
->bp
, XFS_ATTR_FORK
);
1484 * Look up a filename in a node attribute list.
1486 * This routine gets called for any attribute fork that has more than one
1487 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1488 * "remote" values taking up more blocks.
1491 xfs_attr_node_get(xfs_da_args_t
*args
)
1493 xfs_da_state_t
*state
;
1494 xfs_da_state_blk_t
*blk
;
1498 trace_xfs_attr_node_get(args
);
1500 state
= xfs_da_state_alloc();
1502 state
->mp
= args
->dp
->i_mount
;
1503 state
->blocksize
= state
->mp
->m_sb
.sb_blocksize
;
1504 state
->node_ents
= state
->mp
->m_attr_node_ents
;
1507 * Search to see if name exists, and get back a pointer to it.
1509 error
= xfs_da3_node_lookup_int(state
, &retval
);
1512 } else if (retval
== EEXIST
) {
1513 blk
= &state
->path
.blk
[ state
->path
.active
-1 ];
1514 ASSERT(blk
->bp
!= NULL
);
1515 ASSERT(blk
->magic
== XFS_ATTR_LEAF_MAGIC
);
1518 * Get the value, local or "remote"
1520 retval
= xfs_attr3_leaf_getvalue(blk
->bp
, args
);
1521 if (!retval
&& (args
->rmtblkno
> 0)
1522 && !(args
->flags
& ATTR_KERNOVAL
)) {
1523 retval
= xfs_attr_rmtval_get(args
);
1528 * If not in a transaction, we have to release all the buffers.
1530 for (i
= 0; i
< state
->path
.active
; i
++) {
1531 xfs_trans_brelse(args
->trans
, state
->path
.blk
[i
].bp
);
1532 state
->path
.blk
[i
].bp
= NULL
;
1535 xfs_da_state_free(state
);