2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_bmap_btree.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_attr_remote.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_error.h"
40 #include "xfs_trace.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_cksum.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_attr3_leaf_create(struct xfs_da_args *args,
60 xfs_dablk_t which_block, struct xfs_buf **bpp);
61 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
62 struct xfs_attr3_icleaf_hdr *ichdr,
63 struct xfs_da_args *args, int freemap_index);
64 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
65 struct xfs_attr3_icleaf_hdr *ichdr,
66 struct xfs_buf *leaf_buffer);
67 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
68 xfs_da_state_blk_t *blk1,
69 xfs_da_state_blk_t *blk2);
70 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
71 xfs_da_state_blk_t *leaf_blk_1,
72 struct xfs_attr3_icleaf_hdr *ichdr1,
73 xfs_da_state_blk_t *leaf_blk_2,
74 struct xfs_attr3_icleaf_hdr *ichdr2,
75 int *number_entries_in_blk1,
76 int *number_usedbytes_in_blk1);
81 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
82 struct xfs_attr_leafblock *src_leaf,
83 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
84 struct xfs_attr_leafblock *dst_leaf,
85 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
90 * attr3 block 'firstused' conversion helpers.
92 * firstused refers to the offset of the first used byte of the nameval region
93 * of an attr leaf block. The region starts at the tail of the block and expands
94 * backwards towards the middle. As such, firstused is initialized to the block
95 * size for an empty leaf block and is reduced from there.
97 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
98 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
99 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
100 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
101 * the attr block size. The following helpers manage the conversion between the
102 * in-core and on-disk formats.
106 xfs_attr3_leaf_firstused_from_disk(
107 struct xfs_da_geometry *geo,
108 struct xfs_attr3_icleaf_hdr *to,
109 struct xfs_attr_leafblock *from)
111 struct xfs_attr3_leaf_hdr *hdr3;
113 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
114 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
115 to->firstused = be16_to_cpu(hdr3->firstused);
117 to->firstused = be16_to_cpu(from->hdr.firstused);
121 * Convert from the magic fsb size value to actual blocksize. This
122 * should only occur for empty blocks when the block size overflows
125 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
126 ASSERT(!to->count && !to->usedbytes);
127 ASSERT(geo->blksize > USHRT_MAX);
128 to->firstused = geo->blksize;
133 xfs_attr3_leaf_firstused_to_disk(
134 struct xfs_da_geometry *geo,
135 struct xfs_attr_leafblock *to,
136 struct xfs_attr3_icleaf_hdr *from)
138 struct xfs_attr3_leaf_hdr *hdr3;
141 /* magic value should only be seen on disk */
142 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
145 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
146 * value. This only overflows at the max supported value of 64k. Use the
147 * magic on-disk value to represent block size in this case.
149 firstused = from->firstused;
150 if (firstused > USHRT_MAX) {
151 ASSERT(from->firstused == geo->blksize);
152 firstused = XFS_ATTR3_LEAF_NULLOFF;
155 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
156 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
157 hdr3->firstused = cpu_to_be16(firstused);
159 to->hdr.firstused = cpu_to_be16(firstused);
164 xfs_attr3_leaf_hdr_from_disk(
165 struct xfs_da_geometry *geo,
166 struct xfs_attr3_icleaf_hdr *to,
167 struct xfs_attr_leafblock *from)
171 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
172 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
174 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
175 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
177 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
178 to->back = be32_to_cpu(hdr3->info.hdr.back);
179 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
180 to->count = be16_to_cpu(hdr3->count);
181 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
182 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
183 to->holes = hdr3->holes;
185 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
186 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
187 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
191 to->forw = be32_to_cpu(from->hdr.info.forw);
192 to->back = be32_to_cpu(from->hdr.info.back);
193 to->magic = be16_to_cpu(from->hdr.info.magic);
194 to->count = be16_to_cpu(from->hdr.count);
195 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
196 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
197 to->holes = from->hdr.holes;
199 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
200 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
201 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
206 xfs_attr3_leaf_hdr_to_disk(
207 struct xfs_da_geometry *geo,
208 struct xfs_attr_leafblock *to,
209 struct xfs_attr3_icleaf_hdr *from)
213 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
214 from->magic == XFS_ATTR3_LEAF_MAGIC);
216 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
217 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
219 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
220 hdr3->info.hdr.back = cpu_to_be32(from->back);
221 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
222 hdr3->count = cpu_to_be16(from->count);
223 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
224 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
225 hdr3->holes = from->holes;
228 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
229 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
230 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
234 to->hdr.info.forw = cpu_to_be32(from->forw);
235 to->hdr.info.back = cpu_to_be32(from->back);
236 to->hdr.info.magic = cpu_to_be16(from->magic);
237 to->hdr.count = cpu_to_be16(from->count);
238 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
239 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
240 to->hdr.holes = from->holes;
243 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
244 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
245 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
250 xfs_attr3_leaf_verify(
253 struct xfs_mount *mp = bp->b_target->bt_mount;
254 struct xfs_attr_leafblock *leaf = bp->b_addr;
255 struct xfs_attr3_icleaf_hdr ichdr;
257 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
259 if (xfs_sb_version_hascrc(&mp->m_sb)) {
260 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
262 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
265 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
267 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
270 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
273 if (ichdr.count == 0)
276 /* XXX: need to range check rest of attr header values */
277 /* XXX: hash order check? */
283 xfs_attr3_leaf_write_verify(
286 struct xfs_mount *mp = bp->b_target->bt_mount;
287 struct xfs_buf_log_item *bip = bp->b_fspriv;
288 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
290 if (!xfs_attr3_leaf_verify(bp)) {
291 xfs_buf_ioerror(bp, -EFSCORRUPTED);
292 xfs_verifier_error(bp);
296 if (!xfs_sb_version_hascrc(&mp->m_sb))
300 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
302 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
306 * leaf/node format detection on trees is sketchy, so a node read can be done on
307 * leaf level blocks when detection identifies the tree as a node format tree
308 * incorrectly. In this case, we need to swap the verifier to match the correct
309 * format of the block being read.
312 xfs_attr3_leaf_read_verify(
315 struct xfs_mount *mp = bp->b_target->bt_mount;
317 if (xfs_sb_version_hascrc(&mp->m_sb) &&
318 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
319 xfs_buf_ioerror(bp, -EFSBADCRC);
320 else if (!xfs_attr3_leaf_verify(bp))
321 xfs_buf_ioerror(bp, -EFSCORRUPTED);
324 xfs_verifier_error(bp);
327 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
328 .verify_read = xfs_attr3_leaf_read_verify,
329 .verify_write = xfs_attr3_leaf_write_verify,
334 struct xfs_trans *tp,
335 struct xfs_inode *dp,
337 xfs_daddr_t mappedbno,
338 struct xfs_buf **bpp)
342 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
343 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
345 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
349 /*========================================================================
350 * Namespace helper routines
351 *========================================================================*/
354 * If namespace bits don't match return 0.
355 * If all match then return 1.
358 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
360 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
364 /*========================================================================
365 * External routines when attribute fork size < XFS_LITINO(mp).
366 *========================================================================*/
369 * Query whether the requested number of additional bytes of extended
370 * attribute space will be able to fit inline.
372 * Returns zero if not, else the di_forkoff fork offset to be used in the
373 * literal area for attribute data once the new bytes have been added.
375 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
376 * special case for dev/uuid inodes, they have fixed size data forks.
379 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
382 int minforkoff; /* lower limit on valid forkoff locations */
383 int maxforkoff; /* upper limit on valid forkoff locations */
385 xfs_mount_t *mp = dp->i_mount;
388 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
390 switch (dp->i_d.di_format) {
391 case XFS_DINODE_FMT_DEV:
392 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
393 return (offset >= minforkoff) ? minforkoff : 0;
394 case XFS_DINODE_FMT_UUID:
395 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
396 return (offset >= minforkoff) ? minforkoff : 0;
400 * If the requested numbers of bytes is smaller or equal to the
401 * current attribute fork size we can always proceed.
403 * Note that if_bytes in the data fork might actually be larger than
404 * the current data fork size is due to delalloc extents. In that
405 * case either the extent count will go down when they are converted
406 * to real extents, or the delalloc conversion will take care of the
407 * literal area rebalancing.
409 if (bytes <= XFS_IFORK_ASIZE(dp))
410 return dp->i_d.di_forkoff;
413 * For attr2 we can try to move the forkoff if there is space in the
414 * literal area, but for the old format we are done if there is no
415 * space in the fixed attribute fork.
417 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
420 dsize = dp->i_df.if_bytes;
422 switch (dp->i_d.di_format) {
423 case XFS_DINODE_FMT_EXTENTS:
425 * If there is no attr fork and the data fork is extents,
426 * determine if creating the default attr fork will result
427 * in the extents form migrating to btree. If so, the
428 * minimum offset only needs to be the space required for
431 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
432 xfs_default_attroffset(dp))
433 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
435 case XFS_DINODE_FMT_BTREE:
437 * If we have a data btree then keep forkoff if we have one,
438 * otherwise we are adding a new attr, so then we set
439 * minforkoff to where the btree root can finish so we have
440 * plenty of room for attrs
442 if (dp->i_d.di_forkoff) {
443 if (offset < dp->i_d.di_forkoff)
445 return dp->i_d.di_forkoff;
447 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
452 * A data fork btree root must have space for at least
453 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
455 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
456 minforkoff = roundup(minforkoff, 8) >> 3;
458 /* attr fork btree root can have at least this many key/ptr pairs */
459 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
460 XFS_BMDR_SPACE_CALC(MINABTPTRS);
461 maxforkoff = maxforkoff >> 3; /* rounded down */
463 if (offset >= maxforkoff)
465 if (offset >= minforkoff)
471 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
474 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
476 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
477 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
478 spin_lock(&mp->m_sb_lock);
479 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
480 xfs_sb_version_addattr2(&mp->m_sb);
481 spin_unlock(&mp->m_sb_lock);
484 spin_unlock(&mp->m_sb_lock);
489 * Create the initial contents of a shortform attribute list.
492 xfs_attr_shortform_create(xfs_da_args_t *args)
494 xfs_attr_sf_hdr_t *hdr;
498 trace_xfs_attr_sf_create(args);
504 ASSERT(ifp->if_bytes == 0);
505 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
506 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
507 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
508 ifp->if_flags |= XFS_IFINLINE;
510 ASSERT(ifp->if_flags & XFS_IFINLINE);
512 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
513 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
515 hdr->totsize = cpu_to_be16(sizeof(*hdr));
516 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
520 * Add a name/value pair to the shortform attribute list.
521 * Overflow from the inode has already been checked for.
524 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
526 xfs_attr_shortform_t *sf;
527 xfs_attr_sf_entry_t *sfe;
533 trace_xfs_attr_sf_add(args);
537 dp->i_d.di_forkoff = forkoff;
540 ASSERT(ifp->if_flags & XFS_IFINLINE);
541 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
543 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
545 if (sfe->namelen != args->namelen)
547 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
549 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
555 offset = (char *)sfe - (char *)sf;
556 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
557 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
558 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
559 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
561 sfe->namelen = args->namelen;
562 sfe->valuelen = args->valuelen;
563 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
564 memcpy(sfe->nameval, args->name, args->namelen);
565 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
567 be16_add_cpu(&sf->hdr.totsize, size);
568 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
570 xfs_sbversion_add_attr2(mp, args->trans);
574 * After the last attribute is removed revert to original inode format,
575 * making all literal area available to the data fork once more.
578 xfs_attr_fork_remove(
579 struct xfs_inode *ip,
580 struct xfs_trans *tp)
582 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
583 ip->i_d.di_forkoff = 0;
584 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
586 ASSERT(ip->i_d.di_anextents == 0);
587 ASSERT(ip->i_afp == NULL);
589 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
593 * Remove an attribute from the shortform attribute list structure.
596 xfs_attr_shortform_remove(xfs_da_args_t *args)
598 xfs_attr_shortform_t *sf;
599 xfs_attr_sf_entry_t *sfe;
600 int base, size=0, end, totsize, i;
604 trace_xfs_attr_sf_remove(args);
608 base = sizeof(xfs_attr_sf_hdr_t);
609 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
612 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
614 size = XFS_ATTR_SF_ENTSIZE(sfe);
615 if (sfe->namelen != args->namelen)
617 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
619 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
627 * Fix up the attribute fork data, covering the hole
630 totsize = be16_to_cpu(sf->hdr.totsize);
632 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
634 be16_add_cpu(&sf->hdr.totsize, -size);
637 * Fix up the start offset of the attribute fork
640 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
641 (mp->m_flags & XFS_MOUNT_ATTR2) &&
642 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
643 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
644 xfs_attr_fork_remove(dp, args->trans);
646 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
647 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
648 ASSERT(dp->i_d.di_forkoff);
649 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
650 (args->op_flags & XFS_DA_OP_ADDNAME) ||
651 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
652 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
653 xfs_trans_log_inode(args->trans, dp,
654 XFS_ILOG_CORE | XFS_ILOG_ADATA);
657 xfs_sbversion_add_attr2(mp, args->trans);
663 * Look up a name in a shortform attribute list structure.
667 xfs_attr_shortform_lookup(xfs_da_args_t *args)
669 xfs_attr_shortform_t *sf;
670 xfs_attr_sf_entry_t *sfe;
674 trace_xfs_attr_sf_lookup(args);
676 ifp = args->dp->i_afp;
677 ASSERT(ifp->if_flags & XFS_IFINLINE);
678 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
680 for (i = 0; i < sf->hdr.count;
681 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
682 if (sfe->namelen != args->namelen)
684 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
686 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
694 * Look up a name in a shortform attribute list structure.
698 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
700 xfs_attr_shortform_t *sf;
701 xfs_attr_sf_entry_t *sfe;
704 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
705 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
707 for (i = 0; i < sf->hdr.count;
708 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
709 if (sfe->namelen != args->namelen)
711 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
713 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
715 if (args->flags & ATTR_KERNOVAL) {
716 args->valuelen = sfe->valuelen;
719 if (args->valuelen < sfe->valuelen) {
720 args->valuelen = sfe->valuelen;
723 args->valuelen = sfe->valuelen;
724 memcpy(args->value, &sfe->nameval[args->namelen],
732 * Convert from using the shortform to the leaf.
735 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
738 xfs_attr_shortform_t *sf;
739 xfs_attr_sf_entry_t *sfe;
747 trace_xfs_attr_sf_to_leaf(args);
751 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
752 size = be16_to_cpu(sf->hdr.totsize);
753 tmpbuffer = kmem_alloc(size, KM_SLEEP);
754 ASSERT(tmpbuffer != NULL);
755 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
756 sf = (xfs_attr_shortform_t *)tmpbuffer;
758 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
759 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
762 error = xfs_da_grow_inode(args, &blkno);
765 * If we hit an IO error middle of the transaction inside
766 * grow_inode(), we may have inconsistent data. Bail out.
770 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
771 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
776 error = xfs_attr3_leaf_create(args, blkno, &bp);
778 error = xfs_da_shrink_inode(args, 0, bp);
782 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
783 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
787 memset((char *)&nargs, 0, sizeof(nargs));
789 nargs.geo = args->geo;
790 nargs.firstblock = args->firstblock;
791 nargs.flist = args->flist;
792 nargs.total = args->total;
793 nargs.whichfork = XFS_ATTR_FORK;
794 nargs.trans = args->trans;
795 nargs.op_flags = XFS_DA_OP_OKNOENT;
798 for (i = 0; i < sf->hdr.count; i++) {
799 nargs.name = sfe->nameval;
800 nargs.namelen = sfe->namelen;
801 nargs.value = &sfe->nameval[nargs.namelen];
802 nargs.valuelen = sfe->valuelen;
803 nargs.hashval = xfs_da_hashname(sfe->nameval,
805 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
806 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
807 ASSERT(error == -ENOATTR);
808 error = xfs_attr3_leaf_add(bp, &nargs);
809 ASSERT(error != -ENOSPC);
812 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
817 kmem_free(tmpbuffer);
822 * Check a leaf attribute block to see if all the entries would fit into
823 * a shortform attribute list.
826 xfs_attr_shortform_allfit(
828 struct xfs_inode *dp)
830 struct xfs_attr_leafblock *leaf;
831 struct xfs_attr_leaf_entry *entry;
832 xfs_attr_leaf_name_local_t *name_loc;
833 struct xfs_attr3_icleaf_hdr leafhdr;
836 struct xfs_mount *mp = bp->b_target->bt_mount;
839 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
840 entry = xfs_attr3_leaf_entryp(leaf);
842 bytes = sizeof(struct xfs_attr_sf_hdr);
843 for (i = 0; i < leafhdr.count; entry++, i++) {
844 if (entry->flags & XFS_ATTR_INCOMPLETE)
845 continue; /* don't copy partial entries */
846 if (!(entry->flags & XFS_ATTR_LOCAL))
848 name_loc = xfs_attr3_leaf_name_local(leaf, i);
849 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
851 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
853 bytes += sizeof(struct xfs_attr_sf_entry) - 1
855 + be16_to_cpu(name_loc->valuelen);
857 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
858 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
859 (bytes == sizeof(struct xfs_attr_sf_hdr)))
861 return xfs_attr_shortform_bytesfit(dp, bytes);
865 * Convert a leaf attribute list to shortform attribute list
868 xfs_attr3_leaf_to_shortform(
870 struct xfs_da_args *args,
873 struct xfs_attr_leafblock *leaf;
874 struct xfs_attr3_icleaf_hdr ichdr;
875 struct xfs_attr_leaf_entry *entry;
876 struct xfs_attr_leaf_name_local *name_loc;
877 struct xfs_da_args nargs;
878 struct xfs_inode *dp = args->dp;
883 trace_xfs_attr_leaf_to_sf(args);
885 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
889 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
891 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
892 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
893 entry = xfs_attr3_leaf_entryp(leaf);
895 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
896 memset(bp->b_addr, 0, args->geo->blksize);
899 * Clean out the prior contents of the attribute list.
901 error = xfs_da_shrink_inode(args, 0, bp);
906 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
907 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
908 xfs_attr_fork_remove(dp, args->trans);
912 xfs_attr_shortform_create(args);
915 * Copy the attributes
917 memset((char *)&nargs, 0, sizeof(nargs));
918 nargs.geo = args->geo;
920 nargs.firstblock = args->firstblock;
921 nargs.flist = args->flist;
922 nargs.total = args->total;
923 nargs.whichfork = XFS_ATTR_FORK;
924 nargs.trans = args->trans;
925 nargs.op_flags = XFS_DA_OP_OKNOENT;
927 for (i = 0; i < ichdr.count; entry++, i++) {
928 if (entry->flags & XFS_ATTR_INCOMPLETE)
929 continue; /* don't copy partial entries */
932 ASSERT(entry->flags & XFS_ATTR_LOCAL);
933 name_loc = xfs_attr3_leaf_name_local(leaf, i);
934 nargs.name = name_loc->nameval;
935 nargs.namelen = name_loc->namelen;
936 nargs.value = &name_loc->nameval[nargs.namelen];
937 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
938 nargs.hashval = be32_to_cpu(entry->hashval);
939 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
940 xfs_attr_shortform_add(&nargs, forkoff);
945 kmem_free(tmpbuffer);
950 * Convert from using a single leaf to a root node and a leaf.
953 xfs_attr3_leaf_to_node(
954 struct xfs_da_args *args)
956 struct xfs_attr_leafblock *leaf;
957 struct xfs_attr3_icleaf_hdr icleafhdr;
958 struct xfs_attr_leaf_entry *entries;
959 struct xfs_da_node_entry *btree;
960 struct xfs_da3_icnode_hdr icnodehdr;
961 struct xfs_da_intnode *node;
962 struct xfs_inode *dp = args->dp;
963 struct xfs_mount *mp = dp->i_mount;
964 struct xfs_buf *bp1 = NULL;
965 struct xfs_buf *bp2 = NULL;
969 trace_xfs_attr_leaf_to_node(args);
971 error = xfs_da_grow_inode(args, &blkno);
974 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
978 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
982 /* copy leaf to new buffer, update identifiers */
983 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
984 bp2->b_ops = bp1->b_ops;
985 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
986 if (xfs_sb_version_hascrc(&mp->m_sb)) {
987 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
988 hdr3->blkno = cpu_to_be64(bp2->b_bn);
990 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
993 * Set up the new root node.
995 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
999 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1000 btree = dp->d_ops->node_tree_p(node);
1003 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1004 entries = xfs_attr3_leaf_entryp(leaf);
1006 /* both on-disk, don't endian-flip twice */
1007 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1008 btree[0].before = cpu_to_be32(blkno);
1009 icnodehdr.count = 1;
1010 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1011 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1017 /*========================================================================
1018 * Routines used for growing the Btree.
1019 *========================================================================*/
1022 * Create the initial contents of a leaf attribute list
1023 * or a leaf in a node attribute list.
1026 xfs_attr3_leaf_create(
1027 struct xfs_da_args *args,
1029 struct xfs_buf **bpp)
1031 struct xfs_attr_leafblock *leaf;
1032 struct xfs_attr3_icleaf_hdr ichdr;
1033 struct xfs_inode *dp = args->dp;
1034 struct xfs_mount *mp = dp->i_mount;
1038 trace_xfs_attr_leaf_create(args);
1040 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1044 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1045 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1047 memset(leaf, 0, args->geo->blksize);
1049 memset(&ichdr, 0, sizeof(ichdr));
1050 ichdr.firstused = args->geo->blksize;
1052 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1053 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1055 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1057 hdr3->blkno = cpu_to_be64(bp->b_bn);
1058 hdr3->owner = cpu_to_be64(dp->i_ino);
1059 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
1061 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1063 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1064 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1066 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1068 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1069 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1076 * Split the leaf node, rebalance, then add the new entry.
1079 xfs_attr3_leaf_split(
1080 struct xfs_da_state *state,
1081 struct xfs_da_state_blk *oldblk,
1082 struct xfs_da_state_blk *newblk)
1087 trace_xfs_attr_leaf_split(state->args);
1090 * Allocate space for a new leaf node.
1092 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1093 error = xfs_da_grow_inode(state->args, &blkno);
1096 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1099 newblk->blkno = blkno;
1100 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1103 * Rebalance the entries across the two leaves.
1104 * NOTE: rebalance() currently depends on the 2nd block being empty.
1106 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1107 error = xfs_da3_blk_link(state, oldblk, newblk);
1112 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1113 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1114 * "new" attrs info. Will need the "old" info to remove it later.
1116 * Insert the "new" entry in the correct block.
1118 if (state->inleaf) {
1119 trace_xfs_attr_leaf_add_old(state->args);
1120 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1122 trace_xfs_attr_leaf_add_new(state->args);
1123 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1127 * Update last hashval in each block since we added the name.
1129 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1130 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1135 * Add a name to the leaf attribute list structure.
1140 struct xfs_da_args *args)
1142 struct xfs_attr_leafblock *leaf;
1143 struct xfs_attr3_icleaf_hdr ichdr;
1150 trace_xfs_attr_leaf_add(args);
1153 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1154 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1155 entsize = xfs_attr_leaf_newentsize(args, NULL);
1158 * Search through freemap for first-fit on new name length.
1159 * (may need to figure in size of entry struct too)
1161 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1162 + xfs_attr3_leaf_hdr_size(leaf);
1163 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1164 if (tablesize > ichdr.firstused) {
1165 sum += ichdr.freemap[i].size;
1168 if (!ichdr.freemap[i].size)
1169 continue; /* no space in this map */
1171 if (ichdr.freemap[i].base < ichdr.firstused)
1172 tmp += sizeof(xfs_attr_leaf_entry_t);
1173 if (ichdr.freemap[i].size >= tmp) {
1174 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1177 sum += ichdr.freemap[i].size;
1181 * If there are no holes in the address space of the block,
1182 * and we don't have enough freespace, then compaction will do us
1183 * no good and we should just give up.
1185 if (!ichdr.holes && sum < entsize)
1189 * Compact the entries to coalesce free space.
1190 * This may change the hdr->count via dropping INCOMPLETE entries.
1192 xfs_attr3_leaf_compact(args, &ichdr, bp);
1195 * After compaction, the block is guaranteed to have only one
1196 * free region, in freemap[0]. If it is not big enough, give up.
1198 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1203 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1206 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1207 xfs_trans_log_buf(args->trans, bp,
1208 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1209 xfs_attr3_leaf_hdr_size(leaf)));
1214 * Add a name to a leaf attribute list structure.
1217 xfs_attr3_leaf_add_work(
1219 struct xfs_attr3_icleaf_hdr *ichdr,
1220 struct xfs_da_args *args,
1223 struct xfs_attr_leafblock *leaf;
1224 struct xfs_attr_leaf_entry *entry;
1225 struct xfs_attr_leaf_name_local *name_loc;
1226 struct xfs_attr_leaf_name_remote *name_rmt;
1227 struct xfs_mount *mp;
1231 trace_xfs_attr_leaf_add_work(args);
1234 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1235 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1238 * Force open some space in the entry array and fill it in.
1240 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1241 if (args->index < ichdr->count) {
1242 tmp = ichdr->count - args->index;
1243 tmp *= sizeof(xfs_attr_leaf_entry_t);
1244 memmove(entry + 1, entry, tmp);
1245 xfs_trans_log_buf(args->trans, bp,
1246 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1251 * Allocate space for the new string (at the end of the run).
1253 mp = args->trans->t_mountp;
1254 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1255 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1256 ASSERT(ichdr->freemap[mapindex].size >=
1257 xfs_attr_leaf_newentsize(args, NULL));
1258 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1259 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1261 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1263 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1264 ichdr->freemap[mapindex].size);
1265 entry->hashval = cpu_to_be32(args->hashval);
1266 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1267 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1268 if (args->op_flags & XFS_DA_OP_RENAME) {
1269 entry->flags |= XFS_ATTR_INCOMPLETE;
1270 if ((args->blkno2 == args->blkno) &&
1271 (args->index2 <= args->index)) {
1275 xfs_trans_log_buf(args->trans, bp,
1276 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1277 ASSERT((args->index == 0) ||
1278 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1279 ASSERT((args->index == ichdr->count - 1) ||
1280 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1283 * For "remote" attribute values, simply note that we need to
1284 * allocate space for the "remote" value. We can't actually
1285 * allocate the extents in this transaction, and we can't decide
1286 * which blocks they should be as we might allocate more blocks
1287 * as part of this transaction (a split operation for example).
1289 if (entry->flags & XFS_ATTR_LOCAL) {
1290 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1291 name_loc->namelen = args->namelen;
1292 name_loc->valuelen = cpu_to_be16(args->valuelen);
1293 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1294 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1295 be16_to_cpu(name_loc->valuelen));
1297 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1298 name_rmt->namelen = args->namelen;
1299 memcpy((char *)name_rmt->name, args->name, args->namelen);
1300 entry->flags |= XFS_ATTR_INCOMPLETE;
1302 name_rmt->valuelen = 0;
1303 name_rmt->valueblk = 0;
1305 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1306 args->rmtvaluelen = args->valuelen;
1308 xfs_trans_log_buf(args->trans, bp,
1309 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1310 xfs_attr_leaf_entsize(leaf, args->index)));
1313 * Update the control info for this leaf node
1315 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1316 ichdr->firstused = be16_to_cpu(entry->nameidx);
1318 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1319 + xfs_attr3_leaf_hdr_size(leaf));
1320 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1321 + xfs_attr3_leaf_hdr_size(leaf);
1323 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1324 if (ichdr->freemap[i].base == tmp) {
1325 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1326 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1329 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1334 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1337 xfs_attr3_leaf_compact(
1338 struct xfs_da_args *args,
1339 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1342 struct xfs_attr_leafblock *leaf_src;
1343 struct xfs_attr_leafblock *leaf_dst;
1344 struct xfs_attr3_icleaf_hdr ichdr_src;
1345 struct xfs_trans *trans = args->trans;
1348 trace_xfs_attr_leaf_compact(args);
1350 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1351 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1352 memset(bp->b_addr, 0, args->geo->blksize);
1353 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1354 leaf_dst = bp->b_addr;
1357 * Copy the on-disk header back into the destination buffer to ensure
1358 * all the information in the header that is not part of the incore
1359 * header structure is preserved.
1361 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1363 /* Initialise the incore headers */
1364 ichdr_src = *ichdr_dst; /* struct copy */
1365 ichdr_dst->firstused = args->geo->blksize;
1366 ichdr_dst->usedbytes = 0;
1367 ichdr_dst->count = 0;
1368 ichdr_dst->holes = 0;
1369 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1370 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1371 ichdr_dst->freemap[0].base;
1373 /* write the header back to initialise the underlying buffer */
1374 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1377 * Copy all entry's in the same (sorted) order,
1378 * but allocate name/value pairs packed and in sequence.
1380 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1381 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1383 * this logs the entire buffer, but the caller must write the header
1384 * back to the buffer when it is finished modifying it.
1386 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1388 kmem_free(tmpbuffer);
1392 * Compare two leaf blocks "order".
1393 * Return 0 unless leaf2 should go before leaf1.
1396 xfs_attr3_leaf_order(
1397 struct xfs_buf *leaf1_bp,
1398 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1399 struct xfs_buf *leaf2_bp,
1400 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1402 struct xfs_attr_leaf_entry *entries1;
1403 struct xfs_attr_leaf_entry *entries2;
1405 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1406 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1407 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1408 ((be32_to_cpu(entries2[0].hashval) <
1409 be32_to_cpu(entries1[0].hashval)) ||
1410 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1411 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1418 xfs_attr_leaf_order(
1419 struct xfs_buf *leaf1_bp,
1420 struct xfs_buf *leaf2_bp)
1422 struct xfs_attr3_icleaf_hdr ichdr1;
1423 struct xfs_attr3_icleaf_hdr ichdr2;
1424 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1426 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1427 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1428 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1432 * Redistribute the attribute list entries between two leaf nodes,
1433 * taking into account the size of the new entry.
1435 * NOTE: if new block is empty, then it will get the upper half of the
1436 * old block. At present, all (one) callers pass in an empty second block.
1438 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1439 * to match what it is doing in splitting the attribute leaf block. Those
1440 * values are used in "atomic rename" operations on attributes. Note that
1441 * the "new" and "old" values can end up in different blocks.
1444 xfs_attr3_leaf_rebalance(
1445 struct xfs_da_state *state,
1446 struct xfs_da_state_blk *blk1,
1447 struct xfs_da_state_blk *blk2)
1449 struct xfs_da_args *args;
1450 struct xfs_attr_leafblock *leaf1;
1451 struct xfs_attr_leafblock *leaf2;
1452 struct xfs_attr3_icleaf_hdr ichdr1;
1453 struct xfs_attr3_icleaf_hdr ichdr2;
1454 struct xfs_attr_leaf_entry *entries1;
1455 struct xfs_attr_leaf_entry *entries2;
1463 * Set up environment.
1465 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1466 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1467 leaf1 = blk1->bp->b_addr;
1468 leaf2 = blk2->bp->b_addr;
1469 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1470 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1471 ASSERT(ichdr2.count == 0);
1474 trace_xfs_attr_leaf_rebalance(args);
1477 * Check ordering of blocks, reverse if it makes things simpler.
1479 * NOTE: Given that all (current) callers pass in an empty
1480 * second block, this code should never set "swap".
1483 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1484 struct xfs_da_state_blk *tmp_blk;
1485 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1491 /* struct copies to swap them rather than reconverting */
1496 leaf1 = blk1->bp->b_addr;
1497 leaf2 = blk2->bp->b_addr;
1502 * Examine entries until we reduce the absolute difference in
1503 * byte usage between the two blocks to a minimum. Then get
1504 * the direction to copy and the number of elements to move.
1506 * "inleaf" is true if the new entry should be inserted into blk1.
1507 * If "swap" is also true, then reverse the sense of "inleaf".
1509 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1513 state->inleaf = !state->inleaf;
1516 * Move any entries required from leaf to leaf:
1518 if (count < ichdr1.count) {
1520 * Figure the total bytes to be added to the destination leaf.
1522 /* number entries being moved */
1523 count = ichdr1.count - count;
1524 space = ichdr1.usedbytes - totallen;
1525 space += count * sizeof(xfs_attr_leaf_entry_t);
1528 * leaf2 is the destination, compact it if it looks tight.
1530 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1531 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1533 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1536 * Move high entries from leaf1 to low end of leaf2.
1538 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1539 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1541 } else if (count > ichdr1.count) {
1543 * I assert that since all callers pass in an empty
1544 * second buffer, this code should never execute.
1549 * Figure the total bytes to be added to the destination leaf.
1551 /* number entries being moved */
1552 count -= ichdr1.count;
1553 space = totallen - ichdr1.usedbytes;
1554 space += count * sizeof(xfs_attr_leaf_entry_t);
1557 * leaf1 is the destination, compact it if it looks tight.
1559 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1560 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1562 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1565 * Move low entries from leaf2 to high end of leaf1.
1567 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1568 ichdr1.count, count);
1571 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1572 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1573 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1574 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1577 * Copy out last hashval in each block for B-tree code.
1579 entries1 = xfs_attr3_leaf_entryp(leaf1);
1580 entries2 = xfs_attr3_leaf_entryp(leaf2);
1581 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1582 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1585 * Adjust the expected index for insertion.
1586 * NOTE: this code depends on the (current) situation that the
1587 * second block was originally empty.
1589 * If the insertion point moved to the 2nd block, we must adjust
1590 * the index. We must also track the entry just following the
1591 * new entry for use in an "atomic rename" operation, that entry
1592 * is always the "old" entry and the "new" entry is what we are
1593 * inserting. The index/blkno fields refer to the "old" entry,
1594 * while the index2/blkno2 fields refer to the "new" entry.
1596 if (blk1->index > ichdr1.count) {
1597 ASSERT(state->inleaf == 0);
1598 blk2->index = blk1->index - ichdr1.count;
1599 args->index = args->index2 = blk2->index;
1600 args->blkno = args->blkno2 = blk2->blkno;
1601 } else if (blk1->index == ichdr1.count) {
1602 if (state->inleaf) {
1603 args->index = blk1->index;
1604 args->blkno = blk1->blkno;
1606 args->blkno2 = blk2->blkno;
1609 * On a double leaf split, the original attr location
1610 * is already stored in blkno2/index2, so don't
1611 * overwrite it overwise we corrupt the tree.
1613 blk2->index = blk1->index - ichdr1.count;
1614 args->index = blk2->index;
1615 args->blkno = blk2->blkno;
1616 if (!state->extravalid) {
1618 * set the new attr location to match the old
1619 * one and let the higher level split code
1620 * decide where in the leaf to place it.
1622 args->index2 = blk2->index;
1623 args->blkno2 = blk2->blkno;
1627 ASSERT(state->inleaf == 1);
1628 args->index = args->index2 = blk1->index;
1629 args->blkno = args->blkno2 = blk1->blkno;
1634 * Examine entries until we reduce the absolute difference in
1635 * byte usage between the two blocks to a minimum.
1636 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1637 * GROT: there will always be enough room in either block for a new entry.
1638 * GROT: Do a double-split for this case?
1641 xfs_attr3_leaf_figure_balance(
1642 struct xfs_da_state *state,
1643 struct xfs_da_state_blk *blk1,
1644 struct xfs_attr3_icleaf_hdr *ichdr1,
1645 struct xfs_da_state_blk *blk2,
1646 struct xfs_attr3_icleaf_hdr *ichdr2,
1650 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1651 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1652 struct xfs_attr_leaf_entry *entry;
1663 * Examine entries until we reduce the absolute difference in
1664 * byte usage between the two blocks to a minimum.
1666 max = ichdr1->count + ichdr2->count;
1667 half = (max + 1) * sizeof(*entry);
1668 half += ichdr1->usedbytes + ichdr2->usedbytes +
1669 xfs_attr_leaf_newentsize(state->args, NULL);
1671 lastdelta = state->args->geo->blksize;
1672 entry = xfs_attr3_leaf_entryp(leaf1);
1673 for (count = index = 0; count < max; entry++, index++, count++) {
1675 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1677 * The new entry is in the first block, account for it.
1679 if (count == blk1->index) {
1680 tmp = totallen + sizeof(*entry) +
1681 xfs_attr_leaf_newentsize(state->args, NULL);
1682 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1684 lastdelta = XFS_ATTR_ABS(half - tmp);
1690 * Wrap around into the second block if necessary.
1692 if (count == ichdr1->count) {
1694 entry = xfs_attr3_leaf_entryp(leaf1);
1699 * Figure out if next leaf entry would be too much.
1701 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1703 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1705 lastdelta = XFS_ATTR_ABS(half - tmp);
1711 * Calculate the number of usedbytes that will end up in lower block.
1712 * If new entry not in lower block, fix up the count.
1714 totallen -= count * sizeof(*entry);
1716 totallen -= sizeof(*entry) +
1717 xfs_attr_leaf_newentsize(state->args, NULL);
1721 *usedbytesarg = totallen;
1725 /*========================================================================
1726 * Routines used for shrinking the Btree.
1727 *========================================================================*/
1730 * Check a leaf block and its neighbors to see if the block should be
1731 * collapsed into one or the other neighbor. Always keep the block
1732 * with the smaller block number.
1733 * If the current block is over 50% full, don't try to join it, return 0.
1734 * If the block is empty, fill in the state structure and return 2.
1735 * If it can be collapsed, fill in the state structure and return 1.
1736 * If nothing can be done, return 0.
1738 * GROT: allow for INCOMPLETE entries in calculation.
1741 xfs_attr3_leaf_toosmall(
1742 struct xfs_da_state *state,
1745 struct xfs_attr_leafblock *leaf;
1746 struct xfs_da_state_blk *blk;
1747 struct xfs_attr3_icleaf_hdr ichdr;
1756 trace_xfs_attr_leaf_toosmall(state->args);
1759 * Check for the degenerate case of the block being over 50% full.
1760 * If so, it's not worth even looking to see if we might be able
1761 * to coalesce with a sibling.
1763 blk = &state->path.blk[ state->path.active-1 ];
1764 leaf = blk->bp->b_addr;
1765 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1766 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1767 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1769 if (bytes > (state->args->geo->blksize >> 1)) {
1770 *action = 0; /* blk over 50%, don't try to join */
1775 * Check for the degenerate case of the block being empty.
1776 * If the block is empty, we'll simply delete it, no need to
1777 * coalesce it with a sibling block. We choose (arbitrarily)
1778 * to merge with the forward block unless it is NULL.
1780 if (ichdr.count == 0) {
1782 * Make altpath point to the block we want to keep and
1783 * path point to the block we want to drop (this one).
1785 forward = (ichdr.forw != 0);
1786 memcpy(&state->altpath, &state->path, sizeof(state->path));
1787 error = xfs_da3_path_shift(state, &state->altpath, forward,
1800 * Examine each sibling block to see if we can coalesce with
1801 * at least 25% free space to spare. We need to figure out
1802 * whether to merge with the forward or the backward block.
1803 * We prefer coalescing with the lower numbered sibling so as
1804 * to shrink an attribute list over time.
1806 /* start with smaller blk num */
1807 forward = ichdr.forw < ichdr.back;
1808 for (i = 0; i < 2; forward = !forward, i++) {
1809 struct xfs_attr3_icleaf_hdr ichdr2;
1816 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1821 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1823 bytes = state->args->geo->blksize -
1824 (state->args->geo->blksize >> 2) -
1825 ichdr.usedbytes - ichdr2.usedbytes -
1826 ((ichdr.count + ichdr2.count) *
1827 sizeof(xfs_attr_leaf_entry_t)) -
1828 xfs_attr3_leaf_hdr_size(leaf);
1830 xfs_trans_brelse(state->args->trans, bp);
1832 break; /* fits with at least 25% to spare */
1840 * Make altpath point to the block we want to keep (the lower
1841 * numbered block) and path point to the block we want to drop.
1843 memcpy(&state->altpath, &state->path, sizeof(state->path));
1844 if (blkno < blk->blkno) {
1845 error = xfs_da3_path_shift(state, &state->altpath, forward,
1848 error = xfs_da3_path_shift(state, &state->path, forward,
1862 * Remove a name from the leaf attribute list structure.
1864 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1865 * If two leaves are 37% full, when combined they will leave 25% free.
1868 xfs_attr3_leaf_remove(
1870 struct xfs_da_args *args)
1872 struct xfs_attr_leafblock *leaf;
1873 struct xfs_attr3_icleaf_hdr ichdr;
1874 struct xfs_attr_leaf_entry *entry;
1883 trace_xfs_attr_leaf_remove(args);
1886 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1888 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1889 ASSERT(args->index >= 0 && args->index < ichdr.count);
1890 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1891 xfs_attr3_leaf_hdr_size(leaf));
1893 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1895 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1896 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1899 * Scan through free region table:
1900 * check for adjacency of free'd entry with an existing one,
1901 * find smallest free region in case we need to replace it,
1902 * adjust any map that borders the entry table,
1904 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1905 + xfs_attr3_leaf_hdr_size(leaf);
1906 tmp = ichdr.freemap[0].size;
1907 before = after = -1;
1908 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1909 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1910 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1911 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1912 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1913 if (ichdr.freemap[i].base == tablesize) {
1914 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1915 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1918 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1919 be16_to_cpu(entry->nameidx)) {
1921 } else if (ichdr.freemap[i].base ==
1922 (be16_to_cpu(entry->nameidx) + entsize)) {
1924 } else if (ichdr.freemap[i].size < tmp) {
1925 tmp = ichdr.freemap[i].size;
1931 * Coalesce adjacent freemap regions,
1932 * or replace the smallest region.
1934 if ((before >= 0) || (after >= 0)) {
1935 if ((before >= 0) && (after >= 0)) {
1936 ichdr.freemap[before].size += entsize;
1937 ichdr.freemap[before].size += ichdr.freemap[after].size;
1938 ichdr.freemap[after].base = 0;
1939 ichdr.freemap[after].size = 0;
1940 } else if (before >= 0) {
1941 ichdr.freemap[before].size += entsize;
1943 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1944 ichdr.freemap[after].size += entsize;
1948 * Replace smallest region (if it is smaller than free'd entry)
1950 if (ichdr.freemap[smallest].size < entsize) {
1951 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1952 ichdr.freemap[smallest].size = entsize;
1957 * Did we remove the first entry?
1959 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1965 * Compress the remaining entries and zero out the removed stuff.
1967 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1968 ichdr.usedbytes -= entsize;
1969 xfs_trans_log_buf(args->trans, bp,
1970 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1973 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1974 memmove(entry, entry + 1, tmp);
1976 xfs_trans_log_buf(args->trans, bp,
1977 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1979 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1980 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1983 * If we removed the first entry, re-find the first used byte
1984 * in the name area. Note that if the entry was the "firstused",
1985 * then we don't have a "hole" in our block resulting from
1986 * removing the name.
1989 tmp = args->geo->blksize;
1990 entry = xfs_attr3_leaf_entryp(leaf);
1991 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1992 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1993 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1995 if (be16_to_cpu(entry->nameidx) < tmp)
1996 tmp = be16_to_cpu(entry->nameidx);
1998 ichdr.firstused = tmp;
1999 ASSERT(ichdr.firstused != 0);
2001 ichdr.holes = 1; /* mark as needing compaction */
2003 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2004 xfs_trans_log_buf(args->trans, bp,
2005 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2006 xfs_attr3_leaf_hdr_size(leaf)));
2009 * Check if leaf is less than 50% full, caller may want to
2010 * "join" the leaf with a sibling if so.
2012 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2013 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2015 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2019 * Move all the attribute list entries from drop_leaf into save_leaf.
2022 xfs_attr3_leaf_unbalance(
2023 struct xfs_da_state *state,
2024 struct xfs_da_state_blk *drop_blk,
2025 struct xfs_da_state_blk *save_blk)
2027 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2028 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2029 struct xfs_attr3_icleaf_hdr drophdr;
2030 struct xfs_attr3_icleaf_hdr savehdr;
2031 struct xfs_attr_leaf_entry *entry;
2033 trace_xfs_attr_leaf_unbalance(state->args);
2035 drop_leaf = drop_blk->bp->b_addr;
2036 save_leaf = save_blk->bp->b_addr;
2037 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2038 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2039 entry = xfs_attr3_leaf_entryp(drop_leaf);
2042 * Save last hashval from dying block for later Btree fixup.
2044 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2047 * Check if we need a temp buffer, or can we do it in place.
2048 * Note that we don't check "leaf" for holes because we will
2049 * always be dropping it, toosmall() decided that for us already.
2051 if (savehdr.holes == 0) {
2053 * dest leaf has no holes, so we add there. May need
2054 * to make some room in the entry array.
2056 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2057 drop_blk->bp, &drophdr)) {
2058 xfs_attr3_leaf_moveents(state->args,
2059 drop_leaf, &drophdr, 0,
2060 save_leaf, &savehdr, 0,
2063 xfs_attr3_leaf_moveents(state->args,
2064 drop_leaf, &drophdr, 0,
2065 save_leaf, &savehdr,
2066 savehdr.count, drophdr.count);
2070 * Destination has holes, so we make a temporary copy
2071 * of the leaf and add them both to that.
2073 struct xfs_attr_leafblock *tmp_leaf;
2074 struct xfs_attr3_icleaf_hdr tmphdr;
2076 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2079 * Copy the header into the temp leaf so that all the stuff
2080 * not in the incore header is present and gets copied back in
2081 * once we've moved all the entries.
2083 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2085 memset(&tmphdr, 0, sizeof(tmphdr));
2086 tmphdr.magic = savehdr.magic;
2087 tmphdr.forw = savehdr.forw;
2088 tmphdr.back = savehdr.back;
2089 tmphdr.firstused = state->args->geo->blksize;
2091 /* write the header to the temp buffer to initialise it */
2092 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2094 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2095 drop_blk->bp, &drophdr)) {
2096 xfs_attr3_leaf_moveents(state->args,
2097 drop_leaf, &drophdr, 0,
2098 tmp_leaf, &tmphdr, 0,
2100 xfs_attr3_leaf_moveents(state->args,
2101 save_leaf, &savehdr, 0,
2102 tmp_leaf, &tmphdr, tmphdr.count,
2105 xfs_attr3_leaf_moveents(state->args,
2106 save_leaf, &savehdr, 0,
2107 tmp_leaf, &tmphdr, 0,
2109 xfs_attr3_leaf_moveents(state->args,
2110 drop_leaf, &drophdr, 0,
2111 tmp_leaf, &tmphdr, tmphdr.count,
2114 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2115 savehdr = tmphdr; /* struct copy */
2116 kmem_free(tmp_leaf);
2119 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2120 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2121 state->args->geo->blksize - 1);
2124 * Copy out last hashval in each block for B-tree code.
2126 entry = xfs_attr3_leaf_entryp(save_leaf);
2127 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2130 /*========================================================================
2131 * Routines used for finding things in the Btree.
2132 *========================================================================*/
2135 * Look up a name in a leaf attribute list structure.
2136 * This is the internal routine, it uses the caller's buffer.
2138 * Note that duplicate keys are allowed, but only check within the
2139 * current leaf node. The Btree code must check in adjacent leaf nodes.
2141 * Return in args->index the index into the entry[] array of either
2142 * the found entry, or where the entry should have been (insert before
2145 * Don't change the args->value unless we find the attribute.
2148 xfs_attr3_leaf_lookup_int(
2150 struct xfs_da_args *args)
2152 struct xfs_attr_leafblock *leaf;
2153 struct xfs_attr3_icleaf_hdr ichdr;
2154 struct xfs_attr_leaf_entry *entry;
2155 struct xfs_attr_leaf_entry *entries;
2156 struct xfs_attr_leaf_name_local *name_loc;
2157 struct xfs_attr_leaf_name_remote *name_rmt;
2158 xfs_dahash_t hashval;
2162 trace_xfs_attr_leaf_lookup(args);
2165 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2166 entries = xfs_attr3_leaf_entryp(leaf);
2167 ASSERT(ichdr.count < args->geo->blksize / 8);
2170 * Binary search. (note: small blocks will skip this loop)
2172 hashval = args->hashval;
2173 probe = span = ichdr.count / 2;
2174 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2176 if (be32_to_cpu(entry->hashval) < hashval)
2178 else if (be32_to_cpu(entry->hashval) > hashval)
2183 ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2184 ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2187 * Since we may have duplicate hashval's, find the first matching
2188 * hashval in the leaf.
2190 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2194 while (probe < ichdr.count &&
2195 be32_to_cpu(entry->hashval) < hashval) {
2199 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2200 args->index = probe;
2205 * Duplicate keys may be present, so search all of them for a match.
2207 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2210 * GROT: Add code to remove incomplete entries.
2213 * If we are looking for INCOMPLETE entries, show only those.
2214 * If we are looking for complete entries, show only those.
2216 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2217 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2220 if (entry->flags & XFS_ATTR_LOCAL) {
2221 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2222 if (name_loc->namelen != args->namelen)
2224 if (memcmp(args->name, name_loc->nameval,
2225 args->namelen) != 0)
2227 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2229 args->index = probe;
2232 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2233 if (name_rmt->namelen != args->namelen)
2235 if (memcmp(args->name, name_rmt->name,
2236 args->namelen) != 0)
2238 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2240 args->index = probe;
2241 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2242 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2243 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2249 args->index = probe;
2254 * Get the value associated with an attribute name from a leaf attribute
2258 xfs_attr3_leaf_getvalue(
2260 struct xfs_da_args *args)
2262 struct xfs_attr_leafblock *leaf;
2263 struct xfs_attr3_icleaf_hdr ichdr;
2264 struct xfs_attr_leaf_entry *entry;
2265 struct xfs_attr_leaf_name_local *name_loc;
2266 struct xfs_attr_leaf_name_remote *name_rmt;
2270 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2271 ASSERT(ichdr.count < args->geo->blksize / 8);
2272 ASSERT(args->index < ichdr.count);
2274 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2275 if (entry->flags & XFS_ATTR_LOCAL) {
2276 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2277 ASSERT(name_loc->namelen == args->namelen);
2278 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2279 valuelen = be16_to_cpu(name_loc->valuelen);
2280 if (args->flags & ATTR_KERNOVAL) {
2281 args->valuelen = valuelen;
2284 if (args->valuelen < valuelen) {
2285 args->valuelen = valuelen;
2288 args->valuelen = valuelen;
2289 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2291 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2292 ASSERT(name_rmt->namelen == args->namelen);
2293 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2294 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2295 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2296 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2298 if (args->flags & ATTR_KERNOVAL) {
2299 args->valuelen = args->rmtvaluelen;
2302 if (args->valuelen < args->rmtvaluelen) {
2303 args->valuelen = args->rmtvaluelen;
2306 args->valuelen = args->rmtvaluelen;
2311 /*========================================================================
2313 *========================================================================*/
2316 * Move the indicated entries from one leaf to another.
2317 * NOTE: this routine modifies both source and destination leaves.
2321 xfs_attr3_leaf_moveents(
2322 struct xfs_da_args *args,
2323 struct xfs_attr_leafblock *leaf_s,
2324 struct xfs_attr3_icleaf_hdr *ichdr_s,
2326 struct xfs_attr_leafblock *leaf_d,
2327 struct xfs_attr3_icleaf_hdr *ichdr_d,
2331 struct xfs_attr_leaf_entry *entry_s;
2332 struct xfs_attr_leaf_entry *entry_d;
2338 * Check for nothing to do.
2344 * Set up environment.
2346 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2347 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2348 ASSERT(ichdr_s->magic == ichdr_d->magic);
2349 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2350 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2351 + xfs_attr3_leaf_hdr_size(leaf_s));
2352 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2353 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2354 + xfs_attr3_leaf_hdr_size(leaf_d));
2356 ASSERT(start_s < ichdr_s->count);
2357 ASSERT(start_d <= ichdr_d->count);
2358 ASSERT(count <= ichdr_s->count);
2362 * Move the entries in the destination leaf up to make a hole?
2364 if (start_d < ichdr_d->count) {
2365 tmp = ichdr_d->count - start_d;
2366 tmp *= sizeof(xfs_attr_leaf_entry_t);
2367 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2368 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2369 memmove(entry_d, entry_s, tmp);
2373 * Copy all entry's in the same (sorted) order,
2374 * but allocate attribute info packed and in sequence.
2376 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2377 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2379 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2380 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2381 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2384 * Code to drop INCOMPLETE entries. Difficult to use as we
2385 * may also need to change the insertion index. Code turned
2386 * off for 6.2, should be revisited later.
2388 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2389 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2390 ichdr_s->usedbytes -= tmp;
2391 ichdr_s->count -= 1;
2392 entry_d--; /* to compensate for ++ in loop hdr */
2394 if ((start_s + i) < offset)
2395 result++; /* insertion index adjustment */
2398 ichdr_d->firstused -= tmp;
2399 /* both on-disk, don't endian flip twice */
2400 entry_d->hashval = entry_s->hashval;
2401 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2402 entry_d->flags = entry_s->flags;
2403 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2404 <= args->geo->blksize);
2405 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2406 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2407 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2408 <= args->geo->blksize);
2409 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2410 ichdr_s->usedbytes -= tmp;
2411 ichdr_d->usedbytes += tmp;
2412 ichdr_s->count -= 1;
2413 ichdr_d->count += 1;
2414 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2415 + xfs_attr3_leaf_hdr_size(leaf_d);
2416 ASSERT(ichdr_d->firstused >= tmp);
2423 * Zero out the entries we just copied.
2425 if (start_s == ichdr_s->count) {
2426 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2427 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2428 ASSERT(((char *)entry_s + tmp) <=
2429 ((char *)leaf_s + args->geo->blksize));
2430 memset(entry_s, 0, tmp);
2433 * Move the remaining entries down to fill the hole,
2434 * then zero the entries at the top.
2436 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2437 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2438 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2439 memmove(entry_d, entry_s, tmp);
2441 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2442 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2443 ASSERT(((char *)entry_s + tmp) <=
2444 ((char *)leaf_s + args->geo->blksize));
2445 memset(entry_s, 0, tmp);
2449 * Fill in the freemap information
2451 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2452 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2453 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2454 ichdr_d->freemap[1].base = 0;
2455 ichdr_d->freemap[2].base = 0;
2456 ichdr_d->freemap[1].size = 0;
2457 ichdr_d->freemap[2].size = 0;
2458 ichdr_s->holes = 1; /* leaf may not be compact */
2462 * Pick up the last hashvalue from a leaf block.
2465 xfs_attr_leaf_lasthash(
2469 struct xfs_attr3_icleaf_hdr ichdr;
2470 struct xfs_attr_leaf_entry *entries;
2471 struct xfs_mount *mp = bp->b_target->bt_mount;
2473 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2474 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2476 *count = ichdr.count;
2479 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2483 * Calculate the number of bytes used to store the indicated attribute
2484 * (whether local or remote only calculate bytes in this block).
2487 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2489 struct xfs_attr_leaf_entry *entries;
2490 xfs_attr_leaf_name_local_t *name_loc;
2491 xfs_attr_leaf_name_remote_t *name_rmt;
2494 entries = xfs_attr3_leaf_entryp(leaf);
2495 if (entries[index].flags & XFS_ATTR_LOCAL) {
2496 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2497 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2498 be16_to_cpu(name_loc->valuelen));
2500 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2501 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2507 * Calculate the number of bytes that would be required to store the new
2508 * attribute (whether local or remote only calculate bytes in this block).
2509 * This routine decides as a side effect whether the attribute will be
2510 * a "local" or a "remote" attribute.
2513 xfs_attr_leaf_newentsize(
2514 struct xfs_da_args *args,
2519 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2520 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2527 return xfs_attr_leaf_entsize_remote(args->namelen);
2531 /*========================================================================
2532 * Manage the INCOMPLETE flag in a leaf entry
2533 *========================================================================*/
2536 * Clear the INCOMPLETE flag on an entry in a leaf block.
2539 xfs_attr3_leaf_clearflag(
2540 struct xfs_da_args *args)
2542 struct xfs_attr_leafblock *leaf;
2543 struct xfs_attr_leaf_entry *entry;
2544 struct xfs_attr_leaf_name_remote *name_rmt;
2548 struct xfs_attr3_icleaf_hdr ichdr;
2549 xfs_attr_leaf_name_local_t *name_loc;
2554 trace_xfs_attr_leaf_clearflag(args);
2556 * Set up the operation.
2558 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2563 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2564 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2567 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2568 ASSERT(args->index < ichdr.count);
2569 ASSERT(args->index >= 0);
2571 if (entry->flags & XFS_ATTR_LOCAL) {
2572 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2573 namelen = name_loc->namelen;
2574 name = (char *)name_loc->nameval;
2576 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2577 namelen = name_rmt->namelen;
2578 name = (char *)name_rmt->name;
2580 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2581 ASSERT(namelen == args->namelen);
2582 ASSERT(memcmp(name, args->name, namelen) == 0);
2585 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2586 xfs_trans_log_buf(args->trans, bp,
2587 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2589 if (args->rmtblkno) {
2590 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2591 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2592 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2593 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2594 xfs_trans_log_buf(args->trans, bp,
2595 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2599 * Commit the flag value change and start the next trans in series.
2601 return xfs_trans_roll(&args->trans, args->dp);
2605 * Set the INCOMPLETE flag on an entry in a leaf block.
2608 xfs_attr3_leaf_setflag(
2609 struct xfs_da_args *args)
2611 struct xfs_attr_leafblock *leaf;
2612 struct xfs_attr_leaf_entry *entry;
2613 struct xfs_attr_leaf_name_remote *name_rmt;
2617 struct xfs_attr3_icleaf_hdr ichdr;
2620 trace_xfs_attr_leaf_setflag(args);
2623 * Set up the operation.
2625 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2631 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2632 ASSERT(args->index < ichdr.count);
2633 ASSERT(args->index >= 0);
2635 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2637 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2638 entry->flags |= XFS_ATTR_INCOMPLETE;
2639 xfs_trans_log_buf(args->trans, bp,
2640 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2641 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2642 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2643 name_rmt->valueblk = 0;
2644 name_rmt->valuelen = 0;
2645 xfs_trans_log_buf(args->trans, bp,
2646 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2650 * Commit the flag value change and start the next trans in series.
2652 return xfs_trans_roll(&args->trans, args->dp);
2656 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2657 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2658 * entry given by args->blkno2/index2.
2660 * Note that they could be in different blocks, or in the same block.
2663 xfs_attr3_leaf_flipflags(
2664 struct xfs_da_args *args)
2666 struct xfs_attr_leafblock *leaf1;
2667 struct xfs_attr_leafblock *leaf2;
2668 struct xfs_attr_leaf_entry *entry1;
2669 struct xfs_attr_leaf_entry *entry2;
2670 struct xfs_attr_leaf_name_remote *name_rmt;
2671 struct xfs_buf *bp1;
2672 struct xfs_buf *bp2;
2675 struct xfs_attr3_icleaf_hdr ichdr1;
2676 struct xfs_attr3_icleaf_hdr ichdr2;
2677 xfs_attr_leaf_name_local_t *name_loc;
2678 int namelen1, namelen2;
2679 char *name1, *name2;
2682 trace_xfs_attr_leaf_flipflags(args);
2685 * Read the block containing the "old" attr
2687 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2692 * Read the block containing the "new" attr, if it is different
2694 if (args->blkno2 != args->blkno) {
2695 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2703 leaf1 = bp1->b_addr;
2704 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2706 leaf2 = bp2->b_addr;
2707 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2710 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2711 ASSERT(args->index < ichdr1.count);
2712 ASSERT(args->index >= 0);
2714 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2715 ASSERT(args->index2 < ichdr2.count);
2716 ASSERT(args->index2 >= 0);
2718 if (entry1->flags & XFS_ATTR_LOCAL) {
2719 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2720 namelen1 = name_loc->namelen;
2721 name1 = (char *)name_loc->nameval;
2723 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2724 namelen1 = name_rmt->namelen;
2725 name1 = (char *)name_rmt->name;
2727 if (entry2->flags & XFS_ATTR_LOCAL) {
2728 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2729 namelen2 = name_loc->namelen;
2730 name2 = (char *)name_loc->nameval;
2732 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2733 namelen2 = name_rmt->namelen;
2734 name2 = (char *)name_rmt->name;
2736 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2737 ASSERT(namelen1 == namelen2);
2738 ASSERT(memcmp(name1, name2, namelen1) == 0);
2741 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2742 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2744 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2745 xfs_trans_log_buf(args->trans, bp1,
2746 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2747 if (args->rmtblkno) {
2748 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2749 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2750 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2751 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2752 xfs_trans_log_buf(args->trans, bp1,
2753 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2756 entry2->flags |= XFS_ATTR_INCOMPLETE;
2757 xfs_trans_log_buf(args->trans, bp2,
2758 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2759 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2760 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2761 name_rmt->valueblk = 0;
2762 name_rmt->valuelen = 0;
2763 xfs_trans_log_buf(args->trans, bp2,
2764 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2768 * Commit the flag value change and start the next trans in series.
2770 error = xfs_trans_roll(&args->trans, args->dp);