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_format.h"
22 #include "xfs_log_format.h"
23 #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"
43 #include "xfs_dinode.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_attr_leafblock *src_leaf,
82 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
83 struct xfs_attr_leafblock *dst_leaf,
84 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
85 int move_count, struct xfs_mount *mp);
86 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
89 xfs_attr3_leaf_hdr_from_disk(
90 struct xfs_attr3_icleaf_hdr *to,
91 struct xfs_attr_leafblock *from)
95 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
96 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
98 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
99 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
101 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
102 to->back = be32_to_cpu(hdr3->info.hdr.back);
103 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
104 to->count = be16_to_cpu(hdr3->count);
105 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
106 to->firstused = be16_to_cpu(hdr3->firstused);
107 to->holes = hdr3->holes;
109 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
110 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
111 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
115 to->forw = be32_to_cpu(from->hdr.info.forw);
116 to->back = be32_to_cpu(from->hdr.info.back);
117 to->magic = be16_to_cpu(from->hdr.info.magic);
118 to->count = be16_to_cpu(from->hdr.count);
119 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
120 to->firstused = be16_to_cpu(from->hdr.firstused);
121 to->holes = from->hdr.holes;
123 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
124 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
125 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
130 xfs_attr3_leaf_hdr_to_disk(
131 struct xfs_attr_leafblock *to,
132 struct xfs_attr3_icleaf_hdr *from)
136 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
137 from->magic == XFS_ATTR3_LEAF_MAGIC);
139 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
140 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
142 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
143 hdr3->info.hdr.back = cpu_to_be32(from->back);
144 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
145 hdr3->count = cpu_to_be16(from->count);
146 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
147 hdr3->firstused = cpu_to_be16(from->firstused);
148 hdr3->holes = from->holes;
151 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
152 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
153 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
157 to->hdr.info.forw = cpu_to_be32(from->forw);
158 to->hdr.info.back = cpu_to_be32(from->back);
159 to->hdr.info.magic = cpu_to_be16(from->magic);
160 to->hdr.count = cpu_to_be16(from->count);
161 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
162 to->hdr.firstused = cpu_to_be16(from->firstused);
163 to->hdr.holes = from->holes;
166 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
167 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
168 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
173 xfs_attr3_leaf_verify(
176 struct xfs_mount *mp = bp->b_target->bt_mount;
177 struct xfs_attr_leafblock *leaf = bp->b_addr;
178 struct xfs_attr3_icleaf_hdr ichdr;
180 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
182 if (xfs_sb_version_hascrc(&mp->m_sb)) {
183 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
185 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
188 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
190 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
193 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
196 if (ichdr.count == 0)
199 /* XXX: need to range check rest of attr header values */
200 /* XXX: hash order check? */
206 xfs_attr3_leaf_write_verify(
209 struct xfs_mount *mp = bp->b_target->bt_mount;
210 struct xfs_buf_log_item *bip = bp->b_fspriv;
211 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
213 if (!xfs_attr3_leaf_verify(bp)) {
214 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
215 xfs_buf_ioerror(bp, EFSCORRUPTED);
219 if (!xfs_sb_version_hascrc(&mp->m_sb))
223 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
225 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_ATTR3_LEAF_CRC_OFF);
229 * leaf/node format detection on trees is sketchy, so a node read can be done on
230 * leaf level blocks when detection identifies the tree as a node format tree
231 * incorrectly. In this case, we need to swap the verifier to match the correct
232 * format of the block being read.
235 xfs_attr3_leaf_read_verify(
238 struct xfs_mount *mp = bp->b_target->bt_mount;
240 if ((xfs_sb_version_hascrc(&mp->m_sb) &&
241 !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
242 XFS_ATTR3_LEAF_CRC_OFF)) ||
243 !xfs_attr3_leaf_verify(bp)) {
244 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
245 xfs_buf_ioerror(bp, EFSCORRUPTED);
249 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
250 .verify_read = xfs_attr3_leaf_read_verify,
251 .verify_write = xfs_attr3_leaf_write_verify,
256 struct xfs_trans *tp,
257 struct xfs_inode *dp,
259 xfs_daddr_t mappedbno,
260 struct xfs_buf **bpp)
264 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
265 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
267 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
271 /*========================================================================
272 * Namespace helper routines
273 *========================================================================*/
276 * If namespace bits don't match return 0.
277 * If all match then return 1.
280 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
282 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
286 /*========================================================================
287 * External routines when attribute fork size < XFS_LITINO(mp).
288 *========================================================================*/
291 * Query whether the requested number of additional bytes of extended
292 * attribute space will be able to fit inline.
294 * Returns zero if not, else the di_forkoff fork offset to be used in the
295 * literal area for attribute data once the new bytes have been added.
297 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
298 * special case for dev/uuid inodes, they have fixed size data forks.
301 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
304 int minforkoff; /* lower limit on valid forkoff locations */
305 int maxforkoff; /* upper limit on valid forkoff locations */
307 xfs_mount_t *mp = dp->i_mount;
310 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
312 switch (dp->i_d.di_format) {
313 case XFS_DINODE_FMT_DEV:
314 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
315 return (offset >= minforkoff) ? minforkoff : 0;
316 case XFS_DINODE_FMT_UUID:
317 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
318 return (offset >= minforkoff) ? minforkoff : 0;
322 * If the requested numbers of bytes is smaller or equal to the
323 * current attribute fork size we can always proceed.
325 * Note that if_bytes in the data fork might actually be larger than
326 * the current data fork size is due to delalloc extents. In that
327 * case either the extent count will go down when they are converted
328 * to real extents, or the delalloc conversion will take care of the
329 * literal area rebalancing.
331 if (bytes <= XFS_IFORK_ASIZE(dp))
332 return dp->i_d.di_forkoff;
335 * For attr2 we can try to move the forkoff if there is space in the
336 * literal area, but for the old format we are done if there is no
337 * space in the fixed attribute fork.
339 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
342 dsize = dp->i_df.if_bytes;
344 switch (dp->i_d.di_format) {
345 case XFS_DINODE_FMT_EXTENTS:
347 * If there is no attr fork and the data fork is extents,
348 * determine if creating the default attr fork will result
349 * in the extents form migrating to btree. If so, the
350 * minimum offset only needs to be the space required for
353 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
354 xfs_default_attroffset(dp))
355 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
357 case XFS_DINODE_FMT_BTREE:
359 * If we have a data btree then keep forkoff if we have one,
360 * otherwise we are adding a new attr, so then we set
361 * minforkoff to where the btree root can finish so we have
362 * plenty of room for attrs
364 if (dp->i_d.di_forkoff) {
365 if (offset < dp->i_d.di_forkoff)
367 return dp->i_d.di_forkoff;
369 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
374 * A data fork btree root must have space for at least
375 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
377 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
378 minforkoff = roundup(minforkoff, 8) >> 3;
380 /* attr fork btree root can have at least this many key/ptr pairs */
381 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
382 XFS_BMDR_SPACE_CALC(MINABTPTRS);
383 maxforkoff = maxforkoff >> 3; /* rounded down */
385 if (offset >= maxforkoff)
387 if (offset >= minforkoff)
393 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
396 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
398 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
399 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
400 spin_lock(&mp->m_sb_lock);
401 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
402 xfs_sb_version_addattr2(&mp->m_sb);
403 spin_unlock(&mp->m_sb_lock);
404 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
406 spin_unlock(&mp->m_sb_lock);
411 * Create the initial contents of a shortform attribute list.
414 xfs_attr_shortform_create(xfs_da_args_t *args)
416 xfs_attr_sf_hdr_t *hdr;
420 trace_xfs_attr_sf_create(args);
426 ASSERT(ifp->if_bytes == 0);
427 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
428 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
429 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
430 ifp->if_flags |= XFS_IFINLINE;
432 ASSERT(ifp->if_flags & XFS_IFINLINE);
434 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
435 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
437 hdr->totsize = cpu_to_be16(sizeof(*hdr));
438 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
442 * Add a name/value pair to the shortform attribute list.
443 * Overflow from the inode has already been checked for.
446 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
448 xfs_attr_shortform_t *sf;
449 xfs_attr_sf_entry_t *sfe;
455 trace_xfs_attr_sf_add(args);
459 dp->i_d.di_forkoff = forkoff;
462 ASSERT(ifp->if_flags & XFS_IFINLINE);
463 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
465 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
467 if (sfe->namelen != args->namelen)
469 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
471 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
477 offset = (char *)sfe - (char *)sf;
478 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
479 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
480 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
481 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
483 sfe->namelen = args->namelen;
484 sfe->valuelen = args->valuelen;
485 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
486 memcpy(sfe->nameval, args->name, args->namelen);
487 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
489 be16_add_cpu(&sf->hdr.totsize, size);
490 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
492 xfs_sbversion_add_attr2(mp, args->trans);
496 * After the last attribute is removed revert to original inode format,
497 * making all literal area available to the data fork once more.
501 struct xfs_inode *ip,
502 struct xfs_trans *tp)
504 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
505 ip->i_d.di_forkoff = 0;
506 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
508 ASSERT(ip->i_d.di_anextents == 0);
509 ASSERT(ip->i_afp == NULL);
511 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
515 * Remove an attribute from the shortform attribute list structure.
518 xfs_attr_shortform_remove(xfs_da_args_t *args)
520 xfs_attr_shortform_t *sf;
521 xfs_attr_sf_entry_t *sfe;
522 int base, size=0, end, totsize, i;
526 trace_xfs_attr_sf_remove(args);
530 base = sizeof(xfs_attr_sf_hdr_t);
531 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
534 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
536 size = XFS_ATTR_SF_ENTSIZE(sfe);
537 if (sfe->namelen != args->namelen)
539 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
541 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
546 return(XFS_ERROR(ENOATTR));
549 * Fix up the attribute fork data, covering the hole
552 totsize = be16_to_cpu(sf->hdr.totsize);
554 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
556 be16_add_cpu(&sf->hdr.totsize, -size);
559 * Fix up the start offset of the attribute fork
562 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
563 (mp->m_flags & XFS_MOUNT_ATTR2) &&
564 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
565 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
566 xfs_attr_fork_reset(dp, args->trans);
568 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
569 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
570 ASSERT(dp->i_d.di_forkoff);
571 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
572 (args->op_flags & XFS_DA_OP_ADDNAME) ||
573 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
574 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
575 xfs_trans_log_inode(args->trans, dp,
576 XFS_ILOG_CORE | XFS_ILOG_ADATA);
579 xfs_sbversion_add_attr2(mp, args->trans);
585 * Look up a name in a shortform attribute list structure.
589 xfs_attr_shortform_lookup(xfs_da_args_t *args)
591 xfs_attr_shortform_t *sf;
592 xfs_attr_sf_entry_t *sfe;
596 trace_xfs_attr_sf_lookup(args);
598 ifp = args->dp->i_afp;
599 ASSERT(ifp->if_flags & XFS_IFINLINE);
600 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
602 for (i = 0; i < sf->hdr.count;
603 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
604 if (sfe->namelen != args->namelen)
606 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
608 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
610 return(XFS_ERROR(EEXIST));
612 return(XFS_ERROR(ENOATTR));
616 * Look up a name in a shortform attribute list structure.
620 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
622 xfs_attr_shortform_t *sf;
623 xfs_attr_sf_entry_t *sfe;
626 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
627 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
629 for (i = 0; i < sf->hdr.count;
630 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
631 if (sfe->namelen != args->namelen)
633 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
635 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
637 if (args->flags & ATTR_KERNOVAL) {
638 args->valuelen = sfe->valuelen;
639 return(XFS_ERROR(EEXIST));
641 if (args->valuelen < sfe->valuelen) {
642 args->valuelen = sfe->valuelen;
643 return(XFS_ERROR(ERANGE));
645 args->valuelen = sfe->valuelen;
646 memcpy(args->value, &sfe->nameval[args->namelen],
648 return(XFS_ERROR(EEXIST));
650 return(XFS_ERROR(ENOATTR));
654 * Convert from using the shortform to the leaf.
657 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
660 xfs_attr_shortform_t *sf;
661 xfs_attr_sf_entry_t *sfe;
669 trace_xfs_attr_sf_to_leaf(args);
673 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
674 size = be16_to_cpu(sf->hdr.totsize);
675 tmpbuffer = kmem_alloc(size, KM_SLEEP);
676 ASSERT(tmpbuffer != NULL);
677 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
678 sf = (xfs_attr_shortform_t *)tmpbuffer;
680 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
681 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
684 error = xfs_da_grow_inode(args, &blkno);
687 * If we hit an IO error middle of the transaction inside
688 * grow_inode(), we may have inconsistent data. Bail out.
692 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
693 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
698 error = xfs_attr3_leaf_create(args, blkno, &bp);
700 error = xfs_da_shrink_inode(args, 0, bp);
704 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
705 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
709 memset((char *)&nargs, 0, sizeof(nargs));
711 nargs.firstblock = args->firstblock;
712 nargs.flist = args->flist;
713 nargs.total = args->total;
714 nargs.whichfork = XFS_ATTR_FORK;
715 nargs.trans = args->trans;
716 nargs.op_flags = XFS_DA_OP_OKNOENT;
719 for (i = 0; i < sf->hdr.count; i++) {
720 nargs.name = sfe->nameval;
721 nargs.namelen = sfe->namelen;
722 nargs.value = &sfe->nameval[nargs.namelen];
723 nargs.valuelen = sfe->valuelen;
724 nargs.hashval = xfs_da_hashname(sfe->nameval,
726 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
727 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
728 ASSERT(error == ENOATTR);
729 error = xfs_attr3_leaf_add(bp, &nargs);
730 ASSERT(error != ENOSPC);
733 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
738 kmem_free(tmpbuffer);
743 * Check a leaf attribute block to see if all the entries would fit into
744 * a shortform attribute list.
747 xfs_attr_shortform_allfit(
749 struct xfs_inode *dp)
751 struct xfs_attr_leafblock *leaf;
752 struct xfs_attr_leaf_entry *entry;
753 xfs_attr_leaf_name_local_t *name_loc;
754 struct xfs_attr3_icleaf_hdr leafhdr;
759 xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
760 entry = xfs_attr3_leaf_entryp(leaf);
762 bytes = sizeof(struct xfs_attr_sf_hdr);
763 for (i = 0; i < leafhdr.count; entry++, i++) {
764 if (entry->flags & XFS_ATTR_INCOMPLETE)
765 continue; /* don't copy partial entries */
766 if (!(entry->flags & XFS_ATTR_LOCAL))
768 name_loc = xfs_attr3_leaf_name_local(leaf, i);
769 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
771 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
773 bytes += sizeof(struct xfs_attr_sf_entry) - 1
775 + be16_to_cpu(name_loc->valuelen);
777 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
778 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
779 (bytes == sizeof(struct xfs_attr_sf_hdr)))
781 return xfs_attr_shortform_bytesfit(dp, bytes);
785 * Convert a leaf attribute list to shortform attribute list
788 xfs_attr3_leaf_to_shortform(
790 struct xfs_da_args *args,
793 struct xfs_attr_leafblock *leaf;
794 struct xfs_attr3_icleaf_hdr ichdr;
795 struct xfs_attr_leaf_entry *entry;
796 struct xfs_attr_leaf_name_local *name_loc;
797 struct xfs_da_args nargs;
798 struct xfs_inode *dp = args->dp;
803 trace_xfs_attr_leaf_to_sf(args);
805 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
809 memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
811 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
812 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
813 entry = xfs_attr3_leaf_entryp(leaf);
815 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
816 memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
819 * Clean out the prior contents of the attribute list.
821 error = xfs_da_shrink_inode(args, 0, bp);
826 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
827 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
828 xfs_attr_fork_reset(dp, args->trans);
832 xfs_attr_shortform_create(args);
835 * Copy the attributes
837 memset((char *)&nargs, 0, sizeof(nargs));
839 nargs.firstblock = args->firstblock;
840 nargs.flist = args->flist;
841 nargs.total = args->total;
842 nargs.whichfork = XFS_ATTR_FORK;
843 nargs.trans = args->trans;
844 nargs.op_flags = XFS_DA_OP_OKNOENT;
846 for (i = 0; i < ichdr.count; entry++, i++) {
847 if (entry->flags & XFS_ATTR_INCOMPLETE)
848 continue; /* don't copy partial entries */
851 ASSERT(entry->flags & XFS_ATTR_LOCAL);
852 name_loc = xfs_attr3_leaf_name_local(leaf, i);
853 nargs.name = name_loc->nameval;
854 nargs.namelen = name_loc->namelen;
855 nargs.value = &name_loc->nameval[nargs.namelen];
856 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
857 nargs.hashval = be32_to_cpu(entry->hashval);
858 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
859 xfs_attr_shortform_add(&nargs, forkoff);
864 kmem_free(tmpbuffer);
869 * Convert from using a single leaf to a root node and a leaf.
872 xfs_attr3_leaf_to_node(
873 struct xfs_da_args *args)
875 struct xfs_attr_leafblock *leaf;
876 struct xfs_attr3_icleaf_hdr icleafhdr;
877 struct xfs_attr_leaf_entry *entries;
878 struct xfs_da_node_entry *btree;
879 struct xfs_da3_icnode_hdr icnodehdr;
880 struct xfs_da_intnode *node;
881 struct xfs_inode *dp = args->dp;
882 struct xfs_mount *mp = dp->i_mount;
883 struct xfs_buf *bp1 = NULL;
884 struct xfs_buf *bp2 = NULL;
888 trace_xfs_attr_leaf_to_node(args);
890 error = xfs_da_grow_inode(args, &blkno);
893 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
897 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
901 /* copy leaf to new buffer, update identifiers */
902 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
903 bp2->b_ops = bp1->b_ops;
904 memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
905 if (xfs_sb_version_hascrc(&mp->m_sb)) {
906 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
907 hdr3->blkno = cpu_to_be64(bp2->b_bn);
909 xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
912 * Set up the new root node.
914 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
918 xfs_da3_node_hdr_from_disk(&icnodehdr, node);
919 btree = xfs_da3_node_tree_p(node);
922 xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
923 entries = xfs_attr3_leaf_entryp(leaf);
925 /* both on-disk, don't endian-flip twice */
926 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
927 btree[0].before = cpu_to_be32(blkno);
929 xfs_da3_node_hdr_to_disk(node, &icnodehdr);
930 xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
936 /*========================================================================
937 * Routines used for growing the Btree.
938 *========================================================================*/
941 * Create the initial contents of a leaf attribute list
942 * or a leaf in a node attribute list.
945 xfs_attr3_leaf_create(
946 struct xfs_da_args *args,
948 struct xfs_buf **bpp)
950 struct xfs_attr_leafblock *leaf;
951 struct xfs_attr3_icleaf_hdr ichdr;
952 struct xfs_inode *dp = args->dp;
953 struct xfs_mount *mp = dp->i_mount;
957 trace_xfs_attr_leaf_create(args);
959 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
963 bp->b_ops = &xfs_attr3_leaf_buf_ops;
964 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
966 memset(leaf, 0, XFS_LBSIZE(mp));
968 memset(&ichdr, 0, sizeof(ichdr));
969 ichdr.firstused = XFS_LBSIZE(mp);
971 if (xfs_sb_version_hascrc(&mp->m_sb)) {
972 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
974 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
976 hdr3->blkno = cpu_to_be64(bp->b_bn);
977 hdr3->owner = cpu_to_be64(dp->i_ino);
978 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
980 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
982 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
983 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
985 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
987 xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
988 xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
995 * Split the leaf node, rebalance, then add the new entry.
998 xfs_attr3_leaf_split(
999 struct xfs_da_state *state,
1000 struct xfs_da_state_blk *oldblk,
1001 struct xfs_da_state_blk *newblk)
1006 trace_xfs_attr_leaf_split(state->args);
1009 * Allocate space for a new leaf node.
1011 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1012 error = xfs_da_grow_inode(state->args, &blkno);
1015 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1018 newblk->blkno = blkno;
1019 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1022 * Rebalance the entries across the two leaves.
1023 * NOTE: rebalance() currently depends on the 2nd block being empty.
1025 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1026 error = xfs_da3_blk_link(state, oldblk, newblk);
1031 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1032 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1033 * "new" attrs info. Will need the "old" info to remove it later.
1035 * Insert the "new" entry in the correct block.
1037 if (state->inleaf) {
1038 trace_xfs_attr_leaf_add_old(state->args);
1039 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1041 trace_xfs_attr_leaf_add_new(state->args);
1042 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1046 * Update last hashval in each block since we added the name.
1048 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1049 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1054 * Add a name to the leaf attribute list structure.
1059 struct xfs_da_args *args)
1061 struct xfs_attr_leafblock *leaf;
1062 struct xfs_attr3_icleaf_hdr ichdr;
1069 trace_xfs_attr_leaf_add(args);
1072 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1073 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1074 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1075 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1078 * Search through freemap for first-fit on new name length.
1079 * (may need to figure in size of entry struct too)
1081 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1082 + xfs_attr3_leaf_hdr_size(leaf);
1083 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1084 if (tablesize > ichdr.firstused) {
1085 sum += ichdr.freemap[i].size;
1088 if (!ichdr.freemap[i].size)
1089 continue; /* no space in this map */
1091 if (ichdr.freemap[i].base < ichdr.firstused)
1092 tmp += sizeof(xfs_attr_leaf_entry_t);
1093 if (ichdr.freemap[i].size >= tmp) {
1094 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1097 sum += ichdr.freemap[i].size;
1101 * If there are no holes in the address space of the block,
1102 * and we don't have enough freespace, then compaction will do us
1103 * no good and we should just give up.
1105 if (!ichdr.holes && sum < entsize)
1106 return XFS_ERROR(ENOSPC);
1109 * Compact the entries to coalesce free space.
1110 * This may change the hdr->count via dropping INCOMPLETE entries.
1112 xfs_attr3_leaf_compact(args, &ichdr, bp);
1115 * After compaction, the block is guaranteed to have only one
1116 * free region, in freemap[0]. If it is not big enough, give up.
1118 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1123 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1126 xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1127 xfs_trans_log_buf(args->trans, bp,
1128 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1129 xfs_attr3_leaf_hdr_size(leaf)));
1134 * Add a name to a leaf attribute list structure.
1137 xfs_attr3_leaf_add_work(
1139 struct xfs_attr3_icleaf_hdr *ichdr,
1140 struct xfs_da_args *args,
1143 struct xfs_attr_leafblock *leaf;
1144 struct xfs_attr_leaf_entry *entry;
1145 struct xfs_attr_leaf_name_local *name_loc;
1146 struct xfs_attr_leaf_name_remote *name_rmt;
1147 struct xfs_mount *mp;
1151 trace_xfs_attr_leaf_add_work(args);
1154 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1155 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1158 * Force open some space in the entry array and fill it in.
1160 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1161 if (args->index < ichdr->count) {
1162 tmp = ichdr->count - args->index;
1163 tmp *= sizeof(xfs_attr_leaf_entry_t);
1164 memmove(entry + 1, entry, tmp);
1165 xfs_trans_log_buf(args->trans, bp,
1166 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1171 * Allocate space for the new string (at the end of the run).
1173 mp = args->trans->t_mountp;
1174 ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1175 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1176 ASSERT(ichdr->freemap[mapindex].size >=
1177 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1178 mp->m_sb.sb_blocksize, NULL));
1179 ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1180 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1182 ichdr->freemap[mapindex].size -=
1183 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1184 mp->m_sb.sb_blocksize, &tmp);
1186 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1187 ichdr->freemap[mapindex].size);
1188 entry->hashval = cpu_to_be32(args->hashval);
1189 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1190 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1191 if (args->op_flags & XFS_DA_OP_RENAME) {
1192 entry->flags |= XFS_ATTR_INCOMPLETE;
1193 if ((args->blkno2 == args->blkno) &&
1194 (args->index2 <= args->index)) {
1198 xfs_trans_log_buf(args->trans, bp,
1199 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1200 ASSERT((args->index == 0) ||
1201 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1202 ASSERT((args->index == ichdr->count - 1) ||
1203 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1206 * For "remote" attribute values, simply note that we need to
1207 * allocate space for the "remote" value. We can't actually
1208 * allocate the extents in this transaction, and we can't decide
1209 * which blocks they should be as we might allocate more blocks
1210 * as part of this transaction (a split operation for example).
1212 if (entry->flags & XFS_ATTR_LOCAL) {
1213 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1214 name_loc->namelen = args->namelen;
1215 name_loc->valuelen = cpu_to_be16(args->valuelen);
1216 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1217 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1218 be16_to_cpu(name_loc->valuelen));
1220 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1221 name_rmt->namelen = args->namelen;
1222 memcpy((char *)name_rmt->name, args->name, args->namelen);
1223 entry->flags |= XFS_ATTR_INCOMPLETE;
1225 name_rmt->valuelen = 0;
1226 name_rmt->valueblk = 0;
1228 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1230 xfs_trans_log_buf(args->trans, bp,
1231 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1232 xfs_attr_leaf_entsize(leaf, args->index)));
1235 * Update the control info for this leaf node
1237 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1238 ichdr->firstused = be16_to_cpu(entry->nameidx);
1240 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1241 + xfs_attr3_leaf_hdr_size(leaf));
1242 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1243 + xfs_attr3_leaf_hdr_size(leaf);
1245 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1246 if (ichdr->freemap[i].base == tmp) {
1247 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1248 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1251 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1256 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1259 xfs_attr3_leaf_compact(
1260 struct xfs_da_args *args,
1261 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1264 struct xfs_attr_leafblock *leaf_src;
1265 struct xfs_attr_leafblock *leaf_dst;
1266 struct xfs_attr3_icleaf_hdr ichdr_src;
1267 struct xfs_trans *trans = args->trans;
1268 struct xfs_mount *mp = trans->t_mountp;
1271 trace_xfs_attr_leaf_compact(args);
1273 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1274 memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1275 memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1276 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1277 leaf_dst = bp->b_addr;
1280 * Copy the on-disk header back into the destination buffer to ensure
1281 * all the information in the header that is not part of the incore
1282 * header structure is preserved.
1284 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1286 /* Initialise the incore headers */
1287 ichdr_src = *ichdr_dst; /* struct copy */
1288 ichdr_dst->firstused = XFS_LBSIZE(mp);
1289 ichdr_dst->usedbytes = 0;
1290 ichdr_dst->count = 0;
1291 ichdr_dst->holes = 0;
1292 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1293 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1294 ichdr_dst->freemap[0].base;
1296 /* write the header back to initialise the underlying buffer */
1297 xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1300 * Copy all entry's in the same (sorted) order,
1301 * but allocate name/value pairs packed and in sequence.
1303 xfs_attr3_leaf_moveents(leaf_src, &ichdr_src, 0, leaf_dst, ichdr_dst, 0,
1304 ichdr_src.count, mp);
1306 * this logs the entire buffer, but the caller must write the header
1307 * back to the buffer when it is finished modifying it.
1309 xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1311 kmem_free(tmpbuffer);
1315 * Compare two leaf blocks "order".
1316 * Return 0 unless leaf2 should go before leaf1.
1319 xfs_attr3_leaf_order(
1320 struct xfs_buf *leaf1_bp,
1321 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1322 struct xfs_buf *leaf2_bp,
1323 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1325 struct xfs_attr_leaf_entry *entries1;
1326 struct xfs_attr_leaf_entry *entries2;
1328 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1329 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1330 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1331 ((be32_to_cpu(entries2[0].hashval) <
1332 be32_to_cpu(entries1[0].hashval)) ||
1333 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1334 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1341 xfs_attr_leaf_order(
1342 struct xfs_buf *leaf1_bp,
1343 struct xfs_buf *leaf2_bp)
1345 struct xfs_attr3_icleaf_hdr ichdr1;
1346 struct xfs_attr3_icleaf_hdr ichdr2;
1348 xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1349 xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1350 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1354 * Redistribute the attribute list entries between two leaf nodes,
1355 * taking into account the size of the new entry.
1357 * NOTE: if new block is empty, then it will get the upper half of the
1358 * old block. At present, all (one) callers pass in an empty second block.
1360 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1361 * to match what it is doing in splitting the attribute leaf block. Those
1362 * values are used in "atomic rename" operations on attributes. Note that
1363 * the "new" and "old" values can end up in different blocks.
1366 xfs_attr3_leaf_rebalance(
1367 struct xfs_da_state *state,
1368 struct xfs_da_state_blk *blk1,
1369 struct xfs_da_state_blk *blk2)
1371 struct xfs_da_args *args;
1372 struct xfs_attr_leafblock *leaf1;
1373 struct xfs_attr_leafblock *leaf2;
1374 struct xfs_attr3_icleaf_hdr ichdr1;
1375 struct xfs_attr3_icleaf_hdr ichdr2;
1376 struct xfs_attr_leaf_entry *entries1;
1377 struct xfs_attr_leaf_entry *entries2;
1385 * Set up environment.
1387 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1388 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1389 leaf1 = blk1->bp->b_addr;
1390 leaf2 = blk2->bp->b_addr;
1391 xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1392 xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1393 ASSERT(ichdr2.count == 0);
1396 trace_xfs_attr_leaf_rebalance(args);
1399 * Check ordering of blocks, reverse if it makes things simpler.
1401 * NOTE: Given that all (current) callers pass in an empty
1402 * second block, this code should never set "swap".
1405 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1406 struct xfs_da_state_blk *tmp_blk;
1407 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1413 /* struct copies to swap them rather than reconverting */
1418 leaf1 = blk1->bp->b_addr;
1419 leaf2 = blk2->bp->b_addr;
1424 * Examine entries until we reduce the absolute difference in
1425 * byte usage between the two blocks to a minimum. Then get
1426 * the direction to copy and the number of elements to move.
1428 * "inleaf" is true if the new entry should be inserted into blk1.
1429 * If "swap" is also true, then reverse the sense of "inleaf".
1431 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1435 state->inleaf = !state->inleaf;
1438 * Move any entries required from leaf to leaf:
1440 if (count < ichdr1.count) {
1442 * Figure the total bytes to be added to the destination leaf.
1444 /* number entries being moved */
1445 count = ichdr1.count - count;
1446 space = ichdr1.usedbytes - totallen;
1447 space += count * sizeof(xfs_attr_leaf_entry_t);
1450 * leaf2 is the destination, compact it if it looks tight.
1452 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1453 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1455 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1458 * Move high entries from leaf1 to low end of leaf2.
1460 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1461 leaf2, &ichdr2, 0, count, state->mp);
1463 } else if (count > ichdr1.count) {
1465 * I assert that since all callers pass in an empty
1466 * second buffer, this code should never execute.
1471 * Figure the total bytes to be added to the destination leaf.
1473 /* number entries being moved */
1474 count -= ichdr1.count;
1475 space = totallen - ichdr1.usedbytes;
1476 space += count * sizeof(xfs_attr_leaf_entry_t);
1479 * leaf1 is the destination, compact it if it looks tight.
1481 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1482 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1484 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1487 * Move low entries from leaf2 to high end of leaf1.
1489 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1490 ichdr1.count, count, state->mp);
1493 xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1494 xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1495 xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1496 xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1499 * Copy out last hashval in each block for B-tree code.
1501 entries1 = xfs_attr3_leaf_entryp(leaf1);
1502 entries2 = xfs_attr3_leaf_entryp(leaf2);
1503 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1504 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1507 * Adjust the expected index for insertion.
1508 * NOTE: this code depends on the (current) situation that the
1509 * second block was originally empty.
1511 * If the insertion point moved to the 2nd block, we must adjust
1512 * the index. We must also track the entry just following the
1513 * new entry for use in an "atomic rename" operation, that entry
1514 * is always the "old" entry and the "new" entry is what we are
1515 * inserting. The index/blkno fields refer to the "old" entry,
1516 * while the index2/blkno2 fields refer to the "new" entry.
1518 if (blk1->index > ichdr1.count) {
1519 ASSERT(state->inleaf == 0);
1520 blk2->index = blk1->index - ichdr1.count;
1521 args->index = args->index2 = blk2->index;
1522 args->blkno = args->blkno2 = blk2->blkno;
1523 } else if (blk1->index == ichdr1.count) {
1524 if (state->inleaf) {
1525 args->index = blk1->index;
1526 args->blkno = blk1->blkno;
1528 args->blkno2 = blk2->blkno;
1531 * On a double leaf split, the original attr location
1532 * is already stored in blkno2/index2, so don't
1533 * overwrite it overwise we corrupt the tree.
1535 blk2->index = blk1->index - ichdr1.count;
1536 args->index = blk2->index;
1537 args->blkno = blk2->blkno;
1538 if (!state->extravalid) {
1540 * set the new attr location to match the old
1541 * one and let the higher level split code
1542 * decide where in the leaf to place it.
1544 args->index2 = blk2->index;
1545 args->blkno2 = blk2->blkno;
1549 ASSERT(state->inleaf == 1);
1550 args->index = args->index2 = blk1->index;
1551 args->blkno = args->blkno2 = blk1->blkno;
1556 * Examine entries until we reduce the absolute difference in
1557 * byte usage between the two blocks to a minimum.
1558 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1559 * GROT: there will always be enough room in either block for a new entry.
1560 * GROT: Do a double-split for this case?
1563 xfs_attr3_leaf_figure_balance(
1564 struct xfs_da_state *state,
1565 struct xfs_da_state_blk *blk1,
1566 struct xfs_attr3_icleaf_hdr *ichdr1,
1567 struct xfs_da_state_blk *blk2,
1568 struct xfs_attr3_icleaf_hdr *ichdr2,
1572 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1573 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1574 struct xfs_attr_leaf_entry *entry;
1585 * Examine entries until we reduce the absolute difference in
1586 * byte usage between the two blocks to a minimum.
1588 max = ichdr1->count + ichdr2->count;
1589 half = (max + 1) * sizeof(*entry);
1590 half += ichdr1->usedbytes + ichdr2->usedbytes +
1591 xfs_attr_leaf_newentsize(state->args->namelen,
1592 state->args->valuelen,
1593 state->blocksize, NULL);
1595 lastdelta = state->blocksize;
1596 entry = xfs_attr3_leaf_entryp(leaf1);
1597 for (count = index = 0; count < max; entry++, index++, count++) {
1599 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1601 * The new entry is in the first block, account for it.
1603 if (count == blk1->index) {
1604 tmp = totallen + sizeof(*entry) +
1605 xfs_attr_leaf_newentsize(
1606 state->args->namelen,
1607 state->args->valuelen,
1608 state->blocksize, NULL);
1609 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1611 lastdelta = XFS_ATTR_ABS(half - tmp);
1617 * Wrap around into the second block if necessary.
1619 if (count == ichdr1->count) {
1621 entry = xfs_attr3_leaf_entryp(leaf1);
1626 * Figure out if next leaf entry would be too much.
1628 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1630 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1632 lastdelta = XFS_ATTR_ABS(half - tmp);
1638 * Calculate the number of usedbytes that will end up in lower block.
1639 * If new entry not in lower block, fix up the count.
1641 totallen -= count * sizeof(*entry);
1643 totallen -= sizeof(*entry) +
1644 xfs_attr_leaf_newentsize(
1645 state->args->namelen,
1646 state->args->valuelen,
1647 state->blocksize, NULL);
1651 *usedbytesarg = totallen;
1655 /*========================================================================
1656 * Routines used for shrinking the Btree.
1657 *========================================================================*/
1660 * Check a leaf block and its neighbors to see if the block should be
1661 * collapsed into one or the other neighbor. Always keep the block
1662 * with the smaller block number.
1663 * If the current block is over 50% full, don't try to join it, return 0.
1664 * If the block is empty, fill in the state structure and return 2.
1665 * If it can be collapsed, fill in the state structure and return 1.
1666 * If nothing can be done, return 0.
1668 * GROT: allow for INCOMPLETE entries in calculation.
1671 xfs_attr3_leaf_toosmall(
1672 struct xfs_da_state *state,
1675 struct xfs_attr_leafblock *leaf;
1676 struct xfs_da_state_blk *blk;
1677 struct xfs_attr3_icleaf_hdr ichdr;
1686 trace_xfs_attr_leaf_toosmall(state->args);
1689 * Check for the degenerate case of the block being over 50% full.
1690 * If so, it's not worth even looking to see if we might be able
1691 * to coalesce with a sibling.
1693 blk = &state->path.blk[ state->path.active-1 ];
1694 leaf = blk->bp->b_addr;
1695 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1696 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1697 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1699 if (bytes > (state->blocksize >> 1)) {
1700 *action = 0; /* blk over 50%, don't try to join */
1705 * Check for the degenerate case of the block being empty.
1706 * If the block is empty, we'll simply delete it, no need to
1707 * coalesce it with a sibling block. We choose (arbitrarily)
1708 * to merge with the forward block unless it is NULL.
1710 if (ichdr.count == 0) {
1712 * Make altpath point to the block we want to keep and
1713 * path point to the block we want to drop (this one).
1715 forward = (ichdr.forw != 0);
1716 memcpy(&state->altpath, &state->path, sizeof(state->path));
1717 error = xfs_da3_path_shift(state, &state->altpath, forward,
1730 * Examine each sibling block to see if we can coalesce with
1731 * at least 25% free space to spare. We need to figure out
1732 * whether to merge with the forward or the backward block.
1733 * We prefer coalescing with the lower numbered sibling so as
1734 * to shrink an attribute list over time.
1736 /* start with smaller blk num */
1737 forward = ichdr.forw < ichdr.back;
1738 for (i = 0; i < 2; forward = !forward, i++) {
1739 struct xfs_attr3_icleaf_hdr ichdr2;
1746 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1751 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1753 bytes = state->blocksize - (state->blocksize >> 2) -
1754 ichdr.usedbytes - ichdr2.usedbytes -
1755 ((ichdr.count + ichdr2.count) *
1756 sizeof(xfs_attr_leaf_entry_t)) -
1757 xfs_attr3_leaf_hdr_size(leaf);
1759 xfs_trans_brelse(state->args->trans, bp);
1761 break; /* fits with at least 25% to spare */
1769 * Make altpath point to the block we want to keep (the lower
1770 * numbered block) and path point to the block we want to drop.
1772 memcpy(&state->altpath, &state->path, sizeof(state->path));
1773 if (blkno < blk->blkno) {
1774 error = xfs_da3_path_shift(state, &state->altpath, forward,
1777 error = xfs_da3_path_shift(state, &state->path, forward,
1791 * Remove a name from the leaf attribute list structure.
1793 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1794 * If two leaves are 37% full, when combined they will leave 25% free.
1797 xfs_attr3_leaf_remove(
1799 struct xfs_da_args *args)
1801 struct xfs_attr_leafblock *leaf;
1802 struct xfs_attr3_icleaf_hdr ichdr;
1803 struct xfs_attr_leaf_entry *entry;
1804 struct xfs_mount *mp = args->trans->t_mountp;
1813 trace_xfs_attr_leaf_remove(args);
1816 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1818 ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
1819 ASSERT(args->index >= 0 && args->index < ichdr.count);
1820 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1821 xfs_attr3_leaf_hdr_size(leaf));
1823 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1825 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1826 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1829 * Scan through free region table:
1830 * check for adjacency of free'd entry with an existing one,
1831 * find smallest free region in case we need to replace it,
1832 * adjust any map that borders the entry table,
1834 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1835 + xfs_attr3_leaf_hdr_size(leaf);
1836 tmp = ichdr.freemap[0].size;
1837 before = after = -1;
1838 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1839 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1840 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1841 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
1842 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
1843 if (ichdr.freemap[i].base == tablesize) {
1844 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1845 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1848 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1849 be16_to_cpu(entry->nameidx)) {
1851 } else if (ichdr.freemap[i].base ==
1852 (be16_to_cpu(entry->nameidx) + entsize)) {
1854 } else if (ichdr.freemap[i].size < tmp) {
1855 tmp = ichdr.freemap[i].size;
1861 * Coalesce adjacent freemap regions,
1862 * or replace the smallest region.
1864 if ((before >= 0) || (after >= 0)) {
1865 if ((before >= 0) && (after >= 0)) {
1866 ichdr.freemap[before].size += entsize;
1867 ichdr.freemap[before].size += ichdr.freemap[after].size;
1868 ichdr.freemap[after].base = 0;
1869 ichdr.freemap[after].size = 0;
1870 } else if (before >= 0) {
1871 ichdr.freemap[before].size += entsize;
1873 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1874 ichdr.freemap[after].size += entsize;
1878 * Replace smallest region (if it is smaller than free'd entry)
1880 if (ichdr.freemap[smallest].size < entsize) {
1881 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1882 ichdr.freemap[smallest].size = entsize;
1887 * Did we remove the first entry?
1889 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1895 * Compress the remaining entries and zero out the removed stuff.
1897 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1898 ichdr.usedbytes -= entsize;
1899 xfs_trans_log_buf(args->trans, bp,
1900 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1903 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1904 memmove(entry, entry + 1, tmp);
1906 xfs_trans_log_buf(args->trans, bp,
1907 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1909 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1910 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1913 * If we removed the first entry, re-find the first used byte
1914 * in the name area. Note that if the entry was the "firstused",
1915 * then we don't have a "hole" in our block resulting from
1916 * removing the name.
1919 tmp = XFS_LBSIZE(mp);
1920 entry = xfs_attr3_leaf_entryp(leaf);
1921 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1922 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1923 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1925 if (be16_to_cpu(entry->nameidx) < tmp)
1926 tmp = be16_to_cpu(entry->nameidx);
1928 ichdr.firstused = tmp;
1929 if (!ichdr.firstused)
1930 ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
1932 ichdr.holes = 1; /* mark as needing compaction */
1934 xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1935 xfs_trans_log_buf(args->trans, bp,
1936 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1937 xfs_attr3_leaf_hdr_size(leaf)));
1940 * Check if leaf is less than 50% full, caller may want to
1941 * "join" the leaf with a sibling if so.
1943 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
1944 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
1946 return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
1950 * Move all the attribute list entries from drop_leaf into save_leaf.
1953 xfs_attr3_leaf_unbalance(
1954 struct xfs_da_state *state,
1955 struct xfs_da_state_blk *drop_blk,
1956 struct xfs_da_state_blk *save_blk)
1958 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
1959 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
1960 struct xfs_attr3_icleaf_hdr drophdr;
1961 struct xfs_attr3_icleaf_hdr savehdr;
1962 struct xfs_attr_leaf_entry *entry;
1963 struct xfs_mount *mp = state->mp;
1965 trace_xfs_attr_leaf_unbalance(state->args);
1967 drop_leaf = drop_blk->bp->b_addr;
1968 save_leaf = save_blk->bp->b_addr;
1969 xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1970 xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
1971 entry = xfs_attr3_leaf_entryp(drop_leaf);
1974 * Save last hashval from dying block for later Btree fixup.
1976 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
1979 * Check if we need a temp buffer, or can we do it in place.
1980 * Note that we don't check "leaf" for holes because we will
1981 * always be dropping it, toosmall() decided that for us already.
1983 if (savehdr.holes == 0) {
1985 * dest leaf has no holes, so we add there. May need
1986 * to make some room in the entry array.
1988 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
1989 drop_blk->bp, &drophdr)) {
1990 xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1991 save_leaf, &savehdr, 0,
1994 xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1995 save_leaf, &savehdr,
1996 savehdr.count, drophdr.count, mp);
2000 * Destination has holes, so we make a temporary copy
2001 * of the leaf and add them both to that.
2003 struct xfs_attr_leafblock *tmp_leaf;
2004 struct xfs_attr3_icleaf_hdr tmphdr;
2006 tmp_leaf = kmem_zalloc(state->blocksize, KM_SLEEP);
2009 * Copy the header into the temp leaf so that all the stuff
2010 * not in the incore header is present and gets copied back in
2011 * once we've moved all the entries.
2013 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2015 memset(&tmphdr, 0, sizeof(tmphdr));
2016 tmphdr.magic = savehdr.magic;
2017 tmphdr.forw = savehdr.forw;
2018 tmphdr.back = savehdr.back;
2019 tmphdr.firstused = state->blocksize;
2021 /* write the header to the temp buffer to initialise it */
2022 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2024 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2025 drop_blk->bp, &drophdr)) {
2026 xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2027 tmp_leaf, &tmphdr, 0,
2029 xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2030 tmp_leaf, &tmphdr, tmphdr.count,
2033 xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2034 tmp_leaf, &tmphdr, 0,
2036 xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2037 tmp_leaf, &tmphdr, tmphdr.count,
2040 memcpy(save_leaf, tmp_leaf, state->blocksize);
2041 savehdr = tmphdr; /* struct copy */
2042 kmem_free(tmp_leaf);
2045 xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2046 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2047 state->blocksize - 1);
2050 * Copy out last hashval in each block for B-tree code.
2052 entry = xfs_attr3_leaf_entryp(save_leaf);
2053 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2056 /*========================================================================
2057 * Routines used for finding things in the Btree.
2058 *========================================================================*/
2061 * Look up a name in a leaf attribute list structure.
2062 * This is the internal routine, it uses the caller's buffer.
2064 * Note that duplicate keys are allowed, but only check within the
2065 * current leaf node. The Btree code must check in adjacent leaf nodes.
2067 * Return in args->index the index into the entry[] array of either
2068 * the found entry, or where the entry should have been (insert before
2071 * Don't change the args->value unless we find the attribute.
2074 xfs_attr3_leaf_lookup_int(
2076 struct xfs_da_args *args)
2078 struct xfs_attr_leafblock *leaf;
2079 struct xfs_attr3_icleaf_hdr ichdr;
2080 struct xfs_attr_leaf_entry *entry;
2081 struct xfs_attr_leaf_entry *entries;
2082 struct xfs_attr_leaf_name_local *name_loc;
2083 struct xfs_attr_leaf_name_remote *name_rmt;
2084 xfs_dahash_t hashval;
2088 trace_xfs_attr_leaf_lookup(args);
2091 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2092 entries = xfs_attr3_leaf_entryp(leaf);
2093 ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2096 * Binary search. (note: small blocks will skip this loop)
2098 hashval = args->hashval;
2099 probe = span = ichdr.count / 2;
2100 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2102 if (be32_to_cpu(entry->hashval) < hashval)
2104 else if (be32_to_cpu(entry->hashval) > hashval)
2109 ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2110 ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2113 * Since we may have duplicate hashval's, find the first matching
2114 * hashval in the leaf.
2116 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2120 while (probe < ichdr.count &&
2121 be32_to_cpu(entry->hashval) < hashval) {
2125 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2126 args->index = probe;
2127 return XFS_ERROR(ENOATTR);
2131 * Duplicate keys may be present, so search all of them for a match.
2133 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2136 * GROT: Add code to remove incomplete entries.
2139 * If we are looking for INCOMPLETE entries, show only those.
2140 * If we are looking for complete entries, show only those.
2142 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2143 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2146 if (entry->flags & XFS_ATTR_LOCAL) {
2147 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2148 if (name_loc->namelen != args->namelen)
2150 if (memcmp(args->name, name_loc->nameval,
2151 args->namelen) != 0)
2153 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2155 args->index = probe;
2156 return XFS_ERROR(EEXIST);
2158 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2159 if (name_rmt->namelen != args->namelen)
2161 if (memcmp(args->name, name_rmt->name,
2162 args->namelen) != 0)
2164 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2166 args->index = probe;
2167 args->valuelen = be32_to_cpu(name_rmt->valuelen);
2168 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2169 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2172 return XFS_ERROR(EEXIST);
2175 args->index = probe;
2176 return XFS_ERROR(ENOATTR);
2180 * Get the value associated with an attribute name from a leaf attribute
2184 xfs_attr3_leaf_getvalue(
2186 struct xfs_da_args *args)
2188 struct xfs_attr_leafblock *leaf;
2189 struct xfs_attr3_icleaf_hdr ichdr;
2190 struct xfs_attr_leaf_entry *entry;
2191 struct xfs_attr_leaf_name_local *name_loc;
2192 struct xfs_attr_leaf_name_remote *name_rmt;
2196 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2197 ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2198 ASSERT(args->index < ichdr.count);
2200 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2201 if (entry->flags & XFS_ATTR_LOCAL) {
2202 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2203 ASSERT(name_loc->namelen == args->namelen);
2204 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2205 valuelen = be16_to_cpu(name_loc->valuelen);
2206 if (args->flags & ATTR_KERNOVAL) {
2207 args->valuelen = valuelen;
2210 if (args->valuelen < valuelen) {
2211 args->valuelen = valuelen;
2212 return XFS_ERROR(ERANGE);
2214 args->valuelen = valuelen;
2215 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2217 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2218 ASSERT(name_rmt->namelen == args->namelen);
2219 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2220 valuelen = be32_to_cpu(name_rmt->valuelen);
2221 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2222 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2224 if (args->flags & ATTR_KERNOVAL) {
2225 args->valuelen = valuelen;
2228 if (args->valuelen < valuelen) {
2229 args->valuelen = valuelen;
2230 return XFS_ERROR(ERANGE);
2232 args->valuelen = valuelen;
2237 /*========================================================================
2239 *========================================================================*/
2242 * Move the indicated entries from one leaf to another.
2243 * NOTE: this routine modifies both source and destination leaves.
2247 xfs_attr3_leaf_moveents(
2248 struct xfs_attr_leafblock *leaf_s,
2249 struct xfs_attr3_icleaf_hdr *ichdr_s,
2251 struct xfs_attr_leafblock *leaf_d,
2252 struct xfs_attr3_icleaf_hdr *ichdr_d,
2255 struct xfs_mount *mp)
2257 struct xfs_attr_leaf_entry *entry_s;
2258 struct xfs_attr_leaf_entry *entry_d;
2264 * Check for nothing to do.
2270 * Set up environment.
2272 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2273 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2274 ASSERT(ichdr_s->magic == ichdr_d->magic);
2275 ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2276 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2277 + xfs_attr3_leaf_hdr_size(leaf_s));
2278 ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2279 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2280 + xfs_attr3_leaf_hdr_size(leaf_d));
2282 ASSERT(start_s < ichdr_s->count);
2283 ASSERT(start_d <= ichdr_d->count);
2284 ASSERT(count <= ichdr_s->count);
2288 * Move the entries in the destination leaf up to make a hole?
2290 if (start_d < ichdr_d->count) {
2291 tmp = ichdr_d->count - start_d;
2292 tmp *= sizeof(xfs_attr_leaf_entry_t);
2293 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2294 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2295 memmove(entry_d, entry_s, tmp);
2299 * Copy all entry's in the same (sorted) order,
2300 * but allocate attribute info packed and in sequence.
2302 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2303 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2305 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2306 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2307 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2310 * Code to drop INCOMPLETE entries. Difficult to use as we
2311 * may also need to change the insertion index. Code turned
2312 * off for 6.2, should be revisited later.
2314 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2315 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2316 ichdr_s->usedbytes -= tmp;
2317 ichdr_s->count -= 1;
2318 entry_d--; /* to compensate for ++ in loop hdr */
2320 if ((start_s + i) < offset)
2321 result++; /* insertion index adjustment */
2324 ichdr_d->firstused -= tmp;
2325 /* both on-disk, don't endian flip twice */
2326 entry_d->hashval = entry_s->hashval;
2327 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2328 entry_d->flags = entry_s->flags;
2329 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2331 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2332 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2333 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2335 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2336 ichdr_s->usedbytes -= tmp;
2337 ichdr_d->usedbytes += tmp;
2338 ichdr_s->count -= 1;
2339 ichdr_d->count += 1;
2340 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2341 + xfs_attr3_leaf_hdr_size(leaf_d);
2342 ASSERT(ichdr_d->firstused >= tmp);
2349 * Zero out the entries we just copied.
2351 if (start_s == ichdr_s->count) {
2352 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2353 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2354 ASSERT(((char *)entry_s + tmp) <=
2355 ((char *)leaf_s + XFS_LBSIZE(mp)));
2356 memset(entry_s, 0, tmp);
2359 * Move the remaining entries down to fill the hole,
2360 * then zero the entries at the top.
2362 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2363 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2364 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2365 memmove(entry_d, entry_s, tmp);
2367 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2368 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2369 ASSERT(((char *)entry_s + tmp) <=
2370 ((char *)leaf_s + XFS_LBSIZE(mp)));
2371 memset(entry_s, 0, tmp);
2375 * Fill in the freemap information
2377 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2378 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2379 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2380 ichdr_d->freemap[1].base = 0;
2381 ichdr_d->freemap[2].base = 0;
2382 ichdr_d->freemap[1].size = 0;
2383 ichdr_d->freemap[2].size = 0;
2384 ichdr_s->holes = 1; /* leaf may not be compact */
2388 * Pick up the last hashvalue from a leaf block.
2391 xfs_attr_leaf_lasthash(
2395 struct xfs_attr3_icleaf_hdr ichdr;
2396 struct xfs_attr_leaf_entry *entries;
2398 xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2399 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2401 *count = ichdr.count;
2404 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2408 * Calculate the number of bytes used to store the indicated attribute
2409 * (whether local or remote only calculate bytes in this block).
2412 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2414 struct xfs_attr_leaf_entry *entries;
2415 xfs_attr_leaf_name_local_t *name_loc;
2416 xfs_attr_leaf_name_remote_t *name_rmt;
2419 entries = xfs_attr3_leaf_entryp(leaf);
2420 if (entries[index].flags & XFS_ATTR_LOCAL) {
2421 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2422 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2423 be16_to_cpu(name_loc->valuelen));
2425 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2426 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2432 * Calculate the number of bytes that would be required to store the new
2433 * attribute (whether local or remote only calculate bytes in this block).
2434 * This routine decides as a side effect whether the attribute will be
2435 * a "local" or a "remote" attribute.
2438 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2442 size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2443 if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2448 size = xfs_attr_leaf_entsize_remote(namelen);
2457 /*========================================================================
2458 * Manage the INCOMPLETE flag in a leaf entry
2459 *========================================================================*/
2462 * Clear the INCOMPLETE flag on an entry in a leaf block.
2465 xfs_attr3_leaf_clearflag(
2466 struct xfs_da_args *args)
2468 struct xfs_attr_leafblock *leaf;
2469 struct xfs_attr_leaf_entry *entry;
2470 struct xfs_attr_leaf_name_remote *name_rmt;
2474 struct xfs_attr3_icleaf_hdr ichdr;
2475 xfs_attr_leaf_name_local_t *name_loc;
2480 trace_xfs_attr_leaf_clearflag(args);
2482 * Set up the operation.
2484 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2489 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2490 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2493 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2494 ASSERT(args->index < ichdr.count);
2495 ASSERT(args->index >= 0);
2497 if (entry->flags & XFS_ATTR_LOCAL) {
2498 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2499 namelen = name_loc->namelen;
2500 name = (char *)name_loc->nameval;
2502 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2503 namelen = name_rmt->namelen;
2504 name = (char *)name_rmt->name;
2506 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2507 ASSERT(namelen == args->namelen);
2508 ASSERT(memcmp(name, args->name, namelen) == 0);
2511 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2512 xfs_trans_log_buf(args->trans, bp,
2513 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2515 if (args->rmtblkno) {
2516 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2517 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2518 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2519 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2520 xfs_trans_log_buf(args->trans, bp,
2521 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2525 * Commit the flag value change and start the next trans in series.
2527 return xfs_trans_roll(&args->trans, args->dp);
2531 * Set the INCOMPLETE flag on an entry in a leaf block.
2534 xfs_attr3_leaf_setflag(
2535 struct xfs_da_args *args)
2537 struct xfs_attr_leafblock *leaf;
2538 struct xfs_attr_leaf_entry *entry;
2539 struct xfs_attr_leaf_name_remote *name_rmt;
2543 struct xfs_attr3_icleaf_hdr ichdr;
2546 trace_xfs_attr_leaf_setflag(args);
2549 * Set up the operation.
2551 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2557 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2558 ASSERT(args->index < ichdr.count);
2559 ASSERT(args->index >= 0);
2561 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2563 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2564 entry->flags |= XFS_ATTR_INCOMPLETE;
2565 xfs_trans_log_buf(args->trans, bp,
2566 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2567 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2568 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2569 name_rmt->valueblk = 0;
2570 name_rmt->valuelen = 0;
2571 xfs_trans_log_buf(args->trans, bp,
2572 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2576 * Commit the flag value change and start the next trans in series.
2578 return xfs_trans_roll(&args->trans, args->dp);
2582 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2583 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2584 * entry given by args->blkno2/index2.
2586 * Note that they could be in different blocks, or in the same block.
2589 xfs_attr3_leaf_flipflags(
2590 struct xfs_da_args *args)
2592 struct xfs_attr_leafblock *leaf1;
2593 struct xfs_attr_leafblock *leaf2;
2594 struct xfs_attr_leaf_entry *entry1;
2595 struct xfs_attr_leaf_entry *entry2;
2596 struct xfs_attr_leaf_name_remote *name_rmt;
2597 struct xfs_buf *bp1;
2598 struct xfs_buf *bp2;
2601 struct xfs_attr3_icleaf_hdr ichdr1;
2602 struct xfs_attr3_icleaf_hdr ichdr2;
2603 xfs_attr_leaf_name_local_t *name_loc;
2604 int namelen1, namelen2;
2605 char *name1, *name2;
2608 trace_xfs_attr_leaf_flipflags(args);
2611 * Read the block containing the "old" attr
2613 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2618 * Read the block containing the "new" attr, if it is different
2620 if (args->blkno2 != args->blkno) {
2621 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2629 leaf1 = bp1->b_addr;
2630 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2632 leaf2 = bp2->b_addr;
2633 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2636 xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2637 ASSERT(args->index < ichdr1.count);
2638 ASSERT(args->index >= 0);
2640 xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2641 ASSERT(args->index2 < ichdr2.count);
2642 ASSERT(args->index2 >= 0);
2644 if (entry1->flags & XFS_ATTR_LOCAL) {
2645 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2646 namelen1 = name_loc->namelen;
2647 name1 = (char *)name_loc->nameval;
2649 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2650 namelen1 = name_rmt->namelen;
2651 name1 = (char *)name_rmt->name;
2653 if (entry2->flags & XFS_ATTR_LOCAL) {
2654 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2655 namelen2 = name_loc->namelen;
2656 name2 = (char *)name_loc->nameval;
2658 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2659 namelen2 = name_rmt->namelen;
2660 name2 = (char *)name_rmt->name;
2662 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2663 ASSERT(namelen1 == namelen2);
2664 ASSERT(memcmp(name1, name2, namelen1) == 0);
2667 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2668 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2670 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2671 xfs_trans_log_buf(args->trans, bp1,
2672 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2673 if (args->rmtblkno) {
2674 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2675 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2676 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2677 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2678 xfs_trans_log_buf(args->trans, bp1,
2679 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2682 entry2->flags |= XFS_ATTR_INCOMPLETE;
2683 xfs_trans_log_buf(args->trans, bp2,
2684 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2685 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2686 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2687 name_rmt->valueblk = 0;
2688 name_rmt->valuelen = 0;
2689 xfs_trans_log_buf(args->trans, bp2,
2690 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2694 * Commit the flag value change and start the next trans in series.
2696 error = xfs_trans_roll(&args->trans, args->dp);