2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_alloc.h"
36 #include "xfs_btree.h"
37 #include "xfs_dir_sf.h"
38 #include "xfs_dir2_sf.h"
39 #include "xfs_attr_sf.h"
40 #include "xfs_dinode.h"
41 #include "xfs_inode.h"
42 #include "xfs_inode_item.h"
45 #include "xfs_attr_leaf.h"
46 #include "xfs_error.h"
51 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
54 /*========================================================================
55 * Function prototypes for the kernel.
56 *========================================================================*/
59 * Routines used for growing the Btree.
61 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
63 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
65 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
66 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
67 xfs_da_state_blk_t *blk1,
68 xfs_da_state_blk_t *blk2);
69 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
70 xfs_da_state_blk_t *leaf_blk_1,
71 xfs_da_state_blk_t *leaf_blk_2,
72 int *number_entries_in_blk1,
73 int *number_usedbytes_in_blk1);
76 * Routines used for shrinking the Btree.
78 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
79 xfs_dabuf_t *bp, int level);
80 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
82 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
83 xfs_dablk_t blkno, int blkcnt);
88 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
90 xfs_attr_leafblock_t *dst_leaf,
91 int dst_start, int move_count,
93 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
94 STATIC int xfs_attr_put_listent(xfs_attr_list_context_t *context,
95 attrnames_t *, char *name, int namelen,
99 /*========================================================================
100 * External routines when attribute fork size < XFS_LITINO(mp).
101 *========================================================================*/
104 * Query whether the requested number of additional bytes of extended
105 * attribute space will be able to fit inline.
106 * Returns zero if not, else the di_forkoff fork offset to be used in the
107 * literal area for attribute data once the new bytes have been added.
109 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
110 * special case for dev/uuid inodes, they have fixed size data forks.
113 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
116 int minforkoff; /* lower limit on valid forkoff locations */
117 int maxforkoff; /* upper limit on valid forkoff locations */
118 xfs_mount_t *mp = dp->i_mount;
120 offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
122 switch (dp->i_d.di_format) {
123 case XFS_DINODE_FMT_DEV:
124 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
125 return (offset >= minforkoff) ? minforkoff : 0;
126 case XFS_DINODE_FMT_UUID:
127 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
128 return (offset >= minforkoff) ? minforkoff : 0;
131 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) {
132 if (bytes <= XFS_IFORK_ASIZE(dp))
133 return mp->m_attroffset >> 3;
137 /* data fork btree root can have at least this many key/ptr pairs */
138 minforkoff = MAX(dp->i_df.if_bytes, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
139 minforkoff = roundup(minforkoff, 8) >> 3;
141 /* attr fork btree root can have at least this many key/ptr pairs */
142 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
143 maxforkoff = maxforkoff >> 3; /* rounded down */
145 if (offset >= minforkoff && offset < maxforkoff)
147 if (offset >= maxforkoff)
153 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
156 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
160 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
161 !(XFS_SB_VERSION_HASATTR2(&mp->m_sb))) {
163 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
164 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
165 XFS_SB_UNLOCK(mp, s);
166 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
168 XFS_SB_UNLOCK(mp, s);
173 * Create the initial contents of a shortform attribute list.
176 xfs_attr_shortform_create(xfs_da_args_t *args)
178 xfs_attr_sf_hdr_t *hdr;
186 ASSERT(ifp->if_bytes == 0);
187 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
188 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
189 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
190 ifp->if_flags |= XFS_IFINLINE;
192 ASSERT(ifp->if_flags & XFS_IFINLINE);
194 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
195 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
197 hdr->totsize = cpu_to_be16(sizeof(*hdr));
198 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
202 * Add a name/value pair to the shortform attribute list.
203 * Overflow from the inode has already been checked for.
206 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
208 xfs_attr_shortform_t *sf;
209 xfs_attr_sf_entry_t *sfe;
217 dp->i_d.di_forkoff = forkoff;
218 dp->i_df.if_ext_max =
219 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
220 dp->i_afp->if_ext_max =
221 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
224 ASSERT(ifp->if_flags & XFS_IFINLINE);
225 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
227 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
229 if (sfe->namelen != args->namelen)
231 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
233 if (((args->flags & ATTR_SECURE) != 0) !=
234 ((sfe->flags & XFS_ATTR_SECURE) != 0))
236 if (((args->flags & ATTR_ROOT) != 0) !=
237 ((sfe->flags & XFS_ATTR_ROOT) != 0))
243 offset = (char *)sfe - (char *)sf;
244 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
245 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
246 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
247 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
249 sfe->namelen = args->namelen;
250 sfe->valuelen = args->valuelen;
251 sfe->flags = (args->flags & ATTR_SECURE) ? XFS_ATTR_SECURE :
252 ((args->flags & ATTR_ROOT) ? XFS_ATTR_ROOT : 0);
253 memcpy(sfe->nameval, args->name, args->namelen);
254 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
256 be16_add(&sf->hdr.totsize, size);
257 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
259 xfs_sbversion_add_attr2(mp, args->trans);
263 * Remove an attribute from the shortform attribute list structure.
266 xfs_attr_shortform_remove(xfs_da_args_t *args)
268 xfs_attr_shortform_t *sf;
269 xfs_attr_sf_entry_t *sfe;
270 int base, size=0, end, totsize, i;
276 base = sizeof(xfs_attr_sf_hdr_t);
277 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
280 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
282 size = XFS_ATTR_SF_ENTSIZE(sfe);
283 if (sfe->namelen != args->namelen)
285 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
287 if (((args->flags & ATTR_SECURE) != 0) !=
288 ((sfe->flags & XFS_ATTR_SECURE) != 0))
290 if (((args->flags & ATTR_ROOT) != 0) !=
291 ((sfe->flags & XFS_ATTR_ROOT) != 0))
296 return(XFS_ERROR(ENOATTR));
299 * Fix up the attribute fork data, covering the hole
302 totsize = be16_to_cpu(sf->hdr.totsize);
304 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
306 be16_add(&sf->hdr.totsize, -size);
309 * Fix up the start offset of the attribute fork
312 if (totsize == sizeof(xfs_attr_sf_hdr_t) && !args->addname &&
313 (mp->m_flags & XFS_MOUNT_ATTR2)) {
315 * Last attribute now removed, revert to original
316 * inode format making all literal area available
317 * to the data fork once more.
319 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
320 dp->i_d.di_forkoff = 0;
321 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
322 ASSERT(dp->i_d.di_anextents == 0);
323 ASSERT(dp->i_afp == NULL);
324 dp->i_df.if_ext_max =
325 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
326 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
328 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
329 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
330 ASSERT(dp->i_d.di_forkoff);
331 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || args->addname ||
332 !(mp->m_flags & XFS_MOUNT_ATTR2));
333 dp->i_afp->if_ext_max =
334 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
335 dp->i_df.if_ext_max =
336 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
337 xfs_trans_log_inode(args->trans, dp,
338 XFS_ILOG_CORE | XFS_ILOG_ADATA);
341 xfs_sbversion_add_attr2(mp, args->trans);
347 * Look up a name in a shortform attribute list structure.
351 xfs_attr_shortform_lookup(xfs_da_args_t *args)
353 xfs_attr_shortform_t *sf;
354 xfs_attr_sf_entry_t *sfe;
358 ifp = args->dp->i_afp;
359 ASSERT(ifp->if_flags & XFS_IFINLINE);
360 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
362 for (i = 0; i < sf->hdr.count;
363 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
364 if (sfe->namelen != args->namelen)
366 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
368 if (((args->flags & ATTR_SECURE) != 0) !=
369 ((sfe->flags & XFS_ATTR_SECURE) != 0))
371 if (((args->flags & ATTR_ROOT) != 0) !=
372 ((sfe->flags & XFS_ATTR_ROOT) != 0))
374 return(XFS_ERROR(EEXIST));
376 return(XFS_ERROR(ENOATTR));
380 * Look up a name in a shortform attribute list structure.
384 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
386 xfs_attr_shortform_t *sf;
387 xfs_attr_sf_entry_t *sfe;
390 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
391 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
393 for (i = 0; i < sf->hdr.count;
394 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
395 if (sfe->namelen != args->namelen)
397 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
399 if (((args->flags & ATTR_SECURE) != 0) !=
400 ((sfe->flags & XFS_ATTR_SECURE) != 0))
402 if (((args->flags & ATTR_ROOT) != 0) !=
403 ((sfe->flags & XFS_ATTR_ROOT) != 0))
405 if (args->flags & ATTR_KERNOVAL) {
406 args->valuelen = sfe->valuelen;
407 return(XFS_ERROR(EEXIST));
409 if (args->valuelen < sfe->valuelen) {
410 args->valuelen = sfe->valuelen;
411 return(XFS_ERROR(ERANGE));
413 args->valuelen = sfe->valuelen;
414 memcpy(args->value, &sfe->nameval[args->namelen],
416 return(XFS_ERROR(EEXIST));
418 return(XFS_ERROR(ENOATTR));
422 * Convert from using the shortform to the leaf.
425 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
428 xfs_attr_shortform_t *sf;
429 xfs_attr_sf_entry_t *sfe;
439 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
440 size = be16_to_cpu(sf->hdr.totsize);
441 tmpbuffer = kmem_alloc(size, KM_SLEEP);
442 ASSERT(tmpbuffer != NULL);
443 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
444 sf = (xfs_attr_shortform_t *)tmpbuffer;
446 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
448 error = xfs_da_grow_inode(args, &blkno);
451 * If we hit an IO error middle of the transaction inside
452 * grow_inode(), we may have inconsistent data. Bail out.
456 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
457 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
462 error = xfs_attr_leaf_create(args, blkno, &bp);
464 error = xfs_da_shrink_inode(args, 0, bp);
468 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
469 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
473 memset((char *)&nargs, 0, sizeof(nargs));
475 nargs.firstblock = args->firstblock;
476 nargs.flist = args->flist;
477 nargs.total = args->total;
478 nargs.whichfork = XFS_ATTR_FORK;
479 nargs.trans = args->trans;
483 for (i = 0; i < sf->hdr.count; i++) {
484 nargs.name = (char *)sfe->nameval;
485 nargs.namelen = sfe->namelen;
486 nargs.value = (char *)&sfe->nameval[nargs.namelen];
487 nargs.valuelen = sfe->valuelen;
488 nargs.hashval = xfs_da_hashname((char *)sfe->nameval,
490 nargs.flags = (sfe->flags & XFS_ATTR_SECURE) ? ATTR_SECURE :
491 ((sfe->flags & XFS_ATTR_ROOT) ? ATTR_ROOT : 0);
492 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
493 ASSERT(error == ENOATTR);
494 error = xfs_attr_leaf_add(bp, &nargs);
495 ASSERT(error != ENOSPC);
498 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
505 kmem_free(tmpbuffer, size);
510 xfs_attr_shortform_compare(const void *a, const void *b)
512 xfs_attr_sf_sort_t *sa, *sb;
514 sa = (xfs_attr_sf_sort_t *)a;
515 sb = (xfs_attr_sf_sort_t *)b;
516 if (sa->hash < sb->hash) {
518 } else if (sa->hash > sb->hash) {
521 return(sa->entno - sb->entno);
526 * Copy out entries of shortform attribute lists for attr_list().
527 * Shortform attribute lists are not stored in hashval sorted order.
528 * If the output buffer is not large enough to hold them all, then we
529 * we have to calculate each entries' hashvalue and sort them before
530 * we can begin returning them to the user.
534 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
536 attrlist_cursor_kern_t *cursor;
537 xfs_attr_sf_sort_t *sbuf, *sbp;
538 xfs_attr_shortform_t *sf;
539 xfs_attr_sf_entry_t *sfe;
541 int sbsize, nsbuf, count, i;
543 ASSERT(context != NULL);
546 ASSERT(dp->i_afp != NULL);
547 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
551 cursor = context->cursor;
552 ASSERT(cursor != NULL);
554 xfs_attr_trace_l_c("sf start", context);
557 * If the buffer is large enough, do not bother with sorting.
558 * Note the generous fudge factor of 16 overhead bytes per entry.
560 if ((dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize) {
561 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
564 if (((context->flags & ATTR_SECURE) != 0) !=
565 ((sfe->flags & XFS_ATTR_SECURE) != 0) &&
566 !(context->flags & ATTR_KERNORMALS)) {
567 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
570 if (((context->flags & ATTR_ROOT) != 0) !=
571 ((sfe->flags & XFS_ATTR_ROOT) != 0) &&
572 !(context->flags & ATTR_KERNROOTLS)) {
573 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
576 namesp = (sfe->flags & XFS_ATTR_SECURE) ? &attr_secure:
577 ((sfe->flags & XFS_ATTR_ROOT) ? &attr_trusted :
579 if (context->flags & ATTR_KERNOVAL) {
580 ASSERT(context->flags & ATTR_KERNAMELS);
581 context->count += namesp->attr_namelen +
585 if (xfs_attr_put_listent(context, namesp,
586 (char *)sfe->nameval,
591 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
593 xfs_attr_trace_l_c("sf big-gulp", context);
598 * It didn't all fit, so we have to sort everything on hashval.
600 sbsize = sf->hdr.count * sizeof(*sbuf);
601 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP);
604 * Scan the attribute list for the rest of the entries, storing
605 * the relevant info from only those that match into a buffer.
608 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
610 ((char *)sfe < (char *)sf) ||
611 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
612 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
614 context->dp->i_mount, sfe);
615 xfs_attr_trace_l_c("sf corrupted", context);
616 kmem_free(sbuf, sbsize);
617 return XFS_ERROR(EFSCORRUPTED);
619 if (((context->flags & ATTR_SECURE) != 0) !=
620 ((sfe->flags & XFS_ATTR_SECURE) != 0) &&
621 !(context->flags & ATTR_KERNORMALS)) {
622 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
625 if (((context->flags & ATTR_ROOT) != 0) !=
626 ((sfe->flags & XFS_ATTR_ROOT) != 0) &&
627 !(context->flags & ATTR_KERNROOTLS)) {
628 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
632 sbp->hash = xfs_da_hashname((char *)sfe->nameval, sfe->namelen);
633 sbp->name = (char *)sfe->nameval;
634 sbp->namelen = sfe->namelen;
635 /* These are bytes, and both on-disk, don't endian-flip */
636 sbp->valuelen = sfe->valuelen;
637 sbp->flags = sfe->flags;
638 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
644 * Sort the entries on hash then entno.
646 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
649 * Re-find our place IN THE SORTED LIST.
654 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
655 if (sbp->hash == cursor->hashval) {
656 if (cursor->offset == count) {
660 } else if (sbp->hash > cursor->hashval) {
665 kmem_free(sbuf, sbsize);
666 xfs_attr_trace_l_c("blk end", context);
671 * Loop putting entries into the user buffer.
673 for ( ; i < nsbuf; i++, sbp++) {
676 namesp = (sbp->flags & XFS_ATTR_SECURE) ? &attr_secure :
677 ((sbp->flags & XFS_ATTR_ROOT) ? &attr_trusted :
680 if (cursor->hashval != sbp->hash) {
681 cursor->hashval = sbp->hash;
684 if (context->flags & ATTR_KERNOVAL) {
685 ASSERT(context->flags & ATTR_KERNAMELS);
686 context->count += namesp->attr_namelen +
689 if (xfs_attr_put_listent(context, namesp,
690 sbp->name, sbp->namelen,
697 kmem_free(sbuf, sbsize);
698 xfs_attr_trace_l_c("sf E-O-F", context);
703 * Check a leaf attribute block to see if all the entries would fit into
704 * a shortform attribute list.
707 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
709 xfs_attr_leafblock_t *leaf;
710 xfs_attr_leaf_entry_t *entry;
711 xfs_attr_leaf_name_local_t *name_loc;
715 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
717 entry = &leaf->entries[0];
718 bytes = sizeof(struct xfs_attr_sf_hdr);
719 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
720 if (entry->flags & XFS_ATTR_INCOMPLETE)
721 continue; /* don't copy partial entries */
722 if (!(entry->flags & XFS_ATTR_LOCAL))
724 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
725 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
727 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
729 bytes += sizeof(struct xfs_attr_sf_entry)-1
731 + be16_to_cpu(name_loc->valuelen);
733 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
734 (bytes == sizeof(struct xfs_attr_sf_hdr)))
736 return(xfs_attr_shortform_bytesfit(dp, bytes));
740 * Convert a leaf attribute list to shortform attribute list
743 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
745 xfs_attr_leafblock_t *leaf;
746 xfs_attr_leaf_entry_t *entry;
747 xfs_attr_leaf_name_local_t *name_loc;
754 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
755 ASSERT(tmpbuffer != NULL);
758 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
759 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
760 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
761 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
764 * Clean out the prior contents of the attribute list.
766 error = xfs_da_shrink_inode(args, 0, bp);
771 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
774 * Last attribute was removed, revert to original
775 * inode format making all literal area available
776 * to the data fork once more.
778 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
779 dp->i_d.di_forkoff = 0;
780 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
781 ASSERT(dp->i_d.di_anextents == 0);
782 ASSERT(dp->i_afp == NULL);
783 dp->i_df.if_ext_max =
784 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
785 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
789 xfs_attr_shortform_create(args);
792 * Copy the attributes
794 memset((char *)&nargs, 0, sizeof(nargs));
796 nargs.firstblock = args->firstblock;
797 nargs.flist = args->flist;
798 nargs.total = args->total;
799 nargs.whichfork = XFS_ATTR_FORK;
800 nargs.trans = args->trans;
802 entry = &leaf->entries[0];
803 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
804 if (entry->flags & XFS_ATTR_INCOMPLETE)
805 continue; /* don't copy partial entries */
808 ASSERT(entry->flags & XFS_ATTR_LOCAL);
809 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
810 nargs.name = (char *)name_loc->nameval;
811 nargs.namelen = name_loc->namelen;
812 nargs.value = (char *)&name_loc->nameval[nargs.namelen];
813 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
814 nargs.hashval = be32_to_cpu(entry->hashval);
815 nargs.flags = (entry->flags & XFS_ATTR_SECURE) ? ATTR_SECURE :
816 ((entry->flags & XFS_ATTR_ROOT) ? ATTR_ROOT : 0);
817 xfs_attr_shortform_add(&nargs, forkoff);
822 kmem_free(tmpbuffer, XFS_LBSIZE(dp->i_mount));
827 * Convert from using a single leaf to a root node and a leaf.
830 xfs_attr_leaf_to_node(xfs_da_args_t *args)
832 xfs_attr_leafblock_t *leaf;
833 xfs_da_intnode_t *node;
835 xfs_dabuf_t *bp1, *bp2;
841 error = xfs_da_grow_inode(args, &blkno);
844 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
850 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
855 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
856 xfs_da_buf_done(bp1);
858 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
861 * Set up the new root node.
863 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
868 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
869 /* both on-disk, don't endian-flip twice */
870 node->btree[0].hashval =
871 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
872 node->btree[0].before = cpu_to_be32(blkno);
873 node->hdr.count = cpu_to_be16(1);
874 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
878 xfs_da_buf_done(bp1);
880 xfs_da_buf_done(bp2);
885 /*========================================================================
886 * Routines used for growing the Btree.
887 *========================================================================*/
890 * Create the initial contents of a leaf attribute list
891 * or a leaf in a node attribute list.
894 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
896 xfs_attr_leafblock_t *leaf;
897 xfs_attr_leaf_hdr_t *hdr;
904 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
910 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
912 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
913 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
914 if (!hdr->firstused) {
915 hdr->firstused = cpu_to_be16(
916 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
919 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
920 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
921 sizeof(xfs_attr_leaf_hdr_t));
923 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
930 * Split the leaf node, rebalance, then add the new entry.
933 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
934 xfs_da_state_blk_t *newblk)
940 * Allocate space for a new leaf node.
942 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
943 error = xfs_da_grow_inode(state->args, &blkno);
946 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
949 newblk->blkno = blkno;
950 newblk->magic = XFS_ATTR_LEAF_MAGIC;
953 * Rebalance the entries across the two leaves.
954 * NOTE: rebalance() currently depends on the 2nd block being empty.
956 xfs_attr_leaf_rebalance(state, oldblk, newblk);
957 error = xfs_da_blk_link(state, oldblk, newblk);
962 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
963 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
964 * "new" attrs info. Will need the "old" info to remove it later.
966 * Insert the "new" entry in the correct block.
969 error = xfs_attr_leaf_add(oldblk->bp, state->args);
971 error = xfs_attr_leaf_add(newblk->bp, state->args);
974 * Update last hashval in each block since we added the name.
976 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
977 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
982 * Add a name to the leaf attribute list structure.
985 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
987 xfs_attr_leafblock_t *leaf;
988 xfs_attr_leaf_hdr_t *hdr;
989 xfs_attr_leaf_map_t *map;
990 int tablesize, entsize, sum, tmp, i;
993 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
994 ASSERT((args->index >= 0)
995 && (args->index <= be16_to_cpu(leaf->hdr.count)));
997 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
998 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1001 * Search through freemap for first-fit on new name length.
1002 * (may need to figure in size of entry struct too)
1004 tablesize = (be16_to_cpu(hdr->count) + 1)
1005 * sizeof(xfs_attr_leaf_entry_t)
1006 + sizeof(xfs_attr_leaf_hdr_t);
1007 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1008 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1009 if (tablesize > be16_to_cpu(hdr->firstused)) {
1010 sum += be16_to_cpu(map->size);
1014 continue; /* no space in this map */
1016 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1017 tmp += sizeof(xfs_attr_leaf_entry_t);
1018 if (be16_to_cpu(map->size) >= tmp) {
1019 tmp = xfs_attr_leaf_add_work(bp, args, i);
1022 sum += be16_to_cpu(map->size);
1026 * If there are no holes in the address space of the block,
1027 * and we don't have enough freespace, then compaction will do us
1028 * no good and we should just give up.
1030 if (!hdr->holes && (sum < entsize))
1031 return(XFS_ERROR(ENOSPC));
1034 * Compact the entries to coalesce free space.
1035 * This may change the hdr->count via dropping INCOMPLETE entries.
1037 xfs_attr_leaf_compact(args->trans, bp);
1040 * After compaction, the block is guaranteed to have only one
1041 * free region, in freemap[0]. If it is not big enough, give up.
1043 if (be16_to_cpu(hdr->freemap[0].size)
1044 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1045 return(XFS_ERROR(ENOSPC));
1047 return(xfs_attr_leaf_add_work(bp, args, 0));
1051 * Add a name to a leaf attribute list structure.
1054 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1056 xfs_attr_leafblock_t *leaf;
1057 xfs_attr_leaf_hdr_t *hdr;
1058 xfs_attr_leaf_entry_t *entry;
1059 xfs_attr_leaf_name_local_t *name_loc;
1060 xfs_attr_leaf_name_remote_t *name_rmt;
1061 xfs_attr_leaf_map_t *map;
1066 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1068 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1069 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1072 * Force open some space in the entry array and fill it in.
1074 entry = &leaf->entries[args->index];
1075 if (args->index < be16_to_cpu(hdr->count)) {
1076 tmp = be16_to_cpu(hdr->count) - args->index;
1077 tmp *= sizeof(xfs_attr_leaf_entry_t);
1078 memmove((char *)(entry+1), (char *)entry, tmp);
1079 xfs_da_log_buf(args->trans, bp,
1080 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1082 be16_add(&hdr->count, 1);
1085 * Allocate space for the new string (at the end of the run).
1087 map = &hdr->freemap[mapindex];
1088 mp = args->trans->t_mountp;
1089 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1090 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1091 ASSERT(be16_to_cpu(map->size) >=
1092 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1093 mp->m_sb.sb_blocksize, NULL));
1094 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1095 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1096 be16_add(&map->size,
1097 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1098 mp->m_sb.sb_blocksize, &tmp));
1099 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1100 be16_to_cpu(map->size));
1101 entry->hashval = cpu_to_be32(args->hashval);
1102 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1103 entry->flags |= (args->flags & ATTR_SECURE) ? XFS_ATTR_SECURE :
1104 ((args->flags & ATTR_ROOT) ? XFS_ATTR_ROOT : 0);
1106 entry->flags |= XFS_ATTR_INCOMPLETE;
1107 if ((args->blkno2 == args->blkno) &&
1108 (args->index2 <= args->index)) {
1112 xfs_da_log_buf(args->trans, bp,
1113 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1114 ASSERT((args->index == 0) ||
1115 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1116 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1117 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1120 * Copy the attribute name and value into the new space.
1122 * For "remote" attribute values, simply note that we need to
1123 * allocate space for the "remote" value. We can't actually
1124 * allocate the extents in this transaction, and we can't decide
1125 * which blocks they should be as we might allocate more blocks
1126 * as part of this transaction (a split operation for example).
1128 if (entry->flags & XFS_ATTR_LOCAL) {
1129 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
1130 name_loc->namelen = args->namelen;
1131 name_loc->valuelen = cpu_to_be16(args->valuelen);
1132 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1133 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1134 be16_to_cpu(name_loc->valuelen));
1136 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
1137 name_rmt->namelen = args->namelen;
1138 memcpy((char *)name_rmt->name, args->name, args->namelen);
1139 entry->flags |= XFS_ATTR_INCOMPLETE;
1141 name_rmt->valuelen = 0;
1142 name_rmt->valueblk = 0;
1144 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1146 xfs_da_log_buf(args->trans, bp,
1147 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1148 xfs_attr_leaf_entsize(leaf, args->index)));
1151 * Update the control info for this leaf node
1153 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1154 /* both on-disk, don't endian-flip twice */
1155 hdr->firstused = entry->nameidx;
1157 ASSERT(be16_to_cpu(hdr->firstused) >=
1158 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1159 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1160 + sizeof(xfs_attr_leaf_hdr_t);
1161 map = &hdr->freemap[0];
1162 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1163 if (be16_to_cpu(map->base) == tmp) {
1164 be16_add(&map->base, sizeof(xfs_attr_leaf_entry_t));
1165 be16_add(&map->size,
1166 -((int)sizeof(xfs_attr_leaf_entry_t)));
1169 be16_add(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1170 xfs_da_log_buf(args->trans, bp,
1171 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1176 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1179 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1181 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1182 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1186 mp = trans->t_mountp;
1187 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1188 ASSERT(tmpbuffer != NULL);
1189 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1190 memset(bp->data, 0, XFS_LBSIZE(mp));
1193 * Copy basic information
1195 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1197 hdr_s = &leaf_s->hdr;
1198 hdr_d = &leaf_d->hdr;
1199 hdr_d->info = hdr_s->info; /* struct copy */
1200 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1201 /* handle truncation gracefully */
1202 if (!hdr_d->firstused) {
1203 hdr_d->firstused = cpu_to_be16(
1204 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1206 hdr_d->usedbytes = 0;
1209 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1210 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1211 sizeof(xfs_attr_leaf_hdr_t));
1214 * Copy all entry's in the same (sorted) order,
1215 * but allocate name/value pairs packed and in sequence.
1217 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1218 be16_to_cpu(hdr_s->count), mp);
1219 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1221 kmem_free(tmpbuffer, XFS_LBSIZE(mp));
1225 * Redistribute the attribute list entries between two leaf nodes,
1226 * taking into account the size of the new entry.
1228 * NOTE: if new block is empty, then it will get the upper half of the
1229 * old block. At present, all (one) callers pass in an empty second block.
1231 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1232 * to match what it is doing in splitting the attribute leaf block. Those
1233 * values are used in "atomic rename" operations on attributes. Note that
1234 * the "new" and "old" values can end up in different blocks.
1237 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1238 xfs_da_state_blk_t *blk2)
1240 xfs_da_args_t *args;
1241 xfs_da_state_blk_t *tmp_blk;
1242 xfs_attr_leafblock_t *leaf1, *leaf2;
1243 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1244 int count, totallen, max, space, swap;
1247 * Set up environment.
1249 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1250 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1251 leaf1 = blk1->bp->data;
1252 leaf2 = blk2->bp->data;
1253 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1254 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1258 * Check ordering of blocks, reverse if it makes things simpler.
1260 * NOTE: Given that all (current) callers pass in an empty
1261 * second block, this code should never set "swap".
1264 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1268 leaf1 = blk1->bp->data;
1269 leaf2 = blk2->bp->data;
1276 * Examine entries until we reduce the absolute difference in
1277 * byte usage between the two blocks to a minimum. Then get
1278 * the direction to copy and the number of elements to move.
1280 * "inleaf" is true if the new entry should be inserted into blk1.
1281 * If "swap" is also true, then reverse the sense of "inleaf".
1283 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1286 state->inleaf = !state->inleaf;
1289 * Move any entries required from leaf to leaf:
1291 if (count < be16_to_cpu(hdr1->count)) {
1293 * Figure the total bytes to be added to the destination leaf.
1295 /* number entries being moved */
1296 count = be16_to_cpu(hdr1->count) - count;
1297 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1298 space += count * sizeof(xfs_attr_leaf_entry_t);
1301 * leaf2 is the destination, compact it if it looks tight.
1303 max = be16_to_cpu(hdr2->firstused)
1304 - sizeof(xfs_attr_leaf_hdr_t);
1305 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1307 xfs_attr_leaf_compact(args->trans, blk2->bp);
1311 * Move high entries from leaf1 to low end of leaf2.
1313 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1314 leaf2, 0, count, state->mp);
1316 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1317 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1318 } else if (count > be16_to_cpu(hdr1->count)) {
1320 * I assert that since all callers pass in an empty
1321 * second buffer, this code should never execute.
1325 * Figure the total bytes to be added to the destination leaf.
1327 /* number entries being moved */
1328 count -= be16_to_cpu(hdr1->count);
1329 space = totallen - be16_to_cpu(hdr1->usedbytes);
1330 space += count * sizeof(xfs_attr_leaf_entry_t);
1333 * leaf1 is the destination, compact it if it looks tight.
1335 max = be16_to_cpu(hdr1->firstused)
1336 - sizeof(xfs_attr_leaf_hdr_t);
1337 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1339 xfs_attr_leaf_compact(args->trans, blk1->bp);
1343 * Move low entries from leaf2 to high end of leaf1.
1345 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1346 be16_to_cpu(hdr1->count), count, state->mp);
1348 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1349 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1353 * Copy out last hashval in each block for B-tree code.
1355 blk1->hashval = be32_to_cpu(
1356 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1357 blk2->hashval = be32_to_cpu(
1358 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1361 * Adjust the expected index for insertion.
1362 * NOTE: this code depends on the (current) situation that the
1363 * second block was originally empty.
1365 * If the insertion point moved to the 2nd block, we must adjust
1366 * the index. We must also track the entry just following the
1367 * new entry for use in an "atomic rename" operation, that entry
1368 * is always the "old" entry and the "new" entry is what we are
1369 * inserting. The index/blkno fields refer to the "old" entry,
1370 * while the index2/blkno2 fields refer to the "new" entry.
1372 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1373 ASSERT(state->inleaf == 0);
1374 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1375 args->index = args->index2 = blk2->index;
1376 args->blkno = args->blkno2 = blk2->blkno;
1377 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1378 if (state->inleaf) {
1379 args->index = blk1->index;
1380 args->blkno = blk1->blkno;
1382 args->blkno2 = blk2->blkno;
1384 blk2->index = blk1->index
1385 - be16_to_cpu(leaf1->hdr.count);
1386 args->index = args->index2 = blk2->index;
1387 args->blkno = args->blkno2 = blk2->blkno;
1390 ASSERT(state->inleaf == 1);
1391 args->index = args->index2 = blk1->index;
1392 args->blkno = args->blkno2 = blk1->blkno;
1397 * Examine entries until we reduce the absolute difference in
1398 * byte usage between the two blocks to a minimum.
1399 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1400 * GROT: there will always be enough room in either block for a new entry.
1401 * GROT: Do a double-split for this case?
1404 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1405 xfs_da_state_blk_t *blk1,
1406 xfs_da_state_blk_t *blk2,
1407 int *countarg, int *usedbytesarg)
1409 xfs_attr_leafblock_t *leaf1, *leaf2;
1410 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1411 xfs_attr_leaf_entry_t *entry;
1412 int count, max, index, totallen, half;
1413 int lastdelta, foundit, tmp;
1416 * Set up environment.
1418 leaf1 = blk1->bp->data;
1419 leaf2 = blk2->bp->data;
1426 * Examine entries until we reduce the absolute difference in
1427 * byte usage between the two blocks to a minimum.
1429 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1430 half = (max+1) * sizeof(*entry);
1431 half += be16_to_cpu(hdr1->usedbytes) +
1432 be16_to_cpu(hdr2->usedbytes) +
1433 xfs_attr_leaf_newentsize(
1434 state->args->namelen,
1435 state->args->valuelen,
1436 state->blocksize, NULL);
1438 lastdelta = state->blocksize;
1439 entry = &leaf1->entries[0];
1440 for (count = index = 0; count < max; entry++, index++, count++) {
1442 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1444 * The new entry is in the first block, account for it.
1446 if (count == blk1->index) {
1447 tmp = totallen + sizeof(*entry) +
1448 xfs_attr_leaf_newentsize(
1449 state->args->namelen,
1450 state->args->valuelen,
1451 state->blocksize, NULL);
1452 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1454 lastdelta = XFS_ATTR_ABS(half - tmp);
1460 * Wrap around into the second block if necessary.
1462 if (count == be16_to_cpu(hdr1->count)) {
1464 entry = &leaf1->entries[0];
1469 * Figure out if next leaf entry would be too much.
1471 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1473 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1475 lastdelta = XFS_ATTR_ABS(half - tmp);
1481 * Calculate the number of usedbytes that will end up in lower block.
1482 * If new entry not in lower block, fix up the count.
1484 totallen -= count * sizeof(*entry);
1486 totallen -= sizeof(*entry) +
1487 xfs_attr_leaf_newentsize(
1488 state->args->namelen,
1489 state->args->valuelen,
1490 state->blocksize, NULL);
1494 *usedbytesarg = totallen;
1498 /*========================================================================
1499 * Routines used for shrinking the Btree.
1500 *========================================================================*/
1503 * Check a leaf block and its neighbors to see if the block should be
1504 * collapsed into one or the other neighbor. Always keep the block
1505 * with the smaller block number.
1506 * If the current block is over 50% full, don't try to join it, return 0.
1507 * If the block is empty, fill in the state structure and return 2.
1508 * If it can be collapsed, fill in the state structure and return 1.
1509 * If nothing can be done, return 0.
1511 * GROT: allow for INCOMPLETE entries in calculation.
1514 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1516 xfs_attr_leafblock_t *leaf;
1517 xfs_da_state_blk_t *blk;
1518 xfs_da_blkinfo_t *info;
1519 int count, bytes, forward, error, retval, i;
1524 * Check for the degenerate case of the block being over 50% full.
1525 * If so, it's not worth even looking to see if we might be able
1526 * to coalesce with a sibling.
1528 blk = &state->path.blk[ state->path.active-1 ];
1529 info = blk->bp->data;
1530 ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1531 leaf = (xfs_attr_leafblock_t *)info;
1532 count = be16_to_cpu(leaf->hdr.count);
1533 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1534 count * sizeof(xfs_attr_leaf_entry_t) +
1535 be16_to_cpu(leaf->hdr.usedbytes);
1536 if (bytes > (state->blocksize >> 1)) {
1537 *action = 0; /* blk over 50%, don't try to join */
1542 * Check for the degenerate case of the block being empty.
1543 * If the block is empty, we'll simply delete it, no need to
1544 * coalesce it with a sibling block. We choose (arbitrarily)
1545 * to merge with the forward block unless it is NULL.
1549 * Make altpath point to the block we want to keep and
1550 * path point to the block we want to drop (this one).
1552 forward = (info->forw != 0);
1553 memcpy(&state->altpath, &state->path, sizeof(state->path));
1554 error = xfs_da_path_shift(state, &state->altpath, forward,
1567 * Examine each sibling block to see if we can coalesce with
1568 * at least 25% free space to spare. We need to figure out
1569 * whether to merge with the forward or the backward block.
1570 * We prefer coalescing with the lower numbered sibling so as
1571 * to shrink an attribute list over time.
1573 /* start with smaller blk num */
1574 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1575 for (i = 0; i < 2; forward = !forward, i++) {
1577 blkno = be32_to_cpu(info->forw);
1579 blkno = be32_to_cpu(info->back);
1582 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1583 blkno, -1, &bp, XFS_ATTR_FORK);
1588 leaf = (xfs_attr_leafblock_t *)info;
1589 count = be16_to_cpu(leaf->hdr.count);
1590 bytes = state->blocksize - (state->blocksize>>2);
1591 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1593 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1594 count += be16_to_cpu(leaf->hdr.count);
1595 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1596 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1597 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1598 xfs_da_brelse(state->args->trans, bp);
1600 break; /* fits with at least 25% to spare */
1608 * Make altpath point to the block we want to keep (the lower
1609 * numbered block) and path point to the block we want to drop.
1611 memcpy(&state->altpath, &state->path, sizeof(state->path));
1612 if (blkno < blk->blkno) {
1613 error = xfs_da_path_shift(state, &state->altpath, forward,
1616 error = xfs_da_path_shift(state, &state->path, forward,
1630 * Remove a name from the leaf attribute list structure.
1632 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1633 * If two leaves are 37% full, when combined they will leave 25% free.
1636 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1638 xfs_attr_leafblock_t *leaf;
1639 xfs_attr_leaf_hdr_t *hdr;
1640 xfs_attr_leaf_map_t *map;
1641 xfs_attr_leaf_entry_t *entry;
1642 int before, after, smallest, entsize;
1643 int tablesize, tmp, i;
1647 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1649 mp = args->trans->t_mountp;
1650 ASSERT((be16_to_cpu(hdr->count) > 0)
1651 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1652 ASSERT((args->index >= 0)
1653 && (args->index < be16_to_cpu(hdr->count)));
1654 ASSERT(be16_to_cpu(hdr->firstused) >=
1655 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1656 entry = &leaf->entries[args->index];
1657 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1658 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1661 * Scan through free region table:
1662 * check for adjacency of free'd entry with an existing one,
1663 * find smallest free region in case we need to replace it,
1664 * adjust any map that borders the entry table,
1666 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1667 + sizeof(xfs_attr_leaf_hdr_t);
1668 map = &hdr->freemap[0];
1669 tmp = be16_to_cpu(map->size);
1670 before = after = -1;
1671 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1672 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1673 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1674 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1675 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1676 if (be16_to_cpu(map->base) == tablesize) {
1677 be16_add(&map->base,
1678 -((int)sizeof(xfs_attr_leaf_entry_t)));
1679 be16_add(&map->size, sizeof(xfs_attr_leaf_entry_t));
1682 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1683 == be16_to_cpu(entry->nameidx)) {
1685 } else if (be16_to_cpu(map->base)
1686 == (be16_to_cpu(entry->nameidx) + entsize)) {
1688 } else if (be16_to_cpu(map->size) < tmp) {
1689 tmp = be16_to_cpu(map->size);
1695 * Coalesce adjacent freemap regions,
1696 * or replace the smallest region.
1698 if ((before >= 0) || (after >= 0)) {
1699 if ((before >= 0) && (after >= 0)) {
1700 map = &hdr->freemap[before];
1701 be16_add(&map->size, entsize);
1702 be16_add(&map->size,
1703 be16_to_cpu(hdr->freemap[after].size));
1704 hdr->freemap[after].base = 0;
1705 hdr->freemap[after].size = 0;
1706 } else if (before >= 0) {
1707 map = &hdr->freemap[before];
1708 be16_add(&map->size, entsize);
1710 map = &hdr->freemap[after];
1711 /* both on-disk, don't endian flip twice */
1712 map->base = entry->nameidx;
1713 be16_add(&map->size, entsize);
1717 * Replace smallest region (if it is smaller than free'd entry)
1719 map = &hdr->freemap[smallest];
1720 if (be16_to_cpu(map->size) < entsize) {
1721 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1722 map->size = cpu_to_be16(entsize);
1727 * Did we remove the first entry?
1729 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1735 * Compress the remaining entries and zero out the removed stuff.
1737 memset(XFS_ATTR_LEAF_NAME(leaf, args->index), 0, entsize);
1738 be16_add(&hdr->usedbytes, -entsize);
1739 xfs_da_log_buf(args->trans, bp,
1740 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1743 tmp = (be16_to_cpu(hdr->count) - args->index)
1744 * sizeof(xfs_attr_leaf_entry_t);
1745 memmove((char *)entry, (char *)(entry+1), tmp);
1746 be16_add(&hdr->count, -1);
1747 xfs_da_log_buf(args->trans, bp,
1748 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1749 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1750 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1753 * If we removed the first entry, re-find the first used byte
1754 * in the name area. Note that if the entry was the "firstused",
1755 * then we don't have a "hole" in our block resulting from
1756 * removing the name.
1759 tmp = XFS_LBSIZE(mp);
1760 entry = &leaf->entries[0];
1761 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1762 ASSERT(be16_to_cpu(entry->nameidx) >=
1763 be16_to_cpu(hdr->firstused));
1764 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1766 if (be16_to_cpu(entry->nameidx) < tmp)
1767 tmp = be16_to_cpu(entry->nameidx);
1769 hdr->firstused = cpu_to_be16(tmp);
1770 if (!hdr->firstused) {
1771 hdr->firstused = cpu_to_be16(
1772 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1775 hdr->holes = 1; /* mark as needing compaction */
1777 xfs_da_log_buf(args->trans, bp,
1778 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1781 * Check if leaf is less than 50% full, caller may want to
1782 * "join" the leaf with a sibling if so.
1784 tmp = sizeof(xfs_attr_leaf_hdr_t);
1785 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1786 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1787 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1791 * Move all the attribute list entries from drop_leaf into save_leaf.
1794 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1795 xfs_da_state_blk_t *save_blk)
1797 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1798 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1803 * Set up environment.
1806 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1807 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1808 drop_leaf = drop_blk->bp->data;
1809 save_leaf = save_blk->bp->data;
1810 ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1811 ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1812 drop_hdr = &drop_leaf->hdr;
1813 save_hdr = &save_leaf->hdr;
1816 * Save last hashval from dying block for later Btree fixup.
1818 drop_blk->hashval = be32_to_cpu(
1819 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1822 * Check if we need a temp buffer, or can we do it in place.
1823 * Note that we don't check "leaf" for holes because we will
1824 * always be dropping it, toosmall() decided that for us already.
1826 if (save_hdr->holes == 0) {
1828 * dest leaf has no holes, so we add there. May need
1829 * to make some room in the entry array.
1831 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1832 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1833 be16_to_cpu(drop_hdr->count), mp);
1835 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1836 be16_to_cpu(save_hdr->count),
1837 be16_to_cpu(drop_hdr->count), mp);
1841 * Destination has holes, so we make a temporary copy
1842 * of the leaf and add them both to that.
1844 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1845 ASSERT(tmpbuffer != NULL);
1846 memset(tmpbuffer, 0, state->blocksize);
1847 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1848 tmp_hdr = &tmp_leaf->hdr;
1849 tmp_hdr->info = save_hdr->info; /* struct copy */
1851 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1852 if (!tmp_hdr->firstused) {
1853 tmp_hdr->firstused = cpu_to_be16(
1854 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1856 tmp_hdr->usedbytes = 0;
1857 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1858 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1859 be16_to_cpu(drop_hdr->count), mp);
1860 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1861 be16_to_cpu(tmp_leaf->hdr.count),
1862 be16_to_cpu(save_hdr->count), mp);
1864 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1865 be16_to_cpu(save_hdr->count), mp);
1866 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1867 be16_to_cpu(tmp_leaf->hdr.count),
1868 be16_to_cpu(drop_hdr->count), mp);
1870 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1871 kmem_free(tmpbuffer, state->blocksize);
1874 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1875 state->blocksize - 1);
1878 * Copy out last hashval in each block for B-tree code.
1880 save_blk->hashval = be32_to_cpu(
1881 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1884 /*========================================================================
1885 * Routines used for finding things in the Btree.
1886 *========================================================================*/
1889 * Look up a name in a leaf attribute list structure.
1890 * This is the internal routine, it uses the caller's buffer.
1892 * Note that duplicate keys are allowed, but only check within the
1893 * current leaf node. The Btree code must check in adjacent leaf nodes.
1895 * Return in args->index the index into the entry[] array of either
1896 * the found entry, or where the entry should have been (insert before
1899 * Don't change the args->value unless we find the attribute.
1902 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1904 xfs_attr_leafblock_t *leaf;
1905 xfs_attr_leaf_entry_t *entry;
1906 xfs_attr_leaf_name_local_t *name_loc;
1907 xfs_attr_leaf_name_remote_t *name_rmt;
1909 xfs_dahash_t hashval;
1912 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1913 ASSERT(be16_to_cpu(leaf->hdr.count)
1914 < (XFS_LBSIZE(args->dp->i_mount)/8));
1917 * Binary search. (note: small blocks will skip this loop)
1919 hashval = args->hashval;
1920 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1921 for (entry = &leaf->entries[probe]; span > 4;
1922 entry = &leaf->entries[probe]) {
1924 if (be32_to_cpu(entry->hashval) < hashval)
1926 else if (be32_to_cpu(entry->hashval) > hashval)
1931 ASSERT((probe >= 0) &&
1933 || (probe < be16_to_cpu(leaf->hdr.count))));
1934 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1937 * Since we may have duplicate hashval's, find the first matching
1938 * hashval in the leaf.
1940 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1944 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1945 (be32_to_cpu(entry->hashval) < hashval)) {
1949 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1950 (be32_to_cpu(entry->hashval) != hashval)) {
1951 args->index = probe;
1952 return(XFS_ERROR(ENOATTR));
1956 * Duplicate keys may be present, so search all of them for a match.
1958 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1959 (be32_to_cpu(entry->hashval) == hashval);
1962 * GROT: Add code to remove incomplete entries.
1965 * If we are looking for INCOMPLETE entries, show only those.
1966 * If we are looking for complete entries, show only those.
1968 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
1969 (entry->flags & XFS_ATTR_INCOMPLETE)) {
1972 if (entry->flags & XFS_ATTR_LOCAL) {
1973 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, probe);
1974 if (name_loc->namelen != args->namelen)
1976 if (memcmp(args->name, (char *)name_loc->nameval,
1977 args->namelen) != 0)
1979 if (((args->flags & ATTR_SECURE) != 0) !=
1980 ((entry->flags & XFS_ATTR_SECURE) != 0))
1982 if (((args->flags & ATTR_ROOT) != 0) !=
1983 ((entry->flags & XFS_ATTR_ROOT) != 0))
1985 args->index = probe;
1986 return(XFS_ERROR(EEXIST));
1988 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, probe);
1989 if (name_rmt->namelen != args->namelen)
1991 if (memcmp(args->name, (char *)name_rmt->name,
1992 args->namelen) != 0)
1994 if (((args->flags & ATTR_SECURE) != 0) !=
1995 ((entry->flags & XFS_ATTR_SECURE) != 0))
1997 if (((args->flags & ATTR_ROOT) != 0) !=
1998 ((entry->flags & XFS_ATTR_ROOT) != 0))
2000 args->index = probe;
2001 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2002 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2003 be32_to_cpu(name_rmt->valuelen));
2004 return(XFS_ERROR(EEXIST));
2007 args->index = probe;
2008 return(XFS_ERROR(ENOATTR));
2012 * Get the value associated with an attribute name from a leaf attribute
2016 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2019 xfs_attr_leafblock_t *leaf;
2020 xfs_attr_leaf_entry_t *entry;
2021 xfs_attr_leaf_name_local_t *name_loc;
2022 xfs_attr_leaf_name_remote_t *name_rmt;
2025 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2026 ASSERT(be16_to_cpu(leaf->hdr.count)
2027 < (XFS_LBSIZE(args->dp->i_mount)/8));
2028 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2030 entry = &leaf->entries[args->index];
2031 if (entry->flags & XFS_ATTR_LOCAL) {
2032 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2033 ASSERT(name_loc->namelen == args->namelen);
2034 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2035 valuelen = be16_to_cpu(name_loc->valuelen);
2036 if (args->flags & ATTR_KERNOVAL) {
2037 args->valuelen = valuelen;
2040 if (args->valuelen < valuelen) {
2041 args->valuelen = valuelen;
2042 return(XFS_ERROR(ERANGE));
2044 args->valuelen = valuelen;
2045 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2047 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2048 ASSERT(name_rmt->namelen == args->namelen);
2049 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2050 valuelen = be32_to_cpu(name_rmt->valuelen);
2051 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2052 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2053 if (args->flags & ATTR_KERNOVAL) {
2054 args->valuelen = valuelen;
2057 if (args->valuelen < valuelen) {
2058 args->valuelen = valuelen;
2059 return(XFS_ERROR(ERANGE));
2061 args->valuelen = valuelen;
2066 /*========================================================================
2068 *========================================================================*/
2071 * Move the indicated entries from one leaf to another.
2072 * NOTE: this routine modifies both source and destination leaves.
2076 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2077 xfs_attr_leafblock_t *leaf_d, int start_d,
2078 int count, xfs_mount_t *mp)
2080 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2081 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2085 * Check for nothing to do.
2091 * Set up environment.
2093 ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2094 ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2095 hdr_s = &leaf_s->hdr;
2096 hdr_d = &leaf_d->hdr;
2097 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2098 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2099 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2100 ((be16_to_cpu(hdr_s->count)
2101 * sizeof(*entry_s))+sizeof(*hdr_s)));
2102 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2103 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2104 ((be16_to_cpu(hdr_d->count)
2105 * sizeof(*entry_d))+sizeof(*hdr_d)));
2107 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2108 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2109 ASSERT(count <= be16_to_cpu(hdr_s->count));
2112 * Move the entries in the destination leaf up to make a hole?
2114 if (start_d < be16_to_cpu(hdr_d->count)) {
2115 tmp = be16_to_cpu(hdr_d->count) - start_d;
2116 tmp *= sizeof(xfs_attr_leaf_entry_t);
2117 entry_s = &leaf_d->entries[start_d];
2118 entry_d = &leaf_d->entries[start_d + count];
2119 memmove((char *)entry_d, (char *)entry_s, tmp);
2123 * Copy all entry's in the same (sorted) order,
2124 * but allocate attribute info packed and in sequence.
2126 entry_s = &leaf_s->entries[start_s];
2127 entry_d = &leaf_d->entries[start_d];
2129 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2130 ASSERT(be16_to_cpu(entry_s->nameidx)
2131 >= be16_to_cpu(hdr_s->firstused));
2132 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2135 * Code to drop INCOMPLETE entries. Difficult to use as we
2136 * may also need to change the insertion index. Code turned
2137 * off for 6.2, should be revisited later.
2139 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2140 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2141 be16_add(&hdr_s->usedbytes, -tmp);
2142 be16_add(&hdr_s->count, -1);
2143 entry_d--; /* to compensate for ++ in loop hdr */
2145 if ((start_s + i) < offset)
2146 result++; /* insertion index adjustment */
2149 be16_add(&hdr_d->firstused, -tmp);
2150 /* both on-disk, don't endian flip twice */
2151 entry_d->hashval = entry_s->hashval;
2152 /* both on-disk, don't endian flip twice */
2153 entry_d->nameidx = hdr_d->firstused;
2154 entry_d->flags = entry_s->flags;
2155 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2157 memmove(XFS_ATTR_LEAF_NAME(leaf_d, desti),
2158 XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), tmp);
2159 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2161 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2162 be16_add(&hdr_s->usedbytes, -tmp);
2163 be16_add(&hdr_d->usedbytes, tmp);
2164 be16_add(&hdr_s->count, -1);
2165 be16_add(&hdr_d->count, 1);
2166 tmp = be16_to_cpu(hdr_d->count)
2167 * sizeof(xfs_attr_leaf_entry_t)
2168 + sizeof(xfs_attr_leaf_hdr_t);
2169 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2176 * Zero out the entries we just copied.
2178 if (start_s == be16_to_cpu(hdr_s->count)) {
2179 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2180 entry_s = &leaf_s->entries[start_s];
2181 ASSERT(((char *)entry_s + tmp) <=
2182 ((char *)leaf_s + XFS_LBSIZE(mp)));
2183 memset((char *)entry_s, 0, tmp);
2186 * Move the remaining entries down to fill the hole,
2187 * then zero the entries at the top.
2189 tmp = be16_to_cpu(hdr_s->count) - count;
2190 tmp *= sizeof(xfs_attr_leaf_entry_t);
2191 entry_s = &leaf_s->entries[start_s + count];
2192 entry_d = &leaf_s->entries[start_s];
2193 memmove((char *)entry_d, (char *)entry_s, tmp);
2195 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2196 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2197 ASSERT(((char *)entry_s + tmp) <=
2198 ((char *)leaf_s + XFS_LBSIZE(mp)));
2199 memset((char *)entry_s, 0, tmp);
2203 * Fill in the freemap information
2205 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2206 be16_add(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2207 sizeof(xfs_attr_leaf_entry_t));
2208 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2209 - be16_to_cpu(hdr_d->freemap[0].base));
2210 hdr_d->freemap[1].base = 0;
2211 hdr_d->freemap[2].base = 0;
2212 hdr_d->freemap[1].size = 0;
2213 hdr_d->freemap[2].size = 0;
2214 hdr_s->holes = 1; /* leaf may not be compact */
2218 * Compare two leaf blocks "order".
2219 * Return 0 unless leaf2 should go before leaf1.
2222 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2224 xfs_attr_leafblock_t *leaf1, *leaf2;
2226 leaf1 = leaf1_bp->data;
2227 leaf2 = leaf2_bp->data;
2228 ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2229 (be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC));
2230 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2231 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2232 ((be32_to_cpu(leaf2->entries[0].hashval) <
2233 be32_to_cpu(leaf1->entries[0].hashval)) ||
2234 (be32_to_cpu(leaf2->entries[
2235 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2236 be32_to_cpu(leaf1->entries[
2237 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2244 * Pick up the last hashvalue from a leaf block.
2247 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2249 xfs_attr_leafblock_t *leaf;
2252 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2254 *count = be16_to_cpu(leaf->hdr.count);
2255 if (!leaf->hdr.count)
2257 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2261 * Calculate the number of bytes used to store the indicated attribute
2262 * (whether local or remote only calculate bytes in this block).
2265 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2267 xfs_attr_leaf_name_local_t *name_loc;
2268 xfs_attr_leaf_name_remote_t *name_rmt;
2271 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2272 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2273 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, index);
2274 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc->namelen,
2275 be16_to_cpu(name_loc->valuelen));
2277 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, index);
2278 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt->namelen);
2284 * Calculate the number of bytes that would be required to store the new
2285 * attribute (whether local or remote only calculate bytes in this block).
2286 * This routine decides as a side effect whether the attribute will be
2287 * a "local" or a "remote" attribute.
2290 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2294 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen, valuelen);
2295 if (size < XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize)) {
2300 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen);
2309 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2312 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2314 attrlist_cursor_kern_t *cursor;
2315 xfs_attr_leafblock_t *leaf;
2316 xfs_attr_leaf_entry_t *entry;
2317 xfs_attr_leaf_name_local_t *name_loc;
2318 xfs_attr_leaf_name_remote_t *name_rmt;
2323 cursor = context->cursor;
2324 cursor->initted = 1;
2326 xfs_attr_trace_l_cl("blk start", context, leaf);
2329 * Re-find our place in the leaf block if this is a new syscall.
2331 if (context->resynch) {
2332 entry = &leaf->entries[0];
2333 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2334 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2335 if (cursor->offset == context->dupcnt) {
2336 context->dupcnt = 0;
2340 } else if (be32_to_cpu(entry->hashval) >
2342 context->dupcnt = 0;
2346 if (i == be16_to_cpu(leaf->hdr.count)) {
2347 xfs_attr_trace_l_c("not found", context);
2351 entry = &leaf->entries[0];
2354 context->resynch = 0;
2357 * We have found our place, start copying out the new attributes.
2360 for ( ; (i < be16_to_cpu(leaf->hdr.count))
2361 && (retval == 0); entry++, i++) {
2362 attrnames_t *namesp;
2364 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2365 cursor->hashval = be32_to_cpu(entry->hashval);
2369 if (entry->flags & XFS_ATTR_INCOMPLETE)
2370 continue; /* skip incomplete entries */
2371 if (((context->flags & ATTR_SECURE) != 0) !=
2372 ((entry->flags & XFS_ATTR_SECURE) != 0) &&
2373 !(context->flags & ATTR_KERNORMALS))
2374 continue; /* skip non-matching entries */
2375 if (((context->flags & ATTR_ROOT) != 0) !=
2376 ((entry->flags & XFS_ATTR_ROOT) != 0) &&
2377 !(context->flags & ATTR_KERNROOTLS))
2378 continue; /* skip non-matching entries */
2380 namesp = (entry->flags & XFS_ATTR_SECURE) ? &attr_secure :
2381 ((entry->flags & XFS_ATTR_ROOT) ? &attr_trusted :
2384 if (entry->flags & XFS_ATTR_LOCAL) {
2385 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
2386 if (context->flags & ATTR_KERNOVAL) {
2387 ASSERT(context->flags & ATTR_KERNAMELS);
2388 context->count += namesp->attr_namelen +
2389 (int)name_loc->namelen + 1;
2391 retval = xfs_attr_put_listent(context, namesp,
2392 (char *)name_loc->nameval,
2393 (int)name_loc->namelen,
2394 be16_to_cpu(name_loc->valuelen));
2397 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2398 if (context->flags & ATTR_KERNOVAL) {
2399 ASSERT(context->flags & ATTR_KERNAMELS);
2400 context->count += namesp->attr_namelen +
2401 (int)name_rmt->namelen + 1;
2403 retval = xfs_attr_put_listent(context, namesp,
2404 (char *)name_rmt->name,
2405 (int)name_rmt->namelen,
2406 be32_to_cpu(name_rmt->valuelen));
2413 xfs_attr_trace_l_cl("blk end", context, leaf);
2417 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
2418 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
2419 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
2420 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
2421 & ~(sizeof(u_int32_t)-1))
2424 * Format an attribute and copy it out to the user's buffer.
2425 * Take care to check values and protect against them changing later,
2426 * we may be reading them directly out of a user buffer.
2430 xfs_attr_put_listent(xfs_attr_list_context_t *context,
2431 attrnames_t *namesp, char *name, int namelen, int valuelen)
2433 attrlist_ent_t *aep;
2436 ASSERT(!(context->flags & ATTR_KERNOVAL));
2437 if (context->flags & ATTR_KERNAMELS) {
2440 ASSERT(context->count >= 0);
2442 arraytop = context->count + namesp->attr_namelen + namelen + 1;
2443 if (arraytop > context->firstu) {
2444 context->count = -1; /* insufficient space */
2447 offset = (char *)context->alist + context->count;
2448 strncpy(offset, namesp->attr_name, namesp->attr_namelen);
2449 offset += namesp->attr_namelen;
2450 strncpy(offset, name, namelen); /* real name */
2453 context->count += namesp->attr_namelen + namelen + 1;
2457 ASSERT(context->count >= 0);
2458 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
2459 ASSERT(context->firstu >= sizeof(*context->alist));
2460 ASSERT(context->firstu <= context->bufsize);
2462 arraytop = sizeof(*context->alist) +
2463 context->count * sizeof(context->alist->al_offset[0]);
2464 context->firstu -= ATTR_ENTSIZE(namelen);
2465 if (context->firstu < arraytop) {
2466 xfs_attr_trace_l_c("buffer full", context);
2467 context->alist->al_more = 1;
2471 aep = (attrlist_ent_t *)&(((char *)context->alist)[ context->firstu ]);
2472 aep->a_valuelen = valuelen;
2473 memcpy(aep->a_name, name, namelen);
2474 aep->a_name[ namelen ] = 0;
2475 context->alist->al_offset[ context->count++ ] = context->firstu;
2476 context->alist->al_count = context->count;
2477 xfs_attr_trace_l_c("add", context);
2481 /*========================================================================
2482 * Manage the INCOMPLETE flag in a leaf entry
2483 *========================================================================*/
2486 * Clear the INCOMPLETE flag on an entry in a leaf block.
2489 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2491 xfs_attr_leafblock_t *leaf;
2492 xfs_attr_leaf_entry_t *entry;
2493 xfs_attr_leaf_name_remote_t *name_rmt;
2497 xfs_attr_leaf_name_local_t *name_loc;
2503 * Set up the operation.
2505 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2513 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2514 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2515 ASSERT(args->index >= 0);
2516 entry = &leaf->entries[ args->index ];
2517 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2520 if (entry->flags & XFS_ATTR_LOCAL) {
2521 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2522 namelen = name_loc->namelen;
2523 name = (char *)name_loc->nameval;
2525 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2526 namelen = name_rmt->namelen;
2527 name = (char *)name_rmt->name;
2529 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2530 ASSERT(namelen == args->namelen);
2531 ASSERT(memcmp(name, args->name, namelen) == 0);
2534 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2535 xfs_da_log_buf(args->trans, bp,
2536 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2538 if (args->rmtblkno) {
2539 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2540 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2541 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2542 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2543 xfs_da_log_buf(args->trans, bp,
2544 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2546 xfs_da_buf_done(bp);
2549 * Commit the flag value change and start the next trans in series.
2551 error = xfs_attr_rolltrans(&args->trans, args->dp);
2557 * Set the INCOMPLETE flag on an entry in a leaf block.
2560 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2562 xfs_attr_leafblock_t *leaf;
2563 xfs_attr_leaf_entry_t *entry;
2564 xfs_attr_leaf_name_remote_t *name_rmt;
2569 * Set up the operation.
2571 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2579 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2580 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2581 ASSERT(args->index >= 0);
2582 entry = &leaf->entries[ args->index ];
2584 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2585 entry->flags |= XFS_ATTR_INCOMPLETE;
2586 xfs_da_log_buf(args->trans, bp,
2587 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2588 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2589 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2590 name_rmt->valueblk = 0;
2591 name_rmt->valuelen = 0;
2592 xfs_da_log_buf(args->trans, bp,
2593 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2595 xfs_da_buf_done(bp);
2598 * Commit the flag value change and start the next trans in series.
2600 error = xfs_attr_rolltrans(&args->trans, args->dp);
2606 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2607 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2608 * entry given by args->blkno2/index2.
2610 * Note that they could be in different blocks, or in the same block.
2613 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2615 xfs_attr_leafblock_t *leaf1, *leaf2;
2616 xfs_attr_leaf_entry_t *entry1, *entry2;
2617 xfs_attr_leaf_name_remote_t *name_rmt;
2618 xfs_dabuf_t *bp1, *bp2;
2621 xfs_attr_leaf_name_local_t *name_loc;
2622 int namelen1, namelen2;
2623 char *name1, *name2;
2627 * Read the block containing the "old" attr
2629 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2634 ASSERT(bp1 != NULL);
2637 * Read the block containing the "new" attr, if it is different
2639 if (args->blkno2 != args->blkno) {
2640 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2641 -1, &bp2, XFS_ATTR_FORK);
2645 ASSERT(bp2 != NULL);
2651 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2652 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2653 ASSERT(args->index >= 0);
2654 entry1 = &leaf1->entries[ args->index ];
2657 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2658 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2659 ASSERT(args->index2 >= 0);
2660 entry2 = &leaf2->entries[ args->index2 ];
2663 if (entry1->flags & XFS_ATTR_LOCAL) {
2664 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf1, args->index);
2665 namelen1 = name_loc->namelen;
2666 name1 = (char *)name_loc->nameval;
2668 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2669 namelen1 = name_rmt->namelen;
2670 name1 = (char *)name_rmt->name;
2672 if (entry2->flags & XFS_ATTR_LOCAL) {
2673 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf2, args->index2);
2674 namelen2 = name_loc->namelen;
2675 name2 = (char *)name_loc->nameval;
2677 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2678 namelen2 = name_rmt->namelen;
2679 name2 = (char *)name_rmt->name;
2681 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2682 ASSERT(namelen1 == namelen2);
2683 ASSERT(memcmp(name1, name2, namelen1) == 0);
2686 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2687 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2689 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2690 xfs_da_log_buf(args->trans, bp1,
2691 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2692 if (args->rmtblkno) {
2693 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2694 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2695 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2696 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2697 xfs_da_log_buf(args->trans, bp1,
2698 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2701 entry2->flags |= XFS_ATTR_INCOMPLETE;
2702 xfs_da_log_buf(args->trans, bp2,
2703 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2704 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2705 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2706 name_rmt->valueblk = 0;
2707 name_rmt->valuelen = 0;
2708 xfs_da_log_buf(args->trans, bp2,
2709 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2711 xfs_da_buf_done(bp1);
2713 xfs_da_buf_done(bp2);
2716 * Commit the flag value change and start the next trans in series.
2718 error = xfs_attr_rolltrans(&args->trans, args->dp);
2723 /*========================================================================
2724 * Indiscriminately delete the entire attribute fork
2725 *========================================================================*/
2728 * Recurse (gasp!) through the attribute nodes until we find leaves.
2729 * We're doing a depth-first traversal in order to invalidate everything.
2732 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2734 xfs_da_blkinfo_t *info;
2740 * Read block 0 to see what we have to work with.
2741 * We only get here if we have extents, since we remove
2742 * the extents in reverse order the extent containing
2743 * block 0 must still be there.
2745 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2748 blkno = xfs_da_blkno(bp);
2751 * Invalidate the tree, even if the "tree" is only a single leaf block.
2752 * This is a depth-first traversal!
2755 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2756 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2757 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2758 error = xfs_attr_leaf_inactive(trans, dp, bp);
2760 error = XFS_ERROR(EIO);
2761 xfs_da_brelse(*trans, bp);
2767 * Invalidate the incore copy of the root block.
2769 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2772 xfs_da_binval(*trans, bp); /* remove from cache */
2774 * Commit the invalidate and start the next transaction.
2776 error = xfs_attr_rolltrans(trans, dp);
2782 * Recurse (gasp!) through the attribute nodes until we find leaves.
2783 * We're doing a depth-first traversal in order to invalidate everything.
2786 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2789 xfs_da_blkinfo_t *info;
2790 xfs_da_intnode_t *node;
2791 xfs_dablk_t child_fsb;
2792 xfs_daddr_t parent_blkno, child_blkno;
2793 int error, count, i;
2794 xfs_dabuf_t *child_bp;
2797 * Since this code is recursive (gasp!) we must protect ourselves.
2799 if (level > XFS_DA_NODE_MAXDEPTH) {
2800 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2801 return(XFS_ERROR(EIO));
2805 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2806 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2807 count = be16_to_cpu(node->hdr.count);
2809 xfs_da_brelse(*trans, bp);
2812 child_fsb = be32_to_cpu(node->btree[0].before);
2813 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2816 * If this is the node level just above the leaves, simply loop
2817 * over the leaves removing all of them. If this is higher up
2818 * in the tree, recurse downward.
2820 for (i = 0; i < count; i++) {
2822 * Read the subsidiary block to see what we have to work with.
2823 * Don't do this in a transaction. This is a depth-first
2824 * traversal of the tree so we may deal with many blocks
2825 * before we come back to this one.
2827 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2832 /* save for re-read later */
2833 child_blkno = xfs_da_blkno(child_bp);
2836 * Invalidate the subtree, however we have to.
2838 info = child_bp->data;
2839 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2840 error = xfs_attr_node_inactive(trans, dp,
2842 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2843 error = xfs_attr_leaf_inactive(trans, dp,
2846 error = XFS_ERROR(EIO);
2847 xfs_da_brelse(*trans, child_bp);
2853 * Remove the subsidiary block from the cache
2856 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2857 &child_bp, XFS_ATTR_FORK);
2860 xfs_da_binval(*trans, child_bp);
2864 * If we're not done, re-read the parent to get the next
2865 * child block number.
2867 if ((i+1) < count) {
2868 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2869 &bp, XFS_ATTR_FORK);
2872 child_fsb = be32_to_cpu(node->btree[i+1].before);
2873 xfs_da_brelse(*trans, bp);
2876 * Atomically commit the whole invalidate stuff.
2878 if ((error = xfs_attr_rolltrans(trans, dp)))
2886 * Invalidate all of the "remote" value regions pointed to by a particular
2888 * Note that we must release the lock on the buffer so that we are not
2889 * caught holding something that the logging code wants to flush to disk.
2892 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2894 xfs_attr_leafblock_t *leaf;
2895 xfs_attr_leaf_entry_t *entry;
2896 xfs_attr_leaf_name_remote_t *name_rmt;
2897 xfs_attr_inactive_list_t *list, *lp;
2898 int error, count, size, tmp, i;
2901 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2904 * Count the number of "remote" value extents.
2907 entry = &leaf->entries[0];
2908 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2909 if (be16_to_cpu(entry->nameidx) &&
2910 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2911 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2912 if (name_rmt->valueblk)
2918 * If there are no "remote" values, we're done.
2921 xfs_da_brelse(*trans, bp);
2926 * Allocate storage for a list of all the "remote" value extents.
2928 size = count * sizeof(xfs_attr_inactive_list_t);
2929 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2932 * Identify each of the "remote" value extents.
2935 entry = &leaf->entries[0];
2936 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2937 if (be16_to_cpu(entry->nameidx) &&
2938 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2939 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2940 if (name_rmt->valueblk) {
2941 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2942 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2943 be32_to_cpu(name_rmt->valuelen));
2948 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2951 * Invalidate each of the "remote" value extents.
2954 for (lp = list, i = 0; i < count; i++, lp++) {
2955 tmp = xfs_attr_leaf_freextent(trans, dp,
2956 lp->valueblk, lp->valuelen);
2959 error = tmp; /* save only the 1st errno */
2962 kmem_free((xfs_caddr_t)list, size);
2967 * Look at all the extents for this logical region,
2968 * invalidate any buffers that are incore/in transactions.
2971 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2972 xfs_dablk_t blkno, int blkcnt)
2974 xfs_bmbt_irec_t map;
2976 int tblkcnt, dblkcnt, nmap, error;
2981 * Roll through the "value", invalidating the attribute value's
2986 while (tblkcnt > 0) {
2988 * Try to remember where we decided to put the value.
2991 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
2992 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2993 NULL, 0, &map, &nmap, NULL);
2998 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3001 * If it's a hole, these are already unmapped
3002 * so there's nothing to invalidate.
3004 if (map.br_startblock != HOLESTARTBLOCK) {
3006 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3008 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3010 bp = xfs_trans_get_buf(*trans,
3011 dp->i_mount->m_ddev_targp,
3012 dblkno, dblkcnt, XFS_BUF_LOCK);
3013 xfs_trans_binval(*trans, bp);
3015 * Roll to next transaction.
3017 if ((error = xfs_attr_rolltrans(trans, dp)))
3021 tblkno += map.br_blockcount;
3022 tblkcnt -= map.br_blockcount;
3030 * Roll from one trans in the sequence of PERMANENT transactions to the next.
3033 xfs_attr_rolltrans(xfs_trans_t **transp, xfs_inode_t *dp)
3036 unsigned int logres, count;
3040 * Ensure that the inode is always logged.
3043 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
3046 * Copy the critical parameters from one trans to the next.
3048 logres = trans->t_log_res;
3049 count = trans->t_log_count;
3050 *transp = xfs_trans_dup(trans);
3053 * Commit the current transaction.
3054 * If this commit failed, then it'd just unlock those items that
3055 * are not marked ihold. That also means that a filesystem shutdown
3056 * is in progress. The caller takes the responsibility to cancel
3057 * the duplicate transaction that gets returned.
3059 if ((error = xfs_trans_commit(trans, 0, NULL)))
3065 * Reserve space in the log for th next transaction.
3066 * This also pushes items in the "AIL", the list of logged items,
3067 * out to disk if they are taking up space at the tail of the log
3068 * that we want to use. This requires that either nothing be locked
3069 * across this call, or that anything that is locked be logged in
3070 * the prior and the next transactions.
3072 error = xfs_trans_reserve(trans, 0, logres, 0,
3073 XFS_TRANS_PERM_LOG_RES, count);
3075 * Ensure that the inode is in the new transaction and locked.
3078 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
3079 xfs_trans_ihold(trans, dp);