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 INT_SET(hdr->totsize, ARCH_CONVERT, 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 < INT_GET(sf->hdr.count, ARCH_CONVERT);
228 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
230 if (sfe->namelen != args->namelen)
232 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
234 if (((args->flags & ATTR_SECURE) != 0) !=
235 ((sfe->flags & XFS_ATTR_SECURE) != 0))
237 if (((args->flags & ATTR_ROOT) != 0) !=
238 ((sfe->flags & XFS_ATTR_ROOT) != 0))
244 offset = (char *)sfe - (char *)sf;
245 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
246 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
247 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
248 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
250 sfe->namelen = args->namelen;
251 INT_SET(sfe->valuelen, ARCH_CONVERT, args->valuelen);
252 sfe->flags = (args->flags & ATTR_SECURE) ? XFS_ATTR_SECURE :
253 ((args->flags & ATTR_ROOT) ? XFS_ATTR_ROOT : 0);
254 memcpy(sfe->nameval, args->name, args->namelen);
255 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
256 INT_MOD(sf->hdr.count, ARCH_CONVERT, 1);
257 INT_MOD(sf->hdr.totsize, ARCH_CONVERT, size);
258 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
260 xfs_sbversion_add_attr2(mp, args->trans);
264 * Remove an attribute from the shortform attribute list structure.
267 xfs_attr_shortform_remove(xfs_da_args_t *args)
269 xfs_attr_shortform_t *sf;
270 xfs_attr_sf_entry_t *sfe;
271 int base, size=0, end, totsize, i;
277 base = sizeof(xfs_attr_sf_hdr_t);
278 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
280 end = INT_GET(sf->hdr.count, ARCH_CONVERT);
281 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
283 size = XFS_ATTR_SF_ENTSIZE(sfe);
284 if (sfe->namelen != args->namelen)
286 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
288 if (((args->flags & ATTR_SECURE) != 0) !=
289 ((sfe->flags & XFS_ATTR_SECURE) != 0))
291 if (((args->flags & ATTR_ROOT) != 0) !=
292 ((sfe->flags & XFS_ATTR_ROOT) != 0))
297 return(XFS_ERROR(ENOATTR));
300 * Fix up the attribute fork data, covering the hole
303 totsize = INT_GET(sf->hdr.totsize, ARCH_CONVERT);
305 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
306 INT_MOD(sf->hdr.count, ARCH_CONVERT, -1);
307 INT_MOD(sf->hdr.totsize, ARCH_CONVERT, -size);
310 * Fix up the start offset of the attribute fork
313 if (totsize == sizeof(xfs_attr_sf_hdr_t) && !args->addname &&
314 (mp->m_flags & XFS_MOUNT_ATTR2)) {
316 * Last attribute now removed, revert to original
317 * inode format making all literal area available
318 * to the data fork once more.
320 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
321 dp->i_d.di_forkoff = 0;
322 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
323 ASSERT(dp->i_d.di_anextents == 0);
324 ASSERT(dp->i_afp == NULL);
325 dp->i_df.if_ext_max =
326 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
327 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
329 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
330 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
331 ASSERT(dp->i_d.di_forkoff);
332 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || args->addname ||
333 !(mp->m_flags & XFS_MOUNT_ATTR2));
334 dp->i_afp->if_ext_max =
335 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
336 dp->i_df.if_ext_max =
337 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
338 xfs_trans_log_inode(args->trans, dp,
339 XFS_ILOG_CORE | XFS_ILOG_ADATA);
342 xfs_sbversion_add_attr2(mp, args->trans);
348 * Look up a name in a shortform attribute list structure.
352 xfs_attr_shortform_lookup(xfs_da_args_t *args)
354 xfs_attr_shortform_t *sf;
355 xfs_attr_sf_entry_t *sfe;
359 ifp = args->dp->i_afp;
360 ASSERT(ifp->if_flags & XFS_IFINLINE);
361 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
363 for (i = 0; i < INT_GET(sf->hdr.count, ARCH_CONVERT);
364 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
365 if (sfe->namelen != args->namelen)
367 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
369 if (((args->flags & ATTR_SECURE) != 0) !=
370 ((sfe->flags & XFS_ATTR_SECURE) != 0))
372 if (((args->flags & ATTR_ROOT) != 0) !=
373 ((sfe->flags & XFS_ATTR_ROOT) != 0))
375 return(XFS_ERROR(EEXIST));
377 return(XFS_ERROR(ENOATTR));
381 * Look up a name in a shortform attribute list structure.
385 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
387 xfs_attr_shortform_t *sf;
388 xfs_attr_sf_entry_t *sfe;
391 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
392 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
394 for (i = 0; i < INT_GET(sf->hdr.count, ARCH_CONVERT);
395 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
396 if (sfe->namelen != args->namelen)
398 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
400 if (((args->flags & ATTR_SECURE) != 0) !=
401 ((sfe->flags & XFS_ATTR_SECURE) != 0))
403 if (((args->flags & ATTR_ROOT) != 0) !=
404 ((sfe->flags & XFS_ATTR_ROOT) != 0))
406 if (args->flags & ATTR_KERNOVAL) {
407 args->valuelen = INT_GET(sfe->valuelen, ARCH_CONVERT);
408 return(XFS_ERROR(EEXIST));
410 if (args->valuelen < INT_GET(sfe->valuelen, ARCH_CONVERT)) {
411 args->valuelen = INT_GET(sfe->valuelen, ARCH_CONVERT);
412 return(XFS_ERROR(ERANGE));
414 args->valuelen = INT_GET(sfe->valuelen, ARCH_CONVERT);
415 memcpy(args->value, &sfe->nameval[args->namelen],
417 return(XFS_ERROR(EEXIST));
419 return(XFS_ERROR(ENOATTR));
423 * Convert from using the shortform to the leaf.
426 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
429 xfs_attr_shortform_t *sf;
430 xfs_attr_sf_entry_t *sfe;
440 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
441 size = INT_GET(sf->hdr.totsize, ARCH_CONVERT);
442 tmpbuffer = kmem_alloc(size, KM_SLEEP);
443 ASSERT(tmpbuffer != NULL);
444 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
445 sf = (xfs_attr_shortform_t *)tmpbuffer;
447 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
449 error = xfs_da_grow_inode(args, &blkno);
452 * If we hit an IO error middle of the transaction inside
453 * grow_inode(), we may have inconsistent data. Bail out.
457 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
458 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
463 error = xfs_attr_leaf_create(args, blkno, &bp);
465 error = xfs_da_shrink_inode(args, 0, bp);
469 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
470 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
474 memset((char *)&nargs, 0, sizeof(nargs));
476 nargs.firstblock = args->firstblock;
477 nargs.flist = args->flist;
478 nargs.total = args->total;
479 nargs.whichfork = XFS_ATTR_FORK;
480 nargs.trans = args->trans;
484 for (i = 0; i < INT_GET(sf->hdr.count, ARCH_CONVERT); i++) {
485 nargs.name = (char *)sfe->nameval;
486 nargs.namelen = sfe->namelen;
487 nargs.value = (char *)&sfe->nameval[nargs.namelen];
488 nargs.valuelen = INT_GET(sfe->valuelen, ARCH_CONVERT);
489 nargs.hashval = xfs_da_hashname((char *)sfe->nameval,
491 nargs.flags = (sfe->flags & XFS_ATTR_SECURE) ? ATTR_SECURE :
492 ((sfe->flags & XFS_ATTR_ROOT) ? ATTR_ROOT : 0);
493 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
494 ASSERT(error == ENOATTR);
495 error = xfs_attr_leaf_add(bp, &nargs);
496 ASSERT(error != ENOSPC);
499 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
506 kmem_free(tmpbuffer, size);
511 xfs_attr_shortform_compare(const void *a, const void *b)
513 xfs_attr_sf_sort_t *sa, *sb;
515 sa = (xfs_attr_sf_sort_t *)a;
516 sb = (xfs_attr_sf_sort_t *)b;
517 if (INT_GET(sa->hash, ARCH_CONVERT)
518 < INT_GET(sb->hash, ARCH_CONVERT)) {
520 } else if (INT_GET(sa->hash, ARCH_CONVERT)
521 > INT_GET(sb->hash, ARCH_CONVERT)) {
524 return(sa->entno - sb->entno);
529 * Copy out entries of shortform attribute lists for attr_list().
530 * Shortform atrtribute lists are not stored in hashval sorted order.
531 * If the output buffer is not large enough to hold them all, then we
532 * we have to calculate each entries' hashvalue and sort them before
533 * we can begin returning them to the user.
537 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
539 attrlist_cursor_kern_t *cursor;
540 xfs_attr_sf_sort_t *sbuf, *sbp;
541 xfs_attr_shortform_t *sf;
542 xfs_attr_sf_entry_t *sfe;
544 int sbsize, nsbuf, count, i;
546 ASSERT(context != NULL);
549 ASSERT(dp->i_afp != NULL);
550 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
554 cursor = context->cursor;
555 ASSERT(cursor != NULL);
557 xfs_attr_trace_l_c("sf start", context);
560 * If the buffer is large enough, do not bother with sorting.
561 * Note the generous fudge factor of 16 overhead bytes per entry.
563 if ((dp->i_afp->if_bytes + INT_GET(sf->hdr.count, ARCH_CONVERT) * 16)
564 < context->bufsize) {
565 for (i = 0, sfe = &sf->list[0];
566 i < INT_GET(sf->hdr.count, ARCH_CONVERT); i++) {
569 if (((context->flags & ATTR_SECURE) != 0) !=
570 ((sfe->flags & XFS_ATTR_SECURE) != 0) &&
571 !(context->flags & ATTR_KERNORMALS)) {
572 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
575 if (((context->flags & ATTR_ROOT) != 0) !=
576 ((sfe->flags & XFS_ATTR_ROOT) != 0) &&
577 !(context->flags & ATTR_KERNROOTLS)) {
578 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
581 namesp = (sfe->flags & XFS_ATTR_SECURE) ? &attr_secure:
582 ((sfe->flags & XFS_ATTR_ROOT) ? &attr_trusted :
584 if (context->flags & ATTR_KERNOVAL) {
585 ASSERT(context->flags & ATTR_KERNAMELS);
586 context->count += namesp->attr_namelen +
587 INT_GET(sfe->namelen, ARCH_CONVERT) + 1;
590 if (xfs_attr_put_listent(context, namesp,
591 (char *)sfe->nameval,
593 (int)INT_GET(sfe->valuelen,
597 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
599 xfs_attr_trace_l_c("sf big-gulp", context);
604 * It didn't all fit, so we have to sort everything on hashval.
606 sbsize = INT_GET(sf->hdr.count, ARCH_CONVERT) * sizeof(*sbuf);
607 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP);
610 * Scan the attribute list for the rest of the entries, storing
611 * the relevant info from only those that match into a buffer.
614 for (i = 0, sfe = &sf->list[0];
615 i < INT_GET(sf->hdr.count, ARCH_CONVERT); i++) {
617 ((char *)sfe < (char *)sf) ||
618 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
619 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
621 context->dp->i_mount, sfe);
622 xfs_attr_trace_l_c("sf corrupted", context);
623 kmem_free(sbuf, sbsize);
624 return XFS_ERROR(EFSCORRUPTED);
626 if (((context->flags & ATTR_SECURE) != 0) !=
627 ((sfe->flags & XFS_ATTR_SECURE) != 0) &&
628 !(context->flags & ATTR_KERNORMALS)) {
629 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
632 if (((context->flags & ATTR_ROOT) != 0) !=
633 ((sfe->flags & XFS_ATTR_ROOT) != 0) &&
634 !(context->flags & ATTR_KERNROOTLS)) {
635 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
639 INT_SET(sbp->hash, ARCH_CONVERT,
640 xfs_da_hashname((char *)sfe->nameval, sfe->namelen));
641 sbp->name = (char *)sfe->nameval;
642 sbp->namelen = sfe->namelen;
643 /* These are bytes, and both on-disk, don't endian-flip */
644 sbp->valuelen = sfe->valuelen;
645 sbp->flags = sfe->flags;
646 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
652 * Sort the entries on hash then entno.
654 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
657 * Re-find our place IN THE SORTED LIST.
662 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
663 if (INT_GET(sbp->hash, ARCH_CONVERT) == cursor->hashval) {
664 if (cursor->offset == count) {
668 } else if (INT_GET(sbp->hash, ARCH_CONVERT) > cursor->hashval) {
673 kmem_free(sbuf, sbsize);
674 xfs_attr_trace_l_c("blk end", context);
679 * Loop putting entries into the user buffer.
681 for ( ; i < nsbuf; i++, sbp++) {
684 namesp = (sbp->flags & XFS_ATTR_SECURE) ? &attr_secure :
685 ((sbp->flags & XFS_ATTR_ROOT) ? &attr_trusted :
688 if (cursor->hashval != INT_GET(sbp->hash, ARCH_CONVERT)) {
689 cursor->hashval = INT_GET(sbp->hash, ARCH_CONVERT);
692 if (context->flags & ATTR_KERNOVAL) {
693 ASSERT(context->flags & ATTR_KERNAMELS);
694 context->count += namesp->attr_namelen +
697 if (xfs_attr_put_listent(context, namesp,
698 sbp->name, sbp->namelen,
699 INT_GET(sbp->valuelen, ARCH_CONVERT)))
705 kmem_free(sbuf, sbsize);
706 xfs_attr_trace_l_c("sf E-O-F", context);
711 * Check a leaf attribute block to see if all the entries would fit into
712 * a shortform attribute list.
715 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
717 xfs_attr_leafblock_t *leaf;
718 xfs_attr_leaf_entry_t *entry;
719 xfs_attr_leaf_name_local_t *name_loc;
723 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
725 entry = &leaf->entries[0];
726 bytes = sizeof(struct xfs_attr_sf_hdr);
727 for (i = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); entry++, i++) {
728 if (entry->flags & XFS_ATTR_INCOMPLETE)
729 continue; /* don't copy partial entries */
730 if (!(entry->flags & XFS_ATTR_LOCAL))
732 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
733 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
735 if (INT_GET(name_loc->valuelen, ARCH_CONVERT) >= XFS_ATTR_SF_ENTSIZE_MAX)
737 bytes += sizeof(struct xfs_attr_sf_entry)-1
739 + INT_GET(name_loc->valuelen, ARCH_CONVERT);
741 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
742 (bytes == sizeof(struct xfs_attr_sf_hdr)))
744 return(xfs_attr_shortform_bytesfit(dp, bytes));
748 * Convert a leaf attribute list to shortform attribute list
751 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
753 xfs_attr_leafblock_t *leaf;
754 xfs_attr_leaf_entry_t *entry;
755 xfs_attr_leaf_name_local_t *name_loc;
762 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
763 ASSERT(tmpbuffer != NULL);
766 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
767 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
768 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
769 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
772 * Clean out the prior contents of the attribute list.
774 error = xfs_da_shrink_inode(args, 0, bp);
779 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
782 * Last attribute was removed, revert to original
783 * inode format making all literal area available
784 * to the data fork once more.
786 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
787 dp->i_d.di_forkoff = 0;
788 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
789 ASSERT(dp->i_d.di_anextents == 0);
790 ASSERT(dp->i_afp == NULL);
791 dp->i_df.if_ext_max =
792 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
793 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
797 xfs_attr_shortform_create(args);
800 * Copy the attributes
802 memset((char *)&nargs, 0, sizeof(nargs));
804 nargs.firstblock = args->firstblock;
805 nargs.flist = args->flist;
806 nargs.total = args->total;
807 nargs.whichfork = XFS_ATTR_FORK;
808 nargs.trans = args->trans;
810 entry = &leaf->entries[0];
811 for (i = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); entry++, i++) {
812 if (entry->flags & XFS_ATTR_INCOMPLETE)
813 continue; /* don't copy partial entries */
816 ASSERT(entry->flags & XFS_ATTR_LOCAL);
817 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
818 nargs.name = (char *)name_loc->nameval;
819 nargs.namelen = name_loc->namelen;
820 nargs.value = (char *)&name_loc->nameval[nargs.namelen];
821 nargs.valuelen = INT_GET(name_loc->valuelen, ARCH_CONVERT);
822 nargs.hashval = INT_GET(entry->hashval, ARCH_CONVERT);
823 nargs.flags = (entry->flags & XFS_ATTR_SECURE) ? ATTR_SECURE :
824 ((entry->flags & XFS_ATTR_ROOT) ? ATTR_ROOT : 0);
825 xfs_attr_shortform_add(&nargs, forkoff);
830 kmem_free(tmpbuffer, XFS_LBSIZE(dp->i_mount));
835 * Convert from using a single leaf to a root node and a leaf.
838 xfs_attr_leaf_to_node(xfs_da_args_t *args)
840 xfs_attr_leafblock_t *leaf;
841 xfs_da_intnode_t *node;
843 xfs_dabuf_t *bp1, *bp2;
849 error = xfs_da_grow_inode(args, &blkno);
852 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
858 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
863 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
864 xfs_da_buf_done(bp1);
866 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
869 * Set up the new root node.
871 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
876 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
877 /* both on-disk, don't endian-flip twice */
878 node->btree[0].hashval =
879 leaf->entries[INT_GET(leaf->hdr.count, ARCH_CONVERT)-1 ].hashval;
880 INT_SET(node->btree[0].before, ARCH_CONVERT, blkno);
881 INT_SET(node->hdr.count, ARCH_CONVERT, 1);
882 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
886 xfs_da_buf_done(bp1);
888 xfs_da_buf_done(bp2);
893 /*========================================================================
894 * Routines used for growing the Btree.
895 *========================================================================*/
898 * Create the initial contents of a leaf attribute list
899 * or a leaf in a node attribute list.
902 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
904 xfs_attr_leafblock_t *leaf;
905 xfs_attr_leaf_hdr_t *hdr;
912 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
918 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
920 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
921 INT_SET(hdr->firstused, ARCH_CONVERT, XFS_LBSIZE(dp->i_mount));
922 if (!hdr->firstused) {
923 INT_SET(hdr->firstused, ARCH_CONVERT,
924 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
927 INT_SET(hdr->freemap[0].base, ARCH_CONVERT,
928 sizeof(xfs_attr_leaf_hdr_t));
929 INT_SET(hdr->freemap[0].size, ARCH_CONVERT,
930 INT_GET(hdr->firstused, ARCH_CONVERT)
931 - INT_GET(hdr->freemap[0].base,
934 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
941 * Split the leaf node, rebalance, then add the new entry.
944 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
945 xfs_da_state_blk_t *newblk)
951 * Allocate space for a new leaf node.
953 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
954 error = xfs_da_grow_inode(state->args, &blkno);
957 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
960 newblk->blkno = blkno;
961 newblk->magic = XFS_ATTR_LEAF_MAGIC;
964 * Rebalance the entries across the two leaves.
965 * NOTE: rebalance() currently depends on the 2nd block being empty.
967 xfs_attr_leaf_rebalance(state, oldblk, newblk);
968 error = xfs_da_blk_link(state, oldblk, newblk);
973 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
974 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
975 * "new" attrs info. Will need the "old" info to remove it later.
977 * Insert the "new" entry in the correct block.
980 error = xfs_attr_leaf_add(oldblk->bp, state->args);
982 error = xfs_attr_leaf_add(newblk->bp, state->args);
985 * Update last hashval in each block since we added the name.
987 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
988 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
993 * Add a name to the leaf attribute list structure.
996 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
998 xfs_attr_leafblock_t *leaf;
999 xfs_attr_leaf_hdr_t *hdr;
1000 xfs_attr_leaf_map_t *map;
1001 int tablesize, entsize, sum, tmp, i;
1004 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1005 ASSERT((args->index >= 0)
1006 && (args->index <= INT_GET(leaf->hdr.count, ARCH_CONVERT)));
1008 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1009 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1012 * Search through freemap for first-fit on new name length.
1013 * (may need to figure in size of entry struct too)
1015 tablesize = (INT_GET(hdr->count, ARCH_CONVERT) + 1)
1016 * sizeof(xfs_attr_leaf_entry_t)
1017 + sizeof(xfs_attr_leaf_hdr_t);
1018 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1019 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1020 if (tablesize > INT_GET(hdr->firstused, ARCH_CONVERT)) {
1021 sum += INT_GET(map->size, ARCH_CONVERT);
1025 continue; /* no space in this map */
1027 if (INT_GET(map->base, ARCH_CONVERT)
1028 < INT_GET(hdr->firstused, ARCH_CONVERT))
1029 tmp += sizeof(xfs_attr_leaf_entry_t);
1030 if (INT_GET(map->size, ARCH_CONVERT) >= tmp) {
1031 tmp = xfs_attr_leaf_add_work(bp, args, i);
1034 sum += INT_GET(map->size, ARCH_CONVERT);
1038 * If there are no holes in the address space of the block,
1039 * and we don't have enough freespace, then compaction will do us
1040 * no good and we should just give up.
1042 if (!hdr->holes && (sum < entsize))
1043 return(XFS_ERROR(ENOSPC));
1046 * Compact the entries to coalesce free space.
1047 * This may change the hdr->count via dropping INCOMPLETE entries.
1049 xfs_attr_leaf_compact(args->trans, bp);
1052 * After compaction, the block is guaranteed to have only one
1053 * free region, in freemap[0]. If it is not big enough, give up.
1055 if (INT_GET(hdr->freemap[0].size, ARCH_CONVERT)
1056 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1057 return(XFS_ERROR(ENOSPC));
1059 return(xfs_attr_leaf_add_work(bp, args, 0));
1063 * Add a name to a leaf attribute list structure.
1066 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1068 xfs_attr_leafblock_t *leaf;
1069 xfs_attr_leaf_hdr_t *hdr;
1070 xfs_attr_leaf_entry_t *entry;
1071 xfs_attr_leaf_name_local_t *name_loc;
1072 xfs_attr_leaf_name_remote_t *name_rmt;
1073 xfs_attr_leaf_map_t *map;
1078 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1080 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1081 ASSERT((args->index >= 0)
1082 && (args->index <= INT_GET(hdr->count, ARCH_CONVERT)));
1085 * Force open some space in the entry array and fill it in.
1087 entry = &leaf->entries[args->index];
1088 if (args->index < INT_GET(hdr->count, ARCH_CONVERT)) {
1089 tmp = INT_GET(hdr->count, ARCH_CONVERT) - args->index;
1090 tmp *= sizeof(xfs_attr_leaf_entry_t);
1091 memmove((char *)(entry+1), (char *)entry, tmp);
1092 xfs_da_log_buf(args->trans, bp,
1093 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1095 INT_MOD(hdr->count, ARCH_CONVERT, 1);
1098 * Allocate space for the new string (at the end of the run).
1100 map = &hdr->freemap[mapindex];
1101 mp = args->trans->t_mountp;
1102 ASSERT(INT_GET(map->base, ARCH_CONVERT) < XFS_LBSIZE(mp));
1103 ASSERT((INT_GET(map->base, ARCH_CONVERT) & 0x3) == 0);
1104 ASSERT(INT_GET(map->size, ARCH_CONVERT) >=
1105 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1106 mp->m_sb.sb_blocksize, NULL));
1107 ASSERT(INT_GET(map->size, ARCH_CONVERT) < XFS_LBSIZE(mp));
1108 ASSERT((INT_GET(map->size, ARCH_CONVERT) & 0x3) == 0);
1109 INT_MOD(map->size, ARCH_CONVERT,
1110 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1111 mp->m_sb.sb_blocksize, &tmp));
1112 INT_SET(entry->nameidx, ARCH_CONVERT,
1113 INT_GET(map->base, ARCH_CONVERT)
1114 + INT_GET(map->size, ARCH_CONVERT));
1115 INT_SET(entry->hashval, ARCH_CONVERT, args->hashval);
1116 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1117 entry->flags |= (args->flags & ATTR_SECURE) ? XFS_ATTR_SECURE :
1118 ((args->flags & ATTR_ROOT) ? XFS_ATTR_ROOT : 0);
1120 entry->flags |= XFS_ATTR_INCOMPLETE;
1121 if ((args->blkno2 == args->blkno) &&
1122 (args->index2 <= args->index)) {
1126 xfs_da_log_buf(args->trans, bp,
1127 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1128 ASSERT((args->index == 0) || (INT_GET(entry->hashval, ARCH_CONVERT)
1129 >= INT_GET((entry-1)->hashval,
1131 ASSERT((args->index == INT_GET(hdr->count, ARCH_CONVERT)-1) ||
1132 (INT_GET(entry->hashval, ARCH_CONVERT)
1133 <= (INT_GET((entry+1)->hashval, ARCH_CONVERT))));
1136 * Copy the attribute name and value into the new space.
1138 * For "remote" attribute values, simply note that we need to
1139 * allocate space for the "remote" value. We can't actually
1140 * allocate the extents in this transaction, and we can't decide
1141 * which blocks they should be as we might allocate more blocks
1142 * as part of this transaction (a split operation for example).
1144 if (entry->flags & XFS_ATTR_LOCAL) {
1145 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
1146 name_loc->namelen = args->namelen;
1147 INT_SET(name_loc->valuelen, ARCH_CONVERT, args->valuelen);
1148 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1149 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1150 INT_GET(name_loc->valuelen, ARCH_CONVERT));
1152 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
1153 name_rmt->namelen = args->namelen;
1154 memcpy((char *)name_rmt->name, args->name, args->namelen);
1155 entry->flags |= XFS_ATTR_INCOMPLETE;
1157 name_rmt->valuelen = 0;
1158 name_rmt->valueblk = 0;
1160 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1162 xfs_da_log_buf(args->trans, bp,
1163 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1164 xfs_attr_leaf_entsize(leaf, args->index)));
1167 * Update the control info for this leaf node
1169 if (INT_GET(entry->nameidx, ARCH_CONVERT)
1170 < INT_GET(hdr->firstused, ARCH_CONVERT)) {
1171 /* both on-disk, don't endian-flip twice */
1172 hdr->firstused = entry->nameidx;
1174 ASSERT(INT_GET(hdr->firstused, ARCH_CONVERT)
1175 >= ((INT_GET(hdr->count, ARCH_CONVERT)
1176 * sizeof(*entry))+sizeof(*hdr)));
1177 tmp = (INT_GET(hdr->count, ARCH_CONVERT)-1)
1178 * sizeof(xfs_attr_leaf_entry_t)
1179 + sizeof(xfs_attr_leaf_hdr_t);
1180 map = &hdr->freemap[0];
1181 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1182 if (INT_GET(map->base, ARCH_CONVERT) == tmp) {
1183 INT_MOD(map->base, ARCH_CONVERT,
1184 sizeof(xfs_attr_leaf_entry_t));
1185 INT_MOD(map->size, ARCH_CONVERT,
1186 -sizeof(xfs_attr_leaf_entry_t));
1189 INT_MOD(hdr->usedbytes, ARCH_CONVERT,
1190 xfs_attr_leaf_entsize(leaf, args->index));
1191 xfs_da_log_buf(args->trans, bp,
1192 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1197 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1200 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1202 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1203 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1207 mp = trans->t_mountp;
1208 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1209 ASSERT(tmpbuffer != NULL);
1210 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1211 memset(bp->data, 0, XFS_LBSIZE(mp));
1214 * Copy basic information
1216 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1218 hdr_s = &leaf_s->hdr;
1219 hdr_d = &leaf_d->hdr;
1220 hdr_d->info = hdr_s->info; /* struct copy */
1221 INT_SET(hdr_d->firstused, ARCH_CONVERT, XFS_LBSIZE(mp));
1222 /* handle truncation gracefully */
1223 if (!hdr_d->firstused) {
1224 INT_SET(hdr_d->firstused, ARCH_CONVERT,
1225 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1227 hdr_d->usedbytes = 0;
1230 INT_SET(hdr_d->freemap[0].base, ARCH_CONVERT,
1231 sizeof(xfs_attr_leaf_hdr_t));
1232 INT_SET(hdr_d->freemap[0].size, ARCH_CONVERT,
1233 INT_GET(hdr_d->firstused, ARCH_CONVERT)
1234 - INT_GET(hdr_d->freemap[0].base, ARCH_CONVERT));
1237 * Copy all entry's in the same (sorted) order,
1238 * but allocate name/value pairs packed and in sequence.
1240 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1241 (int)INT_GET(hdr_s->count, ARCH_CONVERT), mp);
1243 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1245 kmem_free(tmpbuffer, XFS_LBSIZE(mp));
1249 * Redistribute the attribute list entries between two leaf nodes,
1250 * taking into account the size of the new entry.
1252 * NOTE: if new block is empty, then it will get the upper half of the
1253 * old block. At present, all (one) callers pass in an empty second block.
1255 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1256 * to match what it is doing in splitting the attribute leaf block. Those
1257 * values are used in "atomic rename" operations on attributes. Note that
1258 * the "new" and "old" values can end up in different blocks.
1261 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1262 xfs_da_state_blk_t *blk2)
1264 xfs_da_args_t *args;
1265 xfs_da_state_blk_t *tmp_blk;
1266 xfs_attr_leafblock_t *leaf1, *leaf2;
1267 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1268 int count, totallen, max, space, swap;
1271 * Set up environment.
1273 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1274 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1275 leaf1 = blk1->bp->data;
1276 leaf2 = blk2->bp->data;
1277 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1278 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1282 * Check ordering of blocks, reverse if it makes things simpler.
1284 * NOTE: Given that all (current) callers pass in an empty
1285 * second block, this code should never set "swap".
1288 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1292 leaf1 = blk1->bp->data;
1293 leaf2 = blk2->bp->data;
1300 * Examine entries until we reduce the absolute difference in
1301 * byte usage between the two blocks to a minimum. Then get
1302 * the direction to copy and the number of elements to move.
1304 * "inleaf" is true if the new entry should be inserted into blk1.
1305 * If "swap" is also true, then reverse the sense of "inleaf".
1307 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1310 state->inleaf = !state->inleaf;
1313 * Move any entries required from leaf to leaf:
1315 if (count < INT_GET(hdr1->count, ARCH_CONVERT)) {
1317 * Figure the total bytes to be added to the destination leaf.
1319 /* number entries being moved */
1320 count = INT_GET(hdr1->count, ARCH_CONVERT) - count;
1321 space = INT_GET(hdr1->usedbytes, ARCH_CONVERT) - totallen;
1322 space += count * sizeof(xfs_attr_leaf_entry_t);
1325 * leaf2 is the destination, compact it if it looks tight.
1327 max = INT_GET(hdr2->firstused, ARCH_CONVERT)
1328 - sizeof(xfs_attr_leaf_hdr_t);
1329 max -= INT_GET(hdr2->count, ARCH_CONVERT)
1330 * sizeof(xfs_attr_leaf_entry_t);
1332 xfs_attr_leaf_compact(args->trans, blk2->bp);
1336 * Move high entries from leaf1 to low end of leaf2.
1338 xfs_attr_leaf_moveents(leaf1,
1339 INT_GET(hdr1->count, ARCH_CONVERT)-count,
1340 leaf2, 0, count, state->mp);
1342 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1343 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1344 } else if (count > INT_GET(hdr1->count, ARCH_CONVERT)) {
1346 * I assert that since all callers pass in an empty
1347 * second buffer, this code should never execute.
1351 * Figure the total bytes to be added to the destination leaf.
1353 /* number entries being moved */
1354 count -= INT_GET(hdr1->count, ARCH_CONVERT);
1355 space = totallen - INT_GET(hdr1->usedbytes, ARCH_CONVERT);
1356 space += count * sizeof(xfs_attr_leaf_entry_t);
1359 * leaf1 is the destination, compact it if it looks tight.
1361 max = INT_GET(hdr1->firstused, ARCH_CONVERT)
1362 - sizeof(xfs_attr_leaf_hdr_t);
1363 max -= INT_GET(hdr1->count, ARCH_CONVERT)
1364 * sizeof(xfs_attr_leaf_entry_t);
1366 xfs_attr_leaf_compact(args->trans, blk1->bp);
1370 * Move low entries from leaf2 to high end of leaf1.
1372 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1373 (int)INT_GET(hdr1->count, ARCH_CONVERT), count,
1376 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1377 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1381 * Copy out last hashval in each block for B-tree code.
1384 INT_GET(leaf1->entries[INT_GET(leaf1->hdr.count,
1385 ARCH_CONVERT)-1].hashval, ARCH_CONVERT);
1387 INT_GET(leaf2->entries[INT_GET(leaf2->hdr.count,
1388 ARCH_CONVERT)-1].hashval, ARCH_CONVERT);
1391 * Adjust the expected index for insertion.
1392 * NOTE: this code depends on the (current) situation that the
1393 * second block was originally empty.
1395 * If the insertion point moved to the 2nd block, we must adjust
1396 * the index. We must also track the entry just following the
1397 * new entry for use in an "atomic rename" operation, that entry
1398 * is always the "old" entry and the "new" entry is what we are
1399 * inserting. The index/blkno fields refer to the "old" entry,
1400 * while the index2/blkno2 fields refer to the "new" entry.
1402 if (blk1->index > INT_GET(leaf1->hdr.count, ARCH_CONVERT)) {
1403 ASSERT(state->inleaf == 0);
1404 blk2->index = blk1->index
1405 - INT_GET(leaf1->hdr.count, ARCH_CONVERT);
1406 args->index = args->index2 = blk2->index;
1407 args->blkno = args->blkno2 = blk2->blkno;
1408 } else if (blk1->index == INT_GET(leaf1->hdr.count, ARCH_CONVERT)) {
1409 if (state->inleaf) {
1410 args->index = blk1->index;
1411 args->blkno = blk1->blkno;
1413 args->blkno2 = blk2->blkno;
1415 blk2->index = blk1->index
1416 - INT_GET(leaf1->hdr.count, ARCH_CONVERT);
1417 args->index = args->index2 = blk2->index;
1418 args->blkno = args->blkno2 = blk2->blkno;
1421 ASSERT(state->inleaf == 1);
1422 args->index = args->index2 = blk1->index;
1423 args->blkno = args->blkno2 = blk1->blkno;
1428 * Examine entries until we reduce the absolute difference in
1429 * byte usage between the two blocks to a minimum.
1430 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1431 * GROT: there will always be enough room in either block for a new entry.
1432 * GROT: Do a double-split for this case?
1435 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1436 xfs_da_state_blk_t *blk1,
1437 xfs_da_state_blk_t *blk2,
1438 int *countarg, int *usedbytesarg)
1440 xfs_attr_leafblock_t *leaf1, *leaf2;
1441 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1442 xfs_attr_leaf_entry_t *entry;
1443 int count, max, index, totallen, half;
1444 int lastdelta, foundit, tmp;
1447 * Set up environment.
1449 leaf1 = blk1->bp->data;
1450 leaf2 = blk2->bp->data;
1457 * Examine entries until we reduce the absolute difference in
1458 * byte usage between the two blocks to a minimum.
1460 max = INT_GET(hdr1->count, ARCH_CONVERT)
1461 + INT_GET(hdr2->count, ARCH_CONVERT);
1462 half = (max+1) * sizeof(*entry);
1463 half += INT_GET(hdr1->usedbytes, ARCH_CONVERT)
1464 + INT_GET(hdr2->usedbytes, ARCH_CONVERT)
1465 + xfs_attr_leaf_newentsize(
1466 state->args->namelen,
1467 state->args->valuelen,
1468 state->blocksize, NULL);
1470 lastdelta = state->blocksize;
1471 entry = &leaf1->entries[0];
1472 for (count = index = 0; count < max; entry++, index++, count++) {
1474 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1476 * The new entry is in the first block, account for it.
1478 if (count == blk1->index) {
1479 tmp = totallen + sizeof(*entry) +
1480 xfs_attr_leaf_newentsize(
1481 state->args->namelen,
1482 state->args->valuelen,
1483 state->blocksize, NULL);
1484 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1486 lastdelta = XFS_ATTR_ABS(half - tmp);
1492 * Wrap around into the second block if necessary.
1494 if (count == INT_GET(hdr1->count, ARCH_CONVERT)) {
1496 entry = &leaf1->entries[0];
1501 * Figure out if next leaf entry would be too much.
1503 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1505 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1507 lastdelta = XFS_ATTR_ABS(half - tmp);
1513 * Calculate the number of usedbytes that will end up in lower block.
1514 * If new entry not in lower block, fix up the count.
1516 totallen -= count * sizeof(*entry);
1518 totallen -= sizeof(*entry) +
1519 xfs_attr_leaf_newentsize(
1520 state->args->namelen,
1521 state->args->valuelen,
1522 state->blocksize, NULL);
1526 *usedbytesarg = totallen;
1530 /*========================================================================
1531 * Routines used for shrinking the Btree.
1532 *========================================================================*/
1535 * Check a leaf block and its neighbors to see if the block should be
1536 * collapsed into one or the other neighbor. Always keep the block
1537 * with the smaller block number.
1538 * If the current block is over 50% full, don't try to join it, return 0.
1539 * If the block is empty, fill in the state structure and return 2.
1540 * If it can be collapsed, fill in the state structure and return 1.
1541 * If nothing can be done, return 0.
1543 * GROT: allow for INCOMPLETE entries in calculation.
1546 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1548 xfs_attr_leafblock_t *leaf;
1549 xfs_da_state_blk_t *blk;
1550 xfs_da_blkinfo_t *info;
1551 int count, bytes, forward, error, retval, i;
1556 * Check for the degenerate case of the block being over 50% full.
1557 * If so, it's not worth even looking to see if we might be able
1558 * to coalesce with a sibling.
1560 blk = &state->path.blk[ state->path.active-1 ];
1561 info = blk->bp->data;
1562 ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1563 leaf = (xfs_attr_leafblock_t *)info;
1564 count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
1565 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1566 count * sizeof(xfs_attr_leaf_entry_t) +
1567 INT_GET(leaf->hdr.usedbytes, ARCH_CONVERT);
1568 if (bytes > (state->blocksize >> 1)) {
1569 *action = 0; /* blk over 50%, don't try to join */
1574 * Check for the degenerate case of the block being empty.
1575 * If the block is empty, we'll simply delete it, no need to
1576 * coalesce it with a sibling block. We choose (aribtrarily)
1577 * to merge with the forward block unless it is NULL.
1581 * Make altpath point to the block we want to keep and
1582 * path point to the block we want to drop (this one).
1584 forward = (info->forw != 0);
1585 memcpy(&state->altpath, &state->path, sizeof(state->path));
1586 error = xfs_da_path_shift(state, &state->altpath, forward,
1599 * Examine each sibling block to see if we can coalesce with
1600 * at least 25% free space to spare. We need to figure out
1601 * whether to merge with the forward or the backward block.
1602 * We prefer coalescing with the lower numbered sibling so as
1603 * to shrink an attribute list over time.
1605 /* start with smaller blk num */
1606 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1607 for (i = 0; i < 2; forward = !forward, i++) {
1609 blkno = be32_to_cpu(info->forw);
1611 blkno = be32_to_cpu(info->back);
1614 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1615 blkno, -1, &bp, XFS_ATTR_FORK);
1620 leaf = (xfs_attr_leafblock_t *)info;
1621 count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
1622 bytes = state->blocksize - (state->blocksize>>2);
1623 bytes -= INT_GET(leaf->hdr.usedbytes, ARCH_CONVERT);
1625 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1626 count += INT_GET(leaf->hdr.count, ARCH_CONVERT);
1627 bytes -= INT_GET(leaf->hdr.usedbytes, ARCH_CONVERT);
1628 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1629 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1630 xfs_da_brelse(state->args->trans, bp);
1632 break; /* fits with at least 25% to spare */
1640 * Make altpath point to the block we want to keep (the lower
1641 * numbered block) and path point to the block we want to drop.
1643 memcpy(&state->altpath, &state->path, sizeof(state->path));
1644 if (blkno < blk->blkno) {
1645 error = xfs_da_path_shift(state, &state->altpath, forward,
1648 error = xfs_da_path_shift(state, &state->path, forward,
1662 * Remove a name from the leaf attribute list structure.
1664 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1665 * If two leaves are 37% full, when combined they will leave 25% free.
1668 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1670 xfs_attr_leafblock_t *leaf;
1671 xfs_attr_leaf_hdr_t *hdr;
1672 xfs_attr_leaf_map_t *map;
1673 xfs_attr_leaf_entry_t *entry;
1674 int before, after, smallest, entsize;
1675 int tablesize, tmp, i;
1679 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1681 mp = args->trans->t_mountp;
1682 ASSERT((INT_GET(hdr->count, ARCH_CONVERT) > 0)
1683 && (INT_GET(hdr->count, ARCH_CONVERT) < (XFS_LBSIZE(mp)/8)));
1684 ASSERT((args->index >= 0)
1685 && (args->index < INT_GET(hdr->count, ARCH_CONVERT)));
1686 ASSERT(INT_GET(hdr->firstused, ARCH_CONVERT)
1687 >= ((INT_GET(hdr->count, ARCH_CONVERT)
1688 * sizeof(*entry))+sizeof(*hdr)));
1689 entry = &leaf->entries[args->index];
1690 ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT)
1691 >= INT_GET(hdr->firstused, ARCH_CONVERT));
1692 ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT) < XFS_LBSIZE(mp));
1695 * Scan through free region table:
1696 * check for adjacency of free'd entry with an existing one,
1697 * find smallest free region in case we need to replace it,
1698 * adjust any map that borders the entry table,
1700 tablesize = INT_GET(hdr->count, ARCH_CONVERT)
1701 * sizeof(xfs_attr_leaf_entry_t)
1702 + sizeof(xfs_attr_leaf_hdr_t);
1703 map = &hdr->freemap[0];
1704 tmp = INT_GET(map->size, ARCH_CONVERT);
1705 before = after = -1;
1706 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1707 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1708 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1709 ASSERT(INT_GET(map->base, ARCH_CONVERT) < XFS_LBSIZE(mp));
1710 ASSERT(INT_GET(map->size, ARCH_CONVERT) < XFS_LBSIZE(mp));
1711 if (INT_GET(map->base, ARCH_CONVERT) == tablesize) {
1712 INT_MOD(map->base, ARCH_CONVERT,
1713 -sizeof(xfs_attr_leaf_entry_t));
1714 INT_MOD(map->size, ARCH_CONVERT,
1715 sizeof(xfs_attr_leaf_entry_t));
1718 if ((INT_GET(map->base, ARCH_CONVERT)
1719 + INT_GET(map->size, ARCH_CONVERT))
1720 == INT_GET(entry->nameidx, ARCH_CONVERT)) {
1722 } else if (INT_GET(map->base, ARCH_CONVERT)
1723 == (INT_GET(entry->nameidx, ARCH_CONVERT) + entsize)) {
1725 } else if (INT_GET(map->size, ARCH_CONVERT) < tmp) {
1726 tmp = INT_GET(map->size, ARCH_CONVERT);
1732 * Coalesce adjacent freemap regions,
1733 * or replace the smallest region.
1735 if ((before >= 0) || (after >= 0)) {
1736 if ((before >= 0) && (after >= 0)) {
1737 map = &hdr->freemap[before];
1738 INT_MOD(map->size, ARCH_CONVERT, entsize);
1739 INT_MOD(map->size, ARCH_CONVERT,
1740 INT_GET(hdr->freemap[after].size,
1742 hdr->freemap[after].base = 0;
1743 hdr->freemap[after].size = 0;
1744 } else if (before >= 0) {
1745 map = &hdr->freemap[before];
1746 INT_MOD(map->size, ARCH_CONVERT, entsize);
1748 map = &hdr->freemap[after];
1749 /* both on-disk, don't endian flip twice */
1750 map->base = entry->nameidx;
1751 INT_MOD(map->size, ARCH_CONVERT, entsize);
1755 * Replace smallest region (if it is smaller than free'd entry)
1757 map = &hdr->freemap[smallest];
1758 if (INT_GET(map->size, ARCH_CONVERT) < entsize) {
1759 INT_SET(map->base, ARCH_CONVERT,
1760 INT_GET(entry->nameidx, ARCH_CONVERT));
1761 INT_SET(map->size, ARCH_CONVERT, entsize);
1766 * Did we remove the first entry?
1768 if (INT_GET(entry->nameidx, ARCH_CONVERT)
1769 == INT_GET(hdr->firstused, ARCH_CONVERT))
1775 * Compress the remaining entries and zero out the removed stuff.
1777 memset(XFS_ATTR_LEAF_NAME(leaf, args->index), 0, entsize);
1778 INT_MOD(hdr->usedbytes, ARCH_CONVERT, -entsize);
1779 xfs_da_log_buf(args->trans, bp,
1780 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1783 tmp = (INT_GET(hdr->count, ARCH_CONVERT) - args->index)
1784 * sizeof(xfs_attr_leaf_entry_t);
1785 memmove((char *)entry, (char *)(entry+1), tmp);
1786 INT_MOD(hdr->count, ARCH_CONVERT, -1);
1787 xfs_da_log_buf(args->trans, bp,
1788 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1789 entry = &leaf->entries[INT_GET(hdr->count, ARCH_CONVERT)];
1790 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1793 * If we removed the first entry, re-find the first used byte
1794 * in the name area. Note that if the entry was the "firstused",
1795 * then we don't have a "hole" in our block resulting from
1796 * removing the name.
1799 tmp = XFS_LBSIZE(mp);
1800 entry = &leaf->entries[0];
1801 for (i = INT_GET(hdr->count, ARCH_CONVERT)-1;
1802 i >= 0; entry++, i--) {
1803 ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT)
1804 >= INT_GET(hdr->firstused, ARCH_CONVERT));
1805 ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT)
1807 if (INT_GET(entry->nameidx, ARCH_CONVERT) < tmp)
1808 tmp = INT_GET(entry->nameidx, ARCH_CONVERT);
1810 INT_SET(hdr->firstused, ARCH_CONVERT, tmp);
1811 if (!hdr->firstused) {
1812 INT_SET(hdr->firstused, ARCH_CONVERT,
1813 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1816 hdr->holes = 1; /* mark as needing compaction */
1818 xfs_da_log_buf(args->trans, bp,
1819 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1822 * Check if leaf is less than 50% full, caller may want to
1823 * "join" the leaf with a sibling if so.
1825 tmp = sizeof(xfs_attr_leaf_hdr_t);
1826 tmp += INT_GET(leaf->hdr.count, ARCH_CONVERT)
1827 * sizeof(xfs_attr_leaf_entry_t);
1828 tmp += INT_GET(leaf->hdr.usedbytes, ARCH_CONVERT);
1829 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1833 * Move all the attribute list entries from drop_leaf into save_leaf.
1836 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1837 xfs_da_state_blk_t *save_blk)
1839 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1840 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1845 * Set up environment.
1848 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1849 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1850 drop_leaf = drop_blk->bp->data;
1851 save_leaf = save_blk->bp->data;
1852 ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1853 ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1854 drop_hdr = &drop_leaf->hdr;
1855 save_hdr = &save_leaf->hdr;
1858 * Save last hashval from dying block for later Btree fixup.
1861 INT_GET(drop_leaf->entries[INT_GET(drop_leaf->hdr.count,
1862 ARCH_CONVERT)-1].hashval,
1866 * Check if we need a temp buffer, or can we do it in place.
1867 * Note that we don't check "leaf" for holes because we will
1868 * always be dropping it, toosmall() decided that for us already.
1870 if (save_hdr->holes == 0) {
1872 * dest leaf has no holes, so we add there. May need
1873 * to make some room in the entry array.
1875 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1876 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1877 (int)INT_GET(drop_hdr->count, ARCH_CONVERT), mp);
1879 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1880 INT_GET(save_hdr->count, ARCH_CONVERT),
1881 (int)INT_GET(drop_hdr->count, ARCH_CONVERT),
1886 * Destination has holes, so we make a temporary copy
1887 * of the leaf and add them both to that.
1889 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1890 ASSERT(tmpbuffer != NULL);
1891 memset(tmpbuffer, 0, state->blocksize);
1892 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1893 tmp_hdr = &tmp_leaf->hdr;
1894 tmp_hdr->info = save_hdr->info; /* struct copy */
1896 INT_SET(tmp_hdr->firstused, ARCH_CONVERT, state->blocksize);
1897 if (!tmp_hdr->firstused) {
1898 INT_SET(tmp_hdr->firstused, ARCH_CONVERT,
1899 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1901 tmp_hdr->usedbytes = 0;
1902 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1903 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1904 (int)INT_GET(drop_hdr->count, ARCH_CONVERT),
1906 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1907 INT_GET(tmp_leaf->hdr.count, ARCH_CONVERT),
1908 (int)INT_GET(save_hdr->count, ARCH_CONVERT),
1911 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1912 (int)INT_GET(save_hdr->count, ARCH_CONVERT),
1914 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1915 INT_GET(tmp_leaf->hdr.count, ARCH_CONVERT),
1916 (int)INT_GET(drop_hdr->count, ARCH_CONVERT),
1919 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1920 kmem_free(tmpbuffer, state->blocksize);
1923 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1924 state->blocksize - 1);
1927 * Copy out last hashval in each block for B-tree code.
1930 INT_GET(save_leaf->entries[INT_GET(save_leaf->hdr.count,
1931 ARCH_CONVERT)-1].hashval,
1935 /*========================================================================
1936 * Routines used for finding things in the Btree.
1937 *========================================================================*/
1940 * Look up a name in a leaf attribute list structure.
1941 * This is the internal routine, it uses the caller's buffer.
1943 * Note that duplicate keys are allowed, but only check within the
1944 * current leaf node. The Btree code must check in adjacent leaf nodes.
1946 * Return in args->index the index into the entry[] array of either
1947 * the found entry, or where the entry should have been (insert before
1950 * Don't change the args->value unless we find the attribute.
1953 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1955 xfs_attr_leafblock_t *leaf;
1956 xfs_attr_leaf_entry_t *entry;
1957 xfs_attr_leaf_name_local_t *name_loc;
1958 xfs_attr_leaf_name_remote_t *name_rmt;
1960 xfs_dahash_t hashval;
1963 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1964 ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT)
1965 < (XFS_LBSIZE(args->dp->i_mount)/8));
1968 * Binary search. (note: small blocks will skip this loop)
1970 hashval = args->hashval;
1971 probe = span = INT_GET(leaf->hdr.count, ARCH_CONVERT) / 2;
1972 for (entry = &leaf->entries[probe]; span > 4;
1973 entry = &leaf->entries[probe]) {
1975 if (INT_GET(entry->hashval, ARCH_CONVERT) < hashval)
1977 else if (INT_GET(entry->hashval, ARCH_CONVERT) > hashval)
1982 ASSERT((probe >= 0) &&
1984 || (probe < INT_GET(leaf->hdr.count, ARCH_CONVERT))));
1985 ASSERT((span <= 4) || (INT_GET(entry->hashval, ARCH_CONVERT)
1989 * Since we may have duplicate hashval's, find the first matching
1990 * hashval in the leaf.
1992 while ((probe > 0) && (INT_GET(entry->hashval, ARCH_CONVERT)
1997 while ((probe < INT_GET(leaf->hdr.count, ARCH_CONVERT))
1998 && (INT_GET(entry->hashval, ARCH_CONVERT) < hashval)) {
2002 if ((probe == INT_GET(leaf->hdr.count, ARCH_CONVERT))
2003 || (INT_GET(entry->hashval, ARCH_CONVERT) != hashval)) {
2004 args->index = probe;
2005 return(XFS_ERROR(ENOATTR));
2009 * Duplicate keys may be present, so search all of them for a match.
2011 for ( ; (probe < INT_GET(leaf->hdr.count, ARCH_CONVERT))
2012 && (INT_GET(entry->hashval, ARCH_CONVERT) == hashval);
2015 * GROT: Add code to remove incomplete entries.
2018 * If we are looking for INCOMPLETE entries, show only those.
2019 * If we are looking for complete entries, show only those.
2021 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2022 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2025 if (entry->flags & XFS_ATTR_LOCAL) {
2026 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, probe);
2027 if (name_loc->namelen != args->namelen)
2029 if (memcmp(args->name, (char *)name_loc->nameval,
2030 args->namelen) != 0)
2032 if (((args->flags & ATTR_SECURE) != 0) !=
2033 ((entry->flags & XFS_ATTR_SECURE) != 0))
2035 if (((args->flags & ATTR_ROOT) != 0) !=
2036 ((entry->flags & XFS_ATTR_ROOT) != 0))
2038 args->index = probe;
2039 return(XFS_ERROR(EEXIST));
2041 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, probe);
2042 if (name_rmt->namelen != args->namelen)
2044 if (memcmp(args->name, (char *)name_rmt->name,
2045 args->namelen) != 0)
2047 if (((args->flags & ATTR_SECURE) != 0) !=
2048 ((entry->flags & XFS_ATTR_SECURE) != 0))
2050 if (((args->flags & ATTR_ROOT) != 0) !=
2051 ((entry->flags & XFS_ATTR_ROOT) != 0))
2053 args->index = probe;
2055 = INT_GET(name_rmt->valueblk, ARCH_CONVERT);
2056 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2057 INT_GET(name_rmt->valuelen,
2059 return(XFS_ERROR(EEXIST));
2062 args->index = probe;
2063 return(XFS_ERROR(ENOATTR));
2067 * Get the value associated with an attribute name from a leaf attribute
2071 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2074 xfs_attr_leafblock_t *leaf;
2075 xfs_attr_leaf_entry_t *entry;
2076 xfs_attr_leaf_name_local_t *name_loc;
2077 xfs_attr_leaf_name_remote_t *name_rmt;
2080 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2081 ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT)
2082 < (XFS_LBSIZE(args->dp->i_mount)/8));
2083 ASSERT(args->index < ((int)INT_GET(leaf->hdr.count, ARCH_CONVERT)));
2085 entry = &leaf->entries[args->index];
2086 if (entry->flags & XFS_ATTR_LOCAL) {
2087 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2088 ASSERT(name_loc->namelen == args->namelen);
2089 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2090 valuelen = INT_GET(name_loc->valuelen, ARCH_CONVERT);
2091 if (args->flags & ATTR_KERNOVAL) {
2092 args->valuelen = valuelen;
2095 if (args->valuelen < valuelen) {
2096 args->valuelen = valuelen;
2097 return(XFS_ERROR(ERANGE));
2099 args->valuelen = valuelen;
2100 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2102 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2103 ASSERT(name_rmt->namelen == args->namelen);
2104 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2105 valuelen = INT_GET(name_rmt->valuelen, ARCH_CONVERT);
2106 args->rmtblkno = INT_GET(name_rmt->valueblk, ARCH_CONVERT);
2107 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2108 if (args->flags & ATTR_KERNOVAL) {
2109 args->valuelen = valuelen;
2112 if (args->valuelen < valuelen) {
2113 args->valuelen = valuelen;
2114 return(XFS_ERROR(ERANGE));
2116 args->valuelen = valuelen;
2121 /*========================================================================
2123 *========================================================================*/
2126 * Move the indicated entries from one leaf to another.
2127 * NOTE: this routine modifies both source and destination leaves.
2131 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2132 xfs_attr_leafblock_t *leaf_d, int start_d,
2133 int count, xfs_mount_t *mp)
2135 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2136 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2140 * Check for nothing to do.
2146 * Set up environment.
2148 ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2149 ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2150 hdr_s = &leaf_s->hdr;
2151 hdr_d = &leaf_d->hdr;
2152 ASSERT((INT_GET(hdr_s->count, ARCH_CONVERT) > 0)
2153 && (INT_GET(hdr_s->count, ARCH_CONVERT)
2154 < (XFS_LBSIZE(mp)/8)));
2155 ASSERT(INT_GET(hdr_s->firstused, ARCH_CONVERT) >=
2156 ((INT_GET(hdr_s->count, ARCH_CONVERT)
2157 * sizeof(*entry_s))+sizeof(*hdr_s)));
2158 ASSERT(INT_GET(hdr_d->count, ARCH_CONVERT) < (XFS_LBSIZE(mp)/8));
2159 ASSERT(INT_GET(hdr_d->firstused, ARCH_CONVERT) >=
2160 ((INT_GET(hdr_d->count, ARCH_CONVERT)
2161 * sizeof(*entry_d))+sizeof(*hdr_d)));
2163 ASSERT(start_s < INT_GET(hdr_s->count, ARCH_CONVERT));
2164 ASSERT(start_d <= INT_GET(hdr_d->count, ARCH_CONVERT));
2165 ASSERT(count <= INT_GET(hdr_s->count, ARCH_CONVERT));
2168 * Move the entries in the destination leaf up to make a hole?
2170 if (start_d < INT_GET(hdr_d->count, ARCH_CONVERT)) {
2171 tmp = INT_GET(hdr_d->count, ARCH_CONVERT) - start_d;
2172 tmp *= sizeof(xfs_attr_leaf_entry_t);
2173 entry_s = &leaf_d->entries[start_d];
2174 entry_d = &leaf_d->entries[start_d + count];
2175 memmove((char *)entry_d, (char *)entry_s, tmp);
2179 * Copy all entry's in the same (sorted) order,
2180 * but allocate attribute info packed and in sequence.
2182 entry_s = &leaf_s->entries[start_s];
2183 entry_d = &leaf_d->entries[start_d];
2185 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2186 ASSERT(INT_GET(entry_s->nameidx, ARCH_CONVERT)
2187 >= INT_GET(hdr_s->firstused, ARCH_CONVERT));
2188 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2191 * Code to drop INCOMPLETE entries. Difficult to use as we
2192 * may also need to change the insertion index. Code turned
2193 * off for 6.2, should be revisited later.
2195 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2196 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2197 INT_MOD(hdr_s->usedbytes, ARCH_CONVERT, -tmp);
2198 INT_MOD(hdr_s->count, ARCH_CONVERT, -1);
2199 entry_d--; /* to compensate for ++ in loop hdr */
2201 if ((start_s + i) < offset)
2202 result++; /* insertion index adjustment */
2205 INT_MOD(hdr_d->firstused, ARCH_CONVERT, -tmp);
2206 /* both on-disk, don't endian flip twice */
2207 entry_d->hashval = entry_s->hashval;
2208 /* both on-disk, don't endian flip twice */
2209 entry_d->nameidx = hdr_d->firstused;
2210 entry_d->flags = entry_s->flags;
2211 ASSERT(INT_GET(entry_d->nameidx, ARCH_CONVERT) + tmp
2213 memmove(XFS_ATTR_LEAF_NAME(leaf_d, desti),
2214 XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), tmp);
2215 ASSERT(INT_GET(entry_s->nameidx, ARCH_CONVERT) + tmp
2217 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2218 INT_MOD(hdr_s->usedbytes, ARCH_CONVERT, -tmp);
2219 INT_MOD(hdr_d->usedbytes, ARCH_CONVERT, tmp);
2220 INT_MOD(hdr_s->count, ARCH_CONVERT, -1);
2221 INT_MOD(hdr_d->count, ARCH_CONVERT, 1);
2222 tmp = INT_GET(hdr_d->count, ARCH_CONVERT)
2223 * sizeof(xfs_attr_leaf_entry_t)
2224 + sizeof(xfs_attr_leaf_hdr_t);
2225 ASSERT(INT_GET(hdr_d->firstused, ARCH_CONVERT) >= tmp);
2232 * Zero out the entries we just copied.
2234 if (start_s == INT_GET(hdr_s->count, ARCH_CONVERT)) {
2235 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2236 entry_s = &leaf_s->entries[start_s];
2237 ASSERT(((char *)entry_s + tmp) <=
2238 ((char *)leaf_s + XFS_LBSIZE(mp)));
2239 memset((char *)entry_s, 0, tmp);
2242 * Move the remaining entries down to fill the hole,
2243 * then zero the entries at the top.
2245 tmp = INT_GET(hdr_s->count, ARCH_CONVERT) - count;
2246 tmp *= sizeof(xfs_attr_leaf_entry_t);
2247 entry_s = &leaf_s->entries[start_s + count];
2248 entry_d = &leaf_s->entries[start_s];
2249 memmove((char *)entry_d, (char *)entry_s, tmp);
2251 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2252 entry_s = &leaf_s->entries[INT_GET(hdr_s->count,
2254 ASSERT(((char *)entry_s + tmp) <=
2255 ((char *)leaf_s + XFS_LBSIZE(mp)));
2256 memset((char *)entry_s, 0, tmp);
2260 * Fill in the freemap information
2262 INT_SET(hdr_d->freemap[0].base, ARCH_CONVERT,
2263 sizeof(xfs_attr_leaf_hdr_t));
2264 INT_MOD(hdr_d->freemap[0].base, ARCH_CONVERT,
2265 INT_GET(hdr_d->count, ARCH_CONVERT)
2266 * sizeof(xfs_attr_leaf_entry_t));
2267 INT_SET(hdr_d->freemap[0].size, ARCH_CONVERT,
2268 INT_GET(hdr_d->firstused, ARCH_CONVERT)
2269 - INT_GET(hdr_d->freemap[0].base, ARCH_CONVERT));
2270 hdr_d->freemap[1].base = 0;
2271 hdr_d->freemap[2].base = 0;
2272 hdr_d->freemap[1].size = 0;
2273 hdr_d->freemap[2].size = 0;
2274 hdr_s->holes = 1; /* leaf may not be compact */
2278 * Compare two leaf blocks "order".
2279 * Return 0 unless leaf2 should go before leaf1.
2282 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2284 xfs_attr_leafblock_t *leaf1, *leaf2;
2286 leaf1 = leaf1_bp->data;
2287 leaf2 = leaf2_bp->data;
2288 ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2289 (be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC));
2290 if ( (INT_GET(leaf1->hdr.count, ARCH_CONVERT) > 0)
2291 && (INT_GET(leaf2->hdr.count, ARCH_CONVERT) > 0)
2292 && ( (INT_GET(leaf2->entries[ 0 ].hashval, ARCH_CONVERT) <
2293 INT_GET(leaf1->entries[ 0 ].hashval, ARCH_CONVERT))
2294 || (INT_GET(leaf2->entries[INT_GET(leaf2->hdr.count,
2295 ARCH_CONVERT)-1].hashval, ARCH_CONVERT) <
2296 INT_GET(leaf1->entries[INT_GET(leaf1->hdr.count,
2297 ARCH_CONVERT)-1].hashval, ARCH_CONVERT))) ) {
2304 * Pick up the last hashvalue from a leaf block.
2307 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2309 xfs_attr_leafblock_t *leaf;
2312 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2314 *count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
2315 if (!leaf->hdr.count)
2317 return(INT_GET(leaf->entries[INT_GET(leaf->hdr.count,
2318 ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2322 * Calculate the number of bytes used to store the indicated attribute
2323 * (whether local or remote only calculate bytes in this block).
2326 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2328 xfs_attr_leaf_name_local_t *name_loc;
2329 xfs_attr_leaf_name_remote_t *name_rmt;
2332 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2333 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2334 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, index);
2335 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc->namelen,
2336 INT_GET(name_loc->valuelen,
2339 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, index);
2340 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt->namelen);
2346 * Calculate the number of bytes that would be required to store the new
2347 * attribute (whether local or remote only calculate bytes in this block).
2348 * This routine decides as a side effect whether the attribute will be
2349 * a "local" or a "remote" attribute.
2352 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2356 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen, valuelen);
2357 if (size < XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize)) {
2362 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen);
2371 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2374 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2376 attrlist_cursor_kern_t *cursor;
2377 xfs_attr_leafblock_t *leaf;
2378 xfs_attr_leaf_entry_t *entry;
2379 xfs_attr_leaf_name_local_t *name_loc;
2380 xfs_attr_leaf_name_remote_t *name_rmt;
2385 cursor = context->cursor;
2386 cursor->initted = 1;
2388 xfs_attr_trace_l_cl("blk start", context, leaf);
2391 * Re-find our place in the leaf block if this is a new syscall.
2393 if (context->resynch) {
2394 entry = &leaf->entries[0];
2395 for (i = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT);
2397 if (INT_GET(entry->hashval, ARCH_CONVERT)
2398 == cursor->hashval) {
2399 if (cursor->offset == context->dupcnt) {
2400 context->dupcnt = 0;
2404 } else if (INT_GET(entry->hashval, ARCH_CONVERT)
2405 > cursor->hashval) {
2406 context->dupcnt = 0;
2410 if (i == INT_GET(leaf->hdr.count, ARCH_CONVERT)) {
2411 xfs_attr_trace_l_c("not found", context);
2415 entry = &leaf->entries[0];
2418 context->resynch = 0;
2421 * We have found our place, start copying out the new attributes.
2424 for ( ; (i < INT_GET(leaf->hdr.count, ARCH_CONVERT))
2425 && (retval == 0); entry++, i++) {
2426 attrnames_t *namesp;
2428 if (INT_GET(entry->hashval, ARCH_CONVERT) != cursor->hashval) {
2429 cursor->hashval = INT_GET(entry->hashval, ARCH_CONVERT);
2433 if (entry->flags & XFS_ATTR_INCOMPLETE)
2434 continue; /* skip incomplete entries */
2435 if (((context->flags & ATTR_SECURE) != 0) !=
2436 ((entry->flags & XFS_ATTR_SECURE) != 0) &&
2437 !(context->flags & ATTR_KERNORMALS))
2438 continue; /* skip non-matching entries */
2439 if (((context->flags & ATTR_ROOT) != 0) !=
2440 ((entry->flags & XFS_ATTR_ROOT) != 0) &&
2441 !(context->flags & ATTR_KERNROOTLS))
2442 continue; /* skip non-matching entries */
2444 namesp = (entry->flags & XFS_ATTR_SECURE) ? &attr_secure :
2445 ((entry->flags & XFS_ATTR_ROOT) ? &attr_trusted :
2448 if (entry->flags & XFS_ATTR_LOCAL) {
2449 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
2450 if (context->flags & ATTR_KERNOVAL) {
2451 ASSERT(context->flags & ATTR_KERNAMELS);
2452 context->count += namesp->attr_namelen +
2453 (int)name_loc->namelen + 1;
2455 retval = xfs_attr_put_listent(context, namesp,
2456 (char *)name_loc->nameval,
2457 (int)name_loc->namelen,
2458 (int)INT_GET(name_loc->valuelen,
2462 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2463 if (context->flags & ATTR_KERNOVAL) {
2464 ASSERT(context->flags & ATTR_KERNAMELS);
2465 context->count += namesp->attr_namelen +
2466 (int)name_rmt->namelen + 1;
2468 retval = xfs_attr_put_listent(context, namesp,
2469 (char *)name_rmt->name,
2470 (int)name_rmt->namelen,
2471 (int)INT_GET(name_rmt->valuelen,
2479 xfs_attr_trace_l_cl("blk end", context, leaf);
2483 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
2484 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
2485 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
2486 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
2487 & ~(sizeof(u_int32_t)-1))
2490 * Format an attribute and copy it out to the user's buffer.
2491 * Take care to check values and protect against them changing later,
2492 * we may be reading them directly out of a user buffer.
2496 xfs_attr_put_listent(xfs_attr_list_context_t *context,
2497 attrnames_t *namesp, char *name, int namelen, int valuelen)
2499 attrlist_ent_t *aep;
2502 ASSERT(!(context->flags & ATTR_KERNOVAL));
2503 if (context->flags & ATTR_KERNAMELS) {
2506 ASSERT(context->count >= 0);
2508 arraytop = context->count + namesp->attr_namelen + namelen + 1;
2509 if (arraytop > context->firstu) {
2510 context->count = -1; /* insufficient space */
2513 offset = (char *)context->alist + context->count;
2514 strncpy(offset, namesp->attr_name, namesp->attr_namelen);
2515 offset += namesp->attr_namelen;
2516 strncpy(offset, name, namelen); /* real name */
2519 context->count += namesp->attr_namelen + namelen + 1;
2523 ASSERT(context->count >= 0);
2524 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
2525 ASSERT(context->firstu >= sizeof(*context->alist));
2526 ASSERT(context->firstu <= context->bufsize);
2528 arraytop = sizeof(*context->alist) +
2529 context->count * sizeof(context->alist->al_offset[0]);
2530 context->firstu -= ATTR_ENTSIZE(namelen);
2531 if (context->firstu < arraytop) {
2532 xfs_attr_trace_l_c("buffer full", context);
2533 context->alist->al_more = 1;
2537 aep = (attrlist_ent_t *)&(((char *)context->alist)[ context->firstu ]);
2538 aep->a_valuelen = valuelen;
2539 memcpy(aep->a_name, name, namelen);
2540 aep->a_name[ namelen ] = 0;
2541 context->alist->al_offset[ context->count++ ] = context->firstu;
2542 context->alist->al_count = context->count;
2543 xfs_attr_trace_l_c("add", context);
2547 /*========================================================================
2548 * Manage the INCOMPLETE flag in a leaf entry
2549 *========================================================================*/
2552 * Clear the INCOMPLETE flag on an entry in a leaf block.
2555 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2557 xfs_attr_leafblock_t *leaf;
2558 xfs_attr_leaf_entry_t *entry;
2559 xfs_attr_leaf_name_remote_t *name_rmt;
2563 xfs_attr_leaf_name_local_t *name_loc;
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 < INT_GET(leaf->hdr.count, ARCH_CONVERT));
2581 ASSERT(args->index >= 0);
2582 entry = &leaf->entries[ args->index ];
2583 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2586 if (entry->flags & XFS_ATTR_LOCAL) {
2587 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2588 namelen = name_loc->namelen;
2589 name = (char *)name_loc->nameval;
2591 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2592 namelen = name_rmt->namelen;
2593 name = (char *)name_rmt->name;
2595 ASSERT(INT_GET(entry->hashval, ARCH_CONVERT) == args->hashval);
2596 ASSERT(namelen == args->namelen);
2597 ASSERT(memcmp(name, args->name, namelen) == 0);
2600 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2601 xfs_da_log_buf(args->trans, bp,
2602 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2604 if (args->rmtblkno) {
2605 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2606 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2607 INT_SET(name_rmt->valueblk, ARCH_CONVERT, args->rmtblkno);
2608 INT_SET(name_rmt->valuelen, ARCH_CONVERT, args->valuelen);
2609 xfs_da_log_buf(args->trans, bp,
2610 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2612 xfs_da_buf_done(bp);
2615 * Commit the flag value change and start the next trans in series.
2617 error = xfs_attr_rolltrans(&args->trans, args->dp);
2623 * Set the INCOMPLETE flag on an entry in a leaf block.
2626 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2628 xfs_attr_leafblock_t *leaf;
2629 xfs_attr_leaf_entry_t *entry;
2630 xfs_attr_leaf_name_remote_t *name_rmt;
2635 * Set up the operation.
2637 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2645 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2646 ASSERT(args->index < INT_GET(leaf->hdr.count, ARCH_CONVERT));
2647 ASSERT(args->index >= 0);
2648 entry = &leaf->entries[ args->index ];
2650 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2651 entry->flags |= XFS_ATTR_INCOMPLETE;
2652 xfs_da_log_buf(args->trans, bp,
2653 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2654 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2655 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2656 name_rmt->valueblk = 0;
2657 name_rmt->valuelen = 0;
2658 xfs_da_log_buf(args->trans, bp,
2659 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2661 xfs_da_buf_done(bp);
2664 * Commit the flag value change and start the next trans in series.
2666 error = xfs_attr_rolltrans(&args->trans, args->dp);
2672 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2673 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2674 * entry given by args->blkno2/index2.
2676 * Note that they could be in different blocks, or in the same block.
2679 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2681 xfs_attr_leafblock_t *leaf1, *leaf2;
2682 xfs_attr_leaf_entry_t *entry1, *entry2;
2683 xfs_attr_leaf_name_remote_t *name_rmt;
2684 xfs_dabuf_t *bp1, *bp2;
2687 xfs_attr_leaf_name_local_t *name_loc;
2688 int namelen1, namelen2;
2689 char *name1, *name2;
2693 * Read the block containing the "old" attr
2695 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2700 ASSERT(bp1 != NULL);
2703 * Read the block containing the "new" attr, if it is different
2705 if (args->blkno2 != args->blkno) {
2706 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2707 -1, &bp2, XFS_ATTR_FORK);
2711 ASSERT(bp2 != NULL);
2717 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2718 ASSERT(args->index < INT_GET(leaf1->hdr.count, ARCH_CONVERT));
2719 ASSERT(args->index >= 0);
2720 entry1 = &leaf1->entries[ args->index ];
2723 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2724 ASSERT(args->index2 < INT_GET(leaf2->hdr.count, ARCH_CONVERT));
2725 ASSERT(args->index2 >= 0);
2726 entry2 = &leaf2->entries[ args->index2 ];
2729 if (entry1->flags & XFS_ATTR_LOCAL) {
2730 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf1, args->index);
2731 namelen1 = name_loc->namelen;
2732 name1 = (char *)name_loc->nameval;
2734 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2735 namelen1 = name_rmt->namelen;
2736 name1 = (char *)name_rmt->name;
2738 if (entry2->flags & XFS_ATTR_LOCAL) {
2739 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf2, args->index2);
2740 namelen2 = name_loc->namelen;
2741 name2 = (char *)name_loc->nameval;
2743 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2744 namelen2 = name_rmt->namelen;
2745 name2 = (char *)name_rmt->name;
2747 ASSERT(INT_GET(entry1->hashval, ARCH_CONVERT) == INT_GET(entry2->hashval, ARCH_CONVERT));
2748 ASSERT(namelen1 == namelen2);
2749 ASSERT(memcmp(name1, name2, namelen1) == 0);
2752 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2753 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2755 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2756 xfs_da_log_buf(args->trans, bp1,
2757 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2758 if (args->rmtblkno) {
2759 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2760 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2761 INT_SET(name_rmt->valueblk, ARCH_CONVERT, args->rmtblkno);
2762 INT_SET(name_rmt->valuelen, ARCH_CONVERT, args->valuelen);
2763 xfs_da_log_buf(args->trans, bp1,
2764 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2767 entry2->flags |= XFS_ATTR_INCOMPLETE;
2768 xfs_da_log_buf(args->trans, bp2,
2769 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2770 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2771 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2772 name_rmt->valueblk = 0;
2773 name_rmt->valuelen = 0;
2774 xfs_da_log_buf(args->trans, bp2,
2775 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2777 xfs_da_buf_done(bp1);
2779 xfs_da_buf_done(bp2);
2782 * Commit the flag value change and start the next trans in series.
2784 error = xfs_attr_rolltrans(&args->trans, args->dp);
2789 /*========================================================================
2790 * Indiscriminately delete the entire attribute fork
2791 *========================================================================*/
2794 * Recurse (gasp!) through the attribute nodes until we find leaves.
2795 * We're doing a depth-first traversal in order to invalidate everything.
2798 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2800 xfs_da_blkinfo_t *info;
2806 * Read block 0 to see what we have to work with.
2807 * We only get here if we have extents, since we remove
2808 * the extents in reverse order the extent containing
2809 * block 0 must still be there.
2811 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2814 blkno = xfs_da_blkno(bp);
2817 * Invalidate the tree, even if the "tree" is only a single leaf block.
2818 * This is a depth-first traversal!
2821 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2822 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2823 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2824 error = xfs_attr_leaf_inactive(trans, dp, bp);
2826 error = XFS_ERROR(EIO);
2827 xfs_da_brelse(*trans, bp);
2833 * Invalidate the incore copy of the root block.
2835 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2838 xfs_da_binval(*trans, bp); /* remove from cache */
2840 * Commit the invalidate and start the next transaction.
2842 error = xfs_attr_rolltrans(trans, dp);
2848 * Recurse (gasp!) through the attribute nodes until we find leaves.
2849 * We're doing a depth-first traversal in order to invalidate everything.
2852 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2855 xfs_da_blkinfo_t *info;
2856 xfs_da_intnode_t *node;
2857 xfs_dablk_t child_fsb;
2858 xfs_daddr_t parent_blkno, child_blkno;
2859 int error, count, i;
2860 xfs_dabuf_t *child_bp;
2863 * Since this code is recursive (gasp!) we must protect ourselves.
2865 if (level > XFS_DA_NODE_MAXDEPTH) {
2866 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2867 return(XFS_ERROR(EIO));
2871 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2872 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2873 count = INT_GET(node->hdr.count, ARCH_CONVERT);
2875 xfs_da_brelse(*trans, bp);
2878 child_fsb = INT_GET(node->btree[0].before, ARCH_CONVERT);
2879 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2882 * If this is the node level just above the leaves, simply loop
2883 * over the leaves removing all of them. If this is higher up
2884 * in the tree, recurse downward.
2886 for (i = 0; i < count; i++) {
2888 * Read the subsidiary block to see what we have to work with.
2889 * Don't do this in a transaction. This is a depth-first
2890 * traversal of the tree so we may deal with many blocks
2891 * before we come back to this one.
2893 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2898 /* save for re-read later */
2899 child_blkno = xfs_da_blkno(child_bp);
2902 * Invalidate the subtree, however we have to.
2904 info = child_bp->data;
2905 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2906 error = xfs_attr_node_inactive(trans, dp,
2908 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2909 error = xfs_attr_leaf_inactive(trans, dp,
2912 error = XFS_ERROR(EIO);
2913 xfs_da_brelse(*trans, child_bp);
2919 * Remove the subsidiary block from the cache
2922 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2923 &child_bp, XFS_ATTR_FORK);
2926 xfs_da_binval(*trans, child_bp);
2930 * If we're not done, re-read the parent to get the next
2931 * child block number.
2933 if ((i+1) < count) {
2934 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2935 &bp, XFS_ATTR_FORK);
2938 child_fsb = INT_GET(node->btree[i+1].before, ARCH_CONVERT);
2939 xfs_da_brelse(*trans, bp);
2942 * Atomically commit the whole invalidate stuff.
2944 if ((error = xfs_attr_rolltrans(trans, dp)))
2952 * Invalidate all of the "remote" value regions pointed to by a particular
2954 * Note that we must release the lock on the buffer so that we are not
2955 * caught holding something that the logging code wants to flush to disk.
2958 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2960 xfs_attr_leafblock_t *leaf;
2961 xfs_attr_leaf_entry_t *entry;
2962 xfs_attr_leaf_name_remote_t *name_rmt;
2963 xfs_attr_inactive_list_t *list, *lp;
2964 int error, count, size, tmp, i;
2967 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2970 * Count the number of "remote" value extents.
2973 entry = &leaf->entries[0];
2974 for (i = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); entry++, i++) {
2975 if ( INT_GET(entry->nameidx, ARCH_CONVERT)
2976 && ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2977 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2978 if (name_rmt->valueblk)
2984 * If there are no "remote" values, we're done.
2987 xfs_da_brelse(*trans, bp);
2992 * Allocate storage for a list of all the "remote" value extents.
2994 size = count * sizeof(xfs_attr_inactive_list_t);
2995 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2998 * Identify each of the "remote" value extents.
3001 entry = &leaf->entries[0];
3002 for (i = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); entry++, i++) {
3003 if ( INT_GET(entry->nameidx, ARCH_CONVERT)
3004 && ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3005 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
3006 if (name_rmt->valueblk) {
3007 /* both on-disk, don't endian flip twice */
3008 lp->valueblk = name_rmt->valueblk;
3009 INT_SET(lp->valuelen, ARCH_CONVERT,
3010 XFS_B_TO_FSB(dp->i_mount,
3011 INT_GET(name_rmt->valuelen,
3017 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
3020 * Invalidate each of the "remote" value extents.
3023 for (lp = list, i = 0; i < count; i++, lp++) {
3024 tmp = xfs_attr_leaf_freextent(trans, dp,
3025 INT_GET(lp->valueblk,
3027 INT_GET(lp->valuelen,
3030 error = tmp; /* save only the 1st errno */
3033 kmem_free((xfs_caddr_t)list, size);
3038 * Look at all the extents for this logical region,
3039 * invalidate any buffers that are incore/in transactions.
3042 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
3043 xfs_dablk_t blkno, int blkcnt)
3045 xfs_bmbt_irec_t map;
3047 int tblkcnt, dblkcnt, nmap, error;
3052 * Roll through the "value", invalidating the attribute value's
3057 while (tblkcnt > 0) {
3059 * Try to remember where we decided to put the value.
3062 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
3063 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
3064 NULL, 0, &map, &nmap, NULL);
3069 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3072 * If it's a hole, these are already unmapped
3073 * so there's nothing to invalidate.
3075 if (map.br_startblock != HOLESTARTBLOCK) {
3077 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3079 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3081 bp = xfs_trans_get_buf(*trans,
3082 dp->i_mount->m_ddev_targp,
3083 dblkno, dblkcnt, XFS_BUF_LOCK);
3084 xfs_trans_binval(*trans, bp);
3086 * Roll to next transaction.
3088 if ((error = xfs_attr_rolltrans(trans, dp)))
3092 tblkno += map.br_blockcount;
3093 tblkcnt -= map.br_blockcount;
3101 * Roll from one trans in the sequence of PERMANENT transactions to the next.
3104 xfs_attr_rolltrans(xfs_trans_t **transp, xfs_inode_t *dp)
3107 unsigned int logres, count;
3111 * Ensure that the inode is always logged.
3114 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
3117 * Copy the critical parameters from one trans to the next.
3119 logres = trans->t_log_res;
3120 count = trans->t_log_count;
3121 *transp = xfs_trans_dup(trans);
3124 * Commit the current transaction.
3125 * If this commit failed, then it'd just unlock those items that
3126 * are not marked ihold. That also means that a filesystem shutdown
3127 * is in progress. The caller takes the responsibility to cancel
3128 * the duplicate transaction that gets returned.
3130 if ((error = xfs_trans_commit(trans, 0, NULL)))
3136 * Reserve space in the log for th next transaction.
3137 * This also pushes items in the "AIL", the list of logged items,
3138 * out to disk if they are taking up space at the tail of the log
3139 * that we want to use. This requires that either nothing be locked
3140 * across this call, or that anything that is locked be logged in
3141 * the prior and the next transactions.
3143 error = xfs_trans_reserve(trans, 0, logres, 0,
3144 XFS_TRANS_PERM_LOG_RES, count);
3146 * Ensure that the inode is in the new transaction and locked.
3149 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
3150 xfs_trans_ihold(trans, dp);