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 < be16_to_cpu(leaf->hdr.count); 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 < be16_to_cpu(leaf->hdr.count); 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 = be32_to_cpu(entry->hashval);
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[be16_to_cpu(leaf->hdr.count)-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 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
922 if (!hdr->firstused) {
923 hdr->firstused = cpu_to_be16(
924 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
927 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
928 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
929 sizeof(xfs_attr_leaf_hdr_t));
931 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
938 * Split the leaf node, rebalance, then add the new entry.
941 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
942 xfs_da_state_blk_t *newblk)
948 * Allocate space for a new leaf node.
950 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
951 error = xfs_da_grow_inode(state->args, &blkno);
954 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
957 newblk->blkno = blkno;
958 newblk->magic = XFS_ATTR_LEAF_MAGIC;
961 * Rebalance the entries across the two leaves.
962 * NOTE: rebalance() currently depends on the 2nd block being empty.
964 xfs_attr_leaf_rebalance(state, oldblk, newblk);
965 error = xfs_da_blk_link(state, oldblk, newblk);
970 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
971 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
972 * "new" attrs info. Will need the "old" info to remove it later.
974 * Insert the "new" entry in the correct block.
977 error = xfs_attr_leaf_add(oldblk->bp, state->args);
979 error = xfs_attr_leaf_add(newblk->bp, state->args);
982 * Update last hashval in each block since we added the name.
984 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
985 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
990 * Add a name to the leaf attribute list structure.
993 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
995 xfs_attr_leafblock_t *leaf;
996 xfs_attr_leaf_hdr_t *hdr;
997 xfs_attr_leaf_map_t *map;
998 int tablesize, entsize, sum, tmp, i;
1001 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1002 ASSERT((args->index >= 0)
1003 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1005 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1006 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1009 * Search through freemap for first-fit on new name length.
1010 * (may need to figure in size of entry struct too)
1012 tablesize = (be16_to_cpu(hdr->count) + 1)
1013 * sizeof(xfs_attr_leaf_entry_t)
1014 + sizeof(xfs_attr_leaf_hdr_t);
1015 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1016 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1017 if (tablesize > be16_to_cpu(hdr->firstused)) {
1018 sum += be16_to_cpu(map->size);
1022 continue; /* no space in this map */
1024 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1025 tmp += sizeof(xfs_attr_leaf_entry_t);
1026 if (be16_to_cpu(map->size) >= tmp) {
1027 tmp = xfs_attr_leaf_add_work(bp, args, i);
1030 sum += be16_to_cpu(map->size);
1034 * If there are no holes in the address space of the block,
1035 * and we don't have enough freespace, then compaction will do us
1036 * no good and we should just give up.
1038 if (!hdr->holes && (sum < entsize))
1039 return(XFS_ERROR(ENOSPC));
1042 * Compact the entries to coalesce free space.
1043 * This may change the hdr->count via dropping INCOMPLETE entries.
1045 xfs_attr_leaf_compact(args->trans, bp);
1048 * After compaction, the block is guaranteed to have only one
1049 * free region, in freemap[0]. If it is not big enough, give up.
1051 if (be16_to_cpu(hdr->freemap[0].size)
1052 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1053 return(XFS_ERROR(ENOSPC));
1055 return(xfs_attr_leaf_add_work(bp, args, 0));
1059 * Add a name to a leaf attribute list structure.
1062 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1064 xfs_attr_leafblock_t *leaf;
1065 xfs_attr_leaf_hdr_t *hdr;
1066 xfs_attr_leaf_entry_t *entry;
1067 xfs_attr_leaf_name_local_t *name_loc;
1068 xfs_attr_leaf_name_remote_t *name_rmt;
1069 xfs_attr_leaf_map_t *map;
1074 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1076 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1077 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1080 * Force open some space in the entry array and fill it in.
1082 entry = &leaf->entries[args->index];
1083 if (args->index < be16_to_cpu(hdr->count)) {
1084 tmp = be16_to_cpu(hdr->count) - args->index;
1085 tmp *= sizeof(xfs_attr_leaf_entry_t);
1086 memmove((char *)(entry+1), (char *)entry, tmp);
1087 xfs_da_log_buf(args->trans, bp,
1088 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1090 be16_add(&hdr->count, 1);
1093 * Allocate space for the new string (at the end of the run).
1095 map = &hdr->freemap[mapindex];
1096 mp = args->trans->t_mountp;
1097 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1098 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1099 ASSERT(be16_to_cpu(map->size) >=
1100 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1101 mp->m_sb.sb_blocksize, NULL));
1102 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1103 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1104 be16_add(&map->size,
1105 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1106 mp->m_sb.sb_blocksize, &tmp));
1107 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1108 be16_to_cpu(map->size));
1109 entry->hashval = cpu_to_be32(args->hashval);
1110 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1111 entry->flags |= (args->flags & ATTR_SECURE) ? XFS_ATTR_SECURE :
1112 ((args->flags & ATTR_ROOT) ? XFS_ATTR_ROOT : 0);
1114 entry->flags |= XFS_ATTR_INCOMPLETE;
1115 if ((args->blkno2 == args->blkno) &&
1116 (args->index2 <= args->index)) {
1120 xfs_da_log_buf(args->trans, bp,
1121 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1122 ASSERT((args->index == 0) ||
1123 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1124 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1125 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1128 * Copy the attribute name and value into the new space.
1130 * For "remote" attribute values, simply note that we need to
1131 * allocate space for the "remote" value. We can't actually
1132 * allocate the extents in this transaction, and we can't decide
1133 * which blocks they should be as we might allocate more blocks
1134 * as part of this transaction (a split operation for example).
1136 if (entry->flags & XFS_ATTR_LOCAL) {
1137 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
1138 name_loc->namelen = args->namelen;
1139 INT_SET(name_loc->valuelen, ARCH_CONVERT, args->valuelen);
1140 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1141 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1142 INT_GET(name_loc->valuelen, ARCH_CONVERT));
1144 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
1145 name_rmt->namelen = args->namelen;
1146 memcpy((char *)name_rmt->name, args->name, args->namelen);
1147 entry->flags |= XFS_ATTR_INCOMPLETE;
1149 name_rmt->valuelen = 0;
1150 name_rmt->valueblk = 0;
1152 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1154 xfs_da_log_buf(args->trans, bp,
1155 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1156 xfs_attr_leaf_entsize(leaf, args->index)));
1159 * Update the control info for this leaf node
1161 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1162 /* both on-disk, don't endian-flip twice */
1163 hdr->firstused = entry->nameidx;
1165 ASSERT(be16_to_cpu(hdr->firstused) >=
1166 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1167 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1168 + sizeof(xfs_attr_leaf_hdr_t);
1169 map = &hdr->freemap[0];
1170 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1171 if (be16_to_cpu(map->base) == tmp) {
1172 be16_add(&map->base, sizeof(xfs_attr_leaf_entry_t));
1173 be16_add(&map->size,
1174 -((int)sizeof(xfs_attr_leaf_entry_t)));
1177 be16_add(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1178 xfs_da_log_buf(args->trans, bp,
1179 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1184 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1187 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1189 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1190 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1194 mp = trans->t_mountp;
1195 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1196 ASSERT(tmpbuffer != NULL);
1197 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1198 memset(bp->data, 0, XFS_LBSIZE(mp));
1201 * Copy basic information
1203 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1205 hdr_s = &leaf_s->hdr;
1206 hdr_d = &leaf_d->hdr;
1207 hdr_d->info = hdr_s->info; /* struct copy */
1208 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1209 /* handle truncation gracefully */
1210 if (!hdr_d->firstused) {
1211 hdr_d->firstused = cpu_to_be16(
1212 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1214 hdr_d->usedbytes = 0;
1217 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1218 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1219 sizeof(xfs_attr_leaf_hdr_t));
1222 * Copy all entry's in the same (sorted) order,
1223 * but allocate name/value pairs packed and in sequence.
1225 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1226 be16_to_cpu(hdr_s->count), mp);
1227 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1229 kmem_free(tmpbuffer, XFS_LBSIZE(mp));
1233 * Redistribute the attribute list entries between two leaf nodes,
1234 * taking into account the size of the new entry.
1236 * NOTE: if new block is empty, then it will get the upper half of the
1237 * old block. At present, all (one) callers pass in an empty second block.
1239 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1240 * to match what it is doing in splitting the attribute leaf block. Those
1241 * values are used in "atomic rename" operations on attributes. Note that
1242 * the "new" and "old" values can end up in different blocks.
1245 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1246 xfs_da_state_blk_t *blk2)
1248 xfs_da_args_t *args;
1249 xfs_da_state_blk_t *tmp_blk;
1250 xfs_attr_leafblock_t *leaf1, *leaf2;
1251 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1252 int count, totallen, max, space, swap;
1255 * Set up environment.
1257 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1258 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1259 leaf1 = blk1->bp->data;
1260 leaf2 = blk2->bp->data;
1261 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1262 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1266 * Check ordering of blocks, reverse if it makes things simpler.
1268 * NOTE: Given that all (current) callers pass in an empty
1269 * second block, this code should never set "swap".
1272 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1276 leaf1 = blk1->bp->data;
1277 leaf2 = blk2->bp->data;
1284 * Examine entries until we reduce the absolute difference in
1285 * byte usage between the two blocks to a minimum. Then get
1286 * the direction to copy and the number of elements to move.
1288 * "inleaf" is true if the new entry should be inserted into blk1.
1289 * If "swap" is also true, then reverse the sense of "inleaf".
1291 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1294 state->inleaf = !state->inleaf;
1297 * Move any entries required from leaf to leaf:
1299 if (count < be16_to_cpu(hdr1->count)) {
1301 * Figure the total bytes to be added to the destination leaf.
1303 /* number entries being moved */
1304 count = be16_to_cpu(hdr1->count) - count;
1305 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1306 space += count * sizeof(xfs_attr_leaf_entry_t);
1309 * leaf2 is the destination, compact it if it looks tight.
1311 max = be16_to_cpu(hdr2->firstused)
1312 - sizeof(xfs_attr_leaf_hdr_t);
1313 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1315 xfs_attr_leaf_compact(args->trans, blk2->bp);
1319 * Move high entries from leaf1 to low end of leaf2.
1321 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1322 leaf2, 0, count, state->mp);
1324 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1325 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1326 } else if (count > be16_to_cpu(hdr1->count)) {
1328 * I assert that since all callers pass in an empty
1329 * second buffer, this code should never execute.
1333 * Figure the total bytes to be added to the destination leaf.
1335 /* number entries being moved */
1336 count -= be16_to_cpu(hdr1->count);
1337 space = totallen - be16_to_cpu(hdr1->usedbytes);
1338 space += count * sizeof(xfs_attr_leaf_entry_t);
1341 * leaf1 is the destination, compact it if it looks tight.
1343 max = be16_to_cpu(hdr1->firstused)
1344 - sizeof(xfs_attr_leaf_hdr_t);
1345 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1347 xfs_attr_leaf_compact(args->trans, blk1->bp);
1351 * Move low entries from leaf2 to high end of leaf1.
1353 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1354 be16_to_cpu(hdr1->count), count, state->mp);
1356 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1357 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1361 * Copy out last hashval in each block for B-tree code.
1363 blk1->hashval = be32_to_cpu(
1364 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1365 blk2->hashval = be32_to_cpu(
1366 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1369 * Adjust the expected index for insertion.
1370 * NOTE: this code depends on the (current) situation that the
1371 * second block was originally empty.
1373 * If the insertion point moved to the 2nd block, we must adjust
1374 * the index. We must also track the entry just following the
1375 * new entry for use in an "atomic rename" operation, that entry
1376 * is always the "old" entry and the "new" entry is what we are
1377 * inserting. The index/blkno fields refer to the "old" entry,
1378 * while the index2/blkno2 fields refer to the "new" entry.
1380 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1381 ASSERT(state->inleaf == 0);
1382 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1383 args->index = args->index2 = blk2->index;
1384 args->blkno = args->blkno2 = blk2->blkno;
1385 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1386 if (state->inleaf) {
1387 args->index = blk1->index;
1388 args->blkno = blk1->blkno;
1390 args->blkno2 = blk2->blkno;
1392 blk2->index = blk1->index
1393 - be16_to_cpu(leaf1->hdr.count);
1394 args->index = args->index2 = blk2->index;
1395 args->blkno = args->blkno2 = blk2->blkno;
1398 ASSERT(state->inleaf == 1);
1399 args->index = args->index2 = blk1->index;
1400 args->blkno = args->blkno2 = blk1->blkno;
1405 * Examine entries until we reduce the absolute difference in
1406 * byte usage between the two blocks to a minimum.
1407 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1408 * GROT: there will always be enough room in either block for a new entry.
1409 * GROT: Do a double-split for this case?
1412 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1413 xfs_da_state_blk_t *blk1,
1414 xfs_da_state_blk_t *blk2,
1415 int *countarg, int *usedbytesarg)
1417 xfs_attr_leafblock_t *leaf1, *leaf2;
1418 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1419 xfs_attr_leaf_entry_t *entry;
1420 int count, max, index, totallen, half;
1421 int lastdelta, foundit, tmp;
1424 * Set up environment.
1426 leaf1 = blk1->bp->data;
1427 leaf2 = blk2->bp->data;
1434 * Examine entries until we reduce the absolute difference in
1435 * byte usage between the two blocks to a minimum.
1437 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1438 half = (max+1) * sizeof(*entry);
1439 half += be16_to_cpu(hdr1->usedbytes) +
1440 be16_to_cpu(hdr2->usedbytes) +
1441 xfs_attr_leaf_newentsize(
1442 state->args->namelen,
1443 state->args->valuelen,
1444 state->blocksize, NULL);
1446 lastdelta = state->blocksize;
1447 entry = &leaf1->entries[0];
1448 for (count = index = 0; count < max; entry++, index++, count++) {
1450 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1452 * The new entry is in the first block, account for it.
1454 if (count == blk1->index) {
1455 tmp = totallen + sizeof(*entry) +
1456 xfs_attr_leaf_newentsize(
1457 state->args->namelen,
1458 state->args->valuelen,
1459 state->blocksize, NULL);
1460 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1462 lastdelta = XFS_ATTR_ABS(half - tmp);
1468 * Wrap around into the second block if necessary.
1470 if (count == be16_to_cpu(hdr1->count)) {
1472 entry = &leaf1->entries[0];
1477 * Figure out if next leaf entry would be too much.
1479 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1481 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1483 lastdelta = XFS_ATTR_ABS(half - tmp);
1489 * Calculate the number of usedbytes that will end up in lower block.
1490 * If new entry not in lower block, fix up the count.
1492 totallen -= count * sizeof(*entry);
1494 totallen -= sizeof(*entry) +
1495 xfs_attr_leaf_newentsize(
1496 state->args->namelen,
1497 state->args->valuelen,
1498 state->blocksize, NULL);
1502 *usedbytesarg = totallen;
1506 /*========================================================================
1507 * Routines used for shrinking the Btree.
1508 *========================================================================*/
1511 * Check a leaf block and its neighbors to see if the block should be
1512 * collapsed into one or the other neighbor. Always keep the block
1513 * with the smaller block number.
1514 * If the current block is over 50% full, don't try to join it, return 0.
1515 * If the block is empty, fill in the state structure and return 2.
1516 * If it can be collapsed, fill in the state structure and return 1.
1517 * If nothing can be done, return 0.
1519 * GROT: allow for INCOMPLETE entries in calculation.
1522 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1524 xfs_attr_leafblock_t *leaf;
1525 xfs_da_state_blk_t *blk;
1526 xfs_da_blkinfo_t *info;
1527 int count, bytes, forward, error, retval, i;
1532 * Check for the degenerate case of the block being over 50% full.
1533 * If so, it's not worth even looking to see if we might be able
1534 * to coalesce with a sibling.
1536 blk = &state->path.blk[ state->path.active-1 ];
1537 info = blk->bp->data;
1538 ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1539 leaf = (xfs_attr_leafblock_t *)info;
1540 count = be16_to_cpu(leaf->hdr.count);
1541 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1542 count * sizeof(xfs_attr_leaf_entry_t) +
1543 be16_to_cpu(leaf->hdr.usedbytes);
1544 if (bytes > (state->blocksize >> 1)) {
1545 *action = 0; /* blk over 50%, don't try to join */
1550 * Check for the degenerate case of the block being empty.
1551 * If the block is empty, we'll simply delete it, no need to
1552 * coalesce it with a sibling block. We choose (aribtrarily)
1553 * to merge with the forward block unless it is NULL.
1557 * Make altpath point to the block we want to keep and
1558 * path point to the block we want to drop (this one).
1560 forward = (info->forw != 0);
1561 memcpy(&state->altpath, &state->path, sizeof(state->path));
1562 error = xfs_da_path_shift(state, &state->altpath, forward,
1575 * Examine each sibling block to see if we can coalesce with
1576 * at least 25% free space to spare. We need to figure out
1577 * whether to merge with the forward or the backward block.
1578 * We prefer coalescing with the lower numbered sibling so as
1579 * to shrink an attribute list over time.
1581 /* start with smaller blk num */
1582 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1583 for (i = 0; i < 2; forward = !forward, i++) {
1585 blkno = be32_to_cpu(info->forw);
1587 blkno = be32_to_cpu(info->back);
1590 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1591 blkno, -1, &bp, XFS_ATTR_FORK);
1596 leaf = (xfs_attr_leafblock_t *)info;
1597 count = be16_to_cpu(leaf->hdr.count);
1598 bytes = state->blocksize - (state->blocksize>>2);
1599 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1601 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1602 count += be16_to_cpu(leaf->hdr.count);
1603 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1604 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1605 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1606 xfs_da_brelse(state->args->trans, bp);
1608 break; /* fits with at least 25% to spare */
1616 * Make altpath point to the block we want to keep (the lower
1617 * numbered block) and path point to the block we want to drop.
1619 memcpy(&state->altpath, &state->path, sizeof(state->path));
1620 if (blkno < blk->blkno) {
1621 error = xfs_da_path_shift(state, &state->altpath, forward,
1624 error = xfs_da_path_shift(state, &state->path, forward,
1638 * Remove a name from the leaf attribute list structure.
1640 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1641 * If two leaves are 37% full, when combined they will leave 25% free.
1644 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1646 xfs_attr_leafblock_t *leaf;
1647 xfs_attr_leaf_hdr_t *hdr;
1648 xfs_attr_leaf_map_t *map;
1649 xfs_attr_leaf_entry_t *entry;
1650 int before, after, smallest, entsize;
1651 int tablesize, tmp, i;
1655 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1657 mp = args->trans->t_mountp;
1658 ASSERT((be16_to_cpu(hdr->count) > 0)
1659 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1660 ASSERT((args->index >= 0)
1661 && (args->index < be16_to_cpu(hdr->count)));
1662 ASSERT(be16_to_cpu(hdr->firstused) >=
1663 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1664 entry = &leaf->entries[args->index];
1665 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1666 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1669 * Scan through free region table:
1670 * check for adjacency of free'd entry with an existing one,
1671 * find smallest free region in case we need to replace it,
1672 * adjust any map that borders the entry table,
1674 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1675 + sizeof(xfs_attr_leaf_hdr_t);
1676 map = &hdr->freemap[0];
1677 tmp = be16_to_cpu(map->size);
1678 before = after = -1;
1679 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1680 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1681 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1682 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1683 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1684 if (be16_to_cpu(map->base) == tablesize) {
1685 be16_add(&map->base,
1686 -((int)sizeof(xfs_attr_leaf_entry_t)));
1687 be16_add(&map->size, sizeof(xfs_attr_leaf_entry_t));
1690 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1691 == be16_to_cpu(entry->nameidx)) {
1693 } else if (be16_to_cpu(map->base)
1694 == (be16_to_cpu(entry->nameidx) + entsize)) {
1696 } else if (be16_to_cpu(map->size) < tmp) {
1697 tmp = be16_to_cpu(map->size);
1703 * Coalesce adjacent freemap regions,
1704 * or replace the smallest region.
1706 if ((before >= 0) || (after >= 0)) {
1707 if ((before >= 0) && (after >= 0)) {
1708 map = &hdr->freemap[before];
1709 be16_add(&map->size, entsize);
1710 be16_add(&map->size,
1711 be16_to_cpu(hdr->freemap[after].size));
1712 hdr->freemap[after].base = 0;
1713 hdr->freemap[after].size = 0;
1714 } else if (before >= 0) {
1715 map = &hdr->freemap[before];
1716 be16_add(&map->size, entsize);
1718 map = &hdr->freemap[after];
1719 /* both on-disk, don't endian flip twice */
1720 map->base = entry->nameidx;
1721 be16_add(&map->size, entsize);
1725 * Replace smallest region (if it is smaller than free'd entry)
1727 map = &hdr->freemap[smallest];
1728 if (be16_to_cpu(map->size) < entsize) {
1729 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1730 map->size = cpu_to_be16(entsize);
1735 * Did we remove the first entry?
1737 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1743 * Compress the remaining entries and zero out the removed stuff.
1745 memset(XFS_ATTR_LEAF_NAME(leaf, args->index), 0, entsize);
1746 be16_add(&hdr->usedbytes, -entsize);
1747 xfs_da_log_buf(args->trans, bp,
1748 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1751 tmp = (be16_to_cpu(hdr->count) - args->index)
1752 * sizeof(xfs_attr_leaf_entry_t);
1753 memmove((char *)entry, (char *)(entry+1), tmp);
1754 be16_add(&hdr->count, -1);
1755 xfs_da_log_buf(args->trans, bp,
1756 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1757 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1758 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1761 * If we removed the first entry, re-find the first used byte
1762 * in the name area. Note that if the entry was the "firstused",
1763 * then we don't have a "hole" in our block resulting from
1764 * removing the name.
1767 tmp = XFS_LBSIZE(mp);
1768 entry = &leaf->entries[0];
1769 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1770 ASSERT(be16_to_cpu(entry->nameidx) >=
1771 be16_to_cpu(hdr->firstused));
1772 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1774 if (be16_to_cpu(entry->nameidx) < tmp)
1775 tmp = be16_to_cpu(entry->nameidx);
1777 hdr->firstused = cpu_to_be16(tmp);
1778 if (!hdr->firstused) {
1779 hdr->firstused = cpu_to_be16(
1780 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1783 hdr->holes = 1; /* mark as needing compaction */
1785 xfs_da_log_buf(args->trans, bp,
1786 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1789 * Check if leaf is less than 50% full, caller may want to
1790 * "join" the leaf with a sibling if so.
1792 tmp = sizeof(xfs_attr_leaf_hdr_t);
1793 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1794 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1795 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1799 * Move all the attribute list entries from drop_leaf into save_leaf.
1802 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1803 xfs_da_state_blk_t *save_blk)
1805 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1806 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1811 * Set up environment.
1814 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1815 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1816 drop_leaf = drop_blk->bp->data;
1817 save_leaf = save_blk->bp->data;
1818 ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1819 ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1820 drop_hdr = &drop_leaf->hdr;
1821 save_hdr = &save_leaf->hdr;
1824 * Save last hashval from dying block for later Btree fixup.
1826 drop_blk->hashval = be32_to_cpu(
1827 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1830 * Check if we need a temp buffer, or can we do it in place.
1831 * Note that we don't check "leaf" for holes because we will
1832 * always be dropping it, toosmall() decided that for us already.
1834 if (save_hdr->holes == 0) {
1836 * dest leaf has no holes, so we add there. May need
1837 * to make some room in the entry array.
1839 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1840 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1841 be16_to_cpu(drop_hdr->count), mp);
1843 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1844 be16_to_cpu(save_hdr->count),
1845 be16_to_cpu(drop_hdr->count), mp);
1849 * Destination has holes, so we make a temporary copy
1850 * of the leaf and add them both to that.
1852 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1853 ASSERT(tmpbuffer != NULL);
1854 memset(tmpbuffer, 0, state->blocksize);
1855 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1856 tmp_hdr = &tmp_leaf->hdr;
1857 tmp_hdr->info = save_hdr->info; /* struct copy */
1859 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1860 if (!tmp_hdr->firstused) {
1861 tmp_hdr->firstused = cpu_to_be16(
1862 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1864 tmp_hdr->usedbytes = 0;
1865 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1866 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1867 be16_to_cpu(drop_hdr->count), mp);
1868 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1869 be16_to_cpu(tmp_leaf->hdr.count),
1870 be16_to_cpu(save_hdr->count), mp);
1872 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1873 be16_to_cpu(save_hdr->count), mp);
1874 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1875 be16_to_cpu(tmp_leaf->hdr.count),
1876 be16_to_cpu(drop_hdr->count), mp);
1878 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1879 kmem_free(tmpbuffer, state->blocksize);
1882 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1883 state->blocksize - 1);
1886 * Copy out last hashval in each block for B-tree code.
1888 save_blk->hashval = be32_to_cpu(
1889 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1892 /*========================================================================
1893 * Routines used for finding things in the Btree.
1894 *========================================================================*/
1897 * Look up a name in a leaf attribute list structure.
1898 * This is the internal routine, it uses the caller's buffer.
1900 * Note that duplicate keys are allowed, but only check within the
1901 * current leaf node. The Btree code must check in adjacent leaf nodes.
1903 * Return in args->index the index into the entry[] array of either
1904 * the found entry, or where the entry should have been (insert before
1907 * Don't change the args->value unless we find the attribute.
1910 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1912 xfs_attr_leafblock_t *leaf;
1913 xfs_attr_leaf_entry_t *entry;
1914 xfs_attr_leaf_name_local_t *name_loc;
1915 xfs_attr_leaf_name_remote_t *name_rmt;
1917 xfs_dahash_t hashval;
1920 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1921 ASSERT(be16_to_cpu(leaf->hdr.count)
1922 < (XFS_LBSIZE(args->dp->i_mount)/8));
1925 * Binary search. (note: small blocks will skip this loop)
1927 hashval = args->hashval;
1928 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1929 for (entry = &leaf->entries[probe]; span > 4;
1930 entry = &leaf->entries[probe]) {
1932 if (be32_to_cpu(entry->hashval) < hashval)
1934 else if (be32_to_cpu(entry->hashval) > hashval)
1939 ASSERT((probe >= 0) &&
1941 || (probe < be16_to_cpu(leaf->hdr.count))));
1942 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1945 * Since we may have duplicate hashval's, find the first matching
1946 * hashval in the leaf.
1948 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1952 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1953 (be32_to_cpu(entry->hashval) < hashval)) {
1957 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1958 (be32_to_cpu(entry->hashval) != hashval)) {
1959 args->index = probe;
1960 return(XFS_ERROR(ENOATTR));
1964 * Duplicate keys may be present, so search all of them for a match.
1966 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1967 (be32_to_cpu(entry->hashval) == hashval);
1970 * GROT: Add code to remove incomplete entries.
1973 * If we are looking for INCOMPLETE entries, show only those.
1974 * If we are looking for complete entries, show only those.
1976 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
1977 (entry->flags & XFS_ATTR_INCOMPLETE)) {
1980 if (entry->flags & XFS_ATTR_LOCAL) {
1981 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, probe);
1982 if (name_loc->namelen != args->namelen)
1984 if (memcmp(args->name, (char *)name_loc->nameval,
1985 args->namelen) != 0)
1987 if (((args->flags & ATTR_SECURE) != 0) !=
1988 ((entry->flags & XFS_ATTR_SECURE) != 0))
1990 if (((args->flags & ATTR_ROOT) != 0) !=
1991 ((entry->flags & XFS_ATTR_ROOT) != 0))
1993 args->index = probe;
1994 return(XFS_ERROR(EEXIST));
1996 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, probe);
1997 if (name_rmt->namelen != args->namelen)
1999 if (memcmp(args->name, (char *)name_rmt->name,
2000 args->namelen) != 0)
2002 if (((args->flags & ATTR_SECURE) != 0) !=
2003 ((entry->flags & XFS_ATTR_SECURE) != 0))
2005 if (((args->flags & ATTR_ROOT) != 0) !=
2006 ((entry->flags & XFS_ATTR_ROOT) != 0))
2008 args->index = probe;
2010 = INT_GET(name_rmt->valueblk, ARCH_CONVERT);
2011 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2012 INT_GET(name_rmt->valuelen,
2014 return(XFS_ERROR(EEXIST));
2017 args->index = probe;
2018 return(XFS_ERROR(ENOATTR));
2022 * Get the value associated with an attribute name from a leaf attribute
2026 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2029 xfs_attr_leafblock_t *leaf;
2030 xfs_attr_leaf_entry_t *entry;
2031 xfs_attr_leaf_name_local_t *name_loc;
2032 xfs_attr_leaf_name_remote_t *name_rmt;
2035 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2036 ASSERT(be16_to_cpu(leaf->hdr.count)
2037 < (XFS_LBSIZE(args->dp->i_mount)/8));
2038 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2040 entry = &leaf->entries[args->index];
2041 if (entry->flags & XFS_ATTR_LOCAL) {
2042 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2043 ASSERT(name_loc->namelen == args->namelen);
2044 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2045 valuelen = INT_GET(name_loc->valuelen, ARCH_CONVERT);
2046 if (args->flags & ATTR_KERNOVAL) {
2047 args->valuelen = valuelen;
2050 if (args->valuelen < valuelen) {
2051 args->valuelen = valuelen;
2052 return(XFS_ERROR(ERANGE));
2054 args->valuelen = valuelen;
2055 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2057 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2058 ASSERT(name_rmt->namelen == args->namelen);
2059 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2060 valuelen = INT_GET(name_rmt->valuelen, ARCH_CONVERT);
2061 args->rmtblkno = INT_GET(name_rmt->valueblk, ARCH_CONVERT);
2062 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2063 if (args->flags & ATTR_KERNOVAL) {
2064 args->valuelen = valuelen;
2067 if (args->valuelen < valuelen) {
2068 args->valuelen = valuelen;
2069 return(XFS_ERROR(ERANGE));
2071 args->valuelen = valuelen;
2076 /*========================================================================
2078 *========================================================================*/
2081 * Move the indicated entries from one leaf to another.
2082 * NOTE: this routine modifies both source and destination leaves.
2086 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2087 xfs_attr_leafblock_t *leaf_d, int start_d,
2088 int count, xfs_mount_t *mp)
2090 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2091 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2095 * Check for nothing to do.
2101 * Set up environment.
2103 ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2104 ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2105 hdr_s = &leaf_s->hdr;
2106 hdr_d = &leaf_d->hdr;
2107 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2108 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2109 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2110 ((be16_to_cpu(hdr_s->count)
2111 * sizeof(*entry_s))+sizeof(*hdr_s)));
2112 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2113 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2114 ((be16_to_cpu(hdr_d->count)
2115 * sizeof(*entry_d))+sizeof(*hdr_d)));
2117 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2118 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2119 ASSERT(count <= be16_to_cpu(hdr_s->count));
2122 * Move the entries in the destination leaf up to make a hole?
2124 if (start_d < be16_to_cpu(hdr_d->count)) {
2125 tmp = be16_to_cpu(hdr_d->count) - start_d;
2126 tmp *= sizeof(xfs_attr_leaf_entry_t);
2127 entry_s = &leaf_d->entries[start_d];
2128 entry_d = &leaf_d->entries[start_d + count];
2129 memmove((char *)entry_d, (char *)entry_s, tmp);
2133 * Copy all entry's in the same (sorted) order,
2134 * but allocate attribute info packed and in sequence.
2136 entry_s = &leaf_s->entries[start_s];
2137 entry_d = &leaf_d->entries[start_d];
2139 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2140 ASSERT(be16_to_cpu(entry_s->nameidx)
2141 >= be16_to_cpu(hdr_s->firstused));
2142 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2145 * Code to drop INCOMPLETE entries. Difficult to use as we
2146 * may also need to change the insertion index. Code turned
2147 * off for 6.2, should be revisited later.
2149 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2150 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2151 be16_add(&hdr_s->usedbytes, -tmp);
2152 be16_add(&hdr_s->count, -1);
2153 entry_d--; /* to compensate for ++ in loop hdr */
2155 if ((start_s + i) < offset)
2156 result++; /* insertion index adjustment */
2159 be16_add(&hdr_d->firstused, -tmp);
2160 /* both on-disk, don't endian flip twice */
2161 entry_d->hashval = entry_s->hashval;
2162 /* both on-disk, don't endian flip twice */
2163 entry_d->nameidx = hdr_d->firstused;
2164 entry_d->flags = entry_s->flags;
2165 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2167 memmove(XFS_ATTR_LEAF_NAME(leaf_d, desti),
2168 XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), tmp);
2169 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2171 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2172 be16_add(&hdr_s->usedbytes, -tmp);
2173 be16_add(&hdr_d->usedbytes, tmp);
2174 be16_add(&hdr_s->count, -1);
2175 be16_add(&hdr_d->count, 1);
2176 tmp = be16_to_cpu(hdr_d->count)
2177 * sizeof(xfs_attr_leaf_entry_t)
2178 + sizeof(xfs_attr_leaf_hdr_t);
2179 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2186 * Zero out the entries we just copied.
2188 if (start_s == be16_to_cpu(hdr_s->count)) {
2189 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2190 entry_s = &leaf_s->entries[start_s];
2191 ASSERT(((char *)entry_s + tmp) <=
2192 ((char *)leaf_s + XFS_LBSIZE(mp)));
2193 memset((char *)entry_s, 0, tmp);
2196 * Move the remaining entries down to fill the hole,
2197 * then zero the entries at the top.
2199 tmp = be16_to_cpu(hdr_s->count) - count;
2200 tmp *= sizeof(xfs_attr_leaf_entry_t);
2201 entry_s = &leaf_s->entries[start_s + count];
2202 entry_d = &leaf_s->entries[start_s];
2203 memmove((char *)entry_d, (char *)entry_s, tmp);
2205 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2206 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2207 ASSERT(((char *)entry_s + tmp) <=
2208 ((char *)leaf_s + XFS_LBSIZE(mp)));
2209 memset((char *)entry_s, 0, tmp);
2213 * Fill in the freemap information
2215 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2216 be16_add(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2217 sizeof(xfs_attr_leaf_entry_t));
2218 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2219 - be16_to_cpu(hdr_d->freemap[0].base));
2220 hdr_d->freemap[1].base = 0;
2221 hdr_d->freemap[2].base = 0;
2222 hdr_d->freemap[1].size = 0;
2223 hdr_d->freemap[2].size = 0;
2224 hdr_s->holes = 1; /* leaf may not be compact */
2228 * Compare two leaf blocks "order".
2229 * Return 0 unless leaf2 should go before leaf1.
2232 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2234 xfs_attr_leafblock_t *leaf1, *leaf2;
2236 leaf1 = leaf1_bp->data;
2237 leaf2 = leaf2_bp->data;
2238 ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2239 (be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC));
2240 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2241 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2242 ((be32_to_cpu(leaf2->entries[0].hashval) <
2243 be32_to_cpu(leaf1->entries[0].hashval)) ||
2244 (be32_to_cpu(leaf2->entries[
2245 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2246 be32_to_cpu(leaf1->entries[
2247 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2254 * Pick up the last hashvalue from a leaf block.
2257 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2259 xfs_attr_leafblock_t *leaf;
2262 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2264 *count = be16_to_cpu(leaf->hdr.count);
2265 if (!leaf->hdr.count)
2267 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2271 * Calculate the number of bytes used to store the indicated attribute
2272 * (whether local or remote only calculate bytes in this block).
2275 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2277 xfs_attr_leaf_name_local_t *name_loc;
2278 xfs_attr_leaf_name_remote_t *name_rmt;
2281 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2282 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2283 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, index);
2284 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc->namelen,
2285 INT_GET(name_loc->valuelen,
2288 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, index);
2289 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt->namelen);
2295 * Calculate the number of bytes that would be required to store the new
2296 * attribute (whether local or remote only calculate bytes in this block).
2297 * This routine decides as a side effect whether the attribute will be
2298 * a "local" or a "remote" attribute.
2301 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2305 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen, valuelen);
2306 if (size < XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize)) {
2311 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen);
2320 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2323 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2325 attrlist_cursor_kern_t *cursor;
2326 xfs_attr_leafblock_t *leaf;
2327 xfs_attr_leaf_entry_t *entry;
2328 xfs_attr_leaf_name_local_t *name_loc;
2329 xfs_attr_leaf_name_remote_t *name_rmt;
2334 cursor = context->cursor;
2335 cursor->initted = 1;
2337 xfs_attr_trace_l_cl("blk start", context, leaf);
2340 * Re-find our place in the leaf block if this is a new syscall.
2342 if (context->resynch) {
2343 entry = &leaf->entries[0];
2344 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2345 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2346 if (cursor->offset == context->dupcnt) {
2347 context->dupcnt = 0;
2351 } else if (be32_to_cpu(entry->hashval) >
2353 context->dupcnt = 0;
2357 if (i == be16_to_cpu(leaf->hdr.count)) {
2358 xfs_attr_trace_l_c("not found", context);
2362 entry = &leaf->entries[0];
2365 context->resynch = 0;
2368 * We have found our place, start copying out the new attributes.
2371 for ( ; (i < be16_to_cpu(leaf->hdr.count))
2372 && (retval == 0); entry++, i++) {
2373 attrnames_t *namesp;
2375 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2376 cursor->hashval = be32_to_cpu(entry->hashval);
2380 if (entry->flags & XFS_ATTR_INCOMPLETE)
2381 continue; /* skip incomplete entries */
2382 if (((context->flags & ATTR_SECURE) != 0) !=
2383 ((entry->flags & XFS_ATTR_SECURE) != 0) &&
2384 !(context->flags & ATTR_KERNORMALS))
2385 continue; /* skip non-matching entries */
2386 if (((context->flags & ATTR_ROOT) != 0) !=
2387 ((entry->flags & XFS_ATTR_ROOT) != 0) &&
2388 !(context->flags & ATTR_KERNROOTLS))
2389 continue; /* skip non-matching entries */
2391 namesp = (entry->flags & XFS_ATTR_SECURE) ? &attr_secure :
2392 ((entry->flags & XFS_ATTR_ROOT) ? &attr_trusted :
2395 if (entry->flags & XFS_ATTR_LOCAL) {
2396 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
2397 if (context->flags & ATTR_KERNOVAL) {
2398 ASSERT(context->flags & ATTR_KERNAMELS);
2399 context->count += namesp->attr_namelen +
2400 (int)name_loc->namelen + 1;
2402 retval = xfs_attr_put_listent(context, namesp,
2403 (char *)name_loc->nameval,
2404 (int)name_loc->namelen,
2405 (int)INT_GET(name_loc->valuelen,
2409 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2410 if (context->flags & ATTR_KERNOVAL) {
2411 ASSERT(context->flags & ATTR_KERNAMELS);
2412 context->count += namesp->attr_namelen +
2413 (int)name_rmt->namelen + 1;
2415 retval = xfs_attr_put_listent(context, namesp,
2416 (char *)name_rmt->name,
2417 (int)name_rmt->namelen,
2418 (int)INT_GET(name_rmt->valuelen,
2426 xfs_attr_trace_l_cl("blk end", context, leaf);
2430 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
2431 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
2432 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
2433 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
2434 & ~(sizeof(u_int32_t)-1))
2437 * Format an attribute and copy it out to the user's buffer.
2438 * Take care to check values and protect against them changing later,
2439 * we may be reading them directly out of a user buffer.
2443 xfs_attr_put_listent(xfs_attr_list_context_t *context,
2444 attrnames_t *namesp, char *name, int namelen, int valuelen)
2446 attrlist_ent_t *aep;
2449 ASSERT(!(context->flags & ATTR_KERNOVAL));
2450 if (context->flags & ATTR_KERNAMELS) {
2453 ASSERT(context->count >= 0);
2455 arraytop = context->count + namesp->attr_namelen + namelen + 1;
2456 if (arraytop > context->firstu) {
2457 context->count = -1; /* insufficient space */
2460 offset = (char *)context->alist + context->count;
2461 strncpy(offset, namesp->attr_name, namesp->attr_namelen);
2462 offset += namesp->attr_namelen;
2463 strncpy(offset, name, namelen); /* real name */
2466 context->count += namesp->attr_namelen + namelen + 1;
2470 ASSERT(context->count >= 0);
2471 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
2472 ASSERT(context->firstu >= sizeof(*context->alist));
2473 ASSERT(context->firstu <= context->bufsize);
2475 arraytop = sizeof(*context->alist) +
2476 context->count * sizeof(context->alist->al_offset[0]);
2477 context->firstu -= ATTR_ENTSIZE(namelen);
2478 if (context->firstu < arraytop) {
2479 xfs_attr_trace_l_c("buffer full", context);
2480 context->alist->al_more = 1;
2484 aep = (attrlist_ent_t *)&(((char *)context->alist)[ context->firstu ]);
2485 aep->a_valuelen = valuelen;
2486 memcpy(aep->a_name, name, namelen);
2487 aep->a_name[ namelen ] = 0;
2488 context->alist->al_offset[ context->count++ ] = context->firstu;
2489 context->alist->al_count = context->count;
2490 xfs_attr_trace_l_c("add", context);
2494 /*========================================================================
2495 * Manage the INCOMPLETE flag in a leaf entry
2496 *========================================================================*/
2499 * Clear the INCOMPLETE flag on an entry in a leaf block.
2502 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2504 xfs_attr_leafblock_t *leaf;
2505 xfs_attr_leaf_entry_t *entry;
2506 xfs_attr_leaf_name_remote_t *name_rmt;
2510 xfs_attr_leaf_name_local_t *name_loc;
2516 * Set up the operation.
2518 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2526 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2527 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2528 ASSERT(args->index >= 0);
2529 entry = &leaf->entries[ args->index ];
2530 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2533 if (entry->flags & XFS_ATTR_LOCAL) {
2534 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2535 namelen = name_loc->namelen;
2536 name = (char *)name_loc->nameval;
2538 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2539 namelen = name_rmt->namelen;
2540 name = (char *)name_rmt->name;
2542 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2543 ASSERT(namelen == args->namelen);
2544 ASSERT(memcmp(name, args->name, namelen) == 0);
2547 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2548 xfs_da_log_buf(args->trans, bp,
2549 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2551 if (args->rmtblkno) {
2552 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2553 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2554 INT_SET(name_rmt->valueblk, ARCH_CONVERT, args->rmtblkno);
2555 INT_SET(name_rmt->valuelen, ARCH_CONVERT, args->valuelen);
2556 xfs_da_log_buf(args->trans, bp,
2557 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2559 xfs_da_buf_done(bp);
2562 * Commit the flag value change and start the next trans in series.
2564 error = xfs_attr_rolltrans(&args->trans, args->dp);
2570 * Set the INCOMPLETE flag on an entry in a leaf block.
2573 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2575 xfs_attr_leafblock_t *leaf;
2576 xfs_attr_leaf_entry_t *entry;
2577 xfs_attr_leaf_name_remote_t *name_rmt;
2582 * Set up the operation.
2584 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2592 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2593 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2594 ASSERT(args->index >= 0);
2595 entry = &leaf->entries[ args->index ];
2597 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2598 entry->flags |= XFS_ATTR_INCOMPLETE;
2599 xfs_da_log_buf(args->trans, bp,
2600 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2601 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2602 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2603 name_rmt->valueblk = 0;
2604 name_rmt->valuelen = 0;
2605 xfs_da_log_buf(args->trans, bp,
2606 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2608 xfs_da_buf_done(bp);
2611 * Commit the flag value change and start the next trans in series.
2613 error = xfs_attr_rolltrans(&args->trans, args->dp);
2619 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2620 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2621 * entry given by args->blkno2/index2.
2623 * Note that they could be in different blocks, or in the same block.
2626 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2628 xfs_attr_leafblock_t *leaf1, *leaf2;
2629 xfs_attr_leaf_entry_t *entry1, *entry2;
2630 xfs_attr_leaf_name_remote_t *name_rmt;
2631 xfs_dabuf_t *bp1, *bp2;
2634 xfs_attr_leaf_name_local_t *name_loc;
2635 int namelen1, namelen2;
2636 char *name1, *name2;
2640 * Read the block containing the "old" attr
2642 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2647 ASSERT(bp1 != NULL);
2650 * Read the block containing the "new" attr, if it is different
2652 if (args->blkno2 != args->blkno) {
2653 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2654 -1, &bp2, XFS_ATTR_FORK);
2658 ASSERT(bp2 != NULL);
2664 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2665 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2666 ASSERT(args->index >= 0);
2667 entry1 = &leaf1->entries[ args->index ];
2670 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2671 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2672 ASSERT(args->index2 >= 0);
2673 entry2 = &leaf2->entries[ args->index2 ];
2676 if (entry1->flags & XFS_ATTR_LOCAL) {
2677 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf1, args->index);
2678 namelen1 = name_loc->namelen;
2679 name1 = (char *)name_loc->nameval;
2681 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2682 namelen1 = name_rmt->namelen;
2683 name1 = (char *)name_rmt->name;
2685 if (entry2->flags & XFS_ATTR_LOCAL) {
2686 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf2, args->index2);
2687 namelen2 = name_loc->namelen;
2688 name2 = (char *)name_loc->nameval;
2690 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2691 namelen2 = name_rmt->namelen;
2692 name2 = (char *)name_rmt->name;
2694 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2695 ASSERT(namelen1 == namelen2);
2696 ASSERT(memcmp(name1, name2, namelen1) == 0);
2699 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2700 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2702 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2703 xfs_da_log_buf(args->trans, bp1,
2704 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2705 if (args->rmtblkno) {
2706 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2707 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2708 INT_SET(name_rmt->valueblk, ARCH_CONVERT, args->rmtblkno);
2709 INT_SET(name_rmt->valuelen, ARCH_CONVERT, args->valuelen);
2710 xfs_da_log_buf(args->trans, bp1,
2711 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2714 entry2->flags |= XFS_ATTR_INCOMPLETE;
2715 xfs_da_log_buf(args->trans, bp2,
2716 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2717 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2718 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2719 name_rmt->valueblk = 0;
2720 name_rmt->valuelen = 0;
2721 xfs_da_log_buf(args->trans, bp2,
2722 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2724 xfs_da_buf_done(bp1);
2726 xfs_da_buf_done(bp2);
2729 * Commit the flag value change and start the next trans in series.
2731 error = xfs_attr_rolltrans(&args->trans, args->dp);
2736 /*========================================================================
2737 * Indiscriminately delete the entire attribute fork
2738 *========================================================================*/
2741 * Recurse (gasp!) through the attribute nodes until we find leaves.
2742 * We're doing a depth-first traversal in order to invalidate everything.
2745 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2747 xfs_da_blkinfo_t *info;
2753 * Read block 0 to see what we have to work with.
2754 * We only get here if we have extents, since we remove
2755 * the extents in reverse order the extent containing
2756 * block 0 must still be there.
2758 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2761 blkno = xfs_da_blkno(bp);
2764 * Invalidate the tree, even if the "tree" is only a single leaf block.
2765 * This is a depth-first traversal!
2768 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2769 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2770 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2771 error = xfs_attr_leaf_inactive(trans, dp, bp);
2773 error = XFS_ERROR(EIO);
2774 xfs_da_brelse(*trans, bp);
2780 * Invalidate the incore copy of the root block.
2782 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2785 xfs_da_binval(*trans, bp); /* remove from cache */
2787 * Commit the invalidate and start the next transaction.
2789 error = xfs_attr_rolltrans(trans, dp);
2795 * Recurse (gasp!) through the attribute nodes until we find leaves.
2796 * We're doing a depth-first traversal in order to invalidate everything.
2799 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2802 xfs_da_blkinfo_t *info;
2803 xfs_da_intnode_t *node;
2804 xfs_dablk_t child_fsb;
2805 xfs_daddr_t parent_blkno, child_blkno;
2806 int error, count, i;
2807 xfs_dabuf_t *child_bp;
2810 * Since this code is recursive (gasp!) we must protect ourselves.
2812 if (level > XFS_DA_NODE_MAXDEPTH) {
2813 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2814 return(XFS_ERROR(EIO));
2818 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2819 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2820 count = INT_GET(node->hdr.count, ARCH_CONVERT);
2822 xfs_da_brelse(*trans, bp);
2825 child_fsb = INT_GET(node->btree[0].before, ARCH_CONVERT);
2826 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2829 * If this is the node level just above the leaves, simply loop
2830 * over the leaves removing all of them. If this is higher up
2831 * in the tree, recurse downward.
2833 for (i = 0; i < count; i++) {
2835 * Read the subsidiary block to see what we have to work with.
2836 * Don't do this in a transaction. This is a depth-first
2837 * traversal of the tree so we may deal with many blocks
2838 * before we come back to this one.
2840 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2845 /* save for re-read later */
2846 child_blkno = xfs_da_blkno(child_bp);
2849 * Invalidate the subtree, however we have to.
2851 info = child_bp->data;
2852 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2853 error = xfs_attr_node_inactive(trans, dp,
2855 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2856 error = xfs_attr_leaf_inactive(trans, dp,
2859 error = XFS_ERROR(EIO);
2860 xfs_da_brelse(*trans, child_bp);
2866 * Remove the subsidiary block from the cache
2869 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2870 &child_bp, XFS_ATTR_FORK);
2873 xfs_da_binval(*trans, child_bp);
2877 * If we're not done, re-read the parent to get the next
2878 * child block number.
2880 if ((i+1) < count) {
2881 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2882 &bp, XFS_ATTR_FORK);
2885 child_fsb = INT_GET(node->btree[i+1].before, ARCH_CONVERT);
2886 xfs_da_brelse(*trans, bp);
2889 * Atomically commit the whole invalidate stuff.
2891 if ((error = xfs_attr_rolltrans(trans, dp)))
2899 * Invalidate all of the "remote" value regions pointed to by a particular
2901 * Note that we must release the lock on the buffer so that we are not
2902 * caught holding something that the logging code wants to flush to disk.
2905 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2907 xfs_attr_leafblock_t *leaf;
2908 xfs_attr_leaf_entry_t *entry;
2909 xfs_attr_leaf_name_remote_t *name_rmt;
2910 xfs_attr_inactive_list_t *list, *lp;
2911 int error, count, size, tmp, i;
2914 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2917 * Count the number of "remote" value extents.
2920 entry = &leaf->entries[0];
2921 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2922 if (be16_to_cpu(entry->nameidx) &&
2923 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2924 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2925 if (name_rmt->valueblk)
2931 * If there are no "remote" values, we're done.
2934 xfs_da_brelse(*trans, bp);
2939 * Allocate storage for a list of all the "remote" value extents.
2941 size = count * sizeof(xfs_attr_inactive_list_t);
2942 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2945 * Identify each of the "remote" value extents.
2948 entry = &leaf->entries[0];
2949 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2950 if (be16_to_cpu(entry->nameidx) &&
2951 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2952 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2953 if (name_rmt->valueblk) {
2954 /* both on-disk, don't endian flip twice */
2955 lp->valueblk = name_rmt->valueblk;
2956 INT_SET(lp->valuelen, ARCH_CONVERT,
2957 XFS_B_TO_FSB(dp->i_mount,
2958 INT_GET(name_rmt->valuelen,
2964 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2967 * Invalidate each of the "remote" value extents.
2970 for (lp = list, i = 0; i < count; i++, lp++) {
2971 tmp = xfs_attr_leaf_freextent(trans, dp,
2972 INT_GET(lp->valueblk,
2974 INT_GET(lp->valuelen,
2977 error = tmp; /* save only the 1st errno */
2980 kmem_free((xfs_caddr_t)list, size);
2985 * Look at all the extents for this logical region,
2986 * invalidate any buffers that are incore/in transactions.
2989 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2990 xfs_dablk_t blkno, int blkcnt)
2992 xfs_bmbt_irec_t map;
2994 int tblkcnt, dblkcnt, nmap, error;
2999 * Roll through the "value", invalidating the attribute value's
3004 while (tblkcnt > 0) {
3006 * Try to remember where we decided to put the value.
3009 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
3010 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
3011 NULL, 0, &map, &nmap, NULL);
3016 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3019 * If it's a hole, these are already unmapped
3020 * so there's nothing to invalidate.
3022 if (map.br_startblock != HOLESTARTBLOCK) {
3024 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3026 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3028 bp = xfs_trans_get_buf(*trans,
3029 dp->i_mount->m_ddev_targp,
3030 dblkno, dblkcnt, XFS_BUF_LOCK);
3031 xfs_trans_binval(*trans, bp);
3033 * Roll to next transaction.
3035 if ((error = xfs_attr_rolltrans(trans, dp)))
3039 tblkno += map.br_blockcount;
3040 tblkcnt -= map.br_blockcount;
3048 * Roll from one trans in the sequence of PERMANENT transactions to the next.
3051 xfs_attr_rolltrans(xfs_trans_t **transp, xfs_inode_t *dp)
3054 unsigned int logres, count;
3058 * Ensure that the inode is always logged.
3061 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
3064 * Copy the critical parameters from one trans to the next.
3066 logres = trans->t_log_res;
3067 count = trans->t_log_count;
3068 *transp = xfs_trans_dup(trans);
3071 * Commit the current transaction.
3072 * If this commit failed, then it'd just unlock those items that
3073 * are not marked ihold. That also means that a filesystem shutdown
3074 * is in progress. The caller takes the responsibility to cancel
3075 * the duplicate transaction that gets returned.
3077 if ((error = xfs_trans_commit(trans, 0, NULL)))
3083 * Reserve space in the log for th next transaction.
3084 * This also pushes items in the "AIL", the list of logged items,
3085 * out to disk if they are taking up space at the tail of the log
3086 * that we want to use. This requires that either nothing be locked
3087 * across this call, or that anything that is locked be logged in
3088 * the prior and the next transactions.
3090 error = xfs_trans_reserve(trans, 0, logres, 0,
3091 XFS_TRANS_PERM_LOG_RES, count);
3093 * Ensure that the inode is in the new transaction and locked.
3096 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
3097 xfs_trans_ihold(trans, dp);