4bfc732bc9c9a8eb3391930e1b92b400b9da7f66
[linux-block.git] / fs / xfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
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.
8  *
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.
13  *
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
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_alloc.h"
32 #include "xfs_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_bmap.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42
43 /*
44  * xfs_attr_leaf.c
45  *
46  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
47  */
48
49 /*========================================================================
50  * Function prototypes for the kernel.
51  *========================================================================*/
52
53 /*
54  * Routines used for growing the Btree.
55  */
56 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
57                                 struct xfs_buf **bpp);
58 STATIC int xfs_attr_leaf_add_work(struct xfs_buf *leaf_buffer,
59                                   xfs_da_args_t *args, int freemap_index);
60 STATIC void xfs_attr_leaf_compact(struct xfs_da_args *args,
61                                   struct xfs_buf *leaf_buffer);
62 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
63                                                    xfs_da_state_blk_t *blk1,
64                                                    xfs_da_state_blk_t *blk2);
65 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
66                                            xfs_da_state_blk_t *leaf_blk_1,
67                                            xfs_da_state_blk_t *leaf_blk_2,
68                                            int *number_entries_in_blk1,
69                                            int *number_usedbytes_in_blk1);
70
71 /*
72  * Routines used for shrinking the Btree.
73  */
74 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
75                                   struct xfs_buf *bp, int level);
76 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
77                                   struct xfs_buf *bp);
78 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
79                                    xfs_dablk_t blkno, int blkcnt);
80
81 /*
82  * Utility routines.
83  */
84 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
85                                          int src_start,
86                                          xfs_attr_leafblock_t *dst_leaf,
87                                          int dst_start, int move_count,
88                                          xfs_mount_t *mp);
89 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
90
91 /*========================================================================
92  * Namespace helper routines
93  *========================================================================*/
94
95 /*
96  * If namespace bits don't match return 0.
97  * If all match then return 1.
98  */
99 STATIC int
100 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
101 {
102         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
103 }
104
105
106 /*========================================================================
107  * External routines when attribute fork size < XFS_LITINO(mp).
108  *========================================================================*/
109
110 /*
111  * Query whether the requested number of additional bytes of extended
112  * attribute space will be able to fit inline.
113  *
114  * Returns zero if not, else the di_forkoff fork offset to be used in the
115  * literal area for attribute data once the new bytes have been added.
116  *
117  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
118  * special case for dev/uuid inodes, they have fixed size data forks.
119  */
120 int
121 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
122 {
123         int offset;
124         int minforkoff; /* lower limit on valid forkoff locations */
125         int maxforkoff; /* upper limit on valid forkoff locations */
126         int dsize;
127         xfs_mount_t *mp = dp->i_mount;
128
129         offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
130
131         switch (dp->i_d.di_format) {
132         case XFS_DINODE_FMT_DEV:
133                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
134                 return (offset >= minforkoff) ? minforkoff : 0;
135         case XFS_DINODE_FMT_UUID:
136                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
137                 return (offset >= minforkoff) ? minforkoff : 0;
138         }
139
140         /*
141          * If the requested numbers of bytes is smaller or equal to the
142          * current attribute fork size we can always proceed.
143          *
144          * Note that if_bytes in the data fork might actually be larger than
145          * the current data fork size is due to delalloc extents. In that
146          * case either the extent count will go down when they are converted
147          * to real extents, or the delalloc conversion will take care of the
148          * literal area rebalancing.
149          */
150         if (bytes <= XFS_IFORK_ASIZE(dp))
151                 return dp->i_d.di_forkoff;
152
153         /*
154          * For attr2 we can try to move the forkoff if there is space in the
155          * literal area, but for the old format we are done if there is no
156          * space in the fixed attribute fork.
157          */
158         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
159                 return 0;
160
161         dsize = dp->i_df.if_bytes;
162
163         switch (dp->i_d.di_format) {
164         case XFS_DINODE_FMT_EXTENTS:
165                 /*
166                  * If there is no attr fork and the data fork is extents, 
167                  * determine if creating the default attr fork will result
168                  * in the extents form migrating to btree. If so, the
169                  * minimum offset only needs to be the space required for
170                  * the btree root.
171                  */
172                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
173                     xfs_default_attroffset(dp))
174                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
175                 break;
176         case XFS_DINODE_FMT_BTREE:
177                 /*
178                  * If we have a data btree then keep forkoff if we have one,
179                  * otherwise we are adding a new attr, so then we set
180                  * minforkoff to where the btree root can finish so we have
181                  * plenty of room for attrs
182                  */
183                 if (dp->i_d.di_forkoff) {
184                         if (offset < dp->i_d.di_forkoff)
185                                 return 0;
186                         return dp->i_d.di_forkoff;
187                 }
188                 dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
189                 break;
190         }
191
192         /*
193          * A data fork btree root must have space for at least
194          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
195          */
196         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
197         minforkoff = roundup(minforkoff, 8) >> 3;
198
199         /* attr fork btree root can have at least this many key/ptr pairs */
200         maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
201         maxforkoff = maxforkoff >> 3;   /* rounded down */
202
203         if (offset >= maxforkoff)
204                 return maxforkoff;
205         if (offset >= minforkoff)
206                 return offset;
207         return 0;
208 }
209
210 /*
211  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
212  */
213 STATIC void
214 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
215 {
216         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
217             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
218                 spin_lock(&mp->m_sb_lock);
219                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
220                         xfs_sb_version_addattr2(&mp->m_sb);
221                         spin_unlock(&mp->m_sb_lock);
222                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
223                 } else
224                         spin_unlock(&mp->m_sb_lock);
225         }
226 }
227
228 /*
229  * Create the initial contents of a shortform attribute list.
230  */
231 void
232 xfs_attr_shortform_create(xfs_da_args_t *args)
233 {
234         xfs_attr_sf_hdr_t *hdr;
235         xfs_inode_t *dp;
236         xfs_ifork_t *ifp;
237
238         trace_xfs_attr_sf_create(args);
239
240         dp = args->dp;
241         ASSERT(dp != NULL);
242         ifp = dp->i_afp;
243         ASSERT(ifp != NULL);
244         ASSERT(ifp->if_bytes == 0);
245         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
246                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
247                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
248                 ifp->if_flags |= XFS_IFINLINE;
249         } else {
250                 ASSERT(ifp->if_flags & XFS_IFINLINE);
251         }
252         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
253         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
254         hdr->count = 0;
255         hdr->totsize = cpu_to_be16(sizeof(*hdr));
256         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
257 }
258
259 /*
260  * Add a name/value pair to the shortform attribute list.
261  * Overflow from the inode has already been checked for.
262  */
263 void
264 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
265 {
266         xfs_attr_shortform_t *sf;
267         xfs_attr_sf_entry_t *sfe;
268         int i, offset, size;
269         xfs_mount_t *mp;
270         xfs_inode_t *dp;
271         xfs_ifork_t *ifp;
272
273         trace_xfs_attr_sf_add(args);
274
275         dp = args->dp;
276         mp = dp->i_mount;
277         dp->i_d.di_forkoff = forkoff;
278
279         ifp = dp->i_afp;
280         ASSERT(ifp->if_flags & XFS_IFINLINE);
281         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
282         sfe = &sf->list[0];
283         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
284 #ifdef DEBUG
285                 if (sfe->namelen != args->namelen)
286                         continue;
287                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
288                         continue;
289                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
290                         continue;
291                 ASSERT(0);
292 #endif
293         }
294
295         offset = (char *)sfe - (char *)sf;
296         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
297         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
298         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
299         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
300
301         sfe->namelen = args->namelen;
302         sfe->valuelen = args->valuelen;
303         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
304         memcpy(sfe->nameval, args->name, args->namelen);
305         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
306         sf->hdr.count++;
307         be16_add_cpu(&sf->hdr.totsize, size);
308         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
309
310         xfs_sbversion_add_attr2(mp, args->trans);
311 }
312
313 /*
314  * After the last attribute is removed revert to original inode format,
315  * making all literal area available to the data fork once more.
316  */
317 STATIC void
318 xfs_attr_fork_reset(
319         struct xfs_inode        *ip,
320         struct xfs_trans        *tp)
321 {
322         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
323         ip->i_d.di_forkoff = 0;
324         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
325
326         ASSERT(ip->i_d.di_anextents == 0);
327         ASSERT(ip->i_afp == NULL);
328
329         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
330 }
331
332 /*
333  * Remove an attribute from the shortform attribute list structure.
334  */
335 int
336 xfs_attr_shortform_remove(xfs_da_args_t *args)
337 {
338         xfs_attr_shortform_t *sf;
339         xfs_attr_sf_entry_t *sfe;
340         int base, size=0, end, totsize, i;
341         xfs_mount_t *mp;
342         xfs_inode_t *dp;
343
344         trace_xfs_attr_sf_remove(args);
345
346         dp = args->dp;
347         mp = dp->i_mount;
348         base = sizeof(xfs_attr_sf_hdr_t);
349         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
350         sfe = &sf->list[0];
351         end = sf->hdr.count;
352         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
353                                         base += size, i++) {
354                 size = XFS_ATTR_SF_ENTSIZE(sfe);
355                 if (sfe->namelen != args->namelen)
356                         continue;
357                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
358                         continue;
359                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
360                         continue;
361                 break;
362         }
363         if (i == end)
364                 return(XFS_ERROR(ENOATTR));
365
366         /*
367          * Fix up the attribute fork data, covering the hole
368          */
369         end = base + size;
370         totsize = be16_to_cpu(sf->hdr.totsize);
371         if (end != totsize)
372                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
373         sf->hdr.count--;
374         be16_add_cpu(&sf->hdr.totsize, -size);
375
376         /*
377          * Fix up the start offset of the attribute fork
378          */
379         totsize -= size;
380         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
381             (mp->m_flags & XFS_MOUNT_ATTR2) &&
382             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
383             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
384                 xfs_attr_fork_reset(dp, args->trans);
385         } else {
386                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
387                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
388                 ASSERT(dp->i_d.di_forkoff);
389                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
390                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
391                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
392                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
393                 xfs_trans_log_inode(args->trans, dp,
394                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
395         }
396
397         xfs_sbversion_add_attr2(mp, args->trans);
398
399         return(0);
400 }
401
402 /*
403  * Look up a name in a shortform attribute list structure.
404  */
405 /*ARGSUSED*/
406 int
407 xfs_attr_shortform_lookup(xfs_da_args_t *args)
408 {
409         xfs_attr_shortform_t *sf;
410         xfs_attr_sf_entry_t *sfe;
411         int i;
412         xfs_ifork_t *ifp;
413
414         trace_xfs_attr_sf_lookup(args);
415
416         ifp = args->dp->i_afp;
417         ASSERT(ifp->if_flags & XFS_IFINLINE);
418         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
419         sfe = &sf->list[0];
420         for (i = 0; i < sf->hdr.count;
421                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
422                 if (sfe->namelen != args->namelen)
423                         continue;
424                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
425                         continue;
426                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
427                         continue;
428                 return(XFS_ERROR(EEXIST));
429         }
430         return(XFS_ERROR(ENOATTR));
431 }
432
433 /*
434  * Look up a name in a shortform attribute list structure.
435  */
436 /*ARGSUSED*/
437 int
438 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
439 {
440         xfs_attr_shortform_t *sf;
441         xfs_attr_sf_entry_t *sfe;
442         int i;
443
444         ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
445         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
446         sfe = &sf->list[0];
447         for (i = 0; i < sf->hdr.count;
448                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
449                 if (sfe->namelen != args->namelen)
450                         continue;
451                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
452                         continue;
453                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
454                         continue;
455                 if (args->flags & ATTR_KERNOVAL) {
456                         args->valuelen = sfe->valuelen;
457                         return(XFS_ERROR(EEXIST));
458                 }
459                 if (args->valuelen < sfe->valuelen) {
460                         args->valuelen = sfe->valuelen;
461                         return(XFS_ERROR(ERANGE));
462                 }
463                 args->valuelen = sfe->valuelen;
464                 memcpy(args->value, &sfe->nameval[args->namelen],
465                                                     args->valuelen);
466                 return(XFS_ERROR(EEXIST));
467         }
468         return(XFS_ERROR(ENOATTR));
469 }
470
471 /*
472  * Convert from using the shortform to the leaf.
473  */
474 int
475 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
476 {
477         xfs_inode_t *dp;
478         xfs_attr_shortform_t *sf;
479         xfs_attr_sf_entry_t *sfe;
480         xfs_da_args_t nargs;
481         char *tmpbuffer;
482         int error, i, size;
483         xfs_dablk_t blkno;
484         struct xfs_buf *bp;
485         xfs_ifork_t *ifp;
486
487         trace_xfs_attr_sf_to_leaf(args);
488
489         dp = args->dp;
490         ifp = dp->i_afp;
491         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
492         size = be16_to_cpu(sf->hdr.totsize);
493         tmpbuffer = kmem_alloc(size, KM_SLEEP);
494         ASSERT(tmpbuffer != NULL);
495         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
496         sf = (xfs_attr_shortform_t *)tmpbuffer;
497
498         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
499         bp = NULL;
500         error = xfs_da_grow_inode(args, &blkno);
501         if (error) {
502                 /*
503                  * If we hit an IO error middle of the transaction inside
504                  * grow_inode(), we may have inconsistent data. Bail out.
505                  */
506                 if (error == EIO)
507                         goto out;
508                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
509                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
510                 goto out;
511         }
512
513         ASSERT(blkno == 0);
514         error = xfs_attr_leaf_create(args, blkno, &bp);
515         if (error) {
516                 error = xfs_da_shrink_inode(args, 0, bp);
517                 bp = NULL;
518                 if (error)
519                         goto out;
520                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
521                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
522                 goto out;
523         }
524
525         memset((char *)&nargs, 0, sizeof(nargs));
526         nargs.dp = dp;
527         nargs.firstblock = args->firstblock;
528         nargs.flist = args->flist;
529         nargs.total = args->total;
530         nargs.whichfork = XFS_ATTR_FORK;
531         nargs.trans = args->trans;
532         nargs.op_flags = XFS_DA_OP_OKNOENT;
533
534         sfe = &sf->list[0];
535         for (i = 0; i < sf->hdr.count; i++) {
536                 nargs.name = sfe->nameval;
537                 nargs.namelen = sfe->namelen;
538                 nargs.value = &sfe->nameval[nargs.namelen];
539                 nargs.valuelen = sfe->valuelen;
540                 nargs.hashval = xfs_da_hashname(sfe->nameval,
541                                                 sfe->namelen);
542                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
543                 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
544                 ASSERT(error == ENOATTR);
545                 error = xfs_attr_leaf_add(bp, &nargs);
546                 ASSERT(error != ENOSPC);
547                 if (error)
548                         goto out;
549                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
550         }
551         error = 0;
552
553 out:
554         kmem_free(tmpbuffer);
555         return(error);
556 }
557
558 STATIC int
559 xfs_attr_shortform_compare(const void *a, const void *b)
560 {
561         xfs_attr_sf_sort_t *sa, *sb;
562
563         sa = (xfs_attr_sf_sort_t *)a;
564         sb = (xfs_attr_sf_sort_t *)b;
565         if (sa->hash < sb->hash) {
566                 return(-1);
567         } else if (sa->hash > sb->hash) {
568                 return(1);
569         } else {
570                 return(sa->entno - sb->entno);
571         }
572 }
573
574
575 #define XFS_ISRESET_CURSOR(cursor) \
576         (!((cursor)->initted) && !((cursor)->hashval) && \
577          !((cursor)->blkno) && !((cursor)->offset))
578 /*
579  * Copy out entries of shortform attribute lists for attr_list().
580  * Shortform attribute lists are not stored in hashval sorted order.
581  * If the output buffer is not large enough to hold them all, then we
582  * we have to calculate each entries' hashvalue and sort them before
583  * we can begin returning them to the user.
584  */
585 /*ARGSUSED*/
586 int
587 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
588 {
589         attrlist_cursor_kern_t *cursor;
590         xfs_attr_sf_sort_t *sbuf, *sbp;
591         xfs_attr_shortform_t *sf;
592         xfs_attr_sf_entry_t *sfe;
593         xfs_inode_t *dp;
594         int sbsize, nsbuf, count, i;
595         int error;
596
597         ASSERT(context != NULL);
598         dp = context->dp;
599         ASSERT(dp != NULL);
600         ASSERT(dp->i_afp != NULL);
601         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
602         ASSERT(sf != NULL);
603         if (!sf->hdr.count)
604                 return(0);
605         cursor = context->cursor;
606         ASSERT(cursor != NULL);
607
608         trace_xfs_attr_list_sf(context);
609
610         /*
611          * If the buffer is large enough and the cursor is at the start,
612          * do not bother with sorting since we will return everything in
613          * one buffer and another call using the cursor won't need to be
614          * made.
615          * Note the generous fudge factor of 16 overhead bytes per entry.
616          * If bufsize is zero then put_listent must be a search function
617          * and can just scan through what we have.
618          */
619         if (context->bufsize == 0 ||
620             (XFS_ISRESET_CURSOR(cursor) &&
621              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
622                 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
623                         error = context->put_listent(context,
624                                            sfe->flags,
625                                            sfe->nameval,
626                                            (int)sfe->namelen,
627                                            (int)sfe->valuelen,
628                                            &sfe->nameval[sfe->namelen]);
629
630                         /*
631                          * Either search callback finished early or
632                          * didn't fit it all in the buffer after all.
633                          */
634                         if (context->seen_enough)
635                                 break;
636
637                         if (error)
638                                 return error;
639                         sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
640                 }
641                 trace_xfs_attr_list_sf_all(context);
642                 return(0);
643         }
644
645         /* do no more for a search callback */
646         if (context->bufsize == 0)
647                 return 0;
648
649         /*
650          * It didn't all fit, so we have to sort everything on hashval.
651          */
652         sbsize = sf->hdr.count * sizeof(*sbuf);
653         sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
654
655         /*
656          * Scan the attribute list for the rest of the entries, storing
657          * the relevant info from only those that match into a buffer.
658          */
659         nsbuf = 0;
660         for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
661                 if (unlikely(
662                     ((char *)sfe < (char *)sf) ||
663                     ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
664                         XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
665                                              XFS_ERRLEVEL_LOW,
666                                              context->dp->i_mount, sfe);
667                         kmem_free(sbuf);
668                         return XFS_ERROR(EFSCORRUPTED);
669                 }
670
671                 sbp->entno = i;
672                 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
673                 sbp->name = sfe->nameval;
674                 sbp->namelen = sfe->namelen;
675                 /* These are bytes, and both on-disk, don't endian-flip */
676                 sbp->valuelen = sfe->valuelen;
677                 sbp->flags = sfe->flags;
678                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
679                 sbp++;
680                 nsbuf++;
681         }
682
683         /*
684          * Sort the entries on hash then entno.
685          */
686         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
687
688         /*
689          * Re-find our place IN THE SORTED LIST.
690          */
691         count = 0;
692         cursor->initted = 1;
693         cursor->blkno = 0;
694         for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
695                 if (sbp->hash == cursor->hashval) {
696                         if (cursor->offset == count) {
697                                 break;
698                         }
699                         count++;
700                 } else if (sbp->hash > cursor->hashval) {
701                         break;
702                 }
703         }
704         if (i == nsbuf) {
705                 kmem_free(sbuf);
706                 return(0);
707         }
708
709         /*
710          * Loop putting entries into the user buffer.
711          */
712         for ( ; i < nsbuf; i++, sbp++) {
713                 if (cursor->hashval != sbp->hash) {
714                         cursor->hashval = sbp->hash;
715                         cursor->offset = 0;
716                 }
717                 error = context->put_listent(context,
718                                         sbp->flags,
719                                         sbp->name,
720                                         sbp->namelen,
721                                         sbp->valuelen,
722                                         &sbp->name[sbp->namelen]);
723                 if (error)
724                         return error;
725                 if (context->seen_enough)
726                         break;
727                 cursor->offset++;
728         }
729
730         kmem_free(sbuf);
731         return(0);
732 }
733
734 /*
735  * Check a leaf attribute block to see if all the entries would fit into
736  * a shortform attribute list.
737  */
738 int
739 xfs_attr_shortform_allfit(
740         struct xfs_buf  *bp,
741         struct xfs_inode *dp)
742 {
743         xfs_attr_leafblock_t *leaf;
744         xfs_attr_leaf_entry_t *entry;
745         xfs_attr_leaf_name_local_t *name_loc;
746         int bytes, i;
747
748         leaf = bp->b_addr;
749         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
750
751         entry = &leaf->entries[0];
752         bytes = sizeof(struct xfs_attr_sf_hdr);
753         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
754                 if (entry->flags & XFS_ATTR_INCOMPLETE)
755                         continue;               /* don't copy partial entries */
756                 if (!(entry->flags & XFS_ATTR_LOCAL))
757                         return(0);
758                 name_loc = xfs_attr_leaf_name_local(leaf, i);
759                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
760                         return(0);
761                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
762                         return(0);
763                 bytes += sizeof(struct xfs_attr_sf_entry)-1
764                                 + name_loc->namelen
765                                 + be16_to_cpu(name_loc->valuelen);
766         }
767         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
768             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
769             (bytes == sizeof(struct xfs_attr_sf_hdr)))
770                 return(-1);
771         return(xfs_attr_shortform_bytesfit(dp, bytes));
772 }
773
774 /*
775  * Convert a leaf attribute list to shortform attribute list
776  */
777 int
778 xfs_attr_leaf_to_shortform(
779         struct xfs_buf  *bp,
780         xfs_da_args_t   *args,
781         int             forkoff)
782 {
783         xfs_attr_leafblock_t *leaf;
784         xfs_attr_leaf_entry_t *entry;
785         xfs_attr_leaf_name_local_t *name_loc;
786         xfs_da_args_t nargs;
787         xfs_inode_t *dp;
788         char *tmpbuffer;
789         int error, i;
790
791         trace_xfs_attr_leaf_to_sf(args);
792
793         dp = args->dp;
794         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
795         ASSERT(tmpbuffer != NULL);
796
797         ASSERT(bp != NULL);
798         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
799         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
800         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
801         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
802
803         /*
804          * Clean out the prior contents of the attribute list.
805          */
806         error = xfs_da_shrink_inode(args, 0, bp);
807         if (error)
808                 goto out;
809
810         if (forkoff == -1) {
811                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
812                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
813                 xfs_attr_fork_reset(dp, args->trans);
814                 goto out;
815         }
816
817         xfs_attr_shortform_create(args);
818
819         /*
820          * Copy the attributes
821          */
822         memset((char *)&nargs, 0, sizeof(nargs));
823         nargs.dp = dp;
824         nargs.firstblock = args->firstblock;
825         nargs.flist = args->flist;
826         nargs.total = args->total;
827         nargs.whichfork = XFS_ATTR_FORK;
828         nargs.trans = args->trans;
829         nargs.op_flags = XFS_DA_OP_OKNOENT;
830         entry = &leaf->entries[0];
831         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
832                 if (entry->flags & XFS_ATTR_INCOMPLETE)
833                         continue;       /* don't copy partial entries */
834                 if (!entry->nameidx)
835                         continue;
836                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
837                 name_loc = xfs_attr_leaf_name_local(leaf, i);
838                 nargs.name = name_loc->nameval;
839                 nargs.namelen = name_loc->namelen;
840                 nargs.value = &name_loc->nameval[nargs.namelen];
841                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
842                 nargs.hashval = be32_to_cpu(entry->hashval);
843                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
844                 xfs_attr_shortform_add(&nargs, forkoff);
845         }
846         error = 0;
847
848 out:
849         kmem_free(tmpbuffer);
850         return(error);
851 }
852
853 /*
854  * Convert from using a single leaf to a root node and a leaf.
855  */
856 int
857 xfs_attr_leaf_to_node(xfs_da_args_t *args)
858 {
859         xfs_attr_leafblock_t *leaf;
860         xfs_da_intnode_t *node;
861         xfs_inode_t *dp;
862         struct xfs_buf *bp1, *bp2;
863         xfs_dablk_t blkno;
864         int error;
865
866         trace_xfs_attr_leaf_to_node(args);
867
868         dp = args->dp;
869         bp1 = bp2 = NULL;
870         error = xfs_da_grow_inode(args, &blkno);
871         if (error)
872                 goto out;
873         error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
874                                              XFS_ATTR_FORK);
875         if (error)
876                 goto out;
877         ASSERT(bp1 != NULL);
878         bp2 = NULL;
879         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
880                                             XFS_ATTR_FORK);
881         if (error)
882                 goto out;
883         ASSERT(bp2 != NULL);
884         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(dp->i_mount));
885         bp1 = NULL;
886         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
887
888         /*
889          * Set up the new root node.
890          */
891         error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
892         if (error)
893                 goto out;
894         node = bp1->b_addr;
895         leaf = bp2->b_addr;
896         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
897         /* both on-disk, don't endian-flip twice */
898         node->btree[0].hashval =
899                 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
900         node->btree[0].before = cpu_to_be32(blkno);
901         node->hdr.count = cpu_to_be16(1);
902         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
903         error = 0;
904 out:
905         return(error);
906 }
907
908
909 /*========================================================================
910  * Routines used for growing the Btree.
911  *========================================================================*/
912
913 /*
914  * Create the initial contents of a leaf attribute list
915  * or a leaf in a node attribute list.
916  */
917 STATIC int
918 xfs_attr_leaf_create(
919         xfs_da_args_t   *args,
920         xfs_dablk_t     blkno,
921         struct xfs_buf  **bpp)
922 {
923         xfs_attr_leafblock_t *leaf;
924         xfs_attr_leaf_hdr_t *hdr;
925         xfs_inode_t *dp;
926         struct xfs_buf *bp;
927         int error;
928
929         trace_xfs_attr_leaf_create(args);
930
931         dp = args->dp;
932         ASSERT(dp != NULL);
933         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
934                                             XFS_ATTR_FORK);
935         if (error)
936                 return(error);
937         ASSERT(bp != NULL);
938         leaf = bp->b_addr;
939         memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
940         hdr = &leaf->hdr;
941         hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
942         hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
943         if (!hdr->firstused) {
944                 hdr->firstused = cpu_to_be16(
945                         XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
946         }
947
948         hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
949         hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
950                                            sizeof(xfs_attr_leaf_hdr_t));
951
952         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
953
954         *bpp = bp;
955         return(0);
956 }
957
958 /*
959  * Split the leaf node, rebalance, then add the new entry.
960  */
961 int
962 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
963                                    xfs_da_state_blk_t *newblk)
964 {
965         xfs_dablk_t blkno;
966         int error;
967
968         trace_xfs_attr_leaf_split(state->args);
969
970         /*
971          * Allocate space for a new leaf node.
972          */
973         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
974         error = xfs_da_grow_inode(state->args, &blkno);
975         if (error)
976                 return(error);
977         error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
978         if (error)
979                 return(error);
980         newblk->blkno = blkno;
981         newblk->magic = XFS_ATTR_LEAF_MAGIC;
982
983         /*
984          * Rebalance the entries across the two leaves.
985          * NOTE: rebalance() currently depends on the 2nd block being empty.
986          */
987         xfs_attr_leaf_rebalance(state, oldblk, newblk);
988         error = xfs_da_blk_link(state, oldblk, newblk);
989         if (error)
990                 return(error);
991
992         /*
993          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
994          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
995          * "new" attrs info.  Will need the "old" info to remove it later.
996          *
997          * Insert the "new" entry in the correct block.
998          */
999         if (state->inleaf) {
1000                 trace_xfs_attr_leaf_add_old(state->args);
1001                 error = xfs_attr_leaf_add(oldblk->bp, state->args);
1002         } else {
1003                 trace_xfs_attr_leaf_add_new(state->args);
1004                 error = xfs_attr_leaf_add(newblk->bp, state->args);
1005         }
1006
1007         /*
1008          * Update last hashval in each block since we added the name.
1009          */
1010         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1011         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1012         return(error);
1013 }
1014
1015 /*
1016  * Add a name to the leaf attribute list structure.
1017  */
1018 int
1019 xfs_attr_leaf_add(
1020         struct xfs_buf          *bp,
1021         struct xfs_da_args      *args)
1022 {
1023         xfs_attr_leafblock_t *leaf;
1024         xfs_attr_leaf_hdr_t *hdr;
1025         xfs_attr_leaf_map_t *map;
1026         int tablesize, entsize, sum, tmp, i;
1027
1028         trace_xfs_attr_leaf_add(args);
1029
1030         leaf = bp->b_addr;
1031         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1032         ASSERT((args->index >= 0)
1033                 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1034         hdr = &leaf->hdr;
1035         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1036                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1037
1038         /*
1039          * Search through freemap for first-fit on new name length.
1040          * (may need to figure in size of entry struct too)
1041          */
1042         tablesize = (be16_to_cpu(hdr->count) + 1)
1043                                         * sizeof(xfs_attr_leaf_entry_t)
1044                                         + sizeof(xfs_attr_leaf_hdr_t);
1045         map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1046         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1047                 if (tablesize > be16_to_cpu(hdr->firstused)) {
1048                         sum += be16_to_cpu(map->size);
1049                         continue;
1050                 }
1051                 if (!map->size)
1052                         continue;       /* no space in this map */
1053                 tmp = entsize;
1054                 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1055                         tmp += sizeof(xfs_attr_leaf_entry_t);
1056                 if (be16_to_cpu(map->size) >= tmp) {
1057                         tmp = xfs_attr_leaf_add_work(bp, args, i);
1058                         return(tmp);
1059                 }
1060                 sum += be16_to_cpu(map->size);
1061         }
1062
1063         /*
1064          * If there are no holes in the address space of the block,
1065          * and we don't have enough freespace, then compaction will do us
1066          * no good and we should just give up.
1067          */
1068         if (!hdr->holes && (sum < entsize))
1069                 return(XFS_ERROR(ENOSPC));
1070
1071         /*
1072          * Compact the entries to coalesce free space.
1073          * This may change the hdr->count via dropping INCOMPLETE entries.
1074          */
1075         xfs_attr_leaf_compact(args, bp);
1076
1077         /*
1078          * After compaction, the block is guaranteed to have only one
1079          * free region, in freemap[0].  If it is not big enough, give up.
1080          */
1081         if (be16_to_cpu(hdr->freemap[0].size)
1082                                 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1083                 return(XFS_ERROR(ENOSPC));
1084
1085         return(xfs_attr_leaf_add_work(bp, args, 0));
1086 }
1087
1088 /*
1089  * Add a name to a leaf attribute list structure.
1090  */
1091 STATIC int
1092 xfs_attr_leaf_add_work(
1093         struct xfs_buf  *bp,
1094         xfs_da_args_t   *args,
1095         int             mapindex)
1096 {
1097         xfs_attr_leafblock_t *leaf;
1098         xfs_attr_leaf_hdr_t *hdr;
1099         xfs_attr_leaf_entry_t *entry;
1100         xfs_attr_leaf_name_local_t *name_loc;
1101         xfs_attr_leaf_name_remote_t *name_rmt;
1102         xfs_attr_leaf_map_t *map;
1103         xfs_mount_t *mp;
1104         int tmp, i;
1105
1106         trace_xfs_attr_leaf_add_work(args);
1107
1108         leaf = bp->b_addr;
1109         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1110         hdr = &leaf->hdr;
1111         ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1112         ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1113
1114         /*
1115          * Force open some space in the entry array and fill it in.
1116          */
1117         entry = &leaf->entries[args->index];
1118         if (args->index < be16_to_cpu(hdr->count)) {
1119                 tmp  = be16_to_cpu(hdr->count) - args->index;
1120                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1121                 memmove((char *)(entry+1), (char *)entry, tmp);
1122                 xfs_trans_log_buf(args->trans, bp,
1123                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1124         }
1125         be16_add_cpu(&hdr->count, 1);
1126
1127         /*
1128          * Allocate space for the new string (at the end of the run).
1129          */
1130         map = &hdr->freemap[mapindex];
1131         mp = args->trans->t_mountp;
1132         ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1133         ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1134         ASSERT(be16_to_cpu(map->size) >=
1135                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1136                                          mp->m_sb.sb_blocksize, NULL));
1137         ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1138         ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1139         be16_add_cpu(&map->size,
1140                 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1141                                           mp->m_sb.sb_blocksize, &tmp));
1142         entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1143                                      be16_to_cpu(map->size));
1144         entry->hashval = cpu_to_be32(args->hashval);
1145         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1146         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1147         if (args->op_flags & XFS_DA_OP_RENAME) {
1148                 entry->flags |= XFS_ATTR_INCOMPLETE;
1149                 if ((args->blkno2 == args->blkno) &&
1150                     (args->index2 <= args->index)) {
1151                         args->index2++;
1152                 }
1153         }
1154         xfs_trans_log_buf(args->trans, bp,
1155                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1156         ASSERT((args->index == 0) ||
1157                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1158         ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1159                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1160
1161         /*
1162          * For "remote" attribute values, simply note that we need to
1163          * allocate space for the "remote" value.  We can't actually
1164          * allocate the extents in this transaction, and we can't decide
1165          * which blocks they should be as we might allocate more blocks
1166          * as part of this transaction (a split operation for example).
1167          */
1168         if (entry->flags & XFS_ATTR_LOCAL) {
1169                 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
1170                 name_loc->namelen = args->namelen;
1171                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1172                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1173                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1174                                    be16_to_cpu(name_loc->valuelen));
1175         } else {
1176                 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
1177                 name_rmt->namelen = args->namelen;
1178                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1179                 entry->flags |= XFS_ATTR_INCOMPLETE;
1180                 /* just in case */
1181                 name_rmt->valuelen = 0;
1182                 name_rmt->valueblk = 0;
1183                 args->rmtblkno = 1;
1184                 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1185         }
1186         xfs_trans_log_buf(args->trans, bp,
1187              XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1188                                    xfs_attr_leaf_entsize(leaf, args->index)));
1189
1190         /*
1191          * Update the control info for this leaf node
1192          */
1193         if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1194                 /* both on-disk, don't endian-flip twice */
1195                 hdr->firstused = entry->nameidx;
1196         }
1197         ASSERT(be16_to_cpu(hdr->firstused) >=
1198                ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1199         tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1200                                         + sizeof(xfs_attr_leaf_hdr_t);
1201         map = &hdr->freemap[0];
1202         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1203                 if (be16_to_cpu(map->base) == tmp) {
1204                         be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1205                         be16_add_cpu(&map->size,
1206                                  -((int)sizeof(xfs_attr_leaf_entry_t)));
1207                 }
1208         }
1209         be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1210         xfs_trans_log_buf(args->trans, bp,
1211                 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1212         return(0);
1213 }
1214
1215 /*
1216  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1217  */
1218 STATIC void
1219 xfs_attr_leaf_compact(
1220         struct xfs_da_args      *args,
1221         struct xfs_buf          *bp)
1222 {
1223         xfs_attr_leafblock_t    *leaf_s, *leaf_d;
1224         xfs_attr_leaf_hdr_t     *hdr_s, *hdr_d;
1225         struct xfs_trans        *trans = args->trans;
1226         struct xfs_mount        *mp = trans->t_mountp;
1227         char                    *tmpbuffer;
1228
1229         trace_xfs_attr_leaf_compact(args);
1230
1231         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1232         ASSERT(tmpbuffer != NULL);
1233         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1234         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1235
1236         /*
1237          * Copy basic information
1238          */
1239         leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1240         leaf_d = bp->b_addr;
1241         hdr_s = &leaf_s->hdr;
1242         hdr_d = &leaf_d->hdr;
1243         hdr_d->info = hdr_s->info;      /* struct copy */
1244         hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1245         /* handle truncation gracefully */
1246         if (!hdr_d->firstused) {
1247                 hdr_d->firstused = cpu_to_be16(
1248                                 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1249         }
1250         hdr_d->usedbytes = 0;
1251         hdr_d->count = 0;
1252         hdr_d->holes = 0;
1253         hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1254         hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1255                                              sizeof(xfs_attr_leaf_hdr_t));
1256
1257         /*
1258          * Copy all entry's in the same (sorted) order,
1259          * but allocate name/value pairs packed and in sequence.
1260          */
1261         xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1262                                 be16_to_cpu(hdr_s->count), mp);
1263         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1264
1265         kmem_free(tmpbuffer);
1266 }
1267
1268 /*
1269  * Redistribute the attribute list entries between two leaf nodes,
1270  * taking into account the size of the new entry.
1271  *
1272  * NOTE: if new block is empty, then it will get the upper half of the
1273  * old block.  At present, all (one) callers pass in an empty second block.
1274  *
1275  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1276  * to match what it is doing in splitting the attribute leaf block.  Those
1277  * values are used in "atomic rename" operations on attributes.  Note that
1278  * the "new" and "old" values can end up in different blocks.
1279  */
1280 STATIC void
1281 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1282                                        xfs_da_state_blk_t *blk2)
1283 {
1284         xfs_da_args_t *args;
1285         xfs_da_state_blk_t *tmp_blk;
1286         xfs_attr_leafblock_t *leaf1, *leaf2;
1287         xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1288         int count, totallen, max, space, swap;
1289
1290         /*
1291          * Set up environment.
1292          */
1293         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1294         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1295         leaf1 = blk1->bp->b_addr;
1296         leaf2 = blk2->bp->b_addr;
1297         ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1298         ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1299         ASSERT(leaf2->hdr.count == 0);
1300         args = state->args;
1301
1302         trace_xfs_attr_leaf_rebalance(args);
1303
1304         /*
1305          * Check ordering of blocks, reverse if it makes things simpler.
1306          *
1307          * NOTE: Given that all (current) callers pass in an empty
1308          * second block, this code should never set "swap".
1309          */
1310         swap = 0;
1311         if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1312                 tmp_blk = blk1;
1313                 blk1 = blk2;
1314                 blk2 = tmp_blk;
1315                 leaf1 = blk1->bp->b_addr;
1316                 leaf2 = blk2->bp->b_addr;
1317                 swap = 1;
1318         }
1319         hdr1 = &leaf1->hdr;
1320         hdr2 = &leaf2->hdr;
1321
1322         /*
1323          * Examine entries until we reduce the absolute difference in
1324          * byte usage between the two blocks to a minimum.  Then get
1325          * the direction to copy and the number of elements to move.
1326          *
1327          * "inleaf" is true if the new entry should be inserted into blk1.
1328          * If "swap" is also true, then reverse the sense of "inleaf".
1329          */
1330         state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1331                                                             &count, &totallen);
1332         if (swap)
1333                 state->inleaf = !state->inleaf;
1334
1335         /*
1336          * Move any entries required from leaf to leaf:
1337          */
1338         if (count < be16_to_cpu(hdr1->count)) {
1339                 /*
1340                  * Figure the total bytes to be added to the destination leaf.
1341                  */
1342                 /* number entries being moved */
1343                 count = be16_to_cpu(hdr1->count) - count;
1344                 space  = be16_to_cpu(hdr1->usedbytes) - totallen;
1345                 space += count * sizeof(xfs_attr_leaf_entry_t);
1346
1347                 /*
1348                  * leaf2 is the destination, compact it if it looks tight.
1349                  */
1350                 max  = be16_to_cpu(hdr2->firstused)
1351                                                 - sizeof(xfs_attr_leaf_hdr_t);
1352                 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1353                 if (space > max)
1354                         xfs_attr_leaf_compact(args, blk2->bp);
1355
1356                 /*
1357                  * Move high entries from leaf1 to low end of leaf2.
1358                  */
1359                 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1360                                 leaf2, 0, count, state->mp);
1361
1362                 xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1363                 xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1364         } else if (count > be16_to_cpu(hdr1->count)) {
1365                 /*
1366                  * I assert that since all callers pass in an empty
1367                  * second buffer, this code should never execute.
1368                  */
1369                 ASSERT(0);
1370
1371                 /*
1372                  * Figure the total bytes to be added to the destination leaf.
1373                  */
1374                 /* number entries being moved */
1375                 count -= be16_to_cpu(hdr1->count);
1376                 space  = totallen - be16_to_cpu(hdr1->usedbytes);
1377                 space += count * sizeof(xfs_attr_leaf_entry_t);
1378
1379                 /*
1380                  * leaf1 is the destination, compact it if it looks tight.
1381                  */
1382                 max  = be16_to_cpu(hdr1->firstused)
1383                                                 - sizeof(xfs_attr_leaf_hdr_t);
1384                 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1385                 if (space > max)
1386                         xfs_attr_leaf_compact(args, blk1->bp);
1387
1388                 /*
1389                  * Move low entries from leaf2 to high end of leaf1.
1390                  */
1391                 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1392                                 be16_to_cpu(hdr1->count), count, state->mp);
1393
1394                 xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1395                 xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1396         }
1397
1398         /*
1399          * Copy out last hashval in each block for B-tree code.
1400          */
1401         blk1->hashval = be32_to_cpu(
1402                 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1403         blk2->hashval = be32_to_cpu(
1404                 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1405
1406         /*
1407          * Adjust the expected index for insertion.
1408          * NOTE: this code depends on the (current) situation that the
1409          * second block was originally empty.
1410          *
1411          * If the insertion point moved to the 2nd block, we must adjust
1412          * the index.  We must also track the entry just following the
1413          * new entry for use in an "atomic rename" operation, that entry
1414          * is always the "old" entry and the "new" entry is what we are
1415          * inserting.  The index/blkno fields refer to the "old" entry,
1416          * while the index2/blkno2 fields refer to the "new" entry.
1417          */
1418         if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1419                 ASSERT(state->inleaf == 0);
1420                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1421                 args->index = args->index2 = blk2->index;
1422                 args->blkno = args->blkno2 = blk2->blkno;
1423         } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1424                 if (state->inleaf) {
1425                         args->index = blk1->index;
1426                         args->blkno = blk1->blkno;
1427                         args->index2 = 0;
1428                         args->blkno2 = blk2->blkno;
1429                 } else {
1430                         /*
1431                          * On a double leaf split, the original attr location
1432                          * is already stored in blkno2/index2, so don't
1433                          * overwrite it overwise we corrupt the tree.
1434                          */
1435                         blk2->index = blk1->index
1436                                     - be16_to_cpu(leaf1->hdr.count);
1437                         args->index = blk2->index;
1438                         args->blkno = blk2->blkno;
1439                         if (!state->extravalid) {
1440                                 /*
1441                                  * set the new attr location to match the old
1442                                  * one and let the higher level split code
1443                                  * decide where in the leaf to place it.
1444                                  */
1445                                 args->index2 = blk2->index;
1446                                 args->blkno2 = blk2->blkno;
1447                         }
1448                 }
1449         } else {
1450                 ASSERT(state->inleaf == 1);
1451                 args->index = args->index2 = blk1->index;
1452                 args->blkno = args->blkno2 = blk1->blkno;
1453         }
1454 }
1455
1456 /*
1457  * Examine entries until we reduce the absolute difference in
1458  * byte usage between the two blocks to a minimum.
1459  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1460  * GROT: there will always be enough room in either block for a new entry.
1461  * GROT: Do a double-split for this case?
1462  */
1463 STATIC int
1464 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1465                                     xfs_da_state_blk_t *blk1,
1466                                     xfs_da_state_blk_t *blk2,
1467                                     int *countarg, int *usedbytesarg)
1468 {
1469         xfs_attr_leafblock_t *leaf1, *leaf2;
1470         xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1471         xfs_attr_leaf_entry_t *entry;
1472         int count, max, index, totallen, half;
1473         int lastdelta, foundit, tmp;
1474
1475         /*
1476          * Set up environment.
1477          */
1478         leaf1 = blk1->bp->b_addr;
1479         leaf2 = blk2->bp->b_addr;
1480         hdr1 = &leaf1->hdr;
1481         hdr2 = &leaf2->hdr;
1482         foundit = 0;
1483         totallen = 0;
1484
1485         /*
1486          * Examine entries until we reduce the absolute difference in
1487          * byte usage between the two blocks to a minimum.
1488          */
1489         max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1490         half  = (max+1) * sizeof(*entry);
1491         half += be16_to_cpu(hdr1->usedbytes) +
1492                 be16_to_cpu(hdr2->usedbytes) +
1493                 xfs_attr_leaf_newentsize(
1494                                 state->args->namelen,
1495                                 state->args->valuelen,
1496                                 state->blocksize, NULL);
1497         half /= 2;
1498         lastdelta = state->blocksize;
1499         entry = &leaf1->entries[0];
1500         for (count = index = 0; count < max; entry++, index++, count++) {
1501
1502 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1503                 /*
1504                  * The new entry is in the first block, account for it.
1505                  */
1506                 if (count == blk1->index) {
1507                         tmp = totallen + sizeof(*entry) +
1508                                 xfs_attr_leaf_newentsize(
1509                                                 state->args->namelen,
1510                                                 state->args->valuelen,
1511                                                 state->blocksize, NULL);
1512                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1513                                 break;
1514                         lastdelta = XFS_ATTR_ABS(half - tmp);
1515                         totallen = tmp;
1516                         foundit = 1;
1517                 }
1518
1519                 /*
1520                  * Wrap around into the second block if necessary.
1521                  */
1522                 if (count == be16_to_cpu(hdr1->count)) {
1523                         leaf1 = leaf2;
1524                         entry = &leaf1->entries[0];
1525                         index = 0;
1526                 }
1527
1528                 /*
1529                  * Figure out if next leaf entry would be too much.
1530                  */
1531                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1532                                                                         index);
1533                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1534                         break;
1535                 lastdelta = XFS_ATTR_ABS(half - tmp);
1536                 totallen = tmp;
1537 #undef XFS_ATTR_ABS
1538         }
1539
1540         /*
1541          * Calculate the number of usedbytes that will end up in lower block.
1542          * If new entry not in lower block, fix up the count.
1543          */
1544         totallen -= count * sizeof(*entry);
1545         if (foundit) {
1546                 totallen -= sizeof(*entry) +
1547                                 xfs_attr_leaf_newentsize(
1548                                                 state->args->namelen,
1549                                                 state->args->valuelen,
1550                                                 state->blocksize, NULL);
1551         }
1552
1553         *countarg = count;
1554         *usedbytesarg = totallen;
1555         return(foundit);
1556 }
1557
1558 /*========================================================================
1559  * Routines used for shrinking the Btree.
1560  *========================================================================*/
1561
1562 /*
1563  * Check a leaf block and its neighbors to see if the block should be
1564  * collapsed into one or the other neighbor.  Always keep the block
1565  * with the smaller block number.
1566  * If the current block is over 50% full, don't try to join it, return 0.
1567  * If the block is empty, fill in the state structure and return 2.
1568  * If it can be collapsed, fill in the state structure and return 1.
1569  * If nothing can be done, return 0.
1570  *
1571  * GROT: allow for INCOMPLETE entries in calculation.
1572  */
1573 int
1574 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1575 {
1576         xfs_attr_leafblock_t *leaf;
1577         xfs_da_state_blk_t *blk;
1578         xfs_da_blkinfo_t *info;
1579         int count, bytes, forward, error, retval, i;
1580         xfs_dablk_t blkno;
1581         struct xfs_buf *bp;
1582
1583         trace_xfs_attr_leaf_toosmall(state->args);
1584
1585         /*
1586          * Check for the degenerate case of the block being over 50% full.
1587          * If so, it's not worth even looking to see if we might be able
1588          * to coalesce with a sibling.
1589          */
1590         blk = &state->path.blk[ state->path.active-1 ];
1591         info = blk->bp->b_addr;
1592         ASSERT(info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1593         leaf = (xfs_attr_leafblock_t *)info;
1594         count = be16_to_cpu(leaf->hdr.count);
1595         bytes = sizeof(xfs_attr_leaf_hdr_t) +
1596                 count * sizeof(xfs_attr_leaf_entry_t) +
1597                 be16_to_cpu(leaf->hdr.usedbytes);
1598         if (bytes > (state->blocksize >> 1)) {
1599                 *action = 0;    /* blk over 50%, don't try to join */
1600                 return(0);
1601         }
1602
1603         /*
1604          * Check for the degenerate case of the block being empty.
1605          * If the block is empty, we'll simply delete it, no need to
1606          * coalesce it with a sibling block.  We choose (arbitrarily)
1607          * to merge with the forward block unless it is NULL.
1608          */
1609         if (count == 0) {
1610                 /*
1611                  * Make altpath point to the block we want to keep and
1612                  * path point to the block we want to drop (this one).
1613                  */
1614                 forward = (info->forw != 0);
1615                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1616                 error = xfs_da_path_shift(state, &state->altpath, forward,
1617                                                  0, &retval);
1618                 if (error)
1619                         return(error);
1620                 if (retval) {
1621                         *action = 0;
1622                 } else {
1623                         *action = 2;
1624                 }
1625                 return(0);
1626         }
1627
1628         /*
1629          * Examine each sibling block to see if we can coalesce with
1630          * at least 25% free space to spare.  We need to figure out
1631          * whether to merge with the forward or the backward block.
1632          * We prefer coalescing with the lower numbered sibling so as
1633          * to shrink an attribute list over time.
1634          */
1635         /* start with smaller blk num */
1636         forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1637         for (i = 0; i < 2; forward = !forward, i++) {
1638                 if (forward)
1639                         blkno = be32_to_cpu(info->forw);
1640                 else
1641                         blkno = be32_to_cpu(info->back);
1642                 if (blkno == 0)
1643                         continue;
1644                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1645                                         blkno, -1, &bp, XFS_ATTR_FORK);
1646                 if (error)
1647                         return(error);
1648                 ASSERT(bp != NULL);
1649
1650                 leaf = (xfs_attr_leafblock_t *)info;
1651                 count  = be16_to_cpu(leaf->hdr.count);
1652                 bytes  = state->blocksize - (state->blocksize>>2);
1653                 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1654                 leaf = bp->b_addr;
1655                 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1656                 count += be16_to_cpu(leaf->hdr.count);
1657                 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1658                 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1659                 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1660                 xfs_trans_brelse(state->args->trans, bp);
1661                 if (bytes >= 0)
1662                         break;  /* fits with at least 25% to spare */
1663         }
1664         if (i >= 2) {
1665                 *action = 0;
1666                 return(0);
1667         }
1668
1669         /*
1670          * Make altpath point to the block we want to keep (the lower
1671          * numbered block) and path point to the block we want to drop.
1672          */
1673         memcpy(&state->altpath, &state->path, sizeof(state->path));
1674         if (blkno < blk->blkno) {
1675                 error = xfs_da_path_shift(state, &state->altpath, forward,
1676                                                  0, &retval);
1677         } else {
1678                 error = xfs_da_path_shift(state, &state->path, forward,
1679                                                  0, &retval);
1680         }
1681         if (error)
1682                 return(error);
1683         if (retval) {
1684                 *action = 0;
1685         } else {
1686                 *action = 1;
1687         }
1688         return(0);
1689 }
1690
1691 /*
1692  * Remove a name from the leaf attribute list structure.
1693  *
1694  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1695  * If two leaves are 37% full, when combined they will leave 25% free.
1696  */
1697 int
1698 xfs_attr_leaf_remove(
1699         struct xfs_buf  *bp,
1700         xfs_da_args_t   *args)
1701 {
1702         xfs_attr_leafblock_t *leaf;
1703         xfs_attr_leaf_hdr_t *hdr;
1704         xfs_attr_leaf_map_t *map;
1705         xfs_attr_leaf_entry_t *entry;
1706         int before, after, smallest, entsize;
1707         int tablesize, tmp, i;
1708         xfs_mount_t *mp;
1709
1710         trace_xfs_attr_leaf_remove(args);
1711
1712         leaf = bp->b_addr;
1713         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1714         hdr = &leaf->hdr;
1715         mp = args->trans->t_mountp;
1716         ASSERT((be16_to_cpu(hdr->count) > 0)
1717                 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1718         ASSERT((args->index >= 0)
1719                 && (args->index < be16_to_cpu(hdr->count)));
1720         ASSERT(be16_to_cpu(hdr->firstused) >=
1721                ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1722         entry = &leaf->entries[args->index];
1723         ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1724         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1725
1726         /*
1727          * Scan through free region table:
1728          *    check for adjacency of free'd entry with an existing one,
1729          *    find smallest free region in case we need to replace it,
1730          *    adjust any map that borders the entry table,
1731          */
1732         tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1733                                         + sizeof(xfs_attr_leaf_hdr_t);
1734         map = &hdr->freemap[0];
1735         tmp = be16_to_cpu(map->size);
1736         before = after = -1;
1737         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1738         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1739         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1740                 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1741                 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1742                 if (be16_to_cpu(map->base) == tablesize) {
1743                         be16_add_cpu(&map->base,
1744                                  -((int)sizeof(xfs_attr_leaf_entry_t)));
1745                         be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1746                 }
1747
1748                 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1749                                 == be16_to_cpu(entry->nameidx)) {
1750                         before = i;
1751                 } else if (be16_to_cpu(map->base)
1752                         == (be16_to_cpu(entry->nameidx) + entsize)) {
1753                         after = i;
1754                 } else if (be16_to_cpu(map->size) < tmp) {
1755                         tmp = be16_to_cpu(map->size);
1756                         smallest = i;
1757                 }
1758         }
1759
1760         /*
1761          * Coalesce adjacent freemap regions,
1762          * or replace the smallest region.
1763          */
1764         if ((before >= 0) || (after >= 0)) {
1765                 if ((before >= 0) && (after >= 0)) {
1766                         map = &hdr->freemap[before];
1767                         be16_add_cpu(&map->size, entsize);
1768                         be16_add_cpu(&map->size,
1769                                  be16_to_cpu(hdr->freemap[after].size));
1770                         hdr->freemap[after].base = 0;
1771                         hdr->freemap[after].size = 0;
1772                 } else if (before >= 0) {
1773                         map = &hdr->freemap[before];
1774                         be16_add_cpu(&map->size, entsize);
1775                 } else {
1776                         map = &hdr->freemap[after];
1777                         /* both on-disk, don't endian flip twice */
1778                         map->base = entry->nameidx;
1779                         be16_add_cpu(&map->size, entsize);
1780                 }
1781         } else {
1782                 /*
1783                  * Replace smallest region (if it is smaller than free'd entry)
1784                  */
1785                 map = &hdr->freemap[smallest];
1786                 if (be16_to_cpu(map->size) < entsize) {
1787                         map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1788                         map->size = cpu_to_be16(entsize);
1789                 }
1790         }
1791
1792         /*
1793          * Did we remove the first entry?
1794          */
1795         if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1796                 smallest = 1;
1797         else
1798                 smallest = 0;
1799
1800         /*
1801          * Compress the remaining entries and zero out the removed stuff.
1802          */
1803         memset(xfs_attr_leaf_name(leaf, args->index), 0, entsize);
1804         be16_add_cpu(&hdr->usedbytes, -entsize);
1805         xfs_trans_log_buf(args->trans, bp,
1806              XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1807                                    entsize));
1808
1809         tmp = (be16_to_cpu(hdr->count) - args->index)
1810                                         * sizeof(xfs_attr_leaf_entry_t);
1811         memmove((char *)entry, (char *)(entry+1), tmp);
1812         be16_add_cpu(&hdr->count, -1);
1813         xfs_trans_log_buf(args->trans, bp,
1814             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1815         entry = &leaf->entries[be16_to_cpu(hdr->count)];
1816         memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1817
1818         /*
1819          * If we removed the first entry, re-find the first used byte
1820          * in the name area.  Note that if the entry was the "firstused",
1821          * then we don't have a "hole" in our block resulting from
1822          * removing the name.
1823          */
1824         if (smallest) {
1825                 tmp = XFS_LBSIZE(mp);
1826                 entry = &leaf->entries[0];
1827                 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1828                         ASSERT(be16_to_cpu(entry->nameidx) >=
1829                                be16_to_cpu(hdr->firstused));
1830                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1831
1832                         if (be16_to_cpu(entry->nameidx) < tmp)
1833                                 tmp = be16_to_cpu(entry->nameidx);
1834                 }
1835                 hdr->firstused = cpu_to_be16(tmp);
1836                 if (!hdr->firstused) {
1837                         hdr->firstused = cpu_to_be16(
1838                                         tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1839                 }
1840         } else {
1841                 hdr->holes = 1;         /* mark as needing compaction */
1842         }
1843         xfs_trans_log_buf(args->trans, bp,
1844                           XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1845
1846         /*
1847          * Check if leaf is less than 50% full, caller may want to
1848          * "join" the leaf with a sibling if so.
1849          */
1850         tmp  = sizeof(xfs_attr_leaf_hdr_t);
1851         tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1852         tmp += be16_to_cpu(leaf->hdr.usedbytes);
1853         return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1854 }
1855
1856 /*
1857  * Move all the attribute list entries from drop_leaf into save_leaf.
1858  */
1859 void
1860 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1861                                        xfs_da_state_blk_t *save_blk)
1862 {
1863         xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1864         xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1865         xfs_mount_t *mp;
1866         char *tmpbuffer;
1867
1868         trace_xfs_attr_leaf_unbalance(state->args);
1869
1870         /*
1871          * Set up environment.
1872          */
1873         mp = state->mp;
1874         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1875         ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1876         drop_leaf = drop_blk->bp->b_addr;
1877         save_leaf = save_blk->bp->b_addr;
1878         ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1879         ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1880         drop_hdr = &drop_leaf->hdr;
1881         save_hdr = &save_leaf->hdr;
1882
1883         /*
1884          * Save last hashval from dying block for later Btree fixup.
1885          */
1886         drop_blk->hashval = be32_to_cpu(
1887                 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1888
1889         /*
1890          * Check if we need a temp buffer, or can we do it in place.
1891          * Note that we don't check "leaf" for holes because we will
1892          * always be dropping it, toosmall() decided that for us already.
1893          */
1894         if (save_hdr->holes == 0) {
1895                 /*
1896                  * dest leaf has no holes, so we add there.  May need
1897                  * to make some room in the entry array.
1898                  */
1899                 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1900                         xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1901                              be16_to_cpu(drop_hdr->count), mp);
1902                 } else {
1903                         xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1904                                   be16_to_cpu(save_hdr->count),
1905                                   be16_to_cpu(drop_hdr->count), mp);
1906                 }
1907         } else {
1908                 /*
1909                  * Destination has holes, so we make a temporary copy
1910                  * of the leaf and add them both to that.
1911                  */
1912                 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1913                 ASSERT(tmpbuffer != NULL);
1914                 memset(tmpbuffer, 0, state->blocksize);
1915                 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1916                 tmp_hdr = &tmp_leaf->hdr;
1917                 tmp_hdr->info = save_hdr->info; /* struct copy */
1918                 tmp_hdr->count = 0;
1919                 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1920                 if (!tmp_hdr->firstused) {
1921                         tmp_hdr->firstused = cpu_to_be16(
1922                                 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1923                 }
1924                 tmp_hdr->usedbytes = 0;
1925                 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1926                         xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1927                                 be16_to_cpu(drop_hdr->count), mp);
1928                         xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1929                                   be16_to_cpu(tmp_leaf->hdr.count),
1930                                   be16_to_cpu(save_hdr->count), mp);
1931                 } else {
1932                         xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1933                                 be16_to_cpu(save_hdr->count), mp);
1934                         xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1935                                 be16_to_cpu(tmp_leaf->hdr.count),
1936                                 be16_to_cpu(drop_hdr->count), mp);
1937                 }
1938                 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1939                 kmem_free(tmpbuffer);
1940         }
1941
1942         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
1943                                            state->blocksize - 1);
1944
1945         /*
1946          * Copy out last hashval in each block for B-tree code.
1947          */
1948         save_blk->hashval = be32_to_cpu(
1949                 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1950 }
1951
1952 /*========================================================================
1953  * Routines used for finding things in the Btree.
1954  *========================================================================*/
1955
1956 /*
1957  * Look up a name in a leaf attribute list structure.
1958  * This is the internal routine, it uses the caller's buffer.
1959  *
1960  * Note that duplicate keys are allowed, but only check within the
1961  * current leaf node.  The Btree code must check in adjacent leaf nodes.
1962  *
1963  * Return in args->index the index into the entry[] array of either
1964  * the found entry, or where the entry should have been (insert before
1965  * that entry).
1966  *
1967  * Don't change the args->value unless we find the attribute.
1968  */
1969 int
1970 xfs_attr_leaf_lookup_int(
1971         struct xfs_buf  *bp,
1972         xfs_da_args_t   *args)
1973 {
1974         xfs_attr_leafblock_t *leaf;
1975         xfs_attr_leaf_entry_t *entry;
1976         xfs_attr_leaf_name_local_t *name_loc;
1977         xfs_attr_leaf_name_remote_t *name_rmt;
1978         int probe, span;
1979         xfs_dahash_t hashval;
1980
1981         trace_xfs_attr_leaf_lookup(args);
1982
1983         leaf = bp->b_addr;
1984         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1985         ASSERT(be16_to_cpu(leaf->hdr.count)
1986                                         < (XFS_LBSIZE(args->dp->i_mount)/8));
1987
1988         /*
1989          * Binary search.  (note: small blocks will skip this loop)
1990          */
1991         hashval = args->hashval;
1992         probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1993         for (entry = &leaf->entries[probe]; span > 4;
1994                    entry = &leaf->entries[probe]) {
1995                 span /= 2;
1996                 if (be32_to_cpu(entry->hashval) < hashval)
1997                         probe += span;
1998                 else if (be32_to_cpu(entry->hashval) > hashval)
1999                         probe -= span;
2000                 else
2001                         break;
2002         }
2003         ASSERT((probe >= 0) &&
2004                (!leaf->hdr.count
2005                || (probe < be16_to_cpu(leaf->hdr.count))));
2006         ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
2007
2008         /*
2009          * Since we may have duplicate hashval's, find the first matching
2010          * hashval in the leaf.
2011          */
2012         while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
2013                 entry--;
2014                 probe--;
2015         }
2016         while ((probe < be16_to_cpu(leaf->hdr.count)) &&
2017                (be32_to_cpu(entry->hashval) < hashval)) {
2018                 entry++;
2019                 probe++;
2020         }
2021         if ((probe == be16_to_cpu(leaf->hdr.count)) ||
2022             (be32_to_cpu(entry->hashval) != hashval)) {
2023                 args->index = probe;
2024                 return(XFS_ERROR(ENOATTR));
2025         }
2026
2027         /*
2028          * Duplicate keys may be present, so search all of them for a match.
2029          */
2030         for (  ; (probe < be16_to_cpu(leaf->hdr.count)) &&
2031                         (be32_to_cpu(entry->hashval) == hashval);
2032                         entry++, probe++) {
2033 /*
2034  * GROT: Add code to remove incomplete entries.
2035  */
2036                 /*
2037                  * If we are looking for INCOMPLETE entries, show only those.
2038                  * If we are looking for complete entries, show only those.
2039                  */
2040                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2041                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2042                         continue;
2043                 }
2044                 if (entry->flags & XFS_ATTR_LOCAL) {
2045                         name_loc = xfs_attr_leaf_name_local(leaf, probe);
2046                         if (name_loc->namelen != args->namelen)
2047                                 continue;
2048                         if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
2049                                 continue;
2050                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2051                                 continue;
2052                         args->index = probe;
2053                         return(XFS_ERROR(EEXIST));
2054                 } else {
2055                         name_rmt = xfs_attr_leaf_name_remote(leaf, probe);
2056                         if (name_rmt->namelen != args->namelen)
2057                                 continue;
2058                         if (memcmp(args->name, (char *)name_rmt->name,
2059                                              args->namelen) != 0)
2060                                 continue;
2061                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2062                                 continue;
2063                         args->index = probe;
2064                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2065                         args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2066                                                    be32_to_cpu(name_rmt->valuelen));
2067                         return(XFS_ERROR(EEXIST));
2068                 }
2069         }
2070         args->index = probe;
2071         return(XFS_ERROR(ENOATTR));
2072 }
2073
2074 /*
2075  * Get the value associated with an attribute name from a leaf attribute
2076  * list structure.
2077  */
2078 int
2079 xfs_attr_leaf_getvalue(
2080         struct xfs_buf  *bp,
2081         xfs_da_args_t   *args)
2082 {
2083         int valuelen;
2084         xfs_attr_leafblock_t *leaf;
2085         xfs_attr_leaf_entry_t *entry;
2086         xfs_attr_leaf_name_local_t *name_loc;
2087         xfs_attr_leaf_name_remote_t *name_rmt;
2088
2089         leaf = bp->b_addr;
2090         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2091         ASSERT(be16_to_cpu(leaf->hdr.count)
2092                                         < (XFS_LBSIZE(args->dp->i_mount)/8));
2093         ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2094
2095         entry = &leaf->entries[args->index];
2096         if (entry->flags & XFS_ATTR_LOCAL) {
2097                 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2098                 ASSERT(name_loc->namelen == args->namelen);
2099                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2100                 valuelen = be16_to_cpu(name_loc->valuelen);
2101                 if (args->flags & ATTR_KERNOVAL) {
2102                         args->valuelen = valuelen;
2103                         return(0);
2104                 }
2105                 if (args->valuelen < valuelen) {
2106                         args->valuelen = valuelen;
2107                         return(XFS_ERROR(ERANGE));
2108                 }
2109                 args->valuelen = valuelen;
2110                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2111         } else {
2112                 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2113                 ASSERT(name_rmt->namelen == args->namelen);
2114                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2115                 valuelen = be32_to_cpu(name_rmt->valuelen);
2116                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2117                 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2118                 if (args->flags & ATTR_KERNOVAL) {
2119                         args->valuelen = valuelen;
2120                         return(0);
2121                 }
2122                 if (args->valuelen < valuelen) {
2123                         args->valuelen = valuelen;
2124                         return(XFS_ERROR(ERANGE));
2125                 }
2126                 args->valuelen = valuelen;
2127         }
2128         return(0);
2129 }
2130
2131 /*========================================================================
2132  * Utility routines.
2133  *========================================================================*/
2134
2135 /*
2136  * Move the indicated entries from one leaf to another.
2137  * NOTE: this routine modifies both source and destination leaves.
2138  */
2139 /*ARGSUSED*/
2140 STATIC void
2141 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2142                         xfs_attr_leafblock_t *leaf_d, int start_d,
2143                         int count, xfs_mount_t *mp)
2144 {
2145         xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2146         xfs_attr_leaf_entry_t *entry_s, *entry_d;
2147         int desti, tmp, i;
2148
2149         /*
2150          * Check for nothing to do.
2151          */
2152         if (count == 0)
2153                 return;
2154
2155         /*
2156          * Set up environment.
2157          */
2158         ASSERT(leaf_s->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2159         ASSERT(leaf_d->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2160         hdr_s = &leaf_s->hdr;
2161         hdr_d = &leaf_d->hdr;
2162         ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2163                (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2164         ASSERT(be16_to_cpu(hdr_s->firstused) >=
2165                 ((be16_to_cpu(hdr_s->count)
2166                                         * sizeof(*entry_s))+sizeof(*hdr_s)));
2167         ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2168         ASSERT(be16_to_cpu(hdr_d->firstused) >=
2169                 ((be16_to_cpu(hdr_d->count)
2170                                         * sizeof(*entry_d))+sizeof(*hdr_d)));
2171
2172         ASSERT(start_s < be16_to_cpu(hdr_s->count));
2173         ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2174         ASSERT(count <= be16_to_cpu(hdr_s->count));
2175
2176         /*
2177          * Move the entries in the destination leaf up to make a hole?
2178          */
2179         if (start_d < be16_to_cpu(hdr_d->count)) {
2180                 tmp  = be16_to_cpu(hdr_d->count) - start_d;
2181                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2182                 entry_s = &leaf_d->entries[start_d];
2183                 entry_d = &leaf_d->entries[start_d + count];
2184                 memmove((char *)entry_d, (char *)entry_s, tmp);
2185         }
2186
2187         /*
2188          * Copy all entry's in the same (sorted) order,
2189          * but allocate attribute info packed and in sequence.
2190          */
2191         entry_s = &leaf_s->entries[start_s];
2192         entry_d = &leaf_d->entries[start_d];
2193         desti = start_d;
2194         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2195                 ASSERT(be16_to_cpu(entry_s->nameidx)
2196                                 >= be16_to_cpu(hdr_s->firstused));
2197                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2198 #ifdef GROT
2199                 /*
2200                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2201                  * may also need to change the insertion index.  Code turned
2202                  * off for 6.2, should be revisited later.
2203                  */
2204                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2205                         memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2206                         be16_add_cpu(&hdr_s->usedbytes, -tmp);
2207                         be16_add_cpu(&hdr_s->count, -1);
2208                         entry_d--;      /* to compensate for ++ in loop hdr */
2209                         desti--;
2210                         if ((start_s + i) < offset)
2211                                 result++;       /* insertion index adjustment */
2212                 } else {
2213 #endif /* GROT */
2214                         be16_add_cpu(&hdr_d->firstused, -tmp);
2215                         /* both on-disk, don't endian flip twice */
2216                         entry_d->hashval = entry_s->hashval;
2217                         /* both on-disk, don't endian flip twice */
2218                         entry_d->nameidx = hdr_d->firstused;
2219                         entry_d->flags = entry_s->flags;
2220                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2221                                                         <= XFS_LBSIZE(mp));
2222                         memmove(xfs_attr_leaf_name(leaf_d, desti),
2223                                 xfs_attr_leaf_name(leaf_s, start_s + i), tmp);
2224                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2225                                                         <= XFS_LBSIZE(mp));
2226                         memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2227                         be16_add_cpu(&hdr_s->usedbytes, -tmp);
2228                         be16_add_cpu(&hdr_d->usedbytes, tmp);
2229                         be16_add_cpu(&hdr_s->count, -1);
2230                         be16_add_cpu(&hdr_d->count, 1);
2231                         tmp = be16_to_cpu(hdr_d->count)
2232                                                 * sizeof(xfs_attr_leaf_entry_t)
2233                                                 + sizeof(xfs_attr_leaf_hdr_t);
2234                         ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2235 #ifdef GROT
2236                 }
2237 #endif /* GROT */
2238         }
2239
2240         /*
2241          * Zero out the entries we just copied.
2242          */
2243         if (start_s == be16_to_cpu(hdr_s->count)) {
2244                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2245                 entry_s = &leaf_s->entries[start_s];
2246                 ASSERT(((char *)entry_s + tmp) <=
2247                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2248                 memset((char *)entry_s, 0, tmp);
2249         } else {
2250                 /*
2251                  * Move the remaining entries down to fill the hole,
2252                  * then zero the entries at the top.
2253                  */
2254                 tmp  = be16_to_cpu(hdr_s->count) - count;
2255                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2256                 entry_s = &leaf_s->entries[start_s + count];
2257                 entry_d = &leaf_s->entries[start_s];
2258                 memmove((char *)entry_d, (char *)entry_s, tmp);
2259
2260                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2261                 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2262                 ASSERT(((char *)entry_s + tmp) <=
2263                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2264                 memset((char *)entry_s, 0, tmp);
2265         }
2266
2267         /*
2268          * Fill in the freemap information
2269          */
2270         hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2271         be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2272                         sizeof(xfs_attr_leaf_entry_t));
2273         hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2274                               - be16_to_cpu(hdr_d->freemap[0].base));
2275         hdr_d->freemap[1].base = 0;
2276         hdr_d->freemap[2].base = 0;
2277         hdr_d->freemap[1].size = 0;
2278         hdr_d->freemap[2].size = 0;
2279         hdr_s->holes = 1;       /* leaf may not be compact */
2280 }
2281
2282 /*
2283  * Compare two leaf blocks "order".
2284  * Return 0 unless leaf2 should go before leaf1.
2285  */
2286 int
2287 xfs_attr_leaf_order(
2288         struct xfs_buf  *leaf1_bp,
2289         struct xfs_buf  *leaf2_bp)
2290 {
2291         xfs_attr_leafblock_t *leaf1, *leaf2;
2292
2293         leaf1 = leaf1_bp->b_addr;
2294         leaf2 = leaf2_bp->b_addr;
2295         ASSERT((leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) &&
2296                (leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)));
2297         if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2298             (be16_to_cpu(leaf2->hdr.count) > 0) &&
2299             ((be32_to_cpu(leaf2->entries[0].hashval) <
2300               be32_to_cpu(leaf1->entries[0].hashval)) ||
2301              (be32_to_cpu(leaf2->entries[
2302                         be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2303               be32_to_cpu(leaf1->entries[
2304                         be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2305                 return(1);
2306         }
2307         return(0);
2308 }
2309
2310 /*
2311  * Pick up the last hashvalue from a leaf block.
2312  */
2313 xfs_dahash_t
2314 xfs_attr_leaf_lasthash(
2315         struct xfs_buf  *bp,
2316         int             *count)
2317 {
2318         xfs_attr_leafblock_t *leaf;
2319
2320         leaf = bp->b_addr;
2321         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2322         if (count)
2323                 *count = be16_to_cpu(leaf->hdr.count);
2324         if (!leaf->hdr.count)
2325                 return(0);
2326         return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2327 }
2328
2329 /*
2330  * Calculate the number of bytes used to store the indicated attribute
2331  * (whether local or remote only calculate bytes in this block).
2332  */
2333 STATIC int
2334 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2335 {
2336         xfs_attr_leaf_name_local_t *name_loc;
2337         xfs_attr_leaf_name_remote_t *name_rmt;
2338         int size;
2339
2340         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2341         if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2342                 name_loc = xfs_attr_leaf_name_local(leaf, index);
2343                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2344                                                    be16_to_cpu(name_loc->valuelen));
2345         } else {
2346                 name_rmt = xfs_attr_leaf_name_remote(leaf, index);
2347                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2348         }
2349         return(size);
2350 }
2351
2352 /*
2353  * Calculate the number of bytes that would be required to store the new
2354  * attribute (whether local or remote only calculate bytes in this block).
2355  * This routine decides as a side effect whether the attribute will be
2356  * a "local" or a "remote" attribute.
2357  */
2358 int
2359 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2360 {
2361         int size;
2362
2363         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2364         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2365                 if (local) {
2366                         *local = 1;
2367                 }
2368         } else {
2369                 size = xfs_attr_leaf_entsize_remote(namelen);
2370                 if (local) {
2371                         *local = 0;
2372                 }
2373         }
2374         return(size);
2375 }
2376
2377 /*
2378  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2379  */
2380 int
2381 xfs_attr_leaf_list_int(
2382         struct xfs_buf          *bp,
2383         xfs_attr_list_context_t *context)
2384 {
2385         attrlist_cursor_kern_t *cursor;
2386         xfs_attr_leafblock_t *leaf;
2387         xfs_attr_leaf_entry_t *entry;
2388         int retval, i;
2389
2390         ASSERT(bp != NULL);
2391         leaf = bp->b_addr;
2392         cursor = context->cursor;
2393         cursor->initted = 1;
2394
2395         trace_xfs_attr_list_leaf(context);
2396
2397         /*
2398          * Re-find our place in the leaf block if this is a new syscall.
2399          */
2400         if (context->resynch) {
2401                 entry = &leaf->entries[0];
2402                 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2403                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2404                                 if (cursor->offset == context->dupcnt) {
2405                                         context->dupcnt = 0;
2406                                         break;
2407                                 }
2408                                 context->dupcnt++;
2409                         } else if (be32_to_cpu(entry->hashval) >
2410                                         cursor->hashval) {
2411                                 context->dupcnt = 0;
2412                                 break;
2413                         }
2414                 }
2415                 if (i == be16_to_cpu(leaf->hdr.count)) {
2416                         trace_xfs_attr_list_notfound(context);
2417                         return(0);
2418                 }
2419         } else {
2420                 entry = &leaf->entries[0];
2421                 i = 0;
2422         }
2423         context->resynch = 0;
2424
2425         /*
2426          * We have found our place, start copying out the new attributes.
2427          */
2428         retval = 0;
2429         for (  ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2430                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2431                         cursor->hashval = be32_to_cpu(entry->hashval);
2432                         cursor->offset = 0;
2433                 }
2434
2435                 if (entry->flags & XFS_ATTR_INCOMPLETE)
2436                         continue;               /* skip incomplete entries */
2437
2438                 if (entry->flags & XFS_ATTR_LOCAL) {
2439                         xfs_attr_leaf_name_local_t *name_loc =
2440                                 xfs_attr_leaf_name_local(leaf, i);
2441
2442                         retval = context->put_listent(context,
2443                                                 entry->flags,
2444                                                 name_loc->nameval,
2445                                                 (int)name_loc->namelen,
2446                                                 be16_to_cpu(name_loc->valuelen),
2447                                                 &name_loc->nameval[name_loc->namelen]);
2448                         if (retval)
2449                                 return retval;
2450                 } else {
2451                         xfs_attr_leaf_name_remote_t *name_rmt =
2452                                 xfs_attr_leaf_name_remote(leaf, i);
2453
2454                         int valuelen = be32_to_cpu(name_rmt->valuelen);
2455
2456                         if (context->put_value) {
2457                                 xfs_da_args_t args;
2458
2459                                 memset((char *)&args, 0, sizeof(args));
2460                                 args.dp = context->dp;
2461                                 args.whichfork = XFS_ATTR_FORK;
2462                                 args.valuelen = valuelen;
2463                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2464                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2465                                 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2466                                 retval = xfs_attr_rmtval_get(&args);
2467                                 if (retval)
2468                                         return retval;
2469                                 retval = context->put_listent(context,
2470                                                 entry->flags,
2471                                                 name_rmt->name,
2472                                                 (int)name_rmt->namelen,
2473                                                 valuelen,
2474                                                 args.value);
2475                                 kmem_free(args.value);
2476                         } else {
2477                                 retval = context->put_listent(context,
2478                                                 entry->flags,
2479                                                 name_rmt->name,
2480                                                 (int)name_rmt->namelen,
2481                                                 valuelen,
2482                                                 NULL);
2483                         }
2484                         if (retval)
2485                                 return retval;
2486                 }
2487                 if (context->seen_enough)
2488                         break;
2489                 cursor->offset++;
2490         }
2491         trace_xfs_attr_list_leaf_end(context);
2492         return(retval);
2493 }
2494
2495
2496 /*========================================================================
2497  * Manage the INCOMPLETE flag in a leaf entry
2498  *========================================================================*/
2499
2500 /*
2501  * Clear the INCOMPLETE flag on an entry in a leaf block.
2502  */
2503 int
2504 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2505 {
2506         xfs_attr_leafblock_t *leaf;
2507         xfs_attr_leaf_entry_t *entry;
2508         xfs_attr_leaf_name_remote_t *name_rmt;
2509         struct xfs_buf *bp;
2510         int error;
2511 #ifdef DEBUG
2512         xfs_attr_leaf_name_local_t *name_loc;
2513         int namelen;
2514         char *name;
2515 #endif /* DEBUG */
2516
2517         trace_xfs_attr_leaf_clearflag(args);
2518         /*
2519          * Set up the operation.
2520          */
2521         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2522                                              XFS_ATTR_FORK);
2523         if (error) {
2524                 return(error);
2525         }
2526         ASSERT(bp != NULL);
2527
2528         leaf = bp->b_addr;
2529         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2530         ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2531         ASSERT(args->index >= 0);
2532         entry = &leaf->entries[ args->index ];
2533         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2534
2535 #ifdef DEBUG
2536         if (entry->flags & XFS_ATTR_LOCAL) {
2537                 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2538                 namelen = name_loc->namelen;
2539                 name = (char *)name_loc->nameval;
2540         } else {
2541                 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2542                 namelen = name_rmt->namelen;
2543                 name = (char *)name_rmt->name;
2544         }
2545         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2546         ASSERT(namelen == args->namelen);
2547         ASSERT(memcmp(name, args->name, namelen) == 0);
2548 #endif /* DEBUG */
2549
2550         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2551         xfs_trans_log_buf(args->trans, bp,
2552                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2553
2554         if (args->rmtblkno) {
2555                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2556                 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2557                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2558                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2559                 xfs_trans_log_buf(args->trans, bp,
2560                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2561         }
2562
2563         /*
2564          * Commit the flag value change and start the next trans in series.
2565          */
2566         return xfs_trans_roll(&args->trans, args->dp);
2567 }
2568
2569 /*
2570  * Set the INCOMPLETE flag on an entry in a leaf block.
2571  */
2572 int
2573 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2574 {
2575         xfs_attr_leafblock_t *leaf;
2576         xfs_attr_leaf_entry_t *entry;
2577         xfs_attr_leaf_name_remote_t *name_rmt;
2578         struct xfs_buf *bp;
2579         int error;
2580
2581         trace_xfs_attr_leaf_setflag(args);
2582
2583         /*
2584          * Set up the operation.
2585          */
2586         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2587                                              XFS_ATTR_FORK);
2588         if (error) {
2589                 return(error);
2590         }
2591         ASSERT(bp != NULL);
2592
2593         leaf = bp->b_addr;
2594         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2595         ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2596         ASSERT(args->index >= 0);
2597         entry = &leaf->entries[ args->index ];
2598
2599         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2600         entry->flags |= XFS_ATTR_INCOMPLETE;
2601         xfs_trans_log_buf(args->trans, bp,
2602                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2603         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2604                 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2605                 name_rmt->valueblk = 0;
2606                 name_rmt->valuelen = 0;
2607                 xfs_trans_log_buf(args->trans, bp,
2608                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2609         }
2610
2611         /*
2612          * Commit the flag value change and start the next trans in series.
2613          */
2614         return xfs_trans_roll(&args->trans, args->dp);
2615 }
2616
2617 /*
2618  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2619  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2620  * entry given by args->blkno2/index2.
2621  *
2622  * Note that they could be in different blocks, or in the same block.
2623  */
2624 int
2625 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2626 {
2627         xfs_attr_leafblock_t *leaf1, *leaf2;
2628         xfs_attr_leaf_entry_t *entry1, *entry2;
2629         xfs_attr_leaf_name_remote_t *name_rmt;
2630         struct xfs_buf *bp1, *bp2;
2631         int error;
2632 #ifdef DEBUG
2633         xfs_attr_leaf_name_local_t *name_loc;
2634         int namelen1, namelen2;
2635         char *name1, *name2;
2636 #endif /* DEBUG */
2637
2638         trace_xfs_attr_leaf_flipflags(args);
2639
2640         /*
2641          * Read the block containing the "old" attr
2642          */
2643         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2644                                              XFS_ATTR_FORK);
2645         if (error) {
2646                 return(error);
2647         }
2648         ASSERT(bp1 != NULL);
2649
2650         /*
2651          * Read the block containing the "new" attr, if it is different
2652          */
2653         if (args->blkno2 != args->blkno) {
2654                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2655                                         -1, &bp2, XFS_ATTR_FORK);
2656                 if (error) {
2657                         return(error);
2658                 }
2659                 ASSERT(bp2 != NULL);
2660         } else {
2661                 bp2 = bp1;
2662         }
2663
2664         leaf1 = bp1->b_addr;
2665         ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2666         ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2667         ASSERT(args->index >= 0);
2668         entry1 = &leaf1->entries[ args->index ];
2669
2670         leaf2 = bp2->b_addr;
2671         ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2672         ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2673         ASSERT(args->index2 >= 0);
2674         entry2 = &leaf2->entries[ args->index2 ];
2675
2676 #ifdef DEBUG
2677         if (entry1->flags & XFS_ATTR_LOCAL) {
2678                 name_loc = xfs_attr_leaf_name_local(leaf1, args->index);
2679                 namelen1 = name_loc->namelen;
2680                 name1 = (char *)name_loc->nameval;
2681         } else {
2682                 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2683                 namelen1 = name_rmt->namelen;
2684                 name1 = (char *)name_rmt->name;
2685         }
2686         if (entry2->flags & XFS_ATTR_LOCAL) {
2687                 name_loc = xfs_attr_leaf_name_local(leaf2, args->index2);
2688                 namelen2 = name_loc->namelen;
2689                 name2 = (char *)name_loc->nameval;
2690         } else {
2691                 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2692                 namelen2 = name_rmt->namelen;
2693                 name2 = (char *)name_rmt->name;
2694         }
2695         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2696         ASSERT(namelen1 == namelen2);
2697         ASSERT(memcmp(name1, name2, namelen1) == 0);
2698 #endif /* DEBUG */
2699
2700         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2701         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2702
2703         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2704         xfs_trans_log_buf(args->trans, bp1,
2705                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2706         if (args->rmtblkno) {
2707                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2708                 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2709                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2710                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2711                 xfs_trans_log_buf(args->trans, bp1,
2712                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2713         }
2714
2715         entry2->flags |= XFS_ATTR_INCOMPLETE;
2716         xfs_trans_log_buf(args->trans, bp2,
2717                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2718         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2719                 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2720                 name_rmt->valueblk = 0;
2721                 name_rmt->valuelen = 0;
2722                 xfs_trans_log_buf(args->trans, bp2,
2723                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2724         }
2725
2726         /*
2727          * Commit the flag value change and start the next trans in series.
2728          */
2729         error = xfs_trans_roll(&args->trans, args->dp);
2730
2731         return(error);
2732 }
2733
2734 /*========================================================================
2735  * Indiscriminately delete the entire attribute fork
2736  *========================================================================*/
2737
2738 /*
2739  * Recurse (gasp!) through the attribute nodes until we find leaves.
2740  * We're doing a depth-first traversal in order to invalidate everything.
2741  */
2742 int
2743 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2744 {
2745         xfs_da_blkinfo_t *info;
2746         xfs_daddr_t blkno;
2747         struct xfs_buf *bp;
2748         int error;
2749
2750         /*
2751          * Read block 0 to see what we have to work with.
2752          * We only get here if we have extents, since we remove
2753          * the extents in reverse order the extent containing
2754          * block 0 must still be there.
2755          */
2756         error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2757         if (error)
2758                 return(error);
2759         blkno = XFS_BUF_ADDR(bp);
2760
2761         /*
2762          * Invalidate the tree, even if the "tree" is only a single leaf block.
2763          * This is a depth-first traversal!
2764          */
2765         info = bp->b_addr;
2766         if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2767                 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2768         } else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2769                 error = xfs_attr_leaf_inactive(trans, dp, bp);
2770         } else {
2771                 error = XFS_ERROR(EIO);
2772                 xfs_trans_brelse(*trans, bp);
2773         }
2774         if (error)
2775                 return(error);
2776
2777         /*
2778          * Invalidate the incore copy of the root block.
2779          */
2780         error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2781         if (error)
2782                 return(error);
2783         xfs_trans_binval(*trans, bp);   /* remove from cache */
2784         /*
2785          * Commit the invalidate and start the next transaction.
2786          */
2787         error = xfs_trans_roll(trans, dp);
2788
2789         return (error);
2790 }
2791
2792 /*
2793  * Recurse (gasp!) through the attribute nodes until we find leaves.
2794  * We're doing a depth-first traversal in order to invalidate everything.
2795  */
2796 STATIC int
2797 xfs_attr_node_inactive(
2798         struct xfs_trans **trans,
2799         struct xfs_inode *dp,
2800         struct xfs_buf  *bp,
2801         int             level)
2802 {
2803         xfs_da_blkinfo_t *info;
2804         xfs_da_intnode_t *node;
2805         xfs_dablk_t child_fsb;
2806         xfs_daddr_t parent_blkno, child_blkno;
2807         int error, count, i;
2808         struct xfs_buf *child_bp;
2809
2810         /*
2811          * Since this code is recursive (gasp!) we must protect ourselves.
2812          */
2813         if (level > XFS_DA_NODE_MAXDEPTH) {
2814                 xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
2815                 return(XFS_ERROR(EIO));
2816         }
2817
2818         node = bp->b_addr;
2819         ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
2820         parent_blkno = XFS_BUF_ADDR(bp);        /* save for re-read later */
2821         count = be16_to_cpu(node->hdr.count);
2822         if (!count) {
2823                 xfs_trans_brelse(*trans, bp);
2824                 return(0);
2825         }
2826         child_fsb = be32_to_cpu(node->btree[0].before);
2827         xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
2828
2829         /*
2830          * If this is the node level just above the leaves, simply loop
2831          * over the leaves removing all of them.  If this is higher up
2832          * in the tree, recurse downward.
2833          */
2834         for (i = 0; i < count; i++) {
2835                 /*
2836                  * Read the subsidiary block to see what we have to work with.
2837                  * Don't do this in a transaction.  This is a depth-first
2838                  * traversal of the tree so we may deal with many blocks
2839                  * before we come back to this one.
2840                  */
2841                 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2842                                                 XFS_ATTR_FORK);
2843                 if (error)
2844                         return(error);
2845                 if (child_bp) {
2846                                                 /* save for re-read later */
2847                         child_blkno = XFS_BUF_ADDR(child_bp);
2848
2849                         /*
2850                          * Invalidate the subtree, however we have to.
2851                          */
2852                         info = child_bp->b_addr;
2853                         if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2854                                 error = xfs_attr_node_inactive(trans, dp,
2855                                                 child_bp, level+1);
2856                         } else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2857                                 error = xfs_attr_leaf_inactive(trans, dp,
2858                                                 child_bp);
2859                         } else {
2860                                 error = XFS_ERROR(EIO);
2861                                 xfs_trans_brelse(*trans, child_bp);
2862                         }
2863                         if (error)
2864                                 return(error);
2865
2866                         /*
2867                          * Remove the subsidiary block from the cache
2868                          * and from the log.
2869                          */
2870                         error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2871                                 &child_bp, XFS_ATTR_FORK);
2872                         if (error)
2873                                 return(error);
2874                         xfs_trans_binval(*trans, child_bp);
2875                 }
2876
2877                 /*
2878                  * If we're not done, re-read the parent to get the next
2879                  * child block number.
2880                  */
2881                 if ((i+1) < count) {
2882                         error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2883                                 &bp, XFS_ATTR_FORK);
2884                         if (error)
2885                                 return(error);
2886                         child_fsb = be32_to_cpu(node->btree[i+1].before);
2887                         xfs_trans_brelse(*trans, bp);
2888                 }
2889                 /*
2890                  * Atomically commit the whole invalidate stuff.
2891                  */
2892                 error = xfs_trans_roll(trans, dp);
2893                 if (error)
2894                         return (error);
2895         }
2896
2897         return(0);
2898 }
2899
2900 /*
2901  * Invalidate all of the "remote" value regions pointed to by a particular
2902  * leaf block.
2903  * Note that we must release the lock on the buffer so that we are not
2904  * caught holding something that the logging code wants to flush to disk.
2905  */
2906 STATIC int
2907 xfs_attr_leaf_inactive(
2908         struct xfs_trans **trans,
2909         struct xfs_inode *dp,
2910         struct xfs_buf  *bp)
2911 {
2912         xfs_attr_leafblock_t *leaf;
2913         xfs_attr_leaf_entry_t *entry;
2914         xfs_attr_leaf_name_remote_t *name_rmt;
2915         xfs_attr_inactive_list_t *list, *lp;
2916         int error, count, size, tmp, i;
2917
2918         leaf = bp->b_addr;
2919         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2920
2921         /*
2922          * Count the number of "remote" value extents.
2923          */
2924         count = 0;
2925         entry = &leaf->entries[0];
2926         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2927                 if (be16_to_cpu(entry->nameidx) &&
2928                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2929                         name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2930                         if (name_rmt->valueblk)
2931                                 count++;
2932                 }
2933         }
2934
2935         /*
2936          * If there are no "remote" values, we're done.
2937          */
2938         if (count == 0) {
2939                 xfs_trans_brelse(*trans, bp);
2940                 return(0);
2941         }
2942
2943         /*
2944          * Allocate storage for a list of all the "remote" value extents.
2945          */
2946         size = count * sizeof(xfs_attr_inactive_list_t);
2947         list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2948
2949         /*
2950          * Identify each of the "remote" value extents.
2951          */
2952         lp = list;
2953         entry = &leaf->entries[0];
2954         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2955                 if (be16_to_cpu(entry->nameidx) &&
2956                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2957                         name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2958                         if (name_rmt->valueblk) {
2959                                 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2960                                 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2961                                                     be32_to_cpu(name_rmt->valuelen));
2962                                 lp++;
2963                         }
2964                 }
2965         }
2966         xfs_trans_brelse(*trans, bp);   /* unlock for trans. in freextent() */
2967
2968         /*
2969          * Invalidate each of the "remote" value extents.
2970          */
2971         error = 0;
2972         for (lp = list, i = 0; i < count; i++, lp++) {
2973                 tmp = xfs_attr_leaf_freextent(trans, dp,
2974                                 lp->valueblk, lp->valuelen);
2975
2976                 if (error == 0)
2977                         error = tmp;    /* save only the 1st errno */
2978         }
2979
2980         kmem_free((xfs_caddr_t)list);
2981         return(error);
2982 }
2983
2984 /*
2985  * Look at all the extents for this logical region,
2986  * invalidate any buffers that are incore/in transactions.
2987  */
2988 STATIC int
2989 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2990                                     xfs_dablk_t blkno, int blkcnt)
2991 {
2992         xfs_bmbt_irec_t map;
2993         xfs_dablk_t tblkno;
2994         int tblkcnt, dblkcnt, nmap, error;
2995         xfs_daddr_t dblkno;
2996         xfs_buf_t *bp;
2997
2998         /*
2999          * Roll through the "value", invalidating the attribute value's
3000          * blocks.
3001          */
3002         tblkno = blkno;
3003         tblkcnt = blkcnt;
3004         while (tblkcnt > 0) {
3005                 /*
3006                  * Try to remember where we decided to put the value.
3007                  */
3008                 nmap = 1;
3009                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
3010                                        &map, &nmap, XFS_BMAPI_ATTRFORK);
3011                 if (error) {
3012                         return(error);
3013                 }
3014                 ASSERT(nmap == 1);
3015                 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3016
3017                 /*
3018                  * If it's a hole, these are already unmapped
3019                  * so there's nothing to invalidate.
3020                  */
3021                 if (map.br_startblock != HOLESTARTBLOCK) {
3022
3023                         dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3024                                                   map.br_startblock);
3025                         dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3026                                                 map.br_blockcount);
3027                         bp = xfs_trans_get_buf(*trans,
3028                                         dp->i_mount->m_ddev_targp,
3029                                         dblkno, dblkcnt, 0);
3030                         if (!bp)
3031                                 return ENOMEM;
3032                         xfs_trans_binval(*trans, bp);
3033                         /*
3034                          * Roll to next transaction.
3035                          */
3036                         error = xfs_trans_roll(trans, dp);
3037                         if (error)
3038                                 return (error);
3039                 }
3040
3041                 tblkno += map.br_blockcount;
3042                 tblkcnt -= map.br_blockcount;
3043         }
3044
3045         return(0);
3046 }