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