Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
[linux-2.6] / 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_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_alloc.h"
35 #include "xfs_btree.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_bmap.h"
42 #include "xfs_attr.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_error.h"
45
46 /*
47  * xfs_attr_leaf.c
48  *
49  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
50  */
51
52 /*========================================================================
53  * Function prototypes for the kernel.
54  *========================================================================*/
55
56 /*
57  * Routines used for growing the Btree.
58  */
59 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
60                                     xfs_dabuf_t **bpp);
61 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
62                                               int freemap_index);
63 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
64 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
65                                                    xfs_da_state_blk_t *blk1,
66                                                    xfs_da_state_blk_t *blk2);
67 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
68                                            xfs_da_state_blk_t *leaf_blk_1,
69                                            xfs_da_state_blk_t *leaf_blk_2,
70                                            int *number_entries_in_blk1,
71                                            int *number_usedbytes_in_blk1);
72
73 /*
74  * Routines used for shrinking the Btree.
75  */
76 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
77                                   xfs_dabuf_t *bp, int level);
78 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
79                                   xfs_dabuf_t *bp);
80 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
81                                    xfs_dablk_t blkno, int blkcnt);
82
83 /*
84  * Utility routines.
85  */
86 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
87                                          int src_start,
88                                          xfs_attr_leafblock_t *dst_leaf,
89                                          int dst_start, int move_count,
90                                          xfs_mount_t *mp);
91 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
92
93 /*========================================================================
94  * Namespace helper routines
95  *========================================================================*/
96
97 STATIC_INLINE attrnames_t *
98 xfs_attr_flags_namesp(int flags)
99 {
100         return ((flags & XFS_ATTR_SECURE) ? &attr_secure:
101                   ((flags & XFS_ATTR_ROOT) ? &attr_trusted : &attr_user));
102 }
103
104 /*
105  * If namespace bits don't match return 0.
106  * If all match then return 1.
107  */
108 STATIC_INLINE int
109 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
110 {
111         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
112 }
113
114 /*
115  * If namespace bits don't match and we don't have an override for it
116  * then return 0.
117  * If all match or are overridable then return 1.
118  */
119 STATIC_INLINE int
120 xfs_attr_namesp_match_overrides(int arg_flags, int ondisk_flags)
121 {
122         if (((arg_flags & ATTR_SECURE) == 0) !=
123             ((ondisk_flags & XFS_ATTR_SECURE) == 0) &&
124             !(arg_flags & ATTR_KERNORMALS))
125                 return 0;
126         if (((arg_flags & ATTR_ROOT) == 0) !=
127             ((ondisk_flags & XFS_ATTR_ROOT) == 0) &&
128             !(arg_flags & ATTR_KERNROOTLS))
129                 return 0;
130         return 1;
131 }
132
133
134 /*========================================================================
135  * External routines when attribute fork size < XFS_LITINO(mp).
136  *========================================================================*/
137
138 /*
139  * Query whether the requested number of additional bytes of extended
140  * attribute space will be able to fit inline.
141  * Returns zero if not, else the di_forkoff fork offset to be used in the
142  * literal area for attribute data once the new bytes have been added.
143  *
144  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
145  * special case for dev/uuid inodes, they have fixed size data forks.
146  */
147 int
148 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
149 {
150         int offset;
151         int minforkoff; /* lower limit on valid forkoff locations */
152         int maxforkoff; /* upper limit on valid forkoff locations */
153         int dsize;      
154         xfs_mount_t *mp = dp->i_mount;
155
156         offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
157
158         switch (dp->i_d.di_format) {
159         case XFS_DINODE_FMT_DEV:
160                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
161                 return (offset >= minforkoff) ? minforkoff : 0;
162         case XFS_DINODE_FMT_UUID:
163                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
164                 return (offset >= minforkoff) ? minforkoff : 0;
165         }
166
167         if (!(mp->m_flags & XFS_MOUNT_ATTR2)) {
168                 if (bytes <= XFS_IFORK_ASIZE(dp))
169                         return mp->m_attroffset >> 3;
170                 return 0;
171         }
172
173         dsize = dp->i_df.if_bytes;
174         
175         switch (dp->i_d.di_format) {
176         case XFS_DINODE_FMT_EXTENTS:
177                 /* 
178                  * If there is no attr fork and the data fork is extents, 
179                  * determine if creating the default attr fork will result 
180                  * in the extents form migrating to btree. If so, the 
181                  * minimum offset only needs to be the space required for 
182                  * the btree root.
183                  */ 
184                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes > mp->m_attroffset)
185                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
186                 break;
187                 
188         case XFS_DINODE_FMT_BTREE:
189                 /*
190                  * If have data btree then keep forkoff if we have one,
191                  * otherwise we are adding a new attr, so then we set 
192                  * minforkoff to where the btree root can finish so we have 
193                  * plenty of room for attrs
194                  */
195                 if (dp->i_d.di_forkoff) {
196                         if (offset < dp->i_d.di_forkoff) 
197                                 return 0;
198                         else 
199                                 return dp->i_d.di_forkoff;
200                 } else
201                         dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
202                 break;
203         }
204         
205         /* 
206          * A data fork btree root must have space for at least 
207          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
208          */
209         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
210         minforkoff = roundup(minforkoff, 8) >> 3;
211
212         /* attr fork btree root can have at least this many key/ptr pairs */
213         maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
214         maxforkoff = maxforkoff >> 3;   /* rounded down */
215
216         if (offset >= minforkoff && offset < maxforkoff)
217                 return offset;
218         if (offset >= maxforkoff)
219                 return maxforkoff;
220         return 0;
221 }
222
223 /*
224  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
225  */
226 STATIC void
227 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
228 {
229         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
230             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
231                 spin_lock(&mp->m_sb_lock);
232                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
233                         xfs_sb_version_addattr2(&mp->m_sb);
234                         spin_unlock(&mp->m_sb_lock);
235                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
236                 } else
237                         spin_unlock(&mp->m_sb_lock);
238         }
239 }
240
241 /*
242  * Create the initial contents of a shortform attribute list.
243  */
244 void
245 xfs_attr_shortform_create(xfs_da_args_t *args)
246 {
247         xfs_attr_sf_hdr_t *hdr;
248         xfs_inode_t *dp;
249         xfs_ifork_t *ifp;
250
251         dp = args->dp;
252         ASSERT(dp != NULL);
253         ifp = dp->i_afp;
254         ASSERT(ifp != NULL);
255         ASSERT(ifp->if_bytes == 0);
256         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
257                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
258                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
259                 ifp->if_flags |= XFS_IFINLINE;
260         } else {
261                 ASSERT(ifp->if_flags & XFS_IFINLINE);
262         }
263         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
264         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
265         hdr->count = 0;
266         hdr->totsize = cpu_to_be16(sizeof(*hdr));
267         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
268 }
269
270 /*
271  * Add a name/value pair to the shortform attribute list.
272  * Overflow from the inode has already been checked for.
273  */
274 void
275 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
276 {
277         xfs_attr_shortform_t *sf;
278         xfs_attr_sf_entry_t *sfe;
279         int i, offset, size;
280         xfs_mount_t *mp;
281         xfs_inode_t *dp;
282         xfs_ifork_t *ifp;
283
284         dp = args->dp;
285         mp = dp->i_mount;
286         dp->i_d.di_forkoff = forkoff;
287         dp->i_df.if_ext_max =
288                 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
289         dp->i_afp->if_ext_max =
290                 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
291
292         ifp = dp->i_afp;
293         ASSERT(ifp->if_flags & XFS_IFINLINE);
294         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
295         sfe = &sf->list[0];
296         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
297 #ifdef DEBUG
298                 if (sfe->namelen != args->namelen)
299                         continue;
300                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
301                         continue;
302                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
303                         continue;
304                 ASSERT(0);
305 #endif
306         }
307
308         offset = (char *)sfe - (char *)sf;
309         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
310         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
311         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
312         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
313
314         sfe->namelen = args->namelen;
315         sfe->valuelen = args->valuelen;
316         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
317         memcpy(sfe->nameval, args->name, args->namelen);
318         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
319         sf->hdr.count++;
320         be16_add_cpu(&sf->hdr.totsize, size);
321         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
322
323         xfs_sbversion_add_attr2(mp, args->trans);
324 }
325
326 /*
327  * Remove an attribute from the shortform attribute list structure.
328  */
329 int
330 xfs_attr_shortform_remove(xfs_da_args_t *args)
331 {
332         xfs_attr_shortform_t *sf;
333         xfs_attr_sf_entry_t *sfe;
334         int base, size=0, end, totsize, i;
335         xfs_mount_t *mp;
336         xfs_inode_t *dp;
337
338         dp = args->dp;
339         mp = dp->i_mount;
340         base = sizeof(xfs_attr_sf_hdr_t);
341         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
342         sfe = &sf->list[0];
343         end = sf->hdr.count;
344         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
345                                         base += size, i++) {
346                 size = XFS_ATTR_SF_ENTSIZE(sfe);
347                 if (sfe->namelen != args->namelen)
348                         continue;
349                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
350                         continue;
351                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
352                         continue;
353                 break;
354         }
355         if (i == end)
356                 return(XFS_ERROR(ENOATTR));
357
358         /*
359          * Fix up the attribute fork data, covering the hole
360          */
361         end = base + size;
362         totsize = be16_to_cpu(sf->hdr.totsize);
363         if (end != totsize)
364                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
365         sf->hdr.count--;
366         be16_add_cpu(&sf->hdr.totsize, -size);
367
368         /*
369          * Fix up the start offset of the attribute fork
370          */
371         totsize -= size;
372         if (totsize == sizeof(xfs_attr_sf_hdr_t) && !args->addname &&
373             (mp->m_flags & XFS_MOUNT_ATTR2) && 
374             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE)) {
375                 /*
376                  * Last attribute now removed, revert to original
377                  * inode format making all literal area available
378                  * to the data fork once more.
379                  */
380                 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
381                 dp->i_d.di_forkoff = 0;
382                 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
383                 ASSERT(dp->i_d.di_anextents == 0);
384                 ASSERT(dp->i_afp == NULL);
385                 dp->i_df.if_ext_max =
386                         XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
387                 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
388         } else {
389                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
390                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
391                 ASSERT(dp->i_d.di_forkoff);
392                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || args->addname ||
393                         !(mp->m_flags & XFS_MOUNT_ATTR2) ||
394                         dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
395                 dp->i_afp->if_ext_max =
396                         XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
397                 dp->i_df.if_ext_max =
398                         XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
399                 xfs_trans_log_inode(args->trans, dp,
400                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
401         }
402
403         xfs_sbversion_add_attr2(mp, args->trans);
404
405         return(0);
406 }
407
408 /*
409  * Look up a name in a shortform attribute list structure.
410  */
411 /*ARGSUSED*/
412 int
413 xfs_attr_shortform_lookup(xfs_da_args_t *args)
414 {
415         xfs_attr_shortform_t *sf;
416         xfs_attr_sf_entry_t *sfe;
417         int i;
418         xfs_ifork_t *ifp;
419
420         ifp = args->dp->i_afp;
421         ASSERT(ifp->if_flags & XFS_IFINLINE);
422         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
423         sfe = &sf->list[0];
424         for (i = 0; i < sf->hdr.count;
425                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
426                 if (sfe->namelen != args->namelen)
427                         continue;
428                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
429                         continue;
430                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
431                         continue;
432                 return(XFS_ERROR(EEXIST));
433         }
434         return(XFS_ERROR(ENOATTR));
435 }
436
437 /*
438  * Look up a name in a shortform attribute list structure.
439  */
440 /*ARGSUSED*/
441 int
442 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
443 {
444         xfs_attr_shortform_t *sf;
445         xfs_attr_sf_entry_t *sfe;
446         int i;
447
448         ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
449         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
450         sfe = &sf->list[0];
451         for (i = 0; i < sf->hdr.count;
452                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
453                 if (sfe->namelen != args->namelen)
454                         continue;
455                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
456                         continue;
457                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
458                         continue;
459                 if (args->flags & ATTR_KERNOVAL) {
460                         args->valuelen = sfe->valuelen;
461                         return(XFS_ERROR(EEXIST));
462                 }
463                 if (args->valuelen < sfe->valuelen) {
464                         args->valuelen = sfe->valuelen;
465                         return(XFS_ERROR(ERANGE));
466                 }
467                 args->valuelen = sfe->valuelen;
468                 memcpy(args->value, &sfe->nameval[args->namelen],
469                                                     args->valuelen);
470                 return(XFS_ERROR(EEXIST));
471         }
472         return(XFS_ERROR(ENOATTR));
473 }
474
475 /*
476  * Convert from using the shortform to the leaf.
477  */
478 int
479 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
480 {
481         xfs_inode_t *dp;
482         xfs_attr_shortform_t *sf;
483         xfs_attr_sf_entry_t *sfe;
484         xfs_da_args_t nargs;
485         char *tmpbuffer;
486         int error, i, size;
487         xfs_dablk_t blkno;
488         xfs_dabuf_t *bp;
489         xfs_ifork_t *ifp;
490
491         dp = args->dp;
492         ifp = dp->i_afp;
493         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
494         size = be16_to_cpu(sf->hdr.totsize);
495         tmpbuffer = kmem_alloc(size, KM_SLEEP);
496         ASSERT(tmpbuffer != NULL);
497         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
498         sf = (xfs_attr_shortform_t *)tmpbuffer;
499
500         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
501         bp = NULL;
502         error = xfs_da_grow_inode(args, &blkno);
503         if (error) {
504                 /*
505                  * If we hit an IO error middle of the transaction inside
506                  * grow_inode(), we may have inconsistent data. Bail out.
507                  */
508                 if (error == EIO)
509                         goto out;
510                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
511                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
512                 goto out;
513         }
514
515         ASSERT(blkno == 0);
516         error = xfs_attr_leaf_create(args, blkno, &bp);
517         if (error) {
518                 error = xfs_da_shrink_inode(args, 0, bp);
519                 bp = NULL;
520                 if (error)
521                         goto out;
522                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
523                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
524                 goto out;
525         }
526
527         memset((char *)&nargs, 0, sizeof(nargs));
528         nargs.dp = dp;
529         nargs.firstblock = args->firstblock;
530         nargs.flist = args->flist;
531         nargs.total = args->total;
532         nargs.whichfork = XFS_ATTR_FORK;
533         nargs.trans = args->trans;
534         nargs.oknoent = 1;
535
536         sfe = &sf->list[0];
537         for (i = 0; i < sf->hdr.count; i++) {
538                 nargs.name = (char *)sfe->nameval;
539                 nargs.namelen = sfe->namelen;
540                 nargs.value = (char *)&sfe->nameval[nargs.namelen];
541                 nargs.valuelen = sfe->valuelen;
542                 nargs.hashval = xfs_da_hashname((char *)sfe->nameval,
543                                                 sfe->namelen);
544                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
545                 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
546                 ASSERT(error == ENOATTR);
547                 error = xfs_attr_leaf_add(bp, &nargs);
548                 ASSERT(error != ENOSPC);
549                 if (error)
550                         goto out;
551                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
552         }
553         error = 0;
554
555 out:
556         if(bp)
557                 xfs_da_buf_done(bp);
558         kmem_free(tmpbuffer, size);
559         return(error);
560 }
561
562 STATIC int
563 xfs_attr_shortform_compare(const void *a, const void *b)
564 {
565         xfs_attr_sf_sort_t *sa, *sb;
566
567         sa = (xfs_attr_sf_sort_t *)a;
568         sb = (xfs_attr_sf_sort_t *)b;
569         if (sa->hash < sb->hash) {
570                 return(-1);
571         } else if (sa->hash > sb->hash) {
572                 return(1);
573         } else {
574                 return(sa->entno - sb->entno);
575         }
576 }
577
578
579 #define XFS_ISRESET_CURSOR(cursor) \
580         (!((cursor)->initted) && !((cursor)->hashval) && \
581          !((cursor)->blkno) && !((cursor)->offset))
582 /*
583  * Copy out entries of shortform attribute lists for attr_list().
584  * Shortform attribute lists are not stored in hashval sorted order.
585  * If the output buffer is not large enough to hold them all, then we
586  * we have to calculate each entries' hashvalue and sort them before
587  * we can begin returning them to the user.
588  */
589 /*ARGSUSED*/
590 int
591 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
592 {
593         attrlist_cursor_kern_t *cursor;
594         xfs_attr_sf_sort_t *sbuf, *sbp;
595         xfs_attr_shortform_t *sf;
596         xfs_attr_sf_entry_t *sfe;
597         xfs_inode_t *dp;
598         int sbsize, nsbuf, count, i;
599         int error;
600
601         ASSERT(context != NULL);
602         dp = context->dp;
603         ASSERT(dp != NULL);
604         ASSERT(dp->i_afp != NULL);
605         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
606         ASSERT(sf != NULL);
607         if (!sf->hdr.count)
608                 return(0);
609         cursor = context->cursor;
610         ASSERT(cursor != NULL);
611
612         xfs_attr_trace_l_c("sf start", context);
613
614         /*
615          * If the buffer is large enough and the cursor is at the start,
616          * do not bother with sorting since we will return everything in
617          * one buffer and another call using the cursor won't need to be
618          * made.
619          * Note the generous fudge factor of 16 overhead bytes per entry.
620          * If bufsize is zero then put_listent must be a search function
621          * and can just scan through what we have.
622          */
623         if (context->bufsize == 0 ||
624             (XFS_ISRESET_CURSOR(cursor) &&
625              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
626                 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
627                         attrnames_t     *namesp;
628
629                         if (!xfs_attr_namesp_match_overrides(context->flags, sfe->flags)) {
630                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
631                                 continue;
632                         }
633                         namesp = xfs_attr_flags_namesp(sfe->flags);
634                         error = context->put_listent(context,
635                                            namesp,
636                                            (char *)sfe->nameval,
637                                            (int)sfe->namelen,
638                                            (int)sfe->valuelen,
639                                            (char*)&sfe->nameval[sfe->namelen]);
640
641                         /*
642                          * Either search callback finished early or
643                          * didn't fit it all in the buffer after all.
644                          */
645                         if (context->seen_enough)
646                                 break;
647
648                         if (error)
649                                 return error;
650                         sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
651                 }
652                 xfs_attr_trace_l_c("sf big-gulp", context);
653                 return(0);
654         }
655
656         /* do no more for a search callback */
657         if (context->bufsize == 0)
658                 return 0;
659
660         /*
661          * It didn't all fit, so we have to sort everything on hashval.
662          */
663         sbsize = sf->hdr.count * sizeof(*sbuf);
664         sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP);
665
666         /*
667          * Scan the attribute list for the rest of the entries, storing
668          * the relevant info from only those that match into a buffer.
669          */
670         nsbuf = 0;
671         for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
672                 if (unlikely(
673                     ((char *)sfe < (char *)sf) ||
674                     ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
675                         XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
676                                              XFS_ERRLEVEL_LOW,
677                                              context->dp->i_mount, sfe);
678                         xfs_attr_trace_l_c("sf corrupted", context);
679                         kmem_free(sbuf, sbsize);
680                         return XFS_ERROR(EFSCORRUPTED);
681                 }
682                 if (!xfs_attr_namesp_match_overrides(context->flags, sfe->flags)) {
683                         sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
684                         continue;
685                 }
686                 sbp->entno = i;
687                 sbp->hash = xfs_da_hashname((char *)sfe->nameval, sfe->namelen);
688                 sbp->name = (char *)sfe->nameval;
689                 sbp->namelen = sfe->namelen;
690                 /* These are bytes, and both on-disk, don't endian-flip */
691                 sbp->valuelen = sfe->valuelen;
692                 sbp->flags = sfe->flags;
693                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
694                 sbp++;
695                 nsbuf++;
696         }
697
698         /*
699          * Sort the entries on hash then entno.
700          */
701         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
702
703         /*
704          * Re-find our place IN THE SORTED LIST.
705          */
706         count = 0;
707         cursor->initted = 1;
708         cursor->blkno = 0;
709         for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
710                 if (sbp->hash == cursor->hashval) {
711                         if (cursor->offset == count) {
712                                 break;
713                         }
714                         count++;
715                 } else if (sbp->hash > cursor->hashval) {
716                         break;
717                 }
718         }
719         if (i == nsbuf) {
720                 kmem_free(sbuf, sbsize);
721                 xfs_attr_trace_l_c("blk end", context);
722                 return(0);
723         }
724
725         /*
726          * Loop putting entries into the user buffer.
727          */
728         for ( ; i < nsbuf; i++, sbp++) {
729                 attrnames_t     *namesp;
730
731                 namesp = xfs_attr_flags_namesp(sbp->flags);
732
733                 if (cursor->hashval != sbp->hash) {
734                         cursor->hashval = sbp->hash;
735                         cursor->offset = 0;
736                 }
737                 error = context->put_listent(context,
738                                         namesp,
739                                         sbp->name,
740                                         sbp->namelen,
741                                         sbp->valuelen,
742                                         &sbp->name[sbp->namelen]);
743                 if (error)
744                         return error;
745                 if (context->seen_enough)
746                         break;
747                 cursor->offset++;
748         }
749
750         kmem_free(sbuf, sbsize);
751         xfs_attr_trace_l_c("sf E-O-F", context);
752         return(0);
753 }
754
755 /*
756  * Check a leaf attribute block to see if all the entries would fit into
757  * a shortform attribute list.
758  */
759 int
760 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
761 {
762         xfs_attr_leafblock_t *leaf;
763         xfs_attr_leaf_entry_t *entry;
764         xfs_attr_leaf_name_local_t *name_loc;
765         int bytes, i;
766
767         leaf = bp->data;
768         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
769
770         entry = &leaf->entries[0];
771         bytes = sizeof(struct xfs_attr_sf_hdr);
772         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
773                 if (entry->flags & XFS_ATTR_INCOMPLETE)
774                         continue;               /* don't copy partial entries */
775                 if (!(entry->flags & XFS_ATTR_LOCAL))
776                         return(0);
777                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
778                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
779                         return(0);
780                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
781                         return(0);
782                 bytes += sizeof(struct xfs_attr_sf_entry)-1
783                                 + name_loc->namelen
784                                 + be16_to_cpu(name_loc->valuelen);
785         }
786         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
787             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
788             (bytes == sizeof(struct xfs_attr_sf_hdr)))
789                 return(-1);
790         return(xfs_attr_shortform_bytesfit(dp, bytes));
791 }
792
793 /*
794  * Convert a leaf attribute list to shortform attribute list
795  */
796 int
797 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
798 {
799         xfs_attr_leafblock_t *leaf;
800         xfs_attr_leaf_entry_t *entry;
801         xfs_attr_leaf_name_local_t *name_loc;
802         xfs_da_args_t nargs;
803         xfs_inode_t *dp;
804         char *tmpbuffer;
805         int error, i;
806
807         dp = args->dp;
808         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
809         ASSERT(tmpbuffer != NULL);
810
811         ASSERT(bp != NULL);
812         memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
813         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
814         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
815         memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
816
817         /*
818          * Clean out the prior contents of the attribute list.
819          */
820         error = xfs_da_shrink_inode(args, 0, bp);
821         if (error)
822                 goto out;
823
824         if (forkoff == -1) {
825                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
826                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
827
828                 /*
829                  * Last attribute was removed, revert to original
830                  * inode format making all literal area available
831                  * to the data fork once more.
832                  */
833                 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
834                 dp->i_d.di_forkoff = 0;
835                 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
836                 ASSERT(dp->i_d.di_anextents == 0);
837                 ASSERT(dp->i_afp == NULL);
838                 dp->i_df.if_ext_max =
839                         XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
840                 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
841                 goto out;
842         }
843
844         xfs_attr_shortform_create(args);
845
846         /*
847          * Copy the attributes
848          */
849         memset((char *)&nargs, 0, sizeof(nargs));
850         nargs.dp = dp;
851         nargs.firstblock = args->firstblock;
852         nargs.flist = args->flist;
853         nargs.total = args->total;
854         nargs.whichfork = XFS_ATTR_FORK;
855         nargs.trans = args->trans;
856         nargs.oknoent = 1;
857         entry = &leaf->entries[0];
858         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
859                 if (entry->flags & XFS_ATTR_INCOMPLETE)
860                         continue;       /* don't copy partial entries */
861                 if (!entry->nameidx)
862                         continue;
863                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
864                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
865                 nargs.name = (char *)name_loc->nameval;
866                 nargs.namelen = name_loc->namelen;
867                 nargs.value = (char *)&name_loc->nameval[nargs.namelen];
868                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
869                 nargs.hashval = be32_to_cpu(entry->hashval);
870                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
871                 xfs_attr_shortform_add(&nargs, forkoff);
872         }
873         error = 0;
874
875 out:
876         kmem_free(tmpbuffer, XFS_LBSIZE(dp->i_mount));
877         return(error);
878 }
879
880 /*
881  * Convert from using a single leaf to a root node and a leaf.
882  */
883 int
884 xfs_attr_leaf_to_node(xfs_da_args_t *args)
885 {
886         xfs_attr_leafblock_t *leaf;
887         xfs_da_intnode_t *node;
888         xfs_inode_t *dp;
889         xfs_dabuf_t *bp1, *bp2;
890         xfs_dablk_t blkno;
891         int error;
892
893         dp = args->dp;
894         bp1 = bp2 = NULL;
895         error = xfs_da_grow_inode(args, &blkno);
896         if (error)
897                 goto out;
898         error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
899                                              XFS_ATTR_FORK);
900         if (error)
901                 goto out;
902         ASSERT(bp1 != NULL);
903         bp2 = NULL;
904         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
905                                             XFS_ATTR_FORK);
906         if (error)
907                 goto out;
908         ASSERT(bp2 != NULL);
909         memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
910         xfs_da_buf_done(bp1);
911         bp1 = NULL;
912         xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
913
914         /*
915          * Set up the new root node.
916          */
917         error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
918         if (error)
919                 goto out;
920         node = bp1->data;
921         leaf = bp2->data;
922         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
923         /* both on-disk, don't endian-flip twice */
924         node->btree[0].hashval =
925                 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
926         node->btree[0].before = cpu_to_be32(blkno);
927         node->hdr.count = cpu_to_be16(1);
928         xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
929         error = 0;
930 out:
931         if (bp1)
932                 xfs_da_buf_done(bp1);
933         if (bp2)
934                 xfs_da_buf_done(bp2);
935         return(error);
936 }
937
938
939 /*========================================================================
940  * Routines used for growing the Btree.
941  *========================================================================*/
942
943 /*
944  * Create the initial contents of a leaf attribute list
945  * or a leaf in a node attribute list.
946  */
947 STATIC int
948 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
949 {
950         xfs_attr_leafblock_t *leaf;
951         xfs_attr_leaf_hdr_t *hdr;
952         xfs_inode_t *dp;
953         xfs_dabuf_t *bp;
954         int error;
955
956         dp = args->dp;
957         ASSERT(dp != NULL);
958         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
959                                             XFS_ATTR_FORK);
960         if (error)
961                 return(error);
962         ASSERT(bp != NULL);
963         leaf = bp->data;
964         memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
965         hdr = &leaf->hdr;
966         hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
967         hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
968         if (!hdr->firstused) {
969                 hdr->firstused = cpu_to_be16(
970                         XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
971         }
972
973         hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
974         hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
975                                            sizeof(xfs_attr_leaf_hdr_t));
976
977         xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
978
979         *bpp = bp;
980         return(0);
981 }
982
983 /*
984  * Split the leaf node, rebalance, then add the new entry.
985  */
986 int
987 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
988                                    xfs_da_state_blk_t *newblk)
989 {
990         xfs_dablk_t blkno;
991         int error;
992
993         /*
994          * Allocate space for a new leaf node.
995          */
996         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
997         error = xfs_da_grow_inode(state->args, &blkno);
998         if (error)
999                 return(error);
1000         error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
1001         if (error)
1002                 return(error);
1003         newblk->blkno = blkno;
1004         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1005
1006         /*
1007          * Rebalance the entries across the two leaves.
1008          * NOTE: rebalance() currently depends on the 2nd block being empty.
1009          */
1010         xfs_attr_leaf_rebalance(state, oldblk, newblk);
1011         error = xfs_da_blk_link(state, oldblk, newblk);
1012         if (error)
1013                 return(error);
1014
1015         /*
1016          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1017          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1018          * "new" attrs info.  Will need the "old" info to remove it later.
1019          *
1020          * Insert the "new" entry in the correct block.
1021          */
1022         if (state->inleaf)
1023                 error = xfs_attr_leaf_add(oldblk->bp, state->args);
1024         else
1025                 error = xfs_attr_leaf_add(newblk->bp, state->args);
1026
1027         /*
1028          * Update last hashval in each block since we added the name.
1029          */
1030         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1031         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1032         return(error);
1033 }
1034
1035 /*
1036  * Add a name to the leaf attribute list structure.
1037  */
1038 int
1039 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
1040 {
1041         xfs_attr_leafblock_t *leaf;
1042         xfs_attr_leaf_hdr_t *hdr;
1043         xfs_attr_leaf_map_t *map;
1044         int tablesize, entsize, sum, tmp, i;
1045
1046         leaf = bp->data;
1047         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1048         ASSERT((args->index >= 0)
1049                 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1050         hdr = &leaf->hdr;
1051         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1052                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1053
1054         /*
1055          * Search through freemap for first-fit on new name length.
1056          * (may need to figure in size of entry struct too)
1057          */
1058         tablesize = (be16_to_cpu(hdr->count) + 1)
1059                                         * sizeof(xfs_attr_leaf_entry_t)
1060                                         + sizeof(xfs_attr_leaf_hdr_t);
1061         map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1062         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1063                 if (tablesize > be16_to_cpu(hdr->firstused)) {
1064                         sum += be16_to_cpu(map->size);
1065                         continue;
1066                 }
1067                 if (!map->size)
1068                         continue;       /* no space in this map */
1069                 tmp = entsize;
1070                 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1071                         tmp += sizeof(xfs_attr_leaf_entry_t);
1072                 if (be16_to_cpu(map->size) >= tmp) {
1073                         tmp = xfs_attr_leaf_add_work(bp, args, i);
1074                         return(tmp);
1075                 }
1076                 sum += be16_to_cpu(map->size);
1077         }
1078
1079         /*
1080          * If there are no holes in the address space of the block,
1081          * and we don't have enough freespace, then compaction will do us
1082          * no good and we should just give up.
1083          */
1084         if (!hdr->holes && (sum < entsize))
1085                 return(XFS_ERROR(ENOSPC));
1086
1087         /*
1088          * Compact the entries to coalesce free space.
1089          * This may change the hdr->count via dropping INCOMPLETE entries.
1090          */
1091         xfs_attr_leaf_compact(args->trans, bp);
1092
1093         /*
1094          * After compaction, the block is guaranteed to have only one
1095          * free region, in freemap[0].  If it is not big enough, give up.
1096          */
1097         if (be16_to_cpu(hdr->freemap[0].size)
1098                                 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1099                 return(XFS_ERROR(ENOSPC));
1100
1101         return(xfs_attr_leaf_add_work(bp, args, 0));
1102 }
1103
1104 /*
1105  * Add a name to a leaf attribute list structure.
1106  */
1107 STATIC int
1108 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1109 {
1110         xfs_attr_leafblock_t *leaf;
1111         xfs_attr_leaf_hdr_t *hdr;
1112         xfs_attr_leaf_entry_t *entry;
1113         xfs_attr_leaf_name_local_t *name_loc;
1114         xfs_attr_leaf_name_remote_t *name_rmt;
1115         xfs_attr_leaf_map_t *map;
1116         xfs_mount_t *mp;
1117         int tmp, i;
1118
1119         leaf = bp->data;
1120         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1121         hdr = &leaf->hdr;
1122         ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1123         ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1124
1125         /*
1126          * Force open some space in the entry array and fill it in.
1127          */
1128         entry = &leaf->entries[args->index];
1129         if (args->index < be16_to_cpu(hdr->count)) {
1130                 tmp  = be16_to_cpu(hdr->count) - args->index;
1131                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1132                 memmove((char *)(entry+1), (char *)entry, tmp);
1133                 xfs_da_log_buf(args->trans, bp,
1134                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1135         }
1136         be16_add_cpu(&hdr->count, 1);
1137
1138         /*
1139          * Allocate space for the new string (at the end of the run).
1140          */
1141         map = &hdr->freemap[mapindex];
1142         mp = args->trans->t_mountp;
1143         ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1144         ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1145         ASSERT(be16_to_cpu(map->size) >=
1146                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1147                                          mp->m_sb.sb_blocksize, NULL));
1148         ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1149         ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1150         be16_add_cpu(&map->size,
1151                 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1152                                           mp->m_sb.sb_blocksize, &tmp));
1153         entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1154                                      be16_to_cpu(map->size));
1155         entry->hashval = cpu_to_be32(args->hashval);
1156         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1157         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1158         if (args->rename) {
1159                 entry->flags |= XFS_ATTR_INCOMPLETE;
1160                 if ((args->blkno2 == args->blkno) &&
1161                     (args->index2 <= args->index)) {
1162                         args->index2++;
1163                 }
1164         }
1165         xfs_da_log_buf(args->trans, bp,
1166                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1167         ASSERT((args->index == 0) ||
1168                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1169         ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1170                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1171
1172         /*
1173          * Copy the attribute name and value into the new space.
1174          *
1175          * For "remote" attribute values, simply note that we need to
1176          * allocate space for the "remote" value.  We can't actually
1177          * allocate the extents in this transaction, and we can't decide
1178          * which blocks they should be as we might allocate more blocks
1179          * as part of this transaction (a split operation for example).
1180          */
1181         if (entry->flags & XFS_ATTR_LOCAL) {
1182                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
1183                 name_loc->namelen = args->namelen;
1184                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1185                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1186                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1187                                    be16_to_cpu(name_loc->valuelen));
1188         } else {
1189                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
1190                 name_rmt->namelen = args->namelen;
1191                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1192                 entry->flags |= XFS_ATTR_INCOMPLETE;
1193                 /* just in case */
1194                 name_rmt->valuelen = 0;
1195                 name_rmt->valueblk = 0;
1196                 args->rmtblkno = 1;
1197                 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1198         }
1199         xfs_da_log_buf(args->trans, bp,
1200              XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1201                                    xfs_attr_leaf_entsize(leaf, args->index)));
1202
1203         /*
1204          * Update the control info for this leaf node
1205          */
1206         if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1207                 /* both on-disk, don't endian-flip twice */
1208                 hdr->firstused = entry->nameidx;
1209         }
1210         ASSERT(be16_to_cpu(hdr->firstused) >=
1211                ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1212         tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1213                                         + sizeof(xfs_attr_leaf_hdr_t);
1214         map = &hdr->freemap[0];
1215         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1216                 if (be16_to_cpu(map->base) == tmp) {
1217                         be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1218                         be16_add_cpu(&map->size,
1219                                  -((int)sizeof(xfs_attr_leaf_entry_t)));
1220                 }
1221         }
1222         be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1223         xfs_da_log_buf(args->trans, bp,
1224                 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1225         return(0);
1226 }
1227
1228 /*
1229  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1230  */
1231 STATIC void
1232 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1233 {
1234         xfs_attr_leafblock_t *leaf_s, *leaf_d;
1235         xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1236         xfs_mount_t *mp;
1237         char *tmpbuffer;
1238
1239         mp = trans->t_mountp;
1240         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1241         ASSERT(tmpbuffer != NULL);
1242         memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1243         memset(bp->data, 0, XFS_LBSIZE(mp));
1244
1245         /*
1246          * Copy basic information
1247          */
1248         leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1249         leaf_d = bp->data;
1250         hdr_s = &leaf_s->hdr;
1251         hdr_d = &leaf_d->hdr;
1252         hdr_d->info = hdr_s->info;      /* struct copy */
1253         hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1254         /* handle truncation gracefully */
1255         if (!hdr_d->firstused) {
1256                 hdr_d->firstused = cpu_to_be16(
1257                                 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1258         }
1259         hdr_d->usedbytes = 0;
1260         hdr_d->count = 0;
1261         hdr_d->holes = 0;
1262         hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1263         hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1264                                              sizeof(xfs_attr_leaf_hdr_t));
1265
1266         /*
1267          * Copy all entry's in the same (sorted) order,
1268          * but allocate name/value pairs packed and in sequence.
1269          */
1270         xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1271                                 be16_to_cpu(hdr_s->count), mp);
1272         xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1273
1274         kmem_free(tmpbuffer, XFS_LBSIZE(mp));
1275 }
1276
1277 /*
1278  * Redistribute the attribute list entries between two leaf nodes,
1279  * taking into account the size of the new entry.
1280  *
1281  * NOTE: if new block is empty, then it will get the upper half of the
1282  * old block.  At present, all (one) callers pass in an empty second block.
1283  *
1284  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1285  * to match what it is doing in splitting the attribute leaf block.  Those
1286  * values are used in "atomic rename" operations on attributes.  Note that
1287  * the "new" and "old" values can end up in different blocks.
1288  */
1289 STATIC void
1290 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1291                                        xfs_da_state_blk_t *blk2)
1292 {
1293         xfs_da_args_t *args;
1294         xfs_da_state_blk_t *tmp_blk;
1295         xfs_attr_leafblock_t *leaf1, *leaf2;
1296         xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1297         int count, totallen, max, space, swap;
1298
1299         /*
1300          * Set up environment.
1301          */
1302         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1303         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1304         leaf1 = blk1->bp->data;
1305         leaf2 = blk2->bp->data;
1306         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1307         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1308         args = state->args;
1309
1310         /*
1311          * Check ordering of blocks, reverse if it makes things simpler.
1312          *
1313          * NOTE: Given that all (current) callers pass in an empty
1314          * second block, this code should never set "swap".
1315          */
1316         swap = 0;
1317         if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1318                 tmp_blk = blk1;
1319                 blk1 = blk2;
1320                 blk2 = tmp_blk;
1321                 leaf1 = blk1->bp->data;
1322                 leaf2 = blk2->bp->data;
1323                 swap = 1;
1324         }
1325         hdr1 = &leaf1->hdr;
1326         hdr2 = &leaf2->hdr;
1327
1328         /*
1329          * Examine entries until we reduce the absolute difference in
1330          * byte usage between the two blocks to a minimum.  Then get
1331          * the direction to copy and the number of elements to move.
1332          *
1333          * "inleaf" is true if the new entry should be inserted into blk1.
1334          * If "swap" is also true, then reverse the sense of "inleaf".
1335          */
1336         state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1337                                                             &count, &totallen);
1338         if (swap)
1339                 state->inleaf = !state->inleaf;
1340
1341         /*
1342          * Move any entries required from leaf to leaf:
1343          */
1344         if (count < be16_to_cpu(hdr1->count)) {
1345                 /*
1346                  * Figure the total bytes to be added to the destination leaf.
1347                  */
1348                 /* number entries being moved */
1349                 count = be16_to_cpu(hdr1->count) - count;
1350                 space  = be16_to_cpu(hdr1->usedbytes) - totallen;
1351                 space += count * sizeof(xfs_attr_leaf_entry_t);
1352
1353                 /*
1354                  * leaf2 is the destination, compact it if it looks tight.
1355                  */
1356                 max  = be16_to_cpu(hdr2->firstused)
1357                                                 - sizeof(xfs_attr_leaf_hdr_t);
1358                 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1359                 if (space > max) {
1360                         xfs_attr_leaf_compact(args->trans, blk2->bp);
1361                 }
1362
1363                 /*
1364                  * Move high entries from leaf1 to low end of leaf2.
1365                  */
1366                 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1367                                 leaf2, 0, count, state->mp);
1368
1369                 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1370                 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1371         } else if (count > be16_to_cpu(hdr1->count)) {
1372                 /*
1373                  * I assert that since all callers pass in an empty
1374                  * second buffer, this code should never execute.
1375                  */
1376
1377                 /*
1378                  * Figure the total bytes to be added to the destination leaf.
1379                  */
1380                 /* number entries being moved */
1381                 count -= be16_to_cpu(hdr1->count);
1382                 space  = totallen - be16_to_cpu(hdr1->usedbytes);
1383                 space += count * sizeof(xfs_attr_leaf_entry_t);
1384
1385                 /*
1386                  * leaf1 is the destination, compact it if it looks tight.
1387                  */
1388                 max  = be16_to_cpu(hdr1->firstused)
1389                                                 - sizeof(xfs_attr_leaf_hdr_t);
1390                 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1391                 if (space > max) {
1392                         xfs_attr_leaf_compact(args->trans, blk1->bp);
1393                 }
1394
1395                 /*
1396                  * Move low entries from leaf2 to high end of leaf1.
1397                  */
1398                 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1399                                 be16_to_cpu(hdr1->count), count, state->mp);
1400
1401                 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1402                 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1403         }
1404
1405         /*
1406          * Copy out last hashval in each block for B-tree code.
1407          */
1408         blk1->hashval = be32_to_cpu(
1409                 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1410         blk2->hashval = be32_to_cpu(
1411                 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1412
1413         /*
1414          * Adjust the expected index for insertion.
1415          * NOTE: this code depends on the (current) situation that the
1416          * second block was originally empty.
1417          *
1418          * If the insertion point moved to the 2nd block, we must adjust
1419          * the index.  We must also track the entry just following the
1420          * new entry for use in an "atomic rename" operation, that entry
1421          * is always the "old" entry and the "new" entry is what we are
1422          * inserting.  The index/blkno fields refer to the "old" entry,
1423          * while the index2/blkno2 fields refer to the "new" entry.
1424          */
1425         if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1426                 ASSERT(state->inleaf == 0);
1427                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1428                 args->index = args->index2 = blk2->index;
1429                 args->blkno = args->blkno2 = blk2->blkno;
1430         } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1431                 if (state->inleaf) {
1432                         args->index = blk1->index;
1433                         args->blkno = blk1->blkno;
1434                         args->index2 = 0;
1435                         args->blkno2 = blk2->blkno;
1436                 } else {
1437                         blk2->index = blk1->index
1438                                     - be16_to_cpu(leaf1->hdr.count);
1439                         args->index = args->index2 = blk2->index;
1440                         args->blkno = args->blkno2 = blk2->blkno;
1441                 }
1442         } else {
1443                 ASSERT(state->inleaf == 1);
1444                 args->index = args->index2 = blk1->index;
1445                 args->blkno = args->blkno2 = blk1->blkno;
1446         }
1447 }
1448
1449 /*
1450  * Examine entries until we reduce the absolute difference in
1451  * byte usage between the two blocks to a minimum.
1452  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1453  * GROT: there will always be enough room in either block for a new entry.
1454  * GROT: Do a double-split for this case?
1455  */
1456 STATIC int
1457 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1458                                     xfs_da_state_blk_t *blk1,
1459                                     xfs_da_state_blk_t *blk2,
1460                                     int *countarg, int *usedbytesarg)
1461 {
1462         xfs_attr_leafblock_t *leaf1, *leaf2;
1463         xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1464         xfs_attr_leaf_entry_t *entry;
1465         int count, max, index, totallen, half;
1466         int lastdelta, foundit, tmp;
1467
1468         /*
1469          * Set up environment.
1470          */
1471         leaf1 = blk1->bp->data;
1472         leaf2 = blk2->bp->data;
1473         hdr1 = &leaf1->hdr;
1474         hdr2 = &leaf2->hdr;
1475         foundit = 0;
1476         totallen = 0;
1477
1478         /*
1479          * Examine entries until we reduce the absolute difference in
1480          * byte usage between the two blocks to a minimum.
1481          */
1482         max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1483         half  = (max+1) * sizeof(*entry);
1484         half += be16_to_cpu(hdr1->usedbytes) +
1485                 be16_to_cpu(hdr2->usedbytes) +
1486                 xfs_attr_leaf_newentsize(
1487                                 state->args->namelen,
1488                                 state->args->valuelen,
1489                                 state->blocksize, NULL);
1490         half /= 2;
1491         lastdelta = state->blocksize;
1492         entry = &leaf1->entries[0];
1493         for (count = index = 0; count < max; entry++, index++, count++) {
1494
1495 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1496                 /*
1497                  * The new entry is in the first block, account for it.
1498                  */
1499                 if (count == blk1->index) {
1500                         tmp = totallen + sizeof(*entry) +
1501                                 xfs_attr_leaf_newentsize(
1502                                                 state->args->namelen,
1503                                                 state->args->valuelen,
1504                                                 state->blocksize, NULL);
1505                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1506                                 break;
1507                         lastdelta = XFS_ATTR_ABS(half - tmp);
1508                         totallen = tmp;
1509                         foundit = 1;
1510                 }
1511
1512                 /*
1513                  * Wrap around into the second block if necessary.
1514                  */
1515                 if (count == be16_to_cpu(hdr1->count)) {
1516                         leaf1 = leaf2;
1517                         entry = &leaf1->entries[0];
1518                         index = 0;
1519                 }
1520
1521                 /*
1522                  * Figure out if next leaf entry would be too much.
1523                  */
1524                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1525                                                                         index);
1526                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1527                         break;
1528                 lastdelta = XFS_ATTR_ABS(half - tmp);
1529                 totallen = tmp;
1530 #undef XFS_ATTR_ABS
1531         }
1532
1533         /*
1534          * Calculate the number of usedbytes that will end up in lower block.
1535          * If new entry not in lower block, fix up the count.
1536          */
1537         totallen -= count * sizeof(*entry);
1538         if (foundit) {
1539                 totallen -= sizeof(*entry) +
1540                                 xfs_attr_leaf_newentsize(
1541                                                 state->args->namelen,
1542                                                 state->args->valuelen,
1543                                                 state->blocksize, NULL);
1544         }
1545
1546         *countarg = count;
1547         *usedbytesarg = totallen;
1548         return(foundit);
1549 }
1550
1551 /*========================================================================
1552  * Routines used for shrinking the Btree.
1553  *========================================================================*/
1554
1555 /*
1556  * Check a leaf block and its neighbors to see if the block should be
1557  * collapsed into one or the other neighbor.  Always keep the block
1558  * with the smaller block number.
1559  * If the current block is over 50% full, don't try to join it, return 0.
1560  * If the block is empty, fill in the state structure and return 2.
1561  * If it can be collapsed, fill in the state structure and return 1.
1562  * If nothing can be done, return 0.
1563  *
1564  * GROT: allow for INCOMPLETE entries in calculation.
1565  */
1566 int
1567 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1568 {
1569         xfs_attr_leafblock_t *leaf;
1570         xfs_da_state_blk_t *blk;
1571         xfs_da_blkinfo_t *info;
1572         int count, bytes, forward, error, retval, i;
1573         xfs_dablk_t blkno;
1574         xfs_dabuf_t *bp;
1575
1576         /*
1577          * Check for the degenerate case of the block being over 50% full.
1578          * If so, it's not worth even looking to see if we might be able
1579          * to coalesce with a sibling.
1580          */
1581         blk = &state->path.blk[ state->path.active-1 ];
1582         info = blk->bp->data;
1583         ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1584         leaf = (xfs_attr_leafblock_t *)info;
1585         count = be16_to_cpu(leaf->hdr.count);
1586         bytes = sizeof(xfs_attr_leaf_hdr_t) +
1587                 count * sizeof(xfs_attr_leaf_entry_t) +
1588                 be16_to_cpu(leaf->hdr.usedbytes);
1589         if (bytes > (state->blocksize >> 1)) {
1590                 *action = 0;    /* blk over 50%, don't try to join */
1591                 return(0);
1592         }
1593
1594         /*
1595          * Check for the degenerate case of the block being empty.
1596          * If the block is empty, we'll simply delete it, no need to
1597          * coalesce it with a sibling block.  We choose (arbitrarily)
1598          * to merge with the forward block unless it is NULL.
1599          */
1600         if (count == 0) {
1601                 /*
1602                  * Make altpath point to the block we want to keep and
1603                  * path point to the block we want to drop (this one).
1604                  */
1605                 forward = (info->forw != 0);
1606                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1607                 error = xfs_da_path_shift(state, &state->altpath, forward,
1608                                                  0, &retval);
1609                 if (error)
1610                         return(error);
1611                 if (retval) {
1612                         *action = 0;
1613                 } else {
1614                         *action = 2;
1615                 }
1616                 return(0);
1617         }
1618
1619         /*
1620          * Examine each sibling block to see if we can coalesce with
1621          * at least 25% free space to spare.  We need to figure out
1622          * whether to merge with the forward or the backward block.
1623          * We prefer coalescing with the lower numbered sibling so as
1624          * to shrink an attribute list over time.
1625          */
1626         /* start with smaller blk num */
1627         forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1628         for (i = 0; i < 2; forward = !forward, i++) {
1629                 if (forward)
1630                         blkno = be32_to_cpu(info->forw);
1631                 else
1632                         blkno = be32_to_cpu(info->back);
1633                 if (blkno == 0)
1634                         continue;
1635                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1636                                         blkno, -1, &bp, XFS_ATTR_FORK);
1637                 if (error)
1638                         return(error);
1639                 ASSERT(bp != NULL);
1640
1641                 leaf = (xfs_attr_leafblock_t *)info;
1642                 count  = be16_to_cpu(leaf->hdr.count);
1643                 bytes  = state->blocksize - (state->blocksize>>2);
1644                 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1645                 leaf = bp->data;
1646                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1647                 count += be16_to_cpu(leaf->hdr.count);
1648                 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1649                 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1650                 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1651                 xfs_da_brelse(state->args->trans, bp);
1652                 if (bytes >= 0)
1653                         break;  /* fits with at least 25% to spare */
1654         }
1655         if (i >= 2) {
1656                 *action = 0;
1657                 return(0);
1658         }
1659
1660         /*
1661          * Make altpath point to the block we want to keep (the lower
1662          * numbered block) and path point to the block we want to drop.
1663          */
1664         memcpy(&state->altpath, &state->path, sizeof(state->path));
1665         if (blkno < blk->blkno) {
1666                 error = xfs_da_path_shift(state, &state->altpath, forward,
1667                                                  0, &retval);
1668         } else {
1669                 error = xfs_da_path_shift(state, &state->path, forward,
1670                                                  0, &retval);
1671         }
1672         if (error)
1673                 return(error);
1674         if (retval) {
1675                 *action = 0;
1676         } else {
1677                 *action = 1;
1678         }
1679         return(0);
1680 }
1681
1682 /*
1683  * Remove a name from the leaf attribute list structure.
1684  *
1685  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1686  * If two leaves are 37% full, when combined they will leave 25% free.
1687  */
1688 int
1689 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1690 {
1691         xfs_attr_leafblock_t *leaf;
1692         xfs_attr_leaf_hdr_t *hdr;
1693         xfs_attr_leaf_map_t *map;
1694         xfs_attr_leaf_entry_t *entry;
1695         int before, after, smallest, entsize;
1696         int tablesize, tmp, i;
1697         xfs_mount_t *mp;
1698
1699         leaf = bp->data;
1700         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1701         hdr = &leaf->hdr;
1702         mp = args->trans->t_mountp;
1703         ASSERT((be16_to_cpu(hdr->count) > 0)
1704                 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1705         ASSERT((args->index >= 0)
1706                 && (args->index < be16_to_cpu(hdr->count)));
1707         ASSERT(be16_to_cpu(hdr->firstused) >=
1708                ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1709         entry = &leaf->entries[args->index];
1710         ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1711         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1712
1713         /*
1714          * Scan through free region table:
1715          *    check for adjacency of free'd entry with an existing one,
1716          *    find smallest free region in case we need to replace it,
1717          *    adjust any map that borders the entry table,
1718          */
1719         tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1720                                         + sizeof(xfs_attr_leaf_hdr_t);
1721         map = &hdr->freemap[0];
1722         tmp = be16_to_cpu(map->size);
1723         before = after = -1;
1724         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1725         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1726         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1727                 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1728                 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1729                 if (be16_to_cpu(map->base) == tablesize) {
1730                         be16_add_cpu(&map->base,
1731                                  -((int)sizeof(xfs_attr_leaf_entry_t)));
1732                         be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1733                 }
1734
1735                 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1736                                 == be16_to_cpu(entry->nameidx)) {
1737                         before = i;
1738                 } else if (be16_to_cpu(map->base)
1739                         == (be16_to_cpu(entry->nameidx) + entsize)) {
1740                         after = i;
1741                 } else if (be16_to_cpu(map->size) < tmp) {
1742                         tmp = be16_to_cpu(map->size);
1743                         smallest = i;
1744                 }
1745         }
1746
1747         /*
1748          * Coalesce adjacent freemap regions,
1749          * or replace the smallest region.
1750          */
1751         if ((before >= 0) || (after >= 0)) {
1752                 if ((before >= 0) && (after >= 0)) {
1753                         map = &hdr->freemap[before];
1754                         be16_add_cpu(&map->size, entsize);
1755                         be16_add_cpu(&map->size,
1756                                  be16_to_cpu(hdr->freemap[after].size));
1757                         hdr->freemap[after].base = 0;
1758                         hdr->freemap[after].size = 0;
1759                 } else if (before >= 0) {
1760                         map = &hdr->freemap[before];
1761                         be16_add_cpu(&map->size, entsize);
1762                 } else {
1763                         map = &hdr->freemap[after];
1764                         /* both on-disk, don't endian flip twice */
1765                         map->base = entry->nameidx;
1766                         be16_add_cpu(&map->size, entsize);
1767                 }
1768         } else {
1769                 /*
1770                  * Replace smallest region (if it is smaller than free'd entry)
1771                  */
1772                 map = &hdr->freemap[smallest];
1773                 if (be16_to_cpu(map->size) < entsize) {
1774                         map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1775                         map->size = cpu_to_be16(entsize);
1776                 }
1777         }
1778
1779         /*
1780          * Did we remove the first entry?
1781          */
1782         if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1783                 smallest = 1;
1784         else
1785                 smallest = 0;
1786
1787         /*
1788          * Compress the remaining entries and zero out the removed stuff.
1789          */
1790         memset(XFS_ATTR_LEAF_NAME(leaf, args->index), 0, entsize);
1791         be16_add_cpu(&hdr->usedbytes, -entsize);
1792         xfs_da_log_buf(args->trans, bp,
1793              XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1794                                    entsize));
1795
1796         tmp = (be16_to_cpu(hdr->count) - args->index)
1797                                         * sizeof(xfs_attr_leaf_entry_t);
1798         memmove((char *)entry, (char *)(entry+1), tmp);
1799         be16_add_cpu(&hdr->count, -1);
1800         xfs_da_log_buf(args->trans, bp,
1801             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1802         entry = &leaf->entries[be16_to_cpu(hdr->count)];
1803         memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1804
1805         /*
1806          * If we removed the first entry, re-find the first used byte
1807          * in the name area.  Note that if the entry was the "firstused",
1808          * then we don't have a "hole" in our block resulting from
1809          * removing the name.
1810          */
1811         if (smallest) {
1812                 tmp = XFS_LBSIZE(mp);
1813                 entry = &leaf->entries[0];
1814                 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1815                         ASSERT(be16_to_cpu(entry->nameidx) >=
1816                                be16_to_cpu(hdr->firstused));
1817                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1818
1819                         if (be16_to_cpu(entry->nameidx) < tmp)
1820                                 tmp = be16_to_cpu(entry->nameidx);
1821                 }
1822                 hdr->firstused = cpu_to_be16(tmp);
1823                 if (!hdr->firstused) {
1824                         hdr->firstused = cpu_to_be16(
1825                                         tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1826                 }
1827         } else {
1828                 hdr->holes = 1;         /* mark as needing compaction */
1829         }
1830         xfs_da_log_buf(args->trans, bp,
1831                           XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1832
1833         /*
1834          * Check if leaf is less than 50% full, caller may want to
1835          * "join" the leaf with a sibling if so.
1836          */
1837         tmp  = sizeof(xfs_attr_leaf_hdr_t);
1838         tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1839         tmp += be16_to_cpu(leaf->hdr.usedbytes);
1840         return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1841 }
1842
1843 /*
1844  * Move all the attribute list entries from drop_leaf into save_leaf.
1845  */
1846 void
1847 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1848                                        xfs_da_state_blk_t *save_blk)
1849 {
1850         xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1851         xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1852         xfs_mount_t *mp;
1853         char *tmpbuffer;
1854
1855         /*
1856          * Set up environment.
1857          */
1858         mp = state->mp;
1859         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1860         ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1861         drop_leaf = drop_blk->bp->data;
1862         save_leaf = save_blk->bp->data;
1863         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1864         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1865         drop_hdr = &drop_leaf->hdr;
1866         save_hdr = &save_leaf->hdr;
1867
1868         /*
1869          * Save last hashval from dying block for later Btree fixup.
1870          */
1871         drop_blk->hashval = be32_to_cpu(
1872                 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1873
1874         /*
1875          * Check if we need a temp buffer, or can we do it in place.
1876          * Note that we don't check "leaf" for holes because we will
1877          * always be dropping it, toosmall() decided that for us already.
1878          */
1879         if (save_hdr->holes == 0) {
1880                 /*
1881                  * dest leaf has no holes, so we add there.  May need
1882                  * to make some room in the entry array.
1883                  */
1884                 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1885                         xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1886                              be16_to_cpu(drop_hdr->count), mp);
1887                 } else {
1888                         xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1889                                   be16_to_cpu(save_hdr->count),
1890                                   be16_to_cpu(drop_hdr->count), mp);
1891                 }
1892         } else {
1893                 /*
1894                  * Destination has holes, so we make a temporary copy
1895                  * of the leaf and add them both to that.
1896                  */
1897                 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1898                 ASSERT(tmpbuffer != NULL);
1899                 memset(tmpbuffer, 0, state->blocksize);
1900                 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1901                 tmp_hdr = &tmp_leaf->hdr;
1902                 tmp_hdr->info = save_hdr->info; /* struct copy */
1903                 tmp_hdr->count = 0;
1904                 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1905                 if (!tmp_hdr->firstused) {
1906                         tmp_hdr->firstused = cpu_to_be16(
1907                                 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1908                 }
1909                 tmp_hdr->usedbytes = 0;
1910                 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1911                         xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1912                                 be16_to_cpu(drop_hdr->count), mp);
1913                         xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1914                                   be16_to_cpu(tmp_leaf->hdr.count),
1915                                   be16_to_cpu(save_hdr->count), mp);
1916                 } else {
1917                         xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1918                                 be16_to_cpu(save_hdr->count), mp);
1919                         xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1920                                 be16_to_cpu(tmp_leaf->hdr.count),
1921                                 be16_to_cpu(drop_hdr->count), mp);
1922                 }
1923                 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1924                 kmem_free(tmpbuffer, state->blocksize);
1925         }
1926
1927         xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1928                                            state->blocksize - 1);
1929
1930         /*
1931          * Copy out last hashval in each block for B-tree code.
1932          */
1933         save_blk->hashval = be32_to_cpu(
1934                 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1935 }
1936
1937 /*========================================================================
1938  * Routines used for finding things in the Btree.
1939  *========================================================================*/
1940
1941 /*
1942  * Look up a name in a leaf attribute list structure.
1943  * This is the internal routine, it uses the caller's buffer.
1944  *
1945  * Note that duplicate keys are allowed, but only check within the
1946  * current leaf node.  The Btree code must check in adjacent leaf nodes.
1947  *
1948  * Return in args->index the index into the entry[] array of either
1949  * the found entry, or where the entry should have been (insert before
1950  * that entry).
1951  *
1952  * Don't change the args->value unless we find the attribute.
1953  */
1954 int
1955 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1956 {
1957         xfs_attr_leafblock_t *leaf;
1958         xfs_attr_leaf_entry_t *entry;
1959         xfs_attr_leaf_name_local_t *name_loc;
1960         xfs_attr_leaf_name_remote_t *name_rmt;
1961         int probe, span;
1962         xfs_dahash_t hashval;
1963
1964         leaf = bp->data;
1965         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1966         ASSERT(be16_to_cpu(leaf->hdr.count)
1967                                         < (XFS_LBSIZE(args->dp->i_mount)/8));
1968
1969         /*
1970          * Binary search.  (note: small blocks will skip this loop)
1971          */
1972         hashval = args->hashval;
1973         probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1974         for (entry = &leaf->entries[probe]; span > 4;
1975                    entry = &leaf->entries[probe]) {
1976                 span /= 2;
1977                 if (be32_to_cpu(entry->hashval) < hashval)
1978                         probe += span;
1979                 else if (be32_to_cpu(entry->hashval) > hashval)
1980                         probe -= span;
1981                 else
1982                         break;
1983         }
1984         ASSERT((probe >= 0) &&
1985                (!leaf->hdr.count
1986                || (probe < be16_to_cpu(leaf->hdr.count))));
1987         ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1988
1989         /*
1990          * Since we may have duplicate hashval's, find the first matching
1991          * hashval in the leaf.
1992          */
1993         while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1994                 entry--;
1995                 probe--;
1996         }
1997         while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1998                (be32_to_cpu(entry->hashval) < hashval)) {
1999                 entry++;
2000                 probe++;
2001         }
2002         if ((probe == be16_to_cpu(leaf->hdr.count)) ||
2003             (be32_to_cpu(entry->hashval) != hashval)) {
2004                 args->index = probe;
2005                 return(XFS_ERROR(ENOATTR));
2006         }
2007
2008         /*
2009          * Duplicate keys may be present, so search all of them for a match.
2010          */
2011         for (  ; (probe < be16_to_cpu(leaf->hdr.count)) &&
2012                         (be32_to_cpu(entry->hashval) == hashval);
2013                         entry++, probe++) {
2014 /*
2015  * GROT: Add code to remove incomplete entries.
2016  */
2017                 /*
2018                  * If we are looking for INCOMPLETE entries, show only those.
2019                  * If we are looking for complete entries, show only those.
2020                  */
2021                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2022                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2023                         continue;
2024                 }
2025                 if (entry->flags & XFS_ATTR_LOCAL) {
2026                         name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, probe);
2027                         if (name_loc->namelen != args->namelen)
2028                                 continue;
2029                         if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
2030                                 continue;
2031                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2032                                 continue;
2033                         args->index = probe;
2034                         return(XFS_ERROR(EEXIST));
2035                 } else {
2036                         name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, probe);
2037                         if (name_rmt->namelen != args->namelen)
2038                                 continue;
2039                         if (memcmp(args->name, (char *)name_rmt->name,
2040                                              args->namelen) != 0)
2041                                 continue;
2042                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2043                                 continue;
2044                         args->index = probe;
2045                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2046                         args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2047                                                    be32_to_cpu(name_rmt->valuelen));
2048                         return(XFS_ERROR(EEXIST));
2049                 }
2050         }
2051         args->index = probe;
2052         return(XFS_ERROR(ENOATTR));
2053 }
2054
2055 /*
2056  * Get the value associated with an attribute name from a leaf attribute
2057  * list structure.
2058  */
2059 int
2060 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2061 {
2062         int valuelen;
2063         xfs_attr_leafblock_t *leaf;
2064         xfs_attr_leaf_entry_t *entry;
2065         xfs_attr_leaf_name_local_t *name_loc;
2066         xfs_attr_leaf_name_remote_t *name_rmt;
2067
2068         leaf = bp->data;
2069         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2070         ASSERT(be16_to_cpu(leaf->hdr.count)
2071                                         < (XFS_LBSIZE(args->dp->i_mount)/8));
2072         ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2073
2074         entry = &leaf->entries[args->index];
2075         if (entry->flags & XFS_ATTR_LOCAL) {
2076                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2077                 ASSERT(name_loc->namelen == args->namelen);
2078                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2079                 valuelen = be16_to_cpu(name_loc->valuelen);
2080                 if (args->flags & ATTR_KERNOVAL) {
2081                         args->valuelen = valuelen;
2082                         return(0);
2083                 }
2084                 if (args->valuelen < valuelen) {
2085                         args->valuelen = valuelen;
2086                         return(XFS_ERROR(ERANGE));
2087                 }
2088                 args->valuelen = valuelen;
2089                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2090         } else {
2091                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2092                 ASSERT(name_rmt->namelen == args->namelen);
2093                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2094                 valuelen = be32_to_cpu(name_rmt->valuelen);
2095                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2096                 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2097                 if (args->flags & ATTR_KERNOVAL) {
2098                         args->valuelen = valuelen;
2099                         return(0);
2100                 }
2101                 if (args->valuelen < valuelen) {
2102                         args->valuelen = valuelen;
2103                         return(XFS_ERROR(ERANGE));
2104                 }
2105                 args->valuelen = valuelen;
2106         }
2107         return(0);
2108 }
2109
2110 /*========================================================================
2111  * Utility routines.
2112  *========================================================================*/
2113
2114 /*
2115  * Move the indicated entries from one leaf to another.
2116  * NOTE: this routine modifies both source and destination leaves.
2117  */
2118 /*ARGSUSED*/
2119 STATIC void
2120 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2121                         xfs_attr_leafblock_t *leaf_d, int start_d,
2122                         int count, xfs_mount_t *mp)
2123 {
2124         xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2125         xfs_attr_leaf_entry_t *entry_s, *entry_d;
2126         int desti, tmp, i;
2127
2128         /*
2129          * Check for nothing to do.
2130          */
2131         if (count == 0)
2132                 return;
2133
2134         /*
2135          * Set up environment.
2136          */
2137         ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2138         ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2139         hdr_s = &leaf_s->hdr;
2140         hdr_d = &leaf_d->hdr;
2141         ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2142                (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2143         ASSERT(be16_to_cpu(hdr_s->firstused) >=
2144                 ((be16_to_cpu(hdr_s->count)
2145                                         * sizeof(*entry_s))+sizeof(*hdr_s)));
2146         ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2147         ASSERT(be16_to_cpu(hdr_d->firstused) >=
2148                 ((be16_to_cpu(hdr_d->count)
2149                                         * sizeof(*entry_d))+sizeof(*hdr_d)));
2150
2151         ASSERT(start_s < be16_to_cpu(hdr_s->count));
2152         ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2153         ASSERT(count <= be16_to_cpu(hdr_s->count));
2154
2155         /*
2156          * Move the entries in the destination leaf up to make a hole?
2157          */
2158         if (start_d < be16_to_cpu(hdr_d->count)) {
2159                 tmp  = be16_to_cpu(hdr_d->count) - start_d;
2160                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2161                 entry_s = &leaf_d->entries[start_d];
2162                 entry_d = &leaf_d->entries[start_d + count];
2163                 memmove((char *)entry_d, (char *)entry_s, tmp);
2164         }
2165
2166         /*
2167          * Copy all entry's in the same (sorted) order,
2168          * but allocate attribute info packed and in sequence.
2169          */
2170         entry_s = &leaf_s->entries[start_s];
2171         entry_d = &leaf_d->entries[start_d];
2172         desti = start_d;
2173         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2174                 ASSERT(be16_to_cpu(entry_s->nameidx)
2175                                 >= be16_to_cpu(hdr_s->firstused));
2176                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2177 #ifdef GROT
2178                 /*
2179                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2180                  * may also need to change the insertion index.  Code turned
2181                  * off for 6.2, should be revisited later.
2182                  */
2183                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2184                         memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2185                         be16_add_cpu(&hdr_s->usedbytes, -tmp);
2186                         be16_add_cpu(&hdr_s->count, -1);
2187                         entry_d--;      /* to compensate for ++ in loop hdr */
2188                         desti--;
2189                         if ((start_s + i) < offset)
2190                                 result++;       /* insertion index adjustment */
2191                 } else {
2192 #endif /* GROT */
2193                         be16_add_cpu(&hdr_d->firstused, -tmp);
2194                         /* both on-disk, don't endian flip twice */
2195                         entry_d->hashval = entry_s->hashval;
2196                         /* both on-disk, don't endian flip twice */
2197                         entry_d->nameidx = hdr_d->firstused;
2198                         entry_d->flags = entry_s->flags;
2199                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2200                                                         <= XFS_LBSIZE(mp));
2201                         memmove(XFS_ATTR_LEAF_NAME(leaf_d, desti),
2202                                 XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), tmp);
2203                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2204                                                         <= XFS_LBSIZE(mp));
2205                         memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2206                         be16_add_cpu(&hdr_s->usedbytes, -tmp);
2207                         be16_add_cpu(&hdr_d->usedbytes, tmp);
2208                         be16_add_cpu(&hdr_s->count, -1);
2209                         be16_add_cpu(&hdr_d->count, 1);
2210                         tmp = be16_to_cpu(hdr_d->count)
2211                                                 * sizeof(xfs_attr_leaf_entry_t)
2212                                                 + sizeof(xfs_attr_leaf_hdr_t);
2213                         ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2214 #ifdef GROT
2215                 }
2216 #endif /* GROT */
2217         }
2218
2219         /*
2220          * Zero out the entries we just copied.
2221          */
2222         if (start_s == be16_to_cpu(hdr_s->count)) {
2223                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2224                 entry_s = &leaf_s->entries[start_s];
2225                 ASSERT(((char *)entry_s + tmp) <=
2226                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2227                 memset((char *)entry_s, 0, tmp);
2228         } else {
2229                 /*
2230                  * Move the remaining entries down to fill the hole,
2231                  * then zero the entries at the top.
2232                  */
2233                 tmp  = be16_to_cpu(hdr_s->count) - count;
2234                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2235                 entry_s = &leaf_s->entries[start_s + count];
2236                 entry_d = &leaf_s->entries[start_s];
2237                 memmove((char *)entry_d, (char *)entry_s, tmp);
2238
2239                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2240                 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2241                 ASSERT(((char *)entry_s + tmp) <=
2242                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2243                 memset((char *)entry_s, 0, tmp);
2244         }
2245
2246         /*
2247          * Fill in the freemap information
2248          */
2249         hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2250         be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2251                         sizeof(xfs_attr_leaf_entry_t));
2252         hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2253                               - be16_to_cpu(hdr_d->freemap[0].base));
2254         hdr_d->freemap[1].base = 0;
2255         hdr_d->freemap[2].base = 0;
2256         hdr_d->freemap[1].size = 0;
2257         hdr_d->freemap[2].size = 0;
2258         hdr_s->holes = 1;       /* leaf may not be compact */
2259 }
2260
2261 /*
2262  * Compare two leaf blocks "order".
2263  * Return 0 unless leaf2 should go before leaf1.
2264  */
2265 int
2266 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2267 {
2268         xfs_attr_leafblock_t *leaf1, *leaf2;
2269
2270         leaf1 = leaf1_bp->data;
2271         leaf2 = leaf2_bp->data;
2272         ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2273                (be16_to_cpu(leaf2->hdr.info.magic) == 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(xfs_dabuf_t *bp, int *count)
2292 {
2293         xfs_attr_leafblock_t *leaf;
2294
2295         leaf = bp->data;
2296         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2297         if (count)
2298                 *count = be16_to_cpu(leaf->hdr.count);
2299         if (!leaf->hdr.count)
2300                 return(0);
2301         return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2302 }
2303
2304 /*
2305  * Calculate the number of bytes used to store the indicated attribute
2306  * (whether local or remote only calculate bytes in this block).
2307  */
2308 STATIC int
2309 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2310 {
2311         xfs_attr_leaf_name_local_t *name_loc;
2312         xfs_attr_leaf_name_remote_t *name_rmt;
2313         int size;
2314
2315         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2316         if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2317                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, index);
2318                 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc->namelen,
2319                                                    be16_to_cpu(name_loc->valuelen));
2320         } else {
2321                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, index);
2322                 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt->namelen);
2323         }
2324         return(size);
2325 }
2326
2327 /*
2328  * Calculate the number of bytes that would be required to store the new
2329  * attribute (whether local or remote only calculate bytes in this block).
2330  * This routine decides as a side effect whether the attribute will be
2331  * a "local" or a "remote" attribute.
2332  */
2333 int
2334 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2335 {
2336         int size;
2337
2338         size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen, valuelen);
2339         if (size < XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize)) {
2340                 if (local) {
2341                         *local = 1;
2342                 }
2343         } else {
2344                 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen);
2345                 if (local) {
2346                         *local = 0;
2347                 }
2348         }
2349         return(size);
2350 }
2351
2352 /*
2353  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2354  */
2355 int
2356 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2357 {
2358         attrlist_cursor_kern_t *cursor;
2359         xfs_attr_leafblock_t *leaf;
2360         xfs_attr_leaf_entry_t *entry;
2361         int retval, i;
2362
2363         ASSERT(bp != NULL);
2364         leaf = bp->data;
2365         cursor = context->cursor;
2366         cursor->initted = 1;
2367
2368         xfs_attr_trace_l_cl("blk start", context, leaf);
2369
2370         /*
2371          * Re-find our place in the leaf block if this is a new syscall.
2372          */
2373         if (context->resynch) {
2374                 entry = &leaf->entries[0];
2375                 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2376                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2377                                 if (cursor->offset == context->dupcnt) {
2378                                         context->dupcnt = 0;
2379                                         break;
2380                                 }
2381                                 context->dupcnt++;
2382                         } else if (be32_to_cpu(entry->hashval) >
2383                                         cursor->hashval) {
2384                                 context->dupcnt = 0;
2385                                 break;
2386                         }
2387                 }
2388                 if (i == be16_to_cpu(leaf->hdr.count)) {
2389                         xfs_attr_trace_l_c("not found", context);
2390                         return(0);
2391                 }
2392         } else {
2393                 entry = &leaf->entries[0];
2394                 i = 0;
2395         }
2396         context->resynch = 0;
2397
2398         /*
2399          * We have found our place, start copying out the new attributes.
2400          */
2401         retval = 0;
2402         for (  ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2403                 attrnames_t *namesp;
2404
2405                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2406                         cursor->hashval = be32_to_cpu(entry->hashval);
2407                         cursor->offset = 0;
2408                 }
2409
2410                 if (entry->flags & XFS_ATTR_INCOMPLETE)
2411                         continue;               /* skip incomplete entries */
2412                 if (!xfs_attr_namesp_match_overrides(context->flags, entry->flags))
2413                         continue;
2414
2415                 namesp = xfs_attr_flags_namesp(entry->flags);
2416
2417                 if (entry->flags & XFS_ATTR_LOCAL) {
2418                         xfs_attr_leaf_name_local_t *name_loc =
2419                                 XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
2420
2421                         retval = context->put_listent(context,
2422                                                 namesp,
2423                                                 (char *)name_loc->nameval,
2424                                                 (int)name_loc->namelen,
2425                                                 be16_to_cpu(name_loc->valuelen),
2426                                                 (char *)&name_loc->nameval[name_loc->namelen]);
2427                         if (retval)
2428                                 return retval;
2429                 } else {
2430                         xfs_attr_leaf_name_remote_t *name_rmt =
2431                                 XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2432
2433                         int valuelen = be32_to_cpu(name_rmt->valuelen);
2434
2435                         if (context->put_value) {
2436                                 xfs_da_args_t args;
2437
2438                                 memset((char *)&args, 0, sizeof(args));
2439                                 args.dp = context->dp;
2440                                 args.whichfork = XFS_ATTR_FORK;
2441                                 args.valuelen = valuelen;
2442                                 args.value = kmem_alloc(valuelen, KM_SLEEP);
2443                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2444                                 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2445                                 retval = xfs_attr_rmtval_get(&args);
2446                                 if (retval)
2447                                         return retval;
2448                                 retval = context->put_listent(context,
2449                                                 namesp,
2450                                                 (char *)name_rmt->name,
2451                                                 (int)name_rmt->namelen,
2452                                                 valuelen,
2453                                                 (char*)args.value);
2454                                 kmem_free(args.value, valuelen);
2455                         }
2456                         else {
2457                                 retval = context->put_listent(context,
2458                                                 namesp,
2459                                                 (char *)name_rmt->name,
2460                                                 (int)name_rmt->namelen,
2461                                                 valuelen,
2462                                                 NULL);
2463                         }
2464                         if (retval)
2465                                 return retval;
2466                 }
2467                 if (context->seen_enough)
2468                         break;
2469                 cursor->offset++;
2470         }
2471         xfs_attr_trace_l_cl("blk end", context, leaf);
2472         return(retval);
2473 }
2474
2475
2476 /*========================================================================
2477  * Manage the INCOMPLETE flag in a leaf entry
2478  *========================================================================*/
2479
2480 /*
2481  * Clear the INCOMPLETE flag on an entry in a leaf block.
2482  */
2483 int
2484 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2485 {
2486         xfs_attr_leafblock_t *leaf;
2487         xfs_attr_leaf_entry_t *entry;
2488         xfs_attr_leaf_name_remote_t *name_rmt;
2489         xfs_dabuf_t *bp;
2490         int error;
2491 #ifdef DEBUG
2492         xfs_attr_leaf_name_local_t *name_loc;
2493         int namelen;
2494         char *name;
2495 #endif /* DEBUG */
2496
2497         /*
2498          * Set up the operation.
2499          */
2500         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2501                                              XFS_ATTR_FORK);
2502         if (error) {
2503                 return(error);
2504         }
2505         ASSERT(bp != NULL);
2506
2507         leaf = bp->data;
2508         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2509         ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2510         ASSERT(args->index >= 0);
2511         entry = &leaf->entries[ args->index ];
2512         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2513
2514 #ifdef DEBUG
2515         if (entry->flags & XFS_ATTR_LOCAL) {
2516                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2517                 namelen = name_loc->namelen;
2518                 name = (char *)name_loc->nameval;
2519         } else {
2520                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2521                 namelen = name_rmt->namelen;
2522                 name = (char *)name_rmt->name;
2523         }
2524         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2525         ASSERT(namelen == args->namelen);
2526         ASSERT(memcmp(name, args->name, namelen) == 0);
2527 #endif /* DEBUG */
2528
2529         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2530         xfs_da_log_buf(args->trans, bp,
2531                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2532
2533         if (args->rmtblkno) {
2534                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2535                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2536                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2537                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2538                 xfs_da_log_buf(args->trans, bp,
2539                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2540         }
2541         xfs_da_buf_done(bp);
2542
2543         /*
2544          * Commit the flag value change and start the next trans in series.
2545          */
2546         error = xfs_attr_rolltrans(&args->trans, args->dp);
2547
2548         return(error);
2549 }
2550
2551 /*
2552  * Set the INCOMPLETE flag on an entry in a leaf block.
2553  */
2554 int
2555 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2556 {
2557         xfs_attr_leafblock_t *leaf;
2558         xfs_attr_leaf_entry_t *entry;
2559         xfs_attr_leaf_name_remote_t *name_rmt;
2560         xfs_dabuf_t *bp;
2561         int error;
2562
2563         /*
2564          * Set up the operation.
2565          */
2566         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2567                                              XFS_ATTR_FORK);
2568         if (error) {
2569                 return(error);
2570         }
2571         ASSERT(bp != NULL);
2572
2573         leaf = bp->data;
2574         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2575         ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2576         ASSERT(args->index >= 0);
2577         entry = &leaf->entries[ args->index ];
2578
2579         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2580         entry->flags |= XFS_ATTR_INCOMPLETE;
2581         xfs_da_log_buf(args->trans, bp,
2582                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2583         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2584                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2585                 name_rmt->valueblk = 0;
2586                 name_rmt->valuelen = 0;
2587                 xfs_da_log_buf(args->trans, bp,
2588                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2589         }
2590         xfs_da_buf_done(bp);
2591
2592         /*
2593          * Commit the flag value change and start the next trans in series.
2594          */
2595         error = xfs_attr_rolltrans(&args->trans, args->dp);
2596
2597         return(error);
2598 }
2599
2600 /*
2601  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2602  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2603  * entry given by args->blkno2/index2.
2604  *
2605  * Note that they could be in different blocks, or in the same block.
2606  */
2607 int
2608 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2609 {
2610         xfs_attr_leafblock_t *leaf1, *leaf2;
2611         xfs_attr_leaf_entry_t *entry1, *entry2;
2612         xfs_attr_leaf_name_remote_t *name_rmt;
2613         xfs_dabuf_t *bp1, *bp2;
2614         int error;
2615 #ifdef DEBUG
2616         xfs_attr_leaf_name_local_t *name_loc;
2617         int namelen1, namelen2;
2618         char *name1, *name2;
2619 #endif /* DEBUG */
2620
2621         /*
2622          * Read the block containing the "old" attr
2623          */
2624         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2625                                              XFS_ATTR_FORK);
2626         if (error) {
2627                 return(error);
2628         }
2629         ASSERT(bp1 != NULL);
2630
2631         /*
2632          * Read the block containing the "new" attr, if it is different
2633          */
2634         if (args->blkno2 != args->blkno) {
2635                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2636                                         -1, &bp2, XFS_ATTR_FORK);
2637                 if (error) {
2638                         return(error);
2639                 }
2640                 ASSERT(bp2 != NULL);
2641         } else {
2642                 bp2 = bp1;
2643         }
2644
2645         leaf1 = bp1->data;
2646         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2647         ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2648         ASSERT(args->index >= 0);
2649         entry1 = &leaf1->entries[ args->index ];
2650
2651         leaf2 = bp2->data;
2652         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2653         ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2654         ASSERT(args->index2 >= 0);
2655         entry2 = &leaf2->entries[ args->index2 ];
2656
2657 #ifdef DEBUG
2658         if (entry1->flags & XFS_ATTR_LOCAL) {
2659                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf1, args->index);
2660                 namelen1 = name_loc->namelen;
2661                 name1 = (char *)name_loc->nameval;
2662         } else {
2663                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2664                 namelen1 = name_rmt->namelen;
2665                 name1 = (char *)name_rmt->name;
2666         }
2667         if (entry2->flags & XFS_ATTR_LOCAL) {
2668                 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf2, args->index2);
2669                 namelen2 = name_loc->namelen;
2670                 name2 = (char *)name_loc->nameval;
2671         } else {
2672                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2673                 namelen2 = name_rmt->namelen;
2674                 name2 = (char *)name_rmt->name;
2675         }
2676         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2677         ASSERT(namelen1 == namelen2);
2678         ASSERT(memcmp(name1, name2, namelen1) == 0);
2679 #endif /* DEBUG */
2680
2681         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2682         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2683
2684         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2685         xfs_da_log_buf(args->trans, bp1,
2686                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2687         if (args->rmtblkno) {
2688                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2689                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2690                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2691                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2692                 xfs_da_log_buf(args->trans, bp1,
2693                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2694         }
2695
2696         entry2->flags |= XFS_ATTR_INCOMPLETE;
2697         xfs_da_log_buf(args->trans, bp2,
2698                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2699         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2700                 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2701                 name_rmt->valueblk = 0;
2702                 name_rmt->valuelen = 0;
2703                 xfs_da_log_buf(args->trans, bp2,
2704                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2705         }
2706         xfs_da_buf_done(bp1);
2707         if (bp1 != bp2)
2708                 xfs_da_buf_done(bp2);
2709
2710         /*
2711          * Commit the flag value change and start the next trans in series.
2712          */
2713         error = xfs_attr_rolltrans(&args->trans, args->dp);
2714
2715         return(error);
2716 }
2717
2718 /*========================================================================
2719  * Indiscriminately delete the entire attribute fork
2720  *========================================================================*/
2721
2722 /*
2723  * Recurse (gasp!) through the attribute nodes until we find leaves.
2724  * We're doing a depth-first traversal in order to invalidate everything.
2725  */
2726 int
2727 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2728 {
2729         xfs_da_blkinfo_t *info;
2730         xfs_daddr_t blkno;
2731         xfs_dabuf_t *bp;
2732         int error;
2733
2734         /*
2735          * Read block 0 to see what we have to work with.
2736          * We only get here if we have extents, since we remove
2737          * the extents in reverse order the extent containing
2738          * block 0 must still be there.
2739          */
2740         error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2741         if (error)
2742                 return(error);
2743         blkno = xfs_da_blkno(bp);
2744
2745         /*
2746          * Invalidate the tree, even if the "tree" is only a single leaf block.
2747          * This is a depth-first traversal!
2748          */
2749         info = bp->data;
2750         if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2751                 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2752         } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2753                 error = xfs_attr_leaf_inactive(trans, dp, bp);
2754         } else {
2755                 error = XFS_ERROR(EIO);
2756                 xfs_da_brelse(*trans, bp);
2757         }
2758         if (error)
2759                 return(error);
2760
2761         /*
2762          * Invalidate the incore copy of the root block.
2763          */
2764         error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2765         if (error)
2766                 return(error);
2767         xfs_da_binval(*trans, bp);      /* remove from cache */
2768         /*
2769          * Commit the invalidate and start the next transaction.
2770          */
2771         error = xfs_attr_rolltrans(trans, dp);
2772
2773         return (error);
2774 }
2775
2776 /*
2777  * Recurse (gasp!) through the attribute nodes until we find leaves.
2778  * We're doing a depth-first traversal in order to invalidate everything.
2779  */
2780 STATIC int
2781 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2782                                    int level)
2783 {
2784         xfs_da_blkinfo_t *info;
2785         xfs_da_intnode_t *node;
2786         xfs_dablk_t child_fsb;
2787         xfs_daddr_t parent_blkno, child_blkno;
2788         int error, count, i;
2789         xfs_dabuf_t *child_bp;
2790
2791         /*
2792          * Since this code is recursive (gasp!) we must protect ourselves.
2793          */
2794         if (level > XFS_DA_NODE_MAXDEPTH) {
2795                 xfs_da_brelse(*trans, bp);      /* no locks for later trans */
2796                 return(XFS_ERROR(EIO));
2797         }
2798
2799         node = bp->data;
2800         ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2801         parent_blkno = xfs_da_blkno(bp);        /* save for re-read later */
2802         count = be16_to_cpu(node->hdr.count);
2803         if (!count) {
2804                 xfs_da_brelse(*trans, bp);
2805                 return(0);
2806         }
2807         child_fsb = be32_to_cpu(node->btree[0].before);
2808         xfs_da_brelse(*trans, bp);      /* no locks for later trans */
2809
2810         /*
2811          * If this is the node level just above the leaves, simply loop
2812          * over the leaves removing all of them.  If this is higher up
2813          * in the tree, recurse downward.
2814          */
2815         for (i = 0; i < count; i++) {
2816                 /*
2817                  * Read the subsidiary block to see what we have to work with.
2818                  * Don't do this in a transaction.  This is a depth-first
2819                  * traversal of the tree so we may deal with many blocks
2820                  * before we come back to this one.
2821                  */
2822                 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2823                                                 XFS_ATTR_FORK);
2824                 if (error)
2825                         return(error);
2826                 if (child_bp) {
2827                                                 /* save for re-read later */
2828                         child_blkno = xfs_da_blkno(child_bp);
2829
2830                         /*
2831                          * Invalidate the subtree, however we have to.
2832                          */
2833                         info = child_bp->data;
2834                         if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2835                                 error = xfs_attr_node_inactive(trans, dp,
2836                                                 child_bp, level+1);
2837                         } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2838                                 error = xfs_attr_leaf_inactive(trans, dp,
2839                                                 child_bp);
2840                         } else {
2841                                 error = XFS_ERROR(EIO);
2842                                 xfs_da_brelse(*trans, child_bp);
2843                         }
2844                         if (error)
2845                                 return(error);
2846
2847                         /*
2848                          * Remove the subsidiary block from the cache
2849                          * and from the log.
2850                          */
2851                         error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2852                                 &child_bp, XFS_ATTR_FORK);
2853                         if (error)
2854                                 return(error);
2855                         xfs_da_binval(*trans, child_bp);
2856                 }
2857
2858                 /*
2859                  * If we're not done, re-read the parent to get the next
2860                  * child block number.
2861                  */
2862                 if ((i+1) < count) {
2863                         error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2864                                 &bp, XFS_ATTR_FORK);
2865                         if (error)
2866                                 return(error);
2867                         child_fsb = be32_to_cpu(node->btree[i+1].before);
2868                         xfs_da_brelse(*trans, bp);
2869                 }
2870                 /*
2871                  * Atomically commit the whole invalidate stuff.
2872                  */
2873                 if ((error = xfs_attr_rolltrans(trans, dp)))
2874                         return (error);
2875         }
2876
2877         return(0);
2878 }
2879
2880 /*
2881  * Invalidate all of the "remote" value regions pointed to by a particular
2882  * leaf block.
2883  * Note that we must release the lock on the buffer so that we are not
2884  * caught holding something that the logging code wants to flush to disk.
2885  */
2886 STATIC int
2887 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *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->data;
2896         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == 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_da_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_da_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, size);
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(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
2987                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2988                                         NULL, 0, &map, &nmap, NULL, NULL);
2989                 if (error) {
2990                         return(error);
2991                 }
2992                 ASSERT(nmap == 1);
2993                 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2994
2995                 /*
2996                  * If it's a hole, these are already unmapped
2997                  * so there's nothing to invalidate.
2998                  */
2999                 if (map.br_startblock != HOLESTARTBLOCK) {
3000
3001                         dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3002                                                   map.br_startblock);
3003                         dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3004                                                 map.br_blockcount);
3005                         bp = xfs_trans_get_buf(*trans,
3006                                         dp->i_mount->m_ddev_targp,
3007                                         dblkno, dblkcnt, XFS_BUF_LOCK);
3008                         xfs_trans_binval(*trans, bp);
3009                         /*
3010                          * Roll to next transaction.
3011                          */
3012                         if ((error = xfs_attr_rolltrans(trans, dp)))
3013                                 return (error);
3014                 }
3015
3016                 tblkno += map.br_blockcount;
3017                 tblkcnt -= map.br_blockcount;
3018         }
3019
3020         return(0);
3021 }
3022
3023
3024 /*
3025  * Roll from one trans in the sequence of PERMANENT transactions to the next.
3026  */
3027 int
3028 xfs_attr_rolltrans(xfs_trans_t **transp, xfs_inode_t *dp)
3029 {
3030         xfs_trans_t *trans;
3031         unsigned int logres, count;
3032         int     error;
3033
3034         /*
3035          * Ensure that the inode is always logged.
3036          */
3037         trans = *transp;
3038         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
3039
3040         /*
3041          * Copy the critical parameters from one trans to the next.
3042          */
3043         logres = trans->t_log_res;
3044         count = trans->t_log_count;
3045         *transp = xfs_trans_dup(trans);
3046
3047         /*
3048          * Commit the current transaction.
3049          * If this commit failed, then it'd just unlock those items that
3050          * are not marked ihold. That also means that a filesystem shutdown
3051          * is in progress. The caller takes the responsibility to cancel
3052          * the duplicate transaction that gets returned.
3053          */
3054         if ((error = xfs_trans_commit(trans, 0)))
3055                 return (error);
3056
3057         trans = *transp;
3058
3059         /*
3060          * Reserve space in the log for th next transaction.
3061          * This also pushes items in the "AIL", the list of logged items,
3062          * out to disk if they are taking up space at the tail of the log
3063          * that we want to use.  This requires that either nothing be locked
3064          * across this call, or that anything that is locked be logged in
3065          * the prior and the next transactions.
3066          */
3067         error = xfs_trans_reserve(trans, 0, logres, 0,
3068                                   XFS_TRANS_PERM_LOG_RES, count);
3069         /*
3070          *  Ensure that the inode is in the new transaction and locked.
3071          */
3072         if (!error) {
3073                 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
3074                 xfs_trans_ihold(trans, dp);
3075         }
3076         return (error);
3077
3078 }